|
1. |
Use of Single and Sequential Chemical Extractants to AssessRadionuclide and Heavy Metal Availability From Soils for RootUptake |
|
Analyst,
Volume 122,
Issue 8,
1997,
Page 89-100
V. H. Kennedy,
Preview
|
|
摘要:
Critical Review Use of Single and Sequential Chemical Extractants to Assess Radionuclide and Heavy Metal Availability From Soils for Root Uptake V. H. Kennedy*a, A. L. Sancheza, D. H. Oughtonb and A. P. Rowlanda a Institute of Terrestrial Ecology, Merlewood Research Station, Windermere Road, Grange-over-Sands, Cumbria UK LA11 6JU b Laboratory for Analytical Chemistry, Agricultural University of Norway, P.O. Box 5026, Ås NLH-7754 Norway Summary of Contents Introduction Role of Extractants in Predicting Bioavailability Single Extractants Sequential Extraction Schemes Other Methods Used to Estimate Soil Bioavailability Experimental Aspects of Extraction Techniques Conclusions Keywords: Extraction methods; bioavailability; plant uptake; soil phases; data interpretation Introduction To predict the persistence and potential mobility of radionuclide and heavy metal contamination in soils it is necessary to devise a robust chemical or biological method to quantify the fraction of soil contamination available for root uptake.Such information is important for predictive models and environmental assessments where the soil–plant pathway is a key contributor to potentially harmful effects to plants and animals, e.g., estimates of pollutant concentrations likely to interfere with normal plant growth and of potential radionuclide dose-to-man. Single and sequential extraction methods have been used for this purpose. This review discusses their application in radioecological and pollutant studies, and considers problems in interpretation of data produced. Following deposition and/or discharge, the distribution of pollutants between soil solution and soil solid phases gradually reaches equilibrium and may become associated with soil–plant nutrient cycling processes.The rate at which equilibrium is reached depends on the form of the original pollution, and on soil characteristics: this process may take years. Soil is an environmental, biochemical reaction system with three important phases: solid (mineral particles, organic debris, plant roots), solution (groundwater, rain water, biological excreta, products of biochemical reactions) and gas (atmospheric, products of biochemical reactions) which move towards equilibrium with one another. Radionuclide and heavy metal ions bound to the solid phase can be mobilised into the solution phase by changes in soil pH, temperature, redox potential, soil organic matter decomposition, leaching and ion exchange processes and by microbial activity.The soil solution phase (soil water) includes nutrients and metals in ionic form and organic and inorganic complexes (charged and/or neutral species). Soil solution is thought to play an important role in plant nutrient and pollutant uptake from soils and soil-to-plant cycling processes. For instance, heavy metals are mobilised into soil solution by a decrease in pH and by decomposition of humus.1 Once radionuclides and heavy metals enter the soil their form changes as they interact with other soil components.2,3 Initially, the bioavailability of soil radionuclide contamination is influenced by the physico-chemical form of the fallout.4 Thereafter, the form/speciation of soil contamination is likely to change with time due to dynamic biological and biochemical soil processes including organic and mineral cation exchange, complex formation, leaching and mineral weathering.For instance, radionuclides and heavy metals may become adsorbed to mineral particles,5–8 or incorporated into micro-organisms9 and/or fungi10–13 or adsorbed to decomposing organic particles or incorporated into organic complexes in the soil.14–16 As soil solution seeps down the soil profile it provides water and nutrition for plants and soil micro-organisms. If practicable, analysis of soil solutions would be an ideal way of measuring soil radionuclide and heavy metal bioavailability with time. Valerie Kennedy is a longstanding member of the Environmental Chemistry Section, Institute of Terrestrial Ecology (Merlewood Research Station).This research group has a proven team expertise in the development and validation of nutrient and pollutant analytical techniques for environmental samples. Her main research interests focus on ways of quantifying soil-to-plant transfer of nutrients and pollutants and on quality assurance methods for analytical techniques. For the last ten years she has also collaborated with the Merlewood Radioecology research team.She was Secretary to the NERC Steering Committee on Environmental Radioactivity (1987–1990) and was a joint guest editor to a special edition of Science of the Total Environment (1993)—Chemical analysis of natural waters in environmental research: analytical solutions to current problems. She is also the ITE Merlewood Safety Adviser and a member of the Institute of Occupational Safety and Health.Analyst, August 1997, Vol. 122 (89R–100R) 89RUnfortunately, it is difficult to obtain samples of soil solution that are representative of a specific volume of soil, and that contain concentrations of soil contaminants above analytical detection limits. A compromise has been the development of laboratory based extraction methods using field soils. The main objective of this review is to discuss the use of extraction methods to quantify soil radionuclide availability for plant uptake.Information about heavy metal extraction methods have been included for comparative purposes because similar problems are encountered. This review also comments on the use of extraction methods to identify pollutant soil phase associations because changes in these can affect potential soil pollutant availability for root uptake. There is also a wide body of literature on sediment extraction techniques; some relevant references have been cited where appropriate, but studies of freshwater or marine sediments are outside the scope of this review.Role of Extractants in Predicting Bioavailability There are no simple analytical methods that simulate the action of nutrient uptake by plants, therefore, a wide variety of empirical extraction techniques have been adopted.17–20 Many extraction methods are based on the assumption that ions associated with soil components will be displaced from adsorption sites by the presence of an excess of competing ions in the extract solution.As discussed by Grimshaw17 nutrient concentrations measured in extractants can only provide an estimate of potential nutrient availability for plant uptake or for cation exchange between soil and soil solutions. The same applies to radionuclide and heavy metal concentrations measured in extract solutions. Heavy metals and radionuclides associated with exchangeable phases, as measured by extractants, are assumed to be easily mobilised by soil–soil solution ion exchange reactions and are therefore available for root uptake.Pollutants associated with oxidisable phases, as measured by more chemically aggressive extractants, are assumed to remain in the soil for longer periods but may be mobilised by decomposition processes21 (e.g., weathering and microbial activity). However, reactions taking place in the laboratory during extraction are non-selective. They are likely to be influenced by the length of extraction time and the soil mass : volume ratio used.22 Therefore, radionuclide and heavy metal concentrations measured in the extract solution are defined by the extractant and the experimental protocol.23,24 They provide a relative empirical measurement of the amounts of radionuclide and heavy metal contamination that may be available for plant uptake at the time of the measurement, or over a longer period of time.Actual uptake by plants is determined by other factors such as the number of competing ions in soil solution and root uptake processes which vary between plant species.A range of extraction solutions have been used because they are thought to bring soil pollutants associated with particular soil fractions into solution (Table 1). It is often implied that pollutant concentrations measured in extract solutions can be Table 1 The role of extractants used to assess the potential bioavailability of radionuclides and heavy metals for plant uptake Soil/sediment Dominant fraction extracted Reference Extractant pH (authors’ description) (examples only) H2O (distilled/deionised) Soils Water soluble exchangeable soil–soil 4, 33, 41, 44, 45 solution cations 0.43 m HOAc Soils More tightly bound inorganic soil 2,77 pH 3.2–7.3 TN adsorption sites (clay minerals; small amounts organic and/or oxide bound). 1 m NH4OAc Soils § 3, 28, 29, 33 õ NH4OAc pH 4.65 Soils õ 39 1 m NH4NO3 * Soils õ 1 0.005 m CaCl2 Soils õ 55 0.01 m CaCl2 † Soils õ 37, 38, 39 0.2 mm DTPA Soils õ 29 0.001 m DTPA Soils õ 28 pH 5.1–7.8 ¢ Exchangeable nutrient and pollutant 0.01 m EDTA–1 m Soils ½ soil–soil solution cations. 28 (NH4)2CO3 pH 5.1–7.8 õ 0.1 m HCl Soils õ 28 pH 5.1–7.8 õ 0.1 m HNO3 Soils õ 28 1 m HNO3 Soils õ 29 0.05 m KCl Soils õ 45 1 m MgCl2 Sediments õ 34, 35, 68, 78 MgCl2 pH 7 Sediments Soils Ä 26, 47 0.4 m EDTA Soils § 79, 80 0.1 m K4P2O7·H2O Soils õ 2 pH 3.2–7.3 ¶ Organically bound 0.05 m NaOH Soils õ 45 0.1 m NaNO3 ‡ Soils Ä 1 0.05 m Na2EDTA Soils § 45 1 m NaOAc adjusted to pH Sediments, Soils ¶ Carbonate bound 4, 26, 34, 35, 41, 79 5 with HOAc Ä 0.04 m NH4OAc in 25% v/v Sediments, Soils Bound to iron and manganese oxides 4, 26, 34, 35, 41, 79 HOAc 0.1 m HOAc Soils Mainly inorganic sites (specific sorption) 2 pH 3.2–7.3 The International Soil Exchange (ISE) organises a quarterly inter-laboratory comparison scheme for: * As, Cd, Cr, Cu, Hg, Ni, Pb, Tl, Zn.† B, Cd, Co, Cr, Cu.‡ Cd, Cu, Ni, Pb, Zn. 90R Analyst, August 1997, Vol. 122used to quantify radionuclides and heavy metals associated with specific soil phase associations; however, extractants may also remove small but variable amounts of pollutants associated with other soil fractions. This lack of scientific rigor has been compounded by the use of published extraction methods and sequences by non-specialists.25 A study by Rigol et al.26 clearly illustrates the difficulty of assigning soil phases to specific extractants.They used 0.1 mol l21 NaOH under N2 and 0.1 mol l21 Na4P2O7, to estimate radiocaesium activity concentrations associated with humic acid in a range of soils. They found that humic acid concentrations extracted by 0.1 mol l21 NaOH were consistently higher than those extracted by 0.1 mol l21 Na4P2O7 (for soils with 14–88% organic matter content). They suggested that the constant ratio between the two extractants for the amount of humic acid extracted for these soils indicated that differences were related to the extractant, rather than differences in soil characteristics.They also found good correlation between percentage organic matter and percentage extracted humic acid for these soils in both extract solutions. However, they concluded that 0.1 mol l21 NaOH extracted humic acid more efficiently because it had a higher correlation with the organic matter content and would therefore give a better estimate of radiocaesium associated with humic acid for most soil types. Single Extractants A number of different chemical extractants have been used to make a relative empirical assessment of heavy metal/radionuclide availability for plant uptake from different components of the soil (Table 1).Radionuclides and heavy metals brought into the extract solution are either defined as ‘plant available’, or attributed to a soil phase such as carbonates or hydrous oxides which may or may not be available for root uptake.Ideally, the extractant selected should not affect the soil solid " solution equilibrium, therefore it should not change the soil pH, form complexes, or take an active part in ion exchange reactions. Unfortunately, extractants are not soil phase-specific. For example, 0.1 m ethylenediaminetetraacetic acid (EDTA) will extract heavy metals associated with carbonate, organic and ‘available’ soil phases.27 Ure27 tabulated information about a number of single extractants where soil extract concentrations have been reported to correlate with plant concentrations for elements such as Cd, Cu, Zn, Ni, Pb, Se, Mo and Cr.Gupta and Aten1 evaluated several extractants for assessing soil bioavailability of Cu, Zn, and Cd for lettuce and rye grass growing on study soils. These included 2 m HNO3, 0.52 m NH4OAc + EDTA, 0.05 m CaCl2, 0.1 m NH4NO3, 0.1 m KNO3 and 0.01 m CaCl2. They recommended the neutral extractant 0.1 m NaNO3 for bioavailability studies of heavy metal soil contamination because it showed a higher correlation with plant heavy metal concentrations when compared with other extractants used in the study.Adriano et al.28 noted that it was important that pollutant concentrations measured in extract solutions should be able to predict the ‘plant availability’ of the ion being studied for a wide range of soil types, and that it should also be sensitive to the ion where availability is suspected to be minimal. Of five extractants evaluated (0.1 m HCl, 0.1 m HNO3, 1 m EDTA– NH4CO3, 0.001 m DTPA and 1 m NH4OAc) for estimating the pattern of plant uptake of 241Am from a sandy loam and a clay loam, 1 m NH4OAc gave consistent correlation coefficients for plant uptake of 241Am for bush bean (Phaseolus vulgaris) and for corn (Zea mays).They suggested that this extractant was a potentially useful tool for diagnosing plant availability of 241Am from contaminated environments. Wilson and Cline29 used four extractants: 1 m NH4OAc, 0.1 m HNO3, 1 m HNO3 and 0.2 mm DTPA (calcium trisodium salt) to evaluate the extractability of 239Pu from three different soils (a slightly calcareous sandy loam, a calcareous silt loam and a medium acid forest soil).All four extractants extracted a very small proportion of the total 239Pu activity from the soil; the highest amount of 239Pu extracted by 0.1 m HNO3 was only 0.64% of the total 239Pu soil activity, therefore, it was not possible to evaluate the extractants for plutonium. Extracted concentrations of 85W were higher; the amount extracted by 1 m NH4OAc followed a similar pattern for all three soils and was therefore their favoured extractant.One molar NH4OAc pH 7 has also been found to be a robust extractant for acidic and neutral soils.17 It has provided useful data on soil nutrient status for agriculturalists assessing fertilisation requirements and for environmental researchers studying semi-natural ecosystems. 20,30 A more recent modification to the use of NH4OAc has been to adjust the extractant to the pH of either the soil or of the soil solution, with acetic acid.31,32 Evans and Dekker33 used water and 1 m NH4OAc extractions in association with a greenhouse experiment and an incubation experiment. The experiments were designed to examine the effects of potassium and calcium carbonates on 137Cs availability for uptake by oats and alfalfa from a range of nine Canadian acidic and alkaline silts and sandy loam soils.Results from water extractions were used to estimate the percentage of soluble 137Cs and results from 1 m NH4OAc extractions to estimate the percentage of exchangeable 137Cs in the experimental soils. They found that the effect of fertiliser applications on 137Cs availability for plant uptake varied with soil type, and with the timing of fertiliser application. Addition of K reduced plant uptake of 137Cs for all the study soils without significantly reducing the amount of extractable 137Cs in their soils. These results illustrate that extraction procedures can only measure what is potentially available for plant uptake; other factors, such as soil nutrient concentrations, may also influence what proportion of soil pollutant is actually taken up by plants growing in the soil.However, these data also demonstrate that empirical extraction techniques can be used to study possible interactions between pollutant and nutrient ions in soil-to-plant cycling processes.Although a number of researchers have opted for 1 m NH4OAc pH 7 as a suitable extractant to estimate ‘exchangeable’ ion concentrations in soils, Tessier et al.34 felt that it was unsuitable because it may also attack carbonates within aquatic sediments. As an alternative they compared two other extractants, 1 m MgCl2 pH 7 and 1 m NaOAc pH 8.2 and concluded that 1 m MgCl2 pH 7 was the most suitable extractant because it extracted a smaller percentage of carbonate bound ions.Similarly, Ure27 noted that 0.1 m EDTA will extract trace metals bound or associated with carbonate phases as well as those associated with organic matter. Martin et al.35 suggested that single extractants are likely to permit dissolution of some non-labile organic and inorganic fractions as well as labile fractions. It is generally accepted that 1 m NH4OAc pH 7 is unsuitable for alkaline soils, although adjusting its pH to 9 can be a viable alternative.17 Paasikallio36 used NH4OAc pH 4.65 for a time series of extractions on soil sub-samples from a two year rye grass pot experiment, using seven soil types.Extractable 137Cs activity decreased slightly for the first few months and then maintained a similar concentration for all soil types except the carex peat which decreased with time throughout the experiment. Strontium- 90 behaved differently; after an initial decrease the amount extracted gradually increased with time for all soils during the experiment.Plant uptake of 90Sr and 137Cs was reported as a percentage of the radioactivity originally applied to the soil, rather than as concentrations brought into solution by the extractant. Surprisingly, no information about relationships between plant concentrations and extracted soil concentrations was presented for either element. Novozamsky et al.37 used single extraction procedures to assess soil availability of Cd, Cu, Ni and Zn to plants.A range of extractants and mineral soil types were included, and Analyst, August 1997, Vol. 122 91Rrelationships between extraction concentrations and plant yield for green beans, lettuce and endives were evaluated. They concluded that 0.01 m CaCl2 was the most useful extractant for their mineral soils because it had approximately the same salt concentration as the ‘average soil solution’, a range of heavy metals and nutrients could be measured in the same extract solution (so that interactions between a range of ions brought into solution could be studied), and because Ca2+ is often the dominant cation competing for soil adsorption complex sites in mineral soils.Similarly, McLaren and Crawford,2 favoured 0.05 m CaCl2 as the first extraction stage of their sequential extraction series because it had less effect on the natural soil pH and extracted fewer ions associated with soil organic matter.Another recent study recommends the use of 0.01 m CaCl2 because it has a similar pH, concentration and composition to soil solutions.38 Correlations between Cd, Cu, Pb, Ni and Zn brought into solution by 0.01 m CaCl2 and the appearance of toxicity symptoms in plants growing on mineral soils were reported to be more significant than relationships between total soil and plant concentrations for the same metals. This extractant has the added advantage that detailed extraction method protocols have been published for B, Cd, Cu, Pb, Ni and Zn.39 A different approach was developed by another research team because extractable radiocaesium activity concentrations in the nutrient deficient soils they were studying were relatively small compared with radiocaesium activity concentrations in the plants growing on their soils.3 For example, radionuclide activity concentrations in plants represented 30% of the total labile radionuclide pool in some Norwegian soil–plant systems. 3 For some soil types (e.g., organic, nutrient deficient) there can be significant seasonal variations in extracted pollutant concentrations due to depletion of the soil labile pollutant pool in spring, and the release of pollutants from fallen litter during decomposition processes in autumn, in some ecosystems. This phenomenon led to the development of the ‘mobility factor’ method (addition of plant extractable and soil extractable radionuclide contamination).They used water and 1 m NH4OAc extractions of soil and vegetation in an attempt to differentiate between ‘inert’ and ‘mobile’ fractions of radionuclides in soil–water–plant systems. The data were then used to calculate a ‘mobility factor’ which was defined as the percentage of deposited radionuclide present in a labile or potentially bioavailable form at a given time. They suggested that the ‘mobility factor’ represented the total fraction of deposited radionuclide in the soil–water–plant system that is present in a labile form, and readily available for plant uptake at the time of analysis.Time dependent studies on the variation in bioavailability at single sites gave promising results; over six years, a decrease in the labile fraction of 137Cs in soils was correlated with a decrease in soil-to-plant transfer.31 Actual plant uptake was found to be dependent on other factors such as vegetation type, season, microbial activity, available potassium and stable caesium. Such seasonal differences can also be quantified by analysis of litter extract solutions, in addition to surface soil extract solutions, and associated total plant concentrations.It is apparent that a number of single extraction methods can provide a relative empirical method for evaluating the potential availability of soil pollutants for plant uptake. Information about relationships between other nutrient and pollutant ions and/or complexes extracted from the soil at the same time can also be obtained.There is a growing body of evidence that 1 m NH4OAc could be adopted as a suitable empirical soil extractant for a wide range of soils and analysed for a range of nutrients and pollutants (including radionuclides). Other extractants such as 0.01 m CaCl2 (B, Cd, Co, Cr, Cu), 1 m NH4NO3 (Cd, Cu, Ni, Pb, Zn) and 0.1 m NaNO3 (As, Cd, Cr, Cu, Hg, Ni, Pb, Tl, Zn) have also been identified as suitable extractants for mineral soils.Use of these extractants has the added advantage that the International Soil Exchange (ISE) organise a quarterly interlaboratory comparison scheme for the elements listed. Sequential Extraction Schemes Heavy metals and radionuclides are present in many forms.2,3 They may be in soil solution (ionic, organic complexes, inorganic complexes), on soil cation exchange sites, more tightly bound to soil adsorption sites, occluded into soil oxide material, incorporated into organic plant litter, incorporated into soil micro-organisms and/or soil animals or in the lattice structure of primary and secondary soil minerals.2,3 Sequential extraction methods were originally designed to study some of these phase associations in soils and sediments.2,34,40,41 However, pollutant concentrations measured in sequential extraction solutions can also be used to differentiate between short and long term soil radionuclide and heavy metal bioavailability.Most sequential extraction schemes are based on the scheme devised by Tessier et al.34 which was designed to bring metal pollutants associated with different aquatic sediment phases into solution.The phases are operationally defined as water soluble (H2O), exchangeable (1 m MgCl2), carbonate bound (1 m NaOAc adjusted to pH 5 with HOAc), bound to Fe–Mn oxides (0.04 m NH4OH, HCl in 25% v/v HOAc), bound to organic matter (H2O2 adjusted to pH 2 with HNO3) and residual (HF–HClO4digestion).The increasing strength of extractants down the extract series can be used to predict pollutant soil associations and hence potential bioavailability. Initial extraction stages in sequential extraction schemes remove soluble and/or ion exchangeable pollutants, therefore, it is not surprising that extractants commonly used for single extraction studies are selected for initial stages in sequential extraction schemes. Thus, pollutant concentrations measured in these extract solutions can be used to make a relative empirical estimate of pollutant availability for root uptake from the soil, although few researchers have directly related their findings to plant uptake (Table 2).The use of extract solutions with a gradual increase in displacement/dissolution strength down the extraction scheme provides additional information about soil pollutant fractions that may be released from soil phase associations and become available for plant uptake because of weathering, decomposition and/or other soil processes, e.g., the pollutant fraction that is oxidisable or mobilised if the pH is lowered to 5, or the fraction that may be trapped within the crystal lattice of soil mineral particles.Hence, sequential extraction can give an indication of the binding strength of soil pollutants and their potential for mobilisation over time due to changes in soil chemistry.4,42,43 Aquatic sediment characteristics, such as organic matter type and amount, redox potential and salinity, differ significantly from soil characteristics. Such differences should be taken into account when adapting the Tessier sequential extraction scheme for soils.Such schemes usually include three to six sequential extraction stages (Table 3). The physical characteristics of study soils often determine the extraction sequence selected. For instance, a carbonate extraction stage needs to be included where carbonates may play an important role in mobilising heavy metals bound to carbonates in sediments34 but may be inappropriate for acidic peats with negligible calcium carbonate concentrations.44 Similarly, the use of a hydrogen peroxide step to oxidise organic matter is not likely to provide useful information for soils with high organic matter content.Such soil differences often account for differences in the number of extraction stages included in each scheme (Tables 2 and 4). In the 1970s, McLaren and Crawford2 studied copper–soil associations and mobility in 24 UK soils with pH values between 3.2 and 7.3.As part of their work they developed a 92R Analyst, August 1997, Vol. 122sequential extraction method using the following sequence: 0.05 m CaCl2 (weakly bound, soluble), 0.43 m HOAc (extractable, mainly inorganic), 1 m K4P2O7·H2O [corrected to 0.1m, J. Soil Sci., 1974, 25 (1), 119; organic], 0.1 m (NH4)2 C2O4·H2O (occluded by free oxides) and HF (residual). They looked at correlations between soil characteristics such as percentage organic carbon, percentage clay and pH and soil copper fractionation as measured by their extraction series (Table 5).Chang et al.45 used a five stage sequential extraction scheme developed by Emmerich et al.46: 0.5 m KNO3 (exchangeable fraction), H2O (adsorbed fraction), 0.5 m NaOH (organic fraction), 0.05 m Na2 EDTA (carbonate fraction) and 4 m HNO3 (sulfide/residual fraction). They looked at soil associations of Cd, Cr, Cu, Ni, Pb and Zn after sludge applications.They also measured total cadmium and zinc concentrations in barley growing on their experimental plots but did not comment on relationships between plant concentrations and concentrations measured in soil extract solutions. Tessier et al.’s34 scheme for sequential extraction of sediment and soil samples was modified by Wilkins et al.41 to identify the soil components with which 137Cs, 239/240Pu, 106Ru, 241Am, 129I and 90Sr were associated. They concluded that 90Sr is mainly associated with the soil ion exchange phase, whereas 241Am was mainly associated with soil oxide and organic phases. Informa- Table 2 Examples of sequential extraction schemes that have been used for heavy metal and radionuclide analysis of soils Number of Soil–plant extraction Sample transfer Method Reference stages preparation Elements discussed evaluated (examples only) 3 Soil, air dried 137Cs, 90Sr, 110Ag No No 47 (traditional) 2 mm sieved 4 Soil, air dried 137Cs, 90Sr, 110Ag No No 47 (organic role) 2 mm sieved 4 Soil, dried at 50 °C 137Cs, 106Ru No Yes 81 2 mm sieved 4 Soil, air dried 137Cs Yes No 42, 43 2 mm sieved 4 Model sediment Pb, Ni, Cu, Zn No Yes 34 4 Soil, dried at 50 °C 137Cs, 106Ru No Yes 81 2 mm sieved 5 Soil, air dried Cu Yes Yes 2 1 mm sieved 5 Soil Pb, Ni, Cu, Zn, Cd, Cr Yes Yes* 45 Field moisture 5 Soil, air dried, 239,240Pu No Yes* 42 1 mm sieve 5 Model sediment Pb, Ni, Cu, Zn No Yes 78 5 Artificial As, Cs, Hf, Zr, No Yes 35 substrate Ce, Co 5 Artificial As, Ce, Co, Cs, Hf, Zr, Th No Yes 26 substrate 5 Soil, fresh 137Cs Yes Yes 44 6 Soil, air dried, 134Cs, 241Am, 106Ru, No Yes* 41 1 mm sieve 90Sr, 239Pu, 129I 6 Soil, air dried 137Cs, 90Sr Yes Yes 4 * Method evaluated independently by other authors.Table 3 Examples of single extractants that have been used to assess the potential bioavailability of radionuclides and trace elements for plant uptake Sample Method Reference Extractant preparation Elements evaluated (examples only) H2O (distilled) Not clear 137Cs Yes* 33 1 m NH4OAc Air dried 239Pu Yes* 29 Dried, 70 °C 137Cs* Yes 3 241Am Yes 28 Air dried 137Cs, 90Sr Yes* 3 NH4OAc pH 4.65 Air dried 137Cs, 90Sr 36 0.1 m HCl Oven dried 241Am Yes 28 70 °C 0.1 m HNO3 Oven dried 241Am Yes 28 70 °C 1 m HNO3 239Pu Yes* 29 0.2 mm (DTPA) 239Pu Yes* 29 0.01 m CaCl2 Oven dried Cd, Cu, Ni, Zn Yes 40 0.001 m DTPA Oven dried 241Am Yes 28 70 °C 0.01 m EDTA– Oven dried 241Am Yes 28 1 m (NH4)2CO3 70 °C * Method evaluated independently by other authors. Analyst, August 1997, Vol. 122 93RTable 4 Examples of extractants that have been used for steps 1–6 of extraction schemes Soil components extracted Extract Reference Extractant (authors’ description) number Elements discussed (examples only) H2O (distilled/deionised) Water soluble 1 134Cs, 241Am, 106Ru, 90Sr, 41 239Pu, 129I 1 137Cs, 90Sr 3, 4 1 137Cs 42, 43 1 and 2 137Cs 44 0.05 m CaCl2 Weakly bound soluble 1 Cu 2 1 m MgCl2 Exchangeable 1 Pb, Ni, Cu, Zn 34 1 Pb, Ni, Cu, Zn 78 1 As, Cs, Hf, Zr, Ce, Co 35 1 m MgCl2 1 As, Ce, Co, Cs, Hf, Zr, Th 26 (adjusted to pH 7) 1 134Cs, 85Sr, 110Ag 47 0.01 m CaCl2 1 239/240Pu 41 1 m NH4OAc pH 7 1 137Cs, 106Ru 81 2 137Cs 42, 43 2 134Cs, 241Am, 106Ru, 90Sr, 41 239Pu, 129I 2 137Cs, 90Sr 4 3 137Cs 44 0.11 m HOAc Exchangeable and 1 134Cs, 85Sr, 110Ag 47 carbonate bound 0.43 m HOAc Extractable, inorganic 2 Cu 2 0.5 m HOAc Inorganically absorbed 2 239/240Pu 41 1 m NaOAc adjusted to Carbonate 2 As, Ce, Co, Cs, Hf, Zr, Th 26 pH 5 with HOOAc 2 Pb, Ni, Cu, Zn 34, 78 2 As, Cs, Hf, Zr, Ce, Co 35 3 134Cs, 241Am, 106Ru, 41 90Sr, 239Pu, 129I 3 137Cs, 90Sr 4 0.04 m NH4OAc in 25% Fe and Mn oxides 2 137Cs, 106Ru 81 v/v HOAc 3 Pb, Ni, Cu, Zn 34 3 Pb, Ni, Cu, Zn 78 3 As, Cs, Hf, Zr, Ce, Co 35 3 As, Ce, Co, Cs, Hf, Zr, Th 26 4 134Cs, 241Am, 106Ru, 41 90Sr, 239Pu, 129I 4 137Cs, 90Sr 4 4 137Cs 44 0.1 m (NH4)2C2O4·H2O Occluded free oxides 4 Cu 2 0.1 m NaOH (In N2) Humic substances 2 134Cs, 85Sr, 110Ag 47 0.1 m Na4P2O7·H2O Organic 2 134Cs, 85Sr, 110Ag 47 3 239/240Pu 40 5 134Cs, 241Am, 106Ru, 41 90Sr, 239Pu, 129I 0.1 m K4P2O7·H2O 3 Cu 2 30% H2O2 then 0.5 m Oxidisible organic 3 137Cs 42, 43 NH4OAc and 0.04 m matter HNO3 30% H2O2 to pH 6 with 3 134Cs, 85Sr, 110Ag 47 20% HNO3 3 137Cs, 106Ru 81 4 Pb, Ni, Cu, Zn 34 4 Pb, Ni, Cu, Zn 78 4 As, Cs, Hf, Zr, Ce, Co 35 4 As, Ce, Co, Cs, Hf, Zr, Th 26 5 137Cs, 90Sr 4 7 m HNO3 Acid digestible 4 137Cs, 106Ru 81 4 137Cs 42, 43 5 137Cs 44 6 134Cs, 241Am, 106Ru, 90Sr, 41 239Pu, 129I 6 137Cs, 90Sr 4 Mixture of HNO3, HF Residual 5 Pb, Ni, Cu, Zn 34 and HClO4 5 Pb, Ni, Cu, Zn 78 5 As, Cs, Hf, Zr, Ce, Co 35 5 As, Ce, Co, Cs, Hf, Zr, Th 26 5 Cu 2 Ashing and HF digestion 5 239/240Pu 41 94R Analyst, August 1997, Vol. 122tion was also presented for 134Cs and 106Ru (Table 6). They suggested that this type of data could be used to provide information about the potential radionuclide availability for plant uptake.Livens et al.40 presented a similar data set for 239,240Pu using different soil types (Table 7). Oughton et al.4 (Tables 2 and 4) used their sequential extraction scheme to identify the soil fraction of 137Cs and 90Sr contamination that was incorporated within uranium oxide fuel particles. Extraction data for Ukrainian, Belarussian and Norwegian soils were used to compare site specific radionuclide soil mobility and transfer factors.Results indicated that radionuclide speciation and 90Sr mobility varied with distance from Chernobyl. These data were used to support their hypothesis that fuel particle weathering would lead to an increase in the bioavailability and plant uptake of 90Sr within the Chernobyl 30 km zone.3,4 Another research team47 compared two sequential extraction schemes (Tables 2 and 4); one designed to isolate organic fractions associated with 134Cs, 90Sr and 110Ag, and the other to look at inorganic associations (based on the work in reference 34) in two soils, a sandy loam and a sandy soil.Their second scheme followed protocols recommended by the Community Bureau of Reference for heavy metals in soils and sediments.48 They suggested that data from this type of experiment could be used to predict radionuclide availability for plant uptake from soils but did not verify this experimentally. Fawaris and Johanson42,43 used a sequential extraction technique (Tables 2 and 4) to estimate the percentage of 137Cs in a Swedish forest soil that was likely to be available for plant uptake with time.They looked at correlations between water extractable and NH4OAc pH 7 extractable 137Cs and organic matter content, but not for correlations between extracted 137Cs and plant 137Cs. Kennedy and Horrill44 developed a sequential extraction method for use with acidic, organic soils, such as peats, that had relatively low Chernobyl 137Cs contamination compared with soils in Scandinavia and central Europe (Tables 2 and 4).They did a series of extractions over time, using fresh sub-samples of three different organic soils, to assess the reproducibility of the method. They then used the method to extract soils from nine heather dominated ecosystems and looked for correlations between extracted 137Cs and heather plant 137Cs activity concentrations. Relationships between 137Cs activity extracted at each stage of the procedure were presented as a simple correlation between 137Cs activity concentrations of heather growing on the soil and 137Cs activity extracted at each stage of the procedure, and then as a cumulative value derived by adding 137Cs activity concentrations from each stage sequentially.Even with the high variability inherent in all soil extraction procedures they found that there was significant correlation between extractable 137Cs activity (0–10 cms) and heather 137Cs activity; r = 0.748** (r > 0.708, p = 0.01) but only poor correlation with total 137Cs soil activity (0–10 cms); r = 0.406.They suggested that if there were similar findings for other soil types and plant species it might be possible to obtain relative estimates of soil bioavailability using a double water extraction and/or an NH4OAc pH 7 extraction as an empirical extraction method. Published sequential extraction data are often presented as quantitative evidence of pollutant soil and sediment phase associations, and assume that no phase exchanges or sample solution matrix effects occur during extraction or analysis.49 Even when spike solutions have been used to validate extraction techniques data obtained are not always used to comment on the implications for data interpretation.For example, one research group recovered only 10% of a lead spike in their acetate extraction.50 Low recoveries such as this can be due to a number of reasons, e.g., interactions between the soil and spike solution during extraction, interference(s) during analysis etc.Such inherent problems mean that spike solutions may not be a useful measure of extraction efficiency. A better method is probably the use of laboratory reference soils, and participation in interlaboratory comparison studies where possible. Specific extractant efficiency varies for soils with different mineralogical and organic composition therefore each stage of sequential extraction needs to be validated for a range of soil types.49 Even then data from sequential extraction schemes can only provide a relative empirical prediction of pollutant soil phase associations, and the potentially labile fraction available for root uptake.Pollutant concentrations measured in the final stage of sequential extraction schemes are less variable than those Table 5 Correlation data for copper soil fractionation and soil components for 24 soils, as fractions of percentage total Cu, using a sequential extraction scheme2 Cu Cu Cu Cu % % Free Free AAC PYR OX RES Org C Clay pH Fe Mn Cu-CA 0.67** 0.08 0.24 20.37 20.12 20.22 20.64** 20.09 20.21 Cu-AAC 20.12 0.07 20.01 20.31 20.21 20.44* 20.11 20.19 Cu-PYR 20.32 20.37 0.53** 20.38 20.13 20.50 20.60** Cu-OX 20.32 20.28 20.26 20.07 0.44* 0.22 Cu-RES 20.16 0.42* 0.31 20.07 20.06 Key: Cu-CA = soil solution and exchangeable copper; Cu-AAC = bound by mainly inorganic sites (specific absorption); Cu-PYR = bound by organic sites (specific absorption); Cu-OX = occluded by free oxides; Cu-RES = residual; * significant at 5% level; ** significant at 1% level Table 6 Examples of relative radionuclide soil phase association data obtained from sequential extraction procedures for 134Cs (expressed as a percentage of the total)41 Clay Loam Sand Peat Soil phase 134Cs 106Ru 134Cs 106Ru 134Cs 106Ru 134Cs 106Ru Water soluble 5 7 4 4 6 6 5 2 Exchangeable < 1 2 2 1 4 < 1 36 3 Carbonate < 1 2 < 1 8 < 1 3 11 1 Oxide < 1 31 3 30 3 32 25 13 Organic < 1 49 21 51 27 38 12 75 Residual 95 84 70 60 60 21 11 5 Analyst, August 1997, Vol. 122 95Rattributed to earlier stages in the sequence.49 Therefore, pollutant concentrations from this stage can be used to estimate the soil pollutant fraction that is likely to be strongly resistant to weathering processes, and unavailable for root uptake in the short to medium term (0–10, or even 20 years in temperate ecosystems).Although sequential extraction methods were originally designed to study heavy metal and radionuclide phase associations in sediments they are now widely used to look at changes in both soil and sediment pollutant phase associations. Information from early stages in the extraction sequence is similar to that obtained from single extraction methods so it can also be used to help predict potential short-term soil pollutant bioavailability by relating extracted soil concentrations to plant concentrations. Data from later stages in the sequence could also be developed into a useful tool to predict potential longer term radionuclide and heavy metal release into the soil labile pollutant pool (due to soil weathering and decomposition processes).Other Methods Used to Estimate Soil Bioavailability (Not Extraction Methods) In radioecology, transfer factors [plant activity concentration (Bq kg21)]/[total soil activity concentration (Bq kg21)] have been widely used in an attempt to quantify the availability of soil radionuclides for plant uptake. This method gives a relatively crude estimate of potential soil radionuclide bioavailability because soil-to-plant transfer can vary considerably between different plant species, and the proportion of soil radionuclides available for root uptake also varies with season and over time.Even when transfer factors are species specific there is often a poor correlation between total soil radioactivity concentrations and plant radionuclide activity concentrations. 17,18,51 In spite of this transfer factors are currently accepted as the most practical way of describing plant uptake in many radiological assessment models. Other experimental techniques have been used to estimate the bioavailable fraction of soil radionuclide contamination, particularly for radiocaesium.The sorption mechanistic method approach developed by Cremers and co-workers7,16,52,53 measures the interception potential of two different sorption sites in soils and sediments for radiocaesium—frayed edge sites associated with illites and regular exchange sites such as organic matter and clay isomorphic substitution sites.This kinetic exchange method takes account of reversible and irreversible pools of radiocaesium in soil and gives an indication of its potential for remobilisation and/or availability for root uptake. The resin technique,54 which was used to study radiocaesium fixation in UK peat soils, also provides information about different sorption sites and release rates for radiocaesium from these soils.A laboratory tracer experiment to estimate the time dependence of 137Cs fixation to clay minerals has also been proposed.55,56 Radiocaesium activity concentrations in the electrolyte solution were assumed to be similar to those likely to occur in soil solutions. They suggested that an understanding of the time dependence of transfer between soil solution and the labile and non-labile phases in the soil must be quantified before models can accurately predict the long-term behaviour of radiocaesium in a range of soil types. A time series such as this takes account of potential changes in soil radionuclide speciation with time and could provide a useful means of assessing radiocaesium bioavailability for a wide range of soils.It is also probable that a similar approach could be adopted for a number of other potential radionuclide contaminants.Both the combined extraction/equilibrium time series and the sorption mechanistic methods take account of changes in soil radionuclide speciation with time but they have only been used for radiocaesium, for a limited range of soils. More work on reproducibility of data for a range of soil types and pollutants is needed before these methods can be widely adopted for pollutant bioavailability studies. They also have the disadvantage that they have not yet been used to look at potential nutrient and pollutant interactions.Centrifugation techniques have also been used to study heavy metals (e.g., Al, Cu, Cd) in soil solutions.57–59 They have the advantage that experimental conditions can be clearly defined and that fresh soil can be extracted in controlled experimental conditions to provide an instantaneous measure of heavy metals in the capillary fraction of soil solution at the time of sampling.Since only small volumes of solution can be obtained pollutant concentrations in some solutions may be below detection limits for analytical methods and it may be difficult to analyse the solution for a wide range of elements. Experimental Aspects of Extraction Techniques Sampling Strategy It is generally assumed that a soil sample submitted for chemical extraction is representative of the larger area from which it is taken. It is therefore necessary to devise a sampling strategy that will produce representative soil samples that can be used to extrapolate information about soil from the larger area.Whether or not a soil sample is representative of the area is dependent on topography, uniformity of soil type and on how the sample has been selected and collected. Soil sampling strategies have often been unsatisfactory.35,60–62 Several reviews have made recommendations about soil sampling protocols, e.g., for chemical analysis,63 for agricultural or forestry analysis64 and for soil surveys.65 Even when soil samples are taken from an area where the soil is classified as a particular soil type large analytical variations can be obtained for replicate samples taken from a small area of apparently similar land.12,65,66 This is a particular problem for soils taken from undisturbed natural and semi-natural ecosystems that have not been ploughed and fertilised.Thus, reported radionuclide activity concentrations and heavy metal measurements in soils from some sites may be solely a function of the original sampling technique.This problem is exacerbated if the area being surveyed includes open grazing areas, scrub, trees, stony areas and boggy patches. A high degree of spatial variability may exist over quite small areas. The characteristics of the area being studied should be taken into account when the sampling regime is being designed. It should also be clearly documented. Unfortunately many papers that present pollutant field data do not provide clear information about field sampling strategies and/or topographical features of their sampling sites that may affect deposition, mobility and accumulation of pollutant contamination.Also, the number of samples that are collected may be constrained by a lack of resources for statistically acceptable field sampling programmes that will provide a representative data set. It is important that the sample set provides the best possible data with respect to the resources Table 7 Examples of relative radionuclide soil phase association data obtained from sequential extraction procedures for 239,349Pu (expressed as a percentage of the total)40 Brown earth Brown earth Alluvial Soil phase Sand (Woodland) (Pasture) gley Exchangeable 1.0 1.7 2.2 2.2 Specific adsorption 1.5 1.3 4.7 2.5 Organic 68 52.7 58.9 63.5 Sesquioxide 18.5 31.0 17.7 18.5 Residual 11.0 13.3 16.4 7.2 96R Analyst, August 1997, Vol. 122available, field site characteristics and the objectives of the study. One of the major problems with errors introduced during sampling is that they are impossible to quantify but may have a significant effect on measured pollutant distribution patterns, particularly for wide ranging field studies or monitoring programmes, where occasional ‘grab’ samples are taken. Sampling methods can cause significant bias in reported nutrient concentrations measured in both mineral and organic soils17 and in soil radionuclide and heavy metal measurements. 66 Sampling problems are further compounded for radionuclide studies where there is thought to be the possibility of ‘hot spots’. Sample Preparation When resources for analysis are limited, composite samples made by homogenising several samples from a sampling site can give more reliable information on average soil characteristics in the area, but, at the cost of information on within-site variability. Extraction methods use a relatively small (2–100 g) sub-sample of soil. It is therefore important to homogenise the bulk sample before selecting a representative sub-sample for analysis.When soils are air dried and sieved (2 mm sieve) prior to extraction they are stabilised but some of their chemical and physical properties may be altered.17,18 Such changes may be unimportant when measuring total pollutant concentrations in soils (a small fraction of some pollutants may be volatilised), but can significantly affect the amount of radionuclide and heavy metal contamination subsequently brought into solution by extractants, particularly if soils are being used for speciation studies.67 For instance, once a soil has been broken up by sieving the number of ion exchange sites in contact with the extractant may be greater than the number of ion exchange sites in contact with soil solution in the field.Drying and sieving processes homogenise soils so that mineral and organic soil particles become more uniform in size and more evenly distributed than they are in the field. Radionuclide and heavy metal concentrations in extract solutions from soils that have been dried at more than 40 °C and milled to a fine powder may bear little relationship to the amount of pollutant that is available for plant uptake from that soil in the field.Such preparation is acceptable for analysis of total pollutant concentrations but creates an even larger surface area for ion exchange processes than that created by passing samples through a 2 mm sieve. Many papers give little information about the way soils were prepared for analysis. In the absence of this it is difficult to assess whether pollutant concentrations measured in extractants are an artifact of sample preparation or whether they are a reasonable approximation of field conditions.Sample Weight : Extractant Volume and Contact Time Most extraction methods involve shaking a known mass of soil with an extractant for a predefined period of time on an endover- end or side-to-side shaker at ambient temperature.Such procedures are essentially equilibrium processes. If the soil mass to extractant volume ratio is too low (e.g., < 1 : 5) radionuclide and heavy metal re-adsorption processes may occur during extraction and soil–extractant equilibrium may not be attained. Grimshaw17 recommended that the best soil : extractant ratio for nutrient studies is one such that doubling or halving the ratio makes no difference to the final result.He suggested that high ratios were better, recommending 1 : 25 for extractants used to measure nutrient concentrations. A number of radioecologists have opted for 1 : 10 when extracting for radionuclides.4,41 Others do not provide clear information about the ratios used or refer to Tessier et al.34 who used a ratio of 1 : 8 for their heavy metal work with aquatic sediments. It is important that the shaking time is fixed, and that the time is sufficient to allow a steady state to be established between the soil and the extractant but not so long that dissolution of other soil fractions occurs.In practice, the contact time needs to be a minimum of 1 h.17 Where only gentle or intermittent shaking is used it may be necessary to extend this up to 24 h. For instance, Maher68 found that Ca concentrations in sediments extracted by NaOAc–HOAc leachate stabilised after 5 h and iron extracted by NH2OH2–HOAc after 6 h.A problem for the analyst is to select a sub-sample that is representative of the field. It may be necessary to analyse several replicate sub-samples to achieve this. It has been suggested that the minimum representative sample mass for heavy metal analysis should contain at least 1000 sample grains, but that the optimum sample mass should be three or four times greater than the minimum.69 In practice, the optimum mass for a highly contaminated sample will be lower than the mass needed for soils with lower contamination levels.For example, a considerably larger sample will be needed for soils being used in radionuclide sequential extraction studies if activity concentrations in initial extract solutions are to be above the detection limits of analytical methods.44 Even for more highly contaminated soils only a relatively small proportion of the total soil pollutants are likely to be brought into solution by weak extractants. Selecting a representative sub-sample from a fresh soil (particularly a peat) is more difficult than from an air dried, sieved mineral soil or finely milled soil.Organic soils often contain small, medium and large fragments of recognisable plant material as well as a small proportion of mineral particles so that it is difficult to retain the original field proportions in weighed sub-samples. Drying completely alters the physical characteristics of highly organic soils such as peats.17 A larger sub-sample will also be needed because fresh soils contain moisture which may account for a higher proportion of the sample mass than organic soil particles in highly organic soils.There is little evidence in the literature that the importance of using fresh sub-samples of such soils is fully appreciated, although some recent studies have taken this into account. 44,55 Separation of Extractant From Sample Extract solutions used for analysis need to be separated from the soil.Most researchers use either centrifugation,3,34,39,70–72 or centrifugation and vacuum filtration73 or gravity filtration.17,44 Both centrifugation and filtration have inherent problems. For example, very fine particles may be left in the ‘solution’ phase if centrifugation speed is insufficient or if the pore size of the filter is too large. This can be significant if a particular extractant is able to attack the soil phase and disperse it into finer components.If the extractant solution is not strongly acidic it should be centrifuged and collected in polycarbonate containers to reduce adsorption of radionuclide and heavy metal ions onto the side of the container. Glass containers may be used if the extract solution is acidified after separation from the soil. For single extraction procedures both centrifugation and filtration have been used to separate the extractant from soil. Researchers using sequential extraction methods tend to favour centrifuging the sample to separate the extractant from the soil because there is less chance of losing soil between extractions.This can be overcome by using a gravity leaching method with the same filter paper for each of the stages of the extraction sequence.44 If a sequential extraction sequence is being used, centrifugation may affect the extraction characteristics of the soil residue used for subsequent extractions because the soil may become compressed during the process.Thus the soil residue used for Analyst, August 1997, Vol. 122 97Rsubsequent stages may not retain the same wetting characteristics as the original soil sample. This could be important for the second extraction stage, in a sequential series, if it is designed to look at slightly less labile ‘root available’ pollutant contamination in organic soils.74 It is not important for the final stages where the more tightly bound fraction, released after long-term decomposition and weathering processes, is being extracted.Quality Control and Method Validation No information about the reproducibility of extraction methods with time has been found in the literature for either radionuclide or heavy metal pollutants. Evidence from nutrient analysis suggests that reproducibility is good in a reputable laboratory when the extractants are prepared by experienced analysts following standard protocols.75 However, this may not be the case when inexperienced students or researchers are following poorly defined published protocols.Even when the same protocols are used on a sub-sample of the same soil a few months later, nutrient ions measured in solution can vary by up to ±15% for some nutrients. This error may be increased to ±100% if ion concentrations in extract solutions are at, or near the detection limit for the analytical method, or if the soil sample has not been well mixed prior to analysis.There are no certified reference samples for radionuclide and heavy metal soil extraction methods although ISE organise a quarterly inter-laboratory scheme for 0.1 m CaCl2 (B, Cd, Co, Cr, Cu), 1 m NH4NO3 (Cd, Cu, Ni, Pb, Zn) and 0.1 m NaNO3 (As, Cd, Cr, Cu, Hg, Ni, Pb, Tl, Zn). It is important that analysts should use an internal laboratory reference soil (with known concentrations of the ions of interest) with each batch of soils analysed. Results for this sample can then be used to provide information about possible bias in particular sets of extraction data, and to validate experimental data sets. In 1992 a workshop on the sequential extraction of trace metals in soils and sediments concluded that it was important to establish an internationally accepted protocol for sequential extraction procedures and to find a soil sample suitable for inter-laboratory comparison exercises with a view to establishing a certified reference material for extraction procedures.76 Interpretation of Extraction Data Data obtained from extraction methods are often difficult to interpret.Ions in extract solutions displace nutrient and pollutant ions from soil ion exchange sites. In the field ion exchange between soil and soil solution is triggered by processes such as changes in field moisture, leaching, root uptake of nutrients and micro-organism activity. Therefore ion exchange processes measured by the use of extractants can only be an approximate, but nevertheless a useful measure of bioavailability.Interpretation of data from sequential extraction schemes has additional problems such as cumulative error throughout the process, and difficulties in assigning specific soil phases to extract solutions in the series.27 Also, sequential extraction results are often presented as a percentage of a variable total. This practice can obscure consistent pollutant concentrations in the residual fraction in time series studies when the residual fraction increases proportionately because total soil pollutant concentrations have decreased over time.49 Conclusions The use of laboratory extraction techniques to predict radionuclides and heavy metal bioavailability will always be open to criticism.Methods vary between laboratories and lack a standard reference material. However, these techniques can give relative empirical information about possible soil phase associations of radionuclides and heavy metals.Extraction studies cannot predict the amount of pollutant that will be transferred from soils to plants during a specific period of time but they can provide a way of quantifying relative site, source and time dependent variations in soil-to-plant transfer of pollutants. Extractants such as 1 m NH4OAc pH 7 are also useful for studies of pollutant nutrient competition and associations (e.g., K+/Cs+ and Ca2+/Sr2+). Single extraction methods were originally developed to provide agriculturalists with a relative empirical method for assessing fertilisation requirements.They were then adapted to estimate the nutrient status of semi-natural ecosystems before being adopted for studies of radionuclide and heavy metal soil bioavailability. It is unlikely that universal extractants applicable to all soil types, for all radionuclides and heavy metals, and associated with specific soil phases can be devised although it is possible that a weak extractant, such as 1 m NH4OAc pH 7, could be used for a wide range of soils and analysed for a wide range of nutrients and pollutants. This extractant could also be buffered to the soil pH to extend the range of soils for which it can be used, but further work is needed to validate this method.Other extractants such as 0.01 m Ca Cl2, 0.1 m Na NO3 and 1 m NH4NO3 have also been found to be suitable for a range of mineral soils and associated heavy metals.Single extraction methods cannot quantify the soil pollutant fraction that will be available for root uptake over time, but they can provide a relative empirical method for evaluating potential availability of soil pollutants for plant uptake. Information about relationships between nutrient and pollutant ions and/or complexes extracted from the soil at the same time can also be obtained. Although sequential extraction methods were originally designed to study heavy metal and radionuclide phase associations in sediments they are now widely used to look at changes in both soil and sediment pollutant phase associations.Information from early stages in the extraction sequence is similar to that obtained from single extraction methods so it can be used to help predict potential short-term soil pollutant bioavailability by relating extracted soil concentrations to plant concentrations. Data from later stages in the sequence could also be developed into a useful tool to predict potential longer term radionuclide and heavy metal release into the soil labile pollutant pool (due to soil weathering and decomposition processes).Most published data for single and sequential extraction methods are site, time and vegetation specific and have not been validated for comparison studies and/or the development of generic models. The use of long-term laboratory reference soils and participation in well founded inter-laboratory comparison schemes would help validate extraction techniques and provide independent data points for data comparison.At present the only inter-laboratory scheme available is that organised by ISE for some heavy metals using 0.01 m Na NO3, 1 m NH4NO3 and 0.01 m CaCl2. Standard protocols also need to be agreed for soil sampling, sample preparation and extraction techniques to provide comparable data. There is little merit in having stringent extraction and analytical quality control procedures, validation and method protocols unless these are complemented by clear method protocols for sampling programmes and sample preparation procedures.When extraction or alternative techniques are used, soil sampling programmes should take account of potential seasonal effects so that seasonal variability can be separated from within site variability and between site variability. Ideally, sampling programmes should cover at least two years so that temporal changes can also be considered, and soil sampling protocols should be carefully defined. It is often difficult to make direct comparisons of published extraction data. Many papers do not clearly define extraction protocols used; in others protocols are clearly defined, but are sig- 98R Analyst, August 1997, Vol. 122nificantly different from one another so that results obtained cannot be readily compared.76 Even so, they can still provide useful information about potential changes in associations of radionuclide and heavy metal soil contamination with particular soil components.They can also be used to make comparative assessments of different soil types if extraction protocols are strictly defined and data are interpreted carefully.41 Thus, extraction methods can provide a relative empirical method for evaluating radionuclide and heavy metal soil contamination that may be available for plant uptake from the soil. They could also be used to provide information about potential soil pollutant bioavailability after accidental environmental contamination.However, it is important to understand the limitations of such techniques when relating pollutant concentrations measured in soil extract solutions to ecosystems. It should also be remembered that however good a method is for estimating what fraction of soil pollutants is potentially available for plant uptake, this is only one pathway for plant pollutant transfer. Others include aerial deposition, and resuspension; these may be important pathways, particularly in the immediate aftermath of accidental pollution releases to the environment.Information presented in this review is partly based on a desk study funded by the Ministry of Agriculture, Fisheries and Food, MAFF. The authors would like to thank Paul Naylor for his support as MAFF project officer and scientific colleagues at ITE Merlewood for encouragement and constructive comments. References 1 Gupta, S.K., and Aten, C., J. Environ. Anal. Chem., 1993, 51, 25. 2 McLaren, R. G., and Crawford, D. V., J. Soil Sci., 1973, 24, 172. 3 Oughton, D. H., and Salbu, B., in Nordic Radioecology: the Transfer of Radionuclides Through Nordic Ecosystems to Man, ed. Dahlgaard, H., Elsevier, Oxford, 1994, pp. 165–184. 4 Oughton, D. H., Salbu, B., Riise, G., Lien, H., Østby, G., and Nøren, A., Analyst, 1992, 117, 481. 5 Graham, E. R., and Killion, D. D., Soil Sci. Soc. Am. Proc., 1962, 26, 545. 6 Lakanen, E., and Paasikallio, A., Ann. Agric. Fenn., 1970, 9, 133. 7 Cremers, A., Elsen, A., De Preter, P., and Maes, A., Nature, 335, 1988, 247. 8 Evans, D. W., Alberts, J. J., and Clark, R. A., Geochim. Cosmochim. Acta, 1983, 47, 1041. 9 Francis, A. J., in Environmental Migration of Long-lived Radionuclides, International Atomic Energy Agency, Vienna, 1982, pp. 415–429. (IAEA-SM-257/72). 10 Haselwandter, K., Berreck, M., and Brunner, P., Trans. Brit. Mycol. Soc., 1988, 90, 171. 11 Horyna, J., and Randa, Z., J. Radioanal. Nucl. Chem. Lett., 1988, 127, 107. 12 Haugen, L. E., and Uhlen, G., in Radioaktivt nedfall fra Tsjernobylulykken, ed Garmo, T. H., and Gunnerod, T. B., Norges Landbruksvitenskapelige Forskningsrad, Oslo, 1992, pp. 43–64. 13 Heinrich, G., J. Environ. Radioact., 1993, 18, 229. 14 Chebotina, M. Y., and Kulikov, N. V., Sov. J. of Ecol., 1973, 4, 84. 15 Kulikov, N. V., Molchanova, I. V., Chebotina, M. Y., and Karavaeva, E. N., Sov. J. Ecol., 1977, 8, 162. 16 Valcke, E., and Cremers, A., Sci. Total Environ., 1994, 157, 275. 17 Grimshaw, H. M., in Chemical Analysis of Ecological Materials, ed. Allen, S. E., Blackwell Scientific Publications, Oxford, 1989, pp. 7–45. 18 Hesse, P. R., A Textbook of Soil Chemical Analysis, John Murray, London, 1971. 19 Page, A. L., Miller, R. H., and Kenney, D. R, Methods of Soil Analysis, Chemical and Micro-biological Properties, American Society of Agronomy, Madison, 2nd edn., 1982. 20 Ministry of Agriculture Fisheries and Food, The Analysis of Agricultural Materials: A Manual of Analytical Methods used by the Agricultural and Development and Advisory Service, (Reference book 427), Her Majesty’s Stationery Office, London, 3rd edn., 1986. 21 Legret, M., Int.J. Environ. Anal. Chem., 1993, 51, 161. 22 F�orstner, U., Int. J. Environ. Anal. Chem., 199 51, 5. 23 Wilkins, B. T., Green, N., Stewart, S. P., Major, R. O., and Dodd, N. J., in Seminar on the Behaviour of Radionuclides in Estuaries.Renesse 17-21 September 1984, Council for the European Community, Luxembourg, 1985, 55–70. 24 Nirel, P., Thomas, A. J., and Martin, J. M., in Speciation of Fission Activation Products, ed. Bulman, R. A., and Cooper, J. R., Elsevier Applied Science, London, 1986, 19–26. 25 Beckett, P. H. T., Adv. Soil Sci., 1989, 9, 143. 26 Rigol, A., Vidal, M., and Rauret, G., J. Radioanal. Nucl. Chem., 1996, 208, 617. 27 Ure, A. M., Mikrochim. Acta, 1991, II, 49. 28 Adriano, D. C., Delaney, M. S., Hoyt, G. D., and Paine, D., Environ. Exp. Bot., 1977, 17, 69. 29 Wilson, D. O., and Cline, J. F., Nature, 1966, 209, 941. 30 Thomas, G. W., in Methods of Soil Analysis Part II Chemical and Microbial Properties, ed. Page, A. L., Miller, R. H., and Keeney, D. R., Madison, Wisconsin, 1982, pp. 159–165. 31 Oughton, D. H., Skipperud, L., and Tronstad, E., Cs and Sr Mobility in Nordic Soils, Nordic Reactor Safety Commission, Risø, 1995. 32 Oughton D.H., Børretzen, P., Salbu, B., and Tronstad, E., Sci. Total Environ., in the press. 33 Evans, E. J., and Dekker, A. J., Can. J. Soil Sci., 1969, 49, 349. 34 Tessier, A., Campbell, P. G. C., and Bisson, M., Anal. Chem., 1979, 51, 844. 35 Martin, J. M., Nirel, P., and Thomas, A. J., Marine Chem., 1987, 22, 313. 36 Paasikallio, A., Ann. Agric. Fenn., 1984, 23, 109. 37 Novozamsky, I., Lexmond, T. M., and Houba, V. J. G., Int. J. Environ. Anal. Chem., 1993, 51, 47. 38 Houba, V. J.G., Lexmond, Th. M., Novozamsky, I., and Van der Lee, J. J., Sci. Total Environ., 1996, 178, 21. 39 Houba, V. J. G., Novozamsky, I., and Temminghoff, E., Soil and Plant Analysis 5A: Soil Analysis Procedures Extraction With 0.01 M CaCl2, Wageningen Agricultural University, Wageningen. 40 Livens, F. R., Baxter, M. S., and Allen, S. E., in Speciation of Fission and Activation Products in the Environment, ed. Bulman, R. A., and Cooper, J. R., Elsevier, London, 1986, pp. 143–150. 41 Wilkins, B. T., Green, N., Stewart, S. P., and Major, R. O., in Speciation of Fission and Activation Products in the Environment, ed. Bulman, R. A., and Cooper, J. R., Elsevier, New York, 1986, pp. 101–113. 42 Fawaris, B. H., and Johanson, K. J., in Behaviour of 137Cs in the Boreal Forest Ecosystem of Central Sweden, PhD Thesis by B. Fawaris IV, Swedish University of Agricultural Science, Uppsala, 1995. 43 Fawaris, B. H., and Johanson, K. J., Sci. Total Environ, 1995, 170, 221. 44 Kennedy, V.H., and Horrill, A. D., in Radiation Doses and Pathways to Man From Semi-natural Ecosystems, ed. McGarry, A., (Commission of the European Communities contract F13-CT920058; sector A25), Radiological Protection Institute of Ireland, Dublin, 1995, pp. 16–30. 45 Chang, A. C., Page, A. L., Warneke, J. E., and Grgurevic, E., J. Environ. Qual., 1984, 13, 33. 46 Emmerich, W. E., Lund, L. J., Page, A. L., and Chang, A. C., J. Environ. Qual., 1982, 11, 178. 47 Vidal, M., Tent, J., Llaurad�o, M., and Rauret, G., J.Radioecol., 1993, 1, 49. 48 Ure, A. M., Quevauviller, P., Muntau, H., and Griepink, B., Int. J. Environ. Anal. Chem., 1993, 51, 135. 49 Nirel, P. M. V., and Morel, F. M. M., Water Res., 1990, 24, 1055. 50 Belzile, N., Lecomte, P., and Tessier, A., Environ. Sci. Technol., 1989, 23, 1015. 51 Nisbet, A. F., and Lembrechts, J. F., in Transfer of Radionuclides in Natural and Semi-natural Environments, ed.Desmet, G., Nassimbeni, P., and Belli, M., Elsevier Applied Science, London, New York, 1990, pp. 371–381. 52 Sweeck, L., Wauters, J., Valke, E., and Cremers, A., in Transfer of Radionuclides In Natural and Semi-natural Environments, ed. Desmet, G., Nassimbeni, P., and Belli, M., Elsevier Applied Science, London, New York, 1990, pp. 249–258. Analyst, August 1997, Vol. 122 99R53 Wauters, J., Madruga, M. J., Vidal, M., and Cremers, A., Sci. Total Environ., 1996, 187, 121. 54 Livens, F. L., Howe, M. T., Hemingway, J.D., Goulding, K. W. T., and Howard, B. J., Eur. J. Soil Sci. 1996, 47, 105. 55 Absalom, J. P., Young, S. D., and Crout, N. M. J., Eur. J. Soil Sci., 1995, 46, 461. 56 Absalom, J. P., Young, S. D., and Crout, N. M. J., Environ. Sci. Technol., 1996, 30(9), 2735. 57 Gillman, G. P., A Centrifuge Method for Obtaining Soil Solution, (Division of Soils Divisional Report 6), Commonwealth Scientific and Industrial Research Organisation: Townsville 1976. 58 Reynolds, B., Plant Soil, 1984, 78, 434. 59 Keller, C., Commun. Soil Sci. Plant Anal., 1995, 26, 1621. 60 Webster, R., and Oliver, M. A., Statistical Methods in Soil and Land Resource Survey, Oxford University Press, Oxford, 1st edn., 1990, 315 pp. 61 Crepin, J., and Johnson, R. L., Soil Sampling for Environmental Assessment, in Soil Sampling and Methods of Analysis, ed. Carter, M. R., Lewis Publishers, London, 1993. 62 Rubio, R., Int. J. Environ. Anal. Chem., 1993, 51, 205. 63 Kratochvil, B., Wallace, D., and Taylor, J.K., Anal. Chem., 1984, 56, 113R. 64 Petersen, R. G., and Calvin, L. D., in Methods of Soil Analysis: Part 1 Physical and Mineralogical Methods, ed. Klute, A., American Society of Agronomy, Madison, 2nd edn., 1986, pp. 33–52. 65 Rubio, R., and Ure, A. M., Int. J. Environ. Anal. Chem., 1993, 51, 205. 66 McGee, E. J., Keatinge, M. J., Synnott, H. J., and Colgan, P. A., Health Phys., 1995, 68, 320. 67 Kennedy, V. H., and Sanchez, A. L., in preparation. 68 Maher, W. A., Bull.Environ. Contam. Toxicol., 1984, 32, 339. 69 Jackson, M. L., Soil Chemical Analysis, Prentice Hall, London, 1958. 70 Riise, G., Bjørnstad, H. E., Lien, H. N., Oughton, D. H., and Salbu, B., J. Radioanal. Nucl. Chem., 1990, 142, 531. 71 Orsini, L., and Bermond, A., Int. J. Environ. Anal. Chem., 1993, 51, 97. 72 Riise, G., Salbu, B., Singh, B. R, and Steinnes, E., Water, Air, Soil Poll., 1994, 78, 286. 73 Gupta, S. K., and Chen, K. Y., Environ. Lett., 1975, 10(2), 129. 74 Harrison, A. F., personal communication. 75 Kennedy, V. H., Rowland, A. P., and Parrington, J., Commun. Soil Sci. Plant Anal., 1994, 25, 1605. 76 Quevauviller, P., Rauret, G., and Griepink, B., Int. J. Environ. Anal. Chem., 1993, 51, 231. 77 Pay�a-P�erez, A., Sala, J., and Mousty, F., Int. J. Environ. Anal. Chem., 1993, 51, 223. 78 Kheboian, C., and Bauer, C. F., Anal. Chem., 1987, 59, 1417. 79 Clayton, P. M., and Tiller, K. G., A chemical method for the determination of the heavy metal content of soils in environmental studies, (Division of Soils Technical Papers, 41).CSIRO, Melbourne, 1979, 17 pp. 80 Haq, A.U., and Miller, M. H., Agron. J., 1972, 64, 779. 