|
1. |
Critical Review. Approaching the use of oscillating reactions for analytical monitoring |
|
Analyst,
Volume 123,
Issue 2,
1998,
Page 1-8
Rafael Jiménez-Prieto,
Preview
|
PDF (121KB)
|
|
摘要:
Critical Review Approaching the use of oscillating reactions for analytical monitoring Rafael Jim�enez-Prieto, Manuel Silva and Dolores P�erez-Bendito* Department of Analytical Chemistry, Faculty of Sciences, University of C�ordoba, E-14004 C�ordoba, Spain Summary of contents Introduction Physico-chemical features of oscillating chemical reactions involved in chemical analysis Belousov–Zhabotinskii reaction Copper oscillators Peroxidase–oxidase biochemical oscillators Oscillating reactions in analytical determinations Analytical use of the Belousov–Zhabotinskii reaction Determination of metal ions Determination of inorganic anions Determinations based on chaotic regimes Analyte pulse perturbation technique Validation in the analysis of real samples Conclusions References Keywords: Review; oscillating reactions; analytical monitoring Dolores P�erez-Bendito has been Professor of the Analytical Chemistry Department of the University of C�ordoba, Spain since 1981.She has been head of this Department for 12 years. Her research interests include trace analysis, molecular spectroscopy, kinetic methods of analysis (with special emphasis on multicomponent analysis, stopped-flow methodology, kinetometrics, chemiluminescence spectroscopy, immunoassay and micellar catalysis) and environmental analysis. She has published extensively on these topics and is co-author of a monograph on Kinetic Methods in Analytical Chemistry and the biannual reviews of the ACS on this subject, as well as of several textbooks.Introduction Kinetic methods of analysis are currently regarded as highly effective tools in analytical chemistry for both the kinetic determination of a single species and the simultaneous kinetic determination of several species in a mixture with no prior separation.1,2 The dynamic regime of the chemical process involved in these methods is mainly monotonic; however, various types of dynamic regime have been explored in recent years in order to characterize non-linear chemical phenomena in the context of theoretical and experimental chemical kinetics. 3–7 These non-linear phenomena, known as ‘oscillating chemical reactions’, include regular oscillations, period doubling, quasi-periodicity and deterministic chaos, among others. Oscillating chemical reactions are complex systems that have so far been primarily examined in physico-chemical terms with a view to elucidating the intricate underlying mechanisms of the oscillations.Basically, an oscillating chemical reaction is one where some species (usually a reaction intermediate) exhibits fluctuations in its concentration; such fluctuations are usually simple (i.e., periodic)—non-periodicity or even chaos is also possible, however, under specific reaction conditions. The fluctuations are reflected in colour changes (if the oscillating species exhibit different colours), pH changes (variations in the H+ or OH2 concentration in the medium), redox potential changes, etc.For a chemical reaction to be the source of an autooscillating system, at least the following requirements must be met:8 (a) the system should be far from thermodynamic equilibrium, i.e., its Gibbs free energy difference (DG) should be large and negative; (b) there should be at least one autocatalytic step or, alternatively, cross-catalysis between two steps of the reaction mechanism; and (c) the system should possess at least two steady states under the initial conditions. Although oscillations are most readily apparent in open systems, closed systems also may exhibit them for a limited time (until thermodynamic equilibrium is achieved).9–11 The open system most widely used to ensure that the reaction will be permanently far from thermodynamic equilibrium is a continuously stirred tank reactor (CSTR); other types of reactors have also been successfully used for this purpose.12 A CSTR can be considered a homogeneous, well stirred system where mass and energy are continuously exchanged with the surrounding environment.Therefore, the mass balance for each initial reactant should include a positive term accounting for the amount of reactant contributed by the feed stream and two negative terms corresponding to reactant consumption by chemical reaction and the removal of unreacted reagent in the reactor’s outgoing stream, i.e., d A d d A d d A d d A d CSTR input reaction output [ ] [ ] [ ] [ ] t t t t æ è ç ö ø ÷ = æ è ç ö ø ÷ - æ è ç ö ø ÷ - æ è ç ö ø ÷ When the variation of the reactant concentration with time is zero, i.e., d[A]/dt = 0, the steady state is reached.Provided the external conditions do not change, such a state will persist indefinitely and the system will oscillate with constant amplitude and period. In the absence of an external reactant supply, Analyst, February 1998, Vol. 123 (1R–8R) 1Rthe mass balance will only include the term for reactant consumption by chemical reaction and the oscillations will be damped.This relaxation process is also interesting since transient oscillations can result from slow relaxation. The reaction can be monitored by fitting an appropriate detection system (e.g., a potentiometer, spectrometer or fluorimeter) to the CSTR. The signal thus recorded can be monitored with a microcomputer equipped with an analog-to-digital converter or a suitable peripheral (e.g., a recorder or plotter).Fig. 1 depicts a typical flow reactor for studying oscillating chemical reactions. Fig. 2 shows the typical oscillation profiles involved in different dynamic regimes (regular oscillation, period doubling and chaos). As shown below, uses of the oscillating reactions for analytical monitoring are based mainly on these non-linear phenomena. A detailed description of these and related aspects including limit cycles, bifurcation point, monostability, bistability, Fourier power spectra and Lyapounov exponents is beyond the scope of this review; interested readers are referred to pertinent books3–5 and papers.7,8 The first paper that considered the use of regular chemical oscillations for analytical monitoring was published by Tikhonova et al.13 in 1978.Since then, little additional research into the analytical potential of oscillating chemical reactions has been carried out.14 Not only regular chemical oscillations have been used for analytical purposes; for example, the potential of chaotic chemical systems was recently evaluated.15 In general, the bridge between the theoretical (physical chemistry) and practical aspects (analytical chemistry) of oscillating chemical reactions is their response to external perturbations.Thus, trace amounts of some substances have been shown to alter the complex dynamics of oscillating chemical reactions; also, the relationship of the oscillation attributes (period, amplitude, Fourier spectrum, largest Lyapunov exponent, etc.) to the external perturbation concentration can be used to construct calibration graphs.In order to improve the analytical features of oscillating reaction-based determinations, a good physico-chemical knowledge of oscillating chemical reactions is needed. In other words, constructing a realistic kinetic scheme for the whole process and performing a subsequent simulation study will be of help in understanding the role of a compound introduced into the reaction.These results are of great interest with a view to evaluating the sensitivity, precision and selectivity of an oscillating chemical reaction in relation to different compounds. Physico-chemical features of oscillating chemical reactions involved in chemical analysis This section deals with oscillating chemical reactions used so far for analytical monitoring purposes. The description is not exhaustive, and only the most relevant physico-chemical aspects related to their analytical potential (e.g., the role of various catalytic species, the effect of experimental variables on the oscillation attributes and the dynamic regimes involved) are considered.Thus, the Belousov–Zabotins and the most widely studied copper oscillator developed by Orb�an, Epstein and co-workers are dealt with in detail since these underlying oscillating chemical reactions have been used as analytical tools for developing the oscillating reaction-based determinations described in the next section.The peroxidase– oxidase biochemical oscillator is also commented on because its mechanism and the experimental parameters that affect it have been studied in depth despite the fact that this oscillator is a specially complex chemical system. Although its analytical potential has also been evaluated from theoretical data, this biochemical oscillator is a highly promising tool for quantitative enzyme analysis.Belousov–Zhabotinskii reaction One of the best known and most thoroughly investigated oscillating chemical reactions is the Belousov–Zhabotinskii (BZ) reaction, which involves the oxidation of an organic compound (usually malonic acid) by bromate ion in concentrated sulfuric acid. This reaction is catalysed by traces of transition metal ions that possess two oxidation states differing in a single electron, whether in free form [e.g., CeIII–CeIV,6,16,17 MnII–MnIII18] or as complexes [e.g., Ru(bpy)3 2+– Ru(bpy)3 3+,19–22 ferroin–ferriin15].During the reaction, the autocatalytic species HBrO2 is formed. The BZ reaction can be monitored via changes in the potential of a Pt electrode against a reference electrode, which result from variations in the concentrations of the oxidized and reduced forms of the metal catalyst or other reaction ingredients, 6,15,20 from changes in colour between the oxidized and reduced forms of the catalyst if the two absorb at different wavelengths6,17 and from changes in the chemiluminescence intensity when the reaction is catalysed by the Ru(bpy)3 2+– Ru(bpy)3 3+ couple.16,18,19,21 One interesting aspect of the BZ reaction is the search for new species that have a catalytic effect on it.Recent examples are tetraazacopper(ii) (and nickel) complexes.23 The oscillating reaction essentially involves the oxidation of the unsaturated CuII macrocyclic complex by bromate ions in a phosphoric acid medium. Song et al.24–26 examined various oscillating systems involving complexes of CuII with a 14-membered tetraaza macrocyclic ligand.In the KBrO3–H2C(COOH)2–H2SO4– Fig. 1 Schematic representation of a typical flow reactor system for studying oscillating chemical reactions. Fig. 2 Typical oscillations exhibited by oscillators used for analytical monitoring: (A) regular oscillations; (B) period doubling; and (C) chaotic regimes. 2R Analyst, February 1998, Vol. 123CuL(ClO4)2 system, where L is the tetraaza ligand, the complex acts as the catalyst for the BZ reaction;24 a non-macrocyclic ligand in phosphoric rather than sulfuric acid has also been found to result in oscillations, however.27 A ferroin complex such as FeII–5-nitro-o-phenanthroline has also been found to catalyse the BZ reaction in a sulfuric acid medium28 and to have a marked effect on the initial oscillating period and amplitude, and also on the period and duration of oscillations relative to other FeII complexes.Finally, one special modification of the BZ reaction uses the aniline–bromate–sulfuric acid system as an oscillator.29,30 The reaction has also been implemented in a flow injection system.17 The goal of much of the research on the BZ reaction has been to elucidate the mechanism of the reaction and to design model equations for accurately predicting the oscillatory behaviour of the system. Virtually all possible dynamic regimes may be experimentally observed in the BZ reaction.Each regime exists over a given range of control parameters, bifurcation points being where one regime changes to another. In regular oscillations, the oscillating period and amplitude depend strongly not only on the concentrations of the reaction ingredients but also on other factors, including temperature and the flow rates of reactants and products entering and leaving the CSTR, respectively. Thus, the oscillating amplitude generally ranges from about 30 to 100 mV with potentiometric measurements and from 0.3 to 0.5 absorbance with spectrophotometric detection.6,17,31 Also, the oscillating period typically ranges from 30 s to a few minutes, depending on the particular conditions.6,17,31 It should be pointed out that the accuracy of the parameters involved is a result of oscillations that take place over a limited range of phase space.31 The BZ reaction exhibits an induction period and, similarly to the above-mentioned oscillation attributes, its length depends on the specific reaction conditions but is usually several minutes.6,17,18 The time interval between oscillations (oscillation period) or the time to the onset of oscillations (induction period) was first used as an analytical parameter of interest by Zhabotinskii.32 One other highly influential experimental factor in this context is the presence of concomitant species in the reaction medium.In fact, changes in the oxygen concentration in the medium can alter the oscillating attributes amplitude and period. Thus, Saigusa21 has shown that when an oxygen perturbation is applied to the BZ reaction in a closed system and is then removed, the reaction exhibits phase transitions between perturbed and unperturbed oscillatory states, the latter exhibiting longer oscillation periods.The results of the perturbation experiments have been expressed in the form of phase response curves.In summary, although the effect of oxygen on the BZ reaction has been extensively studied,4,6,33 its mechanism remains controversial. Chaos has been one of the most exciting topics in chemistry over the past decade.34,35 Chaotic dynamics are built up by the expanding and folding processes of chaotic orbits and may be described in terms of the local expansion rate of nearby orbits.36,37 Chaos exhibits exponential divergence from adjacent starts. If the initial conditions of a chaotic system are known only with some margin of error, then the final outcome becomes unpredictable.This sensitivity to initial conditions or divergence is used as an indicator of chaos; the most usual measure of the divergence is called the Lyapunov exponent. 36–38 Under specific initial conditions, the BZ reaction may proceed into chaotic regimes, which has been exploited for analytical purposes.15 The methodology is based on the high sensitivity of some types of chaotic regimes of the BZ reaction to small perturbations of the initial conditions introduced by low concentrations of some metal ions.Recently, the influence of temperature on the chaotic regime near its generation threshold was investigated for the ferroin-catalysed BZ reaction in a CSTR.39 Temperature changes were found to result in a nonequilibrium transient phase in the reactor that was concomitant with an abrupt change in the reciprocal induction period. Oscillations occurred according to a probabilistic pattern near the phase-transition temperature.Copper oscillators Copper(ii) takes part as a catalyst in a major group of oscillating systems. For example, the reactions of H2O2 with Na2S2O3 and KSCN have been found to exhibit oscillations in pH and redox potential at Pt and selective copper electrodes, oscillations in the redox potential, pH and O2 concentration in the K2S2O8– Na2S2O3 system and oscillations in the redox potential for the ClO22–S22 system, the CuII–CuI redox couple being the catalyst in all cases.40–43 The most widely studied among CuII oscillators is the reaction developed by Orb�an based on the oxidation of KSCN by H2O2 in a strongly alkaline medium, which is catalysed by traces of copper.44 The reaction takes place in both open and closed systems; in the latter, however, oscillations are gradually damped and eventually disappear altogether.It can be monitored photometrically or potentiometrically.The system oscillates with an amplitude of 15–35 mV and a period from 30 s or less to several minutes. As in the BZ reaction, the system is strongly influenced by experimentariables. Thus, an increase in the CuII or H2O2 concentration decreases the oscillating amplitude and period, whereas an increase in the SCN2 concentration has little effect on the amplitude but increases the period. The temperature has a marked effect on the kinetics; thus, an increase of only 5 °C decreases the oscillating period by up to 15% without significantly altering the amplitude.45 The inclusion of foreign species, such as luminol, in the reaction medium gives rise to similar oscillating systems, exhibiting oscillations in the chemiluminescence at 424 nm.46,47 The mechanism for this oscillating reaction was established48 from previous investigations of Luo et al.49 and Wilson and Harris,50,51 on the H2O2–CuSO4 and H2O2–KSCN sub-systems.The model involves 30 kinetic equations and 26 independent variables.The reaction involves the decomposition of H2O2, catalysed by traces of CuII, which is much more favourable in a strongly alkaline than in an acidic medium. Also, the formation of a yellow superoxide–copper(i) complex [HO2–CuI], which exists only above pH 9, plays a crucial role in determining the pathway of copper catalysis under different conditions of pH;49 thus, at a high enough pH and appropriate reactant concentrations, a yellow colour is observed to appear and disappear and the potential of the Pt electrode changes with a regular periodicity.Oscillations stop below pH 9, where the formation of the HO2–CuI complex is hindered. Peroxidase–oxidase biochemical oscillators Peroxidases are a class of enzymes which usually use hydrogen peroxide to oxidize a variety of organic compounds. Some peroxidases, however, can catalyse oxidation using molecular oxygen instead of hydrogen peroxide. This type of reaction is called a peroxidase–oxidase reaction.One such reaction, which oxidizes the common biochemical reducing agent NADH is 2NADH + 2H+ + O2?2NAD+ + 2H2O where NAD+ is the oxidized form of NADH (b-nicotinamide adenine dinucleotide). This oscillating reaction is a very complex system. Since its discovery by Yamazaki et al.52 in 1965, a substantial number of papers on the models developed for reproducing many of the involved non-linear phenomena (e.g., bistability,53 mixed-mode and quasi-periodic oscillations54,55 and period doubling and chaotic behaviour56–60) have been published.Analyst, February 1998, Vol. 123 3RThe peroxidase–oxidase oscillator has a high potential for quantitative enzyme analysis; in fact, because it uses NADH (a common reducing agent in many biochemical reactions), any enzyme reaction that consumes or produces NADH should in theory lend itself to analysis using this oscillator. The potential of this oscillator for analytical monitoring was examined in two interesting papers by Olson and Scheeline.12,61 In the earlier one,61 the theoretical basis for quantitative enzyme determinations was developed from a modification of the Olsen model58 including a competing (analyte) reaction. Using period doubling, a theoretical calibration graph can be constructed on account of the sensitive dependence of the oscillatory oxygen transient on the analyte enzyme concentration.Interesting conclusions on the precision, sensitivity and approaches to the calculation of limit of detection were drawn. Although the limits of detection achieved are not particularly low, they can be improved by using appropriate experimental conditions.In the later study, Olson and Scheeline12 developed and characterized the analytical tools (15 variables were recognized and described) needed to study this particularly complex chemical system. Variables were studied in depth, which ensured good reproducibility of the oscillations by using controlled experimental parameters.It is worth noting that the effect of structurally different peroxidases (viz., horseradish peroxidase, lactoperoxidase, soybean peroxidase and coprinus peroxidase) was recently investigated by using NADH as substrate.62,63 These new data are of great interest with a view to increasing the analytical potential of this biochemical oscillator. Oscillating reactions in analytical determinations As stated above, studies on oscillating chemical reactions have preferentially been approached from the physico-chemical standpoint with a view to elucidating the intricate non-linear behaviours observed in the experimental system. In fact, few analytical applications have so far been reported; however, an appreciable number of papers on this topic have been published in the last few years. The growing interest can be ascribed to two factors, namely: (1) the physico-chemical background of oscillating chemical reactions is becoming better known every day—today, the chemist uses or develops appropriate tools to explain the complex behaviour of these reactions; and (2) the good results recently achieved in the reported analytical determinations have encouraged others to work on this exciting analytical chemical topic.In general, determinations rely on the interaction of the species to be determined with an oscillating reaction; changes in some characteristic of the oscillator (whether the induction period, oscillating amplitude or oscillating period in regular oscillations or even in chaotic regimes) in the presence and absence of the analyte are correlated with its concentration.The response of the system in the absence of analyte is taken as a reference for establishing the quantitative relationships needed to determine the species of interest. Until fairly recently, these reactions were implemented batchwise. Analytical determinations entailed preparing a fresh oscillating system in each; however, fill reactors,12 flow injection17 and the recent inception of the analyte pulse perturbation (APP) technique45 have introduced substantial practical advantages derived from the ability to operate in a continuous manner.The use of oscillating chemical reactions for analytical monitoring has concentrated on two oscillators, namely the BZ reaction and the copper oscillator described in detail above. Analytical use of the Belousov–Zhabotinskii reaction Determination of metal ions As stated above, the catalytic determination of ruthenium based on a self-oscillating chemical reaction developed by Tikhonova et al.13 in 1978 was the first attempt at using oscillating chemical reactions for analytical monitoring.In this system, the sulfates of both RuIII and RuIV increase the frequency of the oscillations (monitored through the absorbance at 360 nm) in the CeIII-catalysed BZ reaction. A linear relationship was observed between the decrease in the period of a single oscillation and the ruthenium concentration over the range 7–330 ng ml21. In addition to high sensitivity, the method exhibits appreciable precision (the RSD is < 2.8% for about 0.25–1 mg of ruthenium) and selectivity for platinium and rhodium; however, palladium interferes at levels below that of ruthenium in the sample.Karavaev et al.16 used this reaction to determine ruthenium on the basis that traces of Ru(bpy)3 2+ enhance the chemiluminescence (CL) intensity in the CeIIIcatalysed BZ reaction.Despite the increased sensitivity of CL measurements, the proposed method is less sensitive than the photometric method, but has a wider dynamic range; from log– log calibration, ruthenium can be determined at levels between 0.2 nm and 0.1 mm (mg ml21 level), although the limit of quantification can be lowered by measuring the combined CL of several oscillations.Thallium(i) and mercury(ii) have also been determined with good selectivity and sensitivity on the basis of their complexes with Br2, a critical species for the behaviour of the reaction.64 The determination of these metal ions is influenced by the temperature, so thallium(i) can be quantified at concentrations between 0.08 and 10 mg ml21 at 26 °C, whereas mercury can be measured over the same range at 35 °C. Oscillations are monitored by using potentiometric measurements with a Pt or Br ion-selective indicator electrode and an SCE reference electrode.Recently, both metal ions were determined by using the BZ reaction in a double-line flow injection (FI) manifold17 (see Fig. 3). The manifold includes a peristaltic pump, a six-port pneumatic injection valve, a 30 ml flow cell and a UV/VIS spectrophotometer. The FI system operates in the stopped-flow mode and reproducible oscillations (0.6–2.2%) can be monitored for several minutes before the oscillatory behaviour becomes damped or gas bubbles (carbon dioxide) interfere with reaction monitoring.Spatial non-homogeneity (the BZ reaction is unstirred) turns the flow-cell dimensions into the key to the oscillator’s performance. With the FI system, mercury(ii) and thallium(i) can be determined by the increase in the induction period of the BZ reaction in the presence of the metal ions in the reaction medium (the induction period is 35 s shorter than that for the reaction in the presence of 400 ng ml21 of mercury).The method exhibits a determination level similar to that of the above method and poor precision (very small variations in the initial conditions lead to larger experimental errors). In spite of these results, this is an interesting contribution to the use of oscillating chemical reactions for analytical monitoring (the analytical potencial of the procedure is clear). Fig. 3 Flow injection system for mercury(ii) and thallium(i) determination based on their perturbations on the BZ reaction.C1, C2 and C3 are 90, 90 and 200 cm long mixing coils, respectively; IV, injection valve. Solutions A (malonic acid–cerium–ferroin) and B (bromate) and the detector flow cell (30 ml) are thermostated at 25 °C. The timer controls the delay time between injection and stopping of the pump. Adapted from ref. 17. 4R Analyst, February 1998, Vol. 123It is also worth noting the determination of copper(ii)65 based on the Brigg–Rauscher reaction (a combination of the BZ and Bray–Liebhafsky reactions). In this reaction, a mixture of hydrogen peroxide and iodate in sulfuric acid is used as the oxidizing agent, manganese(ii) as the catalyst and malonic acid as the organic reaction substrate.66 Because the oscillation times are increased by increasing concentrations of copper(ii), this metal ion can be determined with a limit of detection of 0.1 mg ml21 and an RSD of 2.7% for 0.5 mg ml21 of copper.Determination of inorganic anions In addition to trace amounts of metal ions, anions have been also determined using the BZ reaction.Thus, hexacyanoferrates [Fe(CN)6 32 or Fe(CN)6 42] can significantly decrease the amplitude of this oscillator while the oscillation frequency remains almost unchanged.67 On the basis of potentiometric measurements made at 25 °C, the decrease in the amplitude is linearly proportional to the concentration of hexacyanoferrates over the range 7.0 3 1028–5.0 3 1026 m (the RSD is 2.7% at 1.0 3 1026 m).The method provides good selectivity: of the 40 foreign ions tested, only mercury(ii), thallium(i) and manganese( ii) interfere, although their tolerated limits are favourable [the maximum tolerated mole ratio for these ions relative to Fe(CN)6 32 is 10]. A study was carried out by using cyclic voltammetry to elucidate the interaction of hexacyanoferrates with the BZ reaction. The results revealed that cyanide, which results from the decomposition of hexacyanoferrate, is the main active species.The system was applied to the determination of the analytes in silver plating and photographic solutions. As can be seen from Table 1, the method is a good alternative to the determination of these compounds in this type of sample judging by the consistency between the experimental results and certified values. The inhibitory effect of chloride and iodide on this reaction have also been exploited to develop analytical methods. Chloride ion prevents the regeneration of the cerium(iv) catalyst, thus causing a decrease in the oscillation amplitude and an increase in the period.68 The decrease in the recorded voltage peaks (amplitude) measured with a bromide electrode is rectilinearly related to the chloride content from 1.0 3 1026 to 1 3 1024 m.One interesting application of the method is the determination of trace amounts of chloride in human serum. On the other hand, iodide is a strong inhibitor of the reaction.69 Thus, after the induction period produced by the addition of iodide, the decrease in amplitude of the first oscillation period is linearly related to the iodide concentration over the range 1.0 3 1025–1.0 3 1023 m.Determinations based on chaotic regimes As stated above, chaotic regimes in the BZ reaction have also been exploited for analytical purposes.15 Thus, trace amounts of manganese(ii) were determined by their perturbating effect on some types of chaotic regimes in the ferroin-catalysed BZ reaction.The Lyapunov exponent, lL, calculated by using Borland Turbo Pascal software based on the Wolf algorithm,70 and the time interval Ti from the starting point to the first maximum of the potential–time graph [Fig. 4(A)] were calculated, and the product of lL and Ti was plotted against the logarithm of the manganese(ii) concentration. The calibration graph thus obtained was linear from 1.0 3 10211 to 1.0 3 1025 m manganese [Fig. 4(B)]. As can be seen, the analytical sensitivity, expressed as the slope of the calibration plot, was very low [in other words, the variation of the product of lL and Ti was very small relative to the wide range of manganese(ii) concentrations tested]. As can be seen in Table 2, this considerably reduces the reproducibility of the method by the effect of the high sensitivity of the chaotic regimes to the experimental conditions and the inherent irreproducibility.The authors proposed the following expression for the limit of detection: lLTi = a log(CMn/C* Mn) where a and C* Mn can be calculated by extrapolation of the linear portions of the lLTi–log CMn plots. The limit of detection thus calculated for manganese(ii) was about 10215 m (3 pg ml21) much lower than those typically provided by traditional catalytic kinetic methods (about 100 pg ml21). The limit of detection is closely related to the starting ferroin Table 1 Determination of hexacyanoferrates in silver plating baths and photographic solutions using the BZ reaction.Adapted from ref. 65 Certified value*/ Found†/ Error Sample mg ml21 mg ml21 (%) Silver plating bath solution 1‡ 53.22 53.6 ± 0.32 0.71 Silver plating bath solution 2‡ 71.30 72.3 ± 0.28 1.40 Photographic solution 1 4.24 4.2 ± 0.08 20.94 Photographic solution 2 3.08 3.0 ± 0.08 22.67 * Hexacyanoferrates as hexacyanoferrate(ii) and -(iii) in silver plating baths and photographic solutions, respectively.† Means of five determinations ± s. ‡ Small amounts of potassium bromide were added to precipitate silver(i) before the determination. Fig. 4 (A) Oscillations in the BZ chaotic reaction used for the determination of manganese(ii). The time elapsed between the starting point and the first maximum on the Pt electrode potential versus time plot is shown. (B) Calibration plot for the determination of manganese using this oscillator. Adapted from ref. 15. Table 2 Determination of manganese(ii) by its perturbation on the chaotic BZ chemical system using a lL Ti versus log CMn calibration plot.Adapted from ref. 15 Manganese Manganese Error added/g ml21 found/g ml21 (%) 7.531023 (7.5 ± 1)31023 0.0 1.131025 (9 ± 2)31025 718 4.331026 (4 ± 2)31026 27.5 3.031027 (5 ± 4)31027 67 Analyst, February 1998, Vol. 123 5R(catalyst) concentration in this oscillating chemical reaction. In summary, although this method for maganese(ii) is complicated and cumbersome, it is highly analytically significant because it demonstrates for the first time that some types of chemical chaos lend themselves to analytical applications.Typical oscillation attributes such as the oscillation period and amplitude were recently used to evaluate their use in chemical analysis by using this chaotic system.71 Again, manganese(ii), and also vanadium(iv), were perturbing species; their analytical concentrations were related to the oscillation period.In fact, the chaotic BZ chemical system is very sensitive to low concentrations of these metal ions [ < 1029 m for manganese(ii) and < 1028 m for vanadium(iv)], although in different ways: the periods between oscillations decrease in the presence of vanadium(iv) and increase in the presence of manganese(ii). The calibration graphs (period versus log Cmetal ion) are complicated; in any case, vanadium(iv) can be determined over the range 1028–1025 m, where it exhibits a virtually linear relationship.Analyte pulse perturbation technique The recently introduced APP technique45 uses a CSTR and relies on the sequential perturbation of an oscillating reaction by successive additions of analytes (or standards) after the regular oscillations are restored. Maintaining optimum experimental conditions, the system remains in an oscillating state for at least 8 h, acting as a continuous indicator system. This provides a rapid, simple method for performing many determinations on the same oscillating system.This operating mode offers obvious advantages over discrete systems and endows the APP technique with high practical potential. As can be seen from Fig. 5, the experimental set-up for implementation of the APP technique is very simple. It consists of a CSTR furnished with a thermostating jacket, a Pt indicator electrode and a reference electrode connected to an analog-todigital converter in a microcomputer for monitoring oscillations and their perturbation (by potentiometry), a peristaltic pump for delivery of the reaction ingredients intended to ensure attainment of the steady state in the CSTR (one of the pump lines is used to replenish the reactor) and an autoburette or micropipette for dispensing a small volume (in the microlitre region) of the analyte solution in order to perturb the oscillating system.The analytical performance of the APP technique has been evaluated in the determination of Na2S2O3 from the decrease in the oscillating period and amplitude relative to their values before the system was perturbed by injecting the analyte.The decrease is quantitatively correlated with the analyte concentration. 45 The oscillating reaction used for this purpose was the above-described oxidation of KSCN by alkaline hydrogen peroxide, catalysed by traces of copper(ii). Fig. 6 shows the response of the oscillating system to Na2S2O3 perturbations, and also those of other analytes (e.g., an increase in the oscillating amplitude by effect of a perturbation with gallic acid72 or an increase in the oscillating period by virtue of the perturbation with a given amount of reduced glutathione,73 which resulted in first- and second-order correlations, respectively, between the responses and the injected analyte concentration).As can be seen from Table 3, the determinations are sensitive (the dynamic linear range is typically from a few nanomoles to a few micromoles) and highly precise (the RSD is usually about 1%).Of special practical significance in these determinations is the fairly high throughput achieved by using the APP technique. In fact, up to 10 samples h21 can often be processed, which is acceptable for an oscillating reaction and much better than the typical sample throughput of discrete approaches (1–2 h21 at the most). The APP technique has been applied not only to individual determinations but also to the resolution of binary mixtures of species that elicit differential responses from an oscillating Fig. 5 Manifold for implementing of the analyte pulse perturbation technique by use of a CSTR. 6R Analyst, February 1998, Vol. 123system. Thus, gallic acid and resorcinol can be determined by the increasing oscillating amplitude and period produced by the former and latter, respectively, in relation to the system in the steady state.74 The ensuing method allows both species to be determined at the micromole level with gallic acid-to-resorcinol ratios from 1 : 6 to 1 : 35, even in the presence of a synergistic effect, with acceptable precision (RSD 4.42 and 3.58%, respectively).The above determinations use potentiometric detection to monitor oscillations. Two different detection systems (the above-described potentiometric detector and a chemiluminescent system based on the use of luminol as one of the reaction ingredients) have been compared with a view to increasing their sensitivity.75 The reaction was found to be analytically useful with both types of detection system and, as expected, somewhat more sensitive with chemiluminescence detection.Validation in the analysis of real samples The true measure of the actual potential of an analytical technique can only be obtained by applying it to real samples. In this respect, the APP technique has been found to permit the use of oscillating reactions for routine analyses. Thus, a method for the rapid, straightforward determination for vanillin, paracetamol and ascorbic acid,76 and another for vitamin B6,75 all in food and pharmaceutical samples, have been developed.As can be seen from Table 3, the methods possess favourable analytical features and allow the determination of the above-mentioned species in real samples (Tables 4 and 5). Conclusions Oscillating chemical reactions as dynamic systems have traditionally aroused interest in the context of kinetic methods of analysis.However, they have largely been the subject only of academic investigations owing to their little practical interest. In the last few years, several studies have been published that show the potential of these reactions in analytical-based determinations. From these papers, the following conclusions can be drawn: (1) The key to the analytical use of oscillating chemical reactions seems to be based in their response upon perturbation with different species (analytes), the BZ reaction being one of the most often used.(2) Different non-linear regimes of the BZ reaction perturbed by both metal ions and anions have been used for analytical purposes. In general, lower limits of detection are achieved in the determination of metal ions, especially in chaotic chemical regimes. The understanding of the physico-chemical aspects of the oscillator and those of the subsequent interaction of the Fig. 6 Typical profiles for the H2O2–NaSCN–NaOH–CuSO4 oscillating system in the absence and presence of (A) sodium thiosulfate, (B) gallic acid and (C) reduced glutathione.Arrowheads indicate the times at which oscillations were perturbed. Table 3 Analytical figures of merit of oscillating reaction-based determinations based on the APP technique Dynamic linear RSD Analyte range/mmol (%) Ref. Na2S2O3 1–18 0.71 45 Gallic acid 0.075–2.0 0.67 72 Reduced glutathione 0.1–0.7 0.67 73 Resorcinol 1.5–12 1.53 74 Vanillin 1–30 0.78 76 Paracetamol 0.5–6.0 0.61 76 Ascorbic acid 0.5–5.0 4.65 76 Vitamin B6 0.5–20 3.04 75 Table 4 Use of the analyte pulse perturbation technique in food analysis. Adapted from ref. 76 Standard method*/ Found/ Error Analyte Sample mg g21 mg g21 (%) Vanillin Vanilla sugar 1 78.6 ± 5.2 83.7 ± 4.3 6.43 Vanilla sugar 2 3.6 ± 0.3 3.3 ± 0.2 28.61 Vanilla sugar 3 82.1 ± 2.7 85 ± 12 4.41 Stick vanilla 1 15.6 ± 1.1 15.1 ± 0.3 23.39 Stick vanilla 2 13.8 ± 2.3 14.5 ± 2.3 4.56 Ascorbic acid Orange juice 0.563 ± 0.002 0.527 ± 0.007 26.23 * AOAC and NBS methods for vanillin and ascorbic acid, respectively.Table 5 Use of the analyte pulse perturbation technique in pharmaceutical analysis. Adapted from refs. 75 and 76 Nominal Error Analyte Trade name content Found (%) Paracetamol Termalgin 500 mg/tablet 522 ± 57 4.50 Efferalgan 500 mg/tablet 519 ± 35 3.86 Saldeva 300 mg/tablet 307 ± 15 2.43 Cortafriol 500 mg/tablet 516 ± 46 3.30 Melabon 350 mg/tablet 364 ± 14 4.03 Propalgina plus 500 mg/bag 496 ± 25 20.82 Vitamin B6 Pacium 33.3 mg g21 33 ± 8 21.1 Vertigum 275.1 mg g21 247 ± 13 10.1 Taurobetina 112.4 mg g21 104 ± 5 27.4 Trofimilina 387.1 mg g21 370 ± 30 24.5 Analyst, February 1998, Vol. 123 7Ranalyte with it allows the development of methods with good analytical figures of merits. On the other hand, the type of detection system used to monitor the oscillating reaction is not decisive in this context. (3) The inception of the analyte pulse perturbation technique (which has been reported by using a system other than the BZ reaction) has improved the use of oscillating chemical reactions for analytical monitoring.The fact that an analyte may perturb oscillations (viz., alter their amplitude and/or period) and the ability to develop a continuous system in permanent oscillation that regains its regular state after each analyte (or standard) perturbation have opened up new avenues for practical applications of oscillating reactions, especially for organic analytes, which have not been determined using the BZ reaction. In summary, this technique has enormously facilitated the use of oscillating reactions, formerly of purely theoretical interest, for routine analytical monitoring practices, as shown in this review of their existing applications to real samples.The authors gratefully acknowledge financial funding of this work by the Direcci�on General Interministerial de Ciencia y Tecnolog�ýa (DGICyT), Spain.References 1 Mottola, H. A., Kinetic Aspects of Analytical Chemistry, Wiley, New York, 1988. 2 P�erez-Bendito, D., and Silva, M., Kinetic Methods in Analytical Chemistry, Ellis Horwood, Chichester, 1988. 3 Butenin, N. V., Neimark, Y. I., and Fuf�aev, N. A., Introducci�on a la Teor�ýa de las Oscilaciones no Lineales, Mir, Moscow, 1987. 4 Gray, P., and Scott, S. K., Chemical Oscillations and Instabilities. Non Linear Chemical Kinetics, Clarendon Press, Oxford, 1990. 5 Field, R. J., and Burger, M., Oscillations and Travelling Waves in Chemical Systems, Wiley, New York, 1985. 6 Field, F. J., and Schneider, F. W., J. Chem. Educ., 1989, 66, 195. 7 Scheeline, A., Kirkor, E. S., Kovacs-Boerger, A. E., and Olson, D. L., Mikrochim. Acta, 1995, 118, 1. 8 Melka, R. F., Olsen, B., Beavers, L., and Draeger, J. A., J. Chem. Educ., 1992, 69, 596. 9 Nagy, A., and Treindl, L., Chem. Listy, 1988, 82, 1097. 10 Yoshimoto, M., Yoshikawa, K., Mori, Y., and Hanazaki, I., Chem.Phys. Lett., 1992, 189, 18. 11 Epstein, I. R., J. Chem. Educ., 1989, 66, 191. 12 Olson, D. L., and Scheeline, A., Anal. Chim. Acta, 1993, 283, 703. 13 Tikhonova, L. P., Zakrevskaya, L.N., and Yatsimirskii, K. B., J. Anal. Chem. USSR, 1978, 33, 1991. 14 Yatsimirskii, K. B., J. Anal. Chem. USSR, 1987, 42, 1743. 15 Yatsimirskii, K. B., Strizhak, P. E., and Ivaschenko, T. S., Talanta, 1993, 40, 1227. 16 Karavaev, A. D., Kazakov, V. P., Tolstikov, G.A., Yakshin, V. V., and Khokhlova, N. L., J. Anal. Chem. USSR, 1986, 41, 42. 17 Echols, R. T., Caroll, M. K., and Tyson, J. F., Anal. Proc., 1995, 32, 3. 18 Zhuravlev, A. I., and Trainin, V. M., J. Biolumin. Chemilumin., 1990, 5, 227. 19 Weight, H. R., Angew. Chem., Int. Ed. Engl., 1992, 31, 355. 20 Kazakov, V. P., Karavayev, A. D., and Vakhidova, S. R., React. Kinet. Catal. Lett., 1991, 45, 199. 21 Saigusa, K., Chem. Phys. Lett., 1989, 157, 251. 22 Karavaev, A. D., and Kazakov, V.P., React. Kinet. Catal. Lett., 1987, 34, 15. 23 Zhao, X. D., Xu, Z. Q., Zhao, L., Wang, Y. F., Xie, F. X., and Ni, S. S., Chim. Chem. Lett., 1994, 5, 199. 24 Song, J., Ni, S., and Xu, J., Huaxue Wuli Xuebao, 1994, 7, 192. 25 Song, J., Ni, S., and Xu, J., Huaxue Wuli Xuebao, 1994, 7, 566. 26 Song, J., Ni, S., and Xu, J., Huaxue Wuli Xuebao, 1994, 10, 48. 27 Xie, F., Xu, Z., Yia, X., and Ni, S., Gaodeng Xuexiao Huaxue Xuebao, 1993, 15, 1439. 28 Gao, Q., and Yuan, W., Dalian Ligong Daxue Xuebao, 1991, 31, 545. 29 Handlirova, M., React. Kinet. Catal. Lett., 1988, 36, 207. 30 Handlirova, M., React. Kinet. Catal. Lett., 1988, 37, 145. 31 Scott, S. K., Chemical Chaos, Clarendon Press, Oxford, 1991, ch. 8. 32 Zhabotinskii, A. M., Zh. Anal. Khim., 1972, 27, 437. 33 Gyorgyi, L., Deutsch, T., and Koros, E., Int. J. Chem. React., 1987, 19, 435. 34 Tomlin, A. S., Anal. Proc., 1993, 30, 307. 35 Bishop, S. R., Anal. Proc., 1993, 30, 310. 36 Nicolis, G., J.Phys.: Condens. Matter, 1990, 2, SA47. 37 Peng, B., Petrov, V., and Showalter, K., J. Phys. Chem., 1991, 95, 4957. 38 Scott, S. K., Chemical Chaos, Clarendon Press, Oxford, 1991, ch. 2. 39 Strizhak, P. E., and Didenko, O. Z., Teor. Eksp. Khim., 1994, 30, 147. 40 Orb�an, M., React. Kinet. Catal. Lett., 1990, 42, 343. 41 Orb�an, M., and Epstein, I. R., J. Am. Chem. Soc., 1989, 111, 2891. 42 Koehler, J. M., Z. Phys. Chem. (Leipzig), 1989, 270, 545. 43 Strizhak, P. E., Izv.Akad. Nauk SSSR, Ser. Khim., 1991, 11, 2474. 44 Orb�an, M., J. Am. Chem. Soc., 1986, 108, 6893. 45 Jim�enez-Prieto, R., Silva, M., and P�erez-Bendito, D., Anal. Chem., 1995, 67, 729. 46 Amrehn, J., Resch, P., and Schneider, F. W., J. Phys. Chem., 1988, 92, 3318. 47 Sattar, S., and Epstein, I. R., J. Phys. Chem., 1990, 94, 275. 48 Luo, Y., Orb�an, M., Kustin, K., and Epstein, I. R., J. Am. Chem. Soc., 1989, 111, 4541. 49 Luo, Y., Kustin, K., and Epstein, I. R., Inorg. Chem., 1988, 27, 2489. 50 Wilson, I. R., and Harris, G. M., J. Am. Chem. Soc., 1960, 82, 4515. 51 Wilson, I. R., and Harris, G. M., J. Am. Chem. Soc., 1961, 83, 286. 52 Yamazaki, I., Yokota, K., and Nakajima, R., Biochem. Biophys. Res. Commun., 1965, 21, 582. 53 Fedkina, V. R., and Bronnikova, T. B., Biofizika, 1995, 40, 36. 54 Hauck, T., and Schneider, F. W., J. Phys. Chem., 1993, 97, 391. 55 Hauser, M. J. B., and Olsen, L. F., J. Chem. Soc., Faraday Trans., 1996, 92, 2857. 56 Olsen, L. F., and Degn, H., Nature (London), 1977, 267, 177. 57 Olsen, L. F., Z. Naturforsch., Teil A, 1979, 34, 1544. 58 Olsen, L. F., Phys. Lett. A, 1983, 94, 454. 59 Geest, T., Steinmetz, C. G., Larter, R., and Olsen, L. F., J. Phys. Chem., 1992, 96, 5678. 60 Bronnikova, T. V., Fedkina, V. R., Schaffer, W. M., and Olsen, L. F., J. Phys. Chem., 1995, 99, 9309. 61 Olson, D. L., and Scheeline, A., Anal. Chim. Acta, 1990, 237, 381. 62 Valeur, K. R., and Olsen, L. F., Biochim. Biophys. Acta, 1996, 1289, 377. 63 Kummer, U., Valeur, K. R., Baier, G., Wegmann, K., and Olsen, L. F., Biochim. Biophys. Acta, 1996, 1289, 397. 64 Liang, Y., and Yu, R., Gaodeng Xuexiao Huaxue Xuebao, 1988, 9, 881. 65 Sekimoto, T., Hirayama, K., and Unohara, N., Nihon Daigaku Kogakubu Kiyo Bunrui A, 1990, 31, 105. 66 Briggs, T. S., and Rauscher, W. C., J. Chem. Educ., 1973, 50, 496. 67 Jiang, M., Li, Y., Zhou, X., Zhao, Z., Wang, H., and Mo, J., Anal. Chim. Acta, 1990, 236, 411. 68 Zhang, Q., and Chen, J., Fenxi Shiyanshi, 1988, 7, 4. 69 An, C., Fang, X., Liu, Y., Liu, Z., and Cai, R., Wuhan Daxue Xuebao, Ziran Hexuebau, 1993, 4, 56. 70 Wolf, A., Swift, J. B., Swinney, H. L., and Vastano, J. A., Physica D, 1985, 16, 285. 71 Yatsimirskii, K. B., Strizhak, P. E., Ivashchenko, T. S., and Didenko, O. Z., Quim. Anal. (Barcelona), 1996, 15, 292. 72 Jim�enez-Prieto, R., Silva, M., and P�erez-Bendito, D., Anal. Chim. Acta, 1996, 321, 53. 73 Jim�enez-Prieto, R., Silva, M., and P�erez-Bendito, D., Analyst, 1996, 121, 563. 74 Jim�enez-Prieto, R., Silva, M., and P�erez-Bendito, D., Anal. Chim. Acta, 1996, 334, 323. 75 Jim�enez-Prieto, R., Silva, M., and P�erez-Bendito, D., Talanta, 1997, 44, 1463. 76 Jim�enez-Prieto, R., Silva, M., and P�erez-Bendito, D., Analyst, 1997, 122, 287. Paper 7/03354K Received May 15, 1997 Accepted September
