摘要:

Critical Review Recent developments in food authentication† M. John Dennis CSL Food Science Laboratory, Norwich Research Park, Colney, Norwich, UK NR4 7UQ Received 17th April 1998, Accepted 30th June 1998 1 Introduction 2 Classification of authenticity issues 2.1 Species of origin 2.2 Geographical region of origin 2.3 Commercial treatment 2.4 Water 2.5 Brands 3 Recent developments in meat authentication 4 Recent developments in fish authentication 5 Recent developments in milk and cheese authentication 6 Recent developments in vegetable oil authentication 7 Recent developments in essential oil authentication 8 Recent developments in fruit products authentication 9 Effects of food fraud 10 References 1 Introduction Food authentication is the process by which a food is verified as complying with its label description.Labelling and compositional regulations, which may differ from country to country, have a fundamental place in determining which scientific tests are appropriate for a particular issue. As an example of this, the European Community does not permit the use of pulpwash (an aqueous extract of the albido) in orange juice whereas the use of “in-line pulpwash” (i.e., made as part of the juice process) is permitted in the USA.Thus different tests are appropriate to establish the authenticity of “pure orange juice” in the different countries. The question of whether in-line pulpwash can be differentiated chemically from off-line pulpwash is an issue which, as yet, has not been considered.Labelling legislation is there to ensure that food is properly described. It seeks to protect the consumer from being sold an inferior product with a false description in addition to protecting honest traders from unfair competition. Enforcement of this legislation ensures that correctly described products remain available to the consumer and that consumer confidence is maintained, which in turn ensures a market place for these foods.Thus the availability of sound analytical methods which can ensure the authenticity of foods plays a fundamental role in the operation of modern society. The desire to make a fraudulent profit from the misrepresentation of food has been a feature of society from historical times. One of the earliest scientific surveys of the authenticity of food was undertaken by Arthur Hill Hassall in 1861.1 He employed microscopy, which until this time had been reserved for medical studies, to investigate the authenticity of coffee, an extremely expensive commodity at that time.His survey found that 31 out of 34 samples contained adulterants such as chicory, roasted wheat and burnt sugar. In 1995 the UK Ministry of Agriculture Fisheries and Food (MAFF) undertook a survey of instant coffees and evaluated their authenticity on the basis of their sugar composition; 15% of the coffees examined were considered as not authentic.2 This demonstrates two important points.First, adulteration issues do not go away. If there is the potential for an illegal profit to be made, only continued vigilance provides reassurance against this type of fraud. Second, the development of new technologies will often discover food authenticity issues which current techniques can not. Hassall’s idea of using the microscope for food studies was a good example of this point which subsequent scientific developments in food authentication have regularly confirmed.This review will therefore record some of the recent advances in food authentication. 2 Classification of authenticity issues As the example of pulpwash demonstrates, there can be many different and indeed subtle issues concerning labelling which it may be desirable to check by performing chemical tests. However, it is possible to classify the issues into a number of similar topics. 2.1 Species of origin A common authenticity problem is for the species from which a food was made to be misdescribed. This may take the form of † © Crown copyright.M. John Dennis obtained his PhD from the University of East Anglia in 1976 and joined the Ministry of Agriculture Fisheries and Food, Food Science Laboratory to study the influence of nitrosothiols on nitrosamine formation. He subsequently investigated a number of issues in food safety including the use of the flour improvers potassium bromate and azodicarbonamide and the presence of the contaminants ethylcarbamate and polycyclic aromatic hydrocarbons.His current research interests lie in developing more rapid and more effective stable isotope procedures for assessing food authenticity and in developing the use of stable isotope analysis for human nutrition studies. He is head of Food Authenticity at CSL and is Chairman of the Methods Sub-Group for MAFF’s - Food Authenticity Working Party. Analyst, 1998, 123, 151R–156R 151Rsubstitution of one species for another. Thus in a survey of battered fried fish sold to the consumer,3 the species of fish was found to be misdescribed in 5% of cases.In this instance there was no evidence of a fraud being carried out for economic benefit since as often as not, an equally expensive fish was substituted. Of course, claims concerning the species of origin are effectively claims concerning the genetic make up of the organism and the definition of a species may make this a rather arbitrary classification.Some claims may go beyond the species barrier to the variety of the organism. Thus claims that beef has come from, for example, Aberdeen Angus may potentially require testing. Such claims confer a commercial advantage because some consumers consider that pure bred beef herds are less likely to have contracted bovine spongiform encephalopathy (BSE). To my knowledge, tests for variety have not yet been established; however, the application of recent developments in DNA technology (described further below) suggest that variety specific authenticity testing could be developed.The introduction of genetically modified organisms (GMOs) into food commerce may constitute a special case for variety specific authenticity testing if labelling makes claims about the presence or absence of these materials. Another authenticity issue which may commonly arise is the need to determine whether food products from one species have been mixed with similar material from a cheaper species.The question of whether durum wheat pasta contains common wheat represents a typical example. In a MAFF survey,4 only one sample out of 249 was found to exceed the limit of 8% common wheat in durum wheat pasta, which was the limit which the methods employed could be confident constituted misdescription. A further 14 samples showed evidence of containing between 3 and 8% common wheat by at least one method. The chosen methods involved the determination of protein composition using either HPLC or electrophoresis. 2.2 Geographical region of origin It is common for certain foodstuffs to be described as coming from a particular country or region. Very often this description is used as a cipher for product quality. The price of good quality wines is often largely based on the region they come from but the same can also be said of cheeses, sausages, olive oil and so on. European legislation (EEC No. 2081/92) has been developed for the protection of geographical indications and designations of origin for agricultural produce and foodstuffs.Chemical tests to determine region of origin remain in the early stages of development and it is likely that, to be practicable, databases of authentic produce parameters would necessarily have to be large. Thus food producers have tended towards the development of quality schemes certified by approved inspection bodies in order to control the quality (and hence the value of the labelling) of specified foodstuffs from a region. 2.3 Commercial treatment There are a number of commercial treatments to which foods may be subject. Some are considered desirable by the consumer, e.g., pasteurisation (although this does not add value to the product). Cold pressed (i.e., virgin) olive oil represents another example of a quality process (although the test for this procedure seeks to demonstrate the absence of refined oil). Other procedures, such as food irradiation or the use of GMOs, seek to benefit the consumer, in these cases through the retardation of spoilage or through the production of cheaper food.However, in much of Europe (and in contrast with the USA) these developments have been treated with considerable suspicion by consumers. 2.4 Water Water was one of the earliest adulterants of milk and beer. It still remains a common extender of foodstuffs and has been detected both in liquid products such as wine and also in meat products. 2.5 Brands Protecting the authenticity of brands is an important issue for most businesses whether it be Rolex watches or Levi jeans and foods are no exception. The Scotch whisky industry goes to considerable lengths to protect its brands from being counterfeited, particularly in some third world countries where Scotch can command premium prices and counterfeiting of brand is rife. 3 Recent developments in meat authentication Authenticity issues in meat and meat products have recently been reviewed by Hargin.5 Geographic origin, meat species (particularly in admixture) and treatment (whether meat described as fresh had in fact previously been frozen) were seen as key issues.Differences in the legislation of meat composition (e.g., use of blood plasma) between EC member states was seen as a potential problem. MAFF has undertaken two surveys in order to establish the prevalence of potential meat frauds in the UK market place.It was discovered that 16 out of 164 samples of cured meat failed to properly declare the amount of added water.6 It was also found that 44 out of 534 fresh meat samples had been previously frozen but were not labelled to this effect.7 This survey employed a comparative measurement of the activity of the enzyme b-hydroxyacyl-CoA dehydrogenase (HADH) before and after freezing the sample. This enzyme is released when mitochondria are disrupted by freezing; thus there is little change in enzyme activity when a previously frozen piece of meat is refrozen and then thawed whereas a considerable difference is found when fresh meat is analysed initially and is then reanalysed after being frozen and thawed.A survey of species substitution of raw and cooked meats carried out by the Florida Department of Agriculture and Consumer Services found 22.9% of cooked products and 15.9% of raw products contained meats of species other than that described, at levels in excess of 1%.Enzyme linked immunosorbent assay (ELISA) and agar-gel immunodiffusion were the techniques employed for the analysis.8 Research continues to develop better methods for determining the authenticity of meat. Myers and Yamazaki9 have investigated a new immunological technique in which antibodies to meat immunoglobulin (IgG) are bound to a polyester cloth. This format was considered superior to traditional microwell formats for ground meat samples containing fine meat particles because, in the latter, the meat particles may retard the diffusion of the sample IgG molecules.The development of DNA methods continues to have a major place in meat authentication. Hunt et al.10 used oligonucleotide probes to identifiy the species of origin of raw and cooked meat. The benefits of this procedure were that the polymerase chain reaction (PCR) approach was not required. This was advantageous because this equipment is not available in all laboratories and it may also give rise to undesirable assay variability. It was also considered superior to immunological techniques because the latter detect soluble plasma proteins.It has been argued that these are not meat and may arise from adventitious contamination with meat juices or blood. The oligonucleotide probe method works on intracellular DNA. A sample wash procedure can therefore be used to eliminate any cross-contamination from blood, making the procedure particularly effective for enforce- 152R Analyst, 1998, 123, 151R–156Rment purposes.Because the probes recognise relatively short segments of DNA, the method is applicable to processed and canned meat products. Consumers have a right to expect properly labelled meat products. However, when the mislabelling issue contravenes ethnic or religious mores, the issue becomes particularly sensitive. Al-Rashood et al.11 describe a method for detecting the presence of pork fat in processed foods.The method employs HPLC of the triacylglycerols and can detect as little as 5% pork fat (on a fat basis) in admixture with other meats. Meat products are potential adulterants in a much wider variety of foods than just other meat products. Agullo and Gelos12 described how the determination of free and bound cholesterol using gas chromatography (GC) can be applied to the detection of bovine blood plasma in egg pastas. 4 Recent developments in fish authentication The main issue in fish authentication is one of species.This is usually in the filleted product since before filleting the morphological features provide a good means of determining the species. The traditional technique for determining the species of raw fish is polyacrylamide gel electrophoresis (PAGE) of the sarcoplasmic proteins. After fish is cooked, the use of a denaturant, sodium dodecyl sulfate (SDS), is necessary to give good electrophoretic profiles. Craig et al.13 described the use of SDS-PAGE for the detection of other fish species in raw, reformed breaded scampi (Nephrops norvegicus).Heavily processed (e.g., autoclaved) products may require cyanogen bromide (CNBr) cleavage of proteins and for different species of closely related families such as tuna or salmon DNA procedures are preferred.14 Rehbein et al.15 reported a method of DNA analysis which could be applied to canned products from closely related species. They applied the PCR to amplify sections of the mitochondrial cytochrome b gene.These were then analysed using single strand conformation polymorphism (SSCP). In this procedure, single stranded DNA is folded such that complementary sections become bound together. Differences in the three dimensional structure caused by alterations in base pair sequences of the single strand lead to different electrophoretic mobilities which are visualised by silver staining. Differentiation of four eel species and three types of caviar was achieved.Ram et al.16 employed a different DNA technique for authentication of canned tuna and bonito. The mitochondria cytochrome b gene was again the focus of attention and the PCR was used to amplify the amount of DNA. However, in this study, sequencing followed by restriction site analysis was performed. This technique uses restriction enzymes which cleave double stranded DNA at defined base pairs to generate species specific DNA fragments. One DNA base pair change between species can therefore be detected by choosing a restriction enzyme which cuts the DNA at the site of this change.Different sized DNA fragments are then produced by the action of the restriction enzyme in the two species. Thus restriction analysis, SSCP and the use of oligonucleotide probes (as described for meat) represent the three major strands of research on the application of DNA methods to food authentication. Each of them has advantages for different authenticity issues and a more detailed description of these important methods has recently been published by Davidson.17 5 Recent developments in milk and cheese authentication The methods which have been applied to milk and cheese authentication are similar to those for meat products.However, since milk from healthy animals does not contain cellular material, the DNA methods are not applicable. The preferred methods are largely based on protein analysis and either involve typical protein chromatographic techniques (such as electrophoresis) or the use of antibody technology.Recio et al.18 have contributed a very useful review of capillary electrophoretic methods. CZE has been used to detect the adulteration of fresh milk with milk powders and for the determination of the fraudulent addition of rennet whey solids to dairy products (through detection of caseinomacropeptide), and is potentially useful for detecting the adulteration of milk with milks from other species.It is also useful in monitoring proteolysis in cheese production and in providing a measure of thermal treatment of milk. An electrophoretic ripening index for the evaluation of proteolysis using PAGE provided a means for assessing the quality of Parmesan cheese.19 Retail samples purchased in Italy were generally of good quality but those purchased in Austria suggested that adulteration with products with low proteolysis (e.g., cheese rind or very young cheese) had occurred.PAGE has also found application in determining bovine milk in Halloumi (ovine) cheese using analysis of the as1-casein to provide a detection limit of 2.5%.20 The issue of detecting the presence of milk from different species has also been addressed using immunological techniques.20,21 These methods can permit detection of as little as 0.1% milk from a foreign species. Geographic origin is an authenticity issue which is of particular concern to purchasers of cheese.The measurement of the stable isotopes 13C and 15N has been shown to be influenced by the region of origin of the milk. This is because the isotope values in milk are related to those of the fodder on which the cows are fed. Milk from regions dominated by grassland typically shows relatively negative d-13C values, but in regions dominated by crop cultivation the d-13C values are more positive. The d-15N values are influenced by factors such as soil conditions, the intensity of agricultural use and the climate.22 As yet, there are insufficient data to determine whether this approach is practicable for certifying the origin of milk and its products; however, the approach shows promise and may form an important component of a suite of tests for geographic origin.Another important area in the authenticity of milk products is the detection of non-milk fat. Ulberth23 applied multivariate regression analysis of fatty acid composition to the detection of tallow in admixture with milk fat.The method was able to determine as little as 1.2% tallow in milk fat using partial least squares (PLS) regression, which is superior to the traditional method of using the butyric acid (C4:0) content alone. The use of advanced statistical techniques is becoming increasingly important to food authentication since they represent the most convenient methods of interpreting data from a number of discrete analytes or methods.A number of different approaches are available and these have been cogently reviewed by Adams.24 6 Recent developments in vegetable oil authentication In common with milk, vegetable oil does not contain sufficient DNA to enable the newer biotechnological techniques to be used to determine the plant origin of the oil. Techniques for authenticating oils have therefore centred on compositional analysis, but there is always the danger that an adulterant can be found which will not be detected by these techniques.Stigmastadiene is a dehydration product of stigmastasterol formed during the refining process. It is therefore a valuable Analyst, 1998, 123, 151R–156R 153Rindicator (with trans fatty acids) of the presence of refined oils in cold pressed oils.25 Methods which detect the presence of a particular component of one oil which is not present in another have more limited usefulness since they merely highlight the potential unsuitability of the oil as an adulterant.Thus the high level of steryl esters in corn or rapeseed oil would permit their detection in a number of other oils such as soybean, groundnut, olive and palm, but would be less effective for detecting admixtures of the latter group.26 Similarly, the detection of tocopherols and tocotrienols in palm and grapeseed would permit their detection in olive, hazelnut, sunflower and soybean at levels as low as 1–2% but would be ineffective at detecting admixture within the latter class.27 Differences in carbon isotope composition are, except in the case of maize oil, largely insufficient for the authentication of oils.Nevertheless, Kelly et al.28 were able to distinguish sunflower oil from two other C3 oils (i.e., plants using the Calvin cycle) on the basis of d-13C values of individual fatty acids. Angerosa et al.29 took this concept of determining the isotope ratio of individual components of oils further. They were able to detect the addition of olive pomace oil to both virgin and refined olive oil at levels as low as 5% by measuring the d-13C value of the aliphatic alcohol fraction.This fraction contains less isoprenoids and methyl sterols in pomace oil, leading to a more negative isotope ratio. This procedure proved superior to existing methods, such as wax analysis, which are currently incorporated in legislation. An area of considerable development over the last few years has been the use of multivariate statistical approaches to interpret spectral data.Clearly the spectroscopic data are related to the composition of the food, but the chemical basis of this relationship is not always interpretable. Three spectroscopic approaches have been developed for authenticating oils. As yet, all these approaches have provided only a preliminary indication of promise and there is a need for a concerted effort from a number of laboratories to establish whether this promise is capable of being fulfilled. Fourier transform infrared (FT-IR) spectroscopy has been applied to the authentication of a number of commodities.It provides a very rapid analytical method which can be inexpensive if a large number of analyses are required. The near-infrared (NIR) region has been applied to compositional analysis30 and also to a successful classification of a small number of oils from different species.31 Downey32 has reviewed both the spectroscopic approach and the statistical procedures with a summary of applications to a range of food commodities. The mid-infrared (MIR) region has been tested for its ability to detect potential adulterants in laboratory generated mixtures of virgin olive oil and walnut or refined olive oil.33 FT-Raman spectroscopy has also been applied to the authentication of virgin olive oil.34 Adulteration with soybean, corn and olive residue oil was detected at 1, 5 and 10%, respectively, with 100% correct discrimination between genuine and adulterated samples.However, the procedures employing vibrational spectroscopy evaluate laboratory generated mixtures and it is not always clear how representative these will be of illegal commercial practices or whether the authentic samples used for generating training sets for the statistical evaluation were also used to prepare the adulterated samples. The application of sensory data from a taste panel has also been used to characterise different extra virgin olive oils.35 With taste providing one of the most important stimuli from food, it may seem surprising that sensory evaluation plays such a little part in authentication.The reason, of course, is that taste is subjective and difficult to quantify. By using the taste panel as an instrument to generate sensory data (rather than to make interpretations) it becomes possible to use the advanced mathematical techniques (such as multivariate statistics and fuzzy logic) applied to other instrumental methods.Linking sensory characteristics with the concentration of chemicals provides a potentially powerful means of authenticating oils which is directly linked to features which the consumer demands.36 Nuclear magnetic resonance (NMR) is the third spectroscopic technique which is being increasingly applied to food authentication. It can be used in a number of different ways. The entire spectrum can be used to generate a database which is subsequently interpreted by statistical techniques.In this it is analogous to using FT-IR or FT-Raman spectroscopy. Alternatively, it can be used to measure small amounts of specific compounds in the sample which are then used as markers of authenticity. Finally, the isotopic specificity of NMR can be exploited to provide a measurement of species specific isotopic ratios. Shaw et al.37 adopted the multivariate approach using 13C NMR spectra and found that they were able to differentiate the cultivar of a number of extra virgin olive oils in over 90% of cases and were also able to give some indication of the region of origin.Sacchi et al.38 were able to identify a number of minor components using high field 1H NMR which were markers of adulteration or were related to oil quality and freshness; however the technique is not yet sufficiently developed to represent an authentication technique. 7 Recent developments in essential oil authentication Essential oils are so called because they are “essences” rather than because they are considered “necessary”.The essential oils tend to have a very high unit cost, making their extension potentially very profitable. The use of NMR to provide isotopic measurements at defined chemical positions of the test molecule plays a key role in ensuring the authenticity of many of these materials. The 2H nucleus represents one of the best studied nuclei and this site-specific approach has been widely adopted by the Eurofins company under its SNIF-NMR trademark.This approach has been used for determining the authenticity of vanillin and p-hydroxybenzaldehyde from vanilla essence.39 As in so many of these cases, inclusion of d-13C data provides an extra analytical dimension and enhances the scope of the authentication. The 2H NMR approach has been adopted for benzaldehyde from bitter almond oil and cinnamon oil40 and for phenylethanol and phenylethyl acetate,41 and the combined 2H and d-13C approach for mustard oils.42 These isotopic methods provide one of the best means of determining the source of food components but the high value of the natural product can make it cost effective to go to considerable lengths to overcome them.Thus Remaud et al.39 demonstrated the presence of a dideuterated methoxyl group in a sample of vanillin which could only have occurred through the synthesis of an isotopically labelled adulterant prepared with the deliberate intention of subverting the test. Another method of authenticating essences is to use the chemical composition of minor components of the essence.This approach has been applied to vanillin.43 Often it is the case that many of the biologically important components of essences are chiral. Thus chiral GC methods have proved valuable and their value can be extended by including on-line d-13C measurements. 44 The use of two isotopic measurements by GC-IRMS (13C and 15N) was found valuable for the authentication of methyl N-methylanthranilate.45 The most notable recent development in authentication of flavours comes from 3H (tritium) analysis.46 This has been applied to the analysis of benzaldehyde where the short half-life of 3H (12 years) means that it is never found in petroleum derived materials.As yet, only the potential of the method has been demonstrated since the methodology used requires a large amount of sample. However, this difficulty might be overcome 154R Analyst, 1998, 123, 151R–156Rby the use of accelerator mass spectrometry, which is designed to measure radioactivity from small samples. 8 Recent developments in fruit products authentication As with fish, fruits are relatively easy to authenticate when they are whole. It is the act of processing them into other products such as fruit juice or wine which gives rise to the possibility of extension with cheaper materials. Fruits are largely composed of simple sugars and the ready availability of commercial sweeteners means that the potential for adulteration is great. 2H NMR and d-13C are the core methods for detecting the addition of beet sugar and cane or corn syrups to wine and fruit juices, respectively, from C3 plants (e.g., orange).47 These methods have also been extended to maple syrup48 and citrus honey49 ( 2H NMR). These methods are not easy to apply to pineapple juice since the d-13C value of pineapple is similar to that found for cane and corn products.Jamin et al.50 addressed this difficulty by comparing the carbon isotope ratio of the juice with those of the organic acids in the juice, since it was considered likely that these compounds would need to be added in order for the juice to retain an acceptable sugar/acid ratio. This approach of using an internal reference is common for isotopic methods and has been collaboratively tested for d-13C measurements from fruit pulps and sugars.51 This principle has recently been applied to detecting the addition of mixtures of beet and cane or corn sugar to fruit juices by analysis of the d-13C content of individual sugars.52 At this time, an area of great interest is the development of rapid, automatable sample treatment processes.Interest has been sparked by the measurement of d-18O in sugars from fruit juice which provides complementary information to other isotopic techniques.53 This research will certainly lead to further developments in food authentication.Isotopic methods seek to detect a signal from the major adulterant of a product. An alternative approach is to look for a minor component which might be present in the potential adulterant but which is present in much smaller amounts in the foodstuff tested. Oligosaccharide analysis has proved particularly useful for detecting the presence of commercial sweeteners in fruit juices.54,55 Initial work in this area used HPLC with pulsed amperometric detection, although it is now more common to use capillary GC of the silyl ethers. These approaches have been reviewed by Low56 and have also been applied to honey57 and maple syrup.58 Anthocyanins represent another key marker for detecting the addition of cheap fruits to more expensive fruit purees,59 particularly when maintaining colour is important.The method has been applied to detecting elderberry in red wine60 and jams.61 For species which do not contain colours, phenolic components such as dihydrochalcones can prove useful markers.62,63 Spectroscopic approaches which consider the entire sample composition have also been applied to fruit products.The approach taken is similar to that used for vegetable oils. Again, the main issue still to be resolved is the reliability of interpretation for unknown samples. If the basis on which samples are being classified is not well understood, then the effects of changes in, e.g., growing environment cannot be taken into account in the future.Nevertheless the approach has demonstrated promise for orange juice (NIR)64 and for fruit pur�ees65,66 and jam67 (MIR), for distinguishing the arabica and robusta coffee varieties68,69 and for detecting the adulteration of instant coffees with carbohydrates.70 It is noteworthy that a chemical basis was proposed for the ability to distinguish the coffee varieties, namely the chlorogenic acid and caffeine contents.An understanding of the reasons for particular classifications in this type of research remains rare. This therefore represents a valuable development because a clear understanding of the basis of a test permits its more general use, for instance on commodities from a country or region which is not represented in the database. Discovering the reasons for particular classifications of sample sets may be easier using NMR than IR. However, less research has been undertaken using NMR and pattern recognition techniques.The approach undoubtedly holds promise, as has been demonstrated for orange juice71 and for apple juice72 by 1H NMR. Colquhoun73 has recently provided a summary of this approach. As yet there are no well characterised methods for determining the region of origin of a product with any degree of certainty. However, the most promising approach seems to be the use of multi-element data together with a pattern recognition approach. The concept relies on the transfer of trace elements from the soil in the region of interest and hence has so far been best evaluated for fruit products.Baxter et al.74 were able unequivocally to classify white Spanish wines from three regions. The accuracy of classification fell to 95% when red and ros�e wines were included in the database. A similar approach has been adopted for the country of origin of orange juice.75 It may be that other analytical parameters (e.g., stable isotopes) wiltend the possibilities for certifying region of origin. 9 Effects of food fraud There can be no argument that consumers have a right to accurate, informative labelling. Studies linking sensory perceptions and chemical composition are therefore helpful in ensuring that molecules which have a sensory effect are present in premium quality foods. A unique study investigated the biological effect of mixing peppermint oil with corn-mint oil.76 Both materials exhibited a wide range of activity against different species of bacteria but both showed consistent spasmolytic activity on guinea-pig ileum suggesting that they are equally effective in treating conditions associated with smooth muscle (e.g., irritable bowel syndrome).Thus admixture of these oils would not necessarily disadvantage consumers who took the oils for this medical purpose. In most cases, the materials used for extending food are innocuous (e.g., sugar or water). However, if honest labelling is not enforced through legislation then the possibility arises that more harmful practices may ensue.The addition of ethylene glycol to Austrian wine some 10 years ago is a good example. What is less well known is that the incorporation of this material in “elixir sulfanilamide” in 1938 led to 105 deaths in the USA and forced fundamental changes in the operation of the Food and Drug Administration.77 The adulteration of food may lead to long term health effects in survivors.A follow up study of survivors of the Spanish toxic oil syndrome found that 58% still suffered symptoms 12 years after the poisoning event.78 There can therefore be no doubt of the need for continued vigilance in the determination of food authenticity. References 1 A. H. Hassall, Adulterations Detected or Plain Instructions for the Discovery of Frauds in Food and Medicine, Longman, Green, Longman and Roberts, London, 1861. 2 Food Surveillance Paper 46: Authenticity of Soluble Coffee, HM Stationery Office, London, 1995. 3 Food Surveillance Paper 44: Authenticity of Fish and Fish Products, HM Stationery Office, London, 1994. 4 Food Surveillance Paper 47: Authenticity of Dried Durum Wheat Pasta, HM Stationery Office, London 1995. 5 K. D. Hargin, Meat Sci., 1996, 43, S277. Analyst, 1998, 123, 151R–156R 155R6 MAFF Food Surveillance Information Sheet Number 132: Survey of Added Water in Cured Meat Products, MAFF, London, 1997. 7 MAFF Food Surveillance Information Sheet Number 87: Labelling of Fresh Meat and Poultry Survey, MAFF, London, 1996. 8 Y-H. P. Hsieh, B. B. Woodward and S.-H. Ho, J. Food Protect., 1995, 58(5), 555. 9 S. Myers and H. Yamazaki, Biotechnol. Tech., 1997, 11(7), 533. 10 D. J. Hunt, H. C. Parkes and I. D. Lumley, Food Chem., 1997, 60(3), 437. 11 K. A. Al-Rashood, E. M. Abdel-Moety, A. Rauf, R. R. Abou-Shaaban and K. I. Al-Khamis, J. Liq. Chromatogr., 1995, 18(13), 2661. 12 E. Agullo and B. S. Gelos, Food Res.Int., 1996, 29(1), 77. 13 H. Craig, A. H. Ritchie and I. M. Mackie, Food Chem., 1995, 52 451. 14 I. M. Mackie, in Food Authentication, ed. P. R. Ashurst and M. J. Dennis, Blackie, London, 1996, pp. 140–170. 15 H. Rehbein, G. Kress and T. Schmidt, J. Sci. Food Agric., 1997, 74(1), 35. 16 J. L. Ram, M. L. Ram and F. F. Baidoun, J. Agric. Food Chem., 1996, 44(8), 2460. 17 W. S. Davidson, in Analytical Methods of Food Authentication, ed. P. R. Ashurst and M. J. Dennis, Blackie, London, 1998, pp. 182–203. 18 I. Recio, L. Amigo and R. Lopez-Fandino, J. Chromatogr. B, 1997, 697(1–2), 231. 19 H. K. Mayer, Milchwissenschaft, 1997, 52(8), 443. 20 A. I. Haza, P. Morales, R. Martin, T. Garcia, G. Anguita, I. Gonzalez, B. Sanz and P. E. Hernandez, J. Food Protect., 1997, 60(8), 973. 21 W. Richter, I. Krause, C. Graf, I. Sperrer, C. Schwarzer and H. Klostermeyer, Z. Lebensm.-Unters. Forsch. A, 1997, 204(1), 21. 22 B. E. Kornexl, T. Werner, A. Rossmann and H.-L.Schmidt, Z. Lebensm.-Unters. Forsch. A, 1997, 205(1), 19. 23 F. Ulberth, J. Agric. Food Chem., 1995, 43(6), 1556. 24 M. J. Adams, in Analytical Methods of Food Authentication, ed. P. R. Ashurst and M. J. Dennis, Blackie, London, 1998, pp. 308–336. 25 L. Bruehlund and H. J. Fiebig, Fett Wiss. Technol., 1995, 97(6), 203. 26 M. H. Gordon and L. A. D. Miller, J. Am. Oil Chem. Soc., 1997, 74(5), 505. 27 F. Dionisi, J. Prodolliet and E. Tagliaferri, J. Am. Oil Chem. Soc., 1995, 72(12), 1505. 28 S. Kelly, I. Parker, M. Sharman, J. Dennis and I. Goodall, Food Chem., 1997, 59(2), 181. 29 F. Angerosa, L. Camera, S. Cumitini, G. Gleixner and F. Reniero, J. Agric. Food Chem., 1997, 45(8), 3044. 30 M. D. Guillen and N. Cabo, J. Sci. Food Agric., 1997, 75(1), 1. 31 K. M. Bewig, A. D. Clarke, C. Roberts and N. Unklesbay, J. Am. Oil Chem. Soc.,1994, 71(2), 195. 32 G. Downey, J. Near Infrared Spectrosc., 1996, 4, 47. 33 Y. W. Lai, E. K. Kemsley and R. H. Wilson, Food Chem., 1995, 53(1), 95. 34 V.Baeten, M. Meurens, M. T. Morales and R. Aparicio, J. Agric. Food Chem., 1996, 44(8), 2225. 35 R. Aparicio, J. J. Calvente and M. T. Morales, J. Sci. Food Agric., 1996, 72(4) 435. 36 R. Aparicio, M. T. Morales and V. Alonso, J. Agric. Food Chem., 1997, 45(4), 1076. 37 A. D. Shaw, A. Di Camillo, G. Vlahov, A. Jones, G. Bianchi, J. Rowland and D. B. Kell, Anal. Chim. Acta, 1997, 348(1–3), 357. 38 R. Sacchi, M. Patumi, G. Fontanazza, P. Barone, P.Fiordiponti, L. Mannina, E. Rossi and A. L. Segre, J. Am. Oil Chem. Soc., 1996, 73(6), 747. 39 G. S. Remaud, Y.-L., Martin, G. G. Martin and G. J. Martin, J. Agric. Food Chem., 1997, 45(3), 859. 40 G. Ramaud, A. A. Debon, Y.-L. Martin, G. G. Martin and G. J. Martin, J. Agric. Food Chem., 1997, 45(10), 4042. 41 G. Fronza, C. Fuganti, P. Grasselli, S. Servi, G. Zucchi, M. Barbeni and M. Cisero, J. Agric. Food Chem., 1995, 43(2), 439. 42 G. S. Remaud, Y.-L. Martin, G. G. Martin, N.Naulet and G. J. Martin, J. Agric. Food Chem., 1997, 45(5), 1844. 43 A. Kaunzinger, D. Juchelka and A. Mosandl, J. Agric. Food Chem., 1997, 45(5), 1752. 44 A. Mosandl, Food Rev. Int., 1995, 11(4), 597. 45 S. Faulhaber, U. Hener and A. Mosandl, J. Agric. Food Chem., 1997, 45(7), 2579. 46 M. P. Neary, J. D. Spaulding, J. E. Noakes and R. A. Culp, J. Agric. Food Chem., 1997, 45(6), 2153. 47 G. G. Martin, V. Hanote, M. Lees and Y.-L. Martin, J. AOAC Int., 1996, 79(1), 62. 48 G.G. Martin, Y.-L., Martin, N. Naulet and H. J. D. McManus, J. Agric. Food Chem., 1996, 44(10), 3206. 49 P. Lindner, E. Bermann and B. Gamarnik, J. Agric. Food Chem., 1996, 44(1), 139. 50 E. Jamin, J. Gonzalez, G. Remaud, N. Naulet and G. G. Martin, J. Agric. Food Chem., 1997, 45(10), 3961. 51 A. Rossmann, J. Koziet, G. J. Martin and M. J. Dennis, Anal. Chim. Acta, 1997, 340(1–3), 21. 52 M. P. Day, P. Correia, R. Evans and D. A. Hammond, Fruit Process., 1998, 8(3), 86. 53 J.Koziet and E. Bricout, Sci. Aliments, 1997, 17(4), 403. 54 N. H. Low, J. AOAC Int., 1996, 79(3), 724. 55 T. A. Iuliano, J. AOAC Int., 1996, 79(6), 1381. 56 N. H. Low, in Analytical Methods of Food Authentication, ed. P. R. Ashurst and M. J. Dennis, Blackie, London, 1998, pp. 97–137. 57 N. H. Low and W. South, J. AOAC Int., 1995, 78(5), 1210. 58 J. G. Stuckel and N. H. Low, Food Res. Int., 1996, 29(3–4), 373. 59 A. Versari, D. Barbanti, S. Biesenbruch and P. J. Farnell, Ital. J. Food Sci., 1997, 9(2), 141. 60 P. Bridle and C. Garcia-Viguera, Food Chem., 1996, 55(2), 111. 61 C. Garcia-Viguera, P. Zafrilla and F. A. Tomas-Barberan, J. Sci. Food Agric., 1997, 73(2), 207. 62 A. Versari, D. Barbanti, S. Biesenbruch and P. J. Farnell, Lebensm.- Wiss. Technol., 1997, 30(6), 585. 63 M. L. Bengoechea, A. I. Sancho, B. Bartolome, I. Estrella, C. Gomez- Cordoves and M. T. Hernandez, J. Agric. Food Chem., 1997, 45(10), 4071. 64 M. Twomey, G. Downey and P. B. McNulty, J. Sci. Food Agric., 1995, 67(1), 77. 65 M. Defernez, E. K. Kemsley and R. H. Wilson, J. Agric. Food Chem., 1995, 43(1), 109. 66 E. K. Kemsley, J. K. Holland, M. Defernez and R. H. Wilson, J. Agric. Food Chem., 1996, 44(12), 3864. 67 M. Defernez and R. H. Wilson, J. Sci. Food Agric., 1995, 67(4), 461. 68 R. Briandet, E. K. Kemsley and R. H. Wilson, J. Agric. Food Chem., 1996, 44(1), 170. 69 G. Downey and J. Boussion, J. Sci. Food Agric., 1996, 71(1), 41. 70 R. Briandet, E. K. Kemsley and R. H. Wilson, J. Sci. Food Agric., 1996, 71(3), 359. 71 J. T. W. E. Vogels, L. Terwel, A. C. Tas, F. van den Berg, F. Dukel and J. Van der Greef, J. Agric. Food Chem., 1996, 44(1), 175. 72 P. S. Belton, I. Delgadillo, A. M. Gil, P. Roma, F. Casuscelli, I. J. Colquhoun, M. J. Dennis and M. M. Spraul, Magn. Reson. Chem., 1997, 35, S52. 73 I. J. Colquhoun, Spectrosc. Eur., 1998, 10, 8. 74 M. J. Baxter, H. M. Crews, M. J. Dennis, I. Goodall and D. Anderson, Food Chem., 1997, 60(3), 443. 75 G. J. Martin, J. B. Fournier, P. Allain and Y. Mauras, Analusis, 1997, 25(1), 7. 76 M. Lis-Balchin, S. G. Deans and S. Hart, Med. Sci. Res., 1997, 25(3), 151. 77 P. M. Wax, Ann. Intern. Med., 1995, 122(6), 456. 78 L. D. Kaufman, M. I. Martinez, J. M. Serrano and J. J. Gomez-Reino, J. Rheumatol., 1995 22(2), 282. Paper 8/02892C 156R Analyst, 1998, 123, 151R–156R

