摘要:

ANALYST, JANUARY 1991, VOL. 116 1 Editorial This has been a challenging year in the Analytical Journals office with various staff, procedural (for North American authors) and editorial changes being made, in order to keep The Analyst abreast of current developments in both the style and terminology used in analytical chemistry. Some of these changes may become evident as you read through this first issue of 1991; however, most are minor, hence there is no over-all change in format. These changes are detailed in the revised ‘Instructions to Authors’ given on page 105. One of the most obvious points of note is that The Analyst will no longer publish ‘short papers’, mainly because the criteria for these were exactly the same as those for full papers and the larger page size of recent years has meant that the distinction on the basis of length has become blurred.The main reason for the editorial and style changes is to achieve consistency between all primary journals of the RSC. A major step forward in 1990 was the establishment of a North American connection for The Analyst, with Professor J. F. Tyson being appointed US Associate Editor (see photo- graph). Hopefully his appointment will give The Analyst the foothold necessary to make further progress and keep a high profile in the North American market in order to encourage authors and to recruit referees. Thus, North American authors may now send their papers directly to Professor Tyson. Papers will usually be refereed within North America, thereby saving valuable time in the publication process.Professor Julian F. Tyson, US Associate Editor f o r The Analyst Professor Tyson and Professor J. D. R. Thomas (outgoing Chairman of the Analytical Editorial Board) both contributed Editorials for the July issue of The Analyst about the history and future of The Analyst, respectively. Professor Thomas is now the President of the Analytical Division and has been succeeded as Chairman by Dr. A. G. Fogg (Loughborough University), The Editorial staff would like to take this opportunity to express their appreciation for all the hard work Professor Thomas has put in over the last seven years and for the assistance and advice he has given throughout his term of office. We are certain he will continue to take an active interest in all the Analytical Journals and we wish him success in his new role.L to R: Harpal Minhas, Paul Delaney, Monique Warner, Claire Harris, Judith Egan (front), Brenda Holliday, Sheryl Whitewood, Paula O’Riordan and Roger Young There have also been several staff changes within the office (see photograph). Judith Egan remains Editorial Manager, Analytical, and Editor of the Journal of Analytical Atomic Spectrometry (JAAS) and Roger Young remains the Editor of Analytical Proceedings. However, Harpal Minhas (formerly Senior Assistant Editor) was appointed Editor of The Analyst in May. The former Editor, Janet Dean, has been promoted to the post of Editorial Manager, Dalton Transactions. Paul Delaney is now Senior Assistant Editor, along with Assistant Editors Paula O’Riordan, Sheryl Whitewood and Brenda Holliday. Claire Harris the Editorial Secretary for The Analyst and Analytical Proceedings and Monique Warner the Editorial Secretary for JAAS complete the Editorial team.All these changes inevitably disrupted production. However, everybody, including many of our referees, worked extremely hard and managed to keep The Analyst on schedule for most of the year. Our high standards were maintained with a rejection rate of 48% and average times to publication falling to their lowest since we relocated to Cambridge (in 1988/9) by September. Readers, authors and referees should note that there have also been several changes to the Advisory Board and Regional Advisory Editors, hence, you should consult the inside front cover of The Analyst to ensure that you are aware of your local Advisory Board membededitor. Of course, if you would like further information or to discuss any aspects of the publication procedures, please do not hesitate to contact either myself or Professor Tyson . Finally, The Royal Society of Chemistry celebrates its 150th Anniversary in 1991. To commemorate this event the Analy- tical Divison has organised two symposia (one jointly with the Faraday Division) to coincide with RSC celebrations during the Annual Chemical Congress, at Imperial College from 8th to 12th April, and we hope to see you there. Harpal Minhas, Editor

ISSN:0003-2654    

DOI:10.1039/AN9911600001

出版商:RSC    

年代:1991

数据来源: RSC

摘要:

