Negative Regulators of Angiogenesis. during Wound Healing

Transcription

1 Negative Regulators of Angiogenesis during Wound Healing BY MATEUSZ S. WIETECHA B.S., University of Illinois at Chicago, 2008 Summa Cum Laude, Highest Departmental Distinction, Honors THESIS Submitted as partial fulfillment of the requirements for the degree of Doctor of Philosophy in Oral Sciences in the Graduate College of the University of Illinois at Chicago (UIC) 2015 Chicago, Illinois Defense Committee: Luisa A. DiPietro, DDS, PhD Chair and Advisor, UIC Praveen Gajedrareddy, BDS, PhD Periodontics, UIC Mark Lingen, DDS, PhD Pathology, University of Chicago Steven T. Olson, PhD Molecular Biology of Oral Diseases, UIC Herve Y. Sroussi, DMD, PhD Oral Medicine and Diagnostic Sciences, UIC

2 DEDICATION This Thesis is dedicated to my family, without whose support none of my accomplishments would be possible. To my mother and father, Elżbieta and Adam Wietecha, for sacrifices made when we immigrated to the United States, for their affectionate parenting and guidance, and for their unyielding encouragement in my academic endeavors. To my older siblings for being spectacular role models and life mentors, for always pushing me to realize my full potential. To my brother, Wojtek, for sharing his positive energy and passion for life, for sparking my interest in technology and pushing the boundaries of what is possible. To my brother s wife, Ania, for her kindness and moral support. To my sister, Monika, for her compassion and warmth, for inspiring me to pursue higher education in biomedical science, healthcare and ultimately in dentistry. To my sister s husband, Robert, and their wonderful children Jenna, Alex and Emilie for providing emotional support and much-needed fun throughout the long years of my training. ii

3 ACKNOWLEDGMENTS I thank my Thesis committee Dr. DiPietro, Dr. Gajedrareddy, Dr. Lingen, Dr. Olson, Dr. Sroussi for their advice and encouragement in the realization of my research goals. I especially wish to acknowledge Dr. Luisa DiPietro, my primary mentor and advisor, for helping to develop this research project and for the many years of unwavering support throughout my training. I wish to thank past and present members of the DiPietro laboratory for contributing to a stimulating environment that has been a pleasure to work in. I thank Dr. Lin Chen for his mentorship with experimental design and laboratory technique. I wish to acknowledge several students who worked in Dr. DiPietro s laboratory and helped with different aspects of my research project: Dr. Matthew Ranzer, Dr. Ariel Johnson, Mateusz Krόl, Elizabeth Michalczyk, Veronica Radford, Bianca Flores. I thank the collaborating laboratories of Dr. Tarun Patel and Dr. Steven Olson, especially Richard Swanson for help with protein purification. I thank Dr. Wendy Cerny, Julia Tulley, Katherine Long, and Yan Zhao for administrative assistance. iii

4 CONTRIBUTIONS OF AUTHORS Chapter 1 is a literature review that places my work in the context of the scientific field and represents a published manuscript [1] which was written primarily by me with some assistance from Dr. Wendy Cerny and Dr. Luisa DiPietro. Chapter 2 introduces various candidates for anti-angiogenic factors involved in wound healing that were mined from a microarray dataset obtained from Dr. Lin Chen in Dr. DiPietro s laboratory. Chapter 3 represents a published manuscript [2] for which I was a co-first author with Dr. Lin Chen. I was primarily responsible for data analysis as well as writing the manuscript, Dr. Matthew Ranzer worked closely with both of us on the experimental design and animal work, while the other authors are from Dr. Tarun Patel s laboratory who contributed the recombinant proteins used for in vivo experiments and the in vitro data in endothelial cells. Chapter 4 represents a series of as-yet-unpublished experiments designed and performed primarily by me with the help of Dr. Lin Chen and students in Dr. DiPietro s laboratory, including Mateusz Krόl and Elizabeth Michalczyk. Chapter 5 represents independent effort by me in a novel field of research. iv

5 TABLE OF CONTENTS CHAPTER PAGE 1. INTRODUCTION Blood Vessel Biology: a Matter of Life and Death Angiogenesis: a One-sided Understanding A Distinct Pattern of Angiogenesis and Regression during Wound Healing To be or Not to be: The Endothelial Cell as Battleground for Competing Signals Vessel Maintenance: Mechanisms of Homeostasis Loss of Quiescence: Endothelial Cell Priming, Activation and Sprouting Induced Plasticity and Dynamic Balance of Competing Signals The Anti-Angiogenic Switch: Tipping the Balance toward Regression Regression: Streamlining the Blood Vessel Network Endogenous Mediators of Vessel Regression Remodeling of the Extra-cellular Matrix: Dynamic Reciprocity in Vessel Regression Scar Formation: Does Regression Impair Tissue Remodeling? Endothelial Cell Apoptosis: Mechanisms at the Moment of Death Maturation: Bolstering the Essential Vessels Tumor Angiogenesis: the Case of the Ever-sprouting Blood Vessel Critical Knowledge Gaps and Motivations for Thesis Project IDENTIFICATION OF CANDIDATE ANTI-ANGIOGENIC FACTORS USING HIGH-THROUGHPUT TECHNOLOGY Introduction to the Wound Transcriptome Mining the Wound Transcriptome for Anti-angiogenic Factors Sprouty-2 and PEDF as prime candidates and foci for thesis Candidate for future exploration: Vasostatin-I Extra-cellular matrix-derived peptides as candidates for future exploration SPROUTY AS A PRIMER FOR VESSEL REGRESSION IN WOUNDS Introduction to Sprouty Project Materials and Methods Animals and Wound Model Total RNA extraction and real-time PCR Protein extraction and Western blotting Endothelial cell culture: migration and MAPK activation Treatment of wounds with recombinant proteins Immunofluorescent histochemistry v

6 TABLE OF CONTENTS (continued) CHAPTER PAGE 3.3. Experimental Results Spry2 mrna expression and protein production in murine skin wounds during healing Spry2-positive cell numbers increase in the wound bed during healing Endothelial cell migration and MAPK signaling are inhibited following incubation with recombinant TAT-tagged Spry Wound vascularity is decreased following exogenous application of recombinant TAT-tagged Spry MAPK signaling is decreased in TAT-Spry2 treated wounds, whereas TAT-Spry2Y55F treated wounds exhibit increased MAPK signaling Discussion PIGMENT EPITHELIUM-DERIVED FACTOR (PEDF) AS A DRIVER OF VESSEL REGRESSION IN WOUNDS Introduction to PEDF Project Materials and Methods Animals and Wound Model Culture of human keratinocytes and fibroblasts Purification of recombinant PEDF Treatment of wounds with recombinant proteins and antibodies Wound size measurements Total RNA extraction and real-time PCR for PEDF Protein extraction and ELISA for PEDF Analysis of wound macrovascular density Immunofluorescent histochemistry for PEDF, CD31, Vimentin, and alpha-sma Picrosirius Red analysis for collagen maturity Statistical Analysis Experimental Results Localization and production of PEDF in mouse skin Distinct pattern of PEDF expression and production during excisional wound healing Localization of PEDF in relation to fibroblasts and endothelial cells during excisional wound healing Exogenous PEDF inhibits angiogenesis and promotes collagen maturation during wound proliferation Inhibition of endogenous PEDF delays blood vessel regression and collagen maturation during wound resolution Discussion vi

7 TABLE OF CONTENTS (continued) CHAPTER PAGE 5. FINDING MEANING IN BIOLOGICAL COMPLEXITY: NETWORK ANALYSIS OF HEALING WOUNDS Introduction to Systems Biology of Wound Healing Introduction to Systems Biology and Biological Network Analysis Unraveling the Complexity of Wound Healing Materials and Methods Creation of Angiome, Matrisome, and Angiome/Matrisome protein-protein interaction networks Analysis of wound transcriptome data to generate differential expression values Linking transcriptome data to Angiome/Matrisome networks Experimental Results Visualization and Properties of Angiome and Matrisome networks Visualization, Properties, and Gene Ontology of Angiome/Matrisome sub-network Visualization of differential gene expression in Angiome/Matrisome sub-network separately in Skin and Oral Mucosa Visualization of Differential Gene Expression, Regulatory Motifs, and Gene Ontology in Angiome/Matrisome sub-network between Skin and Oral Mucosal Wounds Discussion CONCLUSIONS APPENDIX CITED LITERATURE vii

8 LIST OF FIGURES FIGURE PAGE 1.1 Blood vessel histology and diseases with angiogenic phenotypes Summary diagrams describing the timelines of tissue repair and of major pro- and anti-angiogenic factors Overview of the angiogenic process during wound repair Mechanisms of endothelial cell quiescence Mechanisms of endothelial cell priming and activation Mechanisms of endothelial cell apoptosis Diagram summarizing the angiogenic profiles in health and disease Diagram summarizing the major converging pathways that direct endothelial cell phenotypes of quiescence, priming/activation, sprouting, maturation, and apoptosis Diagram describing the time-course of healing 1-mm excisional oral and skin wounds Gene expression data of potential anti-angiogenic agents mined from the wound transcriptome Gene expression data of potential ECM-derived anti-angiogenic agents mined from the wound transcriptome Spry2 mrna expression and protein levels in mouse skin wounds during healing Specificity of the anti-spry2 antibody in wounded and unwounded murine skin Spry2-positive cell numbers increase in the dermis of the wound bed during healing Endothelial cell migration and MAPK signaling are inhibited following incubation with recombinant TAT-tagged Spry Wound vascularity is decreased following exogenous application of recombinant TAT-tagged Spry MAPK signaling is decreased in TAT-Spry2 treated wounds, whereas TAT-Spry2 Y55F treated wounds exhibit increased MAPK signaling Spry2 and Spry2 Y55F function in the context of endothelial cell MAPK signaling viii

9 LIST OF FIGURES (continued) FIGURE PAGE 4.1 Immunofluorescent histochemistry for PEDF in unwounded mouse skin Localization of PEDF in unwounded murine skin in relation to dermal fibroblasts and endothelial cells Comparisons of baseline PEDF transcript expression in dermal fibroblasts, keratinocytes, and other cell types Pattern of PEDF mrna and protein production during excisional wound repair Localization of PEDF during excisional wound repair in relation to dermal fibroblasts and endothelial cells Treatment of wounds with rpedf decreases blood vessel density, increases pericyte coverage, and increases collagen maturity during the proliferative phase of healing Treatment of wounds with neutralizing antibody against PEDF increases blood vessel density and decreases collagen maturity during the remodeling phase of healing Proposed Model for PEDF function in the context of wound healing Visualization of the Angiome and Matrisome protein-protein interaction networks Visualization of the shared genes/interactions within the Angiome and Matrisome networks and the resulting Angiome/Matrisome sub-network Fold changes in gene expression in Angiome/Matrisome sub-network in skin and oral mucosal wounds during the phase transitions from Inflammation to Proliferation and Proliferation to Resolution Fold changes in gene expression in Angiome/Matrisome sub-network in skin and oral wounds during the phase transition from Inflammation to Resolution/Remodeling Differential network analysis in the transition from Inflammation to Resolution Differential network analysis in the transition from Proliferation to Resolution ix

10 LIST OF ABBREVIATIONS α-sma alpha-smooth muscle actin ANOVA analysis of variance CXCL10 C-X-C motif chemokine 10 CD31 cluster of differentiation 31 EC endothelial cell ECM extra-cellular matrix FB fibroblast FGF fibroblast growth factor GAPDH glyceraldehyde 3-phophate dehydrogenase GFP green fluorescent protein GF growth factor GO gene ontology HIF-1α hypoxia inducible factor 1-alpha IGF insulin-like growth factor KC keratinocyte KO knockout MAPK mitogen-activated protein kinase MMP matrix metalloproteinase MVEC microvascular endothelial cell NS normal (unwounded) skin PBS phosphate-buffered saline PEDF pigment epithelium-derived factor PDGF platelet-derived growth factor RTK receptor tyrosine kinase RT-PCR reverse transcription polymerase chain reaction SEM standard error of the mean TAT trans-activator of transmission TSP thrombospondin VE-Cadherin vascular endothelial cadherin VEGF vascular endothelial growth factor VSMC vascular smooth muscle cell x

11 1. INTRODUCTION Parts of this chapter have been previously published as [1]: Wietecha MS, Cerny WL, DiPietro LA. Mechanisms of Vessel Regression: Toward an Understanding of the Resolution of Angiogenesis. Current Topics in Microbiology and Immunology. 2013; 367: PubMed ID: doi: /82_2012_287. See Appendix for copyright license information BLOOD VESSEL BIOLOGY: A MATTER OF LIFE AND DEATH Blood vessels are vital to the development and survival of complex multi-cellular life. The marvelously efficient method devised by nature to deliver oxygen to cells, tissues and organs, most of which are located in otherwise anoxic environments, is what makes possible the vertebrates evolutionarily advanced forms and functions. Indeed, because of the physical constraints of gas diffusion, organisms without a functioning vascular system wouldn t be able to grow beyond a dozen cells in any direction [3]; in humans, practically no cell is ever more than four cells away from a blood vessel [4]. The fractal-like branching of the cardiovascular system, from the largest vessel (the aorta) to the smallest (capillary) and back to the largest (the vena cavas), ensures that every one of the roughly fifty trillion cells in the human body is adequately supplied with oxygen and nutrients, and can dispose of its waste products. Given this seminal importance to life, it should not be surprising that mammals have evolved complex mechanisms for the intricate control of blood vessel growth and maintenance. During development of the organism, blood vessels form de novo in a process called vasculogenesis, whereas further sprouting of vessels from pre-existing tubules is called angiogenesis [5]. Besides development, angiogenesis occurs physiologically in adults during the female menstrual cycle in the ovary [6] and the endometrium [7], in skeletal muscle remodeling during exercise [8, 9], and during tissue repair and regeneration [10]. In all of these cases of naturally-occurring, or physiological, angiogenesis, sprouting of tubules is followed by a controlled phase of blood vessel maturation and regression which results in a spatially distributed and completely perfused network that optimally meets a tissue s metabolic demands as required to preserve homeostasis [11, 12]. 1

12 2 Angiogenesis is an integrated function of a variety of cell types and their microenvironment [13] and involves the following stages: activation, sprouting, regression, and maturation. Endothelial cells (ECs) form the inner wall of every blood vessel and are the primary cell type involved in the angiogenic process (Fig. 1.1A). During angiogenesis, ECs are induced to proliferate, migrate out of an existing vessel, differentiate and assemble to form branches of tubules capable of carrying blood and its constituents [14-16]. EC-EC connections are mediated by inter-cellular junctional protein complexes called vascular-endothelial cadherins (VE- Cadherins) which regulate tubule leakage, or vasopermeability [17]. Destabilization of VE- Cadherins promotes blood vessel sprouting, whereas strengthening the EC-EC connections is indispensable during vessel maturation and maintenance. While ECs by themselves initially form tubules, these fragile structures need to be stabilized by supporting mural cells (Fig. 1.1A). The pericyte is the most intimate partner of the EC and, by binding directly to multiple ECs through a shared basement membrane, is capable of express cell-cell communication, promoting vessel barrier integrity and vascular network stability [18]. Another type of mural cell is the vascular smooth muscle cell, which envelopes the basement membrane of the EC tubule and mediates vessel tone for fine control of blood flow to meet local tissue metabolic demands [19]. The entire blood vessel network is surrounded by the extra-cellular matrix (ECM), which acts as a bioactive heterogeneous scaffold transmitting molecular signaling and mechanical cues to the blood vessels through cell membrane integrins and other receptors [20-23] (Fig. 1.1A). The ECM composing the microenvironment of blood vessels is highly dynamic, and it is constantly undergoing remodeling through the actions of fibroblasts, cells located within the ECM which are responsible for its maintenance, regulating its composition and biomechanical properties, thus directly influencing blood vessel biology [24-26]. During vessel sprouting, the ECM microenvironment is an important source of pro-angiogenic factors in addition to its role as a permissive scaffold on which the network can form. During vessel maturation and regression,

13 Figure 1.1: Blood vessel histology and diseases with angiogenic phenotypes. A) Schematic demonstrating that blood vessels are tubes of endothelial cells surrounded by a complex microenvironment composed of mural cells (pericytes, smooth muscle cells, fibroblasts) and proteins of the extra-cellular matrix (collagens, proteoglycans, matri-cellular proteins, cytokines). B) Diagram that shows which major human diseases have angiogenic phenotype, either excessive or insufficient angiogenesis. Highlighted are conditions that will be addressed in this dissertation, with notes that current single-agent therapies are not effective. The major theme of this dissertation, the importance balance between blood vessel growth and regression which leads to tissue homeostasis and health, is highlighted. 3

14 4 the ECM also undergoes remodeling into a more mature phenotype, providing crucial antiangiogenic cues, both molecular [27] and biomechanical [20], which guide the vessels into a spatially distributed and functional network that is optimally suited to maintain tissue homeostasis [11]. Thus, angiogenesis refers not only to the growth but also to the maturation and regression of the blood vessel network, a superbly intricate process involving coordination between multiple cell types and their dynamic microenvironment. Given the complexity of angiogenesis and its importance to life, it should not be surprising that many prominent human diseases feature angiogenic phenotypes [28], whose underlying pathology involves a dysfunction of the myriad of mechanisms that comprise the angiogenic process (Fig. 1.1B). Excessive blood vessel growth is a hallmark of all malignancies, and it is an important phenotype in arthritis, psoriasis, and macular degeneration effecting the onset of blindness. Insufficient angiogenesis occurs in chronic wounds of diabetic patients, and non-healing often leads to amputations of affected limbs [29]. The increasing public health burdens of obesity and diabetes mellitus go hand-in-hand with multiple cardiovascular diseases, including myocardial and peripheral ischemia, which present with deficient angiogenesis leading to infarcts and necrosis of affected tissues [28]. In many explored cases of pathology, there is an imbalance between vessel growth and regression, which perturbs tissue homeostasis (Fig. 1.1B). The paramount importance of angiogenesis in health and disease accentuates the necessity for rigorous research in this area --- it is quite literally a matter of life and death. Scientific progress here, with a goal to fully understand the underlying physiological and pathological processes, will make it possible to successfully intervene therapeutically and to promote systemic health in millions of current and future patients.

15 ANGIOGENESIS: A ONE-SIDED UNDERSTANDING It is widely accepted that angiogenesis is regulated in a spatial and temporal manner via a dynamic balance between pro- and anti-angiogenic factors [30-32]. The particular inducers and inhibitors of angiogenesis belong to many classes of molecules, depending on the specific cellular function which they regulate, and range from large soluble proteins and cleaved ECMderived peptides to intracellular molecules that modulate signaling pathways [33]. Because angiogenesis is an integrated function of multiple cell types as well as their microenvironment [13] (Fig. 1.1A), the overall process as it occurs in vivo is extremely complex. Nonetheless, over the last couple of decades, using a multitude of clever in vitro and animal models [34], investigators have made tremendous progress in unraveling the molecular mechanisms behind the many aspects of angiogenesis [5]. The mechanisms that promote blood vessel growth and sprouting have been explored in multiple systems [35] and are now well-defined [14-16]. In contrast, the essential stop-signals which inhibit further growth of the vasculature and mediate blood vessel regression are not understood [8, 36, 37]. This is partly because the most popular in vivo models for studying angiogenesis, like sponge or tumor implantation, are ones that induce pathological blood vessel growth and by design do not present with appropriate negative feedback responses [34]. Since these models of pathological angiogenesis lack essential physiological responses --- e.g. the production of endogenous anti-angiogenic factors which promote vessel pruning leading to vascular homeostasis --- they have been used to evaluate a whole range of exogenously applied and potentially therapeutic anti-angiogenic agents [34]. Ironically, with so much emphasis on models of pathological angiogenesis in search of the magic bullet molecule which would stop the process in its tracks, very little is known about the endogenous anti-angiogenic factors and mechanisms that are actually utilized by the organism to counter over-exuberant neovascularization [38]. These resolving processes take place during physiological

16 6 angiogenesis, which historically seems to be a less interesting biological phenomenon because of its link to health rather than to disease. However, due to the lack of groundbreaking clinical successes in combating vascular pathologies thus far (Fig. 1.1B), it is becoming apparent that we are missing an important piece of the angiogenesis puzzle. A question could be raised: how are we to therapeutically direct such a complex biological process as angiogenesis without truly understanding how it naturally resolves? We argue here that knowing only one side of the equation is not enough. The purpose of this Introduction is to shed some light on the relatively unknown but crucial aspect of the angiogenic process: blood vessel regression following robust vessel sprouting. The question we wish to answer is: how does the repaired blood vessel network return to vascular homeostasis? To get at a more global view, we will explore vessel regression from multiple perspectives and attempt to fit the pieces back into a framework of basic understanding. This framework will be modeled after a biological process which encompasses the entire angiogenic response, from induction to resolution: wound healing in the skin [10, 39]. The relevant literature from assorted disciplines, featuring studies utilizing various in vitro and animal models, will be reviewed and synthesized in the context of repair and regeneration, assuming for the purpose of this introduction that physiological anti-angiogenic processes share common mechanisms in different tissues. With the global framework in mind, we will identify gaps in our current understanding and suggest directions for further research in this area. We hope that this Introduction will clear the stage for the following chapters which go into experimental detail describing the roles of two anti-angiogenic factors and into the systems analysis of angiogenesis in wounds.

17 A DISTINCT PATTERN OF ANGIOGENESIS AND REGRESSION DURING WOUND HEALING Wound repair follows a well-characterized sequence of events, or phases: hemostasis, inflammation, proliferation, remodeling [40, 41] (Fig. 1.2A). In the many areas and microenvironments of the wound, each phase --- through complex coordination between the cells and ECM of the healing wound --- mechanistically brings about the next phase in the sequence [13, 42]. The phases of wound repair are best described as distinct yet overlapping, and they may be illustrated as gradients, both spatially and temporally [13, 41] (Fig. 1.2A). The proliferative phase of healing is characterized by the formation of granulation tissue, a rather chaotic collection of cells (macrophages, fibroblasts, ECs) and loose connective tissue that quickly replaces the fibrin clot in the wound bed. One of the primary components of granulation tissue is a dense network of capillary loops which comes about as a result of a vigorous angiogenic response starting at the wound margins and spreading into the wound bed [10] (Fig. 1.3A-C). In a little over a week after injury, the density of blood vessels in the wound bed is over three times higher than that of the original, uninjured tissue [43, 44] (Fig. 1.2B). The vast, rapidly forming neovascular network is supported by a fibroblast-secreted temporary fibrous scaffold that is rich in Collagen Type III and other ECM components that promote EC invasion through the matrix during sprouting [13] (Fig. 1.2A). What stimulates and directs blood vessel growth in wounds has been explored in much detail [10]. The critical initial stimulus is low oxygen tension, or hypoxia, experienced by the tissue as a result of damage to the existing blood vessel network [45]. The stress of hypoxia stimulates substantial transcriptional changes in the affected cells of the wound [46]. Nearly all cell types in and around the wound are induced to produce massive amounts of various proangiogenic factors (Fig. 1.2B), which act to promote EC proliferation, migration, and differentiation into a branched tubular network that sprouts toward the hypoxic gradient [47] (Fig. 1.3B-C).