ISSN:0003-2654    

DOI:10.1039/a802892c

出版商:RSC    

年代:1998

数据来源: RSC

摘要:

Factorial correspondence regression applied to multi-way spectral data Nicolas Gouti, Douglas N. Rutledge and Max H. Feinberg* Institut National de la Recherche Agronomique, Laboratoire de Chimie Analytique, 16 Rue Claude Bernard, 75231 Paris Cedex 05, France. E-mail: feinberg@inapg.inra.fr Received 18th February 1998, Accepted 26th June 1998 The principal advantage of factorial correspondence analysis, where rows and columns are processed symmetrically, is the possibility of having in the same factorial space observation (row) and variable (column) projections.This joint plot allows one to find similarities that may exist between variables and observations in term of distances. When dealing with picture sequences, this joint plot is composed of pixel and picture projections. For a sequence of spectra, the joint plot is composed of projections of the wavelengths or frequencies and of the spectra. In the reported study, 2D data sets were formed by outer product between mid- and near-infrared spectroscopic data recorded on edible oil samples with different levels of unsaturation. The association of a regression technique with factorial correspondence analysis gives a convenient way to detect interactions between wavenumbers and wavelengths as a function of the level of unsaturation.This new technique is called factorial correspondence regression. The mathematical procedure is developed and the validation method, which is based on a cross-validation procedure to choose independent variables entering the regression equation, is reported. The results obtained from the proposed method are presented in the form of maps for a graphical interpretation that allows much easier assignments of near-infrared peaks to combinations of mid-infrared peaks.Mid- (MIR) and near-infrared (NIR) spectroscopy are both widely used analytical techniques to provide qualitative and/or quantitative information about many kinds of samples, including food products.MIR is specifically used for the determination of the molecular structure of analytes, whereas NIR is rather used for quality control in industry, mainly in the agrofood sector. With the development of NIR applications, there is a need to understand better the chemical nature of the spectral features selected to build the prediction models used for routine NIR analysis. The question would be almost solved if it was possible simply to assign an MIR wavenumber to an NIR wavelength starting from the theoretical chemical background of both techniques.However, these assignments remain difficult as the NIR absorption bands are not only wide and overlapping but also correspond to combinations and harmonics of the fundamental molecular vibrations observed directly in the MIR region. However, an assignment of NIR bands to particular MIR vibrations could be made possible by correlating the spectra of samples recorded in both domains for a given analytical application.If this assignment is correct and feasible, it could be possible to substitute one of the methods with the other, for instance, in order to reduce the cost of a quality control procedure. Devaux et al.1 applied canonical correlation analysis (CCA) to study the connection between these two spectral domains but this method cannot be used to build prediction models. Another strategy was first described by Rutledge and co-workers2,3 to highlight the interactions that may exist between two spectra obtained on a set of samples.It consists in calculating, for each sample, the product of the intensities for all combinations of the frequencies of the two spectra. This calculation is known as the outer product of the two vectors and allows one to create a matrix where each element represents the cross product of the intensities of two spectral scale units (wavelengths, wavenumbers, shifts, etc.). This matrix can be displayed as an image or a picture and we used this representation to develop new processing techniques derived from more classical image processing techniques.Moreover, when the outer product operation is applied to several samples, a set of matrices is built which can be treated in a way similar to an image sequence. In this work, a procedure based on factorial correspondence analysis (FCA) was applied to such sequences of matrices generated by combining MIR and NIR spectra.It is more classical to apply principal component analysis (PCA) to the multivariate image analysis technique.4 Several applications of PCA to multivariate images have been presented in food and environmental chemistry.5,6 The main purpose of this technique is to highlight negative or positive correlations among different images in a sequence or among pixels from images and to discard the redundant information. However, FCA has also been applied to multivariate image sequences.FCA differs from PCA in that observations and variables are mathematically treated in the same way. Therefore, it is preferred to speak of rows and columns rather than observations and variables when using FCA, to underline this specificity. FCA was applied by Trebbia and Bonnet7 as an image filtering technique for background and noise elimination in electron energy loss spectroscopic elemental mapping. It can be run without any a priori hypothesis about the noise; it merely analyzes the variance between the pixel intensities of an image collection treated as a single set.Furthermore, it makes a distinction between the variance due to the real presence, in some pixels, of the researched signal and the variance induced by both the background and the noise. The main advantage of using FCA on multivariate data sets is the possibility of projecting rows and columns simultaneously on to the same score plot. This joint plot can reveal relationships between these two dimensions.Whatever the technique used, when applied to sets of matrices, it is necessary to pre-process the data before computation. The main step consists in reorganizing each rectangular picture (I by J pixels) into a vector of I 3J elements. Three different kinds of reorganization are possible according to the analytical problem under study.8 However, the assignment of NIR bands to particular MIR vibrations must be performed for a specific analytical purpose because, as we said, it is not possible to establish a deterministic Analyst, 1998, 123, 1783–1790 1783relationship between both these signals.For this application, we used a set of spectra acquired for various edible oils with different unsaturation levels. These oils were composed of oleic and linoleic acids with other fatty acids, at different concentrations, giving different global levels of unsaturation. Our goal was to characterize the links between the MIR and NIR spectra regarding the prediction of the unsaturation level of an edible oil, in order to check if it would be possible to replace the more expensive MIR technique with a cheaper NIR instrument. Theory All computations and programming were carried out using Matlab.9 This high-level matrix-oriented programming environment is very suitable for image and statistical processing.The Matlab programs are available from the authors. The following notation is used in the rest of the text: bold italic upper case (X, Z or Fr) for matrices, bold italic lower case (y, eor z) for vectors, italic upper case (N, P or J) for constants and italic lower case (i, j or xnp) for indices of matrix elements.Factorial correspondence analysis The major differences between the FCA and PCA techniques is that FCA produces a factor space that simultaneously represents rows and columns, as a result of the symmetrical treatment of the multivariate data set.Basically, PCA makes a clear distinction between data matrix dimensions: rows (or observations) are considered to be different from columns (or variables), and the metric used to measure the distance between two points is a Euclidean distance. In contrast, FCA maintains complete symmetry between rows and columns and uses a different metric called a c2 metric. Therefore, in order to adapt the PCA algorithm for FCA, one has to make a transformation of the data set X (N rows 3 P columns) to build a new matrix Z such that the c2 metric used on X is equivalent to a Euclidean metric on Z.7,10 In matrix form, Z can be written as Z = D21 r FD21/2 c (1) where Dr and Dc are the diagonal matrices containing, respectively, the row and column sums of the frequency matrix F that is deduced from X (sum of all elements equal to 1).Z represents the new row coordinates in the vector space defined by the columns. In order to calculate correspondence factors (CFs) from this new matrix, a symmetric matrix V, also called inertia matrix, is calculated as V = ZADrZ 2 AcAc = (D21/2 c FD21/2 r )(D21/2 r FD21/2 c ) 2 AcAc (2) where the vector c represents the main diagonal of Dc.It defines the coordinates of the center of gravity of the N vector rows. Eigenvalue matrix L and eigenvector matrix U are calculated by diagonalizing this symmetric matrix. The rank of V is Q = min(P,N) 2 1 and the smallest latent root is therefore zero. Finally, the row projections Fr and the column projections Fc on to the Q CFs are deduced from the following relations: Fr = ZU = D21 r FD21/2 c U (3) Fc = D21 c FFrL21/2 (4) The projection matrices Fr and Fc are in the same factorial space and can be simultaneously plotted on the same joint plot.Fig. 1 shows how such a joint plot is constructed. Thus, in FCA rows and columns can be transposed since they are used in a symmetrical way for computation. In the case studied, Fr has been defined as the picture projections in the factorial space and Fc as the pixel projections.Data pre-processing and variable coding The only prerequisite for FCA is that the data array contains non-negative numbers and the types of data that can be treated can be either discrete or continuous. Among discrete data, one may use ordinal (or categorical) variables for which the ordering of the values does have a meaning. Continuous variables can be kept as they are or coded as ordinal variables.This coding operation may in some cases result in a loss of metric information but it may also have some advantages.11 For instance, the coding can impose a linear relationship among non-linearly related variables and decrease the influence of the extreme values. In addition, the loss of the metric information is often not important in multivariate statistical analysis when all variables are scaled to give a very small range. The categorical coding algorithm consisted in partitioning each continuous variable distribution into three distinct classes: low, medium and high level.Partitions are defined by searching thresholds in the actual data distribution that increase the information content of each class. For each variable, the method is as follows: (i) sort observations in ascending order; (ii) calculate the difference between successive observations; (iii) search the maximum difference for specified data intervals; and (iv) define category limit as the average value between the two observations with the greatest difference. This scanning procedure is applied to both halves of the distribution and it is possible to define a low to medium class limit and a medium to high class limit in such a way that three classes are built.Among all possible partitions detected, the optimum one is that which gives the lowest standard deviation between class sizes. This coding procedure was called the maximum difference thresholding (MDT) algorithm12 and was used to transform continuous variables into categorical variables by replacing each value by its class number.Definition of a region of interest (ROI) It is possible to extract relevant information from the picture sequence just by classifying the pixels into classes; this process is known as segmentation. Segmentation was applied to the joint plot of the Fc and Fr. In anticipation, all vectors of Fc were folded back to give factorial pictures.Paired intensity values b1 ij and b2 ij from the two first factorial pictures were used as coordinates to construct the joint plot. Similarly, the same Fig. 1 Classification by factor space segmentation. A joint plot of row Fr and column Fc projections is achieved. Column projections are in the form of a density plot. Interesting zones are selected and back-projected to image space as a binary image. 1784 Analyst, 1998, 123, 1783–1790projection was applied to Fr and paired intensity values were denoted by a1 n and a2 n.Because of the very large number of columns, the joint plot was difficult to visualize and it was proposed to illustrate projections as density curves. This density plot shows the presence of clusters, gradients and outlying pixels. Point projection distance can be translated into correlation, whereas this interpretation would be inadvisable when using classical PCA. Then, according to the respective locations of row projections and column projections, a characteristic set of points may be selected by the user; this is a region of interest (ROI).The elements of an ROI are back-projected as a binary (black and white) picture as shown in Fig. 1. A complete segmentation, where each ROI can be related to a particular sample, may lead to consistent conclusions for the set of matrices under study. Factorial correspondence regression (FCR) Multivariate image or picture information can also be used for regression and calibration purposes.The application of principal component regression (PCR) and partial least squares (PLS) regression to matrices such as images has already been demonstrated. As shown earlier, FCA is also a principal component-related analysis resulting in two sets of scores, Fr and Fc. Hence regression techniques may be used on either Fr or Fc, depending on the type of the response vector or matrix that is available. This new regression technique will be called factorial correspondence regression (FCR).For the purposes of the study, the regression equation was established between the vector of the levels of unsaturation and the Fr matrix. Each initial matrix corresponds to an oil sample and the regression model can be used to find columns that are most significantly related to the level of unsaturation. However, the same modeling technique could be applied to Fc as a function of any dependent variable. Let g be the vector of regression coefficients such as y = Fr g+ e y = Z b+ e with b = U g (5) where y is the response vector or dependent variable, Fr are the independent variables to which a column of ones was added in order to determine the intercept and eis the vector of residuals.The classical solution of eqn. (5) giving reasonable values for g has been extensively described in the literature13 and is obtained using the ordinary least squares (OLS) algorithm: g g g 0 1 M Q E I IIII ¢� ¡Æ ¢«¢«¢«¢« g = (FArFr)21FAry with g = The interpretation of g gives the factorial axes that are the most important to characterize the predicted variable. Since in FCA row projections Fr and column projections Fc are defined in the same factorial space, it is possible to apply the regression coefficients to Fc, as shown in the Fig. 2, leading to the following regression equation: Fc g = z The resulting vector zof P = I 3 J elements is expressed in the same unit as y. By inverse reorganization of the vector, we obtain a picture where each pixel value can be interpreted as the influence of the corresponding variable on the dependent variable.A plot of this picture allows a visual interpretation of the overall variable influences. In order to highlight the most significant variables, z is Studentized as zstd = z2 z s z (6) where z is the mean value and sz the standard deviation of the matrix. In this way, high absolute values identify those variables which are most significantly involved with the level of unsaturation.Thus, critical t-values can be defined for an eventual classification of the elements in order to simplify the interpretation. Five classes were created based on two critical tvalues that correspond, respectively, to bilateral risk levels of 10% and 25%: a = 10% ¡í ¢®t1¢® = 1.645 a = 25% ¡í ¢®t2¢® = 1.150 The five classes are as follows: 1. zstd @ 2 Variable intensities decrease greatly when the response increases. 2. 2t1 < zstd @ 2t2 Variable intensities decrease slightly when the response increases. 3. 2t2 < zstd @ t2 Variables with no influence. 4. t2 < zstd @ t1 Variable intensities increase slightly when the response increases. 5. zstd > t1 Variable intensities increase greatly when the response increases. Projections of classes 1, 2, 4 and 5 in picture space may lead to visual interpretation in the form of two maps. In the case studied here, dependent variables represent interactions of NIR and MIR frequencies and this arbitrary mapping technique clearly illustrated frequency couples that evolved significantly with the level of unsaturation.When dealing with any predictive regression model, it is necessary to check the consistency of the predicted values with actually observed values; this process is called model validation. The most effective way to validate a regression model consists in comparing new response values from a new data set with response values predicted with a model built with the old data set. In many situations, collecting new data for validation purposes is not possible. Cross-validation is a technique that can be applied when no new data are available.It consists in randomly discarding about 20% of observations from the data set for model building and then using this left out group as a ¡®new¡� data set in order to validate the model¡�s predictive ability. This process is repeated until each observation has been discarded at least once.14,15 Let Z(2G) be the submatrix of Z in which a group of nG observations have been discarded and y(2G) the corresponding response vector.Let Z(G) and y(G) be their respective complements, i.e., submatrix and vector containing discarded values. An inertia matrix V(2G) is computed, as in eqn. (2), for Z(2G) giving a new eigenvector matrix U(2G). Then a new row projection matrix Fr (2G) is defined as in eqn. (3) and estimated Fig. 2 Principle of factorial correspondence regression. Analyst, 1998, 123, 1783¡©1790 1785OLS coefficients �g(2G) are obtained, leading to estimated coefficients �b(2G) as in eqn.(5). Predicted values of group G are then obtained in matrix form: �y(G) = Z(G) �b(2G) It is possible to compute an error of prediction for each run by this technique. However, it is preferable to compute this error as a function of the number of correspondence factors (CFs) entered in the model. The criterion used is the well known root mean square error of cross-validation (RMSECV), defined for a specific number of CFs A as RMSECV RMSECV PRESS A G G G G G k K A g g g n k K g A G Kn y y Kn Kn G = - [ ]¢ - [ ] = - [ ] = = = =    1 1 2 1 1 y y y y ( ) ( ) ( ) ( ) � � � where yg and �yg are the elements of vectors y(G) and �y(G), respectively, K is the number of iterations performed, nG is the number of observations discarded and PRESS is the predictive error sum of squares criterion.In order to have a better estimate of expected variations when the model is applied, the product K 3 nG is defined in such a way that all observations have been discarded the same number of times. The Matlab program that gives the K subsets of calibration and prediction is available from the authors. An RMSECV value is computed each time a new CF enters the model. The model that gives the first local minimal RMSECV is considered as the optimum model. Two different methods for selecting the CFs were used in order to obtain the best predictive model: the top-down method and a stepwise regression procedure.The top-down method was used with the first A CFs, A being defined as the optimum number of such factors.16 It is well known that too large a value for A leads to overfitting and bad prediction due to the introduction of ‘noise’, whereas too small a value causes underfitting due to systematic error. Since the first CFs used as independent variables are not necessarily the most appropriate to model the response vector y, a stepwise regression procedure may be used in which CFs are selected sequentially according to their absolute correlation coefficients with the part of y not yet modeled.17 More precisely, the stepwise algorithm consists of the following steps: 1.Select K prediction subsets of nG observations each, and deduce the K corresponding calibration subsets. 2. Find the first CF to enter the model: For k = 1 to K (a) Calculate new CFs for the reduced data matrix Z(2G), as in eqns.(2) and (3). (b) Note the index of the CF that is most correlated to the reduced response y(2G). End (c)The CF that is most often correlated to the reduced response vector is the first one to enter the model. 3. Calculate the first RMSECV test value as RMSECVtest = ¢ 1 N y y. 4. For k = 1 to K (a) Same as 2(a). (b) Estimate the regression coefficients �b(2G) using the current model as in eqn.(5): �y(2G) = Z(2G) �b(2G) with �e(2G) = y(2G) 2 �y(2G) (c) Compute the partial PRESS value for the left out group G of nG observations: PRESSk G G G G g g g n y y G = - [ ]¢ - [ ] = - ( ) =  y y y y ( ) ( ) ( ) ( ) � � � 2 1 (d) Note the index of the CF the most correlated to �e (2G), the vector of residuals. End 5. Compute the current model RMSECV as: RMSECV PRESS model = =  1 1 KnG k k k 6. If RMSECVmodel < RMSECVtest, then (a) RMSECVtest = RMSECVmodel, and the CF the most often correlated to the reduced vector of residuals enters the model.Return to step 4. Else (b) the procedure stops and the last CF entering the model is removed. This leads to a model being the best predictive one. Sampling plan Experimental multi-way spectral data The spectral matrices of edible oils used in this work have been analyzed previously by CCA1 and by analysis of variance.2 The 13 samples (the code used for graphical display is given in parentheses) consisted of three grapeseed oils (G), two olive oils (O), two peanut oils (P), two rapeseed oils (R), two maize oils (M), one sunflower oil (S) and one walnut oil (W).A total of 39 spectra in the two infrared spectral ranges were obtained by collecting triplicate spectra for each oil sample. The MIR spectra ranged from 3018 to 2796 cm21 with a step of 2 cm21 (J = 111 wavenumbers). Spectra were obtained on an IFS25 Fourier transform spectrometer (Bruker, Karlsruhe, Germany) equipped with a DTGS detector.An attenuated total reflection cell was used for the sampling. The NIR spectra ranged from 1600 to 2408 nm with a step of 8 nm (I = 101 wavelengths) and were recorded on an Infralyzer 500 instrument (Technicon, Tarrytown, NY, USA); the samples were placed in a Bran + Luebbe liquid cell that allowed transflectance measurements. In order to avoid scaling effect due to recording conditions, each spectrum was scaled to unit variance. Table 1 gives the average fatty acid composition of these oils obtained from the literature.1 For each oil, the value of the calculated unsaturation level (CUL) was computed by summing fatty acid proportions weighted by their corresponding number of double bonds.This scale was used as the observed response for the regression model. The goal was to select the most informative spectral features from both the MIR and NIR spectra. It would have been more interesting to have the actual fatty composition of samples, because of the natural variability of such samples, but this information was not available.Table 1 Fatty acid composition of oils (with the number of double bonds) Oil Saturated Oleic (1) Linoleic (2) Linolenic (3) CUL* Olive 11 77 8 0 93 Peanut 18 57 20 0 97 Rapeseed 10 55 22 7 120 Maize 12 40 45 0 130 Sunflower 9 33 54 0 141 Grapeseed 12 22 65 0 152 Walnut 7 19 66 8 175 * CUL is the calculated unsaturation level obtained by summing the fatty acid proportions weighted by their corresponding number of double bonds. 1786 Analyst, 1998, 123, 1783–1790The number of oil samples and the number of replicates are reasonable for the elaboration of a regression model. Moreover, the range of the CUL values, from 93 to 175, covers the domain of the unsaturation level variabilitproduct matrix and data pre-processing For each sample, all intensity values for one spectral domain were multiplied by the intensity values of the other domain, producing a matrix containing all possible combinations of the intensity values in both domains.In mathematical terms, this operation is the outer product of two vectors (also called Kronecker product) that gives a data matrix which can be presented as a picture when using an arbitrary color (or gray) scale for each element.2,3 The outer product may be used as a general starting point for studying interactions between variables. This technique facilitates interpretation since results may be visualized as pictures or maps.The application of the outer product operation to all the MIR and NIR spectra produced N (39) matrices of size I 3 J (101 3 111 = 11 211) elements as illustrated in Fig. 3. Reasoning in absolute values, each value is directly related to intensity values in both spectral domains: (i) if the intensities are simultaneously high in the MIR and NIR spectra, the product is high; (ii) if the intensities are simultaneously low, the product is low; (iii) if one is high and the other low, the resulting product tends to an intermediate value.This 3D matrix was reorganized into a 2D data matrix by unfolding each matrix as a row vector of P = I 3 J elements. The dimensions of this new matrix were N = 39 rows and P = 11 211 columns. The outer product matrices were pre-processed by applying the maximum difference thresholding procedure to each column so that the quantitative elements were coded as categorical variables.The resulting data matrix X contained categorical variables where each value consisted in one of the three-class integer values. As will be shown below, the categorical coding step is useful for removing any non-linear relationship among original variables by creating new linearly related variables. Results and discussion Spectral assignment by back-projection of the regions of interest Factorial correspondence analysis was applied to the coded data matrix X.The kernel algorithm was used for the computation as it handles very large numbers of variables (P = 11 211) more quickly.18 The first two CFs contain 80% of the overall variance. The rows are oil samples and the columns are the wavenumber–wavelength couples. The joint plot of row and column projections in Fig. 4 shows the specific distribution of oil samples and wavenumber–wavelength couples according to the unsaturation level. It is clear that the oil samples are scattered according to their botanical origin on a diagonal running from the top right down to the bottom left corner.In fact, this distribution can be related to the concentration of fatty acids and therefore to the CUL values calculated in Table 1. For example, olive oils have the lowest CUL (93) and are located opposite walnut oils, which have the highest CUL (175). This trend is illustrated by the isoresponse curves of CUL fitted by multiple linear regression to the scores of the two first CFs.A determination coefficient of r2 = 0.83 was obtained, indicating a good relationship between the unsaturation level and the measured spectra. Column projections are not individually projected, but reported as a density plot with gray-level scale indicating their density. This allows a better visual interpretation and makes pixel selection much easier. Density curves are arranged in a circular shape and classes of interest can be defined.Since most saturated oil samples, such as olive oils, are at the bottom left corner of the figure, a first ROI was selected in that area. The same process was repeated for columns close to walnut oil samples in order to build ROI 2, which is representative of the most unsaturated oils. Two binary pictures were obtained by back-projecting ROI 1 and ROI 2 on to picture space. For better visualization, both binary pictures are represented on the same graph as shown in Fig. 5. Average spectra of the NIR data set and the MIR data set are reported in order to relate interactions to spectral features in each domain.This segmented picture highlights most representative columns of matrix X for each ROI and, consequently, for each oil type. For the ROI 1, frequencies in the MIR domain around 2902 and 2846 cm21, which correspond to the –CH2 symmetric and asymmetric stretching modes, respectively, are linked to 1648 nm in the NIR domain. These interactions are representative of saturated oils since the number of CH2 groups increases when the unsaturation level decreases.For the ROI 2, MIR spectral regions at 2956 and 2864 cm21 are characteristic of the asymmetric and symmetrical stretching vibration modes of the –CH3 group. These vibrations are linked to the NIR region between 2272 and 2400 nm. Although there are fewer –CH2 groups in unsaturated oils, the relative number of –CH3 groups is greater and the intensities of related bands are greater.The region at 3008 cm21 corresponds to the NCH cis stretching vibration. This vibration is also linked to 2144 nm in the NIR domain; this vibration corresponds to a combination of the NCH cis stretching and the CNC stretching vibration that occurs at 1660 cm21. Fig. 3 For each oil sample, a matrix is formed by multiplying all MIR spectrum intensities by all NIR spectrum intensities. It is an outer product or a Kronecker product of two vectors. Fig. 4 Joint plot on correspondence factors 1 and 2.The sample projections are olive (O), peanut (P), rapeseed (R), maize (M), sunflower (S), grapeseed (G) and walnut (W). Isoresponse curves for CUL are indicated as dotted lines. The column projections are indicated by their density using a gray scale. Two selected regions of interest are represented as dashed-frame boxes, ROI 1 characterizing low unsaturated oils and ROI 2 high unsaturated oils. Analyst, 1998, 123, 1783–1790 1787This example demonstrates that it is possible to detect the links between the two spectroscopic techniques and obtain consistent wavelength selection rules by simply selecting an adequate ROI on the joint plot.All these first results were confirmed by projecting columns on higher correspondence factors. Nevertheless, this approach remained subjective since the ROI needs to be defined by the operator. A simple way to avoid any operator intervention consists in applying a continuous classification of the variables by means of regression on to the unsaturation level.The regression technique used was FCR. The objective here is to bring out all possible links between the wavenumbers of MIR and the wavelengths of NIR spectra in a reduced number of maps. Factorial correspondence regression on the level of unsaturation An optimum regression model for Fr was selected by crossvalidation using both techniques proposed to select the CFs. In order to have a good estimate of the RMSECV criterion, the number of runs was set to K = 78 with nG = 7 left out observations at each time.The same K subsets were used in the two CF selection methods. The optimum number of CFs found for the top-down method was five, with the rule of the first local minimum, and the RMSECV value was 6.73. A smaller value of 5.68 was obtained when using the stepwise regression procedure describes above. The bar graph in Fig. 6 shows the results obtained at each iteration step during the cross-validation process.It represents the relative number of times each CF had been retained as the most correlated factor to the part of y not yet modeled. The resulting set of CFs {2, 1, 3, 4, 5, 8, 14, 15, 7} was used for the FCR on the calculated unsaturation level. The FCR method was applied to the data set with the subset of CFs defined above. The resulting Studentized values were split into five classes based on the two arbitrary critical t-values, leading to two distinct maps.The map in Fig. 7 highlights interactions between MIR and NIR domains that decrease when the unsaturation level increases. Conversely, the map of Fig. 8 highlights the interactions that increase with the unsaturation level; this means that the interaction t-value intensities increase when the unsaturation level increases. In both of these maps, there are some regions that are well discriminated. For simplification we refer to (2) when interactions are in Fig. 7 and (+) when interactions are in Fig. 8. As a first observation on Fig. 7, the interactions involving –CH2 group bands (2914 and 2850 cm21) are mainly linked (2) to the Fig. 5 Back-projection in the picture space of ROI 1 and ROI 2 for low and high unsaturation level samples, respectively. Fig. 6 Performance of stepwise selection technique. Most often selected correspondence factors are shown as a gray cylinder. Fig. 7 Back-projection of zstd into picture space.Only classes 1 and 2 are reported; they represent the interactions whose intensities decrease when the unsaturation level increases. Fig. 8 Back-projection of zstd into picture space. Only classes 4 and 5 are reported; they represent the interactions whose intensities increase when the unsaturation level increases. 1788 Analyst, 1998, 123, 1783–1790unsaturation level. On the other hand, in Fig. 8, interactions involving –CH3 (2960 and 2868 cm21) and NCH (3008 cm21) bands are (+) and consequently correlated to the unsaturation level.The features at 1696 and 1760 nm which are related to the –CH3 and NCH vibrations (+) can be assigned to the first overtone of the –CH3 asymmetric stretching and the –CH3 symmetric stretching vibrations, respectively. The peak located at 1736 nm that is linked to the –CH2 vibrations (2) could be assigned to the first overtone of the –CH2 asymmetric stretching or to the combination of both asymmetric and symmetric stretching vibrations.The feature at 2144 nm is clearly related to 3008 cm21 (+) since it is a combination of this wavenumber and the CNC stretching vibration at 1660 cm21. Finally, features at 2304, 2336 and 2384 nm are significantly linked to the –CH3 and NCH stretching vibrations (+). They are assigned to combinations of the methyl stretching and scissoring vibrations. All probable frequency assignments between the MIR and the NIR domains are summarized in Table 2. It can be assumed that the relatively narrow spectral ranges of both of these infrared techniques prevent the observation of some important features. For instance, no measurements are available between 1500 and 1700 cm21 to assign CNC stretching, or below 1660 cm21 for elongations.Furthermore, the resolution in the NIR domain is low (8 nm), making some assignments difficult. A more extensive study will be undertaken using a wider range of MIR frequencies in order to confirm the role of the CNC vibrations and C–H deformations in producing the NIR bands.The comparison of these results with those obtained by the CCA procedure1 and by analysis of variance2 shows that this new method is well adapted to study the different relationships between the two spectral domains. To test the influence of the categorical coding step, the FCR procedure was applied to the NIR and MIR spectrum sets separately before and after coding (Fig. 9). The t-values obtained on both coded spectra are similar to those obtained on the original spectra.The coding step highlights the wavelengths and wavenumbers most characteristic of the unsaturation level. Moreover, results obtained with the data set X coded into five and seven classes showed no real differences from those obtained on a three-class coded data set. The higher rank order statistics gave joint plots with more distinct element clusters, but the nature of the MIR–NIR interactions significantly related to the level of unsaturation remained unchanged.Conclusions The proposed method has been shown to be a useful tool for the study of the relationships between two sets of MIR and NIR spectra. The novel association of a regression technique with factorial correspondence analysis has given more precise results than the simple method based on the back-projection of the ROI in the FCA joint plots. The main advantage of the outer product method is to give results that may be interpreted in the form of pictures or maps.This outer product technique may be generalized and applied when any two multivariate measurements are obtained on the same sample. The assignments of wavelengths in the NIR domain to wavenumbers in the MIR domain are facilitated by the visual interpretation of the results. Applying FCR to the MIR and NIR spectra separately led to some peaks being selected in each spectral domain (absolute values higher than t2) but no direct correspondences could be made between the MIR and NIR spectral ranges.A statistical interpretation is also possible as the resulting maps reveal the level of significance of the relationships between the domains. The advantage of the FCR method over other techniques is to underline spectral variations due only to the response vector used for the regression, other variations being neglected. Nevertheless, an extension of this technique to data sets with more spectra, more samples and larger spectral ranges may be necessary in order to improve the discrimination between oil samples, and therefore to improve the assignment precision between the two domains.Table 2 Spectral assignments detected by factorial correspondence regression. Probable peak assignments are given in parentheses using the following notations: ovt., overtone; d, scissoring; n, stretching; sym., symmetric; asym., asymmetric (2): Fig. 7 NIR wavelength/nm MIR wavenumber/cm21 (+): Fig. 8 1640 2914 (n CH2 asym.) 2 (n C–H); (1st ovt. n CNH cis) 2850 (n CH2 sym.) 2 1696 3008 (n CNC–H cis) + (1st ovt. n CH3 asym.) 2960 (n CH3 asym.) + 2942 + 2930 + 2856 + 1736 2996 2 (1st ovt. n CH2 asym.) 2914 (n CH2 asym.) 2 (n CH2 asym. + n CH2 sym.) 2850 (n CH2 sym.) 2 1760 3008 (n CNC–H cis) + (1st ovt. n CH3 sym.) 2930 + 1816 3008 (n CNC–H cis) + (2 3 n CNC–H cis + d CH2) 2892 2 2850 (n CH2 sym.) 2 1856 3008 (n CNC–H cis) + 2996 2 2978 2 2850–2796 2 2144 3008 (n CNC–H cis) + (n CNC–H cis + n CNC) 2304 3008 (n CNC–H cis) + (n CH3 asym.+ d CH3 asym.) 2960 (n CH3 asym.) + 2942 + 2868 (n CH3 sym.) + 2336 3008 (n CNC–H cis) + (n CH3 asym. + d CH3 sym.); 2960 (n CH3 asym.) + (n CH3 sym. + d CH3 asym.) 2942 + 2868 (n CH3 sym.) + 2384 3008 (n CNC–H cis) + (n CH3 sym. + d CH3 sym.) 2960 (n CH3 asym.) + 2942 + 2868 (n CH3 sym.) + Fig. 9 FCR applied separately to the NIR and MIR spectrum sets before coding (dotted lines) and after coding (solid lines). Only peaks with absolute values higher than t2 are marked.Analyst, 1998, 123, 1783–1790 1789Finally, FCA can be used as an orientation tool (to gain some general knowledge about the relationships between rows and columns of a data table), as a direct synthesis tool (the calculated factors are used as new variables to describe the data) or/and as a multi-stage synthesis tool (the factor characteristics are used for further multivariate statistical analysis, such as multivariate regression). Multivariate image analysis (MIA) is becoming more widely used in analytical chemistry with the advent of imaging techniques to produce image sequences and characterize dynamic processes.Several recent literature reviews on imaging techniques for analytical chemistry describe these new developments. 19 A common characteristic of all these techniques is that they produce large amounts of complex data, which require processing in order to be interpreted. For this reason, it is necessary to develop new processing techniques that can transform a simple data set into information useful for analytical purposes. References 1 Devaux, M. F., Robert, P., Qannari, A., Safar, M., and Vigneau, E., Appl. Spectrosc., 1993, 47, 1024. 2 Barros, A. S., Safar, M., Devaux, M. F., Robert, P., Bertrand, D., and Rutledge, D. N., Appl. Spectrosc., 1997, 51, 21. 3 Rutledge, D. N., Barros, A. S., and Gaudard, F., Magn. Reson. Chem., 1997, 35, 13. 4 Geladi, P., and Grahn, H., Multivariate Image Analysis, Wiley, Chichester, 1996. 5 Geladi, P., Isaksson, H., Lindqvist, L., Wold, S., and Esbensen, K. H., Chemom. Intell. Lab. Syst., 1991, 2, 129. 6 Grahn, H., and Geladi, P., in Signal Treatment and Signal Analysis in NMR, ed. Rutledge, D. N., Elsevier, Amsterdam, 1996, pp. 513– 534. 7 Trebbia, P., and Bonnet, N., Ultramicroscopy, 1990, 34, 165. 8 Henrion, R., Chemom. Intell. Lab. Syst., 1994, 25, 1. 9 MATLAB for Windows Version 4.2.c, MathWorks, Natick, MA, 1994. 10 Lefebvre, J., Introduction aux Analyses Statistiques Multidimensionnelles, Masson, Paris, 1980. 11 Mellinger, M., Chemom. Intell. Lab. Syst., 1987, 2, 61. 12 Bugner, E., and Feinberg, M. H., Chemom. Intell. Lab. Syst., 1991, 12, 21. 13 Basilevsky, A., Statistical Factor Analysis and Related Methods. Theory and Applications, Wiley, New York, 1993. 14 Höskuldsson, A., Chemom. Intell. Lab. Syst., 1996, 32, 37. 15 Kowalski, K.G., Chemom. Intell. Lab. Syst., 1990, 9, 177. 16 Jouan-Rimbaud, D., Walczak, B., Massart, D. L., Last, I. R., and Prebble, K. A., Anal. Chim. Acta, 1995, 304, 285. 17 Draper, N. R., and Smith, H., Applied Regression Analysis, Wiley, New York, 2nd edn., 1981. 18 Wu, W., Massart, D. L., and De Jong, S., Chemom. Intell. Lab. Syst., 1997, 36, 165. 19 Van Espen, P., Janssens, G., Vanhoolst, W., and Geladi, P., Analusis, 1992, 20, 81. Paper 8/01394B 1790 Analyst, 1998, 123, 1783–1790

