Shinichi Usui, 1 Masakazu Nakamura, 2 Kazuhiro Jitsukata, 3 Masayuki Nara, 4 Seijin Hosaki, 1 and Mitsuyo Okazaki 4*

Transcription

1 Clinical Chemistry 46: (2000) Lipids, Lipoproteins, and Cardiovascular Risk Factors Assessment of Between-Instrument Variations in a HPLC Method for Serum Lipoproteins and Its Traceability to Reference Methods for Total Cholesterol and HDL-Cholesterol Shinichi Usui, 1 Masakazu Nakamura, 2 Kazuhiro Jitsukata, 3 Masayuki Nara, 4 Seijin Hosaki, 1 and Mitsuyo Okazaki 4* Background: The main purpose of this study was to evaluate the between-instrument variation of the HPLC method for the measurement of total cholesterol (TC), HDL-cholesterol (HDL-C), LDL-cholesterol (LDL-C), VLDL-cholesterol (VLDL-C), chylomicron cholesterol (CM-C), LDL size, and HDL size. Furthermore, the accuracy of the HPLC was assessed for the determination of TC and HDL-C, compared with CDC reference methods. Methods: We used four HPLC instruments with different column-load numbers from 250 to For accuracy assessment of TC and HDL-C, we used the reference methods recommended by the CDC. Results: The values measured by the four instruments were highly correlated with each other (mean r 0.965), and the absolute mean differences were 4 43 mg/l for TC, 4 30 mg/l for HDL-C, 0 48 mg/l for LDL-C, 7 66 mg/l for VLDL-C, 0 7 mg/l for CM-C, nm for LDL size, and nm for HDL size. For TC, the HPLC instruments showed high correlation and good agreement with the reference method: r 0.997; total error <6.6%; absolute mean bias <1.2%. For HDL-C, the results from the HPLC method were significantly higher 1 School of Allied Health Science, Tokyo Medical and Dental University, Yushima, Bunkyo-ku, Tokyo , Japan. 2 Department of Epidemiology and Mass Examination for CVD, Osaka Medical Center for Cancer and Cardiovascular Diseases, Nakamichi, Higashinari-ku, Osaka , Japan. 3 Department of Clinical Chemistry, SRL Inc., 51 Komiya-cho, Hachiohjishi, Tokyo , Japan. 4 Laboratory of Chemistry, College of Liberal Arts and Sciences, Tokyo Medical and Dental University, , Kohnodai, Ichikawa-shi, Chiba , Japan. *Author for correspondence. Fax ; okazaki.cul@ cul.tmd.ac.jp. Received July 15, 1999; accepted November 2, (10.8% absolute mean bias) than those of the CDC reference method, in spite of good correlation between the two methods (r 0.998). Conclusions: The between-instrument variation in serum lipoprotein analysis by HPLC was confirmed to be very small. This method met the US National Cholesterol Education Program s performance criteria for TC but not for HDL-C American Association for Clinical Chemistry Numerous epidemiological studies have shown that longterm hyperlipidemia is associated with coronary heart disease. High blood cholesterol or increased LDL-cholesterol (LDL-C) 5 and decreased HDL-cholesterol (HDL-C) have received attention because of their links to the increased risk of atherosclerotic disease (1 3). Small dense LDL, one of the subclasses of LDL, has an important clinical significance as well (4, 5). Both quantitative and qualitative approaches to lipoproteins have been clinically required. Lipoprotein particles are classified on the basis of their hydrated density, size, and electrophoretic mobility. Major lipoprotein classes have been defined according to their density. The separation and quantification of lipoproteins by ultracentrifugation is not suited to the clinical laboratory setting, however, because the method is time-consuming and requires special training and large sample volumes. HPLC is a method for classifying and quantifying lipoproteins by particle size (6). In the HPLC method, the 5 Nonstandard abbreviations: LDL-C, LDL-cholesterol; HDL-C, HDL-cholesterol; CM-C, chylomicron cholesterol; VLDL-C, VLDL-cholesterol; TC, total cholesterol; OMC, Osaka Medical Center for Cancer and Cardiovascular Diseases; TG, triglyceride; and NCEP, National Cholesterol Education Program. 63

