“Computational Histopatholomics for Deep Interrogation of Hyperplexed Heterogenous Tumor Data”
Comprehensive genetic profiling of tumors has revealed intrinsic molecular variability, or intratumor heterogeneity (ITH), in multiple cancers, including breast, prostate, glioblastoma, colorectal, and ovarian. Heterogeneity is rooted in both genetic and nongenetic factors and evolves through a supportive tumor microenvironment (TME). Not surprisingly, genetic, phenotypic, and TME heterogeneity present major obstacles to cancer treatment.
While genetic testing flags the presence and type of ITH, the spatial distribution of the heterogeneity can be appreciated only with in situ imaging of tissue sections or tumor microarrays using methods such as fluorescence-based immunohistochemistry. In situtechniques probe the tumor and surrounding tissue for the expression of proteins, DNA, and RNA in the context of individual cells and tissue slices. Although such imaging has typically been restricted to no more than four to seven proteins labeled per slide, new technological advances now allow up to 60 proteins and a few RNA or DNA probes—with a theoretically unlimited number possible—to be labeled on the same multicellular tissue slice of up to 10 mm, thus yielding large-scale, whole-slide hyperplexed fluorescence images. However, imaging at this scale raises several new big-data challenges and opportunities for automated image analysis, including how to quantitate and characterize spatial ITH, how to harmonize computational methodologies across diagnostic labs and clinical trials, and how to translate ITH analysis into research insights and clinical use.
Chennubhotla and colleagues’ computational histopatholomic approach tackles these challenges with algorithms designed for deep investigation of hyperplexed ITH data. The tools they are developing allow end users to characterize disease subtypes, to study phenotypic heterogeneity effects of genomic alterations, and identify potential associations between heterogeneity and clinical features of interest (e.g., surveillance of progression, risk of invasion, risk of metastasis, drug response, etc.). Their image-based prognostic algorithms help reduce intra- and interobserver variability in quantifying heterogeneity at diagnostic laboratories and support the development of an ITH index that, combined with genetic rating scales (e.g., Oncotype Dx), will better predict tumor progression and patient outcome.