摘要:

Critical Review Applications of Chromatography in Art Conservation: Techniques Used for the Analysis and Identification of Proteinaceous and Gum Binding Media Sarah L. Vallance Departments of Chemical and Life Sciences/Historical and Critical Studies, University of Northumbria at Newcastle, Ellison Place, Newcastle upon Tyne, UK NE1 8ST Summary of Contents Introduction Paint Media Analysis of Paint Media Analysis of Proteinaceous Media Analysis of Gum Media Conclusions References Keywords: Gas chromatography; high-performance liquid chromatography; proteinaceous media; polysaccharide gums; art conservation; review Introduction There is a need for the development of accurate and reliable methods for the analysis of samples from works of art in order to meet the specific requirements of the conservator.A detailed awareness of the constituents of paint layers from easel paintings, for example, provides the conservator with the background information required to facilitate the design of the optimum safe conservation/restoration treatment plan, taking into account the nature of the original materials used.From a practical aspect, specific knowledge of the nature of the media in particular works may offer some indication as to why some paintings are in better condition than others of a similar age. This type of information also enables the art historian to come to an educated and informed conclusion regarding the age and potential origin of an unknown work. In addition, it is feasible that a counterfeit work may be revealed, if the artist has been careless enough to use pigments or binders that are historically incorrect.The most important factors in all the techniques to be discussed are the extensive methods of sample preparation deemed necessary, since the chromatographic techniques used in the analyses are not extraordinary. The microscopic samples characteristic of work in this area are notoriously problematic to deal with and sensitivity is paramount, which consequently places great importance on the laboratory skills of the conservation scientists themselves.Paint Media Since man first learned to paint, artists have used a diverse variety of binding media for their pigments, ranging from natural gums and oils to proteinaceous materials such as egg (glair and tempera), milk (casein) and collagen glues made from animal skins and skeletons, for example. Chemically, oils and fats are the glycerol esters of aliphatic acids, typically those of the 18-carbon series.1 Oils tend to be liquid at room temperature, whereas fats are solid or semi-solid and greasy to the touch.If an oil possesses sufficient di- and triunsaturated fatty acids in its triglyceride components it will ‘dry’, i.e., polymerise, giving rise to a semi-solid. The drying oils most widely used in western European art are linseed (obtained from the seeds of the flax), walnut and poppy, although the time when they were first used for painting purposes is not known.Analytical evidence suggests that linseed oil was used in northern Europe from, at latest, the 13th Century, whilst in Italy, where oil painting was introduced in the 15th Century, walnut oil was initially preferred, although linseed oil became more common there from the 16th Century.2 In most recent years, the use of other oils, such as safflower and tung, has become more common.3–5 Collagen6 is the predominant proteinaceous material in animal skeletons (both skin and bone), representing one third of the total protein present in mammalian organisms.Collagen production in the body is preceded by the production of procollagen, a much larger biosynthetic precursor molecule, which is then degraded by specific enzymes resulting in collagen. There are a number of different types of collagen, but they all consist of molecules which contain three polypeptide a- chains in a triple helix conformation.Each a-chain has an amino acid sequence which is mainly a repeating structure, with glycine as every third residue and either proline or hydroxyproline often preceding the glycine residues. The various types of collagen can be distinguished by the slight differences in the sequence of their constituent amino acids. Animal and fish collagen glues are widely used as strong adhesives for wood, binders in the preparation of grounds, size for canvas and pigment binders in decorative paints.7 Preparation is relatively simple, involving the treatment of specific collagen-containing animal or fish tissues with hot water.When the leached solution cools, it forms a gelatinous mass; gelling occurs as a result of the partial decomposition of the tissues. If Sarah Vallance graduated in 1992 with a BSc in Applied Chemistry from the University of Northumbria at Newcastle. After a period in industry, she returned to the university and recently completed her programme of research for the award of PhD.She is at present preparing her thesis, The Development and Application of Chromatographic Methods in the Characterisation of Artists’ Media, for examination. Analyst, June 1997, Vol. 122 (75R–81R) 75Rthe extraction process is performed at a lower temperature (e.g., 80–90 °C), the gelatine solutions formed from the collagen are usually clear and light yellow. A glue which has been prepared under harsher conditions is less pure, much more turbid and viscous and is considerably darker in colour.Casein is a mixture of related phosphoproteins found in milk products.8 It is produced in mammary tissue from amino acids which have been supplied by the blood, contains all the common amino acid residues and is particularly rich in those designated as essential, e.g., leucine. Casein, egg albumin (glair) and yolk (tempera) have found uses as pigment binders, temporary varnishes and sealant over primers or grounds; casein additionally provides one of the strongest natural adhesives known, much used by cabinet makers and joiners. There are numerous recipes for the preparation of egg-based media containing ingredients as varied as linseed oil, fig tree shoots and vinegar, but the preparation of casein is uncomplicated: skimmed milk is heated to 35 °C, then the mixture is acidified to pH 4.8 with hydrochloric acid.It is allowed to stand and the casein solids are separated from the supernatant liquid, then washed with hydrochloric acid (pH 4.8).The casein thus prepared is a technically pure, slightly hygroscopic white powder.8 Gums are a group of non-crystalline, polysaccharide materials which can be found in vegetable matter and are often exuded when a plant is ‘wounded’. They are either water-soluble or water-dispersible compounds with a complex composition, usually consisting of a number of sugars (e.g., galactose and mannose) plus the acids derived from them (e.g., galacturonic acid).Gums differ from gelatines and other proteinaceous materials which form similar mucilaginous solutions in water, in that they contain virtually no nitrogen.9 Plant gums are commonly found as adhesives and binders. Gum arabic is primarily used as a painting medium, but others such as gum tragacanth (used as a medium for painting on linen) and cherry gum (which gives an enamel-like effect when mixed with egg or casein emulsions) are used less frequently.10 Documents show that gums have been used as binding media and sizing materials for centuries: gum was used as a replacement for sun-dried oil as early as the 12th Century.11 The artist’s selection of a binding medium was/is governed not only by the type of pigment used, but also by historical factors such as their location and the period of the piece: fashion and scientific progress have ensured that different materials have enjoyed periodic popularity. Analysis of Paint Media The analyses of oil-based media are well documented,12–15 but any information on the nature of other binding media has primarily been obtained via microscopic staining methods or crude solubility tests.16–21 Existing methods of analysis have provided the basis for separating the general categories of binding media (oil, gum and protein) by qualitative means, e.g., differential staining of cross-sections can distinguish between oil and protein layers.16 If a polysaccharide is heated with acid, a furfural-type compound results; these compounds react with aniline/aromatic amines to yield coloured Schiff’s bases.This reaction forms the basis of a spot test for polysaccharide media.22 Further confirmatory tests17,18 require the use of mixtures of alcohol and neutral iron chloride, or sodium hydroxide with copper sulfate (Fehling’s solution). An advantage of such micro-analytical techniques is their applicability in situ, but the results can be misleading, with false-positive and -negative results being observed.For this reason, these techniques are best used in conjunction with other analytical methods. These simple early qualitative techniques, including paper and thin-layer chromatography,19–21 are sufficient where only the general category of the binding medium is needed or, for example, where the binder may contain unique constituents, such as hydroxyproline in gelatine.However, where the amino acid or sugar composition of a proteinaceous material or gum is insufficiently distinctive, as in the cases of egg and milk proteins, for example, only quantitative chromatographic techniques will allow differentiation between similar binding media. Analysis of Proteinaceous Media Quantitative amino acid analysis by means of ion-exchange chromatography of protein hydrolysates was introduced by Moore and Stein23 in 1954. A sample of hydrolysed protein was applied to a column of sulfonic acid resin (pH 3) and an eluent with increasing pH and salt content was pumped through the column.The eluted amino acids, 18 in total, were detected by their absorbances produced with ninhydrin reagent. The required sample size was in the region of 0.3 mg of protein. In 1969, Keck and Peters24 utilised this method and reported the successful analysis of several antique and modern art specimens. Standard samples of various media were prepared both with and without pigment and dried at room and elevated temperature.After hydrolysis with 6 m HCl, under vacuum, each sample was applied to a sulfonic acid column, which was operated at 60 °C with a pH gradient from 2.87 to 5.00, a citrate concentration gradient from 0.05 to 0.267 m and a chloride concentration gradient from 0.18 to 0.25 m. The eluted amino acids were detected by absorbance measurement with ninhydrin. The percentage amino acid content was calculated for each of the standard media samples and these results were used for the purpose of identifying the unknown samples from works of art.This method facilitated the differentiation between gelatine, casein, glair, tempera and even horn found in the samples taken from works of art. There are a number of major disadvantages associated with ion-exchange chromatography in this area. Specific singlepurpose equipment is very expensive and is probably unattainable by most museums and galleries, where both space and funds are at a premium.A pH gradient, such as that used above, is difficult to control precisely, and the technique also involves a lengthy and complex method of sample preparation, where the required sample size of paint (between 1 and 4 mg to give 300 mg of proteinaceous material) is considered large, when put into the context of a fragment to be removed from a valuable work of art. The gas chromatographic (GC) determination of amino acids, obtained via the hydrolysis of proteinaceous material under acidic conditions, has been achieved using trimethylsilyl derivatives.25,26 Amino acids form two products when silylated, since the carboxyl group is easier to silylate than the amino group: under mild conditions, using hexamethyldisilazane (HMDS) as the silylating agent, the major product is the silyl ester.A stronger donor is usually required by the amino group and silylation has been achieved using N-trimethylsilyldiethylamine (TMSDEA) or N,O-bis(trimethylsilyl)acetamide (BSA), yielding the silylamine silyl ester.25 Masschelein-Kleiner26 described a GC method for the determination of trimethylsilyl derivatives of amino acids.BSA was used to derivatise pure amino acids and those obtained from the hydrolysis of proteinaceous binding materials, namely animal gelatine, casein emulsion, egg albumin and egg yolk. The volatile derivatives were injected on to a silanised Chromosorb W column (containing 10% UCC W982) with temperature programming from 90 to 220 °C at 2 °C min21.The relative amounts of the constituent amino acids were determined, thus allowing the characterisation of aged proteinaceous media. 76R Analyst, June 1997, Vol. 122Further GC methods for amino acids utilised their volatile butyl ester and N-acetylmethyl ester derivatives.27,28 Kenndler et al.27 reported a method for the hydrolysis of proteinaceous binding media from priming and paint layers of 16th and 18th Century easel and wall paintings.The efficiencies of hydrolysis of proteins by hydrochloric acid and an ion-exchange mechanism were compared. Derivatisation of the amino acid residues was achieved in two stages: first, the derivatisation of the carboxylic group, then that of the amino group. The carboxylic functions were converted into butyl esters by the addition of butanol (containing dissolved hydrogen chloride, concentration 3 mol l21) and the amino groups were then acylated with trifluoroacetic anhydride, using dichloromethane as the solvent.After evaporation of the solvent, the residue was dissolved in ethyl acetate, which contained an internal standard, and aliquots of the sample were injected directly on to the column (DB-5 capillary). A linear temperature gradient programme from 100 to 280 °C at 10 °C min21 achieved the separation of the amino acids and, by determining the relative peak areas of each, the characteristic profile of amino acids in each of the binding media was revealed.White28 hydrolysed proteinaceous material, from the gesso, ground layers and upper paint layers of a selection of Italian works, with hydrochloric acid, then de-ionised the samples on a small column of Amberlite IR-120H ion-exchange resin. Methylation of the carboxylic acid function was achieved with methanol in the presence of anhydrous hydrogen chloride gas, then the amino group was acylated with trifluoroacetic anhydride.Samples were concentrated under nitrogen, taking care not to evaporate to dryness as N-acetylmethyl esters of amino acids are extremely volatile and some loss of analytes would be incurred. Aliquots of sample were injected on to the packed column (1% XE-60 cyanosilicone gum on 100–120 mesh Diatomite CQ support, coated from acetone) and a linear temperature gradient from 80 to 210 °C at 3 °C min21 ensured separation of the amino acid components. Relative peak area were used to establish an amino acid fingerprint for each of the protein media types and characteristic amino acid ratios were used to confirm their identities.There are a number of major problems with these methods. The last technique requires a sample approximately three times the size of that for analysis of oil media, since compensation should be made for losses incurred at every stage of the sample preparation. It should also be noted that losses can occur during acid hydrolysis: the presence of sugars and carbohydrates in the sample can cause the elimination of amino acids as humins, which cause darkening of the hydrolysate and the formation of insoluble matter.Humins is the collective term for a mixture of coloured compounds (resembling naturally occurring melanins) produced during the acid hydrolysis of many proteins. A major contributory factor in the formation of humins is the presence of tryptophan and amino sugars (e.g., galactosamine, glucosamine) or carbohydrates. These compounds undergo extensive degradation during acid hydrolysis, resulting in humin production, possibly via the Maillard reaction.29 Humin formation may be prevented if hydrolysis is performed in 80% aqueous ethanol, in the presence of an ionexchange resin, for up to 10 h at 95 °C.30 It is thought that the amino acids are effectively removed from the solution and retained on the resin by their nitrogen function, hence hindering the formation of Schiff’s base condensation products between the sugars’ keto or aldehyde groups and the amino groups.The immobilised amino acids can then be eluted from the ionexchange resin by simply eluting with 10% ammonia solution. The fatty acid content of eggs has been quantified by Mills and White31 via GC analysis: the presence of tempera can be confirmed by the absence of azelate along with the presence of both palmitate and stearate non-drying oils, i.e., those which thicken at elevated temperatures but do not dry to a skin, even after prolonged exposure. Eight year old paint films, containing lead white and egg yolk medium, were analysed following saponification with potassium hydroxide and methylation of the acids with diazomethane.The conditions used (i.e., wide bore BP1 column with on-column injection and temperature programming from 110 to 310 °C at 7 °C min21) revealed the presence of the methyl esters of saturated palmitic and stearic acids, plus a variable amount of unsaturated oleic acid.It was reported that egg fats contain only small amounts of unsaturated acids and the formation of a small amount of azelate is not uncommon in a tempera medium, although amounts can range from negligible to almost one third of the palmitate present: in a pure oil film, the azelate peak would be at least equal to that of the palmitate methyl ester. This obviously leads to ambiguity in the results, further increased when non-drying oils are actually added to a blood-glue preparation.32 Skans and Michelsen33 observed that some animal glue preparations can contain up to 10% non-drying oils, hence it is not inconceivable that a sample could be wrongly identified as egg yolk.Nowik34 reported on a GC method which facilitated the simultaneous determination of both amino acids and fatty acids, which may be found together in mixed media such as tempera. Samples of proteinaceous and oil media were acid hydrolysed (6 m HCl), then neutralised with calcium carbonate.The samples were treated with an aqueous ethanol–pyridine solution prior to derivatisation, which was achieved with ethyl chloroformate (ECF), then the volatile N-(O,S)-ethoxycarbonyl ethyl ester derivatives were extracted with chloroform (containing 1% ECF). Samples were injected (splitting ratio 1 : 20) on to a CP Sil-19 CB capillary column, with a temperature programme from 100 to 300 °C at 30 °C min21.The same method was previously reported for the determination of amino acids alone; the investigations involved comparisons of methods of sample preparation, analysis and column type.35 Schilling and co-workers36–38 recently reported on the GC determination of ethyl chloroformate derivatives of amino acids from proteinaceous media; the studies also included investigations into the effects of pigments and ageing on the identification of proteins from art objects and details of a statistical approach to the interpretation of experimental results.Derivatisation was performed as described above, except that acid hydrolysis of the samples was achieved using hydrochloric acid (6 m) in the vapour phase. Samples were injected on to a capillary column coated with HP-INNOWAX, with a temperature programme from 70 to 250 °C at 27 °C min21. In recent years, the determination of amino acids by reversedphase high-performance liquid chromatography (RP-HPLC) has developed significantly, primarily owing to the speed and sensitivity of the technique compared with more traditional specialised amino acid analysers.Currently, RP-HPLC following pre-column derivatisation is one of the most widely used methods for the determination of amino acids, since it provides a good selection of both derivatising agents and detection techniques. The technique lends itself well to the field of conservation science: the occurrence of proteinaceous material in art objects coupled with the extremely small samples characteristic of the work means that RP-HPLC is an idea tool for the identification of samples taken from works of art.Erhardt et al.,39 reported a method for the determination of phenylthiocarbamyl derivatives of amino acids from proteinaceous material taken from works of art. Samples were hydrolysed under acidic conditions, dried, buffered with triethylamine, redried, then derivatised with phenyl isothiocyanate (PITC).The derivatisation reaction was stopped by the evaporation of the PITC and the samples were determined by RP-HPLC after dissolution in aqueous disodium hydrogenphosphate buffer. Samples were injected on to a C18 column and Analyst, June 1997, Vol. 122 77Ra ternary solvent system was used as the mobile phase (water– acetonitrile–acetate buffer), with a gradient programme from 0 : 6 : 94 to 16 : 28 : 56. A reasonable separation of the amino acids was shown in the paper, but there was very little discussion about the results themselves.Halpine40,41 has made extensive use of RP-HPLC for the amino acid analysis of proteinaceous matter from art objects. In a recent study,40 samples taken from ‘The Annunciation with St. Francis and St. Louis of Toulouse’, a series of 15th Century Italian tempera panels by Cosimo Tura, were submitted for amino acid analysis.42 Hydrolysis of the protein material was performed with acid vapour and derivatisation of the liberated amino acids, following buffering and drying of the hydrolysate, was achieved with PITC.The samples were analysed by RPHPLC42 with a binary solvent system of acetate buffer– acetonitrile. A C18 column was used with a gradient programme to separate the 18 amino acids deemed necessary for the identification of proteinaceous material. The use of norleucine, an unnatural amino acid, as an internal standard facilitated the quantification of the constituent amino acids and, using this method, a number of samples of animal glue media and egg/glue media were identified. A problem with this method is in the choice of derivatising agent: the PITC amino acid derivatives degrade in solution and so must be kept at 24 °C until required for analysis; only then can the samples be solubilised in the dilution buffer.Concern has been expressed with respect to the effects of mineral pigments on the reproducibility of the analyses.43 Halpine40 performed tests in which copper and calcium based pigments were added separately to proteinaceous reference samples, confirming that the presence of copper lowers the yield of all the amino acids whilst calcium reduced the recovery of aspartic and glutamic acids.This means that calcium could hinder the identification of a casein medium and copper could lower the amino acid levels to those associated with background interference, making interpretation of analytical results more difficult. Halpine postulated that the copper somehow affected the hydrolysis procedure, but an alternative explanation is that it prevents complete derivatisation of the amino acids in the hydrolysate owing to the formation of copper–amino acid complexes.Evidence for this is in the improved recovery of amino acids obtained when a small amount of EDTA solution is added to protein hydrolysates prior to derivatisation. The copper preferentially complexes with the EDTA, leaving the amino acids free to participate in the derivatisation reaction.44 It is desirable to remove any pigment from the sample, if possible, and Halpine achieved this using a simple extraction method.41 He sampled an egg tempera panel with gesso ground, which had been prepared in the laboratory, and 30 ml of HPLC grade water was added to each sample in a hydrolysis tube.After thorough mixing, the samples were left to stand for 1 h, sonicated to break up any remaining larger particles, then centrifuged for a maximum of 15 min at full speed. The watersoluble proteinaceous material was then removed, evaporated to dryness and analysed by HPLC after derivatisation with PITC, in addition to the insoluble material which remained in the precipitate.This simple technique, besides removing unwanted contaminants from the samples, proved useful for the determination of mixed media, since the degree of water solubility was found to vary depending on the protein types present.However, paintings are not the only art objects of importance: stone sculptures, frescoes and stuccoes provide other sources of samples posing interesting questions for the conservator. In particular, protein levels in these samples are typically much lower than those seen in easel paintings, for example, so the sensitivity of the method of analysis is paramount. Ronca45 used RP-HPLC, following derivatisation of the amino acid residues with PITC, to analyse both artificial samples and those taken from a number of 13th Century French stone sculptures and Italian stuccoes, a 16th Century Italian external fresco and the gesso ground of a 15th Century Italian wooden statue.Proteinaceous material was extracted from the matrices by a variety of chemical methods, the most successful being the use of 1 m sodium hydroxide solution at 80 °C for 3 h followed by colorimetric determination with Folin’s reagent. The amino acid composition of the extracted protein was achieved via RPHPLC, preceded by direct acid hydrolysis of the proteinaceous material, desalting of the hydrolysate using a sulfonic resin column and final derivatisation with PITC.The analyses revealed the presence of gelatine and egg proteins in the samples, which corroborated historical information on the nature of the materials commonly used by artists in those periods. The use of 9-fluorenylmethyl chloroformate (FMOC) as a derivatising agent in the RP-HPLC analysis of proteinaceous media was first reported by Grzywacz.46 Samples were hydrolysed under acid vapour conditions, dried and diluted with borate buffer (pH 8.5) prior to derivatisation with FMOC.Samples were injected on to a 3 mm Spherex C18 (ODS) column with a binary gradient elution programme based on the method developed by Haynes et al.;47 the eluents were (a) 50 mm sodium acetate and 7 mm triethylammonium acetate with 10% acetonitrile, adjusted to pH 6.5 with acetic acid, and (b) acetonitrile–water (90 + 10, v/v).Standard proteinaceous media and a number of museum samples were analysed and identified using this method. Vallance et al.44 also utilised an RP-HPLC method for the analysis of proteinaceous binding media taken from works of art. Samples were hydrolysed with concentrated hydrochloric acid and buffered, then, following investigations into pigment interferences, an aliquot of EDTA solution was added to any sample containing a copper-based pigment prior to derivatisation of the hydrolysate with FMOC.Separation and analysis were achieved using an ODS2 column, eluting with a mobile phase of acetate buffer (pH 4.2) and acetonitrile and operating a gradient programme from 20 to 100% acetonitrile over 70 min. A number of contributory factors recommend the selection of FMOC as the derivatising agent for the amino acid analysis. It undergoes a rapid reaction with primary and secondary amino acids and favours mild, aqueous conditions.The derivatised product yield is high48,49 and the derivatives themselves are stable at room temperature for at least 2 weeks.50 The FMOC moiety is both a good UV chromophore and highly fluorescent, allowing a choice of emission or absorption detection techniques; the FMOC amino acid derivatives can be detected at limiting levels in the low femtomole range by excitation at 260 nm.51 However, a disadvantage associated with FMOC is that it also reacts with any water present in the sample51 to form the corresponding alcohol as a hydrolysis product; the extent of any undesirable hydrolysis products is minimised by the prompt extraction of the reaction mixture with hexane.Analysis of Gum Media Chemists have employed a variety of analytical techniques for the characterisation of carbohydrate compounds, probably the most widely used in recent years being GC. Since GC is dependent on the volatility of the analytes, it is necessary for the sugars to undergo some form of derivatisation reaction prior to analysis; the various derivatisation techniques employed for this purpose have been widely reported.52–69 An obvious problem associated with the use of any of these reported methods for the analysis of gums used in works of art is the sample size; many of these techniques will not be sufficiently sensitive for the microscopic samples which are available to the conservation scientist.Despite this, progress has been made in the analysis of gum media from art objects, 78R Analyst, June 1997, Vol. 122frequently employing a combination of TLC and GC techniques. Masschelein-Kleiner and co-workers70,71 used TLC and GC to determine the trimethylsilyl derivatives of sugars resulting from the hydrolysis of samples of the surface coating of a wooden Egyptian sarcophagus, dating from the 21st Dynasty. A mixed alumina–silica stationary phase was prepared for TLC, the eluent being propanol–ethyl acetate–water–25% ammonia solution; naphthoresorcinol was used for detection on the plates after separation.Separation and analysis by GC were achieved using an E301 silicone column (length 2.0 m, external diameter 3 mm) with a temperature programme from 160 to 200 °C at 1.7 °C min21. The analysis indicated the presence of gum tragacanth and, when the paint medium itself was analysed, a mixture of honey and gum tragacanth was revealed. Flieder72 utilised TLC to separate and identify gum media from a 16th Century manuscript.The gum samples were hydrolysed under acidic conditions, deacidified with an ionexchange resin, then separated on a silica plate using butanol– ethanol–water (57 + 27 + 16) as the eluent; the sugar components were revealed with naphthoresorcinol. Although not every component was separated, the identification of the medium as gum arabic was facilitated. Birstein73 also used a combination of TLC and GC plus IR spectrometry when studying problems associated with the binding media found in Asian wall paintings.Hydrolysates of samples (1–5 mg of polysaccharide material) were acetylated prior to analysis but, although the technique was sensitive enough to facilitate identification of the media studied, the sample preparation was lengthy, requiring 12 h for hydrolysis and 5 h for derivatisation. Birstein and Tul’chinsky74 employed IR techniques for polysaccharide identification in archaeological samples; spectra of artistically important gums and some samples removed from works of art were published, but the results were not particularly informative, only allowing the distinction between polysaccharides and water-soluble proteins.Furthermore, large samples were required in order to obtain any spectra. Szyszko75 investigated the nature of binding media used in the paint layers of three Egyptian epitaphal stelae, on wooden supports, two of which dated from the second millenium bc and the third thought to be from a much later period.TLC was used in these investigations, the results of which indicated gum tragacanth as the binding medium. Studies of paintings found in the Tomb of Nefertari76,77 at Luxor revealed an interesting local phenomenon. The watersoluble paint medium used in the works was found to contain no rhamnose component, which is usually indicative of gum tragacanth. However, when samples of gum taken from locally growing trees of the Acacia genus were analysed by GC they too were found to be lacking in rhamnose, whilst the remainder of the sugar content matched that seen in the samples taken from the paintings.It was therefore concluded that the paint medium was in fact gum arabic, despite all other commercial sources of the gum showing a rhamnose component. These findings could in turn mean that the samples from the Egyptian epitaphal stelae could actually contain gum arabic from the same local source, rather than gum tragacanth as originally concluded.78 Twilley79 published a report on the analysis and artistic applications of plant gums.Analysis of samples was achieved via a number of techniques, including GC of trimethylsilyl sugar derivatives and TLC. Erhardt et al.39 presented their findings of investigations into the GC analysis of gums to the American Institute for Conservation of Historic and Artistic Works (AIC). Samples were hydrolysed with trifluoroacetic acid (2.3 ml in 7.7 ml of water), then dried in a vacuum desiccator.The sugars were then converted into their oxime form with a pyridine solution of hydroxylamine hydrochloride prior to silylation with trimethylsilylimidazole. Samples were separated on a non-polar DBI (bonded polydimethylsiloxane) capillary column with a temperature programme from 100 to 275 °C at 10 °C min21. The resolution was improved using a more polar DB17 column, although higher temperatures were required, i.e., from 150 to 325 °C at 10 °C min21.In 1996, Bleton et al.80 reported a GC method for the analysis of ink samples from ancient manuscripts. Following methanolysis and silylation with trimethylsilylimidazole, samples were injected on to an SE-52 capillary column with a stepped temperature programme from 40 to 130 °C at 9 °C min21, then from 130 to 290 °C at 2 °C min21, the final temperature being held for a further 30 min. Reference samples of old ink and samples taken from a variety of manuscripts were analysed.Other chromatographic techniques can be utilised for the characterisation of sugar based compounds. Pyrolysis–mass spectrometry (Py–MS) was the technique employed by Wright81 in studies of Egyptian mummy cases. Samples of organic materials used in the construction of the cartonnages were examined by Py–MS: pyrograms were compared with those of a series of standard materials, facilitating the identification of around 50% of the samples. Polysaccharide gums were detected in samples from objects between 2000 and 4000 years old.Derrick and Stulik82 used pyrolysis–gas chromatography (Py–GC) to investigate the separation and identification of natural gums in works of art. Powdered gum samples were pyrolysed using a coil type probe: standard gum samples were pyrolysed at 700 °C, but this temperature was lowered for the analysis of more complex samples. Gums arabic, tragacanth, guar, ghatti and karaya all gave distinguishable and reproducible pyrograms, allowing their identification. It was observed that the pyrograms of gum–pigment mixtures differed from those of the standard gums, their peak patterns and intensities being altered. This effect was minimised by performing the pyrolysis of the samples at a lower temperature of 400 °C.Computational methods of pattern recognition were employed to assist with sample identification. In 1995, Williams and Langdon83 used gel permeation chromatography (GPC) to characterise gum arabic, identifying the three principal molecular mass components of the gum.Ion-exchange chromatography,84–86 be it anion or cation, was popular for a time, but proved to be less sensitive than HPLC methods. In 1976, Rabel et al.87 reported on a normal-phase partition HPLC method, using refractive index (RI) detection, for sugar determination which was sensitive down to around 80 mg; however, much of the work on the use of HPLC techniques for the analysis of sugars was recorded between 1980 and 1987.88–95 A variety of derivatisation techniques (e.g., post-column cuprammonium, dansylhydrazine, pre-column dabsylhydrazine) and detection methods (e.g., UV absorption, fluorescence, RI) have been used and sensitivities as low as 5 pmol have been achieved; it must be noted, however, that none of these latter studies were in the particular area of conservation science. Analytical methods used for the determination of polysaccharides in art objects have been reviewed previously,96 one in particular focusing on chromatographic techniques,97 although obviously neither contain details of the most recent work from 1992 to the present.Conclusions There are a number of important questions that the conservator/ conservation scientist must ask before deciding on an appropriate analytical technique: 1. What do I need to know? Is the exact identity of the binding medium necessary? 2.How much sample is available? 3. What pigments, if any, are likely to be Analyst, June 1997, Vol. 122 79Rpresent in the sample? 4. Have any conservation/restoration treatments been performed on the work previously? If so, what was the nature of any treatment, e.g., consolidation, retouching? The conservation scientist needs to select a technique which will give the maximum amount of information for the minimum amount of sample and sample preparation. For proteinaceous media, this would appear to be amino acid determination by RPHPLC using FMOC as the derivatising agent.Gas chromatography of silylated sugar derivatives is, at present, probably the best method for the identification of natural gums: however, as this is probably the least investigated area so far, major developments in methodology which would greatly improve sensitivity can be expected; the main problem with samples of gum media is the minute amount of actual medium present.Simple qualitative techniques such as TLC, microscopy and staining tests may be sufficient to indicate the basic media type used in a work, but as more and more works of art require conservation/restoration treatments it is crucial that the conservator has as much information as possible on the nature of any materials used by the artist, in order to avoid the loss or spoiling of any valuable and irreplaceable pieces. It is clear that the previous investigations into quantitative analyses are of immense value and as chromatographic techniques are continually developed and improved, increasing both sensitivity and reproducibility, their use in the area of art conservation becomes even more ideal.One possibility is the development analytical methods using capillary zone electrophoresis (CZE), a relatively new technique currently used for the determination of proteins, etc.; it could prove to be a useful method of analysis in this area, owing to its high sensitivity, but a thorough investigation into its suitability would be required.Microbore techniques should also find their use in conservation science, since they obviously suit the sometimes nanomole amounts of samples which are provided for analysis. The main concern of the conservation scientist is the availability of reliable and accurate analytical techniques suitable for use with the minute samples typically seen in this field of work. No doubt further research will result in the simplification of methods of sample preparation, possibly negating the need for clean-up procedures (e.g., the removal of pigments from samples) prior to analysis.However, any improvements which mean that the required sample size is reduced and the loss of valuable sample material is minimised or, ideally, eliminated will be wholeheartedly welcomed by the conservator and conservation scientist alike. References 1 Gettens, R. J., and Stout, G. L., Painting Materials, a Short Encyclopaedia, Dover, New York, 1966, p. 36. 2 Mills, J. S., and White, R., The Organic Chemistry of Museum Objects, Butterworth Heinemann, London, 2nd edn., 1994, pp. 36 and 171–172. 3 Eastlake, C. L., Materials for a History of Oil Painting, Dover, New York, 1960, vol. 2, p. 88. 4 Matsui, E., Sci. Pap. Jpn. Antiques Art Crafts, 1981, 26, 15. 5 Carillo y Gariel, A., Technica de la Pintura de Nueva Espan�a, Imprenta Universitaria, Mexico, 1946, pp. 42–48. 6 The Merck Index, ed. Windholz, M., Merck, Rahway, NJ, 10th edn., 1983, p. 2452. 7 Gettens, R. J., and Stout, G. L., Painting Materials, a Short Encyclopaedia, Dover, New York, 1966, pp. 25–27. 8 Gettens, R. J., and Stout, G. L., Painting Materials, a Short Encyclopaedia, Dover, New York, 1966, pp. 7–8. 9 Gettens, R. J., and Stout, G. L., Painting Materials, a Short Encyclopaedia, Dover, New York, 1966, pp. 28–29. 10 Doerner, M., The Materials of the Artist and Their Use in Painting, Harcourt, Brace, New York, 1934, pp. 223–224. 11 Laurie, A. P., The Materials of the Painters’ Craft, J. B. Lippincott, Philadelphia, 1911, p. 164. 12 Mills, J. S., Stud. Conserv., 1966, 11, 92. 13 Mills, J. S., and White, R., Nat. Galleryech. Bull., 1980, 4, 65. 14 Mills, J. S., and White, R., in Application of Science in the Examination of Works of art, ed. England, P. A., and van Zelst, L., Museum of Fine Arts, Boston, 1985, pp. 29–34. 15 Mills, J. S., and White, R., in Conservation and Restoration of Pictorial Art, ed.Bromelle, N., and Smith, P., Butterworths, London, 1976, pp. 72–76. 16 Gay, M. C., in Conservation and Restoration of Pictorial Art, ed. Bromelle, N., and Smith, P., Butterworths, London, 1976, pp. 78– 83. 17 Butler, C. L., and Cretcher, L. H., J. Am. Chem. Soc., 1929, 51, 1519. 18 Gettens, R. J., and Stout, G. L., Painting Materials, a Short Encyclopaedia, Dover, New York, 1966, p. 29. 19 Hey, M., Stud. Conserv., 1958, 3, 183. 20 Denniger, E., S.Afr. Adv. Sci. Spec. Publ., 1971, 2, 80. 21 Masschelein-Kleiner, L., Stud. Conserv., 1974, 19, 207. 22 Feigl, F., Spot Tests in Organic Analysis, Elsevier, Amsterdam, 5th edn., 1956, p. 372. 23 Moore, S., and Stein, W. H., J. Biol. Chem., 1954, 211, 893. 24 Keck, S., and Peters, T., Stud. Conserv., 1969, 14, 75. 25 Pierce, A. E., Silylation of Organic Compounds, Pierce, Rockford, IL, 1977, pp. 218–243. 26 Masschelein-Kleiner, L., in Conservation and Restoration of Pictorial Art, ed.Bromelle, N., and Smith, P., Butterworths, London, 1976, pp. 84–87. 27 Kenndler, E., Schmidt-Beiwl, K., Mairinger, F., and P�ohm, M., Fresenius’ J. Anal. Chem., 1992, 342, 135. 28 White, R., Nat. Gallery Tech. Bull., 1984, 8, 5. 29 Krampitz, G., Tierphysiol. Tierern�ahr. Futtermittelk, 1960, 15, 227. 30 P�ohm, M., Naturwissenschaften, 1961, 48, 555. 31 Mills, J. S., and White, R., The Organic Chemistry of Museum Objects, Butterworth Heinemann, London, 2nd edn., 1994, p. 171. 32 Brown, C., Bomford, D., Plesters, J., and Mills, J. S., Nat. Gallery Tech. Bull., 1987, 11, 85. 33 Skans, S., and Michelsen, P., Maltech. Restauro, 1986, 92, (2), 63. 34 Nowik, W., Stud. Conserv., 1995, 40, 120. 35 Hu�sek, P., J. Chromatogr., 1991, 552, 289. 36 Schilling, M. R., Khanjian, H. P., and Souza, L. A. C., J. Am. Inst. Conserv., 1996, 35, 45. 37 Schilling, M. R., Khanjian, H. P., and Souza, L. A. C., J. Am. Inst. Conserv., 1996, 35, 123. 38 Schilling, M.R., and Khanjian, H. P., in ICOM Committee for Conservation, 11th Triennial Meeting, Edinburgh, September 1996: Preprints, ed. Bridgland, J., James and James, London, 1996, vol. I, p. 211. 39 Erhardt, D., Hopwood, W., Baker, M., and von Endt, D., in Preprints of Papers Presented at the 6th Annual Meeting, American Institute for Conservation of Historic and Artistic Works, Washington, DC, 1988, pp. 67–84. 40 Halpine, S. M., Stud. Conserv., 1992, 37, 22. 41 Halpine, S. M., Conservation Research 1995: Studies in the History of Art, vol. 51, Monograph Series II, National Gallery of Art, Washington, DC, 1995. 42 Cohen, S. A., Heys, M., and Tarvin, T. L., The Picotag Method—a Manual of Advanced Techniques for Amino Acid Analysis, Millipore, Waters Chromatography Division, Milford, MA, 1989, pp. 2–3 and 11–12. 43 Chemistry and Biochemistry of the Amino Acids, ed. Barrett, G. C., London, 1985, pp. 376–396. 44 Vallance, S. L., Singer, B. W., Hitchen, S. M., and Townsend, J., LC– GC Int., 1997, 10(1), 48. 45 Ronca, F., Stud.Conserv., 1994, 39, 107. 46 Grzywacz, C. M., J. Chromatogr. A, 1994, 676, 177. 47 Haynes, P. A., Sheumack, D., Kibby, J., and Redmond, J. W., J. Chromatogr., 1991, 540, 177. 48 Matzner, M., Kurkjy, R. P., and Cotter, R. T., Chem. Rev., 1964, 64, 645. 49 Hall, M. K., J. Am. Chem. Soc., 1957, 79, 5439. 50 Seiler, N., in Handbook of Derivatives for Chromatography, ed. Blau, K., and Halket, J. M., Wiley, Chichester, 1993, p. 187. 51 Einarsson, S., Josefsson, B., and Lagerkvist, S., J. Chromatogr., 1983, 282, 609. 80R Analyst, June 1997, Vol. 12252 McInnes, A. G., Ball, D. H., Cooper, F. P., and Bishop, C. T., J. Chromatogr., 1958, 1, 556. 53 Bishop, C. T., Adv. Carbohydr. Chem., 1964, 19, 95. 54 Bishop, C. T., and Cooper, F. P., Can. J. Chem., 1960, 38, 388. 55 Sawardeker, J. S., Sloneker, J. H., and Jeanes, A., Anal. Chem., 1967, 39, 121. 56 Bj�orndal, H., Lindberg, B., and Svensson, S., Acta Chem. Scand., 1967, 21, 1801. 57 Sweeley, C. C., Bently, R., Makita, M., and Wells, W. W., J. Am. Chem. Soc., 1963, 85, 2497. 58 Beadle, J. B., J. Agric. Food Chem., 1969, 17, 904. 59 Honda, S., Kakehi, K., and Okada, K., J. Chromatogr., 1979, 176, 367. 60 Sullivan, J. E., and Schewe, L. R., J. Chromatogr. Sci., 1977, 15, 196. 61 Decker, P., and Schweer, H., J. Chromatogr., 1982, 236, 369. 62 Dmitriev, B. A., Backinowsky, L. V., Chizhov, O. S., Zolotarev, B. M., and Kochetkov, N.K., Carbohydr. Res., 1971, 19, 432. 63 Churns, S. C., J. Chromatogr., 1990, 500, 555. 64 Varma, R., Varma, R. S., and Wardi, A. H., J. Chromatogr., 1973, 77, 222. 65 Ha, Y. W., and Thomas, R. L., J. Food Sci., 1988, 53 (2), 574. 66 Aspinall, G. O., and Fairweather, R. M., Carbohydr. Res., 1965, 1, 83. 67 Aspinall, G. O., and McKenna, J. P., Carbohydr. Res., 1968, 7, 244. 68 Reinhold, V. N., Wirtz-Pietz, F., and Biemann, K., Carbohydr. Res., 1974, 37, 203. 69 Wiecko, J., and Sherman, W.R., J. Am. Chem. Soc., 1976, 98, 7631. 70 Masschelein-Kleiner, L., and Tricot-Marckx, E., Bull. Inst. R. Patrimoine Artistique, 1965, 8, 180. 71 Masschelein-Kleiner, L., Heylan, J., and Tricot-Marckx, F., Stud. Conserv., 1968, 13, 105. 72 Flieder, F., Stud. Conserv., 1968, 13, 49. 73 Birstein, V. J., Stud. Conserv., 1975, 20, 8. 74 Birstein, V. J., and Tul’chinsky, V. M., Khim. Prirod. Soedin., 1976, 1, 15. 75 Szyszko, W., Ochr. Zabytkow, 1972, 25, 170. 76 Mora, P., Mora, L., and Porta, E., in ICOM Committee for Conservation, 9th Triennial Meeting, Dresden, 26–31 August 1990: Preprints, ed.Grimstad, K., 1990, p. 518. 77 Palet, A., and Porta, E., in Congresa de Conservaci�on de Bienes Culturales: Valencia, 20–23 Setiembre de 1990, ed. Roig Picazo, P., 1990, p. 452. 78 Mills, J. S., and White, R., The Organic Chemistry of Museum Objects, Butterworth Heinemann, London, 2nd edn., 1994, p. 79. 79 Twilley, J. W., Adv. Chem. Ser., 1984, No. 205, 357. 80 Bleton, J., Coupry, C., and Sansoulet, J., Stud. Conserv., 1996, 41, 95. 81 Wright, M. M., J. Anal. Appl. Pyrol., 1987, 11, 195. 82 Derrick, M. R., and Stulik, D. C., in ICOM Committee for Conservation, 9th Triennial Meeting, Dresden, 26–31 August 1990: Preprints, ed. Grimstad, K., 1990, p. 9. 83 Williams, P. A., and Langdon, M. J., Chromatogr. Anal., 1995, Oct/ Nov, 5. 84 Mrochek, J. E., Dinsmore, S. R., and Waalkes, T. P., Clin. Chem., 1975, 21, 1314. 85 Verhaar, L.A. Th., and Kuster, B. F. M., J. Chromatogr., 1981, 210, 279. 86 Kesler, R. B., Anal. Chem., 1967, 39, 1416. 87 Rabel, F. M., Caputo, A. G., and Butts, E. T., J. Chromatogr., 1976, 126, 731. 88 Binder, H., J. Chromatogr., 1980, 189, 414. 89 Alpenfels, W. F., Anal. Biochem., 1981, 114, 153. 90 Grimble, G. K., Barker, H. M., and Taylor, R. H., Anal. Biochem., 1983, 128, 422. 91 Mopper, K., and Johnson, L., J. Chromatogr., 1983, 256, 27. 92 Vratny, P., Frei, R. W., Brinkman, U. A.Th., and Nielen, M. W. F., J. Chromatogr., 1984, 295, 355. 93 Rosenfelder, G., M�orgelin, M., Chang, J. W., Sch�onenberger, C. A., Braun, D. G., and Towbin, R. H., Anal. Biochem., 1985, 147, 156. 94 Hull, S. R., and Turco, S. J., Anal. Biochem., 1985, 146, 143. 95 Lin, J.-K., and Wu, S.-S., Anal. Chem., 1987, 59, 1320. 96 Mills, J. S., and White, R., The Organic Chemistry of Museum Objects, Butterworth Heinemann, London, 2nd edn., 1994, pp. 78– 80. 97 Matouösov�a, M., and Bucifalov�a, J., Sb.Restaur�at. Praci, 1989, 4, 60. Paper 6/06219I Received September 9, 1996 Accepted March 20, 1997 Analyst, June 1997, Vol. 122 81R Critical Review Applications of Chromatography in Art Conservation: Techniques Used for the Analysis and Identification of Proteinaceous and Gum Binding Media Sarah L. Vallance Departments of Chemical and Life Sciences/Historical and Critical Studies, University of Northumbria at Newcastle, Ellison Place, Newcastle upon Tyne, UK NE1 8ST Summary of Contents Introduction Paint Media Analysis of Paint Media Analysis of Proteinaceoum Media Conclusions References Keywords: Gas chromatography; high-performance liquid chromatography; proteinaceous media; polysaccharide gums; art conservation; review Introduction There is a need for the development of accurate and reliable methods for the analysis of samples from works of art in order to meet the specific requirements of the conservator.A detailed awareness of the constituents of paint layers from easel paintings, for example, provides the conservator with the background information required to facilitate the design of the optimum safe conservation/restoration treatment plan, taking into account the nature of the original materials used. From a practical aspect, specific knowledge of the nature of the media in particular works may offer some indication as to why some paintings are in better condition than others of a similar age.This type of information also enables the art historian to come to an educated and informed conclusion regarding the age and potential origin of an unknown work. In addition, it is feasible that a counterfeit work may be revealed, if the artist has been careless enough to use pigments or binders that are historically incorrect. The most important factors in all the techniques to be discussed are the extensive methods of sample preparation deemed necessary, since the chromatographic techniques used in the analyses are not extraordinary. The microscopic samples characteristic of work in this area are notoriously problematic to deal with and sensitivity is paramount, which consequently places great importance on the laboratory skills of the conservation scientists themselves.Paint Media Since man first learned to paint, artists have used a diverse variety of binding media for their pigments, ranging from natural gums and oils to proteinaceous materials such as egg (glair and tempera), milk (casein) and collagen glues made from animal skins and skeletons, for example. Chemically, oils and fats are the glycerol esters of aliphatic acids, typically those of the 18-carbon series.1 Oils tend to be liquid at room temperature, whereas fats are solid or semi-solid and greasy to the touch.If an oil possesses sufficient di- and triunsaturated fatty acids in its triglyceride components it will ‘dry’, i.e., polymerise, giving rise to a semi-solid.The drying oils most widely used in western European art are linseed (obtained from the seeds of the flax), walnut and poppy, although the time when they were first used for painting purposes is not known. Analytical evidence suggests that linseed oil was used in northern Europe from, at latest, the 13th Century, whilst in Italy, where oil painting was introduced in the 15th Century, walnut oil was initially preferred, although linseed oil became more common there from the 16th Century.2 In most recent years, the use of other oils, such as safflower and tung, has become more common.3–5 Collagen6 is the predominant proteinaceous material in animal skeletons (both skin and bone), representing one third of the total protein present in mammalian organisms.Collagen production in the body is preceded by the production of procollagen, a much larger biosynthetic precursor molecule, which is then degraded by specific enzymes resulting in collagen. There are a number of different types of collagen, but they all consist of molecules which contain three polypeptide a- chains in a triple helix conformation. Each a-chain has an amino acid sequence which is mainly a repeating structure, with glycine as every third residue and either proline or hydroxyproline often preceding the glycine residues.The various types of collagen can be distinguished by the slight differences in the sequence of their constituent amino acids.Animal and fish collagen glues are widely used as strong adhesives for wood, binders in the preparation of grounds, size for canvas and pigment binders in decorative paints.7 Preparation is relatively simple, involving the treatment of specific collagen-containing animal or fish tissues with hot water. When the leached solution cools, it forms a gelatinous mass; gelling occurs as a result of the partial decomposition of the tissues.If Sarah Vallance graduated in 1992 with a BSc in Applied Chemistry from the University of Northumbria at Newcastle. After a period in industry, she returned to the university and recently completed her programme of research for the award of PhD. She is at present preparing her thesis, The Development and Application of Chromatographic Methods in the Characterisation of Artists’ Media, for examination. Analyst, June 1997, Vol. 122 (75R–81R) 75Rthe extraction process is performed at a lower temperature (e.g., 80–90 °C), the gelatine solutions formed from the collagen are usually clear and light yellow.A glue which has been prepared under harsher conditions is less pure, much more turbid and viscous and is considerably darker in colour. Casein is a mixture of related phosphoproteins found in milk products.8 It is produced in mammary tissue from amino acids which have been supplied by the blood, contains all the common amino acid residues and is particularly rich in those designated as essential, e.g., leucine.Casein, egg albumin (glair) and yolk (tempera) have found uses as pigment binders, temporary varnishes and sealant over primers or grounds; casein additionally provides one of the strongest natural adhesives known, much used by cabinet makers and joiners. There are numerous recipes for the preparation of egg-based media containing ingredients as varied as linseed oil, fig tree shoots and vinegar, but the preparation of casein is uncomplicated: skimmed milk is heated to 35 °C, then the mixture is acidified to pH 4.8 with hydrochloric acid.It is allowed to stand and the casein solids are separated from the supernatant liquid, then washed with hydrochloric acid (pH 4.8). The casein thus prepared is a technically pure, slightly hygroscopic white powder.8 Gums are a group of non-crystalline, polysaccharide materials which can be found in vegetable matter and are often exuded when a plant is ‘wounded’.They are either water-soluble or water-dispersible compounds with a complex composition, usually consisting of a number of sugars (e.g., galactose and mannose) plus the acids derived from them (e.g., galacturonic acid). Gums differ from gelatines and other proteinaceous materials which form similar mucilaginous solutions in water, in that they contain virtually no nitrogen.9 Plant gums are commonly found as adhesives and binders. Gum arabic is primarily used as a painting medium, but others such as gum tragacanth (used as a medium for painting on linen) and cherry gum (which gives an enamel-like effect when mixed with egg or casein emulsions) are used less frequently.10 Documents show that gums have been used as binding media and sizing materials for centuries: gum was used as a replacement for sun-dried oil as early as the 12th Century.11 The artist’s selection of a binding medium was/is governed not only by the type of pigment used, but also by historical factors such as their location and the period of the piece: fashion and scientific progress have ensured that different materials have enjoyed periodic popularity.Analysis of Paint Media The analyses of oil-based media are well documented,12–15 but any information on the nature of other binding media has primarily been obtained via microscopic staining methods or crude solubility tests.16–21 Existing methods of analysis have provided the basis for separating the general categories of binding media (oil, gum and protein) by qualitative means, e.g., differential staining of cross-sections can distinguish between oil and protein layers.16 If a polysaccharide is heated with acid, a furfural-type compound results; these compounds react with aniline/aromatic amines to yield coloured Schiff’s bases.This reaction forms the basis of a spot test for polysaccharide media.22 Further confirmatory tests17,18 require the use of mixtures of alcohol and neutral iron chloride, or sodium hydroxide with copper sulfate (Fehling’s solution).An advantage of such micro-analytical techniques is their applicability in situ, but the results can be misleading, with false-positive and -negative results being observed. For this reason, these techniques are best used in conjunction with other analytical methods. These simple early qualitative techniques, including paper and thin-layer chromatography,19–21 are sufficient where only the general category of the binding medium is needed or, for example, where the binder may contain unique constituents, such as hydroxyproline in gelatine.However, where the amino acid or sugar composition of a proteinaceous material or gum is insufficiently distinctive, as in the cases of egg and milk proteins, for example, only quantitative chromatographic techniques will allow differentiation between similar binding media. Analysis of Proteinaceous Media Quantitative amino acid analysis by means of ion-exchange chromatography of protein hydrolysates was introduced by Moore and Stein23 in 1954.A sample of hydrolysed protein was applied to a column of sulfonic acid resin (pH 3) and an eluent with increasing pH and salt content was pumped through the column. The eluted amino acids, 18 in total, were detected by their absorbances produced with ninhydrin reagent. The required sample size was in the region of 0.3 mg of protein. In 1969, Keck and Peters24 utilised this method and reported the successful analysis of several antique and modern art specimens.Standard samples of various media were prepared both with and without pigment and dried at room and elevated temperature. After hydrolysis with 6 m HCl, under vacuum, each sample was applied to a sulfonic acid column, which was operated at 60 °C with a pH gradient from 2.87 to 5.00, a citrate concentration gradient from 0.05 to 0.267 m and a chloride concentration gradient from 0.18 to 0.25 m.The eluted amino acids were detected by absorbance measurement with ninhydrin. The percentage amino acid content was calculated for each of the standard media samples and these results were used for the purpose of identifying the unknown samples from works of art. This method facilitated the differentiation between gelatine, casein, glair, tempera and even horn found in the samples taken from works of art. There are a number of major disadvantages associated with ion-exchange chromatography in this area.Specific singlepurpose equipment is very expensive and is probably unattainable by most museums and galleries, where both space and funds are at a premium. A pH gradient, such as that used above, is difficult to control precisely, and the technique also involves a lengthy and complex method of sample preparation, where the required sample size of paint (between 1 and 4 mg to give 300 mg of proteinaceous material) is considered large, when put into the context of a fragment to be removed from a valuable work of art.The gas chromatographic (GC) determination of amino acids, obtained via the hydrolysis of proteinaceous material under acidic conditions, has been achieved using trimethylsilyl derivatives.25,26 Amino acids form two products when silylated, since the carboxyl group is easier to silylate than the amino group: under mild conditions, using hexamethyldisilazane (HMDS) as the silylating agent, the major product is the silyl ester.A stronger donor is usually required by the amino group and silylation has been achieved using N-trimethylsilyldiethylamine (TMSDEA) or N,O-bis(trimethylsilyl)acetamide (BSA), yielding the silylamine silyl ester.25 Masschelein-Kleiner26 described a GC method for the determination of trimethylsilyl derivatives of amino acids. BSA was used to derivatise pure amino acids and those obtained from the hydrolysis of proteinaceous binding materials, namely animal gelatine, casein emulsion, egg albumin and egg yolk.The volatile derivatives were injected on to a silanised Chromosorb W column (containing 10% UCC W982) with temperature programming from 90 to 220 °C at 2 °C min21. The relative amounts of the constituent amino acids were determined, thus allowing the characterisation of aged proteinaceous media. 76R Analyst, June 1997, Vol. 122Further GC methods for amino acids utilised their volatile butyl ester and N-acetylmethyl ester derivatives.27,28 Kenndler et al.27 reported a method for the hydrolysis of proteinaceous binding media from priming and paint layers of 16th and 18th Century easel and wall paintings.The efficiencies of hydrolysis of proteins by hydrochloric acid and an ion-exchange mechanism were compared. Derivatisation of the amino acid residues was achieved in two stages: first, the derivatisation of the carboxylic group, then that of the amino group.The carboxylic functions were converted into butyl esters by the addition of butanol (containing dissolved hydrogen chloride, concentration 3 mol l21) and the amino groups were then acylated with trifluoroacetic anhydride, using dichloromethane as the solvent. After evaporation of the solvent, the residue was dissolved in ethyl acetate, which contained an internal standard, and aliquots of the sample were injected directly on to the column (DB-5 capillary). A linear temperature gradient programme from 100 to 280 °C at 10 °C min21 achieved the separation of the amino acids and, by determining the relative peak areas of each, the characteristic profile of amino acids in each of the binding media was revealed.White28 hydrolysed proteinaceous material, from the gesso, ground layers and upper paint layers of a selection of Italian works, with hydrochloric acid, then de-ionised the samples on a small column of Amberlite IR-120H ion-exchange resin. Methylation of the carboxylic acid function was achieved with methanol in the presence of anhydrous hydrogen chloride gas, then the amino group was acylated with trifluoroacetic anhydride.Samples were concentrated under nitrogen, taking care not to evaporate to dryness as N-acetylmethyl esters of amino acids are extremely volatile and some loss of analytes would be incurred. Aliquots of sample were injected on to the packed column (1% XE-60 cyanosilicone gum on 100–120 mesh Diatomite CQ support, coated from acetone) and a linear temperature gradient from 80 to 210 °C at 3 °C min21 ensured separation of the amino acid components.Relative peak area were used to establish an amino acid fingerprint for each of the protein media types and characteristic amino acid ratios were used to confirm their identities. There are a number of major problems with these methods. The last technique requires a sample approximately three times the size of that for analysis of oil media, since compensation should be made for losses incurred at every stage of the sample preparation.It should also be noted that losses can occur during acid hydrolysis: the presence of sugars and carbohydrates in the sample can cause the elimination of amino acids as humins, which cause darkening of the hydrolysate and the formation of insoluble matter. Humins is the collective term for a mixture of coloured compounds (resembling naturally occurring melanins) produced during the acid hydrolysis of many proteins.A major contributory factor in the formation of humins is the presence of tryptophan and amino sugars (e.g., galactosamine, glucosamine) or carbohydrates. These compounds undergo extensive degradation during acid hydrolysis, resulting in humin production, possibly via the Maillard reaction.29 Humin formation may be prevented if hydrolysis is performed in 80% aqueous ethanol, in the presence of an ionexchange resin, for up to 10 h at 95 °C.30 It is thought that the amino acids are effectively removed from the solution and retained on the resin by their nitrogen function, hence hindering the formation of Schiff’s base condensation products between the sugars’ keto or aldehyde groups and the amino groups.The immobilised amino acids can then be eluted from the ionexchange resin by simply eluting with 10% ammonia solution. The fatty acid content of eggs has been quantified by Mills and White31 via GC analysis: the presence of tempera can be confirmed by the absence of azelate along with the presence of both palmitate and stearate non-drying oils, i.e., those which thicken at elevated temperatures but do not dry to a skin, even after prolonged exposure.Eight year old paint films, containing lead white and egg yolk medium, were analysed following saponification with potassium hydroxide and methylation of the acids with diazomethane. The conditions used (i.e., wide bore BP1 column with on-column injection and temperature programming from 110 to 310 °C at 7 °C min21) revealed the presence of the methyl esters of saturated palmitic and stearic acids, plus a variable amount of unsaturated oleic acid. It was reported that egg fats contain only small amounts of unsaturated acids and the formation of a small amount of azelate is not uncommon in a tempera medium, although amounts can range from negligible to almost one third of the palmitate present: in a pure oil film, the azelate peak would be at least equal to that of the palmitate methyl ester.This obviously leads to ambiguity in the results, further increased when non-drying oils are actually added to a blood-glue preparation.32 Skans and Michelsen33 observed that some animal glue preparations can contain up to 10% non-drying oils, hence it is not inconceivable that a sample could be wrongly identified as egg yolk. Nowik34 reported on a GC method which facilitated the simultaneous determination of both amino acids and fatty acids, which may be found together in mixed media such as tempera. Samples of proteinaceous and oil media were acid hydrolysed (6 m HCl), then neutralised with calcium carbonate.The samples were treated with an aqueous ethanol–pyridine solution prior to derivatisation, which was achieved with ethyl chloroformate (ECF), then the volatile N-(O,S)-ethoxycarbonyl ethyl ester derivatives were extracted with chloroform (containing 1% ECF).Samples were injected (splitting ratio 1 : 20) on to a CP Sil-19 CB capillary column, with a temperature programme from 100 to 300 °C at 30 °C min21. The same method was previously reported for the determination of amino acids alone; the investigations involved comparisons of methods of sample preparation, analysis and column type.35 Schilling and co-workers36–38 recently reported on the GC determination of ethyl chloroformate derivatives of amino acids from proteinaceous media; the studies also included investigations into the effects of pigments and ageing on the identification of proteins from art objects and details of a statistical approach to the interpretation of experimental results.Derivatisation was performed as described above, except that acid hydrolysis of the samples was achieved using hydrochloric acid (6 m) in the vapour phase. Samples were injected on to a capillary column coated with HP-INNOWAX, with a temperature programme from 70 to 250 °C at 27 °C min21.In recent years, the determination of amino acids by reversedphase high-performance liquid chromatography (RP-HPLC) has developed significantly, primarily owing to the speed and sensitivity of the technique compared with more traditional specialised amino acid analysers. Currently, RP-HPLC following pre-column derivatisation is one of the most widely used methods for the determination of amino acids, since it provides a good selection of both derivatising agents and detection techniques.The technique lends itself well to the field of conservation science: the occurrence of proteinaceous material in art objects coupled with the extremely small samples characteristic of the work means that RP-HPLC is an idea tool for the identification of samples taken from works of art. Erhardt et al.,39 reported a method for the determination of phenylthiocarbamyl derivatives of amino acids from proteinaceous material taken from works of art.Samples were hydrolysed under acidic conditions, dried, buffered with triethylamine, redried, then derivatised with phenyl isothiocyanate (PITC). The derivatisation reaction was stopped by the evaporation of the PITC and the samples were determined by RP-HPLC after dissolution in aqueous disodium hydrogenphosphate buffer. Samples were injected on to a C18 column and Analyst, June 1997, Vol. 122 77Ra ternary solvent system was used as the mobile phase (water– acetonitrile–acetate buffer), with a gradient programme from 0 : 6 : 94 to 16 : 28 : 56.A reasonable separation of the amino acids was shown in the paper, but there was very little discussion about the results themselves. Halpine40,41 has made extensive use of RP-HPLC for the amino acid analysis of proteinaceous matter from art objects. In a recent study,40 samples taken from ‘The Annunciation with St. Francis and St. Louis of Toulouse’, a series of 15th Century Italian tempera panels by Cosimo Tura, were submitted for amino acid analysis.42 Hydrolysis of the protein material was performed with acid vapour and derivatisation of the liberated amino acids, following buffering and drying of the hydrolysate, was achieved with PITC. The samples were analysed by RPHPLC42 with a binary solvent system of acetate buffer– acetonitrile.A C18 column was used with a gradient programme to separate the 18 amino acids deemed necessary for the identification of proteinaceous material.The use of norleucine, an unnatural amino acid, as an internal standard facilitated the quantification of the constituent amino acids and, using this method, a number of samples of animal glue media and egg/glue media were identified. A problem with this method is in the choice of derivatising agent: the PITC amino acid derivatives degrade in solution and so must be kept at 24 °C until required for analysis; only then can the samples be solubilised in the dilution buffer.Concern has been expressed with respect to the effects of mineral pigments on the reproducibility of the analyses.43 Halpine40 performed tests in which copper and calcium based pigments were added separately to proteinaceous reference samples, confirming that the presence of copper lowers the yield of all the amino acids whilst calcium reduced the recovery of aspartic and glutamic acids. This means that calcium could hinder the identification of a casein medium and copper could lower the amino acid levels to those associated with background interference, making interpretation of analytical results more difficult.Halpine postulated that the copper somehow affected the hydrolysis procedure, but an alternative explanation is that it prevents complete derivatisation of the amino acids in the hydrolysate owing to the formation of copper–amino acid complexes. Evidence for this is in the improved recovery of amino acids obtained when a small amount of EDTA solution is added to protein hydrolysates prior to derivatisation.The copper preferentially complexes with the EDTA, leaving the amino acids free to participate in the derivatisation reaction.44 It is desirable to remove any pigment from the sample, if possible, and Halpine achieved this using a simple extraction method.41 He sampled an egg tempera panel with gesso ground, which had been prepared in the laboratory, and 30 ml of HPLC grade water was added to each sample in a hydrolysis tube. After thorough mixing, the samples were left to stand for 1 h, sonicated to break up any remaining larger particles, then centrifuged for a maximum of 15 min at full speed.The watersoluble proteinaceous material was then removed, evaporated to dryness and analysed by HPLC after derivatisation with PITC, in addition to the insoluble material which remained in the precipitate. This simple technique, besides removing unwanted contaminants from the samples, proved useful for the determination of mixed media, since the degree of water solubility was found to vary depending on the protein types present.However, paintings are not the only art objects of importance: stone sculptures, frescoes and stuccoes provide other sources of samples posing interesting questions for the conservator. In particular, protein levels in these samples are typically much lower than those seen in easel paintings, for example, so the sensitivity of the method of analysis is paramount.Ronca45 used RP-HPLC, following derivatisation of the amino acid residues with PITC, to analyse both artificial samples and those taken from a number of 13th Century French stone sculptures and Italian stuccoes, a 16th Century Italian external fresco and the gesso ground of a 15th Century Italian wooden statue. Proteinaceous material was extracted from the matrices by a variety of chemical methods, the most successful being the use of 1 m sodium hydroxide solution at 80 °C for 3 h followed by colorimetric determination with Folin’s reagent.The amino acid composition of the extracted protein was achieved via RPHPLC, preceded by direct acid hydrolysis of the proteinaceous material, desalting of the hydrolysate using a sulfonic resin column and final derivatisation with PITC. The analyses revealed the presence of gelatine and egg proteins in the samples, which corroborated historical information on the nature of the materials commonly used by artists in those periods. The use of 9-fluorenylmethyl chloroformate (FMOC) as a derivatising agent in the RP-HPLC analysis of proteinaceous media was first reported by Grzywacz.46 Samples were hydrolysed under acid vapour conditions, dried and diluted with borate buffer (pH 8.5) prior to derivatisation with FMOC.Samples were injected on to a 3 mm Spherex C18 (ODS) column with a binary gradient elution programme based on the method developed by Haynes et al.;47 the eluents were (a) 50 mm sodium acetate and 7 mm triethylammonium acetate with 10% acetonitrile, adjusted to pH 6.5 with acetic acid, and (b) acetonitrile–water (90 + 10, v/v). Standard proteinaceous media and a number of museum samples were analysed and identified using this method.Vallance et al.44 also utilised an RP-HPLC method for the analysis of proteinaceous binding media taken from works of art. Samples were hydrolysed with concentrated hydrochloric acid and buffered, then, following investigations into pigment interferences, an aliquot of EDTA solution was added to any sample containing a copper-based pigment prior to derivatisation of the hydrolysate with FMOC.Separation and analysis were achieved using an ODS2 column, eluting with a mobile phase of acetate buffer (pH 4.2) and acetonitrile and operating a gradient programme from 20 to 100% acetonitrile over 70 min. A number of contributory factors recommend the selection of FMOC as the derivatising agent for the amino acid analysis.It undergoes a rapid reaction with primary and secondary amino acids and favours mild, aqueous conditions. The derivatised product yield is high48,49 and the derivatives themselves are stable at room temperature for at least 2 weeks.50 The FMOC moiety is both a good UV chromophore and highly fluorescent, allowing a choice of emission or absorption detection techniques; the FMOC amino acid derivatives can be detected at limiting levels in the low femtomole range by excitation at 260 nm.51 However, a disadvantage associated with FMOC is that it also reacts with any water present in the sample51 to form the corresponding alcohol as a hydrolysis product; the extent of any undesirable hydrolysis products is minimised by the prompt extraction of the reaction mixture with hexane.Analysis of Gum Media Chemists have employed a variety of analytical techniques for the characterisation of carbohydrate compounds, probably the most widely used in recent years being GC.Since GC is dependent on the volatility of the analytes, it is necessary for the sugars to undergo some form of derivatisation reaction prior to analysis; the various derivatisation techniques employed for this purpose have been widely reported.52–69 An obvious problem associated with the use of any of these reported methods for the analysis of gums used in works of art is the sample size; many of these techniques will not be sufficiently sensitive for the microscopic samples which are available to the conservation scientist. Despite this, progress has been made in the analysis of gum media from art objects, 78R Analyst, June 1997, Vol. 122frequently employing a combination of TLC and GC techniques. Masschelein-Kleiner and co-workers70,71 used TLC and GC to determine the trimethylsilyl derivatives of sugars resulting from the hydrolysis of samples of the surface coating of a wooden Egyptian sarcophagus, dating from the 21st Dynasty.A mixed alumina–silica stationary phase was prepared for TLC, the eluent being propanol–ethyl acetate–water–25% ammonia solution; naphthoresorcinol was used for detection on the plates after separation. Separation and analysis by GC were achieved using an E301 silicone column (length 2.0 m, external diameter 3 mm) with a temperature programme from 160 to 200 °C at 1.7 °C min21.The analysis indicated the presence of gum tragacanth and, when the paint medium itself was analysed, a mixture of honey and gum tragacanth was revealed. Flieder72 utilised TLC to separate and identify gum media from a 16th Century manuscript. The gum samples were hydrolysed under acidic conditions, deacidified with an ionexchange resin, then separated on a silica plate using butanol– ethanol–water (57 + 27 + 16) as the eluent; the sugar components were revealed with naphthoresorcinol.Although not every component was separated, the identification of the medium as gum arabic was facilitated. Birstein73 also used a combination of TLC and GC plus IR spectrometry when studying problems associated with the binding media found in Asian wall paintings. Hydrolysates of samples (1–5 mg of polysaccharide material) were acetylated prior to analysis but, although the technique was sensitive enough to facilitate identification of the media studied, the sample preparation was lengthy, requiring 12 h for hydrolysis and 5 h for derivatisation.Birstein and Tul’chinsky74 employed IR techniques for polysaccharide identification in archaeological samples; spectra of artistically important gums and some samples removed from works of art were published, but the results were not particularly informative, only allowing the distinction between polysaccharides and water-soluble proteins. Furthermore, large samples were required in order to obtain any spectra. Szyszko75 investigated the nature of binding media used in the paint layers of three Egyptian epitaphal stelae, on wooden supports, two of which dated from the second millenium bc and the third thought to be from a much later period. TLC was used in these investigations, the results of which indicated gum tragacanth as the binding medium.Studies of paintings found in the Tomb of Nefertari76,77 at Luxor revealed an interesting local phenomenon. The watersoluble paint medium used in the works was found to contain no rhamnose component, which is usually indicative of gum tragacanth.However, when samples of gum taken from locally growing trees of the Acacia genus were analysed by GC they too were found to be lacking in rhamnose, whilst the remainder of the sugar content matched that seen in the samples taken from the paintings. It was therefore concluded that the paint medium was in fact gum arabic, despite all other commercial sources of the gum showing a rhamnose component.These findings could in turn mean that the samples from the Egyptian epitaphal stelae could actually contain gum arabic from the same local source, rather than gum tragacanth as originally concluded.78 Twilley79 published a report on the analysis and artistic applications of plant gums. Analysis of samples was achieved via a number of techniques, including GC of trimethylsilyl sugar derivatives and TLC. Erhardt et al.39 presented their findings of investigations into the GC analysis of gums to the American Institute for Conservation of Historic and Artistic Works (AIC). Samples were hydrolysed with trifluoroacetic acid (2.3 ml in 7.7 ml of water), then dried in a vacuum desiccator.The sugars were then converted into their oxime form with a pyridine solution of hydroxylamine hydrochloride prior to silylation with trimethylsilylimidazole. Samples were separated on a non-polar DBI (bonded polydimethylsiloxane) capillary column with a temperature programme from 100 to 275 °C at 10 °C min21.The resolution was improved using a more polar DB17 column, although higher temperatures were required, i.e., from 150 to 325 °C at 10 °C min21. In 1996, Bleton et al.80 reported a GC method for the analysis of ink samples from ancient manuscripts. Following methanolysis and silylation with trimethylsilylimidazole, samples were injected on to an SE-52 capillary column with a stepped temperature programme from 40 to 130 °C at 9 °C min21, then from 130 to 290 °C at 2 °C min21, the final temperature being held for a further 30 min.Reference samples of old ink and samples taken from a variety of manuscripts were analysed. Other chromatographic techniques can be utilised for the characterisation of sugar based compounds. Pyrolysis–mass spectrometry (Py–MS) was the technique employed by Wright81 in studies of Egyptian mummy cases. Samples of organic materials used in the construction of the cartonnages were examined by Py–MS: pyrograms were compared with those of a series of standard materials, facilitating the identification of around 50% of the samples.Polysaccharide gums were detected in samples from objects between 2000 and 4000 years old. Derrick and Stulik82 used pyrolysis–gas chromatography (Py–GC) to investigate the separation and identification of natural gums in works of art. Powdered gum samples were pyrolysed using a coil type probe: standard gum samples were pyrolysed at 700 °C, but this temperature was lowered for the analysis of more complex samples.Gums arabic, tragacanth, guar, ghatti and karaya all gave distinguishable and reproducible pyrograms, allowing their identification. It was observed that the pyrograms of gum–pigment mixtures differed from those of the standard gums, their peak patterns and intensities being altered. This effect was minimised by performing the pyrolysis of the samples at a lower temperature of 400 °C.Computational methods of pattern recognition were employed to assist with sample identification. In 1995, Williams and Langdon83 used gel permeation chromatography (GPC) to characterise gum arabic, identifying the three principal molecular mass components of the gum. Ion-exchange chromatography,84–86 be it anion or cation, was popular for a time, but proved to be less sensitive than HPLC methods. In 1976, Rabel et al.87 reported on a normal-phase partition HPLC method, using refractive index (RI) detection, for sugar determination which was sensitive down to around 80 mg; however, much of the work on the use of HPLC techniques for the analysis of sugars was recorded between 1980 and 1987.88–95 A variety of derivatisation techniques (e.g., post-column cuprammonium, dansylhydrazine, pre-column dabsylhydrazine) and detection methods (e.g., UV absorption, fluorescence, RI) have been used and sensitivities as low as 5 pmol have been achieved; it must be noted, however, that none of these latter studies were in the particular area of conservation science.Analytical methods used for the determination of polysaccharides in art objects have been reviewed previously,96 one in particular focusing on chromatographic techniques,97 although obviously neither contain details of the most recent work from 1992 to the present. Conclusions There are a number of important questions that the conservator/ conservation scientist must ask before deciding on an appropriate analytical technique: 1.What do I need to know? Is the exact identity of the binding medium necessary? 2. How much sample is available? 3. What pigments, if any, are likely to be Analyst, June 1997, Vol. 122 79Rpresent in the sample? 4. Have any conservation/restoration treatments been performed on the work previously? If so, what was the nature of any treatment, e.g., consolidation, retouching? The conservation scientist needs to select a technique which will give the maximum amount of information for the minimum amount of sample and sample preparation.For proteinaceous media, this would appear to be amino acid determination by RPHPLC using FMOC as the derivatising agent. Gas chromatography of silylated sugar derivatives is, at present, probably the best method for the identification of natural gums: however, as this is probably the least investigated area so far, major developments in methodology which would greatly improve sensitivity can be expected; the main problem with samples of gum media is the minute amount of actual medium present.Simple qualitative techniques such as TLC, microscopy and staining tests may be sufficient to indicate the basic media type used in a work, but as more and more works of art require conservation/restoration treatments it is crucial that the conservator has as much information as possible on the nature of any materials used by the artist, in order to avoid the loss or spoiling of any valuable and irreplaceable pieces.It is clear that the previous investigations into quantitative analyses are of immense value and as chromatographic techniques are continually developed and improved, increasing both sensitivity and reproducibility, their use in the area of art conservation becomes even more ideal. One possibility is the development analytical methods using capillary zone electrophoresis (CZE), a relatively new technique currently used for the determination of proteins, etc.; it could prove to be a useful method of analysis in this area, owing to its high sensitivity, but a thorough investigation into its suitability would be required.Microbore techniques should also find their use in conservation science, since they obviously suit the sometimes nanomole amounts of samples which are provided for analysis. The main concern of the conservation scientist is the availability of reliable and accurate analytical techniques suitable for use with the minute samples typically seen in this field of work.No doubt further research will result in the simplification of methods of sample preparation, possibly negating the need for clean-up procedures (e.g., the removal of pigments from samples) prior to analysis. However, any improvements which mean that the required sample size is reduced and the loss of valuable sample material is minimised or, ideally, eliminated will be wholeheartedly welcomed by the conservator and conservation scientist alike. References 1 Gettens, R.J., and Stout, G. L., Painting Materials, a Short Encyclopaedia, Dover, New York, 1966, p. 36. 2 Mills, J. S., and White, R., The Organic Chemistry of Museum Objects, Butterworth Heinemann, London, 2nd edn., 1994, pp. 36 and 171–172. 3 Eastlake, C. L., Materials for a History of Oil Painting, Dover, New York, 1960, vol. 2, p. 88. 4 Matsui, E., Sci.Pap. Jpn. Antiques Art Crafts, 1981, 26, 15. 5 Carillo y Gariel, A., Technica de la Pintura de Nueva Espan�a, Imprenta Universitaria, Mexico, 1946, pp. 42–48. 6 The Merck Index, ed. Windholz, M., Merck, Rahway, NJ, 10th edn., 1983, p. 2452. 7 Gettens, R. J., and Stout, G. L., Painting Materials, a Short Encyclopaedia, Dover, New York, 1966, pp. 25–27. 8 Gettens, R. J., and Stout, G. L., Painting Materials, a Short Encyclopaedia, Dover, New York, 1966, pp. 7–8. 9 Gettens, R. J., and Stout, G.L., Painting Materials, a Short Encyclopaedia, Dover, New York, 1966, pp. 28–29. 10 Doerner, M., The Materials of the Artist and Their Use in Painting, Harcourt, Brace, New York, 1934, pp. 223–224. 11 Laurie, A. P., The Materials of the Painters’ Craft, J. B. Lippincott, Philadelphia, 1911, p. 164. 12 Mills, J. S., Stud. Conserv., 1966, 11, 92. 13 Mills, J. S., and White, R., Nat. Gallery Tech. Bull., 1980, 4, 65. 14 Mills, J. S., and White, R., in Application of Science in the Examination of Works of art, ed.England, P. A., and van Zelst, L., Museum of Fine Arts, Boston, 1985, pp. 29–34. 15 Mills, J. S., and White, R., in Conservation and Restoration of Pictorial Art, ed. Bromelle, N., and Smith, P., Butterworths, London, 1976, pp. 72–76. 16 Gay, M. C., in Conservation and Restoration of Pictorial Art, ed. Bromelle, N., anSmith, P., Butterworths, London, 1976, pp. 78– 83. 17 Butler, C. L., and Cretcher, L. H., J.Am. Chem. Soc., 1929, 51, 1519. 18 Gettens, R. J., and Stout, G. L., Painting Materials, a Short Encyclopaedia, Dover, New York, 1966, p. 29. 19 Hey, M., Stud. Conserv., 1958, 3, 183. 20 Denniger, E., S. Afr. Adv. Sci. Spec. Publ., 1971, 2, 80. 21 Masschelein-Kleiner, L., Stud. Conserv., 1974, 19, 207. 22 Feigl, F., Spot Tests in Organic Analysis, Elsevier, Amsterdam, 5th edn., 1956, p. 372. 23 Moore, S., and Stein, W. H., J. Biol. Chem., 1954, 211, 893. 24 Keck, S., and Peters, T., Stud.Conserv., 1969, 14, 75. 25 Pierce, A. E., Silylation of Organic Compounds, Pierce, Rockford, IL, 1977, pp. 218–243. 26 Masschelein-Kleiner, L., in Conservation and Restoration of Pictorial Art, ed. Bromelle, N., and Smith, P., Butterworths, London, 1976, pp. 84–87. 27 Kenndler, E., Schmidt-Beiwl, K., Mairinger, F., and P�ohm, M., Fresenius’ J. Anal. Chem., 1992, 342, 135. 28 White, R., Nat. Gallery Tech. Bull., 1984, 8, 5. 29 Krampitz, G., Tierphysiol. Tierern�ahr.Futtermittelk, 1960, 15, 227. 30 P�ohm, M., Naturwissenschaften, 1961, 48, 555. 31 Mills, J. S., and White, R., The Organic Chemistry of Museum Objects, Butterworth Heinemann, London, 2nd edn., 1994, p. 171. 32 Brown, C., Bomford, D., Plesters, J., and Mills, J. S., Nat. Gallery Tech. Bull., 1987, 11, 85. 33 Skans, S., and Michelsen, P., Maltech. Restauro, 1986, 92, (2), 63. 34 Nowik, W., Stud. Conserv., 1995, 40, 120. 35 Hu�sek, P., J. Chromatogr., 1991, 552, 289. 36 Schilling, M.R., Khanjian, H. P., and Souza, L. A. C., J. Am. Inst. Conserv., 1996, 35, 45. 37 Schilling, M. R., Khanjian, H. P., and Souza, L. A. C., J. Am. Inst. Conserv., 1996, 35, 123. 38 Schilling, M. R., and Khanjian, H. P., in ICOM Committee for Conservation, 11th Triennial Meeting, Edinburgh, September 1996: Preprints, ed. Bridgland, J., James and James, London, 1996, vol. I, p. 211. 39 Erhardt, D., Hopwood, W., Baker, M., and von Endt, D., in Preprints of Papers Presented at the 6th Annual Meeting, American Institute for Conservation of Historic and Artistic Works, Washington, DC, 1988, pp. 67–84. 40 Halpine, S. M., Stud. Conserv., 1992, 37, 22. 41 Halpine, S. M., Conservation Research 1995: Studies in the History of Art, vol. 51, Monograph Series II, National Gallery of Art, Washington, DC, 1995. 42 Cohen, S. A., Heys, M., and Tarvin, T. L., The Picotag Method—a Manual of Advanced Techniques for Amino Acid Analysis, Millipore, Waters Chromatography Division, Milford, MA, 1989, pp. 2–3 and 11–12. 43 Chemistry and Biochemistry of the Amino Acids, ed. Barrett, G. C., London, 1985, pp. 376–396. 44 Vallance, S. L., Singer, B. W., Hitchen, S. M., and Townsend, J., LC– GC Int., 1997, 10(1), 48. 45 Ronca, F., Stud. Conserv., 1994, 39, 107. 46 Grzywacz, C. M., J. Chromatogr. A, 1994, 676, 177. 47 Haynes, P. A., Sheumack, D., Kibby, J., and Redmond, J. W., J. Chromatogr., 1991, 540, 177. 48 Matzner, M., Kurkjy, R. P., and Cotter, R. T., Chem. Rev., 1964, 64, 645. 49 Hall, M. K., J. Am. Chem. Soc., 1957, 79, 5439. 50 Seiler, N., in Handbook of Derivatives for Chromatography, ed. Blau, K., and Halket, J. M., Wiley, Chichester, 1993, p. 187. 51 Einarsson, S., Josefsson, B., and Lagerkvist, S., J. Chromatogr., 1983, 282, 609. 80R Analyst, June 1997, Vol. 12252 McInnes, A. G., Ball, D. H., Cooper, F. P., and Bishop, C. T., J. Chromatogr., 1958, 1, 556. 53 Bishop, C. T., Adv. Carbohydr. Chem., 1964, 19, 95. 54 Bishop, C. T., and Cooper, F. P., Can. J. Chem., 1960, 38, 388. 55 Sawardeker, J. S., Sloneker, J. H., and Jeanes, A., Anal. Chem., 1967, 39, 121. 56 Bj�orndal, H., Lindberg, B., and Svensson, S., Acta Chem. Scand., 1967, 21, 1801. 57 Sweeley, C. C., Bently, R., Makita, M., and Wells, W. W., J. Am. Chem. Soc., 1963, 85, 2497. 58 Beadle, J. B., J. Agric. Food Chem., 1969, 17, 904. 59 Honda, S., Kakehi, K., and Okada, K., J. Chromatogr., 1979, 176, 367. 60 Sullivan, J. E., and Schewe, L. R., J. Chromatogr. Sci., 1977, 15, 196. 61 Decker, P., and Schweer, H., J. Chromatogr., 1982, 236, 369. 62 Dmitriev, B. A., Backinowsky, L. V., Chizhov, O. S., Zolotarev, B. M., and Kochetkov, N. K., Carbohydr. Res., 1971, 19, 432. 63 Churns, S. C., J. Chromatogr., 1990, 500, 555. 64 Varma, R., Varma, R. S., and Wardi, A. H., J. Chromatogr., 1973, 77, 222. 65 Ha, Y. W., and Thomas, R. L., J. Food Sci., 1988, 53 (2), 574. 66 Aspinall, G. O., and Fairweather, R. M., Carbohydr. Res., 1965, 1, 83. 67 Aspinall, G. O., and McKenna, J. P., Carbohydr. Res., 1968, 7, 244. 68 Reinhold, V. N., Wirtz-Pietz, F., and Biemann, K., Carbohydr. Res., 1974, 37, 203. 69 Wiecko, J., and Sherman, W. R., J. Am. Chem. Soc., 1976, 98, 7631. 70 Masschelein-Kleiner, L., and Tricot-Marckx, E., Bull. Inst. R. Patrimoine Artistique, 1965, 8, 180. 71 Masschelein-Kleiner, L., Heylan, J., and Tricot-Marckx, F., Stud. Conserv., 1968, 13, 105. 72 Flieder, F., Stud. Conserv., 1968, 13, 49. 73 Birstein, V. J., Stud. Conserv., 1975, 20, 8. 74 Birstein, V. J., and Tul’chinsky, V. M., Khim. Prirod. Soedin., 1976, 1, 15. 75 Szyszko, W., Ochr. Zabytkow, 1972, 25, 170. 76 Mora, P., Mora, L., and Porta, E., in ICOM Committee for Conservation, 9th Triennial Meeting, Dresden, 26–31 August 1990: Preprints, ed. Grimstad, K., 1990, p. 518. 77 Palet, A., and Porta, E., in Congresa de Conservaci�on de Bienes Culturales: Valencia, 20–23 Setiembre de 1990, ed. Roig Picazo, P., 1990, p. 452. 78 Mills, J. S., and White, R., The Organic Chemistry of Museum Objects, Butterworth Heinemann, London, 2nd edn., 1994, p. 79. 79 Twilley, J. W., Adv. Chem. Ser., 1984, No. 205, 357. 80 Bleton, J., Coupry, C., and Sansoulet, J., Stud. Conserv., 1996, 41, 95. 81 Wright, M. M., J. Anal. Appl. Pyrol., 1987, 11, 195. 82 Derrick, M. R., and Stulik, D. C., in ICOM Committee for Conservation, 9th Triennial Meeting, Dresden, 26–31 August 1990: Preprints, ed. Grimstad, K., 1990, p. 9. 83 Williams, P. A., and Langdon, M. J., Chromatogr. Anal., 1995, Oct/ Nov, 5. 84 Mrochek, J. E., Dinsmore, S. R., and Waalkes, T. P., Clin. Chem., 1975, 21, 1314. 85 Verhaar, L. A. Th., and Kuster, B. F. M., J. Chromatogr., 1981, 210, 279. 86 Kesler, R. B., Anal. Chem., 1967, 39, 1416. 87 Rabel, F. M., Caputo, A. G., and Butts, E. T., J. Chromatogr., 1976, 126, 731. 88 Binder, H., J. Chromatogr., 1980, 189, 414. 89 Alpenfels, W. F., Anal. Biochem., 1981, 114, 153. 90 Grimble, G. K., Barker, H. M., and Taylor, R. H., Anal. Biochem., 1983, 128, 422. 91 Mopper, K., and Johnson, L., J. Chromatogr., 1983, 256, 27. 92 Vratny, P., Frei, R. W., Brinkman, U. A. Th., and Nielen, M. W. F., J. Chromatogr., 1984, 295, 355. 93 Rosenfelder, G., M�orgelin, M., Chang, J. W., Sch�onenberger, C. A., Braun, D. G., and Towbin, R. H., Anal. Biochem., 1985, 147, 156. 94 Hull, S. R., and Turco, S. J., Anal. Biochem., 1985, 146, 143. 95 Lin, J.-K., and Wu, S.-S., Anal. Chem., 1987, 59, 1320. 96 Mills, J. S., and White, R., The Organic Chemistry of Museum Objects, Butterworth Heinemann, London, 2nd edn., 1994, pp. 78– 80. 97 Matouösov�a, M., and Bucifalov�a, J., Sb. Restaur�at. Praci, 1989, 4, 60. Paper 6/06219I Received September 9, 1996 Accepted March 20, 1997 Analyst, J