2 64 Usui et al.: Variations in a HPLC Method lipoproteins separated by gel permeation columns (TSKgel LipopropakXL; Tosoh) are detected by an on-line enzymatic reaction for cholesterol. Chylomicron cholesterol (CM-C), VLDL-cholesterol (VLDL-C), LDL-C, HDL-C, and total cholesterol (TC) are calculated automatically by a computer (7). The greatest advantages of the HPLC method are direct profiling of lipoproteins in small sample volumes, and rapid and simultaneous determination of their size distribution. For many epidemiological and clinical studies, a reliable and convenient technique is required to allow quantification and characterization of lipoprotein subclasses directly from serum samples. Several studies have compared the analytical performance of the HPLC method for lipoproteins with established methods, including an automated enzymatic assay for TC (8), and homogenous (9, 10), precipitation (9 11), and sequential ultracentrifugation (7) methods for HDL-C. However, the interlaboratory or between-instrument variation of this assay has not been examined. Furthermore, it is necessary to assess the accuracy of the HPLC method in comparison with the reference method to establish traceability to the reference method recommended by CDC (12). The main purpose of this study was to evaluate the between-instrument variation for measurement of TC, HDL-C, LDL-C, VLDL-C, CM-C, LDL size, and HDL size. We also participated in the certification protocols provided by the CDC s Cholesterol Reference Method Laboratory Network to assess the accuracy of the TC and HDL-C assays. Materials and Methods samples For comparison studies, two series (groups A and B) of pooled serum samples were prepared at Osaka Medical Center for Cancer and Cardiovascular Diseases (OMC), an international member of the Cholesterol Reference Method Laboratory Network. Group A (n 50) and group B (n 46) samples had a balanced distribution of TC and HDL-C concentrations, respectively, and had triglyceride (TG) values within the reference interval. Each of the pooled serum samples was obtained over an 2-week period from the daily pools, which consisted of fresh sera from fewer than four outpatients. Aliquots of these samples were kept at 80 C and shipped to our laboratories on dry ice. For the TG interference study, serum samples (group C) were collected from hypertriglyceridemic patients (n 19) with serum TG concentrations 2000 mg/l at Osaka University Hospital after informed consent. Group C samples were kept at 4 C and analyzed within 10 days after sample collection. Precision studies were performed using two different pooled sera (QC1 and QC2) stored at 80 C. QC1 and QC2 were prepared by the same procedure as described above and had TC values of 1370 and 2180 mg/l, respectively. calibration material An in-house pooled serum (lot no ) was prepared as the calibration material for the HPLC method. This material was obtained by combining fresh sera from healthy volunteers after a 12-h fast. Reference values (1770 mg/l for TC and 653 mg/l for HDL-C) were assigned to this material, using the reference methods at OMC. Aliquots of the material were stored at 80 C. hplc method We used an automated TOSOH Lipoprotein Analytical System (Tosoh), which included an AS-8020 autosampler, CCPS and CCPM-2 pumps, and a UV-8020 detector (7). Serum lipoproteins were separated using two connected columns ( mm) of TSKgel LipopropakXL (Tosoh) with TSKeluent LP-1 (Tosoh) as the elution buffer at a flow rate of ml/min. Cholesterol was detected by an on-line enzymatic reaction through the mixing of the effluent from the columns with a commercial kit, Determiner L TC (Kyowa Medex). Reagents 1 and 2 of the commercial kit were continuously pumped at flow rates of and ml/min, respectively. The absorbance at 550 nm was continuously monitored after the enzymatic reaction at 45 C in a reactor coil (7.5 m 0.4 mm i.d.). A 5- L sample was injected at an interval of 16 min per sample. An in-house pooled serum (lot no ) was used as the calibrator. HPLC profiles of the calibration material and a sample from a hyperlipidemic patient (TC, 3240 mg/l; TG, mg/l) are shown in Fig. 1. The cholesterol concentrations in the major lipoprotein classes were estimated from a linear calibration curve of the assigned TC value plotted against the total area with the calibrator, after automatic calculations of total area and individual areas corresponding to HDL, LDL, VLDL, and CM. Calibration of columns was carried out using latex Fig. 1. Cholesterol profiles obtained with the HPLC method (instrument A). Solid line, calibrator with a TC concentration of 1770 mg/l and a TG concentration of 560 mg/l. Dashed line, sample from a hyperlipidemic patient with a TC concentration of 3240 mg/l and a TG concentration of mg/l. Vertical dotted lines, positions of LDL (25.3 nm) and HDL (11.3 nm) in the calibrator.