81 Andersson, K. G., and Roed, J., J. Environ. Radioact., 1994, 22, 183. Paper 7/04133K Received June 13, 1997 Accepted June 27, 1997 100R Analyst, August 1997, Vol. 122 Critical Review Use of Single and Sequential Chemical Extractants to Assess Radionuclide and Heavy Metal Availability From Soils for Root Uptake V.H. Kennedy*a, A. L. Sancheza, D. H. Oughtonb and A. P. Rowlanda a Institute of Terrestrial Ecology, Merlewood Research Station, Windermere Road, Grange-over-Sands, Cumbria UK LA11 6JU b Laboratory for Analytical Chemistry, Agricultural University of Norway, P.O. Box 5026, Ås NLH-7754 Norway Summary of Contents Introduction Role of Extractants in Predicting Bioavailability Single Extractants Sequential Extraction Schemes Other Methods Used to Estimate Soil Bioavailability Experimental Aspects of Extraction Techniques Conclusions Keywords: Extraction methods; bioavailability; plant uptake; soil phases; data interpretation Introduction To predict the persistence and potential mobility of radionuclide and heavy metal contamination in soils it is necessary to devise a robust chemical or biological method to quantify the fraction of soil contamination available for root uptake.Such information is important for predictive models and environmental assessments where the soil–plant pathway is a key contributor to potentially harmful effects to plants and animals, e.g., estimates of pollutant concentrations likely to interfere with normal plant growth and of potential radionuclide dose-to-man.Single and sequenxtraction methods have been used for this purpose. This review discusses their application in radioecological and pollutant studies, and considers problems in interpretation of data produced.Following deposition and/or discharge, the distribution of pollutants between soil solution and soil solid phases gradually reaches equilibrium and may become associated with soil–plant nutrient cycling processes. The rate at which equilibrium is reached depends on the form of the original pollution, and on soil characteristics: this process may take years. Soil is an environmental, biochemical reaction system with three important phases: solid (mineral particles, organic debris, plant roots), solution (groundwater, rain water, biological excreta, products of biochemical reactions) and gas (atmospheric, products of biochemical reactions) which move towards equilibrium with one another. Radionuclide and heavy metal ions bound to the solid phase can be mobilised into the solution phase by changes in soil pH, temperature, redox potential, soil organic matter decomposition, leaching and ion exchange processes and by microbial activity.The soil solution phase (soil water) includes nutrients and metals in ionic form and organic and inorganic complexes (charged and/or neutral species).Soil solution is thought to play an important role in plant nutrient and pollutant uptake from soils and soil-to-plant cycling processes. For instance, heavy metals are mobilised into soil solution by a decrease in pH and by decomposition of humus.1 Once radionuclides and heavy metals enter the soil their form changes as they interact with other soil components.2,3 Initially, the bioavailability of soil radionuclide contamination is influenced by the physico-chemical form of the fallout.4 Thereafter, the form/speciation of soil contamination is likely to change with time due to dynamic biological and biochemical soil processes including organic and mineral cation exchange, complex formation, leaching and mineral weathering.For instance, radionuclides and heavy metals may become adsorbed to mineral particles,5–8 or incorporated into micro-organisms9 and/or fungi10–13 or adsorbed to decomposing organic particles or incorporated into organic complexes in the soil.14–16 As soil solution seeps down the soil profile it provides water and nutrition for plants and soil micro-organisms.If practicable, analysis of soil solutions would be an ideal way of measuring soil radionuclide and heavy metal bioavailability with time. Valerie Kennedy is a longstanding member of the Environmental Chemistry Section, Institute of Terrestrial Ecology (Merlewood Research Station).This research group has a proven team expertise in the development and validation of nutrient and pollutant analytical techniques for environmental samples. Her main research interests focus on ways of quantifying soil-to-plant transfer of nutrients and pollutants and on quality assurance methods for analytical techniques. For the last ten years she has also collaborated with the Merlewood Radioecology research team.She was Secretary to the NERC Steering Committee on Environmental Radioactivity (1987–1990) and was a joint guest editor to a special edition of Science of the Total Environment (1993)—Chemical analysis of natural waters in environmental research: analytical solutions to current problems. She is also the ITE Merlewood Safety Adviser and a member of the Institute of Occupational Safety and Health. Analyst, August 1997, Vol. 122 (89R–100R) 89RUnfortunately, it is difficult to obtain samples of soil solution that are representative of a specific volume of soil, and that contain concentrations of soil contaminants above analytical detection limits.A compromise has been the development of laboratory based extraction methods using field soils. The main objective of this review is to discuss the use of extraction methods to quantify soil radionuclide availability for plant uptake. Information about heavy metal extraction methods have been included for comparative purposes because similar problems are encountered.This review also comments on the use of extraction methods to identify pollutant soil phase associations because changes in these can affect potential soil pollutant availability for root uptake. There is also a wide body of literature on sediment extraction techniques; some relevant references have been cited where appropriate, but studies of freshwater or marine sediments are outside the scope of this review.Role of Extractants in Predicting Bioavailability There are no simple analytical methods that simulate the action of nutrient uptake by plants, therefore, a wide variety of empirical extraction techniques have been adopted.17–20 Many extraction methods are based on the assumption that ions associated with soil components will be displaced from adsorption sites by the presence of an excess of competing ions in the extract solution. As discussed by Grimshaw17 nutrient concentrations measured in extractants can only provide an estimate of potential nutrient availability for plant uptake or for cation exchange between soil and soil solutions. The same applies to radionuclide and heavy metal concentrations measured in extract solutions.Heavy metals and radionuclides associated with exchangeable phases, as measured by extractants, are assumed to be easily mobilised by soil–soil solution ion exchange reactions and are therefore available for root uptake.Pollutants associated with oxidisable phases, as measured by more chemically aggressive extractants, are assumed to remain in the soil for longer periods but may be mobilised by decomposition processes21 (e.g., weathering and microbial activity). However, reactions taking place in the laboratory during extraction are non-selective. They are likely to be influenced by the length of extraction time and the soil mass : volume ratio used.22 Therefore, radionuclide and heavy metal concentrations measured in the extract solution are defined by the extractant and the experimental protocol.23,24 They provide a relative empirical measurement of the amounts of radionuclide and heavy metal contamination that may be available for plant uptake at the time of the measurement, or over a longer period of time.Actual uptake by plants is determined by other factors such as the number of competing ions in soil solution and root uptake processes which vary between plant species.A range of extraction solutions have been used because they are thought to bring soil pollutants associated with particular soil fractions into solution (Table 1). It is often implied that pollutant concentrations measured in extract solutions can be Table 1 The role of extractants used to assess the potential bioavailability of radionuclides and heavy metals for plant uptake Soil/sediment Dominant fraction extracted Reference Extractant pH (authors’ description) (examples only) H2O (distilled/deionised) Soils Water soluble exchangeable soil–soil 4, 33, 41, 44, 45 solution cations 0.43 m HOAc Soils More tightly bound inorganic soil 2,77 pH 3.2–7.3 TN adsorption sites (clay minerals; small amounts organic and/or oxide bound). 1 m NH4OAc Soils § 3, 28, 29, 33 õ NH4OAc pH 4.65 Soils õ 39 1 m NH4NO3 * Soils õ 1 0.005 m CaCl2 Soils õ 55 0.01 m CaCl2 † Soils õ 37, 38, 39 0.2 mm DTPA Soils õ 29 0.001 m DTPA Soils õ 28 pH 5.1–7.8 ¢ Exchangeable nutrient and pollutant 0.01 m EDTA–1 m Soils ½ soil–soil solution cations. 28 (NH4)2CO3 pH 5.1–7.8 õ 0.1 m HCl Soils õ 28 pH 5.1–7.8 õ 0.1 m HNO3 Soils õ 28 1 m HNO3 Soils õ 29 0.05 m KCl Soils õ 45 1 m MgCl2 Sediments õ 34, 35, 68, 78 MgCl2 pH 7 Sediments Soils Ä 26, 47 0.4 m EDTA Soils § 79, 80 0.1 m K4P2O7·H2O Soils õ 2 pH 3.2–7.3 ¶ Organically bound 0.05 m NaOH Soils õ 45 0.1 m NaNO3 ‡ Soils Ä 1 0.05 m Na2EDTA Soils § 45 1 m NaOAc adjusted to pH Sediments, Soils ¶ Carbonate bound 4, 26, 34, 35, 41, 79 5 with HOAc Ä 0.04 m NH4OAc in 25% v/v Sediments, Soils Bound to iron and manganese oxides 4, 26, 34, 35, 41, 79 HOAc 0.1 m HOAc Soils Mainly inorganic sites (specific sorption) 2 pH 3.2–7.3 The International Soil Exchange (ISE) organises a quarterly inter-laboratory comparison scheme for: * As, Cd, Cr, Cu, Hg, Ni, Pb, Tl, Zn.† B, Cd, Co, Cr, Cu. ‡ Cd, Cu, Ni, Pb, Zn. 90R Analyst, August 1997, Vol. 122used to quantify radionuclides and heavy metals associated with specific soil phase associations; however, extractants may also remove small but variable amounts of pollutants associated with other soil fractions.This lack of scientific rigor has been compounded by the use of published extraction methods and sequences by non-specialists.25 A study by Rigol et al.26 clearly illustrates the difficulty of assigning soil phases to specific extractants. They used 0.1 mol l21 NaOH under N2 and 0.1 mol l21 Na4P2O7, to estimate radiocaesium activity concentrations associated with humic acid in a range of soils.They found that humic acid concentrations extracted by 0.1 mol l21 NaOH were consistently higher than those extracted by 0.1 mol l21 Na4P2O7 (for soils with 14–88% organic matter content). They suggested that the constant ratio between the two extractants for the amount of humic acid extracted for these soils indicated that differences were related to the extractant, rather than differences in soil characteristics.They also found good correlation between percentage organic matter and percentage extracted humic acid for these soils in both extract solutions. However, they concluded that 0.1 mol l21 NaOH extracted humic acid more efficiently because it had a higher correlation with the organic matter content and would therefore give a better estimate of radiocaesium associated with humic acid for most soil types. Single Extractants A number of different chemical extractants have been used to make a relative empirical assessment of heavy metal/radionuclide availability for plant uptake from different components of the soil (Table 1).Radionuclides and heavy metals brought into the extract solution are either defined as ‘plant available’, or attributed to a soil phase such as carbonates or hydrous oxides which may or may not be available for root uptake. Ideally, the extractant selected should not affect the soil solid " solution equilibrium, therefore it should not change the soil pH, form complexes, or take an active part in ion exchange reactions.Unfortunately, extractants are not soil phase-specific. For example, 0.1 m ethylenediaminetetraacetic acid (EDTA) will extract heavy metals associated with carbonate, organic and ‘available’ soil phases.27 Ure27 tabulated information about a number of single extractants where soil extract concentrations have been reported to correlate with plant concentrations for elements such as Cd, Cu, Zn, Ni, Pb, Se, Mo and Cr.Gupta and Aten1 evaluated several extractants for assessing soil bioavailability of Cu, Zn, and Cd for lettuce and rye grass growing on study soils. These included 2 m HNO3, 0.52 m NH4OAc + EDTA, 0.05 m CaCl2, 0.1 m NH4NO3, 0.1 m KNO3 and 0.01 m CaCl2. They recommended the neutral extractant 0.1 m NaNO3 for bioavailability studies of heavy metal soil contamination because it showed a higher correlation with plant heavy metal concentrations when compared with other extractants used in the study.Adriano et al.28 noted that it was important that pollutant concentrations measured in extract solutions should be able to predict the ‘plant availability’ of the ion being studied for a wide range of soil types, and that it should also be sensitive to the ion where availability is suspected to be minimal. Of five extractants evaluated (0.1 m HCl, 0.1 m HNO3, 1 m EDTA– NH4CO3, 0.001 m DTPA and 1 m NH4OAc) for estimating the pattern of plant uptake of 241Am from a sandy loam and a clay loam, 1 m NH4OAc gave consistent correlation coefficients for plant uptake of 241Am for bush bean (Phaseolus vulgaris) and for corn (Zea mays).They suggested that this extractant was a potentially useful tool for diagnosing plant availability of 241Am from contaminated environments. Wilson and Cline29 used four extractants: 1 m NH4OAc, 0.1 m HNO3, 1 m HNO3 and 0.2 mm DTPA (calcium trisodium salt) to evaluate the extractability of 239Pu from three different soils (a slightly calcareous sandy loam, a calcareous silt loam and a medium acid forest soil).All four extractants extracted a very small proportion of the total 239Pu activity from the soil; the highest amount of 239Pu extracted by 0.1 m HNO3 was only 0.64% of the total 239Pu soil activity, therefore, it was not possible to evaluate the extractants for plutonium. Extracted concentrations of 85W were higher; the amount extracted by 1 m NH4OAc followed a similar pattern for all three soils and was therefore their favoured extractant.One molar NH4OAc pH 7 has also been found to be a robust extractant for acidic and neutral soils.17 It has provided useful data on soil nutrient status for agriculturalists assessing fertilisation requirements and for environmental researchers studying semi-natural ecosystems. 20,30 A more recent modification to the use of NH4OAc has been to adjust the extractant to the pH of either the soil or of the soil solution, with acetic acid.31,32 Evans and Dekker33 used water and 1 m NH4OAc extractions in association with a greenhouse experiment and an incubation experiment.The experiments were designed to examine the effects of potassium and calcium carbonates on 137Cs availability for uptake by oats and alfalfa from a range of nine Canadian acidic and alkaline silts and sandy loam soils. Results from water extractions were used to estimate the percentage of soluble 137Cs and results from 1 m NH4OAc extractions to estimate the percentage of exchangeable 137Cs in the experimental soils.They found that the effect of fertiliser applications on 137Cs availability for plant uptake varied with soil type, and with the timing of fertiliser application. Addition of K reduced plant uptake of 137Cs for all the study soils without significantly reducing the amount of extractable 137Cs in their soils. These results illustrate that extraction procedures can only measure what is potentially available for plant uptake; other factors, such as soil nutrient concentrations, may also influence what proportion of soil pollutant is actually taken up by plants growing in the soil.However, these data also demonstrate that empirical extraction techniques can be used to study possible interactions between pollutant and nutrient ions in soil-to-plant cycling processes. Although a number of researchers have opted for 1 m NH4OAc pH 7 as a suitable extractant to estimate ‘exchangeable’ ion concentrations in soils, Tessier et al.34 felt that it was unsuitable because it may also attack carbonates within aquatic sediments.As an alternative they compared two other extractants, 1 m MgCl2 pH 7 and 1 m NaOAc pH 8.2 and concluded that 1 m MgCl2 pH 7 was the most suitable extractant because it extracted a smaller percentage of carbonate bound ions. Similarly, Ure27 noted that 0.1 m EDTA will extract trace metals bound or associated with carbonate phases as well as those associated with organic matter.Martin et al.35 suggested that single extractants are likely to permit dissolution of some non-labile organic and inorganic fractions as well as labile fractions. It is generally accepted that 1 m NH4OAc pH 7 is unsuitable for alkaline soils, although adjusting its pH to 9 can be a viable alternative.17 Paasikallio36 used NH4OAc pH 4.65 for a time series of extractions on soil sub-samples from a two year rye grass pot experiment, using seven soil types. Extractable 137Cs activity decreased slightly for the first few months and then maintained a similar concentration for all soil types except the carex peat which decreased with time throughout the experiment.Strontium- 90 behaved differently; after an initial decrease the amount extracted gradually increased with time for all soils during the experiment.Plant uptake of 90Sr and 137Cs was reported as a percentage of the radioactivity originally applied to the soil, rather than as concentrations brought into solution by the extractant. Surprisingly, no information about relationships between plant concentrations and extracted soil concentrations was presented for either element. Novozamsky et al.37 used single extraction procedures to assess soil availability of Cd, Cu, Ni and Zn to plants. A range of extractants and mineral soil types were included, and Analyst, August 1997, Vol. 122 91Rrelationships between extraction concentrations and plant yield for green beans, lettuce and endives were evaluated. They concluded that 0.01 m CaCl2 was the most useful extractant for their mineral soils because it had approximately the same salt concentration as the ‘average soil solution’, a range of heavy metals and nutrients could be measured in the same extract solution (so that interactions between a range of ions brought into solution could be studied), and because Ca2+ is often the dominant cation competing for soil adsorption complex sites in mineral soils. Similarly, McLaren and Crawford,2 favoured 0.05 m CaCl2 as the first extraction stage of their sequential extraction series because it had less effect on the natural soil pH and extracted fewer ions associated with soil organic matter. Another recent study recommends the use of 0.01 m CaCl2 because it has a similar pH, concentration and composition to soil solutions.38 Correlations between Cd, Cu, Pb, Ni and Zn brought into solution by 0.01 m CaCl2 and the appearance of toxicity symptoms in plants growing on mineral soils were reported to be more significant than relationships between total soil and plant concentrations for the same metals.This extractant has the added advantage that detailed extraction method protocols have been published for B, Cd, Cu, Pb, Ni and Zn.39 A different approach was developed by another research team because extractable radiocaesium activity concentrations in the nutrient deficient soils they were studying were relatively small compared with radiocaesium activity concentrations in the plants growing on their soils.3 For example, radionuclide activity concentrations in plants represented 30% of the total labile radionuclide pool in some Norwegian soil–plant systems. 3 For some soil types (e.g., organic, nutrient deficient) there can be significant seasonal variations in extracted pollutant concentrations due to depletion of the soil labile pollutant pool in spring, and the release of pollutants from fallen litter during decomposition processes in autumn, in some ecosystems.This phenomenon led to the development of the ‘mobility factor’ method (addition of plant extractable and soil extractable radionuclide contamination). They used water and 1 m NH4OAc extractions of soil and vegetation in an attempt to differentiate between ‘inert’ and ‘mobile’ fractions of radionuclides in soil–water–plant systems.The data were then used to calculate a ‘mobility factor’ which was defined as the percentage of deposited radionuclide present in a labile or potentially bioavailable form at a given time. They suggested that the ‘mobility factor’ represented the total fraction of deposited radionuclide in the soil–water–plant system that is present in a labile form, and readily available for plant uptake at the time of analysis.Time dependent studies on the variation in bioavailability at single sites gave promising results; over six years, a decrease in the labile fraction of 137Cs in soils was correlated with a decrease in soil-to-plant transfer.31 Actual plant uptake was found to be dependent on other factors such as vegetation type, season, microbial activity, available potassium and stable caesium. Such seasonal differences can also be quantified by analysis of litter extract solutions, in addition to surface soil extract solutions, and associated total plant concentrations.It is apparent that a number of single extraction methods can provide a relative empirical method for evaluating the potential availability of soil pollutants for plant uptake. Information about relationships between other nutrient and pollutant ions and/or complexes extracted from the soil at the same time can also be obtained. There is a growing body of evidence that 1 m NH4OAc could be adopted as a suitable empirical soil extractant for a wide range of soils and analysed for a range of nutrients and pollutants (including radionuclides).Other extractants such as 0.01 m CaCl2 (B, Cd, Co, Cr, Cu), 1 m NH4NO3 (Cd, Cu, Ni, Pb, Zn) and 0.1 m NaNO3 (As, Cd, Cr, Cu, Hg, Ni, Pb, Tl, Zn) have also been identified as suitable extractants for mineral soils. Use of these extractants has the added advantage that the International Soil Exchange (ISE) organise a quarterly interlaboratory comparison scheme for the elements listed.Sequential Extraction Schemes Heavy metals and radionuclides are present in many forms.2,3 They may be in soil solution (ionic, organic complexes, inorganic complexes), on soil cation exchange sites, more tightly bound to soil adsorption sites, occluded into soil oxide material, incorporated into organic plant litter, incorporated into soil micro-organisms and/or soil animals or in the lattice structure of primary and secondary soil minerals.2,3 Sequential extraction methods were originally designed to study some of these phase associations in soils and sediments.2,34,40,41 However, pollutant concentrations measured in sequential extraction solutions can also be used to differentiate between short and long term soil radionuclide and heavy metal bioavailability.Most sequential extraction schemes are based on the scheme devised by Tessier et al.34 which was designed to bring metal pollutants associated with different aquatic sediment phases into solution.The phases are operationally defined as water soluble (H2O), exchangeable (1 m MgCl2), carbonate bound (1 m NaOAc adjusted to pH 5 with HOAc), bound to Fe–Mn oxides (0.04 m NH4OH, HCl in 25% v/v HOAc), bound to organic matter (H2O2 adjusted to pH 2 with HNO3) and residual (HF–HClO4digestion). The increasing strength of extractants down the extract series can be used to predict pollutant soil associations and hence potential bioavailability.Initial extraction stages in sequential extraction schemes remove soluble and/or ion exchangeable pollutants, therefore, it is not surprising that extractants commonly used for single extraction studies are selected for initial stages in sequential extraction schemes. Thus, pollutant concentrations measured in these extract solutions can be used to make a relative empirical estimate of pollutant availability for root uptake from the soil, although few researchers have directly related their findings to plant uptake (Table 2).The use of extract solutions with a gradual increase in displacement/dissolution strength down the extraction scheme provides additional information about soil pollutant fractions that may be released from soil phase associations and become available for plant uptake because of weathering, decomposition and/or other soil processes, e.g., the pollutant fraction that is oxidisable or mobilised if the pH is lowered to 5, or the fraction that may be trapped within the crystal lattice of soil mineral particles.Hence, sequential extraction can give an indication of the binding strength of soil pollutants and their potential for mobilisation over time due to changes in soil chemistry.4,42,43 Aquatic sediment characteristics, such as organic matter type and amount, redox potential and salinity, differ significantly from soil characteristics. Such differences should be taken into account when adapting the Tessier sequential extraction scheme for soils.Such schemes usually include three to six sequential extraction stages (Table 3). The physical characteristics of study soils often determine the extraction sequence selected. For instance, a carbonate extraction stage needs to be included where carbonates may play an important role in mobilising heavy metals bound to carbonates in sediments34 but may be inappropriate for acidic peats with negligible calcium carbonate concentrations.44 Similarly, the use of a hydrogen peroxide step to oxidise organic matter is not likely to provide useful information for soils with high organic matter content.Such soil differences often account for differences in the number of extraction stages included in each scheme (Tables 2 and 4). In the 1970s, McLaren and Crawford2 studied copper–soil associations and mobility in 24 UK soils with pH values between 3.2 and 7.3.As part of their work they developed a 92R Analyst, August 1997, Vol. 122sequential extraction method using the following sequence: 0.05 m CaCl2 (weakly bound, soluble), 0.43 m HOAc (extractable, mainly inorganic), 1 m K4P2O7·H2O [corrected to 0.1m, J. Soil Sci., 1974, 25 (1), 119; organic], 0.1 m (NH4)2 C2O4·H2O (occluded by free oxides) and HF (residual). They looked at correlations between soil characteristics such as percentage organic carbon, percentage clay and pH and soil copper fractionation as measured by their extraction series (Table 5).Chang et al.45 used a five stage sequential extraction scheme developed by Emmerich et al.46: 0.5 m KNO3 (exchangeable fraction), H2O (adsorbed fraction), 0.5 m NaOH (organic fraction), 0.05 m Na2 EDTA (carbonate fraction) and 4 m HNO3 (sulfide/residual fraction). They looked at soil associations of Cd, Cr, Cu, Ni, Pb and Zn after sludge applications. They also measured total cadmium and zinc concentrations in barley growing on their experimental plots but did not comment on relationships between plant concentrations and concentrations measured in soil extract solutions.Tessier et al.’s34 scheme for sequential extraction of sediment and soil samples was modified by Wilkins et al.41 to identify the soil components with which 137Cs, 239/240Pu, 106Ru, 241Am, 129I and 90Sr were associated. They concluded that 90Sr is mainly associated with the soil ion exchange phase, whereas 241Am was mainly associated with soil oxide and organic phases. Informa- Table 2 Examples of sequential extraction schemes that have been used for heavy metal and radionuclide analysis of soils Number of Soil–plant extraction Sample transfer Method Reference stages preparation Elements discussed evaluated (examples only) 3 Soil, air dried 137Cs, 90Sr, 110Ag No No 47 (traditional) 2 mm sieved 4 Soil, air dried 137Cs, 90Sr, 110Ag No No 47 (organic role) 2 mm sieved 4 Soil, dried at 50 °C 137Cs, 106Ru No Yes 81 2 mm sieved 4 Soil, air dried 137Cs Yes No 42, 43 2 mm sieved 4 Model sediment Pb, Ni, Cu, Zn No Yes 34 4 Soil, dried at 50 °C 137Cs, 106Ru No Yes 81 2 mm sieved 5 Soil, air dried Cu Yes Yes 2 1 mm sieved 5 Soil Pb, Ni, Cu, Zn, Cd, Cr Yes Yes* 45 Field moisture 5 Soil, air dried, 239,240Pu No Yes* 42 1 mm sieve 5 Model sediment Pb, Ni, Cu, Zn No Yes 78 5 Artificial As, Cs, Hf, Zr, No Yes 35 substrate Ce, Co 5 Artificial As, Ce, Co, Cs, Hf, Zr, Th No Yes 26 substrate 5 Soil, fresh 137Cs Yes Yes 44 6 Soil, air dried, 134Cs, 241Am, 106Ru, No Yes* 41 1 mm sieve 90Sr, 239Pu, 129I 6 Soil, air dried 137Cs, 90Sr Yes Yes 4 * Method evaluated independently by other authors.Table 3 Examples of single extractants that have been used to assess the potential bioavailability of radionuclides and trace elements for plant uptake Sample Method Reference Extractant preparation Elements evaluated (examples only) H2O (distilled) Not clear 137Cs Yes* 33 1 m NH4OAc Air dried 239Pu Yes* 29 Dried, 70 °C 137Cs* Yes 3 241Am Yes 28 Air dried 137Cs, 90Sr Yes* 3 NH4OAc pH 4.65 Air dried 137Cs, 90Sr 36 0.1 m HCl Oven dried 241Am Yes 28 70 °C 0.1 m HNO3 Oven dried 241Am Yes 28 70 °C 1 m HNO3 239Pu Yes* 29 0.2 mm (DTPA) 239Pu Yes* 29 0.01 m CaCl2 Oven dried Cd, Cu, Ni, Zn Yes 40 0.001 m DTPA Oven dried 241Am Yes 28 70 °C 0.01 m EDTA– Oven dried 241Am Yes 28 1 m (NH4)2CO3 70 °C * Method evaluated independently by other authors. Analyst, August 1997, Vol. 122 93RTable 4 Examples of extractants that have been used for steps 1–6 of extraction schemes Soil components extracted Extract Reference Extractant (authors’ description) number Elements discussed (examples only) H2O (distilled/deionised) Water soluble 1 134Cs, 241Am, 106Ru, 90Sr, 41 239Pu, 129I 1 137Cs, 90Sr 3, 4 1 137Cs 42, 43 1 and 2 137Cs 44 0.05 m CaCl2 Weakly bound soluble 1 Cu 2 1 m MgCl2 Exchangeable 1 Pb, Ni, Cu, Zn 34 1 Pb, Ni, Cu, Zn 78 1 As, Cs, Hf, Zr, Ce, Co 35 1 m MgCl2 1 As, Ce, Co, Cs, Hf, Zr, Th 26 (adjusted to pH 7) 1 134Cs, 85Sr, 110Ag 47 0.01 m CaCl2 1 239/240Pu 41 1 m NH4OAc pH 7 1 137Cs, 106Ru 81 2 137Cs 42, 43 2 134Cs, 241Am, 106Ru, 90Sr, 41 239Pu, 129I 2 137Cs, 90Sr 4 3 137Cs 44 0.11 m HOAc Exchangeable and 1 134Cs, 85Sr, 110Ag 47 carbonate bound 0.43 m HOAc Extractable, inorganic 2 Cu 2 0.5 m HOAc Inorganically absorbed 2 239/240Pu 41 1 m NaOAc adjusted to Carbonate 2 As, Ce, Co, Cs, Hf, Zr, Th 26 pH 5 with HOOAc 2 Pb, Ni, Cu, Zn 34, 78 2 As, Cs, Hf, Zr, Ce, Co 35 3 134Cs, 241Am, 106Ru, 41 90Sr, 239Pu, 129I 3 137Cs, 90Sr 4 0.04 m NH4OAc in 25% Fe and Mn oxides 2 137Cs, 106Ru 81 v/v HOAc 3 Pb, Ni, Cu, Zn 34 3 Pb, Ni, Cu, Zn 78 3 As, Cs, Hf, Zr, Ce, Co 35 3 As, Ce, Co, Cs, Hf, Zr, Th 26 4 134Cs, 241Am, 106Ru, 41 90Sr, 239Pu, 129I 4 137Cs, 90Sr 4 4 137Cs 44 0.1 m (NH4)2C2O4·H2O Occluded free oxides 4 Cu 2 0.1 m NaOH (In N2) Humic substances 2 134Cs, 85Sr, 110Ag 47 0.1 m Na4P2O7·H2O Organic 2 134Cs, 85Sr, 110Ag 47 3 239/240Pu 40 5 134Cs, 241Am, 106Ru, 41 90Sr, 239Pu, 129I 0.1 m K4P2O7·H2O 3 Cu 2 30% H2O2 then 0.5 m Oxidisible organic 3 137Cs 42, 43 NH4OAc and 0.04 m matter HNO3 30% H2O2 to pH 6 with 3 134Cs, 85Sr, 110Ag 47 20% HNO3 3 137Cs, 106Ru 81 4 Pb, Ni, Cu, Zn 34 4 Pb, Ni, Cu, Zn 78 4 As, Cs, Hf, Zr, Ce, Co 35 4 As, Ce, Co, Cs, Hf, Zr, Th 26 5 137Cs, 90Sr 4 7 m HNO3 Acid digestible 4 137Cs, 106Ru 81 4 137Cs 42, 43 5 137Cs 44 6 134Cs, 241Am, 106Ru, 90Sr, 41 239Pu, 129I 6 137Cs, 90Sr 4 Mixture of HNO3, HF Residual 5 Pb, Ni, Cu, Zn 34 and HClO4 5 Pb, Ni, Cu, Zn 78 5 As, Cs, Hf, Zr, Ce, Co 35 5 As, Ce, Co, Cs, Hf, Zr, Th 26 5 Cu 2 Ashing and HF digestion 5 239/240Pu 41 94R Analyst, August 1997, Vol. 122tion was also presented for 134Cs and 106Ru (Table 6). They suggested that this type of data could be used to provide information about the potential radionuclide availability for plant uptake. Livens et al.40 presented a similar data set for 239,240Pu using different soil types (Table 7).Oughton et al.4 (Tables 2 and 4) used their sequential extraction scheme to identify the soil fraction of 137Cs and 90Sr contamination that was incorporated within uranium oxide fuel particles. Extraction data for Ukrainian, Belarussian and Norwegian soils were used to compare site specific radionuclide soil mobility and transfer factors. Results indicated that radionuclide speciation and 90Sr mobility varied with distance from Chernobyl.These data were used to support their hypothesis that fuel particle weathering would lead to an increase in the bioavailability and plant uptake of 90Sr within the Chernobyl 30 km zone.3,4 Another research team47 compared two sequential extraction schemes (Tables 2 and 4); one designed to isolate organic fractions associated with 134Cs, 90Sr and 110Ag, and the other to look at inorganic associations (based on the work in reference 34) in two soils, a sandy loam and a sandy soil.Their second scheme followed protocols recommended by the Community Bureau of Reference for heavy metals in soils and sediments.48 They suggested that data from this type of experiment could be used to predict radionuclide availability for plant uptake from soils but did not verify this experimentally. Fawaris and Johanson42,43 used a sequential extraction technique (Tables 2 and 4) to estimate the percentage of 137Cs in a Swedish forest soil that was likely to be available for plant uptake with time.They looked at correlations between water extractable and NH4OAc pH 7 extractable 137Cs and organic matter content, but not for correlations between extracted 137Cs and plant 137Cs. Kennedy and Horrill44 developed a sequential extraction method for use with acidic, organic soils, such as peats, that had relatively low Chernobyl 137Cs contamination compared with soils in Scandinavia and central Europe (Tables 2 and 4).They did a series of extractions over time, using fresh sub-samples of three different organic soils, to assess the reproducibility of the method. They then used the method to extract soils from nine heather dominated ecosystems and looked for correlations between extracted 137Cs and heather plant 137Cs activity concentrations. Relationships between 137Cs activity extracted at each stage of the procedure were presented as a simple correlation between 137Cs activity concentrations of heather growing on the soil and 137Cs activity extracted at each stage of the procedure, and then as a cumulative value derived by adding 137Cs activity concentrations from each stage sequentially.Even with the high variability inherent in all soil extraction procedures they found that there was significant correlation between extractable 137Cs activity (0–10 cms) and heather 137Cs activity; r = 0.748** (r > 0.708, p = 0.01) but only poor correlation with total 137Cs soil activity (0–10 cms); r = 0.406.They suggested that if there were similar findings for other soil types and plant species it might be possible to obtain relative estimates of soil bioavailability using a double water extraction and/or an NH4OAc pH 7 extraction as an empirical extraction method. Published sequential extraction data are often presented as quantitative evidence of pollutant soil and sediment phase associations, and assume that no phase exchanges or sample solution matrix effects occur during extraction or analysis.49 Even when spike solutions have been used to validate extraction techniques data obtained are not always used to comment on the implications for data interpretation.For example, one research group recovered only 10% of a lead spike in their acetate extraction.50 Low recoveries such as this can be due to a number of reasons, e.g., interactions between the soil and spike solution during extraction, interference(s) during analysis etc.Such inherent problems mean that spike solutions may not be a useful measure of extraction efficiency. A better method is probably the use of laboratory reference soils, and participation in interlaboratory comparison studies where possible. Specific extractant efficiency varies for soils with different mineralogical and organic composition therefore each stage of sequential extraction needs to be validated for a range of soil types.49 Even then data from sequential extraction schemes can only provide a relative empirical prediction of pollutant soil phase associations, and the potentially labile fraction available for root uptake.Pollutant concentrations measured in the final stage of sequential extraction schemes are less variable than those Table 5 Correlation data for copper soil fractionation and soil components for 24 soils, as fractions of percentage total Cu, using a sequential extraction scheme2 Cu Cu Cu Cu % % Free Free AAC PYR OX RES Org C Clay pH Fe Mn Cu-CA 0.67** 0.08 0.24 20.37 20.12 20.22 20.64** 20.09 20.21 Cu-AAC 20.12 0.07 20.01 20.31 20.21 20.44* 20.11 20.19 Cu-PYR 20.32 20.37 0.53** 20.38 20.13 20.50 20.60** Cu-OX 20.32 20.28 20.26 20.07 0.44* 0.22 Cu-RES 20.16 0.42* 0.31 20.07 20.06 Key: Cu-CA = soil solution and exchangeable copper; Cu-AAC = bound by mainly inorganic sites (specific absorption); Cu-PYR = bound by organic sites (specific absorption); Cu-OX = occluded by free oxides; Cu-RES = residual; * significant at 5% level; ** significant at 1% level Table 6 Examples of relative radionuclide soil phase association data obtained from sequential extraction procedures for 134Cs (expressed as a percentage of the total)41 Clay Loam Sand Peat Soil phase 134Cs 106Ru 134Cs 106Ru 134Cs 106Ru 134Cs 106Ru Water soluble 5 7 4 4 6 6 5 2 Exchangeable < 1 2 2 1 4 < 1 36 3 Carbonate < 1 2 < 1 8 < 1 3 11 1 Oxide < 1 31 3 30 3 32 25 13 Organic < 1 49 21 51 27 38 12 75 Residual 95 84 70 60 60 21 11 5 Analyst, August 1997, Vol. 122 95Rattributed to earlier stages in the sequence.49 Therefore, pollutant concentrations from this stage can be used to estimate the soil pollutant fraction that is likely to be strongly resistant to weathering processes, and unavailable for root uptake in the short to medium term (0–10, or even 20 years in temperate ecosystems). Although sequential extraction methods were originally designed to study heavy metal and radionuclide phase associations in sediments they are now widely used to look at changes in both soil and sediment pollutant phase associations.Information from early stages in the extraction sequence is similar to that obtained from single extraction methods so it can also be used to help predict potential short-term soil pollutant bioavailability by relating extracted soil concentrations to plant concentrations. Data from later stages in the sequence could also be developed into a useful tool to predict potential longer term radionuclide and heavy metal release into the soil labile pollutant pool (due to soil weathering and decomposition processes).Other Methods Used to Estimate Soil Bioavailability (Not Extraction Methods) In radioecology, transfer factors [plant activity concentration (Bq kg21)]/[total soil activity concentration (Bq kg21)] have been widely used in an attempt to quantify the availability of soil radionuclides for plant uptake.This method gives a relatively crude estimate of potential soil radionuclide bioavailability because soil-to-plant transfer can vary considerably between different plant species, and the proportion of soil radionuclides available for root uptake also varies with season and over time. Even when transfer factors are species specific there is often a poor correlation between total soil radioactivity concentrations and plant radionuclide activity concentrations. 17,18,51 In spite of this transfer factors are currently accepted as the most practical way of describing plant uptake in many radiological assessment models.Other experimental techniques have been used to estimate the bioavailable fraction of soil radionuclide contamination, particularly for radiocaesium. The sorption mechanistic method approach developed by Cremers and co-workers7,16,52,53 measures the interception potential of two different sorption sites in soils and sediments for radiocaesium—frayed edge sites associated with illites and regular exchange sites such as organic matter and clay isomorphic substitution sites.This kinetic exchange method takes account of reversible and irreversible pools of radiocaesium in soil and gives an indication of its potential for remobilisation and/or availability for root uptake. The resin technique,54 which was used to study radiocaesium fixation in UK peat soils, also provides information about different sorption sites and release rates for radiocaesium from these soils.A laboratory tracer experiment to estimate the time dependence of 137Cs fixation to clay minerals has also been proposed.55,56 Radiocaesium activity concentrations in the electrolyte solution were assumed to be similar to those likely to occur in soil solutions. They suggested that an understanding of the time dependence of transfer between soil solution and the labile and non-labile phases in the soil must be quantified before models can accurately predict the long-term behaviour of radiocaesium in a range of soil types.A time series such as this takes account of potential changes in soil radionuclide speciation with time and could provide a useful means of assessing radiocaesium bioavailability for a wide range of soils. It is also probable that a similar approach could be adopted for a number of other potential radionuclide contaminants. Both the combined extraction/equilibrium time series and the sorption mechanistic methods take account of changes in soil radionuclide speciation with time but they have only been used for radiocaesium, for a limited range of soils.More work on reproducibility of data for a range of soil types and pollutants is needed before these methods can be widely adopted for pollutant bioavailability studies. They also have the disadvantage that they have not yet been used to look at potential nutrient and pollutant interactions. Centrifugation techniques have also been used to study heavy metals (e.g., Al, Cu, Cd) in soil solutions.57–59 They have the advantage that experimental conditions can be clearly defined and that fresh soil can be extracted in controlled experimental conditions to provide an instantaneous measure of heavy metals in the capillary fraction of soil solution at the time of sampling.Since only small volumes of solution can be obtained pollutant concentrations in some solutions may be below detection limits for analytical methods and it may be difficult to analyse the solution for a wide range of elements.Experimental Aspects of Extraction Techniques Sampling Strategy It is generally assumed that a soil sample submitted for chemical extraction is representative of the larger area from which it is taken. It is therefore necessary to devise a sampling strategy that will produce representative soil samples that can be used to extrapolate information about soil from the larger area.Whether or not a soil sample is representative of the area is dependent on topography, uniformity of soil type and on how the sample has been selected and collected. Soil sampling strategies have often been unsatisfactory.35,60–62 Several reviews have made recommendations about soil sampling protocols, e.g., for chemical analysis,63 for agricultural or forestry analysis64 and for soil surveys.65 Even when soil samples are taken from an area where the soil is classified as a particular soil type large analytical variations can be obtained for replicate samples taken from a small area of apparently similar land.12,65,66 This is a particular problem for soils taken from undisturbed natural and semi-natural ecosystems that have not been ploughed and fertilised.Thus, reported radionuclide activity concentrations and heavy metal measurements in soils from some sites may be solely a function of the original sampling technique.This problem is exacerbated if the area being surveyed includes open grazing areas, scrub, trees, stony areas and boggy patches. A high degree of spatial variability may exist over quite small areas. The characteristics of the area being studied should be taken into account when the sampling regime is being designed. It should also be clearly documented. Unfortunately many papers that present pollutant field data do not provide clear information about field sampling strategies and/or topographical features of their sampling sites that may affect deposition, mobility and accumulation of pollutant contamination.Also, the number of samples that are collected may be constrained by a lack of resources for statistically acceptable field sampling programmes that will provide a representative data set. It is important that the sample set provides the best possible data with respect to the resources Table 7 Examples of relative radionuclide soil phase association data obtained from sequential extraction procedures for 239,349Pu (expressed as a percentage of the total)40 Brown earth Brown earth Alluvial Soil phase Sand (Woodland) (Pasture) gley Exchangeable 1.0 1.7 2.2 2.2 Specific adsorption 1.5 1.3 4.7 2.5 Organic 68 52.7 58.9 63.5 Sesquioxide 18.5 31.0 17.7 18.5 Residual 11.0 13.3 16.4 7.2 96R Analyst, August 1997, Vol. 122available, field site characteristics and the objectives of the study. One of the major problems with errors introduced during sampling is that they are impossible to quantify but may have a significant effect on measured pollutant distribution patterns, particularly for wide ranging field studies or monitoring programmes, where occasional ‘grab’ samples are taken.Sampling methods can cause significant bias in reported nutrient concentrations measured in both mineral and organic soils17 and in soil radionuclide and heavy metal measurements. 66 Sampling problems are further compounded for radionuclide studies where there is thought to be the possibility of ‘hot spots’.Sample Preparation When resources for analysis are limited, composite samples made by homogenising several samples from a sampling site can give more reliable information on average soil characteristics in the area, but, at the cost of information on within-site variability. Extraction methods use a relatively small (2–100 g) sub-sample of soil. It is therefore important to homogenise the bulk sample before selecting a representative sub-sample for analysis.When soils are air dried and sieved (2 mm sieve) prior to extraction they are stabilised but some of their chemical and physical properties may be altered.17,18 Such changes may be unimportant when measuring total pollutant concentrations in soils (a small fraction of some pollutants may be volatilised), but can significantly affect the amount of radionuclide and heavy metal contamination subsequently brought into solution by extractants, particularly if soils are being used for speciation studies.67 For instance, once a soil has been broken up by sieving the number of ion exchange sites in contact with the extractant may be greater than the number of ion exchange sites in contact with soil solution in the field.Drying and sieving processes homogenise soils so that mineral and organic soil particles become more uniform in size and more evenly distributed than they are in the field. Radionuclide and heavy metal concentrations in extract solutions from soils that have been dried at more than 40 °C and milled to a fine powder may bear little relationship to the amount of pollutant that is available for plant uptake from that soil in the field.Such preparation is acceptable for analysis of total pollutant concentrations but creates an even larger surface area for ion exchange processes than that created by passing samples through a 2 mm sieve. Many papers give little information about the way soils were prepared for analysis.In the absence of this it is difficult to assess whether pollutant concentrations measured in extractants are an artifact of sample preparation or whether they are a reasonable approximation of field conditions. Sample Weight : Extractant Volume and Contact Time Most extraction methods involve shaking a known mass of soil with an extractant for a predefined period of time on an endover- end or side-to-side shaker at ambient temperature.Such procedures are essentially equilibrium processes. If the soil mass to extractant volume ratio is too low (e.g., < 1 : 5) radionuclide and heavy metal re-adsorption processes may occur during extraction and soil–extractant equilibrium may not be attained. Grimshaw17 recommended that the best soil : extractant ratio for nutrient studies is one such that doubling or halving the ratio makes no difference to the final result.He suggested that high ratios were better, recommending 1 : 25 for extractants used to measure nutrient concentrations. A number of radioecologists have opted for 1 : 10 when extracting for radionuclides.4,41 Others do not provide clear information about the ratios used or refer to Tessier et al.34 who used a ratio of 1 : 8 for their heavy metal work with aquatic sediments. It is important that the shaking time is fixed, and that the time is sufficient to allow a steady state to be established between the soil and the extractant but not so long that dissolution of other soil fractions occurs. In practice, the contact time needs to be a minimum of 1 h.17 Where only gentle or intermittent shaking is used it may be necessary to extend this up to 24 h.For instance, Maher68 found that Ca concentrations in sediments extracted by NaOAc–HOAc leachate stabilised after 5 h and iron extracted by NH2OH2–HOAc after 6 h. A problem for the analyst is to select a sub-sample that is representative of the field.It may be necessary to analyse several replicate sub-samples to achieve this. It has been suggested that the minimum representative sample mass for heavy metal analysis should contain at least 1000 sample grains, but that the optimum sample mass should be three or four times greater than the minimum.69 In practice, the optimum mass for a highly contaminated sample will be lower than the mass needed for soils with lower contamination levels. For example, a considerably larger sample will be needed for soils being used in radionuclide sequential extraction studies if activity concentrations in initial extract solutions are to be above the detection limits of analytical methods.44 Even for more highly contaminated soils only a relatively small proportion of the total soil pollutants are likely to be brought into solution by weak extractants.Selecting a representative sub-sample from a fresh soil (particularly a peat) is more difficult than from an air dried, sieved mineral soil or finely milled soil.Organic soils often contain small, medium and large fragments of recognisable plant material as well as a small proportion of mineral particles so that it is difficult to retain the original field proportions in weighed sub-samples. Drying completely alters the physical characteristics of highly organic soils such as peats.17 A larger sub-sample will also be needed because fresh soils contain moisture which may account for a higher proportion of the sample mass than organic soil particles in highly organic soils.There is little evidence in the literature that the importance of using fresh sub-samples of such soils is fully appreciated, although some recent studies have taken this into account. 44,55 Separation of Extractant From Sample Extract solutions used for analysis need to be separated from the soil. Most researchers use either centrifugation,3,34,39,70–72 or centrifugation and vacuum filtration73 or gravity filtration.17,44 Both centrifugation and filtration have inherent problems.For example, very fine particles may be left in the ‘solution’ phase if centrifugation speed is insufficient or if the pore size of the filter is too large. This can be significant if a particular extractant is able to attack the soil phase and disperse it into finer components. If the extractant solution is not strongly acidic it should be centrifuged and collected in polycarbonate containers to reduce adsorption of radionuclide and heavy metal ions onto the side of the container.Glass containers may be used if the extract solution is acidified after separation from the soil. For single extraction procedures both centrifugation and filtration have been used to separate the extractant from soil. Researchers using sequential extraction methods tend to favour centrifuging the sample to separate the extractant from the soil because there is less chance of losing soil between extractions.This can be overcome by using a gravity leaching method with the same filter paper for each of the stages of the extraction sequence.44 If a sequential extraction sequence is being used, centrifugation may affect the extraction characteristics of the soil residue used for subsequent extractions because the soil may become compressed during the process. Thus the soil residue used for Analyst, August 1997, Vol. 122 97Rsubsequent stages may not retain the same wetting characteristics as the original soil sample.This could be important for the second extraction stage, in a sequential series, if it is designed to look at slightly less labile ‘root available’ pollutant contamination in organic soils.74 It is not important for the final stages where the more tightly bound fraction, released after long-term decomposition and weathering processes, is being extracted.Quality Control and Method Validation No information about the reproducibility of extraction methods with time has been found in the literature for either radionuclide or heavy metal pollutants. Evidence from nutrient analysis suggests that reproducibility is good in a reputable laboratory when the extractants are prepared by experienced analysts following standard protocols.75 However, this may not be the case when inexperienced students or researchers are following poorly defined published protocols.Even when the same protocols are used on a sub-sample of the same soil a few months later, nutrient ions measured in solution can vary by up to ±15% for some nutrients. This error may be increased to ±100% if ion concentrations in extract solutions are at, or near the detection limit for the analytical method, or if the soil sample has not been well mixed prior to analysis. There are no certified reference samples for radionuclide and heavy metal soil extraction methods although ISE organise a quarterly inter-laboratory scheme for 0.1 m CaCl2 (B, Cd, Co, Cr, Cu), 1 m NH4NO3 (Cd, Cu, Ni, Pb, Zn) and 0.1 m NaNO3 (As, Cd, Cr, Cu, Hg, Ni, Pb, Tl, Zn).It is important that analysts should use an internal laboratory reference soil (with known concentrations of the ions of interest) with each batch of soils analysed. Results for this sample can then be used to provide information about possible bias in particular sets of extraction data, and to validate experimental data sets.In 1992 a workshop on the sequential extraction of trace metals in soils and sediments concluded that it was important to establish an internationally accepted protocol for sequential extraction procedures and to find a soil sample suitable for inter-laboratory comparison exercises with a view to establishing a certified reference material for extraction procedures.76 Interpretation of Extraction Data Data obtained from extraction methods are often difficult to interpret.Ions in extract solutions displace nutrient and pollutant ions from soil ion exchange sites. In the field ion exchange between soil and soil solution is triggered by processes such as changes in field moisture, leaching, root uptake of nutrients and micro-organism activity. Therefore ion exchange processes measured by the use of extractants can only be an approximate, but nevertheless a useful measure of bioavailability.Interpretation of data from sequential extraction schemes has additional problems such as cumulative error throughout the process, and difficulties in assigning specific soil phases to extract solutions in the series.27 Also, sequential extraction results are often presented as a percentage of a variable total. This practice can obscure consistent pollutant concentrations in the residual fraction in time series studies when the residual fraction increases proportionately because total soil pollutant concentrations have decreased over time.49 Conclusions The use of laboratory extraction techniques to predict radionuclides and heavy metal bioavailability will always be open to criticism.Methods vary between laboratories and lack a standard reference material. However, these techniques can give relative empirical information about possible soil phase associations of radionuclides and heavy metals. Extraction studies cannot predict the amount of pollutant that will be transferred from soils to plants during a specific period of time but they can provide a way of quantifying relative site, source and time dependent variations in soil-to-plant transfer of pollutants.Extractants such as 1 m NH4OAc pH 7 are also useful for studies of pollutant nutrient competition and associations (e.g., K+/Cs+ and Ca2+/Sr2+). Single extraction methods were originally developed to provide agriculturalists with a relative empirical method for assessing fertilisation requirements.They were then adapted to estimate the nutrient status of semi-natural ecosystems before being adopted for studies of radionuclide and heavy metal soil bioavailability. It is unlikely that universal extractants applicable to all soil types, for all radionuclides and heavy metals, and associated with specific soil phases can be devised although it is possible that a weak extractant, such as 1 m NH4OAc pH 7, could be used for a wide range of soils and analysed for a wide range of nutrients and pollutants.This extractant could also be buffered to the soil pH to extend the range of soils for which it can be used, but further work is needed to validate this method. Other extractants such as 0.01 m Ca Cl2, 0.1 m Na NO3 and 1 m NH4NO3 have also been found to be suitable for a range of mineral soils and associated heavy metals. Single extraction methods cannot quantify the soil pollutant fraction that will be available for root uptake over time, but they can provide a relative empirical method for evaluating potential availability of soil pollutants for plant uptake.Information about relationships between nutrient and pollutant ions and/or complexes extracted from the soil at the same time can also be obtained. Although sequential extraction methods were originally designed to study heavy metal and radionuclide phase associations in sediments they are now widely used to look at changes in both soil and sediment pollutant phase associations.Information from early stages in the extraction sequence is similar to that obtained from single extraction methods so it can be used to help predict potential short-term soil pollutant bioavailability by relating extracted soil concentrations to plant concentrations. Data from later stages in the sequence could also be developed into a useful tool to predict potential longer term radionuclide and heavy metal release into the soil labile pollutant pool (due to soil weathering and decomposition processes).Most published data for single and sequential extraction methods are site, time and vegetation specific and have not been validated for comparison studies and/or the development of generic models. The use of long-term laboratory reference soils and participation in well founded inter-laboratory comparison schemes would help validate extraction techniques and provide independent data points for data comparison.At present the only inter-laboratory scheme available is that organised by ISE for some heavy metals using 0.01 m Na NO3, 1 m NH4NO3 and 0.01 m CaCl2. Standard protocols also need to be agreed for soil sampling, sample preparation and extraction techniques to provide comparable data. There is little merit in having stringent extraction and analytical quality control procedures, validation and method protocols unless these are complemented by clear method protocols for sampling programmes and sample preparation procedures.When extraction or alternative techniques are used, soil sampling programmes should take account of potential seasonal effects so that seasonal variability can be separated from within site variability and between site variability. Ideally, sampling programmes should cover at least two years so that temporal changes can also be considered, and soil sampling protocols should be carefully defined. It is often difficult to make direct comparisons of published extraction data.Many papers do not clearly define extraction protocols used; in others protocols are clearly defined, but are sig- 98R Analyst, August 1997, Vol. 122nificantly different from one another so that results obtained cannot be readily compared.76 Even so, they can still provide useful information about potential changes in associations of radionuclide and heavy metal soil contamination with particular soil components.They can also be used to make comparative assessments of different soil types if extraction protocols are strictly defined and data are interpreted carefully.41 Thus, extraction methods can provide a relative empirical method for evaluating radionuclide and heavy metal soil contamination that may be available for plant uptake from the soil. They could also be used to provide information about potential soil pollutant bioavailability after accidental environmental contamination.However, it is important to understand the limitations of such techniques when relating pollutant concentrations measured in soil extract solutions to ecosystems. It should also be remembered that however good a method is for estimating what fraction of soil pollutants is potentially available for plant uptake, this is only one pathway for plant pollutant transfer. Others include aerial deposition, and resuspension; these may be important pathways, particularly in the immediate aftermath of accidental pollution releases to the environment.Information presented in this review is partly based on a desk study funded by the Ministry of Agriculture, Fisheries and Food, MAFF. The authors would like to thank Paul Naylor for his support as MAFF project officer and scientific colleagues at ITE Merlewood for encouragement and constructive comments. References 1 Gupta, S. K., and Aten, C., J. Environ. Anal. Chem., 1993, 51, 25. 2 McLaren, R. G., and Crawford, D.V., J. Soil Sci., 1973, 24, 172. 3 Oughton, D. H., and Salbu, B., in Nordic Radioecology: the Transfer of Radionuclides Through Nordic Ecosystems to Man, ed. Dahlgaard, H., Elsevier, Oxford, 1994, pp. 165–184. 4 Oughton, D. H., Salbu, B., Riise, G., Lien, H., Østby, G., and Nøren, A., Analyst, 1992, 117, 481. 5 Graham, E. R., and Killion, D. D., Soil Sci. Soc. Am. Proc., 1962, 26, 545. 6 Lakanen, E., and Paasikallio, A., Ann. Agric. Fenn., 1970, 9, 133. 7 Cremers, A., Elsen, A., De Preter, P., and Maes, A., Nature, 335, 1988, 247. 8 Evans, D.W., Alberts, J. J., and Clark, R. A., Geochim. Cosmochim. Acta, 1983, 47, 1041. 9 Francis, A. J., in Environmental Migration of Long-lived Radionuclides, International Atomic Energy Agency, Vienna, 1982, pp. 415–429. (IAEA-SM-257/72). 10 Haselwandter, K., Berreck, M., and Brunner, P., Trans. Brit. Mycol. Soc., 1988, 90, 171. 11 Horyna, J., and Randa, Z., J. Radioanal. Nucl. Chem. Lett., 1988, 127, 107. 12 Haugen, L. E., and Uhlen, G., in Radioaktivt nedfall fra Tsjernobylulykken, ed Garmo, T. H., and Gunnerod, T. B., Norges Landbruksvitenskapelige Forskningsrad, Oslo, 1992, pp. 43–64. 13 Heinrich, G., J. Environ. Radioact., 1993, 18, 229. 14 Chebotina, M. Y., and Kulikov, N. V., Sov. J. of Ecol., 1973, 4, 84. 15 Kulikov, N. V., Molchanova, I. V., Chebotina, M. Y., and Karavaeva, E. N., Sov. J. Ecol., 1977, 8, 162. 16 Valcke, E., and Cremers, A., Sci. Total Environ., 1994, 157, 275. 17 Grimshaw, H. M., in Chemical Analysis of Ecological Materials, ed. Allen, S. E., Blackwell Scientific Publications, Oxford, 1989, pp. 7–45. 18 Hesse, P. R., A Textbook of Soil Chemical Analysis, John Murray, London, 1971. 19 Page, A. L., Miller, R. H., and Kenney, D. R, Methods of Soil Analysis, Chemical and Micro-biological Properties, American Society of Agronomy, Madison, 2nd edn., 1982. 20 Ministry of Agriculture Fisheries and Food, The Analysis of Agricultural Materials: A Manual of Analytical Methods used by the Agricultural and Development and Advisory Service, (Reference book 427), Her Majesty’s Stationery Office, London, 3rd edn., 1986. 21 Legret, M., Int.J. Environ. Anal. Chem., 1993, 51, 161. 22 F�orstner, U., Int. J. Environ. Anal. Chem., 1993, 51, 5. 23 Wilkins, B. T., Green, N., Stewart, S. P., Major, R. O., and Dodd, N. J., in Seminar on the Behaviour of Radionuclides in Estuaries. Renesse 17-21 September 1984, Council for the European Community, Luxembourg, 1985, 55–70. 24 Nirel, P., Thomas, A. J., and Martin, J. M., in Speciation of Fission Activation Products, ed. Bulman, R. A., and Cooper, J. R., Elsevier Applied Science, London, 1986, 19–26. 25 Beckett, P. H. T., Adv. Soil Sci., 1989, 9, 143. 26 Rigol, A., Vidal, M., and Rauret, G., J. Radioanal. Nucl. Chem., 1996, 208, 617. 27 Ure, A. M., Mikrochim. Acta, 1991, II, 49. 28 Adriano, D. C., Delaney, M. S., Hoyt, G. D., and Paine, D., Environ. Exp. Bot., 1977, 17, 69. 29 Wilson, D. O., and Cline, J. F., Nature, 1966, 209, 941. 30 Thomas, G. W., in Methods of Soil Analysis Part II Chemical and Microbial Properties, ed. Page, A. L., Miller, R. H., and Keeney, D. R., Madison, Wisconsin, 1982, pp. 159–165. 31 Oughton, D. H., Skipperud, L., and Tronstad, E., Cs and Sr Mobility in Nordic Soils, Nordic Reactor Safety Commission, Risø, 1995. 32 Oughton D. H., Børretzen, P., Salbu, B., and Tronstad, E., Sci. Total Environ., in the press. 33 Evans, E. J., and Dekker, A. J., Can. J. Soil Sci., 1969, 49, 349. 34 Tessier, A., Campbell, P. G. C., and Bisson, M., Anal. Chem., 1979, 51, 844. 35 Martin, J. M., Nirel, P., and Thomas, A. J Marine Chem., 1987, 22, 313. 36 Paasikallio, A., Ann. Agric. Fenn., 1984, 23, 109. 37 Novozamsky, I., Lexmond, T. M., and Houba, V. J. G., Int. J. Environ. Anal. Chem., 1993, 51, 47. 38 Houba, V. J. G., Lexmond, Th. M., Novozamsky, I., and Van der Lee, J. J., Sci. Total Environ., 1996, 178, 21. 39 Houba, V. J. G., Novozamsky, I., and Temminghoff, E., Soil and Plant Analysis 5A: Soil Analysis Procedures Extraction With 0.01 M CaCl2, Wageningen Agricultural University, Wageningen. 40 Livens, F. R., Baxter, M. S., and Allen, S. E., in Speciation of Fission and Activation Products in the Environment, ed. Bulman, R. A., and Cooper, J. R., Elsevier, London, 1986, pp. 143–150. 41 Wilkins, B. T., Green, N., Stewart, S. P., and Major, R. O., in Speciation of Fission and Activation Products in the Environment, ed. Bulman, R. A., and Cooper, J. R., Elsevier, New York, 1986, pp. 101–113. 42 Fawaris, B. H., and Johanson, K. J., in Behaviour of 137Cs in the Boreal Forest Ecosystem of Central Sweden, PhD Thesis by B. Fawaris IV, Swedish University of Agricultural Science, Uppsala, 1995. 43 Fawaris, B. H., and Johanson, K. J., Sci. Total Environ, 1995, 170, 221. 44 Kennedy, V. H., and Horrill, A. D., in Radiation Doses and Pathways to Man From Semi-natural Ecosystems, ed. McGarry, A., (Commission of the European Communities contract F13-CT920058; sector A25), Radiological Protection Institute of Ireland, Dublin, 1995, pp. 16–30. 45 Chang, A. C., Page, A. L., Warneke, J. E., and Grgurevic, E., J. Environ. Qual., 1984, 13, 33. 46 Emmerich, W. E., Lund, L. J., Page, A. L., and Chang, A. C., J. Environ. Qual., 1982, 11, 178. 47 Vidal, M., Tent, J., Llaurad�o, M., and Rauret, G., J. Radioecol., 1993, 1, 49. 48 Ure, A. M., Quevauviller, P., Muntau, H., and Griepink, B., Int. J. Environ. Anal. Chem., 1993, 51, 135. 49 Nirel, P. M. V., and Morel, F. M. M., Water Res., 1990, 24, 1055. 50 Belzile, N., Lecomte, P., and Tessier, A., Environ. Sci. Technol., 1989, 23, 1015. 51 Nisbet, A. F., and Lembrechts, J. F., in Transfer of Radionuclides in Natural and Semi-natural Environments, ed.Desmet, G., Nassimbeni, P., and Belli, M., Elsevier Applied Science, London, New York, 1990, pp. 371–381. 52 Sweeck, L., Wauters, J., Valke, E., and Cremers, A., in Transfer of Radionuclides In Natural and Semi-natural Environments, ed. Desmet, G., Nassimbeni, P., and Belli, M., Elsevier Applied Science, London, New York, 1990, pp. 249–258. Analyst, August 1997, Vol. 122 99R53 Wauters, J., Madruga, M. J., Vidal, M., and Cremers, A., Sci. Total Environ., 1996, 187, 121. 54 Livens, F. L., Howe, M. T., Hemingway, J. D., Goulding, K. W. T., and Howard, B. J., Eur. J. Soil Sci. 1996, 47, 105. 55 Absalom, J. P., Young, S. D., and Crout, N. M. J., Eur. J. Soil Sci., 1995, 46, 461. 56 Absalom, J. P., Young, S. D., and Crout, N. M. J., Environ. Sci. Technol., 1996, 30(9), 2735. 57 Gillman, G. P., A Centrifuge Method for Obtaining Soil Solution, (Division of Soils Divisional Report 6), Commonwealth Scientific and Industrial Research Organisation: Townsville 1976. 58 Reynolds, B., Plant Soil, 1984, 78, 434. 59 Keller, C., Commun. Soil Sci. Plant Anal., 1995, 26, 1621. 60 Webster, R., and Oliver, M. A., Statistical Methods in Soil and Land Resource Survey, Oxford University Press, Oxford, 1st edn., 1990, 315 pp. 61 Crepin, J., and Johnson, R. L., Soil Sampling for Environmental Assessment, in Soil Sampling and Methods of Analysis, ed. Carter, M. R., Lewis Publishers, London, 1993. 62 Rubio, R., Int. J. Environ. Anal. Chem., 1993, 51, 205. 63 Kratochvil, B., Wallace, D., and Taylor, J. K., Anal. Chem., 1984, 56, 113R. 64 Petersen, R. G., and Calvin, L. D., in Methods of Soil Analysis: Part 1 Physical and Mineralogical Methods, ed. Klute, A., American Society of Agronomy, Madison, 2nd edn., 1986, pp. 33–52. 65 Rubio, R., and Ure, A. M., Int. J. Environ. Anal. Chem., 1993, 51, 205. 66 McGee, E. J., Keatinge, M. J., Synnott, H. J., and Colgan, P. A., Health Phys., 1995, 68, 320. 67 Kennedy, V. H., and Sanchez, A. L., in preparation. 68 Maher, W. A., Bull. Environ. Contam. Toxicol., 1984, 32, 339. 69 Jackson, M. L., Soil Chemical Analysis, Prentice Hall, London, 1958. 70 Riise, G., Bjørnstad, H. E., Lien, H. N., Oughton, D. H., and Salbu, B., J. Radioanal. Nucl. Chem., 1990, 142, 531. 71 Orsini, L., and Bermond, A., Int. J. Environ. Anal. Chem., 1993, 51, 97. 72 Riise, G., Salbu, B., Singh, B. R, and Steinnes, E., Water, Air, Soil Poll., 1994, 78, 286. 73 Gupta, S. K., and Chen, K. Y., Environ. Lett., 1975, 10(2), 129. 74 Harrison, A. F., personal communication. 75 Kennedy, V. H., Rowland, A. P., and Parrington, J., Commun. Soil Sci. Plant Anal., 1994, 25, 1605. 76 Quevauviller, P., Rauret, G., and Griepink, B., Int. J. Environ. Anal. Chem., 1993, 51, 231. 77 Pay�a-P�erez, A., Sala, J., and Mousty, F., Int. J. Environ. Anal. Chem., 1993, 51, 223. 78 Kheboian, C., and Bauer, C. F., Anal. Chem., 1987, 59, 1417. 79 Clayton, P. M., and Tiller, K. G., A chemical method for the determination of the heavy metal content of soils in environmental studies, (Division of Soils Technical Papers, 41). CSIRO, Melbourne, 1979, 17 pp. 80 Haq, A.U., and Miller, M. H., Agron. J., 1972, 64, 779. 81 Andersson, K. G., and Roed, J., J. Environ. Radioact., 1994, 22, 183. Paper 7/04133K Received June 13, 1997 Accepted June 27, 1997 100R Analyst, August 1997, Vo
ISSN:0003-2654
DOI:10.1039/a704133k
出版商:RSC
年代:1997
数据来源: RSC
|
2. |
Liquid Chromatography With Ultraviolet Absorbance Detection ofEthylenethiourea in Blood Serum After Microwave Irradiation as anAuxiliary Cleanup Step |
|
Analyst,
Volume 122,
Issue 8,
1997,
Page 733-735
Paulo Cícero do Nascimento,
Preview
|
|
摘要:
