| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Articles |
1
Department of Clinical Chemistry, University Hospital MAS, S-20502 Malmö, Sweden.
a Author for correspondence. Fax 46-40-33-62-86;
Joyce.Carlson{at}klkemi.mas.lu.se
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Methods: Using the Beckman Paragon CZE 2000 instrument, we
analyzed 711 routine clinical samples, including 95 monoclonal
components (MCs) and 9 cases of Bence Jones myeloma, in both the CZE
and HRAGE systems. Mathematical algorithms developed for the detection
of monoclonal immunoglobulins (MCs) in the 
- and ß-regions
of the electropherogram were tested on the entire material. Additional
algorithms evaluating oligoclonality and polyclonal concentrations of
immunoglobulins were also tested.
Results: CZE electropherograms corresponded well with HRAGE. Only one IgG MC of 1 g/L, visible on HRAGE, was not visible after CZE. Algorithms detected 94 of 95 MCs (98.9%) and 100% of those visible after CZE. Of 607 samples lacking an MC on HRAGE, only 3 were identified by the algorithms (specificity, 99%). Algorithms evaluating total gammaglobulinemia and oligoclonality also identified several cases of Bence Jones myeloma.
Conclusions: The use of capillary electrophoresis provides a modern, rapid, and cost-effective method of analyzing serum proteins. The additional option of computerized decision support, which provides rapid and standardized interpretations, should increase the clinical availability and usefulness of protein analyses in the future.
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
In this study, CZE was used to separate serum proteins, and several computerized algorithms were developed to automatically identify samples with abnormal distribution of immunoglobulins so that further evaluation could later be performed. Major emphasis is placed on identification of MCs. The more subtle distinction of oligoclonal banding and quantitative evaluation of polyclonal immunoglobulins are also addressed.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
1 to 79 g/L
from 83 individual patients (Fig. 1
). Two patients were represented by 3 samples each, 8 patients
were represented by 2 samples, and 73 patients by 1 sample each. Nine
samples were from eight patients with Bence Jones myeloma.
Oligoclonality was seen in 43 samples and was graded from mild (+) to
severe (+++). The remaining 564 samples were referred to as normal,
regardless of immunoglobulin concentrations.
|
capillary electrophoresis
Samples were stored at 8 °C for a maximum of 4 days before
preparation for capillary electrophoresis (CE). Fibrinogen was then
precipitated from 300 µL of plasma by the addition of 6 µL of an
aqueous solution of thrombin (2.3 units/µL; cat. no. T-4648;
Sigma-Aldrich) and incubation for 30 min at 20–22 °C; samples were
centrifuged at 20 800g for 10 min. Separation by CE was
performed on a Paragon CZE 2000 (Beckman Coulter) containing seven
capillaries [each 20 cm x 25 ± 3 µm (i.d.)], with
absorbance detection at 214 nm. The sample (10 µL) was diluted in 20
volumes of signal solution containing two internal standards selected
to define the informative part of the absorbance curve. The original
signal solution was diluted 1:1 with deionized water to reduce ionic
strength, and the injection time was increased from 1.0 to 1.1 s.
Electrophoresis was performed according to the manufacturer’s
instructions, and absorbance at 214 nm was detected at regular
intervals of 2 Hz. Peaks were detected in the order signal substance 1
(SS1); the -, ß-, and 
-globulins; albumin; prealbumin; and
signal substance 2.
data interpretation
Each electropherogram comprising 506 raw absorbance measurements
was imported to a Pentium II personal computer. The Paragon CZE 2000
software, Ver. 1.5, includes an option to export aligned curve
information to another computer through the RS-232 interface.
Mathematical algorithms to analyze the data were developed and tested
using Mat Lab, Ver. 5.2 (Math Works). For some algorithms, raw
absorbance data were distributed among 150 time points along the time
axis between SS1 and albumin, using a preexisting resampling function
in Mat Lab, based on polyphase implementation to produce aligned
curves.
