MUC1 Gene–Derived Glycoprotein Assays for Monitoring Breast Cancer (CA 15-3, CA 27.29, BR): Are They Measuring the Same Antigen?

Immunoassays often have notoriously large between-method differences in reported values. Some of the differences are related to standardization, while others are due to epitope binding differences of the antisera used. Circulating tumor markers often have problems posed by between-method differences in providing consistent results for patient management. Most of these analytes are measured by immunoassays using monoclonal or polyclonal antibodies. Since tumor markers are used to follow the course of a patient's disease, multiple measurements will typically be made over a period of months or years. Changes in their values are assumed to indicate disease progression or remission and may lead to therapeutic interventions. Patients may have their tests performed in several different laboratories, so different assays may be used to measure the same tumor marker. If those assays are not equivalent, erroneous information based on assay methodology, rather than the patient's clinical status, may be generated.

Serum assays that measure circulating levels of MUC1 gene–derived glycoprotein in patients with breast cancer illustrate how between-method differences may arise. The human MUC1 gene codes for a large (300–400 kd) mucin glycoprotein that is expressed in most glandular epithelia.1,2 These mucins also are called polymorphic epithelial mucins (PEMs). In malignant cells, PEM is overexpressed on the entire cell surface, and increasing amounts are shed into the circulation. Tumors involving glandular organs, such as the breast, can release high concentrations of PEM into serum, making serum assays potentially useful as tumor markers.3–6 

The polypeptide core of PEM contains a 69-amino acid cytoplasmic domain and a much larger extracellular domain, consisting of 20 amino acid tandem repeats.7 The number of tandem repeats varies between individuals, and different alleles of the human MUC1 gene may code for 25 to 100 or more tandem repeats. More than half of the molecule is carbohydrate, and the degree of glycosylation is variable as well. Therefore, PEM is very heterogeneous in structure.

Multiple commercial assays are available for measuring MUC1 gene–derived glycoprotein in serum. However, only 2 basic assay formats are used: immunometric (sandwich) and competitive formats.8–11 Both of these assay formats are defined in terms of the antigens to which their monoclonal antibodies bind. The quantity called cancer antigen (CA) 15-3 is defined as the glycoprotein that binds with 2 monoclonal antibodies, namely, DF3 and 115D8.9,12 The DF3 antibody, which serves as the detection signal in the sandwich assays, recognizes an 8-amino acid sequence in the 20-amino acid tandem-repeat segment of PEM. Figure 1 depicts DF3 as the signal antibody binding to the sequence DTRPAPGS, which corresponds to the amino acids Asp-Thr-Arg-Pro-Ala-Pro-Gly-Ser.1 The 115D8 monoclonal antibody is the solid-phase capture antibody, which binds to a peptide-carbohydrate epitope on the same repeat. The 2 antibodies together make a sandwich immunoassay with a positively sloped dose-response curve.

The second assay format is a competitive immunoassay based on inhibition of the binding of a specific form of PEM to a labeled antibody. This specific form of PEM is called CA 27.29, or breast antigen (BR). This antigen is empirically defined as the glycoprotein that displaces labeled monoclonal B27.29 antibody from an immobilized antigen purified from ZR-75-1 breast cancer cells.12,13 The labeled antibody binds to an 8-amino acid sequence of PEM that partially overlaps the antigenic binding site of DF3, but is shifted 3 amino acids closer to the N-terminal end of the 20-amino acid tandem repeat. This sequence is depicted in Figure 1 as SAPDTRPA, which corresponds to amino acids Ser-Ala-Pro-Asp-Thr-Arg-Pro-Ala. The combination of the labeled B27.29 antibody and the solid-phase antigen purified from breast cancer cells forms a competitive assay with a decreasing exponential dose-response curve.

These 2 assay formats (sandwich and competitive) theoretically should produce the same results when measuring purified glycosylated tandem repeats of PEM, assuming that the circulating forms of PEM contain equal numbers of immunoreactive sites for both binding sequences. To test this assumption, we evaluated the performance of assays for CA 15-3 and CA 27.29 by reviewing proficiency survey summary reports from the College of American Pathologists (CAP). To further understand the causes of assay differences, we compared the performance of 3 commercial assays using actual patient specimens.