ISSN:0003-2654
DOI:10.1039/a703354k
出版商:RSC
年代:1998
数据来源: RSC
|
2. |
Critical Review. High-performance liquid chromatography of nitrated polycyclic aromatic hydrocarbons |
|
Analyst,
Volume 123,
Issue 2,
1998,
Page 9-18
Josef Cvačka,
Preview
|
PDF (106KB)
|
|
摘要:
Critical Review High-performance liquid chromatography of nitrated polycyclic aromatic hydrocarbons Josef Cva�ckaa, Ji�r�ý Barek*a, Arnold G. Foggb, Josino C. Moreirac and Ji�r�ý Zimaa aUNESCO Laboratory of Environmental Electrochemistry, Department of Analytical Chemistry, Charles University, Hlavova 2030, 128 43 Prague 2, Czech Republic b Department of Chemistry, Loughborough University, Loughborough, Leicestershire, UK LE11 3TU c CESTEH/ENSP/FIOCRUZ, Rua Leopoldo Bulhoes 1480, Manguinhos, 21041-210 Rio de Janeiro, Brazil Summary of Contents Introduction Sample preparation Chromatographic systems, stationary and mobile phases Detection techniques Spectrophotometric detection Electrochemical detection Fluorescence detection Chemiluminescence detection Mass spectrometric detection Standards and reference materials Appendix Abbreviations References Keywords: Review; high-performance liquid chromatography; nitrated polycyclic aromatic hydrocarbons Introduction It is nearly two decades since Jager1 and Pitts et al.2 discovered independently that polycyclic aromatic hydrocarbons (PAHs; a list of abbreviations is given at the end of the paper) can undergo atmospheric reactions with nitrogen oxides to form nitro derivatives.Nitrated PAHs (NPAHs) are also directly emitted by diesel and petrol engines, and therefore their concentrations are raised in cities with heavy traffic and are further increased during smog episodes. So far, these compounds have been found also in carbon black and photocopier toners, fly ash, exhaust emissions from waste incineration plants, products of coal combustion, natural and waste waters, sediments, cigarette smoke and some foodstuffs (see reviews3–6).Aromatic systems of NPAHs found in various matrices have typically from two to five rings and one, two or three nitro groups. Samples contain a number of isomers, but the abundances of individual compounds in various types of sample are different. In the air the most volatile NPAHs, such as nitrobiphenyls and nitronaphthalenes occur predominantly in the gas phase, whereas ‘heavier’ isomers (nitro derivatives of pyrene, fluoranthene, anthracene, chrysene and others) are associated mainly with the particulate phase.In complex vehicle exhaust samples, nitropyrenes, nitroanthracenes, nitrophenanthrenes, nitrofluoranthenes, nitrofluorenes, nitronaphthalenes and others are found. The common concentrations of NPAHs in the atmosphere are at pg m23 levels and diesel exhaust particles contain ng g21 levels of NPAH isomers.The concentrations of dinitro- PAHs are approximately 10 times lower in both types of sample. Extensive efforts have been made to model NPAH formation and decay in the atmosphere.7–10 NPAHs are of particular interest because of their genotoxicity. Early investigators11 pointed out that extracts of diesel and air particulate matter exhibit strong direct mutagenicity when tested in the Ames Salmonella typhimurium assay.Nonsubstituted PAHs, which are abundant in such samples, are mutagenic only after metabolic activation. It was shown that the main portion of direct-acting mutagenicity of diesel and air particulates is associated with NPAHs.2,11,12 Some NPAHs, such as the dinitropyrenes, are among the most potent mutagens ever tested.13,14 According to the International Agency for Research on Cancer,15 some NPAHs are possibly carcinogenic to humans.Biological properties of NPAHs are discussed in detail elsewhere.13,16,17 Various methods for the determination of NPAHs have been reviewed by Moreira and co-workers.3,6 Owing to the complexity of environmental samples and to the need to distinguish among isomers of NPAHs with different biological properties, chromatographic techniques have mainly been used. Gas chromatography (GC) and high-performance liquid chromatography (HPLC) have been employed in more than 90% of all analyses.6 The main advantage of GC is the higher separation efficiency, which allows the separation of a larger number of compounds.Several hundred thousand theoretical plates can be generated with capillary GC compared with HPLC (up to Josef Cva�cka obtained his BSc degree in chemistry in 1993 at the Faculty of Science, Charles University, Prague, Czech Republic. He then graduated in 1995 with an MSc degree in analytical chemistry from the Department of Analytical Chemistry, Charles University. His diploma thesis dealt with the optimisation of HPLC separation of azodyes. At present he is a PhD student in the same Department.He is interested in the development of new HPLC methods for the determination of biologically-active organic substances at trace concentration levels. His thesis will deal with the HPLC determination of nitrated polycyclic aromatic hydrocarbons in waters using chemiluminescence detection. Analyst, February 1998, Vol. 123 (9R–18R) 9R20 000 theoretical plates using conventional equipment).18 Moreover, the combination of GC with mass spectrometry (MS) provides an excellent and powerful tool for the identification of individual compounds. The limitations of the GC are connected with the low volatility and instability of some compounds. It was found that partial decomposition of NPAHs occurs not only in the injector,19 but also in the column20 and in the GC–MS interface.20 This effect made the identification and quantification of isomers difficult, especially at the low levels common in environmental samples.This problem is overcomed by using HPLC, usually carried out at room temperature. HPLC is able to separate both small, volatile molecules and large, unstable molecules, as illustrated by the example of the separation of 22 NPAHs (Fig. 1). Reliable results can be obtained with relatively simple, inexpensive LC equipment, so that HPLC is often employed in the determination of NPAHs.Chemiluminescence and partially fluorescence and electrochemical detection are comparable in terms of selectivity and sensitivity to GC detection techniques such as NPD or NICIMS and NIAPIMS. Moreover, sample clean-up is very simple in many HPLC analyses. The continuing use of LC methods for the determination of NPAHs is documented by the increase in the number of published papers on this topic from one in 1980 to nine in 1996. Different types of sample have been analysed using HPLC: atmospheric (34%) and vehicle exhaust (29%) samples, biological materials (16%), foodstuffs (7%), reaction mixtures (5%) and others (see Table 1).Sample preparation The preparation of samples depends not only on the nature of the sample but also on the HPLC detection technique which is going to be used. Generally, the more specific the detector, the simpler is the sample preparation. Therefore, the use of electrochemical, fluorescence or chemiluminescence detection allows the preparation of samples to be simplified compared with the use of a universal spectrophotometric detector.Atmospheric and diesel or gasoline exhaust samples are treated in a similar way. The first step, extraction of particulate materials on glass-fibre filters or of gaseous substances trapped on polyurethane foam, is performed by Soxhlet or ultrasonic extraction. Time-consuming Soxhlet extraction, usually performed for 12–24 h, is increasingly being replaced by ultrasonication for 15–30 min.The recoveries with ultrasonic extraction are in the range 90–95%.21 Dichloromethane is the most common solvent for Soxhlet extraction and benzene– ethanol (4 + 1)22 or toluene18 have also been used. Ultrasonication is performed with benzene–ethanol (3 + 1),23–26 dichloromethane27,28 or 40% dichloromethane in hexane.29 Supercritical fluid extraction can also efficiently extract NPAHs from complex matrices, but it is employed only rarely. The highest recoveries are obtained using pure CHClF2 or CO2 modified with 10% of toluene.30 The extract is filtered through a membrane filter and concentrated, usually in a rotary evaporator under reduced pressure.The remaining organic solvent is removed under a gentle stream of inert gas, usually nitrogen. There is the possibility of direct injection of the redissolved extract into the,31 but, owing to the complexity of such types of matrices, further clean-up or fractionation of the sample is usually necessary.Simple clean-up was applied to atmospheric and diesel exhaust samples24,25 prior to HPLC with chemiluminescence detection. The benzene–ethanol (3 + 1) extract was purified by extraction with 5% NaOH, 20% H2SO4 and water. After the benzene phase was evaporated to dryness, the residue was dissolved in ethanol and NPAHs were reduced by refluxing with NaHS. The reaction mixture was extracted with benzene. Several drops of a saturated solution of ascorbic acid were added and mixture was evaporated to dryness.The residue was dissolved in 0.2 ml acetonitrile containing ascorbic acid and injected into the HPLC system. Fractionation of the sample allows one to simplify the matrix and hence to remove interfering compounds. Schleibinger et al.21 compared different fractionation procedures for atmospheric particulate samples. It was found that relatively high recoveries are achieved using solid-phase extraction (87–95%), column chromatography (87–91%) or preparative HPLC (83–92%), whereas low values are obtained using TLC (55–60%).The most popular fractionation technique is solidphase extraction on silica gel Sep-Pak cartridges (Waters, Milford, MA, USA). The common preparation scheme is as follows. An extract dissolved in the minimum volume of dichloromethane is applied to the top of the cartridge. The cartridge is then eluted stepwise with 3 ml of hexane, 6 ml of dichloromethane and finally 3 ml of methanol. The hexane eluate contains aliphatic hydrocarbons and PAHs and this fraction is discarded.Dichloromethane elutes oxy- and nitro- PAHs and this fraction is further dried under nitrogen, the residue is dissolved in the mobile phase or methanol and an aliquot of the solution is injected on to the HPLC column. Methanol elutes more polar compounds from the sample. The recovery of such a preparation scheme for NPAHs was reported to be 95%.32 A similar fractionation procedure involves a Bond Elut silica (Analytical International, Harbor City, CA, USA) cartridge.33 Elution is performed with 6 ml of cyclohexane followed by 6 ml of dichloromethane and an average recovery of about 85% for NPAHs was achieved.Different organic solvents are used for fractionation on an alumina Sep-Pak column.23 This column is eluted with benzene (discarded), followed by trichloromethane. The trichloromethane fraction is evaporated to dryness, the residue is dissolved in methanol and an aliquot of the solution is injected into the HPLC system.TLC fractionation on silica gel has been used in sample preparation for dinitropyrene analysis.34 After the TLC separation with toluene–hexane (5 + 1), the plates were dried and appropriate regions were extracted with methanol. Fractionation can also be performed using HPLC.29 The extract purified on silica gel is separated on amino-modified silica gel (mBondapak-NH2, Waters) with 10% dichloromethane in pentane.The fraction eluting between 40 and 70 ml is collected and concentrated and the residue is dissolved in methanol. The sample preparation procedure can also involve chemical reduction of NPAHs to APAHs, when fluorescence or chemiluminescence detection is used. Reduction is performed Fig. 1 An example of the separation of 22 NPAH standards. Peaks: 3 = 6-NQ; 4 = 5-NQ; 7 = 5-N-6-MQ; 8 = 8-N-7-MQ; 9 = 1,8-DNN; 12 = 1,5-DNN; 14 = 1-NN; 20 = 2-NN; 23 = 2-NB; 24 = 1-N-2-MN; 25 = 2,7-DNF; 27 = 4-NB; 28 = 3-NB; 31 = 2-NF; 33 = 9-NPH; 34 = 9-NA; 35 = 3-NPH; 37 = 1,6-DNP; 38 = 1-NP; 42 = 4-N-p-T; 43 = 6-NC; 44 = 3-NPer; 45 = 6-NBaP.Separation conditions: Alltech ODS column (250 32.1 mm id); 31 min gradient from 24 to 80% acetonitrile in water at 35 °C; spectrophotometric detection at 254 nm. Reprinted from ref. 54 with courtesy of Elsevier Science. 10R Analyst, February 1998, Vol. 123Table 1 Selected HPLC methods for the determination of NPAHs.AS, air and air particulate sample; CB, carbon black; RM, reaction mixture; VES, vehicle exhaust sample; DCM, dichloromethane; PBA, pyrenebutyric acid amide on LiChrosorb Si 100 Compounds Matrix Sample Preparation Stationary phase Mobile phase Detection Ref. 1-NN, 2-NN, 1,5-DNN, 1,8-DNN RM – ODS LiChrosorb, 250 3 4 mm id, 5 mm MeOH–water (75 + 25) Spectrophotometric at 254 nm 63 2-NB, 3-NB, 4-NB, 1-NN, 2-NN, 2-NF, 9-NA, 1,3-DNP, 1,3,6-TNP, 2,7-DNF and others VES Soxhlet extraction with DCM, filtration, evaporation, silica gel fractionation, drying, dissolution in mobile phase ODS Ultrasphere, 250 3 4 mm id, 5 mm 5% propanol in 0.05 m monochloroacetic acid– sodium acetate buffer (pH 3.8), 50 °C Electrochemical, TLC flow cell with glassy carbon working electrode at 20.6 V vs.Ag/AgCl 23 2-NN, 9-NA, 1-NP AS Ultrasonic extraction with DCM, evaporation, drying, dissolution in DCM, silica gel fractionation, evaporation, dissolution in MeOH ODS Nucleosil, 250 3 4 mm id, 10 mm MeCN–MeOH–acetic acid–sodium acetate buffer (pH 4.0) (3 + 4 + 3) Electrochemical, wall-jet flow cell with glassy carbon working electrode at 20.65 V vs.Ag/AgCl 28 1-NN, 2-NF, 4-NB, 1-NP, 2,7-DNF, 1,3-DNP and others VES Ultrasonic extraction with DCM, filtration, evaporation, drying, silica gel fractionation, drying, dissolution in MF Two ODS Alltech, 500 3 1 mm id, 10 mm Monochloroacetic acid– sodium acetate buffer (pH 4.7–MeCN (3 + 7) Electrochemical, TLC flow cell with glassy carbon working electrode at 20.6 V vs.Ag/AgCl 27 1-NP, 6-NBaP AS, VES Soxhlet extraction with benzene–EtOH (8 + 2), evaporation, drying, dissolution, reduction with NaBH4–CuCl2, extraction with benzene, evaporation, dissolution in MeCN ODS Zorbax, 250 3 4.6 mm id MeCN–water (65 + 35) buffered with NH3 or Bu4NPO4 Fluorescence after off-line reduction; 365/430 nm (1-NP); 430/497 nm (6-NBaP) 22 1-NN, 2-NN, 2-NF, 9-NA, 1-NP and others VES Soxhlet extraction with DCM, evaporation, dissolution in MeOH– H2SO4, dilution with MeOH ODS Spherisorb, 150 3 6 mm id, 3 mm MeOH–1 mm H2SO4 (85 + 15) Fluorescence after on-line electrochemical reduction 31 2-NF, 9-NA, 1-NP, 1,3-DNP, 7-NBaA, 6-NC AS, VES Soxhlet extraction with DCM, evaporation, fractionation on silica gel, evaporation, dissolution, fractionation on –NH2 silica gel, evaporation, dissolution in MeOH ODS Zorbax, 250 3 4.6 mm id MeOH–water gradient Fluorescence after on-line reduction on column packed with Zn–silica 29 1-NP VES Reflux with toluene, filtration, evaporation, dimethyl sulfoxide– cyclohexane separation, fractionation on silica gel column, evaporation, dissolution in MeOH Column switching system with PBA (250 3 4 mm id, 7 mm) and ODS (Seibersdorf) (250 3 4 mm id, 5 mm) MeOH–water Fluorescence after on-line reduction on a column packed with Rh–Pt catalyst on alumina (354/433 nm) 60 1-NP, 1,3-DNP, 1,6-DNP, 1,8-DNP, BaP AS Ultrasonic extraction with benzene–ethanol (3 + 1), filtration, extraction with 5% NaOH, 20% H2SO4, water, evaporation, dissolution in MeOH, reduction with NaHS, extraction with benzene, evaporation, dissolution in MeCN Column switching system with two ODS columns for simultaneous determination BaP and NPAHs Imidazole–perchloric acid buffer (pH 7.6)–MeCN (1 + 1) for NPAHs, water– MeCN (1 + 3) for BaP TCPO–H2O2 chemiluminescence after off-line reduction 24 2-NFA, 1-NP, 2-NP, 4-NP, 6-NC AS Ultrasonic extraction with benzene–ethanol (3 + 1), filtration, evaporation, dissolution in MeCN Column switching system with two ODS separation columns (Cosmosil 5C18AR, 250 3 4.6 mm id, 5 mm and Cosmosil 5C18MS, 250 3 4.6 mm id, 5 mm), reduction column [Zn–glass beads (1 + 1)] and ODS concentration column (Cosmosil 5C18MS) (1) Imidazole–perchloric acid buffer (pH 6.7)–MeCN (1 + 4) (2) Imidazole–perchloric acid buffer (pH 7.6)–MeCN (1 + 1) (3) 10 mm ascorbic acid TCPO–H2O2 chemiluminescence after on-line reduction 26 continued over— Analyst, February 1998, Vol. 123 11Ron the crude extract22 in cases when no further fractionation is to be performed, or on one of the fractions.35 Sample preparation can also be incorporated into the chromatographic system. Tejada et al.36 injected directly an extract dissolved in 50% methanol in dichloroethane without further purification. They used a multi-column HPLC system with two ODS columns.On the first column, NPAHs were separated from other types of compounds and transferred to the second analytical column. The advantage of such sample treatment is the very high recovery. Tejada et al.36 reported a 102% recovery for 1-NP. An interesting approach to the fractionation of complex environmental samples is bioassay-directed fractionation.37–39 This is a potent technique for selecting bioactive compounds based on the Ames mutagenicity assay performed on individual fractions.Biologically active fractions are subjected to further analysis. Carbon black is a complex matrix containing NPAHs. Jin et al.40 compared dichloromethane, benzene, toluene and chlorobenzene as organic solvents for Soxhlet extraction of carbon black. Chlorobenzene was found to extract the highest mass of organic compounds. For spectrophotometric detection, 40 further fractionation of the extract on a silica gel cartridge is necessary. The cartridge is eluted successively with hexane, dichloromethane and methanol; the dichloromethane fraction is concentrated, dried, diluted in methanol and analysed by HPLC.When chemiluminescence detection is used,41 it is possible to inject the evaporated and redissolved toluene extract without further purification. Soil and sediment samples can be extracted with dichloromethane42,43 or benzene.44 Maggard et al.42 compared Soxhlet and blender extraction with dichloromethane for extraction of NPAHs from soils.It was found that the blender procedure gives a higher recovery ( > 90%) than Soxhlet extraction ( > 80%). The extract can be further purified using gel permeation chromatography.43 After evaporation, the residue is dissolved in acetonitrile and injected on to the HPLC column. There is very little information about the preparation of water samples. The papers available deal with degradation or stability study of nitropyrenes.44,45 The concentrations of NPAHs are relatively high and therefore simple sample preparation is performed.Water samples are extracted three times by simple liquid–liquid extraction with benzene–ethanol (4 + 1)45 or ethyl acetate.44 The extracts are dried and the residue is dissolved in a solvent and analysed by HPLC. Foodstuff samples are extracted ultrasonically with benzene– ethanol (4 + 1)46 or acetonitrile;47 Soxhlet extraction with acetonitrile47 is also possible. The extract is then cleaned using size-exclusion chromatography on an SX-3 column with cyclohexane–ethyl acetate (1 + 1)47 or fractionated into neutral, acidic and basic fractions.46 NPAHs present in the neutral fraction are reduced to APAHs and analysed by HPLC with fluorescence detection. Urine samples48 are purified using solid-phase extraction on octadecyl-modified silica gel Sep-Pak cartridges (Waters).The sample is applied to a cartridge pre-washed with methanol and water. After washing the cartridge with water, NPAHs are eluted with methanol.The recovery is about 89%. The recovery of the sample preparation procedure is considerably lower than 100%, and therefore an internal standard should be used for accurate quantification of NPAHs. Surprisingly, an internal standard has been used in only a few analyses. Murahashi et al.24 and Hayakawa et al.25 used 2-fluoro-7-nitrofluorene as an internal standard in analysis of air particulates for 1-NP, 1,3-DNP, 1,6-DNP and 1,8-DNP. As NPAHs are not photochemically stable, it is recommended to carry out all manipulations with sample and standards in the dark.Chromatographic systems, stationary and mobile phases Most analyses are performed in the reversed-phase mode, utilising octadecyl-modified silica as the stationary phase. Mobile phases consist of acetonitrile or methanol and water or buffer solution. Both isocratic and gradient elution are employed in NPAHs separations. Grosse-Rhode et al. described the chromatography of NPAHs on several chemically modified copolymer49 and anthryl-modified silica50 stationary phases.A cyano column with hexane–propan-2-ol mobile phases has also been used.39,51 Normal-phase chromatography is not common in NPAH analysis; these systems are employed to ensure better compatibility with the detection method used.38 Some studies of the retention of NPAHs in normal-phase chromatography have been reported.52,53 Greibrokk et al.53 found a silica column together with dichloromethane–hexane mobile phases to be very useful for separations of NPAH positional isomers.Pirkletype chiral stationary phases52 do not facilitate the separation of NPAHs. Retention indices54 and retention data49,53,55 have been published for different stationary and mobile phases. Fu et al.52 discussed some relationships between the structure of NPAHs and retention order. The advantages of computer assistance in new chromatographic method development were demonstrated using the DryLab G simulation program.54,56 Almost all separations are carried out using conventional columns of 4.0 or 4.6 mm id.Narrow-bore or microbore columns facilitate a several-fold decrease in mobile phase consumption and improvements in sample mass-detection sensitivity, but have been employed only rarely,27,54,55,57 so far, Table 1—continued Compounds Matrix Sample Preparation Stationary phase Mobile phase Detection Ref. 1-NP, 1,3-DNP, 1,6-DNP, 1,8-DNP AS Ultrasonic extraction with benzene–ethanol (3 + 1), filtration, extraction with 5% NaOH, 20% H2SO4, water, evaporation, dissolution in EtOH, reduction with NaHS, extraction with benzene, evaporation, dissolution in MeCN ODS Cosmosil, 250 3 4.6 mm id 10 mm imidazole buffer (pH 7.6)–MeCN (1 + 1) TCPO–H2O2 chemiluminescence after off-line reduction 25 1-NN, 2-NF, 9-NA, 1-NP, 6-NC, 3-NPer CB Soxhlet extraction with toluene, evaporation, dissolution in DCM ODS Zorbax, 250 3 4.6 mm id MeCN–TRIS hydrochloride buffer (pH 6.5) (78 + 22) TCPO–H2O2 chemiluminescence after on-line reduction on column packed with Zn and glass beads 41 12R Analyst, February 1998, Vol. 123Fluorescence 37% Spectrophotometry 26% Chemiluminescence 22% Amperometry 9% Mass Spectrometry 3% Other 3% in the LC of NPAHs, and then mostly for theoretical studies.54,55,57 A column of 10 mm id has also been used in a system including fraction collection for off-line MS detection. 38 Analyses of real samples are usually performed using relatively simple experimental arrangements of HPLC equipment.Column-switching techniques, however, are also employed. 18,26,58–61 These systems, which are sometimes complex, are constructed in different ways. The use of two columns with different properties18,60 allows the incorporation of a sample pre-treatment step in the HPLC analysis and thus saves time. A system consisting of an initial column containing a reducing agent for conversion of NPAHs to APAHs, with a second analytical column,58,60 provides another example of column switching.Column-switching systems can also be hybridized into two subsystems with different conditions for parallel NPAH and PAH analysis.59 Detection techniques As shown in Fig. 2, different detection techniques are used in the HPLC of NPAHs. In the analysis of real samples, sensitivity of detection is one of the most important factors. Fig. 3 depicts the detection limits of NPAHs for all techniques used in HPLC systems.It can be seen that the most sensitive technique appears to be chemiluminescence, with DLs between 100 fg and 10 pg. Fluorescence is also very sensitive, with DLs of 1–10 pg. Slightly higher DLs are obtained when on-line conversion of NPAHs to APAHs is less than 100%, e.g., when an electrochemical reductor is employed (see below). Amperometric detectors, operated in the reductive mode, provide DLs from tens to hundreds of picograms, depending on the experimental arrangement.The few references dealing with the LC–MS of NPAHs provide little information concerning sensitivity. Under negative ion chemical ionization many substances are detected at picogram levels, but some NPAHs are not detected even at microgram levels. Spectrophotometric detection Spectrophotometric detectors are almost universal, because they are simple, cheap and reliable. Therefore, they are used widely in LC. Unfortunately, their sensitivity is not high enough for the trace analysis of NPAHs.Nevertheless, spectrophotometric detectors are very convenient for use in cases where the demands on sensitivity are not so high. These detectors are useful for the testing of new columns and mobile phases49,53 and for evaluating the effects of mobile phase composition and temperature.55 Investigations of retention mechanisms62 and the relationships between structure and retention order52 are carried out spectrophotometrically.Spectrophotometric detection is also suitable for the analysis of reaction mixtures and for determination of reaction products. This detection technique has been used for the rapid determination of 1-NN, 2-NN, 1,5-DNN, 1,8-DNN and naphthalene63 and methyl derivatives of 4-NB64 in reaction mixtures. Microsomal incubation products of 6-NBaP have been separated and detected by spectrophotometry at 254 nm.65 The stability of 1-NP and 1,6-DNP in environmental water samples and soil suspensions was examined by HPLC measurements of their degradation products.45 Spectrophotometric detection was also used in a study of the biodegradation of 1-NP.44 A spectrophotometric detector has been used for the confirmation of the presence of nitropyrenes in carbon blacks and toners51 and for separation of 2-NF and other mutagenic compounds from sterilised soil.42 Diode-array detectors can provide additional information about substances by the on-line recording of UV/VIS spectra.This approach has been applied in the determination of 1-NP in diesel particulate extracts18 and in the analysis of carbon black.40,66 Carbon black samples were examined for nine NPAHs.40 Several peaks had retention times close to those of standard samples of these compounds, but only one (3-NFO) had a UV/VIS spectrum matching its standard. Electrochemical detection Electrochemical detection is based on electrochemical reaction of the determinand at the electrode surface and requires several conditions to be fulfilled.The mobile phase must be electrically conductive to support charge transfer between the mobile phase and electrode, and the mobile phases and samples must be free of oxygen when working at negative potentials. Oxygen reduction in the chromatographic system can give rise to high background current levels, which limits the useful working potential range. To avoid the presence of oxygen, stainless-steel tubing must be used throughout with no Teflon tubing, and mobile phases must be de-gassed.Several approaches have been used for mobile phase de-gassing: Galceran and Moyano28 de-gassed the mobile phase with helium for 2 h at 50 °C and helium pre-saturated with the deoxygenated mobile phase was also bubbled throughout the chromatographic process; Rappaport et al.32 continuously heated the mobile phase under nitrogen in a flask fitted with a reflux condenser. Also, a special oxygen scrubber column29 or a porous graphite guard cell, operated at a very low potential,31 can be inserted between the pump and the injector.Glassy carbon electrodes are sensitive to passivation by heavy metals, so that chemicals for mobile phase preparation must be of very high purity. Even when using very pure chemicals, repolishing of the electrode is recommended monthly27 or weekly.32 Electrochemical cells used for NPAH detection are of the thin-layer27,29,32 or wall-jet28 type, each with a three electrode system, i.e., working, auxiliary and reference electrodes.Fig. 2 Detection techniques used in the HPLC analysis of NPAHs. Fig. 3 Detection limits of HPLC detection techniques for NPAHs. Analyst, February 1998, Vol. 123 13RTypical detectors contain a glassy carbon27,28,32 or a gold/ mercury29 working electrode with an Ag/AgCl reference electrode, or a porous graphite working electrode67 with a palladium-based modified H2/H+ reference electrode.Almost all electrochemical detectors are operated in the constant reductive amperometric mode with the working electrodes typically in the range between 2500 and 2650 mV versus the reference electrode. Attempts to use electrochemical detectors in the differential-pulse mode were not very successful.29 Measurement of hydrodynamic voltammograms can help to confirm the analytes’ identity.28,29,32 Chromatographic separations with electrochemical detection are usually carried out in the isocratic mode, but measurement in the gradient mode is also possible.29 Nitro-substituted and oxygenated PAHs have been determined in atmospheric samples from Barcelona28 with DLs of 200–1600 pg for 2-NN, 9-NA and 1-NP.Ang et al.67 described the determination of nitro- and oxy-derivatives of PAHs in ambient air particulates in Singapore with DLs at subnanogram levels. Rappaport et al.32 looked for 16 NPAHs in diesel exhaust (2-ANFO, 7-NFCA, 2,7-DNFO, 3-N-9-F, 2,7-DNF, 1-NN, 2-NB, 2-NN, 3-NB, 4-NB, 2-NF, 1,3,6-TNP, 9-NA, 1,3-DNP, 1-NP, 4-NFA).Confirmation of the identity of these compounds was based on comparisons of hydrodynamic voltammograms. Using this approach, the presence of 1-NP was confirmed and its concentration quantified. The DLs were 10–100 pg for most substances. A further increase in sensitivity can be achieved by using microbore LC.27 Compared with a similar system employing a conventional column,32 sensitivities in the micro-HPLC system (0.91 ml thin-layer flow cell, two 500 31 mm id columns, injection valve and connections of minimum internal dead volume) were 3–7 times higher.Column efficiencies in this system for NPAHs varied between 26 000–30 000 theoretical plates. MacCrehan et al.29 compared amperometric, differential-pulse amperometric and fluorescence detection techniques for the analysis of air and diesel particulate matter samples; 1-NP was determined by amperometric detection with a DL of 60 pg.A large-area porous graphite working electrode31 not only allowed the determination of NPAHs in the reductive amperometric mode, but also the high degree of electrochemical conversion to APAHs can be used in subsequent fluorescence detection. A gold/mercury thin layer electrochemical cell,29 operated in the differential-pulse mode under gradient elution conditions, gave very poor results. Although several NPAHs were identified, there were experimental difficulties with coordinating the gradient-elution and the base-potential programs.Fluorescence detection Most trace HPLC analyses of NPAHs are accomplished using fluorescence detection. Because of the strong electron-withdrawing effect of the nitro group, NPAHs themselves are not fluorescent. Therefore, it is necessary to convert NPAHs into a fluorescent species, and this is done mostly by reduction of the nitro group into an amino group. APAHs are strongly fluorescent, so that the sensitivity of determination is very high.Table 2 gives the optimum excitation and emission wavelengths for 35 amino analogues of NPAHs. Two approaches are used with NPAH reduction. This reaction can be performed before an HPLC analysis (off-line methods) or directly in a chromatographic system (on-line methods). On-line methods require more complicated HPLC equipment, involving a reduction column29,33,36,47,54,56,60,68 or a large-area porous graphite electrode operated at a negative potential.23,31,68 Off-line methods are time-consuming, laborious and difficult to automate for routine analyses. Both on-line and off-line methods of reduction are almost equally employed in the HPLC analysis of NPAHs.Different chemical reagents have been used for off-line reduction. The reaction can be accomplished using (i) sodium tetrahydroborate with copper(ii) chloride22,34 or (ii) copper(i) chloride21 as a catalyst at room temperature for 3–16 h, (iii) aqueous sodium hydrosulfide24,69–71 for 1–1.5 h or (iv) zinc powder in hydrochloric acid.72,73 Other metal powders (cadmium, copper and platinum) have also been tested,68 but apart from zinc only cadmium–copper (1 + 1) has a high reduction efficiency.Kinouchi et al.46 used specific nitroreductase purified from Bacteroides fragilis for reduction of NPAHs. APAHs can be further derivatized with 6-aminoquinolyl-Nhydroxysuccinimidyl carbamate.43 Several reducing materials for the on-line reduction of NPAHs have been tested.Columns packed with zinc powder or mixtures of zinc powder with silica29,54 or glass beads68 have limited lifetimes of a few days. Zinc is consumed by the reaction and small voids are formed in the packing, so that the chromatographic efficiency decreases. Zinc columns operate at room temperature and reduction yields are typically greater than 99%. A certain amount of supporting electrolyte or buffer is necessary in the mobile phase to facilitate the reaction.29,41 Other types of reduction columns employed in NPAH analysis are based on catalytic reactions.The packing is not consumed during the reaction and these columns have a very long, almost indefinite, lifetime.36 Catalytic reduction columns perform best when they are maintained at elevated temperatures, typically at 60–80 °C.36 Noble metals on a solid support are the main catalysts reported. A ‘three-way’ catalyst,36 designed to reduce hydrocarbons in automobile exhaust emission, is active only in methanol–water mobile phase.No reduction was observed in acetonitrile–water mixtures. Pt–Rh on alumina36,60,74 was Table 2 Excitation and emission wavelengths for fluorescence detection of reduced NPAHs Compound lex/lem/nm 2-Nitrobiphenyl 227/39454 3-Nitrobiphenyl 232/39954 4-Nitrobiphenyl 285/36754 2,2A-Dinitrobiphenyl 228/37254 1-Nitronaphthalene 243/42936 1-Nitro-2-methylnaphthalene 244/41454 2-Nitronaphthalene 234/40336 1,3-Dinitronaphthalene 247/42054 1,5-Dinitronaphthalene 231/39036 1,8-Dinitronaphthalene 229/41754 1,3,6,8-Tetranitronaphthalene 264/41754 1-Nitrofluorene 285/370;36 290/36521 2-Nitrofluorene 285/37054 2,7-Dinitrofluorene 292/38754 2-Nitro-9-fluorenone 290/36554 3-Nitro-9-fluorenone 245/40754 2,7-Dinitro-9-fluorenone 232/38754 2-Nitroanthracene 260/49536 9-Nitroanthracene 263/50536 9-Methyl-10-nitroanthracene 267/53436 9,10-Dinitroanthracene 264/49554 9-Nitrophenanthrene 247/430;54 345/43021 1-Methyl-9-nitrophenanthrene 254/44036 1-Nitropyrene 360/430;36 345/43321 4-Methyl-3-nitropyrene 355/42036 1,3-Dinitropyrene 395/44536 1,6-Dinitropyrene 369/44236 1,8-Dinitropyrene 395/45436 1,3,6-Trinitropyrene 396/46036 1,3,6,8-Tetranitropyrene 397/46536 3-Nitrofluoranthene 300/530;36 295/51521 8-Nitrofluoranthene 300/55036 6-Nitrochrysene 273/437;36 345/43021 6-Nitrobenzo[a]pyrene 420/47536 3-Nitroperylene 227/54054 14R Analyst, February 1998, Vol. 123CI CI CI O C C O CI CI O O CI + H2O2 CI CI CI OH + C C O O O O + C C O O O O fluorophor 2 CO2 + fluorophor* 2 observed to be more active and to have a higher reductive capacity than the ‘three-way’ catalyst.The degree of conversion for 1-NP and other NPAHs approaches 100%.36 The use of a Pt catalyst supported on alumina has also been reported.33 Other catalyst materials were found to be inferior to these. When tested in methanol–water mobile phases, commercially available Rh on alumina also reduced 1-NP, but neither Pd on alumina nor Rh metal powder worked.36 Several systems23,31,68 utilize electrochemical reduction of NPAHs to amino derivatives.The extent of conversion using this technique is not very high and it depends on the flow rate of the mobile phase.31 Murayama and Dasgupta31 reported an approximately 5% reductive conversion, whereas, in another system,68 values of about 13–14% were found. The reductive column can be placed either before29,47,54 or after33,68,75 the analytical column.Electrochemical reduction is performed in the post-column arrangement.23,31,68 More sophisticated equipment employs a reductive column in a columnswitching system.36,60 Fluorescence detection has been employed in the analyses of various types of samples. Concentrations of 1-NP, 1,3-DNP, 1,6-DNP and 1,8-DNP were measured in atmospheric dust particles76 and in the air24 in Japan; 1-NP, 1,6-DNP and 1,8-DNP were detected in an ambient aerosol at an urban site and suburban site in Michigan.34 Levels of 1-NP, 2-NF and 3-NFA were quantified in suspended particulate matter in Berlin.21 Ten mononitro-PAHs were identified in atmospheric samples collected in Paris.74 Atmospheric samples were also analysed for 1-NP,35 1-NP and 3-NFA,69 1-NP and 6-NBaP.22 Hisamatsu et al.71 identified mutagenes formed by photochemical reaction of pyrene with nitrogen dioxide.The following substances were determined in vehicle engine exhausts: 1-NP, 6-NBaP and 9-NA in diesel particulate extract,33 1,3-DNP, 1,6-DNP and 1-NP in sooty emissions of cars,23 9-NA and other NPAHs in exhausts from diesel engines;31 1-NP in particulate emissions from vehicles.36 Concentrations of 1-NP, 2-NF, 9-NA, 7-NBaA and 6-NBaP were determined in NIST Standard Reference Material SRM1650.29 Fluorescence detection has also been used in analyses of sediments and soils (1-NP, 9-NA, 2-NN, 6-NC, 6-NBaP)43 and foodstuffs (1-NP).46,47 1-NP and its metabolites were determined in biological samples68 and amounts of 1-NP were also measured in leaves from trees growing under various traffic conditions.70 Chemiluminescence detection At present, TCPO–H2O2 chemiluminescence is the most sensitive HPLC detection technique in NPAH analysis.The DLs are typically 1–2 orders of magnitude lower than those which can be achieved in the same chromatographic system using fluorescence detection. Moreover, the high degree of selectivity allows the direct determination of NPAHs in complex matrices with minimum sample preparation.The exceptional selectivity of chemiluminescence detection is documented by the chromatogram in Fig. 4, obtained by direct injection of a crude extract from airborne particulates after offline reduction. This detection approach is based on reaction between corresponding amino derivatives of NPAHs (fluorophore) with a mixture of bis(2,4,6-trichlorophenyl)oxalate (TCPO) and hydrogen peroxide,77 as shown in the reaction scheme.The high sensitivity of detection is due to a combination of high efficiences in the excitation step and high chemiluminescence quantum yields. As no light source is used, a low background is obtained. A basic catalyst in the mobile phase is necessary to accomplish the reaction. Several basic buffers have been tested.23,41 Tris(hydroxymethyl)aminomethane hydrochloride41 or imidazole–perchloric acid23–26,59,78,79 buffers appeared to be the most suitable for this purpose.Acetonitrile is a typical organic modifier for this type of analysis. The conversion of NPAHs into APAHs is performed as in fluorescence analysis. Off-line reduction with NaHS,24,25,78–81 an on-line catalyst58 or Zn–glass bead26,41 column and electrochemical reduction23 at glassy carbon working electrode have been used. The chemiluminescence radiation can be detected by commercially available chemiluminescence detectors23–26,78 or by fluorescence detectors with the light source turned off.41,58 Various samples have been analysed for NPAHs by means of chemiluminescence.Diurnal concentrations of 1,3-DNP, 1,6-DNP, 1,8-DNP, 1-NP and benzo[a]pyrene were measured in the air of Kanazawa city.24,25,80 Murahashi et al.59 published a method for the simultaneous determination of NPAHs and PAHs in airborne particulates with column switching equipment. Another column switching system designed by Murahashi et al.26 is suitable for the sensitive determination of 1-NP, 2-NP, 4-NP, 2-NFA and 6-NC in airborne particulates.Concentrations of nitrated pyrenes and their derivatives were measured in sooty emission from cars;23 mono- and dinitro- PAHs were determined in diesel exhaust particle extracts.58 Dinitropyrenes and nitropyrenes were determined in emission particulates from vehicles with diesel and gasoline engines.78,79 NPAHs were also detected in carbon black.41 A different type of chemiluminescence detector, originally designed for a GC system, has been evaluated in conjunction with HPLC.Details of this NO·–O3 gas-phase detector57,82 and the interface83 are described elsewhere. Unfortunately, this Fig. 4 Chromatogram of benzene–ethanol extract from airborne particulates after off-line reduction. Peaks: 1 = 1,6-DNP; 2 = 1,8-DNP; 3 = 1,3-DNP; 4 = 2-fluoronitrofluorene (internal standard); 5 = 1-NP. Separation conditions: Cosmosil ODS column 25034.6 mm id); 10 mm imidazole buffer (pH 7.6) – acetonitrile (1 + 1 v/v) mobile phase; TCPO chemiluminescence detection. Reprinted from ref. 25 with courtesy of the American Chemical Society. Analyst, February 1998, Vol. 123 15Rdetector is much less sensitive (10–30 ng in ref. 57, 50 pg in ref. 54) than the TCPO–H2O2 chemiluminescence detector. Together with the commercial unavailability of the detector, this is probably the main reason why no practical applications have been published. Mass spectrometric detection Analysis using mass spectrometry is very convenient and provides considerable information about the nature of the analyte.Although some LC–MS instruments are on the market, this method is not yet well established. Moreover, the equipment is expensive so that only a few analytical procedures have been described so far. LC–MS with a nebulizer and a moving belt was used to characterize 2-NF, its metabolites and related compounds.48 The interface affects the separation only slightly.The equipment, operating in the electron ionization mode, gives acceptable mass spectra down to 65 ng of injected substance and the responses are linear in the range 0.2–1.2 mg. A particle beam interface was employed to identify hydrocarbons, PAHs and NPAHs in extracts of fossil fuel combustion emissions.84 The best results for NPAHs were obtained under negative ion chemical ionization. The DLs determined in the FIA–MS mode for 1,8-DNN, 9-NA, 3-NFO, 2,2A-DNB, 1-NP, 2,7-DNF, 2,7-DNFO were at the picogram level. Some compounds, such as 1-NN and 2-NN, were not detectable even at microgram levels.The calibration curves exhibited good linearity over a range of greater than two orders of magnitude. Mass spectrometry is also very useful off-line, when selected fractions from HPLC runs are analysed. Using this approach, more than 50 NPAHs have been tentatively identified in an extract of diesel exhaust particulates.38 Standards and reference materials Precise and accurate determinations of NPAHs require the use of appropriate standards and certified reference materials.The Institute for Reference Materials and Measurements (IRMM) prepared a series of eight nitropolyarenes (1-NP, 1-NN, 2-NN, 9-NA, 6-NC, 3-NFA, 6-NBaP, 2-NMNF) with a purity better then 99% as certified reference materials.85 J. T. Baker86 supply 23 mono-NPAHs and nine di- and tri-NPAHs as neat reference materials in 95–99% purity, Dr. Ehrenstorfer87 offers 42 mono-, di- and trinitro-PAHs (Dr.Ehrenstorfer Reference Materials). AccuStandard Inc. sell 17 mono-NPAHs and 12 di- and trinitro- PAHs as neat substances or solutions in toluene (100 mg ml21).88 Standard reference materials SRM 1587 (nitrated polycyclic aromatic hydrocarbons in methanol; 2-NF, 9-NA, 3-NFA, 1-NP, 7-NBaA, 6-NC, 6-NBaP), SRM1596 (dinitropyrene isomers and 1-nitropyrene in methylene chloride; 1-NP, 1,3-DNP, 1,6-DNP, 1,8-DNP) and SRM1650 (diesel particulate matter, with certified content of 1-NP) are available from NIST.89 A review90 dealing with the synthesis of nitropolyarenes is useful for cases where the requisite isomer is not commercially available.Appendix Abbreviations APAHs Amino-polycyclic aromatic hydrocarbons 2-ANFO 2-Acetamido-3-nitro-9-fluorenone DL Detection limit 2,2A-DNB 2,2A-Dinitrobiphenyl 2,7-DNF 2,7-Dinitrofluorene 2,7-DNFO 2,7-Dinitro-9-fluorenone 1,5-DNN 1,5-Dinitronaphthalene 1,8-DNN 1,8-Dinitronaphthalene 1,3-DNP 1,3-Dinitropyrene 1,6-DNP 1,6-Dinitropyrene 1,8-DNP 1,8-Dinitropyrene FIA Flow-injection analysis GC Gas chromatography HPLC High-performance liquid chromatography MS Mass spectrometry 9-NA 9-Nitroanthracene 2-NB 2-Nitrobiphenyl 3-NB 3-Nitrobiphenyl 4-NB 4-Nitrobiphenyl 7-NBaA 7-Nitrobenzo[a]anthracene 6-NBaP 6-Nitrobenzo[a]pyrene 6-NC 6-Nitrochrysene 2-NF 2-Nitrofluorene 3-N-9-F 3-Nitro-9-fluorene 2-NFA 2-Nitrofluoranthene 3-NFA 3-Nitrofluoranthene 4-NFA 4-Nitrofluoranthene 7-NFCA 7-Nitrofluorene-1-carboxylic acid 3-NFO 3-Nitro-9-fluorenone NICIMS Negative ion chemical ionization mass spectrometry NIAPIMS Negative ion atmospheric pressure ionization mass spectrometry 2-NMNF 2-Nitro-7-methoxynaphtho-(2,1-b)furan 5-N-6-MQ 5-Nitro-6-methylquinoline 8-N-7-MQ 8-Nitro-7-methylquinoline 1-NN 1-Nitronaphthalene 2-NN 2-Nitronaphthalene 4-N-p-T 4-Nitro-p-terphenyl 1-NP 1-Nitropyrene 2-NP 2-Nitropyrene 4-NP 4-Nitropyrene NPD Nitrogen–phosphorus detector 3-NPer 3-Nitroperylene 3-NPH 3-Nitrophenanthrene 9-NPH 9-Nitrophenanthrene 5-NQ 5-Nitroquinoline 6-NQ 6-Nitroquinoline NPAHs Nitrated polycyclic aromatic hydrocarbons PAHs Polycyclic aromatic hydrocarbons 1,3,6-TNP 1,3,6-Trinitropyrene TCPO Bis(2,4,6-trichlorophenyl) oxalate UV/VIS Ultraviolet/visible J.C., J.B.and J.Z. thank the University Development Fund of the Czech Ministry of Education for financial support (Grant No. 1230/1997). A.G.F. thanks the Department of Analytical Chemistry, Charles University, Prague for support as a visiting professor in 1994–96.References 1 Jager, J., J. Chromatogr., 1978, 152, 575. 