ISSN:0003-2654    

DOI:10.1039/a801394b

出版商:RSC    

年代:1998

数据来源: RSC

摘要:

Evaluation of sampling mass for well mixed bulk materials with consideration of analytical variance Zhi Gao,a Xiwen He,*a Guizhu Zhang,a Yuhuan Lib and Xiaoyu Wub a Department of Chemistry, Nankai University, Tianjin, 300071, China b Tianjin Geological Academy, Tianjin, 300181, China Received 5th March 1998, Accepted 24th June 1998 An equation was derived for the evaluation of the critical sampling mass (CSM) for well mixed bulk materials. The CSM is the sampling mass at the critical situation that the total variance is not significantly different from the analytical variance, which is tested by the F-test method.The CSM, which is connected with the sampling constant, the content of the component to be determined and the analytical variance, was found to be very useful in estimating the representativity of sampling. The equation was applied to sampling for chemical analysis of silicon carbide with satisfactory results. It is applicable to any well mixed bulk material.Sampling is an important aspect of analytical chemistry which is attracting increasing attention.123 One of the most important facets of sampling is to establish the quantitative relationship between the sampling error and sampling properties, such as the sample mass and the number of increments.4–7 In order to reflect correctly the properties of a population by the analytical results, the representativity of the sample is the most important aspect to be considered.However, in investigating the sampling and its representativity, workers usually concentrate on the sampling step only, without considering the analytical method, which results in some limitations in practice. The sampling representativity is tested by the final analytical results. Because chemical analysis is an integrated procedure, it should be related to every step of the overall analytical procedure. In addition, we also need to specify the degree of representativity of the sample in practical sampling operations.The precision of the final result is expressed by the total variance. Since the total variance is the sum of the sampling variance and analytical variance,8 it is obvious that the analytical variance is the limit of the total variance that can be reached if the analytical method (including the sample pulverization, mixing and measurement) is specified. However, the total variance will never be equal to the analytical variance algebraically. It can be considered to be not significantly different from the analytical variance.In this instance, the sampling variance is ignored and the corresponding sampling mass is considered to be representative of the population. In this work, the F-test variance ratio statistical method was used to compare the total variance with the analytical variance. An equation for the critical sampling mass (CSM) is presented for well mixed bulk material at the critical situation that the total variance is not significantly different from the analytical variance.It can be concluded that the sample is statistically representative of the population if the sampling mass is greater than the CSM. Otherwise, it is not representative of the population statistically. The CSM is connected with the sampling constant, the analytical variance and the concentration of the component to be determined. The equation was applied successfully to sampling for the chemical analysis of silicon carbide material.It is applicable to the sampling of any well mixed bulk material. Theoretical Total variance Consider n samples taken from a population. Each sample is analyzed once for the content of the component of interest. For well mixed bulk material, the whole process of analysis can be divided into two steps, sampling from the population and the subsequent chemical analysis. The total error of sample i is the sum of the errors of each step: Eti = Esi + Eai (1) where Eti is the total error, Esi is the sampling error and Eai is the analytical error.We refer the errors here only to the random errors without considering the systematic errors. Squaring and expanding for all samples, we have E E E EE i i i ii i n i n i n i n t s a sa 2 2 2 1 1 1 1 2 = + + = = = = Â Â Â Â (2) Since the sampling and the chemical analysis are independent, 2 0 1 E Ei i i n s a = = Â (3) Eqn. (2) becomes E E E i i n i i n i i n t s a 2 1 2 1 2 1 = + = = = Â Â Â (4) For the variances, we have s s n E s i i n t a 2 2 2 1 1 1 = + - = Â (5) where st 2 is the total variance and the ss 2 is the sampling variance.Analytical variance For simplicity and practical application, the second term of the right-hand side of eqn. (5) that is related only to the analytical errors need to be specified. Since Analyst, 1998, 123, 1791–1794 1791Eai = wi 2 � wi (6) where wi is the determined concentration of sample i, and � wi is the true concentration of sample i.Although Eai is related to sample i other than the population, we can conclude that [1/(n21)] equals sa 2, the analytical variance of E i i n a 2 1 = Â the population as a first order of approximation as the following. As stated in the literature,9,10 the analytical standard deviation has a linear relationship with concentration: sa = s0 + kc (7) where sa and s0 are the analytical standard deviations at concentrations c and zero, respectively, and k is a constant.Logically, we can infer that the single analytical deviation Eai at concentration � wi is related to the single analytical deviation Eai(0) at concentration � w by a linear function: Eai = Eai(0) + x( � wi 2 � w) (8) where x is a constant. x( � wi 2 � w) is much smaller than Eai(0). Squaring and expanding for all samples, we have E E E w w w w i i n i i i i n i n i i n a a a 2 1 2 1 1 2 2 1 0 2 0 = + - + - = = = = Â Â Â Â ( ) ( )( ) ( ) x x (9) For well mixed bulk materials, Eai(0) and � wi 2 � w are random errors, so 2 0 0 1 x E w w i i i n a ( )( ) - = = Â (10) Ignoring x2 ( � wi 2 � w)2 for a first order of approximation, we have 1 1 1 1 0 2 1 2 2 1 n E n E s i i n i i n - = - = = = Â Â a a a ( ) (11) Subsequently, eqn.(5) becomes st 2 = ss 2 + sa 2 (12) Comparison of st 2 and sa 2 As stated above, both st 2 and sa 2 are variances of the population. It is valid to compare st 2 with sa 2. Since the total variance will never equal the analytical variance algebraically, but can be treated as being not significantly different from the analytical variance, at this critical instance, the sampling variance is negligible.This is an F-test problem. The following null hypothesis is tested: H0 : st 2 = sa 2. An alternative hypothesis states that H1 : st 2 > sa 2. The F-value is calculated: F = st 2/sa 2 (13) What we compare here is the calculated F-ratio with the critical value F12a obtained from statistical tables at a confidence level of 1 2 a.If F@F12a, there is no significant difference between st 2 and sa 2, so ss 2 can be ignored. If F > F12a, si2 is significantly greater than sa 2, and ss 2 is also an important contributor to st 2. For F12a = st 2/sa 2, eqn. (12) is rearranged to obtain the critical ss 2: ssc 2 = (F12a 2 1)sa 2 (14) The sampling mass for ssc 2 is defined as the critical sampling mass (CSM). Critical sampling mass For well mixed bulk materials, the relative sampling standard deviation (R) is related to the sampling mass by the equation4 msR2 = Ks (15) where ms is the sample mass and Ks, is the sampling constant, which is defined as the mass of the sample to be taken to give a sampling relative standard deviation of 1% at the 68% confidence level.Since R = 100ssw21 (16) where ss is the sampling standard deviation and w is the concentration of the component to be determined, eqn. (15) is substituted to obtain the sampling variance ss 2: ss 2 = Ksw2 3 1024 ms21 (17) Eqn.(14) is substituted with eqn. (17) to obtain the critical sampling mass: msc = Ksw2sa22 (F12a 2 1)21 3 1024 (18) When the sampling constant, the analytical variance and the concentration of the component are known, the CSM can be calculated. If the sampling mass is greater than the CSM, the total variance is not srom the analytical variance at confidence level of 1 2a, and the sampling variance can be ignored even though it still contributes to the total variance in this case. That is, the total variance is controlled at the same level of analytical variance. Hence the CSM is regarded as the minimum sampling mass representative of the population.Experimental Silicon carbide bulk material (low content of SiC, size < 3 mm) which had been well mixed was chosen as the population. Increments with different masses were withdrawn randomly from the population. Subsequently, these increments were dried at 105 °C and pulverized until all of them passed a sieve with a mesh size of 0.147 mm (mesh number 100) to make samples for chemical analysis.The loss on ignition (LOI) was determined by the gravimetric method after ignition at 750–800 °C.11 The residue of the ignited sample was decomposed with HF–HNO3– H2SO4 followed by fusion with potassium pyrosulfate (K2S2O7) and dissolution with HCl. The precipitate was filtered and ignited at 750–800 °C to determine the content of SiC gravimetrically.11 The filtrate was used to determine Fe2O3 by spectrophotometry with sulfosalicylic acid and to determine CaO, MgO and Cu by flame atomic absorption spectrometry.The sample was decomposed with HF–HNO3–H2SO4 after ignition at 750–800 °C. Subsequently, the residue was dissolved in HCl to determine Na2O and K2O by flame photometry. Results and discussion Analytical variance Ten samples of mass 300 g each were taken randomly from the population and mixed thoroughly to make a composite sample of the population.The samples were ground until they all passed a mesh size of 1 mm. Ten sub-samples of mass 100 g each were taken randomly from the composite sample and mixed thoroughly. The following procedure was the same as in the Experimental section to make a sample of the population. Ten parallel portions of the same laboratory sample were analyzed to determine the concentrations of eight components.The statistical results are given in Table 1. As can be seen, the analytical precision varies with the components. The precisions for the high content components SiC and LOI are better than 1792 Analyst, 1998, 123, 1791–1794those for the lower content components. The precision for micro-amounts of Cu is the worst. Estimation of sampling constants A set of 10 parallel increments for a certain sampling mass were withdrawn from the population of silicon carbide material. Five different sampling masses were tested.The total variance for each component was calculated. Subtraction of the analytical variance from the total variance gave the sampling variance. The relationship between the sampling variance and the sampling mass for each component is illustrated in Figs. 1–3. As can be seen, the sampling variance decreases with increase in the sample mass for each component. These curves can be fitted as hyperbolas very well with the function in eqn.(17). The fitted results and the average sampling constants are given in Table 2. The sampling constants are different for each of the components, although they are in the same population. This indicates that the micro-homogeneity is different for each component in the material. Therefore, the required sampling mass is different for each component in order to meet a fixed sampling error. Evaluation of the critical sampling mass The CSM was calculated for each component according to eqn.(18). The results are given in Table 3. As can been seen, the CSM is different for each of the components at the same confidence level. This means the sample mass should be more than the greatest CSM in order to meet the requirement of all components in the same population. For the same component, the CSM decreases with increase in the required confidence level. It is obvious that if the overall precision is permitted to be low, only a small sample mass need be taken from the population.By comparing the CSM with the sampling constant, we also find that the CSM is not related to the sampling constant in a simple way. As shown in Tables 2 and 3, the main components SiC and LOI have smaller sampling constants than the others, but their CSMs are greater because of their good analytical precision. Copper, with a larger sampling constant, has a smaller CSM because of its poor analytical precision. Conclusion A new equation for calculating the CSM for well mixed bulk materials has been derived. The equation established the relationship of the sampling mass with the sampling constant, the analytical variance and the content of the component to be Table 1 Statistical results of the chemical analysis of silicon carbide (degrees of freedom f = 9) Component Analytical variance Content (% m/m) RSD (%) SiC 1.3 3 1025 34.84 1.03 LOI 1.7 3 1025 29.38 1.40 Fe2O3 6.7 3 1028 1.08 2.40 CaO 8.3 3 1028 0.96 3.00 MgO 9.0 3 1029 0.30 3.16 Na2O 6.7 3 1029 0.29 2.82 K2O 1.3 3 1028 0.39 2.92 Cu 4.1 3 10212 0.0031 6.53 Fig. 1 Relationship between the sampling variance (ss 2) and the sampling mass (ms) for a, SiC (4) and b, LOI (~). Fig. 2 Relationship between the sampling variance (ss 2) and the sampling mass (ms) for c, CaO (8), d, Fe2O3 (-), e, MgO (½), f, K2O (3) and g, Na2O (+). Fig. 3 Relationship between the sampling variance (ss 2) and the sampling mass (ms) for h, Cu (2). Table 2 Results fitted with eqn. (17) for curves in Figs. 1–3 and the sampling constants Component Coefficient, Ksw2 3 1024/g Sampling constant, Ks/g SiC 4.81 3 1023 396 LOI 2.88 3 1023 341 Fe2O3 7.95 3 1026 682 CaO 8.68 3 1026 941 MgO 3.88 3 1026 4311 Na2O 7.09 3 1027 843 K2O 4.85 3 1027 3190 Cu 1.21 3 10210 1259 Table 3 Critical sampling mass for components in silicon carbide bulk material at different confidence levels. The degrees of freedom of the total variance and the analytical variance are both 9 CSM at different confidence levels 1 2 a = 99% 95% 90% 75% Component F1 2 a = 5.35 3.18 2.44 1.59 SiC 85 170 257 627 LOI 39 78 117 286 Fe2O3 27 55 83 202 CaO 24 48 73 178 MgO 99 198 299 731 Na2O 24 49 74 180 K2O 8.5 17 26 63 Cu 6.9 14 21 51 Analyst, 1998, 123, 1791–1794 1793determined, and proved to be effective in elucidating the sampling representativity of the population.The equation was used to investigate the sampling of silicon carbide bulk material successfully. The classical and robust variance analysis usually uses duplicate determinations for each sample to study the analytical variance.12,13 However, it only provides the mean square deviation of all samples as an estimate of the analytical variance contributing to the total variance.In this paper, the analytical variance based on a single determination for each sample was studied. The relationship of the total variance with the sampling variance and the analytical variance of the population was established, which set the basis of the variance comparison.From the view of fitness-for-purpose,14,15 any sample is representative of the population to some extent as long as it meet the purpose. However, this requirement of the total variance is usually arbitrary. The CSM presented here is an objective requirement of sampling at the critical situation that the total variance is not significantly different from the analytical variance at a given confidence level. It also provide a statistical method for dealing with the representativity of the sample taken from the population.With regard to statistics, if the sample mass is greater than the CSM, the total variance is not significantly different from the analytical variance, which is determined mainly by the analytical method. Therefore, improving the precision of the analytical method will effectively decrease the error in the final result. However, if the sampling mass is less than the CSM, the total variance is determined by both the sampling and analytical variance.The sampling error cannot be neglected. The CSM is different for each of the component to be determined in the same material. Hence the sampling operation can be made more economical with the knowledge of the CSM of the component, especially for valuable materials. The CSM, different from the sampling constant, which reflects only the sampling properties of the population, is related to the sampling constant, the concentration of the component and the precision of chemical analysis.It effectively reflects the representativity of the sample and the precision of the final result to be reached in practice. The CSM makes the sources of the total error clear. When the sampling and the chemical analysis are carried out by different workers, the CSM is helpful in specifying the responsibility for each worker, especially when the final result is doubted or even wrong. The equation for the CSM is applicable to any well mixed bulk material. We acknowledge the financial support of the Chinese Natural Science Foundation and the Doctoral Program Foundation of the China Ministry of Education. References 1 Gy, P. M., Analusis, 1995, 23, 497. 2 He, X.-W., and Guo, W., Chin. J. Anal. Chem., 1995, 23, 1456. 3 Kratochvil, B., Wallace, D., and Taylor, J. K., Anal. Chem., 1984, 56, 113R. 4 Ingamells, C. O., and Switzer, P., Talanta, 1973, 20, 547. 5 Visman, J., Mater. Res. Stand., 1969, 11, 8. 6 Wallace, D., and Kratochvil, B., Anal. Chem., 1987, 59, 226. 7 Kratochvil, B., and Taylor, J. K., Anal. Chem., 1981, 53, 924A. 8 Youden, W. J., Statistical Methods for Chemists, Krieger, New York, 1977. 9 Thompson, M., Analyst, 1988, 113, 1579. 10 Thompson, M., and Howarth, R. J., Analyst, 1976, 101, 690. 11 American National Standard, Standard Methods of Chemical Analysis of Silicon Carbide Abrasive Grain and Abrasive Crude, ANSI B74.15, 1971. 12 Ramsey, M. H., Thompson, M., and Hale, M., J. Geochem. Explor., 1992, 44, 23. 13 Thompson, M., and Maguire, M., Analyst, 1993, 118, 1107. 14 Thompson, M., and Ramsey, M. H., Analyst, 1995, 120, 261. 15 Thompson, M., and Fearn, T., Analyst, 1996, 121, 275. Paper 8/01822G 1794 Analyst, 1998, 123, 1791–1794

