Clin Biochem, Vol. 23, pp. 213-219, 1990 Printed in Canada. All rights reserved.
0009.9120/90 $3.00 + .00 Copyright © 1990 The Canadian Society of Clinical Chemists.
An Evaluation of an Enzyme Immunoassay Method for Immunoreactive Trypsin in Dried Blood Spots GIULIO CABRINI, 1 FABIO PEDERZINI, 1 LAURA PEROBELLI, 2 and GIANNI MASTELLA 1 1Cystic Fibrosis Center and 2Laboratorio di Chimica Clinica e di Ematologia, Ospedale Civile Maggiore, Verona, Italy A novel monoclonal antibody based enzyme immunoassay (EIA) method for the measurement of the human cationic trypsinogen (NeoScreen, AGEN Biomedical Ltd., Acacia Ridge, Australia) in dried blood spots for the neonatal screening of cystic fibrosis was evaluated. The calibration standards provided as dried blood spots by AGEN are highly unstable and must be replaced with user prepared materials. Reference values from control individuals were obtained by parametric methods. A preliminary comparison with a polyclonal antibody based RIA method (Trypsik, SORIN Biomedica, Saluggia, Italy) was performed. Regression analysis between the RIA and the EIA methods gave a coefficient of correlation of 0.58 for RIA values < 40 p.g/L and of 0.77 for RIA values -> 40 p.g/L. Average CV of the within-run imprecision for the EIA method was 19.6% and for the RIA method 28.8%. CVs of the between-run imprecision at low, intermediate and high values for the EIA method were 23.7%, 15.8%, 15.6% and for the RIA method 20.6%, 14.4%, 11.2%. The diagnostic accuracy analyzed by a Receiver Operating Characteristics (ROC) curve of the RIA method gave a maximum accuracy of 190.9 while that of a simulated ROC curve for the EIA method was 193.0. We found that the precision and the diagnostic accuracy of the EIA method (AGEN) are equal to or better than those of one of the RIA methods.
KEY WORDS: trypsinogen; immunoreactive trypsin; neonatal screening; cystic fibrosis. Introduction he observation of a persistently elevated concentration of immunoreactive trypsin (IRT) in T the blood of newborn babies affected by cystic fibrosis (CF) (1) is now utilized as a diagnostic tool in CF neonatal screening. Some RIA methods based on the binding of polyclonal antibodies to a combination of different molecular species (mainly cationic trypsinogen and/or trypsin) from h u m a n blood that are usually termed IRT are commercially available as reagent kits and adopted by several laboratories (2). In order to develop a quicker and less labour intensive assay, a novel EIA method using microtiter plates was recently proposed (3,4). Since the EIA method is based on the binding of two monoclonal
Correspondence: Dr. Giulio Cabrini, Cystic Fibrosis Center, Ospedale Civile Maggiore, Piazzale Stefani 1, 37126 Verona, Italy. Manuscript received May 23, 1989; revised October 31, 1989; accepted November 2, 1989. CLINICAL BIOCHEMISTRY, VOLUME 23, JUNE 1990
antibodies to the h u m a n trypsinogen, it may also have a better analytical specificity in comparison with that of the polyclonal antibody-based RIA methods. The EIA method is now commercially available as reagent kit (NeoScreen) and distributed by AGEN (Acacia Ridge, Australia). The aim of this study is to discuss the limitations and advantages of the AGEN reagent kit in comparison with a widely used RIA reagent kit. Materials a n d m e t h o d s BLOOD SAMPLES
Dried blood samples were collected on filter paper (Schleicher and Schuell 2992 from Pabish, Pero, Italy) by heel prick from the newborn babies (age: 3-5 days) of the Veneto Region and forwarded to the laboratory for the EIA and/or RIA assays. Trypsinogen retesting, meconium lactase and sweat test for the diagnosis of CF were performed according to the diagnostic strategy previously described by Mastella et al. (5). IV[_ETHODS
Trypsinogen by EIA was performed as described (4) except that the calibration standards were both those obtained from AGEN and an in-house preparation. The in-house standards were prepared by the following method, using the same antigen as the AGEN standards (trypsin from h u m a n pancreas purified by Calbiochem, La Jolla, CA). Heparinized blood (100 mL) obtained from the blood bank was centrifuged at 800 x g for 10 min at 4 °C. Plasma was discarded and the cells were resuspended with an equal volume of a phosphate buffered saline solution (154 mmol/L NaC1, 10 mmol/L sodium phosphate, pH 7.4) and centrifuged as before. These steps were performed three times and the final blood cell suspension was adjusted to an hematocrit value of 86%. One volume of this suspension was added to one volume of 154 mmol/L NaC1, 40 mg/mL bovine serum albumin (Hoechst-Behring, Marburg, FRG) containing variable amounts of trypsin purified 213
CABRINI, PEDERZINI, PEROBELLI, AND MASTELLA
from h u m a n pancreas in order to obtain the desired final concentration of trypsin. This suspension was distributed to 50 ~L aliquots onto the same type of filter paper as t h a t used for the patients and allowed to dry at room temperature for 6 h, then frozen at - 70 °C until use. Except when expressly stated, the trypsinogen concentration of the samples is calculated from calibration curves obtained by fitting with the least square method duplicate determinations of five concentrations of in-house calibration standards spanning the analytical range, expressed as ~Lg/L whole blood. The standards provided by AGEN are expressed as ~Lg/Lplasma. IRT by RIA method was performed as described (6) and subsequently modified by Burlina et al. (7) using the reagent kit obtained from SORIN Biomedica without further modifications. Sample values are expressed as ~Lg/Lwhole blood. Trypsinogen from specimens and calibration standards (for EIA and RIA) was measured on punched spots of 4 mm diameter. Reagents other t h a n those listed above were obtained from Sigma Chemical Co. (St. Louis, MO). APPARATUS
The absorbance at 420 nm of the transformed substrate (EIA method) was read in a MR 700 microplate reader (Dynatech Laboratories, Channel Islands, GB). Absorbance values were collected and analyzed through an IBM XT model 286 personal computer. RIA samples were counted in a Multiprias gamma counter (Packard Instruments, Downers Grove, IL). Results
EVALUATION OF CALIBRATIONSTANDARDS
During the first period of evaluation of the AGEN kit, specimens on some plates showed absorbance values below the zero calibrator values of the calibration curve obtained with the standards provided by AGEN. In order to understand this finding, 10 specimens have been measured in duplicate with the plates, reagents and standards of three different lots as shown in Table 1. The absorbance values of the specimens have been transformed into concentration values (even if negative) to show the range of variability. The possibility t h a t the calibration standards by AGEN were responsible for this variability was examined by analyzing on the same plate and the same reagents the calibration curves from AGEN standards obtained from five different lots. Table 2 reports the slope and intercept of the five curves fitted by the least square method (r >- 0.98); a high variability is evident particularly for the intercept with the Y axis, which represents the zero calibrator values. A further analysis of the stability character-
214
TABLE 1
Specimen Measurement by Different Lots Trypsinogen, ~tg/L Plasma Range of
Specimen
(Number) 1 2 3 4 5 6 7 8 9 10
1st Lot
2nd Lot
3rd Lot
Variation
-52 -27 -48 -57 -40 -53 -11 -38 -59 -57
2 30 0 -8 5 -10 47 26 -7 -14
15 35 5 0 7 3 46 28 0 0
67 62 53 57 47 56 58 66 59 57
Ten different specimens were measured in duplicate with the plates, reagents and standards of three different lots. Absorbance data were transformed into concentration values by the straight line fitted to five standards of each calibration curve. For significance of negative values, see text. istics of the AGEN standards is shown in Figure 1 where the absorbance values of the zero standard are plotted as a function of time after opening the standard package. It is apparent t h a t the AGEN standards changed and produced an increasing amount of immunoreactive or interfering substances. The results to be presented were all obtained with the in-house standards. REFERENCE VALUES
The frequency distribution of 5682 control individuals (representing all those tested over a 2month period except for two CF patients) is shown in Figure 2 (A). The distribution is not Gaussian (mean = 11.0, SD = 6.58, coefficient of skewness = 3.838, of kurtosis = 44.826). Therefore, the original data TABLE 2
Comparison of the Parameters of the Calibration Curves of Different Lots Lot
Slopea
Interceptb
1st
0.0047
0.332
2nd 3rd 4th 5th
0.0051 0.0032 0.0038 0.0045
0.673 0.546 0.495 0.278
Mean Range
0.0043 0.0019
0.456 0.395
Calibration standards by AGEN from five different lots were run in the same plato. A straight line was fitted by the least square method to five standards in duplicate. a Slope = dA]d[trypsinogen]. b Intercept of the Y axis = A at 420 nm.
CLINICAL BIOCHEMISTRY, VOLUME 23, JUNE 1990
EIA M E T H O D
F O R IRT
0.600 0.500 (D
/
eAGEN colibrotor zero [3User p r e p l
/
r
0.400
r
r'~ 0
0.300 0.200
_D [ 3 ~ [ 3
0.100 0.000
[3
u-u0
13
D-
I
I
5
10
I
[30
[]
[]
[] I
15 20 TIME (doys)
I
25
30
Figure 1--Absorbance of the zero calibrators as a function of time. New lot of standards (AGEN and in-house) commenced at time zero. The daily mean value of the standards with [trypsinogen] = 0 is plotted. were corrected for skewness and kurtosis according to Solberg (8) with addition of constant values, log and power transformations. The transformation which gave a distribution least dissimilar to a Gaussian one is Log (X + 1.9) where X is the original value of trypsinogen. The transformed distribution is shown in Figure 2 (B). (mean = 1.067, SD = 0.192, coefficient of skewness = -0.009, of kurtosis = 3.908). The cumulative frequency of the transformed experimental distribution is graphically compared with t h a t of a Gaussian distribution with the same mean and SD in Figure 2(C), showing a very small difference between theoretical and experimental distribution. This is particularly evident for the right tail of the distribution (e.g., fractile 0.995), where the cut-off value between controls and patients must be chosen. Kolmogorov-Smirnov test (Dmax = 0.004, critical value at 0.01 level = 0.014) and Chi-squared goodness-of-fit test (corrected Chi-squared = 10.129, critical value at 0.01 level = 34.805) confirm the hypothesis t h a t the transformed experimental distribution fit a Gaussian distribution at the confidence limit of 1% (9). Parametric fractiles calculated from the right tail of the Gaussian distribution of the transformed data (e.g., fractile 0.990 = 30.6 ~g/L trypsinogen, 0.995 = 34.4, 0.999 = 42.0) were strictly corresponding to the actual frequency of the experimental distribution of the 5682 control individuals. REGRESSION ANALYSIS OF THE T W O M E T H O D S
A regression analysis between the RIA (X) and EIA (Y) methods was calculated from 178 specimens spanning the whole analytical range, subdivided into two groups (above and below the cut-off value for the RIA method = 40 ~g/L whole blood). The
CLINICAL BIOCHEMISTRY, VOLUME 23, J U N E 1990
coefficients of correlation were low either for RIA values < 40 ~g/L (r = 0.58, Y = 3.9 + 0.345.X, Syx = 3.53) and for RIA values -> 40 ~g/L (r = 0.77, Y = - 0 . 5 + 0.577.X, Syx = 20.34). PRECISION OF THE T W O M E T H O D S
Within-run imprecision of the two methods is shown in Table 3. Mean values methods were calculated on the difference of duplicate values of 80 specimens spanning the analytical ranges according to Buttner et al. (10), while the values at different concentration intervals (low, intermediate, elevated) were calculated by subdividing the whole analytical range into defined intervals. The EIA method is less imprecise t h a n the RIA method, either considering the whole analytical range and the partial intervals. Table 4 reports the between-run imprecision t h a t was calculated on the value of three specimens spanning the analytical range. Each sample of the three specimens was measured in at least 25 consecutive working days according to Buttner et al. (10). The between-run imprecision for the RIA method is slightly lower t h a n t h a t for the EIA method. DIAGNOSTIC A C C U R A C Y
Figure 3 (A) shows the frequency distribution of 5682 control individuals and the value of trypsinogen of 18 CF patients measured by the EIA method. In order to evaluate the sensitivity and specificity in the diagnosis of CF, a Receiver Operating Characteristic (ROC) curve for the RIA method is plotted together with a simulated curve for the EIA method, as shown in Figure 3 (B), according to Swets (11). ROC curves of individual control distribution from 5682 (EIA) and 5422 (RIA) subjects served as nega-
215
CABRINI, PEDERZINI, PEROBELLI, AND MASTELLA 1400. 1200.
~
1000.
0
800.
0
BOO'
o h 0
F.,
400.
m
3~
200.
Z
0 0
5
15
10
20
25
30
I
!
35
40
TRYPSINOGEN,/4g/L whole blood 1400 1200
looo ..J
o
800
o0
SO0
b.
o
Ili
400 200
Z 0
0.28
t'"
7
LOG(ll~fPSINOGEN+1.9),/~g/L whole blood
2.08
1.0 • observed frequency
Z 0
0.8 0.8 :D Z
0.4 0.2-
o 0.0 0.280
LOO0"RYPSINOOEN+I .g), H,g/L whole blood
1.780
Figure 2--Frequency distribution of blood trypsinogen in 5682 control individuals (EIA method). (A) Original data; (B) Transformed data; (C) Comparison of the cumulative frequency of the experimental data after the transformation and a Gaussian distribution with the same mean and SD.
tive events at defined cut-off values. The values of 78 CF subjects measured by the RIA method in the period preceding the introduction of the EIA method served as positive events. Since the average number of CF subjects diagnosed per year by neonatal screening in the laboratory is approximately 12 and the EIA method was introduced only in the last 6 months, the evaluation of diagnostic accuracy by the EIA method with a comparable number of CF subjects m a y need at least 6 years. Therefore the RIA values of the 78 CF subjects have been converted into EIA values on the basis of the regression
216
analysis calculated above the cut-off level of the RIA method (40 }xg/L whole blood) described in the previous section. For this reason the ROC curve for the EIA method must be assumed as a simulated curve giving a preliminary indication. This simulated curve predicts that the EIA method might have a similar or even better sensitivity and specificity than the RIA method. The minimum mistake cut-off level, calculated graphically at the intercept of the ROC curves with their tangents at 45 °C was at the cut-off level of 45 ~g/L for the RIA method and 25 ~g/L for the EIA method. The m a x i m u m accuracy, that is the
CLINICAL BIOCHEMISTRY, VOLUME 23, J U N E 1990
EIA METHOD FOR IRT TABLE 4 Between-Run Imprecision of the EIA and RIA Methods
TABLE 3 Within-Run Imprecision of the EIA and RIA Methods Trypsinogen, ~g/L Whole Blood Method EIA (AGEN)
Specimen CV, n %
Interval
Mean
SD
0-120
29.5
5.8
80
19.6
0-10 11-20 21-40 41-120
6.3 15.7 29.9 61.2
1.4 1.9 2.3 9.8
17 13 33 17
22.2 12.0 7.6 16.0
0-160
44.1
12.7
80
28.8
0-20 21-40 41-60 61-100 101-160
10.0 33.4 49.4 77.8 126.5
3.0 8.3 5.7 10.1 23.8
19 19 28 10 4
30.0 24.8 11.5 12.9 18.8
Trypsinogen, ~g/L Whole Blood Mean SD
Method
n
CV, %
EIA (AGEN)
10.8 22.1 103.6
2.6 3.5 16.1
31 27 25
23.7 15.8 15.6
RIA (SORIN)
6.3 55.8 84.4
1.3 8.0 9.5
25 25 25
20.6 14.4 11.2
n = number of consecutive days. RIA (SORIN)
1400bJ
1200
(/)
1000
value where the sum of % sensitivity and specificity is maximum, was 190.9 for the RIA and 193.0 for the EIA method. Discussion IRT in blood spots can be assayed by at least three polyclonal antibody based RIA methods (2). Because
EIA METHOD r - 1 CONTROL SUBJECTS (n = 5682) • CYSTICRBROSIS SUBJECTS (n = 18)
-J
o 0
u b.. 0
,,,.
800 600 400.
IJJ
I]3
'~
200.
Z
0 20
0
40
60
80
120
1 O0
140
TRYPSINOGEN,/~g/L whole blood
SPECIRCnY,
1O0 100
98
96
=
i
94 i
92
90
i
95 9O
8O
/ ,
7O 65
J [ /
J
TRYPSINOGEN CUT OFF VALUES ~ • RIA method J ~ = EIA method RIA RI/ control oontrol IndMduole I nd Ivid ,= 5422 EiA Ei~ control control lndMduals 1nd ~ d == 5682 CF ~mple8 = 78
60
Figure 3--Evaluation of diagnositic accuracy of blood trypsinogen. (A) Specimens from five cystic fibrosis patients collected from August 1988 (upper row) are shown together with 13 specimens that were collected before August 1988 (lower row) and kept frozen at - 2 0 °C for a maximum of 18 months. (B) ROC curve for RIA method plotted against simulated curve for EIA method. CLINICAL BIOCHEMISTRY, VOLUME 23, JUNE 1990
217
CABRINI, PEDERZINI, PEROBELLI,AND MASTELLA of differences between antibodies and procedures and the absence of an International Reference Standard for the h u m a n trypsinogen, reference values and cut-off levels of these methods are not homogeneous and comparative analytical accuracy and detection limit studies are not possible. The introduction of a new monoclonal antibody-based EIA method performed in microtiter plates (3,4) does not solve these problems but might have the advantage of being simpler and quicker than a RIA method. The use of a colorimetric detection system instead of a radioactive tracer makes the organization of the work much easier. The calibration standards of the AGEN kit seem unreliable when stored at 4 °C as recommended by the manufacturer. Specimens with low values can be over- or underestimated with an oscillation between 40 and 70 ~g/L plasma depending on the Lot as shown in Table 1. This is related to an increase of the absorbance of the zero calibrators (see Table 2) which also shows a dependence upon the time (see Figure 1). The increase of the zero calibrator values (intercept) with conserved slope of the calibration curves is caused by an increase of the nonspecific "binding in the sandwich assay. This could be caused by the type of proteins that are added in the standard, to the storage temperature that AGEN suggests or to the type of preparation (dried spots instead of lyophilized preparation). Since the reagents and procedure of preparation of the calibration standards supplied by AGEN are not known (except for the type and source of the antigen), the cause remains unclear. In contrast, the in-house calibration standards are stable and reliable when prepared as described in the method section and stored as dried spots at - 7 0 °C. The frequency distribution of the control individuals is not Gaussian. Actually a non-Gaussian distribution is very common in biological measurements (12). It was, however, possible to transform the data to obtain a Gaussian distribution and parametric reference values. The regression analysis of the data obtained by measuring 178 specimens with the EIA and RIA method gives a poor coefficient of correlation. This might be expected, however, because of the different origin of the antigens and the differences in antibodies and assay type. The within-run imprecision of the EIA method is in our hands higher than previously reported (CV of 19.6% vs. 8.0%) (4). However, the imprecision profile shows an acceptable value (CV of 7.6%) in the analytical interval where the cut-off level must be chosen (21-40 ~g/L whole blood) and where the minimum imprecision is desirable. The within-run imprecision of the EIA method appeared to be less than that of the RIA method in the 21-40 interval, b u t was otherwise comparable. The between-run imprecision of the EIA method is also higher than that reported (4) and similar to that of the RIA method (13). Even if the precision of the EIA method is better 218
or equal to that of the RIA method, the CVs reported in Tables 3 and 4 seem quite high in comparison to similar EIA determinations in serum or plasma. Additional sources of variability could be the standardization of the blood sampling procedure (which is performed by hundreds of different nurses from 84 hospitals of the Veneto Region) and variability in the elution of the trypsinogen from the filter paper. The control distribution together with the values of the CF patients of Figure 3 (A) gives a general impression of a very low overlap of the two populations. It should also be noted that the CF subject with the lowest trypsinogen concentration shown in Figure 3 (18 ~g/L) was affected by meconium ileus, which can be a cause of false negative cases in CF neonatal screening (14). Also, sensitivity and specificity in the diagnosis of CF are better evaluated at several cut-off levels as shown in the profile of a ROC curve. The profile for the RIA method (Figure 3 (B)) shows a very high diagnostic accuracy. A ROC curve for the EIA method obtained with actual measurements will be available only after some years of collection of data from CF patients. Considering the assumptions and limitations associated with simulations, the simulated ROC curve for the EIA method also shows a very high diagnostic accuracy. In conclusion, the calibration standard material in the EIA method as commercially distributed by AGEN must be improved. The method is quick and simple and shows precision characteristics and diagnostic accuracy equal to or better than those of an RIA method. Acknowledgments We wish to thank: Prof. U. Lippi for revising the manuscript, Prof. G. Berton, Dr. M. C. Dechecchi, Dr. A. Tamanini for helpful discussions, Dr. C. Parmelli for the software for data transformation, V. Stanzial for assistance in data analysis, A. Facchin, P. Faggionato, L. Pattini, S. Pizzini, C. Zampini for excellent technical assistance, E. Bravi for assistance in graphic presentation. This work was partly supported by grants from the "Progetto Finalizzato: Ingegneria Genetica e Basi Molecolari delle Malattie Ereditarie," CNR, Rome, Italy. References
1. Crossley JR, Elliot RB, Smith PA. Dried-blood spot screening for cystic fibrosis in the newborn. Lancet 1979; 1: 472--4. 2. Davidson AGF, Wong LTK, Kirby LT, Applegarth DA. Immunoreactive trypsin in cystic fibrosis. J Ped Gastroenterol Nutr 1984; 3 (Suppl. 1): $79-88. 3. Bowling FG, Watson ARA, Rylatt DB, Elliott JE, Bunch RJ, Bundesen PG. Monoclonal antibody-based enzyme immunoassay for trypsinogen in neonatal screening for cystic fibrosis. Lancet 1987; 1: 826-7. 4. Bowling FG, Brown ARD. Newborn screening for cystic fibrosis using an enzyme linked immunoabsorbent assay (ELISA) technique. Clin Chim Acta 1988; 171: 257-62. CLINICAL BIOCHEMISTRY,VOLUME 23, JUNE 1990
EIA METHOD FOR IRT 5. Mastella G, Pederzini F, Girella E, Righetti G, Zanchetta M, Rizzotti P. Cystic fibrosis screening policy. Lancet 1984; 2: 575-6. 6. Malvano R, Marchisio M, Massaglia A, et al. Radioimmunoassay of trypsin-like substance in human serum. Scand J Gastroenterol 1980; 15 (Suppl. 62): 3-10. 7. Burlina A, Tonon M, Perobelli L, Zanchetta M, Rizzotti P, Plebani M. A new blood spot method for immunotrypsin. IRCS Med Sci 1982; 10: 951-2. 8. Solberg HE. The theory of reference values. Part 5. Statistical treatment of collected reference values. Determination of reference limits. J Clin Chem Clin Biochem 1983; 23: 749-60. 9. Armitage P. Statistical methods in medical research. Oxford: Blackwell Scientific Publications, 1971. 10. Buttner J, Borth R, Bourtwell HJ, Broughton PMG,
CLINICAL BIOCHEMISTRY, VOLUME 23, JUNE 1990
11. 12. 13. 14.
Bowyer RC. Approved recommendation (1978) on quality control in clinical chemistry. Part 2. Assessment of analytical methods for routine use. J Clin Chem Clin Biochem 1980; 18: 78-88. Swets JA. Measuring the accuracy of diagnostic systems. Science 1988; 240: 1285-93. Fuentes-Arderiu J, Sierra MT, Panadero AM. No more assumptions about whether control results have a Gaussian distribution. Clin Chem 1988; 34: 769-70. Zucchelli GC, Piro MA, Malvano R. Radioimmunoassay of trypsin-like immunoreactivity in dried blood spots. J Nucl Med All Sci 1982; 26: 35-9. Heeley AF, Heeley MF. Biochemical screening of the neonatal population for the early detection of CF in east Anglia 1980-1986. Insights into Paediatrics 1987; 1: 35-43.
219
0009.9120/90 $3.00 + .00 Copyright © 1990 The Canadian Society of Clinical Chemists.
An Evaluation of an Enzyme Immunoassay Method for Immunoreactive Trypsin in Dried Blood Spots GIULIO CABRINI, 1 FABIO PEDERZINI, 1 LAURA PEROBELLI, 2 and GIANNI MASTELLA 1 1Cystic Fibrosis Center and 2Laboratorio di Chimica Clinica e di Ematologia, Ospedale Civile Maggiore, Verona, Italy A novel monoclonal antibody based enzyme immunoassay (EIA) method for the measurement of the human cationic trypsinogen (NeoScreen, AGEN Biomedical Ltd., Acacia Ridge, Australia) in dried blood spots for the neonatal screening of cystic fibrosis was evaluated. The calibration standards provided as dried blood spots by AGEN are highly unstable and must be replaced with user prepared materials. Reference values from control individuals were obtained by parametric methods. A preliminary comparison with a polyclonal antibody based RIA method (Trypsik, SORIN Biomedica, Saluggia, Italy) was performed. Regression analysis between the RIA and the EIA methods gave a coefficient of correlation of 0.58 for RIA values < 40 p.g/L and of 0.77 for RIA values -> 40 p.g/L. Average CV of the within-run imprecision for the EIA method was 19.6% and for the RIA method 28.8%. CVs of the between-run imprecision at low, intermediate and high values for the EIA method were 23.7%, 15.8%, 15.6% and for the RIA method 20.6%, 14.4%, 11.2%. The diagnostic accuracy analyzed by a Receiver Operating Characteristics (ROC) curve of the RIA method gave a maximum accuracy of 190.9 while that of a simulated ROC curve for the EIA method was 193.0. We found that the precision and the diagnostic accuracy of the EIA method (AGEN) are equal to or better than those of one of the RIA methods.
KEY WORDS: trypsinogen; immunoreactive trypsin; neonatal screening; cystic fibrosis. Introduction he observation of a persistently elevated concentration of immunoreactive trypsin (IRT) in T the blood of newborn babies affected by cystic fibrosis (CF) (1) is now utilized as a diagnostic tool in CF neonatal screening. Some RIA methods based on the binding of polyclonal antibodies to a combination of different molecular species (mainly cationic trypsinogen and/or trypsin) from h u m a n blood that are usually termed IRT are commercially available as reagent kits and adopted by several laboratories (2). In order to develop a quicker and less labour intensive assay, a novel EIA method using microtiter plates was recently proposed (3,4). Since the EIA method is based on the binding of two monoclonal
Correspondence: Dr. Giulio Cabrini, Cystic Fibrosis Center, Ospedale Civile Maggiore, Piazzale Stefani 1, 37126 Verona, Italy. Manuscript received May 23, 1989; revised October 31, 1989; accepted November 2, 1989. CLINICAL BIOCHEMISTRY, VOLUME 23, JUNE 1990
antibodies to the h u m a n trypsinogen, it may also have a better analytical specificity in comparison with that of the polyclonal antibody-based RIA methods. The EIA method is now commercially available as reagent kit (NeoScreen) and distributed by AGEN (Acacia Ridge, Australia). The aim of this study is to discuss the limitations and advantages of the AGEN reagent kit in comparison with a widely used RIA reagent kit. Materials a n d m e t h o d s BLOOD SAMPLES
Dried blood samples were collected on filter paper (Schleicher and Schuell 2992 from Pabish, Pero, Italy) by heel prick from the newborn babies (age: 3-5 days) of the Veneto Region and forwarded to the laboratory for the EIA and/or RIA assays. Trypsinogen retesting, meconium lactase and sweat test for the diagnosis of CF were performed according to the diagnostic strategy previously described by Mastella et al. (5). IV[_ETHODS
Trypsinogen by EIA was performed as described (4) except that the calibration standards were both those obtained from AGEN and an in-house preparation. The in-house standards were prepared by the following method, using the same antigen as the AGEN standards (trypsin from h u m a n pancreas purified by Calbiochem, La Jolla, CA). Heparinized blood (100 mL) obtained from the blood bank was centrifuged at 800 x g for 10 min at 4 °C. Plasma was discarded and the cells were resuspended with an equal volume of a phosphate buffered saline solution (154 mmol/L NaC1, 10 mmol/L sodium phosphate, pH 7.4) and centrifuged as before. These steps were performed three times and the final blood cell suspension was adjusted to an hematocrit value of 86%. One volume of this suspension was added to one volume of 154 mmol/L NaC1, 40 mg/mL bovine serum albumin (Hoechst-Behring, Marburg, FRG) containing variable amounts of trypsin purified 213
CABRINI, PEDERZINI, PEROBELLI, AND MASTELLA
from h u m a n pancreas in order to obtain the desired final concentration of trypsin. This suspension was distributed to 50 ~L aliquots onto the same type of filter paper as t h a t used for the patients and allowed to dry at room temperature for 6 h, then frozen at - 70 °C until use. Except when expressly stated, the trypsinogen concentration of the samples is calculated from calibration curves obtained by fitting with the least square method duplicate determinations of five concentrations of in-house calibration standards spanning the analytical range, expressed as ~Lg/L whole blood. The standards provided by AGEN are expressed as ~Lg/Lplasma. IRT by RIA method was performed as described (6) and subsequently modified by Burlina et al. (7) using the reagent kit obtained from SORIN Biomedica without further modifications. Sample values are expressed as ~Lg/Lwhole blood. Trypsinogen from specimens and calibration standards (for EIA and RIA) was measured on punched spots of 4 mm diameter. Reagents other t h a n those listed above were obtained from Sigma Chemical Co. (St. Louis, MO). APPARATUS
The absorbance at 420 nm of the transformed substrate (EIA method) was read in a MR 700 microplate reader (Dynatech Laboratories, Channel Islands, GB). Absorbance values were collected and analyzed through an IBM XT model 286 personal computer. RIA samples were counted in a Multiprias gamma counter (Packard Instruments, Downers Grove, IL). Results
EVALUATION OF CALIBRATIONSTANDARDS
During the first period of evaluation of the AGEN kit, specimens on some plates showed absorbance values below the zero calibrator values of the calibration curve obtained with the standards provided by AGEN. In order to understand this finding, 10 specimens have been measured in duplicate with the plates, reagents and standards of three different lots as shown in Table 1. The absorbance values of the specimens have been transformed into concentration values (even if negative) to show the range of variability. The possibility t h a t the calibration standards by AGEN were responsible for this variability was examined by analyzing on the same plate and the same reagents the calibration curves from AGEN standards obtained from five different lots. Table 2 reports the slope and intercept of the five curves fitted by the least square method (r >- 0.98); a high variability is evident particularly for the intercept with the Y axis, which represents the zero calibrator values. A further analysis of the stability character-
214
TABLE 1
Specimen Measurement by Different Lots Trypsinogen, ~tg/L Plasma Range of
Specimen
(Number) 1 2 3 4 5 6 7 8 9 10
1st Lot
2nd Lot
3rd Lot
Variation
-52 -27 -48 -57 -40 -53 -11 -38 -59 -57
2 30 0 -8 5 -10 47 26 -7 -14
15 35 5 0 7 3 46 28 0 0
67 62 53 57 47 56 58 66 59 57
Ten different specimens were measured in duplicate with the plates, reagents and standards of three different lots. Absorbance data were transformed into concentration values by the straight line fitted to five standards of each calibration curve. For significance of negative values, see text. istics of the AGEN standards is shown in Figure 1 where the absorbance values of the zero standard are plotted as a function of time after opening the standard package. It is apparent t h a t the AGEN standards changed and produced an increasing amount of immunoreactive or interfering substances. The results to be presented were all obtained with the in-house standards. REFERENCE VALUES
The frequency distribution of 5682 control individuals (representing all those tested over a 2month period except for two CF patients) is shown in Figure 2 (A). The distribution is not Gaussian (mean = 11.0, SD = 6.58, coefficient of skewness = 3.838, of kurtosis = 44.826). Therefore, the original data TABLE 2
Comparison of the Parameters of the Calibration Curves of Different Lots Lot
Slopea
Interceptb
1st
0.0047
0.332
2nd 3rd 4th 5th
0.0051 0.0032 0.0038 0.0045
0.673 0.546 0.495 0.278
Mean Range
0.0043 0.0019
0.456 0.395
Calibration standards by AGEN from five different lots were run in the same plato. A straight line was fitted by the least square method to five standards in duplicate. a Slope = dA]d[trypsinogen]. b Intercept of the Y axis = A at 420 nm.
CLINICAL BIOCHEMISTRY, VOLUME 23, JUNE 1990
EIA M E T H O D
F O R IRT
0.600 0.500 (D
/
eAGEN colibrotor zero [3User p r e p l
/
r
0.400
r
r'~ 0
0.300 0.200
_D [ 3 ~ [ 3
0.100 0.000
[3
u-u0
13
D-
I
I
5
10
I
[30
[]
[]
[] I
15 20 TIME (doys)
I
25
30
Figure 1--Absorbance of the zero calibrators as a function of time. New lot of standards (AGEN and in-house) commenced at time zero. The daily mean value of the standards with [trypsinogen] = 0 is plotted. were corrected for skewness and kurtosis according to Solberg (8) with addition of constant values, log and power transformations. The transformation which gave a distribution least dissimilar to a Gaussian one is Log (X + 1.9) where X is the original value of trypsinogen. The transformed distribution is shown in Figure 2 (B). (mean = 1.067, SD = 0.192, coefficient of skewness = -0.009, of kurtosis = 3.908). The cumulative frequency of the transformed experimental distribution is graphically compared with t h a t of a Gaussian distribution with the same mean and SD in Figure 2(C), showing a very small difference between theoretical and experimental distribution. This is particularly evident for the right tail of the distribution (e.g., fractile 0.995), where the cut-off value between controls and patients must be chosen. Kolmogorov-Smirnov test (Dmax = 0.004, critical value at 0.01 level = 0.014) and Chi-squared goodness-of-fit test (corrected Chi-squared = 10.129, critical value at 0.01 level = 34.805) confirm the hypothesis t h a t the transformed experimental distribution fit a Gaussian distribution at the confidence limit of 1% (9). Parametric fractiles calculated from the right tail of the Gaussian distribution of the transformed data (e.g., fractile 0.990 = 30.6 ~g/L trypsinogen, 0.995 = 34.4, 0.999 = 42.0) were strictly corresponding to the actual frequency of the experimental distribution of the 5682 control individuals. REGRESSION ANALYSIS OF THE T W O M E T H O D S
A regression analysis between the RIA (X) and EIA (Y) methods was calculated from 178 specimens spanning the whole analytical range, subdivided into two groups (above and below the cut-off value for the RIA method = 40 ~g/L whole blood). The
CLINICAL BIOCHEMISTRY, VOLUME 23, J U N E 1990
coefficients of correlation were low either for RIA values < 40 ~g/L (r = 0.58, Y = 3.9 + 0.345.X, Syx = 3.53) and for RIA values -> 40 ~g/L (r = 0.77, Y = - 0 . 5 + 0.577.X, Syx = 20.34). PRECISION OF THE T W O M E T H O D S
Within-run imprecision of the two methods is shown in Table 3. Mean values methods were calculated on the difference of duplicate values of 80 specimens spanning the analytical ranges according to Buttner et al. (10), while the values at different concentration intervals (low, intermediate, elevated) were calculated by subdividing the whole analytical range into defined intervals. The EIA method is less imprecise t h a n the RIA method, either considering the whole analytical range and the partial intervals. Table 4 reports the between-run imprecision t h a t was calculated on the value of three specimens spanning the analytical range. Each sample of the three specimens was measured in at least 25 consecutive working days according to Buttner et al. (10). The between-run imprecision for the RIA method is slightly lower t h a n t h a t for the EIA method. DIAGNOSTIC A C C U R A C Y
Figure 3 (A) shows the frequency distribution of 5682 control individuals and the value of trypsinogen of 18 CF patients measured by the EIA method. In order to evaluate the sensitivity and specificity in the diagnosis of CF, a Receiver Operating Characteristic (ROC) curve for the RIA method is plotted together with a simulated curve for the EIA method, as shown in Figure 3 (B), according to Swets (11). ROC curves of individual control distribution from 5682 (EIA) and 5422 (RIA) subjects served as nega-
215
CABRINI, PEDERZINI, PEROBELLI, AND MASTELLA 1400. 1200.
~
1000.
0
800.
0
BOO'
o h 0
F.,
400.
m
3~
200.
Z
0 0
5
15
10
20
25
30
I
!
35
40
TRYPSINOGEN,/4g/L whole blood 1400 1200
looo ..J
o
800
o0
SO0
b.
o
Ili
400 200
Z 0
0.28
t'"
7
LOG(ll~fPSINOGEN+1.9),/~g/L whole blood
2.08
1.0 • observed frequency
Z 0
0.8 0.8 :D Z
0.4 0.2-
o 0.0 0.280
LOO0"RYPSINOOEN+I .g), H,g/L whole blood
1.780
Figure 2--Frequency distribution of blood trypsinogen in 5682 control individuals (EIA method). (A) Original data; (B) Transformed data; (C) Comparison of the cumulative frequency of the experimental data after the transformation and a Gaussian distribution with the same mean and SD.
tive events at defined cut-off values. The values of 78 CF subjects measured by the RIA method in the period preceding the introduction of the EIA method served as positive events. Since the average number of CF subjects diagnosed per year by neonatal screening in the laboratory is approximately 12 and the EIA method was introduced only in the last 6 months, the evaluation of diagnostic accuracy by the EIA method with a comparable number of CF subjects m a y need at least 6 years. Therefore the RIA values of the 78 CF subjects have been converted into EIA values on the basis of the regression
216
analysis calculated above the cut-off level of the RIA method (40 }xg/L whole blood) described in the previous section. For this reason the ROC curve for the EIA method must be assumed as a simulated curve giving a preliminary indication. This simulated curve predicts that the EIA method might have a similar or even better sensitivity and specificity than the RIA method. The minimum mistake cut-off level, calculated graphically at the intercept of the ROC curves with their tangents at 45 °C was at the cut-off level of 45 ~g/L for the RIA method and 25 ~g/L for the EIA method. The m a x i m u m accuracy, that is the
CLINICAL BIOCHEMISTRY, VOLUME 23, J U N E 1990
EIA METHOD FOR IRT TABLE 4 Between-Run Imprecision of the EIA and RIA Methods
TABLE 3 Within-Run Imprecision of the EIA and RIA Methods Trypsinogen, ~g/L Whole Blood Method EIA (AGEN)
Specimen CV, n %
Interval
Mean
SD
0-120
29.5
5.8
80
19.6
0-10 11-20 21-40 41-120
6.3 15.7 29.9 61.2
1.4 1.9 2.3 9.8
17 13 33 17
22.2 12.0 7.6 16.0
0-160
44.1
12.7
80
28.8
0-20 21-40 41-60 61-100 101-160
10.0 33.4 49.4 77.8 126.5
3.0 8.3 5.7 10.1 23.8
19 19 28 10 4
30.0 24.8 11.5 12.9 18.8
Trypsinogen, ~g/L Whole Blood Mean SD
Method
n
CV, %
EIA (AGEN)
10.8 22.1 103.6
2.6 3.5 16.1
31 27 25
23.7 15.8 15.6
RIA (SORIN)
6.3 55.8 84.4
1.3 8.0 9.5
25 25 25
20.6 14.4 11.2
n = number of consecutive days. RIA (SORIN)
1400bJ
1200
(/)
1000
value where the sum of % sensitivity and specificity is maximum, was 190.9 for the RIA and 193.0 for the EIA method. Discussion IRT in blood spots can be assayed by at least three polyclonal antibody based RIA methods (2). Because
EIA METHOD r - 1 CONTROL SUBJECTS (n = 5682) • CYSTICRBROSIS SUBJECTS (n = 18)
-J
o 0
u b.. 0
,,,.
800 600 400.
IJJ
I]3
'~
200.
Z
0 20
0
40
60
80
120
1 O0
140
TRYPSINOGEN,/~g/L whole blood
SPECIRCnY,
1O0 100
98
96
=
i
94 i
92
90
i
95 9O
8O
/ ,
7O 65
J [ /
J
TRYPSINOGEN CUT OFF VALUES ~ • RIA method J ~ = EIA method RIA RI/ control oontrol IndMduole I nd Ivid ,= 5422 EiA Ei~ control control lndMduals 1nd ~ d == 5682 CF ~mple8 = 78
60
Figure 3--Evaluation of diagnositic accuracy of blood trypsinogen. (A) Specimens from five cystic fibrosis patients collected from August 1988 (upper row) are shown together with 13 specimens that were collected before August 1988 (lower row) and kept frozen at - 2 0 °C for a maximum of 18 months. (B) ROC curve for RIA method plotted against simulated curve for EIA method. CLINICAL BIOCHEMISTRY, VOLUME 23, JUNE 1990
217
CABRINI, PEDERZINI, PEROBELLI,AND MASTELLA of differences between antibodies and procedures and the absence of an International Reference Standard for the h u m a n trypsinogen, reference values and cut-off levels of these methods are not homogeneous and comparative analytical accuracy and detection limit studies are not possible. The introduction of a new monoclonal antibody-based EIA method performed in microtiter plates (3,4) does not solve these problems but might have the advantage of being simpler and quicker than a RIA method. The use of a colorimetric detection system instead of a radioactive tracer makes the organization of the work much easier. The calibration standards of the AGEN kit seem unreliable when stored at 4 °C as recommended by the manufacturer. Specimens with low values can be over- or underestimated with an oscillation between 40 and 70 ~g/L plasma depending on the Lot as shown in Table 1. This is related to an increase of the absorbance of the zero calibrators (see Table 2) which also shows a dependence upon the time (see Figure 1). The increase of the zero calibrator values (intercept) with conserved slope of the calibration curves is caused by an increase of the nonspecific "binding in the sandwich assay. This could be caused by the type of proteins that are added in the standard, to the storage temperature that AGEN suggests or to the type of preparation (dried spots instead of lyophilized preparation). Since the reagents and procedure of preparation of the calibration standards supplied by AGEN are not known (except for the type and source of the antigen), the cause remains unclear. In contrast, the in-house calibration standards are stable and reliable when prepared as described in the method section and stored as dried spots at - 7 0 °C. The frequency distribution of the control individuals is not Gaussian. Actually a non-Gaussian distribution is very common in biological measurements (12). It was, however, possible to transform the data to obtain a Gaussian distribution and parametric reference values. The regression analysis of the data obtained by measuring 178 specimens with the EIA and RIA method gives a poor coefficient of correlation. This might be expected, however, because of the different origin of the antigens and the differences in antibodies and assay type. The within-run imprecision of the EIA method is in our hands higher than previously reported (CV of 19.6% vs. 8.0%) (4). However, the imprecision profile shows an acceptable value (CV of 7.6%) in the analytical interval where the cut-off level must be chosen (21-40 ~g/L whole blood) and where the minimum imprecision is desirable. The within-run imprecision of the EIA method appeared to be less than that of the RIA method in the 21-40 interval, b u t was otherwise comparable. The between-run imprecision of the EIA method is also higher than that reported (4) and similar to that of the RIA method (13). Even if the precision of the EIA method is better 218
or equal to that of the RIA method, the CVs reported in Tables 3 and 4 seem quite high in comparison to similar EIA determinations in serum or plasma. Additional sources of variability could be the standardization of the blood sampling procedure (which is performed by hundreds of different nurses from 84 hospitals of the Veneto Region) and variability in the elution of the trypsinogen from the filter paper. The control distribution together with the values of the CF patients of Figure 3 (A) gives a general impression of a very low overlap of the two populations. It should also be noted that the CF subject with the lowest trypsinogen concentration shown in Figure 3 (18 ~g/L) was affected by meconium ileus, which can be a cause of false negative cases in CF neonatal screening (14). Also, sensitivity and specificity in the diagnosis of CF are better evaluated at several cut-off levels as shown in the profile of a ROC curve. The profile for the RIA method (Figure 3 (B)) shows a very high diagnostic accuracy. A ROC curve for the EIA method obtained with actual measurements will be available only after some years of collection of data from CF patients. Considering the assumptions and limitations associated with simulations, the simulated ROC curve for the EIA method also shows a very high diagnostic accuracy. In conclusion, the calibration standard material in the EIA method as commercially distributed by AGEN must be improved. The method is quick and simple and shows precision characteristics and diagnostic accuracy equal to or better than those of an RIA method. Acknowledgments We wish to thank: Prof. U. Lippi for revising the manuscript, Prof. G. Berton, Dr. M. C. Dechecchi, Dr. A. Tamanini for helpful discussions, Dr. C. Parmelli for the software for data transformation, V. Stanzial for assistance in data analysis, A. Facchin, P. Faggionato, L. Pattini, S. Pizzini, C. Zampini for excellent technical assistance, E. Bravi for assistance in graphic presentation. This work was partly supported by grants from the "Progetto Finalizzato: Ingegneria Genetica e Basi Molecolari delle Malattie Ereditarie," CNR, Rome, Italy. References
1. Crossley JR, Elliot RB, Smith PA. Dried-blood spot screening for cystic fibrosis in the newborn. Lancet 1979; 1: 472--4. 2. Davidson AGF, Wong LTK, Kirby LT, Applegarth DA. Immunoreactive trypsin in cystic fibrosis. J Ped Gastroenterol Nutr 1984; 3 (Suppl. 1): $79-88. 3. Bowling FG, Watson ARA, Rylatt DB, Elliott JE, Bunch RJ, Bundesen PG. Monoclonal antibody-based enzyme immunoassay for trypsinogen in neonatal screening for cystic fibrosis. Lancet 1987; 1: 826-7. 4. Bowling FG, Brown ARD. Newborn screening for cystic fibrosis using an enzyme linked immunoabsorbent assay (ELISA) technique. Clin Chim Acta 1988; 171: 257-62. CLINICAL BIOCHEMISTRY,VOLUME 23, JUNE 1990
EIA METHOD FOR IRT 5. Mastella G, Pederzini F, Girella E, Righetti G, Zanchetta M, Rizzotti P. Cystic fibrosis screening policy. Lancet 1984; 2: 575-6. 6. Malvano R, Marchisio M, Massaglia A, et al. Radioimmunoassay of trypsin-like substance in human serum. Scand J Gastroenterol 1980; 15 (Suppl. 62): 3-10. 7. Burlina A, Tonon M, Perobelli L, Zanchetta M, Rizzotti P, Plebani M. A new blood spot method for immunotrypsin. IRCS Med Sci 1982; 10: 951-2. 8. Solberg HE. The theory of reference values. Part 5. Statistical treatment of collected reference values. Determination of reference limits. J Clin Chem Clin Biochem 1983; 23: 749-60. 9. Armitage P. Statistical methods in medical research. Oxford: Blackwell Scientific Publications, 1971. 10. Buttner J, Borth R, Bourtwell HJ, Broughton PMG,
CLINICAL BIOCHEMISTRY, VOLUME 23, JUNE 1990
11. 12. 13. 14.
Bowyer RC. Approved recommendation (1978) on quality control in clinical chemistry. Part 2. Assessment of analytical methods for routine use. J Clin Chem Clin Biochem 1980; 18: 78-88. Swets JA. Measuring the accuracy of diagnostic systems. Science 1988; 240: 1285-93. Fuentes-Arderiu J, Sierra MT, Panadero AM. No more assumptions about whether control results have a Gaussian distribution. Clin Chem 1988; 34: 769-70. Zucchelli GC, Piro MA, Malvano R. Radioimmunoassay of trypsin-like immunoreactivity in dried blood spots. J Nucl Med All Sci 1982; 26: 35-9. Heeley AF, Heeley MF. Biochemical screening of the neonatal population for the early detection of CF in east Anglia 1980-1986. Insights into Paediatrics 1987; 1: 35-43.
219