2 Pitts, J. N. Jr., van Cauwenberghe, K. A., Grosjean, D., Schmid, J. P., Fitz, D. R., Belser, W. L., Jr., Knudson, G. B., Hynds, P. M., Science, 1978, 202, 515. 3 Moreira, J. C., and Barek, J., Qu�ým. Nova, 1995, 18, 362. 16R Analyst, February 1998, Vol. 1234 Jacob, J., Karcher, W., Belliardo, J. J., Dumler, R., and Boenke, A., Fresenius’ J.Anal. Chem., 1991, 340, 755. 5 Moreira, J. C., Kuriyama, G. S., and Barek, J., in Proceedings of the National Conference on Polycyclic Aromatic Hydrocarbons in General and Industrial Environment, Garga, 1996, p. 63. 6 Barek, J., Cva�cka, J., Moreira, J. C., and Zima, J., Chem. Listy, 1996, 90, 805. 7 Pitts, J. N., Jr., Atmos. Environ., 1987, 21, 2531. 8 Fan, Z., Chen, D., Birla, P., and Kamens, R. M., Atmos. Environ., 1995, 29, 1171. 9 Arey, J., Zielinska, B., Atkinson, R., Winer, A.M., Ramdahl, T., and Pitts, J. N., Jr., Atmos. Environ., 1986, 20, 2339. 10 Kamens, R. M., Zhi-Hua, F., Chen, D., Chen, S., and Vartiainen M., Chemosphere, 1994, 28, 1623. 11 Finlayson-Pitts, B. J., and Pitts, J. N., Jr., Science, 1997, 276, 1045, and references cited therein. 12 Schuetzle, D., and Lewtas, J., Anal. Chem., 1986, 58, 1060A. 13 Rosenkranz, H. S., and Mermelstein, R., Mutat. Res., 1983, 114, 217. 14 Rosenkranz, H. S., Mutat. Res., 1982, 101, 1. 15 IARC Monographs on the Evaluation of Carcinogenic Risk to Humans. Diesel and Gasoline Engine Exhausts and Some Nitroarenes, IARC Lyon, 1989. 16 Sigvardson, K. W., and Birks, J. W., Anal. Chem., 1983, 55, 432. 17 Delgado-Rodriguez, A., Ort�ýz-Martelo, R., Graf, U., Villalobos- Pietrini, R., and G�omez-Arroyo, S., Mutat. Res., 1995, 341, 235. 18 Linder, W., Posch, W., Wolfbeis, O. S., and Tritthart, P., Chromatographia, 1985, 20, 213. 19 Tong, H. Y., Sweetman, J.A., and Karasek, F. W., J. Chromatogr., 1983, 264, 231. 20 Sweetman, J. A., Karasek, F. W., and Schuetzle, D., J. Chromatogr., 1982, 247, 245. 21 Schleibinger, H., Leberl, C., and Ruden, H., Zentralbl. Bakteriol. Mikrobiol. Hyg., Ser. B, 1988, 187, 44. 22 Gibson, T. L., Atmos. Environ., 1982, 16, 2037. 23 Imaizumi, N., Hayakawa, K., Suzuki, Y., and Miyazaki, M., Biomed. Chromatogr., 1990, 4, 108. 24 Murahashi, T., Miyazaki, M., Kakizawa, R., Yamagishi, Y., Kitamura, M., and Hayakawa, K., Jpn.J. Toxicol. Environ. Health., 1995, 41, 328. 25 Hayakawa, K., Murahashi, T., Butoh, M., and Miyazaki, M., Environ. Sci. Technol., 1995, 29, 928. 26 Murahashi, T., and Hayakawa, K., Anal. Chim. Acta, 1997, 343, 251. 27 Jin, Z., and Rappaport, S. M., Anal. Chem., 1983, 55, 1778. 28 Galceran, M. T., and Moyano, E., Talanta, 1993, 40, 615. 29 MacCrehan, W. A., May, W. E., Yang, S. D., and Benner, B. A., Jr., Anal.1988, 60, 194. 30 Paschke, T., Hawthorne, S.B., Miller, D. J., and Wenclawiak, B., J. Chromatogr., 1992, 609, 333. 31 Murayama, M., and Dasgupta, P. K., Anal. Chem., 1996, 68, 1226. 32 Rappaport, S. M., Jin, Z. L., and Xu, X. B., J. Chromatogr., 1982, 240, 145. 33 Hartung, A., Kraft, J., Schulze, J., Kiess, H., and Lies, K.-H., Chromatographia, 1984, 19, 269. 34 Siak, J., Chan, T. L., Gibson, T. L., and Wolff, G. T., Atmos. Environ., 1985, 19, 369. 35 Xu, X. B., and Jin, Z. L., J. Chromatogr., 1984, 317, 545. 36 Tejada, B. S., Zweidinger, R. B., and Sigsby, J. E., Jr., Anal. Chem., 1986, 58, 1827. 37 Greenberg, A., Lwo, J.-H., Atherholt, T. B., Rosen, R., Hartman, T., Butler J., and Louis, J., Atmos. Environ., Part A., 1993, 27, 1609. 38 Xu, X. B., Nachtman, J. P., Jin, Z. L., Wei, E. T., and Rappaport, S. M., Anal. Chim. Acta, 1982, 136, 163. 39 Nakagawa, R., Kitamori, S., Horikawa, K., Nakashima, K., and Tokiwa, H., Mutat. Res., 1983, 124, 201. 40 Jin, Z., Dong, S., Xu, W., Li, Y., and Xu, X., J.Chromatogr., 1987, 386, 185. 41 Sigvardson, K. W., and Birks, J. W., J. Chromatogr., 1984, 316, 507. 42 Maggard, L. A., Brown, K. W., and Donnelly, K. C., Chemosphere, 1987, 16, 1243. 43 Ne�ca, J., and Machala, M., in Proceedings of International Conference TOCOEN’96, 1996, p. 249. 44 Heitkamp, M. A., Freeman, J. P., Miller, D. W., and Cerniglia, C. E., Arch. Microbiol., 1991, 156, 223. 45 Tahara, I., Kataoka, K., Kinouchi, T., and Ohnishi, Y., Mutat. Res., 1995, 343, 109. 46 Kinouchi, T., Tsutsui, H., and Ohnishi, Y., Mutat. Res., 1986, 171, 105. 47 Schlemitz, S., and Pfannhauser, W., Food Addit. Contam., 1996, 13, 969. 48 Moller, L., and Gustafsson, J.-A., Biomed. Environ. Mass Spectrom., 1986, 13, 681. 49 Grosse-Rhode, C., Kicinski, H. G., and Kettrup, A., Chromatographia, 1988, 26, 209. 50 Grosse-Rhode, C., Kicinski, H. G., and Kettrup, A., Chromatographia, 1990, 29, 489. 51 Rosenkranz, H. S., McCoy, E. C., Sanders, D. S., Butler, M., Kiriazides, D.K., and Mermelstein, R., Science, 1980, 209, 1039. 52 Fu, P. P., Zhang, Y., Yuh-Lin, M., Von Tungeln, L. S., Kim, Y., Jung, H., and Jun, M-L., J. Chromatogr., 1993, 642, 107. 53 Greibrokk, T., Iversen, B., Johansen, E. J., Ronningsen, H.-P., and Svendsen, H., J. High. Resolution Chromatogr. Commun., 1984, 7, 671. 54 Liu, T-Y, and Robbat, A., Jr., J. Chromatogr., 1991, 539, 1. 55 Robbat, A., Jr., and Liu, T.-Y., J. Chromatogr., 1990, 513, 117. 56 Thompson, D. J., and Ellenson, W.D., J. Chromatogr., 1989, 485, 607. 57 Robbat, A., Jr., Corso, N. P., and Liu, T-Y., Anal. Chem., 1988, 60, 173. 58 Li, H., and Westerholm, R., J. Chromatogr. A., 1994, 664, 177. 59 Murahashi, T., Hayakawa, K., Iwamoto, Y., and Miyazaki, M., Bunseki Kagaku., 1994, 43, 1017; Chem. Abstr., 1995, 122, 37513h. 60 Veigl, E., Posch, W., Lindner, W., and Tritthart, P., Chromatographia, 1994, 38, 199. 61 Murahashi, T., Miyazaki, M., Kakizawa, R., Yamagishi, Y., Kitamura, M., and Hayakawa, K., Jpn.J. Toxicol. Environ. Health, 1995, 41, 328; Chem., Abstr., 1996, 124, 14247b. 62 Lafleur, A. L., and Wornat, M. J., Anal. Chem., 1988, 60, 1096. 63 Hill, P., Newbery, J. E., and Parry Jones, R., J. High Resolut. Chromatogr. Chromatogr. Commun., 1983, 6, 625. 64 You, J. M., Sun, X. J., Zheng, G. X., and Lu, C. Y., Sepu, 1995, 13, 292. 65 Raha, C., Hart Anstey, M., and Bresnick, E., J. Liq. Chromatogr., 1986, 9, 2945. 66 Jin, Z., Dong, S., Li, Y., Xu, W., and Xu, X., Huanjing Huaxue, 1988, 7, 28; Chem..Abstr., 1988, 109, 103885r. 67 Ang, K. P., Tay, B. T., and Gunasingham, H., Int. J. Environ. Stud., 1987, 29, 163; Chem. Abstr., 1987, 107, 140008m. 68 Hayakawa, K., Terai, N., Suzuki, K., Dinning, P. G., Yamada, M., and Miyazaki, M., Biomed. Chromatogr., 1993, 7, 262. 69 Kamiura, T., Kawaraya, T., Tanaka, M., and Nakadoi, T., Anal. Chim. Acta, 1991, 254, 27. 70 Nakajima, D., Teshima, T., Ochiai, M., Tabata, M., Suzuki, M., and Suzuki, S., Bull. Environ. Contam. Toxicol., 1994, 53, 888. 71 Hisamatsu, Y., Nishimura, T., Tanabe, K., and Matsushita, H., Mutat. Res., 1986, 172, 19. 72 Imaizumi, N., Hayakawa, K., and Miyazaki, M., Eisei Kagaku, 1989, 35, P4. 73 Xu, X. B., and Jin, Z. L., J. Chromatogr., 1984, 317, 545. 74 Wortham, H. M., Masclet, P. A., and Mouvier, G., Analusis, 1990, 18, 536; Chem. Abstr., 1991, 114, 68151c. 75 Schuetzle, D., and Perez, J. M., J. Air Pollut. Control Assoc., 1983, 33, 751. 76 Saitoh, N., Koizumi, A., and Kamiyama, S., Iwate-ken Eisei Kenkyusho Nenpo, 1988, 31, 24; Chem. Abstr., 1990, 113, 35964w. 77 Sigvardson, K. W., Kennish, J. M., and Birks, J. W., Anal. Chem., 1984, 56, 1096. 78 Hayakawa, K., Butoh, M., and Miyazaki, M., Anal. Chim. Acta, 1992, 266, 251. 79 Hayakawa, K., Butoh, M., and Miyazaki, M., Jpn. J. Toxicol. Environ. Health, 1993, 39, P19. 80 Hayakawa, K., and Miyazaki, M., in Proceedings of the 8th International Symposium on Bioluminescence and Chemiluminescence, 1994, p. 72; Chem. Abstr., 1996, 124, 350937r. 81 Levsen, K., Puttins, U., Schilhabel, J., and Priess, B., Fresenius’ J. Anal. Chem., 1988, 330, 527. 82 Robbat, A., Corso, N. P., Doherty, P. J., and Wolf, M. H., Anal. Chem., 58, 1986, 2078. 83 Robbat, A., Jr., and Corso, N. P., US Pat., 4 801 430, 1989. Analyst, February 1998, Vol. 123 17R84 Bonfanti, L., Careri, M., Mangia, A., Manini, P., and Maspero, M., J.Chromatogr. A, 1996, 728, 359. 85 BCR Reference Materials, Institute for Reference Materials and Measurements, Retieseweg, 1996. 86 J. T. Baker Organic Standards Catalogue 95/96, J. T. Baker, Deventer, 1995. 87 Directory of Environmental Standards 1996. Reference Materials for Residue Analysis, Labor Dr. Ehrenstorfer-Sch�afers, Augsburg, 1996. 88 AccuStandard Chemical Reference Materials, AccuStandard Inc., New Haven, CT, USA, 1997. 89 Standard Reference Materials Catalog 1995–96, National Institute of Standards and Technology. Gaithersburg, MD, 1995. 90 Cho, B.P., Org. Presep. Proced. Int., 1995, 27, 243. Paper 7/05097F Received July 16, 1997 Accepted September 24, 1997 18R Analyst, February 1998, Vol.
ISSN:0003-2654
DOI:10.1039/a705097f
出版商:RSC
年代:1998
数据来源: RSC
|
3. |
Statistically-based performance characteristics in laboratory performance studies |
|
Analyst,
Volume 123,
Issue 2,
1998,
Page 167-172
Steffen Uhlig,
Preview
|
PDF (71KB)
|
|
摘要:
Statistically-based performance characteristics in laboratory performance studies Steffen Uhlig*a and Peter Lischerb a Free University Berlin, Institute of Statistics and Econometrics, D-14195 Berlin, Germany b Swiss Federal Research Station for Agroecology and Agriculture, Institute of Environmental Protection and Agriculture IUL, CH-3003 Bern, Switzerland The International Harmonized Protocol for the Proficiency Testing of (Chemical) Analytical Laboratories (Thomson, M., and Wood, R., J.AOAC Int., 1993, 76, 926) lays down the minimum requirements for proficiency testing schemes. A prerequisite of statistically-based laboratory-performance characteristics is a suitable statistical model. A detailed investigation of statistical modelling, laboratory performance characteristics and quality limits is presented in this paper. Keywords: z-score; sum of squared z-scores; quality limit; reproducibility standard deviation; proficiency test; collaborative trial; quality requirement; laboratory performance characteristics In addition to internal quality assurance, quality management in chemical analysis includes (1) validation of the analytical method, (2) certification of reference materials and (3) laboratory assessment.Interlaboratory tests constitute an essential instrument for implementing these three elements of quality management. Interlaboratory tests for the validation of an analytical method, also referred to as collaborative studies, focus on the quality level of the method under analysis.This level is quantified by the repeatability and reproducibility standard deviation and, where available, by the recovery rate. Procedures for the conduct and evaluation of collaborative studies are set out, e.g., in ISO 57251 and in the Swiss Food Manual.2 Interlaboratory tests for laboratory assessment, also referred to as laboratory performance studies, serve to ascertain quality variations between laboratories.Requirements for laboratory performance studies and various assessment schemes have been set out in the joint protocol of the AOAC, ISO and IUPAC.3 Such a scheme may be employed both for the certification and licencing of laboratories as well as for deployment planning within the framework of monitoring programmes. A prerequisite of statistically-based laboratory performance characteristics is a statistical model for the test results. A detailed investigation of modelling, assessment parameters and quality limits has been undertaken in a research project of the German Federal Institute for Consumer Health Protection and Veterinary Medicine on behalf of the German Federal Environment Agency.4 Some of the results of this project are summarized in this paper.Burden of proof Let us consider a case where a laboratory is not performing satisfactorily. The assessment of the laboratory in question may be undertaken through a statistical significance test.In accordance with the harmonized protocol,3 the underlying statistical hypotheses may basically be formulated as follows: H0 (null hypothesis): The laboratory meets the quality requirements. H1(alternative hypothesis): The laboratory does not meet the quality requirements. This arrangement of the hypotheses means that it first has to be proved that the laboratory in question is not performing correctly, i.e., the burden of proof lies with the organizer of the proficiency test and the laboratory must be given the benefit of the doubt.When a laboratory with non-satisfactory performance produces sharply deviating results, the null hypothesis may be discarded after analysing just two or three different test materials. However, if the laboratory concerned is on the borderline, that is, if the results produced are bad yet without extreme deviations, it would be much more difficult to classify this laboratory. In this case, the null hypothesis may be refuted only after a larger number of different test materials have been analysed.The conclusion we may first draw therefrom is that preference should always be given to a laboratory assessment based on as many test materials and as many laboratory performance studies—within a reasonable timespan—as possible. However, in cases where some laboratories had not participated in all the laboratory performance studies because they had, for instance, started on the analytical tests concerned only later, these laboratories are treated more ‘mildly’ than others which had participated from the beginning.This is clearly not a plausible option. The statistical hypothesis mentioned above draw attention to yet another aspect: the performance of laboratories for which the null hypothesis had been rejected may actually be rated unsatisfactory with an error probability of 5%. On the other hand, for laboratories for which the null hypothesis had not been rejected, and which consequently remained within the quality limits, there is no clear indication as to whether their performance is satisfactory.Yet, it is precisely this aspect that should be clarified within a well-directed programme of quality management. This can only be ensured if the above hypotheses are reversed: H0: The laboratory does not meet the quality requirements. H1: The laboratory meets the quality requirements. With this, the burden of proof is also duly reversed. If the null hypothesis is now rejected, it follows that the laboratory actually meets the quality requirements with a pre-determined error probability of, for instance, 5%.Further, this reversal also means that with increasing numbers of test materials analysed the assessment of a laboratory becomes progressively milder. In our opinion, this aspect is extremely important for the practical application of laboratory assessment schemes. Statistical model In collaborative studies conducted to examine an analytical method for repeatability and reproducibility, for the statistical modelling it is common to assume that the laboratories participating in the test had been randomly selected from a population and are, consequently, representative of this population.This is also the prerequisite for the repeatability and Analyst, February 1998, Vol. 123 (167–172) 167reproducibility standard deviation determined in the collaborative study being intrepreted as an estimate for the standard deviation in the underlying population and, consequently, as the parameter characterising the method examined.Whereas laboratories participating in collaborative studies ‘only’ serve as a random sample from which inferences can be made for the population of all laboratories, laboratory performance studies are undertaken specifically to examine the technical competence of individual laboratories. In this case, the primary objective is not to determine the quality level of the method, but the performance level of individual laboratories.If the laboratories reveal extremely divergent levels of competence, it is hardly meaningful to group them all under one population without drawing any distinctions. Rather, the laboratory shall be considered not as a ‘random’ factor in the statistical model but as a ‘fixed’ factor. Introducing a laboratory-specific relative variability gj p > 0 dependent on analyte p, in this paper we consider the statistical model Yj ip = mip + gj pej ip with j = 1, .. ., J the laboratory, i = 1, . . ., I the test material and p = 1, . . ., P the analyte. Further, mip is the expected value and ej ip is the random deviation of the laboratory j, with mean zero and variance s2 ip. The random deviations e1 ip, . . ., eJ ip are assumed to be independent random variables with identical distributions. gj p = 1 corresponds to an average analytical performance, higher values of gj p correspond to a lower analytical performance, smaller values to a higher performance. The product sipgj p is the reproducibility standard deviation associated with the performance level of the laboratory j.Therefore, this model enables statistically-based indicators to be constructed for the individual competence of the laboratories. In order to ensure a statistical treatment that is as simple as possible, systematic differences in recovery rates are not taken into consideration.Therefore, statistical evaluation should be preceded by an examination of systematic differences in recovery rates. Finally, it must be noted that the statistical prerequisite of representativity cannot be taken as met: because the results of a laboratory performance study may have substantial economic implications for the participating laboratories, it must be assumed that participation in laboratory performance studies hinges on individual considerations. Consequently, the reproducibility standard deviation determined in laboratory performance studies cannot unconditionally be taken as the estimated value of the underlying population of all the laboratories.Rather, it should only be regarded as the mean standard deviation of laboratories participating in the test. z-scores In the harmonized protocol,3 the so-called z-score forms the basis of all statistical analyses, with Z = Y 2 m s where Y is the measured value, m the best available estimate for the content and s the standard deviation. Z has an inherently simple interpretation.If Y is normally distributed with the content m and standard deviation s, then the probability of the absolute value of Z not exceeding the value 2 is about 95%. This suggests that fixing the value of 2 as a ‘quality limit’ for the zscore would be meaningful. Thus, the quality criterion is exactly met when the absolute value of Z does not exceed the value 2. Sometimes, values of 2.5 or 3 are also used as limits instead of 2.It must be pointed out that the z-score reacts sensitively to fluctuations in both the relative standard deviation and the mean value. For fixing the mean value m, both the certified content and an estimate computed on the basis of the test results of all the laboratories or a section of them may be used. The standard deviation can be determined on the basis of the Horwitz curve,5 s/m 3 100% = 2120.5 log10 m or with an estimate on the basis of the test results.Classical estimation methods for m and s based on the normal distribution model are not recommended because they assume that all laboratories have the same performance. The alternatives are robust methods of estimation for m and s in which outlier elimination may be dispensed with (see, e.g., refs. 2, 4 and 6–8). However, one should be careful in the choice of the method, since there is no generally accepted definition of ‘statistical robustness’; several ‘robust methods’ provide different results and not necessarily close results. The Q-estimator for the reproducibility standard deviation (see refs. 4 and 8–10) is particularly recommended, since this estimator is highly efficient and is able to handle a large number of ‘outliers’. Further, it must be noted that the use of normal z-scores for large relative standard deviations s/m is questionable because in this case one must assume a right-skewed distribution rather than a symmetrical one.In ref. 11 a method has been presented whereby this defect is rectified through a simple modification of the z-scores. Combination scores Introduction Generally, laboratory performance studies involve the examination of not just one but several parameters. Given this, the question before us is how z-scores calculated for each test material–parameter combination analysed, may be used to draw conclusions about the quality of the laboratory. What is required are so-called combination scores which compress the information contained in the scores into a single numerical value.Further, there is also the task of interpreting the results, framing suitable rules for decisions and fixing classifications. A series of combination scores have been described in the harmonized protocol mentioned above.3 At the same time the protocol also warns against an overinterpretation or misinterpretation of the results. Basically, the following main types of combination scores may be distinguished: 1.Variability scores measure the mean scatter of the test results of a laboratory. In the harmonized protocol for instance, the sum of squared z-scores is taken as the variability score. 2. Distortion scores measure systematic ‘upward’ or ‘downward’ distortions in the results of a laboratory. For instance, the harmonized protocol3 uses the rescaled sum of the z-scores, RSZ, as a distortion score. 3. Frequency scores denote the absolute or relative frequencies with which a prescribed quality criterion, relating to a single material–analyte combination in each case, is not met.In this paper we focus on variability scores. Sum of squared z-scores, SZZ It has been recommended in the harmonized protocol that the variability of a laboratory j should be ascertained through the sum of the squared z-scores SSZp j ip j i I Z = = Â ( )2 1 with Zj p representing the z-score of the laboratory j, given analyte p and test material i.SSZj p can be used as the testing parameter of a c2-test examining the laboratory variability. If the sum of squared scores exceeds the 95%-quantile of the c2- distribution with I degrees of freedom, it follows at the significance level of 5% that the concerned laboratory j reveals an excessive variance or is, in other words, an outlier laboratory. 168 Analyst, February 1998, Vol. 123It should be noted that the burden of proof in this method rests not with the laboratory but with the assessing agency.In particular, only those laboratories which deviate sharply from the others can be accurately identified with this method. It must also be pointed out that the sum of squared scores, SSZ, can only be used when there is no correlation between the z-scores; however, where the correlation coefficient between all the z-scores shows the same value r, the modified sum of squared z-scores L ( )SSZ p j p j i p j i p j i i I k k Z Z = - < å 1 2 1 2 1 2 1 2 r r ( ) @ @ suggested in ref. 7 may be used alternatively. Relative laboratory performance, RLP Both sums of squared scores (SSZ and L) have an inherent disadvantage in that the numerical values by themselves have no meaning since the statistical distribution depends on the number of test materials. This drawback can be rectified if instead of the sum of scores the square root of the mean value of single squared scores is considered. Thus, we have RLP SSZ p j p j I = / with RLP standing for relative laboratory performance, while I represents the number of test materials.RLPj p measures the relative variability of the laboratory j for the analyte p, and may be interpreted as the estimator for the parameter gj p. An average competence yields a value close to 1 while an above average performance competence gives rise to a relative variability below 1. RLPj p denotes that factor around which the scatter of the measurement values would be greater if all the laboratories were to show the same competence as laboratory j.The geometric interpretation of RLPj p also merits attention: let us first take up the case where only two test materials have been analysed. Here, the related z-scores Z21 and Z22 calculated for a specific analyte and a specific laboratory may be considered as the cartesian coordinates of a two-dimensional diagram. The distance from the origin (0,0) may be expressed as AZ21 + Z22 = 2ARLPj p, i.e., RLPj p can be characterised as the distance between the point (Z1, Z2) and the origin.Equivalent to this is the representation in a Youden plot. If the distance lies above the value 3, that is, if RLP j p > 3/A2, then the conclusion to be drawn therefrom according to Youden is that the laboratory concerned either shows an obvious systematic error or an insufficient degree of precision. An analogous interpretation is also arrived at on analysing three or more test materials: up to a constant factor, RLPj p always denotes the distance between the point represented by the z-scores and the origin.If a statistically validated decision is to be made as to whether a laboratory meets the given quality requirements, it is recommended that the following test problem be examined: H0 : gjp ! 2 and H1 : gj p < 2 This means that every laboratory should prove that its theoretical relative variability does not exceed twice the predetermined standard deviation s.With a pre-determined error probability a and under the further condition that the z-scores relating to different test materials are uncorrelated, the requisite proof shall be considered produced if it is established that RLPj p < 2Ac2 I,a/I Here, c2I ,a represents the a-quantile of the chi-squared distribution with I degrees of freedom. In Table 1, the values 2Ac2 I,0.05/I and 2Ac2I ,0.10/I required for testing the quality criterion are presented for I = 1, 2, .. ., 15, 20, 30, 100 and the significance levels a = 0.05 and 0.10. In order to prove statistically after analysing just three test materials, for instance, that the theoretical variability of the laboratory is lower than twice the value of the pre-determined standard deviation s, the coefficient of variability RLPj p must be smaller than 0.441 3 2 = 0.882 (for the significance level a = 0.10). However, if a larger number of test materials are analysed, then the requirements to be met by the coefficient of variability RLPj p are reduced.Thus, after the analysis of five test materials, RLPj p must be smaller than 1.136, after ten materials it must be smaller than 1.4 and after 15 materials smaller than 1.51. Hence, laboratories whose performance is relatively poor must have a large number of test materials analysed in order to prove that they have fulfilled the statistical quality of requirement. The result of the test is at the same time a classification of the laboratories into those which have been rated H1 and those for which H0 has been retained. Laboratories which have been classified under H1 meet the quality requirements set out in H1 with a statistical certainty of at least 90 or 95%.For laboratories for which H0 has been retained it is doubtful whether they meet the quality requirements, although this possibility cannot be entirely ruled out. These laboratories have not yet proved that they fulfil the quality requirement. Participation in further laboratory performance studies will show whether these laboratories actually meet the requirements.A reduction of the significance level 5% is desirable, although it would mean that a larger number of test materials would have to be analysed before the quality criterion can be taken as met. It goes without saying that instead of the quality limit 2 used here, a smaller or greater value may also be used. If the standard deviation s is fixed on the basis of the Horwitz curve, then the quality limit 2 would mean that the variability of the laboratory must be lower than twice the value of the Horwitz curve.This quality limit may be derived from the observation by Horwitz that the reproducibility standard deviation ascertained in the interlaboratory test is almost always below twice the value of the Horwitz curve.5 The probability of wrongly retaining the null hypothesis (when the theoretical quality limit is not exceeded) may be expressed by the power function p(g).p(g) depends on the number of test materials, the quality limit and the signficance level a, and it describes the probability of rating H1 a laboratory which has the theoretical variability g. For example, with ten Table 1 Critical values 2Ac2I ,0.05/I and 2Ac2I ,0.10/I required for testing the quality criterion H0 : gj p !2 against H1 : gj p < 2 I 2Ac2I ,0.05/I 2Ac2I ,0.10/I 1 0.126 0.252 2 0.452 0.650 3 0.684 0.882 4 0.844 1.032 5 0.958 1.136 6 1.044 1.212 7 1.132 1.272 8 1.168 1.320 9 1.218 1.362 10 1.256 1.400 11 1.290 1.424 12 1.320 1.450 13 1.346 1.472 14 1.370 1.492 15 1.392 1.510 20 1.474 1.578 30 1.570 1.658 100 1.766 1.816 Analyst, February 1998, Vol. 123 169test materials, the quality limit 2 and the significance level a = 10% we have p(1) = 0.967, p(1.5) = 0.44, p(2) = 0.10 ( = significance level) and p(3) = 0.022. Therefore, at g = 1 the laboratory risk, i.e., the probability of wrongly retaining H0, equals 1 2 p(1) = 0.033.It should be noted that in the underlying statistical test it is assumed that the theoretical relative variability of the laboratory is constant over all test materials. Hence, one should be careful in interpreting the results when the concentrations and the matrices of the test materials are different. Finally, it should be noted that the statistical test described in this section is equivalent to the calculation of the respective (1 2 2a) confidence interval for gj p; if, and only if, the upper endpoint of the confidence interval is below the quality limit will the laboratory be rated H1.This principle is applied later (see under Example). Relative standard deviation of the laboratory, RSDL One may pose the question as to what the relative reproducibility standard deviation determined in interlaboratory studies would be if all laboratories reveal the same quality standard as laboratory j.To estimate this relative standard deviation for test material i and analyte p we may use RSDL RLP ip j ip ip p j s m = � � 100 with mip representing the estimated content and sip the related standard deviation. This value is referred to as the relative standard deviation of the laboratory for laboratory j, given test material i and analyte p. If the test materials analysed have a similar matrix and similar content of the analyte, it may be assumed that the theoretical (true) relative standard deviation sip/mip is not dependent on the material j.Then, the empirical relative standard deviation sip/mip may be replaced by the relative standard deviation RSDp ip ip i I I s m = � = å 1 100 2 2 1 / % and we obtain an indicator not dependent on the test material RSDLj p = RLPj p 3 RSDp RSDLj p is obviously a measure of the variability of the laboratory which is dependent on (1) the relative competence of the individual laboratory vis-a-vis the rest of the laboratories collectively and (2) on the mean relative reproducibility standard deviation.RSDLj p is not only used to estimate the ‘true’ relative standard deviation of the laboratory, gj p(RSDp) of laboratory j, thereby being interpretable as an empirical relative standard deviation of the laboratory, but can also be used for laboratory classification. The following quality requirement makes the task of statistical analysis simple: the true relative standard deviation of the laboratory concerned should be smaller than 40% (or some other percentage), formally expressed as gj p 3 RSDp < 40%.For statistical proof that this quality requirement has been met, an appropriate statistical test must be employed to test H0 : gj p 3 RSDp !40% against H1 : gj p 3 RSDp < 40% An obvious choice is a test based on the test statistic RSDLj p: the null hypothesis is rejected at the significance level a if RSDLj p < 40% 3Ac2 I,a/I i.e., the laboratory has proved with the error probability a that it meets the statistical quality requirement.In order to prove statistically after analysing just three test materials that the theoretical laboratory relative standard deviation is below 40%, the mean laboratory relative standard deviation RSDLj p of the laboratory concerned must be smaller than 0.441 3 40% = 17.6%, provided that the significance level is 10%. However, with a larger number of test materials the requirements to be met by the laboratory relative standard deviation are reduced: thus, after five test material analyses, the RSDLj p must be smaller than 22.7%, after ten analyses smaller than 28% and after 15 analyses smaller than 30.2%.Another approach, which allows a simplified calculation of the RSDL, is presented in ref. 12. The main idea of this method is a variance-stablising transformation of the data followed by analysis of variance. The method can be applied if there is a functional relationship between the error variance and the expected value of the test results.The Horwitz function is an example of such a relationship. Laboratory performance relating to several analytes Particularly when only a few test materials are analysed, the statistical methods employed will naturally yield results of low statistical accuracy if the statisticlysis is related to just one analyte. It may be necessary to opt for combination scores not specific to single analytes, with which a higher degree of evidence of statistical results may be achieved—at the risk of a lower level of specificity.The relative laboratory performance covering several analytes may be defined as RLP RLP j p j p P P = = å 1 2 1 ( ) with all material–analyte combinations (i,p) being included therein. It would be meaningful to calculate such a combination score for homogeneous groups of analytes for which it may be assumed that the analytical competence of a laboratory does not depend on the analyte and level of concentration.In such cases RLPj can be interpreted as the estimate for the all-analytes relative variability of the laboratory j. If a statistically validated decision is to be made on the basis of the all-analytes laboratory variability RLPj as to whether a laboratory shows a variability which is lower than twice the prescribed standard deviation, then the following test problem should be examined: H0 : gj !2 against gj < 2 with gj the theoretical variability of the laboratory which is assumed to be independent of the analyte p, i.e., gj 1 = .. . = gj P = gj. This means that every laboratory must prove that its theoretical variability does not exceed twice the prescribed standard deviation s. If one can assume that the test results of various test materials are not correlated whereas the different analytes for the same test material are correlated with a constant correlation coefficient r, then the following approximate statistical test may be recommended.For all laboratories j with RLPj < 2 3A[1 + r2(P 2 1)]c2f ,a/(IP) where f = IP/[1 + r2(P 2 1)], a decision is taken in favour of the alternative H1. In other words in such a case the laboratory has proved with the error probability a that it fulfils the statistical requirement. While using Box’s approximation13 for the distribution of a linear combination of independent cdistributed random variables, this test derives from the assumption that the z-scores for gj = 1 have a standard-normal distribution.It must be noted that the assumption of having a constant correlation between all analytes is crucial to the statistical test 170 Analyst, February 1998, Vol. 123presented and should only be used if the individual correlation coefficients between the analytes are not significantly different. Example Within the framework of the nationwide Programme for Monitoring Food Contaminants, the Central Bureau for the Registration and Assessment of Environmental Chemicals in the German Federal Health Office conducted a total of eight laboratory performance studies with four test materials each during the period 1992–94 in order to examine the laboratories’ competence to analyse plant and animal foods.In the process, the heavy metals Cd, Hg and Pb were determined along with a series of organic substances. For 16 out of a total of 32 test materials, a determination of heavy metals was performed.On the whole 17 laboratories participated in the tests, although not every laboratory participated in every test. A statistical analysis was conducted as described under Relative Laboratory Performance and under Laboratory Performance Relating to Several Analytes, with the mean values m and the standard deviations s for the z-scores being calculated with the robust method described in ref. 2. Owing to lack of space, only a section of the results can be presented here.Fig. 1 shows the relative laboratory performance RLPj Pb ascertained for Pb. The boxes assigned to individual laboratories describe the asymmetric (2% + 10%) confidence interval for gj Pb in each case while the line across the middle corresponds to the relative laboratory performance RLPj Pb. For all those laboratories whose upper box-edge falls below the value 2 (corresponding to the upper broken line), the quality requirement described under z-scores, whereby the variability of the laboratory must fall below twice the prescribed standard deviation, is met with an error probability of 10%.In the case represented in Fig. 1, only laboratories 07, 12, 15 and 16 fail to meet the requirement. At the same time it is also possible to identify those laboratories which record a variability significantly higher than the standard deviation s (in keeping with the procedure suggested in ref. 3, with an error probability of 2%).The laboratories falling under this category are those whose lower box-edge falls above value 1: once again laboratories 12, 15 and 16. Particular note should be taken of laboratory 07. This laboratory meets the quality requirement, although the related variability RLP07 Pb is just as great as for laboratory 15. However, because laboratory 15 had analysed just nine test materials for Pb while laboratory 07 had altogether analysed 16 test materials, the confidence interval for laboratory 07 is considerably smaller. This results in the variability of laboratory 07 being significantly greater than 1, that is, if the method suggested in ref. 3 were to be used, laboratory 07 would be an outlier laboratory while laboratory 15 would lie exactly on the limit of significance. Viewed from this angle, laboratory 07 gets a negative rating only because it has participated more often in laboratory performance studies than laboratory 15. For the majority of laboratories, the confidence intervals for Cd, Pb and Hg do overlap.Hence, it may be assumed that the theoretical variability gj is independent of the analyte in each case, that is, gj Cd = gj Pb = gj Hg. The related confidence intervals are presented in Fig. 2. The correlation coefficient between the test results for Cd, Hg and Pb was found to be r = 0.18. Owing to the larger data set sampled, the confidence intervals for relative laboratory performance were found to have become shorter, with the result that, except for laboratories 11 and 10, the variability is significantly smaller than 2s.Concluding remarks For assessing laboratories on the basis of laboratory performance studies the relative laboratory performance RLP proves more flexible and easy to interpret than the usual sum of squared scores SSZ. On the basis of the statistical model described in this paper it is possible to check whether the variability of a laboratory falls significantly below a prescribed limit using simple means.With that, the burden of proof for quality assessment is shifted to the laboratories, so that in borderline cases it should be up to the laboratory itself to remove doubts about its analytical competence by participating in a larger number of laboratory performance studies. Provided that analyte-specific coefficients of variability do not differ significantly from each other, it is also possible to determine a variability covering all analytes, thereby arriving at assessment data which are on the one hand unspecific and relating to all analytes yet at the same time statistically more reliable.The performance measures presented in this paper can easily be calculated with the software programs, RING 4.2 and ProLab 97, compiled for the statistical analysis of collaborative studies and laboratory performance studies. (The address of the supplier may be obtained from the first author.) This work was funded by the German Federal Institute for Consumer Health Protection and Veterinary Medicine (BgVV) on behalf of the German Federal Environment Agency (UBA). S.U. thanks Professor Henschel (UBA) and Professor Arnold (BgVV) for many helpful disucssions. P.L. thanks the BgVV, where he was a visiting researcher during the work on this paper. Special thanks go to one of the referees for his Fig. 1 Relative laboratory performance for lead. Fig. 2 Relative laboratory performance for heavy metals. Analyst, February 1998, Vol. 123 171constructive comments and suggestions, which led to an improved version of the paper. References 1 ISO 5725-2, 1994, Accuracy (Trueness and Precision) of Test Measurements. 2 Schweizerisches Lebensmittelbuch, Kapitel 60, Statistik und Ringversuche, Eidgen�ossische Drucksachen- und Materialzentrale, Bern, 1989. 3 Thompson, , and Wood, R., J. AOAC Int., 1993, 76, 926. 4 Uhlig, S., Entwicklung und DV-m�assige Implementierung eines Programms zur Auswertung von analytischen Laborvergleichsuntersuchungen gem�ass international harmonisierter Protokolle, Research Report, BgVV, Berlin, 1995. 5 Horwitz, W., Pure Appl. Chem., 1982, 54, 67A, 54. 6 Analytical Methods Committee, Analyst, 1989, 114, 1699. 7 Lischer, P., in Robust Statistics, Data Analysis, and Computer Intensive Methods, ed. Rieder, H., Springer-Verlag, Berlin, 1996, pp. 251–264. 8 Uhlig, S., in: Industrial Statistics. Aims and Computational Aspects, ed. Kitsos, C. P., and Edler, L., Physica-Verlag, Heidelberg, 1997, pp. 65–73. 9 M�uller, Ch., and Uhlig, S., submitted for publication. 10 Uhlig, S., Habilitation Thesis, Freie Universit�at Berlin, Fachbereich Wirtschaftswissenschaft, 1997. 11 Uhlig, S., and Henschel, P., Fresenius’ J. Anal. Chem., 1997, 358, 761. 12 Lischer, P., and Uhlig, S., submitted for publication. 13 Box, G. E. P., Ann. Math. Stat., 1954, 290. Paper 7/05432G Received July 28, 1997 Accepted October 15, 1997 172 Analyst, February 1998,
