Direct measure of the low-density fractions of ... - ACS Publications

Chemistry Department, Oklahoma State University, Stillwater, Oklahoma 74078- ... States. The magnitude of the program for screening the general public...
0 downloads 0 Views 678KB Size
Anal. Chem. 1991, 63,2947-2951

2947

Direct Measure of the Low-Density Fractions of Serum Cholesterol Neil Purdie* and Laura H. Murphy

Chemistry Department, Oklahoma State University, Stillwater, Oklahoma 74078-0447 Robin B. Purdie

Wellness Center, Oklahoma State University, Stillwater, Oklahoma 74078-0447

Data on total chdesteroi (TC) and its dlstribution among the three solubilizing ilpld fractions in human serum have been obtained from three independent laboratories, and binary Ilnear corrdatbs between TC a d each of the various fractions are compared. Two sources used the approved doubie-enzymatk multbtep Aliain-Trlnder reactbn wlth absorption d e tection. I n the third method, which is entirely new, a nonenzymatk chromogenic reactlon and circular dichroism (CD) detection were used. TC results from all three sources are In excelient agreement. HDLC values measured by both enzymatic methods also agree In their correlations with TC but these are quite different from the correiatlon observed between HDLC and TC values obtained by the new nonenzymaticp”. Reas0nsaregh”suggestthatthe m n z y m a t k method Is more accurate for the meawremnt of the lowdensity llpld fractions and why health risk determinations that are based upon calculated values for this variable should be deemphasized until a more dependable procedure is approved for use.

INTRODUCTION Determination of serum cholesterol is one of the most prevalent clinical diagnostic measurements made in the United States. The magnitude of the program for screening the general public is so immense that automated methods are necessary. Methods presently in use differ in complexity from the simple dipstick approach to the sophisticated lipid profile tests in which the distributions of cholesterol among the three solubilizing macromolecules are determined (I1. In a report (2) from the Laboratory Standardization Panel ( U P ) of the National Cholesterol Education Program (NCEP) the risk of coronary artery disease has been correlated with three ranges of total cholesterol (TC): low risk if less than 200 mg/dL; marginal in the range 200-239 mg/dL; and high if greater than 240 mg/dL. Other risk factors, such as age, gender, heredity, and the use of tobacco and alcohol, are added as bases for further patient counseling (3,4).This relatively simple approach replaces an earlier recommendation (5) in which relative risk was established using the calculated ratio of TC to the high-density lipoprotein cholesterol (HDL-C) subfraction which is customarily measured in a second independent test. The same report (2) discloses the serious inaccuracies that exist in the measured amounts of TC in human serum reference standards. In data from 1500 laboratories, 47% failed to measure the true value to within a coefficient of variance (CV) of &5%, and 18%of these failed at a CV of *lo%. The outcome is the LSP has recommended that an improvement to within &3% CV for TC should be achieved by 1992. In-