ISSN:0003-2654    

DOI:10.1039/a606219i

出版商:RSC    

年代:1997

数据来源: RSC

摘要:

Determination of Trace Metal Distributions in the Iron Oxide Phases of Red Bed Sandstones by Chemometric Analysis of Whole Rock and Selective Leachate Data Mark R. Cave* and Karen Harmon Analytical Geochemistry Group, British Geological Survey, Keyworth, Nottingham, UK NG12 5GG Hematite is known to be a sink for trace metals and study of the trace metal distributions within hematite-rich formations can provide evidence of past groundwater activity. The aim of this study is to develop a method specifically for determining the trace metal content of naturally occurring hematite in a sandstone formation.The elemental compositions of twelve samples of permo-triassic Red Bed sandstones from the St. Bees formation, Sellafield, Cumbria, were determined. An acid digestion procedure was used to produce solutions suitable for ICP-AES analysis for the major and trace elements (Al, Fe, Mg, Ca, Na, K, Ti, Mn, Ba, Sr). The same solution was used for ICP-MS analysis for lower abundance trace elements (Cr, Ni, Pb, Sn, Th, U, V, Zr).Fusion with LiBO2, followed by an ion exchange preconcentration–separation procedure, provided solutions for rare earth elements, La, and Y determination by ICP-AES. Solutions for ICP-AES analysis of Si in the whole rock were prepared by a modified fusion procedure using 4 + 1 mixture of LiBO2 and Li2B4O7. A leaching procedure was developed to selectively extract the hematite phase of the rock samples. This procedure consisted of a preliminary extraction with 1 m acetic acid for 20 h at 20 °C to remove carbonates, followed by an extraction with 0.1 m oxalic acid–0.175 m ammonium oxalate (Tamm’s reagent) for 20 h at 70 °C to dissolve the iron oxide.The Tamm’s reagent leachate was analysed by ICP-AES and ICP-MS for the same suite of elements as the whole rock. Principal component analysis was applied to the whole rock and leachate data matrices. Qualitative assessment of the resulting principal components revealed elemental groupings which can be assigned to minerals known to exist in this rock type and showed the leaching procedure was not fully selective for the iron oxide phase.Chemometric mixture decomposition applied to the leachate data sets identified the extraction of four different mineral sources tentatively assigned to hematite, dolomite, chlorite and iron oxy-hydroxides. Quantitative estimates of the composition and proportion each of these components in each rock sample were calculated.Keywords: Inductively coupled plasma atomic emission spectrometry; inductively coupled plasma mass spectrometry; Red Bed sandstones; rare earth elements; selective extraction; principal component analysis; mixture modelling In order to understand and accurately model the migration of trace elements in groundwater systems it is essential to know how these elements are partitioned between the rock and the water. Hematite is known to be a sink for trace metals and study of the trace metal distributions within hematite-rich formations can provide evidence of past groundwater activity. The aim of this study is to develop a method specifically for determining the trace metal content of naturally occurring hematite in a sandstone formation. The distribution of trace elements in rocks can be identified by techniques with the spatial resolution necessary to analyse individual mineral grains on a mm scale.Such techniques are widely used in geochemical and mineralogical studies and include: proton induced X-ray emission (PIXE), scanning electron microscopy (SEM), energy dispersive X-ray emission (EDAX) and laser ablation microprobe ICP mass spectrometry (LAMP–ICP-MS).The data obtained by these methods give information on a small scale which does not necessarily reflect the distribution of trace elements in the bulk rock. In contrast, methods which homogenise the rock matrix before analysis, such as XRF analysis of pressed powder pellets and fusion discs or ICP emission/mass spectrometry of acid and fusion digests, give an integrated picture of the total trace element composition but very little information about the minerals in the rock and their associated trace metals.To obtain information on the trace metal associations in the bulk rock, a number of workers1,2 have developed extraction schemes in which mineral phases are selectively dissolved with carefully chosen reagents.By analysis of the extraction media the concentration of trace elements associated with the target mineral phase can be determined. However, two major weaknesses of the extraction technique have been demonstrated. 3–7 (i) The so called ‘selective extraction reagents’ are not specific for one mineral phase; therefore the associated analysis is not a true representation of the trace elements from a single phase. (ii) The chemical leaching process does not quantitatively extract the true total trace element concentration from the target phase as many trace elements are re-adsorbed onto the rock matrix during the leaching process. This study used a number of samples of permo-triassic Red Bed sandstones from the St.Bees formation, Sellafield, Cumbria, from one borehole to show that chemometric analysis of multi-element data from whole rock analysis and selective extraction can provide information on the mineral components of the bulk rock and help to overcome the shortcomings of selective extraction identified above.The samples, in depth order are labelled SC1–SC12. Experimental The sample preparation, digestion and analysis of the whole rock samples and the leachates has been carried out using standard techniques which are fully described in Appendix 1. Selective Extraction Procedures Following an examination of published results from previous work and reviews, e.g., Tessier et al.2 and Zielinski et al.,8 four reagents were used for trial experiments.For removal of carbonates, prior to iron oxide extraction, 1 m ammonium acetate (adjusted to pH 5 with glacial acetic acid) and 1 m acetic Analyst, June 1997, Vol. 122 (501–512) 501acid were chosen. For selective extraction of iron oxides Tamm’s reagent9 (0.175 m ammonium oxalate–0.1 m oxalic acid) and 0.1 m HCl were selected. Sodium dithionate–citrate mixtures have been used in some studies8 but were not considered in this work for the reasons outlined by Tessier et al.:2 (i) extraction with dithionate citrate can lead to the precipitation of trace metals because of the formation of sulfides as a result of disproportionation of dithionate during extraction; (ii) sodium dithionate cannot be purchased in a pure form and contains high levels of trace metals at or above the levels to be leached from the rock.Purification of the dithionate by ion exchange is a difficult and lengthy process; and (iii) sodium dithionate solutions are not compatible with atomic spectrometry sample introduction systems which quickly become clogged (and cause instrument malfunction).The oxalic acid–ammonium oxalate extract used by a number of workers (e.g., Zielinski et al.8), overcomes some of these disadvantages. It is claimed to be selective for iron oxides, can be obtained in high purity and the organic part of the matrix can be destroyed after the extraction to minimise sample introduction problems.The sample to extractant ratio was 5 g to 30 ml. The extractions were performed in 50 ml Nalgene Teflon FEP centrifuge tubes. Extractions at 70 °C were carried out in a water bath with occasional shaking; extractions at room temperature were continually shaken on a mechanical shaker. Samples were centrifuged between extractions and the supernatant liquor poured off. Between the acetic acid–ammonium acetate leach and the final leach, the sample was washed with 5 ml of deionised water.The Tamm’s reagent leachates were analysed for major elements, Cr and V by ICP-AES; 10 ml of the remaining solution were then removed for analysis by ICP-MS. Because of the low tolerance of the latter technique to high dissolved solids, it was necessary to digest the oxalate matrix with acid H2O2 prior to analysis. To the 10 ml of sample in a 15 ml polycarbonate test tube, 1 ml of H2O2 and 0.1 ml of concentrated HNO3 were added. The tube was capped loosely and left overnight in a water bath at 70 °C.This procedure was repeated three times, reducing the organic carbon content to less than one tenth of its original value. The final solution was diluted 1 + 1 with deionised water prior to analysis by ICPMS. The duplicate Tamm’s reagent leachate was acidified to 10% v/v with respect to concentrated HCl. The rare earth elements (REE) in this solution were determined by ICP-AES following ion exchange separation and preconcentration.Results and Discussion Development of Selective Extraction Procedure The objectives of this study were to develop a leaching procedure to selectively extract the iron oxide phases from the sandstones and to measure their trace metal content. The main criteria which determined method development were: (1) the selective leaching scheme should be specific for the iron oxide phase; (2) the trace elements to be determined in the rock leachate (U, Th, REEs, Ni, Sn and Pb) would be in the mg kg21 to mg kg21 range.This required a choice of extraction reagents which would minimise both sample contamination and potential interference in the chosen methods of analysis, in this case ICPAES and ICP-MS; and (3) the extraction scheme necessary to produce the required data for geochemical interpretation should also include an extraction to remove carbonate phases prior to an extraction to selectively remove the target iron oxide phases. This allowed the trace metal content of the iron oxides to be determined without contamination from trace metals in carbonates.A preliminary trial was designed with two main objectives: (i) to determine the best combination of acetic acid and ammonium acetate to selectively remove the carbonate phase without attacking the iron oxide or silicate phases; and (ii) to select the conditions for reaction and to compare the selectivity and efficiency of either HCl or oxalic acid/ammonium oxalate (Tamm’s reagent) for removing iron oxides.Initial leaching experiments were performed only on one core sample (SC1). The carbonate extraction experiments (Table 1) showed that the ammonium acetate–acetic acid (pH 5) and the 1 m acetic acid media were equally effective in removing calcium carbonate. The analyses of the two extractants are similar for all elements determined. The comparability of the Si and Fe values between these two reagents shows that the more aggressive acetic acid had not attacked the silica or iron oxide matrix more aggressively than the ammonium acetate solution.The 1 m acetic acid was therefore chosen for carbonate extraction because of its faster reaction rate (Table 1). Dissolution of iron oxide using Tamm’s reagent at room temperature was very slow, as shown by the marginal increase in Fe content of the leach solution over 46 h (Table 1). The red colouration of the sandstone appeared unchanged over the same period.The 1 m HCl at 70 °C showed much higher levels of Fe, which steadily increased over the period it was applied, although the rock still appeared red. High silicon values suggested that the silicate matrix was being attacked. On heating the Tamm’s reagent to 70 °C the efficiency of the iron oxide extraction was greatly enhanced, as indicated by the increase in Fe concentration combined with the marked decolouration of the sandstone material to pale grey with a thin red layer of finer clay material on top.The heated Tamm’s reagent had a higher efficiency of extraction than the heated HCl and a reduced tendency to attack the silicate matrix, as shown by the lower Si values. The Tamm’s reagent at 70 °C was therefore preferable to HCl as the iron oxide extraction reagent. The relative selectivity and effectiveness of Tamm’s reagent and 0.3 m oxalic acid removing the iron oxide phase were Table 1 Extraction media, conditions of extraction and mass (mg kg21) of major elements extracted during the preliminary extraction trial Conditions Extractant Ca Mg Na K Si Mn Fe 24 h @ 20 °C NH4OAc 13 289 84.94 19.57 50.98 18.92 36.60 7.52 (pH 5) 2 h @ 20 °C HOAc (1m) 13 115 88.96 17.53 36.50 23.51 37.35 10.88 27 h @ 20 °C Tamm’s 1 2.88 13.37 2.43 21.94 65.37 8.40 31.71 46 h @ 20 °C Tamm’s 2 2.33 16.35 3.18 32.97 101 8.03 35.55 46 h @ 20 °C + 24 h @ 70 °C Tamm’s 3 2.99 175 8.57 91.67 508 14.62 2445 27 h @ 70 °C HCl (0.1m) 1.448 362 16.05 219 1116 17.33 1016 46 h @ 70 °C HCl (0.1m) 1.391 452 17.73 399 1285 19.58 1619 70 h @ 70 °C HCl (0.1m) 1.298 561 22.65 718 1043 19.93 2171 502 Analyst, June 1997, Vol. 122compared in a second trial. This used 2 3 5 g splits of the second shallowest sample. Each split was extracted with 30 ml of 1 m acetic acid at 20 °C for 23 h followed by 2 35 ml washes of distilled water to remove carbonate, prior to the final iron oxide extraction. This used 30 ml of Tamm’s reagent for one split and 30 ml of 0.3 m oxalic acid for the other.The rates of dissolution of Fe, Mg, Si and Al were monitored by taking 1 ml aliquots of each solution over the 20 h extraction. These were diluted to 10 ml and analysed by ICP AES (Fig. 1). The 0.3 m oxalic acid was much more aggressive than Tamm’s reagent. The oxalic acid decolourised both the coarser sand material and the finer clay layer. The amount of iron extracted after 20 h was also greater and the rate of extraction faster.The oxalic acid was, however, probably too aggressive, as the increased rate of dissolution and final amounts of Si and Al leached indicate silicate matrix attack. Steady state for both the oxalate media was reached in about 16–20 hours. On the basis of the data from these preliminary trials the leaching procedure summarised in Table 2 was formulated. For each sample, 2 35 g splits of < 400 mm sandstone powder were weighed into 2 Nalgene 50 ml Teflon FEP centrifuge tubes. The second Tamm’s reagent leach was incorporated because, after the first leach, some of the samples were still red indicating that not all the iron oxide phase was being removed.This leachate was subsequently preserved and analysed as a separate leach solution. The first Tamm’s reagent leachate was used to determine the major elements, Cr and V by ICP-AES and Cr, Ni, Sn, Pb, Th and U by ICP-MS. The second split of the same sample was used for the ion exchange separation and concentration of the REEs which were subsequently analysed by ICP-AES.These methods are described in detail in Appendix 1. The results of the chemical analysis of the whole rock and the two Tamm’s reagent leaches are summarised in Tables 3, 4 and 5, respectively. Fig. 1 Extraction of major elements from sample SC1; (a) 0.3 m oxalic acid; (b) Tamm’s reagent. A, Fe, B, Mg; C, Si; D, Al. Table 3 Summary of chemical analysis data for the whole rock samples Element Units SC1 SC2 SC3 SC4 SC5 SC6 SC7 SC8 SC9 SC10 SC11 SC12 SiO2 % oxide 76.9 78.0 69.4 64.0 70.3 75.3 64.8 72.0 79.3 70.6 76.1 72.8 Al2O3 % oxide 7.05 5.24 8.98 7.76 12.21 5.19 9.90 8.56 8.06 8.63 6.46 5.47 Fe2O3 % oxide 0.95 0.93 2.61 4.29 1.39 0.82 3.02 2.15 1.78 3.02 2.28 1.99 MgO % oxide 0.309 0.213 0.525 0.502 0.796 0.768 0.879 1.95 0.518 2.27 1.59 2.27 CaO % oxide 1.60 3.91 4.64 6.97 1.36 5.48 7.20 2.45 0.21 2.34 2.32 3.60 Na2O % oxide 1.15 0.96 1.83 1.53 2.20 1.23 1.79 1.99 1.60 1.58 1.08 0.92 K2O % oxide 3.29 2.63 3.68 3.31 4.60 2.58 3.67 3.39 3.60 3.48 3.41 2.94 TiO2 % oxide 0.201 0.134 0.567 0.884 0.541 0.114 0.447 0.320 0.224 0.356 0.279 0.239 MnO % oxide 0.009 0.021 0.036 0.052 0.013 0.073 0.052 0.057 0.007 0.049 0.048 0.084 Ba mg kg21 469 413 410 323 516 268 373 403 442 373 324 279 Cr mg kg21 12.1 10.8 49.5 68.2 33.6 6.4 23.1 16.6 12.9 20.9 19.8 14.8 Ni mg kg21 5.49 3.50 8.82 8.48 14.41 2.75 18.46 7.63 12.01 16.18 6.70 5.37 Pb mg kg21 11.62 11.81 9.62 13.22 9.84 7.20 6.95 10.14 11.27 10.89 11.19 8.55 Sn mg kg21 1.45 1.20 2.41 2.47 1.99 0.89 1.49 1.26 1.34 2.20 2.50 1.79 Sr mg kg21 84.4 78.2 105.5 100.4 120.0 61.7 106.7 95.5 88.3 85.0 65.6 58.1 Th mg kg21 3.97 3.30 8.62 11.05 7.45 2.73 5.66 4.92 4.32 8.29 6.10 4.57 U mg kg21 1.05 1.00 2.41 3.35 2.29 0.64 1.59 1.54 1.19 2.10 1.40 1.19 V mg kg21 < 9.0 15.28 48.70 61.64 48.72 < 9.0 44.53 27.04 16.94 37.63 23.36 16.35 Zr mg kg21 77.3 49.4 195 303 134 29.0 100 93.9 62.0 91.1 79.3 60.9 Ce mg kg21 21.7 16.9 51.3 62.4 48.7 18.8 53.0 31.3 27.7 47.4 28.7 27.1 Dy mg kg21 4.06 3.87 6.15 10.36 6.40 4.06 6.77 6.54 5.51 6.41 5.44 4.27 Er mg kg21 0.70 0.81 1.78 3.57 1.39 0.63 1.43 0.87 0.69 1.09 0.73 0.98 Eu mg kg21 0.67 0.63 1.17 1.74 0.83 0.69 1.19 0.86 0.59 0.88 0.65 0.74 Gd mg kg21 < 0.94 < 0.94 1.03 3.50 < 0.97 < 0.93 < 0.98 < 0.95 < 0.93 < 0.94 < 0.97 < 1.0 Ho mg kg21 < 0.67 < 0.67 0.85 1.73 < 0.70 < 0.67 < 0.68 0.69 < 0.66 < 0.67 < 0.70 < 0.71 La mg kg21 11.68 9.54 26.09 31.47 26.84 11.81 28.57 15.81 13.92 23.73 14.01 15.34 Lu mg kg21 0.15 0.15 0.36 0.71 0.27 0.17 0.33 0.27 0.18 0.29 0.18 0.22 Nd mg kg21 10.36 8.77 23.16 33.37 19.78 10.85 24.02 17.71 12.87 21.13 13.56 13.71 Pr mg kg21 2.50 2.56 5.42 6.89 5.05 2.85 6.18 3.73 3.45 4.57 2.46 2.76 Sm mg kg21 2.34 2.41 5.32 8.04 4.26 2.79 4.95 3.98 2.79 4.79 2.63 3.27 Tb mg kg21 < 1.6 < 1.6 < 1.6 < 1.6 < 1.7 < 1.6 < 1.7 < 1.6 < 1.6 < 1.61 < 1.7 < 1.7 Tm mg kg21 < 0.67 < 0.67 < 0.69 1.00 0.72 0.75 < 0.70 < 0.68 < 0.66 < 0.67 < 0.7 < 0.71 Y mg kg21 10.6 9.7 23.9 45.3 19.2 9.9 21.4 14.2 9.9 16.4 12.1 14.6 Yb mg kg21 1.13 1.01 2.60 4.42 2.10 1.06 2.26 1.62 1.03 1.75 1.23 1.38 Table 2 Summary of leaching media and conditions used for the finalised leaching procedure Leach Leaching Time/ Temperature/ Leachate no.solution h °C preservation 1 1 mAcetic acid 20 20 None 2 Tamm’s reagent 20 70 1% HCl 3 Tamm’s reagent 20 70 1% HCl Analyst, June 1997, Vol. 122 503Table 4 Summary of chemical analysis data for the first Tamm’s reagent leach Element Units SC1 SC2 SC3 SC4 SC5 SC6 SC7 SC8 SC9 SC10 SC11 SC12 SiO2 % oxide 0.1 0.1 0.1 0.1 0.2 0.1 0.18 0.06 0.15 0.15 0.07 0.05 Al2O3 % oxide 0.05 0.04 0.05 0.05 0.14 0.04 0.17 0.03 0.15 0.15 0.04 0.03 Fe2O3 % oxide 0.25 0.30 0.66 1.02 0.19 0.26 1.30 0.49 0.84 0.87 0.82 0.72 MgO % oxide 0.039 0.031 0.047 0.048 0.134 0.153 0.2 0.5 0.2 0.2 0.1 0.4 CaO % oxide 0.0008 0.0007 0.0009 0.0009 0.0006 0.0008 0.0007 0.0006 0.0010 0.0009 0.0007 0.0009 Na2O % oxide 0.0007 0.0010 0.0028 0.0022 0.0034 0.0026 0.0035 0.0049 0.0038 0.0041 0.0048 0.0030 K2O % oxide 0.01 0.01 0.01 0.01 0.03 0.00 0.0154 0.0075 0.0106 0.0108 0.0060 0.0039 TiO2 % oxide 0.008 0.011 0.026 0.040 0.004 0.009 0.0483 0.0191 0.0325 0.0321 0.0290 0.0248 Cr mg kg21 4.860 5.160 14.328 22.055 4.307 4.500 18.08 8.42 9.93 10.90 10.85 8.20 Ni mg kg21 0.862 0.530 1.012 1.276 2.444 1.074 3.67 0.97 2.89 3.72 1.42 1.14 Pb mg kg21 0.4 0.5 0.6 0.9 0.3 0.2 1.119 0.462 0.680 0.766 1.128 0.661 Sn mg kg21 0.06 0.07 0.18 0.24 0.04 0.13 0.14 0.11 0.15 0.17 0.37 0.31 Th mg kg21 0.47 0.46 1.21 1.68 0.95 0.70 1.39 1.11 1.07 1.14 2.06 1.38 U mg kg21 0.1 0.1 0.3 0.4 0.2 0.1 0.36 0.24 0.22 0.26 0.32 0.24 V mg kg21 5.02 8.50 17.33 22.43 4.75 7.15 21.06 8.67 16.94 16.71 9.85 8.48 Ce mg kg21 0.13 0.14 0.51 0.49 1.02 1.23 0.83 1.12 2.00 2.02 0.26 0.10 Dy mg kg21 0.04 0.06 0.08 0.09 0.12 0.07 0.153 0.078 0.213 0.237 0.116 0.059 Er mg kg21 0.014 0.019 0.016 0.025 0.038 < 0.007 0.035 0.013 0.052 < 0.007 0.032 < 0.007 Eu mg kg21 0.010 0.015 0.022 0.023 0.032 0.047 0.038 0.054 0.092 0.094 0.032 0.013 Gd mg kg21 < 0.016 < 0.016 < 0.016 < 0.016 0.02 0.04 0.071 0.059 0.201 0.223 0.037 < 0.016 Ho mg kg21 < 0.012 0.01 0.01 0.01 0.02 0.01 0.015 < 0.012 0.021 0.030 0.018 < 0.012 La mg kg21 0.06 0.08 0.20 0.19 0.36 0.38 0.348 0.293 0.662 0.641 0.118 0.049 Lu mg kg21 0.0031 0.0040 0.0060 0.0064 0.0073 0.0055 0.009 0.008 0.010 0.018 0.007 0.009 Nd mg kg21 0.12 0.17 0.42 0.36 0.69 1.09 0.568 1.034 1.687 1.724 0.213 0.089 Pr mg kg21 0.03 0.05 0.09 0.08 0.14 0.22 0.117 0.203 0.334 0.395 0.050 0.116 Sm mg kg21 0.03 0.05 0.09 0.09 0.16 0.24 0.157 0.263 0.423 0.442 0.117 0.045 Tb mg kg21 < 0.028 < 0.028 < 0.028 < 0.028 < 0.028 < 0.028 < 0.028 < 0.029 < 0.029 < 0.029 < 0.028 < 0.028 Tm mg kg21 < 0.012 < 0.012 < 0.012 < 0.012 < 0.012 < 0.012 0.014 < 0.012 < 0.012 < 0.012 < 0.012 0.013 Y mg kg21 0.17 0.28 0.32 0.37 0.49 0.30 0.539 0.356 0.856 0.938 0.419 0.283 Yb mg kg21 0.0170 0.0232 0.0318 0.0397 0.0529 0.0282 0.050 0.042 0.061 0.071 0.040 0.029 Table 5 Summary of chemical analysis data for the second Tamm’s reagent leach Element Units SC1 SC2 SC3 SC4 SC5 SC6 SC7 SC8 SC9 SC10 SC11 SC12 SiO2 % oxide 0.055 0.045 0.036 0.037 0.064 0.027 0.11 0.05 0.10 0.10 0.03 0.02 Al2O3 % oxide 0.036 0.027 0.022 0.020 0.053 0.017 0.10 0.04 0.09 0.10 0.02 0.01 Fe2O3 % oxide 0.11 0.10 0.32 0.51 0.06 0.04 0.317 0.421 0.179 0.193 0.272 0.209 MgO % oxide 0.029 0.024 0.023 0.022 0.051 0.023 0.128 0.210 0.129 0.133 0.016 0.091 CaO % oxide 0.0006 0.0006 0.0007 0.0006 0.0008 0.0008 0.0008 0.0007 0.0008 0.0008 0.0006 0.0006 Na2O % oxide 0.0002 0.0002 0.0004 0.0003 0.0004 0.0004 0.0007 0.0008 0.0005 0.0005 0.0004 0.0004 K2O % oxide 0.0028 0.0028 0.0031 0.0021 0.0083 0.0009 0.0081 0.0045 0.0049 0.0051 0.0014 0.0009 Cr mg kg21 1.695 1.404 6.229 9.186 1.053 < 1.5 4.17 5.92 2.49 2.42 2.96 1.97 Ni mg kg21 0.420 0.336 0.618 0.530 1.008 0.262 2.24 0.881 1.36 1.66 0.633 0.575 Pb mg kg21 0.196 0.149 0.266 0.536 0.191 0.034 0.31 0.23 0.11 0.12 0.37 0.22 Sn mg kg21 0.04 0.04 0.12 0.18 0.01 0.02 0.045 0.117 0.040 0.046 0.112 0.091 Th mg kg21 0.13 0.10 0.29 0.38 0.14 0.10 0.234 0.517 0.266 0.284 0.210 0.161 U mg kg21 0.021 0.019 0.075 0.106 0.031 0.011 0.04 0.07 0.02 0.03 0.04 0.03 V mg kg21 2.1 2.7 9.6 12.6 1.3 0.8 7.89 12.76 3.04 3.23 3.67 2.77 Ce mg kg21 0.256 0.203 0.316 0.330 0.547 0.400 0.79 0.97 1.14 1.28 0.11 0.06 Dy mg kg21 0.023 0.022 0.027 0.030 0.032 0.016 0.06 0.04 0.05 0.05 0.03 0.01 Er mg kg21 < 0.008 < 0.008 < 0.008 < 0.008 0.01 < 0.007 < 0.009 < 0.008 < 0.011 < 0.007 < 0.009 < 0.007 Eu mg kg21 0.009 0.010 0.010 0.011 0.012 0.011 0.021 0.036 0.036 0.039 0.008 0.003 Gd mg kg21 < 0.016 < 0.016 < 0.015 < 0.015 < 0.018 < 0.018 < 0.020 0.016 0.032 0.037 < 0.019 < 0.016 Ho mg kg21 < 0.011 < 0.012 < 0.012 < 0.012 < 0.013 < 0.012 < 0.013 < 0.011 < 0.014 < 0.014 < 0.013 < 0.011 La mg kg21 0.113 0.108 0.131 0.134 0.198 0.110 0.288 0.250 0.348 0.393 0.053 0.019 Lu mg kg21 0.0015 0.0019 0.0013 0.0034 0.0011 < 0.0014 0.006 0.007 0.003 0.002 0.001 0.001 Nd mg kg21 0.1493 0.1505 0.2255 0.2340 0.3286 0.2886 0.538 0.882 0.983 1.068 0.069 0.025 Pr mg kg21 0.023 0.024 0.052 0.049 0.088 0.063 0.141 0.201 0.209 0.209 0.033 0.027 Sm mg kg21 0.044 0.040 0.050 0.055 0.066 0.057 0.122 0.215 0.211 0.220 0.030 < 0.014 Tb mg kg21 < 0.027 < 0.027 < 0.026 < 0.026 < 0.026 < 0.027 < 0.027 < 0.027 < 0.027 < 0.027 < 0.027 < 0.027 Tm mg kg21 < 0.011 < 0.011 < 0.011 < 0.011 < 0.011 < 0.011 < 0.012 < 0.011 < 0.011 < 0.011 < 0.011 < 0.012 Y mg kg21 0.102 0.111 0.100 0.117 0.102 0.062 0.184 0.141 0.151 0.149 0.081 0.056 Yb mg kg21 0.008 0.008 0.009 0.013 0.008 0.004 0.015 0.014 0.011 0.009 0.006 0.004 504 Analyst, June 1997, Vol. 122Principal Component Analysis of the Whole Rock and Leached Data Principal component analysis10 (PCA) was applied to the multielement data matrix of the whole rock analyses and the combined data from the two leaches of the sandstones samples. The analysis of the whole rock samples and the oxalate leachates yielded two data matrices of element concentration versus sample.The rock samples can be assumed to be made up of varying proportions of similar mineral groups depending on the location of the sample within the borehole. Each of the elemental analyses for the twelve samples therefore reflects different proportions of these mineral components. Similarly, the oxalate leaching medium can be assumed to have leached out different amounts of hematite (and possibly other minerals) from each of the slightly different samples.The composition of the components of the mixture or the amount of each component present cannot be calculated from the elemental composition of a single rock sample or leachate solution. If, however, the data for all of the whole rock analyses or the leachate analyses are examined by multivariate statistical methods it is possible, in theory, to estimate the number of components responsible for the two data matrices, the composition of those components and the different proportions of each component in each sample.In the first stage of the data reduction, PCA identifies a vector or abstract variable which has the highest correlation coefficient with all the chemical elements in the matrix and, hence, accounts for the greatest proportion of the total variance. This vector is called the first principal component (PC1).A second principal component (PC2) is constructed from the residual matrix, orthogonal (i.e., uncorrelated) to PC1 and is calculated so as to account for the maximum proportion of the remaining variance. Further PCs up to the original number of samples (in this case 12 for the whole rock and 24 for the leachate data) can be calculated and will account for the total variance in the data set. The first PCs describe underlying properties of the data, in this case the geochemical components of the whole rock or leachate system, and the remaining PCs are associated with additional sources of variation attributable to sampling, the sample preparation and analysis.The data were pre-treated by subtracting the mean of each matrix column from the values in each column (a procedure known as data centering) and scaled by dividing each of the resulting values by the standard deviation for that particular column. This procedure gives all variables an equal weighting so that the major elements do not dominate and hide trends in the trace element data.The PCA deconvolution of the data was carried out on the scaled matrices using the NIPALS algorithm described by Wold et al.10 programmed in Microsoft Visual Basic and using Statsoft Statistica (V5.0). Stage two of the data reduction involves determining how many true geochemical components there are in the system by applying statistical and empirical tests on the PCs.For this study three methods were used to identify the number of true components. For each PC its associated eigenvalue describes its variance and hence its relative importance. By taking the ratio of successive eigenvalues, a number of workers11–13 have shown that the number of significant components is indicated by the maximum value of the ratios, if there are more than one local maxima the second local maximum is chosen in order of decreasing PC number.Malinowski14 suggested an empirical method where a minimum value of a function, calculated from the eigenvalues and the number of rows and columns in the data matrix indicates the number of true components. The results of these two methods for the whole rock and leachate data are summarised in Table 6. In the whole rock data both methods indicate there are four components. The situation in the leachate data is not quite so clear as the eigenvalue ratio suggests four components and the indicator function 11.To try and clarify the position for the leachate data set a third test, as suggested by Hopke,15 was carried out on the leachate data set. In this method the PCs are varimax rotated15 and the number of components is indicated by the number of eigenvalues, calculated from the rotated PCs, greater than one. Table 6 shows that there are four components with eigenvalues greater than one and the fifth eigenvalue close to 1 suggesting that four to five true geochemical components exist in the data.The varimax rotation option was not available on the Visual Basic programme and had to be carried out using the Statistica software. This varimax rotation test could not be performed on the whole rock data set as the Statistica programme requires more samples than variables to carry out the required calculations. Once the number of components has been established, the next step involves placing physical significance on the abstract factors (PCs) that have been shown to be true factors.At this stage the PCs are not directly representative of the true components but are linearly related. The ultimate solution to the deconvolution involves discovering the mathematical rotations necessary to convert the abstract PC vectors into true factors. This is discussed in the section on quantitative interpretation. Table 6 Summary of test parameters used to determine the number of true components in the whole rock and leachate data sets Whole rock data Combined leach data Varimax Number Indicator Number Indicator rotated of factors Ratio function of factors Ratio function eigenvalue 1 4.08 0.00494 1 3.01 0.001201 7.62 2 1.81 0.00469 2 2.85 0.000957 7.79 3 1.49 0.00458 3 1.41 0.000874 3.52 4 3.05 0.00426 4 1.92 0.000788 1.25 5 1.11 0.00477 5 1.18 0.000751 0.96 6 2.84 0.00496 6 1.36 0.000689 0.71 7 1.36 0.00630 7 1.74 0.000611 0.08 8 1.38 0.00854 8 1.45 0.000550 0.39 9 1.39 0.01262 9 3.02 0.000476 0.29 10 2.01 0.02003 10 1.10 0.000480 0.20 11 1.75 0.000463 0.05 12 1.29 0.000469 0.02 13 1.68 0.000467 0.07 14 1.43 0.000486 0.01 Analyst, June 1997, Vol. 122 505The initial approach to giving physical significance to the abstract PCs is a qualitative method. Qualitative Interpretation For each PC there is a score vector which is a representation of the original composition in terms of the new, abstract, PC variable and is linearly related to the true geochemical components in the rock samples being studied.The weights vector associated with each PC represents a correlation between each element measured and the abstract PC component. By examining the element patterns in the weight vectors of significant principal components, in conjunction with some basic geochemical and mineralogical knowledge of the materials under study, the PCA data can reveal qualitative information regarding the geochemical phases present and the distribution of trace metals amongst them.For the whole rock data a cluster plot, produced with the Statistica software, of the four significant weights vectors using euclidean distance and Ward’s clustering algorithm16 is shown in Fig. 2. By cutting the distance axis at ca. 44% four clusters are formed representing the four geochemical components in the whole rock data. These can be tentatively assigned to geochemical/ mineralogical sources. The major elements Na, Al and K are associated with Sr, Ba, and Ni and probably represent a feldspar grouping.Titanium, Ca, REEs, V, Cr and Zr form a group which probably represents resistate minerals. Magnesium groups with Mn possibly represents dolomite which is known to be present in the deeper samples.17 Iron, Si, Sn, U, Th and Pb probably represent a combination of hematite, clay minerals and Fe-oxy-hydroxides. As there are only 12 samples the PCA cannot be expected to identify all geochemical components but it is clear that the actinide elements are associated with a different geochemical phase from the lanthanides.The interpretation of the leachate data has been clarified by the use of varimax rotation on the PCA weights. This rotation tends to drive variable weights toward either zero or one on a given component. It precludes the emergence of a strong general component by making the variance explained by the individual components more equal, resulting in orthogonal components which are often more readily identifiable as specific source components.18 Table 7 shows the varimax rotated weights and clearly indicates that component 1 is REE dominated with some association with Si, Al and Ni probably related to clay materials.Component 2 is associated with Fe, Ti, Cr, Pb, Sn, Th, U and V probably derived from hematite dissolution. Component 3 is Si, Al, K and to some extent Ni dominated and is possibly related to feldspar dissolution. Component 4 is Mg dominated and probably derives from dolomite dissolution. Finally component 5, which may not be significant, is associated with Ca, possibly from a residual calcite source.The qualitative interpretation of the data clearly shows that more than one geochemical component is being extracted by the Tamm’s reagent and that the trace element distributions within these components are quite different. Quantitative Interpretation Theory Having identified that the procedure for extracting the iron oxide phase has extracted more than one physico-chemical component the next stage is to use the abstract components identified by PCA to quantify the composition of these components and how much of each component is in each sample.To simplify the deconvolution step it was decided to split the extraction data into two data sets, the first and second leach data, and deal with quantification of each separately. Although the mean centering and scaling of the data as previously described enables a clear qualitative interpretation, quantitative deconvolution of the PCA scores and weights obtained from data scaled in this manner can be complicated.A simplified scaling procedure was applied to each data set. Each determinand was scaled to a percentage of the highest value of that determinand found in the original rock sample, thereby giving a greater weighting to determinands which are preferentially extracted by the Tamm’s reagent. Each data set was subjected to PCA deconvolution as described for the qualitative data reduction.To carry out the quantification step from the abstract components identified by the PCA a modified procedure similar to that described by Malinowski19 was used. In chromatographic procedures with two way data (e.g, time profile/UV spectra) where there is poor resolution of components PCA related techniques have been used to resolve the components. This has been carried out by obtaining estimates of the chromatographic profiles by analysis of eigenvalues of the data set (evolving factor analysis, EFA) and refining these chromatographic profiles using the pseudoinverse calculation and the original raw data set to produce an estimate of the component concentrations.By forward and Fig. 2 Cluster plot of weights for the combined Tamm’s reagent leachate data set. Table 7 Varimax rotated weights for the combined leachate data set Factor Factor Factor Factor Factor 1 2 3 4 5 SiO2 0.46 0.24 0.83* 20.02 0.08 Al2O3 0.56 0.17 0.77 20.09 0.10 Fe2O3 0.21 0.94 0.19 0.09 0.07 MgO 0.35 0.17 20.01 0.79 20.04 CaO 0.36 0.31 0.07 20.01 0.87 Na2O 0.34 0.57 0.32 0.64 0.00 K2O 0.18 0.16 0.90 0.16 20.06 TiO2 0.23 0.95 0.10 20.02 0.02 Cr 0.08 0.95 0.15 20.02 0.08 Ni 0.57 0.36 0.67 0.02 0.09 Pb 20.01 0.92 0.26 0.17 0.04 Sn 20.08 0.80 20.20 0.39 0.18 Th 0.07 0.81 0.23 0.47 0.16 U 0.09 0.89 0.26 0.28 0.10 V 0.29 0.90 0.12 20.14 0.03 Ce 0.95 0.03 0.26 0.07 0.07 Dy 0.63 0.50 0.49 0.15 0.11 Eu 0.91 0.25 0.21 0.19 0.05 La 0.91 0.10 0.37 0.00 0.08 Nd 0.97 0.03 0.16 0.10 0.07 Pr 0.95 0.06 0.14 0.18 0.17 Sm 0.95 0.14 0.17 0.17 0.05 Y 0.62 0.51 0.45 0.20 0.09 * Numbers in bold are > 0.7. 506 Analyst, June 1997, Vol. 122backward iterations, each time eliminating values outside boundary conditions (e.g., negative concentrations), true chromatographic profiles and component concentrations can be calculated. In this study EFA is not appropriate to estimate the component depth profiles as the geochemical variations are not well behaved gaussian peaks as in chromatography.The first estimate of the geochemical component depth profiles was obtained by looking at the correlation of the abstract component depth profiles, derived from PCA, with the scaled profiles for each determinand, using the determinand with the highest correlation coefficient with the abstract profile as the first estimate. The determinand profiles used were Ti, Mg, Al and Sn for the first leachate data set and V, Al, Mg and K for the second leachate data set.If the leachate data are represented by a mixture matrix in which the chemical constituents are represented by the rows and the samples by columns (matrix M). Then the mixture matrix can be considered to be the product of a column matrix C in which each column represents the composition one of the chemical components in the mixture and a row matrix P in which each row represents the proportions of each component in the mixture.In matrix notation the relationship between the three matrices is shown in eqn. (1). M = CP (1) Given that M is known and that an initial guess at P can be made from the element profiles indicated in Tables 4 and 5 then first estimate of the concentration matrix (C) is given by eqn. (2) C = MP21 (2) where P21 represents the matrix inverse of P. This solution is only feasible if P is a square matrix, which is not the case for the leach data.This can be solved, however, using a least squares solution or ‘pseudoinverse’ method shown in eqn. (3) C = MPA[PAP]21 (3) where PA is the transpose of P. This allows the direct calculation of the concentration matrix. Since the proportion matrix is only a first approximation it is likely that calculation of the concentration matrix calculated from these values will give non-realistic values for some samples (e.g., concentrations < 0). In order to arrive at a meaningful solution it is necessary to use an iterative approach similar to that described by Gamp et al.20 In this method a first approximation of C is calculated from eqn.(3) using an estimation for matrix P. Values in C (which through scaling is expressed as a percentage) less than 0 are corrected to 0 and values greater than 100 are corrected to 100. Using the modified C matrix the pseudoinverse expression [shown in eqn. (4)] is used to calculate a new P matrix. P = [CAC]21 CAM (4) Negative concentrations in P are corrected to 0 and values greater than 1 corrected to 1 and the iterative procedure is started again producing the next approximation for C.Iterative modifications of P and C are continued until meaningful values for both matrices are produced. This procedure was carried out for both leach data matrices using 4 and 5 components as indicated in Table 6. In practice it was found that physically meaningful components could only be obtained using the 4 components for both leach data sets.This suggests that Hopke’s test15 (see Table 6) for defining the number of components in the leachate data set as 4 was correct. During the forward and backward iteration process it was also found that the speed of convergence was improved if a maximum boundary for the concentration matrix of 30% was imposed for the first few iterations which was relaxed to 100% for the final iterations. As a final refinement, the elemental composition matrix from the final iteration was subjected to iterative target factor analysis (50 iterations) as described by Malinowski19 to ensure that the data fitted the PCA model.These data were used to calculate the final component profile matrix using the pseudoinverse, as previously described. The final concentration and profile matrices, corrected for detection limits, for both leachate data sets are shown in Tables 8 and 9. Interpretation of the Quantitative data Assuming that the major oxides make up the majority of each component, Table 10 shows the compositions expressed as percentage major oxide.Two components from each leach Table 8 Resolved components from the Leach 1 data set Com- Com- Com- Component 1 ponent 2 ponent 3 ponent 4 SiO2 Leach 1 0.043 0.012 0.293 0.168 Leach 2 0.000 0.480 0.180 0.106 Al2O3 Leach 1 0.016 0.000 0.298 0.123 Leach 2 0.000 0.288 0.159 0.103 Fe2O3 Leach 1 0.962 0.152 0.661 1.003 Leach 2 0.397 2.312 0.605 0.094 MgO Leach 1 0.021 0.694 0.356 0.074 Leach 2 0.024 0.195 0.622 0.004 CaO Leach 1 0.001 0.001 0.001 0.001 Leach 2 0.000 0.010 0.003 0.001 Na2O Leach 1 0.002 0.004 0.006 0.007 Leach 2 0.000 0.003 0.002 0.000 K2O Leach 1 0.004 0.003 0.027 0.020 Leach 2 0.003 0.006 0.005 0.008 TiO2 Leach 1 0.039 0.005 0.022 0.028 Leach 2 0.026 0.071 0.017 0.005 Cr Leach 1 20.701 0.000 0.830 3.932 Leach 2 9.023 23.234 3.692 1.273 Ni Leach 1 0.555 0.314 6.277 4.296 Leach 2 0.128 5.868 2.176 2.001 Pb Leach 1 0.856 0.012 0.523 1.794 Leach 2 0.197 3.918 0.217 0.264 Sn Leach 1 0.215 0.095 0.000 0.634 Leach 2 0.082 1.282 0.231 0.000 Th Leach 1 1.417 0.610 0.690 3.190 Leach 2 0.289 1.936 1.136 0.013 U Leach 1 0.360 0.098 0.123 0.341 Leach 2 0.095 0.347 0.058 0.017 V Leach 1 21.767 0.282 10.283 0.747 Leach 2 18.074 1.454 8.418 0.009 Ce Leach 1 0.000 1.043 3.579 0.199 Leach 2 0.040 2.225 3.238 0.554 Dy Leach 1 0.046 0.030 0.354 0.250 Leach 2 0.006 0.289 0.110 0.051 Er Leach 1 0.019 0.000 0.041 0.058 Leach 2 0.000 0.063 0.015 0.004 Eu Leach 1 0.006 0.046 0.150 0.040 Leach 2 0.000 0.104 0.122 0.010 La Leach 1 0.045 0.234 1.146 0.193 Leach 2 0.031 0.936 0.794 0.257 Lu Leach 1 0.004 0.008 0.021 0.014 Leach 2 0.006 0.000 0.009 0.001 Nd Leach 1 0.000 1.017 2.935 0.000 Leach 2 0.007 1.514 3.059 0.272 Pr Leach 1 0.000 0.252 0.608 0.058 Leach 2 0.009 0.363 0.649 0.069 Sm Leach 1 0.001 0.243 0.726 0.131 Leach 2 0.018 0.331 0.692 0.045 Y Leach 1 0.190 0.218 1.381 0.857 Leach 2 0.056 0.926 0.336 0.147 Yb Leach 1 0.023 0.028 0.103 0.073 Leach 2 0.012 0.045 0.022 0.009 Analyst, June 1997, Vol. 122 507(components 1 and 4 from leach 1 and components 1 and 2 from leach 2) are predominantly made up of iron oxide. Component 1 from leach 1 and 2 and components 4 and 2 from leach 1 and 2 data have very similar composition and it is assumed to represent the same mineral host in each leach. Examination of the profiles of these four components (Table 9) shows that component 1 from both leach 1 and 2 show a similar trend with the first leach having a minimum at sample 5.In a description of the mineralogy of these core samples, Kemp and Pearce17 describes sample 5 as being anomalous (drab grey sandstone) lacking in the characteristic red colouration of the other samples and confirms, by XRD, that it is depleted in hematite compared to the other samples. It can therefore be concluded that component 1 probably represents hematite in both the first and second leaches.The other iron containing host represented by components 4 and 2 in leaches 1 and 2 has lower total iron oxide content and higher Si and Al content than the hematite component and makes up a smaller proportion of the total iron leached. This is probably amorphous iron oxide or iron oxyhydroxides which are known to exist in these samples. 16,20 Component 2 and component 3 in leach 1 and 2 are dominated by magnesium oxide and appears predominantly in the deeper samples (samples 6–12).Kemp and Pearce17 show that dolomite (Mg/Ca carbonate) is present in these deeper samples. The fact that the Ca content of these components is low can be explained by the oxalic acid in the Tamm’s reagent causing precipitation of insoluble Ca oxalate. It is therefore assumed that dolomite is the mineral source of these components. Component 3 and component 4 of leach 1 and 2 appear to be iron aluminosilicates. M�ossbauer spectroscopy of whole rock and clay fractions21 of these samples shows iron is present as hematite and as iron alumino-silicate.Kemp and Pearce17 suggest that chlorite is present in these samples and the composition of these components is consistent with reported analysis of chlorite samples.22 Having estimated the compositions of the components in the leach solutions and the proportions of each component in each sample and made a tentative identification of their mineral source, the purpose of this study has been achieved, namely to find out the chemical composition of the iron oxide phases of the sandstone formation.The profiles for the two leaches show similar patterns (Table 9) and therefore for clarity only leach 1 data has been plotted. Fig. 3 shows example profiles for Fe, U, Ce and Ni derived from: (i) the total concentrations found in leach 1; (ii) the concentration found in the hematite component of leach 1; (iii) the concentration found in the Fe oxy-hydroxide component of leach 1; and (iv) the sum of the concentrations found in both the hematite and Fe oxy-hydroxide components of leach 1.The profiles are plotted against sample number (which are in depth order) rather than actual depth as this shows up the patterns of the profiles more clearly. The Fe profile shows that the total Fe leached from the shallow samples (1–6) is derived almost entirely from iron oxides; however, for the deeper samples a significant proportion of the total iron leached is derived from non-iron oxide sources.The majority of the iron is found in the hematite component. The actinides as shown by U behave similarly to iron. In contrast, total Ce leached, which is representative of the rare earth elements, is almost exclusively derived from non-iron oxide sources. Nickel behaves in a manner intermediate between the actinides and the rare earth elements, again the majority of the Ni in total leached data is derived from non-iron oxide sources.These profiles illustrate the problem, noted by other investigators, of the poor specificity of the selective extractant. Clearly, if the unprocessed data alone had been used to make an Table 9 Proportions of resolved components from the leach 1 and leach 2 data sets Sample Component Component Component Component No. 1 2 3 4 Leach 1— 1 0.198 0.00 0.077 0.038 2 0.305 0.00 0.085 0.00 3 0.711 0.004 0.101 0.001 4 1.01 0.00 0.077 0.041 5 0.103 0.041 0.260 0.093 6 0.218 0.154 0.153 0.018 7 0.820 0.00 0.435 0.051 8 0.371 0.639 0.088 0.006 9 0.525 0.026 0.461 0.00 10 0.505 0.00 0.537 0.040 11 0.465 0.176 0.00 0.401 12 0.385 0.596 0.00 0.223 Leach 2— 1 0.098 0.022 0.034 0.111 2 0.129 0.014 0.032 0.082 3 0.520 0.051 0.011 0.107 4 0.704 0.107 0.00 0.00 5 0.036 0.00 0.078 0.439 6 0.022 0.004 0.049 0.071 7 0.353 0.00 0.170 0.947 8 0.556 0.00 0.311 0.061 9 0.076 0.00 0.220 0.484 10 0.077 0.00 0.229 0.616 11 0.188 0.075 0.00 0.077 12 0.103 0.044 0.089 0.025 Table 10 Major oxide percentage composition of resolved components in leach 1 and 2 and their probable mineral source Fe-Alumino Fe-Alumino silicate Fe-Oxy- Fe-oxy- silicate Hematite Dolomite (chlorite) hydroxides Hematite hydroxides Dolomite (chlorite) Probable Leach 1 Leach 2 mineral source Component 1 Component 2 Component 3 Component 4 Component 1 Component 2 Component 3 Component 4 SiO2 3.9 1.4 17.6 11.8 0.0 14.3 11.3 33.1 Al2O3 1.4 0.0 17.9 8.6 0.0 8.6 10.0 32.1 Fe2O3 88.5 17.4 39.7 70.5 88.0 68.7 38.0 29.2 MgO 2.0 79.8 21.4 5.2 5.3 5.8 39.1 1.3 Ca 0.1 0.1 0.1 0.1 0.0 0.3 0.2 0.3 Na2O 0.2 0.5 0.4 0.5 0.1 0.1 0.1 0.1 K2O 0.4 0.3 1.6 1.4 0.7 0.2 0.3 2.5 TiO2 3.5 0.6 1.3 1.9 5.8 2.1 1.0 1.4 508 Analyst, June 1997, Vol. 122interpretation of the geochemistry of the iron oxide phases, the conclusions would have been erroneous. Another reported problem with the sequential leach methodology is the re-absorption of trace metals onto the undissolved solid phase during the leaching process.To investigate this, the major and trace metals in the hematite and iron oxy-hydroxy phases from both the first and second leach were calculated as a percentage of the total amount of each element from the sum of the two leaches. Subsequently, to take into account the different amounts of each component in leach 1 and 2 the percentage compositions were ratioed to the iron content of each respective leachate. The results of this are shown in the bar charts Fig. 4. Each bar represents the ratio of the percentages (relative to iron) in leach 2 compared with leach 1. For iron the scaled values give an equal distribution of iron in leach 1 and 2. For those elements with ratios > 1 there is a relative enrichment in leach 2 compared with leach 1. Ratios < 1 indicate a relative depletion in leach 2 compared with leach 1. If re-absorption was Fig. 3 Concentration profiles for Fe, U, Ce, and Ni in the resolved components of the first Tamm’s reagent leach.Sample number in increasing depth order. 5, Total extracted by Tamm’s reagent; ~, total associated with Fe oxide sources; +, total associated with hematite; total associated with Fe-oxyhydroxides. Fig. 4 Ratio of the proportions of each element (relative to Fe) in the first and second Tamm’s reagent leach. Analyst, June 1997, Vol. 122 509the dominating factor then enrichment in leach 2 would be expected. Fig. 4 shows that for some elements there is evidence for re-absorption in both iron oxide phases (e.g., Ce), however, some elements show a relative enrichment f leach 1 (e.g., Th) and for many of the trace metals the relative proportions in each leach is very similar.In summary there is no clear evidence for systematic re-absorption during the leaching procedure used in this study. Conclusions Chemometric processing of whole rock and leachate data of related rock samples gives ‘added value’ to the bare analytical results.Analysis of multi-element data obtained from whole rock samples from the same formation, using chemometric methods, provides macroscale information on trace metal associations in the rock which is complimentary to microanalysis methods. Factor analysis of multi-element data obtained from extraction of selected mineral phases can be used to show both qualitative and quantitative information about the chemical composition of selected mineral phases.The problem of nonselectivity of the leaching reagent can be minimised by chemometric processing of the data. The use of chemometric processing of leaching data, produced in a similar manner to that described in this work, could provide an important tool for measuring the trace element distributions in rocks soils and sediments for geochemical and environmental applications. For example a non-specific reagent with the ability to hold trace elements in solution (e.g., a mineral acid with a pH < 1) could be used as the extractant, separating the different phases by their rate of reaction with this media by time based sampling.The overlap between the dissolution of different phases could then be resolved by factor analysis of the time series data. The authors would like to thank UK Nirex for funding this study, the NERC ICP-MS facility at Royal Holloway University of London for the use of the ICP-MS and Dr. M. Thompson, Dr. A. H. Bath, Mr.D. L. Miles, Dr. Richard Metcalfe and Mr. S. Reeder for helpful comments and suggestions on the text of this paper. This paper is published with the approval of the Director, British Geological Survey (NERC). Appendix 1 Sample Preparation and Analytical Methods Instrumentation The ICP spectrometer used in this work was a Perkin-Elmer Plasma II sequential scanning system with twin 1 m vacuum monochromators: monochromator A, with a 3600 line/mm grating and wavelength range of 160–400 nm, and monochromator B, with an 1800 line/mm grating and wavelength range of 160–800 nm.The monochromator gratings, plasma power, plasma gas flows, plasma viewing height, 50-position autosampler, and nebuliser peristaltic pump were all under computer control. The analytical emission lines and plasma operating conditions used in this work are summarised in Table 11. The ICP-MS instrumentation used was a VG PlasmaQuad 2 operated in scanning mode over the mass range 50–245 u.The isotopes used for analysis and the system operating conditions are given in Table 12. Reagents The chemical reagents used were: glacial acetic acid, (Merck, Poole, Dorset, UK; Aristar grade), ammonium acetate (Aldrich Gold label), ammonium oxalate monohydrate (Aldrich, Gillingham, Dorset, UK), 35% w/w hydrochloric acid (Merck; Aristar grade), 40% m/m hydrofluoric acid (Merck; Aristar grade), 30% m/m hydrogen peroxide (Merck; Aristar grade), anhydrous lithium metaborate (Merck; Spectrosol), 4 + 1 lithium metaborate –lithium tetraborate mixture [Spectroflux (R) 100B Johnson Matthey, Royston, Herts, UK], 70% m/v nitric acid (Merck; Aristar grade), oxalic acid (Merck; Aristar grade) and 70% m/m perchloric acid (Merck; Aristar grade).Table 11 ICP-AES operating conditions Mono- Plasma Element Wavelength Viewing chromator source Al 396.152 15 B stdcond Ba 455.403 15 B stdcond Ca 315.887 15 B stdcond Cr 205.552 8 B OPT Fe 259.940 15 A stdcond K 766.490 9 B stdcond Mg 279.079 15 A stdcond Mn 257.610 15 A stdcond Na 589.592 15 B stdcond Ni 231.604 11 B OPT S 180.731 8 A OPT Si 251.611 15 B stdcond Sr 407.771 15 B stdcond Ti 334.941 15 A stdcond Zr 343.823 15 B stdcond V 292.402 15 A stdcond Ce 418.660 11 B Ree Dv 353.170 11 A Ree Er 390.631 11 B Ree Eu 381.962 11 B Ree Gd 335.047 11 A Ree Ho 345.6 13 A Ree La 333.749 11 A Ree Lu 261.537 12 A Ree Nd 430.357 11 B Ree Pr 422.535 11 B Ree *Sc 424.683 n/a† n/a n/a Sm 359.267 11 B Ree Tb 350.904 11 A Ree Tm 313.118 11 A Ree Y 371.026 11 B Ree Yb 328.924 9 A Ree Plasma source file conditions— Nebuliser Plasma flow/ flow/ Auxiliary/ Nebuliser/ Source Power/W l min21 l min21 l min21 l min21 stdcond 1000 1.00 15.0 1.0 1.0 OPT 1410 0.88 15.0 1.0 1.0 Ree 1200 1.05 15.0 1.0 1.0 * Myers–Tracy signal compensation used.† n/a, Not applicable. Table 12 ICP-MS Operating Conditions Plasma power 1250 W Nebuliser gas flow 0.8 l min21 Ar Plasma gas flow 13.75 l min21 Ar Auxiliary gas flow 0.41 l min21 Nebuliser pump rate 0.9 ml min21 Distance of load coil to sampling cone 10 mm Sampling Cone aperture 0.7 mm Skimmer Cone aperture 1.0 mm Element Mass/u Cr 52 Ni 60 Sn 118 Pb 208 Th 232 U 238 510 Analyst, June 1997, Vol. 122High purity analytical reagent grade water (resistivity 18 MW cm) was prepared using a commercial laboratory reverse osmosis/deioniser system (Elga, High Wycombe, Bucks, UK). Reagent blanks were analysed for all procedures.To check the accuracy of the whole rock analysis a certified reference material, USGS Andesite AGV-1, prepared by the US Geological Survey, was analysed. Sample preparation Twelve samples from 9 borehole cores were selected for analysis. The foil and wax coating were removed from the middle portion of each core, which then was split into discs perpendicular to the core axis with a plastic-covered hammer and chisel. The contaminated outer surface was removed, leaving several clean, round sections of rock; these were broken into pieces of 1–4 cm in diameter. To prevent contamination, the samples were allowed to come in contact with only non-metallic tools and containers.Equipment was washed with an alkaline detergent solution (Micro), soaked in 25% v/v HNO3, and rinsed with deionised water before use. Sample powders were prepared from the 1–4 cm fragments of whole rock in the following stages: (i) the 1–4 cm sample was wrapped in thick plastic and crushed with a covered hammer to pieces < 1 cm in size; (ii) the < 1 cm pieces were ground with a mechanical mortar and pestle in stages until all the powder passed through a 400 mm nylon mesh sieve held by a plastic frame and collected in a plastic pan; (iii) the < 400 mm sample powder was transferred to a sheet of glazed paper, coned and quartered until a 100 g portion was separated; (iv) this 100 g portion of < 400 mm powder was ground in stages and passed through a 125 mm nylon mesh sieve; (v) both < 400 mm and < 125 mm powder fractions were weighed, stored in glass jars with plastic lids. The < 400mm fraction was bagged and refrigerated.The < 125 mm fraction was dried at 105 °C for 24 h and stored in a desiccator. Whole rock dissolution Acid digestion and fusion procedures were used in the whole rock analysis. The acid digestion procedure was used to produce solutions suitable for ICP-AES analysis for the major and trace elements (Al, Fe, Mg, Ca, Na, K, Ti, Mn, Ba, Sr) and for ICPMS analysis for lower abundance trace elements (Cr, Ni, Pb, Sn, Th, U, V, Zr).The fusion procedure, followed by an ion exchange preconcentration–separation procedure, was utilised to provide solutions for REEs, La, and Y determination by ICPAES or ICP-MS. Solutions for ICP-AES analysis for Si in the whole rock were prepared by a modified fusion procedure. The acid digestion procedure consisted of the following steps: (1) approximately 0.5 g of dry, < 125 mm powder was weighed accurately into a Teflon PFA beaker with a graphite base; (2) 15ml of concentrated HF was added to the beaker.The beaker was covered, and the powder digested at 20 °C for 48 h and then at 95 °C in a water bath for 3 h; (3) the beaker was uncovered and transferred to a hot plate. A 10 ml volume of concentrated HClO4 and 10 ml of concentrated HNO3 were added (to the solution in the beaker) and heated gently to dryness; (4) the residue was dissolved in 15 ml of concentrated HNO3 and again taken to dryness; (5) a second 15 ml volume of concentrated HNO3 was added to the beaker to dissolve the residue.This solution was transferred into a calibrated flask and made up to 50 ml with deionised water. The final solution was transferred to a high density polyethylene (HDPE) bottle for storage. The LiBO2 fusion–dissolution procedure used was based on that developed by Govindaraju and Mevelle.23 The method used was: (1) approximately 0.25 g of dry, < 125 mm powder and 1.0 g of LiBO2 flux were weighed accurately into a platinum crucible and mixed thoroughly; (2) the crucible was covered, placed in a preheated muffle furnace, and heated at 1000 °C for 2 h.The molten sample was swirled gently at 30 min intervals. On removal from the furnace, the crucible base was plunged into water to quench the melt and fracture the bead; (3) 20 ml of dissolving solution (100 ml of concentrated HCl, 10 g of (COOH)2·P2H2O and 5 ml of H2O2 diluted to 1 l with deionised water) was added to the crucible.The crucible was placed on a magnetic stirrer and the solution stirred for 30 min with a PTFE bar, after which it was transferred to a 100 ml calibrated flask. A second 20 ml aliquot of dissolving solution was added to the crucible, stirred for 30 min and transferred to the flask. The process was repeated until the bead was dissolved; (4) the flask was filled to the 100 ml mark with dissolving solution previously used to wash the sides and lid of the crucible.The sample solution was then transferred to a HDPE bottle pending further processing to separate and preconcentrate REEs. LiBO2, Li2B4O7 fusion–dissolution procedure for silicon Because of the high percentage of silica in the samples, the LiBO2 fusion procedure described previously had to be modified to achieve complete dissolution of Si. A 4 + 1 mixture of LiBO2 and Li2B4O7 fluxes was substituted for the LiBO2 flux; only 0.2 g of sample powder was fused and 25% HNO3 was used as the dissolving solution.All other conditions were identical to those outlined in the LiBO2 fusion procedure. After dissolution the sample was transferred to an HDPE bottle to await analysis. Ion exchange procedure for separation and preconcentration of REEs The REEs were separated using a ‘mini-column’ ion exchange procedure.23 The sample solution obtained by the fusion method described above (or the oxalate leachate) was ‘loaded’ onto columns constructed of 1 ml disposable plastic pipette tips packed with 2 g of Duolite 225 (SRC 16) (Merck) cation exchange resin.The detailed procedure is summarised in Table 13. Steps 1 and 2 were omitted if the columns had been used previously. The method for whole rock fusion solutions had to be modified for the oxalate leachates. The leach solution matrix was a more efficient eluent than the rock dissolving solution for moving the REEs down the column and a wash of smaller volume was required to give acceptable recoveries.Under these conditions the recovery for the three heaviest REEs (Tm, Yb, Lu) as tested on a spiked leachate matrix was still 10 to 15% low. Analytical protocols for ICP-AES and ICP-MS analysis For all analysis the ICP-AES instrument was recalibrated every 10 samples. For the whole rock analysis and REE separated fusion samples, a USGS standard rock (AGV-1) was processed Table 13 REE ion exchange separation procedure Acid Pump Step Acid concentration Volume of speed/ number type (% v/v) acid/ml ml min21 Comment 1 HCl 40 10 0.4 Conditioning 2 HCl 10 10 0.4 Conditioning 3 HCl 40 10 0.4 Conditioning 4 Sample 10 30 0.4 Loading 5 HNO3 12 18 (fusion) 0.4 Wash 10 (leachate) 0.4 Wash 6 25 10 0.4 Elution Analyst, June 1997, Vol. 122 511using the same dissolution procedures as the sandstone samples and analysed to check for accuracy and to maintain quality control.Internal standardisation (using Sc) was used on the whole rock digests to correct for matrix effects and to improve precision. Analysis of the whole rock and the leachates for the REEs required the use of matrix-matched standards because analytical emission lines (for the REEs) or the sample matrix (oxalate leachates) interfered with the Sc internal standard line. The solution sample preparation and the method of calibration for each of the different solution types is described below.For the acid-digested whole rock sample, 1 ml of solution was diluted to 10 ml with deionised water followed by the addition 0.4 ml of 1000 mg l21 Sc as internal standard. The samples were analysed against standards made up in 1% HNO3 containing the same concentration of Sc. The oxalate leachates were diluted 1 + 1 with deionised water and analysed against matrix matched standards. The REE eluates, in 25% HNO3, were analysed with no further preparation against matrix-matched standards.The ICP-MS instrument was calibrated every five samples with standards made up in 1% HNO3. The USGS standard rock AGV-1 was used for quality control for the whole rock analysis. The whole rock acid digestion samples were diluted 1 + 9 with deionised water before analysis. The oxalate leachate samples were prepared by digesting with H2O2 to reduce the dissolved solids content as described previously and then diluted 1 + 1 with deionised water.Analysis of standard rock material (USGS AGV-1) An international rock standard was prepared and analysed by the same procedures and methods as the sandstone samples, excluding initial crushing and grinding. The resulting analytical data are shown in Table 14, where they are compared with recommended values.24 All results are within the 95% confidence limits given in the data compiled by Gladney et al.,24 apart from Sr and Dy which are high and are probably due to contamination effects and Gd which is low and may be due to poor background correction in the ICP-AES analysis.Despite these small discrepancies the methodology was considered to be sufficiently accurate for the purposes of the studies carried out in this work. References 1 Chester, R., and Hughes, M. J., Chem. Geol., 1967, 2, 249. 2 Tessier, A., Campbell, P. G. C., and Bisson, M., Anal. Chem., 1979, 51, 844. 3 Rendell, P. S., Batley, G. E., and Cameron, A.J., Environ. Sci. Technol., 1980, 14, 314. 4 Tipping, E., Hetherington, N. B., Hilton, J., Thompson, D. W., Bowles E., and Hamilton-Taylor, J., Anal. Chem., 1985, 57, 844. 5 Rapin, F., Tessier, A., Campbell, P. G. C., and Carignan, R., Environ. Sci. Technol., 1986, 20, 836. 6 Kheboian, C., and Bauer, C. F., Anal. Chem., 1987, 59, 1417. 7 Sholkovitz, E. R., Chem. Geol., 1989, 77, 47. 8 Zielinski, R. A., Bloch, S., and Walker, T. R., Econ. Geol., 1983, 78, 1574. 9 Tamm, O., Medd. Skogsf�orsoeksues., 1922, 19, 387. 10 Wold, S., Ebenson, K., and Geladi, P., Chemom. Intell. Lab. Syst., 1987, 2, 37. 11 He, X., Li, H., and Shi, H., Fenxi Huaxue, 1986, 14, 34. 12 Wirsz, D. F., and Blades, M. W., Anal. Chem., 1986, 58, 51. 13 Zhang, P., Dudley, N., Ure, A. M., Littlejohn, D., Anal. Chim. Acta, 1992, 258, 1. 14 Malinowski, E. R., Anal. Chem., 1977, 49, 612. 15 Hopke, P. K., Atmos. Environ., 1982, 16, 1379. 16 Ward, J. H., J. Am. Stat. Assoc., 1963, 58, 236. 17 Kemp, S.J., and Pearce, J. M., UK Nirex Safety Studies Asessment Report, NSS/R239, in preparation. 18 Thurston, G. D., and Spengler, J. D., Atmos. Environ., 1985, 19, 9. 19 Malinowski, E. R., Factor Analysis in Chemistry, Wiley, New York, 1991. 20 Gamp, H., Maeder, M., Meyer, C. J., and Zuberbuhler, A. D., Talanta, 1985, 32, 1133. 21 Longworth, G., UK Nirex Safety Studies Assessment Report, NSS/ R303, in preparation. 22 Deer, W. A., Howie, R. A., and Zussman, An Introduction to the Rock Forming Minerals, Longman, 1982. 23 Govindaraju, K., and Mevelle, G., J. Anal. At. Spectrom., 1987, 2, 615. 