18 Figure 1.2: Summary diagrams describing the timelines of tissue repair and of major proand anti-angiogenic factors. A) Timeline of phases of healing in this wound model, temporally correlated to phenotypes of connective tissue remodeling and blood vessel physiology. NS = normal skin (unwounded), ECM = extra-cellular matrix. B) Timeline of important pro- and antiangiogenic factors in excisional wounds, temporally correlated to phases of healing and relevant phenotypes in (A). Pro-angiogenic factors include: platelet-derived growth factor (PDGF), vascular endothelial growth factor (VEGF), basic fibroblast growth factor (FGF-2) [8, 30, 39]. Antiangiogenic factors include: Sprouty2 [present work Chapter 3], Interferon Gamma-induced Protein 10 (IP-10, or CXCL10) [66], thrombospondin-2 (TSP-2) [69], pigment epithelium-derived factor (PEDF) [present work Chapter 4]. 8

19 9 The majority of the newly sprouted vessels are mere EC tubules whose network architecture is tortuous and often blind-ended [4]. Functionally, these neovessels are immature, lacking tight EC-EC contacts and possessing scant coverage by pericytes, and have been found to be barely perfused and leaky, contributing to granulation tissue s moist appearance [48, 49] (Fig. 1.3C). Importantly, the disorganized and non-functional nature of the neovessel network during this phase of wound healing has been compared to the vasculature of a solid tumor [38, 50] (Fig. 1.7). Once wound closure and maximum granulation tissue formation are achieved, the wound enters the final phase of healing: remodeling or resolution (Fig. 1.2A). An important component of the remodeling phase is the systematic breakdown of the provisional, angiopermissive Collagen Type III-rich ECM and its gradual replacement with one that is composed primarily of Collagen Type I, initiating a lengthy process of modification of the wound ECM to an architecture whose composition and bio-mechanics resemble the pre-wounded, angio-restrictive state [41]. With the completion of wound closure, the barrier function of the epithelium is achieved, leading to decreased microbial invasion and thus a much reduced oxygen demand by the inflammatory cells [45]. As granulation tissue fills up the space of the wound bed, the cells involved experience contact inhibition and begin to wind down their metabolically demanding proliferative and migratory behaviors. With decreased oxygen demand in the wound, there is a decline in tissue hypoxia, which results in a diminished production and activity of soluble proangiogenic mediators (Fig. 1.2B). Thus, during the remodeling phase, the wound switches to an anti-angiogenic state, and this switch is characterized by a number of critical changes to the vessel microenvironment (Fig. 1.2A).

20 Figure 1.3: Overview of the angiogenic process during wound repair, showing relationships between endothelial cells (EC), mural cells (pericytes and fibroblasts), vessel perfusion, and extra-cellular matrix (ECM) remodeling. Black boxes correspond to the detailed mechanistic images in Figures A) The default state of the vasculature when tissue is at homeostasis is that of quiescence, maintained in part by adequate perfusion and ample pericyte coverage. See Figure 1.4 for detailed representation of the mechanisms of EC survival and quiescence. B) Injury perturbs the homeostasis of affected tissue, leading to inadequate oxygen supply and hypoxia-driven EC activation in nearby vessels. See Figure 1.5 for detailed representation of the mechanisms of EC priming and activation. C) Pro-angiogenic factors and an angio-permissive provisional ECM promote robust sprouting of neovessels toward the hypoxic gradient of the wound, resulting in a chaotic network of immature and leaky tubules characteristic of granulation tissue formation. D) As the wound enters the remodeling phase, vessels which have reconnected to a viable network and are perfused are selected for maturation via pericyte coverage, while those inessential tubules that are not perfused are eliminated via systematic pruning by various anti-angiogenic factors. See Figure 1.6 for detailed representation of the mechanisms leading up to EC apoptosis during vessel regression. E) Over time, all extraneous tubules are eliminated, and the vessel network as well as the ECM are remodeled and return to an architecture and functionality resembling the original, pre-wound homeostatic state. Previously published in [1] as Figure 1. 10

21 11 After a peak in vessel density is reached in the wound bed, the anti-angiogenic phenotype prevents further vessel sprouting and mediates the regression of blood vessels back to baseline levels [41, 44] (Fig. 1.2B). While a minority of newly sprouted vessels which have successfully integrated into the existing perfused network undergo maturation [31], the majority that are not perfused and functional are predisposed to pruning [51] (Fig. 1.3D). The most accepted theory for how pruning of extraneous vessels occurs is that the ECs comprising these tubules are induced to undergo programmed cell death, or apoptosis [52, 53]. Though the exact mechanisms leading to and mediating EC apoptosis during wound healing are largely unknown, they likely come about as a result of a convergence of a number of biological processes. Besides the oxygen-associated changes to the vessel microenvironment, certain active antiangiogenic signals are thought to fine-tune the proper pruning of the blood vessel network during the remodeling phase. The known regression signals include soluble and ECM-derived anti-angiogenic mediators which lead to specific intracellular signaling pathways that result in the cellular and microenvironmental changes associated with vessel regression.

22 TO BE OR NOT TO BE: THE ENDOTHELIAL CELL AS BATTLEGROUND FOR COMPETING SIGNALS Vessel Maintenance: Mechanisms of Homeostasis In the adult, the default state for blood vessels is that of quiescence (Fig. 1.3A). Quiescent ECs are characterized by a remarkably low proliferative potential, stable VE- Cadherin complexes and ample coverage by mural cells, especially by pericytes [17, 18]. Importantly, stabilized vessel networks have been shown to be resistant to both pro- and antiangiogenic stimuli [54]. This protects tissues from random sprouting of superfluous vasculature, while at the same time protecting the existing blood vessel network from inadvertent pruning. This state of quiescence, critical for homeostasis, is maintained by multiple mechanisms (reviewed in [55]) (Fig. 1.4). Major disturbances to the homeostatic state, such as injury and/or hypoxia, can prime ECs comprising the affected tubules for either angiogenesis or pruning. Thus, the mechanisms which have been identified for blood vessel network stability are likely lost or lacking in areas of active angiogenesis and regression [55]. During wound healing, one of the central end-points of physiological vessel growth and regression is a return to a vascular state of quiescence, as this state contributes largely to tissue homeostasis (Fig. 1.3E). Vascular Endothelial Growth Factor (VEGF) is thought to be the most potent proangiogenic mediator during wound repair [56], but it also has significant roles in vessel maintenance [55] (Fig. 1.4). When tissue is at homeostasis, basal levels of VEGF support the survival of the stable vascular network in an autocrine fashion by maintaining critical antiapoptotic Akt signaling in ECs [16]. Basal levels of the primary pro-angiogenic VEGF receptor (VEGFR2) are also required for survival of quiescent vessels, as the shear stress in these perfused tubules activates VEGF-independent anti-apoptotic signaling [51].

23 13 Figure 1.4: Mechanisms of endothelial cell quiescence. Factors contributing to EC survival and quiescence in homeostasis include normoxia, stable EC integrin contacts with a mature extra-cellular matrix (ECM), stabilization by adherent pericytes, laminar flow causing shear stress on the luminal surface, tight EC-EC connections via VE-Cadherin junctional complexes, and maintaining baseline autocrine VEGF signaling while preventing pro-angiogenic overstimulation by trapping extra-cellular VEGF with soluble VEGFR1 and internalizing extraneous membrane VEGFR2. Previously published in [1] as Figure 2A. Quiescent vasculature is protected from VEGF overstimulation via modification of signaling events by VE-Cadherin, including its deactivation of over-expressed VEGFR2 [55]. Simultaneously, a competing VEGF receptor, the soluble VEGFR1, is expressed in quiescence and functions as a VEGF trap, further limiting stimulation of ECs by VEGF [16]. In addition, vessels surrounded by pericytes are remarkably immune to VEGF overstimulation and EC

24 14 destabilization [54]. These and other mechanisms assure the long-term survival of a quiescent, properly perfused blood vessel network that optimally supplies the surrounding tissues (Fig. 1.3A, Fig. 1.4) Loss of Quiescence: Endothelial Cell Priming, Activation and Sprouting Loss of EC quiescence following injury is largely mediated by hypoxia and the ischemic state of the affected cells [46] (Fig. 1.3B, Fig. 1.5). The oxygen-sensing Prolyl Hydroxylase Domain (PHD) family of proteins are important intracellular mediators of vessel quiescence through their suppression of pro-angiogenic master transcriptional regulator Hypoxia Inducible Factor 1 Alpha (HIF-1α) [57]. Whereas under conditions of homeostasis PHDs continually target HIF-1α for proteasomal degradation, in hypoxia PHDs become inactivated, allowing HIF-1α to accumulate and overturn the transcriptional profile of the EC from quiescent to what might be called primed. Priming makes ECs much more sensitive to outside stimuli, especially proangiogenic ones. Multiple cell types including keratinocytes, fibroblasts, and macrophages in the early hypoxic wound produce large amounts of VEGF, while ECs comprising vessels in the wound margins react to hypoxia by over-expressing VEGFR2 on their membranes [57] (Fig. 1.5). Binding of VEGF to VEGFR2, a receptor tyrosine kinase, initiates mitogen-activated protein kinase (MAPK) signaling pathways, further overturning the cellular machinery and effectively activating the ECs. Activated ECs lose their protective mural cells and lose their stabilizing VE- Cadherin connections, contributing to their increased sensitivity to pro-angiogenic factors. Additional stimulation by VEGF and co-activation by sufficient matrix contacts to EC integrins leads to EC proliferation, migration, and differentiation into a branched tubular network that sprouts toward the VEGF gradient in the wound bed (reviewed in [14, 16, 47, 56]) (Fig. 1.2B, Fig. 1.3C). The leakiness of the neovessels in granulation tissue is mediated in large part

25 15 Figure 1.5: Mechanisms of endothelial cell priming and activation. Hypoxia leads to the priming and activation of the EC, resulting in the activation of master transcription factor HIF-1α, ample production of pro-angiogenic VEGFR2 that binds to abundant VEGF secreted by nearby hypoxic cells, co-activation by integrin binding to ECM, and leading to the loss of EC quiescence by dissociation of EC-EC and EC-pericyte junctional complexes. The activated EC is not only more sensitive to pro-angiogenic stimuli but also to apoptotic signals by concurrent production of the death receptor, Fas. Previously published in [1] as Figure 2B. by VEGF, which, through nitric oxide synthase and cyclooxygenase signaling [47], further dissociates VE-Cadherin junctional complexes [17]. The net result of hypoxia-induced proangiogenic activity is the remarkably vessel-rich, moist granulation tissue that fills the wound bed within a few days after injury (Fig. 1.2). Since tumors are known to stall at this stage of the angiogenic process [50] (Fig. 1.1B, Fig. 1.7), the vast majority of research in the field of angiogenesis also stops at this point [38].

26 16 Models of physiological vascular regression, such as healing wounds, provide knowledge about the mechanisms by which the tortuous and leaky neovessel network undergoes pruning back to baseline, leading to vascular and tissue homeostasis (Fig. 1.2, Fig. 1.7) Induced Plasticity and Dynamic Balance of Competing Signals Paradoxically, the pro-angiogenic mediator VEGF may be one of the factors responsible for the initiation of vessel regression in the post-proliferative phase of healing. Studies have found that activation of ECs by VEGF simultaneously marks these cells for death by inducing the expression of the death receptor Fas, also known as CD95, which, if bound by its ligand, initiates apoptotic signaling pathways [58, 59] (Fig. 1.6, Fig. 1.8). As a consequence, activated ECs enter a state of plasticity, making them not only more sensitive to pro-angiogenic stimuli but also less resistant to death by apoptosis-promoting signals. Once the ECs are primed by hypoxia and activated by VEGF during the proliferative phase, competing signaling pathways steer them toward multiple potential cellular behaviors: sprouting, maturation, or apoptosis (Fig. 1.8). The resulting phenotype may be a consequence of the dynamic balance between pro- and anti-angiogenic stimuli acting upon the ECs from their immediate microenvironment (Fig. 1.2B). The microenvironment in the wound is highly dynamic and changes drastically from the proliferative to remodeling phase; whereas the balance clearly favors pro-angiogenesis early on, the scale tips toward anti-angiogenesis in the later wound (Fig. 1.2). What happens at the junction between these two extremes, when pro-angiogenic factors are still present and antiangiogenic mediators have not yet become dominant, is an area of active investigation. How do the ECs of the wound neovessel network integrate multiple competing stimuli and eventually take the plunge toward regression?

27 The Anti-Angiogenic Switch: Tipping the Balance toward Regression A type of mechanism that likely aids during this critical period of the wound angiogenic profile --- a period which might be called the anti-angiogenic switch --- is a well-conserved one in biology: negative feedback. Indeed, there are probably several redundant intracellular negative feedback mechanisms protecting ECs from VEGF overstimulation during the postproliferative and remodeling phases of healing which help to guide them into regression. One such mechanism is mediated by a family of intracellular Sprouty proteins, which are known to inhibit pro-angiogenic MAPK signaling pathways like those initiated by VEGF in ECs [60]. The investigation of the function of Spry2 in dermal wounds is a major part of this dissertation and will be discussed in Chapter 3. In a similar fashion, vasohibin may modulate VEGF signaling in a negative feedback, although this mechanism is yet to be explored in the context of physiological angiogenesis [61]. These and other negative feedback mechanisms may play an important role as buffers to pro-angiogenic stimuli, making ECs immediately less sensitive to further pro-angiogenic stimulation, thus down-regulating the signaling pathways leading to mitogenic cell behavior, while simultaneously giving anti-angiogenic stimuli and their corresponding signaling pathways an advantage during the critical transitional period into the remodeling phase of healing (Fig. 1.8) Regression: Streamlining the Blood Vessel Network The leaky neovessel network of the granulation tissue is largely non-perfused [49] (Fig. 1.3C, Fig. 1.7). Insufficient perfusion likely leads to reduced shear stress and inadequate biomechanically-induced EC survival signaling, and this may be an essential mechanism mediating physiological vessel regression [20, 51] (Fig. 1.6). Furthermore, it is well-established that pro-angiogenic factors such as VEGF are reduced during the vessel regression phase of wound repair [44, 62] (Fig. 1.2B), although it is unknown whether the autocrine VEGF survival

28 18 Figure 1.6: Mechanisms of endothelial cell apoptosis. Factors contributing to EC apoptosis during vessel regression consist of several anti-angiogenic stimuli in the immediate wound microenvironment, including: soluble and matricellular molecules that can trigger the production of the death ligand, FasL, inducing apoptosis upon binding to Fas receptor; fibroblast-secreted matrix metalloproteinases (MMPs) that cleave ECM molecules as part of matrix remodeling, resulting in unstable ECM contacts with EC integrins, in the collapse of a stable scaffold, as well as in the release of anti-angiogenic ECM-derived peptides; turbulent or stagnant flow on the luminal surface causing loss of survival signaling. Previously published in [1] as Figure 2C. signaling is reduced along with the potent extracellular component of the VEGF stimulus. Several studies have shown that VEGF is required at baseline levels for the continued stability of some established vessel networks and that total VEGF inhibition causes EC apoptosis and regression [55, 63]. These data certainly point to a potentially pivotal role for the loss of molecular and especially biomechanical survival signals in the process of physiological vessel

29 19 involution, although these ideas have yet to be investigated in an appropriate in vivo model, such as wound healing. In contrast, there is good evidence that wound resolution involves the generation of an actively anti-angiogenic environment. Exogenous application of VEGF and other pro-angiogenic factors into a resolving wound does not prevent blood vessel regression from naturally occurring [64]. Indeed, while exogenous pro-angiogenic factors cause an even more exuberant angiogenic response during the proliferative phase of healing, the extra blood vessels subsequently undergo remarkably rapid pruning, and vessel density levels of treated wounds match those of the controls within a few days following the peak of granulation tissue formation [64]. It appears that the post-proliferative wound not only becomes more resistant to proangiogenic stimuli by the negative feedback mechanisms just discussed, but also generates active anti-angiogenic signals. During periods of active vessel regression, as occurs during the remodeling phase of wound repair, there exist potent endogenous anti-angiogenic factors which ensure proper pruning of the blood vessel network regardless of the presence of competing proangiogenic signals (Fig. 1.6) Endogenous Mediators of Vessel Regression Spurred by the promise of anti-angiogenesis in cancer therapy [65], dozens of endogenous inhibitors of angiogenesis have been identified over the last three decades [66]. Even though these molecules are derived from naturally occurring circulating or matrix components, the majority of characterization has been performed either exclusively in vitro or in models of pathological angiogenesis, e.g. in xenografted tumors. Thus, while the list of endogenous anti-angiogenic factors is extensive, very little is known about their actual functions in the control of physiological angiogenesis and vessel regression. How they may fit mechanistically into the larger process of remodeling during wound repair can only be

30 20 Figure 1.7: Diagram summarizing the angiogenic profiles in health and disease, incorporating the integrity of vessels as a spectrum, and the usual end-points of each condition: Oral wounds end in regeneration; dermal wounds repair with scar formation; chronic wounds are underhealing ; cancer can be described as an overhealing wound. speculated based on their origins (circulating, soluble or matrix-derived) and what limited information can be gleaned from available studies [66]. We will focus our discussion on those anti-angiogenic factors which have been explored in multiple systems and are thus most likely to play significant roles in physiological vessel regression. Cytokines, or small soluble molecules secreted by various cell types, are probable mediators of vessel regression, as many well-known cytokines that are ubiquitous in the wound environment have been found to have anti-angiogenic properties [67] (Fig. 1.6). One of the best studied cytokines in the context of vessel involution is the C-X-C motif chemokine 10 (CXCL10), also known as Interferon Gamma-induced Protein 10 (IP-10). CXCL10 binds CXCR3, a receptor commonly expressed on ECs [68], and knock-out studies for CXCR3 revealed a significant

31 21 delay in physiological vessel regression in murine dermal wound models [69, 70]. Interestingly, CXCR3-null wounds also exhibited a marked delay in ECM remodeling to a mature phenotype, further suggesting a strong association between these two processes. Mechanistic studies in vitro and in the matrigel plug model of pathological angiogenesis revealed that CXCL10 caused a dissociation of newly formed vessels via disruption of critical integrins, secondarily leading to EC apoptosis [69]. The anti-angiogenic effects were found to be strong even in the presence of pro-angiogenic factors, giving yet more credence to the view that a potent anti-angiogenic phenotype dominates in the resolving wound microenvironment. Finally, since CXCL10 is known to be produced in the post-proliferative and beginning of the remodeling phases of healing [70], the aggregate data strongly suggest a significant role for CXCL10 in physiological vessel regression. An important class of proteins, called matricellular proteins [71], which are capable of binding to the ECM microenvironment and acting upon resident ECs through specific receptors, are very likely to be instrumental in regulating physiological vessel involution (Fig. 1.6). Two members of the Thrombospondin family, TSP-1 and TSP-2, are relatively well-studied potent anti-angiogenic factors. As matricellular proteins, they are capable of binding multiple ECM components and thus regulating critical cell-cell and cell-ecm interactions [72]. These molecules have been found to inhibit angiogenesis by down-regulating EC proliferation and migration, inhibiting VEGF signaling, and initiating apoptosis [73]. Whereas TSP-1 is produced during the early phases of healing and likely functions to attenuate VEGF-mediated proangiogenic signals, TSP-2 is produced during the remodeling phase and is likely more involved in ECM remodeling-associated vessel regression [74]. Integrating into the ECM microenvironment of the blood vessels ensures prime real estate for TSPs spatio-temporal down-regulation of wound angiogenesis. However, a recent evaluation of wound healing in TSP-2-null mice did not find differences in EC apoptosis as compared to wild-type mice;

32 22 instead, TSP-2 was implicated in the regulation of ECM remodeling in wounds [75], probably by tipping the matrix protease balance toward the creation of the angio-restrictive ECM scaffold of the resolving wounds. Pigment Epithelium-Derived Factor (PEDF) is one of the most potent endogenous antiangiogenic mediators, able to inhibit EC mitogenic behavior as well as induce EC apoptosis [59, 76]. Similarly to TSPs, PEDF is capable of binding multiple ECM components, especially Collagen Type I, an ubiquitous component of the resolving wound [77]. Many of PEDF s properties, including its ability to specifically target neovessels without disrupting intact mature vasculature [59], as well as its ability to reduce vasopermeability [78], imply an over-arching role of promoting vascular homeostasis in many tissues [76]. Though PEDF s in vivo functions have only been explored in models of pathological angiogenesis, such as in tumors and in ocular neovascularization, the aggregate data strongly suggest that PEDF also plays an important role in the process of physiological blood vessel regression during wound healing. The investigation of PEDF s function in healing dermal wounds is a major part of this dissertation and will be discussed in Chapter 4. Other molecules likely involved in some way in the regulation of wound angiogenesis during wound repair are angiostatin and vasostatin [66]. Vasostatin is a fragment of calreticulin, a multifunctional molecule in itself, and has been found to be anti-angiogenic when exogenously applied to excisional dermal wounds [79], although it is unknown how its activity may integrate into the overall process of vessel regression. Angiostatin is a potent anti-angiogenic fragment of an ubiquitous protein present in the wound microenvironment, plasminogen (which functions to dissolve the fibrin clot) [80], but its function has not yet been investigated in a model of physiological angiogenesis. The discussion of angiostatin and vasostatin is valuable because it brings to light a potentially important role for the proteolytic processing of parent proteins into functional

33 23 fragments in the down-regulation of angiogenesis [80]. Besides the soluble and matricellular factors, an essential class of anti-angiogenic molecules are those derived from ECM components, generated when specific matrix proteases cleave large ECM proteins into bioactive peptides [33]. Indeed, proteolytic processing of structural proteins is a characteristic component of the ECM remodeling that occurs during wound resolution and in coordination with vessel regression Remodeling of the Extra-cellular Matrix: Dynamic Reciprocity in Vessel Regression The ECM microenvironment of blood vessels includes the immediate basement membrane, composed primarily of Collagen Type IV and laminin, and the surrounding connective tissue, composed of fibrillar collagens (Type I and III) and other ECM-associated molecules, like fibronectin [21]. The synthesis and remodeling of the various components of the ECM have been observed to drastically affect EC biology [22], and, reciprocally, the resident ECs themselves participate in the regulation of their immediate ECM microenvironment [13]. During vessel growth, the EC basement membrane as well as the provisional matrix is dissolved ahead of the invading neovessel sprout. During vessel regression, the ECM as a whole undergoes remodeling from an angio-permissive scaffold into a more mature phenotype, which provides crucial anti-angiogenic cues, both molecular and biomechanical, that guide the vessels into a spatially distributed and functional network [11, 20] (Fig. 1.2A). The primary cell type responsible for the maintenance and remodeling of the ECM is the fibroblast (Fig. 1.1A). Multiple studies have found that fibroblasts may be important regulators of angiogenesis through the production and activation of specific soluble and matrix components [24, 26, 81]. Gene expression profiles differ dramatically in proliferating/migrating versus quiescent fibroblasts, with activated fibroblasts shown to produce pro-angiogenic matrix proteases and soluble proteins, such as VEGF; in contrast, quiescent fibroblasts exhibit an

34 24 elevated expression of multiple ECM precursors of potent anti-angiogenic peptides (to be discussed later), as well as anti-angiogenic matricellular proteins TSP2 and PEDF [26]. Coculture studies of ECs and fibroblasts have shown a temporal regulation of EC function by fibroblasts both directly by expression of specific angiogenic factors like VEGF, and indirectly by altering the mechanical microenvironment via matrix disruption, deposition, and remodeling [24]. Finally, different fibroblast subpopulations may play a role in determining the fibroblast s proversus anti-angiogenic functions. Fibroblasts derived from the papillary dermis and co-cultured with ECs are angio-permissive, stimulating robust vessel growth, whereas reticular fibroblasts from the deeper tissue are angio-restrictive, presumably as a result of non-soluble factors such as the composition of the secreted ECM [81]. It could be speculated from these in vitro studies that during wound healing, fibroblasts promote angiogenic growth in the proliferative phase by being themselves activated by the hypoxic environment. At the onset of the wound resolution phase, fibroblasts may switch to an anti-angiogenic phenotype --- due to contact inhibition and normalizing oxygen levels --- to regulate ECM remodeling, indirectly mediating vessel regression through their action upon the ECM microenvironment (Fig. 1.6, Fig. 1.8). ECM remodeling by fibroblasts involves both the systematic breakdown and deposition of new ECM components. The ECM itself has been demonstrated to be a rich source of endogenous anti-angiogenic mediators. Soluble proteases, called matrix metalloproteinases (MMPs), enzymatically cleave larger ECM components [82, 83]. Cleavage of the provisional ECM scaffold by MMPs may disrupt stable ECM contacts with EC integrins, causing a decrease in survival signaling and reduced pro-angiogenic integrin co-activation. Widespread proteolysis of various components of the ECM scaffold may also undermine the support for parts of the blood vessel network, leading to structural collapse and pruning [83] (Fig. 1.6). Importantly, cleavage of certain ECM components by specific MMPs may generate bioactive peptide fragments which have been shown to be anti-angiogenic [33, 66, 84]. One of

35 25 the better studied anti-angiogenic ECM-derived peptides is endostatin, a fragment of EC basement membrane component Collagen Type XVIII [66]. Endostatin is readily cleaved from its parent matrix protein by multiple MMPs and is a potent inhibitor of EC proliferation and migration, also inducing EC apoptosis. Early studies of endostatin s angiogenic function during wound healing have produced conflicting results, and a recent study using Collagen Type XVIIInull and endostatin over-expressing mice found that endostatin s role in physiological wound angiogenesis may be indirect, causing changes in vessel quality rather than quantity [85]. Another component of the basement membrane, Collagen Type IV, can be cleaved into several potent anti-angiogenic peptides, including arrestin, canstatin, and tumstatin [33, 84]. Other ECM components that may be cleaved to anti-angiogenic peptides include fibronectin (yielding anastellin), Collagen Type VIII (yielding vastatin), and heparan sulfate proteoglycans (yielding endorepellin) [33]. Although the roles of these anti-angiogenic ECM-derived peptides in regulating physiological angiogenesis has yet to be investigated, the sheer number and variety suggests that their release into the neovessel microenvironment during the remodeling phase of wound healing may be an important mechanism for the spatial control of vessel regression (Fig. 1.6). Chapter 2 will briefly explore expression patterns of these ECM-derived peptides during dermal wound repair. Whether EC apoptosis is the direct cause of vessel pruning or happens as a result of EC dissociation from the existing network is still a matter of debate [37]. For instance, CXCL10 has been shown to induce EC apoptosis only after its disruption of a key integrin that keeps the ECs anchored to the supporting ECM [69]. Similar integrin-disrupting mechanisms have been observed in vitro for other endogenous anti-angiogenic mediators, particularly ECM-derived peptides such as endostatin, tumstatin, canstatin, and arresten [66]. Importantly, it has been shown that interruption of stable cell-ecm connections leads to EC apoptosis [27] (Fig. 1.6). These data suggest a critical role for integrins in the resolution of angiogenesis.