ISSN:0003-2654    

DOI:10.1039/a801822g

出版商:RSC    

年代:1998

数据来源: RSC

摘要:

Comparison of various adsorbents for long-term diffusive sampling of volatile organic compounds Naciye Kilic*a and James A. Ballantineb a University of Uludag, Art and Science Faculty, Chemistry Department, 16059 Bursa, Turkey b Monitoring Unit, Chemistry Department, University of Wales, Swansea, UK SA2 8PP Received 16th February 1998, Accepted 7th July 1998 The relative performances of diffusive samplers packed with Tenax TA, Chromosorb 106 and Carbotrap for the long-term diffusive sampling of several typical volatile organic compounds were compared.In order to measure the magnitudes of the decreases in mass uptakes in a real long-term diffusive sampling situation, the results for 7 and 14 d sampling periods were compared with the sum of consecutive 1 d diffusive sampling results, performed in parallel for the same period of time. The results of this study gives information about the magnitudes of the decreases in mass uptakes in a real sampling situation and show the limitations of the adsorbents tested for the sampling of volatile organic compounds from different classes such as acetone and dichloromethane as compounds with a functional group, toluene as an example of aromatic compounds and pentane, hexane, nonane, decane and undecane as examples of straight chain hydrocarbons (C5–C11).It was shown that the data obtained from long-term diffusive sampling are subject to significant error, particularly for low boiling compounds.Chromosorb 106 could be the adsorbent of choice for this sampling situation, as it gives the highest percentage of masses (ng) collected for longer exposure periods relative to the sum of the masses (ng) collected over each exposure period among the other adsorbents tested. Diffusive sampling is an accepted method for the determination of time-weighted average concentrations of volatile organic compound vapours in both workplace and environmental atmospheres.1,2 Diffusive samplers were initially developed for the measurement of occupational exposure to volatile organic compounds.Therefore, uptake rates for some adsorbate– adsorbent pairs have been determined experimentally for an 8 h exposure time and at constant concentration.3 However, the uptake rate of a diffusive sampler changes with concentration and exposure time.4 There have been long-term monitoring studies using diffusive samplers and exposure periods longer than 8 h.5,6 The investigation of the performance of diffusive samplers over a 7 month period collecting volatile organic compounds showed that there is a correlation between the amount of loss from back-diffusion and the volatility of the compound.7 It has been observed in practice that uptake rates decrease with increasing time and concentration. This effect was explained as a result of the adsorption characteristics of porous polymers.Adsorbents such as Tenax and Porapak have approximately linear adsorption isotherms, which means a vapor pressure of adsorbate will exist at the surface of the adsorbent and reduce the concentration gradient of the adsorbate, hence the uptake rate will be decreased.Obviously, more adsorbed vapor will also result in a more reduced uptake rate. An experimental procedure presented by van den Hoed and Halmans8 for the selection of suitable adsorbate–adsorbent pairs showed that uptake rates can easily be calculated from the physical dimensions of the samplers and the diffusion coefficient of the compound, provided that the adsorbent inside the samplers has been correctly matched to that compound.They suggested that for non-optimum conditions, adsorption isotherms have to be taken into account. For non-ideal adsorbate–adsorbent pairs, the effective uptake rates would be different from the calculated ideal uptake rates and need to be known for different sampling conditions such as concentration, adsorbent type and exposure time in order to be able to monitor a wide range of compounds using diffusive samplers with reasonably quantitative results.Experimentally measured effective uptake rates are available for different concentrations and maximum exposure times up to 8 h.3 Experimental Sampling strategy Sampling was performed by placing 18 Perkin-Elmer (Norwalk, CT, USA) stainless-steel diffusion tubes in an organic research laboratory in which considerable levels of volatile organic compounds could be found in the atmosphere.This sampling strategy represents a real-life situation in which a workplace with varying amounts of a variety of volatile organic compounds is being monitored for both short and long periods by diffusive sampling. Six tubes were packed with Carbotrap graphitized carbon black, 20–40 mesh, 100 m2 g21 surface area, six with Chromosorb 106, 60–80 mesh, 750 m2 g21 surface area, and six with Tenax TA, 60–80 mesh, 35 m2 g21 surface area.The adsorbent bed length was 50 mm for each adsorbent. Two tubes fitted with a Swagelock end-cap with a PTFE ferrule at one end and a diffusion end-cap with a silicone membrane at the other were used for each 1, 7 and 14 d sampling period. Sampling was started by replacing the three 1 d tubes packed with different adsorbents (a total of six in two sampling sets) with freshly conditioned tubes after exactly 1 d. The same procedure was repeated with 1 d intervals for 14 d. The tubes left for the 7 d exposure period were replaced with freshly conditioned tubes on the seventh day of this period.The tubes were analyzed on the same day of sampling. On the 14th day of the experiment, tubes exposed for 14 d, together with the tubes exposed for the second 7 d of this sampling period and the tubes for the 1-day sampling period, were removed and analyzed. The levels of the analytes collected were calculated in terms of adsorbed masses (ng). Analyst, 1998, 123, 1795–1797 1795Instrumentation and analytical parameters A Perkin-Elmer Model ATD 400 automatic thermal desorption injector system and a Hewlett-Packard (Avondale, PA, USA) Model 5890 II gas chromatograph fitted with a flame ionization detector were used for the subsequent analysis of the collected samples.The results were plotted on a Hewlett-Packard HP 3396A printer/plotter integrator and the data were stored in an Elonex 286 M-120 PC using Hewlett-Packard Peak 96 software.ATD 400 automatic thermal desorber parameters The following conditions were used: desorption time, 10 min, desorption temperature, 250 °C (for Tenax TA and Chromosorb 106) and 320 °C (for Carbotrap); trap hold time, 5 min, trap high temperature, 250 °C; trap low temperature, 230 °C; trap packing, Tenax TA; and splitting factor, 5.6 (17.21% to GC). GC conditions A 25 m 3 0.3 mm id 1.0 WCOT capillary column with a mm film thickness of polydimethylsiloxane was used with a flow rate of the carrier gas (helium) of 1.8 ml min21.The column temperature was held for 2 min at 40 °C, then increased at 10 °C min21 to 200 °C, which was held for 5 min. Identification of analytes and calibration Individual analytes in the samples were identified by their mass spectrometric fragmentation patterns obtained by GC–MS using a VG Masslab (Winsford, Cheshire, UK) Model 12-223 quadropole mass spectrometer, fitted with a Chrompack (Middelburg, The Netherlands) thermal desorption cold trap injector and library search facility.A four-level external calibration of the analytes of interest were carried out by spiking adsorption tubes in a flowing stream of nitrogen with a multicomponent standard mixture made up in methanol. The solvent (methanol) was purged with nitrogen flow to avoid overloading the cold trap and column. Since methanol has a low retention volume on Tenax TA, most of it broke through whereas the other compounds were adsorbed.The possible breakthrough of the compounds during nitrogen purging was checked and the optimum flow rate and volume were used. Nitrogen was purged through the adsorbent tube at 33 ml min21 rate for 5 min to ensure adsorption of the compounds of interest and removal of methanol. Results The masses (ng) obtained for the 7 and 14 d periods were compared with the sums of the individual 1 d results taken each day during the 7 and 14 d periods. In this way, the losses associated with the long-term diffusion sampling could be quantified for a real-time monitoring situation where the levels of the volatile organic compounds varied on a day to day basis.The results for each analyte on each adsorbent tube, expressed as the percentage collected, are given in Table 1. Low percentages are due to a decrease in uptake rates with time and loss of adsorbed compounds by back-diffusion and simply show that the adsorbent is not suitable for the collection of these compounds.The results revealed that the effect of adsorbent type on losses of compounds during long exposure (7 and 14 d) are significant for acetone and dichloromethane. The amounts of straight-chain hydrocarbons (pentane, hexane, nonane, and undecane) collected on all three adsorbents increase as the volatility of the compound decreases. Carbotrap gave the highest percentage values for pentane, hexane and toluene among the three adsorbents for each exposure period.Although the results obtained for nonane, decane and undecane on Carbotrap were slightly lower relative to those obtained for the other two adsorbents, it is clear that this decrease is due to incomplete desorption of the high boiling compounds which are held Table 1 Influence of adsorbent type on mass collected during long-term diffusive sampling expressed as % of summed 1 d sampling Compound Exposure time/d Adsorbent Acetone Dichloromethane Pentane Hexane 7 (1) Carbotrap 23 9 64 88 Chromosorb 106 54 52 56 73 Tenax TA 48 38 51 70 7 (2) Carbotrap 17 11 60 77 Chromosorb 106 47 49 50 71 Tenax TA 36 37 40 67 14 Carbotrap 6 4 43 67 Chromosorb 106 32 37 40 57 Tenax TA 19 20 33 56 Compound Toluene Nonane Decane Undecane 7 (1) Carbotrap 101 96 93 77 Chromosorb 106 104 115 113 106 Tenax TA 95 103 101 104 7 (2) Carbotrap 92 85 91 71 Chromosorb 106 95 103 90 79 Tenax TA 82 97 96 98 14 Carbotrap 93 89 86 74 Chromosorb 106 89 100 94 94 Tenax TA 75 92 89 94 Fig. 1 Percentage of the compounds collected on each adsorbent over 14 d. 1796 Analyst, 1998, 123, 1795–1797strongly on the Carbotrap adsorbent. Incomplete desorption was also seen on Chromosorb 106 for decane and undecane. Fig. 1 shows the percentage of compounds collected during the14 d sampling period. The percentage of the compounds with functional groups collected on each adsorbent are significantly different for all exposure periods. The analytical data obtained from this study also give information about the fluctuations in the levels of organic compounds in an organic research laboratory air over a 14 d period.Fig. 2 illustrates the variations of four of the analytes taken from the 1 d diffusion tubes which were analyzed on each day of the 14 d period. It can be seen in this real-life situation that there were considerable variations in the masses collected for some of the analytes, particularly pentane and toluene. Conclusions The results of this study give information about the magnitudes of the decreases in mass uptakes in a real sampling situation for several typical volatile organic compounds on three different adsorbents, Carbotrap, Chromosorb 106 and Tenax TA.The effect of adsorbent type and exposure period on the mass uptakes can be seen from the percentages of the compounds collected given in Table 1. It is clear that the uptake rates decrease with increasing exposure time and more on Tenax TA than on Chromosorb 106.The loss in mass collected during long-term diffusive sampling is due to the decrease in uptake rates and back-diffusion of the compounds adsorbed. This study showed that the data obtained from long-term diffusive sampling are subject to significant error, particularly for low boiling compounds. Acetone and dichloromethane were not retained on Carbotrap and therefore showed the lowest percentage masses collected relative to the other two adsorbents. This result was confirmed by the adsorbtion characteristics of the Carbotrap adsorbent, which, unlike Tenax TA and Chromosorb 106, has no localized charges for interaction with the functional groups.Hence the effect of adsorbent type on the decreases in uptake rates during long-term diffusive sampling has been confirmed. This work was performed in real sampling situations and shows the limitations of the adsorbents tested for volatile organic compounds from different classes, namely acetone, dichloromethane, toluene, pentane, hexane, nonane, decane and undecane.The results showed that Chromosorb 106 could be the adsorbent of choice for this sampling situation, as it gave the highest percentages of the masses (ng) collected for longer exposure periods relative to the sum of the masses (ng) collected on each individual 1 d exposure period during corresponding longer exposure periods. We did not evaluate practical diffusive uptake rates, which would be necessary to calculate actual exposure concentrations. Such uptake rates should be established from the literature cited or by using an established protocol analogous to EN838. References 1 Mixed Hydrocarbons (C5 to C10) in Air, MDHS 66, Health and Safety Executive, Occupational Medicine and Hygiene Laboratory, London, 1989. 2 Benzene in Air, MDHS 50, Health and Safety Executive, Occupational Medicine and Hygiene Laboratory, London, 1985. 3 Wright M. D., BCR Certification of Reference Materials—Organics on Tenax—A Critical Examination of Sources of Error, Health and Safety Executive, Occupational Medicine and Hygiene Laboratory, Report IR/L/IA/93/3, London, 1993. 4 Brown R. H., and Walkin, K. J., Anal. Proc., 1981, 18, 205. 5 Crump D. R., and Madany I. M., Proc. Indoor Air, 1993, 2, 15. 6 Brown V. M., Crump, D. R., and Gardiner, D., Environ. Technol., 13, 1992, 367. 7 Brown, V. M., and Crump, D. R., in Proceedings of the International Conference on Volatile Organic Compounds in the Environment, ed. Leslie, G., and Perry, R., IAI, Rothenfluh, 1993, pp. 241–249. 8 van den Hoed, N., and Halmans, M. T. H., J. Am. Ind. Hyg. Assoc., 1987, 48, 364. Paper 8/01339J Fig. 2 Variations in the levels (ng) of compounds collected during 14 d (1–14). Analyst, 1998, 123, 1795–1797 1797

ISSN:0003-2654    

DOI:10.1039/a801339j

出版商:RSC    

年代:1998

数据来源: RSC

摘要:

Rapid preconcentration method for the determination of pyrethroid insecticides in vegetable oils and butter fat and simultaneous determination by gas chromatography–electron capture detection and gas chromatography–mass spectrometry A. Ramesh* and M. Balasubramanian Department of Pesticide Chemistry, Fredrick Institute of Plant Protection and Toxicology, Padappai-601 301, Chennai, Tamil Nadu, India Received 24th April 1998, Accepted 6th July 1998 A simple and rapid solid-phase extraction (SPE)–GC method for the preconcentration and quantification of pyrethroids at low nanogram levels in oils and high fat content samples is presented.The method was studied using seven highly persistent pyrethroid insecticides, viz., cypermethrin, deltamethrin, fenvalerate, cyfluthrin, allethrin, cyhalothrin and permethrin. Preconcentration was achieved by treating the oil samples with methyltrioctylammonium chloride and subsequent elution of the pyrethroid molecules from a graphitised carbon black SPE cartridge using 5 ml of acetonitrile.Pyrethroid quantification was achieved by GC with electron capture detection. Recoveries of the pyrethroids at fortification levels of 0.05–0.5 ppm were 94–105%. Storage on graphitised carbon black for 30 d lowered the recovery of the pyrethroids by only 3–6%. The method compared well with results obtained by a GC–MS method. The relative standard deviation at a concentration level of 0.05–0.2 mg ml21 ranged from 1.31 to 5.16%.The limit of detection achieved was 0.002 mg ml21 without any additional clean-up and with little interference from lipids during analysis. Synthetic pyrethroids, an important group of insecticides widely used in many countries, have received considerable attention because of their greater photostability, enhanced insecticidal activity and relatively low toxicity when compared with organochlorine and organophosphorus insecticides.1 Many countries are invovled in research into the fate of these compounds in the environment following their application.The Food and Agricultural Organization and the World Health Organization have prescribed residue limits for some of these pyrethroids in agricultural and livestock products.2 Many methods have been reported in the literature for the determination of pyrethroid residues, either as individual compounds or simultaneously, by using various analytical techniques such as gas chromatography (GC) and high-performance liquid chromatography –mass spectrometry (HPLC–MS).3–5 Baker and Bottomley6 developed a multiresidue LC–UV method for the simultaneous determination of the pyrethroid insecticides bioresmethrin, cisresmethrin, cypermethrin, deltamethrin, fenpropathrin, fenvalerate, permethrin, phenothrin and resmethrin in fruits and vegetables.Bengston et al.7 and Haddad et al.8 determined deltamethrin, fenvalerate, permethrin and phenothrin residues in stored wheat grain.Pang and co-workers1,4,9 modified the AOAC multiresidue method for the simultaneous determination of pyrethroid residues in fruits, vegetables and grain. Immunoassay methods have also been developed for the quantification of pyrethroid molecules.10,11 The separation of trace levels of pesticides from edible oils in preparation for analysis by GC requires the use of sample preparation techniques that separate the pesticides from lipids. Even a small amount of lipids can cause deterioration of the column and contamination of the GC–electron capture detection (ECD) system.Conventional gravity flow Florisil column and mixed column12,13 methods were reported to remove residual fat from sample extracts after partial clean-up by liquid–liquid partition for the quantification of organophosphorus, organochlorine and pyrethroid compounds. In general, an acetonitrile– hexane partition step has been used to remove the lipids from oil/fat samples followed by Florisil column chromatography.However, this procedure cannot be used repeatedly, owing to the possible co-elution of pyrethroids and lipids.14 This paper describes a method that attempts to overcome these drawbacks. The method allows the simultaneous determination of seven pyrethroids in oils, lipids and high fat content samples at nanogram levels. Particular emphasis was placed on the simplification of the usually cumbersome and tedious sample preparation methods.Extraction and clean-up procedures were developed to determine the seven pyrethroids in oil and fat samples in a simple and rapid way with a limit of quantification of 2 ppb. The method was confirmed by GC– ECD and GC–MS analysis. Experimental The pesticides investigated were as follows: deltamethrin (purity 99.6%; Chem Service, West Chester, PA, USA); cypermethrin (purity 94.5%; Gharda Chemicals, Thane 421 203, India); fenvalerate (purity 99.1%; Chem Service); cyfluthrin (purity 98.7%; Bayer, Leverkusen, Germany); allethrin (purity 92.4%; Sumitomo, Osaka, Japan); cyhalothrin (purity 98.7%; Zeneca ICI Agrochemicals, Chennai 600 028, India); and permethrin (purity 94%; Gharda Chemicals).Solvents Trace analysis grade acetone and acetonitrile were purchased from Merck (Darmstadt, Germany). Apparatus A Shimadzu Model GC-14A instrument (Shimadzu, Kyoto, Japan) equipped with a 63Ni electron capture detector and a methyl silicone coated fused-silica capillary column (CBP-1, 25 m 3 0.2 mm id; film thickness, 0.25 mm) was used.Analyst, 1998, 123, 1799–1802 1799Operating conditions were: column temperature, 250 °C; injector and detector temperature, 270 °C; nitrogen (Iolar grade I, purity 99.9%) as make-up and carrier gas: flow rate, 1.2 ml min21; split ratio, 1 : 10; and injection volume, 2 ml. A Hewlett-Packard (Avondale, PA, USA) GC–MS 5970 instrument equipped with a mass selective detector and a J&W Scientific (Rancho Cordova, CA, USA) DB-5 capillary column (30 m 3 0.25 mm id; film thickness, 0.25 mm) was also used.Standard solutions Individual stock solutions of all the pyrethroids studied were made by dissolving 25 mg of each compound in 25 ml of acetone and were stored in a refrigerator. Working standards were prepared by diluting the stock solutions to obtain a final concentration of 10 mg ml21 of each compound. These standards were used for the preparation of both calibration solution and fortified samples.Solid-phase extraction Graphitised carbon black has been shown to be a valuable sorbent material for solid-phase extraction (SPE) for a variety of pollutants in water.15–18 Graphitised carbon black (500 mg) from Indo-National (Chennai 600 034, India) was loaded into a 1 cm stainless-steel/glass cartridge between two Teflon frits. The cartridge was connected to a solvent recovery flask through a vacuum pump and was conditioned by rinsing with 10 ml of acetonitrile followed by 10 ml of water and 2 ml of acetone.Preconcentration of pyrethroids on solid-phase extraction cartridge A 10 ml volume of oil sample was placed in a test-tube. Known concentrations, viz., 0.05, 0.1, 0.2 and 0.5 and 0.5 mg ml21, of pyrethroid mixtures were added to the oil samples, which were mixed thoroughly and allowed to stand for 5 min. Five drops of methyltrioctylammonium chloride (MTOAC) were added to the oil sample and each tube was shaken vigorously for 3 min.The oil sample was transferred to the SPE column reservoir and allowed to percolate for 5 min after which a vacuum was applied to drain the oil completely. The column was then eluted slowly with 5 ml of acetonitrile. The eluate was collected, filtered, evaporated under vacuum and reconstituted in acetone for quantitative analysis by GC. Results and discussion The described procedure consists in the purification of pyrethroids by SPE in place of the usual liquid–liquid partition step followed by column clean-up.Relatively large volumes of oil sample can be passed through the cartridge, as pesticides present in small amounts will be concentrated on the surface of the sorbent. No major interference from lipids was observed during the process. In addition, the columns can be used repeatedly (at least 10 times) by simply washing with 10 ml of acetone followed by 10 ml of water. The elution of oil takes place efficiently and also faster after mixing the oil with 0.5 ml of MTOAC.Fat samples were heated to around 50–60 °C and eluted through the column rapidly. The exact role of MTOAC was not studied. However, an excess of MTOAC led to a low yield in recovery studies. Absolute recoveries were determined by using external calibration. The results of the recovery studies are presented in Table 1. The method was successfully tested on different types of oil such as soybean oil, groundnut oil, sunflower oil, olive oil and butter fat.Irrespective of the nature and source of the oil, the recoveries varied by only 5%. The viscosity of the oils had no effect on the recoveries. Further, in the analysis of oil samples by GC–ECD, no impurities were encountered during the separation of the pyrethroid molecules. Effect of storage All cartridges containing pyrethroids were stored at three different temperatures, viz., 4, 20 and 30 °C, for 30 d to determine the effect of storage conditions on pyrethroid stability.Cartridge samples were stored in a temperaturecontrolled oven/refrigerator. After 30 d, the cartridges were removed from the oven/refrigerator and allowed to come to room temperature. The pyrethroids were eluted using 5 ml of acetonitrile and analysed by GC–ECD. Results are presented in Table 2. Analysis of the results showed that samples stored at 4 °C for 30 d retained the initial recovery levels. At 20 °C, allethrin and permethrin showed an average of 5% lower recoveries.All the other pyrethroids stored at this temperature showed < 3% difference from the initial recovery. Allethrin and permethrin samples stored at 30 °C showed 10% lower recoveries, whereas the other samples exhibited a loss of only 5%. This is probably due to the greater susceptibility of the ester Table 1 Recovery of pyrethroids in groundnut, soybean, sunflower oil, butter fat and olive oil Groundnut oil Soybean oil Sunflower oil Average Average Average Pyrethroid Added (ppm) recovery (%) RSD* (%) recovery (%) RSD* (%) recovery (%) RSD* (%) Allethrin 0.05 96.33 1.92 96.83 1.51 95.27 2.82 Cyhalothrin 0.1 95.87 2.85 96.2 1.51 95.9 2.91 Permethrin 0.1 96.03 1.42 95.5 3.30 96.57 1.76 Cyfluthrin 0.05 97.00 2.48 97.93 1.76 95.617 1.38 Cypermethrin 0.2 103.13 4.85 104.93 3.14 102.3 3.91 Fenvalerate 0.2 96.73 1.97 96.67 1.51 96.1 1.50 Deltamethrin 0.05 105.03 1.53 104.33 2.73 104.6 2.10 Butter fat Olive oil Average Average Pyrethroid Added (ppm) recovery (%) RSD* (%) recovery (%) RSD* (%) Allethrin 0.2 95.37 2.17 96.4 1.82 Cyhalothrin 0.5 96.33 1.92 97.4 1.81 Permethrin 0.2 93.73 4.29 95.4 1.95 Cyfluthrin 0.5 94.93 1.34 97.5 5.16 Cypermethrin 0.2 104.9 1.73 103.0 2.00 Fenvalerate 0.2 96.53 1.93 94.2 1.90 Deltamethrin 0.15 95.5 2.17 94.5 2.11 * Average of six replicate analyses. 1800 Analyst, 1998, 123, 1799–1802linkage in permethrin to hydrolysis. In addition, allethrin, which was the first direct chrysanthemic acid derivative to be synthesized, is less stable than compounds with a CH3- or halovinyl backbone as illustrated by cypermethrin (dichloro-) or tefluthrin (chlorotrifluoro-).Gas chromatography–mass spectrometry Samples obtained after SPE were analysed by GC–MS. Compounds were identified by matching their retention times and characteristic ions with those of standards. Identification was based on two criteria. Co-elution of characteristic ions must be within ±0.02 min in terms of retention time and the relative abundance of the selected masses must be within 20% of those in the mass spectrum of the standard. A representative chromatogram is illustrated in Fig. 1. Conclusion It was found that MTOAC plays an important role in the preconcentration of pyrethroid molecules from oil/high fat content samples. The role of graphitised carbon black as a SPE material for preconcentration of pyrethroids was also established. The proposed method is suitable for the determination of trace amounts of pyrethroids in oil samples.The method is rapid, accurate and inexpensive, and can also be applied to different oils. As there is no simple and specific method for the preconcentration of pyrethroid molecules from oil and fat samples, the proposed method has significant advantages over other techniques. The actual role of MTOAC in the separation technique has yet to be established. Similar results were observed for some organophosphorus and organochlorine pesticides when tested under the same conditions.Further studies are currently in progress and the results will be published elsewhere. The authors thank the management, Director and colleagues, FIPPAT, for their co-operation in conducting this study. References 1 Pang, G. F., Fan, C. L., Chao, Y. Z., and Zhao, T. S., JAOAC Int., 1994, 77, 738. 2 Guide to Codex Recommendations Concerning Pesticide Residues, Food and Agriculture Organization, World Health Organization, Rome, 1985. 3 Miyamoto, J., Pure Appl. Chem., 1981, 53, 1967. 4 Pang, G. F., Chao, Y. Z., Shanliu, X., and Fan, C. L., JAOAC Int., 1995, 78, 1474. 5 Tsumura, Y., Wada, I., Fjiwara, Y., Nakamura, Y., Tonogai, Y., and Ito, Y., J. Agric. Food Chem., 1994, 42, 2922. 6 Baker, P. G., and Bottomley, P., Analyst, 1982, 107, 206. 7 Bengston, M., Davis, R. A. H., Desmarchelier, J. M., Henning, R., Murray, W., Simpson, B. W., Snelson, J. T., Sticker, R., and Wallbank, B. E., Pestic. Sci., 1983, 14, 373. 8 Haddad, P. R., Brayan, J. G., Sharp, G. J., and Dilli, S., J. Chromatogr., 1989, 461, 337. 9 Pang, G. F., Chao, Y. Z., Fan, C. L., Zhang, J. J., and Li, X. M., JAOAC Int., 1997, 80, 63. 10 Stanker, L. H., Bigbee, C., Emon, J. V., Watkins, B., Jensen, R. H., Morris, C., and Vanderlaan, M., J. Agric. Food Chem., 1989, 37, 834. 11 Skerritt, J. H., Hill, A. S., McAdam, D. P., and Stanker, L. H., J. Agric. Food Chem., 1992, 40, 1287. 12 Gillespie, A. M., Daly, S.L., Gilvydis, D. M., Schneider, F., and Walters, S. M., JAOAC Inbt., 1995, 78, 431. 13 Nakamura, Y., Tsumura, Y., Tonogai, Y., and Shibata, T., J. Food Hyg. Soc. Jpn., 1996, 37(3), 151. Table 2 Effect of storage on pyrethroid concentration at 4, 20, and 30 °C Groundnut oil Soybean oil Sunflower oil Average Average Average Average Average Average Average Average Average recovery recovery recovery recovery recovery recovery recovery recovery recovery Added at 4 °C at 20 °C at 30 °C at 4 °C at 20 °C at 30 °C at 4 °C at 20 °C at 30 °C Pyrethroid (ppm) (%) (%) (%) (%) (%) (%) (%) (%) (%) Allethrin 0.05 96.56 92.17 88.32 96.73 93.28 89.17 95.17 91.25 89.38 Cyhalothrin 0.1 95.35 92.84 91.65 96.2 94.25 93.47 95.96 93.34 91.45 Permethrin 0.1 96.28 91.52 87.34 95.56 90.21 87.52 96.47 90.25 87.28 Cyfluthrin 0.05 96.84 95.24 94.26 97.83 96.32 94.36 95.06 93.16 92.19 Cypermethrin 0.2 103.15 102.27 100.32 104.83 102.35 100.67 102.22 100.23 99.38 Fenvalerate 0.2 97.21 96.12 96.23 96.53 95.31 93.27 96.02 94.32 93.54 Deltamethrin 0.05 105.11 104.6 103.4 104.03 103.2 101.7 104.48 103.1 101.62 Butter fat Olive oil Average Average Average Average Average Average Added recovery recovery recovery recovery recovery recovery Pyrethroid (ppm) at 4 °C (%) at 20 °C (%) at 30 °C (%) at 4 °C (%) at 20 °C (%) at 30 °C (%) Allethrin 0.2 95.30 92.37 90.65 96.4 91.37 90.62 Cyhalothrin 0.2 96.33 94.36 93.25 97.4 95.34 93.45 Permethrin 0.2 93.73 90.24 87.27 95.4 90.25 86.51 Cyfluthrin 0.2 94.93 93.16 92.31 97.5 95.32 92.31 Cypermethrin 0.2 104.9 102.36 100.12 103.0 101.3 100.22 Fenvalerate 0.2 96.53 93.61 93.45 94.2 92.31 91.28 Deltamethrin 0.15 95.5 93.21 92.14 94.5 93.26 92.31 Fig. 1 a, Gas chromatogram of untreated oil sample. b, Gas chromatogram of olive oil spiked with pesticides (each 0.2 mg ml21). 1, Allethrin; 2, cyhalothrin*; 3, permethrin*; 4, cypermethrin*; 5, fenvalerate*; 6, deltamethrin. An asterisk denotes a pesticide with more than one peak. Analyst, 1998, 123, 1799–1802 180114 Sasaki, K., Suzuki, T., and Saito, Y., J. Assoc. Off. Anal. Chem., 1987, 70, 460. 15 Crescenzi, C., Dicorcia, A., Passariello, G. M., Samperi, R., and Turnes Carou, M. I., J. Chromatogr., 1996, 733, 41. 16 Dicorcia, A., Marcomini, A., and Samperi, R., Environ. Sci. Technol., 1994, 28, 850. 17 Dicorcia, A., Crescenzi, C., Samperi, R., and Scappaticcio, L., Anal. Chem., 1997, 69, 2819. 18 Bucheli, T. D., Gruebler, F. C., Muller, S. R., and Schwarzenbach, R. P., Anal. Chem., 1997, 69, 1569. Paper 8/03097I 1802 Analyst, 1998, 123, 1799–1802