Liquid Chromatography With Ultraviolet Absorbance Detection of Ethylenethiourea in Blood Serum After Microwave Irradiation as an Auxiliary Cleanup Step Paulo C�ýcero do Nascimento*, Denise Bohrer, Solange Garcia and Aneti Fernanda Ritzel Department of Chemistry, Federal University of Santa Maria, Santa Maria, RS 97110-000, Brazil, E-mail: npaulo@quimica.ufsm.br The report describes a method of deproteinizing serum that combines the action of acids with microwave irradiation. The acid concentration is ten times smaller than in the usual acid deproteinization method and the results are similar.The main advantages of the proposed method are the maintenance of the pH of the supernatant at around 5, and the reduction of the concentration of the deproteinizing agent. The procedure was applied to the determination of ethylenethiourea in serum samples by HPLC with spectrophotometric detection. Using 0.5 ml of serum, 0.04 mg of ethylenethiourea were determined with recoveries between 89 and 109%.Keywords: Ethylenethiourea; blood analysis; microwave irradiation; protein precipitation Ethylenethiourea (ETU) (imidazolidine-2-thione) is the most important degradation product of ethylenebisdithiocarbamate (EBDC) fungicides1 that has been shown to be teratogenic and carcinogenic,2 and it is capable of producing genotoxic effects.3 Although it has been shown that levels of ETU on raw agricultural products treated with EBDCs are quite low and ETU is readly biodegradable under field conditions, concern has been expressed that food processing, particulary heat treatment, could increase the conversion of EBDC residues so that the level of ETU in processed products can be higher than on the raw material.4 ETU may also be produced in the smoke from tobacco containing high EBDC residues.5 Determination of ETU in biological fluids such as blood6 and urine7 is important since it is an ubiquitous impurity of the EBDC fungicides and the ETU levels can be an indicator of exposure to them.GLC and HPLC methods are the most useful for determining ETU residues in many samples, and HPLC has been reported to have advantages over the more widely used GLC techniques in the determination of traces of ETU, because derivatization procedures can be omitted.8210 However, cleanup procedures in blood analysis cannot be avoided, due to the complexity of this matrix. Kobayashi et al.6 have used HPLC in the determination of ETU in rat plasma after an extraction procedure that includes a protein precipitation with ethanol followed by a timeconsuming partition and column separation.For the HPLC analysis of blood samples, the removal of proteins is the most important cleanup step because they can precipitate when in contact with the mobile phase and thereby block tubing, causing increases in back pressure or deterioration of column performance.11 Methods for deproteinization of plasma or serum include precipitation with acids, organic solvents and inorganic salts.The best results are obtained when strong acids like trichloroacetic or perchloric are used in comparison with organic solvents or inorganic salts.12 In spite of their relatively low efficacy, organic solvents (such as methanol, acetonitrile, acetone and ethanol) have been popular as precipitants in HPLC analysis mainly because of their widespread use as mobile phase components, and because when strong acids are used the analyte must be stable at the low pH values encountered in the supernatant.However, they may dilute the sample and affect analyte solubility. In the present report, we describe a method of deproteinizing serum that combines the action of acids (in lower concentration than the ones used in the conventional method) with the use of energy as a heat source, and its application in the determination of trace amounts of ETU in serum samples. Microwave (MW) irradiation has been shown to be a very efficient technique for heating during chemical reactions.It has been successfully applied to both sample wet ashing,13 and organic synthesis14 as a reaction medium heating system, but the use of irradiation for improving the results of chemical reactions is not so widespread.15 The advantage of coupling an acidic reagent with heating is that a considerably smaller acid concentrations are required for the deproteinization. As a result, the supernatant remains near physiological pH values, important for the stability of ETU.16,17 Experimental Apparatus A domestic Philco (S�ao Paulo, Brazil) MW oven operating with a variable power between 175 and 800 W, a Lambda 16 UV/VIS spectrometer (Perkin Elmer, � Uberlingen, Germany) and a Jouan C4-12 centrifuge (Saint-Herblain, France) were used.The chromatographic equipment consisted of a Dionex DX- 300 gradient chromatography system (Sunnyvale, United States), equipped with a variable-wavelength SPD-10AV UV/ VIS detector (Shimadzu, Kyoto, Japan), fitted with a 10 ul flow cell and set at 233 nm, and a C-R6A data processor (Shimadzu).Serum Samples All deproteinization procedures were performed with a single lot of pooled human serum collected in the Santa Maria University Hospital. Reagents For protein precipitation. Trichloroacetic acid (TCA) 10% (m/v) and 1% (m/v); perchloric acid 6% (m/v) and 0.5% (m/v); sulfuric acid 10% (v/v) and 1% (v/v); nitric acid 10% (v/v) and 1% (v/v) were used.For chromatographic analysis. Methanol and acetonitrile of HPLC grade and ETU were used. All reagents were of analytical-reagent grade from Merck (Darmstadt, Germany). The water used was distilled, deionized and further purified by a Milli-Q high purity water device (Millipore, Bedford, USA). Analyst, August 1997, Vol. 122 (733–735) 733Calibration of Microwave Equipment The available power in the cavity was determined by measuring the absolute change in the temperature of 1 kg of distilled water, in a polyethylene beaker, for 2 min, at each power setting (in triplicate), and the apparent power absorbed was determined by the relationship given by Kingston.13 Sample Cleanup In the proposed procedure, a serum sample (0.5 ml or 1.0 ml) was added to a 10 3 75 mm centrifuge tube followed by the deproteinizing agent (see Table 1).The assisted deproteinization was performed in a closed vessel (centrifuge tube with cover) that was always kept at the same position in the oven.The samples were individually submitted to the irradiation, and a beaker with 100 ml of water was also set in the oven to avoid a sudden heating of the sample. The program described in Table 2 was used for the heating process and after irradiation, the closed vessel was cooled in an ice-water bath to reduce the internal pressure prior to opening. The contents were then centrifuged at 2400g for 15 min, the supernatant collected and a 50 ml aliquot was assayed for protein content by the Coomassie Blue procedure.18 The conventional procedure was carried out with the acid concentrations and volumes described in Table 1, using the same quantities of serum as described above.After deproteinization, the samples were centrifuged at 2400g for 15 min. All samples were prepared in triplicate and after deproteinizing, the pH of the combined triplicate supernatants was measured.To evaluate the smaller concentration of the acids that, in combination with the irradiation, was efficient, concentrations between 0.1 and 5% (m/v) of each acid were used and the protein content of the supernatant was measured. Chromatographic Separation All separations were performed on a Lichrospher C-18 (5 mm, 250 3 4.6 mm) column (Merck, Darmstadt, Germany), with a 20 ml loop. Water was used as mobile phase with a flow rate of 1 ml min21. After each chromatographic separation the column was washed with acetonitrile by means of a gradient program.ETU Determination in Serum The serum samples were spiked with 2.0, 5.0 and 10 mg ETU (per ml of sample), and after deproteinizing (by both methods), the supernatants were filtered through a 0.45 mm membrane prefilter (Sartorius, G�ottingen, Germany) and injected onto the column. The recoveries of ETU were calculated from calibration graphs that were constructed from the concentration and peak area of the chromatogained with ETU standards.The effect of volume changes due to the displacement of the proteins from solution was calculated according to Fielding.19 All fortified samples were prepared in triplicate. Blank analyses were performed in order to check interferences from the sample and from deproteinizing agents. Results and Discussion The total protein content of the pooled serum sample was 69 000 g l21 (Biuret assay). The efficacy of the investigated acids in removing the proteins from serum samples is shown in Table 3.The acid concentrations listed in Table 3 are the smallest concentrations that promoted a deproteinization higher than 99.9% of the original concentration. The results indicated that the combination of the less concentrated acids with irradiation showed about the same efficacy as the more concentrated ones, without irradiation, but the pH of the supernatant stayed 3–4 units higher than in the conventional procedure.The time used for both methods was the same, around 20 min per sample. Although nitric acid itself is not normally used for deproteinizing serum samples, it showed a good performance in combination with irradiation. While the supernatant appeared cloudy after the precipitation with sulfuric acid by the usual method, it was clear when irradiation was used. The absorption of MW energy by dipolar molecular resonance may lead to intense local heating, improving reactions yields.Table 1 Conditions for protein precipitation: concentration of the precipitant agent and volume per ml of serum for the two procedures, with and without MW irradiation Without MW MW assisted Precipitant Precipitant concentration (%)* Volume of agent/ml (per ml of serum) Precipitant concentration (%)* Volume of agent/ml (per ml of serum) TCA 10 0.4 1 1.0 HClO4 6 0.8 0.5 1.0 HNO3 10 1.0 1 0.5 H2SO4 10 1.0 0.5 0.5 * TCA (m/v); HClO4, HNO3, H2SO4 (v/v). Table 2 Microwave program for protein precipitation Temperature/°C Step Time/s Power/W* Initial Final 1† 50 174 24 56 2‡ 45 244–767 24 56 * Apparent power absorbed determined according Kingston13 † After this step the vessel was cooled in an ice-water bath.‡ Five s at 244 W; 296 W; 349 W; 389 W; 453 W; 523 W; 558 W; 610 W; 767 W. Table 3 Efficacy of protein precipitation, expressed as remaining proteins in supernatant, and pH of supernatant after deproteinization Without MW MW assisted Protein Protein Precipitant remaining/ mg l21 Supernatant pH remaining/ mg 121 Supernatant pH TCA 8.7 ± 3.1 1.89 11.0 ± 6.6 4.92 HClO4 19.0 ± 2.7 1.43 51.3 ± 2.5 4.82 HNO3 7.3 ± 1.5 0.86 32.7 ± 5.9 4.51 H2SO4 * — — 56.7 ± 3.1 5.21 * Shows no effective deproteinizing action up to 10% concentration. 734 Analyst, August 1997, Vol. 122In the deproteinization of serum samples that combine acid action with heating in a water bath, the temperature is maintained at 70 °C and the acid concentration remains as high as in conventional methods.20 In the present method, the temperature reached 56 °C at the end of irradiation steps 1 and 2 (see Table 2).The use of one step irradiation produced a nonreproducible protein precipitation. Chromatographic Determination of ETU ETU has a sharp absorbance at 233 nm and this was adopted as the wavelength setting on the UV monitoring system. ETU was readily eluted from a 25 cm 3 4.6 mm id C18-bonded silica column with water–methanol solutions (90 + 10, 95 + 5, 99 + 1 v/v) as mobile phase.A bonded stationary phase exhibiting apolar characteristics appeared to be suitable for this determination. However, for the determination of ETU in samples, of which the matrix was not totally eliminated by cleanup procedures, rapid elution seems to be disadvantageous once the analyte signal can be overlapped by the baseline disturbance due to the injection. When the mobile phase is pure water, the retention time of ETU is higher than with water–methanol mixtures, so that pure water was used as mobile phase throughout this work.With water as mobile phase the retention time of ETU was 6.0 min. The calibration graph was rectilinear for 0.005–0.30 mg of ETU (r = 0.999), and the RSD (n = 3) was 0.48%. Recoveries of ETU from fortified serum samples, by means of acid deproteinization, with and without irradiation are showed in Table 4. With the exception of TCA, which signal overlaps the one of ETU, both procedures are adequate for ETU determination. However, best recoveries were obtained from deproteinization with more dilute acids and irradiation (Table 4).Typical chromatograms obtained after acid deproteinization (HNO3), with and without irradiation, are shown in Fig. 1. Because of a small decrease in the retention time of ETU after successive measurements of serum samples, a gradient program was carried out to wash the column between the injections. With this procedure, one chromatographic determination takes 12 min and up to 130 determinations were made with the same column. The best results were obtained with nitric and sulfuric acids.TCA showed interferences and HClO4, for higher fortification levels, presented poor recoveries. Conclusions The proposed method is simple and offers advantages over the conventional protein-precipitation procedure, due to the possibility of using a ten-fold less acid concentration, compared with the usual concentration, and the pH of the supernatant is maintained near to physiological values.The use of HPLC means that the time-consuming derivatization procedures for the ETU determination can be omitted. When procedures with many steps are necessary for an effective cleanup, they can introduce losses due to sample handling. In the case of serum, the injection of the sample onto the HPLC system without a cleanup step is not possible. The use of acidic protein precipitants, with irradiation, to assist sample cleanup prior to HPLC injection, is an effective procedure for ETU trace residue analysis, because little sample preparation and dilution are required, and the pH remains at a value that guarantees the stability of the analyte.Acknowledgements to CNPq for the scholarships and to GTZ/ RFG and CNPq for the financial support. References 1 Newsome, W. H., in Analytical Methods for Pesticides and Plant Growth Regulators, Academic Press, New York, 1980, Vol.XI, p. 197. 2 Fishbein, L., J. Toxicol. Environ. Health, 1976, 1, 713. 3 Dearfield, K. L., Mutat. Res., 1994, 317, 111. 4 Bottomley, P., Hoodless, R. A., and Smart, N. A., in Residue Reviews ed. Gunther, F. A., Springer Verlag, New York, 1985, p. 73. 5 Lentza-Rios, C., Rev. Environ. Contam. Toxicol., 1990, 115, 1. 6 Kobayashi, H., Matano, O., and Goto, S., J. Chromatogr., 1981, 207, 281. 7 Kurttio, P., and Savolainen, K., Scand. J. Work, Environ. Health , 1990, 16, 203. 8 Farrington, D. S., and Hopkins, R. G., Analyst, 1979, 104, 111. 9 Onley, J. H., J. Assoc. Off. Anal. Chem., 1977, 60, 1111. 10 Lehotay, J., Brandsteterov�a, E., and Oktavec, D., J. Liq. Chromatogr., 1992, 15, 525. 11 McDowall, R. D., J. Chromatogr., 1989, 492, 3. 12 Blanchard, J., J. Chromatogr., 1981, 226, 455. 13 Kingston, H. M., and Jassie, L. B., Anal. Chem., 1986, 58, 2534. 14 Gedye, R., Smith, F., Westaway, K., Ali, H., Baldisera, L., Laberge, L., and Roussel, J., Tetrahedron Lett., 1986, 27, 279. 15 Ganzler, K., Salg�o, A., and Valk�o, K., J. Chromatogr., 1986, 371, 299. 16 Cruickshank, P. A., and Jarrow, H. C., J. Agr. Food Chem., 1973, 21, 33. 17 Marshall, W., J. Agric. Food Chem., 1977, 25, 357. 18 Bradford, M. M., Anal. Biochem., 1976, 72, 248. 19 Fielding, J., and Ryall, R. G., Clin. Chim. Acta, 1971, 33, 235. 20 Mathieu, C., and Mathieu, H., Clin. Chem., 1995, 41, 625. Paper 7/01011G Received February 12, 1997 Accepted April 29, 1997 Table 4 Recoveries of ETU from fortified serum samples, with and without MW irradiation as adjuvant for acid deproteinization Recovery (%)* Acid ETU added/mg 121 MW assisted without MW HNO3 2.0 98.15 ± 0.84 — 5.0 88.99 ± 10.39 78.55 ± 4.32 10.0 93.92 ± 2.83 — H2SO4 2.0 106.45 ± 2.76 — 5.0 99.4 ± 2.12 — 10.0 89.13 ±— HClO4 3.0 109.16 ± 3.08 — 7.5 94.29 ± 2.53 77.27 ± 3.89 15.0 60.48 ± 5.68 — * Recovery ± RSD (n = 3).Fig. 1 Chromatograms of serum samples fortified with ETU, after deproteinization. (a) Without MW irradiation; (b) with MW irradiation.Analyst, August 1997, Vol. 122 735 Liquid Chromatography With Ultraviolet Absorbance Detection of Ethylenethiourea in Blood Serum After Microwave Irradiation as an Auxiliary Cleanup Step Paulo C�ýcero do Nascimento*, Denise Bohrer, Solange Garcia and Aneti Fernanda Ritzel Department of Chemistry, Federal University of Santa Maria, Santa Maria, RS 97110-000, Brazil, E-mail: npaulo@quimica.ufsm.br The report describes a method of deproteinizing serum that combines the action of acids with microwave irradiation.The acid concentration is ten times smaller than in the usual acid deproteinization method and the results are similar. The main advantages of the proposed method are the maintenance of the pH of the supernatant at around 5, and the reduction of the concentration of the deproteinizing agent. The procedure was applied to the determination of ethylenethiourea in serum samples by HPLC with spectrophotometric detection. Using 0.5 ml of serum, 0.04 mg of ethylenethiourea were determined with recoveries between 89 and 109%.Keywords: Ethylenethiourea; blood analysis; microwave irradiation; protein precipitation Ethylenethiourea (ETU) (imidazolidine-2-thione) is the most important degradation product of ethylenebisdithiocarbamate (EBDC) fungicides1 that has been shown to be teratogenic and carcinogenic,2 and it is capable of producing genotoxic effects.3 Although it has been shown that levels of ETU on raw agricultural products treated with EBDCs are quite low and ETU is readly biodegradable under field conditions, concern has been expressed that food processing, particulary heat treatment, could increase the conversion of EBDC residues so that the level of ETU in processed products can be higher than on the raw material.4 ETU may also be produced in the smoke from tobacco containing high EBDC residues.5 Determination of ETU in biological fluids such as blood6 and urine7 is important since it is an ubiquitous impurity of the EBDC fungicides and the ETU levels can be an indicator of exposure to them.GLC and HPLC methods are the most useful for determining ETU residues in many samples, and HPLC has been reported to have advantages over the more widely used GLC techniques in the determination of traces of ETU, because derivatization procedures can be omitted.8210 However, cleanup procedures in blood analysis cannot be avoided, due to the complexity of this matrix.Kobayashi et al.6 have used HPLC in the determination of ETU in rat plasma after an extraction procedure that includes a protein precipitation with ethanol followed by a timeconsuming partition and column separation. For the HPLC analysis of blood samples, the removal of proteins is the most important cleanup step because they can precipitate when in contact with the mobile phase and thereby block tubing, causing increases in back pressure or deterioration of column performance.11 Methods for deproteinization of plasma or serum include precipitation with acids, organic solvents and inorganic salts.The best results are obtained when strong acids like trichloroacetic or perchloric are used in comparison with organic solvents or inorganic salts.12 In spite of their relatively low efficacy, organic solvents (such as methanol, acetonitrile, acetone and ethanol) have been popular as precipitants in HPLC analysis mainly because of their widespread use as mobile phase components, and because when strong acids are used the analyte must be stable at the low pH values encountered in the supernatant. However, they may dilute the sample and affect analyte solubility.In the present report, we describe a method of deproteinizing serum that combines the action of acids (in lower concentration than the ones used in the conventional method) with the use of energy as a heat source, and its application in the determination of trace amounts of ETU in serum samples.Microwave (MW) irradiation has been shown to be a very efficient technique for heating during chemical reactions. It has been successfully applied to both sample wet ashing,13 and organic synthesis14 as a reaction medium heating system, but the use of irradiation for improving the results of chemical reactions is not so widespread.15 The advantage of coupling an acidic reagent with heating is that a considerably smaller acid concentrations are required for the deproteinization.As a result, the supernatant remains near physiological pH values, important for the stability of ETU.16,17 Experimental Apparatus A domestic Philco (S�ao Paulo, Brazil) MW oven operating with a variable power between 175 and 800 W, a Lambda 16 UV/VIS spectrometer (Perkin Elmer, � Uberlingen, Germany) and a Jouan C4-12 centrifuge (Saint-Herblain, France) were used.The chromatographic equipment consisted of a Dionex DX- 300 gradient chromatography system (Sunnyvale, United States), equipped with a variable-wavelength SPD-10AV UV/ VIS detector (Shimadzu, Kyoto, Japan), fitted with a 10 ul flow cell and set at 233 nm, and a C-R6A data processor (Shimadzu). Serum Samples All deproteinization procedures were performed with a single lot of pooled human serum collected in the Santa Maria University Hospital. Reagents For protein precipitation.Trichloroacetic acid (TCA) 10% (m/v) and 1% (m/v); perchloric acid 6% (m/v) and 0.5% (m/v); sulfuric acid 10% (v/v) and 1% (v/v); nitric acid 10% (v/v) and 1% (v/v) were used. For chromatographic analysis. Methanol and acetonitrile of HPLC grade and ETU were used. All reagents were of analytical-reagent grade from Merck (Darmstadt, Germany). The water used was distilled, deionized and further purified by a Milli-Q high purity water device (Millipore, Bedford, USA). Analyst, August 1997, Vol. 122 (733–735) 733Calibration of Microwave Equipment The available power in the cavity was determined by measuring the absolute change in the temperature of 1 kg of distilled water, in a polyethylene beaker, for 2 min, at each power setting (in triplicate), and the apparent power absorbed was determined by the relationship given by Kingston.13 Sample Cleanup In the proposed procedure, a serum sample (0.5 ml or 1.0 ml) was added to a 10 3 75 mm centrifuge tube followed by the deproteinizing agent (see Table 1).The assisted deproteinization was performed in a closed vessel (centrifuge tube with cover) that was always kept at the same position in the oven. The samples were individually submitted to the irradiation, and a beaker with 100 ml of water was also set in the oven to avoid a sudden heating of the sample. The program described in Table 2 was used for the heating process and after irradiation, the closed vessel was cooled in an ice-water bath to reduce the internal pressure prior to opening.The contents were then centrifuged at 2400g for 15 min, the supernatant collected and a 50 ml aliquot was assayed for protein content by the Coomassie Blue procedure.18 The conventional procedure was carried out with the acid concentrations and volumes described in Table 1, using the same quantities of serum as described above. After deproteinization, the samples were centrifuged at 2400g for 15 min. All samples were prepared in triplicate and after deproteinizing, the pH of the combined triplicate supernatants was measured.To evaluate the smaller concentration of the acids that, in combination with the irradiation, was efficient, concentrations between 0.1 and 5% (m/v) of each acid were used and the protein content of the supernatant was measured. Chromatographic Separation All separations were performed on a Lichrospher C-18 (5 mm, 250 3 4.6 mm) column (Merck, Darmstadt, Germany), with a 20 ml loop.Water was used as mobile phase with a flow rate of 1 ml min21. After each chromatographic separation the column was washed with acetonitrile by means of a gradient program. ETU Determination in Serum The serum samples were spiked with 2.0, 5.0 and 10 mg ETU (per mample), and after deproteinizing (by both methods), the supernatants were filtered through a 0.45 mm membrane prefilter (Sartorius, G�ottingen, Germany) and injected onto the column.The recoveries of ETU were calculated from calibration graphs that were constructed from the concentration and peak area of the chromatograms obtained with ETU standards. The effect of volume changes due to the displacement of the proteins from solution was calculated according to Fielding.19 All fortified samples were prepared in triplicate. Blank analyses were performed in order to check interferences from the sample and from deproteinizing agents. Results and Discussion The total protein content of the pooled serum sample was 69 000 g l21 (Biuret assay).The efficacy of the investigated acids in removing the proteins from serum samples is shown in Table 3. The acid concentrations listed in Table 3 are the smallest concentrations that promoted a deproteinization higher than 99.9% of the original concentration. The results indicated that the combination of the less concentrated acids with irradiation showed about the same efficacy as the more concentrated ones, without irradiation, but the pH of the supernatant stayed 3–4 units higher than in the conventional procedure.The time used for both methods was the same, around 20 min per sample. Although nitric acid itself is not normally used for deproteinizing serum samples, it showed a good performance in combination with irradiation. While the supernatant appeared cloudy after the precipitation with sulfuric acid by the usual method, it was clear when irradiation was used.The absorption of MW energy by dipolar molecular resonance may lead to intense local heating, improving reactions yields. Table 1 Conditions for protein precipitation: concentration of the precipitant agent and volume per ml of serum for the two procedures, with and without MW irradiation Without MW MW assisted Precipitant Precipitant concentration (%)* Volume of agent/ml (per ml of serum) Precipitant concentration (%)* Volume of agent/ml (per ml of serum) TCA 10 0.4 1 1.0 HClO4 6 0.8 0.5 1.0 HNO3 10 1.0 1 0.5 H2SO4 10 1.0 0.5 0.5 * TCA (m/v); HClO4, HNO3, H2SO4 (v/v).Table 2 Microwave program for protein precipitation Temperature/°C Step Time/s Power/W* Initial Final 1† 50 174 24 56 2‡ 45 244–767 24 56 * Apparent power absorbed determined according Kingston13 † After this step the vessel was cooled in an ice-water bath. ‡ Five s at 244 W; 296 W; 349 W; 389 W; 453 W; 523 W; 558 W; 610 W; 767 W. Table 3 Efficacy of protein precipitation, expressed as remaining proteins in supernatant, and pH of supernatant after deproteinization Without MW MW assisted Protein Protein Precipitant remaining/ mg l21 Supernatant pH remaining/ mg 121 Supernatant pH TCA 8.7 ± 3.1 1.89 11.0 ± 6.6 4.92 HClO4 19.0 ± 2.7 1.43 51.3 ± 2.5 4.82 HNO3 7.3 ± 1.5 0.86 32.7 ± 5.9 4.51 H2SO4 * — — 56.7 ± 3.1 5.21 * Shows no effective deproteinizing action up to 10% concentration. 734 Analyst, August 1997, Vol. 122In the deproteinization of serum samples that combine acid action with heating in a water bath, the temperature is maintained at 70 °C and the acid concentration remains as high as in conventional methods.20 In the present method, the temperature reached 56 °C at the end of irradiation steps 1 and 2 (see Table 2).The use of one step irradiation produced a nonreproducible protein precipitation. Chromatographic Determination of ETU ETU has a sharp absorbance at 233 nm and this was adopted as the wavelength setting on the UV monitoring system.ETU was readily eluted from a 25 cm 3 4.6 mm id C18-bonded silica column with water–methanol solutions (90 + 10, 95 + 5, 99 + 1 v/v) as mobile phase. A bonded stationary phase exhibiting apolar characteristics appeared to be suitable for this determination. However, for the determination of ETU in samples, of which the matrix was not totally eliminated by cleanup procedures, rapid elution seems to be disadvantageous once the analyte signal can be overlapped by the baseline disturbance due to the injection.When the mobile phase is pure water, the retention time of ETU is higher than with water–methanol mixtures, so that pure water was used as mobile phase throughout this work. With water as mobile phase the retention time of ETU was 6.0 min. The calibration graph was rectilinear for 0.005–0.30 mg of ETU (r = 0.999), and the RSD (n = 3) was 0.48%. Recoveries of ETU from fortified serum samples, by means of acid deproteinization, with and without irradiation are showed in Table 4.With the exception of TCA, which signal overlaps the one of ETU, both procedures are adequate for ETU determination. However, best recoveries were obtained from deproteinization with more dilute acids and irradiation (Table 4). Typical chromatograms obtained after acid deproteinization (HNO3), with and without irradiation, are shown in Fig. 1. Because of a small decrease in the retention time of ETU after successive measurements of serum samples, a gradient program was carried out to wash the column between the injections.With this procedure, one chromatographic determination takes 12 min and up to 130 determinations were made with the same column. The best results were obtained with nitric and sulfuric acids. TCA showed interferences and HClO4, for higher fortification levels, presented poor recoveries. Conclusions The proposed method is simple and offers advantages over the conventional protein-precipitation procedure, due to the possibility of using a ten-fold less acid concentration, compared with the usual concentration, and the pH of the supernatant is maintained near to physiological values.The use of HPLC means that the time-consuming derivatization procedures for the ETU determination can be omitted. When procedures with many steps are necessary for an effective cleanup, they can introduce losses due to sample handling. In the case of serum, the injection of the sample onto the HPLC system without a cleanup step is not possible.The use of acidic protein precipitants, with irradiation, to assist sample cleanup prior to HPLC injection, is an effective procedure for ETU trace residue analysis, because little sample preparation and dilution are required, and the pH remains at a value that guarantees the stability of the analyte. Acknowledgements to CNPq for the scholarships and to GTZ/ RFG and CNPq for the financial support.References 1 Newsome, W. H., in Analytical Methods for Pesticides and Plant Growth Regulators, Academic Press, New York, 1980, Vol. XI, p. 197. 2 Fishbein, L., J. Toxicol. Environ. Health, 1976, 1, 713. 3 Dearfield, K. L., Mutat. Res., 1994, 317, 111. 4 Bottomley, P., Hoodless, R. A., and Smart, N. A., in Residue Reviews ed. Gunther, F. A., Springer Verlag, New York, 1985, p. 73. 5 Lentza-Rios, C., Rev. Environ. Contam. Toxicol., 1990, 115, 1. 6 Kobayashi, H., Matano, O., and Goto, S., J. Chromatogr., 1981, 207, 281. 7 Kurttio, P., and Savolainen, K., Scand. J. Work, Environ. Health , 1990, 16, 203. 8 Farrington, D. S., and Hopkins, R. G., Analyst, 1979, 104, 111. 9 Onley, J. H., J. Assoc. Off. Anal. Chem., 1977, 60, 1111. 10 Lehotay, J., Brandsteterov�a, E., and Oktavec, D., J. Liq. Chromatogr., 1992, 15, 525. 11 McDowall, R. D., J. Chromatogr., 1989, 492, 3. 12 Blanchard, J., J. Chromatogr., 1981, 226, 455. 13 Kingston, H. M., and Jassie, L. B., Anal. Chem., 1986, 58, 2534. 14 Gedye, R., Smith, F., Westaway, K., Ali, H., Baldisera, L., Laberge, L., and Roussel, J., Tetrahedron Lett., 1986, 27, 279. 15 Ganzler, K., Salg�o, A., and Valk�o, K., J. Chromatogr., 1986, 371, 299. 16 Cruickshank, P. A., and Jarrow, H. C., J. Agr. Food Chem., 1973, 21, 33. 17 Marshall, W., J. Agric. Food Chem., 1977, 25, 357. 18 Bradford, M. M., Anal. Biochem., 1976, 72, 248. 19 Fielding, J., and Ryall, R. G., Clin. Chim. Acta, 1971, 33, 235. 20 Mathieu, C., and Mathieu, H., Clin. Chem., 1995, 41, 625. Paper 7/01011G Received February 12, 1997 Accepted April 29, 1997 Table 4 Recoveries of ETU from fortified serum samples, witthout MW irradiation as adjuvant for acid deproteinization Recovery (%)* Acid ETU added/mg 121 MW assisted without MW HNO3 2.0 98.15 ± 0.84 — 5.0 88.99 ± 10.39 78.55 ± 4.32 10.0 93.92 ± 2.83 — H2SO4 2.0 106.45 ± 2.76 — 5.0 99.4 ± 2.12 — 10.0 89.13 ± 1.57 — HClO4 3.0 109.16 ± 3.08 — 7.5 94.29 ± 2.53 77.27 ± 3.89 15.0 60.48 ± 5.68 — * Recovery ± RSD (n = 3). Fig. 1 Chromatograms of serum samples fortified with ETU, after deproteinization. (a) Without MW irradiation; (b) with MW irradiation. Analyst, August 1997, Vol. 122 735
ISSN:0003-2654
DOI:10.1039/a701011g
出版商:RSC
年代:1997
数据来源: RSC
|
3. |
Multivariate Statistical Process Control Applied to SulfateDetermination by Sequential Injection Analysis |
|
Analyst,
Volume 122,
Issue 8,
1997,
Page 737-741
A. Rius,
Preview
|
|
摘要:
Multivariate Statistical Process Control Applied to Sulfate Determination by Sequential Injection Analysis A. Rius*, M. P. Callao and F. X. Rius Departament de Qu�ýmica, Universitat Rovira i Virgili, Pl. Imperial T`arraco 1, 43005 Tarragona, Spain. E-mail: arius@quimica.urv.es A methodology was developed for determining sulfates in water at levels up to 500 mg l21 using a sequential injection analysis system. The multivariate calibration model avoids the need for the separation of interferents and sample pre-treatment.The trueness of the multivariate calibration model was assessed by comparing the predictions of the model with reference concentrations determined by a reference method using the joint interval test for the slope and intercept of the regression line with errors on both axes. The accuracy, evaluated by the root mean square error of prediction, reached 6.9%. Multivariate statistical process control techniques were used to check the system’s stability before developing the model and the validity of the model when it is used to predict the concentrations of unknown samples.Keywords: Multivariate statistical process control; sequential injection analysis; multivariate calibration; sulfates The importance of sulfate determination from the standpoint of pollution monitoring is reflected in the maximum admissible concentration of 250 mg l21 in natural waters established by legislation;1 hence the development of a rapid, robust methodology for determining sulfates is of great interest.One of the techniques that meets the necessary requirements for being considered useful in process analysis is flow injection analysis (FIA),2,3 although only a few applications have been designed for process flow injection analysis (PFIA).4,5 Several FIA methodologies have been published on sulfate determination based on turbidimetric,6,7 spectrofluorimetric,8 atomic absorption9 or spectrophotometric10 detection.However, the main reason why FIA analysers are not used in process lines is the need for too much maintenance in such environments, especially the tubes of peristaltic pumps.11,12 A related technique, sequential injection analysis (SIA), was introduced to overcome this problem. Some of the advantages of this technique over FIA are simpler manifolds, less maintenance and stable flow rates. However, the sampling frequency is often reduced.11213 This paper discusses the development of an SIA system for determining sulfates in natural waters.The methodology is based on a previously reported FIA method10 that used univariate spectrophotometric determination of the cation complex FeSO4 +,14 the main interferents being chloride, fluoride and organic compounds. The use of multivariate calibration in the present methodology does not require previous separation of interferents or sample pre-treatment. However, the development of a multivariate model is often a costly and time-consuming process; many calibration samples, necessary to build the model, must be analysed by a reference method.Therefore, it is important to ensure that the analytical system works under statistical control during the modelling process, and that it remains under control while the unknown samples are analysed. To do so, a representative control sample is analysed repeatedly before and during the multivariate model development in order to check the system’s stability.For univariate responses, control charts may be used to monitor the system, but when using multivariate responses, the tool to be used is multivariate statistical process control (MSPC).15–17 The development of the methodology consists of the following steps: first, once the operational parameters for the sequential injection analyser (SIA) have been established, the system’s stability over time is checked using MSPC; second, the sulfate concentration in the calibration set of samples is determined by using the reference method, and these samples are analysed by SIA.The recorded spectra and determined concentrations are used to develop the PLS model. Finally, the model can be used to predict concentrations in unknown samples. The method can determine concentrations up to 500 mg l21, so it is useful beyond the maximum allowed level (250 mg l21). The trueness of the multivariate calibration model was assessed by comparing the cross-validation predicted values of the model with the reference concentrations using the joint interval test for the slope and intercept of the regression line with errors on both axes.The accuracy was evaluated by the root mean square error of prediction (RMSEP). Theory Multivariate Statistical Control of the SIA System The multivariate responses for the analysed control sample are arranged in an X matrix (I 3 J) (I = control sample spectra and J = wavelengths), which is decomposed by principal components analysis (PCA) into a product of two new matrices: X = Tk Pk T + E (1) Tk being the matrix of scores, Pk the matrix of loadings, k the number of factors included in the model and E the model error or residuals.The lack of model fit was assessed by means of the Q statistic.15,18 Q is a scalar [eqn. (2)] that for a given sample xi, measures the amount of variation not accounted for by the PCA model and therefore tests whether the system is operating beyond the common causes for variability: Qi = ei ei T = xi (I 2 PkPk T) xi T (2) where ei is the ith row of E, I is the identity matrix, Pk is the matrix of the k loadings vectors retained in the PCA model and xi is the ith row in X. The Q confidence limits19 can be calculated according to the equation Qa =Q1 ca 2Q2h0 2 Q1 +1+ Q2h0(h0 -1) Q1 2 é ë êê ù û úú 1 h0 (3) where Qi= lj i for i = 1, 2, 3 j=k+1 n å (4) and h0 = 1 2Q1Q3 3Q2 2 (5) Analyst, August 1997, Vol. 122 (737–741) 737k being the number of principal components retained in the model, lj the eigenvalue associated with the jth principal component, n the total number of principal components and ca the standard normal deviation corresponding to a given a. It is assumed for the validity of the Q statistic and its confidence limits that the measurement errors are independent and normally distributed and that the rank of the PCA model is correct. The Hotelling T2 statistic,15,18 which measures the variation in each sample within the PCA model, is calculated according to Ti 2 = xi Pk l21 Pk T xi T (6) where l is the diagonal matrix containing the eigenvalues associated with the eigenvectors included in the PCA model.The statistical confidence limit for T2 can be calculated18 by means of the equation T k m m k F k m k m k , , , , ( ) a a 2 = - - - 1 (7) where m is the number of samples used in the PCA model, k the number of principal components included in the model and Fk,m2k,a the value for the F-distribution for k and m2k degrees of freedom and a given a.Information about the control sample can also be obtained from the plot of scores for the relevant principal components. When there is a change in the system, the scores of the new spectrum for the control sample will be very different from the previous scores, and the change will be detected. However, this information is also included in the Hotelling T2 statistic since it is calculated using the scores.Moreover, the Q statistic gives us additional information which is not included in the scores plot, because it is related to the variation which is not accounted for by the model, and the plots of Q and T2 are a hypothesis test which clearly signals any out of control sample whereas the inspection of the scores plot is a qualitative tool. Multivariate Calibration Partial least squares (PLS) was used to build the multivariate calibration model20 that relates sulfate concentration values in natural water samples to their absorption UV/VIS spectra.The number of factors to be included in the model was selected according to the root mean square error of prediction (RMSEPCV) criterion [eqn. (8)] using cross-validation (leaveone- out) to evaluate predicted concentrations: RMSEPCV = S( �c - c)2 N (8) where �c is the predicted concentration, ce reference concentration and N is the number of calibration samples. The plot of the regression coefficients versus response variables for each factor is also used to aid in the selection of the appropriate number of factors. The accuracy of the method was evaluated by calculating the relative error of prediction, estimated according to RRMSEPCV = 100 c S( �c - c)2 N (9) where �c is the mean of the reference concentrations.The trueness was checked using linear regression with errors on both axes of the predicted concentration values versus the reference concentration values for each sample.Bivariate least squares (BLS) was used in order to take into account the errors for both methods.Theoretically, it should fit a straight line with a slope of unity and zero intercept. The joint confidence interval test for the slope and the intercept21 assesses whether the theoretical value (1,0) is within the confidence ellipse. The ellipse describing the confidence interval is calculated using the equation 1 2 2 2 1 2 2 1 2 2 2 1 2 2 1 s b b x s b b m m x s m m s F i i i i N i i N i N i N e e e a = = - - = å å å - + - - + - = ( � ) ( � )( � ) ( � ) ( , ) (10) where N is the number of samples, b = 0, m = 1, �b is the intercept, m is the slope, xi is the reference concentration, sei 2 is the variance of the residuals and s is the residual standard deviation.21 The value of a which is necessary for the point (0,1) to be inside the confidence region is calculated. Experimental Instrumentation A Hach (Loveland, CO, USA) Model 2100A turbidimeter was used to determine the reference sulfate concentration in samples. The sequential injection analyser has been described elsewhere22 and consists of the following parts: CAVRO (Sunnyvale, CA, USA) XL 3000 stepper motor-driven syringe pump, connected to a PC with an RS-232 interface; a sixposition Eurosas (Paris, France) EPS 1306 BPB automatic valve connected to the computer through a PCL-711S PC-Lab-Card; Omnifit (Cambridge, UK) PTFE tubing reaction coil (70 cm 3 0.8 mm id); holding coil (200 cm 3 0.8 mm id); a Hewlett- Packard (Avondale, PA, USA) HP 8452A diode-array spectrophotometer, controlled by an HP Vectra 386/25 computer equipped with an HP-IB IEEE 488 interface for communications; and a Hellma (Jamaica, NY, USA) Model 178.711QS flow-through cell.A scheme of the analyser is shown in Fig. 1. Samples Thirty-four samples of bottled and underground water and their mixtures were used as the calibration set and covered the range from 0 to 500 mg l21 of sulfate.Chloride is the main interferent in the method10 and therefore several levels of chloride were added to the samples in order to examine its interference within the usual range found in water (0–800 mg l21). A two-factor, five-level complete experimental design (five different levels of sulfate and five levels of chloride concentration) plus another nine samples in the most representative zone were applied. The reference concentration of sulfate in all samples was determined by the turbidimetric reference method.23 One of these samples was used for the stability study of the SIA system, prior to developing the multivariate calibration.Fig. 1 Scheme of the sequential injection analyser. 738 Analyst, August 1997, Vol. 122Reagents and Procedure The SIA method is based on the reaction of sulfate anion with FeIII to form the cationic complex FeSO4 +, which is detected spectrophotometrically.The reagent used is a 0.02 m Fe(NO3)3 and 0.34 m HClO4 solution. The carrier is 0.1 m HClO4 solution. The optimum aspiration order of sample and reagent, the aspiration times, the concentrations of reagent and carrier and the flush-out time to the detector were determined experimentally using several sulfate standards in order to obtain an adequate response (absorbances between 0.2 and 1.0) using the minimum volumes of reagents and ensuring differences in the spectra in the range of concentrations studied (0–500 mg l21 of sulfate).The sequence of the analysis is as follows: (1) carrier aspiration for 55 s; (2) sample aspiration for 2 s; (3) reagent aspiration for 5 s; (4) flush out to detector for 20 s; and (5) recording of the spectrum 20 s after the flow stops. The flow rate is 2 ml min21 for all steps. The spectrum is recorded every 2 nm in the range from 300 to 400 nm, with an integration time of 0.1 s. The sampling frequency achieved is 20 h21.For the turbidimetric method, the buffer solution, reagent and standard solution of sulfates are prepared as described elsewhere. 23 Software HP 89531A software was used to record and store the spectra. Customized software was used to control the syringe pump and the valve. The multivariate calibration models used were programmed using MATLAB functions.24 Results Study of the Stability of the SIA System Using the MSPC Techniques The control sample taken from the calibration set was repeatedly analysed 60 times for a period of 2 d, which is longer than necessary to analyse all the calibration set samples.Thirtyseven analyses were prepared on the first day and 23 on the second. During the analysis of these samples, some disturbances were visually observed in the system: an odd peak in the spectra for sample 38, probably due to a disturbance in the detector, and changes in the absorbance values due to air bubbles in the system during the analysis of sample 54.Moreover, an air bubble was observed in the flow cell between the analyses of 9 and 10; obviously, the spectrum obtained was not considered. After each of these disturbances had been observed, the system was immediately readjusted. A PCA model was built with the 37 spectra obtained on the first day.The plot of the scores for these 37 samples is shown in Fig. 2. It can be seen that sample 10 is detected as an outlier, probably owing to the recent adjustment of the system, as was sample 36, for which there is no known experimental reason why it is detected as an outlier.According to these results, a new model was built with 35 samples (outliers were not detected in the new score plot for this model), and the limits for the Q and T2 statistics were calculated using eqns. (3) and (7) (with a = 0.05). This model, with two factors explaining 99.9% of the variance, was used to calculate the Q and T2 statistics for all 60 control samples, using eqns.(2) and (6). The results are shown in Fig. 3. Fig. 3(a) shows the plot of the Hotelling statistic versus the sample number, and it can be seen that there are six samples beyond the 95% confidence limit. Fig. 3(b) shows a plot of the Q statistic versus the sample number, and it can be seen that there are five samples beyond the 95% confidence limit. From these data, it can be deduced that the three replicates of the control sample with assignable special causes of variability (numbers 10, 38 and 54) are clearly detected in the two charts.There are also other samples beyond the limits of both control charts: one of them is sample 36, which is one of the samples not included in the model, as discussed before. Moreover, sample 37 (beyond the limits in both plots) and sample 32 (beyond the limit in the Hotelling statistic plot) do not have assignable causes of variability and both would be considered normal samples at the 99% confidence level since they are very close to the confidence limits.Therefore, the application of MSPC indicates the occurrence of any disturbance by means of the Q and T2 charts. This way of detecting out of control samples has advantages over only inspecting the spectrum or the scores plot. Fig. 4 shows the spectra of the control samples included in the PCA model together with the spectra of samples 10, 36 and 54. From inspection of this plot (and also from the plot of the scores, see Fig. 2), it could be interpreted that these samples were analysed when the system was out of control, but this conclusion is much more evident from the charts of the Q and T2 statistics (Fig. 3). Moreover, the computation of these statistics can be easily automated so that when an out of control sample occurs it can be automatically detected, facilitating the revision of the system and the resuknown samples analysed since the last control sample. This means that this methodology enables Fig. 2 Plot of the scores for the PCA model built from the first-day control sample spectra. Fig. 3 (a) Plot of the Hotelling statistic versus sample number; (b) plot of the Q statistic versus sample number. Broken lines, 95% confidence limits; and dotted lines, change of scale. Analyst, August 1997, Vol. 122 739disturbances to be detected during the process of recording spectra for the calibration samples to be used for the development of the multivariate model.Because the calibration sample spectra are not still registered during the application of the MSPC techniques, these are the only tools to ensure that no special causes of variability are occurring at this stage. In general, the SIA system has been shown to be stable over time, except in special cases. Multivariate Calibration Validation The concentration and spectra of samples in the calibration set (34 samples) were used to build the multivariate model PLS-1; the spectra were recorded with the system working under statistical control, i.e., when only the unavoidable variability is present. Fig. 5(a) shows the root mean squared prediction error of cross-validation (RMSEPCV) versus the number of factors for the model. As can be observed, the minimum error is reached at five factors. However, the plot of regression coefficients versus response variables for each factor [Fig. 5(b)] shows a smooth trend for four factors while a sharp wandering trend is obtained for five factors, indicating that the fifth factor might include a certain amount of noise.Therefore, four factors were chosen to build the model. The explained variance of the four-factor model is 100% for X-variables and 99.6% for the Yvariable. The plot of the sample scores in the reduced space of the first two factors, which explains 99.8% of the variance of the independent variables, shows that neither outliers nor submodels can be observed [Fig. 6(a)].The plot of the residual variance versus object leverage values confirms the previous conclusions [Fig. 6(b)] The accuracy, given by the relative error of prediction of the model [eqn. (9)] when four factors are included is 6.9%. The trueness of the method was assessed through the joint confidence interval test of slope and intercept for the straight line obtained by regressing the predicted SIA values (obtained by cross-validation) on the reference concentrations values taking into account errors in both axes [eqn.(10)]. The uncertainty associated with the sulfate concentration for the xaxis (reference method) is calculated from the regression line of the turbidimetric method while the uncertainty associated with the y-axis is evaluated from the MSEPCV values of the multivariate model. It is assumed that the variances are uniform for all samples. The value of 12a calculated using eqn. (10) is 0.71, so it can be concluded that there is no significant difference in relation to slope 1 and intercept 0, i.e., no systematic error was detected, for a given a = 0.29.The PLS model developed was used to predict the sulfate concentrations of the control sample spectra which had been used to study the stability of the system. The results are shown in Fig. 7. It can be seen that the predicted values for samples 10, 38 and 54 clearly incorporate bias, but these are samples that were previously identified as outliers by the MSPC control charts (Fig. 3). Fig. 4 Spectra of the 37 control samples included in the PCA model, together with the spectra of some abnormal samples. Fig. 5 (a) Plot of RMSEPCV versus the number of factors for the PLS model; (b) plot of the regression coefficients versus independent variables for the fourth and fifth factors. Fig. 6 (a) Plot of the scores for the PLS model for the 34 samples of the calibration set; (b) plot of the residual variance versus leverage for the calibration set samples.Fig. 7 Plot of the predicted sulfate concentration for the 60 control samples versus sample number. 740 Analyst, August 1997, Vol. 122Conclusions An SIA method has been developed for determining sulfates in natural waters at concentrations up to 500 mg l21. The low consumption of carrier (1 ml per analysis) and reagent (0.17 ml per analysis) compared with other techniques, the frequency of analysis (20 samples per hour) and the use of multivariate calibration (which avoids the need for sample pre-treatment and separation of interferents) make this methodology useful for routine determinations.The study of the system’s stability overtime during the model development demonstrated its stability, and the use of MSPC techniques allows one to check if it is working under statistical control. A study of the multivariate and univariate control graphs, obtained by applying the model to the signals of the control sample, shows that the information from both of them is comparable.This means that the signals obtained before and during the development of the model can be used as prior data to plot a control graph of concentrations while the model is being used. The values of accuracy and trueness obtained show the quality performance of the system. Financial support from the Spanish Ministry of Education and Science (DGICyT, project No. BP93-0366) is acknowledged. References 1 Bolet�ýn Oficial del Estado, 20 September 1990, R.D. 1138/90, p. 27493. 2 Valcarcel, M., and Luque de Castro, M. D., Flow Injection Analysis. Principles and Applications, Ellis Horwood, Chichester, 1987. 3 Ruzicka, J., and Hansen, E. H., Flow Injection Analysis, Wiley- Interscience, New York, 2nd edn., 1988. 4 Yalvac, E. D., and Bredeweg, R. A., Proc. Contr. Qual., 1994, 6, 81. 5 Beebe, K. R., Blaser, W. W., Bredeweg, R. A., Chauvel, J. P., Harner, R. S., LaPack, M., Leugers, A., Martin, D. P., Wrignt, L. G., and Yalvac, E.D., Anal. Chem., 1993, 65, 199R. 6 Santelli, R. E., Salgado Lopes, P. R., Leme Santelli, R. C., and Rebello Wagener, A. de L., Anal. Chim. Acta, 1995, 300, 149. 7 Zhi, A. L., Rios, A., and Valcarcel, M., Quim. Anal., 1994, 13, 121. 8 Chimpalee, N., Chimpalee, D., Suparuknari, S., Boonyanitchayakul, B., and Thorburn Burns, D., Anal. Chim. Acta, 1994, 298, 401. 9 Gallego, M., and Valcarcel, M., Mikrochim. Acta, 1991, III (4–6), 163. 10 Kojlo, A., Michalowski, J., and Trojanowicz, M., Anal.Chim. Acta, 1990, 228, 287. 11 Ruzicka, J., Marshall, G. D., and Christian, G. D., Anal. Chem., 1990, 62, 1861. 12 Marshall, G. D., and van Staden, J. F., Proc. Contr. Qual., 1992, 3, 251–261. 13 Ruzicka, J., and Marshall, G. D., Anal. Chim. Acta, 1990, 237, 329. 14 Goguel, R., Anal. Chem., 1969, 41, 1034. 15 Wise, B. M., Ricker, N. L., Veltkamp, D. F., and Kowalski, B. R., Proc. Contr. Qual., 1990, 1, 41. 16 Veltkamp, D. J., Proc. Contr. Qual., 1993, 5, 205. 17 Kourti, T., and MacGregor, J. F., Chemom. Intell. Lab. Syst., 1995, 28, 3. 18 McLennan, F., and Kowalski, B. R., Process Analytical Chemistry, Chapman and Hall, London, 1995. 19 Jackson, J. E., and Mudholkar, G. S., Technometrics, 1979, 21, 341. 20 Martens, H., and Naes, T., Multivariate Calibration, Wiley, Chichester, 1989. 21 Riu, J., and Rius, F. X., Anal. Chem., 1996, 68, 1851. 22 Rius, A., Callao, M. P., and Rius, F. X., Anal. Chim. Acta, 1995, 316, 27. 23 Standard Methods for the Examination of Water and Wastewater, ed.Franson, M. A. H., APHA, AWWA and WPCF, Washington, DC, 17 edn., 1992, Sect. 4500. 24 MATLAB, The Mathworks, South Natick, MA, USA. Paper 6/07954G Received November 25, 1996 Accepted April 16, 1997 Analyst, August 1997, Vol. 122 741 Multivariate Statistical Process Control Applied to Sulfate Determination by Sequential Injection Analysis A. Rius*, M. P. Callao and F. X. Rius Departament de Qu�ýmica, Universitat Rovira i Virgili, Pl.Imperial T`arraco 1, 43005 Tarragona, Spain. E-mail: arius@quimica.urv.es A methodology was developed for determining sulfates in water at levels up to 500 mg l21 using a sequential injection analysis system. The multivariate calibration model avoids the need for the separation of interferents and sample pre-treatment. The trueness of the multivariate calibration model was ased by comparing the predictions of the model with reference concentrations determined by a reference method using the joint interval test for the slope and intercept of the regression line with errors on both axes.The accuracy, evaluated by the root mean square error of prediction, reached 6.9%. Multivariate statistical process control techniques were used to check the system’s stability before developing the model and the validity of the model when it is used to predict the concentrations of unknown samples. Keywords: Multivariate statistical process control; sequential injection analysis; multivariate calibration; sulfates The importance of sulfate determination from the standpoint of pollution monitoring is reflected in the maximum admissible concentration of 250 mg l21 in natural waters established by legislation;1 hence the development of a rapid, robust methodology for determining sulfates is of great interest. One of the techniques that meets the necessary requirements for being considered useful in process analysis is flow injection analysis (FIA),2,3 although only a few applications have been designed for process flow injection analysis (PFIA).4,5 Several FIA methodologies have been published on sulfate determination based on turbidimetric,6,7 spectrofluorimetric,8 atomic absorption9 or spectrophotometric10 detection.However, the main reason why FIA analysers are not used in process lines is the need for too much maintenance in such environments, especially the tubes of peristaltic pumps.11,12 A related technique, sequential injection analysis (SIA), was introduced to overcome this problem.Some of the advantages of this technique over FIA are simpler manifolds, less maintenance and stable flow rates. However, the sampling frequency is often reduced.11213 This paper discusses the development of an SIA system for determining sulfates in natural waters. The methodology is based on a previously reported FIA method10 that used univariate spectrophotometric determination of the cation complex FeSO4 +,14 the main interferents being chloride, fluoride and organic compounds.The use of multivariate calibration in the present methodology does not require previous separation of interferents or sample pre-treatment. However, the development of a multivariate model is often a costly and time-consuming process; many calibration samples, necessary to build the model, must be analysed by a reference method. Therefore, it is important to ensure that the analytical system works under statistical control during the modelling process, and that it remains under control while the unknown samples are analysed. To do so, a representative control sample is analysed repeatedly before and during the multivariate model development in order to check the system’s stability. For univariate responses, control charts may be used to monitor the system, but when using multivariate responses, the tool to be used is multivariate statistical process control (MSPC).15–17 The development of the methodology consists of the following steps: first, once the operational parameters for the sequential injection analyser (SIA) have been established, the system’s stability over time is checked using MSPC; second, the sulfate concentration in the calibration set of samples is determined by using the reference method, and these samples are analysed by SIA.The recorded spectra and determined concentrations are used to develop the PLS model.Finally, the model can be used to predict concentrations in unknown samples. The method can determine concentrations up to 500 mg l21, so it is useful beyond the maximum allowed level (250 mg l21). The trueness of the multivariate calibration model was assessed by comparing the cross-validation predicted values of the model with the reference concentrations using the joint interval test for the slope and intercept of the regression line with errors on both axes.The accuracy was evaluated by the root mean square error of prediction (RMSEP).Theory Multivariate Statistical Control of the SIA System The multivariate responses for the analysed control sample are arranged in an X matrix (I 3 J) (I = control sample spectra and J = wavelengths), which is decomposed by principal components analysis (PCA) into a product of two new matrices: X = Tk Pk T + E (1) Tk being the matrix of scores, Pk the matrix of loadings, k the number of factors included in the model and E the model error or residuals.The lack of model fit was assessed by means of the Q statistic.15,18 Q is a scalar [eqn. (2)] that for a given sample xi, measures the amount of variation not accounted for by the PCA model and therefore tests whether the system is operating beyond the common causes for variability: Qi = ei ei T = xi (I 2 PkPk T) xi T (2) where ei is the ith row of E, I is the identity matrix, Pk is the matrix of the k loadings vectors retained in the PCA model and xi is the ith row in X.The Q confidence limits19 can be calculated according to the equation Qa =Q1 ca 2Q2h0 2 Q1 +1+ Q2h0(h0 -1) Q1 2 é ë êê ù û úú 1 h0 (3) where Qi= lj i for i = 1, 2, 3 j=k+1 n å (4) and h0 = 1 2Q1Q3 3Q2 2 (5) Analyst, August 1997, Vol. 122 (737–741) 737k being the number of principal components retained in the model, lj the eigenvalue associated with the jth principal component, n the total number of principal components and ca the standard normal deviation corresponding to a given a.It is assumed for the validity of the Q statistic and its confidence limits that the measurement errors are independent and normally distributed and that the rank of the PCA model is correct. The Hotelling T2 statistic,15,18 which measures the variation in each sample within the PCA model, is calculated according to Ti 2 = xi Pk l21 Pk T xi T (6) where l is the diagonal matrix containing the eigenvalues associated with the eigenvectors included in the PCA model.The statistical confidence limit for T2 can be calculated18 by means of the equation T k m m k F k m k m k , , , , ( ) a a 2 = - - - 1 (7) where m is the number of samples used in the PCA model, k the number of principal components included in the model and Fk,m2k,a the value for the F-distribution for k and m2k degrees of freedom and a given a. Information about the control sample can also be obtained from the plot of scores for the relevant principal components.When there is a change in the system, the scores of the new spectrum for the control sample will be very different from the previous scores, and the change will be detected. However, this information is also included in the Hotelling T2 statistic since it is calculated using the scores. Moreover, the Q statistic gives us additional information which is not included in the scores plot, because it is related to the variation which is not accounted for by the model, and the plots of Q and T2 are a hypothesis test which clearly signals any out of control sample whereas the inspection of the scores plot is a qualitative tool. Multivariate Calibration Partial least squares (PLS) was used to build the multivariate calibration model20 that relates sulfate concentration values in natural water samples to their absorption UV/VIS spectra.The number of factors to be included in the model was selected according to the root mean square error of prediction (RMSEPCV) criterion [eqn.(8)] using cross-validation (leaveone- out) to evaluate predicted concentrations: RMSEPCV = S( �c - c)2 N (8) where �c is the predicted concentration, c is the reference concentration and N is the number of calibration samples. The plot of the regression coefficients versus response variables for each factor is also used to aid in the selection of the appropriate number of factors. The accuracy of the method was evaluated by calculating the relative error of prediction, estimated according to RRMSEPCV = 100 c S( �c - c)2 N (9) where �c is the mean of the reference concentrations. The trueness was checked using linear regression with errors on both axes of the predicted concentration values versus the reference concentration valuesh sample.Bivariate least squares (BLS) was used in order to take into account the errors for both methods.Theoretically, it should fit a straight line with a slope of unity and zero intercept.The joint confidence interval test for the slope and the intercept21 assesses whether the theoretical value (1,0) is within the confidence ellipse. The ellipse describing the confidence interval is calculated using the equation 1 2 2 2 1 2 2 1 2 2 2 1 2 2 1 s b b x s b b m m x s m m s F i i i i N i i N i N i N e e e a = = - - = å å å - + - - + - = ( � ) ( � )( � ) ( � ) ( , ) (10) where N is the number of samples, b = 0, m = 1, �b is the intercept, m is the slope, xi is the reference concentration, sei 2 is the variance of the residuals and s is the residual standard deviation.21 The value of a which is necessary for the point (0,1) to be inside the confidence region is calculated.Experimental Instrumentation A Hach (Loveland, CO, USA) Model 2100A turbidimeter was used to determine the reference sulfate concentration in samples. The sequential injection analyser has been described elsewhere22 and consists of the following parts: CAVRO (Sunnyvale, CA, USA) XL 3000 stepper motor-driven syringe pump, connected to a PC with an RS-232 interface; a sixposition Eurosas (Paris, France) EPS 1306 BPB automatic valve connected to the computer through a PCL-711S PC-Lab-Card; Omnifit (Cambridge, UK) PTFE tubing reaction coil (70 cm 3 0.8 mm id); holding coil (200 cm 3 0.8 mm id); a Hewlett- Packard (Avondale, PA, USA) HP 8452A diode-array spectrophotometer, controlled by an HP Vectra 386/25 computer equipped with an HP-IB IEEE 488 interface for communications; and a Hellma (Jamaica, NY, USA) Model 178.711QS flow-through cell.A scheme of the analyser is shown in Fig. 1. Samples Thirty-four samples of bottled and underground water and their mixtures were used as the calibration set and covered the range from 0 to 500 mg l21 of sulfate. Chloride is the main interferent in the method10 and therefore several levels of chloride were added to the samples in order to examine its interference within the usual range found in water (0–800 mg l21).A two-factor, five-level complete experimental design (five different levels of sulfate and five levels of chloride concentration) plus another nine samples in the most representative zone were applied. The reference concentration of sulfate in all samples was determined by the turbidimetric reference method.23 One of these samples was used for the stability study of the SIA system, prior to developing the multivariate calibration.Fig. 1 Scheme of the sequential injection analyser. 738 Analyst, August 1997, Vol. 122Reagents and Procedure The SIA method is based on the reaction of sulfate anion with FeIII to form the cationic complex FeSO4 +, which is detected spectrophotometrically. The reagent used is a 0.02 m Fe(NO3)3 and 0.34 m HClO4 solution. The carrier is 0.1 m HClO4 solution. The optimum aspiration order of sample and reagent, the aspiration times, the concentrations of reagent and carrier and the flush-out time to the detector were determined experimentally using several sulfate standards in order to obtain an adequate response (absorbances between 0.2 and 1.0) using the minimum volumes of reagents and ensuring differences in the spectra in the range of concentrations studied (0–500 mg l21 of sulfate).The sequence of the analysis is as follows: (1) carrier aspiration for 55 s; (2) sample aspiration for 2 s; (3) reagent aspiration for 5 s; (4) flush out to detector for 20 s; and (5) recording of the spectrum 20 s after the flow stops.The flow rate is 2 ml min21 for all steps. The spectrum is recorded every 2 nm in the range from 300 to 400 nm, with an integration time of 0.1 s. The sampling frequency achieved is 20 h21. For the turbidimetric method, the buffer solution, reagent and standard solution of sulfates are prepared as described elsewhere. 23 Software HP 89531A software was used to record and store the spectra.Customized software was used to control the syringe pump and the valve. The multivariate calibration models used were programmed using MATLAB functions.24 Results Study of the Stability of the SIA System Using the MSPC Techniques The control sample taken from the calibration set was repeatedly analysed 60 times for a period of 2 d, which is longer than necessary to analyse all the calibration set samples. Thirtyseven analyses were prepared on the first day and 23 on the second.During the analysis of these samples, some disturbances were visually observed in the system: an odd peak in the spectra for sample 38, probably due to a disturbance in the detector, and changes in the absorbance values due to air bubbles in the system during the analysis of sample 54. Moreover, an air bubble was observed in the flow cell between the analyses of 9 and 10; obviously, the spectrum obtained was not considered.After each of these disturbances had been observed, the system was immediately readjusted. A PCA model was built with the 37 spectra obtained on the first day.The plot of the scores for these 37 samples is shown in Fig. 2. It can be seen that sample 10 is detected as an outlier, probably owing to the recent adjustment of the system, as was sample 36, for which there is no known experimental reason why it is detected as an outlier. According to these results, a new model was built with 35 samples (outliers were not detected in the new score plot for this model), and the limits for the Q and T2 statistics were calculated using eqns.(3) and (7) (with a = 0.05). This model, with two factors explaining 99.9% of the variance, was used to calculate the Q and T2 statistics for all 60 control samples, using eqns. (2) and (6). The results are shown in Fig. 3. Fig. 3(a) shows the plot of the Hotelling statistic versus the sample number, and it can be seen that there are six samples beyond the 95% confidence limit.Fig. 3(b) shows a plot of the Q statistic versus the sample number, and it can be seen that there are five samples beyond the 95% confidence limit. From these data, it can be deduced that the three replicates of the control sample with assignable special causes of variability (numbers 10, 38 and 54) are clearly detected in the two charts. There are also other samples beyond the limits of both control charts: one of them is sample 36, which is one of the samples not included in the model, as discussed before.Moreover, sample 37 (beyond the limits in both plots) and sample 32 (beyond the limit in the Hotelling statistic plot) do not have assignable causes of variability and both would be considered normal samples at the 99% confidence level since they are very close to the confidence limits. Therefore, the application of MSPC indicates the occurrence of any disturbance by means of the Q and T2 charts.This way of detecting out of control samples has advantages over only inspecting the spectrum or the scores plot. Fig. 4 shows the spectra of the control samples included in the PCA model together with the spectra of samples 10, 36 and 54. From inspection of this plot (and also from the plot of the scores, see Fig. 2), it could be interpreted that these samples were analysed when the system was out of control, but this conclusion is much more evident from the charts of the Q and T2 statistics (Fig. 3). Moreover, the computation of these statistics can be easily automated so that when an out of control sample occurs it can be automatically detected, facilitating the revision of the system and the results for the unknown samples analysed since the last control sample. This means that this methodology enables Fig. 2 Plot of the scores for the PCA model built from the first-day control sample spectra. Fig. 3 (a) Plot of the Hotelling statistic versus sample number; (b) plot of the Q statistic versus sample number.Broken lines, 95% confidence limits; and dotted lines, change of scale. Analyst, August 1997, Vol. 122 739disturbances to be detected during the process of recording spectra for the calibration samples to be used for the development of the multivariate model. Because the calibratiopectra are not still registered during the application of the MSPC techniques, these are the only tools to ensure that no special causes of variability are occurring at this stage.In general, the SIA system has been shown to be stable over time, except in special cases. Multivariate Calibration Validation The concentration and spectra of samples in the calibration set (34 samples) were used to build the multivariate model PLS-1; the spectra were recorded with the system working under statistical control, i.e., when only the unavoidable variability is present. Fig. 5(a) shows the root mean squared prediction error of cross-validation (RMSEPCV) versus the number of factors for the model.As can be observed, the minimum error is reached at five factors. However, the plot of regression coefficients versus response variables for each factor [Fig. 5(b)] shows a smooth trend for four factors while a sharp wandering trend is obtained for five factors, indicating that the fifth factor might include a certain amount of noise. Therefore, four factors were chosen to build the model.The explained variance of the four-factor model is 100% for X-variables and 99.6% for the Yvariable. The plot of the sample scores in the reduced space of the first two factors, which explains 99.8% of the variance of the independent variables, shows that neither outliers nor submodels can be observed [Fig. 6(a)]. The plot of the residual variance versus object leverage values confirms the previous conclusions [Fig. 6(b)] The accuracy, given by the relative error of prediction of the model [eqn.(9)] when four factors are included is 6.9%. The trueness of the method was assessed through the joint confidence interval test of slope and intercept for the straight line obtained by regressing the predicted SIA values (obtained by cross-validation) on the reference concentrations values taking into account errors in both axes [eqn. (10)]. The uncertainty associated with the sulfate concentration for the xaxis (reference method) is calculated from the regression line of the turbidimetric method while the uncertainty associated with the y-axis is evaluated from the MSEPCV values of the multivariate model.It is assumed that the variances are uniform for all samples. The value of 12a calculated using eqn. (10) is 0.71, so it can be concluded that there is no significant difference in relation to slope 1 and intercept 0, i.e., no systematic error was detected, for a given a = 0.29. The PLS model developed was used to predict the sulfate concentrations of the control sample spectra which had been used to study the stability of the system.The results are shown in Fig. 7. It can be seen that the predicted values for samples 10, 38 and 54 clearly incorporate bias, but these are samples that were previously identified as outliers by the MSPC control charts (Fig. 3). Fig. 4 Spectra of the 37 control samples included in the PCA model, together with the spectra of some abnormal samples.Fig. 5 (a) Plot of RMSEPCV versus the number of factors for the PLS model; (b) plot of the regression coefficients versus independent variables for the fourth and fifth factors. Fig. 6 (a) Plot of the scores for the PLS model for the 34 samples of the calibration set; (b) plot of the residual variance versus leverage for the calibration set samples. Fig. 7 Plot of the predicted sulfate concentration for the 60 control samples versus sample number. 740 Analyst, August 1997, Vol. 122Conclusions An SIA method has been developed for determining sulfates in natural waters at concentrations up to 500 mg l21. The low consumption of carrier (1 ml per analysis) and reagent (0.17 ml per analysis) compared with other techniques, the frequency of analysis (20 samples per hour) and the use of multivariate calibration (which avoids the need for sample pre-treatment and separation of interferents) make this methodology useful for routine determinations. The study of the system’s stability overtime during the model development demonstrated its stability, and the use of MSPC techniques allows one to check if it is working under statistical control.A study of the multivariate and univariate control graphs, obtained by applying the model to the signals of the control sample, shows that the information from both of them is comparable. This means that the signals obtained before and during the development of the model can be used as prior data to plot a control graph of concentrations while the model is being used.The values of accuracy and trueness obtained show the quality performance of the system. Financial support from the Spanish Ministry of Education and Science (DGICyT, project No. BP93-0366) is acknowledged. References 1 Bolet�ýn Oficial del Estado, 20 September 1990, R. D. 1138/90, p. 27493. 2 Valcarcel, M., and Luque de Castro, M. D., Flow Injection Analysis. Principles and Applications, Ellis Horwood, Chichester, 1987. 3 Ruzicka, J., and Hansen, E. H., Flow Injection Analysis, Wiley- Interscience, New York, 2nd edn., 1988. 4 Yalvac, E. D., and Bredeweg, R. A., Proc. Contr. Qual., 1994, 6, 81. 5 Beebe, K. R., Blaser, W. W., Bredeweg, R. A., Chauvel, J. P., Harner, R. S., LaPack, M., Leugers, A., Martin, D. P., Wrignt, L. G., and Yalvac, E. D., Anal. Chem., 1993, 65, 199R. 6 Santelli, R. E., Salgado Lopes, P. R., Leme Santelli, R. C., and Rebello Wagener, A. de L., Anal. Chim. Acta, 1995, 300, 149. 7 Zhi, A. L., Rios, A., and Valcarcel, M., Quim. Anal., 1994, 13, 121. 8 Chimpalee, N., Chimpalee, D., Suparuknari, S., Boonyanitchayakul, B., and Thorburn Burns, D., Anal. Chim. Acta, 1994, 298, 401. 9 Gallego, M., and Valcarcel, M., Mikrochim. Acta, 1991, III (4–6), 163. 10 Kojlo, A., Michalowski, J., and Trojanowicz, M., Anal. Chim. Acta, 1990, 228, 287. 11 Ruzicka, J., Marshall, G. D., and Christian, G. D., Anal. Chem., 1990, 62, 1861. 12 Marshall, G. D., and van Staden, J. F., Proc. Contr. Qual., 1992, 3, 251–261. 13 Ruzicka, J., and Marshall, G. D., Anal. Chim. Acta, 1990, 237, 329. 14 Goguel, R., Anal. Chem., 1969, 41, 1034. 15 Wise, B. M., Ricker, N. L., Veltkamp, D. F., and Kowalski, B. R., Proc. Contr. Qual., 1990, 1, 41. 16 Veltkamp, D. J., Proc. Contr. Qual., 1993, 5, 205. 17 Kourti, T., and MacGregor, J. F., Chemom. Intell. Lab. Syst., 1995, 28, 3. 18 McLennan, F., and Kowalski, B. R., Process Analytical Chemistry, Chapman and Hall, London, 1995. 19 Jackson, J. E., and Mudholkar, G. S., Technometrics, 1979, 21, 341. 20 Martens, H., and Naes, T., Multivariate Calibration, Wiley, Chichester, 1989. 21 Riu, J., and Rius, F. X., Anal. Chem., 1996, 68, 1851. 22 Rius, A., Callao, M. P., and Rius, F. X., Anal. Chim. Acta, 1995, 316, 27. 23 Standard Methods for the Examination of Water and Wastewater, ed. Franson, M. A. H., APHA, AWWA and WPCF, Washington, DC, 17 edn., 1992, Sect. 4500. 24 MATLAB, The Mathworks, South Natick, MA, USA. Paper 6/07954G Received November 25, 1996 Accepted April 16, 1997 Analyst, August 1997, Vol. 122 7
