Tumor microenvironment gene expression dynamics: Keystone cell types and non-cancer cell driver genes

Cancer is not a disease, rather, of multi-scale heterogeneity, i.e., each victim of cancer is of unique evolutionary dynamics at distinct levels. Indeed, cancer therapeutics has been evolving from non-specific chemo and radiation therapy, mutation-specific targeted therapy, immunotherapy, as well as combination therapies, to ongoing tumor microenvironment (TME) specific interventions, in a broader sense, targeting tumor ecosystems. However, the critical schemata of TME with cell-type specific gene expression dynamics underpinning mechanistic druggable key driver genes remains largely obscure. Here, we present an integrated approach of AI-powered modeling TME gene expression dynamics via borrowing ecology models, namely, treating cell types as the counterpart of species in ecological settings. We represent each TME as a tensor with public non-small cell lung cancer (NSCLC) data settings, mainly focusing on lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUSC) with scRNA-seq, snRNA-seq, and spatial transcriptomics data. Given that up to 72 distinct cell types/states were found in LUSC and 57 in LUAD, we validated these distinct cell types/states with the uniLung database via wavelet-transformed cell cluster analysis. Furthermore, we treated these cell types as species via incorporating a network ecology model (40 spatial transcriptomics data for NSCLC study, E-MTAB-13530, among other data settings). We found distinct keystone cell types in driving cancer progression in addition to key driver gene sets. These patient specific keystone cell types are not limited to cancer cells, rather many immune and stromal cell types are indeed ecology drivers, such as certain subtypes of cancer associated fibroblast (CAF). Distinct to LUAD, LUSC is of more loss of function of tumor suppressors, such as chromatin remodeling genes SMARCA4 and PBRM1 . Of note, DNMT3A, TET2 and ASXL1 are of immune cell driver genes, much individualized. Thus, tumor driver genes are not limited to cancer cells, yet to discover more non-cancer cell tumor driver gene sets. We propose validation and qualification strategies with our available bioassay platforms, spanning from genomics, cell-based assays and biomarkers. We continue to cross-validate this ecology network model with expanded distinct data setting. Of particular interest, we have been examining these complex interactions via cancer organoid models. Taken together, targeting tumor ecosystems with individualized biologics alongside geography ecology modeling of TME, we advocate integrated AI-powered single-cell level temporospatial analytics infrastructure as we have been building central laboratory logistics for fresh and frozen cancer samples, clinical pathology, automation, high throughput biomarker assays, scRNA-seq and snRNA-seq, spatial transcriptomics, flow cytometry, Elispot assay, cell potency, as well as organoid platforms.