24 Gladney, E. S., Jones, E. A., and Nickell, E. J., Geostand. Newsl., 1992, 16, 111. Paper 6/07953I Received November 25, 1996 Accepted February 18, 1997 Table 14 Comparison of AGV-1 analysis by the procedures described in this work to certified values 95% Detec- AGV-1 Recom- Contion (this mended fidence Element Unit limit work) values24 limits SiO2 % oxide — — 58.84 0.56 Al2O3 % oxide 0.003 17.1 17.15 0.3 Fe2O3 oxide 0.003 7.01 6.77 0.34 MgO % oxide 0.011 1.47 1.53 0.2 CaO % oxide 0.003 5.01 4.94 0.28 NaO2 % oxide 0.003 4.41 4.26 0.22 K2O % oxide 0.033 3.06 2.92 0.16 TiO2 % oxide 0.001 1.1 1.05 0.1 MnO % oxide 0.00001 0.097 0.09 0.02 Ba mg kg21 1.6 1224 1226 34 Cr mg kg21 0.05 9.4 10.1 4.8 Ni mg kg21 0.2 12.6 16 6 Pb mg kg21 0.2 35.3 36 10 Sn mg kg21 0.1 4.59 4.2 2.2 Sr* mg kg21 0.1 695 662 18 Th mg kg21 0.1 6.18 6.5 1 U mg kg21 0.1 1.99 1.92 0.3 V mg kg21 9 106 121 22 Zr mg kg21 3 249 227 36 Ce mg kg21 1.4 66 67 10 Dy* mg kg21 0.41 6.71 3.6 0.6 Er mg kg21 0.41 1.1 1.7 0.4 Eu mg kg21 0.05 1.7 1.64 0.2 Gd* mg kg21 < 0.96 5 1 Ho mg kg21 0.68 < 0.68 0.67 0.2 La mg kg21 0.41 36.1 38 6 Lu mg kg21 0.07 0.21 0.27 0.06 Nd mg kg21 1.51 28.6 33 6 Pr mg kg21 1.23 7.21 7.6 2.2 Sm mg kg21 0.68 6 5.9 0.8 Tb mg kg21 1.6 < 1.6 0.7 0.2 Tm mg kg21 0.7 < 0.7 0.34 0.12 Y mg kg21 0.1 17.5 20 6 Yb mg kg21 0.1 1.62 1.72 0.38 * See explanation in text. 512 Analyst, June 1997, Vol. 122 Determination of Trace Metal Distributions in the Iron Oxide Phases of Red Bed Sandstones by Chemometric Analysis of Whole Rock and Selective Leachate Data Mark R. Cave* and Karen Harmon Analytical Geochemistry Group, British Geological Survey, Keyworth, Nottingham, UK NG12 5GG Hematite is known to be a sink for trace metals and study of the trace metal distributions within hematite-rich formations can provide evidence of past groundwater activity.The aim of this study is to develop a method specifically for determining the trace metal content of naturally occurring hematite in a sandstone formation. The elemental compositions of twelve samples of permo-triassic Red Bed sandstones from the St. Bees formation, Sellafield, Cumbria, were determined. An acid digestion procedure was used to produce solutions suitable for ICP-AES analysis for the major and trace elements (Al, Fe, Mg, Ca, Na, K, Ti, Mn, Ba, Sr).The same solution was used for ICP-MS analysis for lower abundance trace elements (Cr, Ni, Pb, Sn, Th, U, V, Zr). Fusion with LiBO2, followed by an ion exchange preconcentration–separation procedure, provided solutions for rare earth elements, La, and Y determination by ICP-AES. Solutions for ICP-AES analysis of Si in the whole rock were prepared by a modified fusion procedure using 4 + 1 mixture of LiBO2 and Li2B4O7. A leaching procedure was developed to selectively extract the hematite phase of the rock samples.This procedure consisted of a preliminary extraction with 1 m acetic acid for 20 h at 20 °C to remove carbonates, followed by an extraction with 0.1 m oxalic acid–0.175 m ammonium oxalate (Tamm’s reagent) for 20 h at 70 °C to dissolve the iron oxide. The Tamm’s reagent leachate was analysed by ICP-AES and ICP-MS for the same suite of elements as the whole rock. Principal component analysis was applied to the whole rock and leachate data matrices.Qualitative assessment of the resulting principal components revealed elemental groupings which can be assigned to minerals known to exist in this rock type and showed the leaching procedure was not fully selective for the iron oxide phase. Chemometric mixture decomposition applied to the leachate data sets identified the extraction of four different mineral sources tentatively assigned to hematite, dolomite, chlorite and iron oxy-hydroxides. Quantitative estimates of the composition and proportion each of these components in each rock sample were calculated.Keywords: Inductively coupled plasma atomic emission spectrometry; inductively coupled plasma mass spectrometry; Red Bed sandstones; rare earth elements; selective extraction; principal component analysis; mixture modelling In order to understand and accurately model the migration of trace elements in groundwater systems it is essential to know how these elements are partitioned between the rock and the water.Hematite is known to be a sink for trace metals and study of the trace metal distributions within hematite-rich formations can provide evidence of past groundwater activity. The aim of this study is to develop a method specifically for determining the trace metal content of naturally occurring hematite in a sandstone formation. The distribution of trace elements in rocks can be identified by techniques with the spatial resolution necessary to analyse individual mineral grains on a mm scale.Such techniques are widely used in geochemical and mineralogical studies and include: proton induced X-ray emission (PIXE), scanning electron microscopy (SEM), energy dispersive X-ray emission (EDAX) and laser ablation microprobe ICP mass spectrometry (LAMP–ICP-MS). The data obtained by these methods give information on a small scale which does not necessarily reflect the distribution of trace elements in the bulk rock.In contrast, methods which homogenise the rock matrix before analysis, such as XRF analysis of pressed powder pellets and fusion discs or ICP emission/mass spectrometry of acid and fusion digests, give an integrated picture of the total trace element composition but very little information about the minerals in the rock and their associated trace metals. To obtain information on the trace metal associations in the bulk rock, a number of workers1,2 have developed extraction schemes in which mineral phases are selectively dissolved with carefully chosen reagents.By analysis of the extraction media the concentration of trace elements associated with the target mineral phase can be determined. However, two major weaknesses of the extraction technique have been demonstrated. 3–7 (i) The so called ‘selective extraction reagents’ are not specific for one mineral phase; therefore the associated analysis is not a true representation of the trace elements from a single phase.(ii) The chemical leaching process does not quantitatively extract the true total trace element concentration from the target phase as many trace elements are re-adsorbed onto the rock matrix during the leaching process. This study used a number of samples of permo-triassic Red Bed sandstones from the St. Bees formation, Sellafield, Cumbria, from one borehole to show that chemometric analysis of multi-element data from whole rock analysis and selective extraction can provide information on the mineral components of the bulk rock and help to overcome the shortcomings of selective extraction identified above.The samples, in depth order are labelled SC1–SC12. Experimental The sample preparation, digestion and analysis of the whole rock samples and the leachates has been carried out using standard techniques which are fully described in Appendix 1. Selective Extraction Procedures Following an examination of published results from previous work and reviews, e.g., Tessier et al.2 and Zielinski et al.,8 four reagents were used for trial experiments.For removal of carbonates, prior to iron oxide extraction, 1 m ammonium acetate (adjusted to pH 5 with glacial acetic acid) and 1 m acetic Analyst, June 1997, Vol. 122 (501–512) 501acid were chosen. For selective extraction of iron oxides Tamm’s reagent9 (0.175 m ammonium oxalate–0.1 m oxalic acid) and 0.1 m HCl were selected. Sodium dithionate–citrate mixtures have been used in some studies8 but were not considered in this work for the reasons outlined by Tessier et al.:2 (i) extraction with dithionate citrate can lead to the precipitation of trace metals because of the formation of sulfides as a result of disproportionation of dithionate during extraction; (ii) sodium dithionate cannot be purchased in a pure form and contains high levels of trace metals at or above the levels to be leached from the rock.Purification of the dithionate by ion exchange is a difficult and lengthy process; and (iii) sodium dithionate solutions are not compatible with atomic spectrometry sample introduction systems which quickly become clogged (and cause instrument malfunction).The oxalic acid–ammonium oxalate extract used by a number of workers (e.g., Zielinski et al.8), overcomes some of these disadvantages. It is claimed to be selective for iron oxides, can be obtained in high purity and the organic part of the matrix can be destroyed after the extraction to minimise sample introduction problems.The sample to extractant ratio was 5 g to 30 ml. The extractions were performed in 50 ml Nalgene Teflon FEP centrifuge tubes. Extractions at 70 °C were carried out in a water bath with occasional shaking; extractions at room temperature were continually shaken on a mechanical shaker. Samples were centrifuged between extractions and the supernatant liquor poured off. Between the acetic acid–ammonium acetate leach and the final leach, the sample was washed with 5 ml of deionised water.The Tamm’s reagent leachates were analysed for major elements, Cr and V by ICP-AES; 10 ml of the remaining solution were then removed for analysis by ICP-MS. Because of the low tolerance of the latter technique to high dissolved solids, it was necessary to digest the oxalate matrix with acid H2O2 prior to analysis. To the 10 ml of sample in a 15 ml polycarbonate test tube, 1 ml of H2O2 and 0.1 ml of concentrated HNO3 were added.The tube was capped loosely and left overnight in a water bath at 70 °C. This procedure was repeated three times, reducing the organic carbon content to less than one tenth of its original value. The final solution was diluted 1 + 1 with deionised water prior to analysis by ICPMS. The duplicate Tamm’s reagent leachate was acidified to 10% v/v with respect to concentrated HCl. The rare earth elements (REE) in this solution were determined by ICP-AES following ion exchange separation and preconcentration. Results and Discussion Development of Selective Extraction Procedure The objectives of this study were to develop a leaching procedure to selectively extract the iron oxide phases from the sandstones and to measure their trace metal content.The main criteria which determined method development were: (1) the selective leaching scheme should be specific for the iron oxide phase; (2) the trace elements to be determined in the rock leachate (U, Th, REEs, Ni, Sn and Pb) would be in the mg kg21 to mg kg21 range.This required a choice of extraction reagents which would minimise both sample contamination and potential interference in the chosen methods of analysis, in this case ICPAES and ICP-MS; and (3) the extraction scheme necessary to produce the required data for geochemical interpretation should also include an extraction to remove carbonate phases prior to an extraction to selectively remove the target iron oxide phases.This allowed the trace metal content of the iron oxides to be determined without contamination from trace metals in carbonates. A preliminary trial was designed with two main objectives: (i) to determine the best combination of acetic acid and ammonium acetate to selectively remove the carbonate phase without attacking the iron oxide or silicate phases; and (ii) to select the conditions for reaction and to compare the selectivity and efficiency of either HCl or oxalic acid/ammonium oxalate (Tamm’s reagent) for removing iron oxides.Initial leaching experiments were performed only on one core sample (SC1). The carbonate extraction experiments (Table 1) showed that the ammonium acetate–acetic acid (pH 5) and the 1 m acetic acid media were equally effective in removing calcium carbonate. The analyses of the two extractants are similar for all elements determined. The comparability of the Si and Fe values between these two reagents shows that the more aggressive acetic acid had not attacked the silica or iron oxide matrix more aggressively than the ammonium acetate solution.The 1 m acetic acid was therefore chosen for carbonate extraction because of its faster reaction rate (Table 1). Dissolution of iron oxide using Tamm’s reagent at room temperature was very slow, as shown by the marginal increase in Fe content of the leach solution over 46 h (Table 1). The red colouration of the sandstone appeared unchanged over the same period.The 1 m HCl at 70 °C showed much higher levels of Fe, which steadily increased over the period it was applied, although the rock still appeared red. High silicon values suggested that the silicate matrix was being attacked. On heating the Tamm’s reagent to 70 °C the efficiency of the iron oxide extraction was greatly enhanced, as indicated by the increase in Fe concentration combined with the marked decolouration of the sandstone material to pale grey with a thin red layer of finer clay material on top.The heated Tamm’s reagent had a higher efficiency of extraction than the heated HCl and a reduced tendency to attack the silicate matrix, as shown by the lower Si values. The Tamm’s reagent at 70 °C was therefore preferable to HCl as the iron oxide extraction reagent. The relative selectivity and effectiveness of Tamm’s reagent and 0.3 m oxalic acid removing the iron oxide phase were Table 1 Extraction media, conditions of extraction and mass (mg kg21) of major elements extracted during the preliminary extraction trial Conditions Extractant Ca Mg Na K Si Mn Fe 24 h @ 20 °C NH4OAc 13 289 84.94 19.57 50.98 18.92 36.60 7.52 (pH 5) 2 h @ 20 °C HOAc (1m) 13 115 88.96 17.53 36.50 23.51 37.35 10.88 27 h @ 20 °C Tamm’s 1 2.88 13.37 2.43 21.94 65.37 8.40 31.71 46 h @ 20 °C Tamm’s 2 2.33 16.35 3.18 32.97 101 8.03 35.55 46 h @ 20 °C + 24 h @ 70 °C Tamm’s 3 2.99 175 8.57 91.67 508 14.62 2445 27 h @ 70 °C HCl (0.1m) 1.448 362 16.05 219 1116 17.33 1016 46 h @ 70 °C HCl (0.1m) 1.391 452 17.73 399 1285 19.58 1619 70 h @ 70 °C HCl (0.1m) 1.298 561 22.65 718 1043 19.93 2171 502 Analyst, June 1997, Vol. 122compared in a second trial. This used 2 3 5 g splits of the second shallowest sample. Each split was extracted with 30 ml of 1 m acetic acid at 20 °C for 23 h followed by 2 35 ml washes of distilled water to remove carbonate, prior to the final iron oxide extraction. This used 30 ml of Tamm’s reagent for one split and 30 ml of 0.3 m oxalic acid for the other.The rates of dissolution of Fe, Mg, Si and Al were monitored by taking 1 ml aliquots of each solution over the 20 h extraction. These were diluted to 10 ml and analysed by ICP AES (Fig. 1). The 0.3 m oxalic acid was much more aggressive than Tamm’s reagent. The oxalic acid decolourised both the coarser sand material and the finer clay layer. The amount of iron extracted after 20 h was also greater and the rate of extraction faster.The oxalic acid was, however, probably too aggressive, as the increased rate of dissolution and final amounts of Si and Al leached indicate silicate matrix attack. Steady state for both the oxalate media was reached in about 16–20 hours. On the basis of the data from these preliminary trials the leaching procedure summarised in Table 2 was formulated. For each sample, 2 35 g splits of < 400 mm sandstone powder were weighed into 2 Nalgene 50 ml Teflon FEP centrifuge tubes.The second Tamm’s reagent leach was incorporated because, after the first leach, some of the samples were still red indicating that not all the iron oxide phase was being removed. This leachate was subsequently preserved and analysed as a separate leach solution. The first Tamm’s reagent leachate was used to determine the major elements, Cr and V by ICP-AES and Cr, Ni, Sn, Pb, Th and U by ICP-MS. The second split of the same sample was used for the ion exchange separation and concentration of the REEs which were subsequently analysed by ICP-AES. These methods are described in detail in Appendix 1.The results of the chemical analysis of the whole rock and the two Tamm’s reagent leaches are summarised in Tables 3, 4 and 5, respectively. Fig. 1 Extraction of major elements from sample SC1; (a) 0.3 m oxalic acid; (b) Tamm’s reagent. A, Fe, B, Mg; C, Si; D, Al. Table 3 Summary of chemical analysis data for the whole rock samples Element Units SC1 SC2 SC3 SC4 SC5 SC6 SC7 SC8 SC9 SC10 SC11 SC12 SiO2 % oxide 76.9 78.0 69.4 64.0 70.3 75.3 64.8 72.0 79.3 70.6 76.1 72.8 Al2O3 % oxide 7.05 5.24 8.98 7.76 12.21 5.19 9.90 8.56 8.06 8.63 6.46 5.47 Fe2O3 % oxide 0.95 0.93 2.61 4.29 1.39 0.82 3.02 2.15 1.78 3.02 2.28 1.99 MgO % oxide 0.309 0.213 0.525 0.502 0.796 0.768 0.879 1.95 0.518 2.27 1.59 2.27 CaO % oxide 1.60 3.91 4.64 6.97 1.36 5.48 7.20 2.45 0.21 2.34 2.32 3.60 Na2O % oxide 1.15 0.96 1.83 1.53 2.20 1.23 1.79 1.99 1.60 1.58 1.08 0.92 K2O % oxide 3.29 2.63 3.68 3.31 4.60 2.58 3.67 3.39 3.60 3.48 3.41 2.94 TiO2 % oxide 0.201 0.134 0.567 0.884 0.541 0.114 0.447 0.320 0.224 0.356 0.279 0.239 MnO % oxide 0.009 0.021 0.036 0.052 0.013 0.073 0.052 0.057 0.007 0.049 0.048 0.084 Ba mg kg21 469 413 410 323 516 268 373 403 442 373 324 279 Cr mg kg21 12.1 10.8 49.5 68.2 33.6 6.4 23.1 16.6 12.9 20.9 19.8 14.8 Ni mg kg21 5.49 3.50 8.82 8.48 14.41 2.75 18.46 7.63 12.01 16.18 6.70 5.37 Pb mg kg21 11.62 11.81 9.62 13.22 9.84 7.20 6.95 10.14 11.27 10.89 11.19 8.55 Sn mg kg21 1.45 1.20 2.41 2.47 1.99 0.89 1.49 1.26 1.34 2.20 2.50 1.79 Sr mg kg21 84.4 78.2 105.5 100.4 120.0 61.7 106.7 95.5 88.3 85.0 65.6 58.1 Th mg kg21 3.97 3.30 8.62 11.05 7.45 2.73 5.66 4.92 4.32 8.29 6.10 4.57 U mg kg21 1.05 1.00 2.41 3.35 2.29 0.64 1.59 1.54 1.19 2.10 1.40 1.19 V mg kg21 < 9.0 15.28 48.70 61.64 48.72 < 9.0 44.53 27.04 16.94 37.63 23.36 16.35 Zr mg kg21 77.3 49.4 195 303 134 29.0 100 93.9 62.0 91.1 79.3 60.9 Ce mg kg21 21.7 16.9 51.3 62.4 48.7 18.8 53.0 31.3 27.7 47.4 28.7 27.1 Dy mg kg21 4.06 3.87 6.15 10.36 6.40 4.06 6.77 6.54 5.51 6.41 5.44 4.27 Er mg kg21 0.70 0.81 1.78 3.57 1.39 0.63 1.43 0.87 0.69 1.09 0.73 0.98 Eu mg kg21 0.67 0.63 1.17 1.74 0.83 0.69 1.19 0.86 0.59 0.88 0.65 0.74 Gd mg kg21 < 0.94 < 0.94 1.03 3.50 < 0.97 < 0.93 < 0.98 < 0.95 < 0.93 < 0.94 < 0.97 < 1.0 Ho mg kg21 < 0.67 < 0.67 0.85 1.73 < 0.70 < 0.67 < 0.68 0.69 < 0.66 < 0.67 < 0.70 < 0.71 La mg kg21 11.68 9.54 26.09 31.47 26.84 11.81 28.57 15.81 13.92 23.73 14.01 15.34 Lu mg kg21 0.15 0.15 0.36 0.71 0.27 0.17 0.33 0.27 0.18 0.29 0.18 0.22 Nd mg kg21 10.36 8.77 23.16 33.37 19.78 10.85 24.02 17.71 12.87 21.13 13.56 13.71 Pr mg kg21 2.50 2.56 5.42 6.89 5.05 2.85 6.18 3.73 3.45 4.57 2.46 2.76 Sm mg kg21 2.34 2.41 5.32 8.04 4.26 2.79 4.95 3.98 2.79 4.79 2.63 3.27 Tb mg kg21 < 1.6 < 1.6 < 1.6 < 1.6 < 1.7 < 1.6 < 1.7 < 1.6 < 1.6 < 1.61 < 1.7 < 1.7 Tm mg kg21 < 0.67 < 0.67 < 0.69 1.00 0.72 0.75 < 0.70 < 0.68 < 0.66 < 0.67 < 0.7 < 0.71 Y mg kg21 10.6 9.7 23.9 45.3 19.2 9.9 21.4 14.2 9.9 16.4 12.1 14.6 Yb mg kg21 1.13 1.01 2.60 4.42 2.10 1.06 2.26 1.62 1.03 1.75 1.23 1.38 Table 2 Summary of leaching media and conditions used for the finalised leaching procedure Leach Leaching Time/ Temperature/ Leachate no. solution h °C preservation 1 1 mAcetic acid 20 20 None 2 Tamm’s reagent 20 70 1% HCl 3 Tamm’s reagent 20 70 1% HCl Analyst, June 1997, Vol. 122 503Table 4 Summary of chemical analysis data for the first Tamm’s reagent leach Element Units SC1 SC2 SC3 SC4 SC5 SC6 SC7 SC8 SC9 SC10 SC11 SC12 SiO2 % oxide 0.1 0.1 0.1 0.1 0.2 0.1 0.18 0.06 0.15 0.15 0.07 0.05 Al2O3 % oxide 0.05 0.04 0.05 0.05 0.14 0.04 0.17 0.03 0.15 0.15 0.04 0.03 Fe2O3 % oxide 0.25 0.30 0.66 1.02 0.19 0.26 1.30 0.49 0.84 0.87 0.82 0.72 MgO % oxide 0.039 0.031 0.047 0.048 0.134 0.153 0.2 0.5 0.2 0.2 0.1 0.4 CaO % oxide 0.0008 0.0007 0.0009 0.0009 0.0006 0.0008 0.0007 0.0006 0.0010 0.0009 0.0007 0.0009 Na2O % oxide 0.0007 0.0010 0.0028 0.0022 0.0034 0.0026 0.0035 0.0049 0.0038 0.0041 0.0048 0.0030 K2O % oxide 0.01 0.01 0.01 0.01 0.03 0.00 0.0154 0.0075 0.0106 0.0108 0.0060 0.0039 TiO2 % oxide 0.008 0.011 0.026 0.040 0.004 0.009 0.0483 0.0191 0.0325 0.0321 0.0290 0.0248 Cr mg kg21 4.860 5.160 14.328 22.055 4.307 4.500 18.08 8.42 9.93 10.90 10.85 8.20 Ni mg kg21 0.862 0.530 1.012 1.276 2.444 1.074 3.67 0.97 2.89 3.72 1.42 1.14 Pb mg kg21 0.4 0.5 0.6 0.9 0.3 0.2 1.119 0.462 0.680 0.766 1.128 0.661 Sn mg kg21 0.06 0.07 0.18 0.24 0.04 0.13 0.14 0.11 0.15 0.17 0.37 0.31 Th mg kg21 0.47 0.46 1.21 1.68 0.95 0.70 1.39 1.11 1.07 1.14 2.06 1.38 U mg kg21 0.1 0.1 0.3 0.4 0.2 0.1 0.36 0.24 0.22 0.26 0.32 0.24 V mg kg21 5.02 8.50 17.33 22.43 4.75 7.15 21.06 8.67 16.94 16.71 9.85 8.48 Ce mg kg21 0.13 0.14 0.51 0.49 1.02 1.23 0.83 1.12 2.00 2.02 0.26 0.10 Dy mg kg21 0.04 0.06 0.08 0.09 0.12 0.07 0.153 0.078 0.213 0.237 0.116 0.059 Er mg kg21 0.014 0.019 0.016 0.025 0.038 < 0.007 0.035 0.013 0.052 < 0.007 0.032 < 0.007 Eu mg kg21 0.010 0.015 0.022 0.023 0.032 0.047 0.038 0.054 0.092 0.094 0.032 0.013 Gd mg kg21 < 0.016 < 0.016 < 0.016 < 0.016 0.02 0.04 0.071 0.059 0.201 0.223 0.037 < 0.016 Ho mg kg21 < 0.012 0.01 0.01 0.01 0.02 0.01 0.015 < 0.012 0.021 0.030 0.018 < 0.012 La mg kg21 0.06 0.08 0.20 0.19 0.36 0.38 0.348 0.293 0.662 0.641 0.118 0.049 Lu mg kg21 0.0031 0.0040 0.0060 0.0064 0.0073 0.0055 0.009 0.008 0.010 0.018 0.007 0.009 Nd mg kg21 0.12 0.17 0.42 0.36 0.69 1.09 0.568 1.034 1.687 1.724 0.213 0.089 Pr mg kg21 0.03 0.05 0.09 0.08 0.14 0.22 0.117 0.203 0.334 0.395 0.050 0.116 Sm mg kg21 0.03 0.05 0.09 0.09 0.16 0.24 0.157 0.263 0.423 0.442 0.117 0.045 Tb mg kg21 < 0.028 < 0.028 < 0.028 < 0.028 < 0.028 < 0.028 < 0.028 < 0.029 < 0.029 < 0.029 < 0.028 < 0.028 Tm mg kg21 < 0.012 < 0.012 < 0.012 < 0.012 < 0.012 < 0.012 0.014 < 0.012 < 0.012 < 0.012 < 0.012 0.013 Y mg kg21 0.17 0.28 0.32 0.37 0.49 0.30 0.539 0.356 0.856 0.938 0.419 0.283 Yb mg kg21 0.0170 0.0232 0.0318 0.0397 0.0529 0.0282 0.050 0.042 0.061 0.071 0.040 0.029 Table 5 Summary of chemical analysis data for the second Tamm’s reagent leach Element Units SC1 SC2 SC3 SC4 SC5 SC6 SC7 SC8 SC9 SC10 SC11 SC12 SiO2 % oxide 0.055 0.045 0.036 0.037 0.064 0.027 0.11 0.05 0.10 0.10 0.03 0.02 Al2O3 % oxide 0.036 0.027 0.022 0.020 0.053 0.017 0.10 0.04 0.09 0.10 0.02 0.01 Fe2O3 % oxide 0.11 0.10 0.32 0.51 0.06 0.04 0.317 0.421 0.179 0.193 0.272 0.209 MgO % oxide 0.029 0.024 0.023 0.022 0.051 0.023 0.128 0.210 0.129 0.133 0.016 0.091 CaO % oxide 0.0006 0.0006 0.0007 0.0006 0.0008 0.0008 0.0008 0.0007 0.0008 0.0008 0.0006 0.0006 Na2O % oxide 0.0002 0.0002 0.0004 0.0003 0.0004 0.0004 0.0007 0.0008 0.0005 0.0005 0.0004 0.0004 K2O % oxide 0.0028 0.0028 0.0031 0.0021 0.0083 0.0009 0.0081 0.0045 0.0049 0.0051 0.0014 0.0009 Cr mg kg21 1.695 1.404 6.229 9.186 1.053 < 1.5 4.17 5.92 2.49 2.42 2.96 1.97 Ni mg kg21 0.420 0.336 0.618 0.530 1.008 0.262 2.24 0.881 1.36 1.66 0.633 0.575 Pb mg kg21 0.196 0.149 0.266 0.536 0.191 0.034 0.31 0.23 0.11 0.12 0.37 0.22 Sn mg kg21 0.04 0.04 0.12 0.18 0.01 0.02 0.045 0.117 0.040 0.046 0.112 0.091 Th mg kg21 0.13 0.10 0.29 0.38 0.14 0.10 0.234 0.517 0.266 0.284 0.210 0.161 U mg kg21 0.021 0.019 0.075 0.106 0.031 0.011 0.04 0.07 0.02 0.03 0.04 0.03 V mg kg21 2.1 2.7 9.6 12.6 1.3 0.8 7.89 12.76 3.04 3.23 3.67 2.77 Ce mg kg21 0.256 0.203 0.316 0.330 0.547 0.400 0.79 0.97 1.14 1.28 0.11 0.06 Dy mg kg21 0.023 0.022 0.027 0.030 0.032 0.016 0.06 0.04 0.05 0.05 0.03 0.01 Er mg kg21 < 0.008 < 0.008 < 0.008 < 0.008 0.01 < 0.007 < 0.009 < 0.008 < 0.011 < 0.007 < 0.009 < 0.007 Eu mg kg21 0.009 0.010 0.010 0.011 0.012 0.011 0.021 0.036 0.036 0.039 0.008 0.003 Gd mg kg21 < 0.016 < 0.016 < 0.015 < 0.015 < 0.018 < 0.018 < 0.020 0.016 0.032 0.037 < 0.019 < 0.016 Ho mg kg21 < 0.011 < 0.012 < 0.012 < 0.012 < 0.013 < 0.012 < 0.013 < 0.011 < 0.014 < 0.014 < 0.013 < 0.011 La mg kg21 0.113 0.108 0.131 0.134 0.198 0.110 0.288 0.250 0.348 0.393 0.053 0.019 Lu mg kg21 0.0015 0.0019 0.0013 0.0034 0.0011 < 0.0014 0.006 0.007 0.003 0.002 0.001 0.001 Nd mg kg21 0.1493 0.1505 0.2255 0.2340 0.3286 0.2886 0.538 0.882 0.983 1.068 0.069 0.025 Pr mg kg21 0.023 0.024 0.052 0.049 0.088 0.063 0.141 0.201 0.209 0.209 0.033 0.027 Sm mg kg21 0.044 0.040 0.050 0.055 0.066 0.057 0.122 0.215 0.211 0.220 0.030 < 0.014 Tb mg kg21 < 0.027 < 0.027 < 0.026 < 0.026 < 0.026 < 0.027 < 0.027 < 0.027 < 0.027 < 0.027 < 0.027 < 0.027 Tm mg kg21 < 0.011 < 0.011 < 0.011 < 0.011 < 0.011 < 0.011 < 0.012 < 0.011 < 0.011 < 0.011 < 0.011 < 0.012 Y mg kg21 0.102 0.111 0.100 0.117 0.102 0.062 0.184 0.141 0.151 0.149 0.081 0.056 Yb mg kg21 0.008 0.008 0.009 0.013 0.008 0.004 0.015 0.014 0.011 0.009 0.006 0.004 504 Analyst, June 1997, Vol. 122Principal Component Analysis of the Whole Rock and Leached Data Principal component analysis10 (PCA) was applied to the multielement data matrix of the whole rock analyses and the combined data from the two leaches of the sandstones samples. The analysis of the whole rock samples and the oxalate leachates yielded two data matrices of element concentration versus sample.The rock samples can be assumed to be made up of varying proportions of similar mineral groups depending on the location of the sample within the borehole. Each of the elemental analyses for the twelve samples therefore reflects different proportions of these mineral components.Similarly, the oxalate leaching medium can be assumed to have leached out different amounts of hematite (and possibly other minerals) from each of the slightly different samples. The composition of the components of the mixture or the amount of each component present cannot be calculated from the elemental composition of a single rock sample or leachate solution. If, however, the data for all of the whole rock analyses or the leachate analyses are examined by multivariate statistical methods it is possible, in theory, to estimate the number of components responsible for the two data matrices, the composition of those components and the different proportions of each component in each sample.In the first stage of the data reduction, PCA identifies a vector or abstract variable which has the highest correlation coefficient with all the chemical elements in the matrix and, hence, accounts for the greatest proportion of the total variance.This vector is called the first principal component (PC1). A second principal component (PC2) is constructed from the residual matrix, orthogonal (i.e., uncorrelated) to PC1 and is calculated so as to account for the maximum proportion of the remaining variance. Further PCs up to the original number of samples (in this case 12 for the whole rock and 24 for the leachate data) can be calculated and will account for the total variance in the data set. The first PCs describe underlying properties of the data, in this case the geochemical components of the whole rock or leachate system, and the remaining PCs are associated with additional sources of variation attributable to sampling, the sample preparation and analysis.The data were pre-treated by subtracting the mean of each matrix column from the values in each column (a procedure known as data centering) and scaled by dividing each of the resulting values by the standard deviation for that particular column.This procedure gives all variables an equal weighting so that the major elements do not dominate and hide trends in the trace element data. The PCA deconvolution of the data was carried out on the scaled matrices using the NIPALS algorithm described by Wold et al.10 programmed in Microsoft Visual Basic and using Statsoft Statistica (V5.0). Stage two of the data reduction involves determining how many true geochemical components there are in the system by applying statistical and empirical tests on the PCs.For this study three methods were used to identify the number of true components. For each PC its associated eigenvalue describes its variance and hence its relative importance. By taking the ratio of successive eigenvalues, a number of workers11–13 have shown that the number of significant components is indicated by the maximum value of the ratios, if there are more than one local maxima the second local maximum is chosen in order of decreasing PC number.Malinowski14 suggested an empirical method where a minimum value of a function, calculated from the eigenvalues and the number of rows and columns in the data matrix indicates the number of true components. The results of these two methods for the whole rock and leachate data are summarised in Table 6. In the whole rock data both methods indicate there are four components. The situation in the leachate data is not quite so clear as the eigenvalue ratio suggests four components and the indicator function 11. To try and clarify the position for the leachate data set a third test, as suggested by Hopke,15 was carried out on the leachate data set.In this method the PCs are varimax rotated15 and the number of components is indicated by the number of eigenvalues, calculated from the rotated PCs, greater than one. Table 6 shows that there are four components with eigenvalues greater than one and the fifth eigenvalue close to 1 suggesting that four to five true geochemical components exist in the data.The varimax rotation option was not available on the Visual Basic programme and had to be carried out using the Statistica software. This varimax rotation test could not be performed on the whole rock data set as the Statistica programme requires more samples than variables to carry out the required calculations. Once the number of components has been established, the next step involves placing physical significance on the abstract factors (PCs) that have been shown to be true factors.At this stage the PCs are not directly representative of the true components but are linearly related. The ultimate solution to the deconvolution involves discovering the mathematical rotations necessary to convert the abstract PC vectors into true factors. This is discussed in the section on quantitative interpretation. Table 6 Summary of test parameters used to determine the number of true components in the whole rock and leachate data sets Whole rock data Combined leach data Varimax Number Indicator Number Indicator rotated of factors Ratio function of factors Ratio function eigenvalue 1 4.08 0.00494 1 3.01 0.001201 7.62 2 1.81 0.00469 2 2.85 0.000957 7.79 3 1.49 0.00458 3 1.41 0.000874 3.52 4 3.05 0.00426 4 1.92 0.000788 1.25 5 1.11 0.00477 5 1.18 0.000751 0.96 6 2.84 0.00496 6 1.36 0.000689 0.71 7 1.36 0.00630 7 1.74 0.000611 0.08 8 1.38 0.00854 8 1.45 0.000550 0.39 9 1.39 0.01262 9 3.02 0.000476 0.29 10 2.01 0.02003 10 1.10 0.000480 0.20 11 1.75 0.000463 0.05 12 1.29 0.000469 0.02 13 1.68 0.000467 0.07 14 1.43 0.000486 0.01 Analyst, June 1997, Vol. 122 505The initial approach to giving physical significance to the abstract PCs is a qualitative method. Qualitative Interpretation For each PC there is a score vector which is a representation of the original composition in terms of the new, abstract, PC variable and is linearly related to the true geochemical components in the rock samples being studied. The weights vector associated with each PC represents a correlation between each element measured and the abstract PC component.By examining the element patterns in the weight vectors of significant principal components, in conjunction with some basic geochemical and mineralogical knowledge of the materials under study, the PCA data can reveal qualitative information regarding the geochemical phases present and the distribution of trace metals amongst them. For the whole rock data a cluster plot, produced with the Statistica software, of the four significant weights vectors using euclidean distance and Ward’s clustering algorithm16 is shown in Fig. 2. By cutting the distance axis at ca. 44% four clusters are formed representing the four geochemical components in the whole rock data. These can be tentatively assigned to geochemical/ mineralogical sources. The major elements Na, Al and K are associated with Sr, Ba, and Ni and probably represent a feldspar grouping.Titanium, Ca, REEs, V, Cr and Zr form a group which probably represents resistate minerals. Magnesium groups with Mn possibly represents dolomite which is known to be present in the deeper samples.17 Iron, Si, Sn, U, Th and Pb probably represent a combination of hematite, clay minerals and Fe-oxy-hydroxides. As there are only 12 samples the PCA cannot be expected to identify all geochemical components but it is clear that the actinide elements are associated with a different geochemical phase from the lanthanides.The interpretation of the leachate data has been clarified by the use of varimax rotation on the PCA weights. This rotation tends to drive variable weights toward either zero or one on a given component. It precludes the emergence of a strong general component by making the variance explained by the individual components more equal, resulting in orthogonal components which are often more readily identifiable as specific source components.18 Table 7 shows the varimax rotated weights and clearly indicates that component 1 is REE dominated with some association with Si, Al and Ni probably related to clay materials.Component 2 is associated with Fe, Ti, Cr, Pb, Sn, Th, U and V probably derived from hematite dissolution. Component 3 is Si, Al, K and to some extent Ni dominated and is possibly related to feldspar dissolution. Component 4 is Mg dominated and probably derives from dolomite dissolution.Finally component 5, which may not be significant, is associated with Ca, possibly from a residual calcite source. The qualitative interpretation of the data clearly shows that more than one geochemical component is being extracted by the Tamm’s reagent and that the trace element distributions within these components are quite different. Quantitative Interpretation Theory Having identified that the procedure for extracting the iron oxide phase has extracted more than one physico-chemical component the next stage is to use the abstract components identified by PCA to quantify the composition of these components and how much of each component is in each sample.To simplify the deconvolution step it was decided to split the extraction data into two data sets, the first and second leach data, and deal with quantification of each separately. Although the mean centering and scaling of the data as previously described enables a clear qualitative interpretation, quantitative deconvolution of the PCA scores and weights obtained from data scaled in this manner can be complicated.A simplified scaling procedure was applied to each data set. Each determinand was scaled to a percentage of the highest value of that determinand found in the original rock sample, thereby giving a greater weighting to determinands which are preferentially extracted by the Tamm’s reagent.Each data set was subjected to PCA deconvolution as described for the qualitative data reduction. To carry out the quantification step from the abstract components identified by the PCA a modified procedure similar to that described by Malinowski19 was used. In chromatographic procedures with two way data (e.g, time profile/UV spectra) where there is poor resolution of components PCA related techniques have been used to resolve the components. This has been carried out by obtaining estimates of the chromatographic profiles by analysis of eigenvalues of the data set (evolving factor analysis, EFA) and refining these chromatographic profiles using the pseudoinverse calculation and the original raw data set to produce an estimate of the component concentrations.By forward and Fig. 2 Cluster plot of weights for the combined Tamm’s reagent leachate data set. Table 7 Varimax rotated weights for the combined leachate data set Factor Factor Factor Factor Factor 1 2 3 4 5 SiO2 0.46 0.24 0.83* 20.02 0.08 Al2O3 0.56 0.17 0.77 20.09 0.10 Fe2O3 0.21 0.94 0.19 0.09 0.07 MgO 0.35 0.17 20.01 0.79 20.04 CaO 0.36 0.31 0.07 20.01 0.87 Na2O 0.34 0.57 0.32 0.64 0.00 K2O 0.18 0.16 0.90 0.16 20.06 TiO2 0.23 0.95 0.10 20.02 0.02 Cr 0.08 0.95 0.15 20.02 0.08 Ni 0.57 0.36 0.67 0.02 0.09 Pb 20.01 0.92 0.26 0.17 0.04 Sn 20.08 0.80 20.20 0.39 0.18 Th 0.07 0.81 0.23 0.47 0.16 U 0.09 0.89 0.26 0.28 0.10 V 0.29 0.90 0.12 20.14 0.03 Ce 0.95 0.03 0.26 0.07 0.07 Dy 0.63 0.50 0.49 0.15 0.11 Eu 0.91 0.25 0.21 0.19 0.05 La 0.91 0.10 0.37 0.00 0.08 Nd 0.97 0.03 0.16 0.10 0.07 Pr 0.95 0.06 0.14 0.18 0.17 Sm 0.95 0.14 0.17 0.17 0.05 Y 0.62 0.51 0.45 0.20 0.09 * Numbers in bold are > 0.7. 506 Analyst, June 1997, Vol. 122backward iterations, each time eliminating values outside boundary conditions (e.g., negative concentrations), true chromatographic profiles and component concentrations can be calculated. In this study EFA is not appropriate to estimate the component depth profiles as the geochemical variations are not well behaved gaussian peaks as in chromatography.The first estimate of the geochemical component depth profiles was obtained by looking at the correlation of the abstract component depth profiles, derived from PCA, with the scaled profiles for each determinand, using the determinand with the highest correlation coefficient with the abstract profile as the first estimate. The determinand profiles used were Ti, Mg, Al and Sn for the first leachate data set and V, Al, Mg and K for the second leachate data set.If the leachate data are represented by a mixture matrix in which the chemical constituents are represented by the rows and the samples by columns (matrix M). Then the mixture matrix can be considered to be the product of a column matrix C in which each column represents the composition one of the chemical components in the mixture and a row matrix P in which each row represents the proportions of each component in the mixture.In matrix notation the relationship between the three matrices is shown in eqn. (1). M = CP (1) Given that M is known and that an initial guess at P can be made from the element profiles indicated in Tables 4 and 5 then first estimate of the concentration matrix (C) is given by eqn. (2) C = MP21 (2) where P21 represents the matrix inverse of P. This solution is only feasible if P is a square matrix, which is not the case for the leach data.This can be solved, however, using a least squares solution or ‘pseudoinverse’ method shown in eqn. (3) C = MPA[PAP]21 (3) where PA is the transpose of P. This allows the direct calculation of the concentration matrix. Since the proportion matrix is only a first approximation it is likely that calculation of the concentration matrix calculated from these values will give non-realistic values for some samples (e.g., concentrations < 0). In order to arrive at a meaningful solution it is necessary to use an iterative approach similar to that described by Gamp et al.20 In this method a first approximation of C is calculated from eqn.(3) using an estimation for matrix P. Values in C (which through scaling is expressed as a percentage) less than 0 are corrected to 0 and values greater than 100 are corrected to 100. Using the modified C matrix the pseudoinverse expression [shown in eqn. (4)] is used to calculate a new P matrix.P = [CAC]21 CAM (4) Negative concentrations in P are corrected to 0 and values greater than 1 corrected to 1 and the iterative procedure is started again producing the next approximation for C. Iterative modifications of P and C are continued until meaningful values for both matrices are produced. This procedure was carried out for both leach data matrices using 4 and 5 components as indicated in Table 6. In practice it was found that physically meaningful components could only be obtained using the 4 components for both leach data sets.This suggests that Hopke’s test15 (see Table 6) for defining the number of components in the leachate data set as 4 was correct. During the forward and backward iteration process it was also found that the speed of convergence was improved if a maximum boundary for the concentration matrix of 30% was imposed for the first few iterations which was relaxed to 100% for the final iterations. As a final refinement, the elemental composition matrix from the final iteration was subjected to iterative target factor analysis (50 iterations) as described by Malinowski19 to ensure that the data fitted the PCA model.These data were used to calculate the final component profile matrix using the pseudoinverse, as previously described. The final concentration and profile matrices, corrected for detection limits, for both leachate data sets are shown in Tables 8 and 9. Interpretation of the Quantitative data Assuming that the major oxides make up the majority of each component, Table 10 shows the compositions expressed as percentage major oxide.Two components from each leach Table 8 Resolved components from the Leach 1 data set Com- Com- Com- Component 1 ponent 2 ponent 3 ponent 4 SiO2 Leach 1 0.043 0.012 0.293 0.168 Leach 2 0.000 0.480 0.180 0.106 Al2O3 Leach 1 0.016 0.000 0.298 0.123 Leach 2 0.000 0.288 0.159 0.103 Fe2O3 Leach 1 0.962 0.152 0.661 1.003 Leach 2 0.397 2.312 0.605 0.094 MgO Leach 1 0.021 0.694 0.356 0.074 Leach 2 0.024 0.195 0.622 0.004 CaO Leach 1 0.001 0.001 0.001 0.001 Leach 2 0.000 0.010 0.003 0.001 Na2O Leach 1 0.002 0.004 0.006 0.007 Leach 2 0.000 0.003 0.002 0.000 K2O Leach 1 0.004 0.003 0.027 0.020 Leach 2 0.003 0.006 0.005 0.008 TiO2 Leach 1 0.039 0.005 0.022 0.028 Leach 2 0.026 0.071 0.017 0.005 Cr Leach 1 20.701 0.000 0.830 3.932 Leach 2 9.023 23.234 3.692 1.273 Ni Leach 1 0.555 0.314 6.277 4.296 Leach 2 0.128 5.868 2.176 2.001 Pb Leach 1 0.856 0.012 0.523 1.794 Leach 2 0.197 3.918 0.217 0.264 Sn Leach 1 0.215 0.095 0.000 0.634 Leach 2 0.082 1.282 0.231 0.000 Th Leach 1 1.417 0.610 0.690 3.190 Leach 2 0.289 1.936 1.136 0.013 U Leach 1 0.360 0.098 0.123 0.341 Leach 2 0.095 0.347 0.058 0.017 V Leach 1 21.767 0.282 10.283 0.747 Leach 2 18.074 1.454 8.418 0.009 Ce Leach 1 0.000 1.043 3.579 0.199 Leach 2 0.040 2.225 3.238 0.554 Dy Leach 1 0.046 0.030 0.354 0.250 Leach 2 0.006 0.289 0.110 0.051 Er Leach 1 0.019 0.000 0.041 0.058 Leach 2 0.000 0.063 0.015 0.004 Eu Leach 1 0.006 0.046 0.150 0.040 Leach 2 0.000 0.104 0.122 0.010 La Leach 1 0.045 0.234 1.146 0.193 Leach 2 0.031 0.936 0.794 0.257 Lu Leach 1 0.004 0.008 0.021 0.014 Leach 2 0.006 0.000 0.009 0.001 Nd Leach 1 0.000 1.017 2.935 0.000 Leach 2 0.007 1.514 3.059 0.272 Pr Leach 1 0.000 0.252 0.608 0.058 Leach 2 0.009 0.363 0.649 0.069 Sm Leach 1 0.001 0.243 0.726 0.131 Leach 2 0.018 0.331 0.692 0.045 Y Leach 1 0.190 0.218 1.381 0.857 Leach 2 0.056 0.926 0.336 0.147 Yb Leach 1 0.023 0.028 0.103 0.073 Leach 2 0.012 0.045 0.022 0.009 Analyst, June 1997, Vol. 122 507(components 1 and 4 from leach 1 and components 1 and 2 from leach 2) are predominantly made up of iron oxide. Component 1 from leach 1 and 2 and components 4 and 2 from leach 1 and 2 data have very similar composition and it is assumed to represent the same mineral host in each leach. Examination of the profiles of these four components (Table 9) shows that component 1 from both leach 1 and 2 show a similar trend with the first leach having a minimum at sample 5.In a description of the mineralogy of these core samples, Kemp and Pearce17 describes sample 5 as being anomalous (drab grey sandstone) lacking in the characteristic red colouration of the other samples and confirms, by XRD, that it is depleted in hematite compared to the other samples. It can therefore be concluded that component 1 probably represents hematite in both the first and second leaches.The other iron containing host represented by components 4 and 2 in leaches 1 and 2 has lower total iron oxide content and higher Si and Al content than the hematite component and makes up a smaller proportion of the total iron leached. This is probably amorphous iron oxide or iron oxyhydroxides which are known to exist in these samples. 16,20 Component 2 and component 3 in leach 1 and 2 are dominated by magnesium oxide and appears predominantly in the deeper samples (samples 6–12). Kemp and Pearce17 show that dolomite (Mg/Ca carbonate) is present in these deeper samples.The fact that the Ca content of these components is low can be explained by the oxalic acid in the Tamm’s reagent causing precipitation of insoluble Ca oxalate. It is therefore assumed that dolomite is the mineral source of these components. Component 3 and component 4 of leach 1 and 2 appear to be iron aluminosilicates. M�ossbauer spectroscopy of whole rock and clay fractions21 of these samples shows iron is present as hematite and as iron alumino-silicate. Kemp and Pearce17 suggest that chlorite is present in these samples and the composition of these components is consistent with reported analysis of chlorite samples.22 Having estimated the compositions of the components in the leach solutions and the proportions of each component in each sample and made a tentative identification of their mineral source, the purpose of this study has been achieved, namely to find out the chemical composition of the iron oxide phases of the sandstone formation.The profiles for the two leaches show similar patterns (Table 9) and therefore for clarity only leach 1 data has been plotted. Fig. 3 shows example profiles for Fe, U, Ce and Ni derived from: (i) the total concentrations found in leach 1; (ii) the concentration found in the hematite component of leach 1; (iii) the concentration found in the Fe oxy-hydroxide component of leach 1; and (iv) the sum of the concentrations found in both the hematite and Fe oxy-hydroxide components of leach 1.The profiles are plotted against sample number (which are in depth order) rather than actual depth as this shows up the patterns of the profiles more clearly. The Fe profile shows that the total Fe leached from the shallow samples (1–6) is derived almost entirely from iron oxides; however, for the deeper samples a significant proportion of the total iron leached is derived from non-iron oxide sources. The majority of the iron is found in the hematite component.The actinides as shown by U behave similarly to iron. In contrast, total Ce leached, which is representative of the rare earth elements, is almost exclusively derived from non-iron oxide sources. Nickel behaves in a manner intermediate between the actinides and the rare earth elements, again the majority of the Ni in total leached data is derived from non-iron oxide sources.These profiles illustrate the problem, noted by other investigators, of the poor specificity of the selective extractant. Clearly, if the unprocessed data alone had been used to make an Table 9 Proportions of resolved components from the leach 1 and leach 2 data sets Sample Component Component Component Component No. 1 2 3 4 Leach 1— 1 0.198 0.00 0.077 0.038 2 0.305 0.00 0.085 0.00 3 0.711 0.004 0.101 0.001 4 1.01 0.00 0.077 0.041 5 0.103 0.041 0.260 0.093 6 0.218 0.154 0.153 0.018 7 0.820 0.00 0.435 0.051 8 0.371 0.639 0.088 0.006 9 0.525 0.026 0.461 0.00 10 0.505 0.00 0.537 0.040 11 0.465 0.176 0.00 0.401 12 0.385 0.596 0.00 0.223 Leach 2— 1 0.098 0.022 0.034 0.111 2 0.129 0.014 0.032 0.082 3 0.520 0.051 0.011 0.107 4 0.704 0.107 0.00 0.00 5 0.036 0.00 0.078 0.439 6 0.022 0.004 0.049 0.071 7 0.353 0.00 0.170 0.947 8 0.556 0.00 0.311 0.061 9 0.076 0.00 0.220 0.484 10 0.077 0.00 0.229 0.616 11 0.188 0.075 0.00 0.077 12 0.103 0.044 0.089 0.025 Table 10 Major oxide percentage composition of resolved components in leach 1 and 2 and their probable mineral source Fe-Alumino Fe-Alumino silicate Fe-Oxy- Fe-oxy- silicate Hematite Dolomite (chlorite) hydroxides Hematite hydroxides Dolomite (chlorite) Probable Leach 1 Leach 2 mineral source Component 1 Component 2 Component 3 Component 4 Component 1 Component 2 Component 3 Component 4 SiO2 3.9 1.4 17.6 11.8 0.0 14.3 11.3 33.1 Al2O3 1.4 0.0 17.9 8.6 0.0 8.6 10.0 32.1 Fe2O3 88.5 17.4 39.7 70.5 88.0 68.7 38.0 29.2 MgO 2.0 79.8 21.4 5.2 5.3 5.8 39.1 1.3 Ca 0.1 0.1 0.1 0.1 0.0 0.3 0.2 0.3 Na2O 0.2 0.5 0.4 0.5 0.1 0.1 0.1 0.1 K2O 0.4 0.3 1.6 1.4 0.7 0.2 0.3 2.5 TiO2 3.5 0.6 1.3 1.9 5.8 2.1 1.0 1.4 508 Analyst, June 1997, Vol. 122interpretation of the geochemistry of the iron oxide phases, the conclusions would have been erroneous. Another reported problem with the sequential leach methodology is the re-absorption of trace metals onto the undissolved solid phase during the leaching process.To investigate this, the major and trace metals in the hematite and iron oxy-hydroxy phases from both the first and second leach were calculated as a percentage of the total amount of each element from the sum of the two leaches. Subsequently, to take into account the different amounts of each component in leach 1 and 2 the percentage compositions were ratioed to the iron content of each respective leachate.The results of this are shown in the bar charts Fig. 4. Each bar represents the ratio of the percentages (relative to iron) in leach 2 compared with lea 1. For iron the scaled values give an equal distribution of iron in leach 1 and 2. For those elements with ratios > 1 there is a relative enrichment in leach 2 compared with leach 1. Ratios < 1 indicate a relative depletion in leach 2 compared with leach 1. If re-absorption was Fig. 3 Concentration profiles for Fe, U, Ce, and Ni in the resolved components of the first Tamm’s reagent leach.Sample number in increasing depth order. 5, Total extracted by Tamm’s reagent; ~, total associated with Fe oxide sources; +, total associated with hematite; total associated with Fe-oxyhydroxides. Fig. 4 Ratio of the proportions of each element (relative to Fe) in the first and second Tamm’s reagent leach. Analyst, June 1997, Vol. 122 509the dominating factor then enrichment in leach 2 would be expected.Fig. 4 shows that for some elements there is evidence for re-absorption in both iron oxide phases (e.g., Ce), however, some elements show a relative enrichment for leach 1 (e.g., Th) and for many of the trace metals the relative proportions in each leach is very similar. In summary there is no clear evidence for systematic re-absorption during the leaching procedure used in this study. Conclusions Chemometric processing of whole rock and leachate data of related rock samples gives ‘added value’ to the bare analytical results.Analysis of multi-element data obtained from whole rock samples from the same formation, using chemometric methods, provides macroscale information on trace metal associations in the rock which is complimentary to microanalysis methods. Factor analysis of multi-element data obtained from extraction of selected mineral phases can be used to show both qualitative and quantitative information about the chemical composition of selected mineral phases.The problem of nonselectivity of the leaching reagent can be minimised by chemometric processing of the data. The use of chemometric processing of leaching data, produced in a similar manner to that described in this work, could provide an important tool for measuring the trace element distributions in rocks soils and sediments for geochemical and environmental applications. For example a non-specific reagent with the ability to hold trace elements in solution (e.g., a mineral acid with a pH < 1) could be used as the extractant, separating the different phases by their rate of reaction with this media by time based sampling.The overlap between the dissolution of different phases could then be resolved by factor analysis of the time series data. The authors would like to thank UK Nirex for funding this study, the NERC ICP-MS facility at Royal Holloway University of London for the use of the ICP-MS and Dr. M. Thompson, Dr.A. H. Bath, Mr. D. L. Miles, Dr. Richard Metcalfe and Mr. S. Reeder for helpful comments and suggestions on the text of this paper. This paper is published with the approval of the Director, British Geological Survey (NERC). Appendix 1 Sample Preparation and Analytical Methods Instrumentation The ICP spectrometer used in this work was a Perkin-Elmer Plasma II sequential scanning system with twin 1 m vacuum monochromators: monochromator A, with a 3600 line/mm grating and wavelength range of 160–400 nm, and monochromator B, with an 1800 line/mm grating and wavelength range of 160–800 nm.The monochromator gratings, plasma power, plasma gas flows, plasma viewing height, 50-position autosampler, and nebuliser peristaltic pump were all under computer control. The analytical emission lines and plasma operating conditions used in this work are summarised in Table 11. The ICP-MS instrumentation used was a VG PlasmaQuad 2 operated in scanning mode over the mass range 50–245 u.The isotopes used for analysis and the system operating conditions are given in Table 12. Reagents The chemical reagents used were: glacial acetic acid, (Merck, Poole, Dorset, UK; Aristar grade), ammonium acetate (Aldrich Gold label), ammonium oxalate monohydrate (Aldrich, Gillingham, Dorset, UK), 35% w/w hydrochloric acid (Merck; Aristar grade), 40% m/m hydrofluoric acid (Merck; Aristar grade), 30% m/m hydrogen peroxide (Merck; Aristar grade), anhydrous lithium metaborate (Merck; Spectrosol), 4 + 1 lithium metaborate –lithium tetraborate mixture [Spectroflux (R) 100B Johnson Matthey, Royston, Herts, UK], 70% m/v nitric acid (Merck; Aristar grade), oxalic acid (Merck; Aristar grade) and 70% m/m perchloric acid (Merck; Aristar grade).Table 11 ICP-AES operating conditions Mono- Plasma Element Wavelength Viewing chromator source Al 396.152 15 B stdcond Ba 455.403 15 B stdcond Ca 315.887 15 B stdcond Cr 205.552 8 B OPT Fe 259.940 15 A stdcond K 766.490 9 B stdcond Mg 279.079 15 A stdcond Mn 257.610 15 A stdcond Na 589.592 15 B stdcond Ni 231.604 11 B OPT S 180.731 8 A OPT Si 251.611 15 B stdcond Sr 407.771 15 B stdcond Ti 334.941 15 A stdcond Zr 343.823 15 B stdcond V 292.402 15 A stdcond Ce 418.660 11 B Ree Dv 353.170 11 A Ree Er 390.631 11 B Ree Eu 381.962 11 B Ree Gd 335.047 11 A Ree Ho 345.6 13 A Ree La 333.749 11 A Ree Lu 261.537 12 A Ree Nd 430.357 11 B Ree Pr 422.535 11 B Ree *Sc 424.683 n/a† n/a n/a Sm 359.267 11 B Ree Tb 350.904 11 A Ree Tm 313.118 11 A Ree Y 371.026 11 B Ree Yb 328.924 9 A Ree Plasma source file conditions— Nebuliser Plasma flow/ flow/ Auxiliary/ Nebuliser/ Source Power/W l min21 l min21 l min21 l min21 stdcond 1000 1.00 15.0 1.0 1.0 OPT 1410 0.88 15.0 1.0 1.0 Ree 1200 1.05 15.0 1.0 1.0 * Myers–Tracy signal compensation used.† n/a, Not applicable. Table 12 ICP-MS Operating Conditions Plasma power 1250 W Nebuliser gas flow 0.8 l min21 Ar Plasma gas flow 13.75 l min21 Ar Auxiliary gas flow 0.41 l min21 Nebuliser pump rate 0.9 ml min21 Distance of load coil to sampling cone 10 mm Sampling Cone aperture 0.7 mm Skimmer Cone aperture 1.0 mm Element Mass/u Cr 52 Ni 60 Sn 118 Pb 208 Th 232 U 238 510 Analyst, June 1997, Vol. 122High purity analytical reagent grade water (resistivity 18 MW cm) was prepared using a commercial laboratory reverse osmosis/deioniser system (Elga, High Wycombe, Bucks, UK). Reagent blanks were analysed for all procedures.To check the accuracy of the whole rock analysis a certified reference material, USGS Andesite AGV-1, prepared by the US Geological Survey, was analysed. Sample preparation Twelve samples from 9 borehole cores were selected for analysis. The foil and wax coating were removed from the middle portion of each core, which then was split into discs perpendicular to the core axis with a plastic-covered hammer and chisel. The contaminated outer surface was removed, leaving several clean, round sections of rock; these were broken into pieces of 1–4 cm in diameter.To prevent contamination, the samples were allowed to come in contact with only non-metallic tools and containers. Equipment was washed with an alkaline detergent solution (Micro), soaked in 25% v/v HNO3, and rinsed with deionised water before use. Sample powders were prepared from the 1–4 cm fragments of whole rock in the following stages: (i) the 1–4 cm sample was wrapped in thick plastic and crushed with a covered hammer to pieces < 1 cm in size; (ii) the < 1 cm pieces were ground with a mechanical mortar and pestle in stages until all the powder passed through a 400 mm nylon mesh sieve held by a plastic frame and collected in a plastic pan; (iii) the < 400 mm sample powder was transferred to a sheet of glazed paper, coned and quartered until a 100 g portion was separated; (iv) this 100 g portion of < 400 mm powder was ground in stages and passed through a 125 mm nylon mesh sieve; (v) both < 400 mm and < 125 mm powder fractions were weighed, stored in glass jars with plastic lids. The < 400mm fraction was bagged and refrigerated.The < 125 mm fraction was dried at 105 °C for 24 h and stored in a desiccator. Whole rock dissolution Acid digestion and fusion procedures were used in the whole rock analysis. The acid digestion procedure was used to produce solutions suitable for ICP-AES analysis for the major and trace elements (Al, Fe, Mg, Ca, Na, K, Ti, Mn, Ba, Sr) and for ICPMS analysis for lower abundance trace elements (Cr, Ni, Pb, Sn, Th, U, V, Zr).The fusion procedure, followed by an ion exchange preconcentration–separation procedure, was utilised to provide solutions for REEs, La, and Y determination by ICPAES or ICP-MS. Solutions for ICP-AES analysis for Si in the whole rock were prepared by a modified fusion procedure. The acid digestion procedure consisted of the following steps: (1) approximately 0.5 g of dry, < 125 mm powder was weighed accurately into a Teflon PFA beaker with a graphite base; (2) 15ml of concentrated HF was added to the beaker.The beaker was covered, and the powder digested at 20 °C for 48 h and then at 95 °C in a water bath for 3 h; (3) the beaker was uncovered and transferred to a hot plate. A 10 ml volume of concentrated HClO4 and 10 ml of concentrated HNO3 were added (to the solution in the beaker) and heated gently to dryness; (4) the residue was dissolved in 15 ml of concentrated HNO3 and again taken to dryness; (5) a second 15 ml volume of concentrated HNO3 was added to the beaker to dissolve the residue.This solution was transferred into a calibrated flask and made up to 50 ml with deionised water. The final solution was transferred to a high density polyethylene (HDPE) bottle for storage. The LiBO2 fusion–dissolution procedure used was based on that developed by Govindaraju and Mevelle.23 The method used was: (1) approximately 0.25 g of dry, < 125 mm powder and 1.0 g of LiBO2 flux were weighed accurately into a platinum crucible and mixed thoroughly; (2) the crucible was covered, placed in a preheated muffle furnace, and heated at 1000 °C for 2 h.The molten sample was swirled gently at 30 min intervals. On removal from the furnace, the crucible base was plunged into water to quench the melt and fracture the bead; (3) 20 ml of dissolving solution (100 ml of concentrated HCl, 10 g of (COOH)2·P2H2O and 5 ml of H2O2 diluted to 1 l with deionised water) was added to the crucible.The crucible was placed on a magnetic stirrer and the solution stirred for 30 min with a PTFE bar, after which it was transferred to a 100 ml calibrated flask. A second 20 ml aliquot of dissolving solution was added to the crucible, stirred for 30 min and transferred to the flask. The process was repeated until the bead was dissolved; (4) the flask was filled to the 100 ml mark with dissolving solution previously used to wash the sides and lid of the crucible.The sample solution was then transferred to a HDPE bottle pending further processing to separate and preconcentrate REEs. LiBO2, Li2B4O7 fusion–dissolution procedure for silicon Because of the high percentage of silica in the samples, the LiBO2 fusion procedure described previously had to be modified to achieve complete dissolution of Si. A 4 + 1 mixture of LiBO2 and Li2B4O7 fluxes was substituted for the LiBO2 flux; only 0.2 g of sample powder was fused and 25% HNO3 was used as the dissolving solution.All other conditions were identical to those outlined in the LiBO2 fusion procedure. After dissolution the sample was transferred to an HDPE bottle to await analysis. Ion exchange procedure for separation and preconcentration of REEs The REEs were separated using a ‘mini-column’ ion exchange procedure.23 The sample solution obtained by the fusion method described above (or the oxalate leachate) was ‘loaded’ onto columns constructed of 1 ml disposable plastic pipette tips packed with 2 g of Duolite 225 (SRC 16) (Merck) cation exchange resin.The detailed procedure is summarised in Table 13. Steps 1 and 2 were omitted if the columns had been used previously. The method for whole rock fusion solutions had to be modified for the oxalate leachates. The leach solution matrix was a more efficient eluent than the rock dissolving solution for moving the REEs down the column and a wash of smaller volume was required to give acceptable recoveries.Under these conditions the recovery for the three heaviest REEs (Tm, Yb, Lu) as tested on a spiked leachate matrix was still 10 to 15% low. Analytical protocols for ICP-AES and ICP-MS analysis For all analysis the ICP-AES instrument was recalibrated every 10 samples. For the whole rock analysis and REE separated fusion samples, a USGS standard rock (AGV-1) was processed Table 13 REE ion exchange separation procedure Acid Pump Step Acid concentration Volume of speed/ number type (% v/v) acid/ml ml min21 Comment 1 HCl 40 10 0.4 Conditioning 2 HCl 10 10 0.4 Conditioning 3 HCl 40 10 0.4 Conditioning 4 Sample 10 30 0.4 Loading 5 HNO3 12 18 (fusion) 0.4 Wash 10 (leachate) 0.4 Wash 6 25 10 0.4 Elution Analyst, June 1997, Vol. 122 511using the same dissolution procedures as the sandstone samples and analysed to check for accuracy and to maintain quality control. Internal standardisation (using Sc) was used on the whole rock digests to correct for matrix effects and to improve precision. Analysis of the whole rock and the leachates for the REEs required the use of matrix-matched standards because analytical emission lines (for the REEs) or the sample matrix (oxalate leachates) interfered with the Sc internal standard line. The solution sample preparation and the method of calibration for each of the different solution types is described below. For the acid-digested whole rock sample, 1 ml of solution was diluted to 10 ml with deionised water followed by the addition 0.4 ml of 1000 mg l21 Sc as internal standard. The samples were analysed against standards made up in 1% HNO3 containing the same concentration of Sc. The oxalate leachates were diluted 1 + 1 with deionised water and analysed against matrix matched standards. The REE eluates, in 25% HNO3, were analysed with no further preparation against matrix-matched standards. The ICP-MS instrument was calibrated every five samples with standards made up in 1% HNO3. The USGS standard rock AGV-1 was used for quality control for the whole rock analysis. The whole rock acid digestion samples were diluted 1 + 9 with deionised water before analysis. The oxalate leachate samples were prepared by digesting with H2O2 to reduce the dissolved solids content as described previously and then diluted 1 + 1 with deionised water. Analysis of standard rock material (USGS AGV-1) An international rock standard was prepared and analysed by the same procedures and methods as the sandstone samples, excluding initial crushing and grinding. The resulting analytical data are shown in Table 14, where they are compared with recommended values.24 All results are within the 95% confidence limits given in the data compiled by Gladney et al.,24 apart from Sr and Dy which are high and are probably due to contamination effects and Gd which is low and may be due to poor background correction in the ICP-AES analysis. Despite these small discrepancies the methodology was considered to be sufficiently accurate for the purposes of the studies carried out in this work. References 1 Chester, R., and Hughes, M. J., Chem. Geol., 1967, 2, 249. 2 Tessier, A., Campbell, P. G. C., and Bisson, M., Anal. Chem., 1979, 51, 844. 3 Rendell, P. S., Batley, G. E., and Cameron, A. J., Environ. Sci. Technol., 1980, 14, 314. 4 Tipping, E., Hetherington, N. B., Hilton, J., Thompson, D. W., Bowles E., and Hamilton-Taylor, J., Anal. Chem., 1985, 57, 844. 5 Rapin, F., Tessier, A., Campbell, P. G. C., and Carignan, R., Environ. Sci. Technol., 1986, 20, 836. 6 Kheboian, C., and Bauer, C. F., Anal. Chem., 1987, 59, 1417. 7 Sholkovitz, E. R., Chem. Geol., 1989, 77, 47. 8 Zielinski, R. A., Bloch, S., and Walker, T. R., Econ. Geol., 1983, 78, 1574. 9 Tamm, O., Medd. Skogsf�orsoeksues., 1922, 19, 387. 10 Wold, S., Ebenson, K., and Geladi, P., Chemom. Intell. Lab. Syst., 1987, 2, 37. 11 He, X., Li, H., and Shi, H., Fenxi Huaxue, 1986, 14, 34. 12 Wirsz, D. F., and Blades, M. W., Anal. Chem., 1986, 58, 51. 13 Zhang, P., Dudley, N., Ure, A. M., Littlejohn, D., Anal. Chim. Acta, 1992, 258, 1. 14 Malinowski, E. R., Anal. Chem., 1977, 49, 612. 15 Hopke, P. K., Atmos. Environ., 1982, 16, 1379. 16 Ward, J. H., J. Am. Stat. Assoc., 1963, 58, 236. 17 Kemp, S. J., and Pearce, J. M., UKirex Safety Studies Asessment Report, NSS/R239, in preparation. 18 Thurston, G. D., and Spengler, J. D., Atmos. Environ., 1985, 19, 9. 19 Malinowski, E. R., Factor Analysis in Chemistry, Wiley, New York, 1991. 20 Gamp, H., Maeder, M., Meyer, C. J., and Zuberbuhler, A. D., Talanta, 1985, 32, 1133. 21 Longworth, G., UK Nirex Safety Studies Assessment Report, NSS/ R303, in preparation. 22 Deer, W. A., Howie, R. A., and Zussman, An Introduction to the Rock Forming Minerals, Longman, 1982. 23 Govindaraju, K., and Mevelle, G., J. Anal. At. Spectrom., 1987, 2, 615. 24 Gladney, E. S., Jones, E. A., and Nickell, E. J., Geostand. Newsl., 1992, 16, 111. Paper 6/07953I Received November 25, 1996 Accepted February 18, 1997 Table 14 Comparison of AGV-1 analysis by the procedures described in this work to certified values 95% Detec- AGV-1 Recom- Contion (this mended fidence Element Unit limit work) values24 limits SiO2 % oxide — — 58.84 0.56 Al2O3 % oxide 0.003 17.1 17.15 0.3 Fe2O3 % oxide 0.003 7.01 6.77 0.34 MgO % oxide 0.011 1.47 1.53 0.2 CaO % oxide 0.003 5.01 4.94 0.28 NaO2 % oxide 0.003 4.41 4.26 0.22 K2O % oxide 0.033 3.06 2.92 0.16 TiO2 % oxide 0.001 1.1 1.05 0.1 MnO % oxide 0.00001 0.097 0.09 0.02 Ba mg kg21 1.6 1224 1226 34 Cr mg kg21 0.05 9.4 10.1 4.8 Ni mg kg21 0.2 12.6 16 6 Pb mg kg21 0.2 35.3 36 10 Sn mg kg21 0.1 4.59 4.2 2.2 Sr* mg kg21 0.1 695 662 18 Th mg kg21 0.1 6.18 6.5 1 U mg kg21 0.1 1.99 1.92 0.3 V mg kg21 9 106 121 22 Zr mg kg21 3 249 227 36 Ce mg kg21 1.4 66 67 10 Dy* mg kg21 0.41 6.71 3.6 0.6 Er mg kg21 0.41 1.1 1.7 0.4 Eu mg kg21 0.05 1.7 1.64 0.2 Gd* mg kg21 < 0.96 5 1 Ho mg kg21 0.68 < 0.68 0.67 0.2 La mg kg21 0.41 36.1 38 6 Lu mg kg21 0.07 0.21 0.27 0.06 Nd mg kg21 1.51 28.6 33 6 Pr mg kg21 1.23 7.21 7.6 2.2 Sm mg kg21 0.68 6 5.9 0.8 Tb mg kg21 1.6 < 1.6 0.7 0.2 Tm mg kg21 0.7 < 0.7 0.34 0.12 Y mg kg21 0.1 17.5 20 6 Yb mg kg21 0.1 1.62 1.72 0.38 * See explanation in text. 512 Analyst, June 1997, Vol. 122

