- Email: [email protected]
A New Insulin Immunoassay Specific for the Rapid-Acting Insulin Analog, Insulin Aspart, Suitable for Bioavailability, Bioequivalence, and Pharmacokinetic Studies
Clinical Biochemistry, Vol. 33, No. 8, 627– 633, 2000 Copyright © 2001 The Canadian Society of Clinical Chemists Printed in the USA. All rights reserved 0009-9120/00/$–see front matter
PII 0009-9120(00)00183-1
A New Insulin Immunoassay Specific for the Rapid-Acting Insulin Analog, Insulin Aspart, Suitable for Bioavailability, Bioequivalence, and Pharmacokinetic Studies LENNART ANDERSEN, PEER NOBERT JØRGENSEN, LISBETH BJERRING JENSEN, and DECLAN WALSH Novo Nordisk A/S, Copenhagen, Denmark Objectives: To validate a specific enzyme-linked immunosorbent assay for the rapid-acting human insulin analogue, insulin aspart, in human serum, human plasma, and porcine plasma. Design and methods: For the enzyme-linked immunosorbent assay, two murine monoclonal antibodies were developed that bind to two different epitopes on the insulin aspart molecule. Key parameters for validation were imprecision, accuracy, matrix effects, dilution-linearity, and cross-reactivity. Results: No cross-reactivity was found with human and porcine insulin, human proinsulin, or human C-peptide. The assay is sensitive (limit of quantification ⫽ 11.5 pmol/L), accurate (95–107% recovery with human serum, human plasma, and porcine plasma in the range 16 – 800 pmol/L), and has a 14.7% total imprecision within the entire analytical range. Dilution of samples gave linear results with human serum as the diluent. Conclusions: The insulin aspart-specific enzyme-linked immunosorbent assay described in this study is well suited to study the bioavailability, bioequivalence, and pharmacokinetics of this insulin analogue. Copyright © 2001 The Canadian Society of Clinical Chemists
KEY WORDS: monoclonal antibodies; enzyme-linked immunosorbent assay; ELISA; insulin analogue assay; specific insulin aspart assay.
Introduction nsulin aspart—Asp(B28)-human insulin—is an analogue of human insulin in which the proline at position 28 of the B chain is replaced by aspartic acid. This change reduces spontaneous self-association into hexamers in neutral solution (1), which is a problem with human insulin in that it delays the absorption of subcutaneously injected insulin (2). Previously the lack of a specific immunoassay for quantification of insulin aspart in serum or plasma samples has led to the use of radioimmunoassays
I
(RIAs) that cannot distinguish between endogenous insulin and insulin aspart (3,4). In addition, the human radioimmunoassay (RIA) did not measure insulin aspart linearly so a correction for nonlinearity was necessary to obtain reliable estimates of the insulin aspart concentration (4). The new specific enzyme-linked immunosorbent assay (ELISA) described here involves two monoclonal antibodies (MCA), one that binds to an epitope common for human insulin and insulin aspart, and the other which binds to an epitope specific for insulin aspart. The studies outlined below were developed to validate the new ELISA for the quantification of insulin aspart in bioavailability, bioequivalence, and pharmacokinetic investigations. The validated key parameters presented in this paper are imprecision, accuracy, matrix effects, dilution-linearity, and cross-reactivity for substances, which are relevant in connection with preclinical and clinical studies. Possible interference by insulin antibodies was not investigated by using this insulin aspart ELISA as it was shown in an earlier assay (using the same buffers and employing two insulin antibodies in the same concentrations as this assay) that the measured insulin was similar to free insulin concentrations at low insulin-antibody binding capacity (⬍ 25% binding) (5). Finally, a comparison was needed with the currently employed nonspecific RIA (4). Materials and methods MONOCLONAL
ANTIBODIES
Correspondence: Lennart Andersen, Novo Nordisk A/S, Hagedornsvej 1 (HAB3.66), DK-2820 Copenhagen, Denmark. E-mail: [email protected] Manuscript received May 30, 2000; revised and accepted October 11, 2000.
The monoclonal antibodies (MCA) X14-6F34 and HUI-018 were developed by hybridoma technology from mice immunized with insulin aspart and human insulin, respectively. The epitope for X14-6F34 is near the C-terminal end of the B-chain. The affinity for insulin aspart is 107 L/mol. The antibody
CLINICAL BIOCHEMISTRY, VOLUME 33, NOVEMBER 2000
627
ANDERSEN
does not bind human insulin. The epitope for HUI018 is centred around the A-loop. The affinity for human insulin is 108 L/mol. The epitopes were determined in a competitive ELISA by comparing 50% inhibition concentrations of human insulin and insulin analogues with single amino acid substitutions. PEPTIDES Insulin aspart, human insulin, porcine insulin, human proinsulin, and human C-peptide were from Novo Nordisk (Copenhagen, Denmark). ASSAY
PROCEDURE
The insulin aspart ELISA was performed at room temperature. Microtiter wells were coated overnight with 100 L coating buffer containing 10 mg/L HUI-018 specific for human and aspart insulin. The coated wells were emptied and blocked with 200 L of blocking buffer for 1 h with agitation and washed four times with washing buffer. The insulin determination was performed in duplicate. To each well was added 25 L calibrator or serum/plasma samples and 75 L of antibody-detecting buffer with 0.5–2 mg/L biotinylated X14-6F34 specific for insulin aspart. All calibrators were prepared in a pool of fasted human serum, which was also used for the dilution of samples with a high insulin aspart concentration. After 20- to 24-h incubation with shaking, the plates were washed four times. After this, 100 L of 0.5 mg/L avidin D-horseradish peroxidase (Vector Laboratories, Burlingame, CA, USA; A-2004) in conjugate buffer was applied to each well. After agitated incubation for 30 min the plates were washed four times. Enzyme activity in wells was estimated by adding 100 L/well of 3,3⬘,5,5⬘-tetramethylbenzidine (Sigma Chemical Co., St. Louis, Mo, USA; T-3405) in substrate solution. After shaking for 10 min, the reaction was terminated by the addition of 100 L of 4 mol/L H3PO4. Absorbance values were read at 450 nm. All buffers, chemicals, and the biotinylation method were as previously described for insulin ELISA (5). DAKO insulin ELISA (DAKO Diagnostics, Denmark House, Ely, Cambridgeshire, UK) was used in compliance with the kit instructions.
ET AL.
