Genome sequencing studies of normal tissues have revealed the accumulation of oncogenic mutations with age without evident malignancy. These findings raise questions on when and why tumor development sets in, how it proceeds, and whether there are detectable early warning signs. To address these problems, we are studying the proliferation and differentiation dynamics of normal and transformed stem cells in vivo by combining non-invasive cell lineage tracing with mathematical modeling of clonal dynamics and statistical inference. In this talk, I will discuss recent results for hematological malignancies and solid tumors.

During development, a multi-potent stem cell gives rise to the three major functional cell lineages in the pancreas, the exocrine acinar cells, the ductal cells and the neuroendocrine islet cells. Mouse ablation studies have demonstrated the capacity of each cell lineage to repopulate the other cell type. In addition, lineage tracing experiments have demonstrated that pancreatic ductal adenocarcinoma (PDAC), the most lethal and common form of pancreas cancer, arises most commonly from acinar cells that have undergone a metaplastic transdifferentiation to the ductal cell lineage. We have further demonstrated a second transdifferentiation event where ductal adenocarcinoma cells can undergo neuroendocrine differentiation. PDAC with higher levels of neuroendocrine differentiation is associated with early recurrence and decreased overall patient survival; and standard-of-care chemotherapeutics induce PDAC neuroendocrine differentiation. We are studying ductal – neuroendocrine transdifferentiation in PDAC at multiple scales from the nanoscale, where we are studying spatial gene regulation, to the macroscale, where we are modeling tumor – microenvironmental crosstalk using 3-dimensional bioprinting. At the nanoscale, we find that the c-MYC transcription factor supports ductal – neuroendocrine transdifferentiation and that MYC activity is controlled by dynamic spatial regulation where MYC’s subnuclear localization to the inner basket of the nuclear pore complex controls genes that are poised to respond to extracellular signals. At the macroscale, we are investigating the influence of stromal cell types on PDAC cell phenotypes and drug responsiveness using a 3D bioprinter that incorporates multiple cell types into a scaffold free tissue with defined architecture. Together, this research is shedding light on mechanisms that induce lineage plasticity in response to environmental signals and drug stress. We anticipate our research will inform new therapeutic approaches that can be efficacious in complex tissues that better represent PDAC tumors, with the hope of impacting clinical care of patients suffering from the devastation of pancreas cancer.  

The ongoing explosion of genomic data and analysis tools has led to an urgent need for an environment that allows researchers to encapsulate the entirety of a research narrative – text, images, multimedia, executable code – within a single document. GenePattern Notebook is an environment that integrates the popular Jupyter Notebook system with the GenePattern platform for integrative genomics, allowing researchers to create, collaborate on, and disseminate inception-to-publication research narratives including sophisticated machine learning methods, without the need to write code. The online GenePattern Notebook workspace provides built-in access to high-performance compute resources and cloud-based data storage to support compute-intensive analyses.

Building mechanistic systems biology models requires a laborious process of manual information gathering and model implementation. This creates a gap between the scope of typical models, and the scale of high-throughput data they are meant to help interpret. To address this problem, we developed INDRA (the Integrated Network and Dynamical Reasoning Assembler), a system for automatically assembling models of molecular mechanisms directly from English language, including from the scientific literature. INDRA interfaces with natural language processing systems to extract descriptions of mechanisms from text (e.g., "GRB2 binds EGFR that is phosphorylated on tyrosine"), and uses knowledge assembly algorithms to fix certain errors, resolve redundant knowledge, infer missing information, and assess belief. INDRA then uses this knowledge base to produce models in different formalisms, including rule-based dynamical models, and causal networks. We present applications of INDRA including (i) a model automatically built from an English language description of mechanisms which can reproduce resistance to Vemurafenib in BRAF-V600E melanoma cells (ii) a model of growth-factor signaling built by machine-reading ~95,000 relevant publications, which can provide mechanistic explanations to drug perturbations in a melanoma cell line (iii) a human-machine dialogue system in which a user can gather information and build a mechanistic model to explain an experimental observation by talking with a computer partner.