3 Clinical Chemistry 46, No. 1, beads (Magsphere) 25 and 37 nm in diameter, and highmolecular weight calibrators (Pharmacia Biotech) containing thyroglobulin (17 nm), ferritin (12.2 nm), catalase (9.2 nm), albumin (7.1 nm), and ovalbumin (6.1 nm). The mean particle sizes of LDL and HDL in the calibrators were 25.3 and 11.3 nm, respectively, based on the calibration curve. The elution times of the observed peaks on a chromatogram were converted to particle sizes, using a linear calibration curve of the logarithm of the particle sizes plotted against the elution times corresponding to LDL and HDL in the calibrator. In this study, we used four HPLC systems (instruments A, B, S1, and S2) with different column-load numbers. The term column-load number indicates the number of samples injected into the columns. At the beginning of this study, the column-load numbers for each instrument were 5000 (column nos. T9056 and T9057) for instrument A, 250 (column nos. 84SM2Y0005 and 84SM2Y0006) for B, 2000 (column nos. 84SM1Y0023 and 84SM1Y0024) for S1, and 2500 (column nos. T9091 and T9092) for S2. The HPLC method was performed at Tokyo Medical and Dental University with instruments A and B (in Tokyo and Chiba, respectively), and at SRL Inc., in Tokyo, with instruments S1 and S2. The same calibrator was used for all instruments. All samples were analyzed in duplicate. cdc reference methods for tc and hdl-c The reference method for cholesterol is the Abell-Levy- Brodie-Kendall assay, as modified by the CDC, described previously in detail (13). In this method, serum is saponified with ethanolic KOH and extracted with hexane. An aliquot of the extract is evaporated, and the residue is reacted with Liebermann-Burchard reagent to develop the color, which is measured at 620 nm. The reference method for HDL-C consists of a threestep procedure involving ultracentrifugation, precipitation, and cholesterol analysis (14). After triglyceride-rich lipoproteins are removed from the serum by ultracentrifugation against a solution with a density of kg/l, apolipoprotein B-containing lipoproteins in the ultracentrifugal infranate are precipitated with heparin and MnCl 2, and the cholesterol in the heparin-mncl 2 supernatant is measured by the Abell-Kendall method at CDC. The reference methods were performed in duplicate at OMC in Japan. tg analysis TG values were determined enzymatically in a single analysis with a Hitachi 7170 automated analyzer, using reagent kit L Type Wako TG-H (Wako Pure Chemical Industries). The TG assay was corrected for the presence of endogenous glycerol. Results precision The results of the precision studies are summarized in Table 1. Two different materials (QC1 and QC2) with low (1370 mg/l) and high (2180 mg/l) TC concentrations were analyzed to assess the precision of the four HPLC systems. One aliquot of each pooled serum was assayed by each instrument in five replicates each day for 4 days and in duplicate each day for 16 days. The within-day variance (V within ) was determined as the mean of the daily variances at the first 4 days. The between-day variance (V between ) was determined as the variance (V) of the daily means at 20 days, which was adjusted for the within-day variance component; V between V (V within /Number of Table 1. Summary of the analytical precision studies of four HPLC systems. a Within-day Between-day Total Sample Mean SD b CV, % SD CV, % SD CV, % TC, mg/l QC QC HDL-C, mg/l QC QC LDL-C, mg/l QC QC VLDL-C, mg/l QC QC CM-C, mg/l QC QC LDL size, nm QC QC HDL size, nm QC QC a The results were obtained from four HPLC systems (A, B, S1, and S2) with two different pooled sera (QC1 and QC2) analyzed in five replicates for 4 days and in duplicate for 16 days. Within-day imprecision values were based on the results for 4 days, and between-day imprecision values were based on the results for 20 days. Mean values are the means of the results from all instruments. b SD and CV are the minimum and maximum values of the results from all instruments.