ASSAY
IMPRECISION AND ACCURACY
Human serum and heparin plasma samples from five normal donors (two male, three female) and heparin plasma samples from five pigs were used. Each sample was spiked with insulin aspart at four concentration levels: ⬃16, 170, 400, and 700 pmol/L. All samples were analyzed twice in double determinations. This procedure was repeated three times in different assays on separate days. The total number of double determinations was 480. ASSAY
SENSITIVITY
The lower limit of quantification (LOQ) is defined as the lowest concentration (Z) that can be measured with a coefficient of variation (CV) of 20% (CV ⫽ SDZ/Z ⫽ 0.2). The imprecision results were used to calculate the LOQ. LOQ was also validated independently by using 15.7 pmol/L insulin aspart-spiked serum samples from seven healthy subjects. These samples were analyzed twice in each of the five assay runs. DILUTION
LINEARITY
Human serum samples from five normal donors spiked with ⬃700 pmol/L insulin aspart were diluted with a pool of human serum to reach relative concentrations of ⬃1.0, 0.5, 0.25, 0.125, and 0.0625. Each sample was analyzed in double determinations in each assay. ASSAY
SELECTIVITY
Human serum samples were spiked with insulin aspart in five concentrations ranging from 20 to 646 pmol/L. In addition, human serum samples were spiked with the peptides human insulin, porcine insulin, human proinsulin, and human C-peptide, respectively, at 20 to 200,000 pmol/L. An equal volume of each serum containing insulin aspart was mixed with an equal volume of serum spiked with each peptide. The effect of the peptides on the insulin aspart measurements was expressed as the difference between measured concentration in the sample with and without the tested peptide. The difference is expressed as a percentage of the concentration mean of four duplicate measurements in samples without the tested peptide. COMPARISON
SPIKING
WITH INSULIN RADIOIMMUNOASSAY
OF SERUM AND PLASMA SAMPLES WITH INSULIN
The insulin aspart used for spiking was a highperformance liquid chromatography (HPLC) standard defined as 1.00 IU/6.00 nmol (6,7). All volumes used in the preparation of spiked samples were weighed and the concentrations were calculated. Each spiked sample was stored at ⫺18 °C until assay.
By reanalysing serum samples from a previous clinical trial (8) the insulin aspart-specific ELISA was compared to the RIA used previously (Pharmacia insulin RIA 100, Pharmacia Diagnostics, Uppsala, Sweden). The samples from each subject were taken at 28 time points between ⫺90 and 600 min after injection. The RIA measures both endogenous human insulin and insulin aspart. Therefore, the sum of insulin
628
CLINICAL BIOCHEMISTRY, VOLUME 33, NOVEMBER 2000
ASPART
INSULIN ASPART-SPECIFIC ELISA
Figure 1 — Calibration curve for the insulin aspart ELISA. The fitted logistic curve was: Absorbance ⫽ D ⫹ (A ⫺ D)/[1 ⫹ (conc/C)B]. The calibrator curve follows a logistic function, which includes the zero calibrator point. Conc ⫽ calibrator concentration; D ⫽ estimated blank absorbance; A ⫽ estimated asymptotic absorbance; B ⫽ slope parameter; C ⫽ turning point.
aspart measured by insulin aspart ELISA and the endogenous human insulin measured by DAKO insulin ELISA was used to make a fair comparison. The DAKO assay was tested for cross-reaction with insulin aspart by spiking human serum with human insulin at the concentration levels 125, 250, and 500 pmol/L. These were further spiked with 2 ⫻ 101, 2 ⫻ 102, 2 ⫻ 103, 2 ⫻ 104, and 2 ⫻ 105 pmol/L insulin aspart. Human insulin was measured in the samples both with and without insulin aspart but with the same human insulin concentration. The effect of insulin aspart on the human insulin measurements was calculated as mentioned under Assay Selectivity.
Figure 2 — Assay results from the imprecision and accuracy experiments plotted against the spiked insulin aspart concentration. It is apparent that the analytical error is proportional to the spiked concentration.
pmol/L, respectively, in a human serum pool. The CVs of the back fit of the calibrator insulin aspart concentrations from 44 consecutive assays on 19 different days were 8.3, 4.2, 4.0, 3.3, 6.9, 2.7, and 2.8%, respectively. ASSAY
IMPRECISION
Figure 1 shows a typical calibration curve. A four parameter logistic function (MultiCalc v.2.50; Wallac Oy, Turku, Finland), shown as the continuous curve, gives an excellent fit for the absorbance of the eight calibrators consisting of insulin aspart spiked at 0, 22.3, 68.1, 135.7, 179.2, 234.0, 552.1, and 906.7
The ratio between the measured and spiked concentration was analyzed statistically according to the ANOVA model yhijk ⫽ ␣h ⫹ i ⫹ ␦j ⫹ k ⫹ ⑀hijk: where ␣h is the mean response for spike level h; i is the mean response for matrix i; ␦j is the random effect of variation in assay condition; k is the random effect of variation between samples; and ⑀hijk is the residual random. The measurements showed a proportional increasing variation with increasing spiked concentration (Figure 2). Thus the analysis was based on measured concentration divided by the spiked concentration. The data revealed some inconsistency mainly due to one pig plasma sample which showed unusually high concentrations and was suspected to be contaminated with insulin aspart. No results from this individual were used. Tests for the statistical analysis showed variance homogeneity between matrix, spiking levels, and assay setups. A variance component analysis was performed; the variance components corresponding to residual, assay and sample variation were 0.104, 0.053, and 0.089, respectively. The CV of within-assay repeatability (residual) and total imprecision (sample, assay and residual) were estimated as 10.4% and 14.7%, respectively (range 16 –729 pmol/L). Table 1 shows the within-assay and total imprecision (SD) in the normal concentration range for human insulin.
CLINICAL BIOCHEMISTRY, VOLUME 33, NOVEMBER 2000
629
STATISTICAL
ANALYSIS
SAS version 6.11 Proc Univariate and Proc Mixed was used for calculation of variance statistics. APL version 22 was used for regression analysis and calculation of the mean and standard deviation (SD). Results ASSAY
CHARACTERISTICS
ANDERSEN
ET AL.
