摘要:

Critical Review Methods for the detection of polyether ionophore residues in poultry Christopher T. Elliott*, D. Glenn Kennedy and W. John McCaughey Veterinary Sciences Division, Stoney Road, Belfast, UK BT4 3SD. E-mail: elliotc@dani.gov.uk Summary of Contents Coccidiosis and coccidiostats Coccidiosis Drugs used to combat coccidiosis Structure of carboxylic acid ionophores Function of carboxylic acid ionophores Use of carboxylic acid ionophores as coccidiostats Coccidiostat residues and human health Legislative backgrounds to residue testing Analytical methods for ionophore residue detection Screening methods for the detection of ionophores In vitro assays Screening by thin-layer chromatography–bioautography Ionophore residue screening by immunoassay Antibody production Extraction procedures used in immunoassay procedures Sensitivities of immunoassays for ionophores Chemical methods for the detection of ionophores residues HPLC methods Methods requiring derivatisation 9-Anthryldiazomethane (ADAM) derivatisation Aromatic aldehyde derivatisation 1-Bromoacetylpyrene (BAP) derivatives Pyridinium dichromate oxidation Hydrazone derivative Methods requiring no derivatisation HPLC methods: summary Mass spectrometric methods Mass spectrometric methods: summary Pharmacokinetics of ionophores Monensin studies Salinomycin Lasalocid and maduramicin Narasin and semduramicin Summary Current position of ionophore residue testing in European national reference and control laboratories Range of substances monitored Analytical methods utilised Target matrices used for residue surveillance Comments and suggestions from survey References Keywords: Polyether ionophore; residue; health; immunoassay; bioassay; high-performance liquid chromatography; mass spectrometry; reference laboratory; review Coccidiosis and coccidiostats Coccidiosis Coccidiosis is a parasitic disease, caused by protozoa resident in the intestinal epithelium, which occurs wherever animals arehoused in small areas that are contaminated with coccidial oocysts.Historically, poultry have shown the greatest susceptibility to coccidiosis, because of the intensive nature of most of the poultry industry. In poultry, coccidiosis is caused by Eimeria spp., of which eight are known to cause serious clinical disease. A further 22 species cause less severe clinical effects in this species. The symptoms of coccidiosis in poultry may be one or more of the following: bloody diarrhoea, high mortality, reduction in feed and water intake, emaciation and loss of egg production.Much of the economic loss that is associated with coccidiosis is incurred prior to diagnosis. This makes prevention more important than treatment. Drugs used to combat coccidiosis A wide range of drugs is available for the prevention and treatment of coccidiosis. However, continuous prophylactic use of coccidiostats can cause a progressive loss of efficacy because of emerging drug resistance in the parasite population.Various Christopher Elliott is a Principal Scientific Officer in the drug residue laboratory of the Veterinary Sciences Division, Belfast, a European National Reference Laboratory for veterinary drug residue surveillance. He is the leader of a team developing new analytical methods capable of detecting drug residues in both laboratory and on-line environments. Current research is focused on the use of dry chemistry and biosensor based immunoassay techniques. He graduated from the University of Ulster with an M.Sc.in Biomedical Sciences and from the Queen’s University of Belfast with a Ph.D. in Veterinary Sciences. Analyst, June 1998, Vol. 123 (45R–56R) 45RO O O O CH3 H3C OH H H O H3C COOH CH3 H3C HO H H3CO H3C CH3 H H CH2CH3 CH2OH Monensin O COOH O O O O H CH3 H3C CH3 H OH OH O CH3 H CH3 CH3 OH CH3 H CH3 Salinomycin HO H3C COOH O O O H H3C H H3CH2C CH2CH3 H CH3 CH2CH3 CH3 OH H CH3 H OH H Lasalocid O COOH O O O O H CH3 H3C CH3 H OH OH O CH3 H CH3 CH3 OH CH3 H CH3 Narasin H3C CH3 CH3 CH3 CH3 O O H OCH3 CH3 Maduramicin H3C OH COOH OCH3 H3C O O O O CH3 HO H H O OH CH3 H CH3 H CH3 O OCH3 OCH3 CH3 CH3 O O H OCH3 CH3 Semduramicin H3C OH COOH OH H3C O O O O CH3 HO H H O OH CH3 H CH3 H CH3 O OCH3 CH3 CH3 H H strategies have evolved to cope with this, including changing drugs used in starter and grower feeds and rotation of drugs between successive batches of chickens.The first drugs used to treat coccidiosis were the sulfonamides. Subsequently, a wide range of compounds, such as clopidol, decoquinate and methylbenzoquate; nicarbazin; toltrazuril and diclazuril; robenidine; halofuginone; amprolium and ethopabate have replaced these. However, by far the most widely used compounds are the carboxylic acid ionophores. Structure of carboxylic acid ionophores The term ‘ionophore’ (ion bearer) was coined in 19671 to describe this group of naturally occurring compounds.The finding, in 1968, that monensin was effective as an anticoccidial prompted a search for other compounds possessing similar properties. Six members of this family: monensin, narasin, lasalocid, salinomycin, maduramicin and semduramicin (Fig. 1) have become widely used as anticoccidial drugs, particularly in the poultry industry. Chemically, the carboxylic acid ionophores appear to be open-chain molecules consisting of an array of heterocyclic ether-containing rings.When present as deprotonated anions, these compounds form stable, electrically neutral complexes with alkali metal cations. However, the involvement of the ionised carboxyl group is not always essential for metal binding. It is required for metal binding by lasalocid2 but is not for binding of Na+ by monensin.3 The carboxylic acid ionophores differ from each other in their affinity for individual metal ions. However, all of these compounds, with the exception of lasalocid, bind monovalent cations (e.g., Na+ and K+).Lasalocid has a tendency to form dimers,4 and can form complexes with divalent cations such as Mg2+ and Ca2+. The formation of metal complexes results in all of these compounds adopting a quasi-cyclic formation consequent to head-to-tail hydrogen bonding (Fig. 2). Function of carboxylic acid ionophores In normal cells, the intracellular concentration of Na+ is low and that of K+ is high.In the extracellular space, their relative concentrations are reversed. The concentration of Ca2+ is similar on both sides of the cell membrane. However, the free intracellular Ca2+ concentration is up to four orders of magnitude lower than that outside the cell. All of the carboxylic Fig. 1 Structures of the carboxylic acid ionophores. 46R Analyst, June 1998, Vol. 123O O CH3 CH3 HO H H H3C O O O O O CH2OH H3CH2C H H3C OH H H3C H3CO CH3 H – Na+ acid ionophores, including lasalocid, promote perturbations in the intracellular cation balance: Na+ increases and K+ decreases.Given the intracellular Na+ and K+ concentrations, the relative change in the intracellular concentration of Na+ is greater than that of K+. The increase in the intracellular concentration of Na+ is believed to result in Na+ : Ca2+ exchange, leading to a dramatic increase in the intracellular free Ca2+ concentration.5 Use of carboxylic acid ionophores as coccidiostats The carboxylic acid ionophores can dramatically reduce the mortality associated with coccidiosis.For example, in a floor pen trial,6 salinomycin, fed to broilers at a concentration of 100 mg kg21, resulted in mortality decreased to a rate of 0.1% compared with a 20.5% mortality rate found in an infected group that received no salinomycin. In the same birds the feed conversion ratio (ratio of feed consumed to liveweight increase) was 2.36 in the untreated, infected birds and 2.14 in the salinomycin-treated, infected birds.The feed conversion ratio is of major importance to poultry producers as it combines their greatest profit element (bird weight) with their greatest cost element (feed). Any reduction in the feed conversion ratio is inextricably linked to an increase in net income. This has produced the incentive for the manufacturers of veterinary pharmaceuticals to develop and market new carboxylic acid ionophores as coccidiostats. Of all of the coccidiostats developed to date, the carboxylic acid ionophores have proved to be remarkably free from problems of drug resistance, perhaps as a result of their rather non-specific mechanism of action.There has, however, been a gradual reduction in the sensitivity of coccidia to these compounds, indicating that resistant strains may be beginning to emerge. Unfortunately, there is as yet, no clear alternative to the carboxylic acid ionophores. Coccidiostat residues and human health Pressman and Fahim7 reviewed this area in 1983.Carboxylic acid ionophores are potent pharmacological agents, exerting marked cardiovascular effects in experimental animal systems. Most of these effects have been characterised using monensin as the model for the whole group. The principal effect is an increase in coronary flow, indicative of coronary dilatation. It has been estimated that a threshold dose for increased coronary flow in the dog, following injection of monensin, is 1.0 mg kg21.The threshold dose in man, following oral administration of monensin, in food, will inevitably exceed 1.0 mg kg21. In normal individuals, coronary dilatation is unlikely to have any adverse effect. However, it has been suggested7 that victims of coronary artery disease may be at an increased risk. In ischaemic areas of cardiac tissue, blood flow is already maximised in an attempt to maintain optimum perfusion of these areas. Induced dilatation, by an ionophore, of normal coronary vessels would tend to reduce further the perfusion of the partially occluded myocardium, an effect known as ‘coronary steal’.Any instances of adverse reactions to monensin would, inevitably, be swamped by the spontaneous occurrence of hypoxia in victims of coronary artery disease. Nonetheless, the possibility that monensin or other carboxylic acid ionophores in food could exacerbate the condition of affected individuals remains. Legislative background to residue testing The control of residues of veterinary medicines in cattle, sheep and pigs has been a cornerstone of the European Union’s (EU’s) agriculture policies to ensure consumer protection and promote even competition for markets for many years.Member States are required to monitor food animals for a range of legal and illegal compounds to provide assurance to consumers about the safety and wholesomeness of their food (Council Directive 96/23/EC). Currently, the EU is assessing the safety of all pharmacologically active compounds that are administered to food-producing animals, and is attempting to set a legally binding maximum residue limit (MRL) for each compound.With effect from January 1, 2000, only those compounds listed in Annexes I, II and III of Council Regulation 2377/90/EC can be used in food-producing animals. Annexe I lists compounds with an established MRL, Annexe II lists those compounds that are generally recognised as safe and therefore need no MRL, and Annexe III lists those compounds with a provisional MRL. No MRLs have yet been set by the EU for any of the carboxylic acid ionophores. The EU legislation on the control of residues in poultry meat (Directive 96/23/EC) came into effect on July 1, 1997, and will be implemented by Member States from January 1, 1998.Member States will include coccidiostats in their national residues testing programmes. For cost-effective detection of most other classes of veterinary drugs, Member States have adopted, wherever possible, a two-tier testing system of: screening and confirmatory tests (Commission Decision 93/256/EEC).Screening tests are rapid, high volume, low cost tests that are designed to select samples for confirmatory analysis. They are biased to produce no false negatives, but may produce a low level of false positives. In effect, they classify large numbers of samples as being either ‘negative’ or ‘potentially positive’.All samples in the latter category are then subjected to a confirmatory test. These are low volume, high cost tests geared to produce no false positives, and a mimimal rate of false negatives. Taken together, this combination of screening and confirmatory tests provides an efficient and costeffective means to control veterinary drug residues in foodproducing animals. In anticipation of the forthcoming inclusion of carboxylic acid ionophores in national residue testing programmes, this review assesses available screening and confirmatory methods and the pharmacokinetics of these compounds.Analytical methods for ionophore residue detection Screening methods for the detection of ionophores The development of an individual analytical method, with the complex and interrelated requirements of a screening procedure (fast, reliable, broad spectrum, sensitive, inexpensive), poses a formidable challenge. The method developed will relate to the specific needs of the end-user and the analytical technologies available at the time.The analytical techniques used to detect ionophore residues not surprisingly date back to the time of the discovery of monensin. Despite the variety and depth of coccidiostat screening assays, the topic has not been reviewed since 1985 Fig. 2 Binding of sodium by monensin. Analyst, June 1998, Vol. 123 47Rwhen Weiss and MacDonald8 summarised detailed information on the bioassays available at that time.As with many types of screening assays for drug residues there has been a change in developmental emphasis from bioassays to immunoassays, a trend that is likely to continue. In vitro assays Amongst some of the earliest methods designed to detect the presence of ionophore drugs were cell culture procedures. Strout and Ouellette9 and McDougald and Galloway10 described similar techniques whereby the effects of the presence of anticoccidial agents, including the ionophores, on the growth of Eimeria tenella in cell culture were monitored.Kidney cells in culture were inoculated with sporozoites and incubated for 72 h. After this time the cultures were fixed and stained and viewed under a light microscope. In untreated controls, normal coccidia development was noted whereas in cultures with ionophores added there was a predominance of abnormal sporozoites. Detection limits of the order of 10 000 mg l21 for monensin were claimed. In a similar assay, the detection method was altered to allow a more quantitative measure of anticoccidial activity.11 An ELISA (enzyme-linked immunosorbent assay) was performed following the cell culture stage of the procedure.This allowed the detection of mature Eimeria tenella antigens present in culture supernatants. The inhibition of growth elicited by the ionophore compounds caused a reduction in the presence of these antigens. Ionophore concentrations as low as 0.5 mg l21 could be detected by this technique.In summary, these in vitro assays were able to differentiate between many classes of anticoccidial drugs and, in particular, ionophores. This was achieved by observing the particular effects on the growth inhibition detected during culture. All the studies mentioned performed the assays with drugs in buffer solutions and were primarily designed to define the nature of the anticoccidial activity of various drugs. In addition, it should be noted that there are requirements for lengthy incubation periods in these procedures, making them less attractive for routine analytical testing. There is, however, great scope to use refined versions of these assays to develop their potential to be used as receptor-based assays for ionophore residue presence in complex biological samples.Such assays would have the capability of being able to detect multi-ionophore residues at high sensitivity. Screening by thin-layer chromatography–bioautography (Table 1) The ionophores, being antibiotics, inhibit the growth of certain bacteria.The earliest ionophore residue detection methods developed used this property.17 However, the basic microbiological assay technique was unsuccessful in reliably detecting monensin residues, owing to the interference of sample components, even after extensive sample extraction, and the use of thin-layer chromatography (TLC) as a clean-up procedure became the accepted method by many workers.In their review article of TLC methods for monensin residue testing, Weiss and MacDonald8 detail the evolution of this methodology. Only minor changes appear in later published techniques. The first description of a TLC–bioautographic assay and the model for many to follow appeared in 1967.12 Tissue samples (muscle, liver and kidney) were sequentially extracted into methanol and carbon tetrachloride before spotting onto silica gel thin-layer plates. Fat samples required additional purification prior to analysis by means of solid-phase columns packed with silica gel.The bioautography was performed by melting agar over the surface of the TLC plate seeded with Bacillus subtilis innoculum. Following an overnight incubation, the sizes of zones of inhibition were measured to determine monensin presence. The sensitivity of this procedure was determined as being about 25 mg kg21. Modifications to this assay were devised by Okada et al.18 These included a more refined sample clean-up, improvement in the agar and the use of an alternative developing solvent.The net result was to achieve improved chromatography of monensin and enhanced zones of inhibition, leading to improved detection limits of about 10 mg kg21 in muscle, liver, kidney and fat. Further improvements to the TLC–bioautography assay were introduced by Vanderkop and MacNeil,13 who identified the silylation of glassware used in extractions as an important means of increasing the amount of monensin recovered.In addition, further optimisation of the TLC solvent system employed and bioassay conditions resulted in a relatively fast and simple method capable of routine use. Salinomycin residues have also been successfully detected by TLC–bioautography.14 An extraction procedure based on acetone and ethanol partitioning and zone visualisation following TLC separation utilising the organism B. stearothermophilus resulted in a procedure capable of detecting down to 10 mg kg21 salinomycin. Vanderkop and MacNeil15 developed a multi-TLC–bioautography method that allowed the identification of monensin, salinomycin and lasalocid residues.This Table 1 TLC–bioassay for ionophore residues LOD‡/ Ionophore* Matrix† Extraction Clean-up TLC conditions Organism mg kg21 Ref. Mo M,L,K,F MeOH and CCl4 Silica-based SPE§ on fat extracts only Silica gel plates developed in carbon tetrachloride–benzene–methyl cellosolve (80 + 10 + 5) Bacillus subtilis Å 50 12 Mo Unspecified poultry tissues MeOH and CCl4 None 6D silica gel developed in chloroform– methanol–acetone–glycerol (98 + 60 + 40 + 2) Bacillus subtilis 250 13 Sa M,L,K,F Acetone and light petroleum None Unspecified plates developed in hexane– diethyl ether–methanol–acetic acid (70 + 30 + 4 + 0.5) Bacillis stearothermophilis Å 10 14 Mo, Las and Sa L, K MeOH and CCl4 None Silica gel developed in ethyl acetate– acetonitrile (50 + 50) Bacillus subtilis 45–1000 15 Mo M,L,K,F MeOH and CCl4 Silica SPE§ except for Mo Silica gel 60 plates developed in ethyl acetate Bacillus subtilis 10–12.5 16 * Mo = Monensin, Sa = salinomycin, Las = lasalocid.† M = muscle, L = liver, K = kidney, F = fat. ‡ LOD = Limit of detection. § SPE = Solid-phase extraction. 48R Analyst, June 1998, Vol. 123differed from their previous method13 for monensin by altering the developing solvents to ethyl acetate and acetonitrile (50 + 50). Sensitivities ranging from 450 to 1000 mg kg21 in chicken liver and kidney extracts were achieved. In another procedure, capable of simultaneously detecting monensin and lasalocid residues, samples (serum and lung) were extracted using a chloroform-based procedure.19 The organism B.stearothermophilus was used to determine residue presence. The detection levels of the assay were not given. In summary, there are TLC–bioautographic procedures capable of detecting most of the ionophore residues in tissue.These methods rely heavily on lengthy solvent-based extraction systems and the sensitivities achieved vary widely. Ionophore residue screening by immunoassay Antibody production. Owing to the requirement for extensive clean up prior to analysis by TLC–bioassay workers sought methodologies that would act as alternative ionophore screening tests. As with many other forms of veterinary drug screening analysis, attention focused on the use of immunoassays. The prerequisite analytical tool for this methodology is an antibody that recognises the ionophore in a sensitive and specific manner.To produce such antibodies the ionophore must be linked to a suitable carrier protein to enable the complex to be recognised as foreign and thus evoke a humoral immune response in immunised animals. Ionophore conjugates have been successfully produced by a variety of synthetic approaches (Table 2). Heitzman et al.20 converted monensin salt to an acid and conjugated this to ovalbumin (OVA) by means of a mixed anhydride reaction.30 Rabbits were immunised with 1 mg aliquots of the conjugate and a polyclonal serum was harvested after several months.Pauillac et al.21 also used the mixed anhydride conjugation method to couple a succinylated monensin derivative to both OVA and bovine serum albumin (BSA) to prepare polyclonal (rabbit) and monoclonal (mouse) antibodies. Mount and Failla22 used an alternative means of preparing a rabbit polyclonal serum to the same compound. In this procedure monensin salt was converted to a reactive bromoacetate derivative which spontaneously coupled to BSA.In a more recent study, Godfrey et al.24 coupled the carboxylic acid form of monensin to haemocynanin and thyroglobulin using the reactive N-hydroxysuccinamide. In summary, antibodies (both polyclonal and monoclonal) have been raised to most of the ionophores. These tend to be highly specific and, with the exception of salinomycin/narasin, antibodies, do not have significant cross reactivity profiles.It should be noted, however, that little data were produced to show the cross reactivities against ionophore metabolites. Extraction procedures used in immunoassay procedures (Table 3). Only a limited number of the ELISA procedures have been developed and validated to detect ionophore residues in poultry samples.23,24,27–29,31 The matrix most often chosen in these assays has been liver.The extraction procedures have generally relied on solvent partitioning. However, one excep- Table 2 Methods and procedures employed in the production of monoclonal and polyclonal antibodies to ionophores Conjugation Ionophore Derivative produced procedure Carrier protein Species Cross-reactivity Ref. Monensin Monensin acid Mixed anhydride Ovalbumin Rabbit O-Desmethylmonensin (42%) 20 Monosuccinyl monensin Mixed anhydride BSA* Rabbit and mouse None found 21 Monensin bromoacetate Direct BSA Rabbit None found 22 Monensin acid Active ester Keyhole limpet haemocyanin Rabbit None found 23 Monensin acid Mixed anhydride Transferrin Rabbit None found 24 Salinomycin Salinomycin hemisuccinate Mixed anhydride BSA Mouse Narasin ( > 100%) 25 Salinomycin hydrazide Active ester Not stated Mouse 26 None Carbodiimide Human serum albumin Rabbit Narasin (100%) 27 Lasalocid None Carbodiimide Human serum albumin Sheep None found 28 Maduramicin None Carbodiimide Human serum albumin Rabbit None found 29 * BSA = Bovine serum albumin. Table 3 Extraction procedures applied to ELISA methods for poultry sample analysis for ionophore residues.Assay performance data are also outlined for individual methods LOD‡/mg kg21 LOQ§/mg kg21 Ionophore Matrix* Extraction Clean-up RSD† (%) or mg l21 or mg l21 Ref. Monensin L Water–acetonitrile Hexane–diethyl ether 8.5–19.6 2.9 4.6 24 M, L, K, F, S Proteolytic digest Immunoaffinity 8.7–16.3 0.09–1.99 No data 23 Salinomycin L Water–methanol Dichloromethane 19.9 No data Å 50 31 Se, M, L Water–acetonitrile hexane 11–31 0.016–0.09 0.028–0.15 27 Lasalocid Se, M, L Water-acetonitrile Hexane 5–34 0.1–0.18 0.16–0.29 28 Maduramicin Se, M, L Ethyl acetate or water– acetonitrile Diethyl ether 5–33 0.01–0.02 No data 29 * Se = Serum, M = muscle, L = liver, K = kidney, F = fat, S = skin.† RSD = Relative standard deviation. ‡ LOD = Limit of detection. § LOQ = Limit of quantification.Analyst, June 1998, Vol. 123 49Rtion to this is the method of Godfrey et al.,23 who utilised the technique of immunoaffinity chromatography. This procedure greatly reduced the volumes of organic solvents required in the extraction process, allowing for the possibility of some form of automation to be introduced into the process. The disadvantage of this technique was the requirement to perform a proteolytic digest on the sample material prior to extraction. Sensitivities of immunoassays for ionophores (Table 3 ).In general, the ELISA procedures developed to detect ionophore residues were significantly more sensitive compared with the earlier TLC–bioautography based screening assays. In summary, a number of immunoassays have been developed which can detect trace levels of ionophore residues. The extraction procedures required for these assays are similar to those of the TLC procedures. However, in contrast to the TLC methods, which were capable of being multi-ionophore residue detection methods, the ELISAs normally detect only one, or occasionally two ionophore compounds.The other important difference between these two methodologies relates to the detection of ionophore metabolites. The TLC–bioautographic technique relies on biological activity of the ionophores and is unlikely to measure significant amounts of biologically inactive metabolites. The ELISA methods, owing to unavailability of metabolite reference standards, are largely untested for the cross reactivity profiles of the antibodies against them, i.e., it is not known if the ELISAs also detect the presence of these metabolites present in samples.However, examples of structural similarities which exist between parent ionophores and metabolites which have been documented32 are recognised by the antibodies to at least some degree. In summary, a wide variety of screening assays have been developed to detect residues of ionophores. The choice of method used in a laboratory will often be dictated by the availability of suitable expertise, suitable equipment and the ability to purchase appropriate commercial products.Chemical methods for the detection of ionophore residues During the last decade, there has been a dramatic increase in the number of published HPLC and mass spectrometric methods for the detection of carboxylic acid ionophores. However, there are still relatively few sensitive and specific methods from which regulatory analysts can choose.The chemical analysis of the carboxylic acid ionophores has been reviewed on two previous occasions, in 19858 and 1995.33 HPLC methods Lasalocid, alone among the carboxylic acid ionophores, has a fluorescent chromophore. All of the HPLC-based assays that have been developed to determine the other carboxylic acid ionophores in tissue require derivatisation to introduce a suitable chromophore. Methods requiring derivatisation (Table 4) 9-Anthryldiazomethane (ADAM) derivatisation. ADAM reacts with carboxylic acid groups to form a highly fluorescent derivative.It was first used for the analysis of monensin residues in bovine muscle in 1985.34 Monensin, extracted from beef liver, was first acetylated with acetic anhydride to form acetyl esters from the two hydroxyl groups in monensin. It is not clear why this step was needed. This derivative was then reacted with ADAM to form monensin–9-anthryldiazomethane.An additional silica clean-up was required prior to HPLC analysis. No sample traces were presented. Since then, three other methods have been described35–37 that have used ADAM as a derivatising agent. The method described by Hoshino et al.37 was similar to the method described above,34 although the acetylation step was not included. No traces from negative tissues were presented. Martinez and Shimoda subsequently described a modification of their original assay34 that was capable of detecting monensin, salinomycin, narasin and lasalocid.36 Acetylation of sample extracts was apparently necessary for the determination of monensin, narasin and salinomycin.However, they claimed that lasalocid did not form an ester derivative following treatment with acetic anhydride, and so this step could be omitted when attempting to determine lasalocid. Using their method, narasin and salinomycin were incompletely resolved. Furthermore, lasalocid eluted on the tailing edge of a very major matrix peak.Most of the methods that use ADAM derivatisation suffer from poor sensitivity (limit of quantification > 100 mg kg21),37, or poor recoveries35 ( < 60%) or both.36 In addition, with one exception,37 clean-up procedures are lengthy. However, the simpler clean up procedure described by Hoshino et al.37 was accompanied by very poor sensitivity. All of these methods required an Table 4 HPLC methods for carboxylic acid ionophores requiring derivatisation Derivatisation Detection LOQ‡/ Recovery Analyte* mode Derivatising agent Extraction Clean-up method† mg kg21 (%) Ref.Mo Off-line Anthryldiazomethane MeOH–H2O Alumina/CH2Cl2/LH-20/acetylate Fl 365/418 50 71–96 34 Mo off-line Anthryldiazomethane MeOH–H2O CHCl3/silica/CHCl3 Fl 365/412 10 46–78 35§ Las, Ma, Na, Sa Off-line Anthryldiazomethane MeOH–H2O Alumina/CH2Cl2/LH-20/acetylate Fl 365/418 150 57–90 36¶ Mo Off-line Anthryldiazomethane Orthophosphoric acid CHCl3 Fl 365/412 500 68–82 37 Mo Post-column Vanillin MeOH–H2O CCl4/silica VIS 520 25 82–96 38 Se Post-column Vanillin MeOH–ammonia C8/silica VIS 522 40 82–107 39 Las, Mo Na, Sa Post-column Vanillin Isooctane–ethyl acetate Silica VIS 520 3–10 71–94 40· Sa Post-column Vanillin MeOH–H2O CH2Cl2 VIS 520 100 89 31, 41 Sa Post-column Vanillin Acetone Light petroleum VIS 520 10 83–98 42 Sa Post-column Dimethylaminobenzaldehyde Ethanol–propan- 2-ol Microwave VIS 592 10 87–100 43 Mo, Sa Off-line 1-Bromoacetylpyrene Acetonitrile Ethyl acetate/silica/derivatise/ Florisil Fl 360/450 100 66–96 44 Sa Off-line Pyridinium dichromate MeOH CCl4/silica/C18/derivatise/silica UV 225 100 95–102 45 * Mo = Monensin, Sa = salinomycin, Las = lasalocid, Ma = maduramicin.† Fl = Fluorescence, VIS = visible, UV = ultraviolet; all values in nm. ‡ LOQ = Limit of quantification. § Recoveries very low (46 and 56%) at 10 and 100 mg g21. ¶ Acetylation not needed for lasalocid.· Lasalocid not derivatised in muscle. Separate extraction used for lasalocid in liver. 50R Analyst, June 1998, Vol. 123additional clean-up step, using silica gel, following derivatisation to remove unreacted ADAM, increasing the complexity of the method. Aromatic aldehyde derivatisation. Vanillin and dimethylaminobenzaldehyde react with hydroxyl groups in the Komarowsky reaction. These post-column reagents, prepared in methanolic sulfuric acid, react with the carboxylic acid ionophores at elevated temperatures.The ionophores decompose in a poorly understood reaction, to form coloured products. This property was first utilised in the determination of carboxylic acid ionophores in 1973, when a colorimetric method for monensin in feeds, based on the reaction between vanillin and monensin, was reported.46 Although subsequently applied to the determination of monensin, narasin and salinomycin in animal feeds in 1985,47 the use of vanillin as a derivatising agent for carboxylic acid ionophore residues in tissues has only been reported recently.33,38–42 These methods offer a significant improvement over the earlier ADAM-based assays.Much simpler one- or two-step clean-up procedures are possible coupled with, in most cases, improved sensitivity. Typically, the extraction step consists of either a liquid–liquid3,41,42 or a silica gel solid-phase extraction.40 This method, for lasalocid, monensin, narasin and salinomycin, claimed limits of quantification of between 3 and 10 ng g21, with excellent recoveries.However, the detection of lasalocid required a significant alteration to be made to the extraction procedure, diminishing the claim that the four main carboxylic acid ionophores are included. Derivatisation with vanillin has also been applied to the determination of semduramicin, in an assay that employed sequential C8 and silica gel solid-phase clean-up steps. Chromatograms were very clean and free from possible interfering matrix components.This is the only chemical method that has been described for this compound, to date, and has a limit of quantification of 40 mg kg21 One promising method for salinomycin43 combines a very simple extraction procedure with derivatisation with dimethylaminobenzaldehyde. Samples are mixed with ethanol and propan-2-ol and salinomycin is extracted by a brief pulse of microwave energy from a domestic microwave oven. This assay has a limit of quantification of 10 mg kg21 and recoveries range from 87 to 100%.However, the authors did not compare their novel extraction procedure with more conventional approaches using incurred positive samples. Such derivatisation/degradation methods are limited by the fact that other hydroxylcontaining compounds in the sample extract will also yield coloured products. This may unacceptably increase the false positive rate of the assay. 1-Bromoacetylpyrene (BAP) derivatives.The Kryptofix K222-catalysed formation of pyrenacyl derivatives of carboxylic acid ionophores has been the subject of study by Asukabe’s group. In 1984, they reported the formation of fluorescent ester derivatives of salinomycin and monensin in standard solutions.48 Subsequent studies showed that this technique could also be used for lasalocid and narasin.49 Although this group went on to develop a method for the determination of lasalocid, monensin and salinomycin in animal feeds,50 another Japanese group described the use of the same derivative for the analysis of residues of monensin and salinomycin.44 Their method produced well-resolved peaks, with no interfering peaks at the retention time of either compound.This method, with a limit of quantification of 100 mg kg21 was not, however, particularly sensitive. Recoveries ranged from 66 to 96%. Pyridinium dichromate oxidation. Dimenna et al.45 described a method that gave good recoveries, but poor sensitivity, for salinomycin in chicken skin/fat.Following extraction with methanol, and clean up by partitioning into carbon tetrachloride, followed by silica gel and C18 solid-phase extraction, salinomycin was derivatised using pyridinium dichromate. This oxidises the allylic hydroxyl group to form an a,b-unsaturated ketone, with strong UV absorbance at 225 nm. However, the limit of quantification of this method was poor, at 100 mg kg21. Hydrazone derivative. Another derivatising reagent has been applied to the determination of salinomycin in standard solutions.51 It relied on the formation of a hydrazone derivative, with maximum absorbance at 419 nm.However, this reagent has not been applied to the analysis of carboxylic acid ionophore residues in tissues. Methods requiring no derivatisation (Table 5) The only intrinsically fluorescent carboxylic acid ionophore is lasalocid. If excited at l = 308–315 nm, lasalocid fluoresces at 400–430 nm.This intrinsic fluorescence of lasalocid is highly pH-dependent. At pH 8.3 the intensity of emission is some two orders of magnitude greater than that at pH 3.0. However, the corrosive nature of such an alkaline pH towards silica columns has limited the applicability of this property. Weiss et al.52 developed a method for the determination of lasalocid in tissues that was an extension of a similar method developed for the determination of lasalocid in blood.57 Lasalocid was extracted into acetonitrile, and defatted using hexane.An aliquot was dried and residues were reconstituted in water, saturated with HPLC mobile phase. The lasalocid was then extracted into the mobile phase, which was a complex mixture of tetrahydrofuran, methanol, hexane and ammonia. Lasalocid was, however, incompletely resolved from a matrix component by the use of two silica columns (Whatman Partisil 10) connected in series. Despite this, a limit of quantification of 25 mg kg21 was achieved, with recoveries averaging 72%.Traces of negative samples were not presented. This method was subsequently found to be robust in two multi-laboratory studies using chicken fat58 and bovine liver59 as the test matrix. A similar extraction procedure, column and mobile phase was used by Kozak and Wisniewska-Dymytrow,53 a method that had a comparable limit of quantification (20 ng g21) and recovery (85%). No traces were presented. Ishikuro54 developed a method that involved extraction with acetonitrile and defatting with hexane.In this case, however, a silica Sep-Pak cartridge was used to effect clean-up. Reversed-phase HPLC (Unisil Pack 5C18) was used to resolve lasalocid. The pH of the mobile phase was 3.0. The limit of quantification was 50 mg kg21, although the detection limit was approximately 5 mg kg21. The recovery achieved ranged from 85 to 92%. Blank tissue traces were not presented. Table 5 HPLC methods for lasalocid residue detection requiring no derivatisation Ionophore Extraction Clean-up Detection method* LOQ†/mg kg21 Recovery (%) Ref.Lasalocid Acetonitrile Hexane wash/dry/extract into basic mobile phase Fl 310/430 25 72 52 Lasalocid Acetonitrile Hexane wash/dry/extract into basic mobile phase Fl 310/440 20 85 53 Lasalocid Acetonitrile Hexane wash/silica Sep-Pak Fl 310/420 50 85–92 54 Lasalocid Methanol Extract with CCl4/silica Sep-Pak Fl 310/420 400 86–100 55 Lasalocid Acetonitrile Extract with CCl4/silica Sep-Pak Fl 310/425 2 66-76 56 * Fl = Fluorescence; all values in nm.† LOQ = Limit of quantification based on lowest concentration at which the method was validated. Analyst, June 1998, Vol. 123 51RHorii et al.55 described a method in which liver was extracted with methanol, and then extracted with carbon tetrachloride. Following silica Sep-Pak clean-up, samples were run on a Nucleosil 100 C18 reversed-phase column in a mobile phase at pH 7.0.Although these workers claimed a limit of quantification of 10 ng g21, validation data were produced only at 400 and 2000 ng g21. Blank tissue traces were not shown. Tarbin and Shearer56 presented a well-validated method in which lasalocid was extracted from tissues and eggs using acetonitrile. The extract was partitioned, with salting-out, into carbon tetrachloride–acetonitrile and dried. The extract was further cleaned up using a silica solid phase column. HPLC was carried out in a basic mobile phase using either a polymeric PLRP-S column or a porous graphitic carbon column (Hypercarb), both of which were stable over a wide pH range.The polymeric column proved unsuitable for the determination of lasalocid in eggs, because of the presence of an interfering peak. However, the porous graphitic column, which required a mobile phase containing 5% tetramethylguanidine in acetonitrile to achieve good peak shape, was unaffected by this interference.The limit of quantification of the assay was 2 mg kg21 in tissue and 10 mg kg21 in egg. Recoveries ranged from 66 to 72% (egg) and 76% (tissue). HPLC methods: summary No truly multi-residue HPLC method for all of the carboxylic acid ionophores has yet been developed. Monensin, narasin and salinomycin can readily be combined in a single analytical procedure. To include lasalocid residue detection requires alterations to the extraction procedure.36,40 Derivatisation with vanillin, with UV/VIS detection, appears to offer the best sensitivity and simplest extraction procedure.For lasalocid, however, HPLC with fluorescence detection is the method of choice. The great simplicity of the microwave extraction procedure developed by Akhtar and Croteau43 merits further examination, given the limit of quantification and recoveries obtained, particularly if the assay is as effective in measuring incurred residues of salinomycin as it is in the analysis of salinomycin spikes. Some of the more sensitive HPLC methods that have been described rely on the use of solvents such as chloroform and carbon tetrachloride.24,38,56 These methods may be of limited value in the modern analytical laboratory, given the increasing costs of waste disposal.Mass spectrometric methods (Table 6) Increasingly, legislative demands are requiring analysts, wherever possible, to use mass spectrometric detection for the confirmation of residues of veterinary drugs. However, few mass spectrometric methods have been published for the determination of carboxylic acid ionophores in tissues in the last 15 years.The lack of published methods reflects, in part, the instability and relative insensitivity of these compounds in GC systems. Most of the published methods have relied on the emerging technology of LC–MS. The earliest methods for the mass spectrometric determination of these compounds did not have the advantages offered by LC coupled to MS.As a consequence, these tended to be cumbersome. In 1983, Weiss et al.60 described a pyrolysis GC– mass spectrometric method for lasalocid in bovine liver. Samples were extracted and cleaned up using semi-preparative HPLC. Following pyrolysis of a ditrimethylsilyl derivative, two GC peaks were produced for lasalocid following a retroaldol cleavage. For the retroaldol aldehyde, which contained both trimethylsilyl (TMS) groups, two ions (m/z 381, MH+; and m/z 291, MH+ 2 TMS) were monitored.The MH+ ion of the retroaldol ketone could not be measured because of matrix interference, but ions corresponding to the loss of water (m/z 337) and to a second decomposition product (m/z 211) were monitored. Validation data were not presented, but the method was capable of detecting lasalocid at a concentration of 51 mg kg21 in an incurred liver sample. A fast atom bombardment (FAB) MS method, capable of measuring 10 mg kg21 monensin in chicken fat, was described by Blomquist et al.61 In addition to [M+H]+, ions corresponding to a sodium adduct and to loss of water could be detected.Validation data were not presented. Takatsuki et al.35 described the determination of a trimethylsilyl ester of monensin using GC–MS. However, the method was very insensitive, being capable of detecting monensin only at concentrations above 5000 ng g21. Validation data were not presented. Monensin is, however, not well suited to GC–MS because of its high molecular mass and their low volatility.Since then, the increasing availability of LC–MS has enabled sensitive methods for the determination of such compounds to be developed. Horii et al.62 described a thermospray LC–MS–MS method that could detect 60 mg kg21 lasalocid in chicken liver. The molecular ion was not a significant ion. Instead, a fragment at m/z 337 was chosen as the parent ion, and daughter ions at m/z 319, 281 and 237 were measured. Validation data were not presented. The determination of maduramicin in chicken fat using thermospray LC–MS and thermospray LC–MS–MS has also been reported.63 Fat was extracted with acetonitrile and partitioned with hexane.The extract was cleaned up using an alumina column prior to analysis. This clean-up procedure was insufficient to permit analysis using LC–MS. However, the increased selectivity offered by tandem mass spectrometry permitted the detection of maduramicin at concentrations down to approximately 50 mg kg21.Schneider et al.64 were the first to apply electrospray LC–MS to the determination of carboxylic acid ionophores. They described a method that could detect semduramicin at concentrations of approximately 30 mg kg21 in chicken liver. In addition to the [M+Na]+ ion at m/z 895, daughter ions at m/z 852 and 834, corresponding to the sequential loss of CO2 and H2O, were detected. Table 6 Mass spectrometric methods for the determination of carboxylic acid ionophores Ionophore Extraction Clean-up Detection method* LOQ†/mg kg21 Recovery (%) Ref.Las Acetonitrile Hexane wash/dry/extract into basic mobile phase Pyrolysis GC–MS > 50 NS‡ 60 Mo Acetonitrile Hexane wash/dry/extract into basic mobile phase FAB–MS > 10 NS 61 Mo Chloroform Silica gel GC–MS > 5000 NS 35 La NS NS Thermospray LC–MS > 60 NS 62 Ma Acetonitrile Partition with hexane/alumina Thermospray LC–MS 250 61–80 63 Se Methanol C8 and silica-solid phase cartridges Electrospray LC–MS NS NS 64 Las Acetonitrile Hexane–toluene Electrospray LC–MS 5 77–88 65 Mo.Sa, Na Methanol Hexane–toluene Electrospray LC–MS 5 85–117 66 * Mo = Monensin, Sa = salinomycin, Las = lasalocid , Ma = maduramicin, Se = semduramicin, Na = narasin. † LOQ = Limit of quantification. ‡ NS = Not stated. 52R Analyst, June 1998, Vol. 123More recently, this laboratory has published methods for the determination of lasalocid65 and monensin, salinomycin and narasin66 in biological matrices using electrospray LC–MS.In both instances, the extraction procedures were relatively simple. For lasalocid,65 acidified sample homogenates were extracted into acetonitrile, and then cleaned-up by liquid–liquid extraction into hexane–toluene (1 + 1, v/v). For monensin, salinomycin and narasin,66 samples were homogenised with methanol, and the drugs extracted into hexane–toluene (1 + 1, v/v) under basic conditions. These simple extraction procedures, combined with the compound-specific selectivity offered by electrospray LC–MS, resulted in the production of very clean traces.The limit of quantification in each assay was 5 mg kg21. In both assays, a pseudo-molecular ion, corresponding to [M + Na]+, was the most prominent ion. It is often possible to adjust the amount of analyte fragmentation by increasing the skimmer cone voltage. However, for all four of these compounds, increasing the cone voltage had only a limited effect on the degree of fragmentation obtained.Mass spectrometric methods: summary Electrospray LC–MS appears to offer the best possibility for the confirmation of low levels of the carboxylic acid ionophores. The published methods are very sensitive, and require very simple sample extraction and clean-up steps. However, as with the HPLC methods described above, no truly multi-residue method has yet been developed. In summary, as with the screening assays for ionophore residues, a wide variety of chemical procedures have been developed.The choice of method used in a laboratory will be dictated by the availability of suitable expertise and equipment. Pharmacokinetics of ionophores An important aspect of all drug incorporation in an animal’s diet is the pharmacokinetic characteristics displayed; i.e., the absorption, distribution, metabolism and clearance of the drug from the treated animal. These parameters all play a part in determining the matrix most suitable to monitor a particular drug residue.In addition, MRL legislation also determines whether or not the parent drug, metabolite(s) or both must be measured. However, MRLs for ionophores have not yet been established and the interpretation of pharmacokinetic information in this review will be restricted to distribution and clearance aspects. A number of pharmacokinetic studies have been performed on ionophores in poultry. Some of the data generated have given conflicting results with regards to residue concentrations detected.The most likely explanations for these variations lie in the differences in analytical techniques used and, specifically, in wide variations in method sensitivities, extraction recoveries and the ability to differentiate metabolites from the parent compound. Monensin studies The earliest reported pharmacokinetic study of monensin was that of Herberg and Van Duyn.67 Following administration of [3H]monensin orally at a concentration of 360 mg kg21 for two days, chickens were slaughtered at zero, two and four days after withdrawal. The radioactivity present in a wide range of samples was measured and converted to monensin equivalents.The highest residue concentrations were found in liver samples (7660 mg kg21) taken without withdrawal from the drug. Residues were still detectable in liver after the fourth day of withdrawal. (Table 7). From this study it was clear that the main excretion route for monensin was faecal as opposed to urinary (99 : 1 ratio).A similar study was performed by Donoho68 with the exception that a more typical monensin medication regime was employed, i.e., 120 mg kg21 for 2 weeks. Substantially lower residue concentrations were detected in this latter study but residues in liver at day 0 (495 mg kg21) remained the highest concentrations found (Table 7). In contrast to these radiolabelled studies, pharmacokinetic experiments performed using Table 7 Pharmacokinetic data on monensin residue concentrations present in poultry samples Matrix/sample type Mean concentration detected/ Medication mg kg21 or mg l21 level/mg Withdrawal Analytical per kg of feed Duration period/d assay Serum Fat Liver Muscle Ref. 360 2 d 0 Radiolabel 1840 2320 7660 2160 67 2 630 440 1470 1120 4 340 140 720 700 120 1week 0 TLC–bioassay 110 39 29 18 1 17 ND* ND 2 ND ND ND 120 2 weeks 0 TLC–bioassay 106 60 68 1 ND ND 0 Radiolabel 100 495 10 1 12 272 11 2 66 222 ND 3 50 149 ND 5 50 106 ND 120 2 weeks 0 TLC–bioassay 1320 900 1300 700 69 1 ND 100 180 ND 2 ND ND ND ND 100–120 30 d 0 ELISA 79 15 26 23 120 2 weeks 0 ELISA 93 1 18 24 2 111 3 ND * ND = Not detected.Analyst, June 1998, Vol. 123 53RTLC–bioassay and ELISA as the means of residue quantification18,24,69 gave more consistent results (Table 7). A noticeable exception to this was the results shown by Godfrey et al.,23 who found lower concentrations of monensin residues in liver compared with fat and muscle.This finding may be due to the longer medication given during this study (30 d). A study employing administration of [3H]monensin gave clear evidence of extensive drug metabolism. Donoho et al.70 concluded that only 70 and 7% of the radioactivity found in fat and liver samples, respectively, was due to the presence of the parent form of monensin. Analysis of liver extracts by chromatography revealed that many monensin metabolites were present.In summary, monensin residues were readily detected in liver, fat and muscle for between zero and two days post drug withdrawal with liver generally giving the highest concentrations. After this time only the more sensitive methods were able to detect the presence of monensin residues. These data suggest that the withdrawal periods recommended by monensin manufacturers of 3 d are adequate to deliver meat with low residue content, provided good farm management is employed.Salinomycin (Table 8) In all salinomycin pharmacokinetic studies reviewed, a 2 week medication period at doses ranging from 30 to 132 mg kg21 was given. Wide variations in the concentrations found between these studies were reported. Atef and co-workers69,71 reported 1100 mg kg21 salinomycin in the liver of chickens given no withdrawal; this compared with 3 mg kg21 found by Kennedy et al.27 under similar conditions. Such extensive differences may be best explained by the different analytical methods used in these studies.The salinomycin data (Table 8) clearly show that after a one day withdrawal from the drug, the concentrations present in the edible tissues of the animals are very low and barely detectable. Lasalocid and maduramicin (Table 9) Pharmacokinetic data for lasalocid and maduramicin residues in poultry were available only in two and one studies, respectively. However, from this limited information it appears that lasalocid residues persist in edible tissues for longer than either salinomycin or monensin and may require a substantially longer withdrawal period than any of the other ionophore compounds studied to date.Narasin and semduramicin No pharmacokinetic information was found on either of these compounds. However, owing to structural similarities between narasin and salinomycin and semduramicin and maduramicin, similar residue profiles may be anticipated. Summary The concentrations of the various ionophores added to poultry feed in the studies reviewed varied in concentration from low to high ppm levels; the duration of medication ranged from a few days to a few weeks.There did not, however, seem to be simple correlations between the residue content found in samples when compared with either the size of dose or duration of treatment. This may be, in part, due to the various types of analytical techniques used to quantify the residue content and, in part, to differences in the procedures used to medicate and house the experimental animals.What is clear, however, is that following withdrawal of medicated feed, residue concentrations fall quickly and that the time periods specified in manufacturer’s guidelines will normally ensure that high residue levels do not reach the food chain. This is assuming that the correct level of medication has been given for the correct time period and that possible sources of cross-contamination of the drug have been removed.If any of these factors are not adequately controlled, then the likelihood of high residue concentrations in edible tissues increases sharply. One important area for drug residue depletion studies that appears to have been overlooked is that of eggs. Reports have shown72,73 that in both caged and free-range birds a significant number of positives have been found in eggs. This is Table 8 Pharmacokinetic data on salinomycin residue concentrations present in poultry samples Matrix/sample type Mean concentration detected/ Medication mg kg21 or mg l21 level/mg Duration/ Withdrawal Analytical per kg of feed weeks period/d assay Serum Fat Liver Muscle Ref. 60 2 0 TLC 700 900 1100 670 71 1 ND* 110 100 ND 2 ND ND ND ND 30 2 0 HPLC 73 9 ND 42 1 37 ND ND 3 ND ND ND 60 2 0 HPLC 162 13 ND 1 62 ND ND 3 ND ND ND 90 0 HPLC 201 24 60 1 73 ND ND 3 ND ND ND 60 2 0 ELISA 8 3 1 27 66 2 0 ELISA 203 33 1 ND 132 0 351 1 ND * ND = Not detected. 54R Analyst, June 1998, Vol. 123particularly true of lasalocid. Work in this matrix is at an early stage. Current position of ionophore residue testing in European national reference and control laboratories A survey of all EU National Reference Laboratories (NRLs) and Control Laboratories (CLs) associated with drug residue analysis was performed, by the authors of this review, between May and September, 1997. This exercise was intended to provide an overview of the extent of coccidiostat residue analysis in Europe and a breakdown of the analytical methods currently employed.Of the 28 laboratories surveyed, replies were received from 23. All EU member states with the exception of Germany which carried out the survey provided a written response. In total 15 of the laboratories were actively involved in coccidiostat residue analysis; a further three had an interest in the subject and provided comments and opinions on the subject. This review will only deal with the array of information provided that was relevant to ionophore residue testing.A full report of the survey findings will appear elsewhere. Range of substances monitored Assays for monensin and salinomycin were performed in ten of the laboratories surveyed, eight of which also included narasin in their monitoring programmes. Only four laboratories had analytical methods to detect lasalocid residues; only one had methods for maduramicin, and semduramicin residue analysis was not performed in any laboratory.At the time of this survey, none of the European laboratories that replied was able to measure residues of the whole group of ionophores. Furthermore, only one laboratory predicted that it would have the analytical capability to detect five of the six compounds in 1998. Analytical methods utilised The most frequently used analytical method employed to determine the presence of ionophore residues was undoubtedly HPLC. Many laboratories indicated that this technique was used as both a screening and confirmatory tool (49 and 35%, respectively).The second most popular analytical methodology was that of TLC and TLC–bioassays; these again were used in screening and confirmatory modes (27 and 39%, respectively). The other form of screening test used was immunoassay (15%). The more widespread use of this method was hampered by the lack of commercial test-kit availability. A small proportion of laboratories also indicated that LC-MS screening and confirmatory assays were available (9 and 26%, respectively).Target matrices used for residue surveillance The majority of laboratories chose to perform ionophore residue analysis on either muscle or liver samples. A smaller proportion of those surveyed also tested feed, egg and kidney samples for these substances. In only a single case was urine tested and no laboratory indicated that serum analysis was performed. Comments and suggestions from survey The recipients of the survey were asked for their opinions on the subject of the current state of coccidiostat residue analysis, what future changes are likely to occur within their own establishments and what progress they would like to see on a European scale.There was virtually unanimous agreement that there was a great need for improvement in the field of coccidiostat monitoring. Whilst some laboratories thought that their current array of analytical methods was satisfactory, it was clearly felt that there was wide scope for the development of methods, both screening and confirmatory, which would offer greater sensitivity and be multi-residue in nature.A second important aspect of the survey was to determine the interest in organising interlaboratory studies and a workshop on the theme of coccidiostat residue analysis. There was widespread approval of these two suggestions and a general feeling was conveyed that an organised provision of literature and analytical methods would be of considerable help to laboratories.In summary, the area of coccidiostat monitoring in European residue testing laboratories has apparently been given a lower priority than other classes of veterinary drugs. This is most likely because there has not previously been a requirement under National Surveillance Schemes to include coccidiostat monitoring. There also appears to be a realisation that this will change in 1998 and many laboratories are looking for guidance and help with the implementation of satisfactory residue monitoring programmes which will enable them to comply with European legislative requirements.In conclusion, the ionophore drugs have been, and continue to be, widely used in the control of coccidiosis in poultry. Analytical methods have been developed to detect trace levels of ionophore residues in treated birds. There has been a move towards ELISA for screening tests and LC–MS for confirmatory procedures.However, despite these improvements there is a lack of fast and reliable multi-residue methods available. With a requirement for coccidiostat residue monitoring appearing in new European Directives there is a clear need to move in this direction. The authors thank Dr. Sabine Hahnau of the BGVV, Berlin, for assisting in the preparation and subsequent analysis of informa- Table 9 Pharmacokinetic data on lasalocid and maduramicin residue concentrations present in poultry samples Matrix/sample type Mean concentration detected/ mg kg21 or mg l21 Medication level/ Duration/ Withdrawal Analytical Ionophore mg per kg of feed weeks period/d assay Serum Lung Liver Muscle Ref.Lasalocid 90 2 0 ELISA 250 450 1 28 1 90 80 5 2 30 70 1 3 10 60 1 Lasalocid 75 1 0 Bioassay 1360 6.9 19 Maduramicin 5 2 0 ELISA 70 106 28 29 1 30 60 6 2 20 20 3 3 10 20 3 Analyst, June 1998, Vol. 123 55Rtion provided in the survey of European residue testing laboratories. References 1 Pressman, B.C., Harris, E. J., Jagger, W. S., and Johnson, J. H., Proc. Natl. Acad. Sci. USA, 1967, 58, 1949. 2 Reed, P. W., Methods Enzymol., 1979, 55, 435. 3 Pressman, B. C., Annu. Rev. Biochem., 1976, 45, 501. 4 Johnson, S. M., Herrin, J., Liu, S. J., and Paul, I. C., J. Am. Chem. Soc., 1970, 92, 4428. 5 Pressman, B. C., and Fahim, M., Annu. Rev. Pharmacol. Toxicol., 1982, 22, 465. 6 Danforth, H. D., Ruff, M. D., Reid, W. M., and Johnson, J., J.Poultry Sci., 1977, 56, 933. 7 Pressman, B. C., and Fahim, M., Adv. Exp. Med. Biol., 1983, 161, 543. 8 Weiss, G., and MacDonald, R., J. Assoc. Off. Anal. Chem.,1985, 68, 971. 9 Strout, R. G., and Ouellette, C. A., Exp. Parasitol., 1973, 33, 477. 10 McDougald, L. R., and Galloway, R. B., Exp. Parasitol., 1973, 34, 189. 11 Gilbert, J. M., Fuller A. L., and McDougald, A., Colloq. INRA, 1989, 49, 253. 12 Donoho, A. L., and Kline, R. M., Antimicrob. Agents Chemother., 1967, 763. 13 Vanderkop, R. A., and MacNeil, J. D., J. Assoc. Off. Anal. Chem., 1989, 72, 735. 14 Heil, K., Fleur, P., and Cieleszky, S., J. Agric. Food Chem., 1984, 32, 997. 15 Vanderkop, R. A., and MacNeil, J. D., J. Chromatogr., 1990, 508, 386. 16 Kondo, F. J., Food Prot., 1980, 51, 786. 17 Kline, R. M., Stricker, R. E., Coffman, J. D., and Bikin, H., J. Assoc. Off. Anal. Chem., 1970 , 53, 49. 18 Okada, J., Higuchi, I., and Kondo, S., J. Food Hyg. Soc., 1980, 21, 177. 19 Stipkovits, L., and Juhasz, S., Acta Vet.Hung., 1987, 35, 237. 20 Heitzman, R. J., Carter, A. P., and Cottingham, J. D., Br. Vet. J., 1986, 142, 516. 21 Pauillac, S., Halmos, T., Labrousse, H., Antonakis, K., and Avrameas, S. J., Immunol. Methods, 1993, 164, 165. 22 Mount, M. E., and Failla, D. L., J. Assoc. Off. Anal. Chem., 1986, 70, 201. 23 Godfrey, M. A. J., Luckey, M. F., and Kwasowski, P., Food Addit. Contam., 1997, 14, 281. 24 Crooks, S. R. H., Traynor, I. M., Elliott, C.T., and McCaughey, W. J., Analyst, 1997, 122, 161. 25 Miller, G. A., Clark, R. C., and Jessee, W., Hybridoma, 1986, 5, 353. 26 Elissalde, M. H., Beier, R. C., Rowe, L. D., and Stanker, L. H., J.Agric. Food Chem., 1993, 41, 2167. 27 Kennedy, D. G., Blanchflower, W. J., and O’Dornan, B. C., Food Addit. and Contam., 1995, 12, 93. 28 Kennedy, D. G., Blanchflower, W. J., and O’Dornan, B. C., Food Addit. Contam., 1995, 12, 83. 29 Kennedy, D. G., Blanchflower, W. J., and O’Dornan, B.C., Food Addit. Contam., 1995, 14, 27. 30 Erlanger, B. F., Borek, F., Beiser, S. M., and Lieberman, S., J. Biol. Chem., 1957, 228, 479. 31 Muldoon, M. T., Elissalde, M. H., Beier, R. C., and Stanker, L. H., J. Agric. Food Chem., 1995, 43, 1745. 32 Herberg, R., Manthey, J., Richardson, L., Cooley, C., and Donoho, A., J. Agric. Food Chem., 1978, 26, 1087. 33 Asukabe, H., and Harada, K. I., in: Chemical Analysis for Antibiotics used in Agriculture, ed. Oka, H., Nakazawa, H. Harada, K.I., and MacNeil, J. D., AOAC International, Arlington, VA, 1995, p. 121. 34 Martinez, E. E., and Shimoda, W., J. Assoc. Off. Anal. Chem.,1985, 68, 1149. 35 Takatsuki, K., Suzuki, S., and Ushizawa, I., J. Assoc. Off. Anal. Chem., 1986, 69, 443. 36 Martinez, E. E., and Shimoda, W., J. Assoc. Off. Anal. Chem., 1986, 69, 637. 37 Hoshino, Y., Horie, M., Nose, N., and Iwasaki, H., J. Food Hyg. Soc. Jpn., 1985, 26, 585. 38 Moran, J. W., Rodewald, J. W., Donoho, A. L., and Coleman, M. R., JAOAC Int., 1994, 77, 885. 39 Ericson, J. F., Calcagni, A., and Lynch, M. J., JAOAC Int., 1994, 77, 577. 40 Gerhardt, G. G., Salisbury, C. D. C., and Campbell, H. M., Food Addit. Contam., 1995, 12, 731. 41 Muldoon, M. T., Elissalde, M. H., Beier, R. C., and Stanker, L. H., Immunoassays for Residue Analysis, Am. Chem. Soc., Symp. Ser., vol. 621, 1996, ch. 9, p. 109. 42 Akhtar, M. H., El-Sooud, K. A., and Shehata, A. A., Food Addit. Contam., 1996, 13, 897. 43 Akhtar, M. H., Croteau, L. G., J. Environ. Sci. Health, 1996, B31, 117. 44 Miyakawa, H., Horii, S., and Kokubo, Y., J. Food Hyg Soc., 1995, 36, 725. 45 Dimenna, G. P., Creegan, J. A., Turnbull, L. B., and Wright, G. J., J. Agric. Food Chem., 1986, 34, 805. 46 Golab, T., Barton, S. J., and Scrogs, R. E., J. Assoc. Off. Anal. Chem., 1973, 56, 171. 47 Blanchflower, W. J., Rice, D. A., and Hamilton, J. T. G., Analyst,1985, 110,1283. 48 Asukabe, H., Sasaki, T., Harada, K.-I., and Suzuki, M., .J. Chromatogr., 1984, 295, 453. 49 Asukabe, H., Murata, H., Harada, K.-I., and Suzuki, M., J. Chromatogr. A, 1993, 657, 349. 50 Asukabe, A., Murata, H., Harada, K.-I., Suzuki, M., Oka, H., and Ikai, Y., J. Agric. Food Chem., 1994, 42, 112. 51 Mathur, A. K., J. Chromatogr. A, 1994, 664, 284. 52 Weiss, G., Felicito, N. R., Kayakaty, M., Chen, G., Caruso, A., Hargroves, E., Cowley, C., and MacDonald, A., J. Agric. Food Chem., 1983, 31, 75. 53 Kozak, A., and Wisniewska-Dymytrow, H., Med. Weter., 1995, 51, 160. 54 Ishikuro, E., Eisigaku Zasshi, 1989, 30, 160. 55 Horii, S., Miyahara, K., and Momma, C., J. Liq. Chromatogr., 1990, 13, 1411. 56 Tarbin, J. A., and Shearer, G., J. Chromatogr., 1992, 579, 177. 57 Kayakaty, M., and Weiss, G., J. Agric. Food Chem., 1983, 31, 81. 58 Frank, L. R., and Barnes, C. J., J. Assoc. Off. Anal. Chem., 1989, 72, 584. 59 Newkirk, D. R., and Barnes, C. J., J. Assoc. Off. Anal. Chem., 1989, 72, 581. 60 Weiss, G., Kaykaty, M., and Miwa, B., J. Agric. Food Chem., 1983, 31, 78. 61 Blomquist, G. B., Jansson, M., Ryhage, E. R., and Osterdahl, B.-G., J. Agric. Food Chem., 1986, 34, 274. 62 Horii, S., Miyahara, K., and Maruyama, T., Shokuhin Eisigaku Zasshi, 1991, 32, 30. 63 Stout, S. J., Wilson, L. A., Kleiner, A. I., da Cunha, A. R., and Francl, T. J., Biomed. Environ. Mass Spectrom., 1989, 18, 57. 64 Schneider, R. P., Lynch, M. J., Ericson, J. F., and Founda, H. G., Anal. Chem., 1991, 63, 1789. 65 Blanchflower, W. J., and Kennedy, D. G., Analyst, 1995, 120, 1129. 66 Blanchflower, W. J., and Kennedy, D. G., J. Chromatogr. 1996, 675, 225. 67 Herberg, R. J., and Van Duyn, E., J. Agric. Food Chem. 1969 17, 853. 68 Donoho, A. L., J. Anim. Sci., 1982, 58, 1528. 69 Atef, M., Ramadan, A. and Abo-El-Sooud, K., Br. Poultry Sci., 1993, 34, 195. 70 Donoho, A. L., Herberg R. J., Zornes, L. L., and Van Duyn, E., J. Agric. Food Chem., 1982, 30, 909. 71 Atef, M., Ramadan, A., Youssef, S. A. H., and El-Sooud, A., Res. Vet Sci., 1993, 54, 179. 72 Medicines Act Veterinary Information Service, Veterinary Medicines Directorate, UK, 21st edn., John C. Alborough Ltd., Ringshall, Suffolk, UK, January, 1997. 73 Kennedy, D. G., Blanchflower, W. J., Hughes, P. J., and McCaughey, W. J., Food Addit. Contam., 1996, 13, 787. Paper 7/08698I Received December 2, 1997 Accepted February 23, 1998 56R Analyst, June 1998, Vol. 123

