From tumor transcriptomes to underlying cell type proportions to better predict prognosis and response to treatments

Y. Blum,S. Job,A. Kamoun,N. Elarouci,M. Ayadi,F. Petitprez,R. Nicolle,E. Brambilla,A. de Reyniès

DOI: https://doi.org/10.1093/annonc/mdz269.048

IF: 51.769

2019-10-01

Annals of Oncology

Abstract:Background Tumor pathologists classify tumors according to cell-level and tissue-level criteria, using molecular markers in addition to morphological patterns, as reported in WHO tumor classifications. Their work describes to some extent both inter-tumor and intra-tumor heterogeneity, but it is a tedious task with potential reproducibility issues. The molecular subtyping of tumors represents a sometimes parallel and sometimes convergent effort to describe the heterogeneity of tumors. It is automated, which attenuates the limitations of pathological scoring mentioned above, but so far it is poorly adapted to describe intra-tumor heterogeneity. Methods We propose a novel method, WISP (Weighted In Silico Pathology), which finely measures intra-tumor heterogeneity by automatically estimating the proportions of cell types present in a bulk tumor sample, these cell types being predefined based on histological or high-throughput molecular criterions. Results We illustrate the relevance of our approach in several tumor types including lung cancer, glioblastoma and pancreatic cancer. We show that the cell types that describe tumors for a given cancer type can fairly well recapitulate existing molecular classifications and are very consistent with an independent pathological scoring. We show that our method offers a more standardized and finer-grained solution for describing tumor heterogeneity than either pathological scoring or molecular subtyping. More importantly, we show that the proportion of certain cell types is strongly associated to prognosis and drug response. Conclusions This study provides a framework for standardized molecular pathology and fully reshapes the way in which we think about pathology or molecular subtyping. We believe that our results are a proof of concept demonstrating the importance of considering cell types intra-tumor proportions for personalized clinical care. Legal entity responsible for the study The authors. Funding CIT Program - Ligue Contre le Cancer. Disclosure All authors have declared no conflicts of interest.

oncology