TABLE 1 Within-Assay and Total Assay Imprecision in the Normal Concentration Range for Human Insulin Normal rangea (Human Insulin) Lowb Lowc Upperc Abovec
Insulin Aspart (pmol/L)
Within-Assay Run SD (pmol/L)
Total SD (pmol/L)
Total CV (%)
15.7 16.0 100 500
1.74 1.66 10.4 52.0
2.25 2.35 14.7 73.5
14.3 14.7 14.7 14.7
a
Insulin aspart has no normal range as it does not normally occur in the blood. Estimated by using samples spiked with 15.7 pmol/L. c Estimated by using a statistical method based on concentration ratios. b
ASSAY
ACCURACY
The analysis of accuracy was tested according to the model outlined under Assay Imprecision and Accuracy. The hypothesis of homogeneity of the mean between spiking levels was rejected at a 0.1% significance level by standard F-test. The mean at lowest, low middle, high middle, and highest spiking level was 1.05 (95% confidence interval [CI] 0.98⫺1.12), 1.00 (95% CI 0.93⫺1.07), 1.00 (95% CI 0.94⫺1.07), and 0.99 (95% CI 0.92⫺1.05), respectively. The hypothesis of homogeneity of the mean between matrices was rejected at a 5% significance level by standard F-test. The mean for human serum, human plasma, and pig plasma was 1.01 (95% CI 0.92⫺1.10), 0.95 (95% CI 0.86⫺1.04), and 1.07 (95% CI 0.97⫺1.17), respectively. ASSAY
SENSITIVITY
The estimated total SD of the insulin aspart concentrations measured in samples spiked with 16 pmol/L insulin aspart is 2.352 pmol/L corresponds to CV 14.7% according to the imprecision experiment, giving a LOQ of 11.8 pmol/L. The variance components analysis of the measurements on seven individual serum samples spiked with 15.7 pmol/L insulin aspart showed residual, assay, and sample variation to be 3.025, 0.278, and 1.748 (pmol/L)2, respectively. SD of within-assay repeatability and total imprecision was estimated as 1.739 and 2.247 pmol/L, respectively, leading to an LOQ of 11.2 pmol/L. The mean value of LOQ is 11.5 pmol/L. DILUTION
LINEARITY AND RECOVERY
Table 2 shows the recovery of insulin aspart expressed as percentage of the amount of insulin aspart in the samples. The recovery of insulin aspart is linear within the dilution range. No statistically significant (p ⫽ 0.49) slope was found by regression analysis. The mean recovery was 94.9% with 7.3% SD. ASSAY
SELECTIVITY
Table 3 shows the effect of human insulin, porcine insulin, human proinsulin, and human 630
C-peptide on insulin aspart quantification. It appears that the effects do not depend on the insulin aspart concentration but depend only on the nature and the concentration of the added peptide. As shown in Table 3, human insulin, porcine insulin, and human proinsulin have a minor or negative effect on the insulin aspart measurements. Human C-peptide has a positive effect on the insulin aspart measurements at a concentration of 105 pmol/L. COMPARISON
WITH INSULIN RADIOIMMUNOASSAY
The DAKO insulin ELISA used to measure the endogenous human insulin in the serum samples from the clinical study was not significantly affected by the insulin aspart administered to the patients. The mean effect of insulin aspart on human insulin recovery in the tested samples was 1.2, 2.2, 7.2, ⫺6.6, and ⫺59.8% with insulin aspart concentrations of 2 ⫻ 101, 2 ⫻ 102, 2 ⫻ 103, 2 ⫻ 104, and 2 ⫻ 105 pmol/L, respectively. Comparison of the RIA and combined ELISA results from the analyzed study showed very close agreement in all cases. Log-transformation of the two analytical data sets secured homogeneity of variance. The RIA and the ELISA results were compared by using linear regression (Figure 3) of the differences between the log-concentrations vs. the sum of their log-concentrations (9). No statistically significant slope (p ⬎ 0.4) was found; therefore, the mean difference and the 95% CI could be calculated. Back transformed, the mean difference TABLE 2 Percent Recovery after Dilution of Insulin Aspart-Spiked Human Serum Samples Fold Dilution Sample ID 1 ID 2 ID 3 ID 4 ID 5 Mean SD
1
2
4
8
16
93.9% 92.2% 90.9% 93.8% 105.2% 95.2% 5.7%
96.2% 86.7% 88.6% 98.2% 106.9% 95.3% 8.1%
92.6% 91.4% 90.2% 93.9% 105.3% 94.7% 6.1%
88.4% 84.4% 86.1% 86.8% 103.3% 89.8% 7.7%
91.5% 95.0% 93.3% 110.3% 104.8% 99.0% 8.2%
CLINICAL BIOCHEMISTRY, VOLUME 33, NOVEMBER 2000
INSULIN ASPART-SPECIFIC ELISA
TABLE 3 Difference Between Measured and Expected Insulin Aspart Concentration Expressed as a Percentage of Expected Concentration Cross-Reacting Peptide Human insulin
Porcine insulin
Human proinsulin
Human C-peptide
Concentration (pmol/L) of Cross-Reacting Peptide
Expected Insulin Aspart (pmol/L)
10
102
103
104
105
10.4 23.5 93.3 187.4 322.9 10.4 23.5 93.3 187.4 322.9 10.4 23.5 93.3 187.4 322.9 10.4 23.5 93.3 187.4 322.9
⫺13.3% 1.4% 9.3% ⫺2.3% ⫺4.6% ⫺1.7% 1.0% 1.0% 4.6% ⫺9.2% ⫺2.7% 5.6% 1.6% ⫺4.5% ⫺6.1% 7.8% 3.9% ⫺0.3% ⫺1.8% ⫺2.1%
⫺18.0% ⫺0.7% ⫺9.7% ⫺2.9% 10.3% ⫺1.7% 1.0% ⫺4.4% ⫺3.9% 2.5% ⫺2.7% 2.7% ⫺7.0% ⫺5.0% 3.6% ⫺1.7% 1.4% 1.3% ⫺3.9% ⫺3.4%
⫺13.3% ⫺4.2% ⫺5.1% ⫺5.5% ⫺7.4% 0.2% ⫺11.4% ⫺0.7% ⫺2.3% ⫺5.2% 1.2% 4.4% ⫺4.8% ⫺5.5% ⫺14.5% 2.2% 6.9% ⫺1.7% 1.3% ⫺3.7%
⫺35.4% ⫺36.1% ⫺16.6% ⫺15.2% ⫺15.8% 7.0% ⫺16.5% ⫺15.0% ⫺14.1% ⫺18.8% ⫺15.2% ⫺13.1% ⫺16.8% ⫺18.4% ⫺17.0% 21.4% ⫺14.0% 0.9% 0.9% 0.0%
⫺100% ⫺90.6% ⫺86.9% ⫺77.8% ⫺76.9% ⫺93.3% ⫺85.1% ⫺81.1% ⫺80.1% ⫺78.5% ⫺100% ⫺82.5% ⫺77.4% ⫺76.5% ⫺75.3% 145% 60.2% 14.7% 12.6% 12.7%
1
of the ratio between the ELISA and RIA results was 0.034 (95% CI 0.014⫺0.054). The 95% CI does not include 0% indicating a significant bias. This was below 5%, however, and therefore not likely to be relevant.