ISSN:0003-2654
DOI:10.1039/a705432g
出版商:RSC
年代:1998
数据来源: RSC
|
4. |
Assessment of detection methods in trace analysis by means of a statistically based in-house validation concept |
|
Analyst,
Volume 123,
Issue 2,
1998,
Page 173-179
Bernd Jülicher,
Preview
|
PDF (129KB)
|
|
摘要:
Assessment of detection methods in trace analysis by means of a statistically based in-house validation concept Bernd J� ulicher*a, Petra Gowika and Steffen Uhligb a EU Reference Laboratory for Residues of Veterinary Drugs, Division of Animal Drug Registration, Residue Control and Feed Additives, Federal Institute for Health Protection of Consumers and Verterinary Medicine, Diedersdorfer Weg 1, D-12277 Berlin, Germany b Institute for Statistics and Econometrics, Free University Berlin, Garystrasse 21, D-14195 Berlin, Germany A matrix-considering in-house validation concept for analytical methods is presented which takes into account the uncertainty due to matrix- and time-induced deviations. It is based on a variance component model for univariate quantitative measurement data that can be adapted to both screening and confirmation methods and to both zero-tolerance and threshold decisions.The model allows the calculation of critical concentrations for given a-errors and the calculation of the corresponding power function to evaluate the performance of an analytical method. The model is applied to a real-life validation experiment.Keywords: Validation; calibration; variance component model; power function; uncertainty; trace analysis; residue analysis; limit of detection; critical concentration In view of the globalization of markets, the reliability of analytical methods used for regulatory purposes (e.g., food control, environmental monitoring) has become increasingly important.To protect consumers from intolerable health hazards and to prevent market distortions, the competent authorities have to guarantee the comparable, reliable control of established threshold levels. One approach to ensure the analytical quality is to recommend or even to require the use of precisely described methods as the Codex Alimentarius Commission1 and the US Food Safety and Inspection Service2 do, whereas another approach is to establish minimum quality criteria for the different analytical techniques as the European Union3 does in the field of veterinary drug residue control.A widely accepted prerequisite for the application or even recommendation of an analytical method is its validation.1,4–8 Although virtually all procedures for the validation of methods are based on the determination of specific performance characteristics, e.g., the limits of ‘detection and quantification’, 7–13 there is no commonly accepted validation procedure for their assessment.Even within the concepts that are based on the construction of a calibration curve,7,8,12 there is no consensus about the choice of calibration samples and the number of replicates. Most validation procedures include the performance of a collaborative study1,4,6–8,12–16 to evaluate the reproducibility of the method, which means that enormous efforts have to be made. Consequently, the availability of fully validated methods for regulatory purposes is limited.In order to gain flexibility and to reduce time and costs, an inhouse validation procedure for the processing of quantitative data is presented here that takes into account matrix- and timeinduced deviations based on a statistical variance component model. By applying this model to a real-life validation experiment, the practicability of the in-house validation procedure is demonstrated. The reliability of the examined method is evaluated by calculating the critical concentrations for given aand b-errors.Requirements for validation procedures in trace analysis A basic requirement for analytical methods used in trace analysis and especially in residue control is their ability to distinguish between upwards and downwards deviations from a threshold level in accordance with a defined error probability. Therefore, the method should show as low a variance as possible so that these decisions can be made with the appropriate sharpness and security. At the beginning of the validation, preliminary examinations have to be carried out to determine the substances detectable by the method, the present and potential future matrix populations of the laboratory and the measurement range to which the validation will apply.Any change of the matrix resulting, for instance, from different origins of the samples or different freshness or storage conditions, could have an unexpected influence on the analysis and therefore reduce the evidential value of the results.Hence it has to be defined in advance under which conditions the method will be considered to be valid. Most validation procedures1,5 –11,13–16 assume that the distribution of the measurement values of the matrix of the unknown sample equals the distribution of the measurement values of the matrix of the calibration samples. Hence a huge amount of laboratory work and costs would be unavoidable if one of these procedures was to be repeated for each matrix being considered.These considerations suggest an investigation of the matrix variability, which so far has not been undertaken. Additional variability caused by time-related effects such as measurement series deviations due to instrument instabilities can occur. This additional variability must also be taken into account when performance characteristics are determined. Usually, attempts are made to compensate for this by the construction of a calibration curve within each measurement series and by the addition of an internal standard if available.However, even when internal standards are used, time effects may be observed. Therefore, the validity of a validation calibration curve should be guaranteed for a period to be defined by the validation procedure itself. Another crucial point in validation is the determination of the blank value. For several analytical techniques, e.g., mass spectrometry, it is problematic to determine.Establishing a virtual blank value by means of linear extrapolation of the calibration curve6–8 is questionable, because adsorption or contamination effects, especially in the region of the blank value, may result in a pronounced non-linear course of the calibration curve. It seems more reasonable to adapt the model to the purpose of analysis. Analyst, February 1998, Vol. 123 (173–179) 173Finally, the reliability of the analysis should be sufficient for court proceedings, i.e., the underlying concept should adequately reflect the true conditions.On the one hand, possible compensation claims resulting from false-positive decisions (aerror) have to be excluded to the greatest extent possible. On the other hand, consumer safety has to be guaranteed, i.e., virtually all contaminated samples should be discovered. The analytical strategy used so far to comply with these requirements is a combination of screening and confirmatory methods.17 One requirement for a successful screening method is the minimisation of the false negative rate (b-error); for a satisfactory confirmatory method, it is the minimisation of both the b- and the a-error.Consequently, the intended use of a method, e.g., screening or confirmation, influences the validation criteria. Definition of the maximum tolerated error probabilities for the analytical methods applied is a prerequisite for comparability of the analytical results.The different analytical communities, e.g., environmental monitoring, residue control, food control, pharmaceutical or forensic chemistry, should agree on error probabilities for common use in their respective fields. Statistical model For the validation of an analytical method, the determination of the method-specific parameters has to be based on a statistical model that reflects the existing analytical conditions, e.g., system instabilities, matrix variability and the distribution of measurement values.In the model presented (see also Appendix), it is assumed that the fundamental relationship between measurement Y, concentration xi and matrix j is given by Yij = m(xi) + �aj + �b j x + eij, (1) where x1 @ x2 @ ... @ xI represent the spike concentrations, i = 1, ..., I denote the spike levels and j = 1, ..., J denote the matrix of the calibration sample. Y may represent the measured raw (height or area of peaks) and also estimated concentration values obtained using a standard calibration curve established within the measurement series.m(x) is called the overall calibration curve. It is assumed to be linear within the calibration range, i.e., m(x) = a + bx within the interval [x1; xI]. The definition of the matrix levels j depends on the calibration experiment; j could, for instance, represent an animal or a species, or different parts of an animal such as muscle or liver. Additionally, each j could represent another measurement series.A clear definition of the different j levels is essential for the validation procedure. The measurement error is denoted by eij, which represents the variability of Yij when matrix j is fixed. The interpretation of eij is closely related to the experiment and to the definition of j. In either case, eij is a random variable with zero mean. Its variance can depend on concentration xi and on the level j, i.e., on compartments, species or instrument conditions.In the model presented, the variance of eij is assumed to be constant, Var(eij) = s0 2, �aj and �bj are random effects (possibly correlated) with zero mean and variances sa 2 and sb 2, respectively. Additionally, all random variables, eij, �aj and �bj, are assumed to have a normal distribution. If estimates �aj and �b j of the parameters aj = a + �aj and bj = b + �b j are available, the concentration x of the unknown sample with signal Y can be estimated from x = (Y 2 �aj)/�b j.However, the matrix-specific calibration curve normally fails for practical reasons such as the availability of sample material. Therefore, in the model presented, the overall calibration curve m(x) is used, where m(x) determines the relationship between x and Y on average over all matrices of the defined population. The contaminant concentration x of the unknown sample with signal y will be estimated using x = (y 2 �a)/�b , where �a and �b denote the estimated coefficients of the overall calibration curve.Calibration experiment The model was applied to a quantitative method used in veterinary drug residue control. In order to assess the influence of different operational conditions (freshness, storage conditions) and different matrices (muscles of calves, cows, pigs and turkeys from different feeding conditions with different fat contents), 26 calibration runs were performed, each at four concentration levels, x = 0.3, 0.6, 0.9 and 1.2 mg kg21 chloramphenicol (CAP), in muscle.The operating conditions and the matrices were chosen randomly. In the European Union the use of CAP in food-producing animals is banned and therefore the validation had to be carried out around the zero concentration range. The measuring results were obtained by means of a sample preparation consisting of several liquid- and solid-phase extraction steps followed by a derivatisation step to prepare the sample for GC–MS analysis.Deuterated chloramphenicol-d5 was used as internal standard. Complying with the prerequisite of an unambiguous identification of an analyte, a measuring result was only considered acceptable if the identification criteria for GC–MS had been fulfilled, e.g., the presence of four characteristic fragments (diagnostic ions) of the analyte within given margins at the correct retention time.3 Fragmentation took place in a negative chemical ionisation source. The reactant gas was ammonia.Quantitative evaluation was carried out by using the internal standard according to the peak-area mode customary in gas chromatography. For this purpose, the area ratio of the most intense ion of the analyte and of the internal standard, respectively, were used. Statistical analysis of the calibration experiment For each calibration run j the corresponding calibration function �aj + �b jx and the residual standard deviation sj were calculated.Table 1 gives the results. The measurement data and the calibration functions �aj + �b jx of the 26 calibration runs are presented graphically in Fig. 1. There was no indication that the measurement data were not a random sample from a normal population. The influence of different matrices and different operating conditions on the constant �aj and the slope �b j was examined by several t-tests (Fig. 2). No significant effect of different matrices or operating conditions was detected (details omitted).If there were no effects resulting from different matrices, operating conditions or calibration runs, the theoretical (unknown) calibration function of run j at concentration x would equal the theoretical overall calibration function: Yij = aj + bjxi + eij = a + bxi + eij where eij denotes the random error. Applying this restricted model, the (1 2 a) prediction interval for measurement values at concentration x could be calculated: � � ( , ) , a bx s t x M x + ± + Ê Ë Á � � � - - 0 52 1 1 1 1 26 1 1 a/2 where M x x x i i i = Ê Ë Á � � � = Ê Ë Á � � � 4 4 3 3 2 7 2 S S S .denotes the information matrix for (�aj, �b j); t52.1–a/2 denotes the 1 2 a/2-quantile of the t-distribution with 52 degrees of freedom (which are derived from the 52 degrees of freedom of s0 2). The resulting 98% prediction interval is shown in Fig. 3. Ten out of the 4 3 26 measurement values are not in the interval, which is far more than 2%. It can be concluded that 174 Analyst, February 1998, Vol. 123there are additional sources of error which model (1) does not take into account. For a more detailed analysis of the error, the scatter of the 26 calibration functions has to be investigated. These calibration functions can be written �aj + �b jx = aj + bjx + estimation error where �aj + �b jx denotes the observed (estimated) calibration function and aj + bjx denotes the unknown, theoretical calibration function at calibration run j.The variance of the observed calibration function is the sum of the variance of the theoretical calibration function and the variance of the estimation error, i.e., Var(�aj + �b jx) = Var(aj + bjx) + Var(estimation error) where j is assumed to be randomly chosen. The estimation error refers to the estimation of the calibration function at concentration x for one calibration run. Its variance can be computed as Var estimation error) = ( ( , ) s0 2 1 1 1 x M x - Ê Ë Á � � � where s20 denotes the variance of the error eij.Therefore, the variance of the theoretical calibration function aj + bjx can be estimated: Var( ) ( , ) � � a b x s s x M x j j a b x j j + = - æ è ç ö ø ÷ + - 2 0 2 1 1 1 where s2� aj + �b jx denotes the empirical variance of the estimated calibration functions and s20 = 1 26·s2j denotes the residual variance. Hence an overall calibration function can be estimated as m( ) � � � � x a bx a b x j j j = + = +  1 26 The respective variances were calculated at the limits, x = 0.3 mg kg21 and x = 1.2 mg kg21, and at the centre of the calibration range, x = 0.75 mg kg21.The results are given in Table 2. At concentration x = 0.3 mg kg21, the empirical variance s2� aj + �b jx is slightly lower than the variance of the estimation error. Since variances are non-negative, the variance of the calibration function at x = 0.3 mg kg21 can be estimated, Var(aj + bj30.3) Å 0, i.e., there is no indication that the calibration function at the lower limit of the calibration range is dependent on the run j.In contrast to this result, at concentration x = 0.75 mg kg21, the empirical variance is significantly higher than the variance of the estimation error, i.e., Var(aj + bj 3 0.75) > 0. This result is in accordance with the observed scatter of the calibration functions (Fig. 1). Apparently the dispersion becomes larger at higher concentrations. This is just an empirical result— analytical implications will not be discussed in this paper.Other analytical methods, matrices and substances may yield different results. Based on model (1), the (1 2 a) prediction interval for the measurement values can be taken into account. Then the (1 2 a) prediction interval for the measurement values can be computed as � � ( , ) ( ) , / a bx t s x M x a b x j j + ± + æ è ç ö ø ÷ + + - - 25 1 2 0 2 1 1 1 1 a Var The resulting 98% prediction interval is shown in Fig. 4.This interval covers the empirical distribution of the measurement values very well. This is a clear indication that the general model proposed in this paper is adequate to deal with matrixand time-induced deviations. Computation of the critical concentrations for banned substances Based on the assumption that the calibration function can be extrapolated linearly, the critical concentration CCa could be determined as illustrated in Fig. 5(a). As discussed above, this assumption is, however, questionable. For confirmatory purposes it may be more appropriate to assume a worst case scenario, as shown in Fig. 5(b). Table 1 Results of the calibration experiment Run No. 0.3 mg kg21 0.6 mg kg21 0.9 mg kg21 1.2 mg kg21 �aj ^bj sj 1 0.36 0.66 1.06 1.31 0.035 1.08 0.0433 2 0.28 0.62 0.90 1.24 20.030 1.05 0.0190 3 0.31 0.65 0.92 1.09 0.090 0.87 0.0603 4 0.31 0.61 0.88 1.12 0.055 0.90 0.0212 5 0.30 0.66 0.95 1.26 0.00 1.06 0.0227 6 0.35 0.72 1.05 1.27 0.075 1.03 0.0542 7 0.31 0.64 0.99 1.31 20.025 1.12 0.0087 8 0.32 0.67 0.87 1.26 0.025 1.01 0.0556 9 0.26 0.58 0.92 1.20 20.050 1.05 0.0190 10 0.31 0.64 0.84 1.19 0.035 0.95 0.0448 11 0.33 0.62 0.90 1.19 0.045 0.95 0.0032 12 0.33 0.69 0.86 1.21 0.070 0.94 0.0586 13 0.28 0.60 0.89 1.17 20.005 0.99 0.0145 14 0.32 0.67 0.93 1.17 0.070 0.94 0.0404 15 0.36 0.50 0.88 1.13 0.045 0.90 0.0702 16 0.35 0.69 1.03 1.30 0.045 1.06 0.0271 17 0.28 0.57 0.91 1.24 20.055 1.07 0.0170 18 0.32 0.58 0.87 1.14 0.040 0.92 0.0087 19 0.32 0.69 0.88 1.25 0.040 0.99 0.0569 20 0.32 0.70 1.02 1.19 0.075 0.98 0.0756 21 0.31 0.60 0.75 1.14 0.040 0.88 0.0697 22 0.35 0.66 1.10 1.27 0.045 1.07 0.0803 23 0.29 0.62 0.92 1.25 20.025 1.06 0.0095 24 0.31 0.73 0.91 1.30 0.025 1.05 0.0719 25 0.37 0.74 1.04 1.45 0.015 1.18 0.0318 26 0.29 0.56 0.89 1.28 20.070 1.10 0.0424 Analyst, February 1998, Vol. 123 175If the measured value exceeds the critical concentration, it will be concluded that the analyte is detected. The error probability a for making a wrong decision (false-positive rate) Fig. 1 Calibration functions of the calibration experiment.Fig. 2 Flow chart of the preliminary analysis of the data of the calibration experiment. Fig. 3 98% prediction interval (without matrix- and time-induced deviations). Table 2 Variance components of the estimated calibration functions Concentration x/ mg kg21 s2 �aj + �bjx Var(estimation error) Var(aj + bjx) 0.3 0.00119 0.00148 0 0.75 0.00236 0.00053 0.00188 1.2 0.00619 0.00148 0.00471 Fig. 4 98% prediction interval (with matrix- and time-induced deviations). Fig. 5 (a) Graphical determination of CCa using linear extrapolation. (b) Graphical determination of CCa for confirmatory purposes. 176 Analyst, February 1998, Vol. 123depends on the prediction level. For a 98% prediction interval the error probability is 1%.For the method examined, the critical concentration is CCa = 0.42 mg kg21, with an underlying error probability of a = 0.01. According to the model presented for confirmatory purposes (type II calibration curves, see Appendix), it is clear that the CCa cannot be smaller than the lowest concentration in the calibration experiment. This has to be considered when defining the calibration points during the preliminary examinations. Each analytical method involves not only a false-positive rate but also a false-negative rate which depends on the true concentration x of the analyte.The false-negative rate b = 1 2 p(x) can be calculated from the power function p(x), which describes the probability of detecting the analyte when its true concentration is x. It characterizes the performance of the analytical method for detecting the analyte. The derivation of the power function will not be discussed in detail here. It is obvious that it depends on the decision criterion for the detection, and hence on the critical concentration CCa and on the false positive rate a.For confirmatory purposes, the power function can be calculated: p x F t a b x s a bx a b x s a bx J J j j l j j x ( ) ( ) ( � � ) ( ) ( � � ) ; , ( = - + + + + + + + + æ è çç ö ø ÷÷ - - - 1 1 11 0 2 1 0 2 1 d a c) Var Var Var Var where x1 denotes the lowest concentration level in the calibration experiment and FJ 2 1; d (c) x denotes the distribution function of the t-distribution with J 2 1 degrees of freedom and the non-centrality parameter dx j j b x x a b x s a bx ( ) ( ) ( ) ( � � ) c Var Var = - + + + + 1 1 0 2 1 Figure 6 shows the power function calculated by means of the measuring results of the calibration experiment.As illustrated, the power function provides the critical concentration CCb at which the false-negative rate equals b, where b = 0.01 or b = 0.05 is given. The calculated values for the examined CAP confirmatory method are CCb = 0.50 and 0.55 mg kg21 for b = 0.05 and 0.01, respectively.The CCb can be used as a validation criterion: as long as CCb is below a given limit, it is guaranteed that for true concentrations above the limit the falsenegative rate does not exceed b. Discussion In trace analysis, the analytical results are usually affected by the type of matrix or by time-related conditions of the analytical system. A validation based on the presented variance component model refers not only to samples that correspond to the particular matrix used in the validation procedure but also, within certain time limits, to all future samples belonging to the defined population.As demonstrated in the Appendix, the model can be adapted not only to the different performance levels of methods, screening or confirmatory, but also to zero-tolerance and threshold decisions based on quantitative measurement data. The proposed concept of a matrix-considering in-house validation procedure should be considered with regard to the current discussion about the uncertainty of measurement.At present, there are different approaches under discussion about how to quantify uncertainty.18,19 The uncertainty of a measurement can be defined ‘as the interval on the measurement scale within which the true value lies with a specified probability, when all sources of error have been taken into account’.19 By taking into consideration matrix- and time-induced deviations, as done in the model presented, essential components of the uncertainty resulting from the systematic error are already covered by applying the in-house validation concept.By a recently published approach to the quantification of uncertainty, 19 the analytical error can be broken down into four components: the method bias, the laboratory bias, the run bias and the random measurement error.19 According to the concept presented, the run bias refers to the time-induced deviations and the method bias is additionally split into two components, the matrix-induced deviations and remaining method-induced biases.The decomposition of the uncertainty of a measuring result can be illustrated by the uncertainty tree in Fig. 7. It should be noted that neither collaborative studies nor in-house validation procedures are able to cover all components of uncertainty. Inhouse validation procedures do not cover the laboratory bias, whereas collaborative studies do not cover matrix-induced deviations.(There are some hints that the laboratory bias in comparison with the time effects can sometimes be neglected.20 When considering the laboratory bias as less important than other error components, it could be sufficient to determine it within proficiency tests, which are necessary anyway for the continuous assessment of the technical competence of the participating test laboratories.) Conclusion and outlook The application of the concept presented provides the analyst with comprehensive information about the performance of the method in question.Critical concentrations and error probabilities are calculated by means of a power functia variance component model. Ongoing validation Fig. 6 Power function of the examined CAP confirmatory method. Fig. 7 Uncertainty tree. Analyst, February 1998, Vol. 123 177experiments using different matrices, substance groups and analytical techniques are expected to confirm the practicability of the presented concept.Future work will be dedicated to developing the necessary statistical procedures for the integration of this validation concept into the overall framework of quality management. Financial support by the European Commission is gratefully acknowledged. Appendix: derivation of formulae Calibration function The statistical model considered in this paper is Y = Yij = m(xi) + �aj + �b jxi + eij (1) where Yij denote the measurement value at contaminant level i, i = 1, .. ., I and matrix j, j = 1, . . ., J. x1 @ x2 @ . . . @ xI denote the contaminant values in the calibration experiment and �aj + �b jx represents the matrix specific correction term of the calibration function m(x). The latter is assumed to be linear within the calibration interval, i.e., m(x) = a + bx for x1 @ x @ xI. Out of the interval two different extrapolations are applied. For screening methods the curve will be linearly extrapolated (type I calibration), as long as a + bx!0, i.e., m(x) = max{a + bx, 0}.For confirmatory methods (type II calibration) a worst case scenario is considered: m( ) x a bx x x a bx x x = + < + ÏÌÓ 1 1 1 if if ! The calculation of the critical concentration depends on the assumptions concerning the calibration function and on the threshold value. However, in each case it is based on a significance test: The critical concentration represents the measured concentration from which on the threshold value xthreshold is exceeded significantly.The threshold value is assumed to be greater or equal to zero. If it is zero (as in the example presented), the significance test consists in investigating the question whether the substance can be detected in the sample. However, one should take into account that the threshold value could also be a target value or a maximum admissible concentration. The underlying test hypotheses can be formulated as follows: H0 : x @ xthreshold against H1 : x > xthreshold where x denotes the unknown concentration of a sample with measurement value Y.Additionally a modified threshold is defined as x x x x 0 = ìíî threshold threshold 1 at type I calibration , } at type II calibration max{ which is equivalent to xthreshold with respect to the calibration function, i.e., m(x0) = m (xthreshold). According to the statistical model, the variance of the measurement value Y = Y(x) at concentration x can be written Var(Y) = s2 0 + s2 1(x), where s2 1(x) denotes the variance of the matrix-induced error and s2 0 denotes the measurement variance.In practice, these variance components are unknown and have to be estimated. For the moment we assume that they are exactly known. Let c = m(x0) = a + bx0 denote the expected signal at the threshold level and let �c = �a + �bx0 denote the estimate of c based on the data of the calibration experiment. The corresponding error variance of �c is denoted by s2� c.Then at the significance level a the null hypothesis H0 can be rejected if Y c x z c - + + > - � ( ) , � s s s a 2 1 2 0 0 2 1 where z1 2 a denotes the (1 2a)-quantile of the standard normal distribution. This formula does not take into account that �c might be negative, which is not allowed since m(x) ! 0. Replacing �c by max{�c, 0} provides the rejection rule Y c x z c - + + > - max{ , 0} � ( ) , � s s s a 2 1 2 0 0 2 1 The critical concentration CCa is the corresponding value on the x-axis, i.e.CC z b x c a b c a a s s s = + + + - - 1 2 1 2 0 0 2 0 � ( ) �, } � � � max{ Random sampling In practice the variance components s2� c, s2 0 and s2 1(x0) have to be estimated. We assume that J matrices are chosen at random. Each matrix sample is divided into I test portions, which will be spiked at concentration levels x1 . . ., xI. Linear regression of submodels Yij = aj + bjxi + eij = (a + aj) + (b + �b j)x + eij with fixed j provides estimates �aj, �b j and the corresponding residual variances s2j = SI i (Yij 2 �aj 2 �b jxi)2/(I 2 2). These parameters provide parameters of the overall calibration curve, the constant �a = 1 JSJj �aj, the slope �b = 1 JSJ j �b j and the estimated error variance �s2 0 = SJ j = 1 s2j /J.In order to obtain an estimator of the variance of s2 1(x0) of cj = aj + bjx0, we consider the estimation error gj of the estimator �cj = �aj + �bjx0 = aj + bj x0 + gj.Using the information matrix M = s2 the variance of the I x x x i i I i i I i i I å å å æ è çççç ö ø ÷÷÷÷ 2 estimation error gj can be calculated according to s s s g 2 0 2 0 0 0 2 0 1 2 2 1 1 2 1 1 1 1 1 = × æ è ç ö ø ÷ = + - - æ è çç ö ø ÷÷ é ë êêê ù û úúú ì í ï î ï ü ý ï � ï = = = å å å ( , ) ( ) / x x I x I x x I x i i I i i I i i I M-1 Replacing s2 0 by the estimator �s2 0 provides an unbiased estimator �s2 g of s2 g. Since the variance of �cj can be estimated by the empirical variance s2c of �c1, .. ., �cJ an unbiased estimator of s2 1(x0) is �s2 1(x0) = s2c 2 �s2g . If the result is negative, let �s2 1(x0) = 0. These estimators may be applied for the estimation of the sum of variances s2� c + s2 1(x0) + s2 0 by �s2� c + �s2 1(x0) + �s2 0. The latter is a linear combination of stochastically independent c2-distributed random variables. If these estimators replace the true variances in the formula for the CCa, the quantile z1 2 a has to be corrected, too.A conservative correction (which guarantees that the actual significance level never exceeds the given a) replaces z1 2 a by the critical value of the t-distribution with J 2 1 degrees of freedom, i.e., the CCa can be calculated CC t b x c a b J c a a s s s = + + + - - - 11 2 1 2 0 0 2 0 , � � � � ( ) � max{�, } � � 178 Analyst, February 1998, Vol. 123Power function The power function p(x) represents the probability of the exceedance of the critical concentration CCa in the case the sample measured has a true concentration x > x0.The power function depends on a, s0, s1, the type of calibration curve, the sampling design, etc. In order to determine the power function we have to compute the probability p(x) of Y c x t c J - + + > - - max{�, } � � ( ) � � , 0 2 1 2 0 0 2 1 1 s s s a where Y represents the measurement value of a sample with concentration x. It is assumed that according to the statistical model Y is normally distributed with expectation max{a + bx, 0} and variance Var(Y) = s2 0 + s2 1(x), where s2 1(x) denotes the variance component induced by random matrix effects at concentration x.Moreover it is assumed that the probability of negative estimates �c is neglectable. At first we consider the distribution of �s2� c + �s2 1(x) + �s2 0. Unfortunately, there is no closed form expression for its distribution. In fact, it is a linear combination of c2-distributed variables.Worst case considerations lead to the conservative approximation � � ( ) � ( ) ~ /( ) � � s s s s s s c c c J x x J 2 1 2 0 2 2 1 2 0 2 1 2 1 + + + + - - and bearing in mind that Y c x N b x x x c c - + + - + + æ è çç ö ø ÷÷ � ( ) ~ ( ) ( ) , � � s s s s s s 2 1 2 0 2 0 2 1 2 0 2 1 we obtain Y c x t b x x x c J x c - + + = - + + - � � � ( ) � ~ ( ) ( ) ( ) ( � � s s s d d s s s 2 1 2 0 2 1 0 2 1 2 0 2 where non-centrality parameter). x Here tJ 2 1(dx) denotes the t-distribution with J 2 1 degrees of freedom and non-centrality parameter dx.Because p x P Y c x t F t x x c J J J c c x ( ) � � � ( ) � ( ) ( ) � , , , � � = - + + &g a d a 2 1 2 0 0 2 1 1 1 11 2 1 2 0 0 2 2 1 2 0 2 1 the power function at the true concentration value x can be estimated p x F t x x J J c c x ( ) � � ( ) � � � ( ) � , , � � = - + + + + Ê Ë ÁÁ � � �� - - - 1 1 1 1 2 1 2 0 0 2 2 1 2 0 2 d a s s s s s s where FJ 2 1,dx denotes the distribution function of the tdistribution with J 2 1 degrees of freedom and non-centrality parameter d s s s x c b x x x = - + + ( ) � � ( ) � � 0 2 1 2 0 2 References 1 Codex Alimentarius, Volume Three, Residues of Veterinary Drugs in Foods, Codex Alimentarius Commission, Rome, 2nd edn., 1996. 2 Analytical Chemistry Laboratory, Guidebook—Residue Chemistry, USDA Food Safety and Inspection Service, Washington, DC, 1991. 3 EEC Directive 93/256, Off.J. Eur. Commun., 1993, No. L 118/64. 4 EEC Directive 85/591, Off. J. Eur. Commun., 1985, No. L 372/50. 5 Accreditation for Chemical Laboratories: Guidance on the Interpretation of the EN 45000 Series of Standards and ISO/IEC Guide 25, EURACHEM Secretariat, Teddington, 1993. 6 Food and Drug Administration, Guideline for Industry, Text on Validation of Analytical Procedures, 60 FR 11260, Center for Drug Evaluation and Research, Rockville, MD, 1995. 7 Food and Drug Administration, Draft Guideline on the Validation of Analytical Procedures: Methodology; Availability; Notice, 61 FR 9315, Center for Drug Evaluation and Research, Rockville, MD, 1996. 8 IUPAC, ISO and AOAC, J. Assoc. Off. Anal. Chem., 1989, 72, 694. 9 Capability of Detection—Part 1: Terms and Definitions, ISO/DIS 11843-1, International Standards Organization, Geneva, 1995. 10 Deutsches Institut f�ur Normung, DIN 32645, Chemical Analysis; Decision Limit, Detection Limit and Determination Limit; Estimation in Case of Repeatability, Terms, Methods, Evaluation, Beuth Verlag, Berlin, 1994. 11 Deutsches Institut f�ur Normung, DIN 55350, Teil 34, Concepts in the Field of Quality and Statistics; Limit of Detection, Limit of Determination and Capability of a Method for Determination, Beuth Verlag, Berlin, 1991. 12 IUPAC, Pure Appl. Chem., 1990, 62(1), 149. 13 Guidelines for Collaborative Studies, J. Assoc. Off. Anal. Chem., 1989, 72(4), 694. 14 Accuracy (Trueness and Precision) of Measurement Methods and Results; General Principles and Definitions, ISO 5725-1, International Standards Organization, Geneva, 1994. 15 Bundesgesundheitsamt, Planning and Statistical Evaluation of Ring Tests, Collection of official methods under Article 35 of the German Federal Foods Act, Beuth Verlag, Berlin, 1983. 16 Wernimont, G. T., in Use of Statistics to Develop and Evaluate Analytical Methods, ed. Spendley, W., Association of Official Analytical Chemists, Arlington, VA, 4th edn., 1993. 17 Stephany, R. W., in Residues of Veterinary Drugs in Food: Proceedings of the EuroResidue Conference in Noordwijkerhout, The Netherlands, May 21–23, 1990, ed. Haagsma, A., Ruiter, A., and Czedik-Eysenberg, P. B., University of Utrecht, Utrecht, 1990, pp. 76–85. 18 Quantifying Uncertainty in Analytical Measurement, EURACHEM Secretariat, Teddington, 1995 19 Analytical Methods Committee, Analyst, 1995, 120, 2303. 20 Zimmermann, R., and von Lengerken, J., Arch. Anim. Nutr., 1987, 37, 723. Paper 7/07281C Accepted October 10, 1997
ISSN:0003-2654
DOI:10.1039/a707281c
出版商:RSC
年代:1998
数据来源: RSC
|
5. |
Multivariate calibration on designed mixtures of four pharmaceuticals |
|
Analyst,
Volume 123,
Issue 2,
1998,
Page 181-189
Cevdet Demir,
Preview
|
PDF (125KB)
|
|
摘要:
N CH3O N H H HO CH CH2 Quinidine H2NOC HO CHCH2NHCH OH CH3 CH2CH Labetalol CH2CH CH2 OCH2CHCH2NHCH(CH3)2 OH Alprenolol HO CHCHNHCHCH2O OH CH3 CH3 Isoxsuprine Multivariate calibration on designed mixtures of four pharmaceuticals Cevdet Demir and Richard G. Brereton* School of Chemistry University of Bristol, Cantock’s Close, Bristol, UK BS8 1TS The use of multivariate calibration for estimating the concentration of mixtures by electronic absorption spectroscopy, as illustrated by quinidine, labetalol, alprenolol and isoxsuprine, recorded at five concentration levels, is considered.The importance of designing calibration experiments is described, including weighing and dilution schemes, consideration about spectral overlap and the properties of the design matrix. Several calibration methods are applied. A function for weighted regression is proposed. The ability to predict small weighing errors is studied. Keywords: Chemometrics; electronic absorption spectroscopy; multivariate calibration; weighted regression There is a very large literature on multivariate calibration of mixtures.1–10 Much recent work concerns applications to spectroscopic techniques such as electronic absorption spectroscopy (EAS)11,12 and mid-infrared spectroscopy, (MIR)13,14 where peaks are relatively broad and overlapping.Under such circumstances, most components absorb in most regions of a spectrum. However, much of the calibration literature is fairly empirical, not emphasizing factors such as design or spectral similarity as important in the quality of predictions. The purpose of this paper is to discuss these factors in a systematic manner, so investigating in a stepwise manner the design of the mixture experiments,15,16 choice of components in the mixture (if relevant), choice of regression method and so on.In addition, there have been several papers on weighted regression17,20 and this paper also investigates the improvements obtained by using a weighting function, plus the design of such a function.Methods Experimental HPLC-grade methanol was obtained from Rathburn (Walkerburn, UK) and quinidine, labetalol, alprenolol and isoxsuprine from Sigma (Poole, Dorset, UK). Structures are shown in Fig. 1. A Pharmacia LKB Biochrom Ultrospec III UV/VIS spectrometer, Model 80-2097-62 (Pharmacia, Uppsala, Sweden), equipped with deuterium (200–325 nm) and tungsten–halogen (325–900 nm) lamps were used to record electronic absorption spectra over the wavelength range 250–350 nm with a resolution of 1 nm.The wavelength reproducibility is ±0.5 nm and the photometric reproducibility is within 0.5% of the absorbance value. The Wavescan software package (Pharmacia LKB, Cambridge, UK) provided by the instrument supplier was used for data acquisition. Experimental design Absorptivity coefficients and concentration range The first step is to estimate the absorptivity coefficients of the four compounds.Five solutions in methanol with known concentrations were prepared for each compound (Table 1). Electronic absorption spectra were recorded over the wavelength range 250–350 nm. Superimposed spectra of the compounds at concentrations of 103, 104, 103 and 102 mg ml21 are shown in Fig. 2. From these measurements an equation linking absorbance to concentration for each compound can be obtained as follows ajkl = cklejk (1) where j is the wavelength, k is the compound, l the concentration level, c is the concentration and e is the absorptivity coefficient in absorbance cm21 mg21 ml.The absorptivity coefficient at Fig. 1 Structures of compounds studied. Table 1 Concentrations of standard compounds (mg ml21) Quinidine Labetalol Alprenolol Isoxsuprine 20.6 20.8 20.6 20.4 41.2 41.6 41.2 40.8 61.8 62.4 61.8 61.2 82.4 83.2 82.4 81.6 103.0 104.0 103.0 102.0 Analyst, February 1998, Vol. 123 (181–189) 181wavelength j is given by e jk jk k l L L a c L = = - - -å( ) / ( ) / 1 2 1 2 (2) where L is total number of concentration levels (L = 5), numbered from 2(L 2 1)/2 to +(L 2 1)/2, i.e., from 22 to +2 in the case discussed below.The absorption of a mixture of compounds at wavelength j is then given by A c j k jk k K = = å e 1 (3) In order to design a series of mixtures, the maximum and minimum concentrations of each compound must first be estimated and are given by cmax k and cmin k. By changing these values, an upper and lower limit to the summed absorbance are obtained as follows: A c j k jk k K max max = = å e 1 (4) A c j k jk k K min min = = å e 1 (5) The limits of absorbances are shown in Fig. 3. In no case is the absorbance greater than 0.7 and until around 330 nm the maximum and minimum absorbance graphs are fairly flat (it is impossible to achieve a completely flat graph for these particular compounds). Design A five concentration level design was chosen (a reduced 54 design) (see Table 4). The reduction was performed by choosing a subset of all possible 625 experiments with a number of properties.There are five mixtures for each compound at each concentration level. The design is nearly orthogonal. Although more orthogonal designs could be proposed, dependent on the number of compounds, the number of concentration levels and the number of samples. This design spans the mixture space fairly well. The nearly evenly distributed design over the experimental domain is given in Fig. 4. The correlation coefficients are calculated between compounds. The coefficients are 20.38 between quinidine and labetalol, 20.12 between quinidine and alprenolol, 20.12 between quinidine and isoxsuprine, 20.38 between labetalol and alprenolol, 20.12 between labetalol and isoxsuprine, and 20.38 between alprenolol and isoxsuprine. A complete five level factorial design would be prohibitive. Each level corresponds to a concentration, the lowest level (22) to the minimum and the highest (+2) to the maximum.The concentrations are calculated by c l c l c kl k k = + ÊË �� + - ÊË �� 2 4 2 4 max min (6) where l is the concentration level, ckl is the concentration for compound k at concentration level l and cmax k and cmin k are the maximum and minimum concentrations for compound k. The true concentrations of each compound at the five concentration levels are given in Table 2. Weighing and dilution scheme Five mother solutions (A–E) for each compound were prepared in 500 ml of methanol.The amount of compounds weighed for each of the mother solutions are given in Table 3. Each mother solution was diluted to give two stock solutions so that 10 stock solutions (Fig. 5) were obtained for each compound. From these stock solutions, 25 mixture of samples were prepared at five Fig. 2 Superimposed spectra of four compounds at concentrations given in the text. A, Quinidine; B, labetalol; C, alprenolol; and D isoxsuprine.Fig. 3 Absorption spectra of compounds at minimum and maximum concentrations in the range 250–350 nm. Fig. 4 Mixture space of two typical compounds, quinidine and labetalol. 182 Analyst, February 1998, Vol. 123concentration levels. The preparation scheme is shown in Table 4. For example, sample 1 consists of 30.50 mg ml21 quinidine, 6.08 mg ml21 labetalol, 14.25 mg ml21 alprenolol and 11.92 mg ml21 isoxsuprine (see Table 3). The spectra of mixtures were recorded in a random order. This design allowed each concentration level to arise from two independent dilutions and weighings so spreading the sample preparation error.As will be discussed below, calibration methods can be employed to determine weighing error approximately. Calibration Single wavelength multiple linear regression Calibration is a fundamental step in the calculation of the unknown concentrations of compounds in spectroscopy. The most commonly used calibration methods are univariate in the sense that they produce models that are linear functions of the measurements.There is an extensive literature on this subject. 