4 66 Usui et al.: Variations in a HPLC Method Table 2. Between-instrument comparison of the HPLC method for cholesterol content and particle size in major lipoprotein classes. x y Slope a Intercept b r x b y b difference b Absolute mean TC A B A S A S B S B S S1 S HDL-C A B A S A S B S B S S1 S LDL-C A B A S A S B S B S S1 S VLDL-C A B A S A S B S B S S1 S CM-C A B A S A S B S B S S1 S LDL size A B A S A S B S B S S1 S HDL size A B A S A S B S B S S1 S a Slopes and intercepts were calculated by the Deming regression analysis. b Intercept, mean values, and differences are given in mg/l for cholesterol concentrations, and in nm for particle sizes. repeats per day). Total variance (V total ) was the value of V within V between. At the low TC concentration (QC1), the total imprecision (CV) values for A, B, S1, and S2, respectively, were as follows: 1.5%, 2.7%, 1.6%, and 2.0% for TC; 1.9%, 3.3%, 2.0%, and 2.4% for HDL-C; 1.2%, 2.2%, 1.6%, and 1.4% for LDL-C; 2.3%, 5.2%, 4.5%, and 5.4% for VLDL-C; 12%, 19%, 43%, and 32% for CM-C; 0.35%, 0.37%, 0.39%, and 0.33% for LDL size; and 0.55%, 0.92%, 0.82%, and 0.43% for HDL size. At the high TC concentration (QC2), the total imprecision values for A, B, S1, and S2, respectively, were as follows: 1.7%, 1.8%, 0.8%, and 0.8% for TC; 2.8%, 2.8%, 0.9%, and 1.2% for HDL-C; 1.8%, 1.9%, 1.3%, and 1.5% for LDL-C; 2.5%, 4.7%, 4.2%, and 5.9% for VLDL-C; 17%, 11%, 37%, and 29% for CM-C; 0.37%, 0.39%, 0.50%, and 0.31% for LDL size; and 0.73%, 0.78%, 0.79%, and 0.41% for HDL size. Because the CM-C concentrations in the samples used were 7 mg/l, the CVs were relatively higher than

5 Clinical Chemistry 46, No. 1, Fig. 2. Scatter (A) and bias (B) plot for the TC assay (n 50). We used four HPLC instruments: A (F),B(E), S1 ( ), and S2 ( ). The dotted line in A represents y x; the dotted line in B represents y 0. Bias (%), [(HPLC Reference method)/reference method] 100. for the other analytes; the SD for the CM-C assay was mg/l. between-instrument comparison Using four HPLC instruments and group-a samples (n 50), we estimated between-instrument variation for the measurement of lipoprotein cholesterol concentrations (TC, HDL-C, LDL-C, VLDL-C, and CM-C) and the particle sizes of LDL and HDL. To examine the degree of agreement among the values obtained by the four instruments, we calculated linear regression equations by Deming regression analysis and the absolute mean differences between each apparatus (Table 2). All four instruments correlated highly with each other and produced good correlation coefficient values of for TC, for HDL-C, for LDL-C, for VLDL-C, for CM-C, for LDL size, and for HDL size. The absolute mean differences among HPLC apparatuses were 4 43 mg/l for TC, 4 30 mg/l for HDL-C, 0 48 mg/l for LDL-C, 7 66 mg/l for VLDL-C, 0 7 mg/l for CM-C, nm for LDL size, and nm for HDL size. Statistical differences as assessed by the Student paired t-test were not observed between the TC values obtained from instruments S1 and S2, between the HDL-C values obtained from B and S2, between the LDL-C values obtained from B, S1, and S2, and between CM-C values obtained from A and S2. For the VLDL-C, LDL size, and HDL size assays, the differences between the values obtained from the four instruments were significant (P 0.05) but slight. LDL and HDL were detected in all samples by all four instruments, but CM was detected in only 15 samples by A, 24 samples by B, 12 samples by S1, and 16 samples by S2. VLDL was not detected by any of the instruments. Between-instrument imprecision values, which were calculated by an analysis of variance, were 1.9% for TC, 3.1% for HDL-C, 3.0% for LDL-C, 10% for VLDL-C, 49% for CM-C, 0.6% for LDL size, and 0.8% for HDL size. Table 3. Correlations and differences between four HPLC systems and the reference method for TC. Instrument n Slope (SE) a Intercept (SE), a mg/l r Mean bias, b % Total error, c % A (0.0114) 12 (26.0) B d (0.0103) 44 (23.6) S (0.0084) 33 (19.3) S (0.0087) 17 (19.9) a Slopes and intercepts were calculated by least-squares linear regression analyses. b Mean bias, %: mean of [(HPLC Reference method)/reference method] 100. c Total error, %: (Mean bias Total imprecision) d Significantly different from 1.00.