ANALYST, JANUARY 1991, VOL. 116 3 Basic Statistical Methods for Analytical Chemistry Part 2. Calibration and Regression Methods* A Review James N. Miller Department of Chemistry, L ough borough University of Techno logy, L ough bo roug h, L eicestershire LE77 3TU, UK Summary of Contents Linear calibration Co r re I a t i o n coefficient ’Least squares’ line Errors and confidence limits Method of standard additions Limit of detection and sensitivity Intersection of two straight lines Residuals in regression analysis Regression techniques in the comparison of analytical methods Robust and non-parametric regression methods Analysis of variance in linear regression Weighted linear regression methods Partly straight, partly curved calibration plots Treatment of non-linear data by transformations Curvilinear regression Spline functions and other robust non-linear regression methods Keywords: Analytical calibration method; statistics and rectilinear graph; curve fitting method; robust and non -parametric method; re view Introduction Most methods in modern analytical science involve the use of optical, electrical, thermal or other instruments in addition to the manipulative ‘wet chemistry’ skills which are an essential part of the analyst’s training.Instrumental methods bring chemical benefits, such as the ability to study a wide range of concentrations, achieve very low limits of detection and perhaps study two or more analytes simultaneously. They also bring the practical benefits of lower unit costs and increased speed of analysis, perhaps through partial or complete automation.The results of instrumental analyses are evalu- ated by using calibration methods that bring about and reflect these advantages and are, to some extent, distinct from the statistical approaches discussed in Part 1 of this review.1 Nonetheless many of the concepts summarized in Part 1 are also applied in the statistics of calibration methods, and familiarity with these concepts is assumed here. A typical calibration experiment (a single analyte-multi- variate calibration has recently been surveyed2) is performed by making up a series of standard solutions containing known amounts of the analyte and taking each solution separately through an instrumental analysis procedure with a well defined protocol. For each solution, the instrument generates a signal, and these signals are plotted on the y-axis of a calibration graph, with the standard concentrations on the x-axis.A straight line or curve is drawn through the calibration points and may then be used for the determination of a test (‘unknown’) sample. The unknown is taken through exactly the same analysis protocol as the standards, the instrument signal is recorded and the test concentration estimated from the calibration graph by interpolation-and not , with one * For Part 1 of this series see reference 1. special exception described below, by extrapolation. It is apparent that one calibration graph can be used in the determination of many test samples, provided that instrument conditions and the experimental protocol do not change.This approach thus offers the desired feature of being able to analyse many samples rapidly over a range of concentrations; less obvious advantages include the ability to estimate limits of detection (see below) and eliminate the effects of some types of systematic error. For example, if the monochromator in a spectrophotometer has an error in its wavelength scale, errors in calculated concentrations using this instiument should cancel out between the standards and the samples. This approach to the determination of concentrations poses several problems. What type of line-straight, curved, or part-straight, part-curved-should be drawn through the calibration points? Given that the instrument signals obtained from the standards will be subject to random errors, what is the best straight line or curve through those points? What are the errors in test concentrations determined by interpolation? What is the limit of detection of the analysis? These and other statistical questions posed by calibration experiments still generate new methods and excite considerable controversy.Not surprisingly, it is in the area of curve-fitting that most new procedures are being introduced , but linear regression methods also generate their own original literature, as will become apparent. Linear Calibration Correlation Coefficient Many analytical procedures are carefully designed to give a linear calibration graph over the concentration range of interest, and analysts who use such methods routinely may assume linearity with only occasional checks.In the develop-4 ANALYST, JANUARY 1991, VOL. 116 ment of new methods, and in any other case where there is the least uncertainty, the assumption of linearity must be carefully investigated. It is always valuable to inspect the calibration graph visually on graph paper or on a computer monitor, as gentle curvature that might otherwise go unnoticed is often detected in this way (see below). Here, and in many other aspects of calibration statistics, the low-cost computer pro- grams available for most personal computers are very valu- able. As will be seen, it is important to plot the graph with the instrument response on the y-axis and the concentrations of the standards on the x-axis. One of the calibration points should normally be a ‘blank’, i.e., a sample containing all the reagents, solvents, etc., present in the other standards, but no analyte.It is poor practice to subtract the blank signal from those of the other standards before plotting the graph. The blank point is subject to errors as are all the other points and should be treated in the same way. As shown in Part 1 of this review,’ if two results, x1 and x2, have random errors el and e2, then the random error in x1 - x2 is not el - e2. Thus, subtraction of the blank seriously complicates the proper estimation of the random errors of the calibration graph. Moreover, even if the blank signal is subtracted from the other measurements, the resulting graph may not pass exactly through the origin. Linearity is often tested using the correlation coefficient, r .This quantity, whose full title is the ‘product-moment correla- tion coefficient’, is given by where the points on the graph are (xl, yl), ( x 2 , y 2 ) , . . (xi, yi), . . . (xn, yn), and X and J are, as usual, the mean values of xi and yi respectively. It may be shown that -1 d r d +l. In the hypothetical situation when r = - 1, all the points on the graph would lie on a perfect straight line of negative slope; if r = +1, all the points would lie exactly on a line of positive slope; and r = 0 indicates no linear correlation between x and y . Even rather ‘poor’ calibration graphs, i.e., with significant y-direction errors, will have r values close to 1 (or - l), values of Irl< about 0.98 being unusual. Worse, points that clearly lie on a gentle curve can easily give high values of IT-(.So the magnitude of r , considered alone, is a poor guide to linearity. A study of the ‘y-residuals’ (see below) is a simple and instructive test of whether a linear plot is appropriate. A recent report of the Analytical Methods Committee3 provides a useful critique of the uses of r , and suggests an alternative method of testing linearity, based on the weighted least squares method (see below). ‘Least Squares’ Line If a linear plot is valid, the analyst must plot the ‘best’ straight line through the points generated by the standard solutions. The common approach to this problem (not necessarily the best!) is to use the unweighted linear least squares method, which utilizes three assumptions. These are (i) that all the errors occur in the y-direction, i.e., that errors in making up the standards are negligible compared with the errors in measuring instrument signals, (ii) that the y-direction errors are normally distributed, and (iii) that the variation in the y-direction errors is the same at all values of x.Assumption (ii) is probably justified in most experiments (although robust and non-parametric calibration methods which minimize its sig- nificance are available, see below), but the other two assumptions merit closer examination. The assumption that errors only occur in the y-direction is effectively valid in many experiments; errors in instrument signals are often at least 2-3% [relative standard deviation (RSD)], whereas the errors in making up the standards should be not more than one-tenth of this.However, modern automatic techniques are dramatically improving the precision of many instrumental methods; flow injection analysis, for example, shows many examples of RSDs of 0.5% or less.4 In such cases, it may be necessary either to abandon assumption (i) (again, suitable statistical methods are available-see below), or to maintain the validity of the assumption by making up the standards gravimetrically rather than volu- metrically, i.e., with an even greater accuracy than usual. If the assumption is valid, the line calculated as shown below, is called the line of regression of y on x, and has the general formula y = bx + a , where b and a are, respectively, its slope and intercept. This line is calculated by minimizing the sums of the squares of the distances between the standard points and the line in the y-direction.(Hence the term ‘least squares’ for this method.) It is important to note that the line of regression of x on y would seek to minimize the squares of x-direction errors, and therefore would be entirely inappropriate when the signal is plotted on the y-axis. (The two lines are not the same except in the hypothetical situation when all the points lie exactly on a straight line.) The y-direction distances between each calibration point and the point on the calculated line at the same value of x are known as the y-residuals and are of great importance in several calculations, as will be shown later in this paper. Assumption (iii), that the y-direction errors are equal, is also open to comment.In statistical terms it means that all the points on the graph are of equal weight, i. e., equal importance in the calculation of the best line-hence the term ‘un- weighted’ least squares. In recent years this assumption has been tested for several different types of instrumental analysis, and in many cases it is found that the y-direction errors tend to increase as x increases, though not necessarily in linear proportion. Such findings should encourage the use of weighted least squares methods, in which greater weight is given to those points with the smallest experimental errors. These points are discussed further in a later section. If assumptions (i)-(iii) are accepted then the slope, b, and intercept, a , of the unweighted least squares line are found from a = J - b x (3) The equations show that, when b has been determined, a can be calculated by using the fact that the fitted line passes through the centroid, (X, J ) .These results are proved in reference 5 , a classic text on the mathematics of regression methods. The values of a and 6 can be simply applied to the determination of the concentration of a test sample from the corresponding instrument output. Errors and Confidence Limits The concentration value for a test sample calculated by interpolation from the least squares line is of little value unless it is accompanied by an estimate of its random variation. To understand how such error estimates are made, it is first important to appreciate that analytical scientists use the line of regression of y on x in an unusual and complex way.This is best appreciated by considering a conventional application of the line in a non-chemical field. Suppose that the weights of a series of infants are plotted against their ages. In this case the weights would be subject to measurement errors and to inter-individual variations (e.g., all 3 month old infants would not weigh the same), so would be correctly plotted on the y-axis: the infants’ ages, which would presumably be known exactly, would be plotted on the x-axis. The resulting plot would be used to predict the average weight (y) of a child of given age (x). That is, the graph would be used to estimate a y-value from an input x-value. The y-value obtained would of course be subject to error, because the least squares line itselfANALYST, JANUARY 1991, VOL.116 is subject to uncertainty. The graph would not normally be used to estimate the age of a child from its weight! In analytical work, however, the calibration graph is used in the inverse way-an experimental value of y ('yo, the instru- ment signal for a test sample) is input, and the corresponding value of x (xo, the concentration of the test sample) is determined by interpolation. The important difference is that xo is subject to error for two reasons, (1) the errors in the calibration line, as in the weight versus age example, and (2) the random error in the input yo value. Error calculations involving this 'inverse regression' methods are thus far from simple and indeed involve approximations (see below).First, we must estimate the random errors of the slope and intercept of the regression line itself. These involve the preliminary calculation of the important statistic sY/,, which is given by (4) In this equation, each yi value is a measured signal value from the analytical instrument, while the corresponding ji is the value of y on the fitted straight line at the same value of x. Each (yi - j$) value is thus a y-residual (see above). It is clear that equation (4) is similar to the equation for the standard deviation of a series of replicate results, except that the term ( n - 2) appears in the denominator as the number of degrees of freedom of the data, rather than n - 1. This difference is explained below in the discussion of analysis of variance applied to regression calculations.After syjx has been deter- mined, the standard deviation of the slope, sb, and the standard deviation of the intercept, s, can be determined from These standard deviations can then be used to estimate the confidence limits for the true slope and intercept values. The confidence limits for the slope are given by b + tsb, where the value of t is chosen at the desired confidence level (two-tailed values) and with n - 2 degrees of freedom. Similarly, the confidence limits for the intercept are given by a k ts,. These confidence limits are often of practical value in determining whether the slope or intercept of a line differs significantly from a particular or predicted value. For example, to test whether the intercept of a line differs significantly from 0 at the 95% confidence level, we need only see whether or not the 95% confidence interval for a includes zero.The statistic syjx is also used to provide equations for the confidence interval of the mean value of yo at a particular xo value, and for the (wider) confidence interval for a new and single value of yo measured at x = xo. These equations are of limited value in analytical work, as already noted, and examples are given in standard texts.5-7 Estimating the confidence limits for the entire line is more complex, as a combined confidence region for a and b is required. This problem was apparently first addressed by Working and Hotelling8 in 1929, and there is a useful summary of their method and of related studies in the often-cited paper by H ~ n t e r .~ The general form of the confidence limits is shown in Fig. l(a), from which it is clear that the confidence limits are at their narrowest (best) in the region of (X, p), as the regression line must pass through this point. We can now reconsider the principal analytical problem, that of estimating the standard deviation, sxo, and the confidence interval of a single concentration value xo derived from an instrument signal yo. As shown diagrammatically in Fig. 1, the confidence interval for this xo value results from the uncertainty in the measurement of yo, combined with the confidence interval for the regression line at that yo value. The standard deviation sxo is given by (7) It can be shown5 that this equation is an approximation that is only valid when the function t2 has a value less than about 0.05.For g to have low values it is clearly necessary for b and ?(xi - X ) 2 to be relatively large and sY/, to be small. In an analytical experiment with reasonable precision and a good calibration plot these results are indeed obtained; for example the data given in reference 9 yield a g value of 0.002. In a typical analysis, the value of yo might be obtained as the mean of m observations of a test sample, rather than as a single observation. In which case, the (approximate) equation for sxo becomes I (9) After sxo has been calculated, the confidence limits for xo can be determined as xo _+ tsxo, with t again chosen at a desired confidence level and n - 2 degrees of freedom. Inspection of equations (7) and (9) provides important guidance on the performance of a calibration experiment, presuming that we wish to minimize sxo.In cases where rn = 1, the first of the three terms within the bracket in these equations is generally the largest. Thus, making only a small number of replicate determinations of y o can dramatically improve the precision of xo. Similarly, increasing the number of calibration points, n , is beneficial. If considerations of time, material availability, etc. limit the total number of experiments (rn + n) that can be XO X Fig. 1 Confidence limits in linear regression: (a) shows the hyperbolic form of the confidence limits for a predicted y-value; and (b) shows how these confidence limits combine with the uncertainty in yo to yield a confidence interval for a predicted x-value, xo6 ANALYST, JANUARY 1991, VOL. 116 performed, the sum of the first two components of the bracketed term in (7) and (9) is minimized by setting rn = n.However, small values of n are to be avoided for a separate reason, viz., that the use of n - 2 degrees of freedom then leads to very large values of t and correspondingly wide confidence intervals. Calculation shows that, in the simple case where yo = y , then for any given values of sylx and 6 , the priority (at the 95% confidence level) is to avoid values of n < 5 because of the high values o f t associated with <3 degrees of freedom. When n 3 5 , maximum precision from a fixed number of measurements is obtained when rn = n. The last bracketed term in equations (7) and (9) shows that precision (for fixed rn and n) is maximized when yo is as close as possible to 7 (this is expected in view of the confidence interval variation shown in Fig.l), and when 7 (xi - X ) 2 is as large as possible. The latter finding suggests ;hat calibration graphs might best be plotted with a cluster of points near the origin, and another cluster at the upper limit of the linear range of interest [Fig. 2(a)]. If n calibration points are determined in two clusters of n/2 points at the extremes of a straight line, the value of the term 7 (xi - X)’ is increased by a factor [3(n - l)/(n + l)] compared Lith the case in which then points are equally spaced along the same line [Fig. 2(b)]. In practice it is usual to use a calibration graph with points roughly equally distributed over the concentration range of interest.The use of two clusters of points gives no assurance of the linearity of the plot between the two extreme x values; moreover, the term [ ( y o - 7)2/b2 ?(xi - i ) 2 ] is often the smallest of the three bracketed terms in equations (7) and (9), so reducing its value further may have only a marginal over-all effect on the precision of no. I Method of Standard Additions In several analytical methods (e.g., potentiometry, atomic and molecular spectroscopy) matrix effects on the measured signal demand the use of the method of standard additions. Known amounts of analyte are added (with allowance for any dilution effects) to aliquots of the test sample itself, and the calibration graph (Fig. 3) shows the variation of the measured signal with the amount of analyte added.In this way some matrix effects are equalized between the sample and the standards. The concentration of the test sample, x,, is given by the intercept on the x-axis, which is clearly the ratio of the X y-axis intercept and the slope of the calibration line, calculated using equations (2) and ( 3 ) , i.e. , x, = alb (10) The standard deviation of x,, sxe, is given by a modified form of equation (7): This standard deviation can as always be converted into a confidence interval using the appropriate t value. It might be expected that such confidence intervals would be wider for this extrapolation method than for a conventional interpolation method. In reality, however, this is not so, as the uncertainty in the value of x, derives only from the random errors of the regression line itself, the corresponding value of y being fixed at zero in this case.The real disadvantages of the method of standard additions are that each calibration line is valid for only a single test sample, larger amounts of the test sample may be needed and automation is difficult. The slope of a standard additions plot is normally different from that of the conventional calibration plot for the same sample. The slope ratio is a measure of the proportional systematic error produced by the matrix effect, a principle used in many ‘recovery’ experiments.’” The use of the conventional standard additions method has been discussed at length by Cardone. 11,12 The generalized standard additions method (GSAM)13 is applicable to multicomponent analysis problems, but belongs to the realm of chemometrics.*4 Limit of Detection and Sensitivity The ability to detect minute amounts of analyte is a feature of many instrumental techniques and is often the major reason for their use.Moreover, the concept of a limit of detection (LOD) seems obvious: it is the least amount of material the analyst can detect because it yields an instrument response significantly greater than a blank. Nonetheless, the definition and measurement of LODs has caused great controversy in recent years, with additional and considerable confusion over nomenclature, and there have been many publications by statutory bodies and official committees in efforts to clarify the situation. Ironically, the significance of LODs, at least in the strict quantitative sense, is probably overestimated.There is clearly a need for a means of expressing that (for example) spectrofluorimetry at its best is capable of determining lower amounts of analytes than absorptiometry, and the principal use of LODs in the literature appears to be to show that a newly discovered method is indeed ‘better’ than its predeces- sors. But there are many reasons why the LOD of a particular method will be different in different laboratories, when Fig. 2 low concentrations; and (b) equally spaced standards Calibration graphs with: ( a ) clusters of standards at high and Fig. 3 point 0 is due to the original sample; for details see text Calibration graph for the method of standard additions.TheANALYST, JANUARY 1991, VOL. 116 7 applied to different samples or used by different workers. Not least among the problems is, as always, the occurrence of hidden and possibly large systematic errors, a point rightly emphasized by a recent report of the Analytical Methods Committee of the Analytical Division of the Royal Society of Chemistry.15 It can thus be very misleading to read too much into the absolute value of a detection limit. One principle on which all authorities are agreed is that the sensitivity of a method is not the same as its LOD. The sensitivity is simply the slope of the calibration plot. As calibration plots are often curved (see below) the concentra- tion range over which the sensitivity applies should be quoted. In practice the concept of sensitivity is of limited value in comparing methods, as it depends so much on experimental conditions (e.g., the sensitivity of a spectrophotometric determination can simply be increased by increasing the optical path length).Comparisons of closely related methods -for example, of spectrophotometric methods for iron(Ir1) using three organic chelating reagents with different molar absorptivities in 10 mm cuvettesl6-may be of value. The most common definitions of the LOD take the form in which the lowest detectable instrument signal, yL, is given by YL = yB + ksB (12) where yB and sB are, respectively, the blank signal and its standard deviation. Any sample yielding a signal greater than y~ is held to contain some analyte, while samples yielding signals <yL are reported to contain no detectable analyte. The constant, k , is at the discretion of the analyst, and it cannot be emphasized too strongly that there is no single, ‘correct’, definition of the LOD.It is thus essential that, whenever an LOD is quoted, its definition should also be given. After yL has been established, it can readily be converted into a mass or concentration LOD, cL, by using the equation CL = kSB/b (13) This equation shows the relationship between the LOD and sensitivity, the latter being given by the slope of the calibration graph, b, if the graph is linear throughout. Kaiserl’ suggested that k should have a value of 3 (although other workers have, at least until recently, used k = 2, k = 23’2, etc.). This recommendation has been reinforced by the International Union of Pure and Applied Chemistry (1UPAC)lR and others, and is now very common.It is important to clarify the significance of this definition. Fig. 4(a) illustrates the distribution of the random errors of yB, the standard deviation oB being estimated by sB, as always. The probability that a blank sample will yield a signal greater than yB + 3sg is given by the shaded area, readily shown, using tables of the standard normal distribution, to be 0.00135, i.e., 0.135%. This is the probability that a false positive result will occur, i.e., that analyte will be said to be present when it is in fact absent. This is analogous to a type I error in conventional statistical tests.’ However, the second type of error (type I1 error), that of obtaining a false negative result, i.e., deducing that analyte is absent when it is in fact present, can also occur.If numerous measurements are made on a solution that contains analyte at the LOD level, cL, the instrument responses will be normally distributed about yL, with the same standard deviation (estimated by sB) as the blank. (This assumption of equal standard deviations was noted earlier in this review and has thus far been used throughout.) Half the measurements on a sample with concentration cL will thus yield instrument responses below yL. If any sample, yielding a signal less than yL, is reported as containing no detectable analyte, the probability of a type I1 error is clearly 50% ; this would always be true, irrespective of the separation of yB and YL.Many workers, therefore, separately define a ‘limit of decision’, a point between yB and yL, to establish more sensible levels of type I and type I1 errors. This procedure is 0 Y Limit of Limit of decision detection Y Fig. 4 Limits of detection; for details see text again analogous to simple statistical tests; in this case the null hypothesis is that there is no analyte present, and the alternative hypothesis is that an analyte concentration CL is present. The critical value for testing these hypotheses is set by the establishment of the limit of decision. In most cases the assumption is made that type I and type I1 errors are equally to be minimized (although it is easy to imagine practical instances where this is not appropriate), so the limit of decision would be at yB + 1 .5 ~ ~ [Fig. 4(b)]. If analyte is reported present or absent at y-values, respectively, above and below this limit, there is a probability of 6.7% for each type of error. Many analysts feel this to be a reasonable criterion; it is not much different from the 95% confidence level routinely used in most statistical tests. Clearly, if the probability of both types of error is to be reduced to 0.135%, the limit of decision must be at yB + 3sB, and the LOD at yB + 6s~.16 Some workers further define a ‘limit of determination’, i. e., a concentration which can be determined with a particular RSD. Using the IUPAC definition of LOD, it is clear that the RSD at the limit of detection is 33.33%. A common definition of the limit of determination is yB + 10sB, indicating an RSD of 10%.It is to be noted that this result again assumes that the standard deviation sB applies to measurements at all levels of y. The effects on LODs of departures from a uniform standard deviation have been considered by Liteanu and Rica19 and by the Analytical Methods Committee.15 If the LOD definition of yB + 3sB is accepted, it remains to discuss the estimation of yB and sB themselves. The blank signal, yB might be obtained either as the average of several readings of a ‘field blank’ (i.e., a sample containing solvent, reagents and sample matrix but no analyte, and examined by the same protocol as all other samples), or by utilizing the intercept value, a, from the least squares calculation. If all the assumptions involved in the latter calculation are valid, these two methods should yield values for yB that do not differ significantly.Repeated measurements on a field blank will also provide the value of sB. Only if a field blank is unobtainable (a not uncommon situation) should the intercept, a, be used as the measure of Y B , in this situation sYfx will provide an estimate of sB.8 ANALYST, JANUARY 1991, VOL. 116 Intersection of Two Straight Lines Analytical scientists frequently use methods requiring the determination of the point of intersection of two straight lines. This approach is used in Job’s method20 and in other studies of molecular interactions such as drug-protein binding. The usual requirement is to determine the x-value (often a concentration ratio rather than a single concentration in this case) of the point of intersection.If the two lines, each determined by the methods described above, are given by y = blx + a1 and y = b2x + a2, the intersection point, xx, is easily shown to be given by x, = (a2 - M b l - b2) (14) The confidence interval for x, has been calculated in several ways (reviewed in reference 19), and continues to excite interest;21 it is clearly related to the hyperbolic curves representing the confidence intervals for each line (Fig. 5). For the line y = blx + al these curves are given by This equation yields the confidence limits for the true mean value of y at any given value of x . The t value is taken at the desired confidence level (usually 95%) and nl - 2 degrees of freedom. A similar equation applies to the line y = b2x + a2.One reasonable definition for the lower confidence limit for xx, ( x L ) , is the abscissa value of the point of intersection of the upper confidence limit of line 1 and the lower confidence limit of line 2 (Fig. 5). At this point which can be solved for xL. An analogous equation can be written for xu, which is similarly defined by the intersection of the lower confidence limit for line 1 and the upper confidence limit for line 2 As the confidence intervals for lines 1 and 2 may be of different width, and as the two lines may interesect at any angle, the confidence limits for x, may not be symmetrical about x, itself. It should also be noted that the confidence limits for x, derived from (for example) the 95% confidence limits for the two separate lines are not necessarily the 95% confidence limits for x,.As the estimation method used above assumes the worst case in combining the random errors of the two lines, the derived confidence limits are on the pessimistic (ie., realistic!) side. Finally it is important to note that the practical applications of this method utilize extrapolations of the two straight lines to the intersection point. These extrapolations are generally short, and care is usually taken to perform the experiments in conditions where the extrapola- tions are believed to be valid. However, if this belief is erroneous (e.g., in studies of drug-protein binding where there is more than one class of binding site instead of the single class often assumed), even the best statistical methods cannot produce chemically valid results. Residuals in Regression Statistics Previous sections of this review have shown that the un- weighted regression methods in common use in analytical chemistry are based on several assumptions which merit XL xx xu X Fig.5 Confidence limits for the point of intersection of two straight lines critical examination and that it is not a straightforward matter to decide whether a straight line or a curve should be drawn through a set of calibration points. Important additional information on both these topics can be obtained from the y-residuals, the 0, - 9 ) values which represent the differences between the experimental y-values and the fitted y-values. The residuals thus represent the random experimental errors in the measurements of y, if the statistical model used (the unweighted regression line of y on x ) is correct.Many statistical tests can be applied to these residuals (a comprehen- sive survey is given in reference 5) but for routine work it is often sufficient to plot the individual residuals against 9 or against x . Many regression programs for personal computers offer this facility and some provide additional refinements, e.g., the inclusion of lines showing the standard deviations of the residuals. It can be shown that, if the calibration line is calculated from the equation y = bx + a (but not if it is forced through the origin by using the form y = bx), the residuals always total zero, allowing for rounding errors. As already noted, the residuals are assumed to be normally distributed.Fig. 6(a) shows the form that the residuals should thus take if the unweighted regression line is a good model for the experimen- tal data. Fig. 6(b) and ( c ) indicates possible results if the unweighted regression line is inappropriate. If the residuals tend to become larger as y (or x ) increases, the use of a weighted regression line (see below) is indicated, and if the residuals tend to fall on a curve, the use of a curved calibration graph rather than a linear one is desirable. In the latter case the signs (+ or -) of the residuals, which should be in random order if an appropriate statistical model has been used, will tend to occur in sequence (‘runs’); in the example given, there is clearly a sequence of positive residuals, followed by a sequence of negative ones followed by a second positive sequence.The number of ‘runs’ (three in the example given) is thus significantly less than if the signs of the residuals had been + and - in random order. The Wald-Wolfowitz method tests for the significance of the number of runs in a set of data5.22 by comparing the observed number of runs with tabulated data,23 but it cannot be used if there are fewer than nine points in the calibration graph. Like the other residual diagnostic methods described here, the test is thus of restricted value in instrumental analysis, where the number of calibration points is frequently less than this. Tests on residuals are not, however, limited to linear regression plots: they can also be applied to non-linear plots, and indeed to any situation in which experimental data are fitted to a statistical model and some unexplained variations occur.Examination of the residuals may shed light on a further problem, that of outliers among the data. The first part of thisANALYST, JANUARY 1991, VOL. 116 9 3 Fig. 6 Residuals in regression; for details see text review1 emphasized the importance of examining possible outliers carefully before rejecting them, if only because an observation that appears to be an outlier if one statistical model (e.g., linear regression with normally distributed errors) is used, might not be an outlier if an alternative model (e.g., a weighted or a polynomial regression equation) is fitted. After the residuals of a calibration graph have been calculated, it is usually easy to identify any that are exception- ally large.Again, many personal computer programs ‘flag’ such data points automatically. Unfortunately it is not legitimate simply to examine the residuals by the @test’ or related outlier tests, as the residuals are not independent measurements (they must total zero). However, several methods have been developed for studying potential outliers in regression.5.7 These methods transform the residuals before examination, and will not be treated in detail here. Perhaps the best-known approach involves the estimation of ‘Cook’s Distance’24 for the suspect point. This distance is a measure of the influence of an observation, i.e., of how much the regression line would be altered by omission of the observa- tion from the usual calculations of a and b.A discussion of this method, along with a BASIC computer program which implements it, has recently been published.25 The problem of outliers can alternatively be by-passed by the use of the robust and non-parametric methods described below. Regression Techniques in the Comparison of Analytical Methods When a novel analytical method is developed there is a natural desire to validate it by comparing it with well established methods. This is normally achieved by applying both new and established methods to the analysis of the same group of test samples. As calibration methods are designed for use over wide concentration ranges, these samples ‘ will properly contain widely differing amounts of the analyte under study. The question then arises, how are the paired results (i.e., each sample examined by each of the two methods) evaluated for systematic errors? The paired t-test1 cannot be used, as it ascribes the same weight to any given difference between a pair of results, irrespective of the absolute value measured.The approach most commonly used is to plot the results of the two methods on the two axes of a regression graph; each point on the graph thus represents a single sample measured by the two techniques being compared. It is clear that, if both methods give identical results for all samples, the resulting graph will be a straight line of unit slope and zero intercept, with the correlation coefficient I = +l. Some departure from these idealized results is inevitable in practice, and the usual requirements if the new method is to be regarded as satisfactory are that I is close to +1, that the confidence interval for the intercept, a, includes zero, and that the confidence interval for the slope, b , includes 1.The new method is tested most rigorously if the comparison is made with a considerable number of samples covering in a roughly uniform way the concentration range of interest. There are several ways in which the plotted line can deviate from the ideal characteristics summarized above. Sometimes one method will give results which are higher or lower than those of the other method by a constant amount (i.e., b = 1, a > or <O). In other cases there is a constant relative difference between the two methods (a = 0, b > or < 1).These two types of error can occur simultaneously ( a > or < 0, b > or < l), and there are instances in which there is excellent agreement between the two methods over part of the range of interest, but disagreement at, e.g. , very high or very low concentrations. Finally, there are experiments where some of the points lie close to the ideal line (b = 1, a = 0), but another group of samples give widely divergent points; speciation problems are the most probable cause of this result. These possibilities have been summarized by Thompson ,26 whose paper also studies an important problem in the use of conventional regression lines in method comparisons. The line of regression of y on x assumes that the random errors of x are zero. This is clearly not the case when two experimental methods are being compared, so despite its almost universal use in this context the conventional regression line is not a proper statistical tool for such comparisons. (The line of regression of x on y would be equally unsuitable.) It has, however, been shown that by using Monte Carlo simulation methods,26 the consequences of this unsoundness are not serious provided that at least ten samples covering the concentration range of interest fairly uniformly are used in the comparison, and the results from the method with the smaller random errors are plotted on the x-axis.A rigorous solution of the method comparison problem would be a calculation of the best straight line through a series of points with both x and y values subject to random errors.Over a century ago, Adcock27 offered a solution which assumed that the x- and y- direction errors were equal. A complete solution, based on maximum likelihood methods, has been proposed by Ripley and Thompson .2* Their technique utilizes the statistical weight of each point (thus requiring more information-see below) and as it is a computer-based iterative approach, it may not command ready acceptance for routine use. Robust and Non-parametric Regression Methods Previous sections of this review have shown that the un- weighted least-squares regression line of y on x may not be appropriate if the y-direction errors are not normally distri- buted, or if both x- and y-direction errors occur (as in method comparisons). Moreover, the calculation and interpretation of this line are complicated by the presence of possible outliers.Statisticians use the term robustness to describe the property of insensitivity to departures from an assumed statistical model; a number of robust regression methods have been developed in recent years. One obvious approach is to use non-parametric methods, which do not assume any particular distribution for the population of which the experimental data are a sample. (Note that robust methods are not necessarily non-parametric, but non-parametric methods are generally robust. )10 ANALYST, JANUARY 1991, VOL. 116 Perhaps the best-known non-parametric regression method is that developed by Theil.29 He suggested that the slope, b, of the regression line should be estimated by determining the median of the slopes of all the lines joining pairs of points.(The median of a set of measurements is the middle value when an odd number of measurements is arranged in numerical order, or the average of the two middle values when there is an even number of measurements.) A graph with n points will thus have [n(n + 1)/2] independent estimates of the slope. After b has been determined, n estimates of the intercept, a , can be obtained from the equation ai = yi - hi. The median of these n values of a is taken as the value of the intercept estimate. Theil’s procedure is open to two objec- tions. Firstly, it seems that the slope estimates of points with well separated xi values should carry more weight than those from neighbouring pairs of points; JacckeP has proposed a modified method that achieves this.Secondly, computation of the median slope value becomes tedious even for fairly moderate values of n; a graph with ten points will yield 55 separate estimates of the slope to be determined and sorted into numerical order. This problem is overcome by the use of a shorter technique (Theil’s abbreviated or incomplete method) in which slope estimates are obtained from xi and the first point above the median value of x, from x2 and the second point above the median, and so on. (If n is odd, the middle xi value is not used at all in the slope calculation.) After the slope of the line has been estimated in this way, the intercept is estimated as in the ‘complete’ method. An example of this approach, which also illustrates its robustness towards out- liers, is given in reference 22.The Theil methods make no assumptions about the directions of the errors, so are suitable for method comparisons (see above). Hussain and Sprent31 have shown that the Theil (complete) method is almost as efficient as the least-squares method when the errors are normally distributed, and much more efficient, especially when n is small, when the errors are not normally distributed. (Efficiency in statistics is a relative concept: it allows the comparison of two statistical tests in their ability to detect alternative hypotheses which are close to the null hypothesis, H,.3*) Maritz33 has reviewed Theil’s method and Sprent32 and Maritz34 have surveyed other robust and non-parametric regression methods, including those which handle curved graphs.It should be noted that the spline techniques of curve-fitting discussed below are non-parametric methods. When a line is to be drawn through a large number of points (e.g., in method comparisons), a relatively rapid and prelimi- nary method of plotting it may be of value. The points are divided as nearly as possible into three equal groups of x-values, and the median x-value for each of the three groups is identified. These three data points are known as the summary points. The slope of the resistant line (i.e., resistant to outliers) through all the points is then estimated by the slope of the line joining the two outermost summary points. The intercept of the line is calculated as the average of the three intercepts obtained from the determined slope and the three summary points (cf.Theil’s method above). The values of the slope and intercept may be polished by iterative minimization of the y-residuals. This method (and many other ‘quick and dirty’ methods used in exploratory data analysis) is discussed in reference 35. In some experiments the results cannot be expressed in quantitative terms, but only in terms of a rank order.’ Examples-perhaps infrequent in analytical work-include the preferences of laboratory workers for different pieces of equipment, the state of health of laboratory animals or the taste quality of a food or drink sample. Relationships in such cases are studied using rank correlation methods. Spearman’s rank correlation coefficient, p,36 is famous as being the first statistical method to use ranks, and is readily shown33 to be the product-moment correlation coefficient, Y, converted for use when both x and y variables are expressed as ranks.Spearman’s p is given by 6Cdi2 p = l - - n(n2 - 1) and, like Y, lies in the range - 1 d p d + 1. In equation (18) di is the difference between the x and y rankings for the ith measurement. If the calculated value of p exceeds the critical value (taken from tables) at the appropriate confidence level and value of n, a significant correlation between x and y is established (although not necessarily a causal relationship). As in other ranking methods, tied ranks (i. e., observations of equal rank in x or y) are given mid-rank values. Thus, if several food samples were ranked, with the two best samples judged equal in quality, the ranking would be 1.5,1.5,3,4, .. . instead of 1, 2, 3, 4, . . . The Kendall rank correlation coefficient, ~ , 3 7 is based on a different idea. If, in a ranking experiment, high x-values are generally associated with high y-values, we expect that yj > yi if xj > xi. Pairs of observations having this property are said to be concordant, while observations where the x and y values are in opposite order are said to be discordant (if xi = xj or yi = yj a tie has occurred). Kendall’s method involves examining each of the [n(n - 1)/2] pairs of data and evaluating n,, the number of concordances, and nd, the number of discordances. The rank correlation coefficient is then given by (19) nc - nd n(n - 1)/2 t = Again, it is evident that -1 d t d +1, the value -1 corresponding to all the pairs of data points giving discor- dances, and + 1 corresponding to complete concordance.Intermediate values are again compared with tabulated values. Kendall’s method has the advantage that the data do not have to be converted into ranks for n, and nd to be calculated. Moreover, the computation can be further simpli- fied38 if the term [n(n - 1)/2] is omitted, the test statistic being simply calculated as T = n, - nd. It is rare, however, for the results of the Kendall and Spearman methods to disagree, and both have been used successfully as tests for trend, i.e., to examine whether there is a correlation when one of the variables is time. The concept of concordance introduced by Kendall can be extended to problems with more than two variables; examples are given in standard texts.38 Analysis of Variance in Linear Regression Analysis of variance (ANOVA) is a powerful and very general method which separates the contributions to the over-all variation in a set of experimental data and tests their significance.The sources of variation (one of which is invariably the random measurement error) are each charac- terized by a sum of squares (SS), i.e., the sum of a number of squared terms representing the variation in question, a number of degrees of freedom (DF) (as defined in reference l), and a mean square, which is the former divided by the latter and can be used to test the significance of the variation contribution by means of the F-test.’ The mean square and the number of degrees of freedom for the over-all variation are, respectively, the sums of the mean squares and degrees of freedom of the several contributing sources of variation: this additive property greatly simplifies the calculations, which are now widely available on personal computer software.In analytical calibration experiments, variation in the y-direction only is considered. This variation is expressed as the sum of the squares of the distances of each calibration point from the mean y value, j , i.e., by C(yi - j ) 2 . This is the total SS about J . There are two contributions to this over-all variation. One is the SS due to regression, i.e., that part of the variation due to the relationship between y and x; each term in this SS is clearly of the form (ji - j)2.This SS has just one DF,ANALYST, JANUARY 1991, VOL. 116 11 as just one function of the yi values, i.e., the slope, 6 , will calculate C(j+ - j ) 2 from b2 Z(xi - X ) 2 . The second source of variation is the SS about regression, i. e., the variation due to deviations from the calibration line, each term in the SS being of the form (yi - j i ) 2 . This SS has (n - 2) DF, reflecting the fact that the residuals come from a model requiring the estimation of two parameters, a and b. In accordance with the additivity principle described above, it is possible to show5 that Total SS about J = SS due to regression + SS about Moreover, the number of DF for the total SS is ( n - 2) + 1 = ( n - 1). This result is expected, as only (n - 1) yi values are needed to determine the total SS about y , as Z(yi - 7 ) = 0 by definition.A typical ANOVA table for a linear regression plot is shown in Table 1. This is a one-way ANOVA calculation, there being only one source of variation in addition to the inevitable experimental error. The significance of the correla- tion can be tested by using the F-test, i.e., by calculating regression (20) F1, ( n - 2 ) = MSreg/MSres (21) In practice this is rarely necessary (though readily available in software packages), as the F-values are generally vastly greater than the critical values. A more common estimate of the goodness of fit is given by the statistic R2, sometimes known as the (multiple) coefficient of determination or the (multiple) correlation coefficient.The prefix ‘multiple’ occurs because R2 can also be used in curvilinear regression (see below). If the regression line (straight or curved) is to be a good fit to the experimental points, the SS due to regression should be a high proportion of the total SS about J . This is expressed quantitatively using the equation R2 = SS due to regressiordtotal SS about J (22) R2 clearly lies between 0 and 1 (although it can never reach 1 if there are multiple determinations of yi at given xi valuess), however, it is often alternatively expressed as a percentage- the percentage of goodness of fit provided by a regression equation. It can be shown5 that, for a straight line plot, R2 = r2, the square of the product-moment correlation coefficient. The application of R2 to non-linear regression methods is considered further below.Weighted Linear Regression Methods The preceding discussion of regression methods has assumed that all the points on the regression line have equal weight (i.e., equal importance) when the regression line is plotted. This is a reflection of the assumption that the y-direction errors are equal at all the values of x used in the calibration graph. In practice this is often an unrealistic assumption, as it is very common for the standard deviations of the measure- ments to alter with x. As already noted, RSDs will be high at analyte levels just above the LOD. However, there may be other variations at much higher concentrations. In some cases the standard deviation is expected to rise in proportion to the concentration, i.e., the RSD is approximately constant,39 while in other cases the standard deviation rises, though less rapidly than the concentration.40 Many attempts have been made to formulate rules and equations for this concentration related behaviour of the standard deviation for different methods.41.42 In practice, however, it will frequently be better to rely on the analyst’s experience of a particular method, instrument, etc.in this respect. If experience suggests that the standard deviation of replicate measurements does indeed vary significantly with x (heteroscedastic data) , a weighted regression line should be plotted. The equations for this line differ from equations (2)- (7) because a weighting factor, wi, must be associated with each calibration point xi, yi. This factor is inversely propor- tional to the variance of yi, si2, and must either be estimated from a suitable model (see above), or determined directly from replicate measurements of yi: wi = si-2/(Csi-2/n) I (23) Equation (23) conveniently scales the weighting factors so that their sum is equal to n, the number of xi values.The slope and intercept of the weighted regression line are then given, respectively, by C wix,yi - njiwJw Zwixi2 - n(xw)2 b= (24) a = jjw - bXW (25) 1 Both these equations use the coordinates of the weighted centroid, (Xw, Jw), given by X, = zw,xi/n and Jw = Cwiyi/n, respectively; the weighted regression line must pass through this point. The standard deviation, sxow, and hence a con- fidence interval of a concentration estimated from a weighted regression line is given by 1 I where wo is an interpolated weight appropriate to the experimental yo value, and s(ylx)w is given by The confidence limits for weighted regression lines have the general form shown in Fig.7, with the weighted centroid closer to the origin than its unweighted counterpart. Calculations of weighted regression lines are evidently more complex than unweighted regression computations, and only the more advanced computer software packages provide suitable programs. The slope and intercept of a weighted regression line are often very similar to those obtained when Table 1 Anova table for linear regression Degrees of Source of variation freedom Sum of squares Mean square (MS) n c (jji - y)2 i = 1 Regression 1 PI bi - pi)* About regression ( i .e . , n - 2 1 = 1 ? (ri - pip $Ix = - 2 residual) n Total n - 1 ,bi - J)212 ANALYST, JANUARY 1991, VOL. 116 X Fig. 7 Confidence limits in weighted regression. The point 0 is the weighted centroid (Zw, Jw) unweighted calculations are applied to the same data, but only the weighted line provides proper estimates of standard deviations and confidence limits when the weights vary significantly with xi. Weighted regression calculations must also be used when curvilinear data are converted into rectilinear data by a suitable algebraic transformation (see below). The Analytical Methods Committee has recently suggested3 the use of the weighted regression equations to test for the linearity of a calibration graph. The weighted residual sum of squares (i.e., the squares of the y-residuals, multiplied by their weights and then summed) should, if the plot is linear, have a chi-squared distribution with (n - 2) DF. Significantly high values of chi squared thus suggest non-linearity. The same principle can be extended to test the fit of non-linear plots (see below). Partly Straight, Partly Curved Calibration Plots In many analytical methods a typical calibration plot is linear at low concentrations, but shows curvature towards the x-axis at higher analyte levels. This is often because an intrinsically non-linear relationship between instrument signal and concen- tration approximates to a linear function near the origin (e.g., fluorescence spectrometry). In such cases it would be logical to fit all the data, including the low-concentration points, to a curve.In practice, however, linear regression equations have been regarded as so much simpler to calculate than non-linear ones that the former are usually used over as wide a concentration range as possible. This gives rise to the question-to what upper concentration limit is the linear approximation satisfactory? A simple exploratory approach to this problem is exempli- fied in reference 22. It involves calculating with equations (1)-(3) the correlation coefficient, slope, intercept and SS of the residuals, first for all the calibration points, and then for the data sets with the highest, next-highest, etc. points being omitted successively. If the highest point(s) lie on a signifi- cantly non-linear portion of the plot, omitting them from the calculations will produce large reductions in the SS of residuals, significant increases of Y towards 1 (note again that absolute values are of little significance) and changes of a and b towards the values suggested by the calibration points near the origin.When the omission of further points produces only minor changes in r and the other parameters mentioned, then the linear portion of the graph has been successfully identified. During this stepwise removal of calibration points, a situation may arise where the omission of a point produces only slight improvements in the value of Y etc. , at the expense of a further restriction in the accepted linear range of the experiment. Judgement must then be exercised to balance the advantages of an increased linear range against the possible loss of accuracy and precision.Treatment of Non-linear Data by Transformations A large number of analytical methods are known to produce non-linear calibration graphs. In some cases these curvilinear relationships are only important at higher analyte concentra- tions (see above), but in other cases (e.g., immunoassays) the entire plot is non-linear. Methods for fitting such data to curves are well established (see below) but the simplicity of linear methods has encouraged numerous workers to try to transform their data so that a rectilinear graph can be plotted. The most common transformations involve the use of logarithms (i. e., plotting y against In x , or In y against In x ) and exponentials. Less commonly used transformations include reciprocals, square roots and trigonometric functions.Two or more of these functions are sometimes used in combination, especially in calibration programs supplied with commercial analytical instruments. This topic has been surveyed by Bysouth and Tys0n.~3 Draper and Smiths have reviewed such procedures at some length, and logical approaches to the best transformations have been surveyed by Box and Cox44 and by Carroll and Ruppert.45 It is important to note that all such transformations may affect the relative magnitudes of the errors at different points on the plot. Thus a non-linear plot with approximately equal errors at all values of x (homoscedastic data) may be transformed into a linear plot with heteroscedastic errors. It can be shown5,46 that if a function y = f(x) with homoscedastic errors is transformed into the function Y = BX + A , then the weighting factors, wi used in equations (24)-(27) above are calculated from wi = [ l , ( 3 ] ’ In some cases transformations make the use of a weighted regression plot less necessary.Thus a line of the form y = bx with y-direction errors dependent on x may be subjected to a log-log transformation. The errors in log y then depend less markedly on log x and, therefore, a homoscedastic approach may be reasonable. Similarly, Kurtz et al.47 have applied a series of power functions to transform chromatographic data to constant variance. In some analytical methods data transformations are so common that specialist software may be available to perform the calculations and present the results graphically.This is particularly true in the field of competitive binding immuno- assays, where several different (but related) transformations are in common use, including those involving the logit function { logit x = In[x/( 1 - x ) ] } and logistic functions such as y = A/(B + Ce-DX), with the A-D values to be found by iteration. (Immunoassay practitioners also use spline func- tions-see below .) These methods have been reviewed.48 Curvilinear Regression In many analytical methods, non-linear regression plots arise from several experimental factors, making it impossible to predict a model equation for the curve. For example in molecular fluorimetry, theory49 shows that a plot of fluores- cence intensity against concentration should be linear-but only as the result of several assumptions about the optical system used and the sample under study, and with the aid of a mathematical approximation.In practice some or all of these assumptions will fail, but to an unpredictable extent, giving a calibration graph that may approximate to a straight line near the origin (see above) but is, in reality, a curve throughout. In such cases it is sensible to adopt an empirical fit of a curve toANALYST, JANUARY 1991, VOL. 116 13 the observed data. This is most commonly attempted by using a polynomial equation of the form y = a + bx + cx2 + dx3 + . . . (29) The advantage of this type of equation is that, after the number of terms has been established, matrix manipulation allows an exact solution for a, b, c, etc.if the least-squares fitting criterion (see above) is used. Most computer packages offer such polynomial curve-fitting programs, so, in practice the major problem for the experimental scientist is to decide on the appropriate number of terms to be used in equation (29). The number of terms must clearly be <n for the equation to have any physical meaning and common sense suggests that the least number of terms providing a satisfactory fit to the data should always be used (quadratic or cubic fits are frequently excellent). Several approaches to this problem are available, the simplest (though probably not the best) being the use of the coefficient of determination, R2. As described above, this coefficient expresses the extent to which the total SS about jj can be explained by the regression equation under scrutiny.Values of R2 close to 1 (or 100%+omputer packages often present the result in this form) are thus apparently indicative of a good fit between the chosen equation and the experimen- tal data. In practice we would thus examine in turn the values of R2 obtained for the quadratic, cubic, quartic, etc. fits, and then make a judgement on the most appropriate polynomial. This method is open to two objections. The first is that, like the related correlation coefficient, r (see above), R2 can take very high values (>> 0.9) even when visual inspectinn shows that the fit is obviously indifferent. More seriously still, it may be shown that R2 always increases as successive terms are added to the polynomial equation, even if the latter are of no real value.(Draper and Smiths point out that this presents particular dangers if the data are grouped, i.e., if we have several y-values at each of only a few x-values. The number of terms in the polynomial must then be less than the number of x-values.) Thus, if this method is to be used, it is essential to attach little importance to absolute R2 values and to continue adding terms to the polynomial only if this leads to substantial increases in R2. An alternative and probably more satisfactory curve-fitting criterion is the use of the ‘adjusted R2’ statistic, given by50 R2 (adjusted) = 1 - (residual MS/total MS) (30) The use of the mean square (MS) instead of SS terms, allows the number of degrees of freedom (n - p ) , and hence the number of fitted parameters (p), to be taken into account.For any given data set and polynomial function, adjusted R2 always has a lower value than R2 itself; many computer packages provide both calculations. Among several other methods for establishing the best polynomial fit537 the simple use of the F-test1 has much to commend it. In this application (often referred to as a partial F-test), Fis used to test the null hypothesis that the addition of an extra polynomial term to equation (29) does not signifi- cantly improve the goodness of fit of the curve, when compared with the curve obtained without the extra term. Thus (extra SS due to addingx” term)/l residual MS for n-order model F = (31) The calculated Fvalue is compared with the tabulated value of Fl, n - p at the desired probability level; p is again the number of parameters to be determined, e.g., 3 (a, b, c), in a test to see whether a quadratic term is desirable.This test is simplified if the ANOVA calculation breaks down the SS due to regression into its component parts, i.e., the SS due to the term in x , the additional SS due to the term in x2 etc. It is worth noting that, if the form of a curvilinear plot is known, the calculation of the parameter values can be regarded as an optimization problem and can be tackled by methods such as simplex optimization.51 This approach offers no particular benefit in cases where exact solutions are also available by matrix manipulation, such as the polynomial equation (29), but may be very advantageous in other cases where exact solutions are not accessible.Again commercially available software is plentiful. A recent paper52 describes the calculation of confidence limits for calibration lines calculated using the simplex method. ?Vhichevermodel and calculation method is chosen to plot a calibration graph, it is desirable to examine the residuals generated and use them to study the validity of the chosen model. If the latter is suitable the residuals should show no marked trend in sign, spread or value when plotted against corresponding x or y values. As for linear regression, outlying points can also be studied. Spline Functions and Other Robust Non-linear Regression Methods As non-linear calibration plots often arise from a combination of physico-chemical effects (see above) and failure of mathem- atical approximations etc., it is perhaps unrealistic to expect that any single curve will adequktely describe such data.It may thus be better to plot a curve which is a continuous series of shorter curved portions. The most popular approach of this type is the cubic spline method, which seeks to fit the experimental data with a series of curved portions each of cubic form, y = a + bx + cx2 + dx3. These portions are connected at points called ‘knots’, at each of which the two linked cubic functions and their first two derivatives must be continuous. In practice, the knots may coincide with experimental calibration points, but this is not essential and a variety of approaches to the selection of the number and positions of the knots is available.Spline function calculations are provided by several software packages, and their applica- tion to analytical problems has been reviewed by Wold53 and by Weg~cheider.5~ A group of additional regression methods now attracting considerable attention relies on the use of fitting criteria other than or in addition to the least squares approach. In particular the ‘reweighted least squares’ method described in detail by Rousseeuw and Leroy55 utilizes the least median of squares criterion (i.e., minimization of the median of the squared residuals) to identify large residuals. The least squares curve is then fitted to the remaining points. These modern robust methods can of course be applied to straight line graphs as well as curves, and despite their requirement for more advanced computer programs they are already attracting the attention of analytical scientists (e.g., reference 56).Such developments provide timely reminders that the apparently simple task of fitting a straight line or a curve to a set of analytical data is still provoking much original research. I thank Dr. Jane C. Miller for many invaluable discussions on the material of this review. References 1 Miller, J. C., and Miller, J. N., Analyst, 1988, 113, 1351. 2 Martens, H., and Naes, T., Multivariate Calibration, Wiley, Chichester, 1989. 3 Analytical Methods Committee, Analyst, 1988, 113, 1469. 4 RGiiEka, J., and Hansen, E. H., Flow Injection Analysis, Wiley, New York, 2nd edn., 1988. 5 Draper, N. R., and Smith, H., Applied Regression Analysis, Wiley, New York, 2nd edn., 1981.6 Edwards, A. L., A n Introduction to Linear Regression and Correlation, Freeman, New York, 2nd edn., 1984. 7 Kleinbaum, D. G., Kupper, L. L., and Muller, K. E., Applied Regression Analysis and Other Multivariable Methods, PWS-Kent, Boston, 2nd edn., 1988.ANALYST, JANUARY 1991, VOL. 116 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 Working, H., and Hotelling, H., J. Am. Stat. Assoc. (Suppl.), 1929,24,73. Hunter, J. S., J. Assoc. Off. Anal. Chem., 1981, 64, 574. Massart, D. L., Vandeginste, B. G. M., Deming, S. N., Michotte, Y., and Kaufman, L., Chemometrics: A Textbook, Elsevier, Amsterdam, 1988. Cardone, M. J., Anal. Chem., 1986, 58, 433. Cardone, M. J., Anal. Chem., 1986, 58, 438.Jochum, C., Jochum, P., and Kowalski, B. R., Anal. Chem., 1981, 53, 85. Brereton, R. G., Analyst, 1987, 112, 1635. Analytical Methods Committee, Analyst, 1987, 112, 199. Specker, H., Angew. Chem. Int. Ed. Engl., 1968, 7, 252. Kaiser, H., The Limit of Detection of a Complete Analytical Procedure, Adam Hilger, London, 1968. IUPAC, Nomenclature, Symbols, Units and Their Usage in Spectrochemical Analysis-11, Spectrochim. Acta, Part B , 1978, 33, 242. Liteanu, C., and Rica, I., Statistical Theory and Methodology of Trace Analysis, Ellis Horwood, Chichester, 1980. Hadjiioannou, T. P., Christian, G. D., Efstathiou, C. E., and Nikolelis, D. P., Problem Solving in Analytical Chemistry, Pergamon Press, Oxford, 1988. Jandera, P., Kolda, S., and Kotrly, S., Talunta, 1970, 17, 443. Miller, J. C., and Miller, J. N., Statistics for Analytical Chemistry, Ellis Horwood, Chichester, 2nd edn., 1988. Swed, F. S., and Eisenhart, C., Ann. Math. Stat., 1943, 14,66. Cook, R. D., Technometrics, 1977, 19, 15. Rius, F. X., Smeyers-Verbeke, J., and Massart, D. L., Trends Anal. Chem., 1989,8,8. Thompson, M., Analyst, 1982, 107, 1169. Adcock, R. J., Analyst (Des Moines, Iowa, USA), 1878, 5, 53. Ripley, B. D., and Thompson, M., Analyst, 1987, 112, 377. Theil, H., Proc. K. Ned. Akad. Wet., Part A , 1950, 53, 386. Jaeckel, L. A., Ann. Math. Stat., 1972, 43, 1449. Hussain, S. S., and Sprent, P., J. R. Stat. SOC., Part A , 1983, 146, 182. Sprent, P., Applied Nonparametric Statistical Methods, Chap- man and Hall, London, 1989. Maritz, J. S., Aust. J. Stat., 1979, 21, 30. Maritz, J. S., Distribution-Free Statistical Methods, Chapman and Hall, London, 1981. Velleman, P. F., and Hoaglin, D. C., Applications, Basics and Computing of Exploratory Data Analysis, Duxbury Press, Boston, 1981. 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 Spearman, C., Am. J. Psychol., 1904, 15, 72. Kendall, M. G., Biometriku, 1938, 30, 81. Conover, W. J., Practical Nonparametric Statistics, Wiley, New York, 2nd edn., 1980. Francke, J.-P., de Zeeuw, R. A., and Hakkert, R., Anal. Chem., 1978,50, 1374. Garden, J. S., Mitchell, D. G., and Mills, W. N., Anal. Chem., 1980,52, 2310. Hughes, H., and Hurley, P. W., Analyst, 1987, 112, 1445. Thompson, M., Analyst, 1988, 113, 1579. Bysouth, S. R., and Tyson, J. F., J. Anal. At. Spectrom., 1986, 1, 85. Box, G. E. P., and Cox, D. R., J. Am. Stat. Assoc., 1984, 79, 209. Carroll, R. J., and Ruppert, D., J. Am. Stat. Assoc., 1984, 79, 321. Thompson, M., and Howarth, R. J., Analyst, 1980, 105, 1188. Kurtz, D. A., Rosenberger. J. L., and Tamayo, G. J., in Trace Residue Analysis, ed. Kurtz, D. A., American Chemical Society Symposium Series No. 284, American Chemical Society, Washington, DC, 1985. Wood, W. G., and Sokolowski, G.. Radioimmunoassay in Theory and Practice, A Handbook for Laboratory Personnel, Schnetztor-Verlag, Konstanz, 1981. Miller, J. N., in Standards in Fluorescence Spectrometry, ed. Miller, J. N., Chapman and Hall, London, 1981. Chatfield, C., Problem Solving: A Statistician's Guide, Chap- man and Hall, London, 1988. Morgan, S. L., and Deming, S. N., Anal. Chem., 1974,46,1170. Phillips, G. R., and Eyring, E. M., Anal. Chem., 1988,60,738. Wold, S.. Technometrics, 1974, 16, 1. Wegscheider, W., in Trace Residue Analysis, ed. Kurtz, D. A., American Chemical Society Symposium Series No. 284, American Chemical Society, Washington, DC, 1985. Rousseeuw, P. J., and Leroy, A. M., Robust Regression and Outlier Detection, Wiley, New York, 1987. Phillips, G. R., and Eyring, E. M., Anal. Chem., 1983,55,1134. Paper 01031 07K Received July loth, 1990 Accepted August 30th, 1990