36 26 During wound healing, vessel regression occurs concurrently with ECM remodeling (Fig. 1.2A). The process of ECM remodeling involves the cleavage of multiple ECM components by MMPs, leading to the simultaneous disruption of EC-ECM contacts (freeing integrins of stable scaffold-associated ligands) and the generation of ECM-derived anti-angiogenic peptides. This process may contribute to vessel regression in a variety of ways (Fig. 1.6): 1. Systematic disruption of EC connections to the matrix scaffold may cause structural collapse of the affected tubules via imbalance of transmitted forces, leading to dissociation from the neovessel network [83]. 2. The density and stiffness of the remodeled, more mature ECM has been shown to alter EC behavior toward an anti-angiogenic phenotype via biomechanical pathways [20]. 3. The cleaved ECM-derived soluble anti-angiogenic peptides may competitively bind to the free integrin receptors on ECs, initiating anti-angiogenic signaling pathways [66]. 4. Since amplification of pro-angiogenic and pro-survival signaling pathways are often dependent on integrin co-activation [86], loss of stable integrin-ecm contacts during remodeling may promote EC apoptosis. 5. Activated ECs may find themselves in an ECM microenvironment that has already remodeled to a more mature phenotype, making their integrin repertoire inappropriate, and leading to insufficient integrin co-stimulation to prolong survival signals [27]. These concepts, while logical and supported by indirect evidence, need to be investigated in the context of physiological vessel regression Scar Formation: Does Regression Impair Tissue Remodeling? The seemingly intimate relationship between the processes of ECM remodeling and vessel regression may help to explain the varieties of scarring outcomes between tissues. Oral mucosal and fetal dermal wounds are known to heal in a scarless fashion when compared to

37 27 similar size wounds in the adult skin [87]. One remarkable common feature is the much reduced angiogenic response of both oral mucosal [43] and fetal wounds [88] during the proliferative phase of healing as compared to dermal wounds in adult animals (Fig. 1.7). Although the association between increased angiogenesis and scarring is notable, the potential mechanisms underlying this relationship are unknown. The reduced angiogenesis in non-scar forming wounds translates to a reduced necessity for vessel regression in these wounds. It is tempting to speculate that the mechanisms of vessel regression explored here contribute to less efficient ECM remodeling, leading to scarring rather than regeneration in adult dermal wounds. These largely unexplored questions underscore the implications for greater understanding of the resolving processes of physiological angiogenesis, which will likely have significant clinical applications to the field of tissue regeneration. In Chapter 5, we will explore systems-level regulatory differences in angiogenesis and ECM remodeling between oral and skin wounds Endothelial Cell Apoptosis: Mechanisms at the Moment of Death The method by which ECs are eliminated during vessel regression is programmed cell death, or apoptosis [53]. The characteristic morphological changes in cells undergoing apoptosis include cell shrinkage, chromatin condensation, membrane blebbing, and nuclear fragmentation [58]. The proteases that perform these destructive actions are called caspases, which are present in cells in their inactive pro-enzyme form and must be activated in large enough numbers to promote apoptosis. Caspase activation results from signaling pathways that are induced by specific intra- or extra-cellular signals. Induction of apoptosis in ECs is mainly via the extrinsic pathway, which involves the expression of a death receptor, Fas, and the binding to it by Fas ligand (FasL). As was discussed, activation by VEGF in hypoxic conditions has been found to increase the expression of Fas on ECs, thus sensitizing them to both mitogenic and apoptotic stimuli [59] (Fig. 1.6, Fig. 1.8). The dynamic ECM microenvironment is also an

38 Figure 1.8: Diagram summarizing the major converging pathways that direct endothelial cell phenotypes of quiescence, priming/activation, sprouting, maturation, and apoptosis. The default EC state is that of quiescence, which is maintained by several mechanisms, including pericyte coverage and normoxia. Hypoxia triggers changes in the transcriptional profile of the EC, priming it for stimulation by hypoxia-induced extra-cellular VEGF. VEGF activates the EC, which enters into a state of plasticity, becoming sensitive to both pro-angiogenic (via VEGFR2) and anti-angiogenic (via Fas receptor) stimuli. The dynamic balance between these two sets of stimuli in the immediate microenvironment of the activated EC determines its fate: sprouting, maturation, or apoptosis. Sprouting results from hypoxia-driven pro-angiogenic signaling pathways like the mitogen-activated protein kinase (MAPK) pathway, which induces EC proliferation and migration. Apoptosis results from anti-angiogenic signaling pathways like those induced by soluble and ECM-derived molecules and by the remodeling of the extracellular matrix (ECM) itself leading to reduced integrin stability. The major determining factor for whether an EC undergoes apoptosis or maturation seems to be vessel perfusion, and maturation pathways aim to return the EC to a quiescent state. Previously published in [1] as Figure 3. 28

39 29 important regulator of EC fate, as it can promote either survival or death depending on the stability of the EC integrin-ecm connections [27]. While many of the endogenous antiangiogenic molecules discussed have been found to induce caspase-mediated EC apoptosis, they go about it in remarkably different ways, and what actually initiates apoptosis in ECs comprising the neovessel network during healing is likely very context-specific (Fig. 1.6, Fig. 1.8).The apoptotic signaling mechanisms that are initiated by the various anti-angiogenic stimuli are complex and diverse, and in-depth discussion of this topic is beyond the scope of this review. Here, we simply wish to give some examples to illustrate the incredible redundancy involved. The proposed apoptotic pathway initiated by TSPs involves binding to the EC receptor CD36, propagating p38 Kinase and Jun Kinase (JNK) signals, leading to FasL production [89]. PEDF has been shown to promote the production of FasL via various signaling mechanisms in ECs, including PPAR-ɣ, NF-κB, JNK and/or p53, depending on the context [59, 90-92]. The proposed mechanism by which CXCL10 causes apoptosis is different: binding to CXCR3 activates µ-calpain, which cleaves a crucial integrin, leading to EC dissociation from the vessel network and apoptosis induced by detachment from the ECM [69]. Similar integrin-dependent apoptotic pathways have been implicated in cases where ECs lose stable contacts with the ECM during remodeling by MMPs [27]. Furthermore, several of the ECM-derived antiangiogenic peptides function by binding and blocking key EC integrins, presumably initiating apoptosis in a related fashion [66]. It is also the case that loss of integrin-mediated survival signaling, both on the abluminal (via the ECM) and luminal (via shear stress caused by flow) sides of the EC, further sensitizes the cell to extrinsic apoptotic signals [20, 51, 58]. Recent explorations into EC apoptosis have found that besides the canonical signaling pathways, unique EC-specific signaling proteins may be involved in vessel regression [36]. One such protein, called FGD5 (FYVE, Rho guanine exchange factor, and pleckstrin homology

40 30 domain containing 5), is expressed in ECs in a variety of tissues and promotes the activation of GTPase Cdc42 [93]. FGD5 activation of Cdc42 mediates vessel regression by two mechanisms: downregulation of pro-angiogenic VEGFR2 coincident with the upregulation of VEGFR1 and subsequent sequestration of VEGF, and by initiation of Hey-1/p53-dependent EC apoptosis. However, this mechanism appears to be context-specific, as FGD5 has previously been shown to promote tubule formation in the presence of VEGF [36]. The actual response likely depends on a convergence of upstream signaling cascades initiated by anti-angiogenic stimuli leading up to FGD5, which have yet to be investigated. While the mechanistic data is robust, the study utilized two in vivo models of pathological angiogenesis to explore the anti-angiogenic function of FGD5, and it remains to be seen whether FGD5 is involved in physiological vessel regression Maturation: Bolstering the Essential Vessels During the remodeling phase of healing, regression and maturation of neovessels occur simultaneously (Fig. 1.2A). While the vast majority of the wound vasculature undergoes pruning by the mechanisms just discussed (Fig. 1.3D), a minority gets chosen for the opposite fate: survival and maturation (Fig. 1.3E, Fig. 1.7). What are the determinants of vessel maturation in the complex neovessel network of the remodeling granulation tissue? Vessel perfusion likely plays a critical role in the induction of vessel maturation [16, 20, 51]. Tubules which reconnect with a functional network re-establish a stable, laminar flow, leading to shear stress-mediated survival signaling (Fig. 1.3D-E, Fig. 1.4). This signaling may give these vessels a competitive advantage over their non-perfused neighbors (Fig. 1.8). Furthermore, the microenvironment of the perfused vessel will over time be less hypoxic due to more efficient delivery of oxygenated blood, leading to the down-regulation of pro-angiogenic stimuli. In these ECs, increased survival signaling, coupled with decreased mitogenic signaling,

41 31 may promote their maturation into vessels that will make up the permanent blood vessel network of the healed tissue, a network that architecturally and functionally resembles that of the pre-wounded state (Fig. 1.3E, Fig. 1.7). Mechanisms of vessel maturation have been explored in some detail (reviewed in [94-96]). The overarching processes involve the recruitment and attachment of network-stabilizing mural cells, such as pericytes [18] and vascular smooth muscle cells [97], as well as the bolstering of EC-EC connections through VE-Cadherins [17] (Fig. 1.4). In a general sense, maturation aims at a return to a state of EC quiescence and vascular homeostasis. The recruitment and differentiation of mural cells is largely mediated by Platelet-Derived Growth Factor (PDGF), which is produced by ECs undergoing maturation and binds to the PDGF receptor on nearby pericytes, promoting their migration toward and adhesion to vessel walls [94]. The presence of pericytes stimulates the production by ECs of Transforming Growth Factor Beta (TGF-β), a multifunctional protein that attenuates EC response to pro-angiogenic stimuli and stimulates mural cell differentiation and activity [55]. The sphingolipid metabolite, Sphingosine-1-phosphate (S1P), has been found to be important during vessel maturation. ECderived S1P strengthens EC-pericyte interactions through the assembly of N-Cadherin between the EC and pericyte [16], and S1P promotes Akt-associated survival signaling as well as Rac1- mediated assembly of VE-Cadherin in ECs [55]. Finally, Angiopoietin-1 produced by mural cells interacts with ECs through Tie-2 receptors and promotes vessel barrier integrity by attenuating VEGF-induced vasopermeability, stabilizing VE-Cadherin, and promoting pericyte adhesion [98, 99]. These and other mechanisms of vessel maturation promote the survival of a perfused and functional blood vessel network in the highly anti-angiogenic environment of the resolving wound. Whilst protecting the ECs from pro- and anti-angiogenic stimuli, these mechanisms bolster EC-EC connections, reducing vascular leakage. Throughout wound resolution, while the

42 32 majority of non-perfused and blind-ended neovessels undergo regression and are eliminated (Fig. 1.3D, Fig. 1.6), the processes of vessel maturation promote the stabilization of the remaining vessel network, restoring original architecture and function and thus helping to return the tissue to homeostasis (Fig. 1.3E, Fig. 1.7) Tumor Angiogenesis: The Case of the Ever-Sprouting Blood Vessel The exuberant angiogenesis that occurs in dermal wounds during the proliferative phase and results in the characteristic granulation tissue formation has been compared to the angiogenesis that occurs in solid tumors [38, 50, 65, 100, 101] (Fig. 1.1B, 1.7). In these tumors, there is a seemingly endless cycle of blood vessel growth and regression, with the tissue experiencing constant hypoxia and never quite being able to achieve homeostasis. That is, the tumor could be described as a wound that gets stuck in the proliferative phase, never fully transitioning into resolution and never quite adopting an anti-angiogenic phenotype as occurs in normal wounds. Based on our discussion of the microenvironmental and cellular changes that normally take place during the later phases of healing, several hypotheses can be made regarding the mechanisms that may be dysregulated in tumors as compared to wounds. Attempts at treatments of tumors using anti-angiogenic therapy have yielded mixed results. Despite decades of rigorous research, no magic bullet molecule has been found that effectively stops pathologic angiogenesis consistently and is able to starve the tumor in animal models and in the clinic (Fig. 1.1B). The biology inherent in the process of angiogenesis is extremely complex, with various cell types, countless redundant molecules and pathways, and cross-talk with a highly dynamic ECM microenvironment. A comprehensive understanding of not only the individual parts but how they come together may be necessary before we are able to predictably steer the angiogenic process in a desired direction. Importantly, this must include rigorous investigation into the mechanisms of vessel regression. By understanding how the

43 33 organism deals with exuberant blood vessel growth in health, we may be able to recruit these natural resolving processes in cases of pathology. Indeed, the field of cancer therapy appears to be approaching a similar conclusion. Recent observations have found that by normalizing tumor vasculature, via the induction of maturation mechanisms just discussed, the tumors respond better to therapy [ ]. The idea is that the leaky and non-perfused vasculature of tumors inefficiently delivers chemotherapeutic agents to the cancer cells (Fig. 1.7). Furthermore, the implication of the dynamic reciprocity of the vasculature and its microenvironment is that active promotion of a more normal phenotype in vessels may induce normalization of the tumor tissue itself, halting the viscous cycle of hypoxia-driven growth. In light of these exciting concepts, it becomes clear that the field of cancer therapy may have much to learn from wound healing, particularly from more rigorous exploration of the mechanisms of wound resolution, and the tissue remodeling and vessel regression processes that naturally occur during healing CRITICAL KNOWLEDGE GAPS AND MOTIVATIONS FOR THESIS PROJECT Wound healing is a robust model for studying the physiological control of angiogenesis in vivo [10]. During experimentally-induced wound repair, the vascular network expands and regresses in a well-characterized and remarkably reproducible temporal pattern. Angiogenesis during wound repair is induced, controlled, and ultimately resolved to vascular and tissue homeostasis (Fig. 1.2). Information inferred from the available literature suggests that spatiotemporal control of blood vessel growth and regression occurs through competing signals acting upon ECs, signals derived from the highly dynamic wound microenvironment. These pro- and anti-angiogenic stimuli are both molecular and biomechanical in nature, and the convergence of the induced signaling pathways determines EC fate: sprouting, regression, or maturation (Fig. 1.8).

44 34 Blood vessel regression likely occurs via various mechanisms (Fig. 1.6, Fig. 1.8). Throughout the process there is significant cross-talk between the different cells and their microenvironment (Fig. 1.1A). Blood vessels that are not perfused and experience low shear stress within the lumen are pruned. As the oxygen tension within the healing tissue normalizes, decreased hypoxia leads to a decrease in the production of pro-angiogenic factors. At the same time, endothelial cells which have been continually stimulated by pro-angiogenic factors begin to produce and activate inhibitors to pro-angiogenic signaling pathways in a negative feedback mechanism. The mounting intracellular inhibition makes these cells less sensitive to further proangiogenic stimulation while at the same time making them more sensitive to progressively accumulating anti-angiogenic stimuli, thus causing an anti-angiogenic switch. Many direct antiangiogenic mediators are produced in the microenvironment of the remodeling tissue. Soluble molecules, such as cytokines, as well as matricellular proteins, act directly upon ECs through specific receptors and downstream pathways. Often these then activate pro-apoptotic signaling pathways leading to systematic EC death. Another class of anti-angiogenic factors includes bioactive fragments of ECM components. These fragments are released into the microenvironment through the action of various matrix proteases which cleave ECM proteins at specific sites, and directly antagonize nearby ECs. In this process of matrix remodeling, support for blood vessels may become undermined, leading to their collapse and regression. Meanwhile, as the matrix remodels from a provisional to a more mature composition, the biomechanical properties of the matrix favor the anti-angiogenic phenotype. The net result is fine spatio-temporal control of blood vessel pruning occurring simultaneously and in cooperation with matrix remodeling, leading to an optimally distributed vessel network that adequately supplies the healing tissue. This summary of the mechanisms of vessel regression is logical and based on mountains of experimental data. This big-picture understanding was inferred from the synthesis

45 35 of the available literature. Unfortunately, this understanding cannot be easily tested due to the limitations of reductionistic methods. At the same time, the focus of reductionistic investigations have been heavily skewed toward pro-angiogenesis, so while the factors and mechanisms controlling EC sprouting and maturation are relatively well-described in the context of wound healing, mechanisms of regression are mostly unknown. The gaps in knowledge are thus two-fold, comprising two separate but mutuallysupporting themes of scientific study: reductionistic and systems-level. With respect to the first theme, there is a real gap in the knowledge of the individual parts and gears of vessel regression, the factors and mechanisms responsible for the phenotype of EC apoptosis. These can be determined through traditional methods of reductionistic science, via systematic identification of additional anti-angiogenic factors, their addition or removal from the system, and the measurement of the effects of this perturbation which may reveal the potential mechanisms through which they work. Chapter 2 of this thesis describes an approach for identifying likely candidate factors for further experimental exploration. Chapters 3 and 4 describe projects in the reductionistic theme that characterize two novel anti-angiogenic factors, Sprouty-2 and PEDF, in the context of dermal wound repair. These projects aim to fill the gap of missing parts and gears in vessel regression. With respect to the second theme of scientific inquiry, systems-level approaches to wound healing are extremely scarce. As the introduction made clear, the sheer complexity of tissue repair, with the myriad of cells and molecules interacting within a dynamic matrix, makes it likely that the phenotypes of angiogenesis and vessel regression are emergent phenomena that take place in the finely orchestrated environment of whole healing tissue. With this apparent synergism, the whole is greater than the sum of its parts, and a purely reductionistic approach is unlikely to ever result in a true understanding of this system. To tackle this issue, Chapter 5 of this thesis introduces the concept of systems biology and utilizes a novel tool called network

46 36 analysis to explore the processes of angiogenesis and matrix remodeling from a top-down perspective. This project aims to fill the gap of the missing machine that is made up of all the parts and gears that reductionistic science has discovered. In the end, it is clear that both themes of scientific inquiry are necessary. Without the parts, there can be no machine ; without the machine, the parts make no sense. With blood vessel regression, both are missing. These critical gaps in the knowledge and understanding of such an important physiological process severely impair therapeutic options for all the pathologies that feature angiogenic phenotypes (Fig. 1.1B). Progress in the attainment of more knowledge and understanding of this process which naturally restores tissue homeostasis and health will provide novel insights to the pathogenesis and treatment of these diseases.

47 2. IDENTIFICATION OF CANDIDATE ANTI-ANGIOGENIC FACTORS USING HIGH-THROUGHPUT TECHNOLOGY 2.1. INTRODUCTION TO THE WOUND TRANSCRIPTOME Many anti-angiogenic factors have been identified in studies that focus on controlling the pathological angiogenesis that occurs in tumors and in ocular neovascularization, but no such analysis has yet been completed for the healing wound. Recently, our laboratory completed an extensive temporal microarray analysis of dermal and tongue wounds in mice [105]. Sizematched 1-mm excisional wounds were made on the dorsum and tongue of mice, and the healing wounds were harvested at 6 hours, 12 hours, 1, 3, 5, 7, 10 days post-injury. These timepoints correspond to specific phases of healing identical to those described for 3-mm wounds in Chapter 1 (Fig. 1.2), although at a shorter time-scale. Figure 2.1 describes the approximate phenotypes in terms of angiogenesis and ECM remodeling in the 1-mm excisional oral and skin wounds. Figure 2.1: Diagram describing the time-course of healing 1-mm excisional oral and skin wounds. Phases of healing are labeled on the bottom, whereas the gradients of angiogenic and ECM remodeling phenotypes are shown with respect to the skin (orange) and oral mucosal (blue) wounds, which heal faster. 37

48 MINING THE WOUND TRANSCRIPTOME FOR ANTI-ANGIOGENIC FACTORS Sprouty-2 and PEDF as prime candidates and foci for thesis We first searched the literature for known anti-angiogenic agents that have been studied in other systems, from in vitro to in vivo studies in other organisms. Most of these agents were described in Chapter 1. We then used the wound transcriptome to search for potential genes that may be involved in vessel regression during wound healing. First, we looked for genes which have differential expression during the time-course of dermal wound repair, and found several candidates that have well-characterized anti-angiogenic functions in other systems. In this way, we identified Sprouty-2 as a differentially-expressed gene in the early-proliferative phase of repair (Fig. 2.2). Sprouty-2 was briefly introduced in Chapter 1 as an intra-cellular factor that may be involved in the anti-angiogenic switch in wounds. We decided to pursue this candidate for further analysis. Chapter 3 of this thesis is entirely dedicated to fully describing the background and experimental methods for evaluating the role of Sprouty-2 during dermal wound healing. We further screened the list of candidate anti-angiogenic agents against genes in the transcriptome that are up-regulated during the last 2-3 time-points, when vascular regression is known to occur in this wound model (Fig 2.1). In this way, we identified PEDF and Vasostatin-I (Fig. 2.2). PEDF was briefly introduced in Chapter 1 as a matri-cellular factor that may be involved in mediating vessel regression in wounds. We decided to pursue this candidate for further analysis. Chapter 4 of this thesis is entirely dedicated to fully describing the background and experimental methods for evaluating the role of PEDF during dermal wound healing.