ISSN:0003-2654    

DOI:10.1039/a708698i

出版商:RSC    

年代:1998

数据来源: RSC

摘要:

Tutorial Review Electrochemical analysis of clinical blood-gases, gases and vapours C. E. W. Hahn Nuffield Department of Anaesthetics, University of Oxford, Radcliffe Infirmary, Woodstock Road, Oxford, UK OX2 6HE This tutorial review charts the development of electrochemical sensors for the analysis of blood-gases, gases and vapours in clinical medicine over the past four decades. The development of each sensor is set in its historical and clinical context, and the first part of the review concentrates on aqueous electrolyte electrochemistry and on those sensors which have made a major impact on the clinical measurement of the partial pressures of oxygen and carbon dioxide in the blood.The electrochemical interference effects of anaesthetic agents on these measurements are also described. Those electrochemical sensors which have failed, in the past, to make a clear impact in this area are not considered, but the few attempts to devise aqueous electrolyte electrochemical sensors for anaesthetic agent measurement are reviewed.The second part of the review describes the chequered history of the development of non-aqueous solvent electrochemical sensors to measure the partial pressures of oxygen and carbon dioxide, in both the presence and absence of each other, in the gas phase. The last part of the review examines various attempts, using non-aqueous solvent electrochemistry, to measure the concentration of inhalational anaesthetic vapours in the gas phase. These sensors have yet to make an impact on clinical practice.Throughout this tutorial review, theoretical models of membrane-covered electrochemical sensors are described where appropriate. This review represents a personal view of the development of electrochemical sensors for clinical measurement, and it is therefore necessarily selective in its approach and emphasis. Keywords: Review; clinical medicine; blood-gases; gases; vapours; electrochemical sensors Clive Hahn is Professor of Anaesthetic Science in the Nuffield Department of Anaesthetics at the University of Oxford, and is a Consultant in Clinical Measurement in the Oxford Radcliffe Hospital NHS Trust.He trained in Manchester, Sheffield and Oxford. He originally took up NHS posts in the Oxford United Hospitals, before becoming a University Lecturer in Anaesthetics and then gaining his present appointments. His major research interest, over many years, has been in the field of cardiopulmonary gas exchange in the sick and healthy lung.In common with others in the Biological and Medical Sciences, his interest in molecular sensors grew out of the frustration of not being able to measure gases of clinical interest with any degree of certainty or precision, especially when anaesthetic gases and vapours were present. He is the author of numerous articles on cardio-respiratory gas exchange, clinical measurement and blood-gas analysis. He is also the British Medical Journal Books Series Editor for their ‘Principles and Practice’ series of books on Clinical Measurement.Introduction The massive contribution made by chemists to the development and understanding of blood-gas and acid-base balance over the past centuries is chronicalled in the book The History of Blood- Gases, Acids and Bases by Poul Astrup and John Severinghaus1 and in their historical essays on blood-gas analysis.2–6 The list is long and very distinguished but, right up to the 1950s, much of this basic research and development work remained confined to chemistry, biochemistry and physiology laboratories.Furthermore, despite the fact that dissolved gases had been extracted from blood by vacuum techniques for more than 300 years, with their identity known for 200 years and their contents having been measured for more than 150 years, these blood-gas content measurements had contributed relatively little to patient care over this period of time.5 However, a succession of external ‘forcing’ factors have contributed, over the past 60 years, to the increasing utilisation of blood-gas measurements in the treatment and care of patients and to the concomitant development of modern blood-gas electrochemical analysis techniques.First, physiological research into high altitude acclimatisation, deep sea diving and aviation medicine began to burgeon during World War II, and the limitations of the classical volumetric analysis of gases dissolved in blood (i.e., the blood-gases) became so apparent that respiratory physiologists and clinicians began to seek other ways of making these measurements.(The chemical analyses of anaesthetic gases and vapours were to follow some decades later.) Following this initial war interest, two unconnected events, both in the 1950s, led to the rapid development of modern electrochemical sensors for measuring blood-gases and, amazingly, these early electrochemical devices were so good that they have remained relatively unchanged over the succeeding four decades.The first event, the massive poliomyelitis (polio) epidemic that struck both Copenhagen and the USA in 1952, focused the minds of chemists, physiologists and clinicians very sharply. Polio was a very important disease which consumed the resources of many hospital departments, and this disease led directly to the development of intensive care units, modern Analyst, June 1998, Vol. 123 (57R–86R) 57Rartificial ventilation and the potentiometric electrochemical analysis of the partial pressure of carbon dioxide (PCO2) dissolved in blood. The inventor of this electrochemical technique, Richard Stow, was a scientist working in the Physical Medicine Department at Ohio State University, where there were desperate attempts to measure the blood-gases of polio patients who were receiving artificial ventilation. His pioneering work was developed further by John Severinghaus, an anaesthetist in San Francisco, who successfully developed the first practical electrochemical PCO2 sensor (or ‘electrode’) for clinical use.4 The second event was the development, in the 1950s, of blood bubble oxygenators for human use.It was quickly realised, at that time, that there was a concurrent necessity to measure the partial pressure of oxygen (PO2) dissolved in blood, in order to check that the oxygenators themselves were working efficiently and properly.5 A biochemist and physiologist, Leland Clark, not only developed some of the first of these new oxygenators, but also invented the first practical electrochemical sensor for measuring the PO2 of whole blood.5,6 So, the first two practical electrochemical medical devices were developed out of acute clinical necessity by clinical scientists and clinicians working in hospital laboratories, but these workers had built upon an experimental framework which had already been firmly established by the electrochemist in the chemistry laboratory.1 From this point on, major contributions to the development of practical electrochemical blood-gas, gas and anaesthetic vapour sensors were made by chemists, physiologists, bioengineers and clinicians alike.Unfortunately, it is, and has been, generally true that clinicians and electrochemists rarely seem to read each other’s research papers, or appreciate the real value of each other’s contribution to the field, and so there is still today a great communication gap between the clinician and the chemist.The author of this tutorial review hopes to bridge this gap somewhat and (i) to introduce the chemist to some of the very real problems facing the use and development of blood-gas sensors in clinical practice and (ii) to correct some misconceptions concerning the measurement of gas and vapour concentrations in the blood. Physiological background Before considering the electrochemical techniques themselves, the whole subject matter needs to be placed within its relevant physiological and clinical context, since this is the raison d’�etre for the development of the electrochemical gas and blood-gas measurement techniques themselves.Cardiopulmonary homeostasis The primary respiratory functions of blood are to transport oxygen from the lungs to body tissues, carbon dioxide from the tisss to the lungs and hydrogen ions from the tissues to the kidneys.The function of the lung is to facilitate the exchange of these gases between the blood and the outside atmosphere. The determination of blood acid–base status (i.e., clinical chemistry) is extremely important in the critically ill patient, but that subject, including the measurement of hydrogen ion concentration, is outside the brief of this tutorial review. This review is concerned solely with the determination of the blood-gases and the inspired or expired gases and vapours.Here, the blood-gases referred to are the partial pressures of oxygen or carbon dioxide in blood (PO2 and PCO2 respectively) and this tutorial review will exclude the measurement of the partial pressures of anaesthetic gases dissolved in blood (for reasons given below). The respiratory blood gases themselves reflect cardiopulmonary homeostasis. This is the ability of the cardiopulmonary system to maintain a constant relationship between respiration in the cells of the body, the supply of oxygen from the lungs to the blood and the elimination of carbon dioxide by the lungs.During inspiration, fresh gas in drawn into the lungs and gas exchange takes place between the gas in the lung and the blood flow entering and leaving the lung. The blood which leaves the lung is freshly oxygenated and is pumped, by the left side of the heart, through a system of arteries to the main body organs and the body tissues. In these body tissues, internal respiration takes place and gas exchange also occurs.The respiratory product of this internal respiratory system, carbon dioxide, is transported back to the right side of the heart, and is pumped back through the lungs (as mixed-venous blood) by the right ventricle. When this venous blood enters the lung, carbon dioxide is offloaded and is expelled in expired air. For a normal human being, about 350 l of carbon dioxide are expired every day! Thus, the body constantly consumes oxygen and produces carbon dioxide, and there is a ‘normal’ balance between the absolute contents and the partial pressures of these two gases in the blood.Cardiopulmonary function acts as a complex feedback system to keep the blood-gas partial pressures of O2 and CO2 at their ‘normal’ physiological values. This balance can be upset by many physiological or clinical factors or problems, but cardiopulmonary function has a remarkable ability to adapt to compensate for these changes, especially to keep the blood-gases at their ‘normal’ values.Carbon dioxide homeostasis Broadly, the PCO2 of arterial blood (i.e., oxygenated blood which is leaving the lung) reflects the adequacy of the ventilation of gas within the lung. Arterial blood normally has a PCO2 of 5.3 kPa and mixed-venous blood has a PCO2 of 6.1 kPa for a man breathing room air. These PCO2 values are kept constant by the lung ventilation rate, which is about 7 dm3 min21. Carbon dioxide homeostasis is concerned with balancing the metabolic rate of production of CO2 in the body against the effectiveness of lung ventilation.If metabolic CO2 production increases, lung ventilation must be increased to excrete the CO2 produced by the body, and so keep arterial PCO2 constant. There is an inverse law relationship between lung ventilation rate and arterial PCO2, at a constant metabolic CO2 production rate. In fact, very little CO2 is carried in blood as a dissolved gas, since by far the majority is carried as hydrogencarbonate or by combination with haemoglobin in the red blood cells.This chemically stored CO2 is released in the lung, in gaseous form, through the action of carbonic anhydrase, and this CO2 is expelled in expired air. The actual total CO2 content (i.e., chemically bound and dissolved) in blood is related to the PCO2 through well defined but complex biochemical relationships, and this is a key part of blood acid-base chemistry.As stated previously, this is outside the remit of this teaching review. It is sufficient to note here that the magnitude of the arterial blood PCO2 is the key clinical measure of the effectiveness of the lungs to expel CO2. Oxygen homeostasis As with CO2, there is also a very well defined physiological relationship between PO2 and oxygen content in the blood. This well known relationship is illustrated in the shape of the oxyhaemoglobin dissociation curve (the blood oxygen content– PO2 relationship) shown in Fig. 1 for both foetal and adult haemoglobin. The sigmoid shape is due to the way in which oxygen combines reversibly with haemoglobin in the red blood cells, and the chemistry of this relationship has exercised the minds of chemists and physiologists for decades.1 The ways in which this oxygen content–PO2 relationship can be compromised are legion, and the interested reader can be referred to a readable applied respiratory physiology textbook for this 58R Analyst, June 1998, Vol. 123purpose.7 For the purpose of this review, it is sufficient to state that the measure of the PO2 in arterial blood is one important indication of how efficient the lungs are in oxygenating the blood, but the measurement of arterial blood PO2 is not, in itself, an indication of adequate tissue oxygenation. The composition of the inspired gases, the affinity of haemoglobin for oxygen, the amount of haemoglobin present in the blood and the heart cardiac output all play major roles in the adequate oxygenation of the body tissues.1,7 Respiratory support Respiratory support, or artificial ventilation, is applied to patients when their lungs begin to fail, because of clinical reasons such as neuromuscular weakness, a decrease in the respiratory drive signals emanating from the brain, an increase in the impedance to respiration or acute lung disease in the severely sick.7,8 All these conditions lead to a failure of the lung to clear the body's CO2 production, together with a decrease in the arterial blood PO2 when the patient is breathing air.A mechanical device, commonly called a ventilator, is therefore connected to the patient’s airway to inflate the lungs and to expel the CO2. Supplementary oxygen is added to the input of the ventilator, so that the patient is oxygenated to the desired level. The efficiency of this artificial ventilation respiratory support is ascertained by obtaining arterial O2 and CO2 bloodgas samples, measuring their partial pressures and then comparing these with the values expected for efficient gas exchange in the lung.As stated previously, the polio epidemic in the 1950s led directly to the development of mechanical ventilators and modern intensive care units. There is now a galaxy of mechanical and electromechanical high-technology ventilators in current use, and the principles and practice of these devices have been very well described by Sir Keith Sykes in his book Principles and Practice of Respiratory Support.8 Intensive care units for sick babies, especially prematurely born infants, have also developed rapidly over the past three decades.These units, too, have placed growing demands on the technologist (including the chemist) to make urgent blood-gas and acid–base analyses with minute blood samples, typically less than 50 ml. The frightening speed at which respiratory disease in both adult and neonatal patients can degenerate has also led to a perceived need to measure arterial PO2 and PCO2 in vivo with intravascular electrochemical transducers, because it is felt that the patient might deteriorate within the analysis time taken by serial discrete blood samples.Whether this need is true or not is a matter of clinical judgement,9,10 but this need has placed an even greater burden on the technologist because it is no easy matter to make on-line electrochemical measurements in such a hostile environment as flowing blood in a very sick patient.Hence the demand for blood-gas analysis, whether using discrete samples or on-line, is of paramount importance for both adult and the neonatal intensive care, and the needs of these two different clinical units can tax the ingenuity of even the cleverest developer of blood-gas sensors. Some perceived clinical demands may just be wishful thinking, as alluded to above, and may be totally impossible to meet with current electrochemical techniques.Before embarking upon a fruitless development exercise, the chemist needs to think long and hard before he or she sets off on a journey to devise the perfect sensor–because it may not, in reality, be needed and might also never be used in clinical practice9,10 (i.e., outside the experimental laboratory). Inhalational anaesthesia Patients undergoing major surgery are still mostly anaesthetised by inhalational anaesthetic agents, delivered by an anaesthetic machine.Inhalational anaesthesia has had a long history, lasting over 150 years now, and the ideal modern anaesthetic agent should induce anaesthesia rapidly, maintain it with the minimum of unwanted side effects for the duration of the operation and should present both the patient and the anaesthetist with the minimum of complications at the end of the operation. It is, of course, impossible to achieve all these ideals since the agent is a drug and, of necessity, this means that it has the potential to harm the patient if administered in too high a concentration or if the equipment delivering the drug to the patient fails to work properly.Because there are so many interconnections between an anaesthetic machine, the ventilator and the patient, and because the agent is delivered by a vaporiser which contains mechanical and/or electronic components, the possibility of mishap is always present. The fact that anaesthesia is a very safe process is a testimony to the reliability of modern instrumentation and to the skill of the clinical anaesthetist, but anaesthetic agent measurement devices are needed to monitor not just the patient, but also the anaesthetic machine itself, if the risk of mishap is to be minimised.It is important, at this point, to correct a misconception which has grown concerning a perceived necessity to measure the partial pressure of inhalational anaesthetic agents in the blood.11 Although the direct molecular mechanism whereby the anaesthetic agent induces a deep state of hypnosis is still not yet fully understood, it has been clear for decades now that it is the partial pressure of a given anaesthetic agent which determines the depth of hypnosis in a given patient.This anaesthetic partial pressure in the blood determines the quantity of anaesthetic which passes across the blood–brain barrier, and the clinical effect of the anaesthetic is titrated against this partial pressure by the anaesthetist using clinical judgement.The important point to grasp here is that, because the agent is breathed in and out of the lungs by spontaneous or artificial respiration, the partial pressure of an inhalational agent in the blood cannot exceed that of the inspired partial pressure (i.e., that partial pressure delivered by the anaesthetic machine). The patient Fig. 1 Oxyhaemoglobin dissociation curves for normal adult haemoglobin compared with foetal haemoglobin.After birth, foetal haemoglobin is progressively replaced with adult haemoglobin and the foetal blood dissociation curve gradually moves across to the right to form the adult curve. Point A represents serious hypoxia which requires urgent treatment in an adult patient with a normal haemoglobin concentration and normal circulation. Point B shows the arterial PO2 which corresponds to the threshold of loss of consciousness from hypoxia.(Taken from ref. 7.) Analyst, June 1998, Vol. 123 59Rtherefore eventually comes into partial pressure equilibration with the setting delivered by the anaesthetic machine via an exponential time process, and neither the patient's expired anaesthetic gas nor the anaesthetic blood partial pressures can rise above the partial pressure set by the anaesthetic machine, because of the law of mass balance. The actual concentration of the agent in the patient's lungs and body organs is governed, of course, by the tissue–gas partition coefficient of that particular agent in the body tissue—and either much, or little, of the anaesthetic agent is stored in the body tissues, according to the individual solubility of the anaesthetic agent in the body tissues.An agent with a low tissue–gas solubility coefficient provides a fast induction for, and recovery from, an inhalational anaesthetic, and this is another ideal aim for the chemist and pharmacologist designing new anaesthetic drugs.The point at issue here is that, unlike oxygen and carbon dioxide (which are chemically bound to haemoglobin and whose partial pressures in expired gas, and in arterial and venous blood, are governed by complex physiological relationships), the inhalational agent will (theoretically) eventually come into partial pressure equilibrium throughout the whole body (i.e., in both inspired and expired gas, and in arterial and venous blood). This does not mean that the uptake and excretion of inhalational agents are not governed by complex physiological/ pharmacological relationships, but it is assumed for pharmacokinetic modelling purposes that they are ‘inert’.Therefore, measurement of the partial pressure of the agent in inspired and expired gas will provide the anaesthetist with the information which is required concerning patient safety. Since these measurements confirm the vaporiser setting, it is unnecessary to measure the blood-gas partial pressure of an inhalational anaesthetic.Such measurements are only required for research purposes, in order to develop new, or confirm old, models of anaesthetic uptake and elimination by human beings and animals. Even in the case of low-flow, or semi-closed, anaesthesia, where it is possible to accumulate an anaesthetic agent through the inappropriate use of certain vaporisers, the present day recommended practice is to measure the inspired and expired agent gas concentrations at the patient's mouth to ensure that a dangerous (or too low, in some circumstances) level of anaesthetic is not delivered.Although it might be argued by some practitioners that monitoring the partial pressure of the volatile agent in the patient's blood on a frequent basis might aid the anaesthetist, the clinical demand is simply not there at the moment. The above arguments do not apply when a non-inhalational anaesthetic drug is delivered intravenously, since a drug delivered in this fashion can build up inexorably within the body, if the body metabolism does not break the drug down into safe by-products in sufficient time, i.e., the patient can be ‘overdosed’.Great care must therefore be taken concerning the infusion of anaesthetic agents—but the topic of total intravenous anaesthesia is outside the brief of this tutorial review. Measurement summary In summary, therefore, it is clinically important to measure the inspired and expired partial pressures of oxygen and carbon dioxide in the critically ill adult and infant patient.It is equally important to measure the PO2 and PCO2 of arterial blood, but here a choice has to be made whether blood samples are taken and analysed by a static laboratory bench analyser, or whether it is possible (or economically viable) to make these measurements in vivo. Sometimes, it will be necessary to measure the PO2 and PCO2 of mixed-venous blood (that is, the blood which is returning directly to the lungs).Sometimes it may be possible to deduce the blood partial pressures by making in situ measurements on the skin of a patient (so-called transcutaneous measurements, described in a later section). Lastly, it will be important to measure the inspired and expired concentrations of oxygen, carbon dioxide and any present inhalational anaesthetic agent when an anaesthetic machine and/or a ventilator is connected to the patient and is thus taking on a life support function.It is important to note that some of these measurements can be made relatively slowly, e.g., the inspired oxygen and inhalational anaesthetic agent concentrations delivered by a machine, and static blood-gas analysis in a bench analyser. In other instances, very rapid response times of the order of 100 ms are required, such as for breath-by-breath inspired and expired gas analysis. Similarly, in vivo blood-gas sensors should respond reasonably rapidly, preferably with a response time of the order of seconds, if the measurements are to have any significant physiological meaning.Before turning to the electrochemical analysis of the respiratory and anaesthetic gases, it must be pointed out that electrochemical techniques have serious rivals in the particular case of respiratory gas-phase measurements. Breath-by-breath O2 gas analysis is now routinely performed by fast paramagnetic and magnetoacoustic O2 analysers, and these analysers also incorporate infrared and photoacoustic gas analysis for CO2 and the inhalational anaesthetic agents.12 As matters now stand, these analysers will not be superseded by presentday electrochemical techniques.On the other hand, ‘static’ or slow response O2, CO2 and anaesthetic agent gas analysis could be performed electrochemically, and this could offer a cheap and viable alternative for monitoring the gas output concentration of the anaesthetic machine.One final comment in this measurement summary section is on the question ‘Why devise new sensors to make “old” measurements?’. Every reviewer will have his/her own opinion on this matter. By the time the reader has finished reading this review, it will be clear that all electrochemical sensors have their own drawbacks (some more major than others) and it is therefore perfectly legitimate for the chemist to explore other electrochemical avenues, or techniques, to make the same measurements.This is how research and development progresses in any science. If the new technique works better, is more reliable, is less invasive and is much better or cheaper, then the old technique will be abandoned and relegated to the electrochemical museum. If the new technique is no better than the old, then at least this avenue has been explored and tested. For this reason alone, clinicians and electrochemists alike have continued over the past four decades their relentless pursuit of developing the ideal chemical sensor for the respiratory and anaesthetic gases.So, the clinical demands have already been outlined and have remained essentially unchanged over the past four decades. Many more chemical sensors have been developed over the years for PO2 and PCO2 measurements than can be mentioned in this tutorial review. They have all had their moments of glory and some examples include quinhydrone, conductometric, potentiometric, and CHEMFET sensors.Many of these have been described in past reviews13–15 and in the historical series by Severinghaus and Astrup.1–6 Also, the past two decades has witnessed a burgeoning interest in the measurement of O2 and CO2 using optical techniques, especially in the field of in vivo ‘intravascular’ blood-gas analysis, and a fibre-optic in vivo blood-gas monitor is now available for hospital use.12 This particular topic is, again, outside the brief of this tutorial review and so will not be considered here.Space constraints permit only a selection of electrochemical sensors to be described, and this reviewer has made his choice selectively, concentrating on those devices which have made the greatest impact or have shown the most promise. Electrochemical techniques still remain supreme in the laboratory 60R Analyst, June 1998, Vol. 123bench blood-gas analyser, and they are likely to do so for the foreseeable future. Analytical techniques Gas phase Because steady state gas-phase measurements require only a relatively slow response time, electrochemical gas sensors are usually placed at the gas output of the gas delivery system (i.e., the anaesthetic machine or the ventilator). The actual position of the gas sampling site can be important as far as patient safety is concerned.It is possible to sample at a given point in the patient’s gas delivery system, in the belief that this measurement point is the correct one for monitoring the safety of the patient, and yet place the patient in danger—because an undetected gas delivery disconnection can occur between the patient and the point of gas measurement.These issues and potential mistakes are subtle in clinical practice, and have been discussed and illustrated by Sykes.16 The ideal measurement point is at the patient's mouth, but this is impractical for electrochemical sensors, because of their size and the constant risk of infection—clinical electrochemical gas-phase sensors are not amenable to sterilisation.Furthermore, gas measurement at the mouth must be made with a very fast time response device, since the measurements must be able to follow both the inspiratory and the expiratory gas concentration patterns. A slow time response sensor will simply measure a running average of the inspiratory and expiratory gases, rendering the measurement useless for respiratory analysis. Thus, in the absence of very fast time response electrochemical sensors, measurements at the mouth are out of the question (even with side-stream remote analysis), and so electrochemical gas measurements must be confined to the parts of the gas delivery system which are distal to the patient's airway.Blood phase Here, three techniques are commonly used. The major technique, and that most used, is in vitro analysis of the blood, using electrochemical sensors installed in a blood-gas analyser situated either close to, or some distance away from, the patient.A blood sample is taken from the patient’s artery in a gas-tight syringe, and this sample is then introduced to three electrochemical sensors which determine PO2, PCO2 and hydrogen ion concentration separately. Less commonly used are in vivo techniques, which can be sub-divided into two approaches. The first approach (which is rarely used in the adult patient) is to insert an electrochemical sensor into the patient’s artery so that blood-gas measurements can be made in vivo.The alternative approach is a non-invasive (transcutaneous) method of measuring blood-gas partial pressures, where the electrochemical sensor is placed on the patient’s skin, and the measurement relies on O2 and CO2 diffusing through the skin to the electrochemical sensor. Each different measurement approach has certain advantages and disadvantages, but all methods employ the same electrochemical principles to determine the gases in question.Almost without exception, separate electrochemical sensors are used to measure PO2 and PCO2, and these sensors invariably use a twoelectrode cell, with a working electrode to measure the variable and a reference electrode to standardise the measurement. A three-electrode cell is rarely employed, and this tutorial review will concentrate on the working electrode, at which the O2, CO2 or anaesthetic agent measurement is made. The reference electrodes (which are standard Ag/AgCl or calomel reference electrodes) will not be considered any further.Aqueous electrolyte solution electrochemistry:carbon dioxide determination Since the history and development of the electrochemical determination of CO2 is less varied than that of O2 determination, this topic will be considered first. Apart from very recent developments (to be described later), CO2 determination in the blood has always been performed by potentiometric means using aqueous electrolytes.The PCO2 sensor, first described by Stow et al.17 in 1957 and then by Gertz and Loeschcke18 in 1958, was developed into its present form by Severinghaus and Bradley19 in 1958. This CO2 sensor is commonly referred to as the Stow–Severinghaus electrode. The Stow–Severinghaus sensor This sensor is, in effect, a glass pH electrode housed behind a thin membrane which is permeable to CO2, with a thin hydrogencarbonate layer placed between the pH electrode and the membrane.The working principle of this sensor can be illustrated by reference to Figs. 2 and 3. Fig. 2 shows a cross- Fig. 2 A schematic outline of the Stow–Severinghaus PCO2 sensor, suitable for in vitro blood-gas measurement. Fig. 3 Schematic diagram of the tip of the Stow-Severinghaus PCO2 sensor showing the membrane, buffer electrolyte layer and the pH-sensitive glass of the pH electrode. The equilibria equations for CO2 transport across the membrane are illustrated.Analyst, June 1998, Vol. 123 61Rsection of a practical PCO2 sensor. The end of the pH glass electrode is covered by a matrix which holds a thin layer of solution containing NaHCO3 with some NaCl, and sometimes AgCl, also added. An Ag/AgCl reference electrode, in physical contact with the hydrogencarbonate solution, completes the electrical circuit to the pH working electrode. The potential difference between the working electrode and the reference electrode is measured by a standard high-impedance voltage amplifier and display system.Working principle The pH sensor is separated from the blood, which is to be analysed, by a membrane (typically 20 mm thick Teflon) which is highly permeable to CO2. When in practical use, the PCO2 of the blood comes into equilibrium with the PCO2 of the hydrogencarbonate solution in the matrix immediately adjacent to the H+ ion-sensitive glass. This process is shown diagrammatically in Fig. 3. The equilibria now existing in this thin hydrogencarbonate layer are CO2 + H2O " H2CO3 " H+ + HCO32 HCO32 " H+ + CO3 22 NaHCO3 " Na+ + HCO32 By writing the dissociation constants K, K1 and K2 in terms of the chemical activity of each species present, and noting that the ionic products for the dissociation of water are given by KW = aH+aOH2 it can be shown that the CO2 partial pressure, PCO2, in this thin film is given by P a a a K K K a CO H Na H w H 2 2 1 2 1 2 = + - + ( ) + + + + a / (1) where a is the solubility coefficient of CO2 in the thin film of hydrogencarbonate, and it is assumed that Henry's solubility law holds in this thin film.This principle was first determined by Stow in 1952, although his work was not published until 1957.17 However, Stow’s sensor utilised a thin layer of water between the membrane and the pH working electrode, and the sensor did not work well—it was unstable in practice, presumably because the pH of the distilled-water electrolyte changed very easily with slight contamination.Severinghaus’s key contribution to the development of the sensor was to add sodium hydrogencarbonate to the water in order to stabilise the pH and to increase the sensor sensitivity. Accordingly, Severinghaus and Bradley19 demonstrated that the addition of a 5–20 mm hydrogencarbonate concentration greatly increased the sensitivity of the sensor, and under these conditions, eqn. (1) simplified to P a a K CO Na H 2 1 = + + a (2) Severinghaus was able to demonstrate that when the PCO2 in the hydrogencarbonate layer changed to a new value, then [by taking logarithms of both sides of eqn.(2) and noting that both K1 and aNa+ are constants, and therefore cancel out when two aH+ values are inserted], the relationship between the changes in the PCO2 and pH in the bicarbonate layer was given by DlogPCO2 = 2DpH (3) where the D terms represent the change in the logarithm of the PCO2 and the change in pH.It must be noted that -Dlog aH+ = DpH, by definition, in the hydrogencarbonate layer. These authors also defined the ratio Dlog PCO2/DpH as the sensitivity, S, of the sensor. Thus, if S is 21.00, then each 10-fold change in PCO2 will theoretically induce a one unit change in the pH of the hydrogencarbonate solution. At a steady temperature of 37 °C, this would be registered as a voltage change of 61.5 mV, as predicted by the Nernst equation. Severinghaus and Bradley found that the sensor sensitivity was slightly less than this predicted change, and that S could typically be 20.97.It is clear from eqn. (3) that since the relationship between pH and potential difference is linear, and the relationship between PCO2 and pH is logarithmic, then the carbon dioxide sensor produces a logarithmic relationship between PCO2 and output voltage. All blood-gas analysers therefore employ an algorithm to linearise this relationship, so that PCO2 output is displayed linearly.Stability It is important to note that the bulk of the electrolyte, shown in Fig. 2, plays no part in the equilibration process and merely serves as a reservoir and as electrical conductor. All active processes take place only in the thin hydrogencarbonate layer trapped adjacent to the glass electrode. It is therefore important to keep a stable calibrating gas, or liquid, solution in the sample chamber between readings, so that the sensor always works from a stable pH reference point in the hydrogencarbonate layer and returns to this same point after sampling.Sensor instability or drift will result if this regime is not adhered to, and modern blood-gas analysers pass through a sample analysis/wash/ reference calibration point regime when used in clinical practice. If a sample is retained too long in the analysing chamber adjacent to the sensor membrane, the bulk of the sensor electrolyte solution will begin to come into equilibrium with the PCO2 of the sample, and this will destroy the stability of the sensor. Liquid-gas difference Because all the electrode reactions described above are reversible, and since pH measurement itself is potentiometric, the overall electrochemical CO2 process does not consume the analyte.Since the PCO2 sensor does not consume CO2 from the sample, it follows that, in principle, any liquid or gas sample which has the same CO2 partial pressure will register the same output when analysed with the same sensor.There is therefore no so-called ‘liquid–gas measurement difference’ for the potentiometric Stow–Severinghaus PCO2 sensor, and this is a major advantage over new rival amperometric techniques. Calibration The absence of a liquid–gas difference for the potentiometric CO2 sensor means that the sensor can be calibrated with either a gas or liquid before it is used to measure the PCO2 of a blood sample. Since the PCO2 sensor output is logarithmically related to PCO2, a zero PCO2 calibration point cannot be used.It is usually convenient to calibrate the sensor with two gases, or two liquids equilibrated with these gases, which have CO2 contents which span the expected range of the blood PCO2 values. Since the clinical range for patient blood PCO2 values is expected to be between 2.7 and 8.0 kPa (or above in extreme cases), the two PCO2 calibration values will normally lie between 3 and 10 kPa. Uses and limitations The Stow–Severinghaus CO2 sensor, as described above, has a typical outside body diameter of 10 mm and an overall sensor length of 35 mm.It is very difficult to reduce these overall dimensions because of the very nature of the construction of pH glass electrodes. However, Parker et al.,20 in 1978, made a 62R Analyst, June 1998, Vol. 123heroic attempt to develop an intravascular dual O2/CO2 sensor, and they managed to reduce the overall diameter of the sensor to 1.6 mm. However, this sensor was obviously very difficult to construct and must have been extremely fragile, and it does not appear to have been developed any further.Thus, the standard PCO2 sensor is used solely for in vitro or transcutaneous (see later) CO2 measurements in clinical practice, and it cannot be used for breath-by-breath CO2 gas analysis because of its slow response time. Nonetheless, this sensor has stood the severe test of time, and has been used (outside blood-gas measurement in medicine) for ambient gas monitoring, tissue studies, industrial fermentation control and even for satellite atmosphere monitoring.4 As highlighted by Severinghaus and Astrup, Stow’s development of the CO2 sensor was a result of logical thinking and did not depend on a long series of prior discoveries, apart from the previous development of the glass pH electrode.The sensor was not patented, was not disputed and the inventor did not profit from his discovery4—an unusual occurrence in today’s society.Determination of oxygen Danneel,21 in 1897, reported that oxygen in aqueous solution reacted with negatively charged inert metals, and he demonstrated an approximately linear relationship between oxygen partial pressure and current when using two large polarised platinum electrodes. He, and others, attempted to use platinum electrodes for oxygen measurement in biological media, but found that the electrodes were rapidly poisoned, or else were coated with protein, causing decrease in the oxygen reduction current. Because the measurements were not dependable, Danneel's discovery fell into disuse.Much later, the discovery of the dropping-mercury electrode, and the subsequent measurement of dissolved oxygen with this electrode, was the prelude to a burgeoning interest in the measurement of dissolved oxygen. According to Severinghaus and Astrup,6 it was work by Blinks and Skow22 (in which they used platinum or gold instead of mercury as the working electrode) which led to the use of solid metal working electrodes for the measurement of dissolved oxygen, and thus the subsequent measurement of an ever-broadening range of other gases.Although Heyrovsky (the discoverer of the dropping-mercury electrode and the founder of polarography) had strongly objected to the use of solid metal working electrodes, because of their instability due to the ‘poisoning’ of the cathode surface,5 Blinks and Skow persevered with their investigations of the use of platinum electrodes, and they demonstrated that they could produce results which were identical with those obtained with the dropping-mercury electrode. In fact they found that good oxygen reduction plateaux could be obtained in stirred solutions, with good current linearity over a range of oxygen concentration from 0 to 99.5% v/v.They further reported that oxygen was reduced to hydrogen peroxide during their investigations into oxygen evolution and consumption in plant cells, and they demonstrated that plant catalyse immediately broke down the hydrogen peroxide.Thus, Blinks and Skow appear to be the first workers to initiate the use of platinum working electrodes, at fixed polarising potentials, for the amperometric measurement of oxygen partial pressure in biological solutions, and they were also the first to demonstrate the linearity of the oxygen reduction current with PO2. Unfortunately, these workers appear to have been frequently overlooked when the history of the amperometric oxygen sensor has been reported in the past.Thereafter, a whole succession of physiologists and chemists attempted to use solid wire electrodes to measure PO2, with little consistent success in biological media, and the solution to the problem of the measurement of PO2 in liquids and biological solutions was finally proposed by Leland Clark.23 Clark’s great contribution to this field was the imposition of a gas-permeable membrane between the liquid sample and the working electrode, with the reference electrode also housed behind the same membrane. This innovative, but simple, step solved the problem of contamination or poisoning of the working electrode in one stroke.This single change in design was a historical turning point for respiratory physiology, and led rapidly to the development of the modern blood-gas analysis machine. There followed an unparalleled explosion in the use of the oxygen electrode in clinical medicine, and it was rapidly used to measure gaseous oxygen, blood PO2 and blood oxygen content and in, the study of the oxyhaemoglobin dissociation curve, and its use was also extended to the food, alcohol, aircraft and the space industries, to soil chemistry and to waste water and sewerage management. 5,6,13,24–27 The development of this membrane-covered electrode also led to a variety of electrochemical control techniques being applied to the working electrode, in order to obviate, or overcome, some of the inevitable practical difficulties which would be encountered when the sensor was applied to practical clinical problems. It has also led to a research interest in the development of mathematical and computer sensor simulation models, in order to describe the reaction–diffusion processes taking place in the sensor and at the working electrode surface. The Clark sensor The fundamental principles of the Clark amperometric sensor are illustrated in Fig. 4. As with the Stow–Severinghaus CO2 sensor, the Clark sensor comprises a container which houses the electrolyte, the working electrode (cathode) and the reference electrode. The external electrical circuit consists of a fixed voltage source and a current-to-voltage converter to measure the oxygen reduction current. A gas-permeable membrane separates the electrolyte from the oxygen sample and the working electrode is polarised at a steady negative voltage with respect to the reference electrode (which is almost invariably an Ag/ AgCl electrode).The electrolyte solution is, typically, an aqueous buffer solution with the addition of Cl2 ions, and a pH is generally chosen to be in the region of 7. When used in a blood-gas analyser, the complete system is thermostatically Fig. 4 Schematic outline of the Clark PO2 sensor. Analyst, June 1998, Vol. 123 63Rcontrolled at 37 °C, so that the blood-gas PO2 measurements are made at normal body temperature.28 Electrode reactions There has always been discussion and disagreement about the exact nature of the electrochemical reactions taking place at the electrode surface in the aqueous electrolyte solutions used in the Clark sensor.14,27,28 Perhaps there will never be agreement, because the electrode reactions (and any combination of them) appear to depend upon poorly reproducible or uncontrollable conditions such as the past history of the cathode surface, the pH of the electrolyte, the metal used for the cathode and even the geometry of the cathodic compartment.27 Furthermore, it could well be true that the combination of the reactions taking place at micro-cathodes (diameters less than 20 mm) might not be the same as those taking place at the conventional macro-cathodes (diameters greater than 200 mm) which are conventionally used for monitoring industrial processes. Various schemes have been proposed in the past,14 with varying degrees of complexity, but it seems clear that the two products for the reduction of O2 at noble metal surfaces, in alkaline media, are either the hydrogen peroxide ion, HO22, or the hydroxyl ion, OH2.A much simplified reaction scheme is given by O2 +H2O + 2e ? HO22 + OH2 (4a) HO22 + H2O + 2e ? 3OH2 (4b) catalytic decomposition HO22 –––––––––––––––––? 1 2O2 + OH2 (4c) ––––? ýýýýýý ý ý ý –––––––––––––––––––––––––––– or direct O2 + 2H2O + 4e ––––? 4OH2 (4d) where the HO22 ion is either reduced to OH2, is catalytically converted back to oxygen, which can further react on the cathode surface, or else escapes from the cathode surface into the bulk of the electrolyte, in the immediate vicinity of the cathode.Examples of complex pathways relating to these reactions have been illustrated by Jacq and Bloch,29 Wroblowa et al.,30Appleby and Savy31 and by Linek et al.27 Although these workers confined their findings to macro-electrodes, there is no reason to suppose that their general conclusions cannot be applied, in part, to the processes taking place at micro-cathode surfaces.The net result of these processes is that the number of electrons, n, involved in the reduction of O2 can vary between the limits 2 � n � 4. Hahn et al.32 examined the way in which n varied with electrolyte pH and cathode metal, with a rotating ring-disc electrode system. The conclusion was that the amount of HO22 reduced as the negative polarising voltage was increased, for all values of pH.The problem with the production of the unwanted HO22 ion is that it can build up in the bulk electrolyte if it escapes from the cathode surface. It can then diffuse back to the cathode to be reduced to OH2 ions, and so constitute an uted current. This takes the form of a hysteresis effect, when a low PO2 sample has been introduced to the sensor immediately following a high PO2 sample. In this instance, there is a long tail in the sensor response time before it registers the ‘true’ low PO2 value, and this introduces a long time constant into the measurement procedure.Both the hysteresis and the over-long time constant effects are reduced as n ? 4, as the polarising voltage is increased to high negative values.32 Eqns. (4a) and (4b) suggest that two clearly separated O2 reduction waves should be observed when O2 is reduced in aqueous electrolyte solutions. However, this is not necessarily the case even for unshielded electrodes.The electrode material, the electrode size and the pH of the electrolyte all play a part in the reduction process scheme, and the presence of a shielding membrane also appears to change the nature of the processes occurring. For instance, when O2 is reduced on unshielded gold macroelectrodes, two O2 reduction waves [corresponding to eqns. (4a) and (4b)] are seen. When shielded with a membrane (as in a Clark sensor), only one wave [corresponding to eqn.(4d)] is seen. When Pt is used as the cathode material, only one reduction wave is seen for both unshielded and membranecovered cases. When 20 mm diameter micro-electrodes are employed, the pH of the electrolyte plays a key role in determining the shape and position of the O2 reduction voltammogram. Fig. 5 shows voltammograms for the same 20 mm diameter Pt micro-disc cathode, shielded by the same 25 mm polypropylene membrane, for O2 reduction in pH 6.8 and 11.2 electrolyte solutions at 37 °C.The voltammogram with pH 6.8 electrolyte shows no clearly defined plateau at the voltage normally used to poise Clark blood-gas PO2 sensors, namely 20.6 V versus Ag/AgCl, whereas with the pH 11.2 electrolyte a clear plateau is seen at polarising voltages more negative than 20.8 V. Furthermore, a large degree of hysteresis is evident between the upward and decreasing sweeps for pH 6.8, and this hysteresis is removed on the plateau part of the wave for pH 11.2.Despite this evidence, manufacturers to this day use electrolyte solutions with pH Å 7 in their Pt microdisc blood-gas PO2 sensors and poise them at about 20.6 to 20.7 V, i.e., sensors appear to work on a voltammogram which does not display a clear diffusion plateau at that voltage. Cathode size and the stirring problem The polyethylene membrane covering Clark’s electrode not only completely avoided the cathode poisoning problem, and made measurements of gases (and not just liquids) possible (because both cathode and reference electrode were located within a single electrochemical cell) but also, very importantly, the membrane was relatively impermeable to oxygen.Limiting the amount of oxygen consumed by the cathode enabled the sensor to measure the oxygen partial pressure in solution. Fig. 5 Voltammograms for a 20 mm Pt micro-electrode, when covered with a 25 mm polypropylene membrane, at 37 °C (normal body temperature), when using a pH 6.8 electrolyte (top) and a pH 11.2 electrolyte (bottom). 64R Analyst, June 1998, Vol. 123However, the first sensor had a relatively large cathode (2 mm diameter) and the sensor reading in stagnant water was 25% too low when the sensor was calibrated with a gas with the same PO2.5 This ‘liquid–gas’ difference, to be called later the ‘stirring effect’, was enormous. Severinghaus and Bradley19 immediately saw one solution to this problem, and introduced a magnetic stirring bar in front of the electrode, which greatly reduced the liquid–gas difference.Thereafter, many magnetic stirring systems were described, but alternative methods of solving the stirring problem soon appeared. Some methods used thick polyethylene membranes, some used Mylar (6 mm) as the material, but the disadvantage of tackling the problem by making the membrane more impermeable to oxygen was that the time response of the electrode became unacceptably large ( > 2 min).The final solution of the stirring problem came from attempts to make the electrode small enough for intravascular use or for use with very small in vitro samples. The emergence of highgain current amplifiers made accurate measurement possible for currents of the order of several pA. Staub,33 in 1961, designed a micro-electrode in order that it might fit into a very small cuvette, and he embedded a 50 mm platinum wire in a glass rod—the first reported such small micro-electrode Clark PO2 sensor.This small electrode drew immediate attention to the possibility of reducing the stirring effect by reducing the cathode diameter and thus the quantity of oxygen consumed by the sensor. It also enabled researchers to use thinner membranes and so reduce the electrode response time to the order of seconds. Very soon, cathodes with diameters of 12–25 mm were produced, and when these were covered with 25 mm polyethylene, the stirring effect was found to be only about 3–5%.This was an acceptable level for clinical and most research uses, without stirring. Other early approaches included mechanically agitating the sample so that it moves to and fro across the membrane face adjacent to the cathode surface, or pulsing the polarising voltage with a fixed duty cycle and thus using pulsed amperometry to control the electrode. Although sufficiently short pulses have the advantage that they can control the expanding diffusion field, emanating from the vicinity of the cathode, in a space of defined diffusion conditions (e.g., within the membrane or with a defined unstirred layer of the solution), the electrode control circuitry is complex.Furthermore, it is only recently (see below) that realistic two or three layer digital simulation models have been developed to describe the complex relationship between the cathode time-dependent current, the cathode diameter and the electrolyte and membrane layer thicknesses and permeabilities.Therefore, over the past 40 years, blood-gas analyser manufacturers have accepted the blood-gas difference as inevitable, and strike an empirical compromise between the cathode diameter, the type of membrane material and the membrane thickness. This has inevitably led to a 3–5% difference between gas and blood samples which have the same PO2, but this difference is automatically accounted for in the software of the analyser. Calibration and quality control In vitro blood PO2 sensors can be calibrated either with two gases, or with two liquids which have themselves been brought into equilibrium with known gas mixtures.One of these gases (or liquids) will have a zero PO2, in order to set a ‘zero point’ for the PO2 sensor. A two-point calibration procedure must be employed, because the quiescent current present in a microcathode PO2 sensor is a sizeable fraction of the oxygen reduction current produced when blood samples are introduced into the analyser. Typically, the current generated by a blood-gas PO2 sensor is of the order of 70–100 pA per kPa PO2.With such a small current sensitivity, it is clear that even a quiescent background current of 10–20 pA has to be ‘backed off’ by the control circuitry. The ‘high’ calibration point will typically be set somewhere between a PO2 of 10 and 21 kPa. The other main gas in the two calibration mixtures will be CO2, with ‘low’ and ‘high’ PCO2 values, for calibrating the Stow–Severinghaus CO2 sensor.A modern blood-gas analyser will go through an automatic two-point calibration procedure every 2–4 h, to account for the inevitable drift of the output signals from these electrochemical sensors and to correct for changes in their sensitivity to PO2 and PCO2; however, they will also go through a ‘single point’ check calibration after each blood sample has been analysed. Ways in which the accuracy of blood PO2 (and PCO2) measurements can be assessed has always been controversial, and this subject matter really belongs in the domain of clinical biochemistry.The ‘gold standard’ should be samples of fresh tonometered human blood, but this is an artform in itself and many laboratories have not had, in the past, the facility for conducting this work. Instead, ampoules of commercial quality control materials became widely available, based on aqueous solutions with O2 solubilities equivalent to water, haemoglobincontaining products, and perfluorocarbon-containing emulsions with an O2 solubility several times higher than that of water.According to Hansen and Fiel,34 it is not unusual to find differences of !20% between model-specific mean PO2 values for a given set of quality control ampoules. In their own work, they found that the perfluorocarbon emulsion materials performed better than the other materials. However, good modern practice now involves the use of tonometered non-human or human blood, with a high quality tonometer sited close to the blood-gas analyser unit, so that the performance of the PO2 and PCO2 electrodes can be checked at will.Theoretical models Theoretical models describing the output current–PO2 relationship of the Clark PO2 sensor have changed dramatically over the past four decades. Perhaps the major problem, at least in the early years, arose from the application of one-dimensional diffusion theory to membrane-covered micro-cathode electrodes.Once it had been realised that this was a great mistake, hemispherical and cylindrical diffusion models began to emerge. All the analytical solutions described below in this section are based on various solutions of the Fick equation: d d p t P p = Ñ2 (5) where P is the oxygen permeability in the medium (P = aD, where a is the oxygen solubility and D is the oxygen diffusion coefficient in the medium) and p is the oxygen partial pressure in the medium (written with a lower case p here to avoid confusion with the permeability term, P).Steady-state models One-dimensional linear diffusion model. This three-layer model, which only strictly applies to macro-cathodes employed in gas-phase PO2 sensors, is described in Fig. 6. In this model, the constraint is normally made that the cathode surface is polarised sufficiently negatively enough to ensure that all oxygen molecules reaching it are immediately destroyed. For linear diffusion, eqn. 5 reduced to d d d d p t P p z = 2 2 and the current at the (disc) cathode surface is given by evaluating the oxygen flux to the cathode. The steady-state limiting current, iL, is given by Analyst, June 1998, Vol. 123 65Ri nF R p d P d P L s e e m m = + p 2 / / (6) where Pm and Pe are the oxygen permeabilities in the membrane and electrolyte layer respectively, ps is the prevailing PO2 in the sample, n is the number of electrons involved in the reaction, F is Faraday's constant, R is the radius of the disc electrode and de and dm are the thicknesses of the electrolyte and membrane layers, respectively.Since, in practice, dm Å 20 mm, de Å 5 mm, Pm Å 8 3 10211 m2 s21 atm21 and Pe Å 2.7 31029m2 s21 atm21, it is clear that the ratio dm/Pm is much greater than de/Pe and eqn. (6) reduces to i nF R P P d L m s m = p 2 (7) Eqn. (7) describes adequately the steady-state behaviour of macro-disc gas-phase PO2 electrodes, but the current generated by this equation fails, by at least an order of magnitude, to approach those measured experimentally for in vitro blood-gas electrodes which have cathode diameters of the order of 20 mm.In this case, ‘edge diffusion’ of oxygen towards the cathode surface is of paramount importance, and models employing hemispherical or cylindrical polar coordinates have to be employed. One-dimensional hemispherical diffusion model. A simplified one-dimensional diffusion model which has been used to describe the behaviour of microdisc electrodes is the semihemispherical model.Hahn,14,35 in 1974, described such a model which included a first-order reaction rate term to describe the reduction of oxygen at the cathode surface, and so allow the model to generate voltammograms. A schematic diagram of this model is shown in Fig. 7, which also describes the sensor parameters. Eqn. (5) in this instance is d d d d d d p t P p r r p r = + � æ è ç ö ø ÷ 2 2 2 and the current is derived from half the oxygen flux at the cathode surface.In the limiting current case, where all the oxygen molecules reaching the cathode surface are immediately destroyed (i.e., the micro-disc cathode is polarised sufficiently negative enough for the reaction current to be diffusionlimited), then the sensor current equation simplifies to i nFp d P r r d P r R = + 2p s m m m e e e e (8) This model is obviously not a true representation of physical reality, but its analytical simplicity enables the effects of varying parameters such as R, de, dm, Pm and Pe to be investigated with ease.The sensor currents generated by this simple theoretical model, for micro-disc diameters of the order of 20 mm, can agree closely with those measured by experiment, if the electrolyte layer thickness (de) is assumed to be about 5–10 mm. This simple hemispherical model (excluding the reaction rate terms) was replicated by Ultman et al.36 in 1981, where it can be seen that their eqn.