The mean insulin profiles measured by ELISA and RIA are shown in Figure 4. It appears that the insulin measured by the ELISA and the RIA give similar results independent of the kinetic time course.
Figure 3 — Comparison between the combined results with specific ELISAs and the results with the insulin RIA 100 (Pharmacia) for measurements of insulin aspart ⫹ human insulin. These are plotted as the difference between the loge concentration from the two assay methods vs. the sum of the loge concentrations from the two assays. Only samples with insulin aspart ⬎ 10 pmol/L are included.
Figure 4 — Comparison of the total insulin (insulin aspart ⫹ human insulin) measured with ELISA or RIA. Samples from 10 individuals were taken between ⫺90 and 600 min after insulin aspart injection. Insulin concentrations were measured with either method and are shown as mean concentration and SD. It is apparent that the concentration difference in the kinetic time course between the two methods is insignificant.
CLINICAL BIOCHEMISTRY, VOLUME 33, NOVEMBER 2000
631
ANDERSEN
Discussion The insulin aspart-specific ELISA as described in this study proved to be suitable for preclinical and clinical studies on the bioavailability, bioequivalence, and pharmacokinetics of insulin aspart. It may also be useful in other studies, especially when endogenous human insulin and insulin aspart need to be measured separately. Eight insulin aspart calibrators in the range 0 to 1000 pmol/L and the use of a four-parameter logistic dose-response model is sufficient to establish an accurate calibration curve with acceptable imprecision. The data used to evaluate imprecision and accuracy showed proportionality between SD and the measured concentration. Variance homogeneity between spiking levels was secured by using the measured concentration divided by the spiked concentration. By this normalization of the data the statistical analysis was performed in the most optimal way, giving it more statistical power compared to the analysis of each spiking level separately. The imprecision analysis showed a constant CV between 16 pmol/L and the upper end of the calibrator curve. The CV of 14.7% in this assay range is acceptable for bioavailability, bioequivalence, and pharmacokinetic studies (10) because these parameters vary more between individuals. It was proven that the LOQ of 11.5 pmol/L could be calculated using the statistically estimated 14.7% CV at 16 pmol/L because the experiment with repeated measurements of seven individual insulinspiked samples at 15.7 pmol/L gave a non-significantly different LOQ value. In addition, this indicates that data below the lowest nonzero calibrator and the LOQ are accurate and precise. Even although the statistical analysis showed a significant accuracy bias with regard to both spiking levels and matrices, it was relatively low and within acceptable limits (10) for both parameters. Dilution of samples up to 16-fold with insulin aspart-free human serum does not affect the accuracy. Effects of human insulin, porcine insulin, human proinsulin, or human C-peptide on the insulin aspart quantification in samples are small and generally below the CV of the ELISA and independent of the concentration of insulin aspart at physiologic concentrations. However, if the concentration rises to 10,000 pmol/L or higher the cooperative effects become pronounced. The negative effects on insulin aspart measurements seen with 10,000 pmol/L to 100,000 pmol/L human insulin, porcine insulin, or human proinsulin may arise from competition between binding of insulin aspart and these cross-reactive insulins for the common epitope on the monoclonal antibody coated on the microtiter plate. There is no good explanation as to why human C-peptide showed positive effects above 10,000 pmol/L as this pep632
ET AL.
tide is not known to bind to insulin aspart or to share any homologous epitopes. Most of the effects regarding interference, as seen with peptides with homologous insulin epitopes, are due to the assay design. The design of this assay is based on a relatively low affinity (about 107 L/mol) of the detecting antibody, which only binds insulin aspart. Attempts to make the assay more accurate were not successful. Too much variability was found in the assay when the detection antibody was used as catcher antibody or when a washing step was introduced between sample incubation and detection antibody addition. Furthermore, adjustments to the assay were limited as there was only one monoclonal antibody available that allowed discrimination between two insulins with only one amino-acid difference. The comparison of the RIA and the combined ELISA results showed good agreement between the two methods with only a small difference (⬍5%). This difference is negligible in the context of clinical studies. The insulin aspart ELISA is available from MediLab A/S (Adelgade 5, DK-1304, Copenhagen, Denmark). The combination of the specific DAKO insulin ELISA, which measures endogenous human insulin only, and this new insulin aspart ELISA, which does not measure human insulin, may prove to be useful for studies in which both types of insulin may need to be assessed individually. Acknowledgement The authors thank Aage Vølund for critically reading the manuscript and Rene Tabanera y Palacios for performing the statistical analysis in SAS (both from the Statistics-Diabetes Department, Novo Nordisk). We also thank Birgitte Skafte and Birgitte Green Holm (Immunochemistry Department, Novo Nordisk) for performing the DAKO insulin ELISA analysis.
References 1. Brange J, Ribel U, Hansen JF, et al. Monomeric insulins obtained by protein engineering and their medical implications. Nature 1988; 333: 679 – 82. 2. Barnett AH, Owens DR. Insulin analogues. Lancet 1997; 349: 47–51. 3. UK Prospective Diabetes Study (UKPDS) XI. Biochemical risk factors in type 2 diabetic patients at diagnosis compared with age-matched normal subjects. Diabetic Med 1994; 11: 534⫺54. 4. Andersen L, Vølund A, Olsen KJ, Plum A, Walsh D. Validity and use of a non-parallel insulin assay for pharmacokinetic studies of a rapid-acting insulin analog, insulin aspart. J Immunoassay. In press. 5. Andersen L, Dinesen B, Jørgensen PN, Poulsen F, Røder ME. Enzyme immunoassay for intact human insulin in serum or plasma. Clin Chem 1993; 39: 578 – 82. 6. Vølund A. Conversion of insulin units to SI units [Letter]. Am J Clin Nutr 1993; 58: 714 –15. CLINICAL BIOCHEMISTRY, VOLUME 33, NOVEMBER 2000
INSULIN ASPART-SPECIFIC ELISA
7. Vølund A, Brange J, Drejer K, et al. In vitro and in vivo potency of insulin analogues designed for clinical use. Diabetic Med 1991; 8: 839 – 47. 8. Heinemann L, Weyer C, Rauhaus S, Heinrichs S, Heise T. Variability of the metabolic effect of soluble insulin and the rapid acting insulin analogue insulin aspart. Diabetes Care 1998; 21: 1910⫺14.
9. Bland JM, Altman DG. Statistical method for assessing agreement between two methods of clinical measurement. Lancet 1986; 1: 307⫺10. 10. Shah VP, Midha KK, Dighe S, et al. Analytical methods validation: Bioavailability, bioequivalence and pharmacokinetic studies. Pharm Res 1992; 9: 588 –92.