ISSN:0003-2654    

DOI:10.1039/AN9911600003

出版商:RSC    

年代:1991

数据来源: RSC

摘要:

ANALYST, JANUARY 1991, VOL. 116 15 Liquid-Solid Extraction of Tributyltin From Marine Samples" Otis Evans, Betty J. Jacobs and Arnold L. Cohen Environmental Monitoring Systems Laboratory, US Environmental Protection Agency, 26 West Martin Luther King Drive, Cincinnati, OH 45268, USA An analytical method based on a liquid-solid extraction (LSE) procedure followed by gas chromatography (splitless 'hot' injection mode) with electron-capture detection (GC-ECD) has been developed and evaluated for tributyltin (TBT) chloride in fresh and marine waters. Splitless 'hot' injection GC requires the rigorous exclusion of water from the eluting solvent and the sample extract solutions. Tributyltin chloride is thermally sensitive and tends to degrade in the injection liner. This thermal degradation, possibly a debutylation reaction, is enhanced in the presence of trace amounts of water.Electron capture detection, although offering adequate sensitivity, has insufficient specificity for TBT chloride at trace level concentrations. Also, the solvent, ethyl acetate, can extract compounds from the LSE column or cartridge housings. These compounds, phthalates, cresols, adipates, etc., may interfere with the measurement and detection of the analyte. Chromatograms of extracts of LSE discs and procedural blanks have 'clean' backgrounds in contrast to many of the LSE columns and cartridges. Extraction discs mounted in a tandem arrangement show no breakthrough of analyte from the front to the rear disc for sample volumes which ranged from 100 to 500 ml.Liquid-solid extraction appears to meet the need for sample collection (on-site), preservation/storage (column or disc) and convenient and inexpensive sample shipment. The TBT chloride can be preserved on column(s) or disc(s) for at least 1 month. Recoveries of the tin analyte at the 0.1 ng ml-1 level range between 91 and 104%. Keywords: Liquid-solid extraction; tributyltin chloride; gas chromatography with electron-capture detection; splitless 'hot' injection; speciation Butyltin compounds, because of their versatility, have per- meated many aspects of human society, exerting a profound effect on the environment and the economy. These com- pounds have found extensive use in industry, agriculture, aquatic areas (fresh water and marine) and medicine.' The principal butyltin compounds are the tributyltins (TBTs).The TBTs are used as insecticides, fungicides, acaricides and preservatives for many different types of materials. However, their use in antifouling paints (as biocides) on ships, boats and docks and as slimicides in cooling towers has raised many issues and concerns.2-5 The organotin- based paints release the TBTs directly into the aquatic environment. Tributyltins are effective biocides against marine fouling organisms; however, they are non-specific and extremely toxic to non-target animal and plant species. These compounds have also caused major pollution problems in areas with restricted water circulation and significant rec- reational boating activity. As a result of the adverse manner in which these organo- tin compounds affect the environment, an awareness has developed of the need for reliable analytical methodology to identify and quantify them.The analytical research community has responded with a plethora of methods and techniques which address the speciation of organotins in biological tissues ,610 sediments7.9-'5 and various aqueous matrices. 14,1G2* The chemical literature is growing rapidly with the introduction of new and more sophisticated instrumentation29-32 which has the capability to detect and identify organotins at extremely low levels. This research was initiated by the US Environmental Protection Agency in order to meet the mandates of its monitoring programme and provide methods for the commun- ity to use. The work involved a review and evaluation of the technical literature, and in-house development of standard- ized methodology for organotin speciation. The general method criteria are rapid and inexpensive procedures which are selective for specific organotin compound(s).The current analytical needs of the Agency are: firstly, stan- dardized methodology for TBT in marine and fresh waters; secondly, a method detection limit (MDL) of 2-5 ng 1-l; thirdly, the identification and determination of TBT decom- position products and mixed butyltin compounds; and fourthly, methods for TBT and other organotins in sediments and biotissues. The liquid-solid extraction (LSE) procedure described here, initially reported by Junk and Richard,17>33 appears to meet the criteria of the Agency and current analytical needs with respect to TBT.However, the measurement-detection [splitless injection gas chromatography (GC) with electron- capture detection (ECD)] portion of the procedure presents certain problems that make consistent and accurate quantita- tive measurement difficult.34 We are also, to our knowledge, presenting the first scientific report, in the chemical literature, on the application of LSE discs [chemically bonded silica particles enmeshed in a polytetrafluorethylene (PTFE) matrix] for organometallic compounds in environmental matrices, i. e. , the extraction of TBT.34735 Recent publica- tions36,37 report the use of LSE discs for the extraction of organic analytes from aqueous matrices. Experimental? Materials and Reagents The solvents used were ethyl acetate and methanol (Fisher Scientific, Fairlawn, NJ, USA), Ultrex hydrochloric acid and ammonia solution (J.T. Baker, Phillipsburg, NJ, USA) and distilled, de-ionized water, prepared by passing distilled water through mixed bed cation- and anion-exchange resins. Liquid-solid extraction columns, Bakerbond [octadecyl (C,,) 40 pm, 1 ml, 100 mg (J. T. Baker)] and discs, Empore [extraction discs, octadecyl 25 mm (Analytichem Inter- national, Harbor City, CA, USA)] were used in the tributyltin chloride extractions. * This paper was presented in part at the 198th American Chemical Society National Meeting, Division of Environmental Chemistry, Miami Beach, FL, USA, September 13th, 1989. t Mention of trade names or commercial products does not constitute endorsement or recommendation for use by the US Environmental Protection Agency.16 ANALYST, JANUARY 1991, VOL.116 Standards The tributyltin chloride stock standard solution (Alfa Pro- ducts, Danvers, MA, USA) was prepared in methanol at a concentration of 100 mg 1-1 and stored in PTFE at 4°C. Subsequent dilutions were made with methanol and the solutions stored in glass at the same temperature. Individual working external standard solutions were prepared in ethyl acetate containing 15 p1 of 20% hydrochloric acid per 50 ml of solvent. Standards taken through the LSE procedure were prepared in distilled, de-ionized water adjusted to pH 4.5. Artificial sea-water (Forty Fathoms, Marine Enterprises, Towson, MD, USA) was used to simulate waters from the natural ocean environment.Instrumentation A Model GC-5880A gas chromatograph equipped with an electron-capture detector, and a 7672A automatic sampler (Hewlett-Packard, Avondale, PA, USA) were used. The 5880A was equipped with a split-splitless injection port operated in the splitless mode. A DB-1 fused silica capillary column (30 m x 0.32 mm i.d., 0.25 pm thickness) (J and W Scientific, Rancho Cordova, CA, USA) was used with helium as the carrier gas at a flow-rate of 3.9 ml min-1. The linear velocity was 20.8 cm s-1 at 160 "C (measured isothermally). The argon-methane (5% methane in argon) make-up gas flow-rate was 30 ml min-1. The injector and detector port temperatures were 200 and 260 "C, respectively. After injec- tion the column temperature was held isothermally at 80 "C for 1.0 min, then the temperature was increased at a rate of 15.0 "C min-1 to 180 "C and held for 10.0 min, and finally the temperature was increased again at a rate of 20.0 "C min-1 to 230°C and held there for 8.0 min in order to reduce column bleed interferences in ensuing runs.Samples were extracted with a Baker LSE column processor vacuum manifold (J. T. Baker) (stainless-steel basin), together with a vacuum hose fitting, a cover with Luer fittings and gasket, a vacuum gauge controller and manifold Luer plugs. A 304 stainless-steel syringe (pressure filter) holder, 25 mm, was used to support the LSE discs. Extraction column reservoirs (75 ml) and adapters (1,3 and 6 ml) were used when applying the sample to the LSE columns. A glass syringe (50 ml capacity) with a stainless-steel Luer tip was used as a reservoir when applying the sample to the extraction discs.A digital pipette (Brinkman Instruments, Westbury, NY, USA) was used to dispense specified volumes during standard and sample solution preparations. Extraction Procedure Adjust a 100 ml sample containing 1-2% v/v methanol to pH 4.5 (add 1-2 ml of methanol per 100 ml of sample). Take a 1 ml, 100 mg silica (C18) column or 25 mm PTFE enmeshed extraction disc (Clg) and add three column volumes (about 3 ml) of non-acidified ethyl acetate using a squeeze bottle. (After pre-conditioning the PTFE filter disc with ethyl acetate, create a vacuum in order to pull air through it for 5 min. The disc must not be allowed to become dry with subsequent conditioning and sample application.) Add four column volumes (about 4 ml) of methanol, 2 column volumes (2 ml) of de-ionized water and 2 column volumes of pH 4.5 de-ionized water.(Do not allow the column to become dry during additions of column conditioners and before the sample is added.) Next attach the sample reservoir to the column. (If the disc is used, the reservoir is attached prior to conditioning.) Add the sample solution and adjust the flow-rate to about 5 ml min-1. Following sample application, dry the LSE column andor the disc by drawing room air in through the device, using the vacuum manifold for at least 30 min, then place the LSE disc and/or column in a desiccator (charged with calcium sulphate) overnight to effect complete removal of all residual water. (It is essential that all residual water be removed from LSE discs and columns prior to elution of TBT with the eluting solvent.) The analyte (TBT) is eluted with two 250 p1 portions of ethyl acetate (acidified with 15 pL of HC1 perS0 ml of ethyl acetate) into a calibrated GC glass sample vial.[Each portion of ethyl acetate (HC1 acidified34) remains in contact with the column for at least 30 s.] The final volume of eluate (columddisc extract) is adjusted to 0.5 ml with a few drops of ethyl acetate (HC1 acidified). Finally, the sample vial is refrigerated overnight (4°C) to allow the extract solution to equilibrate. It was found that allowing the extract to stand overnight (refrigerated) resulted in more stable and consistent responses being achieved upon sample analysis. Caution. Tributyltin chloride is an extremely toxic and combustible substance.Pure standard material (liquid) and stock standard solutions of this compound should be handled with suitable protection for skin, eyes, etc. Results and Discussion The purpose of this research was to develop an analytical procedure for the detection and quantitative determination of tributyltin chloride in marine waters. Splitless injection capillary column gas chromatography with electron-capture detection was chosen as the determinate step because such equipment is common to most environmental laboratorie~.~9 On-column injection GC and element-selective detectors have been shown to improve the quality of the data and enhance sensitivity and selectivity and should be used whenever possible. Details concerning the many factors, i.e., interferences, moisture, thermal decomposition, analyte adsorption, and special conditioning requirements, affecting and influencing the LSE procedure and subsequent GC separation and detection, have been presented elsewhere.34 Fig.1 shows a chromatogram for a tributylin chloride external standard prepared in pure solvent. The analyte peak has a retention time of approximately 14.6 min. However, samples (and standards) taken through the extraction procedure can be affected by interferences which can originate from commer- 0 14 Time/min 28 Fig. 1 GC-ECD trace of a 100 ppb solution of tributyltin chloride (retention time, 14.65 min)ANALYST, JANUARY 1991, VOL. 116 17 cially available LSE columns and cartridges.40 The eluting solvent may extract these potentially interfering compounds from the polypropylene housing, polyethylene frit and CI8 bonded porous silica during the critical elution step (Fig.2). The number of compounds extracted, and the number of potential interferences varies between manufacturers and also between and within a particular batch of columns. A representative number of columns should be analysed from each batch before use in order to determine background variability. Extraneous peaks, if present, can be deleterious to trace level measurements of tributyltin chloride when using an electron-capture detector. The exceptional sensitivity of the detector can be compromised by its insufficient specificity at the parts per trillion (ppt) concentration level, Table 1 presents the results that can be obtained using this procedure, providing certain criteria are met.34 The mean recoveries are greater than 95% at the 25 ppt level.The recent introduction of LSE discs provides an opportun- ity for a comparison to be made between the two extraction technologies, and for the procedure to be evaluated further. Fig. 3 presents a series of chromatograms which show the apparent advantage of the LSE discs. Fewer extraneous peaks are observed, resulting in fewer potential interferences in the area of analyte elution (identified by the arrow in Fig. 3). However, a disadvantage of the discs is that they cannot be air dried (by vacuum) easily and as a consequence the extraction process is lengthened. Moisture (water) must be completely eliminated, because its presence is detrimental to the splitless injection GC of a thermally sensitive compound, such as t v) 0 a ,I, 0 14 28 0 14 Tirnelmin 28 Fig.2 porous silica columns GC-ECD traces of extracts from commercial c18 bonded tributyltin chloride. Accuracy and precision data for analyte measurements at the 50 ppt level, using the discs, are shown in Table 2. The results are comparable to those obtained by use of the extraction columns. Further studies were performed to determine the potential breakthrough volume of this particular extraction device. The typical sample volume used was 100 ml; however, to increase the sensitivity and lower the detection limit it was necessary to extract larger amounts of solution. Table 3 presents the results for discs mounted in a tandem arrangement for measuring breakthrough.There was no measurable breakthrough of the analyte from the front to the rear disc for all of the volumes attempted. As the final extract volume is 0.5 ml, enhancement factors of up to 1000 are achievable with 500 ml sample solutions. 0 14 28 Ti me/mi n Fig. 3 GC-ECD traces of LSE solutions from c18, 25 mm discs. (a) Extract from pre-conditioned disc; ( b ) extract from laboratory reagent blank; and ( c ) extract from sea-water containing 0.05 pg 1-1 tributyltin chloride Table 1 Accuracy and precision data for 18 determinations of the analyte (tributyltin chloride) at 0.025 pg 1-1 with LSE and GC-ECD, using a c18 silica-based column (100 mg) Concentration after extraction Relative Mean Standard standard accuracy Volume of Expected/ Mean observed/ deviation/ deviation (% true Technique sampldml pg 1-1 Yg 1-1 Pg I-' (%) concentration) LSE 200 10 10.3 0.8 7.7 102.9 GC-ECD 200 12.5 12.0 1.4 11.7 96.118 ANALYST, JANUARY 1991, VOL.116 Table 2 Accuracy and precision data for eight determinations of the analyte (tributyltin chloride) using a 25 mm C18 PTFE enmeshed extraction (filter) disc* Relative Mean True Mean Standard standard accuracy concentration observed/ deviation/ deviation (% true ng 1-1 ng 1-1 ng 1-1 (YO) concentration * 200 ml sample volume, standards taken through the LSE 50 52.7 3.79 7.2 105 procedure. Table 3 Recovery from tandem mounted discs.* Expected recovery, 20 pg 1-1; final volume (V,) (extract), 0.5 ml; enhancement factors (EF) (theoretical), 200, 400 and 10o0, respectively; 6-7 determina- tions per volume of sample; and ND = not detected Mean observed Volume of concen- sample/ Disc tration ml position pgl-1 100t Front 18.7 Rear (back) ND 200$ Front 22.3 Rear (back) ND 5009 Front 18.8 Rear (back) ND * 25 mm, Cls.t 0.1 pg 1-1 tributyltin chloride. $ 0.05 pg 1-1 tributyltin chloride. § 0.02 pg I-' tributyltin chloride. Mean Relative accuracy Standard standard (Yo true deviation/ deviation concen- pg 1-1 (Yo) tration) 0.97 5.2 93.5 1.52 6.8 111.5 0.92 4.9 94 The results indicate that the LSE of TBT can avoid the possible complications associated with a comparable liquid- liquid extraction (LLE). Liquid-liquid extractions tend to be time consuming and laborious. They are also expensive as large volumes of solvent may be required and many of these solvents are potential health hazards.The volume of eluting solvent in this procedure was less than 1 ml. Of course, slightly larger amounts were used during column conditioning. Other potential advantages that LSEs appeared to offer were further explored. Table 4 shows data for the preservation and storage of TBT on the extraction columns for approxi- mately 1 month. Comparable results have been achieved for analogous experiments using the discs. The space occupied by the latter is minimal and avoids the storage of bulky containers and the manpower required to handle them. The implication is that these devices would be convenient and inexpensive for the shipment of samples. Several research groups have investigated the preservation of TBT via freezing to determine applicability for storage and exchange of samples.28 However, an acknowledged disadvantage is the possible loss of analyte should the samples tend to thaw during transport.In our experiments, the extraction procedure was followed to the desiccation step, then the columns were refrigerated. For the evaluation phase the procedure was continued at the stage of elution into the GC vial. Subsequent experiments, of a shorter duration, indicated that TBT contained on an LSE disc or column could be stored at room temperature (in a desiccator). No further research is planned in this area. Conclusion The LSE of TBT from aqueous solution offers simplicity in extracting, concentrating, preserving and storing an important environmental pollutant of great concern.The discs were found to have a low background, good collection efficiency and no observable breakthrough. An electron capture detec- tor does not have the specificity to allow measurement of TBT at trace concentrations in environmental samples. Even Table 4 Tributyltin preservation on LSE columns. Initial concentra- tion 0.1 ng ml-l, 100 ml sample volume, and expected recovery 20 ng ml-1 Concen- tration after Relative Recovery No. of LSE, mean Standard standard (% true determina- observed/ deviation/ deviation concen- Week tions ng ml-1 ng ml-1 (%) tration) 1 4 19.8 1.1 5.5 99 2 4 19.2 0.8 4.0 96 3 3 19.5 1.1 5.8 97.5 4 3 18.6 0.4 2.3 93 though the results presented here are good, determinations of TBT by using splitless injection GC are tedious and require a considerable amount of time.Therefore, its use is not recommended for routine analytical measurements of this analyte. Element-selective detectors, atomic absorption, flame photometric, etc., should be strongly considered for complex aqueous matrices. The authors thank Craig Markell (3M Company) for the liquid-solid extraction discs, and Gregor Junk (Ames Laboratory-US Department of the Energy) for his helpful discussions and comments during the course of this work. We gratefully acknowledge the assistance and advice of Thomas Bellar, Thomas Behymer, William Middleton, Jerry Bashe, Jeffrey Collins and other members of the Chemistry Research Division of the Environmental Monitoring Systems Labora- tory and Technology Applications, Inc., Cincinnati, Ohio.1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 References Saxena. A. K., Appl. Organomet. Chem., 1987. 1, 39, and references cited therein. Proceedings of the Organotin Symposium of the Oceans, 86 Conference, Washington, DC, September 23-25, 1986. The Institute of Electrical and Electronics Engineers, New York, 1986. Proceedings of the International Organotin Symposium of the Oceans, 87 Conference, Halifax, Nova Scotia, Canada, Septem- ber 28-October I , 1987. US Environmental Protection Agency, Office of Research and Development, Environmental Research Laboratories, Ambient Aquatic Life Water Quality Advisories for Tributyltin, US Environmental Protection Agency, Duluth. MN, 1987, and references cited therein. Clark, E. A . , Sterritt, R.M., and Lester, J . N., Environ. Sci. Technol., 1988,22, 600, and references cited therein. Han, J. S . , and Weber, J. H., Anal. Chem., 1988, 60, 316. Stephenson, M. D., and Smith, D. R., Anal. Chem., 1988, 60, 696. Sullivan, J . J., Torkelson, J . D., Wekell, M. M., Hollingworth, T. A., Saxton, W. L . , Miller, G. A., Panaro, K. W., and Uhler, A. D., Anal. Chem., 1988, 60, 626. Tsuda, T., Nakanishi, H., Aoki, S., and Takebayashi, J., J. Chromatogr., 1987, 387, 361. Tsuda, T., Nakanishi, H., Morita, T., and Takebayashi. J., J. Assoc. Off. Anal. Chem., 1986, 69, 981. Maguire, R. J . , Environ. Sci. Technol., 1984, 18, 291. Muller, M. D., Anal. Chem., 1987, 59, 617. Epler, K. S . , O'Haver, T. C., Turk, G. C., and MacCrehan, W. A., Anal. Chem., 1988, 60, 2062.Hattori, Y., Kobayaski, A., Takemoto, S., Takami, K., Kuge, Y . , Sugimae, A., and Nakamoto, M.. J. Chromatogr., 1984, 315, 341. Cooney, J. J . , Kronick, A. T., Olson, G. J., Blair, W. R., and Brinckman, F. E., Chemosphere, 1988, 17, 1795. Matthias, C. L., Bellama, J . M., Olson, G. J., and Brinckman, F. E . , Environ. Sci. Technol., 1986, 20. 609. Junk, G. A., and Richard, J. J . , Chemosphere, 1987, 16, 61. Jewett, K. L., and Brinckman, F. E., J. Chromatogr. Sci., 1981, 19, 583. Ebdon. J . , Hill, S. J . , and Jones, P., Analyst, 1985, 110, 515.ANALYST, JANUARY 1991, VOL. 116 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Donard, 0. F. X., Rapsomanikis, S., and Weber, J. H., Anal. Chem., 1986,58, 772. Randall, L., Donard, 0. F. X., and Weber, J. H., Anal.Chim. Acta, 1986, 184, 197. Ebdon, L., and Alonso, J. I. G., Analyst, 1987, 112, 1551. Chau, Y. K., Wong, P. T. S., and Bengert, G. A., Anal. Chem., 1982, 54, 246. Hasebe, K., Yamomota, Y., and Kambara, T., Fresenius 2. Anal. Chem., 1982,310,234. Wang, J., and Zadeii, J., Talanta, 1987,34, 909. Maguire, R. J., and Huneault, H., J. Chromatogr., 1981, 209, 458. Chamsaz, M., and Winefordner, J. D., J. Anal. At. Spectrom., 1988, 3, 119. Valkirs, A. O., Seligman, P. F., Olsern, G. J., Brinkman, F. E., Matthias, C. L., and Bellama, J. M., Analyst, 1987, 112, 17. Lindsay, S. S . , and Pesek, J. J., J . Liq. Chromatogr., 1989, 12, 2367. Suyani, H., Creed, J., Davidson, T., and Caruso, J., J. Chromatogr. Sci., 1989,27, 139. Siu, K. W. M., Gardner, G. J., and Berman, S. S., Anal. Chem., 1989,61, 2320. Suyani, H., Heitkemper, D., Creed, J., and Caruso, J., Appl. Spectrosc., 1989,43, 962. 33 34 35 36 37 38 39 40 Junk, G. A., and Richard, J. J., J. Res. Natl. Bur. Stand., 1988, 93, 274. Evans, O., Jacobs, B., and Cohen, A., US Environmental Protection Agency Method 282.3, The Determination of Tri- butyltin Chloride in Marine and Freshwaters by Liquid-Solid Extraction and Gas Chromatography with Electron-Capture Detection (GCIECD), US Environmental Protection Agency, Cincinnati, OH, 1989. Evans, O., Jacobs, B., and Cohen, A. L., The Liquid-Solid Extraction of Tributyltin (TBT) from Marine Samples, Ameri- can Chemical Society (ENVR), Washington, DC, 1989. Brouwer, E. R., Lingeman, H., and Brinkman, U. A. Th., Chromatographia, 1990, 29,415. Hagen, D. F., Markell, C. G., Schmitt, G. A., and Blerins, D. D., Anal. Chim. Acta, 1990,236, 157. Wells, M. J. M., and Michael, J. L., J. Chromatogr. Sci., 1987, 25, 345. DeVeaux, R. D., and Szelewski, M., J. Chromatogr. Sci., 1989, 27, 513. Junk, G. A., Avery, M. J., and Richard, J. J., Anal. Chem., 1988,60, 1347. Paper 0103560B Received August 6th, 1990 Accepted August 30th, 1990