* To whom correspondence should be addressed. 0003-2700/91/0363-2947$02.50/0

accuracies associated with the determination of the dietribution of cholesterol among the various lipid and lipoprotein subfractions were not reported either at that time or in a subsequent report (6). Cholesterol is distributed in human serum bonded to high-density (HDL-C) and low-density lipoprotein (LDL-C) fractions and to triglycerides (TGL) as the very low density (VLDL-C) lipoprotein form. Statistical evidence from a number of long-term clinical tests seems to justify that high proportions of HDLC and low proportions of LDLC correlate with lower relative risk (I, 2). In a typical lipid profile study, TC and HDL-C are measured directly in two separate testa, VLDL-C is taken to be a fixed fraction (e.g. either 0.16 or 0.2) of the TGL measured directly in a third test, and LDL-C is calculated as (TC - HDL-C - O.2TGL). The very poor proficiency and lack of reliability in the measurement of serum or plasma HDL-C has been independently addressed and interlaboratory Cv’s as high as 38% were reported (7,8).The propagation of errors in each of the three independent measurements makes LDL-C the fraction known with least overall accuracy. Because of this it is difficult to meaningfully monitor and justify that clinical progress has been made in LDL-C reduction therapy with time. The purpose of this work is to make a comparative analysis of data from different laboratories that profile the cholesterol lipoprotein distribution using minor variations of the same method, to confirm in another way the gravity of the experimental error in the measurement of HDL-C, and to propose for consideration a different chromogenic reaction and a different detector for serum cholesterol measurements. EXPERIMENTAL SECTION Absorbance in the visible region is the detection method that is used in the clinically approved diagnostic methods for serum cholesterol, all of which use the Allain-Trinder ( 9 , I O ) reaction scheme as the chromogenic process. The Allain part of this reaction (9) is a double-enzymatic two-step process in which the final products are cholest-5-en-3-oneand hydrogen peroxide. The hydrogen peroxide combines in a third step, i.e. the Trinder reaction (IO),in which the end product is the red form of a quinonimine dye that is not structurally related to cholesterol but has an absorbance intensity that is proportional to the cholesterol concentration. One of the external diagnostic laboratories used the Abbott Vision clinical autoanalyzer, subsequently designated as A, and the other the DuPont aca or D autoanalyzer. HDL-C measurements are made using the same color reaction after the selective precipitation of the combined LDL-C and VLDL-C fractions with a prepared aliquot of dextran sulfate-Mg. Absorbance measurements are made at a single wavelength. The A data set consisted of 538 serum samples, and the D set had 130. TGL values were not measured for all of the samples but an overall sample size of 270 is taken to be fairly representative of the total number. TC and HDL-C measurements for each sample were made only once before the specimens were released to our laboratory. On receipt the serum layer was withdrawn and stored at 0-5 “C. 0 1991 American Chemical Society

2048

ANALYTICAL CHEMISTRY, VOL. 63,NO. 24, DECEMBER 15, 1991

The second procedure was developed in our laboratory and forms the basis for a patent application that was filed with the U S . Patent Office, Jan 1990. From our many experiences with the obvious analytical selectivity that is associated with CD detection ( I I ) , our objective was to use CD to discriminate among the three lipoprotein fractions simultaneouslywithout performing the precipitation step. The cholesterol molecule has a fairly strong CD-active band that maximizes around 210 nm. It is of no value for analytical detection unless all of the strong UV absorbers are first removed from the serum. For CD detection to work, the cholesterol molecule has to be derivatized, and for the most convenience the induced CD band should be located in the visible region of the spectrum. The colored products of the AllainTrinder and the Liebermann-Burchard (12)reactions are not CD active and cannot be considered. The chromogenic reaction that was finally selected is that attributed to Chugaev (13). The reagent as described by Chugaev is a two to one mixture of 20% w/v anhydrous ZnClz in glacial acetic acid and 98% acetyl chloride. The colored end product is a CD-active unsaturated steroid whose precise structure is unknown although a mechanism has been suggested in which the B ring of the steroid nucleus is believed to open to produce an analogue of vitamin D (14). In CD detection differences between the absorbances of two coincident circularly polarized incident beams by the active medium are measured as a function of wavelength (11). It is more selective than absorbance detection. Potential interferences are other CD-active compounds and strong absorbers that adversely affect the signal/noise ratio. Other steroids do react with the Chugaev reagent but do not interfere with the cholesterol test because serum levels are much too low and each has its own very unique CD spectrum (14,15).High levels of hemoglobin released from hemolyzed samples do affect the signal/noise. Turbid specimens do not affect the CD spectrum in the same way an absorbance spectrum is affected because scattering of both incident beams is canceled in the subtraction of left and right absorbances. Spectral measurements were made on a JASCO-500A CD automatic spectropolarimeter equipped with a DP-SOON data processor. In serum sample measurements, 2 mL of the zinc reagent is added to a 50-wL aliquot of serum placed in a 10-mL vial. To this 1 mL of acetyl chloride is carefully added. The thoroughly mixed solution is incubated at 67 "C for 8 min during which time a reddish orange color develops. Chloroform (1mL) is added after cooling, the mixture is centrifuged for 1 min and transferred to the spectrophotometric cuvette, and the CD spectrum measured from 625 to 325 nm. Serum and reagent volumes and the experimental conditions used in this developmental work were dictated by the fact that the 1 cm path length cuvette used had a total volume of 3 mL. Miniaturization is entirely possible and is included as part of the patent application. Everything would be scaled down accordingly, and it is conceivable that a lipid profile assay might be done on a serum sample as small as a finger stick. The data set that was measured by CD numbered 150 samples and was composed of subsets taken at random from both the A and the D data sets. A lack of funds prevented us from making comparative measurements for the same serum samples using all three methods which we fully understand detracts from the completeness of the analysis. It will be seen however that arguments to combine the enzymatic data into one set are quite valid. TGL data were not available for all of the 150 samples. RESULTS AND DISCUSSION The CD spectrum for the colored product from the Chugaev reaction with the NI3S cholesterol standard reference material (SRM 911a) in chloroform is shown in Figure 1. Data from a series of dilutions of the standard can be used to prepare calibration curves at a number of wavelengths. Reactions with serum cholesterol and with standard solutions of cholesteryl fatty acid esters in chloroform produce analogous spectra which suggests to us that the cholesterol is totally converted to the acetate ester under the conditions of the Chugaev reaction. The selectivity of the CD detector is such that it is possible to measure the HDL-C and the combined (VLDL + LDL)-C fractions separately and directly in a single experiment that