(5) is identical with that published by Hahn previously.4,35 They later extended the steady-state hemispherical model to a three-layer model, where a flowing liquid medium constituted the third layer.37 Linek et al.27 relaxed the steady-state constraint, previously applied to the three-layer hemispherical model, and developed a series of much more complex time-dependent equations which described the sensor current response to a step-change in the oxygen concentration in front of the sensor membrane (i.e., in the liquid layer).These equations are fully described in their book, and they also describe Clark sensor current relationships (for macro-cathodes) when there is disturbed four-electron stoichiometry of the oxygen reduction process at the cathode surface.27 Two-dimensional cylindrical diffusion models. The preceding models, based on oxygen transport by one-dimensional diffusion processes, produce analytical solutions, but they do not describe the real electrochemical situations occurring in sensors with micro-cathodes.The contribution of ‘edge’, or radial, diffusion towards the micro-cathode is best described by a twodimensional oxygen diffusion model, which is based on a combination of two one-dimensional diffusion oxygen flows in mutually perpendicular directions. This model involves solving the Fick equation in two separate layers, the first layer comprising the membrane, with diffusion taking place perpendicular to the cathode surface, and the second layer comprising the electrolyte solution, with oxygen diffusing radially towards the micro-cathode surface.This two-layer model is described schematically in Fig. 8, and was proposed by Hahn14,35 in 1974 as a one-dimensional representation of a micro-electrode oxygen sensor. Again, this early model incorporated a firstorder electrochemical reaction, and was used to generate voltammograms describing the overall sensor current–polarising voltage relationship. The success of this model lies in the fact that since Pm is much less that Pe, then the steady-state equations in the two layers can be described by Fig. 6 Coordinate system for a one-dimensional diffusion model for a macro-electrode PO2 sensor, with a large cathode (radius R 9 de; R 9 dm) and electrolyte and membrane thicknesses de and dm, respectively.Fig. 7 Schematic diagram of the hemispherical diffusion model for a PO2 sensor, with cathode, electrolyte layer and membrane radii Rc, re and rm, respectively.The electrolyte and membrane layer thicknesses are de and dm, respectively. 66R Analyst, June 1998, Vol. 123m e z e P p z e z P p r r p r p z r R p z z r p p z r R D p z kp r z m e p p r z e P + > > = > > + � + æ è ç ö ø ÷ = � = = ® ¥ = = £ = ( ) = + = = : : : , : : , : : m e s e s d d d d d d d d z = 0, d d d d all 2 2 2 2 2 2 0 0 1 0 0 0 0 0 m m e e d d d d p z P p z &aelrave; öø = æè öø where k is the reaction rate at the cathode surface.The sensor current is given by i nFk p r r r R = ( ) ò 2 0 0 p , d (9) but analytical solutions to these equations are unmanageable, and so the simultaneous equations must be solved by digital simulation techniques. Once more, this type of model was greatly extended and modified by Linek et al.,27 particularly for use with macro-cathodes which could be used for oxygen gas analysis. Later, Jenson et al.,38 in 1978, made an important contribution to this model when they solved, and balanced, the contributions from the membrane and electrolyte layers, to produce a simplified analytical solution for the electrode current given by i nF R P d p x K x K x s = + � é ë ê ù û ú p 2 1 0 1 2 m m R R R ( ) ( ) (10) The first part of this relationship: nF R P d p m m s p 2 (11) represents the current corresponding to one-dimensional diffusion through the membrane, i.e., eqn.(7). The second part of the equation: 2 2 1 0 pnFR P d p K x x K x s m m R R R ( ) ( ) (12) corresponds to the current due to the radial flux of oxygen through the electrolyte, and it is this term which dominates for R < dm.Linek et al.27 have also further modified this relationship to compensate for the possible over-estimation of the importance of the radial contribution to the total oxygen flux by the assumption of an infinitely rapid axial diffusion of oxygen through the electrolyte layer.They circumvented this drawback by introducing a ‘resistance’ to the diffusion in the zdirection, and produced modified equations to describe their compensation technique.27 Time-dependent models In most situations, the Clark PO2 sensor is continuously polarised and therefore any time-dependence in the Fick equation [eqn. (5)] normally refers to step changes in the sample PO2 at the face of the membrane. However, in other instances, a step change is forced on the polarising voltage (i.e., pulse chronoamperometry), from a potential where oxygen is not reduced to a potential where oxygen is fully reduced. In this case it is assumed that the sample PO2 (ps) remains constant at the membrane interface, and the Fick equation [eqn.(5)] is then solved, in two or more layers, to produce the sensor pulse amperometry current–time response. These two different timedependent models will be considered separately. Step-change in PO2. In this situation, a step-change in ps is imposed at the membrane surface, and the Fick equation is (usually) solved for the one-dimensional case (i.e., applying to gas-phase Clark sensors with large cathodes).Hitchman13 analysed this situation, again making the assumption that the effect of the electrolyte layer may be ignored. His solution for the electrode current at time t after a step change in the sample PO2 is given by i t i i i n D t d n n ( ) ( ) exp - - = + - - æ è ç ö ø ÷ ¥ = ¥å0 0 2 2 2 1 1 2 1 p m m (13) where i0 is the electrode current immediately preceding the step change in PO2 and i° is the final current reached, i.e., the steadystate current.Considering the simplest example, where the oxygen step change is from zero to ps, eqn. (13) simplifies to i t nF R p P d n D t d s n n ( ) ( ) exp = + - - æ è ç ö ø ÷ é ë êê ù û úú = ¥å p p 2 2 2 2 1 1 2 1 m m m m (14) and for large t approaches the same steady-state value as given in eqn. (7). Eq. (14) is the simplest example of a Clark sensor time response for a step change in sample PO2, and when solving this equation workers normally assume four-electron (i.e., n = 4) stoichiometry for the reduction of oxygen at the cathode, and therefore take no account of disturbed stoichiometry or multidimensional diffusion.These extra complications have been described, however, in some depth, in the book by Linek et al.,27 where models with hemispherical and cylindrical coordinates are considered. Furthermore, these more complex models have also been adapted by Linek et al.to take account of the influence of the various combinations of the electrochemical reactions (2 � n � 4) taking place within the electrolyte layer. Pulse amperometry. The ‘switch-on’ current transient, which occurs when the polarising voltage is pulsed, is the most complex of all the models. An analytical description of the sensor current transient, following the polarising voltage switch-on, for the onedimensional diffusion model described by Fig. 6, was first described by Mancy et al.39 in 1962. They produced analytical current-time solutions for a macro-cathode membrane-covered Fig. 8 Co-ordinate system of the cylindrical diffusion model for a PO2 sensor, with micro-disc cathode (radius R) and electrolyte and membrane layer thicknesses e and m, respectively. Diffusion in the membrane is assumed to be one-dimensional. Analyst, June 1998, Vol. 123 67RClark sensor, with a steady sample PO2 at the membrane surface. Their analytical solutions can be broadly divided into three time intervals, namely at very short times when oxygen diffusion was limited to the electrolyte layer alone, at intermediate times when the diffusion layer had entered the membrane and there was joint transport control and, finally, at long time intervals when the diffusion layer had spread right into the membrane and was approaching the outer face of the membrane.By noting that, in practice, de < dm and Dm Å 1022 De, the Mancy solutions can be summarised as follows.14 (1) At very short time intervals, diffusion in the electrolyte layer alone is rate limiting and the sensor current is given by i nF R D t p n d D t s n = æè öø + - æ è ç ö ø ÷ é ë êê ù û úú = ¥å p p a 2 1 2 2 2 1 1 2 e e e e / exp (15) where ae is the oxygen solubility in the electrolyte layer. Since de is typically 5–10 mm and dm is typically about 20 mm, then eqn.(15) will only hold for t < 0.1 s, since the diffusion layer must be confined to the electrolyte film. Under these conditions, i nF R D t p » æè öø p a p 2 1 2 e e s / (16) (2) At short time intervals, when the diffusion layer has entered the membrane (but has not reached the face of the membrane adjacent to the sample) there is joint transport control and the sensor current is given by i nF R D t p n d D t n = æè öø + - æ è ç ö ø ÷ é ë êê ù û úú = ¥å p p a 2 1 2 2 2 1 1 2 m m s m m / exp (17) This equation can be reduced to a simpler form by noting that typically b Å 20 mm and dm Å 1023 m2 s21, and thus the sensor current can be approximated to i nF R D t p = æè öø p a p 2 1 2 m m s / (18) Eqns.(16) and (18) are similar because, in both cases, diffusional transport is controlling. (3) At long time intervals, the diffusion layer will have spread right into the membrane and will be approaching the outer face of the membrane. The current will therefore be approaching its steady-state value, and under these conditions Mancy et al.gave the current as i nF R P d n D t d n = + æ è ç ö ø ÷ é ë êê ù û úú = ¥å p p 2 2 2 2 1 1 2 m m m m exp (19) and the steady-state current, when the exponential term becomes negligible, is therefore i nF R P d p = p 2 m m s (20) This is exactly the same solution as given by eqns. (7) and (11) when transport in the membrane only is of importance. Fig. 9 shows a theoretical plot of sensor current against time,13 together with the details of the section of the time response corresponding to the Mancy equations.The values for the physical parameters of the electrode used in this model are given in the book by Hitchman.13 Eqns. (16) and (18) are expressions of the well knownell equation, and Myland and Oldham40 later reworked the Mancy theory to describe the current–time behaviour and discussed the role of the various geometrical, transport and solubility factors which could affect this behaviour.Again, this analysis was confined to one-dimensional diffusion (i.e., macrocathodes) and the Myland and Oldham equations were analysed to predict the duration of the sensor Cottrellian behaviour, and the interval before the onset of the steady-state sensor current, for gas-phase or liquid-phase sensors. Although both Mancy et al.39 and Myland and Oldham40 made it clear that their theory was applicable only to one-dimensional diffusion in all three layers, their theoretical predictions have been mistakenly extended to include membrane-covered micro-disc electrodes for which diffusion is two-dimensional.The reason for this is possibly because both unshielded micro-disc electrodes and shielded macro-disc electrodes demonstrate a Cottrellian transient, following a step change in potential. This makes it tempting to assume implicitly that the same behaviour will follow when a micro-disc electrode is shielded by a membrane. 41 Unfortunately, this supposition is erroneous, and we cannot extrapolate the (one-dimensional) Mancy theory to the type of microdisc cathode Clark PO2 sensor used in medicine and biology.41 This is unfortunate, since inspection of eqn. (16) reveals important potential theoretical advantages for a pulsed Clark sensor. Eqn. (16) shows that the sensor current should be proportional to t21 2 for the short time intervals before the diffusion layer reaches the membrane, and therefore a plot of i against t21 2 should reveal a straight line of zero slope for this time epoch.This epoch should therefore define the time span (after the onset of the pulse) during which the sensor current is independent of the membrane characteristics. If the sensor characteristics were to be confined to electrolyte parameters only (or, at the worst, only weakly dependent on the membrane), then a thin membrane which has a fast time response to changes in oxygen concentration could be employed with the sensor.This ideal and theoretical scenario (i.e., membrane independence) would reduce sensor calibration drift, reduce the problem of membrane fouling, which can ruin sensor performance, and minimise (or theoretically eliminate) the problem of the bloodgas difference effect. Another crucial advantage would have been that the sensor could be used as an absolute measuring device [i.e., current output could be determined from a knowledge of the physical parameters of the sensor contained in eqn.(16)]. It could therefore be calibrated in the gas phase and then used in the blood phase without need of further recalibration, because the current output would be independent of the membrane material or properties such as ageing, stretching.41 This would be particularly advantageous for intravascular, or in vivo sensors, of the type described in the in vivo analysis. Unfortunately, all these ideal attributes have proved to be wishful thinking for the micro-cathode Clark sensors used in clinical medicine.There are two reasons for this. First, membrane-covered Clark sensors, when pulsed in practical situations, produce a non-Faradaic current for an ill- Fig. 9 Variation of current per unit area with time for the one-dimensional macro-cathode PO2 sensor theory. The values for the physical parameters of the electrode model are given in the text. (Taken from ref. 13.) 68R Analyst, June 1998, Vol. 123defined time epoch immediately following the voltage pulse.41 The important thing to note here is that an aqueous solution has been used as the electrolyte and that a four-electron reaction has been assumed to take place. Experiments conducted with either macro- or micro-cathode membrane-covered Clark sensors indicate that the simple ‘Faraday current only’ assumption, common to all the one- and multi-dimensional mathematical models, breaks down in real practice. This phenomenon has been reported time and again, and various suggestions for this non-Faradaic time current have been put forward, including the electrode resistance–capacitance discharge time constant, sensor geometry and design and the simultaneous electrochemical reduction of some other species in the electrolyte.14,27,41 The fact that the same micro-disc sensor with the same membrane, but with a non-aqueous solvent such as dimethyl sulfoxide (DMSO) used as the electrolyte solution,41 does not demonstrate the long-lasting non-Faradaic current phenomena when pulsed might indicate that the problem lies with disturbed fourelectron reaction stoichiometry.However, since this longlasting current is still seen when oxygen is totally excluded from the sensor (i.e., when the sensor is pulsed in the presence of nitrogen only), the problem cannot simply be due to disturbed reaction stoichiometry. The non-Faradaic current problem, observed using aqueous electrolytes, is still unresolved.Second, and very important, the failure of the micro-cathode Clark sensor to display Cottrellian behaviour at any time point during the duration of the voltage pulse is due to the radial diffusion of oxygen to the micro-cathode through the electrolyte layer. Gavaghan and co-workers41–44 have reworked the threelayer cylindrical diffusion model, using digital simulation techniques to describe the relationship of the sensor current to micro-cathode size and electrolyte and membrane layer characteristics, when the polarising voltage is pulsed.Their solution involved solving the Fick equation: d d d d d d d d p t P p r r p r p z = + � + é ë ê ù û ú 2 2 2 2 1 in the separate electrolyte, membrane and sample layers using similar boundary conditions to those shown in the steady-state cylindrical diffusion model section, with the exception that the partial pressure at the micro-disc cathode was now taken to be zero immediately following the polarising voltage switch-on. Also, the previous constraint that diffusion was onedimensional in the membrane layer was relaxed, and diffusion was treated as two-dimensional in this layer—as in the sample and electrolyte layers.All equations, in all layers, were solved numerically and simultaneously as functions of time, t. The sensor time-varying current, in this case, is given by i t nFP p z r r R z ( ) = æè ç öø ÷ ò = 2 0 p d d e o d (21) Gavaghan and Rollett45 used an alternating direction implicit (ADI) method to solve the simultaneous equations in a uniform rectangular mesh superimposed on the finite region of Fig. 10, which was then matched to a truncated series solution in the region of a singularity which occurs at the cathode edge. The boundary conditions on z = 0 ensure that there is a discontinuity in the first derivative of the oxygen partial pressure at the point r = R, z = 0, and this forms a ‘boundary singularity’ at this point.The numerical simulation of these equations show that the i-t21/2 behaviour for a micro-disc cathode (diameter < 100 mm) is decidedly non-Cottrellian for all t values. The conclusion of this work was therefore that analytical solutions to the Fick diffusion equation would not produce predictions which matched practice and that, for the type of micro-disc Clark sensors used in medicine and biology, it is necessary to resort to less transparent numerical computations in order to predict the electrochemical sensor behaviour.This becomes even more important when predicting the behaviour of a Clark PO2 sensor when it is switched both on and off with a variable duty cycle. Gavaghan et al.44 later extended their work to produce a theoretical computer simulation of this practical situation, and it is clear that future electroanalytical chemistry models describing non steady-state membranecovered micro-disc cathode behaviour must be based on computer simulation and not on analytical expressions.In-vivo analysis As stated in the Introduction, there is still disagreement between clinicians over the clinical necessity, or cost efctiveness, of measuring PO2 in the blood with intravascular sensors. Since the use of an in vivo sensor necessitates the electrochemical transducer into a patient's artery, such measurements are not undertaken lightly and there must be compelling reasons for performing such measurements.Historically, such a compelling reason arose from the need to monitor, on-line, the arterial PO2 in newborn infants suffering from respiratory distress.46–48 Arterial PO2 measurements are particularly important in the neonatal period (the first month of life) since a high arterial PO2 ( > 11.3–13.3 kPa) will lead to damage to the capillaries of the retina in the eye, producing irreversible blindness (retrolental fibroplasia), and too low a PO2 ( < 7.3 kPa) can lead to tissue hypoxia and brain damage.Early intravascular PO2 electrodes were therefore designed with infants, and not adults, in mind.48 The size of these electrochemical sensors, typically 1.4 and 1.7 mm od, presented no great difficulty in neonatal work, since they could be inserted safely into the umbilical artery of the infant. However, these sensor sizes would present great difficulty for safe insertion into an adult artery. Intravascular PO2 sensors small enough to be inserted into adult arteries (0.6 mm diameter) were produced later,49 but their use has been severely limited to those centres which have the technical expertise to calibrate them and use them.The above ethical and practical reservations do not apply to physiological investigations in non-human subjects, and research into the development of in vivo animal PO2 sensors began, remarkably, immediately following the original Clark publications. In the late 1950s, Kreuzer and Nessler50 published a description of a catheter-tip PO2 sensor, using a 0.8 mm diameter cathode in a polyethylene catheter, with a thin Teflon membrane held on the end with a stainless-steel ring.The ring Fig. 10 Two-dimensional solution region and equations used to model the time-dependent oxygen diffusion processes in a membrane-covered microdisc cathode PO2 sensor. The oxygen partial pressures have been normalised to pA = p/ps, where ps is the initial sample partial pressure of oxygen.Analyst, June 1998, Vol. 123 69Rdid not provide enough tension on the membrane, making it sensitive to arterial blood pulse pressure. One year later, Krog and Johanson51 described an intravascular sensor constructed with platinum and silver electrodes, on the tip of a cardiac catheter, and using Teflon as the membrane. Following these pioneering efforts, a multitude of intravascular PO2 sensors were devised, mainly for use in animals, but sometimes used in human studies lasting several hours.52 Although these early catheter electrodes were plagued by pressure sensitivity, this sensitivity was finally eliminated by the use of membranes tightly stretched over the cathode.The construction and development of these devices, up to the 1980s, has been described by Kreuzer et al.25 and by Hahn.15 There has not been much progress in this field since the 1980s, perhaps owing to a combination of several disparate factors. First, there is the inevitable clinical reluctance to use intravascular devices which can cause blood clotting in an artery and create thrombosis, or introduce a potential source of infection.Other major practical problems include thrombous formation on the sensor membrane itself, which degrades sensor performance, the slow response time of the biocompatible membrane material, sensor sensitivity to blood flow, effects of variations in the patient's temperature on the sensor output characteristics and the sheer magnitude of the problems associated with calibrating the sensor in vivo.Most of these problems, including those of calibration, have still to be overcome. In vitro electrochemical sensors, built into bench blood-gas analysers, can be calibrated at will and their performance can be checked and monitored by quality control material. This is not the case for an intravascular electrochemical sensor, which must be calibrated in situ. This will involve taking an arterial blood sample from the patient, inserting it into a ‘gold standard’ bench analyser and then inserting the true PO2 value into the electrochemical control instrumentation controlling the intravascular sensor.Out of sheer necessity, this will involve a ‘single-point’ calibration, and the clinician will have no idea whether this is adequate or not, or whether the transducer is truly linear in its response to changes in PO2. The question of time response is also difficult to deal with.A membrane which is highly permeable to oxygen will give a fast time response (say of the order of 1 s), but this will increase the oxygen flux through the membrane very considerably and will therefore introduce a large blood-gas difference effect. This, in turn, will make the sensor highly sensitive to blood flow across the membrane face. A less permeable membrane will reduce this flow effect, but will inevitably lead to a long time response to changes in arterial PO2.An inevitable compromise therefore has to be struck, and a material such as polyethylene or polystyrene is chosen as the membrane, in order to minimise the flow effect. The usual consequence of choosing this material is that the sensor response time to a step change in PO2 is typically of the order of 60–90 s to 95% of the final response.15 This is certainly too slow to follow fast physiological changes, but will provide an indication of slower trends in arterial PO2, over periods of several hours.As explained in the mathematical model section, the possibility of using pulse amperometry to control the electrode has not proved successful in practice. Theoretically, pulsing would appear to be an ideal electrochemical control modality, but the non-Faradaic current following the onset on the pulse appears to render the technique useless for present-day intravascular sensors which use aqueous electrolyte layers.41 The development of electrochemical intravascular PO2 sensors has therefore remained fairly static over the past decade, with little progress being made either in their utilisation or development.Two sensors which have stood the test of time are those first described by Parker and co-workers46,47 in 1971 and Mindt and co-workers in 197349 and 1979.53 These are shown in Figs. 11 and 12, since they illustrate two separate approaches to sensor design. Fig. 11 shows the Parker dip-coated catheter PO2 sensor, built into a bi-lumen poly(vinyl chloride) (PVC) catheter. The electrode is at the tip of one lumen of this doublelumen tube, and a rubber-modified polystyrene membrane is dip-coated on to the tip.Prior to this dip-coating, KCl crystals are dip-coated on to the cathode-reference surface and the electrode is sterilised by gamma irradiation after packing and stored dry. The cathode and the anode are silver. The sensor in activated in situ by water transport across the membrane, after it has been inserted into the artery. The second lumen is used to obtain blood samples, which are then introduced into the bench blood-gas analyser to calibrate the catheter sensor.The outside diameter of the sensor is either 1.4 or 1.7 mm, making this sensor suitable for insertion into the umbilical artery of an infant. The characteristics and time response of this sensor could best be described by linear one-dimensional diffusion. A later variation in intravascular sensor design was that of Mindt and co-workers,49,53 illustrated in Fig. 12. The geometry of this sensor was completely different to that of the conventional end-on sensors. The Mindt sensor obeys cylindrical diffusion, since the catheter body itself forms the membrane, and the end of the sensor is designed to be thick enough to be relatively impermeable to oxygen, as illustrated in Fig. 12. Both the anode and the cathode are fine silver wires, and the electrolyte solution is sealed into the catheter with epoxy resin.Hahn15 described the steady-state oxygen transport equations for this type of sensor, and derived the sensor current output as i nFLp r r P r r P = + [ ] 2 2 1 3 2 p s e m ln( / ) / ln( / ) / (22) where r1 is the wire cathode radius, L is the wire cathode length, r2 is the inner catheter membrane radius and r3 is the outer membrane radius. The simplicity of the Mindt design and the stability of its performance characteristics have ensured its longevity.It is Fig. 11 Dip-coated catheter PO2 sensor, built into a bi-lumen PVC catheter. One lumen houses the electral connections to the sensor and the other lumen is used to obtain blood samples for calibrating the sensor in situ. Fig. 12 In vivo PO2 sensor which uses a polyethylene catheter as the membrane. Both the anode and cathode electrodes are silver wires, and the electrolyte solution is permanently sealed into the catheter body. A solid ‘plug’ at the catheter tip ensures that only radial oxygen diffusion reaches the silver wire cathode. 70R Analyst, June 1998, Vol. 123commercially available for use in adult human monitoring, since its outside diameter is 0.65 mm. Its other advantage is that its electrolyte is in liquid form, and the electrode does not need to be activated in situ.53 Sterilisation is achieved by gamma irradiation, as with the Parker design. One important factor to note about the above two sensors is that they both use silver as the cathode material.This presents another, previously unmentioned, problem concerning the use of intravascular PO2 sensors. Since these sensors are only going to be used, out of clinical necessity, with severely ill patients, it is almost inevitable that they are going to be used in some patients who have recently been anaesthetised, or are undergoing anaesthesia. Two of the commonly used anaesthetic agents, nitrous oxide (dinitrogen oxide) and halothane, have been discovered to be extremely electroactive on silver cathode surfaces, and either of these agents can give rise to a ‘pseudooxygen’ signal on the sensor display system.This therefore causes the sensor to ‘fail dangerously’, and this topic will be dealt with in the Anaesthetic agents section. Since the electrochemistry of the catheter tip intravascular sensors is identical with that for the in vitro sensors, there is no new electrochemistry to add here. Because the electrochemistry is established, the haemo-compatibility of the intravascular sensor itself remains the key chemical problem to be solved, if these sensors are to realise their clinical potential.The most important unresolved component of an intravascular sensor is the membrane which is in direct contact with the blood. It is therefore important to try to ensure that the various blood components (water, proteins, red blood cells, white blood cells, platelets and ions) do not react with the membrane surface, and so alter its characteristics.This problem has still not been resolved successfully over the past four decades, but perhaps the most significant recent advance in this area has been made by Zhang and co-workers.54,55 This group has experimented with coating conventional intravascular PO2 sensors with a novel copolymer, poly(MPC-co-BMA), with an electrically neutral head group, 2-methacrylyloxyethyl phosphorylcholine (MPC) copolymerised with butyl methacrylate (BMA), which mimics the lipid bilayer structure of the red cell membrane.This material, and some other biomembrane mimetic polymer surfaces,55 can significantly reduce protein absorption, platelet adhesion and thrombous formation. Zhang and co-workers54,55 therefore coated two types of intravascular sensor, one based on the conventional intravascular sensor (Fig. 11) and the other based on the Mindt design (Fig. 12) with poly(MPC-co-BMA) and showed dramatic improvements in sensor performance when compared with similar sensors not coated with the copolymer.Scanning electron microscopy showed that sensors coated with MPC-co-BMA were effectively free of absorption of blood components on their surfaces.56 Since MPC-co-BMA has poor mechanical properties, it was coated on to conventional intravascular sensor membranes such as polyetherurethane (PU) or PVC. An up-to-date account of the haemocompatibility of various invasive sensors can be found in a recent review by Benmakroha et al.57 If these recent developments do prove successful, there may well be a resurgence in the interest of developing intravascular PO2 sensors, especially if the sensor response time can be reduced to @1 s by using highpermeability membranes, and then reducing the blood-gas, or flow effect, by means of electrochemical control techniques such as pulse amperometry or high-speed voltammetry, where the electrochemistry only occurs for a defined period and is then switched off during a relaxation period.Transcutaneous sensors Gerlach,58 in 1851, investigated experimentally the O2 and CO2 gas exchange between skin and ambient air in animals and humans. One hundred years later, Baumberger and Goodfriend59 reported the determination of arterial PO2 in humans through the intact skin, by immersing a finger into a phosphate buffer solution at 45 °C and measuring its PO2 using a droppingmercury electrode. These findings were confirmed in 1957 by Rooth et al.,60 who used a large bare platinum electrode to measure the PO2.59 A decade later, Huch et al.61 showed that after drug-induced hyperaemia, PO2 values very close to arterial oxygen partial pressures could be measured with surface PO2 sensors on the skin of newborn babies.Practical PO2 sensors were then developed using electrical heating to warm the skin to 42–43 °C, and these first PO2 sensors were called ‘transcutaneous’ sensors.62,63 The sensor (shown schematically in Fig. 13) is basically a Clark sensor housed behind a membrane, which also incorporates an electrical heating element to warm the skin. Two thermocouples measure the temperature of the heating element and the temperature of the skin, to ensure that the patient is not burned. Later developments of this sensor led to the development of a similar sensor for CO2 64 (based on the Stow–Severinghaus sensor) and then to the development and testing of combined O2–CO2 transcutaneous sensors,65 as shown in Fig. 14. Despite the obvious, and attractive, advantages that such sensors do not involve taking a blood sample from the patient or inserting a catheter into an artery, they have proved to be successful only in babies and small children. The potential risk of burning the patient is always present, and the location of the sensor needs to be changed every 2 h or so, to ensure that a blister-forming injury (second degree burn) does not form. Obviously, the incidence of burns is a function of the electrode temperature and the length of time the electrode is left in the same location on the patient’s skin.However, the PO2 measured Fig. 13 Schematic cross-sectional diagram of a transcutaneous PO2 sensor, showing the main constituent parts. The platinum cathode is typically 20 mm in diameter. A small heating element and an NTC resistor are located inside the silver anode for measurement and control of the sensor temperature. 1, epoxy resin; 2, retaining ring; 3, O-ring; 4, Teflon membrane; 5, cuprophane spacer; 6, platinum cathode; 7, silver anode; 8, heating element; 9, NTC resistor; and 10, electrolyte chamber. (From Friis Hansen, B., Marstrand-Christiansen, P., Vesterager, P., and Jacobsen, E., Scand. J. Clin. Lab. Invest., 1976, 37, Suppl., 146.) Fig. 14 Schematic diagram of a combined O2–CO2 transcutaneous sensor. (From Mahutte, C. K., Michiels, T. M., Hassel, K. T., and Trueblood, D. M., Crit.Care Med., 1984, 12, 1063.) Analyst, June 1998, Vol. 123 71Rby a transcutaneous sensor not only reflects the arterial blood PO2, but is also influenced by physiological changes such as peripheral blood perfusion and the core temperature of the patient. When the patient is in shock, or is hypothermic, the transcutaneous PO2 measurement no longer correlates with central arterial PO2, and the device will under-read the arterial PO2. This can be potentially dangerous with some patients, since the transcutaneous PO2 may indicate that the patient is hypoxic (because the sensor is under-reading the true PO2) and the patient might then be given added oxygen to breath.For chronically sick patients with some respiratory diseases, this procedure can endanger the patient's life. Hence, the transcutaneous sensor is not used routinely with adults, and its use has been mainly confined to special care baby units. Since the electrochemical principles of the Clark O2 and Stow–Severinghaus CO2 electrodes contained in transcutaneous sensors have already been described in previous sections, no further comments will be made on this particular type of sensor.Anaesthetic agents Two inhalational anaesthetic agents, nitrous oxide (dinitrogen oxide; N2O) and halothane (CF3CHClBr), have been found to be electrochemically active, in aqueous electrolytes, on the cathodes of Clark PO2 sensors, at the polarising voltages commonly used with them.The sensor currents generated by these two anaesthetic agents have been found to be additive to that of oxygen reduction, thereby producing a ‘pseudo-oxygen’ current which suggests that there is more oxygen present in the sample than there is in reality. The oxygen measuring devices can therefore over-read, and this is a highly dangerous situation since a life-threatening hypoxic episode may be occurring in the patient and yet the O2 sensor is indicating that the blood oxygen partial pressure is ‘normal’ or even ‘high’.In fact, it was the observation of suspiciously ‘high’ blood PO2’s, in the presence of N2O and/or halothane when there was a much lower inspired oxygen gas partial pressure, which alerted clinicians and scientists to the fact that conventional Clark O2 sensors were also reducing molecules other than oxygen.66–73 Apart from nitrous oxide and halothane, no other inhalational agent has been found to be electroactive using the polarising voltage window available in aqueous electrolytes, and these other anaesthetic agents can be treated as ‘inert’ gases as far as the sensor aqueous solvent electrochemistry is concerned.Similarly, no inhalational anaesthetic agent has yet been found to interfere electrochemically with the Stow–Severinghaus PCO2 sensor, since its measurement principle is based on potentiometry. The electrochemical reduction of nitrous oxide and halothane in aqueous solvents appears to depend critically upon the metal used as the working electrode, in addition to the polarising voltage employed.This will be described in the following two sections. Nitrous oxide The history of the electrochemical reduction of nitrous oxide in aqueous solutions has followed a staccato pattern over the past three decades.74–80 Early attempts to reduce N2O with Pt, Au, Hg or Pb electrodes showed that it was either inactive, or at best gave only a transient reduction current.81 Much of this early work was conducted in acid solutions, using macro-electrodes and in the absence of oxygen, presumably because the reduction of O2 might interfere with the interpretation of the results obtained.During this period of initial interest, Johnson and Sawyer80 specifically examined the electrochemical reduction of N2O on platinum electrodes in alkaline solutions of pH 10 and 14. Their experiments did not reveal any N2O electrode activity on Pt surfaces until the surface was deliberately pretreated.When their platinum electrode had obtained a fresh platinum surface, formed by the prior reduction of a platinum oxide film, the reduction of N2O was found to proceed at a polarising potential of -0.8 V (pH 14) or -0.65 V (pH 10). Their data clearly established that the N2O molecule had to be absorbed on the Pt electrode surface before reduction would take place, and that the dependence of the N2O reduction current on electrolyte composition was due to the change of potential of desorption towards more positive values as pH increased.They therefore proposed that the reduction of N2O on a PtO surface followed the sequence80 PtO + 2e2 + H2O ? Pt* + 2OH2 (23a) Pt* + N2O ? Pt(N2O) (23b) Pt(N2O) + 2e2 + H2O ? Pt* + 2OH2 + N2 (23c) Overall, the process could be expressed as N2O + H2O + 2e2 ? N2 + 2OH2 (23d) This work clearly indicated that N2 was formed as a product of the reaction, in aqueous alkaline solutions, their results corroborated those of Zagorski and Suwalski,79 who had, a year earlier, examined the electroreduction of N2O with a droppingmercury electrode in both acidic and alkaline solutions. Although Zagorski and Suwalski failed to produce limiting currents for N2O reduction, their experimental results showed clear erratic N2O reduction current oscillations, whose amplitude grew more and more as the potential was driven more negative.They concluded that the average current intensity, at a chosen potential, was proportional to the concentration of N2O, and we can now deduce that the reduction current oscillations were due to the formation of N2O bubbles from the reduction process eqn.(23). Matters seem to have lain at rest at this point, since most Clark sensors used in medicine and biology employed Pt electrodes and aqueous electrolytes with a pH close to 7.0, and Johnson and Sawyer's work had concluded that N2O would be electrochemically inert on non-pre-treated Pt surfaces.Under these circumstances, N2O would not be expected to be electrochemically active and, in any case, the solvent decomposition potential would be certainly too low to enable the N2O molecule to be reduced. It was therefore clearly assumed, at that time, that N2O was not reducible and would not interfere with blood or gas PO2 measurements. However, a second wave of interest in the electrochemical activity of N2O was already beginning to emerge, when clinicians reported that PO2 measurements could, in fact, be affected by the level of N2O in a blood or gas sample.70,71,73 This ‘interference’ effect was reported70 to be intermittently present in conventional Clark Pt cathode PO2 sensors, but was found to be consistently present in an Au cathode respiratory PO2 sensor.This effect, which corresponded to a ‘pseudo-oxygen’ signal of 7–15% (depending on the polarising voltage) in the presence of pure N2O gas was confirmed on the examination of a Beckman LB1 O2 gas analyser, which incorporated a Clark-type sensor with a gold cathode.When pure N2O was introduced to the gas analyser, the analyser signal read between ‘1% and 5% O2’. The most puzzling feature of this N2O interference effect was that it appeared to occur on gold electrodes used in the clinical environment, but not on gold electrodes used in the electrochemistry laboratory. This dilemma was resolved when it was realised that the clinical sensor Ag/AgCl reference electrode was the culprit, with silver ions from the reference electrode being plated out on the working cathode of the clinical sensor.70 Silver can be easily plated onto either Pt or Au, but the layer is so thin that it is invisible to the eye.However, this silver layer is certainly very electroactive. Detailed studies soon revealed that N2O was reduced quantitatively in the potential region 72R Analyst, June 1998, Vol. 12321.5 to 21.6 V (versus Ag/AgCl) on pure silver electrodes, and that when the reduction process proceeded gas bubbles began to appear on the electrode surface.It was found that the half-wave potential was independent of pH indicating that the ratedetermining step in the reduction was the transfer of the first electron to N2O and this reaction scheme was written as follows: rate determining step: N2O + e ? N2O2 fast steps: N2O2 + H2O + e ? N2 + 2OH2 and so the overall reaction is N2O + H2O + 2e ? N2 + 2OH2 agreeing with Johnson and Sawyers’ conclusion [eqn.(23d)]. The fact that the reaction took place on Ag, and not on Pt or Au, and that the transfer coefficient was 0.28 (i.e., significantly less than 0.5), showed that the first step was not an outer-sphere electron transfer, but that the N2O must have been absorbed on the electrode surface probably through a bond between the oxygen atom and silver.70 This work was conducted in the presence of oxygen, and this highlights the important axiom that electrochemical experiments on clinical sensors must be conducted under those conditions which are expected to be found in clinical practice–for example, experimental results obtained in the absence of oxygen might be totally different to those obtained when oxygen is present in abundance. These studies on the electrochemical reduction of N2O and O2 in the presence of each other revealed clearly separated O2 and N2O reduction waves,70 as illustrated in Fig. 15, which shows voltammograms obtained with a 0.66 mm diameter Ag disc cathode, covered with a 12 mm Teflon membrane, when housed in a Clark sensor body (Fig. 4) at 25 °C. The O2–N2O gas mixture was varied between 0 and 100% v/v for each gas, with the other making up the balance. The net result was a series of O2 and N2O voltammograms as shown in Fig. 15, each with clearly identified and separated half-wave potentials and good plateaux. The absolute size of the current is a reflection of the transport properties of the membrane for O2 and N2O and, apart from the number of electrons involved, the ratio of the N2O and O2 currents is governed by the permeabilities of these two gases in Teflon.When the limiting currents for O2 and N2O (corrected in this case for the O2 current) were plotted against the percentage of each gas in the mixture, straight line relationships were obtained showing that nitrous oxide concentration (in addition to oxygen) could now be measured electrochemically. Following the report which clearly indicated that N2O could be reduced electrochemically on silver contaminated surfaces,70 other reports then began to appear on oxygen gas analysers failing dangerously in the presence of N2O, on blood-gas analysers becoming sensitive to N2O and so leading to the potential non-diagnosis of arterial hypoxaemia,71,73 and on an intravascular PO2 blood sensor grossly over-reading in patients breathing O2 and N2O gas mixtures.82,83 The intravascular blood PO2 sensors which were commercially available in the 1970s often employed silver cathodes.Because silver has a much wider electrochemical reduction voltage window than Pt (at the same electrolyte pH), silver cathode PO2 sensors were (and are) exquisitely sensitive to N2O, if the polarising potential is not kept low enough to escape the rising portion of the second reduction (N2O) wave (Fig. 15). Clinical reports, published at that time, which used Ag cathode intravascular PO2 sensors in the presence of N2O, appeared to be unaware of this ‘faildangerous’ phenomenon.84 One positive outcome of this work was that it was realised that a Clark-type sensor, employing a silver micro-disc electrode and a conventional aqueous electrolyte solution, could be used for the simultaneous determination of O2 and N2O in gas or blood.70,85 Since the outcome of the N2O reduction process was the formation of N2 gas bubbles [eqn.(23d)], which rapidly occluded the electrode surface and prevented any further electrochemistry occurring, the Ag micro-cathode membranecovered sensor could not be polarised in the conventional constant-potential manner. The solution to this problem was to employ a train of polarising voltages in a pulsing regime (with an appropriate duty cycle) to avoid the build-up of N2 on the micro-electrode surface.86 The first voltage pulse was used to reduce O2 and the second to reduce both O2 + N2O.Since the simultaneous O2 and N2O reduction processes did not interfere with each other, the current obtained from the second pulse was simply an addition of the O2 and N2O reducing currents.87 An appropriate computer algorithm was therefore employed to deconvolute the O2 and N2O current–concentration contributions. The determination of the appropriate duty cycle to employ under these experimental conditions can be obtained from analysis of the digital pulse simulation procedure described in the computer model simulation section.In 1982, a novel prototype O2–N2O ‘sandwich electrode’ was described,88 based on a metallised membrane PO2 sensor by Bergman,89–91 to measure O2 and N2O in gas mixtures. The outline design is shown in Fig. 16 and it comprised two separate electrochemical sensors. The first sensor was a Teflon mem- Fig. 15 Voltammograms for O2–N2O gas mixtures, obtained with a membrane-covered (12 mm Teflon) 0.66 mm diameter electrode, showing distinct and separate plateaux for O2 and N2O reduction.Fig. 16 Schematic cross-sectional view of a composite O2–N2O sandwich electrode. A nylon spacer is inserted between the metallised membrane and the inner (second) macrodisc cathode. (Taken from ref. 88.) Analyst, June 1998, Vol. 123 73Rbrane with one surface coated with a thin layer (5–10 nm) of gold, deposited by vacuum evaporation. This layer, when polarised at -0.6 V formed an oxygen filter, which effectively scrubbed all oxygen from any gas sample diffusing through the membrane.The limiting current generated at the metallised layer was proportional to the PO2 of the gas sample. If the gas sample contained N2O, these molecules passed through the membrane (since they were not reduced at 20.6 V, on Au surfaces) to the second sensor, which consisted of a silver disc cathode (3 mm diameter) which was pulsed with a regime similar to that described above for the intravascular sensors.Thus, the N2O concentration in the gas mixture was measured by pulse amperometry at the second cathode. It was hoped that this design would also facilitate the electrochemical measurement of halothane (see the next section) and so create a composite O2 + N2O + halothane sensor, but technical difficulties precluded any further development, despite its success at measuring O2 + N2O simultaneously. More recently, there has been yet another resurgence of interest in the electrochemical reduction of N2O on metal electrodes, but this more recent work (with one exception) has been confined to the non-clinical area.92–96 The one exception is the discovery that N2O can be reduced on gold micro-electrodes in aprotic media, and this will be described further in the nonaqueous section of this review.Halothane Halothane was the first satisfactory non-explosive and potent inhalational anaesthetic agent. It is a fluorinated hydrocarbon, liquid at room temperature, with the structural formula CF3CHClBr, and was synthesised in 1951.97 By the late 1950s, clinical evaluations of halothane showed that it was effective and safe in a wide assortment of clinical conditions and in all age groups, and its great versatility became apparent. Halothane therefore rapidly spread into worldwide use, and it is still in use today despite worries concerning its hepatotoxicity.What was not realised, in the early days following its introduction to clinical practice, was that halothane was electrochemically active on both silver and gold surfaces.The reasons for not observing this electrochemical activity are as follows. Halothane is a large molecule, with a relative molecular mass of 197, and so the membranes used in conventional Clark blood-gas sensors, such as polypropylene and Teflon, provide a natural barrier to the potential penetration of the halothane molecule to the sensor cathode surface.Furthermore, if a conventional Pt surface is uncontaminated from the deposition of Ag+ ions, the halothane molecule is ‘reluctant’ to be reduced on the ‘clean’ (platinum) surface. It needs a long dwell time of a blood sample, containing halothane, in the analysis chamber of a blood-gas analyser, combined with a membrane reasonably permeable to halothane and a silver-contaminated platinum micro-cathode surface, before electrochemical activity is observed. However, these particular and necessary requirements can be met in clinical practice and Severinghaus et al.,98 in 1971, reported that conventional Pt micro-cathode membrane-covered Clark sensors could become extremely sensitive to the presence of halothane vapour (over-reading the PO2 for room air by up to 1600%), following a 5 min exposure of the sensor to a gaseous sample of halothane vapour in N2.The sensor was covered with a 25 mm polyethylene membrane but when it was covered with a 25 mm polypropylene membrane the equilibrium with halothane was delayed by about 20 min, probably owing to the reduced permeability of the polypropylene.(When covered with this material, one sensor did eventually over-read by 800%, under the same conditions as before.) Severinghaus et al. found that the sensitivity of the O2 sensor to halothane depended upon the polarising voltage, the membrane material, the electrolyte pH and the deposition of Ag on the Pt cathode surface. Similar studies conducted with oxygenated blood, brought into equilibrium with 2.6% v/v halothane, demonstrated a slow upward drift of the sensor PO2 reading by 100% after 1 min.Thus, an early clear warning was established that the presence of halothane, in clinical concentrations, could cause PO2 sensors to over-read dangerously. This report was soon followed by further examples. Bates et al.,99 in 1975, reported that intravascular PO2 sensors with gold micro-cathodes and covered with a hydrophilic membrane (Hydron) were sensitive to halothane when polarised at 20.7 V, and the time required to achieve maximum halothane effect was approximately 18 min.Soon after this, Dent and Netter100 demonstrated that halothane (in the presence of oxygen) could also be reduced on gold micro-electrodes which had been specifically developed for measuring oxygen partial pressures in micro-areas of living tissue during anaesthesia. Since it was apparent from their work that the halothane and oxygen reduction waves overlapped, it was impossible to distinguish between the halothane and oxygen reduction currents, and so this rendered the sensor useless for oxygen tissue measurements in the presence of halothane anaesthesia.Their only solution to this dilemma was the hope that a coating material permeable to oxygen, but impermeable to halothane, might be developed and so permit oxygen measurements to be made with their microelectrode in the presence of halothane.100 These reports were soon followed by a clear clinical warning, in a letter to The Lancet in December 1978,72 which highlighted the need for vigilance in exposing commercial blood-gas analysers to blood samples containing halothane.The letter described a gradual upward drift in the analyser PO2 readings, following halothane exposure, and vigilance was emphasised because the errors induced dangerous over-reading of PO2 in critically ill patients. Following this warning, steps were taken either to abrade the cathode surface to remove silver deposition, or else to clean the Pt surface regularly in a nitric acid solution in order to remove potentially harmful Ag deposits.However, the problem of halothane reduction on Ag or Au micro-cathodes used in intravascular sensors still remained unsolved, since these sensors inevitably used hydrophobic membranes which were very permeable to halothane molecules. The only solution was to not use these types of intravascular PO2 sensors in the presence of halothane anaesthesia.The chemistry of the reduction of halothane is not clear-cut, since both the reaction mechanism and the role of the cathode metal have been a matter of controversy. Early studies were concerned with the reduction of halothane on the droppingmercury electrode and are of little relevance to clinical sensors.101–103 The fact that these matters are still unresolved is perhaps due to the way in which studies in different centres have employed different experimental conditions, with large rotating ring-disc electrodes, large stationary wire electrodes or microelectrodes being employed in some studies, and some conducted in the absence and some in the presence of oxygen.Most of the studies were conducted at unshielded electrodes, with just two reported with membrane-covered micro-disc electrodes. 104,105 Hence it is possible that the wide variety of results and mechanisms proposed are due solely to the particular experimental conditions encountered when the studies were conducted.Albery and coworkers,104 in 1981, reported on the amperometric reduction of halothane in 1 m KOH solution using a 1.35 mm diameter silver disc rotating-disc electrode and a membrane- covered 50 mm diameter silver micro-disc electrode, with 25 mm silicone film and 75 mm silastic rubber membranes.104 Measurements were made both in the presence and absence of oxygen, but their results showed that a micro-disc silver cathode electrode covered with a suitable polymer membrane, such as silicone film, might be used to measure halothane concentration 74R Analyst, June 1998, Vol. 123in the absence of oxygen, but not in the presence of oxygen.The difficulty was that the half-wave potentials for oxygen and halothane were too close, 20.425 and 20.565 V (versus SCE), respectively. Fig. 17 shows the reduction wave for halothane at a silver rotating disc electrode in 1 m KOH solution, taken from the work of Albery et al.104 The halothane reduction effect commences early at approximately 20.3 V versus SCE and the current is transport limited for potentials more negative than 20.7 V.Thus, the halothane wave overlaps that of oxygen (cf., Fig. 5) in addition to that of nitrous oxide (cf., Fig. 15) in alkaline electrolyte solutions. Their conclusion was that the reaction of halothane, in the absence of oxygen, is a two-electron reduction: Ag + e2 + CHClBrCF3 ¡ slow – (Ag···Br···CHClCF3)2 ¡ fast, H2O, e2 – Ag + Br2 + CH2ClCF3 + OH2 The overall mechanism therefore leads to 2-chloro-1,1,1-trifluorethane and bromide, with the overall equation written as: CHClBrCF3 + 2e2 + H2O ? CH2ClCF3 + Br2 + OH2 (24) with the bond split in the reaction almost certainly being the C– Br bond, since bromide is detected as a reaction product.This work suggested that the silver electrode might have a particular affinity for Br.104 Later studies suggested that gold was a better cathode material to use,105 since O2 and halothane displayed clearly separate reduction waves on gold.Unlike on silver, where oxygen reduction appears as a single diffusion-controlled wave, on gold two plateaux of similar magnitude were observed, attributed to two two-electron steps: O2 + H2O + 2e ? HO22 + OH2 (25a) HO22 + H2O + 2e ? 3OH2 (25b) A comparison of the current–voltage plots for oxygen and halothane revealed two potential domains of interest, where oxygen alone was reduced between 20.3 and 20.8 V versus SCE [eqn.(25a)], and between 21.1 and 21.6 V where the second oxygen wave [eqn. (25b)] was coincident with the halothane reduction. The separation between these two waves therefore appeared suitable for the application of a double potential pulse experiment, of the type already described for the oxygen/nitrous oxide assay on a membrane-covered silver micro-disc electrode.(This approach was also encouraged by the absence of a nitrous oxide reduction wave on gold, under the prescribed experimental conditions.) However, when these conclusions from the rotating disc experiments were tested with a membrane-covered 125 mm diameter gold disc micro-electrode, current–voltage plots revealed that the magnitude of the current on the first oxygen wave was depressed (at a constant O2 concentration) in the presence of halothane, as shown in Fig. 18.The depression was non-linear, with the largest reduction being observed between 0 and 1% v/v halothane. Therefore, unlike the rotating disc electrode, the membrane-covered sensor assay of O2 was not independent of halothane, and the presence of the anaesthetic agent led to large discrepancies in the determination of the oxygen concentration. Hall et al.105 developed a strategy for assaying halothane in the presence of oxygen by applying a potential step to a polarising voltage where, for short times, the oxygen reduction was under diffusion control whereas the halothane reduction was under mixed kinetic and diffusion control.Oxygen and halothane were thus assayed by the application of a single rectangular wave polarising voltage and current sampling at two points, one at very short times and the other at longer times when both analytes were approaching diffusion control. The resultant current–concentration plot was linear for halothane up to a saturation value, but the sensor still suffered from problems when the gold electrode was contaminated with trace amounts of silver.A second attempt at the development of a practical halothane sensor, to be used in the presence of oxygen, was reported in 1989, again using a rotating ring-disc electrode for fundamental studies and then a 125 mm gold micro-disc electrode, covered with a 6 mm thick silastic membrane.106 Again, results with the unshielded and shielded electrodes differed (with one problem being that the reduction of oxygen on a gold electrode behind a membrane with only a thin electrolyte layer appeared as a single wave, in contradistinction to the two waves which appeared at an unshielded electrode), and the membrane-covered sensor appeared to give a linear halothane current response (up to 4% v/v halothane) only as long as the prevailing oxygen concentration was above 60% v/v.As with the previous attempt,105 the development of a sensor which would unambiguously assay both oxygen and halothane in the presence of each other was proving to be an extremely difficult task.At the beginning of the 1990s, a third attempt was made, this time by Mount and co-workers,107,108 to study the electrochemical reduction of halothane and to develop a practical O2/ halothane sensor. Accordingly, the first studies were conducted with silver, platinum, gold and glassy carbon ring-disc electrodes, with silver used as the ring in each case and with copper disc electrodes.In all cases, the diameter of the disc was Fig. 17 Current–voltage curve for halothane reduction on a rotating-disc Ag electrode in 1 m KOH electrolyte solution. Fig. 18 Voltammograms for 0–4% halothane in oxygen, at a 125 mm diameter gold cathode membrane-covered Clark sensor at a constant PO2. The depression of the first reduction wave (O2 reduction) is clearly seen as the halothane concentration is increased. Analyst, June 1998, Vol. 123 75Rapproximately 7 mm. The mechanistic studies were conducted in the absence of oxygen in an electrolyte solution of 0.1 m potassium hydroxide. Halothane was supplied from an anaesthetic vaporiser in the range 1–4% v/v, and the studies concluded that, for a wide variety of electrode materials, electrochemical reduction of halothane involved two electrons and resulted in the production of a bromide ion at the silver ring electrode. As far as the mechanism was concerned, the conclusion was that the first electron transfer was irreversible and was the slowest of the two, followed by a chemical step which was always fast and never rate-determining, followed by a second electron transfer which might be irreversible or by a subsequent fast chemical step.The following mechanism was postulated:107 CHClBrCF3 + e2 ? (CHClBrCF3).2 (26a) (CHClBrCF3).2 ? (CHClCF3). + Br2 (26b) (CHClCF3). + e2 ? (CHClCF3)2 (26c) (CHClCF3)2 + H2O ? CH2ClCF3 + OH2 (26d) and so, overall, CF3CHClBr + H2O + 2e2? CH2ClCF3 + Br2 + OH2 (26e) agreeing with the conclusions of Albery et al.104 [eqn.