ISSN:0003-2654
DOI:10.1039/a607954g
出版商:RSC
年代:1997
数据来源: RSC
|
4. |
Evaluation of Portable X-ray Fluorescence Instrumentation forin situ Measurements of Lead on Contaminated Land |
|
Analyst,
Volume 122,
Issue 8,
1997,
Page 743-749
Ariadni Argyraki,
Preview
|
|
摘要:
Evaluation of Portable X-ray Fluorescence Instrumentation for in situMeasurements of Lead on Contaminated Land Ariadni Argyrakia, Michael H. Ramseya and Philip J. Pottsb a Department of Geology, Imperial College, London, UK SW7 2BP b Department of Earth Science, The Open University, Walton Hall, Milton Keynes, UK MK7 6AA The performance of the Spectrace TN 9000 portable X-ray fluorescence (P-XRF) instrument for in situ sampling and analysis of contaminated soil was evaluated.The method was compared with laboratory analysis of the samples using ICP-AES and XRF as established methods for the assessment of contaminated land. The trueness of the field-based P-XRF results was affected by the soil moisture and the surface roughness of the in situ samples, after the correction of which, no bias was observed between the analytical results of the comparative methods. Relatively large measurement uncertainty (±55% for Pb) was caused by the small sample mass analysed and the small scale heterogeneity of the sample.This uncertainty was quantified using duplicate measurements and does not impair the delineation of ‘hot spots’ of contamination as it contributes less than 20% to the total variance. General advantages and limitations of the P-XRF methodology for the investigation of contaminated land were assessed and suggestions are made for the optimisation of the methodology. Keywords: Portable X-ray fluorescence instrumentation; in situ measurement; lead; contaminated land In situ measurements of soils at contaminated sites potentially have the advantages of giving both a rapid assessment of the concentration of the contaminant and also first hand information on its spatial distribution and the degree of heterogeneity in an undisturbed position.The disadvantage of in situ measurement is often the high degree of uncertainty associated with the measurement, caused by factors such as the heterogeneity.If, however, this uncertainty can be quantified realistically, then these measurements can be fit for many purposes of environmental interpretation. Conventional schemes of analysis can usually be divided into two basic stages: (i) the collection of samples in the field, and (ii) the preparation and analysis of these samples in the laboratory, using a variety of analytical methods. The selection of the most appropriate sample preparation and analysis method depends on the sampling medium, the scope of the investigation and the availability of facilities.One well established procedure for the analysis of contaminated soils is to collect samples of top soil (e.g., 0–150 mm) using a hand auger, sieve and collect the < 2 mm fraction, grind to a fine powder and then use an acid extraction procedure, followed by ICP-AES to determine selected analytes. 1 However, it is important to appreciate that the estimated values of contaminant concentration from such investigations may vary due to sampling errors associated with the particular sampling design used in the field.2 X-ray fluorescence (XRF) is also a well established methodology for the assessment of contaminated land.Samples may be dried and then analysed as loose powders or alternatively prepared as compressed powder pellets. Ease of sample preparation as well as analytical trueness and high precision are advantages of this methodology.3 Field-based portable XRF (P-XRF) instrumentation offers potential advantages over other laboratory techniques.By undertaking analyses of contaminated soil in situ, and thereby avoiding the necessity of removing samples, P-XRF has the potential of giving both a rapid assessment of the concentration of the contaminant and also immediate information on its spatial distribution and degree of heterogeneity without disturbing the location. One disadvantage of in situ measurements is the high degree of uncertainty that may be associated with the measurement, caused by sample heterogeneity effects.If, however, this uncertainty can be quantified realistically, then these measurements may be fit for many purposes of environmental interpretation. As noted for in situ measurements in general, the potential problem of high measurement uncertainty can be addressed by its realistic quantification and recognition in environmental interpretation. Field-based P-XRF instruments are becoming increasingly important in assessments of contaminated land.Their performance for the investigation of contaminated land has recently been evaluated by several workers but after removal of the soils to a field laboratory. Specifically, P-XRF devices have been used for the immediate delineation of Cr source contamination ‘hot spots’4 and the spatial distribution of Pb concentration in residential soils.5 P-XRF methodology has also been used to provide data for remedial activities at sites contaminated with Pb and As6 and the determination of Pb in urban soil and dust samples.7 Comparisons of the estimates of elemental concentrations made using P-XRF with estimates from other analytical methods such as ICP-AES and AAS have revealed data of acceptable quality when the portable methodology is used properly.However, a slight bias between the XRF results and those of the other methods has been reported.4–9 A significant problem of the P-XRF methodology, addressed in most of the above studies, is the use of appropriate calibration samples. This paper uses in situ rather than laboratory-based measurements using a Spectrace TN 9000 P-XRF instrument for the determination of Pb in soil.The main aim of this work was to evaluate the performance of a field-based P-XRF instrument for the determination of Pb in contaminated soil in comparison with an established method of analysis, i.e., hand augering followed by laboratory-based ICP-AES.To facilitate the evaluation, the field chosen for this assessment had previously been studied in some detail as part of an on-going study of environmental sampling procedures.2,10,11 The use of P-XRF for spatial mapping, for locating contamination ‘hot-spots’ and for flexibility in the design of sampling protocols was also assessed. Results are used to demonstrate both the advantages and limitations of the technique. The rational optimisation of the sampling and analytical protocol using P-XRF for the investigation of contaminated land, both in situ and in the laboratory, was another objective of this work. The last objective was to identify needs for further development work, which will enable the ultimate limits of performance of P-XRF to be approached.Analyst, August 1997, Vol. 122 (743–749) 743Experimental P-XRF Field measurements were undertaken with a Tracor Northern Spectrace TN 9000 P-XRF instrument on hire from Thermo Unicam (Cambridge, UK).The instrument consisted of a handheld analyser unit and a portable spectrum acquisition and data processing unit (Fig. 1). The analyser unit incorporated three radioactive isotope excitation sources: 55Fe, 109Cd and 241Am and a mercury(ii) iodide X-ray detector. To undertake an analysis in the field, the analyser unit was placed against the sample surface which was excited with each of the three sources in turn. For the analysis of samples in the laboratory (e.g., powdered soil), the analyser unit was mounted in a laboratory stand with the analyser window pointing vertically and covered by a sample lid, the position of which (open or closed) was interlocked to the operation of the instrument.Samples were then conveniently contained in conventional XRF sample cups, placed in position over the window, and analysed once the lid was closed. Whatever the form of sample presentation, spectra were accumulated in the spectrum acquisition and data processing unit and at the end of the preset count time, fluorescence intensities were measured using a spectrum deconvolution procedure based on region-of-interest integration. Data were then corrected for matrix effects and quantified using a fundamental parameter procedure.Quantitative results were displayed for immediate appraisal and stored for subsequent downloading to an external microcomputer via an RS232 interface. The performance of this XRF technique in the laboratory has been described by Potts et al.3 These earlier studies were undertaken on a range of silicate rock reference materials prepared as compressed powder pellet samples, and were designed to demonstrate the basic performance characteristics of the technique.Using a count time (live time) of 200 s per source, repeatability precision was found to be in the range 0.45–1.8% for the major elements and 2–5% for trace elements. A high degree of linearity was also achieved in the relationship between analysed and expected values for the 70 international reference materials investigated.The detection limit for Pb was found to be 39 mg g21, representative of a 200 s count time. ICP-AES Field-based P-XRF results were compared with an established ICP-AES technique. Samples representing the top 0–150 mm surface of the soil were removed with a hand auger and returned to the laboratory for analysis. There, samples were dried, ground in a pestle and mortar and analysed by ICP-AES after acid extraction.Full details of the procedure can be found elsewhere.1 The Site The site selected for study was a fallow field at Bolehill near Wirksworth, Derbyshire, UK. This site was used in Medieval times for Pb smelting. Recent interest arose, therefore, in the possibility of using the presence of high levels of Pb contamination known to be present in the soil to model the mobility of this contaminant in the environment.12 Previous investigations by auger sampling and ICP-AES analysis had revealed mean Pb concentrations of 6229 mg g21 in the top 0–150 mm soil layer over the site.2 The spatial distribution of the element was also well known from both this pilot study (Fig. 2), and also a proficiency testing and collaborative sampling trial which had been based on part of the same field.10,11 Useful information about the sampling variance was available from these trials as well as analytical bias in the contributed data based on the analysis of six CRMs and SRMs (BCR CRM 141, BCR CRM 142, BCR CRM 143, NIST SRM 2709, NIST SRM 2710 and NIST SRM 2711).Sampling Protocol On the basis of these previous site investigations, a field of approximately 10 000 m2 in area was selected for study. This area was known to include an area containing the highest concentration of contaminants (contamination ‘hot spot’) as well as lower concentrations and was, therefore, judged to be both suitable for evaluating the performance of P-XRF against ICP-AES as well as testing the capabilities of P-XRF for ‘hot spot’ location.The field was first marked out with a 30 3 30 m grid using a 30 m triangulation tape (Fig. 3). Three sets of measurements were then undertaken. (i) An orienteering survey at the nodes of the 30 m grid. (ii) On the basis of these measurements a second, more dense 10 3 10 m grid was investigated on the same day, focusing on two areas identified as potential ‘hot spots’ of contamination on the basis of the first set of measurements.(iii) Finally, measurements were taken on two 1 m2 areas to characterise the small scale lateral variation of the metals in soil and to elucidate further the problems of sample heterogeneity. These last two areas were selected at both a ‘hot spot’ and a lower concentration area at the field. Fig. 1 Schematic representation of P-XRF instrumentation for in situ soil analysis. Fig. 2 Sample locations and Pb concentration in soil from previous study of the contaminated field in Derbyshire. 744 Analyst, August 1997, Vol. 122At each location selected for analysis, turf was removed to a depth of about 30 mm over an area of approximately 150 3 150 mm using a spade. The removal of the turf was necessary to expose unvegetated topsoil and to achieve a reasonably flat surface for P-XRF analysis, keeping in mind that the XRF signal for Pb originates from the top 1.5 mm surface layer. Duplicate measurements were taken at each sampling point at a distance of 2 m from the initial sampling location, so that an estimate could be made of measurement precision (including sampling error).This distance was selected to represent the location error of the sampling locations on the field given the surveying technology employed (30 m tape, triangulated). Auger samples 0–150 mm deep and 25 mm in diameter were also taken in each of the measuring sites for laboratory analysis by XRF and ICP-AES.Analysis Protocol Relatively short count times were selected for the P-XRF instrument to maximise the rate at which sites could be measured rather than to optimise detection limits. Selected count times were, therefore, 50 s for the 109Cd source, 20 s for 55Fe and 5 s for 241Am. The relatively high detection limit for Pb obtained under these conditions (calculated to be about 39 mg g21 ) did not affect the results significantly because of the high concentration of the contaminant in the soil at this site.The time required for duplicate P-XRF measurements at each of the sampling locations was approximately 3–5 min, allowing 24 sampling locations to be covered in half a working day. The samples collected for laboratory analysis were dried at 65 °C (to prevent excessive ‘caking’ of soil that results if 100 °C is used), disaggregated in a pestle and mortar and the < 2 mm fraction was ground (to < 100 mm) using an agate ring grinder.Analytical test portions of these samples were analysed as loose powders by laboratory-based P-XRF, using a sample cup fitted with a 6 mm thick polyester film. The measurements were taken for 100, 40 and 20 s using 109Cd, 55Fe and 241Am radioisotope sources, respectively, simulating the conditions in a field laboratory and improving on the measuring times used in situ. The same samples were analysed using ICP-AES after treatment with a mixture of nitric and perchloric acids to solubilize the analytes, which were presented for analysis in dilute hydrochloric acid solutions.1 For the analytical quality control, four reference materials (HRM 1, HRM 2, NIST SRM 2710 and HRM 31) in the form of compressed pellets were run periodically between the measurements in the field.In the laboratory-based XRF analysis, five reference materials were used (HRM 1, HRM 2, HRM 31, GXR-2 and NIST SRM 2710) and for the ICP-AES analysis seven reference materials (BCR CRM 141, BCR CRM 142, BCR CRM 143, NIST SRM 2709, NIST SRM 2710, NIST SRM 2711 and HRM 31) were employed.HRMs are house reference materials prepared at Imperial College for measuring the analytical bias. They are less expensive than the CRMs and their accepted value has been determined against existing CRMs. In particular, HRM 31 has a matrix composition that matches the sample composition of this particular sampling target. It was prepared using a composite material taken previously from the same field.The concentrations of both the analytes and the matrix elements in this HRM were in very close agreement with those of the samples. Duplicate analytical measurements were undertaken on each of the sampling duplicates in the laboratory, for ten of the sampling locations, in order to provide estimates of both the sampling and analytical precision. Results Measurement and Sampling Variance The first evaluation of performance was to estimate the proportion of the total variance of elemental concentration in the three data sets that could be apportioned to sampling and analysis.Pb data from the three sets of measurements (field measurements by P-XRF, laboratory measurements by P-XRF and laboratory measurements by ICP-AES) were evaluated by an analysis of variance (ANOVA) routine and judged against criteria proposed by Ramsey et al.13 These criteria, developed to ensure the appropriate interpretation particularly of environmental data, propose that the sampling variance plus analytical variance (together called the measurement variance) should not exceed 20% of the total variance of a data set. The balance, ‘geochemical’ variance, contains information about the geochemical or distribution behaviour of an analyte which can only be interpreted with confidence if this measurement variance threshold is not exceeded.Accordingly, the proportion of the total variance of Pb measurements contributed by the processes of sampling and analysis for all three techniques is shown in Fig. 4. The proposed limit is not exceeded for any of the sets of data. For the field-based P-XRF measurements, it was not possible to separate the analytical and sampling variance components of the measurement variance [Fig. 4(a)], but it is clear that sampling is the dominant contributor for both laboratory-based P-XRF and ICP-AES [Fig. 4(b) and (c)]. The similarity between the percentage of measurement variance for both the field- and laboratory-based P-XRF also suggests that this conclusion would be true for the field-based P-XRF measurements. The uncertainty on the measurements of the field-based P-XRF was also calculated as 200 Smeas/øx , where Smeas is the estimated measurement variance by robust ANOVA and øx the robust mean of the duplicate measurements.The value of this uncertainty was ±55% of the mean value. Bias Bias was evaluated from NIST SRM 2710, which was analysed for all contributing measurement techniques.This reference material was selected because the matrix and analyte concentrations match approximately the samples under consideration. Results for Pb are listed in Table 1. For Pb, the bias of 25.5% for ICP-AES probably originates in under-correction of matrix effects. For P-XRF, the bias of 211% for both field and laboratory measurements may reflect a mis-match between the Fig. 3 Sample locations for the in situ P-XRF measurements and collection of auger samples, showing the three sets of measurements: the initial 30 m grid, the 10 3 10 m grid and the two 1 m2 areas.Analyst, August 1997, Vol. 122 7450 0 5000 2000 6000 8000 10000 12000 4000 10000 15000 20000 25000 Pb/mg g–1 ICP-AES data Pb/mg g–1 P-XRF-field data y = 0.4271 x + 556.14 St.Err. on slope: 0.06 composition and degree of compaction of this reference material and that of a ‘typical’ soil used to determine the fundamental parameter coefficients required for the matrix correction procedure.The fact that the reference material used was pelletized, but the samples analysed in laboratory-based PXRF were not, could have been a potential source of bias. However, when powdered samples were analysed by both laboratory-based P-XRF and ICP-AES no significant bias was detected. When evaluating these data sets, it must be stated that there are some limitations in the use of a dry, finely ground, pelleted reference material to estimate the bias of field measurements by P-XRF made on moist, uncompacted and unground soils with a flat but irregular surface.Nominally, however, these levels of bias are probably acceptable and ‘fit-for-purpose’ in the context of this work, but further investigations of the origin of this bias are presented below. Bias can also be estimated by comparing the three sets of measurements against each other (i.e., relative bias). A direct comparison by simple linear regression of the measurements for Pb by field-based P-XRF against those by ICP-AES show a rotational ‘bias’ of 257 ± 6% (1s) in the former (Fig. 5). However, this ‘bias’ includes a measure of sampling bias as well as purely analytical bias, because the form and nature of the samples used in the comparison were not identical. The fieldbased P-XRF measurements are representative of the soil surface composition of Pb to a maximum depth of about 1.5 mm, whereas the ICP-AES results represent the homogenised 0–150 mm auger sample.Furthermore, the field-based P-XRF measurements were made on wet, uncompacted soils, whereas the ICP-AES measurements were made on dried, sieved and homogenised powders. A bias between the two techniques is to be expected, therefore, but to understand the capabilities and limitations of the P-XRF technique, it is relevant to investigate the components of this bias. When the bias for Pb between field-based P-XRF and laboratory-based P-XRF is compared, a similar rotational ‘bias’ is apparent, namely 252 ± 7% (1s), compared with 57 ± 6% for the field-based P-XRF versus laboratory-based ICP comparison.This suggests that the bias is not produced by a difference in the performance of the analytical technique, which is identical here, but is due to one of these other factors of sampling and sample preparation. The use of simple least squares linear regression has been shown to have limitations in the detection of bias between analytical methods,14 but such limitations are considered not to have affected this broad conclusion. Origins of Bias One hypothesis for the apparent low ‘bias’ of the field-based PXRF measurements is that the Pb content of the top few millimetres of the soil is systematically lower than that in the Fig. 4 ANOVA pie charts for Pb in soil, showing the proportions of sampling, analytical and geochemical variances for all three analytical techniques. (a) P-XRF, field data; (b) P-XRF, laboratory data; (c) ICP-AES data.The 20% limit for the measurement variance is not exceeded for any of the analytical methods. Table 1 Comparison of results for the analysis of NIST SRM 2710 by the three methods, showing acceptable levels of analytical bias. Bias for ICPAES probably originates in under-correction of matrix effects. For P-XRF the bias may reflect a mis-match between the composition of the reference material and that of the soil used in the matrix correction procedure (n: number of analyses) NIST SRM 2710 Pb Accepted value/mg g21 5532 Measured by field-based P-XRF/mg g21 4935 (n = 10) Bias (%) 210.8 Measured by laboratory-based P-XRF/mg g21 4898 (n = 4) Bias (%) 211.5 Measured by ICP-AES/mg g21 5228 (n = 2) Bias (%) 25.5 Fig. 5 Simple linear regression of field-based P-XRF Pb measurements against ICP-AES Pb measurements showing a rotational bias of 257 ± 6% between the two analytical methods. 746 Analyst, August 1997, Vol. 122150 mm sample taken for both laboratory-based P-XRF and ICP-AES analysis. To test this hypothesis, a 62 mm core of topsoil was taken in an area of the field with high Pb concentration, and progressively cut to remove 11 slices of around 6 mm thickness. The core of soil was analysed in situ in the field by P-XRF before each slice was removed (Fig. 6). Each slice was also analysed as a dried powder by laboratory-based ICP-AES. The Pb profiles produced by this ‘sliced core experiment’ show little systematic spatial trend with depth of core (Fig. 7), and in particular, there is no particular depletion in the top few millimetres, as would be required to confirm the hypothesis. A second core showed a different vertical variation in Pb concentration, but again no significant depletion at the surface. It is concluded, therefore, that at the site investigated in this study, surface analysis by P-XRF gives results that are not biased due to vertical variation in Pb concentration in the soil.Two other potential causes for the low ‘bias’ of the fieldbased P-XRF measurements are the moisture content of the soil and the rough surface of the soil used for excitation. These factors were corrected to assess their relative contributions to the bias. The moisture content of each slice of the core experiment was determined and ranged from 6.83% to 19.09% m/m. When a correction was made for moisture to the fieldbased P-XRF measurements, an initial bias of 219.9% against ICP-AES determinations for this particular experiment was reduced to 27.2% (Fig. 8). In order to compensate for the roughness of the soil surface, a Rayleigh scatter correction15 was made. After this correction, the bias was effectively eliminated, giving a non-significant bias of +0.8% over the 19 slices of both cores. It would appear, therefore, that the cause of the apparent ‘bias’ of the field-based P-XRF measurements, at least for these two cores, is entirely due to the combined effects of soil moisture and surface roughness.In practice, the Rayleigh scatter correction can be applied routinely with minimal extra measurement time, but the determination of moisture correction will need to be either a retrospective correction, or require the deployment of a field probe. Mapping The possibility of mapping the spatial distribution of metal contamination across a site in real time is one of the significant advantages of using field-based P-XRF.A further strength of the technique is that ‘hot spots’ can be investigated in more detail during the same visit at the sampling target. Measurements from the initial 30 m grid indicated two ‘hot spots’ with Pb concentrations of over 8000 mg g21 (Fig. 9). To investigate these ‘hot spots’ further, two secondary 10 m grids were set up centred on the candidate ‘hot spots’ (Fig. 9). Interestingly, the existence of the eastern ‘hot spot’ was substantiated by a cluster of similarly high Pb concentrations. The western ‘hot spot’, however, was not substantiated, with most of that 10 m grid being lower than 6000 mg g21.Given the high sampling variance (i.e., within-location) found at this site (15% of total), it would be expected that occasional isolated high values will occur by chance, without being confirmed by high data from adjacent sites to substantiate a ‘hot spot’. The use of field-based P-XRF allows such ‘spurious hot spots’ to be eliminated at the time of the survey.Any bias in the measurements would not directly affect the investigation of this spatial coherence of the ‘hot spots’, but only affect their intensity and extent. Fitnessfor- purpose criteria should be considered in order to determine the significance of any bias in the measurements. In this instance, the estimated bias of 257% for Pb between the in situ P-XRF measurements and those made in the laboratory does not affect the spatial delineation of the ‘hot spots’ since the observed metal concentrations are well above the limits for contaminated land.In general, given the uncertainty of ±55% on Fig. 6 Schematic representation of the ‘core experiment’ set-up. Fig. 7 Pb concentration in soil as determined in the core by the two analytical techniques, showing little systematic spatial trend in the core and no depletion in the top few millimetres. Fig. 8 Comparison data for Pb showing the ratio of XRF to ICP-AES analyses for the raw data, XRF data corrected for moisture and XRF data after moisture and Rayleigh scatter corrections.The bias after the corrections is effectively reduced to a non-significant level of +0.8%. Analyst, August 1997, Vol. 122 747the field-based P-XRF measurements, the Pb concentration would have to be over 4000 mg g21 in order to be recognised as a ‘hot spot’ against the background of 2000 mg g21. Spatial mapping on the smallest scale (i.e., < 1 m) using field-based P-XRF was used to investigate the heterogeneity of Pb on what is called the ‘within-location’ scale, in the larger scale surveys. The square metre surrounding the eastern ‘hot spot’ was surveyed using 16 samples, four taken at each of the distances from the centre of 500, 250, 120 and 50 mm (Fig. 10). The pattern of Pb concentrations measured does not conform to a single smooth maximum. Although high concentrations ( > 14 000 mg g21 Pb) dominate the central area, there is also a similar high value at the eastern edge, and also a low value (8810 mg g21) close to the centre.If the variance of the Pb concentration is estimated from the four replicates at each distance, then there is no systematic change with distance (RSD = 28, 24, 32 and 18% for the four distances mentioned above). Furthermore, this variance within 1 m is not significantly different from that measured at the 10 or 30 m scales, when expressed in units relative to the mean value at the corresponding scale. Similar results were extracted from the second 1 m2 site located in the area with the lowest metal concentrations.Discussion The potential of P-XRF for the in situ investigation of metal contaminated land has been evaluated, showing the strengths and disadvantages of the methodology. The possibility of immediate estimates of metal contamination of soil by in situ measurement is the most important advantage of the method over previous studies which have used P-XRF with the soil removed to an on-site laboratory.Large amounts of data can be obtained in a timely and cost-efficient manner during a single visit to the sampling target. Detection limits were sufficiently low to measure elevated levels of contamination with sufficient precision for several metals of interest (e.g., Pb, Zn and potentially Cu, Ni, As, Ba) in the particular field under study. The fact that the contaminants were present in high concentrations enhanced the abilities of the method but, since the P-XRF detection limits are well below the regulatory limits for most metals on contaminated land, the methodology is considered to be very useful for such investigations.The rapid analysis time for a suite of several metals (e.g., 2 min) is another advantage. However, the measurement time may need to be extended for sampling targets where elemental concentrations are lower than in this study. Because of its ability for rapid analysis, P-XRF methodology facilitates the use of iterative sampling designs.For example, a follow-up survey can be implemented to investigate any apparent ‘hot spots’, during a single site visit. Satisfactory analytical quality was observed in the in situ analysis, giving good analytical precision (e.g., 1%). However, the overall measurement precision is determined by the degree of heterogeneity of the sample site, effectively increasing the uncertainty of the analytical measurements.Another important advantage of the P-XRF methodology is the potential for the study of the heterogeneity of the sampling target at any relevant scale. In fact, P-XRF provides a tool for the investigation of in situ heterogeneity of the metal concentration across the sampling target, down to the centimetre scale. Disadvantages of the method include the very small size of sample from which the XRF signal originates ( Å 1 cm3) which, although useful for mapping the heterogeneity, may give results that are not representative of the surface composition to within an acceptable measurement precision.However, at the site studied here, although the measurement variance by field-based P-XRF was high, it was still a sufficiently small fraction of the total variance to be acceptable for the purposes of this study. If the small depth of an analysis (1.5 mm) is not considered representative, there is the possibility of averaging measurements over a greater depth, as in the sliced core study described above, or indeed extending measurements over even greater depths by the analysis of drill cores.Bias in the estimates of Pb concentrations by field-based PXRF was quantified by comparison of the measurements with those from the laboratory analysis of the corresponding samples using ICP-AES and P-XRF. The problem of apparent bias was overcome in this instance by making corrections for soil moisture and surface roughness of the soil surface analysed.In general, recognition of the sources of bias is important for the evaluation and improvement of the P-XRF methodology. Fig. 9 Spatial mapping of Pb concentration determined by P-XRF, showing the iterative sampling design with initial 30 m grid followed by the 10 m grid centred on candidate ‘hot spots’ (Pb > 8000 mg g21). The eastern ‘hot spot’ is confirmed by a cluster of high values, whereas the western ‘hot spot’ is not confirmed.Fig. 10 Concentration of Pb in soil in the 1 m2 area over the contamination ‘hot spot’ on the eastern side of the field, showing no systematic trend in the highest values. 748 Analyst, August 1997, Vol. 122Detection limits are not sufficiently low to quantify some elements (e.g., Cd) at typical background concentrations in soils (e.g., 0.1–0.5 mg g21 16). Also, a slightly longer survey time was required on the site compared with that required for the simple taking of soil samples.The method gains credit, however, in the total time spent for analysis, since the time for sample preparation is minimised. Finally, disadvantages related to the logistics of the use of the method include the relatively high capital cost of equipment for occasional use. Also, the time and expense required for compliance with licensing and safety regulations necessary for the use of instrumentation containing radioactive excitation sources is a potential problem that will vary between different countries.When laboratory-based P-XRF is used with larger, dried and homogenised samples, many of these disadvantages disappear, but with the simultaneous loss of one of the main advantages of the technique: that of immediate measurement output and information on in situ variations in concentration whilst on the sampling target. Conclusions The performance of P-XRF using Spectrace TN 9000 instrumentation as a field sampling and analytical technique has been evaluated for the in situ assessment of contaminated land.The technique has proved to be a useful and fit-for-purpose, powerful tool, capable of giving precise and very rapid analytical results for the determination of Pb concentrations in soil. The method was compared with both ICP-AES analysis of solutions derived from the soil samples and laboratory analysis of powdered samples using XRF, as established methods for the assessment of contaminated land.A rotational bias of 257% was calculated between the in situ field-based P-XRF and the ICP-AES measurements for Pb. The magnitude of bias between the field-based P-XRF measurements and the laboratory-based P-XRF results was estimated to be 252% for the same element. The origin of this bias was shown to be due largely to factors related to the nature of the sample. Specifically, the soil moisture and the surface roughness of the samples are the two factors which, without correction, affected the trueness of the results. When a correction was applied for these factors at two locations within the field, the bias for Pb measurements was effectively reduced to a statistically insignificant level.Field-based P-XRF measurements were capable of rapidly locating contamination ‘hot spots’ in the field. Furthermore, the detected contamination ‘hot spots’ on the initial 30 m grid could be investigated during the same working day by setting up a more dense sampling grid around the areas with the highest metal concentrations.Bias in uncorrected field-based P-XRF results did not affect the spatial coherence of the ‘hot spots’ but only their intensity and extent. Consequently, the effect of the bias, if uncorrected, would be vital in evaluating the extent of metal contamination if concentrations were close to a regulatory limit. An important characteristic of the field-based P-XRF methodology is the ability to investigate the small scale variability of metal content in soil.This is particularly useful in sampling targets showing a high degree of heterogeneity with respect to the metal distributions. The degree of heterogeneity can easily be quantified by estimating the variability in different spatial scales within the field. The limitations of the P-XRF methodology at this particular sampling target include the small effective sample size ( Å 1 cm3) and the small depth of analysis that might result in poor representativity, the bias on the estimated concentration values and the relatively high cost of the equipment.The first two limitations can be overcome by taking measurements on a finer grid so that variability can be observed at a smaller scale, and by excavation to extend analytical data to depth. Bias in comparison with laboratory results was corrected by accounting for the moisture content and applying a correction for surface roughness. Further work is needed for the improvement of the performance of the method.The use of appropriate reference materials could be useful for the elimination of bias of the in situ measurements. However, it would be difficult to maintain materials simulating the soil in its natural state, i.e., containing sufficient moisture and with a matching grain size distribution. Comparisons of the concentration estimates with those from other established analytical methods are very important in the initial stages of development of an in situ analytical method.Finally, good planning of the survey is needed before using PXRF for in situ field-based analysis so as to be able to gain as much information as possible on estimates of both concentration values and uncertainties, to permit a realistic interpretation of the extent of contamination at the site. The authors are very grateful to Peter Webb and Olwen Williams-Thorpe for field assistance with some of the P-XRF measurements. One of the authors (A.A.) is indebted to the Greek Scholarship Foundation for their support of this research.References 1 Thompson, M., and Walsh, J. N., A Handbook of Inductively Coupled Plasma Spectrometry, Blackie, Glasgow, 1989. 2 Ramsey, M. H., Argyraki, A., and Thompson, M., Analyst, 1995, 120, 1353. 3 Potts, P. J, Webb, P. C., Williams-Thorpe, O., and Kilworth, R., Analyst, 1995, 120, 1273. 4 Puls, R. W., Clark, D. A., Carlson, C., and Vardy, J., Ground Water Monit. Remed., 1994, 14, 111. 5 Kuharic, C. A., Cole, W. H., Singh, A. K., and Gonzales, D., EPA Report, EPA/600/R-93/073, US Environmental Protection Agency, Washington, DC, 1993. 6 Swift, R. P., Spectroscopy, 1995, 10, 31. 7 Vincent, H. A., and Boyer, D. M., ASTM Special Technical Publication, STP 1226, American Society for Testing and Materials, Philadelphia, 1995, p. 215. 8 Kuharic, C. A., and Cole, W. H., Adv. X-Ray Anal., 1995, 38, 725. 9 Pyle, S. M., Nocerino, J. M., Deming, S. N., Palasota, J.A., Palasota, J. M., Miller, E. L., Hillman, D. C., Kuharic, C. A., Cole, W. H., Watson, M. A., and Nichols, K. D., Environ. Sci. Technol., 1995, 30, 204. 10 Argyraki, A., Ramsey, M. H., and Thompson, M., Analyst, 1995, 120, 2799. 11 Ramsey, M. H., Argyraki, A., and Thompson, M., Analyst, 1995, 120, 2309. 12 Maskall, J., and Thornton, I., Land Contam. Reclam., 1993, 1, 92. 13 Ramsey, M. H., Thompson, M., and Hale, M. J., Geochem. Explor., 1992, 44, 23. 14 Ripley, B. D., and Thompson, M., Analyst, 1987, 112, 377. 15 Potts, P. J., A Handbook of Silicate Rock Analysis, Blackie, Glasgow, 1987, p. 252. 16 Rose, A. W., Hawkes, H. E., and Webb, J. S., Geochemistry in Mineral Exploration, Academic Press, London, 1979. Paper 7/00746I Received February 3, 1997 Accepted May 8, 1997 Analyst, August 1997, Vol. 122 749 Evaluation of Portable X-ray Fluorescence Instrumentation for in situMeasurements of Lead on Contaminated Land Ariadni Argyrakia, Michael H.Ramseya and Philip J. Pottsb a Department of Geology, Imperial College, London, UK SW7 2BP b Department of Earth Science, The Open University, Walton Hall, Milton Keynes, UK MK7 6AA The performance of the Spectrace TN 9000 portable X-ray fluorescence (P-XRF) instrument for in situ sampling and analysis of contaminated soil was evaluated. The method was compared with laboratory analysis of the samples using ICP-AES and XRF as established methods for the assessment of contaminated land.The trueness of the field-based P-XRF results was affected by the soil moisture and the surface roughness of the in situ samples, after the correction of which, no bias was observed between the analytical results of the comparative methods. Relatively large measurement uncertainty (±55% for Pb) was caused by the small sample mass analysed and the small scale heterogeneity of the sample. This uncertainty was quantified using duplicate measurements and does not impair the delineation of ‘hot spots’ of contamination as it contributes less than 20% to the total variance. General advantages and limitations of the P-XRF methodology for the investigation of contaminated land were assessed and suggestions are made for the optimisation of the methodology.Keywords: Portable X-ray fluorescence instrumentation; in situ measurement; lead; contaminated land In situ measurements of soils at contaminated sites potentially have the advantages of giving both a rapid assessment of the concentration of the contaminant and also first hand information on its spatial distribution and the degree of heterogeneity in an undisturbed position.The disadvantage of in situ measurement is often the high degree of uncertainty associated with the measurement, caused by factors such as the heterogeneity. If, however, this uncertainty can be quantified realistically, then these measurements can be fit for many purposes of environmental interpretation. Conventional schemes of analysis can usually be divided into two basic stages: (i) the collection of samples in the field, and (ii) the preparation and analysis of these samples in the laboratory, using a variety of analytical methods.The selection of the most appropriate sample preparation and analysis method depends on the sampling medium, the scope of the investigation and the availability of facilities. One well established procedure for the analysis of contaminated soils is to collect samples of top soil (e.g., 0–150 mm) using a hand auger, sieve and collect the < 2 mm fraction, grind to a fine powder and then use an acid extraction procedure, followed by ICP-AES to determine selected analytes. 1 However, it is important to appreciate that the estimated values of contaminant concentration from such investigations may vary due to sampling errors associated with the particular sampling design used in the field.2 X-ray fluorescence (XRF) is also a well established methodology for the assessment of contaminated land.Samples may be dried and then analysed as loose powders or alternatively prepared as compressed powder pellets. Ease of sample preparation as well as analytical trueness and high precision are advantages of this methodology.3 Field-based portable XRF (P-XRF) instrumentation offers potential advantages over other laboratory techniques. By undertaking analyses of contaminated soil in situ, and thereby avoiding the necessity of removing samples, P-XRF has the potential of giving both a rapid assessment of the concentration of the contaminant and also immediate information on its spatial distribution and degree of heterogeneity without disturbing the location.One disadvantage of in situ measurements is the high degree of uncertainty that may be associated with the measurement, caused by sample heterogeneity effects. If, however, this uncertainty can be quantified realistically, then these measurements may be fit for many purposes of environmental interpretation.As noted for in situ measurements in general, the potential problem of high measurement uncertainty can be addressed by its realistic quantification and recognition in environmental interpretation. Field-based P-XRF instruments are becoming increasingly important in assessments of contaminated land. Their performance for the investigation of contaminated land has recently been evaluated by several workers but after removal of the soils to a field laboratory.Specifically, P-XRF devices have been used for the immediate delineation of Cr source contamination ‘hot spots’4 and the spatial distribution of Pb concentration in residential soils.5 P-XRF methodology has also been used to provide data for remedial activities at sites contaminated with Pb and As6 and the determination of Pb in urban soil and dust samples.7 Comparisons of the estimates of elemental concentrations made using P-XRF with estimates from other analytical methods such as ICP-AES and AAS have revealed data of acceptable quality when the portable methodology is used properly.However, a slight bias between the XRF results and those of the other methods has been reported.4–9 A significant problem of the P-XRF methodology, addressed in most of the above studies, is the use of appropriate calibration samples. This paper uses in situ rather than laboratory-based measurements using a Spectrace TN 9000 P-XRF instrument for the determination of Pb in soil.The main aim of this work was to evaluate the performance of a field-based P-XRF instrument for the determination of Pb in contaminated soil in comparison with an established method of analysis, i.e., hand augering followed by laboratory-based ICP-AES. To facilitate the evaluation, the field chosen for this assessment had previously been studied in some detail as part of an on-going study of environmental sampling procedures.2,10,11 The use of P-XRF for spatial mapping, for locating contamination ‘hot-spots’ and for flexibility in the design of sampling protocols was also assessed.Results are used to demonstrate both the advantages and limitations of the technique. The rational optimisation of the sampling and analytical protocol using P-XRF for the investigation of contaminated land, both in situ and in the laboratory, was another objective of this work.The last objective was to identify needs for further development work, which will enable the ultimate limits of performance of P-XRF to be approached. Analyst, August 1997, Vol. 122 (743–749) 743Experimental P-XRF Field measurements were undertaken with a Tracor Northern Spectrace TN 9000 P-XRF instrument on hire from Thermo Unicam (Cambridge, UK). The instrument consisted of a handheld analyser unit and a portable spectrum acquisition and data processing unit (Fig. 1).The analyser unit incorporated three radioactive isotope excitation sources: 55Fe, 109Cd and 241Am and a mercury(ii) iodide X-ray detector. To undertake an analysis in the field, the analyser unit was placed against the sample surface which was excited with each of the three sources in turn. For the analysis of samples in the laboratory (e.g., powdered soil), the analyser unit was mounted in a laboratory stand with the analyser window pointing vertically and covered by a sample lid, the position of which (open or closed) was interlocked to the operation of the instrument. Samples were then conveniently contained in conventional XRF sample cups, placed in position over the window, and analysed once the lid was closed.Whatever the form of sample presentation, spectra were accumulated in the spectrum acquisition and data processing unit and at the end of the preset count time, fluorescence intensities were measured using a spectrum deconvolution procedure based on region-of-interest integration.Data were then corrected for matrix effects and quantified using a fundamental parameter procedure. Quantitative results were displayed for immediate appraisal and stored for subsequent downloading to an external microcomputer via an RS232 interface. The performance of this XRF technique in the laboratory has been described by Potts et al.3 These earlier studies were undertaken on a range of silicate rock reference materials prepared as compressed powder pellet samples, and were designed to demonstrate the basic performance characteristics of the technique.Using a count time (live time) of 200 s per source, repeatability precision was found to be in the range 0.45–1.8% for the major elements and 2–5% for trace elements. A high degree of linearity was also achieved in the relationship between analysed and expected values for the 70 international reference materials investigated. The detection limit for Pb was found to be 39 mg g21, representative of a 200 s count time.ICP-AES Field-based P-XRF results were compared with an established ICP-AES technique. Samples representing the top 0–150 mm surface of the soil were removed with a hand auger and returned to the laboratory for analysis. There, samples were dried, ground in a pestle and mortar and analysed by ICP-AES after acid extraction. Full details of the procedure can be found elsewhere.1 The Site The site selected for study was a fallow field at Bolehill near Wirksworth, Derbyshire, UK.This site was used in Medieval times for Pb smelting. Recent interest arose, therefore, in the possibility of using the presence of high levels of Pb contamination known to be present in the soil to model the mobility of this contaminant in the environment.12 Previous investigations by auger sampling and ICP-AES analysis had revealed mean Pb concentrations of 6229 mg g21 in the top 0–150 mm soil layer over the site.2 The spatial distribution of the element was also well known from both this pilot study (Fig. 2), and also a proficiency testing and collaborative sampling trial which had been based on part of the same field.10,11 Useful information about the sampling variance was available from these trials as well as analytical bias in the contributed data based on the analysis of six CRMs and SRMs (BCR CRM 141, BCR CRM 142, BCR CRM 143, NIST SRM 2709, NIST SRM 2710 and NIST SRM 2711). Sampling Protocol On the basis of these previous site investigations, a field of approximately 10 000 m2 in area was selected for study.This area was known to include an area containing the highest concentration of contaminants (contamination ‘hot spot’) as well as lower concentrations and was, therefore, judged to be both suitable for evaluating the performance of P-XRF against ICP-AES as well as testing the capabilities of P-XRF for ‘hot spot’ location. The field was first marked out with a 30 3 30 m grid using a 30 m triangulation tape (Fig. 3). Three sets of measurements were then undertaken. (i) An orienteering survey at the nodes of the 30 m grid. (ii) On the basis of these measurements a second, more dense 10 3 10 m grid was investigated on the same day, focusing on two areas identified as potential ‘hot spots’ of contamination on the basis of the first set of measurements. (iii) Finally, measurements were taken on two 1 m2 areas to characterise the small scale lateral variation of the metals in soil and to elucidate further the problems of sample heterogeneity.These last two areas were selected at both a ‘hot spot’ and a lower concentration area at the field. Fig. 1 Schematic representation of P-XRF instrumentation for in situ soil analysis. Fig. 2 Sample locations and Pb concentration in soil from previous study of the contaminated field in Derbyshire. 744 Analyst, August 1997, Vol. 122At each location selected for analysis, turf was removed to a depth of about 30 mm over an area of approximately 150 3 150 mm using a spade.The removal of the turf was necessary to expose unvegetated topsoil and to achieve a reasonably flat surface for P-XRF analysis, keeping in mind that the XRF signal for Pb originates from the top 1.5 mm surface layer. Duplicate measurements were taken at each sampling point at a distance of 2 m from the initial sampling location, so that an estimate could be made of measurement precision (including sampling error).This distance was selected to represent the location error of the sampling locations on the field given the surveying technology employed (30 m tape, triangulated). Auger samples 0–150 mm deep and 25 mm in diameter were also taken in each of the measuring sites for laboratory analysis by XRF and ICP-AES. Analysis Protocol Relatively short count times were selected for the P-XRF instrument to maximise the rate at which sites could be measured rather than to optimise detection limits.Selected count times were, therefore, 50 s for the 109Cd source, 20 s for 55Fe and 5 s for 241Am. The relatively high detection limit for Pb obtained under these conditions (calculated to be about 39 mg g21 ) did not affect the results significantly because of the high concentration of the contaminant in the soil at this site. The time required for duplicate P-XRF measurements at each of the sampling locations was approximately 3–5 min, allowing 24 sampling locations to be covered in half a working day.The samples collected for laboratory analysis were dried at 65 °C (to prevent excessive ‘caking’ of soil that results if 100 °C is used), disaggregated in a pestle and mortar and the < 2 mm fraction was ground (to < 100 mm) using an agate ring grinder. Analytical test portions of these samples were analysed as loose powders by laboratory-based P-XRF, using a sample cup fitted with a 6 mm thick polyester film.The measurements were taken for 100, 40 and 20 s using 109Cd, 55Fe and 241Am radioisotope sources, respectively, simulating the conditions in a field laboratory and improving on the measuring times used in situ. The same samples were analysed using ICP-AES after treatment with a mixture of nitric and perchloric acids to solubilize the analytes, which were presented for analysis in dilute hydrochloric acid solutions.1 For the analytical quality control, four reference materials (HRM 1, HRM 2, NIST SRM 2710 and HRM 31) in the form of compressed pellets were run periodically between the measurements in the field.In the laboratory-based XRF analysis, five reference materials were used (HRM 1, HRM 2, HRM 31, GXR-2 and NIST SRM 2710) and for the ICP-AES analysis seven reference materials (BCR CRM 141, BCR CRM 142, BCR CRM 143, NIST SRM 2709, NIST SRM 2710, NIST SRM 2711 and HRM 31) were employed. HRMs are house reference materials prepared at Imperial College for measuring the analytical bias.They are less expensive than the CRMs and their accepted value has been determined against existing CRMs. In particular, HRM 31 has a matrix composition that matches the sample composition of this particular sampling target. It was prepared using a composite material taken previously from the same field. The concentrations of both the analytes and the matrix elements in this HRM were in very close agreement with those of the samples.Duplicate analytical measurements were undertaken on each of the sampling duplicates in the laboratory, for ten of the sampling locations, in order to provide estimates of both the sampling and analytical precision. Results Measurement and Sampling Variance The first evaluation of performance was to estimate the proportion of the total variance of elemental concentration in the three data sets that could be apportioned to sampling and analysis. Pb data from the three sets of measurements (field measurements by P-XRF, laboratory measurements by P-XRF and laboratory measurements by ICP-AES) were evaluated by an analysis of variance (ANOVA) routine and judged against criteria proposed by Ramsey et al.13 These criteria, developed to ensure the appropriate interpretation particularly of environmental data, propose that the sampling variance plus analytical variance (together called the measurement variance) should not exceed 20% of the total variance of a data set.The balance, ‘geochemical’ variance, contains information about the geochemical or distribution behaviour of an analyte which can only be interpreted with confidence if this measurement variance threshold is not exceeded. Accordingly, the proportion of the total variance of Pb measurements contributed by the processes of sampling and analysis for all three techniques is shown in Fig. 4. The proposed limit is not exceeded for any of the sets of data.For the field-based P-XRF measurements, it was not possible to separate the analytical and sampling variance components of the measurement variance [Fig. 4(a)], but it is clear that sampling is the dominant contributor for both laboratory-based P-XRF and ICP-AES [Fig. 4(b) and (c)]. The similarity between the percentage of measurement variance for both the field- and laboratory-based P-XRF also suggests that this conclusion would be true for the field-based P-XRF measurements.The uncertainty on the measurements of the field-based P-XRF was also calculated as 200 Smeas/øx , where Smeas is the estimated measurement variance by robust ANOVA and øx the robust mean of the duplicate measurements. The value of this uncertainty was ±55% of the mean value. Bias Bias was evaluated from NIST SRM 2710, which was analysed for all contributing measurement techniques. This reference material was selected because the matrix and analyte concentrations match approximately the samples under consideration. Results for Pb are listed in Table 1.For Pb, the bias of 25.5% for ICP-AES probably originates in under-correction of matrix effects. For P-XRF, the bias of 211% for both field and laboratory measurements may reflect a mis-match between the Fig. 3 Sample locations for the in situ P-XRF measurements and collection of auger samples, showing the three sets of measurements: the initial 30 m grid, the 10 3 10 m grid and the two 1 m2 areas.Analyst, August 1997, Vol. 122 7450 0 5000 2000 6000 8000 10000 12000 4000 10000 15000 20000 25000 Pb/mg g–1 ICP-AES data Pb/mg g–1 P-XRF-field data y = 0.4271 x + 556.14 St.Err. on slope: 0.06 composition and degree of compaction of this reference material and that of a ‘typical’ soil used to determine the fundamental parameter coefficients required for the matrix correction procedure. The fact that the reference material used was pelletized, but the samples analysed in laboratory-based PXRF were not, could have been a potential source of bias.However, when powdered samples were analysed by both laboratory-based P-XRF and ICP-AES no significant bias was detected. When evaluating these data sets, it must be stated that there are some limitations in the use of a dry, finely ground, pelleted reference material to estimate the bias of field measurements by P-XRF made on moist, uncompacted and unground soils with a flat but irregular surface.Nominally, however, these levels of bias are probably acceptable and ‘fit-for-purpose’ in the context of this work, but further investigations of the origin of this bias are presented below. Bias can also be estimated by comparing the three sets of measurements against each other (i.e., relative bias). A direct comparison by simple linear regression of the measurements for Pb by field-based P-XRF against those by ICP-AES show a rotational ‘bias’ of 257 ± 6% (1s) in the former (Fig. 5). However, this ‘bias’ includes a measure of sampling bias as well as purely analytical bias, because the form and nature of the samples used in the comparison were not identical. The fieldbased P-XRF measurements are representative of the soil surface composition of Pb to a maximum depth of about 1.5 mm, whereas the ICP-AES results represent the homogenised 0–150 mm auger sample. Furthermore, the field-based P-XRF measurements were made on wet, uncompacted soils, whereas the ICP-AES measurements were made on dried, sieved and homogenised powders.A bias between the two techniques is to be expected, therefore, but to understand the capabilities and limitations of the P-XRF technique, it is relevant to investigate the components of this bias. When the bias for Pb between field-based P-XRF and laboratory-based P-XRF is compared, a similar rotational ‘bias’ is apparent, namely 252 ± 7% (1s), compared with 57 ± 6% for the field-based P-XRF versus laboratory-based ICP comparison. This suggests that the bias is not produced by a difference in the performance of the analytical technique, which is identical here, but is due to one of these other factors of sampling and sample preparation.The use of simple least squares linear regression has been shown to have limitations in the detection of bias between analytical methods,14 but such limitations are considered not to have affected this broad conclusion. Origins of Bias One hypothesis for the apparent low ‘bias’ of the field-based PXRF measurements is that the Pb content of the top few millimetres of the soil is systematically lower than that in the Fig. 4 ANOVA pie charts for Pb in soil, showing the proportions of sampling, analytical and geochemical variances for all three analytical techniques. (a) P-XRF, field data; (b) P-XRF, laboratory data; (c) ICP-AES data. The 20% limit for the measurement variance is not exceeded for any of the analytical methods.Table 1 Comparison of results for the analysis of NIST SRM 2710 by the three methods, showing acceptable levels of analytical bias. Bias for ICPAES probably originates in under-correction of matrix effects. For P-XRF the bias may reflect a mis-match between the composition of the reference material and that of the soil used in the matrix correction procedure (n: number of analyses) NIST SRM 2710 Pb Accepted value/mg g21 5532 Measured by field-based P-XRF/mg g21 4935 (n = 10) Bias (%) 210.8 Measured by laboratory-based P-XRF/mg g21 4898 (n = 4) Bias (%) 211.5 Measured by ICP-AES/mg g21 5228 (n = 2) Bias (%) 25.5 Fig. 5 Simple linear regression of field-based P-XRF Pb measurements against ICP-AES Pb measurements showing a rotational bias of 257 ± 6% between the two analytical methods. 746 Analyst, August 1997, Vol. 122150 mm sample taken for both laboratory-based P-XRF and ICP-AES analysis. To test this hypothesis, a 62 mm core of topsoil was taken in an area of the field with high Pb concentration, and progressively cut to remove 11 slices of around 6 mm thickness. The core of soil was analysed in situ in the field by P-XRF before each slice was removed (Fig. 6). Each slice was also analysed as a dried powder by laboratory-based ICP-AES. The Pb profiles produced by this ‘sliced core experiment’ show little systematic spatial trend with depth of core (Fig. 7), and in particular, there is no particular depletion in the top few millimetres, as would be required to confirm the hypothesis.A second core showed a different vertical variation in Pb concentration, but again no significant depletion at the surface. It is concluded, therefore, that at the site investigated in this study, surface analysis by P-XRF gives results that are not biased due to vertical variation in Pb concentration in the soil. Two other potential causes for the low ‘bias’ of the fieldbased P-XRF measurements are the moisture content of the soil and the rough surface of the soil used for excitation.These factors were corrected to assess their relative contributions to the bias. The moisture content of each slice of the core experiment was determined and ranged from 6.83% to 19.09% m/m. When a correction was made for moisture to the fieldbased P-XRF measurements, an initial bias of 219.9% against ICP-AES determinations for this particular experiment was reduced to 27.2% (Fig. 8). In order to compensate for the roughness of the soil surface, a Rayleigh scatter correction15 was made.After this correction, the bias was effectively eliminated, giving a non-significant bias of +0.8% over the 19 slices of both cores. It would appear, therefore, that the cause of the apparent ‘bias’ of the field-based P-XRF measurements, at least for these two cores, is entirely due to the combined effects of soil moisture and surface roughness. In practice, the Rayleigh scatter correction can be applied routinely with minimal extra measurement time, but the determination of moisture correction will need to be either a retrospective correction, or require the deployment of a field probe.Mapping The possibility of mapping the spatial distribution of metal contamination across a site in real time is one of the significant advantages of using field-based P-XRF. A further strength of the technique is that ‘hot spots’ can be investigated in more detail during the same visit at the sampling target.Measurements from the initial 30 m grid indicated two ‘hot spots’ with Pb concentrations of over 8000 mg g21 (Fig. 9). To investigate these ‘hot spots’ further, two secondary 10 m grids were set up centred on the candidate ‘hot spots’ (Fig. 9). Interestingly, the existence of the eastern ‘hot spot’ was substantiated by a cluster of similarly high Pb concentrations. The western ‘hot spot’, however, was not substantiated, with most of that 10 m grid being lower than 6000 mg g21. Given the high sampling variance (i.e., within-location) found at this site (15% of total), it would be expected that occasional isolated high values will occur by chance, without being confirmed by high data from adjacent sites to substantiate a ‘hot spot’.The use of field-based P-XRF allows such ‘spurious hot spots’ to be eliminated at the time of the survey. Any bias in the measurements would not directly affect the investigation of this spatial coherence of the ‘hot spots’, but only affect their intensity and extent.Fitnessfor- purpose criteria should be considered in order to determine the significance of any bias in the measurements. In this instance, the estimated bias of 257% for Pb between the in situ P-XRF measurements and those made in the laboratory does not affect the spatial delineation of the ‘hot spots’ since the observed metal concentrations are well above the limits for contaminated land.In general, given the uncertainty of ±55% on Fig. 6 Schematic representation of the ‘core experiment’ set-up. Fig. 7 Pb concentration in soil as determined in the core by the two analytical techniques, showing little systematic spatial trend in the core and no depletion in the top few millimetres. Fig. 8 Comparison data for Pb showing the ratio of XRF to ICP-AES analyses for the raw data, XRF data corrected for moisture and XRF data after moisture and Rayleigh scatter corrections. The bias after the corrections is effectively reduced to a non-significant level of +0.8%.Analyst, August 1997, Vol. 122 747the field-based P-XRF measurements, the Pb concentration would have to be over 4000 mg g21 in order to be recognised as a ‘hot spot’ against the background of 2000 mg g21. Spatial mapping on the smallest scale (i.e., < 1 m) using field-based P-XRF was used to investigate the heterogeneity of Pb on what is called the ‘within-location’ scale, in the larger scale surveys.The square metre surrounding the eastern ‘hot spot’ was surveyed using 16 samples, four taken at each of the distances from the centre of 500, 250, 120 and 50 mm (Fig. 10). The pattern of Pb concentrations measured does not conform to a single smooth maximum. Although high concentrations ( > 14 000 mg g21 Pb) dominate the central area, there is also a similar high value at the eastern edge, and also a low value (8810 mg g21) close to the centre.If the variance of the Pb concentration is estimated from the four replicates at each distance, then there is no systematic change with distance (RSD = 28, 24, 32 and 18% for the four distances mentioned above). Furthermore, this variance within 1 m is not significantly different from that measured at the 10 or 30 m scales, when expressed in units relative to the mean value at the corresponding scale. Similar results were extracted from the second 1 m2 site located in the area with the lowest metal concentrations. Discussion The potential of P-XRF for the in situ investigation of metal contaminated land has been evaluated, showing the strengths and disadvantages of the methodology.The possibility of immediate estimates of metal contamination of soil by in situ measurement is the most important advantage of the method over previous studies which have used P-XRF with the soil removed to an on-site laboratory. Large amounts of data can be obtained in a timely and cost-efficient manner during a single visit to the sampling target.Detection limits were sufficiently low to measure elevated levels of contamination with sufficient precision for several metals of interest (e.g., Pb, Zn and potentially Cu, Ni, As, Ba) in the particular field under study. The fact that the contaminants were present in high concentrations enhanced the abilities of the method but, since the P-XRF detection limits are well below the regulatory limits for most metals on contaminated land, the methodology is considered to be very useful for such investigations.The rapid analysis time for a suite of several metals (e.g., 2 min) is another advantage. However, the measurement time may need to be extended for sampling targets where elemental concentrations are lower than in this study. Because of its ability for rapid analysis, P-XRF methodology facilitates the use of iterative sampling designs. For example, a follow-up survey can be implemented to investigate any apparent ‘hot spots’, during a single site visit.Satisfactory analytical quality was observed in the in situ analysis, giving good analytical precision (e.g., 1%). However, the overall measurement precision is determined by the degree of heterogeneity of the sample site, effectively increasing the uncertainty of the analytical measurements. Another important advantage of the P-XRF methodology is the potential for the study of the heterogeneity of the sampling target at any relevant scale.In fact, P-XRF provides a tool for the investigation of in situ heterogeneity of the metal concentration across the sampling target, down to the centimetre scale. Disadvantages of the method include the very small size of sample from which the XRF signal originates ( Å 1 cm3) which, although useful for mapping the heterogeneity, may give results that are not representative of the surface composition to within an acceptable measurement precision.However, at the site studied here, although the measurement variance by field-based P-XRF was high, it was still a sufficiently small fraction of the total variance to be acceptable for the purposes of this study. If the small depth of an analysis (1.5 mm) is not considered representative, there is the possibility of averaging measurements over a greater depth, as in the sliced core study described above, or indeed extending measurements over even greater depths by the analysis of drill cores. Bias in the estimates of Pb concentrations by field-based PXRF was quantified by comparison of the measurements with those from the laboratory analysis of the corresponding samples using ICP-AES and P-XRF.The problem of apparent bias was overcome in this instance by making corrections for soil moisture and surface roughness of the soil surface analysed. In general, recognition of the sources of bias is important for the evaluation and improvement of the P-XRF methodology.Fig. 9 Spatial mapping of Pb concentration determined by P-XRF, showing the iterative sampling design with initial 30 m grid followed by the 10 m grid centred on candidate ‘hot spots’ (Pb > 8000 mg g21). The eastern ‘hot spot’ is confirmed by a cluster of high values, whereas the western ‘hot spot’ is not confirmed. Fig. 10 Concentration of Pb in soil in the 1 m2 area over the contamination ‘hot spot’ on the eastern side of the field, showing no systematic trend in the highest values. 748 Analyst, August 1997, Vol. 122Detection limits are not sufficiently low to quantify some elements (e.g., Cd) at typical background concentrations in soils (e.g., 0.1–0.5 mg g21 16). Also, a slightly longer survey time was required on the site compared with that required for the simple taking of soil samples. The method gains credit, however, in the total time spent for analysis, since the time for sample preparation is minimised.Finally, disadvantages related to the logistics of the use of the method include the relatively high capital cost of equipment for occasional use. Also, the time and expense required for compliance with licensing and safety regulations necessary for the use of instrumentation containing radioactive excitation sources is a potential problem that will vary between different countries. When laboratory-based P-XRF is used with larger, dried and homogenised samples, many of these disadvantages disappear, but with the simultaneous loss of one of the main advantages of the technique: that of immediate measurement output and information on in situ variations in concentration whilst on the sampling target.Conclusions The performance of P-XRF using Spectrace TN 9000 instrumentation as a field sampling and analytical technique has been evaluated for the in situ assessment of contaminated land. The technique has proved to be a useful and fit-for-purpose, powerful tool, capable of giving precise and very rapid analytical results for the determination of Pb concentrations in soil.The method was compared with both ICP-AES analysis of solutions derived from the soil samples and laboratory analysis of powdered samples using XRF, as established methods for the assessment of contaminated land. A rotational bias of 257% was calculated between the in situ field-based P-XRF and the ICP-AES measurements for Pb. The magnitude of bias between the field-based P-XRF measurements and the laboratory-based P-XRF results was estimated to be 252% for the same element.The origin of this bias was shown to be due largely to factors related to the nature of the sample. Specifically, the soil moisture and the surface roughness of the samples are the two factors which, without correction, affected the trueness of the results. When a correction was applied for these factors at two locations within the field, the bias for Pb measurements was effectively reduced to a statistically insignificant level.Field-based P-XRF measurements were capable of rapidly locating contamination ‘hot spots’ in the field. Furthermore, the detected contamination ‘hot spots’ on the initial 30 m grid could be investigated during the same working day by setting up a more dense sampling grid around the areas with the highest metal concentrations. Bias in uncorrected field-based P-XRF results did not affect the spatial coherence of the ‘hot spots’ but only their intensity and extent.Consequently, the effect of the bias, if uncorrected, would be vital in evaluating the extent of metal contamination if concentrations were close to a regulatory limit. An important characteristic of the field-based P-XRF methodology is the ability to investigate the small scale variability of metal content in soil. This is particularly useful in sampling targets showing a high degree of heterogeneity with respect to the metal distributions. The degree of heterogeneity can easily be quantified by estimating the variability in different spatial scales within the field. The limitations of the P-XRF methodology at this particular sampling target include the small effective sample size ( Å 1 cm3) and the small depth of analysis that might result in poor representativity, the bias on the estimated concentration values and the relatively high cost of the equipment. The first two limitations can be overcome by taking measurements on a finer grid so that variability can be observed at a smaller scale, and by excavation to extend analytical data to depth. Bias in comparison with laboratory results was corrected by accounting for the moisture content and applying a correction for surface roughness. Further work is needed for the improvement of the performance of the method. The use of appropriate reference materials could be useful for the elimination of bias of the in situ measurements. However, it would be difficult to maintain materials simulating the soil in its natural state, i.e., containing sufficient moisture and with a matching grain size distribution. Comparisons of the concentration estimates with those from other established analytical methods are very important in the initial stages of development of an in situ analytical method. Finally, good planning of the survey is needed before using PXRF for in situ field-based analysis so as to be able to gain as much information as possible on estimates of both concentration values and uncertainties, to permit a realistic interpretation of the extent of contamination at the site. The authors are very grateful to Peter Webb and Olwen Williams-Thorpe for field assistance with some of the P-XRF measurements. One of the authors (A.A.) is indebted to the Greek Scholarship Foundation for their support of this research. References 1 Thompson, M., and Walsh, J. N., A Handbook of Inductively Coupled Plasma Spectrometry, Blackie, Glasgow, 1989. 2 Ramsey, M. H., Argyraki, A., and Thompson, M., Analyst, 1995, 120, 1353. 3 Potts, P. J, Webb, P. C., Williams-Thorpe, O., and Kilworth, R., Analyst, 1995, 120, 1273. 4 Puls, R. W., Clark, D. A., Carlson, C., and Vardy, J., Ground Water Monit. Remed., 1994, 14, 111. 5 Kuharic, C. A., Cole, W. H., Singh, A. K., and Gonzales, D., EPA Report, EPA/600/R-93/073, US Environmental Protection Agency, Washington, DC, 1993. 6 Swift, R. P., Spectroscopy, 1995, 10, 31. 7 Vincent, H. A., and Boyer, D. M., ASTM Special Technical Publication, STP 1226, American Society for Testing and Materials, Philadelphia, 1995, p. 215. 8 Kuharic, C. A., and Cole, W. H., Adv. X-Ray Anal., 1995, 38, 725. 9 Pyle, S. M., Nocerino, J. M., Deming, S. N., Palasota, J. A., Palasota, J. M., Miller, E. L., Hillman, D. C., Kuharic, C. A., Cole, W. H., Watson, M. A., and Nichols, K. D., Environ. Sci. Technol., 1995, 30, 204. 10 Argyraki, A., Ramsey, M. H., and Thompson, M., Analyst, 1995, 120, 2799. 11 Ramsey, M. H., Argyraki, A., and Thompson, M., Analyst, 1995, 120, 2309. 12 Maskall, J., and Thornton, I., Land Contam. Reclam., 1993, 1, 92. 13 Ramsey, M. H., Thompson, M., and Hale, M. J., Geochem. Explor., 1992, 44, 23. 14 Ripley, B. D., and Thompson, M., Analyst, 1987, 112, 377. 15 Potts, P. J., A Handbook of Silicate Rock Analysis, Blackie, Glasgow, 1987, p. 252. 16 Rose, A. W., Hawkes, H. E., and Webb, J. S., Geochemistry in Mineral Exploration, Academic Press, London, 1979. Paper 7/00746I Received February 3, 1997 Accepted May 8, 1997 Analyst, August 1997, Vol. 122 749