Algorithms were designed to detect MCs in the curve. All 711 curves were exposed to all MC algorithms, but only the 607 curves regarded as oligoclonal or normal according to the reference method were evaluated for oligoclonality. Samples with oligoclonality (n = 43), samples from children <10 years of age (n = 13), and samples lacking any necessary quantitative data (n = 6) were later excluded, leaving 545 samples for quantitative comparisons. Nine sera from patients with Bence Jones proteinuria were evaluated separately. The result of each computer-based algorithm was compared with the conventional results based on quantification and HRAGE.
m-component algorithms
Algorithms developed to identify MCs focused on both the - and
ß-regions of aligned curves to include all MCs in our material.
Algorithm 1 used the first derivative of the electropherogram curve to
identify all peaks in the -region. The "steepness" of
each peak was defined by calculating ratios of the peak absorbance (PA)
with absorbance values at intervals of four measurements before and
after the peak. If a valley occurred within this interval, valley
absorbance was used for comparison. The greater of the two ratios
described the steepness of the peak.
Algorithm 2 divided the baseline integrated the area under the curve
(AUC) in the ß-region by the baseline-integrated AUC in the
-region. Algorithm 3 integrated the valley-valley AUC in the
complement factor C3 peak, and algorithm 4 divided this result by the
PA in the -region. Algorithm 5 determined the number of peaks in the
ß2-region that had a steepness (calculated as
in algorithm 1) >1.33. Samples having more than one such peak were
classified as having an MC. Algorithm 6 integrated the
valley-valley AUC for the transferrin peak.
The principles of algorithms 1–3 are illustrated in Fig. 2
. All 711 electropherograms were subjected to each of
the above algorithms, and the numeric results of samples with MCs were
compared with those without MCs according to the HRAGE evaluation to
determine the discriminator values that provided optimal
sensitivity and specificity for each algorithm (Fig. 3
). The numerical discriminator values determined by this method
are included in Table 1
. Serial dilutions in a normal serum sample of an IgG MC in the
-region with a concentration of 40 g/L were used to evaluate the
detection limit of algorithm 1.
|
|
|
oligoclonality
The number and magnitude of inflection points in the -region of
the curve were used to indicate oligoclonality. A change of sign of the
second derivative of the absorbance curve indicated an inflection
point. The inflection point was included if the slope of the curve as
it passed through zero exceeded a threshold value. A discriminator
value of 0.0150 was chosen for the sum of the slopes of included
inflection points in the -region according to optimal sensitivity
and specificity compared with HRAGE evaluations.
quantification of immunoglobulins
We compared the immunonephelometric values for immunoglobulins
IgA, IgG, and IgM, and their sum with nonaligned (raw) absorbance data,
using several algorithms. PAs and integrated AUC data were in each case
normalized by division with the corresponding values for SS1. The
concentrations of IgG and IgG + IgM were compared with the PA in the
-region. The IgA concentration was compared with the valley
absorbance between C3 and transferrin as well as with a calculated
slope in the -region of the absorbance curve. The sum of the
immunonephelometric values for IgA + IgG + IgM was compared with the
baseline-integrated AUC between the SS1 and haptoglobin peaks after
subtraction of approximated peaks for C3 and transferrin. The
transferrin peak area was approximated as the product of the PA for
transferrin minus the valley absorbance between C3 and transferrin
multiplied by the number of time measurements between these two points.