(24)]. Because both chemical steps were never rate determining, it was impossible to know the precise sequence of the electrochemical and chemical steps. A further paper by the same group108 examined the electrochemical reduction of halothane in the presence of oxygen, and this work concluded that silver or gold was the choice of disc electrode metal and that the detection of the bromide ion product from the halothane reduction (at a silver ring electrode) provided a highly successful method for measuring the concentration of halothane in the presence of oxygen, in any mixture of the two gases used for anaesthesia.In this instance, oxygen concentration would be measured at the silver or gold disc electrode. This work also demonstrated that the pH of the electrolyte solution had to be 11, since some oxygen reduction species were also detected on the ring at pH 13.In this technique, both the ring current due to the formation of solid silver bromide and the subsequent reduction charge required for its removal could be used to measure the halothane concentration.109 Since it was clear that a clinical sensor needed a single stationary working electrode rather than a ring-disc electrode configuration, Mount and Clark described, in a subsequent paper,109 an unshielded halothane/oxygen sensor utilising the chronoamperometric measurement of halothane and oxygen concentrations at a silver electrode.The silver disc electrode was 7 mm in diameter and the studies were conducted in a standard electrochemical reaction cell at room temperature, i.e., the sensor was not membrane covered. Their technique employed a triple potential step regime, where the electrode was initially at a potential of 20.15 V. The electrode was then pulsed to 20.85 V, at which point mass transport-limited reduction of halothane (two electrons) and oxygen (four electrons) occurred.The electrode potential was then stepped to a potential between 0.1 and 0.2 V, at which solid silver bromide was formed on the electrode surface from the bromide ion produced by the halothane reduction which took place during the first pulse. Finally, after a time delay, the electrode was pulsed to its starting potential of 20.15 V where the reduction of the solid silver bromide on the electrode occurred according to the reverse reaction Ag(s) + Br2(aq) " AgBr(s) + e2 (27) From their experimental data, it was clear that the current measured during the second pulse was not a good measure of the amount of halothane present, because of complications of either nucleation effects or secondary oxidation reactions.However, the oxidation of bromide which occurs during the second pulse [and which formed the solid silver bromide according to the forward reaction in eqn. (27)] was reduced according to the reverse reaction during the third pulse and this was found to be a good measure of the amount of halothane present.The authors recognised that their system was ‘idealised’ since the sensor did not have a membrane,109 and a membrane will always complicate gas transport considerably, as highlighted in the O2 Clark sensor section of this teaching review. Thus, the Mount- Clark O2/halothane sensor will have a much more complex current–time response when the sensor is covered by a membrane; transport of the gas in the thin layer of electrolyte will be bounded rather than governed by semi-infinite linear diffusion and the finite rate of transport of gas through the membrane will determine the sensor response. Also, the pH in the thin layer of electrolyte adjacent to the disc may rise more rapidly above pH 11 during the first reduction pulse and then fall more slowly, as radial diffusion of hydroxide ions in the thin electrolyte layer will be less efficient at maintaining the pH at a steady level.However, despite these complications, the triple pulsing regime is clearly a promising way forward, and it should be possible to model this type of sensor using numerical analysis techniques analogous to those described in the Clark O2 sensor section. More recently, Langmaier and Samec110 have questioned the ECEC mechanism proposed in eqn. (26), and have commented that the absence of electrocatalysis in this reaction is surprising.Their criticism is that the mechanism described by eqn. (26) does not account for the strong effect of the nature of the metal on the halothane reduction process. In their examination of the reduction of halothane (in nitrogen) on a wide variety of metal electrodes, they found that the half-wave potentials of halothane reduction at various metal electrodes could differ by several hundred millivolts in both methanol and water. They concluded that such behaviour might be due to the absorption of halothane on the metal surface M, following the scheme (M···S) + CHClBrCF3 " (M···Br···CHClCF3) + S (27a) followed by uptake of the first electron as the rate-determining step: (M···Br···CHClCF3) + e2 ? M + Br2 + (CHClCF3).(27b) where S is the absorbed solvent molecule. These steps would then be followed by the fast reactions in eqns. (26c) and (26d). However, their work confirmed the previous conclusions that the electrochemical reduction of halothane proceeded through a two-electron overall scheme yielding 2-chloro-1,1,1-trifluoroethane and bromide as the main product, with the uptake of the first electron by the halothane molecule being the ratedetermining step.The latest word (up to this present time) in this quest to detect halothane in the presence of oxygen must belong to Caruana and Giglio,111 who have devised a method of measuring halothane by anodic stripping voltammetry, although they appeared to be unaware of the chronoamperometric technique of Mount and Clark.109 The technique of Caruana and Giglio is different to that of other workers in that they used an acidic electrolyte solution (based on citric acid) with a pH of 4.0.Their electrode was a 0.698 cm diameter gold electrode and cyclic voltammograms were measured in the range 20.3 to +0.9 V. These cyclic voltammograms showed a clear irreversible reduction peak at 20.3 V and a sharp peak at +0.62 V, the height of which was linearly dependent on scan rate.This suggested that the oxidised species was absorbed on the electrode surface, and the oxidation peak only occurred when reduction had taken place on the electrode, confirming that that peak at +0.62 V was due to the oxidation of the absorbed products from the reduction of halothane. Initial calibration plots of halothane concentration against the peak oxidation current showed that the peak current 76R Analyst, June 1998, Vol. 123was independent of oxygen concentration in the electrolyte solution, but the linear range for halothane measurement was between 18 and 155 nm, since beyond 155 nm the response saturated at a constant peak height. Since the equivalent halothane concentration in pH 4.0 buffer for 4% v/v halothane would be 0.44 mm, the cyclic voltammetric technique could not be used to measure halothane under clinical conditions. This problem was obviated by adjusting the sensor sensitivity by using a short duration potential step for 50 ms at 20.7 V, followed immediately by a potential sweep from 0 to +0.8 V at a sweep rate of 50 mV s21. The result of this pulsing/sweeping regime was that there was a clear linear relationship between oxidation peak current and halothane concentration between 0 and 4% v/v halothane.Furthermore, this linearity was unaffected by the presence of oxygen. Although this new development is obviously very promising, it must be remembered that these experiments112 were conducted in a reaction cell, and the sensor (like that of Mount and Clark109) was not covered by a membrane permeable to halothane.Furthermore, the sensor would not provide a cotemporal measurement of O2 concentration, unlike the Mount and Clark device. It is clear from the above that a further (and possibly large) step is required before a practical membrane-covered sensor for halothane can be developed, and this development would be greatly aided by a digital simulation model which incorporated both membrane and electrolyte layers, and which would enable both pulsing and sweep control techniques to be modelled.This stage has yet to be realised. There is an ironic practical twist to this quest to develop an electrochemical halothane sensor since the increasing availability of alternative volatile anaesthetic agents, and the worries about the pathophysiology of post-halothane hepatic dysfunction, have led to cautionary advice regarding the medico-legal implications of the repeated use of halothane,112 including calls for it to be made obsolete.113 Although halothane will no doubt remain popular in undeveloped countries, by virtue of its low cost and high anaesthetic potency, opinion seems to be hardening that the days of halothane anaesthesia in the UK are limited.A recent anaesthetic survey has clearly indicated that a combination of the fear of litigation and the availability of newer ‘cleaner’ inhalational anaesthetic agents has brought about a sharp decline in the use of halothane, especially amongst anaesthetists in training.114 Halothane has for years been a reliable agent for use in difficult situations, but as alternative agents are developed for the safe inhalational induction of anaesthesia (a field where halothane has been dominant), then halothane may become little more than an item of historical interest to a new generation of anaesthetists.Unfortunately for the electrochemist, the new inhalational agents, including isoflurane (CHF2OCHClCF3), enflurane (CHF2OCF2CHClF), sevoflurane [CH2FOCH(CF3)2] and desflurane (CHF2OCHFCF3), all appear to be electrochemically inert in aqueous electrolytes.However, all is not lost electrochemically, since it is becoming apparent that most of these agents might be electrochemically active on micro-electrodes in non-aqueous media, as described in the following non-aqueous section.Non-aqueous solution electrochemistry The history of the electrochemical measurement of O2 and CO2 in non-aqueous solutions is much shorter than that in aqueous solutions, as described in the previous sections, and aprotic solutions have certainly not been used in blood-gas sensors so far. Furthermore, aprotic electrochemistry has not been used for the gaseous determination of O2, CO2 or the inhalational anaesthetic agents in clinical medicine, although prototype gas sensors have been developed in recent years.This apparent lack of interest is probably due to the success of the well tried and tested Stow–Severinghaus and Clark aqueous electrolyte sensors, and to the fact that aprotic electrolyte solutions could pose new dangers to the users of blood (or gas-phase) sensors owing to the potential toxicity of these substances. Furthermore, since they would attack the plastic materials currently used in clinical sensors, their use would necessitate new materials and new design procedures.Why, then, should we bother to design new clinical sensors using aprotic solutions? The answer is that aprotic solutions provide a very wide voltage window (up to 23.0 V) for the electrochemical reduction of potentially reducible clinical gases and vapours. This opens up the possibility of developing a single sensor which could measure a mixture of gases and vapours with a single electrochemical technique. This is a laudable aim, but the journey towards realising this aim has been bedeviled with problems so far, notably the crossinterference of the various reaction products.These problems will be described in the following sections. The non-aqueous electrochemistry will be described taking DMSO as the solvent, since DMSO is already used therapeutically in clinical medicine for urinary disorders and as an anti-viral preparation. It therefore does not pose a toxic hazard to the user provided that standard laboratory precautions are employed.Because of the way in which this particular electrochemical history has evolved, the reduction of O2 and CO2 in aprotic media will be described first, and the following two sections describe the electro-reduction of oxygen and carbon dioxide, in the absence of each other, although these conditions are decidedly non-clinical. Special problems exist when these two gases are reduced in the presence of each other, and this is considered in the third section. Finally, preliminary results on the electro-reduction of inhalational anaesthetic agents in DMSO are described.Oxygen reduction The electrochemical reduction of oxygen in aprotic media was studied in the 1960s by a small group of workers.115–117 This, and subsequent, work was extensively discussed by Bauer and Beck,118 in 1972, who reviewed the electrochemical behaviour of O2 in no fewer than 33 solvents and fused salts. Most of these studies were conducted with either Pt or Au solid macroelectrodes, or with the dropping-mercury electrode, using cyclic-voltammetry.In one instance, controlled amounts of water were added to DMSO and the effect of this on the electrochemical reduction of O2 in DMSO was examined.119 The results of these investigations in DMSO, with tetraalkylammonium salts as the supporting electrolyte, showed that two distinctive, and clearly separated, oxygen reduction waves, as shown in Fig. 19. The first wave is due to the electroreduction of oxygen to the superoxide ion, O2 .2 produced by a oneelectron process.The second wave is due to the further reduction of O2 .2 via one electron, to O2 22. Furthermore, since the first process is reversible, the overall process can be written as O2 + e " O2 .2 (28a) O2 .2 + e ? O2 22 (28b) If the electrode polarisation is maintained in the vicinity of the first wave, it has been shown that stable solutions of O2 .2 can be generated, and that O2 .2 undergoes several reactions where it can function as a base, a nucleophile and a one-electron donor.120,121 In the mid-1990s, renewed interest in the electroreduction of oxygen in aprotic media led to investigations on the reactivity of O2 .2, following the electrochemical reduction of oxygen using cyclic voltammetry and rotating-disc electrodes, and to hydrodynamic chronocoulometric studies to determine the diffusion coefficients and concentrations of oxygen in aprotic media.122,123 These studies confirmed the earlier conclusions, namely that oxygen is reduced initially in a one- Analyst, June 1998, Vol. 123 77Relectron reversible diffusion-limited step to the superoxide ion, which is further reduced to the peroxide, O2 .2 This second step was observed as a highly irreversible peak or wave at more negative potentials, indicating that O2 22 is highly unstable.123 Although this work is electrochemically interesting, it has had no impact on the design of oxygen macro-electrode sensors, presumably because the conventional Clark O2 sensor (with aqueous electrolytes) performs so well in clinical practice.Carbon dioxide reduction Early work, again conducted in the 1960s and 1970s, on the electroreduction of CO2 in aprotic media initially revealed ambiguous and conflicting results.124–126 These early studies were conducted with a variety of electrode materials (including mercury, gold and lead) and with a variety of aprotic solvents including dimethylformamide (DMF), acetonitrile and DMSO.Some results indicated a one-electron reduction step and others a two-electron step. The number of electrons transferred appeared to depend upon the applied potential, the nature of the metal, the solvent employed and the presence or absence of water in the solvent. As with the O2 reduction studies, a supporting electrolyte such as tetraethylammonium perchlorate was employed in the electrochemical studies. All these studies revealed a single CO2 reduction wave in aprotic media when CO2 was reduced (in the absence of O2) in N2.The CO2 reduction wave was shifted to much more negative potentials (about 22.5 to 23.0 V versus Ag/AgCl) than the first O2 reduction wave. An illustration of this is also shown in Fig. 19, which compares the positions of the O2 and CO2 reduction waves relative to each other. What was clear, from these early and then subsequent studies, was that a complicated set of competing reactions, with differing pathways, were possible.It was reported that the possible reduction products were C2O4 22 (oxalate), HCO22 (formate), carbon monoxide and carbonate, and possibly glycolate under the influence of residual, or purposely added, water.127–131 Viewed overall, these reactions, in an anhydrous system, were presented as 2CO2 + 2e ? C2O4 22 (29a) 2CO2 + 2e ? CO + CO3 22 (29b) and when water was present as CO2 + H+ + 2e ? HCO22 (29c) However, what was consistent in these early studies was the conclusion that, irrespective of whether water was present or not, the electroreduction of CO2 first involved a one-electron step to form CO2 .2.Thereafter, a variety of competing pathways were possible, and these were summarised by Gressin et al.129 and by Amatore and Sav�eant130 in the scheme shown in Fig. 20. Sav�eant’s group further pointed out that the distribution of the products depended strongly upon the experimental operational factors such as current density, concentration and diffusion layer thickness.130 They also made the important observation that these factors rendered the extrapolation of macro-scale electrolysis results to the context of microelectrolytic techniques as ‘uncertain’ (and the opposite argument obviously applies), and this is an important fact to remember when results are obtained with micro-electrodes.Again, these early studies have had little impact on the development of CO2 sensors using aprotic solvents, for two different reasons.Firstly, the Stow–Severinghaus sensor is so well established that there has been little room for a competing amperometric CO2 sensor. Second, and more important, these macro-electrodes studies on CO2 reduction in nitrogen are rendered ‘academic’ when they are repeated in the ‘real-life’ presence of oxygen, as illustrated in the next section. Reduction of O2 and CO2 in the presence of each other It is all very well examining the reduction of CO2 under anaerobic conditions and the reduction of O2 in the absence of CO2, but (with the exception of inspired gases which do not contain CO2) clinical expired gas and blood-gas analysis will always involve a mixture of both gases.At this point, the electrochemistry of O2 and CO2, in the presence of each other, becomes decidedly complicated at macro-electrode surfaces. (A clearer picture emerges when micro-electrodes are employed, but this will be considered later.) On the basis of the evidence presented in Fig. 16, and on the evidence previously obtained when O2 and N2O were reduced together in aqueous electrolytes (Fig. 15), it would be expected that O2 and CO2 would produce two distinct and clearly separated reduction waves when reduced in the presence of each other. In this theoretical case, the second reduction wave would be expected to be an addition of the CO2 reduction current and the second O2 reduction current wave.However, this is not obtained in practice, and the wave shown in Fig. 21 is obtained instead experimentally132,133 on macro-cathode surfaces. The CO2 reduction wave has now completely disappeared, and it can be seen that the single ‘oxygen’ reduction wave has increased somewhat in magnitude from that shown in Figure 19. The conclusion must be that the superoxide anion radical, O2 .2, reacts irreversibly with the dissolved CO2, and is effective in reducing the concentration of dissolved CO2 prior to its possible reduction at more negative potentials.These experimental studies, which were conducted with macro-electrodes, appeared to confirm that it would be impossible to determine O2 and CO2 electrochemically, in the presence of each other, with a simple electrochemical technique such as voltammetry.133 However, three attempts have been made to circumvent this superoxide interference effect by devising gas-phase O2–CO2 sensors, using macro-electrodes and various degrees of ingenuity.First, Albery and Barron134 tackled this electrochemical conundrum by attempting to ‘scrub’ electrochemically the oxygen from a gas sample and then measure the residual CO2 electrochemically. Accordingly, they modified the O2–N2O metallised membrane electrode system described previously,88 and devised a new double membrane/double solvent layer sensor.134 The outer membrane consisted of a metallised membrane, polarised at 20.7 V, where O2 was reduced in an aqueous media electrolyte layer, buffered at pH 5 using a phthalate electrolyte.The current on the metallised Fig. 19 Linear sweep voltammograms for the independent reduction of O2 in N2 and CO2 in N2 at an unshielded Au working electrode in DMSO. 78R Analyst, June 1998, Vol. 123C2O4 2– CO2 •– HCO2 • CO2 + 1e– HCO2 – C C O O O O• C O O O O C C O O O O C C O O O O C C2O4 2– C2O4 2– + CO2 • CO + CO3 2– + CO2 •– + CO2 + H2O + 1 e– + CO2 •– + 1e– + CO2 •– + 1e– + CO2 •– + CO2 membrane was proportional to thecentration of O2, and it was found that the metallised membrane filtered the O2 with 99% efficiency.The remaining CO2 molecules then passed through the metallised membrane, and through the aqueous electrolyte, to another membrane with an inner compartment containing a non-aqueous electrolyte (usually DMSO) and a silver macro-cathode. The CO2 was reduced to CO2 .2 at this second electrode surface. As is clear from the above description, the electrode construction was complicated and a major problem with this approach was that the two-compartment arrangement introduced a considerable gas diffusion barrier, which manifested itself in a relatively slow response time.Although Albery and Barron134 published results showing this sensor could be used to determine gaseous O2 and CO2 on a breath-by-breath basis, this work could not be replicated by Coombs132 and Clark,133 and the sensor appears not to have been developed any further.In the second attempt to develop a practical sensor, the electrochemical filter technique was abandoned, and a pulsed titration technique was developed in its stead.135,136 The principle of this approach for the electrochemical analysis of mixed gaseous samples of O2 and CO2 is shown schematically in Fig. 22. The method of determination depends on reducing O2 for a fraction of a second, at a potential where CO2 is inactive, in order to generate a known amount of the reactive O2 .2 anion radical, and thereby initiate the fast O2 .2–CO2 reaction.The amount of O2 .2 subsequently left unreacted after a given time was then determined by pulsing the working electrode potential to a more positive potential, and re-oxidising the remaining O2 .2 to O2. The collection efficiency for the O2– O2 .2 redox couple was estimated in the absence of CO2, and then subsequently with CO2 present, and the changes in both the generation and recovery transients due to the deactivation of O2 .2 and the regeneration of O2 were observed. By quantifying these observed changes with respect to a theoretical model of the overall system, it was possible to infer the sample concentrations of O2 and CO2.In this sensor, DMSO was used as the solvent, and a large Au working electrode, approximately 1.5 mm in diameter, was employed. The overall reaction scheme for this sensor is shown in Fig. 23, and several practical problems emerge.First, since the sensor design necessitated a gold macro-cathode where the reaction processes could be contained in the thin layer of solvent trapped between the membrane and the cathode surface, the sensor could only be used for gaseous analysis. It had no future for blood-gas analysis. Second, as is clear from the reaction scheme presented in Fig. 23, the very nature of the O2 .2–CO2 reaction pathways regenerates O2, which is further reduced during the first voltage pulse.This additional O2 elevates the observed overall O2 reduction signal. Therefore, a complex mathematical model136 had to be devised to deconvolute the true O2 and CO2 concentrations, and the analysis system became more and more complicated. Because of this complexity, and the impossibility of using this system for blood-gas analysis, the pulse-titration technique made no further progress. A third attempt was made, this time by Qian et al.,137 to devise an electrochemical method for the simultaneous measurement of CO2 and O2 and they combined two separate amperometric sensors.In this technique, a gaseous sample is drawn through an electrolytic cell designed to scrub the oxygen from the gas sample. This device consisted of a Pt-catalysed Teflon-bonded hydrophobic porous electrode, which had been previously developed for fuel-cell applications. The electrolyte was 0.5 m sulfuric acid. Qian et al.137 demonstrated that all the oxygen was virtually completely consumed by electro-reduction, and the remaining gas then passed to a CO2 sensor consisting of a Pt micro-disc of 60 mm diameter, with a DMSO solution.The gas-permeable membrane employed was either porous Teflon or solid polyethylene. Again, this sensor could Fig. 20 Possible reduction processes for the electrochemical reduction of CO2 in aprotic media with and without the addition of water. (Taken from ref. 128.) Fig. 21 Experimental linear sweep voltammogram for the mixed electrochemical reduction of O2 + CO2 in N2 under the same experimental conditions as in Fig. 19, showing only one reduction wave. Note the absence of the CO2 reduction wave and the second O2 reduction wave, both of which are present in Fig. 19. Analyst, June 1998, Vol. 123 79RCO4 •– C2O6 •– CO4 2– C2O6 2– O2 •– O2 + e– CO2 CO2 CO2 fast fast O2 •– O2 •– O2 O2 electrode solution only be employed for gaseous samples, and since both the O2 and CO2 sensors had different response times (compounded by the transport lag time between the sample gas reaching the two separate cells) the overall system had serious disadvantages.Furthermore, since the quoted response times were 15 and 35 s for 90% changes in O2 and CO2, respectively, the combined sensors were far too slow for the required response time (about 0.1 s) for breath-by-breath O2 and CO2 determination. As with the two previous design attempts, the sensor system could not be used for the determination of dissolved O2 or CO2 in blood.Reduction of O2 and CO2 at micro-electrodes The solution to the electrochemical conundrum of the crossinterference of the O2 and CO2 reduction processes, in nonaqueous media, finally came by returning to the original electrochemical roots of biology and medicine. By the very nature of their work, biological and clinical scientists use microelectrodes to investigate biological phenomena—as described in the early sections on aqueous amperometric sensors.Microelectrodes not only perturb (e.g., consume in this case) as little as possible the biological system in which they are measuring, but they also produce a minimal quantity of reaction byproducts —which are often unwanted in any electrochemical system, anyway. Hence it could be a sensible proposition to suppose that a micro-disc electrode in either an unshielded, or membrane-covered, system might produce amounts of O2 and CO2 reduction products which were too small to interfere significantly with the simultaneous reduction of these two gases in non-aqueous media.This proposition was tested in a series of studies138,139 (which have yet to be completed) which analysed the reduction of O2 and CO2 in the presence of each other, and in the presence of inhalational anaesthetic agents,140 at gold micro-cathodes in DMSO. These studies have been conducted with Au micro-disc electrodes of 2–80 mm diameter, with sweep voltammetry rates varying between 0.1 and 5.0 V s21.It has been demonstrated conclusively that it is most unwise to extrapolate electrochemical results obtained on macro-electrodes to micro-disc electrodes, so confirming the warning by Amatore and Sav�eant.130 Although these micro-disc electrodes replicated the macroelectrode studies, when O2 was reduced in the absence of CO2 and CO2 was reduced in the absence of O2, the results were entirely different when O2 and CO2 were reduced in the presence of each other.138,139 In this instance, in both the unshielded and membrane-covered cases, clearly separated O2 and CO2 reduction waves were obtained.Both produced good linear relationships between the limiting currents of the two waves and the oxygen and CO2 concentrations. Oxygen plus carbon dioxide voltammograms are illustrated in Fig. 24, showing the reduction waves in both the unshielded and membrane-covered sensor cases. The membrane-covered sensor was a conventional Clark-type sensor, as shown in Fig. 4, and the electrodes were gold wires sealed in glass rods, obtained either from commercial sources or from La Trobe University, Melbourne, Australia. They were ‘non-ideal’ for sensor design Fig. 22 Pulsing regime for the pulse-titration technique when (a) O2 only is present and (b) both O2 and CO2 are present. It can be seen in (b) that the recovery signal from the oxidation of O2 .2 is signifantly less in the presence of CO2 and that the O2 signal is enhanced in the presence of CO2, as explained in the text.Fig. 23 General reaction scheme for the homogeneous deactivation of O2 .2 by CO2 in DMSO, which occurs in the solvent layer adjacent to the macro-cathode in the pulse-titration sensor. Fig. 24 Voltammograms for the reduction of CO2 in O2, with the balance being N2, for a 10 mm diameter Au microdisc electrode. (i) Voltammograms for an unshielded electrode for (a) 3, (b) 6, (c) 9, (d) 12 and (e) 15% v/v CO2 in 25% v/v O2.(ii) Voltammograms for the gold micro-cathode when covered with a 12 mm PTFE membrane at CO2 concentrations of (a) 3, (b) 6 and (c) 9% v/v, when the O2 concentration was kept constant at 10% v/v. 80R Analyst, June 1998, Vol. 123since they had flattened surfaces and were not shaped for a Clark sensor, and yet they gave excellent results. This clearly indicates that specially designed sensors could be designed fairly easily. The key to the success of this dual O2–CO2 system obviously lies in the utilisation of micro-disc electrodes, and it seems reasonable to suppose that the smaller the microdisc the better.Recalling the reaction scheme of the O2–CO2 pulsed titration sensor (Fig. 23): CO2 + O2 .2 ? CO4 .2 (30a) CO4 .2 + CO2 ? C2O6 .2 (30b) C2O6 .2 + O2 .2 ? C2O6 22 + O2 (30c) the ‘feed-forward’ process [eqn. (30a)] destroyed the CO2 present in the DMSO layer adjacent to the gold surface, and the ‘feed-back’ process [eqn.(30c)] produced O2, which then elevated the original oxygen signal given by O2 + e2 " O2 .2 To eliminate the ‘feed-forward’ and ‘feed-back’ mechanisms, which destroy any direct relationship between current and gas concentration, it is necessary to prevent eqn. (30c) from occurring and to generate species such as C2O62 via direct electron transfer (heterogeneously at the electrode surface): C2O62 + e ? C2O6 22 (31) rather than homogeneously via eqn. (30c), so that a new process is available which is proportional to CO2 concentration.By using micro-disc electrodes, the product O2 .2 is produced in sufficiently small amounts that eqn. (30c) does not occur to any substantial extent, and by sweeping the polarising voltage there is also insufficient time under steady-state conditions prevailing at the micro-electrode for this reaction to take place homogeneously in the region of the electrode surface. Hence any O2 .2 reaction with CO2 produces very little O2 and this does not then elevate the oxygen reduction signal via the feed-back process in eqn.(30c). Micro-disc electrode studies, in the absence of CO2, have indicated that the reduction of CO2 provides either a drawn-out single process or two barely resolved processes. However, the limiting current obtained represented the summation of both processes and was well defined. The data can be explained by assuming that the short time-scale and low concentration of CO2 .2 lead to conditions where the reactions in Fig. 20: 2CO2 .2 ? CO + CO3 22 CO2 .2 + CO2 .2 ? C2O4 22 2CO2 .2 + H2O ? HCO22 + HCO32 are unimportant, i.e., the CO2 .2 has a finite existence at the electrode surface and can be directly reduced to a transient CO22 moiety: fast CO2 .2 + e ? CO2 .22 –––? products (32) It was clear that the second reduction process in the presence of O2 and CO2 mixtures was not simply a summation of the second O2 reduction process: O2 .2 + e ? O2 22 and the CO2 reduction processes.In the absence of eqn. (30c), direct reaction of the product of the reaction of CO2 and O2 .2 must occur to give a single multi-electron irreversible process. The exact details of this new process are not known, but direct electron transfer at the electrode surface occurs and O2 is not generated homogeneously as is the case at macro-electrodes. As a consequence, when O2 and CO2 gas mixtures are present in clinical concentrations, the limiting current for the first reduction process was found to be proportional to the O2 concentration and the limiting current for the second process was directly proportional to the CO2 concentration.Furthermore, when the membrane-covered sensor studies were repeated after the deliberate addition of 5 and 10% v/v of water to the DMSO solvent, the O2 and CO2 reduction results were essentially unchanged, with the only apparent change being that the magnitude of the CO2 reduction current was decreased on the addition of H2O to the solvent.138 Within the measurement error, the O2 reduction current was essentially the same when either 5 or 10% H2O was added to the DMSO.In both cases, the sensor responded linearly to changes in CO2 concentration, and the reduction processes appeared to occur at the same polarising voltages whether the solvent was wet or dry. It is thus clear that ‘unexpected’ results can be obtained with micro-disc electrodes, and further studies on this new system are required.So far, the reported studies have used 6 and 12 mm Teflon membranes to cover the sensor, and the gold micro-disc electrodes have varied, in practice, between 2 and 10 mm in diameter. The voltage sweep rates have varied between 0.1 and 5 V s21. As stated in the Aqueous electrolyte amperometric PO2 sensor section, the use of thin membranes produces conflicting advantages and disadvantages. A thin membrane will yield a fast time response (and hence a fast analysis update) but will inevitably increase the sensor liquid–gas difference by facilitating O2 transport from the liquid sample to the gas species consuming cathode.This was partly compensated for by employing micro-disc electrodes, but since they were polarised continuously the O2 consumption process was also continuous, and so a compromise had to be struck between electrode size and membrane permeability and thickness. However, in the case of the micro-disc O2–CO2 sensor described above, the liquid–gas current difference can be obviated by the employment of suitable electrochemical techniques.First, the minute micro-disc electrode will reduce the depletion of the gas species from the liquid sample, and, further, the polarising voltage is swept only for a finite time and can then be switched off. After a suitable quiescent time, the duty cycle can be repeated, allowing the overall diffusion system to relax during the ‘pause’ period.This sensor therefore lends itself to the type of ‘on-off’ computer simulation described in the sensor simulation model, where the pulsing regime is replaced by sweep voltammetry, and the characteristics of the membrane and solvent layers are woven into the general equations, together with the size of the micro-disc electrode employed. In this fashion, the actual operating characteristics of a sensor can be optimised by computer simulation to produce the minimum liquid–gas difference effect and yet maximise the analysis update time.Furthermore, the use of an array of micro-disc electrodes, wired in parallel and suitably spaced so that their diffusion-reaction patterns do not overlap each other, would greatly increase the total sensor output signal, and also build ‘redundancy’ into the sensor performance. O2 + N2O reduction The same micro-disc electrode sensor was tested in the presence of O2 and N2O binary gas mixtures.141 Studies were conducted with both shielded and membrane-covered micro-disc gold electrodes, and similar results were obtained in both instances.As with the aqueous electrolyte sensor, two distinct and clearly separated reduction waves were observed for the reduction of O2 and N2O in DMSO. The addition of 5 and 10% v/v of H2O to the DMSO did not appear to affect the reduction processes, although the overall voltage window wherein the reduction processes took place was reduced. Also, as with the aqueous electrolyte studies, N2 appeared to be the Analyst, June 1998, Vol. 123 81Rmain product of the reduction process, and care had to be taken (as with the aqueous solvents) to avoid the problem of N2 bubble formation on the micro-disc surface.141 Fig. 25 shows the voltammograms (sweep rate 0.1 V s21) obtained from a 10 mm diameter Au micro-disc electrode, when shielded with a 12 mm PTFE membrane for O2 and N2O gas mixtures, where both the O2 and N2O concentrations varied from 10 to 90% v/v.Although there does not appear to be an O2 reduction current in Fig. 25(a), it is in fact there as a ‘background current’ on the current scale displayed. When this section of the voltammogram is magnified, as shown in Fig. 25(b), the O2 voltammograms become clearly defined. The steps in the O2 current are due to the resolution of the potentiostat output, representing the last significant bit in the digital output. The N2O and O2 limiting currents from Fig. 25(a) and (b) translate into linear current–gas concentration relationships, and it is therefore clear that a membrane-covered Au micro-disc sensor can be designed to measure O2 and N2O in the presence of each other.The design characteristics of this sensor have yet to be optimised in order to facilitate liquid sample measurements of O2 and N2O. Reduction of O2, CO2 and N2O in the presence of each other The next complication in the quest for a multi-gas sensor is the determination of O2, CO2 and N2O in the presence of each other.Results, published only in patent form at the moment,140 have clearly revealed that three waves are seen on the current– voltage voltammogram when O2, CO2 and N2O are simultaneously reduced in the presence of each other. Fig. 26 illustrates this phenomena clearly, where it is seen that the O2, CO2 and N2O reaction processes are clearly separated and identified by their positions on the voltammogram. Furthermore, the limiting currents, within the gas species concentrations examined so far, have indicated that the three waves produce linear relationships between current and gas concentration for the individual gas species. Clearly, there is still much more work to be done on this new type of sensor, but the indications, so far, are that microdisc sensors, in aprotic media such as DMSO, clearly hold much promise for the future.The range of solvents examined so far has been severely limited, but a practical sensor might well employ a gel-type electrolyte layer, rather than a liquid layer.Inhalational anaesthetic agents It is here that a possible clear distinction appears, once more, between electrochemical results obtained with macro- and micro-electrodes. The first real attempt to devise an electrochemical sensor for an inhalational anaesthetic agent, other than halothane, was made by Compton and co-workers142,143 in 1988. The agent they chose to examine was isoflurane, a fluorinated ether (CHF2OCHClCF3), which is now commonly used in anaesthetic practice worldwide. In their first reports, they used a 0.682 cm diameter macro-electrode in a rotating-disc assembly to investigate the electrode activity, or otherwise, of isoflurane at a range of conventional electrode materials.142,143 Their results, using rotating-disc voltammetry, showed that isoflurane was inert towards reduction at silver, gold, mercury and platinum macro-electrodes, in both aqueous and non-aqueous solutions.The sole exception was on mercury in dimethylformamide solvent, where reduction was observed at around 23.0 V (with reference to a saturated calomel electrode), but this was so close to the potential of the solvent decomposition that any analytical applications were precluded. They therefore directed their attention towards finding a possible mediator for electron transfer. Their investigations demonstrated that the fluoranthene radical anion could mediate the reduction of isoflurane in acetonitrile solution and they then developed a polymer-modified electrode for the reduction of isoflurane based on these observations.In this particular case, the polymer [poly(11-vinylfluoranthene (PVF)] containing electroactive pendant groups was synthesised and deposited on a platinum disc electrode of diameter 0.702 cm. The solvent used was acetonitrile and cyclic voltammetry was employed to investigate the electrochemical processes.Cyclic voltammograms Fig. 25 (a) Voltammograms obtained using a 2 mm diameter Au microdisc electrode, when shielded with a 12 mm PTFE membrane for O2 + N2O gas mixtures. The O2 concentration varied from 10 to 80% v/v and the N2O concentration from 90 to 20% v/v. The O2 reduction current is so minute that it appears as a ‘background current’ on the current scale displayed here (full scale 80 nA). (b) The greatly amplified O2 voltammograms for the data displayed in (a), with the exception that the full-scale current is now 1.8 nA and the voltage scale is from 0 to 21.4 V versus Ag.The steps in the O2 current are due to the ‘last bit’ resolution of the potentiostat output. Fig. 26 Voltammograms for an unshielded 2 mm diameter gold micro-disc electrode in a gas mixture containing 10% v/v O2 and 5, 10, 15 and 20% CO2, with the balance being N2O. The oxygen signal is so low compared with the CO2 and N2O signals that it appears as a ‘baseline’ current.Clearly separated diffusion plateaux are seen for CO2 and N2O. Diffusion plateaux for oxygen could also be seen if the oxygen reduction current part of the voltammogram was magnified sufficiently (see Fig. 25). 82R Analyst, June 1998, Vol. 123obtained in the absence of isoflurane indicated that the reduction process corresponded to the addition of one electron to the fluoranthene pendant groups in the polymer, particularly since the charged passed corresponded quantitatively to the amount of PVF deposited on the surface.In the presence of isoflurane (in the absence of oxygen and nitrous oxide), typical ‘catalytic’ behaviour was apparent, with the reduction peak being considerably enhanced and the oxidation peak being significantly reduced. Although these studies showed promise, the further development of such a modified electrode into a practical sensor was severely restricted by the limited lifetime of the polymer coats, because desorption of the polymer occurred on prolonged potential cycling, and after about 40 cycles no recordable voltammogram remained.The other problem with this device was that both oxygen and nitrous oxide displayed electroactivity at the PVF-modified electrode, making it impossible to deconvolute the O2, N2O and isoflurane components of the sensor current. Compton and Northing144 then made one more attempt to devise an amperometric sensor for the detection of isoflurane and nitrous oxide.In this new work they devised a channelelectrode gas sensor. The conventional channel electrode consists of an electrode set in one wall of a rectangular duct through which electrolyte solution is pumped under laminar flow conditions. The design of Compton and Northing144 incorporated a gas-permeable membrane upstream of the electrode, and the substrate entered the cell from the gas phase through the membrane, and was then carried by a controlled (and well defined) flow to the electrode surface, where it was detected. Because the flow in the channel could be calculated, the signal observed from the working electrode was related to the concentration of the gas at the membrane–solution interface.This, in turn, gave a direct measure of the levels of the substrate bathing the exterior of the membrane in the gas phase. In order to eliminate all oxygen from the substrate, a second oxygen consuming electrode was located upstream of the membrane to ‘scrub’ oxygen from the system.Experiments were conducted to examine whether the reduction of anthraquinone (AQ) on the electrode upstream of the membrane could be used to eliminate oxygen interference on the downstream channel electrode. Their conclusion was that transport-controlled reduction of AQ to AQ22 on the upstream electrode (21.5 V versus Ag) did reduce any oxygen interference to acceptably low levels, with the following reaction taking place and ‘titrating’ any oxygen dissolved in the solution: AQ22 + 2O2 ? AQ + 2O22 These experiments were conducted successfully in the presence of up to 50% v/v oxygen.The next step was to determine whether nitrous oxide could be detected via a mediated reduction by the fluoranthene radical anion in the channel electrode flow cell. Their previous work had already indicated that isoflurane could be detected in this fashion, since it underwent a reduction via an ECA mechanism. Their new studies also showed that N2O underwent mediated reduction according to the following mechanism: electrode: F + e2 ? F.2 solution: F.2 + N2O ? F + products and that a ‘classical’ ECA mechanism operated.The time response of the sensor to changes in isoflurane and nitrous oxide was examined by pulsing the detector electrode between two potentials (21.5 and -1.9 V versus Ag) and allowing the system to reach a steady state at both potentials. A response time of 10 s or less was recorded, corresponding to the time required to establish new anaesthetic concentration profiles within the cell.By modifying the cell geometry to allow very fast flow rates, Compton and Northing expected the time response to improve significantly, with values approaching 1 s being expected. Their studies also indicated that the catalysis of the fluoranthene reduction of nitrous oxide and isoflurane was additive, and that the two mechanisms proceeded in parallel. Their conclusion was that it would therefore be possible to use the channel electrode sensor to analyse mixtures containing both nitrous oxide and isoflurane, provided that an oxygen scrubber was placed upstream of the channel electrode and that an independent means of measuring one of the two anaesthetic gases, for instance using a separate amperometric nitrous oxide sensor, was available.Thus, a complete measuring system would consist of an oxygen scrubber, a channel electrode sensor and an independent amperometric N2O detector.This complicated system illustrates how difficult it is to measure electrochemically two or more anaesthetic agents, simultaneously and separately, in the presence of oxygen. Although the evidence presented above suggests that a simple electrochemical measurement technique is well nigh impossible to devise, some hopes have been raised by the possible use of micro-electrodes and non-aqueous solvents. As stated above, the early work of Compton and colleagues indicated that (at least with macro-electrode surfaces) isoflurane demonstrated a complete lack of electroactivity with a wide range of electrode materials in both aqueous and non-aqueous solvents.Other work, published in patent form,140 has now indicated that not only isoflurane, but also other inhalational agents such as halothane, enflurane and sevoflurane, can be reduced on Au micro-electrodes in DMSO solution. The published work reported so far has been conducted with bare Au microelectrodes (with diameters varying between 2 and 10 mm) in DMSO in a reaction cell, with the anaesthetic agents being supplied by conventional anaesthetic vaporisers in clinical concentrations.This work has indicated that, using Au micro-disc electrodes, there is a window where isoflurane, halothane, enflurane and sevoflurane display diffusion-controlled plateaux on the voltammograms.140 Furthermore, the reduction waves for isoflurane, enflurane and sevoflurane are clearly distinguishable from that of oxygen.Also, similarly to the case of the production of O2 and N2O in DMSO with microelectrodes, the reduction current for 33% oxygen is between one and two orders of magnitude smaller than that of the anaesthetic agents at their usual clinical concentrations (0.6–1.0% v/v). This is illustrated in Fig. 27 for a mixture of 1% v/v enflurane in a 33% v/v oxygen–67% v/v nitrogen mixture. Fig. 28 illustrates a series of voltammograms for enflurane, halothane, isoflurane and sevoflurane, giving a comparison of their respective Fig. 27 A voltammogram for the reduction of 1% v/v enflurane in 33% O2–67% N2 for an unshielded 10 mm Au micro-disc electrode in DMSO solution. Analyst, June 1998, Vol. 123 83Rreduction potentials in DMSO solution. Although this work has yet to be reported in detail, it does illustrate that the use of micro-disc electrodes and aprotic solutions may lead to a whole new understanding of the electrochemistry of anaesthetic agents in clinical gas mixtures.Conclusion This review has attempted to chart a pathway through the undeniably difficult waters of the electrochemistry of clinical gases and vapours. Despite valiant efforts by many researchers, little real electrochemical progress seems to have been made over the past four decades, following the discovery and demonstration of the ever-popular Clark and Stow–Severinghaus O2 and CO2 sensors. It is only, perhaps, when both the electrochemist and the physiologist/clinician turn their eyes back to the roots of electrochemical research in biology and medicine, namely that of the micro-disc electrode, that new progress will be made in this field.As far as this reviewer is concerned, the macro-disc electrode has had its day, perhaps even in the field of gaseous analysis. Although the supply of analyte molecules is essentially ‘infinite’ in a gas sample, and it is therefore tempting to manufacture macro-electrode sensors, the clear indications are that micro-electrodes allow individual gas species in gas mixtures to be separated electrochemically on the same surface, and thus analysed simultaneously and separately.It is beginning to appear that this process might apply not only to simple gas species such as O2 and CO2 but also to the anaesthetic gases and vapours. Only further research and development in this area will prove whether this hypothesis is short- or long-lived.References 1 Astrup, P., and Severinghaus, J. W., The History of Blood Gases, Acids and Bases, Munksgaard, Copenhagen, 1st edn., 1986. 2 Severinghaus, J. W., and Astrup, P., J. Clin. Monit., 1985, 1, 180. 3 Severinghaus, J. W., and Astrup, P., J. Clin. Monit., 1985, 1, 259. 4 Severinghaus, J. W., and Astrup, P., J. Clin. Monit., 1986, 2, 60. 5 Severinghaus, J. W., and Astrup, P., J. Clin. Monit., 1986, 2, 125. 6 Severinghaus, J. W., and Astrup, P., J.Clin. Monit., 1986, 2, 174. 7 Nunn, J. N., Nunn’s Applied Respiratory Physiology, Butterworth Heinemann, London, 4th edn., 1993. 8 Sykes, M. K., Principles and Practice of Respiratory Support, BMJ Publishing Group, London, 1st edn., 1995. 9 Gilbert, H. C., and Vender, J. S., J. Clin. Monit., 1996, 12, 179. 10 Hoffer, J. L., and Norfleet, E. A., J. Clin. Monit., 1996, 12, 183. 11 Market Opportunities for Advanced Sensors in Healthcare, Report to the Department of Trade and Industry, C3326-FR-002a, Cambridge Consultants, Cambridge, 1989, p. 18 and Table 1. 12 Hutton, P., and Prys-Roberts, C., Monitoring in Anaesthesia and Intensive Care, Saunders, Philadelphia, 1st edn., 1994. 13 Hitchman, M. L., Measurement of Dissolved Oxygen, Wiley, Chichester, 1978. 14 Hahn, C. E. W., J. Phys., 1980, 13, 470. 15 Hahn, C. E. W., J. Phys. E, 1981, 14, 783. 16 Sykes, M. K., J. Clin. Monit., 1987, 3, 116. 17 Stow, R. W., Baer, R. F., and Randall, B., Arch. Phys. Med. Rehabil., 1957, 38, 646. 18 Gertz, K. H., and Loeschcke, H. H., Naturwissenschaften, 1958, 148, 115. 19 Severinghaus, J. W., and Bradley, A. F., J. Appl. Physiol., 1958, 13, 515. 20 Parker, D., Delpy, D., and Lewis, M., Med. Biol. Eng. Comput., 1978, 16, 599. 21 Daneel, H. L., Z. Elektrochem., 1897/8, 4, 227. 22 Blinks, L. R., and Skow, R. K., Proc. Natl. Acad. Sci. USA, 1938, 24, 420. 23 Clark, L. C., Trans. Am. Soc. Artif. Intern. Organs, 1956, 2, 41. 24 Fatt, I., Polarographic Oxygen Sensor: Its Theory of Operation and Its Application in Biology, Medicine and Technology, CRC Press, Cleveland, OH, 1976. 25 Kreuzer, F., Kimmich, H.P., and Brezina, M., in Medical and Biological Applications of Electrochemical Devices, Wiley, Chichester, 1980, pp. 173–261. 26 Polarographic Oxygen Sensors: Aquatic and Physiological Applications, ed. Gnaiger, E., and Forstner, H., Springer, New York, 1983. 27 Linek, V., Vacek, V., Snikule, J., and Benes, P., Measurement of Oxygen by Membrane-Covered Probes: Guidelines for Applications in Chemical and Biochemical Engineering, Ellis Horwood, Chichester, 1988. 28 Adams, A. P., and Hahn C. E. W., Principles and Practice of Blood- Gas Analysis, Churchill Livingstone, London, 2nd edn., 1982. 29 Jacq, J., and Bloch, O., Electrochim. Acta, 1964, 9, 551. 30 Wroblowa, H. S., Yen, C.-P., and Razumney, G., J. Electroanal. Chem., 1976, 69, 195. 31 Appleby, A. J., and Savy, M., J. Electroanal. Chem., 1978, 92, 15. 32 Hahn, C.E. W., Davis, A., and Albery, W. J., Respir. Physiol., 1975, 25, 109. 33 Staub, N. C., J. Appl. Physiol., 1961, 16, 192. 34 Hansen, J. E., and Fiel, M. C., Chest, 1988, 94, 49. 35 Hahn, C. E. W., DPhil Thesis, University of Oxford, 1974. 36 Ultman, J. S., Firouztal�e, E., and Skerpon, M. J., J. Electroanal. Chem., 1981, 127, 59. 37 Firouztal�e, E., Skerpon, M. J., and Ultman, J. S., J. Electroanal. Chem., 1982, 134, 1. 38 Jenson, O. J., Jacobson, T., and Thomas, K., J.Electroanal. Chem., 1978, 87, 203. 39 Mancy, K. H., Okun, D. A., and Reilley, C. N., J. Electroanal. Chem., 1962, 4, 65. 40 Myland, J. C., and Oldham, K. B., J. Electrochem. Soc.,1984, 131, 1815. 41 Sutton, L., Gavaghan, D. J., and Hahn, C. E. W., J. Electroanal. Chem., 1996, 408, 21. 42 Gavaghan, D. J., Rollett, J. S., and Hahn, C. E. W., J. Electroanal. Chem., 1992, 325, 23. 43 Gavaghan, D. J., Rollett, J. S., and Hahn, C. E. W., J. Electroanal. Chem., 1993, 348, 1. 44 Gavaghan, D.J., Rollett, J. S., and Hahn, C. E. W., J. Electroanal. Chem., 1993, 348, 15. 45 Gavaghan, D. J., and Rollett, J. S., J. Electroanal. Chem., 1990, 295, 1. 46 Parker, D., Key, A., and Davies, R., Lancet, 1971, i, 952. 47 Parker, D., Key, A., Davies, R. Scopes, J. W., and Marcovitch, H., Biomed. Eng., 1971, 6, 313. 48 Rolfe, P., in IEE Medical Electronics Monographs, ed. Hill, D. W., and Watson, B. W., Peregrinus Press, London, 1976, vol. 18–22, pp. 126–158. 49 Mindt, W., Proc.Jahrestagung D. G. Biomed. Technol. Erlangen, 1973, 29. 50 Kreuzer, F., and Nessler, C. G., Science, 1958, 128, 1005. 51 Krog, J., and Johanson, K., Rev. Sci. Instrum., 1959, 30, 108. Fig. 28 Voltammograms for the reduction of enflurane, halothane, isoflurane and sevoflurane at an unshielded 12 mm Au micro-disc electrode, all on the same voltage scale, giving a comparison of their respective reduction potentials. (a) 1% v/v sevoflurane; (b) 1% v/v enflurane; (c) 0.6% v/v halothane; and (d) 0.6% v/v isoflurane. 84R Analyst, June 1998, Vol. 12352 Charlton, G., Read, D., and Reed, J., J. Appl. Physiol., 1963, 18, 1247. 53 Eberhard, P., Fehlmann, W., and Mindt, W., Biotelemetry Patient Monit., 1979, 6, 16. 54 Zhang, S., Wright, G., Kingston, M. A., and Rolfe, P., Med. Biol. Eng. Comput., 1996, 34, 313. 55 Zhang, S., Benmakroha, Y., and Rolfe, P., Biosens. Bioelectron., 1996, 11, 1019. 56 Chapman, D., Biochem. Soc. Trans., 1993, 21, 258. 57 Benmakroha, Y., Zhang, S., and Rolfe, P., Med. Biol. Eng. Comput., 1995, 33, 811. 58 Gerlach, V., Arch. Anat. Physiol. Lp3, 1851, 431. 59 Baumberger, J. P., and Goodfriend, R. B., Fed. Proc. Fed. Am. Soc. Exp. Biol., 1951, 10, 10. 60 Rooth, G., Sjosted, S., and Caligara, F., Sci. Tools, 1957, 4, 37. 61 Huch, A., Huch, R., and Lubbers, D. W., Arch. Gynaekol., 1969, 207, 443. 62 Huch, A., and Huch, R., Birth Defects, 1979, 15, 5. 63 Eberhart, P., Hammacher, K., and Mindt, W., Med. Technol. (Stuttgart), 1972, 26. 64 Berran, A. V., Huxtable, R. F., and Sperling, D. R., J. Appl. Physiol., 1976, 41, 442. 65 Whitehead, M. D., Halsall, D., Pollitzer, M. J., Delpy, D., Parker, D., and Reynolds, E. O. R., Paediatr. Res., 1980, 14, 1426. 66 Severinghaus, J. W., Weiskopf, R. B., Nishimura, M., and Bradley, A. F., J. Appl. Physiol., 1971, 31, 640. 67 Bates, M. L, Feingold, A., and Gold, M. I., Am. J. Clin. Pathol., 1975, 64, 448. 68 Dent, J. G., and Netter, K. J., Br. J. Anaesth., 1976, 48, 195. 69 Stosseck, K., Anaesthetist, 1977, 26, 1977. 70 Albery, W. J., Brooks, W. N., Gibson, S. P., and Hahn, C. E. W., J. Appl. Physiol., 1978, 45, 637. 71 Evans, M. C., and Cameron, I. R., Lancet, 1978, ii, 1371. 72 Douglas, I. H. S., McKenzie, P. J., McA Ledingham, I., and Smith, G., Lancet, 1978, ii, 1370. 73 Brown, B. R., Anesthesiology, 1979, 30, 146. 74 Norden, A. G. W., and Flynn, F. V., Clin. Chim. Acta, 1979, 99, 229. 75 Eletskii, V. V., Rotenberg, Z. A., and Pleskov, Y. V., Sov. Electrochem., 1969, 5, 428. 76 Stegan, von der J. H. G., Visscher, W., and Hoogland, J. G., Electrochem. Technol., 1966, 4, 564. 77 Dutta, D., and Landolt, D., J. Electrochem. Soc., 1972, 119, 1320. 78 Gadde, R., and Burchenstein, S., J. Electroanal. Chem., 1974, 50, 163. 79 Zakorski, S. P., and Suwalski, J. P., J. Electroanal. Chem., 1973, 46, 353. 80 Johnson, K. E., and Sawyer, D. T., J. Electroanal. Chem., 1974, 49, 95. 81 Gutmann, V., Gritzner G., and Schr�ober, G., Mikrochim. Acta, 1964, 193. 82 Eberhard, P., and Mindt, W., paper presented at the First International Symposium, Marburg, 1978. 83 Sugioka, K. Cattermole, R. W., and Sebel, P.S ., Br. J. Anaesth., 1987, 59, 1548. 84 Armstrong, R. F., Hutchinson, J. M., Lincoln, C., Ingram, D., and Soutter, L., Br. J. Anaesth., 1976, 48, 1005. 85 Albery, W. J., Brooks, W. N., Gibson, S. P., Heslop, M. W., and Hahn, C. E. W., Electrochim. Acta, 1979, 24, 108. 86 Hahn, C. E. W., Brooks, W. N., Albery, W. J., and Rolfe, P., Anaesthesia, 1979, 34, 263. 87 Brooks, W. N., Hahn, C. E. W., Foëx, P., Maynard, P., and Albery, W. J., Br. J. Anaesth., 1980, 52, 715. 88 Hahn, C. E. W., Hall, E. A. H, Maynard, P., and Albery, W. J., Br. J. Anaesth., 1982, 54, 681. 89 Bergman, I., Br. Pat., 1 200 595, 1966. 90 Bergman, I., Nature (London), 1968, 218, 266. 91 Bergman, I., and Windle, D. A., Ann. Occup. Hyg., 1972, 15, 329. 92 Ebert, H., Parsons, R., Ritzoulis, G., and VanderNoot, T., J. Electroanal. Chem., 1989, 264, 181. 93 Ritzoulis, G., J. Electroanal. Chem., 1992, 327, 209. 94 Ahmadi, A., Bracey, E., Evans, R. W., and Attard, G., J. Electroanal. Chem., 1993, 350, 297. 95 Attard, G. A., and Ahmadi, A., J. Electroanal. Chem., 1995, 389, 175. 96 Kudo, A., and Mine, A., J. Electroanal. Chem., 1996, 408, 267. 97 Dobkin, A. B., Development of New Volatile Inhalation Anaesthetics, Elsevier, Amsterdam, 1979, 51. 98 Severinghaus, J. W., Weiskopf, R. B., Nishimura, M., and Bradley, A. F., J. Appl. Physiol., 1971, 31, 640. 99 Bates, M. L., Feingold, A., and Gold, M. I., J. Clin. Pathol., 1975, 64, 448. 100 Dent, J. G., and Netter, K. J., Br. J. Anaesth., 1976, 48, 195. 101 Smirnov, P. N., Izd. Kazansk. Univ., Kazan’, 1962, 120. 102 Smirnov, P. N., Sov. Invent. Illustr., Gen. Org., 1963, 6. 103 Gol'din, M. M., Feoktistov, L. G., Tomilov, A. P., and Smirnov, K. M., J. Gen. Chem. USSR, 1972, 42, 2552. 104 Albery, W. J., Hahn, C. E. W., and Brooks, W. N., Br. J. Anaesth., 1981, 53, 447. 105 Hall, E. A. H., Conhill, E. J., and Hahn, C. E. W., J. Biomed. Eng., 1988, 10, 319. 106 Tebbutt, P., and Hahn, C. E. W., J. Electroanal. Chem., 1989, 261, 205. 107 Mount, A. R., Appleton, M. S and Albery, W. J., J. Electroanal. Chem., 1992, 334, 155. 108 Mount, A. R., Appleton, M. S., Albery, W. J., and Hahn, C. E. W., J. Electroanal. Chem., 1992, 340, 287. 109 Mount, A. R., and Clark, D., J. Electroanal. Chem., 1994, 367, 175. 110 Langmaier, J., and Samec, Z., J. Electroanal. Chem., 1996, 402, 107. 111 Caruana, D. J., and Giglio, J., J. Chem. Soc., Faraday Trans., 1996, 92, 3669. 112 Elliott, R. H., and Strunin, L., Br. J. Anaesth., 1993, 70, 339. 113 Blogg, C. E., Br. Med. J., 1986, 292, 1691. 114 O’Connor, Today’s Anaesth., 1996, 11, 19. 115 Peover, M. E., and White, B. S., Chem. Commun., 1965, 183. 116 Maricle, D. J., and Hodgson, W. G., Abstracts, 149th Meeting of the American Chemical Society, Division of Analytical Chemistry, Detroit, 1965, p. 9B. 117 Sawyer, D. T., and Roberts, J. L., J. Electroanal. Chem., 1966, 12, 90. 118 Bauer, D., and Beck, J.-P., J. Electroanal. Chem., 1972, 40, 233. 119 Fujinaga, T., Izutsu, K., and Adachi, T., Bull. Chem. Soc. Jpn., 1969, 42, 140. 120 Sawyer, D. T., and Gibian, M.-J., Tetrahedron, 1979, 35, 1471. 121 Sawyer, D. T., and Valentine, J. S., Acc. Chem. Res., 1981, 14, 393. 122 Tsushima, M., Tokuda, K., and Ohsaka, T., Anal. Chem., 1994, 66, 4551. 123 Vasudevan, D., and Wendt, H., J. Electroanal. Chem., 1995, 192, 69. 124 Dehn, H., Gutmann, V., Kirch, H., and Schöber, G.tsh. Chem., 1962, 93, 1348. 125 Roberts, J. L., and Sawyer, D. T., J. Electroanal. Chem., 1965, 9, 1. 126 Haynes, L. V., and Sawyer, D. T., Anal. Chem., 1967, 39, 332. 127 Lamy, E., Nadjo, L., and Sav�eant, J. M., J. Electroanal. Chem., 1977, 78, 403. 128 Kaiser, Von U., and Heitz, E., Reaktionslenk. Elektrochem., 1977, 77, 818. 129 Gressin, J. C., Michelet, D., Nadjo, L., and Sav�eant, J. M., Nouv. J. Chim., 1979, 3, 545. 130 Amatore, C., and Sav�eant, J. M., J. Am. Chem. Soc., 1981, 103, 5021. 131 Eggins, B. R., and McNeill, J., J. Electroanal. Chem., 1983, 148, 17. 132 Coombs, A. M., DPhil Thesis, Oxford University, 1985. 133 Clark, D., DPhil Thesis, Oxford University, 1987. 134 Albery, W. J., and Barron, P., J. Electroanal. Chem., 1982, 138, 79. 135 Albery, W. J., Clark, D., Drummond, H. J. J., Coombs, A. J. M, Young, W. K., and Hahn, C. E. W., J. Electroanal. Chem., 1992, 340, 99. 136 Albery, W. J., Clark, D., Young, W. K., and Hahn, C. E. W., J. Electroanal. Chem., 1992, 340, 111. 137 Qian, F., Lu, J., Zhou, Z., and Cha, C., Sens. Actuators B, 1993, 17, 77. 138 Hahn, C. E. W., McPeak, H., Bond, A. M. S., and Clark, D., J. Electroanal. Chem., 1995, 393, 61. 139 Hahn, C. E. W., McPeak, H., and Bond, A. M., J. Electroanal. Chem., 1995, 393, 69. Analyst, June 1998, Vol. 123 85R140 Hahn, C. E. W., and Clark, D., Eur. Pat., 704054B, 1997; US Pat., 5720870, 1998. 141 McPeak, H., and Hahn, C. E. W., J. Electroanal. Chem., 1997, 427, 179. 142 Compton, R. G., Northing, R. J., Waller, A. M., Fleet, G. W. J., and Bashyal, B. P., J. Electroanal. Chem., 1988, 244, 203. 143 Compton, R. G., Northing, R. J., Fleet, G. W., Son, J. C., and Bashyal, B. P., Clin, Phys. Physiol. Meas., 1988, 9, 133. 144 Compton, R. G., and Northing, R. J., J. Chem. Soc., Faraday Trans., 1990, 86, 1077. Paper 7/08951A Accepted December 12, 1997 86R Analyst, June 1998, Vo