ISSN:0003-2654
DOI:10.1039/a700746i
出版商:RSC
年代:1997
数据来源: RSC
|
5. |
Determination of Inorganic and Total Mercury in Biological Tissuesby Electrothermal Vaporization Inductively Coupled Plasma MassSpectrometry |
|
Analyst,
Volume 122,
Issue 8,
1997,
Page 751-754
Scott N. Willie,
Preview
|
|
摘要:
Determination of Inorganic and Total Mercury in Biological Tissues by Electrothermal Vaporization Inductively Coupled Plasma Mass Spectrometry Scott N. Willie*a, D. Conrad Gr�egoireb and Ralph E. Sturgeona a Institute for National Measurement Standards, National Research Council of Canada, Ottawa, Ontario, Canada K1A 0R9 b Geological Survey of Canada, 601 Booth Street, Ottawa, Ontario, Canada K1A 0E8 A rapid method for the determination of total and inorganic mercury in biological tissues is presented using electrothermal vaporization inductively coupled plasma mass spectrometry (ETV ICP–MS).Samples were solubilized using tetramethylammonium hydroxide. For the determination of total mercury sample aliquots (10 ml) are dried and vaporized into the plasma. For the determination of inorganic mercury, iodoacetic acid, sodium thiosulfate and acetic acid are added to the sample, cleaving the methylmercury from the tissue. Volatile methylmercury iodide is formed and removed from the ETV as the sample dries, leaving only inorganic mercury to be quantified.A limit of detection of 0.05 mg g21 in solid samples was obtained. National Research Council of Canada reference materials DORM-2 (dogfish muscle), DOLT-2 (dogfish liver) and TORT-2 (lobster hepatopancreas) were used to assess the accuracy of the method. Keywords: Mercury; inorganic mercury; biological tissues; electrothermal vaporization inductively coupled plasma mass spectrometry; tetramethylammonium hydroxide solubilization Mercury is a ubiquitous element in the environment resulting from both anthropogenic and natural geological activities.In the marine environment the distribution and speciation of Hg is of considerable interest as biomethylation1 and subsequently bioconcentration in the food chain2 can occur. As a result, mercury is predominantly present in marine biological samples as methylmercury. It is well known that organomercury compounds are more toxic than metallic or aquo Hg forms; consequently, considerable effort and progress have been made in the development of techniques capable of separating and identifying the various mercury species.3 Common methods for the determination of total mercury require sample digestion followed by chemical reduction and subsequent gas phase transfer of the element to a detector.Approaches for speciation analysis involve milder sample preparation procedures to preserve the integrity of any organic mercury.4 Frequently, these are variations of the classical West�o�o5 procedure involving solvent extraction, cleanup of the extract, isolation of the organomercurial as the corresponding halide derivative and detection by gas chromatography with electron capture detection (GC–ECD)6 or atomic emission detection (GC–AED).7 Methods using liquid chromatography (LC) also require sample extraction and cleanup steps.8,9 Alternatively, procedures utilizing derivatisation of organomercury via ethylating10 or butylating11 agents followed by isolation and detection have been reported.Generally, these procedures require less sample cleanup than GC or LC methods but separation via a chromatographic column or cryogenic trapping is necessary. Another method that does not require extensive sample extraction and cleanup utilizes headspace sampling. A cleaving agent is used to liberate the methylmercury from the biological tissue, whereupon it is subsequently converted to volatile methylmercury iodide, sampled, desorbed onto a column and detected using GC–MIP (microwave induced plasma).12,13 The determination of Hg by thermal vaporization from solid samples is an attractive approach that requires minimal sample preparation.14 This technique has been applied to environmental samples15 and rocks and sediments16 using quartz tube AAS detection.As well as total Hg, direct methods for the speciation of mercury in soils and sediments17 and biologicals18 have been reported.ETV ICP-MS is well-suited for application to thermal vaporization methodology as the temperature of the graphite tube can be accurately controlled, small samples can be easily injected and the sampling frequency is relatively high. In addition, the tolerance of ICP-MS to the solvent and matrix components in the sample is superior to the less powerful microwave He plasma system.19 The determination of total and inorganic mercury in biological tissues is reported here.Extraction or acid digestion of the samples is replaced by room temperature treatment of the tissue with tetramethylammonium hydroxide (TMAH). TMAH, an alkali tissue solublizer, has been utilized for the speciation analysis of tin and mercury and also for total element analysis using flame,20 graphite furance21,22 and ICP-AES.23 Additionally, a method to determine the inorganic Hg content of the tissues has been developed by capitalizing on the volatility of methylmercury halide derivatives.Iodoacetic acid (IOAc) is added to the alkaline digest and the methylmercury iodide (MeHgI) is distilled from the sample as the graphite tube is gently heated. The inorganic mercury remaining in the graphite tube after heating to 120 °C is then quantified. Experimental Instrument A Perkin-Elmer SCIEX ELAN 5000 ICP mass spectrometer equipped with a HGA-600MS electrothermal vaporizer (ETV) and Model AS-60 autosampler was used.The experimental conditions for the mass spectrometer are given in Table 1. The Table 1 Instrumental operating conditions and data acquisition parameters ICP mass spectrometer— Rf power 1050 W Outer argon flow rate 15.0 l min21 Intermediate argon flow rate 850 ml min21 Carrier argon flow rate 900 ml min21 Sampler-skimmer Nickel Data acquistion— Dwell time 30 ms Scan mode Peak hopping Signal measurement mode Integrated response Analyst, August 1997, Vol. 122 (751–754) 751HGA-600 MS has been described previously24 and was interfaced to the plasma via a 0.8 m length of 6 mm id poly(tetrafluoroethylene) tubing.The 202Hg isotope was monitored. The ETV program is outlined in Table 2. An initial high temperature step cleans the furnace. Following furnace cooling in step 2, the sample is injected. The sample is dried in step 3 and the moisture and vapours are vented through the dosing hole. In step 4 a graphite probe is pneumatically activated to seal the dosing hole and the carrier gas is directed from one end of the ETV workhead to the argon plasma at a flow rate of 900 ml min21 prior to heating the graphite tube to 800 °C.The graphite tube is not cleaned at high temperature immediately after the analytical signal has been obtained, rather cleaning occurs when the next analysis cycle is initiated. This prevents the residue remaining after the 800 °C pyrolysis from entering the plasma and transfer line.Reagents TMAH (30% in methanol, Aldrich, Milwaukee, WI, USA) was used to solubilize the samples. A 37 mg ml21 iodoacetic acid solution (IOAc, Aldrich) was prepared in water. A 25 mg ml21 sodium thiosulfate (Na2S2O3) solution was prepared by dissolution of the salt in high purity water. Reagent grade acetic acid (HOAc; Environmental Grade, Anachemia, Montreal, QC, Canada) was used. High purity water was produced by passage through a commercial reverse osmosis unit followed by a mixed bed ion-exchange system (Barnstead Nanopure, Barnstead, Dubuque, IA, USA).Mercury solutions were prepared by dissolution of HgCl2 (gold star, Alfa Chemicals, Ward Hill, MA, USA) in dilute nitric acid. A methylmercury standard was prepared by dissolving MeHgCl (Alpha Division, Danvers, MA, USA) in an appropriate amount of propan-2-ol. Working solutions were prepared daily by serial dilution with high purity water. National Research Council of Canada (NRCC) reference materials DORM-2 (dogfish muscle), DOLT-2 (dogfish liver) and TORT-2 (lobster hepatopancreas) were used to assess the accuracy of the method.Sample preparation A 0.250 g sample was weighed into a beaker and 4 ml of TMAH was added. Following the reaction of the tissue with the TMAH for approximately 5 min, the solution was transferred to a 25 ml calibrated flask and diluted to volume with high purity water. Samples of DORM-2 were diluted ten fold prianalysis. Total mercury A portion (10 ml) of the above sample was injected into the HGA furnace and the program listed in Table 2 was initiated.Quantification was accomplished by the method of standard additions using either standards of methyl or inorganic mercury. Inorganic mercury A 250 ml aliquot of the sample was pipetted into a sample cup along with 125 ml of the IOAc solution and 125 ml of Na2S2O3 solution. The mixture was acidified by addition of 20 ml of HOAc. The mixture (10 ml) was injected into the graphite tube.Quantification was accomplished by the method of standard additions using standards of inorganic mercury. Results and Discussion TMAH is an efficient reagent for solubilizing tissue samples. Although the resulting sample ‘digest’ is not clear and colourless, an unmodified autosampler can be used to pipette the samples into the graphite tube as ultrasonic mixing of the solution is not required. Thus, the ease of sample preparation using this approach is a distinct advantage over conventional slurry methods and solid sampling techniques where a weighed dry mass is inserted into the graphite tube.Also, imprecision due to inhomogeneity inherent to the sample can be minimized as larger masses can be subsampled. Other reports have suggested that high-pressure homogenization22 or heating23 of the alkaline digest is required to solubilize the tissue; this was not found necessary for this study. On standing, the digest becomes less cloudy in appearance but no difference in final results was found for samples prepared 5 min or 2 months prior to determination. Total mercury (inorganic and organic) could be determined in a solution of solubilized biological tissue using the ETV program outlined in Table 2.No modifiers were required and a single peak was observed using a temperature of 800 °C to vaporize the mercury. This temperature was chosen as the minimum required to give a single peak but not so high that less volatile matrix components would be transferred to the plasma.Lower vaporization temperatures resulted in multiple peaks. Matrix modifiers are important in ETV ICP-MS not only for altering the chemical composition of the sample, but to act as physical carriers to help transfer the analyte from the graphite tube to the plasma.24 In the present study, sufficient matrix components are present in the sample to serve as a physical carrier for the inorganic Hg.The instrument software would not permit the gas flow to be redirected through the dosing hole in the final steps of the furnace program. Thus, to prevent residue from the 800 °C vaporization from entering the transfer line, a high temperature cleaning of the graphite tube was not performed at the end of the furnace program. Rather, prior to sample introduction, each furnace cycle commenced with a high-temperature heating step with gas flow directed through the dosing hole. This ensured removel of residue from the previous sample from the system.Minimisation of sample material entering the transfer line reduced problems encountered from the build up of condensed matrix components which over a period of 40 to 50 cycles reduced the sensitivity. The determination of inorganic mercury was achieved by adding IOAc, Na2S2O3 and HOAc to the sample. IOAc cleaves the methylmercury from the biological tissue and the MeHgI that is formed is volatilized as the sample dries and is removed from the system through the dosing hole.Na2S2O3 was added to prevent degradation of MeHgI and a small amount of HOAc (20 ml) was added to prevent formation of MeHgOH.12 The gas flow was then redirected to the plasma, the tube heated to 800 °C and the inorganic Hg remaining in the sample was volatilized and quantified. Fig. 1 shows the peak obtained for a sample of DOLT-2 with and without a spike of inorganic Hg. The concentrations of IOAc and Na2S2O3 were not reoptimized in this study but chosen from ref. 12. A subsequent paper by Lansens and Baeyens25 improved the headspace chromatography procedure by replacing Na2S2O3 with H2SO4 to enhance cleaving of methylmercury from the Table 2 Electrothermal vaporizer program Time/s Gas to Temperature/ Gas Flow/ Step °C Ramp Hold ml min21 Vent ICP 1 2500 1 5 300 X 2 20 1 20 300 X 3 120 1 25 300 X 4 120 1 10 0 X 5 800 1 7 0 X 6 20 1 10 0 X 752 Analyst, August 1997, Vol. 122biological tissue.This was not utilized here as the conditions used in this study indicated complete conversion of the methylmercury to the iodide form and the use of H2SO4 is detrimental to the lifetime of the graphite tube. The vaporization of methylmercury as well as inorganic mercury from a biological tissue is shown in Fig. 2. To obtain this figure the ETV gas flow was directed to the plasma during the sample drying step and step 3 of the ETV program (Table 2) was modified to a 40 s ramp and 20 s hold to ensure a very slow drying of the sample.Signals due to methylmercury and inorganic mercury occur at 15 and 55 s, respectively. Attempts to separately quantify the organic mercury peak were unsuccessful because, under the conditions used to obtain this data, carryover of inorganic mercury augments the methylmercury peak. It appears that too rapid drying of the sample physically carries a fraction of the inorganic mercury along with water vapour to the plasma.This was confirmed by adding a spike of inorganic 201Hg to the sample and monitoring the 201/202Hg ratio. A slow dry period was required to eliminate this problem. Unfortunately, by increasing the drying time, the peak for methylmercury becomes too broad to be accurately quantified and is not well separated from the inorganic mercury peak. The early peak response from organic mercury is 20 fold less sensitive (peak hight) than that from an inorganic spike that occurs later in the vaporization program.This is a result of the combination of little or no useful physical carrier at low temperatures and the water vapour which enters and perturbs the plasma. Effect of Sample Mass on Analyte Response Fig. 3 shows the effect on the response for various amounts of diluted DORM-2 added to the ETV. Response is normalized to 1 ng of total mercury. At optimum ( Å 60 mg), the signal obtained is 21 fold greater than the response in deionized, distilled water.At lower masses, analyte transport becomes less efficient and at higher masses it is suspected that space charge effects reduce sensitivity or perhaps condensation of matrix components on colder parts of the system scavenges the analyte. Although response varies substantially with sample mass, the certified value for total Hg in DORM-2 was obtained at masses of 25 mg, 60 mg and 250 mg using the method of additions. Figures of Merit A limit of detection of 0.05 mg g21 in the solid tissue is obtained based on 3 s of a blank TMAH solution.The Hg content of the blank varied with the stock solutions of TMAH. Dilutions using a bottle of TMAH stored for several years in the laboratory gave a blank of 23 ng, whereas that from a new bottle of TMAH was 12 ng. The precision of replicate measurement was determined to be ±7.8% using a solution of DORM-2 at a concentration 50 fold above the detection limit. Analytical Results The accuracy of the method was evaluated by analysing a suite of marine biological certified reference materials.6 DORM-2, dogfish flesh material, DOLT-2, dogfish liver tissue and TORT- 2, lobster hepatopancreas, are all certified for total and methylmercury content.The determined values for total mercury agree with the certified values (see Table 3). The determined values for inorganic mercury agree with the Fig. 1 ETV ICP-MS signals for (a) DOLT-2 and (b) DOLT-2 spiked with 9.6 pg inorganic Hg.Fig. 2 ETV ICP-MS signals for DOLT-2 using a slow drying step. Fig. 3 Effect of total mass of DORM-2 on normalized response for 1 ng of inorganic Hg. Table 3 Analytical results Certified/mg g21 Determined/mg g21 Reference Methyl Total Total Inorganic material mercury mercury mercury mercury DORM-2 4.47 ± 0.32 4.64 ± 0.26 4.3 ± 0.5 0.33 ± 0.02 DOLT-2 0.693 ± 0.053 2.14 ± 0.28 2.20 ± 0.11 1.4 ± 0.2 TORT-2 0.152 ± 0.013 0.27 ± 0.06 0.25 ± 0.04 0.10 ± 0.02 Analyst, August 1997, Vol. 122 753difference between the certified total and methyl mercury content. These results demonstrate the ETV device can serve as a thermochemical reactor wherein samples may be subjected to a controlled thermal and chemical atmosphere, permiting speciation based on vapour pressures. References 1 Organometallic Compounds in the Environment, ed. Craig, P. J., Wiley, New York, 1986, p. 65. 2 Surma-Aho, K., Paasivirta, J., Rekolainen, S., and Verta, M., Chemosphere, 1986, 15, 353. 3 Baeyens, W., Trends Anal.Chem., 1992, 11, 245. 4 Kiceniuk, J. W., and Ray S., Analysis of Contaminants in Edible Aquatic Resources, VCH, Weinheim, Germany, 1994, ch. 8. 5 West�o�o, G., Acta. Chem. Scand., 1968, 22B, 2277. 6 Berman, S. S., Siu, K. W. M., Maxwell, P. S., Beauchemin, D., and Clancy, V. P., Fresenius’ Z. Anal. Chem., 1989, 333, 641. 7 Donais, M. K., Uden, P. C., Schantz, M. M., and Wise, S. A., Anal. Chem., 1996, 68, 3859. 8 Huang, C.-W., and Jiang, S.-J., J.Anal. At. Spectrom., 1993, 8, 681. 9 Bloxam, M. J., Gachanja, A., Hill, S. J., and Worsfold, P. J., J. Anal. At. Spectrom., 1996, 11, 145. 10 Rapsomanikis, S., Donard, O. F. X., and Weber, J. H., Anal. Chem., 1986, 58, 35. 11 Bulska, E., Baxter, D. C., and Frech, W., Anal. Chim. Acta, 1991, 249, 545. 12 Decadt, G., Baeyens, W., Bradley, D., and Goeyens, L., Anal. Chem., 1985, 57, 2788. 13 Lansens, P., Mueleman, C., Casais, C., and Baeyens, W., Appl. Organomet. Chem., 1993, 7, 45. 14 Campos, R. C., Curtius, A. J., and Berndt, H., J. Anal. At. Spectrom., 1990, 5, 669. 15 Dumarey, R., and Dams, R., Mikrochim. Acta, 1984, 3, 191. 16 Nicholson, R. A., Analyst, 1977, 102, 399. 17 Bombach, G., Bombach, K., and Klemm, W., Fresenius’ J. Anal. Chem., 1994, 350, 18. 18 Hanamura, S., Smith, B. W., and Winefordner, J. D., Anal. Chem., 1983, 55, 2026. 19 Bauer, C. F., and Natusch, D. F. S., Anal. Chem., 1981, 53, 2020. 20 Jackson, A. J., Michael, L. W., and Schumacher, H.J., Anal. Chem., 1972, 44, 1064. 21 Gross, S. B., and Parkinson, E. A., At. Absorpt. Newsl., 1974, 13, 387. 22 Tan, Y., Blais, J.-S., and Marshall, W. D., Analyst, 1996, 121, 1419. 23 Uchida, T., Isoyama, H., Yamada, K., Oguchi, K., Nakagawa, G., Sugie, H., and Iida, C., Anal. Chim. Acta, 1992, 256, 277. 24 Sturgeon, R. E., Willie, S. N., Zheng, J., Kudo, A., and Gr�egoire, D. C., J. Anal. At. Spectrom., 1993, 8, 1053. 25 Lansens, P., and Baeyens, W., Anal. Chim. Acta, 1990, 228, 93.Paper 7/01169E Received February 19, 1997 Accepted April 15, 1997 754 Analyst, August 1997, Vol. 122 Determination of Inorganic and Total Mercury in Biological Tissues by Electrothermal Vaporization Inductively Coupled Plasma Mass Spectrometry Scott N. Willie*a, D. Conrad Gr�egoireb and Ralph E. Sturgeona a Institute for National Measurement Standards, National Research Council of Canada, Ottawa, Ontario, Canada K1A 0R9 b Geological Survey of Canada, 601 Booth Street, Ottawa, Ontario, Canada K1A 0E8 A rapid method for the determination of total and inorganic mercury in biological tissues is presented using electrothermal vaporization inductively coupled plasma mass spectrometry (ETV ICP–MS). Samples were solubilized using tetramethylammonium hydroxide.For the determination of total mercury sample aliquots (10 ml) are dried and vaporized into the plasma. For the determination of inorganic mercury, iodoacetic acid, sodium thiosulfate and acetic acid are added to the sample, cleaving the methylmercury from the tissue.Volatile methylmercury iodide is formed and removed from the ETV as the sample dries, leaving only inorganic mercury to be quantified. A limit of detection of 0.05 mg g21 in solid samples was obtained. National Research Council of Canada reference materials DORM-2 (dogfish muscle), DOLT-2 (dogfish liver) and TORT-2 (lobster hepatopancreas) were used to assess the accuracy of the method.Keywords: Mercury; inorganic mercury; biological tissues; electrothermal vaporization inductively coupled plasma mass spectrometry; tetramethylammonium hydroxide solubilization Mercury is a ubiquitous element in the environment resulting from both anthropogenic and natural geological activities. In the marine environment the distribution and speciation of Hg is of considerable interest as biomethylation1 and subsequently bioconcentration in the food chain2 can occur. As a result, mercury is predominantly present in marine biological samples as methylmercury.It is well known that organomercury compounds are more toxic than metallic or aquo Hg forms; consequently, considerable effort and progress have been made in the development of techniques capable of separating and identifying the various mercury species.3 Common methods for the determination of total mercury require sample digestion followed by chemical reduction and subsequent gas phase transfer of the element to a detector. Approaches for speciation analysis involve milder sample preparation procedures to preserve the integrity of any organic mercury.4 Frequently, these are variations of the classical West�o�o5 procedure involving solvent extraction, cleanup of the extract, isolation of the organomercurial as the corresponding halide derivative and detection by gas chromatography with electron capture detection (GC–ECD)6 or atomic emission detection (GC–AED).7 Methods using liquid chromatography (LC) also require sample extraction and cleanup steps.8,9 Alternatively, procedures utilizing derivatisation of organomercury via ethylating10 or butylating11 agents followed by isolation and detection have been reported. Generally, these procedures require less sample cleanup than GC or LC methods but separation via a chromatographic column or cryogenic trapping is necessary. Another method that does not require extensive sample extraction and cleanup utilizes headspace sampling.A cleaving agent is used to liberate the methylmercury from the biological tissue, whereupon it is subsequently converted to volatile methylmercury iodide, sampled, desorbed onto a column and detected using GC–MIP (microwave induced plasma).12,13 The determination of Hg by thermal vaporization from solid samples is an attractive approach that requires minimal sample preparation.14 This technique has been applied to environmental samples15 and rocks and sediments16 using quartz tube AAS detection.As well as total Hg, direct methods for the speciation of mercury in soils and sediments17 and biologicals18 have been reported. ETV ICP-MS is well-suited for application to thermal vaporization methodology as the temperature of the graphite tube can be accurately controlled, small samples can be easily injected and the sampling frequency is relatively high. In addition, the tolerance of ICP-MS to the solvent and matrix components in the sample is superior to the less powerful microwave He plasma system.19 The determination of total and inorganic mercury in biological tissues is reported here.Extraction or acid digestion of the samples is replaced by room temperature treatment of the tissue with tetramethylammonium hydroxide (TMAH). TMAH, an alkali tissue solublizer, has been utilized for the speciation analysis of tin and mercury and also for total element analysis using flame,20 graphite furance21,22 and ICP-AES.23 Additionally, a method to determine the inorganic Hg content of the tissues has been developed by capitalizing on the volatility of methylmercury halide derivatives.Iodoacetic acid (IOAc) is added to the alkaline digest and the methylmercury iodide (MeHgI) is distilled from the sample as the graphite tube is gently heated. The inorganic mercury remaining in the graphite tube after heating to 120 °C is then quantified. Experimental Instrument A Perkin-Elmer SCIEX ELAN 5000 ICP mass spectrometer equipped with a HGA-600MS electrothermal vaporizer (ETV) and Model AS-60 autosampler was used.The experimental conditions for the mass spectrometer are given in Table 1. The Table 1 Instrumental operating conditions and data acquisition parameters ICP mass spectrometer— Rf power 1050 W Outer ate 15.0 l min21 Intermediate argon flow rate 850 ml min21 Carrier argon flow rate 900 ml min21 Sampler-skimmer Nickel Data acquistion— Dwell time 30 ms Scan mode Peak hopping Signal measurement mode Integrated response Analyst, August 1997, Vol. 122 (751–754) 751HGA-600 MS has been described previously24 and was interfaced to the plasma via a 0.8 m length of 6 mm id poly(tetrafluoroethylene) tubing. The 202Hg isotope was monitored. The ETV program is outlined in Table 2. An initial high temperature step cleans the furnace. Following furnace cooling in step 2, the sample is injected. The sample is dried in step 3 and the moisture and vapours are vented through the dosing hole. In step 4 a graphite probe is pneumatically activated to seal the dosing hole and the carrier gas is directed from one end of the ETV workhead to the argon plasma at a flow rate of 900 ml min21 prior to heating the graphite tube to 800 °C.The graphite tube is not cleaned at high temperature immediately after the analytical signal has been obtained, rather cleaning occurs when the next analysis cycle is initiated.This prevents the residue remaining after the 800 °C pyrolysis from entering the plasma and transfer line. Reagents TMAH (30% in methanol, Aldrich, Milwaukee, WI, USA) was used to solubilize the samples. A 37 mg ml21 iodoacetic acid solution (IOAc, Aldrich) was prepared in water. A 25 mg ml21 sodium thiosulfate (Na2S2O3) solution was prepared by dissolution of the salt in high purity water. Reagent grade acetic acid (HOAc; Environmental Grade, Anachemia, Montreal, QC, Canada) was used.High purity water was produced by passage through a commercial reverse osmosis unit followed by a mixed bed ion-exchange system (Barnstead Nanopure, Barnstead, Dubuque, IA, USA). Mercury solutions were prepared by dissolution of HgCl2 (gold star, Alfa Chemicals, Ward Hill, MA, USA) in dilute nitric acid. A methylmercury standard was prepared by dissolving MeHgCl (Alpha Division, Danvers, MA, USA) in an appropriate amount of propan-2-ol. Working solutions were prepared daily by serial dilution with high purity water.National Research Council of Canada (NRCC) reference materials DORM-2 (dogfish muscle), DOLT-2 (dogfish liver) and TORT-2 (lobster hepatopancreas) were used to assess the accuracy of the method. Sample preparation A 0.250 g sample was weighed into a beaker and 4 ml of TMAH was added. Following the reaction of the tissue with the TMAH for approximately 5 min, the solution was transferred to a 25 ml calibrated flask and diluted to volume with high purity water.Samples of DORM-2 were diluted ten fold prior to analysis. Total mercury A portion (10 ml) of the above sample was injected into the HGA furnace and the program listed in Table 2 was initiated. Quantification was accomplished by the method of standard additions using either standards of methyl or inorganic mercury. Inorganic mercury A 250 ml aliquot of the sample was pipetted into a sample cup along with 125 ml of the IOAc solution and 125 ml of Na2S2O3 solution.The mixture was acidified by addition of 20 ml of HOAc. The mixture (10 ml) was injected into the graphite tube. Quantification was accomplished by the method of standard additions using standards of inorganic mercury. Results and Discussion TMAH is an efficient reagent for solubilizing tissue samples. Although the resulting sample ‘digest’ is not clear and colourless, an unmodified autosampler can be used to pipette the samples into the graphite tube as ultrasonic mixing of the solution is not required.Thus, the ease of sample preparation using this approach is a distinct advantage over conventional slurry methods and solid sampling techniques where a weighed dry mass is inserted into the graphite tube. Also, imprecision due to inhomogeneity inherent to the sample can be minimized as larger masses can be subsampled. Other reports have suggested that high-pressure homogenization22 or heating23 of the alkaline digest is required to solubilize the tissue; this was not found necessary for this study.On standing, the digest becomes less cloudy in appearance but no difference in final results was found for samples prepared 5 min or 2 months prior to determination. Total mercury (inorganic and organic) could be determined in a solution of solubilized biological tissue using the ETV program outlined in Table 2. No modifiers were required and a single peak was observed using a temperature of 800 °C to vaporize the mercury.This temperature was chosen as the minimum required to give a single peak but not so high that less volatile matrix components would be transferred to the plasma. Lower vaporization temperatures resulted in multiple peaks. Matrix modifiers are important in ETV ICP-MS not only for altering the chemical composition of the sample, but to act as physical carriers to help transfer the analyte from the graphite tube to the plasma.24 In the present study, sufficient matrix components are present in the sample to serve as a physical carrier for the inorganic Hg.The instrument software would not permit the gas flow to be redirected through the dosing hole in the final steps of the furnace program. Thus, to prevent residue from the 800 °C vaporization from entering the transfer line, a high temperature cleaning of the graphite tube was not performed at the end of the furnace program. Rather, prior to sample introduction, each furnace cycle commenced with a high-temperature heating step with gas flow directed through the dosing hole.This ensured removel of residue from the previous sample from the system. Minimisation of sample material entering the transfer line reduced problems encountered from the build up of condensed matrix components which over a period of 40 to 50 cycles reduced the sensitivity. The determination of inorganic mercury was achieved by adding IOAc, Na2S2O3 and HOAc to the sample.IOAc cleaves the methylmercury from the biological tissue and the MeHgI that is formed is volatilized as the sample dries and is removed from the system through the dosing hole. Na2S2O3 was added to prevent degradation of MeHgI and a small amount of HOAc (20 ml) was added to prevent formation of MeHgOH.12 The gas flow was then redirected to the plasma, the tube heated to 800 °C and the inorganic Hg remaining in the sample was volatilized and quantified. Fig. 1 shows the peak obtained for a sample of DOLT-2 with and without a spike of inorganic Hg. The concentrations of IOAc and Na2S2O3 were not reoptimized in this study but chosen from ref. 12. A subsequent paper by Lansens and Baeyens25 improved the headspace chromatography procedure by replacing Na2S2O3 with H2SO4 to enhance cleaving of methylmercury from the Table 2 Electrothermal vaporizer program Time/s Gas to Temperature/ Gas Flow/ Step °C Ramp Hold ml min21 Vent ICP 1 2500 1 5 300 X 2 20 1 20 300 X 3 120 1 25 300 X 4 120 1 10 0 X 5 800 1 7 0 X 6 20 1 10 0 X 752 Analyst, August 1997, Vol. 122biological tissue. This was not utilized here as the conditions used in this study indicated complete conversion of the methylmercury to the iodide form and the use of H2SO4 is detrimental to the lifetime of the graphite tube. The vaporization of methylmercury as well as inorganic mercury from a biological tissue is shown in Fig. 2. To obtain this figure the ETV gas flow was directed to the plasma during the sample drying step and step 3 of the ETV program (Table 2) was modified to a 40 s ramp and 20 s hold to ensure a very slow drying of the sample.Signals due to methylmercury and inorganic mercury occur at 15 and 55 s, respectively. Attempts to separately quantify the organic mercury peak were unsuccessful because, under the conditions used to obtain this data, carryover of inorganic mercury augments the methylmercury peak. It appears that too rapid drying of the sample physically carries a fraction of the inorganic mercury along with water vapour to the plasma.This was confirmed by adding a spike of inorganic 201Hg to the sample and monitoring the 201/202Hg ratio. A slow dry period was required to eliminate this problem. Unfortunately, by increasing the drying time, the peak for methylmercury becomes too broad to be accurately quantified and is not well separated from the inorganic mercury peak. The early peak response from organic mercury is 20 fold less sensitive (peak hight) than that from an inorganic spike that occurs later in the vaporization program.This is a result of the combination of little or no useful physical carrier at low temperatures and the water vapour which enters and perturbs the plasma. Effect of Sample Mass on Analyte Response Fig. 3 shows the effect on the response for various amounts of diluted DORM-2 added to the ETV. Response is normalized to 1 ng of total mercury. At optimum ( Å 60 mg), the signal obtained is 21 fold greater than the response in deionized, distilled water.At lower masses, analyte transport becomes less efficient and at higher masses it is suspected that space charge effects reduce sensitivity or perhaps condensation of matrix components on colder parts of the system scavenges the analyte. Although response varies substantially with sample mass, the certified value for total Hg in DORM-2 was obtained at masses of 25 mg, 60 mg and 250 mg using the method of additions.Figures of Merit A limit of detection of 0.05 mg g21 in the solid tissue is obtained based on 3 s of a blank TMAH solution. The Hg content of the blank varied with the stock solutions of TMAH. Dilutions using a bottle of TMAH stored for several years in the laboratory gave a blank of 23 ng, whereas that from a new bottle of TMAH was 12 ng. The precision of replicate measurement was determined to be ±7.8% using a solution of DORM-2 at a concentration 50 fold above the detection limit.Analytical Results The accuracy of the method was evaluated by analysing a suite of marine biological certified reference materials.6 DORM-2, dogfish flesh material, DOLT-2, dogfish liver tissue and TORT- 2, lobster hepatopancreas, are all certified for total and methylmercury content. The determined values for total mercury agree with the certified values (see Table 3). The determined values for inorganic mercury agree with the Fig. 1 ETV ICP-MS signals for (a) DOLT-2 and (b) DOLT-2 spiked with 9.6 pg inorganic Hg. Fig. 2 ETV ICP-MS signals for DOLT-2 using a slow drying step. Fig. 3 Effect of total mass of DORM-2 on normalized response for 1 ng of inorganic Hg. Table 3 Analytical results Certified/mg g21 Determined/mg g21 Reference Methyl Total Total Inorganic material mercury mercury mercury mercury DORM-2 4.47 ± 0.32 4.64 ± 0.26 4.3 ± 0.5 0.33 ± 0.02 DOLT-2 0.693 ± 0.053 2.14 ± 0.28 2.20 ± 0.11 1.4 ± 0.2 TORT-2 0.152 ± 0.013 0.27 ± 0.06 0.25 ± 0.04 0.10 ± 0.02 Analyst, August 1997, Vol. 122 753difference between the certified total and methyl mercury content. These results demonstrate the ETV device can serve as a thermochemical reactor wherein samples may be subjected to a controlled thermal and chemical atmosphere, permiting speciation based on vapour pressures.References 1 Organometallic Compounds in the Environment, ed. Craig, P. J., Wiley, New York, 1986, p. 65. 2 Surma-Aho, K., Paasivirta, J., Rekolainen, S., and Verta, M., Chemosphere, 1986, 15, 353. 3 Baeyens, W., Trends Anal. Chem., 1992, 11, 245. 4 Kiceniuk, J. W., and Ray S., Analysis of Contaminants in Edible Aquatic Resources, VCH, Weinheim, Germany, 1994, ch. 8. 5 West�o�o, G., Acta. Chem. Scand., 1968, 22B, 2277. 6 Berman, S. S., Siu, K. W. M., Maxwell, P. S., Beauchemin, D., and Clancy, V. P., Fresenius’ Z. Anal. Chem., 1989, 333, 641. 7 Donais, M. K., Uden, P. C., Schantz, M. M., and Wise, S. A., Anal. Chem., 1996, 68, 3859. 8 Huang, C.-W., and Jiang, S.-J., J. Anal. At. Spectrom., 1993, 8, 681. 9 Bloxam, M. J., Gachanja, A., Hill, S. J., and Worsfold, P. J., J. Anal. At. Spectrom., 1996, 11, 145. 10 Rapsomanikis, S., Donard, O. F. X., and Weber, J. H., Anal. Chem., 1986, 58, 35. 11 Bulska, E., Baxter, D. C., and Frech, W., Anal. Chim. Acta, 1991, 249, 545. 12 Decadt, G., Baeyens, W., Bradley, D., and Goeyens, L., Anal. Chem., 1985, 57, 2788. 13 Lansens, P., Mueleman, C., Casais, C., and Baeyens, W., Appl. Organomet. Chem., 1993, 7, 45. 14 Campos, R. C., Curtius, A. J., and Berndt, H., J. Anal. At. Spectrom., 1990, 5, 669. 15 Dumarey, R., and Dams, R., Mikrochim. Acta, 1984, 3, 191. 16 Nicholson, R. A., Analyst, 1977, 102, 399. 17 Bombach, G., Bombach, K., and Klemm, W., Fresenius’ J. Anal. Chem., 1994, 350, 18. 18 Hanamura, S., Smith, B. W., and Winefordner, J. D., Anal. Chem., 1983, 55, 2026. 19 Bauer, C. F., and Natusch, D. F. S., Anal. Chem., 1981, 53, 2020. 20 Jackson, A. J., Michael, L. W., and Schumacher, H. J., Anal. Chem., 1972, 44, 1064. 21 Gross, S. B., and Parkinson, E. A., At. Absorpt. Newsl., 1974, 13, 387. 22 Tan, Y., Blais, J.-S., and Marshall, W. D., Analyst, 1996, 121, 1419. 23 Uchida, T., Isoyama, H., Yamada, K., Oguchi, K., Nakagawa, G., Sugie, H., and Iida, C., Anal. Chim. Acta, 1992, 256, 277. 24 Sturgeon, R. E., Willie, S. N., Zheng, J., Kudo, A., and Gr�egoire, D. C., J. Anal. At. Spectrom., 1993, 8, 1053. 25 Lansens, P., and Baeyens, W., Anal. Chim. Acta, 1990, 228, 93. Paper 7/01169E Received February 19, 1997 Accepted April 15, 1997 754 Analyst, August 1997, Vo
ISSN:0003-2654
DOI:10.1039/a701169e
出版商:RSC
年代:1997
数据来源: RSC
|
6. |
Infrared Discrimination of Enantiomerically Enriched and RacemicSamples of Methamphetamine Salts |
|
Analyst,
Volume 122,
Issue 8,
1997,
Page 755-760
J. S. Chappell,
Preview
|
|
摘要:
Infrared Discrimination of Enantiomerically Enriched and Racemic Samples of Methamphetamine Salts J. S. Chappell Drug Enforcement Administration, Western Laboratory, 390 Main Street, Room 700, San Francisco, CA 94105, USA A relatively rapid and simple means of enantiomer determination is described for the determination of methamphetamine, a common drug of abuse. The method employs the well known technique of infrared transmission spectrometry on solid samples dispersed within an alkali metal halide matrix.This approach exploits the solid-state, ion-exchange reaction between methamphetamine hydrochloride and a potassium iodide matrix and the subsequent formation of the hydriodide salt in situ. The infrared properties of the hydriodide salt are distinct for enantiomerically enriched and racemic samples, and therefore are readily distinguished by infrared transmission spectrometry. This technique uses materials and instrumentation that are generally available to most crime laboratories.The applicability of this method to some other amine drugs is discussed. Keywords: Methamphetamine; hydrochloride salt; hydriodide salt; enantiomeric composition; infrared spectrometry The analysis of illicit drug substances continues to represent a significant workload for many crime laboratories. One aspect of these analyses can be the determination of the enantiomeric composition for a chiral drug substance, which may be required by federal law where it is forensically important for sentencing purposes.Methamphetamine is one of the most common drugs of abuse to have fallen in this category, where the physiological activity of the dextrorotatory isomer (with the absolute configuration S) is at least an order of magnitude greater than the levorotatory isomer (R configuration). (A recent amendment to the federal sentencing guidelines,1 has removed the legal distinction between the enantiomers of methamphetamine; however, isomer determination is strongly encouraged within the forensic community to counter any due process claims or other potential litigation.) Identification of the methamphetamine isomer, or the presence of a racemic mixture, can also provide valuable information as to the method of manufacture, particularly with regard to the synthetic route and starting materials.2 However, enantiomer determination is not widely performed in many laboratories, in part owing to the lack of instrumentation (i.e., a polarimeter).Classically, the enantiomeric composition has been assessed by observing the precipitation of crystal forms by optical microscopy following the reaction of methamphetamine with certain metal salts.3 Methods have since been developed to resolve the enantiomers by NMR4,5 or chromatographic techniques,6–8 including capillary electrophoresis.9,10 Many of these methods rely on the preparation of diastereoisomeric derivatives of the enantiomers.An infrared approach has also been employed to distinguish enantiomerically enriched samples from racemic mixtures of methamphetamine by derivatization with phenyl isothiocyanate, 11,12 an achiral reagent. A more rapid and simple infrared technique is discussed in this paper, which may have wider application. One feature of many chiral substances is the formation of distinct crystalline phases for a single enantiomer and for a racemic mixture of the two enantiomers.This enables the two types of samples to be readily distinguished by various physical properties, including the condensed phase infrared transmission spectrum. Unfortunately, and contrary to the majority of chiral substances,13 some salts of amine drugs, including methamphetamine, do not form racemic crystals. Instead, a mechanical mixture (or conglomerate sample) results in which an equimolecular mixture of enantiomeric crystals of the two isomers is present. Since the two enantiomeric crystal forms possess identical infrared spectra, the racemic mixture presents the same spectrum as that observed for either enantiomer.This behavior prevents infrared transmission spectrometry from distinguishing between enantiomerically enriched and racemic samples of the prevalent chemical form of methamphetamine, the hydrochloride salt. In this paper, a simple procedure is outlined that induces the formation of a racemic phase for methamphetamine, and thereby allows for enantiomerically enriched and racemic samples to be rapidly distinguished and confirmed by infrared spectrometry. This approach relies on the solid-state, ionexchange reaction methamphetamine·HCl + KI ? methamphetamine·HI + KCl (1) which occurs between methamphetamine hydrochloride and potassium iodide when the two powders are mixed and pressure-sintered into a transmission window for infrared measurements.We have previously reported that the hydrochloride salt of the dextrorotatory isomer of methamphetamine is reactive with alkali metal halide matrices,14 whereby the infrared spectrum observed for the amine salt can vary dramatically depending on the anion of the matrix material.It was found that the exchange of iodide ion for chloride was particularly favored for methamphetamine and some structurally related analogs (i.e., amphetamine and some ringsubstituted analogs, including the 3,4-methylenedioxyamphetamines). This phenomenon was recently found also to occur with a racemic sample of methamphetamine, except that the racemic hydriodide salt forms a liquid phase distinct in both physical appearance and infrared spectrum from that of the single enantiomer.Consequently, this behavior allows enantiomerically enriched and racemic samples of methamphetamine hydrochloride to be easily distinguished by infrared spectrometry following sample preparation within a potassium iodide matrix. This chemistry also enables the predominant isomer of an enantiomerically enriched sample to be determined since a racemic sample of methamphetamine hydrochloride may be prepared by blending equal amounts of the two enantiomers.Experimental Sample Preparation The preparation of samples for this method is no different from preparing an infrared transmission window with KCl or KBr, other than using KI as the matrix material. A sample of Analyst, August 1997, Vol. 122 (755–760) 755methamphetamine hydrochloride is ground (generally with a small agate mortar and pestle) to produce a fine powder, followed by grinding an appropriate amount of KI in with the methamphetamine salt.In this study, approximately 2–4 mg of the methamphetamine salt were used with sufficient matrix material (100–200 mg) to prepare a pressed transmission window within a 5 mm 320 mm rectangular opening cut within a blotter card. The KI solid should be ground well and mixed thoroughly with the methamphetamine sample for 30–60 s to ensure that nearly complete ion exchange occurs and methamphetamine hydriodide is formed.This composite powder is then pressed into an infrared transmission window under a load of 12 000–15 000 lb (typically with a Carver press) with or without application of a vacuum to the sample. Windows prepared with KI are generally not as visually transparent as KBr windows, but can still provide a suitable infrared transmission spectrum. The potassium halide salts employed in this study were of either analytical-reagent grade (KCl and KI) or spectroscopic grade (KBr), and gave featureless spectra throughout the mid-infrared region (4000–400 cm21).A Nicolet (Madison, WI, USA) Model 205 Fourier transform infrared spectrometer was used to record all spectra, operating at 32 scans per spectrum with a 4 cm21 resolution. This analysis requries that the sample be sufficiently free of diluents or other contaminants so that the infrared spectrum of the methamphetamine salt is clearly identifiable.Most common diluents are easily removed by dry extraction techniques to give infrared-pure methamphetamine·HCl.15 Complete identification of the predominant isomer for enantiomerically enriched samples further requires a standard for one of the enantiomers to blend with the unknown sample in order to test for the formation of the racemic form. Preferably, a standard of the levorotatory isomer (l or 2 notation) of methamphetamine·HCl would be mixed with a suspected sample of the dextrorotatory isomer (d or +) to observe the spectral change to the racemic form.Approximately equal amounts of the two should be mixed together prior to thoroughly grinding a suitable amount of the KI matrix material with the sample. The spectrum will clearly display the characteristics of the racemic HI salt if the opposite enantiomer was introduced to the original enantiomerically enriched sample. Owing to the inexact nature of weighing and mixing of equal parts of the two enantiomers and the likely presence of a small excess of one isomer, weak spectral features of the enantiomeric HI salt may also be present.Comparative Studies Enantiomerically enriched (d-isomer) and racemic (d,l-isomers) samples of amphetamine and N,N-dimethylamphetamine in addition to methamphetamine were considered in this study. The molecular structures of these compounds are depicted in Table 1.All drug samples were obtained as pharmaceutical grade standards with a high purity ( > 99%). The standards were in the form of the HCl salt, with the exception of damphetamine as the sulfate salt. For comparison purposes, the HI salts were also prepared by the complexation of the respective amine bases with hydriodic acid in concentrated aqueous solutions. The amine bases were formed by the neutralization of the respective HCl or sulfate salts with concentrated ammonia in aqueous solution, followed by extraction of the base with methylene chloride.The solvent of this extract solution was removed by evaporation and the base was collected as an oily liquid. The amine bases were then dissolved in concentrated aqueous solutions of hydriodic acid (approximately 5 mol l21) and the ion pair was extracted with chloroform. This extract was passed through a column of anhydrous sodium sulfate to remove water (anhydrous sodium sulfate is an effective drying agent for chloroform extract solutions, and there is no detectable ion exchange of sulfate for the halide of the soluble amine salt), and the HI salt was recovered by solvent evaporation of the chloroform solution over a steam-bath.The HCl salt of d-amphetamine was prepared similarly with the use of concentrated hydrochloric acid. Most of these salts were subsequently recrystallized as small white crystals with an elongated habit from saturated acetone solutions upon addition of diethyl ether.Relatively sharp melting-points were measured for all of the salt forms with the exception of the racemic samples of amphetamine·HI and methamphetamine·HI, which would only form an oily, liquid material. The melting-points of the recrystallized salts are given in Table 1. All solvents and chemicals used in the formation and recrystallization of the salts were of analyticalreagent grade. Results and Discussion The infrared spectrum of an enantiomerically enriched sample of methamphetamine·HCl (d-isomer) prepared within a KI matrix is illustrated in Fig. 1. Comparative spectra of dmethamphetamine ·HCl (in the unreactive KCl matrix) and dmethamphetamine ·HI (in KI) are also shown to illustrate the conversion that the HCl salt undergoes when prepared in KI. The changes in the spectrum are dramatic, where the spectrum of the HCl salt in KI essentially assumes the same spectral features of the HI salt and loses the characteristics of the original hydrochloride salt.The most salient differences are found in the shape of the strong absorption envelope within the hydrogen stretching region between 3200 and 2600 cm21. The strong absorbance in this region is primarily attributed to the stretching motions of the protons bound to the amine nitrogen,16 which possess considerable oscillator strength since they modulate the intrinsically large dipole of the ion pair of the amine salt. This absorption envelope also displays a complex shape that can be particularly sensitive to changes within the crystalline structure of the salt, including changes in the anion.14 In conjunction with these spectral differences, a single strong peak at 2461 cm21 is observed for the HCl salt (in KCl) that is entirely absent for the HI salt.Numerous spectral differences also exist throughout the fingerprint region (2000–400 cm21), especially with the appearance of a single strong band at 1548 cm21 and a set of three sharp peaks near 810 cm21 in the spectrum of the HI salt.Spectra for a racemic sample of methamphetamine·HCl (d,lisomers) are shown in Fig. 2. Analogously to Fig. 1, the Table 1 Melting-points (°C) for N-methyl-substituted amphetamine salts Enantiomerically enriched Racemic Halide salt sample* sample Amphetamine (R1 = R2 = H) HCl 154–155 146–149† HI 134–136 Oily liquid‡ Methamphetamine (R1 = H, R2 = CH3) HCl 170–172 133–133 HI 97–99 Oily liquid‡ N,N-Dimethylamphetamine (R1 = R2 = CH3) HCl 181–183 156–157 HI 146–147 119–121 * The dextrorotatory isomer (S configuration) is the predominant enantiomer.† Amphetamine·HCl forms a racemic crystal. ‡ Sample exists as a liquid phase at ambient conditions ( ~ 22 °C, ~ 30% relative humidity). 756 Analyst, August 1997, Vol. 122spectrum of the HCl salt is observed to change greatly between the preparations in KCl and KI. The spectrum of the racemic HCl salt in KI displays a steep slope in its baseline, and also exhibits peaks that are notably broader than those of the preparation in a KCl matrix. The broadened peaks correspond well with the features in the spectrum of the racemic HI salt, and are consistent with the amorphous structure of a liquid-like state, as opposed to the long range translational order of the crystalline HCl salt.The strong slope in the baseline is also characteristic of a liquid (or waxy) sample, which inhibits sintering of the matrix material, and results in a visually opaque transmission window with strong background scattering. The transformation to a liquid phase for a racemic sample may actually be recognized in the course of sample preparation as the KI matrix is ground into the racemic HCl salt and the powder can clump as if it were wet.This behavior, unfortunately, does not give a very clear transmission window, but the spectrum is distinct and strikingly different from the spectrum of the enantiomerically enriched sample.The ion-exchange reaction is dependent on the amount of time during which the powders of the HCl salt and KI matrix are in contact, although grinding the two solids together for 30 s appears to allow for essentially complete conversion to the HI salt. The pressure employed to sinter the matrix into a transmission window has no observable effect on the reaction rate. Ion exchange is observed to occur to a lesser extent between methamphetamine·HCl and a KBr matrix, yielding a mixture of the HCl and HBr salts in the transmission window.14 This effect may account for some observations that the spectra for enantiomerically enriched and racemic samples of methamphetamine ·HCl can show some weak differences distinguishing the two.17 The spectra of the enantiomerically enriched and racemic samples of the HBr salt exhibit some real but subtle differences, especially in the shape of the absorption envelope near 2800 cm21 and the series of peaks between 1100 and 1000 cm21.These differences, however, are very weak in the composite spectrum that results from a preparation of the HCl salt in KBr, and would not be a good gauge by which to perform the enantiomer determination. The formation of the HI salt allows enantiomerically enriched and racemic samples of methamphetamine to be distinguished, but an additional step is required to determine the predominant isomer of enantiomerically enriched samples (see Sample Preparation).This approach contrasts a general technique involving the reaction of chiral amines with chiral organic acids, and the prospective formation of chiral salt forms that can possibly distinguish the two enantiomers, in addition to the racemic mixture. Reaction with d-mandelic acid has proved particularly successful for amphetamine, where distinct crystalline phases precipitate for the d- and l-isomers as well as the racemic mixture, and all three salt forms are readily distinguished by infrared spectrometry.18 Unfortunately, mandelic acid does not precipitate any usable salt forms with methamphetamine, nor have any other chiral organic acids been reported to resolve the enantiomers of methamphetamine reliably.The ion-exchange reaction is observed to occur with other amine salts, and specifically with some structurally related analogs of methamphetamine. Observations made on amphetamine and N,N-dimethylamphetamine (and also ephedrine and pseudoephedrine) have also revealed a propensity for the hydrochloride salts to react spontaneously and nearly completly with a KI matrix within 1 min.The corresponding HI salts, however, have been found not to react significantly with a KCl (or KBr) matrix, even with pressed transmission windows stored for many weeks. This chemistry appears at first to suggest that the hydriodide phases are more stable than the hydrochloride, presumably from a greater covalent character to the ion-pair bond between the protons of the amine nitrogen and the iodide anion. This bond, however, is generally recognized to strengthen with the replacement of iodide with chloride in amine salts,16 owing to the stronger ionic attraction between these protons and the chloride anion.The interaction creates a Fig. 1 Comparison of the infrared transmission spectra for enantiomerically enriched (d-isomer) methamphetamine hydrochloride prepared in KCl and KI with the spectrum of the hydriodide prepared in a KI matrix.Fig. 2 Comparison of the infrared transmission spectra for racemic (d,lisomers) methamphetamine hydrochloride prepared in KCl and KI with the spectrum of the hydriodide observed neat between KBr discs. Analyst, August 1997, Vol. 122 757hydrogen bond that links the protonated amine to the halide anion as an ion pair.19–21 The hydrogen bond within the methamphetamine·HCl ion pair is therefore probably stronger than that in the HI salt, which implies a greater stability for the enantiomeric crystal of the HCl salt.Consequently, the driving force behind the ion-exchange reaction is probably determined by other binding preferences, and specifically, a greater affinity for the chloride anion (relative to the iodide) to bind to the hard potassium cation of the matrix. The soft iodide anion exhibits a significantly weaker binding preference than chloride, and is therefore relegated to pairing with the protonated amine.The matrix effects for enantiomerically enriched and racemic samples of amphetamine·HCl in KI are illustrated by the spectra in Figs. 3 and 4, respectively. The behavior observed is analogous to that of methamphetamine·HCl, where the enantiomerically enriched sample precipitates a crystalline HI salt (Fig. 3) whereas the racemic sample converts into a liquid hydriodide phase (Fig. 4). The spectra for the enantiomerically enriched sample in Fig. 3 illustrate an essentially complete conversion of the HCl salt into the HI salt when prepared in a KI matrix.The spectral features of the HI salt are also distinguishable from the HCl salt (in KCl) with several small yet significant differences in the shape of the amine proton stretching envelope (between 3000 and 2400 cm21), and also with the absence of the broad combination band near 2000 cm21 that is present for the HCl salt. There are also many peak differences throughout the fingerprint region for the enantiomerically enriched sample.The spectra for the racemic sample in Fig. 4 likewise show a significant change in the spectral features for the HCl salt when prepared in KI, although the difference is primarily one of broadened peak shapes due to the liquid nature of the racemic HI salt. These features correspond fairly well with those in the spectrum for a racemic sample of amphetamine·HI and, like methamphetamine, offer a clear contrast from that observed for an enantiomerically enriched sample.Unlike methamphetamine, enantiomerically enriched and racemic samples of amphetamine ·HCl (in KCl) are also distinguishable since the HCl salt of d,l-amphetamine forms a racemic crystal with a distinctive infrared spectrum. In complete contast, neither the hydrochloride nor the hydriodide salts of N,N-dimethylamphetamine form a racemic phase that can be distinguished by infrared spectrometry from the enatiomerically enriched sample. The spectra for a racemic sample of N,N-dimethylamphetamine are shown in Fig. 5, where, like amphetamine and methamphetamine, the HCl salt reacts with a KI matrix to give a spectrum equivalent to that of the HI salt. The spectrum of the HCl salt (in KCl) is also different from that of the HI salt (in KI), especially in the appearance of the absorption envelope between 3200 and 2400 cm21. The series of spectra in Fig. 5 are absolutely equivalent to those observed for the corresponding preparations of an enantiomerically enriched sample, indicating that racemic samples of both the HCl and HI salts are purely mechanical mixtures.This observation raises some question as to the nature of the liquid phase observed for the racemic hydriodide salts of methamphetamine and amphetamine. An equimolecular mixture of enantiomeric crystals for a racemic sample can experience a significant depression of its melting-point due to the solubility of each enantiomer in the melt of the other enantiomer.The depression of the meltingpoint of a mechanical mixture of enantiomers is well approximated by the expression13 ln x H R T T x f A » - æ è ç ö ø ÷ < D 1 1 1 f A f for 0.5 < (2) where x is the mole fraction of the more abundant enantiomer, Tf is the temperature at which melting is complete for the mixture, DHf A and Tf A are the heat of fusion and melting-point, respectively, for the pure enantiomer (x = 1) and R is the gas constant (1.987 cal K21 mol21).This equation describes the liquid–solid equilibrium in the phase diagram for a binary mixture of enantiomers, where the racemic mixture (x = 0.5) Fig. 3 Comparison of the infrared transmission spectra for enantiomerically enriched (d-isomer) amphetamine hydrochloride prepared in KCl and KI with the spectrum of the hydriodide prepared in a KI matrix. Fig. 4 Comparison of the infrared transmission spectra for racemic (d,lisomers) amphetamine hydrochloride prepared in KCl and KI with the spectrum of the hydriodide observed neat between KBr discs. 758 Analyst, August 1997, Vol. 122exists as a eutectic, the lowest melting-point observed among the mixtures. The derivation of this equation assumes several conditions, the most significant requiring that the enantiomers exhibit no heat of mixing in the liquid or melt state (i.e., ideal behavior). Although at first glance this may appear to be a serious restriction, in fact the heats of mixing are often observed to be less than 1% of the heat of fusion and, in general, eqn.(2) has been found to be highly reliable in the description of the melting behavior of enantiomeric mixtures.13 The melting-points measured for the salts in this study (Table 1) reveal a depressed melting-point (relative to the enantiomerically enriched samples) for all racemic samples, including the racemic crystal of amphetamine·HCl. It should also be noted that for all three amine compounds, the enantiomerically enriched HI salts melt at a considerably lower temperature than the HCl salt of the same amine.This lower melting point indicates that the enantiomeric crystal of the HI salt is less stable relative to its melted state compared with that of the HCl salt and its melt. This effect is reflected by a significant decrease in the heat of fusion for the HI salt relative to the HCl salt, as can be illustrated by N,N-dimethylamphetamine. A value of 10.8 kcal mol21 can be estimated from eqn.(2) for the heat of fusion of the enantiomeric crystal of the HCl salt, whereas that of the HI salt is calculated to be near 8.4 kcal mol21. This decrease arises from weaker cohesive interactions in the crystalline packing of the enantiomeric crystal of the HI salt relative to that of the HCl salt. The large iodide anion probably disrupts the close packing that is possible with the smaller chloride anion. A decrease in the stability of the enantiomeric crystal of the HI salts has important implications for the stability of a mechanical mixture of methamphetamine·HI within a racemic sample.The heat of fusion for the enantiomeric crystal of methamphetamine·HCl is estimated to be 6.4 kcal mol21 from eqn. (2), considerably lower than that of either salt of N,Ndimethylamphetamine. A value of 5 kcal mol21 for the heat of fusion of the HI salt implies a low melting-point near 63 °C for the racemic mixture, and the heat of fusion for methamphetamine ·HI could plausibly be lower, yielding an even lower meltingpoint. Although it is unlikely that the racemic melting-point is actually depressed below room temperature (DHf A Å 2 kcal mol21), the stability of a racemic mixture of enantiomeric crystals of the HI salt appears not to be much greater than that of the melt at room temperature.The stability of the melt phase can also be increased entropically by the inclusion of impurities, and may be particularly enhanced by the presence of dissolved moisture, which is excluded from the packing structure of the enantiomeric crystal.This effect may allow a hydrated liquid state to become metastable, if not entirely stable, at room temperature. A significant level of water is in fact indicated to be dissolved within the racemic liquid phase, as is evident from the infrared spectrum of the racemic HI salt in Fig. 2 and the large water peak near 3400 cm21. The spectrum of the racemic HCl salt prepared in KI also exhibits a sizeable peak in this region, suggesting an increased level of moisture within the KI preparation compared with that in KCl.Moisture available from the atmosphere, and also possibly adsorbed on the KI reagent, may therefore help stabilize the liquid state of the racemic HI salt at room temperature relative to the enantiomeric crystal. This description appears consistent with observations made on mechanical mixtures prepared from the enantiomeric crystals of the d- and l-isomers of methamphetamine·HI. Preparations made by simply mixing equal portions of powders of the HI salts exhibit a melting-point near 60 °C, and remain solid while stored in a dry environment over phosphorus pentoxide (P2O5).These solid mixtures appear stable for several weeks when stored over P2O5, but are observed to liquefy within 24 h upon standing in a room atmosphere (approximately 22 °C, approximately 30% relative humidity).The liquefaction of the racemic HI salt is also reversible within a dry environment, even though it is a much slower process. Pressed transmission windows containing the liquid racemic HI salt (the HCl salt in KI and the HI salt in KCl) are observed slowly to develop weak spectral features of the enantiomeric HI salt over the course of several weeks when stored over P2O5. This observation indicates that a mechanical mixture of enantiomeric crystals is actually the stable state for the racemic HI salt in the absence of absorbable moisture.The dehydration of the liquid HI salt within a dry environment and the subsequent precipitation of the enantiomeric crystal are extremely slow processes, however, and no racemic sample has yet been found to convert completely into an infrared-pure transmission spectrum of the enantiomeric crystal. Similar behavior is also observed for transmission windows containing racemic amphetamine·HI, although the crystallization of the enantiomeric HI salt within the matrix appears to progress faster than that for methamphetamine ·HI.Interestingly, the enantiomeric crystal of N,N-dimethylamphetamine ·HI is sufficiently stable to ensure that a mechanical mixture is the ground state for a racemic sample at ambient conditions, even when atmospheric moisture is available for hydration of the liquid phase. In summary, this method for the enantiomer determination of methamphetamine relies on a reduced stability for the enantiomeric crystal of the HI salt relative to the cohesive energy of the HCl salt.This destabilization of the HI salt is apparently sufficient to make a racemic mixture of enantiomeric crystals unstable towards liquefaction under ambient conditions, whereby atmospheric moisture appears to enhance further the stability of the melt phase through hydration. This effect is also observed for amphetamine, and is probably intrinsic to many other low molecular mass amine salts.The appearance of matrix effects with KI may then be possibly exploited as a means of enantiomer determination. Some care, however, should be exercised when interpreting spectral differences between enantiomerically enriched and racemic samples of a particular amine salt. Spectral differences may also arise due to variable rates for the ion-exchange reaction in the two samples, even Fig. 5 Comparison of the infrared transmission spectra for racemic (d,lisomers) N,N-dimethylamphetamine hydrochloride prepared in KCl and KI with the spectrum of the hydriodide prepared in a KI matrix.Analyst, August 1997, Vol. 122 759when both samples are converting into the same crystalline phase. Several factors can affect the rate of ion exchange and the subsequent precipitation of a salt with the anion of the matrix; most notable are the average particle size (and size distribution) and the amount of adsorbed moisture over the particle surface.Particle size dictates the area of the contact interface between grains within the matrix, and hence the size of the reaction zone for ion exchange. The moisture content of a powder sample has also been found to affect greatly the rate of ion exchange, since the water adsorbed to the particle surface apparently activates the reaction, presumably by ‘solvating’ the ions at or near the interface between grains. For the amines considered here and their reaction with KI, ion exchange is fairly rapid and conversion to the hydriodide salt is essentially complete within 1 min, so that the rate of the reaction does not become an issue in the spectral interpretation. References 1 Proposed 1995 Guideline Amendments, 60 Federal Register 25074. 