The multiplex bead array approach to identifying serum biomarkers associated with breast cancer

Abstract

Introduction: Breast cancer is the most common type of cancer seen in women in western countries. Thus, diagnostic modalities sensitive to early-stage breast cancer are needed. Antibody-based array platforms of a data-driven type, which are expected to facilitate more rapid and sensitive detection of novel biomarkers, have emerged as a direct, rapid means for profiling cancer-specific signatures using small samples. In line with this concept, our group constructed an antibody bead array panel for 35 analytes that were selected during the discovery step. This study was aimed at testing the performance of this 35-plex array panel in profiling signatures specific for primary non-metastatic breast cancer and validating its diagnostic utility in this independent population.

Methods: Thirty-five analytes were selected from more than 50 markers through screening steps using a serum bank consisting of 4,500 samples from various types of cancer. An antibody-bead array of 35 markers was constructed using the Luminex bead array platform. A study population consisting of 98 breast cancer patients and 96 normal subjects was analysed using this panel. Multivariate classification algorithms were used to find discriminating biomarkers and validated with another independent population of 90 breast cancer and 79 healthy controls.

Results: Serum concentrations of epidermal growth factor, soluble CD40-ligand and proapolipoprotein A1 were increased in breast cancer patients. High-molecular-weight-kininogen, apolipoprotein A1, soluble vascular cell adhesion molecule-1, plasminogen activator inhibitor-1, vitamin-D binding protein and vitronectin were decreased in the cancer group. Multivariate classification algorithms distinguished breast cancer patients from the normal population with high accuracy (91.8% with random forest, 91.5% with support vector machine, 87.6% with linear discriminant analysis). Combinatorial markers also detected breast cancer at an early stage with greater sensitivity.

Conclusions: The current study demonstrated the usefulness of the antibody-bead array approach in finding signatures specific for primary non-metastatic breast cancer and illustrated the potential for early, high sensitivity detection of breast cancer. Further validation is required before array-based technology is used routinely for early detection of breast cancer.

Figure 1

Procedure for constructing the 35-plex panel. Through using 4500 serum samples from cancer patients, five markers were discovered and identified through two-dimensional polyacrylamide gel electrophoresis (2D-PAGE), six markers through surface-enhanced laser desorption/ionisation time-of-flight (SELDI-TOF) and 24 markers through conventional sandwich ELISA method. After optimisation in capturing and detecting the antibody pair for a target analyte, the capturing antibody was conjugated with beads, and the simplex kit was validated for its dynamic range, recovery rate, parallelism with standard curve, interference, matrix effect, and lower and upper limits of detection. Twenty-three simplex kits could be grouped together according to dilution factor and absence of cross reactivity. The remaining 12 were left and used as a simplex kit. BC = breast cancer; SC = stomach cancer; LC = lung cancer; CC = colon cancer; HCC = hepatocellular carcinoma.

Figure 2

Classification performance of combinatorial markers identified through 35-plex panel assay. (a) Principal component analysis (PCA) with 35 markers showed clustering and separation of breast cancer patients (closed circle) and normal subjects (open circle) in the PCA chart using principal component (Comp) 1 and 2. NF = normal female; BC = breast cancer. (b) The area under the curve was calculated for combinatorial markers and a single marker, and compared using a receiver operating curve. CEA = carcinoembryonic antigen; EGF = epidermal growth factor; LDA = linear discriminant analysis; RF = random forests; SVM = support vector machine.