ISSN:0003-2654    

DOI:10.1039/a803097i

出版商:RSC    

年代:1998

数据来源: RSC

摘要:

Monitoring the elimination of gadolinium-based pharmaceuticals. Cloud point preconcentration and spectrophotometric determination of Gd(III)-2-(3,5-dichloro-2-pyridylazo)-5-dimethylaminophenol in urine Mar�ýa F. Silva, Liliana P. Fernandez and Roberto A. Olsina* Universidad Nacional de San Luis, Facultad de Qu�ýmica, Bioqu�ýmica y Farmacia, Departamento de Qu�ýmica Anal�ýtica, Chacabuco y Pedernera, San Luis, 5700 Argentina. E-mail: rolsina@unsl.edu.ar Received 23rd June 1998, Accepted 15th July 1998 An extraction methodology based on cloud point phase separation of non-ionic surfactants has been developed for the preconcentration of ppb amounts of gadolinium in urine as a prior step to its determination by an absorptiometric procedure.A method based on the formation of complexes with 2-(3,5-dichloro-2-pyridylazo)-5- dimethylaminophenol was used for the extraction of Gd(III) in the surfactant-rich phase of non-ionic surfactant polyethyleneglycolmono-p-nonylphenylether (PONPE 7.5).The variables affecting the combined preconcentration-absorptiometric method have been evaluated and optimised. The extraction efficiency, linearity, and the limit of detection (LOD) of the method were determined. The optimised procedure was applied to determine total and free Gd(III) contents in real urine samples of patients after the NMR imaging diagnostic examination with contrast agent. Introduction During the last three decades, the use of rare earth elements (REEs) in manufactured goods have resulted in a wide variety of electromechanical and metallurgical devices to glasses, superconductors, lasers and electronic components. More recently, their application in medicinal chemistry has been evaluated.1–4 Due to the strong tendency of the paramagnetic lanthanide cations to complex with naturally occurring agents, Gd-diethylenetriaminepentaacetate (Gd-DTPA) and 1,4,7,10-tetraazacycledodecane-1,4,7,10-tetraacetic acid (Gd- DOTA) chelates have been introduced as contrast agents in magnetic resonance imaging (MRI) and computer tomography scanning.5–8 The use of NMR contrast agents is an active area of research.9–11 Specific extraneous resonanting nuclei can be introduced to label particular parts of a molecule.Measurement of proton relaxation times is a valuable aspect of NMR studies with lanthanides. The longitudinal relaxation rates of the protons of water are much greater in the presence of Gd(III) and the enhancement of protons at the Gd-binding site is sensitive to conformational changes and other environmental perturbations.Physical chemical and biological results support the use of strongly chelated Gd complexes as NMR contrast agents. These agents are distributed in extracellular water (ECW) but not in intracellular water (ICW). The ECW : ICW ratio is often different in lesions than in normal tissue, leading to differential tissue concentration of contrast agent in lesion and normal.A rapid and complete excretion of the contrast agent is desired when a diagnostic examination of a patient is being carried out. [Gd(DTPA)]22 and [Gd(DOTA)]22 are rapidly excreted mainly into urine, while free Gd(III) is retained, with liver being the main repository.12 The excretion half-life of free Gd is enormously different from that of the Gd complexes: 7 d versus 5 min, respectively. Free Gd(III) ion has very poor acute tolerance and its long-term tolerance is unknown.It is therefore very important to determine free Gd(III) content in a Gd-based pharmaceutical. Analytical analyses are thus necessary to determine with confidence whether or not Gd was injected as [Gd(L), where L = ligands], or together with unchelated Gd(III), and if it was completely eliminated. In addition to uncomplexed Gd in the injected solutions, free metal can arise as a product of the reaction of endogenous elements with the metal complex.13 [Gd(DTPA]22 reacts rapidly with Cu(II) and Zn(II) and free Gd is detected even some minutes after injection.14 Such a situation is not observed with DOTA being the chelating agent.It is therefore very important to point out that kinetics of dissociation in vivo are more important than thermodynamic stability. After testing contrasting agents tolerance in small rodents, free Gd(III) and free DTPA ligand were found to be 25 to 100 times more toxic than the [Gd(DTPA)]22 complex (Gd(OH) LD50 0.1 mmol kg21). The use of micellar solutions in different areas of analytical chemistry has attracted much attention in recent years.Among other micelle-based separation methods, the cloud point extraction is an efficient extractive step for the enrichment of REEs, allowing the quantification of such metals at ppb levels.15,16 Aqueous solutions of many surfactant micellar systems, when subjected to temperature alterations, exhibit critical phenomenon. Non-ionic surfactant solutions have the property of separating into two liquid phases (a surfactant-rich phase and an essentially bulk aqueous phase) when they are heated above a given temperature, called the cloud point.Any analyte solubilized in the hydrophobic core of the micelles, will separate and become concentrated in the small volume of the surfactant-rich phase.17,18 The mechanism by which phase separation occurs is yet to be fully explained. In the present paper we have developed and optimised a high sensitive and low-cost method for the preconcentration and determination of total and free Gd(III) contents in urine in order to monitor the elimination of the metal after the injection of gadolinium-based pharmaceuticals. The results were contrasted against ICP.With the purpose of removing concomitant ions Analyst, 1998, 123, 1803–1807 1803and separate free Gd(III) from the chelated metal, the sample was loaded in a AG50-X8 cation exchange column and a chromatographic method19 was performed.The preconcentration step, mediated by micelles of the non-ionic surfactant polyethyleneglycolmono-p-nonylphenylether (PONPE 7.5), is performed by means of the formation of a Gd(III)-2-(3,5-dichloro- 2-pyridylazo)-5-dimethylaminophenol [Gd(III)-3,5- diClDMPAP] complex. The optimised procedure was successfully applied to determine total and free Gd(III) contents in urine samples of patients after the NMR imaging diagnostic examination with contrast agent.Experimental Apparatus A Gilford Response II spectrophotometer with 10 mm-optical path cells was used to perform the absorptiometric measurements. The ICP measures were made with a sequential inductively coupled plasma spectrometer (Baird ICP 2070, Baird, Bedford, MA, USA). Reagents and solutions A 1 mg ml21 standard solution of Gd(III) was prepared from acidic dissolution of its oxide of analytical-reagent grade (Aldrich, Milwaukee, WI, USA).Stock solutions were standardised by a chelatometric method.20 A 3.75 3 1023 mol l21 solution of purified 2-(2,5-dichloro-2-pyridylazo)-5-dimethylaminophenol (3,5-diClDMPAP) was prepared dissolving the reagent, and made up to 50 ml with distilled ethanol. A 3 3 1023 mol l21 solution of 2-(2-pyridylazo)-5-dimethylaminophenol (Tokio Kasei Industries), DMPAP, was prepared dissolving the reagent, and made up to 50 ml with distilled ethanol. As for the extracting solution, as it is not possible to obtain a real aqueous solution of surfactant polyethyleneglycolmono- p-nonylphenylether (Toko Kasei Industries, Tokyo, Japan), PONPE 7.5 (cloud point below room temperature) it was experimentally convenient to prepare a mother solution (solution A) as follows: 10 ml of PONPE 7.5, 10 ml of NaClO4 (1 mol l21), and 40 ml of distilled ethanol, were mixed and made up to 100 ml with doubly distilled water. In this way the ionic strength was adjusted to 0.1 mol l21 and adequate cloud point temperature (higher than 293 K) and accurate surfactant concentration (0.01%) could be reached.Under these conditions an optimal preconcentration factor was obtained. Other reagents used were octylphenylpoly(ethyleneglycol)ether (E. Merck, Darmstadt, Germany), TX-100, (Merck), Gd-diethylenetriaminepentaacetate ([Gd-(DTPA)]22, 469.0 mg ml21, Opacite, (Shering AG, Germany). Experimental procedure Determination of linium content. 1. 10 ml of urine [patient urine samples or normal urine samples spiked with proper amounts of Opacite and Gd(III)] were collected in a laboratory dish and dried in a water bath at 90 °C. 2. 1 ml of concentrated nitric acid and 1.5 ml of concentrated hydrochloric acid were cautiously added, the mixture was quantitatively transferred to a crucible and heated and boiled until all the nitric acid fumes were spelled and white fumes were evolved. 3. The sample was dried again and step 2 was repeated.At this point all the organic matter was decomposed. 4. 10 ml of doubly distilled water were added. The resultant clear solution was loaded in a AG50-X8 cation-exchange column, and the chromatographic procedure recommended by Crock et al.19 was performed in order to remove Ca(II) and other possible concomitants present in the sample. The elution procedure was as follows: (a) load 10 ml of sample solution; (b) elute with 10 ml of 2 mol l21 hydrochloric acid, followed by 10 ml of 2 mol l21 nitric acid, discarding both eluates; (c) elute with 10 ml of 7.5 mol l21 nitric acid, and collect the eluate for subsequent Gd determination; (d) regenerate the column by washing with 50 ml of 8 mol l21 nitric acid. 5. Finally, Gd(III) was quantified following the developed CPE-absorptiometric methodology. Determination of free Gd(III) content. 1. 10 ml of urine [patient urine samples or normal urine samples spiked with proper amounts of Opacite and Gd(III)] were collected.The sample was previously adjusted to pH 1 with hydrochloric acid. Then it was loaded in the AG50-X8 cation-exchange column and a chromatographic procedure was performed in order to remove Ca(II), other possible concomitants present in the sample and chelated Gd ([Gd-(DTPA)]22). The elution procedure was as follows: (a) load 10 ml of sample solution; (b) elute with 10 ml of 0.5 mol l21 hydrochloric acid, the eluate contains the Gd ([Gd-(DTPA)]22); (c) elute with 10 ml of 2 mol l21 of hydrochloric acid, followed by 10 ml of 2 mol l21 nitric acid, discarding both eluates; (d) elute with 10 ml of 7.5 mol l21 nitric acid, and collect the eluate for subsequent Gd determination; and (e) regenerate the column by washing with 50 ml of 8 mol l21 nitric acid. 2. Free Gd(III) was quantified following the developed CPEabsorptiometric methodology. All absorptiometric measurements were performed against a blank prepared identically, but from a non-gadolinium spiked urine sample.CPE-absorptiometric determination procedure. 1 ml of solution A, the eluate containing the metal, chelating reagent and buffer solution were placed in a centrifuge tube. The solution prepared was kept at 315 K for 10 min for equilibration and then centrifuged (graded Corning plastic centrifuge tube 15 ml capacity with a plastic cap) for 5 min at 2000 rpm (606.06g). After being cooled at 255 K for 5 min the surfactant phase which had separated became a viscous gel and the aqueous phase could be poured off. 1. Standard scale (cuvettes, 3.5 ml capacity). The surfactant phase (0.4 ml) in the tube was then dissolved at ambient temperature by adding 1 ml of ethanol and made up to 3 ml with doubly distilled water. The absorbance at 592 nm of the resultant clear solution was measured in a standard cuvette against a blank of reagents prepared identically. 2. Semi-micro scale (cuvettes with frosted thick wall, 1.4 ml capacity). The surfactant phase (0.4 ml) in the tube was dissolved by adding 0.3 ml of ethanol and made up to 1 ml with doubly distilled water.The absorbance of the resultant clear solution was measured in a semi-micro cuvette against a blank of reagents prepared identically at 592 nm. 3. Micro scale (cuvettes with frosted thick wall, 0.7 ml capacity). The absorbance of the surfactant-rich phase (0.4 ml) was directly measured in a micro cuvette against a blank of reagents prepared identically at 592 nm.Results and discussion Selection of surfactant extracting solution Experiments were carried out in order to verify the Gd(III)- 3,5-diClDMPAP-TX-100 system thermal stability. Above 333 1804 Analyst, 1998, 123, 1803–1807II 1.0 0.8 0.6 0.4 0.2 0.0 0 2 4 6 8 10 12 14 16 18 I 1.0 0.8 0.6 0.4 0.2 0.0 4 6 8 10 12 14 observed for an amphiphile concentration higher than 0.6% (m/m). In order to achieve a good preconcentration factor, 1% (m/m) was chosen as optimal.Centrifugation time. No effect was observed upon extraction parameters when centrifugation time was increased from 1 up to 30 min. Table 1 summarises the optimal experimental conditions for Gd(III) cloud point extraction-absorptiometric determination with 3.5-diClDMPAP and PONPE 7.5. Successive CPE procedures were performed to the aqueous phase in order to verify the extraction efficiency. Quantitative cloud point extraction for 3,5-diClDMPAP-gadolinium chelate was observed under the optimal experimental conditions (extraction percentage higher than 99.9%).Beer’s Law The calibration curves for the standard, semi-micro and micro scales were measured under the optimal experimental conditions. The results are shown in Table 2. Determination of free Gd(III) and [Gd(DTPA)]22 in urine samples Real (patient) urine samples. Application of the proposed methodology to the analysis of urine samples of patients after the NMR imaging diagnostic examination with contrast agent led to the results given in Table 3.Spiked urine samples. Normal urine samples spiked with proper amounts of Opacite and Gd(III) were prepared reproducing the expected samples for a patient injected with the contrast agent. The sample compositions were calculated considering the following factors: kinetics of dissociation in vivo, excretion mechanism, dosage and original concentration of the pharmaceutical. Application of the proposed methodology to the analysis of spiked samples led to the results given in Table 4.Analytical performance of the method Comparing and contrasting the analytical performance, it is clear that the developed method is superior (LOD 5.8 3 1029 mol l21) to existing ICP (LOD 6 3 1028 mol l21)29,30 and GFAAS (LOD 1 3 1027 mol l21)31,32 procedures currently employed. Table 1 Experimental conditions for the CPE-absorptiometric determination of gadolinium Equilibration temperature 315 K Equilibration time 10 min Centrifugation time 5 min Cooling time 5 min Working pH 9.50 Buffer solution sodium tetraborate 1023 mol l21 Ionic strength 0.1 mol l21 (sodium perchlorate) Surfactant PONPE 7.5 (1% m/m) Maximum of reagent absorption 450 nm Maximum of complex absorption 592 nm %Ea 99.98% a Percentage extracted by the successive extraction method.Table 2 Beer’s Law Apparent molar LODc/ Beer’s Law Scale Pa absorptivityb mol l21 fulfilment Standard 3.33 4.60 3 105 mol l21 cm21 4.35 3 1028 3.42–510 mg l21 Semi-micro 10 1.38 3 106 mol l21 cm21 1.45 3 1028 1.13–171 mg l21 Micro 25 3.45 3 106 mol l21 cm21 5.80 3 1029 0.45–68 mg l21 a Preconcentration factor = micellar phase volume (ml)/Aqueous phase volume (ml).b Referred to 10 ml. c Lower limit of detection. Table 3 Determination of [Gd(DTPA)]22 and free Gd(III) in real (patient) urine Sample Scale Total Gd (s)/mol l21a Free Gd(III) (s)/mol l21a 1b Standard 6.78 3 1023 (5 3 1024) 1.02 3 1025 (9.2 3 1027) 2c Micro 3.36 3 1027 (2.42 3 1028) 2.55 3 1027 (1.74 3 1028) a Mean value of six patients, three splits each (n = 27). b Urine samples taken 5 min after injection of 10 ml of Opacite.c Urine samples taken 7 d after injection of 10 ml of Opacite. Table 4 Determination of [Gd(DTPA)]22 and free Gd(III) in urine [Gd(DTPA)]22 Free Gd(III) Sample compositiona Scale Added/mg Found/mg RE(%)b sc Added/mg Found/mg RE(%)b sc Pd = 5 Standard 9.35 9.39 0.4 0.022 0.53 0.54 1.9 0.013 P = 0.5 Semi-micro 1.60 1.63 1.8 0.017 0.91 0.93 2.2 0.008 P = 0.1 Micro 0.160 0.163 1.9 0.009 0.45 0.46 2.2 0.004 P = ( ) [ ] - + Gd L Gd 2 3 : a Referred to 10 ml sample volume.b Relative percentage error. c Standard deviation (n = 6). d relative concentration respect to Gd content. Note: The present results were contrasted against ICP following a direct methodology; synthetic aqueous samples were prepared reproducing the total Gd contents in the urine samples. Standard scale: added 9.89 mg; found 9.92 mg.Semi-micro scale: added 2.51 mg; found 2.48 mg. Micro scale: added 0.61 mg; not found. 1806 Analyst, 1998, 123, 1803–1807Conclusions Cloud point extraction offers an interesting possibility for preconcentrating Gd in urine samples. A safe, low-cost and highly sensitive methodology for monitoring Gd in urine has been developed. Studies given above have demonstrated quantitative cloud point extraction of the metal chelate, high efficiency of the chromatographic step at separating concomitant ions and chelated Gd, and consequently, the possibility to determine total gadolinium content as well as free Gd(III) in a wide concentration range.The results demonstrate the usefulness of the proposed method to effectively determine the speciation of gadolinium. The proposed method can also be applied to the determination of free Gd(III) for quality control of gadolinium-based pharmaceuticals. We gratefully acknowledge support of this work by National University of San Luis (Project No. 7502) and CONICET. References 1 H. Seiler, A. Sigel and H. Sigel, Handbook on Metals in Clinical and Analytical Chemistry, Marcel Dekker, New York, 1994, pp. 354. 2 C. H. Evans, Biochemistry of the Lanthanides, Plenum Press, New York, 1990. 3 Martindale, The Extra Pharmacopea, 26th edn., Pharmaceutical Press, London, 1972. 4 R. W. Deng, J. Wu and L. Long, Bull. Soc. Chim. Belges, 1992, 101, 438. 5 J. C. Bousquet, S. Saini, D. Stark, P.Hahn, M. Nigam, J. Wittenberg and J. Ferrucci, Radiol., 1988, 166, 693. 6 F. L. Van der Vyver and G. V. Peersman, Magn. Res. Imag., 1991, 8, 333. 7 K. Kumar, K. Sukumaran and M. Tweedle, Anal. Chem., 1994, 66, 295. 8 J. Hagan, S. C. Taylor and M. Tweedle, Anal. Chem., 1988, 60, 514. 9 See 1, pp. 365. 10 D. Fornasiero, J. C. Bellen, R. J. Baker and B. R. Chatterton, Invest. Radiol., 1985, 22, 322. 11 Y. K. Adzamli, H. Gries, D. Johnson and M. Blau, J. Med. Chem., 1989, 32, 139. 12 C. G. Bunzli and G. R. Choppin, Lanthanide Probes in Life, Chemical and Earth Sciences, Elsevier, 1989, Chapter 5. 13 M. F. Tweedle, J. J. Hagan, E. V. Dose, S. M. Mantha and S. M. Cicero, Magn. Res. Imag., 1992, 10 : 4, 641. 14 A. E. Martell and R. M. Smith, Critical Stability Constant, Plenum Press, New York, 1974, vol. 3. 15 W. Hinze and E. Pramauro, Crit. Rev. in Anal. Chem., 1993, 24(2), 133. 16 M. F. Silva, L. Fern�andez, R. Olsina and D. Stacchiola, Anal. Chim. Acta, 1997, 342, 229. 17 C. Garc�ýa Pinto, J. L. Perez Pav�on, B. Moreno Cordero, E. Romero Beato and S. Garc�ýa Sanchez, J. Anal. At. Spectrom., 1996, 11, 37. 18 S. Sirimanne, J. Barr, D. Patterson and Li Ma, Anal. Chem., 1996, 68, 1556. 19 J. C. Crock, F. E. Lichte, G. O. Riddle and C. L. Beech, Talanta, 1986, 33, 601. 20 A. Flaschka, EDTA titrations—An Introduction to Theory and Practice, 2nd edn., Pergamon Press, London, 1967. 21 H. Watanabe, T. Saitoh, T. Kamidate and H. Haraguchi, Mikrochim. Acta, 1992, 106, 83. 22 Z. Larson, Phys. Chem., 1967, 56, 173. 23 A. Hrdlika, J. Havel, B. Moreno Cordero and M. Valiente, Anal. Sci., 1991, 7, 925. 24 L. D. Martinez, E. Perino, E. J. Marchewsky and R. A. Olsina, Talanta, 1993, 40(3), 385. 25 L. Fernandez and R. Olsina, Talanta, 1991, 38, 339. 26 M. F. Silva, L. Fernandez and R. Olsina, An. Qu�ýmica. Int. Ed., 1996, 92, 344. 27 L. Fernandez and R. Olsina, Anal. Sci., 1990, 6, 411. 28 L. Fernandez and R. Olsina, Talanta, 1992, 39, 1605. 29 P. W. Boumans, ICP-ES part I, Wiley, New York, 1987, p. 132. 30 A. Varma, Handbook of ICP-AES, CRC Press, New York, 1991, p. 58. 31 P. J. Potts, A. Handbook of Silicate Rock Analysis, Blackie, London, 1992, p. 147. 32 A. Varma, Handbook of Atomic Absorption Spectroscopy, CRC Press, New York, 1990, p. 272. Paper 8/04789H Analyst, 1998, 123, 1803&ndash

ISSN:0003-2654    

DOI:10.1039/a804789h

出版商:RSC    

年代:1998

数据来源: RSC

摘要:

Flow injection spectrophotometric determination of hydrogen peroxide using a crude extract of zucchini (Cucurbita pepo) as a source of peroxidase Iolanda da Cruz Vieira and Orlando Fatibello-Filho* Departamento de Química, Grupo de Química Analítica, Centro de Ciências Exatas e de Tecnologia, Universidade Federal de São Carlos, Caixa Postal 676, CEP 13.560-970, São Carlos, SP, Brazil. E-mail: bello@dq.ufscar.br Received 8th May 1998, Accepted 7th July 1998 A flow injection (FI) spectrophotometric procedure is presented for determining hydrogen peroxide for pharmaceutical use and in swimming pool water samples.Crude extracts of several vegetables such as peach, yam, manioc, artichoke, sweet potato, turnip, horseradish and zucchini were investigated as the source of peroxidase (donor: hydrogen peroxide oxidoreductase, POD; EC 1.11.1.7). Of these, a zucchini crude extract was found to give highest specific activity and was used directly as the carrier solution.This enzyme catalyses the oxidation of guaiacol in the presence of hydrogen peroxide to tetraguaiacol, which shows strong absorbance at 470 nm. For the optimum extraction conditions found, the peroxidase activity in the crude extract did not vary for at least 5 months when stored at 4 °C and decreased by only 2–3% during an 8 h working period at 25 °C. The recovery of hydrogen peroxide from two samples ranged from 97.8 to 103.0% and a rectilinear calibration curve for hydrogen peroxide concentration from 1.6 3 1025 to 6.6 3 1024 mol l21 was obtained.A detection limit of 2.1 3 1026 mol l21 and a sample throughput of 32 h21 were attained. The relative standard deviations were < 0.2% for hydrogen peroxide solutions containing 2.0 3 1024 and 4.0 3 1024 mol l21 (n = 10) and a paired t-test showed that all results obtained for water samples using this FI procedure and permanganate titration agreed at the 95% confidence level. The determination of hydrogen peroxide is of great interest in a large variety of samples in industrial and environmental fields and in clinical control.1,2 Several methods have been proposed for the determination of hydrogen peroxide, including oxidation –reduction titrimetry,3,4 chemiluminescence,5 spectophotometry6 –8 and electrochemical.9–14 Nevertheless, there are no procedures for determining hydrogen peroxide using crude extracts as a source of peroxidase.The first biosensor for hydrogen peroxide was constructed using a bovine liver membrane containing catalase with an oxygen electrode.15 Recently, the development of several biosensors with the use of novel biological materials as biocatalysts has received considerable attention with the aim of replacing pure enzyme.Several tissue biosensors for determining hydrogen peroxide, such as asparagus tissue,16 grape tissue,17 pineapple,18 kohlrabi,19 tobacco callus20 and horseradish root,21–23 have been used as peroxidase sources.This class of biocatalytic materials maintains the enzyme to be used in these biosensors in its natural environment, which results in considerable stabilization of the desired enzymatic activity. Peroxidase (donor: hydrogen peroxide oxidoreductase, POD; EC 1.11.1.7), includes a class of enzyme extensively distributed in higher plants (e.g., horseradish, turnip, fig sap), animals (e.g., tryptophan pyrrolase, iodine peroxidase of thyroid) and microorganisms (e.g., cytochrome c peroxidase of yeast).24 Peroxidase catalysis is associated with four types of activity: peroxidic, oxidative, catalytic and hydroxylation.24–27 We have been developed several biosensors and enzymatic batch and flow injection procedures for determining phenolic compounds,28–30 l-dopa and carbidopa,31 methyldopa and dopamine32 and sulfite33 using crude extracts of various vegetables in place of isolated enzymes.The use of such biological materials is very attractive because of their high stability, high enzyme activity concentration, very low cost and fewer cofactor requirements in comparison with the pure enzymes.In this paper, a simple, sensitive and rapid flow injection (FI) enzymatic procedure is reported for determining hydrogen peroxide for pharmaceutical use and in swimming pool water samples. A crude extract of zucchini (Cucurbita pepo) was used as the enzymatic source of peroxidase (POD; EC 1.11.1.7) directly in the carrier solution.In the peroxidic reaction, this enzyme catalyses the oxidation of various hydrogen donors such as guaiacol in the presence of hydrogen peroxide to tetraguaiacol, which shows a strong absorption at 470 nm (see Fig. 1). The use of an insoluble polyvinylpyrrolidone such as Polyclar SB-100 to remove natural phenolic compounds from the solution in the preparation of the crude extract (homogenate) of zucchini led to a substantial increase in the enzyme activity, storage time and stability of the baseline.Experimental Apparatus A DuPont Instruments (Newtown, CT, USA) Model RC-5B centrifuge, provided with a Model SS-34 rotor, was used in the preparation of the crude extract of the zucchini. A Hewlett-Packard (Boise, ID, USA) Model 8452A UV– visible spectrophotometer with a quartz cell (optical path 1 cm) was used in POD activity and protein determinations. An eight-channel Ismatec (Zurich, Switzerland) Model 7618-40 peristaltic pump supplied with Tygon pump tubing was used for the propulsion of the fluids.The manifold was assembled with polyethylene tubing (0.8 mm id). A Micronal (São Paulo, Brazil) Model B352 automatic proportional com- Analyst, 1998, 123, 1809–1812 1809OCH3 OH OCH3 O O OCH3 O O OCH3 OCH3 4 + 4H2O2 + 8H2O Peroxidase mutator was used for inserted reagent and sample solutions. Spectrophotometric measurements were carried out using a Femto (São Paulo, Brazil) Model 435 spectrophotometer with a glass flow cell (optical path 1.0 cm) connected to a Cole Parmer (Niles, IL, USA) Model 12020000 two-channel strip-chart recorder.The effect of temperature on the enzymatic reaction was evaluated using a Tecnal (Piracicaba, Brazil) Model TE184 thermostatically controlled water-bath. Reagents and solutions All reagents were of analytical-reagent grade and all solutions were prepared with water from a Millipore (Bedford, MA, USA) Milli-Q system (Model UV Plus Ultra-Low Organics Water).Guaiacol and peroxidase from horseradish (type VI, P8375) were purchased from Sigma (St. Louis, MO, USA). Hydrogen peroxide was purchased from Aldrich (Milwaukee, WI, USA); a 1.00 3 1021 mol l21 stock standard solution was prepared daily in 0.1 mol l21 phosphate buffer (pH 7.0) and standardized by a conventional method.3 Working standard solutions of 1.6 3 1025–6.6 3 1024 mol l21 were prepared from the stock standard solution by dilution with 0.1 mol l21 phosphate buffer (pH 7.0).Polyclar SB-100 used as a protective and/or stabilizing agent in the crude extract preparation was kindly donated by GAF (Wayne, NJ, USA). This polyvinylpyrrolidone was first purified essentially as described elsewhere.30–33 Commercial hydrogen peroxide samples for pharmaceutical use were purchased from a local drug store. Swimming pool water samples without any chlorine species were collected and an iced cooler for storage was used during transport to the laboratory.Healthy zucchini (Cucurbita pepo), a variety of squash with a long, narrow shape and a greenish rind, purchased from a local producer were selected, washed, hand-peeled, chopped and cooled in a refrigerator at 4 °C. Zucchini crude extract preparation A 25 g amount of the frozen peeled zucchini was homogenized in a liquefier with 100 ml of 0.1 mol l21 phosphate buffer (pH 6.0) containing 2.5 g of Polyclar SB-100 for 2 min at 4–6 °C. The homogenate was rapidly filtered through four layers of cheesecloth and centrifuged at 13 500 rpm for 15 min at 4 °C.The resulting supernatant was stored at this temperature in a refrigerator and utilized as the enzymatic source after the determination of peroxidase activity and total protein. Peroxidase activity and total protein determinations Peroxidase activity present in the crude extract was determined in triplicate by measurement of the absorbance at 470 nm of tetraguaiacol24–27 produced by the reaction between 0.2 ml of supernatant solution, 2.7 ml of 0.05 mol l21 guaiacol solution and 0.1 ml of 10.0 mmol l21 hydrogen peroxide solution in 0.1 mol l21 phosphate buffer (pH 7.0) at 25 °C.The initial rate of guaiacol peroxidation reaction (Fig. 1) was a linear function of time for 1.5–2.0 min. One activity unit is defined as the amount of enzyme that causes an increase of 0.001 absorbance per minute under the conditions described above. Total protein concentration was determined in triplicate by the method of Lowry et al.34 using bovine serum albumin as a standard.POD solution in phosphate buffer A 270 units ml21 POD solution in 0.1 mol l21 phosphate buffer (pH 7.0) was prepared daily by dilution of 10 ml of a 6750 units ml21 POD solution with 0.1 mol l21 phosphate buffer (pH 7.0) in a 250 ml calibrated flask using the same buffer solution. Sample preparation and FI enzymatic procedure Appropriate dilution of hydrogen peroxide for pharmaceutical use and swimming pool water samples with 0.1 mol l21 phosphate buffer solution (pH 7.0) containing guaiacol at a convenient concentration was performed in order to obtain a concentration of the hydrogen peroxide in the range 1.6 3 1025–6.6 3 1024 mol l21.The single channel spectrophotometric flow system used was similar to that reported previously.30,31,33 In this work, a 270 units ml21 POD solution in 0.1 mol l21 phosphate buffer (pH 7.0) was used as the carrier solution at a flow rate of 1.0 ml min21.A 0.05 mol l21 guaiacol–hydrogen peroxide solution contained in the sample loop (50 cm, 250 ml) was injected and transported by the enzymatic carrier stream. A 250 cm tubular coiled reactor maintained in a 25 °C water-bath was placed in the analytical path in order to provide better reaction conditions and the tetraguaiacol formed (Fig. 1) was measured in the flowthrough spectrophotometric cell at 470 nm. Results and discussion Selection and preparation of the crude extract Peroxidase is widely distributed in the plant kingdom, in certain animal tissues and also in microorganisms.This enzyme has been isolated from several sources such as horseradish, turnip and soybean. The main source of peroxidase, commercialized by various companies, is horseradish.24–27. In this work, vegetable crude extracts such as peach (Prunus persica), yam (Alocasia macrorhiza), manioc (Manihot utilissima), artichoke (Cynara scolymus L.), sweet potato (Ipomoea batatas L.Lam.), turnip (Brassica campestre ssp. rapifera), horseradish (Armoracia rusticana) and zucchini (Cucurbita pepo) were obtained and characterized. The activity and total protein of the crude extracts of the these vegetable materials varied according to the extraction procedure and medium used. The buffer-to-tissue ratio was an important factor in the preparation of POD from all these enzymatic sources. In this study, the enzyme was extracted using ratios varying from 2 : 1 to 6 : 1 ml g21 and the highest specific activity for each was obtained at a ratio 4 : 1 ml g21.The effect of buffer pH on the extraction of POD was also investigated in the pH range 5.0–7.5. The highest enzymatic activity for each was obtained at pH 6.0. To minimize the effect of the natural phenolic compounds responsible for the decrease in the POD activity in these crude extracts, Polyclar SB-100 at a mass ratio of 1 : 10 g g21 was used.28–33. The enzyme activity of the crude extract of zucchini obtained using this PVP did not vary for at least 5 months when stored in a refrigerator at 4 °C, whereas that of a sweet potato crude extract decreased by 5–7% and those of turnip, horseradish and artichoke crude extracts decreased by 15–20% under the same experimental conditions.Table 1 shows the activity (units ml21), total protein (mg ml21) and specific activity (units mg21 of protein) obtained in triplicate using different vegetable crude extracts.As can be seen, the peach crude extract showed the lowest specific activity Fig. 1 Reaction between guaiacol, hydrogen peroxide and peroxidase 1810 Analyst, 1998, 123, 1809–1812whereas with the zucchini crude extract the highest enzymatic activity was obtained. To the best of our knowledge, no work has been published on obtaining peroxidase from zucchini. This is surprising, since the specific activity of this crude extract was about 35% higher than that obtained for horseradish crude extract, a common source of commercial peroxidase.Therefore, the crude extract of zucchini was used in subsequent experiments. Storage time and crude extract stability For the optimum extraction conditions described above, the peroxidase activity in the crude extract did not vary for at least 5 months when the extract was stored in a refrigerator at 4 °C and decreased by only 2–3% after an 8 h working period at 25 °C. Similar long storage times and low background absorbance of the crude extract obtained with Polyclar SB-100 in the present work were also obtained in previous studies,29–33 showing the advantage of the medium, preparation method and biological material used in this work in comparison with other substances normally used such as sodium azide and lcysteine. 35 Reaction between guaiacol, hydrogen peroxide and peroxidase The flow injection procedure for determining hydrogen peroxide is based on the catalytic oxidation of guaiacol by peroxidase in the presence of hydrogen peroxide to tetraguaiacol, which shows strong absorption at 470 nm (Fig. 1). Hence, when hydrogen peroxide–guaiacol solution is inserted in the flow injection system the formation of tetraguaiacol increased with increase in hydrogen peroxide concentration. A preliminary batch study showed that 0.05 mol l21 guaiacol and 5.0 3 10–3 mol l21 peroxidase do not react in a time range of 0–3 h. Therefore, in this study all guaiacol standard and sample solutions were injected together (same solution) in the flow injection system containing zucchini crude extract as the carrier solution.Effect of enzyme concentration, pH and temperature The effect of the POD concentration from 12 to 380 units ml21 on the analytical signal (absorbance) for 5.0 3 1022 mol l21 guaiacol and 2.0 3 1023 mol l21 hydrogen peroxide was investigated. The absorbance signal increased linearly with increase in enzyme solution concentration up to 300 units ml21 POD.Therefore, a concentration of 270 units ml21 was adopted in this work. The effect of pH in the range 5.0–7.5 on the absorbance of 5.0 3 1022 mol l21 guaiacol and 2.0 3 1023 mol l21 hydrogen peroxide solution and 270 units ml21 POD enzyme was also studied. The optimum pH value for POD activity was 7.0. The effect of temperature was studied between 15 and 65 °C. The enzyme exhibited the highest activity in the range 25–45 °C, after which a gradual decline in its activity owing to heat inactivation was observed between 45 and 65 °C.Therefore, a temperature of 25 °C was selected for further experiments. A commercial POD from horseradish under the above experimental conditions showed an optimum pH of 6.5 and maximum activity in the temperature range 20–40 °C. Flow injection parameters and reaction conditions The effect of varying the sample loop length from 25 to 100 cm (125–500 ml) on the analytical response was initially evaluated.The best sample loop length was found to be 50 cm (250 ml). Table 1 Activity, total protein and specific activity obtained from various vegetable crude extracts Vegetable crude extract Activity/ units ml21 Total protein/ mg ml21 Specific activity/ units mg21 protein Peach 262 2.45 107 Yam 1929 5.84 303 Manioc 1738 2.67 651 Artichoke 8905 4.61 1 932 Sweet potato 8688 3.21 2 707 Turnip 4226 0.49 8 624 Horseradish 3381 0.33 10 245 Zucchini 6750 0.49 13 776 Table 2 Calibration equations obtained for hydrogen peroxide and related parameters as a function of guaiacol concentration in mol l21 [Guaiacol]/ mol l21 Equation Linearity range/ 1025 mol l21 Correlation coefficient* (r) Detection limit/ mol l21 1.0 3 1023 A = 0.014 + 985.44[H2O2] 2.4–58.2 0.9986 4.3 3 1026 5.0 3 1023 A = 0.012 +1260.15[H2O2] 1.–60.6 0.9989 2.8 3 1026 5.0 3 1022 A = 0.011 +1571.10[H2O2] 1.6–65.6 0.9993 2.1 3 1026 * n = 6.Table 3 Results of addition–recovery experiments using hydrogen peroxide at three different reference concentrations [Hydrogen peroxide]/mg l21* Recovery Sample Added Found (%) Swimming pool water 2.72 2.70 ± 0.05 99.3 5.44 5.34 ± 0.08 98.2 8.16 8.36 ± 0.04 102.4 10.88 11.21 ± 0.10 103.0 Hydrogen peroxide for pharmaceutical use 2.72 2.66 ± 0.06 97.8 5.44 5.37 ± 0.09 98.7 8.16 8.04 ± 0.13 98.5 10.88 10.86 ± 0.08 99.8 * n = 6.Fig. 2 Transient absorbance signals obtained in triplicate for hydrogen peroxide standard solutions of 1.6, 3.9, 6.6, 13.1, 19.7, 26.2, 39.4, 52.5 and 65.6 3 1025 mol l21, five samples (A, hydrogen peroxide 1; B, hydrogen peroxide 2; C, hydrogen peroxide 3; D, swimming pool water 1, and E, swimming pool water 2) and the standard solutions again.Analyst, 1998, 123, 1809–1812 1811With respect to sensitivity and analytical frequency, the optimum compromise was attained using a coiled reactor 250 cm long and a flow rate of 1.0 ml min21. The dispersion coefficient of the flow injection system was 1.12.Effect of guaiacol concentration on the calibration curves The effect of guaiacol concentration at 1.0 3 1023; 5.0 3 1023 and 5.0 3 1022 mol l21 on the linearity of the hydrogen peroxide calibration curves and detection limit (three times the signal blank-to-slope ratio) is shown in Table 2. The best linearity of the calibration curve (absorbance versus concentration of hydrogen peroxide) was attained at a guaiacol concentration of 5.0 3 1022 mol l21.At this guaiacol concentration, the greatest linearity range and the lowest detection limit of 2.1 3 1026 mol l21 were obtained. Therefore, this concentration was adopted in all further work. Analytical characteristics, recovery and application The optimum FI conditions established as described above, i.e., sample loop length of 50 cm (250 ml), coiled reactor length of 250 cm, carrier flow rate of 1.0 ml min21, enzyme concentration of 270 units ml21 in phosphate buffer (pH 7.0) and temperature of 25 °C, were adopted in the proposed method.Recoveries of 97.8–103.0% of hydrogen peroxide, from two samples (n = 6), were obtained using the FI spectrophotometric procedure (Table 3). This is good evidence of the absence of matrix effects in the proposed method. In addition, the RSDs were < 0.2% for solutions containing 2.0 3 1024 and 4.0 3 1024 mol l21 of hydrogen peroxide (n = 10). Standard solutions containing from 1.6 31025 to 6.6 31024 mol l21 hydrogen peroxide were employed to construct the calibration curve, A = 0.011 + 1571.10 C (r = 0.9993), where C is the hydrogen peroxide concentration in mol l21 and A is the absorbance.Triplicate signals for nine reference hydrogen peroxide standard solutions and quadruplicate signals for three samples of hydrogen peroxide for pharmaceutical use and two swimming pool water sample solutions demonstrated good precision and baseline stability (Fig. 2).Table 4 presents the results obtained using a permanganate titration3 and the proposed FI enzymatic method. Applying a paired t-test to the results obtained by either procedure, it was found that all results were in agreement at the 95% confidence level. The analytical frequency was 32 samples h21 and the FI enzymatic procedure proposed in this paper is simple, precise, inexpensive and rapid and is suitable for routine analysis. Financial support from FAPESP (Process 91/2637-5), PADCT/ CNPq (Process 62.0060/91-3) and CNPq (Process 50.1638/91-1) and also a scholarship granted by FAPESP (Process 97/04764-0) to I.C.V.are gratefully acknowledged. References 1 Kirk, R. E., and Othmer, D. F., Encyclopedia of Chemical Technology, Wiley, New York, 1981, p. 12. 2 Ruzgas, T., Csöregi, E., Emnéus, J., Gorton, L., and Marko-Varga, G., Anal. Chim. Acta, 1996, 330, 123. 3 Vogel, A. I., Textbook of Quantitative Inorganic Analysis, Longman, New York, 1989, p. 372. 4 Hurdis, E. C., and Romeyn, H., Anal. Chem., 1954, 26, 320. 5 Aizawa, M., Ikariyama, Y., and Kuno, H., Anal. Lett., 1984, 17, 555. 6 Matsubara, C., Kawamoto, N., and Takamura, K., Analyst, 1992, 117, 1781. 7 Matsubara, C., Kudo, K., Kawashita T., and T. Takamura, T., Anal. Chem., 1985, 57, 1107. 8 Clapp, P. A., Evans D. F., and Sheriff, T. S. S., Anal. Chim. Acta, 1989, 218, 331 9 Akmal, N., and Mark, H. B., Jr., Anal. Lett., 1992, 25, 2175. 10 Gao, Z., Ivaska, A., Li, P., Lui, K., and Yang, J., Anal.Chim. Acta, 1992, 259, 211. 11 Lundbäch, H., Johansson, G., and Holst, O., Anal. Chim. Acta, 1983, 155, 47. 12 Aizawa, M., Karube, I., and Suzuki, S., Anal. Chim. Acta, 1974, 69, 431. 13 Moody, G. J., Sanghera, G. S., and Thomas, J. D. R., Analyst, 1986, 111, 605. 14 Bennetto, H. P., Dekeyzer, D. R., Delaney, G. M., Koshy, A., Mason, J. R., Razak, L. A., Stirling, J. L., and Thurston, C. F., Analyst, 1987, 8, 22. 15 Mascini, M., Jannelle, M., and Palleschi, G., Anal.Chim. Acta, 1982, 138, 6. 16 Oungpipat, W., Alexander, P. W., and Southwell-Keely, P., Anal. Chim. Acta, 1995, 309, 35. 17 Wijesuriya, D., Lin, M. S., and Rechnitz, G. A., Anal. Chim. Acta, 1990, 234, 453. 18 Lin, M. S., Tham, S. Y., and Rechnitz, G. A., Electroanalysis, 1990, 2, 511. 19 Chen, L., Lin, M. S., and Hara, M., and Rechnitz, G. A., Anal. Lett., 1991, 24,1. 20 Navarante, A., and Rechnitz, G. A., Anal. Chim. Acta, 1992, 257, 59. 21 Liu, H. Y., Zhang, Z.N., Fan, Y. B., Dai, M., Zhang, X., Wei, J. J., Qiu, Z. N., Li, H. B., Wu, X. X., Deng, J. Q., and Qi, D. Y., Fresenius’ J. Anal. Chem., 1997, 357, 297. 22 Fang, Y., Cai, R. H., Deng, J. Q., and Deng, Z. F., Electroanalysis, 1992, 4, 819. 23 Wang, J., and Lin, M. S., Electroanalysis, 1989, 1, 43. 24 Whitaker, J. R., Principles of Enzymology for the Food Sciences, Marcel Dekker, New York, 1985, p. 592. 25 Robinson, D. S., in Peroxidase and Catalase in Food, ed. Robinson, D. S., and Eskin, N. A. M., Elsevier, New York, 1991, p. 1. 26 Campa, A., in Biological Roles of Plant Peroxidase: Known and Potential Function, ed. Everse, K., and Grisham, M. B., CRC Press, Boca Raton, FL, 1991, vol. 2, p. 25. 27 Dunford, H. B., and Stillman, J. S., Coord. Chem. Rev., 1976, 19, 187. 28 Signori, C. A., and Fatibello-Filho, O., Quím. Nova, 1994, 17, 38. 29 Vieira, I. da C., and Fatibello-Filho, O., Anal. Lett., 1997, 30, 895. 30 Fatibello-Filho, O., and Vieira, I. da C., Anal. Chim. Acta, 1998, 363, 111. 31 Fatibello-Filho, O., and Vieira, I. da C., Analyst, 1997, 122, 345. 32 Vieira, I. da C., and Fatibello-Filho, O., Talanta, 1998, 46, 559. 33 Fatibello-Filho, O., and Vieira, I. da C., Anal. Chim. Acta, 1997, 354, 51. 34 Lowry, O. H., Rosebrough, N. J., Farr, A. L., and Randall, R. J., J. Biol. Chem., 1951, 193, 265. 35 Uchiyama, S., and Suzuki, S., Anal. Chim. Acta, 1992, 261, 361. Paper 8/03478H Table 4 Analysis of samples of hydrogen peroxide for pharmaceutical use and swimming pool water using redox titration with potassium permanganate3 and the proposed FI enzymatic procedure (n = 4, 95% confidence level) [Hydrogen peroxide]/mg ml21 Relative Sample Permanganate Enzymatic error (%) Hydrogen peroxide 1 32.6 ± 0.2 31.1 ± 0.1 24.6 Hydrogen peroxide 2 31.0 ± 0.2 29.6 ± 0.2 24.5 Hydrogen peroxide 3 30.3 ± 0.3 31.0 ± 0.1 +2.3 Swimming pool water 1 (4.5 ± 0.3) 3 1023 (4.6 ± 0.2) 3 1023 +2.2 Swimming pool water 2 (4.5 ± 0.2) 3 1023 (4.4 ± 0.1) 3 1023 22.2 1812 Analyst, 1998, 123, 1809–1812

ISSN:0003-2654    

DOI:10.1039/a803478h

出版商:RSC    

年代:1998

数据来源: RSC

摘要:

Multi-element analysis of environmental samples by X-ray fluorescence spectrometry using a simple thin-layer sample preparation technique Tomohiro Kyotani and Masaaki Iwatsuki* Department of Applied Chemistry and Biotechnology, Faculty of Engineering, Yamanashi University, Takeda, Kofu 400-8511, Japan Received 18th May 1998, Accepted 21st July 1998 A simple and rapid method was developed for the simultaneous multi-element (Na, Mg, Al, Si, S, Cl, K, Ca, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, Br, Cd and Pb) analysis of environmental samples, such as vehicle exhaust particulates and atmospheric dust, by X-ray fluorescence (XRF) spectrometry using a simple thin-layer sample preparation technique.A portion of vehicle exhaust particulates or atmospheric dust CRM was suspended in a small volume of a dispersant on a hydrophilized poly(tetrafluoroethylene) (PTFE) membrane filter, dispersed by air blown from a pump and dried by IR irradiation. The resulting filter was sandwiched between two PTFE rings and a polycarbonate–dichloromethane solution was dropped onto the filter to obtain good resistance to X-ray irradiation.The XRF measurements of the specimens gave reasonable reproducibilities. Calibration graphs of analyte elements were obtained from XRF measurements of thin-layer standard specimens, prepared by impregnating activated carbon powder with standard solutions of the elements and then drying the powder. Analytical results obtained without matrix correction were in good agreement with the certified values.X-ray fluorescence (XRF) analysis of powdered samples, such as rocks, minerals and ceramics, has normally been carried out by several methods: i.e., the pellet (briquette) method1–4 and the glass bead (fusion) method1,2,5–7 for normal amounts of sample, or the filter cake method,8,9 which consists in filtering the suspension of powder or precipitate, for small amounts of sample. The pellet method has been widely used, because of its simplicity, rapidity and non-destructive nature.However, many calibration specimens similar to the real sample or computeraided correction procedures are required for the precise correction of matrix effects, although the Compton scatter correction method can easily be applied for the rough correction of such effects. The glass bead method can provide homogeneous specimens and allows easy preparation of synthetic standards; however, it requires a melting procedure with a considerable amount of flux, and causes some loss of sensitivity for trace elements.On the other hand, the filter cake method allows simple and rapid preparation of thin-layer specimens and sensitive analysis. However, some loss may be caused by dissolution into the dispersing medium and filtration. Giaugue et al.10 reported an ingenious method for the preparation of thin-layer specimens by filtering aerosols of rock, glass and pottery powders. Haupt et al.11,12 also reported a thin-film standard preparation method using an aerosol generator.However, unrepresentative specimens may be obtained for powders containing ‘sticky’ particles by these methods. Renault and McKee13 reported a simple and convenient method of mounting powdered samples onto adhesive tape for the determination of environmental lead. Cohen and Smith14 reported a method for the preparation of a monolayer film of particles for determination of major elements in silicate rocks.However, the homogeneous distribution of the powder or particles is not easy to achieve with these methods . Although thin-film standards are commercially available, they are expensive and difficult to obtain at appropriate concentrations for many elements to be determined. In our previous work,15 homogeneous thin-layer standard specimens were conveniently prepared by impregnating activated carbon powder with standard solutions of the elements of interest followed by drying of the powder for the multi-element XRF analysis of airborne particulate matter.This preparation method can easily provide the desired amounts of all the elements of interest in the standard specimens and allows simple determination free from collection of the matrix. If powdered samples can be prepared as a thin-layer in the same way as these standard specimens, the matrix effects in the XRF intensity measurements should also be absent. Therefore, the application of the proposed preparation method to environmental samples was investigated by using vehicle exhaust particulates and atmospheric dust collected by a bag filter.However, its direct application was difficult and modifications were needed. This paper describes the modified methods, which gave satisfactory results for the analysis of these CRMs. Experimental Samples, reagents and filters The environmental samples used were a CRM from the National Institute for Environmental Studies (NIES), Japan, No. 8 vehicle exhaust particulates16 and a standard sample of atmospheric dust (AS-1)17 collected by a bag filter. The NIES No. 8 sample was used as received without drying. The AS-1 sample was stored in a constant humidity box adjusted to 50% with 43% sulfuric acid. Activated carbon powder (Merck, Darmstadt, Germany) was ground in a tungsten carbide mortar and washed with nitric acid; the fine powder of < 1.0 mm size was collected by a sedimentation method.All reagents used were of analyticalreagent grade. De-ionized, distilled water was used throughout. A 0.01% surfactant (Tween-85) solution was prepared by dissolving the reagent (Wako Pure Chemical Industries, Osaka, Japan) in water, and was stable for 1 month. A 0.02% polycarbonate–dichloromethane solution was prepared by dissolving the resin powder (Teijin Chemicals, Tokyo, Japan) in dichloromethane. Analyst, 1998, 123, 1813–1816 1813Hydrophilized poly(tetrafluoroethylene) (PTFE) membrane filters (Millipore, Bedford, MA, USA; Omnipore-JH, 0.45 mm pore size and 47 mm diameter) were used as supports for the powdered samples and their mixtures with activated carbon powder for XRF analysis, because of their good resistance to Xray irradiation.Apparatus and measuring conditions A Shimadzu (Kyoto, Japan) VF-320A (scanning type) XRF spectrometer with a data processor (DP-32II) was operated with a Rh tube at 40 kV and 60 mA. The XRF specimen was fixed between two aluminium masks with a hole of 40 mm in diameter, and placed between a titanium mask with a hole of 32 mm in diameter and a PTFE inner cup in a sample holder.Then, the Ka (Lb for lead and La for cadmium) peak and background intensities of the elements to be determined were measured for 100 s each by using LiF, PET and TAP analyzing crystals and scintillation and flow proportional counters as appropriate. The net intensities were calculated by subtracting the background intensities at the peak angles as well as blank values, which were measured for the blank XRF specimen.The background intensity at the peak angle was obtained by multiplying the measured intensity at the background angle by the background intensity ratio at the peak and background angles, which was previously determined using a blank XRF specimen. A Mettler (Greifensee, Switzerland) M3 microbalance was used for weighing powdered samples and activated carbon, which were taken in a small boat (12 3 8 3 5 mm) made of aluminium foil.Eppendorf (Hamburg, Germany) micropipettes (10–100 ml) were used for pipetting standard solutions. Recommended procedure for the preparation of thin-layer specimens Vehicle exhaust particulates. Portions of 4–5 and 20–30 mg of the particulates sample were placed in a 30 ml beaker for the determination of minor and trace elements, respectively. A 0.1 ml volume of ethanol and 1.5 ml of the surfactant solution were added to the beaker in that order and the particulates sample was dispersed by ultrasonication for 3 min.The suspension was transferred onto a hydrophilized PTFE membrane filter, which was fixed on a ground-glass plate by a fluororesin-coated funnel15 (35 mm inside diameter), which was prepared by coating the interior wall of a funnel for a membrane filter holder with fluoroalkylsilane. The beaker was washed twice with 0.5 ml of the surfactant solution. A schematic diagram of the preparation method is shown in Fig. 1. The suspension of the particulates was dispersed by blowing air from a pump for 5 min without bubbling, and dried by IR irradiation for 20 min without filtration. The filter with the sample was sandwiched between two PTFE rings (39 mm inside diameter). After air-drying, 0.3 ml of the polycarbonate solution was dropped onto the back side of the filter to avoid re-dispersion of the particulates and airdried. The resulting thin-layer specimens were used for the XRF analysis.A blank XRF specimen was also prepared in the same way as mentioned above, except that the sample was omitted. Atmospheric dust collected by a bag filter. A 4–8 mg portion of the dust samples was placed on a hydrophilized PTFE membrane filter, which was fixed on a ground-glass plate by a fluororesin-coated funnel. A 0.1 ml volume of ethanol and 3.0 ml of the surfactant solution were added to the dust sample in that order.Then, thin-layer specimens were prepared in a similar way to the vehicle exhaust particulates. The resulting thin-layer specimens were used for the XRF analysis. Thin-layer standard specimens for calibration. Several thin-layer standard specimens with 10.0 mg of finely powdered activated carbon were prepared by impregnating activated carbon powder with the desired amounts of standard solutions of all the elements of interest and by dispersing and drying the powder in a similar manner to the real environmental samples. Plots of net counts per second versus amounts of elements were used as calibration graphs.The amounts of elements in real environmental samples were determined from the measured XRF intensities without matrix correction. The detailed procedure for the preparation of the thin-layer standard specimens was reported in a previous paper.15 Results and discussion Determination of minor and trace elements in vehicle exhaust particulates Since the NIES No. 8 vehicle exhaust particulates sample has a marked electrostatic effect, the suppression of this effect was initially investigated. A surfactant solution, ethanol, acetone, cyclohexane and isopropyl alcohol were examined as dispersants for the preparation of homogeneous thin-layer specimens. A 1.5 ml volume of each dispersant was added to 5.0 mg of the sample, and thin-layer specimens were prepared according to the recommended procedure, but omitting the ultrasonication step. Acetone and cyclohexane could not disperse the particulates successfully owing to their high volatility.The surfactant solution caused a significant loss of the solvent owing to the electrostatic effect of the powder, but could maintain the dispersibility during drying. Ethanol and isopropyl alcohol could suppress the electrostatic effect, but caused an uneven sample surface during drying. Therefore, 0.1 ml of ethanol was added in order to suppress the electrostatic effect prior to addition of the surfactant solution.However, a satisfactory specimen could not be prepared simply by airdispersion, because some particles aggregated strongly. Therefore, pre-dispersion by ultrasonication was carried out prior to air-dispersion. About 3 min of ultrasonication in a mixture of 0.1 ml of ethanol and 1.5 ml of the surfactant solution allowed easy and rapid disaggregation to fine particles. Specimens prepared using 4–5 mg of the NIES No. 8 sample were subjected to XRF analysis.Sodium, magnesium, aluminium, potassium, calcium, zinc, silicon, sulfur, chlorine and iron were successfully detected and determined, but other trace elements in the sample were not detected, or were near to the detection limits. Therefore, specimens prepared using 20–30 mg of the sample for trace element analysis were subjected to Fig. 1 Schematic diagram of the method used for the preparation of thinlayer specimens of environmental samples. a, Fluororesin-coated funnel; b, suspension of environmental sample in a mixture of ethanol and 0.01% surfactant (Tween-85) solution; c, Omnipore-JH membrane filter (0.45 mm, 47 mm f); d, ground-glass plate. 1814 Analyst, 1998, 123, 1813–1816XRF analysis.The analytical results in Table 1 show that the proposed method can be successfully applied to the XRF determination of minor and trace elements in vehicle exhaust particulates. Although the minimum mass of test portion recommended by the supplier of the NIES No. 8 sample is 300 mg, our analytical results obtained from seven independent measurements demonstrate the homogeneity of the sample for the elements determined in this study. Determination of major and minor elements in atmospheric dust collected by a bag filter Matrix effects will be absent for XRF analysis of airborne particulate matter collected as a thin-layer on a membrane filter.15 Therefore, a method for the preparation of thin-layer specimens of atmospheric dust collected by a bag filter was investigated.Since atmospheric dust usually contains elemental carbon and organic matter,18 activated carbon was not added as a carrier. Thin-layer specimens were prepared by using 5.0 mg of AS-1 sample and the surfactant solution as a dispersant according to the recommended procedure. However, significant loss of the solvent was observed, owing to the electrostatic effect of the atmospheric dust, although to a lesser extent than with the vehicle exhaust particulates.Therefore, 0.1 ml of ethanol was added initially in order to suppress the electrostatic effect, after which the surfactant solution was added. Since 1.5–2.0 ml of the surfactant solution gave a heterogeneous sample, 3.0 ml of the dispersant were used in the recommended procedure. Fig. 2 shows the effect of the amount of AS-1 on the analytical results for light elements, which may be strongly influenced by the matrix. Analytical results were normalized, with those of the specimen prepared using 4.18 mg of AS-1 being 1.Specimens prepared using more than 10 mg gave low values for sodium, magnesium, aluminium, silicon and sulfur owing to the absorption effect. Because 2.0 mg of sample were not sufficient to form a homogeneous specimen, a specimen prepared using 2.16 mg gave somewhat lower values. Therefore, 4–8 mg were adopted as the amount of sample, which can be analysed without matrix correction and allows easy preparation of homogeneous specimens. Analytical results for AS-1 in Table 2 show that the proposed method can be successfully applied to the XRF determination of major and minor elements in atmospheric dust collected by a bag filter.The method can also be applied to other samples such as sediments. Conclusion A simple and rapid method for the preparation of thin-layer specimens has been developed for the multi-element analysis of environmental samples by XRF.The method was applied to vehicle exhaust particulates and atmospheric dust as typical environmental samples. Minor and trace elements in vehicle exhaust particulates and major, minor and trace elements in atmospheric dust were successfully determined by the proposed method. The advantages of the method are freedom from matrix correction, the ability to determine many elements in very small amounts of sample and the simplicity in preparing both thinlayer and standard specimens.Table 1 Analytical results for the vehicle exhaust particulates CRM, NIES No. 8 Contenta Element Certified valueb Proposed methodc Recovery (%) Na 0.192 ± 0.008 0.19 ± 0.01 99 ± 8 Mg 0.101 ± 0.005 0.10 ± 0.01 96 ± 9 Al 0.33 ± 0.02 0.32 ± 0.03 96 ± 8 K 0.115 ± 0.008 0.12 ± 0.01 105 ± 8 Ca 0.53 ± 0.02 0.58 ± 0.04 109 ± 7 Zn 0.104 ± 0.005 0.11 ± 0.01 102 ± 13 Si — d 1.19 ± 0.04 S — 1.90 ± 0.18 Cl — 0.052 ± 0.005 Fe — 0.43 ± 0.02 Ti — 290 ± 21 V 17 ± 2 16.4 ± 0.6 108 ± 10 Cr 25.5 ± 1.5 24.3 ± 2.3 96 ± 9 Mn — 111 ± 16 Co 3.3 ± 0.3 ND e Ni 18.5 ± 1.5 16.4 ± 0.6 89 ± 3 Cu 67 ± 3 63 ± 4 94 ± 6 Br 56 f 74 ± 6 130 ± 10 Pb 219 ± 9 220 ± 2 101 ± 1 Cd 1.1 ± 0.1 ND a Values for Na, Mg, Al, K, Ca, Zn, Si, S, Cl and Fe in %; values for Ti, V, Cr, Mn, Co, Ni, Cu, Br, Pb and Cd in mg g21.b From ref. 16. c Mean and standard deviation were calculated from seven independent measurements. d —: Not tested. e ND: Not detected. f Reference value. Fig. 2 Effect of amount of atmospheric dust sample on analytical results for light elements.Atmospheric dust sample: AS-1, -: Na, 5: Mg, :: Al, < : Si, 8: S, 2: K, ½: Ca. Analytical results were normalized, with those of samples prepared using 4.18 mg being 1. Table 2 Analytical results for the atmospheric dust CRM, AS-1 Content/mg g21 Element Literaturea Proposed methodb Na 14 ± 1 14 ± 1.1 Mg 17 ± 2 14 ± 1.1 Al 50 ± 7 70 ± 9 Si —c 150 ± 14 S — 13 ± 0.7 K 9.7 7.6 ± 0.7 Ca 56 ± 5 50 ± 5 Ti 4.2 ± 1.1 3.3 ± 0.4 V 0.23 ± 0.07 0.25 ± 0.02 Cr 0.34 ± 0.03 0.43 ± 0.07 Mn 1.2 ± 0.1 1.0 ± 0.14 Fe 45 ± 3 41 ± 7 Co 0.026 ± 0.004 0.024 ± 0.001 Ni 0.2 ± 0.03 0.14 ± 0.02 Cu 0.4 ± 0.14 0.3 ± 0.05 Zn 3.4 ± 0.5 3.7 ± 0.4 Br 0.34 ± 0.009 0.35 ± 0.07 Pb — 1.6 ± 0.2 Cd — ND d a From ref. 17. b Mean and standard deviation were calculated from four independent sample preparation and measurements. c —: Not tested. d ND: Not detected. Analyst, 1998, 123, 1813–1816 1815The authors thank Dr.Jun Yoshinaga and Professor Yoshikazu Hashimoto for the use of the vehicle exhaust particulates (NIES No. 8) and the atmospheric dust (AS-1), respectively. References 1 R. Tertian and F. Claisse, Principles of Quantitative X-ray Fluorescence Analysis, Heyden, London, 1982, pp. 317–333. 2 F. Feret and R. Jenkins, in A Practical Guide for the Preparation of Specimens for X-ray Fluorescence and X-ray Diffraction Analysis, ed. V. E. Buhrke, R. Jenkins and D. K. Smith, Wiley, New York, 1998, pp. 35–122. 3 K. Matsumoto and K. Fuwa, Anal. Chem., 1979, 51, 2355. 4 M. Guevara and S. P. Verma, X-Ray Spectrom., 1987, 16, 87. 5 K. Norrish and G. M. Thompson, X-Ray Spectrom., 1990, 19, 67. 6 Y. N. Hua and C. T. Yap, Anal. Sci., 1994, 10, 867. 7 H. M. West, J. Cawley and R. Wills, Analyst, 1995, 120, 1267. 8 T. Kitamura, S. Tanimoto, M. Iwatsuki and S. Nishida, Bunseki Kagaku, 1998, 47, 211. 9 W. K. Stankiewicz, Z. A. Mzyk and B. M. Roter, Mikrochim. Acta, 1996, 123, 137. 10 R. D. Giaugue, F. S. Goulding, J. M. Jaklevic and R. H. Pehl, Anal. Chem., 1973, 45, 671. 11 O. Haupt, B. Klaue, C. Schaefer and W. Dannecker, X-Ray Spectrom., 1995, 24, 267. 12 O. Haupt, C. Schaefer, S. Strauss and W. Dannecker, Fresenius’ J. Anal. Chem., 1996, 355, 375. 13 J. Renault and C. McKee, Analyst, 1995, 120, 1261. 14 L. H. Cohen and D. K. Smith, Anal. Chem., 1989, 61, 1837. 15 M. Iwatsuki, T. Kyotani and S. Koshimizu, Anal. Sci., 1997, 13, 807. 16 K. Okamoto, Kikan Kankyo Kenkyu, 1987, 66, 124. 17 Y. Hashimoto, T. Otoshi and K. Oikawa, Environ. Sci. Technol., 1976, 10, 815. 18 A. Mizohata, Y. Matsuda, K. Sakamoto and S. Kadowaki, J. Jpn. Soc. Air Pollut., 1986, 21, 83. Paper 8/03706J 1816 Analyst, 1998, 123, 1813–1816

ISSN:0003-2654    

DOI:10.1039/a803706j

出版商:RSC    

年代:1998

数据来源: RSC

摘要:

Vapour phase Fourier transform infrared spectrometric determination of carbonate in sediments Amparo P�erez-Ponce, Jose M. Garrigues, Salvador Garrigues and Miguel de la Guardia* Department of Analytical Chemistry, University of Valencia, 50 Dr. Moliner St., 46100 Burjasot, Valencia, Spain Received 24th April 1998, Accepted 7th July 1998 A rapid, sensitive and direct procedure for the determination of carbonate in sediments based on vapour-generation Fourier transform infrared spectrometry is described.A 1 ml volume of hydrochloric acid (0.25 m) was injected into a vessel, heated at 40 °C, containing 10 mg of sediment. The CO2 evolved under these conditions was swept by a stream of nitrogen to an infrared gas cell. The flow injection (FI) recordings were registered as a function of time and the areas of the FI recording obtained in the wavenumber range 2500–2150 cm21 were measured and interpolated in a calibration equation obtained from known amounts of CaCO3 treated in the same way as the samples. The method provided a limit of detection of 0.2 mg of CaCO3, a sampling frequency of 20 h21 and an RSD of 0.7% for three independent analyses of 20 mg of sediment.Results obtained for a series of natural sediment samples compared well with those obtained using a back-titrimetric reference method. Inorganic carbonate exists in soils mainly as the sparingly soluble alkaline earth metal carbonates calcite (CaCO3) and dolomite (CaCO3·MgCO3).Calcite is commonly the main component. Soluble inorganic carbonates also exist in soils, primarily in highly sodic (alkali metal) soils of arid regions, but their content is generally low compared with the alkaline earth metal carbonates,1 and for this reason the content of inorganic carbonate in sediments is referred to the percentage of CaCO3.2 The determination of carbonate in sediments can be achieved in two ways. The preferred technique is the direct measurement of inorganic carbon.3–14 However, it is also possible to determine carbonates by the difference between total carbon, in untreated samples, and the organic carbon measured in samples previously acidified in order to remove the inorganic carbon. 15,16 Methods involving the determination of total inorganic carbonate include (i) dissolution of carbonate in acid and determination of the evolved CO2 by volume3 or pressure4 measurements of CO2 or by thermal conductivity5,6 and (ii) dry ashing of the sediment in oxygen at temperatures higher than 1000 °C and thermogravimetric determination of CO2 by using a thermal conductivity detector8 or by IR detection9,10 or possibly continuous titration of CO2 absorbed in water.11–14 In general, these techniques involve the use of specific instrumentation devoted to this kind of determination and, additionally, these methods are expensive and time consuming.In this work, based on the use of common FTIR instrumentation, we developed a simple and rapid procedure for the determination of carbonate in sediments based on previous experience with the flow injection analysis–FTIR determination of carbonate in waters through on-line acid treatment and CO2 measurement in the vapour phase generated by convective17 or microwave-assisted18 heating.Experimental Apparatus and reagents A Magna 750 FTIR spectrometer from Nicolet (Madison, WI, USA), equipped with a temperature-stabilized DGTS detector, a long-lasting Ever-Glo source and a KBr beamsplitter, was employed for spectral measurements obtained by accumulating three scans at a resolution of 16 cm21, using a laboratory-made gas cell made of PTFE with an internal volume of 490 ml and a bandpass of 39 mm, equipped with ZnSe windows. Version 2.1 of Omnic software, developed by Nicolet, was employed to control the instrument, for data acquisition and also for processing the analytical results. The manifold employed for vapour-generation FTIR measurements (Fig. 1) was a single-channel manifold with a nitrogen carrier flow, which included a removable glass sample vessel of 13 ml internal volume with a gas inlet, a gas outlet and a septum. An appropriate acid volume was injected inside the vessel through the septum using Hamilton (Reno, NV, USA) microsyringes of different volumes. Sample vessels were introduced inside a hot water-bath, the temperature of which was controlled by means of a thermocouple and operated using a laboratory-made electrically controlled heater.The temperature of the N2 flow was measured with another thermocouple and fixed using an electrically controlled heater from AKO (Barcelona, Spain). A ganged distribution valve from Omnifit (Manchester, UK) was used as a by-pass system when sample vessels were Fig. 1 Manifold employed for vapour-generation FTIR determination of carbonate in sediments. Analyst, 1998, 123, 1817–1821 1817changed, in order to ensure a permanent N2 flow through the IR cell.All tubes employed in the manifold were made of PTFE of 1/32 in id and 1/16 in od. To avoid the presence of water drops inside the measurement cell, a 1.6 ml piece of glass was installed at the exit of the glass reactor to act as a water trap in which the excess water vapour condensed before passing inside the gas cell. Analytical-reagent grade CaCO3 from Panreac (Barcelona, Spain) was employed for the preparation of standard solutions and 37% HCl from Fluka (Buchs, Switzerland) was used for sample and standard solution acidification.Nitrogen C-45 (99.995% purity) from Carburos Metálicos (Barcelona, Spain) was employed as the carrier gas and de-ionized water with a resistance > 10 MW obtained using a Milli-RX system from Millipore (Molsheim, France) was used throughout. Vapour phase FTIR analysis Amounts of 20 mg of samples were accurately weighed into the glass vessels that were employed as sample vessels and incorporated in the manifold depicted in Fig. 1.Prior to any measurement, an N2 carrier flow was passed through an empty vessel and the corresponding background was obtained. The empty vessel was then removed by by-passing the N2 flow using the ganged distribution valve, and a new vessel containing sample or standard solution was placed inside the water-bath at 40 °C; the N2 flow was passed through the vessel again and 1 ml of 0.25 m HCl was injected into the vessel and the CO2 generated was transported inside the gas cell of the FTIR spectrometer using a nitrogen carrier flow of 50 ml min21, without the need for an equilibrium time before measurement.The FTIR spectra were continuously recorded, as a function of time, between 3000 and 1000 cm21, which includes the CO2 absorption band at 2350 cm21. From the absorbance data in the 2500–2150 cm21 range, integrated absorbance measurements were obtained, providing transient recordings which correspond to the absorbance of CO2, in the selected wavenumber range.The areas of the aforementioned peaks, obtained with a baseline correction established between two sequential injections, were interpolated in the calibration line obtained with peaks corresponding to known amounts of CaCO3 treated in the same way as the samples. Reference back-titrimetric procedure The method recommended by the American Society of Agronomy1 is based on the neutralization of carbonates with H2SO4 and back-titration of the excess of acid.To 0.1 g of sediment, 9 ml of 1% H2SO4 and 40 ml of de-ionized water were added and the mixture was shaken and heated until boiling to eliminate the H2CO3 formed. After cooling, a few drops of phenolphthalein were added and the solution was titrated with 0.1m NaOH, adding a drop of titrant every 2–3 s until the appearance of a pink colour. The percentage of CaCO3 in sediment was calculated using the expression CaCO m/m NaOH sed 3 16 9 5 (% ) .= - VM m where V is to the reading volume, MNaOH the NaOH molar concentration and msed the mass of sediment. Results and discussion FTIR spectra of sediments in solid and vapour phase It is known that carbonate provides two sensitive and well defined bands corresponding to the C–O vibration, one at 1400 cm21, with strong intensity, and another at 870 cm21, with medium intensity. Both bands suffer sll shifts depending on the cation present in carbonate and large shifts when CO3 22 changes to HCO32 as a function of pH [Fig. 2(A)]. Furthermore, natural soil and sediment samples contain a mixture of carbonates and hydrogencarbonates with different cations, and other inorganic anions that show absorption in this spectral Fig. 2 Solid and vapour phase FTIR spectra of carbonate and hydrogencarbonate. (A) Solid phase spectra obtained from KBr pellets of (a) sodium hydrogencarbonate, (b) sodium carbonate, (c) calcium carbonate, (d) magnesium carbonate and (e) a sediment sample.Experimental conditions: spectra obtained from 50 accumulated scans at a resolution of 16 cm21 establishing the background with air. (B) Vapour phase co-added spectra obtained from 2 min recording obtained for (a) sodium hydrogencarbonate, (b) sodium carbonate, (c) calcium carbonate, (d) magnesium carbonate and (e) a sediment sample. Experimental conditions: spectra obtained from three accumulated scans at a resolution of 16 cm21, using a gas cell of 39 mm pathlength with ZnSe windows, for a solid mass equivalent to 4 mg of CaCO3, an injection volume of 1 ml of 0.25 m HCl using 13 ml reactors heated at 40 °C and a 50 ml min21 N2 carrier gas flow rate.The spectral background was obtained from an N2 flow passed through an empty vessel. The co-added spectra were obtained during 2 min from the starting point of the transient signals obtained after the injection of HCl into the sample reactors installed in the manifold. 1818 Analyst, 1998, 123, 1817–1821region, such as silicates. All this makes it difficult to determine the total carbonate content by direct measurement of IR absorbance in solids. Additionally, quantitative analysis using alkali metal halide pellets needs the use of internal references or rigorous control of the bandpass. When carbonate standards and natural sediment samples were treated with acid and the spectra of CO2 generated by decomposition of carbonates were recordered in the range 3000–1000 cm21, a sensitive double band was obtained between 2400 and 2300 cm21 that corresponds to the asymmetric vibration of CO2 [Fig. 2(B)]. In this spectral range, water shows a very low absorption and for this reason the use of vapour phase generation of CO2 from carbonate in soils and sediments provides a simple and sensitive approach for the determination of carbonate in natural samples. Effect of HCl on CO2 generation The most important step in the determination of carbonate by vapour phase FTIR spectrometry is the acidification of sediments in order to generate CO2, so the concentration and volume of HCl necessary for a complete generation of CO2 from CaCO3 was evaluated. Table 1 shows the effect of HCl concentration and volume on the area of the integrated absorbance signals obtained for 10 mg of CaCO3, analysis being carried out in a monoparametric mode with a volume of 1 ml and a concentration of 0.25 m for natural sample analysis.In theory, 10 mg of CaCO3 needs 2 3 1024 mol of HCl, i.e.,1 ml of 0.2 m HCl or 0.8 ml of 0.25 m HCl, to be converted quantitatively into CO2. The results in Table 1 demonstrate that, in this method, rapid and complete CO2 formation only needs a stoichiometric amount of HCl. Therefore, 1 ml of 0.25 m HCl was selected for the treatment of 10 mg of CaCO3 or a natural sediment, although for sediment a slight excess was used. However the aforementioned amount of acid was taken as a general criterion without observing deleterious effects on the analytical signals due to the presence of a small excess of acid.Effect of reactor temperature The reaction between carbonate and HCl and the release of CO2 are very efficient and rapid processes and the effect of temperature is not a decisive factor with regard to thermodynamic aspects, but it affects the kinetics of the reaction. As can be seen in the inset in Fig. 3, the peak width decreases and the peak height increases when the reactor temperature increases, thus improving the reaction speed.However, from the experimental point of view, the extremely fast measurement of CO2 creates problems with respect to monitoring absorption spectra and the peak area decreases at reactor temperatures higher than 60 °C; therefore a temperature of 40 °C was selected in order to obtain a compromise between speed, sensitivity and reproducibility of the analytical signals. Effect of reactor volume As can be seen in Table 2, an increase in the sample reactor volume, which increases the CO2 dispersion by increasing the total volume of the system, has a very small influence on the analytical signals obtained from the areas of the transient peaks.A reactor volume of 13 ml, which corresponds to standard glass vials, can be recommended for the determination of carbonate in sediments. The only problem created by the use of small reactors is the transport of water vapour inside the measurement cell, but this can be solved easily by incorporating a water trap consisting of a 1.6 ml glass cell which helps to condense the water before the measurement cell.Effect of volume of water added to samples To prevent acid splashing and to improve the homogeneous mixture between solid samples and acid, different volumes of water, from 1 to 5 ml, were added to the solid samples before acidification. As can be seen in Table 3, the volume of water used does not seriously affect the peak areas corresponding to 10 mg of CaCO3 or those of a sediment treated following the general procedure.However, the addition of water decreases the reaction speed and consequently the frequency of analysis. Table 1 Effect of HCl concentration and volume on the vapour phase FTIR determination of carbonate in sediments. Experimental conditions: in both series of measurements we employed 10 mg of CaCO3, 13 ml reactor volume and 60 °C reactor temperature, using 120 and 100 ml min21 N2 flow rates for HCl concentration effect and HCl volume effect, respectively.Area values, expressed in arbitrary units, were measured from the absorbance in the wavenumber range 2500–2150 cm21 and the standard deviation values correspond to three independent measurements HCl HCl concentration/m volume/ml (V = 1 ml) Area ± sn21 * (C = 0.25 m) Area ± sn21 * 0.125 17.2 ± 0.8 0.5 20.63 ± 0.10 0.25 22.35 ± 0.21 0.75 26.9 ± 0.7 0.5 21.8 ± 0.7 1 26.7 ± 0.8 1 22.6 ± 1.6 1.5 26.9 ± 0.7 1.5 23.7 ± 1.4 2 27.6 ± 0.4 2 23.6 ± 0.9 5 22.7 ± 1.2 * Sn21: Standard deviation for n 2 1 degrees of freedom.Fig. 3 Effect of the reactor temperature on vapour-generation FTIR determination of carbonate in sediments. Inset: peaks corresponding to the wavenumber range 2500–2150 cm21, obtained for 10 mg of CaCO3 at different reactor temperatures. Experimental conditions: 10 mg of CaCO3, 1 ml of 0.25 m HCl, 13 ml reactor volume and 120 ml min21 N2 flow rate.Areas, expressed in arbitrary units, were measured from the absorbance in the range 2500–2150 cm21. Error bars indicate the interval of the average value ± the corresponding standard deviation of three independent measurements. Table 2 Effect of reactor volume on vapour phase FTIR determination of carbonate in sediments. Experimental conditions: 10 mg of CaCO3, 1 ml of 0.25 m HCl, 40 °C reactor temperature and 120 ml min21 N2 flow rate. Areas, expressed in arbitrary units, were measured from the absorbance in the wavenumber range 2500–2150 cm21 and standard deviation values correspond to three independent measurements Reactor volume/ml Area ± sn21 5.3 17.8 ± 0.9 6.7 16.8 ± 2.0 12.8 19.4 ± 1.1 13.2 19.4 ± 1.1 21.5 19.3 ± 1.6 28 20.5 ± 1.2 Analyst, 1998, 123, 1817–1821 1819Because of this, it is preferable to add HCl directly to the solid in order to improve the sampling frequency.Effect of sample mass The effect of sediment mass on the peak areas of the transient signals of evolved CO2 was evaluated in order to ensure the appropriate sensitivity and precision of analysis by vapour phase FTIR spectrometry.For these experiments, a 3 ml volume of 0.4 m HCl was fixed, which provides a stoichiometric amount of acid to provide the total neutralization of 60 mg of pure CaCO3. The results Table 4 indicate that on increasing the sediment mass, the peak area also increases without changing the sampling frequency, but the reproducibility of the signals corresponding to a high sample mass decreases.A typical regression line, A = (3.1 ± 1.0) + (2.44 ± 0.03)m, where A is peak area and m the sample mass (mg), with a regression coefficient r = 0.9998, was obtained for increasing sediment mass. This indicates that the system works correctly for different sediment masses, thus providing a good means of improving the analytical sensitivity in the analysis of samples containing a very low carbonate concentration.Effect of N2 carrier flow rate The carrier gas flow rate is a critical parameter in vapour phase FTIR spectrometric analyses17–20 which affects the analytical sensitivity and the sampling frequency. This parameter affects the volatility of the analytes and controls the speed of vapour introduction into the measurement cell. An increase in the carrier flow rate causes a decrease in sensitivity (Fig. 4), but increases the sampling frequency, as can be seen in the inset, from 10.9 h21 for 11 ml min21 to 30 h21 for 100 ml min21.Hence a nitrogen flow rate of 50 ml min21, which allows a 20 h21 sampling frequency, was selected in order to achieve a compromise between analytical sensitivity and sample throughput. Use of a fixed reaction time A series of additional experiments were carried out by isolating the reaction vessel from the N2 carrier flow, after the addition of HCl and introducing the CO2 into the carrier stream after a fixed reaction time. Times between 0.6 and 2 min were tried and, as can be seen in the inset in Fig. 5, an increase in the reaction time leads to an increase in the peak height and a decrease in the peak width.However, there is no variation of the peak area, hence the analytical sensitivity remains constant or is slightly lowered, probably owing to small leaks of CO2 through the manifold connections. Therefore, in order to obtain as reproducible data as possible, it is recommended to work in the continuous flow mode after the injection of HCl using a constant flow of N2 to introduce the evolved CO2 continuously into the measurement cell. Table 3.Effect of water volume added to samples, before acidification, on vapour phase FTIR determination of carbonate in sediments. Experimental conditions: 10 mg of CaCO3 or sediment, 1 ml of 0.25 m HCl, 13 ml reactor volume, 40 °C reactor temperature and 120 ml min21 N2 flow rate. Areas, expressed in arbitrary units, were measured from the absorbance in the wavenumber range 2500–2150 cm21 and the standard deviation values correspond to three independent measurements. Sampling frequency was established from the peak width CaCO3 Sediment Sampling Sampling Water Area frequency/ Area frequency/ volume/ml ± sn21 h21 ± sn21 h21 0 19.3 ± 0.7 40 8.1 ± 0.3 46 5 19.6 ± 0.3 20 7.7 ± 0.4 30 1 20.9 ± 0.3 20 8.5 ± 0.3 24 2 21.4 ± 0.3 20 7.8 ± 0.1 20 3 20.9 ± 0.4 20 8.2 ± 0.3 20 Table 4 Effect of sample mass on vapour phase FTIR determination of carbonate in sediments.Experimental conditions: 3 ml 0.4 M HCl, 13 ml reactor volume, 40 °C reactor temperature and 40 ml min21 N2 flow rate. Areas, expressed in arbitrary units, were measured from the absorbance in the wavenumber range 2500–2150 cm21 and the standard deviation values correspond to three independent measurements Sample Sampling mass/mg Area ± sn21 frequency/h21 5 15.7 ± 0.8 20 10 27.8 ± 0.5 20 20 52.3 ± 0.7 20 40 99.0 ± 1.3 20 60 151 ± 4 20 Fig. 4 Effect of the N2 carrier flow rate on vapour phase FTIR carbonate determination. Inset: peaks corresponding to the wavenumber range 2500–2150 cm21 obtained for 10 mg of sediment using different carrier gas flow rates. Experimental conditions: 10 mg of sediment sample, 1 ml of 0.25 m HCl, 13 ml reactor volume and 40 °C reactor temperature. Areas, expressed in arbitrary units, were measured from the absorbance in the range 2500–2150 cm21 and the standard deviation values were obtained from three independent measurements.Fig. 5 Effect of the reaction time on vapour phase FTIR determination of carbonate. Inset: each peak corresponds to the measurement of CO2 evolved from a separate sample of 10 mg of CaCO3. Arrows indicate the injection of HCl. The assays were carried out on different CaCO3 portions weighed in different glass vials, which were installed separately in the manifold. Experimental conditions: 10 mg of CaCO3, 1 ml of 0.25 m HCl, 13 ml reactor volume, 40 °C reactor temperature and 32 ml min21 N2 flow rate.Areas, expressed in arbitrary units, were measured from the absorbance in the range 2500–2150 cm21 and the standard deviation values correspond to three independent measurements. 1820 Analyst, 1998, 123, 1817–1821Determination of carbonate in natural samples The developed vapour-generation FTIR procedure was employed to analyse six natural sediments and two sewage sludge samples, containing different carbonate concentrations from 23 to 78% m/m (expressed as calcium carbonate).The values obtained were compared with those found by a back-titrimetric reference method,1 and results are summarized in Table 5, which gives the average values ± the standard deviation corresponding to three independent analyses of each sample. The regression between values found by the developed procedure (y) and those obtained by the reference method (x) provided the regression equation y = (22.6 ± 1.9) + (1.01 ± 0.04)x, with a regression coefficient r = 0.996.This regression clearly indicates that the developed procedure does not require any blank correction, because the intercept of this line is statistically comparable to zero,21 and it does not have constant relative errors, the slope being statistically equal to unity. Analytical figures of merit The main analytical characteristics of the method were established from typical calibration lines, obtained under the optimum experimental conditions and from the analysis of natural samples.A typical expression for a calibration obtained up to 16 mg of CaCO3 is A = (21.4 ± 1.4) + (5.95 ± 0.14) m, with r = 0.996, where A is the area of the transient peaks and m the mass of CaCO3 (mg), and from this, and taking into consideration the repeatability of blank measurements, a limit of detection of 0.2 mg of CaCO3 can be established for a probability level of 99.6%.The precision of carbonate determination in natural samples of sediments can be established from the relative standard deviation (RSD) of three independent analyses and, as can be seen from the data in Table 5, an average RSD of < 2% was obtained. The sampling frequency of the method is 15 h21, including the time required for the installation of glass sample vials in the manifold. Conclusion This work confirmes the applicability of the vapour phase generation methodology for the FTIR analysis of solid samples without the need for a preliminary treatment of samples and offers a good alternative for the direct determination of total carbonate in sediment samples.The authors acknowledge financial support from the DGES, Project No. PB96-0779, and the Generalitat Valenciana, Project GV 3218/95. Amparo Pérez-Ponce acknowledges a grant from the Conselleria de Cultura, Educación y Ciencia de la Generalitat Valenciana. References 1 Nelson, R.E., Methods of Soil Analysis, Part 2, American Society of Agronomy, Madison, WI, 1982, p. 182. 2 Primo Yufera, E., and Carrasco Dorrién, J. M., Química Agraria, Part 1, Alhambra, Madrid, 1973, p. 348. 3 British Standards Institution, British Standard, BS 7755:Section 3.10, 1995 [ISO10693, 1995], 12. 4 Jones, G. A., and Kaiteris, P., J. Sediment. Petrol., 1983, 53, 655. 5 Amundson, R. G., Trask, J., and Pendall, E., Soil Sci. Soc. Am. J., 1988, 52, 880. 6 Weliky, K., Suess, E., Ungerer, C.A., Muller, P. J., and Fischer, K., Limnol. Oceanogr., 1983, 28, 1252. 7 Barker, J. F., and Chatten, S., Chem. Geol., 1982, 36, 317. 8 Krom, M. D., and Berner, R. A., J. Sediment. Petrol., 1983, 53, 660. 9 Birkelbach, M., and Ohls, K., GIT Fachz. Lab., 1995, 39, 1125. 10 Schwartz. V., Fresenius’ J. Anal. Chem., 1995, 351, 629. 11 Paulik, F., and Arnold, M., J. Thermal Anal., 1993, 39, 1079. 12 Paulik, J., Paulik, F., and Arnold, M., J. Thermal Anal., 1984, 29, 345. 13 Paulik, F., Paulik, J., and Arnold, M., J. Thermal Anal., 1984, 29, 333. 14 Paulik, J., Paulik, F., and Arnold, M., J. Thermal Anal., 1982, 25, 327. 15 Neal, R. H., and Younglove, T., Commun. Soil Sci. Plant Anal., 1993, 24, 2733. 16 Snyder, J. D., and Trofymow, J. A., Commun. Soil Sci. Plant Anal., 1984, 15, 587. 17 Pérez-Ponce, A., Garrigues, S., and de la Guardia, M., Vibr. Spectrosc., 1998, 16, 61. 10 Pérez-Ponce, A., Garrigues, S., and de la Guardia, M., Anal. Chim. Acta, 1998, 358, 235. 19 López-Anreus, E., Garrigues, S., and de la Guardia, M., Anal. Chim. Acta, 1996, 333, 157. 20 Pérez-Ponce, A., Garrigues, S., and de la Guardia, M., Analyst, 1996, 121, 923. 21 de la Guardia, M., Salvador, A., and Berenguer, V., An. Quím. Ser. B, 1981, 77, 129. Paper 8/03098G Table 5 Results obtained for the determination of carbonate in sediment samples by vapour phase FTIR spectrometry and by a reference titrimetric method.1 Experimental conditions: 20 mg of sample, 1 ml of 0.4 m HCl, 40 °C reactor temperature, 13 ml reactor volume and 50 ml min21 N2 flow rate. Carbonate (% m/m CaCO3) Sample Found ± sn21 Reference ± sn21 Sediment 1 34.4 ± 0.5 38.6 ± 1.0 Sediment 2 33.8 ± 1.3 36.0 ± 1.9 Sediment 3 50.9 ± 0.5 50.6 ± 1.2 Sediment 4 76.1 ± 3.0 78.0 ± 1.2 Sediment 5 22.7 ± 0.4 23.4 ± 0.6 Sediment 6 33.6 ± 0.3 37.5 ± 0.7 Sewage sludge 1 52.2 ± 1.0 53.1 ± 1.4 Sewage sludge 2 48.9 ± 0.8 51.4 ± 0.9 Analyst, 1998, 123, 1817–1821 1821