The C3 peak was similarly approximated using the PA for C3 minus the
valley absorbance between C3 and the -region and the time between
these two measurements.
reproducibility
To test reproducibility, we analyzed serum protein calibrators
(Dako A/S) each 10 times daily on 5 days using all capillaries and
under routine operating conditions. The imprecision (CV) for the Dako
X0940 calibrator with an albumin concentration of 96 g/L was
4.2% for PA measured from baseline and 4.6% for the
baseline-integrated AUC of the albumin peak. Corresponding values for
the Dako X0908 calibrator (albumin, 43.2 g/L) were 6.8% and 6.6%,
respectively. These same electropherograms were used to evaluate the
precision of the numerical results of MC algorithms 1 and 2 and the
algorithm to approximate the sum of polyclonal IgA + IgG + IgM. The CVs
for these algorithms were 0.97%, 4.5%, and 5.4% for the Dako X0940
calibrator and 0.94%, 6.8%, and 5.0% for the Dako X0908 calibrator,
respectively. Imprecision was not calculated for algorithms 3, 4, and
6, which were dependent on differentiation of peaks for C3 and
transferrin, as the C3 peak is degraded in the commercial protein
calibrator solutions used. Imprecision was not relevant for algorithm
5, which produced an integral result.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
. Most MCs were detected by several algorithms. The
sequential use of all six algorithms for MCs on the total 711 samples
detected a total of 94 of the 95 samples judged to contain MCs by
HRAGE, corresponding to a sensitivity of 98.9%. No monoclonal peak or
irregularity could be seen in the CZE curve for one sample from a
92-year-old woman containing an IgG MC estimated as 1 g/L with normal
background immunoglobulin synthesis and no cryoglobulin. Serial
dilutions of a sample with a 40 g/L IgG MC gave a detectable peak
identifiable by algorithm 1 at a concentration of 1 g/L and by the
oligoclonality algorithm at 0.5 g/L.
The algorithms for MCs also identified five of nine samples from
patients with Bence Jones proteinuria. All five contained identifiable
peaks attributable to circulating light chains, despite a lack of
visible bands on HRAGE in two cases. Three of 607 samples lacking MCs
by HRAGE evaluation were also identified, giving a total specificity of
99.5%. The CE curves for these patients are shown in Fig. 4
, A–C. Samples from a 37-year-old patient with inflammation,
hypergammaglobulinemia, and a high C3 concentration (Fig. 4A
), and from
a 55-year-old woman with inflammation and immunosuppressive treatment
(Fig. 4B
) were recognized because of high C3 concentrations by
algorithms 3 and 4. One sample from a 73-year-old man (Fig. 4C
) judged
as +++ oligoclonal on HRAGE had at least three distinct IgG 
bands
visible after IFE (Fig. 4D
).
|
oligoclonality
All 607 samples without MCs were exposed to the algorithm designed
to recognize oligoclonality. The sensitivity of this algorithm for
oligoclonality as a whole was low, 37% (16 of 43) compared with the
subjective evaluation of HRAGE gels with an optimal discriminator value
of 0.0150, but agreement increased from 11% (2 of 19) for mild
oligoclonality (+) to 47% (8 of 17) for moderate (++) and 85% (6 of
7) for severe (+++) oligoclonality, confirming a correlation between
the visual impression of oligoclonality in the gel and irregularity in
the CZE–absorbance curve. An example of a (+++) oligoclonal sample is
presented in Fig. 4
, C and D. Four curves from samples that had been
judged to be normal were identified by the oligoclonality algorithm.
quantification of immunoglobulins
Linear regression analysis was used to evaluate the correctness of
the algorithms for total-immunoglobulin quantification in all samples
lacking an MC or oligoclonality (n = 545). The relationships
between immunonephelometric results in g/L and absorbance data
providing the best correlation coefficients of all algorithms tested
are shown in Table 2
.
