Circulating tumor cells (CTCs) have become increasingly acceptable as a prognostic marker in stratifying metastatic cancer patients for treatment and as a predictive marker in monitoring therapeutic response. CTC enumeration is an established prognostic marker (gold standard) in metastatic prostate, breast and colorectal cancer. However, due to the heterogeneity with respect to CTC phenotypic expression, epithelial-mesenchymal transition, and morphologic variability of different cancer cells, it is impossible to simply use the counts of well defined cells to characterize a wide spectrum of cancer status and progression. In addition, manual counting of CTCs also introduces operator bias on the size, shape and expression levels. We have developed an automated CTC characterization system that extracts enumeration, cell morphology and expression level of all intact, irregular and fragmented CTCs in an automatic fashion. A multiparameter classification model was then developed to characterize patient clinical outcome. Whole blood from advanced stage cancer patients and non-diseased controls was processed through anti-EpCAM antibody coated OnQChipsTM. Chips were imaged at low magnification on a fully calibrated rapid automated platform. Captured CTC candidate events were processed by an automated CTC detection algorithm using a set of spatial and spectral features to initially remove non-cellular events and then to indentify CTC subclasses. All CTC subclasses as well as artifact classes were manually labeled and verified at high magnification by trained imaging technologists. Manual labels were used to assess performance of the automated algorithms. A multivariate model based on CART (Classification and Regression Trees) was used for the classifier development. A total of 27 prostate cancer patients and 33 normal controls with 7.5mL blood samples per patient were used to develop and validate the initial techniques. The preliminary results show that the automated CTC event detection algorithm achieved a sensitivity of 96% and specificity of 89%. The CTC subclass classification algorithm achieved classification accuracy from 82% to 95% across all subclasses. The algorithms were N-fold cross-validated with 80/20 random sampling. The preliminary clinical model achieved sensitivity and specificity values of 90% and 82% respectively for patient vs. normal classification. A method for automated patient CTC classification and clinical model has been developed. The performance data from all classification algorithms is very encouraging. The multivariate patient model discriminates cancer patients from normal donor samples with high sensitivity and specificity. Future work includes incorporation of image-based features as well as clinical patient data into the model to improve sensitivity and specificity and address specific clinical needs.

Citation Format: Chunsheng Jiang, Oleg Gusyatin, David Tims, Aladin Milutinovic, Kam Sprott, Michael Stocum. Multiparameter CTC characterization using dual capture microfluidic chips. [abstract]. In: Proceedings of the 104th Annual Meeting of the American Association for Cancer Research; 2013 Apr 6-10; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2013;73(8 Suppl):Abstract nr 3507. doi:10.1158/1538-7445.AM2013-3507