2 Allen, A., and Cantrell, T.S., Forensic Sci. Int., 1989, 42, 183. 3 Official Methods of Analysis of the Association of Official Analytical Chemists, ed. Horwitz, W., Association of Official Analysis Chemists, Arlington, VA, 10th edn., 1965, p. 597. 4 Kram, T. C., and Lurie, I. S., Forensic Sci. Int., 1992, 55, 131. 5 LaBelle, M. J., Savard, C., Dawson, B. A., Black, D. B., Katyal, L. K., Zrcek, F., and By, A. W., Forensic Sci. Int., 1995, 71, 215. 6 Wells, C. E., J. Ass. Off. Anal. Chem., 1970, 53, 113. 7 Noggle, F. T., and Clark, C. R., J. Forensic Sci., 1986, 31, 732. 8 Makino, Y., Ohta, S., and Hirobe, M., Forensic Sci. Int., 1996, 78, 65. 9 Lurie, I. S., J. Chromatogr., 1992, 605, 269. 10 Lurie, I., Klein, R., Dal Cason, T., LaBelle, M., Brenneisen, R., and Weinberger, R., Anal.Chem., 1994, 66, 4019. 11 Allen, A., and Cooper, D., Microgram, 1979, 12, 24. 12 Shriner, R. L., Fuson, R. C., and Curtin, D. Y., The Systematic Identification of Organic Compounds, Wiley, New York, 1964, pp. 256–259. 13 Jacques, J., Collet, A., and Wilen, S. H., Enantiomers, Racemates, and Resolutions, Krieger, Malabar, FL, 1994, pp. 43–48. 14 Chappell, J. S., Forensic Sci. Int., 1995, 75, 1. 15 Ely, R. A., Drug Enforcement Administration, San Francisco, CA, USA, personal communication, 1993. 16 Colthup, N. B., Daly, L. H., and Wiberly, S. E., Introduction to Infrared and Raman Spectroscopy, Academic Press, New York, 1964, pp. 281–282. 17 Heagy, J. A., Drug Enforcement Administration, San Francisco, CA, USA, personal communication, 1996. 18 Heagy, J. A., Anal. Chem., 1970, 42, 1459. 19 For review, see Pauling, L., The Nature of the Chemical Bond, Cornell University Press, Ithaca, NY, 3rd edn., 1960, pp. 449–504. 20 Chenon, B., and Sandorfy, C., Can. J. Chem., 1958, 36, 1181. 21 Brissette, C., and Sandorfy, C., Can. J. Chem., 1960, 38, 34. Paper 7/00122C Received January 6, 1997 Accepted April 15, 1997 760 Analyst, August 1997, Vol. 122 Infrared Discrimination of Enantiomerically Enriched and Racemic Samples of Methamphetamine Salts J. S. Chappell Drug Enforcement Administration, Western Laboratory, 390 Main Street, Room 700, San Francisco, CA 94105, USA A relatively rapid and simple means of enantiomer determination is described for the determination of methamphetamine, a common drug of abuse.The method employs the well known technique of infrared transmission spectrometry on solid samples dispersed within an alkali metal halide matrix. This approach exploits the solid-state, ion-exchange reaction between methamphetamine hydrochloride and a potassium iodide matrix and the subsequent formation of the hydriodide salt in situ.The infrared properties of the hydriodide salt are distinct for enantiomerically enriched and racemic samples, and therefore are readily distinguished by infrared transmission spectrometry. This technique uses materials and instrumentation that are generally available to most crime laboratories. The applicability of this method to some other amine drugs is discussed. Keywords: Methamphetamine; hydrochloride salt; hydriodide salt; enantiomeric composition; infrared spectrometry The analysis of illicit drug substances continues to represent a significant workload for many crime laboratories.One aspect of these analyses can be the determination of the enantiomeric composition for a chiral drug substance, which may be required by federal law where it is forensically important for sentencing purposes. Methamphetamine is one of the most common drugs of abuse to have fallen in this category, where the physiological activity of the dextrorotatory isomer (with the absolute configuration S) is at least an order of magnitude greater than the levorotatory isomer (R configuration). (A recent amendment to the federal sentencing guidelines,1 has removed the legal distinction between the enantiomers of methamphetamine; however, isomer determination is strongly encouraged within the forensic community to counter any due process claims or other potential litigation.) Identification of the methamphetamine isomer, or the presence of a racemic mixture, can also provide valuable information as to the method of manufacture, particularly with regard to the synthetic route and starting materials.2 However, enantiomer determination is not widely performed in many laboratories, in part owing to the lack of instrumentation (i.e., a polarimeter).Classically, the enantiomeric composition has been assessed by observing the precipitation of crystal forms by optical microscopy following the reaction of methamphetamine with certain metal salts.3 Methods have since been developed to resolve the enantiomers by NMR4,5 or chromatographic techniques,6–8 including capillary electrophoresis.9,10 Many of these methods rely on the preparation of diastereoisomeric derivatives of the enantiomers.An infrared approach has also been employed to distinguish enantiomerically enriched samples from racemic mixtures of methamphetamine by derivatization with phenyl isothiocyanate, 11,12 an achiral reagent.A more rapid and simple infrared technique is discussed in this paper, which may have wider application. One feature of many chiral substances is the formation of distinct crystalline phases for a single enantiomer and for a racemic mixture of the two enantiomers. This enables the two types of samples to be readily distinguished by various physical properties, including the condensed phase infrared transmission spectrum. Unfortunately, and contrary to the majority of chiral substances,13 some salts of amine drugs, including methamphetamine, do not form racemic crystals.Instead, a mechanical mixture (or conglomerate sample) results in which an equimolecular mixture of enantiomeric crystals of the two isomers is present. Since the two enantiomeric crystal forms possess identical infrared spectra, the racemic mixture presents the same spectrum as that observed for either enantiomer. This behavior prevents infrared transmission spectrometry from distinguishing between enantiomerically enriched and racemic samples of the prevalent chemical form of methamphetamine, the hydrochloride salt.In this paper, a simple procedure is outlined that induces the formation of a racemic phase for methamphetamine, and thereby allows for enantiomerically enriched and racemic samples to be rapidly distinguished and confirmed by infrared spectrometry. This approach relies on the solid-state, ionexchange reaction methamphetamine·HCl + KI ? methamphetamine·HI + KCl (1) which occurs between methamphetamine hydrochloride and potassium iodide when the two powders are mixed and pressure-sintered into a transmission window for infrared measurements. We have previously reported that the hydrochloride salt of the dextrorotatory isomer of methamphetamine is reactive with alkali metal halide matrices,14 whereby the infrared spectrum observed for the amine salt can vary dramatically depending on the anion of the matrix material.It was found that the exchange of iodide ion for chloride was particularly favored for methamphetamine and some structurally related analogs (i.e., amphetamine and some ringsubstituted analogs, including the 3,4-methylenedioxyamphetamines). This phenomenon was recently found also to occur with a racemic sample of methamphetamine, except that the racemic hydriodide salt forms a liquid phase distinct in both physical appearance and infrared spectrum from that of the single enantiomer.Consequently, this behavior allows enantiomerically enriched and racemic samples of methamphetamine hydrochloride to be easily distinguished by infrared spectrometry following sample preparation within a potassium iodide matrix. This chemistry also enables the predominant isomer of an enantiomerically enriched sample to be determined since a racemic sample of methamphetamine hydrochloride may be prepared by blending equal amounts of the two enantiomers.Experimental Sample Preparation The preparation of samples for this method is no different from preparing an infrared transmission window with KCl or KBr, other than using KI as the matrix material. A sample of Analyst, August 1997, Vol. 122 (755–760) 755methamphetamine hydrochloride is ground (generally with a small agate mortar and pestle) to produce a fine powder, followed by grinding an appropriate amount of KI in with the methamphetamine salt. In this study, approximately 2–4 mg of the methamphetamine salt were used with sufficient matrix material (100–200 mg) to prepare a pressed transmission window within a 5 mm 320 mm rectangular opening cut within a blotter card.The KI solid should be ground well and mixed thoroughly with the methamphetamine sample for 30–60 s to ensure that nearly complete ion exchange occurs and methamphetamine hydriodide is formed. This composite powder is then pressed into an infrared transmission window under a load of 12 000–15 000 lb (typically with a Carver press) with or without application of a vacuum to the sample.Windows prepared with KI are generally not as visually transparent as KBr windows, but can still provide a suitable infrared transmission spectrum. The potassium halide salts employed in this study were of either analytical-reagent grade (KCl and KI) or spectroscopic grade (KBr), and gave featureless spectra throughout the mid-infrared region (4000–400 cm21).A Nicolet (Madison, WI, USA) Model 205 Fourier transform infrared spectrometer was used to record all spectra, operating at 32 scans per spectrum with a 4 cm21 resolution. This analysis requries that the sample be sufficiently free of diluents or other contaminants so that the infrared spectrum of the methamphetamine salt is clearly identifiable. Most common diluents are easily removed by dry extraction techniques to give infrared-pure methamphetamine·HCl.15 Complete identification of the predominant isomer for enantiomerically enriched samples further requires a standard for one of the enantiomers to blend with the unknown sample in order to test for the formation of the racemic form.Preferably, a standard of the levorotatory isomer (l or 2 notation) of methamphetamine·HCl would be mixed with a suspected sample of the dextrorotatory isomer (d or +) to observe the spectral change to the racemic form. Approximately equal amounts of the two should be mixed together prior to thoroughly grinding a suitable amount of the KI matrix material with the sample.The spectrum will clearly display the characteristics of the racemic HI salt if the opposite enantiomer was introduced to the original enantiomerically enriched sample. Owing to the inexact nature of weighing and mixing of equal parts of the two enantiomers and the likely presence of a small excess of one isomer, weak spectral features of the enantiomeric HI salt may also be present.Comparative Studies Enantiomerically enriched (d-isomer) and racemic (d,l-isomers) samples of amphetamine and N,N-dimethylamphetamine in addition to methamphetamine were considered in this study. The molecular structures of these compounds are depicted in Table 1. All drug samples were obtained as pharmaceutical grade standards with a high purity ( > 99%). The standards were in the form of the HCl salt, with the exception of damphetamine as the sulfate salt.For comparison purposes, the HI salts were also prepared by the complexation of the respective amine bases with hydriodic acid in concentrated aqueous solutions. The amine bases were formed by the neutralization of the respective HCl or sulfate salts with concentrated ammonia in aqueous solution, followed by extraction of the base with methylene chloride. The solvent of this extract solution was removed by evaporation and the base was collected as an oily liquid. The amine bases were then dissolved in concentrated aqueous solutions of hydriodic acid (approximately 5 mol l21) and the ion pair was extracted with chloroform.This extract was passed through a column of anhydrous sodium sulfate to remove water (anhydrous sodium sulfate is an effective drying agent for chloroform extract solutions, and there is no detectable ion exchange of sulfate for the halide of the soluble amine salt), and the HI salt was recovered by solvent evaporation of the chloroform solution over a steam-bath.The HCl salt of d-amphetamine was prepared similarly with the use of concentrated hydrochloric acid. Most of these salts were subsequently recrystallized as small white crystals with an elongated habit from saturated acetone solutions upon addition of diethyl ether. Relatively sharp melting-points were measured for all of the salt forms with the exception of the racemic samples of amphetamine·HI and methamphetamine·HI, which would only form an oily, liquid material.The melting-points of the recrystallized salts are given in Table 1. All solvents and chemicals used in the formation and recrystallization of the salts were of analyticalreagent grade. Results and Discussion The infrared spectrum of an enantiomerically enriched sample of methamphetamine·HCl (d-isomer) prepared within a KI matrix is illustrated in Fig. 1. Comparative spectra of dmethamphetamine ·HCl (in the unreactive KCl matrix) and dmethamphetamine ·HI (in KI) are also shown to illustrate the conversion that the HCl salt undergoes when prepared in KI.The changes in the spectrum are dramatic, where the spectrum of the HCl salt in KI essentially assumes the same spectral features of the HI salt and loses the characteristics of the original hydrochloride salt. The most salient differences are found in the shape of the strong absorption envelope within the hydrogen stretching region between 3200 and 2600 cm21. The strong absorbance in this region is primarily attributed to the stretching motions of the protons bound to the amine nitrogen,16 which possess considerable oscillator strength since they modulate the intrinsically large dipole of the ion pair of the amine salt.This absorption envelope also displays a complex shape that can be particularly sensitive to changes within the crystalline structure of the salt, including changes in the anion.14 In conjunction with these spectral differences, a single strong peak at 2461 cm21 is observed for the HCl salt (in KCl) that is entirely absent for the HI salt.Numerous spectral differences also exist throughout the fingerprint region (2000–400 cm21), especially with the appearance of a single strong band at 1548 cm21 and a set of three sharp peaks near 810 cm21 in the spectrum of the HI salt. Spectra for a racemic sample of methamphetamine·HCl (d,lisomers) are shown in Fig. 2. Analogously to Fig. 1, the Table 1 Melting-points (°C) for N-methyl-substituted amphetamine salts Enantiomerically enriched Racemic Halide salt sample* sample Amphetamine (R1 = R2 = H) HCl 154–155 146–149† HI 134–136 Oily liquid‡ Methamphetamine (R1 = H, R2 = CH3) HCl 170–172 133–133 HI 97–99 Oily liquid‡ N,N-Dimethylamphetamine (R1 = R2 = CH3) HCl 181–183 156–157 HI 146–147 119–121 * The dextrorotatory isomer (S configuration) is the predominant enantiomer.† Amphetamine·HCl forms a racemic crystal. ‡ Sample exists as a liquid phase at ambient conditions ( ~ 22 °C, ~ 30% relative humidity). 756 Analyst, August 1997, Vol. 122spectrum of the HCl salt is observed to change greatly between the preparations in KCl and KI. The spectrum of the racemic HCl salt in KI displays a steep slope in its baseline, and also exhibits peaks that are notably broader than those of the preparation in a KCl matrix. The broadened peaks correspond well with the features in the spectrum of the racemic HI salt, and are consistent with the amorphous structure of a liquid-like state, as opposed to the long range translational order of the crystalline HCl salt.The strong slope in the baseline is also characteristic of a liquid (or waxy) sample, which inhibits sintering of the matrix material, and results in a visually opaque transmission window with strong background scattering. The transformation to a liquid phase for a racemic sample may actually be recognized in the course of sample preparation as the KI matrix is ground into the racemic HCl salt and the powder can clump as if it were wet.This behavior, unfortunately, does not give a very clear transmission window, but the spectrum is distinct and strikingly different from the spectrum of the enantiomerically enriched sample. The ion-exchange reaction is dependent on the amount of time during which the powders of the HCl salt and KI matrix are in contact, although grinding the two solids together for 30 s appears to allow for essentially complete conversion to the HI salt.The pressure employed to sinter the matrix into a transmission window has no observable effect on the reaction rate. Ion exchange is observed to occur to a lesser extent between methamphetamine·HCl and a KBr matrix, yielding a mixture of the HCl and HBr salts in the transmission window.14 This effect may account for some observations that the spectra for enantiomerically enriched and racemic samples of methamphetamine ·HCl can show some weak differences distinguishing the two.17 The spectra of the enantiomerically enriched and racemic samples of the HBr salt exhibit some real but subtle differences, especially in the shape of the absorption envelope near 2800 cm21 and the series of peaks between 1100 and 1000 cm21.These differences, however, are very weak in the composite spectrum that results from a preparation of the HCl salt in KBr, and would not be a good gauge by which to perform the enantiomer determination.The formation of the HI salt allows enantiomerically enriched and racemic samples of methamphetamine to be distinguished, but an additional step is required to determine the predominant isomer of enantiomerically enriched samples (see Sample Preparation). This approach contrasts a general technique involving the reaction of chiral amines with chiral organic acids, and the prospective formation of chiral salt forms that can possibly distinguish the two enantiomers, in addition to the racemic mixture. Reaction with d-mandelic acid has proved particularly successful for amphetamine, where distinct crystalline phases precipitate for the d- and l-isomers as well as the racemic mixture, and all three salt forms are readily distinguished by infrared spectrometry.18 Unfortunately, mandelic acid does not precipitate any usable salt forms with methamphetamine, nor have any other chiral organic acids been reported to resolve the enantiomers of methamphetamine reliably. The ion-exchange reaction is observed to occur with other amine salts, and specifically with some structurally related analogs of methamphetamine.Observations made on amphetamine and N,N-dimethylamphetamine (and also ephedrine and pseudoephedrine) have also revealed a propensity for the hydrochloride salts to react spontaneously and nearly completly with a KI matrix within 1 min. The corresponding HI salts, however, have been found not to react significantly with a KCl (or KBr) matrix, even with pressed transmission windows stored for many weeks.This chemistry appears at first to suggest that the hydriodide phases are more stable than the hydrochloride, presumably from a greater covalent character to the ion-pair bond between the protons of the amine nitrogen and the iodide anion. This bond, however, is generally recognized to strengthen with the replacement of iodide with chloride in amine salts,16 owing to the stronger ionic attraction between these protons and the chloride anion.The interaction creates a Fig. 1 Comparison of the infrared transmission spectra for enantiomerically enriched (d-isomer) methamphetamine hydrochloride prepared in KCl and KI with the spectrum of the hydriodide prepared in a KI matrix. Fig. 2 Comparison of the infrared transmission spectra for racemic (d,lisomers) methamphetamine hydrochloride prepared in KCl and KI with the spectrum of the hydriodide observed neat between KBr discs.Analyst, August 1997, Vol. 122 757hydrogen bond that links the protonated amine to the halide anion as an ion pair.19–21 The hydrogen bond within the methamphetamine·HCl ion pair is therefore probably stronger than that in the HI salt, which implies a greater stability for the enantiomeric crystal of the HCl salt. Consequently, the driving force behind the ion-exchange reaction is probably determined by other binding preferences, and specifically, a greater affinity for the chloride anion (relative to the iodide) to bind to the hard potassium cation of the matrix.The soft iodide anion exhibits a significantly weaker binding preference than chloride, and is therefore relegated to pairing with the protonated amine. The matrix effects for enantiomerically enriched and racemic samples of amphetamine·HCl in KI are illustrated by the spectra in Figs. 3 and 4, respectively. The behavior observed is analogous to that of methamphetamine·HCl, where the enantiomerically enriched sample precipitates a crystalline HI salt (Fig. 3) whereas the racemic sample converts into a liquid hydriodide phase (Fig. 4). The spectra for the enantiomerically enriched sample in Fig. 3 illustrate an essentially complete conversion of the HCl salt into the HI salt when prepared in a KI matrix. The spectral features of the HI salt are also distinguishable from the HCl salt (in KCl) with several small yet significant differences in the shape of the amine proton stretching envelope (between 3000 and 2400 cm21), and also with the absence of the broad combination band near 2000 cm21 that is present for the HCl salt.There are also many peak differences throughout the fingerprint region for the enantiomerically enriched sample. The spectra for the racemic sample in Fig. 4 likewise show a significant change in the spectral features for the HCl salt when prepared in KI, although the difference is primarily one of broadened peak shapes due to the liquid nature of the racemic HI salt.These features correspond fairly well with those in the spectrum for a racemic sample of amphetamine·HI and, like methamphetamine, offer a clear contrast from that observed for an enantiomerically enriched sample. Unlike methamphetamine, enantiomerically enriched and racemic samples of amphetamine ·HCl (in KCl) are also distinguishable since the HCl salt of d,l-amphetamine forms a racemic crystal with a distinctive infrared spectrum.In complete contast, neither the hydrochloride nor the hydriodide salts of N,N-dimethylamphetamine form a racemic phase that can be distinguished by infrared spectrometry from the enatiomerically enriched sample. The spectra for a racemic sample of N,N-dimethylamphetamine are shown in Fig. 5, where, like amphetamine and methamphetamine, the HCl salt reacts with a KI matrix to give a spectrum equivalent to that of the HI salt. The spectrum of the HCl salt (in KCl) is also different from that of the HI salt (in KI), especially in the appearance of the absorption envelope between 3200 and 2400 cm21.The series of spectra in Fig. 5 are absolutely equivalent to those observed for the corresponding preparations of an enantiomerically enriched sample, indicating that racemic samples of both the HCl and HI salts are purely mechanical mixtures. This observation raises some question as to the nature of the liquid phase observed for the racemic hydriodide salts of methamphetamine and amphetamine.An equimolecular mixture of enantiomeric crystals for a racemic sample can experience a significant depression of its melting-point due to the solubility of each enantiomer in the melt of the other enantiomer. The depression of the meltingpoint of a mechanical mixture of enantiomers is well approximated by the expression13 ln x H R T T x f A » - æ è ç ö ø ÷ < D 1 1 1 f A f for 0.5 < (2) where x is the mole fraction of the more abundant enantiomer, Tf is the temperature at which melting is complete for the mixture, DHf A and Tf A are the heat of fusion and melting-point, respectively, for the pure enantiomer (x = 1) and R is the gas constant (1.987 cal K21 mol21). This equation describes the liquid–solid equilibrium in the phase diagram for a binary mixture of enantiomers, where the racemic mixture (x = 0.5) Fig. 3 Comparison of the infrared transmission spectra for enantiomerically enriched (d-isomer) amphetamine hydrochloride prepared in KCl and KI with the spectrum of the hydriodide prepared in a KI matrix.Fig. 4 Comparison of the infrared transmission spectra for racemic (d,lisomers) amphetamine hydrochloride prepared in KCl and KI with the spectrum of the hydriodide observed neat between KBr discs. 758 Analyst, August 1997, Vol. 122exists as a eutectic, the lowest melting-point observed among the mixtures.The derivation of this equation assumes several conditions, the most significant requiring that the enantiomers exhibit no heat of mixing in the liquid or melt state (i.e., ideal behavior). Although at first glance this may appear to be a serious restriction, in fact the heats of mixing are often observed to be less than 1% of the heat of fusion and, in general, eqn. (2) has been found to be highly reliable in the description of the melting behavior of enantiomeric mixtures.13 The melting-points measured for the salts in this study (Table 1) reveal a depressed melting-point (relative to the enantiomerically enriched samples) for all racemic samples, including the racemic crystal of amphetamine·HCl.It should also be noted that for all three amine compounds, the enantiomerically enriched HI salts melt at a considerably lower temperature than the HCl salt of the same amine. This lower melting point indicates that the enantiomeric crystal of the HI salt is less stable relative to its melted state compared with that of the HCl salt and its melt.This effect is reflected by a significant decrease in the heat of fusion for the HI salt relative to the HCl salt, as can be illustrated by N,N-dimethylamphetamine. A value of 10.8 kcal mol21 can be estimated from eqn. (2) for the heat of fusion of the enantiomeric crystal of the HCl salt, whereas that of the HI salt is calculated to be near 8.4 kcal mol21. This decrease arises from weaker cohesive interactions in the crystalline packing of the enantiomeric crystal of the HI salt relative to that of the HCl salt.The large iodide anion probably disrupts the close packing that is possible with the smaller chloride anion. A decrease in the stability of the enantiomeric crystal of the HI salts has important implications for the stability of a mechanical mixture of methamphetamine·HI within a racemic sample. The heat of fusion for the enantiomeric crystal of methamphetamine·HCl is estimated to be 6.4 kcal mol21 from eqn.(2), considerably lower than that of either salt of N,Ndimethylamphetamine. A value of 5 kcal mol21 for the heat of fusion of the HI salt implies a low melting-point near 63 °C for the racemic mixture, and the heat of fusion for methamphetamine ·HI could plausibly be lower, yielding an even lower meltingpoint. Although it is unlikely that the racemic melting-point is actually depressed below room temperature (DHf A Å 2 kcal mol21), the stability of a racemic mixture of enantiomeric crystals of the HI salt appears not to be much greater than that of the melt at room temperature.The stability of the melt phase can also be increased entropically by the inclusion of impurities, and may be particularly enhanced by the presence of dissolved moisture, which is excluded from the packing structure of the enantiomeric crystal. This effect may allow a hydrated liquid state to become metastable, if not entirely stable, at room temperature.A significant level of water is in fact indicated to be dissolved within the racemic liquid phase, as is evident from the infrared spectrum of the racemic HI salt in Fig. 2 and the large water peak near 3400 cm21. The spectrum of the racemic HCl salt prepared in KI also exhibits a sizeable peak in this region, suggesting an increased level of moisture within the KI preparation compared with that in KCl. Moisture available from the atmosphere, and also possibly adsorbed on the KI reagent, may therefore help stabilize the liquid state of the racemic HI salt at room temperature relative to the enantiomeric crystal.This description appears consistent with observations made on mechanical mixtures prepared from the enantiomeric crystals of the d- and l-isomers of methamphetamine·HI. Preparations made by simply mixing equal portions of powders of the HI salts exhibit a melting-point near 60 °C, and remain solid while stored in a dry environment over phosphorus pentoxide (P2O5).These solid mixtures appear stable for several weeks when stored over P2O5, but are observed to liquefy within 24 h upon standing in a room atmosphere (approximately 22 °C, approximately 30% relative humidity). The liquefaction of the racemic HI salt is also reversible within a dry environment, even though it is a much slower process. Pressed transmission windows containing the liquid racemic HI salt (the HCl salt in KI and the HI salt in KCl) are observed slowly to develop weak spectral features of the enantiomeric HI salt over the course of several weeks when stored over P2O5.This observation indicates that a mechanical mixture of enantiomeric crystals is actually the stable state for the racemic HI salt in the absence of absorbable moisture. The dehydration of the liquid HI salt within a dry environment and the subsequent precipitation of the enantiomeric crystal are extremely slow processes, however, and no racemic sample has yet been found to convert completely into an infrared-pure transmission spectrum of the enantiomeric crystal.Similar behavior is also observed for transmission windows containing racemic amphetamine·HI, although the crystallization of the enantiomeric HI salt within the matrix appears to progress faster than that for methamphetamine ·HI. Interestingly, the enantiomeric crystal of N,N-dimethylamphetamine ·HI is sufficiently stable to ensure that a mechanical mixture is the ground state for a racemic sample at ambient conditions, even when atmospheric moisture is available for hydration of the liquid phase.In summary, this method for the enantiomer determination of methamphetamine relies on a reduced stability for the enantiomeric crystal of the HI salt relative to the cohesive energy of the HCl salt. This destabilization of the HI salt is apparently sufficient to make a racemic mixture of enantiomeric crystals unstable towards liquefaction under ambient conditions, whereby atmospheric moisture appears to enhance further the stability of the melt phase through hydration.This effect is also observed for amphetamine, and is probably intrinsic to many other low molecular mass amine salts. The appearance of matrix effects with KI may then be possibly exploited as a means of enantiomer determination. Some care, however, should be exercised when interpreting spectral differences between enantiomerically enriched and racemic samples of a particular amine salt.Spectral differences may also arise due to variable rates for the ion-exchange reaction in the two samples, even Fig. 5 Comparison of the infrared transmission spectra for racemic (d,lisomers) N,N-dimethylamphetamine hydrochloride prepared in KCl and KI with the spectrum of the hydriodide prepared in a KI matrix. Analyst, August 1997, Vol. 122 759when both samples are converting into the same crystalline phase. Several factors can affect the rate of ion exchange and the subsequent precipitation of a salt with the anion of the matrix; most notable are the average particle size (and size distribution) and the amount of adsorbed moisture over the particle surface. Particle size dictates the area of the contact interface between grains within the matrix, and hence the size of the reaction zone for ion exchange. The moisture content of a powder sample has also been found to affect greatly the rate of ion exchange, since the water adsorbed to the particle surface apparently activates the reaction, presumably by ‘solvating’ the ions at or near the interface between grains. For the amines considered here and their reaction with KI, ion exchange is fairly rapid and conversion to the hydriodide salt is essentially complete within 1 min, so that the rate of the reaction does not become an issue in the spectral interpretation. References 1 Proposed 1995 Guideline Amendments, 60 Federal Register 25074. 2 Allen, A., and Cantrell, T. S., Forensic Sci. Int., 1989, 42, 183. 3 Official Methods of Analysis of the Association of Official Analytical Chemists, ed. Horwitz, W., Association of Official Analysis Chemists, Arlington, VA, 10th edn., 1965, p. 597. 4 Kram, T. C., and Lurie, I. S., Forensic Sci. Int., 1992, 55, 131. 5 LaBelle, M. J., Savard, C., Dawson, B. A., Black, D. B., Katyal, L. K., Zrcek, F., and By, A. W., Forensic Sci. Int., 1995, 71, 215. 6 Wells, C. E., J. Ass. Off. Anal. Chem., 1970, 53, 113. 7 Noggle, F. T., and Clark, C. R., J. Forensic Sci., 1986, 31, 732. 8 Makino, Y., Ohta, S., and Hirobe, M., Forensic Sci. Int., 1996, 78, 65. 9 Lurie, I. S., J. Chromatogr., 1992, 605, 269. 10 Lurie, I., Klein, R., Dal Cason, T., LaBelle, M., Brenneisen, R., and Weinberger, R., Anal. Chem., 1994, 66, 4019. 11 Allen, A., and Cooper, D., Microgram, 1979, 12, 24. 12 Shriner, R. L., Fuson, R. C., and Curtin, D. Y., The Systematic Identification of Organic Compounds, Wiley, New York, 1964, pp. 256–259. 13 Jacques, J., Collet, A., and Wilen, S. H., Enantiomers, Racemates, and Resolutions, Krieger, Malabar, FL, 1994, pp. 43–48. 14 Chappell, J. S., Forensic Sci. Int., 1995, 75, 1. 15 Ely, R. A., Drug Enforcement Administration, San Francisco, CA, USA, personal communication, 1993. 16 Colthup, N. B., Daly, L. H., and Wiberly, S. E., Introduction to Infrared and Raman Spectroscopy, Academic Press, New York, 1964, pp. 281–282. 17 Heagy, J. A., Drug Enforcement Administration, San Francisco, CA, USA, personal communication, 1996. 18 Heagy, J. A., Anal. Chem., 1970, 42, 1459. 19 For review, see Pauling, L., The Nature of the Chemical Bond, Cornell University Press, Ithaca, NY, 3rd edn., 1960, pp. 449–504. 20 Chenon, B., and Sandorfy, C., Can. J. Chem., 1958, 36, 1181. 21 Brissette, C., and Sandorfy, C., Can. J. Chem., 1960, 38, 34. Paper 7/00122C Received January 6, 1997 Accepted April 15, 1997 760 Analyst, August 1997, Vol. 122
ISSN:0003-2654
DOI:10.1039/a700122c
出版商:RSC
年代:1997
数据来源: RSC
|
7. |
Strategies for Constructing the Calibration Set in theDetermination of Active Principles in Pharmaceuticals by Near InfraredDiffuse Reflectance Spectrometry |
|
Analyst,
Volume 122,
Issue 8,
1997,
Page 761-765
M. Blanco,
Preview
|
|
摘要:
Strategies for Constructing the Calibration Set in the Determination of Active Principles in Pharmaceuticals by Near Infrared Diffuse Reflectance Spectrometry M. Blanco*, J. Coello, H. Iturriaga, S. Maspoch and C. de la Pezuela Departamento de Qu�ýmica, Unidad de Qu�ýmica Anal�ýtica, Universidad Aut�onoma de Barcelona, E-08193 Bellaterra, Barcelona, Spain The active principle in the blended phase of a commercially available pharmaceutical preparation was determined using near infrared diffuse reflectance spectrometry in combination with a fibre optic probe and multivariate calibration by partial least-squares regression. Two different ways of preparing laboratory samples spanning an appropriate concentration range for constructing the calibration set were compared.One of the procedures involves preparing synthetic samples by weighing and the other using under- and overdosed production samples. Although the results provided by the two strategies were not significantly different, the second was judged more effective because it has a less marked effect on those physical properties of the samples that affect the IR spectrum.The prediction errors obtained (less than 1%) indicate the suitability of the proposed sample preparation procedure, which is faster than the usual method of choice and provides comparable results. Keywords: Miokamycin; near infrared spectrometry; pharmaceutical; fibre optic probe; calibration for quantitative analysis; partial least squares regression Near infrared diffuse reflectance spectrometry (NIRRS) has developed very rapidly since its inception, so much so that it is currently a frequent choice for the analysis of agricultural foods1 and a powerful tool for control analyses in the textile,2 polymer,3 biomedical4 and pharmaceutical fields,5 amongst others.The dramatic spread of the NIRRS technique can be ascribed to its suitability for qualitative and quantitative determinations of physical and chemical parameters, its ability to process solid samples with minimal or no manipulation and its compatibility with fibre optic probes as signal transmitters, which facilitates analyses.The NIRRS technique has aroused special interest in the pharmaceutical industry, where it is commonly used for the identification of both raw materials and end products,6–8 and in moisture measurements;9,10 however, it is equally applicable to finished products,11–13 with the advantages inherent in its nondestructive, expeditious, non-invasive nature. Although the use of a fibre optic probe minimizes sample manipulation by the analyst, the loss of energy in travelling through fibres increases signal noise and consequently decreases the signal-to-noise ratio, which can have a significant adverse effect on quantitative analyses.Despite this, fibre optic probes can still be suitable for quantitative purposes and provide accuracy and precision similar to those of the traditional spectral recording procedure.14 Because the NIRRS signal is dependent on both the physical and chemical properties of the sample, differences between spectra for a given preparation can arise from changes in both the physical properties of the samples (particle size, grain size distribution, compactness, etc.) and the concentrations of their components.One of the hindrances to the use of NIRRS in quantitative pharmaceutical analyses is the difficulty in obtaining an appropriate sample set for constructing the calibration matrix; in fact, production samples contain concentrations that vary only slightly around central values that are close to the stated contents, so they hardly encompass a wide enough range for appropriate calibration. Expanding such a narrow concentration range without altering other factors that influence spectra entails using an effective strategy to prepare synthetic or dosed samples (e.g., manufactured samples on pilot-scale equipment;15,16 samples prepared at the laboratory by mixing all the pharmaceutical ingredients over concentration ranges close to the manufacturer’s stated contents;17 or a procedure for preparing underand overdosed samples from different production batches18,19).The first strategy probably gives the calibration samples with the closest resemblance to production samples; however, no new batches can be produced as the need for calibration arises, nor can one assure that the samples thus obtained will match production samples in every respect.The second procedure has the advantage that the different analyte concentrations required to compile the calibration set are easy to obtain; however, the laboratory procedures used to grind and mix ingredients, for example, can vary markedly from those employed in the production chain, so the physical properties of the two types of sample, and hence their NIRRS spectra, may differ.The use of a calibration set containing normal production samples in addition to under- and overdosed production samples places extra demands on the calibration and so the production of synthetic or dosed samples should pay due regard to physical properties and, provided strict control is exercised, inadvertent differences between production and synthetic/dosed samples can be minimised and accommodated. This paper compares the performance of two strategies for preparing laboratory samples spanning an appropriate concentration range for constructing a calibration set.One involves making synthetic samples by weighing the required amount of each ingredient; the other uses under- and overdosed samples from different production batches. The active principle in the blended phase of the pharmaceutical preparation assayed was determined by partial least-squares regression (PLSR)20, using a calibration set consisting of laboratory samples prepared as described above that was expanded with untreated samples from different production batches in order to consider the variability inherent in the manufacturing process.Experimental Samples and Reagents Miokamycin (9,3A-diacetylmiodecamycin) is a macrolide antibiotic derived from miodecamycin A1. The amorphous form, containing 0.2% hydroxypropylmethylcellulose, is the active principle of a pharmaceutical that is commercially available in tablets. Microcrystalline cellulose, magnesium stearate, carboxymethyl starch, methylpropylcellulose and aluminium glycinate are the excipients.Analyst, August 1997, Vol. 122 (761–765) 761The determination of miokamycin which has a nominal content of 640 mg g21 in the product, is performed in the blended phase before compression. Four samples each from 19 different production batches were selected over a two-month period. These were subsequently allocated to samples subsets for use in developing and testing the NIRRS calibration (see Results and Discussion). The reagents used included P.A.grade methyl alcohol and 1 m hydrochloric acid from Panreac (Barcelona, Spain). The active principle, excipients and production samples were all supplied by Laboratorios Menarini (Badalona, Spain). Apparatus NIRRS spectra were recorded on a NIRSystems 6500 near infrared spectrophotometer equipped with a reflectance detector and a fibre optic probe (model AP6645 ANO3P) for qualitative and quantitative work.The instrument was controlled via the software NSAS v. 3.20, which includes routines for spectral acquisition and processing. The UV spectra used in the reference method were recorded on a Hewlett–Packard HP 8452A diode array spectrophotometer. The instrument’s bundled software, HP 89530 MSDOS UV/VIS, includes facilities for controlling it and acquiring and processing spectra. An Alresa centrifuge and a Shaker Mixer TURBULA Type T2 C from WAB (Basel, Switzerland) were also used. Software Multivariate calibration was performed with the Unscrambler v. 5.03 software package, from CAMO A/S (Trondheim, Norway), which affords principal components regression (PCR) and PLSR, with selection of variables and detection of outliers. The NIRRS spectra recorded were exported from the instrument to Unscrambler with the aid of Perconv v. 2.05, also from CAMO A/S. The multicomponent analysis software MC, developed nidad de Qu�ýmica Anal�ýtica at the Universidad Aut�onoma de Barcelona, was used in the UV analysis.It allows the concentrations of mixture components to be determined by multiple linear regression (MLR) fitting of the mixed spectrum from the spectra for the pure components. Preparation of Laboratory Samples The concentration range of the calibration set was expanded in two ways. One involved preparing 20 samples by weighing variable amounts of the active principle and the five excipients over a range spanning concentrations within 10% above and below the stated contents. The other involved preparing a set consisting of 21 over- or underdosed samples by adding a known amount of the active principle or a mixture of the excipients, respectively, to an also known amount of preparation.In both cases the resulting mixture weighed about 8 g and was homogenized in the shaker mixer before its NIRRS spectrum was recorded. Complete mixing was checked using the method proposed by Ciurczak.21 The samples obtained with the former and latter procedure will henceforward be referred to as ‘synthetic samples’ and ‘dosed samples’; those supplied by the manufacturer were labelled ‘production samples’. UV Reference Procedure The following UV spectrophotometric procedure was used as reference in the determination of the active principle (miokamycin) in dosed and production samples: about 0.5 g of sample was dissolved in 80 ml of methanol, stirred magnetically for 15 min and made to 100 ml with the same solvent. An aliquot of the resulting solution was centrifuged at 3000 rpm for 5 min and 5 ml of the supernatant were diluted to 50 ml with methanol.To 5 ml of this solution 2.5 ml of 1 m HCl were added and the solution made up to 50 ml with methanol. Finally, the spectrum for this last solution was recorded between 200 and 350 nm against a blank consisting of 1 + 20 1 m HCl–methanol. The programme MC was used to quantify the samples from their first-derivative spectrum over the wavelength range 220–270 nm, using the first-derivative spectrum for miokamycin as the sole standard (Fig. 1). Recording of NIRRS Spectra Spectral scanning was performed by introducing the tip of the fibre optic probe into the same wide mouth polyethylene bottle in which the sample was weighed and mixed, turning over the sample with a spatula between successive scans. Each sample was used to record three spectra over the wavelength range 1100–2500 nm.The three spectra thus obtained were averaged and the resulting spectrum was employed to quantify the active principle in the blended phase of the pharmaceutical preparation. Production batches for inclusion in the calibration set were selected from the mean spectrum for each batch, which was obtained from three spectra for each of the four samples in the batch. Fig. 2 shows the absorbance and first-derivative NIRRS spectra for the active principle and a sample of the pharmaceutical preparation.Data Processing The active principle concentration used as reference was the weighing value for synthetic samples and the mean value of two determinations with the above-described UV reference procedure for dosed and production samples. To assure that there were no significant differences between the weighing and UV reference methods, three synthetic samples were prepared at low, medium and high concentration levels and three aliquots of each one were analysed by the UV reference procedure.Differences between weighed and UV found values were lower than ±0.6% in all cases. The error of the UV reference method was established from the following expression: SDD = (di - dm)2 i=1 nå n -1 (1) Fig. 1 First-derivative UV spectra for (1) a 20.0 mg l21 miokamycin solution and (2) a solution of the pharmaceutical preparation containing 49.3 mg l21 in 1 m HCl–methanol (1 + 20). 762 Analyst, August 1997, Vol. 122where SDD is the standard deviation of difference, di the difference between two replicates for sample i, dm the mean of all the differences and n the number of samples used.22 The obtained value was 5.2 mg g21, 0.8% of the nominal content of miokamycin.All PLSR models were constructed by cross-validation, using as many segments as samples were in the calibration set. The optimum number of PLS components for quantitation was estimated on the basis of the statistical criterion of Haaland and Thomas.23 The goodness of the results provided by the different PLSR models studied was assessed in terms of the relative standard error of prediction, %RSEP:24 % ( ) RSEP c c c i i i i n i n = - � = = å å LAB NIRRS LAB 2 1 2 1 100 (2) where cLABi and cNIRRSi are the concentration of the active principle provided by the reference method and that calculated by the PLSR model, respectively, for sample i, and n is the number of samples.In order to ensure that the results provided by the reference method and the proposed NIRRS method were not significantly different, a t-test for paired data was applied to the results for the production samples in sub-sets P2 and P3 (see Results and Discussion), and the error for the reference method, SDD, was compared with the standard error of prediction, SEP, for the samples in the prediction set, defined as SEP c c n i i i n = - - = å ( ) LAB NIRRS 2 1 1 (3) where CLABi and CNIRRSi are the concentrations of the active principle provided by the reference method and calculated by the PLSR model, respectively, for sample i, and n is the number of samples.SEP must be roughly equal to SDD if the reference and proposed method are to provide similar results.22 Results and Discussion The calibration set used to determine the active principle should contain normal production samples from different batches in addition to synthetic or dosed samples in order to ensure that the set is representative of the most common sources of variability in the manufacturing process (particle size, moisture, supplier of the raw materials, etc.).Using a mixed calibration matrix affords the implementation of methodology that quantifies concentrations close to the stated content (640 mg/g) or falling within the manufacturer’s accepted range with the same accuracy. Because NIRRS spectra were recorded directly from the solid samples, the scattering observed must be minimized by using some mathematical treatment.The first spectral derivative is the readiest choice in this respect and has the advantage that it cancels baseline shifts; however, the second derivative is more widely used and allows correction of constant variations, at the expense of increased noise. On the other hand, while the use of a fibre optic probe considerably facilitates recording of spectra, it also raises signal noise such that the 2200–2500 nm wavelength range is rendered essentially unusable in practice.In order to select production batches representative of the variability sources, which should thus be included in the calibration set, the mean second-derivative spectra over the range 1100–2200 nm for 14 randomly chosen batches among the 19 available were subjected to principal component analysis (PCA);20 the other 5 production batches were not included in the PCA in order that they could form a completely independent sample set to test the robustness of the calibration.Based on the results, the first two principal components (PCs) accounted for 88.8% of the variance. Fig. 3 shows a plot of score PC1 against PC2; as can be seen, the production batches clustered in three blocks, which suggests differences between them. Such differences cannot be ascribed to the concentration because all the values were close to the stated content (641.0–662.5 mg g21), but rather to slight changes in the physical properties of the samples that influence the NIRRS spectrum.Therefore, for a calibration set to include variability sources inherent in the manufacturing process as far as possible, it should contain a sample from at least each of the production batches representing the extremes of variability, e.g., those in boldface in Fig. 3. To determine the active principlng the synthetic samples four sets were used: a calibration set (C1) consisting of 20 samples (7 production samples from 7 different batches and Fig. 2 Absorbance and first-derivative NIRRS spectra for (1) the active principle and (2) a sample of the commercially available preparation. Fig. 3 Plot of scores for the first and second PC for the production batches. Production batches included in (filled circles) and excluded from (open circles) the calibration set. Analyst, August 1997, Vol. 122 76313 synthetic samples) and three prediction sets, viz. P1 (7 synthetic samples), P2 (49 production samples from 14 different batches) and P3 (20 production samples from 5 other batches). The four sets used with under- and overdosed samples were as follows: a calibration set (C2) consisting of 20 samples (7 production samples from 7 different batches and 13 dosed samples) and the following three prediction sets: P4 (8 dosed samples) and the above-described P2 and P3.The synthetic and dosed samples in prediction sets P1 and P4 were those which were not used in the calibration sets.P2 contained those production samples used in the PCA that were not included in the calibration sets. Finally, in order to assess the robustness of the analytical method, P3 was constructed from batches totally independent of those used in C1, C2 and P2. The samples in prediction set P3 belonged to the four production batches that were not included in the PCA performed to select production samples representatives of the variability in the manufacture process. After the different sample sets to be processed were formed, the PLSR calibration models used were optimized by selecting the most appropriate spectral mode and wavelength range.The criterion used to choose the working wavelength range was to discard very noisy spectral regions (viz., the range from 2200 to 2500 nm) and those where the active principle absorbed negligibly (viz., the 1850–2200 nm range). In this way, the PLSR procedure was applied to absorbance, first-derivative and second-derivative spectra over the wavelength ranges 1100–2200 and 1100–1850 nm.After the different calibration models were constructed, the most suitable method for the intended determination was selected on the basis of being the most straightforward procedure (viz., that using the fewest PLS components) which lead to the lowest relative standard error of prediction (%RSEP) for prediction sets P1 and P4 or P2. Table 1 compares the %RSEP values obtained with the different calibration models studied and both calibration sets.No significant differences in %RSEP among the strategies tested in the different wavelength ranges and spectral modes were observed; however, the models obtained from firstderivative spectra recorded over the shorter wavelength range (1100–1850 nm) were simpler (i.e., they required fewer PLS components) and also, presumably, were more robust. The calibration set consisting of dosed and production samples (C2) always gave lower %RSEP, and, occasionally, simpler models than the calibration set consisting of synthetic and production samples (C1).This may have been the result of the physical properties of the under- and overdosed samples being more similar than those of the synthetic samples to those of the production specimens. Therefore, the under- and overdosing procedure is proposed as the most suitable choice for obtaining a wide enough concentration range for calibration in determining the active principle. The results given below were obtained by using the underand overdosing procedure, and calibrating from first-derivative spectra obtained over the wavelength range 1100–1850 nm.The regression parameters for NIRRS against the reference method for sets C2 and P4 were as follows: intercepts = 3.6 ± 24.0 (C2), 28.1 ± 100.7 (P4); slopes = 0.99 ± 0.04 (C2), 0.96 ± 0.15 (P4); r = 0.9972 (C2), 0.9873 (P4). At a significance level of 95%, the calculated slope and intercept were not significantly different from 1 and 0, respectively, so no systematic difference between the NIRRS results and those provided by the reference method for the calibration samples and the dosed (prediction) samples existed. In order to check whether the proposed method provided concentration values for the production samples (set P2) not significantly different from those of the reference method, a ttest for paired data was used.Based on the tcalc value obtained, 0.145, there were no significant differences between the concentrations obtained with both (t0.05/2,48 = 2.011).The active principle was determined in the samples of P3 (Table 2). Application of the t-test for paired data gave tcalc = 1.234, so the concentrations provided by the reference and NIRRS method were not significantly different either (t0.05/2,19 = 2.093). Finally, the proposed NIRRS method was validated as a viable alternative to the reference spectrophotometric method; in fact, the error of the laboratory method (SDD = 5.2 mg g21) was of the same order of magnitude as the standard error of prediction (SEP = 5.9 mg g21).Conclusions The use of a calibration set containing production samples in addition to under- and overdosed samples apparently provides simpler models subject to smaller errors than one including Table 1 Relative standard errors of prediction calculated for the calibration and prediction sets by using different PLSR models Calibration set synthetic + production— Range/nm Spectral mode PLScomponents RESPC1 RSEPP1 RSEPP2 1100–2200 Absorbance 5 0.69 1.48 1.36 Derivative 1 4 0.59 1.41 1.40 Derivative 2 5 0.53 1.42 1.56 1100–1850 Absorbance 4 0.72 1.40 1.14 Derivative 1 3 0.66 1.46 1.21 Derivative 2 4 0.58 1.15 1.56 Calibration set: dosed + production— Range/nm Spectral mode PLScomponents RESPC1 RSEPP1 RSEPP2 1100–2200 Absorbance 5 0.46 0.91 0.91 Derivative 1 4 0.47 1.02 1.08 Derivative 2 4 0.52 1.08 1.23 1100–1850 Absorbance 4 0.41 0.92 0.89 Derivative 1 2 0.48 0.93 0.91 Derivative 2 3 0.54 1.01 1.09 Table 2 Determination of the active principle in the samples of prediction set P3 by use of the PLSR model obtained from first-derivative spectra recorded over the wavelength range 1100–1850 nm.Sample LAB/mg g21 NIRRS/mg g21 Relative error (%) 1 638.6 635.4 0.5 2 639.4 644.1 20.7 3 645.1 640.4 0.7 4 646.7 651.6 20.8 5 632.9 630.7 0.3 6 633.3 634.6 20.2 7 633.6 640.2 21.0 8 639.8 631.1 1.4 9 630.0 628.8 0.2 10 630.5 640.3 21.5 11 632.1 628.6 0.6 12 635.1 636.7 20.2 13 628.6 630.6 20.3 14 630.4 624.8 0.9 15 631.4 626.8 0.7 16 633.7 632.9 0.1 17 629.1 629.5 20.1 18 631.6 623.0 1.4 19 632.5 624.6 1.3 20 635.4 624.7 1.7 RSEP% = 0.88 764 Analyst, August 1997, Vol. 122production and synthetic samples. However, this conclusion cannot be unequivocally reached from the data presented since the calibrations for the two methods of producing the required compositional variations in the sample sets were based on mixed reference analytical data i.e., some based on masses and some based on UV determinations. Although the comparative data on a small number of samples showed minimal differences (less than ±0.6%) between mass and UV data, these differences contribute extra variance to the reference data, which affects the NIR calibrations and hence reduces the certainty of the conclusions drawn about performance of the NIR methods on the test sets.The active principle in the pharmaceutical preparation can be quantified by NIRRS with no prior manipulation of the sample; spectra can be recorded in situ with errors of prediction less than 1% and substantially reduced analysis times by using a fibre optic probe. The proposed method is thus a viable alternative to the spectrophotometric determination of the preparation. The authors are grateful to Spain’s DGYCyT (project PB 93-899) and to the Direcci�o General de Recerca de la Generalitat de Catalunya (1995 SGR-50) for financial support granted for the realization of this work. Thanks are also due to Laboratorios Menarini S.A.for kindly supplying the samples used. References 1 Osborne, B. G., and Fearn, T., Near Infrared Spectroscopyn Food Analysis, Longman Scientific and Technical, Harlow, 1986. 2 Blanco, M., Coello, J., Iturriaga, H., Maspoch, S., and Bertran, E., Analyst, 1994, 119, 1779. 3 Bunding Lee, K. A., Appl. Spectrosc. Rev., 1993, 28, 231. 4 Corti, P., and Dreassi, E., Il Farmaco, 1993, 48, 3. 5 Morisseau, K. M., and Rhodes, C. T., Drug Dev. Ind. Pharm., 1995, 21, 1071. 6 Dreassi, E., Ceramelli, G., Corti, P., Lonardi, S., and Perrucio, P. L., Analyst, 1995, 120, 1005. 7 Plugge, W., and Van der Vlies, C., J. Pharm. Biomed. Anal., 1993, 11, 435. 8 Corti, P., Savini, L., Dreassi, E., Petriconi, S., Genga, R., Montecchi, L., and Lonardi, S., Process Control Qual., 1992, 2, 131. 9 Sinsheimer, J. E., and Poswalk, N.M., J. Pharm. Sci., 1986, 57, 2007. 10 Warren, R. J., Zarembo, J. E., Chong, C. W., and Robinson, M. J., J. Pharm. Sci., 1970, 59, 109. 11 Dempster, M. A., MacDonald, B. F., Gemperline, P. J., and Boyer, N. R., Anal. Chim. Acta, 1995, 310, 43. 12 Jones, J. A., Last, I. R., MacDonald, B. F., and Prebble, K. A., J. Pharm. Biomed. Anal., 1993, 11, 1227. 13 Drennen, J. K., and Lodder, R. A., J. Pharm. Sci., 1990, 79, 622. 14 Blanco, M., Coello, J., Iturriaga, H., Maspoch, S., De la Pezuela, C., and Russo, E., Anal.Chim. Acta, 1994, 298, 183. 15 Jouan-Rimbaud, D., Khots, M. S., Massart, D. L., Last, I. R., and Prebble, K. A., Anal. Chim. Acta, 1995, 315, 257. 16 Jouan-Rimbaud, D., Walczak, B., Massart, D. L., Last, I. R., and Prebble, K. A., Anal. Chim. Acta, 1995, 304, 285. 17 Blanco, M., Coello, J., Iturriaga, H., Maspoch, S., and De la Pezuela, C., Anal. Chim. Acta, 1996, 333, 147. 18 Dreassi, E., Ceramelli, G., Savini, L., Corti, P., Perruccio, P.L., and Lonardi, S., Analyst, 1995, 120, 319. 19 Dreassi, E., Ceramelli, G., Corti, P., Perruccio, P. L., and Lonardi, S., Analyst, 1996, 121, 219. 20 Martens, H., and Naes, T., Multivariate Calibration, Wiley, New York, 1991. 21 Ciurczak, E., Pharm. Technol, 1991, 15, 140. 22 Esbensen, K., Midtgaard, T., and Sch�onkopf, S., Multivariate Analysis-In Practice, CAMO AS, Trondheim, 1994. 23 Haaland, D. M., and Thomas, E. V., Anal. Chem., 1988, 60, 1193. 24 Otto, M., and Wegscheider, W., Anal.Chem., 1985, 57, 63. Paper 7/00630F Received January 28, 1997 Accepted April 25, 1997 Analyst, August 1997, Vol. 122 765 Strategies for Constructing the Calibration Set in the Determination of Active Principles in Pharmaceuticals by Near Infrared Diffuse Reflectance Spectrometry M. Blanco*, J. Coello, H. Iturriaga, S. Maspoch and C. de la Pezuela Departamento de Qu�ýmica, Unidad de Qu�ýmica Anal�ýtica, Universidad Aut�onoma de Barcelona, E-08193 Bellaterra, Barcelona, Spain The active principle in the blended phase of a commercially available pharmaceutical preparation was determined using near infrared diffuse reflectance spectrometry in combination with a fibre optic probe and multivariate calibration by partial least-squares regression.Two different ways of preparing laboratory samples spanning an appropriate concentration range for constructing the calibration set were compared. One of the procedures involves preparing synthetic samples by weighing and the other using under- and overdosed production samples.Although the results provided by the two strategies were not significantly different, the second was judged more effective because it has a less marked effect on those physical properties of the samples that affect the IR spectrum. The prediction errors obtained (less than 1%) indicate the suitability of the proposed sample preparation procedure, which is faster than the usual method of choice and provides comparable results.Keywords: Miokamycin; near infrared spectrometry; pharmaceutical; fibre optic probe; calibration for quantitative analysis; partial least squares regression Near infrared diffuse reflectance spectrometry (NIRRS) has developed very rapidly since its inception, so much so that it is currently a frequent choice for the analysis of agricultural foods1 and a powerful tool for control analyses in the textile,2 polymer,3 biomedical4 and pharmaceutical fields,5 amongst others.The dramatic spread of the NIRRS technique can be ascribed to its suitability for qualitative and quantitative determinations of physical and chemical parameters, its ability to process solid samples with minimal or no manipulation and its compatibility with fibre optic probes as signal transmitters, which facilitates analyses. The NIRRS technique has aroused special interest in the pharmaceutical industry, where it is commonly used for the identification of both raw materials and end products,6–8 and in moisture measurements;9,10 however, it is equally applicable to finished products,11–13 with the advantages inherent in its nondestructive, expeditious, non-invasive nature.Although the use of a fibre optic probe minimizes sample manipulation by the analyst, the loss of energy in travelling through fibres increases signal noise and consequently decreases the signal-to-noise ratio, which can have a significant adverse effect on quantitative analyses. Despite this, fibre optic probes can still be suitable for quantitative purposes and provide accuracy and precision similar to those of the traditional spectral recording procedure.14 Because the NIRRS signal is dependent on both the physical and chemical properties of the sample, differences between spectra for a given preparation can arise from changes in both the physical properties of the samples (particle size, grain size distribution, compactness, etc.) and the concentrations of their components.One of the hindrances to the use of NIRRS in quantitative pharmaceutical analyses is the difficulty in obtaining an appropriate sample set for constructing the calibration matrix; in fact, production samples contain concentrations that vary only slightly around central values that are close to the stated contents, so they hardly encompass a wide enough range for appropriate calibration. Expanding such a narrow concentration range without altering other factors that influence spectra entails using an effective strategy to prepare synthetic or dosed samples (e.g., manufactured samples on pilot-scale equipment;15,16 samples prepared at the laboratory by mixing all the pharmaceutical ingredients over concentration ranges close to the manufacturer’s stated contents;17 or a procedure for preparing underand overdosed samples from different production batches18,19).The first strategy probably gives the calibration samples with the closest resemblance to production samples; however, no new batches can be produced as the need for calibration arises, nor can one assure that the samples thus obtained will match production samples in every respect.The second procedure has the advantage that the different analyte concentrations required to compile the calibration set are easy to obtain; however, the laboratory procedures used to grind and mix ingredients, for example, can vary markedly from those employed in the production chain, so the physical properties of the two types of sample, and hence their NIRRS spectra, may differ.The use of a calibration set containing normal production samples in addition to under- and overdosed production samples places extra demands on the calibration and so the production of synthetic or dosed samples should pay due regard to physical properties and, provided strict control is exercised, inadvertent differences between production and synthetic/dosed samples can be minimised and accommodated.This paper compares the performance of two strategies for preparing laboratory samples spanning an appropriate concentration range for constructing a calibration set. One involves making synthetic samples by weighing the required amount of each ingredient; the other uses under- and overdosed samples from different production batches. The active principle in the blended phase of the pharmaceutical preparation assayed was determined by partial least-squares regression (PLSR)20, using a calibration set consisting of laboratory samples prepared as described above that was expanded with untreated samples from different production batches i consider the variability inherent in the manufacturing process.Experimental Samples and Reagents Miokamycin (9,3A-diacetylmiodecamycin) is a macrolide antibiotic derived from miodecamycin A1. The amorphous form, containing 0.2% hydroxypropylmethylcellulose, is the active principle of a pharmaceutical that is commercially available in tablets.Microcrystalline cellulose, magnesium stearate, carboxymethyl starch, methylpropylcellulose and aluminium glycinate are the excipients. Analyst, August 1997, Vol. 122 (761–765) 761The determination of miokamycin which has a nominal content of 640 mg g21 in the product, is performed in the blended phase before compression. Four samples each from 19 different production batches were selected over a two-month period.These were subsequently allocated to samples subsets for use in developing and testing the NIRRS calibration (see Results and Discussion). The reagents used included P.A. grade methyl alcohol and 1 m hydrochloric acid from Panreac (Barcelona, Spain). The active principle, excipients and production samples were all supplied by Laboratorios Menarini (Badalona, Spain). Apparatus NIRRS spectra were recorded on a NIRSystems 6500 near infrared spectrophotometer equipped with a reflectance detector and a fibre optic probe (model AP6645 ANO3P) for qualitative and quantitative work.The instrument was controlled via the software NSAS v. 3.20, which includes routines for spectral acquisition and processing. The UV spectra used in the reference method were recorded on a Hewlett–Packard HP 8452A diode array spectrophotometer. The instrument’s bundled software, HP 89530 MSDOS UV/VIS, includes facilities for controlling it and acquiring and processing spectra.An Alresa centrifuge and a Shaker Mixer TURBULA Type T2 C from WAB (Basel, Switzerland) were also used. Software Multivariate calibration was performed with the Unscrambler v. 5.03 software package, from CAMO A/S (Trondheim, Norway), which affords principal components regression (PCR) and PLSR, with selection of variables and detection of outliers. The NIRRS spectra recorded were exported from the instrument to Unscrambler with the aid of Perconv v. 2.05, also from CAMO A/S. The multicomponent analysis software MC, developed by the Unidad de Qu�ýmica Anal�ýtica at the Universidad Aut�onoma de Barcelona, was used in the UV analysis. It allows the concentrations of mixture components to be determined by multiple linear regression (MLR) fitting of the mixed spectrum from the spectra for the pure components. Preparation of Laboratory Samples The concentration range of the calibration set was expanded in two ways.One involved preparing 20 samples by weighing variable amounts of the active principle and the five excipients over a range spanning concentrations within 10% above and below the stated contents. The other involved preparing a set consisting of 21 over- or underdosed samples by adding a known amount of the active principle or a mixture of the excipients, respectively, to an also known amount of preparation. In both cases the resulting mixture weighed about 8 g and was homogenized in the shaker mixer before its NIRRS spectrum was recorded.Complete mixing was checked using the method proposed by Ciurczak.21 The samples obtained with the former and latter procedure will henceforward be referred to as ‘synthetic samples’ and ‘dosed samples’; those supplied by the manufacturer were labelled ‘production samples’. UV Reference Procedure The following UV spectrophotometric procedure was used as reference in the determination of the active principle (miokamycin) in dosed and production samples: about 0.5 g of sample was dissolved in 80 ml of methanol, stirred magnetically for 15 min and made to 100 ml with the same solvent.An aliquot of the resulting solution was centrifuged at 3000 rpm for 5 min and 5 ml of the supernatant were diluted to 50 ml with methanol. To 5 ml of this solution 2.5 ml of 1 m HCl were added and the solution made up to 50 ml with methanol. Finally, the spectrum for this last solution was recorded between 200 and 350 nm against a blank consisting of 1 + 20 1 m HCl–methanol.The programme MC was used to quantify the samples from their first-derivative spectrum over the wavelength range 220–270 nm, using the first-derivative spectrum for miokamycin as the sole standard (Fig. 1). Recording of NIRRS Spectra Spectral scanning was performed by introducing the tip of the fibre optic probe into the same wide mouth polyethylene bottle in which the sample was weighed and mixed, turning over the sample with a spatula between successive scans.Each sample was used to record three spectra over the wavelength range 1100–2500 nm. The three spectra thus obtained were averaged and the resulting spectrum was employed to quantify the active principle in the blended phase of the pharmaceutical preparation. Production batches for inclusion in the calibration set were selected from the mean spectrum for each batch, which was obtained from three spectra for each of the four samples in the batch.Fig. 2 shows the absorbance and first-derivative NIRRS spectra for the active principle and a sample of the pharmaceutical preparation. Data Processing The active principle concentration used as reference was the weighing value for synthetic samples and the mean value of two determinations with the above-described UV reference procedure for dosed and production samples. To assure that there were no significant differences between the weighing and UV reference methods, three synthetic samples were prepared at low, medium and high concentration levels and three aliquots of each one were analysed by the UV reference procedure.Differences between weighed and UV found values were lower than ±0.6% in all cases. The error of the UV reference method was established from the following expression: SDD = (di - dm)2 i=1 nå n -1 (1) Fig. 1 First-derivative UV spectra for (1) a 20.0 mg l21 miokamycin solution and (2) a solution of the pharmaceutical preparation containing 49.3 mg l21 in 1 m HCl–methanol (1 + 20). 