ellipticity (mdeg)

I

350

400

450

/

wavelength (nm)

Figure 1. CD spectrum for the product of the color reaction between the Chugaev reagent and an NBS cholesterol standard reference material.

wuaveleoeh

-20

a

370

420

470

520

570

Figure 2. CD spectra for (a)total serum cholesterol, (b) the (VLDL + LDL)-C fraction, equal to (a) minus (c),(c) the HDL-C fraction after the addition of phosphotungstate-Mg precipitating reagent. does not involve a precipitation step. The CD spectrum in Figure 2a is representative of the spectrum for total serum cholesterol in all its forms. Figure 2c is the spectrum of the fraction that remains after the selective precipitation of the combined low-density fractions by the addition of phosphotungstate-Mg precipitating reagent according to the Sigma353-2 procedure; i.e. the remainder is the HDL-C fraction. Figure 2b is the difference between the spectra for TC and HDL-C and is therefore representative of the (VLDL + LDL)-C fraction. The precipitating agent was added only to correlate CD bands in the total spectrum with contributions from particular subfractions, and was not a part of the routine measurement of serum cholesterol. Further confirmation of the band assignments is obtained from the similarity between Figure 2c and the CD spectrum for the a-lipoprotein fraction (associated with HDL-C) which is selectively separated on a heparinagarose stationary phase according to the procedure developed by ISOLAB. The validity of the Chugaev-CD procedure and the accuracy of the method for the determination of TC are further vindicated in a comparison with TC data for Standard Reference Materials (SRM) for Cholesterol in Human Serum (frozen) that are available from what was then NBS (now NIST) and listed in their literature as SRM (1951-1) (210.36 f 2.46 mg/dL), SRM (1951-2) (242.29 f 1.53 mg/dL), and SRM (1951-3) (281.97 i 1.82 mg/dL). According to the NBS literature the SRM samples were donated by the Center for Disease Control (CDC) and the TC measurements were made at NBS. From the accompanying Certificate of Analysis these data are reported to compare extremely well with the CDC determinations in which a modified Abell-Kendall (16) method was used. Corresponding TC values that were obtained using the Chugaev-CD procedure, calculated by adding together the two fractions were 206, 241.1, and 286 mg/dL, respectively. We believe that this is convincing evidence for the validity of the new procedure. In the Chugaev-CD procedure values for the combined (VLDL + LDL)-C fractions are calculated from data taken

ANALYTICAL CHEMISTRY, VOL. 63, NO. 24, DECEMBER 15, 1991

2940

400

t

TC (CD)

m.

cholesterol E'aCtlOM (mg/dL)

200.

100 . TC(AmdB)

n

100

0

200

300

400

Figure 3. Total cholesterol (CD) vs total cholesterol (A) and (D); least-squares equatlon is y = -10.21 4- 1 . 0 0 5 5 ~( R 2 = 0.835).