ISSN:0003-2654    

DOI:10.1039/a607953i

出版商:RSC    

年代:1997

数据来源: RSC

摘要:

Comparison of Chemometric Methods: Derivative Ratio Spectra and Multivariate Methods (CLS, PCR and PLS) for the Resolution of Ternary Mixtures of the Pesticides Carbofuran Carbaryl and Phenamifos After Their Extraction into Chloroform T. Galeano D�ýaz, A. Guiberteau*, J. M. Ort�ýz Burguillos and F. Salinas Analytical Chemistry Department, University of Extremadura, Badajoz 06071, Spain The resolution of a ternary mixture of the pesticides carbofuran, carbaryl and fenamiphos in heterogeneous media (after extraction into CHCl3) by the application of different chemometric methods such as derivative ratio spectra (DD), classical least squares (CLS), principal components regression (PCR) and partial least squares (PLS) was performed.CLS, PCR and PLS were applied with the absorption spectra or with their transformations (logarithm or first or second derivative). Also, different data preprocessing algorithms were examined. Second-derivative spectra were used for DD.A comparison of the results obtained in the analysis of these compounds by the different methods was made by using analysis of variance (ANOVA). These methods were successfully applied to the analysis of spiked river water samples. Keywords: Chemometrics; derivative ratio spectra; multivariate methods; pesticides; water analysis Carbofuran (2,3-dihydro-2,2-dimethyl-7-benzofuranyl methylcarbamate), carbaryl (1-naphthyl methylcarbamate) and fenamiphos [ethyl 3-methyl-4-(methylthio)phenyl isopropylphosphoramidate] are insecticides widely utilized in agriculture.There are reports on the determination of carbofuran and carbaryl in a mixture in aqueous medium by the spectrophotometric derivative method1 and by chemometric methods,2,3 specifically partial least squares (PLS) methods, applied either to spectrophotometric or polarographic signals. A mixture of fenamiphos and folpet4 and a mixture of carbaryl and chlorpyrifos5 have also been analyzed by using PLS methods, but there is no reference to the resolution of a ternary mixture of carbofuran, carbaryl and fenamiphos by chemometric methods.In this paper, we report on the resolution of this ternary mixture in heterogeneous media by the application of different chemometric methods: derivative ratio spectra (DD) and the multivariate methods classical least squares (CLS), principal component regression (PCR) and partial least squares (PLS) (types PLS-1 and PLS-2).CLS, PCR and PLS were used with the absorption spectra and with their transformation into first- and second-derivative and logarithmic absorption spectra, after extraction into CHCl3. These methods were subsequently applied to the resolution of carbofuran, carbaryl and fenamiphos in spiked river water samples. The abilities of different chemometric methods to resolve mixtures of different compounds whose signals (i.e., absorption spectra) are overlapped have been widely utilized.The main advantage of multicomponent analysis by using multivariate calibration is the speed of the determination for the components in mixtures, avoiding the need for a prior separation that is otherwise necessary owing to the overlapping of the signals. Salinas et al.6 developed a spectrophotometric method for resolving binary mixtures when the spectra of the components are overlapped. The method is based on the use of the first derivative of the ratios of the spectra.The absorption spectrum of the mixture is obtained and divided (amplitude by amplitude at appropriate wavelengths) by the absorption spectrum of a standard solution of one of the components (previously stored in a computer), and the first derivative (or another derivative order) of the ratio spectrum is obtained. The concentration of the other component is then determined from a calibration graph. Later, the method was extended to the resolution of ternary mixtures in combination with zero-crossing methods.7 On the other hand, owing to the the increase in the resolving power of analytical instrumentation and the easier access to microcomputers with appropriate software, in recent years the use of multivariate calibration data, that is, of the analytical signal depending on two or more variables, has become more general.Methods such as CLS, PCR and PLS have frequently been used in quantitative spectral analysis to obtain very selective information from unselective data.All these methods, named full-spectrum methods, assume a linear relationship between the absorbance values and the concentrations of the components in the mixture, although PCR and PLS can also be applied to non-linear systems.8–10 Each method needs a calibration step, where the relationship between the spectra and the concentrations of the components is deduced from a set of reference samples, followed by a prediction step in which the results of the calibration are used to determine the concentrations of the components from the spectra of the analyzed samples.The CLS method is the easiest of the multivariate methods, and is based on calibration by multiple linear regression. Its main disadvantage is that it is a rigid model that needs the knowledge of all the components in the mixture and their concentrations, and that there should be no chemical or physical interaction between the components in the mixture or with other compounds present in the matrix.For the application of the PLS and PCR methods, which are more flexible and do not need these requirements for their application, it is necessary to make a previous spectral decomposition. They are methods based on factors analysis and their objective is to obtain the spectrum of the mixture from a determined number of variable spectra named loadings and the different amounts of each of them that must be added to reconstruct the original spectrum and that are known as scores.The difference between the PCR and PLS methods is the following: in the PCR method only the information in the matrix of signals is used in the spectral decomposition, but in the PLS method the concentration data matrix is also used in this step. When the decomposition and regression are made at the same time, for all the components, the type is named PLS-2, but if this decomposition and regression are made separately for each component the name given is PLS-1.The main advantage Analyst, June 1997, Vol. 122 (513–517) 513of PLS-2 is the higher speed and less complexity, but these advantages have became less important owing to the increase in the calculation capacity of computers today. Also, for complex systems or with a great variability of the components, the use of PLS-1 is preferred. Different workers11–14 have performed an exhaustive mathematical treatment of the algorithms implied in these methods. Some researchers have proposed the use of the multivariate calibration methods in combination with the derivative techniques.However, their convenience is contradictory.15–17 In this work, the resolution of the ternary mixtures of carbofuran, carbaryl and fenamiphos by extraction–UV spectrophotometry was carried out, using the chemometric methods second-derivative ratio spectra, CLS, PCR, and PLS (types 1 and 2). A comparison of the results obtained in the analysis of synthetic samples by the different methods applied was made by using analysis of variance (ANOVA).The data used were the absorption spectra and their transformations (logarithm or derivative) and these data were preprocessed in different ways (mean centring or variance scaling). Experimental Apparatus A Beckman (Fullerton, CA, USA) DU-50 spectrophotometer connected via an RS-232 to an Olivetti PC 286 microcomputer was used for all absorption measurements.Beckman Data Leader Software, version 3.0,18 was used for spectral acquisition, storage, manipulation and analysis of the spectrophotometric data.The calculation of the first derivative absorption spectra was performed by the Savitzky–Golay simplified leastsquares method of spectral smoothing and differentiation.19,20 The Grams/386 Software Package, version A 1.01, and the PLS plus version 2.0 Application software21 were used for the statistical treatment of the data and the application of PLS method.Reagent and Chemicals Standard solutions (2 3 1025 m) of carbofuran, carbaryl and fenamiphos (supplied by Sigma, St. Louis, MO, USA) were prepared by weighing of appropriate amounts and dissolution in HPLC-grade water. A 0.5 m AcOH–AcONa buffer solution of pH 4.7 and a 0.3 m NaCl solution were used. All other reagents were of analytical-reagent grade. Procedure for the Determination of the Ternary Mixture in River Water Samples The samples of river water were stored at low temperature and in the dark and filtered through 0.45 mm nylon filters before analysis.In a separating funnel an aliquot of the sample containing between 3 3 1026 and 1.3 3 1025 m for each component was taken and 10.0 ml of 3 m NaCl, 12.0 ml of buffer (pH 4.7) and the necessary volume of deionized water to complete the solution to 50.0 ml were added. Subsequently the mixture was extracted with 5.0 ml of CHCl3 by shaking vigorously for 2 min.The organic phase was separated, centrifuged and filtered.Finally, the absorption spectrum between 200 and 350 nm against an identically prepared sample but without the pesticides was recorded. The optimized calibration matrix calculated by application of the multivariate methods (PCR and PLS) was applied to analyse the spectrum obtained. Results and Discussion CLS, PCR and PLS Methods Carbofuran, carbaryl and fenamiphos absorb in the UV region, with lmax 277, 279 and 249 nm and molar absorptivities � of 2.6 3 103, 6.1 3 103 and 10.7 3 103 l mol21 cm21, respectively, in aqueous medium.Owing to its convenience, to improve the sensitivity of the methods to determine these pesticides extraction with CHCl3, with a phase ratio of 10 : 1 was used. The absorption spectra of carbofuran, carbaryl and fenamiphos obtained by extraction of aqueous solutions (pH 4.7 with acetic acid–acetate buffer) with CHCl3 (phase ratio 10 : 1) were recorded between 200 and 350 nm with a scan rate of 500 nm min21.In Fig. 1 the overlapped peaks of the three compounds are shown, together with the spectrum of a mixture of the three compounds. These spectra, and also all others subsequently registered, were filtered through a seven experimental points window. The multivariate methods (CLS, PCR and PLS) were applied with the absorption spectra, with their logarithm, and with the derivative spectra (first or second derivative, by using a Dl of 4 nm), after the extraction process. The results obtained by applying different methods were compared.For the application of these multivariate methods a training set of 14 ternary samples (series 1) with different concentrations of each component in the range 3.0 3 1026–1.3 3 1025 m) was prepared, following the same procedure as detailed for the determination of the three pesticides in river water samples, to carry out the calibration step. The absorption spectra were obtained after extraction into CHCl3 (Fig. 2).In Table 1 the Fig. 1. Absorption spectra after extraction into CHCl3 (phase ratio 10 : 1, pH 4.7) of (1) carbofuran (1.30 3 1025 m), (2) carbaryl (1.05 3 1025 m), (3) fenamiphos (5.50 3 1026 m) and (4) the mixture. Fig. 2 Absorption spectra after extraction into CHCl3 (phase ratio 10 : 1, pH 4.7) of ternary mixtures of carbofuran, carbaryl and fenamiphos (series 1). The spectra are filtered through seven experimental points. 514 Analyst, June 1997, Vol. 122composition of the ternary mixtures employed (series 1) is summarized.The spectral region between 245 and 345 nm was selected for the analysis, which implies working with 101 experimental points per spectrum. In Fig. 2 the absorption spectra of these ternary mixtures and in Fig. 3 the firstderivative spectra (as an example of the assayed transformations) are shown. A second series of eight samples (series 2) was prepared as the prediction set. The composition of these samples, prepared in triplicate, is also summarized in Table 1.To select the number of factors in the PCR or PLS algorithms, the cross-validation method, leaving out one sample at a time,22 was used. The prediction error sum of squares (PRESS) was calculated each time a new factor was added giving rise to different PCR or PLS models. In our opinion, a good criterion in the selection of the optimum number of factors is to compare the minimum PRESS, corresponding to a model with h* factors, with the PRESS of the other models, and select the model with the smallest number of factors, h factors, such that their PRESS is not significantly greater than minimum PRESS, using to establish this criterion the F-statistic and the Haaland and Thomas criterion.14 In conclusion we selected, as the optimum, the number of factors for the first PRESS value the F-ratio probability of which drops below 0.75, avoiding in this way some overfitting. The statistical parameters rmsd (root mean square deviation), which is an indication of the average error in the analysis, for each component, and r2 (square of the correlation coefficient), which shows how the plots of actual versus predicted concentrations fit to a straight line, were used to evaluate the different methods.As already indicated the data used were the absorption spectra or their logarithm or derivative. The r2 and rmsd values obtained were considerably worse when the logarithm of the spectra was used, and only slightly better results were obtained when first- or second- derivative spectra were used. Henceforth, we only examined the absorption spectra.In Table 2, the number of factors (n) for the training set of standards and the statistical parameters (rmsd and r2) are summarized for the different multivariate methods used. The Lab-Cal software package carries out, as the default choice, a mean centring of the data. This is the only preprocessing algorithm used to obtain the results in Table 2.In addition, none of the other preprocessing algorithms were tried, such as baseline correction, the variance scaling (scale to unit variance) or multiplicative scatter correction (MSC) in combination with the PCR method. Except when we applied MSC, we observed that the successive application of these preprocessing algorithms slightly improves the r2 and rmsd values. The multivariate methods were applied to the resolution of the ternary synthetic samples (series 2, Table 1).The results obtained, which are very close to a 100% of recovery in all instances, are discussed below. Resolution of Ternary Mixtures by Applying Second-derivative Ratio Spectra The stored spectra of the ternary mixtures were divided by a standard spectrum of carbofuran (2.5 3 1025 m) obtained under the same conditions. The ratio spectra thus obtained were smoothed through the use of 25 experimental points and the second derivatives were calculated with Dl = 4 nm (selected values once the optimization of instrumental variables had been performed).In Fig. 4, a spectrum of a ternary mixture (a), a ratio spectrum between these and the spectrum of a sample of carbofuran (b) and the second-derivative ratio spectrum (c) are shown as example. Since the absorbance values at the different wavelengths of a dissolution of the three compounds in which the concentrations of carbofuran (A), carbaryl (B) and fenamiphos (C) are cA, cB and cC is AM,li = eA,li c A + eB,licB + eC,licC where b = 1, when we divide this spectrum by the spectrum of a standard dissolution of carbofuran (of concentration c°A), that is, by A°A,li = �°A,licA we obtain AM,li Ao M,li = cA co A + cB co A ÎB,li ÎA,li + cC co A ÎC,li ÎA,li Table 1 Composition of the samples of the calibration set (series 1, samples 1–14) and prediction set (series 2, samples 15–38) Concentration/1026 m Sample Carbofuran Carbaryl Fenamiphos 1 3.00 13.00 13.00 2 5.50 10.50 13.00 3 8.00 8.00 13.00 4 10.50 5.50 13.00 5 13.00 3.00 13.00 6 13.00 8.00 8.00 7 13.00 13.00 3.00 8 10.50 13.00 5.50 9 8.00 13.00 8.00 10 10.50 10.50 8.00 11 8.00 10.50 10.50 12 5.00 5.00 10.00 13 5.00 10.00 5.00 14 9.50 5.50 5.00 15–17 10.00 10.00 6.00 18–20 10.00 6.00 10.00 21–23 6.00 10.00 10.00 24–26 5.00 5.50 5.00 27–29 5.00 7.50 7.50 30–32 7.50 5.00 7.50 33–35 7.50 7..00 36–38 9.00 5.00 7.00 Fig. 3 First-derivative spectra (Dl = 4 nm) after extraction into CHCl3 (phase ratio 10 : 1, pH 4.7) of ternary mixtures of carbofuran, carbaryl and fenamiphos (series 1).Analyst, June 1997, Vol. 122 515and the obtaining of the second derivative of this spectrum ratio gives d2 dl2 AM,li Ao A,li æ è ç ö ø ÷ = cB co A d2 dl2 ÎB,li ÎA,li æ è ç ö ø ÷ + cC co A d2 dl2 ÎC,li ÎA,li æ è ç ö ø ÷ In this equation we can see that the derivative ratio spectrum is dependent only on c°A, cB and cC. Later the zero-crossing wavelengths of B and C are determined, that is, the wavelength at which one of the two terms in the previous equation takes a zero value.At these wavelengths, the signal measured in the second-derivative ratio spectrum (DD) is dependent only on the concentration of one of the components of the ternary mixture. In our case, using the above-mentioned spectrum as divisor, the amplitude at DD278.10 was proportional to the concentration of carbaryl and the amplitude at DD267.24 was proportional to the concentration of fenamiphos.On the other hand, for determining carbofuran, stored spectra of the mixture were divided by a standard spectrum of carbaryl (5 3 1026 m). The ratio spectra thus obtained, following the same treatment as detailed before, gives the concentration of carbofuran by measuring the amplitude at DD283.77. We would obtain in the same manner the concentration of fenamiphos, but this is unnecessary since it has already been determined. Also, it is unnecessary to divide by a standard spectrum of fenamiphos.The statistical parameters of the regression model obtained are given in Table 3. Comparison Between the Different Chemometric Methods Applied ANOVA was applied to the comparison of two factors, the factor method and the factor component, using as the statistical population the percentage recoveries of the analytes in the synthetic samples. Before applying the ANOVA methods, we applied Bartlett’s criterion, verifying the variance homogeneity.In Table 4 the means of the recovery values ( � R), standard error and confidence interval for a 95% confidence level (IC95) for the mean of each method and each component are shown. The results of the ANOVA are summarized in Table 5. Fig. 4 (a) Spectrum of a ternary mixture of carbofuran (3 3 1026 m), carbaryl (1.3 31025 m) and fenamiphos (1.3 31026 m). (b) Ratio spectrum of (a) and carbofuran (2.5 3 1025 m) as divisor and (c) second-derivative ratio spectrum.Table 2 General characteristics of the matrices Components Carbofuran Carbaryl Fenamiphos Matrix r2 n* rmsd r2 n* rmsd r2 n* rmsd CLS 0.9872 — 0.3536 0.9907 — 0.3152 0.9984 — 0.1377 PCR 0.9904 4 0.3068 0.9934 4 0.2649 0.9986 4 0.1272 PLS-1 0.9963 5 0.1894 0.9976 5 0.1599 0.9987 4 0.1259 PLS-2 0.9961 5 0.1960 0.9978 5 0.1537 0.9984 5 0.1357 * n is the number of factors utilized (see text). Table 3 Statistical data for calibration models (derivative ratio spectra) Parameter Carbofuran Carbaryl Fenamiphos b1 (slope) 992.0 792.0 3667.8 sm (standard deviation 23.35 12.22 32.61 of the slope) b0 (intercept) 4.440 3 1024 27.600 3 1025 1.636 3 1023 sb (standard deviation 2.060 3 1024 1.069 3 1024 4.602 3 1024 of intercept) r (regression 0.9992 0.9996 0.9997 coefficient) s (standard error) 1.862 3 1024 9.661 3 1025 4.159 3 1024 LOD*/m* 6.237 3 1027 4.049 3 1027 3.768 3 1027 * Winefordner and Long method23 with k = 3.Table 4 Mean recoveries for synthetic samples Standard error (%) Level R – (%) (internal) IC95 (%) Method— CLS 100.7333 1.5366 98.8349–102.6324 PCR 101.5958 0.9142 99.6967–103.4949 PLS-1 101.4667 0.6744 99.5676–103.3658 PLS-2 101.2417 0.6640 99.3426–103.1408 DD 97.4250 0.7020 95.5259–99.3241 Components— Carbofuran 99.8625 0.4615 98.3915–101.3335 Carbaryl 100.3525 0.5996 98.8815–101.8235 Fenamiphos 101.2625 0.3641 99.7915–102.7335 516 Analyst, June 1997, Vol. 122The F-test is used to see whether the estimates of variance differ significantly.Comparing the between-component mean squares with the residual mean squares gives F2/113 = 0.916; the critical value23 is 3.079, indicating that there is not a statistically significant difference between the means by the different components at the 95% confidence level. With respect to the factor method, the value obtained is F4/113 = 3.320, the critical value being 2.459 (p = 0.05). Hence there is a statistically significant difference between the methods.Thus, making a multiple range test, it is established that only the mean obtained by the ratio derivative spectra method (DD) cannot be considered equal to those obtained by the other methods, since the differences between them are higher than the calculated limit value24 of 2.6857, for the level of significance established. These differences between the means for the groups CLS–DD, PCR–DD, PLS–1–DD and PLS–2–DD are 3.3083, 4.1708, 4.0417 and 3.8167, respectively.Similar multifactorial analysis of two factors was carried out to compare, on the one hand, the factor transformation of spectra or preprocessing algorithm and, on the other, the factor component, the conclusion being that no significant difference exists in any case. The chemometric methods described above were applied successfully to the analysis of samples of spiked river water according with the above-described procedure. The recoveries were 100–116% for carbofuran, 99–110% for carbaryl and 96–111% for fenamiphos.Conclusions All the methods utilized (CLS, PCR, PLS and DD) can be used in the determination of the three pesticides in their mixtures by using the UV absorption spectra or the first- or secondderivative spectra (CLS, PCR and PLS) or second-derivative ratio spectra (DD) after extraction into CHCl3. However, the easier treatment of the data by using CLS, PCR or PLS makes these preferable to the DD method. On the other hand, one must take into account the better statistical parameters obtained (r2 and rmsd) when the PLS or PCR is used compared with CLS.The data enhancement methods, mean centring and variance scaling, or the other preprocessing algorithm used, baseline correction, give rise to only a small improvement in the prediction ability of the different multivariate calibration methods. However, in the validation of the multivariate methods used with synthetic samples, no significant differences (by application of the ANOVA method) are observed with the different pretreatments of the data used.The authors are grateful to the CICYT (Proyect ALI95-1538) and to the Consejer�ýa de Educaci�on y Juventud de la Comunidad de Extremadura (Project EIA94-35) for financial support. References 1 Salvador, A., De Benzo, Z. A., and De la Guardia, M., Mikrochem. J., 1990, 42, 187. 2 Ort�ýz Burguillos, J. M., Tesis Doctoral, University of Extremadura, 1995. 3 Guiberteau A., Galeano D�ýaz, T., Salinas, F., and Ort�ýz, J.M., Anal. Chim. Acta, 1995, 305, 219. 4 Cervantes Oca�na, D., Gil Garc�ýa, M. D., Mart�ýn-Galera, M., and Martinez Vidal, J. L., Bull. Soc. Chim. Belg., 1993, 102, 431. 5 Espinosa-Mansilla, A. Salinas, F., and Zamoro, A., Analyst, 1994, 119, 1183. 6 Salinas, F., Berzas, J. J., and Espinosa Mansilla, A., Talanta, 1990, 37, 347. 7 Berzas Nevado, J. J., Guiberteau Cabanillas, C., and Salinas, F., Talanta, 1992, 39, 547. 8 Vogts, N. B., Chemom.Intell. Lab. Syst., 1989, 7, 119. 9 Wold, S. , Kettaned-Wold, N., and Shagerberg, B., Chemom. Intell. Lab. Syst., 1989, 7, 53. 10 Naes , T., Isaksson, T., and Kowalski, B. R., Anal Chem., 1990, 62, 664. 11 Martens, H., and Naes, T., Multivariate Calibration, Wiley, Chichester, 1989. 12 Thomas, E. V., and Haaland, D. M., Anal. Chem., 1990, 62, 1091. 13 Tauler, R., and Casassas, E., Anal. Chim. Acta, 1989, 223, 257. 14 Haaland, D. M., and Thomas, E. V., Anal. Chem., 1988, 60, 1193. 15 Jones, R., Coomber, T. J., McCormick, J. P., Fell, A. F., and Clark, B. 988, 25, 381. 16 MacLaurin, P., Worsfold, P. J., Crane, M., and Norman, P., Anal. Proc., 1992, 29, 65. 17 Dur�an-Mer�as, I., Mu�noz de la Pe�na, A., Espinosa-Mansilla, A., and Salinas, F., Analyst, 1993, 118, 807. 18 Data Leader Software Package, Beckman, Fullerton, CA, 1989. 19 Savitzky, A., and Golay, M. J. E., Anal. Chem., 1964, 36, 1627. 20 Stainer, J., Termonia, Y., and Deltour, J., Anal.Chem., 1972, 44, 1906. 21 GRAMS/386 Software Package, Galactic Industries, Salem, NH, 1991 (web site: www.galactic.com). 22 Stone, W., J. R. Statist. Soc., 1974, B38, 111. 23 Winefordner, J. D., and Long, G. L., Anal. Chem., 1983, 55, 712A. 24 Walpole, R. E., and Myers, R. H., Probabilidad y Estad�ýstica para Ingenieros, Interamericana, M�exico City, 3rd edn., 1990. Paper 6/07955E Received November 25, 1996 Accepted March 11, 1997 Table 5 Results of ANOVA for two factors: ratio of the recoveries obtained by the applied method and the nature of the components Sum of Degrees of Mean Source of variation squares freedom square F-ratio Between method 292.6837 4 73.1709 3.320 Between component 40.3760 2 20.1880 0.916 Residual 2490.7836 113 22.0423 Total 2823.8433 119 Analyst, June 1997, Vol. 122 517 Comparison of Chemometric Methods: Derivative Ratio Spectra and Multivariate Methods (CLS, PCR and PLS) for the Resolution of Ternary Mixtures of the Pesticides Carbofuran Carbaryl and Phenamifos After Their Extraction into Chloroform T.Galeano D�ýaz, A. Guiberteau*, J. M. Ort�ýz Burguillos and F. Salinas Analytical Chemistry Department, University of Extremadura, Badajoz 06071, Spain The resolution of a ternary mixture of the pesticides carbofuran, carbaryl and fenamiphos in heterogeneous media (after extraction into CHCl3) by the application of different chemometric methods such as derivative ratio spectra (DD), classical least squares (CLS), principal components regression (PCR) and partial least squares (PLS) was performed.CLS, PCR and PLS were applied with the absorption spectra or with their transformations (logarithm or first or second derivative). Also, different data preprocessing algorithms were examined. Second-derivative spectra were used for DD. A comparison of the results obtained in the analysis of these compounds by the different methods was made by using analysis of variance (ANOVA).These methods were successfully applied to the analysis of spiked river water samples. Keywords: Chemometrics; derivative ratio spectra; multivariate methods; pesticides; water analysis Carbofuran (2,3-dihydro-2,2-dimethyl-7-benzofuranyl methylcarbamate), carbaryl (1-naphthyl methylcarbamate) and fenamiphos [ethyl 3-methyl-4-(methylthio)phenyl isopropylphosphoramidate] are insecticides widely utilized in agriculture. There are reports on the determination of carbofuran and carbaryl in a mixture in aqueous medium by the spectrophotometric derivative method1 and by chemometric methods,2,3 specifically partial least squares (PLS) methods, applied either to spectrophotometric or polarographic signals.A mixture of fenamiphos and folpet4 and a mixture of carbaryl and chlorpyrifos5 have also been analyzed by using PLS methods, but there is no reference to the resolution of a ternary mixture of carbofuran, carbaryl and fenamiphos by chemometric methods.In this paper, we report on the resolution of this ternary mixture in heterogeneous media by the application of different chemometric methods: derivative ratio spectra (DD) and the multivariate methods classical least squares (CLS), principal component regression (PCR) and partial least squares (PLS) (types PLS-1 and PLS-2). CLS, PCR and PLS were used with the absorption spectra and with their transformation into first- and second-derivative and logarithmic absorption spectra, after extraction into CHCl3.These methods were subsequently applied to the resolution of carbofuran, carbaryl and fenamiphos in spiked river water samples. The abilities of different chemometric methods to resolve mixtures of different compounds whose signals (i.e., absorption spectra) are overlapped have been widely utilized. The main advantage of multicomponent analysis by using multivariate calibration is the speed of the determination for the components in mixtures, avoiding the need for a prior separation that is otherwise necessary owing to the overlapping of the signals.Salinas et al.6 developed a spectrophotometric method for resolving binary mixtures when the spectra of the components are overlapped. The method is based on the use of the first derivative of the ratios of the spectra. The absorption spectrum of the mixture is obtained and divided (amplitude by amplitude at appropriate wavelengths) by the absorption spectrum of a standard solution of one of the components (previously stored in a computer), and the first derivative (or another derivative order) of the ratio spectrum is obtained.The concentration of the other component is then determined from a calibration graph. Later, the method was extended to the resolution of ternary mixtures in combination with zero-crossing methods.7 On the other hand, owing to the the increase in the resolving power of analytical instrumentation and the easier access to microcomputers with appropriate software, in recent years the use of multivariate calibration data, that is, of the analytical signal depending on two or more variables, has become more general. Methods such as CLS, PCR and PLS have frequently been used in quantitative spectral analysis to obtain very selective information from unselective data. All these methods, named full-spectrum methods, assume a linear relationship between the absorbance values and the concentrations of the components in the mixture, although PCR and PLS can also be applied to non-linear systems.8–10 Each method needs a calibration step, where the relationship between the spectra and the concentrations of the components is deduced from a set of reference samples, followed by a prediction step in which the results of the calibration are used to determine the concentrations of the components from the spectra of the analyzed samples.The CLS method is the easiest of the multivariate methods, and is based on calibration by multiple linear regression.Its main disadvantage is that it is a rigid model that needs the knowledge of all the components in the mixture and their concentrations, and that there should be no chemical or physical interaction between the components in the mixture or with other compounds present in the matrix. For the application of the PLS and PCR methods, which are more flexible and do not need these requirements for their application, it is necessary to make a previous spectral decomposition.They are methods based on factors analysis and their objective is to obtain the spectrum of the mixture from a determined number of variable spectra named loadings and the different amounts of each of them that must be added to reconstruct the original spectrum and that are known as scores. The difference between the PCR and PLS methods is the following: in the PCR method only the information in the matrix of signals is used in the spectral decomposition, but in the PLS method the concentration data matrix is also used in this step.When the decomposition and regression are made at the same time, for all the components, the type is named PLS-2, but if this decomposition and regression are made separately for each component the name given is PLS-1. The main advantage Analyst, June 1997, Vol. 122 (513–517) 513of PLS-2 is the higher speed and less complexity, but these advantages have became less important owing to the increase in the calculation capacity of computers today.Also, for complex systems or with a great variability of the components, the use of PLS-1 is preferred. Different workers11–14 have performed an exhaustive mathematical treatment of the algorithms implied in these methods. Some researchers have proposed the use of the multivariate calibration methods in combination with the derivative techniques. However, their convenience is contradictory.15–17 In this work, the resolution of the ter carbofuran, carbaryl and fenamiphos by extraction–UV spectrophotometry was carried out, using the chemometric methods second-derivative ratio spectra, CLS, PCR, and PLS (types 1 and 2).A comparison of the results obtained in the analysis of synthetic samples by the different methods applied was made by using analysis of variance (ANOVA).The data used were the absorption spectra and their transformations (logarithm or derivative) and these data were preprocessed in different ways (mean centring or variance scaling).Experimental Apparatus A Beckman (Fullerton, CA, USA) DU-50 spectrophotometer connected via an RS-232 to an Olivetti PC 286 microcomputer was used for all absorption measurements. Beckman Data Leader Software, version 3.0,18 was used for spectral acquisition, storage, manipulation and analysis of the spectrophotometric data. The calculation of the first derivative absorption spectra was performed by the Savitzky–Golay simplified leastsquares method of spectral smoothing and differentiation.19,20 The Grams/386 Software Package, version A 1.01, and the PLS plus version 2.0 Application software21 were used for the statistical treatment of the data and the application of the PLS method. Reagent and Chemicals Standard solutions (2 3 1025 m) of carbofuran, carbaryl and fenamiphos (supplied by Sigma, St.Louis, MO, USA) were prepared by weighing of appropriate amounts and dissolution in HPLC-grade water.A 0.5 m AcOH–AcONa buffer solution of pH 4.7 and a 0.3 m NaCl solution were used. All other reagents were of analytical-reagent grade. Procedure for the Determination of the Ternary Mixture in River Water Samples The samples of river water were stored at low temperature and in the dark and filtered through 0.45 mm nylon filters before analysis. In a separating funnel an aliquot of the sample containing between 3 3 1026 and 1.3 3 1025 m for each component was taken and 10.0 ml of 3 m NaCl, 12.0 ml of buffer (pH 4.7) and the necessary volume of deionized water to complete the solution to 50.0 ml were added.Subsequently the mixture was extracted with 5.0 ml of CHCl3 by shaking vigorously for 2 min.The organic phase was separated, centrifuged and filtered. Finally, the absorption spectrum between 200 and 350 nm against an identically prepared sample but without the pesticides was recorded.The optimized calibration matrix calculated by application of the multivariate methods (PCR and PLS) was applied to analyse the spectrum obtained. Results and Discussion CLS, PCR and PLS Methods Carbofuran, carbaryl and fenamiphos absorb in the UV region, with lmax 277, 279 and 249 nm and molar absorptivities � of 2.6 3 103, 6.1 3 103 and 10.7 3 103 l mol21 cm21, respectively, in aqueous medium. Owing to its convenience, to improve the sensitivity of the methods to determine these pesticides extraction with CHCl3, with a phase ratio of 10 : 1 was used.The absorption spectra of carbofuran, carbaryl and fenamiphos obtained by extraction of aqueous solutions (pH 4.7 with acetic acid–acetate buffer) with CHCl3 (phase ratio 10 : 1) were recorded between 200 and 350 nm with a scan rate of 500 nm min21. In Fig. 1 the overlapped peaks of the three compounds are shown, together with the spectrum of a mixture of the three compounds.These spectra, and also all others subsequently registered, were filtered through a seven experimental points window. The multivariate methods (CLS, PCR and PLS) were applied with the absorption spectra, with their logarithm, and with the derivative spectra (first or second derivative, by using a Dl of 4 nm), after the extraction process. The results obtained by applying different methods were compared. For the application of these multivariate methods a training set of 14 ternary samples (series 1) with different concentrations of each component in the range 3.0 3 1026–1.3 3 1025 m) was prepared, following the same procedure as detailed for the determination of the three pesticides in river water samples, to carry out the calibration step. The absorption spectra were obtained after extraction into CHCl3 (Fig. 2). In Table 1 the Fig. 1. Absorption spectra after extraction into CHCl3 (phase ratio 10 : 1, pH 4.7) of (1) carbofuran (1.30 3 1025 m), (2) carbaryl (1.05 3 1025 m), (3) fenamiphos (5.50 3 1026 m) and (4) the mixture.Fig. 2 Absorption spectra after extraction into CHCl3 (phase ratio 10 : 1, pH 4.7) of ternary mixtures of carbofuran, carbaryl and fenamiphos (series 1). The spectra are filtered through seven experimental points. 514 Analyst, June 1997, Vol. 122composition of the ternary mixtures employed (series 1) is summarized. The spectral region between 245 and 345 nm was selected for the analysis, which implies working with 101 experimental points per spectrum.In Fig. 2 the absorption spectra of these ternary mixtures and in Fig. 3 the firstderivative spectra (as an example of the assayed transformations) are shown. A second series of eight samples (series 2) was prepared as the prediction set. The composition of these samples, prepared in triplicate, is also summarized in Table 1. To select the number of factors in the PCR or PLS algorithms, the cross-validation method, leaving out one sample at a time,22 was used.The prediction error sum of squares (PRESS) was calculated each time a new factor was added giving rise to different PCR or PLS models. In our opinion, a good criterion in the selection of the optimum number of factors is to compare the minimum PRESS, corresponding to a model with h* factors, with the PRESS of the other models, and select the model with the smallest number of factors, h factors, such that their PRESS is not significantly greater than minimum PRESS, using to establish this criterion the F-statistic and the Haaland and Thomas criterion.14 In conclusion we selected, as the optimum, the number of factors for the first PRESS value the F-ratio probability of which drops below 0.75, avoiding in this way some overfitting.The statistical parameters rmsd (root mean square deviation), which is an indication of the average error in the analysis, for each component, and r2 (square of the correlation coefficient), which shows how the plots of actual versus predicted concentrations fit to a straight line, were used to evaluate the different methods.As already indicated the data used were the absorption spectra or their logarithm or derivative. The r2 and rmsd values obtained were considerably worse when the logarithm of the spectra was used, and only slightly better results were obtained when first- or second- derivative spectra were used. Henceforth, we only examined the absorption spectra.In Table 2, the number of factors (n) for the training set of standards and the statistical parameters (rmsd and r2) are summarized for the different multivariate methods used. The Lab-Cal software package carries out, as the default choice, a mean centring of the data. This is the only preprocessing algorithm used to obtain the results in Table 2. In addition, none of the other preprocessing algorithms were tried, such as baseline correction, the variance scaling (scale to unit variance) or multiplicative scatter correction (MSC) in combination with the PCR method.Except when we applied MSC, we observed that the successive application of these preprocessing algorithms slightly improves the r2 and rmsd values. The multivariate methods were applied to the resolution of the ternary synthetic samples (series 2, Table 1). The results obtained, which are very close to a 100% of recovery in all instances, are discussed below. Resolution of Ternary Mixtures by Applying Second-derivative Ratio Spectra The stored spectra of the ternary mixtures were divided by a standard spectrum of carbofuran (2.5 3 1025 m) obtained under the same conditions.The ratio spectra thus obtained were smoothed through the use of 25 experimental points and the second derivatives were calculated with Dl = 4 nm (selected values once the optimization of instrumental variables had been performed). In Fig. 4, a spectrum of a ternary mixture (a), a ratio spectrum between these and the spectrum of a sample of carbofuran (b) and the send-derivative ratio spectrum (c) are shown as example.Since the absorbance values at the different wavelengths of a dissolution of the three compounds in which the concentrations of carbofuran (A), carbaryl (B) and fenamiphos (C) are cA, cB and cC is AM,li = eA,li c A + eB,licB + eC,licC where b = 1, when we divide this spectrum by the spectrum of a standard dissolution of carbofuran (of concentration c°A), that is, by A°A,li = �°A,licA we obtain AM,li Ao M,li = cA co A + cB co A ÎB,li ÎA,li + cC co A ÎC,li ÎA,li Table 1 Composition of the samples of the calibration set (series 1, samples 1–14) and prediction set (series 2, samples 15–38) Concentration/1026 m Sample Carbofuran Carbaryl Fenamiphos 1 3.00 13.00 13.00 2 5.50 10.50 13.00 3 8.00 8.00 13.00 4 10.50 5.50 13.00 5 13.00 3.00 13.00 6 13.00 8.00 8.00 7 13.00 13.00 3.00 8 10.50 13.00 5.50 9 8.00 13.00 8.00 10 10.50 10.50 8.00 11 8.00 10.50 10.50 12 5.00 5.00 10.00 13 5.00 10.00 5.00 14 9.50 5.50 5.00 15–17 10.00 10.00 6.00 18–20 10.00 6.00 10.00 21–23 6.00 10.00 10.00 24–26 5.00 5.50 5.00 27–29 5.00 7.50 7.50 30–32 7.50 5.00 7.50 33–35 7.50 7.50 5.00 36–38 9.00 5.00 7.00 Fig. 3 First-derivative spectra (Dl = 4 nm) after extraction into CHCl3 (phase ratio 10 : 1, pH 4.7) of ternary mixtures of carbofuran, carbaryl and fenamiphos (series 1). Analyst, June 1997, Vol. 122 515and the obtaining of the second derivative of this spectrum ratio gives d2 dl2 AM,li Ao A,li æ è ç ö ø ÷ = cB co A d2 dl2 ÎB,li ÎA,li æ è ç ö ø ÷ + cC co A d2 dl2 ÎC,li ÎA,li æ è ç ö ø ÷ In this equation we can see that the derivative ratio spectrum is dependent only on c°A, cB and cC.Later the zero-crossing wavelengths of B and C are determined, that is, the wavelength at which one of the two terms in the previous equation takes a zero value. At these wavelengths, the signal measured in the second-derivative ratio spectrum (DD) is dependent only on the concentration of one of the components of the ternary mixture.In our case, using the above-mentioned spectrum as divisor, the amplitude at DD278.10 was proportional to the concentration of carbaryl and the amplitude at DD267.24 was proportional to the concentration of fenamiphos. On the other hand, for determining carbofuran, stored spectra of the mixture were divided by a standard spectrum of carbaryl (5 3 1026 m).The ratio spectra thus obtained, following the same treatment as detailed before, gives the concentration of carbofuran by measuring the amplitude at DD283.77. We would obtain in the same manner the concentration of fenamiphos, but this is unnecessary since it has already been determined. Also, it is unnecessary to divide by a standard spectrum of fenamiphos. The statistical parameters of the regression model obtained are given in Table 3. Comparison Between the Different Chemometric Methods Applied ANOVA was applied to the comparison of two factors, the factor method and the factor component, using as the statistical population the percentage recoveries of the analytes in the synthetic samples.Before applying the ANOVA methods, we applied Bartlett’s criterion, verifying the variance homogeneity. In Table 4 the means of the recovery values ( � R), standard error and confidence interval for a 95% confidence level (IC95) for the mean of each method and each component are shown.The results of the ANOVA are summarized in Table 5. Fig. 4 (a) Spectrum of a ternary mixture of carbofuran (3 3 1026 m), carbaryl (1.3 31025 m) and fenamiphos (1.3 31026 m). (b) Ratio spectrum of (a) and carbofuran (2.5 3 1025 m) as divisor and (c) second-derivative ratio spectrum. Table 2 General characteristics of the matrices Components Carbofuran Carbaryl Fenamiphos Matrix r2 n* rmsd r2 n* rmsd r2 n* rmsd CLS 0.9872 — 0.3536 0.9907 — 0.3152 0.9984 — 0.1377 PCR 0.9904 4 0.3068 0.9934 4 0.2649 0.9986 4 0.1272 PLS-1 0.9963 5 0.1894 0.9976 5 0.1599 0.9987 4 0.1259 PLS-2 0.9961 5 0.1960 0.9978 5 0.1537 0.9984 5 0.1357 * n is the number of factors utilized (see text).Table 3 Statistical data for calibration models (derivative ratio spectra) Parameter Carbofuran Carbaryl Fenamiphos b1 (slope) 992.0 792.0 3667.8 sm (standard deviation 23.35 12.22 32.61 of the slope) b0 (intercept) 4.440 3 1024 27.600 3 1025 1.636 3 1023 sb (standard deviation 2.060 3 1024 1.069 3 1024 4.602 3 1024 of intercept) r (regression 0.9992 0.9996 0.9997 coefficient) s (standard error) 1.862 3 1024 9.661 3 1025 4.159 3 1024 LOD*/m* 6.237 3 1027 4.049 3 1027 3.768 3 1027 * Winefordner and Long method23 with k = 3.Table 4 Mean recoveries for synthetic samples Standard error (%) Level R – (%) (internal) IC95 (%) Method— CLS 100.7333 1.5366 98.8349–102.6324 PCR 101.5958 0.9142 99.6967–103.4949 PLS-1 101.4667 0.6744 99.5676–103.3658 PLS-2 101.2417 0.6640 99.3426–103.1408 DD 97.4250 0.7020 95.5259–99.3241 Components— Carbofuran 99.8625 0.4615 98.3915–101.3335 Carbaryl 100.3525 0.5996 98.8815–101.8235 Fenamiphos 101.2625 0.3641 99.7915–102.7335 516 Analyst, June 1997, Vol. 122The F-test is used to see whether the estimates of variance differ significantly. Comparing the between-component mean squares with the residual mean squares gives F2/113 = 0.916; the critical value23 is 3.079, indicating that there is not a statistically significant difference between the means by the different components at the 95% confidence level.With respect to the factor method, the value obtained is F4/113 = 3.320, the critical value being 2.459 (p = 0.05). Hence there is a statistically significant difference between the methods. Thus, making a multiple range test, it is established that only the mean obtained by the ratio derivative spectra method (DD) cannot be considered equal to those obtained by the other methods, since the differences between them are higher than the calculated limit value24 of 2.6857, for the level of significance established.These differences between the means for the groups CLS–DD, PCR–DD, PLS–1–DD and PLS–2–DD are 3.3083, 4.1708, 4.0417 and 3.8167, respectively. Similar multifactorial analysis of two factors was carried out to compare, on the one hand, the factor transformation of spectra or preprocessing algorithm and, on the other, the factor component, the conclusion being that no significant difference exists in any case.The chemometric methods described above were applied successfully to the analysis of samples of spiked river water according with the above-described procedure. The recoveries were 100–116% for carbofuran, 99–110% for carbaryl and 96–111% for fenamiphos. Conclusions All the methods utilized (CLS, PCR, PLS and DD) can be used in the determination of the three pesticides in their mixtures by using the UV absorption spectra or the first- or secondderivative spectra (CLS, PCR and PLS) or second-derivative ratio spectra (DD) after extraction into CHCl3.However, the easier treatment of the data by using CLS, PCR or PLS makes these preferable to the DD method. On the other hand, one must take into account the better statistical parameters obtained (r2 and rmsd) when the PLS or PCR is used compared with CLS. The data enhancement methods, mean centring and variance scaling, or the other preprocessing algorithm used, baseline correction, give rise to only a small improvement in the prediction ability of the different multivariate calibration methods.However, in the validation of the multivariate methods used with synthetic samples, no significant differences (by application of the ANOVA method) are observed with the different pretreatments of the data used. The authors are grateful to the CICYT (Proyect ALI95-1538) and to the Consejer�ýa de Educaci�on y Juventud de la Comunidad de Extremadura (Project EIA94-35) for financial support. References 1 Salvador, A., De Benzo, Z. A., and De la Guardia, M., Mikroc 1990, 42, 187. 2 Ort�ýz Burguillos, J. M., Tesis Doctoral, University of Extremadura, 1995. 3 Guiberteau A., Galeano D�ýaz, T., Salinas, F., and Ort�ýz, J. M., Anal. Chim. Acta, 1995, 305, 219. 4 Cervantes Oca�na, D., Gil Garc�ýa, M. D., Mart�ýn-Galera, M., and Martinez Vidal, J. L., Bull. Soc. Chim. Belg., 1993, 102, 431. 5 Espinosa-Mansilla, A. Salinas, F., and Zamoro, A., Analyst, 1994, 119, 1183. 6 Salinas, F., Berzas, J. J., and Espinosa Mansilla, A., Talanta, 1990, 37, 347. 7 Berzas Nevado, J. J., Guiberteau Cabanillas, C., and Salinas, F., Talanta, 1992, 39, 547. 8 Vogts, N. B., Chemom. Intell. Lab. Syst., 1989, 7, 119. 9 Wold, S. , Kettaned-Wold, N., and Shagerberg, B., Chemom. Intell. Lab. Syst., 1989, 7, 53. 10 Naes , T., Isaksson, T., and Kowalski, B. R., Anal Chem., 1990, 62, 664. 11 Martens, H., and Naes, T., Multivariate Calibration, Wiley, Chichester, 1989. 12 Thomas, E. V., and Haaland, D. M., Anal. Chem., 1990, 62, 1091. 13 Tauler, R., and Casassas, E., Anal. Chim. Acta, 1989, 223, 257. 14 Haaland, D. M., and Thomas, E. V., Anal. Chem., 1988, 60, 1193. 15 Jones, R., Coomber, T. J., McCormick, J. P., Fell, A. F., and Clark, B. J., Anal. Proc., 1988, 25, 381. 16 MacLaurin, P., Worsfold, P. J., Crane, M., and Norman, P., Anal. Proc., 1992, 29, 65. 17 Dur�an-Mer�as, I., Mu�noz de la Pe�na, A., Espinosa-Mansilla, A., and Salinas, F., Analyst, 1993, 118, 807. 18 Data Leader Software Package, Beckman, Fullerton, CA, 1989. 19 Savitzky, A., and Golay, M. J. E., Anal. Chem., 1964, 36, 1627. 20 Stainer, J., Termonia, Y., and Deltour, J., Anal. Chem., 1972, 44, 1906. 21 GRAMS/386 Software Package, Galactic Industries, Salem, NH, 1991 (web site: www.galactic.com). 22 Stone, W., J. R. Statist. Soc., 1974, B38, 111. 23 Winefordner, J. D., and Long, G. L., Anal. Chem., 1983, 55, 712A. 24 Walpole, R. E., and Myers, R. H., Probabilidad y Estad�ýstica para Ingenieros, Interamericana, M�exico City, 3rd edn., 1990. Paper 6/07955E Received November 25, 1996 Accepted March 11, 1997 Table 5 Results of ANOVA for two factors: ratio of the recoveries obtained by the applied method and the nature of the components Sum of Degrees of Mean Source of variation squares freedom square F-ratio Between method 292.6837 4 73.1709 3.320 Between component 40.3760 2 20.1880 0.916 Residual 2490.7836 113 22.0423 Total 2823.8433 119 Analyst,

ISSN:0003-2654    

DOI:10.1039/a607955e

出版商:RSC    

年代:1997

数据来源: RSC

摘要:

Individual Kinetic Determinations Using Partial Least Squares Calibration Guillermo L�opez-Cuetoa, Jos�e F. Rodr�ýguez-Medinab and Carlos Ubide*b a Departamento de Qu�ýmica Anal�ýtica, Facultad de Ciencias, Universidad de Alicante, Apdo. 99, 03080-Alicante, Spain b Departamento de Qu�ýmica Aplicade, Facultad de Quimica, Universidad del Pa�ýs Vasco, Apdo. 1072, 20080-San Sebasti�an, Spain. E-mail: qapubsec@sq.ehu.es The slow oxidation reactions of cysteine, tyrosine and tryptophan by hexacyanomanganate(IV) were used to investigate partial least squares calibration (PLS) as a method to be applied to the kinetic determination of individual species.All the three reactions are first order in hexacyanomanganate(IV) concentration and the reactions with cysteine and tyrosine are first order in amino acid concentration. The stoichiometry was studied in each case. The determination results were always compared with those obtained by applying the initial rate method to the same kinetic runs.When suitable kinetic curves are obtained and the initial rate is easily determined, both methods give similar results; however, when the initial rate is not easily determined (too fast reaction, slope that changes very quickly, etc.) the results obtained with PLS are better (better precision and wider dynamic range). Moreover, and in contrast to the initial rate method, PLS allowed tryptophan to be determined in the presence of some different feed matrices, provided the calibration set was of the same nature as the samples.This advantage is of general application and makes the analytical method more robust. PLS was tested on uncatalysed reactions but it could equally be applied to catalysed reactions. Keywords: Hexacyanomanganate(IV); cysteine; tyrosine; tryptophan; kinetic determination; partial least squares calibration Kinetic methods of determination are of increasing interest owing to the developement of instrumental methods which are able to follow any change in reaction mixtures and to the advent of high-speed digital computers that allow one to acquire and handle large amounts of data and adopt new approaches.The way to relate the kinetic curve to the concentration of the species in the reaction mixture has traditionally been through some kinetic parameter, which is then considerered as the kinetic signal.1,2 Different approaches can be used to find such a signal, but the most important are (1) the initial reaction rate method (differential), (2) the rate constant method (integral, and only applicable to catalytic determinations if pseudo-order is maintained with time), (3) the fixed-time method, where the change in the instrumental signal value is taken at some previously prefixed time interval and (4) the variable-time method, where the parameter is the time elapsed to obtain some previously prefixed change in the value of the instrumental signal.All of them have advantages and disadvantages; for instance, the initial rate method takes a short time, but the experimental signal must be known during the first stages of the reaction; the rate constant method can be very precise, but most of the kinetic curve is needed and so it takes a longer time; the fixed- and the variabletime methods do not need data handling and can easily be automated, but their precision is very dependent on, among others, the capacity to reproduce the value of the initial experimental signal.In both catalysed and uncatalysed determinations the initialrate method seems to be the most popular; nevertheless, it is difficult to apply when the reaction is too fast (the first stages of reaction are then missed) or when the initial straight part of the kinetic curve is very short (in this case pseudo-zero-order conditions cannot be applied in practice). In both cases the calibration plot will probably be a curved line, because the initial rate method uses only data from the first part of the reaction and ignores the rest; nevertheless, the kinetic profile contains a large amount of information that is not used.In order to use the whole kinetic profile, computer-based methods have been used for processing kinetic data; these methods include curve fitting,3 non-linear regression4 and the Kalman filter.5 The main drawback of all these methods is that some parameters of the system (rate constants, absorptivities, etc.) must be known in advance.This is not the case when some of the multivariate calibration methods6 are used. These latter methods are finding increasing use in analytical work because they do not need a previous knowledge of the kinetic model, rate constants, absorptivities, etc. On the other hand, some of them can be applied not only to linear but also to non-linear systems. In kinetic determinations they have found application especially in the simultaneous determination of multiple components of interest.For instance, binary and ternary mixtures, with a very complex kinetic behaviour, have been resolved by kinetic analysis;7 also, they have been used in order to overcome interferences from species reacting with a general reagent.8 However, it does not seem that they have been used for individual kinetic determinations. Nevertheless, in these cases they still look attractive because a large number of data points can be obtained at different wavelengths and multivariate calibration methods can handle them with the aforementioned advantages; this can open up new possibilities, especially when conventional kinetic methods fail and/or when some complexity from the system can be expected (non-linearity, complex matrices, etc.).This is the reason why partial least squares (PLS) calibration was chosen in this work as an alternative method to the classical initial reaction rate method. In this paper, the oxidation of three amino acids (cysteine, tyrosine and tryptophan) by hexacyanomanganate(iv) in acidic medium is used to show how PLS can be applied, with good results, to individual kinetic determinations, even when the initial rate method gives poor results.Hexacyanomanganate(iv) has a relatively high redox potential of 0.85 V (versus SCE)9 and has sometimes been used as an oxidant for analytical purposes.10 All the results were compared with those obtained by applying the initial rate method to the same kinetic runs.Finally, the possibility of overcoming the effect of a matrix with several interferences at different concentrations (feed samples in this case) by using PLS is considered. Analyst, June 1997, Vol. 122 (519–523) 519Experimental Apparatus A Hewlett-Packard (Avondale, PA, USA) HP 8452A diodearray spectrophotometer with a 1.0 cm pathlength fused-silica cell was used. The cell temperature was controlled (±0.2 °C) with an HP 89090A Peltier temperature control accessory.The solution in the cell was continuously stirred with a small magnetic stirrer incorporated in the Peltier system. All the data were acquired with a Hewlett-Packard HP Vectra 386/25N computer coupled to the spectrophotometer. For pH measurements, a Metrohm (Herisau, Switzerland) Model 716 DMS Titrino was used. All volumes less than 0.1 ml were added with Brand (Wertheim, Germany) micropipettes and all volumes less than 10 ml were added with Eppendorf (Hamburg, Germany) or Gilson (Villiers-le-Bel, France) micropipettes.Reagents All the chemicals were of analytical-reagent grade (Fluka, Buchs, Switzerland or Merck, Darmstadt, Germany) and used as received. Doubly distilled water was used throughout. Potassium hexacyanomanganate(III), K3Mn(CN)6. Prepared according to the method of Lower and Fernelius.11 Spectrophotometric assays by the methods reported previously12 indicated a content of at least 95%.Hexacyanomanganate(IV) solution, 5.2 3 1023 mol l21. Prepared by dissolving 0.0171 g of K3Mn(CN)6 in a dark 10 ml calibrated flask with 0.1 mol l21 perchloric acid previously cooled to 0 °C. The solution was diluted to volume with further 0.1 mol l21 perchloric acid. Before use, the solution must be kept at about 0 °C for about 1 h to allow the MnIII disproportionation reaction to proceed to completion.7 The MnIV solution thus obtained caneriod of 6–8 h if kept at about 0 °C.Cysteine, tyrosine and tryptophan solutions, 1.04 3 1022 mol l21. Prepared by dissolving 0.0303 g of cysteine, 0.0471 g of tyrosine or 0.0531 g of tryptophan in 0.1 mol l21 HClO4 in a 25 ml calibrated flask. The solutions were stored in a refrigerator. From these, working standard solutions ranging from 1.04 3 1025 to 5.20 3 1023 mol l21 were prepared by diluting with 0.1 mol l21 HClO4. Perchloric acid solution, 0.1 mol l21, ionic strength 1.0. Prepared by dissolving 32.87 g of NaClO4·H2O in water and enough perchloric acid to make 250 ml of solution.Computers and Programs The absorbance versus time data from the spectrophotometer were converted with a laboratory-written program, in QUICKBASIC, into an ASCII format to be treated with the UNSCRAMBLER version 5.5 software package (Camo, Trondheim, Norway) which allowed the application of PLS calibration. Multivariate analysis of the data was performed on a 486DX PC computer having 8 Mbyte of RAM.Procedures Amino acid determination Volumes of (2.5 2 x) ml of the perchloric acid solution were placed in a 1 cm cuvette, the PTFE lid was put on and the cell was left to stand for 5 min in the controlled-temperature cell holder of the spectrophotometer (25 °C). A volume between 0.1 and 0.25 ml of the amino acid solution (x ml) was added and the absorbance (A) was set to zero. The cell was removed and 0.1 ml of the hexacyanomanganate(iv) solution was added carefully to avoid mixing, the lid was put on and the cell was then inverted two or three times (this point was taken as zero time). The reaction was followed by monitoring the decrease in the absorbance between 376 and 400 nm (seven wavelengths) every 4 nm and every 1 s for 300 s.These wavelengths correspond to an absorption band of the hexacyanomanganate(iv)13 and neither the amino acids considered here nor their reaction products show absorption in that band. For the determination of tryptophan in feed sample matrices (see below), all data were used, but in all other instances (either initial rates or PLS treatment, for all three amino acids) only the run at 388 nm (maximum of the band) was considered.Because UV radiation decomposes hexacyanomanganate(iv)14 the shutter of the spectrophotometer was ‘on’, that is, no light impinged on the cell unless a measurement was taking place (integration time 0.2 s). The solution in the cell was continuously stirred at 1000 rpm.The initial rate was calculated by applying linear regression analysis to the first few instrumental data points (pseudo-zeroorder conditions). Feed sample treatment Samples containing, among others, several amino acids (lysine, proline, glycine, alanine, arginine, etc.) at different concentrations8 were subjected to a solid–liquid extraction procedure in order to recover the water-soluble matrix: 0.75 g of the feed sample was extracted with 15 ml water for 30 min in a conical flask with magnetic stirring.The extract was centrifugted for 15 min and filtered through a Millipore (Bedford, MA, USA) filterpaper of 0.45 mm pore size. The filtered extract solutions were stored in a refrigerator. Tryptophan determination in feed sample matrices The procedure was the same as in Amino acids determination, but (2.4 2 x) ml of the perchloric acid solution and 0.1 ml of the feed sample extract solution were used (for sheep feed, only 0.05 ml was taken).Results and Discussion Oxidation of Amino Acids. Stoichiometry The oxidation reactions of cysteine (cys), tyrosine (tyr) and tryptophan (trp) depend on both the oxidant and the reaction conditions.15 The stoichiometry of the reaction between each amino acid and MnIV was studied in excess MnIV. Reaction mixtures were followed spectrophotometrically at 388 nm; once the reaction has gone to completion, the lnA–t plot becomes straight because Mn(CN)6 22 slowly decomposes in a pseudofirst- order manner with respect to the cyano complex concentration. 13 The extrapolation of this linear portion to zero time allows the MnIV unreacted with amino acid (Fig. 1) and the Fig. 1 Stoichiometry with excess Mn(CN)6 22. 1, Kinetic run; 2, extrapolation. Initial conditions: Mn(CN)6 22, 1024 mol l21; cysteine, 1025 mol l21; pH, 1; ionic strength, 1; temperature, 50 °C. 520 Analyst, June 1997, Vol. 122stoichiometry of the reaction to be determined. This procedure is similar to the logarithmic extrapolation method for the resolution of mixtures by differential reaction rate methods,16 although here one of the paths for the consumption of the reagent is its own decomposition, and so the result will be approximate.The following results were obtained: 4Mn(CN)6 22 + cys?products 3Mn(CN)6 22 + 2tyr?products 3Mn(CN)6 22 + trp?products The oxidation products were not identified. The stoichiometry was calculated for an initial concentration ratio R = [MnIV]0/ [amino acid]0 ranging between 4 and 10.Kinetics and Initial Reaction Rate Plots Kinetic runs for the reactions between MnIV and cysteine, tyrosine and tryptophan are shown in Fig. 2. In all cases plots of the initial rate, v0, versus the initial concentration of MnIV (for a constant amino acid concentration) were straight, at least between 0.5 and 2.0 3 1024 mol l21 MnIV, establishing a firstorder dependence on [MnIV]. On the other hand, plots of v0 versus [amino acid] (for a constant MnIV initial concentration) were also straight up to 2.0 3 1024 mol l21 cysteine and 1.5 3 1024 mol l21 tyrosine.According to this, and at least during the first stages of reaction, the rate law for the oxidation of cysteine and tyrosine will have the form d[MnIV] rate = = kobs[amino acid][MnIV] (4) dt Subsequently, the reaction order can be disrupted, especially with tyrosine, and the lnA–t plots are no longer straight. The results for tryptophan are different and the v0–[amino acid] plot is curved in the range 0.5 3 1025–5.0 3 1025 mol l21.This last reaction is very fast during the first stages and the initial rate cannot be calculated because when the first point is acquired (3 s) between 20 and 40% of the reaction has been missed; this could be the reason why the calibration plot curves downwards when the first experimental ‘initial rate’ is used. On the other hand, the kinetic profile slows very quickly and the tangent to the curve must be calculated with very few points and consequently with higher imprecision.The v0–[amino acid] plots can be used as calibration plots (in the ranges considered above) and the following regression lines were found, where v0 is in absorbance s21 and parameters are given with standard error values (the calibration equation for tryptophan comes from an empirical fit): cysteine: v0 = (0.14 ± 0.02)31023 + (1.03 ± 0.02)310 [cys] (r = 0.998) tyrosine: v0 = (0.29 ± 0.07)31023 + (2.47 ± 0.08)310 [tyr] (r = 0.997) tryptophan: v0 = (0.6 ± 0.2)31023 + (2.0 ± 0.2)3102[tryp] 2 (2.0 ± 0.4)3106[trp]2 (r = 0.991) It can be said, in general, that all three amino acids can be determined with acceptable precision but with some restrictions, viz., (1) the precision for low tyrosine concentrations (under 5 3 1025 mol l21) is very low, probably because, under these conditions, the initial rate changes very quickly, and (2) the errors for tryptophan tend to be higher than those for cysteine and tyrosine, as could be expected from the higher random errors that affect the parameters of the calibration equation for tryptophan.PLS Calibration PLS is one of the forms of multivariate calibration that has become popular for handling the large amount of data created by some of the new analytical instruments and stored by personal computers.17 PLS has been used in multicomponent analysis to resolve simultaneously mixtures of analytes either by equilibrium or by kinetic-based methods; it is, in fact, a linear calibration method but it has also been used to resolve nonlinear systems by using a large number of principal components; if the non-linearity is not too severe, fairly good results can be obtained and only in cases with a high degree of nonlinearity are some non-linear methods said to give better results.18–20 In this work, we applied PLS mainly as a method that uses the whole set of absorbance data, either at one or at several wavelengths, during the reaction time; this makes it less important whether the first stages of reaction are taken into account or not.On the other hand, PLS can implicitly model some interferences,8 whenever the calibration solutions and samples have similar compositions; this allows the interferences to be overcome without previous separation and makes the method more robust. In these cases PLS becomes a method for determining a single species in a mixture with some interfering species.One of the main features of PLS treatment, in every case, is the choice of the number of principal components or factors (PCs). If the number of PCs is too small a significant part of the variance is not explained and systematic effects can be unaccounted for; if, on the other hand, too many PCs are chosen, overfitting (noise modelling) is produced. The correct number of PCs was calculated here by applying the cross-validation procedure to minimize the error of prediction calculated by means of the square error prediction (SEP): SEP = (Ccalc i=1 nå - Cadd )2 n where Ccalc and Cadd are the calculated (or estimated) and added concentrations of analyte, respectively, for sample i, and n is the number of samples.When the SEP values obtained for each amino acid were plotted versus the number of PCs used for constructing the model, Fig. 3 was obtained. The correct number of PCs is the smallest number whose SEP value does not differ significantly from the minimum value of SEP.The criterion used to find it was that proposed by Haaland and Thomas21 which uses an F-test with a probability P = 0.25. The number of PCs used was three for cysteine and tryptophan and two for tyrosine. Keeping in mind the kinetic profiles (Fig. 2), it is odd that cysteine and tryptophan need the same number of PCs whereas tyrosine needs one PC less; the third PC in the case Fig. 2 Kinetic runs for reactions between Mn(CN)6 22 and 1, 1024 mol l21 cysteine; 2, 8 3 1025 mol l21 tyrosine and 3, 3 3 1025 mol l21 tryptophan.Other conditions as in Fig. 1, except temperature, 25 °C. Analyst, June 1997, Vol. 122 521of cysteine seems to be necessary to model low concentrations properly; one possible explanation is that cysteine does not obey Beer’s law, this could be due to the fact that cysteine, in acidic solution, tends to be oxidized to the dimer cystine, although the spectra of cysteine solutions remained stable for several weeks.Table 1 shows the results obtained by applying PLS calibration to the same runs considered when initial rates were applied. Cysteine and tyrosine gave similar errors when both methods were applied, the only difference being that PLS allows low concentrations of tyrosine to be correctly determined. In no case did either the lack of precision or the type of calibration preclude the correct application of PLS within the range studied (Table 1).Another way of comparing the precisions of the two methods was by determining the amino acid concentration of ten replicates at pre-fixed concetration levels. The results are given in Table 2. No significant difference was found between the standard deviations (F-test, P = 0.05) for cysteine and tyrosine and so the precision was similar for both methods and was always better than 5%. For tryptophan PLS gives a precision significantly better (1.8%) than the initial rate method (7.1%).This is probably due to the inherent lack of precision with which the initial rate is calculated in kinetic runs with a rapidly changing slope. In all cases principal components regression (PCR) gave similar results. On the other hand, the use of several wavelengths, instead of just one, did not produce better results. Tryptophan Determination in Feed Samples One of the theoretical advantages of PLS (but not of PCR) is that it can model interferences whenever they are included in the calibration set. This allows determinations to be made without previous separations, even when the interferences are present in the matrix in variable amounts.17,22 In order to test this possibility, in kinetic determinations, and to compare the results with those obtained by the initial rate method, some feed samples were selected to determine tryptophan.These matrices were chosen because they contain species that can be expected to behave as kinetic interferences in samples where tryptophan can be found and so they can be useful for testing the effectiveness of the proposed method.Kinetic profiles of the matrices used in the absence of extra tryptophan showed that all of them were different. Tryptophan was determined after it had been added to the water-soluble matrices obtained from the feed samples (see Procedures). To obtain the calibration set, different amounts of tryptophan were added to some of the feed sample solutions, trying to include different kind of matrices.In the initial rate method, the calibration results were adjusted to the following straight line: v0 = (0.01 ± 0.22)31023 + (1.11 ± 0.09)3102[trp] (r = 0.979) where v0 is in absorbance s21. PLS needed seven factors (PCs) owing to the existence of a larger number of systematic effects (including matrix interferences) and the kinetics was followed at seven wavelengths (see Procedures); otherwise, higher errors were obtained. The results (together with those obtained from an external set) are given in Table 3.The final total amount was all the time virtually equal to the amount added, indicating that the tryptophan concentration in these samples was originally very low.5 The initial rate method properly predicted pig feed 5 samples (the matrix most often used in calibration) but it usually failed for other matrices. In contrast, the results obtained with PLS were much better, indicating a higher predictive capacity in Fig. 3 SEP versus number of PCs for the PLS model. Table 1 Errors in the calibration and external sets after applying the PLS* calibration method. Concentrations are in mol l21 3 105 PLS Regression Amino acid Taken value Error (%) Cysteine† 1.00§ 1.05 4.6 2.0 2.0 0.5 4.0§ 4.1 3.5 5.0 5.1 2.2 7.0§ 6.8 2.9 8.0 8.0 0.3 10.0§ 9.8 2.9 12.5 12.7 1.8 15.0 15.6 4.2 20.0§ 19.7 1.6 Tyrosine‡ 1.00§ 1.04 4.4 2.0 1.8 7.8 4.0§ 3.9 2.9 5.0 4.8 3.6 7.0§ 7.0 0.6 8.0 7.9 1.0 10.0§ 10.1 1.0 12.0 12.4 3.6 12.5§ 12.8 2.1 15.0§ 14.7 1.7 Tryptophan† 0.50§ 0.50 0.6 0.70§ 0.64 8.0 0.80 0.75 6.3 1.00§ 1.05 5.0 1.5 1.3 14.0 2.0§ 2.0 0.5 2.5 2.5 0.5 3.0§ 3.0 1.3 3.5 3.5 0.6 4.0§ 4.0 0.9 4.5 4.6 1.5 5.0§ 5.0 0.2 * For every run all data were previously mean-centred and autoscaled.† PLS model defined with three factors. ‡ PLS model defined with two factors. § Calibration set. Table 2 Precision in the determination of amino acids using the initial rate and PLS calibration methods (ten replicates). Concentrations are in mol l21 3 105 Initial rate PLS Amino acid Taken Found RSD (%) Found RSD (%) Cysteine 10.0 9.7 3.5 9.6 3.3 Tyrosine 8.0 8.9 3.4 7.9 4.5 Tryptophan 3.00 2.86 7.1 3.04 1.8 522 Analyst, June 1997, Vol. 122the presence of interferences. All this proves that PLS can be a robust calibration method that allows direct kinetic determinations of individual species in matrices with variable amounts of some interferences, whenever they are included in the calibration set, whereas the traditional methods (such as the initial rate method) can not.Conclusions PLS can be used as a calibration method for individual kinetic determinations and no previous knowledge of reaction rates, rate constants, etc., is needed. When the initial slope of the kinetic profile can be easily determined, the initial rate and PLS calibration methods give similar results, but when the initial slope cannot be determined (very fast initial rates) or when the precision is poor (reaction order ill-defined, influence of the products on the reaction kinetics, etc.) PLS gives better results than the initial rate calibration method (better precision and wider dynamic range).Moreover, PLS calibration can overcome some interferences present in variable concentrations if those interferences are included in the calibration set, because they are implicitly modelled. This makes PLS a robust method of calibration for the kinetic determination of individual species.In this work only uncatalysed reactions were considered, but PLS can equally be applied to catalysed reactions. The authors are grateful to Dr. S. Hern�andez-Cassou for providing the feed samples and to Dr. S. Maspoch for helpful discussions. References 1 Mottola, H. A., Kinetic Aspects of Analytical Chemistry, Wiley, New York, 1988. 2 P�erez-Bendito, D., and Silva, M., Kinetic Methods in Analytical Chemistry, Ellis Horwood, Chichester, 1988. 3 Ridder, G. M., and Margerum, D. W., Anal. Chem., 1979, 49, 2090. 4 Cladera, A., G�omez, E., Estela, J. M., Cerd�a, V., and Cerd�a, J. L., Anal. Chim. Acta, 1993, 272, 339. 5 Otto, M., Analyst, 1990, 115, 685. 6 Thomas, E. V., Anal. Chem., 1994, 66. 795A. 7 L�opez-Cueto, G., Maspoch, S., Rodr�ýguez-Medina, J. F., and Ubide, C., Analyst, 1996, 121, 407. 8 Saurina, J., and Hern�andez-Cassou, S., Anal. Chim. Acta, 1993, 281, 593. 9 L�opez-Cueto, G., and Ubide, C., Can. J. Chem., 1991, 69, 2112. 10 L�opez-Cueto, G., and Ubide, C., Talanta, 1990, 37, 849. 11 Lower, J. A., and Fernelius, W. C., Inorg. Synth., 1946, 2, 213. 12 L�opez-Cueto, G., Alonso-Mateos, A., Ubide, C., and del Campo Mart�ýnez, G., Talanta, 1988, 35, 795. 13 L�opez-Cueto, G., and Ubide, C., Can. J. Chem., 1986, 64, 2301. 14 L�opez-Cueto, G., Rodr�ýguez-Medina J. F., and Ubide, C., Talanta, 1996, 43, 2101. 15 Lunec, J., in Encyclopedia of Analytical Science, ed.Townshend, A., Academic Press, San Diego, 1995, p. 3686. 16 Mottola, H. A., Kinetic Aspects of Analytical Chemistry, Wiley, New York, 1988, p. 124. 17 Martens, H., and Naes, T., Multivariate Calibration, Wiley, New York, 1989. 18 Sekulic, S., Seasholtz, M. B.,Wang, Z., and Kowalski, B. R., Anal. Chem., 1993, 65, 835A. 19 Blanco, M., Coello, J., Iturriaga, H., Maspoch, S., and Red�on, M., Anal. Chem., 1995, 67, 4477. 20 Blanco, M., Coello, J., Iturriaga, H., Maspoch, S., Red�on, M., and Villegas, N., Analyst, 1996, 121, 395. 21 Haaland, D. M., and Thomas, E. V., Anal. Chem., 1988, 60, 1193. 22 Computer Methods in UV, Visible and IR Spectroscopy, ed. George, W. O., and Willis, H. A., Royal Society of Chemistry, Cambridge, 1990, p. 55. Paper 7/00718C Received January 31, 1997 Accepted March 19, 1997 Table 3 Errors in the tryptophan concentration when determined in the presence of feed matrices and after applying both the initial rate and PLS* calibration methods.Concentrations are in mol l21 3 105 Initial rate PLS Matrix Added Found Error (%) Found Error (%) Pig Feed 5† 0.50 0.46 8.2 0.49 2.0 Pig Feed 5 0.70 0.74 5.6 0.67 4.3 Pig Feed 5 0.75 0.80 6.3 0.98 31 Pig Feed 5 0.75 0.74 1.9 0.87 16 Pig feed 5† 0.80 0.76 5.4 0.77 3.8 Pig feed 5 1.00 1.02 2.4 1.03 3.0 Pig feed 5† 1.50 1.31 12 1.61 7.3 Pig feed 5† 1.50 1.33 11 1.58 5.3 Pig feed 5 2.00 1.93 3.4 1.94 3.0 Pig feed 5 2.50 2.45 2.1 2.47 1.2 Pig feed 5 3.00 3.00 0.1 3.13 4.3 Pig feed 5† 3.50 3.82 9.1 3.54 1.1 Pig feed 5 4.00 3.71 7.3 4.11 2.8 Pig feed 5† 4.50 4.16 7.6 4.39 2.4 Pig feed 1 2.40 2.36 1.6 1.73 22 Pig feed 1 3.80 1.58 58 3.66 3.7 Pig feed 2 2.80 2.71 3.3 2.20 21 Pig feed 2 3.50 1.92 45 2.97 15 Pig feed 3 1.50 1.30 15 1.38 8.0 Pig feed 3 3.00 2.70 9.9 2.23 26 Pig feed 4 1.00 1.76 76 1.10 10 Pig feed 4† 3.00 2.82 6.1 3.10 3.3 Pig feed 6 2.00 2.63 32 1.96 2.0 Pig feed 6 4.00 4.70 18 3.57 10 Pig feed 7† 2.20 2.37 7.6 2.10 4.5 Pig feed 7 4.20 5.27 26 3.57 15 Sheep feed 1.72 1.13 34 1.78 3.5 Sheep feed 3.30 1.93 42 3.09 6.3 Chicken feed†,‡ 0.80 2.95 269 0.73 8.7 Chicken feed 3.80 5.99 58 3.70 2.6 Horse feed† 2.50 2.98 19 2.52 0.8 Horse feed 3.50 4.33 24 3.31 5.4 * All data were previously autoscaled and the PLS model was defined with seven factors.† Samples used in the calibration set in both methods. ‡ This sample could not be used in the initial rate calibration (see the error column) but it was used in the PLS calibration.Analyst, June 1997, Vol. 122 523 Individual Kinetic Determinations Using Partial Least Squares Calibration Guillermo L�opez-Cuetoa, Jos�e F. Rodr�ýguez-Medinab and Carlos Ubide*b a Departamento de Qu�ýmica Anal�ýtica, Facultad de Ciencias, Universidad de Alicante, Apdo. 99, 03080-Alicante, Spain b Departamento de Qu�ýmica Aplicade, Facultad de Quimica, Universidad del Pa�ýs Vasco, Apdo. 1072, 20080-San Sebasti�an, Spain. E-mail: qapubsec@sq.ehu.es The slow oxidation reactions of cysteine, tyrosine and tryptophan by hexacyanomanganate(IV) were used to investigate partial least squares calibration (PLS) as a method to be applied to the kinetic determination of individual species.All the three reactions are first order in hexacyanomanganate(IV) concentration and the reactions with cysteine and tyrosine are first order in amino acid concentration. The stoichiometry was studied in each case. The determination results were always compared with those obtained by applying the initial rate method to the same kinetic runs. When suitable kinetic curves are obtained and the initial rate is easily determined, both methods give similar results; however, when the initial rate is not easily determined (too fast reaction, slope that changes very quickly, etc.) the results obtained with PLS are better (better precision and wider dynamic range).Moreover, and in contrast to the initial rate method, PLS allowed tryptophan to be determined in the presence of some different feed matrices, provided the calibration set was of the same nature as the samples.This advantage is of general application and makes the analytical method more robust. PLS was tested on uncatalysed reactions but it could equally be applied to catalysed reactions. Keywords: Hexacyanomanganate(IV); cysteine; tyrosine; tryptophan; kinetic determination; partial least squares calibration Kinetic methods of determination are of increasing interest owing to the developement of instrumental methods which are able to follow any change in reaction mixtures and to the advent of high-speed digital computers that allow one to acquire and handle large amounts of data and adopt new approaches.The way to relate the kinetic curve to the concentration of the species in the reaction mixture has traditionally been through some kinetic parameter, which is then considerered as the kinetic signal.1,2 Different approaches can be used to find such a signal, but the most important are (1) the initial reaction rate method (differential), (2) the rate constant method (integral, and only applicable to catalytic determinations if pseudo-order is maintained with time), (3) the fixed-time method, where the change in the instrumental signal value is taken at some previously prefixed time interval and (4) the variable-time method, where the parameter is the time elapsed to obtain some previously prefixed change in the value of the instrumental signal.All of them have advantages and disadvantages; for instance, the initial rate method takes a short time, but the experimental signal must be known during the first stages of the reaction; the rate constant method can be very precise, but most of the kinetic curve is needed and so it takes a longer time; the fixed- and the variabletime methods do not need data handling and can easily be automated, but their precision is very dependent on, among others, the capacity to reproduce the value of the initial experimental signal. In both catalysed and uncatalysed determinations the initialrate method seems to be the most popular; nevertheless, it is difficult to apply when the reaction is too fast (the first stages of reaction are then missed) or when the initial straight part of the kinetic curve is very short (in this case pseudo-zero-order conditions cannot be applied in practice).In both cases the calibration pitial rate method uses only data from the first part of the reaction and ignores the rest; nevertheless, the kinetic profile contains a large amount of information that is not used. In order to use the whole kinetic profile, computer-based methods have been used for processing kinetic data; these methods include curve fitting,3 non-linear regression4 and the Kalman filter.5 The main drawback of all these methods is that some parameters of the system (rate constants, absorptivities, etc.) must be known in advance.This is not the case when some of the multivariate calibration methods6 are used. These latter methods are finding increasing use in analytical work because they do not need a previous knowledge of the kinetic model, rate constants, absorptivities, etc. On the other hand, some of them can be applied not only to linear but also to non-linear systems.In kinetic determinations they have found application especially in the simultaneous determination of multiple components of interest. For instance, binary and ternary mixtures, with a very complex kinetic behaviour, have been resolved by kinetic analysis;7 also, they have been used in order to overcome interferences from species reacting with a general reagent.8 However, it does not seem that they have been used for individual kinetic determinations. Nevertheless, in these cases they still look attractive because a large number of data points can be obtained at different wavelengths and multivariate calibration methods can handle them with the aforementioned advantages; this can open up new possibilities, especially when conventional kinetic methods fail and/or when some complexity from the system can be expected (non-linearity, complex matrices, etc.). This is the reason why partial least squares (PLS) calibration was chosen in this work as an alternative method to the classical initial reaction rate method. In this paper, the oxidation of three amino acids (cysteine, tyrosine and tryptophan) by hexacyanomanganate(iv) in acidic medium is used to show how PLS can be applied, with good results, to individual kinetic determinations, even when the initial rate method gives poor results.Hexacyanomanganate(iv) has a relatively high redox potential of 0.85 V (versus SCE)9 and has sometimes been used as an oxidant for analytical purposes.10 All the results were compared with those obtained by applying the initial rate method to the same kinetic runs.Finally, the possibility of overcoming the effect of a matrix with several interferences at different concentrations (feed samples in this case) by using PLS is considered. Analyst, June 1997, Vol. 122 (519–523) 519Experimental Apparatus A Hewlett-Packard (Avondale, PA, USA) HP 8452A diodearray spectrophotometer with a 1.0 cm pathlength fused-silica cell was used.The cell temperature was controlled (±0.2 °C) with an HP 89090A Peltier temperature control accessory. The solution in the cell was continuously stirred with a small magnetic stirrer incorporated in the Peltier system. All the data were acquired with a Hewlett-Packard HP Vectra 386/25N computer coupled to the spectrophotometer. For pH measurements, a Metrohm (Herisau, Switzerland) Model 716 DMS Titrino was used. All volumes less than 0.1 ml were added with Brand (Wertheim, Germany) micropipettes and all volumes less than 10 ml were added with Eppendorf (Hamburg, Germany) or Gilson (Villiers-le-Bel, France) micropipettes.Reagents All the chemicals were of analytical-reagent grade (Fluka, Buchs, Switzerland or Merck, Darmstadt, Germany) and used as received. Doubly distilled water was used throughout. Potassium hexacyanomanganate(III), K3Mn(CN)6. Prepared according to the method of Lower and Fernelius.11 Spectrophotometric assays by the methods reported previously12 indicated a content of at least 95%.Hexacyanomanganate(IV) solution, 5.2 3 1023 mol l21. Prepared by dissolving 0.0171 g of K3Mn(CN)6 in a dark 10 ml calibrated flask with 0.1 mol l21 perchloric acid previously cooled to 0 °C. The solution was diluted to volume with further 0.1 mol l21 perchloric acid. Before use, the solution must be kept at about 0 °C for about 1 h to allow the MnIII disproportionation reaction to proceed to completion.7 The MnIV solution thus obtained can be used for a period of 6–8 h if kept at about 0 °C.Cysteine, tyrosine and tryptophan solutions, 1.04 3 1022 mol l21. Prepared by dissolving 0.0303 g of cysteine, 0.0471 g of tyrosine or 0.0531 g of tryptophan in 0.1 mol l21 HClO4 in a 25 ml calibrated flask. The solutions were stored in a refrigerator. From these, working standard solutions ranging from 1.04 3 1025 to 5.20 3 1023 mol l21 were prepared by diluting with 0.1 mol l21 HClO4.Perchloric acid solution, 0.1 mol l21, ionic strength 1.0. Prepared by dissolving 32.87 g of NaClO4·H2O in water and enough perchloric acid to make 250 ml of solution. Computers and Programs The absorbance versus time data from the spectrophotometer were converted with a laboratory-written program, in QUICKBASIC, into an ASCII format to be treated with the UNSCRAMBLER version 5.5 software package (Camo, Trondheim, Norway) which allowed the application of PLS calibration.Multivariate analysis of the data was performed on a 486DX PC computer having 8 Mbyte of RAM. Procedures Amino acid determination Volumes of (2.5 2 x) ml of the perchloric acid solution were placed in a 1 cm cuvette, the PTFE lid was put on and the cell was left to stand for 5 min in the controlled-temperature cell holder of the spectrophotometer (25 °C). A volume between 0.1 and 0.25 ml of the amino acid solution (x ml) was added and the absorbance (A) was set to zero. The cell was removed and 0.1 ml of the hexacyanomanganate(iv) solution was added carefully to avoid mixing, the lid was put on and the cell was then inverted two or three times (this point was taken as zero time).The reaction was followed by monitoring the decrease in the absorbance between 376 and 400 nm (seven wavelengths) every 4 nm and every 1 s for 300 s. These wavelengths correspond to an absorption band of the hexacyanomanganate(iv)13 and neither the amino acids considered here nor their reaction products show absorption in that band.For the determination of tryptophan in feed sample matrices (see below), all data were used, but in all other instances (either initial rates or PLS treatment, for all three amino acids) only the run at 388 nm (maximum of the band) was considered. Because UV radiation decomposes hexacyanomanganate(iv)14 the shutter of the spectrophotometer was ‘on’, that is, no light impinged on the cell unless a measurement was taking place (integration time 0.2 s).The solution in the cell was continuously stirred at 1000 rpm. The initial rate was calculated by applying linear regression analysis to the first few instrumental data points (pseudo-zeroorder conditions). Feed sample treatment Samples containing, among others, several amino acids (lysine, proline, glycine, alanine, arginine, etc.) at different concentrations8 were subjected to a solid–liquid extraction procedure in order to recover the water-soluble matrix: 0.75 g of the feed sample was extracted with 15 ml water for 30 min in a conical flask with magnetic stirring.The extract was centrifugted for 15 min and filtered through a Millipore (Bedford, MA, USA) filterpaper of 0.45 mm pore size. The filtered extract solutions were stored in a refrigerator. Tryptophan determination in feed sample matrices The procedure was the same as in Amino acids determination, but (2.4 2 x) ml of the perchloric acid solution and 0.1 ml of the feed sample extract solution were used (for sheep feed, only 0.05 ml was taken).Results and Discussion Oxidation of Amino Acids. Stoichiometry The oxidation reactions of cysteine (cys), tyrosine (tyr) and tryptophan (trp) depend on both the oxidant and the reaction conditions.15 The stoichiometry of the reaction between each amino acid and MnIV was studied in excess MnIV. Reaction mixtures were followed spectrophotometrically at 388 nm; once the reaction has gone to completion, the lnA–t plot becomes straight because Mn(CN)6 22 slowly decomposes in a pseudofirst- order manner with respect to the cyano complex concentration. 13 The extrapolation of this linear portion to zero time allows the MnIV unreacted with amino acid (Fig. 1) and the Fig. 1 Stoichiometry with excess Mn(CN)6 22. 1, Kinetic run; 2, extrapolation. Initial conditions: Mn(CN)6 22, 1024 mol l21; cysteine, 1025 mol l21; pH, 1; ionic strength, 1; temperature, 50 °C. 520 Analyst, June 1997, Vol. 122stoichiometry of the reaction to be determined. This procedure is similar to the logarithmic extrapolation method for the resolution of mixtures by differential reaction rate methods,16 although here one of the paths for the consumption of the reagent is its own decomposition, and so the result will be approximate. The following results were obtained: 4Mn(CN)6 22 + cys?products 3Mn(CN)6 22 + 2tyr?products 3Mn(CN)6 22 + trp?products The oxidation products were not identified. The stoichiometry was calculated for an initial concentration ratio R = [MnIV]0/ [amino acid]0 ranging between 4 and 10.Kinetics and Initial Reaction Rate Plots Kinetic runs for the reactions between MnIV and cysteine, tyrosine and tryptophan are shown in Fig. 2. In all cases plots of the initial rate, v0, versus the initial concentration of MnIV (for a constant amino acid concentration) were straight, at least between 0.5 and 2.0 3 1024 mol l21 MnIV, establishing a firstorder dependence on [MnIV].On the other hand, plots of v0 versus [amino acid] (for a constant MnIV initial concentration) were also straight up to 2.0 3 1024 mol l21 cysteine and 1.5 3 1024 mol l21 tyrosine. According to this, and at least during the first stages of reaction, the rate law for the oxidation of cysteine and tyrosine will have the form d[MnIV] rate = = kobs[amino acid][MnIV] (4) dt Subsequently, the reaction order can be disrupted, especially with tyrosine, and the lnA–t plots are no longer straight.The results for tryptophan are different and the v0–[amino acid] plot is curved in the range 0.5 3 1025–5.0 3 1025 mol l21. This last reaction is very fast during the first stages and the initial rate cannot be calculated because when the first point is acquired (3 s) between 20 and 40% of the reaction has been missed; this could be the reason why the calibration plot curves downwards when the first experimental ‘initial rate’ is used.On the other hand, the kinetic profile slows very quickly and the tangent to the curve must be calculated with very few points and consequently with higher imprecision. The v0–[amino acid] plots can be used as calibration plots (in the ranges considered above) and the following regression lines were found, where v0 is in absorbance s21 and parameters are given with standard error values (the calibration equation for tryptophan comes from an empirical fit): cysteine: v0 = (0.14 ± 0.02)31023 + (1.03 ± 0.02)310 [cys] (r = 0.998) tyrosine: v0 = (0.29 ± 0.07)31023 + (2.47 ± 0.08)310 [tyr] (r = 0.997) tryptophan: v0 = (0.6 ± 0.2)31023 + (2.0 ± 0.2)3102[tryp] 2 (2.0 ± 0.4)3106[trp]2 (r = 0.991) It can be said, in general, that all three amino acids can be determined with acceptable precision but with some restrictions, viz., (1) the precision for low tyrosine concentrations (under 5 3 1025 mol l21) is very low, probably because, under these conditions, the initial rate changes very quickly, and (2) the errors for tryptophan tend to be higher than those for cysteine and tyrosine, as could be expected from the higher random errors that affect the parameters of the calibration equation for tryptophan. PLS Calibration PLS is one of the forms of multivariate calibration that has become popular for handling the large amount of data created by some of the new analytical instruments and stored by personal computers.17 PLS has been used in multicomponent analysis to resolve simultaneously mixtures of analytes either by equilibrium or by kinetic-based methods; it is, in fact, a linear calibration method but it has also been used to resolve nonlinear systems by using a large number of principal components; if the non-linearity is not too severe, fairly good results can be obtained and only in cases with a high degree of nonlinearity are some non-linear methods said to give better results.18–20 In this work, we applied PLS mainly as a method that uses the whole set of absorbance data, either at one or at several wavelengths, during the reaction time; this makes it less important whether the first stages of reaction are taken into account or not.On the other hand, PLS can implicitly model some interferences,8 whenever the calibration solutions and samples have similar compositions; this allows the interferences to be overcome without previous separation and makes the method more robust.In these cases PLS becomes a method for determining a single species in a mixture with some interfering species. One of the main features of PLS treatment, in every case, is the choice of the number of principal components or factors (PCs). If the number of PCs is too small a significant part of the variance is not explained and systematic effects can be unaccounted for; if, on the other hand, too many PCs are chosen, overfitting (noise modelling) is produced.The correct number of PCs was calculated here by applying the cross-validation procedure to minimize the error of prediction calculated by means of the square error prediction (SEP): SEP = (Ccalc i=1 nå - Cadd )2 n where Ccalc and Cadd are the calculated (or estimated) and added concentrations of analyte, respectively, for sample i, and n is the number of samples. When the SEP values obtained for each amino acid were plotted versus the number of PCs used for constructing the model, Fig. 3 was obtained. The correct number of PCs is the smallest number whose SEP value does not differ significantly from the minimum value of SEP. The criterion used to find it was that proposed by Haaland and Thomas21 which uses an F-test with a probability P = 0.25. The number of PCs used was three for cysteine and tryptophan and two for tyrosine. Keeping in mind the kinetic profiles (Fig. 2), it is odd that cysteine and tryptophan need the same number of PCs whereas tyrosine needs one PC less; the third PC in the case Fig. 2 Kinetic runs for reactions between Mn(CN)6 22 and 1, 1024 mol l21 cysteine; 2, 8 3 1025 mol l21 tyrosine and 3, 3 3 1025 mol l21 tryptophan. Other conditions as in Fig. 1, except temperature, 25 °C. Analyst, June 1997, Vol. 122 521of cysteine seems to be necessary to model low concentrations properly; one possible explanation is that cysteine does not obey Beer’s law, this could be due to the fact that cysteine, in acidic solution, tends to be oxidized to the dimer cystine, although the spectra of cysteine solutions remained stable for several weeks.Table 1 shows the results obtained by applying PLS calibration to the same runs considered when initial rates were applied. Cysteine and tyrosine gave similar errors when both methods were applied, the only difference being that PLS allows low concentrations of tyrosine to be correctly determined. In no case did either the lack of precision or the type of calibration preclude the correct application of PLS within the range studied (Table 1).Another way of comparing the precisions of the two methods was by determining the amino acid concentration of ten replicates at pre-fixed concetration levels. The results are given in Table 2. No significant difference was found between the standard deviations (F-test, P = 0.05) for cysteine and tyrosine and so the precision was similar for both methods and was always better than 5%.For tryptophan PLS gives a precision significantly better (1.8%) than the initial rate method (7.1%). This is probably due to the inherent lack of precision with which the initial rate is calculated in kinetic runs with a rapidly changing slope. In all cases principal components regression (PCR) gave similar results. On the other hand, the use of several wavelengths, instead of just one, did not produce better results. Tryptophan Determination in Feed Samples One of the theoretical advantages of PLS (but not of PCR) is that it can model interferences whenever they are included in the calibration set.This allows determinations to be made without previous separations, even when the interferences are present in the matrix in variable amounts.17,22 In order to test this possibility, in kinetic determinations, and to compare the results with those obtained by the initial rate method, some feed samples were selected to determine tryptophan.These matrices were chosen because they contain species that can be expected to behave as kinetic interferences in samples where tryptophan can be found and so they can be useful for testing the effectiveness of the proposed method. Kinetic profiles of the matrices used in the absence of extra tryptophan showed that all of them were different. Tryptophan was determined after it had been added to the water-soluble matrices obtained from the feed samples (see Procedures). To obtain the calibration set, different amounts of tryptophan were added to some of the feed sample solutions, trying to include different kind of matrices.In the initial rate method, the calibration results were adjusted to the following straight line: v0 = (0.01 ± 0.22)31023 + (1.11 ± 0.09)3102[trp] (r = 0.979) where v0 is in absorbance s21. PLS needed seven factors (PCs) owing to the existence of a larger number of systematic effects (including matrix interferences) and the kinetics was followed at seven wavelengths (see Procedures); otherwise, higher errors were obtained.The results (together with those obtained from an external set) are given in Table 3. The final total amount was all the time virtually equal to the amount added, indicating that the tryptophan concentration in these samples was originally very low.5 The initial rate method properly predicted pig feed 5 samples (the matrix most often used in calibration) but it usually failed for other matrices.In contrast, the results obtained with PLS were much better, indicating a higher predictive capacity in Fig. 3 SEP versus number of PCs for the PLS model. Table 1 Errors in the calibration and external sets after applying the PLS* calibration method. Concentrations are in mol l21 3 105 PLS Regression Amino acid Taken value Error (%) Cysteine† 1.00§ 1.05 4.6 2.0 2.0 0.5 4.0§ 4.1 3.5 5.0 5.1 2.2 7.0§ 6.8 2.9 8.0 8.0 0.3 10.0§ 9.8 2.9 12.5 12.7 1.8 15.0 15.6 4.2 20.0§ 19.7 1.6 Tyrosine‡ 1.00§ 1.04 4.4 2.0 1.8 7.8 4.0§ 3.9 2.9 5.0 4.8 3.6 7.0§ 7.0 0.6 8.0 7.9 1.0 10.0§ 10.1 1.0 12.0 12.4 3.6 12.5§ 12.8 2.1 15.0§ 14.7 1.7 Tryptophan† 0.50§ 0.50 0.6 0.70§ 0.64 8.0 0.80 0.75 6.3 1.00§ 1.05 5.0 1.5 1.3 14.0 2.0§ 2.0 0.5 2.5 2.5 0.5 3.0§ 3.0 1.3 3.5 3.5 0.6 4.0§ 4.0 0.9 4.5 4.6 1.5 5.0§ 5.0 0.2 * For every run all data were previously mean-centred and autoscaled.† PLS model defined with three factors. ‡ PLS model defined with two factors.§ Calibration set. Table 2 Precision in the determination of amino acids using the initial rate and PLS calibration methods (ten replicates). Concentrations are in mol l21 3 105 Initial rate PLS Amino acid Taken Found RSD (%) Found RSD (%) Cysteine 10.0 9.7 3.5 9.6 3.3 Tyrosine 8.0 8.9 3.4 7.9 4.5 Tryptophan 3.00 2.86 7.1 3.04 1.8 522 Analyst, June 1997, Vol. 122the presence of interferences. All this proves that PLS can be a robust calibration method that allows direct kinetic determinations of individual species in matrices with variable amounts of some interferences, whenever they are included in the calibration set, whereas the traditional methods (such as the initial rate method) can not.Conclusions PLS can be used as a calibration method for individual kinetic determinations and no previous knowledge of reaction rates, rate constants, etc., is needed. When the initial slope of the kinetic profile can be easily determined, the initial rate and PLS calibration methods give similar results, but when the initial slope cannot be determined (very fast initial rates) or when the precision is poor (reaction order ill-defined, influence of the products on the reaction kinetics, etc.) PLS gives better results than the initial rate calibration method (better precision and wider dynamic range).Moreover, PLS calibration can overcome some interferences present in variable concentrations if those interferences are included in the calibration set, because they are implicitly modelled.This makes PLS a robust method of calibration for the kinetic determination of individual species. In this work only uncatalysed reactions were considered, but PLS can equally be applied to catalysed reactions. The authors are grateful to Dr. S. Hern�andez-Cassou for providing the feed samples and to Dr. S. Maspoch for helpful discussions. References 1 Mottola, H. A., Kinetic Aspects of Analytical Chemistry, Wiley, New York, 1988. 2 P�erez-Bendito, D., and Silva, M., Kinetic Methods in Analytical Chemistry, Ellis Horwood, Chichester, 1988. 3 Ridder, G. M., and Margerum, D. W., Anal. Chem., 1979, 49, 2090. 4 Cladera, A., G�omez, E., Estela, J. M., Cerd�a, V., and Cerd�a, J. L., Anal. Chim. Acta, 1993, 272, 339. 5 Otto, M., Analyst, 1990, 115, 685. 6 Thomas, E. V., Anal. Chem., 1994, 66. 795A. 7 L�opez-Cueto, G., Maspoch, S., Rodr�ýguez-Medina, J. F., and Ubide, C., Analyst, 1996, 121, 407. 8 Saurina, J., and Hern�andez-Cassou, S., Anal. Chim. Acta, 1993, 281, 593. 9 L�opez-Cueto, G., and Ubide, C., Can. J. Chem., 1991, 69, 2112. 10 L�opez-Cueto, G., and Ubide, C., Talanta, 1990, 37, 849. 11 Lower, J. A., and Fernelius, W. C., Inorg. Synth., 1946, 2, 213. 12 L�opez-Cueto, G., Alonso-Mateos, A., Ubide, C., and del Campo Mart�ýnez, G., Talanta, 1988, 35, 795. 13 L�opez-Cueto, G., and Ubide, C., Can. J. Chem., 1986, 64, 2301. 14 L�opez-Cueto, G., Rodr�ýguez-Medina J. F., and Ubide, C., Talanta, 1996, 43, 2101. 15 Lunec, J., in Encyclopedia of Analytical Science, ed.Townshend, A., Academic Press, San Diego, 1995, p. 3686. 16 Mottola, H. A., Kinetic Aspects of Analytical Chemistry, Wiley, New York, 1988, p. 124. 17 Martens, H., and Naes, T., Multivariate Calibration, Wiley, New York, 1989. 18 Sekulic, S., Seasholtz, M. B.,Wang, Z., and Kowalski, B. R., Anal. Chem., 1993, 65, 835A. 19 Blanco, M., Coello, J., Iturriaga, H., Maspoch, S., and Red�on, M., Anal. Chem., 1995, 67, 4477. 20 Blanco, M., Coello, J., Iturriaga, H., Maspoch, S., Red�on, M., and Villegas, N., Analyst, 1996, 121, 395. 21 Haaland, D. M., and Thomas, E. V., Anal. Chem., 1988, 60, 1193. 22 Computer Methods in UV, Visible and IR Spectroscopy, ed. George, W. O., and Willis, H. A., Royal Society of Chemistry, Cambridge, 1990, p. 55. Paper 7/00718C Received January 31, 1997 Accepted March 19, 1997 Table 3 Errors in the tryptophan concentration when determined in the presence of feed matrices and after applying both the initial rate and PLS* calibration methods. Concentrations are in mol l21 3 105 Initial rate PLS Matrix Added Found Error (%) Found Error (%) Pig Feed 5† 0.50 0.46 8.2 0.49 2.0 Pig Feed 5 0.70 0.74 5.6 0.67 4.3 Pig Feed 5 0.75 0.80 6.3 0.98 31 Pig Feed 5 0.75 0.74 1.9 0.87 16 Pig feed 5† 0.80 0.76 5.4 0.77 3.8 Pig feed 5 1.00 1.02 2.4 1.03 3.0 Pig feed 5† 1.50 1.31 12 1.61 7.3 Pig feed 5† 1.50 1.33 11 1.58 5.3 Pig feed 5 2.00 1.93 3.4 1.94 3.0 Pig feed 5 2.50 2.45 2.1 2.47 1.2 Pig feed 5 3.00 3.00 0.1 3.13 4.3 Pig feed 5† 3.50 3.82 9.1 3.54 1.1 Pig feed 5 4.00 3.71 7.3 4.11 2.8 Pig feed 5† 4.50 4.16 7.6 4.39 2.4 Pig feed 1 2.40 2.36 1.6 1.73 22 Pig feed 1 3.80 1.58 58 3.66 3.7 Pig feed 2 2.80 2.71 3.3 2.20 21 Pig feed 2 3.50 1.92 45 2.97 15 Pig feed 3 1.50 1.30 15 1.38 8.0 Pig feed 3 3.00 2.70 9.9 2.23 26 Pig feed 4 1.00 1.76 76 1.10 10 Pig feed 4† 3.00 2.82 6.1 3.10 3.3 Pig feed 6 2.00 2.63 32 1.96 2.0 Pig feed 6 4.00 4.70 18 3.57 10 Pig feed 7† 2.20 2.37 7.6 2.10 4.5 Pig feed 7 4.20 5.27 26 3.57 15 Sheep feed 1.72 1.13 34 1.78 3.5 Sheep feed 3.30 1.93 42 3.09 6.3 Chicken feed†,‡ed 3.80 5.99 58 3.70 2.6 Horse feed† 2.50 2.98 19 2.52 0.8 Horse feed 3.50 4.33 24 3.31 5.4 * All data were previously autoscaled and the PLS model was defined with seven factors. † Samples used in the calibration set in both methods. ‡ This sample could not be used in the initial rate calibration (see the error column) but it was used in the PLS calibration. Analyst, June 1997, Vol. 122 523

ISSN:0003-2654    

DOI:10.1039/a700718c

出版商:RSC    

年代:1997

数据来源: RSC

摘要:

Modulation of the pH in the Determination of Phosphate With Flow Injection and Fourier Transform Infrared Detection R. Vonach, B. Lendl and R. Kellner* Institute for Analytical Chemistry, Vienna University of Technology, Getreidemarkt 9/151, A-1060 Vienna, Austria Flow injection (FI) with FTIR detection is proposed as a versatile technique for the determination of phosphate in aqueous solutions. The modulation of the pH was effected by introducing the appropriate buffer solutions into the sample line of the manifold.Spectral changes of the phosphate vibrations in the region 900–1300 cm21 due to the dissociation or association of hydrogen ions were utilized for the determination of phosphate. Two modes were investigated, the conversion of dihydrogenphosphate (pH 5) into monohydrogenphosphate (pH 10) and into the tertiary phosphate (pH > 13). Calibration graphs of standard solutions were obtained in the concentration range 100–1000 mg l21 (1–10 mmol l21) phosphate.The standard deviation of the method was 2.0 mg l21. The high selectivity of the method was demonstrated by the analysis of six sugar and non-nutritive sweetened soft drink samples containing 350–600 mg l21 phosphate. The average deviation of the results from those given by ion chromatography as an external reference method was 1.7% for non-nutritive sweetened samples and 3% for sugar containing samples. A sample frequency of 60 h21 was obtained. Keywords: Fourier transform infrared spectrometry; flow injection; phosphate determination; soft drink analysis; process analysis The versatility of flow analysis with different kinds of detectors has been impressively demonstrated over the past two decades of use.Despite the high demand for the direct qualitative and quantitative multi-component analysis of organic and inorganic substances, Fourier transform infrared (FTIR) spectrometers have found only limited use as detectors in flow injection (FI) so far. Most applications dealing with FTIR detection in FI use a simple single line manifold as a tool for automated reproducible transport of the sample to the detector.The first papers on such a coupling of FI with FTIR referred to the determination of dioctylsulfosuccinate (DOSS)1 and phenyl isocyanate.2 The determination of choline compounds3 and acetaminophen4 has been successfully performed in strongly absorbing aqueous solutions. Using more IR-transparent organic carriers, commonly hexane, the usefulness of the FI–FTIR approach has been demonstrated with various applications, such as the quantification of o-, m- and p-xylene in xylol5,6 and the simultaneous determination of the three components toluene, methyl tertbutyl ether (MTBE) and benzene in gasolines.7–10 After a dissolution or extraction step with chlorinated hydrocarbons, the determination of ibuprofen,11 acetylsalicylic acid and caffeine,12 and paracetamol13 in pharmaceuticals is feasible, and also oil and greases in spiked water samples.14 Further improvements to decrease the influence of the solvent absorption have been made by solid phase preconcentration for the determination of the pesticide carbaryl15 and its metabolite 1-naphthol.16 The solvent absorption has been completely eliminated by measuring in the gas phase after the evaporation of the sample and introduction with a nitrogen carrier flow into the gas cell. With this approach, the determination of toluene, MTBE and benzene in gasolines17 and of ethanol in beverages18 has been successfully performed.Recently, our group has introduced an FI–FTIR system incorporating enzymic reactions to determine glucose, urea and sucrose in complex matrices. 19,20 A further approach for reactive FI–FTIR is the determination of organic and inorganic acids in aqueous solution with modulation of the pH. Oxygen containing acids show vibration bands in the spectral range 1800–1000 cm21 that change in intensity and wavelength owing to the dissociation or association of hydrogen ions.The spectral change is obtained by introducing selected buffer solutions by means of FI. Since ionic reactions proceed with short time constants and with negligible temperature dependence, the simplicity and robustness of this flow injection manifold are a great advantage over other reactive FI methods. For the automated determination of phosphate there are many different methods available. They are based on a variety of analytical techniques such as FI, ion chromatography (IC), ionselective electrodes and more recently capillary zone electrophoresis (CZE).Atomic emission spectrometry with an inductively coupled plasma (ICP-AES) has also been applied to the determination of total P, but this is an expensive alternative, especially considering a single analyte determination. Phosphate selective electrodes are based on Ag3PO4 or HgHPO4.21,22 However, the poor selectivity over other common anions has not yet been overcome, which excludes them from commercial applications.Enzyme electrodes exhibit low detection limits of 1028 m,23 but they are generally considered too complicated for practical use. The determination of phosphate was one of the first applications of FI presented by Ruzicka and Hansen in 1975,24 with numerous papers following on this topic. This conventional FI approach utilizes the molybdenum blue complex,25–35 or less often the yellow colour of molybdophosphovanadate.36 Both FI and ion chromatography are well established in automated environmental analysis, covering the concentration range from sub-ppb to a few ppm ( < 10 ppm).For process analysis of higher concentrations, an automated dilution step becomes necessary. This is inconvenient and difficult to achieve if a high precision of < 1% is required. The objective of this work was the development of a method that allows the automated determination of phosphate in the range 100–1000 mg l21.In addition, the influence of spectral interferences from sugars were investigated, and also the applicability to process analysis of soft drinks containing sugar and other sweeteners. Experimental Apparatus The flow set-up is shown in Fig. 1. The dual line FI manifold consists of two electrically actuated six-way valves [Valco Analyst, June 1997, Vol. 122 (525–530) 525Instruments (Houston, TX, USA) C22Z-3186EH] and two peristaltic pumps, one for a continuous carrier and reagent flow [Gilson (Worthington, OH, USA) Minipuls 3] and a second one for the automatic filling of the reagent loop [Ismatec (Glattbrugg- Z�urich, Switzerland) MS Reglo].The injection volumes were chosen as 1.5 ml for the sample loop and 100 ml for the reagent loop. Flow rates of 1.1 and 0.055 ml min21 for the sample and the reagent line were achieved by choosing PVC tubing (Tygon) with the appropriate inner diameters of 2 and 0.35 mm, respectively.The high flow ratio (20 : 1) ensures minimum sample dilution and thereby both maximum concentration and maximum precision of the analyte concentration in the flow cell. PTFE tubing of 0.5 and 0.25 mm id was used for all connections. Owing to the short time constant of ionic reactions there was no need for a reaction coil. The connection between the valves and the detection cell was kept as short as possible in order to reduce the axial dispersion of sample and reagents.All IR spectra were obtained on a Bruker (Billerica, MA, USA) IFS 88 FTIR spectrometer equipped with a liquid nitrogen cooled narrow band MCT detector (D* = 2 3 1010 cm Hz1/2 W21). A transmission cell with 25 mm pathlength and CaF2 windows (of 2 mm thickness) was applied, yielding an approximately 1/e attenuation (absorbance approximately 0.4) of the water background absorption at 1100 cm21. Transmission cells with ZnSe windows and other optical pathlengths (50 mm) and the ATR mode (45° ZnSe crystal, 25 internal reflections) were also investigated, leading to a lower S/N for measurements in aqueous solutions in the spectral region 1400–1000 cm21.The introduction of an InSb low wave pass filter (5% cut-on 1370 cm21) into the optical set-up provides a further approximately 3–4-fold increase in the S/N in the spectral region invesgated. Further details on the optical filtering were given in a previous paper.37 Spectra with satisfactory S/N were obtained by recording and co-adding 100 scans at a resolution of 4 cm21 in less than 15 s.A laboratory-built electronic interface was used for the synchronization of the pumps, the valves, the FTIR control and data acquisition, allowing completely automated realization of the measurement cycle. Reagents The phosphate solutions for recording the spectra of the individual dissociation states were prepared from analyticalreagent grade NaH2PO4·H2O and Na2HPO4·2H2O, purchased from Merck (Darmstadt, Germany).The solutions for the [H3PO4] and [PO4]32 spectra were prepared by acidifying with HCl and adding NaOH, respectively. Phosphate calibration solutions were prepared by dissolving the appropriate amount of sodium dihydrogenphosphate (NaH2PO4·H2O, analytical-reagent grade, guaranteed concentration 99–102%; Merck) in distilled water. The accuracy and long term stability of the stock solution was confirmed by reference measurements of industrially prepared phosphate solutions (Titriplex; Merck).An acetate buffer solution containing 100 mmol l21 acetic acid and 300 mmol l21 sodium acetate was used for adjusting the pH to approximately 5. The sodium carbonate buffer solution for adjusting the pH to approximately 10–10.5 was prepared with 100 mmol l21 sodium hydrogencarbonate and 600 mmol l21 sodium carbonate. A solution of 2.6 mol l21 NaOH was used for adjusting the pH to > 13.Glucose, fructose and sucrose solutions for interference studies were prepared by dissolving 80 g of each compound (concentration > 99%, for biochemical use; Merck) in 1 l water. The liquids were stabilized with 50 mg l21 NaN3 and stored for 48 h before use in order to prevent mutarotational effects. The industrial samples studied were various soft drinks containing sugar and non-nutritive sweeteners. Prior to analysis the samples were degassed in an ultrasonic bath as the only sample preparation step.Procedure Quantitative analysis in reactive FI–FTIR is based on the evaluation of the difference signal of the analyte before and after the FI-induced chemical or biochemical reaction. As the IR spectrum changes significantly owing to the ionic reaction, distinct features of the difference spectrum can be used for quantification. In contrast to most FI applications, the axial dispersion was kept as small as possible (D ? 1), not just at the maximum of the FI peak, but during the whole recording time of the spectra.This required a comparatively high sample volume of 1.5 ml but ensured both maximum sensitivity and minimum variation of the dispersion. The pH of the injected sample was adjusted to 5 after mixing with the acetate buffer, which was supplied by the reagent line. After a lag phase of 15 s the first spectrum (reference spectrum) of the sample was recorded. By switching the reagent valve a plug of a carbonate buffer solution or sodium hydroxide solution was inserted into the reagent line.Another 15 s of equilibration were necessary to reach the desired pH of 10 or > 13 in the flow cell and the second (sample) spectrum was recorded. Subsequently both valves were switched back to the loading position and the system was ready for the next run. The duration of a complete measurement cycle was 60 s. Reference measurements were performed by ion chromatography with a standard anionic chromatographic set-up and a conductivity detector subsequent to a 1 + 99 dilution.Results and Discussion Water Absorption and its pH Dependence In aqueous solution, large parts of the mid-IR region are not accessible for quantitative IR spectrometry, hence absorption of the analyte has to be well apart from the OH stretching band n1,3 (around 3200–3400 cm21), OH bending band n2 (approximately 1640 cm21) and the libration band nL (around 750 cm21) of liquid water.Although the spectral region investigated is located between n2 and nL, a large unspecific background absorption in the range 900-1500 cm21 also has to be taken into consideration. This background remains fairly constant in the pH range 2–12, but both high H+ and OH2 concentrations influence the water background spectrum (Fig. 2). The observed spectral change is proportional to the concentrations of H+ and Fig. 1 FI–FTIR set up. 526 Analyst, June 1997, Vol. 122OH2.At low pH a broad band with a maximum at 1200 cm21 is formed, which is assigned to the symmetric bending vibration ds of the [H3O]+ ion.38 In alkaline media a baseline shift is visible, which we assume is due to an unspecific change in the water background absorption. Phosphate Absorption The water corrected IR spectra of the individual dissociation states of phosphate are shown in Fig. 3. In case of [H3PO4] and [PO4]32, water of the same pH (1 and 13.5, respectively) was used as the background spectrum.Since the water background absorption increases in the region below 900 cm21 due nL, the bands at lower wavenumbers cannot be used for quantitative analysis. The wavenumbers, the widths [full width at half maximum (FWHM)] and the extinction coefficients (e) of the IR-active phosphate stretching vibrations were extracted from the above spectra and are listed in Table 1. The available literature data on the absorption maxima are additionally presented for comparison.The symmetry of the tertiary phosphate [PO4]32 is completely Td and because of this it has only two IR-active bands, n3 (asymmetric stretch) and n4 (asymmetric deformation) at 1017 and 567 cm21 according to the literature.39,40 Although n4 is not accessible with our experimental set-up, the peak at 1009 cm21 shows good agreement with the literature data. The symmetry of [HPO4]22 in aqueous solution can be seen simplified as a type ZXY3 molecule having C3v symmetry. The bands at 980 and 1080 cm21 correspond to the literature data for ns(PO3) and nas(PO3) of 991 and 1080 cm21, whereas n(POH) is beyond the measurement range at 870 cm21.40 The symmetry of [H2PO4]2 corresponds to the C2v symmetry of a simplified ZX2Y2 molecule.Three of the four stretching vibrations are above 900 cm21 (Fig. 3) with the most intensive nsPO2 at 1078 cm21. No literature data concerning the assignment of the IR absorption of phosphoric acid were available, but the 1178 cm21 vibration is in the range of the PNO stretching vibration of organophosphorus compounds, which is located between 1140 and 1380 cm21.41 The other band at 1010 cm21 is then assigned to the asymmetric P(OH)3 stretching vibration.Modulation of pH The pH modulation was performed in two different modes. First the change from pH 5 to > 13 was investigated. On converting [H2PO4]2 into [PO4]32, the most intensive phosphate vibration n3([PO4]32) is feasible for the phosphate determination.Subtracting the first spectrum ([H2PO4]2) from the second ([PO4]32), a difference spectrum is obtained displaying positive [PO4]32 (1009 cm21) and the negative [H2PO4]2 bands (1157, 1078 cm21). Additionally, a significant baseline shift due to the pH > 13 of the [PO4]32 solution was observed. The acetate buffer of the [H2PO4]2 solution has its absorption bands (CNO and COO2) mainly above 1400 cm21 and does not interfere with the phosphate absorption below 1200 cm21.The second case studied was the conversion of [H2PO4]2 into [HPO4]22 on changing the pH from 5 to 10. Here the specific spectral change of phosphate is smaller. First, the extinction coefficient of the [HPO4]22 band at 1080 cm21 is smaller than that of the corresponding [PO4]32 vibration. Second, the bands of both [H2PO4]2 and [HPO4]22 are almost at the same wavelength and, as a result, the difference spectrum only displays the enlargement of the [H2PO4]2 band at 1078 cm21 to the [HPO4]22 band at 1080 cm21.On the other hand, no interference of the pH on the absorption spectra occurs. The absorption of the carbonate buffer, with its maximum at approximately 1380 cm21, does not influence the phosphate absorption, which makes that method free from reagent interferences. Additionally, the spectral noise at 1090 cm21 ([H2PO4]2 ? [HPO4]22) is less than that at 1004 cm21 ([H2PO4]2 ? [PO4]32), which is due to the increasing water and CaF2 background absorption. Calibration Eight calibration solutions containing 0–1000 mg l21 phosphate were analysed with six repetitions each.Selected difference spectra are shown in Fig. 4 for the alkaline method (pH5 ? > 13) and in Fig. 5 for the carbonate buffer method (pH 5 ? 10). Various evaluation methods were investigated, leading to Fig. 2 Spectral change of the water background absorption at very low and high pH. Spectra a, b and c refer to pH 1.5, 1.0 and 0.5 and spectra d, e and f to pH 12.5, 13.0 and 13.5.Fig. 3 Vibrational spectra of the individual dissociation states of phosphate: a, [H3PO4] (pH 1); b, [H2PO4]2 (pH 5); c, [HPO4]22 (pH 10); d, [PO4]32 (pH > 13). All concentrations 50 mmol l21. Table 1 IR-active stretching vibrations of phosphate Literature Width e/ Maximum maximum (FWHM) l mol21 Species Assignment40 n/cm21 n/cm21 cm21 cm21 [PO4]32 n3 1009 101739 52 1700 [HPO4]22 nas(PO3) 1080 108040 46 1110 ns(PO3) 991 98040 19 470 n(POH) —* 87040 — — [H2PO4]2 nas(PO2) 1159 115040 65 360 ns(PO2) 1078 107040 23 610 nas[P(OH)2] 941 94540 37 310 ns[P(OH)2] —* 88040 — — [H3PO4] n(PNO)† 1178 — 60 190 nas[P(OH)3]† 1010 — 48 430 * Not within the experimentally accessible wavelength range of 1400–900 cm21.† Proposed by the present authors. Analyst, June 1997, Vol. 122 527the best results by using two baseline points for the alkaline method and one baseline point for the carbonate buffered method.In order to reduce the spectral noise, the peak maximum and the baseline points were averaged within a 10–20 wavenumber range. The calibration parameters and results are given in Table 2. The precision of both methods was evaluated by the standard deviation of the method (sx0), calculated as the ratio of the residual standard deviation (sy) and the slope (b). In both cases exact linearity was obtained with r2 > 0.9999. Despite the twofold higher sensitivity of the alkaline method (slope of the calibration curve), sx0 is about equal for both methods.First, the stronger water absorption causes higher spectral noise at lower wavenumbers. Second, the variation in the baseline shift at pH > 13 induces an additional imprecision. Both methods exhibit a repeatability of 0.5% for a concentration of 500 mg l21, which makes applications in process analysis feasible. Interference Study The range 900–1200 cm21 is part of the fingerprint region with interfering absorption bands of numerous compounds.In this study, we focused on the interference of common sugars, such as sucrose, glucose and fructose, as they occur in the food industry. In order to do so, solutions of 80 g l21 of each compound (234 mmol l21 sucrose, 444 mmol l21 glucose and fructose) were prepared and treated by FI–FTIR with modulation of the pH. The acidic properties of sugars in concentrated NaOH solution are well known.42 According to Urban and Shaffer,43 the first apparent dissociation constants, pKA1, are 11.7 for fructose, 12.1 for glucose and 12.6 for sucrose.(In the apparent dissociation constant the activity coefficient of the salt is included.) The spectral changes in the alkaline method [Fig. 6(a)] are therefore proposed to be due to the dissociation reactions of the individual sugars. The spectra exhibit spectral changes partially higher than 0.1. These features are one order of magnitude higher than the phosphate specific absorption change, precluding a phosphate determination by modulation of the pH with sodium hydroxide.Concerning the modulation of pH with carbonate buffer, comparatively weak sugar specific absorption changes were observed [Fig. 6(b)]. Besides the absorption of the carbonate buffer (approximately 1380 cm21), sucrose and glucose show very small absorption changes. The difference spectrum of fructose has the strongest spectral change (0.008), but this is still just one twentieth of that in sodium hydroxide method.The fructose difference spectra of both pH modulations are qualitatively similar, which indicates that the dissociation of Fig. 4 Difference spectra of phosphate calibration solutions obtained by pH modulation with sodium hydroxide (pH 5 ? > 13). Fig. 5 Difference spectra of phosphate calibration solutions obtained by pH modulation with carbonate buffer (pH 5 ? 10). Table 2 Calibration parameters and results Modulation mode pH 5? > 13 pH5?10 Phosphate concentration/mg l21 0, 100, 200, 400, 500, 600, 800, 1000 No.of repetitions 6 6 Evaluated peak, n/cm21 999–1009 1085–1095 Baseline point(s), n/cm21 1072–1082 1155–1165 930–950 Slope, b/mg21 l 5.1531025 2.4831025 Intercept, a 1.131023 23.831024 Regression coefficient, r2 0.99994 0.99996 Residual standard deviation, sy 1.331024 4.931025 Standard deviation of the method, sx0/mg l21 2.6 2.0 Fig. 6 Spectral changes obtained by pH modulation of common sugars with (a) sodium hydroxide (pH 5 ? pH > 13) and (b) carbonate buffer solutions (pH 5 ?10).Spectra 1, 2 and 3 refer to 80 g l21 sucrose, glucose and fructose. For better visualization an offset of 0.01 and 0.1 was applied for each spectrum of the sodium hydroxide and the carbonate buffer method, respectively. Note the difference in the absorbance scale. 528 Analyst, June 1997, Vol. 122fructose (pKA1 = 11.7) begins partially at the pH of the carbonate buffer (approximately pH 10.5).Real Samples Phosphate in industrial samples was determined by the proposed FI–FTIR method. Six samples of various soft drinks, three non-nutritive and three sugar-containing sweeteners, were examined. The selectivity enhancement due to the modulation of the pH is demonstrated in Figs. 7 and 8. Whereas other compounds such as the non-nutritive sweeteners (acesulfame K, cyclamate) and the coloring matter interfere with the phosphate band,44 the difference spectrum with pH modulation is reduced to phosphate specific absorption features.At sugar concentrations of approximately 100 g l21 the phosphate bands are masked by the sugar absorption of > 0.4. However, in the 20-fold enlarged difference spectrum obtained with pH modulation, the almost interference-free phosphate specific absorption change becomes clearly visible. The results of the FI–FTIR analyses of real samples are given in Table 3. Ion chromatography was chosen as an external reference method.Concerning the non-nutritive sweetened samples, both the alkaline and the carbonate buffered method are in satisfactory accordance with the reference method, with an average deviation of 1.7% from the reference values. The sugar-containing samples were analysed by the carbonate buffered pH modulation method, with higher deviations of 1.2–5.3%. The repeatability (standard deviation of multiple injections) was in all cases better than the accordance with the reference method, namely 0.5% for the non-nutritive sweetened samples and 1.6% for sugar-containing samples.Conclusion From the results presented in this paper, it is concluded that the modulation of the pH by means of FI–FTIR is a promising technique for the selective, rapid and automated determination of 100–1000 mg l21 phosphate in aqueous solutions. The proposed ionic reaction fulfils the basic requirements for a fast and reproducible analysis using reactive FI–FTIR.Hence both a high sample throughput (60 h21) and a high precision ( < 1%) are feasible. A significant selectivity enhancement in FTIR was achieved by acquiring a difference spectrum of pH-dependent spectral features. Furthermore, sensitivity enhancement was achieved by choosing the pH of the appropriate dissociation state of the analyte. Oxygen-containing acids, especially organic acids, have stretching vibrations in the 1800–1000 cm21 region, which change in wavelength and intensity owing to the dissociation or association of hydrogen ions.Further experiments will be Fig. 7 Spectra of a non-nutritive sweetened real world sample containing 500 mg l21 phosphate at pH 5 (A) and 10 (B). The interfering absorptions (1040, 1170 cm21) are eliminated in the difference spectrum (C), allowing the quantification of phosphate analogous to the calibration solutions (Fig. 6). An offset of 0.01 was chosen for (B) and 0.03 for (C) for better visualization.Table 3 Phosphate concentrations of soft drinks determined by FI–FTIR and ion chromatography as an external reference method. The phosphate concentrations and standard deviations are expressed in mg l21 and the recoveries as % of the reference value FI–FTIR (pH modulation) Reference method pH 5?10 pH 5? > 13 (ion chromatography) Concentration* Recovery Concentration* Recovery Concentration Non-nutritive sweetened samples— A 498.2 ± 1.7 100.5 485.8 ± 2.8 98.0 495.9 ± 3.3 B 364.4 ± 1.7 103.4 345.9 ± 2.5 98.1 352.5 ± 3.3 C 453.9 ± 2.3 101.3 445.5 ± 2.5 99.4 448.3 ± 3.3 Sugar-containing samples— D 571 ± 7 94.7 —† —† 602.5 ± 3.3 E 492 ± 8 98.8 —† —† 497.8 ± 3.3 F 535 ± 8 102.4 —† —† 522.2 ± 3.3 * The standard deviation was calculated from six repetitions.† No available data because of sugar dissociation upon pH modulation. Fig. 8 Spectra of a sugar-containing (approximately 100 g l21) real sample at pH 5 (A) and 10 (B). Whereas the phosphate bands are hidden in the strong sugar absorbance, the phosphate specific absorption change is visible in the difference spectrum (C).Analyst, June 1997, Vol. 122 529carried out in order to investigate the potential of pH modulation for the determination of other ionic compounds. The authors gratefully thank the Forschungsf�orderungsfonds f�ur die gewerbliche Wirtschaft and the Fonds zur F�orderung der wissenschaftlichen Forschung, P11338 � OCH, for support of this work.References 1 Morgan, D. K., Danielson, N. D., and Katon, J. E., Anal. Lett., 1985, 18(A16), 1979. 2 Curran, D. J., and Collier, W. G., Anal. Chim. Acta, 1985, 177, 259. 3 Miller, B. E., Danielson, N. D., and Katon, J. E., Appl. Spectrosc., 1988, 42, 401. 4 Ramos, M. L., Tyson, J. F., and Curran, D. J., Anal. Proc., 1995, 32, 175. 5 de la Guardia, M., Garrigues, S., and Gallignani, M., Anal. Chim. Acta, 1992, 261, 53. 6 Garrigues, S., Gallignani, M., and de la Guardia, M., Analyst, 1992, 117 1849. 7 Gallignani, M., Garrigues, S., and de la Guardia, M., Anal. Chim. Acta, 1993, 274, 267. 8 de la Guardia, M., Gallignani, M., and Garrigues, S., Anal. Chim. Acta, 1993, 282, 543. 9 Gallignani, M., Garrigues, S., de la Guardia, M., Burguera, J. L., and Burguera, M., Talanta, 1994, 41, 739. 10 Gallignani, M., Garrigues, S., and de la Guardia, M., Analyst, 1994, 119, 653. 11 Garrigues, S., Gallignani, M., and de la Guardia, M., Talanta, 1993, 40, 89. 12 Garrigues, S., Gallignani, M., and de la Guardia, M., Talanta, 1993, 40, 1799. 13 Bouhsain, Z., Garrigues, S., and de la Guardia, M., Analyst, 1996, 121, 635. 14 Daghbouche, Y., Garrigues, S., and de la Guardia, M., Analyst, 1996, 121, 1031. 15 Garrigues, S., Vidal, M. T., Gallignani, M., and de la Guardia, M., Analyst, 1994, 119, 659. 16 Daghbouche, Y., Garrigues, S., and de la Guardia, M., Anal. Chim. Acta, 1995, 314, 203. 17 L�opez-Anreus, E., Garrigues, S., and de la Guardia, M., Anal.Chim. Acta, 1996, 333, 157. 18 P�erez-Ponce, A., Garrigues, S., and de la Guardia, M., Analyst, 1996, 121, 923. 19 Rosenberg, E., and Kellner, R., J. Mol. Struct., 1993, 294, 9. 20 Lendl, B., and Kellner, R., Mikrochim. Acta, 1995, 119, 73. 21 Vermes, I., and Grabner, E. W., J. Electroanal. Chem., 1990, 284, 315. 22 Encyclopedia of Analytical Science, ed. Townshend, A., Academic Press, New York, 1995, pp. 3957–3960. 23 Conrath, N., Gr�undig, B., H�uwel, St., and Cammann, K., Anal.Chim. Acta, 1995, 309, 47. 24 Ruzicka, J., and Hansen, E. H., Anal. Chim. Acta, 1975, 78, 145. 25 Sun, F., and Korenaga, T., Appl. Spectrosc., 1996, 50, 1145. 26 McKelvie, I. D., Peat, D. M. W., and Worsfold, P. J., Anal. Proc., 1995, 32, 437. 27 Woo, L., and Maher, W., Anal. Chim. Acta, 1995, 315, 123. 28 Agudo, M., R�ýos, A., and Valc�arcel, M., Trends Anal. Chem., 1994, 13, 409. 29 Herrero, M. A., Atienza, J., Maquieira, A., and Puchades, R., Analyst., 1992, 117, 1019. 30 Linares, P., Luque de Castro, M. D., and Valc�arcel, M., Anal. Chem., 1986, 58, 120. 31 Janse, T. A. H. M., Van der Wiel, P. F. A., and Kateman, G., Anal. Chim. Acta, 1983, 155, 89. 32 Motomizu, S., Wakimoto, T., and T�oei, K., Talanta, 1983, 30, 333. 33 Johnson, K. S., and Petty, R. L., Anal. Chem., 1982, 54, 1185. 34 Hirai, Y., Yoza, N., and Ohashi, S., Anal. Chim. Acta, 1980, 115, 269. 35 Hansen, E. H., and Ruzicka, J., Anal. Chim. Acta, 1976, 87, 353. 36 Røyset, O., Anal.Chim. Acta, 1985, 178, 217. 37 Krieg, P., Lendl, B., Vonach R., and Kellner, R., Fresenius’ J. Anal. Chem., 1996, 356, 504. 38 Falk, M., and Grigu`ere, P. A., Can. J. Chem., 1957, 35, 1195. 39 Nakamoto, K., Infrared and Raman Spectra of Inorganic and Coordination Compounds, Wiley, New York, 4th edn., 1986. 40 Steger, E., and Herzog, K., Anorg. Allg. Chem., 1964, 331, 169. 41 Colthup, N. B., and Daly, L. H., Introduction to Infrared and Raman Spectroscopy, Academic Press, Boston, 3rd edn., 1990. 42 Kilde, G., and Wynne-Jones, W. F. K., Trans. Faraday Soc., 1953, 49, 243. 43 Urban, F., and Shaffer, P., J. Biol. Chem., 1932, 94, 697. 44 Vonach, R., Kellner, R., and Lippitsch, M., in Proceedings of EURO FOOD CHEM VIII, ed. Sontag, G., and Pfannhauser, W., Austrian Chemical Society, Vienna, 1995, p. 573. Paper 6/08540G Received December 23, 1996 Accepted February 10, 1997 530 Analyst, June 1997, Vol. 122 Modulation of the pH in the Determination of Phosphate With Flow Injection and Fourier Transform Infrared Detection R.Vonach, B. Lendl and R. Kellner* Institute for Analytical Chemistry, Vienna University of Technology, Getreidemarkt 9/151, A-1060 Vienna, Austria Flow injection (FI) with FTIR detection is proposed as a versatile technique for the determination of phosphate in aqueous solutions. The modulation of the pH was effected by introducing the appropriate buffer solutions into the sample line of the manifold.Spectral changes of the phosphate vibrations in the region 900–1300 cm21 due to the dissociation or association of hydrogen ions were utilized for the determination of phosphate. Two modes were investigated, the conversion of dihydrogenphosphate (pH 5) into monohydrogenphosphate (pH 10) and into the tertiary phosphate (pH > 13). Calibration graphs of standard solutions were obtained in the concentration range 100–1000 mg l21 (1–10 mmol l21) phosphate. The standard deviation of the method was 2.0 mg l21.The high selectivity of the method was demonstrated by the analysis of six sugar and non-nutritive sweetened soft drink samples containing 350–600 mg l21 phosphate. The average deviation of the results from those given by ion chromatography as an external reference method was 1.7% for non-nutritive sweetened samples and 3% for sugar containing samples. A sample frequency of 60 h21 was obtained. Keywords: Fourier transform infrared spectrometry; flow injection; phosphate determination; soft drink analysis; process analysis The versatility of flow analysis with different kinds of detectors has been impressively demonstrated over the past two decades of use.Despite the high demand for the direct qualitative and quantitative multi-component analysis of organic and inorganic substances, Fourier transform infrared (FTIR) spectrometers have found only limited use as detectors in flow injection (FI) so far. Most applications dealing with FTIR detection in FI use a simple single line manifold as a tool for automated reproducible transport of the sample to the detector. The first papers on such a coupling of FI with FTIR referred to the determination of dioctylsulfosuccinate (DOSS)1 and phenyl isocyanate.2 The determination of choline compounds3 and acetaminophen4 has been successfully performed in strongly absorbing aqueous solutions.Using more IR-transparent organic carriers, commonly hexane, the usefulness of the FI&been demonstrated with various applications, such as the quantification of o-, m- and p-xylene in xylol5,6 and the simultaneous determination of the three components toluene, methyl tertbutyl ether (MTBE) and benzene in gasolines.7–10 After a dissolution or extraction step with chlorinated hydrocarbons, the determination of ibuprofen,11 acetylsalicylic acid and caffeine,12 and paracetamol13 in pharmaceuticals is feasible, and also oil and greases in spiked water samples.14 Further improvements to decrease the influence of the solvent absorption have been made by solid phase preconcentration for the determination of the pesticide carbaryl15 and its metabolite 1-naphthol.16 The solvent absorption has been completely eliminated by measuring in the gas phase after the evaporation of the sample and introduction with a nitrogen carrier flow into the gas cell. With this approach, the determination of toluene, MTBE and benzene in gasolines17 and of ethanol in beverages18 has been successfully performed.Recently, our group has introduced an FI–FTIR system incorporating enzymic reactions to determine glucose, urea and sucrose in complex matrices. 19,20 A further approach for reactive FI–FTIR is the determination of organic and inorganic acids in aqueous solution with modulation of the pH. Oxygen containing acids show vibration bands in the spectral range 1800–1000 cm21 that change in intensity and wavelength owing to the dissociation or association of hydrogen ions.The spectral change is obtained by introducing selected buffer solutions by means of FI. Since ionic reactions proceed with short time constants and with negligible temperature dependence, the simplicity and robustness of this flow injection manifold are a great advantage over other reactive FI methods. For the automated determination of phosphate there are many different methods available. They are based on a variety of analytical techniques such as FI, ion chromatography (IC), ionselective electrodes and more recently capillary zone electrophoresis (CZE).Atomic emission spectrometry with an inductively coupled plasma (ICP-AES) has also been applied to the determination of total P, but this is an expensive alternative, especially considering a single analyte determination. Phosphate selective electrodes are based on Ag3PO4 or HgHPO4.21,22 However, the poor selectivity over other common anions has not yet been overcome, which excludes them from commercial applications.Enzyme electrodes exhibit low detection limits of 1028 m,23 but they are generally considered too complicated for practical use. The determination of phosphate was one of the first applications of FI presented by Ruzicka and Hansen in 1975,24 with numerous papers following on this topic. This conventional FI approach utilizes the molybdenum blue complex,25–35 or less often the yellow colour of molybdophosphovanadate.36 Both FI and ion chromatography are well established in automated environmental analysis, covering the concentration range from sub-ppb to a few ppm ( < 10 ppm). For process analysis of higher concentrations, an automated dilution step becomes necessary.This is inconvenient and difficult to achieve if a high precision of < 1% is required. The objective of this work was the development of a method that allows the automated determination of phosphate in the range 100–1000 mg l21.In addition, the influence of spectral interferences from sugars were investigated, and also the applicability to process analysis of soft drinks containing sugar and other sweeteners. Experimental Apparatus The flow set-up is shown in Fig. 1. The dual line FI manifold consists of two electrically actuated six-way valves [Valco Analyst, June 1997, Vol. 122 (525–530) 525Instruments (Houston, TX, USA) C22Z-3186EH] and two peristaltic pumps, one for a continuous carrier and reagent flow [Gilson (Worthington, OH, USA) Minipuls 3] and a second one for the automatic filling of the reagent loop [Ismatec (Glattbrugg- Z�urich, Switzerland) MS Reglo].The injection volumes were chosen as 1.5 ml for the sample loop and 100 ml for the reagent loop. Flow rates of 1.1 and 0.055 ml min21 for the sample and the reagent line were achieved by choosing PVC tubing (Tygon) with the appropriate inner diameters of 2 and 0.35 mm, respectively.The high flow ratio (20 : 1) ensures minimum sample dilution and thereby both maximum concentration and maximum precision of the analyte concentration in the flow cell. PTFE tubing of 0.5 and 0.25 mm id was used for all connections. Owing to the short time constant of ionic reactions there was no need for a reaction coil. The connection between the valves and the detection cell was kept as short as possible in order to reduce the axial dispersion of sample and reagents.All IR spectra were obtained on a Bruker (Billerica, MA, USA) IFS 88 FTIR spectrometer equipped with a liquid nitrogen cooled narrow band MCT detector (D* = 2 3 1010 cm Hz1/2 W21). A transmission cell with 25 mm pathlength and CaF2 windows (of 2 mm thickness) was applied, yielding an approximately 1/e attenuation (absorbance approximately 0.4) of the water background absorption at 1100 cm21. Transmission cells with ZnSe windows and other optical pathlengths (50 mm) and the ATR mode (45° ZnSe crystal, 25 internal reflections) were also investigated, leading to a lower S/N for measurements in aqueous solutions in the spectral region 1400–1000 cm21.The introduction of an InSb low wave pass filter (5% cut-on 1370 cm21) into the optical set-up provides a further approximately 3–4-fold increase in the S/N in the spectral region investigated. Further details on the optical filtering were given in a previous paper.37 Spectra with satisfactory S/N were obtained by recording and co-adding 100 scans at a resolution of 4 cm21 in less than 15 s.A laboratory-built electronic interface was used for the synchronization of the pumps, the valves, the FTIR control and data acquisition, allowing completely automated realization of the measurement cycle. Reagents The phosphate solutions for recording the spectra of the individual dissociation states were prepared from analyticalreagent grade NaH2PO4·H2O and Na2HPO4·2H2O, purchased from Merck (Darmstadt, Germany).The solutions for the [H3PO4] and [PO4]32 spectra were prepared by acidifying with HCl and adding NaOH, respectively. Phosphate calibration solutions were prepared by dissolving the appropriate amount of sodium dihydrogenphosphate (NaH2PO4·H2O, analytical-reagent grade, guaranteed concentration 99–102%; Merck) in distilled water. The accuracy and long term stability of the stock solution was confirmed by reference measurements of industrially prepared phosphate solutions (Titriplex; Merck).An acetate buffer solution containing 100 mmol l21 acetic acid and 300 mmol l21 sodium acetate was used for adjusting the pH to approximately 5. The sodium carbonate buffer solution for adjusting the pH to approximately 10–10.5 was prepared with 100 mmol l21 sodium hydrogencarbonate and 600 mmol l21 sodium carbonate. A solution of 2.6 mol l21 NaOH was used for adjusting the pH to > 13. Glucose, fructose and sucrose solutions for interference studies were prepared by dissolving 80 g of each compound (concentration > 99%, for biochemical use; Merck) in 1 l water.The liquids were stabilized with 50 mg l21 NaN3 and stored for 48 h before use in order to prevent mutarotational effects. The industrial samples studied were various soft drinks containing sugar and non-nutritive sweeteners. Prior to analysis the samples were degassed in an ultrasonic bath as the only sample preparation step.Procedure Quantitative analysis in reactive FI–FTIR is based on the evaluation of the difference signal of the analyte before and after the FI-induced chemical or biochemical reaction. As the IR spectrum changes significantly owing to the ionic reaction, distinct features of the difference spectrum can be used for quantification. In contrast to most FI applications, the axial dispersion was kept as small as possible (D ? 1), not just at the maximumf the FI peak, but during the whole recording time of the spectra.This required a comparatively high sample volume of 1.5 ml but ensured both maximum sensitivity and minimum variation of the dispersion. The pH of the injected sample was adjusted to 5 after mixing with the acetate buffer, which was supplied by the reagent line. After a lag phase of 15 s the first spectrum (reference spectrum) of the sample was recorded. By switching the reagent valve a plug of a carbonate buffer solution or sodium hydroxide solution was inserted into the reagent line.Another 15 s of equilibration were necessary to reach the desired pH of 10 or > 13 in the flow cell and the second (sample) spectrum was recorded. Subsequently both valves were switched back to the loading position and the system was ready for the next run. The duration of a complete measurement cycle was 60 s. Reference measurements were performed by ion chromatography with a standard anionic chromatographic set-up and a conductivity detector subsequent to a 1 + 99 dilution.Results and Discussion Water Absorption and its pH Dependence In aqueous solution, large parts of the mid-IR region are not accessible for quantitative IR spectrometry, hence absorption of the analyte has to be well apart from the OH stretching band n1,3 (around 3200–3400 cm21), OH bending band n2 (approximately 1640 cm21) and the libration band nL (around 750 cm21) of liquid water. Although the spectral region investigated is located between n2 and nL, a large unspecific background absorption in the range 900-1500 cm21 also has to be taken into consideration.This background remains fairly constant in the pH range 2–12, but both high H+ and OH2 concentrations influence the water background spectrum (Fig. 2). The observed spectral change is proportional to the concentrations of H+ and Fig. 1 FI–FTIR set up. 526 Analyst, June 1997, Vol. 122OH2. At low pH a broad band with a maximum at 1200 cm21 is formed, which is assigned to the symmetric bending vibration ds of the [H3O]+ ion.38 In alkaline media a baseline shift is visible, which we assume is due to an unspecific change in the water background absorption.Phosphate Absorption The water corrected IR spectra of the individual dissociation states of phosphate are shown in Fig. 3. In case of [H3PO4] and [PO4]32, water of the same pH (1 and 13.5, respectively) was used as the background spectrum.Since the water background absorption increases in the region below 900 cm21 due nL, the bands at lower wavenumbers cannot be used for quantitative analysis. The wavenumbers, the widths [full width at half maximum (FWHM)] and the extinction coefficients (e) of the IR-active phosphate stretching vibrations were extracted from the above spectra and are listed in Table 1. The available literature data on the absorption maxima are additionally presented for comparison. The symmetry of the tertiary phosphate [PO4]32 is completely Td and because of this it has only two IR-active bands, n3 (asymmetric stretch) and n4 (asymmetric deformation) at 1017 and 567 cm21 according to the literature.39,40 Although n4 is not accessible with our experimental set-up, the peak at 1009 cm21 shows good agreement with the literature data.The symmetry of [HPO4]22 in aqueous solution can be seen simplified as a type ZXY3 molecule having C3v symmetry. The bands at 980 and 1080 cm21 correspond to the literature data for ns(PO3) and nas(PO3) of 991 and 1080 cm21, whereas n(POH) is beyond the measurement range at 870 cm21.40 The symmetry of [H2PO4]2 corresponds to the C2v symmetry of a simplified ZX2Y2 molecule.Three of the four stretching vibrations are above 900 cm21 (Fig. 3) with the most intensive nsPO2 at 1078 cm21. No literature data concerning the assignment of the IR absorption of phosphoric acid were available, but the 1178 cm21 vibration is in the range of the PNO stretching vibration of organophosphorus compounds, which is located between 1140 and 1380 cm21.41 The other band at 1010 cm21 is then assigned to the asymmetric P(OH)3 stretching vibration.Modulation of pH The pH modulation was performed in two different modes. First the change from pH 5 to > 13 was investigated. On converting [H2PO4]2 into [PO4]32, the most intensive phosphate vibration n3([PO4]32) is feasible for the phosphate determination.Subtracting the first spectrum ([H2PO4]2) from the second ([PO4]32), a difference spectrum is obtained displaying positive [PO4]32 (1009 cm21) and the negative [H2PO4]2 bands (1157, 1078 cm21). Additionally, a significant baseline shift due to the pH > 13 of the [PO4]32 solution was observed. The acetate buffer of the [H2PO4]2 solution has its absorption bands (CNO and COO2) mainly above 1400 cm21 and does not interfere with the phosphate absorption below 1200 cm21. The second case studied was the conversion of [H2PO4]2 into [HPO4]22 on changing the pH from 5 to 10.Here the specific spectral change of phosphate is smaller. First, the extinction coefficient of the [HPO4]22 band at 1080 cm21 is smaller than that of the corresponding [PO4]32 vibration. Second, the bands of both [H2PO4]2 and [HPO4]22 are almost at the same wavelength and, as a result, the difference spectrum only displays the enlargement of the [H2PO4]2 band at 1078 cm21 to the [HPO4]22 band at 1080 cm21.On the other hand, no interference of the pH on the absorption spectra occurs. The absorption of the carbonate buffer, with its maximum at approximately 1380 cm21, does not influence the phosphate absorption, which makes that method free from reagent interferences. Additionally, the spectral noise at 1090 cm21 ([H2PO4]2 ? [HPO4]22) is less than that at 1004 cm21 ([H2PO4]2 ? [PO4]32), which is due to the increasing water and CaF2 background absorption. Calibration Eight calibration solutions containing 0–1000 mg l21 phosphate were analysed with six repetitions each.Selected difference spectra are shown in Fig. 4 for the alkaline method (pH5 ? > 13) and in Fig. 5 for the carbonate buffer method (pH 5 ? 10). Various evaluation methods were investigated, leading to Fig. 2 Spectral change of the water background absorption at very low and high pH. Spectra a, b and c refer to pH 1.5, 1.0 and 0.5 and spectra d, e and f to pH 12.5, 13.0 and 13.5.Fig. 3 Vibrational spectra of the individual dissociation states of phosphate: a, [H3PO4] (pH 1); b, [H2PO4]2 (pH 5); c, [HPO4]22 (pH 10); d, [PO4]32 (pH > 13). All concentrations 50 mmol l21. Table 1 IR-active stretching vibrations of phosphate Literature Width e/ Maximum maximum (FWHM) l mol21 Species Assignment40 n/cm21 n/cm21 cm21 cm21 [PO4]32 n3 1009 101739 52 1700 [HPO4]22 nas(PO3) 1080 108040 46 1110 ns(PO3) 991 98040 19 470 n(POH) —* 87040 — — [H2PO4]2 nas(PO2) 1159 115040 65 360 ns(PO2) 1078 107040 23 610 nas[P(OH)2] 941 94540 37 310 ns[P(OH)2] —* 88040 — — [H3PO4] n(PNO)† 1178 — 60 190 nas[P(OH)3]† 1010 — 48 430 * Not within the experimentally accessible wavelength range of 1400–900 cm21.† Proposed by the present authors. Analyst, June 1997, Vol. 122 527the best results by using two baseline points for the alkaline method and one baseline point for the carbonate buffered method. In order to reduce the spectral noise, the peak maximum and the baseline points were averaged within a 10–20 wavenumber range.The calibration parameters and results are given in Table 2. The precision of both methods was evaluated by the standard deviation of the method (sx0), calculated as the ratio of the residual standard deviation (sy) and the slope (b). In both cases exact linearity was obtained with r2 > 0.9999. Despite the twofold higher sensitivity of the alkaline method (slope of the calibration curve), sx0 is about equal for both methods.First, the stronger water absorption causes higher spectral noise at lower wavenumbers. Second, the variation in the baseline shift at pH > 13 induces an additional imprecision. Both methods exhibit a repeatability of 0.5% for a concentration of 500 mg l21, which makes applications in process analysis feasible. Interference Study The range 900–1200 cm21 is part of the fingerprint region with interfering absorption bands of numerous compounds. In this study, we focused on the interference of common sugars, such as sucrose, glucose and fructose, as they occur in the food industry.In order to do so, solutions of 80 g l21 of each compound (234 mmol l21 sucrose, 444 mmol l21 glucose and fructose) were prepared and treated by FI–FTIR with modulation of the pH. The acidic properties of sugars in concentrated NaOH solution are well known.42 According to Urban and Shaffer,43 the first apparent dissociation constants, pKA1, are 11.7 for fructose, 12.1 for glucose and 12.6 for sucrose. (In the apparent dissociation constant the activity coefficient of the salt is included.) The spectral changes in the alkaline method [Fig. 6(a)] are therefore proposed to be due to the dissociation reactions of the individual sugars. The spectra exhibit spectral changes partially higher than 0.1. These features are one order of magnitude higher than the phosphate specific absorption change, precluding a phosphate determination by modulation of the pH with sodium hydroxide.Concerning the modulation of pH with carbonate buffer, comparatively weak sugar specific absorption changes were observed [Fig. 6(b)]. Besides the absorption of the carbonate buffer (approximately 1380 cm21), sucrose and glucose show very small absorption changes. The difference spectrum of fructose has the strongest spectral change (0.008), but this is still just one twentieth of that in sodium hydroxide method.The fructose difference spectra of both pH modulations are qualitatively similar, which indicates that the dissociation of Fig. 4 Difference spectra of phosphate calibration solutions obtained by pH modulation with sodium hydroxide (pH 5 ? > 13). Fig. 5 Difference spectra of phosphate calibration solutions obtained by pH modulation with carbonate buffer (pH 5 ? 10). Table 2 Calibration parameters and results Modulation mode pH 5? > 13 pH5?10 Phosphate concentration/mg l21 0, 100, 200, 400, 500, 600, 800, 1000 No.of repetitions 6 6 Evaluated peak, n/cm21 999–1009 1085–1095 Baseline point(s), n/cm21 1072–1082 1155–1165 930–950 Slope, b/mg21 l 5.1531025 2.4831025 Intercept, a 1.131023 23.831024 Regression coefficient, r2 0.99994 0.99996 Residual standard deviation, sy 1.331024 4.931025 Standard deviation of the method, sx0/mg l21 2.6 2.0 Fig. 6 Spectral changes obtained by pH modulation of common sugars with (a) sodium hydroxide (pH 5 ? pH > 13) and (b) carbonate buffer solutions (pH 5 ?10).Spectra 1, 2 and 3 refer to 80 g l21 sucrose, glucose and fructose. For better visualization an offset of 0.01 and 0.1 was applied for each spectrum of the sodium hydroxide and the carbonate buffer method, respectively. Note the difference in the absorbance scale. 528 Analyst, June 1997, Vol. 122fructose (pKA1 = 11.7) begins partially at the pH of the carbonate buffer (approximately pH 10.5). Real Samples Phosphate in industrial samples was determined by the proposed FI–FTIR method.Six samples of various soft drinks, three non-nutritive and three sugar-containing sweeteners, were examined. The selectivity enhancement due to the modulation of the pH is demonstrated in Figs. 7 and 8. Whereas other compounds such as the non-nutritive sweeteners (acesulfame K, cyclamate) and the coloring matter interfere with the phosphate band,44 the difference spectrum with pH modulation is reduced to phosphate specific absorption features.At sugar concentrations of approximately 100 g l21 the phosphate bands are masked by the sugar absorption of > 0.4. However, in the 20-fold enlarged difference spectrum obtained with pH modulation, the almost interference-free phosphate specific absorption change becomes clearly visible. The results of the FI–FTIR analyses of real samples are given in Table 3. Ion chromatography was chosen as an external reference method. Concerning the non-nutritive sweetened samples, both the alkaline and the carbonate buffered method are in satisfactory accordance with the reference method, with an average deviation of 1.7% from the reference values.The sugar-containing samples were analysed by the carbonate buffered pH modulation method, with higher deviations of 1.2–5.3%. The repeatability (standard deviation of multiple injections) was in all cases better than the accordance with the reference method, namely 0.5% for the non-nutritive sweetened samples and 1.6% for sugar-containing samples.Conclusion From the results presented in this paper, it is concluded that the modulation of the pH by means of FI–FTIR is a promising technique for the selective, rapid and automated determination of 100–1000 mg l21 phosphate in aqueous solutions. The proposed ionic reaction fulfils the basic requirements for a fast and reproducible analysis using reactive FI–FTIR. Hence both a high sample throughput (60 h21) and a high precision ( < 1%) are feasible.A significant selectivity enhancement in FTIR was achieved by acquiring a difference spectrum of pH-dependent spectral features. Furthermore, sensitivity enhancement was achieved by choosing the pH of the appropriate dissociation state of the analyte. Oxygen-containing acids, especially organic acids, have stretching vibrations in the 1800–1000 cm21 region, which change in wavelength and intensity owing to the dissociation or association of hydrogen ions.Further experiments will be Fig. 7 Spectra of a non-nutritive sweetened real world sample containing 500 mg l21 phosphate at pH 5 (A) and 10 (B). The interfering absorptions (1040, 1170 cm21) are eliminated in the difference spectrum (C), allowing the quantification of phosphate analogous to the calibration solutions (Fig. 6). An offset of 0.01 was chosen for (B) and 0.03 for (C) for better visualization. Table 3 Phosphate concentrations of soft drinks determined by FI–FTIR and ion chromatography as an external reference method.The phosphate concentrations and standard deviations are expressed in mg l21 and the recoveries as % of the reference value FI–FTIR (pH modulation) Reference method pH 5?10 pH 5? > 13 (ion chromatography) Concentration* Recovery Concentration* Recovery Concentration Non-nutritive sweetened samples— A 498.2 ± 1.7 100.5 485.8 ± 2.8 98.0 495.9 ± 3.3 B 364.4 ± 1.7 103.4 345.9 ± 2.5 98.1 352.5 ± 3.3 C 453.9 ± 2.3 101.3 445.5 ± 2.5 99.4 448.3 ± 3.3 Sugar-containing samples— D 571 ± 7 94.7 —† —† 602.5 ± 3.3 E 492 ± 8 98.8 —† —† 497.8 ± 3.3 F 535 ± 8 102.4 —† —† 522.2 ± 3.3 * The standard deviation was calculated from six repetitions.† No available data because of sugar dissociation upon pH modulation. Fig. 8 Spectra of a sugar-containing (approximately 100 g l21) real sample at pH 5 (A) and 10 (B). Whereas the phosphate bands are hidden in the strong sugar absorbance, the phosphate specific absorption change is visible in the difference spectrum (C).Analyst, June 1997, Vol. 122 529carried out in order to investigate the potential of pH modulation for the determination of other ionic compounds. The authors gratefully thank the Forschungsf�orderungsfonds f�ur die gewerbliche Wirtschaft and the Fonds zur F�orderung der wissenschaftlichen Forschung, P11338 � OCH, for support of this work. References 1 Morgan, D. K., Danielson, N.D., and Katon, J. E., Anal. Lett., 1985, 18(A16), 1979. 2 Curran, D. J., and Collier, W. G., Anal. Chim. Acta, 1985, 177, 259. 3 Miller, B. E., Danielson, N. D., and Katon, J. E., Appl. Spectrosc., 1988, 42, 401. 4 Ramos, M. L., Tyson, J. F., and Curran, D. J., Anal. Proc., 1995, 32, 175. 5 de la Guardia, M., Garrigues, S., and Gallignani, M., Anal. Chim. Acta, 1992, 261, 53. 6 Garrigues, S., Gallignani, M., and de la Guardia, M., Analyst, 1992, 117 1849. 7 Gallignani, M., Garrigues, S., and de la Guardia, M., Anal.Chim. Acta, 1993, 274, 267. 8 de la Guardia, M., Gallignani, M., and Garrigues, S., Anal. Chim. Acta, 1, 543. 9 Gallignani, M., Garrigues, S., de la Guardia, M., Burguera, J. L., and Burguera, M., Talanta, 1994, 41, 739. 10 Gallignani, M., Garrigues, S., and de la Guardia, M., Analyst, 1994, 119, 653. 11 Garrigues, S., Gallignani, M., and de la Guardia, M., Talanta, 1993, 40, 89. 12 Garrigues, S., Gallignani, M., and de la Guardia, M., Talanta, 1993, 40, 1799. 13 Bouhsain, Z., Garrigues, S., and de la Guardia, M., Analyst, 1996, 121, 635. 14 Daghbouche, Y., Garrigues, S., and de la Guardia, M., Analyst, 1996, 121, 1031. 15 Garrigues, S., Vidal, M. T., Gallignani, M., and de la Guardia, M., Analyst, 1994, 119, 659. 16 Daghbouche, Y., Garrigues, S., and de la Guardia, M., Anal. Chim. Acta, 1995, 314, 203. 17 L�opez-Anreus, E., Garrigues, S., and de la Guardia, M., Anal. Chim. Acta, 1996, 333, 157. 18 P�erez-Ponce, A., Garrigues, S., and de la Guardia, M., Analyst, 1996, 121, 923. 19 Rosenberg, E., and Kellner, R., J. Mol. Struct., 1993, 294, 9. 20 Lendl, B., and Kellner, R., Mikrochim. Acta, 1995, 119, 73. 21 Vermes, I., and Grabner, E. W., J. Electroanal. Chem., 1990, 284, 315. 22 Encyclopedia of Analytical Science, ed. Townshend, A., Academic Press, New York, 1995, pp. 3957–3960. 23 Conrath, N., Gr�undig, B., H�uwel, St., and Cammann, K., Anal. Chim. Acta, 1995, 309, 47. 24 Ruzicka, J., and Hansen, E. H., Anal. Chim. Acta, 1975, 78, 145. 25 Sun, F., and Korenaga, T., Appl. Spectrosc., 1996, 50, 1145. 26 McKelvie, I. D., Peat, D. M. W., and Worsfold, P. J., Anal. Proc., 1995, 32, 437. 27 Woo, L., and Maher, W., Anal. Chim. Acta, 1995, 315, 123. 28 Agudo, M., R�ýos, A., and Valc�arcel, M., Trends Anal. Chem., 1994, 13, 409. 29 Herrero, M. A., Atienza, J., Maquieira, A., and Puchades, R., Analyst., 1992, 117, 1019. 30 Linares, P., Luque de Castro, M. D., and Valc�arcel, M., Anal. Chem., 1986, 58, 120. 31 Janse, T. A. H. M., Van der Wiel, P. F. A., and Kateman, G., Anal. Chim. Acta, 1983, 155, 89. 32 Motomizu, S., Wakimoto, T., and T�oei, K., Talanta, 1983, 30, 333. 33 Johnson, K. S., and Petty, R. L., Anal. Chem., 1982, 54, 1185. 34 Hirai, Y., Yoza, N., and Ohashi, S., Anal. Chim. Acta, 1980, 115, 269. 35 Hansen, E. H., and Ruzicka, J., Anal. Chim. Acta, 1976, 87, 353. 36 Røyset, O., Anal. Chim. Acta, 1985, 178, 217. 37 Krieg, P., Lendl, B., Vonach R., and Kellner, R., Fresenius’ J. Anal. Chem., 1996, 356, 504. 38 Falk, M., and Grigu`ere, P. A., Can. J. Chem., 1957, 35, 1195. 39 Nakamoto, K., Infrared and Raman Spectra of Inorganic and Coordination Compounds, Wiley, New York, 4th edn., 1986. 40 Steger, E., and Herzog, K., Anorg. Allg. Chem., 1964, 331, 169. 41 Colthup, N. B., and Daly, L. H., Introduction to Infrared and Raman Spectroscopy, Academic Press, Boston, 3rd edn., 1990. 42 Kilde, G., and Wynne-Jones, W. F. K., Trans. Faraday Soc., 1953, 49, 243. 43 Urban, F., and Shaffer, P., J. Biol. Chem., 1932, 94, 697. 44 Vonach, R., Kellner, R., and Lippitsch, M., in Proceedings of EURO FOOD CHEM VIII, ed. Sontag, G., and Pfannhauser, W., Austrian Chemical Society, Vienna, 1995, p. 573. Paper 6/08540G Received December 23, 1996 Accepted February 10, 1997 530 Analyst, Jun