CLINICAL BIOCHEMISTRY, VOLUME 33, NOVEMBER 2000
633
PII 0009-9120(00)00183-1
A New Insulin Immunoassay Specific for the Rapid-Acting Insulin Analog, Insulin Aspart, Suitable for Bioavailability, Bioequivalence, and Pharmacokinetic Studies LENNART ANDERSEN, PEER NOBERT JØRGENSEN, LISBETH BJERRING JENSEN, and DECLAN WALSH Novo Nordisk A/S, Copenhagen, Denmark Objectives: To validate a specific enzyme-linked immunosorbent assay for the rapid-acting human insulin analogue, insulin aspart, in human serum, human plasma, and porcine plasma. Design and methods: For the enzyme-linked immunosorbent assay, two murine monoclonal antibodies were developed that bind to two different epitopes on the insulin aspart molecule. Key parameters for validation were imprecision, accuracy, matrix effects, dilution-linearity, and cross-reactivity. Results: No cross-reactivity was found with human and porcine insulin, human proinsulin, or human C-peptide. The assay is sensitive (limit of quantification ⫽ 11.5 pmol/L), accurate (95–107% recovery with human serum, human plasma, and porcine plasma in the range 16 – 800 pmol/L), and has a 14.7% total imprecision within the entire analytical range. Dilution of samples gave linear results with human serum as the diluent. Conclusions: The insulin aspart-specific enzyme-linked immunosorbent assay described in this study is well suited to study the bioavailability, bioequivalence, and pharmacokinetics of this insulin analogue. Copyright © 2001 The Canadian Society of Clinical Chemists
KEY WORDS: monoclonal antibodies; enzyme-linked immunosorbent assay; ELISA; insulin analogue assay; specific insulin aspart assay.
Introduction nsulin aspart—Asp(B28)-human insulin—is an analogue of human insulin in which the proline at position 28 of the B chain is replaced by aspartic acid. This change reduces spontaneous self-association into hexamers in neutral solution (1), which is a problem with human insulin in that it delays the absorption of subcutaneously injected insulin (2). Previously the lack of a specific immunoassay for quantification of insulin aspart in serum or plasma samples has led to the use of radioimmunoassays
I
(RIAs) that cannot distinguish between endogenous insulin and insulin aspart (3,4). In addition, the human radioimmunoassay (RIA) did not measure insulin aspart linearly so a correction for nonlinearity was necessary to obtain reliable estimates of the insulin aspart concentration (4). The new specific enzyme-linked immunosorbent assay (ELISA) described here involves two monoclonal antibodies (MCA), one that binds to an epitope common for human insulin and insulin aspart, and the other which binds to an epitope specific for insulin aspart. The studies outlined below were developed to validate the new ELISA for the quantification of insulin aspart in bioavailability, bioequivalence, and pharmacokinetic investigations. The validated key parameters presented in this paper are imprecision, accuracy, matrix effects, dilution-linearity, and cross-reactivity for substances, which are relevant in connection with preclinical and clinical studies. Possible interference by insulin antibodies was not investigated by using this insulin aspart ELISA as it was shown in an earlier assay (using the same buffers and employing two insulin antibodies in the same concentrations as this assay) that the measured insulin was similar to free insulin concentrations at low insulin-antibody binding capacity (⬍ 25% binding) (5). Finally, a comparison was needed with the currently employed nonspecific RIA (4). Materials and methods MONOCLONAL
ANTIBODIES
Correspondence: Lennart Andersen, Novo Nordisk A/S, Hagedornsvej 1 (HAB3.66), DK-2820 Copenhagen, Denmark. E-mail: [email protected] Manuscript received May 30, 2000; revised and accepted October 11, 2000.
The monoclonal antibodies (MCA) X14-6F34 and HUI-018 were developed by hybridoma technology from mice immunized with insulin aspart and human insulin, respectively. The epitope for X14-6F34 is near the C-terminal end of the B-chain. The affinity for insulin aspart is 107 L/mol. The antibody
CLINICAL BIOCHEMISTRY, VOLUME 33, NOVEMBER 2000
627
ANDERSEN
does not bind human insulin. The epitope for HUI018 is centred around the A-loop. The affinity for human insulin is 108 L/mol. The epitopes were determined in a competitive ELISA by comparing 50% inhibition concentrations of human insulin and insulin analogues with single amino acid substitutions. PEPTIDES Insulin aspart, human insulin, porcine insulin, human proinsulin, and human C-peptide were from Novo Nordisk (Copenhagen, Denmark). ASSAY
PROCEDURE
The insulin aspart ELISA was performed at room temperature. Microtiter wells were coated overnight with 100 L coating buffer containing 10 mg/L HUI-018 specific for human and aspart insulin. The coated wells were emptied and blocked with 200 L of blocking buffer for 1 h with agitation and washed four times with washing buffer. The insulin determination was performed in duplicate. To each well was added 25 L calibrator or serum/plasma samples and 75 L of antibody-detecting buffer with 0.5–2 mg/L biotinylated X14-6F34 specific for insulin aspart. All calibrators were prepared in a pool of fasted human serum, which was also used for the dilution of samples with a high insulin aspart concentration. After 20- to 24-h incubation with shaking, the plates were washed four times. After this, 100 L of 0.5 mg/L avidin D-horseradish peroxidase (Vector Laboratories, Burlingame, CA, USA; A-2004) in conjugate buffer was applied to each well. After agitated incubation for 30 min the plates were washed four times. Enzyme activity in wells was estimated by adding 100 L/well of 3,3⬘,5,5⬘-tetramethylbenzidine (Sigma Chemical Co., St. Louis, Mo, USA; T-3405) in substrate solution. After shaking for 10 min, the reaction was terminated by the addition of 100 L of 4 mol/L H3PO4. Absorbance values were read at 450 nm. All buffers, chemicals, and the biotinylation method were as previously described for insulin ELISA (5). DAKO insulin ELISA (DAKO Diagnostics, Denmark House, Ely, Cambridgeshire, UK) was used in compliance with the kit instructions.
ET AL.