at 525 nm and HDL-C values are obtained from data measured at either 490 or 390 nm or both. The most precise numbers for HDL-C are obtained when the calibration is done using the difference between the data at 490 and 390 nm which alleviates problems associated with drift in the instrument baseline from sample to sample. The sum of these two cholesterol subfractions gives the value for TC. Discrimination between the VLDL-C and the LDL-C subfractions in the data gathered at 525 nm is not possible at this time. Enzymatic procedures for the measurement of serum cholesterol are attractive because of the specificity of enzymes used as reagents. Absorbance is not a selective detector and has many potential interferences. In contrast the current nonenzymatic procedure involves a reagent that is specific to steroids and a detector which in the full spectrum mode is selective enough to discriminate among these steroids, as well as between the combined high- and low-density lipoprotein cholesterol subfractions. Interferences are apparently absent. There was no evidence to suggest that the resulta from CD detection were affected by the choice of either heparin or EDTA as the anticoagulating agent used by either of the source laboratories. Up to 30 independent repeat measurements were made for two individual serum samples both of which were relatively low in HDL-C (21 and 27 mg/dL) and in (VLDL LDL)-C (126 and 165 mg/dL). These low levels are representative of the most adverse analytical cases because signal sizes are among the smallest measured. Imprecisions were determined to be i 6 . 3 % CV for HDL-C using the difference data at 390 and 475 nm. For (VLDL + LDL)-C the observed CV was i2.3%. Measurements were made over periods of 2-3 weeks. Imprecisions in the measurement of the (VLDL LDL)-C fraction are already lower than the 1992 recommended range of *3.0% CV for TC proposed by the LSP. The imprecision in the measurement of HDL-C using the difference data is an improvement over the figures quoted above in the Introduction. Figures in the LSP Feb 1990 (6) report include data from laboratories that use both the A and the D procedures, and we take these figures to be typical of those expected from the two external laboratories contributing to this research. The most significant result from the CD procedure is that the combined low-density lipid cholesterol level is measured both by direct methods and with excellent precision. For the purposes of monitoring LDL-C reduction therapies, it is unfortunate that the information for the VLDL-C cannot be separated from the LDL-C fraction. The same problem is however encountered with the ISOLAB procedure where both are eluted together as the combined &lipoprotein fraction. The Chugaev-CD procedure is arguably superior to the Allain-Trinder-absorption procedure because it is more precise in the measurement of HDL-C and offers a direct measure-

+

+

cholesterol E'aCtions (m%dL)

200

100

m Figure 5. TC vs HDLCC (D), VLDL-C (V), LDL-C (m), and (VLDL LDL)-C (e) for lab B. Correlation equations are y = 51.6 0 . 0 0 2 ~ 0

100

+

200

+

+

+

(R2 = O.O), y = -2.4 0 . 1 3 x ( R 2 = 0.247), y = -49.9 0.87x(R2 = 0.93), and y = -51.6 0.99x(R2 = 0.922), respectively. All data were used for the regression analysis, but for clarity, data for only 88

+

patients are plotted.

ment of the combined low-density fraction. The remainder of the discussion is devoted to more specific descriptions of the interrelationships that exist among the three data sets accumulated in this work. Correlations between TC from the AUain-Trinder (measured values) and the Chugaev-CD (calculated values) methods are excellent (Figure 3). The correlation slope and intercept are +1.005 and -10 mg/dL, respectively. The combination of TC values measured by the A and D procedures into one group is considered valid because the wet chemistries are the same and both methods are clinically approved. The evidence is clear that the CD method is valid for the measurement of TC. Since no new method with which to measure VLDL-C is presented here, comparisons between the Allain-Trinder and Chugaev methods ought to be limited to only HDL-C and (VLDL + LDL)-C vs TC. Data for these subfractions are plotted as a function of TC for A in Figure 4, for D in Figure 5, and for CD in Figure 6. Figures 4 and 5 also include correlations for the separate VLDL-C and LDL-C terms. Correlations are the same for the A and D data sets, that is all are linearly dependent on TC with the exception of HDL-C. All of the HDL-C data measured enzymatically can, in fact, be fitted by the single value of 50 f 10 mg/dL over the whole concentration range of TC. Because HDL-C is an independent measurement used to calculate LDL-C, the imprecisions in HDL-C are carried over. This is immediately obvious in the nonzero correlation intercept (-50 mg/dL) and

2950

ANALYTICAL CHEMISTRY, VOL.