ISSN:0003-2654    

DOI:10.1039/AN9911600015

出版商:RSC    

年代:1991

数据来源: RSC

摘要:

ANALYST, JANUARY 1991, VOL. 116 21 Application of Tryptamine as a Derivatizing Agent for Airborne Isocyanate Determination Part 4." Evaluation of Major High-performance Liquid Chromatographic Methods Regarding Airborne Isocyanate Determination With Specific Investigation of the Competitive Rate of Derivatization Weh S. Wu, Robert E. Stoyanofft and Virindar S. Gaind Occupational Health Laboratory, Ontario Ministry of Labour, I0 1 Resources Road, Weston, Ontario M9P 3T1, Canada Several amines were investigated to determine their competitive rates in the derivatization of isocyanates. The a mines studied included N-(p-nitro benzy1)-N-propylam i ne, 1 -(2-pyridyl)piperazi ne, 1 -( 2-met hoxy- phenyl)piperazine, N-a-methyl tryptamine and tryptamine. Phenyl isocyanate, which might find application in future studies on sampling isocyanate vapour on solid absorbents, because it possesses a much higher vapour pressure than any of the industrially used isocyanates, was employed as the reference isocyanate.This approach was adopted because only the relative, rather than actual, rates of derivatization were of interest. By comparing the significant features of the methods for the determination of isocyanates using high-performance liquid chromatographic techniques, it was concluded that the proposed method, which uses tryptamine (and possibly N-co-tryptamine), was the most promising for practical application. The theoretical background of the proposed method was based on the isolation of a selected n-system in a derivative for specific detection. This approach should be applicable to other areas involving analysis with chromatographic techniques.The competitive rate study has also provided a better assessment in selecting a particular amine for further research in personal sampling of isocyanates on amine-coated solid absorbents. Keywords: Isocyanate derivatization; competitive derivatization rate; isolation of selected n-system High-performance liquid chromatography (HPLC) tech- niques for determining airborne isocyanates have been routinely practiced in occupational health laboratories. The reason for this is that high-performance liquid chromato- graphy is capable of handling larger molecules or compounds that are unstable to heat such as the industrially used isocyanates or their common derivatives.This has become even more important in recent years because the monitoring of airborne polymeric isocyanates in workplaces has been in demand. In spite of the many published isocyanate monitoring methods using HPLC techniques, the representative methods that have involved major developments are few and are discussed below in order to draw some distinct comparisons to the attention of the reader. Dunlap er al.1 developed the earliest significant method using N-(p-nitrobenzy1)-N-propylamine (NNNP) to derivatize isocyanates. Quantification of the derivatives is carried out by reversed-phase high-performance liquid chromatography with ultraviolet (UV) detection. This method has been applied almost exclusively to the determination of monomeric isocy- anates.The NNNP reagent, however, can only be purchased in the form of a salt owing to the instability of the amine itself. An additional disadvantage of this method is that it lacks the sensitivity of detection often required for the determination of extremely low levels of airborne isocyanates. Despite the drawbacks of this method, viewed by present day standards, many laboratories are still accustomed to applying it as the ultimate regulatory method for monitoring isocyanates in the workplace. Subsequently, several HPLC methods have been published, with the emphasis on improving the sensitivity of detection, * For Part 3 of this series see reference 6. t Present address: Concord Scientific Corporation, 2 Tippett Road, Toronto, Ontario M3H 2V2, Canada. but not for determining polymeric isocyanates. The most important methods are those of Levin er a1.,2 which uses the fluorescent reagent 1-naphthalenemethylamine to derivatize isocyanates, and Goldberg et al.,3 which uses 1-(2-pyridyl)- piperazine (PP) as the derivatizing agent to enhance the UV chromophore for detection. Interestingly, the latter workers have also used PP to derivatize a number of polymeric isocyanates.4 However, procedures for quantifying the deri- vatized polymeric components have not been addressed. In order to satisfy the increasing demand for monitoring both monomeric and polymeric isocyanates, Bagon et al.5 have devised a dual-detection HPLC method using 1-(2- methoxypheny1)piperazine (MPP) as the derivatizing agent. Polymeric components of a particular isocyanate are ident- ified by the ratio of the responses from both the amperometric oxidation and UV detectors. Each of these components is quantified to the corresponding amount of -NCO by calibra- tion with the parent monomer.However, this method has some major drawbacks; hence the reliability of the analysis is often open to question. Establishing a reliable response ratio from amperometric oxidation and UV detection is, in general, not an easy task. The reason for this is that the former is regarded as a detector with 'low stability' and the latter lacks both sensitivity and selectivity. This method, regardless of its drawbacks, has, nevertheless, provided an unprecedented approach to the determination of various polymeric iso- cyanates together with their monomers.In fact, in the UK, this method has been designated as the official regulatory method for monitoring total isocyanates exposure in work- places apparently because no other methods are available. Recently, Wu et a1.6 developed a method for determining total isocyanates with considerable reliability. This method uses tryptamine to react with all the isocyanates and the derivatives are quantified by a dual-detection system consist- ing of a fluorescence and an amperometric oxidation detector. As both detectors are very sensitive and selective, very low22 ANALYST, JANUARY 1991, VOL. 116 detection levels can be attained with minimum detection interferences. The uniqueness of this method is that all the tryptamine-derivatized isocyanates can be quantified by cali- bration against a single, pure standard such as tryptamine- derivatized toluene diisocyanate.The amount of isocyanato groups (-NCO) for individual components of the sample can be quantified without necessarily identifying the appropriate type of isocyanate. The theoretical basis of this method is that the n-orbitals of the indolyl moiety of tryptamine, responsible for the fluorescence and amperometric oxidation activities, are unperturbed before and after derivatization.7.8 In the present study this approach has been generalized to analysis with chromatographic techniques as the isolation of a selected n-system in a derivative for specific detection. The evaluation of analytical methods for determining isocyanates would be more conclusive if the derivatization reaction rates could be compared.When collecting airborne isocyanates in solutions of various derivatizing agents, the derivatization reaction rates would affect the over-all ef- ficiency of the methods. This is particularly important for sampling airborne isocyanates using a solid absorbent coated with a derivatizing agent, which would become indispensable for personal sampling in the workplace. A relatively slow reaction involving a reagent coated on a solid absorbent would be even slower than the same reaction occurring in solution. Unfortunately, information regarding competitive derivatiza- tion rates has not appeared in any of the published methods, which clearly indicates that some research in this area is necessary. Derivatization of an isocyanate (RNCO) with an amine (Am) is a second-order chemical reaction and the reaction rate can be written as follows: -d[RNCO]ldt = k[KNCO] [Am] (1) where k is the rate constant.Assuming the initial concentra- tions of amine and isocyanate are a and i, respectively, and the concentration of the derivative formed after a given time, t , is x, the reaction rate can be expressed as duldt = k(a - ~ ) ( i - X) ( 2 ) However, if a second amine is also involved in the derivatiza- tion to compete with the first amine, the rate kinetics for both reactions are given by equations ( 3 ) and (4), respectively duldt = k(a - X) (i - x - y ) dyldt = k’(b - y ) (i - x - y ) (3) (4) where b is the initial concentration of the second amine, y the concentration of the corresponding amine-derivatized iso- cyanate and k’ the corresponding rate constant. As both reactions proceed concurrently at different rates, the combined rate kinetics can be expressed by equation (5) with the time variable, t , being considered as a constant (5) By appropriately integrating equation ( 5 ) as shown in equa- tion (6), the relative rate (klk’) for the competitive derivatiza- tion reaction9910 is finally obtained [equation (7)] klk’ = log[(a - x ) / ~ ] / l o g [ ( b - y)lb] (7) Experimentally, it is unnecessary to conduct the rate study on industrially used diisocyanates because multiple derivatives would be produced due to random attack on the isocyanato groups by the amines thus complicating the investigation.As only the relative derivatization rate with the isocyanato functional group was of interest, a relatively pure and stable mono-isocyanate, i.e . , phenyl isocyanate, was preferred for this work. Experimental and Results Chemicals and Apparatus Tryptamine and N-a-methyl tryptamine (NMTP) were pur- chased from Sigma (St. Louis, MO, USA). Phenyl isocyanate, 1-(2-pyridyl)piperazine (PP) and 1-(2-methoxyphenyl)piper- azine (MPP) were from Aldrich (Milwaukee, WI, USA). N-4-Nitrobenzyl-N-propylamine (NNNP) was a Regis prod- uct, obtained through Caledon Laboratories (Georgetown, Ontario, Canada) as the hydrochloride salt. Amine solutions of NNNP were freshly prepared immediately before use. Final dilutions of these solutions were made in acetonitrile. All solvents were of glass-distilled quality and were obtained from Caledon Laboratories.Water was doubly distilled after treatment with KMn04. The HPLC system consisted of a Beckman 112 solvent delivery module, a Scientific System LP-21 pulse damper and a Shoeffel 970 fluorescence detector. The excitation wavelength was set at 275 nm and the emission wavelength was filtered at 320 nm. A 5 pm Hypersil-ODS column (25 cm x 4.6 mm i.d.) from Chromatography Science was used. The mobile phase was acetonitrile-water (50 + 50). The flow-rate was set at 0.8 ml min-1. The infrared (IR) spectrometer was a Beckman Model 4240 instrument and the spectrum of tryptamine-derivatized phenyl isocyanate (KBr disc) was obtained at a scan rate of 600 cm-1 min-1. Preparation of the Tryptamine Derivative of Phenyl Isocyanate A solution of phenyl isocyanate (1 g) in 10 ml of acetonitrile was added dropwise, with stirring, to 100 ml of an acetonitrile solution containing 0.5 g of tryptamine.The solution was allowed to stand for 1 h and the derivative was recrystallized from acetonitrile. The urea derivative [m.p. 196 “C (decomp.)] was also identified by the IR band at about 1650 cm-1. Study of Competitive Reaction Rates by Derivatizing Phenyl Isocyanate With Amines There were several options for conducting the experiments for this study. The competitive reaction rate is a relative rate between two competing amine reagents reacting concurrently with the isocyanate. Therefore, only two derivatizing agents were used for each specific rate study. To a set of 50 ml calibrated flasks, each containing a mixture of amines, 5 ml aliquots of acetonitrile solutions of phenyl isocyanate were added individually. For the sake of simplicity, the amount of the individual amines in each flask was kept constant while the amount of phenyl isocyanate added was varied.The contents of the flasks were allowed to react for 1 h before diluting to volume with acetonitrile. All solutions were diluted with acetonitrile to the appropriate concentrations for HPLC evaluation. A small amount of acetic anhydride (in acetonitrile), sufficient to remove the excess of amine, was also added to each flask before HPLC analysis. The data for the competitive derivatization rates for tryptamine and MPP, tryptamine and PP, tryptamine and NNNP, and NMTP and MPP are listed in Tables 1 4 . A typical HPLC trace of a solution of phenyl isocyanate containing tryptamine and MPP is shown in Fig.1. Study of Competitive Reaction Rates by Derivatizing Phenyl Isocyanate With Tryptamine and Water As water is known to react with isocyanates and exists in the atmosphere, it is important to have some knowledge about its relative reaction rate in comparison with those of the amines. It is also known that isocyanates react much more slowly with water than with amines. Therefore, the amount of water used for the experiments has to be fairly large in order toANALYST, JANUARY 1991, VOL. 116 Table 1 Relative rate for derivatizing phenyl isocyanate with tryptamine and MPP 23 Phenyl iso- Tryptamine MPP (b)l cyanate (i)/ PI-TP* (x)l PI-MPPt (y)l (a)lpmol p o l pmol pmol p o l Log[(a - x)/a] Log[@ - y)/b] 1.563 1.115 1.050 0.599 0.451 - 0.2 1006 - 0.22485 -0.17512 1.563 1.115 0.840 0.470 0.370 -0.15534 1.563 1.115 0.630 0.349 0.281 -0.10974 -0.12611 - 0.08240 1.563 1.115 0.420 0.227 0.193 -0.06825 1.563 1.115 0.210 0.105 0.105 -0.03029 - 0.04282 klk' = 0.979$ * Tryptamine derivative of phenyl isocyanate.t MPP derivative of phenyl isocyanate, obtained with the assumption that i = x + y . $ Obtained from the slope of the log[(a - x)/a] versus log[(b - y)/b] plot. Table 2 Relative rate for derivatizing phenyl isocyanate with tryptamine and PP Phenyl iso- Tryptamine PP (b)/ cyanate PI-TP* ( x ) l PI-PPt ( y ) l (a)/ pmol pmol (i)/pmol pmol pmol Log[(a - x)/a] Log[(b - y)lb] 1.563 1.154 1.050 0.724 0.326 -0.27020 -0.14418 1.563 1.154 0.840 0.569 0.271 -0.19657 -0.11625 1.563 1.154 0.630 0.407 0.223 -0.13100 -0.09326 1.563 1.154 0.420 0.280 0.140 -0.08573 - 0.056 17 1.563 1.154 0.210 0.128 0.082 -0.03711 - 0.03201 klk' = 2.011 * Tryptamine derivative of phenyl isocyanate.t PP derivative of phenyl isocyanate, obtained with the assumption that i = x + y . Table 3 Relative rate €or derivatizing phenyl isocyanate with tryptamine and NNNP Phenyl iso- Tryptamine NNNP (b)/ cyanate PI-TP* (x)/ PI-NNNP? (y)/ (a)lpmol pmol (i)/pmol pmol pmol Log[@ - x)/a] Log[@ - y ) h ] -0.00976 1.563 15.03 1.050 0.716 0.334 - 0.26608 1.563 15.03 0.840 0.560 0.280 -0.19266 - 0.0081 7 1.563 15.03 0.630 0.449 0.181 -0.14707 - 0 .00526 -0.09949 -0.00290 1.563 15.03 0.420 0.320 0. 100 1.563 15.03 0.210 0.163 0.047 -0.04783 -0.00136 klk' = 23.63 * Tryptamine derivative of phenyl isocyanate.t NNNP derivative of phenyl isocyanate, obtained with the assumption that i = x + y . Table 4 Relative rate for derivatizing phenyl isocyanate with NMTP and MPP Phenyl iso- NMTP (a)/ MPP (b)l cyanate PI-NMTP* PI-MPPt pmol pmol (i)/pmol bJ)/Pmol (x)lpmol Log[(a - x>la] Log[(b - y ) h ] - - 1.437 O$ 1.050$ 1.050 0 1.437 2.800 1.050 0.640 0.410 -0.25600 -0.06876 1.437 2.800 0.840 0.512 0.328 -0.19132 - 0.0541 1 - 0.03790 1.437 2.800 0.630 0.396 0.234 -0.14001 1.437 2.800 0.420 0.259 0.161 -0.08631 -0.02572 -0.01 195 1.437 2.800 0.210 0.134 0.076 - 0.0425 1 klk = 3.744 * NMTP derivative of phenyl isocyanate. t MPP derivative of phenyl isocyanate. $ Derivatization without addition of MPP to the reaction mixture, used for calibrating the amount of PI-NMTP produced in the set.differentiate the rate from that of the amine. The competitive reaction rate for tryptamine and water is shown in Table 5. Summary of Competitive Derivatization Rates and Over-all Comparison of the Methods The competitive derivatization rates obtained above would be more meaningful if relative rates could be assigned to all the amines investigated. Table 6 shows the relative rate constants listed in descending order; the rate for MPP was arbitrarily assigned a value of 100. An over-all comparison of the various HPLC methods is presented in Table 7. Experimental data reflecting the relative rate constant of klk' are plotted in Fig. 2. Discussion Although many HPLC methods for determining airborne isocyanates have been published, no study of the competitive derivatization rate of amines has been reported.This work has shown that the relative rates for the derivatization of phenyl isocyanate by MPP and tryptamine are almost identical and24 ANALYST, JANUARY 1991, VOL. 116 -TP ,TP tlmin Fig. 1 Chromatogram of competitive derivatization of phenyliso- cyanate with tryptamine and MPP. (a) Solution containing 1.05, 1.56 and 1.12 pmol of phenyl isocyanate, tryptamine and MPP, respec- tively, before dilution; and (b) solution containing 0.63, 1.56 and 1.12 , pmol of phenyl isocyanate, tryptamine and MPP, respectively, before dilution Table 5 Relative rate for derivatizing phenyl isocyanate with tryptamine and water Phenyl iso- Tryptamine H20 (6)l cyanate (i)l PI-TP* (x)/ PI-H20t (a)lpmol pmol pmol pmol (y)/pmol 1.563 1.39 x 104 1.050 AS B§ 1.563 1.39 x 104 0.840 A B 1.563 1.39 x 104 0.630 A B 1.563 1 .3 9 ~ 104 0.420 A B 1.563 1.39 x 104 0.210 A B Log[(a - x)/a] > - 0.47 log[(b - y)/6] 2 - 3 x 10-67; klk' > 1 x 10s * Tryptamine derivative of phenyl isocyanate. t Derivative of phenyl isocyanate with water. S Approximately 100% yield. 0 Approximately zero. 7 Assuming the HPLC technique fails to differentiate up to a 5% yield of PI-H20. Table 6 Relative reaction rates for derivatizing phenyl isocyanate with various amines Derivatizing agent Relative rate constant, k NMTP MPP Tryptamine PP NNNP Water 374 100 98 49 4 a x 10-5 are two and 25 times faster than those for PP and NNNP, respectively. Experiments conducted earliela by using a Test Atmosphere Generation System indicated that the recoveries of airborne toluene diisocyanate in the respective impinger solutions of MPP and tryptamine were indeed very competi- tive.As can be seen from Table 6, NMTP is the most reactive of the amines investigated, being about four times more reactive than MPP. However, a detailed study of the application of NMTP to the determination of isocyanates has not been conducted because of the high cost of this amine. Further, the amines used for derivatization are always present in a large excess, which is unlikely to affect the over-all yield of the derivatives caused by reactions with slightly slower rates. On the other hand, a very much slower derivatization rate was obtained with NNNP. The main concern is that the NNNP method is still being used widely for quantifying airborne isocyanates in workplaces and may not be able to reflect the true exposure.The results indicate that a much smaller amount of tryptamine or MPP is required than of NNNP to efficiently derivatize an equivalent amount of isocyanate. This would also be of benefit in the HPLC system, because less material would need to be loaded on to the column if tryptamine were to be used for derivatization. It has been observed in previous experimental work6 that the derivatization of isocyanates with NNNP is less efficient than with tryptamine or MPP, as reflected by the consistently lower results. It was suspected that the solvated water might have been partly responsible as NNNP has to be extracted from water before use.However, the relative derivatization rate study indicates that this is not the situation as it would have affected the rate of the other competing amines 300 I L I I 0 100 200 [-Log(a - x)/al x 100 Fig. 2 Plot of second-order reaction kinetics for competitive rates for derivatizing phenyl isocyanate with amines. A, Tryptamine and NNNP; B, NMTP and MPP; C, tryptamine and PP; and D, tryptamine and MPP Table 7 Comparison of representative HPLC methods for determining isocyanate Derivatizing agent Availability for Type of No. of specific determining of unidentified detection detection modes pol yisocyanate isocyanate Availability for quantifying NCO* NNNP U1 traviolet NAMAS Fluorescence PP Ultraviolet MPP Ultraviolet and amperometric TryptamineP Fluorescence and amperometric ot 1 0 1 2 No No No Yes Yes No No No No Yes * Reactive isocyanato functional group.t UV absorption common for organic chemicals, regarded as a non-specific detection mode. $ 1-Naphthalenemethylamine. P Employed in the proposed method.ANALYST, JANUARY 1991, VOL. 116 25 simultaneously if the derivatization rate for water were competitive. In fact, the relative rate for water to compete with amine for the derivatization of isocyanate is negligible. For instance, our work shows that the derivatization of phenyl isocyanate by tryptamine is at least 1 x 105 times faster than by water. A slower derivatization rate for NNNP could also be caused by the instability of the amine form of the reagent. However, as the NNNP solutions for derivatization were always pre- pared freshly before use, the instability of the amine form of the reagent does not appear to be the reason for the slower rate.Therefore, it is more probable that the nature of the reaction kinetics is responsible, but the exact cause is still unknown. It should be noted that in all the experiments conducted on competitive derivatization rates the actual yield of tryptamine- derivatized phenyl isocyanate was used as the basis for all the necessary calculations. This was because of the highly fluorescent nature of this derivative. The analytical inter- ferences on high-performance liquid chromatography would be minimized by fluorescence detection. One of the most important reasons for performing the relative rate study was to provide the best possible assessment in developing personal sampling techniques for isocyanate exposure in the workplace.The preferred air collection media for personal sampling at work sites are solid absorbents instead of impinger solutions. For a reagent-coated solid absorbent, this implies that the derivatization would occur in the unfavourable solid phase of the reagent. Amines with faster derivatization rates would therefore be more suitable for coating the solid absorbents. Considering all the factors such as the sensitivity and selectivity of detection of the methods discussed for isocyanates, the use of tryptamine is recommended for future studies on sampling airborne iso- cyanates using solid absorbents. By comparing all the features of the HPLC methods shown in Table 7, the proposed method is the most suitable for practical applications. Moreover, the proposed concept of the isolation of a selected x-system in a derivative for specific detection has promoted a new area for exploration in analysis with chromatographic techniques. References 1 Dunlap, D . A . , Sandridge, R. L., and Keller, J . , Anal. Chem., 1976,48, 497. 2 Levine, S. P., Hoggatt, J . H . , Chladek, E., Jungclaus, G . , and Gerlock, J . L., Anal., Chem., 1979, 51, 1106. 3 Goldberg, P. A . , Walker, R. F., Ellwood, P. A . , and Hardy, H. L., J. Chromatogr., 1981,212,93. 4 Walker, R. F., Ellwood, P. A., Hardy, H. L., and Goldberg, P. A . , J. Chromatogr., 1984,301,485. 5 Bagon, D. A . , Warwick, C. J . , and Brown, R. H., Am. Znd. Hyg. Assoc. J . , 1984, 45, 39. 6 Wu, W. S., Stoyanoff, R. E., Szklar, R. S . , Gaind, V. S . , and Rakanovic, M., Analyst, 1990, 115, 801. 7 Wu, W. S . , Nazar, M. A . , Gaind, V. S . , and Calovini, L., Analyst, 1987, 112, 863. 8 Wu, W. S . , Szklar, R. S., and Gaind, V. S., Analyst, 1988,113, 1209. 9 Wibaut, M. J. P., Rec. Trav. Chim., 1915, 34, 241. 10 Ingold, C. K., and Smith, M. S . , J. Chem. SOC., 1938,905. NOTE-References 6, 7 and 8 are to Parts 3, 1 and 2 of this series, respectively. Paper 01031 62 C Received July 13th, 1990 Accepted September 7th, 1990

ISSN:0003-2654    

DOI:10.1039/AN9911600021

出版商:RSC    

年代:1991

数据来源: RSC

摘要:

ANALYST, JANUARY 1991. VOL. 116 27 Determination of Titanium(1v) in River Water by Ion-pair Reversed-phase High-performance Liquid Chromatography With 4,4' -Dia n ti pyry I met ha ne Nobuo Uehara, Kazuhiro Morimoto and Yoshio Shijo Department of Applied Chemistry, Faculty of Engineering, Utsunomiya University, Ishii-cho, Utsunomiya 32 I , Japan An ion-pair reversed-phase hig h-performance liquid chromatographic method for the selective determina- tion of Tiiv with 4,4'-diantipyrylmethane (DAM) is described. The TiIV-DAM complex was separated on an ODS column using acetonitrile-water (30 + 70) containing 1 x 10-4 mol dm-3 DAM, 0.01 mol dm-3 ammonium iodide and 0.02 mol dm-3 chloroacetate (pH 2.25). The detection limit for TiIV with the proposed method is 1.8 pg I-'. Titanium(iv) in river water can be determined without the interference of foreign ions after pr e-co ncen t rat ion.Keywords : Titanium( I v) ; 4,4 '-dian tip yrylm ethane; ion -pair re versed-p hase high -perf0 rmance liquid chroma t- ography; river water; pre-concentration 4,4'-Diantipyrylmethane (DAM) is the most popular reagent for the spectrophotometric determination of Ti" because of its selectivity.1J The reagent has been applied to the deter- mination of Ti1" in cements,3 ferro-niobium,4 silicate rocks5 and plant materials.6 However, it is difficult to use DAM for the determination of low levels of Ti" because the molar absorptivity of the Ti"-DAM complex is not very large. Recently, more sensitive spectrophotometric reagents for Ti'" have been reported. Inoue et al.7 described an extraction spectrophotometric method for Ti1' with N-p-octyloxy- benzoyl-N-phenylhydroxylamine and phenylfluorone. Marini et al. 8 used 2-(5-chioro-2-pyridylazo)-5-dimethylaminophenol and hydrogen peroxide for the determination of Ti". Gregorowicz el aZ.9 determined Ti'" in steel with Eriochrome Azurol G. However, these reagents are not as selective as DAM. Most of the work combining high-performance liquid chromatography (HPLC) with spectrophotometric detection has been aimed at the development of sensitive and selective analytical methods for determining metal ions. However, there are relatively few reports on the application of HPLC to Ti". Main and Fritz15 determined Ti'" by HPLC with bis(quaternary ammonium hydrazones) of 2,6-diacetylpyri- dine. The detection limit of Ti'" was 8 x 10-8 mol dm-3 when 2,6-diacetylpyridinebis(N-methylenepyridinohydrazone) was used as the chelating agent.However, the molar absorptivities of these Ti"-bishydrazone complexes are smaller than that of the Ti'"-DAM complex. Therefore, DAM appears to be more sensitive than these bishydrazones for the determination of Ti'" by HPLC. This paper describes the use of DAM as a pre-column chelating agent for the sensitive determination of Ti" by ion-pair reversed-phase HPLC. The proposed method was applied to the determination of Ti'" in river and rain water after pre-concentration using a simple and rapid evaporation of the sample solution. Experimental Apparatus The liquid chromatographic system consisted of a Nihon Seimitsu Kagaku (Tokyo, Japan) NSP-800-3U pump, a Japan Spectroscopic (Tokyo, Japan) hexane dumper, a Japan Spectroscopic 870-UV spectrophotometric detector, a Rheodyne 7125 loop injector with a 20 p1 sample loop and a Shimadzu (Kyoto, Japan) U-125 MU recorder. A Yamamura Kagaku (Tokyo, Japan) YMC R-ODs-5 column (250 X 4.6 mm i.d.) was used for all experiments.The simple laboratory- built evaporator shown in Fig. 1 was used for pre-concentra- tion; it was of the same type as that described in a previous paper. 16 Reagents Distilled, de-ionized water was purified further by means of a Millipore Milli-Q system. Analytical-reagent grade aceto- nitrile and methanol were filtered through a Millipore filter (0.45 pm) after distillation. The DAM was obtained from Dojindo (Kumamoto, Japan).The TitV standard solution (1000 mg 1 - 1 ) for atomic absorption spectrometry was obtained from Wako Pure Chemicals (Osaka, Japan). All other chemicals were of guaranteed-reagent grade. Evaporation Glass cylinder U Cooling t t /water Fig. 1 Schematic diagram of the laboratory-built evaporator28 ANALYST, JANUARY 1991, VOL. 116 Eluent and Chromatographic Conditions The eluent used was acetonitrile-water (30 + 70) containing 1 x 10-4 rnol dm-3 DAM, 0.01 mol dm-3 ammonium iodide and 0.02 mol dm-3 chloroacetate buffer (pH 2.25). The pH of the eluent was adjusted before addition of acetonitrile. The flow-rate of the eluent was 1.0 ml min-1. The eluate was monitored at 390 nm. Procedure Ten millilitres of 60% nitric acid were added to 1.0 1 of sampled river water and the solution was filtered through a Millipore filter (0.45 pm).To 10 ml of river water sample, 1.0 ml of 60% nitric acid and 0.1 ml of 3 mol dm-3 sulphuric acid were added. The solution was evaporated to dryness with the laboratory-built evaporator. The residue was dissolved in 0.4 ml of 2% DAM solution (0.5 mol dm-3 HC1) by shaking for 2 min and the solution was allowed to stand for 20 min. To a 0.3 ml aliquot of the solution, 0.1 ml of 4 mol dm-3 sodium chloroacetate solution and 0.1 ml of methanol were added. An aliquot of the solution (20 pl) was injected on to the HPLC column. Results and Discussion Derivatization Studies The TP-DAM complex was formed within 15 min in an acidic medium of 0.5 rnol dm-3 HCI. However, direct injection of 0.5 mol dm-3 HC1 solution into the chromatograph is inadequate.Therefore, the pH of the solution was adjusted to 2.25, which is a suitable pH for the ODS column. The solution must be injected into the chromatograph immediately after adjustment of the pH because the complex decomposes gradually at pH 3 9c L.LJ. HPLC Studies Fig. 2 shows a typical chromatogram of the TP-DAM complex. All the other metal ions studied, viz., A P , Crvl, CuII, Fell1, InlI1, NP, Mn", Vv and Z P , gave no peaks on the chromatogram under the conditions used. Iodide ion, which has been used for the ion-pair extraction of TitV with DAM was examined as a counter anion for HPLC of the TP-DAM complex, as the complex has a positive charge.2 Fig. 3 shows the effect of ammonium iodide concentration on the mass distribution ratio of the Ti'"-DAM complex.As the ammonium iodide concentration increased the mass distribution ratio of the complex also increased, which indicated that ammonium iodide acted as an ion-pair t v) 0 n F E 8 a L al I I I 1 I L 0 2 4 6 8 10 Time/m i n Fig. 2 Chromatogram of the TP-DAM complex. Column, YMC R-ODS-5 (250 x 4.6 mm i.d.); eluent, acetonitrile-water (30 + 70) containing 1 x mol dm-3 DAM, 0.01 mol dm-3 ammonium iodide and 0.02 mol dm-3 chloroacetate (pH 2.25). Ti'", 0.1 mg 1-1; flow-rate, 1.0 ml min-1; detection wavelength, 390 nm; detector sensitivity, 0.01 a.u.f.s.; and injection volume, 20 p1 reagent. The peak of the complex was well resolved from the solvent front above an ammonium iodide concentration of 1 X 10-2 mol dm-3.Acetonitrile was a good organic modifier for the retention of the TP-DAM complex. The mass distribution ratio of the complex decreased as the content of acetonitrile in the mobile phase increased, as shown in Fig. 4. An acetonitrile content of 30% v/v was selected in order to obtain a suitable chromato- gram for the detection of TiIV. The retention behaviour of the TP-DAM complex was investigated for various pH values of the eluent. As the pH of the eluent increased, the mass distribution ratio of the complex also increased and the peak broadened, as shown in Fig. 5 Sach behaviour appears to be due to hydrolysis of the complex and interaction between the residual silanol groups of the ODS column and the complex. The DAM concentration of the eluent also affected the retention of the TP-DAM complex.The peak height of the complex increased and the mass distribution ratio decreased with an increase in the DAM concentration. The addition of DAM to the eluent prevented the decomposition of the complex and allowed its rapid elution. 8 t 6 QE 4 2 0 -3 -2 -1 Log (cNH,,/mol dm-3) Fig. 3 distribution ratio. For chromatographic conditions see Fig. 2 Effect of concentration of ammonium iodide on the mass 4 I\ d 2 0 ' I 20 30 40 Acetonitrile content (% v/v) Fig. 4 For chromatographic conditions see Fig. 2 Effect of acetonitrile content on the mass distribution ratio. 2 & 1 1 lo0g .- a r Y I : 2.0 3.0 4.0 PH Fig. 5 peak height. For chromatographic conditions see Fig. 2 Effect of pH of the eluent on the mass distribution ratio andANALYST, JANUARY 1991, VOL.116 29 Table 1 Effect of foreign ions on the determination of TitV. Concentration of Ti1V added, 0.1 mg 1-l Concentration added/ Error* Foreign ion mgl-l W) 100 100 100 100 50 50 20 20 20 20 20 20 20 20 20 20 0.5 +1.3 +2.5 +2.2 -3.1 +2.9 -0.9 +1.3 +0.2 +0.9 +2.9 -3.3 -3.6 -0.9 -3.5 -1.2 +1.8 -4.9 * A plus sign indicates a positive error, a minus sign indicates a t Ascorbic acid was added as a masking agent. negative error. Calibration Graph and Detection Limit The calibration graph of peak height versus TiIV concentration was a straight line in the concentration range 0.01-0.3 mg 1-1 TiLV with 20 p1 injections. The detection limit for TiIV was 1.8 pg 1-1 at a signal to noise ratio of 3. The reproducibility of the method is 3.9%, expressed as the relative standard deviation for ten replicate analyses of solutions containing 0.1 mg 1 - 1 TiIV .Interferences The effect of foreign ions on the determination of TiIV with HPLC was studied. Table 1 shows the results. None of the metal ions tested interfered with the determination of TitV at concentrations commonly found in river water.17 Only Bi"' interfered seriously with the determination of TiIV. However, the interference from Bi"' would not be a problem in practice because the concentration of Bi"' in river water is very low. The presence of a large amount of FelI1 gave a broad peak which interfered with the determination of T P . However, the Fellr peak disappeared completely on adding ascorbic acid to the sample solution, because Fe" does not react with DAM.Applications The simple evaporation system used in a previous paper16 was employed for pre-concentration because the HPLC method Table 2 Results for the determination of TiIV in river and rain water Concentration of Ti1"*/ I - % - ' Sample Watarase river 1.85?0.10(n=4) Kinu river 0.76 k 0.09 ( n = 3) Rain water? 2.66 k 0.27 ( n = 3) * Mean k standard deviation. t Sampled at Utsunomiya University. was not sufficiently sensitive to determine Ti'" in river water directly. The 15-fold enrichment achieved by pre-concentra- tion enabled TitV to be determined in river water by HPLC. The water sampled at the Watarase and Kinu rivers (both in Tochigi, Japan) and also rain water were analysed by the standard additions method. To 10 ml of the water sample, 0-80 pl of 1 mg 1-1 TiIV standard solution were added.The results are summarized in Table 2. The determination of TiIV with the proposed method is sensitive and free from interfer- ences. The technique can be used for the determination of pg 1-l levels of TiIV in the presence of large amounts of foreign ions after pre-concentration. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 References Cheng, K. L., Ueno, K., and Imamura, T., Handbook of Organic Analytical Reagents, CRC Press, Boca Raton, FL, 1982, p. 23. Ishii, H . , Bunseki Kagaku, 1972, 21, 665. Ishii, H., Bunseki Kagaku, 1967, 16, 110. Kajiyama, R., and Yamaguchi. K., Bunseki Kagaku, 1967, 16, 908. Chung, C., Anal. Chirn. Acta, 1983, 154, 259. Tusl, J . , Chem. Listy, 1988, 82, 1303. Inoue, S., Takahashi, T., Hoshi, S . , and Matsubara, M., Bunseki Kagaku, 1988, 37, 316. Marini, H. J., Anton, R. I . , and Olsina, A., Bull. Chem. SOC. Jpn., 1987, 60, 2635. Gregorowicz, Z., Cebura, J., Gorka, P . , and Kowalski, S., Chem. Anal. (Warsaw), 1987, 32, 505. Nickless, G., J. Chrornatogr., 1985, 313, 129. O'Laughlin, J. W., J. Liq. Chrornatogr., 1984, 7, 137. Willeford, B. R., and Veening, H., J. Chromatogr., 1982, 251, 61. Suzuki, N., and Saitoh, K., Kagaku no Ryoiki Zokan, 1983, 138, 127. Yotsuyanagi, T., and Hoshino, H . , Bunseki, 1983, 556. Main, M. V., and Fritz, J. S., Anal. Chern., 1989, 61, 1272. Uehara, N . , Morimoto, K., Shimizu, T., and Shijo, Y., Chern. Lett., 1989, 411. Ogura, N., and Fukushima, K., Bunseki, 1990, 181. Paper OlO33.536 Received July 24th, 1990 Accepted September loth, 1990

ISSN:0003-2654    

DOI:10.1039/AN9911600027

出版商:RSC    

年代:1991

数据来源: RSC

摘要:

31 ANALYST, JANUARY 1991, VOL. 116 - Simultaneous Determination of Silver and Copper by Flame Atomic Absorption Spectrometry With Alternate Irradiation by Two Hollow Cathode Lamps LS2 . Osamu Sakurada, Shunitz Tanaka and Mitsuhiko Taga" Department of Chemistry, Faculty of Science, Hokkaido University, Sapporo 060, Japan Tei ji Ka kiza ki Laboratory of Chemistry, Hokkaido University of Education at Iwamizawa, lwamizawa 068, Japan i PI0 ' CPU The simultaneous flame atomic absorption spectrometric determination of silver and copper is described. The resonance lines of both silver (328.07 nm) and copper (327.40 nm) were introduced within the same bandpass of the monochromator. Therefore, it was possible to measure the absorption signals of silver and copper simultaneously, when the silver and copper hollow cathode lamps were alternately irradiated. Acquisition of the absorption data was synchronized with the irradiations using a computer.For the introduction of the sample into the flame, a discrete nebulization method was investigated in order to minimize the sample volume required and the analysis time. Keywords: Simultaneous atomic absorption spectrometry; flame atomization; discrete nebulization; silver; copper N D Atomic absorption spectrometry (AAS) is an excellent analytical method in terms of selectivity and sensitivity. The selectivity of AAS is attributed to the use of an analyte-specific resonance line being emitted from the radiation source, a hollow cathode lamp (HCL). However, this indicates that for simultaneous multi-element determination by AAS, it is necessary to prepare multi-channel radiation sources and optical systems for each element.Simultaneous multi-element determination by AAS has been attempted by many workersl.2 and the instrumentation is now commercially available.3 Harnly et al. proposed a multi-element AAS system consisting of a continuum light source such as an Xe arc lamp, a high-resolution echelle polychromator and a computerized high-speed data system. Nakamura and Kubotas also reported an instrument for multi-element AAS consisting of a specific multi-element HCL, a single detection channel with one photomultiplier tube (PMT) and a time-divided high-speed data acquisition system. However, these instruments are very complicated and expensive to use for routine analyses. On the other hand, the spectral interferences that arise in AAS from the overlap of the absorption lines of the analytes consequently lead to large experimental errors .6 Therefore, a graphite furnace AAS method has been developed for the simple simultaneous determination of copper and silver by the use of their neighbouring resonance lines.' This method is based on the difference in the appearance temperatures of the analytes in the graphite furnace.Silver is atomized at a lower temperature than copper. The mixed radiation of the resonance lines of silver (328.07 nm) and copper (327.40 nm) is introduced into the graphite furnace atomizer simultaneously. By measurement of the peak heights of silver and copper on a chromatographic absorption-time profile, the simultaneous determination of silver and copper is achieved.In this work, the simultaneous determination of silver and copper by flame AAS was investigated using the neighbouring resonance lines. In flame AAS, the difference in the appear- ance temperature, used in the graphite furnace method, cannot be expected. Therefore, alternate irradiation of the sample with the silver and copper HCLs was attempted. By utilizing a computerized high-speed data acquisition system to * To whom correspondence should be addressed. collect the absorption signals alternately, the absorption signal for each element can be obtained with one measurement. A discrete nebulization methods for the introduction of the sample into the flame was also investigated in order to minimize the sample volume required and the analysis time.The proposed simultaneous method is simpler and less expensive than other methods. The method was applied to the determination of silver and copper in commercially available silver brazing filler metals used for welding. Experimental Apparatus A Hitachi Model 170-50 atomic absorption spectrometer was used, with a pre-mix burner for the air-acetylene flame. In order to introduce the mixed radiation of the resonance lines of silver and copper into the flame, the optical system for deuterium background correction was used as the secondary radiation source. As shown in Fig. 1, the silver and copper HCLs were placed in the positions of the HCL and the deuterium lamp (in the normal mode), respectively. The light beams from the silver and copper HCLs were pulsed at 50 Hz alternately, as shown in Fig.2. The light beams were made spatially coincident using the half-mirror and subsequently passed through the flame atomizer into the spectrometer. The HCL2 - 1 HCLl M PMT32 I HCL2 Cu HCL ANALYST, JANUARY 1991, VOL. 116 HCL1 . Ag HCL spectral bandpass of the monochromator was set at 1.1 nm to detect both the silver and copper resonance lines. The voltage analogues of the logarithmic converted circuit of the PMT (Hamamatsu R456) were converted into digital data by a 12 bit analogue to digital converter ( N D ) [Contec AD12-16A(98)]. Timing of the A/D cycle was obtained using software-controlled radiation pulses from the HCLs, with a parallel input-output module [Contec (PI0-48W( 98)].The vertical scale in Figs. 3-6 is shown in volts (V) and is comparable to absorbance. The peak area integrated the peak height with the time (s). A Model PC-9801UV2 personal computer (Nippon Electric) was used for data acquisition. Software to perform the data acquisition and analysis was written in BASIC and incorporated a Machine Code sub- routine for rapid data acquisition.9 The discrete nebulization method was used to introduce the sample solution into the flame. The device for this was assembled from a miniature polytetrafluoroethylene funnel connected to the nebulizer capillary. The sample solution was injected into the funnel and nebulized in the flame. The computer-controlled data acquisition was triggered by the detection of the sample passing through the nebulizer capillary by means of a photocoupler (Omron Model EE-SV3) registering the change in the light transmission. The absorbance-time signals were digitally stored in the memory of the computer.The stored time, namely, the time required for the sample volume, was pre-set, normally 4 s for a 100 p1 injection. The system for the detection of the injection was a modification of the system proposed by Goto et al.lO Micropipettes (Eppendorf 4700 and 4710) were used for sample injection. 1 Reagents Standard solutions of silver and copper were prepared by dissolving silver nitrate (analytical-reagent grade, Wako Pure Chemicals) in lmol dm-3 HN03 and by dissolving copper metal (99.999% pure, Mitsuwa Pure Chemicals) in concentrated nitric acid and diluting with water to a final concentration of 1 mol dm-3 HN03.Other reagents were of analytical-reagent grade. Doubly distilled water was used throughout. - Results and Discussion Flame Conditions For the simultaneous atomic absorption spectrometric determination of silver and copper, it is important to determine the optimum flame conditions for the two elements. In order to do this, response surface graphs11312 were constructed from the results of the atomic absorption spectrometric determination of silver and copper plotted as burner height versus acetylene pressure. The burner height values were indicated by the scale reading on the AAS apparatus. In the distribution pattern of silver and copper shown in Fig. 3, the maximum absorbance peaks are located at Off (1.1 4ms 4ms Fig.2 HCLs and ( b ) signal sampling (a) Timing chart of emission signals from the silver and copper a flame height of approximately 2.5 and an acetylene pressure of 0.25 kg cm-2. In subsequent experiments, the following experimental conditions were chosen as the optimum flame conditions; air flow pressure, 1.6 kg cm-2; acetylene flow pressure, 0.25 kg cm-2; and burner height, 2.5 (arbitrary units). Absorbance Signals of Copper and Silver Obtained by Alternate Irradiation of the Sample Silver and copper are simultaneously atomized and hence there is usually no time difference in the appearance of the signal, in conventional AAS. Therefore, the different absorp- tion signals cannot be distinguished when the mixed resonance lines of silver and copper pass through the flame at the same I I 6 3l L 0 a L .- Y 3 a L 0 0.2 0.3 0.4 0.5 0.6 0.7 P C ~ H J ~ S cm-* 2 0 0.2 0.3 0.4 0.5 0.6 0.7 P C 2 H J b cm-2 Fig.3 Response surfaces of silver and copper obtained by continu- ous nebulization. Sample solution of (a) silver ( 5 ppm); and (b) copper (5 ppm). Burner height value was indicated by the scale reading on the AAS apparatusANALYST, JANUARY 1991, VOL. 116 33 time. However, if the silver and copper HCLs alternately irradiate the sample and acquisition of the absorption data is synchronized with the irradiation, it is possible to distinguish between the absorption signals of silver and copper. The two HCLs are pulsed at 50 Hz alternately as shown in Fig. 2 and both radiations pass through the flame. By utilizing a computerized high-speed data acquisition system to collect both of the absorption signals alternately, each element could be determined with a single measurement.The absorbance- time profiles for silver and copper, obtained simultaneously using the instrument previously mentioned, are shown in Fig. 4(a). A sample solution of silver (5 ppm) and copper ( 5 ppm) *I 0 0 m Time Fig. 4 Absorption profiles of silver (top) and copper (bottom). (a) Continuous-flow nebulization method. (b) Discrete nebulization method. Sample injection volume was 100 pl. The numbers on the profiles refer to sample composition and correspond to the concentration of the metal in ppm z Y z 2 a 1 0 1 0 50 100 150 Injection vol u rne/pl Fig. 5 Effect of injection volume on peak height and peak area of (a) silver and ( b ) copper obtained with 5 ppm silver and 5 ppm copper mixed solutions in 0.1 mol dm-3 HN03 was injected by using a conventional method.The absorbance-time profiles obtained in the mixed solution were compatible with the signals obtained in each single element solution. Mutual interference due to the coexistence of silver and copper was not observed. Conse- quently, it is clear that the simultaneous determination of silver and copper can be performed with the alternate irradiation method. In conventional flame AAS, the continuous flow injection method is popular for sample injection. However, it has the disadvantage that a large sample volume is required. In order to minimize the sample volume required and the analysis time, a discrete nebulization method was investigated.The signal peak shape obtained with this method is shown in Fig. 4(b). Effect of Injection Volume The dependence of the peak height and the peak area on various injection volumes of the mixed solution, containing 5 ppm of Ag and 5 ppm of Cu, is shown in Fig. 5. The peak height increases with the volume of sample solution injected up to about 100 p1. Thereafter, the limit of the peak height is the same as the height obtained with continuous aspiration. 6 3 s E 0 r 0) .- Y m n 2 1 0 0 1 2 3 4 5 6 Concentration (ppm) Fig. 6 Calibration graphs for (a) silver and ( b ) copper obtained with a constant volume (100 pl) of mixed solutions of silver and copper Table 1 Results for the determination of silver and copper in silver brazing filler metals. Figures in parentheses represent the relative standard deviations (YO) of five analyses.The proposed method of simultaneous flame AAS used a 100 p1 injection volume; flame AAS was the conventional single-element flame method using continuous-flow nebulization. Reference values are as reported by the manufacturer Content (YO m/m) Proposed method Peak Peak Sample Element height area Low melt* Ag 43.9 44.1 (0.5) (2.4) Cu 15.6 15.2 (2.7) (1.8) SampleNo. 318t Ag 25.7 25.2 c u 34.4 33.7 (2.3) (1.3) (2.1) (2.1) * Obtained from Nippon Yushi. t Obtained from Kinzokuyouzai. Reference Flame value AAS (%m/m) 45.1 44-46 15.3 14-16 25.7 5 2 5 34.8 =35 (1.1) (2.5) (1.4) (2.5)34 ANALYST, JANUARY 1991, VOL. 116 The reproducibility of the signal is improved by increasing the injection volume.With an injection volume of greater than 100 pl, the relative standard deviation for ten determinations of peak height is less than 1%. The peak-area value shows a linear relationship to the injection volume. Calibration Graph The calibration graphs for silver and copper, as the absorbance-time profiles, are shown in Fig. 6. The peak height and peak area gave straight lines for silver and copper. As a constant injection volume was used, there was a linear relationship between the peak-area value and the concentration. Application The proposed simultaneous AAS method using discrete nebulization was applied to the determination of silver and copper in two types of silver brazing filler metal alloys that consisted of silver, copper, zinc and cadmium.Samples were dissolved in nitric acid. The results shown in Table 1 were obtained by measuring the peak height and peak area and are in agreement with the reference values reported by the manufacturers. In addition, the proposed method was vali- dated by conventional single-element flame AAS using continuous-flow nebulization. In applying the Student’s t-test to the two methods, there was no significant statistical difference in the results obtained using the two methods. The simultaneous determination of silver and copper by flame AAS was also performed using the neighbouring resonance line. The method requires only a simple modifi- cation to the normal AAS system. The proposed method can be employed in order to minimize both the sample volume and the analysis time. 1 2 3 4 5 6 7 8 9 10 11 12 References Busch, K. W., and Morrison, G. H., Anal. Chem.. 1973. 45, 712A. Harnly, J . M., Anal. Chem., 1986, 58, 933A. Okumoto, T., Tsukada, M., Tobe, H., Kitagawa, M., Yone- tani, A., and Sawakabu, H., The Hitachi Scientific Instrument News, 1987,30, 2787. Harnly. J. M., O’Haver, T. C., Golden, B., and Wolf, W. R., Anal. Chem., 1979, 51, 2007. Nakamura, S., and Kubota, M., Analyst, 1990, 115. 283. Vajda, F., Anal. Chim. Acta, 1981, 128, 31. Taga, M., Tanaka, S., and Sakurada, O., Bunseki Kagaku, 1989, 38,403. Cresser, M. S., Prog. Anal. At. Spectrosc., 1981, 4, 219. Kakizaki. T., Hikma, S., and Hasebe, K., Anal. Sci., 1989, 5, 781. Goto, K., Uchida. T., and Iida, C., Rev. Sci. Instrum., 1983,54, 291. Mossholder, N. V., Fassel, V. A., and Kniseley, R. N., Anal. Chem., 1973, 45, 1614. Fujiwara, K., Haraguchi, H., and Fuwa, K., Anal. Chem., 1975, 47, 743. Paper 0103357J Received July 24th, 1990 Accepted September 11 th, 1990

ISSN:0003-2654    

DOI:10.1039/AN9911600031

出版商:RSC    

年代:1991

数据来源: RSC

摘要:

ANALYST, JANUARY 1991. VOL. 116 35 Two-step Coprecipitation Method for Differentiating Chromium Species in Water Followed by Determination of Chromium by Neutron Activation Ana I ysis Chi-Ren Lan, Chia-Liang Tseng and Mo-Hsiung Yang* Institute of Nuclear Science, National Tsing Hua University, 30043 Hsinchu, Taiwan Zeer B. Alfassi Department of Nuclear Engineering, Ben Gurion University, 84 102 Beer Sheva, Israel A two-step method for the determination of Crvl and Crlll in natural waters was developed. The method is based on the variation of the coprecipitation yields with Pb(PDC)2 (PDC = pyrrolidine dithiocarbamate) as a function of pH. By using two different pH values both species can be determined separately. Firstly, Crvl was coprecipitated at pH 4.0 and then Cr"1 was separated at pH 9.Total chromium was determined by reduction of Crvl followed by coprecipitation at pH 9. The validity of the procedure was checked with the National Institute of Standards and Technology Standard Reference Material 1643b Trace Elements in Water and the result was found to be in good agreement with the certified value. Keywords: Water; chromium species analysis; ammonium pyrrolidine dithiocarbamate; coprecipitation; neutron activation analysis Chromium is present in natural waters in two different oxidation states, Cr111 and CrVI. The former is considered an essential element in mammals, whereas the latter is con- sidered to be a toxic material.14 Thermodynamic calculations indicated that in natural waters Cr should exist almost exclusively as CrV1.7 However, it was found experimentally that the actual ratio of Cr"' to CrV1 in natural waters varied from 0.02 to 0.99.8 Arrhenius and Bonattig pointed out that this variation and contradiction with theory might be due to the in situ coprecipitation of chromate only, with strontium or barium sulphate.This selective coprecipitation can also be used as an analytical tool for the separation of the two species followed by their individual determination. Chuecas and Riley10 studied the coprecipitation of Cr from water using W r as a radiotracer. They found that both aluminium and iron(m) hydroxides (hydrated oxides) will coprecipitate CrlI1 effi- ciently. The pH range for 99% coprecipitation is considerably larger for iron(1ri) hydroxide (pH 7.0-9.0) than for aluminium hydroxide (pH 7.5-8.0).When coprecipitating CrV1 spiked with slCrv1, about 1.2% of the Cr was precipitated. This might be due to partial coprecipitation caused by a small amount of slCr111 in the radiotracer. Fukai11 measured both Cr111 and CrV1 in sea-water by coprecipitation with iron(r1r) hydroxide first from the untreated water and then after reduction of the sample with sodium sulphite in acidic medium. Several studies have been carried out on the extraction of Cr species by means of ammonium pyrrolidine dithiocarbamate (APDC)-ethyl methyl ketone (EMK) or diethyl dithiocarbamate(DDTC) with EMK. However, the percentage extraction varied considerably (5&100%).12,13 De Jong and Brinkman14 selectively determined CrVI and Cr111 in sea-water using solvent extraction,. They found that CrVI was extracted with high efficiency (>99%) from various acidic solutions with tertiary amines.These workers used a pH of 2 (0.01 mol dm-3 HCI) and Aliquat 336 as the extractant; Aliquat 336 is a mixture of methyl trialkylammonium chlorides with alkyl groups that are mainly C8-Cl0. The extracting organic solvent was toluene. Chromium(1ii) was not extracted at all in this medium. However, by using the same extractant at pH 6-8 and in the presence of at least 1 mol dm-3 thiocyanate (in CC14 rather than toluene, in order to dissolve * To whom correspondence should be addressed. the KSCN), CrI11 was quantitatively extracted while none of the CrV1 was extracted in this pH range. One of the methods of Cr speciation recommended by the US Environmental Protec- tion Agency (EPA)15 involves the extraction of CrVI with APDC into isobutyl methyl ketone (IBMK); however, several problems are associated with this method.16 Isozoki et a1.17 studied the different ionic species of Cr in natural water using ion-exchange chromatography; the work was similar to that of Naranjit et a1.18 and Leyden et a1.19 Isozoki et al.used a chelating column of Chelex 100 for the quantitative adsorption of Cr"1 while CrVI was not absorbed and passed through the column. Chromium(v1) was deter- mined after reduction to Cr"1. Miyazaki and Barnes20 used a poly(dithi0carbamate) chelating resin. Only CrV* was retained on the column, the Cr"' passed through. Total Cr was obtained after oxidizing CrIII to CrV' with KMn04 in acidic media.Wai et a1.21 differentiated between Cr"' and CrVI by extracting CrVI from natural waters into chloroform with DDTC followed by back-extraction into aqueous HgII solu- tion for determination by graphite furnace atomic absorption spectrometry. The Cr"1 remaining in the extracted solution can be oxidized to CrVI with KMn04 and then extracted with DDTC. Subramanian22 developed procedures using the APDC- IBMK extraction system for the selective determination of Cr111, and the simultaneous determination of Cr"* plus CrV1, without the need to convert Cr"1 into CrV1. He used phthalate buffer and found that at about 0.02-0.1% of phthalate, both species were extracted efficiently, whereas above 0.8% phthalate, only CrVI was quantitatively extracted (all in the pH range 2.5-4.0).For both atomic absorption spectrometry and inductively coupled plasma atomic emission spectrometry and also for spectrophotometric determination it is preferable to obtain the concentrated sample in a liquid phase; however, for neutron activation analysis (NAA) and also for X-ray fluor- escence spectrometry it is preferable to have the sample in the form of a solid. This is particularly true for Cr as thermal neutron activation leads to two radionuclides of which the short-lived radio- nuclide "Cr has a low abundance of the parent isotope (2.36%), a relatively small cross-section for formation (0.36 barn) and, most significantly, it is almost a pure (3-emitter and emits very few y-rays (0.043%). The lower limit of detection using this radionuclide is very high and measurement of Cr by36 ANALYST, JANUARY 1991, VOL.116 NAA is carried out using the long-lived radionuclide 51Cr (half-life, 27.71 d). In order to obtain high sensitivity the sample should be irradiated at high fluxes for long periods of time (several days or at least 10-20 h). With these long irradiation times liquid samples will suffer considerable radiolysis, leading to the formation of large amounts of gases, which are likely to lead to explosion of the irradiation ampoule. Hence a solid sample should be used and, rather than drying the liquid sample, it is preferable to pre-concen- trate the trace elements by coprecipitation. The best copreci- pitants for NAA will be compounds that have small thermal neotron absorption cross-sections and which do not form radioisotopes on neutron absorption or where the radioiso- topes formed are very short-lived and/or are only (3-emitters.Materials that fulfil these criteria are organic compounds of lead and bismuth. In earlier w0rks~23-27 the precipitation of several trace elements from natural waters and biological fluids with Pb(PDC)2 and Bi(PDC)3 (PDC = pyrrolidine dithiocarbamate) have been described; the present work is concerned with the two species of Cr, viz. , Cr"' and CrvI. Nakayama et ~ ~ 1 . ~ 8 found that in the pH range 4-10 both CrIII and CrVI were coprecipitated with bismuth oxide. Hence they first coprecipitated Cr"1 with iron(m) hydroxide at about pH 8, and then CrV1 was collected at the same pH with bismuth oxide. Pik et a1.29 first coprecipitated CrIII with iron(rI1) hydroxide at pH 8.5 and then coprecipitated CrVI from the remaining solution with Co(PDC)2 at pH 4.Experimental All chemicals used were of analytical-reagent grade and the solutions were prepared using doubly distilled, de-ionized water. Two methods were used to determine the coprecipitation yields of CrIII and CrVI with Pb(PDC)2. In the first, the radiotracer 51Cr was used. The coprecipitation yield was measured from the radioactivity counts [with an NaI(T1) detector] of the original radiotracer solution and the precipi- tate. Solutions of 5lCrvI and 51C1-111 were prepared by the method of Collins et a1.30 About 50 mg of K2Cr04 or K2Cr207 were irradiated in the reactor for 2-3 d. The irradiated salt was dissolved in 5 ml of a solution containing 15 mg of Zn(N03)2 and 50 mg of Cr03.After complete dissolution, 3 ml of 1 mol dm-3 KOH were added and the solution was heated at 90-95 "C for about 30 min. While the reaction mixture was still hot, 1 ml of a solution containing 3 mg of Zn(N03)2 was added, with stirring. The resulting suspension was filtered through a 10 x 6 mm i.d. column of alumina or celite washed with 5 ml of 1 mol dm-3 KOH and 5 ml of 1 X 10-4 mol dm-3 KOH, and kept wet with a KOH solution of pH 10. The filtered solution was used as a radiotracer for CrVI. Some of the 51Cr changed its valency due to the Szilard-Chalmers process and was retained on the column as Cr"1. The 51CrIII was eluted from the column with 5 ml of 1 mol dm-3 HC1 and was used as a radiotracer for CrIII.In the second method, standard solutions of unlabelled CrIII and CrVI were used and the standard solution and the precipitate were analysed simultaneously by NAA. The coprecipitation studies were carried out by the addition of 1 ml of a standard solution containing 1 mg ml-1 of either CrI" or CrVI to 250-500 ml of tap water (for the radiotracer experiments) or distilled water (for the NAA experiments) followed by the addition of 2 mg of Pb(N03)2, 5 ml of 1 mol dm-3 acetate buffer and adjustment of the pH with HN03. Then, 100 mg of APDC were added to the solution which was stirred for 30 min. The precipitates were filtered through a 0.45 pm Millipore filter (Gelman) and dried in a desiccator containing silica gel until completely dry. In the NAA experiments the dried samples were introduced into polyethylene vials which were heat-sealed. A standard was prepared by introducing 1 ml of the standard solution into a polyethylene vial and heating to dryness under an infrared lamp.The precipitates and the standard were placed in an irradiation capsule and irradiated for 30 h at a flux of 5 x 1012 n cm-2 s-1. The radioactivity induced in the samples was measured by a Ge(Li) detector connected to a multi-channel analyser, after 1 week of cooling. The coprecipitation yields were calculated as the radioactivity ratio of the samples to the standard. Results and Discussion The results for the recovery of Cr"1 and CrVI by coprecipita- tion with Pb(PDC)2 as a function of pH for 250 ml of distilled water, tap water and sea-water are given in Table 1.As can be seen from these results, it is possible to precipitate CrVI almost exclusively in the pH range 2.5-4.5. Chromium(Ir1) cannot be coprecipitated alone, as at pH 9 about 1644% of CrVI is also precipitated. However, a two-step coprecipitation on the same sample can give information about the concentrations of both CrVI and Cr"'. The sample is adjusted to pH 4 and Pb(N03)2 (2 mg) and APDC (100 mg) are added to the solution. The solution is stirred for 30 min and then filtered on a 47 mm Millipore filter (0.45 pm). The precipitate on the filter-paper is used to determine CrVI. The filtrate is adjusted to pH 9 with 25% ammonia solutions (about 3 ml). Then, 100 mg of APDC and 2 mg of Pb(N03)2 are added, the solution is stirred (for 30 min) and filtered on a 47 mm diameter 0.45 pm Millipore filter and the precipitate is used for the determina- tion of CrIII.Table 2 gives the results for the recovery of CrIII and CrVI from 2 1 of sea-water or tap water spiked with CrIII or CrVI at a total concentration of 20 ng ml-1. As can be seen from this table even at this low concentration there is nearly a 100% recovery (about 95-105%). Total Cr Determination Total Cr can be determined either by oxidation of Cr"1 to CrVI and coprecipitation at pH 4 or by reduction of CrVI to Cr"' and coprecipitation at pH 9.0. It is easier to reduce CrVI than to oxidize CrIII and consequently the latter option was preferred. Chromium(v1) was reduced with NaHS03. Table 1 Recovery of CrlI1 and Crw by coprecipitation with Pb(PDC)2 as a function of pH. All values in per cent.Conditions: water sample, 250 ml; Cr"' or Crw, 100 pg; APDC, 100 mg; and Pb(N03)2, 2 mg Distilled water Tap water Sea-water PH 3 4 5 6 7 8 9 Cr"1 CrVI Cr"' Crw Cr"' Crm 0.8 92 0.8 92 0.5 95 1.8 98 1.8 95 0.5 95 3.6 95 3.6 95 0.5 95 13.2 95 13 95 4 34 88 58 66 58 67 14 93 52 93 52 96 14 97 44 97 44 95 16 Table 2 Recovery tests carried out by spiking sea-water and tap water samples with Cr"' or CrVI. Water sample, 2 1; Cr, 20 ng ml-l Recovery (%) Sample CrVI Sea-water 101.6 95.7 105.3 Aver age : 100.9 Tap water 102.9 101 .o 100.7 Average : 101.5 Cr"1 98.5 99.2 102.1 99.9 98.6 95.9 103.1 99.2ANALYST, JANUARY 1991, VOL. 116 37 Table 3 Recovery of CrVI by reduction and precipitation at pH 9 as a function of the amount of NaHS03 added (CrVI, 104 pg) AmountofNaHS03/mg 1.7 6.7 16.7 33.3 66.6 83.5 167 500 Recovery (% ) 25.4 28.7 41.5 56.3 100 99.3 97.6 99.9 Sample (2 I) w pH = 4.0 100 mg of APDC 2 mg of Pb(NO3I2 stirred for 30 rnin 62 mg of NaHS03 boiled for 5 min i o o mg of APDC 2 mg of Pb(NO3j2 stirred for 30 min pH = 9.0 2 mg of Pb(N03j2 stirred for 30 min mg of APDC NAA NAA NAA CrVl only Crlll only Total amount of Cr Fig 1 Scheme for the speciation of Cr"1 and CrVI and for the determination of total Cr Different amounts of NaHS03 were added to a series of solutions containing 104 pg of CrvI.The solutions were stirred and then placed in a microwave oven for heating (2 min at 100% power followed by 3 min at 60% power). After boiling for 5 min the solutions were rapidly cooled in an ice-bath.The pH was adjusted to 9 with 1 ml of 25% ammonia solution and the Cr"1 in the solution was then determined as described above. Table 3 shows the percentage recovery of CrVI obtained with this method as a function of the amount of NaHS03 added. It can be seen that 66.6 mg of NaHS03 are sufficient for complete reduction and recovery. Consequently in later experiments, about 70 mg of NaHS03 were used for the reduction of CrVI in the procedure for the determination of total Cr. Fig. 1 illustrates schematically the determination of the two species of Cr and of total Cr. This method of separate determination of the two species of Cr has advantages over most of the previous methods in that it does not require the oxidation of Cr"1 to Crvr or the reduction of Crvr to Cr"' which is usually carried out with an excess of the reagents.The proposed method does, however, suffer from the disadvan- tage that even if one is interested only in Cr"', one must first coprecipitate CrVI. If only CrIII is required, then coprecipita- tion with hydrated iron(i1i) hydroxide might be a better procedure. Real Samples Three samples of sea-water, well water and tap water were analysed according to the scheme illustrated in Fig. 1. The results are given in Table 4. It can be seen that the agreement between the values for CrVI plus Cr"1 and total Cr is good. A certified reference water, viz., National Institute of Standards and Technology Standard Reference Material 1643b Trace Elements in Water, was also analysed as a quality control material during the measurement of the total amount of Cr.Table 4 Results for the determination of CrW3-W in natural waters. Results given are mean k standard deviation ( n = 3) Foundng ml-I Total Cr Sample CrV1 Cr"' Sea-water 0.10 f 0.01 0.49 f 0.04 0.54 k 0.03 Well water 0.13 2 0.05 0.11 _t 0.02 0.25 _+ 0.02 Tap water 0.14 k 0.04 0.20 k 0.05 0.33 k 0.05 The resulting value (18.1 _t 1.5 ng ml-1, mean of three sample analyses f standard deviation) was in good agreement with the certified value (18.6 ng ml-1). Limit of Detection Blank values were measured by using tap water as the metal-free solution after coprecipitation of total Cr. Analysis of five samples of the blank gave an average value of 0.261 k 0.006 ng ml-1. Employing the usual convention, that the detection limit is 4.7 times the square root of the background, or 4.7 times the standard deviation, gave a limit of detection of 0.03 ng ml-1.Mechanism The variation of the amount of CrVI with pH might be associated with the equilibrium between Cr042- and Cr2O72-. At acidic pH, Cr042- is the major species and it is this species that is coprecipitated, while Cr2O72- is not precipitated to any great extent. The precipitation of CrI" at pH 9 cannot be in the form of Cr(PDC)3. It is possible that the insoluble hydrated CrVI oxide is coprecipitated with Pb(PDC)2. It is known31 that Cr"I exists at acidic pH as the hexaaqua ion [Cr(H20)#+, which has a pK of 4. At higher pH the hydroxide ion [Cr(H20),(OH)]2+ is formed, which can give soluble dimers and polymers. At even higher pH values, dark green gels are formed which are coprecipitated with Pb(PDC)2.The authors thank the National Science Council of the Republic of China for financial support of this work. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 References Bowen, H. J., Trace Elements in Biochemistry, Academic Press, New York, 1966. Pankon, J. F., and Janauer, G. E . , Anal. Chim. Acta, 1974,69, 97. Valkovic, V., Trace Element Analysis, Taylor and Francis, London, 1975, p. 107. Schwartz, K . , and Mertz, W., Arch. Biochem. Biophys., 1957, 72, 515. Schwartz, K . , and Mertz, W., Arch. Biochem. Biophys., 1959, 85,292. Mertz, W., Phys. Rev., 1969, 49, 163. Elderfield, H., Earth Planet. Sci. Lett., 1970, 9, 10. Florence, T. M., and Batley, G. E., CRC Crit. Rev. Anal. Chem., 1980,9, 219.Arrhenius, G., and Bonatti, E., Prog. Oceanogr., 1965, 3, 7. Chuecas, L., and Riley, J. P., Anal. Chim. Acta, 1966,35,240. Fukai, R., Nature (London), 1967, 213, 901. Filbert, T. R., and Clay, A. M., Anal. Chim. Acta, 1973, 67, 289. Hiro, R., Owa, T., Takaoka, M., Tanaka, T., and Kawahara, A., Bunseki Kagaku, 1976, 25, 122. De Jong, G. J . , and Brinkman, U. A. Th., Anal. Chim. Acta, 1978, 98,243. Methods for Chemical Analysis of Water and Wastes, US Environmental Protection Agency, Cincinnati, OH, 1979. Sturgeon, R. E., Berman, S. S., Desaulniers, A,, and Russel, D. S . , Talanta, 1980, 27, 85.38 ANALYST, JANUARY 1991, VOL. 116 17 18 19 20 21 22 23 24 25 Isozoki, A., Kumazai, K., and Utsume, S., Anal. Chim. Acta, 1983, 153, 15. Naranjit, D., Thomassen, Y., and Van Loon, J. C., Anal. Chim. Acta, 1979, 110, 307. Leyden, D. E., Channell, R. E., and Blount, C. W., Anal. Chem., 1972, 44, 607. Miyazaki, A., and Barnes, R. M., Anal. Chem., 1981,53, 364. Wai, C. H.. Tsay, L. H., and Yu, J. C., Mikrochim. Acta, Part 11, 1987, 73. Subramanian, K. S., Anal. Chem., 1988, 60, 11. Lavi, N., and Alfassi, Z. B., J. Radioanal. Nucl. Chem, 1988, 126, 361. Lavi, N., Mantel, M., and Alfassi, Z. B., Analyst, 1988, 113, 1855. Lavi, N., and Alfassi, Z. B., J. Radioanal. Nucl. Chem., 1989, 130,71. 26 Lavi, N., and Alfassi, Z. B., Analyst, 1990, 115, 817. 27 Lan, C. R., Sun, Y. C., Chao, J. H., Chung, C., Yang, M. H., Lavi, N., and Alfassi, Z. B., Radiochim. Acta, in the press. 28 Nakayama, E., Kuwamoto, T., Tokoru, H., and Fujingawa, T., Anal. Chim. Acta, 1981, 131, 247. 29 Pik, A. J., Eckert, J. M., and Williams, K. L., Anal. Chim. Acta. 1981, 124, 351. 30 Collins, K. E., Collins, C. H., Yang, M. H., Ke, C. N., Lo, J. M., and Yeh, S. J., J. Radioanal. Chem., 1972, 10, 197. 31 Cotton, F. A., and Wilkinson, G., Advanced Inorganic Chem- istry, Wiley, New York, 4th edn., 1980, p. 727. Paper 0102883E Received June 26th, 1990 Accepted September 13th, I990

ISSN:0003-2654    

DOI:10.1039/AN9911600035

出版商:RSC    

年代:1991

数据来源: RSC

摘要:

ANALYST, JANUARY 1991, VOL. 116 39 Ion-selective Electrodes in Organic Analysis: Determination of Alkyl Halides via ln Situ Generation of S-Alkylisothiouronium Salts Wing Hong Chan,* Albert Wai Ming Lee* and Yu Man Cheung Department of Chemistry, Hong Kong Baptist College, 224 Waterloo Road, Kowloon, Hong Kong Low relative molecular mass alkyl halides, after in situ derivatization to the corresponding S-alkyl- isothiouronium salts in the presence of an excess of thiourea, were determined with a poly(viny1 chloride) membrane S-alkylisothiouronium-selective electrode based on S-butylisothiouronium tetraphenylborate. This membrane electrode exhibited Nernstian response in the range 1.0 x 10-1-1.6 x 10-4 mol dm-3with an average cationic (positive) slope of 58.8 mV per concentration decade at 25 "C.The electrode had a reasonably wide working pH range (6.5-8.5), fast dynamic response time (30 s-2 min), stable response for at least 2 months and high selectivity for the S-alkylisothiouronium ion in the presence of many inorganic and organic ions. The electrode functioned satisfactorily for the determination of primary and secondary alkyl halides, excluding alkyl fluorides. Keywords: Alkyl halide determination; ion-selective electrode; organic analysis; in situ S-alkyl- is0 th io u ro n iu m salt genera ti on Organohalides are essential industrial chemicals. They are important intermediates in many chemical reactions and are used extensively as solvents. Other significant uses of this class of compound are as anaesthetics, refrigerants, and grain and fruit fumigants.1 There is a wide range of published methods for the determination of organohalides. The basic approach involves the decomposition of an organohalide sample to halide ions. After decomposition, the liberated halide ions are determined by using a halide ion-selective electrode,2 grav- imetry or visual titrimetry.3 For compounds containing a tightly bound halogen atom such as alkyl halides, the oxygen flask combustion technique is used.3 All of these methods require either a fairly tedious sample preparation procedure or a large sample size. In order to increase the use of ion-selective electrodes (ISEs) in organic analysis, a new strategy has been proposed for converting a covalent organic compound into a water- soluble ionic derivative which is amenable to potential measurement.Many important organic functional groups such as esters, alcohols, aldehydes and amines, after derivatization to ionic species, can be subjected to selective electrode measurement.4 However, a general method for the determi- nation of alkyl halides using an ISE has not yet been developed. The aim of this investigation was to devclop a poly(viny1 chloride) (PVC) membrane S-alkylisothiouronium- selective electrode for the determination of alkyl halides via in situ generation of the corresponding S-alkylisothiouronium salts in 95% ethanol by means of a bimolecular nucleophilic substitution reaction,s viz., + S NH2 X- II 95 % II RX + H2N-C-NH2 - R-S-C-NH? ethanol where X is Br or I. For the less reactive alkyl halides, such as primary alkyl chlorides and secondary alkyl bromides, it was first necessary to convert these compounds into the corresponding alkyl iodides by refluxing overnight with an excess of sodium iodide in 95% ethanol.h The crude iodo compound, without purifica- tion, was then converted into the S-alkylisothiouronium salt by treatment with an excess of thiourea.* To whom correspondence should be addressed. Experimental Apparatus Potentiometric measurements were made at a constant temperature in the range 20-25 "C with an Orion digital pH/ion meter (Model SA720). A platinized platinum elec- trode (Model 3401) from Yellow Springs Instruments was used as an internal reference electrode. A saturated calomel electrode (SCE) from Orion (Model 9006) was used as an external reference electrode. For pH measurements, a Sen- sonex combined pH electrode (Model 5200C) was used. The nuclear magnetic resonance (NMR) spectra were recorded with a Jeol NMR spectrometer (60 MHz) (Model PMX 60SI).Reagents All solutions were prepared with analytical-reagent grade reagents in distilled, de-ionized water unless stated other- wise. The organic solvents and reagents were also of ana- lytical-reagent grade. Tetrahydrofuran (THF) was distilled over sodium under nitrogen before being used. High relative molecular mass PVC was obtained from Aldrich. Sodium tetraphenylborate and bis(2-ethylhexyl) phthalate (DEHP) were obtained from Fluka. The Britton-Robinson buffer of pH 7.5 consisted of 57.5 ml of 0.2 mol dm-3 sodium hydroxide and 100 ml of a stock solution containing 0.04 mol dm-3 acetic acid, 0.04 mol dm-3 orthophosphoric acid and 0.04 mol dm-3 boric acid.S-Butylisothiouronium Tetraphenylborate-PVC Matrix Membrane Electrode Preparation of the sensing material S-Butylisothiouronium bromide (0.01 mol) was dissolved in 20 ml of distilled, de-ionized water. An equimolar amount of sodium tetraphenylborate was dissolved in another 20 ml of distilled, de-ionized water. The solutions were mixed and stirred for 10 min. The white precipitate of S-butyl- isothiouronium tetraphenylborate was filtered off by suction filtration, washed with distilled, de-ionized water, air-dried for 24 h and ground to a fine powder. Preparation of the sensing membrane A mixture of S-butylisothiouronium tetraphenylborate (0.04 g), PVC powder (0.26 g) and plasticizer (DEHP) (0.10 g) was dissolved in 20 ml of distilled THF.The solution was poured into a 75 mm i.d. Petri dish and covered with a piece of40 Internal reference electrode (Pt-Pt) ANALYST, JANUARY 1991, VOL. 116 Internal PVC matrix reference membrane with solution, S-butylisothio- 0.01 mol dm-3 uronium SBB at pH 7.5 tetraphenylborate filter-paper. After all the THF had evaporated, an S-butyliso- thiouronium tetraphenylborate-PVC matrix membrane sheet was obtained. Standard solution at pH 7.5 Assembly of the electrode A portion of the sensor membrane (diameter, 1 cm) was cut and fitted into a screw-cap adaptor with an O-ring placed above the membrane as the electrode body.7 A solution of 1.00 x 10-2 mol dm-3 S-butylisothiouronium bromide at pH 7.5 was used as the internal reference solution and a platinized platinum electrode was used as an internal reference elec- trode.Saturated KC1 solution Conditioning and storage procedure The assembled electrode was conditioned by soaking it in 1 x 10-2 mol dm-3 S-butylisothiouronium bromide solution at pH 7.5 for 1 d before use; the electrode was also stored in the same solution when not in use. Construction of the ISE system The assembled electrode was immersed in an S-butyl- isothiouronium bromide (SBB) solution and acted as a half-cell. The other half of the cell was formed by inserting an SCE into a saturated KC1 solution. The ISE system was completed by connecting the two half-cells by a KC1 salt- I (ion pair) External reference electrode W E ) In Situ Derivatization of Alkyl Halides In the bimolecular nucleophilic substitution reaction, the order of reactivity follows the sequence RI > RBr > RC1 (R = alkyl).For a given halide, the reactivity decreases in the order primary > secondary > tertiary. Based on the difference in reactivity, alkyl halides can be grouped into two classes in the in situ derivatization reaction. Class 1. Halides with higher reactivity, i.e., alkyl iodides and The alkyl halide (about 0.5 g) was weighed accurately in a 100 ml round-bottomed flask and 1.2 equiv. of thiourea were added. The mixture was dissolved in 30 ml of 95% ethanol and the solution was refluxed for a certain period of time (90-150 min). After refluxing, the solvent was removed under reduced pressure to leave a colourless oily liquid, which was dissolved in a pH 7.5 buffer solution which consisted of boric acid, acetic acid, orthophosphoric acid and sodium hydroxide.The resulting solution was diluted to the mark with the buffer in a 100 ml calibrated flask and then potential measurements were performed. primary alkyl bromides Class 2. Halides with lower reactivity, i .e., alkyl chlorides and The alkyl halide (about 0.5 g) was accurately weighed into a 100 ml round-bottomed flask and 3 equiv. of sodium iodide were added. The mixture was dissolved in 40 ml of 95% ethanol and the solution was refluxed overnight. Thiourea secondary alkyl bromides (1.5 equiv.) was then added to the cool solution and the mixture was refluxed for 150 min. After removal of the solvent in vacuo, the oily residue was dissolved in pH 7.5 buffer and diluted to the mark with the same buffer in a 100 ml calibrated flask.The solution was then ready for potential measurement. Calibration The butyl bromide sample, after the aforementioned in situ derivatization, was used to prepare a series of standard solutions in the concentration range 1 X 10-1-1 x 10-5 mol dm-3. Aliquots (80 ml) of the standard solutions were used for the ISE measurements. The potentials of the stirred solutions were recorded when they became stable and were plotted as a function of the logarithm of the butyl bromide concentration. The graph was used for subsequent determina- tion of alkyl halides. Results and Discussion Nature and Composition of the Membrane S-Butylisothiouronium bromide reacts readily with sodium tetraphenylborate to form a stable, crystalline 1 : 1 ion-pair complex whose composition was unambiguously verified by proton NMR spectroscopy.The integration ratio and the multiplicity of the various proton signals of the NMR spectrum agree well with the proposed structure (Fig. 1). The response characteristics of the PVC membrane doped with various amounts of S-butylisothiouronium tetraphenylborate were systematically investigated. The suitability and sensitivity of the membrane, based on both the slope of the calibration graph and the limit of detection, were studied. The calibra- tions were performed in Britton-Robinson buffer (pH 7 . 9 , in which all of the S-butylisothiouronium ions are in the univalent form.As shown in Table 1, the response was dependent on the proportion of active compound contained in the membrane formulation. For optimum performance, 9-12% m/m of the sensor in the membrane formulation was required. When less carrier was used, the response deviated considerably from the Nernst equation. In the fabrication of the PVC sensing membrane, about 25% m/m of plasticizer (DEHP) was added to improve the plasticity of the membrane. In Situ Derivatization Reaction In the presence of an excess of thiourea, the water-soluble ionic S-alkylisothiouronium salts can be prepared quan- titatively from alkyl halides by heating under reflux. For compounds containing a tightly bound halogen atom, the alkyl halides are first converted into the corresponding alkyl iodides by refluxing overnight with 3 equiv.of sodium iodide in 95% ethanol. For quantitative work, it is essential that the extent of S-alkylisothiouronium salt formation is reproducible under a given set of conditions. In order to ensure this and to determine the time required for quantitative conversion of the alkyl halide into the S-alkylisothiouronium salt, the reaction was monitored by potentiometric measurement using the selective electrode. The measured potential readings were plotted as a function of reaction time (Fig. 2). It was found that quantitative (or constant) conversion of butyl bromide into S-butylisothiouronium bromide was obtained when the 6 2.50(s) + / 6 1.38(m) 6 3.15(t) NH2 \ \ I/ C H3-C H 2-C H2-C H 2-S-C-N H 2 B-( CeH 5 / I \ 6 0.89(t) 6 1.58(m) b 2.50(s) 6 6.7&7.18(m) Fig.1 Assignment of the proton NMR data for the sensor (in CDC13)ANALYST, JANUARY 1991, VOL. 116 41 Table 1 Response characteristics of a PVC membrane doped with various amounts of S-butylisothiouronium tetraphenylborate (sensor) Total mass Slope/mV Detection Sensor (PVC + DEHP + per concentration limit/ Correlation (% m/m) sensor)/g Sample decade 10-4 mol dm-3 coefficient 2.5 6.6 9.5 12.3 24.8 0.4091 0.4112 0.4075 0.4105 0.4048 1 2 1 2 1 2 1 2 1 2 38.4 37.2 51.5 52.7 58.0 57.7 57.4 57.6 53.7 52.4 3.2 3.5 2.3 2.0 2.2 2.0 2.0 1.6 1.3 1.6 0.9989 0.9985 0.9990 0.9985 0.9986 0.9983 0.9990 0.9988 0.9931 0.9970 Table 2 Conditions for the in situ derivatization of different alkyl halide samples Primary alkyl iodide 1-Iodobutane Primary alkyl bromide 1-Bromobutane Primary alkyl chloride 1-Chlorobutane Secondary alkyl bromide 2-Bromobutane 1 -Bromopropane Reflux with excess of thiourea in 30 ml of 95% ethanol for 90 min Reflux with excess of thiourea in 30 ml of 95% ethanol for 150 min Reflux with 3 equiv.of NaI in 40 ml of 95% ethanol for 24 h, then reflux with excess of thiourea for 150 min &I -80 0 v) v) I? E ki 2 -100 > -120 I I i 50 100 150 200 250 300 Time/min Fig. 2 Time required for the quantitative conversion of butyl bromide into the S-butylisothiouronium salt (monitored with the ISE) -50 1 w .u 3 ; -100 2 -150 2 B L a -200 / I -Detection limit I -250 -6 -5 - 4 -3 -2 -1 Log of concentration Response of the electrode to the concentration of butyl Fig. 3 bromide in Britton-Robinson buffer of pH 7.5 former was refluxed with an excess of thiourea in 95% ethanol for 150 min.Similarly, the reaction conditions required for quantitative conversion of other alkyl halides into the corre- sponding S-alkylisothiouronium salts were determined and these are given in Table 2. The results obtained are consistent with the relative reactivities of alkyl halides in nucleophilic substitution reactions. Characteristics of the Electrodes As shown in Fig. 3, the PVC membrane electrode exhibited an average Nernstian slope of 58.8 mV per concentration decade 0 - 50 Lu u v) 9 2 -100 2 > E -150 3 5 7 9 11 PH Fig. 4 Effect of pH on the potential of the S-butylisothiouronium tetraphenylborate-PVC matrix membrane electrode. A, 5 X B, 5 x 10-3; and C, 5 x 10-4 mol dm-3 butyl bromide over five determinations (standard deviation = 1.17 mV) with good linearity (correlation coefficient = 0.9991) from 1.0 X 10-1 to 1.6 x 10-4 mol dm-3 butyl bromide.Both the slope of the calibration graph and the correlation coefficient demon- strate the suitability and sensitivity of this membrane for the determination of alkyl halides. Effect of pH The influence of pH on the response of the S-butylisothiouronium tetraphenylborate-PVC matrix membrane electrode was evaluated by performing the potential measurements on the derivatization product of butyl bromide at different pH values. The electrode showed stable and constant readings (+1 mV) in the pH range 6.5-8.5 for various concentrations of butyl bromide (Fig. 4). The S-butylisothiouronium ion can be protonated in a strongly acidic medium to give a divalent cation, whereas it is readily hydrolysed to butanethiol in a strongly alkaline medium +NH2 +NH2 11 + H+ 11 high C4HgS-C-NH3 C4HgS-C-NH2 0 PH C4HgSH + NH2-C-NH2 I142 ANALYST, JANUARY 1991.VOL. 116 Therefore, subsequent potential measurements were made at pH 7.5, at which nearly all of the S-butylisothiouronium ions are in the univalent form. Response Time and Stability of the Electrode The S-butylisothiouronium tetraphenylborane-PVC matrix membrane electrode has a rapid response time. The response time of the standard solutions was recorded in increased order of their concentration. The results indicated that the average dynamic response time was 30 s and 2 min for a concentrated (>1 X 10-3 rnol dm-3) and dilute (<1 X 10-3 rnol dm-3) solution, respectively (Fig.5 ) . On the other hand, ageing of the membrane was not a serious problem. The performance of the electrode in terms of linearity and Nernstian response was reproducible over a period of 2 months, after repeated measurements. Effect of Interfering Ions The response of the electrode to S-alkylisothiouronium ion in the presence of various foreign ions was examined. The 8 -50 B In- C r -100 > -200 0 100 200 300 Ti me/s Fig. 5 Response time of the membrane electrode for different concentrations of butyl bromide in Britton-Robinson buffer of pH 7.5 (stable reading indicated by the arrow). A, 1 X 10-1; B, 1 X 10-2; C, 1 X 10-3; and D, 1 x rnol dm-3 butyl bromide Table 3 Selectivity coefficients ( k r ; ) for the S-butylisothiouronium tetraphenylborate-PVC matrix membrane electrode.Concentration of each foreign ion, 1.0 x 10-2 mol dm-3 kpot Interfering compound (B) S.B Thiourea O* Urea 0* Tetrabutylammonium hydrogen sulphate 0.5 Ammonium chloride 0" * Identical calibration graphs were obtained both in the presence and absence of the foreign ion. potential given by solutions each containing 1 x 10-2 rnol dm-3 of the foreign compound and various S-butylisothio- uronium concentrations in the range 1 x 10-1-1 x 10-5 rnol dm-3 was measured. The selectivity coefficients were calculated by using the fixed interference method.* The results obtained (Table 3) demonstrated that no significant effect was caused by organic compounds such as thiourea and urea, and by inorganic ions such as ammonium and chloride.As an excess of thiourea was used in the derivatization of the alkyl halides, it was fortuitous that thiourea did not interfere with the determination. The tetrabutylammonium ion interfered only when present at concentration levels at least several times greater than that of the S-butylisothiouronium ion. In addition, the calibrations were carried out in Britton- Robinson buffer solution which consisted of boric acid, acetic acid, orthophosphoric acid and sodium hydroxide; none of these inorganic components of the buffer solution caused any interference. Determination of Alkyl Halides At the onset of this investigation, it was expected that the sensing material of the electrode could be used to detect both primary and secondary aliphatic alkyl halides.S- Alkyliso- thiouronium solutions in the concentration range 1 x 10-1- 5 X 10-4 rnol dm-3 were prepared from the corresponding alkyl halides and determined by using the S-butyliso- thiouronium tetraphenylborate-PVC matrix membrane elec- trode. The potentials given by these solutions were compared with the calibration graph prepared from the butyl bromide derivatization product in order to assess the accuracy and reproducibility of the method. The results obtained for five samples (Table 4), each analysed in triplicate, showed that l-iodobutane and l-bromobutane could be quantitatively converted into the S-butylisothiouronium salts and deter- mined with the membrane electrode. The average recovery was 98.1% with a mean standard deviation of 0.88%.The lower recovery of l-bromopropane (Table 4), which is a reactive alkyl halide, may be due to its greater volatility. For the less reactive alkyl halides, l-chlorobutane, and 2-bromo- butane, the two-step derivatization reaction is more prone to side-reactions, such as an elimination reaction. The recovery of these two halides was 86.6 and 61.3%, respectively. However, the excellent precision observed for this electrode method rendered the determination of these less reactive halides equally feasible. The determination of primary chloro- alkanes and secondary bromoalkanes, using the same calibra- tion graph, adjusting the amount of the less reactive halides found with the electrode by a factor of 0.87 and 0.61, respectively, will give the equivalent amount of the halide in the sample.Moreover, the absolute potentials recorded by the electrode for various alkyl halides at the same concentration level were very similar. Therefore, the electrode can be used Table 4 Determination of alkyl halides using the S-butylisothiouronium tetraphenylborate-PVC matrix membrane electrode Sample Trial 1 -1odobutane l-Bromobutane 3 2 3 l-Chlorobutane 1 2 3 2-Bromobutane 1 2 3 l-Bromopropane 1 Mass used1 g 0.9228 0.3809 0.3123 0.6906 0.5571 0.2503 0.1682 0.5369 0.5893 0.4317 0.1079 0.4662 0.6063 0.6790 0.1516 Mass measured by ISEIg 0.8933 0.3790 0.3067 0.6726 0.5460 0.2473 0.1600 0.5235 0.5592 0.3708 0.0940 0.4042 0.3771 0.4094 0.0934 Standard Recovery Mean (Yo) (Yo) deviation (%) 96.8 99.5 98.2 1.35 98.2 97.4 98.0 98.1 0.70 98.8 95.1 97.5 95.8 1.45 94.9 85.9 87.1 86.6 0.61 86.7 62.2 60.3 61.3 0.97 61.6ANALYST, JANUARY 1991, VOL.116 43 either to determine the concentration of an individual alkyl halide or to measure the total concentration of a mixture of alkyl halides. Conclusion An indirect ISE system for the determination of alkyl halides has been described. The in situ generation of the ionic S-alkylisothiouronium salt from the corresponding covalent alkyl halide in the presence of an excess of thiourea is the key factor in the viability of the method. This highly selective electrochemical method has been shown to be applicable to the determination of low relative molecular mass primary and secondary alkyl halides, excluding alkyl fluorides. References 1 Gessner, G. N., The Condensed Chemical Dictionary, Van Nostrand Reinhold, New York, 8th edn., 1971, p. 359. 2 Ma, T. S., and Hassan, S. S. M., Organic Analysis Using Ion Selective Electrodes, Academic Press, London, 1982, vol. 2, p. 14. 3 Ma, T. S., and Robert, C. R., Modern Organic Elemental Analysis, Marcel Dekker, New York, 1979, pp. 1.58-206. 4 Chan, W. H., Lee, A. W. M., and Chan, L. K., Analyst, 1990, 115,201, and references cited therein. .5 Vogel, A. I., Qualitative Organic Analysis, Longman, London, 2nd edn., 1966, p. 98. 6 Fieser, L. F., and Fieser, M . , Reagents for Organic Synthesis, Wiley, New York, 1967, vol. 1, p. 1087. 7 Chan, W. H., Wong, M. S., and Yip, C. W., J . Chem., Educ., 1986, 63, 91.5. 8 Guilbault, G. G., Ion-Sel. Electrode Rev., 1979, 1, 139. Paper 01031 081 Received July 1 Oth, 1990 Accepted August 17th, I990

ISSN:0003-2654    

DOI:10.1039/AN9911600039

出版商:RSC    

年代:1991

数据来源: RSC

摘要:

ANALYST, JANUARY 1991, VOL. 116 45 Masking of Zirconium(1v) in the Determination of Fluoride With an Ion-selective Electrode: Application to Zirconium(1v) Fluoride-based Glasses Akio Yuchi, Jun-ichi Baba, Hiroko Wada and Genkichi Nakagawa Department of Applied Chemistry, Nagoya Institute of Technology, Gokiso, Showa, Nagoya 466, Japan The performance of six chelating reagents [ethylenediamine-N,N,N’,N’-tetraacetic acid (EDTA); trans-1,2- cy c I o h exa n ed i a m i n e- N, N, N ’, N ’-tetra acetic N ’- ( 2- h y d r ox yet h y I ) et h y I e n ed i a m i n e- N, N, N ’- triacetic acid (HEDTA); trie”yIen“etraamine-N,N,N’,N’~N’’’,N’’’-hexaacetic acid (TTHA); diethylenetriamine- N,N,N’,N”,N’ipentaacetic acid (DTPA); and citrate] has been studied for masking zirconium(1v) in the determination of fluoride with an ion-selective electrode.Citrate was not suitable because it produced a prolonged electrode response. Of the aminopolycarboxylates, DTPA has a much greater masking ability than the others. Using DTPA at pH 5-6, fluoride was successfully determined at a concentration of 1 X 10-5 rnol dm-3 in the presence of up to 4 x 10-6 rnol dm-3 zirconium(1v). The proposed method was applied to the analysis of a number of zirconium(1v) fluoride compounds and ZrF4-based glasses after fusion with sodium carbonate. Keywords: Fluoride determination; ion-selective electrode; fluoride glass; zirconium masking; dieth ylenetriamine-N, N, N ’ , N”, N”-pentaacetic acid a c i d ( C DTA) ; In the last decade, heavy metal fluoride glasses, which gener- ally contain zirconium(1v) , hafnium(1v) or thorium(1v) , have attracted much attention in relation to their potential use in long-distance fibre optics.192 In accordance with progress in this area, the determination of fluoride in the presence of these metal ions has increasingly been required; e.g. , chemical durability testing is essential to assess the utility of each fluoride glass, as the transparency of the glass in the middle infrared region (0.2-8 pm) easily deteriorates as a result of attack from environmental water. To monitor the dissolution rate of a fluoride glass in an aqueous solution, a recent review3 recommends following the appearance of the dissolution products over a period of time, by analysis of the soaking solution, instead of following the loss in the mass of the glass.Metal components have been determined by various spectro- scopic methods, and fluoride by potentiometry with a fluoride ion-selective electrode.&g Zirconium(1v) , hafnium(1v) or thorium(rv) contained in these glasses has an extremely high affinity for fluoridel”15 and thus is expected to interfere seriously with the determination of fluoride. To eliminate the interference, commercially available TISAB (total ionic strength adjustment buffer solution), which contains citrate or truns-l,2-cyclohexanediamine-N, N , N’, ”-tetraacetic acid (CDTA) as a masking reagent for metal ions,16,17 has been used. These reagents are effective for masking common metal ions such as Al”, Fe3+ and Ca2+ but are not always suitable for metal ions in higher oxidation states.For example, Simmons and Simmons8 have observed that all the fluoride ions were not in the free form when CDTA was used as a masking reagent for zirconium(1v). No further work appears to have been undertaken since that report. In previous papers,l8-20 Yuchi and co-workers have studied the reaction of the fluoro complexes of trivalent metal ions with various masking reagents and found that mixed ligand complexes were generally present with a masking reagent and fluoride. Reagents forming less stable mixed ligand complexes are more efficient for the masking of a metal ion in the determination of fluoride by potentiometry using a fluoride ion-selective electrode. It has been found that ethylene- diamine-N, N , N‘, ”-tetraacetic acid (EDTA) complexes of tetravalent metal ions also form stable mixed ligand com- plexes with fluoride.21 In the present paper, six chelating reagents [EDTA, CDTA, N’-(2-hydroxyethyl)ethylenediamine-N, N , N‘-tri- acetic acid (HEDTA), diethylenetriamine-N,N,N’,N”,N’- pentaacetic acid (DTPA), triethylenetetraamine-N, N , N‘, - N’, N”, N”-hexaacetic acid (TTHA) and citrate] have been examined for the determination of fluoride in the presence of zirconium(1v).Using DTPA as a masking reagent, fluoride was successfully determined in some zirconium fluoride compounds and ZrF4-based glasses. As fluoride in these materials is prone to be replaced by oxygen-containing species such as OH- and 0 2 - ions, these data will be complementary to the results for the determination of oxygen in fluoride glasses by charged particle activation analysis.22 Experimental Reagents All the reagents used were of analytical-reagent grade.Potassium nitrate was recrystallized twice. Carbonate-free potassium hydroxide solution was prepared as described elsewhere .23 Potassium fluoride was dried in a platinum crucible for 24 h at 110°C. Fluoride solutions were stored in polyethylene containers. Zirconium(1v) stock solution was prepared by dissolving zirconium(1v) oxide nitrate, ZTO(NO~)~, in a 4 mol dm-3 nitric acid solution, which prevents the formation of polymeric hydrolysed species .24 The concentration of zirconium(1v) was determined by titration with EDTA in 1 rnol dm-3 HN03 at 90°C using Xylenol Orange. The zirconium tetrafluoride (Morita Kagaku Kogyo) and potassium hexafluorozirconate (Kanto Chemicals) used as samples were of technical grade.Measurement The equipment used was the same as that described previ- ously. 1 ~ 1 All the potentiometric measurements were perfor- med at 25 “C and at an ionic strength of 0.1 rnol dm-3 KN03. The effects of pH and the concentrations of fluoride and zirconium(1v) on the electrode response were studied by utilizing various masking reagents; a series of solutions containing 2.5 x 10-6-2.5 x 10-4 rnol dm-3 zirconium(Iv), 1.25 X 10-5-1.25 x 10-3 mol dm-3 fluoride ion and 5 X 1 0 - h l x 10-1 rnol dm-3 masking reagent were titrated with 0.1 rnol dm-3 potassium hydroxide. After each addition the46 ANALYST, JANUARY 1991, VOL. 116 100 80 - u L 0 cc 60 40 3 4 5 6 7 A n A " 0 0 0 0 0 o u A P,,,, PH 3 4 5 6 7 3 4 5 6 7 Fig. 1 R or R'(%) versus pH.Masking reagent: (a) none; ( 6 ) EDTA; and 1.25 X R: 0, A and 0; and R': 0. cF:cZr = 5. cL: cZr = and 0, 0, 1.25 X pF and pH were measured with a fluoride ion-selective electrode and a fluoride-resistant glass electrode. The recovery of fluoride, R = [F-]/cF (cF = total fluoride concentration), was calculated from the pF values. The recovery, taking the protonation of fluoride into account, R', was also calculated by using the relevant constants and both pF and pH values's ( K is the formation constant in each instance). R' = ([F-] + [HF] + ~[HF~-])/CF = ([F-] -k KHF[H+][F-] + 2KHF2KHF[H+][F-]2)/CF Recommended Procedure A 0.05 g portion of the sample to be analysed is placed in a platinum crucible and covered with 1 g of Na2C03.The crucible is heated at 900 "C for 15 min. The cooled melt is digested with 60-70 cm3 of 1 X rnol dm-3 DTPA. After dissolution, 20 cm3 of 1 rnol dm-3 HN03 are added, and the solution is diluted to 250 cm3. After a further 500-fold dilution the solution is analysed for fluoride. Results and Discussion Zirconium(1v) seriously interferes with the determination of fluoride as shown in Fig. l(a). Although R increases with an increase in pH or with dilution of the sample, it does not reach 100% in the pH range suited to the use of the fluoride ion-selective electrodes. Addition of a masking reagent generally improves the recovery. The effects of pH and the concentrations of fluoride, zirconium(1v) and masking reagent were studied for each system (Figs.1-3). Effects of pH and Fluoride Concentration In the presence of a masking reagent, R also increases with pH, steeply in an acidic medium and gradually in a neutral medium [Fig. l(b) and (c)]. The increase in R at pH <5 is due to the deprotonation of HF. As the formation of HF is negligible above pH 5 , a plateau or an inflection point appears in the graph of R versus pH depending on the concentrations of fluoride and zirconium(1v). At higher concentrations of fluoride and zirconium(rv), the formation of ZrF, is mainly responsible for the interference in a slightly acidic medium, whereas at lower concentrations of fluoride and zirconium(1v) the interference is caused by ZrLF. [As the fully deprotonated ligands (L) used in this study have different electric charges, the net charges on the zirconium complexes are different to each other and therefore have been omitted for simplicity.] ZrF, + L' ZrLF + ( n - 1)F- (1) ZrLF ZrL + F- (2) (c) DTPA.cF/mol dm-3: 0, 1.25 x 10-3; A , - 3 : 80 1 ," 6ot; I , , I 3 4 5 5 7 PH Fig. 2 Comparison of aminopolycarboxylates as masking reagents for ZrIV. Masking reagent: 0, DTPA; 0, TTHA; A, HEDTA; 0, CDTA; and ., EDTA. cF = 1.25 X 10-5 rnol dm-3. cZr = 2.5 x 10-6 rnol dm-3. cL = 5 x 10-6 mol dm-3 Both equilibria, particularly that given by equation (l), shift to the right by simple dilution of the sample. Hence, the sample should be diluted as much as possible within the dynamic range of the fluoride ion-selective electrodes. Such an effect has also been utilized to eliminate interference from alumi- nium .20,25 The increase in R with pH found for the EDTA system [Fig.l(b)] at pH >6 is ascribed to the replacement of fluoride in the mixed ligand complexes by hydroxyl ions to form ZrL(0H) or Zr2b(OH)2. ZrLF + OH- e ZrL(0H) + F- lr 1/2[ Zr2L2 ( OH121 (3) The R versus pH curves obtained agree well with those calculated using the relevant stability constants.21J6-28 Comparison of Masking Reagents Fig. 2 shows the masking abilities of aminopolycarboxylates for 0.25 x 10-5 rnol dm-3 zirconium(1v) at a total fluoride concentration of 1.25 x 10-5 mol dm-3. Satisfactory recovery was obtained only with DTPA at pH >5. As R exceeds 80% for a sample with cF : cZr = 5 in the neutral pH region (Fig. 2), the average number of fluoride ions bound to zirconium is less than 1.A higher concentration of an aminopolycarboxylate did not give a higher recovery. Hence, the equilibrium [equation (l)] is completely shifted to the right. In such solutions, the following relationships hold: (4) CF = [ZrLF] + [F-] cZr = [ZrL] + [ZrLF] ( 5 ) KFZrLF = [ZrLF]/[ZrL][F-] (6)ANALYST, JANUARY 1991, VOL. 116 47 100 I s 80 Q 60 I 0 0 I . 1 I I 1 3 4 5 6 7 PH Fig. 3 CF = 1.25 X 0, 0.1; a, 0.01; and 0, 0.001 Effect of citrate concentration on the recovery of fluoride. rnol dm-3. cZr = 2.5 x 10-6 mol dm-3. cLlmol dm-3: 0 ' 1 I 1 I 4 5 6 -Log (cz,/mol dm-3) Fig. 4 Calculated recovery of fluoride versus -log cZr when masking zirconium with A, DTPA and B, EDTA. CF = 1.25 x lo-' mol dm-3 For a solution containing known concentrations of fluoride and zirconium(iv), the recovery of fluoride can be calculated from these equations.The formation constant of the mixed ligand complex, I(FZrLF, for DTPA was found to be 103.80, whereas that for EDTA has been previously reported2I to be 104.52. Recovery of 1.25 x 10-5 rnol dm-3 fluoride in the presence of various concentrations of zirconium(1v) was calculated and is shown in Fig. 4. When DTPA was used as a masking reagent, the tolerable amounts of zirconium(1v) for 99 and 98% recovery of fluoride are 2 X 10-6 and 4 X 10-6 mol dm-3, respectively. These values correspond to 16 and 32% as the molar ratio of zirconium(1v) to total fluoride and are sufficient for the analysis of fluoride glasses, because the molar ratios of these glasses are generally lower than 25%.Using EDTA, on the other hand, the tolerable amounts of zirconium(1v) are 4 X 10-7 and 8 x rnol dm-3 corresponding to only 3 and 6%, respectively. The potentially octadentate ligand, DTPA, may be the correct size to form a stable and coordination-saturated complex with zirconium(1v) similar to bis(nitrilotriacetate)zirconium,zg and the resultant complex has a much lower affinity for fluoride ion. For citrate (Fig. 3), a slight increase in recovery with an increase in the concentration of citrate from 0.001 to 0.1 rnol dm-3 indicates a different reaction schenx. Even 0.1 rnol dm-3 citrate solution, however, has a -nasking ability inferior to DTPA. Moreover, a higher concentration of citrate results in a prolonged response time of the fluoride ion- selective electrodes, which has been pointed out in relation to the masking of alurninium.30,31 Pre-treatment Dissolution of zirconium fluoride compounds is not easy; e.g., 0.004 g of finely powdered ZrF, suspended in 100 cm3 of a 0.5 x 10-3 rnol dm-3 DTPA solution stirred continuously required 7 h at pH 6-7 and 3 h at pH 3 to dissolve.Fusion with sodium carbonate was examined as a general method of pre-treatment for the dissolution of zirconium fluoride com- Table 1 Determination of fluoride in samples containing zirconium Fluoride (%) Sample Measured Calculated ZrF4 43.5,43.6,44.0 45.5 ZB glass* 35.1,35.1,35.2 37.3 K2ZrF6 37.6,37.9,38.3 40.2 ZBLAN glass? 40.7,41.2,41.7 39.1 * ZrF4: BaF2 = 2 : 1. t ZrF4 : BaF2 : LaF3 : AIF3 : NaF = 53 : 20 : 4 : 3 : 20.pounds. After fusion, even a fluoride glass sample could be dissolved in a DTPA solution. As fusion for 15, 30 or 60 min produced the same results, 15 min proved to be sufficient. It was necessary to treat the cooled melt with a DTPA solution before neutralization with nitric acid, in order to avoid prolonging the dissolution time. For samples containing relatively large amounts of zirconium, small amounts of a white precipitate, zirconium hydroxide or hydrated zirconium oxide, were formed during the neutralization of the DTPA solution with nitric acid. As potassium hexafluorozirconate was soluble in water, fluoride could be determined without fusion. When the same sample was pre-treated as described above, a white precipitate was formed but the analytical results were in good agreement with each other; thus the amount of fluoride in the precipitate is negligible. The addition of nitric acid results in a pH of about 6, which gradually increases with time owing to the evolution of C02.Neither the precipitation of zirconium compounds nor the increase in pH interferes with the subsequent determination of fluoride. The solutions thus obtained can be kept for at least 1 week without any deterioration. Determination of Fluoride in Samples Containing Zirconium Fluoride in commercially available zirconium fluoride com- pounds of technical grade and in fluoride glasses was determined using the proposed procedure. Diethylene- triamine-N, N , N ' , A'", N"-pentaacetic acid was effective for masking Ba2+, La3+ and relatively small amounts of AP+.The results in Table 1 show a satisfactory reproducibility. Purities of commercial ZrF, and K2ZrF6 were 96.2 and 94.3%, respectively. The infrared absorption band at 1640 cm-1 suggests the presence of strongly adsorbed water molecules .22 Prior separation of fluoride by conventional steam distil- lation was virtually impossible, although a recent paper32 describes a modified method, which is effective in the presence of zirconium(1v). As demonstrated above, poten- tiometry with a fluoride ion-selective electrode by using DTPA as a masking reagent is a more convenient and less time consuming method for the determination of fluoride in samples containing zirconium. We thank Professor Y. Kawamoto and Asahi Glass Company Ltd. for providing the samples.This study was supported by a Grant-in-Aid for Scientific Research (No. 02640444) from the Ministry of Education, Science and Culture, Japan. References 1 2 3 4 5 Comyns, A. E., ed., Fluoride Glasses, Wiley, New York, 1989. Almeida, R . M., Halide Glasses for Infrared Fiberoptics, NATO AS1 Series E-123, Martinus Nijhoff, Dordrecht, 1985. Moynihan, C. T., and Loehr, S. R., Muter. Sci. Forum, 1988, 32133, 243. Simmons, C. J., Sutter. H . , Simmons, J . H., and Tran, D. C., Muter. Rex Bull., 1982, 17, 1203. Doremus, R . H . , Murphy, D., Bansal, N. P., Lanford, W. A., and Burman, C., J . Muter. Sci., 1985, 20, 4445.48 ANALYST, JANUARY 1991, VOL. 116 6 Tregoat, D., Liepmann, M. J., Fonteneau, G., Lucas, J., and Mackenzie, J. D., J. Non-Cryst. Solids, 1986, 83, 282.7 Ravaine, D., and Perera, G., J. Am. Ceram. SOC., 1986,69,852. 8 Simmons, C. J., and Simmons, J. H., J. Am. Ceram. SOC., 1986, 69, 661. 9 Simmons, C. J., J. Am. Ceram. SOC., 1987, 70, 295. 10 Noren, B., Acta Chem. Scand., 1967, 21, 2457. 11 Noren, B., Acta Chem. Scand., 1969, 23, 379. 12 Noren, B., Acta Chem. Scand., 1967, 21, 2435. 13 Noren, B., Acta Chem. Scand., 1967, 21, 2449. 14 Baumann, E. W., J. Inorg. Nucl. Chem., 1970, 32,3823. 15 Klotz, P., Mukherji, A., Feldberg, S., and Newman, L., Inorg. Chem., 1971, 10, 740. 16 Frant, M. S., and Ross, J . W., Anal. Chem., 1968,40, 1169. 17 Harwood, J. E., Water Res., 1969, 3, 273. 18 Yuchi, A., Hotta, H., Wada, H., and Nakagawa, G., Bull. Chem. SOC. Jpn., 1987, 60, 1379. 19 Yuchi, A., Ueda, K., Wada, H., and Nakagawa, G., Anal. Chim. Acta, 1986, 186, 313. 20 Yuchi, A., Yanai, N., Wada, H., and Nakagawa, G., Analyst, 1988, 113, 1405. 21 Yuchi, A., Ban, T., Wada, H., and Nakagawa, G., Inorg. Chem., 1990, 29, 136. 22 Mitachi, S., and Tick, P. A., Mater. Sci. Forum, 1988, 32/33, 197. 23 Powell, J. E., and Hiller, M. A., J. Chem. Educ., 1957,34,330. 24 Noren, B., Acta Chem. Scand., 1973, 27, 1369. 25 Shiraishi, N., Murata, Y., Nakagawa, G., and Kodama, K., Anal. Lett., 1973, 6 , 893. 26 Intorre, B. I., and Martell, A. E., J. Am. Chem. SOC., 1960,82, 358. 27 Intorre, B. I., and Martell, A. E., Inorg. Chem., 1964, 3, 81. 28 Bottari, E., and Anderegg, G., Helv. Chim. Acta, 1967, 50, 2349. 29 Hoard, J. L., Willstadter, E., and Silverton, J. V., J. Am. Chem. SOC., 1965, 87, 1610. 30 Edmond, C. R., Anal. Chem., 1969,41, 1327. 31 Ingram, B. L., Anal. Chem., 1970,42, 1825. 32 Reig, F. B., Moreno, A. C., Pardillo, M. B., Martinez, V. P., and Adelantado, J. V. G., Mikrochim. Acta, Part 111, 1989,49. Paper 01031 09G Received July 1 Oth, 1990 Accepted September 1 Oth, 1990