|
A specific attempt was made to identify samples with IgA deficiency
(0.09 g/L IgA) from 558 samples, including those from children. Three
samples with IgA 0.09 g/L were identified (sensitivity, 100%) by a
valley absorbance between C3 and transferrin <0.0220. This algorithm
also identified nine false-positive samples with a mean IgA
concentration of 0.76 ± 0.36 g/L (specificity, 546 of 555, or
98%). An algorithm to detect increased polyclonal IgA (>3.65 g/L) at
optimal discrimination (valley absorbance >0.0482) performed with a
sensitivity of 89.7%, and a specificity of 80.1%. The 86 false
positives had a mean IgA concentration of 2.76 ± 0.57 g/L, giving
a positive predictive value (PPV) of only 50%. A similar attempt to
find samples with IgG concentrations above the upper reference limit
(14.9 g/L) for adults >50 years of age, using PAs in the -region
>0.139, identified 13 of 14 such samples (sensitivity, 92.9%) among
398 samples, with 5 false positives, having a mean IgG of 13.7 ±
0.95 g/L, for a specificity of 99% (PPV = 72%).
The association of Bence Jones proteinuria with low total-immunoglobulin concentrations was tested using the baseline-integrated AUC between haptoglobin and SS1 minus the C3 and transferrin areas (see above) normalized with respect to the AUC for SS1. Optimal discrimination between Bence Jones myeloma and other samples lacking MCs occurred at a cutoff of 0.244. Below this value, the sensitivity for detection of Bence Jones proteinuria was 44% with a specificity of 99%. Adjustment of the discriminator value to produce 100% sensitivity identified 70 false-positive samples, greatly reducing the specificity and PPV of the test.
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Although CE was introduced in 1967 by Hjerten (8), it was
first used for the separation of serum proteins in 1983 (9).
Using other capillaries and buffers on the Beckman P/ACE, a single
capillary instrument with highly efficient cooling, other authors
(2)(4)(7) have demonstrated
excellent correlations between the AUC after CE and densitometric
measurements after HRAGE, using a delimit function to compare the
classic -, ß-, and -electrophoretic regions. These authors have
generally noted the disappearance of some very small MCs on CE, as well
as detection of small, typically IgA, MCs by CE that were hidden
because of comigration with transferrin or C3 on HRAGE. IgM MCs were
sometimes distorted or absent from CE, but could be visualized by
alterations in buffer pH and/or ionic strength.
Jenkins et al. (2) have suggested that the extreme pH of the buffer and stringent washing protocols are necessary to minimize adhesion of proteins to the capillary surface. These authors identified all but one (IgM) MC by immunosubtraction CE (IS-CE) and found one IgA MC not seen on HRAGE using this method. Jenkins et al. (2) reported a sensitivity of 94.6% (70 of 74), and Katzmann et al. (4), also using the P/ACE instrument, reported a sensitivity of 98% and specificity of 91% for the detection of MCs, which were superior to the sensitivity obtained by densitometric scans of both HRAGE and cellulose acetate gels. Thus, the method of CE has been well documented to perform reliably for the analysis of serum proteins, with minor discrepancies noted by several authors compared with HRAGE for the detection of MCs.
The Paragon CZE 2000 instrument, with seven parallel capillaries [20 cm x 25 µm (i.d.)], allows rapid throughput for routine clinical analyses. Bienvenu et al. (5) have assessed the performance of this instrument, noting the disappearance of fibrinogen from plasma samples. Fibrinogen was clearly visible in an electropherogram from the P/ACE instrument when identical buffer and running conditions were used. The only apparent difference between instruments is cooling efficiency, suggesting that fibrinogen may precipitate at higher temperatures and may be a cause of the error codes frequently observed by us and others when analyzing plasma samples. After improvements in software, more stringent specifications for capillary diameter, and a more efficient rinsing of capillaries between runs—all modifications recently introduced by the manufacturer—and after the precipitation of fibrinogen from the original plasma samples, we have found that the precision of the method corresponds well with that obtained for specific protein quantification. We could verify excellent precision, not only for evaluation of the albumin peak, which is subject to very little interference by overlapping protein fractions, but also for the specific algorithms discussed in this report. Such precision is a prerequisite for reliability if algorithms are to be useful in routine automated interpretation of CZE curves. It also allows meaningful longitudinal comparisons of repeated samples throughout the clinical course of a single patient.