ISSN:0003-2654    

DOI:10.1039/a803098g

出版商:RSC    

年代:1998

数据来源: RSC

摘要:

The in-situ analysis of lipsticks by surface enhanced resonance Raman scattering C. Rodger, V. Rutherford, D. Broughton, P. C. White* and W. E. Smith Department of Pure and Applied Chemistry, University of Strathclyde, Glasgow, UK G1 1XL Received 7th July 1998, Accepted 23rd July 1998 The use of surface enhanced resonance Raman scattering (SERRS) spectroscopy is reported for the in-situ characterization of chromophores in lipstick smears on glass and cotton surfaces. A surfactant is required to obtain SERRS spectra of the dyes and pigments in these waxy samples.Of the surfactants tested, poly(l-lysine) is preferred for this purpose and serves a dual function, since it also produces the required aggregation of the silver colloid. The method is quick, effective and sensitive, and with the silver colloid distributed on the surfaces tested, no appreciable background fluorescence from the substrates is detected. For six commercial lipstick samples examined by this in-situ SERRS method, discrimination between the samples could be achieved and it was possible to identify some of the individual pigments present, thus indicating the potential of the technique for forensic and quality control applications.Introduction A common problem in forensic science is to identify and establish the provenance of trace samples of lipstick smears deposited on a variety of surfaces including glass, paper, cigarette butts and garments. In addition, the manufacturer of lipstick requires methods which can provide quick and effective sample identification for quality control.A lipstick typically consists of 65% castor oil, 15% beeswax, 10% carnuba wax, 5% lanolin, a number of soluble and insoluble dyes, pigments and perfume.1 Standard methods of analysis either involve an assessment of the perceived colour by microscopy and microspectrophotometry, 2,3 or separation techniques such as thinlayer and high-pressure liquid chromatography.3,4 However these techniques are not entirely satisfactory since they are insensitive and either involve human opinion, or require a complicated extraction procedure to isolate the dyes and pigments from the waxy matrix.Contamination or incomplete extraction can result due to the insolubility of any pigments and the nature of the waxy matrix. Additionally, during the extraction process the sample is susceptible to dissolution, modification, evaporation and absorption of contaminants.Therefore, a technique which does not require an extraction step or the use of solvents would be of value. Modern Raman spectroscopy is now simple and effective but fluorescence from both the matrix and the chromophores in lipsticks limits the use of resonance Raman scattering to determine the chromophore mix selectively and in-situ. However, by using the SERRS technique, fluorescence is usually quenched and therefore, should overcome these problems. SERRS is obtained when a molecule with a chromophore is adsorbed onto or is in close proximity to a suitable metal surface and the excitation wavelength is tuned to the molecular resonance frequency of the analyte.The enhancement obtained is very much greater than with either resonance or surface enhancement alone and the spectra obtained are unique. Since a SERRS spectrum is characteristic of the molecule the technique has been used to discriminate between dyes and identify dyes in mixtures, even when the dyes have very similar chemical structures.5,6 Recently, an in-situ SERRS method has been reported for the detection of a reactive dye covalently bound to cotton, whereby a fibre was treated with colloid.7 Since the scattering from the chromophores is much stronger than that from any other component on or close to the colloid surface, and SERRS can discriminate mixtures of structurally similar compounds, an insitu determination of the chromophores used in lipsticks without any separation procedures was considered to be feasible.To confirm this an in-situ method was devised to enable SERRS detection of the colourants in lipstick smears on both glass and cotton surfaces. SERRS analyses were also performed on dye extracts from lipsticks and the results obtained from these studies are presented and discussed. Results generated from the analysis of five other lipsticks are also included to illustrate potential forensic and quality control applications of this in-situ technique.Experimental Raman scattering was recorded using the Renishaw 2000 Micro Probe Raman Spectrometer and a modified Cary 81 system described previously.8 The Renishaw system was used to study solid samples of each lipstick. An argon laser with an excitation wavelength of 514.5 nm and power output of approximately 20 mW was used to irradiate samples and the Raman scattering was collected using ten, ten second accumulations. Six different areas were analysed from each sample.Any fluorescent background was subtracted using the standard background subtraction programme provided with the software. The Cary 81 system was used to collect the spectra from lipstick extracts contained in a cuvette with a pathlength of 1 cm. The slit width and amplification were set to ‘4’ and ‘300 K’ respectively and an argon laser (100 mW) was used to provide the excitation wavelength of 514.5 nm. Silver colloid was prepared using a modified Lee and Meisel procedure.9,10 The nature of this colloid varies from laboratory to laboratory.In this laboratory, the colloid is almost mono disperse and consists of hexagonal particles with a longest dimension of 36 nm. Quality control is obtained by testing every batch using UV/VIS spectroscopy. An absorption maximum between 404 and 410 nm with a half width of less than 60 nm is required. This colloid is stable and usable for a period of at least six months.For in-situ SERRS analyses the aqueous colloid was concentrated by centrifuging an aliquot of colloid (15 ml) at 2750 rpm for 30 min, then removing 98% of the supernatant and resuspending the colloidal silver in the Analyst, 1998, 123, 1823–1826 1823remainder of the supernatant to give approximately 300 ml of the concentrated silver colloid solution. The lipstick samples examined in this study included; Yardley ‘28’ Holly Red, Outdoor Girl ‘37’ Cocktail Cherry, Outdoor Girl ‘64’ Summer Fruits and three samples (A, B and C), which were supplied by Boots (Airdrie, Scotland).For the in-situ analyses, lipstick samples were smeared onto glass microscope slides or samples of cotton (1 cm 31 cm). The smears were treated with approximately 70 ml of an aqueous 0.01% (v/v) solution of the surfactant poly(l-lysine) hydrobromide (Sigma, Poole, Dorset, UK). Concentrated colloid (300 ml) was added directly to the surfactant treated surface and allowed to dry out naturally.SERRS spectra were collected directly from six separate areas of the surface on each sample using the microscope to focus on a specific area. Other surfactants examined included aqueous and ethanolic solutions of hexamine at a concentration of 1024 m and an aqueous solution of dodecylamine hydrochloride (1024 m). These surfactants were obtained from BDH, Poole, Dorset, UK. To prepare extracts of the lipsticks, a sample was smeared on the inside of a beaker into which 10 ml of ethanol was added.Extraction of the dyes and pigments occurred upon vigorous shaking. No attempt was made to determine the concentration of the lipstick dyes and pigments and consequently the method remains qualitative. SERRS was obtained by adding a small aliquot of the extract to the aqueous silver colloid which had been aggregated with an aqueous poly(l-lysine) solution [0.01% (v/v)]. Ultimately, the size of the aliquot required depended upon the concentration of the dyes and pigments in the extract.In these particular studies, where the concentrations were unknown a standard experimental procedure was adopted. With an aliquot of typically 150 ml the volumes of colloid and poly(l-lysine) used were 2 ml and 150 ml respectively. Results and discussion All the lipsticks studied were red in colour and observation of the smears with the microprobe revealed that they contained a very fine distribution of red particles with a few larger aggregates randomly distributed throughout.Resonance Raman scattering could not be obtained readily from these smears or from the extracts due to strong fluorescence. However, using the procedures discussed below, fluorescence quenching was achieved and good SERRS was obtained either by addition of colloid onto the surface of a smear or into a solution of an extract. Initially, SERRS from the surface of the smear was obtained by painting the concentrated silver colloid suspension onto the surface and allowing it to dry at ambient temperature. Since the colloidal particles are charged and surrounded by a polar layer their adherence to the waxy surface of the lipstick is poor and, although good scattering could be obtained by focusing the microprobe on one of the small aggregates of silver on the surface, it was difficult to obtain reproducible results.To overcome this problem, the surface was treated with a surfactant before the colloid was introduced.Using ‘Holly Red’ lipstick smeared on both glass and cotton as the test samples, three surfactants were assessed to identify which of these were the most effective in achieving adherence between the lipstick and colloid. The surfactants tested included poly(l-lysine) hydrobromide, hexamine, and dodecylamine hydrochloride, but some experimental problems were encountered in their application. With the exception of hexamine prepared in ethanol, there was a tendency for the aqueous solutions of the surfactants to roll off the waxy sample.Consequently, great care was required during the application to achieve reproducible coverage. Observation of the coated samples under the microscope indicated areas where the colloid had covered the sample completely and uniformly. These areas appeared grey and shiny and SERRS signals collected from them were strong with little or no fluorescence. Areas of incomplete colloid coverage produced poor SERRS and strong fluorescent backgrounds. In addition, the absolute intensity of the SERRS was reproducible from any one spot but varied from spot to spot.However, the relative intensities of the peaks were approximately constant from spot to spot. The effects of using each surfactant on the SERRS signals generated from ‘Holly Red’ samples of lipstick on glass and cotton surfaces are shown in Fig. 1 and 2. All of the surfactants produced good quality SERRS signals but poly(l-lysine) and dodecylamine, when applied under the conditions as described in Experimental, were found to be the most effective.Hexamine was applied both as an aqueous solution and an ethanolic solution. By using the latter, an improvement in the surface wetting of the sample was expected with an enhancement of the SERRS signals, but only minor improvements were obtained. Poly(l-lysine) was therefore selected as the surfactant to be used for further studies because of its effectiveness and that it has been used successfully on previous occasions as an aggregating agent for solution and in-situ SERRS spectroscopic studies.5,7 SERRS from the ethanol extract of ‘Holly Red’ (Fig. 3) were collected using a Cary 81 instrument fitted with an argon laser producing an excitation wavelength of 514.5 nm (power 100 mW). This instrument was used because it was equipped to measure solution or suspension Raman scattering quantitatively. Poly(l-lysine) was selected as the aggregating agent because it enables SERRS to be obtained from a wide range of dyes, irrespective of the charge on the dye.Many other aggregants are unsuitable because they tend to discriminate against some analytes, particularly if the analyte has the same charge as the silver surface. Under these conditions the lipstick extract gave acceptable SERRS, although a fluorescence background was observed. Fig. 1 Comparison of in-situ SERRS of ‘Holly Red’ on glass using four different surfactants collected using 514.5 nm excitation. 1824 Analyst, 1998, 123, 1823–1826Provided the silver colloid surface is correctly treated so as to attract the dyes, the concentration in the extract can be low, with concentrations of 1028 m or lower capable of being analysed routinely.8,11,12 The fluorescence quenching observed in SERRS requires surface adhesion of the analyte normally at monolayer coverage or below, therefore the fluorescent background observed suggested that some of the chromophores did not adhere.A modern system such as the Renishaw (Renishaw plc., Wotton-under-Edge, Gloucestershire, UK) equipped with a solution cell could comfortably remove this degree of background thus allowing the analysis of sample extracts. However, this approach was considered unsuitable because of the problem in obtaining complete extraction of all the chromophores from the sample. Nonetheless, six replicate analyses from the ‘Holly Red’ sample did produce spectra with identical peak positions and relative intensities and reasonably reproducible absolute intensities at low concentration.With further development of this method it may be possible to provide a simple way of detecting products from more conventional assays for which separation steps are a prerequisite, and in particular, in more routine analyses where the dye compositions are better defined. Examples of the SERRS spectra obtained from ‘Holy Red’ lipstick extracted from cotton and glass surfaces are shown in Fig. 4. From these results it can be observed that there are some additional peaks in the in-situ spectra that may represent an insoluble component not extracted by the ethanol. The ability of SERRS to provide a good molecular fingerprint for dyes and to discriminate between mixtures of closely related chromophores has been demonstrated for mixtures of four or five azo dyes.6 Therefore discrimination between the dyes and pigments used in lipsticks should be possible without separation of the components. A comparison with data from previous SERRS results indicate that the colourants present in the lipstick are typical of the rhodamine class of dyes or pigments.8,11,13 The in-situ studies from the cotton and glass samples gave spectra with very similar relative intensities of the SERRS signals, thus indicating no specific surface effects which Fig. 2 Comparison of in-situ SERRS of ‘Holly Red’ on glass using four different surfactants collected using 514.5 nm excitation.Fig. 3 Solution SERRS from an ethanol extract of ‘Holly Red’ lipstick collected using 514.5 nm excitation. Fig. 4 In-situ SERRS from ‘Holly Red’ lipstick smeared on cotton and glass collected using 514.5 nm excitation. Fig. 5 In-situ SERRS collected from ‘Cherry Cocktail’ and ‘Summer Fruits’ lipstick samples smeared on glass and cotton using 514.5 nm excitation. Analyst, 1998, 123, 1823–1826 1825prevent comparison or recognition of the chromophores on these different surfaces.Some variations of the intensities were observed when these results were compared with the spectra obtained for the extracted samples but the frequencies of the main peaks remained the same. This variation in intensities can possibly be attributed to the degree of control on the angle of the dye to the silver colloid surface under the solution and in-situ conditions. In a recent study of rhodamine dyes in solution it was reported that the nature of the chemisorption process and, in particular the angle the dye subtends from the surface, can affect the relative intensity of the SERRS signals.14 Under the in-situ conditions, the matrix is more likely to control the angle of the dye but some variation in intensity of the scattered light would still be expected because there would still be a random distribution of angles between the dye and the surface of the silver colloid.To establish the generality of the results, a further five lipsticks were investigated using the in-situ method and in each case good SERRS spectra were obtained from smears deposited on glass and cotton surfaces (Fig. 5 and 6). With some of the samples a background fluorescence can be observed. Since the spectra indicate that the lipsticks contain similar chromophores, this fluorescence probably arises from areas of the substrate where there has been incomplete coverage with the silver colloid.Although fluorescence background can be removed from spectra these results emphasise the need for efficient adsorption of the colloid onto the sample surface. These results do however show that each lipstick generates a different SERRS ‘fingerprint’ thus indicating the discriminative power of the technique. Since this has been achieved on extremely small quantities of samples, and without any prior separation of the chromophores, this in-situ method offers considerable advantages over techniques that are currently in use.Conclusions SERRS spectroscopy was used successfully for the in-situ analysis of lipstick smears on glass and cotton surfaces. To obtain good quality spectra from the waxy samples a surfactant was required before the colloid was introduced. Of the surfactants studied poly(l-lysine) was determined to be the most favourable to use in conjunction with the concentrated colloid. The reduction of background fluorescence from a sample can be achieved by ensuring good adsorption of the colloid onto the chromophore in the sample area being analysed.The technique is simple, fast, sensitive, selective and requires a minute amount of sample and there are no pre-separation steps. Spectra characteristic of each lipstick tested were obtained indicating the discriminative power of the technique and a degree of universality not common with techniques based on Raman scattering. A comparison of this in-situ method with an extracted sample illustrated the difficulties inherent in any analytical method which requires complete extraction and identification of a mixed dye and pigment system of chromophores from a waxy matrix.However, such methods are employed to analyse the total chromophore mix and SERRS could be used effectively to obtain a quick indication of the success of the extraction procedure required to separate the chromophores from the matrix without the need for further and difficult steps required to isolate each component.Overall, the results from this study clearly indicate that this in-situ SERRS spectroscopic method has considerable advantages over other qualitative analytical techniques used currently for the examination of lipsticks by forensic scientists and manufacturers. References 1 A. M. L. Barker and P. D. B. Clarke, Forensic Sci., 1972, 12, 449. 2 D. J. Reuland and A. E. Welch, J. Forensic Sci. Soc., 1980, 20, 111. 3 M. Y. Choudhry, J. Forensic Sci., 1991, 36, 366. 4 D. J. Reuland and W. A. Trinler, J. Forensic Sci. Soc., 1984, 24, 509. 5 C. H. Munro, W. E. Smith and P. C. White, Analyst, 1993, 118, 731. 6 C. H. Munro, W. E. Smith and P. C. White, Analyst, 1995, 120, 993. 7 C. H. Munro, W. E. Smith, M. Garner, J. Clarkson and P. C. White, Langmuir, 1995, 11, 3712. 8 C. Rodger, W. E. Smith, G. Dent and M. Edmondson, J. Chem. Soc., Dalton Trans., 1996, 5, 791. 9 P. C. Lee and D. Meisel, J. Raman Spectrosc., 1986, 17, 55. 10 P. C. White, C. H. Munro and W. E. Smith, Analyst, 1996, 121, 3835. 11 P. Hildebrandt and M. Stockburger, J. Raman Spectrosc., 1986, 17, 55. 12 K. Kneipp, Y. Wang, R. R. Dasari and M. S. Feld, Appl. Spectrosc., 1995, 49, 780. 13 P. Hildebrandt and M. Stockburger, J. Phys. Chem., 1984, 88, 5935. 14 C. Rodger, V. Rutherford, P. C. White and W. E. Smith, J. Raman Spectrosc., 1988, 29, 601. Paper 8/05275A Fig. 6 In-situ SERRS collected from ‘A’, ‘B’ and ‘C’ lipstick samples smeared on glass and cotton using 514.5 nm excitation. 1826 Analyst, 1998, 123, 1823–1826

ISSN:0003-2654    

DOI:10.1039/a805275a

出版商:RSC    

年代:1998

数据来源: RSC