ISSN:0003-2654    

DOI:10.1039/AN9911600045

出版商:RSC    

年代:1991

数据来源: RSC

摘要:

ANALYST, JANUARY 1991, VOL. 116 49 Indirect Determination of Chloride by Gas-diffusion Flow With Amperometric Detection SneZana D. Nikolic and Emil B. Milosavljevic Faculty of Chemistry, University of Belgrade, P.O. Box 550, 1 101 Belgrade, Yugoslavia James L. Hendrix and John H. Nelson Departments of Chemistry and Chemical and Metallurgical Engineering, Macka y University of Nevada, Reno, NV 89557, USA Injection School of Mines, A rapid, indirect gas-diffusion flow injection (FI) method with amperometric detection has been developed for the selective and sensitive determination of CI-. The method is based on permanganate oxidation of CI- to chlorine. The chlorine diffuses through the micro-porous membrane and is quantified amperometrically at a platinum working electrode. Calibration graphs were linear up to the maximum concentration of CI- investigated (10 mmol dm-3).The precision of the technique was better than a relative standard deviation of 1% at 2 mmol dm-3 levels and better than 2% at 10 pmol dm-3, with a throughput of 30 samples h-1. At elevated temperatures (50 "C) and higher acidities (5 mol dm-3 H2S04), the detection limit was 0.1 pmol dm-3 (0.7 ng of CI-). The effects of temperature, sample acidity, working potential and interferents on the FI signals were studied. The method was successfully applied to the determination of CI- in natural and tap waters. Keywords : Gas-diffusion flow injection method; ampe rome tric detection; indirect chloride determination The ubiquitous nature of the chloride ion, and its importance, makes the determination of C1- one of the most frequently required analyses. It is not surprising, therefore, that a number of flow injection (FI) methods have been developed for this analyte.One of the first FI publications1 describes the spectrophotometric determination of CI- in brackish waters. Other papers utilizing spectrophotometric detection fol- lowed.2-11 An interesting approach to the determination of C1- is the combination of FI dialysis with spectrophotometric detection.2.12 Martinez-Jimenez et al. *3,14 determined C1- and mixtures of C1- and I - , with use of FI in conjunction with indirect atomic absorption spectrometric detection. The combination of FI and potentiometry for the determination of CI - has also been extensively studied. Various ion-selective electrodes'5-26 and a copper-wire indicator electrode27 were used for this purpose.Zaitsu et aZ.28 combined FI with turbidimetry in order to develop a method for the determina- tion of CI-. The same analyte was determined by FI methods, with use of chemiluminescence29 and condu~tivity~~ detection. A literature search revealed that there are only two FI amperometric methods designed for the determination of CI-. Polta and Johnson31 utilized pulsed amperometric detection to determine C1-, which alters the rate of surface oxide formation at the platinum working electrode. A triple-step potential waveform had to be used, hence the method required an apparatus more sophisticated than that required for single-potential amperometry . Also, the method suffers from various interferents and, as these workers pointed out, it is more suitable for detection in liquid chromatographic analysis where sufficient resolution of species is provided.The FI method, with direct amperometric detection of C1- at a silver working electrode, developed by Frenzel et al. ,32 is also non-selective as any species that forms either insoluble silver salts or stable complexes with Ag+ would necessarily inter- fere. Hence, special calibration methods were required. The present paper describes an approach to the use of amperometric detection for the determination of C1-. In the FI manifold developed, the injected analyte is converted on-line into chlorine, which diffuses from the donor stream through the micro-porous membrane into the acceptor solu- tion.The latter carries chlorine to the flow-through ampero- metric detector, where it is reduced at a platinum working electrode. The cathodic current measured is proportional to the concentration of C1- in the original sample or standard. To the best of our knowledge, there are only two FI publica- tions33.34 that combine gas diffusion with amperometric detection. This is surprising, as the inherent sensitivity of amperometry and selectivity of the gas-diffusion processes render this combination a powerful analytical tool. Experimental Reagents and Materials All the chemicals used were of analytical-reagent grade. The aqueous reagent and standard solutions were stored in polyethylene bottles. De-ionized water was used throughout. A saturated solution of KMn04 served as the oxidizing agent.It was prepared by boiling the saturated solution with subsequent filtering in order to remove MnOz and any excess of KMn04 that might be present. A stock solution of 0.1 mol dm-3 NaCl was prepared from BDH (Poole, Dorset, UK) concentrated volumetric standards, which are certified to have an accuracy within the factor limits of 0.999 and 1.001. Standard C1- solutions, which in most of the experiments were made in 3 mol dm-3 H2S04, were prepared by diluting aliquots of the stock solution to the appropriate volumes. Instrumentation and Apparatus The FI manifold is illustrated in Fig. 1. Two peristaltic pumps were used. One was a Model Mini S-840 (Ismatec, Zurich, Switzerland) and the other was a Model HPB 5400 (Iskra, Kranj, Yugoslavia).The injection valve was a Model 5020 (Rheodyne, Cotati, CA, USA) equipped with a 200 vl sample loop. The gas-diffusion unit, which was obtained from Shenyang Film-Projector Reflector Factory (Shenyang, China), is similar in construction to the Tecator (Hoganas, Sweden) Chemifold V gas-diffusion cell. The membrane used, which was of Teflon, was supplied with the unit. All connections were made with 0.5 mm i.d. Teflon tubing except for the long mixing coil (MC1), which was made from a 0.8 mm i.d. Teflon tube. The flow-through amperometric cell (Dionex, Sunnyvale, CA, USA), described earlier,35 consisted of platinum working and counter electrodes. The reference electrode was an Ag-AgCI (1 mol dm-3 NaCI) electrode and it was separated from the flowing stream by an ion-exchange Nafion mem-50 ANALYST, JANUARY 1991, VOL.116 RF I P 7 7 - U ' Fig. 1 F1 manifold used for the indirect determination of chloride: C, carrier (3 rnol dm-3 H2S04); R, reagent (saturated KMnO,); AS, acceptor solution (0.01 rnol dm-3 H2S04); P, peristaltic pump; PD, pulse damper; I, injection valve; MC1, long mixing coil (2 m x 0.8 mm i.d.); D, diffusion cell; CTB, constant-temperature bath; MC2, short mixing coil (0.3 m x 0.5 mm i.d.); EC, electrochemical flow-through cell; PO, potentiostat; RE, recorder; and W, waste. Flow-rates are given in ml min-1 brane (all electrode potentials are reported versus this reference electrode). The platinum working electrode was polished occasionally with a small amount of toothpaste and a paper tissue.The potential to the flow-through amperometric cell was applied and currents were measured with a Model MA5450 polarograph (Iskra, Kranj, Yugoslavia); the result- ing F1 signals were recorded on a Model 61 Servograph (Radiometer, Copenhagen, Denmark) strip-chart recorder. The measurements were made with both donor and acceptor streams continuously flowing. Temperature regulation was achieved with a constant- temperature bath , type NBE (VEB Prufgerate-Werl, Medingen, Germany). Results and Discussion The rate of oxidation of CI- to chlorine by permanganate is slow. Hence, in order to apply this reaction in FI, steps must be taken to increase the reaction rate. The logical step is to use the saturated KMn04 solution as the oxidant, as has been carried out recently in the non-FI method developed for the determination of C1- by flame infrared emission .36 The effects of several parameters on the performance of the FI system, illustrated in Fig.1, designed for the indirect determination of CI-, were studied. I n order to find a suitable acceptor solution, several potential candidates were tested (H20, NaOH, Na2C03, KN03 and H2S04). It was found that the optimum signal to noise ratio was obtained with 0.01 mol dm-3 H2SO4, and in all the subsequent experiments this medium was used as the acceptor solution. The effect of the applied potential at the working platinum electrode was investigated in the range +0.10 to +0.70 V versus an Ag-AgC1 reference electrode. The hydrodynamic voltammogram for a 2.50 mmol dm-3 sodium chloride standard in 3 mol dm-3 H2SO4 is shown in Fig.2. As can be seen, the optimum potential is +0.30 V. However, if selectivity concerns dictate otherwise, slight variations of the applied potential are possible, bearing in mind that at potentials lower than about +O. 15 V, the background current becomes too high, probably as a result of the onset of oxonium ion reduction. The effect of H2SO4 concentration on the peak currents was studied by injecting the same CI- standard (2.50 mmol dm-3) while varying the concentration of the acid from 1.0 to 5.0 rnol dm-3. The data obtained are shown in Fig. 3. As can be I r 0 0 0 0 0 0 I I I I 1 I 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 PotentialN versus Ag-Ag CI Fig. 2 2.50 mmol dm-3 sodium chloride standard Hydrodynamic voltammogram for a 200 p1 injection of a 2.8 1 2.4 2.0 a 2 5 1.2 z u 0.8 0.4 0 1.0 2.0 3.Q 4.0 5.0 [H2SO$/mol dm-3 Fig.3 Variation of peak current as a function of sulphuric acid concentration. 0, Experimental data points; and -, calculated according to the equation i = exp(-3.50 + 0.86~). For details see text ,4min, Scan - Fig. 4 Response of the amperometric detector to: (a) nine repetitive injections of a 2.00 mmol dm-3 chloride standard; and ( b ) five repetitive injections of a 10.0 pmol dm-3 chloride standard seen, the current increases exponentially with an increase in H2S04 concentration. The simple equation of the type: i = exp(-3.50 + 0 . 8 6 ~ ) (where i is the current in PA, and c is the concentration of &SO4 in rnol dm-3) fits the data very well. The corresponding correlation coefficient was found to be 0.9989.The temperature effects were studied by injecting a 2.50 mmol dm-3 sodium chloride standard made in 3 rnol dm-3 H2S04, while varying the temperature in the interval 30-60 "C. A linear relationship was obtained with an increase in sensitivity of (0.187 -t 0.009) yA "C-1, with a correlation coefficient of 0.9975. This finding is interesting, as the change in temperature not only affects the oxidation rate, but also the solubility of the chlorine formed, the diffusion process and theANALYST, JANUARY 1991, VOL. 116 51 Table 1 Solutions tested for their possible interference* Compound NHjSCN CH3COONa NazHPOl Na2EDTAt NaF NH4N03 Concentratiodmol dm-3 0.01 0.1 0.1 0.01 0.1 0.1 * The response of the amperometric detector to 200 pl injections of the solutions tested could not be distinguished from the background noise.T Na2EDTA = Disodium ethylenediaminetetraacetate. Table 2 Comparison of FI results for the determination of C1- in the presence of Br- and I- [all samples contained 2.50 mmol dm-3 (88.6 pg ml-1) C1-] Sample c1- C1 -/Br- 'r c1 -/I - $ C1- /Br- 8 FI/pg ml- Difference (%)* 88.6 5 0.3 0 96.5 k 0.9 +8.9 88.2 -t 0.7 -0.45 84.3 k 0.3 -4.8 * Compared to a pure C1- (88.6 pg ml-1) standard. t C1- + 1 pg ml-1 of Br-. $ C1- + 1 pg ml-* of I-. Q C1- + 1 pg ml-1 of Br- + 0.5 mmol dm-3 103- (bromine expelled by boiling the solution for 10 min). Table 3 Comparison of C1- determination by FI and argentimetric titration in three water samples Conccntratiordpg ml- ~ Sample Argentimetric FI Label Belgrade tap water 22.0 k 0.1 21.2 k 0.2 Prolom" 4.60 k 0.08 4.50 k 0.1 32.0 Knjaz MiloS" 14.0 2 0.2 12.9 -t 0.2 12.0 * Commercial mineral waters. reduction of chlorine at a platinum working electrode.The last process is by itself a complicated one. Temperature changes affect the working potential (they change the potential of the reference electrode), and also the diffusive and migratory properties of the analyte, etc. As satisfactory sensitivity is achieved even at the lowest temperature studied (30 "C), most of the subsequent experiments were carried out at this temperature. The linearity studies were conducted by injecting in triplicate a total of eight standards between 0.10 and 10 mmol dm-3 made in 3 mol dm-3 H2S04. The linear regression equation for a typical calibration run was: i = (-5.05 k 0.04) x 10-3 + (0.164 k 0.007) x c (i is the peak current in PA, and c is the concentration of C1- in mmol dm-3), with a correlation coefficient of 0.9992 (all the statistics were calculated for a 95% confidence level).The repeatability of the analytical system is illustrated in Fig. 4. For example, the relative standard deviation for a 2.00 mmol dm-3 standard was found to be 0.8% ( n = 9). The detection limit under these experimental conditions (30 "C; 3 mol dm-3 H2SO4), calcu- lated according to the recommended procedure,37 was 5 pmol dm-3 of CI-. At elevated temperatures (50 "C) and higher acidities (standards were made in 5 mol dm-3 H2S04) the detection limit was 0.1 pmol dm-3, which corresponded to 0.71 ng of CI- (the sample loop volume was 200 pl).It has been established previously that PTFE membranes used in the FI gas-diffusion studies are effective barriers for ionic species.38.39 Nevertheless, a number of anions were tested. The concentrations of these species given in Table 1 are the maximum concentrations at which they were tested. As expected, in all these examples the response of the amperometric detector could not be distinguished from the baseline. It has been established previously34 that anions such as NOI-, S032-, C03"-, S2032-, CN- and S2-, which, when acidified, form acidic gases, could potentially interfere when a particular amperometric flow-through cell is used. These anions at sufficiently high concentrations, even if they are not electroactive at the working potential, could interfere in an indirect manner.The Ag-AgC1 reference electrode in the configuration used is separated from the flowing stream by the ion-exchange Nafion membrane. If the buffer capacity of the acceptor solution is too low, there will be a significant change in pH when the acidic gases, formed by on-line acidification in the FI manifold, diffuse through the Teflon membrane and are trapped in the acceptor solution. This pH change alters the potential of the reference electrode assembly, which is probably induced by a shift in the ion-exchange equilibria at the Nafion membrane. However, in the method developed in this work, the standards and samples are acidified off-line as they are made in 3 mol dm-3 H2S04, so that the aforemen- tioned anions would not interfere in the determination of CI- , as they are evolved prior to injection.(Danger! If some of the aforementioned anions are present, the acidification of the samples should be performed with all due precautions as in some instances poisonous gases are formed.) Other potential interferents are the anions that can be oxidized on-line by permanganate to form molecular species. If these species diffuse through the membrane and are reducible at the platinum electrode at the applied potential, they would cause a positive error in the determination of CI-. Likely candidates are Br- and I-. It has been established that Br- interferes, whereas I- does not. Under the experimental conditions used for the determination of C1-, it is probable that Br- is mainly oxidized to bromine, while I- yields higher oxidation states, which form ionic species.Bromide and I- can be present in natural waters up to levels of 1 and 0.1 pg ml-1, respectively. As can be seen from Table 2, a 2.50 mmol dm-3 C1- standard spiked with Br- at 1 pg ml-1 levels increases the signal by 8.9%. Kubala et a1.,36 in their flame infrared emission method for the determination of CI-, utilized the iodate pre-treatment method, which is usually applied to the determination of CI- by classical argentimetric procedures: 103- + 6Br- + 6H+ -+ I- + 3Br2 + 3H20 This pre-treatment produces bromine, which can be boiled out of the solution, and I - , which was shown not to interfere with the FI method for the determination of CI-. The applicability of this pre-treatment method is also illustrated in Table 2, from which it can be seen that iodate treatment decreases the absolute percentage difference from the pure C1- standard by about 50% (see Table 2).In order to illustrate the potential of the indirect FUgas- diffusiodamperometric method for the determination of CI- , three water samples have been analysed. Table 3 compares the values obtained with those given by argentimetric titrations. It is interesting to note that FI results always gave slightly lower values than those of argentimetric titrations. This could be explained by the fact that argentimetric titrations are influ- enced by the presence of anions such as phosphate, which might be present in the water samples analysed. Therefore, considering that the titration method is not error free, the agreement between the results obtained by the two methods is very good.The authors acknowledge the financial support of the United States Bureau of Mines under the Mining and Mineral Resources Institute Generic Center programme (Grant num- ber G1125132-3205, Mineral Industry Waste Treatment and Recovery Generic Center) and the Serbian Republic Research Fund.52 ANALYST, JANUARY 1991, VOL. 116 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 References RgZiEka, J., Stewart, J. W. B., and Zagatto, E. A. , Anal. Chim. Acta, 1976,81, 387. Hansen, E. H., and RgiiEka, J., Anal. Chim. Acta, 1976, 87, 353 * Slanina, J., Bakker, F., Bruyn-Hes, A., and Mols, J. J., Anal. Chim. Acta, 1980, 113, 331. Krug, F.J . , Pessenda, L. C. R., Zagatto, E. A. G., Jacintho, A. O., and Reis, B. F., Anal. Chim. Acta, 1981, 130, 409. Rosner, B., and Schwedt, G., Fresenius Z. Anal. Chem., 1983, 315, 197. Alexander, P. W., and Thalib, A., Anal. Chem., 1983,55,497. van Staden, J. F., Fresenius Z. Anal. Chem., 1985,322, 36. Faizullah, A. T., and Townshend, A., Anal. Chim. Acta, 1986, 179, 233. Whitman, D. A., Christian, G. D., and RfiiiEka, J., Analyst, 1988, 113, 1821. Frenzel, W., Fresenius Z. Anal. Chem., 1988, 329, 668. Eremina, I. D., Shpigun, L. K., and Zolotov, Yu. A., Zh. Anal. Khirn., 1989, 44, 399. Sundqvist, U., Fresenius Z. Anal. Chem., 1988, 329, 688. Martinez-Jimenez, P., Gallego, M., and Valcarcel, M., J . Anal. At. Spectrom., 1987, 2, 211. Martinez-Jimenez, P., Gallego, M., and Valcarcel, M., Anal.Chim. Acta, 1987, 193, 127. Miiller, H., Anal. Chem. Symp. Ser., 1981, 8, 279. Virtanen, R., Anal. Chem. Symp. Ser., 1981, 8,375. Gao, Z., and Lu, M., Huanjing Kexue, 1981, 2, 376. Trojanowicz, M., and Matuszevski, W., Anal. Chim. Acta, 1982, 138, 71. Trojanowicz, M., and Matuszevski, W., Anal. Chim. Acta, 1983, 151, 77. Ilcheva, L., and Cammann, K., Fresenius 2. Anal. Chem., 1985, 322, 232. 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 van Staden, J. F., Anal. Chim. Acta, 1986, 179, 407. van Staden, J. F., Anal. Lett., 1986, 19, 1407. van Staden, J. F., Analyst, 1986, 111, 1231. Alegret, S., Alonso, J., Bartoli, J., Garcia-Raurich, J., and Martinez-Fabregas, E., J . Pharm. Biomed. Anal., 1988,6,749. Cammann, K., Fresenius 2. Anal. Chem., 1988, 329, 691. Cardwell, T. J., Cattrall, R. W., Hansen, P. C., and Hamilton, I. C., Anal. Chim. Acta, 1988, 214, 359. Alexander, P. W., Haddad, P. R., and Trojanowicz, M., Anal. Chem., 1984,56, 2417. Zaitsu, T., Maehara, M., and Toei, K., Bunseki Kagaku, 1984, 33, 149. Cooper, M. M., and Spurlin, S. R., Anal. Lett., 1986, 19,2221. Lach, G., and Bachmann, K., Anal. Chim. Acta, 1987,196,163. Polta, J. A., and Johnson, D. C., Anal. Chem., 1985,57,1373. Frenzel, F., Almhofer, N., Citroni, G., and Hager, W., Fresenius Z. Anal. Chem., 1988, 330,489. Granados, M., Maspoch, S. , and Blanco, M., Anal. Chim. Acta, 1986, 179, 445. MilosavljeviC, E. B., Solujie, L., Hendrix, J. L., and Nelson, J. H., Anal. Chem., 1988, 60, 2791. Rocklin, R. D., and Johnson, D. C., Anal. Chem., 1983,55,4. Kubala, S. W.. Tilotta, D. C., Busch, M. A., and Busch, K. W., Anal. Chem., 1989,61,2785. Analytical Methods Committee, Analyst, 1987, 112, 199. van der Linden, W. E., Anal. Chim. Acta, 1983, 151, 359. Hollowell, D. A,, Pacey, G. E., and Gordon, G., Anal. Chem., 1985,57,2851. Paper 01028641 Received June 26th, 1990 Accepted August lst, 1990

ISSN:0003-2654    

DOI:10.1039/AN9911600049

出版商:RSC    

年代:1991

数据来源: RSC