6 68 Usui et al.: Variations in a HPLC Method Fig. 3. Scatter (A) and bias (B) plot for the HDL-C assay (n 48). We used three HPLC instruments: A ( ),B( ), and S1 ( ). The dotted line in A represents y x; the dotted line in B represents y 0. Bias (%), [(HPLC Reference method)/reference method] 100. accuracy for tc and hdl-c The results obtained for the TC assay with all four instruments and the reference method are shown in Fig. 2 and Table 3. Linear regression analysis by the leastsquares method showed that only instrument B gave a slope that was significantly 1.0, but the intercepts did not differ significantly from 0 mg/l for any of the instruments. The estimated values at 2200 mg/l (the upper end of the reference interval), using the regression lines, were 2212 mg/l for A, 2222 mg/l for B, 2189 mg/l for S1, and 2195 mg/l for S2, which corresponded to 0.5%, 1.0%, 0.5%, and 0.2% bias, respectively. The mean biases for A, B, S1, and S2 were 0.3%, 1.2%, 0.6%, and 0.5%, respectively. The total error (%), which is equal to the systematic error [absolute mean bias (%)] plus the random error [total imprecision (%) 1.96], was 3.6% for A, 6.6% for B, 3.7% for S1, and 4.3% for S2 (Table 3). Three instruments (A, B, and S1) were used for the comparison with the reference method for HDL-C. Group B samples (n 46) were analyzed using the same conditions as for the TC assay. The results are shown in Fig. 3 and Table 4. Although the two methods correlated very well, with correlation coefficients of , the slopes and intercepts of the regression lines were significantly different from 1.0 and 0 mg/l, respectively. The HPLC method overestimated HDL-C values at all points between 208 to 853 mg/l when compared with the reference method. Estimated values at 350 mg/l (the lower end of the reference interval), using the regression lines, were 396 mg/l (13.1% bias) for A, 381 mg/l (8.9% bias) for B, and 395 mg/l (12.9% bias) for S1. Mean biases of 12.3%, 8.4%, and 11.6% were found for instruments A, B, and S1, respectively, for the HDL-C assay. The total error was 17.2% for A, 15.0% for B, and 13.6% for S1. tg interference We used instrument A and group C samples (n 19) to assess the performance of the HPLC method for patients with a wide range of serum TG concentrations from 2110 to mg/l. With one sample that had a TG concentration of mg/l, the TC value was remarkably overestimated by the HPLC method relative to the refer- Table 4. Correlations and differences between three HPLC systems and the reference method for HDL-C. Instrument n Slope (SE) a Intercept (SE), a mg/l r Mean bias, b % Total error, c % A d (0.0087) 18 e (4.3) B d (0.0143) 17 e (7.1) S d (0.0057) 34 e (2.8) a Slopes and intercepts were calculated by least-squares linear regression analyses. b Mean bias, %: mean of [(HPLC Reference method)/reference method] 100. c Total error, %: (Mean bias Total imprecision) d Significantly different from e Significantly different from 0.

7 Clinical Chemistry 46, No. 1, Fig. 4. HPLC method compared with the reference methods for TC (A) and HDL-C (B) by least-squares linear regression analysis. Samples were from hypertriglyceridemic patients (n 18). Solid lines are regression lines: (A), y 1.06x 152 (r 0.951); (B), y 1.04x 52.9 (r 0.985). The dotted lines indicate y x. ence method: the TC values determined by the reference and HPLC methods were 2945 and 4306 mg/l, respectively. Because the bias of 46.2% (1361 mg/l) exceeded the sum of the mean bias plus SD multiplied by 3, the sample was excluded from all data analyses, although the reason for the discrepancy was not clear. In Fig. 4A, the reference (x) and HPLC (y) methods yield a linear regression line of y 1.06x 152 (r 0.951; n 18) for TC assay. Fig. 5A, in which the biases vs the reference method are plotted as a function of serum TG concentrations, shows an increasingly negative bias (r 0.743; P 0.001) with increasing TG concentration. Samples from patients with high TG concentrations often give high VLDL-C and/or high CM-C values. Plotted as a function of CM-C concentration as determined by the HPLC method rather than serum TG concentrations, TC assay biases were inversely correlated (r 0.963: P 0.001) with CM-C concentrations (Fig. 5B). Therefore, we assumed that the detection of CM-C might be incomplete in the HPLC method, which would lead to underestimation of TC. The mean bias, however, was 0.1% (6 mg/l) and small in an overall range of TG concentrations up to mg/l. For the HDL-C assay, the reference (x) and HPLC (y) Fig. 5. Effect of increased serum TG on the HPLC methods for TC (A and B) and HDL-C (C). Samples were from hypertriglyceridemic patients (n 18). Bias [(HPLC Reference method)/reference method 100] was plotted as a function of TG (A and C) and CM-C (B) concentrations. CM-C values were obtained by the HPLC method. Solid lines are regression lines: (A), r 0.743; P 0.001; (B), r 0.963; P 0.001; (C), r 0.720; P Dotted lines indicate 0% bias.