Knüppel et al. (10) were the first to apply a
rule-based system for the identification of MCs; they reported a
sensitivity of 76%. Kratzer et al. (11) used an artificial
neural network (ANN) to evaluate multiple aspects of densitometric
curves after protein electrophoresis of 446 plasma samples. Although
they did not report on electrophoretic technique or on the number of
monoclonal gammopathies included in the study, they did report a
training set comprising four positive samples and seven samples
with suspected monoclonal gammopathies. These authors used a
preliminary Fourier transformation of the -region of the
densitometric curve and reported 100% success in identifying MCs. They
did not comment on the possibility of MCs being located outside of the
classical -electrophoretic region, nor did they mention the
concentration range of their MCs. Another study by Männer et al.
(12), using a back-propagation ANN that accepts 94 online
input elements from the -region of densitometrically scanned data
(Hitesystem 310; Olympus), achieved a correct diagnosis in 90% (45 of
50) of pathological or suspect samples and a correct diagnosis in 234
of 256 samples lacking MCs (specificity, 91%; PPV, 71%). To attain
these results, a preliminary smoothing transformation was necessary.
ANNs are dependent on access to training sets containing a sufficient
number of examples of each desired output diagnosis to allow
generalization in pattern recognition (13). This condition
is difficult to meet for small MCs in atypical positions. In our
initial experience attempting to create an ANN for MCs, we failed to
identify atypical samples. Our network also failed to identify certain
obvious large MCs in the -region, perhaps confusing them with
polyclonal hypergammaglobulinemia. For this reason, an algorithmic
approach was selected.
One may argue that design of algorithms to identify specific known MCs
in patient sera produces algorithms unsuitable for an unknown
population of samples. In our material, isolated MCs could not be
detected without the specific algorithms 2–6, which address the
ß2-region. The fact that these algorithms were
designed for unique small MCs (as seen in parentheses in Table 2
) did
not prevent their general usefulness for identifying many other cases.
The use of these algorithms has substantially improved our sensitivity
for small components without decreasing specificity. In our material,
no M-component appeared in the 1- or
2-regions, and therefore no algorithms could
be tested in these regions. In analogy to algorithms for the
ß-region, algorithms comparing AUC (2) to
AUC () and AUC (2) to AUC
(1) to detect discrepancies in inflammatory
pattern patterns might be useful.
Although most new cases of myeloma present with large MCs and suppressed background immunoglobulin synthesis, small monoclonal fractions are of major medical interest. They may be present in monoclonal gammopathies of undetermined significance, and in such cases they should be followed because 10–25% of such cases successively develop myeloma (14). Single or multiple small MCs may also be seen in cases of lymphoma and in AL amyloidosis, despite relatively normal immunoglobulin concentrations (15).
In this study, no attempt was made to verify all MCs using the IS-CE
system available with the Paragon CZE 2000 instrument. In our
experience, most MCs >5 g/L can be classified solely on the basis of
quantitative (immunonephelometric) analyses. The small components,
which may comigrate with transferrin or C3, should be verified and
classified by immunofixation. In these cases we have traditionally
performed immunofixation after HRAGE (IFE). Bienvenu et al.
(5) demonstrated complete agreement between IFE and IS-CE,
performed by the separate addition of anti-, anti-, anti-µ,
anti-, and anti-
antisera to the sample and precipitation onto a
solid support before CZE, only for MCs >10 g/L; with IS-CE, they
experienced a 54% failure rate for small components. Results
of IS-CE are particularly difficult to interpret in cases with a high
polyclonal immunoglobulin background and in cases where small peaks
comigrate with other peaks. For these reasons, IS-CE was not performed
in our study.