21,22 Univariate calibration is based on a single measurement. It implies, for example, measuring instrumental response x at a single wavelength and converting it into a concentration by a sample calibration line. There are two approaches to univariate calibration. The first involves predicting a single concentration from absorbance at a single wavelength.The weakness of this is that it will fail if there are unknown interferences or completely overlapping spectra as is the case for three of the compounds. An alternative is to predict the concentrations using a matrix of all four concentrations, so that �Y = xjbj (7) where xj represents the absorbance at wavelength j, �Y is the estimated concentration matrix whose columns are the concentrations of four compounds and bj is the regression coefficient to be determined in the calibration.The regression coefficient can be obtained by multiple linear regression (MLR) as follows bj = (xAjxj)21xAjY (8) Once bj has been obtained, then the concentration of compound k can be estimated from the absorbance. The matrices are represented graphically in Fig. 6. Single compound multiple linear regression Another alternative is to predict the concentration of compound Table 2 True concentrations of compounds in the mixture using two weighing replicates (mg ml21) Level origin Quinidine Labetalol Alprenolol Isoxsuprine +2 (A) 30.50 14.20 22.30 24.10 +2 (D) 31.20 14.50 22.35 24.40 +1 (B) 24.12 12.13 18.28 20.33 +1 (D) 24.96 12.43 17.96 19.87 0 (C) 18.15 10.61 14.25 16.60 0 (E) 18.69 9.96 14.35 16.47 21 (B) 12.06 8.08 10.25 11.92 21 (E) 12.46 8.48 9.98 12.35 22 (A) 6.10 6.08 8.03 8.03 22 (C) 6.05 5.98 6.11 8.30 Table 3 Amounts of compounds weighed for the experimental design (mg) Sample Quinidine Labetalol Alprenolol Isoxsuprine A 61.0 28.4 44.6 48.2 B 60.3 28.3 43.9 47.7 C 60.5 27.9 44.8 49.8 D 62.4 29.0 44.7 48.8 E 62.3 29.7 45.1 49.4 Fig. 5 Preparation of 10 stock solutions, where 1–5 correspond to levels +2 to 22. Table 4 Preparation of mixtures. A–E correspond to weighings in Table 3. There are five levels for each compound, the highest being denoted by +2 and the lowest by 22 No. of samples Quinidine Labetalol Alprenolol Isoxsuprine 1 +2 (A) 22 (A) 0 (C) 21 (B) 2 +2 (D) 0 (C) 22 (A) +1 (B) 3 +2 (A) 21 (B) +1 (B) 0 (C) 4 +2 (D) 0 (E) 21 (B) +1 (D) 5 +2 (A) +1 (B) 21 (E) 22 (A) 6 +1 (B) +2 (A) 22 (C) 0 (E) 7 +1 (D) 21 (E) +2 (A) 0 (C) 8 +1 (B) 0 (E) 22 (A) +2 (A) 9 +1 (D) 22 (C) +2 (D) 0 (E) 10 +1 (B) 21 (B) 22 (C) 0 (C) 11 0 (C) 21 (E) +1 (D) +2 (D) 12 0 (E) 22 (A) +1 (B) 21 (E) 13 0 (C) 22 (C) +2 (A) 21 (B) 14 0 (E) 21 (B) +1 (D) 22 (C) 15 0 (C) +2 (D) +1 (B) 21 (E) 16 21 (B) +1 (D) +2 (D) 22 (A) 17 21 (E) +2 (A) 0 (E) 22 (C) 18 21 (B) +1 (B) 0 (C) 22 (A) 19 21 (E) +1 (D) 22 (A) +2 (A) 20 21 (B) 22 (A) 0 (E) +2 (D) 21 22 (A) 0 (E) 21 (B) +1 (B) 22 22 (C) +1 (B) 21 (E) +2 (A) 23 22 (A) +2 (D) 21 (B) +1 (D) 24 22 (C) +2 (A) 0 (C) 21 (B) 25 22 (A) 0 (C) +2 (A) +1 (B) Fig. 6 Graphical representation of matrices used in this paper. Analyst, February 1998, Vol. 123 183k using the entire spectrum by �yk = Xbk (9) where �yk is the estimated concentration for compound k, X is the UV data matrix and bk is the regression coefficient for compound k.Provided that the inverse of XAX exists, bk may be estimated by bk = (XAX)21 XAyk (10) where yk in this case is the concentration of compound k. When there are more samples than variables, this approach is referred to as multiple linear regression (MLR).23 If the number of variables is larger than the number of samples and the variables are linearly dependent or if they are highly correlated, the matrix is singular or nearly singular, which means that the computed regression coefficients are relatively imprecise.This collinearity can cause MLR to fail when applied to data sets containing highly correlated variables. In the application of our study, MLR was not effective because of high correlation between absorbances, especially at high wavelengths which are dominated by quinidine. Principal component regression The idea of principal component regression (PCR)24,25 is to form a model between a concentration of a compound and the principal components (PCs) of a data matrix, i.e., spectra.It differs from MLR in that the spectra are first reduced to a few orthogonal PCs. This means that even if there is a high correlation between wavelengths, PCR is nevertheless effective. PCR is performed using only one concentration vector at a time. When there is more than one concentration of interest, PCR is applied to each concentration vector separately. In PCR the data matrix X is first decomposed as in PCA. The decomposition is performed as follows: �X = TPA (11) where T and P are the scores and loadings of matrix X with dimensions I 3 J.The dimensions of T and P are I 3 N and J 3 N, respectively, where I is the number of samples, J is the number of variables and N is the number of components. PCR is obtained by regressing yk on T. The estimated regression coefficients bk are obtained simply via the pseudoinverse as follows bk = (TAT)21TAyk (12) Since the scores T are uncorrelated, the inverse of TAT exists.The coefficients bk can then be used to predict the concentration �yk. The same calculations are repeated for four compounds individually. If it is desired, the predicted concentration �yk is calculated for further principal components. Partial least-squares regression (PLS1) PLS is well described in the literature.26-28. The PLS decomposition most often used in calibration is called PLS129,30 and it performs the decomposition and regression for each compound individually.PLS is performed using NIPALS algorithm. It differs from PCR in that the covariance between the spectral matrix X and concentration vector yk is used for decomposition given by �X = TkPkA (13) and yk = UkqkA (14) where Tk and Uk are the scores of matrix X and the concentration vector yk for compound k, and Pk and qk are the corresponding loadings. The dimensions of Uk and qk are I 3 N and 1 3 N, respectively. When both X and yk are used to estimate components, the components for the X and yk have the following relationship: Unk = tnkbnk (15) where Unk and tnk represent the nth PLS component for compound k and bnk is a coefficient.PLS algorithms consist of two steps, calibration and prediction. In contrast to other regression methods, a matrix inversion is performed not in the calibration step but in the prediction step. The important part of regression is predicting the concentration vector �yk from the X matrix. This is done by decomposing the matrix X and building up the predicted concentration vector �yk.The regression model can be obtained for prediction of compound k as follows: �yk = Xbk (16) where the coefficient bk is calculated by bk = W(PAW)21qkA (17) Partial least squares regression (PLS2) The PLS2 algorithm31,32 was designed for the case when several concentration vectors yk are to be fitted using the same measured spectra X. The vectors are collected as columns in a matrix Y. PLS2 is performed using linear combinations of the Y variables.In contrast to most PLS1 algorithm, PLS2 must involve an iterative step for each of the components. The decomposition of the Y matrix is now given by Y = UQA (18) where Q is the loadings matrix with dimensions K 3 N for matrix Y and K is the number of compounds. The aim of PLS2 is to find a good linear model for all the concentration vectors simultaneously. The model is given as follows �Y = XB (19) The coefficients B are calculated by B = W(PAW)21QA (20) It should be noted that the same PCs occur in the model for each concentration vector, only the regression coefficients change.Weighted regression Application to regression Improvement of regression methods can be achieved by weighting the wavelengths according to significance. In this paper we explore the possibilities of weighted regression. The data matrix X can be weighted at each wavelength individually by a weighting vector denoted vjk. This vector differs according to compound k.A new data matrix kZijk is then obtained as follows: kZijk = Xijvjk (21) This data matrix is used for the PCR and PLS regressions as discussed in eqns. (11) and (13). In PCR, the decomposition of weighted data matrix kZ can be defined by k �Z = kTkPA (22) followed by regression the concentrations on to the scores. A also be obtained for PLS: k �Z = kTkkPkA (23) 184 Analyst, February 1998, Vol. 123The same concentration vectors yk and the matrix Y are used for the decomposition of kZ as for unweighted regression.Computation of weighting function In this paper, we assume that the higher the correlation between concentration and absorbance, the more useful the wavelength is in regression. The correlation coefficient between independent variables yk and dependent variables x at wavelength j for compound k can be defined by r x x y y x x y y jk ij j ik k i I ij j ik k i I i I = - - - - = = =    ( )( ) ( ) ( ) 1 2 2 1 1 (24) One function that influences the effectiveness of wavelengths in calibration is their correlation with the concentration matrix.However, it is debatable what to do if the correlation coefficients are close to zero or negative. Any weighting function must be positive. Standardizing the 101 correlation coefficients over all wavelengths ranks the correlation coefficients and puts them on a common scale as follows: s r r r r r I jk jk k jk k i I = - - - = å ( ) ( ) 2 1 1 (25) where srjk is standardized correlation coefficient for compound k at wavelength j and �r k is mean correlation coefficient over all wavelengths. The reason for standardizing is first to select about half the wavelengths, which is done by eliminating those with negative standardized correlation coefficients. The second is that some compounds such as quinidine will exhibit high correlations whereas other may exhibit correlations that are low because of spectral overlap.All wavelengths with standardized correlation coefficients less than zero were eliminated, keeping approximately half the wavelengths. The remaining positive standardized correlation coefficients were used to obtain the weighting function. Finally, the weighting function is calculated as follows so that low intensity wavelengths are reduced in significance: vjk = srjkejk (26) where vjk is the weighting function for compound k at wavelength j and the absorptivity coefficients e were calculated using eqn.(1). The weighting function vjk obtained above was used to calculate the data matrix Z in eqns. (22) and (23). The weighting functions are shown for each compound in Fig. 7. Determination of number of components An important aspect of multivariate calibration is to determine the optimum number of principal components. In PCR and PLS, sufficient principal components are required to ensure that the most information is present.In order to model the calibration without overfitting the concentration data, cross-validation can be used. There are several methods for performing crossvalidation. In this paper, we use the method of leaving out one sample at a time.33 One sample is removed from the training set, a model is constructed based on the remaining samples and a prediction is made on the left out sample. Each sample is left out in turn so that a new model is constructed and a new prediction is performed on the second sample, etc.This process continues until each sample has been left out once. To determine the optimum number of principal components, a root mean square validation error (RMSVE) is calculated. This describes how well the calibration model predicts an unknown sample. A minimum RMSVE indicates the optimum number of principal components. The root mean square modelling error, RMSME, can also be calculated. The calculation of RMSVE has been described previously.34 The calibration error (RMSE) is calculated for all spectra together.This differs from the modelling error calculated only on the training set. Weighing errors One of the important sources of error in calibration is due to sample preparation. In order to determine this type of error each level is replicated twice using two weighings. As has been explained in the section Weighing and dilution scheme, five weighing replicates were performed for each compound.These replicates are evenly spaced in the experimental design at five concentration levels. Although the balance shows an accurate concentration accyik for each level, in this paper we assume that the concentration errors are unknown. An average concentration over all 25 samples is given by y y I k acc ik i I = = å1 (27) Calibration is performed using average concentrations, avyik, rather than the accurate concentrations. An aim of this work is to see how well small weighing errors are predicted. The average concentration at level l and for compound k is calculated according to the following equation: av max min max min y y c c l l l kl k k k = + - - ( ) ( ) (28) where lmax and lmin are maximum and minimum concentration levels. There are five concentration levels (Table 2) for each compound.The concentration where the weighing errors have been averaged out is defined by avyik = avykl (29) Fig. 7 Weighting functions used for weighted calibrations.A, Quinidine; B, labetalol; C, alprenolol; and D, isoxsuprine. Analyst, February 1998, Vol. 123 185Predicted concentrations for each compound are obtained using calibration methods discussed above. The predicted concentration using the averaged weighing is given by av�yik. Note that av�yik is not the same for each level; it is expected that the magnitude is influenced by small weighing errors. The aim of this work was to see if the weighing error can be predicted. For each level there are two separate weighings, each of which results in a set of wklm mixtures, where k is the compound, l is the level and n is the set of weighings.The value of wklm = 2 or 3. The deviation from the average weight for set kln will be given by Dkln = avykl 2 accykln (30) where accyklm is the accurate value of a concentration for compound k, level l and weighings n. For example, for level 22 and weighing A (sample 1) for quinidine, w1,22,1 = 3 and accy1,22,1 = 30.50.The predicted weighing error is different. For weighing set kln the average predicted weight is given by � � y y k k ik i ik ln ln ln ln = '  (31) and the average weight over all five dilutions simply by � � y y kl ikl i kl = ' å 5 (32) The difference between �� ykl and �� ykln is that the former is the average predicted over all five mixtures at level l for compound k and the latter is simply the prediction at weighing n. The difference can be calculated as follows: �D kln = �� ykl 2 �� ykln (33) This is an estimate of the weighing error.Hence the root mean square error in estimated weighing error for compound k is given by Ek k k n l = - Ñ = = - + å å ( � ) D D ln ln 2 1 2 2 2 (34) where Î is the total number of weighings and levels, which equals 10 in this case. Data processing Matlab Version 4.2c for MS Windows (Mathworks, Natick, MA, USA, 1994) and Microsoft Excel were used for data processing. PCR, PLS1 and PLS2 calibrations were performed and validated where appropriate against PLS_Toolbox published by B.Wise (Eigenvector Research Inc.). Weighted calibrations were carried out by modification of existing algorithms. Results Single wavelength multiple linear regression The errors from the method in the section Single wavelength multiple linear regression are illustrated in Fig. 8. The minimum errors are at wavelength 335 nm for quinidine and 303 nm for labetalol. The errors for alprenolol and isoxsuprine are at the same wavelength, 271 nm.Quinidine has no significant interferents, so 335 nm is a compromise high wavelength. However, noise dominates where the intensity is lowest. The other three compounds are almost completely overlapping and they have higher errors, but the minimum errors in all cases are approximately at the position of maximum absorbance (Fig. 2). Note that the maximum absorbances of alprenolol and isoxsuprine correspond closely to a minimum in the spectrum of labetalol, and the maximum absorbance of labetalol corresponds to a local minimum in the spectrum of quinidine.Note also that alprenolol and isoxsuprine always have a higher error than quinidine and labetalol. This is a consequence of their spectral characteristics. Quinidine, with a pronounced composition 1 regionted calibration Multivariate calibration methods, PCR, PLS1 and PLS2, were applied to the UV/VIS spectral data of a mixture of four compounds.In order to compare the methods, root mean square errors were calculated between true and predicted concentrations. The RMSE of prediction of PCR, PLS1 and PLS2 calibrations for the four compounds using eight principal components are presented in Table 5. It can be seen that the values of the RMSE are similar for PCR, PLS1 and PLS2 in all cases. Among the calibration methods, PLS2 has slightly better prediction error for all the compounds. This is because of the interference of compounds in the spectra is taken into account. In PLS1, the calibration is performed using only one concentration at a time, whereas in PLS2 all of the concentrations are used simultaneously.As can be seen from Table 5, the lowest RMSE is obtained for quinidine in all calibration methods. Higher prediction errors are obtained for alprenolol and isoxsuprine. This is because the spectra of quinidine and labetalol overlap only slightly and exhibit good composition 1 regions; however, alprenolol and isoxsuprine completely overlap.To determine the number of PCR, PLS1 and PLS2 components, a cross-validation method, leaving out one sample at a time, was performed. Having a set of 25 samples, calibrations were performed on 24 samples, and using this calibration, the concentrations of compounds in the sample left out during calibration were predicted. Each of the 25 samples was left out in turn and the predicted concentrations of compounds in each sample were compared with the true concentrations.The RMSVE was then used as a measure to determine the optimum number of components. The values of RMSVE as a function of principal components are shown in Figs. 9, 10 and 11 for PCR, PLS1 and PLS2, respectively. In PCR, for the particular example the minimum Fig. 8 Over-all univariate calibration errors for all wavelengths. 2, Quinidine; ½, labetalol; . alprenolol; and 8, isoxsuprine. 186 Analyst, February 1998, Vol. 123error was found for component six for quinidine and labelatolol, seven for isoxsuprine and eight for alprenolol. Similar results were obtained in PLS1 and PLS2 cross-validations. The same optimum number of PLS components were found for PLS1 and PLS2 (Figs. 10 and 11). The values of RMSVE are slightly different from PCR. The optimum numbers of PLS1 and PLS2 components are five for labetalol, six for quinidine, seven for isoxsuprine and eight for alprenolol. The only difference for PCR is that the number of component is six for labetalol, whereas this number is five for the same compound in PLS1 and PLS2.Interestingly, all graphs show a sharp fall-off up to three components for quinidine and labetalol and four for isoxsuprine and alprenolol. The difference in behaviour can be explained by the similar spectra of isoxsuprine and alprenolol, which might be confused when quinidine and labetalol are analysed, suggesting only three different components in the mixture.In the case of isoxsuprine and alprenolol, the influence of all three other components is felt. It would be interesting to speculate how the design of the experiments could influence the number of significant components. Weighted calibration Weighted calibration was carried out by weighting the wavelengths as described above. Four weighting functions were obtained, one for each compound. Weighted principal component regression (WPCR) and weighted partial least squares, WPLS1 and WPLS2, were performed.The WPLS1 results are given in Table 6. Similar trends were observed for WPLS2 and WPCR but have been omitted for brevity. As in unweighted calibration, the values of RMSE are slightly different for WPCR, WPLS and WPLS2 calibrations. However, weighted calibration gave a substantial improvement over the unweighted calibration. The best results were obtained for quinidine. It may be concluded that weighting the wavelengths gives better predictions for less overlapped spectra. However, weighted calibration also gives better predictions for other compounds. Note that the RMSE is higher for labetalol, alprenolol and isoxsuprine using a weighting function based on quinidine than using unweighted methods.This is because this function is based on the spectrum of quinidine. For weighting based on labetalol, the compound would have the least prediction error but the corresponding conclusions are not valid for alprenolol and isoxsuprine (weighting functions based on alprenolol and isoxsuprine).The RMSE is less in weighted than unweighted calibration in all cases. However, weighting the wavelengths Fig. 9 RMSVE as a function of number of PCR components. 2, Quinidine; ½, labetalol; ., alprenolol; and 8 isoxsuprine. Fig. 10 RMSVE as a function of number of PLS1 components. 2, Quinidine; ½, labetalol; ., alprenolol; and 8 isoxsuprine. Table 5 Root mean square errors of unweighted PCR, PLS1 and PLS2 calibrations (mg ml21) Quinidine Labetalol Alprenolol Isoxsuprine Components PCR PLS1 PLS2 PCR PLS1 PLS2 PCR PLS1 PLS2 PCR PLS1 PLS2 1 3.5522 3.5678 3.5311 4.5199 4.5157 4.5054 6.8986 6.8907 6.8751 7.2247 7.2137 7.1907 2 1.6698 1.4978 1.2217 3.9306 3.8016 2.5697 4.9981 5.0255 4.9509 4.1977 4.1895 4.1617 3 0.4260 0.4284 0.4198 0.6026 0.6008 0.5980 4.9831 4.9797 4.3077 4.1894 4.1892 3.5148 4 0.4171 0.4235 0.2232 0.6017 0.6007 0.4391 1.8100 1.4773 1.3953 1.6546 1.5268 1.4815 5 0.2929 0.2174 0.1857 0.5248 0.5063 0.4041 1.4733 1.0083 1.0166 1.6475 1.4976 0.8290 6 0.1939 0.2126 0.1817 0.4531 0.4705 0.4032 1.3062 0.8070 0.7854 1.6328 0.8406 0.7269 7 0.1917 0.0987 0.1809 0.4352 0.2903 0.3990 1.0200 0.5814 0.7549 1.0158 0.5554 0.6687 8 0.1873 0.0820 0.1621 0.4348 0.2397 0.3715 1.0068 0.4099 0.4650 1.0142 0.4024 0.5125 Fig. 11 RMSVE as a function of number of PLS2 components. 2, Quinidine; ½, labetalol; ., alprenolol; and 8 isoxsuprine. Analyst, February 1998, Vol. 123 187does not effect the predictions sufficiently in the case of overlapped spectra, as expected. Cross-validation results for WPLS1 are shown in Fig. 12. The weighting function corresponds to the compound whose concentration is estimated. For example, the curve for quinidine corresponds only to the quinidine weighting function. The optimum number of components is reduced by one in all cases. Instead of a flat plateau over three components, the corresponding fall off is observed after only two components. An explanation is that the weighting functions emphasize composition 2 regions where only two significant components overlap.Weighing errors The graphs of root mean square errors in the estimated weighing error for PLS1 are given in Fig. 13. Several conclusions can be drawn. The first is that, in general, the more components there are the better the weighing error is estimated. For quinidine and labetalol an obvious minimum is observed at three components.These errors are small so it gives good confidence in the prediction that these have been estimated fairly well. Similar conclusions can be obtained using other approaches but only one is illustrated here for brevity. The weighing errors are small, and it is remarkable that the regression methods are able to predict these small errors. Some errors will be compounded by unknown sample preparation errors, such as those due to dilution, which have not been systematically studied in this work.However, in many practical situations, it is assumed that errors in the concentration axis are unknown. Calibration can provide the experimenter with strong guidelines as to the accuracy of his/her experiments, especially if a procedure is repeated over time. It is recommended that a procedure such as that recommended in this paper is performed first to see whether these experimental errors can be reasonably well predicted and, if so, to use this procedure later to predict unknown sample preparation errors in real situations. Conclusion Much of the literature on multivariate calibration is fairly empirical, reflecting the statistical background to this area.In conventional statistics it is hard to put physical meaning to effects; for example, economic and social statistics do not normally result from planned experiments. However, in the area of the spectroscopy of mixtures there are very systematic and planned effects.An obvious example is spectral overlap; different compounds show different levels of spectral similarity and so differences in composition 1 regions. This in turn influences the effectiveness of methods for multivariate calibration and hence appropriate weighting functions. It is important to understand, in a systematic manner, how these factors influence multivariate calibration. It is important not to rely too greatly on any single method. In some cases univariate single wavelength calibration can perform perfectly well, for example, in the case of quinidine, where the best wavelength (see Fig. 8) gives an error of less than 1 mg ml21, less than the calibration error, for example, of PLS1 using two components (Table 5).However, in the other cases, Table 6 Root mean square errors of WPLS1 calibration (mg ml21) Weighting function Combased on ponents Quinidine Labetalol Alprenolol Isoxsuprine Quinidine 1 1.2700 5.3877 7.9892 8.7701 2 0.4246 0.5911 7.0109 7.1381 3 0.3635 0.5467 6.1029 5.0316 4 0.2258 0.4619 4.8949 4.4388 5 0.1833 0.3584 4.0650 3.4985 6 0.1483 0.2863 3.3162 3.0476 7 0.1113 0.2208 2.7305 2.6354 8 0.0936 0.1803 2.3171 2.4109 Labetalol 1 4.7669 3.7226 7.1320 7.4415 2 0.4736 0.6735 6.8610 6.8267 3 0.3804 0.5764 6.0784 4.2538 4 0.2418 0.5565 4.8376 1.6105 5 0.1921 0.4681 4.4671 1.3138 6 0.1636 0.4074 3.7914 1.0552 7 0.1417 0.3635 3.5003 0.9098 8 0.1162 0.3103 3.1307 0.7211 Alprenolol 1 5.2782 4.3414 6.0015 6.0681 2 2.2207 4.2234 4.2139 4.8260 3 1.9216 3.4642 3.2697 3.5542 4 1.6136 2.8043 2.7465 3.0219 5 1.4190 2.4187 2.5814 2.7979 6 1.3665 2.3497 2.1902 2.4481 7 1.2938 2.1980 1.9126 2.1788 8 1.1755 2.0154 1.8303 2.1051 Isoxsuprine 1 5.2158 4.3442 6.0392 6.0924 2 2.4263 4.2787 4.1192 5.0931 3 1.8586 3.1965 3.0381 3.2425 4 1.0299 2.2102 2.1679 1.9796 5 0.9151 2.0232 1.7970 1.4483 6 0.7894 1.8074 1.4268 1.1640 7 0.6622 1.4836 1.0955 0.9452 8 0.4951 1.2932 0.9774 0.8185 Fig. 12 RMSVE as a function of number of WPLS1 components. 2, Quinidine; ½, labetalol; ., alprenolol; and 8isoxsuprine. The curves show the corresponding weighting function for each compound. For example, the curve for quinidine corresponds only to the quinidine weighting function. Fig. 13 RMSE (mg ml21) in estimated weighing error for PLS1. (a) Quinidine; (b) labetalol; (c) alprenolol; and (d) isoxsuprine. 188 Analyst, February 1998, Vol. 123where there is complete spectral overlap, univariate methods perform less well.To a certain extent, the performance of univariate approaches depends, in part, on both the design of the calibration experiments and spectral overlap. PLS2 has very few advantages over PLS1 (Table 5) and, in fact, in many cases performs slightly worse, whereas PLS1 is much better than PCR. Weighting functions are not very commonly used in multivariate calibration, but can show some advantages when the pure spectra are known. However, if two spectra are very similar, there is no magic approach that will give good results, even if the experimental design is optimal.If the experiments are designed poorly, then conclusions about the relative effectiveness of different algorithms may be unreliable. It is possible to report calibration errors in a very empirical way, but this may simply imply that the experiment is poorly designed rather than that any particular method is performing well. This problem is particularly serious when analysing field samples, rather than laboratory mixtures, as the compound distributions may not be ideal, so certain algorithms may work unexpectedly well with a test set but fail to predict a complete unknown simply because the mixture space has not been spanned adequately.This paper provides a good case study for designing and analysing multivariate spectroscopic calibration experiments, and it is recommended that investigators follow some or all of the steps above in order to develop systematic calibration models for spectroscopic mixtures.Uludag University (Bursa, Turkey) is thanked for financial support for this project (C.D.). References 1 Bebe, K. R., and Kowalski, B. R., Anal. Chem., 1987, 59, 1007A. 2 Haaland, D. M., Anal. Chem., 1988, 60, 1208. 3 Thomas, E. V., and Haaland, D. M., Anal. Chem., 1990, 62, 1091. 4 Brown, P. J., Anal. Proc., 1990, 27, 303. 5 Kowalski, B. R., and Seasholtz, M. B., J. Chemom., 1991, 5, 129. 6 Espinosa-Mansilla, A., Mu�noz de la Pe�na, A., Salinas, F., and Zamoro, A., Anal. Chim. Acta, 1992, 258, 47. 7 Dur�an-Mer�as, I., Mu�noz de la Pe�na, A., Espinosa-Mansilla, A., and Salinas, F., Analyst, 1993, 118, 807. 8 Thomas, E. V., Anal. Chem., 1994, 66, 795A. 9 Andrew, K. N., and Worsfold, P. J., Analyst, 1994, 119, 1541. 10 Zhang, P., and Littlejohn, D., Chemom. Intell. Lab. Syst., 1996, 34, 203. 11 Cirovic, D. A., Brereton, R. G., Walsh, P. T., Ellwood, J. A., and Scobble, E., Analyst, 1996, 121, 575. 12 Araujo, P. W., Cirovic, D. A., and Brereton, R. G., Analyst, 1996, 121, 581. 13 Sun, J., J. Chemom., 1995, 9, 21. 14 Vigneau, E., Bertrand, D., and Qannari, M., Chemom. Intell. Lab. Syst., 1996, 35, 231. 15 Araujo, P. W., Kavianpour, K., and Brereton, R. G., Analyst, 1995, 120, 295. 16 Araujo, P. W., and Brereton, R. G., Analyst, 1995, 120, 2497. 17 Cleveland, W. S., and Devlin, S. J., J. Am. Stat. Assoc., 1988, 83, 596. 18 Carroll, R. J., and Ruppert, D., Transformation and Weighting in Regression, Chapman and Hall, New York, 1988. 19 Næs, T., and Isaksson, T., Anal. Chem., 1990, 62, 664. 20 Wang, Z., Isaksson, T., and Kowalski, B. R., Anal. Chem., 1994, 66, 249. 21 Riu, J., and Rius, F. X., J. Chemom., 1995, 9, 343. 22 Booksh, K. S., and Kowalski, B. R., Anal. Chem., 1994, 66, 782A. 23 Draper, N. R., and Smith, H., Applied Regression Analysis, Wiley, New York, 2nd edn., 1981. 24 Gemperline, P. J., and Salt, A., J. Chemom., 1989, 3, 343. 25 Sun, J., J. Chemom., 1996, 10, 1. 26 Wold, S., Ruhe, A., Wold, H., and Dunn, W., SIAM J. Sci. Stat. Comput., 1984, 5, 735. 27 Geladi, P., and Kowalski, B. R., Anal. Chim. Acta, 1986, 185, 1. 28 H�oskuldsson, A., J. Chemom., 1988, 2, 211. 29 Lorber, A., Wangen, L., and Kowalski, B. R., J. Chemom., 1987, 1, 19. 30 Martens, H., and Næs, T., Multivariate Calibration, Wiley, New York, 1989. 31 Geladi, P., and Kowalski, B. R., Anal. Chim. Acta., 1986, 185, 19. 32 Manne, R., Chemom. Intell. Lab. Syst., 1987, 2, 187. 33 Lorber, A., and Kowalski, B. R., Appl. Spectrosc., 1990, 44, 1464. 34 Demir, C., and Brereton, R. G., Analyst, 1997, 122, 631. Paper 7/05010K Received July 14, 1997 Accepted October 15, 1997 Analyst, February 1998
ISSN:0003-2654
DOI:10.1039/a705010k
出版商:RSC
年代:1998
数据来源: RSC
|
6. |
Liquid–liquid extraction in flow injection analysis using an open-phase separator for the spectrophotometric determination of copper in plant digests |
|
Analyst,
Volume 123,
Issue 2,
1998,
Page 191-193
Telma Blanco,
Preview
|
PDF (65KB)
|
|
摘要:
Liquid–liquid extraction in flow injection analysis using an open-phase separator for the spectrophotometric determination of copper in plant digests Telma Blanco†, Nelson Maniasso, Maria Fernanda Gin�e* and Antonio O. Jacintho Centro de Energia Nuclear na Agricultura, Universidade de S�ao Paulo CENA-USP, Av Centen�ario 303, Piracicaba, C. Postal 96, 13400-970 Brazil A flow injection system with an open-phase separation device is presented for the liquid–liquid extraction of the diethyldithiocarbamate–copper complex (DDTC–Cu) using carbon tetrachloride (CCl4).An open tube enclosed by computer-controlled solenoid valves was installed vertically to promote fast phase separation by differences in density. The solenoid valves were used to introduce the reacted sample zone into the tube, to select a precise volume of the extracted complex to be detected and to pass the remaining solution to waste. A preconcentration factor of 10 was attained, corresponding to the aqueous to organic phase volume ratio. The solenoid valve control allowed the use of a small solvent volume (0.7 ml) per determination.The DDTC–Cu complex was detected at 436 nm. A sample throughput of 30 determinations per hour for Cu ranging from 50 to 400 mg l21 in real samples was achieved. A limit of detection of 5 mg l21 of Cu was calculated using 3s. A precision characterised by an RSD of less than 3% was typical for a sample with a copper concentration of 145 mg l21.The accuracy was assessed with NIST SRM 1572 and no difference at the 95% confidence level was found. Keywords: Liquid–liquid extraction; plant analysis; copper determination; flow injection; spectrophotometry Automation in analytical chemistry procedures has been beneficial in minimising environmental effects by decreasing the amounts of reagents and waste.1 Recent proposals using flow injection analysis have demonstrated these characteristics in addition to showing improvements in sample throughput, sensitivity, precision and accuracy.2,3 In the search for highly sensitive flow systems, several on-line procedures for matrix separations and preconcentration have been described.4 Preconcentration methods using ion exchangers have been used more than liquid–liquid extraction procedures.The major limitation to performing an efficient solvent extraction in a flow system probably resides in the phase separation step. The first papers using phase separation in flow systems used T-shaped gravitational devices.5,6 Problems with compensation of the separator flow rates affected the precision of measurements. Also, poor efficiency of the separation process was associated with the high sample dispersion inside the T-devices.Another inconvenience was the high solvent consumption of the flow system, most of it being wasted. Studies concerning mechanisms of extraction and dispersion in liquid–liquid segmented flow analysis demonstrated the importance of film formation along the analytical path.7 Solvent extraction in flow systems using phase separation in PTFE microporous membranes has been reported.7 However, devices using membranes were considered to have short lifetimes.Lucy and Yeung8 described the formation of a wetting film of the organic phase inside the extraction coil by using solvents of high viscosity and low interfacial tension. Recently, the wetting film process was performed in a flow system using sequential injection of the solvent, the sample and the eluting solution.9 However, the wetting film proposal was demonstrated for only some solvents, the most critical aspects being related to the thickness and stability of the wetting film.In this work, the liquid–liquid extraction process was performed in a flow system using an extraction coil and an open–phase separator device based on differences in density. Automation of the process was necessary to reduce the volume of the organic solvent to attain sufficient sensitivity for the determination of Cu in plant material.Experimental Apparatus A peristaltic pump (IPC-8.V1.34, Ismatec, Z�urich, Switzerland) provided with Tygon and Acidiflex pump tubing, an automatic injector (Model 352, Micronal, S�ao Paulo, Brazil) and a spectrophotometer (Model 432, Femto, S�ao Paulo, Brazil) furnished with a flow cell (10 mm optical path, 80 ml inner volume) coupled to a strip-chart recorder (REC 61, Radiometer, Copenhagen, Denmark) were used.A microcomputer (IBM compatible 486) equipped with an interface card (PCL-711s, American Advantech, San Jose, CA, USA) controlled a set of four three-way solenoid valves (Model 161T031, NResearch, Stow, MA, USA). The flow system manifold included a mixing coil of polyethylene tubing (0.8 mm id), an extraction coil made of PTFE tubing (0.5 mm id, wall thickness < 0.2 mm), flow lines of polyethylene (0.8 mm id) and PTFE (0.5 mm id) tubing and a laboratory-made phase separator.The phase separator was constructed from a 20 cm 3 8 mm id glass tube. The higher extremity of the tube was completely open. At the other extreme the tube presented a slightly concave bottom, with two lateral exits at opposite sides of approximately 1 mm id. Software written in QuickBasic 4.5 was used to control the solenoid valves. Reagents and solutions All chemicals were of analytical-reagent grade and freshly distilled, de-ionized water was used throughout.A stock standard solution of 1000 mg l21 Cu2+ was prepared by dissolving 1.9650 g of CuSO4·5H2O in water, adding 5 ml of concentrated H2SO4 and diluting to 500 ml. Working standard solutions with concentrations from 50 to 400 mg l21 of Cu were prepared by diluting the stock standard solution with 0.25 m HClO4. Aqueous solutions of 0.25 m HClO4 as the carrier for the sample, EDTA (5% m/v) of pH 9–10, adjusted with concentrated ammonia solution, to form cationic complexes and sodium diethyldithiocarbamate (NaDDTC) (5% m/v) to pro- † Instituto de Qu�ýmica de S�ao Carlos, Universidade de S�ao Paulo, S�ao Paulo, Brazil.Analyst, February 1998, Vol. 123 (191–193) 191duce the Cu extractable compound were used. The solvent employed was CCl4. Stock standard solutions of 1000 mg l21 of Fe3+, Mn2+, Ni2+, Co2+, Mo6+ and V5+ were prepared from the respective oxides. A mixed solution of cations was prepared from the stock standard solution in 0.25 m HClO4, containing 200 mg l21 Cu2+ + 30 mg l21 Mn2+ + 3.0 mg l21 Ni2+ + 12 mg l21 Fe3+ + 1.5 mg l21 Mo6+ + 0.5 mg l21 Co2+ + 0.5 mg l21 V5+.Plant samples from different species, including NIST SRM 1572 Citrus Leaves reference material, were digested following the nitric acid–perchloric acid methodology described earlier. 10 Flow system The formation of the DDTC–Cu complex followed by extraction with CCl4 and phase separation were performed in the flow system shown in Fig. 1. The injector device is shown resting in the sample loading stage. The solution pumped at S fills the loop L while the solutions placed at C1, C2, C3 and C4 are introduced into the system through the injector and confluence points x and y flowing towards the waste through valve V1. All the solenoid valves are switched off and the free route is indicated by the solid lines. In this position, CCl4 is aspirated flowing through valve V2 to the spectrophotometer D, being recycled after passing valve V3.Meanwhile water is aspirated through valve V4. After shifting down the central part of the injector, the sample in L is carried towards point x, where the solution of EDTA is added. The two solutions are mixed while being transported through coil MC. At point y the reagent solution NaDDTC and the solvent CCl4 are merged simultaneously with the sample zone. The reaction and solvent extraction take place during passage through the extraction coil EC.When the reacted sample zone reaches V1, the valve is opened to collect the solutions at the phase separator. After a delay of 1 min, valve V1 is closed, the injector is returned to the original sampling position and valves V2 and V3 are switched on together. In this way, the organic phase is aspirated from the phase separator, passing through the fln of the coloured complex at 436 nm. Valve V2 is closed after 20 s, opening V4 to pass the solutions remaining in the phase separator to waste.Valve V4 remains opened for 40 s. Valve V3 is closed 10 s after V2 to start recycling the pure CCl4. A timing diagram for the devices is shown at the bottom of Fig. 1. Results and discussion The flow rates and concentrations of the sample and reagents were defined based on the manually performed procedure.11 To ensure good EDTA complex formation, mixing coils of different lengths (25–100 cm) were tested. The absorbance obtained by using the mixed solution containing different cations in addition to Cu was compared with that of a pure 200 mg l21 Cu solution in order to evaluate the degree of complexation.A mixing coil of 75 cm was used, as with smaller coils complete EDTA complexation of concomitant species was not attained. Hence the signals were characterized by low precision, higher than that of Cu, indicating a positive interference. No effect was observed with the simultaneous addition of the reagent solution NaDDTC and CCl4 to the sample zone.The extraction process was more efficient when the length of the extraction coil was increased from 100 to 250 cm and when the inner diameter of the tube was decreased from 0.7 to 0.5 mm. A coil of 250 cm 3 0.5 mm id was adopted. No problems of leakage at the connections as a consequence of the increased internal pressure were observed. The time elapsed for the sample to reach valve V1 (25 s) was adjusted in a preliminary test using a dye solution.The flow rates of the aqueous and solvent solutions were set to a ratio of 10. To attain a sample concentration factor of 10, both the injected sample volume and the time elapsed during collection of the solutions at the phase separator (valve V1 opened) must be accurately adjusted to select just the reacted sample bulk. The volumes collected at the phase separator must be sufficient to fill the flow cell in a reproducible way without affecting the sample throughput.Sample loops containing up to 2.0 ml were tested; 1.5 ml was chosen as the volume collected at the phase separator during 1 min satisfied the above-mentioned conditions. In addition, the collected volume of the organic phase (0.32 ml) filled about 0.7 cm of the tube, diminishing the risk of aspirating the aqueous phase to the detector. The synchronisation of the injector and the valves was adjusted so as to carry out 30 determinations per hour. This sample throughput was possible because the time for filling the sample loop, injection and transportation as far as V1 was overlapped with the delays to fill and deliver solutions from the phase separation device (Fig. 1). A volume of solvent of less than 0.7 ml was used per determination, not justifying the online recovery proposed earlier.3 The CCl4 collected was recovered in a batch process. The peaks registered during a routine analysis are shown in Fig. 2. The linearity of the calibration curve was characterized by a correlation coefficient of 0.998.Signals corresponding to 10 replicates of a real sample presented an RSD of < 3%. The calculated limit of detection (3s) of Cu was 5 mg l21. Fig. 1 Flow system for liquid–liquid extraction and timing diagram for the injector and valves. In the flow diagram, the three rectangles at the left represent the injector in the sampling position. S = aspirated sample at 3.9 ml min21; L = 300 cm (1.5 ml) sampling loop; C1 = 0.25 m HClO4 sample carrier stream (1.8 ml min21); C2 = 5% m/v EDTA solution of pH 8–9 (1.8 ml min21); C3 = 0.5% m/v NaDDTC solution (0.6 ml min21); C4 = CCl4 (0.32 ml min21); x and y = confluence points; MC = 75 cm coil; EC = 250 cm coil; V1–V4 = solenoid valves; PS = phase separator; D = spectrophotometer set at 436 nm; W = waste; and Re = reservoir.The arrows on the lines indicate the pumping direction and the large arrow below the injector represents the movement of the central part to the alternative position. In the time diagram, the shaded area in the first and second lines indicates the injector resting time in the sampling and injection position, respectively. For the other lines, the shaded area represents the delays for valves being switched on.Valves V1–V4 are switched on to perform the stages of the process indicated on the right. 192 Analyst, February 1998, Vol. 123Results for the analysis of several plant samples including five independent digests of NIST SRM 1572 Citrus Leaves are presented in Table 1.The results were in agreement with the certified value at the 95% confidence level as verified by a ttest. Conclusions A flow injection procedure for the liquid–liquid extraction of the DDTC–Cu complex in CCl4 was demonstrated by analysing real samples of plant materials. The synchronised filling and draining of the proposed phase separator allowed an enrichment factor of 10 to be attained, using a minimum volume of solvent and an efficient sample throughput without memory effects.Accurate performance was achieved using computer-controlled three-way solenoid valves combined with an electronically controlled injector. This work was supported by the Fundaç�ao de Amparo `a Pesquisa do Estado de S�ao Paulo (FAPESP) and a scholarship from the Conselho Nacional de Desenvolvimento Cient�ýfico e Tecnol�ogico (CNPq). References 1 de la Guardia, M., and Ruzicka, J., Analyst, 1995, 120, 17N. 2 Reis, B. F., Gin�e, M. F., Zagatto, E. A. G., Lima, J. L. F. C., and Lapa, R. A., Anal. Chim. Acta, 1994, 293, 129. 3 Bouhsain, Z., Garrigues, S., and de la Guardia, M., Analyst, 1997, 122, 441. 4 Fang, Z., Flow Injection Separation and Preconcentration, VCH, New York, 1993. 5 Karlberg, B., and Thelander, S., Anal. Chim. Acta, 1978, 98, 1. 6 Bergamin Fo., H., Medeiros, J. X., Reis, B. F., and Zagatto, E. A. G., Anal. Chim. Acta, 1978, 101, 9. 7 Kawase, J., Nakae, A., and Yamanaka, M., Anal.Chem., 1979, 51, 1640. 8 Lucy, C. A., and Yeung, K. K. C., Anal. Chem., 1994, 66, 2220. 9 Luo, Y., Al-Othman, R., Ruzicka, J., and Christian, G. D., Analyst, 1996, 121, 601. 10 Zagatto, E. A. G., Krug, F. J., Bergamin Fo., H., and Jørgensen, S. S., in Flow Injection Atomic Spectroscopy, ed. Burguera, J. L., Marcel Dekker, New York, 1989, ch. 6, pp. 225–257. 11 Marczenko, Z., Separation and Spectrophotometric Determination of Elements, Ellis Horwood, Chichester, 2nd edn., 1986, ch. 19, pp. 257–271. Paper 7/05219G Received July 21, 1997 Accepted October 2, 1997 Fig. 2 Recorder tracing for Cu determination in plant digests. From right to left, the peaks in triplicate correspond to standard solutions of 0, 50, 100, 200, 300 and 400 mg l21 of Cu followed by 10 signals obtained by processing a sample solution, triplicate signals of four different samples and another series of standard solutions. Table 1 Copper content in different plant materials and NIST SRM 1572 Citrus Leaves (data on a dry basis; n = 5) Concentration found/ Sample mg g21 Bean leaves 14.4 ± 0.58 Maize leaves 13.9 ± 0.75 Soybean leaves 16.9 ± 0.81 Cotton leaves 12.7 ± 1.00 Coffee leaves 25.2 ± 1.53 NIST SRM 1572 Citrus Leaves 15.8 ± 0.88* * Certified Cu content 16.5 mg g21. tcalc = 1.50; ttabulated = 2.78. Analyst, February 1998, Vo