8 70 Usui et al.: Variations in a HPLC Method Table 5. HDL-C values determined before and after ultracentrifugation (n 18). HPLC Reference Method Before UC a After UC Mean, mg/l SD, mg/l Maximum, mg/l Minimum, mg/l a UC, ultracentrifugation. methods were highly correlated (r 0.985) and produced a linear regression line of y 1.04x 52.9 (n 18), as shown in Fig. 4B. The HPLC method gave a mean bias of 20.0% vs the reference method for group C samples, and the bias was positively higher than for samples with TG concentrations within the reference interval. Fig. 5C shows the HDL-C assay biases plotted as a function of serum TG concentrations, suggesting that there is correlation (r 0.720; P 0.001) between HDL-C assay bias and TG concentration. We also compared the HDL-C values obtained by the HPLC method from samples before and after ultracentrifugation at OMC (Table 5). Although statistical analysis by the Student paired t-test gave a significantly higher (not lower) result after ultracentrifugation than before ultracentrifugation, the mean difference (HDL-C results after ultracentrifugation minus HDL-C results before ultracentrifugation) was very small (16 mg/l). If TG-rich lipoproteins of d kg/l had made the HPLC method overestimate the HDL-C values, the results after ultracentrifugation would have been lower than those before ultracentrifugation. The HPLC method for HDL-C, therefore, was considered less affected by TG-rich lipoproteins. Discussion We and our colleagues have used HPLC to obtain lipoprotein profiles for serum or plasma samples from human and nonhuman subjects (15). HPLC methods have been applied to the analysis of lipoproteins in low concentrations, such as the unbound lipoproteins present in the supernatants of mixtures of serum and immunoaffinity gels containing antibodies to apolipoproteins (16), and secreted lipoproteins in the supernatants of HepG2 cell cultures or hepatocytes obtained from the Watanabe heritable hyperlipidemic rabbit (17). The main advantages of the HPLC method are direct profiling of the lipoprotein cholesterol distribution in a small sample volume (5 L) without special techniques, and automatic calculation of the major lipoprotein cholesterol concentrations and particle sizes of LDL and HDL in a single analytical run of 16 min. In Japan, some clinical laboratories have begun to use this method, and the need to assess the precision and accuracy of the HPLC method led to this study. It has been difficult to compare the HPLC method with the recommended reference methods because the reference methods require relatively large sample volumes and are time-consuming and labor-intensive. In addition, reference methods have not been established for VLDL-C, CM-C, LDL size, and HDL size. In the precision study, the four HPLC systems met the performance goals for measurement of TC (CV 3%), HDL-C (CV 4% at HDL-C 420 mg/l), and LDL-C (CV 4%) required by the recent National Cholesterol Education Program (NCEP) guidelines (18, 19). It was, however, impossible to assess the precision of measurements of VLDL-C, CM-C, and particle sizes by the HPLC method because at present there are no NCEP guidelines for these assays. Our results did reveal that the HPLC method produced better precision (mean CV 4.6%) for the VLDL-C assay than the single vertical spin auto profiler (CV 4.8%) (20) and fast lipoprotein chromatography (CV 5.8%) (21). In our between-instrument comparison study, a small variation was found in determinations of lipoprotein cholesterol concentrations and particle sizes. However, we did not find a tendency for column-load numbers to have a direct effect on the lipoprotein profile. The precision remained high for column-load numbers in a wide range from 250 to According to the NCEP guidelines for TC measurements, the goals for total error and absolute mean bias are 8.9% and 3%, respectively (19). The results in the accuracy study revealed that all HPLC systems, with columns with different column-load numbers, met the NCEP performance criteria for the TC assay when the same calibrator (lot no ) was used, and suggested that the recovery of lipoproteins from columns was satisfactory. The NCEP performance goals for HDL-C require a total error of 13% and an absolute mean bias of 5% (18). These criteria were not met by three HPLC systems, which had total errors 13.6% and absolute mean biases 8.4%. Considering