Abstract 2182: Reconstructing evolutionary models of tumor progression from single-cell heterogeneity data

Studies of genetic variation in solid tumors have revealed massive cell-to-cell heterogeneity within single tumors, leading to a surge of interest in reconstructing processes of tumor progression at the cellular level. These efforts are hindered by a lack of accurate quantitative models of tumor evolution processes and computer algorithms to analyze tumor heterogeneity data with respect to these models. We have previously developed algorithms to infer likely tumor phylogenetic trees from single-cell fluorescence in situ hybridization (FISH) gene copy number data, which has the advantage of making it practical to profile thousands of single cells and thus to survey hundreds of cells per tumor across tens of tumors. We have further shown that these phylogenetic reconstructions of a tumor9s past evolution have significant predictive power for future progression of those tumors in a variety of cancer types. In the present work, we address the need for accurate models of tumor evolution through new methods to infer FISH-based tumor phylogenies while simultaneously learning tumor-specific models of evolution by copy number variation at the single gene, single chromosome and whole genome scales. We have applied these algorithms to four FISH data sets: i) cervical cancers collected from paired primary and metastasis from 16 patients and probed at for four genes (LAMP3, PROX1, PRKAA1 and CCND1) measured for up to 250 cells per tumor, ii) tongue cancers collected from 65 patients at four tumor stages and probed at for four genes (TERC, CCND1, EGFR and TP53) measured for up to 250 cells per patient, iii) prostate cancers collected from 6 non-progressive and 7 progressive carcinomas and probed at six genes (TBL1XR1, CTTNBP2, MYC, PTEN, MEN1 and PDGFB) measured in up to 407 cells per patient, and iv) ductal carcinoma in situ and invasive ductal carcinoma of the breast collected as paired samples from 13 patients and probed at eight genes (COX2, MYC, CCND1, HER2, ZNF217, DBC2, CDH1 and TP53) and measured on up to 220 cells per tumor. Statistical analysis of inferred cervical cancer trees allowed us to distinguish progressing from non-progressing primary tumors with 86% accuracy, while similar statistics derived from tongue cancers allowed us to significantly classify tumors with good versus bad prognosis with respect to overall (P-value = 0.0000394) and disease-free (P-value = 0.000117) survival, independent of tumor stage and smoking behavior. These results improve on our best prior accuracies for each of these tasks, providing validation for the model inference approach while also providing new methods relevant to the problem of predicting future progression of tumors. Citation Format: Salim A. Chowdhury, E. Michael Gertz, Darawalee Wangsa, Kerstin Heselmeyer-Haddad, Thomas Ried, Alejandro Schaeffer, Russell Schwartz. Reconstructing evolutionary models of tumor progression from single-cell heterogeneity data. [abstract]. In: Proceedings of the 106th Annual Meeting of the American Association for Cancer Research; 2015 Apr 18-22; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2015;75(15 Suppl):Abstract nr 2182. doi:10.1158/1538-7445.AM2015-2182