03,

NO. 24, DECEMBER 15, 1991

.A 200-

-

1Q)

0

100

200

400

300

+

Flgurr 6. TC vs HDL-C ( O ) , VLDL-C ( 0 ) ,LDL-C (a),and (VLDL LDL)-C (e) for CD. Correlation equations are y = 7.2 0 . 1 5 x ( R 2 = 0.342), y = -1 1.7 0 . 1 8 ~( R 2 = 0.247), y = 3.9 0 . 6 8 (~R 2 = 0.7531, and y = -7.3 0 . 8 5 ~( R 2 = 0.9451, respectively. All data were used for the regression analysis, but for clarity, data for only 74

+ +

+

+

patients are plotted. the correlation slope close to 1.00 for plots of (VLDL + LDL)-C vs TC in both Figures 4 and 5. Given that HDL-C can be fitted with a common value of 50 mg/dL for all samples measured enzymatically, the intercept bias is self-evident, as is the unit slope, because the plot of (VLDL LDL)-C vs TC is essentially a plot of (TC minus a constant) vs TC. Intercepts are also about -50 mg/dL for the LDL-C alone plotted against TC correlations, and the slopes are reduced to ca. 0.85 because of the corrections that are due to the linear dependences of VLDL-C with TC. It is clear that LDL-C values that are calculated using measured HDL-C values can be estimated just as well by assuming a constant value for HDL-C. Excellent correlations are observed for both (VLDL + LDL)-C and HDL-C as a function of TC determined by the CD method (Figure 6). Correlation intercepts are close to zero and the slopes correspond with figures which are often considered to be reasonable and acceptable distributions of total cholesterol among the various subfractions on the basis of ultracentrifugation data i.e. (VLDL LDL)-C (ca. 85%) and HDL - C (ca. 15%). If VLDL-C data from TGL measurements are used to separate the low-density fractions in the CD measurements then the derived LDL-C vs TC plot is linear with a correlation intercept of 4 mg/dL and a (LDL-C)/TC slope of 0.68 (Figure 6). Since the VLDL-C amounts to only one-fourth of the "average" LDL-C fraction for this population, the sum of the low-density fractions is a fair approximation to the LDL-C fraction alone and might be used in its place to make LDL-C reduction therapy decisions. The recent LSP report (6)includes a proposal that defines risk categories in terms of LDL-C levels rather than on figures for TC. Summarizing these in order of increasing risk the ranges are 240 mg/dL for TC. Recommended ranges that are based upon LDL-C figures are 160 mg/dL. In other words, the ranges are based on the assumption that LDL-C is about 66% of the total cholesterol which is just what we have determined the correlation slope to be for the variation of LDL-C with TC using the CD procedure. Imprecisions in HDL-C determinations using the AllainTrinder reaction will have to be improved considerably if the recommendation from the LSP is to be met. Otherwise one can do equally well by taking HDL-C to be constant and changing the 30 mg/dL interval in the LDL-C ranges to reflect this deficiency. A very good alternative is to measure only (VLDL + LDL)-C directly by the CD procedure and redefine the ranges at 85% of the cutoff values for TC risk assessments,

+

(In,

+

namely at 170 and 204 mg/dL, which gets around the assumption that everyone, regardless of the state of health, has a VLDL-C that is exactly equal to 0.2TGL. CONCLUSIONS In summary therefore we observe that a good correspondence is found to exist between the TC values measured by the enzymatic and nonenzymatic procedures. Values for HDL-C as measured by the two enzymatic procedures A and D also compare very well. A dependence on TC within individual populations is observed for all the lipid fractions with the exception of HDL-C for the A and D assays. This result suggests to us that there is a basic systematic error in the determination of HDL-C using the enzymatic method of analysis. One possible explanation is that the method involves a precipitation step which is difficult to reproduce consistently. Numerous other possible contributing factors have also been considered (5), among which is the inconsistency observed in HDL-C measurements when different precipitating agents are used. We had expected to discover that the greatest experimental errors occurred for the HDL-C measurements. This expectation was based upon the fact that because such large relative inaccuracies (CV>*5%) had been reported for the measurement of TC, even larger errors would result when the same method is used to measure the much smaller numerical values associated with HDL-C, not to mention the additional problems associated with the removal of low-density fractions (5). However we had not anticipated a correlation of zero. For the measurements made using an alternative color reaction coupled with CD detection, good linearity is seen for both HDL-C and (VLDL LDL)-C correlations with TC. The apparent improvement in the quality of the HDL-C measurements when CD is used is conceivably due to the fact that precipitation is not a part of the assay and neither CD-inactive hemolyzed red blood cells nor very high TGL levels interfere with CD detection. With this improvement in the precision for HDL-C data, the idea of using a ratio of TC to HDL-C as a second diagnostic parameter can be reconsidered. In addition the imprecisions in (VLDL + LDL)-C are reduced and this sum may be a viable substitute for LDL-C in reduction therapy studies. A fair conclusion from this work is that the present NCEP recommendation to base patient risk judgments only upon TC data is justified and that there is little reason to perpetuate the measurement of HDL-C for diagnostic purposes as long as the measurements are so imprecise. The assumption that (VLDL + LDL)-C i= LDL-C for diagnostic purposes, is a reasonable one because both correlate with TC in an equivalent manner and LDL-C is generally much larger than VLDL-C. Registry No. Cholesterol, 57-88-5.