ISSN:0003-2654    

DOI:10.1039/a608540g

出版商:RSC    

年代:1997

数据来源: RSC

摘要:

Determination of Amyloglucosidase Activity Using Flow Injection Analysis With Fourier Transform Infrared Spectrometric detection R. Schindler, B. Lendl and R. Kellner* Institute for Analytical Chemistry, Vienna University of Technology, Getreidemarkt 9/151, A-1060 Vienna, Austria An automated method for the determination of amyloglucosidase (glucan 1,4-a-glucosidase) activity in aqueous solution is proposed. The method is based on the direct FTIR spectrometric monitoring of starch hydrolysis catalyzed by amyloglucosidase.It is shown that spectral changes caused by the enzyme action can be directly related to the enzyme activity, hence eliminating the need for additional consecutive reaction steps which are normally required to derive a reaction product detectable by conventional techniques such as UV/VIS spectrometry or electrochemistry. Unspecific background absorption was eliminated by calculation of a difference spectrum of the FTIR spectrum of unreacted starch and the FTIR spectrum of the reaction mixture.To guarantee optimum handling of the viscous starch solution and to obtain high sensitivity, a flow injection analysis system combining the merging zone approach and the stopped flow technique was developed. Using a stopped flow time of 5 min and a 55 g l21 starch substrate solution, linear calibration curves from 50 to 2000 U21 (830 - 33 300 nkat l21) were obtained. Recoveries for spiked fermentation broth samples ranged from 102 to 98% and were determined by the standard addition method.Keywords: Enzyme activity; glucan 1,4-a-glucosidase; amyloglucosidase; Fourier transform infrared spectrometry; flow injection analysis Glucan 1,4-a-glucosidase (exo-1,4-a-d-glucan glucohydrolase, amyloglucosidase, EC 3.2.1.3) is a glycoprotein which contains carbohydrate residues that are glycosidically linked through dmannose to the hydroxy groups of serine and threonine in the polypeptide chain of the enzyme.1 The enzyme hydrolyzes the a(1,4)- and, at a significantly slower rate, a(1,6)-glucan bonds of polysaccharides, e.g., starch, releasing b-d-glucose units from the non-reducing end of the molecule.Amyloglucosidase is produced from Aspergillus niger by submerged fermentation and is frequently used in industry, e.g., for starch liquefaction in the food and pharmaceutical industries.2 Most methods for the determination of amyloglucosidase activity rely on the hydrolysis of maltose or glycogen and the quantification of the glucose produced, usually by enzymatic methods.3 An alternative approach is the use of special colored substrates such as pnitrophenol a-d-glucopyranoside and measurement of p-nitrophenol released by the enzyme action.4,5 Normally these procedures are carried out in a batch, but flow injection analysis (FIA) 6–8 and sequential injection analysis (SIA)7 systems for amyloglucosidase determination have also been reported. Application of FTIR spectrometric detection takes advantage of the inherent molecular specificity of the mid-IR spectral region to simplify the determination of amyloglucosidase activity.In the proposed method, no expensive substrates or additional reaction steps are required because cheap, unmodified starch can be used as an enzyme substrate. Enzymatic degradation of starch causes significant changes in the mid IR spectrum mainly in the range between 1000 and 1200 cm21. This fact is exploited here for the determination of amyloglucosidase in aqueous solution in the range 50–2000 U l21 (830–33 300 nkat l21; 1 U = 16.67 nkat).For this purpose, two FTIR spectra are recorded, the first for the unreacted substrate and the second after a reaction time of typically 5 min. Subtraction of the IR spectrum of the reaction mixture from that recorded before reaction delivers a difference spectrum corresponding to the degree of enzymatic action and hence can be exploited to quantify the enzyme activity.Application of FTIR spectrometry to molecular specific detection in flow injection analysis has already been demonstrated for enzyme substrate determinations using immobilized enzymes.9,10 Furthermore, the use of FTIR spectrometry for the determination of a- amylase activity in human sera 11 and the assay of bile saltstimulated lipase activity in reversed micelles12 using batch methods has been reported. Combination of an FTIR spectrometer with an FIA manifold is of advantage as a high degree of automation is achieved, opening up the possibility of using this coupled technique for process analysis and process control.Experimental Reagents All reagents were of analytical-reagent grade. The substrate (starch) solution (55 g l21) was prepared by dissolving an appropriate amount of starch (Merck, Darmstadt, Germany) in 0.1 mol l21 acetate buffer solution of pH 4.3 and boiling for 5 min. The clear solution obtained was filtered over a 1 cm thick layer of calcined diatomaceous earth (Hyflo) and diluted to volume with acetate buffer solution.The starch solution was prepared freshly every day. An amyloglucosidase stock standard solution was prepared by dissolving 83.3 mg of lyophilized enzyme (obtained from Boehringer, Mannheim, Germany; 6 U mg21) in 250 ml of distilled water. The solution was stored at 4 °C and was stable for at least 1 month. Aliquots of the stock standard solution were diluted as required with distilled water to give working standard solutions.The enzyme activities stated in this paper are related to the enzyme activity of the enzyme stock standard solution which was prepared as described above and are valid for a temperature of 25 °C. For reference measurements (25 °C), glycogen (Type II from oyster; Sigma, St. Louis, MO, USA) was used as a substrate followed by an enzymatic determination of the glucose produced according to the method described by Bergmeyer.3 FTIR Spectrometer, Flow-through Cell.A Bruker (Karlsruhe, Germany) IFS 88 FTIR spectrometer equipped with a high-sensitivity narrow-band mercury cadmium telluride (MCT) detector was used. In order to exploit the dynamic range of the MCT detector more efficiently, a low wave pass filter with a 5% cut on at 1370 cm21 was placed in Analyst, June 1997, Vol. 122 (531–534) 531the sample compartment immediately behind the flow cell. By this means, virtually no IR radiation above 1400 cm21 could reach the MCT detector.Hence it was possible to increase the light intensity in the region of interest (950–1300 cm21) by opening the aperture of the FTIR spectrometer without oversaturating the detector, which resulted in an improved signal-to-noise ratio. Measurements were carried out using a conventional liquid flow through cell equipped with one CaF2 and one ZnSe window (each 2 mm thick) and a lead spacer providing an optical pathlength of 49 mm.The use of a flow cell containing two different window materials (CaF2 and ZnSe) is of advantage compared with cells containing either two CaF2 or ZnSe windows. ZnSe is more transparent for IR-radiation below 1000 cm21 than CaF2 but has the drawback of a higher refractive index. Therefore, when two ZnSe windows were used, double reflection of the IR radiation in the flow cell occurred. Hence small changes in the optical pathlength due to slight pressure fluctuations in the flow system caused interference fringes which increased the noise level significantly.This effect was not observed when CaF2 windows were used. Therefore, a combination of both materials yielded the optimum signal-to-noise ratio especially at lower wavenumbers as here more light intensity was available and no interference fringes were recorded. FIA-manifold and procedure The manifold used is depicted in Fig. 1. The FIA system was set up with two Rheodyne (Cotati, CA, USA) six-port injection valves, PTFE tubing, Tygon pump tubing and FIA fittings.An Ismatec (Glattbrugg-Z�urich, Switzerland) MS Reglo peristaltic pump and an M3 LAUDA (K�onigshofen, Germany) thermostated water-bath were also used. The enzyme substrate (starch) was injected by the first injection valve whereas the sample was introduced by the second valve. The valves were mechanically connected to each other and automatically switched by means of a labory-made device. Automated operation of the FIA– FTIR system was achieved by a small Macro routine written within the OPUS software controlling the FTIR spectrometer.The Macro routine allowed exact timing of spectrum acquisition, switching the injection valves and starting/stopping the peristaltic pump by setting external TTL signals. Because of the viscous nature of concentrated starch solutions, special care was paid to prevent clogging of the flow-through cell and to ensure delivery of a constant substrate concentration during the experiments.Cell clogging was prevented by application of the merging zone approach, that is, simultaneous injection of sample (amyloglucosidase) and enzyme substrate (starch) and simultaneous merging of the two injected zones downstream in the reaction coil. The advantage of this approach is that the flow cell is rinsed with carrier (distilled water) before and after each measurement. Injection of a constant starch concentration was achieved by continuous stirring of the starch solution before introduction into the injection valve.In order to increase the reaction time and hence the recorded signal, the stopped flow technique was also applied. The flow was halted as soon as the reaction plug had filled the reaction coil, which was immersed in the water-bath for temperature control. After restarting the flow, FTIR spectra (128 scans each, spectral resolution 8 cm21 and Blackmann Harris three-term apodization) were recorded as the reaction plug passed the flow cell.Difference spectra were calculated by subtracting a spectrum of pure starch from the spectrum of the reaction mixture. Results and Discussion FTIR Spectra Spectrum A in Fig. 2(a) represents undegraded starch and was recorded on injecting substrate (55 g l21 starch) solution and distilled water by the two injection valves. This spectrum served as a reference spectrum for the analysis of amyloglucosidase in pure aqueous solutions.Spectrum B was obtained on injecting 2000 U l21amyloglucosidase instead of distilled water and applying a stopped flow time of 20 min at a reaction temperature of 25 °C. According to the literature, the absorption bands around 1153–904 cm21 can be assigned to C–O and C–C stretching modes.13 Repetitive analysis of the 2000 U l21 standard but applying different stopped flow times enabled starch degradation to be followed as a function of time. The corresponding difference spectra which were calculated by subtracting the reference spectrum from those recorded after different reaction times are shown in Fig 2(b).The most pronounced changes in the starch spectrum, due to partial hydrolysis, occur at 1078 cm21 where an increase and at 1020 cm21 where a decrease in absorption is found, resulting in positive and negative peaks in the calculated difference spectra, respectively. The FTIR spectrum of b-d- glucose, the product of the enzymatic reaction, characteristically absorbs at 1080 cm21,14 owing to the C–O stretching mode of the anomeric C– O–H group.This explains the increase in absorption at 1078 cm21 during hydrolysis and hence the negative peak in the calculated difference spectrum. This observation was confirmed by Bellon-Maurell et al.,15 who stated that the peak at 1076 Fig: 1 FIA manifold. Injection volumes, 250 ml each; length of reactor in thermostated bath, l3 = 250 cm (0.5 mm id); all other tubing was made as short as possible (l1 = l2 = 10 cm, 0.5 mm id).Fig. 2 Resulting FTIR spectra. (a) Spectra of 27.5 g l21 starch before (A) and after (B) enzymatic action, at 25 °C and with a 20 min reaction time for an enzyme standard of 1000 U l21. (b) Difference spectra at 25 °C for different times for an enzyme standard of 1000 U l21; A, 20; B, 12; C, 8; D, 4; and E, 2 min. 532 Analyst, June 1997, Vol. 122cm21 shows a decrease as the degree of polymerization increases.The decrease in absorption at 1020 cm21 could be explained by the enzymatic hydrolysis of the a-(1,4)-C–O–Cbridge, which exhibits C–O and C–C stretching modes in this region.14 Under the experimental conditions described here, enzymatic hydrolysis of starch catalyzed by amyloglucosidase causing the cleavage of b-d-glucose from the non-reducing end of the polymer chain, it was clearly observed that the peak at 1020 cm21 is affected by the degree of polymerization, which is contradictory to that reported by Bellon-Maurell et al.,15 who stated that the this peak is not affected at all by the degree of polymerization.Calculation of the difference spectra offers an additional interesting feature for the determination of amyloglucosidase activity by direct quantification of the glucose produced. The products of the reaction under study are b-d-glucose and shortened starch units. It is assumed that at the beginning of the reaction the shape of the FTIR spectrum of the high molecular mass starch does not change significantly because only small units are cleaved off from the ends, but the main high molecular mass part does not change. The decrease in the absorption intensity of the starch is accompanied by the appearance of absorption of b-d-glucose which is produced during reaction.Enzyme activity is commonly stated in units (U), where 1 U corresponds to the amount of enzyme catalyzing the turnover of 1 mmol of substrate per minute under given reaction conditions [according to the SI definition, the enzyme activity (katal) is defined as the amount of enzyme which catalyzes under defined reaction conditions the turnover of 1 mol of substrate per second].In the example given, an amyloglucosidase activity of 1000 U l21 (25 °C) should therefore produce 20 mmol l21 glucose (corresponding to 3.6 g l21) within a 20 min reaction time at 25 °C. These assumptions were proved by calculating the difference spectrum between the spectrum of a mixture of b-d-glucose (3.60 g l21) and starch (23.9 g l21) and the spectrum of starch (27.5 g l21).The b-d-glucose–starch mixture was measured 10 min after dissolution to exclude mutarotation effects, which could influence a measurement after dissolving b-d-glucose in the starch solution. In Fig. 3 one difference spectrum calculated for 1000 U l21 amyloglucosidase with 27.5 g l21 starch solution after a 20 min reaction time at 25 °C and the difference spectrum calculated from the b-d-glucose–starch mixture are shown.The almost perfect match of the two spectra is obvious, and a direct calculation of the enzyme activity without using enzyme standards for calibration by evaluation of the glucose produced can be carried out in principle. The differences in the region of 1050–1000 cm21 can be attributed to the fact that for the starch– glucose mixture undegraded starch was used, whereas the starch in the reaction mixture has been degraded to some extent, which is also assumed to cause spectral changes.The resulting value of the enzyme activity with this method agreed perfectly with the 1000 U l21 enzyme activity found with the calibration curve. Nevertheless, special care has to be taken in the preparation of the b-d-glucose–starch mixture to exclude mutarotation effects; a calibration with enzyme standard solution is more convenient for normal use, especially in routine analysis.Optimization of Reaction Conditions and Flow Parameters The starch delivery was automated in order to guarantee a constant supply of starch for reaction. To determine the variation of the amount of starch reaching the flow cell at the time of measurement, the RSD of the peak height at 1025 cm21 was calculated. An RSD of 0.5% was found for 10 replicate injections of 55 g l21 starch and distilled water by the two coupled injection valves showing the high precision of the FIA– FTIR system.The influence of different starch concentrations on the performance of the FIA–FTIR system was investigated. A high starch concentration increases the linear range of the method because substrate saturation is guaranteed,11 but the problem of cell clogging is also increased. A concentration of 55 g l21 was selected, being the highest concentration which did not cause practical problems. The flow rate for each stream was set to 0.88 ml min21, which was slow enough to allow reproducible timing by the Macro routine.With the help of a dye solution (Methylene Blue, 100 mg l21), the length of the tubing was optimized to guarantee complete mixing of sample and substrate zones. With a reactor volume of 490 ml (250 cm 3 0.5 mm id) and the other tubing as short as possible, sufficient mixing was guaranteed without an increase in dispersion. The time the mixed zones needed to reach the reactor and the flow cell was also determined with the dye solution. A stopped flow time of 5 min was chosen as a compromise between sensitivity and sample throughput.A longer stopped flow time would give higher sensitivity at the cost of a reduced sample throughput. The influence of temperature was investigated in the range 25–85 °C (Fig. 4). The data agreed well with the literature data.1 Calibration Curves. As the analytical readout, the difference in absorbances at 1078 and 1020 cm21 in the calculated difference spectra was taken. Other pairs of wavenumbers or different spectral ranges could be used, but this pair of wavenumbers was selected for achieving the optimum signal to noise ratio.Calibration curves were measured for two reaction temperatures, 37 and 54 °C, using six and nine enzyme standards, respectively. Each enzyme standard was subjected to triple analysis. For 37 °C a calibration curve ranging from 125 to 2000 U l21 was recorded with y (U l21) = 20.0015 + 2.22 3 1025 x with a standard deviation of the method sx0 = 22.5 U l21 and Fig. 3 Difference spectra: A, calculated from b-d-glucose and starch as described in the text; and B, enzyme standard (1000 U l21) at 25 °C, 20 min reaction time. Fig. 4 Temperature dependence of the enzyme activity in the FIA–FTIR setup (2000 U l21 amyloglucosidase solution, 27.5 g l21 starch, absorbance difference 1020–1078 cm21, n = 3). Analyst, June 1997, Vol. 122 533a regression coefficient r = 0.9995. At 54 °C the calibration range could be extended to 25–2000 U l21, giving a linear calibration curve y (U l21), = 0.00047 + 4.84 3 1025 x (sx0 = 36.2 U l21, r = 0.9987).The enzyme standards were prepared by diluting aliquots of the enzyme stock solution. In order to verify the enzyme activities obtained, an external reference method was used. As the reference method, the procedure described in ref. 3 was applied to the analysis of selected enzyme standards using a reaction temperature of 25 °C.Triple analysis of three standards (500, 1000 and 2000 U l21) yielded enzyme activities of 480, 880 and 1740 U l21, respectively (RSD for n = 3 : 8 1, 7.1 and 8.1%), hence confirming the enzyme activities of the standards. The lower values found for the reference method can be explained by the different substrates, especially because of the constant difference between the values (11–14%). Spiked Real Samples The method was tested in the same way as described by Hansen et al.8 by spiking a fermentation broth with known amounts of amyloglucosidase.This broth was obtained from the Institute for Process Engineering of this university and originated from an acetone–butanol fermentation. It contained around 3% glucose (as carbon source) and degradation products such as nbutanol (around 7%), acetone (around 4%), acetic acid (around 4%), butyric acid (around 1%) and ethanol (around 1%). The broth was spiked with different amounts of amyloglucosidase stock solution and diluted with distilled water to give a 1:1 diluted broth.A 1:1 diluted fermentation broth was injected at 25 °C without applying a stopped–flow time and the spectrum obtained was used as the background. In this way a short and hence negligible reaction time (15 s) was obtained. For each temperature of the two calibration curves a sample was measured, evaluating its activity by using the corresponding calibration curve. To investigate the influence of the matrix on the results of the developed method, standard additions were performed for each sample, adding increments of 125, 250 and 500 U l21 amyloglucosidase. The results of the analysis of the spiked real samples are given in Table 1.The FIA–FTIR approach offers an effective way of eliminating the matrix, as can be seen from the results in Table 1. The lower recoveries obtained when using the calibration curves established from pure aqueous enzyme standards can be explained by a matrix effect.Especially glucose, which is present in the broth, is known to be a competitive inhibitor of the enzyme. The effect of inhibitors will be more pronounced for lower concentrations, which can also be seen in Table 1. However, the successful application of standard additions shows that the developed method can be used for the determination of amyloglucosidase activity in real samples. Conclusions An automated system capable of the determination of amyloglucosidase activity in aqueous solution based on the coupling of FIA and FTIR spectrometry was developed.The FIA–FTIR system is based on direct monitoring of starch hydrolysis catalyzed by amyloglucosidase hence eliminating the need for consecutive indicator reactions or expensive, specially designed enzyme substrates usually needed to assay amyloglucosidase activity. The coupling of the FTIR spectrometer with a flow system capable of performing chemical reactions in an automated, repeatable and reproducible way proved to enhance the detection power of the FTIR instrument, as using this approach a parameter which is not accessible by a single FTIR measurement could be successfully determined.Furthermore, this coupling has demonstrated new possibilities for the application of mid-IR FTIR spectrometry for remote process monitoring and process control. References 1 Enzyme Handbook, ed. Schomburg, D., and Salzmann, M., Springer, Berlin, 1991, p. 3.2.1.3. 2 Enzymes, Biomass, Food and Feed, in Biotechnology, ed.Reed, G., and Nagodawithana, T. W., 2nd edn., VCH, Weinheim, 1995. 3 Methods in Enzymatic Analysis, ed. Bergmeyer, H. U., VCH, Weinheim, 3rd edn., 1984, vol. IV. 4 Elder, M. T., and Montgomery, R. S., J. AOAC Int., 1995, 78(2), 398. 5 Holm, K. A., Anal. Chim. Acta, 1980, 117, 359. 6 Holm, K. A., Analyst, 1986, 111, 927. 7 Hartford, C. G., Muntz, D. W., Evans, J. V., and Blair, G. T., J. Food Compos. Anal., 1989, 2(4), 364. 8 Hansen, E. H., Willumsen, B., Winther, S. K., and Drabøl, H., Talanta, 1994, 41(11), 1881. 9 Rosenberg, E., and Kellner, R., J. Mol. Struct., 1993, 294, 9. 10 Lendl, B., and Kellner, R., Mikrochim. Acta, 1995, 119, 73. 11 Krieg, P., Lendl, B., Vonach, R., and Kellner, R., Fresenius’ Z. Anal. Chem., 1996, 356, 507. 12 O’Conner, C. J., and Cleverly, D. R., J. Chem. Technol Biotechnol., 1994, 61, 209. 13 Hineno, M., Carbohydr. Res., 1977, 56, 219. 14 Back, D. M., Michalska, D. F., and Polavarapu, P.L., Appl. Spectrosc., 1984, 38(2), 173. 15 Bellon-Maurell, V., Vallant, C., and Goffinet, D., Appl. Spectrosc., 1995, 49, 556. Paper 7/00432J Received January 20, 1997 Accepted March 12, 1997 Table 1: Results for spiked real samples Temperature 37 °C 54 °C Activity added/U l21 500 200 Activity found with standard addition/ U l21 509 196 Recovery (%) 101.8 98.2 Slope of calibration curve for standard addition, b/U l21 2.23 3 1025 4.55 3 1025 Activity found with calibration curve/ U l21 479 180 Recovery (%) 95.8 90.0 534 Analyst, June 1997, Vol. 122 Determination of Amyloglucosidase Activity Using Flow Injection Analysis With Fourier Transform Infrared Spectrometric detection R. Schindler, B. Lendl and R. Kellner* Institute for Analytical Chemistry, Vienna University of Technology, Getreidemarkt 9/151, A-1060 Vienna, Austria An automated method for the determination of amyloglucosidase (glucan 1,4-a-glucosidase) activity in aqueous solution is proposed.The method is based on the direct FTIR spectrometric monitoring of starch hydrolysis catalyzed by amyloglucosidase. It is shown that spectral changes caused by the enzyme action can be directly related to the enzyme activity, hence eliminating the need for additional consecutive reaction steps which are normally required to derive a reaction product detectable by conventional techniques such as UV/VIS spectrometry or electrochemistry. Unspecific background absorption was eliminated by calculation of a difference spectrum of the FTIR spectrum of unreacted starch and the FTIR spectrum of the reaction mixture.To guarantee optimum handling of the viscous starch solution and to obtain high sensitivity, a flow injection analysis system combining the merging zone approach and the stopped flow technique was developed. Using a stopped flow time of 5 min and a 55 g l21 starch substrate solution, linear calibration curves from 50 to 2000 U21 (830 - 33 300 nkat l21) were obtained.Recoveries for spiked fermentation broth samples ranged from 102 to 98% and were determined by the standard addition method. Keywords: Enzyme activity; glucan 1,4-a-glucosidase; amyloglucosidase; Fourier transform infrared spectrometry; flow injection analysis Glucan 1,4-a-glucosidase (exo-1,4-a-d-glucan glucohydrolase, amyloglucosidase, EC 3.2.1.3) is a glycoprotein which contains carbohydrate residues that are glycosidically linked through dmannose to the hydroxy groups of serine and threonine in the polypeptide chain of the enzyme.1 The enzyme hydrolyzes the a(1,4)- and, at a significantly slower rate, a(1,6)-glucan bonds of polysaccharides, e.g., starch, releasing b-d-glucose units from the non-reducing end of the molecule.Amyloglucosidase is produced from Aspergillus niger by submerged fermentation and is frequently used in industry, e.g., for starch liquefaction in the food and pharmaceutical industries.2 Most methods for the determination of amyloglucosidase activity rely on the hydrolysis of maltose or glycogen and the quantification of the glucose produced, usually by enzymatic methods.3 An alternative approach is the use of special colored substrates such as pnitrophenol a-d-glucopyranoside and measurement of p-nitrophenol released by the enzyme action.4,5 Normally these procedures are carried out in a batch, but flow injection analysis (FIA) 6–8 and sequential injection analysis (SIA)7 systems for amyloglucosidase determination have also been reported.Application of FTIR spectrometric detection takes advantage of the inherent molecular specificity of the mid-IR spectral region to simplify the determination of amyloglucosidase activity. In the proposed method, no expensive substrates or additional reaction steps are required because cheap, unmodified starch can be used as an enzyme substrate. Enzymatic degradation of starch causes significant changes in the mid IR spectrum mainly in the range between 1000 and 1200 cm21.This fact is exploited here for the determination of amyloglucosidase in aqueous solution in the range 50–2000 U l21 (830–33 300 nkat l21; 1 U = 16.67 nkat). For this purpose, two FTIR spectra are recorded, the first for the unreacted substrate and the second after a reaction time of typically 5 min. Subtraction of the IR spectrum of the reaction mixture from that recorded before reaction delivers a difference spectrum corresponding to the degree of enzymatic action and hence can be exploited to quantify the enzyme activity.Application of FTIR spectrometry to molecular specific detection in flow injection analysis has already been demonstrated for enzyme substrate determinations using immobilized enzymes.9,10 Furthermore, the use of FTIR spectrometry for the determination of a- amylase activity in human sera 11 and the assay of bile saltstimulated lipase activity in reversed micelles12 using batch methods has been reported.Combination of an FTIR spectrometer with an FIA manifold is of advantage as a high degree of automation is achieved, opening up the possibility of using this coupled technique for process analysis and process control. Experimental Reagents All reagents were of analytical-reagent grade. The substrate (starch) solution (55 g l21) was prepared by dissolving an appropriate amount of starch (Merck, Darmstadt, Germany) in 0.1 mol l21 acetate buffer solution of pH 4.3 and boiling for 5 min.The clear solution obtained was filtered over a 1 cm thick layer of calcined diatomaceous earth (Hyflo) and diluted to volume with acetate buffer solution. The starch solution was prepared freshly every day. An amyloglucosidase stock standard solution was prepared by dissolving 83.3 mg of lyophilized enzyme (obtained from Boehringer, Mannheim, Germany; 6 U mg21) in 250 ml of distilled water. The solution was stored at 4 °C and was stable for at least 1 month.Aliquots of the stock standard solution were diluted as required with distilled water to give working standard solutions. The enzyme activities stated in this paper are related to the enzyme activity of the enzyme stock standard solution which was prepared as described above and are valid for a temperature of 25 °C. For reference measurements (25 °C), glycogen (Type II from oyster; Sigma, St. Louis, MO, USA) was used as a substrate followed by an enzymatic determination of the glucose produced according to the method described by Bergmeyer.3 FTIR Spectrometer, Flow-through Cell.A Bruker (Karlsruhe, Germany) IFS 88 FTIR spectrometer equipped with a high-sensitivity narrow-band mercury cadmium telluride (MCT) detector was used. In order to exploit the dynamic range of the MCT detector more efficiently, a low wave pass filter with a 5% cut on at 1370 cm21 was placed in Analyst, June 1997, Vol. 122 (531–534) 531the sample compartment immediately behind the flow cell. By this means, virtually no IR radiation above 1400 cm21 could reach the MCT detector. Hence it was possible to increase the light intensity in the region of interest (950–1300 cm21) by opening the aperture of the FTIR spectrometer without oversaturating the detector, which resulted in an improved signal-to-noise ratio. Measurements were carried out using a conventional liquid flow through cell equipped with one CaF2 and one ZnSe window (each 2 mm thick) and a lead spacer providing an optical pathlength of 49 mm.The use of a flow cell containing two different window materials (CaF2 and ZnSe) is of advantage compared with cells containing either two CaF2 or ZnSe windows. ZnSe is more transparent for IR-radiation below 1000 cm21 than CaF2 but has the drawback of a higher refractive index. Therefore, when two ZnSe windows were used, double reflection of the IR radiation in the flow cell occurred.Hence small changes in the optical pathlength due to slight pressure fluctuations in the flow system caused interference fringes which increased the noise level significantly. This effect was not observed when CaF2 windows were used. Therefore, a combination of both materials yielded the optimum signal-to-noise ratio especially at lower wavenumbers as here more light intensity was available and no interference fringes were recorded. FIA-manifold and procedure The manifold used is depicted in Fig. 1. The FIA system was set up with two Rheodyne (Cotati, CA, USA) six-port injection valves, PTFE tubing, Tygon pump tubing and FIA fittings. An Ismatec (Glattbrugg-Z�urich, Switzerland) MS Reglo peristaltic pump and an M3 LAUDA (K�onigshofen, Germany) thermostated water-bath were also used. The enzyme substrate (starch) was injected by the first injection valve whereas the sample was introduced by the second valve. The valves were mechanically connected to each other and automatically switched by means of a laboratory-made device.Automated operation of the FIA– FTIR system was achieved by a small Macro routine written within the OPUS software controlling the FTIR spectrometer. The Macro routine allowed exact timing of spectrum acquisition, switching the injection valves and starting/stopping the peristaltic pump by setting external TTL signals. Because of the viscous nature of concentrated starch solutions, special care was paid to prevent clogging of the flow-through cell and to ensure delivery of a constant substrate concentration during the experiments.Cell clogging was prevented by application of the merging zone approach, that is, simultaneous injection of sample (amyloglucosidaand enzyme substrate (starch) and simultaneous merging of the two injected zones downstream in the reaction coil. The advantage of this approach is that the flow cell is rinsed with carrier (distilled water) before and after each measurement.Injection of a constant starch concentration was achieved by continuous stirring of the starch solution before introduction into the injection valve. In order to increase the reaction time and hence the recorded signal, the stopped flow technique was also applied. The flow was halted as soon as the reaction plug had filled the reaction coil, which was immersed in the water-bath for temperature control. After restarting the flow, FTIR spectra (128 scans each, spectral resolution 8 cm21 and Blackmann Harris three-term apodization) were recorded as the reaction plug passed the flow cell.Difference spectra were calculated by subtracting a spectrum of pure starch from the spectrum of the reaction mixture. Results and Discussion FTIR Spectra Spectrum A in Fig. 2(a) represents undegraded starch and was recorded on injecting substrate (55 g l21 starch) solution and distilled water by the two injection valves.This spectrum served as a reference spectrum for the analysis of amyloglucosidase in pure aqueous solutions. Spectrum B was obtained on injecting 2000 U l21amyloglucosidase instead of distilled water and applying a stopped flow time of 20 min at a reaction temperature of 25 °C. According to the literature, the absorption bands around 1153–904 cm21 can be assigned to C–O and C–C stretching modes.13 Repetitive analysis of the 2000 U l21 standard but applying different stopped flow times enabled starch degradation to be followed as a function of time. The corresponding difference spectra which were calculated by subtracting the reference spectrum from those recorded after different reaction times are shown in Fig 2(b).The most pronounced changes in the starch spectrum, due to partial hydrolysis, occur at 1078 cm21 where an increase and at 1020 cm21 where a decrease in absorption is found, resulting in positive and negative peaks in the calculated difference spectra, respectively.The FTIR spectrum of b-d- glucose, the product of the enzymatic reaction, characteristically absorbs at 1080 cm21,14 owing to the C–O stretching mode of the anomeric C– O–H group. This explains the increase in absorption at 1078 cm21 during hydrolysis and hence the negative peak in the calculated difference spectrum. This observation was confirmed by Bellon-Maurell et al.,15 who stated that the peak at 1076 Fig: 1 FIA manifold. Injection volumes, 250 ml each; length of reactor in thermostated bath, l3 = 250 cm (0.5 mm id); all other tubing was made as short as possible (l1 = l2 = 10 cm, 0.5 mm id).Fig. 2 Resulting FTIR spectra. (a) Spectra of 27.5 g l21 starch before (A) and after (B) enzymatic action, at 25 °C and with a 20 min reaction time for an enzyme standard of 1000 U l21. (b) Difference spectra at 25 °C for different times for an enzyme standard of 1000 U l21; A, 20; B, 12; C, 8; D, 4; and E, 2 min. 532 Analyst, June 1997, Vol. 122cm21 shows a decrease as the degree of polymerization increases. The decrease in absorption at 1020 cm21 could be explained by the enzymatic hydrolysis of the a-(1,4)-C–O–Cbridge, which exhibits C–O and C–C stretching modes in this region.14 Under the experimental conditions described here, enzymatic hydrolysis of starch catalyzed by amyloglucosidase causing the cleavage of b-d-glucose from the non-reducing end of the polymer chain, it was clearly observed that the peak at 1020 cm21 is affected by the degree of polymerization, which is contradictory to that reported by Bellon-Maurell et al.,15 who stated that the this peak is not affected at all by the degree of polymerization.Calculation of the difference spectra offers an additional interesting feature for the determination of amyloglucosidase activity by direct quantification of the glucose produced. The products of the reaction under study are b-d-glucose and shortened starch units.It is assumed that at the beginning of the reaction the shape of the FTIR spectrum of the high molecular mass starch does not change significantly because only small units are cleaved off from the ends, but the main high molecular mass part does not change. The decrease in the absorption intensity of the starch is accompanied by the appearance of absorption of b-d-glucose which is produced during reaction. Enzyme activity is commonly stated in units (U), where 1 U corresponds to the amount of enzyme catalyzing the turnover of 1 mmol of substrate per minute under given reaction conditions [according to the SI definition, the enzyme activity (katal) is defined as the amount of enzyme which catalyzes under defined reaction conditions the turnover of 1 mol of substrate per second].In the example given, an amyloglucosidase activity of 1000 U l21 (25 °C) should therefore produce 20 mmol l21 glucose (corresponding to 3.6 g l21) within a 20 min reaction time at 25 °C.These assumptions were proved by calculating the difference spectrum between the spectrum of a mixture of b-d-glucose (3.60 g l21) and starch (23.9 g l21) and the spectrum of starch (27.5 g l21). The b-d-glucose–starch mixture was measured 10 min after dissolution to exclude mutarotation effects, which could influence a measurement after dissolving b-d-glucose in the starch solution. In Fig. 3 one difference spectrum calculated for 1000 U l21 amyloglucosidase with 27.5 g l21 starch solution after a 20 min reaction time at 25 °C and the difference spectrum calculated from the b-d-glucose–starch mixture are shown.The almost perfect match of the two spectra is obvious, and a direct calculation of the enzyme activity without using enzyme standards for calibration by evaluation of the glucose produced can be carried out in principle. The differences in the region of 1050–1000 cm21 can be attributed to the fact that for the starch– glucose mixture undegraded starch was used, whereas the starch in the reaction mixture has been degraded to some extent, which is also assumed to cause spectral changes.The resulting value of the enzyme activity with this method agreed perfectly with the 1000 U l21 enzyme activity found with the calibration curve. Nevertheless, special care has to be taken in the preparation of the b-d-glucose–starch mixture to exclude mutarotation effects; a calibration with enzyme standard solution is more convenient for normal use, especially in routine analysis.Optimization of Reaction Conditions and Flow Parameters The starch delivery was automated in order to guarantee a constant supply of starch for reaction. To determine the variation of the amount of starch reaching the flow cell at the time of measurement, the RSD of the peak height at 1025 cm21 was calculated. An RSD of 0.5% was found for 10 replicate injections of 55 g l21 starch and distilled water by the two coupled injection valves showing the high precision of the FIA– FTIR system.The influence of different starch concentrations on the performance of the FIA–FTIR system was investigated. A high starch concentration increases the linear range of the method because substrate saturation is guaranteed,11 but the problem of cell clogging is also increased. A concentration of 55 g l21 was selected, being the highest concentration which did not cause practical problems.The flow rate for each stream was set to 0.88 ml min21, which was slow enough to allow reproducible timing by the Macro routine. With the help of a dye solution (Methylene Blue, 100 mg l21), the length of the tubing was optimized to guarantee complete mixing of sample and substrate zones. With a reactor volume of 490 ml (250 cm 3 0.5 mm id) and the other tubing as short as possible, sufficient mixing was guaranteed without an increase in dispersion. The time the mixed zones needed to reach the reactor and the flow cell was also determined with the dye solution.A stopped flow time of 5 min was chosen as a compromise between sensitivity and sample throughput. A longer stopped flow time would give higher sensitivity at the cost of a reduced sample throughput. The influence of temperature was investigated in the range 25–85 °C (Fig. 4). The data agreed well with the literature data.1 Calibration Curves. As the analytical readout, the difference in absorbances at 1078 and 1020 cm21 in the calculated difference spectra was taken.Other pairs of wavenumbers or different spectral ranges could be used, but this pair of wavenumbers was selected for achieving the optimum signal to noise ratio. Calibration curves were measured for two reaction temperatures, 37 and 54 °C, using six and nine enzyme standards, respectively. Each enzyme standard was subjected to triple analysis. For 37 °C a calibration curve ranging from 125 to 2000 U l21 was recorded with y (U l21) = 20.0015 + 2.22 3 1025 x with a standard deviation of the method sx0 = 22.5 U l21 and Fig. 3 Difference spectra: A, calculated from b-d-glucose and starch as described in the text; and B, enzyme standard (1000 U l21) at 25 °C, 20 min reaction time. Fig. 4 Temperature dependence of the enzyme activity in the FIA–FTIR setup (2000 U l21 amyloglucosidase solution, 27.5 g l21 starch, absorbance difference 1020–1078 cm21, n = 3). Analyst, June 1997, Vol. 122 533a regression coefficient r = 0.9995. At 54 °C the calibration range could be extended to 25–2000 U l21, giving a linear calibration curve y (U l21), = 0.00047 + 4.84 3 1025 x (sx0 = 36.2 U l21, r = 0.9987). The enzyme standards were prepared by diluting aliquots of the enzyme stock solution. In order to verify the enzyme activities obtained, an external reference method was used. As the reference method, the procedure described in ref. 3 was applied to the analysis of selected enzyme standards using a reaction temperature of 25 °C. Triple analysis of three standards (500, 1000 and 2000 U l21) yielded enzyme activities of 480, 880 and 1740 U l21, respectively (RSD for n = 3 : 8 1, 7.1 and 8.1%), hence confirming the enzyme activities of the standards.The lower values found for the reference method can be explained by the different substrates, especially because of the constant difference between the values (11–14%). Spiked Real Samples The method was tested in the same way as described by Hansen et al.8 by spiking a fermentation broth with known amounts of amyloglucosidase. This broth was obtained from the Institute for Process Engineering of this university and originated from an acetone–butanol fermentation.It contained around 3% glucose (as carbon source) and degradation products such as nbutanol (around 7%), acetone (around 4%), acetic acid (around 4%), butyric acid (around 1%) and ethanol (around 1%).The broth was spiked with different amounts of amyloglucosidase stock solution and diluted with distilled water to give a 1:1 diluted broth. A 1:1 diluted fermentation broth was injected at 25 °C without applying a stopped–flow time and the spectrum obtained was used as the background. In this way a short and hence negligible reaction time (15 s) was obtained. For each temperature of the two calibration curves a sample was measured, evaluating its activity by using the corresponding calibration curve.To investigate the influence of the matrix on the results of the developed method, standard additions were performed for each sample, adding increments of 125, 250 and 500 U l21 amyloglucosidase. The results of the analysis of the spiked real samples are given in Table 1. The FIA–FTIR approach offers an effective way of eliminating the matrix, as can be seen from the results in Table 1. The lower recoveries obtained when using the calibration curves established from pure aqueous enzyme standards can be explained by a matrix effect.Especially glucose, which is present in the broth, is known to be a competitive inhibitor of the enzyme. The effect of inhibitors will be more pronounced for lower concentrations, which can also be seen in Table 1. However, the successful application of standard additions shows that the developed method can be used for the determination of amyloglucosidase activity in real samples. Conclusions An automated system capable of the determination of amyloglucosidase activity in aqueous solution based on the coupling of FIA and FTIR spectrometry was developed. The FIA–FTIR system is based on direct monitoring of starch hydrolysis catalyzed by amyloglucosidase hence eliminating the need for consecutive indicator reactions or expensive, specially designed enzyme substrates usually needed to assay amyloglucosidase activity.The coupling of the FTIR spectrometer with a flow system capable of performing chemical reactions in an automated, repeatable and reproducible way proved to enhance the detection power of the FTIR instrument, as using this approach a parameter which is not accessible by a single FTIR measurement could be successfully determined. Furthermore, this coupling has demonstrated new possibilities for the application of mid-IR FTIR spectrometry for remote process monitoring and process control. References 1 Enzyme Handbook, ed. Schomburg, D., and Salzmann, M., Springer, Berlin, 1991, p. 3.2.1.3. 2 Enzymes, Biomass, Food and Feed, in Biotechnology, ed. Reed, G., and Nagodawithana, T. W., 2nd edn., VCH, Weinheim, 1995. 3 Methods in Enzymatic Analysis, ed. Bergmeyer, H. U., VCH, Weinheim, 3rd edn., 1984, vol. IV. 4 Elder, M. T., and Montgomery, R. S., J. AOAC Int., 1995, 78(2), 398. 5 Holm, K. A., Anal. Chim. Acta, 1980, 117, 359. 6 Holm, K. A., Analyst, 1986, 111, 927. 7 Hartford, C. G., Muntz, D. W., Evans, J. V., and Blair, G. T., J. Food Compos. Anal., 1989, 2(4), 364. 8 Hansen, E. H., Willumsen, B., Winther, S. K., and Drabøl, H., Talanta, 1994, 41(11), 1881. 9 Rosenberg, E., and Kellner, R., J. Mol. Struct., 1993, 294, 9. 10 Lendl, B., and Kellner, R., Mikrochim. Acta, 1995, 119, 73. 11 Krieg, P., Lendl, B., Vonach, R., and Kellner, R., Fresenius’ Z. Anal. Chem., 1996, 356, 507. 12 O’Conner, C. J., and Cleverly, D. R., J. Chem. Technol Biotechnol., 1994, 61, 209. 13 Hineno, M., Carbohydr. Res., 1977, 56, 219. 14 Back, D. M., Michalska, D. F., and Polavarapu, P. L., Appl. Spectrosc., 1984, 38(2), 173. 15 Bellon-Maurell, V., Vallant, C., and Goffinet, D., Appl. Spectrosc., 1995, 49, 556. Paper 7/00432J Received January 20, 1997 Accepted March 12, 1997 Table 1: Results for spiked real samples Temperature 37 °C 54 °C Activity added/U l21 500 200 Activity found with standard addition/ U l21 509 196 Recovery (%) 101.8 98.2 Slope of calibration curve for standard addition, b/U l21 2.23 3 1025 4.55 3 1025 Activity found with calibration curve/ U l21 479 180 Recovery (%) 95.8 90.0 534 Analyst, June 1997, Vol. 122

ISSN:0003-2654    

DOI:10.1039/a700432j

出版商:RSC    

年代:1997

数据来源: RSC

摘要:

Determination of Iodine in Milk and Oyster Tissue Samples Using Combustion and Peroxydisulfate Oxidation F. Gua, A. A. Marchetti*b and T. Straumeb a Shanghai Institute of Radiation Medicine, Shanghai 200032, China b Health and Ecological Assessment Division, Lawrence Livermore National Laboratory, P.O. Box 808, Livermore, CA 94551, USA Two methods are described for the preparation of samples for total iodine measurement in biological matrices. In the first method, the samples were combusted in a stream of oxygen to release iodine that, subsequently, was trapped in a solution as iodide.The second method is a new approach in which the samples were oxidized in a basic solution of peroxydisulfate. In this case, the iodine was retained in solution as iodate. Total iodine was measured by gas chromatographic analysis of the 2-iodopentan-3-one derivative. The methods were tested using Standard Reference Materials (SRMs) 1549 Non-Fat Milk Powder, and 1566a and 1566 Oyster Tissue.Also, KI and KIO3 were used for testing the procedures. The results obtained for the SRMs, given as average ± standard deviation in mg g21, were: 3.39 ± 0.14 and 3.40 ± 0.23 for SRM 1549; 4.60 ± 0.42 and 4.51 ± 0.45 for SRM 1566a; and 2.84 ± 0.16 and 2.76 ± 0.06 for SRM 1566; values corresponding to combustion and wet oxidation, respectively. Overall, the absolute recoveries varied between 91 and 103%. These methods can also be used in the preparation of targets for the measurement of 129I using accelerator mass spectrometry.Keywords: Iodine; iodine-129; gas chromatography; combustion; peroxydisulfate; milk; oyster tissue Iodine is an essential micronutrient. Deficiency of iodine has been associated with thyroid diseases. Therefore, measurements of iodine in biological samples1,2 as well as in foods3 are of considerable interest as they can be utilized as intake monitors. In addition, it has been suggested that the concentration of iodine, among other elements, in thyroid tissue could be used as a chemical marker for cancer diagnosis.4,5 Radioiodine released into the environment by accidents such as Chernobyl has resulted in a significant dose to the thyroid in exposed populations and subsequent production of thyroid cancers.6 In those cases, short-lived 131I was mainly responsible for the dose.However, measurements of long-lived 129I may be used in retrospective assessments of deposition patterns and intensities of shorter-lived radioiodines.7 Because 129I can be measured with very high sensitivity using accelerator mass spectrometry (AMS), there is interest in the application of this isotope as an environmental, geochemical and biological tracer.Most iodine in biological matrices is covalently bonded and requires mineralization prior to analysis. This is a critical step because iodine can be lost easily by volatilization during procedures such as ashing. Also, iodine has three oxidation states that are stable in aqueous solution (iodide, iodine and iodate).Molecular iodine is not very soluble in water, is volatile and associates readily with organic matter. Solutions resulting from mineralization of a biological matrix can be complex and have to be conditioned carefully to avoid iodine losses. There are relatively few certified standards for iodine in biological and other complex matrices.3,8 Independent methods are necessary to establish reliable concentrations of iodine in these types of matrices. In this work, two different methods of sample preparation are presented. The first method consists in combusting the sample in a stream of oxygen and collecting the iodine liberated in a trapping solution.The second method is a new approach using peroxydisulfate in basic solution to oxidize the sample matrix and transform the total iodine present to iodate in solution. Sample solutions resulting from both methods were analyzed for iodine by gas chromatography,9–13 after preparation and extraction of the 2-iodopentan-3-one derivative. Standard Reference Materials (SRMs) 1549 Non-Fat Milk Powder, and 1566a and 1566 Oyster Tissue from the National Institute of Standards and Technology (NIST) were used for testing the procedures.In addition, the peroxydisulfate method was tested using KI and KIO3; a similar test was conducted for the combustion method in a previous study.14 Experimental Sample Combustion All reagents used were of analytical-reagent grade and solutions were prepared with de-ionized water.The hexane was Baxter B&J High Purity Solvent UV grade and the pentan-3-one was EM Science technical grade. The combustion apparatus has been described previously as applied to soil samples.14 Combustions were carried out in a 1000 °C Blue M tube furnace with an attached small auxiliary furnace (Watlow ceramic fiber heater). Samples were weighed in either porcelain or quartz combustion boats and placed in a 2.5 cm diameter quartz tube at the center of the furnace.Oxygen was supplied at a flow rate of about 80 ml min21. Downstream from the sample, there was a 10 cm long quartz wool plug which was pre-heated at 1000 °C with the auxiliary furnace. Further downsteam, the quartz tube was connected to a gas wash-bottle filled with 75 ml of trapping solution. Sample masses were kept between 0.25 and 0.50 g for Non- Fat Milk Powder and between 0.10 and 0.25 g for Oyster Tissue.The temperature ramp had to be controlled carefully to avoid a sudden ignition. During preliminary tests, sudden ignitions caused the quartz wool plug to move downstream away from the auxiliary furnace and possibly some back flow. These tests resulted in very little or no iodine being detected. The same effect was noted whenever a discoloration appeared in the quartz tube due to condensation of partially combusted products. That is, a complete and smooth combustion was necessary for a good recovery of iodine in our system.The following conditions, determined by trial-and-error, resulted in reproducible results and good recoveries. The initial furnace temperature was set to 180 °C. For Oyster Tissue, the temperature was raised at 3 °C min21 to 300 °C and then at 30 °C min21 to 1000 °C. For Non-Fat Milk Powder, the temperature was raised at 1 °C min21 to 300 °C and then at Analyst, June 1997, Vol. 122 (535–537) 53510 °C min21 to 1000 °C. The temperature was kept at 1000 °C for 90 min. The trapping solution consisted of 75 ml of 0.1 m KOH in which a weighed amount of Na2SO3 was dissolved. There should be sufficient Na2SO3 in the trapping solution to assure the reduction to iodide of all the iodine carried by the gas stream. However, a large excess is not desirable since it would have to be eliminated in the oxidation step of the iodine derivative preparation. The presence of sulfite after combustion was checked by the reaction of Å 1 ml of the trapping solution with 1–2 drops of 0.1 m KMnO4 in Å 1 m H2SO4.Tests were conducted using increments of 0.5 g of Na2SO3 in the trapping solution. It was found that Non-Fat Milk Powder samples required about 6 g of Na2SO3 to give a positive reaction while Oyster Tissue samples required about 1.5 g. After the combustion had been completed, the trapping solution was transferred into a 100 ml calibrated flask and diluted to the mark with water.Peroxydisulfate Oxidation The decomposition of peroxydisulfate ion in solution occurs with the net production of two protons according to: S2O8 22+H2O?2SO4 22 + 2H++1 2O2 Because iodine is retained better in basic than in acidic solutions, approximately double the amount of base required to neutralize the protons was added initially. It was expected that the strong oxidizing and basic conditions would transform the iodine present in the sample to iodate in solution.The amount of peroxydisulfate required to oxidize a given mass of sample was calculated roughly considering a minimum of 1.5 peroxydisulfate ions per carbon atom for mineralization, as estimated in the determination of dissolved organic carbon in waters.15 It was assumed that the samples contain about 80% of carbon and the reagent was added in about 30% excess of the calculated value. The following procedure applied to a mass of Å 0.1 g of Oyster Tissue or Non-Fat Milk Powder.A 250 ml flask with a standard tapered joint was filled with 80 ml of water and placed on a combined heater and magnetic stirrer plate; 3.4 g of KOH were added and the stirrer was started. An aliquot of sample was weighed and incorporated into the solution. Once the sample was distributed homogeneously throughout the solution, 4 g of K2S2O8 were added and the mixture was heated to a gentle boil for about 60 min. A reflux condenser was connected to the flask to avoid iodine losses during the heating step.It took 30–40 min for the solution to become clear. The solution was transferred into a 100 ml calibrated flask along with 1.0 g of Na2SO3 to reduce the iodate to iodide. The flask was filled to the mark with water. Tests performed using KI and KIO3 were carried out in the same manner except that the sample was substituted by an aliquot of KI or KIO3 solution containing 0.80 mg of iodine. When samples of Å 0.25 g were prepared, the reagent amounts were increased proportionally to the sample mass.The volume of the final solution also had to be increased to 250 ml to prevent the formation of precipitates. Measurement of Iodine Details of the preparation of the iodo derivative and its determination by gas chromatography have been described elsewhere.14,16 The same procedure was applied to the final sample solutions from both preparation methods. To prepare the iodo derivative, a 25 ml aliquot of sample solution was combined with 1 ml of 4% pentan-3-one, 1 ml of 5 m H2SO4 and 1 ml of 30% H2O2.Because the solutions obtained by peroxydisulfate oxidation were more basic than those obtained by combustion, 2 ml of 5 m H2SO4 were added to ensure a pH of about 1. After 10 min, the 2-iodopentan-3-one formed was extracted in 10 ml of hexane. The iodo derivative was measured using a Hewlett-Packard 5880A gas chromatograph with a J&W Scientific DB-624 megabore capillary column (30 m 3 0.53 mm id, 3 mm film) and a 63Ni electron-capture detector.The carrier gas was He (5.45 ml min21) and the make-up gas for the detector was Ar (5% CH4). The injector port temperature was 150 °C. The detector was operated at 300 °C. The oven temperature profile was 1 min at 50 °C, ramped at 10 °C min21 to 150 °C, and 1 min at 150 °C. The samples were introduced by splitless injection of 2 ml of the hexane extract using a Hewlett- Packard 7673 automatic sampler.Peak areas were measured using a Hewlett-Packard 3396A integrator. Three 2 ml injections per extract were performed. Each sample area was determined by averaging the six areas corresponding to the extracts of two 25 ml aliquots. To calibrate the instrument, standards were prepared by dilution of a stock solution containing 1000 mg ml21 iodine as KI. Amounts of iodine varying between 0.02 and 1 mg were taken to 25 ml with water and processed as indicated for the sample aliquots.The peak area was plotted as a function of the total iodine present in the 25 ml aliquot. The linear calibration equation was obtained by weighted least squares. The practical quantification limit was Å 0.02 mg of iodine in a 25 ml aliquot, i.e., a concentration of iodine of Å 6 3 1029 m. Blank determinations were performed using both methods. With the combustion method, iodine could not be detected in any of the blanks. On the other hand, with the peroxydisulfate method, small iodine peaks were observed ( Å 0.01 mg) in the blanks.In the latter case, an average blank area was subtracted from the measured sample areas. The uncertainties in the peak areas introduced by the apparatus are less than 1% relative standard deviation. Uncertainties due to the derivatization and extraction procedure are estimated at Å 1% relative standard deviation. Chromatograms of a standard, a sample prepared using the combustion method and a sample prepared using the peroxydisulfate method are shown in Fig. 1. Results and Discussion The results of the iodine determinations using both sample preparation methods are summarized in Table 1. These are presented as the average and standard deviation of the given number of samples analyzed. The results for the SRMs have been corrected to a dry mass basis as indicated in the certificates of analysis. The mass loss from drying was 4.20, 4.01 and 4.09%, for SRMs 1549, 1566a and 1566, respectively. There is a very good agreement between the two methods and the reference values, as seen in Table 1.The tests conducted for the Fig. 1 Chromatograms corresponding to 2 ml injections of hexane extracts of (a) a standard containing 0.4 mg of iodine in 25 ml, (b) a 0.56 g sample of Non-Fat Milk Powder (SRM 1549) prepared using the combustion method, and (c) a 0.14 g sample of Non-Fat Milk Powder (SRM 1549) prepared using the peroxydisulfate method. The peaks marked by an arrow correspond to the 2-iodopentan-3-one derivative. 536 Analyst, June 1997, Vol. 122peroxydisulfate method using 0.80 mg of iodine as KI and KIO3 resulted in 0.799 ± 0.025 (n = 5) and 0.763 ± 0.023 (n = 5) mg, respectively. Previous results of a similar test performed using the combustion method14 are included in Table 1 for completeness. In general, the recoveries vary between 91 and 103%. The combustion method was used previously by our group in the determination of total iodine and 129I in soil samples.7,14 In extending this method to organic matrices, the procedure had to be modified by trial-and-error so that the combustion would proceed smoothly and thoroughly to obtain a good recovery of iodine.Since these are constraints inherent to this particular apparatus, it could be possible to modify the furnace set-up to accommodate larger samples and reduce the time for sample preparation. Sulfite in the trapping solution had to be in excess during the combustion procedure to assure the reduction of iodine to iodide.However, the amount of sulfite needed experimentally was much larger than that required by the reduction of iodine alone. A possible explanation is that the oxygen that flowed through the trapping solution oxidized part of the sulfite. In addition, there may be other reactions involving combustion products from the samples. Further studies are needed to understand the mechanism of sulfite depletion. Overall, the combustion method has shown very good recoveries for iodine with various matrices.The new approach to separate iodine by wet oxidation using peroxydisulfate seems to be a good alternative to combustion, in particular with organic matrices. It has also shown good recoveries and it is very simple to implement. It is important to note that tracers were not used to estimate recoveries and correct sample concentrations; therefore, the results reflect the actual yields. Finally, both methods result in final solutions of iodine as iodide that are suitable for the precipitation of AgI which is the target material for the measurement of 129I by AMS.Because iodine in the sample is present at the microgram level, AgI is precipitated generally after the addition of a few milligrams of 127I carrier for handling purposes. The combustion method has been applied successfully to this measurement7,14 and the same is expected from the peroxydisulfate method. This work was performed under the auspices of the United States Department of Energy by the Lawrence Livermore Natinal Laboratory under contract W-7405-En-48, with support from the Office of International Health Programs, US Department of Energy.References 1 Spate, V. L., Morris, J. S., Chickos, S., Baskett, C. K., Mason, M. M., Cheng, T. P., Reams, C. L., West, C., Furnee, C., Willett, W., and Horn-Ross, P., J. Radioanal. Nucl. Chem., 1995, 195, 21. 2 Mason, M. M., Spate, V. L., Morris, J.S., Baskett, C. K., Cheng, T. P., Reams, C. L., Le Marchand, L., Henderson, B. E., and Kolonel, L. N., J. Radioanal. Nucl. Chem., 1995, 195, 57. 3 Heckman, M. M., J. Assoc. Off. Anal. Chem., 1979, 62, 1045. 4 Nishida, M., Sakurai, H., Tezuka, U., Kawada, J., Koyama, M., and Takada, J., Clin. Chim. Acta, 1990, 187, 181. 5 Zaichick, V. Y., Tsyb, A. F., and Vtyurin, B. M., Analyst, 1995, 120, 817. 6 Williams, D., Nature (London), 1994, 371, 556. 7 Straume, T., Marchetti, A.A., Anspaugh, L. R., Khrouch, V. T., Gavrilin, Y. I., Shinkarev, S. M., Drozdovitch, V. V., Ulanovsky, A. V., Korneev, S. V., Brekeshev, M. K., Leonov, E. S., Voigt, G., Pachenko, S. V., and Minenko, V. F., Health Phys., 1996, 71, 733. 8 Norman, B. R., and Iyengar, G. V., Fresenius’ J. Anal. Chem., 1994, 348, 430. 9 Hasty, R. A., Mikrochim. Acta, 1971, 348. 10 Hasty, R. A., Mikrochim. Acta, 1973, 621. 11 Grys, S., J. Chromatogr., 1974, 100, 43. 12 Maros, L., K�aldy, M., and Igaz, S., Anal.Chem., 1989, 61, 733. 13 Mitsuhashi, T., and Kaneda, Y., J. Assoc. Off. Anal. Chem., 1990, 73, 790. 14 Marchetti, A. A., Gu, F., Robl, R., and Straume, T., Nucl. Instrum. Methods, in the press. 15 Peyton, G. R., Mar. Chem., 1993, 41, 91. 16 Marchetti, A. A., Rose, L., and Straume, T., Anal. Chim. Acta, 1994, 296, 243. Paper 6/07555J Received November 6, 1996 Accepted February 26, 1997 Table 1 Measurements of total iodine* Measured Sample Reference value/ reference Sample n Method† size/g Measured/mg g21 mg g21 (%) NIST SRM 1549 8 C 0.25–0.50 3.39 ± 0.14 3.38 ± 0.02 100 Non-Fat Milk Powder 6 P 0.12–0.22 3.40 ± 0.23 101 NIST SRM 1566a 5 C 0.10–0.25 4.60 ± 0.42 4.46 ± 0.42 103 Oyster Tissue 6 P 0.10–0.25 4.51 ± 0.45 101 NIST SRM 1566 7 C 0.10–0.14 2.84 ± 0.16 2.8‡ 101 Oyster Tissue 6 P 0.12–0.14 2.76 ± 0.06 99 KI 8 C 1.82 ± 0.12 mg§ 2.00 mg 91 ‘Baker Analyzed’ 5 P 0.799 ± 0.025 mg 0.80 mg 100 KIO3 5 C 1.85 ± 0.06 mg§ 2.00 mg 93 ‘Baker Analyzed’ 5 P 0.763 ± 0.023 mg 0.80 mg 95 *Iodine was determined by reaction with pentan-3-one, extraction in hexane and measurement by gas chromatography.The results are expressed as the average ± standard deviation for a number of samples n. † Sample preparation method: combustion (C), peroxydisulfate oxidation (P). ‡ Non-certified value. § The results corresponding to the combustion method are from Marchetti et al.14 Analyst, June 1997, Vol. 122 537 Determination of Iodine in Milk and Oyster Tissue Samples Using Combustion and Peroxydisulfate Oxidation F.Gua, A. A. Marchetti*b and T. Straumeb a Shanghai Institute of Radiation Medicine, Shanghai 200032, China b Health and Ecological Assessment Division, Lawrence Livermore National Laboratory, P.O. Box 808, Livermore, CA 94551, USA Two methods are described for the preparation of samples for total iodine measurement in biological matrices. In the first method, the samples were combusted in a stream of oxygen to release iodine that, subsequently, was trapped in a solution as iodide. The second method is a new approach in which the samples were oxidized in a basic solution of peroxydisulfate.In this case, the iodine was retained in solution as iodate. Total iodine was measured by gas chromatographic analysis of the 2-iodopentan-3-one derivative. The methods were tested using Standard Reference Materials (SRMs) 1549 Non-Fat Milk Powder, and 1566a and 1566 Oyster Tissue. Also, KI and KIO3 were used for testing the procedures.The results obtained for the SRMs, given as average ± standard deviation in mg g21, were: 3.39 ± 0.14 and 3.40 ± 0.23 for SRM 1549; 4.60 ± 0.42 and 4.51 ± 0.45 for SRM 1566a; and 2.84 ± 0.16 and 2.76 ± 0.06 for SRM 1566; values corresponding to combustion and wet oxidation, respectively. Overall, the absolute recoveries varied between 91 and 103%. These methods can also be used in the preparation of targets for the measurement of 129I using accelerator mass spectrometry.Keywords: Iodine; iodine-129; gas chromatography; combustion; peroxydisulfate; milk; oyster tissue Iodine is an essential micronutrient. Deficiency of iodine has been associated with thyroid diseases. Therefore, measurements of iodine in biological samples1,2 as well as in foods3 are of considerable interest as they can be utilized as intake monitors. In addition, it has been suggested that the concentration of iodine, among other elements, in thyroid tissue could be used as a chemical marker for cancer diagnosis.4,5 Radioiodine released into the environment by accidents such as Chernobyl has resulted in a significant dose to the thyroid in exposed populations and subsequent production of thyroid cancers.6 In those cases, short-lived 131I was mainly responsible for the dose.However, measurements of long-lived 129I may be used in retrospective assessments of deposition patterns and intensities of shorter-lived radioiodines.7 Because 129I can be measured with very high sensitivity using accelerator mass spectrometry (AMS), there is interest in the application of this isotope as an environmental, geochemical and biological tracer.Most iodine in biological matrices is covalently bonded and requires mineralization prior to analysis. This is a critical step because iodine can be lost easily by volatilization during procedures such as ashing. Also, iodine has three oxidation states that are stable in aqueous solution (iodide, iodine and iodate).Molecular iodine is not very soluble in water, is volatile and associates readily with organic matter. Solutions resulting from mineralization of a biological matrix can be complex and have to be conditioned carefully to avoid iodine losses. There are relatively few certified standards for iodine in biological and other complex matrices.3,8 Independent methods are necessary to establish reliable concentrations of iodine in these types of matrices.In this work, two different methods of sample preparation are presented. The first method consists in combusting the sample in a stream of oxygen and collecting the iodine liberated in a trapping solution. The second method is a new approach using peroxydisulfate in basic solution to oxidize the sample matrix and transform the total iodine present to iodate in solution. Sample solutions resulting from both methods were analyzed for iodine by gas chromatography,9–13 after preparation and extraction of the 2-iodopentan-3-one derivative.Standard Reference Materials (SRMs) 1549 Non-Fat Milk Powder, and 1566a and 1566 Oyster Tissue from the National Institute of Standards and Technology (NIST) were used for testing the procedures. In addition, the peroxydisulfate method was tested using KI and KIO3; a similar test was conducted for the combustion method in a previous study.14 Experimental Sample Combustion All reagents used were of analytical-reagent grade and solutions were prepared with de-ionized water.The hexane was Baxter B&J High Purity Solvent UV grade and the pentan-3-one was EM Science technical grade. The combustion apparatus has been described previously as applied to soil samples.14 Combustions were carried out in a 1000 °C Blue M tube furnace with an attached small auxiliary furnace (Watlow ceramic fiber heater). Samples were weighed in either porcelain or quartz combustion boats and placed in a 2.5 cm diameter quartz tube at the center of the furnace.Oxygen was supplied at a flow rate of about 80 ml min21. Downstream from the sample, there was a 10 cm long quartz wool plug which was pre-heated at 1000 °C with the auxiliary furnace. Further downsteam, the quartz tube was connected to a gas wash-bottle filled with 75 ml of trapping solution. Sample masses were kept between 0.25 and 0.50 g for Non- Fat Milk Powder and between 0.10 and 0.25 g for Oyster Tissue.The temperature ramp had to be controlled carefully to avoid a sudden ignition. During preliminary tests, sudden ignitions caused the quartz wool plug to move downstream away from the auxiliary furnace and possibly some back flow. These tests resulted in very little or no iodine being detected. The same effect was noted whenever a discoloration appeared in the quartz tube due to condensation of partially combusted products. That is, a complete and smooth combustion was necessary for a good recovery of iodine in our system.The following condions, determined by trial-and-error, resulted in reproducible results and good recoveries. The initial furnace temperature was set to 180 °C. For Oyster Tissue, the temperature was raised at 3 °C min21 to 300 °C and then at 30 °C min21 to 1000 °C. For Non-Fat Milk Powder, the temperature was raised at 1 °C min21 to 300 °C and then at Analyst, June 1997, Vol. 122 (535–537) 53510 °C min21 to 1000 °C.The temperature was kept at 1000 °C for 90 min. The trapping solution consisted of 75 ml of 0.1 m KOH in which a weighed amount of Na2SO3 was dissolved. There should be sufficient Na2SO3 in the trapping solution to assure the reduction to iodide of all the iodine carried by the gas stream. However, a large excess is not desirable since it would have to be eliminated in the oxidation step of the iodine derivative preparation. The presence of sulfite after combustion was checked by the reaction of Å 1 ml of the trapping solution with 1–2 drops of 0.1 m KMnO4 in Å 1 m H2SO4.Tests were conducted using increments of 0.5 g of Na2SO3 in the trapping solution. It was found that Non-Fat Milk Powder samples required about 6 g of Na2SO3 to give a positive reaction while Oyster Tissue samples required about 1.5 g. After the combustion had been completed, the trapping solution was transferred into a 100 ml calibrated flask and diluted to the mark with water.Peroxydisulfate Oxidation The decomposition of peroxydisulfate ion in solution occurs with the net production of two protons according to: S2O8 22+H2O?2SO4 22 + 2H++1 2O2 Because iodine is retained better in basic than in acidic solutions, approximately double the amount of base required to neutralize the protons was added initially. It was expected that the strong oxidizing and basic conditions would transform the iodine present in the sample to iodate in solution.The amount of peroxydisulfate required to oxidize a given mass of sample was calculated roughly considering a minimum of 1.5 peroxydisulfate ions per carbon atom for mineralization, as estimated in the determination of dissolved organic carbon in waters.15 It was assumed that the samples contain about 80% of carbon and the reagent was added in about 30% excess of the calculated value. The following procedure applied to a mass of Å 0.1 g of Oyster Tissue or Non-Fat Milk Powder.A 250 ml flask with a standard tapered joint was filled with 80 ml of water and placed on a combined heater and magnetic stirrer plate; 3.4 g of KOH were added and the stirrer was started. An aliquot of sample was weighed and incorporated into the solution. Once the sample was distributed homogeneously throughout the solution, 4 g of K2S2O8 were added and the mixture was heated to a gentle boil for about 60 min. A reflux condenser was connected to the flask to avoid iodine losses during the heating step.It took 30–40 min for the solution to become clear. The solution was transferred into a 100 ml calibrated flask along with 1.0 g of Na2SO3 to reduce the iodate to iodide. The flask was filled to the mark with water. Tests performed using KI and KIO3 were carried out in the same manner except that the sample was substituted by an aliquot of KI or KIO3 solution containing 0.80 mg of iodine. When samples of Å 0.25 g were prepared, the reagent amounts were increased proportionally to the sample mass.The volume of the final solution also had to be increased to 250 ml to prevent the formation of precipitates. Measurement of Iodine Details of the preparation of the iodo derivative and its determination by gas chromatography have been described elsewhere.14,16 The same procedure was applied to the final sample solutions from both preparation methods. To prepare the iodo derivative, a 25 ml aliquot of sample solution was combined with 1 ml of 4% pentan-3-one, 1 ml of 5 m H2SO4 and 1 ml of 30% H2O2.Because the solutions obtained by peroxydisulfate oxidation were more basic than those obtained by combustion, 2 ml of 5 m H2SO4 were added to ensure a pH of about 1. After 10 min, the 2-iodopentan-3-one formed was extracted in 10 ml of hexane. The iodo derivative was measured using a Hewlett-Packard 5880A gas chromatograph with a J&W Scientific DB-624 megabore capillary column (30 m 3 0.53 mm id, 3 mm film) and a 63Ni electron-capture detector.The carrier gas was He (5.45 ml min21) and the make-up gas for the detector was Ar (5% CH4). The injector port temperature was 150 °C. The detector was operated at 300 °C. The oven temperature profile was 1 min at 50 °C, ramped at 10 °C min21 to 150 °C, and 1 min at 150 °C. The samples were introduced by splitless injection of 2 ml of the hexane extract using a Hewlett- Packard 7673 automatic sampler.Peak areas were measured using a Hewlett-Packard 3396A integrator. Three 2 ml injections per extract were performed. Each sample area was determined by averaging the six areas corresponding to the extracts of two 25 ml aliquots. To calibrate the instrument, standards were prepared by dilution of a stock solution containing 1000 mg ml21 iodine as KI. Amounts of iodine varying between 0.02 and 1 mg were taken to 25 ml with water and processed as indicated for the sample aliquots.The peak area was plotted as a function of the total iodine present in the 25 ml aliquot. The linear calibration equation was obtained by weighted least squares. The practical quantification limit was Å 0.02 mg of iodine in a 25 ml aliquot, i.e., a concentration of iodine of Å 6 3 1029 m. Blank determinations were performed using both methods. With the combustion method, iodine could not be detected in any of the blanks. On the other hand, with the peroxydisulfate method, small iodine peaks were observed ( Å 0.01 mg) in the blanks.In the latter case, an average blank area was subtracted from the measured sample areas. The uncertainties in the peak areas introduced by the apparatus are less than 1% relative standard deviation. Uncertainties due to the derivatization and extraction procedure are estimated at Å 1% relative standard deviation. Chromatograms of a standard, a sample prepared using the combustion method and a sample prepared using the peroxydisulfate method are shown in Fig. 1. Results and Discussion The results of the iodine determinations using both sample preparation methods are summarized in Table 1. These are presented as the average and standard deviation of the given number of samples analyzed. The results for the SRMs have been corrected to a dry mass basis as indicated in the certificates of analysis. The mass loss from drying was 4.20, 4.01 and 4.09%, for SRMs 1549, 1566a and 1566, respectively. There is a very good agreement between the two methods and the reference values, as seen in Table 1.The tests conducted for the Fig. 1 Chromatograms corresponding to 2 ml injections of hexane extracts of (a) a standard containing 0.4 mg of iodine in 25 ml, (b) a 0.56 g sample of Non-Fat Milk Powder (SRM 1549) prepared using the combustion method, and (c) a 0.14 g sample of Non-Fat Milk Powder (SRM 1549) prepared using the peroxydisulfate method. The peaks marked by an arrow correspond to the 2-iodopentan-3-one derivative. 536 Analyst, June 1997, Vol. 122peroxydisulfate method using 0.80 mg of iodine as KI and KIO3 resulted in 0.799 ± 0.025 (n = 5) and 0.763 ± 0.023 (n = 5) mg, respectively. Previous results of a similar test performed using the combustion method14 are included in Table 1 for completeness. In general, the recoveries vary between 91 and 103%. The combustion method was used previously by our group in the determination of total iodine and 129I in soil samples.7,14 In extending this method to organic matrices, the procedure had to be modified by trial-and-error so that the combustion would proceed smoothly and thoroughly to obtain a good recovery of iodine.Since these are constraints inherent to this particular apparatus, it could be possible to modify the furnace set-up to accommodate larger samples and reduce the time for sample preparation. Sulfite in the trapping solution had to be in excess during the combustion procedure to assure the reduction of iodine to iodide.However, the amount of sulfite needed experimentally was much larger than that required by the reduction of iodine alone. A possible explanation is that the oxygen that flowed through the trapping solution oxidized part of the sulfite. In addition, there may be other reactions involving combustion products from the samples. Further studies are needed to understand the mechanism of sulfite depletion. Overall, the combustion method has shown very good recoveries for iodine with various matrices.The new approach to separate iodine by wet oxidation using peroxydisulfate seems to be a good alternative to combustion, in particular with organic matrices. It has also shown good recoveries and it is very simple to implement. It is important to note that tracers were not used to estimate recoveries and correct sample concentrations; therefore, the results reflect the actual yields. Finally, both methods result in final solutions of iodine as iodide that are suitable for the precipitation of AgI which is the target material for the measurement of 129I by AMS.Because iodine in the sample is present at the microgram level, AgI is precipitated generally after the addition of a few milligrams of 127I carrier for handling purposes. The combustion method has been applied successfully to this measurement7,14 and the same is expected from the peroxydisulfate method. This work was performed under the auspices of the United States Department of Energy by the Lawrence Livermore Natinal Laboratory under contract W-7405-En-48, with support from the Office of International Health Programs, US Department of Energy. References 1 Spate, V.L., Morris, J. S., Chickos, S., Baskett, C. K., Mason, M. M., Cheng, T. P., Reams, C. L., West, C., Furnee, C., Willett, W., and Horn-Ross, P., J. Radioanal. Nucl. Chem., 1995, 195, 21. 2 Mason, M. M., Spate, V. L., Morris, J.S., Baskett, C. K., Cheng, T. P., Reams, C. L., Le Marchand, L., Henderson, B. E., and Kolonel, L. N., J. Radioanal. Nucl. Chem., 1995, 195, 57. 3 Heckman, M. M., J. Assoc. Off. Anal. Chem., 1979, 62, 1045. 4 Nishida, M., Sakurai, H., Tezuka, U., Kawada, J., Koyama, M., and Takada, J., Clin. Chim. Acta, 1990, 187, 181. 5 Zaichick, V. Y., Tsyb, A. F., and Vtyurin, B. M., Analyst, 1995, 120, 817. 6 Williams, D., Nature (London), 1994, 371, 556. 7 Straume, T., Marchetti, A. A., Anspaugh, L. R., Khrouch, V. T., Gavrilin, Y. I., Shinkarev, S. M., Drozdovitch, V. V., Ulanovsky, A. V., Korneev, S. V., Brekeshev, M. K., Leonov, E. S., Voigt, G., Pachenko, S. V., and Minenko, V. F., Health Phys., 1996, 71, 733. 8 Norman, B. R., and Iyengar, G. V., Fresenius’ J. Anal. Chem., 1994, 348, 430. 9 Hasty, R. A., Mikrochim. Acta, 1971, 348. 10 Hasty, R. A., Mikrochim. Acta, 1973, 621. 11 Grys, S., J. Chromatogr., 1974, 100, 43. 12 Maros, L., K�aldy, M., and Igaz, S., Anal. Chem., 1989, 61, 733. 13 Mitsuhashi, T., and Kaneda, Y., J. Assoc. Off. Anal. Chem., 1990, 73, 790. 14 Marchetti, A. A., Gu, F., Robl, R., and Straume, T., Nucl. Instrum. Methods, in the press. 15 Peyton, G. R., Mar. Chem., 1993, 41, 91. 16 Marchetti, A. A., Rose, L., and Straume, T., Anal. Chim. Acta, 1994, 296, 243. Paper 6/07555J Received November 6, 1996 Accepted February 26, 1997 Table 1 Measurements of total iodine* Measured Sample Reference value/ reference Sample n Method† size/g Measured/mg g21 mg g21 (%) NIST SRM 1549 8 C 0.25–0.50 3.39 ± 0.14 3.38 ± 0.02 100 Non-Fat Milk Powder 6 P 0.12–0.22 3.40 ± 0.23 101 NIST SRM 1566a 5 C 0.10–0.25 4.60 ± 0.42 4.46 ± 0.42 103 Oyster Tissue 6 P 0.10–0.25 4.51 ± 0.45 101 NIST SRM 1566 7 C 0.10–0.14 2.84 ± 0.16 2.8‡ 101 Oyster Tissue 6 P 0.12–0.14 2.76 ± 0.06 99 KI 8 C 1.82 ± 0.12 mg§ 2.00 mg 91 ‘Baker Analyzed’ 5 P 0.799 ± 0.025 mg 0.80 mg 100 KIO3 5 C 1.85 ± 0.06 mg§ 2.00 mg 93 ‘Baker Analyzed’ 5 P 0.763 ± 0.023 mg 0.80 mg 95 *Iodine was determined by reaction with pentan-3-one, extraction in hexane and measurement by gas chromatography. The results are expressed as the average ± standard deviation for a number of samples n. † Sample preparation method: combustion (C), peroxydisulfate oxidation (P). ‡ Non-certified value. § The results corresponding to the combustion method are from Marchetti et al.14 Analyst, June 1997, Vol. 122 5

ISSN:0003-2654    

DOI:10.1039/a607555j

出版商:RSC    

年代:1997

数据来源: RSC

摘要:

Slurry Sampling for Hydride Generation Atomic Absorption Spectrometric Determination of Arsenic in Cigarette Tobaccos Jerzy Mierzwa†a, Samuel B. Adelojua and Harkirat S. Dhindsab a Centre for Electrochemical Research and Analytical Technology, and Department of Chemistry, University of Western Sydney, Nepean, P.O. Box 10, Kingswood, New South Wales 2747, Australia b Department of Chemistry, University of New England, Armidale, New South Wales 2351, Australia The development of a slurry sampling hydride generation atomic absorption spectrometric (HGAAS) method for the determination of arsenic in cigarette tobacco samples is described.The method is relatively simple and has been shown to give values of total arsenic close to those obtained using methods requiring total dissolution and decomposition of all vegetable matter before analysis. Pre-treatment of samples slurried in nitric acid by ultrasonication permitted the extraction of about 90% of the total arsenic from tobacco samples.Further improvement in the recovery efficiency (up to 93–94%) was accomplished by the use of an additional step of short microwave-accelerated treatment. l-Cysteine was used as a pre-reduction agent. The accuracy and precision of the slurry sampling HGAAS method were studied using the certified reference material (CRM) CTA-OTL-1 Oriental Tobacco Leaves. Under the optimum conditions, as little as 2.6 ng of arsenic can be detected. The relative standard deviation of the overall procedure was calculated to be below 7.6% at arsenic concentration levels of 0.5–0.9 mg kg21 and the analytical results obtained for the CRM agreed with the certified value.The main factors that influenced the reliability of the method were sample homogeneity, particle size and slurry concentration. Keywords: Arsenic; atomic absorption spectrometry; hydride generation; slurried sample; cigarette tobacco Arsenic is an element that has created considerable concern and interest in biological and environmental monitoring for many years (see, e.g., ref. 1). Comprehensive data on the toxicology of arsenic are available.2 In particular, the interest in determining arsenic in biological and environmental samples has accentuated the need for a rapid and sensitive analytical procedure for the control of the concentration of this element. Arsenic has been identified as a potential human carcinogen and smoking can be one of the pathways for the transfer of arsenic to humans; therefore, its control in tobacco samples is important.The possible sources of arsenic in plants (tobacco leaves) presumably include surface contamination by industrial activities, natural uptake of arsenic from soil, and even the use of arsenical pesticides in countries where these are still permitted.3 The transfer of elements from cigarette tobacco to smoke condensate has been studied, and it was concluded that the total content of arsenic was transferred to the smoke.4 Atomic absorption spectrometry, both with hydride generation (HGAAS) and electrothermal atomization (ETAAS), is widely used for the determination of arsenic at trace levels in plant samples.The use of either ETAAS5,6 or HGAAS7–11 for the determination of arsenic in these samples requires dissolution and decomposition of samples. There are, usually, timeconsuming procedures and a risk of contamination and/or loss of the element is associated with this mode of sample preparation.A slurry sampling technique combined with HGAAS has recently been proposed as an attractive alternative for avoiding the need for dissolution and mineralization steps. In the authors opinion, such an analytical technique is based on a very efficient extraction of the analyte into the liquid phase, but results reported suggested that analyte adsorbed onto solids can also take part in the hydride generation reaction. This approach has been successfully used for the determination of arsenic in light sandy soil,12 sewage sludge,12 fly ash,12–14 city waste incineration ash12 and diatomaceous earth.14 To the best of our knowledge, no work has been reported on the determination of arsenic in plant samples by slurry sampling combined with hydride generation.Some different potential problems might be expected with plant samples, e.g., tobacco leaf samples. Arsenic is likely to be encapsulated within cell walls, and it is unclear whether these would be broken down by the combined effects of dilute acid attack and ultrasonication.It is also not entirely certain if organic forms of arsenic would be broken down to inorganic arsenic unless a total oxidative acid digestion is carried out. This paper describes a rapid method for the determination of total arsenic in tobacco samples by a relatively simple pretreatment and further sampling of slurries combined with HGAAS. The reliable utilization of this method was considered by careful optimization of the main experimental parameters, e.g., sample grinding time, slurry ultrasonication time, and slurry and reagent concentrations.Experimental Apparatus All measurements were performed on a GBC Model 902 atomic absorption spectrometer (GBC, Dandenong, Victoria, Australia) operated in the single beam mode. A deuterium lamp background correction system was used. An arsenic hollow cathode lamp (Photron, Narre Warren, Victoria, Australia) operated at 10 mA was used as the primary radiation source. The 193.7 nm spectral line and a spectral band-width of 0.7 nm were selected. Slurry mixing was accomplished with an ultrasonic bath (Branson, Danbury, CT, USA) and an MT-19 vortex mixer (Chiltern Scientific, Sydney, NSW, Australia). A 650 W domestic microwave oven (NEC, Tokyo, Japan) was used for the additional pre-treatment of slurries.Slurries were manually injected into a laboratory-made batch hydride generation system connected to a conventional (glass) gas–liquid separator † On leave from: Central Laboratory, University of M.Curie-Sklodowska, P1 20-031 Lublin, Poland. Present address: Department of Nuclear Science, National Tsing-Hua University, 30043 Hsinchu, Taiwan. Analyst, June 1997, Vol. 122 (539–542) 539and electrically heated quartz T-tube. All measurements were based on peak area absorbance and a 30 s integration time. All glassware and plastic containers were soaked in 2 m nitric acid for 24 h and rinsed three times with Milli-Q water before use.Reagents and Samples Nitric acid (65%) of spectral purity (Merck, Darmstadt, Germany), Triton X-100 (Merck) and Milli-Q water were used. Working standards were obtained from stock solutions of arsenic(iii) by serial dilution with Milli-Q water and acidified with nitric acid. Sodium tetrahydroborate solution was prepared by dissolving the reagent in 1.0% sodium hydroxide and filtering through a 0.45 mm filter.This solution was stored for up to 1 week in a refrigerator. l-Cysteine (Merck) solution was prepared by dissolving the reagent in Milli-Q water. All reagents used were of analytical-reagent grade or better. A mixture of 99% argon +1% oxygen was used as the carrier gas. Two different brands of cigarette tobacco were analysed: the first was collected from cigarettes commercially manufactured in Poland (sample A) and the second from cigarettes manufactured in Australia (sample B).The certified reference material CTA-OTL-1 Oriental Tobacco Leaves (IChTJ, Warsaw, Poland) was used for the verification of the method. Analytical Procedure For each brand of cigarettes the tobacco content of a cigarette was mixed well and about 500–600 mg of tobacco were collected. Initially, all samples were dried for 12 h at a temperature of 85 °C. To ensure homogeneity, the samples were ground in an agate mortar for approximately 20 min.The effectiveness of grinding was controlled by observations under an optical microscope. After grinding, the average diameter for most of the particles was below 60 mm. Slurries in the range 0.5–2.5% m/v were then prepared. The ground samples were weighed accurately (50–250 mg of sample) into poly- (propylene) screw-cap bottles (volume, Å 15 ml) to which 10 ml of 6.5 m nitric acid with 0.005% Triton X-100 were added. A precise (±0.01 mg) electronic balance was used for all weighings.A volume/volume factor (volume of solid/volume of diluent) as proposed by Miller-Ihli15 was taken into account during slurry preparation. The slurries were treated for 40 s in a microwave oven (at full power mode, i.e., 650 W). [Warning: The caps of the poly(propylene) bottles were not screwed tight prior to the microwave treatment.] Slurries were then ultrasonicated for 12 min in an ultrasonic bath at 75–76 °C (maximum obtainable temperature for this bath).Powdered lcysteine was added prior to the slurry ultrasonication. Blanks were prepared in a similar fashion, but omitting samples. The slurries were vortex mixed for 5–10 s prior to each measurement. An aliquot (0.5–1 ml) of the slurry was then taken and placed in the reaction vessel together with 4 ml of water. A 0.5 ml volume of 1.2% NaBH4 solution was then added by manual pipetting. The carrier gas (a mixture of 99% Ar and 1% O2) was bubbled through the reaction vessel so that the arsenic hydride produced was swept to the electrically heated (to 900 °C) quartz tube.All measurements were repeated at least 4–5 times. Results and Discussion Optimization of HGAAS Measurement For the HGAAS determination of arsenic in tobacco samples, the optimum concentration of acid in the reaction vessel was in the range 1.0–1.5 m and the concentration of sodium tetrahydroborate was 1.2%. Arsenic was determined as arsenic(iii) and l-cysteine was used as a reducing agent to convert arsenic(v) to arsenic(iii).16–20 It was found experimentally that the optimum l-cysteine concentration was in the range 1.7–3.2%; a 2.5% concentration of this reagent was finally chosen. As there was a tendency for the suspension to be blown towards the atomizer, especially for the organic matrix slurries considered in this study, the gas flow rate had to be kept to a minimum.A carrier gas flow rate of 0.8 l min21 was therefore chosen. The use of a larger volume of reaction solution (i.e., greater than 5.0–5.5 ml in our hydride generation system) is not advisable due to extensive foaming of the solution.No significant (i.e., greater than 3% of the analytical signal), suppressive interferences were observed. Transition metal (main source of such interferences) concentrations are normally at relatively low levels in tobacco samples and are therefore unlikely to cause analytical problems. It is well known that lcysteine can also reduce interferences in the determination of arsenic(iii).16–20 Sample Homogeneity and Particle Size Fig. 1 shows the influence of sample grinding time on the arsenic signal obtained by the slurry sampling HGAAS method. For the CTA-OTL-1 tobacco leaves sample, the plateau of the arsenic signal was obtained after 10 min of grinding. The grinding time must be longer (minimum 15 min) for the real samples of cigarette tobacco (see also Fig. 1). The relative standard deviation (RSD) of the determination of arsenic in sample B was 2.4% better (n = 5) for a slurry made from a portion of sample after 15 min grinding than for a slurry made from a portion of the same sample after 10 min grinding only.This can be explained by the fact that certified reference materials are distributed after some pre-treatment and homogenization. The small difference in the extent of grinding required for two cigarette tobacco samples may be due to the variation in particle size within each sample and the difference in arsenic concentration.Sample A, having a higher arsenic concentration, may require slightly more grinding and hence, finer particles to ensure maximum extraction. The other factors that may influence the required grinding time are the amount and brand of the tobacco sample. The use of mechanical grinding (e.g., a special mixer mill or vibrating mill) of tobacco leaves can further reduce the grinding time and hence the time for the whole analytical procedure (see Fig. 1 Influence of grinding time on arsenic signal: -, ( Å 150 mg) CTAOTL- 1 sample; +, ( Å 50 mg) tobacco A sample. 540 Analyst, June 1997, Vol. 122also ref. 21). The time needed to reach a particle size below 60 mm can be (as already suggested in some independent experiments) reduced by half in comparison with manual grinding. Partitioning of Arsenic in Slurries The use of nitric acid as a liquid medium leads to the extraction of the analyte into the liquid phase of the slurry.The degree of arsenic extraction into the liquid phase was evaluated by an examination of the slurries prepared with various concentrations of acid. The percentage recovery accomplished with each acid concentration was determined by comparing the arsenic concentration in the supernatant with the total concentration in the slurry. A range of acid concentrations from 4.5 to 7.5 m was investigated and the results showed that the amount of arsenic extracted into the liquid phase depends on both the nitric acid concentration and the nature of the sample.The optimum acid concentration required for the extraction of approximately 90% of the arsenic from all tobacco samples was about 6.5 m (460 ml of 65% acid +540 ml of water to prepare 1 l). Ultrasonic and Microwave Treatment The only initial pre-treatment of the slurries was ultrasonication using an ultrasonic bath. Fig. 2 shows that the arsenic signal increases with an increase in the ultrasonication time.This is undoubtedly due to the increasing extraction of the analyte with further ultrasonication. The optimum signal was obtained with 15 min of ultrasonication at room temperature ( Å 20 °C), or with only 12 min of ultrasonication at about 75–76 °C (see Fig. 2). The amount of extracted arsenic was improved if a microwave-assisted heating step was introduced as the first step of sample pre-treatment. A relatively short time of only 40 s (at the full power of the microwave oven) was applied because a longer treatment was not possible owing to the small volume of slurry, and the limited heat resistance of the plastic bottles used and to avoid strong boiling of the acid. Interestingly, it was observed that the inclusion of even the short initial microwave treatment improves the amount of arsenic extracted into the liquid phase.This results in the additional extraction of 3–4% of arsenic for the samples studied, as shown in Table 1.Optimization of Slurry Concentrations To study the range of slurry concentrations useful for the determination of arsenic, several slurries with different concentrations were prepared and analysed. The relationship between slurry concentration and analytical signal for the certified reference material CTA-OTL-1 and sample A is displayed in Fig. 3. It can be seen that a linear relationship (the best analytical range of slurry concentration) was obtained for both materials for slurry concentrations from 0.5 to 3% m/v.Quantification and Comparison with Reference Analysis The certified reference material CTA-OTL-1, tobacco leaf sample A and tobacco leaf sample B were analysed using the optimized slurry preparation procedure and HGAAS quantification based on either a prepared calibration graph or standard additions calibration. The correlation coefficients were 0.9994 and 0.9991 for the calibration graph and standard additions graph, respectively.The linear range of the calibration graph was 0–40 ng ml21 of arsenic. Results for arsenic concentrations obtained using the two calibration techniques were then compared with the certified value for the reference material and the values obtained for samples A and B using a reference analytical technique based on analysis using Zeeman-effect background-corrected ETAAS after wet-acid decomposition. The results are presented in Table 2. In all cases, the mean results for 12 analyses were close to the mean reference value ( > 92%).Results of the standard additions method were slightly lower than those calculated from the calibration graph ( > 94%). A simple calibration technique based on aqueous standard solutions can therefore be recommended. Arsenic concentrations in the range 0.5–1.0 mg kg21 (similar concentrations for all samples studied) were obtained. Generally, our results were in good agreement with the reference Fig. 2 Influence of ultrasonication time on arsenic signal (for CTA-OTL- 1 sample): +, ultrasonication at room temperature; -, ultrasonication at 75–76 °C.Table 1 Effect of microwave pre-treatment on the percentage of arsenic extracted into the supernatant Arsenic extracted into the liquid phase (%)* Without With microwave microwave pre-treatment pre-treatment Tobacco A 89 93 Tobacco B 91 94 CTA-OTL-1 90 93 * Mean value of three extractions. Fig. 3 Relationship between analytical signal of arsenic and certified reference material CTA-OTL-1 (-), and tobacco A (+) slurry concentration.Analyst, June 1997, Vol. 122 541value for the CTA-OTL-1 tobacco. The absolute detection limit for the determination of arsenic(iii) by slurry sampling HGAAS, calculated using the IUPAC recommendation (based on the 3s criterion), was 2.6 ng. The RSD of the proposed analytical procedure was lower than 7.6% for arsenic concentrations at levels between 0.5 and 0.9 mg kg21 (for four individual measurements of three independently prepared slurries).Conclusion The proposed slurry sampling HGAAS method provides a rapid and relatively simple approach for the determination of arsenic in tobacco samples without the need for sample decomposition and total dissolution. The two critical factors in obtaining reliable results seem to be particle size and sample homogeneity, but the slurry concentration range must also be taken into account. The combination of hot ultrasonication and microwave treatment of samples was useful in improving the efficiency of the described method.The method can be used for the rapid and sensitive control of arsenic levels in tobacco samples. Although the proposed method is probably applicable to diverse plant samples, it is recommended that the performance of the procedure should be optimized prior to the analysis of samples with a considerably different matrix composition (e.g., plant leaves and roots).The authors thank Bert Aarts for technical assistance. J. M. thanks the University of Western Sydney, Nepean, for the award of a visiting research fellowship. References 1 Arsenic in the Environment. Part I: Cycling and Characterization, ed. Nriagu, J. O., Wiley, New York, 1994. 2 Registry of Toxic Effects of Chemical Substances, ed. Tatken, R. L., and Lewis, R. J., US Department of Health and Human Services, Cincinnati, OH, 1983. 3 Kabata-Pendias, A., and Pendias, H., Trace Elements in Soils and Plants, CRC Press, Boca Raton, FL, 1984. 4 Ahmed, S., Chaudhry, M. S., and Qureshi, I. H., J. Radioanal. Chem., 1979, 54, 331. 5 Dupire, S., and Hoenig, M., Analusis, 1980, 8, 153. 6 Hoenig, M., and Vanhoegweghen, P., Spectrochim. Acta, Part B, 1982, 37, 817. 7 Griffin, H. R., Hocking, M. B., and Lowery, D. G., Anal. Chem., 1975, 47, 229. 8 Vijan, P. N., Rayner, A. C., Sturgis, D., and Wood, G. R., Anal. Chim. Acta, 1976, 82, 329. 9 Smith, R. G., Van Loon, J.C., Knechtel, J. R., Fraser, J. L., Pitts, A. E., and Hodges, A. E., Anal. Chim. Acta, 1977, 93, 61. 10 Kuldvere, A., At. Spectrosc., 1980, 1, 138. 11 Arafat, N. M., and Glooschenko, W. A., Analyst, 1981, 106, 1174. 12 Haswell, S. J., Mendham, J., Butler, M. J., and Smith, D. C., J. Anal. At. Spectrom., 1988, 3, 731. 13 Nerin, C., Zufiaurre, R., and Cacho, J., Mikrochim. Acta, 1992, 108, 241. 14 Lopez Garcia, I., Arroyo Cortez, J., and Hernandez Cordoba, M., At. Spectrosc., 1993, 14, 144. 15 Miller-Ihli, N. J., At. Spectrosc., 1992, 13, 1. 16 Boampong, C., Brindle, I. D., Le, X., Pidwerbesky, L., and Ceccerelli Ponzoni, C. M., Anal. Chem., 1988, 60, 1185. 17 Brindle, I. D., and Le, X., Anal. Chem., 1989, 1, 1175. 18 Chen, H., Brindle, I. D., and Le, X., Anal. Chem., 1992, 64, 687. 19 Welz, B., and Sucmanov�a, M., Analyst., 1993, 118, 1417. 20 Welz, B., and Sucmanov�a, M., Analyst., 1993, 118, 1425. 21 Dobrowolski, R., and Mierzwa, J., Fresenius’ J.Anal. Chem., 1992, 344, 340. Paper 6/08246G Received December 6, 1996 Accepted February 19, 1997 Table 2 Results of arsenic determination in tobacco samples Arsenic concentration/mg kg21 (dry mass) Slurry* Calibration Standard Reference Sample graph additions value Tobacco A 0.89(0) ± 0.064 0.87(5) ± 0.062 0.900 ± 0.052† Tobacco B 0.55(6) ± 0.035 0.54(0) ± 0.037 0.582 ± 0.033† CTA-OTL-1 0.51(0) ± 0.038 0.50(1) ± 0.036 0.539 ± 0.060‡ * Mean value ± standard deviation (n = 12).The uncertainty in the last figure is emphasized by using parentheses. † Values obtained after wetacid decomposition of the sample and Zeeman-effect background-corrected ETAAS determination of arsenic. ‡ Certified value. 542 Analyst, June 1997, Vol. 122 Slurry Sampling for Hydride Generation Atomic Absorption Spectrometric Determination of Arsenic in Cigarette Tobaccos Jerzy Mierzwa†a, Samuel B. Adelojua and Harkirat S. Dhindsab a Centre for Electrochemical Research and Analytical Technology, and Department of Chemistry, University of Western Sydney, Nepean, P.O. Box 10, Kingswood, New South Wales 2747, Australia b Department of Chemistry, University of New England, Armidale, New South Wales 2351, Australia The development of a slurry sampling hydride generation atomic absorption spectrometric (HGAAS) method for the determination of arsenic in cigarette tobacco samples is described.The method is relatively simple and has been shown to give values of total arsenic close to those obtained using methods requiring total dissolution and decomposition of all vegetable matter before analysis.Pre-treatment of samples slurried in nitric acid by ultrasonication permitted the extraction of about 90% of the total arsenic from tobacco samples. Further improvement in the recovery efficiency (up to 93–94%) was accomplished by the use of an additional step of short microwave-accelerated treatment. l-Cysteine was used as a pre-reduction agent.The accuracy and precision of the slurry sampling HGAAS method were studied using the certified reference material (CRM) CTA-OTL-1 Oriental Tobacco Leaves. Under the optimum conditions, as little as 2.6 ng of arsenic can be detected. The relative standard deviation of the overall procedure was calculated to be below 7.6% at arsenic concentration levels of 0.5–0.9 mg kg21 and the analytical results obtained for the CRM agreed with the certified value.The main factors that influenced the reliability of the method were sample homogeneity, particle size and slurry concentration. Keywords: Arsenic; atomic absorption spectrometry; hydride generation; slurried sample; cigarette tobacco Arsenic is an element that has created considerable concern and interest in biological and environmental monitoring for many years (see, e.g., ref. 1). Comprehensive data on the toxicology of arsenic are available.2 In particular, the interest in determining arsenic in biological and environmental samples has accentuated the need for a rapid and sensitive analytical procedure for the control of the concentration of this element.Arsenic has been identified as a potential human carcinogen and smoking can be one of the pathways for the transfer of arsenic to humans; therefore, its control in tobacco samples is important. The possible sources of arsenic in plants (tobacco leaves) presumably include surface contamination by industrial activities, natural uptake of arsenic from soil, and even the use of arsenical pesticides in countries where these are still permitted.3 The transfer of elements from cigarette tobacco to smoke condensate has been studied, and it was concluded that the total content of arsenic was transferred to the smoke.4 Atomic absorption spectrometry, both with hydride generation (HGAAS) and electrothermal atomization (ETAAS), is widely used for the determination of arsenic at trace levels in plant samples. The use of either ETAAS5,6 or HGAAS7–11 for the determination of arsenic in these samples requires dissolution and decomposition of samples.There are, usually, timeconsuming procedures and a riskcontamination and/or loss of the element is associated with this mode of sample preparation. A slurry sampling technique combined with HGAAS has recently been proposed as an attractive alternative for avoiding the need for dissolution and mineralization steps.In the authors opinion, such an analytical technique is based on a very efficient extraction of the analyte into the liquid phase, but results reported suggested that analyte adsorbed onto solids can also take part in the hydride generation reaction. This approach has been successfully used for the determination of arsenic in light sandy soil,12 sewage sludge,12 fly ash,12–14 city waste incineration ash12 and diatomaceous earth.14 To the best of our knowledge, no work has been reported on the determination of arsenic in plant samples by slurry sampling combined with hydride generation.Some different potential problems might be expected with plant samples, e.g., tobacco leaf samples. Arsenic is likely to be encapsulated within cell walls, and it is unclear whether these would be broken down by the combined effects of dilute acid attack and ultrasonication. It is also not entirely certain if organic forms of arsenic would be broken down to inorganic arsenic unless a total oxidative acid digestion is carried out.This paper describes a rapid method for the determination of total arsenic in tobacco samples by a relatively simple pretreatment and further sampling of slurries combined with HGAAS. The reliable utilization of this method was considered by careful optimization of the main experimental parameters, e.g., sample grinding time, slurry ultrasonication time, and slurry and reagent concentrations. Experimental Apparatus All measurements were performed on a GBC Model 902 atomic absorption spectrometer (GBC, Dandenong, Victoria, Australia) operated in the single beam mode.A deuterium lamp background correction system was used. An arsenic hollow cathode lamp (Photron, Narre Warren, Victoria, Australia) operated at 10 mA was used as the primary radiation source. The 193.7 nm spectral line and a spectral band-width of 0.7 nm were selected. Slurry mixing was accomplished with an ultrasonic bath (Branson, Danbury, CT, USA) and an MT-19 vortex mixer (Chiltern Scientific, Sydney, NSW, Australia).A 650 W domestic microwave oven (NEC, Tokyo, Japan) was used for the additional pre-treatment of slurries. Slurries were manually injected into a laboratory-made batch hydride generation system connected to a conventional (glass) gas–liquid separator † On leave from: Central Laboratory, University of M. Curie-Sklodowska, P1 20-031 Lublin, Poland. Present address: Department of Nuclear Science, National Tsing-Hua University, 30043 Hsinchu, Taiwan.Analyst, June 1997, Vol. 122 (539–542) 539and electrically heated quartz T-tube. All measurements were based on peak area absorbance and a 30 s integration time. All glassware and plastic containers were soaked in 2 m nitric acid for 24 h and rinsed three times with Milli-Q water before use. Reagents and Samples Nitric acid (65%) of spectral purity (Merck, Darmstadt, Germany), Triton X-100 (Merck) and Milli-Q water were used.Working standards were obtained from stock solutions of arsenic(iii) by serial dilution with Milli-Q water and acidified with nitric acid. Sodium tetrahydroborate solution was prepared by dissolving the reagent in 1.0% sodium hydroxide and filtering through a 0.45 mm filter. This solution was stored for up to 1 week in a refrigerator. l-Cysteine (Merck) solution was prepared by dissolving the reagent in Milli-Q water. All reagents used were of analytical-reagent grade or better.A mixture of 99% argon +1% oxygen was used as the carrier gas. Two different brands of cigarette tobacco were analysed: the first was collected from cigarettes commercially manufactured in Poland (sample A) and the second from cigarettes manufactured in Australia (sample B). The certified reference material CTA-OTL-1 Oriental Tobacco Leaves (IChTJ, Warsaw, Poland) was used for the verification of the method. Analytical Procedure For each brand of cigarettes the tobacco content of a cigarette was mixed well and about 500–600 mg of tobacco were collected.Initially, all samples were dried for 12 h at a temperature of 85 °C. To ensure homogeneity, the samples were ground in an agate mortar for approximately 20 min. The effectiveness of grinding was controlled by observations under an optical microscope. After grinding, the average diameter for most of the particles was below 60 mm. Slurries in the range 0.5–2.5% m/v were then prepared.The ground samples were weighed accurately (50–250 mg of sample) into poly- (propylene) screw-cap bottles (volume, Å 15 ml) to which 10 ml of 6.5 m nitric acid with 0.005% Triton X-100 were added. A precise (±0.01 mg) electronic balance was used for all weighings. A volume/volume factor (volume of solid/volume of diluent) as proposed by Miller-Ihli15 was taken into account during slurry preparation. The slurries were treated for 40 s in a microwave oven (at full power mode, i.e., 650 W). [Warning: The caps of the poly(propylene) bottles were not screwed tight prior to the microwave treatment.] Slurries were then ultrasonicated for 12 min in an ultrasonic bath at 75–76 °C (maximum obtainable temperature for this bath). Powdered lcysteine was added prior to the slurry ultrasonication.Blanks were prepared in a similar fashion, but omitting samples. The slurries were vortex mixed for 5–10 s prior to each measurement. An aliquot (0.5–1 ml) of the slurry was then taken and placed in the reaction vessel together with 4 ml of water. A 0.5 ml volume of 1.2% NaBH4 solution was then added by manual pipetting. The carrier gas (a mixture of 99% Ar and 1% O2) was bubbled through the reaction vessel so that the arsenic hydride produced was swept to the electrically heated (to 900 °C) quartz tube.All measurements were repeated at least 4–5 times. Results and Discussion Optimization of HGAAS Measurement For the HGAAS determination of arsenic in tobacco samples, the optimum concentration of acid in the reaction vessel was in the range 1.0–1.5 m and the concentration of sodium tetrahydroborate was 1.2%.Arsenic was determined as arsenic(iii) and l-cysteine was used as a reducing agent to convert arsenic(v) to arsenic(iii).16–20 It was found experimentally that the optimum l-cysteine concentration was in the range 1.7–3.2%; a 2.5% concentration of this reagent was finally chosen.As there was a tendency for the suspension to be blown towards the atomizer, especially for the organic matrix slurries considered in this study, the gas flow rate had to be kept to a minimum. A carrier gas flow rate of 0.8 l min21 was therefore chosen. The use of a larger volume of reaction solution (i.e., greater than 5.0–5.5 ml in our hydride generation system) is not advisable due to extensive foaming of the solution. No significant (i.e., greater than 3% of the analytical signal), suppressive interferences were observed.Transition metal (main source of such interferences) concentrations are normally at relatively low levels in tobacco samples and are therefore unlikely to cause analytical problems. It is well known that lcysteine can also reduce interferences in the determination of arsenic(iii).16–20 Sample Homogeneity and Particle Size Fig. 1 shows the influence of sample grinding time on the arsenic signal obtained by the slurry sampling HGAAS method.For the CTA-OTL-1 tobacco leaves sample, the plateau of the arsenic signal was obtained after 10 min of grinding. The grinding time must be longer (minimum 15 min) for the real samples of cigarette tobacco (see also Fig. 1). The relative standard deviation (RSD) of the determination of arsenic in sample B was 2.4% better (n = 5) for a slurry made from a portion of sample after 15 min grinding than for a slurry made from a portion of the same sample after 10 min grinding only.This can be explained by the fact that certified reference materials are distributed after some pre-treatment and homogenization. The small difference in the extent of grinding required for two cigarette tobacco samples may be due to the variation in particle size within each sample and the difference in arsenic concentration. Sample A, having a higher arsenic concentration, may require slightly more grinding and hence, finer particles to ensure maximum extraction.The other factors that may influence the required grinding time are the amount and brand of the tobacco sample. The use of mechanical grinding (e.g., a special mixer mill or vibrating mill) of tobacco leaves can further reduce the grinding time and hence the time for the whole analytical procedure (see Fig. 1 Influence of grinding time on arsenic signal: -, ( Å 150 mg) CTAOTL- 1 sample; +, ( Å 50 mg) tobacco A sample. 540 Analyst, June 1997, Vol. 122also ref. 21).The time needed to reach a particle size below 60 mm can be (as already suggested in some independent experiments) reduced by half in comparison with manual grinding. Partitioning of Arsenic in Slurries The use of nitric acid as a liquid medium leads to the extraction of the analyte into the liquid phase of the slurry. The degree of arsenic extraction into the liquid phase was evaluated by an examination of the slurries prepared with various concentrations of acid. The percentage recovery accomplished with each acid concentration was determined by comparing the arsenic concentration in the supernatant with the total concentration in the slurry.A range of acid concentrations from 4.5 to 7.5 m was investigated and the results showed that the amount of arsenic extracted into the liquid phase depends on both the nitric acid concentration and the nature of the sample. The optimum acid concentration required for the extraction of approximately 90% of the arsenic from all tobacco samples was about 6.5 m (460 ml of 65% acid +540 ml of water to prepare 1 l).Ultrasonic and Microwave Treatment The only initial pre-treatment of the slurries was ultrasonication using an ultrasonic bath. Fig. 2 shows that the arsenic signal increases with an increase in the ultrasonication time. This is undoubtedly due to the increasing extraction of the analyte with further ultrasonication. The optimum signal was obtained with 15 min of ultrasonication at room temperature ( Å 20 °C), or with only 12 min of ultrasonication at about 75–76 °C (see Fig. 2). The amount of extracted arsenic was improved if a microwave-assisted heating step was introduced as the first step of sample pre-treatment. A relatively short time of only 40 s (at the full power of the microwave oven) was applied because a longer treatment was not possible owing to the small volume of slurry, and the limited heat resistance of the plastic bottles used and to avoid strong boiling of the acid.Interestingly, it was observed that the inclusion of even the short initial microwave treatment improves the amount of arsenic extracted into the liquid phase. This results in the additional extraction of 3–4% of arsenic for the samples studied, as shown in Table 1. Optimization of Slurry Concentrations To study the range of slurry concentrations useful for the determination of arsenic, several slurries with different concentrations were prepared and analysed.The relationship between slurry concentration and analytical signal for the certified reference material CTA-OTL-1 and sample A is displayed in Fig. 3. It can be seen that a linear relationship (the best analytical range of slurry concentration) was obtained for both materials for slurry concentrations from 0.5 to 3% m/v. Quantification and Comparison with Reference Analysis The certified reference material CTA-OTL-1, tobacco leaf sample A and tobacco leaf sample B were analysed using the optimized slurry preparation procedure and HGAAS quantification based on either a prepared calibration graph or standard additions calibration. The correlation coefficients were 0.9994 and 0.9991 for the calibration graph and standard additions graph, respectively.The linear range of the calibration graph was 0–40 ng ml21 of arsenic. Results for arsenic concentrations obtained using the two calibration techniques were then compared with the certified value for the reference material and the values obtained for samples A and B using a reference analytical technique based on analysis using Zeeman-effect background-corrected ETAAS after wet-acid decomposition.The results are presented in Table 2. In all cases, the mean results for 12 analyses were close to the mean reference value ( > 92%). Results of the standard additions method were slightly lower than those calculated from the calibration graph ( > 94%).A simple calibration technique based on aqueous standard solutions can therefore be recommended. Arsenic concentrations in the range 0.5–1.0 mg kg21 (similar concentrations for all samples studied) were obtained. Generally, our results were in good agreement with the reference Fig. 2 Influence of ultrasonication time on arsenic signal (for CTA-OTL- 1 sample): +, ultrasonication at room temperature; -, ultrasonication at 75–76 °C. Table 1 Effect of microwave pre-treatment on the percentage of arsenic extracted into the supernatant Arsenic extracted into the liquid phase (%)* Without With microwave microwave pre-treatment pre-treatment Tobacco A 89 93 Tobacco B 91 94 CTA-OTL-1 90 93 * Mean value of three extractions.Fig. 3 Relationship between analytical signal of arsenic and certified reference material CTA-OTL-1 (-), and tobacco A (+) slurry concentration. Analyst, June 1997, Vol. 122 541value for the CTA-OTL-1 tobacco. The absolute detection limit for the determination of arsenic(iii) by slurry sampling HGAAS, calculated using the IUPAC recommendation (based on the 3s criterion), was 2.6 ng.The RSD of the proposed analytical procedure was lower than 7.6% for arsenic concentrations at levels between 0.5 and 0.9 mg kg21 (for four individual measurements of three independently prepared slurries). Conclusion The proposed slurry sampling HGAAS method provides a rapid and relatively simple approach for the determination of arsenic in tobacco samples without the need for sample decomposition and total dissolution.The two critical factors in obtaining reliable results seem to be particle size and sample homogeneity, but the slurry concentration range must also be taken into account. The combination of hot ultrasonication and microwave treatment of samples was useful in improving the efficiency of the described method. The method can be used for the rapid and sensitive control of arsenic levels in tobacco samples.Although the proposed method is probably applicable to diverse plant samples, it is recommended that the performance of the procedure should be optimized prior to the analysis of samples with a considerably different matrix composition (e.g., plant leaves and roots). The authors thank Bert Aarts for technical assistance. J. M. thanks the University of Western Sydney, Nepean, for the award of a visiting research fellowship. References 1 Arsenic in the Environment. Part I: Cycling and Characterization, ed.Nriagu, J. O., Wiley, New York, 1994. 2 Registry of Toxic Effects of Chemical Substances, ed. Tatken, R. L., and Lewis, R. J., US Department of Health and Human Services, Cincinnati, OH, 1983. 3 Kabata-Pendias, A., and Pendias, H., Trace Elements in Soils and Plants, CRC Press, Boca Raton, FL, 1984. 4 Ahmed, S., Chaudhry, M. S., and Qureshi, I. H., J. Radioanal. Chem., 1979, 54, 331. 5 Dupire, S., and Hoenig, M., Analusis, 1980, 8, 153. 6 Hoenig, M., and Vanhoegweghen, P., Spectrochim. Acta, Part B, 1982, 37, 817. 7 Griffin, H. R., Hocking, M. B., and Lowery, D. G., Anal. Chem., 1975, 47, 229. 8 Vijan, P. N., Rayner, A. C., Sturgis, D., and Wood, G. R., Anal. Chim. Acta, 1976, 82, 329. 9 Smith, R. G., Van Loon, J. C., Knechtel, J. R., Fraser, J. L., Pitts, A. E., and Hodges, A. E., Anal. Chim. Acta, 1977, 93, 61. 10 Kuldvere, A., At. Spectrosc., 1980, 1, 138. 11 Arafat, N. M., and Glooschenko, W. A., Analyst, 1981, 106, 1174. 12 Haswell, S. J., Mendham, J., Butler, M. J., and Smith, D. C., J. Anal. At. Spectrom., 1988, 3, 731. 13 Nerin, C., Zufiaurre, R., and Cacho, J., Mikrochim. Acta, 1992, 108, 241. 14 Lopez Garcia, I., Arroyo Cortez, J., and Hernandez Cordoba, M., At. Spectrosc., 1993, 14, 144. 15 Miller-Ihli, N. J., At. Spectrosc., 1992, 13, 1. 16 Boampong, C., Brindle, I. D., Le, X., Pidwerbesky, L., and Ceccerelli Ponzoni, C. M., Anal. Chem., 1988, 60, 1185. 17 Brindle, I. D., and Le, X., Anal. Chem., 1989, 1, 1175. 18 Chen, H., Brindle, I. D., and Le, X., Anal. Chem., 1992, 64, 687. 19 Welz, B., and Sucmanov�a, M., Analyst., 1993, 118, 1417. 20 Welz, B., and Sucmanov�a, M., Analyst., 1993, 118, 1425. 21 Dobrowolski, R., and Mierzwa, J., Fresenius’ J. Anal. Chem., 1992, 344, 340. Paper 6/08246G Received December 6, 1996 Accepted February 19, 1997 Table 2 Results of arsenic determination in tobacco samples Arsenic concentration/mg kg21 (dry mass) Slurry* Calibration Standard Reference Sample graph additions value Tobacco A 0.89(0) ± 0.064 0.87(5) ± 0.062 0.900 ± 0.052† Tobacco B 0.55(6) ± 0.035 0.54(0) ± 0.037 0.582 ± 0.033† CTA-OTL-1 0.51(0) ± 0.038 0.50(1) ± 0.036 0.539 ± 0.060‡ * Mean value ± standard deviation (n = 12). The uncertainty in the last figure is emphasized by using parentheses. † Values obtained after wetacid decomposition of the sample and Zeeman-effect background-corrected ETAAS determination of arsenic. ‡ Certified value. 542 Analyst, June 1997, Vol.

