Autophagy is a highly regulated catabolic process that degrades cytosolic organelles and macromolecules. Deregulation of the autophagic process has been linked to the development of several pathologies, including cancer, where it is associated with both pro- and anti-tumor functions. Although multiple studies have provided mechanistic insight into how autophagy suppresses tumorigenesis, the mechanisms explaining a pro-tumor role for autophagy have been limited to its capacity to promote tumor cell viability in response to stress. We hypothesized that autophagy contributes to additional cellular processes that support oncogenic transformation, and here we identify previously unrecognized pro-tumor functions for autophagy in the context of oncogenic Ras.

Using autophagy deficient cells, generated through genetic deletion or RNAi-mediated depletion of critical autophagy genes, we have identified a unique function for autophagy in supporting Ras-mediated anchorage-independent growth and proliferation. Furthermore, in the context of constitutively active Ras autophagy facilitates an enhanced rate of glycolysis, which supports robust anchorage-independent growth. Using a three-dimensional epithelial culture model, we demonstrate a requirement for autophagy in facilitating tumor cell invasion and motility in Ras transformed cells. This reduction in invasive capacity correlates with a partial restoration in cell-cell junctional integrity and polarized secretion of basement membrane proteins. In addition, autophagy supports Ras-driven mesenchymal differentiation and the production of multiple secreted factors required for cell migration and invasion. Overall, these data reveal multiple tumor-promoting functions for autophagy during Ras transformation. Based on our findings we propose that in addition to its widely established function in promoting tumor cell survival, autophagy facilitates a larger repertoire of pro-tumor activities that support cancer progression and metastasis.

Apoptosis, a form of programmed cell death, has been shown to be critical to the pathogenesis of a number of diseases, including neurodegenerative diseases. I investigated the role of BAX/BAK-dependent apoptosis in neurons in the context of neurodegeneration and maturation. BAX/BAK deficiency in the central nervous system of mice provides the ideal model by which to study the role of apoptosis in neurodegeneration. In a mouse model of Amyotrophic Lateral Sclerosis--a neurodegenerative disease that specifically affects motor neurons--deficiency of BAX and BAK preserves viability and function of neurons for an extended period. Moreover, this correlated

with a delay in symptom onset and extension of lifespan. Therefore, BAX/BAK dependent apoptosis is activated early in the disease process and represents a viable therapeutic target for neurodegenerative diseases. Regenerative medicine is currently one of the therapeutic strategies being considered for treatment of neurodegenerative diseases. Using this method, the patient's own stem cells are differentiated in vitro into the affected neuronal

subtype and re-implanted back into the patient. Because of the promise of regenerative medicine, it is critical to understand the signaling pathways mediating the differentiation process in stem cells. Caspase activity has been reported to mediate the differentiation of embryonic stem cells (ESCs), and I discovered that this process requires BAX and BAK. BAX/BAK deficient ESCs

undergoing differentiation express lower levels of neural markers and higher levels of pluripotency markers. These results indicate that ESCs lacking BAX and BAK are far less effective at differentiating down the neural lineage. Furthermore, engagement of BAX/BAK dependent apoptosis induces differentiation of ESCs. Therefore, the BAX/BAK apoptotic pathway mediates differentiation of stem cells.

These results underline the importance of the BAX/BAK apoptotic pathway in the development and survival of neurons. Deficiency of BAX and BAK prolongs survival in the context of neurodegeneration and delays the maturation of stem cells down the neural lineage. The findings of my thesis provide insight into the basic biology of neurons, and how to better treat neurologic disorders.

Much of the recent progress on visual processing has been driven by deep learning and its bicameral heart of composition and end-to-end optimization. The machinery of convolutional networks is now ubiquitous. Its diffusion however was neither instantaneous nor effortless. To advance across the frontiers of vision, deep learning had to be equipped with the right structures: the true, intrinsic structures of the visual world.