Many other authors have compared their findings with those of densitometry after HRAGE or IFE. Densitometry of stained gels is subject to variability, depending on the affinity of the stain used to the specific protein examined and nonspecificity with overlapping protein fractions. Saturation of staining at areas of high focal protein concentrations is also a problem (16). For these reasons, densitometry is not used routinely in our laboratory; instead, the quantities of MCs are deduced from immunonephelometric concentrations and the appearance of background staining, and are verified by IFE with visual evaluation in cases of small MCs of uncertain class. The algorithms presented in this report are intended to raise a strong suspicion of an MC and to direct such samples to further investigation. They are not intended to be used as the sole evaluation of potential MCs.
In our experience, Bence Jones myelomas are often diagnosed late, occasionally after extensive evaluation for decreased renal function. In such cases with reduced glomerular filtration, discrete or multiple bands corresponding to circulating light chains may be seen on HRAGE. More often, the only evidence of this condition is hypogammaglobulinemia. It is therefore encouraging that an algorithm evaluating total-immunoglobulin concentration was successful in identifying 44% of cases of Bence Jones and that the presence of very small peaks attributable to circulating light chains could be detected by MC algorithms. The detection of this condition based solely on analysis of serum samples, however, remains highly unreliable and demonstrates the continuing need for clinical awareness and the performance of urine analysis for light immunoglobulin chains.
Identification of oligoclonal or multiple small monoclonal bands after
electrophoresis in agarose or cellulose acetate gels is highly
subjective. Although the presence of such bands may occur early
in an immunization process, to be followed by polyclonal
hypergammaglobulinemia, they may also indicate the presence of
lymphoproliferative disease or specific immunization against viral,
tumor, or autoantigens (17)(18). Their detection
is dependent on both the resolution of the electrophoretic system and
the interest and observation of the interpreter. Using a mathematical
algorithm to evaluate inflection points in the -region of the CZE
absorbance curve, we found good agreement with samples judged to have
significant oligoclonality by visual assessment. The numerical value of
such an algorithm may be increased or decreased to adjust sensitivity,
but at any given value it provides a reproducible objective measure not
otherwise possible in visual evaluations.
In conclusion, mathematical algorithms for the analysis of the absorbance curve produced by CE of serum samples could accurately identify 99% of samples with MCs as well as most sera from patients with Bence Jones myeloma, and they can mathematically define criteria for oligoclonality. The algorithms cannot replace immunofixation for the determination of immunoglobulin class or urine analysis for the identification and quantification of Bence Jones proteinuria. Algorithms function well as an initial screening of polyclonal immunoglobulin concentrations, but they cannot differentiate between IgG and IgM. They may, however, indicate suitable samples for specific quantification according to the clinical situation. The use of CE provides a modern, rapid, and cost-effective method of analyzing serum proteins. The additional option of computerized decision support, providing rapid and standardized interpretations, may increase the clinical availability and usefulness of protein analyses in the future. We anticipate further development of interpretive algorithms and validation of their use in clinical decision making.
|
Acknowledgments |
|---|
|
Footnotes |
|---|
|
References |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The following articles in journals at HighWire Press have cited this article:
|
|
 |
|
  P. Luraschi, I. Infusino, I. Zorzoli, G. Merlini, C. Fundaro and, and C. Franzini Heavy Chain Disease Can Be Detected by Capillary Zone Electrophoresis Clin. Chem., January 1, 2005; 51(1): 247 - 249. [Full Text] [PDF]
|
|
|
|
|
|
B. Lumbreras-Lacarra, J. M. Ramos-Rincon, and I. Hernandez-Aguado Methodology in Diagnostic Laboratory Test Research in Clinical Chemistry and Clinical Chemistry and Laboratory Medicine Clin. Chem., March 1, 2004; 50(3): 530 - 536. [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]
|
|
|
|
|
|
M. Jonsson and J. Carlson Computer-supported Interpretation of Protein Profiles after Capillary Electrophoresis Clin. Chem., July 1, 2002; 48(7): 1084 - 1093. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
