Abstract Background – Glioblastoma Multiforme (GBM) is a highly aggressive and heterogeneous disease in which survival of patients is measured only in months. Pathologic features associated with patient outcome are still incompletely understood. Until recently, phenotypic features of single cells have not been investigated due to technological limitations. However, with the advent of high-content slide scanning coupled with cognitive and scriptable algorithms, researchers are now positioned to identify and quantify single cell features that may provide unique insight on tumor behavior. In this project we sought to identify variations in cellular phenotypes which correlate with time of post diagnostic survival in patients with GBM. Methods – A total of 157 GBM patients were selected for study on the basis of short (SS) (median K-M survival = 6 months; N=81) or long (LS) (median K-M survival = 33 months, N=76) survival times. Median age was 56 years and all patients underwent current standard of care therapy for GBM (surgery, radiation and temozolomide). Among the 157 cases, 11 were excluded due to poor sample quality (final n=146). For each case, the diagnostic H&E slide was digitally scanned using the ScanScope XT (Aperio, Vista, CA, USA) with a 200x/0.8NA objective lens. Definiens TissueStudio v3.0 (Munich, Germany) was used to identify viable tumor regions. Individual cells were segmented in areas of viable tumor and twenty separate features extracted from each tumor cell (subcellular compartmentalization, nucleus to cytoplasm area ratio, nuclear size and shape, etc). In total, the 20 features were extracted in thousands of single cells for each GBM case. Output was evaluated by Matlab (The MathWorks, Inc,, Natick, MA) using a heatmap approach by first normalizing the scales of each feature to a range of 0-1, and assigning a color from green (0) to red (1) (5 classes) for each cell or compartment (x) and each feature (y). Results – This study was completed in a series of three stages including training followed by two replication sets. In the training set (N=50), four of the evaluated features associated with the size and shape of the cancer cell nuclei (i.e. width [μm], circularity, ellipticity and hematoxylin intensity), were found to distinguish the SS group (15/25) from the LS group (6/25) based on supervised classification. A similar pattern was observed in replication set 1 (15/24 and 8/28, respectively) and replication set 2 (15/23 and 8/21, respectively). Overall, 66% of cases were correctly classified with respect to survival time on the basis of these cellular features (p=0.0001). Conclusions – Quantitative image analysis may be useful in the identification of novel prognostic features in GBM with potential for gaining new biological insights on the behavior of these tumors. Citation Format: Mark C. Lloyd, Melissa H. Madden, L. Burton Nabors, Reid C. Thompson, Jeffrey J. Olson, Steven L. Carroll, James Browning, Tamir Epstein, Robert A. Gatenby, Kathleen M. Egan. Single cell phenotypic heterogeneity as a prognostic factor in glioblastoma. [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 714. doi:10.1158/1538-7445.AM2013-714
Read full abstract