ISSN:0003-2654    

DOI:10.1039/a708951a

出版商:RSC    

年代:1998

数据来源: RSC

摘要:

Partial least-squares calibration of two-way diode-array high-performance liquid chromatograms: influence of calibration design, noise and peak separation Konstantinos D. Zissisa, Richard G. Brereton*a and Richard Escottb a School of Chemistry, University of Bristol, Cantock’s Close, Bristol, UK BS8 1TS b SmithKline Beecham Pharmaceuticals, Old Powder Mills, Near Leigh, Tonbridge, Kent, UK TN11 9AN An approach for the calibration of two-way diode-array high-performance liquid chromatograms is described, involving unfolding a three-way data matrix and performing partial least-squares (PLS) calibration.The properties of loadings summed over time and wavelength are discussed. The influence of calibration design, noise levels and peak separation are investigated, using pseudosimulations, both by calculating prediction and test errors and by graphical representation of the summed loadings. The importance of using an independent test set is emphasized. Calibration design is shown to have a major effect both on the appearance of the loadings and on the PLS errors.Keywords: High-performance liquid chromatography; partial least squares; calibration; chemometrics Partial least squares (PLS) is commonly employed for the quantification of components in mixtures. In chromatography, this method is an important alternative to univariate approaches such as the vertical divisor and triangulation. It can be particularly crucial, for example, when a small peak is buried within a large one.However, coupled chromatograms are multivariate in nature, and each chromatogram could be represented by a matrix with the columns representing different wavelengths and the rows different points in time. Unlike spectroscopy, a single vector of univariate parameters, such as a set of concentrations, is calibrated to a tensor (or ‘box’) consisting of absorbances as a function of both elution time and wavelength for the corresponding mixture chromatograms. There are a number of methods1–4 for overcoming this, one of which involves unfolding the data matrix to a two-way matrix for normal PLS calibration, as described in this paper.Careful scaling and centring of the data are required for this procedure to be successful. This paper describes one such approach. The method proposed below is based on an approach first used for the calibration of GC-MS data.4 In order to illustrate the method we use pseudosimulations, in which the datasets are closely based on real data.Two-way chromatograms are obtained in which real spectra of two closely eluting compounds of pharmaceutical interest are used. Noise distributions and peak separations relate to those experimentally obtained, but the approach in this paper allows us to change these parameters systematically and examine the influence on PLS predictions. The paper also demonstrates that absolute quantification of co-eluting components can be achieved with careful selection of calibration designs.Methods PLS calibration PLS calibration is one of the best known regression techniques for multivariate data analysis.5–8 The main advantage over other similar multivariate approaches, such as principal component regression (PCR), is that it takes into consideration errors that are likely to occur in both the main ‘X’ data (often a matrix of absorbance values at successive time units and various wavelengths) and the ‘y’ data (often a concentration vector for one of the compounds present in a mixture).9,10 Over the past 10 years, numerous PLS algorithms have been developed and present a great challenge, as they can be applied to various different applications.11–18 Calibration can be performed on either univariate,19 twoway11 or three-way data,20,21 as illustrated in Fig. 1. An example of univariate calibration involves simply varying the concentration of a compound, y, and monitoring its absorbance at a single wavelength.From this, a linear model of absorbance versus compound concentration can be constructed. In two-way PLS calibration, many different applications are encountered.17,18 One such example of applications in HPLC is calibrating the sum of the area of a chromatographic peak at J wavelengths, aj, Fig. 1 Schematic representations of (a) univariate calibration, (b) two-way PLS calibration and (c) three-way PLS calibration. Analyst, June 1998, Vol. 123 (1165–1173) 1165or the elution profile of a chromatographic peak, summed over all wavelengths, bi to compound concentration, y, where a x x j ij i I I i ij j J J = = = = Â Â 1 2 1 2 and b In these cases, the X data matrices would have dimensions M 3 J and M 3 I, respectively, where M is the number of samples, I the number of points in time and J the number of wavelengths. Three-way PLS calibration is a more elaborate technique, which is based on a tensor Z, with dimensions M 3 I 3 J.There are several approaches to this, one of which involves unfolding the tensor to a long two-way matrix, as described in the next section. In this paper, we will be exclusively concerned with this form of three-way PLS calibration, where y is a univariate concentration block. The PLS decomposition most often used in calibration is called PLS1 and is applied to each compound separately.22 For a typical two-way PLS calibration, PLS1 decomposes the X matrix and y vector as follows: X = TPA + E and y = uqA + f where T and u are the scores of matrix X and vector y, P and q their associate loadings and E and f the residual matrices.Before starting any operation, both the X matrix and the y vector are normally mean centred. Then, PLS1 calculates the loadings weights, w, the scores, t, and loadings, p, for the first PLS component and the value of a contribution to the predicted concentration vector, nn, for component n. New values of X and y can then be estimated by subtracting the contribution of the first PLS component to the X matrix, tpA, from the X matrix, and nn from y. The algorithm can then be repeated for further PLS components, so that the m predicted values, �y, for N PLS components are given by � , , y y m N m n n N = + = Â n 1 where �ym,N is the predicted concentration for sample m after N PLS components have been extracted and �y is the mean concentration of the compound over the samples.PLS2 is an extension to PLS1, and its main difference is that several y vectors can be taken into consideration in the calculation.In the work presented in this paper, the PLS1 algorithm developed by Wold et al.11 was used exclusively. Unfolding In three-way PLS calibration, M data matrices of (I 3 J) dimensions give rise to a tensor, Z, of M 3 I 3 J dimensions. Before performing PLS1, it is usual to unfold this 3-D, Z, tensor into a 2-D matrix.2,23,24 To achieve this, the rows of Z are concatenated to give a row vector. After unfolding has taken place, the 2-D X matrix would have dimensions M 3 (I.J), where M is the number of samples, I the number of points in time and J the number of wavelengths.A schematic representation of this procedure is shown in Fig. 2. The scores, t, and loadings, p, of the resultant X matrix will have dimensions M and I.J, respectively. Time dependent and wavelength dependent loadings PLS calibration is performed for each compound separately.In three-way PLS1, most of the information about a particular set of data is hidden in the scores and loadings of the various PLS components. Of particular interest is the information located in the loadings. For example, if X contains information about a mixture of two compounds, summing the loadings over time would result in a new (I 3 N) matrix, time I,N P, according to the equation time p p i n ij n j J , , = = å1 where the PLS components are numbered 1. ..n. . .N. The information given out by this (I.N) matrix would correspond to the elution profile of the compound PLS is performed on. In contrast, summing the loadings, I.J,NP over wavelengths would result in a new (J 3 N) matrix, l J,NP, according to the equation l p p j n ij n i I , , = = Â1 By plotting this new matrix versus wavelength, the spectrum of the compound on which PLS is performed ated. A schematic representation is given in Fig. 3. Compounds The two compounds whose spectra were used in this study were SKF-101468-A (ropinirole) (I) and its synthetically associated impurity, SKF-96266-A (II).The compounds were synthesised in-house, at SmithKline Beecham (Tonbridge, Kent, UK)25 and their structures are shown in Fig. 4. The normalised experimental spectrum, n 1,J�sk, of each compound was used, as depicted in Fig. 5. The spectra were obtained from the chromatographic analysis of the pure compounds, explained in detail in a previous paper.26 In total, the number of wavelengths used was 31, ranging from 230 to 290 nm in 2 nm increments.Fig. 2 Schematic representation of unfolding a three-way Ztensor into a two-way X matrix. 1166 Analyst, June 1998, Vol. 123NH O N NH O N O SKF-101468-A (I) SKF-96266-A (II) Simulations To generate elution profiles for the two compounds, symmetric simulations were performed, based on the basic equation for Gaussian peaks: c A i t i k k k k , ( ) = - - é ë ê ù û ú exp 2 2 s where Ak is an absorbance value at the point of maximum intensity, i is the number of the data point in time, sk is a factor relating to the width of the peak at half its height and tk is the retention time at the maximum of the peak.In simulating symmetric elution profiles, Ak was given a value of 1 for both compounds, whereas sk had a value of 6 for the two compounds. For the initial simulations, t1 was set at 14 points in time and t2 at 26 (separation of 12 points in time).The total number of points in time was 46, with a digital resolution of 1 s. The simulated elution profile for each peak, given by I,1�ck, was then multiplied by the true normalised experimental spectrum of each compound, n I.J�sk to generate data matrices I,J �X 1 and �X 2 respectively, based on the following equation: I,J �X k = I,1�ck 1,J n�sk Experimental design A total of 25 simulated mixtures were used. The calibration designs were based on five levels, which were coded between 22 and +2 for each compound present in the mixture, in increments of 1.The levels relate directly to the concentrations of compounds, according to the following equation: y y l l k k , max, = + 3 5 where yl,k is the concentration of compound k at a coded level l, and ymax,k is the maximum true chromatographic concentration of compound k. The two ymax,k values were set at 0.6 and 0.4103 mm for compounds I and II, respectively, so that the two peaks, in the summed elution profiles over 230–290 nm (2 nm increments), at the same coded level will have identical heights.The values of concentrations at the different levels are given in Table 1. The various designs can be represented by two vectors d1 and d2. The designs were selected so that a range of correlation coefficients, r12, between d1 and d2 at values from 0 to 1 were employed. To generate a design matrix with any desired correlation coefficient, a first level, a permuter and a difference vector have to be carefully selected (Table 2).The construction of multi-level, multi-factor calibration designs is described in detail elsewhere.27 A typical 25-experiment five-level design for two compounds is shown in Table 3, each column representing vectors d1 and d2, respectively. This design has a value of r12 = 0, so the two concentration vectors are orthogonal28,29 to one another, which implies that the predictions are even throughout the mixture space.When r12 = 1, the two concentration vectors are confounded, so that it is impossible to distinguish the effects of the concentration of compound one increasing and of the concentration of compound two increasing and vice versa. For any given value of r12, the design consists of 25 chromatograms, each of two closely eluting peaks in different Fig. 3 Schematic representation of converting the loadings matrixI,J,N P into a matrix of summed loadings over time, time I,N P and a matrix of summed loadings over wavelength, J,N lP.Fig. 4 Structures of the compounds whose spectra were used in this study. Fig. 5 Normalised experimental spectra of compounds SKF-101468-A (I) and SKF-96266-A (II). Table 1 Values of concentrations for compounds I and II at the five coded levels Coded level, l yl,1/mm yl,2/mm 22 0.1200 0.0820 21 0.2400 0.1641 0 0.3200 0.2461 1 0.4800 0.3282 2 0.6000 0.4103 Table 2 Correlation coefficients, first levels, permuters and difference vectors used in the calibration designs Correlation First level of Difference coefficient design Permuter vector 0.0 0, 0 22, 21, 2, 1, 22 0, 2, 3, 1 0.2 0, 0 2, 21, 2, 1, 22 0, 1, 3, 2 0.4 0, 0 22, 21, 2, 1, 22 0, 1, 2, 3 0.5 22, 22 21, 0, 2, 1, 21 0, 1, 3, 2 0.6 0, 0 22,21, 2, 1, 22 0, 3, 1, 2 0.8 0, 0 2, 1, 21, 2,22 0, 2, 3, 1 1.0 0, 0 22, 21, 2, 1, 22 0, 2, 3, 1 Analyst, June 1998, Vol. 123 1167proportions. For each PLS1 calculation, the y vectors for the two compounds consist of 25 concentrations, given by vectors M,1y1 and M,1y2, and derived from d1 and d2, as described above.The 25 two-compound X matrices arise from multiplying each I,J �X 1 and I,J �X 2 matrices by the values of y1 and y2, adding them up and also adding noise to them. This gives rise to 3-D tensor Z, according to the following equation: m,I,J Z = M,1y1 # I,J �X 1 + M,1y2 # I,J �X 2 + M,I,J N The noise tensor, M,I,J N, generated was based on a Gaussian function with a mean of zero and a standard deviation relating to the true chromatographic noise.The seed was nonreproducible, so that the noise profile was different for each chromatogram. Simulation Parameters The influence of the following parameters on the PLS predictions was investigated: (a) correlation coefficient of design (b) noise and (c) relative peak positions (chromatographic resolution). The range of values to which the parameters were set is given in Table 4. A reference chromatogram was chosen with values of 0.5 for the correlation coefficient of the design, 3 3 1024 AU for the standard deviation of the noise and 12 s for the separation of the two peaks.The standard deviation of the noise used was equivalent to the noise typically encountered in a Beckman System Gold chromatograph (Model 126 pump, Model 507 autosampler), although this could be influenced by a number of factors (equilibrating the system, proper maintenance). A peak separation of 12 s (t1 = 14, t2 = 26) was very close to that found experimentally for the two compounds,26 whereas for the calibration design, one with a value of r12 at 0.5 was thought appropriate.Generation of test sets and assessment of PLS predictions Autopredictions PLS predictions (autopredictions) were calculated for various calibration training sets. For each design, 25 different predictions were obtained for n PLS components, based on a root mean square error (RMSE) which was calculated according to the following equation: RMSE = - = =å ( � ) , , , y y l k m n k m m m 2 1 25 25 where ylm,k is the concentration for sample m of compound k which is at level l and �ym,n,k is the predicted concentration for sample m and compound k, using n PLS components.RMSE values (in mm) were calculated for both compounds I and II and for one, two and three PLS components.After two PLS components were extracted, the RMSEs were found to give very low values (of the order of 1026 mm).Hence they are not reported in any table, as they were not deemed important. Test sets To see how well the various calibration training sets predict the concentrations of the two compounds, independent test sets were generated. All other training sets were then used to try and predict the concentrations of the two compounds in the test sets, and the quality of the predictions was contrasted with that of the autopredictions. When testing to see how well the calibration models work, the following observations were taken into consideration: (i) The 3-D tensor M,I,J test Zgeneratedfor the test set was unfolded on to a 2-D matrix M,I,J test X.This matrix was used exclusively throughout testing of all training sets. The corresponding 2-D matrix for each calibration training set training M,I,J X (mean centred along M to give a one cocted from it, according to the equation test corrected M,I.J X = test M,I.J X 2 training 1,I.J �x and test corrected M,I.J X was used as the ‘X’ data block during testing. (ii) For the PLS predictions, the values of p and w estimated for each calibration training set were used and the same set of concentrations, testylm,k (based on the calibration design for the particular test set), were predicted by each training set.To test each training set, a value of RMSEP was calculated, according to the following equation: RMSEP test test = - ( ) = =å y y l k m n k m m m, , , � 2 1 25 25 where test�ym,n,k is the predicted concentration for sample m of the test set, and compound k, using n PLS components for each training set.In total, 10 test sets were generated, as listed in Table 5. Testing to see how well the models work is very crucial, as a Table 3 A typical 25-experiment, five-level design matrix for two compounds (vectors d1 and d2) 0 0 0 22 22 22 22 2 2 21 21 2 2 0 0 21 21 21 21 1 1 2 2 1 1 0 0 2 2 2 2 22 22 1 1 22 22 0 0 1 1 1 1 21 21 22 22 21 21 0 Table 4 Values of parameters in the simulations whose effect in PLS predictions was investigated (values in bold are those of the reference chromatogram) Correlation coefficient of Standard deviation calibration design, r12 Peak separation/s of noise (AU) 0.0 0 3 3 1026 0.2 4 3 3 1025 0.4 8 3 3 1024 0.5 12 3 3 1023 0.6 16 3 3 1022 0.8 1.0 1168 Analyst, June 1998, Vol. 123model might predict itself with a sufficiently low error even using cross-validation, but when it is used to predict the concentrations of other unknown compounds the error might be substantial.Results Changing correlation coefficient In total, seven different calibration training sets were generated with values of r12 between 0.0 and 1.0, a standard deviation of noise of 3 310-4 AU and a peak separation of 12 s. The RMSEs of the autopredictions for one PLS component are shown in Table 6. These appear to be low at the extreme values of r12 and high in the middle.This trend could lead to misleading conclusions about the ability of a model to predict concentrations of unknown compounds. This is why testing the models using independent test sets was thought appropriate. The two independent test sets (3 and 4) had values of r12 at 0.0 and 0.8, respectively, a standard deviation of noise of 3 3 10-4 AU and the same peak separation as the corresponding training sets discussed above. The results of testing how well the calibration models work are also shown in Table 6.Both sets of results show the same trend, in that RMSEP values increase as the value of the correlation coefficient of the calibration model increases. This is not difficult to comprehend, as a well constructed design (e.g., one with r12 at 0.0) would give the lowest errors, when predicting both test sets with r12 at 0.0 and 0.8, as opposed to a badly constructed model (e.g., one with r12 at 1.0), whose errors are considerably higher. Additionally, the RMSEP values of the training sets predicting the test set with r12 at 0.0 were significantly higher, than those obtained for predicting the test set with r12 at 0.8.This is because a less-well constructed test set (r12 = 0.8) is easier to predict by any model, whereas a well constructed text set will be hard to predict. Three graphs of time dependent loadings for compound I are shown in Fig. 6 and represent calibration models with r12 values at 1.0, 0.5 and 0.0. The corresponding graphs for compound II are shown in Fig. 7. From these, it can be concluded that the amount of information given out about the elution profile of a compound varies for the different calibration designs. For a calibration design with a value of r12 at 1.0 and predictions for compound k, the superimposed elution profiles of both compounds I and II are obtained in the first PLS component, in equal heights. In the second PLS component, the values were very low (all of the order of 10-5 mm).For a design of medium r12 and predictions for compound k, the first PLS component gives the superimposed elution profiles of both compounds, but this time the ratio of relative heights of compound k and the other compound increases as the value of r12 decreases. A calibration design with a value of r12 at 0.0 and predictions for compound k corresponds to the elution profile of pure k (first PLS component), whereas the elution profile of the other compound comes as a negative peak (second PLS component).The same principle is true for a plot of wavelength dependent loadings. Fig. 8 shows three such graphs for compound I and r12 values of 1.0, 0.5 and 0.0, whereas Fig. 9 shows the corresponding graphs for compound II. Choosing the correct experimental design could be of particular importance in calibration, as the information given out in the PLS predictions is instantaneously maximised. Additionally, minor impurities can be detected easily using the summed loadings over time method and a suitable calibration design.For example, when we are dealing with a hypothetically pure compound k and apply a calibration design with a value of r12 at 0.0 on it, then the slightest impurity in k would result in a substantial second peak present in the plot of summed loadings versus time for the first PLS component. For example, consider the case where compound I is contaminated by 0.5% of compound II. This might be common in synthetic analysis, where the main compound contains a small impurity of the second compound, and completely pure samples are difficult to obtain, especially when developing new synthetic methods.Fig. 10 represents the difference between the time dependent loadings plot for compound I (r12 = 0.0, standard deviation of noise of 3 3 10-4 AU, peak separation of 20 points in time) for a 0% and a 0.5% impurity of compound II introduced to the compound I chromatogram. This is equivalent to performing calibration where one of the components is in itself contaminated with small amounts of the other component, as often happens in exploratory synthetic method development.It can be seen that the first PLS component shows an obvious second peak at high time values, with the reverse for the second component. Provided that peak separation and noise levels are sufficiently low, the methods advocated in this paper are powerful approaches for the detection of small amounts of impurities.Table 5 List of independent test sets used in assessing how well the calibration models predict the concentrations of compounds I and II Correlation Standard coefficient deviation of Peak Test set No. of design, r12 noise (AU) separation/s 1 0.0 3 3 1024 16 2 0.8 3 3 1024 16 3 0.0 3 3 1024 12 4 0.8 3 3 1024 12 5 0.0 3 3 1024 8 6 0.8 3 3 1024 8 7 0.0 3 3 1024 4 8 0.8 3 3 1024 4 9 0.0 3 3 1024 0 10 0.8 3 3 1024 0 Table 6 RMSEs (mm) for the prediction of the concentration vectors of compounds I and II (autoprediction and testing) by calibration training sets with seven different correlation coefficients, a standard deviation of noise of 3 3 1024 AU and a peak separation of 12 s (one PLS component only).Test set 3 has r12 = 0.0 and test set 4 has r12 = 0.8 Compound I Compound II Correlation Test Test Test Test coefficient Autoprediction set 3 set 4 Autoprediction set 3 set 4 0.0 0.024440 0.024440 0.021490 0.010472 0.010472 0.009691 0.2 0.064040 0.070566 0.038304 0.022608 0.024327 0.016389 0.4 0.079001 0.101726 0.046203 0.029785 0.037352 0.019477 0.5 0.079600 0.112527 0.050411 0.031216 0.043223 0.020925 0.6 0.076657 0.121206 0.054210 0.031142 0.048563 0.022515 0.8 0.060238 0.134594 0.060241 0.025885 0.057641 0.025883 1.0 0.000020 0.145118 0.064902 0.000010 0.064884 0.029014 Analyst, June 1998, Vol. 123 1169Changing noise The effect of changing the noise of the system to the PLS predictions was also investigated. For a value of r12 at 0.5, and a peak separation of 12 s, five calibration training sets were generated, in which the standard deviation of the noise ranged from 3 3 10-2 to 3 3 10-6 AU, as shown in Table 4.The results of the autopredictions for compounds I and II are shown in Table 7. From these, it can be seen that increasing the noise of the system increases the Rlues in the autopredictions in a linear relationship. This trend is observed using two PLS components, as the first PLS component does not show any obvious trend. To test the five calibration training sets with the different noise levels, the two independent test sets (3 and 4) described in the section Changing correlation coefficient were used.The results of seeing how well the five training sets predict the concentrations of the two compounds in test sets 3 and 4 are also Fig. 6 Time dependent loadings for a model with r12 = (a) 1.0, (b) 0.5 and (c) 0.0 and compound I.Fig. 7 Time dependent loadings for a model with r12 = (a) 1.0, (b) 0.5 and (c) 0.0 and compound II. Fig. 8 Wavelength dependent loadings for a model with r12 = (a) 1.0, (b) 0.5 and (c) 0.0 and compound I. Fig. 9 Wavelength dependent loadings for a model with r12 = (a) 1.0, (b) 0.5 and (c) 0.0 and compound II. 1170 Analyst, June 1998, Vol. 123shown in Table 7. Exactly the same trends are observed as when using autopredictions, but this time the increase of errors (RMSEP) with increasing noise is less linear (using two PLS components).By comparing the errors in the first PLS component (for both autoprediction and testing), it is seen that using a training set with r12 at 0.5 would give higher errors when predicting a test set with r12 at 0.0, whereas the errors would be lower for the autopredictions (r12 at 0.5) and significantly lower for predicting a test set with r12 at 0.8. Changing peak separation Finally, the effect of changing the separation of the two peaks with respect to one another on the PLS predictions was investigated.Five calibration models were generated with separations of 0, 4, 8, 12 and 16 s. All five training sets had a standard deviation of noise at 3 3 10-4 AU and were based on a design with r12 at 0.5. The results of the autopredictions are shown in Table 8. From these, it is evident that increasing peak separation for the two compounds results in a decrease in RMSE values for the concentration predictions. This observation is made for the first PLS component only, as when using two PLS components the RMSEs were virtually zero in all designs (approximately 10-5 mm).The five calibration models were then tested against some independent test sets to check on the validity of their predictions. Each of the five calibration training sets was tested against a test set with the same peak separation as itself, but with a value of r12 at 0.0 or 0.8, and a standard deviation of noise at 3 3 1024 AU.In total, 10 different test sets were used (1–10), so that the same peak separation is featured in each pair of test and training sets. The results were as expected, namely that the smaller the separation between the peaks, the higher were the errors (RMSEP) in the predictions. As before, the concentration prediction errors of the models were high when predicting the test sets with r12 at 0.0, moderate when predicting themselves (r12 at 0.5) and low when predicting the test sets with r12 at 1.0.Conclusions This paper has described a potentially useful approach for the calibration and quantification of diode-array HPLC data, which can easily be applied to real experimental situations. A great deal is learnt about the effectiveness of PLS for quantitative prediction which can be extended to more general situations. Above, it is shown that the size of the residual after two PLS components have been computed is related to the noise level, as expected, for the data in this paper, which are relatively easy to analyse.It is important to recognise that baseline effects, and small underlying impurities could also influence the size of the residual. However, experimental design and the nature of the test set are seen to be of major importance when assessing the quality of Fig. 10 Difference in summed loadings over time between calibration experiments formed with a 0.5% impurity of II in I and pure compound I, for a model with r12 = 0.0, a standard deviation of noise at 3 3 1024 AU and a peak separation of 20 points in time.Table 7 RMSEs for the prediction of the concentration vectors of compounds I and II (autoprediction and testing) by calibration training sets with five different noise levels, an r12 value of 0.5 and a peak separation of 12 s (for one and two PlS components) Standard Autoprediction Test set 3 Test set 4 deviation of Compound noise (AU) N = 1 N = 2 N = 1 N = 2 N = 1 N = 2 I 3 3 1022 0.079637 2.61 3 1023 0.112547 3.45 3 1023 0.050362 2.07 3 1023 3 3 1023 0.079592 2.32 3 1024 0.112516 9.52 3 1025 0.050404 7.68 3 1025 3 3 1024 0.079600 3.19 3 1025 0.112527 3.36 3 1025 0.050411 3.08 3 1025 3 3 1025 0.079601 2.59 3 1026 0.112526 2.78 3 1025 0.050410 2.77 31025 3 3 1026 0.079600 2.33 3 1027 0.112526 2.76 3 1025 0.050410 2.75 3 1025 II 3 3 1022 0.030958 1.20 3 1023 0.043223 1.70 3 1023 0.020735 8.90 3 1024 3 3 1023 0.031212 1.30 3 1024 0.043250 8.15 3 1025 0.020950 6.90 3 1025 3 3 1024 0.031216 1.74 3 1025 0.043223 1.56 3 1025 0.020925 1.46 3 1025 3 3 1025 0.031214 1.34 3 1026 0.043223 1.41 3 1025 0.020924 1.39 3 1025 3 3 1026 0.031213 2.02 3 1027 0.043222 1.38 3 1025 0.020924 1.37 3 1025 Table 8 RMSEs for the prediction of the concentration vectors of compounds I and II (autoprediction and testing) by calibration training sets with five different peak separations, an r12 value of 0.5 and a standard deviation of noise of 3 3 1024 AU (one PLS component) Compound I Compound II Separation of Test sets Test sets Test sets Test sets peaks/s Autoprediction 9, 7, 5, 3, 1 10, 8, 6, 4, 2 Autoprediction 9, 7, 5, 3, 1 10, 8, 6, 4, 2 0 0.093174 0.131619 0.059123 0.048965 0.069071 0.031207 4 0.091180 0.128865 0.057777 0.046181 0.065031 0.029589 8 0.085691 0.121182 0.054197 0.038823 0.054299 0.025351 12 0.079600 0.112527 0.050411 0.031216 0.043223 0.020925 16 0.076409 0.107943 0.048485 0.027458 0.037794 0.018683 Analyst, June 1998, Vol. 123 1171models. Cross-validation often produces an over-optimistic assessment of prediction quality. For example, if a calibration data set is correlated, it may predict itself fairly well, but not a general set of all possible uncorrelated correlograms. Note that for a good uncorrelated design and five concentration levels (which is the minimum recommended for calibration), 25 experiments should be performed for two components.Smaller calibration sets (typical in most analytical laboratories) risk correlation between components, and so a false sense of security. For more than two components in a mixture, good design is mandatory. Although the calibration and prediction errors were assessed on only one PLS component, under-estimating the number of significant components is fairly common in many situations. For example, if there are several compounds in a mixture, in the presence of noise, it is common to be able to model the data well with less components than compounds.More significantly, if there are correlations between the concentrations in the calibration data set (which in practice happens in most real situations), this will reduce the apparent dimensionality. When two compounds are completely correlated there appears to be only one PLS component. For a typical correlation of 0.7–0.8, in the presence of high noise levels, it would be common to model the data satisfactorily using fewer components.The concentration levels modelled in this paper are fairly high, resulting in maximum absorbances over all wavelengths and times of around 1 AU. In addition, the relative average concentrations of both compounds are approximately equal. For impurity monitoring, one component may be present at much lower relative concentrations. Nevertheless, the numbers in this paper give some guidance as to the level of prediction errors found. For example, in Table 7, the error of prediction of compound I using test set 2, two PLS components, and at the highest noise level is about 1.09% ( = 0.0035/0.32).Note that if the level of this compound is low (e.g., a 0.1% impurity) the prediction error would be correspondingly much higher. Note also that percentage prediction error becomes much higher if one component is recorded in low relative concentration. Nevertheless, this paper provides guidelines on how to estimate prediction errors as a function of noise level, peak separation and calibration design. Finally, the PLS loadings plots are seen to be very diagnostic of the spectra and elution profiles of the pure compounds.The appearance is influenced in addition by calibration design. If it is desired to obtain the pure spectra by these means, it is important to have as orthogonal a design as possible. If a series of chromatograms are not orthogonal, a possible approach would be to remove a variable number of chromatograms from the data set and perform PLS as described above, but to calculate the loadings on different subsets of the data with different correlation coefficients.The further the correlation coefficients are from zero, the more mixed the loadings plots are. By visually comparing a series of such graphs, it should be possible to determine the features of each pure component in the mixture. Appendix List of notations X Matrix of absorbance values at successive time points and various wavelengths y Vector of compound concentration J Total number of wavelengths I Total number of points in time aj Sum of the area of a chromatographic peak at j wavelengths bi Elution profile of a chromatographic peak, summed over all wavelengths xij Point in data matrix IJX at time i and wavelength j M Number of experiments T Scores matrix after performing PLS1 on matrix X P Loadings matrix after performing PLS1 on matrix X E Residual matrix after performing PLS1 on matrix X u Scores vector for concentration vector y q Loadings vector for concentration vector y f Residual vector for concentration vector y.Predicted concentration vector after performing PLS calibration N Number of PLS components extracted nn Contribution to the true concentration y for n PLS components �ym,N Predicted concentration for sample m after N PLS components are extracted �y Mean compound concentration time I,N P Matrix of summed loadings over time timepi,n Point in time I,N P matrix, at time i and PLS component n J,N lP Matrix of summed loadings over wavelength lpj,n Point in l J,NP matrix at wavelength j and PLS component n n 1,J �sk True experimental normalised spectrum of pure compound k at j wavelengths, and averaged between points I1 and I2 Ak Absorbance value at point of maximum intensity for symmetrically simulated elution profile of compound k tk Retention time at point of maximum intensity for symmetrically simulated elution profile of compound k sk Factor relation to width of the peak at half its height ci,k Point of simulation elution profile for symmetric peaks and compound k I,1�ck Symmetrically simulated elution profile for pure compounds k, based on a Gaussian peak shape I,J �X k X matrix for compound k, obtained by multiplying the simulated elution profile for compound k with its true normalised experimental spectrum yl,k Concentration of compound k at a coded level l ymax,k Maximum true chromatographic concentration of compound k Md1 Calibration design with coded concentrations for compound I Md2 Calibration design with coded concentrations for compound II M,I,J Z Three-way tensor comprising of M two-way matrices containing both compounds, mixed in various proportions M,1y1 True concentration vector for compound I derived from Md1 M,1y2 True concentration vector for compound II derived from Md2 M,I,J N Three-way tensor, comprising of M noise matrices of dimensions I 3 J r12 Correlation coefficient of calibration design RMSE Root-mean-square error between predicted and true concentrations for autopredictions ylm,k Concentration for sample m of compound k at level l �ym,n,k Predicted concentration of sample m and compound k, using n PLS components 1172 Analyst, June 1998, Vol. 123RMSEP Root-mean-square error between predicted and true concentrations for testing the calibration training sets using independent test sets testylm,k Concentration of sample m of test set and compound k at level l test�ym,n,k Predicted concentration of sample m of test set and compound k, using n PLS components and a calibration training set References 1 Bro, R., and Heimdal H., Chemom.Intell. Lab. Syst., 1996, 34, 85. 2 Bro, R., J. Chemom., 1996, 10, 47. 3 Smilde, A. K., J. Chemom., 1997, 11, 367. 4 Demir, C., and Brereton, R. G., Analyst, 1997, 122, 631. 5 Kowalski, B. R., Gerlach, R., and Wold, H., in Chemical Systems Under Indirect Observation, ed. Joreskog, K., and Wold, S., North- Holland, Amsterdam, 1982. 6 Wold, S., Ruhe, A., Wold, H., and Dunn, W. J., III, J.Sci. Statist. Comput., 1984, 5, 735. 7 Delaney, M. F., Chemom. Intell. Lab. Syst., 1988, 3, 45. 8 Alsberg, B. K., Winson, M. K., and Kell, D. B., Chemom. Intell. Lab. Syst., 1997, 36, 95. 9 Geladi, P., and Kowalski, B. R., Anal. Chim. Acta, 1986, 185, 1. 10 Haaland, D. M., Anal. Chem., 1988, 60, 1208. 11 Wold, S., Albano, C., Dunn, W. J., III, Esbensen, K., Hellberg, S., Johansson, E., and Sjøstrom, M., in Food Research and Data Analysis, ed. Martens, H., Russworm, H., Applied Science, London, 1983. 12 Næs, T., and Martens, H., Commun. Statist. Simul. Comput., 1985, 14, 545. 13 Helland, I. S., Rep. Depart. Math.Statist. Agric. Univ. Norway, 1986, 21, 44. 14 Lorber, A., Wangen, L. E., and Kowalski, B. R., J. Chemom., 1987, 1, 19. 15 Lindgren, F., Geladi, P., and Wold, S., J. Chemom., 1993, 7, 7. 16 Brown, P. J., Anal. Proc., 1990, 27, 303. 17 Araujo, P. A., Cirovic, D. A., and Brereton, R. G., Analyst, 1996, 121, 581. 18 Cirovic, D. A., Brereton, R .G., and Walsh, P. T., Analyst, 1996, 121, 575. 19 Riu, J., and Rius, F. X., J. Chemom., 1995, 9, 343. 20 Henrion, R., Chemom. Intell. Lab. Syst., 1994, 25, 1. 21 Ståhle, L., Chemom. Intell. Lab. Syst., 1989, 7, 95. 22 Sun, J., J. Chemom., 1996, 10, 1. 23 Smilde, A. K., and Doornbos, D. A., J. Chemom., 1991, 5, 345. 24 Geladi, P., Chemom. Intell. Lab. Syst., 1989, 7, 11. 25 Bryant, D. K., Kingswood, M. D., and Belenguer, A., J. Chemom., 1996, 721, 41. 26 Zissis, K. D., Brereton, R. G., and Escott, R., Analyst, 1997, 122, 1009. 27 Brereton, R. G., Analyst, 1997, 122, 1521. 28 Euler, L., Memoir presented to Academy of Science of St. Petersburg on 8th March 1779, published as Leonardi Euleri Opera Omnia, Serie 1, 1932, 7, 291. 29 Plackett R. L., and Burman J. P., Biometrika, 1946, 33, 305. Paper 8/01285G Received February 13, 1998 Accepted March 24, 1998 Analyst, June 1998,

ISSN:0003-2654    

DOI:10.1039/a801285g

出版商:RSC    

年代:1998

数据来源: RSC

摘要:

Evaluation procedures for reversed-phase high-performance liquid chromatographic columns in the analysis of strongly basic compounds using principal components analysis for data assessment Richard G. Breretona and David V. McCalleyb a School of Chemistry, University of Bristol, Cantock’s Close, Bristol UK BS8 1TS b Faculty of Applied Sciences, University of the West of England, Frenchay, Bristol UK BS16 1QY The performance of eight high-purity silica-based reversed-phase columns was analysed using chemometric pattern recognition.Tests were performed at pH 3.0 and 7.0, using 10 compounds, three mobile phase modifiers (methanol, tetrahydofuran and acetonitrile) and four parameters {retention factor (kA), column efficiency (N), the Dorsey–Foley efficiency [N(df)] and the asymmetry factor (As])}. Principal components analysis was performed on subsets of the data. Features of both scores and loadings plots, together with the corresponding correlation coefficients between chromatographic tests, are discussed in detail.It is concluded that methanol and acetonitrile exhibit similar properties, the relative performance at each pH differs, the number of compounds used in the test can be reduced to a recommended five, and several of the original tests are redundant. Procedures are recommended for determining a range of tests that provide information on the column performance, which can be applied to other situations. Keywords: High-performance liquid chromatography; chemometrics; column assessment; pattern recognition; pharmaceuticals The assessment of the quality of silica-based reversed-phase (RP) columns for the analysis of basic substances is of considerable importance, owing to the widespread use of HPLC, particularly in the pharmaceutical industry and in the biomedical science area, for the analysis of these compounds.Bases often give asymmetric peaks and poor efficiency, owing to undesirable column interactions, particularly with silanol groups.1 Although many workers have proposed different test compounds which are claimed to be suitable for this purpose, there is no commonly accepted method for carrying out such evaluations.Perhaps the best known procedure is the Engelhardt test,2,3 in which a mixture of compounds including acidic, basic and neutral species is analysed using an unbuffered methanol–water mobile phase. Although this procedure has much merit, particularly for the distinction of the older type of RP column from newer, high-purity silicas which in general have much less activity towards bases, the test does have some limitations.For example, it has been shown that compounds of even moderately high pKa can generate distorted peaks in unbuffered mobile phases, attributable to variable solute ionisation.4 Alternatively, few compounds of low pKa present challenging tests for the newer columns, although it seems these columns can still differ substantially in their inertness towards bases.Furthermore, we have shown that there is not necessarily a relationship between the performance of a particular column with a given test compound and its performance with other solutes and hence that it is highly unlikely that one or even two basic types of test compound can give a satisfactory overall assessment. In two recent papers,4,5 we have published data containing retention factor (kA), column efficiency (N) and asymmetry factor (As) for 10 different test compounds, using phosphate buffers at pH 3.0 and 7.0 modified in each case with methanol, acetonitrile and tetrahydrofuran (THF) (i.e., six different mobile phases). It might be possible to isolate significant compounds and conditions from this large data set to enable a reasonably simple and meaningful test to be performed. There have been a few papers in which chemometric pattern recognition methods were applied to analyse HPLC information, 6–9 but many of these concentrated on retention data.In an interesting publication, Vervoort et al.10 used chemometric procedures to isolate a test set of four significant probes from a database consisting of 32 retention and peak-shape measurements from proprietary pharmaceutical compounds which were then used to evaluate RP columns. Our data set differs, in that it consists of compounds which are readily available and have greater structural differences than those employed by Vervoort et al. Furthermore, because our data set contains compounds which have been widely used by many different workers and column manufacturers in column tests (although in general only as single probes) such as pyridine, codeine and a tricyclic antidepressant, it could be argued that already it contains compounds which are likely to show up significant column differences.Furthermore, we have used quinine as a column evaluation probe for many years.11 Our data set also includes performance data using both acetonitrile and THF, in addition to methanol. The aim of this work was to use chemometric procedures to analyse this data set, more specifically as follows. 1. To investigate whether testing in one organic modifier could give sufficient information about column performance. We previously noted that differences in column performance can occur depending on the choice of modifier, particularly at pH 7.0. Although it was demonstrated that performance at pH 7.0 was generally worse with acetonitrile and best with THF, we have not so far investigated whether these are mostly differences in the magnitude of the results, or whether there are also significant changes in column ranking with organic modifier. 2. To determine whether testing at one pH value is sufficient to evaluate a column. We have shown previously that most solutes gave better peak shape at pH 3.0 than pH 7.0. However, we have not investigated in detail changes in column ranking that could possibly occur with change in pH. 3. To establish groups of basic compounds which could be used to test the newer generation of reversed phases. While our previous work has demonstrated that column assessment is Analyst, June 1998, Vol. 123 (1175–1185) 1175solute dependent, we have not previously attempted to identify a group of compounds that give apparently unrelated effects and could be usefully included in a test designed to predict the overall activity of a column towards bases.Experimental Full details of the HPLC equipment and columns used have been given previously.4,5 The datasets are summarized in Table 1. Eight chromatographic columns were tested, at two pH values (7.0 and 3.0). Ten test compounds were used, of which nine were tested at pH 7.0 and eight at pH 3.0. Single-letter mnemonics will be used below, for brevity, as indicated in Table 1. Three isoeluotropic mobile phases were used at each pH (Table 1), and in all cases isocratic conditions using a single solvent were employed.Finally, three column evaluation parameters, kA , N and As, were recorded for the probes at pH 3.0 and 7.0. The following equations were used to calculate these parameters kA = (tr 2 t0)/t0 (1) where tr is the retention time and t0 the column void volume (measured with uracil), and N = 5.54 (tr/w0.5)2 (2) where w0.5 is the peak width at half-height. As was calculated from the ratio of the widths of the leading and trailing edge of the peak, measured at 10% of the peak height.At pH 3.0, column efficiency was also measured according to the Dorsey-Foley procedure5 N(df) = 41.7 [(tr/w0.1)2/(As + 1.25)] (3) where w0.1 is the peak width at 10% of the height. Chemometrics Principal components analysis The column assessment data were mostly treated separately for each pH. A matrix consisting of eight rows (corresponding to the chromatographic columns) and a column for each of the assessment parameters was obtained.At pH 3.0 this matrix consisted of 96 ( = 8 34 33) columns, one for each compound (8), assessment parameter (4) and mobile phase (3), and at pH 7.0 it consisted of 81 ( = 9 3 3 3 3) columns. The columns of the matrix were labelled with the mnemonics above, with kA, N, N(df) and As corresponding to the four assessment parameters. For example, AAsc represents the value of asymmetry for compound A (amphetamine) and condition c (acetonitrile).This raw matrix will be denoted by C with elements cij, where i is the column and j the column assessment parameter. In this study, we did not log scale the data. It is well established that different trends are revealed in the data by log scaling,12 for example using spectral map analysis, and it would be constructive to compare both logarithmic and non-logarithmic methods for preprocessing the data. However, other investigators use raw data as input to principal components analysis (PCA) programmes for the analysis of retention and other related data.8,10 There is no hard and fast rule.One potential weakness of log scaling in the case studied in this paper is that some retention data have kA values of 0, i.e., when a compound elutes with the solvent front, which would cause some problems for log scaling. Because each of the efficiency parameters was measured on different scales, the first step was to standardise each column of the matrix, to give a new matrix X as follows: x c c s ij ij j j = - ( ) (4) where �c j is the mean value of parameter cij over all eight columns and sj is the corresponding standard deviation.PCA12–15 was performed on this matrix in the normal way to give scores and loadings matrices: X = T.P + E (5) All software was applied in Excel validated against a variety of other packages. PCA scores and loadings plots were obtained as appropriate. In order to simplify the plots, subsets of the original data were selected, e.g., all the column parameters for a single mobile phase.The data were treated in a similar manner as above, but PCA was performed independently, and appropriate graphs obtained. However, one problem encountered in practice is that the signs of principal components often invert in an unpredictable manner. This is due to the algorithm employed and has no physical significance. We have recently commented on this in the context of procrustes analysis.16 To allow comparability of plots, the signs of the PC scores and loadings were inverted manually to result in graphs of scores under different chromatographic conditions whose axes were similar in sign.In no case was there swap-over of principal components, the only variation being in the signs. Table 1 Datasets used in this paper Columns— Number Column 1 Inertsil ODS 2 Inertsil ODS-2 3 Inertsil ODS-3 4 Kromasil C18 5 Kromasil C8 6 Symmetry C18 7 Supelco ABZ+ 8 Purospher Test compounds— Compound Letter pH 3.0 pH 7.0 Pyridine P ] ] Nicotine N ] ] Amphetamine A ] ] Codeine C ] ] Diphenhydramine D ] ] Nortriptyline R ] ] Procainamide M 3 ] Quinine Q ] ] Benzylamine B ] 3 2-[N-methyl-N- E 3 ] (2-pyridyl) amino]ethanol Test conditions— Mobile phase a3 and a7 Methanol–phosphate buffer, pH 3.0 and 7.0 b3 and b7 THF–phosphate buffer, pH 3.0 and 7.0 c3 and c7 Acetonitrile–phosphate buffer, pH 3.0 and 7.0 As = asymmetry factor, kA = retention factor, N = column efficiency, N(df) = Dorsey–Foley column efficiency. Thus QNa3 signifies column efficiency measurement for quinine using methanol–phosphate buffer at pH 3.0 1176 Analyst, June 1998, Vol. 123Correlation coefficients Where appropriate, correlation coefficients were computed between the 96 (pH 3.0) and 81 (pH 7.0) column assessment measurements. Principal components plots In most cases, several components are required to describe the full variance of the data. However, as is common with most workers, we chose to represent the patterns graphically using a single plot of PC2 versus PC1 in all cases.For the majority of column assessment parameters, the first two PCs describe by far the most variance as assessed by the size of the eigenvalues of the components, and those parameters that characterise the final components are likely to be atypical or rogue variables. Another important reason for restricting the graphs to those of the first two PCs is that, by using different conditions, the amount of variance described by these PCs differs.Strictly, under one set of conditions, the data may more accurately be described by a three-dimensional plot and under another set by a twodimensional plot, but it is not practicable to compare graphs of different dimensionality directly. A final, and important, reason is provided when examining the loadings plots for the column assessment parameters. Whereas the first two PCs provide good separation of the parameters into well defined groups, with the first PC primarily responsible for the separation of the As values from the others, and the second PC for further separating kA and N values, there is no such clear and reproducible separation using the third PC.Results and discussion There are numerous possible approaches to organisation of the data. In this study we chose to select the rows as chromatographic columns at a given pH. As will be shown, it is a mistake to mix column performance parameters at different pHs and so a short, fat matrix consisting of all common chromatographic parameters and both pHs will not give a sensible assessment of column performance.Relative performance changes according to pH. It is, possible however, to arrange the data in several ways. An alternative approach may involve multimode factor analysis (e.g., PARAFAC) in which compounds, chromatographic assessment parameters, pH and columns are different factors, but the resultant graphical information will be hard to interpret.The methods in this paper could be regarded as the result of unfolded PCA. Column evaluation at pH 3.0 Fig. 1 shows a score plot for all eight compounds, using all three mobile phase conditions and four column performance parameters [kA, As, N and N(df)] at pH 3.0. The scores suggest that using these parameters, Purospher is an outlier. This may because this column was not sufficiently wetted by the mobile phases employed, leading to unusual values of kA and N as discussed earlier.4 The three Inertsil columns appear reasonably close together in the score plot; they are made by the same manufacturer but employ different silicas and bonding procedures, so this result is as expected. Supelco ABZ+ is an electrostatically shielded phase, unlike the others, which may explain its somewhat outlying position.Numerous observations and deductions are possible by computing the principal components under each of the three chromatographic conditions independently. 1. The scores plots are illustrated in Fig. 2. Those for methanol and acetonitrile mobile phases are extremely similar; for example, the correlation coefficient between the scores in the two solvents is 0.982 for the first PC and 0.963 for the second PC. The order of the columns along the first principal component (x-axis) for both conditions is almost identical with Kromasil C8 at one extreme, the three Inertsil columns showing very similar scores, then Supelco ABZ+ and finally Purospher at the other extreme.It is important to recognise that the first PC is much more significant than the second, so the differences between the positions of the columns along the second PC (y- Fig. 1 Scores plot at pH 3.0, all conditions. In this and other figures, the first PC is represented by the horizontal axis and the second by the vertical axis. Fig. 2 Scores plots, pH 3.0: (a) methanol, (b) THF and (c) acetonitrile mobile phase.Analyst, June 1998, Vol. 123 1177axis) are not as consequential; for methanol the sum of squares of the scores of the first PC is 95.0 (42.4%) as opposed to 50.2 (22.4%) for the second PC, for example meaning that the first PC is around twice as significant as the second. Nevertheless, the ordering of the columns along the second component is still very similar, with Inertsil ODS-2 and Supelco ABZ+ at one end and Symmetry C18 followed by Kromasil C18 at the other end in both cases.The order of columns along the first PC for THF is similar to the order in methanol and acetonitrile, apart from Symmetry C18. The improvement in performance shown by many of the columns with THF5,17 as opposed to methanol and acetonitrile was not really shown by Symmetry C18, hence the positions of the other columns in the scores plot could be considered to me as a group, leaving this column ‘behind’. Similarly, considering the position of the columns along the second PC, Kromasil C18 moves from a somewhat outlying position in methanol and acetonitrile to be close to several other columns in THF.This may be attributed to the unusually large improvement in the performance of this column with THF noted previously. It appears that while the relative performance of columns at pH 3.0 can be influenced by the organic modifier, these differences are not sufficiently large to invalidate an evaluation based on one mobile phase condition at pH 3.0.THF, which gives the most different behaviour, is rarely used owing to practical difficulties; hence it can be rejected because of the hazards and inconvenience associated with this solvent. It can be concluded, therefore, that a test at pH 3.0 could be performed equally well with acetonitrile or methanol. 2. The positions of the loadings (Fig. 3) can be analysed in detail and are discussed in the case of methanol mobile phase [Figure 3(a)].The angles between the loadings in Fig. 3(a) relate to the correlation between the parameters. (i) Parameters that are opposed i.e., at approximately 180° in the loadings plot, measure equivalent but opposite trends and have correlations close to 21; for this reason, it may be unnecessary to record both parameters. Two examples (Table 2) are PN(df)a3 and PAsa3 which have a correlation coefficient of 20.90. The highest values of As are given by Kromasil C8 and Kromasil C18, followed by Symmetry C18, with the lowest asymmetry being exhibited by Supelco ABZ+; the column efficiencies show an opposite trend.(ii) Parameters that measure unrelated trends are at approximately 90°. Such parameters are prime candidates as complementary measures for inclusion in a test procedure since they may measure unique properties. CAsa3 and DAsa3 are examples; the highest value of DAsa3 corresponds to an intermediate value of CAsa3 and vice versa.Therefore, measuring CAsa3 cannot predict the values of DAsa3, so each parameter adds complementary information. The correlation between these two parameters is 20.12, or close to zero. (iii) Parameters that measure related trends have a small angle between them; again, it is likely that duplication of information is occurring and the correlation coefficients are close to 1. Dka3 and Rka3 are two such examples, with a correlation of 0.99. Note that the size of the kA values are different for the two compounds, but the relative magnitudes are extremely similar, with Kromasil C18 giving the highest kA and Purospher the lowest, both for diphenhydramine and for nortriptyline.Under these circumstances there is no point in measuring both parameters. 3. The loadings plots for methanol mobile phase show similar trends to acetonitrile, comparing Fig. 3(a) and (c). There are slight differences with THF [Fig. 3(b)], as is also evident in the scores plots discussed above.We will restrict the discussion to methanol. 4. The loadings tend to cluster into three groups, namely kA, As and N [including N(df)] for all three mobile phases, [see Fig. 3(a)]. However, the kA values cluster very closely so it can be concluded that retention factors are not particularly good for distinguishing this group of columns as described in 2(iii) above. Furthermore, the angles between the cluster of kA values and those for the various As measurements would imply that there is little correlation between retention factor and asymmetry factor for a given solute.This is confirmed by the correlation coefficients, e.g., PAsa3 : Pka3 = 0.39 and RAsa3 : Rka3 = 0.22. This would imply that columns that show a high kA for a particular compound do not necessarily show high As for that compound; this observation may be a consequence of the relative inertness of the columns used in this study and the suppression of silanol ionisation at low pH.This result confirms findings from a previous study.10 5. There is very little distinction between N and N(df) values. Hence, even though it may be difficult to measure N accurately for asymmetric peak shapes and these numbers are considerably different in magnitude from N(df) , both measure similar trends. The mean value of the eight correlation coefficients between N(df) and N for all the test compounds was 0.89. For example, the parameters QN(df)a3 and QNa3 are separated by only a small angle in the loadings plot, and have a correlation coefficient of 0.96.The ranking of columns using QNa3 measurements is Inertsil ODS-2 (highest efficiency) > Kromasil C8 > Inertsil ODS-3 > Kromasil C18 > Symmetry C18 > Inertsil ODS > Supelco ABZ+ > Purospher (lowest efficiency), whereas with QN(df)a3 it is Inertsil ODS-2 (highest efficiency) > Kromasil C8 > Inertsil ODS > Inertsil ODS-3 > Kromasil C18 > Supelco ABZ+ > Symmetry C18 > Purospher (lowest efficiency).Apart from Inertsil ODS, the rankings are very similar. The values of N for the intermediately ranked columns are very close. In conclusion, it would appear unnecessary to measure both parameters. 6. It appears that As values are in many cases diametrically opposed to N values, suggesting that these parameters in general measure opposite trends. This is to be expected, since as the peak asymmetry increases, the column efficiency should decrease.For example, the parameters for pyridine, PNa3 and PAsa3 have an angle of nearly 180° between them in Fig. 3(a). The correlation coefficient of 20.81 confirms this. Rankings with PNa3 are Inertsil ODS (highest efficiency) > Inertsil ODS-3 > Purospher > Inertsil ODS-2 > Supelco ABZ+ > Symmetry C18 > Kromsasil C18 > Kromasil C8 (lowest efficiency) and with PAsa3 Kromasil C8 (highest asymmetry) > Kromasil C18 > Symmetry C18 > Inertsil ODS-3 > Inertsil ODS > Inertsil ODS-2 > Purospher > Supelco ABZ+ (lowest asymmetry).The rankings are not entirely opposite but the three columns with lowest efficiency have highest asymmetry. There are, nevertheless, a few small changes in ranking for some of the other chromatographic columns. However, the vectors for As and N are not exactly 180° for corresponding compounds in all cases. The mean value of the correlation coefficient between N and As is 20.44.Hence it appears necessary to monitor both N and As values to characterise fully the relative column performance. This is a confirmation of the conclusion reached earlier by a more casual inspection of the results.5 The average angle between N(df) and As is closer to 180° in the loading plots, and indeed the mean value of the correlation coefficient is 20.69. This closer approach to the value of 21.0 is hardly surprising in the light of the equation for calculation of the Dorsey–Foley efficiency parameter, and it could be argued that if only one parameter was to be chosen, N(df) might be a suitable single replacement parameter for N and As.However, the relative infrequency of use of N(df) mitigates against consideration of this in more detail. 7. The angle between the loadings can be used to select test compounds which measure apparently different column properties while eliminating compounds which merely duplicate evaluations.For example, the vectors RAsa3 and DAsa3 have a relatively small angle between them. The correlation coefficient for these parameters r = 0.83. The ranking using RAsa3 is 1178 Analyst, June 1998, Vol. 123Kromasil C18 (highest asymmetry) > Inertsil ODS-3 > Purospher > Kromasil C8 > Supelco ABZ+ > Inertsil ODS-2 > Symmetry C18 > Inertsil ODS (lowest asymmetry) whereas with DAsa3 it is Kromasil C18 (highest asymmetry) > Kromasil C8 > Inertsil ODS > Inertsil ODS-3 > Purospher > Supelco ABZ+ > Inertsil ODS-2 > Symmetry C18 (lowest asymmetry).These are fairly similar, only Kromasil C8 and Inertsil ODS being out of order; the difference for Kromasil is small. Hence, it may not be necessary to include both diphenhydramine (D) and nortriptyline (R) in a test mixture, since they behave in a similar fashion. Alternatively, as the angle between the vectors increases, the behaviour should become less similar. For example, the angle between RAsa3 and CAsa3 approaches 90° and the corresponding correlation is 0.13.The ranking of the columns with CAsa3 is Purospher (highest asymmetry) > Kromasil C8 > Supelco ABZ+ > Inertsil ODS-2 > Symmetry C18 > Kromasil C18 > Inertsil ODS-3 > Inertsil ODS (lowest asymmetry), which bears little relationship to that of RAsa3 (see above). Remember that a crude ranking of the columns as shown above gives no indication of the magnitude of the Fig. 3 Loadings plots pH 3.0: (a) methanol, (b) THF and (c) acetonitrile mobile phase. Analyst, June 1998, Vol. 123 1179differences between the columns and hence is probably less informative than the correlation coefficient. However, these results provide clear evidence for solute-dependent performance, and indicate that it would be desirable to have several compounds of different types in a test procedure. 8. Another interesting feature relates to the distance of the loadings from the centre of the graph [e.g., see Fig. 3(a)].The closer they are to the centre the less strong are the trends. Although ranks may be similar for two parameters related along a straight line, the magnitude will be higher the further away from the centre. The advantage of taking parameters further from the centre is that there is less risk of experimental error. Parameters close to the centre may simply give a fortuitous pattern due to a smaller scatter of variables.Because all variables are standardised, if the data were described by only two principal components, all variables would lie on a circle. Variables near the centre do not describe the pattern illustrated by two PCs, and variables near the centre are likely to show high loadings on the third or later principal component. Fig. 4 is a graph of distance from the centre against loadings on the third PC with methanol as a mobile phase. Under such circumstances, the correlation coefficients will not fully correspond to the angles in the PC plot.Parameters close to the centre such as NNa3 are in the top left hand corner of the graph, and so probably are not so suited to measure the main trends in the data. 9. By comparison of the scores and loadings plots, it is possible to see which chromatographic efficiency parameter is most associated with a given column. Compare, for example, Figs. 2(a) and 3(a) for methanol (condition a3). CAsa3 is in a similar direction to Purospher, DAsa3 to Kromasil C18 and PN(df)a3 to Supelco ABZ+.In all cases the maximum value of the chromatographic evaluation parameter is obtained when measured on the corresponding column. Hence this correspondence between column efficiency parameter and chromatographic column (sometimes displayed in a biplot) provides guidance as to which conditions are most appropriate for testing an individual column or group of columns. 10. From the loadings plots (Fig. 3), it is possible to recommend a reduction in the number of tests necessary. For the case of methanol as mobile phase modifier, if asymmetry is chosen, it would appear that the top right hand corner is well spanned by codeine, quinine, amphetamine, nortriptyline and diphenhydramine, resulting in spanning of over 90° in the loadings plots. Nicotine and benzylamine are rejected because they are too close to the centre. Pyridine is a possible alternative to diphenhydramine. Using methanol as the mobile phase modifier, there is not a great deal to choose between these two components.Pyridine spans a slightly wider space, but only 90° is necessary. If the loadings plot using acetonitrile as a mobile phase is also taken into consideration, diphenhydramine and nortryptyline are even closer together, with a correlation coefficient of 0.99, suggesting that their behaviour in acetonitrile is extremely similar. In methanol their correlation is 0.83 (see item 7 above), although pyridine is slightly closer to the centre.However, if solutes are to be chosen for both methanol and acetonitrile, pyridine is preferable to diphenhydramine. 11. The orders of the solutes are approximately the same for both As and N and when reading in an anticlockwise direction are codeine, quinine, amphetamine, nortriptyline, (nicotine), Table 2 Values of six chromatographic performance parameters in methanol Column PN(df)a3 PAsa3a CAsa3 Dasa3 Dka3 Rka3 Inertsil ODS 2650 2.27 1.18 1.71 2.65 8.62 Inertsil ODS-2 2820 2.11 1.42 1.39 1.72 5.02 Inertsil ODS-3 2320 2.53 1.28 1.6 2.73 9.1 Kromasil C18 293 5.35 1.37 2.64 2.75 9.25 Kromasil C8 229 6.46 1.58 1.79 2.27 6.67 Symmetry C18 944 3.13 1.38 1.39 2.54 7.9 Supelco ABZ+ 3660 1.96 1.49 1.49 0.55 1.8 Purospher 2780 2.08 1.66 1.57 0.35 1.45 Fig. 3 Continued 1180 Analyst, June 1998, Vol. 123diphenhydramine, (benzylamine), pyridine and (nicotine), quinine, codeine, benzylamine, nortriptyline, amphetamine, diphenhydramine, pyridine, with parentheses indicating solutes close to the centre of the plots.In both cases codeine and quinine are at one extreme, with pyridine at the other extreme. Note that nicotine is very close to the centre, suggesting that its position is unreliable, and benzylamine is likewise close to the centre for the asymmetry measurements. These arguments can help reduce the number of test compounds significantly. Interestingly, the As measurements are, in general, more spread out than the N measurements.It can be concluded that apart from pyridine, results using column efficiency are not as distinctive as measurement of asymmetry factor, and thus if only one parameter is measured, asymmetry factor may be a better choice. Considering their position in the loading plots and the arguments above, it would seem that codeine, quinine, amphetamine, nortriptyline and pyridine are candidates for giving a measure of the activity of these columns towards different basic solutes.In general, the findings of the present study are different from those of Vervoort et al.10 At pH 3.0, it was reported that the four test compounds used yielded similar information. This difference may be a reflection of the greater structural diversity of the compounds in the present study. Column evaluation at pH 7.0 Many conclusions can be obtained from these data. 1. The scores plot for the full data set at pH 7.0 for nine compounds and three performance parameters (kA, N and As) is shown in Fig. 5. It is immediately apparent that the clustering of the columns is different from that at pH 3.0. This would indicate that there are considerable differences in the ranking of the columns at pH 3.0 compared with pH 7.0. It should be noted that a slightly different set of compounds was used at each pH, and no N(df) values were included in the pH 7.0 data. 2. The seven compounds common to both pH studied and also the three common column performance parameters (63 parameters) were selected to produce the Fig. 6(a). Interestingly, pH is overwhelmingly important in separating the columns, and is a major factor contributing to principal component 1, with all chromatographic columns at pH 3.0 exhibiting negative scores, and those at pH 7.0 exhibiting positive scores in the first principal component. Hence pH, in this instance, is the prime factor that influences column performance. In addition, the relationship between the columns when considered altogether is similar to the relationship at each individual pH.Fig. 6(b) shows the scores plots of Figs. 1 and 6 superpositioned, with data at pH 3.0 on the left and pH 7.0 on the right and the axes reflected/rotated as appropriate (this is necessary because the sign of each component varies unpredictably according to algorithm). In both Figs. 6(a) and (b), for example, Purospher is at pH 3.0 is in the top left corner, whereas at pH 7.0 it is in the bottom right corner.It is important to realise, of course, that the score plots are not completely comparable because only 63 of the 114 possible conditions or slightly over 50% are in common since one of the four tests is used only at pH 3.0 and only seven of the 10 test compounds are common. However, the reasonably good correspondence suggests that the overall patterns obtained using a subset of the tests are robust, and that the tests used in this paper are more than sufficient to come to conclusions about the relative performance characteristics of the columns and both pHs.Hence we can be fairly certain that the conclusions about relative column performance are real and not simply due to an inadequate number of tests. 3. The patterns at pH 3.0 and 7.0 are demonstrably different, for either the full set of tests or the reduced set of 63 common tests. For example, Purospher and Supelco ABZ+ are close to each other at pH 3.0 (note that the second component is not as significant as the first) whereas they are almost diametrically opposed at pH 7.0.Symmetry C18 is close to the centre at pH 7.0 but far out at pH 3.0. Some features such as the close clustering of the three Inertsil columns remain at both pHs, but the differences are sufficiently reproducible and robust to suggest that there is very different relative column behaviour at the two pHs studied, even if different subsets of the tests are selected.Note that the performance at pH 3.0 is significantly better for all columns, as proposed previously,5 but the patterns change also. Fig. 4 Graph of normalised distance from centre of PC2 versus PC1 plot (horizontal axis) against proportion of variance described by the third principal component (vertical axis) for data at pH 3.0 using methanol as the mobile phase modifier. Analyst, June 1998, Vol. 123 1181This suggests that for the each column there is a different degree of improvement in performance with pH, and so it is important to study the column performance at more than one pH. 4. The scores plot for the individual mobile phases at pH 7.0 is given in Fig. 5. The scores plot for methanol mobile phase is very similar to the scores plot using all three mobile phases given in Fig. 7. There are some differences between the acetonitrile and methanol mobile phases with relative movements of Purospher and Intersil ODS-3. The pattern for THF, however, is different to the other two, suggesting, that at both pH 7.0 and 3.0, the use of THF gives anomalous results.On the whole, column performance improves on using THF, but the columns show very different relative improvements. For example, Kromasil C18 and Purospher show substantial improvements in THF, whereas the Symmetry column shows very limited improvement. 5. The corresponding loadings plots are given in Fig. 8. Several conclusions can be obtained, primarily by considering the results for methanol [Fig. 8(a)] and comparison with those at pH 3.0 [Fig. 3(a)] as appropriate. (i) The kA parameters are more spread out a pH 7.0 than pH 3.0, implying greater differences in the ordering of the columns depending on the particular solute chosen. (ii) However, the kA, N and As values are mutually clustered as at pH 3.0. The average angle between the As and kA values for corresponding compounds is 65° since they are relatively uncorrelated as at pH 3.0.The angles between N and As average at 147°, exhibiting an average correlation coefficient of 20.84, compared with 20.44 at pH 3.0. This suggests that N and As values measure similar but opposite properties and there is less reason to use both parameters at pH 7.0 as opposed to pH 3.0. (iii) Compared with pH 3.0, there is a smaller spread in As values for the different test compounds, suggesting that asymmetry is less useful for discriminating between the columns.The cyclical order of the compounds is approximately the same for both As and N values, with the exception of 2-[Nmethyl- N-(2-pyridyl)amino]ethanol, which is not a sufficiently demanding test compound, as noted previously.4 (iv) If N values were to be used for testing, a set of four compounds might include one of codeine and procainamide, quinine, one of nicotine, diphenydramine or nortriptyline, and one of amphetamine or pyridine. Each group contains one of the five proposed test compounds for pH 3.0, leaving codeine, quinine, nortriptyline and pyridine.The possible extension to Fig. 5 Scores plot at pH 7.0, all conditions Fig. 6 (a) Scores plot for columns at pH 7.0 and 3.0, 63 common conditions. (b) Scores plot for columns at pH 7.0 and 3.0, separately calculated, reflected and superpositioned, using all the conditions (based on Fig. 1 and 5). Fig. 7 Scores plots, pH 7.0: (a) methanol, (b) THF and (c) acetonitrile mobile phase. 1182 Analyst, June 1998, Vol. 123amphetamine results in a similar group of test compounds at both pHs, but a different set of parameters. It is preferable to reduce the number of test compounds but use different parameters than to remain with a single parameter but have to use a different range of test compounds at each pH for obvious experimental reasons. 6. The loadings plot for acetonitrile is similar to that for methanol, with only a few small differences, a movement of the position of 2-[N-methyl-N-(2-pyridyl)amino]ethanol (which is rejected above as a good test compound), a movement of amphetamine from the end to the centre of the N value cluster plus slight overlap between kA and As values.However, the five test compounds recommended above for methanol are similarly useful for acetonitrile, as can be seen by inspection of Fig. 8(c). Conclusions Chromatographic implications Many conclusions can be obtained from the detailed analysis above. 1.Meaningful evaluation of columns must be performed using at least two pH values, since the relative performance of columns is significantly different at ‘high’ and ‘low’ pH. Fig. 8 Loadings plots, pH 7.0: (a) methanol, (b) THF and (c) acetonitrile mobile phase. Analyst, June 1998, Vol. 123 11832. Assessment can be performed for the columns described in this paper with either methanol or acetonitrile as modifier. While the performance of most columns is poorer in acetonitrile at pH 7,0, the relative performance is mostly similar to that in methanol, although a few columns show anomalous behaviour.The relative performance in THF is by no means entirely dissimilar to that in methanol or acetonitrile, bur it does give the most anomalous results. Owing to the relative infrequency of use of THF, however, it can be rejected as a test modifier. Nevertheless, we would encourage more work with THF since it does tend to produce the best peak symmetries. 3. Column efficiency is negatively correlated with asymmetry factor, particularly at pH 7.0, hence only one of these parameters could be chosen for testing. The asymmetry factor seems to be a better measurement at pH 3.0, whereas column efficiency seems more useful at pH 7.0. Owing to the ease of making these measurements simultaneously with modern chromatography data systems, both of these measurements can be recorded. Although measurements of column efficiency (N) made at half-height are no doubt inaccurate, they seem adequate as they indicate similar trends to more sophisticated methods such as the Dorsey–Foley procedure [N(df)]. 4. Codeine, quinine, nortriptyline, amphetamine and pyridine are a set of probe compounds which are reasonably suitable for column evaluation at both high and low pH. It is shown clearly that the use of only one type of probe compound is inadequate for column characterisation owing to the strong dependence of performance on solute nature. Chemometric implications This paper has highlighted the importance of chemometrics for chromatographic column assessment.A vast wealth of information is present in the original data, which may not necessarily be obvious when examining the raw tables of numbers. Correct graphical presentation of results brings out these trends, but must, as always, be interpreted carefully and with caution. Reproducible trends, for example, when using different mobile phases or pHs are encouraging signs that the chemometrics has revealed true trends in the data.In the case of column evaluation measurements, absolute numbers are less useful than a graphical display, which shows both the relationship between different columns and also the relationship between column evaluation parameters. This paper suggests that pattern recognition has a very powerful role to play in assessing the quality of chromatographic columns and, in particular, in guiding the experimentalist as to which measurements are most useful. In this work we chose to use exploratory approaches. It is important to recognise that subsequent to this analysis more sophisticated methods could be employed. For example, hierarchical cluster analysis might be used to group the column evaluation test parameters in the form of a dendrogram. With sufficient columns, supervised methods for classification could be employed, and the variables that are most useful for this purpose identified, for example, those with high discriminatory or modelling power. However, the exploratory and graphical methods in this paper are advocated as a first step. References 1 Nawrocki, J., J. Chromatogr. A, 1997, 779, 29. 2 Engelhardt, H., and Jungheim, M., Chromatographia, 1990, 29, 59. 3 Engelhardt, H., Arangio M., and Lobert, T., LC-GC Int., 1997, 10, 803. 4 McCalley, D. V., J. Chromatogr. A, 1996, 738, 169. 5 McCalley, D. V., J. Chromatogr. A, 1997, 769, 169. 6 Walczak, B., Morin-Allory, L., Lafosse, M., Dreux, M., and Chretien, J. R., J. Chromatogr., 1987, 395, 183. 7 Walczak, B., Chretien, J. R., Dreux, M., Morin-Allory, L., and Lafosse, M., Chemom. Intell. Lab. Syst., 1987, 1, 177. 8 Rigezza, M., and Chretien, J. R., J. Chromatogr., 1991, 556, 169. 9 Hamoir, T., Sanchez F. C., Bourguignon, B., and Massart, D. L., J. Chromatogr. Sci., 1994, 32, 488 10 Vervoort, R., Derksen, M. W., and Debets, A. J., J. Chromatogr. A, 1997, 765, 157. 11 McCalley, D. V., J. Chromatogr., 1986, 357, 221. 12 Multivariate Pattern Recognition in Chemometrics, Illustrated by Case Studies, ed. Brereton R. G., Elsevier, Amsterdam, 1992 13 Brereton, R. G., Chemometrics: Applications of Mathematics and Statistics to Laboratory Systems, Ellis Horwood, Chichester, 1993. Fig. 8 Continued 1184 Analyst, June 1998, Vol. 12314 Wold, S., Esbensen, K., and Geladi, P., Chemom. Intell. Lab. Syst., 1987, 2, 37. 15 Mardia, K. V., Kent, J. T., and Bibby, J., Multivariate Analysis, Academic Press, London, 1979. 16 Demir, C., Hindmarch, P., and Brereton, R. G., Analyst, 1996, 121, 1443. 17 McCalley, D. V., J. Chromatogr. A, 1995, 708, 185. Paper 8/00149I Received January 5, 1998 Accepted March 24, 1998 Analyst, June 1998, Vol. 123 1185

ISSN:0003-2654    

DOI:10.1039/a800149i

出版商:RSC    

年代:1998

数据来源: RSC

摘要:

Characterisation of apple cider cultivars by chemometric techniques using data from high-performance liquid chromatography and flow-injection analysis D. Blanco-Gomisa, I. Herrero-Sáncheza and J. J. Mangas Alonsob a Departamento de Química Física y Analítica, Facultad de Química, Universidad de Oviedo, E-33006 Oviedo, Spain b Centro de Investigación Aplicada y Tecnología Agroalimentaria, E-33300 Villaviciosa, Spain Analytical techniques (HPLC and flow-injection analysis) for determining sugars, organic acids, polyphenols and pectins in apples, were employed along with chemometrics in the ripening and classification studies of cider apples.The use of principal component analysis allowed the authors to reduce the dimensionality of the data matrix; three new variables were obtained that accounted for 76% of variance. The projection of the apple cultivars in the reduced space allowed us to visualize the data structure on the basis of the degree of ripening and technological characteristics of the cider apple varieties monitored.Linear discriminant analysis computed a canonical variable with a prediction capacity of 93%, using three groups for cancellation in order to validate the method. The use of modelling techniques, such as SIMCA and partial least squares made an adequate grouping of apple cultivars feasible on the basis of their degree of ripening. Keywords: High-performance liquid chromatography; flow-injection analysis; chemometrics; ripening; apples Cider is one of the most popular drinks all around the world, and especially in France, the United Kingdom, Germany and Spain.In these countries, the apple crop and its subsequent transformation in order to obtain derivatives (brandy, vinegar, apple juice, etc.), is of enormous commercial, economic as well as social relevance. Therefore, any study related to the improvement of the conditions of production and more profitable farming practice, is of real importance and utility.During the apple growing process, different biochemical transformations are produced. These changes exert a special influence on the contents of different compounds such as sugars, acids, polyphenols and pectins.1–10 The importance, incidence and evolution of these compounds in the growing and ripening process have been under constant study in recent years.11–15 Due to the high number of variables which are used in fruit ripening studies,16–18 mathematical multivariate methods are needed to help us analyse and interpret the information that is generated.19–21 Current technological advances, especially in the field of microelectronics, have led to the manufacture of faster computers which simplify mathematical calculations, thus enabling the development of chemometrics as a multivariate analysis technique.22,23 In the present study, pattern recognition techniques were employed to establish discrimination rules that enable us to differentiate apples into technological and ripening groups from the values of certain chemical descriptors.24 These parameters were selected according to their relevance in the ripening process as well as for their capacity as quality indicators.All these studies will allow the cider producer to have more control over the raw material and to obtain higher quality products. Experimental Samples Traditional cider in Spain is made from the juice of a mixture of cider apples which have different sensory properties, the result being a natural, acidic juice.The apple varieties used in this study were grown in the Agricultural Experiment Station orchards in Villaviciosa (Asturias, North of Spain), a village which is well known for the excellent quality of its cider apples and derivatives. Four different types of apples, namely Collaos (mild sharp), Durón Arroes and Picona Rayada (sweet), Meana (bittersharp) and Raxao (sharp), were chosen in this study for their agronomical and technological suitability.Twenty-eight samples of these cider apple varieties were taken at different moments of their ripening process. Sample preparation The apple extracts were prepared using Richmond’s method25 with some modifications26 so as to ensure the complete extraction of the apple compounds. In order to achieve the said goal, 1 kg of apples were randomly chosen, of which a quarter was mashed into pieces and introduced into a mixer.The mixture was then covered with an adequate volume of ethanol and hydrochloric acid, reaching a final concentration of about 80% (m/m) and 0.1% (v/v), respectively. The fruit was crushed and vigorously mixed with these solvents for 2 min. The mush obtained was flow-back shaken in an inert atmosphere (nitrogen) for 2 h. The extract was subsequently filtered through a cellulose filter (Whatman, Maidstone, UK) and was washed with 80% (v/v) ethanol. The eluate was then reduced to a final volume of approximately 250 ml in a rotary evaporator at 37 °C.After this, the concentrate was diluted with 80% (v/v) ethanol to a final volume of 500 ml. Finally, the apple extracts were filtered through a 0.45 mm filter (Millex, Millipore Watford, UK) before analysis. Analytical procedures Sugars Fructose, glucose, sucrose and sorbitol were determined according to the method described by Blanco et al.26 by highperformance liquid chromatography using a Sugar Pak-I column (300 mm 3 6.0 mm id, 10 mm) and an aqueous solution of 50 mg ml21 of calcium salt of EDTA as mobile phase; flow rate, 0.5 ml min21; temperature, 90 °C. A refractive index detector was used as the detection system.Analyst, June 1998, Vol. 123 (1187–1191) 1187Malic acid This component was analysed using the following chromatographic method optimised by Blanco et al.:27 Spherisorb column (ODS-2 250 mm 3 4.5 mm id, 3 mm); mobile phase, phosphate buffer, pH 2.25 and 1022 m ionic strength; flow rate, 0.5 ml min21; temperature, 36 °C and UV detection at 206 nm.Polyphenols An FIA (flow injection analysis) system was used in the determination of polyphenols according to the method described by Mangas et al.,28 the following configuration being used: carrier, Folin Ciocalteu reagent (1/100) in a 1% Na2CO3 solution; carrier flow rate, 0.2 ml min21; reagent, 1% Na2CO3 solution; reagent flow, 0.6 ml min21; channel dimension, 300 3 0.5 mm, 300 3 0.7 mm; process temperature, 25 °C; colorimetric detection at 673 nm; injection volume, 40 ml.Pectins The different pectin fractions [water-soluble pectins (WSP), chelator-soluble pectins (CSP) and hydrochloric acid-soluble pectins (HASP)], were estimated as total galacturonic acid by incubation with sulfuric acid and subsequent colour development with alkaline m-hydroxydiphenyl.29 Reagents All the reagents, solvents and standards that were used were of analytical quality (99% minimal purity) and were supplied by Fluka (Buchs, Switzerland), Merck (Darmstadt, Germany) and Sigma-Aldrich (Madrid, Spain).Results and discussion Data processing The data was processed by means of the PARVUS statistical package.30 A matrix was constructed with rows (28) representing apple samples and columns (9) corresponding to sugars (sucrose, glucose, fructose), sorbitol, malic acid, total polyphenols and pectins (water-soluble pectin, chelator-soluble pectin and hydrochloric acid-soluble pectin). Samples were categorised as Ripe (R, 9) and Unripe (U, 19) according to the starch–iodine test criterion.31 The data were subjected to autoscaling before statistical analysis.Univariate analysis Table 1 shows the concentration of the variables monitored during the ripening of the apples. Univariate characterisation of the apples was carried out on the basis of Fisher’s weight (FW). Fructose and water-soluble pectin were the two most discriminant variables (FWfructose : 1.31; FWwater-soluble pectin : 1.02).However, the use of the most discriminant variables, fructose, did not allow us to differentiate between these two categories (see Fig. 1), so multivariate analysis was needed. Factor analysis A principal component analysis (PCA) was performed in order to establish the relationship between variables and observations, as well as to recognise the data structure. The application of this display method meant that we were able to observe the data structure using three principal components which accounted for 75.7% of the variance.An oblique axis rotation facilitates the interpretation of the data. The non-orthogonal varivectors were obtained from a linear Table 1 Concentration (g kg21) of sugars, total polyphenols, pectin fractions and malic acid in apple cultivars Apple Class S G F So TP WSP CSP HASP M C1 U 9.9 19.7 46.0 4.57 3.46 0.11 0.24 3.00 3.40 C2 U 8.3 18.8 42.9 3.51 2.89 0.06 0.10 3.48 2.45 C3 U 13.9 17.1 45.1 3.38 2.97 0.08 0.14 5.64 3.42 C4 U 15.4 16.7 43.9 3.06 2.66 0.06 0.15 4.40 2.61 C5 R 15.0 18.0 44.0 3.56 2.80 0.09 0.19 4.31 1.81 C6 R 18.7 19.1 46.8 3.93 3.25 0.28 0.46 4.13 1.28 D1 U 7.3 13.5 33.3 4.41 3.23 0.03 0.17 4.51 3.96 D2 U 11.7 12.5 35.6 4.53 3.15 0.03 0.10 2.92 2.78 D3 U 11.7 12.0 36.3 13.99 2.60 0.03 0.12 3.17 2.70 D4 U 12.7 13.4 34.8 4.38 2.75 0.03 0.08 2.45 3.72 D5 U 18.3 15.5 38.3 6.89 1.87 0.04 0.20 5.08 2.31 D6 U 24.0 15.1 36.2 4.87 2.66 0.04 0.18 5.52 2.51 D7 U 30.1 17.4 41.1 7.31 3.13 0.08 0.20 5.91 1.71 M1 U 21.4 6.5 41.3 4.49 4.16 0.08 0.14 3.90 5.65 M2 U 27.2 11.9 49.7 3.02 4.22 0.04 0.08 3.81 4.90 M3 U 26.6 10.0 47.8 2.61 3.69 0.06 0.17 4.17 3.53 M4 U 33.0 6.4 47.7 3.02 3.39 0.06 0.19 4.18 4.50 M5 R 29.6 11.0 56.6 3.65 4.15 1.11 1.26 7.28 3.31 P1 U 9.1 17.8 56.2 5.44 3.30 0.27 0.30 16.09 1.74 P2 U 8.0 21.7 49.8 2.43 1.92 0.15 0.43 18.42 4.60 P3 R 15.8 20.6 69.0 5.00 2.82 0.18 0.24 13.60 2.43 P4 R 16.1 20.1 75.0 5.73 3.32 0.91 0.62 18.18 2.14 P5 R 9.5 15.9 56.7 5.53 3.38 1.29 0.55 12.54 1.72 R1 U 7.4 26.5 50.0 3.86 2.59 0.13 0.23 8.43 5.37 R2 U 10.7 22.5 48.1 3.59 2.40 0.15 0.30 13.47 3.62 R3 R 15.1 22.4 63.0 6.38 3.03 0.11 0.10 6.97 2.14 R4 R 11.4 20.9 49.5 3.29 2.17 0.25 0.57 8.53 5.66 R5 R 9.4 20.3 50.0 3.58 2.01 0.49 0.58 7.48 3.41 Abbreviations used for variables: S, sucrose; G, glucose; F, fructose; So, sorbitol; TP, total polyphenols; WSP, water-soluble pectin; CSP, chelatorsoluble pectin; HASP, hydrochloric acid-soluble pectin; M, malic acid.Abbreviations used for apple samples: R, Raxao; M, Meana; P, Picona Rayada; C, Collaos; D, Dur�on Arroes. The subindex shows the degree of ripening within each variety, the highest subindex corresponding to the ripest apples. U, unripe class; R, ripe class. 1188 Analyst, June 1998, Vol. 123transformation of the original orthogonal factors. As can be seen in Fig. 2, where samples and variables are projected onto the plane of the first two factors, the said factors differentiate the apple varieties on the basis of their degree of ripening.Thus, the ripest apples (higher subindex) have, in general, smaller scores for the first factor than unripe ones. The variables most correlated with the first factor were fructose and pectins. Consequently, we can establish that the processes of fructose accumulation and apple softening are very important stages in fruit senescence when determining the degree of ripening.We can also see in this representation how the samples are reasonably well structured in variety groups. At the same time, the technological characteristics of the apples are visualised when the observations are projected onto the map formed by the second and third varivectors. As can be seen in Fig. 3, bittersharp apple varieties, such as Meana, had lower scores for the second component and negative values for the third varivector.On the other hand, more acidic apples, such as Raxao, had positive scores for the second principal component and negative values for the third varivector. Sweet and mild sharp cultivars, such as Picona Rayada, Durón Arroes and Collaos, did not exhibit a differentiated distribution when they were projected onto this map (Fig. 3). In this figure, it can also be seen that the second varivector is close to phenolic compounds (negative correlation) whereas malic acid is correlated with the third varivector (negative correlation). Hence, we may state that PCA has allowed us to visualise the data structure, since apple cultivars were reasonably well grouped in accord with their technological characteristics and degree of ripening.Linear discriminant analysis (LDA) LDA is a supervised method that can be used for visualising the data contained in complex data-bases. Mathematical decision rules obtained from a training set are used to classify unknown samples.As can be seen in Table 2, 92.9% of correct predictions were obtained using an internal cross-validation method with three groups for cancellation. A basic problem in LDA is deciding which variables should be included in the analysis. This may be achieved using Wilks’ lambda (l) as selection criterion, and an F statistic to determine the significance of the changes in lambda when a new variable is tested. Generally, values of 3.84 as F-to-enter and 2.71 as Fto- remove are used, which correspond to a confidence level of 90%.In accord with the results obtained in the univariate treatment, fructose and water-soluble pectin were the most relevant variables. The Wilks’ lambda value obtained was 0.49, which means that 51% of total variance is explained by within-group differences; 82.1% of correct predictions were obtained using these two variables to perform the cross-validated LDA method. Soft independent modelling of class analogy (SIMCA) The SIMCA technique is used for constructing an enclosure for each category using a principal component model.A reduced model with four significant principal components, obtained from a single-cross full validation, was employed, which accounted for 88.2% of the variance for the unripe model, and 95.8% of the variance for the ripe model. The first and second class errors (a and b), related to model sensitivity (1-a) and specificity (1-b), were also taken into account.As can be seen in Table 3, the unripe class is more sensitive and specific than Fig. 1 Box–Whisker plots for the fructose variable. U: unripe class; R: ripe class. Fig. 2 Projection of the variables and samples onto the plane formed by the two first factors. R, Raxao; M, Meana; P, Picona Rayada; C, Collaos; D, Dur�on Arroes. The subindex shows the degree of ripening within each variety, the highest subindex corresponding to the ripest apples. 1, Sucrose; 2, glucose; 3, fructose; 4, sorbitol; 5, total polyphenols; 6, water-soluble pectin; 7, chelator-soluble pectin; 8, hydrochloric acid-soluble pectin; and 9, malic acid.Fig. 3 Projection of the variables and samples onto the plane formed by the second and third factors. Abbreviations used, see Fig. 2. Table 2 Prediction matrix for the LDA method (three groups for cancellation) Assigned Category True Category U R Hits (%) U 18 1 94.7 R 1 8 88.9 Overall 92.9 Analyst, June 1998, Vol. 123 1189the ripe class.These results can be easily seen on a Coomans’ plot. As is shown in Fig. 4, three samples (marked with an arrow) are incorrectly classified (belonging to the unripe class); furthermore three outliers (marked with an asterisk) were detected, one of them being the misclassified, unripe sample. Partial least squares (PLS) The determination of the mathematical relationship between two variables groups (predictor variables and criterion variables), is carried out using multivariate regression techniques.In particular, the PLS method is specially recommended when the number of observations is small in relation to the number of variables. We established a binary answer (assigning value 1 to the unripe category and 2 to the ripe category), and then carried out a multivariate regression between this criterion variable (ripening) and the predictor variables (fructose, sorbitol, glucose, sucrose, malic acid, polyphenols, water-soluble pectin, chelator- soluble pectin, and hydrochloric acid-soluble pectin).The model constructed using the PLS regression consisted of four latent variables estimated by cross-validation with three groups for cancellation. The percentages of cross-validated explained variance, explained variance and correlation coefficient were 55, 66 and 71%, respectively. Fig. 5 shows two Box– Whisker plots for each caty using the PLS estimated value. As we can see, the PLS technique enables correct discrimination of apple varieties based on their degree of ripening.Conclusions The use of factor analysis and classification and modelling methods to analyse the data obtained by means of the analytical techniques of HPLC and FIA have enabled us to discriminate between apples on the basis of their technological characteristics and their state of ripening. Whereas fructose and pectins were found to be the most significant variables in the apple ripening process, malic acid and polyphenols enabled the technological characterisation of the apple varieties.This work was made possible by financial support from the CICYT (ALI 92-1027-C03). References 1 Pal, D. K., and Kumar, P. S., J. Hortic. Sci., 1995, 70, 569. 2 Karaoulanis, G. D., and Dilley, D., Int. J. Refrig., 1993, 16, 364. 3 Wu, Q. D., Szakacsdobozi, M., Hammat, M., and Harazdina, G., Plant Physiol., 1993, 102, 219. 4 Gussman, C. D., Goffreda, J. C., and Gianfagna, T. J., Hortscience, 1993, 28, 135. 5 Percy, A. E., Melton, L. D., and Jameson, P. E., Plant Sci., 1997, 125, 31. 6 Mayr, U., and Treutter, D., Erwerbsobstbau, 1996, 38, 164. 7 Fischer, M., and Amado, R., Carbohyd. Polym., 1994, 25, 161. 8 Fischer, M., Arrigoni, E., and Amado, R., Carbohyd. Polym., 1994, 25, 167. 9 Percy, A. E., Obrien, I. E. W., Jameson, P. E., Melton, L. D., Macrae, E. A., and Redgwell, R. J., Physiol. Plantarum, 1996, 96, 43. 10 Garriz, P. I., Alvarez, H. L., and Bartusch, A. M., Turrialba, 1995, 45, 101. 11 Chen, H., Duprat, F., Grotte, M., Loonis, D., and Pietri, E., J. Texture Stud., 1996, 27, 123. 12 Yamada, H., Ohmura, H., Arau, C., and Terui, M., J. Am. Soc. Hortic. Sci., 1994, 119, 1208. 13 Autio, W. R., Hayden, R. A., Micke, W. C., and Brown, G. R., Fruit Varieties J., 1996, 50, 45. 14 Murata, M., Tsurutani, M., Tomita, M., Homma, S., and Kaneko, K., J. Agric. Food Chem., 1995, 43, 1115. 15 Larrigaudiere, C., and Vendrell, M., Sci. Hortic., 1993, 55, 263. 16 Golias, J., and Bataille, B., Facultas Horticulturae, 1994, 9, 69. 17 Brackmann, A., Streif, J., and Bangerth, F., Gartenbauwiessenschaf., 1995, 60, 1. 18 Brackmann, A., Streif, J., and Bangerth, F., Gartenbauwiessenschaf., 1994, 59, 252. 19 Dever, M. D., Cliff, M. A., and Hall, J. W., J. Sci. Food Agric., 1995, 69, 329. 20 Defernez, M., Kemsley, E. K., and Wilson, R. H., J. Agric. Food Chem., 1995, 43 109. 21 Defernez, M., and Wilson, R. H., J. Sci. Food Agric., 1995, 67, 461. 22 Massart, D. L., Vandeginste, B. G. M., Deming, S. N., Michotte, Y., and Kaufman, L., Chemometrics: a Textbook, Elsevier, Amsterdam, 1988. 23 Martens, H., and Naes, T., Multivariate Calibration, John Wiley & Sons, Chichester, 1989. 24 Seiden, P., Bro, R., Poll, L., and Munck, L., J. Agric. Food Chem., 1996, 44, 3203. 25 Richmond, M. L., Brandao, S. C. C., Gray, J. I., Markais, P., and Stine, C. M., J. Agric. Food Chem., 1981, 29, 4. 26 Blanco, D., Gutiérrez, M. D., and Mangas, J. J., Chromatographia, 1988, 25, 701. 27 Blanco, D., Morán, M. J., Gutiérrez, M. D., and Mangas, J. J., Chromatographia, 1988, 25, 1054. Table 3 Sensitivities (S), specificities (Sp) and classification capacity of the SIMCA technique Explained Model variance (%) S (%) Sp (%) Hits (%) U 88.2 89.5 88.9 84.2 R 95.8 88.9 78.9 100.0 Fig. 4 SIMCA analysis. Coomans’ plot. (*), Outlier; (–), misclassified; U; unripe class; R, ripe class. Fig. 5 Box–Whisker plots for the PLS estimated value. U, Unripe class; R, ripe class. 1190 Analyst, June 1998, Vol. 12328 Mangas, J. J., Suárez, B., and Blanco, D., Z. Lebensm. Unters. Forsch., 1993, 197, 424. 29 Mangas, J. J., Dapena, E., Rodríguez, M. S., Moreno, J., Gutiérrez, M. D., and Blanco, D., Hortscience, 1992, 27, 328. 30 Forina, M., Leardi, R., Armanino, C., and Lanteri, S., PARVUS. An Extendable Package of Programs for Data Exploration, Classification and Correlation, Elsevier, Amsterdam, 1988. 31 Blanco, D., Morán, M. J., Gutiérrez, M. D., Moreno, J., Dapena, E., and Mangas, J. J., Z. Lebensm.-Unters. Forsch., 1992, 194, 33. Paper 7/08534F Received November 26, 1997 Accepted March 24, 1998 Analyst, June 1998, Vol. 123 1191

ISSN:0003-2654    

DOI:10.1039/a708534f

出版商:RSC    

年代:1998

数据来源: RSC

摘要:

Quantitative analysis of overlapped Raman spectra by target factor analysis and evolving factor analysis S. �Sa�si�c Faculty of Physical Chemistry, University of Belgrade, Studentski trg 16, POB 137, 11001 Belgrade, Yugoslavia A system of overlapped Raman spectra was formed on the basis of three experimentally recorded basic bands and quantitatively analysed by two statistical methods: target factor analysis (TFA) and evolving factor analysis (EFA). The results were compared with one another and with the true values.For intense bands, the values obtained by both methods were in good agreement with the expected values, while for small shoulders of strong bands quantitative analysis by EFA apparently gave better results than TFA. Keywords: Raman spectroscopy; target factor analysis; evolving factor analysis Factor analysis is a powerful method for determining the number of independent bands in a system of overlapped spectra. The ability of this method to determine the number of constituent bands in overlapped Raman spectra has been presented in studies by Shurvell and co-workers.1–6 Mann and co-workers7,8 have used factor analysis for structural studies of crystalline linear polyethylenes.However, in addition to determining the number of independent components, it is possible to carry out a quantitative analysis of the spectra by additional procedures. By the routines known as target factor analysis (TFA)9–11 and evolving factor analysis (EFA),12–14 in addition to determining the number of components, it is possible to find the concentration distribution of these components across the investigated mixtures.In addition to these two methods, there are several other methods of factor analysis by which the components can be quantitatively determined. Some of these rely on EFA, such as rank annihilation evolving factor analysis15 (RAEFA), window factor analysis16 (WFA), fixed size window evolving factor analysis17 (FSW-EFA) and eigenstructure tracking analysis.18 There is also generalized rank annihilation factor analysis19 (GRAFA), a method for analysing individual components in a mixture, and SIMPLISMA, 20 a method for self modelling mixture analysis. Only TFA and EFA are included in the present work, because they are simple and based on ideas easily comprehended by Raman spectroscopists.These methods have not apparently been used in quantitative investigations of Raman spectra, where band deconvolution is the most frequently used technique.To our knowledge there is only one example of calculation of equilibrium constants from Raman spectra: that of zinc bromide complexes in dimethyl sulfoxide, using a method called factor analysis with equilibria constraints,21,22 which demands some initial values for stability constants in the investigated system. The scarce application of factor analysis in Raman spectroscopy can be attributed to the significantly lower intensity and higher background and noise in Raman spectra compared with absorption spectra, which are frequently analysed by factor analysis.The advantage of Raman spectra could be their simplicity. Systems in which quantitative analysis is based on interpreting Raman spectra within the spectral range of a characteristic vibration rarely feature more than four bands. Many chromatographic data analysed by factor analysis are substantially more complex than any Raman data can ever be.One of the aims of this paper is to encourage the application of quantitative FA in Raman spectroscopy, regardless of the fact that the general features of the spectra used here (band intensity ratios) are rarely met in laboratory practice. In order to illustrate the capability of TFA and EFA, an analysis was carried out of a system consisting of 19 overlapped Raman spectra, where only three pure component bands are experimentally recorded and digitized, while all other spectra are obtained as a numerical combination of exactly known fractions of each of the three component bands.Hence, the result of the analysis is known in advance, allowing for a reliable estimate of the quality of the analysis and a mutual comparison of these two methods. The results obtained are completely valid for any system of spectra similar to that used here. The fact that Raman spectra, and not some other kind, were used in the model is of no relevance.Experimental Pure component bands were recorded in the spectral region 250–350 cm21. These were bands of dichloromethane at 285 cm21 (component I), 1,2-dichloroethane at 301 cm21 (component II) and carbon tetrachloride at 314 cm21 (component III), all obtained from Sigma. All the bands were excited using the 514.5 nm excitation line of a Spectra-Physics Model 2020 Ar+ ion laser and recorded with a Spex 1401 spectrometer. The power of the incident beam was 200 mW and the slit-width corresponded to 4 cm21.The remaining 16 spectra are a combination of these three bands. The fractions of every component band in the analyzed spectra are listed below the results of the analysis for each component in Table 1. In this way, a system of 19 spectra obtained from 19 samples is synthesized, hereafter referred to as ‘mixtures’, whose compositions are given in Table 1. Some of these spectra are shown in Fig. 1. It should be noted that fractions of components vary regularly, which is an important condition for applying factor analysis. Prior to treatment, the baselines are corrected and the spectra normalized to the same area.The data are processed by programs for factor analysis, TFN and EFN, developed in our laboratory. Results and discussion The experiment in which Raman spectra are recorded from several mixtures, with regularly varying concentrations of a few scattering components, can be mathematically presented as D = A · C (1) where D is the experimental matrix in which every column is one recorded spectrum, A is a matrix where each column corresponds to a spectrum of a pure component, and C is a Analyst, June 1998, Vol. 123 (1193–1197) 1193matrix where each row represents the varying intensity as a function of mixture composition. Factor analysis can be performed on such a system. The first step is to find the eigenvectors of both matrices DTD and TDD (singular value decomposition). The eigenvectors obtained after removing noise from the second product represent, to a first approximation, the concentration distribution of components across mixtures, c0i , and eigenvectors obtained from the first product are first expressions of component bands, a0i .The eigenvectors corresponding to component bands are given in Fig. 2. They obviously have no physical meaning and TFA and EFA are actually methods for rotation of the matrix consisting of these eigenvectors and the matrix of concentration eigenvectors, which, analogously to bands in Fig. 2, are not real either, to physically justified shapes. TFA method The TFA method9,11 is based on the assumption that the component bands are known. Those bands serve as ‘targets’ (a1A, a2A, …, anA, where n is number of components) to which we try to rotate the first expression bands by the least squares method. In general, one could make an arbitrary choice of the component band, and the degree of similarity between the ‘target’ and the rotated eigenvector, a0i A, will show the validity of this assumption. In the present case the component bands are known and the rotated vectors are very close to the ‘targets’, as shown in Fig. 3: a1A Å a01 A, a2A Å a02A and a3A Å a03 A. The differences between the rotated vectors and components I and III are minimal, while for component II the difference is more pronounced. The reason could be the lower intensity of that band compared with the other two, and the fact that, in contrast to the other two bands, this component never appears alone.The matrix obtained, which rotates the first expressions of bands into bands similar to experimental bands, is inverted, and the concentration vectors are multiplied by this new matrix. The result for each component is listed in the second row of Table 1. The values obtainedery close to the true values, in agreement with the identity of the bands. The precision is notably better for higher fractions of components in overlapped bands.The discrepancies in these cases usually do not exceed 1%, except for component III, where they reach 3%. The precision of the results decreases with decreasing component ratio. The error gradually increases and reaches 10 or 20% for bands where fractions of components I and II are close to zero. For component III, errors are higher, clearly observable in mixtures 7–10. These errors are probably a consequence of a fault in the rotated band of component II.The fraction of Fig. 1 Spectra of mixture numbers 5, 9, 10, 11 and 14 (from top to bottom). Table 1 Compositions of mixtures and the results of the analysis Mixture No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 Component I— True values 1.000 0.950 0.930 0.900 0.850 0.800 0.600 0.500 0.400 0.350 0.150 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Found by TFA 0.995 0.945 0.925 0.891 0.839 0.787 0.580 0.476 0.377 0.327 0.127 20.02 20.01 20.01 0.000 0.000 0.003 0.003 0.004 Found by EFA 1.000 0.950 0.930 0.902 0.853 0.803 0.606 0.508 0.405 0.354 0.149 0.007 20.01 20.01 20.02 20.02 20.02 20.03 20.03 Component II— True values 0.000 0.000 0.000 0.100 0.150 0.200 0.400 0.500 0.500 0.500 0.550 0.450 0.350 0.300 0.200 0.100 0.000 0.000 0.000 Found by TFA 20.02 20.02 20.02 0.090 0.141 0.191 0.390 0.495 0.498 0.500 0.556 0.461 0.363 0.314 0.216 0.117 0.037 0.036 0.039 Found by EFA 0.001 0.001 0.001 0.101 0.151 0.201 0.400 0.500 0.500 0.500 0.549 0.449 0.349 0.299 0.199 0.099 0.000 0.000 0.000 Component III— True values 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.100 0.150 0.350 0.550 0.650 0.700 0.800 0.900 0.930 0.950 1.000 Found by TFA 0.004 0.004 0.004 20.01 20.02 20.03 20.07 20.09 0.012 0.063 0.259 0.482 0.603 0.664 0.785 0.907 0.956 0.973 1.030 Found by EFA 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.101 0.151 0.352 0.552 0.652 0.702 0.801 0.901 0.930 0.950 1.000 Fig. 2 First expressions of component bands. 1194 Analyst, June 1998, Vol. 123component II is highest in mixtures 8–12 and errors made in the rotation cause its fraction to increase at the expense of component III. This observation indicates that the method gives preference to strong bands and ‘minimizes’ low intensity bands. It can generally be concluded that the TFA method gives very good results for components that dominate in the spectra, but shows less success in the analysis of small shoulders of intense bands.EFA method The basic idea of EFA is to determine the ‘concentration windows’14 for the components investigated, outside which these components do not exist. This could be done by the analysis of eigenvalues of quadratic sub-matrices whose dimensions change from 2 to the total number of mixtures.12,13 We start with matrix tDD, where D consists of spectra of only the first two mixtures, and add spectra of other mixtures, one by one, to D, increasing the dimensions of tDD.At the beginning there is only one eigenvalue, significantly above a threshold level. When a new component in this process appears, another eigenvalue emerges. This is repeated for each new component. In Fig. 4(a), a logarithmic plot of eigenvalues is presented versus the number of mixtures, going from pure component I to component III. If we change the order of mixtures, going now from mixture 19 to mixture 1, a new plot of eigenvalues appears, as shown in Fig. 4(b). The plots in Fig. 4(a) and (b) can be interrelated by joining some of the points within them. The Ith line in the plot in Fig. 4(a) should be inter-related to the (N + 1 2 I) line in the plot in Fig. 4(b),14 where N is the total number of components, by which the sequence of existence of all components with respect to the number of mixtures is obtained. Actually, such a process joining the points in this case is straightforward only for component II. For components I and III the reasoning is as follows.From the plot in Fig. 4(a) it can be seen that component I exists in the first few mixtures, while from the plot in Fig. 4(b) it is observable that it appears in mixture number 11. This means that component I appears in mixtures 1 through 11. For the same reason component III is present in mixtures 9–19. The results of this analysis are shown in Fig. 4(c). This is a very useful plot, because we can now say that the concentration vector, say of component I, obtained by decomposition of tDD dimension 19, should be rotated so that in mixtures 11–19 it equals zero.Hence, we obtain for all components a homogeneous system of equations with nontrivial solutions. The solutions give a rotation vector, ri, for each component. These vectors form the columns of the rotation matrix R, which transforms the concentration vectors into vectors proportional to the real concentration distribution, column vector ci real, via a numerical constant ki.Creal Å C·R (2) where the ith column of R is now given as ki ri and Creal is a matrix with a real concentration distribution of the components. The next step is the inversion of the matrix R and multiplying the inverted matrix by the first expression component bands. The result of this operation are bands proportional to the real bands (Fig. 5). From Fig. 5 it is clear that the factor of proportionality can easily be found by dividing the area of the experimental bands by the area of the calculated bands, or by comparing the peaks.Fig. 3 Comparison of experimental component bands (circles) and the rotated vectors (solid lines) for components I, II and III. Fig. 4 (a) Plot of eigenvalues versus solutions in the forward order. (b) Same dependence as in (a) calculated in the opposite order. (c) Sequence of existence of components I (8), II (2) and III (½). Fig. 5 Comparison of component bands (circles) with EFA bands (solid lines).Analyst, June 1998, Vol. 123 1195This is valid for all three components, since we knew in advance the band of component II. Therefore, in general, for components that do not appear alone, finding the ratio between the experimental and the calculated bands is not so straightforward. It is necessary to do a deconvolution and extract the component band. Another problem is to determine the concentration of the component. This could be resolved if the concentrations of the other components that appear alone in the system were known.Since EFA can be applied only to a system with an evolution process, it is always possible to start or end the process with a species of known concentration. In the present case the bands calculated for components I, II and III should be multiplied by 22.615, 1.335 and 21.283, respectively, to obtain the real bands. The concentration vectors are multiplied by reciprocal values of these numbers and the final concentration distribution of components found by EFA is given for all components in the third row of Table 1.For component I, the errors practically do not exceed 1% for all mixtures in which the ratio of this component is greater than zero. In mixtures 12–19, the errors increase moderately, reaching a maximum with mixture No. 19. For components II and III, the results are better. For both components and all the mixtures the differences between the true and the obtained values are 1 or 2 on the third decimal place.These higher discrepancies in the results for component I compared with the results for components II and III are a consequence of solutions of the system of equations that are more compact for component III (there is only one solution for component II), while for component I it is not compact: the solutions are dispersed within an interval, which causes faults in some mixtures. A comparison of the results obtained by EFA and TFA suggests that EFA is a more reliable technique.For all components and for all solutions the EFA results are closer to the true values. Owing to the very precisely determined concentration windows of the components, the numerical method used in EFA shows an advantage over the least squares method of minimized deviations between the real and the calculated bands. In addition, it seems that TFA is inherently weak in quantitative treatment of small bands. In calculating the acetone–phenol complex constant in a binary mixture,23 it was found that differences in the areas of bands obtained by TFA and deconvolution (which yields more reliable data in this case) are largest at low acetone or phenol concentrations. The inaccuracy of the rotation of the eigenvector that corresponds to component II is even larger if matrix D is composed of experimentally recorded spectra of mixtures with significant differences in intensities.24 These data indicate that the overall intensity distribution and relative contributions of the investigated components in the initial matrix play a major role in TFA. In EFA, however, finding the existence domain of the system components is of crucial significance.This is done by the analysis of eigenvalues of a series of sub-matrices in which the most important step is defining the eigenvalue level that corresponds to noise.25 The success of EFA, in essence, depends on whether the system can be analysed by factor analysis or not.For sufficiently large differences between components and noise, the differences in eigenvectors and eigenvalues are obvious, while, if that condition is not met, experimental judgement decides. This intervention is, however, not of such magnitude as in TFA, which includes the interaction of the experimenter with the system, due to target vector choice. In the model presented here, eigenvalues that correspond to noise were several orders of magnitude lower than those that belong to components, and were not plotted in Fig. 4. Generally, in a system similar to the present one, EFA yields better results, quantitatively and qualitatively. This method is also advantageous due to the possibility of obtaining correct concentration vectors on the basis of system concentration characteristics, and then determining component bands without any previous assumptions. Hence, deconvolution is avoided, which is inevitable in TFA for bands that do not appear alone, while comparison of calculated bands with experimental bands, wherever possible, is an additional criterion for evaluating the quality of the analysis. On the other hand, the TFA method is more intuitive, and in Raman spectroscopy, especially of inorganic solutions, it is often easy to choose the target vectors (although not all of them).The TFA procedure is fairly fast and simple. After defining the target vectors it is easy to rotate the experimental matrix, while the quality of rotation can be evaluated visually, in addition to the empirical criteria.11 Nevertheless, owing to the high quality of spectra (intensity interval was from 200 to 10 000 counts s21 over a band) and constant peak positions, both methods gave very good results in analysis of this spectral model.They could be improved, especially TFA, by an iterative process in which a simultaneous variation of spectra and concentrations is performed until convergence is achieved.Conclusion A system of 19 overlapped spectra containing three basic bands was successfully quantitatively analyzed by EFA and TFA. The overall result obtained by EFA is better than by TFA, particularly for low intensity bands. The maximum errors in EFA are 1% of the true values for component I and even smaller for the other two components. TFA gives poorer results with serious errors for component III in some cases. This error is ascribed to the lower sensitivity of this method and the error in rotation of component II.An iterative treatment, which would remove the error in the rotated vector of component II, would improve the results of this method. This work was supported by a grant from the Research Fund of Serbia. References 1 Ng, J. B., and Shurvell, H. F., J. Phys. Chem., 1987 91, 496. 2 Ng, J. B., Shurvell, H. F., and Petelenz, B., Can. J. Chem., 1988, 66,1912. 3 Rintoul, L., and Shurvell, H. F. J. Raman Spectrosc., 1990, 21, 501. 4 Ng, J. C. F., Park, Y. S., and Shurvell, H. F., Spectrochim. Acta, 1992, 48A, 1139. 5 Ng, J. C. F., Park, Y. S., and Shurvell, H. F., J. Raman Spectrosc., 1992, 23, 229. 6 Pemberton, R. S., and Shurvell, H. F., J. Raman Spectrosc., 1995, 26, 373. 7 Shen, C., Peacock, A. J., Alamo, R. G., Vickers, T. J., Mandekern, L., and Mann, C. K., Appl. Spectrosc., 1992, 46, 1226. 8 Shen, C., Vickers, T. J., and Mann, C. K., Appl. Spectrosc., 1992, 46, 772. 9 Malinowski, E. R., Anal. Chim. Acta, 1978, 103, 339. 10 Malinowski, E. R., and McCue, M., Anal. Chem., 1977, 49, 284. 11 Malinowski, E. R., Factor Analysis in Chemistry, Wiley, New York, 2nd edn., 1991. 12 Gampp, H., Maeder, M., Meyer, C. J., and Zuberbuehler, A. D., Talanta, 1985, 32, 1133. 13 Gampp, H., Maeder, M., Meyer, C. J., and Zuberbuehler, A. D., Talanta, 1986, 33, 946. 14 Maeder, M., Anal. Chem., 1987, 59, 527. 15 Gampp, H., Maeder, M., Meyer, C. J., and Zuberbuehler, A.D., Anal. Chim. Acta, 1987, 193, 287. 16 Malinowski, E. R., J. Chemom., 1992, 6, 29. 17 Keller, H. R., Massart, D. L., and De Beer, J. O., Anal. Chem., 1993, 65, 471. 18 Cuesta Sanches, F., Toft, J., Kvalheim, O. L., and Massart, D. L., Anal. Chim. Acta, 1995, 314, 131. 19 Sanches, E., and Kowalski, B. R., Anal. Chem., 1986, 58, 496. 1196 Analyst, June 1998, Vol. 12320 Windig, W., and Guilment, J., Anal. Chem., 1991, 63, 1425. 21 van Heumen, J., Ozeki, T., and Irish, D. E., Can, J. Chem., 1989, 67, 2030. 22 Ozeki, T., Kihara, H., and Hikime, S., Anal. Chem., 1987, 59, 945. 23 �Sa�si�c, S., and Kuzmanovi�c, M., J. Raman Spectrosc., in the press. 24 �Sa�si�c, S., unpublished work. 25 Darj, M. M., and Malinowski, E. R., Anal. Chem., 1996, 68, 1593. Paper 8/00019K Received January 2, 1998 Accepted March 24, 1998 Analyst, June 1998,

ISSN:0003-2654    

DOI:10.1039/a800019k

出版商:RSC    

年代:1998

数据来源: RSC

摘要:

Statistical method to evaluate clean-up procedures in polychlorinated biphenyl analysis M. C. Pietrogrande*a, D. Ghedinia, G. Veladab and F. Dondia a Department of Chemistry, University of Ferrara, Via L. Borsari, 46, I-44100 Ferrara, Italy b Department of Analytical Chemistry, Faculty of Chemistry, Santiago de Campostela, Spain The clean-up procedure involved in the trace analysis of polychlorinated biphenyls (PCBs) was evaluated by studying GC–MS chromatograms of PCB mixtures after the subsequent steps of the procedure.The complex chromatograms obtained were evaluated using two methods: the individual PCB congener method and a chemometric approach based on a study of the autocovariance function. The results obtained are statistically comparable, proving that the statistical approach is able to determine the total PCB content in the sample quantitatively. Moreover, since the autocovariance method is based on statistical evaluation of the whole complex chromatogram, it can overcome the problems that usually arise because of peak overlapping in PCB mixture chromatograms.It also provides accurate results when the chromatogram shows interfering peaks and low resolution. This is the case with Aroclor 1242, where some compounds released by the cartridges strongly interfere with the analysis, resulting in errors in the quantification of individual PCBs. Highly chlorinated PCBs (i.e., Aroclor 1260) can be quantitatively recovered (mean recovery, 100 ± 1%).In contrast, lower proportions (less than 75%) of the less chlorinated compounds (i.e., those containing two and three chlorine atoms) are recovered because they are selectively retained by silica and alumina columns. Keywords: Polychlorinated biphenyl analysis; clean-up procedure; chemometrics; multicomponent chromatograms; procedure evaluation Polychlorinated biphenyls (PCBs) are ubiquitous environmental pollutants world-wide: they can be found in soils, oils, sediments, waters, animal tissues and airborne particulates.The determination of PCBs in various complex matrices is mandatory for environmental monitoring, and for studying their toxic effects and biodegradation processes. The determination of trace levels of these organic compounds is a complicated procedure, consisting of many steps—sample homogenisation, extraction from the matrix, clean-up and concentration, gas chromatographic separation and detection—each of which can significantly contribute to the total error in the final determination. 1–9 The clean-up procedure has proved fundamental in removing interferences and increasing the accuracy and precision of the final result.1,9 Co-extracted compounds can interfere with the final determination of PCBs, since contamination can overload a high-resolution gas chromatographic (HRGC) column, create negative peaks or an erratic response when electron-capture detection (ECD) is used, or can lead to co-elution with PCBs, resulting in misidentification and incorrect determination.1 Several methods have been proposed to remove interfering, coextracted compounds.In particular, solid-phase extraction (SPE) has been widely employed with various chromatographic phases such as silica, Florisil, alumina and graphitic carbon. Moreover, the availability of inexpensive commercial cartridges for SPE has prompted widespread application to various complex matrices.1,3,6 Considerable attention has recently been paid to the quality of analytical results—in particular when they are derived from multi-step procedures—in order to ensure that the chemical measurements are comparable and accurate.A number of interlaboratory studies have been organised with the following objectives: to determine variations in PCB determinations; to identify what sources cause these variations; and to reduce them through a step-by-step learning process.2–8 The sources of systematic, random errors produced by the various steps in the method can be evaluated by following the traceability chain of the analytical procedure in detail.10–12 In particular, the different validation methods determine the loss of selected target compounds (single PCB congener method) and possible interferences introduced in the subsequent analysis steps.Current methods are based on area or height determination of selected target peaks present in the chromatogram.However, these quantities can be affected by several biasing factors: mutual overlapping of single component peaks, column efficiency degradation and sample overloading. In this paper, an alternative approach is presented for evaluating the clean-up procedure, using SPE with silica and alumina cartridges. In essence the following properties of the chromatographic separation are estimated: (1) the number of single components (SCs), m, present in the mixture analysed; (2) their relative abundance compared with an internal standard; and (3) the peak capacity, nc, i.e., a check of column performance and any degradation.In this manner, each step of the clean-up procedure can be completely evaluated by controlling: the variation of m, the mean loss, or increase, in PCB concentration, and the efficiency of the separation system, affected either by any degradation in column efficiency or by sample overloading. The method consists of a chemometric approach to evaluate multicomponent chromatograms, whenever there is a high degree of overlapping.13–16 Distinct features of the method, based on a study of the autocovariance function,17–25 are its ability to determine the mean number of single components (m) which is usually greater than the number of observed peaks, because of the severe peak overlapping present in multicomponent chromatograms.13 The method was tested by simulation20 and by application to test mixtures:21,22 it was found to give unbiased estimation of both the peak capacity of the separation and the single component number of the analysed mixture.In the present paper, the simplified graphical version is applied,23–25 which does not require complex mathematics: it gives estimates of comparable precision and accuracy and it was also shown to be able to detect column overloading effects.23 The aim of the present study was to test how the individual steps of a multi-step procedure affect the results of the final determination; this goal may be achieved by defining a method for a rapid and accurate estimation of multicomponent chromatograms.The final purpose of the proposed method was its application to real environmental samples; here, a preliminary study is reported referring to a simple synthetic sample. Analyst, June 1998, Vol. 123 (1199–1204) 1199Experimental Reagents and apparatus Aroclor 1242 and 1260 standard mixtures were obtained from Alltech Italia (Milan, Italy).Commercially available chromatographic cartridges of silica and neutral alumina were also supplied by Alltech Italia. The chromatograph used was a Mega Series 5160 fitted with a 30 m 30.25 mm id column coated with a 0.25 mm film of DB- 5 (J & W Scientific, Rancho Cordova, CA, USA); it was equipped with a QMD1000 quadrupole mass spectrometer (Fisons Instruments, Milan, Italy). Analytical methodology Aroclor 1242 and 1260 standard mixtures (2 ml of 50 ppm solutions, samples 1242St and 1260St) were first eluted in silica cartridges with 4 ml of hexane (samples 1242Si and 1260Si, respectively), the solution was reduced to a volume of 2 ml in a Kuderna–Danish apparatus and then eluted in alumina cartridges with 4 ml of hexane–dichloromethane (98 + 2, v/v) (samples 1242Al and 1260Al, respectively). The same procedure was carried out on cartridges pre-washed with 3 ml of hexane before use (samples 1242Si pw, 1242Al pw, 1260Si pw, 1260Al pw).Solutions of Aroclor 1242 (50 ppm) in hexane and hexane–dichloromethane (98 + 2, v/v) were concentrated in a Kuderna–Danish apparatus (from 6 to 2 ml) and the samples 1242Hex and 1242CH2Cl2, respectively, were obtained. A standard solution of octachloronaphthalene at 5 ppm was added to each sample prior to chromatographic analysis as an internal standard in order to normalise the total area of all chromatograms. Each mixture was submitted to a procedure three times so that three chromatograms were obtained for each sample: the reported values are mean values calculated on the replicate determinations.All the samples were analysed in the GC–MS system under programmed capillary GC conditions:22 after 1 min at 90 °C, the column temperature was increased from 90 to 160 °C at 25 °C min21, and then from 160 to 300 °C at 3 °C min21; the final temperature of 300 °C was maintained for 30 min. Sample volumes (2 ml) were injected with splitting: the split ratio was 20 : 1, and the splitless time was 1 min.The electron impact (EI) mass spectra were recorded, and the total ion current (TIC) chromatograms were automatically converted by an ADC (with Dt = 1 s) and stored in the GC–MS system. Obviously, a GC–MS analysis under selected ion monitoring (SIM) conditions may be much more selective than TIC detection:22 in this paper, TIC detection was exploited in order to investigate general effects of sample contamination and sample loss during the steps of the clean-up procedure. Procedure for chromatogram evaluation Study of the complex chromatogram The final step of PCB analysis always involves evaluation of a complex chromatogram since, in environmental matrices, PCB congeners usually occur as multicomponent mixtures of 40–50 congeners.1 It has been demonstrated13 that when a large number of single components (SCs) are present in a mixture, complex retention patterns, determined by peak overlapping, are always obtained even when HRGC systems are employed for separation.Quantitative analysis of PCB compounds in these complex chromatograms requires identification and quantification of some resolved peaks corresponding to single congeners (individual congener method): this determination is difficult, as well as time consuming, and can cause errors since it is not possible to achieve complete PCB separation.1,2,6 Autocovariance method Different statistical methods have been developed to evaluate the properties of multicomponent chromatograms.13–25 Among them, Dondi and co-workers proposed a procedure, based on a study of the use of the autocovariance function to estimate the attributes describing the chromatographic separation.17–25 The method was widely investigated:18,21 it was tested by using computer-simulated chromatograms and the validity of the results obtained was confirmed;22,23 also, the applicability to experimental chromatograms21,22,25 was verified. From the chromatogram, acquired in digitised form, the experimental autocovariance function (EACVF) can be directly calculated, according to the expression: C M Y Y Y Y k M j j k j N k ( ) ( �)( �) , , , ,....t = - - = - + = -å 1 0 1 2 3 1 1 (1) where Yj and Yj + k are the heights in the digitised chromatogram at the j and (j + k) retention time positions, respectively; k is the fixed interdistance used for correlation, N the number of points in the digitised chromatogram, M - 1 the maximum extension over which EACVF is calculated and �Y the mean calculated from the chromatogram as: � Y Y N i i N = å (2) The autocovariance function expresses the intercorrelation between peak heights within the chromatogram as a function of time span, t, i.e., the interdistance between subsequent chromatographic peaks, expressed either by k or by the equation: t = kDt (3) where Dt is the interdistance between subsequent sampled points. In this paper the value Dt = 1 s was used; the time range of interest for all the chromatographic separations was 1200 s; therefore, the number of points N on which C(t) was calculated was 1200.It has been demonstrated that the most significant information is contained in the first part of the C(t) plot; therefore, it was calculated in the interval 0 @ t @ 16 s, i.e., M = 32, since in the chromatograms studied s is approximately 2 s (see below). The autocovariance function can be plotted against the time span, t.Fig. 1 shows the EACVF plot calculated from the chromatogram of a PCB mixture, Aroclor 1260, reported in Fig. 1 Plot of the autocovariance function C(t) calculated for a PCB mixture [sample 1260St, Fig. 2(a)]: C(0) and d1/2 are evaluated from graphical inspection and parameters m and w1/2 are calculated [eqns. (4) and (5)]. 1200 Analyst, June 1998, Vol. 123Fig. 2(a). Such a plot appears as half of an approximately Gaussian peak, the properties of which proved to be related to two basic chromatographic separation parameters:23 m, the number of components present in the mixture and s, the mean standard deviation of the chromatographic peak of a single component.In particular: The value of EACVF at the origin, C(0) in Fig. 1, estimates m: m A a C d X T m m = + 2 2 2 1 2 1 0 2 129 ( ) ( ) . / s (4) where: s2 m/a2 m is the relative dispersion of heights of peak maxima, am the mean value of heights of peak maxima and s2 m the standard deviation of peak maximum distribution; AT is the total area of the acquired chromatogram (see below); X is the total time span in which the chromatographic separation is performed.The width, at half height, of the EACVF peak, d1/2 in Fig. 1, gives an estimate of s, the standard deviation of a pure single component peak, which is expressed in the chromatogram by the width at half the peak height: w1/2 = d1/21.414 (5) AT, the total area of the chromatogram, can be calculated from the digitised chromatogram according to the equation: A Y T i i N =å (6) where N is the number of points of the acquired chromatogram (here, N = 1200). From the two basic quantities, m and s, all the other chromatographic parameters can be determined so that all PCB mixture properties can be evaluated.21–24 It must be emphasized that the proposed method can only be applied when the SC peak shape is constant, this is usually the case with programmed elution techniques.Moreover, since the method is based on determining the width of an EACVF peak at half height, it is sufficiently robust towards moderate SC peak shape asymmetry as is usually found in temperature programmed HRGC analysis of PCBs.22–25 In addition, the method was able to detect column overloading effects.23 A basic parameter, describing the effectiveness of any chromatographic system, is the peak capacity, nc, which, for constant peak width conditions, is:19 n X R c S = 4s (7) nc expresses how many SC peaks the chromatographic system can separate at a given resolution, Rs, when they are sequential, with a constant peak standard deviation, s, over the chromatographic space, X.This parameter explains the performance of the separation system and can be determined and controlled during the procedure. In this paper an Rs value of 0.5—describing peak maximum separation—is assumed. The extent of separation achieved is expressed by g, the separation extent, i.e., the ratio of SCs appearing in the chromatogram as peaks: g = p m (8) Owing to peak overlapping, the number of peaks seen in the final chromatogram, p, is not the number m of components present in the mixture—i.e., when each peak is a pure peak formed by a single component (SC)—but only a proportion of them since, in addition to the pure peaks, there are others formed by two or more SCs.The mean area Am of a single component can be calculated as: A A m m T = (9) The total area AT expresses the total amount of PCBs present in the mixture, while Am represents the mean amount of each congener present in the mixture.This amount is really an average value, also accounting for variations in detector sensitivity. To make different chromatograms comparable, the chromatographic response (peak heights or areas) must be normalised. In this work, this was achieved by referring the AT values of all the chromatograms studied to the peak area of the internal standard (octachloronaphthalene, 5 ppm), which was added to all the samples immediately before the chromatographic analysis.This procedure increases the reproducibility of the measurements: the total area values of the normalised chromatograms exhibited a reproducibility of ±2%. Recoveation To evaluate the clean-up procedure, the PCB content in the original and cleaned-up mixtures was quantitatively calculated and compared. Two methods were used: Fig. 2 Chromatograms obtained from 1260 mixture (50 ppm) after the different steps of the clean-up procedure: (a) original sample (sample 1260St); (b) after elution on silica cartridge (sample 1260Si); (c) after subsequent clean-up on silica and alumina (sample 1260Al). Analyst, June 1998, Vol. 123 1201Individual PCB congener method.1 In the complex chromatograms obtained, some resolved peaks are selected as corresponding to single PCB congeners. This procedure is fairly complex and time consuming: in chromatograms of Aroclor 1242, where peak overlapping is severe, only ten congeners could be identified with absolute certainty, while for the 1260 mixture, 16 PCBs were found.The recovery of each PCB is calculated by comparing areas of the corresponding peaks in the original and cleaned-up samples: the recovery of each mixture is calculated as the mean of the values for the selected congeners. Autocovariance function method.17–25 From the digitised chromatogram, AT values are calculated and normalised relative to the area of the internal standard, octachloronaphthalene.Using the autocovariance method, the number of components present in the mixture, m, can be estimated according to eqn. (4) and from it the mean area, Am, is calculated [eqn. (9)]. Both AT and Am can be used to calculate the PCB content in the mixture by comparing them to corresponding areas (total and mean, respectively) in the standard mixtures (samples 1242St and 1260St): the values obtained were determined with a precision of ±2%.The percentage variation of AT and Am among the original sample and those eluted on cartridges indicates the percentage recovery during the clean-up step. Results and discussion PCB mixtures were analysed in a system acquiring a TIC chromatogram. The GC–MS chromatograms obtained from the original solutions (samples 1242St and 1260St) were compared with those obtained after silica clean-up (samples 1242Si and 1260Si) and after subsequent elution on alumina (samples 1242Al and 1260Al).Chromatograms obtained using this procedure for Aroclor 1260 are reported in Fig. 2: original sample (a), mixture after elution on silica cartridge (b) and after subsequent elution on alumina (c). These steps were performed on non-pre-washed cartridges. From the reported chromatograms it is evident that, when the mixture is eluted on a silica cartridge, nearly 15 peaks appear in the first part of the chromatogram (10–30 min) [Fig. 2(b)]. The same peaks are present at a higher concentration in the sample subsequently eluted on alumina [Fig. 2(c)]. They are also present in procedural blanks and chromatograms obtained from the 1242 mixture after clean-up. Moreover, these compounds severely interfere with the analysis of such a mixture, since they elute in the same chromatographic space as do the lowchlorinated PCBs contained in Aroclor 1242. In order to reduce this interference, cartridges were pre-washed with the solvent (3 ml of hexane) prior to use, as recommended by the supplier.1 The chromatograms obtained (1242Si pw, 1242Al pw, 1260Si pw, 1260Al pw) showed that pre-washing moderately reduces the number of interfering peaks.MS analysis identified the interfering compounds as alkenes and alkyl phthalates: it has been reported that these compounds are used in the poly- (propylene) housing and polyethylene frit,26 or are impurities in the stationary phase,27 and are thus released by the cartridges themselves. Chromatogram evaluation The characteristics of the chromatograms studied are reported in Table 1.Some values are directly calculated from the experimental chromatogram, i.e., the number of observed peaks, p, and the total area, AT. Other parameters can be estimated by the EACVF method, i.e., the number of components, m, the mean chromatogram area, Am, the mean width at half the chromatographic peak height, w1/2, the capacity factor, nc, and the separation extent g.All the reported results refer to the chromatographic space actually used for the separation (10–30 and 30–50 min for 1242 and 1260 mixtures, respectively; hence, X = 1200 s for all the chromatograms). With the w1/2 values it is possible to evaluate separation system performance and follow it through the various analysis steps. The data obtained are small (4.5 ± 0.1 s) and constant in all the chromatograms studied: consequently, the nc values [derived according to eqn.(7)] are also constant (nc = 316 ± 8). These results prove that the chromatographic system used performed well in all the separations reported. This control is important, as mentioned above, since it may single out a possible decline in column efficiency, a poor column choice, injection problems or uncontrolled sample overloading.23 Among these problems, the last-named is particularly significant since it can substantially change the conditions under which single components can be quantified by evaluating area or height and can significantly affect the final analytical result.From the data in Table 1 it is apparent how interfering, cartridge-released compounds cause errors in the analysis of Aroclor 1242 samples: the observable parameters, i.e., the number of peaks, p, and the total area, AT, increase significantly in the mixtures eluted on cartridges. It is possible to calculate the total PCB content in each mixture by comparing the total chromatogram area, AT, with the area of a standard Aroclor chromatogram (sample 1242St).The results obtained (second column in Table 1) show that for cleaned-up 1242 samples (samples 1242Si, 1242Si pw, 1242Al, 1242Al pw), the PCB content is erroneously over-estimated (72–74 ppm of total PCBs); the AT values are higher because there are interfering compounds in the chromatographic space. The EACVF method was applied to evaluate m [eqn. (4)]: the results obtained (third column in Table 1) agree with the data describing Aroclor mixture composition.1,22 It should be noted that, since it is a parameter statistically evaluated, it is affected by an error of ABm.17 Nearly 20 additional components are estimated in the 1242 mixture after elution on the silica and alumina cartridges: this is why the PCB content was overestimated. The Am values [calculated from AT and m according to eqn.(9) and reported in the fifth column of Table 1] are not so severely affected by interference as are the AT values since, in this case, higher AT values are compensated for by higher m.From Am, the total PCB content can be calculated (sixth column in Table 1) with reference to the Am of standard mixtures (samples 1242St and 1260St): the results obtained are not affected by over-estimation and can be considered a correct estimation of the total PCB content of the samples analysed. Table 1 Properties of chromatograms of Aroclor samples obtained after different clean-up steps: p, number of observed peaks; PCB content calculated from AT values; m, number of components; w1/2, mean value of peak width at half the chromatographic peak height; Am, mean area [eqn.(9)]; PCB content calculated from Am values; g, separation extent [eqn. (8) PCB content PCB content from AT from Am Sample p (ppm) m w1/2/s Am (ppm) g 1242St 35 50 45 4.5 64 676 50 0.78 1242Hex 36 50 46 4.5 63 663 50 0.78 1242CH2Cl2 36 51 45 4.5 64 532 50 0.80 1242Si 42 72 66 4.5 51 741 40 0.63 1242Si pw 40 73 62 4.5 59 735 42 0.64 1242Al 41 74 65 4.5 45 272 35 0.63 1242Al pw 39 72 61 4.5 47 859 37 0.64 1260St 38 50 47 4.5 42 242 50 0.81 1260Si 38 51 48 4.5 42 115 50 0.79 1260Si pw 36 50 46 4.5 42 199 50 0.78 1260Al 37 40 47 4.5 41 397 49 0.79 1260Al pw 37 51 46 4.5 41 927 50 0.80 1202 Analyst, June 1998, Vol. 123The g values (seventh column in Table 1) provide information on global peak overlapping in the chromatogram. In the chromatogram of cleaned-up 1242 mixtures, the extent of separation attained is significantly lower than in standard Aroclor (g decreases from 0.8 to 0.6).This is due to the interference of compounds released by the cartridges. Prewashing the cartridges does not significantly reduce the number of interfering compounds (compare samples 1242Si and 124Al with samples 1242Si pw and 1242Al pw, respectively): therefore, in all the chromatograms obtained after clean-up procedures, the degree of overlapping is so severe that only 60% of the compounds present in Aroclor 1242 produce chromatographic peaks.For Aroclor 1260, the chromatographic separation attained is not affected by interferences released during the clean-up procedure, as described by the constancy of the parameters p, m, g and PCB contents reported in Table 1. This result is expected, since compounds derived from cartridge elution are not arranged in the chromatographic space where the PCBs of the sample are placed [see Fig. 2(a)–(c)]: for this mixture a separation extent of 0.8 is always achieved. Recovery determination PCB recovery after different clean-up steps was determined using individual PCB and autocovariance function methods. The results are illustrated in Table 2: mean recovery and reproducibility (% RSD) from triplicate measurements are reported and compared. It should be noted that by use of the individual PCB method the PCB content of each mixture is determined as the mean value of only ten PCBs for Aroclor 1242, the chromatograms of which exhibit the most severe peak overlapping (g values nearly 0.6, Table 1).It is apparent that most of the results obtained by the two methods are statistically comparable, proving that the chemometric approach based on ‘statistical evaluation’ estimates nearly the same PCB content as does the classical individual PCB method. Some discrepancies exist in the data concerning 1242 samples: recovery values calculated by the individual PCB method are significantly higher than those estimated by autocovariance.This result is due to over-estimation of individual PCB content—the basis of the single PCB method— in the samples eluted on the cartridges, a consequence of interference and the low separation attained. It should be noted that the autocovariance method provides the more accurate results, since it is based on a statistical evaluation of the whole complex chromatogram and it is able to determine the properties hidden in the complex retention pattern due to peak overlapping.24 Moreover, from Table 2 it is apparent that recovery values estimated with the autocovariance method show a better reproducibility, with relative standard deviations below 3%.The results obtained for Aroclor 1260 show that the clean-up procedure is efficient, since all the PCBs are quantitatively recovered (mean recovery, 100 ± 2%).For Aroclor 1242, mean recovery values near 80% were achieved for all samples. Less chlorinated compounds (i.e., those containing two and three chlorine atoms) are recovered in lower proportions (lower than 75%) than the others.28 A possible reason for this low recovery may be the selective evaporation of lighter PCB congeners during concentration in the Kuderna–Danish apparatus. To test this hypothesis, the concentration steps with hexane (sample 1242Hex) and with hexane–dichloromethane (sample 1242CH2Cl2) were examined separately.In this case, quantitative recoveries were obtained (see Table 2): this proves that evaporation is not the cause of PCB loss.21 Therefore, the low recovery obtained must be ascribed to different causes, i.e., the fact that PCBs with lower chlorination levels are selectively retained by the silica and alumina columns.29 Conclusions The results obtained have demonstrated that the chemometric approach, based on a study of the autocovariance function, is a rapid, simple and precise method for estimating the properties of complex chromatograms.It overcomes the problems usually encountered in such evaluations due to peak overlapping present therein: the complexity of the mixture can be correctly estimated and the performance of the separation system can also be checked. Therefore, the approach can be proposed as a tool with which to evaluate an analytical procedure based on a chromatographic determination of multicomponent mixtures; it can be applied to monitor the traceability chain of any given method—by testing how the different steps involved in the procedure affect the final result—to eliminate bias and provide long-term stability and accuracy of measurements.One drawback of the autocovariance method is that it is based on a ‘statistical evaluation’ of the number of components, m, realised on a large number of single components. In contrast, the ‘individual PCB congener’ method is based on the assumption that each peak is made up of a single component, whereas the peak might be formed by two or more components.Owing to peak overlapping,13 this effect is severe in multicomponent chromatograms, such as those of PCB mixtures, and causes erroneous analytical results. In this work, MS detection under TIC conditions was investigated, although different detection systems—such as MS under SIM conditions and ECD—with different linearity and sensitivity variations, may also be considered in a subsequent handling.The autocovariance method appears very promising for ECD—afflicted with limited linearity range—since it can immediately detect loss of system efficiency due to chromatographic column overloading or ageing. Moreover, it has been verified that even a widely used procedure, such as clean-up with SPE cartridges, may introduce significant systematic errors in the final determination, because of the release of interfering compounds and low recovery.A strategy for ensuring the best quality of analytical measurements is always required in order to evaluate each individual step present in the methodology. References 1 Lang, V., J. Chromatogr., 1992, 595, 1. 2 de Boer, J., Duiniker, J. C., Calder, J. A., and van der Meer, J., J. Assoc. Off. Anal. Chem., 1992, 75, 1054. 3 de Boer, J., van der Meer, J., Reutergardh, L., and Calder, J. A., J. Assoc. Off. Anal.Chem., 1994, 77, 1411. Table 2 PCB % recovery (% relative standard deviation of three repeated measurements) obtained with the individual PCB method and the proposed method Sample Individual PCB method Proposed method 1242Hex 101* (5) 99 (3) 1242CH2Cl2 98* (4) 99 (2) 1242Si 89* (4) 80 (2) 1242Si pw 93* (3) 85 (3) 1242Al 80* (7) 70 (2) 1242Al pw 84* (5) 74 (3) 1260Si 99† (4) 100 (2) 1260Si pw 100† (5) 100 (2) 1260Al 101† (7) 98 (3) 1260Al pw 99† (5) 99 (2) * Mean value calculated on ten PCBs. † Mean value calculated on 16 PCBs.Analyst, June 1998, Vol. 123 12034 de Boer, J., van der Meer, J., and Brinkman, U. A. Th., J. Assoc. Off. Anal. Chem., 1996, 79, 83. 5 Kimbrough, D. E., Chin, R., and Wakakuwa, J., Analyst, 1994, 119, 1277. 6 Kimbrough, D. E., Chin, R., and Wakakuwa, J., Analyst, 1994, 119, 1283. 7 Kimbrough, D. E., Chin, R., and Wakakuwa, J., Analyst, 1994, 119, 1293. 8 Huthe, F., Musial, C. J., and Misra, R. K., J. Assoc. Off. Anal. Chem., 1988, 71, 369. 9 Folch, I., Vaquero, M. T., Comellas, L., and Broto-Puig, F., J. Chromatogr., 1996, 719, 121. 10 Valc�arcel, M., and Rios, A., Analyst, 1995, 120, 2291. 11 Reed, W. P., Fresenius’ J. Anal. Chem., 1995, 352, 250. 12 King, B., Ann. Chim., 1997, 87, 199. 13 Davis, J. M., and Giddings, J. C., Anal. Chem., 1985, 57, 2168. 14 Nagels, L. C., Creten, W. L., and Vanpeperstraete, P. M., Anal. Chem., 1983, 55, 216. 15 Herman, D. P., Gonnord, M. F., and Guiochon, G., Anal. Chem., 1984, 56, 995. 16 Martin, M., Hermann, D. P., and Guiochon, G., Anal. Chem., 1986, 58, 2200. 17 Felinger, A., Pasti, L., and Dondi, F., Anal. Chem., 1990, 62, 1846. 18 Felinger, A., Pasti, L., Reschiglian, P., and Dondi, F., Anal. Chem., 1990, 62, 1854. 19 Felinger, A., Pasti, L., and Dondi, F., Anal. Chem., 1991, 63, 2627. 20 Felinger, A., Pasti, L., and Dondi, F., Anal. Chem., 1992, 64, 2164. 21 Dondi, F., Betti, A., Pasti, L., Pietrogrande, M. C., and Felinger, A., Anal. Chem., 1993, 65, 2209. 22 Pietrogrande, M. C., Pasti, L., Dondi, F., Bollain Rodriguez, M. H., and Carro Diaz, M. A., J. High Resolut. Chromatogr., 1994, 17, 839. 23ietrogrande, M. C., Felinger, A., and Dondi, F., J. High Resolut. Chromatogr., 1996, 19, 327. 24 Dondi, F., Pietrogrande, M. C., and Felinger, A., Chromatographia, 1997, 45, 435. 25 Pietrogrande, M. C., Beneventi, C., Morselli, L., and Dondi, F., Ann. Chim., 1997, 87, 753. 26 Yunk, G. A., Avery, M. J., and Richard, J. J., Anal. Chem., 1988, 60, 1374. 27 Satsmadjis, J., Georgakopulos-Gregoriades, E., and Voutsinou- Taliaduori, F., J. Chromatogr., 1988, 437, 254. 28 Sevcik, K., J. Chromatogr., 1996, 752, 197. 29 Barnal, J. L., Del Nozal, M. J., and Jimenez, J. J., J. Chromatogr., 1992, 607, 303. Paper 8/00554K Received January 20, 1998 Accepted March 24, 1998 1204 Analyst, June 1998, Vol. 123

ISSN:0003-2654    

DOI:10.1039/a800554k

出版商:RSC    

年代:1998

数据来源: RSC

摘要:

Accelerated solvent extraction of the antioxidant Irganox 1076 in linear low density polyethylene (LLDPE) granules before and after g-irradiation Monica Waldebäck, Charlotte Jansson, Francisco J. Señoráns and Karin E. Markides* Department of Analytical Chemistry, Uppsala University, P.O. Box 531 Uppsala, SE-751 21, Sweden To achieve effective and reproducible extractions of the antioxidant Irganox 1076 from linear low density polyethylene (LLDPE) components at reduced time and cost, the accelerated solvent extraction (ASE) technique was evaluated with solvent composition and temperatures as test variables.The aim was to find proper extraction conditions for the balance between high diffusion rate of the analytes versus solvation of the plastic material. High extraction yields and reproducible results, RSDs between 9.7–12.5%, were obtained using ethyl acetate mixed with hexane 90 : 10 (v : v) at 100 °C. The results showed that no grinding of 4 mm granules was necessary before extraction nor was concentration of the extracts needed before LC analysis.By using the established extraction method, a comparison between g-irradiated and non-irradiated LLDPE granules showed that more than 96% of the Irganox 1076 was lost after g-irradiation (2 3 25 kGy). Keywords: Polyethylene; antioxidant; Irganox 1076; accelerated solvent extraction; g-irradiation; food and drug packagings; optimisation; liquid chromatography Chemical additives are frequently used to enhance the lifetime and minimise degradation of plastic materials.Analysis of polymer additives is thus important in both research and quality control for manufacturers and users of various polymers, especially for those used as packaging material for drugs and food. The first step in an analysis involves the isolation of the additives from the polymer matrix, usually by some type of liquid–solid extraction. The most common extraction method, the Soxhlet extraction, has the drawback of being both time1,2 and solvent consuming.Both heating in the solvent decalin3 and supercritical fluid extraction (SFE)4,5 have been applied as alternative approaches to the extraction of antioxidants from polyethylene. In addition, the recently developed accelerated solvent extraction (ASE) technology has also quickly become an attractive alternative to the traditional extraction techniques6. By applying pressure to the extraction cell, the solvents are kept in the liquid state at temperatures above their atmospheric boiling point.The use of higher temperatures increases the rate of diffusion of the components within the polymer particles as well as their transfer rate from the particles to the extraction solvent.7 Moreover the solubility of most analytes in the extraction solvent is increased. The aim of this study has been to investigate the possibility of using ASE technology to perform reproducible extractions of the antioxidant Irganox 1076 [octadecyl-3-(3,5-di-tert-butyl- 4-hydroxyhyphenyl)propionate] in linear low density polyethylene (LLDPE) granules. An additional aim was to use this extraction method to compare the content of Irganox 1076 in non-irradiated LLDPE granules and LLDPE granules sterilised by g-radiation.The resulting method should be simple, accurate and fast. Experimental Chemicals Acetonitrile and methanol were of LiChrosolv quality, while ethyl acetate, hexane, methylisobutylketone (MIBK), propan- 2-ol and tetrahydrofuran (THF) were of pro analysi grade.All solvents were purchased from Merck, Darmstadt, Germany, except MilliQ-water, which was obtained from a purification system (Millipore, Watford, Herts., UK). Nitrogen, of quality 99.996% of volume, Aga, Stockholm, Sweden and the Irganox 1076 standard was from Ciba Geigy, Basel, Switzerland. Sample Granules and processed samples of LLDPE from several batches, containing the analyte Irganox 1076, with an original concentration according to the manufacturer in the range of 240–460 ppm (Batch A) and 200–300 ppm (Batch B), were used.The polyethylene was received from Pharmacia & Upjohn AB, Uppsala, Sweden. The approximate diameter of a sample granule was 4 mm and the weight was about 36 mg. The irradiated granules used in this study were g-radiated by 2 3 25 kGy. Instrumentation A schematic illustration of the accelerated solvent extractor employed, ASE 200, Dionex (Sunnyvale, CA, USA) is shown in Fig. 1. The extraction cells (Dionex) were of stainless steel with a volume of 11 ml and capped with PEEK seals and stainless steel frits. Procedure The extraction process consisted of five steps: (i) filling and pressurising the cell with solvent at selected temperature; (ii) preheating the cell at selected temperature for equilibration at Fig. 1 Schematic illustration of the ASE system. Analyst, June 1998, Vol. 123 (1205–1207) 1205constant pressure for 5 min; (iii) static extraction at constant pressure and temperature over a selected period of time; (iv) flushing the cell with fresh solvent with a selected volume expressed as a percentage of the cell volume [the steps (iii) and (iv) are called a cycle, and can be repeated several times]; and (v) purging of the cell with nitrogen.The extraction cell was filled with a cellulose filter (Dionex, 1.91 cm diameter) and with 2 g of LLDPE granules. All extractions started with preheating for 5 min at a pressure of 10.5 MPa.The effect of different temperatures and solvents was studied. Each cycle was finished by a 60% (6.6 ml) fresh solvent flush. Finally the cell was purged with an inert gas, nitrogen, for 60 s and the analytes and the solvent from the cell and the lines were collected into a vial. The extracts were filtered by a 4 mm Millex-FH, hydrophobic PTFE-membrane with pore size 0.5 mm (Millipore). All extracts were analysed using an LC system (Waters, Stockport, UK) consisting of a pump (616), an autosampler (717), a 600 s controller and a tunable absorbance detector (486).A SYMMETRY C8, 3.9 3 150 mm (particle diameter 5 mm) column (Waters, Stockport, UK) was used and as a guard column a SYMMETRY C8, 3.9 3 20 mm precolumn was employed. The LC parameters were as follows: mobile phase, methanol–water (96 : 4), flow rate, 1.5 ml min21; injection volume, 10 ml; and UV detection at 230 nm.Calibration plots were made from peak area of a standard of Irganox 1076 in ethyl acetate–hexane (90 : 10). The retention times for the Irganox 1076 peak in the extract and in the standard solutions were equal within + 2%. Millenium software, version 2.15.3 was used for data acquisition and processing. Results and Discussion The first step in an ASE method development is to select a proper solvent composition for the extraction. The solvents recommended for Soxhlet extractions have generally been the solvents of choice also in ASE.6.Due to the higher temperature used in an ASE extraction and the fact that the matrix in this case is a polymer, this choice might be unwise. The solvent successfully applied in Soxhlet extractions is usually also a good swelling agent for the polymer matrix,7 that when used at the higher temperature in an ASE extractor, may result in melting of large portions of the polymer. The risk of irreproducible results and plugging of the tubing within the instrument will therefore be high.In an initial screening study, where the matrix consisted of a processed sample of LLDPE, cut into pieces of about 5 35 mm, various conditions were tested for extraction of Irganox 1076. The extraction yield for the different combinations of solvents and temperatures is shown in Table 1. The original value of Irganox in the LLDPE from the manufacturer was said to be 200–300 ppm . The fact that the extraction yields detected always are less than 100% of the added antioxidant is probably due to losses of Irganox 1076 during the manufacturing, processing and storage of the plastic.8 From Table 1 it can be seen that higher temperature during the extraction generally increases the yield.These results were consistent with published data showing that an increase in temperature gives improved recovery.9 At temperatures above 100 °C the polymer started to melt, and this fact ruled out the use of higher temperatures.When higher temperatures were tried, the tubing of the instrument was plugged. Ethyl acetate was found to give the highest yield at the maximum temperature of 100 °C and was therefore chosen for further studies. With THF as a solvent the matrix showed traces of melting already at 75 °C, and at 100 °C the melted polymer started to plug the tubing of the ASE extractor. This excluded THF from further use although it resulted in the highest yield from all solvents tested for extractions at 75 °C.Since the particle size is considered to be an important parameter influencing the maximum yield and the required extraction time, especially if the extraction rate is limited by diffusion within the particles,7 ground granules, whole granules and processed LLDPE cut into 5 3 5 mm were extracted at different temperatures. The extraction yield was higher for the ground granules at low temperatures, but at temperatures above 80 °C these granules started to melt, while the whole granules could withstand fast extractions at 100 °C.As the grinding step, in this way, could be omitted, whole granules were used in all the subsequent experiments, which rendered a simpler and faster method according to the goal of this study. In addition the method avoided any risk of antioxidant degradation during grinding. In order to find a proper static time for the extractions, three cycle runs of 5, 15 or 25 min were examined.When the static time was increased from 15 to 25 min no further improvement of the extraction recoveries was observed. Using 5 min static time for each cycle, a 36% lower yield was obtained. Therefore, in the following experiments a static time of 3 3 15 min was used. The extracts required no further concentration before LC analysis. In this way losses of Irganox 1076 due to further treatment was eliminated. A typical chromatogram from the LC analysis of an extract is seen in Fig. 2. In an effort to raise the extraction yield of the antioxidant, ethyl acetate was mixed with a swelling agent, hexane, at different concentrations and the extractions were performed at several temperatures using these mixtures. An experimental design, a full factorial design at three levels, was chosen to investigate the main variables, temperature and percentage of hexane, in ethyl acetate. The original values from the manufacturer of added Irganox 1076 in the LLDPE granules used in this study were 240–460 ppm (Batch A).The obtained results Table 1 Extraction yield in mg g21 of Irganox 1076 obtained in the screening study of processed sample of LLDPE Temper- Aceto- Ethyl Propanature °C Methanol nitrile acetate 2-ol THF MIBK 50 13 11 34 — — — 75 43 53 111 48 123 37 100 128 125 173 54 — 57 Fig. 2 A typical chromatogram from an LC analysis of an extract. Column, C8, 5 mm particle diameter, 3.9 3 150 mm; mobile phase, methanol–MilliQ-water (96 : 4); flow rate, 1.5 ml min21; UV detection, 230 nm; injection volume, 10 ml. 1206 Analyst, June 1998, Vol. 123for the 11 experiments are shown in Table 2. The data was fitted to a quadratic model with multiple linear regression (MLR), and the relative standard deviation of the model was 12.8%. The significant variables were temperature and percentage hexane in the solvent used, and no significant interaction between them could be seen at 95% confidence level. From Table 2 it can be concluded that the effect of the percentage of hexane is larger at lower temperatures, i.e., higher percentage of hexane gives higher yield of Irganox 1076 at lower temperatures.It is suggested that hexane works as a good swelling agent and this effect is especially important at lower temperatures. It can also be seen from Table 2 that the yield is enhanced at higher temperatures, a result which is in agreement with the results in Table 1 for every solvent composition.To show the dependence of the response as a function of temperature and percentage of hexane, a response surface plot was made with the model obtained from the data in Table 2 (see Fig. 3). From this figure it can be concluded that the temperature has a greater influence on the yield of Irganox 1076 than the concentration of hexane. Consequently, at 100 °C, the yield is high and the percentage of hexane does not seem to markedly affect the yield at this temperature. At temperatures above 100 °C and with hexane in excess of 25% the polymer started to melt and even dissolve.Even though the melted plastic did not block the accelerated solvent extractor there was obviously a risk of plugging the tubing of the instrument. In addition, these extracts were very cloudy from dissolved plastic material, and there would be a risk that the extracted analyte Irganox 1076 could get redissolved into the co-extracted plastic material. A repeatability study was performed using the optimised parameters found in this study, where whole pellets were extracted at 100 °C with ethyl acetate–hexane 90 : 10 (v/v) as solvent and with a static time of 3 3 15 min.Two different batches, Batches A and B, were tested, with five replicates of each batch extracted. Reproducible results were obtained with RSD values between 9.7–12.5%, see Table 3. Five replicates of g-irradiated (2 3 25 kGy) LLDPE granules from Batch B were extracted in the same way and the average yield from the g-irradiated granules was 4.4 mg g21.The high RSD value of 33.3 results from an extreme value in sample 4, possibly caused by non-homogenous degradation of the antioxidant. More than 96% of the extractable Irganox 1076 could not be extracted after the g-irradiation. The data are comparable with the results obtained by Yagoubi et al.,10 who observed a loss of 95% in the Irganox 1076 content, after polyethylene vinyl acetate was radiotreated by 25 kGy.The difference in the Irganox 1076 yield is probably caused by scission of this molecule as pointed out by Yagoubi,10 and by the increase of crosslinking within the polymer due to the irradiation. The optimised ASE method has shown to be adequate to estimate the losses of the antioxidant Irganox when LLDPE is treated with girradiation, and will be used in further experiments to determine the effect of different treatments on Irganox 1076 degradation. This study has proved ASE to be a fast and suitable technique for a reproducible extraction of Irganox 1076 in unground samples of LLDPE (see Table 3).One sample can be extracted and analysed within less than 1 h, compared with the Soxhlet extraction where only the extraction step needs between 6 and 48 h.1,2 No time is needed for grinding of the LLDPE granules or to reduce the volume of the extract before LC analysis. In addition the ASE instrument used in this study could take up to 24 samples at a time for increased sample throughput.The authors acknowledge the support from Dionex (Salt Lake City, UT, USA) for making a Dionex ASE 200 instrument available, and Pharmacia & Upjohn for activly supplying this project with samples, running experiments and creative discussions. References 1 Crompton, T. R., Eur. Polym. J., 1968, 4, 473. 2 Haney, M. A., and Dark, W. A., J. Chromatogr. Sci., 1980, 18, 655. 3 Chabron, J. F., and Fenska, L. E., Anal. Chem., 1980, 52, 1411. 4 Ashraf-Khorassani, M., and Levy, J. M., J. High Resolut. Chromatogr., 1990, 13, 742. 5 Lou, X., Janssen, H-G., and Cramers, C. A., J. Microcolumn Sep., 1995, 7(4), 303. 6 Richter, B. E., Jones, B. A., Ezzell, J. L., and Porter N. L., Anal. Chem., 1996, 68, 1033. 7 Lou, X., Janssen, H-G., and Cramers, C. A., Anal. Chem., 1997, 69, 1598. 8 Plastics Additives Handbook, ed. Gächter, R., and M�uller, H., Hanser Publishers, Munich, 1984, p. 19. 9 Richter, B. E., Ezzell, L. J., Felix, D., Roberts, K. A., and Later, D. W., Am. Lab., February, 1995. 10 Yagoubi, N., Baillet, A., Pellerin, F., and Ferrier, D., Nucl. Instrum. Methods Phys. Res., Sect. B, 1995, 105, 340. Paper 8/01530I Received February 23, 1998 Accepted April 8, 1998 Table 2 Extraction yield in mg g21 of Irganox 1076 obtained with the experimental design from LLDPE granule Batch A Temperature/°C 5% Hexane 25% Hexane 45% Hexane 80 106 141 160 90 147 163, 179, 161 201 100 203 239 216 Fig. 3 Response surface plot of the extraction yield of Irganox 1076 using temperature and percentage of hexane in ethyl acetate as variables. Table 3 Study of the reproducibility of the proposed method; extraction yield of Irganox 1076 in mg g21 in LLDPE granules from Batch A and Batch B Sample of LLDPE 1 2 3 4 5 Average Std. RSD (%) Batch A 252 218 203 198 213 217 21.2 9.7 Batch B 116 140 142 135 165 140 17.5 12.5 Table 4 Extraction yield of Irganox 1076 in mg g21 in g-irradiated LLDPE granules Sample of LLDPE 1 2 3 4 5 Average Std. RSD (%) Batch B 4.4 3.6 3.1 6.8 3.9 4.4 1.5 33.3 g-irradiated Analyst, June 1998, Vol. 123 1207

ISSN:0003-2654    

DOI:10.1039/a801530i

出版商:RSC    

年代:1998

数据来源: RSC

摘要:

Study of the parameters in microwave dissolution methods using a magnetic stirring device in the microwave unit. Application to dissolution of high-carbon ferrochromium Aurora G. Coedo, Teresa Dorado* and Isabel Padilla Centro Nacional de Investigaciones Metalúrgicas (CSIC), Gregorio del Amo 8, Madrid, 28040 Spain Microwave sample preparation methods have come a long way since the earliest experiments with domestic microwave ovens. This study concerned the decomposition of ferrochromium samples containing a high proportion of insoluble carbides by means of a focused microwave digestion system.The experimental results showed that samples were completely decomposed in HNO3–HF mixed acids. Optimization of the microwave dissolution programmes involved the study of acid mixture volume, the control of pressure and temperature inside a reference vessel, power control and the influence of magnetic stirring during the dissolution processes. Chromium was determined potentiometrically in each of the intermediate digests obtained during optimization of the procedure, for demonstration of the successive recoveries of this element.The selected dissolution procedure provided a recovery of 99–100% Cr under the following conditions: 5 min at 250 W/30 min at 400 W/15 min at 500 W (without stirring) and 5 min at 250 W/15 min at 400 W/5 min at 500 W (with stirring) and showed excellent reproducibility with a slightly improved RSD when magnetic stirring was used.The accuracy was determined using three high-carbon ferrochromium reference materials: ECRM 585-1 from the Bureau of Analysed Samples, BAM 530-1 from the Bundesanstalt für Materialprüfung and BCS 204-4 from British Chemical Standards. The results were compared with those obtained using a classical dissolution procedure. The microwave solutions could also be used for the simultaneous determination of major and minor elements in FeCr samples by spectroscopic techniques.Keywords: High-carbon ferrochromium; microwave dissolution; digestion conditions In analytical chemistry, quality control is becoming increasingly significant. However, it is applied principally to measuring techniques rather than to sample preparation. Most quantitative analytical techniques require samples in liquid form compatible with sample introduction methods. Therefore, regardless of its nature, each sample has to be subjected to a specific treatment. The classical dissolution techniques used for metallurgical materials involve a time-consuming digestion because they involve multi-step, labour-intensive procedures. At present, one of the most important sample preparation methods in elemental analysis is microwave-assisted wet digestion in closed pressurized vessels.1 The use of this system offers all the advantages of open-vessel digestion but dramatically shortens the time required to digest nearly all types of sample.In order to investigate the completeness of the dissolution of some difficult metallurgical samples, such as high-carbon ferrochromium (FeCr), the alternative microwave dissolutionbased wet digestion technique was evaluated, through comparison with those currently used.There are several procedures and standard methods for the analysis of FeCr with a view to determining major constituents or minor elements.2–4 These methods exhibit good recoveries for almost all elements, determined subsequently not only by FAAS but also by ICP-AES or classical wet chemical methods.The dissolution procedures include an initial digestion, in a mixture of acids or oxidizing agents, and a fusion procedure for the treatment of the insoluble residue. However, they need constant attention and are subject to possible contamination and potential loss of some volatile elements. The application of microwave energy in closed vessels, for rapid sample dissolution, seemed to be a very attractive procedure to dissolve high-carbon FeC since carbide compounds present in these samples are extremely acid resistant and it is not possible to dissolve them completely by conventional methods (only by a fusion procedure).The use of microwave energy as the heat source in acid digestions was first demonstrated in 1975.5 Early work using open vessels, involved risks of environmental contamination and mechanical or volatile loss of analyte, in addition to which the maximum sample temperature was limited to the boiling-points of the acids.The use of closed digestion vessels to obtain high temperatures and pressures made it possible to improve the digestion of many types of samples, such as geological species, soils, steels and biological samples.629 Because of the differing chemistries of metallurgical materials, a variety of procedures are required for the digestion of these kinds of samples. Many studies have been performed using this dissolution approach and good recoveries have been reported for certain elements in various siderurgical products.10212 Most of these studies were performed without temperature or pressure control.The control of these parameters is very important because too high temperature and/or pressure conditions may cause serious deformation or breaking of the vessels and, at the same time, they are responsible for the efficiency of the decomposition reactions. In this study, direct wet dissolution procedures for highcarbon FeCr with compounds with low solubility in acids, such as silicon carbide, were explored by using appropriate acid mixtures at high temperatures and pressures.A closed-vessel laboratory microwave system equipped with temperature and pressure regulation and with an in-vessel magnetic stirring device was used to investigate the possibilities and problems of applying this simple method under optimized conditions. The use of the newly developed stirring module for microwave units opens up new possibilities in sample preparation.As a result of applying heating plus stirring, faster reactions are expected owing to the increased surface area of contact between samples and reagents. We evaluated the efficiencies of different FeCr sample decomposition processes using several acid mixtures in micro- Analyst, June 1998, Vol. 123 (1209–1214) 1209wave ovens or through conventional heating. Multiple programming steps were applied and temperature and pressure were measured during chemical digestions, with and without magnetic stirring.The contents of certain elements in the decomposed samples, Cr and minor elements, could be determined by classical potentiometric or spectroscopic techniques. The decomposition processes selected and the optimized digestion parameters allowed reliable results to be obtained, as confirmed Fig. 1 Pressure and temperature versus time data for microwave digestion programme 2 (see text) applied to four different acid mixtures in separate batches.Data were collecting using a control vessel containing 5 ml of each of the acid mixtures: 1, HF–HNO3; 2, HCl–HNO3; 3, HF–H2SO4–H3PO4–HNO3; and 4, HF–HClO4. Fig. 2 Temperature and pressure profiles for 0.1 g of ECRM 585-1 digested in 3 ml of HNO3 + 1 ml of HF + 1 ml of H2O. Numbers (2)–(5) refer to microwave digestion programmes 2–5 (see text). 1210 Analyst, June 1998, Vol. 123by determining the major element, Cr, in several certified reference materials.Experimental Instrumentation The microwave digestion system used for sample preparation was a Milestone MLS-1200 MEGA (Salisole BG, Italy) microwave oven, modified for stirring and equipped with temperature and pressure regulation through a sensor vessel. The power range of the oven may be set in 10 W increments up to a maximum of 1000 W. The ASM-400, stirring device is built-in below the bottom plate of the cavity of the microwave oven.Powerful motor-driven magnets rotate below the cavity, generating a rotating magnetic field in it. The variable stirring speed is controlled manually by a control knob. The unit is equipped with an MDR 1000/6/110 microwave digestion rotor, with a pressure and temperature reference vessel package and up to five high-pressure digestion vessels. The vessels are constructed of TFM (Tetrafluormethaxil; Registered Trademark of Hoechst) and are capable of withstanding temperatures of up to 260 °C and pressures of up to 110 bar.Temperature and pressure are measured by a thermocouple and a transducer, respectively, which are located in the sensor vessel. The maximum values that may be set are pressures of up to 50 bar and temperatures of up to 240 °C. The heating programme is controlled and monitored by a Model 240 remote control terminal. The software installed allows time, power, pressure and internal temperature to be selected for each of the 10 possible heating steps and also makes it possible to regulate and maintain the pre-set pressure and internal temperature values, once they are reached, through the control of microwave power emission.Digestions were carried out in closed 100 ml TFM vessels which had been cleaned by leaching in hot hydrochloric acid (1 + 1) followed by hot nitric acid (1 + 1) and finally rinsed and filled with distilled, de-ionized water until required for use. Reagents and standard reference materials All reagents were obtained from Merck (Darmstadt, Germany); 32% m/v HCl, 65% m/v HNO3, and 48% m/v HF (all of which were distilled, ppb/Teflon grade) were used.Distilled, de- Fig. 3 Temperature and pressure profiles for microwave digestion programme (see text). Data were collected using a control vessel containing 0.1 g of ECRM 585-1 dissolved in (1) 5 ml of HF–HNO3 and (2) 10 ml of HF–HNO3. Table 2 Stirring effect on microwave digestion.Programmes tested: 1–4 (see text). Test conditions: 0.1 g of CRM 585-1 + 10 ml of HF–HNO3–HY2O (1 + 3 + 2) With stirring Without stirring Programme Pressure/ Temperature/ Recovery Pressure/ Temperature/ Recovery No. Time*/min bar °C (%) Time*/min bar °C (%) 1 — 10 198 < 50 — 2.9 152 < 50 2 9 15.7 200 80 — 5.9 178 60 3 11 16.0 200 99 — 7.6 190 80 4 10 16.8 200 100 25 9.6 200 85 * Time in the second step, where the pre-set temperature is reached. Table 1 Effect of microwave heating programmes (1–6) on the recovery of Cr.Data were collected on a control vessel containing 0.1 g of ECRM 585-1 (certified Cr content 57.6%) in 5 ml of acid mixture (tests 1–6) and 10 ml of acid mixture (test 7). Programme Time/ Power/ Cr found Recovery No. min W Test No. (%) (%) 1 5 250 1 19.58 34 10 400 2 5 250 2 37.44 65 10 400 5 500 3 5 250 3 47.23 82 15 400 10 500 4 5 250 4 50.12 87 30 400 5 500 5 5 250 5 51.26 89 30 400 10 500 6 5 250 6 45.50* 79 30 400 15 500 6 5 250 7 57.71 100.2 30 400 15 500 * If acid fumes were released from the valves, the solutions became almost dry and were redissolved and filtered before applying the potentiometric method.Analyst, June 1998, Vol. 123 1211ionized water (DDW) was obtained from a Milli-Q Plus system (Millipore, Bedford, MA, USA) with 18 M½ cm specific resistivity. The certified reference materials used were ECRM 585-1 from the Bureau of Analytical Samples, (Newham Hall, Middlesbrough, UK) BAM 530-1 from the Bundesanstalt f�ur Materialpr�ufung (Berlin-Dahlem) and BCS 204-4 high-carbon ferrochromium from British Chemical Standards (Issued by Bureau of Analytical Samples).Sample dissolution Approximately 0.1 g of ferrochromium, particle size < 100 mm, weighed to the nearest 0.1 mg, was placed in a TFM container including the sensor vessel. Then 6 ml of HNO3, 2 ml of HF and 2 ml of H2O were added to each FeCr sample (a minimum of 10 ml of solution is required to immerse the temperature probe when the stirring device is used).All the experiments were carried out by heating six samples at a time in the microwave oven. The digestion vessels, including blank and sensor vessel, were placed in the PEEK (polyetherether-ketone) protection shield and inserted in the niches along the six-position rotor body. The vessels were sealed in accordance with the manufacturer’s procedure, using a tension wrench or a capping tool, until a clicking sound indicated that the vessel was locked inside its niche.The heating programme consisted of a three-stage power– time setting, with pre-set temperature and pressure in the sensor vessel of 200 °C and 20 bar, respectively. The system was operated at 250 W unpulsed power for 5 min, to start a soft reaction, at 400 W for 30 min and at 500 W for 15 min. On completion of the heating cycle, a ventilation step was applied for 10 min in order to cool the vessels inside the microwave chamber.The rotor was removed from the oven and cooled in water-bath for approximately 15 min. When cool, the vessels were uncapped and diluted to 100 ml with DDW. Table 3 Comparison of dissolution procedures on recovery of Cr in three high carbon FeCr CRMs. Results are mean recoveries ± s (% m/m) (n = 6) Dissolution method Microwave Microwave Certified* without with CRM (%) Classical stirring stirring ECRM 585-1 576.6 (0.2) 57.26 ± 0.44 57.06 ± 0.20 57.85 ± 0.10 BAM 530-1 64.9 (0.1) 64.41 ± 0.38 64.80 ± 0.12 65.20 ± 0.08 BCS 204-4 71.9 71.68 ± 0.25 71.60 ± 0.14 72.10 ± 0.09 * Values in parentheses are standard deviations of the interlaboratory means.Table 4 Minor elements (%) in ferrochromium CRM samples Element Sample Certified* Found (n = 6)† Mn ECRM 585-1 0.86 (0.02) 0.87 (0.01) BAM 530-1 0.16 (0.01) 0.165 (0.008) Ni ECRM 585-1 0.197 (0.012) 0.20 (0.01) BAM 530-1 0.19 (0.01) 0.18 (0.01) Co ECRM 585-1 0.044 (0.003) 0.045 (0.001) BAM 530-1 0.038 (0.004) 0.039 (0.001) V ECRM 585-1 0.33 (0.02) 0.34 (0.01) BAM 530-1 0.23 (0.03) 0.24 (0.01) Si ECRM 585-1 2.76 (0.08) 2.82 (0.05) BAM 530-1 0.49 (0.03) 0.51 (0.02) P ECRM 585-1 0.018 (0.003) 0.018 (0.001) BAM 530-1 0.016 (0.002) 0.017 (0.001) Ti ECRM 585-1 0.36 (0.02) 0.362 (0.009) BAM 530-1 0.05 (0.01) 0.051 (0.002) * Values in parentheses are standard deviations of the interlaboratory means.† Values in parentheses are precision expressed as standard deviation. Fig. 4 Effect of stirring on microwave digestions programmes 2 and 4 (see Table 2). Pressure and temperature profiles (a) and (c) without stirring and (b) and (d) with stirring. 1212 Analyst, June 1998, Vol. 123Sample analysis Samples were analysed for chromium content using the potentiometric method based on an ISO standard.2 For this purpose, microwave solutions were fumed gently with 10 ml of sulfuric acid and 5 ml of phosphoric acid. After dilution with water, chromium was oxidized with ammonium peroxodisulfate in the presence of silver ions and was determined by potentiometric titration with ammonium iron(ii) sulfate.This amount of ammonium iron(ii) sulfate corresponds to Cr + V (in this case, the V content is very low). If vanadium is present, it is necessary to perform reoxidation of vanadium with potassium permanganate followed by potentiometric titration of this element with ammonium iron(ii) sulfate. Chromium is then determined by difference.If the determination of minor elements (Si, P, Mn, Ni, Co, Ti, V) is required, spectroscopic techniques would be used. The operating parameters for instrumental techniques would be optimized to obtain a good performance. Results and discussion Dissolution conditions Various acid mixtures have been studied to determine the best dissolution procedures for different siderurgical products.10 Numerous previous studies have been performed to select the most suitable acids for high-carbon FeCr dissolution.Good results were reported when HF–HNO3 (1 + 3) was used. When other acid mixtures were employed, such as aqua regia, H2SO4–H3PO4–HNO3 (0.5 + 3 + 0.5) or HF–HClO4 (3 + 1), it was not possible to obtain reproducible solutions or complete decomposition of the samples. Fig. 1 shows the pressure and temperature profiles for all the acid mixtures tested (1, HF– HNO3; 2, aqua regia; 3, H2SO4–H3PO4–HNO3; and 4, HF– HClO4) when using a common microwave programme: 5 min at 250 W/10 min at 400 W/5 min at 500 W.The volume of all acid mixtures was 5 ml (including 1 ml of H2O). When the temperature of the solutions reached approximately 90 °C, the pressure inside the vessels increased fr when using acid mixture 1, 2 or 4. This was due to the relatively low boilingpoints and large partial pressures of HNO3, HCl and HF (the pressure and temperature reached during digestion with HF– HClO4 were influenced more by a high proportion of HF than of HClO4).H3PO4 and H2SO4, in contrast, have lower partial pressures at comparable temperatures and higher boiling points. As the pre-set values for pressure and temperature were 20 bar and 200 °C, and the pre-set value for pressure was not reached with any of the acid mixtures, the microwave power was regulated by the pre-set temperature of 200 °C. The digestion processes varied considerably depending on the pressure reached, and improved as pressure increased. Although HF– HNO3 cannot be said to be the only acid mixture to dissolve all the FeCr samples, it was used in this work to study the decomposition conditions and to establish the microwave digestion procedure.The pre-set values of pressure and temperature were 20 bar and 200 °C for all the assays carried out in this work. Using HF–HNO3 digestion, different heating programmes were tested to verify the difference in the recoveries for Cr. As the digestion system was equipped with a stirring module, a study was made of the difference in the recoveries of elements due to the stirring inside the vessels during the chemical reactions throughout the entire microwave oven digestion procedure.Data were collected using a control vessel containing 0.1 g of ECRM 585-1 FeCr sample in 5 ml of HF–HNO3– H2O (1 + 3 + 1). All the heating programmes tested started with 5 mins at a 250 W ‘unpulsed’ power setting (initiation of a smooth reaction) and continued with different pulsed power settings (1000 W are delivered with the On/Off control of the magnetron for fractions of a cycle range; the standard setting for pulse width is 5 s, meaning that, for example, 400 W is equal to 2 s On and 3 s Off).The applied heating programmes were as follows: 1, 5 min at 250-W/10 min at 400 W; 2, 5 min at 250-W/ 10 min at 400 W/5 min at 500 W; 3, 5 min at 250 W/15 min at 400W/10 min at 500 W; 4, 5 min at 250 W/30 min at 400 W/5 min at 500 W; 5, 5 min at 250 W/30 min at 400 W/10 min at 500 W.Fig. 2 shows the heating programmes applied and the data obtained for pressure and temperature values versus time. Programme 1 is not shown owing to the low digestion efficiency. As may be appreciated, the digestion processes were controlled by the temperature limit as the pressure did not exceed the pre-set limit. Only minor changes in pressure were noted up to a temperature of about 100 °C. After this point the temperature inside the vessel increased faster than the pressure, and when the pre-set value of 200 °C was reached, the pressure value did not exceed about 12 bar in any case.In all of these programmes the FeCr samples were partially decomposed, but the digestion efficiency improved as larger digestion times were applied. The solutions obtained after application of each programme were conditioned for the potentiometric determination of chromium. The recoveries of Cr obtained for the ECRM 585-1 FeCr sample using programmes 1–5 were about 34, 65, 82, 87 and 89%, respectively.In an attempt to improve these recoveries, a new programme was applied, increasing the heating time to 15 min in the third step of the last programme (programme 6, 5 min at 250 W/30 min at 400W/15 min at 500 W). Tests showed that insoluble particles or precipitates remained in the digestates only occasionally when less than 7–8 ml of acids, for 0.1 g of sample, were used. During the dissolution process under these conditions, a study was made of the minimum amount of acids required for digestion, in order to reduce possible sources of contamination and to have the lowest concentration of acids in final solutions for subsequent analysis of minor or trace elements by instrumental techniques.Incomplete digestion was produced when 5 ml of acid mixture, for 0.1 g of sample, was used. Maintaining the temperature of 200 °C reached in the last step of the programme for longer than 10 min did not result in any improvement as regards the completeness of digestion because, occasionally, acid fumes were released from valves due to overheating and the sample were almost dried out.On the other hand, a minimum of 10 ml of solution was required to immerse the temperature probe when the stirring device was used. Therefore, 10 ml of acid mixture, for 0.1 g of sample, were adopted to develop the microwave oven procedure. Fig. 3 shows the variations of the temperature and pressure profiles for the digestion of ECRM 585-1, with 5 and 10 ml of acid mixture.The pressure and temperature reached during the first step were, in both cases, about 0.8 bar and 110 °C, respectively. During the last 10 min of the second step, the pre-set maximum temperature of 200 °C was reached and the pressures were held at 11.5 and 15.5 bar for 5 and 10 ml, respectively. These values were maintained throughout the third step in order to complete the digestion of FeCr.By using 10 ml of acid mixture, clear and quantitative FeCr solutions were obtained and the recoveries of Cr were > 99%. The results in Table 1 show the effect of the different microwave heating programmes applied on the recovery of Cr. As the high-carbon FeCr samples showed different behaviours depending on the type of insoluble carbides, samples of similar composition were digested in separate batches in order to ensure that the sensor and sample vessels contained similar matrices.It was also seen that small variations in the temperature profile may exist between digestion batches and also between samples in the digestion vessels and sensor vessel in the same batch. These factors might be the cause of variations in digestion efficiency. Because when applying these digestion programmes the microwave power was controlled by the pre-set Analyst, June 1998, Vol. 123 1213temperature for a significant length of time, it was observed that the longer the digestion the better the digestion efficiency.A reproducible procedure could be established by using 10 ml of acid mixture and maintaining the temperature of 200 °C when applying either of the two last digestion programmes (5 or 6), depending on the proportion of insoluble carbides. A similar study was carried out throughout the entire microwave digestion procedure to determine the influence of stirring in the microwave unit during the chemical reactions.For this study, a magnetic bar was placed in each vessel and a stirring effect was obtained. Digestion programmes 1–4 were tested for the dissolution of 0.1 g of CRM 585-1 using 10 ml of the same acid mixture. The pre-set temperature and pressure values were 200 °C and 20 bar. Table 2 shows a comparison of the temperature and pressure values attained with the same programmes, with and without stirring. When the temperature reached the pre-set value of 200 °C, with magnetic stirring 15.7 bar of pressure had accumulated in 9 min (since the beginning of the first step at pulsed power); however, with a similar digestion time but without stirring, lower values were attained (approximately 160 °C and 3.5 bar). This means that faster reactions were produced when using stirring, owing to the increased surface contact of sample particles with the reagents, along with a more homogeneous temperature distribution.When stirring was used, the temperature and pressure values remained constant once the pre-set temperature was reached.Moreover, the temperature and pressure values given in Table 2, without stirring, correspond to those attained at the end of each heating programme. Good recoveries were obtained on applying shorter programmes when the stirring device was used (30 min in programme 3 are sufficient to dissolve FeCr samples). Fig. 4 shows a comparison between the profiles of temperature and pressure using programmes 2 and 4, with and without magnetic stirring.Comparison of chromium results from different dissolution procedures The purpose of this work was to investigate the validity of the microwave oven procedure as an alternative wet-digestion technique to those currently used. Chromium was determined, using a potentiometric method, in solutions obtained using the classical wet chemical method and the microwave oven procedure, for the CRMs: ECRM 585-1, BAM 530-1 and BCS 204-4. Six replicates per sample were prepared simultaneously for analysis using each of the two dissolution methods.The analytical results in Table 3 were in good agreement with the certified values and indicated that the concentration differences for chromium between the classical dissolution and microwave procedures (with and without stirring) were not significant. Microwave solutions were also used for the determination of minor elements: Mn, Ni, Co and V were determined directly, in these microwave solutions, by FAAS.Si, P and Ti were determined by spectrophotometric methods, after appropriate treatment of the solutions. The analytical results in Table 4, allow a comparison of the certified and the mean values obtained when using microwave solutions for ECRM 585-1 and BAM 530-1. BCS 204-4 is not shown because the minor elements are not the same as those in the other two samples and its certificate of chemical analysis is not expressed with the standard deviation of the interlaboratory mean values.Conclusions The critical point of this work was the dissolution of the highcarbon FeCr samples. It was demonstrated that if samples, including carbides, were well dissolved the results obtained for Cr were in agreement regardless of whether conventional heating systems or fusion or microwave digestion were used. Nevertheless, the experiments, with and without stirring, demonstrated that the technique is sufficiently powerful to dissolve carbides without fusion.By using this simple and fast microwave procedure, it is possible to obtain solutions for the determination of minor elements in the same digest by spectroscopic techniques (FAAS, ICP-AES, ICP-MS, etc.) with little risk of contamination and the use of a minimum volume of acids. The times were reduced from a few hours when using wet mineralization or fusion to about 60 min (including the cooling time) when the microwave heating source was used. The fusion procedure did not require expensive laboratory equipment and it was not sensitive to the different natures of the samples.The microwave oven mixed acid digestion system has proved to be a very rapid and accurate method for decomposition of FeCr prior to elemental or multi-element determination. The recoveries were found to be dependent on sample type, acid mixture, digestion time, power programme and stirring during digestion. The microwave digestion procedure developed provides low sample and reagent consumption with a higher frequency of sample analyses per unit time. The solutions prepared are suitable for analysis by spectroscopic techniques. Financial support provided by the European Community for Steel and Carbon (ECSC) and the Comisión Interministerial de Ciencia y Tecnología (CICYT) is gratefully acknowledged. References 1 Introduction to Microwave Sample Ppreparation. Theory and Practice, ed. Kingston, H. M. and Jassie, L. B., American Chemical Society, Washington, DC, 1988. 2 International Standard ISO 4140-1979, International Standards Organization, Geneva, 1979. 3 Coedo, A. G., Dorado, M. T., and Videl Maeso, A., Spectrochim. Acta, Part B, 1986, 41, 193. 4 ASTM Standards. Chemical Analysis of Metals and Metalbearing Ores, American Society for Testing and Materials, Philadelphia, PA, 1989, Vol. 03.05, Method E.350. 5 Abu-Samra, A., Morris, J. S., and Koirtyohann, S. R., Anal. Chem., 1975, 47, 1475. 6 Fischer, L. B., Anal. Chem., 1986, 58, 261. 7 Papp, C. S. E., and Fischer, L. B., Analyst 1987, 112, 337. 8 Fernando, L. A., Heavner, W. D., and Gavrielli, C. C., Anal. Chem., 1986, 58, 511. 9 Kingston, H. M., and Jassie, L. B., Anal. Chem., 1986, 58, 2534. 10 Dorado, T., Del Monte, M. G., Falciani R., and Tamba, A., European Commission EUR 15503 EN and EUR 15900 IT, Office for Official Publications of the European Communities, Luxembourg, 1996. 11 Berglund, B., and Wichardt, Ch., Anal. Chim. Acta, 1990, 236, 399. 12 Matthes, S. A., Farrel, R. F., and Mackie, A. J., Bur. Mines. An. Support Service Prog. Tech. Prog. Rep., 1983, 120. Paper 8/00151K Received January 5, 1998 Accepted March 19, 1998 1214 Analyst, June 1998, Vol. 123

ISSN:0003-2654    

DOI:10.1039/a800151k

出版商:RSC    

年代:1998

数据来源: RSC

摘要:

Determination of total mercury in biological tissues by flow injection cold vapour generation atomic absorption spectrometry following tetramethylammonium hydroxide digestion Guanhong Tao†, Scott N. Willie and Ralph E. Sturgeon* Institute for National Measurement Standards, National Research Council of Canada, Ottawa, Ontario, Canada K1A 0R9 A simple, rapid and reliable method was developed for the determination of total mercury in biological samples. Samples were solubilized using tetramethylammonium hydroxide (TMAH).The organically bound mercury was cleaved and converted to inorganic mercury by on-line addition of KMnO4. The decomposed mercury together with inorganic mercury originally present in samples was determined by flow injection cold vapour atomic absorption spectrometry after reduction to elemental mercury vapour using NaBH4. A sample throughput of 100 measurements per hour was achieved after a 30 min dissolution with TMAH. The relative standard deviation for 20 mg l21 Hg was 1.3% (n = 11) and the limit of detection was 0.1 mg l21 (3s).The proposed method was validated by the analysis of a suite of certified marine biological reference materials, DORM-2 (dogfish muscle), DOLT-2 (dogfish liver) and TORT-2 (lobster hepatopancreas), with calibration against simple HgII standards. Keywords: Mercury; biological samples; tetramethylammonium hydroxide digestion; flow injection; cold vapour atomic absorption spectrometry There is increasing demand for rapid and sensitive techniques for the determination of toxic elements in biological materials.Mercury is of considerable interest because of its toxic nature and ability to bioaccumulate in many organisms. For the determination of mercury in biological samples, cold vapour atomic absorption spectrometry (CVAAS) is one of the most commonly used methods, owing to its high sensitivity and its ease of operation.1 However, an important prerequisite is that mercury is present in the +2 oxidation state in order for reduction to elemental mercury (Hg0) to occur by a suitable reductant such as tin(ii) chloride.2–5 Quantitation can then be performed by sweeping the mercury vapour formed into a quartz atomizer for atomic absorption detection.Organomercury compounds, however, are not reduced to metallic mercury by SnCl2 or not completely by NaBH4 and so quantitation is impossible unless suitable pre-treatment of the sample is undertaken.4–11 For the analysis of biological samples by CVAAS, the pre-treatment must achieve two objectives.Firstly, the organic matter in the sample must be sufficiently oxidized to liberate the mercury species from the sample matrix, and secondly, the liberated mercury must be fully oxidized to HgII. Two types of decomposition methods have been employed for this purpose, i.e., wet digestion and dry ashing. A bewildering variety of combinations of strong acids (HCl, H2SO4, HNO3), oxidants (H2O2, KMnO4, K2Cr2O7, K2S2O8), elevated temperatures, UV irradiation and microwave exposure have been used and recommended.2–12 However, due to the well-known problems associated with the mobility of this element and the inherent risk of contamination, volatilization and adsorption losses, care must be taken during the sample pre-treatment.Recently, systems were reported for the on-line digestion of biological samples with or without microwave assistance and determination of mercury by CVAAS, which provided an automated, contamination-free enclosed sample handling system. 5,13–17 They have been successfully applied to the determination of mercury in fluid samples, such as blood, urine and saliva. Although solid samples could also be run by the above systems, they must be slurried and homogenized prior to the introduction to the system. This required tedious slurry preparation, which is prone to contamination errors. Therefore, hitherto, these systems were limited to the analysis of liquid samples.Tetramethylammonium hydroxide (TMAH) has been used as a ‘tissue solubilizer’ for various biological samples prior to analysis for major and minor inorganic elements by flame,18–21 furnace atomic absorption spectrometry,21–25 and inductively coupled plasma (ICP) atomic emission spectrometry, 26,27 and recently electrothermal vaporization ICP-mass spectrometry.28 This alkaline digestion with TMAH offers a rapid and simple approach to the preparation of a homogenized sample solution, which is a distinct advantage over conventional slurry preparation methods.In the present work, a rapid and simple method is presented for the analysis of total mercury in solid biological samples. TMAH was used to solubilize the samples and mercury was then determined by flow injection CVAAS with on-line decomposition of organomercury using KMnO4. Experimental Instrumentation A Perkin-Elmer (Norwalk, CT, USA) Model 4100ZL atomic absorption spectrometer in conjunction with a Perkin-Elmer FIAS-400 flow injection system and an AS-90 autosampler was used in this study.A Perkin-Elmer mercury electrodeless discharge lamp operated at 180 mA was used as the line source. The mercury absorbance was measured at 253.6 nm. The flow injection system is shown in Fig. 1, which consists of a 4 : 5 port injection valve, two peristaltic pumps, a reagentsample mixing chemifold and a glass gas–liquid separator (Part No.B09193772). A quartz cell with a path-length of 160 mm and a diameter of 7 mm was used as atomizer. The cell was heated to 200 °C to prevent condensation of moisture. Tygon pump tubings were used to deliver sample, reagents and withdraw waste. The reaction coils and connections were made of 0.9 mm id PTFE tubing. Sample and reagent flow-rates, including the concentrations of the reagents and argon stripping † On leave from Flow Injection Analysis Research Center, Department of Chemistry, Northeastern University, Shenyang, 110006, China.Analyst, June 1998, Vol. 123 (1215–1218) 1215gas are also shown in Fig. 1. The flow injection program used is shown in Table 1. Reagents and standard solution All chemicals used were of analytical-reagent grade unless specified otherwise. Deionized distilled high purity water (18 MW cm) was obtained from a Nanopure system (Barnstead– Thermolyne, Dubuque, IA, USA) TMAH (30% in methanol, Aldrich, Milwaukee, WI, USA) was used to solubilize the samples.An antifoaming agent (defoamer product No. 4528R158), obtained in the form of a household carpet cleaning additive from a local hardware store (Home Hardware Stores, Burford, Ontario, Canada), was used. Although any silicone-based antifoaming agent, such as Dow Corning DB110A, could be equally effective, this antifoaming agent is more cost-efficient. Due to its high viscosity and thus difficulties to pipette, the agent was diluted ten times (w/v) before use.NaBH4 (0.2%, m/v) solution was prepared daily (Alfa Chemicals Inc., Newburyport, MA, USA, 01950, caplets) in 0.05% (w/v) NaOH. To 200 ml of this solution, 1 ml of the diluted antifoaming agent was added. KMnO4 (0.2%, m/v) was also prepared daily in 15% sub-boiling nitric acid produced in-house from reagent grade feedstocks. This solution was kept in a dark brown bottle to prevent it from decomposing. Nitric acid (0.1 mol l21) was used as carrier.Mercury standard solution was prepared by dissolution of HgCl2 (gold star, Alfa Chemicals) in dilute nitric acid. A methyl mercury standard was prepared by dissolving MeHgCl (Alpha Division, Danvers, MA, USA) in an appropriate amount of propan-2-ol. Ethylmercury and phenylmercury stock solutions were prepared as described elsewhere.4 Working solutions were prepared daily by serial dilution with high purity water. The final solutions contained 4% (v/v) TMAH.National Research Council of Canada (NRCC) certified reference materials, DORM-2 (dogfish flesh), DOLT-2 (dogfish liver) and TORT-2 (lobster hepatopancreas) were used to assess the accuracy of the method. Sample preparation Nominal 0.25 g sub-samples of reference material biological tissues were weighed into 50 ml pre-cleaned screw-capped poly(propylene) bottles and 4 ml of TMAH added. Following the reaction of the tissue with the TMAH for approximately 5 min, high purity water was added to bring the volume to 25.0 ml (mass basis).Blanks were processed through an identical procedure. The resulting samples were ready to be analyzed in 30 min. Samples of DORM-2 were further diluted 4 times prior to analysis. Hg measurement The procedures for flow injection CVAAS are shown in Fig. 1 and Table 1. After the sample loop (500 ml) was filled with sample in step 1, the injection valve was switched to the inject position to introduce sample into the carrier stream (0.1 mol l21 HNO3) where it was then mixed with KMnO4 and NaBH4 solutions in sequence in the chemifold.The decomposition of organomercury by KMnO4 occurred in reaction coil L1 and mercury vapour was generated in reaction coil L2. The mercury vapour formed was separated in the gas–liquid separator and transferred by the argon carrier into the quartz cell assembly for detection. The peak height measurement mode was used. Quantitation was achieved by comparison of response against a simple calibration curve prepared from HgII standards processed in the same manner as above.Results and discussion Optimization of experimental parameters TMAH is an efficient reagent for solubilizing tissue samples. Compared to conventional slurry preparation methods, which require ultra-sonication or long-term stirring, this method is relatively simple, fast and less prone to contamination. In the present work, the fish muscle, liver and lobster tissues could be solubilized and homogenized in a few minutes, although the resulting sample ‘digest’ is neither clear nor colorless. On standing, the digest becomes less cloudy in appearance but no difference in final results was found for samples prepared 30 min or 3 months prior to determination. In addition, such TMAH digested solutions have been found to be stable for at least 1 year after preparation in the usual laboratory environment. 29 After the sample is digested in TMAH, the mercury is retained in its original species.Although NaBH4 can reduce both inorganic and organic forms of mercury, their resulting CVAAS sensitivities were different.30 This was also confirmed in this study. The slopes of methyl- and ethylmercury calibration curves were about 75% and 60%, respectively, of that for inorganic Hg whereas the phenylmercury slope was about 35% of the inorganic mercury slope, even when the quartz tube atomizer was heated to 800 °C. The possible reasons for lower sensitivities for organomercury could be a slower reduction process, an accompanying reduced rate of mercury release, and lower atomization efficiency.It was thus impossible to measure the total mercury if the sample was directly run after TMAH digestion. Therefore, it was necessary that all forms of mercury in the sample be oxidatively converted to HgII prior to reduction to elemental Hg. In the present work, KMnO4 was used to decompose the organomercury species.The decomposition was achieved by on-line addition of KMnO4. Since the sample was dissolved in alkaline TMAH media, nitric acid was added to the KMnO4 so as to acidify the sample digest and provide a favorable environment for decomposition of organomercury. The effects of the KMnO4 and nitric acid concentration on the decomposition efficiency of organomercuric species are shown in Fig. 2 and Fig. 3, respectively. Fig. 1 Flow injection manifold for CVAAS with on-line KMnO4 oxidation.GLS, gas-liquid separator; L1, L2, reaction coils, 15 and 50 cm long, respectively; P1, P2, peristaltic pumps; V, injection valve; W, waste. Table 1 FIAS-400 programme Step. No. Time/s Pump 1 Pump 2 Valve position Read Prefill 15 Off On Fill 1 10 On On Fill 2 15 On Off Inject Read 1216 Analyst, June 1998, Vol. 123Methyl-, ethyl- and phenyl-mercury were chosen to investigate the decomposition efficiency. It should be mentioned that these organic compounds used in the decomposition test cannot represent all of the organic mercurials that may possibly exist in marine biological samples, although methyl-, ethyl- and phenylmercury are the most commonly reported organomercurials. 31–33 Of these, methylmercury is the predominant and most toxic organic species in biological samples.Thus, the decomposition test should be representative of the majority of cases. As shown in Fig. 2, increasing the concentration of KMnO4 resulted in increased decomposition of the organomercury species. Complete decomposition was obtained using a concentration of 0.2% or higher, while the sensitivity for inorganic mercury remained constant.Therefore, 0.2% KMnO4 was chosen for further study. In addition to its effective decomposition of organomercurials, KMnO4 enhanced the HgII signal by about 20% (Fig. 2), in agreement with observations by Guo and Baasner.15,16 Thus, despite the dilution of the sample plug in the carrier stream, which was unavoidable during the on-line addition process, both the peak height and area of the signal increased rather than decreased when KMnO4 was added.Fig. 3 shows the effect of nitric acid concentration (in KMnO4) on the sensitivities of inorganic and methyl mercury. Since the ethyl- and phenyl- mercury followed the same trend as methylmercury, only methylmercury was selected for illustration. As seen in Fig. 3, 15% nitric acid (in 0.2% KMnO4) provides favorable conditions for both decomposition and cold vapour generation.The effect of NaBH4 concentration on the mercury sensitivity was also investigated. As shown in Fig. 4, the response remained almost constant beyond 0.2%. Due to the organic matter which still remained in the sample after TMAH digestion, the reduction reaction became so vigorous when the concentration of NaBH4 was higher than 0.3% that excessive foam was created. This made the gas–liquid separation ineffective. A 0.2% NaBH4 solution was thus chosen for further study.In the present system, the time for the oxidation reaction between KMnO4 and the organomercuric species was determined by the reaction coil, L1 in Fig. 1. This reaction was very fast, permitting a short length of 15 cm to be used while ensuring complete decomposition. Figures of merit The system was calibrated with a series of HgII standards having concentrations up to 30 mg l21 . Calibration graphs obeyed the equation H = 9.82 3 1023 C + 1.15 3 1023 (correlation coefficient r2 = 0.999), where H is peak-height absorbance and C is the mercury concentration in mg l21.A blank, limited by instrumental noise of 0.3 ± 0.04 mg l21 was obtained, yielding a limit of detection of 0.1 mg l21 (based on 3 s of a blank TMAH solution). The relative standard deviation of the signal at a level of 20 mg l21 Hg was 1.3% (n = 11). After the sample is digested, sample throughput is about 100 h21, i.e., about 30 samples per hour measured in triplicate.Accuracy The accuracy of the method was evaluated by analyzing a suite of certified marine biological reference materials, i.e., DORM- 2, dogfish flesh material; DOLT-2, dogfish liver tissue; and TORT-2, lobster hepatopancreas. The determined values for total mercury agree with the certified values (see Table 2). Conclusions It has been demonstrated that FI–CVAAS with on-line decomposition of organomercury is a fast and reliable method Fig. 2 Effect of KMnO4 concentration on the mercury signal for HgII, methyl-, ethyl-, and phenyl-mercury.The concentrations of these four species were 20 mg l21 Hg. The concentration of HNO3 in KMnO4 was 15%. Other conditions are the same as shown in Fig. 1. /Hg; -methyl Hg; : ethyl Hg; 5 phenyl Hg. Fig. 3 Effect of HNO3 concentration on the mercury signal for HgII and methylmercury. The concentration of KMnO4 was 0.2%. Other conditions are the same as shown in Fig. 1. / Hg; - methyl Hg.Fig. 4 Effect of NaBH4 concentration on the inorganic mercury signal. Other conditions are the same as shown in Fig. 1. Table 2 Analytical results for certified reference materials (mg g21) Sample Certified This work* DORM-2 4.64 ± 0.26 4.53 ± 0.072 DOLT-2 2.14 ± 0.28 2.04 ± 0.052 TORT-2 0.27 ± 0.06 0.27 ± 0.014 * Mean values ± standard deviations (n = 3). Analyst, June 1998, Vol. 123 1217for the routine analysis of total mercury in solid biological samples. Due to the simplicity of slurry preparation using TMAH and the fully automated measurement of mercury by FI– CVAAS, the risks of analyte loss and contamination are considerably reduced and large batches of samples can be rapidly processed.References 1 Vandecasteele C., and Block, C. B., Modern Methods for Trace Element Determination, Wiley, Chichester, 1993, ch. 5. 2 Farey, B. J., Nelson, L. A., and Rolph, M. G., Analyst, 1976, 103, 656. 3 Hawley, J. E., and Ingle, J. D., Jr., Anal.Chem., 1975, 47, 719. 4 Baxter, D. C., and Frech, W., Anal. Chim. Acta, 1990, 236, 377. 5 Hanna, C. P., Tyson, J. F., and McIntosh, S. A., Anal. Chem., 1993, 65, 653. 6 Ping, L., and Dasgupta, P. K., Anal. Chem., 1989, 61, 1230. 7 Ahmed, R., May, K., and Stoeppler, M., Fresenius’ Z. Anal. Chem., 1987, 326, 510. 8 Hatch, W. R., and Ott, W. L., Anal. Chem., 1968, 40, 2085. 9 Brandenberger, H., and Bader, H., At. Absorpt. Newsl., 1967, 6 101. 10 Murphy, J., Jones, P., and Hill, S.J., Spectrochim. Acta, Part B, 1996, 51, 1867. 11 Harri, L., Jauhiainen, T., and Peramaki, P., At. Spectrosc., 1997, 18, 102. 12 United States Environmental Protection Agency, Mercury Method 245.1 (Manual Cold Vapour Technique), EPA, Washington, DC, 1974, p.1. 13 Welz, B., Tsalev, D. L., and Sperling, M., Anal. Chim. Acta, 1992, 261 91. 14 Guo, G.-Z., and Baasner, J., Anal. Chim. Acta, 1993, 278, 189. 15 Guo, G.-Z., and Baasner, J., Talanta, 1993, 40, 1927. 16 Guo, G.-Z., and Baasner, J., Anal.Chim. Acta, 1996, 320, 171. 17 Hanna, C. P., and McIntosh, S. A., At. Spectrosc., 1995, 16, 106. 18 Jackson, A. J., Michael, L. M., and Schumacher, H. J., Anal. Chem., 1972, 44, 1064. 19 Murthy, L., Menden, E. E., Eller, P. M., and Pertering, H. G., Anal. Biochem., 1973, 53, 365. 20 Kaplan, P. D., and Blackstone, M., Arch. Environ. Health, 1973, 27, 387. 21 Zhou, Y., Wong, M. K., Koh, L. L., and Wee, Y. C., Talanta, 1996, 41, 1061. 22 Gross, S. B., and Parkinson, E. S., At. Absorpt. Newsl., 1974, 13, 107. 23 Alt, F., and Massmann, H., Spectrochim. Acta, Part B, 1978, 33, 337. 24 Tan, Y.-X., Marshall, W. D., and Blais, J.-S., Analyst, 1996, 121, 483. 25 Tan, Y.-X., and Marshall, W. D., Analyst, 1997, 122, 13. 26 De Boer, J. L. M., and Maessen, F. L. M. J., Spectrochim. Acta, Part B, 1983, 38, 739. 27 Uchida, T., Isoyama, H., Yamada, K., Oguchi, K., and Nakagawa, G., Anal. Chim. Acta, 1992, 256, 277. 28 Willie, S. N., Gregoire, D. C., and Sturgeon, R. E., Analyst, 1997, 122, 751. 29 Jimenez, M. S., and Sturgeon, R. E., J. Anal. At. Spectrom., 1997, 12, 597. 30 Oda, C. E., and Ingle, J. D., Jr., Anal. Chem., 1981, 53, 2305. 31 Tseng, C.-M., de Diego, A., Martin, F. M., Amouroux, D., and Donard, O. F. X., J. Anal. At. Spectrom., 1997, 12, 743. 32 Organometallic Compounds in the Environment, ed. Craig, P. T., Wiley, New York, 1986, pp. 65–110. 33 Analysis of Contaminants in Edible Aquatic Resources, ed. Kiceniuk, J. W., and Ray, S., VCH, New York, 1994, pp. 175–204. Paper 8/02000K Received February 13, 1998 Accepted April 3, 1998 1218 Analyst, June 1998, Vol. 123

ISSN:0003-2654    

DOI:10.1039/a802000k

出版商:RSC    

年代:1998

数据来源: RSC