ASSAY
IMPRECISION AND ACCURACY
Human serum and heparin plasma samples from five normal donors (two male, three female) and heparin plasma samples from five pigs were used. Each sample was spiked with insulin aspart at four concentration levels: ⬃16, 170, 400, and 700 pmol/L. All samples were analyzed twice in double determinations. This procedure was repeated three times in different assays on separate days. The total number of double determinations was 480. ASSAY
SENSITIVITY
The lower limit of quantification (LOQ) is defined as the lowest concentration (Z) that can be measured with a coefficient of variation (CV) of 20% (CV ⫽ SDZ/Z ⫽ 0.2). The imprecision results were used to calculate the LOQ. LOQ was also validated independently by using 15.7 pmol/L insulin aspart-spiked serum samples from seven healthy subjects. These samples were analyzed twice in each of the five assay runs. DILUTION
LINEARITY
Human serum samples from five normal donors spiked with ⬃700 pmol/L insulin aspart were diluted with a pool of human serum to reach relative concentrations of ⬃1.0, 0.5, 0.25, 0.125, and 0.0625. Each sample was analyzed in double determinations in each assay. ASSAY
SELECTIVITY
Human serum samples were spiked with insulin aspart in five concentrations ranging from 20 to 646 pmol/L. In addition, human serum samples were spiked with the peptides human insulin, porcine insulin, human proinsulin, and human C-peptide, respectively, at 20 to 200,000 pmol/L. An equal volume of each serum containing insulin aspart was mixed with an equal volume of serum spiked with each peptide. The effect of the peptides on the insulin aspart measurements was expressed as the difference between measured concentration in the sample with and without the tested peptide. The difference is expressed as a percentage of the concentration mean of four duplicate measurements in samples without the tested peptide. COMPARISON
SPIKING
WITH INSULIN RADIOIMMUNOASSAY
OF SERUM AND PLASMA SAMPLES WITH INSULIN
The insulin aspart used for spiking was a highperformance liquid chromatography (HPLC) standard defined as 1.00 IU/6.00 nmol (6,7). All volumes used in the preparation of spiked samples were weighed and the concentrations were calculated. Each spiked sample was stored at ⫺18 °C until assay.
By reanalysing serum samples from a previous clinical trial (8) the insulin aspart-specific ELISA was compared to the RIA used previously (Pharmacia insulin RIA 100, Pharmacia Diagnostics, Uppsala, Sweden). The samples from each subject were taken at 28 time points between ⫺90 and 600 min after injection. The RIA measures both endogenous human insulin and insulin aspart. Therefore, the sum of insulin
628
CLINICAL BIOCHEMISTRY, VOLUME 33, NOVEMBER 2000
ASPART
INSULIN ASPART-SPECIFIC ELISA
Figure 1 — Calibration curve for the insulin aspart ELISA. The fitted logistic curve was: Absorbance ⫽ D ⫹ (A ⫺ D)/[1 ⫹ (conc/C)B]. The calibrator curve follows a logistic function, which includes the zero calibrator point. Conc ⫽ calibrator concentration; D ⫽ estimated blank absorbance; A ⫽ estimated asymptotic absorbance; B ⫽ slope parameter; C ⫽ turning point.
aspart measured by insulin aspart ELISA and the endogenous human insulin measured by DAKO insulin ELISA was used to make a fair comparison. The DAKO assay was tested for cross-reaction with insulin aspart by spiking human serum with human insulin at the concentration levels 125, 250, and 500 pmol/L. These were further spiked with 2 ⫻ 101, 2 ⫻ 102, 2 ⫻ 103, 2 ⫻ 104, and 2 ⫻ 105 pmol/L insulin aspart. Human insulin was measured in the samples both with and without insulin aspart but with the same human insulin concentration. The effect of insulin aspart on the human insulin measurements was calculated as mentioned under Assay Selectivity.
Figure 2 — Assay results from the imprecision and accuracy experiments plotted against the spiked insulin aspart concentration. It is apparent that the analytical error is proportional to the spiked concentration.
pmol/L, respectively, in a human serum pool. The CVs of the back fit of the calibrator insulin aspart concentrations from 44 consecutive assays on 19 different days were 8.3, 4.2, 4.0, 3.3, 6.9, 2.7, and 2.8%, respectively. ASSAY
IMPRECISION
Figure 1 shows a typical calibration curve. A four parameter logistic function (MultiCalc v.2.50; Wallac Oy, Turku, Finland), shown as the continuous curve, gives an excellent fit for the absorbance of the eight calibrators consisting of insulin aspart spiked at 0, 22.3, 68.1, 135.7, 179.2, 234.0, 552.1, and 906.7
The ratio between the measured and spiked concentration was analyzed statistically according to the ANOVA model yhijk ⫽ ␣h ⫹ i ⫹ ␦j ⫹ k ⫹ ⑀hijk: where ␣h is the mean response for spike level h; i is the mean response for matrix i; ␦j is the random effect of variation in assay condition; k is the random effect of variation between samples; and ⑀hijk is the residual random. The measurements showed a proportional increasing variation with increasing spiked concentration (Figure 2). Thus the analysis was based on measured concentration divided by the spiked concentration. The data revealed some inconsistency mainly due to one pig plasma sample which showed unusually high concentrations and was suspected to be contaminated with insulin aspart. No results from this individual were used. Tests for the statistical analysis showed variance homogeneity between matrix, spiking levels, and assay setups. A variance component analysis was performed; the variance components corresponding to residual, assay and sample variation were 0.104, 0.053, and 0.089, respectively. The CV of within-assay repeatability (residual) and total imprecision (sample, assay and residual) were estimated as 10.4% and 14.7%, respectively (range 16 –729 pmol/L). Table 1 shows the within-assay and total imprecision (SD) in the normal concentration range for human insulin.
CLINICAL BIOCHEMISTRY, VOLUME 33, NOVEMBER 2000
629
STATISTICAL
ANALYSIS
SAS version 6.11 Proc Univariate and Proc Mixed was used for calculation of variance statistics. APL version 22 was used for regression analysis and calculation of the mean and standard deviation (SD). Results ASSAY
CHARACTERISTICS
ANDERSEN
ET AL.
TABLE 1 Within-Assay and Total Assay Imprecision in the Normal Concentration Range for Human Insulin Normal rangea (Human Insulin) Lowb Lowc Upperc Abovec
Insulin Aspart (pmol/L)
Within-Assay Run SD (pmol/L)
Total SD (pmol/L)
Total CV (%)
15.7 16.0 100 500
1.74 1.66 10.4 52.0
2.25 2.35 14.7 73.5
14.3 14.7 14.7 14.7
a
Insulin aspart has no normal range as it does not normally occur in the blood. Estimated by using samples spiked with 15.7 pmol/L. c Estimated by using a statistical method based on concentration ratios. b
ASSAY
ACCURACY
The analysis of accuracy was tested according to the model outlined under Assay Imprecision and Accuracy. The hypothesis of homogeneity of the mean between spiking levels was rejected at a 0.1% significance level by standard F-test. The mean at lowest, low middle, high middle, and highest spiking level was 1.05 (95% confidence interval [CI] 0.98⫺1.12), 1.00 (95% CI 0.93⫺1.07), 1.00 (95% CI 0.94⫺1.07), and 0.99 (95% CI 0.92⫺1.05), respectively. The hypothesis of homogeneity of the mean between matrices was rejected at a 5% significance level by standard F-test. The mean for human serum, human plasma, and pig plasma was 1.01 (95% CI 0.92⫺1.10), 0.95 (95% CI 0.86⫺1.04), and 1.07 (95% CI 0.97⫺1.17), respectively. ASSAY
SENSITIVITY
The estimated total SD of the insulin aspart concentrations measured in samples spiked with 16 pmol/L insulin aspart is 2.352 pmol/L corresponds to CV 14.7% according to the imprecision experiment, giving a LOQ of 11.8 pmol/L. The variance components analysis of the measurements on seven individual serum samples spiked with 15.7 pmol/L insulin aspart showed residual, assay, and sample variation to be 3.025, 0.278, and 1.748 (pmol/L)2, respectively. SD of within-assay repeatability and total imprecision was estimated as 1.739 and 2.247 pmol/L, respectively, leading to an LOQ of 11.2 pmol/L. The mean value of LOQ is 11.5 pmol/L. DILUTION
LINEARITY AND RECOVERY
Table 2 shows the recovery of insulin aspart expressed as percentage of the amount of insulin aspart in the samples. The recovery of insulin aspart is linear within the dilution range. No statistically significant (p ⫽ 0.49) slope was found by regression analysis. The mean recovery was 94.9% with 7.3% SD. ASSAY
SELECTIVITY
Table 3 shows the effect of human insulin, porcine insulin, human proinsulin, and human 630
C-peptide on insulin aspart quantification. It appears that the effects do not depend on the insulin aspart concentration but depend only on the nature and the concentration of the added peptide. As shown in Table 3, human insulin, porcine insulin, and human proinsulin have a minor or negative effect on the insulin aspart measurements. Human C-peptide has a positive effect on the insulin aspart measurements at a concentration of 105 pmol/L. COMPARISON
WITH INSULIN RADIOIMMUNOASSAY
The DAKO insulin ELISA used to measure the endogenous human insulin in the serum samples from the clinical study was not significantly affected by the insulin aspart administered to the patients. The mean effect of insulin aspart on human insulin recovery in the tested samples was 1.2, 2.2, 7.2, ⫺6.6, and ⫺59.8% with insulin aspart concentrations of 2 ⫻ 101, 2 ⫻ 102, 2 ⫻ 103, 2 ⫻ 104, and 2 ⫻ 105 pmol/L, respectively. Comparison of the RIA and combined ELISA results from the analyzed study showed very close agreement in all cases. Log-transformation of the two analytical data sets secured homogeneity of variance. The RIA and the ELISA results were compared by using linear regression (Figure 3) of the differences between the log-concentrations vs. the sum of their log-concentrations (9). No statistically significant slope (p ⬎ 0.4) was found; therefore, the mean difference and the 95% CI could be calculated. Back transformed, the mean difference TABLE 2 Percent Recovery after Dilution of Insulin Aspart-Spiked Human Serum Samples Fold Dilution Sample ID 1 ID 2 ID 3 ID 4 ID 5 Mean SD
1
2
4
8
16
93.9% 92.2% 90.9% 93.8% 105.2% 95.2% 5.7%
96.2% 86.7% 88.6% 98.2% 106.9% 95.3% 8.1%
92.6% 91.4% 90.2% 93.9% 105.3% 94.7% 6.1%
88.4% 84.4% 86.1% 86.8% 103.3% 89.8% 7.7%
91.5% 95.0% 93.3% 110.3% 104.8% 99.0% 8.2%
CLINICAL BIOCHEMISTRY, VOLUME 33, NOVEMBER 2000
INSULIN ASPART-SPECIFIC ELISA
TABLE 3 Difference Between Measured and Expected Insulin Aspart Concentration Expressed as a Percentage of Expected Concentration Cross-Reacting Peptide Human insulin
Porcine insulin
Human proinsulin
Human C-peptide
Concentration (pmol/L) of Cross-Reacting Peptide
Expected Insulin Aspart (pmol/L)
10
102
103
104
105
10.4 23.5 93.3 187.4 322.9 10.4 23.5 93.3 187.4 322.9 10.4 23.5 93.3 187.4 322.9 10.4 23.5 93.3 187.4 322.9
⫺13.3% 1.4% 9.3% ⫺2.3% ⫺4.6% ⫺1.7% 1.0% 1.0% 4.6% ⫺9.2% ⫺2.7% 5.6% 1.6% ⫺4.5% ⫺6.1% 7.8% 3.9% ⫺0.3% ⫺1.8% ⫺2.1%
⫺18.0% ⫺0.7% ⫺9.7% ⫺2.9% 10.3% ⫺1.7% 1.0% ⫺4.4% ⫺3.9% 2.5% ⫺2.7% 2.7% ⫺7.0% ⫺5.0% 3.6% ⫺1.7% 1.4% 1.3% ⫺3.9% ⫺3.4%
⫺13.3% ⫺4.2% ⫺5.1% ⫺5.5% ⫺7.4% 0.2% ⫺11.4% ⫺0.7% ⫺2.3% ⫺5.2% 1.2% 4.4% ⫺4.8% ⫺5.5% ⫺14.5% 2.2% 6.9% ⫺1.7% 1.3% ⫺3.7%
⫺35.4% ⫺36.1% ⫺16.6% ⫺15.2% ⫺15.8% 7.0% ⫺16.5% ⫺15.0% ⫺14.1% ⫺18.8% ⫺15.2% ⫺13.1% ⫺16.8% ⫺18.4% ⫺17.0% 21.4% ⫺14.0% 0.9% 0.9% 0.0%
⫺100% ⫺90.6% ⫺86.9% ⫺77.8% ⫺76.9% ⫺93.3% ⫺85.1% ⫺81.1% ⫺80.1% ⫺78.5% ⫺100% ⫺82.5% ⫺77.4% ⫺76.5% ⫺75.3% 145% 60.2% 14.7% 12.6% 12.7%
1
of the ratio between the ELISA and RIA results was 0.034 (95% CI 0.014⫺0.054). The 95% CI does not include 0% indicating a significant bias. This was below 5%, however, and therefore not likely to be relevant.
The mean insulin profiles measured by ELISA and RIA are shown in Figure 4. It appears that the insulin measured by the ELISA and the RIA give similar results independent of the kinetic time course.