762 Analyst, August 1997, Vol. 122where SDD is the standard deviation of difference, di the difference between two replicates for sample i, dm the mean of all the differences and n the number of samples used.22 The obtained value was 5.2 mg g21, 0.8% of the nominal content of miokamycin. All PLSR models were constructed by cross-validation, using as many segments as samples were in the calibration set.The optimum number of PLS components for quantitation was estimated on the basis of the statistical criterion of Haaland and Thomas.23 The goodness of the results provided by the different PLSR models studied was assessed in terms of the relative standard error of prediction, %RSEP:24 % ( ) RSEP c c c i i i i n i n = - � = = å å LAB NIRRS LAB 2 1 2 1 100 (2) where cLABi and cNIRRSi are the concentration of the active principle provided by the reference method and that calculated by the PLSR model, respectively, for sample i, and n is the number of samples.In order to ensure that the results provided by the reference method and the proposed NIRRS method were not significantly different, a t-test for paired data was applied to the results for the production samples in sub-sets P2 and P3 (see Results and Discussion), and the error for the reference method, SDD, was compared with the standard error of prediction, SEP, for the samples in the prediction set, defined as SEP c c n i i i n = - - = å ( ) LAB NIRRS 2 1 1 (3) where CLABi and CNIRRSi are the concentrations of the active principle provided by the reference method and calculated by the PLSR model, respectively, for sample i, and n is the number of samples.SEP must be roughly equal to SDD if the reference and proposed method are to provide similar results.22 Results and Discussion The calibration set used to determine the active principle should contain normal production samples from different batches in addition to synthetic or dosed samples in order to ensure that the set is representativ most common sources of variability in the manufacturing process (particle size, moisture, supplier of the raw materials, etc.).Using a mixed calibration matrix affords the implementation of methodology that quantifies concentrations close to the stated content (640 mg/g) or falling within the manufacturer’s accepted range with the same accuracy.Because NIRRS spectra were recorded directly from the solid samples, the scattering observed must be minimized by using some mathematical treatment. The first spectral derivative is the readiest choice in this respect and has the advantage that it cancels baseline shifts; however, the second derivative is more widely used and allows correction of constant variations, at the expense of increased noise. On the other hand, while the use of a fibre optic probe considerably facilitates recording of spectra, it also raises signal noise such that the 2200–2500 nm wavelength range is rendered essentially unusable in practice.In order to select production batches representative of the variability sources, which should thus be included in the calibration set, the mean second-derivative spectra over the range 1100–2200 nm for 14 randomly chosen batches among the 19 available were subjected to principal component analysis (PCA);20 the other 5 production batches were not included in the PCA in order that they could form a completely independent sample set to test the robustness of the calibration.Based on the results, the first two principal components (PCs) accounted for 88.8% of the variance. Fig. 3 shows a plot of score PC1 against PC2; as can be seen, the production batches clustered in three blocks, which suggests differences between them. Such differences cannot be ascribed to the concentration because all the values were close to the stated content (641.0–662.5 mg g21), but rather to slight changes in the physical properties of the samples that influence the NIRRS spectrum.Therefore, for a calibration set to include variability sources inherent in the manufacturing process as far as possible, it should contain a sample from at least each of the production batches representing the extremes of variability, e.g., those in boldface in Fig. 3.To determine the active principle by using the synthetic samples four sets were used: a calibration set (C1) consisting of 20 samples (7 production samples from 7 different batches and Fig. 2 Absorbance and first-derivative NIRRS spectra for (1) the active principle and (2) a sample of the commercially available preparation. Fig. 3 Plot of scores for the first and second PC for the production batches. Production batches included in (filled circles) and excluded from (open circles) the calibration set.Analyst, August 1997, Vol. 122 76313 synthetic samples) and three prediction sets, viz. P1 (7 synthetic samples), P2 (49 production samples from 14 different batches) and P3 (20 production samples from 5 other batches). The four sets used with under- and overdosed samples were as follows: a calibration set (C2) consisting of 20 samples (7 production samples from 7 different batches and 13 dosed samples) and the following three prediction sets: P4 (8 dosed samples) and the above-described P2 and P3.The synthetic and dosed samples in prediction sets P1 and P4 were those which were not used in the calibration sets. P2 contained those production samples used in the PCA that were not included in the calibration sets. Finally, in order to assess the robustness of the analytical method, P3 was constructed from batches totally independent of those used in C1, C2 and P2. The samples in prediction set P3 belonged to the four production batches that were not included in the PCA performed to select production samples representatives of the variability in the manufacture process.After the different sample sets to be processed were formed, the PLSR calibration models used were optimized by selecting the most appropriate spectral mode and wavelength range. The criterion used to choose the working wavelength range was to discard very noisy spectral regions (viz., the range from 2200 to 2500 nm) and those where the active principle absorbed negligibly (viz., the 1850–2200 nm range).In this way, the PLSR procedure was applied to absorbance, first-derivative and second-derivative spectra over the wavelength ranges 1100–2200 and 1100–1850 nm. After the different calibration models were constructed, the most suitable method for the intended determination was selected on the basis of being the most straightforward procedure (viz., that using the fewest PLS components) which lead to the lowest relative standard error of prediction (%RSEP) for prediction sets P1 and P4 or P2.Table 1 compares the %RSEP values obtained with the different calibration models studied and both calibration sets. No significant differences in %RSEP among the strategies tested in the different wavelength ranges and spectral modes were observed; however, the models obtained from firstderivative spectra recorded over the shorter wavelength range (1100–1850 nm) were simpler (i.e., they required fewer PLS components) and also, presumably, were more robust.The calibration set consisting of dosed and production samples (C2) always gave lower %RSEP, and, occasionally, simpler models than the calibration set consisting of synthetic and production samples (C1). This may have been the result of the physical properties of the under- and overdosed samples being more similar than those of the synthetic samples to those of the production specimens. Therefore, the under- and overdosing procedure is proposed as the most suitable choice for obtaining a wide enough concentration range for calibration in determining the active principle.The results given below were obtained by using the underand overdosing procedure, and calibrating from first-derivative spectra obtained over the wavelength range 1100–1850 nm. The regression parameters for NIRRS against the reference method for sets C2 and P4 were as follows: intercepts = 3.6 ± 24.0 (C2), 28.1 ± 100.7 (P4); slopes = 0.99 ± 0.04 (C2), 0.96 ± 0.15 (P4); r = 0.9972 (C2), 0.9873 (P4).At a significance level of 95%, the calculated slope and intercept were not significantly different from 1 and 0, respectively, so no systematic difference between the NIRRS results and those provided by the reference method for the calibration samples and the dosed (prediction) samples existed. In order to check whether the proposed method provided concentration values for the production samples (set P2) not significantly different from those of the reference method, a ttest for paired data was used.Based on the tcalc value obtained, 0.145, there were no significant differences between the concentrations obtained with both (t0.05/2,48 = 2.011). The active principle was determined in the samples of P3 (Table 2). Application of the t-test for paired data gave tcalc = 1.234, so the concentrations provided by the reference and NIRRS method were not significantly different either (t0.05/2,19 = 2.093).Finally, the proposed NIRRS method was validated as a viable alternative to the reference spectrophotometric method; in fact, the error of the laboratory method (SDD = 5.2 mg g21) was of the same order of magnitude as the standard error of prediction (SEP = 5.9 mg g21). Conclusions The use of a calibration set containing production samples in addition to under- and overdosed samples apparently provides simpler models subject to smaller errors than one including Table 1 Relative standard errors of prediction calculated for the calibration and prediction sets by using different PLSR models Calibration set synthetic + production— Range/nm Spectral mode PLScomponents RESPC1 RSEPP1 RSEPP2 1100–2200 Absorbance 5 0.69 1.48 1.36 Derivative 1 4 0.59 1.41 1.40 Derivative 2 5 0.53 1.42 1.56 1100–1850 Absorbance 4 0.72 1.40 1.14 Derivative 1 3 0.66 1.46 1.21 Derivative 2 4 0.58 1.15 1.56 Calibration set: dosed + production— Range/nm Spectral mode PLScomponents RESPC1 RSEPP1 RSEPP2 1100–2200 Absorbance 5 0.46 0.91 0.91 Derivative 1 4 0.47 1.02 1.08 Derivative 2 4 0.52 1.08 1.23 1100–1850 Absorbance 4 0.41 0.92 0.89 Derivative 1 2 0.48 0.93 0.91 Derivative 2 3 0.54 1.01 1.09 Table 2 Determination of the active principle in the samples of prediction set P3 by use of the PLSR model obtained from first-derivative spectra recorded over the wavelength range 1100–1850 nm.Sample LAB/mg g21 NIRRS/mg g21 Relative error (%) 1 638.6 635.4 0.5 2 639.4 644.1 20.7 3 645.1 640.4 0.7 4 646.7 651.6 20.8 5 632.9 630.7 0.3 6 633.3 634.6 20.2 7 633.6 640.2 21.0 8 639.8 631.1 1.4 9 630.0 628.8 0.2 10 630.5 640.3 21.5 11 632.1 628.6 0.6 12 635.1 636.7 20.2 13 628.6 630.6 20.3 14 630.4 624.8 0.9 15 631.4 626.8 0.7 16 633.7 632.9 0.1 17 629.1 629.5 20.1 18 631.6 623.0 1.4 19 632.5 624.6 1.3 20 635.4 624.7 1.7 RSEP% = 0.88 764 Analyst, August 1997, Vol. 122production and synthetic samples.However, this conclusion cannot be unequivocally reached from the data presented since the calibrations for the two methods of producing the required compositional variations in the sample sets were based on mixed reference analytical data i.e., some based on masses and some based on UV determinations. Although the comparative data on a small number of samples showed minimal differences (less than ±0.6%) between mass and UV data, these differences contribute extra variance to the reference data, which affects the NIR calibrations and hence reduces the certainty of the conclusions drawn about performance of the NIR methods on the test sets.The active principle in the pharmaceutical preparation can be quantified by NIRRS with no prior manipulation of the sample; spectra can be recorded in situ with errors of prediction less than 1% and substantially reduced analysis times by using a fibre optic probe. The proposed method is thus a viable alternative to the spectrophotometric determination of the preparation. The authors are grateful to Spain’s DGYCyT (project PB 93-899) and to the Direcci�o General de Recerca de la Generalitat de Catalunya (1995 SGR-50) for financial support granted for the realization of this work. Thanks are also due to Laboratorios Menarini S.A. for kindly supplying the samples used. References 1 Osborne, B. G., and Fearn, T., Near Infrared Spectroscopy in Food Analysis, Longman Scientific and Technical, Harlow, 1986. 2 Blanco, M., Coello, J., Iturriaga, H., Maspoch, S., and Bertran, E., Analyst, 1994, 119, 1779. 3 Bunding Lee, K. A., Appl. Spectrosc. Rev., 1993, 28, 231. 4 Corti, P., and Dreassi, E., Il Farmaco, 1993, 48, 3. 5 Morisseau, K. M., and Rhodes, C. T., Drug Dev. Ind. Pharm., 1995, 21, 1071. 6 Dreassi, E., Ceramelli, G., Corti, P., Lonardi, S., and Perrucio, P. L., Analyst, 1995, 120, 1005. 7 Plugge, W., and Van der Vlies, C., J. Pharm. Biomed. Anal., 1993, 11, 435. 8 Corti, P., Savini, L., Dreassi, E., Petriconi, S., Genga, R., Montecchi, L., and Lonardi, S., Process Control Qual., 1992, 2, 131. 9 Sinsheimer, J. E., and Poswalk, N. M., J. Pharm. Sci., 1986, 57, 2007. 10 Warren, R. J., Zarembo, J. E., Chong, C. W., and Robinson, M. J., J. Pharm. Sci., 1970, 59, 109. 11 Dempster, M. A., MacDonald, B. F., Gemperline, P. J., and Boyer, N. R., Anal. Chim. Acta, 1995, 310, 43. 12 Jones, J. A., Last, I. R., MacDonald, B. F., and Prebble, K. A., J. Pharm. Biomed. Anal., 1993, 11, 1227. 13 Drennen, J. K., and Lodder, R. A., J. Pharm. Sci., 1990, 79, 622. 14 Blanco, M., Coello, J., Iturriaga, H., Maspoch, S., De la Pezuela, C., and Russo, E., Anal. Chim. Acta, 1994, 298, 183. 15 Jouan-Rimbaud, D., Khots, M. S., Massart, D. L., Last, I. R., and Prebble, K. A., Anal. Chim. Acta, 1995, 315, 257. 16 Jouan-Rimbaud, D., Walczak, B., Massart, D. L., Last, I. R., and Prebble, K. A., Anal. Chim. Acta, 1995, 304, 285. 17 Blanco, M., Coello, J., Iturriaga, H., Maspoch, S., and De la Pezuela, C., Anal. Chim. Acta, 1996, 333, 147. 18 Dreassi, E., Ceramelli, G., Savini, L., Corti, P., Perruccio, P. L., and Lonardi, S., Analyst, 1995, 120, 319. 19 Dreassi, E., Ceramelli, G., Corti, P., Perruccio, P. L., and Lonardi, S., Analyst, 1996, 121, 219. 20 Martens, H., and Naes, T., Multivariate Calibration, Wiley, New York, 1991. 21 Ciurczak, E., Pharm. Technol, 1991, 15, 140. 22 Esbensen, K., Midtgaard, T., and Sch�onkopf, S., Multivariate Analysis-In Practice, CAMO AS, Trondheim, 1994. 23 Haaland, D. M., and Thomas, E. V., Anal. Chem., 1988, 60, 1193. 24 Otto, M., and Wegscheider, W., Anal. Chem., 1985, 57, 63. Paper 7/00630F Received January 28, 1997 Accepted April 25, 1997 Analyst, August 1997, Vol. 122
ISSN:0003-2654
DOI:10.1039/a700630f
出版商:RSC
年代:1997
数据来源: RSC
|
8. |
Application of Near-infrared Reflectance Spectrometry in the Studyof AtopyPart 1. Investigation of Skin Spectra |
|
Analyst,
Volume 122,
Issue 8,
1997,
Page 767-770
E. Dreassi,
Preview
|
|
摘要:
Application of Near-infrared Reflectance Spectrometry in the Study of Atopy Part 1. Investigation of Skin Spectra E. Dreassia, G. Ceramellia, L. Fabbrib, F. Vocionib, P. Bartalinic and P. Corti*a a Department of Chemical and Pharmaceutical Technology, Siena University, via Banchi di Sotto 55, 53100 Siena, Italy b Military Pharmaceutical Establishment, via R. Giuliani 201, 50144 Florence, Italy c Dermatology Clinic, Siena University, viale Bracci, 53100 Siena, Italy An investigation into the existence of spectral differences and differences in response in terms of water and lipid content between normal and atopic skin is described.Since NIR radiation penetrates complex structured matrices down to a depth of 0.15–0.20 mm, it is evident that the method lends itself to spectral detection of skin components down to the deepest level. First the reproducibility of readings made with the instrument was tested and it was also checked whether the use of the probe caused changes in skin equilibrium due to occlusion.Analysis of the NIR spectra did not enable normal and atopic subjects to be distinguished unequivocally but provided important information on the use of NIR spectrometry in these subjects and insights into the stratum corneum. Although the responses of water and lipid structures could not be read directly from the spectra, it was possible to decompose the global spectral information into components by principal components analysis.It was possible to observe a fraction of variance associated in different ways with water. Keywords: Near-infrared reflectance spectrometry; atopy; skin spectra; stratum corneum Near-infrared reflectance spectrometry (NIRS) has found increasing application for qualitative and quantitative control in the pharmaceutical industry and for measuring clinical parameters in the biomedical sector. Fundamental studies in the former sector were made by Ciurczak and co-workers1–9 and Lodder and co-workers.10–13 In the biomedical sector, NIRS has been used principally to measure blood parameters.14–20 The most interesting feature of this method is the possibility of analysing solid systems without special treatment of the sample. Instruments with fibre-optic probes enable the results to be read by simple analyte–probe contact.The NIR region offers interesting possibilities as it permits the absorbances of overtones and combinations of nitrogen–hydrogen, carbon– hydrogen and oxygen–hydrogen bonds to be read.In the framework of a broad programme aimed at the development of NIRS for qualitative and quantitative control in the pharmaceutical industry, we recently examined the possibility of using the method to study the skin. Since evidence exists to show that NIR radiation penetrates complex structured matrices down to a depth of 0.15–0.20 mm, it is evident that the method lends itself to spectral detection of skin components down to the deepest level.Our research was directed at the possibility of detecting any significant differences between the spectra of normal and atopic subjects and to seeing how the skin responds to contact with chemical agents in everyday use, pharmaceutical excipients and occluding agents in dermatology. Even when their skin is apparently healthy, atopic subjects differ from normal subjects in the water distribution of the stratum corneum (SC). Some workers21–23 have reported values that show a clear alteration in the barrier function of the SC, with accentuated transcutaneous water loss.The phenomenon is related to morphological changes in the keratinocytes and differences in the composition of intercellular lipids, due to defects in enzymes involved in the metabolism of essential fatty acids.24 The keratinocyte changes are visible as pores and furrows in the external surface, villous projections on the inner surface and reduced cohesion between keratinocytes.25 Atopic skin tends to be dry, less elastic and loses water more readily.Its altered barrier function makes it more sensitive to exposure and allergens. In this paper we report the results of an investigation into the existence of spectral differences and differences in response in terms of water and lipid content between normal and atopic skin. Although other spectroscopic methods have been used to study the skin,26,27 nothing was found in the literature on the use of NIRS in atopy. Experimental NIRS Analysis Readings were made in the 1100–2060 nm range in 4 nm steps using a Bran + Luebbe (Nordestedt, Germany) InfraAlyzer 500 with an EDAPT optical fibre probe from the same company.Processing of Spectral Data The spectral data were processed with the following programmes: IDAS-PC version 1.41 (Bran + Luebbe), UNSCRAMBLER version 3.40 (CAMO, Trondheim, Norway) and StatGraphics Plus version 6.0 (Manugistics, Rockville, MD, USA).Methodology The subjects were ten volunteers, five women and five men, five of whom were normal and five of whom were atopic. The latter were identified on the basis of medical history. The readings were taken at three points, situated in the stretch from 5 cm above the wrist fold to the elbow, on each forearm. The procedure was repeated at 1 week intervals for 5 months (November to March). Results and Discussion Reproducibility of Readings First we tested the reproducibility of readings made with the instrument.In one subject from each class, we took five spectral Analyst, August 1997, Vol. 122 (767–770) 767readings at 30 min intervals in a single day at each of the six points. These readings were repeated for five consecutive days. In both subjects, the RSD was calculated each day for each of the six points. These values were used to check the reproducibility of the method and any differences in pattern between the different spectral regions and between subjects in the two categories.Significant differences for p = 0.05 were not found between measurements on different days or in subjects of different classes. The RSD did not exceed 1.5% in any case. The pattern of the mean RSD of the five days and the six points for the entire spectral band is shown in Fig. 1. We also checked whether use of the probe caused changes in skin equilibrium due to occlusion during the collection of spectra. This was done by recording spectra at each point for a subject from each class. Three spectra were recorded at each point, waiting 30 s between each spectrum and then repeating the procedure with a waiting time of 30 min (such an interval guarantees the return of the skin to the original state).The spectra obtained were virtually identical, with an RSD that never exceeded 0.9%. This suggests that during the time required to take a reading (about 40 s), application of the probe does not cause alterations due to occlusion.Analysis of Spectra The zero-order mean spectra for the two classes of subjects are shown in Fig. 2(a). Apart from the large responses around 1450 and 1900 nm, corresponding to the first overtone and the combination band of the absorption of water hydroxyls, respectively, and a band at 1208 nm attributable to the second overtone of the C–H bonds of the lipid component of the SC, there were no useful bands for the identification of other functional groups.In other words, with zero-order spectra we did not find any differences useful for distinguishing normal and atopic subjects. Nor did the first and second derivatives of the skin spectra [Fig. 2(b) and (c)] provide any element that distinguished the two groups. Also, the application of discriminant analysis (DA) or clustering techniques to the raw data or to the first and second derivatives of the spectra did not provide useful information. However, interesting results were obtained by principal component analysis (PCA) of the spectral data.Three models were developed by this method: a general model obtained by applying PCA to the spectra of both classes of subject and two partial models obtained using the separate spectral data of the two classes. The two partial models yielded information on the most important components of variance in the two classes of subjects and on the significance of differences between them. Figs. 3–6 show loadings of the first four principal components (PC) and their first and second derivatives in the two partial models.Some interesting observations emerge from careful comparison of the individual components of the two models. The first component (Fig. 3), representing the largest fraction of the variance in both models (about 90%), did not show elements of differentiation for the two classes (the patterns were virtually superimposed). Since this component was independent of the class of subject and contributed substantially to the total variance, it may be regarded as strongly correlated with the fundamental state of the skin.The pattern of this component was similar to that of the average skin spectrum and there was a clear response corresponding to absorbance due to water, which occurs at different levels in the matrix. This response almost completely dominated the spectrum. On the other hand, the peak around 1200 nm corresponding to the lipid component was clearly reduced.In the first and second derivatives, the lipid and water peaks in normal subjects were more intense. In the second derivative, there was an interesting displacement of about 8 nm of the negative and positive peaks in the 1350–1500 nm region attributable to the first overtone of water, whereas this was not observed in the resolution of the combination band (1800–1950 nm). In normal subjects (Fig. 4), the loading of the second component (about 6% of the total variance in both classes) showed water absorption bands and also significant bands around 1200 and 1700–1800 nm, attributable to C–H bonds (second and first overtones, respectively).This means that C–H bonds are more important in the SC of normal subjects than in atopic subjects. The first and second derivatives (especially the latter) confirmed this. In the second derivative spectra, there was an interesting difference between the two classes of subjects as far as the peaks at 1212 and 1716 nm were concerned. It therefore seems possible to conclude that in Fig. 1 Pattern of the mean RSD of the five days and the six points for the entire spectral band for the normal (2) and atopic (5) subjects. Fig. 2 Mean spectra for the normal (2) and atopic (5) subjects. (a) Zeroorder spectra, (b) first derivative and (c) second derivative spectra. Fig. 3 Loadings of the first PC for the two partial models. (a) Loadings, (b) first derivative and (c) second derivative of loadings. 768 Analyst, August 1997, Vol. 122normal subjects there is a more significant structuring of the lipid component than in atopic subjects. The third component (Fig. 5) (about 3% of the total variance) did not show substantial differences in the two classes, although water had a greater influence in atopic than in normal subjects, and vice versa for the lipid component. This was confirmed by the derivative spectra. There was a significant difference between the two classes at 1212 nm and the bands attributable to water were displaced 8 nm.The fourth component (Fig. 6), although contributing only about 1% to the explained variance in both classes, showed large differences between the two classes. In normal subjects it was completely dominated by bands attributable to the vibrations of C–H bonds (1212 nm, second overtone; 1404 nm, double peak due to stretching–bending; 1716/1764 nm, double peak due to first overtone). In atopic subjects, there were only weak peaks at 1212 nm and 1716/1760 nm and a strong peak at 1440 nm, attributable to water.These observations suggest interesting reflections on the sources of variability in the two classes of subjects. We observed that the variance attributed to lipid structures was on the whole greater in normal subjects, whereas the variance in atopic subjects seemed to depend more on the water component which was present in a significant manner in all four PCA components.This is probably due to the fast water equilibrium that develops in atopic subjects and that determines the largest source of variability in the SC of this class. In the model developed from all the spectra recorded, we obtained information on the relationships between spectra, and thus between the two classes of subjects. Figs. 7 and 8 show the loadings of the first four PCs (88.6%, 4.4%, 3.7% and 1.1%, respectively, of the total variance) and the respective first and second derivatives.Examination of the figures shows that the first component is dominated by variance due to absorption bands of water (Fig. 7). In the second component (Fig. 7), we observe absorption bands characteristic of C–H bonds (1212, 1412 and 1716/1764 nm), and two unresolved bands in the 1400–1500 nm region (probably due to water and O–H bond absorption), the variance of which is due to variations in lipid structures. In the third component (Fig. 8), there are bands attributable to lipid components (with negative coefficients) and water (with positive coefficients). The variance associated with this component is therefore presumably due to interactions between hydrophilic and lipophilic structures in the SC.The pattern in the 1810–1830 nm region is interesting because this is the band in which signals attributable to fatty acids with a trans structure, absent in the cis isomer, are found. The fourth component (Fig. 8) has an evident C–H peak at 1212 nm and other bands which are difficult to interpret.Fig. 4 Loadings of the second PC for the two partial models. (a) Loadings, (b) first derivative and (c) second derivative of loadings. Fig. 5 Loadings of the third PC for the two partial models. (a) Loadings, (b) first derivative and (c) second derivative of loadings. Fig. 6 Loadings of the fourth PC for the two partial models. (a) Loadings, (b) first derivative and (c) second derivative of loadings. Fig. 7 Loading of the first (solid line) and second (dotted line) PC [(a) and their first (b) and second (c) derivatives].Fig. 8 Loading of the third (solid line) and fourth (dotted line) PC [(a) and their first (b) and second (c) derivatives]. Analyst, August 1997, Vol. 122 769This shows that by far the greatest source of differences was again water, followed by lipophilic components and interactions between hydrophilic and lipophilic components. We tried to discriminate between normal and atopic subjects on the basis of the scores of the above principal components (PC).Fig. 9 shows the distribution of scores in the plane of the first two PCs. Normal subjects are clearly divided into two subgroups, not correlated with sex, and these subgroups are not particularly dispersed along the two components. On the other hand, atopic subjects do not form a compact group but are dispersed along the axis of the first PC. This further confirms the extreme variability and mobility characterizing the fast water equilibrium of atopic subjects.Although it was not possible to obtain a satisfactory separation of the two groups of subjects, it emerges that atopic subjects do not always differ substantially from normal subjects, and that any differences, whether spectroscopic or clinical, principally concern the water equilibrium of the SC. It also emerges that normal subjects are divided into two distinct subgroups, differing mainly in the second PC which is principally related to lipid structures in the SC.In atopic subjects, this component shows little dispersion. Conclusions Analysis of the NIR spectra did not enable normal and atopic subjects to be distinguished unequivocally but provided important information on the use of NIRS in these subjects and insights into the SC. Although the responses of water and lipid structures could not be read directly from the spectra, it was possible to decompose the global spectral information into components by PCA.For example, it was possible to observe a fraction of variance associated in different ways with water. This variance was present in more than one PC and could therefore reflect different interactions between water and other structures present in the SC (e.g., lipid and protein structures). It could also reflect compartmentalization between surface and deep water within the SC. NIRS therefore seems capable of giving spectral responses that are correlated with the overall water balance and with interactions between water and other components of the SC.References 1 Ciurczak, E. W., and Maldacker, T. A., Spectroscopy, 1986, 1, 36. 2 Ciurczak, E. W., and Torlini, R. P., Spectroscopy, 1987, 2, 41. 3 Ciurczak, E. W., and Shintani-Young, T., paper presented at the Pittsburgh Conference on Analytical Chemistry and Applied Spectroscopy, New Orleans, LA, 1985. 4 Ciurczak, E. W., paper presented at the 1st Panamerican Chemistry Conference, San Juan, Puerto Rico, 1985. 5 Ciurczak, E. W., paper presented at FACSS, Philadelphia, PA, 1985. 6 Ciurczak, E. W., paper presented at the 26th Annual Conference on Pharmaceutical Analysis, Merrimac, WI, 1985. 7 Ciurczak, E. W., paper presented at the 8th Annual Symposium on NIRA, Technicon, Tarrytown, NY, 1985. 8 Buchanan, B. R., Ciurczak, E. W., and Grunke, A. Q., Honigs Spectrosc., 1988, 3, 54. 9 Kradjel, C., and Ciurczak, E. W., paper presented at the 1st Panamerican Chemistry Conference, San Juan, Puerto Rico, 1985. 10 Lodder, R. A., Selby, M., and Hieftje, G. M., Anal. Chem., 1987, 59, 1921. 11 Lodder, R. A., and Hieftje, G. M., Appl. Spectrosc., 1988, 42, 556. 12 Lodder, R. A., and Hieftje, G. M., Appl. Spectrosc., 1988, 42, 1512. 13 Drennen, J. K., and Lodder, R. A., J. Pharm. Sci., 1990, 79, 662. 14 Tamura, M., Seyama, A., and Hazeki, O., Adv. Exp. Med. Biol., 1987, 215, 297. 15 Tamura, M., Hazeki, O., Nioka, S., Chance, B., and Smith, D.S., Adv. Exp. Med. Biol., 1988, 222, 359. 16 Hazeki, O., Seyama, A., and Tamura, M., Adv. Exp. Med. Biol., 1987, 215, 297. 17 Van Toorenenbergen, A. W., Blijenberg, B. G., and Leijnse, B., J. Clin. Chem. Clin. Biochem., 1988, 26, 209. 18 Peuchant, E., Salles, C., and Jensen, R., Anal. Chem., 1987, 59, 1816. 19 Lodder, R. A., Hieftje, G. M., Gary, M., Moorehead, W., Robertson, S. P., and Rand, P., Talanta, 1989, 36, 193. 20 Araki, R., and Nashimoto, I., Adv.Exp. Med. Biol., 1989, 248, 11. 21 Mathias, T., Wilson, S., and Maibach, H., J. Invest. Dermatol., 1981, 77, 219. 22 Tagami, H., Kanamuru, Y., and Inque, K., J. Invest. Dermatol., 1982, 78, 425. 23 Werner, Y., and Lindberg, M., Acta Dermatovenereol., 1985, 65, 102. 24 Monku, M. S., Horrobin, D. F., Morse, N. L., Wright, S., and Burton, J. L., J. Dermatol., 1984, 110, 643. 25 Rajka, G., J. Pediat. Dermatol., 1989, 8, 182. 26 Potts, R. O., Guzek, D.B., Harris, R. R., and McKie, J.E., Arch. Dermatol. Res., 1985, 277, 489. 27 Bommannan, D., Potts, R. O., and Guy, R. H., J. Controlled Release, 1991, 16, 299. Paper 7/01254C Received February 24, 1997 Accepted May 1, 1997 Fig. 9 Distribution of scores of the normal (+) and atopic (Ò) subjects in the plane of the first two PCs. 770 Analyst, August 1997, Vol. 122 Application of Near-infrared Reflectance Spectrometry in the Study of Atopy Part 1. Investigation of Skin Spectra E. Dreassia, G.Ceramellia, L. Fabbrib, F. Vocionib, P. Bartalinic and P. Corti*a a Department of Chemical and Pharmaceutical Technology, Siena University, via Banchi di Sotto 55, 53100 Siena, Italy b Military Pharmaceutical Establishment, via R. Giuliani 201, 50144 Florence, Italy c Dermatology Clinic, Siena University, viale Bracci, 53100 Siena, Italy An investigation into the existence of spectral differences and differences in response in terms of water and lipid content between normal and atopic skin is described.Since NIR radiation penetrates complex structured matrices down to a depth of 0.15–0.20 mm, it is evident that the method lends itself to spectral detection of skin components down to the deepest level. First the reproducibility of readings made with the instrument was tested and it was also checked whether the use of the probe caused changes in skin equilibrium due to occlusion. Analysis of the NIR spectra did not enable normal and atopic subjects to be distinguished unequivocally but provided important information on the use of NIR spectrometry in these subjects and insights into the stratum corneum.Although the responses of water and lipid structures could not be read directly from the spectra, it was possible to decompose the global spectral information into components by principal components analysis. It was possible to observe a fraction of variance associated in different ways with water. Keywords: Near-infrared reflectance spectrometry; atopy; skin spectra; stratum corneum Near-infrared reflectance spectrometry (NIRS) has found increasing application for qualitative and quantitative control in the pharmaceutical industry and for measuring clinical parameters in the biomedical sector. Fundamental studies in the former sector were made by Ciurczak and co-workers1–9 and Lodder and co-workers.10–13 In the biomedical sector, NIRS has been used principally to measure blood parameters.14–20 The most interesting feature of this method is the possibility of analysing solid systems without special treatment of the sample.Instruments with fibre-optic probes enable the results to be read by simple analyte–probe contact. The NIR region offers interesting possibilities as it permits the absorbances of overtones and combinations of nitrogen–hydrogen, carbon– hydrogen and oxygen–hydrogen bonds to be read. In the framework of a broad programme aimed at the development of NIRS for qualitative and quantitative control in the pharmaceutical industry, we recently examined the possibility of using the method to study the skin.Since evidence exists to show that NIR radiation penetrates complex structured matrices down to a depth of 0.15–0.20 mm, it is evident that the method lends itself to spectral detection of skin components down to the deepest level. Our research was directed at the possibility of detecting any significant differences between the spectra of normal and atopic subjects and to seeing how the skin responds to contact with chemical agents in everyday use, pharmaceutical excipients and occluding agents in dermatology.Even when their skin is apparently healthy, atopic subjects differ from normal subjects in the water distribution of the stratum corneum (SC). Some workers21–23 have reported values that show a clear alteration in the barrier function of the SC, with accentuated transcutaneous water loss.The phenomenon is related to morphological changes in the keratinocytes and differences in the composition of intercellular lipids, due to defects in enzymes involved in the metabolism of essential fatty acids.24 The keratinocyte changes are visible as pores and furrows in the external surface, villous projections on the inner surface and reduced cohesion between keratinocytes.25 Atopic skin tends to be dry, less elastic and loses water more readily. Its altered barrier function makes it more sensitive to exposure and allergens.In this paper we report the results of an investigation into the existence of spectral differences and differences in response in terms of water and lipid content between normal and atopic skin. Although other spectroscopic methods have been used to study the skin,26,27 nothing was found in the literature on the use of NIRS in atopy. Experimental NIRS Analysis Readings were made in the 1100–2060 nm range in 4 nm steps using a Bran + Luebbe (Nordestedt, Germany) InfraAlyzer 500 with an EDAPT optical fibre probe from the same company.Processing of Spectral Data The spectral data were processed with the following programmes: IDAS-PC version 1.41 (Bran + Luebbe), UNSCRAMBLER version 3.40 (CAMO, Trondheim, Norway) and StatGraphics Plus version 6.0 (Manugistics, Rockville, MD, USA). Methodology The subjects were ten volunteers, five women and five men, five of whom were normal and five of whom were atopic.The latter were identified on the basis of medical history. The readings were taken at three points, situated in the stretch from 5 cm above the wrist fold to the elbow, on each forearm. The procedure was repeated at 1 week intervals for 5 months (November to March). Results and Discussion Reproducibility of Readings First we tested the reproducibility of readings made with the instrument. In one subject from each class, we took five spectral Analyst, August 1997, Vol. 122 (767–770) 767readings at 30 min intervals in a single day at each of the six points. These readings were repeated for five consecutive days. In both subjects, the RSD was calculated each day for each of the six points. These values were used to check the reproducibility of the method and any differences in pattern between the different spectral regions and between subjects in the two categories. Significant differences for p = 0.05 were not found between measurements on different days or in subjects of different classes.The RSD did not exceed 1.5% in any case. The pattern of the mean RSD of the five days and the six points for the entire spectral band is shown in Fig. 1. We also checked whether use of the probe caused changes in skin equilibrium due to occlusion during the collection of spectra. This was done by recording spectra at each point for a subject from each class. Three spectra were recorded at each point, waiting 30 s between each spectrum and then repeating the procedure with a waiting time of 30 min (such an interval guarantees the return of the skin to the original state). The spectra obtained were virtually identical, with an RSD that never exceeded 0.9%.This suggests that during the time required to take a reading (about 40 s), application of the probe does not cause alterations due to occlusion. Analysis of Spectra The zero-order mean spectra for the two classes of subjects are shown in Fig. 2(a). Apart from the large responses around 1450 and 1900 nm, corresponding to the first overtone and the combination band of the absorption of water hydroxyls, respectively, and a band at 1208 nm attributable to the second overtone of the C–H bonds of the lipid component of the SC, there were no useful bands for the identification of other functional groups. In other words, with zero-order spectra we did not find any differences useful for distinguishing normal and atopic subjects.Nor did the first and second derivatives of the skin spectra [Fig. 2(b) and (c)] provide any element that distinguished the two groups. Also, the application of discriminant analysis (DA) or clustering techniques to the raw data or to the first and second derivatives of the spectra did not provide useful information. However, interesting results were obtained by principal component analysis (PCA) of the spectral data. Three models were developed by this method: a general model obtained by applying PCA to the spectra of both classes of subject and two partial models obtained using the separate spectral data of the two classes.The two partial models yielded information on the most important components of variance in the two classes of subjects and on the significance of differences between them. Figs. 3–6 show loadings of the first four principal components (PC) and their first and second derivatives in the two partial models. Some interesting observations emerge from careful comparison of the individual components of the two models. The first component (Fig. 3), representing the largest fraction of the variance in both models (about 90%), did not show elements of differentiation for the two classes (the patterns were virtually superimposed). Since this component was independent of the class of subject and contributed substantially to the total variance, it may be regarded as strongly correlated with the fundamental state of the skin.The pattern of this component was similar to that of the average skin spectrum and there was a clear response corresponding to absorbance due to water, which occurs at different levels in the matrix. This response almost completely dominated the spectrum. On the other hand, the peak around 1200 nm corresponding to the lipid component was clearly reduced. In the first and second derivatives, the lipid and water peaks in normal subjects were more intense. In the second derivative, there was an interesting displacement of about 8 nm of the negative and positive peaks in the 1350–1500 nm region attributable to the first overtone of water, whereas this was not observed in the resolution of the combination band (1800–1950 nm).In normal subjects (Fig. 4), the loading of the second component (about 6% of the total variance in both classes) showed water absorption bands and also significant bands around 1200 and 1700–1800 nm, attributable to C–H bonds (second and first overtones, respectively). This means that C–H bonds are more important in the SC of normal subjects than in atopic subjects.The first and second derivatives (especially the latter) confirmed this. In the second derivative spectra, there was an interesting difference between the two classes of subjects as far as the peaks at 1212 and 1716 nm were concerned. It therefore seems possible to conclude that in Fig. 1 Pattern of the mean RSD of the five days and the six points for the entire spectral band for the normal (2) and atopic (5) subjects.Fig. 2 Mean spectra for the normal (2) and atopic (5) subjects. (a) Zeroorder spectra, (b) first derivative and (c) second derivative spectra. Fig. 3 Loadings of the first PC for the two partial models. (a) Loadings, (b) first derivative and (c) second derivative of loadings. 768 Analyst, August 1997, Vol. 122normal subjects there is a more significant structuring of the lipid component than in atopic subjects.The third component (Fig. 5) (about 3% of the total variance) did not show substantial differences in the two classes, although water had a greater influence in atopic than in normal subjects, and vice versa for the lipid component. This was confirmed by the derivative spectra. There was a significant difference between the two classes at 1212 nm and the bands attributable to water were displaced 8 nm. The fourth component (Fig. 6), although contributing only about 1% to the explained variance in both classes, showed large differences between the two classes.In normal subjects it was completely dominated by bands attributable to the vibrations of C–H bonds (1212 nm, second overtone; 1404 nm, double peak due to stretching–bending; 1716/1764 nm, double peak due to first overtone). In atopic subjects, there were only weak peaks at 1212 nm and 1716/1760 nm and a strong peak at 1440 nm, attributable to water. These observations suggest interesting reflections on the sources of variability in the two classes of subjects.We observed that the variance attributed to lipid structures was on the whole greater in normal subjects, whereas the variance in atopic subjects seemed to depend more on the water component which was present in a significant manner in all four PCA components. This is probably due to the fast water equilibrium that develops in atopic subjects and that determines the largest source of variability in the SC of this class.In the model developed from all the spectra recorded, we obtained information on the relationships between spectra, and thus between the two classes of subjects. Figs. 7 and 8 show the loadings of the first four PCs (88.6%, 4.4%, 3.7% and 1.1%, respectively, of the total variance) and the respective first and second derivatives. Examination of the figures shows that the first component is dominated by variance due to absorption bands of water (Fig. 7). In the second component (Fig. 7), we observe absorption bands characteristic of C–H bonds (1212, 1412 and 1716/1764 nm), and two unresolved bands in the 1400–1500 nm region (probably due to water and O–H bond absorption), the variance of which is due to variations in lipid structures. In the third component (Fig. 8), there are bands attributable to lipid components (with negative coefficients) and water (with positive coefficients). The variance associated with this component is therefore presumably due to interactions between hydrophilic and lipophilic structures in the SC.The pattern in the 1810–1830 nm region is interesting because this is the band in which signals attributable to fatty acids with a trans structure, absent in the cis isomer, are found. The fourth component (Fig. 8) has an evident C–H peak at 1212 nm and other bands which are difficult to interpret. Fig. 4 Loadings of the second PC for the two partial models. (a) Loadings, (b) first derivative and (c) second derivative of loadings.Fig. 5 Loadings of the third PC for the two partial models. (a) Loadings, (b) first derivative and (c) second derivative of loadings. Fig. 6 Loadings of the fourth PC for the two partial models. (a) Loadings, (b) first derivative and (c) second derivative of loadings. Fig. 7 Loading of the first (solid line) and second (dotted line) PC [(a) and their first (b) and second (c) derivatives]. Fig. 8 Loading of the third (solid line) and fourth (dotted line) PC [(a) and their first (b) and second (c) derivatives].Analyst, August 1997, Vol. 122 769This shows that by far the greatest source of differences was again water, followed by lipophilic components and interactions between hydrophilic and lipophilic components. We tried to discriminate between normal and atopic subjects on the basis of the scores of the above principal components (PC). Fig. 9 shows the distribution of scores in the plane of the first two PCs.Normal subjects are clearly divided into two subgroups, not correlated with sex, and these subgroups are not particularly dispersed along the two components. On the other hand, atopic subjects do not form a compact group but are dispersed along the axis of the first PC. This further confirms the extreme variability and mobility characterizing the fast water equilibrium of atopic subjects. Although it was not possible to obtain a satisfactory separation of the two groups of subjects, it emerges that atopic subjects do not always differ substantially from normal subjects, and that any differences, whether spectroscopic or clinical, principally concern the water equilibrium of the SC.It also emerges that normal subjects are divided into two distinct subgroups, differing mainly in the second PC which is principally related to lipid structures in the SC. In atopic subjects, this component shows little dispersion.Conclusions Analysis of the NIR spectra did not enable normal and atopic subjects to be distinguished unequivocally but provided important information on the use of NIRS in these subjects and insights into the SC. Although the responses of water and lipid structures could not be read directly from the spectra, it was possible to decompose the global spectral information into components by PCA. For example, it was possible to observe a fraction of variance associated in different ways with water.This variance was present in more than one PC and could therefore reflect different interactions between water and other structures present in the SC (e.g., lipid and protein structures). It could also reflect compartmentalization between surface and deep water within the SC. NIRS therefore seems capable of giving spectral responses that are correlated with the overall water balance and with interactions between water and other components of the SC. References 1 Ciurczak, E.W., and Maldacker, T. A., Spectroscopy, 1986, 1, 36. 2 Ciurczak, E. W., and Torlini, R. P., Spectroscopy, 1987, 2, 41. 3 Ciurczak, E. W., and Shintani-Young, T., paper presented at the Pittsburgh Conference on Analytical Chemistry and Applied Spectroscopy, New Orleans, LA, 1985. 4 Ciurczak, E. W., paper presented at the 1st Panamerican Chemistry Conference, San Juan, Puerto Rico, 1985. 5 Ciurczak, E. W., paper presented at FACSS, Philadelphia, PA, 1985. 6 Ciurczak, E. W., paper presented at the 26th Annual Conference on Pharmaceutical Analysis, Merrimac, WI, 1985. 7 Ciurczak, E. W., paper presented at the 8th Annual Symposium on NIRA, Technicon, Tarrytown, NY, 1985. 8 Buchanan, B. R., Ciurczak, E. W., and Grunke, A. Q., Honigs Spectrosc., 1988, 3, 54. 9 Kradjel, C., and Ciurczak, E. W., paper presented at the 1st Panamerican Chemistry Conference, San Juan, Puerto Rico, 1985. 10 Lodder, R. A., Selby, M., and Hieftje, G. M., Anal. Chem., 1987, 59, 1921. 11 Lodder, R. A., and Hieftje, G. M., Appl. Spectrosc., 1988, 42, 556. 12 Lodder, R. A., and Hieftje, G. M., Appl. Spectrosc., 1988, 42, 1512. 13 Drennen, J. K., and Lodder, R. A., J. Pharm. Sci., 1990, 79, 662. 14 Tamura, M., Seyama, A., and Hazeki, O., Adv. Exp. Med. Biol., 1987, 215, 297. 15 Tamura, M., Hazeki, O., Nioka, S., Chance, B., and Smith, D. S., Adv. Exp. Med. Biol., 1988, 222, 359. 16 Hazeki, O., Seyama, A., and Tamura, M., Adv. Exp. Med. Biol., 1987, 215, 297. 17 Van Toorenenbergen, A. W., Blijenberg, B. G., and Leijnse, B., J. Clin. Chem. Clin. Biochem., 1988, 26, 209. 18 Peuchant, E., Salles, C., and Jensen, R., Anal. Chem., 1987, 59, 1816. 19 Lodder, R. A., Hieftje, G. M., Gary, M., Moorehead, W., Robertson, S. P., and Rand, P., Talanta, 1989, 36, 193. 20 Araki, R., and Nashimoto, I., Adv. Exp. Med. Biol., 1989, 248, 11. 21 Mathias, T., Wilson, S., and Maibach, H., J. Invest. Dermatol., 1981, 77, 219. 22 Tagami, H., Kanamuru, Y., and Inque, K., J. Invest. Dermatol., 1982, 78, 425. 23 Werner, Y., and Lindberg, M., Acta Dermatovenereol., 1985, 65, 102. 24 Monku, M. S., Horrobin, D. F., Morse, N. L., Wright, S., and Burton, J. L., J. Dermatol., 1984, 110, 643. 25 Rajka, G., J. Pediat. Dermatol., 1989, 8, 182. 26 Potts, R. O., Guzek, D.B., Harris, R. R., and McKie, J. E., Arch. Dermatol. Res., 1985, 277, 489. 27 Bommannan, D., Potts, R. O., and Guy, R. H., J. Controlled Release, 1991, 16, 299. Paper 7/01254C Received February 24, 1997 Accepted May 1, 1997 Fig. 9 Distribution of scores of the normal (+) and atopic (Ò) subjects in the plane of the first two PCs. 770 Analyst, August 1997, Vol. 122
ISSN:0003-2654
DOI:10.1039/a701254c
出版商:RSC
年代:1997
数据来源: RSC
|
9. |
Near-infrared Reflectance Spectrometry in the Studyof AtopyPart2. Interactions Between the Skin and Polyethylene Glycol400, Isopropyl Myristate and Hydrogel |
|
Analyst,
Volume 122,
Issue 8,
1997,
Page 771-776
E. Dreassi,
Preview
|
|
摘要:
Near-infrared Reflectance Spectrometry in the Study of Atopy Part 2.† Interactions Between the Skin and Polyethylene Glycol 400, Isopropyl Myristate and Hydrogel E. Dreassia, G. Ceramellia, P. Murab, P. L. Perruccioc, F. Vocionic, P. Bartalinid and P. Corti*a a Department of Chemical and Pharmaceutical Technology, Siena University, via Banchi di Sotto 55, 53100 Siena, Italy b Department of Pharmacological Science, Florence University, via G. Capponi 9, 50121 Florence, Italy c Military Pharmaceutical Establishment, via R.Giuliani 201, 50144 Florence, Italy d Dermatology Clinic, Siena University, viale Bracci, 53100 Siena, Italy An investigation into the existence of spectral differences and differences in response in terms of water and lipid content between normal and atopic skin after interaction with chemical agents is described. Three compounds were taken as models: a prevalently hydrophilic solvent (polyethylene glycol, PEG 400), a prevalently lipophilic solvent (isopropyl myristate) and a hydrophilic pharmaceutical (gel) used to promote contact in electrocardiography. Using principal component analysis it was possible to distinguish atopic and normal subjects by simple contact of the skin with chemical agents.Keywords: Near-infrared spectrometry; atopy; skin; polyethylene glycol 400; isopropyl myristate; hydrogel The use of spectrometry to study the skin and skin changes in response to contact with chemical agents is a developing field of great interest. The main method used to date has been attenuated total reflectance infrared spectrometry.1–3 This method and others such as differential scanning calorimetry,4,5 are the main approaches used for this type of research.It is known that interactions between the skin and chemical agents, such as solvents currently used as excipients in pharmaceuticals for topical use, cause changes in skin lipids and in the moisture balance of the stratum corneum and deeper layers. These features are interesting when near-infrared reflectance spectrometry (NIRS) is used.This method operates in a spectral region in which the responses of aliphatic chains are specific and easily identified, and the response of water is also specific and intense. These components are among the most important for differentiating normal and atopic skin. Research by our group in the field of percutaneous absorption of drugs6–11 and the use of NIRS for the qualitative and quantitative control of drugs12–17 led us to investigate whether NIRS could be used to detect skin changes and differences between the skin of normal and atopic subjects.Three compounds were taken as models: a prevalently hydrophilic solvent (polyethylene glycol, PEG 400), a prevalently lipophilic solvent (isopropyl myristate, IPM) and a hydrophilic pharmaceutical (gel) used to promote contact in electrocardiography. The last substance was used to determine the effect of keeping water in contact with the skin.The aim of this study was to determine whether NIRS is a suitable tool for detecting skin changes in response to excipients in preformulation studies, and to investigate whether the reactions of normal and atopic skin are different. Experimental NIRS Analysis The spectrometer used was a Bran + Luebbe (Nordestedt, Germany) InfraAlyzer 500 with an EDAPT optical fibre probe from the same company. The instrument was operated in the 1100–2060 nm range in 4 nm steps.Processing of Spectral Data The spectral data were processed using IDAS-PC version 1.41 (Bran + Luebbe), UNSCRAMBLER version 3.40 programmes (CAMO, Trondheim, Norway) and StatGraphics Plus version 6.0 (Manugistics, Rockville, MD, USA). Methodology The subjects were ten volunteers (five women, five men), five of whom were normal and five atopic, identified on the basis of medical history. PEG 400 (Merck, Darmstadt, Germany), IPM (Merck) and the gel Elektrogel (Ugo Palmerio, Bergamo, Italy) were used.The solvents and gel were applied at a dose of 1 g cm23 to specific, marked areas of the medial surface of the forearm, using cotton-wool. The compounds were removed 10 min later using cotton-wool in order to not irritate or abrade the skin, and the NIRS reading was taken. Other readings were taken 20, 30, 40 and 50 min after application of the compounds. A first reading was made before application of the compounds.Each volunteer repeated this test eight times (every 15 d) in the period from January to April. Results and Discussion Analysis of the raw spectra (Figs. 1 and 2) did not reveal any elements of differentiation between atopic and normal subjects or between skin response to the different compounds. However, the following observations are possible: after the treatment, there was an overall modification in spectral absorption in normal and atopic subjects; the manner in which this occurred varied with the compound tested; and the manner in which spectral absorption increased with time differed between normal and atopic subjects.Principal components analysis (PCA) of the spectra provided more significant information on possible mechanisms of † For Part 1, see p. 767. Analyst, August 1997, Vol. 122 (771–776) 771interaction between the skin and the three compounds. PCA was performed using three different approaches: (i) creating a model for each treatment including both normal and atopic subjects; (ii) creating a separate model for each treatment for normal and atopic subjects; and (iii) creating a single model including all treatment and all subjects.Approach I The following considerations arise from [Fig. 3(a)–(d)]. The curve of the first principal component (PC) [Fig. 3(a)] (93%, 94% and 96% of total variance for PEG, IPM and gel, respectively) does not show variations except for a peak at 1208–1212 nm for PEG, attributable to absorption of the C–H bonds (first overtone).The component can be identified as spectral information related to the physical state of the skin, with peaks at the absorption wavelengths of water (see Part I). The second PC [Fig. 3(b)] (3%, 3% and 2% of total variance for PEG, IPM and gel, respectively) showed marked variations. The whole wavelength band showed large differences between loadings for the three compounds, in terms of peak position and intensity.There was no peak at 1212 nm (lipids) for PEG; this information is provided by the first component. Marked differences were found in the absorption band of water, around 1450 nm, where all peaks had relative shifts (1444, 1460 and 1476 nm for gel; 1452 and 1468 nm for IPM; 1436 and 1452 nm for PEG). The absorption band of the first overtone of C–H bonds also showed shifts between loadings of the different compounds. The third PC [Fig. 3(c)] (2%, 2% and 1% of total variance for PEG, IPM and gel, respectively), although it retains information related to water bands, shows an increased influence of C–H Fig. 1 Spectra of the skin of normal subjects: blank (-) and subsequent readings [10 (½), 20 (2), 30 (5), 40 (:) and 50 min (3)]. (a) Treatment with gel; (b) treatment with IPM; and (c) treatment with PEG. Fig. 2 Spectra of the skin of the atopic subjects: blank (-) and subsequent readings [10 (½), 20 (2), 30 (5), 40 (:) and 50 min (3)]. (a) Treatment with gel; (b) treatment with IPM; and (c) treatment with PEG. 772 Analyst, August 1997, Vol. 122bond absorption bands with respect to the second PC, especially as regards the peaks at 1200 and 1724–1764 nm. For the fourth PC [Fig. 3(d)], the loadings of the compounds were considerably different, but the small percentage of the total variance for which this component accounted (about 1% in all cases) and residual background noise made it difficult to assign all the peaks with certainty.There was a significant shift of the peak around 1400 nm (probably attributable to a combination of stretching and bending of C–H bonds): 1404 nm for gel, 1380 nm for IPM and 1396 nm for PEG. The various PCs can be interpreted as follows. The first component expresses overall changes in the physical state of the skin. The second and third PCs describe the main chemical components influencing the spectrum, namely water and lipids. The differences in these two PCs indicate that the various skin changes induced by application of the compounds can largely be detected by NIRS.The three compounds seem to act with different mechanisms, as is evident in the second and third PCs. Approach II In the comparison of normal and atopic subjects, compound by compound [Fig. 4(a)–(f)], it emerged that for the gel [Fig. 4(a) and (b)], the difference between normal and atopic subjects was large in the second PC in the 1100–1200 nm band and in the band of the first overtone of water.The third and fourth PCs [Fig. 4(b)] also showed a different interaction between compound and skin in the two groups of subjects. IPM [Fig. 4(c) and (d)] caused a very different response in the third and fourth PC between atopic and normal subjects, the main feature of which was for the third PC a wide, intense response between 1300 and 1600 nm. The above observations for IPM also apply to PEG [Fig. 4(e) and (f)]. The first four principal components showed that normal and atopic subjects had different spectral modifications for a given compound, especially evident in the second and third components.They also showed that different compounds caused different skin changes. The biological phenomenon is complex and its interpretation is beyond the scope of this study; however, the NIR spectra of treated skin were certainly related to skin reaction. Water and lipids were the skin parameters most affected by treatment. Using the scores for the principal components obtained by the above approaches (I and II), we endeavoured to follow the processes occurring in skin at the end of treatment and in the subsequent period.Our aim was to determine the changes following each of the three treatments for normal and atopic subjects (Fig. 5), whether the three compounds produced different responses in time in the two classes of subjects (Fig. 6) and whether it was possible to identify the conditions under which atopic subjects gave different responses from normal subjects (Fig. 7, Tables 1 and 2). NIRS readings of the skin in the 50 min after treatment showed the following: Gel, normal subjects: treatment was not associated with significant dispersion in the distribution of the scores of the first two principal components immediately after contact. Twenty minutes from the beginning of treatment, however, the scores for blanks and treatments began to diverge [Fig. 5(a)]. Gel, atopic subjects: treatment produced an immediate increase in the dispersion of the data.After 20 min the scores of the treated group diverged from those of the blanks [Fig. 5(b)]. IPM, normal subjects: treatment did not cause a significant increase in data dispersion immediately after contact; however, by 50 min the blanks and treatment scores had diverged into two separate groups [Fig. 5(c)]. IPM, atopic subjects: immediately after treatment, atopic subjects already had a greater dispersion than normal subjects, and the blanks and treatment groups began to separate, continuing up to 30 min [Fig. 5(d)]. PEG, normal subjects: blanks and treatment data separated into two groups at about 40 min [Fig. 5(e)]. Fig. 3 Loadings of the first four PCs dividing data according to the compound tested: GEL (3); IPM (:); and PEG 400 (½). (a) 1st PC; (b) 2nd PC; (c) 3rd PC; and (d) 4th PC. Analyst, August 1997, Vol. 122 773PEG, atopic subjects: blanks and treatment did not separate under any condition [Fig. 5(f)]. Comparison of the patterns in normal subjects [Fig. 5(a), (c) and (e)] and atopic subjects [Fig. 5(b), (d) and (f)] treated with the different compounds gave rise to the following considerations: for gel treatment, the dispersion of the data for atopic subjects was always much larger than for normal subjects; the same was observed for IPM treatment and it is remarkable that the lipophilic solvent has a much more rapid effect on the skin of atopic subjects than on normal skin, for the former causing substantial spectral modification immediately after the end of the treatment whereas the latter was affected only after 50 min.The present results show that each compound has different effects on the skin according to the class of subject. Atopic subjects seem to be much more sensitive than normal subjects: their skin reacts more strongly and the effects are lasting. During the 50 min of the experiments, the skin changes did not show a genuine kinetic pattern.It was clear, however, that the skin, perturbed by contact with the chemical agent, changed progressively with time without showing signs of recovery in the 50 min period. Approach III Using the PCs from the single model including all treatments on all subjects, we examined whether the changes caused by each compound were different in the two classes of subjects on the first four principal components by applying discriminant analysis, and the results were as follows.In normal subjects there was a random distribution of data in the first 20 min; after 30 min this distribution separated into three groups corresponding to the three compounds tested (Fig. 6), with only one case of treatment with PEG falling in the IPM group. In atopic subjects, the data separated into three groups [Fig. 6(b)], in relation to the three compounds, at the end of the period of contact (10 min). This separation was maintained for the rest of the observation period.This comparison shows that atopic subjects have a greater sensitivity to chemical agents and persistence of effects than normal subjects. The fact that the possibility of differentiating the effects of the different chemicals is time dependent suggests that the effects of the three compounds have time courses which are different from each other and different for normal and atopic subjects. The fact that normal and atopic subjects can be distinguished within the three treatment groups emerged when cluster analysis (CA) was applied to the PCA data, using the scores of the first four principal components, divided according to the time elapsing since application of the compounds.Fig. 7 and Table 1 show the results obtained in the assignment to the two classes by CA at various times after application of the compounds for normal and atopic subjects. Again, the close relationship between reading time and separation of the two classes of subjects was evident, as was the fact that this relationship varied according to the compound tested.Fig. 4 Loadings of the first four PCs dividing data according to the compound tested and class of subject (normal and atopic). (a), (c) and (e) 1st (8, atopic; -, normal) and 2nd (2, normal; 5, atopic) PCs for gel, IPM and PEG, respectively; (b), (d) and (f) 3rd (8, atopic; -, normal) and 4th (2, normal; 5, atopic) PCs for gel, IPM and PEG, respectively. 774 Analyst, August 1997, Vol. 122The best results, in terms of correct assignment to the appropriate class of subjects, were obtained after 30 min for gel [100% correct assignment, Fig. 7(a)], and after 20 min for IPM and PEG [90% correct assignment, Fig. 7(b) and (c)]. Discriminant analysis confirmed the results of cluster analysis, giving correct assignment percentages similar to the above, and better in the cases of IPM and PEG (Table 2). The differences noted previously in the pattern of the compounds tested were confirmed by discriminant analysis, especially as regards the times at which discrimination between the two classes of subjects was a maximum.Fig. 5 Changes caused by treatment with the different compounds with respect to initial skin state for normal and atopic subjects. Representation of blanks (2) and readings at the time of maximum differentiation 20 (2), 30 (3), 40 (4) and 50 (5) min in the plane of the first two PCs for (a) gel and normal subjects, (b) gel and atopic subjects, (c) IPM and normal subjects, (d) IPM and atopic subjects, (e) PEG and normal subjects and (f) PEG and atopic subjects. Fig. 6 Comparison of changes induced by skin treatment [gel (1), IPM (2) and PEG (3)] in normal and atopic subjects. Discriminant analysis of the scores of the first four PCs: (a) normal subjects and (b) atopic subjects. Fig. 7 Differentiation between normal and atopic subjects by cluster analysis. (a) Gel, 30 min; (b) IPM, 20 min; and (c) PEG, 20 min.Table 1 Results in terms of correct assignment to the appropriate class of subjects by cluster analysis Correct assignment (%) Time/ min PEG IPM Gel 10 55 40 55 20 90 90 75 30 90 90 100 40 60 50 55 50 35 30 35 Table 2 Results in terms of correct assignment to the appropriate class of subjects by discriminant analysis Correct assignment (%) Time/ min PEG IPM Gel 10 95 100 75 20 100 100 90 30 100 100 90 40 60 60 60 50 40 50 30 Analyst, August 1997, Vol. 122 775Conclusions Although the present study is of a preliminary nature, the results have several interesting features. In the spectral range from 1100 to 2060 nm, NIRS provides analytical data that, after appropriate statistical analysis, enable normal and atopic subjects to be distinguished on the basis of skin changes in response to each of the three compounds tested. The compounds caused marked changes in the spectral responses of the skin. Irrespective of the compound applied and the class of subject, there was a considerable change in the skin spectrum after exposure to the chemical agent for 10 min.When the spectral data were analysed by PCA, four principal components that distinguished atopic and normal skin were identified. The first component was independent of the compound tested and the class of subject. Since it accounted for a high percentage of total variance in each of the three approaches, it is presumably related to the basic state of the skin.The second and third components showed considerable differences and are explained by the different distributions of water in the surface and deeper layers of the skin and by the different interaction between water and lipids. NIRS was therefore confirmed as suitable for evaluating the skin moisture balance and changes in lipid content. None of the principal components was related to the compounds tested. Since percutaneous absorption of drugs is generally affected by the excipients in the formulation, the present results show that each of the compounds (enhancers) tested has a clear effect on the state of the skin and would therefore affect drug absorption.The mobilization and redistribution of epidermal and dermal moisture, observed to vary in relation to the compound applied and the class of the subject, certainly underlie the changes in absorption observed for a given drug with different excipients. The results also give rise to some biological considerations.Fig. 7 shows that normal and atopic subjects have different skin reactions to contact with the compounds tested. For gel and IPM, both subjects had an initial increase in data dispersion, followed by different reorganization phases leading to separation of the two classes of subject. The process seemed faster but more transient for the gel, but in any case the patterns were different. This is in line with the data in Table 1. The three compounds tested find use both as common enhancers for promoting percutaneous absorption and as excipients in topical preparations. This raises the question of whether drug formulations containing these excipients have the same effect on normal and atopic subjects, in view of the different skin changes observed in the present study.References 1 Potts, R. O., Guzek, D. B., Harris, R. R., and McKie, J. E., Arch. Dermatol. Res., 1985, 277, 489. 2 Mak, V. H. W., Potts, R. O., and Guy, R.H., J. Controlled Release, 1990, 12, 67. 3 Bommannan, D., Potts, R. O., and Guy, R. H., J. Controlled Release, 1991, 16, 299. 4 Golden, G. M., Guzek, D. B., Harris, R. R., McKie, J. E., and Potts, R. O., J. Invest. Dermatol., 1986, 86, 255. 5 Golden, G. M., McKie, J. E., and Potts, R. O., J. Pharm. Sci., 1987, 76, 25. 6 Corti, P., Savini, L., Dreassi, E., Ceramelli, G., Montecchi, L., and Lonardi, S., Pharm. Acta Helv., 1992, 67, 57. 7 Corti, P., Savini, L., Dreassi, E., Petriconi, S., Genga, R., Montecchi, L., and Lonardi, S., Process Control Qual., 1992, 2, 131. 8 Dreassi, E., Ceramelli, G., Savini, L., Corti, P., Perruccio, P. L., and Lonardi, S., Analyst, 1995, 120, 319. 9 Dreassi, E., Ceramelli, G., Corti, P., Lonardi S., and Perruccio, P. L., Analyst, 1995, 120, 1005. 