A protein chip system for parallel analysis of multi-tumor markers and its application in cancer detection.

BACKGROUND Tumor markers are routinely measured in clinical oncology. However, their value in cancer detection has been controversial largely because no single tumor marker is sensitive and specific enough to meet strict diagnostic criteria. One strategy to overcome the shortcomings of single tumor markers is to measure a combination of tumor markers to increase sensitivity and look for distinct patterns to increase specificity. This study aimed to develop a system for parallel detection of tumor markers as a tool for tumor detection in both cancer patients and asymptomatic populations at high risk. MATERIALS AND METHODS A protein chip was fabricated with twelve monoclonal antibodies against the following tumor markers respectively: CA125, CA15-3, CA19-9, CA242, CEA, AFP, PSA, free-PSA, HGH, beta-HCG, NSE and ferritin. Tumor markers were captured after the protein chip was incubated with serum samples. A secondary antibody conjugated with HRP was used to detect the captured tumor markers using chemiluminescence technique. Quantification of the tumor markers was obtained after calibration with standard curves. RESULTS The chip system showed an overall sensitivity of 68.18% after testing 1147 cancer patients, with high sensitivities for liver, pancreas and ovarian tumors and low sensitivities for gastrointestinal tumors, and a specificity of 97.1% after testing 793 healthy individuals. Application of the chip system in physical checkups of 15,867 individuals resulted in 16 cases that were subsequently confirmed as having cancers. Analysis of the detection results with a Support Vector Machine algorithm considerably increased the specificity of the system as reflected in healthy individuals and hepatitis/cirrhosis patients, but only modestly decreased the sensitivity for cancer patients. CONCLUSION This protein chip system is a potential tool for assisting cancer diagnosis and for screening cancer in high-risk populations.