+

LITERATURE CITED Abbott, R. D.; Garrison, R. J.; Wilson, P. W. F.; et al. Arteriosclerosis 1983, 3 , 260-272. Laboratory Standardizatlon Panel, NCEP A report from the Laboratory standardization Panel of ttw National Cholesterol Education Program. Clin. Chem. 1988, 3 4 , 193-201. Kannei, W. B.; Castelli, W. P.; Gordon, T.; et al. Ann. Intern. Med. 1871, 74, 1-11. Casteiii, W. P.; Garrison, R. J.; Wilson, W. F.;Abbott, R. D.; Kaiousdian, S.; Kannei, W. B. JAMA 1986. 256, 2835-2838. Superko, H. R.; Bachorik, P. S.; Wood, P. D. JAMA 1888, 256, 2714-2717. Laboratory Standardization Panel, NCEP: Recommendations for Improving Cholesterol Measurement. NIH Pubilcatlon No. 90-2964; US. Department of Health and Human Services, PHS: Washlngton, DC, Feb 1990. Warnick, G. R.; Aibers, J. J.; Teng-Leary. E. Clin. Chem. 1880. 26, 169-170. Grundy, S. M.; Goodman, D. W.; Rifkind, B. M.; Cleeman. J. I . Arch. I.n t .. .Med. . .. .._- . 1989. .- - -, 749 . .- , 505-510 - - - - .- . Allain, C. A.; Poon, L. S.; Chan, C. S. G.;Richmond, W.; Fu, P. C. Clin. Chem. 1974, 20, 470-475.

Anal. Chem. 1991, 63, 2951-2955 (10) Trlnder, P. A. Ann. Clln. Blochem. 1071. 37,38-41. (11) Purdle, N.; Swallows, K. A. Anal. Chem. 1080, 67, 77-MA. (12) Burke, W.; Diamondstone, E. I.; Velapoldl, R. A.; Menis, 0. Clln. Chem. 1074, 20, 794-801. (13) Chugaev (Tshugaev), L.; Gastev, A. Chem. Eer. 1010, 4 2 , AWRI-ARRA (14) Cox, R. H.; Spencer, E. Y. Can. J . Chem. 1051, 29, 217-222. (15) Purdie, N.; Purdie, R. N. Unpublished results, Chemistry Department, OSU, 1989.

2951

(16) Abell, L. L.; Levey, B. 6.; Brodie, 8. 6.; Kendall, F. F. J . E b l . Chem. 1852, 795, 357-366. (17) Tletz, N. W., Ed. Fundementals of Clinical Chemisby. 3rd ed.: WB Saunders Co.: Philadelphia, PA, 1987; p 457.

.--. .--..

RECEIVED

for review June 5, 1991. Accepted September 19,

1991.

Elemental Specificity in Dual-Channel Flame Photometric Detection of Gas Chromatographic Peaks Walter A. Aue,* Brian Millier, a n d Xun-Yun S u n Department of Chemistry, Dalhousie University, Halifax, Nova Scotia, B3H 4 J 3 Canada

Slgnals from a duakhannel flame photometric detector (FPD) can be dlgitally processed by a conditional-access algorithm (CONDAC) to yiekl chromatograms that are speclwlc (meanlng infinltely selective) for any chosen FPD-active element. Usually, one chromatographic separatlon stored In the computer can thus be examlned successlvely for the presence of several heteroatoms.