ISSN:0003-2654
DOI:10.1039/a705219g
出版商:RSC
年代:1998
数据来源: RSC
|
7. |
Use of solid-phase extraction in the determination of benzene, toluene, ethylbenzene, xylene and cumene in spiked soil and investigation of soil spiking methods |
|
Analyst,
Volume 123,
Issue 2,
1998,
Page 195-200
Katie M. Meney,
Preview
|
PDF (73KB)
|
|
摘要:
Use of solid-phase extraction in the determination of benzene, toluene, ethylbenzene, xylene and cumene in spiked soil and investigation of soil spiking methods Katie M. Meney, Christine M. Davidson* and David Littlejohn Department of Pure and Applied Chemistry, University of Strathclyde, 295 Cathedral Street, Glasgow, UK G1 1XL A method has been developed for the determination of benzene, toluene, ethylbenzene, xylene and cumene (BTEXC) in soils, based on methanol extraction, solid-phase extraction (SPE) of the diluted extract and gas chromatography. Quantitative recoveries of BTEXC were obtained from methanol extracts provided the solvent composition was adjusted to methanol–water (50 + 50) prior to SPE, and care was taken to avoid the development of headspace into which analytes could partition.Cartridge (500 mg) load volumes of up to 20 ml of methanol–water extract were possible for all the analytes, except benzene (7 ml), without significant loss due to volatilization.The minimum elution volume for 100% removal of the analytes from the SPE cartridge was 1.5 ml of dichloromethane. It was possible to recover > 90% of analytes added as a concentrated methanolic solution to a dry, clay soil, but the recoveries decreased if field-moist soil was used and if the soil was spiked with petrol. Recoveries were also reduced if the soil and spiking solution were left in contact for extended periods (as would occur in the event of a real contaminant spillage). Over a 17 d period, more than 30% of the BTEXC added to a soil as a dilute solution in methanol–water (50 + 50) became too tightly bound for removal by a single aliquot of extractant.When the method of vapour fortification was used to produce performance evaluation materials, both uptake of BTEXC and stability of the analyte concentrations after spiking were found to depend strongly on the soil type. Keywords: Solvent extraction; solid-phase extraction; gas chromatography; contaminated soil; benzene; toluene; ethylbenzene; xylene; cumene; vapour fortification Volatile organic compounds (VOCs), such as benzene, toluene, ethylbenzene, xylene and cumene (BTEXC), are important environmental contaminants because of their toxicity and widespread occurrence.They are present in aviation fuel and petrol (gasoline) and are widely used industrial solvents and raw materials.1 Benzene, toluene and ethylbenzene are amongst the compounds designated ‘priority pollutants’ by the US EPA and action levels for BTEX are listed in the Dutch Government quality standards for assessment of land contamination.2 Soils become contaminated with BTEXC through spillage of industrial solvents, leakage of petrol from storage facilities (particularly underground holding tanks) and deposition from contaminated air.1 Most methods for contamination assessment are based on the extraction of VOCs from soil or sediment for quantification by GC or GC–MS.A wide variety of extraction techniques have been used, including headspace analysis,3,4 purge-and-trap analysis,5 thermal desorption,6 vacuum distillation, 7 solvent extraction (including Soxhlet extraction, use of sonication8,9 and microwave assistance10) and supercritical fluid extraction.11 Apart from solvent extraction, all of these methods suffer from the limitation that only small (typically < 5 g) and potentially unrepresentative samples can be processed.An exception is the novel apparatus for headspace analysis of bulk sediment samples (up to 150 g) described by Bianchi and Varney.12 An important drawback of solvent extraction is that relatively large quantities of liquid are required. A high extractant-to-soil ratio reduces the analytical sensitivity unless the extract is later concentrated. Furthermore, environmental considerations, including costs of disposal, make the large-scale use of many organic solvents unacceptable.Solid-phase extraction (SPE) has been applied widely in clean-up and/or preconcentration of liquid samples and extracts. 13 Cartridges containing C18 sorbent have been used in the isolation of benzene and toluene from sea-water14 and of BTEX from acidic industrial effluents.15 Solid-phase micro-extraction (SPME) has been applied to isolate BTEX from water samples, either by direct sorption from the liquid16,17 or via sampling of headspace.18 SPE has also found applications in soil analysis. Both Redondo et al.19 and Huang20 used SPE to collect a suite of pesticides from soil extracts, and Mills and Thurman21 isolated triazine metabolites from soil and aquifer sediments using methanol–water (4 + 1) and automated SPE.Few SPE methods for the collection of VOCs from soil have been reported, however. Loconto22 used a multi-stage SPE procedure to isolate 1,4-dichlorobenzene and 1,2,4-trichlorobenzene from spiked soils, but the recoveries were poor (5.8 and 26.1%, respectively).Headspace SPME has been successfully applied to spiked soils,23,24 but it is not yet clear whether the method can be used in quantitative analyses for VOCs in real soil matrices containing variable proportions of organic matter. Soil sampling is a critical factor in VOC analysis, with large (up to 100%) negative biases being reported owing to loss of analytes through vaporization during sampling, transport and storage.25 Considerable evidence exists to support the need for immediate, on-site immersion of soil samples in a stabilizing solvent such as methanol,26 but this is not yet common practice because it lowers the sensitivity and also limits the range of extraction techniques which can be applied to the sample once returned to the laboratory. In this paper, a method is described based on the approach recommended by Bone27 in which methanol added to contaminated soil in the field serves as both the sample preservation medium and the extractant.Methanol extracts were diluted with water and processed by SPE, prior to quantification of BTEXC by GC. The potential to improve sensitivity via preconcentration during SPE was investigated and the influence on analyte recoveries of soil type, moisture content, spiking method and length of contact between analytes and the soil matrix were studied. Experimental Apparatus A high-resolution gas chromatograph from Carlo Erba (Milan, Italy) capable of split, splitless and on-column injection was used with a flame ionization detector.The column used was a Analyst, February 1998, Vol. 123 (195–200) 195CP-SIL 8 fused-silica capillary, 25 m 3 0.32 mm id, 5 mm film thickness (Chrompack, Middelburg, The Netherlands). Integration was performed with a PU 4810 integrator (ATI Unicam, Cambridge, UK). The temperature programme and conditions were as follows: 100 °C for 5 min, then a ramp at 4 °C min21 to 180 °C; injection port temperature 225 °C; detector temperature, 225 °C; carrier gas (helium) flow rate, 1.8 ml min21; and splitting ratio, 45 : 1.SPE was performed with 500 mg C18 Bond Elut cartridges (Analytichem/Varian, Harbor City, CA, USA). Reagents Methanol (HPLC grade), ethylbenzene and dichloromethane (technical grade) were obtained from Fisons (Loughborough, Leics., UK). Toluene, benzene and cumene (AnalaR) were obtained from Merck (Poole, Dorset, UK) and p-xylene (technical grade) from Aldrich (Gillingham, Dorset, UK).The soils used in method development were a commercial peat, two typical topsoils (a sandy loam and a clay soil) and a ‘made ground’ sample (a mixture of topsoil, stones and clinker) obtained from a former industrial site. They either were used in a field-moist condition or were air dried at room temperature and sieved through a 2 mm mesh. Soil characteristics are summarized in Table 1. None of the soils was intrinsically contaminated with detectable quantities of the analytes.Unleaded petrol was obtained from a commercial filling station. Procedures Solution preparation A solution of dichlorobenzene in dichloromethane (2000 mg ml21) was used as an external standard for quantification of the analytes in the SPE extracts by GC. Standard solutions (approximately 150 mg ml21 of each analyte) were prepared by weighing accurately (by difference) benzene, toluene, ethylbenzene, p-xylene, cumene and dichlorobenzene into a calibrated flask containing dichloromethane and then diluting to volume with further solvent.A concentrated stock standard solution (containing about 6000 mg ml21 of each of the analytes in methanol) was also prepared. Aliquots of this solution were diluted with water or methanol to obtain the working standard solutions used to spike soils and during SPE method development. Headspace minimization Dilute stock standard solutions were stored for a minimum time in calibrated flasks to prevent loss of analytes through partitioning into the headspace.Once prepared, solutions used in SPE method development were immediately taken up in replicate 5 ml glass syringes, the tips of the which were then sealed with Parafilm. If this procedure was not followed it was found that significant (up to 40%) loss of analytes occurred. Syringes containing aliquots of soil extracts were also sealed in this way prior to SPE. Solvent extraction Soils (25 or 50 g) spiked with solutions of the analytes or with petrol were extracted in 100 ml centrifuge tubes.Vapour fortified soils (15 g) were extracted in 50 ml centrifuge tubes. Methanol–water (50 + 50) was used to extract soils which had been spiked with solutions of the analytes, whereas methanol alone was used for soils spiked or vapour fortified with petrol. Polyethylene centrifuge tubes were used in this study, although for some applications glass tubes would be preferable, because their caps could easily be modified to include a septum.When the extractant was added to the soil, care was taken to fill the tube completely with liquid (allowing air to escape by piercing the septum with a needle) and prevent the development of headspace into which the analytes could partition. The volume of extractant added depended on the mass of soil. For 25 and 50 g of soil in the 100 ml tubes, the volumes of extractant were about 75 and 50 ml, respectively; about 35 ml of extractant were added to the 15 g of soil in the 50 ml tube.The exact volume added to each tube was noted to permit the determination of analyte concentrations. The centrifuge tubes were shaken for 1 h on an end-over-end shaker and then centrifuged [2000 rpm; MSE (Loughborough, UK) Mistral 1000 bench centrifuge] and aliquots of the supernatant (generally 3 ml) were taken up in 5 or 20 ml glass syringes for loading on to SPE cartridges. Solid-phase extraction The procedure was based on a method for the extraction of volatile organic compounds from acidic industrial effluents.15 A C18 SPE cartridge was washed with 3 ml of methanol then conditioned with 3 ml of a 1% solution of methanol in water.The standard solutions or soil extracts (3 ml unless stated otherwise) were loaded and the analytes eluted, normally in 3 ml of dichloromethane. A 100 ml aliquot of dichlorobenzene external standard solution was then added to the eluate, which was analysed by GC.Vapour fortification The procedure was derived from that described by Hewitt.28 Multiple 15 g aliquots of soil were weighed into 20 ml glass vials and placed (uncapped) in a 400 ml glass desiccator, along with 10 ml of petrol in a shallow Petri dish. The desiccator was sealed and the soils were exposed to the vapours from the petrol for 14 days. At the end of this period, the soils were transferred to amber-glass bottles with PTFE-lined caps, which were filled to capacity to minimize the headspace.Results and discussion Optimization and characterization of solid-phase extraction procedure Before applying SPE to soil extracts, optimum conditions for the retention of the analytes on the C18 sorbent were determined. Mixed standard solutions (3 ml), containing 70 mg ml21 of each analyte, but with solvent composition varying between 100% water and 100% methanol, were passed through SPE cartridges. The best performance, i.e., close to quantitative recoveries for all analytes, was obtained with methanol–water mixture (50 + 50) (Fig. 1). A higher proportion of methanol [methanol–water (45 + 65) or (40 + 60)] gave analyte recoveries > 100%. The poorer recoveries obtained at < 40% methanol may reflect the low solubilities of the analytes in water, which could prevent quantitative transfer of the analytes to the SPE cartridge. Higher methanol concentrations (75%) led to enhanced recoveries of the less volatile analytes but losses of Table 1 Soil characteristics Parameter Sandy loam Clay Peat Made-ground Moisture content (%)a 14–20 18–27 52–66 19 Loss on ignition (%)b 13 16 32 22.7 pH 5.5 5.3 3.1 6.0 a Obtained by drying at 105 °C overnight; a range of values indicates soil obtained on different sampling trips.b Obtained by ashing at 450 °C overnight. 196 Analyst, February 1998, Vol. 123benzene and toluene. At 100% methanol, the recoveries were uniformly low ( < 30%), perhaps because of poor retention of the analytes by the C18 sorbent when methanol is used as the solvent.Therefore, when soil samples are preserved by field immersion in methanol, it will be necessary to dilute the extracts with water prior to SPE. These recoveries are an improvement on those obtained for extraction of benzene and toluene from sea-water by Saner et al.14 (22 ± 4 and 85 ± 6%, respectively, n = 6), and those achieved for an aqueous standard, with an internally cooled SPME device, by Zhang and Pawliszyn23 (benzene 42 ± 18, toluene 72 ± 8, ethylbenzene 85 ± 6, xylene about 98 ± 5%, n = 3).However, they are similar to those reported for toluene, chlorobenzene and xylene isomers (100 ± 4, 104 ± 5 and 98 ± 5%, respectively, n = 3) when the minimal headspace procedure used in the current study was applied to industrial effluent samples.15 Quantitative recoveries were also achieved when 3 ml aliquots of mixed standard solutions containing up to 650 mg ml21 of each analyte in a methanol–water (50 : 50) matrix were processed. This is equivalent to the concentration which would be obtained by extracting completely 25 g of a soil containing 2070 mg g21 of each analyte with 75 ml of methanol– water.For comparison, the current Dutch ‘C’ (intervention) values for benzene, toluene, ethylbenzene and xylene are 5, 50, 30 and 50 mg kg21, respectively.2 The limit of detection for the same mass of soil and extractant volume was calculated to be around 0.7 mg kg21.To assess the potential of the SPE method for analyte preconcentration, the maximum loading volume and minimum elution volume of the cartridge were investigated. Replicate 3 ml aliquots of mixed standard solutions containing 60 mg ml21 of each analyte were diluted by drawing additional methanol– water (50 + 50) into glass syringes, to give final volumes of 7, 10, 12, 15, 17 and 20 ml. No significant loss of toluene, ethylbenzene, p-xylene or cumene occurred when these solutions were processed by SPE, but benzene recoveries were reduced ( < 60%) when volumes greater than 7 ml were loaded.As might be expected, an important limitation of the SPE method appeared to be loss of the most volatile analyte during loading of the cartridge. When replicate cartridges loaded with 3 ml of the same mixed standard were eluted with different volumes of dichloromethane, 1.5 ml was adequate to remove all analytes.Hence preconcentration factors of approximately 35 and 313 could be obtained for benzene and the other analytes, respectively, without adversely affecting the efficiency of the extraction. When mixed standard solutions [70 mg ml21 of each analyte in methanol–water (50 + 50)] were prepared from the 6000 mg ml21 methanolic stock solution and extracted in duplicate on five successive days, overall recoveries of 92–102% were obtained (RSD < 6%, n = 10). Investigation of soil spiking methods It is common to validate extraction methods via analysis of certified reference materials or the use of ‘spike and recovery’ tests.Unfortunately, no soil reference materials certified for VOC content are available28 and so soils spiked using a variety of methods were investigated using the SPE method developed. Spiking with a concentrated methanol solution Duplicate 25 g aliquots of dry clay soil were spiked with 1 ml of a methanol solution containing 6000 mg ml21 of each analyte and were extracted immediately.The extractant used was methanol–water (50 + 50), since this had been found to be optimal for SPE. Except for benzene in one sample, recoveries of > 90% were obtained (Table 2). However, up to 40% of the benzene and up to 23% of the other analytes were lost when field-moist samples (moisture content 17.5%) were used. The presence of the aqueous phase may limit analyte access to binding sites in the soil matrix, resulting in enhanced loss of, in particular, the most volatile analyte.Spiking with methanol–water (50 + 50) solution A solution (75 ml) with the optimum solvent composition for SPE [methanol–water (50 + 50)], containing about 60 mg ml21 of each analyte, was added to 25 g of field-moist clay soil in a centrifuge tube, taking care to fill the tube completely to prevent the development of headspace. Prior to addition of the BTEXC solution, a small volume of methanol was added to the tube to compensate for the soil moisture initially present and maintain a constant solvent composition.Quantitative recoveries could be obtained provided that the mixture was immediately centrifuged and an aliquot of the supernatant removed for SPE. However, if the soil and solution were shaken together, on an end-over-end shaker, for 1 h before centrifugation, losses occurred. Analyte recoveries were benzene 92, toluene 89, ethylbenzene 87, p-xylene 87, and cumene 89%. Recoveries of 96–98% were obtained when a tube containing the BTEXC solution but no soil was treated similarly, suggesting that losses are due to analyte–soil interactions, and not to volatilization or sorption on the centrifuge tube walls.To investigate this further, samples of three soils were spiked with the same methanol–water solution and left to stand, in sealed centrifuge tubes with no headspace, for various periods. At the end of the contact time, aliquots of the supernatant were removed for analysis, and each soil residue was extracted with a second volume of methanol–water (50 + 50).The contents of blank tubes, containing the methanolic solution but no soil, were also analysed to assess losses due to vaporization or adsorption of the analytes by the tube walls during the experiment. These values were used to calculate the amounts of Fig. 1 Recoveries of benzene, ethylbenzene and cumene from a standard solution using the minimum headspace SPE procedure.Table 2 Recoveries (%) obtained following spiking clay soil with a concentrated methanol solution of the analytes Compound Dry soil (n = 2) Field-moist soil (n = 2) Benzene 88, 95 64, 60 Toluene 103, 99 84, 77 Ethylbenzene 94, 96 84, 76 p-Xylene 92, 98 84, 78 Cumene 93, 97 89, 83 Analyst, February 1998, Vol. 123 197analyte adsorbed by the soils over periods of 6 and 17 d (Table 3). After 6 d, the peat had adsorbed the largest amounts of analytes, probably owing to the larger proportion of organic matter present.Non-polar compounds are well known to have an affinity for the organic fraction of moist soil.1 When separate tubes were analysed after 17 d, results for the peat showed little change, but further analyte sorption was evident in both clay and sandy soils. At the end of the experiment, up to 38% of the analytes added to the peat and up to 30% of those added to the other soils could neither be recovered nor accounted for by analysis of the blank.It must be presumed, therefore, that these had become too tightly bound to the soil matrix for removal with a single aliquot of methanol–water (50 + 50). Losses from the blank tube were of the order of 20% over 17 d, and were greater for the less volatile analytes. This suggests that partitioning into the polypropylene walls, rather than volatilization, was the main mechanism responsible for analyte loss in the absence of soil. Similar contaminant ageing experiments were carried out over a longer period.After 10 weeks, < 50% of the benzene and toluene and < 30% of the other analytes remained in the spiking solution. In addition to analysis of the supernatant, four consecutive extractions were performed on each soil residue, and progressively smaller amounts of the analytes were recovered by each subsequent extraction (Fig. 2). When the levels of analytes which could be recovered via the four extractions were added to those remaining in the supernatant at the end of the contact period and corrected for blank tube losses, recoveries > 100% were obtained for some soils (Table 4).A possible explanation is overestimation of the blank, which could occur if the soil competed effectively with the tube walls for adsorption of the analytes, but retained them in forms which could be removed by the methanol–water extractant. The madeground sample appeared less able to bind these analytes than the true soils.This is probably due to the unusual nature of this matrix, which consisted of a mixture of clinker and topsoil (sandy loam and sandy silt loam). Alexander29 questioned the validity of ‘spike-and-recover’ experiments for the assessment of soil extraction methods on the grounds that they do not simulate accurately the manner in which contamination occurs in the environment. When a solvent or petrol spillage occurs, the soil and VOC often have a considerable amount of time to interact.Under these conditions, ‘ageing’ effect (in which substances are gradually altered to more stable forms or diffuse into inaccessible locations, within the soil matrix) can become important. This effect is of particular concern in contaminated land clean-up, where the presence of recalcitrant forms of a pollutant may make it impossible to achieve adequate soil remediation. It is also important in analytical science. Although quantitative recoveries may be obtained when a soil is spiked and promptly extracted, use of the same method could seriously underestimate contamination in a field sample.30 Spiking with petrol Spillage of petrol is a common source of BTEXC in the environment, and it is important that any extraction method be applicable to petrol-contaminated soil.However, when 500 ml aliquots of unleaded petrol (of known composition) were added to 25 g samples of a field-moist clay soil (moisture content 27%), difficulties were experienced in recovering the compounds of interest using methanol–water (50 + 50).On addition of the extractant, a strong smell of petrol was immediately apparent and immiscible droplets could be seen on the surface of the liquid. A tarry residue remained in the centrifuge tube at the end of the extraction. Recoveries of the less volatile analytes were slightly reduced relative to those obtained when spiking with methanolic solutions (Table 5), presumably because of the presence of high concentrations of non-polar components in petrol, which compete effectively with the extractant for the Table 3 Concentrations of analytes adsorbed by soils from methanol–water (50 + 50) solutions following various contact times (mg g21 soil dry mass) Sandy loama Claya Peatb Compound 6 d 17 d 6 d 17 d 6 d 17 d Benzene 3 10 2 8 19 16 Toluene 7 18 9 17 41 37 Ethylbenzene 16 24 19 24 59 55 p-Xylene 10 17 16 20 53 51 Cumene 13 13 13 12 52 38 a Spiking conditions: 75 ml of 52 mg ml21 (of each analyte) solution added to 50 g of soil. b Spiking conditions: 73 ml of 49 mg ml21 (of each analyte) solution added to 25 g of soil.Fig. 2 Analyte recoveries obtained by successive extraction of a clay soil spiked with a methanol–water (50 + 50) solution and left to age for 10 weeks. Table 4 Overall recoveries (%) of analytes from soil and made-ground following contact with a methanol–water (50 + 50) solution of BTEXC for 10 weeks. Results shown are blank corrected and represent the sum of analyte levels in the supernatant + amounts released by four consecutive extractions of the soil versus the spike concentration Compound Sandy loam Clay Made-ground Benzene 120 120 73 Toluene 110 103 60 Ethylbenzene 119 113 78 p-Xylene 135 124 74 Cumene 138 133 82 Table 5 Recoveries (%) obtained with different extractants following spiking of field-moist clay soil with petrol Methanol–water Compound (50 + 50) (n = 2) Methanol (n = 2) Benzene 58, 59 81, 86 Toluene 88, 84 92, 95 Ethylbenzene 76, 72 90, 92 p-Xylene 83, 74 93, 95 Cumene 74, 64 94, 98 198 Analyst, February 1998, Vol. 123analytes. Improved recoveries could be obtained if the soil was extracted with methanol alone, although there was still significant (up to 19%) loss of benzene. To achieve optimum solvent composition for the SPE procedure, a 3 ml aliquot of the methanol extract was drawn into a glass syringe and then diluted by drawing up 3 ml water.In order to simulate more closely the manner in which soils become contaminated in the field, two large masses of soil (5 kg each of the sandy loam and the clay) were deliberately contaminated with unleaded petrol (200 ml). Each was thoroughly mixed after spiking, then transferred to a container, covered with turf and placed in the open for 8 weeks. Periodically, soil (about 80 g) was removed from each quarter of the container, quickly mixed to give a bulk sample, then four 50 g aliquots were taken for extraction.The amounts of analytes extracted from the sandy loam in general decreased with time but no similar decrease was apparent in the clay soil. In both experiments the reproducibility was exceedingly poor (RSD up to 100% for n = 4). It was difficult to spike the soils homogeneously and mixing of the bulk samples had to be minimized to avoid loss of analytes via vaporization. Where extractable analyte levels decreased, it was not possible to determine whether this was due to conversion to recalcitrant species or to vaporization, and the initial spiking level could not be determined accurately since evaporation of the petrol occurred during the procedure.Hence, although spiking with petrol represents a close approximation of field contamination, it proved too complex a system in which to evaluate the soil extraction methodology. Vapour fortification with petrol Vapour fortification is a spiking method developed by Hewitt.28 Soils are exposed to VOC vapour, in a closed container, which simulates closely the manner in which contamination often occurs in the vadose zone.The absolute level of analytes taken up by the soil is unknown, but vapour fortification can produce homogeneously spiked soils which are stable for over 60 d in sealed vials, and can therefore be used in inter-laboratory trials or to compare the performance of different extraction methods. 31 In this work, three soils (the sandy loam, clay soil and peat) were vapour fortified and used to study interactions between BTEXC and the soil matrix and to evaluate the effects of these on extraction efficiency.All extractions were performed with 100% methanol, in preference to methanol–water (50 + 50) since higher recoveries were obtained (Table 6). The three types of soil differed in their ability to absorb BTEXC when exposed to similar amounts of petrol, and in their stability after fortification.At the end of a 14 d fortification of dry soil samples, the highest levels of all the analytes were recovered from the peat and the lowest from the sandy soil (Table 7). As noted previously for spiking with methanol–water (50 + 50) solutions, a high organic matter content appears to aid rapid sorption of the analytes. In contrast to the work of Hewitt28 and Hewitt and Grant,31 these spiked soils were not stable in terms of their extractable analyte content. Significant losses (up to 27% for benzene, 17% for the other analytes) occurred on storage for only 7 d and, after 42 d, < 50% of the original extractable benzene remained in each soil.Stability varied between soils and between analytes. Recoveries for the more volatile compounds decreased more rapidly than for those with higher boiling points, suggesting that volatilization (rather than conversion to non-extractable forms) was occurring. Over time, the vapour fortified sand retained a higher proportion of the analytes initially present than the clay soil, and the lowest recoveries were obtained for the peat.Hence, although the peat soil was initially able to take up the highest amounts of analytes, these were more easily lost. Reversible binding of the analytes to organic matter may explain the difference in stability observed for soils vapour fortified in the present work and those prepared by Hewitt and Grant,31 who used material of low organic carbon content ( < 7%).31 Overall, however, the repeatability of the vapour fortification method was much better than for conventional (bulk) spiking with liquid petrol (RSD < 7% for sandy loam and clay, < 15% for peat).Over time, however, the precision degraded (RSD up to 44% after 42 d). As might be expected, this was most marked for the more volatile analytes and for the peaty soil. When the experiment was repeated with field-moist soil, the peat, again, was found to contain the highest levels of extractable analytes immediately after vapour fortification (Table 7).The recoveries decreased with time and, in general, were poorest for the more volatile analytes. An exception was Table 7 Concentrations of analytes recovered from vapour fortified soils after storage for various periods (mg g21 soil dry mass). Values in parentheses are RSDs for n = 6 (i.e., extraction of triplicate soil samples and duplicate analysis of each extract) Sandy loam Clay soil Peat Compound day 1 day 7 day 14 day 42 day 1 day 7 day 14 day 42 day 1 day 7 day 14 day 42 Dry soil— Benzene 243 (7) 188 (4) 168 (8) 107 (26) 318 (6) 232 (11) 255 (2) 109 (20) 348 (9) 286 (10) 268 (6) 98 (44) Toluene 868 (4) 761 (4) 744 (5) 601 (14) 1166 (3) 973 (8) 937 (4) 779 (10) 1651 (11) 1379 (4) 1283 (2) 905 (32) Ethylbenzene 348 (5) 326 (4) 317 (6) 301 (9) 480 (4) 417 (11) 379 (4) 320 (3) 657 (14) 550 (8) 514 (4) 446 (23) p-Xylene 534 (4) 500 (4) 460 (11) 480 (9) 725 (4) 635 (9) 572 (4) 579 (4) 1072 (14) 887 (7) 841 (47) 804 (21) Cumene 656 (5) 621 (5) 571 (11) 590 (9) 878 (4) 766 (11) 690 (5) 715 (2) 1287 (14) 1077 (9) 1010 (4.3) 956 (19) Moist soil— Benzene 69 (7) 38 (10) 27 (23) 66 (4) 56 (11) 21 (3) 272 (29) 158 (23) 68 (10) Toluene 175 (6) 81 (9) 75 (6) 255 (6) 165 (10) 127 (9) 1125 (25) 885 (17) 486 (9) Ethylbenzene 175 (7) 43 (10) 36 (7) 94 (6) 72 (9) 57 (7) 411 (22) 435 (10) 285 (8) p-Xylene 94 (5) 81 (11) 64 (8) 157 (5) 133 (9) 105 (6) 895 (16) 775 (9) 547 (5) Cumene 102 (5) 87 (11) 73 (9) 165 (6) 155 (11) 122 (4) 988 (14) 895 (7) 684 (5) Table 6 Concentrations of analytes recovered from vapour fortified dry clay soil on extraction with different solvents (mg g21 soil dry mass). Results are for soils vapour fortified at the same time and in the same desiccator Methanol–water Compound (50 + 50) (n = 2) Methanol (n = 2) Benzene 206, 212 308, 311 Toluene 647, 666 954, 966 Ethylbenzene 157, 169 280, 297 p-Xylene 240, 254 419, 448 Cumene 171, 180 362, 389 Analyst, February 1998, Vol. 123 199the marked loss of ethylbenzene from the sandy loam over 42 d. The soils which were vapour fortified in field-moist conditions were less able to take up and retain BTEXC over the study period than those which were previously air dried. Soil behaves as a dual sorbent towards VOC vapour.32 Both partitioning into organic matter and sorption on mineral matter can occur, and both processes are influenced by the presence of moisture.At high relative humidity, sorbed organic vapours are displaced from the mineral fraction by water and only uptake by soil organic matter occurs. Hence, in the presence of moisture, less VOC can be taken up. Further, if, as suggested above, binding by organic matter is relatively weak in the soils examined, then moist soils would be expected to have a lower capacity to retain BTEXC once sorbed. Finally, a dry vapour fortified clay was extracted with four successive aliquots of methanol, each of which was diluted to methanol–water (50 + 50), processed by SPE and analysed separately.Of the total amounts of analytes which could be removed by the four extractions, the largest proportions (68–74%) were found in the first extract and significant levels (19–25%) in the second. Little additional BTEXC was removed by the third and fourth treatments (about 5% and < 2%, respectively). Since the initial spiking level was unknown, it was not possible to determine whether the soil had been exhaustively extracted.However, it is clear that a single methanol extraction would underestimate soil contamination, as observed for soil spiked with BTEXC in methanol–water (50 + 50). Conclusions BTEXC can be recovered quantitatively and with good precision from methanol extracts of soil by SPE, provided the extract is first diluted to methanol–water (50 + 50). Analyte preconcentration is possible and the cartridge capacity is sufficient to allow analysis at levels well above legislative trigger concentrations for soil contamination.The method can also be used to quantify analytes at sub-mg kg21 levels. A number of difficulties were encountered when attempts were made to produce performance evaluation materials to test the method developed. Quantitative recoveries of all analytes except benzene could be obtained when soils spiked with methanolic or methanol–water (50 + 50) solutions, or with petrol, were extracted immediately.However, the extraction efficiencies were lowered, for methanol-spiked soils, by the presence of moisture and, more important, were observed to depend on the length of time for which the soil and analytes had been in contact. The existence of such ‘ageing effects’ cast serious doubt on the use of spike-and-(prompt)-recovery procedures for validation of methods to extract VOCs from soil. However, preparation of aged reference materials was limited by the volatility of analytes, and the use of blanks to quantify and correct for losses also presented difficulties since BTEXC did not behave in a similar way in the presence and absence of soil.Studies with a methanol–water (50 + 50) spike and with petrol vapour fortification showed that soils differed markedly in their ability to take up VOCs, and also indicated that a single solvent extraction could significantly underestimate contamination. Vapour fortified soils were found to be unstable over periods of only a few days, and the stability was both soil and analyte dependent.The study suggests that none of the spiking methods used is entirely satisfactory for these analytes and further work is necessary to develop alternative approaches. Vapour fortification is probably the best of the methods investigated, provided that soils are extracted soon after exposure, because it most closely simulates the manner in which contamination occurs in the field.The use of spike-and-(prompt)-recovery procedures produces data of little value and is to be discouraged. K.M.M. acknowledges the financial support of ICI plc and the EPSRC under the CASE studentship scheme. References 1 Kliest, J. J. G., in Chemistry and Analysis of VOCs in the Environment, ed. Bloemen, H. J. Th., and Burn, J., Blackie, Glasgow, 1993, pp. 202–236. 2 Environment Quality Standards for Soil and Water, Netherlands Ministry of Housing, Physical Planning and Environment, Leidschendam, 1986 and 1991. 3 Roe, V. D., Lacy, M. J., Stuart, J. D., and Robbins, G. A., Anal. Chem., 1989, 61, 2584. 4 Kolb, B., LC–GC Int., 1995, 8, 512. 5 Yan, X., Carney, K. R., and Overton, E. B., J. Chromatogr. Sci., 1992, 30, 491. 6 Moreton, E. P., Walsh, P. R., and Lawlor, L. J., Groundwater Manage., 1991, 8, 75. 7 Hiatt, M. H., Youngman, D. R., and Donnelly, J. R., Anal. Chem., 1994, 66, 905. 8 Donaldson, S. G., Miller, G. C., and Miller, W. W., J. Assoc. Off.Anal. Chem., 1990, 73, 306. 9 US Environmental Protection Agency, Test Methods for Evaluating Sold Waste, Report No. SW-846, EPA, Washington, DC, 1986. 10 Poole, C. F., and Poole, S. K., Anal. Commun., 1996, 33, 11H. 11 Burford, M. D., Hawthorne, S. B., and Miller, D. J., J. Chromatogr. A, 1994, 685, 95. 12 Bianchi, A., and Varney, M. S., Analyst, 1989, 114, 47. 13 Berrueta, L. A., Gallo, B., and Vicente, F., Chromatographia, 1995, 40, 474. 14 Saner, W. A., Jadamec, J. R., Sager, R. W., and Killeen, T. J., Anal. Chem., 1979, 51, 2180. 15 Deans, I. S., Davidson, C. M., Littlejohn, D., and Brown, I., Analyst, 1993, 118, 1375. 16 Arthur, C. L., and Pawliszyn, J., Anal. Chem., 1990, 62, 2145. 17 Arthur, C. L., Pratt, K., Motlagh, S., Pawliszyn, J., and Belardi, R. P., J. High. Res. Chromatogr., 1992, 15, 741. 18 Zhang, Z., and Pawliszyn, J., Anal. Chem., 1993, 65, 1843. 19 Redondo, M. J., Ruiz, M. J., Boluda, R., and Font, G., Chromatographia, 1993, 36, 187. 20 Huang, L. Q., J. Assoc. Off. Anal. Chem., 1989, 72, 349. 21 Mills, M. S., and Thurman, E. M., Anal. Chem., 1992, 64, 1985. 22 Loconto, P., LC–GC Int., 1991, 4, 10. 23 Zhang, Z., and Pawliszyn, J., Anal. Chem., 1995, 67, 34. 24 Fromberg, A., Nilsson, T., Larsen, B. R., Montanarella, L., Facchetti, S., and Madsen, J. O., J Chromatogr., 1996, 746, 71. 25 Siegrist, R. L., and Jenssen, P. D., Environ. Sci. Technol., 1990, 24, 1387. 26 Smith, J. M., Eng, L., Comeau, J., Rose, C., Schulte, R. M., Barcelona, M. J., Klopp, K., Pilgrim, M. J., Minnich, M., Feenstra, S., Urban, M. J., Moore, M. B., Maskarinee, M. P., Siegrist, R., Parr, J., and Claff, R. E., in Principles of Environmental Sampling, ed. Keith, L. H., Americal Chemical Society, Washington, DC, 3rd edn., 1996, ch. 34, p. 693. 27 Bone, L. I., in Principles of Environmental Sampling, ed. Keith, L. H., Americal Chemical Society, Washington, DC, 3rd edn., 1996, ch. 37, p. 737. 28 Hewitt, A. D., J. AOAC Int., 1994, 77, 735. 29 Alexander, M., Environ. Sci. Technol., 1995, 29, 2713. 30 Pavlostathis, S. G., and Mathavan, G. N., Environ. Sci. Technol., 1992, 26, 532. 31 Hewitt, A. D., and Grant, C. L., Environ. Sci. Technol., 1995, 29, 769. 32 Chiou, C. T., and Shoup, T. D., Environ. Sci. Technol., 1985, 19, 1196. Paper 7/06258C Received August 27, 1997 Accepted October 30, 1997 200 Analyst, February 1998, Vol. 123
ISSN:0003-2654
DOI:10.1039/a706258c
出版商:RSC
年代:1998
数据来源: RSC
|
8. |
Acyclic and cyclic polythiamonoaza- and polythiadiaza-alkane hydrazone derivatives as chromogenic extractants for silver ion |
|
Analyst,
Volume 123,
Issue 2,
1998,
Page 201-207
Junichi Ishikawa,
Preview
|
PDF (97KB)
|
|
摘要:
