Data-driven learning how oncogenic gene expression locally alters heterocellular networks

Abstract

1 ingly relies on mechanistic mathematical modeling and simu-2 lation. In immuno-oncology, models that capture causal rela-3 tions among genetic drivers of oncogenesis, functional plastic-4 ity, and host immunity provide an important complement to wet 5 experiments, given the cellular complexity and dynamics within 6 the tumor microenvironment. Unfortunately, formulating such 7 mechanistic cell-level models currently relies on hand curation 8 by experts, which can bias how data is interpreted or the pri-9 ority of drug targets. In modeling molecular-level networks, 10 rules and algorithms have been developed to limit a priori bi-11 ases in formulating mechanistic models. To realize an equivalent 12 approach for cell-level networks, we combined digital cytome-13 try with Bayesian network inference to generate causal models 14 that link an increase in gene expression associated with onco-15 genesis with alterations in stromal and immune cell subsets di-16 rectly from bulk transcriptomic datasets. To illustrate, we pre-17 dicted how an increase in expression of Cell Communication 18 Network factor 4 (CCN4/WISP1) altered the tumor microenvi-19 ronment using data from patients diagnosed with breast cancer 20 and melanoma. Network predictions were then tested using two 21 immunocompetent mouse models for melanoma. In contrast to 22 hand-curated approaches, we posit that combining digital cy-23 tometry with Bayesian network inference provides a less biased 24 approach for elaborating mechanistic cell-level models directly 25 from data. 26 Heterocellular networks | digital cytometry | deconvolution | anti-tumor 27 immunity | Bayesian network inference | functional plasticity Introduction 30 Tissues are dynamic structures where different cell types or-31 ganize to maintain function in a changing environment. For 32 instance, the mammary epithelium reorganizes during dis-33 tinct stages of the ovarian cycle in preparation for lactation 34 (Klinke, 2016). At the same time, immune cells clear dead 35 cells and defend against pathogens present in the tissue mi-36 croenvironment. Ultimately, the number and functional ori-37 entation of different cell types within a tissue interact to cre-38 ate a network, that is a heterocellular network. This hetero-39 cellular network is essential for creating and maintaining tis-40 sue homeostasis. While we know that tissue homeostasis is 41 disrupted during oncogenesis, our understanding of how ge-42 netic alterations quantitatively and dynamically influence the 43 heterocellular network within malignant tissues in humans 44 is not well developed despite large efforts, like The Cancer 45 Genome Atlas (TCGA), to characterize the genomic and tran-46 scriptomic landscape in human malignancy (Hoadley et al., 47 2018; Wells and Wiley, 2018). In parallel with these large 48 scale data gathering efforts, two informatic developments, 49 namely digital cytometry and Bayesian network inference, 50 may be helpful in interrogating these datasets. 51 87 Increases in size and information content of t...

Head-to-head comparison of CCN4, DNMT3A, PTPN11, and SPARC as suppressors of anti-tumor immunity

“…Recent work to identify possible mediators of treatment resistance and immunosuppression that could be utilized as therapeutic targets has begun to rely on CRISPR/Cas9 screenings as their primary method (24, 25). However, while these screenings are high throughputin identifying potential therapeutic targets, they also identify intercellular proteins that may be difficult to target in therapy due to lack of access or that may carry increased risk for toxicity or adverse reactions based on their function in normal cells.…”