ISSN:0003-2654    

DOI:10.1039/a608246g

出版商:RSC    

年代:1997

数据来源: RSC

摘要:

Determination of Trace Amounts of Rare Earth Elements in High-purity Cerium Oxide by Inductively Coupled Plasma Mass Spectrometry After Separation by Solvent Extraction Bing Li,* Yan Zhang and Ming Yin Institute of Rock and Mineral Analysis, Chinese Academy of Geological Sciences, 26 Baiwanzhuang Road, 100037, Beijing, China A method was developed for the determination of trace amounts of REEs in high-purity cerium oxide by ICP-MS after separation by solvent extraction. Based on the specificity of 2-ethylhexyl hydrogen-2-ethylhexylphosphonate, the Ce matrix in a tetravalent ionic state was separated from all other trivalent ionic state REEs.After removal of the matrix, the polyatomic interferences in ICP-MS were negligible. More than 99.5% of the Ce matrix was removed with 94–102% recoveries for 14 REE impurities in a spiked sample. The precision was 0.7–2.7% (RSD). The detection limits for 14 rare earth elements were 0.026–0.003 ng cm23 in solution and the limits of quantification were 0.02–0.09 mg g21 in solid cerium oxide.The method is rapid, precise and reliable and is well suited for the determination of REE impurities in high-purity cerium oxide. Keywords: Inductively coupled plasma mass spectrometry; high-purity cerium oxide; solvent extraction; rare earth impurity; 2-ethylhexyl hydrogen-2-ethylhexylphosphonate Rare earth compounds have attracted considerable practical interest and in some high-technology fields, their purity is of great importance. As the demand for high purity of REEs is increasing, the development of reliable analytical methods is essential.It is well known that the determination of ultra-trace impurities in high-purity cerium oxide is difficult. Optical emission spectrometry is generally hampered by complex optical spectra and serious interference effects.1–3 Further, the sensitivity is also insufficient for ultra-trace levels of REEs. Of many available analytical techniques, ICP-MS provides very attractive features including extremely low detection limits, simple spectra and freedom from most interferences, high sample throughput and a wide dynamic range.ICP-MS has been widely applied in various application areas for trace element determinations.4 However, only a few papers on high-purity REE applications have appeared.5–8 Yin et al.7 reported the direct determination of 14 trace REEs in high-purity of Y2O3, Sc2O3, Lu2O3, Ho2O3 and Tm2O3, but it is not easy to apply the method to other REE matrices because of severe polyatomic interferences.9 In order to eliminate the polyatomic interferences, Shitaba et al.5 demonstrated the use of electrothermal vaporization (ETV) ICP-MS for the determination of REE impurities in pure Gd2O3.Because the water in the aqueous sample was removed by ETV, the severity of certain polyatomic interferences was greatly reduced. Kawabata et al.6 combined ion chromatography and ICP-MS for the determination of REE impurities in pure Gd2O3 and La2O3.Panday et al.8 reported a method for the trace determination of REEs and other impurities in high-purity scandium by ICP-MS after liquid–liquid extraction of the matrix by means of bis(2- ethylhexyl)orthophosphoric acid. The major potential analytical problem encountered in the analysis of high-purity CeO2 is interferences due to polyatomic ions, e.g., 140CeH+ interferes with monoisotopic 141Pr and CeO+ and CeOH+ overlap the isotopes of Gd and Tb.Because these interferences are relatively large compared with the analyte contribution, the determination could become very difficult or even impossible. In order to obtain reliable results, matrix separation prior to the determination is an essential step. It is well known that 2-ethylhexyl hydrogen-2-ethylhexylphosphonate (EHEHP) is a very effective extraction reagent for REE and has been widely used in the REE industry. To our knowledge, methods for the determination of all REE impurities in high-purity CeO2 by EHEHP solvent extraction followed by ICP-MS have not been published.However, some methods for extraction chromatographic separation using EHEHP resin and ICP-AES for some pure rare earth oxides have been developed. 10 However, most of these off-line ion extraction chromatographic methods involve large elution volumes, which can lead to high blanks and long sample preparation times. In addition, large amounts of sample are generally needed in ICPAES, which is often an important consideration in the analysis of high-purity rare earth oxides.The aim of this work was to develop a rapid and effective separation method based on EHEHP solvent extraction for trace REEs in a high-purity matrix of CeO2 prior to determination by ICP-MS. Experimental Instrumentation The ICP-MS instrument used was a Plasma-Quad (VG Elemental, Winsford, UK). The instrumental parameters are given in Table 1.Table 1 PlasmaQuad operative conditions Inductively coupled plasma— Forward power/kW 1.3 Reflected power/W < 5 Coolant (Ar) flow rate/dm3 min21 12 Auxiliary (Ar) flow rate/dm3 min21 0.7 Carrier gas (Ar) flow rate/dm3 min21 0.78 Sample uptake rate/cm3 min21 1.5 Sampling depth/mm 10 Data acquisition— Measurement mode Multichannel scanning Mass range 88–180 Internal standard 133Cs Channels 2048 Sweeps 120 Dwell time 500 ms Calibration strategy External calibration with internal standardization Analyst, June 1997, Vol. 122 (543–547) 543Reagents, Standard Solutions and Equipment EHEHP and cyclohexane (analytical-reagent grade) were purchased from Hongxing Chemical Plant (Beijing, China). Working solutions were prepared by mixing EHEHP with cyclohexane to give about a 0.05 mol dm23 concentration of EHEHP and purifying the solution by shaking with 10% v/v HNO3. All acids and other reagents used were of ultrapure grade. De-ionized water was further purified by quartz subboiling distillation.The REE calibration standard solution and internal standard solution of Cs (100 ng cm23) were prepared by serial dilution of 1000 mg cm23 stock solutions with 2% v/v nitric acid. Stock standard solutions were prepared using highpurity chemicals. Acetic acid–acetate buffer solution (pH 4), monochloroacetic acid solution (pH 3) and KMnO4 solution (5 mg cm23) were prepared with analytical-reagent grade chemicals. All beakers, separating funnels and other glassware were cleaned carefully before use.First, glassware was soaked in liquid detergent and rinsed thoroughly with de-ionized water to remove all detergent, then boiled for 2 h in HNO3 (1 + 1) and washed with de-ionized water and then sub-boiling distilled water. Finally they were dried in a dustless oven and stored in a plastic bag before use. All of the sample preparations were performed in a clean room. Sample Preparation A 0.2 g amount of high-purity cerium oxide (supplied by Beijing General Research Institute for Non-ferrous Metals, Beijing, China) was digested using an HNO3–H2O2 open acid digestion procedure.The final solution was 2 mg cm23 CeO2 in 1% v/v nitric acid. Separation Procedure A 20 cm3 volume of 0.05 mol dm23EHEHP was placed in a 60 cm3 cleaned separating funnel, followed by 20 cm3 of water (pH 4) and 2 cm3 of KMnO4 solution and shaken for 1 min. The aqueous layer was discarded and the organic phase was subsequently used.An aliquot of the sample digest (5 cm3) was adjusted to pH Å 4 with 2% m/v NaOH solution, then 2 cm3 of KMnO4 solution and 8 cm3 of buffer solution (pH 4) were added. The sample solution was transferred into separating funnel and shaken for 3 min, then the aqueous layer was discarded after the layers had been allowed to separate for 5 min. The organic phase should be deep yellow owing to the colour of Ce4+. The organic layer was back-extracted twice with one 20 cm3 and one 5 cm3 portion of 10% v/v HNO3 for 10 min.The aqueous phase was collected in a beaker and evaporated to approximately 2 cm3. The solution was transferred into a volumetric flask and a 1 cm3 of Cs internal standard solution (1 mg cm23) was added, then diluted to 10 cm3 with water. This solution was then ready for analysis. Blanks were prepared in exactly the same way as in the sample procedure. Results and Discussion Theoretical Considerations and Concentration of EHEHP EHEHP is a very efficient and selective agent for REEs.It is an H-form phosphonate organic solvent. Because the hydroxide radical in the molecule is not esterified, the H+ in the hydroxide radical can readily be replaced by metal ions at slight acidity (pH 3–6) according to the following simplified equation: [Ln2+]a + 3[HP]o = [Ln(P)3]o + 3[H+]a (1) where [Ln2+]a is the concentration of the metal ion in the aqueous phase and [HP]o is the concentration of EHEHP in the organic phase.[Ln(P3)]o and [H+] are the concentrations of the reaction products. It is clear that as the pH is increased, the equilibrium will shift to the right, resulting in a high efficiency of extraction. Hence the H+ produced during the extraction process must be neutralized to ensure optimum efficiency of extraction. However, for high concentrations of REEs the precipitation probably occurs at high pH. Another important characteristic of EHEHP is selectivity as the tetravalent cerium extracted into the organic phase is difficult to back-extract with an appropriate acid, while other trivalent REEs are readily back-extracted.Based on this consideration, the separation of all REE impurities from the cerium matrix can be achieved. According to the reaction molar ratio, 20 cm3 of organic solvent containing 0.05 mol dm23 of EHEHP in cycloexane were used in all experiments. This is more than the real reaction molar ratio when 10 mg of CeO2 sample were extracted.Choice of Oxidant As described above, the cerium in solution must be oxidized to the tetravalent form prior to separation. KBrO3 and KMnO4 were tested as an oxidants for this purpose. It was observed that KMnO4 is more effective and stable with respect to the efficiency of oxidation than KBrO3. Therefore, KMnO4 was chosen and the amount added was fixed at 100 mg, which is a large excess over the concentration of Ce. Effect of pH and Buffer Solution on Extraction In a preliminary test with pH in the range 3–6, pH 4 was found to be the most suitable for all REE extractions.The recoveries of all REEs in the presence and absence of the Ce matrix at pH 4 were studied first. The results are given in Table 2. As can be seen, the recoveries of all REEs without the Ce matrix are acceptable, but the recoveries of some lighter REEs with the Ce matrix are very low. To establish the cause, the concentrations of REEs in aqueous phase were examined.As expected, La, Pr, Nd and Sm were not completely extracted into the organic phase in the presence of the Ce matrix. This result is consistent with the process description based on theoretical considerations. The acidity of the aqueous phase was increased markedly owing to the large amounts of H+ on EHEHP replaced by Ce4+ during extraction, resulting in a decreased extraction efficiency. In order to confirm further the optimum pH of the aqueous phase and maintain the pH range effectively, acetate buffer solution (pH 4) and monochloroacetic acid buffer solution (pH Table 2 Recoveries of REEs (100 ng added) with and without a Ce matrix Recovery in discarded Recovery (%) aqueous phase (%) Element Without Ce With Ce Without Ce With Ce La 90.5 30.4 9.4 63.3 Pr 94.8 72.1 2.5 27.3 Nd 94.5 81.1 < 0.1 11.7 Sm 97.5 92.1 < 0.1 7.4 Eu 93.7 95.4 < 0.1 < 0.1 Gd 108 97.1 < 0.1 < 0.1 Tb 94.7 104 < 0.1 < 0.1 Dy 93.0 94.2 < 0.1 < 0.1 Ho 96.4 103 < 0.1 < 0.1 Er 96.8 107 < 0.1 < 0.1 Tm 97.5 96.9 < 0.1 < 0.1 Yb 97.4 91.6 < 0.1 < 0.1 Lu 95.8 93.3 < 0.1 < 0.1 Y 104 106 < 0.1 < 0.1 544 Analyst, June 1997, Vol. 1223) were checked. The results are given in Table 3. Obviously the acetate buffer solution (pH 4) is more effective than the monochloroacetic acid buffer (pH 3). The amounts of buffer solution tested were between 2 and 12 cm3. The recoveries of La, Pr, Nd and Sm were more than 92% with amounts of buffer solution between 4 and 12 cm3, so 8 cm3 of acetate buffer solution was selected. Effect of Extraction Time The recovery of La was examined in an effort to test the effect of the extraction time on the extraction efficiency.Fig. 1 shows the results. There was no difference in extraction efficiency between 1 and 30 min. The extraction time was therefore fixed at 5 min. Concentration of Nitric Acid for Back-extraction Since lighter REEs are easily back-extracted with dilute HNO3, the concentration of HNO3 can only affect the recoveries of heavier REEs.On the other hand, all REEs could be backextracted quantitatively, provided that the recovery of Lu is acceptable. Fig. 2 shows the effect of the concentration of HNO3 on the back-extraction of Lu. It can be seen that the recovery increased with increase in concentration of HNO3 and reached a plateau at 2 mol dm23 HNO3. It should be noted that the concentration of Ce from the organic to the aqueous phase was also increased with increase in concentration of HNO3.To prevent the back-extraction of Ce at high acidity as far as possible, 3 mol dm23 of HNO3 was chosen for backextraction. Effect of Back-extraction Time Fig. 3 shows the effect of the back-extraction time on the recovery of Lu. The recovery reached an optimum level after shaking for more than 5 min and 10 min was adopted. Matrix Separation Efficiency To assess the matrix separation efficiency, the concentration of Ce remaining in the aqueous phase after back-extraction was determined.The results are given in Table 4. Repeated tests showed a high separation efficiency and good reproducibility. Our experimental results show that large interfering polyatomic ion peaks are almost invisible after more than 99.8% Ce has been separated. However, small polyatomic peaks, such as 140CeO+, 140CeOH+ and 142CeO+, at m/z 156–158 are also observed. In order to avoid such polyatomic interferences, 155Gd was selected for the determination of Gd.In addition, a very small polyatomic peak from 142CeOH+ could affect the measurement of monoisotopic 159Tb, and a correction is therefore necessary. This is most readily done by measuring 157Gd and then calculating the relative contribution of CeOH+ to the peak of m/z 159 using equation (2) Table 3 Effect of buffer solutions on recovery of 100 ng cm23 of REEs Recovery (%) Monochloroacetic acid buffer Acetate buffer Element solution (pH 3) solution (pH 4) La 66.0 92.0 Pr 80.2 94.3 Nd 86.1 99.4 Sm 97.2 94.1 Eu 89.2 93.3 Gd 90.5 105 Tb 88.8 95.0 Dy 90.0 96.9 Ho 89.9 92.0 Er 97.4 92.5 Tm 88.7 93.0 Yb 90.7 89.4 Lu 85.8 91.4 Y 101 92.3 Fig. 1 Effect of extraction time on the efficiency of extraction of La. Fig. 2 Effect of concentration of HNO3 on back-extraction of Lu. Fig. 3 Effect of back-extraction time on the back-extraction of Lu. Table 4 Matrix separation efficiency of five replicate separations Original Remaining Separation No.Ce/mg Ce/mg ratio (%) 1 10.00 0.017 99.83 2 10.00 0.021 99.79 3 10.00 0.018 99.82 4 10.00 0.018 99.82 5 10.00 0.019 99.81 Analyst, June 1997, Vol. 122 545 159Tb =159Mintegral - 142 140 CE � (157Mintegral - 157 155 Gd � 155Mintegral ) =159Mintegral - 0.125 � (157Mintegral -1.06 �155Mintegral ) (2) Accuracy, Precision and Limits of Detection Owing to the lack of certified cerium standard materials, the validity of the method was demonstrated by the separation and analysis of a spiked sample.The accuracy and precision of replicate measurements are given in Table 5. The limits of detection (LOD), calculated as three times the standard deviation of 11 blank measurements, ranged from 0.026 to 0.03 ng cm23 for liquid solution and the limits of quantitation (LOQ) ranged from 0.02 to 0.09 mg g21 for solid cerium oxide, as listed in Table 6. The data in Tables 5 and 6 suggests that the accuracy and precision are satisfactory and the quantification limits allow the determination of 14 REE impurities in high-purity CeO2.Concentrations of REE Impurities in High-purity CeO2 Sample Table 7 summarizes the results obtained on a CeO2 sample of claimed 99.9999% purity using the proposed procedure. Conclusions The combination of solvent extraction by EHEHP with highperformance ICP-MS provides an effective, rapide and reliable technique for the determination of 14 REEs in highpurity cerium oxide.More than 99.5% of the Ce matrix was removed with 95–102% recoveries for 14 REE impurities in a spiked sample. After matrix separation, the polyatomic interferences in ICP-MS were almost eliminated. The sample separation and measurement could be carried out within 30 min for each sample. The use of a buffer solution is an important aspect of the technique as the separation efficiency mainly depends on the acidity of the sample solution.The method is especially useful when small amounts or expensive samples are to be processed. The high sensitivity of ICP-MS and the ultratrace level impurities require that particular attention should be paid to the use of purified reagents and water and a clean laboratory atmosphere. This work was supported by the Chinese National Natural Science Foundation (No. 29475190). The authors are deeply Table 5 Recoveries for a spiked high-purity CeO2 sample Added/ng 0 10 100 Element Determined/ng Determined/ng Recovery (%) Determined/ng Recovery (%) RSD (%)* La 0.40 9.73 93.3 95.6 95.2 0.98 Pr 0.27 10.3 100 101.5 101 1.0 Nd 0.42 9.32 89 98.7 98.3 1.0 Sm < 0.04 10.6 106 99.9 99.9 1.8 Eu 0.11 10.4 103 98.1 98.0 1.2 Gd 0.36 9.92 95.6 97.1 96.7 2.8 Tb 0.09 10.6 105 99.7 99.6 1.6 Dy 0.36 10.2 98.4 101 101 0.73 Ho < 0.02 9.62 96.2 102 102 1.1 Er < 0.09 9.57 95.7 98.5 98.5 1.2 Tm 0.06 9.36 93.0 97.3 97.2 1.7 Yb 0.13 9.07 89.4 94.3 94.2 2.3 Lu 0.15 9.29 91.4 97.6 97.5 2.3 Y 1.17 10.4 92.3 99.1 97.9 1.7 * Mean values calculated from n = 5 (single analysis of five sample preparations).Table 6 Limits of detection for REEs Element m/z LOD*/ng cm23 LOQ†/mg g21 La 139 0.007 0.02 Pr 141 0.019 0.05 Nd 146 0.003 0.06 Sm 147 0.013 0.04 Eu 153 0.01 0.03 Gd 155 0.02 0.07 Tb 159 0.013 0.04 Dy 163 0.024 0.08 Ho 165 0.007 0.02 Er 166 0.026 0.09 Tm 169 0.01 0.03 Yb 174 0.025 0.08 Lu 175 0.01 0.03 Y 89 0.009 0.03 * LOD = Limit of detection, 3s in solution.† LOQ = limit of quantification, 1000 3 dilution, 10s. Table 7 Analytical data (mg g21) for a high-purity CeO2 sample Mean ± SD Mean ± SD Element (n = 5) RSD (%) Element (n = 5) RSD (%) Y 0.18 ± 0.02 11 Tb < 0.04 — (0.14)* La 0.38 ± 0.02 18 Dy < 0.08 — Pr 0.11 ± 0.02 18 Ho < 0.02 — Nd 0.15 ± 0.02 13 Er < 0.09 — Sm < 0.04 — Tm < 0.03 — Eu < 0.03 — Yb < 0.08 — Gd < 0.07 — Lu < 0.03 — * Not corrected. 546 Analyst, June 1997, Vol. 122indebted to Professor Pengyuan Yang for reading the manuscript. References 1 Li, B., Yin, M., Zhang, Z.-G., Wang, X.-R., Yang, P.-Y., Zhuang, Z.- X., and Huang, B.-L., Fenxi Ceshi Yiqi Tongxun, 1996, 6, 63. 2 Yuan, P., Qi, W.-D., and Cheng, X.-H., Guangpuxue Yu Guangpu Fenxi, 1992, 12, 75. 3 Wang, S.-Y., Liu, J., and Cheng, X.-H., Fenxi Huaxue, 1992, 20, 1273. 4 Jarvis, K. E., Gray, A. L., and Houk, R. S., Handbook of ICP-MS, Blackie, Glasgow, 1992. 5 Shitaba, N., Fudagawa, N., and Kubota, M., Anal.Chem., 1991, 63, 636. 6 Kawabata, K., Kishi, Y., Kawaguch, O., Watanabe, Y., and Inoue, Y., Anal. Chem., 1991, 63, 2137. 7 Yin, M., Li, B., and Fu, T.-F., Fenxi Kexue Xuebao, 1995, 11, 13. 8 Panday, V. K., Becker, J. S., and Dietze, J.-J., Fresenius. J. Anal. Chem., 1995, 352, 327. 9 Yin, M., and Li, B., Yankuang Ceshi, 1994, 13, 81. 10 Center of Rare Earth Analysis of China National Nuclear Corporation, Analytical Methods for the Determination of the Trace Elements in High Purity Y2O3, Eu2O3, Sc2O3, La2O3, Nd2O3, Dy2O3 and Tb4O7 (Compilation), Atomic Energy Press, Beijing, 1993.Paper 7/00634I Received January 28, 1997 Accepted March 11, 1997 Analyst, June 1997, Vol. 122 547 Determination of Trace Amounts of Rare Earth Elements in High-purity Cerium Oxide by Inductively Coupled Plasma Mass Spectrometry After Separation by Solvent Extraction Bing Li,* Yan Zhang and Ming Yin Institute of Rock and Mineral Analysis, Chinese Academy of Geological Sciences, 26 Baiwanzhuang Road, 100037, Beijing, China A method was developed for the determination of trace amounts of REEs in high-purity cerium oxide by ICP-MS after separation by solvent extraction.Based on the specificity of 2-ethylhexyl hydrogen-2-ethylhexylphosphonate, the Ce matrix in a tetravalent ionic state was separated from all other trivalent ionic state REEs. After removal of the matrix, the polyatomic interferences in ICP-MS were negligible.More than 99.5% of the Ce matrix was removed with 94–102% recoveries for 14 REE impurities in a spiked sample. The precision was 0.7–2.7% (RSD). The detection limits for 14 rare earth elements were 0.026–0.003 ng cm23 in solution and the limits of quantification were 0.02–0.09 mg g21 in solid cerium oxide. The method is rapid, precise and reliable and is well suited for the determination of REE impurities in high-purity cerium oxide. Keywords: Inductively coupled plasma mass spectrometry; high-purity cerium oxide; solvent extraction; rare earth impurity; 2-ethylhexyl hydrogen-2-ethylhexylphosphonate Rare earth compounds have attracted considerable practical interest and in some high-technology fields, their purity is of great importance.As the demand for high purity of REEs is increasing, the development of reliable analytical methods is essential. It is well known that the determination of ultra-trace impurities in high-purity cerium oxide is difficult.Optical emission spectrometry is generally hampered by complex optical spectra and serious interference effects.1–3 Further, the sensitivity is also insufficient for ultra-trace levels of REEs. Of many available analytical techniques, ICP-MS provides very attractive features including extremely low detection limits, simple spectra and freedom from most interferences, high sample throughput and a wide dynamic range. ICP-MS has been widely applied in various application areas for trace element determinations.4 However, only a few papers on high-purity REE applications have appeared.5–8 Yin et al.7 reported the direct determination of 14 trace REEs in high-purity of Y2O3, Sc2O3, Lu2O3, Ho2O3 and Tm2O3, but it is not easy to apply the method to other REE matrices because of severe polyatomic interferences.9 In order to eliminate the polyatomic interferences, Shitaba et al.5 demonstrated the use of electrothermal vaporization (ETV) ICP-MS for the determination of REE impurities in pure Gd2O3. Because the water in the aqueous sample was removed by ETV, the severity of certain polyatomic interferences was greatly reduced.Kawabata et al.6 combined ion chromatography and ICP-MS for the determination of REE impurities in pure Gd2O3 and La2O3. Panday et al.8 reported a method for the trace determination of REEs and other impurities in high-purity scandium by ICP-MS after liquid–liquid extraction of the matrix by means of bis(2- ethylhexyl)orthophosphoric acid. The major potential analytical problem encountered in the analysis of high-purity CeO2 is interferences due to polyatomic ions, e.g., 140CeH+ interferes with monoisotopic 141Pr and CeO+ and CeOH+ overlap the isotopes of Gd and Tb. Because these interferences are relatively large compared with the analyte contribution, the determination could become very difficult or even impossible. In order to obtain reliable results, matrix separation prior to the determination is an essential step.It is well known that 2-ethylhexyl hydrogen-2-ethylhexylphosphonate (EHEHP) is a very effective extraction reagent for REE and has been widely used in the REE industry. To our knowledge, methods for the determination of all REE impurities in high-purity CeO2 by EHEHP solvent extraction followed by ICP-MS have not been published. However, some methods for extraction chromatographic separation using EHEHP resin and ICP-AES for some pure rare earth oxides have been developed. 10 However, most of these off-line ion extraction chromatographic methods involve large elution volumes, which can lead to high blanks and long sample preparation times. In addition, large amounts of sample are generally needed in ICPAES, which is often an important consideration in the analysis of high-purity rare earth oxides. The aim of this work was to develop a rapid and effective separation method based on EHEHP solvent extraction for trace REEs in a high-purity matrix of CeO2 prior to determination by ICP-MS.Experimental Instrumentation The ICP-MS instrument used was a Plasma-Quad (VG Elemental, Winsford, UK). The instrumental parameters are given in Table 1. Table 1 PlasmaQuad operative conditions Inductively coupled plasma— Forward power/kW 1.3 Reflected power/W < 5 Coolant (Ar) flow rate/dm3 min21 12 Auxiliary (Ar) flow rate/dm3 min21 0.7 Carrier gas (Ar) flow rate/dm3 min21 0.78 Sample uptake rate/cm3 min21 1.5 Sampling depth/mm 10 Data acquisition— Measurement mode Multichannel scanning Mass range 88–180 Internal standard 133Cs Channels 2048 Sweeps 120 Dwell time 500 ms Calibration strategy External calibration with internal standardization Analyst, June 1997, Vol. 122 (543–547) 543Reagents, Standard Solutions and Equipment EHEHP and cyclohexane (analytical-reagent grade) were purchased from Hongxing Chemical Plant (Beijing, China). Working solutions were prepared by mixing EHEHP with cyclohexane to give about a 0.05 mol dm23 concentration of EHEHP and purifying the solution by shaking with 10% v/v HNO3.All acids and other reagents used were of ultrapure grade. De-ionized water was further purified by quartz subboiling distillation. The REE calibration standard solution and internal standard solution of Cs (100 ng cm23) were prepared by serial dilution of 1000 mg cm23 stock solutions with 2% v/v nitric acid. Stock standard solutions were prepared using highpurity chemicals.Acetic acid–acetate buffer solution (pH 4), monochloroacetic acid solution (pH 3) and KMnO4 solution (5 mg cm23) were prepared with analytical-reagent grade chemicals. All beakers, separating funnels and other glassware were cleaned carefully before use. First, glassware was soaked in liquid detergent and rinsed thoroughly with de-ionized water to remove all detergent, then boiled for 2 h in HNO3 (1 + 1) and washed with de-ionized water and then sub-boiling distilled water.Finally they were dried in a dustless oven and stored in a plastic bag before use. All of the sample preparations were performed in a clean room. Sample Preparation A 0.2 g amount of high-purity cerium oxide (supplied by Beijing General Research Institute for Non-ferrous Metals, Beijing, China) was digested using an HNO3–H2O2 open acid digestion procedure. The final solution was 2 mg cm23 CeO2 in 1% v/v nitric acid. Separation Procedure A 20 cm3 volume of 0.05 mol dm23EHEHP was placed in a 60 cm3 cleaned separating funnel, followed by 20 cm3 of water (pH 4) and 2 cm3 of KMnO4 solution and shaken for 1 min.The aqueous layer was discarded and the organic phase was subsequently used. An aliquot of the sample digest (5 cm3) was adjusted to pH Å 4 with 2% m/v NaOH solution, then 2 cm3 of KMnO4 solution and 8 cm3 of buffer solution (pH 4) were added. The sample solution was transferred into separating funnel and shaken for 3 min, then the aqueous layer was discarded after the layers had been allowed to separate for 5 min.The organic phase should be deep yellow owing to the colour of Ce4+. The organic layer was back-extracted twice with one 20 cm3 and one 5 cm3 portion of 10% v/v HNO3 for 10 min. The aqueous phase was collected in a beaker and evaporated to approximately 2 cm3. The solution was transferred into a volumetric flask and a 1 cm3 of Cs internal standard solution (1 mg cm23) was added, then diluted to 10 cm3 with water.This solution was then ready for analysis. Blanks were prepared in exactly the same way as in the sample procedure. Results and Discussion Theoretical Considerations and Concentration of EHEHP EHEHP is a very efficient and selective agent for REEs. It is an H-form phosphonate organic solvent. Because the hydroxide radical in the molecule is not esterified, the H+ in the hydroxide radical can readily be replaced by metal ions at slight acidity (pH 3–6) according to the following simplified equation: [Ln2+]a + 3[HP]o = [Ln(P)3]o + 3[H+]a (1) where [Ln2+]a is the concentration of the metal ion in the aqueous phase and [HP]o is the concentration of EHEHP in the organic phase.[Ln(P3)]o and [H+] are the concentrations of the reaction products. It is clear that as the pH is increased, the equilibrium will shift to the right, resulting in a high efficiency of extraction. Hence the H+ produced during the extraction process must be neutralized to ensure optimum efficiency of extraction.However, for high concentrations of REEs the precipitation probably occurs at high pH. Another important characteristic of EHEHP is selectivity as the tetravalent cerium extracted into the organic phase is difficult to back-extract with an appropriate acid, while other trivalent REEs are readily back-extracted. Based on this consideration, the separation of all REE impurities from the cerium matrix can be achieved. According to the reaction molar ratio, 20 cm3 of organic solvent containing 0.05 mol dm23 of EHEHP in cycloexane were used in all experiments.This is more than the real reaction molar ratio when 10 mg of CeO2 sample were extracted. Choice of Oxidant As described above, the cerium in solution must be oxidized to the tetravalent form prior to separation. KBrO3 and KMnO4 were tested as an oxidants for this purpose. It was observed that KMnO4 is more effective and stable with respect to the efficiency of oxidation than KBrO3.Therefore, KMnO4 was chosen and the amount added was fixed at 100 mg, which is a large excess over the concentration of Ce. Effect of pH and Buffer Solution on Extraction In a preliminary test with pH in the range 3–6, pH 4 was found to be the most suitable for all REE extractions. The recoveries of all REEs in the presence and absence of the Ce matrix at pH 4 were studied first. The results are given in Table 2.As can be seen, the recoveries of all REEs without the Ce matrix are acceptable, but the recoveries of some lighter REEs with the Ce matrix are very low. To establish the cause, the concentrations of REEs in aqueous phase were examined. As expected, La, Pr, Nd and Sm were not completely extracted into the organic phase in the presence of the Ce matrix. This result is consistent with the process description based on theoretical considerations. The acidity of the aqueous phase was increased markedly owing to the large amounts of H+ on EHEHP replaced by Ce4+ during extraction, resulting in a decreased extraction efficiency.In order to confirm further the optimum pH of the aqueous phase and maintain the pH range effectively, acetate buffer solution (pH 4) and monochloroacetic acid buffer solution (pH Table 2 Recoveries of REEs (100 ng added) with and without a Ce matrix Recovery in discarded Recovery (%) aqueous phase (%) Element Without Ce With Ce Without Ce With Ce La 90.5 30.4 9.4 63.3 Pr 94.8 72.1 2.5 27.3 Nd 94.5 81.1 < 0.1 11.7 Sm 97.5 92.1 < 0.1 7.4 Eu 93.7 95.4 < 0.1 < 0.1 Gd 108 97.1 < 0.1 < 0.1 Tb 94.7 104 < 0.1 < 0.1 Dy 93.0 94.2 < 0.1 < 0.1 Ho 96.4 103 < 0.1 < 0.1 Er 96.8 107 < 0.1 < 0.1 Tm 97.5 96.9 < 0.1 < 0.1 Yb 97.4 91.6 < 0.1 < 0.1 Lu 95.8 93.3 < 0.1 < 0.1 Y 104 106 < 0.1 < 0.1 544 Analyst, June 1997, Vol. 1223) were checked.The results are given in Table 3. Obviously the acetate buffer solution (pH 4) is more effective than the monochloroacetic acid buffer (pH 3). The amounts of buffer solution tested were between 2 and 12 cm3. The recoveries of La, Pr, Nd and Sm were more than 92% with amounts of buffer solution between 4 and 12 cm3, so 8 cm3 of acetate buffer solution was selected. Effect of Extraction Time The recovery of La was examined in an effort to test the effect of the extraction time on the extraction efficiency. Fig. 1 shows the results. There was no difference in extraction efficiency between 1 and 30 min. The extraction time was therefore fixed at 5 min. Concentration of Nitric Acid for Back-extraction Since lighter REEs are easily back-extracted with dilute HNO3, the concentration of HNO3 can only affect the recoveries of heavier REEs. On the other hand, all REEs could be backextracted quantitatively, provided that the recovery of Lu is acceptable. Fig. 2 shows the effect of the concentration of HNO3 on the back-extraction of Lu.It can be seen that the recovery increased with increase in concentration of HNO3 and reached a plateau at 2 mol dm23 HNO3. It should be noted that the concentration of Ce from the organic to the aqueous phase was also increased with increase in concentration of HNO3. To prevent the back-extraction of Ce at high acidity as far as possible, 3 mol dm23 of HNO3 was chosen for backextraction. Effect of Back-extraction Time Fig. 3 shows the effect of the back-extraction time on the recovery of Lu. The recovery reached an optimum level after shaking for more than 5 min and 10 min was adopted. Matrix Separation Efficiency To assess the matrix separation efficiency, the concentration of Ce remaining in the aqueous phase after back-extraction was determined. The results are given in Table 4. Repeated tests showed a high separation efficiency and good reproducibility. Our experimental results show that large interfering polyatomic ion peaks are almost invisible after more than 99.8% Ce has been separated.However, small polyatomic peaks, such as 140CeO+, 140CeOH+ and 142CeO+, at m/z 156–158 are also observed. In order to avoid such polyatomic interferences, 155Gd was selected for the determination of Gd. In addition, a very small polyatomic peak from 142CeOH+ could affect the measurement of monoisotopic 159Tb, and a correction is therefore necessary. This is most readily done by measuring 157Gd and then calculating the relative contribution of CeOH+ to the peak of m/z 159 using equation (2) Table 3 Effect of buffer solutions on recovery of 100 ng cm23 of REEs Recovery (%) Monochloroacetic acid buffer Acetate buffer Element solution (pH 3) solution (pH 4) La 66.0 92.0 Pr 80.2 94.3 Nd 86.1 99.4 Sm 97.2 94.1 Eu 89.2 93.3 Gd 90.5 105 Tb 88.8 95.0 Dy 90.0 96.9 Ho 89.9 92.0 Er 97.4 92.5 Tm 88.7 93.0 Yb 90.7 89.4 Lu 85.8 91.4 Y 101 92.3 Fig. 1 Effect of extraction time on the efficiency of extraction of La.Fig. 2 Effect of concentration of HNO3 on back-extraction of Lu. Fig. 3 Effect of back-extraction time on the back-extraction of Lu. Table 4 Matrix separation efficiency of five replicate separations Original Remaining Separation No. Ce/mg Ce/mg ratio (%) 1 10.00 0.017 99.83 2 10.00 0.021 99.79 3 10.00 0.018 99.82 4 10.00 0.018 99.82 5 10.00 0.019 99.81 Analyst, June 1997, Vol. 122 545 159Tb =159Mintegral - 142 140 CE � (157Mintegral - 157 155 Gd � 155Mintegral ) =159Mintegral - 0.125 � (157Mintegral -1.06 �155Mintegral ) (2) Accuracy, Precision and Limits of Detection Owing to the lack of certified cerium standard materials, the validity of the method was demonstrated by the separation and analysis of a spiked sample. The accuracy and precision of replicate measurements are given in Table 5.The limits of detection (LOD), calculated as three times the standard deviation of 11 blank measurements, ranged from 0.026 to 0.03 ng cm23 for liquid solution and the limits of quantitation (LOQ) ranged from 0.02 to 0.09 mg g21 for solid cerium oxide, as listed in Table 6.The data in Tables 5 and 6 suggests that the accuracy and precision are satisfactory and the quantification limits allow the determination of 14 REE impurities in high-purity CeO2. Concentrations of REE Impurities in High-purity CeO2 Sample Table 7 summarizes the results obtained on a CeO2 sample of claimed 99.9999% purity using the proposed procedure. Conclusions The combination of solvent extraction by EHEHP with highperformance ICP-MS provides an effective, rapid, precise and reliable technique for the determination of 14 REEs in highpurity cerium oxide.More than 99.5% of the Ce matrix was removed with 95–102% recoveries for 14 REE impurities in a spiked sample. After matrix separation, the polyatomic interferences in ICP-MS were almost eliminated. The sample separation and measurement could be carried out within 30 min for each sample.The use of a buffer solution is an important aspect of the technique as the separation efficiency mainly depends on the acidity of the sample solution. The method is especially useful when small amounts or expensive samples are to be processed. The high sensitivity of ICP-MS and the ultratrace level impurities require that particular attention should be paid to the use of purified reagents and water and a clean laboratory atmosphere. This work was supported by the Chinese National Natural Science Foundation (No. 29475190). The authors are deeply Table 5 Recoveries for a spiked high-purity CeO2 sample Added/ng 0 10 100 Element Determined/ng Determined/ng Recovery (%) Determined/ng Recovery (%) RSD (%)* La 0.40 9.73 93.3 95.6 95.2 0.98 Pr 0.27 10.3 100 101.5 101 1.0 Nd 0.42 9.32 89 98.7 98.3 1.0 Sm < 0.04 10.6 106 99.9 99.9 1.8 Eu 0.11 10.4 103 98.1 98.0 1.2 Gd 0.36 9.92 95.6 97.1 96.7 2.8 Tb 0.09 10.6 105 99.7 99.6 1.6 Dy 0.36 10.2 98.4 101 101 0.73 Ho < 0.02 9.62 96.2 102 102 1.1 Er < 0.09 9.57 95.7 98.5 98.5 1.2 Tm 0.06 9.36 93.0 97.3 97.2 1.7 Yb 0.13 9.07 89.4 94.3 94.2 2.3 Lu 0.15 9.29 91.4 97.6 97.5 2.3 Y 1.17 10.4 92.3 99.1 97.9 1.7 * Mean values calculated from n = 5 (single analysis of five sample preparations). Table 6 Limits of detection for REEs Element m/z LOD*/ng cm23 LOQ†/mg g21 La 139 0.007 0.02 Pr 141 0.019 0.05 Nd 146 0.003 0.06 Sm 147 0.013 0.04 Eu 153 0.01 0.03 Gd 155 0.02 0.07 Tb 159 0.013 0.04 Dy 163 0.024 0.08 Ho 165 0.007 0.02 Er 166 0.026 0.09 Tm 169 0.01 0.03 Yb 174 0.025 0.08 Lu 175 0.01 0.03 Y 89 0.009 0.03 * LOD = Limit of detection, 3s in solution. † LOQ = limit of quantification, 1000 3 dilution, 10s. Table 7 Analytical data (mg g21) for a high-purity CeO2 sample Mean ± SD Mean ± SD Element (n = 5) RSD (%) Element (n = 5) RSD (%) Y 0.18 ± 0.02 11 Tb < 0.04 — (0.14)* La 0.38 ± 0.02 18 Dy < 0.08 — Pr 0.11 ± 0.02 18 Ho < 0.02 — Nd 0.15 ± 0.02 13 Er < 0.09 — Sm < 0.04 — Tm < 0.03 — Eu < 0.03 — Yb < 0.08 — Gd < 0.07 — Lu < 0.03 — * Not corrected. 546 Analyst, June 1997, Vol. 122indebted to Professor Pengyuan Yang for reading the manuscript. References 1 Li, B., Yin, M., Zhang, Z.-G., Wang, X.-R., Yang, P.-Y., Zhuang, Z.- X., and Huang, B.-L., Fenxi Ceshi Yiqi Tongxun, 1996, 6, 63. 2 Yuan, P., Qi, W.-D., and Cheng, X.-H., Guangpuxue Yu Guangpu Fenxi, 1992, 12, 75. 3 Wang, S.-Y., Liu, J., and Cheng, X.-H., Fenxi Huaxue, 1992, 20, 1273. 4 Jarvis, K. E., Gray, A. L., and Houk, R. S., Handbook of ICP-MS, Blackie, Glasgow, 1992. 5 Shitaba, N., Fudagawa, N., and Kubota, M., Anal. Chem., 1991, 63, 636. 6 Kawabata, K., Kishi, Y., Kawaguch, O., Watanabe, Y., and Inoue, Y., Anal. Chem., 1991, 63, 2137. 7 Yin, M., Li, B., and Fu, T.-F., Fenxi Kexue Xuebao, 1995, 11, 13. 8 Panday, V. K., Becker, J. S., and Dietze, J.-J., Fresenius. J. Anal. Chem., 1995, 352, 327. 9 Yin, M., and Li, B., Yankuang Ceshi, 1994, 13, 81. 10 Center of Rare Earth Analysis of China National Nuclear Corporation, Analytical Methods for the Determination of the Trace Elements in High Purity Y2O3, Eu2O3, Sc2O3, La2O3, Nd2O3, Dy2O3 and Tb4O7 (Compilation), Atomic Energy Press, Beijing, 1993. Paper 7/00634I Received January 28, 1997 Accepted March 11, 1997 Analyst, June 1997, Vol.

ISSN:0003-2654    

DOI:10.1039/a700634i

出版商:RSC    

年代:1997

数据来源: RSC

摘要:

Determination of End-points for Polymorph Conversions of Crystalline Organic Compounds Using On-line Near-infrared Spectroscopy Timothy Norris*, Paul K. Aldridge and S. Sonja Sekulic Pfizer Central Research Laboratories, Groton, CT 06340, USA Trovafloxacin mesylate exists in two anhydrous polymorphic forms, I and II. The fundamental band vibration spectra information does not readily differentiate between the two forms, although the crystal lattice of each form has a distinct X-ray powder diffraction pattern.Subtle spectral differences between the two polymorphs can be detected in the NIR spectral region and the interconversion of the two forms can be monitored in hot solvent crystal slurries using NIR spectroscopy and an on-line fibre optic probe. A generally applicable method has been developed to monitor the conversion of form I into form II as a crystal slurry. The free energy profile in principal component space has been modelled as the process proceeds to give a computer graphic of the change which readily shows when the conversion is complete.Keywords: Polymorphism; near-infrared monitoring; near-infrared spectroscopy; infrared spectroscopy; hydrates; end-point determination; on-line analysis; principal component analysis: hierarchical cluster analysis; X-ray powder diffraction Polymorphism in crystals of organic compounds is caused by different packing arrangements of the same molecule in the crystal lattices. This gives rise to different crystal shapes and forms for the same molecular structure.If the molecular interactions in the crystal lattice are fairly strong, arising from a phenomenon such as hydrogen bonding, differences in polymorphic forms can be differentiated with techniques such as FTIR, Raman spectroscopy or differential scanning calorimetry (DSC) as well as the usually definitive techniques of X-ray powder diffraction pattern and single crystal X-ray analysis. In more subtle examples of polymorphism in organic crystals, differences in the crystal lattice may only be detected definitively by single crystal X-ray analysis or X-ray powder diffraction patterns, because vibrational spectra show only minor peak differences and DSC data cannot detect readily the small energy differences between the forms. A recent comprehensive review1 discusses analysis of organic polymorphs in detail and the problem of the various definitions and classifications used in the field as well as the advantages and disadvantages of various techniques used to examine polymorphs.Trovafloxacin mesylate, a new potent antibiotic, provides an example of a subtle manifestation of polymorphism described above. The anhydrous crystal can exist in two polymorphic forms, I and II, characterized by unique X-ray powder diffraction patterns, the principal peaks of which are noted in Table 1. Crystal polymorphism of compounds in the European Pharmacopoeia has been reviewed,2 the frequency of its occurrence highlighting the extent with which this phenomenon has to be addressed for manufacture of medicines. In pharmaceutical operations it is often desirable to convert the bulk drug substance into a specific polymorph which may have specific desirable chemical or physical properties such as resistance to thermal degradation or absorption of moisture that may aid its formulation as a medicine.It is good practice to define fully the polymorphic properties of a drug substance to ensure control of the manufacturing process in terms of physical properties.Also important, but normally a much rarer occurrence, is the effect of polymorphism on bioavailability of the compound under investigation. In principle, the different polymorphic forms of a compound have different energies and this could affect their bioavailability. However, this is only likely to be significant when the energy differences of the different forms are large.In cases of subtle polymorphism, where there is little difference in the lattice energies and hence melting-points or DSC profiles, as with trovafloxacin mesylate (the polymorphs of trovafloxacin mesylate melt with decomposition in the range 253–256 °C), this would not be a concern. For a discussion of the significance of polymorphism on bioavailability the reader is referred to refs. 3 and 4. In order to monitor polymorphic conversions in batch reactors it would be useful to develop an on-line in-process technique.NIR spectroscopy has been used to distinguish polymorphic forms of a drug substance using pattern recognition techniques.5 It is shown here that NIR between 1100 and 2100 nm can be used to monitor the subtle polymorph conversion of trovafloxacin mesylate polymorph I into poly- Table 1 X-ray powder diffraction patterns for anhydrous trovafloxacin mesylate polymorphs I and II and the monohydrate (principal peaks) Peak 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Polymorph I: 2q (°) Cu 5.0 9.8 13.0 14.8 19.7 20.9 22.0 23.0 28.1 29.3 d space 17.9 9.0 6.8 6.0 4.5 4.2 4.0 3.9 3.2 3.0 Polymorph II: 2q (°) Cu 4.5 7.7 9.1 13.6 15.0 18.2 18.6 22.8 d space 19.5 11.5 9.7 6.5 5.9 4.9 4.8 3.9 Monohydrate: 2q (°) Cu 4.7 9.4 12.4 13.1 13.6 14.2 17.0 17.9 18.7 21.0 22.0 24.2 25.5 26.6 27.2 d space 18.7 9.4 7.1 6.7 6.5 6.3 5.2 5.0 4.7 4.2 4.0 3.7 3.5 3.3 3.3 Analyst, June 1997, Vol. 122 (549–552) 549morph II, using a fibre optic probe connected to a spectrometer.Study of this conversion provides the basis of an on-line monitoring technique that can be used generally for this type of phenomenon. Experimental NIR crystal slurry spectra were recorded on an NIRSystems 6500 spectrometer fitted with a variable pathlength fibre optic probe and an auto-gain adapter manufactured by Perstorp Analytical (Silver Spring, MD, USA). The gap between the probe lens and the mirror was set at maximum, i.e., 1.0 cm, so that the crystal slurry could be probed to its fullest extent using an indeterminate light path.This was found to be the convenient mode of operation for heterogeneous systems, such as crystal slurries. The polymorphic conversions were studied isothermally at 95 °C in butanol and in propan-2-ol at 80 °C at a concentration of Å 100 g l21. NIR dry crystal spectra were also collected on an NIRSystems 6500 spectrometer, fitted with a reflectance detector and auto-gain adaptor.The sample was presented to the beam in a 1.5 cm diameter circular glass cell. Commercial software used to generate hierarchical cluster analysis dendrograms and principal component analysis scores plots was either Pirouette, Multivariate Data Analysis for IBM PC Systems, version 1.2 (Infometrix, Seattle, WA, USA), or Matlab, High Performance Numeric Computation and Visualization Software version 4.2C1 (Math Works, Natick, MA, USA). Typical apparatus used is shown schematically in Fig. 1. Power X-ray data were collected using a Siemens (Iselin, NJ, USA) D5000 spectrometer. IR spectra were measured using a Nicolet (Madison, WI, USA) Magna 500 spectrometer with a DRIFTS autosampler accessory. Results and Discussion When trovafloxacin mesylate, 2, is prepared from hydrolysis of trovafloxacin ethyl ester, 1, with methanesulfonic acid in aqueous tetrahydrofuran and dried, it is initially isolated as polymorph I. However, when form I is heated in a number of solvents such as propan-2-ol or butanol, it is converted into polymorph II which is thermodynamically more stable (see Scheme 1).Polymorphs I and II can be hydrated to form the same monohydrate. The X-ray powder diffraction pattern of trovafloxacin mesylate monohydrate is distinct from those of polymorphs I and II (see Table 1). Some monohydrate is usually present when polymorph I is isolated from the above hydrolysis procedure. In fact polymorph I hydrates more readily than polymorph II.When monohydrate crystals are dehydrated under vacuum, polymorph I is formed. Polymorph I crystals used in these experiments contained some monohydrate crystals. This could be seen in the NIR spectrum of polymorph I as absorptions centred at 1905 and 1952 nm, which are characteristic of water in the monohydrate. These absorptions are absent in the NIR spectrum of pure polymorph I crystals which is otherwise similar to that of the monohydrate (see Table 2).Polymorph I crystals with water contents as low as 0.2% m/m, which mainly consist of pure polymorph I crystals, still undergo conversion to polymorph II. The conversion can still be monitored using NIR on-line even though the spectral absorption differences between polymorphs I and II are fairly subtle. Typically, however, transformations were studied on polymorph I crystals that contained water in the range 0.6–1.2% m/m. NIR spectral data collected from the three types of crystal involved in the trovafloxacin mesylate system are summarized in Table 2 showing four distinct groups of absorption peaks indicated by regions A, B, C and D, defined as 1300–1500, 1600–1800, 1900–2000 and 2100–2400 nm, respectively.Polymorphs I and II have distinct B and D regions and the monohydrate is easily distinguished by absorptions characteristic of water in region C. The monohydrate and polymorph I, in contrast, are very similar in regions B and D. This pair can reversibly lose and gain a molecule of water which results in the absence and presence of the absorptions at 1906 and 1952 nm in region C.When collecting spectral information from the polymorphic transformation on-line as a crystal slurry in a solvent such as butanol, the attenuation of optical fibres restricts the collection of data to the wavelength range 1100–2100 nm. Fig. 1 Apparatus used to monitor trovagloxacin mesylate polymorph interconversions. Table 2 NIR peak positions obtained for anhydrous trovafloxacin mesylate polymorphs I and II, and monohydrate crystals Principal absorption peaks/nm Form A B C D Polymorph I 1392 1662 1722 — — — — 2206 2266 2308 Polymorph II 1370 1664 1720 1758 — — 2176 2218 2260 2300 Monohydrate 1418 1660 1722 — 1906 1952 — 2208 2262 2304 Scheme 1 550 Analyst, June 1997, Vol. 122Thus the detection of change relies on the characteristic polymorph II absorption at 1758 nm in region B and the water absorptions in region C. Fortunately, this is not swamped by the butanol contribution to the matrix spectra. IR spectra of the three forms of trovafloxacin mesylate studied are very similar in the range 2000–500 cm21.Spectra were collected using the DRIFTS method which is generally regarded as one of the methods of choice for examination of polymorphs.6,7 Table 3 highlights the minor differences observed in the IR spectra. These are noted as regions E, F and G which correspond to 500–600, 1050–1100 and 1300–1400 cm21, respectively.As the differences are minor, and no easily distinguishable peaks are available for characterization, the IR region of the spectrum is not useful in distinguishing the polymorphs of this compound. This is in contrast with spectral information available in the near-infra red region where distinct regions of characterization are found. The situation found in the study of trovafloxacin mesylate is unusual, but illustrates the subtle nature of its polymorphism.It is more often observed that both NIR and IR spectra contain regions in which the polymorph under study has useful characteristic differences. Using a technique developed for reaction monitoring,8 it was possible to monitor the spectroscopic changes and mimic the free energy steady-state path of the crystal conversion of polymorph I into polymorph II until it reaches its steady-state minimum energy. This profile, which is analogous to a macroscopic reaction coordinate, is generated in principal component space, and can be used to define the end-point of the polymorph conversion which would otherwise be very difficult to track.The graphical mimic of the change can be generated as the spectra are collected using commercially available software such as Matlab as the crystal transformation proceeds. It is particularly useful because polymorph crystal slurries in organic solvents can remain in their metastable forms for indeterminate amounts of time before enough of the thermodynamically stable form is present to cause the entire crystal slurry mass to undergo complete conversion.The current practice for batch processes involving polymorph conversions of this type is to determine experimentally an extended thermal conversion period that will accommodate the change with a low probability of rejection at the analysis checkpoint on the dry crystal. Use of the NIR monitoring method allows a manufacturer to optimize operation time to an efficient minimum.NIR spectra between 1100 and 2100 nm were usually collected at 5 min intervals during the polymorph conversions. Typical spectral results are shown in Fig. 2. The NIR spectra of a crystal slurry of polymorph I in butanol at 95 °C change systematically over time until the steady state is reached and the spectra once again superimpose on each other.9 At this stage the crystals in the slurry are polymorph II.This was verified by X-ray powder diffraction patterns collected on solvent-free crystals obtained from the slurry. Typically, microscopy reveals that the initial polymorph I crystals are essentially rhombic in nature and after conversion to polymorph II appear as hexagonal prisms. The detection of the difference in the crystal slurry can be accounted for in part by the baseline shifts due to the change in shape of the crystals during the conversion, as well as from the inherent differences in polymorph I and II crystals noted in NIR regions A and B as defined in Table 2.Some of the change arises due to the presence of trovafloxacin mesylate monohydrate in the polymorph I crystals. However, it has been shown that the change from polymorph I into polymorph II can be monitored by this method even when the water content of the polymorph I crystals is < 0.2% m/m as measured by the Karl Fischer method. Principal component analysis can be performed on the spectral data during the change.Fig. 3 and 4 show that the metastable polymorph I crystal slurry and the thermodynamically stable polymorph II crystal slurry can be defined by scores plots in principal component space. The free energy path Table 3 Regions of difference in IR spectra (DRIFTS) obtained for anhydrous trovafloxacin mesylate polymorphs I and II, and monohydrate crystals Principal absorption peaks/cm21 Form E F G Polymorph I 525 536 — 556 1033 1044 — — — 1344 1359 1381 Polymorph II 523 535 544 564 1035 1051 1090 1306 1329 1344 1357 1381 Monohydrate 525 537 — 556 1030 1044 1088 — — 1344 1359 1380 Fig. 2 Typical crystal slurry spectra collected during conversion of trovafloxacin mesylate polymorph I into polymorph II in butanol at 95 °C.(61 spectra collected every 5 min; spectra 50–61 comprise the polymorph II slurry; spectra 1–38 comprise the polymorph I slurry; spectra 39–49 cover the period of polymorph transition.) Fig. 3 Scores plot of principal component 1 (PC 1) versus principal component 2 (PC 2) for typical trovafloxacin mesylate polymorph conversion. Analyst, June 1997, Vol. 122 551of the transition of polymorph I into polymorph II is mimicked clearly by plotting the first three principal components of the spectral data. These can be generated by commercial software packages to yield graphical images on-line shortly after the event has occurred. The specific data in Figs. 3 and 4 show that complete conversion had occurred after spectrum number 50 had been taken, after the slurry had been held at 95 °C in butanol for 250 min.This time can vary from run to run. The end-point was confirmed by an X-ray powder diffraction pattern on the isolated crystal and supported by direct microscopic examination, which showed that the crystals had changed from the rhombic to the hexagonal form. The change can also be modelled using hierarchical cluster analysis in commercial software packages, such as Pirouette (see under Experimental), which again gives a graphical representation of the metastable slurry, the conversion period, definition of the end-point of the conversion and continued stability without further change of the polymorph II slurry (see Fig. 5). Conclusions The methodology described here provides a simple easy-to-use technique for on-line monitoring and determination of when a polymorph conversion in a crystal slurry matrix is complete. This cannot be achieved easily by other methods; for example, use of X-ray powder diffraction analysis results in considerable delay before a batch can be checked to see whether it has been converted to the desired polymorph if this was used as a process monitoring tool.The proposed method allows routine production to proceed with confidence and decreases significantly the probability of allowing an unconverted batch to be processed through to bulk dry crystal before being detected by off-line characterization tests.It is particularly useful for systems such as trovafloxacin mesylate where the amount of time spent in the initial thermodynamically metastable state is variable on a batch-to-batch basis. Use of an NIR monitoring tool in the way described allows a manufacturer to optimize the polymorph conversion time since it can be monitored on a batch-to-batch basis. This contrasts with a more conventional protocol in which it is necessary to continue the conversion through an extended time which has been shown historically to result in conversion to the desired polymorph.The authors thank Philip J. Johnson for experimental assistance and Dr. Jon Bordner and Debra L. De Costa for X-ray powder analysis. References 1 Threlfall, T. L., Analyst, 1995, 120, 2435. 2 Borka, L., Pharm. Acta Helv., 1991, 66, 16. 3 Aguiar, A. J., Krc, J., Kinkel, A. W., and Samyn, J. C., Pharm. Sci., 1967, 56, 847. 4 Burger, A., in Topics in Pharmaceutical Sciences, ed.Bremer, D. D., and Speiser, P., Elsevier, Amsterdam, 1983, p. 347. 5 Aldridge, P. K., Evans, C. L., Ward, H. W., Colgan, S. T., Boyer, N., and Gemperline, P. J., Anal. Chem., 1996, 68, 997. 6 Roston, D. A., Walters, M. C., Rhinebarger, R. P., and Ferro, L. J., J. Pharm. Biomed. Anal., 1993, 11, 293. 7 Hartauer, K. J., Miller, E. S., and Guillory, J. K., Int. J. Pharm., 1992, 85, 163. 8 Norris, T., and Aldridge, P. K., Analyst, 1996, 121, 1003. 9 Hailey, P.A., personal communication. Paper 7/00782E Received February 3, 1997 Accepted March 5, 1997 Fig. 4 Scores plot of principal component 1 (PC 1) versus principal component 3 (PC 3) for typical trovafloxacin mesylate polymorph conversion. Fig. 5 Typical dendrogram obtained for trovafloxacin mesylate polymorph conversion. (Spectra numbers run consecutively from 1 to 53 from the start of the experiment; the order is then 56, 54, 58, 60, 57, 59, 55, 61.) 552 Analyst, June 1997, Vol. 122 Determination of End-points for Polymorph Conversions of Crystalline Organic Compounds Using On-line Near-infrared Spectroscopy Timothy Norris*, Paul K. Aldridge and S. Sonja Sekulic Pfizer Central Research Laboratories, Groton, CT 06340, USA Trovafloxacin mesylate exists in two anhydrous polymorphic forms, I and II. The fundamental band vibration spectra information does not readily differentiate between the two forms, although the crystal lattice of each form has a distinct X-ray powder diffraction pattern.Subtle spectral differences between the two polymorphs can be detected in the NIR spectral region and the interconversion of the two forms can be monitored in hot solvent crystal slurries using NIR spectroscopy and an on-line fibre optic probe. A generally applicable method has been developed to monitor the conversion of form I into form II as a crystal slurry. The free energy profile in principal component space has been modelled as the process proceeds to give a computer graphic of the change which readily shows when the conversion is complete.Keywords: Polymorphism; near-infrared monitoring; near-infrared spectroscopy; infrared spectroscopy; hydrates; end-point determination; on-line analysis; principal component analysis: hierarchical cluster analysis; X-ray powder diffraction Polymorphism in crystals of organic compounds is caused by different packing arrangements of the same molecule in the crystal lattices.This gives rise to different crystal shapes and forms for the same molecular structure. If the molecular interactions in the crystal lattice are fairly strong, arising from a phenomenon such as hydrogen bonding, differences in polymorphic forms can be differentiated with techniques such as FTIR, Raman spectroscopy or differential scanning calorimetry (DSC) as well as the usually definitive techniques of X-ray powder diffraction pattern and single crystal X-ray analysis.In more subtle examples of polymorphism in organic crystals, differences in the crystal lattice may only be detected definitively by single crystal X-ray analysis or X-ray powder diffraction patterns, because vibrational spectra show only minor peak differences and DSC data cannot detect readily the small energy differences between the forms. A recent comprehensive review1 discusses analysis of organic polymorphs in detail and the problem of the various definitions and classifications used in the field as well as the advantages and disadvantages of various techniques used to examine polymorphs.Trovafloxacin mesylate, a new potent antibiotic, provides an example of a subtle manifestation of polymorphism described above. The anhydrous crystal can exist in two polymorphic forms, I and II, characterized by unique X-ray powder diffraction patterns, the principal peaks of which are noted in Table 1. Crystal polymorphism of compounds in the European Pharmacopoeia has been reviewed,2 the frequency of its occurrence highlighting the extent with which this phenomenon has to be addressed for manufacture of medicines. In pharmaceutical operations it is often desirable to convert the bulk drug substance into a specific polymorph which may have specific desirable chemical or physical properties such as resistance to thermal degradation or absorption of moisture that may aid its formulation as a medicine.It is good practice to define fully the polymorphic properties of a drug substance to ensure control of the manufacturing process in terms of physical properties.Also important, but normally a much rarer occurrence, is the effect of polymorphism on bioavailability of the compound under investigation. In principle, the different polymorphic forms of a compound have different energies and this could affect their bioavailability. However, this is only likely to be significant when the energy differences of the different forms are large.In cases of subtle polymorphism, where there is little difference in the lattice energies and hence melting-points or DSC profiles, as with trovafloxacin mesylate (the polymorphs of trovafloxacin mesylate melt with decomposition in the range 253–256 °C), this would not be a concern. For a discussion of the significance of polymorphism on bioavailability the reader is referred to refs. 3 and 4. In order to monitor polymorphic conversions in batch reactors it would be useful to develop an on-line in-process technique.NIR spectroscopy has been used to distinguish polymorphic forms of a drug substance using pattern recognition techniques.5 It is shown here that NIR between 1100 and 2100 nm can be used to monitor the subtle polymorph conversion of trovafloxacin mesylate polymorph I into poly- Table 1 X-ray powder diffraction patterns for anhydrous trovafloxacin mesylate polymorphs I and II and the monohydrate (principal peaks) Peak 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Polymorph I: 2q (°) Cu 5.0 9.8 13.0 14.8 19.7 20.9 22.0 23.0 28.1 29.3 d space 17.9 9.0 6.8 6.0 4.5 4.2 4.0 3.9 3.2 3.0 Polymorph II: 2q (°) Cu 4.5 7.7 9.1 13.6 15.0 18.2 18.6 22.8 d space 19.5 11.5 9.7 6.5 5.9 4.9 4.8 3.9 Monohydrate: 2q (°) Cu 4.7 9.4 12.4 13.1 13.6 14.2 17.0 17.9 18.7 21.0 22.0 24.2 25.5 26.6 27.2 d space 18.7 9.4 7.1 6.7 6.5 6.3 5.2 5.0 4.7 4.2 4.0 3.7 3.5 3.3 3.3 Analyst, June 1997, Vol. 122 (549–552) 549morph II, using a fibre optic probe connected to a spectrometer.Study of this conversion provides the basis of an on-line monitoring technique that can be used generally for this type of phenomenon. Experimental NIR crystal slurry spectra were recorded on an NIRSystems 6500 spectrometer fitted with a variable pathlength fibre optic probe and an auto-gain adapter manufactured by Perstorp Analytical (Silver Spring, MD, USA). The gap between the probe lens and the mirror was set at maximum, i.e., 1.0 cm, so that the crystal slurry could be probed to its fullest extent using an indeterminate light path.This was found to be the convenient mode of operation for heterogeneous systems, such as crystal slurries. The polymorphic conversions were studied isothermally at 95 °C in butanol and in propan-2-ol at 80 °C at a concentration of Å 100 g l21. NIR dry crystal spectra were also collected on an NIRSystems 6500 spectrometer, fitted with a reflectance detector and auto-gain adaptor.The sample was presented to the beam in a 1.5 cm diameter circular glass cell. Commercial software used to generate hierarchical cluster analysis dendrograms and principal component analysis scores plots was either Pirouette, Multivariate Data Analysis for IBM PC Systems, version 1.2 (Infometrix, Seattle, WA, USA), or Matlab, High Performance Numeric Computation and Visualization Software version 4.2C1 (Math Works, Natick, MA, USA). Typical apparatus used is shown schematically in Fig. 1. Power X-ray data were collected using a Siemens (Iselin, NJ, USA) D5000 spectrometer. IR spectra were measured using a Nicolet (Madison, WI, USA) Magna 500 spectrometer with a DRIFTS autosampler accessory. Results and Discussion When trovafloxacin mesylate, 2, is prepared from hydrolysis of trovafloxacin ethyl ester, 1, with methanesulfonic acid in aqueous tetrahydrofuran and dried, it is initially isolated as polymorph I. However, when form I is heated in a number of solvents such as propan-2-ol or butanol, it is converted into polymorph II which is thermodynamically more stable (see Scheme 1).Polymorphs I and II can be hydrated to form the same monohydrate. The X-ray powder diffraction pattern of trovafloxacin mesylate monohydrate is distinct from those of polymorphs I and II (see Table 1). Some monohydrate is usually present when polymorph I is isolated from the above hydrolysis procedure. In fact polymorph I hydrates more readily than polymorph II.When monohydrate crystals are dehydrated under vacuum, polymorph I is formed. Polymorph I crystals used in these experiments contained some monohydrate crystals. This could be seen in the NIR spectrum of polymorph I as absorptions centred at 1905 and 1952 nm, which are characteristic of water in the monohydrate. These absorptions are absent in the NIR spectrum of pure polymorph I crystals which is otherwise similar to that of the monohydrate (see Table 2).Polymorph I crystals with water contents as low as 0.2% m/m, which mainly consist of pure polymorph I crystals, still undergo conversion to polymorph II. The conversion can still be monitored using NIR on-line even though the spectral absorption differences between polymorphs I and II are fairly subtle. Typically, however, transformations were studied on polymorph I crystals that contained water in the range 0.6–1.2% m/m. NIR spectral data collected from the three types of crystal involved in the trovafloxacin mesylate system are summarized in Table 2 showing four distinct groups of absorption peaks indicated by regions A, B, C and D, defined as 1300–1500, 1600–1800, 1900–2000 and 2100–2400 nm, respectively.Polymorphs I and II have distinct B and D regions and the monohydrate is easily distinguished by absorptions characteristic of water in region C. The monohydrate and polymorph I, in contrast, are very similar in regions B and D.This pair can reversibly lose and gain a molecule of water which results in the absence and presence of the absorptions at 1906 and 1952 nm in region C. When collecting spectral information from the polymorphic transformation on-line as a crystal slurry in a solvent such as butanol, the attenuation of optical fibres restricts the collection of data to the wavelength range 1100–2100 nm. Fig. 1 Apparatus used to monitor trovagloxacin mesylate polymorph interconversions.Table 2 NIR peak positions obtained for anhydrous trovafloxacin mesylate polymorphs I and II, and monohydrate crystals Principal absorption peaks/nm Form A B C D Polymorph I 1392 1662 1722 — — — — 2206 2266 2308 Polymorph II 1370 1664 1720 1758 — — 2176 2218 2260 2300 Monohydrate 1418 1660 1722 — 1906 1952 — 2208 2262 2304 Scheme 1 550 Analyst, June 1997, Vol. 122Thus the detection of change relies on the characteristic polymorph II absorption at 1758 nm in region B and the water absorptions in region C. Fortunately, this is not swamped by the butanol contribution to the matrix spectra. IR spectra of the three forms of trovafloxacin mesylate studied are very similar in the range 2000–500 cm21.Spectra were collected using the DRIFTS method which is generally regarded as one of the methods of choice for examination of polymorphs.6,7 Table 3 highlights the minor differences observed in the IR spectra. These are noted as regions E, F and G which correspond to 500–600, 1050–1100 and 1300–1400 cm21, respectively.As the differences are minor, and no easily distinguishable peaks are available for characterization, the IR region of the spectrum is not useful in distinguishing the polymorphs of this compound. This is in contrast with spectral information available in the near-infra red region where distinct regions of characterization are found. The situation found in the study of trovafloxacin mesylate is unusual, but illustrates the subtle nature of its polymorphism.It is more often observed that both NIR and IR spectra contain regions in which the polymorph under study has useful characteristic differences. Using a technique developed for reaction monitoring,8 it was possible to monitor the spectroscopic changes and mimic the free energy steady-state path of the crystal conversion of polymorph I into polymorph II until it reaches its steady-state minimum energy. This profile, which is analogous to a macroscopic reaction coordinate, is generated in principal component space, and can be used to define the end-point of the polymorph conversion which would otherwise be very difficult to track.The graphical mimic of the change can be generated as the spectra are collected using commercially available software such as Matlab as the crystal transformation proceeds. It is particularly useful because polymorph crystal slurries in organic solvents can remain in their metastable forms for indeterminate amounts of time before enough of the thermodynamically stable form is present to cause the entire crystal slurry mass to undergo complete conversion.The current practice for batch processes involving polymorph conversions of this type is to determine experimentally an extended thermal conversion period that will accommodate the change with a low probability of rejection at the analysis checkpoint on the dry crystal. Use of the NIR monitoring method allows a manufacturer to optimize operation time to an efficient minimum.NIR spectra between 1100 and 2100 nm were usually collected at 5 min intervals during the polymorph conversions. Typical spectral results are shown in Fig. 2. The NIR spectra of a crystal slurry of polymorph I in butanol at 95 °C change systematically over time until the steady state is reached and the spectra once again superimpose on each other.9 At this stage the crystals in the slurry are polymorph II.This was verified by X-ray powder diffraction patterns collected on solvent-free crystals obtained from the slurry. Typically, microscopy reveals that the initial polymorph I crystals are essentially rhombic in nature and after conversion to polymorph II appear as hexagonal prisms. The detection of the difference in the crystal slurry can be accounted for in part by the baseline shifts due to the change in shape of the crystals during the conversion, as well as from the inherent differences in polymorph I and II crystals noted in NIR regions A and B as defined in Table 2.Some of the change arises due to the presence of trovafloxacin mesylate monohydrate in the polymorph I crystals. However, it has been shown that the change from polymorph I into polymorph II can be monitored by this method even when the water content of the polymorph I crystals is < 0.2% m/m as measured by the Karl Fischer method. Principal component analysis can be performed on the spectral data during the change.Fig. 3 and 4 show that the metastable polymorph I crystal slurry and the thermodynamically stable polymorph II crystal slurry can be defined by scores plots in principal component space. The free energy path Table 3 Regions of difference in IR spectra (DRIFTS) obtained for anhydrous trovafloxacin mesylate polymorphs I and II, and monohydrate crystals Principal absorption peaks/cm21 Form E F G Polymorph I 525 536 — 556 1033 1044 — — — 1344 1359 1381 Polymorph II 523 535 544 564 1035 1051 1090 1306 1329 1344 1357 1381 Monohydrate 525 537 — 556 1030 1044 1088 — — 1344 1359 1380 Fig. 2 Typical crystal slurry spectra collected during conversion of trovafloxacin mesylate polymorph I into polymorph II in butanol at 95 °C.(61 spectra collected every 5 min; spectra 50–61 comprise the polymorph II slurry; spectra 1–38 comprise the polymorph I slurry; spectra 39–49 cover the period of polymorph transition.) Fig. 3 Scores plot of principal component 1 (PC 1) versus principal component 2 (PC 2) for typical trovafloxacin mesylate polymorph conversion. Analyst, June 1997, Vol. 122 551of the transition of polymorph I into polymorph II is mimicked clearly by plotting the first three principal components of the spectral data. These can be generated by commercial software packages to yield graphical images on-line shortly after the event has occurred. The specific data in Figs. 3 and 4 show that complete conversion had occurred after spectrum number 50 had been taken, after the slurry had been held at 95 °C in butanol for 250 min. This time can vary from run to run. The end-point was confirmed by an X-ray powder diffraction pattern on the isolated crystal and supported by direct microscopic examination, which showed that the crystals had changed from the rhombic to the hexagonal form. The change can also be modelled using hierarchical cluster analysis in commercial software packages, such as Pirouette (see under Experimental), which again gives a graphical representation of the metastable slurry, the conversion period, definition of the end-point of the conversion and continued stability without further change of the polymorph II slurry (see Fig. 5). Conclusions The methodology described here provides a simple easy-to-use technique for on-line monitoring and determination of when a polymorph conversion in a crystal slurry matrix is complete.This cannot be achieved easily by other methods; for example, use of X-ray powder diffraction analysis results in considerable delay before a batch can be checked to see whether it has been converted to the desired polymorph if this was used as a process monitoring tool. The proposed method allows routine production to proceed with confidence and decreases significantly the probability of allowing an unconverted batch to be processed through to bulk dry crystal before being detected by off-line characterization tests. It is particularly useful for systems such as trovafloxacin mesylate where the amount of time spent in the initial thermodynamically metastable state is variable on a batch-to-batch basis. Use of an NIR monitoring tool in the way described allows a manufacturer to optimize the polymorph conversion time since it can be monitored on a batch-to-batch basis. This contrasts with a more conventional protocol in which it is necessary to continue the conversion through an extended time which has been shown historically to result in conversion to the desired polymorph. The authors thank Philip J. Johnson for experimental assistance and Dr. Jon Bordner and Debra L. De Costa for X-ray powder analysis. References 1 Threlfall, T. L., Analyst, 1995, 120, 2435. 2 Borka, L., Pharm. Acta Helv., 1991, 66, 16. 3 Aguiar, A. J., Krc, J., Kinkel, A. W., and Samyn, J. C., Pharm. Sci., 1967, 56, 847. 4 Burger, A., in Topics in Pharmaceutical Sciences, ed. Bremer, D. D., and Speiser, P., Elsevier, Amsterdam, 1983, p. 347. 5 Aldridge, P. K., Evans, C. L., Ward, H. W., Colgan, S. T., Boyer, N., and Gemperline, P. J., Anal. Chem., 1996, 68, 997. 6 Roston, D. A., Walters, M. C., Rhinebarger, R. P., and Ferro, L. J., J. Pharm. Biomed. Anal., 1993, 11, 293. 7 Hartauer, K. J., Miller, E. S., and Guillory, J. K., Int. J. Pharm., 1992, 85, 163. 8 Norris, T., and Aldridge, P. K., Analyst, 1996, 121, 1003. 9 Hailey, P. A., personal communication. Paper 7/00782E Received February 3, 1997 Accepted March 5, 1997 Fig. 4 Scores plot of principal component 1 (PC 1) versus principal component 3 (PC 3) for typical trovafloxacin mesylate polymorph conversion. Fig. 5 Typical dendrogram obtained for trovafloxacin mesylate polymorph conversion. (Spectra numbers run consecutively from 1 to 53 from the start of the experiment; the order is then 56, 54, 58, 60, 57, 59, 55, 61.) 552 Analyst, June 1997, Vol. 122

ISSN:0003-2654    

DOI:10.1039/a700782e

出版商:RSC    

年代:1997

数据来源: RSC