Acyclic and cyclic polythiamonoaza- and polythiadiaza-alkane hydrazone derivatives as chromogenic extractants for silver ion Junichi Ishikawa, Hidefumi Sakamoto*†, Tamao Mizuno, Kunio Doi and Makoto Otomo Department of Applied Chemistry, Nagoya Institute of Technology, Gokiso-cho, Showa-ku, Nagoya 466, Japan Acyclic and cyclic dithiaza-, tetrathiaza- and tetrathiadiazaalkane derivatives bearing 6-trifluoromethyl-2,4-dinitrophenylhydrazono moieties as chromogenic groups on the nitrogen atoms of the alkanes were synthesized and used as extractants for some class ab and b metal ions in liquid–liquid extraction.The extraction behavior was estimated by using the spectral change in the organic phase containing the chromoionophore. All of these compounds exhibited high Ag+ selectivity relative to other class ab and b metal ions, which are soft Lewis acids, even against Hg2+. In the extraction of silver nitrate from aqueous acidic and basic phases, interestingly, the absorption maxima of the organic phases containing the chromoionophores in the visible absorption region shifted to shorter and longer wavelengths, respectively.Among the chromoionophores used here, the acyclic tetrathiazaalkane hydrazone 4b extracted Ag+ most effectively from an aqueous phase into a 1,2-dichloroethane phase under all the extraction conditions studied. A linear calibration graph up to 5.5 3 1025 mol dm23 Ag+ in the aqueous neutral phase was obtained spectrophotometrically by using the acyclic tetrathiazaalkane hydrazone 4b and the determination of Ag+ in a silver solder sample was achieved from the calibration graph.Keywords: Chromoionophore; ionizable ionophore; thiazaalkane; solvent extraction; silver(I) ion; extractant; separation Numerous chromogenic crown ether derivatives have been synthesized and investigated as chromoionophores for particular alkali metal ions.1–3 They are classified into two categories: (1) ionophores bearing proton-dissociable chromogenic moieties, such as a nitrophenol group,2,3 and (2) ionophores bearing neutral chromogenic moieties, such as a phenylazobenzene group.1 A number of chromogenic crown ethers have been investigated as extractants for the detection of alkali metal ions and amines in liquid–liquid extraction, flow injection analysis (FIA)4,5 and optical sensors, (optodes).6–10 On the other hand, most thiazacrown derivatives, which incorporate sulfur and nitrogen atoms as donor atoms instead of oxygen atoms, have much higher complexabilities for soft Lewis acids, such as class ab and b metal ions, than crown ethers.11–14 Several series of macrocycles incorporating oxygen, nitrogen and/or sulfur atoms have been synthesized and the complexabilities of the compounds for some of these metal ions have been investigated by several groups.13–16 It was found that as the number of sulfur atoms as donor atoms in these compounds increased, the higher were their complexabilities for class b metal ions, such as Ag+ and Hg2+.15,16 In addition, compounds incorporating sulfur atoms showed high complexabilities for class b metal ions even in aqueous solution.12 Only a few chromogenic ionophores incorporating sulfur atoms have been synthesized so far.17,18 They have protondissociable chromogenic moieties and have been used only as extractants for class b metal ions.We have previously reported that 7-phenyldithia-7-aza-9-crown-3 (1a), 13-phenyltetrathia- 13-aza-15-crown-5 (2a) and 7,16-diphenyltetrathia-7,16-diaza- 18-crown-6 have high selectivities for Ag+ in liquid–liquid extraction.19 We have also synthesized the chromogenic compounds, 1b–4b, derived from the cyclic and acyclic polythiamonoazaalkane derivatives, 1a–4a, respectively, by coupling with a hydrazone moiety.The chromoionophores 1b– 4b exhibited high Ag+ complexabilities even in an aqueous acidic 1,4-dioxane solution. Considerable hypsochromic shifts of the visible absorption spectra, as in the case of the complexation of donor–acceptor type chromoionophores, of solutions containing 1b–4b were observed when they formed complexes with Ag+ in acidic media.20 We reported that in liquid–liquid extraction using silver nitrate, the visible absorption spectra of the organic phases containing tetrathiazaalkane hydrazones 2b and 4b as the extractants showed hypsochromic and bathochromic shifts under conditions where the aqueous phases were acidic, below pH 4.0, and neutral, pH 6.0–7.0, respectively.21 Considering the solvent extractabilities of the complexes, the species formed under acidic conditions should be ion-pair complexes such as a proton–undissociated chromoionophore– Ag+ complex with NO32.On the other hand, under neutral conditions the chromoionophores complex Ag+ and simultaneously dissociate a proton from the hydrazone moiety to form neutral complexes without a counter anion.This paper describes the extraction behavior of chromoionophores for some class ab and b metal ions using dihexadecyl phosphate as a lipophilic counter anion and provides details of Ag+ extraction under acidic aqueous phase conditions. The liquid–liquid extraction properties of the chromogenic cyclic and acyclic polythiazaalkane derivatives, 2b and 4b–6b, for some class ab and b metal ions at various aqueous phase acidities were studied in detail. Under neutral and basic aqueous phase conditions, the extractabilities of the chromoionophores, 1b–6b and 7b and 8b, which are crown ether hydrazones, for some metal ions were examined and excellent Ag+ selectivities were observed for 2b and 4b–6b.The determination of the Ag+ content in an actual sample using the chromoionophore 4b as an extractant under neutral conditions is also described. Experimental Reagents Unless otherwise specified all reagents were of the best available grade and were used as received.Water was doubly distilled. 1,2-Dichloroethane was distilled and then saturated with distilled water immediately before use. All metal salts were † Present address: Department of Applied Chemistry, Faculty of Systems Engineering, Wakayama University, Sakaedani 930, Wakayama-shi, 640, Japan. E-mail: skmt@sys.wakayama-u.ac.jp Analyst, February 1998, Vol. 123 (201–207) 201S S S N S R n n 1a; n = O, R = H, 2a; n = 1, R = H 1b; n = O, R = hy, 2b; n = 1, R = hy S S S N S R n n 3a; n = O, R = H, 4a; n = 1, R = H 3b; n = O, R = hy, 4b; n = 1, R = hy S S N N S S R R 5a; R = H, 5b; R = hy S S N N S S R R 6a; R = H, 6b; R = hy O O O N O R 7a; R = H, 7b; R = hy O O N N O O R R 8a; R = H, 8b; R = hy O2N F3C NO2 N N H H hy •• of analytical-reagent grade. All chromoionophores used in this work were synthesized in a similar manner to that described previously.20 Spectroscopic and physical methods Visible absorption spectra were obtained on a Hitachi 150-20 spectrophotometer with 1 cm quartz cells.pH measurements were conducted by using a TOA HM-30S pH-meter equipped with a TOA GST-5311C glass electrode. Atomic absorption and ICP atomic emission spectra were recorded with a Seiko SAS/ 727 spectrometer and a Seiko SPS 7000A spectrometer, respectively. Liquid–liquid extraction The procedures adopted here for the extraction experiments are classified into two categories. Extraction of metal ions from aqueous acidic phases Tetrathiaza- and tetrathiadiazaalkane derivatives bearing hydrazone moieties, 2b and 4b–6b, behave as electrically neutral extractants in the solvent extraction of the metal ion from an aqueous acidic phase using a lipophilic anion.In the extraction of the metal ions, dihexadecyl hydrogenphosphate was used to supply a highly lipophilic counter anion for the chromoionophore –metal ion complex cation. For evaluation of the ion extraction selectivity, the following two extraction procedures were employed: (a) The aqueous phase was adjusted to pH 4.2 with a 1.0 3 1022 mol dm23 MES [2-(N- morpholino)ethanesulfonic acid] buffer and contained 1.0 3 1023 mol dm23 metal ion as the sulfate.The ionic strength of the aqueous phase was adjusted to 0.2 with potassium sulfate, which was not extracted into the organic phase in the presence of any of the chromoionophores, including the monoaza-15-crown-5 and diaza-18-crown-6 derivatives, 7b and 8b, respectively, under the experimental conditions used.The organic phase contained 2.0 3 1025 mol dm23 chromoionophore in 1,2-dichloroethane and 2.5 3 1024 mol dm23 dihexadecyl hydrogenphosphate. Twelve millilitres each of the aqueous and organic phases were placed in a 50 ml centrifuge tube. The mixture was shaken in an incubator for 1 h at 25.0 ± 0.2 °C. After standing for 2 h at 25.0 ± 0.2 °C to attain complete phase separation, 4 ml of the organic phase were taken and the absorption spectrum was recorded.(b) The aqueous phase was adjusted to pH 4.2 with 1.0 3 1022 mol dm23 MES buffer and contained 2.0 3 1025 mol dm23 metal ion as the sulfate and potassium sulfate as mentioned above. The organic phase contained 1.0 3 1024 mol dm23 chromoionophore in 1,2-dichloroethane (5 ml) and 2.5 3 1024 mol dm23 dihexadecyl hydrogenphosphate. A mixture of the solutions (5 ml each) was shaken in a similar manner to that described above. After separation of the two phases, an aliquot (0.1 ml) of the aqueous phase was taken and subjected to AAS, except for Hg2+ and Tl+ which were detected by ICP-AES. The Ag+ extraction behavior of the tetrathiazaalkane hydrazones, 2b and 4b, was investigated using an aqueous phase (12 ml) of 1.0 31024 mol dm23 silver sulfate, containing potassium sulfate to adjust the ionic strength to 0.2, and a 1,2-dichloroethane solution (12 ml) containing 1.0 3 1025 mol dm23 ligand and 2.5 3 1024 mol dm23 dihexadecyl hydrogenphosphate.Determination of Ag+ was effected by spectrophotometry. The absorbance of the organic phase was measured at 440 nm under conditions where the pH of the aqueous phase was varied between 1.7 and 4.5; the pH was adjusted by using sulfuric acid or 1.0 3 1022 mol dm23 MES–KOH buffer. Extraction of metal ions from aqueous neutral and basic phases The following two procedures were adopted for a qualitative comparison of the metal ion extractabilities of the chromoionophores: (a) The aqueous phase (12 ml) containing 1.0 3 1023 mol dm23 metal ion as the sulfate and potassium sulfate (ionic strength 0.2) was shaken with a 1,2-dichloroethane (12 ml) solution of 2.0 3 1025 mol dm23 chromoionophore.The pH of the aqueous phase was adjusted to 6.0 with 1.0 3 1022 mol dm23 MES–KOH buffer. Four millilitres of the organic phase were taken and the absorption spectrum was recorded. (b) The aqueous phase (5 ml) containing 2.0 3 1025 mol dm23 metal ion as the sulfate and potassium sulfate (ion strength, 0.2), adjusted to pH 6.0, was shaken with a 1,2-dichloroethane solution (5 ml) of 1.0 3 1024 mol dm23 chromoionophore. The concentration of the metal ion distributed into the organic phase was determined by measuring the metal content of the aqueous phase by AAS or ICP-AES. In order to estimate the composition of the complexes that were formed between the chromoionophores, 2b and 4b–6b, and Ag+ and extracted into the organic phase, the extractions were carried out by varying the chromoionophore concentration and keeping the Ag+ concentration constant at 5.0 3 1026 mol dm23.The chromoionophore concentrations were 1.5 3 1026–4.0 3 1025 mol dm23 for 2b and 4b, and 1.5 3 1025–1.0 3 1024 mol dm23 for 5b and 6b. The pH of the aqueous phase was adjusted to 6.8 for 2b and 4b with 1.0 3 1022 mol dm23 MES–KOH buffer, and to 7.7 for 5a and 6a with 1.0 3 1022 mol dm23 MOPS [3-(N-morpholino)propanesulfonic acid]– KOH buffer.Determination of Ag+ in the aqueous phase was effected by AAS. Extraction experiments for the determination of extraction constants were carried out using aqueous phases of 202 Analyst, February 1998, Vol. 1232.0 3 1024 and 1.0 3 1025 mol dm23 Ag+ as the sulfate containing potassium sulfate (ionic strength 0.2), and 1,2-dichloroethane solutions of the chromoionophores at concentrations of 1.0 3 1025 and 2.0 3 1024 mol dm23, respectively. The absorbance of the organic phase was measured at 440 and 500 nm under conditions where the pH of the aqueous phase was varied between 2.5 and 8.0 for 1b–6b with Good’s buffer and KOH.The extraction constants were calculated from the observed absorbance changes. Determination of Ag+ in silver solder A 0.6 g sample of silver solder was dissolved in concentrated nitric acid (10 ml), and the solution was heated until fumes were observed. Water was added to the solution after it had been cooled to room temperature, and the solution was again heated to fumes.This treatment was repeated three times to ensure complete dissolution of the sample. The resulting solution was diluted to 100 ml, and 1.0 ml of this solution was taken for dilution to 100 ml with 0.2 mol dm23 citric acid–KOH buffer and for adjustment of the pH to 6.0, followed by a 10-fold dilution with water. The sample solution (12 ml) and a 1,2-dichloroethane solution (12 ml) containing 1.0 3 1023 mol dm23 of the acyclic tetrathiazaalkane hydrazone 4b were transferred into a 50 ml centrifuge tube.The mixture was shaken for 1 h. After phase separation, the absorbance of the organic phase was measured at 524 nm. Results and discussion Extraction of Ag+ from aqueous phases The liquid–liquid extraction of Ag+ from an aqueous into an organic phase with the chromoionophores was carried out at various pH values between 2.6 and 7.1. The aqueous phase contained 2.0 3 1024 mol dm23 silver nitrate and potassium sulfate (ionic strength 0.2).The organic phase was a 1,2-dichloroethane solution containing 1.1 3 1025 mol dm23 chromoionophore. No extraction of Ag+ into the organic phase in the absence of the chromoionophore was observed by AAS analysis of the aqueous phase and by spectrophotometric analysis of the organic phase. In the extraction of Ag+ from an aqueous acidic phase (below pH 4.0), the spectra in the organic phase shifted hypsochromically.For 4b the complex extracted into the organic phase should be a ternary complex of proton-undissociated 4b–Ag+– NO3 –, as shown in Fig. 1(b), that is, the proton of the hydrazone group should be hardly dissociated below pH 4. The hypsochromic shift of the visible absorption spectrum of the extracted complex is caused by the electron-withdrawing effect of Ag+ as a result of the interaction of Ag+ with the nitrogen atom in the tetrathiazaalkane chain. The interaction between Ag+ and the nitrogen atom of 4a was supported by 1H NMR spectrometry, which showed changes in the chemical shifts of the phenyl protons after complexation.22 Such a hypsochromic shift of the absorption spectrum of 4b caused by complexation with Ag+ is similar to the spectral change in donor–acceptor type chromogenic crown ethers, such as N-(4-nitrophenylazo)phenyl-aza- 18-crown-6.1 On the other hand, considerable bathochromic shifts of the absorption spectra in the organic phase containing 4b were observed in the extraction of Ag+ from an aqueous neutral phase (pH 6.0–7.0).The bathochromic shifts of the absorption spectra are primarily caused by deprotonation of the hydrazone moiety and formation of a neutral complex with Ag+, 4b2–Ag+, [Fig. 1(c)] because the basicity of the imino group in the hydrazone moiety is decreased by interaction between Ag+ and the nitrogen atom. Similar phenomena were found in the extraction of univalent class b metal ions with chromogenic thiacrown ethers bearing a proton-dissociable chromogenic group, such as 4A-picrylaminobenzo-1,4,8,11-tetrathiacyclopentadec- 13-ene, as the extractant.18 A similar change in the absorption spectra was also found for 2b.Extraction of metal ions from aqueous acidic phase The liquid–liquid extractions of class ab metal ions, viz., Mn2+, Co2+, Ni2+, Cu2+ and Tl+, and class b metal ions, such as Cd2+, Ag+, and Hg2+, using tetrathiaza- and tetrathiadiazaalkane hydrazones were performed at pH 4.2 in the presence of a lipophilic counter anion.Dihexadecyl phosphate was chosen as the counter ion, because it has sufficient lipophilicity to dissolve in 1,2-dichloroethane and its hydrogenated form, dihexadecyl hydrogenphosphate, has an appropriate proton-dissociation constant (pKa), viz., about 1.3, which is estimated from that of diethyl hydrogenphosphate, for investigation of the extraction equilibria. Under the conditions investigated, the protonundissociated chromoionophore–metal ion complexes were extracted as the ion-pair complexes with the counter anion.No proton dissociation of the chromoionophore was observed spectrophotometrically for the extracted complexes. The absorption spectra of the organic phases containing 4b for the extractions of some metal ions at pH 4.2 are shown in Fig. 2. A marked hypsochromic shift of the absorption spectra in the organic phase containing the acyclic tetrathiazaalkane hydrazone 4b was found in the extraction of Ag+.The results obtained in the liquid–liquid extractions of the metal ions with the chromoionophores, 2b and 4b–6b, are shown in Table 1. No extraction of metal ions, except for Cu2+ and Hg2+, which were extracted only slightly by dihexadecyl phosphate, was found in the absence of the ligand. It is clear from Table 1 that Ag+ is selectively extracted by the acyclic tetrathiazaalkane hydrazone 4b while Hg2+, which is known to have an affinity for thiacrown compounds, is also slightly extracted.A similar extraction behavior was found for 2b and 6b. With 5b, the extractability for Ag+ decreased dramatically, probably because of the highly distorted conformation of the alkane ring.19 The extraction behavior of 2b and 4b for Ag+ was examined by changing the pH of the aqueous phase. If the composition of the extracted species is assumed to be 1 : 1 : 1 for Ag+ : chromoionophore : dihexadecyl phosphate ion, the extraction equilibrium and the extraction constant (Kex) are defined by the following equations: Kex Ag HL HA Ag + + + + + ( ) ( ) [ ( ) ] o o o HL A H [| (1) Kex o + + o o Ag(HL)A] H Ag HL HA = [ [ ] [ ][ ] [ ] (2) where HL, HA, A2 and Ag(HL)A are the chromoionophore, dihexadecyl hydrogenphosphate, dihexadecyl phosphate ion and the extracted complex, respectively.The subscript ‘o’ denotes the organic phase while the absence of the subscript denotes the aqueous phase.Since the distribution of the chromoionophores into the aqueous phase is negligible, the total concentrations of Ag+ (CAg) and the chromoionophore (CHL) are, respectively, represented as CAg = [Ag+] + [Ag(HL)A]o (3) CHL = [HL]o + [Ag(HL)A]o (4) If the concentration of Ag(HL)A is much lower than that of Ag+, eqn.(3) can be written as CAg = [Ag+] (5) The absorbance (A) of the organic phase is given by A = eHL[HL]o + eAg(HL)A[Ag(HL)A]o (6) Analyst, February 1998, Vol. 123 203S N S S S H H N N NO2 O2N F3C ( a) S N S S S H H N N NO2 O2N F3C ( b) Ag+ AgNO3 aqueous acidic phase organic phase S N S S S H H N N NO2 O2N F3C ( a) S N S S S H N N ( c) Ag+ O2N F3C NO2 aqueous neutral phase organic phase Ag+ H+ NO3 – where eHL and eAg(HL)A are the molar absorptivities of the species represented by the subscripts.Because the value of eHL is known, the values of eAg(HL)A and Kex can be simultaneously obtained by minimizing the error square sum (U) defined by U = S(Aobs,i2Acal,i)2 (7) where Acal,i and Aobs,i are the calculated and the experimentally observed absorbances, respectively, at 437 nm for 2b and 439 nm for 4b.The extraction constants (log Kex) of 2b and 4b for Ag+ are 3.68 and 4.94, respectively. The extractability of the latter is 18 times higher than that of the former. The acyclic tetrathiazaalkane hydrazone 4b has a sufficiently flexible structure to adopt a suitable conformation for complexation with Ag+. On the other hand, it is evident from the examination of a molecular model that the cyclic tetrathiazaalkane hydrazone 2b has a rigid conformation and the lone pairs of the sulfur atoms point away from the ring because of the much larger size of this atom.Since, consequently, Ag+ may lie in contact with the plane made by the arrangement of the heteroatoms in the tetrathiazaalkane ring of 2b, the complexability of 2b is not particularly high and hence the extractability of 2b should be less than that of 4b.Extraction of metal ions from aqueous neutral and basic phases The liquid–liquid extractions of some class ab and b metal ions with chromoionophores, 1b–8b, from aqueous neutral phases were carried out in the absence of a lipophilic anion. The absorption spectra of the organic phase containing 4b in the extraction of some metal ions at pH 6.0 are shown in Fig. 3. Only for the extraction of Ag+, was a large bathochromic shift of the absorption maximum observed.The other tetrathiazaalkane and tetrathiadiazaalkane hydrazones, 2b, and 5b and 6b, respectively, exhibited similar spectral changes. As mentioned above, these chromoionophores dissociate a proton of the Fig. 1 Postulated structures of complexes extracted from aqueous acidic and neutral phases into the organic phase. 204 Analyst, February 1998, Vol. 123hydrazone moiety to form uncharged complexes with Ag+ and then distribute into the organic phase. It is clear from Fig. 3 that spectral changes for the extractions of the other metal ions were hardly observed even for Hg2+. The extraction behavior of chromoionophores 1b–8b for the metal ions was also examined, the results being summarized in Table 2. The dithiazaalkane hydrazones, 1b and 3b, and the azacrown ether hydrazones, 7b and 8b, hardly extracted any metal ions. On the other hand, the tetrathiazaalkane hydrazones, 2b and 4b, and the tetrathiadiazaalkane hydrazones, 5b and 6b, extracted Ag+ selectively.The results demonstrate that the hydrazone moieties in these ligands could not provide a coordination site for any metal ions under the experimental conditions used, although hydrazone derivatives have been shown to be coordinating ligands for many transition metal ions.23,24 The extraction constants of the chromoionophores, 2b and 4b–6b, for Ag+ were evaluated by two procedures, as follows: (1) simultaneous determination of the extraction constants and the compositions of the extracted complexes using a logarithmic plot of the distribution constant (DAg) versus the concentration of chromoionophore; (2) direct determination using spectral changes in the organic phase by varying the pH of the aqueous phase in the presence of a large excess of Ag+ over the chromoionophore.Assuming the composition of the extracted species, in the first method, to be a 1 :m complex of Ag+ : chromoionophore, the composition and the extraction constant (Kex) are defined by the following equations: Kex m m m Ag (HL) AgL(HL) H + o o + + + [|[ ] – 1 (8) K m m m ex AgL(HL H Ag ][HL = - + + [ ) ] [ ] [ ] 1 0 0 (9) DAg m m = + - + - [ ) ] [ ) ] AgL(HL Ag ] [AgL(HL 1 0 1 (10) where AgL(HL)m21 is a 1 :m complex of Ag+ : chromoionophore, and DAg represents the distribution constants of Ag+.When [Ag+] >> [AgL(HL)m21], the logarithmic form of eqn. (9) can be rewritten by the substitution of eqn. (10) as follows: logDAg = logKexm + mlog[HL]o + pH (11) The slope, m, of the straight lines for plots of logDAg versus log[HL]o affords the ratio of the chromoionophore to Ag+ of the extracted complex.Plots for 4b–6b showed linear relationships with a slope of unity, m = 1, demonstrating that 1 : 1 complexes, AgL, were extracted into the organic phase. On the other hand, the plots for 2b exhibited a break point between two linear relationships, with m = 1 and 2 for the slopes of each line. At lower and higher concentrations than about 2 31025 mol dm23 of 2b, where the concentration of 2b in the organic phase is at least ten times that of Ag+ in the aqueous phase, 2b formed, respectively, 1 : 1 and 1 : 2 complexes with Ag+ which were extracted into the organic phase.As mentioned above, because the cavity size of the cyclic tetrathiazaalkane moiety of 2b is not sufficiently large to capture Ag+ in the cavity, two crown units interpose one Ag+ ion to form a 1 : 2 sandwich-type complex in the presence of a large excess of 2b.The results obtained by applying the second method are shown in Fig. 4, indicating the variation of the apparent molar absorptivity (at 500 nm) of the organic phase containing Fig. 2 Absorption spectra of the organic phase in the extraction of some metal ions from an aqueous acidic phase with 4b and dihexadecyl phosphate ion. Table 1 Extraction of metal ions using tetrathiaza- and tetrathiadiaza-alkane hydrazones and dihexadecyl hydrogenphosphate* under acidic aqueous solution conditions Extraction (%)† Metal Classification ion of metal ion 2b 4b 5b 6b Blank Mn2+ ab 1 0 5 3 0 Co2+ ab 0 1 2 0 0 Ni2+ ab 8 7 0 7 0 Cu2+ ab 8 4 7 3 4 Tl+ ab 0 0 0 0 0 Cd2+ b 4 4 4 4 0 Ag+ b 64 89 14 50 0 Hg2+ b 25 18 21 23 < 10 * Organic solution: [chromoionophore] = 1.0 3 1024 mol dm23, [dihexadecyl hydrogenphosphate] = 2.0 3 1024 mol dm23, in 1,2-dichloroethane.Aqueous solution: [metal ion] = 2.0 3 1025 mol dm23, pH 4.2.† Evaluated from metal ion concentration in the aqueous solution. Fig. 3 Absorption spectra of the organic phase in the extraction of some metal ions from an aqueous neutral phase with 4b. Table 2 Extraction of metal ions using chromoionophores under neutral aqueous solution conditions* Extraction (%)† Metal Classification ion of metal ion 1b 2b 3b 4b 5b 6b 7b 8b Mn2+ ab 0 0 0 0 0 0 0 0 Co2+ ab 0 0 0 0 0 0 0 0 Ni2+ ab 0 0 0 0 6 2 0 0 Cu2+ ab 0 4 0 2 2 2 0 0 Zn2+ ab 0 2 0 1 0 0 0 0 Tl+ ab 0 3 0 4 5 7 0 0 Cd2+ b 0 0 0 0 0 0 0 0 Ag+ b 0 30 0 84 14 33 0 1 Hg2+ b < 5 < 5 < 5 < 5 < 5 < 5 < 5 < 5 * Organic solution: [chromoionophore] = 1.0 3 1024 mol dm23, in 1,2-dichloroethane.Aqueous solution: [metal ion] = 2.0 31025 mol dm23, pH 6.0. † Values were obtained from determination of the metal ion concentration remaining in the aqueous phase. Analyst, February 1998, Vol. 123 205chromoionophores 1b–4b with the pH of the aqueous phase.No extraction of Ag+ into the organic phase was observed below pH 2.5, which demonstrates that Ag+ is hardly extracted into the organic phase in the liquid–liquid extraction system used unless the chromoionophore can dissociate a proton to form an uncharged complex with Ag+. A similar spectral change to that of 4b was observed in the extraction of Ag+ with 2b, whereas chromoionophores 5b and 6b, which have two hydrazone moieties, showed fewer spectral changes than 2b and 4b.The extraction equilibrium and the extraction constant are defined by eqns. (8) and (9), respectively. The total concentrations of Ag+ (CAg) and the chromoionophore (CHL) are, respectively, defined as CAg = [Ag+] + [AgL]o + [AgL(HL)]o (12) CHL = [HL]o + [AgL]o + 2[AgL(HL)]o (13) where AgL and AgL(HL) are 1 : 1 and 1 : 2 complexes of Ag+ : chromoionophore, respectively. If the concentration of Ag+ is much higher than those of AgL and AgL(HL), eqn. (12) can be written as CAg = [Ag+] (14) On the other hand, if the concentration of the chromoionophore is much higher than those of AgL and AgL(HL), eqn.(13) can be written as CHL = [HL]o (15) The extraction constants of the chromoionophores, 2b and 4b– 6b, for Ag+ were calculated using systems in which the concentrations of both Ag+ and the chromoionophore were varied. The absorbance of the organic phase is given by A = eHL[HL]o + eAgL[AgL]o + eAgL(HL)[AgL(HL)]o (16) As the value of eHL is known, the values of eAgL, eAgL(HL) and Kex can be simultaneously obtained in a similar manner to that described above. The extraction constants of the chromoionophores, 2b and 4b–6b, for Ag+ are summarized in Table 3.The values of the extraction constants were consistent with those obtained from the plots of logDAg versus log[HL]o in the extraction of Ag+. The value of the extraction constant for 5b was determined only by plotting logDAg versus log[HL]o, because the spectral changes in the organic phase were too small to obtain a reasonable extraction constant.The extractabilities of the chromoionophores for Ag+ decreased in the order 4b > 2b Å 6b > 5b. The tendency of the extractabilities of 2b and 4b is similar to that in the extractions using dihexadecyl phosphate ion as a lipophilic counter-anion described above. A comparison of the extraction constant of the cyclic tetrathiadiazaalkane hydrazone 5b with that of the acyclic analog 6b for Ag+ shows that the latter is ten times larger than the former.As mentioned above, the structure of the cyclic chromoionophore 5b is too distorted for it to adopt a conformation that would allow it to form a stable complex with Ag+ because of the steric hindrance of the adjacent phenyl groups.19 In contrast, the structure of the acyclic compound 6b is sufficiently flexible to allow the sulfur atoms of the alkane chain to be favorably located to interact with Ag+. Determination of Ag+ with chromoionophore 4b A calibration graph for the extractive spectrophotometric determination of Ag+ using the chromoionophore 4b was prepared in the following manner: The aqueous phase containing Ag+ was adjusted to pH 6.0 with 0.2 mol dm23 citric acid– KOH buffer, which also acts as a masking agent to prevent hydrolysis of the other metal ions.The organic phase contained 1.0 3 1023 mol dm23 4b. The extraction was carried out as described under Experimental and the absorbance of the organic phase was measured at 524 nm against a reagent blank.The calibration graph obtained is shown in Fig. 5, and exhibits a straight line below 5.5 31025 mol dm23 Ag+. Up to 5.0 31023 mol dm23 each of Fe3+, Co2+, Ni2+, Cu2+, Zn2+, Cd2+ and Hg2+ in the aqueous phase had no effect on the calibration graph for Ag+. A commercial silver solder was analyzed for Ag+ by the Fig. 4 Plots of molar absorptivity of the organic phase at 500 nm versus pH of the aqueous phase in the extraction of Ag+ with 1b–4b. 2: 1b, 5: 2b, «: 3b, ~: 4b. Table 3 Extraction constants of tetrathiaza- and tetrathiadiaza-alkane hydrazones for Ag+ eHL/104 Ligand Log Kex1 Log Kex2 (lmax/nm) eAgL */104 eAg(HL)L */104 2b 22.53 2.38 2.66 3.62 (437) 4.98 2.69 4b 20.92 (439) 3.87 3.27 5b 23.54 † (437) 5.72 6b 22.53 (442) 3.66 * At 500 nm. † Value obtained from plots of logDAg versus log[H2L]o. Fig. 5 Calibration plots for Ag+ in the liquid–liquid extraction using 4b. 2: Without foreign metal ions, 5: [metal ion] = 5.0 3 1023 mol dm23; metal ions: Fe3+, Co2+, Ni2+, Cu2+, Zn2+, Cd2+, Hg2+. 206 Analyst, February 1998, Vol. 123proposed method. The nominal composition of the solder was: silver, 56%; zinc, 17%; tin, 5%; copper, 22%. The analytical result for Ag+ was 56.9 ± 0.1% (average of triplicate determinations), which was in fair agreement with the nominal value. References 1 L�ohr, H., and V�ogtle, F., Acc. Chem. Res., 1985, 18, 65. 2 Takagi, M., Cation Binding by Macrocycles, ed.Inoue, Y., and Gokel, G. W., Marcel Dekker, New York, 1991, p. 465. 3 Hayashita, T., and Takagi, M., in Comprehensive Supramolecular Chemistry, ed. Lehn, J.-M., Pergamon, Oxford, 1996, vol. 1, p. 635. 4 Kimura, K., Iketani, S., Sakamoto, H., and Shono,T., Analyst, 1990, 115, 1251. 5 Nakashima, K., Muraki, K., Nakatsuji, S., Akiyama, S., Kaneda, T., and Misumi, S., Analyst, 1989, 114, 501. 6 van Gent, J., Sudh�olter, E. J. R., Lambeck, P. V., Popma, T. J. A., Gerritsma, G. J., and Reinhoudt, D. N. J., J. Chem. Soc., Chem. Commun., 1988, 893. 7 Al-Amir, S. M. S., Ashworth, D. C., and Narayanaswamy, R., Talanta, 1989, 36, 645. 8 Alder, J. F., Ashworth, D. C., Narayanaswamy, R., Moss, R. E., and Sutherland, I. O., Analyst, 1987, 112, 1191. 9 Janata, J., Anal. Chem., 1992, 64, 921A. 10 Arnold, M. A., Anal. Chem., 1992, 64, 1015A. 11 Inoue, Y., Liu, Y., and Hakushi, T., in Cation Binding by Macrocycles, ed. Inoue, Y., and Gokel, G. W., Marcel Dekker, New York, 1990, p. 1. 12 Izatt, R. M., Pawlak, K., Bradshaw, J. S., and Bruening, R. L., Chem. Rev., 1991, 91, 1721. 13 Adam, K. R., Baldwin, D. S., Duckworth, P. A., Lindoy, L. F., McPartlin, M., Bashall, A., Powell, H. R., and Tasker, P. A., J. Chem. Soc., Dalton Trans. 1995, 1127. 14 Reid, G., and Schr�oder, M., Chem. Soc. Rev., 1990, 19, 239. 15 Lindoy, L. F., Pure Appl. Chem., 1989, 61, 1575. 16 Heitzsch, O., Gloe, K., Stephan, H., and Weber, E., Solvent Extr. Ion Exch., 1994, 12, 475. 17 Muroi, M., Kamiki, T., and Sekido, E., Bull. Chem. Soc. Jpn., 1989, 62,8 Sekido, E., Chayama, K., and Muroi, M., Talanta 1985, 32, 797. 19 Sakamoto, J., Ishikawa, J., and Otomo, M., Bull. Chem. Soc. Jpn., 1995, 68, 2831. 20 Ishikawa, J., Sakamoto, H., Mizuno, T., and Otomo, M., Bull. Chem. Soc. Jpn., 1995, 68, 3071. 21 Sakamoto, H., Ishikawa, J., Mizuno, T., Doi, K., and Otomo, M., Chem. Lett., 1993, 609. 22 Sakamoto H., and Ishikawa, J., unpublished work. 23 Singh, R. B., Jain, P., and Singh, R. P., Talanta, 1982, 29, 77. 24 Taya, T., Mukouyama, Y., Doi, K., and Otomo, M., Bull. Chem. Soc. Jpn., 1994, 67, 710. Paper 7/06798D Received September 18, 1997 Accepted October 27, 1997 Analyst, February 1998, Vol. 123 207
ISSN:0003-2654
DOI:10.1039/a706798d
出版商:RSC
年代:1998
数据来源: RSC
|
9. |
Development and application of an automated gas chromatographic sampling system for verification of test vapor stream concentrations for chemical sensor studies |
|
Analyst,
Volume 123,
Issue 2,
1998,
Page 209-215
Timothy Torkelson,
Preview
|
PDF (101KB)
|
|
摘要:
Development and application of an automated gas chromatographic sampling system for verification of test vapor stream concentrations for chemical sensor studies Timothy Torkelson and David S. Ballantine* Department of Chemistry, Northern Illinois University, DeKalb, IL 60115, USA An automated GC sampling system is described which permits real-time verification of vapor stream concentrations used in the testing and development of chemical sensors. The system consists of a computer-actuated injector loop placed in-line between a VG-400 vapor generator (Microsensor Systems) and a surface acoustic wave (SAW) sensor.Optimization of the system was performed by comparing the predicted VG-400 vapor output with the GC-determined vapor stream concentrations. The system was utilized in the identification of hystersis effects in a polymer-coated SAW sensor. Keywords: Automated analysis; sensors; chromatography; calibration; vapor generation As a result of the increasing power and decreasing cost of computers and other electronic equipment, automated sampling and analysis are growing.Automated methods provide faster, more reliable analyses when compared with the slow sample turnaround and human error associated with manual procedures. Automated systems also have the advantage of being on-task continuously, and can operate in environments in which it is difficult or dangerous for a human to function. For these reasons, automated sampling and analyses are very popular in environmental monitoring, occupational safety, process control, chemical identification and sensor development.1–10 In chemical sensor studies, the generation of well defined vapor streams over a wide range of concentrations is important in order to study and calibrate the sensor response.Generation of a vapor stream of constant concentration can be achieved by a variety of methods.11,12 Each of several previously reported methods for creating test vapors suffers from limitations which affect the range, accuracy and/or reproducibility of the generated vapor stream.Static methods, such as dilution chambers, can suffer from incomplete solvent evaporation and/or adsorption on the chamber wall, particularly for solvents with low vapor pressures, and are not suitable for the continuous generation of constant vapor concentrations. Dynamic methods, on the other hand, generally have negligible wall adsorption errors, and can easily maintain a continuous stream of a desired concentration.Permeation and diffusion tubes are very useful for low concentrations ( < 1000 ppm). The accuracy of the generated vapor stream can be adversely affected by changes in temperature, pressure and carrier flow rate, and also by the aging of the permeable membrane. Evaporation and dilution methods can provide very broad concentration ranges, but can suffer from instability of rotameters. Mass flow controllers, although more accurate than rotameters, may still be subject to 5–10% error for multiple dilution systems.12 The instrument used for vapor stream generation in our laboratory, a VG-400 vapor generator (Microsensor Systems, Bowling Green, KY, USA), depends on the dilution of a saturated vapor stream by pulse-width modulation.13 A detailed discussion of the operation principles and limitations of this system is provided in the Experimental section.For any chemical sensor, a reliable account of vapor concentrations is critical in order to evaluate the sensor response behavior or to calibrate the sensor accurately. The frequency response of acoustic wave devices, including thickness shear mode (TSM),14 acoustic plate mode (APM)15 and surface acoustic wave (SAW)16–18 sensors, can depend on both mass loading into the chemically selective film and on viscoelastic changes occurring in the film due to vapor-induced swelling and/or plasticization.While the former can be predicted from experimentally observed sorption isotherms or from thermodynamic partition coefficients,16,17 the latter is a function of both film thickness and modulus,14,15,19 and therefore is not directly proportional to vapor concentration.Consequently, vapor concentrations must be well identified in order to evaluate adequately both mass and viscoelastic response behavior. This paper describes an automated vapor sampling system to overcome the limitations of the VG-400 and other vapor generation systems, and to increase the reliability and accuracy of the delivered vapor stream.The system consists of an automated injection valve in-line between the VG-400 vapor generator and the chemical sensor. This injection valve provides an interface between the vapor generation–sensor system and a gas chromatograph. The automated system runs concurrently with standard sensor analyses by using a computer for instrument control and data acquisition, and provides precise real-time verification of vapor stream concentrations used in sensor studies.Experimental The experimental system is illustrated in Fig. 1. Each of the components included in the system is described below. Fig. 1 Schematic diagram of automated GC sampling system. Individual items are labelled; bold arrows indicate communication and data transfer between system components. Analyst, February 1998, Vol. 123 (209–215) 209Vapor generator The VG-400 vapor generator system contains four solvent reservoir bubblers housed in an aluminum block heat sink maintained at a constant subambient temperature (15 °C) by a circulating water-bath to prevent downstream condensation.For these studies only two bubblers, containing toluene and isooctane solvents, were used. All exposed surfaces in the vapor stream path are either glass, Teflon or nickel to minimize adsorption of vapor and corrosion of carrier lines. The VG-400 is controlled by a Z-80 microcontroller.Communication between the VG-400 and the host computer is accomplished via an RS-232C port. Desired dilution schedules are created and stored in data files on the host computer, and are downloaded during system operation to the VG-400. A constant carrier gas flow rate through the selected bubbler produces a saturated vapor stream at a flow rate of 100 ml min21. Dilution of this vapor stream is accomplished by pneumatic pulse width modulation through a series of three dilution chambers.20 The actual dilution factor depends on the duty cycle, i.e., the relative amount of time that the vapor and diluting air streams are on-line for each of the dilution chambers. For example, a 50% duty cycle would introduce alternating pulses, of equal width, of vapor and diluting carrier gas.Thus a 50% saturated vapor stream would be produced in the first diluting chamber. This 50% stream would then travel to the second dilution chamber (using the same duty cycle) to produce an additional factor of 2 dilution, resulting in a 25% vapor stream.This 25% stream is then sent to the third dilution chamber and mixed with diluting gas to produce a 12.5% vapor stream (dilution factor = 23 = 8). Dilution factors from 1 (pure vapor) to 125 000 are achievable. The output can be verified gravimetrically on an activated charcoal trap. Gravimetric determinations, however, provide only a time-averaged mass flow rather than a real-time indication of vapor concentrations.Long-term fluctuations in the ambient temperature and the solvent level in the bubbler can cause minor variations in the delivered mass flow. Accuracy of the VG-400 calibration can also be adversely affected by pressure differentials between the vapor and carrier sides of the dilution chamber.13 In-line flow constrictions caused by small diameter tubing, for example, can create sizable pressure differentials and upset the pressure balance in the system.Fluctuations in the pressure balance may also occur when a sensor cell is switched in place of a gravimetric collection tube. Unless the system is continuously adjusted to compensate for such changes, the delivered vapor concentrations during experimental studies can vary significantly from the calibrated values. Gas chromatograph The gas chromatograph used was a Varian 3400CX with a 4 m 3 320 mm id capillary test column (DB-1 stationary phase) and a flame ionization detector.The carrier gas (helium) flow rate was 5 ml min21. The hydrogen fuel and air oxidant flow rates were 30 and 300 ml min21, respectively. All flow rates were controlled with constant flow valves which minimized variations in the fuel-to-oxidant ratio during normal operation. The GC column oven temperature was 120 °C and the injection port and detector housing temperatures were 220 °C. The GC system was interfaced to a Benchmark 486 PC (MicroSolutions, DeKalb, IL, USA) via a Keithley MetraByte (Taunton, MA, USA) DAS1202 Analog/Digital I–O board.Acquisition of chromatographic data was performed using Varian (Palo Alto, CA, USA) Star Chromatography Workstation (v. 4.0). The GC response to toluene and isooctane test vapors was calibrated by direct injection of a series of standard solutions. Both toluene and isooctane were diltued with nonane to obtain solutions of known concentrations. Isooctane was used as an internal standard for calibration of toluene, whereas toluene was used as the internal standard for calibration of isooctane.The solutions were prepared by a 30% serial dilution from 93.3% to 0.0680% v/v analyte. The internal standard was 6.67% v/v for each concentration level. Actual solution concentrations are summarized in Tables 1 and 2. During a typical analysis, the temperature of the GC column was ramped from 50 to 120 °C at 50 °C min21 and then held at 120 °C for 2 min.Four replicate injections of 0.1 ml were performed for each concentation level. The calibration curves for toluene and isooctane had correlation coefficients r2 = 0.9991 and 0.9998, respectively. Through knowledge of the amount of standard used and the average peak size of the standard from the calibration injections, a direct conversion from area counts to micrograms of vapor could be found. Peak area counts obtained by sample loop injection to the gas chromatograph for toluene and isooctane vapors could then be converted into the masses and vapor concentrations (in mg l21) using these conversion factors.Automated injection valve A Valco (Houston, TX, USA) six-position electronic actuator injection valve with 1 8 in fittings, shown in detail in Fig. 2, was placed in-line between the VG-400 and the SAW sensor, and used to introduce samples of the vapor generator output to the GC system. Removable 25 ml (0.44 mm id) and 250 ml (1.5 mm id) stainless-steel sample loops (Alltech, Deerfield, IL, USA) were used in evaluating and calibrating the system.The actuating valve and sample loop were enclosed in an insulated box with an electric heater included to regulate the temperature of the valve. A YSI Series 400 temperature probe (Cole-Parmer, Niles, IL, USA) was used to monitor the temperature of the valve housing. The automated injector was tested under two injection configurations. In the forward mode, the sample loop is in-line between the VG-400 and the SAW sensor (Load position).A sample injection is performed by actuating the valve to the Inject position for a designated time interval. In the reverse mode, the sample loop is left off-line (in the Inject position) Table 1 Solution data for the GC calibration of toluene Toluene (analyte) (%) Toluene per injection/mg Isooctane (internal standard) (%) Isooctane per injection/mg Toluene relative to internal standard (%) 93.3 80.9 6.67 4.62 17.5 28.0 24.3 6.67 4.62 6.06 8.40 7.28 6.67 4.62 1.82 2.52 2.18 6.67 4.62 0.545 0.756 0.655 6.67 4.62 0.164 0.227 0.197 6.67 4.62 0.491 0.0680 0.0590 6.67 4.62 0.0147 Table 2 Solution data for the GC calibration of isooctane Isooctane (analyte) (%) Isooctane per injection/mg Toluene (internal standard) (%) Toluene per injection/mg Isooctane relative to internal standard (%) 93.3 64.6 6.67 5.78 14.0 28.0 19.4 6.67 5.78 4.20 8.40 5.81 6.67 5.78 1.26 2.52 1.74 6.67 5.78 0.378 0.756 0.523 6.67 5.78 0.113 0.227 0.157 6.67 5.78 0.0340 0.0680 0.0471 6.67 5.78 0.0102 210 Analyst, February 1998, Vol. 1234.0 s 100 0 3.0 s 2.5 s 2.0 s 1.5 s 1.0 s 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Time/min Peak area (%) until an injection is desired. The actuator valve is then switched to the Load position for a specified time interval to fill the sample loop with vapor. When the loading interval is completed, the actuator switches back to the Inject position and the sample is flushed on to the GC column.Both configurations were evaluated during system optimization. SAW sensor The 158 MHz dual delay line SAW sensor and electronics packaging were obtained from Microsensor Systems. A solution of polyisobutylene (PIB) in toluene was air brushed on to the surface of the SAW to deposit a thin polymer film sufficient to cause a frequency shift of about 55 kHz. PIB was selected as the test coating because the response behavior of PIB-coated SAWs to toluene and isooctane has been documented previously. 