10 Dreassi, E., Ceramelli, G., Corti, P., Massacesi, M., and Perruccio, P. L., Analyst, 1995, 120, 2361. 11 Dreassi, E., Ceramelli, G., Corti, P., Perruccio, P. L., and Lonardi, S., Analyst, 1996, 121, 219. 12 Mura, P., Liguori, A., Bramanti, G., Corti, P., Murratzu, C., and Celesti, L., Pharm. Acta Helv., 1990, 65, 298. 13 Celesti, L., Murratzu, C., Valoti, M., Corti, P., and Sgaragli, G. P., Methods Find. Exp. Clin. Pharmacol., 1992, 14, 701. 14 Celesti, L., Murratzu, C., Valoti, M., Sgaragli, G. P., and Corti, P., Methods Find. Exp. Clin. Pharmacol., 1993, 15, 49. 15 Mura, P., Celesti, L., Valoti, M., Corti, P., and Santoni, P., Eur. J. Pharm. Biopharm., 1994, 40, 90. 16 Mura, P., Celesti, L., Proietti, D., Corsi, S., Furlanetto, S., and Corti, P., Acta Tecnol. Legis Med., 1993, 4, 121. 17 Mura, P., Nassini, C., Proietti, D., Manderioli, A., and Corti, P., Pharm. Acta Helv., 1995, 70, 175. Paper 7/01257H Received February 2, 1997 Accepted May 1, 1997 776 Analyst, August 1997, Vol. 122 Near-infrared Reflectance Spectrometry in the Study of Atopy Part 2.† Interactions Between the Skin and Polyethylene Glycol 400, Isopropyl Myristate and Hydrogel E.Dreassia, G. Ceramellia, P. Murab, P. L. Perruccioc, F. Vocionic, P. Bartalinid and P. Corti*a a Department of Chemical and Pharmaceutical Technology, Siena University, via Banchi di Sotto 55, 53100 Siena, Italy b Department of Pharmacological Science, Florence University, via G. Capponi 9, 50121 Florence, Italy c Military Pharmaceutical Establishment, via R. Giuliani 201, 50144 Florence, Italy d Dermatology Clinic, Siena University, viale Bracci, 53100 Siena, Italy An investigation into the existence of spectral differences and differences in response in terms of water and lipid content between normal and atopic skin after interaction with chemical agents is described.Three compounds were taken as models: a prevalently hydrophilic solvent (polyethylene glycol, PEG 400), a prevalently lipophilic solvent (isopropyl myristate) and a hydrophilic pharmaceutical (gel) used to promote contact in electrocardiography. Using principal component analysis it was possible to distinguish atopic and normal subjects by simple contact of the skin with chemical agents.Keywords: Near-infrared spectrometry; atopy; skin; polyethylene glycol 400; isopropyl myristate; hydrogel The use of spectrometry to study the skin and skin changes in response to contact with chemical agents is a developing field of great interest. The main method used to date has been attenuated total reflectance infrared spectrometry.1–3 This method and others such as differential scanning calorimetry,4,5 are the main approaches used for this type of research.It is known that interactions between the skin and chemical agents, such as solvents currently used as excipients in pharmaceuticals for topical use, cause changes in skin lipids and in the moisture balance of the stratum corneum and deeper layers. These features are interesting when near-infrared reflectance spectrometry (NIRS) is used. This method operates in a spectral region in which the responses of aliphatic chains are specific and easily identified, and the response of water is also specific and intense.These components are among the most important for differentiating normal and atopic skin. Research by our group in the field of percutaneous absorption of drugs6–11 and the use of NIRS for the qualitative and quantitative control of drugs12–17 led us to investigate whether NIRS could be used to detect skin changes and differences between the skin of normal and atopic subjects. Three compounds were taken as models: a prevalently hydrophilic solvent (polyethylene glycol, PEG 400), a prevalently lipophilic solvent (isopropyl myristate, IPM) and a hydrophilic pharmaceutical (gel) used to promote contact in electrocardiography. The last substance was used to determine the effect of keeping water in contact with the skin.The aim of this study was to determine whether NIRS is a suitable tool for detecting skin changes in response to excipients in preformulation studies, and to investigate whether the reactions of normal and atopic skin are different.Experimental NIRS Analysis The spectrometer used was a Bran + Luebbe (Nordestedt, Germany) InfraAlyzer 500 with an EDAPT optical fibre probe from the same company. The instrument was operated in the 1100–2060 nm range in 4 nm steps. Processing of Spectral Data The spectral data were processed using IDAS-PC version 1.41 (Bran + Luebbe), UNSCRAMBLER version 3.40 programmes (CAMO, Trondheim, Norway) and StatGraphics Plus version 6.0 (Manugistics, Rockville, MD, USA).Methodology The subjects were ten volunteers (five women, five men), five of whom were normal and five atopic, identified on the basis of medical history. PEG 400 (Merck, Darmstadt, Germany), IPM (Merck) and the gel Elektrogel (Ugo Palmerio, Bergamo, Italy) were used. The solvents and gel were applied at a dose of 1 g cm23 to specific, marked areas of the medial surface of the forearm, using cotton-wool.The compounds were removed 10 min later using cotton-wool in order to not irritate or abrade the skin, and the NIRS reading was taken. Other readings were taken 20, 30, 40 and 50 min after application of the compounds. A first reading was made before application of the compounds. Each volunteer repeated this test eight times (every 15 d) in the period from January to April. Results and Discussion Analysis of the raw spectra (Figs. 1 and 2) did not reveal any elements of differentiation between atopic and normal subjects or between skin response to the different compounds. However, the following observations are possible: after the treatment, there was an overall modification in spectral absorption in normal and atopic subjects; the manner in which this occurred varied with the compound tested; and the manner in which spectral absorption increased with time differed between normal and atopic subjects.Principal components analysis (PCA) of the spectra provided more significant information on possible mechanisms of † For Part 1, see p. 767. Analyst, August 1997, Vol. 122 (771–776) 771interaction between the skin and the three compounds. PCA was performed using three different approaches: (i) creating a model for each treatment including both normal and atopic subjects; (ii) creating a separate model for each treatment for normal and atopic subjects; and (iii) creating a single model including all treatment and all subjects.Approach I The following considerations arise from [Fig. 3(a)–(d)]. The curve of the first principal component (PC) [Fig. 3(a)] (93%, 94% and 96% of total variance for PEG, IPM and gel, respectively) does not show variations except for a peak at 1208–1212 nm for PEG, attributable to absorption of the C–H bonds (first overtone). The component can be identified as spectral information related to the physical state of the skin, with peaks at the absorption wavelengths of water (see Part I).The second PC [Fig. 3(b)] (3%, 3% and 2% of total variance for PEG, IPM and gel, respectively) showed marked variations. The whole wavelength band showed large differences between loadings for the three compounds, in terms of peak position and intensity. There was no peak at 1212 nm (lipids) for PEG; this information is provided by the first component. Marked differences were found in the absorption band of water, around 1450 nm, where all peaks had relative shifts (1444, 1460 and 1476 nm for gel; 1452 and 1468 nm for IPM; 1436 and 1452 nm for PEG).The absorption band of the first overtone of C–H bonds also showed shifts between loadings of the different compounds. The third PC [Fig. 3(c)] (2%, 2% and 1% of total variance for PEG, IPM and gel, respectively), although it retains information related to water bands, shows an increased influence of C–H Fig. 1 Spectra of the skin of normal subjects: blank (-) and subsequent readings [10 (½), 20 (2), 30 (5), 40 (:) and 50 min (3)].(a) Treatment with gel; (b) treatment with IPM; and (c) treatment with PEG. Fig. 2 Spectra of the skin of the atopic subjects: blank (-) and subsequent readings [10 (½), 20 (2), 30 (5), 40 (:) and 50 min (3)]. (a) Treatment with gel; (b) treatment with IPM; and (c) treatment with PEG. 772 Analyst, August 1997, Vol. 122bond absorption bands with respect to the second PC, especially as regards the peaks at 1200 and 1724–1764 nm.For the fourth PC [Fig. 3(d)], the loadings of the compounds were considerably different, but the small percentage of the total variance for which this component accounted (about 1% in all cases) and residual background noise made it difficult to assign all the peaks with certainty. There was a significant shift of the peak around 1400 nm (probably attributable to a combination of stretching and bending of C–H bonds): 1404 nm for gel, 1380 nm for IPM and 1396 nm for PEG. The various PCs can be interpreted as follows.The first component expresses overall changes in the physical state of the skin. The second and third PCs describe the main chemical components influencing the spectrum, namely water and lipids. The differences in these two PCs indicate that the various skin changes induced by application of the compounds can largely be detected by NIRS. The three compounds seem to act with different mechanisms, as is evident in the second and third PCs.Approach II In the comparison of normal and atopic subjects, compound by compound [Fig. 4(a)–(f)], it emerged that for the gel [Fig. 4(a) and (b)], the difference between normal and atopic subjects was large in the second PC in the 1100–1200 nm band and in the band of the first overtone of water. The third and fourth PCs [Fig. 4(b)] also showed a different interaction between compound and skin in the two groups of subjects.IPM [Fig. 4(c) and (d)] caused a very different response in the third and fourth PC between atopic and normal subjects, the main feature of which was for the third PC a wide, intense response between 1300 and 1600 nm. The above observations for IPM also apply to PEG [Fig. 4(e) and (f)]. The first four principal components showed that normal and atopic subjects had different spectral modifications for a given compound, especially evident in the second and third components.They also showed that different compounds caused different skin changes. The biological phenomenon is complex and its interpretation is beyond the scope of this study; however, the NIR spectra of treated skin were certainly related to skin reaction. Water and lipids were the skin parameters most affected by treatment. Using the scores for the principal components obtained by the above approaches (I and II), we endeavoured to follow the processes occurring in skin at the end of treatment and in the subsequent period.Our aim was to determine the changes following each of the three treatments for normal and atopic subjects (Fig. 5), whether the three compounds produced different responses in time in the two classes of subjects (Fig. 6) and whether it was possible to identify the conditions under which atopic subjects gave different responses from normal subjects (Fig. 7, Tables 1 and 2). NIRS readings of the skin in the 50 min after treatment showed the following: Gel, normal subjects: treatment was not associated with significant dispersion in the distribution of the scores of the first two principal components immediately after contact.Twenty minutes from the beginning of treatment, however, the scores for blanks and treatments began to diverge [Fig. 5(a)]. Gel, atopic subjects: treatment produced an immediate increase in the dispersion of the data. After 20 min the scores of the treated group diverged from those of the blanks [Fig. 5(b)]. IPM, normal subjects: treatment did not cause a significant increase in data dispersion immediately after contact; however, by 50 min the blanks and treatment scores had diverged into two separate groups [Fig. 5(c)]. IPM, atopic subjects: immediately after treatment, atopic subjects already had a greater dispersion than normal subjects, and the blanks and treatment groups began to separate, continuing up to 30 min [Fig. 5(d)]. PEG, normal subjects: blanks and treatment data separated into two groups at about 40 min [Fig. 5(e)]. Fig. 3 Loadings of the first four PCs dividing data according to the compound tested: GEL (3); IPM (:); and PEG 400 (½). (a) 1st PC; (b) 2nd PC; (c) 3rd PC; and (d) 4th PC. Analyst, August 1997, Vol. 122 773PEG, atopic subjects: blanks and treatment did not separate under any condition [Fig. 5(f)]. Comparison of the patterns in normal subjects [Fig. 5(a), (c) and (e)] and atopic subjects [Fig. 5(b), (d) and (f)] treated with the different compounds gave rise to the following considerations: for gel treatment, the dispersion of the data for atopic subjects was always much larger than for normal subjects; the same was observed for IPM treatment and it is remarkable that the lipophilic solvent has a much more rapid effect on the skin of atopic subjects than on normal skin, for the former causing substantial spectral modification immediately after the end of the treatment whereas the latter was affected only after 50 min.The present results show that each compound has different effects on the skin according to the class of subject. Atopic subjects seem to be much more sensitive than normal subjects: their skin reacts more strongly and the effects are lasting. During the 50 min of the experiments, the skin changes did not show a genuine kinetic pattern. It was clear, however, that the skin, perturbed by contact with the chemical agent, changed progressively with time without showing signs of recovery in the 50 min period. Approach III Using the PCs from the single model including all treatments on all subjects, we examined whether the changes caused by each compound were different in the two classes of subjects on the first four principal components by applying discriminant analysis, and the results were as follows.In normal subjects there was a random distribution of data in the first 20 min; after 30 min this distribution separated into three groups corresponding to the three compounds tested (Fig. 6), with only one case of treatment with PEG falling in the IPM group. In atopic subjects, the data separated into three groups [Fig. 6(b)], in relation to the three compounds, at the end of the period of contact (10 min). This separation was maintained for the rest of the observation period. This comparison shows that atopic subjects have a greater sensitivity to chemical agents and persistence of effects than normal subjects.The fact that the possibility of differentiating the effects of the different chemicals is time dependent suggests that the effects of the three compounds have time courses which are different from each other and different for normal and atopic subjects. The fact that normal and atopic subjects can be distinguished within the three treatment groups emerged when cluster analysis (CA) was applied to the PCA data, using the scores of the first four principal components, divided according to the time elapsing since application of the compounds.Fig. 7 and Table 1 show the results obtained in the assignment to the two classes by CA at various times after application of the compounds for normal and atopic subjects. Again, the close relationship between reading time and separation of the two classes of subjects was evident, as was the fact that this relationship varied according to the compound tested. Fig. 4 Loadings of the first four PCs dividing data according to the compound tested and class of subject (normal and atopic).(a), (c) and (e) 1st (8, atopic; -, normal) and 2nd (2, normal; 5, atopic) PCs for gel, IPM and PEG, respectively; (b), (d) and (f) 3rd (8, atopic; -, normal) and 4th (2, normal; 5, atopic) PCs for gel, IPM and PEG, respectively. 774 Analyst, August 1997, Vol. 122The best results, in terms of correct assignment to the appropriate class of subjects, were obtained after 30 min for gel [100% correct assignment, Fig. 7(a)], and after 20 min for IPM and PEG [90% correct assignment, Fig. 7(b) and (c)]. Discriminant analysis confirmed the results of cluster analysis, giving correct assignment percentages similar to the above, and better in the cases of IPM and PEG (Table 2). The differences noted previously in the pattern of the compounds tested were confirmed by discriminant analysis, especially as regards the times at which discrimination between the two classes of subjects was a maximum.Fig. 5 Changes caused by treatment with the different compounds with respect to initial skin state for normal and atopic subjects. Representation of blanks (2) and readings at the time of maximum differentiation 20 (2), 30 (3), 40 (4) and 50 (5) min in the plane of the first two PCs for (a) gel and normal subjects, (b) gel and atopic subjects, (c) IPM and normal subjects, (d) IPM and atopic subjects, (e) PEG and normal subjects and (f) PEG and atopic subjects.Fig. 6 Comparison of changes induced by skin treatment [gel (1), IPM (2) and PEG (3)] in normal and atopic subjects. Discriminant analysis of the scores of the first four PCs: (a) normal subjects and (b) atopic subjects. Fig. 7 Differentiation between normal and atopic subjects by cluster analysis. (a) Gel, 30 min; (b) IPM, 20 min; and (c) PEG, 20 min. Table 1 Results in terms of correct assignment to the appropriate class of subjects by cluster analysis Correct assignment (%) Time/ min PEG IPM Gel 10 55 40 55 20 90 90 75 30 90 90 100 40 60 50 55 50 35 30 35 Table 2 Results in terms of correct assignment to the appropriate class of subjects by discriminant analysis Correct assignment (%) Time/ min PEG IPM Gel 10 95 100 75 20 100 100 90 30 100 100 90 40 60 60 60 50 40 50 30 Analyst, August 1997, Vol. 122 775Conclusions Although the present study is of a preliminary nature, the results have several interesting features.In the spectral range from 1100 to 2060 nm, NIRS provides analytical data that, after appropriate statistical analysis, enable normal and atopic subjects to be distinguished on the basis of skin changes in response to each of the three compounds tested. The compounds caused marked changes in the spectral responses of the skin. Irrespective of the compound applied and the class of subject, there was a considerable change in the skin spectrum after exposure to the chemical agent for 10 min.When the spectral data were analysed by PCA, four principal components that distinguished atopic and normal skin were identified. The first component was independent of the compound tested and the class of subject. Since it accounted for a high percentage of total variance in each of the three approaches, it is presumably related to the basic state of the skin. The second and third components showed considerable differences and are explained by the different distributions of water in the surface and deeper layers of the skin and by the different interaction between water and lipids.NIRS was therefore confirmed as suitable for evaluating the skin moisture balance and changes in lipid content. None of the principal components was related to the compounds tested. Since percutaneous absorption of drugs is generally affected by the excipients in the formulation, the present results show that each of the compounds (enhancers) tested has a clear effect on the state of the skin and would therefore affect drug absorption.The mobilization and redistribution of epidermal and dermal moisture, observed to vary in relation to the compound applied and the class of the subject, certainly underlie the changes in absorption observed for a given drug with different excipients. The results also give rise to some biological considerations. Fig. 7 shows that normal and atopic subjects have different skin reactions to contact with the compounds tested.For gel and IPM, both subjects had an initial increase in data dispersion, followed by different reorganization phases leading to separation of the two classes of subject. The process seemed faster but more transient for the gel, but in any case the patterns were different. This is in line with the data in Table 1. The three compounds tested find use both as common enhancers for promoting percutaneous absorption and as excipients in topical preparations. This raises the question of whether drug formulations containing these excipients have the same effect on normal and atopic subjects, in view of the different skin changes observed in the present study. References 1 Potts, R. O., Guzek, D. B., Harris, R. R., and McKie, J. E., Arch. Dermatol. Res., 1985, 277, 489. 2 Mak, V. H. W., Potts, R. O., and Guy, R. H., J. Controlled Release, 1990, 12, 67. 3 Bommannan, D., Potts, R. O., and Guy, R. H., J. Controlled Release, 1991, 16, 299. 4 Golden, G. M., Guzek, D. B., Harris, R. R., McKie, J. E., and Potts, R. O., J. Invest. Dermatol., 1986, 86, 255. 5 Golden, G. M., McKie, J. E., and Potts, R. O., J. Pharm. Sci., 1987, 76, 25. 6 Corti, P., Savini, L., Dreassi, E., Ceramelli, G., Montecchi, L., and Lonardi, S., Pharm. Acta Helv., 1992, 67, 57. 7 Corti, P., Savini, L., Dreassi, E., Petriconi, S., Genga, R., Montecchi, L., and Lonardi, S., Process Control Qual., 1992, 2, 131. 8 Dreassi, E., Ceramelli, G., Savini, L., Corti, P., Perruccio, P. L., and Lonardi, S., Analyst, 1995, 120, 319. 9 Dreassi, E., Ceramelli, G., Corti, P., Lonardi S., and Perruccio, P. L., Analyst, 1995, 120, 1005. 10 Dreassi, E., Ceramelli, G., Corti, P., Massacesi, M., and Perruccio, P. L., Analyst, 1995, 120, 2361. 11 Dreassi, E., Ceramelli, G., Corti, P., Perruccio, P. L., and Lonardi, S., Analyst, 1996, 121, 219. 12 Mura, P., Liguori, A., Bramanti, G., Corti, P., Murratzu, C., and Celesti, L., Pharm. Acta Helv., 1990, 65, 298. 13 Celesti, L., Murratzu, C., Valoti, M., Corti, P., and Sgaragli, G. P., Methods Find. Exp. Clin. Pharmacol., 1992, 14, 701. 14 Celesti, L., Murratzu, C., Valoti, M., Sgaragli, G. P., and Corti, P., Methods Find. Exp. Clin. Pharmacol., 1993, 15, 49. 15 Mura, P., Celesti, L., Valoti, M., Corti, P., and Santoni, P., Eur. J. Pharm. Biopharm., 1994, 40, 90. 16 Mura, P., Celesti, L., Proietti, D., Corsi, S., Furlanetto, S., and Corti, P., Acta Tecnol. Legis Med., 1993, 4, 121. 17 Mura, P., Nassini, C., Proietti, D., Manderioli, A., and Corti, P., Pharm. Acta Helv., 1995, 70, 175. Paper 7/01257H Received February 2, 1997 Accepted May 1, 1997 776 Analyst, August 1997, Vol. 122
ISSN:0003-2654
DOI:10.1039/a701257h
出版商:RSC
年代:1997
数据来源: RSC
|
10. |
Determination of Finishing Oils in Acrylic Fibres by Near-infraredReflectance Spectrometry |
|
Analyst,
Volume 122,
Issue 8,
1997,
Page 777-781
Marcelo Blanco,
Preview
|
|
摘要:
Determination of Finishing Oils in Acrylic Fibres by Near-infrared Reflectance Spectrometry Marcelo Blanco*, Jordi Coello, Jos�e M. Garc�ýa Fraga†, Hortensia Iturriaga, Santiago Maspoch and Jordi Pag`es Departament de Qu�ýmica, Unitat de Qu�ýmica Anal�ýtica, Universitat Aut`onoma de Barcelona, E-08193 Bellaterra, Barcelona, Spain. E-mail: iqanl@cc.uab.es The potential of near-infrared diffuse reflectance spectrometry for quality control analyses in the textile industry was explored with a view to the quantification of finishing oils in acrylic fibres by partial least-squares regression, using a rotary cuvette system for recording spectra.Calibration was performed with a set of samples that encompassed every source of variability (linear density of the fibres, colour, concentration of the finishing oil), and the wavelength region where absorption was mostly due to the oil was used to construct several models from which that leading to the minimum relative standard error for a sample test set was selected.The results provided by various mathematical treatments [secondderivative, standard normal variate (SNV)] used to minimize scattering resulting from the differential linear density of the samples revealed no significant differences between prediction errors (only in the number of partial least-squares components). The model was used to quantify levels of finishing oil in routinely manufactured samples for a period of 6 months, during which time two batches showed poor predictions due to a new component appearing in the product.Modification of the calibration model to account for this component substantially increased robustness and allowed the accurate quantification of all batches manufactured after the model has been developed. Keywords: Near-infrared spectrometry; finishing oil; acrylic fibres; standard normal variate; partial least-squares calibration The ease with which sample spectra can be recorded, the need for no sample treatment and the ability to determine both physical and chemical parameters have made near-infrared diffuse reflectance spectrometry (NIRRS) a highly appealing technique for industrial control analyses.The substantially increased productivity derived from its use and the resulting expeditiousness with which useful information can be acquired and educated decisions made as a result have fostered its application in domains that could hardly have envisaged its use only a few years ago.In fact, NIR spectrometry has proved a powerful analytical tool for analysing a wide variety of samples used by the pharmaceutical,1–3 agricultural,4,5 nutritional,6,7 petrochemical8,9 and textile industries,10–13 among others, and is gaining ground in the clinical14,15 and related domains. An NIR spectrum consists of the bands for overtones and combinations of tones of fundamental bands (mainly hydrogen bonds, which absorb in the mid-IR region).The spectrum contains not only chemical information on the sample constituents, but also physical information (e.g., crystal habit, particle size, degree of crystallinity) that influences the former. Correlating the spectrum of a sample with its composition is often a difficult task that entails using an appropriate chemometric approach in order to ensure correct quantification. When a chemical compound is to be quantified, the dependence of the signal on the physical variability of the sample must be corrected.There are a number of available mathematical treatments for correcting these effects; most rely on derivative spectra,16,17 which exhibit better band resolution and decreased signal-to-noise ratio, making them unsuitable for determining analytes present at low concentrations. Some textile fibres are subjected to finishing procedures in order to improve properties such as abrasion, wrinkling and fire resistance, appearance and feel.The treatment applied depends on the particular type of fibre and its intended use. During the manufacturing process, fibres are subjected to treatments that introduce substances intended to condition them for the subsequent processing steps. The products used for this purpose are known as ‘finishing oils’ and facilitate lubrication, reduce static electricity, improve filament cohesion, etc. The large physical and chemical differences between natural and synthetic fibres, and hence between their applications, have fostered the use of a variety of products for these purposes, either alone or in combination.Ensuring optimum quality in a product requires careful control of the amount of finishing oil that is deposited on the fibres so that these behave as expected in the textile process; in fact, using too much or too little oil can lead to various problems and considerably diminished productivity. A rapid, reliable determination of the finishing oil content in fibres is thus a suitable means for identifying the need to correct the operating conditions of the manufacturing process in order to ensure the expected product quality.Currently available methods for fibre analysis typically use chlorinated solvents to remove the finishing oil from the fibres and measure the resulting solution by IR spectrometry. This methodology is laborious and time consuming; also, chlorinated solvents are toxic and environmentally aggressive, so their use is gradually being discouraged (e.g., in the EPA regulations for organic solvents and solvent waste).Recently,18 supercritical CO2 was used as a replacement solvent in the extraction and determination of finishing oils in order to avoid these shortcomings; however, the ensuing method is slow and entails optimizing the extraction conditions for each type of finishing. Ideally, an analytical method should use no organic solvents; also, it should be expeditious and provide reliable, reproducible results for the intended purpose. In this work, the joint use of NIRRS and multivariate calibration techniques [partial least-squares (PLS)] for determining finishing oils in acrylic fibres was explored.The effect of fibre properties on the quantification parameters was examined in order to develop a robust method fit for continuous implementation at a production line. Experimental Samples The samples used consisted of acrylic fibre filaments composed of acetonitrile–vinyl acetate (90 + 10 m/m) copolymer.The † Present address: Departamento de Qu�ýmica Anal�ýtica, Nutrici�on y Bromatolog�ýa, Universidad de La Laguna, E-38071 La Laguna, Tenerife, Spain. Analyst, August 1997, Vol. 122 (777–781) 777acrylic fibre studied is obtained by passing a concentrated solution of the copolymer through a spinning nozzle immersed in a cold water bath in order to produce filaments that are subjected to various stretching and drying operations, the diameter of the fibres obtained depending on the extent of stretching of the filaments—the linear density of the fibres used ranged from 1.0 to 5.40 dtex (dtex = tenths of milligram per linear metre of fibre).Then, fibre bundles are immersed in a dyeing bath to obtain the desired colour. Samples vary widely in colour, brightness (matt, semi-matt, glossy) and colour depth. Finally, the fibres are immersed in a finishing oil bath (an aqueous dispersion of the product) and dried. The finishing oil contents in the samples studied ranged from 0.22 to 0.62% by mass, depending on linear density and intended use.Apparatus The experimental set-up used consisted of a NIRSystems 6500 NIR spectrometer equipped with a reflectance detector and a spinning module that afforded recording of spectra while the cuvette was spun and provided an average spectrum after a preset number of scans. In this way, the potential effect of sample heterogeneity was suppressed. A Nicolet FT-IR 50 spectrometer and cuvettes with CaF2 windows were used to record the spectra of the finishing oil, using the reference method. Software Two different types of software were used, namely: (i) NSAS (Near-Infrared Spectral Analysis Software) version 3.25, from NIRSystems, which allows acquisition of the sample spectrum and mathematical processeraging, obtainment of derivative spectra, etc.).(ii) Unscrambler version 5.03, from CAMO A/S, which was used to implement partial least-squares regression (PLSR), identify potential outliers and locate the spectral regions of analytical interest.It also allows whole spectra and selected spectral regions to be statistically processed. The spectra thus recorded were converted to the format used by Unscrambler using JCAMP. Determination of Finishing Oil in Acrylic Fibres Reference procedure A 1 g amount of dry fibre was extracted with 50 ml of CCl4 for 10 min under magnetic stirring.The extract was filtered, its IR spectrum recorded and the absorbance in the spectral region 2800–3000 cm21 measured. The absorbance values thus obtained were interpolated into a calibration graph obtained by applying the same procedure to a set of standard samples containing known amounts of finishing oil. The standard samples used to construct the calibration graph were analysed by using a gravimetric method involving the extraction of a larger amount of sample and evaporating the solvent to dryness. All samples were analysed in duplicate.Recording of NIR Spectra Each spectrum was the average of 32 scans over the wavelength range 1100–2500 nm at 2 nm intervals. Prior to each sample spectrum, a reference spectrum was obtained for a porcelain dish. The spectra subsequently processed were the average of three aliquots of sample. The spectrum of the finishing oil (Fig. 1) was obtained by extracting the aqueous dispersion, used to soak the fibres, into CCl4 and concentrating the organic phase.Data Processing The calibration model used to determine the finishing oil was obtained by using PLSR methodology.19,20 The calibration model was constructed by optimizing the spectral mode, working wavelength range and number of principal components used in the regression in order to minimize the error of prediction for the samples. As can be seen from Fig. 1, the finishing oil only exhibited absorption in the spectral regions 1700–1800 and 2250–2370 nm.The bands in the former region were rather weak and masked by the signal of the acrylic fibre copolymer, so only the latter region was used to construct the calibration model. In order to compare the predictive capacity of the different models tested, the relative standard error for the calibration samples (%RSEC), test samples (%RSET) and external validation samples (%RSEEV) was calculated from % ( ) RSE NIR REF REF = - � = = å å C C C i i i n i n 2 1 2 1 100 (1) where n is the number of samples included in the calibration matrix, the test set or the external validation set, and CREF and CNIR are the sample concentrations provided by the reference and NIR methods, respectively.As noted in the Introduction, diffuse reflectance measurements are influenced by scattered light, the amount of which depends on particle size, crystal habit, etc.21,22 Fig. 2 shows the NIR spectra of five samples containing the same concentration of finishing oil but possessing a different linear density.As can be seen, the effect of the linear density of the fibres was comparable to the baseline shift caused by differences in particle size. In order to minimize this effect, various mathematical treatments based on suppression of the baseline shift were tested. Results and Discussion In order to ensure that the calibration model used would provide accurate predictions, the sample set was split into two sub-sets Fig. 1 NIR spectrum of the finishing oil in CCl4. 778 Analyst, August 1997, Vol. 122consisting of 83 samples for calibration (calibration set) and another 94 for determining the predictive capacity of the model (test set). The best choice was selected from among a large number of PLS calibration models. Finding such a model entailed considering the following parameters: finishing oil concentration, fibre linear density, working wavelength range, spectral mode and number of PLS factors.The samples initially included in the calibration matrix were chosen in such a way as to encompass every possible source of variability in the spectra in order to make it representative of the whole sample population. Thus, the calibration samples were evenly distributed over the concentration range of the finishing oil (0.22–0.62% by mass) and linear density range studied (1.0–5.4 dtex). The selected wavelength range, 2250–2370 nm, was the sole spectral region where the finishing oil absorbed significantly.The best model was selected in terms of performance and the optimum number of factors was determined by comparing models constructed from 1 to 14 factors. The best choice was that leading to the lowest relative error for the 94 samples in the test set (%RSETS). Provision was also made for loadings in order to assess visually the potential noise that PLS components introduced as they were included in the calibration model.The best model based on apparent absorbance spectra (log 1/R) was obtained by using nine PLS components. The first component accounted for 99.99% of the spectral variance; this was a result of the component accounting for spectral scattering associated with linear density variations among the samples in the calibration set, which was the main source of spectral variability. However, no noise in the loadings was observed up to the tenth component. The results thus obtained were acceptable as a whole.However, all black and deeply coloured samples were quantified with high absolute errors (above 0.06%), which suggests that colour contributes to the spectral signal. Reviewing the samples included in the calibration matrix revealed that none was black or deeply coloured. Exchanging some samples between the calibration and test sets in order to include these colour characteristics led to a new model with a minimum %RSETS for nine PLS components and a decreased %RSETS value owing to the correct quantification of black and deeply coloured samples.This suggests that the absorption of pigments and dyes over the working wavelength range entails including samples that encompass the whole colour and colour depth range in the calibration matrix. Table 1 gives the RSE values obtained for the calibration and test sets by using this model. While the predictive capacity of the model was acceptable, various mathematical treatments were tested in order to lessen the effect of the fibre linear density on the calibration matrix and thus construct a more simple and robust method.Of the different available treatments,23,24 both classical derivative methods and new approaches such as standard normal variate (SNV) and multiplicative scatter correction (MSC)25–27 were tested. The first PLS component in the model constructed from second-derivative spectra accounted for 99.8% of the overall variance of the spectrum matrix; therefore, the second derivative was also ineffective in correcting the effect of scattering arising from variations in fibre linear density.The optimum second-derivative method, constructed from 11 PLS components, provided a %RSETS value of 7.27%. The increased number of components can be ascribed to the fact that the effect of linear density was not corrected and that noise was augmented by derivation. MSC and SNV treatments are reportedly equivalent and linearly related.28 Because the SNV mode is a set-independent transformation, no new calculations need be made when a sample is removed from or included in the calibration matrix.This is not the case with MSC, a set-dependent transformation. The SNV treatment is, therefore, more computationally convenient. The results obtained with both types of treatment were identical, so only SNV was used in subsequent work. Fig. 3 shows the spectra of Fig. 2 following application of the SNV treatment, which efficiently corrected the baseline shift arising from the linear density differences between samples.Thus, the first PLS component in a model constructed from spectra processed with SNV accounted for only 40.6% of the variance of the spectral matrix, while the second and third components accounted for 35.1 and 20.7%, respectively. Because the effect othe baseline shift was minimized and the signal-to-noise ratio was not diminished, the optimum number of PLS components was reduced to seven, with no detriment to the predictive capacity.No loading noise was observed with this mathematical treatment up to the eighth PLS component. Therefore, visual inspection of the loadings confirmed that the optimum number of PLS components was seven. Table 1 shows the results obtained for the samples in the calibration and test sets by using the new mathematical treatments. Note that the RSE value for the samples that were analysed in duplicate with the reference method was 8.39%, similar to those provided by all the models employed.The SNV procedure was thus adopted as the optimum mathematical treatment as it reduced the number of PLS Fig. 2 NIR spectrum of acrylic fibres of 1.0 (1), 1.3 (2), 2.2 (3), 3.3 (4) and 5.4 (5) dtex. Table 1 Figures of merit for the different models constructed from the calibration and test sets Number of Treatment PLS components RSEC (%) RSETS (%) Absorbance 9 5.75 6.33 Second derivative 11 6.81 7.27 SNV 7 5.95 6.28 Fig. 3 Spectra of the acrylic fibres of Fig. 2 following application of the SNV treatment. Analyst, August 1997, Vol. 122 779components relative to the absorbance mode, with no loss of predictive capacity. No second-derivative spectra were used because they increased the number of PLS components required and did not lead to improved predictions. Validation of the Method From the results obtained it follows that the calibration model used possesses a good predictive capacity for the finishing oil concentration in samples manufactured over the same period as those employed for calibration.However, the performance of the method in the long run should be assessed in order to detect potential uncontrolled variations (e.g., in the copolymer or finishing oil composition, colour, fibre properties) during the manufacturing process that would be reflected in the spectra and might affect the results. Thus, an overall 340 validation samples were collected from various production lines after the calibration model had been developed. None had been used to optimize the model and all were collected by individuals who had taken no part in the study in order to ensure randomness in the selection process.The sample collection process was started after the calibration model had been established and spanned 6 months. A variable number of samples manufactured each month was collected. The reference values for the samples were obtained from a single measurement of each sample and their spectra were recorded by using the same methodology as for the calibration and test samples. The errors of prediction for the first three sample batches were very similar to those for the test set; however, the fourth and fifth batches exhibited large systematic errors that disappeared in the sixth batch.Fig. 4 shows the differences between the reference and NIR values for the different batches.In order to cancel this error, the joint use of SNV and secondderivative spectra (SNV–second derivative) was considered. While the systematic errors in the fourth and fifth batches were cancelled and the errors for the other samples were similar to the previous errors, this choice was discarded because it required the use of 13 PLS components. The origin of the problem was traced by obtaining the differential spectrum for two samples containing some and no finishing oil, respectively.The differential spectrum (Fig. 5) exhibited a band not corresponding to the oil that might be the source of the deviations. A new model was therefore constructed from a shorter wavelength range (2280–2370 nm), the optimum number of PLS components for which was eight. As can be seen from Table 2, the large RSEEV values for the fourth and fifth batches of external validation samples disappeared, because the systematic error was cancelled, and the other samples were quantified with errors similar to those provided by the SNV model and the previous wavelength range.Fig. 6 shows the differences between the reference values and the NIR results obtained with the new model using a shorter wavelength range for the whole external validation set. As can be seen, very few samples exhibited deviations beyond the accepted limits for control analyses. In order to assess the ability of the proposed method to produce good results, the results provided by the NIR and reference methods were compared on the basis of Student’s tvalues for the paired sample set.There were no significant differences between the two methods at a 95% confidence level. Fig. 4 Differences between reference values for the validation samples and those obtained with the proposed NIR method using the wavelength range 2250–2370 nm. Arrows indicate samples in the different batches. Fig. 5 (1) Spectrum of the finishing oil. (2) Differential spectrum of a sample containing the finishing oil and another containing none, following application of the SNV treatment.Table 2 RSEEV for the different batches of the validation set using different PLS models 2250–2370 nm 2280–2370 nm Number of SNV–second Batch samples SNV derivative SNV 1 119 6.52 7.49 7.46 2 93 8.33 9.90 9.27 3 66 8.70 9.74 7.82 4 10 17.56 11.02 7.75 5 22 14.58 6.54 8.15 6 30 7.74 6.78 8.87 Fig. 6 Differences between reference values for the validation samples and the NIR values provided by the proposed NIR method using the wavelength range 2280–2370 nm.Arrows indicate samples in the different batches. 780 Analyst, August 1997, Vol. 122Conclusions The joint use of the NIRRS technique and chemometric methods is a powerful means for the expeditious determination of finishing oils in acrylic fibres and hence a preferential choice for quality control analyses. While the mathematical treatments tested (SNV, second derivative, SNV–second derivative) improve very little on the results obtained from absorbance values, they do affect the optimum number of PLS components.On the other hand, reducing the working wavelength range to the absorption band of the finishing oil allows one to construct a robust model that ensures correct quantification of each batch of external validation samples, with no systematic errors. The prediction errors thus obtained are acceptable and comparable to those provided by the reference method.In this respect, the proposed NIRRS method is superior to more expensive and time-consuming techniques that use toxic, polluting reagents. The authors are grateful to the firm Courtaulds Espa�na, S.A., which provided pertinent technical information and analysed all the samples studied by using the reference method. This work was conducted in the framework of Project PB93-0899, funded by Spain’s Direcci�on General de Investigaci�on Cient�ýfica y T�ecnica (DGICyT).References 1 Galante, L. J., Brinkley, M. A., Drennen, J. K., and Lodder, R. A., Anal. Chem. 1990, 62, 2514. 2 Blanco, M., Coello, J., Iturriaga, H., Maspoch, S., de la Pezuela, C., and Russo, E., Anal. Chim. Acta, 1994, 298, 183. 3 Plugge, W., and Van der Vlies, C., J. Pharm. Biomed. Anal., 1993, 11, 435. 4 Barton, F. E., Burton, G. W., and Monson, W. G., J. Assoc. Off. Anal. Chem., 1990, 73, 312. 5 Aastveit, A. H., and Marum, P., Appl. Spectrosc., 1993, 47, 463. 6 Evans, D.G., Scotter, C. N. G., Day, L. Z., and Hall, M. N., J. Near Infrared Spectrosc., 1993, 1, 33. 7 Canivato, A. G., Mayes, D. M., Ge, Z., and Callis, J. B., Anal. Chem., 1990, 62, 1977. 8 Crawford, N. R., Hellmuth, W. W., Marcellus, D. H., and Chou, K. J., Process Control Quality, 1992, 4, 13. 9 Kelly, J. J., Barlow, C. H., Jinguji, T. M., and Callis, J. B., Anal. Chem., 1989, 61, 313. 10 Blanco, M., Coello, J., Iturriaga, H., Maspoch, S., and Bertran, E., Analyst, 1994, 119, 1779. 11 Honigs, D. E., Hieftje, G. M., and Hirschfield, T., Appl. Spectrosc., 1984, 38, 317. 12 Aasworth, C. M., Kirkbright, G. F., and Spillane, D. E. M., Analyst, 1983, 108, 1481. 13 Boguskavsky, A., Botha, A., and Hunter, L., NIR News, 1993, 4, 1. 14 Marbach, R., Koschinsky, Th., Gries, F. A., and Heise, H. M., Appl. Spectrosc., 1993, 47, 8lon, J., Yan, S. H., Tong, J., Neurens, M., and Hast, J., Appl. Spectrosc., 1994, 48, 190. 16 Davies, A. M. C., NIR News, 1993, 4 (4), 10. 17 Hildrum, K. I., Isaksson, T., Naes, T., and Tandberg, A., Near Infrared Spectroscopy, Ellis Horwood, Chichester, 1992. 18 Kirschner, C. H., Jordan, S. L., Taylor, L. T., and Seemuth, P. D., Anal. Chem., 1994, 66, 882. 19 Haaland, D. M., and Thomas, E. V., Anal. Chem., 1988, 60, 1193. 20 Geladi, P., and Kowalski, B. R., Anal. Chim. Acta, 1986, 185, 1. 21 Bull, C. R., Analyst, 1991, 116, 781. 22 Blanco, M., Coello, J., Iturriaga, H., Maspoch, S., and de la Pezuela, C., Appl. Spectrosc., 1997, 51, 240. 23 Aucott, L. S., Garthwaite, P. H., and Buckland, S. T., Analyst, 1988, 113, 1849. 24 Dhanoa, M. S., Lister, S. J., and Barnes, R. J., Appl. Spectrosc., 1995, 49, 765. 25 Barnes, R. J., Dhanoa, M. S., and Lister, S. J., Appl. Spectrosc., 1989, 43, 772. 26 Geladi, P., MacDougall, D., and Martens, H., Appl. Spectrosc., 1985, 39, 491. 27 Isaksson, T., and Naes, T., Appl. Spectrosc., 1988, 42, 1273. 28 Dhanoa, M. S., Lister, S. J., Sanderson, R., and Barnes, R.J., J. Near Infrared Spectrosc., 1994, 2, 43. Paper 7/00409E Received January 17, 1997 Accepted May 6, 1997 Analyst, August 1997, Vol. 122 781 Determination of Finishing Oils in Acrylic Fibres by Near-infrared Reflectance Spectrometry Marcelo Blanco*, Jordi Coello, Jos�e M. Garc�ýa Fraga†, Hortensia Iturriaga, Santiago Maspoch and Jordi Pag`es Departament de Qu�ýmica, Unitat de Qu�ýmica Anal�ýtica, Universitat Aut`onoma de Barcelona, E-08193 Bellaterra, Barcelona, Spain.E-mail: iqanl@cc.uab.es The potential of near-infrared diffuse reflectance spectrometry for quality control analyses in the textile industry was explored with a view to the quantification of finishing oils in acrylic fibres by partial least-squares regression, using a rotary cuvette system for recording spectra. Calibration was performed with a set of samples that encompassed every source of variability (linear density of the fibres, colour, concentration of the finishing oil), and the wavelength region where absorption was mostly due to the oil was used to construct several models from which that leading to the minimum relative standard error for a sample test set was selected.The results provided by various mathematical treatments [secondderivative, standard normal variate (SNV)] used to minimize scattering resulting from the differential linear density of the samples revealed no significant differences between prediction errors (only in the number of partial least-squares components).The model was used to quantify levels of finishing oil in routinely manufactured samples for a period of 6 months, during which time two batches showed poor predictions due to a new component appearing in the product. Modification of the calibration model to account for this component substantially increased robustness and allowed the accurate quantification of all batches manufactured after the model has been developed.Keywords: Near-infrared spectrometry; finishing oil; acrylic fibres; standard normal variate; partial least-squares calibration The ease with which sample spectra can be recorded, the need for no sample treatment and the ability to determine both physical and chemical parameters have made near-infrared diffuse reflectance spectrometry (NIRRS) a highly appealing technique for industrial control analyses. The substantially increased productivity derived from its use and the resulting expeditiousness with which useful information can be acquired and educated decisions made as a result have fostered its application in domains that could hardly have envisaged its use only a few years ago.In fact, NIR spectrometry has proved a powerful analytical tool for analysing a wide variety of samples used by the pharmaceutical,1–3 agricultural,4,5 nutritional,6,7 petrochemical8,9 and textile industries,10–13 among others, and is gaining ground in the clinical14,15 and related domains.An NIR spectrum consists of the bands for overtones and combinations of tones of fundamental bands (mainly hydrogen bonds, which absorb in the mid-IR region). The spectrum contains not only chemical information on the sample constituents, but also physical information (e.g., crystal habit, particle size, degree of crystallinity) that influences the former. Correlating the spectrum of a sample with its composition is often a difficult task that entails using an appropriate chemometric approach in order to ensure correct quantification.When a chemical compound is to be quantified, the dependence of the signal on the physical variability of the sample must be corrected. There are a number of available mathematical treatments for correcting these effects; most rely on derivative spectra,16,17 which exhibit better band resolution and decreased signal-to-noise ratio, making them unsuitable for determining analytes present at low concentrations.Some textile fibres are subjected to finishing procedures in order to improve properties such as abrasion, wrinkling and fire resistance, appearance and feel. The treatment applied depends on the particular type of fibre and its intended use. During the manufacturing process, fibres are subjected to treatments that introduce substances intended to condition them for the subsequent processing steps. The products used for this purpose are known as ‘finishing oils’ and facilitate lubrication, reduce static electricity, improve filament cohesion, etc.The large physical and chemical differences between natural and synthetic fibres, and hence between their applications, have fostered the use of a variety of products for these purposes, either alone or in combination. Ensuring optimum quality in a product requires careful control of the amount of finishing oil that is deposited on the fibres so that these behave as expected in the textile process; in fact, using too much or too little oil can lead to various problems and considerably diminished productivity.A rapid, reliable determination of the finishing oil content in fibres is thus a suitable means for identifying the need to correct the operating conditions of the manufacturing process in order to ensure the expected product quality. Currently available methods for fibre analysis typically use chlorinated solvents to remove the finishing oil from the fibres and measure the resulting solution by IR spectrometry.This methodology is laborious and time consuming; also, chlorinated solvents are toxic and environmentally aggressive, so their use is gradually being discouraged (e.g., in the EPA regulations for organic solvents and solvent waste). Recently,18 supercritical CO2 was used as a replacement solvent in the extraction and determination of finishing oils in order to avoid these shortcomings; however, the ensuing method is slow and entails optimizing the extraction conditions for each type of finishing.Ideally, an analytical method should use no organic solvents; also, it should be expeditious and provide reliable, reproducible results for the intended purpose. In this work, the joint use of NIRRS and multivariate calibration techniques [partial least-squares (PLS)] for determining finishing oils in acrylic fibres was explored. The effect of fibre properties on the quantification parameters was examined in order to develop a robust method fit for continuous implementation at a production line.Experimental Samples The samples used consisted of acrylic fibre filaments composed of acetonitrile–vinyl acetate (90 + 10 m/m) copolymer. The † Present address: Departamento de Qu�ýmica Anal�ýtica, Nutrici�on y Bromatolog�ýa, Universidad de La Laguna, E-38071 La Laguna, Tenerife, Spain. Analyst, August 1997, Vol. 122 (777–781) 777acrylic fibre studied is obtained by passing a concentrated solution of the copolymer through a spinning nozzle immersed in a cold water bath in order to produce filaments that are subjected to various stretching and drying operations, the diameter of the fibres obtained depending on the extent of stretching oash;the linear density of the fibres used ranged from 1.0 to 5.40 dtex (dtex = tenths of milligram per linear metre of fibre).Then, fibre bundles are immersed in a dyeing bath to obtain the desired colour.Samples vary widely in colour, brightness (matt, semi-matt, glossy) and colour depth. Finally, the fibres are immersed in a finishing oil bath (an aqueous dispersion of the product) and dried. The finishing oil contents in the samples studied ranged from 0.22 to 0.62% by mass, depending on linear density and intended use. Apparatus The experimental set-up used consisted of a NIRSystems 6500 NIR spectrometer equipped with a reflectance detector and a spinning module that afforded recording of spectra while the cuvette was spun and provided an average spectrum after a preset number of scans.In this way, the potential effect of sample heterogeneity was suppressed. A Nicolet FT-IR 50 spectrometer and cuvettes with CaF2 windows were used to record the spectra of the finishing oil, using the reference method. Software Two different types of software were used, namely: (i) NSAS (Near-Infrared Spectral Analysis Software) version 3.25, from NIRSystems, which allows acquisition of the sample spectrum and mathematical processing of spectra (averaging, obtainment of derivative spectra, etc.).(ii) Unscrambler version 5.03, from CAMO A/S, which was used to implement partial least-squares regression (PLSR), identify potential outliers and locate the spectral regions of analytical interest. It also allows whole spectra and selected spectral regions to be statistically processed. The spectra thus recorded were converted to the format used by Unscrambler using JCAMP.Determination of Finishing Oil in Acrylic Fibres Reference procedure A 1 g amount of dry fibre was extracted with 50 ml of CCl4 for 10 min under magnetic stirring. The extract was filtered, its IR spectrum recorded and the absorbance in the spectral region 2800–3000 cm21 measured. The absorbance values thus obtained were interpolated into a calibration graph obtained by applying the same procedure to a set of standard samples containing known amounts of finishing oil.The standard samples used to construct the calibration graph were analysed by using a gravimetric method involving the extraction of a larger amount of sample and evaporating the solvent to dryness. All samples were analysed in duplicate. Recording of NIR Spectra Each spectrum was the average of 32 scans over the wavelength range 1100–2500 nm at 2 nm intervals. Prior to each sample spectrum, a reference spectrum was obtained for a porcelain dish.The spectra subsequently processed were the average of three aliquots of sample. The spectrum of the finishing oil (Fig. 1) was obtained by extracting the aqueous dispersion, used to soak the fibres, into CCl4 and concentrating the organic phase. Data Processing The calibration model used to determine the finishing oil was obtained by using PLSR methodology.19,20 The calibration model was constructed by optimizing the spectral mode, working wavelength range and number of principal components used in the regression in order to minimize the error of prediction for the samples.As can be seen from Fig. 1, the finishing oil only exhibited absorption in the spectral regions 1700–1800 and 2250–2370 nm. The bands in the former region were rather weak and masked by the signal of the acrylic fibre copolymer, so only the latter region was used to construct the calibration model. In order to compare the predictive capacity of the different models tested, the relative standard error for the calibration samples (%RSEC), test samples (%RSET) and external validation samples (%RSEEV) was calculated from % ( ) RSE NIR REF REF = - � = = å å C C C i i i n i n 2 1 2 1 100 (1) where n is the number of samples included in the calibration matrix, the test set or the external validation set, and CREF and CNIR are the sample concentrations provided by the reference and NIR methods, respectively.As noted in the Introduction, diffuse reflectance measurements are influenced by scattered light, the amount of which depends on particle size, crystal habit, etc.21,22 Fig. 2 shows the NIR spectra of five samples containing the same concentration of finishing oil but possessing a different linear density. As can be seen, the effect of the linear density of the fibres was comparable to the baseline shift caused by differences in particle size. In order to minimize this effect, various mathematical treatments based on suppression of the baseline shift were tested.Results and Discussion In order to ensure that the calibration model used would provide accurate predictions, the sample set was split into two sub-sets Fig. 1 NIR spectrum of the finishing oil in CCl4. 778 Analyst, August 1997, Vol. 122consisting of 83 samples for calibration (calibration set) and another 94 for determining the predictive capacity of the model (test set). The best choice was selected from among a large number of PLS calibration models.Finding such a model entailed considering the following parameters: finishing oil concentration, fibre linear density, working wavelength range, spectral mode and number of PLS factors. The samples initially included in the calibration matrix were chosen in such a way as to encompass every possible source of variability in the spectra in order to make it representative of the whole sample population. Thus, the calibration samples were evenly distributed over the concentration range of the finishing oil (0.22–0.62% by mass) and linear density range studied (1.0–5.4 dtex). The selected wavelength range, 2250–2370 nm, was the sole spectral region where the finishing oil absorbed significantly.The best model was selected in terms of performance and the optimum number of factors was determined by comparing models constructed from 1 to 14 factors. The best choice was that leading to the lowest relative error for the 94 samples in the test set (%RSETS).Provision was also made for loadings in order to assess visually the potential noise that PLS components introduced as they were included in the calibration model. The best model based on apparent absorbance spectra (log 1/R) was obtained by using nine PLS components. The first component accounted for 99.99% of the spectral variance; this was a result of the component accounting for spectral scattering associated with linear density variations among the samples in the calibration set, which was the main source of spectral variability.However, no noise in the loadings was observed up to the tenth component. The results thus obtained were acceptable as a whole. However, all black and deeply coloured samples were quantified with high absolute errors (above 0.06%), which suggests that colour contributes to the spectral signal. Reviewing the samples included in the calibration matrix revealed that none was black or deeply coloured. Exchanging some samples between the calibration and test sets in order to include these colour characteristics led to a new model with a minimum %RSETS for nine PLS components and a decreased %RSETS value owing to the correct quantification of black and deeply coloured samples.This suggests that the absorption of pigments and dyes over the working wavelength range entails including samples that encompass the whole colour and colour depth range in the calibration matrix.Table 1 gives the RSE values obtained for the calibration and test sets by using this model. While the predictive capacity of the model was acceptable, various mathematical treatments were tested in order to lessen the effect of the fibre linear density on the calibration matrix and thus construct a more simple and robust method. Of the different available treatments,23,24 both classical derivative methods and new approaches such as standard normal variate (SNV) and multiplicative scatter correction (MSC)25–27 were tested.The first PLS component in the model constructed from second-derivative spectra accounted for 99.8% of the overall variance of the spectrum matrix; therefore, the second derivative was also ineffective in corrting the effect of scattering arising from variations in fibre linear density. The optimum second-derivative method, constructed from 11 PLS components, provided a %RSETS value of 7.27%. The increased number of components can be ascribed to the fact that the effect of linear density was not corrected and that noise was augmented by derivation. MSC and SNV treatments are reportedly equivalent and linearly related.28 Because the SNV mode is a set-independent transformation, no new calculations need be made when a sample is removed from or included in the calibration matrix. This is not the case with MSC, a set-dependent transformation.The SNV treatment is, therefore, more computationally convenient.The results obtained with both types of treatment were identical, so only SNV was used in subsequent work. Fig. 3 shows the spectra of Fig. 2 following application of the SNV treatment, which efficiently corrected the baseline shift arising from the linear density differences between samples. Thus, the first PLS component in a model constructed from spectra processed with SNV accounted for only 40.6% of the variance of the spectral matrix, while the second and third components accounted for 35.1 and 20.7%, respectively.Because the effect of the baseline shift was minimized and the signal-to-noise ratio was not diminished, the optimum number of PLS components was reduced to seven, with no detriment to the predictive capacity. No loading noise was observed with this mathematical treatment up to the eighth PLS component. Therefore, visual inspection of the loadings confirmed that the optimum number of PLS components was seven. Table 1 shows the results obtained for the samples in the calibration and test sets by using the new mathematical treatments.Note that the RSE value for the samples that were analysed in duplicate with the reference method was 8.39%, similar to those provided by all the models employed. The SNV procedure was thus adopted as the optimum mathematical treatment as it reduced the number of PLS Fig. 2 NIR spectrum of acrylic fibres of 1.0 (1), 1.3 (2), 2.2 (3), 3.3 (4) and 5.4 (5) dtex. Table 1 Figures of merit for the different models constructed from the calibration and test sets Number of Treatment PLS components RSEC (%) RSETS (%) Absorbance 9 5.75 6.33 Second derivative 11 6.81 7.27 SNV 7 5.95 6.28 Fig. 3 Spectra of the acrylic fibres of Fig. 2 following application of the SNV treatment. Analyst, August 1997, Vol. 122 779components relative to the absorbance mode, with no loss of predictive capacity. No second-derivative spectra were used because they increased the number of PLS components required and did not lead to improved predictions.Validation of the Method From the results obtained it follows that the calibration model used possesses a good predictive capacity for the finishing oil concentration in samples manufactured over the same period as those employed for calibration. However, the performance of the method in the long run should be assessed in order to detect potential uncontrolled variations (e.g., in the copolymer or finishing oil composition, colour, fibre properties) during the manufacturing process that would be reflected in the spectra and might affect the results.Thus, an overall 340 validation samples were collected from various production lines after the calibration model had been developed. None had been used to optimize the model and all were collected by individuals who had taken no part in the study in order to ensure randomness in the selection process. The sample collection process was started after the calibration model had been established and spanned 6 months.A variable number of samples manufactured each month was collected. The reference values for the samples were obtained from a single measurement of each sample and their spectra were recorded by using the same methodology as for the calibration and test samples. The errors of prediction for the first three sample batches were very similar to those for the test set; however, the fourth and fifth batches exhibited large systematic errors that disappeared in the sixth batch.Fig. 4 shows the differences between the reference and NIR values for the different batches. In order to cancel this error, the joint use of SNV and secondderivative spectra (SNV–second derivative) was considered. While the systematic errors in the fourth and fifth batches were cancelled and the errors for the other samples were similar to the previous errors, this choice was discarded because it required the use of 13 PLS components.The origin of the problem was traced by obtaining the differential spectrum for two samples containing some and no finishing oil, respectively. The differential spectrum (Fig. 5) exhibited a band not corresponding to the oil that might be the source of the deviations. A new model was therefore constructed from a shorter wavelength range (2280–2370 nm), the optimum number of PLS components for which was eight. As can be seen from Table 2, the large RSEEV values for the fourth and fifth batches of external validation samples disappeared, because the systematic error was cancelled, and the other samples were quantified with errors similar to those provided by the SNV model and the previous wavelength range.Fig. 6 shows the differences between the reference values and the NIR results obtained with the new model using a shorter wavelength range for the whole external validation set. As can be seen, very few samples exhibited deviations beyond the accepted limits for control analyses.In order to assess the ability of the proposed method to produce good results, the results provided by the NIR and reference methods were compared on the basis of Student’s tvalues for the paired sample set. There were no significant differences between the two methods at a 95% confidence level. Fig. 4 Differences between reference values for the validation samples and those obtained with the proposed NIR method using the wavelength range 2250–2370 nm.Arrows indicate samples in the different batches. Fig. 5 (1) Spectrum of the finishing oil. (2) Differential spectrum of a sample containing the finishing oil and another containing none, following application of the SNV treatment. Table 2 RSEEV for the different batches of the validation set using different PLS models 2250–2370 nm 2280–2370 nm Number of SNV–second Batch samples SNV derivative SNV 1 119 6.52 7.49 7.46 2 93 8.33 9.90 9.27 3 66 8.70 9.74 7.82 4 10 17.56 11.02 7.75 5 22 14.58 6.54 8.15 6 30 7.74 6.78 8.87 Fig. 6 Differences between reference values for the validation samples and the NIR values provided by the proposed NIR method using the wavelength range 2280–2370 nm.Arrows indicate samples in the different batches. 780 Analyst, August 1997, Vol. 122Conclusions The joint use of the NIRRS technique and chemometric methods is a powerful means for the expeditious determination of finishing oils in acrylic fibres and hence a preferential choice for quality control analyses. While the mathematical treatments tested (SNV, second derivative, SNV–second derivative) improve very little on the results obtained from absorbance values, they do affect the optimum number of PLS components.On the other hand, reducing the working wavelength range to the absorption band of the finishing oil allows one to construct a robust model that ensures correct quantification of each batch of external validation samples, with no systematic errors. The prediction errors thus obtained are acceptable and comparable to those provided by the reference method.In this respect, the proposed NIRRS method is superior to more expensive and time-consuming techniques that use toxic, polluting reagents. The authors are grateful to the firm Courtaulds Espa�na, S.A., which provided pertinent technical information and analysed all the samples studied by using the reference method.This work was conducted in the framework of Project PB93-0899, funded by Spain’s Direcci�on General de Investigaci�on Cient�ýfica y T�ecnica (DGICyT). References 1 Galante, L. J., Brinkley, M. A., Dren, and Lodder, R. A., Anal. Chem. 1990, 62, 2514. 2 Blanco, M., Coello, J., Iturriaga, H., Maspoch, S., de la Pezuela, C., and Russo, E., Anal. Chim. Acta, 1994, 298, 183. 3 Plugge, W., and Van der Vlies, C., J. Pharm. Biomed. Anal., 1993, 11, 435. 4 Barton, F. E., Burton, G. W., and Monson, W. G., J. Assoc. Off. Anal. Chem., 1990, 73, 312. 5 Aastveit, A. H., and Marum, P., Appl. Spectrosc., 1993, 47, 463. 6 Evans, D. G., Scotter, C. N. G., Day, L. Z., and Hall, M. N., J. Near Infrared Spectrosc., 1993, 1, 33. 7 Canivato, A. G., Mayes, D. M., Ge, Z., and Callis, J. B., Anal. Chem., 1990, 62, 1977. 8 Crawford, N. R., Hellmuth, W. W., Marcellus, D. H., and Chou, K. J., Process Control Quality, 1992, 4, 13. 9 Kelly, J. J., Barlow, C. H., Jinguji, T. M., and Callis, J. B., Anal. Chem., 1989, 61, 313. 10 Blanco, M., Coello, J., Iturriaga, H., Maspoch, S., and Bertran, E., Analyst, 1994, 119, 1779. 11 Honigs, D. E., Hieftje, G. M., and Hirschfield, T., Appl. Spectrosc., 1984, 38, 317. 12 Aasworth, C. M., Kirkbright, G. F., and Spillane, D. E. M., Analyst, 1983, 108, 1481. 13 Boguskavsky, A., Botha, A., and Hunter, L., NIR News, 1993, 4, 1. 14 Marbach, R., Koschinsky, Th., Gries, F. A., and Heise, H. M., Appl. Spectrosc., 1993, 47, 875. 15 Wallon, J., Yan, S. H., Tong, J., Neurens, M., and Hast, J., Appl. Spectrosc., 1994, 48, 190. 16 Davies, A. M. C., NIR News, 1993, 4 (4), 10. 17 Hildrum, K. I., Isaksson, T., Naes, T., and Tandberg, A., Near Infrared Spectroscopy, Ellis Horwood, Chichester, 1992. 18 Kirschner, C. H., Jordan, S. L., Taylor, L. T., and Seemuth, P. D., Anal. Chem., 1994, 66, 882. 19 Haaland, D. M., and Thomas, E. V., Anal. Chem., 1988, 60, 1193. 20 Geladi, P., and Kowalski, B. R., Anal. Chim. Acta, 1986, 185, 1. 21 Bull, C. R., Analyst, 1991, 116, 781. 22 Blanco, M., Coello, J., Iturriaga, H., Maspoch, S., and de la Pezuela, C., Appl. Spectrosc., 1997, 51, 240. 23 Aucott, L. S., Garthwaite, P. H., and Buckland, S. T., Analyst, 1988, 113, 1849. 24 Dhanoa, M. S., Lister, S. J., and Barnes, R. J., Appl. Spectrosc., 1995, 49, 765. 25 Barnes, R. J., Dhanoa, M. S., and Lister, S. J., Appl. Spectrosc., 1989, 43, 772. 26 Geladi, P., MacDougall, D., and Martens, H., Appl. Spectrosc., 1985, 39, 491. 27 Isaksson, T., and Naes, T., Appl. Spectrosc., 1988, 42, 1273. 28 Dhanoa, M. S., Lister, S. J., Sanderson, R., and Barnes, R. J., J. Near Infrared Spectrosc., 1994, 2, 43. Paper 7/00409E Received January 17, 1997 Accepted May 6, 1997 Analyst, August 1997, Vol. 122 781
ISSN:0003-2654
DOI:10.1039/a700409e
出版商:RSC
年代:1997
数据来源: RSC
|
|