Project Summary
Tumor-stroma/immune cell signaling communications within the tumor microenvironment (TME) play important
roles in tumor development and responses to targeted and immunotherapies. However, our knowledge of
complex signaling communications within TME, and their roles in tumor development, drug and
immunotherapy response is limited. Effective molecular targets are still missing that can inhibit the tumor-
stroma signaling communications. Single cell RNA sequencing (scRNA-seq) has been being a powerful
technology to capture transcriptional changes in individual tumor, stroma, immune cells within TME. While
scRNA-seq datasets of human cancer are rapidly growing in number, which is leading to many basic and
translational discoveries, the study of dynamic tumor-stroma signaling communications is limited. Limiting
factors include: 1) static and single time-point snapshots of the complex interactions within the TME, and 2)
difficulty in perturbing a large number of related signaling targets; and measuring corresponding functional
effects to these perturbations in mouse or tumor tissues (to identify novel therapeutic targets and treatments).
To resolve these challenges, in this study, we propose to combine the cutting-edge technologies, including
novel artificial intelligence (AI) models, scRNA-seq, crispr-based single or double knockouts (CDKOs), 3D
tumor-CAF-TAM co-culture assays, and genetic mouse models, in a systems biology manner. Specifically, (in
Aim 1), we will develop novel network AI models using valuable large sets of scRNA-seq data of PDAC human
tumors at WashU to identify static core tumor-CAF-TAM interaction (TCTi) signaling networks (multi-cell intra-
/inter-cellular signaling networks of TCTi); and an initial set of anti-TCTi targets. In Aim 2, we will further
develop another network AI model (M-Step) to infer the better anti-TCTi targets using the functional validation
feedbacks in Aim 3; and predict synergistic drug combinations (inhibiting multiple key anti-TCTi targets). In Aim
3, the predicted targets and drugs will be efficiently evaluated using scalable 3D Tumor-CAF-TAM co-culture
assays and crispr-based knockouts (E-step) with 3 measurable metrics, i.e., tumor proliferation, migration,
angiogenesis. The M-step (modeling) and E-step (validation) forms an E-M process to identify key anti-
TCTi targets and drugs iteratively. We will apply these AI models in Pancreatic ductal adenocarcinoma
(PDAC) because 1) there have been very limited responses to immunotherapy; 2) no effective treatment; 3)
nearly all patients will develop chemo-resistant and metastatic tumors within 2 years of diagnosis. Also
(feasibility), 4) we have a strong cross-disciplinary team studying the PDAC TME (supported by NCI SPORE
and human tumor atlas network (HTAN)), with the valuable state-of-the-art resources. The success of this
study will identify novel anti-tumor-TAM-CAF targets and drug cocktails for PDAC treatment. The AI models,
supporting the novel E-M systems biology, can be applied to other cancers and diseases.