Participant summary data were obtained from 22 specimens distributed from 1999 through 2002 by the CAP in their Tumor Marker Survey. These proficiency testing specimens were distributed as 2 vials per mailing, with 3 mailings per year. The MUC1 gene–derived glycoprotein assays for breast cancer reported in these surveys are tabulated in 2 categories, labeled as CA 27.29 and CA 15-3. Each category is subgrouped according to the vendor supplying the reagents. The number of participating laboratories, median value, low value, and high value were obtained from the CAP summary reports for each method. The mean value, standard deviation, and coefficient of variation were tabulated for all methods with at least 10 participants. The capture and signal attributes for these assays were obtained from the package inserts provided by the companies (Table 1).

The source of the antigen was the same for all 22 CAP survey specimens. The antigen for spiking the specimens designated as CA 15-3 and CA 27.29 was derived from a cell line grown by aseptic tissue culture. Since these proficiency samples also are spiked with CA 19-9, CA 125, and CA 72-4, some of the CA 15-3 or CA 27.29 immunoreactivity may have come from cross-reactivity with the spiking material used for these other analytes. Some of the activity may be endogenous to the base pools, which are made from human plasma.

The performance of 3 of these assays also was compared using clinical specimens. Residual serum samples and test results were retrieved from 58 patients who had the CA 27.29 breast cancer antigen test (Centaur BR assay, Bayer Diagnostics, Tarrytown, NY) requested in the Mayo Clinic clinical laboratory. The specimens were held frozen at −20°C until analysis. All clinical identifiers were removed from these specimens because of institutional review board and Health Insurance Portability and Accountability Act of 1996 (HIPAA) regulations. These specimens were thawed and measured by the AxSym CA 15-3 (Abbott Laboratories, Abbott Park, Ill) and the Vitros Eci CA 15-3 (Ortho-Clinical Diagnostics, Rochester, NY) assays.

Linear correlation analysis and cross-plots with regression statistics were used to compare the methods. The correlation analyses were performed using both zero intercepts and nonzero intercepts. The correlation and regression statistics with the zero intercept help to evaluate the agreement of the assays, while the unconstrained correlation and regression statistics help to evaluate the agreement in measurement specificity for assays that may have calibration offsets and differences in reference materials. For both sets of analyses (survey material and patient specimens), the Abbott AxSym assay was plotted on the horizontal (x) axis for the comparisons. The Abbott AxSym method was arbitrarily selected for the comparison plots because this method had the largest number of users at the time of the proficiency surveys. The cross-plots and regression statistics were performed with Microsoft Excel software.

Table 1 lists the assays reported in the CAP surveys, separated according to sandwich and competitive assay formats. The capture antibody and signal antibody for the sandwich assays are listed in the table. For the competitive assays, the separation system and signal system are listed. Some of the companies changed names between 1999 and 2002. When the equipment and reagent distributions were not the same over time, the 2002 name of the company was used. Some of the methods were listed as both CA 15-3 and CA 27.29 on the CAP summary reports. This article summarizes only the data listed as CA 15-3 for the sandwich assays and only the data listed as CA 27.29 for the competitive assays.

Table 2 shows the regression statistics comparing the medians of the 22 proficiency test samples for 9 assay methods compared with the median of Abbott AxSym CA 15-3 proficiency test results. The correlation plots for 3 of the assays are shown in Figure 2, A and B. All of the assays correlated very well, with most of the correlation coefficients exceeding 0.99. Some of the cross-comparisons of CA 15-3 with CA 27.29 had regression slopes close to 1.0, whereas some of the assays with the same test name (either CA 15-3 or CA 27.29) had more discordant regression slopes. For example, the Bayer ACS:180 CA 27.29 had regression slopes of 1.015 to 1.021 against the Abbott AxSym CA 15-3, while the Roche Elecsys CA 15-3 had regression slopes of 0.836 to 0.886, and the Bayer ADVIA Centaur CA 27.29 assay had regression slopes of 1.095 to 1.099 against the Abbott AxSym CA 15-3.

Figure 2 contrasts the regression plots of the medians of the CAP proficiency testing samples to regression plots for the measurement of the patient samples for the 3 assays evaluated. In each of the sets of data, the linear correlation coefficients (Rho) were high. In comparing the Bayer ADVIA Centaur assay with the Abbott AxSym assay, both the proficiency test specimens and the patient samples gave higher results on the Bayer system (slope = 1.095 and r = 0.997 on proficiency tests, slope = 1.091 and r = 0.939 on full set of patients, and slope = 1.242 and r = 0.870 on subset of 53 patients with Abbott values <70 U/mL). However, for the Ortho-Clinical Vitros Eci, proficiency tests have lower results (slope = 0.914 and r = 0.998), similar to the full set of patient samples (slope = 0.961 and r = 0.986), whereas the subset of 53 patients with Abbott values less than 70 U/mL, gave higher values (slope = 1.222 and r = 0.972).

The patient values have much greater scatter than the CAP proficiency samples, partly because they compare individual measurements rather than the median of groups and partly because they represent more heterogeneous mixtures of the PEM glycoprotein. The differences in the regression slopes presumably are due to the differences in the glycoproteins found in patients compared to the purified cell-line protein. The full set of patient correlations, which includes values greater than 70 U/mL, gives correlation coefficients and slopes that are closer to the values seen with the CAP survey samples. Since these regression slopes are driven mainly by the few samples with high values, these high concentrations of MUC1 gene–derived glycoprotein found in these samples may be more similar to that from the purified cell line.

Between-method differences in immunoassays affect health care systems in multiple ways. Within a laboratory, changes in assay methods often are not well received by clinicians, because these method changes generally shift the patient test values. Differences between laboratories within an integrated health care network may cause confusion and potential treatment errors because both patients and physicians may cross between health care sites. The sharing of electronic medical records across medical facilities with different laboratories makes it difficult to merge laboratory test values for the same analyte measured by different methods. Clinicians generally try to compensate for these differences by mentally adjusting their decision limits, but this requires special diligence and may lead to errors in diagnosis and treatment.

The European Union is attempting to minimize the between-method differences for assays used in every European Union country by requiring these assays to be traceable to a higher metrological reference standard.14 This European Union In Vitro Diagnostic Directive required diagnostic reagent manufacturers to begin providing traceability documentation after December 2003. This initiative is stimulating the development of reference methods and reference standards for many analytes. For tumor markers and hormones, this standardization also requires further definition of which biochemical forms of the analytes should be measured to provide the best support for clinical decisions.

Our review of method performance on the CAP Tumor Marker Survey for the years 1999–2002 allows us to draw several conclusions about assays for CA 15-3 and CA 27.29. First, the median values reported by laboratories that offer these assays are highly correlated, so it appears that each assay listed in Table 2 is measuring the same antigen. Second, the agreement of individual values, measured in U/ mL, is generally good. However, the slopes of the regression lines vary from a low of 0.836 to a high of 1.099, which indicates that the actual value obtained is method dependent.

Comparisons based on CAP proficiency testing material may not represent the relationship of these methods for real patient samples. The graphs in Figure 2 demonstrate that patient specimens show more scatter about the line than do the medians of the survey specimens. Moreover, the slopes of the lines for patient specimen data are farther from unity than the slopes of the lines for the survey specimens. These differences probably are caused by differences between the purified antigen used in the proficiency test material and the heterologous antigens produced by patients. This is particularly true for the MUC1 gene–derived glycoproteins that are measured in the CA 15-3 and CA 27.29 assays.

This study illustrates that standardization of assays to purified reference materials may not lead to harmonization of patient test results. This could have implications for the traceability requirements for the European Union In Vitro Diagnostic Directive directive. However, it is encouraging that the general relationships found with the CAP proficiency specimens were similar to the relationships found with patient specimens. It is interesting to note that some assays with different names (CA 15-3 vs CA 27.29) agree better than some assays with the same name (CA 15-3). The comparability of assays for PEM, regardless of the assay format, also is affected by the reference material used for calibration.1 Matrix differences between the material used for calibration and patient specimens also can cause discordance among assays.