49 39 Figure 2.2: Gene expression data of potential anti-angiogenic agents mined from the wound transcriptome. All genes demonstrate statistically significant changes in expression (p < 0.05 by one-way ANOVA; n=3). Expression is relative to house-keeping genes over the timecourse of 1-mm excisional wound healing in the skin. In these graphs, 0h refers to unwounded, normal skin Candidate for future exploration: Vasostatin-I The rest of this chapter will provide information on the other potential anti-angiogenic agents which were identified to be differentially up-regulated during wound resolution. The purpose is to provide further experimental insight into the complexity of the mechanisms of vessel regression, the relationship between vessel regression and ECM remodeling, and to set the stage for Chapter 5. This final chapter will use novel systems biology methods in conjunction with the wound transcriptome data in an attempt to unravel the complexity of angiogenesis and ECM remodeling during oral mucosal and skin wound healing.

50 40 An angiogenic inhibitor identified in this screen which has not been investigated in the context of wound healing is vasostatin-i, an N-terminal peptide fragment of chromogranin-a. The parent protein Chromogranin-A is released by diffuse neuroendocrine cells and is a component of blood serum [106]. Vasostatin-I has been known to promote relaxation of constricted blood vessels as well as to modulate adhesive interactions of fibroblasts and smooth muscle cells ECM proteins [106]. Importantly, a recent study found that vasostatin-i has significant inhibitory effects on VEGF-induced endothelial cell proliferation and migration in vitro and Matrigel vessel formation in vivo [107]. Furthermore, it has been shown that plasmin, an ubiquitous enzyme in the later phases of wound healing, is capable of cleaving chromogranin-a to produce vasostatin-i [108]. These published results, combined with novel data which shows an evident upregulation of chromogranin-a and plasminogen activator (converts plasminogen to the active form plasmin) during the later phases of healing (Fig. 2.2), suggest that vasostatin-i could play a significant role in vascular regression during wound healing and should be a focus of future scientific inquiry Extra-cellular matrix-derived peptides as candidates for future exploration As described in detail in Chapter 1, the mural cells, particularly fibroblasts (FB), are critical to the establishment, maintenance, and remodeling of the wound microenvironment. The ECM is highly dynamic during the different phases of wound repair; the ECM, in turn, provides biochemical and mechanical cues to the ECs of the dynamic wound vascular network. The provisional ECM during the early phases of wound repair seems to be highly conducive to proangiogenic EC behavior, while the remodeling ECM during the later phases of repair seems to have potent anti-angiogenic properties. The proliferative state of the resident FBs has been implicated in this temporal control of the wound ECM, in that gene expression profiles differ dramatically in proliferating or migrating versus quiescent FB. In particular, quiescent FB exhibit

51 41 high expression of multiple ECM precursors of potent anti-angiogenic peptides and their respective protease activators in vitro [26]. The expression profile found in quiescent fibroblasts matches our wound transcriptome data, in that there is a strikingly similar up-regulation of certain genes during the resolving, post-proliferative phases of healing, when ECM maturation and vascular regression are known to occur (Fig. 1.2, Fig. 2.1). Many of these genes are known anti-angiogenic agents and are actually bio-active fragments of various ECM proteins. A major component of the basement membrane, Collagen Type IV, can be cleaved into several potent anti-angiogenic peptides, including arrestin, canstatin, and tumstatin [33, 84]. Other ECM components that may be cleaved to anti-angiogenic peptides include fibronectin (yielding anastellin), Collagen Type VIII (yielding vastatin), and heparan sulfate proteoglycans (yielding endorepellin) [33]. Although the roles of these anti-angiogenic ECM-derived peptides in regulating physiological angiogenesis has yet to be investigated, the sheer number and variety suggests that their release into the neovessel microenvironment during the remodeling phase of wound healing may be an important mechanism for the control of vessel regression. As described in Chapter 1, it seems likely that in addition to the documented direct roles of intra- (e.g. Sprouty) and extra-cellular (e.g. TSP, CXCL10, PEDF) factors in vessel involution, the remodeling of the ECM itself during wound resolution releases potent anti-angiogenic peptides into the wound microenvironment from ECM-derived precursors produced by quiescent FB which are in part responsible for coordinating the return to vascular homeostasis. The ECMderived anti-angiogenic peptides which were identified to be up-regulated during wound resolution are (Fig. 2.3): anastellin (derived from fibronectin 1) [109] arrestin (derived from collagen type IV alpha 1) [110] vastatin (derived from collagen type VIII alpha 1) [111] endorepellin (derived from heparan sulfate proteoglycan 2) [112]

52 42 Figure 2.3: Gene expression data of potential ECM-derived anti-angiogenic agents mined from the wound transcriptome. All genes demonstrate statistically significant changes in expression (p < 0.05 by one-way ANOVA; n=3). Expression is relative to house-keeping genes over the time-course of 1-mm excisional wound healing in the skin. In these graphs, 0h refers to unwounded, normal skin. Several MMPs which are known to cleave the parent ECM proteins into these peptides are also up-regulated during wound resolution (data not shown). With this initial screen of the wound transcriptome, we were able to identify a number of potentially novel anti-angiogenic factors that may be involved in the process of physiological vascular regression during wound healing. Two of these, Sprouty-2 and PEDF, are described and evaluated in Chapter 3 and 4, respectively. Chapter 5 integrates the wound transcriptome data to evaluate large-scale temporal changes in gene expression in the context of biological networks that model the complex processes of angiogenesis and ECM remodeling.

53 3. SPROUTY AS A PRIMER FOR VESSEL REGRESSION IN WOUNDS Parts of this chapter have been previously published as [2]: Wietecha MS, Chen L, Ranzer MJ, Anderson K, Ying C, Patel TB, DiPietro LA. Sprouty2 downregulates angiogenesis during mouse skin wound healing. American Journal of Physiology: Heart and Circulatory Physiology Feb; 300(2): H PubMed ID: doi: /ajpheart See Appendix for copyright license information INTRODUCTION TO SPROUTY PROJECT Angiogenesis, or the growth of new blood vessels by way of sprouting from pre-existing vasculature, is a tightly regulated and complex biological process involving endothelial cell (EC) and vascular smooth muscle cell (VSMC) proliferation, differentiation, migration, and organization into a branched tubular network [5, 113]. Re-establishment of a viable vascular network following trauma is one of the most important components of successful wound repair subsequent to hemostasis and inflammation [113]. In the healing wound, robust angiogenesis is induced following the endogenous production and release of high amounts of pro-angiogenic agents, including platelet-derived growth factor (PDGF), fibroblast growth factor-2 (FGF-2), and vascular endothelial growth factor (VEGF) [62, 113, 114]. These growth factors, via downstream up-regulation and amplification of signaling proteins in the Raf/Mek/Erk pathway in EC and VSMC, promote the dramatic cellular changes required for angiogenesis [5]. During the pro-angiogenic phase of dermal healing, vessel density in the wound more than doubles compared to vascularity levels observed in normal, uninjured skin [44]. Regression of this newly formed vasculature is a critical component of the resolving dermal wound. After reaching a maximum vessel density, blood vessels in the wound are pruned back to levels observed in normal tissue until vascular homeostasis is achieved [44]. Whereas the mechanisms that drive blood vessel formation in physiological and pathological wound models have been the topic of much investigation, inhibition of angiogenesis followed by vascular regression in the context of wound healing have not been well studied and are thus not well understood. 43

54 44 Sprouty is an intra-cellular protein which includes four mammalian homologs (Spry1-4) that are expressed in most fetal and adult tissues during and after development [115]. Besides their specific documented roles in the proper development of the murine hearing apparatus [116] and enteric neural network [117], Sprouty proteins are also ubiquitously produced in EC and VSMC [118], suggesting a widespread and varied physiological function spanning multiple cell types. In general, mammalian Spry proteins are negative feedback loop modulators of the Raf/Mek/Erk-associated signaling pathways downstream of major growth factor stimuli such as FGF, VEGF, and PDGF and have been found to regulate tubular morphogenesis [60, ]. Upon activation by growth factor binding to its compatible receptor tyrosine kinase (RTK), Spry is induced to translocate to the inner plasma membrane [ ], where it interacts with various early mitogen-activated protein kinase (MAPK) pathway-associated proteins in a celland context-specific manner [60, 119, 121]. Early in vitro studies showed that Spry inhibits FGF- and VEGF-induced EC proliferation and differentiation via the down-regulation of the Raf/Mek/Erk signaling pathway [122]. Spry is up-regulated concurrently with FGF down-regulation during EC morphogenesis in 3D collagen matrices [124]. Additionally, Spry2 has been shown to inhibit VSMC proliferation and migration [125]. Lee et al [126] demonstrated that overexpression of Spry4 inhibited EC migration in vitro as well as the branching and sprouting of blood vessels in murine embryos. Recently, Taniguchi et al [127] showed via in vivo murine knockout (KO) and knockdown analyses that Spry2 and Spry4 are negative regulators of angiogenesis. Specifically, Spry4 KO mice were more resistant to hind limb ischemia with increased blood vessel density in both muscle and skin, and in vivo shrna knockdown of Spry2/4 accelerated angiogenesis in the murine hind limb ischemia model [127]. These previous studies suggest that Sprouty may be an important endogenous antiangiogenic agent in mammals. We hypothesized that Spry2 may be involved in down-regulating

55 45 angiogenesis during the post-proliferative stage of murine dermal wound repair leading to blood vessel regression. In this study, we show that Spry2 is produced in the wound bed during the later phases of healing coincident with the onset of vascular regression in this wound model. Cell permeable, trans-activator of transmission (TAT)-tagged Spry2 inhibits EC migration and MAPK signaling. Furthermore, we show that topical application of exogenous, cell permeable Spry2 onto dermal murine wounds down-regulates angiogenesis as well as MAPK signaling MATERIALS AND METHODS Animals and Wound Models FVB-strain 6-to-8 week-old female mice (Jackson Laboratory, Bar Harbor, ME) were anesthetized using an intraperitoneal injection of 100 mg/kg ketamine and 5 mg/kg xylazine solution, and their dorsal skin was shaved and cleansed with 70% isopropyl alcohol. Six excisional full-thickness dermal wounds were made on the dorsal surface of each mouse, three on both sides of the midline, using a sterile 3-mm punch-biopsy instrument (Acu Punch, Acuderm, Ft. Lauderdale, FL). Standard aseptic techniques were followed. The excised skin was used as normal, unwounded skin control. At 1, 3, 5, 7, 10, 14, 21, and 28 days after injury, five mice per time point were sacrificed and the wounds harvested. Two samples per mouse were placed in RNAlater (Sigma, St. Louis, MO) solution for real-time reverse transcription polymerase chain reaction (RT-PCR) analysis, two were snap-frozen for protein analysis, and two were placed in optimal cutting temperature (OCT) compound (Sakura Finetechnical, Tokyo, Japan) and snap-frozen for cryo-sectioning and immunofluorescent histochemistry. To control for the contraction of murine excisional wounds, 5-mm punch-biopsy instruments were used for the collection of wound samples until the day 5 time point, and 3-mm punch-biopsy instruments were used for wound harvesting during later time points; thus, all harvested wound samples from the various time points ended up with approximately the same amount of surrounding

56 46 unwounded tissue relative to the wound area. The under-side of the skin with its unique postwounding revascularization pattern as well as the presence of a scar (characterized by the lack of hair) was used for the identification of the wound area during later time points, especially on days 21 and 28 post-wounding. Mice were housed in groups of five at 22 to 24 C on a 12:12 hour light:dark cycle; food and water were provided ad libitum. Animal protocols used in these studies were reviewed and approved by the Institutional Animal Care and Use Committee of the University of Illinois at Chicago. All animal procedures were conducted in accordance with the Guide for the Care and Use of Laboratory Animals (National Institutes of Health) Total RNA extraction and real-time PCR Wound samples stored in RNAlater solution (Sigma) were homogenized in TRIzol reagent (Invitrogen, Carlsbad, CA). Total RNA was isolated and treated with DNase I according to the Invitrogen protocol, checked for purity, and its concentration was quantified spectrophotometrically. Total RNA (1 µg) was reverse transcribed to cdna using the RETROscript RT kit (Invitrogen). GAPDH primers were published previously [128]. Spry2 primers were designed using SciTools PrimerQuest software (Integrated DNA Technologies, Coralville, IA); Spry2 primer sequences are as follows (25 nm each): forward 5'- ACTGCTCCAATGACGATGAGGACA-3', reverse 5'-CCTGGCACAATTTAAGGCAACCCT-3'. cdna samples, upstream and downstream primers for both the endogenous control gene (glyceraldehyde 3-phophate dehydrogenase (GAPDH)) and the target gene (Spry2), and SYBR Green PCR Master Mix (Applied Biosystems, Foster City, CA) were loaded onto MicroAmp 96- well PCR reaction plates (Applied Biosystems), and the amplification protocol was run using the ABI Prism 7000 and StepOnePlus Real Time PCR systems (Applied Biosystems). Raw Ct data was analyzed using the delta-delta-ct method, as described in [129]. Values generated for each sample are normalized to GAPDH at each time point, and the data is expressed as fold increase

57 47 in gene expression relative to normal, unwounded skin. Relative RNA expression was subjected to statistical analysis by one-way analysis of variance (ANOVA) and Bonferroni s post-tests using GraphPad Prism 4.0 software (GraphPad Software, San Diego, CA) Protein extraction and Western blotting Three-mm wound samples that had been kept frozen at -80 C were homogenized in 500 µl of RadioImmuno Precipitation Assay buffer with a protease inhibitor cocktail (1/100 dilution; Sigma). Samples were centrifuged at 13,000 rpm at 4 C for 15 min. The resulting supernatants were collected and the protein concentrations were quantified using a bicinchoninic acid (BCA) protein assay kit (Pierce, Rockford, IL). Protein extracts were mixed with sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) buffer and 2-mercaptoethanol (5% of total) and boiled for 3 min. Protein samples (30-55 µg per lane) were loaded into corresponding wells in a 10% Tris-glycine acrylamide gel (Bio-Rad, Hercules, CA). Separated proteins were transferred to a nitrocellulose membrane and blocked with 5% skim milk in Tris-buffered saline (TBS). Antibodies were applied to the membrane for one hour and washed with TBS-Tween 20. For the study for Spry2 presence, rabbit anti-human Spry2 (1/500; Sigma) was used. For the study for MAPK signaling proteins, rabbit anti-rat p44/42 MAPK (t-erk1/2; Cell Signaling Technology, Danvers, MA) or rabbit anti-rat phospho-p44/42 MAPK (p-erk1/2; Cell Signaling Technology) were used. Rabbit anti-human alpha-tubulin (1/3000; Abcam, Cambridge, MA) was used as a loading control for both studies. Finally, the membrane was incubated with goat antirabbit HRP (1/2000; Bio-Rad) followed by enhanced chemi-luminescence (ECL) for detection of positive bands. Imaging and relative protein quantification of the resulting membranes was obtained using ChemiDoc (Bio-Rad).

58 Endothelial cell culture: migration and MAPK activation Recombinant, cell permeable, TAT-tagged green fluorescent protein (GFP), hspry2, and dominant-negative mutant of Spry2 (Y55F) utilized in the described experiments were expressed and purified as described previously [125, 130]. Cell Migration: Mouse embryonic endothelial cells (MEEC) (gift from Dr. Cuevas, Loyola University, Chicago) were maintained in Dulbecco's modified Eagle's medium containing 3% fetal bovine serum (FBS) and incubated at 37 C in a humidified atmosphere of 5% CO 2 and 95% air. Cells were grown to confluency in a 96-well cell culture plate (25,000 cells/well) and serum-starved for 5 hr. Cells were pre-treated for 5 hr with 20 µg/ml each of TAT-GFP (control), TAT-Spry2, or TAT-Spry2 Y55F prior to introducing a wound into the monolayer with a 10 µl pipette tip. Cells were washed once with serum-free media, and the TAT-protein containing media was returned to respective wells. 10% FBS was added to induce migration of cells in one group; another group of cells was not treated with FBS. The same fields were photographed 15 hr after FBS treatment and scratch widths measured using Photoshop software (Adobe Systems, San Jose, CA). Scratches were photographed with a Nikon (Tokyo, Japan) digital camera mounted on an Olympus CKX41 culture microscope (Olympus, Center Valley, PA). Percent closure measurements were subjected to statistical analysis by one-way ANOVA and Bonferroni s post-tests using GraphPad Prism 4.0 software (GraphPad Software). MAPK activation: Human umbilical vein endothelial cells (HUVEC) (gift from Dr. Greisler, Loyola University, Chicago) were seeded onto 1 µg/cm 2 fibronectin-coated flasks and maintained in EGM-2 basal medium supplemented with 2% (v/v) FBS, 0.04% hydrocortisone, 0.4% human fibroblast growth factor (hfgf)-b, 0.1% VEGF, 0.1% R3-human Insulin-like Growth Factor (IGF-1), 0.1% ascorbic acid, 0.1% human Epidermal Growth Factor (hegf), 0.05% gentamicin, 0.05% amphotericin-b, and 0.1% heparin (EGM-2 BulletKit CC-3162; Lonza, Basel, Switzerland) at 37 C and 5% CO 2. HUVEC (10 5 cells/35mm dish) were serum-starved in

59 49 EGM-2 medium containing 0.1% FBS overnight and exposed to 10 μg/ml each of TAT-GFP (control), TAT-Spry2, or TAT-Spry2 Y55F for 1 hr at 37 C. The cells were then stimulated with VEGF (50 ng/ml; Millipore, Billerica, MA) for 10 or 30 min and lysed in SDS sample buffer. Protein concentrations were determined and equal amounts of protein were analyzed by Western blotting. Nitrocellulose membranes were immunoblotted with mouse monoclonal phospho-p44/42 MAPK antibody (p-erk1/2; Cell Signaling Technology, Danvers, MA), and after stripping, blots were re-probed with rabbit polyclonal Erk1/2 antibody (Millipore). TATtagged proteins were detected by Anti-HA-Peroxidase (Roche, Indianapolis, IN) antibodies, as described in [130] Treatment of wounds with recombinant proteins FVB mice were wounded, as described. At day 5 post-injury, 20 µl of controlled-release gel containing 2 µg of recombinant, cell permeable, TAT-tagged GFP (control), hspry2, or dominant-negative Spry2 Y55F mutant was applied onto each 3-mm wound and covered with Tegaderm dressing (3M, St. Paul, MN). 25% Pluronic gel was prepared by mixing Pluronic F- 127 powder (Sigma) with a 20mM HEPES, 150mM NaCl, 10% glycerol buffer at ph 8.0; the mixture is liquid below 15 C and gelatinous above 20 C [131, 132]. Wound samples were harvested 10 days post-injury; three were placed in OCT compound (Sakura Finetechnical) for immunofluorescent histochemistry and three were snap-frozen for Western blot analysis (as described) Immunofluorescent histochemistry Frozen samples in OCT compound were sectioned at 8 µm thickness, with five sections per slide. Wound sections were air-dried for 10 min, re-hydrated in PBS for 10 min, and fixed in pre-cooled acetone for 10 min. After washing with 1X PBS two times for 3 min, sections were first blocked using normal goat serum (1/10 dilution; Sigma) to prevent non-specific binding of secondary antibody, followed by a 3 min 1X PBS wash. For Spry2 localization studies, slides

60 50 were stained using polyclonal rabbit anti-human Sprouty2 primary antibody (1/200; Rockland Immunochemicals, Gilbertsville, PA) or rabbit IgG negative control (1/1200; Vector Laboratories, Burlingame, CA), followed by Alexa Fluor 488 goat anti-rabbit IgG fluorescent secondary antibody (1/500; Invitrogen). Specificity of the anti-spry2 antibody in wounded and unwounded murine skin tissue was assessed by pre-absorption of anti-spry2 with a recombinant peptide specific to the antibody, as previously described [125]. Serial cryo-sections of normal, unwounded skin and skin 14 days post-injury were incubated with neutralized or non-neutralized anti-spry2, or with an IgG isotype control, followed by Alexa Fluor 488 fluorescent secondary antibody. For wound vascularity studies, slides were stained using purified rat anti-mouse CD31 (PECAM-1) primary antibody (1/100; BD Pharmingen, San Diego, CA) or rat IgG negative control (1/20; BD Pharmingen), followed by Alexa Fluor 594 goat anti-rat IgG fluorescent secondary antibody (1/1000; Invitrogen). After each 45 min incubation with primary and secondary antibodies, the slides were washed for 3 min, three times, in 1X PBS (ph 7.4) and mounted in 50% glycerol in PBS. All incubations and washes were performed at room temperature. Slides were visualized and the staining quantified using Scion Image (Scion Corporation, Frederick, MD). For Spry2 localization studies, images of sections taken at 20X magnification were sorted into wound bed and margin categories based on anatomical features. Spry2-positive cells were counted by two investigators and the results were normalized to wound area and averaged. For wound vascularity studies, quantification of cluster of differentiation 31 (CD31, or PECAM-1) staining was performed as previously described [43]. Relative Spry2 and CD31 expressions were subjected to statistical analysis by one-way ANOVA and Bonferroni s post-tests using GraphPad Prism 4.0 software (GraphPad Software).

61 EXPERIMENTAL RESULTS Spry2 mrna expression and protein production in murine skin wounds during healing To determine whether Spry2 may play a significant role in the process of murine excisional skin wound healing, a time course study of Spry2 levels in the wound was performed. We followed the experimental design of previous studies which investigated healing in a wellestablished murine 3-mm excisional dermal wound model [44], and examined Spry2 mrna and protein levels at eight time points post-wounding: 1, 3, 5, 7, 10, 14, 21, and 28 days. These time points have been found to correlate with defined stages of the healing process in this wound model: epithelial closure is complete by day 5 post-wounding, while collagen content and vascularity peak at day 10 [44]. Whole wound samples were harvested at these time points and analyzed for the presence and relative expression of Spry2. Spry2 mrna expression was analyzed using real-time RT-PCR. The amount of Spry2 mrna relative to NS was normalized to the GAPDH endogenous control at each time point. Relative levels of Spry2 mrna were stable until day 5, after which they began to increase: at day 10 a 1.7-fold increase was observed, and at day 14 Spry2 mrna levels increased 3.1-fold relative to unwounded skin (Fig. 3.1A, p < 0.05). Spry2 mrna expression peaked at day 14 post-injury and then gradually decreased to the approximate level observed at day 10 during the last two time points that were investigated (Days 21 and 28, Fig. 3.1A). To determine if time-course levels of Spry2 protein are similar to the mrna expression that was observed, the protein content of Spry2 in whole wound samples was examined using Western blot analysis. Little to no Spry2 was detected in NS and during the earliest time points until day 5 post-wounding. Spry2 protein levels then increased until day 14, after which they were observed to decrease (Fig. 3.1B). This temporal pattern of Spry2 content is similar to that observed in our mrna expression study (Fig. 3.1A), indicating that Spry2 production increases dramatically during the post-proliferative stage of wound healing.