INTRODUCTION When it first became commercially available (I),the flame photometric detector (FPD) (2) offered selective detection for only two elements: phosphorus and sulfur. Since then, some 20 more elements have been shown to produce significant FPD response. (For this purpose a t hand, “significant response” is defined as being at least 10 times stronger than expected from the carbon portion of the molecule.) Figure 1 presents an overview of “best” molar detection limits as reported in the literature (B, P, S, Cr,As, I) or taken from the measurements of our own group. Similar to other selective GC detectors, the FPD is often used to scan complex environmental or medical samples that may show several hundred peaks in capillary gas chromatography. Out of an overwhelming number of hydrocarbonaceous matrix components, this type of survey analysis can pinpoint trace amounts of compounds that contain a particular, biologically interesting heteroelement. Missing a compound of that element or mistaking it for that of another thus becomes a real possibility. While the FPD was once restricted to volatile analytes, its recent use with techniques such as supercritical fluid chromatography and capillary HPLC has opened the way for the selective detection of many more compounds, including nonvolatile organometallics of industrial or biochemical importance. Because many more of these compounds can now be separated, the need for new detector modes of improved selectivity and unambiguous elemental recognition has greatly increased. An ideal FPD would respond in a specific manner to any element for which its interference filter had been chosen. (Note that the term “specific” is commonly used to mean “designed for” or “selective for” in the current FPD literature. In this study, “specific” retains its original analytical definition of “infinitely selective”-meaning that no other than the chosen analyte element is seen to respond.) Unfortunately, true specificity is found rarely if at all among the selective GC detectors, and in this regard the flame photometric detector is no exception. The spectra of most FPD-active ele0003-270019110383-295 1$02.5010

ments overlap severely, and there exist several wavelength regions where practically all species radiate to a major or minor degree. In fact, the increase in selectivity gained by using an interference filter is often quite small. Elemental selectivity, in all kinds of analytical systems, can be significantly improved by dual-channel differential operation (e.g. ref 3). In the FPD this approach can be most effectively used to discriminate against a complex hydrocarbon matrix and to confirm the identity of suspected heteroatoms. An example can be found in our recent determination of a manganese antiknock compound in gasoline (4). While this dual-channel subtraction method excels at eliminating response from compounds of one particular element, it proves cumbersome and time-consuming to use with multielement samples (5). In these, the analyst often needs to establish the presence of only one element (or of only one element at a time). This study examines the theory that elemental specificity can be achieved for resolved GC peaks from a dual-channel flame photometric detector. While tested only in the FPD, the described approach should be applicable wherever two synchronous channels of a suitably diverse and time-dependent information content are available from some sensing device. EXPERIMENTAL SECTION A gas chromatograph with a dual-channel flame photometric detector (Shimadzu GC-4BMPF) was used with a packed column (100 X 0.3 cm i.d. glass, 5% OV-101 on Chromosorb W, 100/120 mesh) under a nitrogen flow of 22 mL/min. The detector, with its standard quartz chimney removed, was run with 300 mL/min of hydrogen and 60 mL/min of air plus an additional 18 mL/min of nitrogen, under an efficient exhaust duct. The output from the two Hamamatsu R 268 photomultiplier tubes was amplified by two slightly modified GC electrometers (Shimadzu) and, via their “integrator” ports, passed onto a laboratory-made electrometer/computer interface. After being dampened by a simple RC filter to prevent aliasing noise in the digital data, the signal voltage was converted to pulses and sampled every tenth of a second. A full-scale input produced a digital count of 16K, about equal to a 14-bit analog-to-digital converter. The dual-channel counts were processed by a 12-MHzAT-compatible computer with the help of a 1-Mbyte memory, 40 Mbyte hard disk, 80 287 math coprocessor, VGA display adapter, and Multi-Sync monitor. The two necessary programs, named CHROM and CORR, were written in compiled Quick Basic to generate a reasonably fast code. All data acquisition was performed by assembly language routines using interrupts for timing accuracy. CHROM is an existing, laboratory-developed program for highaccuracy display and manipulation of dual-channel chromatograms. It offers up to 1600 s (26 min 40 sec) of chromatographic 0 1991 American Chemical Society