16,17 During vapour exposure studies, the oscillating frequency of the SAW was monitored by a Fluke (Everett, WA, USA) PM- 6674 universal frequency counter interfaced to a dedicated 486 personal computer via an IEEE 488 bus. SAW sensor data collection and analyses were performed using software developed in-house with LabVIEW laboratory software (National Instruments, Austin, TX, USA). Automated system control System control software was developed using Microsoft Visual Basic, running under Windows 3.1.Arrows are included in Fig. 1 to indicate the flow of data and/or control during system operation. A typical experiment would run as follows. The VG- 400 is programmed to run a schedule of vapor dilutions. Whenever user-designated conditions are met, the computer actuates the injector valve to perform a sample injection in either the forward or reverse mode and record the sample loop temperature. The actuator signals the gas-chromatograph when the injection is complete, and the latter begins a chromatographic run and the computer begins data collection.When data collection is complete, the computer produces a report for that injection on disk and on the printer. The GC data were compared with previously prepared calibration curves to determine the actual mass of vapor delivered during a given injection. These mass/concentration data were then compared with concentrations calculated using thermodynamic data, and the effects of system configuration and operating conditions on the observed performance were determined.Results and discussion System optimization and calibration Optimization of the automated sample injection system involved evaluation of loop temperature, sample injection time and injection mode. Each of these factors was optimized individually, as discussed below. Actuator loop temperature The loop temperature can affect the amount of vapor contained in the sample loop (ideal gas law) and the relative amount of vapor adsorbed on the wall of the sample loop.Higher temperatures can minimize adsorbed vapor and permit rapid desorption, thereby giving narrower peaks. Adsorption effects appeared to be minimal, since loop temperature studies indicated that there was not much difference in the peak widths or relative areas at loop temperatures ranging from 30 to 100 °C. This being the case, the temperature was selected to obtain the largest range of vapor concentrations.Using a 250 ml loop and a loop temperature of 100 °C would permit GC analysis over the broad concentration ranges used during standard sensor studies, except for vapor streams near 100% saturation. Loading and injection times For each injection mode, the time interval for injecting in the forward mode or loading in the reverse mode was examined. In the forward mode, assuming a flow rate of 5.0 ml min21 of the GC carrier gas, it should take about 0.3 s to flush vapor from the 25 ml sample loop and 3 s to flush the 250 ml sample loop.With the 250 ml sample loop in place and a constant concentration of toluene from the VG-400, the effect of injection time on GC peak area was evaluated. The results are shown in Fig. 3. As expected, the peak size and area increase as the injection time increases from 1 to 3 s. Above an injection time of 3.0 s, however, all of the mass is transferred from the sample loop and the peak size and area remain constant, consistent with the Fig. 2 Schematic detail of two-position electronic actuator injection valve. In the Load position, the sample loop is in-line with the vapor generator. In the Inject position, the sample loop is in line with the GC system. Fig. 3 Plot illustrating the effect of injection time (i.e., the time the actuator is held in the inject position) on the GC peak size for toluene. Results are shown for the 250 ml sample loop.Analyst, February 1998, Vol. 123 211minimum injection time calculated above. Therefore, when using the system in the forward mode an actuation time of 3 s was used to ensure complete sample transfer from the sample loop to the gas chromatograph. In the reverse mode, the sample loop is initially in-line with the gas chromatograph. With a VG- 400 vapor stream flow rate of 100 ml min21 and a 250 ml sample loop, it should take only about 0.2 s to fill the loop completely with a vapor sample upon actuation.Once the sample loop is filled, it is switched back in-line with the GC system. Loading times were varied from 0.5 to 4.0 s. As expected, the peak size was unrelated to loading times over this range. However, because of the mechanics of the electronic actuator, a loading time of 2 s was used for all reverse mode injections in order to allow enough time for the actuator to switch positions fully. Injection mode The effect of injection mode on VG-400 performance was evaluated by comparing theoretically calculated vapor concentrations with generated vapor stream concentrations determined by automated injection and GC analysis.Calculations of the vapor concentrations determined by GC were performed with the following equation: vapor concentration ( g l GC counts count g loop size ( l) loop temp. (K) vapor temp. (K) -1 vapor -1 m m m m ) ( ) = ¥ RF (1) where the GC counts are peak area counts obtained from the GC data analysis, RFvapor is the GC response factor [(5.8 ± 0.21) 3 104 and (4.8 ± 0.17) 3 104 for toluene and isooctane, respectively] and the loop size is the volume of the sample loop (25 or 250 ml).The temperature conversion is needed to compare the concentration of the vapor sample injected at the loop temperature (373 K) with the vapor output of the VG-400 (288 K). These equations produce concentrations in mg l21, but vapor concentrations are more typically reported in mg l21. Errors associated with the response factors (±3.6%) and with reproducibility of loop injection peak area counts (±1.0%) result in anticipated vapor concentration errors of ±3.7%. To calculate the theoretical concentration of vapor streams produced by the vapor generator, the saturated vapor pressure of the solvent was calculated as log P A K B = - + 0 2185 .(2) where P is the saturated vapor pressure (Torr) at temperature K (kelvin) and a is the molar heat of vaporization (cal mol21). Values of a and B are solvent dependent and were obtained from ref. 21. For toluene these values are a = 8580.5 and B = 7.7719 392 and for isooctane a = 8548.0 and B = 7.934 852. (Note: a values reported in other references having units of J mol21 will differ for the values reported here.) These values are valid over temperature ranges from 226.7 up to 319 °C for toluene and from 236.5 up to 99.2 °C for isooctane. These ranges adequately span the operating temperature for vapor generation (15 °C).From the saturated vapor presure and the ideal gas law, the vapor concentration (in mg l21) was calculated. For toluene, the vapor pressure and 100% saturated vapor concentration at 15 °C were calculated to be 16.83 Torr and 86.27 mg l21, respectively. For isooctane, these values were 8.38 Torr and 180.4 mg l21, respectively. The theoretical vapor concentrations were then compared with the actual concentrations determined by GC with the automated injection value on-line.Results for toluene using the 25 and 250 ml sample loops in both the forward and reverse modes are shown in Figs. 4 and 5, respectively, and summarized in Table 3. For the 25 ml loop, displayed in Fig. 4, the actual vapor concentration delivered by the VG-400 deviates significantly from the theoretical value in both the forward and reverse modes. This is probably due to the flow restriction caused by the small diameter of this sample loop, which disrupts the pressure balance on the solvent reservoir side of the VG-400 flow stream.At low concentrations, for which the duty cycle requires less vapor input, the delivered concentration approaches 90% of the theoretically determined value. At higher concentrations, however, the delivered concentration drops to 75–80% of the theoretical value. It is worth noting that the deviation appears to be greater in the forward mode, during which the sample loop is in-line with the VG-400 during vapor equilibration.In contrast, delivered vapor concentrations are generally in good agreement with theoretical values over the concentration range studied for the 250 ml loop (see Fig. 5). Becuse of the larger sample size, the upper limit of this range was limited to 75% saturated vapor for toluene. Again, the delivered concentration was routinely higher in the reverse mode than in the forward mode. Similar results were observed for isooctane. It is worth noting for the 250 ml loop in the reverse mode that the delivered concentrations were consistently greater than the Fig. 4 Plot comparing the theoretical vapor concentrations (0), calculated using eqn. (2) and the ideal gas law, with actual delivered vapor concentrations, determined by GC using eqn (1). Results are shown for toluene using the 25 ml sample loop in both the forward (1) and reverse (4) injector modes. Fig. 5 Plot comparing the theoretical vapor concentrations (0), calculated using eqn.(2) and the ideal gas law, with actual delivered vapor concentrations, determined by GC using eqn. (1). Results are shown for toluene using the 250 ml sample loop in both the forward (1) and reverse (4) injector modes. 212 Analyst, February 1998, Vol. 123theoretical values. This may be due to the procedure used to calibrate the VG-400. Normal operation of the VG-400 requires identical flow rates of vapor and diluting gas into the dilution chambers. Flow controllers are adjusted to deliver 100 ml min21 to both the solvent reservoirs and to the dilution flow streams.The general assumption is that generation of solvent vapor does not significantly alter the flow rate from the bubbler. Although this assumption is valid for solvents of low or moderate vapor pressure, it may not be valid for volatile solvents such as toluene and isooctane. For example, the saturated vapor pressures for toluene and isooctane at 15 °C are 16.83 and 28.38 Torr, respectively.The contribution of vapor for these solvents can increase the flow rate from the bubbler by 2–4%. As a result, the mass of vapor delivered to the dilution chamber during the vapor generation cycle would be slightly greater than the theoretically calculated value. It is also worth noting that, in all cases, the best agreement between theoretical and delivered concentrations occurs at 12.5% saturation. As discussion earlier, this corresponds to a 50% duty cycle, during which vapor and dilution streams are introduced into the dilution chambers for equal amounts of time.Under these conditions, perturbations due to flow constrictions would be equal for both flow streams. Better linearity was observed using a loop with a larger inside diameter and operating in the reverse mode. Therefore, the 250 ml sample loop in the reverse mode was used for the SAW studies reported below. SAW studies After optimization and calibration of the components, the system was connected in-line with a SAW sensor.Using toluene, the frequency response of the 55 kHz PIB SAW sensor was studied at various vapor concentrations generated by the VG-400. Fig. 6 displays frequency shifts plotted against both theoretical vapor concentrations (circles), calculated as described previously, and the actual vapor concentrations (squares) determined chromatographically. Again, there is good agreement between the theoretical and actual concentrations at low percentage saturation, but significant deviations are noted at higher concentrations. As observed in Fig. 5, the actual delivered concentration at 70% saturated vapor is about 7% greater (70 mg l21) than the theoretical value (65 mg l21). The frequency response varies linearly with concentration up to 45 mg l21, with better linearity observed using the actual (GC) concentrations compared with the theoretical concentrations. In both cases, however, positive deviations from linearity appear above 50 mg l21.This non-linearity is believed to be related to significant changes in the viscoelastic properties of the PIB coating under high vapor loading, and will be discussed later. During typical SAW sensor testing the VG-400 is programmed for long-term, continuous vapor generation. For example, the SAW sensor may be subjected to multiple exposures of different solvent vapor streams covering broad concentration ranges, with the entire experimental session requiring close to 48 h of uninterrupted operation. During such long-term studies, variations in the test conditions can introduce errors into the calculated results.Although the VG-400 bubblers are thermostated at sub-ambient temperature (15 °C), the efficiency of this thermostated bath may decrease with variations in the ambient temperature, leading to variations in the vapor pressure of the solvent and, consequently, the concentration of the generated vapor stream. Also, the ambient temperature of the SAW will affect the thermodynamic partition coefficient of the vapor into the polymer coating, which will result in variations in the observed frequency shift for a given vapor concentration.In addition, long-term continuous vapor generation leads to a significant decrease of solvent in the bubbler, which may decrease the percentage saturation of the generated vapor stream. Continuous monitoring of the actual generated vapor concentrations can facilitate the interpretation of sensor responses in terms of such environmental factors.To demonstrate the advantages of the automated system, a coated SAW device was exposed to a long-term sequence of toluene vapors. The SAW frequency responses versus toluene vapor concentrations (determined by GC) are summarized in Fig. 7. Starting with a highly saturated vapor stream, the toluene concentration was systematically decreased (1-down) and then increased again to approximately 70 mg l21 (2-up). The frequency of the SAW was continuously monitored and the frequency shifts were noted as this sequence was repeated (3-down and 4-up).Four GC injections were performed using the auto-injection loop at each vapor concentration to determine the actual toluene concentration delivered to the SAW sensor. The entire experiment required over 14 h of continuous vapor generation. From the first run to the last run, the frequency response of the SAW systematically decreases by nearly 30%.By plotting the frequency shifts against the actual vapor concentrations, determined using the automated GC system, it can be seen that the decreases in the SAW frequency response are not due to decreases in the delivered vapor concentration. Although some of the decrease in SAW response might be attributable to increases in ambient temperature, the magnitude of the response decrease is too large to be due entirely to temperature effects. The systematic decrease in response may be attributable to hysteresis effects, and would be consistent with the viscoelastic Table 3 Comparison of theoretical (VG-400) and observed (GC) toluene vapor concentrations.Observed concentrations are reported in mg l21 and as a fraction of theoretical value (in parentheses). Data are also plotted in Figs. 4 and 5. Estimated errors for observed vapor concentrations are ±3.7% Observed/mg l21 25 ml loop 250 ml loop Vapor Theoretical/ (%) mg l21 Forward Reverse Forward Reverse 100.0 86.27 64.97 (0.76) 69.25 (0.80) — — 75.0 64.7 — — — 70.20 (1.08) 66.7 57.51 — — 55.37 (0.96) — 62.5 53.92 — — — 57.51 (1.07) 50.0 43.14 33.10 (0.77) 34.26 (0.79) 41.02 (0.95) 45.20 (1.05) 37.5 32.35 — — — 33.50 (1.04) 33.3 28.76 25.27 (0.88) 25.32 (0.88) 24.93 (0.87) — 25.0 21.57 19.41 (0.90) 19.74 (0.92) 17.51 (0.81) 23.50 (1.09) 16.6 14.35 12.72 (0.89) — — — 12.5 10.78 — 9.96 (0.92) 10.80 (1.01) 10.78 (1.00) 6.25 5.39 — — 6.35 (1.18) 5.80 (1.08) Analyst, February 1998, Vol. 123 213behavior of thin polymer films. Upon deposition of the polymer film, interfacial stress can affect the SAW response behavior. Under vapor loading conditions, the polymer film expands and relaxes, releasing this stress and leading to hysteresis in the SAW response. The behavior displayed in Fig. 7 is consistent with this explanation. The initial exposure to toluene resulted in large frequency shifts. As the film relaxed under continuous vapor loading, the amount of stress decreased and, hence, the magnitude of the frequency shift decreased. Once the stress has been alleviated, the SAW response becomes more reproducible. To verify this hypothesis, the coated SAW sensor was exposed to extended thermal treatment at above-ambient temperatures ( > 90 °C) to provide sufficient thermal expansion to alleviate stress within the polymer film.The coated SAW was then exposed to another series of vapor concentration cycles while the frequency shifts were recorded.These results are displayed in Fig. 8, and indicate good reproducibility between cycles. There is still a positive deviation from linearity at high vapor concentrations, and may be due to film resonance effects under high vapor loading.18 Comparison of SAW results An effective partition coefficient can be calculated from SAW frequency shifts and vapor concentration data as follows:16,17 K f f C SAW v s s v = D D r (3) where Dfv is the frequency shift observed upon sorption of vapor (in Hz), rs is the density of the coating (in kg m23), Dfs is the frequency shift of the SAW device due to the mass of the coating (in kHz) and Cv is the vapor concentration (in mg l21).16 Note that these KSAW values do not correspond to the thermodynamic partition coefficient which would be calculated from GC retention studies (KGC). The differences are due to viscoelastic factors that contribute to the observed SAW frequency response and result in consistently greater apparent KSAW values than the KGC values.KGC and KSAW values for a PIB-coated SAW device have been reported previously for isooctane and toluene by Grate et al.16 The values they obtained were taken at 25 °C using a 158 MHz SAW with a 280 kHz PIB coating. Values for log KSAW are reported in Table 4; KGC values16 are included for comparison. The KSAW values obtained for the 55 kHz PIB-coated SAWs are slightly larger than those reported by Grate et al.16 for the 280 kHz PIB-coated SAWs.These discrepancies may be due to interfacial adsorption on the surface of the SAW substrate. The devices used by Grate et al. were plasma cleaned prior to coating, whereas our devices were solvent cleaned in an ultrasonic bath. This could affect the interfacial adsorption for the devices. Also, since much thinner films were used in this study, interfacial and surface adsorption may represent a much larger percentage of the total response than would be the case for the thicker polymer film, in which bulk absorption would predominate. This behavior is consistent with SAW responses reported previously.17 Conclusion The goal of this work was to develop and evaluate an automated in-line GC system for verification of vapor stream concentra- Fig. 6 SAW calibration curve of sensor frequency shift for a 55 kHz PIBcoated SAW plotted versus toluene vapor concentration, using both theoretical (0) and actual (1) vapor concentrations. Results shown were obtained using the 250 ml sample loop in the reverse mode.The dashed line indicating a linear isotherm is included to demonstate positive deviations from linearity at highest toluene concentrations. Fig. 7 Plot of SAW frequency versus actual toluene vapor concentration during four continuous cycles of vapor exposure. Lines with arrows are used to indicate the experimental cycle. Series 1 (0) begins with a high concentration of toluene and cycles down to lower concentrations.Series 2 (7) begins with low toluene concentrations and increases to high concentrations. This cycle is repeated with decreasing concentrations for series 3 (4) and increasing concentrations for series 4 (1). The systematic decrease in SAW response is attributed to hysteresis effects as discussed in the text. Results are for a 55 kHz PIB-coated SAW using a 250 ml sample loop in the reverse mode. Fig. 8 Plot of SAW frequency versus actual toluene concentration after the sensor coating has been thermally relaxed.Results are shown for three repeated cycles as indicated, using a 250 ml sample loop in the reverse mode. Table 4 Comparison of log KSAW values Vapor Log KSAW (GC system, this work) Log KSAW (Grate et al.16) Log KGC (Grate et al.16) Toluene 3.65 3.44 2.76 Isooctane 3.32 3.12 2.28 214 Analyst, February 1998, Vol. 123tions. This was accomplished using a Varian gas chromatograph, a Valco electronic actuator and an IBM compatible computer using developed and purchased software.The automated operation of this system provided for unattended analysis and easy operation. The experimental studies showed that the system could validate vapor concentrations used in sensor testing. These concentrations proved to be more accurate than theoretically calculated values. The automated system can provide real-time information about the vapor stream concentrations, allowing for more accurate sensor calibrations and better interpretation of the frequency response behavior.In addition, the GC analysis is well suited to the verification of vapor mixtures, in contrast to other on-line methods. The use of sample loops of different volumes permits the use of this system over a broad vapor concentration range. This system has many applications, including the analysis of multi-component vapor streams used for sensor evaluation, process control and continuous, remote site monitoring.The hardware set-up is open to modification to ensure compatibility with the analytical task. The control program, written in Visual Basic, is simple to modify or expand to custom fit any application. The authors thank Larry Gregerson and John Tobias for their assistance with the construction and modification of an injection port and the temperature probe circuitry. Funding was provided from a Government Assistance in Areas of National Need grant and the NIU Chemistry Department. The work reported was performed in partial fulfillment of requirements for the MS degree from Northern Illinois University (T. Torkelson). References 1 Neill, G. P., Davies, N. W., and McLean, S., J. Chromatogr., 1991, 565, 207. 2 Stan, H. J., and Schwarzer, F., J. Chromatogr. A, 1993, 653, 45. 3 Grate, J. W., Rose-Pehrsson, S., and Venezky, D. L., Anal. Chem., 1993, 65, 1868. 4 Szakasits, J. J., and Robinson, R. E., Anal. Chem., 1991, 63, 114. 5 Filippini, C., Moser, J. U., Sonnleitner, B., and Fiechter, A., Anal. Chim. Acta, 1991, 255, 91. 6 Stansbridge, E. M., Mills, G. A., and Walker, V., J. Chromatogr., 1993, 621, 7. 7 Greenberg, J. P., Lee, B., Helmig, D., and Simmerman, P. R., J. Chromatogr. A, 1994, 676, 389. 8 Gan, J., Yates, S. R., Spencer, W. F., and Yates, M. V., J. Chromatogr. A, 1994, 684, 121. 9 Youchi, Y., Bandow, H., and Akimoto, H., J. Chromatogr., 1993, 642, 401. 10 Annino, R., J. Chromatogr., 1994, 678, 678. 11 Barrat, R. S., Analyst, 1981, 106, 817, and references cited therein. 12 Nelson, G. O., Gas Mixtures: Preparation and Control, Lewis, Chelsea, MI, 1992. 13 Grate, J. W., and Klusty, M., Naval Research Laboratory Report No. AD-A229, US Government Printing Office, Washington, DC, 1990. 14 Martin, S. J., and Frye, G. C., Proc. IEEE Ultrason. Symp., 1991, 393. 15 Martin, S. J., and Frye, G. C., Proc. IEEE Ultrason. Symp., 1992, 27. 16 Grate, J. W., Klusty, M., McGill, R. A., Abraham, M. H., Whiting G., and Andonian-Haftvan, J., Anal. Chem., 1992, 64, 610. 17 Ballantine, D. S., Anal. Chem., 1992, 64, 309. 18 Martin, S. J., Frye, G. C., and Senturia, S. D., Anal. Chem., 1994, 66, 2201. 19 Frye, G. C., Martin, S. J., Proc. Electrochem. Soc., 1993, 93, 51. 20 VG-400 Automatic Vapor Generation System Operating Manual, Microsensor Systems, Bowling Green, KY, 1990. 21 CRC Handbook of Chemistry and Physics, ed. Weast, R. C., CRC Press, Cleveland, OH, 52nd edn., 1972, p. D-146. Paper 7/05280D Received July 22, 1997 Accepted October 21, 1997 Analyst, February 1998, Vol. 123 215
ISSN:0003-2654
DOI:10.1039/a705280d
出版商:RSC
年代:1998
数据来源: RSC
|
10. |
On-line gas-diffusion separation and fluorimetric detection for the determination of acid dissociable cyanide |
|
Analyst,
Volume 123,
Issue 2,
1998,
Page 217-220
Esther Miralles,
Preview
|
PDF (64KB)
|
|
摘要:
On-line gas-diffusion separation and fluorimetric detection for the determination of acid dissociable cyanide Esther Miralles, Dolors Prat*, Ramon Compa�n�o and Merc`e Granados Departament de Qu�ýmica Anal�ýtica, Universitat de Barcelona, Diagonal 647, E-08028 Barcelona, Spain A flow injection system with gas diffusion separation and spectrofluorimetric detection is described for the determination of acid dissociable cyanide in waters. Cyanide diffuses through a microporous PTFE membrane from an acidic donor stream into a sodium hydroxide acceptor stream.The cyanide transferred reacts with o-phthalaldehyde and glycine to form a highly fluorescent isoindole derivative. Complete recovery of cyanide was found for Zn(CN)4 22, Cu(CN)3 22, Cd(CN)4 22, Hg(CN)4 22, Hg(CN)2 and Ag(CN)22 complexes and low recovery from Ni(CN)4 22. No recovery was obtained from the species that are considered as non-free cyanide producing, viz., Fe(CN)6 42, Fe(CN)6 32 and Co(CN)6 32.The sampling frequency was 10 h21 and the detection limit was 0.5 mg l21. The method tolerates a 40-fold excess of sulfide. The results of cyanide determination in water samples obtained with the proposed method are consistent with those obtained with APHA Method 4500 CN2 I for weak acid dissociable cyanide. Keywords: Cyanide; flow injection analysis; gas diffusion separation; fluorimetry; water Cyanide species in the environment originate mainly from industrial wastes. Their high toxicity makes it necessary to develop or improve methods for their determination at very low concentrations. Normally, cyanide testing requirements call for the determination of total cyanide, which includes free cyanide (CN2 and HCN) and metal–cyanide complexes. The US Environmental Protection Agency (EPA) has recently accepted that a distinction should be made between labile cyanides, which are readily bioavailable and extremely toxic, and total cyanides, which include those inert species of low toxicity that do not dissociate to cyanide ions.1 For this reason the procedure for the measurement of weak acid dissociable (WAD) cyanide is gaining acceptance.This method measures the cyanide released under weakly acidic conditions, which includes free cyanide and relatively unstable cyanide complexes such as those with Zn, Cd and Cu. Flow injection analysis (FIA) methods with gas diffusion separation have been shown to be efficient in liberating and totally recovering labile cyanide.1–3 Gas diffusion separations use hydrophobic microporous membranes through which the gas molecules diffuse.4 The hydrophobicity of the polymer [PTFE or poly(propylene)] prevents the filling of the pores by the reactor liquid phase.The advantage of the membrane is that all interferences related to ionic species are removed because only gaseous molecules diffuse. Hydrogen cyanide (HCN) is thus transferred across the membrane from an acidic donor solution into an alkaline receptor channel, where it is converted to CN2 and determined spectrophotometrically2,5,6 or amperometrically. 6,7 The method, however, does not separate cyanide from sulfide since H2S also diffuses across the membrane and must be previously precipitated with PbII8,9 or separated by ion chromatography (IC)7 if it interferes with the detection system.The additional use of ligand exchange reagents to improve the recoveries of cyanide species of medium stability has also been described in the literature. Dithizone and tetraethylenepentamine (TEP) were added to the sample to recover Hg and Ni cyanide species, respectively,3 and 1,10-phenanthroline was added to the acidic solution to release cyanide from the metal– cyanide complexes.2 Recently, we reported the use of o-phthalaldehyde (OPA) as a sensitive fluorogenic reagent for the detection of cyanide in flow systems.10 The aim of this paper was to apply this fluorimetric detection to determine acid dissociable cyanide in industrial waste water and river water by means of a flow injection system with gas diffusion separation.The method developed is rapid, sensitive, selective, reproducible and easy to automate and can be used as a simple routine method for acid dissociable cyanide monitoring. Experimental Apparatus The flow injection manifold consisted of a Minipuls 3 peristaltic pump (Gilson, Villiers le Bel, France), a Model 5041 injection valve (Rheodyne, Cotati, CA, USA) with a sample loop volume of 170 ml, and a laboratory-made gas diffusion module.Reaction and mixing coils were made from PTFE tubing of 0.5 mm id, and Tygon tubes were used for pumping the solutions. The gas diffusion unit consisted of two methacrylate blocks, each with a straight groove of 7.8 cm length, 2 mm width and 0.25 mm depth, between which the microporous Teflon membranes were placed. The two blocks were pressed together by six screws.The on-line detection was carried out with an LS-50 fluorescence spectrometer (Perkin-Elmer, Beaconsfield, Bucks., UK), equipped with a xenon lamp and a Model 176.752 flow cell (25 ml inner volume) (Hellma, M�ullheim, Germany), operated at excitation and emission wavelengths of 331 and 379 nm, respectively. A DR. LANGE LP2W (Neurteck Medioambiente, Zarautz, Spain) spectrophotometer was used for the determination of cyanide by the standard method. Reagents and solutions All chemicals were of analytical-reagent grade unless stated otherwise.Solutions of sodium cyanide (Carlo Erba, Milan, Italy) were prepared at 1000 mg l21 as CN2 in 0.1 m sodium hydroxide. K3Fe(CN)6 (Merck, Darmstadt, Germany), K4Fe(CN)6·3H2O (Merck), K2Ni(CN)4 (Aldrich, Milwaukee, WI, USA), K3Co(CN)6 (Aldrich), KAg(CN)2 (Johnson Matthey Chemicals, Karlsruhe, Germany) and Hg(CN)2 (Merck) were directly weighed and dissolved in 0.01 m NaOH. Solutions of Zn(CN)4 22, Hg(CN)4 22 and Cd(CN)4 22 were prepared by Analyst, February 1998, Vol. 123 (217–220) 217adding stoichiometric amounts of Zn(CN)2 (Aldrich), Hg(CN)2 (Merck) and Cd(NO3)2 (Merck, Standard Solution) to an NaCN solution. The cyanide species CuCN (Johnson Matthey Chemicals) was dissolved in a known excess of NaCN in 0.01 m NaOH to form the Cu(CN)3 22 species. All solutions were stored under refrigeration at 4 °C in amber-coloured glass bottles. Working solutions were prepared daily by dilution of the stock solutions in 1023 m sodium hydroxide. A stock 4 3 1022 m solution of OPA (Fluka, Buchs, Switzerland) was prepared in ethanol and diluted with borate buffer of pH 8.2 to give a 2 31023 m solution. Glycine (Merck) was used as a 2 3 1023 m solution in the same buffer prepared by dilution of a stock 2 3 1022 m aqueous solution.Working solutions were prepared daily. Chloramine-T (Merck), 1% m/v in water, was prepared weekly. Barbituric acid–pyridine solution was prepared by dissolving 15 g of barbituric acid (Merck), 75 ml of pyridine (Carlo Erba) and 15 ml of hydrochloric acid in 250 ml of water.Both solutions were stored under refrigeration. TEP (Fluka) was prepared as a 0.1% v/v solution in water. Hydrochloric acid (1 m) and 1022 m sodium hydroxide were used as donor and acceptor solutions, respectively, in the diffusion unit. Zinc acetate, sodium acetate and acetic acid were used in the distillation step of the standard method.Ultrapure water, MilliQ-plus (Millipore, Molsheim, France), of resistivity 18.2 MW cm21 was used throughout. All glassware used for the experiments was previously soaked in 10% v/v HNO3 for 24 h and rinsed in ultrapure water. Procedures Sample treatment Water samples were collected in 2.5 l glass bottles, preserved with sodium hydroxide (pH = 11–11.5), filtered if necessary and stored at 4 °C in the dark until analysis. FIA method The FIA system was set up as shown in Fig. 1(b). The sample was continuously introduced into the system at a flow rate of 0.85 ml min21. After acidification by merging with a flow of 1 m hydrochloric acid at 0.2 ml min21, the HCN liberated diffused through the Teflon membrane of the separator and was absorbed by an acceptor solution of 1022 m sodium hydroxide at 0.2 ml min21, which was continuously filling the injection loop. The diffused cyanide was then injected intos carrier stream and determined using the OPA–glycine fluorimetric method.10 Method 4500 CN2 I The sample (500 ml) was refluxed for 1 h in a macrocyanide distillation apparatus at pH 4.5, adjusted with acetate buffer solution, and in the presence of zinc acetate as described in the standard method.8 The distillate was collected into NaOH and analysed colorimetrically with the chloramine-T–pyridine– barbituric acid reagent at 578 nm in a 1 cm cell.Results and discussion Optimisation of the experimental parameters affecting cyanide diffusion Several gas diffusion FIA manifolds have been proposed in the literature in order to separate cyanide from the matrix and interferences.In the conventional FIA manifolds a volume of sample is injected into the donor stream; it diffuses through the membrane and is then collected in a flowing acceptor stream [Fig. 1(a)].6,11 Other manifolds improve the diffusion efficiency by means of an enrichment step.2,5 This can be achieved by continuously pumping the sample into the system and either working with a large sample : acceptor flow rate ratio using the manifold shown in Fig. 1(b), or using a stationary acceptor solution in the groove of the diffusion module, which also serves as the injection loop. In the latter, the sample is preconcentrated into a small volume of the stationary acceptor solution. Some preliminary experiments were performed with two different manifolds (Fig. 1). Working with similar hydrodynamic conditions in both systems, the peaks obtained with manifold 1(b) were six times higher than those obtained with manifold 1(a).In the conventional FIA manifold [Fig. 1(a)], the results showed a marked loss of sensitivity because of low efficiency in the diffusion of HCN through the membrane. These results led us to select and optimise the manifold shown in Fig. 1(b) where, in spite of the low efficiency of diffusion, an enrichment occurs in the slow-flowing acceptor channel. Two commercial hydrophobic microporous PTFE membranes were tested: Millipore FGLP (0.2 mm pore diameter, 175 mm thickness, porosity 80%) and Teflon tape (average pore size 10 mm, 110 mm thickness). Data on the Millipore FGLP membrane were provided by the supplier, whereas those on the Teflon tape were estimated from observations by electron microscopy. Although the porosity of the Teflon tape was not determined, a comparison of the electron microscopy images of both membranes revealed that the porosity of the Teflon tape was much lower than that of the Millipore FGLP membrane.Both membranes were tested in cross-current and normal current modes, but no significant differences were observed between the two operational modes, and further experiments were performed with both streams flowing in the same direction. Since the acceptor channel was flowing continuously, it was expected that the lower the flow rate, the greater would be the preconcentration factor of the HCN in the acceptor chamber.Fig. 2(a) shows the effect of flow rate on the amount of HCN diffusing across the membrane. A flow rate of 0.2 ml min21 for the acceptor stream was chosen since lower flow rates would make the analysis time too long. The effect of the flow rate of the hydrochloric acid channel was also investigated and it was Fig. 1 Diagrams of the flow injection systems. P, Peristaltic pump; G, gas diffusion unit; M, membrane; V, injection valve; L1, mixing coil (a) 60 cm 3 0.5 mm id, (b) 5 m 3 0.5 mm id; L2, reaction coil (2 m 3 0.5 mm id); L3, reaction coil (5 m 3 0.5 mm id); F, fluorimeter; and W, waste. 218 Analyst, February 1998, Vol. 123observed that an increase in the flow rate caused a slight decrease in the peak height because the contact time of the sample solution with the acceptor chamber was shorter. A flow rate of 0.2 ml min21 in the acid channel was selected. The effect of the flow rate of the sample stream was investigated. It can be observed from Fig. 2(b) that the higher the sampling flow rate the higher the peak height obtained, but no significant differences were found at flow rates > 1.5 ml min21. A flow rate of 0.85 ml min21 was chosen because of the availability of pumping tubes and to minimise sample waste. When the length of mixing coil L1 was varied no significant differences in the peak height were observed if free cyanide solutions were analysed. However, a 5 m 3 0.5 mm id coil was selected because if the cyanide is complexed it is expected to require a longer reaction time to be released from the complex.The ratio of the signal of a sample injected into the carrier stream without diffusion separation to that of a sample passing through the diffusion cell was about 1 for the Millipore FGLP membrane and 0.65 for the Teflon tape membrane. Therefore, there was no significant loss of sensitivity with respect to a system without diffusion. The efficiency of the diffusion of HCN across the membrane was estimated from the volume of sample pumped (during the time necessary to fill the injection loop), the volume of the injection loop, the signal obtained with the diffusion step and the signal from direct injection into the carrier stream.The transfer of HCN across the FGLP membrane was about 23% and across the Teflon tape about 14% at a 0.85 ml min21 sample loading flow rate and a 0.2 ml min21 flow rate in the donor and acceptor streams. The efficiency in the transport of HCN through the membrane was relatively low but it does not differ significantly from values reported in the literature.4 Although the FGLP membrane provided a slightly higher diffusion efficiency than the Teflon tape, the latter was selected for further experiments since it is easier to work with.Because of the thickness of the FGLP membrane it was necessary to tighten the two blocks of the diffusion module considerably to avoid leakage and this might easily cause fissures in the membrane.The reproducibility of the membrane behaviour was monitored for ten consecutive days. A standard of 50 mg l21 cyanide was injected daily into the diffusion manifold and the signal was compared with that obtained from the injection of the same standard without the diffusion separation. The ratio was 0.62 ± 0.07 . After working with the same membrane for more than 1 month, no memory effects or loss of efficiency were observed.Nevertheless, it is advisable to monitor the efficiency weekly. No significant variations were found when replacing the membrane, which shows that the Teflon tape is fairly homogeneous. Characteristics of the method The calibration graph was linear up to the maximum concentration tested, 200 mg l21 (If = 0.075 + 0.073 CCN2, r = 0.9996). The detection limit was 0.5 mg l21 cyanide, calculated as the concentration corresponding to three times the standard deviation of six repeated injections of a blank solution. The relative standard deviation obtained from ten successive injections of a 50 mg l21 cyanide solution was 2.2%.The gas diffusion technique separates cyanide from most ionic species in the sample but sulfide also diffuses through the membrane and reacts with OPA and glycine to form a fluorescent isoindole derivative. In order to study its interference, solutions of 10, 50 and 100 mg l21 cyanide were spiked with sodium sulfide. Sulfide gave no interference up to about a 40-fold excess.Carbonate was also studied as a potential interferent because it generates CO2. Carbonate did not interfere up to about 400 mg l21 when it was added to a solution containing 50 mg l21 cyanide. Above this value, the signal decreased, probably because the large amount of CO2 produced hinders the diffusion of HCN through the membrane. Nevertheless, the typical levels of carbonates in industrial effluent samples or river water are not expected to cause any problems.Taking into account that about 5 min were necessary for the system to reach a steady state, the throughput of the method was estimated to be 10 samples h21. Dissociation of metal–cyanide complexes Solutions of different metal–cyanide complexes at 50 mg l21 of cyanide were prepared and analysed by the proposed FIA method. The results (Table 1) show that, with 1 m HCl as donor solution, complete recovery of cyanide was obtained from the labile or relatively unstable complexes of CuII, ZnII, CdII, HgII and AgI low recovery from the Ni complex and no recovery from the more inert Fe and Co cyanide complexes.Fig. 2 (a) Influence of acceptor flow rate on peak height. (b) Influence of sample flow rate on peak height. A, FGLP membrane and B, Teflon tape. Conditions: 100 mg l21 CN2, 2 3 1023 m OPA, 2 3 1023 m glycine, 1022 m NaOH, 1 m HCl. Table 1 Recovery of cyanide species at different acidifying concentrations Recovery (%) Species (50 mg l21 as CN2) 0.1 m HCl 1 m HCl Cu(CN)3 22 96 101 Ni(CN)4 22 10 20 Zn(CN)4 22 99 98 Cd(CN)4 22 98 97 Hg(CN)2 60 99 Hg(CN)4 22 83 102 Ag(CN)22 101 100 Fe(CN)6 42 0 0 Fe(CN)6 32 0 0 Co(CN)6 32 0 0 Analyst, February 1998, Vol. 123 219Hence, the FIA method for acid dissociable cyanide excludes the amount of cyanide that is complexed with Ni, with respect to methods involving distillation, but includes complete recovery of cyanide from Hg complexes.1,8 Table 1 also shows that 0.1 m HCl as donor solution completely released cyanide from CuI ZnII, CdII and AgI complexes but not from HgII complexes.The recovery from the NiII complex was also lower than that obtained with 1 m HCl, indicating that HgII and NiII complexes have medium stability. Hence, the measurement of acid dissociable cyanide is system-dependent. These results are consistent with recoveries reported in the literature using gas diffusion separation with other detection systems.3,6 Cyanide can be effectively displaced from the nickel–cyanide complex if TEP is added to the sample before the analysis.3 Therefore, the use of TEP was also investigated.The recovery of nickel cyanide increased from 20 to 95% if 100 ml of 0.1% v/v TEP were added to a 50 ml sample of Ni(CN)4 22 containing 50 and 100 mg l21 of CN2. Accuracy The proposed method was tested by analysing industrial water samples (S1–S3) and comparing the results with those given by the WAD 4500 CN2 I standard method.8 Samples S1 and S2 came from plating baths and were expected to contain free cyanide plus Cd, Cu and Zn complexes, as well as low concentrations of Ni and Fe cyanide complexes, which originate from contact of the cyanide solutions with the surfaces being coated, but in relatively small proportion with respect to labile species.As these samples contained large amounts of cyanide, they were diluted before analysis by the FIA method. To simulate a polluted river water, sample S3 was prepared by dilution of sample S1 with cyanide-free river water.The results (Table 2) show that the difference between the two methods is not greater than 10%. Thus, the proposed FIA method rapidly provides a value of acid dissociable cyanide at room temperature which is related to the most toxic species, avoiding the tedious (1 h) distillation of the standard method. The accuracy was also evaluated from recovery studies of spiked cyanide-free natural river water samples.Samples S4 and S5 were spiked with known amounts of NaCN and S6–S8 were spiked with mixtures of labile metal–cyanide complexes at different proportions. The results, given in Table 2, show that the recoveries are usually above 90%. Although the results are satisfactory, recoveries were slightly low, which could be related to sample pre-treatment. Addition of sodium hydroxide to preserve the samples causes precipitation of metal hydroxides which may adsorb some cyanide and hence there may be losses in the filtration step.Conclusions A combined gas diffusion–fluorimetric detection method for the determination of labile cyanide species in water samples has been developed. The method gives results in a few minutes whereas the time for a cyanide determination using the standard distillation procedure is at least 2 h. The high sensitivity of the fluorimetric detection and the preconcentration step involved in the gas diffusion unit allow a detection limit of 0.5 mg l21 to be achieved.Other advantages of the fluorimetric detection include less hazardous reagents and less interference from sulfide with respect to the standard method. Although the method has been tested for the analysis of individual samples, it can be applied to near real-time monitoring and continuous surveillance, with the additional advantage of lower errors due to transport and sample storage. The authors thank the Chemical Engineering Department, ETSEIB, of the Universitat Polit`ecnica de Catalunya and the Aig�ues de Barcelona company for their collaboration.They also thank CICYT (AMB97-0387) for supporting this study and E.M. thanks CIRIT for a FI grant. References 1 Milosavljevic, E. B., Solujic, L., and Hendrix, J. L., Environ. Sci. Technol., 1995, 29, 426. 2 Zhu, Z., and Fang, Z., Anal. Chim. Acta, 1987, 198, 25. 3 Sebroski, J. R., and Ode, R. H., Environ. Sci. Technol., 1997, 31, 52. 4 van der Linden, W. E., Anal. Chim. Acta, 1983, 151, 359. 5 Kub�an, V., Anal. Chim. Acta, 1992, 259, 45. 6 Christmann, D., Renn, C., and Berman, R., Int. Lab., 1993, 23(8), 23. 7 Liu, Y., Rocklin, R. D., Joyce, R. J., and Doyle, M. J., Anal. Chem., 1990, 62, 766. 8 Standard Methods for the Examination of Water and Wastewater, ed. Clesceri, L. S., Greenberg, A. E., and Trussell, R. R., American Public Health Association; Washington, DC, 17th edn., 1989. 9 Wilmot, J. C., Solujic, L., Milosavljevic, E. B., Hendrix, J. L., and Rader, W. S., Analyst, 1996, 121, 799. 10 Miralles, E., and Prat, D., Compa�n�o, R., and Granados, M., Analyst, 1997, 122, 553. 11 Figuerola, E., Florido, A., Aguilar, M., and de Pablo, J., Anal. Chim. Acta, 1988, 215, 277. Paper 7/06925A Received September 24, 1997 Accepted October 30, 1997 Table 2 Comparison of WAD cyanide methods and recoveries from spiked real samples FIA method* Sample Method 4500 CN21 I Added/mg l21 Found/mg l21 S1 4.16 mg l21 — 3.90 ± 0.07 mg l21 S2 0.62 mg l21 — 0.56 ± 0.03 mg l21 S3 81 mg l21 — 73 ± 4 43 mg l21 — 34 ± 2 S4 — 10 8.7 ± 0.4 S5 — 50 45.5 ± 2 S6 — 10 8.8 ± 0.8 S7 — 50 46 ± 3 S8 — 100 96.3 ± 2.1 * S1–S3: mean of three determinations; S4–S8: mean of three separate spiked samples. 220 Analyst, February 1998,
ISSN:0003-2654
DOI:10.1039/a706925a
出版商:RSC
年代:1998
数据来源: RSC
|
|