From my perspective, there are 5 major questions in the clinical management of cancer: (1) How can we stratify the risk of neoplasms? (2) How can we prevent cancer? (3) How can we detect cancer early enough to safely remove or treat it? (4) How can we cure cancers? and (5) How can we prevent therapeutic resistance from killing the patient? Because cancer biology is actually a sub-discipline of evolutionary biology, there are evolutionary approaches to each of these questions. I will present recent or ongoing research into evolutionary approaches for each of these questions, including work that spans scales from cancer genomics, and somatic cell level evolution, to comparative oncology studies across hundreds of species.

Microenvironmental conditions contribute to the pathogenesis of cancer and include altered cellular composition, extracellular matrix (ECM) deposition, and mechanical cues. However, our understanding of the specific mechanisms by which these microenvironmental perturbations impact the development, progression, and therapy response of cancer is relatively limited. More intricate models are needed to better understand the complex biochemical and biophysical interactions that drive tumor initiation, growth, metastasis, metabolic adaptation, and immune evasion. The fields of biomaterials and tissue engineering provide increasingly sophisticated tools and strategies to recapitulate and monitor relevant properties of tumor-microenvironment interactions. These approaches not only bear tremendous potential to advance our current understanding of cancer, but are also increasingly explored for more clinically relevant drug testing. Indeed, combining patient-specific cells with engineered culture systems promises to enhance the predictive power of precision medicine pipelines. This talk will highlight specific examples of how the microenvironment regulates the highly dynamic nature of cancer and will outline opportunities and challenges of the field of tumor engineering.

A vital challenge that the vast majority of biological research must address is how to translate observations from one physiological context to another—most commonly from experimental animals (e.g., rodents, primates) or technological constructs (e.g., organ-on-chip platforms) to human subjects.  This is typically required for understanding human biology because of the strong constraints on measurements and perturbations in human in vivo contexts.  Direct translation of observations from experimental animals to human subjects is generally unsatisfactory because of significant differences among organisms at all levels of molecular properties from genome to transcriptome to proteome and so forth.  Accordingly, addressing inter-species translation requires an integrated experimental/computational approach for mapping comparable but not identical molecule-to-phenotype relationships.  This presentation will describe methods we have developed for a variety of cross-species translation examples, demonstrated on applications in inflammatory pathologies and cancer.

While several methods exist for sampling tissues in clinical contexts, without high-fidelity tools for comprehensively profiling them, we are both limited in our capacity to understand how constituent cells and their interactions impact prognosis, and to select and develop precision therapeutics. Recent years have witnessed transformative and intersecting advances in nanofabrication and molecular biology that now enable deep profiling of low-input samples. Collectively, these afford new and exciting opportunities to study cellular heterogeneity, starting from the level of the single cell, and may unlock the diagnostic, prognostic, and discovery potential of clinical isolates. Illustratively, I will introduce how we can leverage single-cell genomic approaches – and, in particular, single-cell RNA-Seq – to explore the extensive functional diversity between cells, uncovering, from the “bottom-up,” distinct cell states and their molecular drivers. Moreover, I will discuss high-throughput experimental strategies and demonstrate, in the context of Acute Lymphoblastic Leukemia, how they can be leveraged to achieve the statistical power necessary to reconstruct intracellular circuits, enumerate and redefine cell states and types, and transform our understanding of cellular decision-making in health and disease on a genomic scale.

Glioblastoma is well described genomically, but expression of lineage and signaling proteins has not been extensively characterized in single cells. Here, 34 different phospho-proteins, transcription factors, and lineage proteins were evaluated in millions of cells from resected glioblastomas using mass cytometry and a new, unsupervised, machine learning algorithm that identifies single-cell phenotypes and assesses clinical risk stratification as a continuous variable. This Risk Assessment Population IDentification (RAPID) approach identified tumor cell subsets that independently and continuously stratified patient outcomes, were distinguished by abnormal co-expression of lineage proteins, and harbored opposing signaling axes, suggesting the opportunity to match precision medicine approaches to these prognostic signatures. Once revealed by high-dimensional cytometry, the glioblastoma signaling profiles were apparent in tissue microarray cores using traditional pathological approaches. 

Pre-print: https://www.biorxiv.org/content/10.1101/632208v3

T cells became the central focus of new cancer therapeutics. Immune checkpoint inhibitors targeting T cell signaling pathways and cellular therapeutics utilizing chimeric antigen receptors (CARs) have shown success in the clinic. Discovery of previously unknown genes that modulate T cell function is of urgent need to open different avenues for immunotherapies, as a large fraction of patients still do not respond to, or have undesired side effects to currently approved ones. Systematic approaches to identify new regulators of T cell functions in vivo can provide potentially orthogonal or complementary opportunities. We recently performed in vivo CRISPR screens in CD8 cytotoxic T cells, both in a genome-scale and in a focused manner, in tumor models of immunotherapy (Dong et al. 2019 Cell). Furthermore, we developed a novel AAV-SleepingBeauty system that enhanced the power of genetic screening in primary T cells (Ye et al. 2019 Nature Biotechnology). Our screen re-discovered prime immunotherapy targets such as PD-1, TIM-3 and LAG3, as well as previously undocumented targets. We characterized novel targets such as DHX37, ODC1, MGAT5 and PDIA3 for their activity in mouse and human CD8 T cells including CAR-Ts.

Current major types of immunotherapy include checkpoint blockade, adoptive cell transfer, human recombinant cytokines, and cancer vaccines. However, immunotherapy has met challenges in immunologically cold tumors. These challenges urge for new types of immunotherapies that are more potent and potentially less toxic. Recently, we have developed CRISPRa-mediated Multiplexed Activation of Endogenous Genes as an Immunotherapy (MAEGI) (Wang et al. 2019 Nature Immunology, in press). The CRISPR activation (CRISPRa) system uses a catalytically inactive Cas9 (dCas9), enabling simple and flexible gene expression regulation through dCas9-transcriptional activators paired with single guide RNAs (sgRNAs). This enables precise targeting of large gene pools of endogenous genes in a flexible manner. We demonstrate that MAEGI has therapeutic efficacy across three tumor types. Mechanistically, our preliminary work showed that MAEGI treatment elicits anti-tumor immune responses by recruiting effector T cells and remodeling the tumor microenvironment. We will perform advanced development, characterization and optimization of MAEGI, as a novel immune-gene therapy approach to elicit a potent and specific immune response to tumors based on their unique genetic composition.

Wnt signaling is necessary for tissue development and homeostasis, but aberrant, overactive signaling is present in almost half of known human cancers. Colorectal cancer is the signature tumor type where chronically high Wnt signaling causes more than 80-90% of cases. The mechanism by which Wnt signaling is overactivated is well understood, but there is a lack of understanding what level of Wnt signaling drives tumors to progress and metastasize. For example, patients with the poorest prognosis can harbor the lowest levels of Wnt signaling. Further, biomarkers of active Wnt signaling are heterogeneous in human tumors. Recently, we made the surprising observation that Wnt activity in human colon tumors xenografted into mice is not only heterogeneous but spatially patterned with biomarkers of high Wnt activity are embedded among cells with lower Wnt activity, forming quasi-periodic arrays even though the injected cells were clonally re-derived, and thus genetically homogeneous at the time of injection. Similar patterns were seen by antibody staining for Pyruvate Dehydrogenase Kinase (PDK1) activity, a marker of glycolysis and the lactate transporter MCT-1. Since PDK1 and MCT-1 are Wnt target genes, this suggests a direct link between Wnt signaling and Warburg metabolism. Overall, the data suggest that Wnt-high spots and Wnt-low surrounds represent metabolically distinct cell subsets. Analysis of the tumors at the single cell level using scRNAseq reveals that the xenograft tumors are strikingly heterogeneous and that Wnt and metabolic heterogeneity is transcriptomic at the individual cell level. Differential expression of signaling molecules and receptors among these populations suggests mechanisms for how cancer cells interact with each other and with stromal cells and how the self-organized, patterned communities emerge in these experiments. To test these hypotheses, we develop a multispecies mathematical model in which transitions between cell types are driven by the levels of Wnt signaling. The model also incorporates interactions among the tumor and stromal cells, guided by the scRNAseq analysis. Signaling from FGFs, Wnts and ligands from TGFb/BMP/GDF family were notable including their expression in the stroma.  We found that the model reproduces the spatially heterogeneous spotted patterns of Wnt signaling we observe in subcutaneous colon xenograft tumors and that the model predicts changes in the patterning and tumor sizes when Wnt signaling is perturbed consistent with experimental observations. Strikingly, the model predicts that signaling from the stroma enhances instability of the tumor-host interface and leads to increased stromal content within the tumor that is also observed in experiments. Finally, we use the model to predict the effectiveness of different strategies for therapy.