the high precision of the HPLC method for HDL-C (CV 3.3% at HDL-C 420 mg/l), the disagreement between the reference and HPLC methods must have been caused by systematic error, not random error. The ratio of HDL-C to TC in the calibrator was 36.9%, calculated from the reference values, whereas the ratio of HDL area to total area of the HPLC pattern obtained by the four instruments was 38.6% on average. All HPLC systems overestimated HDL-C in the calibrator, and this positive bias can in part explain the systematic error. When we calculated the HDL-C concentration using a linear calibration curve of the assigned HDL-C value plotted against the area corresponding to the HDL in the calibrator, the absolute mean bias was 5.2% instead of 12.3%. Although we confirmed that the traceability of the HPLC method to the reference method for HDL-C was improved by use of the reference material mentioned above, the problem remains as to whether the HDL classified by HPLC is the same as the HDL classified by the reference method. The HDL separated by the HPLC

9 Clinical Chemistry 46, No. 1, may be different from that in the supernatant obtained after precipitation with the reference method. HDL particles are heterogeneous in density, size, charge, and apolipoprotein component. We have reported previously that the phosphotungstic acid/mgcl 2 precipitation assay underestimated HDL-C values relative to the HPLC method because of the high concentration of MgCl 2 in a commercial kit, whereas a homogeneous assay produced values close to the HPLC method (10). On the other hand, heparin-mn 2 reagent is not likely to precipitate HDL subclasses such as apolipoprotein E-containing HDL particles (22). Further studies with individual samples will be required to answer this question. When TG concentrations were increased, the TC assay bias was acceptable, at 3%, with some exceptions: the HPLC method became negatively biased, especially in the presence of increasing CM-C concentrations. This finding indicates that the determination of CM-C by the HPLC method is incomplete, probably because of inadequate enzymatic detection of CM-C. On the other hand, HDL-C concentrations were significantly overestimated by the HPLC method compared with the reference method, and the positive bias was correlated with serum TG concentration. Ultracentrifugation to remove TG-rich lipoproteins prior to analysis, however, did not eliminate this problem. We therefore concluded that the HPLC method for HDL-C was affected only minimally by the presence of TG-rich lipoproteins. Methods based on different principles will produce different values. Wiebe and Smith (23) described how it might be possible to adjust reagent concentrations, ionic strength, buffer ph, or other characteristics of the HDL isolation reagents so that the HDL-C values would mimic those obtained with another method. It is important in clinical sites that values obtained by new methods be close to the target values. In terms of standardization of HDL-C measurement, the accuracy should be based on the CDC reference method, which has yielded coronary heart disease risk estimations and population distributions. In summary, the between-instrument variation in serum lipoprotein analysis by the HPLC method was confirmed to be very small. This method met the NCEP criteria for the TC assay. As for the HDL-C assay, the values obtained with the HPLC method were significantly higher than those obtained with the reference method, in spite of good correlation between the two methods. The reason for this difference is not clear at present. However, the HPLC method gives considerable information, such as size distribution of heterogeneous lipoproteins, in a chromatogram, as shown in Fig. 1. We believe that additional information that other methods do not provide is present on the chromatogram. Analysis of the chromatogram for both quantitative and qualitative information may be useful for studies on the serum lipoproteins that play a vital role in the development of atherosclerosis. We thank Drs. Shizuya Yamashita and Yuji Matsuzawa of Osaka University Medical School for useful discussions and the gift of a sample. We also thank Noriko Sato of SRL Inc. and Norikazu Komoriya of Yamagata University for technical assistance. References 1. Miller GJ, Miller NE. Plasma-high-density-lipoprotein concentration and development of