62 52 Figure 3.1: Spry2 mrna expression and protein levels in mouse skin wounds during healing. Wound samples were harvested at eight time points following 3-mm dermal excisional punch biopsy and subjected to biochemical analysis. A: Spry2 mrna transcript abundance was measured using real-time RT-PCR, normalized to GAPDH endogenous control at each time point, and compared to normal, unwounded skin (NS). Data are expressed as mean ± SE; n=5 for all time points except day 10 (n=4). *P < 0.05 at day 14 vs. day 0, when a peak in Spry2 mrna is observed, by one-way ANOVA and Bonferroni s post-test. B: Western blotting shows that Spry2 protein levels follow the general pattern of Spry2 mrna expression during wound healing. Separated bands were visualized using anti-hspry2 and anti-alpha-tubulin antibodies. Alpha-tubulin was utilized as a loading control. Representative blot is from four independent experiments with similar results. NS = normal, unwounded skin. Previously published in [2] as Figure Spry2-positive cell numbers increase in the wound bed during healing Having elucidated the general pattern of Spry2 production in whole wound samples over the time course of murine excisional skin wound healing, we sought to localize the protein in the wound. Assessment of Spry2 antibody specificity by peptide pre-absorption showed that, both in wounded as well as in unwounded skin, Spry2 was positively expressed only in the dermis of

63 53 murine skin; autofluorescence of the epidermis was observed in the neutralized antibody and isotype controls (Fig. 3.2). Immunofluorescent histochemical analysis for Spry2-positive cells was performed on 20X magnification fields from the wound bed and wound margin. Whereas Spry2-positive cell numbers in the wound margin did not change over the course of healing (Fig. 3.3A bottom), Spry2-positive cell numbers increased dramatically after day 7 in the wound bed (Fig. 3.3B). In the wound bed, a 17.1-fold increase in Spry2-expressing cells was observed on day 28 compared to day 7 post-injury (Fig. 3.3A top, p < 0.05). These results suggest that the number of cells in the dermis of the wound bed that produce Spry2 increases during the later stages of healing up to levels observed in normal, unwounded skin (Fig. 3.3A top, NS vs. day 28), whereas the number of Spry2-producing cells in the dermis of the wound margin remains unchanged during healing from levels observed in unwounded skin (Fig. 3.3A bottom, NS vs. all time points). Figure 3.2: Specificity of the anti-spry2 antibody in wounded and unwounded murine skin. Immunofluorescent histochemical staining of serial cryo-sections of normal skin (unwounded; panels A, B, C) and skin 14 days post-wounding (panels D, E, F). A, D: Staining with anti-spry2 antibody after pre-absorption of anti-spry2 with a recombinant peptide specific to the antibody. B, E: Staining with the anti-spry2 antibody without pre-absorption. C, F: Staining with an isotype control of the anti-spry2 antibody. Previously published in [2].

64 54 Figure 3.3: Spry2-positive cell numbers increase in the dermis of the wound bed during healing. Immunofluorescent histochemical analysis for Spry2 was performed on cryo-sections from whole wound samples following harvest at eight time points after 3-mm dermal excisional punch biopsy. A: Quantification of Spry2-positive cells in the wound bed (top) and wound margin (bottom). Data are expressed as mean ± SE; n=3 for all time points except days 1 and 5 (n=2) and day 10 (n=4). *P < 0.05 at day 28 vs. day 7 in the wound bed by one-way ANOVA and Bonferroni s post-test. NS = normal, unwounded skin. B: Panels show representative photomicrographs of Spry2-positive immunofluorescence in the dermis of the wound bed from days 7, 14, and 28 post-wounding; a negative control (NC) using a rabbit IgG primary antibody from a day 28 wound is shown. Scale bar = 50µm. Previously published in [2] as Figure 2. All results up to this point show a specific profile of Spry2 production during the timecourse of murine excisional dermal wound repair. During the inflammatory and early proliferative phases of wound healing when VEGF levels are at their highest and when angiogenesis is occurring at a rapid rate [44], Spry2 production and Spry2-producing cells are at very low levels. During the post-proliferative phase when vascularity is known to peak and subsequently subside [44], Spry2 production peaks while Spry2-producing cells increase in the dermis of the wound bed.

65 Endothelial cell migration and MAPK signaling are inhibited following incubation with recombinant TAT-tagged Spry2 Having found that Spry2 production changes considerably in the wound bed during the time course of dermal healing and that its expression correlates with the known angiogenic profile in this wound model [44], we sought a method of modulating Spry2 levels in the wound so as to determine whether Spry2 has a function in regulating angiogenesis. Simple exogenous application of recombinant Spry2 is unlikely to elicit a biological response because Sprouty proteins are intracellular and known to function by binding to intracellular MAPK-associated signaling proteins. We have previously developed an experimental method of modifying recombinant Spry2 with a transduction domain derived from the HIV TAT protein so as to make Spry2 cell permeable, but this work was done using HeLa cancer cells [130] and VSMC [125], not EC. Therefore, it was important to establish the efficacy of TAT-tagged Spry2 on EC migration and Raf/Mek/Erk signaling pathway activation in vitro prior to making conclusions about the effect of TAT-tagged Spry2 on angiogenesis in vivo. To assess EC migration, mouse embryonic endothelial cells (MEEC) were pre-treated with TAT-GFP, TAT-Spry2, or TAT-tagged dominant-negative mutant of Spry2 (TAT-Spry2 Y55F ) prior to introducing a wound into the monolayer. Scratch wounds made on MEEC not exposed to FBS exhibited very low closure rates and no significant differences were seen between TATprotein treated groups 15 hours after injury (Fig. 3.4A left). Cell migration was dramatically induced by FBS in MEEC pre-treated with TAT-GFP and TAT-Spry2 Y55F ; cells treated with the dominant-negative Y55F mutant of Spry2 showed a slight but not significant increase in closure relative to the GFP-treated control group (Fig. 3.4A right). In contrast, MEEC pre-treated with TAT-Spry2 exhibited significantly reduced scratch wound closures compared to the TAT-GFP treated controls (Fig. 3.4A right, p < 0.01), suggesting that TAT-tagged Spry2 successfully permeated the MEEC membranes to inhibit their migration ability in response to serum.

66 56 Figure 3.4: Endothelial cell migration and MAPK signaling are inhibited following incubation with recombinant TAT-tagged Spry2. A: TAT-Spry2 inhibits the migration of MEEC in response to serum. MEEC grown to confluency were pre-treated with 20 µg/ml of TAT-proteins for 5 hrs before making scratch wounds and monitoring cell migration in response to serum (10%) while in the presence of TAT-proteins, as described in Materials and Methods. Marked fields were photographed at time zero and 15 hrs after making the scratches, and migration of cells was calculated as percent closure of scratch wound. The mean ± SEM of 2 separate experiments is shown. *P < 0.01 for Spry2 vs. GFP in the serum group by one-way ANOVA and Bonferroni s post-test. B: TAT-Spry2 inhibits MAPK signaling of HUVEC in response to VEGF. HUVEC were serum-starved in EGM-2 medium containing 0.1% FBS overnight and pre-treated with 10 μg/ml each of TAT-GFP, TAT-Spry2 or TAT-Spry2 Y55F for 1 hr at 37 C before incubating with VEGF (50 ng/ml) for 10 min or 30 min. Total cell lysates were subjected to SDS-PAGE and immunoblotted with anti-phospho-erk1/2 (p-erk1/2), anti-total Erk1/2 (t-erk1/2), and anti-ha-peroxidase (to detect TAT-tagged proteins) conjugated antibodies. Previously published in [2] as Figure 3. To assess EC MAPK activation, human umbilical vein endothelial cells (HUVEC) were pre-treated with TAT-GFP, TAT-Spry2 or TAT-Spry2 Y55F, incubated with VEGF for 10 or 30 minutes, and analyzed via Western blotting for the presence of total Erk1/2 and phosphorylated Erk1/2 signaling proteins. As predicted, the level of total Erk1/2 was relatively unchanged across all treatment groups (Fig. 3.4B, t-erk1/2). A decrease in p- Erk1/2 was observed in TAT-Spry2

67 57 treated HUVEC relative to the TAT-GFP treated cells after 10 minutes of VEGF incubation; no p-erk1/2 was detected in HUVEC not incubated with VEGF nor in those incubated with VEGF for 30 minutes (Fig. 3.4B, p-erk1/2). However, TAT-Spry2 Y55F treated HUVEC did not exhibit an increase in p-erk1/2 content relative to the TAT-GFP treated controls (Fig. 3.4B, p-erk1/2). Positive HA peroxidase staining confirmed the presence of TAT-tagged proteins in cell cultures after treatment. These results suggest that TAT-tagged Spry2 successfully permeated the HUVEC membranes to inhibit their Raf/Mek/Erk signaling pathway activation in response to VEGF stimulation. Thus, cell permeable TAT-Spry2 is an effective treatment for the downregulation of important angiogenesis-related functions in EC Wound vascularity is decreased following exogenous application of recombinant TATtagged Spry2 To test the hypothesis that Spry2 functions to down-regulate angiogenesis during murine dermal wound repair, controlled-release gel containing recombinant, cell permeable, TATtagged GFP (control), Spry2, or dominant-negative mutant of Spry2 (Y55F) proteins was exogenously applied to wounds at day 5 after injury. Day 5 was chosen as the time point for treatment because it correlates with peak production of VEGF as well as the onset of angiogenesis in this wound model [44]. Whole wound samples were harvested at day 10 postinjury (a time point that correlates with a peak in wound vascularity in this model [44]) and analyzed via immunofluorescent histochemistry for the endothelial cell marker CD31. An abundance of CD31 correlates with blood vessel density and is thus an experimental measure of angiogenesis [44]. CD31 area in the wound was significantly decreased by 30% in Spry2- treated wounds relative to the GFP-treated controls (Fig. 3.5, p = 0.05). Wounds treated with the dominant-negative mutant of Spry2 (Y55F) exhibited a moderate (15%) increase in CD31

68 58 Figure 3.5: Wound vascularity is decreased following exogenous application of recombinant TAT-tagged Spry2. Immunofluorescent histochemical analysis for endothelial cell marker CD31 (PECAM-1) was performed on cryo-sections from whole wound samples harvested at day 10 post-injury from 3-mm dermal excisional punch biopsy and after exogenous application at day 5 post-wounding of controlled-release gel containing 2 µg of recombinant cell permeable, TAT-tagged GFP (control), Spry2, or dominant-negative mutant of Spry2 (Y55F). n=6 for GFP, n=5 for Spry2, n=6 for Spry2 Y55F. A: Panels show representative photomicrographs of CD31-positive immunofluorescence in GFP-, Spry2-, and Spry2 Y55F -treated wounds. Scale bar = 50µm. B: Quantification of CD31 immunofluorescence shows a moderate increase in vascularity in Spry2 Y55F -treated wounds and a significant decrease in vascularity in Spry2-treated wounds relative to the GFP-treated control group. Percent CD31 area per field was normalized and compared to the GFP control group to yield fold change in CD31 area; data are expressed as mean ± SE. *P = 0.05 for Spry2 vs. GFP by one-way ANOVA and Bonferroni s post-test. Previously published in [2] as Figure 4. expression in the wound compared to the GFP-treated controls (Fig. 3.5). This result is consistent with previous data that showed that the Spry2 Y55F mutant acts as a dominantnegative and reverses the inhibitory functions of endogenous Spry2 [ ]. Overall, these results indicate that Spry2 functions to down-regulate vascularity in the healing murine wound and that adding more of the protein to the wound during the proliferative phase of wound repair significantly promotes this anti-angiogenic phenotype.

69 MAPK signaling is decreased in TAT-Spry2 treated wounds, whereas TAT-Spry2 Y55F treated wounds exhibit increased MAPK signaling We next assessed whether the observed decrease in wound vascularity after Spry2 treatment is concurrent with an inhibition of the Raf/Mek/Erk signaling pathway in vivo. The levels of total Erk1/2 and phosphorylated Erk1/2 signaling proteins were determined by Western blot analyses of whole wound tissue samples harvested at day 10 post-injury following exogenous application at day 5 post-injury of recombinant, TAT-tagged GFP control, TAT- Spry2, or TAT- Spry2 Y55F. As predicted, the level of total Erk1/2 was relatively unchanged across all treatment groups (Fig. 3.6B, t-erk1/2). A decrease in phosphorylated Erk1/2 was observed in Spry2-treated wounds relative to the GFP-treated control group; in contrast, Spry2 Y55F -treated wounds exhibited an increase in phosphorylated Erk1/2 content (Fig. 3.6A, p-erk1/2). These results suggest that Spry2 may inhibit the MAPK signaling pathway during murine excisional skin wound healing simultaneously with a down-regulation in angiogenesis. 3.4 DISCUSSION This is the first study to characterize Sprouty protein production and function in the context of in vivo wound repair and wound angiogenesis. We show that Spry2 mrna and protein levels increased significantly during the post-proliferative phase of healing coincident with the onset of vascular regression in this model. Spry2 production was localized to cells in the dermis of the wound bed, where angiogenesis is known to occur during wound healing. Application of exogenous, cell permeable Spry2 to the wound during the angiogenic phase of healing significantly reduced vessel density and simultaneously reduced MAPK signaling. These results indicate that endogenous Spry2 may function to down-regulate angiogenesis in the healing murine skin wound, potentially by inhibiting the MAPK signaling pathway.

70 60 Figure 3.6: MAPK signaling is decreased in TAT-Spry2 treated wounds, whereas TAT- Spry2 Y55F treated wounds exhibit increased MAPK signaling. Levels of signaling proteins were analyzed by Western blotting performed on whole wound samples harvested at day 10 post-injury from 3-mm dermal excisional punch biopsy and after exogenous treatment with TATtagged recombinant proteins at day 5 post-wounding. Data represent results from one of four independent experiments with similar results. A and B (top): Representative Western blots showing expression of phosphorylated Erk1/2 (p-erk1/2) (A) and total Erk1/2 (t-erk1/2) (B) signaling proteins in GFP-, Spry2 Y55F -, Spry2-treated wounds. Alpha-tubulin was used as a loading control. A and B (bottom): Quantification of respective Western blots showing phospho- Erk1/2 (A) and total Erk1/2 (B) protein expression normalized to alpha-tubulin and compared to the GFP control group. Previously published in [2] as Figure 5. Sprouty is a negative feedback loop inhibitor of RTK-associated signaling pathways that converge in the Raf/Mek/Erk pathway and are known to promote cellular changes associated with angiogenesis in EC (Fig. 3.7). Three RTK ligands known to be pro-angiogenic in the context of wound repair are FGF-2, PDGF, and VEGF. Upon injury, sequestered FGF-2 is released into the wound environment and soluble FGF-2 is capable of stimulating early angiogenic events, including the proliferation and migration of EC for blood vessel sprouting, as well as inducing the production of VEGF [62, 114]. Previous studies in our laboratory show that FGF-2 production is increased progressively

71 61 Figure 3.7: Spry2 and Spry2 Y55F function in the context of endothelial cell MAPK signaling. Upon GF binding to its respective RTK, Spry2 translocates to the inner plasma membrane, gets activated via phosphorylation on the Y55, and functions by interacting with various MAPK signaling pathway-associated proteins. When RTK signaling is Ras-dependent (left side of diagram), Spry2 inhibition of this pathway is thought to occur at the level of Grb2. When RTK signaling is Ras-independent (right side of diagram), Spry2 inhibition of this pathway is thought to occur at the level of Raf1. In both cases, py55 is required for Spry2 inhibition of the Raf/Mek/Erk pathway. The dominant-negative Y55F mutant of Spry2 inhibits endogenous Spry2 action, thereby promoting the Raf/Mek/Erk pathway. EC = endothelial cell; RTK = receptor tyrosine kinase; GF = growth factor; py = phosphorylated tyrosine. Adapted from [60]. Previously published in [2] as Figure 6. following 3-mm excisional skin injury in the murine model [44]. PDGF is released by the degranulation of platelets and has been shown to be pro-angiogenic during wound healing by inducing production of VEGF and particularly by promoting the maturation of blood vessels via the recruitment of pericytes and VSMC [42]. While keratinocytes are known to produce basal levels of VEGF in uninjured skin [136, 137], there is no VEGF present in the wound site immediately after injury [44, 62]. Production of VEGF is induced by hypoxia [138] in the early

72 62 phases of wound repair, and previous studies in our laboratory show that VEGF production peaks 5 days post-injury in the 3-mm excisional skin wound model [44]. Thus, soluble FGF-2, PDGF, and VEGF are all present and active in the wound during the proliferative phase of healing. Their binding to respective RTKs may trigger not only the propagation of the pro-angiogenic Raf/Mek/Erk pathway initially but also the production and activation of Spry2 intracellularly in EC and VSMC. Indeed, growth factors have been shown to increase Spry2 expression [60, 119, 120], and elevation in Spry2 content may inhibit the ability of RTKs to further activate the Raf/Mek/Erk signaling pathway, resulting in the eventual downregulation of pro-angiogenic cellular behavior (Fig. 3.7). Previous work in our laboratory [44] has shown that blood vessel density peaks at day 10 post-injury in murine 3-mm excisional skin wounds. Starting at day 14, relative vessel density begins to decrease and regression occurs [44]. The temporal pattern of Spry2 production that we observed in this wound model correlates with this pattern of angiogenesis, in that Spry2 mrna and protein levels were seen to peak at day 14 post-wounding. Furthermore, the numbers of Spry2-producing cells in the dermis of the wound bed increased after day 7 postinjury. Importantly, the current studies showed that Spry2 functions to inhibit in vivo angiogenesis as well as MAPK signaling in the murine dermal wound. These findings are strongly supported by the observation that both angiogenesis and MAPK signaling were moderately up-regulated following the addition of the dominant-negative mutant of Spry2 to the wounds. The Spry2 Y55F mutant is thought to function by forming heterodimers with endogenous Spry2 and interfering with the binding of endogenous Spry2 to its targets in the MAPK pathways, thereby promoting the opposite phenotype of the wild type protein (Fig. 6) [ ]. Interestingly, we observed marginal stimulatory effects of Spry2 Y55F on in vitro EC migration and MAPK activation; one potential reason for this result is relatively low endogenous production of wild type Spry2 (to which the Spry2 mutant can bind and inhibit) by EC cultures in response to a

73 63 single controlled stimulus compared with the in vivo wound environment where multiple stimuli are involved. The function of Spry2 in wound healing that is described here is consistent with two other in vivo models of murine angiogenesis, namely embryo and hind limb ischemia [126, 127]. Our in vitro data utilizing TAT-tagged Spry2 is also consistent with other EC culture studies [122, 126]. The previous and current data support the conclusion that Spry2 down-regulates angiogenesis during wound repair via its inhibitory action on the FGF-2-, PDGF-, and VEGFstimulated Raf/Mek/Erk signaling pathway in EC and VSMC. The mechanisms that drive blood vessel formation in physiological and pathological wound models have been the topic of much investigation; however, the mechanisms of angiogenic down-regulation and subsequent vessel regression in the context of wound healing are not well understood. VEGF levels decrease in the resolving wound [44], and this simple loss of pro-angiogenic and vessel survival signals has been suggested to be the major cause of vascular regression [139]. Recent studies in our laboratory demonstrate that late phase exogenous administration of supra-physiological levels of the pro-angiogenic factors including VEGF, FGF-2, and PDGF cannot prevent vessel regression in the wound [64]. This data suggests that active anti-angiogenic signals which strongly counteract the pro-angiogenic stimuli may be present in the resolving wound. Thus, at any particular time point in healing, the balance between pro- and anti-angiogenic factors favors one phenotype over the other [32], with the local environment of the resolving wound critically favoring the mechanisms that lead to angiogenic down-regulation and subsequent blood vessel regression. Many anti-angiogenic factors have been identified in studies that focus on controlling the pathological angiogenesis that occurs in tumors [140] and in ocular neovascularization [141], but no such extensive analysis is yet available for the healing wound. Only thrombospondins [142] and most recently IP-10 [69] have been implicated as endogenous physiological inhibitors

74 64 of angiogenesis in the context of wound repair. In this study we identify Spry2 as another endogenous anti-angiogenic agent present and active in the resolving dermal wound. Thrombospondins 1 and 2 (TSP1/2) have both been implicated in down-regulating wound angiogenesis [72, 142]. IP-10, a CXCR3 ligand, has been shown to actively promote wound blood vessel regression even in the presence of pro-angiogenic factors [69]. TSPs and IP-10 are described further in Chapter 1. Given the documented roles for TSP and IP-10, it seems clear that Spry2 accounts for only a part of the late-phase regulatory mechanism of wound angiogenesis. TSP and IP-10 molecules are extracellular mediators of angiogenesis. TSP2 and IP-10 are both expressed during the later stages of wound repair and function to inhibit angiogenesis and even promote vascular regression via extracellular binding to EC. Although Spry2 is expressed during similar time points during wound healing, it functions to down-regulate angiogenesis in an intracellular fashion (Fig. 3.7). By inhibiting the potentiation of RTKstimulated pro-angiogenic MAPK signaling pathways, Spry2 may contribute to the shift in the balance between pro- and anti-angiogenic stimuli in EC and VSMC. This shift may make these cells more sensitive to extracellular anti-angiogenic factors like TSP and IP-10, resulting in the observed phenotypes of wound blood vessel involution and eventual vascular homeostasis. It is important to note that the current study investigated the function of one of four mammalian homologs of the Sprouty family of proteins. Given what is currently known about them [60, 119, 126, 127], it is likely that Spry1 and especially Spry4 have analogous physiological functions to those described here for Spry2. The contributions of these and other Sprouty-related proteins i.e. the Spred family [143] to the regulation of wound angiogenesis remain to be elucidated.

75 4. PIGMENT EPITHELIUM-DERIVED FACTOR (PEDF) AS A DRIVER OF VESSEL REGRESSION IN WOUNDS 4.1. INTRODUCTION TO PEDF PROJECT Angiogenesis is a complex and tightly regulated biological process involving interactions between several cell types and their microenvironment [13]. Endothelial cells (EC) form into tubular structures that are supported by mural cells, including pericytes, vascular smooth muscle cells (VSMC), and fibroblasts (FB) [1] (Fig. 1.1A). Vascular networks exist in a dynamic microenvironment, and they are influenced on the luminal side by shear stress and on the periphery by FB-regulated bio-chemical and -mechanical properties of intertwined extra-cellular matrix (ECM) components such as collagens and proteoglycans [21], and non-structural secreted molecules such as growth factors, cytokines, bioactive peptides and matrix metalloproteinases (MMP) [1]. Dysfunctional angiogenesis is an important phenotype in multiple common pathologies [28]: excessive blood vessel growth occurs in malignancy [144] and macular degeneration [141], whereas insufficient angiogenesis occurs in cardiac disease [145] and chronic wounds [146], particularly in patients with diabetes [147] (Fig. 1.1B). To more effectively treat these diseases, it is thus imperative to better understand the mechanisms that positively and negatively regulate angiogenesis [1, 148]. Wound healing is an orchestrated biological process consisting of several distinct yet overlapping phases that return the affected tissue to homeostasis [41] (Fig. 1.2A). During experimentally-induced dermal wound repair, the vascular network expands and regresses in a well-characterized and reproducible temporal pattern [1] (Fig. 1.2B). Following a burst of angiogenesis during the proliferative phase of healing, neovessels that are immature and leaky [49] are systemically targeted for removal in the remodeling phase [1], while vessels essential for tissue homeostasis are preserved and reinforced by the recruitment of mural cells [96]. Importantly, cancer has been described as an over-healing wound that is stalled in a pro- 65

76 66 angiogenic proliferative phase with no transition to vessel regression nor ECM remodeling as occur naturally in wounds [50] (Fig. 1.7). Angiogenesis during healing is regulated in a spatiotemporal manner via a dynamic balance between pro- and anti-angiogenic factors [32] (Fig. 1.2B). While pro-angiogenic mechanisms are now well-defined [5, 149], the anti-angiogenic factors that inhibit further growth of the vasculature and promote blood vessel regression are mainly unknown [1]. Since vessel regression occurs predictably in healing skin [44], and because dermal wounds are open to experimental manipulation [10], we have chosen this model to investigate the mechanisms by which anti-angiogenesis is regulated. Pigment epithelium-derived factor (PEDF), also known as early population doubling level cdna-1 (EPC-1) and encoded by the SERPINF1 gene, is a 50-kDa glycoprotein and member of the non-inhibitory serpin family [76, 150]. Although initially described as a neuro-trophic and - differentiation factor [151], PEDF is now best recognized as one of the most potent endogenous anti-angiogenic agents [152]. Importantly, PEDF has been shown to inhibit malignancy in a wide range of tumors, either by directly causing cancer cell death or by targeting the tumors abnormal vasculature [ ]. PEDF is expressed in many mammalian tissues [155] and exists at high levels in human dermis [156, 157] and in the blood [158]. Produced by quiescent fibroblasts [26, 159] and keratinocytes [160], PEDF is readily secreted into the ECM. Analyses of its amino acid sequence [150] and its crystal structure [161] have revealed distinct binding sites for ECM components collagen-1 [162, 163] and glycosaminoglycans [164], including heparin [165, 166] and hyaluronan [167]. Previous data suggests that binding of PEDF to collagen-1 promotes anti-angiogenesis directly [77] or indirectly by disrupting ECM-cell adhesion interactions which are crucial for angiogenesis [166]. Intriguingly, PEDF targets immature neovessels while sparing perfused vasculature in the cornea [152]. The aggregate evidence of PEDF s functions supports its recent classification as a multi-factorial matri-cellular

77 67 tissue homeostatic agent [76]. We hypothesized that PEDF may function to maintain vascular homeostasis in skin by regulating blood vessel regression in resolving dermal wounds. The anti-angiogenic activity of PEDF has been investigated in animal models of induced pathological angiogenesis, including ocular neovascularization and various cancers in rodents [153]. These models do not present with vessel regression nor ECM remodeling as occur physiologically in dermal wounds [1]. In this study, we evaluate PEDF s contribution to the control of angiogenesis during dermal wound repair in the mouse. Our results demonstrate that PEDF is a key endogenous anti-angiogenic factor in healing wounds that controls vessel regression, promoting a return to dermal homeostasis by influencing the vascular microenvironment MATERIALS AND METHODS Animals and Wound Model BALB/c-strain 6-to-8 week-old female mice (Harlan Laboratories, Indianapolis, IN, USA) were anesthetized using an intraperitoneal injection of 100mg/kg ketamine and 5mg/kg xylazine solution, and their dorsal skin was shaved and cleansed with 70% isopropyl alcohol. Six or four excisional full-thickness dermal wounds (depending on experiment) were made on the dorsal surface of each mouse, symmetrically on both sides of the midline, using a sterile 3-mm punchbiopsy instrument (Acu Punch, Acuderm, Ft. Lauderdale, FL, USA). Standard aseptic techniques were followed. The excised skin during wounding was used as normal, unwounded skin control. At different timepoint post-injury (depending on the experiment), animals were sacrificed and the wounds harvested. For real-time RT-PCR analyses, samples were placed in RNAlater (Sigma, St. Louis, MO, USA) and stored at -20 C. For enzyme-linked immunosorbent assay (ELISA) protein analyses, wound samples were snap-frozen and stored at -80 C. For immuno-

78 68 fluorescent histochemical analyses, wound samples were embedded in HistoPrep compound (Fisher Scientific, Waltham, MA, USA), snap-frozen, and stored at -80 C. To account for contraction of murine excisional wounds and standardize the amount of unwounded tissue surrounding each excised wound sample, 5-mm punch-biopsy instruments were used for the collection of samples until the Day 5 timepoint, and 3-mm punch-biopsy instruments were used for wound harvesting during later timepoints. To identify the wound area during later timepoints (i.e. after Day 20 post-injury), the presence of a scar, often characterized by the lack of hair or unusual pattern of hair re-growth, was observed; additionally, photographs taken of the animals throughout the course of the experiments aided in the tracking of wound locations. Mice were housed in groups of five at 22 to 24 C on a 12:12 hour light:dark cycle; food and water were provided ad libitum. Animal protocols used in these studies were reviewed and approved by the Institutional Animal Care and Use Committee of the University of Illinois at Chicago. All animal procedures were conducted in accordance with the Guide for the Care and Use of Laboratory Animals (National Institutes of Health) Culture of human keratinocytes and fibroblasts Normal Human Epidermal Keratinocytes (NHEK) (ATCC, Manassas, VA, USA) were cultured in Dermal Cell Basal Medium and Keratinocyte Growth Kit (ATCC). Normal Human Fibroblasts (NHFB) (PromoCell, Heidelberg, Germany) were cultured in DMEM with 10% FCS. Cell cultures were grown in 6-well plates to 70-80% confluency and harvested using TriZol (Invitrogen, Carlsbad, CA, USA) for total RNA extraction Purification of human recombinant PEDF Purification of human recombinant PEDF (rpedf) from the medium of stable baby hamster kidney cell transfectants overexpressing and secreting the protein was performed as described previously [168]. Purity of protein extracts was verified using spectrophotometry and by SDS-PAGE followed by Coomassie Blue staining and immunoblotting using anti-pedf

79 69 antibody (BioProducts MD, Middletown, MD, USA) and commercial recombinant PEDF as positive control (BioProducts MD). Purified rpedf was further tested for biological activity in vitro and was found to induce apoptosis in cultured human microvascular endothelial cells (data not shown). Recent published work from our laboratory used the same batch of purified rpedf to assess its effects on human keratinocytes, further validating rpedf s biological activity [160]. Amino acid alignment analysis using BLAST was performed to assess similarity between human and mouse PEDF orthologs; results are as follows: 87% identity, 94% similarity, 0% gaps in alignment, 0.0 E-value. The extremely high degree of similarity between PEDF orthologs ensures the reliability of using human PEDF in mouse studies Treatment of wounds with recombinant proteins and antibodies Treatment of healing dermal wounds was performed while animals were under anesthesia via isoflurane inhalation using a SurgiVet isoflurane vaporizer and oxygen mixing apparatus (Smiths Medical, Dublin, OH, USA). Mice were randomly selected into experimental and control groups prior to treatment. Photographs were taken of each mouse on a daily basis for tracking of wounds and measurements of wound closure. Treatment of wounds with rpedf: Purified rpedf was applied to wounds daily following injury at a dose of 2µg per wound, a time-course and concentration determined most effective in preliminary studies (data not shown). rpedf was applied topically onto the open wound prior to scab formation, and after 3 days post-injury rpedf was administered directly into each wound via intradermal injection using a short 3/10cc insulin syringe with a 30gauge 8mm needle. For topical applications, rpedf was dissolved in a controlled-release Pluronic gel (made from Pluronic F-127 (Sigma) to a consistency of 25% wt/v, as described previously [2]) to a concentration of 200µg/mL (10µL applied per wound). For intradermal injections, rpedf was dissolved in sterile PBS to a concentration of 100µg/mL (20µL injected per wound); the control

80 70 group was vehicle (Pluronic gel for topical; PBS for injection). Wound samples were harvested at Day 10 post-injury. Treatment of wounds with neutralizing antibody against PEDF (PEDF-ab): To inhibit endogenous PEDF, healing wounds were treated with a neutralizing antibody against human PEDF (PEDF-ab) (BioProducts MD) [169]. To keep immune reaction to a minimum, antibodies were applied four days apart and at a maximum dose of 0.5µg per mouse. PEDF-ab was administered directly into each wound via intradermal injection using a short 3/10cc insulin syringe with a 30gauge 8mm needle. PEDF-ab was dissolved in sterile PBS to a concentration of 6.25µg/mL (20µL injected per wound). The control group received intradermal injections of Rabbit IgG (Sigma) at the same concentration. Both antibodies did not contain sodium azide preservative. Timepoints for administration of antibodies were Days 8,12,16,20 post-wounding; wound samples were harvested at Days 16,20,24 post-injury. Five mice from each group were randomly selected for sacrifice and tissue harvest at Days 16 and Wound size measurements Throughout wound treatment experiments, photographs of all animals were taken from a set distance in standardized conditions and camera settings, with a ruler in the field of view as reference. Photographs were opened in Fiji image processing software ( the scale was calibrated to ruler, and sizes of wounds determined for each animal. Values were exported to Microsoft Excel; changes in wound area were expressed as percent of original wound area and calculated as follows: (wound area)/(original wound area)x100 All four wounds per animal were analyzed and the four values averaged to produce a unique value for each animal.

81 Total RNA extraction and real-time RT-PCR for PEDF Total RNA extracted using TriZol (Invitrogen) was treated with DNAse I and subjected to reverse transcription using a Retro-script kit (Invitrogen). Semi-quantitative mrna expression of PEDF was examined using a SYBR Green PCR mix and gene specific primers. The sequences of mouse PEDF (Serpinf1) and Gapdh primers were previously published [128, 170]. The sequences of human PEDF (SERPINF1) and GAPDH primers were previously published [160]. For in vitro comparison of NHFB versus NHEK, data were normalized to NHEK. For in vivo timecourse studies, data were normalized to normal, unwounded skin Protein extraction and ELISA for PEDF Wound samples were homogenized in 500µL RadioImmuno Precipitation Assay buffer (Sigma) with a protease inhibitor cocktail (1/100 dilution; Sigma). Samples were centrifuged at 13,000rpm at 4 C for 15min. The resulting supernatants were collected and total protein concentrations were determined using the Pierce BCA Protein Assay Kit according to the manufacturer s instructions (Thermo Fisher Scientific, Rockford, IL, USA). PEDF protein content was evaluated using the ELISAquant PEDF Sandwich ELISA Antigen Detection Kit according to the manufacturer s instructions (BioProducts MD). PEDF content was normalized to each sample s total protein concentration; normalized PEDF content values for all animals per timepoint were averaged to produce a unique value for each group Analysis of wound macrovascular density At time of wound tissue harvest, whole dorsal skin was excised from the animal, turned over on a dissection table, fatty hypodermis wiped off, and photographs taken from set distance in standardized conditions and camera settings. Images of wounds were cropped to same dimension and equally enhanced with contrast to visualize macrovessels. Image files of individual wounds were randomized to minimize bias. For all images, each visible vessel was manually traced in Microsoft PowerPoint using same brush width, and resulting traced images

82 72 exported to Fiji, where area of tracing was quantified relative to total area of image. Percent macrovascular area was calculated as follows: (traced area)/(total area of image)x100 All four wounds per animal were analyzed and the four values averaged to produce a unique value for each group Immunofluorescent histochemistry for PEDF, CD31, Vimentin, and alpha-smooth Muscle Actin (a-sma) Frozen wound samples were sectioned at 8µm thickness, with four sections/slide. All incubations and washes were performed at room temperature. PEDF localization studies: Wound sections were air-dried for 10min and re-hydrated in PBS for 10min. Sections were fixed in pre-cooled acetone for 2min followed by 80% methanol for 5min, washed 3x3min with PBS, and blocked using Normal Goat Serum (10% in PBS; Sigma) for 30min. Slides were double-stained, overnight, using 1) rabbit anti-human PEDF (BioProducts MD) and 2) rat anti-mouse CD31 (BD Pharmingen, San Diego, CA, USA) or 2) chicken anti-mouse Vimentin (abcam, Cambridge, MA, USA) primary antibodies; Rabbit IgG was used as an isotype control for PEDF (Vector Laboratories, Inc., Burlingame, CA, USA). Slides were washed 3x5min with PBS and incubated for 45min using 1) Alexa Fluor 594 goat anti-rabbit, and 2) Alexa Fluor 488 goat anti-rat or 2) FITC-conjugated goat anti-chicken fluorescent secondary antibodies (Invitrogen). Slides were washed 3x5min with PBS and mounted using 50% glycerol containing DAPI for staining of cell nuclei. Vessel density and maturity studies: Wound sections were air-dried for 10min and rehydrated in PBS for 10min. Sections were fixed in pre-cooled acetone for 10min, washed 3x3min with PBS, and blocked for 30min using Normal Goat Serum (Sigma). Slides were double-stained using 1) rat anti-mouse CD31 antibody (BD Pharmingen) and 2) mouse antimouse FITC-conjugated alpha-smooth muscle actin antibody (Sigma) for 1hr, washed 3x5min

83 73 with PBS, followed by 45min incubation with Alexa Fluor 594 goat anti-rat secondary antibody (Invitrogen). Slides were mounted using with 50% glycerol containing DAPI. Visualization and quantification: All slides were visualized under a Carl Zeiss microscope at X10, X20, X40, or X100 magnifications and multi-photon images were taken using AxioVision software (Carl Zeiss Microscopy, LLC, Thornwood, NY, USA). Image files were randomized to minimize bias and a coding system was used to identify analyzed values. Photomicrographs were evaluated and fluorescence staining was quantified using Fiji. Quantification of PEDF and CD31 co-localization: Images of sections taken using different filters at magnification of X40 were opened separately in Fiji, converted to 16-bit, and threshold was applied via the Otsu algorithm at standardized intensity level. Images from the PEDF and CD31 filters were combined to generate tri-color images showing separate colors for PEDF, CD31 and their co-localization. Pixel area of each color was quantified. Percent CD31 co-localized with PEDF was calculated as follows: (co-localized area)/[(co-localized area)+(cd31 area)]x100 At least three histological slides per animal and two images per slide were analyzed and averaged to produce a unique value for each animal. Quantification of CD31 area: For rpedf-treated wounds harvested at Day 10 post-injury, images of sections were taken at magnification of X20 centered on wound bed. For PEDF antibody-treated wounds harvested at Days 16,20,24 post-injury, images of sections were taken at magnification of X10 to better identify wound margins and ensure that analysis of CD31 area be restricted to the wound bed. For both analyses, images showing CD31 fluorescence were opened side-by-side with DAPI, the two windows synced and DAPI-stained image was used to select the area of wound bed simultaneously in the CD31-stained image. Threshold was applied to selected area via the Otsu algorithm at standardized intensity level. Pixel areas for total

84 74 wound bed and for CD31 staining within wound bed were quantified. Percent CD31 area was calculated as follows: (CD31 area in wound bed)/(total area of wound bed)x100 At least four histological slides per animal and two images per slide were analyzed and averaged to produce a unique value for each animal. Quantification of CD31 and a-sma co-localization: Images of sections taken using different filters at magnification of X20 were opened in Fiji. Analysis was identical to that described above for the co-localization of PEDF and CD31. Percent CD31 co-localized with a- SMA was calculated as follows: (co-localized area)/[(co-localized area)+(cd31 area)]x100 At least three histological slides per animal and two images per slide were analyzed and averaged to produce a unique value for each animal Picrosirius Red analysis for collagen maturity Wound samples were fixed in formalin, embedded in paraffin, sectioned, and stained using Picrosirius Red via standard methods to visualize collagen content and maturity. Slides were evaluated under a polarized microscope (Carl Zeiss Microscopy, LLC); photomicrographs were taken at X10 and X20 magnification, randomized, and analyzed in Fiji using standardized color thresholds to identify areas of mature (red-orange) and immature (green-yellow) collagen. Percent mature collagen was calculated as follows: (red-orange area)/[(red-orange area)+(green-yellow area)]x100 At least three histological slides per animal and two images per slide were analyzed and averaged to produce a unique value for each animal.

85 Statistical Analysis Average values for each animal were exported to GraphPad Prism 5.0 (GraphPad Software, San Diego, CA, USA) for statistical analyses and graph creation. Data were analyzed to yield mean±sem values for each group. For analysis of experimental group versus control, a two-tailed Student s t-test was used; for analysis of more than two groups (i.e. for multiple timepoints post-injury), one-way ANOVA followed by Bonferroni s post-tests was used. Statistical significance of values between groups was considered when p< EXPERIMENTAL RESULTS Localization and production of PEDF in mouse skin We first sought to investigate the distribution of PEDF in mouse skin. Histologic sections of unwounded skin were stained for PEDF via indirect immunofluorescence with the anti-pedf antibody or an isotype control. Positive staining for PEDF is seen in epidermis, areas of the dermis, and in hair follicles (Fig. 4.1). Previous studies have shown that PEDF is produced by fibroblasts (FBs) in various mammalian tissues [26], including in the uninjured human dermis and associated appendages [156, 157]. Analysis of sections co-stained with PEDF and vimentin, a marker for dermal FBs [171], revealed that PEDF co-localizes with vimentin throughout the tissue sections in spindle-shaped structures that resemble the morphology of dermal FB (Fig. 4.2A white triangles). PEDF present in the dermis that is not co-localized with vimentin appears as patches of positive staining adjacent to dermal FBs, suggesting a considerable extra-cellular component of PEDF in mouse dermis (Fig. 4.2A white dotted outlines).

86 76 Figure 4.1: Immunofluorescent histochemistry for PEDF in unwounded mouse skin. Positive staining using commercial anti-pedf antibody. Major anatomical structures in murine skin are labeled: epidermis, dermis, hair follicles. Triangles indicate defined areas of positive PEDF staining in the dermal layer. Dashed white line indicates boundary of epidermis and dermis. Isotype control demonstrates specificity of primary antibody. Scale bar = 50µm. To determine the primary cell type responsible for PEDF production in mammalian skin, we cultured normal human epidermal keratinocytes (NHEKs) and normal human dermal fibroblasts (NHFBs) to equal confluency and examined relative PEDF transcript levels between the two cell types. Results reveal that PEDF expression in NHFBs is over 300-fold greater than in NHEKs (Fig. 4.3A). This finding is substantiated by mined results from publicly-available microarrays that show consistently greater expression of PEDF by human dermal FBs versus matched KCs (details in discussion, Fig. 4.3B, Fig. 4.3C), as well as by fibroblast, smooth muscle, stromal cell lines versus endothelial and epithelial cell lines (details in discussion, Fig. 4.3D).

87 77 Figure 4.2: Localization of PEDF in unwounded murine skin in relation to dermal fibroblasts and endothelial cells. Murine skin was harvested, cryo-sectioned and doublestained for PEDF and fibroblast marker vimentin (A) or endothelial cell marker CD31 (B); DAPI was used to stain for nuclei. Representative photomicrographs are shown at X40 and X100 magnification. White triangles indicate examples of PEDF and vimentin (A) or CD31 (B) colocalization. Dotted white outlines indicate patches of PEDF localized to the dermal extracellular matrix. Isotype controls demonstrate specificity of primary antibodies. Scale bar = 50µm. PEDF is known to interact with endothelial cells (ECs) via membrane-associated proteins in other tissues [153]. Analysis of mouse skin sections for PEDF and CD31, a marker for ECs, shows that PEDF co-localizes with CD31 in elongated tube-like structures resembling capillaries in the dermis (Fig. 4.2B white triangles). PEDF that is not co-localized with CD31 nor DAPI and appears as patches of positive staining again suggests that much of PEDF is extra-cellular (Fig. 4.2B white dotted outlines).

88 Figure 4.3: Comparisons of baseline PEDF transcript expression in dermal fibroblasts (FB), keratinocytes (KC), and other cell types. A) Normal human epidermal keratinocytes (NHEK) and normal human dermal fibroblasts (NHFB) were cultured to equal confluency and real-time RT-PCR was used to measure PEDF transcript levels in the two cell types using GAPDH as a housekeeping gene. Data were normalized to NHEK and are shown as mean±sem, with n=3 for both groups. B) Relative PEDF transcript expressions in a publicly available microarray of matched human dermal FB and KC from keloids and control donor sites ([172], GSE44270). KC = keratinocyte, FB = fibroblast. C) Relative PEDF transcript expressions in a publicly available microarray of whole human skin, cultured skin substitute, as well as matched dermal FB and KC derived from skin samples ([173], GDS1505, GSE3204). D) Relative PEDF transcript expressions in a publicly available compendium of microarrays of various pure normal human cell cultures (GDS1402, GSE3239). 78

89 79 The aggregate results demonstrate that PEDF is ubiquitous in unwounded mouse skin, its spatial distribution is related to that of dermal FBs and ECs, that much of PEDF is localized in the dermal ECM, and that PEDF is produced primarily by dermal FBs but not appreciably by KCs nor ECs Distinct pattern of PEDF expression and production during excisional wound healing To investigate whether PEDF may play a role during skin healing, we measured levels of PEDF transcript and protein throughout the time-course of excisional wound repair. We previously published transcriptome analyses of 1-mm excisional wounds in mice [105]; we mined this dataset and found a distinct expression profile for PEDF (encoded by the Serpinf1 gene) in dermal wounds (Fig. 4.4A). To validate and extend this data to our standard injury model, full-thickness 3-mm wounds were made on the dorsum of mice and whole wound samples were harvested at 1,3,5,7,10,14,21 and 28 days post-wounding. These timepoints correspond to important milestones in the timeline of healing of experimental 3-mm excisional wounds in mice (Fig. 1.2). Real-time RT-PCR for PEDF shows that unwounded skin contains significantly more PEDF mrna than all timepoints post-injury (Fig. 4.4B). At Days 3-7, timepoints corresponding to the inflammatory and early-proliferative phases of healing (Fig. 1.2), PEDF mrna levels are significantly lower than in Day 21 wounds. PEDF protein levels were then analyzed by ELISA. Results demonstrate that unwounded skin contains significantly more PEDF than all timepoints post-injury except Day 28 (Fig. 4.4C). At Days 3-7, PEDF protein levels are significantly lower than in Day 28 wounds. This pattern is nearly identical to that observed for the PEDF transcript (Fig. 4.4B), with both results signifying a down-regulation of PEDF production in the early phases of healing followed by a gradual increase to baseline levels of PEDF during wound resolution (Fig. 1.2).

90 80 Figure 4.4: Pattern of PEDF mrna and protein production during excisional wound repair. Wound samples were harvested at multiple timepoints post-injury and analyzed for PEDF content. A) PEDF mrna expression profile mined from transcriptome microarray of 1- mm wounds ([105], GSE23006). Data were normalized to unwounded, normal skin (NS); n=3 for all timepoints; p<0.05 by one-way ANOVA. B) PEDF mrna expression levels were determined by real-time RT-PCR relative to GAPDH housekeeping gene on 3-mm wound samples harvested on Days 1,3,5,7,10,14,21 post-injury. Data were normalized to unwounded, normal skin (NS) and are expressed as mean ± SEM; n=3 for all timepoints; *p<0.05, **p<0.01 versus NS; # p<0.05, ## p<0.01 versus Day 21 by one-way ANOVA and Bonferroni s post-tests. B) PEDF protein content was determined by ELISA on 3-mm wound samples harvested on Days 1,3,5,7,10,14,21,28 post-injury. Data were normalized to total protein content and are expressed as mean±sem; n=10 for all timepoints, except: n=5 for Days 1 and 28, n=7 for NS, n=9 for Day 5; **p<0.01 versus NS; # p<0.05 versus Day 28 by one-way ANOVA and Bonferroni s post-tests Localization of PEDF in relation to fibroblasts and endothelial cells during excisional wound healing To investigate patterns of PEDF spatial distribution during healing, we double-stained tissue sections from full-thickness excisional wounds for PEDF and vimentin or CD31. Changes

91 81 in PEDF co-localization with these markers may provide clues to its differential interaction with dermal FB and EC during healing. Day 7 corresponds to the early-proliferative phase of healing, when provisional ECM dominates the wound bed (Fig. 1.2). At this timepoint, PEDF and vimentin staining are relatively weak and scattered, and the spatial pattern of PEDF is similar to that of vimentin (Fig. 4.5A top row). At Day 14, the extra-cellular staining for PEDF is more intense in the proliferating wound dermis (Fig. 4.5A middle row). By Day 28, a timepoint well into the remodeling phase of healing dominated by mature ECM (Fig. 1.2), PEDF staining is confined to certain areas of the wound dermis with significant co-localization of PEDF and vimentin (Fig. 4.5A bottom row), a spatial distribution very similar to that observed in unwounded skin (Fig. 4.2A). While the pattern of PEDF localization in the wound ECM changes considerably as the wound transitions from the proliferative to the remodeling phase, co-localization of PEDF with vimentin remains robust and unchanged at all timepoints (Fig. 4.5A), suggesting continuous production of PEDF by dermal FB during excisional healing. We next turned our attention to the co-localization of PEDF with endothelial structures. The well-described angiogenic profile of excisional dermal wounds [1, 44] was first quantified in this model, demonstrating high CD31 content on Days 5-14 post-injury and a progressive decline thereafter (Fig. 4.5C), corresponding to robust wound angiogenesis during the proliferative phase followed by prolonged vessel regression in the remodeling phase of healing (Fig. 1.2). At Day 7 post-injury, a period of active endothelial sprouting in the 3-mm excisional wound model (Fig. 1.2, Fig. 4.5C), CD31 staining is scattered and resembles the disorganized nature of the nascent vasculature at this timepoint (Fig. 4.5B top row). While PEDF staining is also diffuse, the spatial distribution of PEDF differs significantly from that of CD31, with minimal areas of co-localization in the tissue sections (Fig. 4.5B top row). At Day 14, a timepoint at the

92 Figure 4.5: Localization of PEDF during excisional wound repair in relation to dermal fibroblasts and endothelial cells. A, B) Wounds harvested at Days 7,14,28 post-injury were double-stained for PEDF and fibroblast marker vimentin (A) or endothelial cell marker CD31 (B); DAPI was used to stain for nuclei. Representative fluorescence photomicrographs are shown at X40 magnification. Dashed white lines indicate boundary of epidermis and dermis. White triangles indicate examples of PEDF and vimentin (A) or CD31 (B) co-localization. Dotted grey outlines in left panels of (B) indicate boundaries of CD31 staining transferred to the PEDF panels to delineate differences in spatial distribution between PEDF and CD31. C) Wound vascularity over the time-course of healing was quantified by measuring the percent area of wound bed occupied by CD31 staining. Data are expressed as mean±sem; n=4 for all timepoints, except: n=6 for normal unwounded skin (NS), n=5 for Day 10, n=3 for Day 3; *p<0.05 versus NS by one-way ANOVA and Bonferroni s post-tests. D) Representative composite photomicrographs showing co-localization of PEDF and CD31 in unwounded skin and Days 10, 28 post-injury at X40 magnification; white color represents co-localized pixels. 82

93 83 (Figure 4.5 legend, continued) E) Percent CD31 co-localized with PEDF over the time-course of healing was quantified by measuring the area of co-localized pixels compared to total CD31 area. Data are expressed as mean±sem; n=4 for all timepoints, except: n=6 for NS, n=5 for Day 10, n=3 for Day 3; *p<0.05, **p<0.01 versus NS; # p<0.05, ## p<0.01 versus Day 28 by one-way ANOVA and Bonferroni s post-tests. For all photomicrographs, scale bar = 50µm. transition from angiogenesis to vessel regression (Fig. 1.2), PEDF is distributed throughout the wound ECM, and co-localization between PEDF and CD31 remains sparse (Fig. 4.5B middle row). At Day 28, a timepoint well into the remodeling and vessel regression phase of healing (Fig. 1.2, Fig. 4.5C), CD31 and PEDF are more defined within the maturing wound bed, and robust co-localization of PEDF with CD31 is observed (Fig 4.5B bottom row), a situation that is nearly identical to that observed in unwounded skin (Fig. 4.2B). PEDF and CD31 co-localization was then quantified over the time-course of healing. Representative photomicrographs show that PEDF co-localizes highly with CD31 in unwounded skin and at Day 28, whereas co-localization is sparse at Day 10 (Fig. 4.5D), a timepoint of active angiogenesis during the proliferative phase (Fig. 1.2, Fig. 4.5C). Quantification reveals that nearly 70% of CD31 is co-localized with PEDF in unwounded skin and at Day 28 (Fig. 4.5E). Percent PEDF+ CD31 drops significantly to around 25% on Days 3-14 and progressively rises to 50% at Day 21 before returning to the baseline level observed in unwounded tissue at Day 28 (Fig. 4.5E). These results show that interaction of PEDF with wound microvasculature follows a pattern that is consistent with the hypothesized anti-angiogenic function for PEDF Exogenous PEDF inhibits angiogenesis and promotes collagen maturation during wound proliferation Having established a distinct pattern of PEDF production and localization during excisional wound repair, we sought to investigate the physiological role of PEDF in healing wounds. In a gain-of-function study, we treated mouse skin wounds with exogenous purified

94 84 recombinant PEDF (rpedf) or sterile phosphate-buffered saline (PBS) vehicle until Day 10 post-injury. The period of treatment with exogenous rpedf corresponds to timepoints when physiological levels of endogenous PEDF are low (Fig. 4.4), during the inflammatory and earlyproliferative phases of healing (Fig. 1.2A). Wound sizes were monitored throughout the experiment and no differences were found in macroscopic healing rates between rpedf- and PBS-treated wounds (Fig. 4.6A). To explore the effect of exogenous rpedf on sprouting micro-vessels, wound samples were harvested at Day 10 post-injury, a timepoint during maximum angiogenesis in this wound model (Fig. 1.2, Fig. 4.5C). Cryo-sections of treated wounds were co-stained with CD31 and alpha-smooth muscle actin (a-sma), a marker for vessel-stabilizing pericytes [174]. Representative photomicrographs show that rpedf-treated wounds have less robust staining for CD31 than PBS-treated wounds, while a greater proportion of CD31 is co-localized with a- SMA in rpedf-treated wounds (Fig. 4.6B). Quantification reveals that rpedf-treated wounds have significantly less (about 30%) CD31 content than PBS-treated wounds (Fig. 4.6C). Concurrently, rpedf-treated wounds have significantly more (about twice) a-sma+ CD31 than control wounds (Fig. 4.6D). These results demonstrate that rpedf-treated wounds exhibit less active angiogenesis but with a higher proportion of pericyte-supported vasculature. To investigate whether the anti-angiogenic effect of exogenous rpedf extended to larger macro-vessels, the underside of whole wound tissue samples was photographed and the percent area covered by blood vessels was quantified by first tracing visible vessels. Representative photographs of original and traced tissues show less vasculature in rpedfversus PBS-treated wounds (Fig. 4.6E). Quantification reveals that rpedf-treated wounds contain significantly less macro-vessels per area than PBS-treated wounds (Fig. 4.6F).

95 Figure 4.6: Treatment of wounds with rpedf decreases blood vessel density, increases pericyte coverage, and increases collagen maturity during the proliferative phase of healing. Murine excisional wounds were treated daily via topical application or intradermal injection of purified exogenous recombinant PEDF (rpedf) or PBS vehicle until Day 10 postinjury, when wound samples were harvested for analysis. For all graphs (A,C,D,F,H): data are expressed as mean±sem; n=6 for both treatment groups; *p<0.05, **p<0.01 between groups by two-tailed t-tests. A) Relative macroscopic wound sizes in PBS- and rpedf-treated mice. Data were normalized to original wound size. B) Cryo-sections of treated wounds at Day 10 postinjury were double-stained for endothelial cell marker CD31 and pericyte marker alpha-smooth muscle actin (alpha-sma). Representative fluorescence photomicrographs taken at X20 magnification show areas of CD31 and alpha-sma staining; right panel shows co-localized 85

96 86 (Figure 4.6 legend, continued) pixels in white, indicating pericyte-supported vasculature in the wound bed. Scale bar = 100µm. C) Wound micro-vascularity was quantified by measuring the percent area of wound bed occupied by CD31 staining. D) Percent CD31 co-localized with a- SMA was quantified by measuring the area of co-localized pixels compared to total CD31 area. E) The underside of wound tissues was photographed at time of sample harvest, images of wounds were cropped, and each visible macro-vessel was traced. Representative original and traced photographs show extent of macro-vessels in experimentally treated wound tissue. F) Wound macro-vascularity was quantified by measuring the percent area of wound tissue occupied by traced vessels. G) Representative Picrosirius Red photomicrographs at X10 (for orientation) and X20 (for detail) magnifications show extent of collagen maturation, where green-yellow indicates immature collagen and orange-red indicates mature collagen. Scale bar = 100µm. H) Percent mature collagen was quantified by measuring area of orange-red staining as a percentage of total collagen staining in the wound bed. To examine the effect of exogenous rpedf on the connective tissue in the wound bed, Picrosirius Red analysis was performed on treated wound samples to evaluate collagen content and maturity. Representative photomicrographs show that rpedf-treated wounds exhibit more red-orange staining in the wound bed compared to PBS-treated controls which show more green-yellow staining (Fig. 4.6G). Quantification reveals that rpedf-treated wounds have significantly more (about twice) mature collagen than PBS-treated wounds (Fig. 4.6H) Inhibition of endogenous PEDF delays blood vessel regression and collagen maturation during wound resolution To further explore the physiological role of PEDF in healing wounds, we performed a loss-of-function study where mouse skin wounds were treated with exogenous neutralizing antibody against PEDF (PEDF-ab) or an isotype control antibody (IgG) every four days starting at Day 8 post-injury. The period of treatment with inhibitory PEDF-ab corresponds to timepoints of healing when levels of endogenous PEDF are high (Fig. 4.4) and when PEDF begins to

97 87 Figure 4.7: Treatment of wounds with neutralizing antibody against PEDF increases blood vessel density and decreases collagen maturity during the remodeling phase of healing. Mouse excisional wounds were treated via intradermal injection with exogenous neutralizing antibody against PEDF (PEDF-ab) or its isotype control (IgG) every four days starting at Day 8 until Day 16, 20 or 24 post-injury. A) Relative macroscopic wound sizes in IgGand PEDF-ab-treated mice. Data were normalized to original wound size and are expressed as mean±sem; n=15 for both groups at all timepoints. B) Cryo-sections of treated wounds harvested at Days 16,20,24 post-injury were double-stained for endothelial cell marker CD31 and DAPI. Representative fluorescence photomicrographs taken at X10 magnification show areas of CD31 staining. Scale bar = 100µm. C) Wound vascularity was quantified by measuring the percent area of wound bed occupied by CD31 staining. Data are expressed as mean±sem; n=5 for both treatment groups at all timepoints; **p<0.01 between groups by two-tailed t-test. D) Representative Picrosirius Red photomicrographs at X20 magnification show extent of collagen maturation in wound bed, where green-yellow indicates immature collagen and orange-red indicates mature collagen. Scale bar = 100µm. E) Percent mature collagen was quantified by measuring area of orange-red staining as a percentage of total collagen stained in the wound bed. Data are expressed as mean±sem; n=5 for both treatment groups at all timepoints and for normal skin (NS); *p<0.05 between groups by two-tailed t-test. interact strongly with the wound microvasculature (Fig. 4.5E). Wound sizes were monitored throughout the experiment and show no differences in macroscopic healing rates between PEDF-ab- and IgG-treated wounds (Fig. 4.7A).

98 88 To explore the effect of PEDF inhibition on resolving microvasculature, treated wound samples were harvested on Days 16,20,24 post-injury, timepoints corresponding to vessel regression in this wound model (Fig. 1.2, Fig. 4.5C). Representative photomicrographs show that PEDF-ab-treated wounds have more robust staining for CD31 than IgG-treated wounds at the three timepoints investigated (Fig. 4.7B). Quantification reveals that PEDF-ab-treated wounds have significantly more (about 1.5 times) CD31 content than IgG-treated wounds at Day 16 post-injury (Fig. 4.7C left); the effect appears to be sustained at Days 20 and 24 post-injury, when the difference in CD31 content between PEDF-ab- versus IgG-treated wounds approaches statistical significance (Fig. 4.7C middle and right). These results demonstrate that inhibition of endogenous PEDF delays vessel regression during the remodeling phase of excisional wound healing. To investigate the effect of PEDF inhibition on the wound ECM maturation process, Picrosirius Red analysis was performed on treated wound samples. Representative photomicrographs show that PEDF-ab-treated wounds exhibit less red-orange staining in the wound bed compared to IgG-treated controls at Day 16 post-injury (Fig. 4.7D). Quantification reveals that PEDF-ab-treated wounds have significantly less (about 13%) mature collagen than IgG-treated wounds at Day 16, with no differences seen on Days 20 and 24 (Fig. 4.7E) DISCUSSION While much research has focused on evaluating the therapeutic potential of PEDF against a myriad of tumors [153] and other pathologies [76, 175, 176], little is known about the physiological function of PEDF. We present evidence of an important role for PEDF in the regulation of angiogenesis during dermal wound repair. Our data demonstrate that PEDF is a key natural anti-angiogenic factor in healing wounds that controls vessel regression, promoting a return to dermal homeostasis by influencing the vascular microenvironment.

99 89 We show that PEDF is abundant in mouse skin (Fig. 4.1), is produced at high levels by confluent dermal FB compared to KC (Fig. 4.3A), localizes with dermal FB and in the ECM (Fig. 4.2A), and co-localizes with microvasculature (Fig. 4.2B). The spatial distribution described here corresponds to PEDF s classification as a FB-secreted matri-cellular protein that binds collagen- 1 [76], an ubiquitous structural component of quiescent mammalian skin [13]. Though we recently found that KCs are capable of producing PEDF [160], we wanted to corroborate the finding that PEDF in mammalian skin is produced primarily by FB. To this end, we mined publicly available databases for comparisons of PEDF expression between mammalian dermal FB and KC. Direct comparison of samples from recently published microarray data of matched human dermal FB and KC from keloids and control donor sites [172] (GSE44270) shows that PEDF expression is higher in FB than in matched KC (Fig. 4.3B). Analysis of another published microarray data of whole human skin, cultured skin substitute, as well as matched dermal FB and KC derived from human skin [173] (GDS1505, GSE3204) shows that whole skin and primary FB have similar levels of PEDF expression, and both exhibit at least three times more PEDF expression than matched KC (Fig. 4.3C). With regard to the positive staining in murine hair follicles (Fig. 4.1), a recent study found abundant PEDF content in human hair follicles, especially in inner and outer root sheath cells [156]. These results demonstrate that dermal FB are the primary cell type responsible for the production of PEDF in mammalian skin, with some contribution from hair follicles and from epidermis. Data mined from a compendium of microarrays comparing expression profiles between various pure normal human cell cultures (GDS1402, GSE3239) show significantly higher levels of PEDF expression in fibroblast, smooth muscle and stromal cell lines than in endothelial or epithelial cell lines (Fig. 4.3D). The strong co-staining between PEDF and vasculature in mouse skin (Fig. 4.2B, Fig. 4.5E) thus implies that extracellular FB-secreted PEDF is binding to the microvascular ECs and is not being produced appreciably by them. The aggregate PEDF

100 90 localization data point to a role for PEDF in maintaining skin homeostasis, as PEDF secreted from resident FB and binding to the ECM and neighboring ECs may function to prevent unnecessary sprouting of neovessels in the quiescent dermis. We discovered a spatio-temporal pattern for PEDF production during skin wound healing that is reproduced at the transcript (Fig. 4.4A,B), protein (Fig. 4.4C), and histological (Fig. 4.5) levels. There is existing evidence to explain the observed low levels of PEDF in the early wound. First, dermal FB do not appear in appreciable numbers until inflammation is resolved in skin wounds [41]. Serum-activated, proliferating dermal FB express significantly less PEDF than quiescent FB [26, 177]; conversely, activated FB that first populate the wound bed have a proangiogenic phenotype, secreting VEGF and provisional matrix that promote capillary invasion [26]. Another study found that hypoxia down-regulates PEDF in myocytes and cardiac fibroblasts [178]. A similar paradigm may be at play not only in dermal FB but also in KC, as we recently reported that human KC decrease production of PEDF in response to mechanical injury or a pro-inflammatory environment [160], while these same cells have been shown to be a major source of pro-angiogenic VEGF in the early skin wound [10]. Thus, decreased FB numbers coupled with activated states of KC and FB due to hypoxia and inflammation likely account for the low level of PEDF present in the early excisional skin wound. The subsequent increase in PEDF production during the late proliferating and remodeling phases of healing may be explained by a return to a quiescent phenotype of wound FB and KC resulting from stabilizing oxygen levels and cell-cell contact inhibition after wound closure. The spatial relationship of PEDF to endothelial structures changes considerably during healing. Superposition of timecourse vessel density data with PEDF-CD31 co-localization shows a striking relationship wherein the percentage of PEDF-binding micro-vessels is inversely proportional to total micro-vessel content in the wound bed during the time-course of healing (Fig. 4.5C versus 4.5E). These spatio-temporal differences in PEDF-EC localization are likely

101 91 of functional importance and point to PEDF s hypothesized role as a matri-cellular protein that binds to ECs and activates anti-angiogenic signaling pathways during the remodeling phase of healing, when vessel regression is known to occur (Fig. 1.2). Treatment of wounds with exogenous PEDF significantly inhibits angiogenesis during the proliferative phase of healing (Fig. 4.6C,F), while inhibition of endogenous PEDF delays vessel regression in the remodeling phase (Fig. 4.7C). The effect on wound vessel density of exogenous PEDF during the proliferative phase was somewhat greater than that of inhibiting PEDF during wound resolution. In a previous study, we showed that exogenous application of VEGF and other pro-angiogenic factors to wounds throughout all phases of healing was able to further stimulate angiogenesis in the proliferative phase but could not prevent timely vessel regression [64], thus demonstrating the highly robust anti-angiogenic phenotype in the resolving wound. Indeed, there is emerging evidence for compensatory EC-mitigating mechanisms that activate during wound resolution, including those mediated by intracellular negative feedback against growth factor stimulation (e.g. Sprouty, see Chapter 3 [2]), ECM remodeling processes which affect biomechanical pathways in ECs and generate matrix-derived anti-angiogenic peptides, and secreted extracellular factors that promote EC apoptosis (reviewed in [1]). Antiangiogenic secreted factors include cytokine IP-10 (CXCL10) [69] and matri-cellular proteins such as thrombospondins [73] and now PEDF (Fig. 1.2B). Thus, due to the presence of compensatory mechanisms in the resolving wound, inhibiting just one of these factors had only a moderate effect on vessel regression. In contrast, ECs comprising the nascent vasculature during the proliferative phase are highly plastic due to activation by hypoxia and VEGF (among other pro-angiogenic factors) [46, 56] (Fig. 1.2B), which up-regulate both VEGF receptor and death receptor Fas (CD95) on EC membranes [1] (Fig. 4.8). This sensitizes the activated ECs to competing pro-angiogenic [46] and anti-angiogenic, pro-apoptotic pathways [59]. Indeed, pre-activation by VEGF is necessary

102 Figure 4.8: Proposed Model for PEDF function in the context of wound healing. Left side of diagram: During the inflammatory and proliferative phases of healing, hypoxia-driven activation of keratinocytes (KC), macrophages (Mϕ) and fibroblasts (FB) leads to VEGF release and binding to micro-vascular endothelial cell (MVEC)-associated VEGF receptors [46], causing VEGF-induced vasopermeability, activation of mitogen-activated protein kinase (MAPK) pathway, and mitogenic MVEC behaviors of proliferation and migration leading to angiogenesis [56]. Simultaneously, VEGF stimulation of MVEC causes activation of a negative-feedback mechanism mediated by Sprouty2 (Chapter 3, [2]) and presentation of Fas receptor (CD95) [59] thus sensitizing the activated MVEC to apoptosis [1]. Right side of diagram: During the remodeling phase healing, normoxia-driven quiescence of FB leads to production of PEDF [26] and collagen-1 as part of ECM maturation [13]. ECM-associated PEDF may bind to MVECassociated PEDF receptors [162], activating the ERK5 signaling pathway [179], leading to PPARɣ activation and NF-κB-induced expression of Fas Ligand (FasL, CD95L) [91]; binding of translocated FasL to Fas receptors in previously-activated MVEC initiates the extrinsic apoptotic pathway via Caspase 8 [59], leading to vessel regression. PEDF has also been found to inhibit VEGF-induced MVEC vasopermeability [78, 180] and migration [90] via other signaling pathways. 92

103 93 for the potent anti-angiogenic effects of PEDF on dermal microvascular ECs in vitro [59, 91, 152, 179]. In these cells, PEDF acts through ERK5 to activate PPARɣ [179], NF-κB and then up-regulate death ligand FasL (CD95L) [91], which initiates the Caspase 8 apoptotic pathway upon binding to Fas in activated ECs [179] (Fig. 4.8). This mechanism helps to explain how PEDF targets nascent, activated vessels while sparing mature, pericyte-stabilized blood vessels in the cornea [152], and is key to our finding that PEDF-treated wounds exhibit less total microvessel density (Fig. 4.6C) but a greater percentage of pericyte-supported microvasculature (Fig. 4.6D). The signaling pathways involved in PEDF-induced antiangiogenesis in other types of ECs have been explored (reviewed in [153]). Several studies have also worked out the mechanisms behind the anti-permeability effect of PEDF [78, 180], and these mechanisms are likely important during healing to counteract VEGF-induced sprouting of poorly-perfused vessels in wounds [49] (Fig. 4.8). The proposed mechanisms for the prevention of sprouting of nascent vasculature and maintenance of microvascular quiescence, when considered in light of our study s descriptive and functional data in healing dermal wounds, establish PEDF s role as an important homeostatic factor in mammalian skin (summarized in Fig. 4.8). Our findings are consistent with other published studies of pathological skin conditions treated with PEDF. In one experiment, psoriatic skin lesions treated via intradermal injection with rpedf or PEDF-derived peptides exhibited less microvascular content than controls [181]. In another experiment, PEDF-overexpressing melanoma xenografts in mice showed less total vessel density but more mature vasculature than control tumors. [182]. In those studies, as well as in others evaluating the effect of PEDF against pathological angiogenesis in tissues besides skin, PEDF was therapeutic and improved outcomes [153]. In our study, partial inhibition of physiological angiogenesis by PEDF did not affect overall healing kinetics (Fig. 4.6A) but seemed to improve the quality of the remodeling tissue, as evidenced by more refined

104 94 vasculature (Fig. 4.6D) and greater ECM maturity (Fig. 4.6H). These results are in line with previous studies from our laboratory and others which suggest that angiogenesis in mammalian dermal wounds is exuberant and may in fact be detrimental to the quality of the remodeled tissue, leading to scar formation [183, 184]. From this perspective, it can be argued that partial inhibition of wound angiogenesis, via blocking pro-angiogenic mechanisms and/or promoting anti-angiogenic mechanisms, may be used to improve the regenerative potential of the wound healing process in skin. Two recent studies have implicated circulating serpins in the pathogenesis of diabetesinduced chronic wounds. The study by McBride et al [185] found that Serpina4, also known as kallistatin, was significantly higher in the circulation of diabetic patients with microvascular complications. Kallistatin-overexpressing mice exhibited delayed dermal healing coincident with reduced wound angiogenesis. The study by Qi et al [186] found that circulating PEDF was higher in diabetic patients with foot ulcers. Neutralizing systemic PEDF in diabetic mice via anti- PEDF antibody perfusion or in PEDF knock-out mice improved healing rates coincident with higher wound angiogenesis. In contrast to our work, this study showed changes in healing rates as a result of PEDF modulation, but it is important to note that the study s focus was on systemic modulation of PEDF levels. As a multifunctional protein that is present and active in many adult tissues [76, 175], systemic PEDF would clearly affect other organs besides the healing skin. Our focus was investigating the role of locally-produced PEDF, and we modulated local PEDF levels via direct injection into wounds of rpedf or neutralizing antibody to PEDF; this was done to minimize off-target effects in other organs. Intriguingly, both studies implicated Wnt/beta-catenin signaling in serpin-mediated inhibition of wound angiogenesis [185, 186]. Unfortunately, neither study looked beyond the proliferative phase to explore the effects of systemic modulation of the serpins during wound resolution. Our work thus provides unique data about the role of locally produced PEDF in physiological vessel regression.

105 95 The promising therapeutic results of PEDF against various cancers [153] can finally be put in the context of a biological process that presents with a proliferative phenotype that naturally resolves. Indeed, cancer has been described as an over-healing wound [50] (Fig. 1.7), stalled in the proliferative phase without the onset of resolving mechanisms, including vessel regression and ECM maturation, as occur physiologically in wounds [1] (Fig. 1.2). The extensive poorly-perfused vascular network in wounds during the proliferative phase [49] has been compared to that of solid tumors [50]. In addition to its potent anti-angiogenic activity (Fig. 4.6C, Fig. 4.7C), PEDF s homeostatic effect on the vascular microenvironment, as demonstrated in this study by vascular integrity (Fig. 4.6D) and ECM maturity (Fig. 4.6H, Fig. 4.7E), may help explain its therapeutic efficacy as an anti-tumor agent in skin and other tissues. Importantly, cancer biologists have found that promoting the maturation of the tumor blood vessel microenvironment increases response to therapy via more efficient delivery of chemotherapeutic agents to cancer cells located deep inside tumors [102]. Another promising line of research has identified a central role for cancer-associated fibroblasts which maintain a proliferative, proangiogenic tumor microenvironment [187]. In contrast to cancers, quiescent FBs in wounds mediate the maturation of the wound microenvironment, partly through the production of PEDF, thus contributing to the strong anti-angiogenic phenotype (Fig. 4.8).

106 5. FINDING MEANING IN BIOLOGICAL COMPLEXITY: NETWORK ANALYSIS OF HEALING WOUNDS 5.1. INTRODUCTION TO SYSTEMS BIOLOGY OF WOUND HEALING Introduction to Systems Biology and Biological Network Analysis In the past decade, exponential advances in computing and networking have given rise to the digital/information age, of which the buzz phrase is Big Data. Big Data refers to the ability to efficiently collect in near- real time vast amounts of raw information-rich data from specific yet heterogeneous sources; to store this data in such a way so as to allow immediate access from remote locations; and then to analyze this complex web of data so as to derive patterns, models and real-world meaning as well as to generate predictions from which to more objectively base future decisions. Defined in this way, the utility of Big Data to science in general and to biomedical research in particular should be clear. It is quite remarkable that biology itself is information-rich: DNA and RNA are by their very nature biological information, and this encoded information lends itself to the Big Data initiative. While biologists have for decades been studying and manipulating the DNA, RNA and proteins of organisms to learn the functions of specific genes, the approach has been necessarily bottom-up, or reductionistic. This approach inferred large-scale understandings of biological systems from the independent, tightly-controlled investigations of the minute details of such systems. Since biological systems in mammals are complex, with regulation occurring on multiple levels simultaneously, the reductionist approach with its many assumptions and inferences from limited data could not generate highly accurate models of these biological systems. This approach yielded advances in the treatment of some acute diseases which have simple etiologies and pathogenesis, such as infections and genetic diseases. However, limited understandings of chronic diseases which have complex etiologies and pathogenesis, such as cancer and chronic wounds, resulted in disappointing translational advances in the treatment of these pathologies (Fib. 1.1B). 96

107 97 With recent improvements in DNA/RNA sequencing, mass spectrometry, as well as computing technologies, it has become possible to approach biomedical science questions from a top-down, systems perspective [188]. This perspective aims to explore and take into account large-scale changes in DNA, RNA and/or proteins and uses mathematical approaches to synthesize this heterogeneous data and model the biological system so as to approach datadriven, rather than inferred, large-scale understandings of the biological systems. The understandings of biology reached through systems methods may be more in tune with the reality of the complexity of biology and these insights could lead to more efficient biomedical applications in the treatment of diseases with complex etiologies. With unprecedented availability of vast biological information and computing power as well as mathematical modeling and computer science tools to analyze experimental omic data sets, it is possible to study complex, emergent properties arising from regulatory biological systems at the multidimensional levels of DNA, RNA and protein. With the exponential increase in the amount of data being generated by high-throughput technologies, a major challenge is the proper analysis of this data, including the development of statistical methods, bioinformatics software, and tools for analyzing and visualizing large datasets [189, 190]. One tool that has become especially valuable in systems biology is network analysis. Networks are graphs with vertices defined as nodes and connections between nodes defined as edges, while the analysis of network topology the structure of the network s branching nodes/edges follows the principles defined by Graph Theory [191]. In seminal papers which first applied Graph Theory to biological systems, Barabasi and Albert discovered that such systems, when represented by networks, remarkably follow certain universal properties that are also found in other real-world systems such as the power grid, social networks and the worldwide web [192, 193]. In any network, the degree of a node is defined by how many edges connect directly to it. A degree distribution (represented by P(k)) is a

108 98 numerical property of the network as a whole which gives the fraction of nodes that have certain degree (represented by k). The P(k) is obtained by counting the number of nodes N(k) that have k edges and dividing by the total number of nodes N [194]. The defining property of real-world networks is that they follow a power law in degree distribution, P(k) ~ k ɣ, where the degree exponent ɣ is greater than 1 [191, 194]. When networks follow a power law, there is a high diversity of node degrees, in that most nodes have low degrees, less nodes have medium degrees and a few nodes have high degrees and are thus highly-connected within the network. This scale-free network degree distribution leads to a certain network topology wherein the few highly-connected nodes serve as hubs which hold the rest of the network together [195]. Scalefree networks, which include biological networks, have been described as robust, in that they are remarkably resistant to failure by random removal of nodes but are sensitive to targeted removal of hub nodes [192]. In the past decade, the field of biological network analysis has flourished. Because the nodes and edges of networks can represent various components and levels of interactions, networks have been used to model a myriad of biological systems and processes: metabolism, genetics, epi-genetics, transcription factor binding, protein-protein interactions, phosphorylation, diseases [195]. Indeed, for nearly every ome, there can be a network created to visualize the complex dataset, and multidimensional networks can be created from the synthesis of related heterogeneous datasets. In the area of basic science research, network analysis is becoming a powerful tool for the study of complex biological processes. Detailed and highly predictive genetic, transcription and protein-protein interaction networks have already been developed for simple organisms such as bacteria and yeast [195, 196]. In more complex eukaryotic systems, such as mice and humans, networks are being used to study complicated physiological and pathological sub-

109 99 systems including multi-level gene regulation, development, tumorigenesis, inflammation, angiogenesis, connective tissue remodeling, among many others [ ] Unraveling the Complexity of Wound Healing Wound healing is one of the most complex biological phenomena in mammals, involving multiple cell types and hundreds of molecules, all interacting within a dynamic matrix in a spatially and temporally defined fashion [13] (Fig. 1.2A). Indeed, within healing wounds there occur, nearly simultaneously in space and time, multiple complex biological processes such as inflammation, angiogenesis and ECM remodeling. The previous two chapters describe the identification and characterization of two different anti-angiogenic factors, Sprouty and PEDF, using traditional methods. These methods may be described as reductionistic, measuring changes in specific molecular parameters after altering the levels of each respective factor in wounds. Decades of reductionistic science has identified hundreds of genes and pathways participating in these processes in various cell types, tissues, and/or higher model organisms. Much of this work was accomplished through the use of transgenic models, usually by knocking out or silencing a gene of interest in cells (in vitro) or higher organisms (in vivo). Thus, most of what we think we understand about the wound healing process is inferred from the results of these tightly-controlled studies. Indeed, the Introduction (Chapter 1) to this Thesis is a synthesis of such reductionistic data to infer a large-scale understanding of the mechanisms of blood vessel regression. With so many factors supposedly affecting wound healing, one may wonder what the significance of all this data actually is. It should be obvious from all that has been discussed so far that nothing in biology occurs in a vacuum, and while one or another factor may have been found to be involved in a process, it is only because that factor interacts in some way with other components in the wound. How do all these factors actually interact to bring about wound inflammation, proliferation, remodeling? Out of hundreds and perhaps thousands

110 100 of moving parts in a healing wound, which combinations of factors are most important to the biological process of interest, whether it be re-epithelialization, angiogenesis or ECM remodeling? Because of technological limitations, such questions were, until quite recently, beyond the scope of scientific investigation. Rapid advances in the methods of systems biology in general and biological network analysis in particular are now making it possible to explore large-scale changes in multiple components during the time-course of repair [203, 204]. While omics studies have been performed on mammalian wounds in the past [105, ], to date there has been little effort in the integration of network analysis to explore the coordinated expression of genes/proteins that give rise to emergent biological processes at the healing tissue level [208]. It is the purpose of this chapter to take the first exploratory steps into the area of network analysis of healing wounds in mammals. As a proof-of-concept, this chapter will attempt to answer the following questions: 1) Given that angiogenesis and ECM remodeling occur concurrently during wound repair (Fig. 1.2), what are the biological networks comprising these complex processes and how are they related? 2) Given that oral wounds heal with less angiogenesis and scar formation than skin wounds (Fig. 1.7), what are the large-scale changes in expression within these networks which may help to explain the greater regenerative potential of oral wounds? To explore these questions, we will first introduce several systems biology software and methods that are being used in the field of biological network analysis. We take inspiration from two recently published articles which aim to characterize all the genes and interactions involved in angiogenesis (the Angiome [209]) and ECM remodeling (the Matrisome [210]). The data from these studies will be used as starting points to create biological networks that describe the processes of angiogenesis and ECM remodeling. These networks will then be used to model

111 101 the process of wound healing by linking the wound transcriptome data described in Chapter 2. Specifically, genes that are part of the Angiome and Matrisome will be queried in the wound transcriptome and their respective expression profiles visualized in the dynamic networks. Topological and temporal patterns of expression will be identified and explored, both in relation to the individual tissues and as a way of contrasting oral versus skin healing MATERIALS AND METHODS Creation of Angiome, Matrisome, and Angiome/Matrisome protein-protein interaction networks Genes associated with vessel growth and regression were linked by networks of proteinprotein interactions (PPI), the angiome, via multiple systems biology databases and software, following as a starting point the paper by Chu et al [209]. Similarly, genes associated with ECM remodeling were linked to form the matrisome, following as a starting point the paper by Hynes and Naba [210]. Cytoscape software was utilized in conjunction with MiMI, GeneMANIA and BiNGO plugins to develop comprehensive data-driven networks. Cytoscape is a free, concurrent, plugin/app-enabled platform for importing and analyzing heterogeneous data from publicallyavailable biological databases and for the creation, editing and visualization of networks [ ]. Michigan Molecular Interactions (MiMI) is a meta-database which provides access to the merged and integrated data from numerous curated protein interactions databases, and the Cytoscape plug-in allows for the direct query of this database to find potential protein-protein interactions in a given list of genes [214]. GeneMANIA is a Cytoscape plug-in which allows for the identification of the most related genes to a given list of genes using a guilt-by-association approach which is useful for the building of networks [215]. BiNGO is another Cytoscape plug-in

112 102 which allows for the importation of Gene Ontology (GO) categories to the genes in a given network and determines which GOs are statistically overrepresented in a set of linked genes which is useful for the functional annotation of networks and gene enrichment analysis [216]. First, lists of genes associated with angiogenesis and ECM remodeling were downloaded from the two starting publications [209, 210] and imported into the Cytoscape platform by first converting the data to tab-delimited text files that identify each row using a common gene name readable by Cytoscape. The imported genes were connected via the MiMI and GeneMANIA plug-ins according to stringent criteria for protein-protein interaction evidence. Resulting networks were visualized and network topologies were analyzed using the Cytoscape network analysis function. Common genes between the angiome and matrisome were identified. Angiome and matrisome genes/node data was exported to tab-delimited text files readable by Python, ensuring that each row began with a common gene name readable by Cytoscape. Then, common genes between the two networks were pulled out using a simple Python program that read each consecutive gene in the input text file (i.e. angiome), searched for it in a list text file (i.e. matrisome), and if there was a match, copied the gene name to a new output text file (i.e. common). A sub-network of common/shared genes was created using the output text using Cytoscape and the MiMI and GeneMANIA plug-ins. The resulting angiome/matrisome shared network was visualized and its network topology was analyzed using the Cytoscape network analysis function. Gene Ontology and attribute data for all of the genes in the angiome, matrisome and angiome/matrisome networks were imported and linked via the BiNGO plug-in for Cytoscape. Other relevant gene attributes were imported to Cytoscape from the original angiome and matrisome publications [209, 210] by first converting the data to tab-delimited text files that

113 103 identify each row using a common gene name readable by Cytoscape. Networks showing colorcoded attributes were visualized by Cytoscape Analysis of wound transcriptome data to generate differential expression values Transcriptome data from microarrays of oral and skin wounds at multiple time-points (6h, 12h, 1, 3, 5, 7, 10 days) post-wounding was downloaded from the Gene Expression Omnibus database (GSE23006) [105]. The normalized expression data was sorted and analyzed in Microsoft Excel to identify differentially expressed genes between phases of healing and tissues in the following steps. For probes in the microarray assigned to the same gene, the expression values for all probes were averaged to yield one value per gene. First, to account for the faster response to wounding in the oral mucosa and to simplify the comparative analysis, several timepoints per tissue were averaged to represent phases of healing: inflammation (12h-Day 1), proliferation (Days 3-5), remodeling/resolution (Days 7-10). To study changes in gene expression as wounds transition from one phase to another (i.e. inflammation to proliferation, proliferation to resolution, inflammation to resolution), fold changes in expression from phase to phase were calculated (upregulation > 1; downregulation < 1; no change = 1). To study differences in fold changes in gene expression between tongue and skin, the fold change values for the tongue were subtracted from those for skin (upregulation in tongue < 0; upregulation in skin > 0; no change = 0) Linking transcriptome data to Angiome/Matrisome networks To explore differential regulation of angiogenesis and ECM remodeling, the wound transcriptome data was linked to the angiome, matrisome, and angiome/matrisome networks. First, the gene expression data was converted to tab-delimited text files that identify each row using a common gene name readable by Cytoscape. Next, the expression data was imported as gene attributes to the angiome, matrisome, and angiome/matrisome networks in Cytoscape. Networks showing color-coded attributes were visualized using color threshold settings.

114 104 Because of its lesser complexity compared to the full angiome and matrisome networks, the shared gene angiome/matrisome sub-network was further analyzed to identify potential hubs and/or regulatory motifs during transitions between phases of healing. This analysis was done to the dataset comparing oral and skin wounds EXPERIMENTAL RESULTS Visualization and Properties of Angiome and Matrisome networks The Angiome consists of 1233 genes linked by 6284 protein-protein interactions, similar to what was reported by Chu et al [209]. The network was visualized using Cytoscape and the resulting graph is shown in Fig. 5.1A. The size of the nodes corresponds to their degree, therefore the largest nodes, often found toward the center of the network, have numerous connections and could be considered hubs, while most of the nodes have only a few connections and are found toward the boundary of the network. The red box insert in Fig. 5.1A shows the degree distribution plot of the Angiome, where the x-axis is the degree and y-axis is the number of the nodes which have that degree; both axes are in the log-scale, so a relatively linear degree distribution results in the power law typical of real-world networks. With ɣ = and R 2 = 0.896, it can be said that the Angiome has the topology of a real-world, scale-free network. Given that the current knowledge about angiogenesis is incomplete and that the mammalian protein-protein interaction maps are still being assembled, it is remarkable that this network, developed as it is on limited information, should already resemble real-world networks. The Matrisome consists of 1178 genes linked by 6866 protein-protein interactions. The network was visualized using Cytoscape and the resulting graph is shown in Fig. 5.1B. The green box insert in Fig. 5.1B shows the degree distribution plot of the Matrisome.

115 A 105 B Figure 5.1: Visualization of the Angiome (A) and Matrisome (B) protein-protein interaction networks. Numbers of nodes (genes) and edges (interactions) are indicated. Inserts show degree distribution plots of respective networks, with power law fit parameters.

116 106 With ɣ = and R 2 = 0.835, it can be said that the Matrisome has topology approaching that of a real-world, scale-free network. Compared to the Angiome, the Matrisome has about the same number of nodes but more interactions; the group of highly-connected nodes at the top of the network appears to be skewing the degree distribution away from the power law as evidenced by the outlier in the plot. Besides this outlier group, the rest of the topology parameters are quite similar in the Matrisome and Angiome networks. Since we wish to study the interaction between angiogenesis and ECM remodeling, we must look at the overlap between these two networks Visualization, Properties, and Gene Ontology of Angiome/Matrisome sub-network The Python script identified 270 genes that are shared between the Angiome and Matrisome networks. These shared genes and their immediate interactions are highlighted in orange in Fig. 5.2A. In the Angiome, the shared genes account for 22% of the original nodes and 1676 or 27% of the original interactions. In the Matrisome, the shared genes account for 23% of the original nodes and 2366 or 34% of the original interactions. Interestingly, while in the Angiome the shared genes/interactions seem to be localized to a defined area of the overall network, the shared genes/interactions in the Matrisome appear to be spread out throughout the network. Similarly, the shared genes in the Angiome account for less of the highly-connected hubs than in the Matrisome; in the Matrisome the proportion of shared interactions is nearly 1.5 times greater. The Angiome/Matrisome sub-network consists of 211 nodes and 559 interactions. The sub-network created by Cytoscape contains 59 less nodes than the number of identified shared genes because not all of the shared genes can be connected via protein-protein interactions according the MiMI/GeneMANIA databases. The sub-network was visualized by Cytoscape and the resulting graph is shown in Figure 2B. Similarly to the original full networks, the size of the

117 107 A B Figure 5.2: Visualization of the shared genes/interactions within the Angiome and Matrisome networks (A) and the resulting Angiome/Matrisome sub-network (B), with degree distribution plot insert and color-coded GO attributes assigned to each node according to the insert legend. (#) in the legend corresponds to the total number of nodes in that category.

118 108 nodes corresponds to their degree, with the most highly-connected nodes found toward the center of the sub-network. The box insert in Fig. 5.2B shows the degree distribution plot of the Angiome/Matrisome sub-network, with a ɣ = and R 2 = 0.828, a topology approaching that of a real-world, scale-free network. The sub-network in Fig. 5.2B contains color-coded nodes according to GO attributes imported from BiNGO at two levels (see legend insert). The fill color refers to the type of protein that the respective gene codes for (e.g. collagens, ECM regulators, secreted factors, etc.), and the outline color refers to the angiogenic phenotype of the protein (green for pro-angiogenic, red for anti-angiogenic). There are a number of interesting patterns that emerge from the arrangement of the labeled nodes within the network. Many of the Secreted Factors, most of which are growth factors (e.g. FGFs, IGFs, VEGF) and functionally pro-angiogenic, are clustered together, with less-connected Secreted Factors scattered around the outskirts of the network. The Collagens are also clustered together. In contrast, the Glycoproteins and ECM regulators are, for the most part, scattered throughout the core network. As an exception, there is a small cluster of Serpin-class ECM Regulators in the lower-right corner of the network. The largest, most highly-connected nodes that are found within the core of the network are Glycoproteins (e.g. fibronectin and plasminogen) and ECM regulators (e.g. the MMPs); these are the hubs of the Angiome/Matrisome sub-network. Most of these hubs are also functionally anti-angiogenic, even though pro-angiogenic factors make up the majority of the nodes in the sub-network (72 versus 56).

119 Visualization of differential gene expression in Angiome/Matrisome sub-network separately in Skin and Oral Mucosa The wound transcriptome was re-analyzed to look at dynamic changes in gene expression between the different phases of healing separately in skin and oral mucosa, and these values were linked to the Angiome/Matrisome sub-network, as described in the Materials and Methods. Cytoscape was used to visualize, through color-coding, the up- and downregulation of the genes represented in the sub-network and the resulting graphs are shown in Fig. 5.3 and Fig Figure 5.3A shows the fold-changes in gene expression as the wounds transition from Inflammation to Proliferation, while Fig. 5.3B shows the same but for wounds transitioning from Proliferation to Resolution/Remodeling. In general, there appears to be more prevalent changes in gene expression in Fig. 5.3A than in Fig. 5.3B; that is, there is more upand/or down-regulation especially in the highly-connected core hubs in Fig. 5.3A, whereas these nodes show less changes in expression in Fig. 5.3B. These trends are similar in skin and oral wounds, but differences in the tissues are difficult to analyze by mere visual comparison of the figures side-by-side. Next, the overall changes in gene expression all the way from Inflammation to Resolution, thus skipping Proliferation, were mapped to the Angiome/Matrisome sub-network. The resulting graph is shown in Fig As opposed to the previous graphs, this one shows more prevalent up- and down-regulation throughout the network in both tissues. A general trend in Fig. 5.3 and Fib. 5.4 is that nodes clustered together tend to have similar patterns of expression.

120 110 A B Figure 5.3: Fold changes in gene expression in Angiome/Matrisome sub-network in skin and oral mucosal wounds during the phase transitions from A) Inflammation to Proliferation, and B) Proliferation to Resolution/Remodeling. The color legend at the bottom of the figure indicates: blue = down-regulation, red = up-regulation, green = no change.

121 111 Figure 5.4: Fold changes in gene expression in Angiome/Matrisome sub-network in skin and oral wounds during the phase transition from Inflammation to Resolution/Remodeling. The color legend at the bottom of the figure indicates: dark blue = down-regulation, red = up-regulation, green = no change Visualization of Differential Gene Expression, Regulatory Motifs, and Gene Ontology in Angiome/Matrisome sub-network between Skin and Oral Mucosal Woudns To more specifically explore differences in expression between the skin and oral mucosa, differential network analysis was performed by subtracting the fold-changes in oral mucosa from skin and mapping the values onto the Angiome/Matrisome sub-network. Cytoscape was used to visualize, through color-coding, the up- and down-regulation of the genes represented in the sub-network and the resulting graphs are shown in Fig. 5.5A for the

122 112 transition from Inflammation to Resolution and in Fig. 5.6A for the transition from Proliferation to Resolution. Comparing these figures to Fig. 5.3 and Fig. 5.4 which show skin and oral wounds side-by-side, it is clear that differential network analysis makes large-scale differences in gene expression between skin and oral wounds stand out. Exploring first the largest temporal changes, from the beginning of wound repair to the end, in skin versus oral wounds, Fig. 5.5A shows that most of the differences in expression are clustered together. Comparing the nodes in Fig. 5.5A to the color-coded GO attributes in Fig. 5.5B, the general trend is that the clustering correlates with attribute category. Specifically, the Collagen class of genes are much more differentially up-regulated in skin wounds. Connecting these nodes with their closest neighbors forms a potential regulatory motif highlighted by red edges in Fig. 5.5A and Fig. 5.5B. In addition to the various Collagens, this regulatory motif implicates ECM regulator MMP14; Glycoproteins Fibulin-2, Matrilin-2, Nov/CCN3, von Willebrand Factor; and Secreted Factors IGF1 and IGF2. Interestingly, these genes are mixed between pro- and anti-angiogenic functionality. The proteins implicated in this regulatory motif, through their interaction with one another throughout the time-course of repair, may be particular to the skin s regulation of ECM formation and remodeling, in conjunction with the skin s robust angiogenic response and resultant regression, and this relationship may contribute to scar formation. Exploring the changes in expression during the later phases of repair, from Proliferation to Resolution, allows for a more focused look at the phase during which vessel regression and ECM maturation occur. First, comparing the general trends in Fig. 5.6 to Fig. 5.5, it is clear that the patterns of expression are quite different. In particular, as compared to Fig. 5.5, less robust changes in gene expression occur in the more restricted time scale within wound healing that Fig. 5.6 represents. For example, the Collagens are no longer differentially expressed.

123 113 A B Figure 5.5: Differential network analysis in the transition from Inflammation to Resolution. A) Color legend indicates up-regulation in skin or oral mucosa. Potential transcriptional regulatory motif in skin wounds is shown by orange-colored nodes connected by red edges. B) Color-coded GO attributes are superimposed on potential regulatory motif in skin wounds.

124 114 A B Figure 5.6: Differential network analysis in the transition from Proliferation to Resolution. A) Color legend indicates up-regulation in skin or oral mucosa. Potential transcriptional regulatory motif in oral wounds is shown by blue-colored nodes connected by red edges. B) Color-coded GO attributes are superimposed on potential regulatory motif in oral wounds.