This thesis incorporates locality and scale structure into end-to-end learning for visual recognition. Locality structure is key for addressing image-to-image tasks that take image inputs and return image outputs. Scale structure is ubiquitous, and optimizing over it learns the degree of locality for the task and data. Alongside structure, this thesis examines adaptive computation to help cope with the variability of rich image-to-image prediction problems. These directions are studied through the lens of local recognition tasks that require inference of what and where.

Fully convolutional networks decompose image-to-image learning and inference into local scopes. Factorizing these scopes into structured and free-form parts, and learning both, optimizes their size and shape to control the degree of locality. Adaptive computation across time, computing layers according to their rate of change, exploits temporal locality to improve the efficiency of video processing. Adaptive computation across tasks, extracting a latent representation of local supervision, transcends locality to non-locally guide and correct inference. Locality is the defining principle of our fully convolutional networks. Adaptivity equips our networks to more fully engage with the vastness and variety of vision.

ABSTRACT

Using switchable mouse models to assess the

therapeutic potential of MdmX and E2F3

by

Daniel Garcia

Diverse as tumors are, they share some common features that are hallmarks of cancer. Two of these hallmarks are the evasion of tumor surveillance mechanisms and deregulated proliferation. Aiming to address these problems, we have used novel switchable mouse models to test the therapeutic potential of MdmX and E2F3, two genes directly involved in allowing tumor cells to evade tumor suppression and to proliferate, respectively.

MdmX is a critical negative regulator of p53 whose over-expression in many cancers is thought to block p53 tumor suppressor function. Consequently, inhibiting MdmX has emerged as an intriguing approach to restoring p53 function in established cancers. However, MdmX-deficient mice exhibit p53-dependent death early in embryogenesis, which has precluded elucidation of the consequences of systemic MdmX inhibition. To determine directly the effects of systemic MdmX inhibition in normal adult tissues and in tumors, we crossed mdmX-/- mice into the p53ERTAM knock-in background to obtain adult MdmX-null mice in which p53 function can be restored. p53 became active when transiently restored in these mice, but the effects of this activity were non-lethal and reversible. Moreover, absence of MdmX enhanced p53 restoration therapy in a mouse model of lymphoma. Hence, systemic inhibition of MdmX is a feasible therapeutic strategy for restoring p53 function in tumors that retain wild type p53.

The E2F family of transcription factors plays a key role in a number of cellular processes, including cellular proliferation and tumorigenesis. E2Fs are deregulated in the vast majority of cancer and appear to be essential conduits for upstream oncogenic lesions. Recent genetic studies in mice have established E2F3 as the most essential activating E2F for cellular proliferation and embryonic viability. However, the embryonic lethality of E2F3-null mice has prevented the study of the contribution of E2F3 to tumor proliferation and the assessment of E2F3 inhibition as a therapeutic avenue. To directly study E2F3 biology in adult tissues and tumors, we established a novel tet-repressor system that allows control of endogenous e2f3 expression in adult mice.

Although induction of blood vessel growth is acknowledged as a pivotal requirement for the evolution of macroscopic tumors, the events that trigger onset of tumor angiogenesis remain largely obscure. The pervasive Myc oncoprotein is itself a potent inducer of angiogenesis in a wide range of tissues. We have used a reversibly switchable mouse transgenic model of Myc-dependent beta-cell carcinogenesis to delineate the kinetics and causal sequence of angiogenic processes following acute Myc activation. We show that onset of endothelial cell proliferation is induced shortly after Myc-induced cell cycle entry of beta cells. Endothelial cell proliferation is not indirectly induced by local tissue hypoxia but instead via a diffusible angiogenic signal produced by Myc-expressing beta cells. This signal triggers the release of pre-existing, sequestered VEGF from the islet extracellular matrix, that then homes to the endothelial compartment where it induces endothelial cell proliferation. Myc activation in beta cells rapidly induces expression and release of the proinflammatory cytokine interleukin 1 beta (IL-1 beta). We show that IL-1 beta is the principal effector downstream of Myc responsible for triggering rapid onset of islet angiogenesis. Together, our data delineate a complete pathway in vivo by which the highly pleiotropic Myc oncoproteins elicits coexpansion of the vascular compartment during tumorigenic progression.