ischaemic heart-disease. Lancet 1975;1: Wilson PW, Abbott RD, Castelli WP. High density lipoprotein cholesterol and mortality. The Framingham Heart Study. Arteriosclerosis 1988;8: National Cholesterol Education Program. Second report of the expert panel on detection, evaluation, and treatment of high blood cholesterol in adults (Adult Treatment Panel II). Circulation 1994; 89: Austin MA, Breslow J, Hennekens C, Buring J, Willett W, Krauss RM. Low-density lipoprotein subclass patterns and risk of myocardial infarction. JAMA 1988;260: Gardner CD, Fortmann SP, Krauss RM. Association of small low-density lipoprotein particles with the incidence of coronary artery disease in men and women. JAMA 1996;276: Hara I, Okazaki M. High-performance liquid chromatography of serum lipoproteins. Methods Enzymol 1986;129: Okazaki M, Sasamoto K, Muramatsu T, Hosaki S. Analysis of plasma lipoproteins by gel permeation chromatography. In: Rifai N, Warnick GR, Dominiczak MH, eds. Handbook of lipoprotein testing. Washington, DC: AACC Press, 1997: Sasamoto K, Okazaki M, Muramatsu T, Yanagisawa T, Ando Y, Kamei S, et al. Quantitation of total cholesterol and HDL-cholesterol by the HPLC method: comparison with the automated enzymatic method and the precipitation method. Jpn J Clin Chem 1996;25: Okazaki M, Sasamoto K, Muramatsu T, Jitsukata K, Horiuchi K, Kubono K. Quantitation of HDL-cholesterol by the improved HPLC method: comparison with ultracentrifugation, precipitation and direct methods. Jpn J Clin Chem 1995;24: Okazaki M, Sasamoto K, Muramatsu T, Hosaki S. Evaluation of precipitation and direct methods for HDL-cholesterol assay by HPLC. Clin Chem 1997;43: Sasamoto K, Okazaki M, Muramatsu T, Kawamura K, Kimura N, Kurihara-A Y, et al. HPLC analysis of HDL-cholesterol by precipitation method: comparison between different precipitation reagents. Jpn J Clin Chem 1996;25: Myers GL, Cooper GR, Henderson LO, Hassemer DJ, Kimberly MM. Standardization of lipid and lipoprotein measurements. In: Rifai N, Warnick GR, Dominiczak MH, eds. Handbook of lipoprotein testing. Washington, DC: AACC Press, 1997: Ellerbe P, Myers GL, Cooper GR, Hertz HS, Sniegoski LT, Welch MJ, et al. A comparison of results for cholesterol in human serum obtained by the reference method and by the definitive method of the national reference system for cholesterol. Clin Chem 1990; 36: Wiebe DA, Warnick GR. Measurement of high-density lipoprotein cholesterol. In: Rifai N, Warnick GR, Dominiczak MH, eds. Handbook of lipoprotein testing. Washington, DC: AACC Press, 1997: Yagyu H, Ishibashi S, Chen Z, Osuga J, Okazaki M, Perrcy S, et al. Overexpressed lipoprotein lipase protects against atherosclerosis in apolipoprotein E knockout mice. J Lipid Res 1999;40: Nakajima K, Okazaki M, Tanaka A, Pullinger CR, Wang T, Nakano T, et al. Separation and determination of remnant-like particles in

10 72 Usui et al.: Variations in a HPLC Method human serum using monoclonal antibodies to apo B-100 and apo A-I. J Clin Ligand Assay 1996;19: Tanaka M, Otani H, Yokode M, Kita T. Regulation of apolipoprotein B secretion in hepatocytes from Watanabe heritable hyperlipidemic rabbit, an animal model for familial hypercholesterolemia. Atherosclerosis 1995;114: Warnick GR, Wood PD, for the National Cholesterol Education Program Working Group on Lipoprotein Measurement. National Cholesterol Education Program recommendations for measurement of high-density lipoprotein cholesterol: executive summary. Clin Chem 1995;41: Bachorik PS, Ross JW, for the National Cholesterol Education Program Working Group on Lipoprotein Measurement. National Cholesterol Education Program recommendations for measurement of low-density lipoprotein cholesterol: executive summary. Clin Chem 1995;41: Chung BH, Segrest JP, Cone JT, Pfau J, Geer JC, Duncan LA. High resolution plasma lipoprotein cholesterol profiles by a rapid, high volume semi-automated method. J Lipid Res 1981;22: März W, Rüdiger S, Hunbert S, Seiffert UB, Gross W. Fast lipoprotein chromatography: new method of analysis for plasma lipoproteins. Clin Chem 1993;39: Gibson JC, Rubinstein A, Brown WV. Precipitation of apo E-containing lipoproteins by precipitation reagents for apolipoprotein B. Clin Chem 1984;30: Wiebe DA, Smith SJ. Six methods for isolating high-density lipoprotein compared, with use of the reference method for quantifying cholesterol in serum. Clin Chem 1985;31: