| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1 Department of Clinical Chemistry, University Hospital MAS, S-205 02 Malmö, Sweden.
aAuthor for correspondence. Fax 46-40-336286; e-mail Joyce.Carlson{at}klkemi.mas.lu.se.
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Methods: Data from Beckman Paragon CZE 2000 electropherograms were compared with quantitative protein data from >800 routine clinical samples. Algorithms were designed to produce semiquantitative analyses of major proteins and to define different patterns of inflammation based on the electropherogram.
Results: The algorithms produced reliable semiquantitative evaluations of prealbumin, albumin, 
1-antitrypsin, haptoglobin, and transferrin, but were less accurate for 1-acid glycoprotein. Some genetic variants of albumin and deficiency variants of 1-antitrypsin were easily recognized. Complex clinical traits such as degree and type of inflammation could be evaluated. When used together with previously developed algorithms addressing immunoglobulins, the new algorithms provide relevant clinical interpretation. Selected outputs indicate the need for reflex testing or evaluation by specialists.
Conclusions: Automation of both electrophoresis and interpretation can provide a rapid, inexpensive, standardized analysis that can hopefully improve the diagnostic information and clinical outcome for large groups of patients. It also provides objective criteria for clinical interpretations, to be validated or adjusted in future clinical studies.
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
In this study, we separated serum proteins by capillary zone electrophoresis (CZE), a rapid, cost-effective method that has been validated against classic gel electrophoresis (4)(5)(6)(7)(8)(9). Capillary electrophoresis produces digital data accessible to mathematical analysis. We have previously shown that computerized algorithms can detect MCs in the absorbance pattern with high sensitivity and specificity (10). In this study, that expert system is extended to produce semiquantitative analyses of other serum proteins. Qualitative observations of genetic variants and drug or bilirubin interaction with albumin (Alb) can also be made. Mathematical factors are suggested to define different patterns of inflammation, and the results of algorithms addressing these patterns in electrophoretic curves are compared with the same algorithms used on quantitative protein data. The goal of the study is to provide the alternative of rapid, inexpensive, fully automated interpretations and to suggest objective criteria for clinical interpretations so that these criteria may be validated or adjusted in future clinical studies.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
1-acid glycoprotein (AAG), 1-antitrypsin (AAT), haptoglobin (Hpt), and C-reactive protein (CRP) in a total of 828 serum samples from adults submitted for routine clinical protein analysis from February to April 2000, and prealbumin (Prealb), 2-macroglobulin (AMG), transferrin (Trf), and complement factor C3 (C3) in a total of 39 samples submitted for routine clinical protein analysis during August 2000 were first quantified on a Beckman Immage instrument (Beckman Coulter Instruments) using the manufacturer’s instructions and reagents calibrated against Certified Reference Material (CRM) 470 (11). The interassay imprecision (CV) of the method is <10% for Alb, <12% for C3, and <4% for other proteins. Age- and gender-specific reference intervals for specific proteins have been recommended by EQUALIS, External Quality Assurance for Laboratory Medicine in Sweden (12). From these reference intervals, means were calculated. For many algorithms, specific protein concentrations in g/L were converted to multiples of this mean (MoM g/L). Simultaneous (within 5 days) analysis of serum proteins and serum bilirubin (S-bili) had been performed for 807 samples analyzed at the laboratory. These samples were used for qualitative evaluation of Alb algorithms. Electropherograms from 84 patients with the common AAT genetic variants protease inhibitor (PI) MM (n = 22), MZ (n = 23), MS (n = 3), SZ (n = 8), FM (n = 2), FZ (n = 2), and ZZ (n = 24), established by isoelectric focusing (13), were available at the laboratory that is the national reference laboratory for AAT phenotyping. Serum iron (S-Fe), total iron binding capacity (TIBC), and iron saturation (S-Fe/S-TIBC) data obtained within 5 days of CZE were available for 136 samples.
Proteins were separated by capillary electrophoresis in a Paragon CZE 2000 (Beckman Coulter), containing seven capillaries, each 20-cm long with an i.d. of 25 ± 3 µm, and with absorbance detection at 214 nm. Electrophoresis was performed using the manufacturer’s instructions and reagents, including two internal standards that migrated outside of the serum protein electropherogram. These substances allow normalization of absorbance curves with reference to absorbance and retention time by version 1.5 software. Absorbance data were exported from version 1.5 through an RS-232 interface to a peripheral personal computer on which mathematical algorithms were developed and tested using Microsoft Visual Basic 6.0 (Microsoft Corporation).
Variability in the position of the Alb peak was evaluated in a histogram, and limits were chosen for acceptance of each curve for further automatic interpretation of the electropherogram. Ranges of retention times for the absorbance peaks of all other proteins were then determined. Examples of measurements done in the electropherogram are illustrated in Fig. 1
.
|
quantification of specific proteins
Results of algorithms calculating peak absorbance (PA), PA minus nearest cathodal valley absorbance (PA-VA), baseline area under the curve (AUCbl), and valley-to-valley integrated areas under the curve (AUCv-v), addressing the position of each specific protein, were compared with results of immunonephelometric analyses of the respective protein by linear regression analysis. The algorithm with the greatest correlation coefficient (r2) or that providing the greatest AUC in a ROC curve (SPSS software, release 10.1.0; SPSS Inc.) was selected (Fig. 2
). The slopes and intercepts of optimal regression curves were used to convert absorbance measurements to protein concentrations (g/L). These values were then divided by the age- and gender-appropriate mean protein concentrations, using the established reference intervals for the Swedish population, and are referred to as MoM CZE values. Exceptions to this approach are presented individually below. Discriminator values for some algorithms were chosen to provide optimal sensitivity and specificity.
|
qualitative factors
Alb.
The anodal width of the Alb peak at one-fourth of the PA, the ratio of this width to Alb PA, the ratio of the anodal/cathodal widths at one-fourth PA, and the distribution of these values were evaluated for 680 samples with S-bili <20 µmol/L, 86 samples with 20 
S-bili < 40 µmol/L, and for 41 patients with S-bili >40 µmol/L. Anodal and cathodal widths of the Alb peak were also evaluated for three samples containing known Alb variants (14), compared with the remaining 804 samples for which S-bili was available.
1 region.
Samples with verified wild-type AAT (PI MM) phenotype (n = 22) were first analyzed to establish a range of retention times for the wild-type AAT phenotype. The maximum absorbance within that range was then defined as PA AAT. A peak corresponding to AAG was sought anodally from the AAT peak, and a range of retention times was determined. The ratio of PA AAG to PA AAT was evaluated for samples with known AAT phenotype.
2 region.
Algorithms using PA, PA-VA, AUCbl, and AUCv-v for the 2 region were compared with protein concentrations of Hpt, AMG, and Hpt plus AMG (n = 39), and peak positions were compared with the ratio of Hpt to Hpt plus AMG for these samples. The Hpt peak position was also compared with the positions of Alb and Trf peaks in the same sample. The distribution of retention times for all 808 samples was plotted, and algorithms comparing PAs and AUCbl for three groups of 20 samples each having the shortest, mean, and longest retentiontimes were compared with Hpt concentrations. Discriminator values for the detection of decreased Hpt concentrations were also evaluated.
ß region.
C3 degradation in serum samples was evaluated by incubating six different serum samples at 8 °C the 1st day and at room temperature for 3 additional days. The PA-VA was calculated after daily electrophoresis, and measurements of C3 by immunonephelometry were also performed daily.
complex clinical traits
Inflammation.
To evaluate degree of inflammation, mean MoM CZE values for AAT, AAG, and Hpt were calculated. These values were then compared with the mean MoM g/L derived from immunonephelometry. No inflammation was defined as a mean MoM 1.5, mild inflammation as 1.5 < mean MoM 2.25, moderate inflammation as 2.25 < mean MoM 3, and severe inflammation as mean MoM >3. This classification based on MoM CZE was compared with that based on MoM g/L. Discrimination between degrees of inflammation was also evaluated by comparing the AUCbl and AUCv-v in the (1 + 2) region with the sum of the immunonephelometric quantities for AAT, AAG, and Hpt in g/L. These algorithms and protein quantities were also compared with CRP concentration in g/L.
Additional internal comparisons of MoM CZE values for these three proteins within the same sample were performed. Type of inflammation was arbitrarily defined (1) as "harmonic" if each MoM was within 20% of the mean MoM and as "connective-tissue constellation" if the AAT MoM was at least 20% lower and the Hpt MoM was 20% greater than the mean MoM. "Selective AAG elevation" required that the AAG MoM was >1.2 times the mean MoM, that both the AAT MoM and the Hpt MoM were less than the mean MoM, and that the criteria for harmonic inflammation were not fulfilled. "Liver constellation" (1)(15)(16) was considered if the AAT MoM:AAG MoM ratio was >1.2 and the Hpt MoM was <0.7 times the mean MoM. "Hemolysis" was suspected if the AAT MoM:AAG MoM ratio was >0.8 and <1.2 and the Hpt MoM was <0.5 times the mean MoM. All samples fulfilling none of the above criteria were classified as "other". Results of these classifications based on MoM CZE were compared with those for the same definitions based on immunonephelometric results (MoM g/L) in Table 1
.
|
Estrogen influence.
A protein pattern suggesting an estrogen effect was considered when AAT was >1.68 g/L and Trf was >3.26 g/L (17). Samples derived from female patients 
18 years of age were evaluated for potential estrogen influence by calculating AAT and Trf concentrations from absorbance data using the algorithm of optimal correlation.
Iron deficiency.
Results of the Trf algorithm were compared for 92 samples with normal (17–52%) and 53 with low (<17%) iron saturation. The Trf PA-VA for samples with and without iron deficiency was compared with the degree of inflammation. Samples were simultaneously evaluated for estrogen influence and for M-components in the ß1 region.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
quantification of specific proteins
The optimal correlations between CZE absorbance measurements and immunologic concentrations are presented in Fig. 2
. For AAT and AAG, 31 samples were excluded because they were considered to contain an AAT variant according to the algorithms presented below. Algorithms measuring PA alone correlated considerably better to AAT and AAG concentrations (r2 = 0.84 and 0.62, respectively) than did algorithms subtracting the valley absorbance cathodal to the AAT peak from PAs for these proteins (r2 = 0.76 and 0.29, respectively).
Hpt.
It was noted that Hpt PA occurred at different retention times, dependent on genetic subgroup. For this reason, three subgroups of 20 samples each, representing the extreme anodal, central, and extreme cathodal positions, were evaluated separately. Slopes of linear regression curves comparing AUCbl of the 2 region to Hpt g/L differed for these groups, being 875 000, 919 000, and 1 000 000, which can be compared with molecular masses of 86 kDa for Hpt 1-1, 93 kDa for Hpt 1-2, and 100 kDa for Hpt 2-2. The correlation coefficients, for these curves were 0.80, 0.74, and 0.73, respectively.
We attempted to further improve the algorithms by expressing Hpt peak position as a fraction of the distance from the Trf peak to the Alb peak in the same sample, thus reducing imprecision attributable to capillary diameter or other variables. Relative peak position was plotted against relative Hpt content expressed as the ratio of Hpt g/L to Hpt g/L plus AMG g/L [Fig. S1 in the data supplement available through the online version of this article at Clinical Chemistry Online (http://www.clinchem.org/ content/vol48/issue7/)], further demonstrating that samples with predominant Hpt content had an anodal position and that AMG has an apparent cathodal peak position at 0.38 the distance between the Trf and Alb peaks. By comparing the peak position with published mobilities of Hpt subtypes on starch gel (18), we inferred that the anodal peak position at 0.476 the distance from Trf to Alb corresponded to Hpt subtype 1-1 and that the cathodal peak position at 0.44 the distance from Trf to Alb corresponded to Hpt 2-2. Samples were then sorted according to relative absorbance at these retention times. The ratio of the absorbance of the 0.44 relative retention time peak to the absorbance of the 0.476 peak was <1.10 for 109 samples. For these samples, the absorbance of the 0.476 relative retention time peak correlated to HPT g/L (r2 = 0.95). The absorbance of the 0.476 relative retention time peak for the remaining 676 samples correlated to Hpt g/L (r2 = 0.75), compared with the absorbance of the 0.44 relative retention time peak for these samples (r2 = 0.83). In all cases, correlations between all attempted algorithms for the samples with cathodal Hpt peaks were inferior to those for the anodal peaks because of superimposition of variable amounts of AMG.
Using back-calculations from these algorithms to identify the (10 of 808) cases with Hpt <0.06 g/L, we constructed a ROC curve with an AUC of 0.910 [Fig. S2 in the data supplement available through the online version of this article at Clinical Chemistry Online (http://www.clinchem.org/content/vol48/issue7/)]. Thus, a sensitivity of 90% corresponded to a specificity of 88.5% and produced a positive predictive value (PPV) of 8.9% and a negative predictive value (NPV) of 99%. For 100% sensitivity, the specificity decreased to 66% with a PPV of only 3.6% (NPV of 100%). These values can be compared with those obtained using a complex algorithm (see below and Table 1
).
ß region.
Values for C3 PA-VA and C3 g/L after storage of serum samples are presented in Fig. 3
and show that <50% of the peak remains after 4 days, compared with the immunologic measurements of C3, which increase slightly during the same time.
|
qualitative factors
Alb.
ROC curves were compared to evaluate different algorithms to identify hyperbilirubinemia; the best showed a sensitivity and specificity of the anodal width (at one-fourth PA) for identifying two different concentrations of S-bili, 20 and 40 µmol/L, with AUCs of 0.659 and 0.846, respectively [Fig. S3 in the data supplement available through the online version of this article at Clinical Chemistry Online (http://www.clinchem.org/ content/vol48/issue7/)]. The anodal width (mean ± 1 SD) for the Alb peak was 517 ± 106 for 680 samples with S-bili <20 µmol/L, 555 ± 155 for 86 samples having S-bili of 20–40 µmol/L, and 766 ± 213 for 41 samples with S-bili 40 µmol/L. The corresponding cathodal widths were 516 ± 116, 519 ± 120, and 563 ± 160. A discriminator value of >729 for anodal width identified 31 of 127 samples with S-bili >20 µmol/L (115 ± 85 µmol/L) compared with S-bili of 43 ± 59 µmol/L for the 96 samples not identified (sensitivity, 24%; specificity, 96%; PPV, 54%; and NPV, 87%). The same limit identified 23 of 41 samples with S-bili >40 µmol/L (sensitivity, 56%; specificity, 96%; PPV, 40%; and NPV, 98%). Cathodal widths at one-fourth PA for three samples containing the relatively common Alb Arg-2Cys variant were 1045, 908, and 823 compared with a range of 203–733 (mean, 517 ± 55) for the remaining 805 samples. A discriminator value of >800 for the cathodal width produced 100% sensitivity and specificity compared with visual evaluation of this Alb variant.
1 region.
The maximum absorbance within the defined time range for AAT was defined as PA AAT. The ratio PA AAG:PA AAT was calculated. This ratio was 1.13 ± 0.12 (range, 1.01–1.63) for all cases of PI ZZ (n = 24), 1.00 ± 0.08 (range, 0.85–1.12) for PI SZ (n = 8), 1.03 and 1.06 for PI FZ (n = 2), and 1.16 and 1.09 for PI FM (n = 2). For comparison, the mean PA AAG:PA AAT ratio was 0.871 ± 0.073 for PI MZ (n = 22), 0.854 ± 0.048 for PI MS (n = 3), and 0.728 ± 0.077 for PI MM (n = 22). Use of a discriminator value for the PA AAG:PA AAT ratio >0.98 provided a sensitivity of 100% (24 of 24) and a specificity of 75% (47 of 63) to identify PI ZZ alone, and a sensitivity of 95% (40 of 42) and specificity of 100% (45 of 45) for distinguishing PI ZZ, SZ, FM, and FZ from PI MM, MZ, and MS. This approach provided much better discrimination than PA AAT alone.
complex clinical traits
Inflammation.
The MoM CZE for AAT, AAG, and Hpt individually and the mean MoM CZE for these three proteins together were compared with those for the comparison method with r2s of 0.88, 0.64, 0.85, and 0.86, respectively. Similarly, the comparison of AUCv-v for the (1 + 2) region to the sum of AAT g/L + AAG g/L + Hpt g/L had a r2 of 0.84 and that for AUCbl had a r2 of 0.78. The distribution of mean MoM CZE for groups of samples having no, mild, moderate, or severe inflammation, based on mean MoM g/L, showed only a slight overlap between groups with adjacent degrees of inflammation [Fig. S4 in the data supplement available through the online version of this article at Clinical Chemistry Online (http://www.clinchem.org/content/vol48/issue7/)]. The sum of the immunonephelometric measurements for AAT, AAG, and Hpt was compared with that for CRP (r2 = 0.51). Similar results were found when we compared AUC (1 + 2) to CRP mg/L (r2 = 0.47).
Using the arbitrary definitions presented in Materials and Methods, we evaluated patterns of acute-phase proteins as presented in Table 1
. Each row represents a classification based on the comparison method, and each column represents the classification based on the electropherogram. As can be seen, numerous discrepancies occurred. Thirteen cases judged as "liver" by CZE criteria had AAT MoM:AAG MoM g/L scores between 0.946 and 1.2, disqualifying them for the liver pattern. Eight of these 13 samples had Hpt MoM:mean MoM g/L scores <0.7, compatible with the concept of the liver pattern, and the sample with the lowest AAT:AAG ratio was an AAT PI MZ sample. Similarly, of the 10 samples with Hpt <0.06 g/L, which could be considered to have hemolysis, 4 had a liver pattern (AAT MoM:AAG MoM >1.2) by the comparison method, and 5 had this pattern by CZE data alone.
Iron deficiency.
The absorbance difference between PA for Trf and the cathodal valley was compared between groups of samples with low (<17%) and normal iron saturation (17–52%), showing a considerable overlap between the groups [Fig. S5 in the data supplement available through the online version of this article at Clinical Chemistry Online (http://www.clinchem.org/content/vol48/issue7/)]. When the reference interval for transferrin (1.94–3.26 g/L) was plotted in the curve, there were only three samples in the normal iron saturation group with absorbance above this interval. These three samples were all obtained from women 20–52 years of age, in whom influences from estrogen could be suspected. In both groups, with and without iron deficiency, samples with a Trf PA-VA below the reference interval nearly all showed significant signs of inflammation.
presentation of results and interpretations
Algorithm outputs expressed as phrases (Table 2
) were compiled in the PC-based decision support system to produce a clinically relevant interpretation. Comments concerning degree of inflammation as well as concentration and distribution of immunoglobulins were always provided, even if normal. MCs were described with comments on background immunoglobulin synthesis when present and negated when absent for all patients >40 years of age. Other comments were used only when the algorithm produced abnormal results. Suspected genetic variants of Alb or AAT, MCs, significant oligoclonality, and subnormal immunoglobulin concentrations produced an automatic warning that individual evaluation was required.
|
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
1, 2, ß, and 
regions were compared with a register comprising 29 (24) or 55 (25) specific outputs, each suggesting a clinical condition. Neither of these reports presented an objective evaluation of the results. Ivandic et al. (26) produced a knowledge-based system using quantitative analysis of creatinine, total protein, and four specific proteins in combination with a test strip screening to classify urine samples into one of four major diagnostic groups. These outcomes have been validated against a clinical diagnosis with a reported concordance of 98%.
An ambitious knowledge-based system has been produced from data accumulated over the past 30 years by Ritchie (3), with inputs of demographic data, medical history, results of CRM 470 standardized protein quantification, and autoimmune antibodies. The "tilt" of the acute-phase reaction described by Laurell (1) has been refined to a relative "multiple of the (age and gender specific) median" by Ritchie to allow use of the same algorithms and discriminator values for all patient groups (3)(23). This complex program can generate paragraphs of highly variable text in nearly infinite combinations and permutations. Drawbacks of the system are that it requires numerous quantitative analyses and considerable time to produce a report (3). The MoM concept used here has been adapted to use accessible mean values in the current study.
We designed this automated screening system to produce relevant interpretations with access to no clinical information aside from age and gender. Although we welcome access to specific clinical information and inquiries, it is our experience that information presented on referrals is frequently scanty and preliminary and that evaluation of suspected unusual conditions (e.g., amyloidosis) always requires specific tests not suitable for automation. A fully written referral leading to individualized evaluation and reflex testing thus remains an alternative in these cases.
For correct interpretation, each algorithm must identify the appropriate segment of the absorbance curve. The retention time for the large, easily identified Alb peak migrating late in the electrophoretic pattern was therefore chosen as a criterion for acceptance of curves for automated interpretation. In this report, interpretation of the electrophoretic curve alone is compared with interpretation based on quantitative data from a panel of specific proteins. Determination of protein concentrations by absorbance measurements is at best an approximation, affected by protein concentration (CRP and ceruloplasmin are rarely visible) and projection. The unexpectedly poor correlation (r2 = 0.85) between the area of the freely projected Alb peak and immunonephelometric quantities reflects imprecision in both the former (CV = 6.8%; n = 50) (10) and latter (CV = 9.6%; n = 78) methods. On the other hand. the correlation between the AAG algorithm and g/L (CV = 3.1%) is poor (r2 = 0.62), as expected, because the variable carbohydrate content produces a low flat curve, and this curve overlaps with variable amounts of -lipoprotein. Good correlations to Hpt concentration could be obtained by comparing its peak position to those for Alb and Trf and using specific algorithms for the different genetic subtypes. Hpt 1-1 consists of two 1- and two ß-chains with a molecular mass of 86 kDa, whereas Hpt 2-2 comprises two 2- and two ß-chains, which readily polymerize, giving an average molecular mass of 100–200 kDa. Thus, absorbance measurements at 214 nm, which reflect the number of peptide bonds in a protein, may well agree better with the actual concentration of Hpt in g/L than values derived from antigen-antibody precipitation of a heterogeneous population of molecules. By tradition, AMG has not been routinely quantified as a part of serum electrophoresis investigations; therefore, no major effort was made to design an AMG algorithm.
Algorithms comparing internal differences within a single electrophoretic curve are subject to error in a single dilution/measurement step, compared with compilation of 9–10 different immunonephelometric determinations, each subject to independent imprecision of 3–10%. CZE is not currently calibrated against CRM 470, but back-calculations from linear regressions adjusted for age- and gender-relevant CRM 470-based reference intervals can be used. The imprecision of such approximations is attributable to the combined imprecision of CZE and nephelometric measures.
Qualitative differences can also be evaluated by algorithms. Anodal extension of the Alb peak, as seen in hyperbilirubinemia or drug interactions (penicillin/penicillamine), is of minor importance in the interpretation, but its presence may provide additional clinical information when evident. Although the sensitivity for identifying hyperbilirubinemia (S-bili >20 µmol/L) by this method is poor, its specificity and NPV are good in a representative material of clinical samples. The possibilities of hyperbilirubinemia and/or drug interactions can usually be easily distinguished by the clinician.
The cathodal genetic variants of Alb were easily distinguished because of cathodal broadening of the Alb peak. Whereas genetic variants of Alb apparently lack clinical relevance (27), one form of bisalbuminemia caused by abnormal cleavage of the Alb propeptide in cases of pancreatic disease mimics anodal bisalbuminemia (28) and would be detectable by these algorithms.
The clinically relevant AAT variants PI ZZ, Z -, - -, and SZ enhance the risk for development of emphysema, especially in smokers (29). The ZZ variant is also related to an increased risk for chronic liver disease (30). A PA AAG:PA AAT ratio >0.98 produces an excellent distinction between PI FM, SZ, FZ, and ZZ vs other variants. Visual inspection and/or a reflex order for AAT quantification may then separate the PI FM from the deficiency variants. The common B, C, and D genetic Trf variants lack clinical relevance (31) but may raise suspicion of MCs. Broadening of the peak or a double peak in the ß1 region produces a warning for MC, and the curve is subjected to manual interpretation to distinguish these possibilities (10).
The instability of C3 in serum samples over time is visualized as degradation of the C3 peak in the electropherogram, whereas a concomitant slight increase in C3 is measured by immunonephelometry (Fig. 3
). Cleavage of the C3 molecule alters the electrophoretic mobilities of its products, diminishing the original C3 peak. Cleavage apparently also increases the molarity of epitopes recognized by the polyclonal antisera in the immunologic method.
A combination of algorithms for AAT, AAG, and Hpt was used to evaluate the inflammatory response in the electropherogram. Fibrinogen was excluded because the study concerned serum samples. The mean MoM for AAT, AAG, and Hpt, based on absorbance information, was shown to correlate well to the mean MoM based on immunologic measurements, and the limits chosen for mean MoM CZE to classify mild, moderate, and severe inflammation corresponded well with our current subjective evaluations of inflammation.
The widespread use of CRP as a marker of inflammation has largely replaced the use of erythrocyte sedimentation rate in many clinical settings. Some colleagues suggest that protein electrophoresis for evaluation of inflammation is no longer of interest, because rapid and inexpensive analysis of CRP provides the same information. For this reason it is interesting to note the poor correlations between CRP and degree of inflammation as evaluated by both CZE and by acute-phase protein quantities by immunonephelometry. This depends partly on differences in the half-life of CRP compared with other acute-phase reactant proteins. In addition to degree of inflammation, internal comparisons between acute-phase reactant proteins provides information on type of inflammation. The additional evaluation of immunoglobulins provides a multifaceted perspective not attainable from CRP measurements alone.
Two major patterns have been described for the inflammatory response in experimental systems. A type 1 response, stimulated by interleukin-1, demonstrates increased synthesis of AAG, CRP, C3, and serum amyloid A. A type 2 response, stimulated by interleukin-6, shows increased synthesis of fibrinogen, Hpt, ceruloplasmin, AAT, and antichymotrypsin. A type 2 response may induce a type 1 reaction (but not the reverse) and cause increased synthesis of most known acute-phase reactants (32)(33)(34). In vivo this type of pattern corresponds well to the term harmonic inflammation described by Laurell (1). Catabolism and redistribution also influence the serum concentration of proteins. This is exemplified empirically (1) by a "connective tissue pattern" seen in polymyalgia rheumatica and systemic lupus erythematosus and by selective increases in the freely filtered AAG when glomerular filtration rate is diminished. A liver constellation of acute-phase reactants has been reported in viral hepatitis, chronic active hepatitis, primary biliary cirrhosis, advanced alcohol-related disease, and even in congestive heart failure, but not in biliary obstruction or liver metastases (1)(15)(16).
In this study we used the MoM concept for the acute-phase proteins to compare relative changes to the mean inflammatory score. The impact of the AAG algorithm was minimized where possible, because of its relatively poor correlation to AAG g/L, by comparing the MoM AAT and MoM Hpt to the mean MoM for all three proteins. The discrepancies in classification seen in Table 1
are generally between "neighboring concepts", i.e., samples judged as having liver disease by g/L criteria are never interpreted as connective tissue disease or "selective orosomucoid elevation" by CZE. A significant overlap between the classifications "liver disease" and "hemolysis" occurred with both CZE and quantitative criteria, as expected, reflecting that both classifications are based on increased Hpt consumption. Among samples identified as having liver constellation by CZE but not by protein quantities, several (13 of 26) had AAT MoM:AAG MoM ratios of 0.946–1.2 by immunonephelometry, and 8 of these had a Hpt MoM <0.5 of the mean MoM by both methods, suggestive of hemolysis. Similarly, of the 10 samples with Hpt <0.06 g/L, which may be regarded as indicative of hemolysis, 4 samples had clear "liver patterns" based on g/L values, and 5 had liver interpretation by CZE.
Generally, algorithms for acute-phase reactant patterns based on CZE produced fewer specific diagnostic evaluations than those comparing specific protein quantities, i.e., 46% of samples were regarded as "other" compared with 35% of samples classified based on protein quantities These results are adequate for a test designed for screening and are preferable to a large number of false-positive results. The lack of diagnostic precision is expressed appropriately in interpretive comments (see Table 2
) such as "compatible with but not specific for connective tissue disease" or "signs of mild inflammation compatible with parenchymal liver disease and/or hemolysis". The large number of discrepancies concerning the evaluation "selective AAG elevation" are probably attributable to the poor correlation of the specific AAG algorithm. Because of these discrepancies, automatic comments about the presence and/or clinical implications of this pattern should be avoided. Many of the samples judged as "other" had nearly normal patterns and should not be designated to any of the specific categories.
The presence of estrogen influence should affect evaluations of potential liver disease and/or iron deficiency. In the estrogen recognition algorithm, evaluation is based on sex, AAT, and Trf (17). Ceruloplasmin is often increased in these cases, but is not visible in the electropherogram. Estrogen influence was detected by the algorithm in 20 cases. These were not included in Table 1
because quantitative results for Trf were available for only 30 other patients and comparison between the two methods was therefore not possible. A Trf concentration calculated from an electropherogram above the reference interval with no suspicion of estrogen effects is highly specific for iron deficiency. It is interesting that all samples with low Trf demonstrated moderate to severe inflammation, independent of iron saturation. Because the degree of inflammation is approximated by another algorithm, a low Trf value in the absence of inflammation can be used to indicate iron overload as seen in hemochromatosis, hemolytic anemia, after multiple transfusions, or during iron therapy.
Synthesis of relevant text fragments reflecting results of these algorithms and those addressing immunoglobulins (10) provides an integrated clinical interpretation. Alternative phrasing options are illustrated in Table 2
. Certain comments, e.g., suspected MC or AAT deficiency, automatically produce a warning. All automated interpretations should be labeled as such and should be approved by competent personnel. Warnings can direct samples to specialist evaluation for further editing and can be used to initiate reflex testing, such as AAT or immunoglobulin quantification or immunofixation.
The numerical factors used in the algorithms are accessible for future adjustments according to results of potential clinical validation studies. Generally speaking, sensitivity will be increased by adjusting numerical factors nearer the value 1.0, at the price of decreased specificity.
In conclusion, mathematical algorithms for analysis of the absorbance curve produced by CZE of serum samples can produce useful semiquantitative evaluations of individual proteins. They can direct further reflex testing or evaluation by specialists. Programmed interpretations are further amenable to multicenter clinical validation. Such interpretation is not intended to replace quantification of all individual proteins or individualized analyses in specific clinical settings, but rather to provide the alternative of a rapid, inexpensive, and standardized analysis suitable as a screening test early in diagnostic evaluations. Such use of serum protein analysis would hopefully improve the diagnostic information and clinical outcome for large groups of patients.
note added in press
After electrophoresis of all the samples included in this study, the buffer characteristics were changed slightly by Beckman Coulter. This has caused slight differences in electrophoretic mobilities and peak absorbances.
|
Acknowledgments |
|---|
|
Footnotes |
|---|
1-acid glycoprotein; AAT, 1-antitrypsin; Hpt, haptoglobin; CRP, C-reactive protein; Prealb, prealbumin; AMG, 2-macroglobulin; Trf, transferrin; C3, complement factor 3; CRM, Certified Reference Material; MoM, multiple(s) of mean; S-bili, serum bilirubin; PI, protease inhibitor; S-Fe, serum iron; TIBC, total iron binding capacity; PA, peak absorbance; PA-VA, PA-cathodal valley absorbance; AUCbl, baseline integrated area under the curve; AUCv-v, valley-valley integrated area under the curve; PPV, positive predictive value; and NPV, negative predictive value. 
|
References |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
1-antitrypsin by electrofocusing. Clin Chem 1982;28:219-225.1-Antitrypsin and other acute phase reactants in liver. Acta Med Scand 1980;207:79-83.[ISI][Medline]
[Order article via Infotrieve]
-1-antitrypsin deficiency PiZ. Acta Med Scand 1978;204:345-351.[ISI][Medline]
[Order article via Infotrieve]
1-antitrypsin deficiency. Bianchi Let al eds. Acute and chronic liver diseases: molecular biology and clinics 1996:231-248 Kluwer Academic Publishers Dordrecht. .
The following articles in journals at HighWire Press have cited this article:
|
|
 |
|
  E. Piitulainen, J. Carlson, K. Ohlsson, and T. Sveger {alpha}1-Antitrypsin Deficiency in 26-Year-Old Subjects: Lung, Liver, and Protease/Protease Inhibitor Studies Chest, October 1, 2005; 128(4): 2076 - 2081. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 C. Gay-Bellile, D. Bengoufa, P. Houze, D. Le Carrer, M. Benlakehal, B. Bousquet, B. Gourmel, and T. Le Bricon Automated Multicapillary Electrophoresis for Analysis of Human Serum Proteins Clin. Chem., November 1, 2003; 49(11): 1909 - 1915. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. Regeniter, J. U. Steiger, A. Scholer, P. R. Huber, and W. H. Siede Windows to the Ward: Graphically Oriented Report Forms. Presentation of Complex, Interrelated Laboratory Data for Electrophoresis/Immunofixation, Cerebrospinal Fluid, and Urinary Protein Profiles Clin. Chem., January 1, 2003; 49(1): 41 - 50. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
, ratio of peak height to initial peak height; X, ratio of immunologic measurement to initial immunologic measurement.