Autoantibodies as potential biomarkers for breast cancer

Abstract

Introduction: Only a limited number of tumor markers for breast cancer are currently available. Antibodies to tumor-associated proteins may expand the number of available tumor markers for breast cancer and may be used together in a serum profile to enhance sensitivity and specificity.

Methods: In the present study, we interrogated a breast cancer cDNA T7 phage library for tumor-associated proteins using biopan enrichment techniques with sera from normal individuals and from breast cancer patients. The enrichment of tumor-associated proteins after biopanning was tested using a plaque-lift assay and immunochemical detection. The putative tumor-associated phage clones were collected for PCR and sequencing analysis. Unique and open reading frame phage-expressed proteins were then used to develop phage protein ELISAs to measure corresponding autoantibodies using 87 breast cancer patients and 87 normal serum samples. A logistic regression model and leave-one-out validation were used to evaluate predictive accuracies with a single marker as well as with combined markers. Identities of those selected proteins were revealed through the sequence BLAST program.

Results: We harvested 100 putative tumor-associated phage clones after biopan enrichment. Sequencing analysis revealed that six phage proteins were inframe and unique. Antibodies to these six phage-expressed proteins were measured by ELISAs, and the results showed that three of the phage clones had statistical significance in discriminating patients from normal individuals. BLAST results of the three proteins showed great matches to ASB-9, SERAC1, and RELT. Measurements of the three predictive phage proteins were combined in a logistic regression model that achieved 80% sensitivity and 100% specificity in prediction of sample status, whereas leave-one-out validation achieved 77.0% sensitivity and 82.8% specificity among 87 patient samples and 87 control samples. Receiver operating characteristic curve analysis and the leave-one-out method both showed that combined measurements of the three antibodies were more predictive of disease than any of the single antibodies studied, underscoring the importance of identifying multiple potential markers.

Conclusion: Serum autoantibody profiling is a promising approach for early detection and diagnosis of breast cancer. Rather than one autoantibody, a panel of autoantibodies appears preferable to achieve superior accuracy. Further refinements will need to be made to further improve the accuracy. Once refined, the assay must be applied to a prospective patient population to demonstrate applicability.

Figure 1

Identification of disease-specific phage clones after the biopanning process. Two nitrocellulose membrane disks were placed on and then lifted from the same phage grown plate of biopan 4. (a) One membrane was probed with pooled normal sera and (b) the other was probed with pooled patient sera. After electrogenerated chemiluminescence detection, numerous immunoreactive clones showed more intensified spots on the membrane incubated with patient sera than on the membrane incubated with normal sera. The circle and square indicate the same area on the two membranes.

Figure 2

ELISA of phage-expressed proteins with individual serum samples. Antigen ELISAs were developed with ASB-9-expressing, SERAC1-expressing and RELT-expressing phages. The assays were performed with serially diluted (1:20 to 1:10,240) individual serum samples that were not used in the biopan, to confirm measurements were representative of an antigen-antibody affinity reaction. Representative curves from three patients are shown for each protein. Empty (no inserts) T7 phages were used to show the nonspecific reaction backgrounds.

Figure 3

Comparisons of the specificity and sensitivity of logistic regression models. Data from quantitative ELISAs for three antibodies were evaluated for ability to predict disease. Lower curve: predictive accuracy using the logistic regression model with RELT data alone from 87 patients and from 87 normal persons as the explanatory variable. The area under the curve is 0.727 and the model is significant (P = 0.0001). Upper curve: predictive accuracy with the combination of ASB-9, SERAC1, and RELT as explanatory variables, where P = 0.0001 and the area under the curve is 0.861.