Figure 3 — Comparison between the combined results with specific ELISAs and the results with the insulin RIA 100 (Pharmacia) for measurements of insulin aspart ⫹ human insulin. These are plotted as the difference between the loge concentration from the two assay methods vs. the sum of the loge concentrations from the two assays. Only samples with insulin aspart ⬎ 10 pmol/L are included.
Figure 4 — Comparison of the total insulin (insulin aspart ⫹ human insulin) measured with ELISA or RIA. Samples from 10 individuals were taken between ⫺90 and 600 min after insulin aspart injection. Insulin concentrations were measured with either method and are shown as mean concentration and SD. It is apparent that the concentration difference in the kinetic time course between the two methods is insignificant.
CLINICAL BIOCHEMISTRY, VOLUME 33, NOVEMBER 2000
631
ANDERSEN
Discussion The insulin aspart-specific ELISA as described in this study proved to be suitable for preclinical and clinical studies on the bioavailability, bioequivalence, and pharmacokinetics of insulin aspart. It may also be useful in other studies, especially when endogenous human insulin and insulin aspart need to be measured separately. Eight insulin aspart calibrators in the range 0 to 1000 pmol/L and the use of a four-parameter logistic dose-response model is sufficient to establish an accurate calibration curve with acceptable imprecision. The data used to evaluate imprecision and accuracy showed proportionality between SD and the measured concentration. Variance homogeneity between spiking levels was secured by using the measured concentration divided by the spiked concentration. By this normalization of the data the statistical analysis was performed in the most optimal way, giving it more statistical power compared to the analysis of each spiking level separately. The imprecision analysis showed a constant CV between 16 pmol/L and the upper end of the calibrator curve. The CV of 14.7% in this assay range is acceptable for bioavailability, bioequivalence, and pharmacokinetic studies (10) because these parameters vary more between individuals. It was proven that the LOQ of 11.5 pmol/L could be calculated using the statistically estimated 14.7% CV at 16 pmol/L because the experiment with repeated measurements of seven individual insulinspiked samples at 15.7 pmol/L gave a non-significantly different LOQ value. In addition, this indicates that data below the lowest nonzero calibrator and the LOQ are accurate and precise. Even although the statistical analysis showed a significant accuracy bias with regard to both spiking levels and matrices, it was relatively low and within acceptable limits (10) for both parameters. Dilution of samples up to 16-fold with insulin aspart-free human serum does not affect the accuracy. Effects of human insulin, porcine insulin, human proinsulin, or human C-peptide on the insulin aspart quantification in samples are small and generally below the CV of the ELISA and independent of the concentration of insulin aspart at physiologic concentrations. However, if the concentration rises to 10,000 pmol/L or higher the cooperative effects become pronounced. The negative effects on insulin aspart measurements seen with 10,000 pmol/L to 100,000 pmol/L human insulin, porcine insulin, or human proinsulin may arise from competition between binding of insulin aspart and these cross-reactive insulins for the common epitope on the monoclonal antibody coated on the microtiter plate. There is no good explanation as to why human C-peptide showed positive effects above 10,000 pmol/L as this pep632
ET AL.
tide is not known to bind to insulin aspart or to share any homologous epitopes. Most of the effects regarding interference, as seen with peptides with homologous insulin epitopes, are due to the assay design. The design of this assay is based on a relatively low affinity (about 107 L/mol) of the detecting antibody, which only binds insulin aspart. Attempts to make the assay more accurate were not successful. Too much variability was found in the assay when the detection antibody was used as catcher antibody or when a washing step was introduced between sample incubation and detection antibody addition. Furthermore, adjustments to the assay were limited as there was only one monoclonal antibody available that allowed discrimination between two insulins with only one amino-acid difference. The comparison of the RIA and the combined ELISA results showed good agreement between the two methods with only a small difference (⬍5%). This difference is negligible in the context of clinical studies. The insulin aspart ELISA is available from MediLab A/S (Adelgade 5, DK-1304, Copenhagen, Denmark). The combination of the specific DAKO insulin ELISA, which measures endogenous human insulin only, and this new insulin aspart ELISA, which does not measure human insulin, may prove to be useful for studies in which both types of insulin may need to be assessed individually. Acknowledgement The authors thank Aage Vølund for critically reading the manuscript and Rene Tabanera y Palacios for performing the statistical analysis in SAS (both from the Statistics-Diabetes Department, Novo Nordisk). We also thank Birgitte Skafte and Birgitte Green Holm (Immunochemistry Department, Novo Nordisk) for performing the DAKO insulin ELISA analysis.
References 1. Brange J, Ribel U, Hansen JF, et al. Monomeric insulins obtained by protein engineering and their medical implications. Nature 1988; 333: 679 – 82. 2. Barnett AH, Owens DR. Insulin analogues. Lancet 1997; 349: 47–51. 3. UK Prospective Diabetes Study (UKPDS) XI. Biochemical risk factors in type 2 diabetic patients at diagnosis compared with age-matched normal subjects. Diabetic Med 1994; 11: 534⫺54. 4. Andersen L, Vølund A, Olsen KJ, Plum A, Walsh D. Validity and use of a non-parallel insulin assay for pharmacokinetic studies of a rapid-acting insulin analog, insulin aspart. J Immunoassay. In press. 5. Andersen L, Dinesen B, Jørgensen PN, Poulsen F, Røder ME. Enzyme immunoassay for intact human insulin in serum or plasma. Clin Chem 1993; 39: 578 – 82. 6. Vølund A. Conversion of insulin units to SI units [Letter]. Am J Clin Nutr 1993; 58: 714 –15. CLINICAL BIOCHEMISTRY, VOLUME 33, NOVEMBER 2000
INSULIN ASPART-SPECIFIC ELISA
7. Vølund A, Brange J, Drejer K, et al. In vitro and in vivo potency of insulin analogues designed for clinical use. Diabetic Med 1991; 8: 839 – 47. 8. Heinemann L, Weyer C, Rauhaus S, Heinrichs S, Heise T. Variability of the metabolic effect of soluble insulin and the rapid acting insulin analogue insulin aspart. Diabetes Care 1998; 21: 1910⫺14.
9. Bland JM, Altman DG. Statistical method for assessing agreement between two methods of clinical measurement. Lancet 1986; 1: 307⫺10. 10. Shah VP, Midha KK, Dighe S, et al. Analytical methods validation: Bioavailability, bioequivalence and pharmacokinetic studies. Pharm Res 1992; 9: 588 –92.
CLINICAL BIOCHEMISTRY, VOLUME 33, NOVEMBER 2000
633