Stochasticity and Order: Studies of Keratinocyte Proliferation Amit Roshan MBBS, MRCS(Eng) Cambridge Cancer Centre Research Fellow, Department of Oncology Clare College, University of Cambridge Dissertation submitted towards the degree of Doctor of Philosophy December 2013 i ABSTRACT A central tenet of stem cell biology has been that proliferating tissues are maintained through a cellular hierarchy comprising of self-renewing stem cells at the apex, multiple lineage-restricted short-lived progenitor cells, and post-mitotic differentiated cells. The wide range of colony sizes in cultured human keratinocytes has been taken to support this hypothesis. Contrary to this model, researchers using genetic lineage tracing in mouse epidermis have inferred a single progenitor population for homeostasis, and a quiescent stem cell population activated upon wounding or genetic mutation. To study the proliferative behaviour of human keratinocytes, I used live imaging in vitro at single cell resolution. This shows two modes of proliferation: Type 1 cell division is stochastic with equal odds of generating dividing or non-dividing progeny, while Type 2 cell division predominantly produces two dividing daughters. These two modes are sufficient to explain the entire range of colony sizes seen after 7-12 days of culture and does not require a spectrum of proliferative ability. This insight provides a simple way to study the effects of external factors on cell fate. To exemplify this, I observed the effects of epidermal growth factor (EGF) and the Wnt agonist R-spondin on proliferation. Here I find proliferation in type 2 colonies changes by changing the proportion of cells dividing. This has implications for the limited success of EGF therapies in clinical trials following burns. To examine clonal contributions to wound repair, I used the mouse oesophageal epithelium which is exclusively composed of, and maintained by, a single progenitor population. I developed a micro-endoscopic wounding technique that produced localised superficial wounds. Here, I found that these wounds healed by uniform contribution from surrounding keratinocytes, demonstrating that reserve stem cells are not obligatory for wound repair. In summary, my work shows that human keratinocytes in vitro have two, and only two, modes of proliferation: a stochastic mode that is insensitive to external EGF signalling, and a EGF-sensitive exponential mode. Additionally, proliferation during wound repair can occur with stochastically dividing progenitors, and does not obligate stem cell recruitment in vivo. ii iii PREFACE This dissertation describes work done between October 2010 and October 2013 at the Hutchison/MRC Research Laboratories, Cambridge, UK under the supervision of Dr. Philip H. Jones. Chapter 5 contains some data that has been published, (included as Appendix 1): Doupé DP, Alcolea MP, Roshan A, Zhang G, Klein AM, Simons BD, & Jones PH. (2012) A single progenitor population switches behavior to maintain and repair esophageal epithelium. Science 337(6098), 1091-1093. Declaration of originality This dissertation is my own work and contains nothing that is the outcome of work done in collaboration with others, except as specified in the text and Acknowledgements. The data presented in this dissertation will not be submitted for the purpose of degree or qualification at this or any other university. Statement of length This dissertation does not exceed 60,000 words in length, including tables, figures, references and appendices. AMIT ROSHAN NOVEMBER 2013 vi vii CONTENTS Abstract i Preface iii Acknowledgements v Contents vii List of Figures & Tables xiii Glossary of Abbreviations xvii 1 Introduction 1 1.1 Motivation: Single Cell Analysis Can Define Cell Fate!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 1 1.2 Cellular Organisation of the Epidermis!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 1 1.3 Homeostasis in the Inter-Follicular Epidermis !⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 5 1.3.1 Early Histological Observations!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 5 1.3.2 Mitotic Indices and Colchicine!!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 5 1.3.3 Autoradiography and Cell Kinetics!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 7 1.3.4 The Epidermal Proliferative Unit!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 9 1.3.5 Molecular Evidence for Basal Cell Heterogeneity!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 9 1.3.6 Viral and Genetic Lineage Tracing!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 11 1.4 Lessons from Human Keratinocyte In Vitro Cultures!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 17 1.4.1 Development of Human Keratinocyte Cultures!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 17 1.4.2 Heterogeneity of Keratinocyte Colonies!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 19 1.4.3 Live Imaging of Keratinocytes!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 19 1.5 The Perturbation of Homeostasis: External Signalling !!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 21 1.5.1 The EGF Signalling Pathway!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 23 1.5.2 Keratinocyte Response to Supplemented EGF In Vitro!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 27 viii 1.5.3 Supplementing EGF for the Clinical Treatment of Burns!⋅!!⋅!!⋅!!⋅!!⋅ 29 1.5.4 Effects of Wnt/β-catenin Signalling on Keratinocytes In Vitro!⋅!!⋅ 31 1.6 The Perturbation of Homeostasis: Wounding ⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 33 1.6.1 Mouse IFE is Repaired by Resident Stem and Progenitor Cells!!!⋅ 35 1.6.2 Cells Outside the IFE Contribute to IFE Wound Repair!!!⋅!!⋅!!⋅!!⋅!!⋅ 37 1.6.3 Suprabasal Keratinocytes Contribute to IFE Wound Repair!!⋅!!⋅!!⋅ 39 1.6.4 Diverse Cell Types Contribute to Repair in Other Epithelia!!⋅!!⋅!!⋅! 39 1.7 Objectives of Research Presented !⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 43 2 Materials & Methods 44 2.1 Cell Culture!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 45 2.1.1 Fibroblast Culture and Feeder Layer Preparation!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 45 2.1.2 Human Neonatal Keratinocyte Culture!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 45 2.1.3 Primary Adult Human Keratinocyte Harvest and Culture!⋅!!⋅!!⋅!!⋅ 47 2.1.4 Culture at Clonal Density!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 47 2.1.5 Clonal Cultures for Live Imaging!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 49 2.1.6 PKH26 Dye Labelling of Keratinocytes!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 49 2.1.7 Cultures with altered EGF and R-Spondin!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 51 2.2 Culture Immunostaining, Imaging and Lysates!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 51 2.2.1 Immunofluorescence Staining of Culture Plates!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 51 2.2.2 Detection of EdU!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 51 2.2.3 Microscopy!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 53 2.2.4 Image Acquisition, Processing and Figure Preparation!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 53 2.2.5 Cell Lysate Preparation for Western Blots!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 53 2.3 Data Analysis!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 55 2.3.1 Comparisons of Populations!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 55 2.3.2 Lineage Trees!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 55 2.3.3 Monte Carlo Simulations!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 57 ix 2.3.4 Growth Curve Fitting!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 57 2.3.5 Figure Preparation!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 57 2.4 Combinatorial Statistics Theory!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 59 2.4.1 Assumptions and Parameters!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 59 2.4.2 Colony Size Distribution as Motzkin Paths!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 61 2.4.3 Small Colony Sizes Can Define Proliferative Behaviour!!!⋅!!⋅!!⋅!!⋅!!⋅ 65 2.4.4 A Game of Gambler’s Ruin!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 67 2.5 Mouse Oesophagus Experiments!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 69 2.5.1 Mouse Strains!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 69 2.5.2 Induction of Cre Mediated Recombinase!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 69 2.5.3 Induction of HGFP Expression!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 71 2.5.4 Labelling of S-Phase with EdU!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 71 2.5.5 Oesophageal Microendoscopy!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 71 2.5.6 Wholemount Preparation !!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 73 2.5.7 Wholemount Imaging!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 73 3 Human Keratinocytes Have Two Growth Dynamics In Vitro 74 3.1 Chapter Overview!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 75 3.2 Live Imaging of Cultured Human Neonatal Keratinocytes!!!⋅!!⋅!!⋅ 75 3.3 Live Imaging of Primary Human Adult Keratinocytes!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 87 3.4 Cell Division Outcomes are Independent!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 95 3.5 Two Growth Dynamics Explain the Entire Population Distribution 97 3.6 PKH26 Labelled Keratinocytes Confirm Two Rates of Growth!!⋅ 101 3.7 Small Colonies Confirm Type 1 Proliferative Dynamics!!!⋅!!⋅!!⋅!!⋅!!⋅ 105 3.8 Proliferating Cells Within Colonies at Early Time Points!⋅!!⋅!!⋅!!⋅!!⋅ 105 3.9 Colony Behaviours Have Distinct Transcriptional Profileles!!!!⋅!!⋅ 107 3.10 Discussion!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 111 x 4 Effects of EGF and R-Spondin on Keratinocyte Growth 114 4.1 Chapter Overview!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 115 4.2 Changes in Culture Media Affect Protein Expression!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 115 4.3 Live Imaging of Cultured Keratinocytes Without Added EGF!!!⋅ 117 4.4 A Dividing Rim of Cells Drives Linear Radial Colony Growth!!⋅ 123 4.5 Changes in Colony Size with Supplemented EGF!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 125 4.6 EGF and R-spondin Do Not Affect Type 1 Behaviour!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 127 4.7 Discussion!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 129 5 Cell Proliferation in Injured Mouse Oesophageal Epithelium 132 5.1 Chapter Overview!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 133 5.2 Cell Proliferation in Wounded Oesophagus!!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 133 5.3 Dilution of HGFP in Wounded Mouse Oesophagus!!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 135 5.4 Wounding of Clonally Labelled Oesophagus In Vivo!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 137 5.5 Discussion!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 139 6 Conclusions and Outlook 142 6.1 Key Lessons!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 143 6.2 Outlook: Open Questions and Future Experiments !!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 143 6.2.1 Molecular Mechanisms Underlying Cell Fate!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 143 6.2.2 Lineage Tracing in Human Xenografts: Steady State Ex Vivo!⋅!!⋅!!⋅ 144 6.2.3 Cell Fate Dynamics In Mutant Clones and Carcinogenesis!⋅!!⋅!!⋅!!⋅ 144 6.2.4 Automated Long Term Imaging In Vivo!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 144 References 146 xi Appendices Appendix 1: Published Article A-1 Doupé DP, Alcolea MP, Roshan A, Zhang G, Klein AM, Simons BD & Jones PH. (2012) A single progenitor population switches behaviour to maintain and repair esophageal epithelium. Science 337(6098), 1091-1093. Appendix 2: Pre-published Article A-43 Roshan A, Jones PH & Greenman CD. (2013) An exact time-independent approach to clone size distributions in normal and mutated cells. arXiv:1311.5769 [q-bio.QM] Appendix 3: Published Review A-63 Roshan A & Jones PH. (2012a) Act your age: Tuning cell behaviour to tissue requirements in interfollicular epidermis. Seminars in Cell & Developmental Biology 23(8),884-889. Appendix 4: Published Review A-71 Roshan A & Jones PH. (2012b) Chronic low dose UV exposure and p53 mutation: tilting the odds in early epidermal preneoplasia. International Journal of Radiation Biology 88(10),682-687. xii xiii LIST OF FIGURES AND TABLES 1 Introduction Figure 1.1 Human interfollicular epidermis!!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 2 Figure 1.2 Early recognition that mitosis occurs in the IFE basal layer!!⋅ 4 Figure 1.3 Basal keratinocyte divisions adopt one of three outcomes!⋅!!⋅ 6 Figure 1.4 Inconsistencies with the EPU model for skin homeostasis ⋅!!⋅ 8 Figure 1.5 Distribution of labelled clones in human xenografts !⋅!!⋅!!⋅!!⋅!!⋅ 12 Figure 1.6 Lineage tracing and cell fate in mouse tail IFE!!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 14 Figure 1.7 Clinical use of cultured human keratinocytes!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 16 Figure 1.8 Proliferative heterogeneity in human keratinocytes in vitro ⋅ 18 Figure 1.9 Selective human keratinocyte live tracking !⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!!⋅!!⋅!!⋅ 20 Figure 1.10 HER receptor dimers and their ligands ⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 22 Figure 1.11 The EGFR pathway!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 24 Figure 1.12 Endocytosis and translocation of EGFR!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 26 Figure 1.13 Effect of EGF & TGF-α on large keratinocyte colonies in vitro 28 Figure 1.14 The Wnt/β-catenin pathway!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 30 Figure 1.15 Stem cell contributions to wound healing in the mouse tail IFE 34 Figure 1.16 Contributions from skin appendages to IFE wound repair!!!⋅ 36 2 Materials & Methods Table 1.1 Donor details for primary human keratinocytes⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 46 Figure 2.1 Labelling of keratinocytes with PKH26 ⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 48 Table 2.2 Primary antibodies used for immunofluorescence!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 50 Figure 2.2 Deriving lineage trees from timelapse images ⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 54 Figure 2.3 Lineage trees can be drawn from larger colonies ⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅! 56 Figure 2.4 Cell division as a branching birth-death process!!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 58 xiv Figure 2.5 Motzkin paths are analogous to cell division choices!⋅!!⋅!!⋅!!⋅!!⋅ 60 Figure 2.6 Cell proliferation as a combinatorial branching process!!⋅!!⋅!!⋅ 62 Figure 2.7 Small clone sizes alone can define proliferative behaviour!!!⋅ 64 Figure 2.8 Gambler’s ruin predicts proportion of “everlasting” clones!⋅ 66 Figure 2.9 Mouse oesophageal endoscopic biopsies!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 70 3 Human Keratinocytes Have Two Growth Dynamics In Vitro Figure 3.1 Live tracking of clonal density neonatal keratinocytes!!⋅!!⋅!!⋅! 74 Figure 3.2 Type 1 colonies in neonatal keratinocytes!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 76 Figure 3.3 Lineage trees of neonatal keratinocyte Type 1 colonies⋅!!⋅!!⋅ 77-78 Figure 3.4 Type 2 colonies in neonatal keratinocytes !!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 79 Figure 3.5 Lineage trees of neonatal keratinocyte Type 2 colonies⋅!!⋅!!⋅ 80-81 Figure 3.6 Similar cell cycles across division types ⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 82 Figure 3.7 Cell fate changes in the middle of large colonies!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 84 Figure 3.8 Live tracking of clonal density adult keratinocytes!⋅!!⋅!!⋅!!⋅!!⋅ 86 Figure 3.9 Type 1 colonies in adult keratinocytes!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 88 Figure 3.10 Lineage trees of adult keratinocyte Type 1 colonies!!⋅!!⋅!!⋅!!⋅ 89-90 Figure 3.11 Type 2 colonies in adult keratinocytes!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 91 Figure 3.12 Lineage trees of adult keratinocyte Type 2 colonies!!⋅!!⋅!!⋅!!⋅ 92-93 Figure 3.13 Closely related keratinocytes in a colony stay together!!⋅!!⋅ 94 Figure 3.14 Related keratinocytes make independent outcome choices 96 Figure 3.15 Growth behaviour can predict colony size distributions!!!⋅ 98 Figure 3.16 Microscopic characteristics of day 12 keratinocyte colonies 100 Figure 3.17 PKH26 labelled keratinocytes show two growth patterns!⋅ 102 Figure 3.18 Day 7 keratinocyte colony distribution!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 104 Figure 3.19 Keratinocyte colonies at 72 hours post seeding!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 106 Figure 3.20 Studying early 8-cell keratinocyte colonies ⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 108 Figure 3.21 Protein heterogeneity in keratinocyte colonies ⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 110 xv 4 Effects of EGF and R-Spondin on Keratinocyte Growth 4.1 Culture conditions influence keratinocyte protein expression ⋅!!⋅!!⋅ 114 4.2 Cell cycle distribution in EGF0 is similar to EGF10 !!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 116 4.3 Cell fate in keratinocytes without supplemented EGF ⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 118 4.4 Lineage trees of keratinocytes without supplemented EGF !!⋅!!⋅!!⋅!!⋅ 119-20 4.5 Keratinocyte colonies in EGF0 culture media!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 122 4.6 An exponentially dividing rim gives linear radial colony growth ⋅! 124 4.7 Effect of EGF on area of keratinocyte colonies !⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅ 126 4.8 Effect of EGF and R-Spondin on keratinocyte colony distributions 128 5 Cell Proliferation in Injured Mouse Oesophageal Epithelium 5.1 Cell proliferation in wounded oesophageal epithelium ⋅!!⋅!!⋅!!⋅!!⋅!!⋅!!⋅! 132 5.2 Uniform contribution of cells to oesophageal wound repair !⋅!!⋅!!⋅!!⋅ 134 5.3 Rapid and uniform dilution of HGFP during wound repair !⋅!!⋅!!⋅!!⋅ 136 5.4 Clonal contribution to wound repair in oesophageal epithelium !⋅ 138 xvi xvii GLOSSARY OF ABBREVIATIONS 3H-TdR Tritiated thymidine β-NF β-naphthoflavone AKT Protein kinase B (PKB) APC Adenomatous polyposis coli AREG Amphiregulin BAD bcl-2-associated death promoter BrdU Bromodeoxyuridine BPAG Bullous pemphigoid antigen BSA Bovine serum albumin BTC Betacellulin CI Confidence interval CK1 Caesin kinase 1 CP Committed progenitor DAPI 4’,6-diamidino-2-phenylindole DIC Differential interference contrast DNA Deoxyribonucleic acid D-MEM Dulbecco’s modified Eagle’s medium DSG3 Desmoglein 3 DVL Dishevelled DTT Dithiothreitol DUB De-ubiquitylating enzymes EDTA Ethylene-diamine-tetra-acetic acid EdU 5-Ethynyl-2’-deoxyuridine EGF Epidermal growth factor EGF0 Epidermal growth factor 0ng/mL xviii EGF5 Epidermal growth factor 5ng/mL EGF10 Epidermal growth factor 10ng/mL EGF20 Epidermal growth factor 20ng/mL EGFR Epidermal growth factor receptor EPU Epidermal proliferative unit EGFP Enhanced green fluorescent protein EPGN Epigen EREG Epiregulin ERK Extracellular signal regulated kinase EYFP Enhanced yellow fluorescent protein FABP5 Epidermal fatty acid-binding protein (E-FABP) FOX Forkhead box FSG Fish skin gelatin GPCRs G-protein coupled receptors GSK-3B Glycogen synthase kinase 3 beta H&E Haematoxylin and eosin HB-EGF Heparin binding epidermal growth factor H2B-GFP HIST1H2BJ/ Enhanced green fluorescent fusion protein HER Human EGF receptor HF Hair follicle IFE Inter-follicular epidermis IVL Involucrin K1/5/10/14 Keratin 1/5/10/14 KIF Keratin intermediate filament LRC Label retaining cell LRP Lipoprotein receptor-related protein Lrig1 Leucine-rich repeats and immunoglobulin-like domains MAPK Mitogen activated growth factor xix MCSP Melanoma chondroitin sulphate proteoglycan MF Migratory front mTOR Mammalian target of rapamycin NFκB Nuclear factor kappa B NFSK Neonatal foreskin keratinocyte PBS Phosphate buffered saline PDGF Platelet derived growth factor PFA Paraformaldehyde PI3K Phosphatidylinositide 3-kinases PZ Proliferating zone RFP Red fluorescent protein RNA Ribonucleic acid RTK Receptor tyrosine kinase RSK Ribosomal protein S6 kinase SDS-PAGE Sodium dodecyl sulfate-polyacrylamide gel electrophoresis SG Sweat gland SPR Small proline-rich proteins TA Transit amplifying TCF/LEF T cell factor/lymphoid enhancer protein TG Transglutaminases TGF-α Transforming growth factor-α TBS Tris buffered saline UIM Ubiquitin-interacting protein xx xxi “…when you can measure what you are speaking about and express it in numbers you know something about it; but when you cannot…you have scarcely, in your thoughts, advanced to the stage of science, whatever the matter may be.” Lord Kelvin. Popular Lectures and Addresses Vol 1.(1891) p.80 “Quantification as such has no merit except insofar as it helps to solve problems. To quantify is not to be a scientist.” Sir Peter Medawar. Advice to a Young Scientist. (1979) p.18-19 1 CHAPTER 1 INTRODUCTION 1.1 Motivation: Single Cell Analysis Can Define Cell Fate Efforts to understand the proliferative capacity of adult cells have often been framed around the concept of a stem cell (Lajtha 1979; Potten & Loeffler 1990). As described since the 1970s, stem cells are believed to be rare, long-lived, self-renewing, and quiescent during homeostasis. They were thought, however, to possess the ability to execute a burst of cellular proliferation following wounding. As long-term residents in tissues with rapid turnover, stem cells were also described as the cell of origin for neoplastic transformation (Berenblum 1954). Recent experiments have however challenged a rigid concept of a hierarchically distinct cell with these attributes, with increasing emphasis on functional and behavioural plasticity (Nguyen et al. 2012). Advances in genetic labelling have allowed cells of interest and their progeny to be tracked for long periods in their native habitat (Kretzschmar & Watt 2012; Alcolea & Jones 2013). This allows direct observation of cell lineage in large numbers across a tissue through analysis of cohorts of labelled cells. While this approach has resulted in much revision to our understanding of cellular hierarchies in multicellular tissue, there are limitations on its application (Blanpain & Simons 2013). Tagged clonal distributions can be helpful to identify average long-term behaviour, but may be unrevealing for detailed attributes like cell cycle time distribution. Thus direct observation of quantitative measures that characterise all cell populations should be attempted. Advances in live imaging of cells allow such a study to be undertaken. In parallel, developing mathematical models that are informed by the data allows greater understanding of cellular behaviour. In this dissertation, I explore the power of live imaging of keratinocyte populations informed by complementary novel statistical approaches. 1.2 Cellular Organisation of the Epidermis The human epidermis is a remarkable paper-thin tissue, forming a flexible, but impermeable surface barrier from embryonic development throughout life. The inter- follicular epidermis (IFE) consists of multiple ordered layers of keratinocytes overlying a basement membrane, punctuated by appendages such as hair follicles (HF) and sweat glands (SG) (Fuchs 2007). Even in adulthood, the epidermis has to 2Figure 1.1. Human interfollicular epidermis. (Adapted from Candi et al., 2005). (A) Layers of the normal interfollicular epidermis resting upon the basal lamina (dotted line). The epidermis projects rete ridges in to the dermis, seperated from each other by dermal papillae, resulting in an undulating pattern. (B) Terminal differentiation in the epidermis. Proteins are expressed at characteristic locations in the epidermis during differentiation. Cell division is restricted to the basal layer, while differentiation occurs in the suprabasal layers. The cornified envelope is formed by proteins that are cross linked by transglutaminases (TG) with specific lipids on the outside. BPAG=bullous pemphigoid antigen; SPR=small proline-rich proteins. TG1, TG5, desmoglein-2, -3, -4 Keratin-5, -14, TG2, BPAG1 BPAG2/collagen-17 ն6շ4 integrin, laminin-5 Cornified Upper granular Granular Spinous Basal Basal lamina KeratinFilaggrin Loricrin SPRs LipidsInvolucrin TG3, keratin-1, -2e, -9, -10, desmoglein-1, desmocollin-1 Loricrin, profilaggrin, trichohyalin, involucrin, SPRs, S100A proteins basal granular cornified spinous Rete ridge Dermal papilla A B 3 keep “running to stand still”, being in a constant state of flux as differentiated cells are continually shed from the external surface while new cells are generated in the basal layer (Figure 1.1A). In addition, because the epidermis is continually subject to injury, it must be able to repair itself to restore its protective function. The epidermis is comprised of four layers resting attached to an inner basement membrane (McGrath & Uitto 2010). These layers from internal to external surface are termed the basal layer, the spinous layer, the granular layer and the cornified layer. An additional transitionary layer between the granular layer and the cornified layer exists in the thick epidermis of the palms and soles called the clear layer. Keratinocytes are characterised by producing copious amounts of cytoplasmic heteropolymers known as keratin intermediate filaments (KIF) that are in turn composed of keratin proteins (Fuchs 2007). As cell migrate through the layers of the skin, the expression pattern of keratins changes in a characteristic manner. Cells in the basal layer express keratin 5 (K5) and keratin 14 (K14), but switch to produce keratin 1 (K1) and keratin 10 (K10) in the suprabasal layer (Fuchs & Green 1980) (Figure 1.1B). This switch is considered the most reliable indicator that the epidermal keratinocyte has undergone commitment to terminal differentiation. In the spinous layer, the KIFs in the cytoskeleton bind robustly to tight cell-cell desmosome junctions giving the keratinocytes their “spinous” or “prickle-cell” appearance. As spinous cells migrate to the next granular layer, they undergo a series of structural changes (Candi et al. 2005). They initially form electron-dense keratohyalin granules giving them their descriptive name. These granules are packed with the protein profillagrin, which are eventually processed to cause further KIF bundling and generate large microfibrillar cables promoting the collapse of the cell to a flattened shape. Together, the KIFs and Filaggrin constitute 80-90% of the protein mass of the epidermis. Additional cornified envelope proteins, such as Involucrin (IVL), Loricrin and Trichohyalin, rich in glutamine and lysine residues are synthesised and deposited under the plasma membrane of granular cells. When the cells become permeabilised to calcium, crosslinking of these additional proteins occurs mediated by several Transglutaminases generating N! - γ -glutamyl lysine bonds. This creates a proteinaceous sac holding the KIFs and reinforcing the cornified envelope. Several components of the cornified envelope are still being discovered such as the S100 family of Ca2+ binding proteins and associated Epidermal Fatty Acid-Binding Protein (E-FABP or FABP5) (Ruse et al. 2001; Ruse et al. 2003; Eckert et al. 2004). The final steps in terminal differentiation involve the destruction of cellular organelles including the nucleus and the extrusion of lipid bilayers on the cornified envelope 4Figure 1.2. Early recognition that mitosis occurs in the IFE basal layer (Adapted from (A) Flemming, 1882, and (B) Ljunggren,1898). (A) Flemming made drawings to depict mitosis at various stages in the basal layer of the salamander skin. He later reported the same in human skin (not shown). (B) Ljunggren reported maintaining skin from a child in ascites fluid for three months, and successfully transplanting it back. As evidence of it surviving, he illustrated seeing cell division in both the basal layer and dermis indicated here. A B 5 forming an impermeable seal. The dead cornified cells are lost as squames, and continually replaced as inner layer cells move outwards. Most experimental evidence on the cellular organisation of the epidermis comes from the mouse. The organisation of the mouse epidermis is similar to the human epidermis with a few key differences (Gudjonsson et al. 2007; Rittié et al. 2013). Firstly, mouse epidermis generally comprises of only three cell layers and is <25!m in thickness, whereas the human epidermis commonly has 6-10 cell layers and is >50!m in thickness. Secondly, the IFE in humans undulates projecting in to the dermis in the form of rete ridges, separated by dermal papillae. Mouse epidermis is devoid of this feature. Thirdly, the density and type of appendages are different, with humans having the lowest range of HF density among mammals but having abundant eccrine sweat glands reaching densities of 200-700/cm2 (Sato & Sato 2000). Mouse eccrine sweat glands are limited to their footpads. Despite these differences of scale and density, valuable lessons may be gleaned from the study of murine epidermis, although applied cautiously between animals with a fifty-fold difference in life span. 1.3 Homeostasis in the Inter-follicular Epidermis 1.3.1 Early Histological Observations Mitosis was first recorded in the basal layer of the skin by Walther Flemming in 1882 (Flemming 1882) (Figure 1.2A). He postulated that the new cells produced by these mitosis exert a pressure that results in a movement of cells towards the region of least resistance at the surface of the skin (Branca 1912; Storey & Leblond 1951). The concept of only some cells being capable of proliferation was reported by Adami in 1901, with “proliferous” cells giving rise to “vegetative” cells while maintaining their “embryonic” state (Adami 1901). Ewing used the basal cells of the epidermis as an example of these dividing cells capable of maintaining a tissue (Ewing 1909). Here he highlighted the concept that proliferating basal cells maintained the tissue, but differentiated, non-proliferating suprabasal cells maintained the barrier function. 1.3.2 Mitotic Indices and Colchicine With the recognition that all divisions in the epidermis occurred as a result of basal layer mitoses, early research looked at the frequency of mitotic figures at rest, or after metaphase arrest with colchicine (Leblond & Walker 1956). The number of cells in mitosis after a fixed time with colchicine gave a “mitotic index” that was used to infer 6Figure 1.3. Basal keratinocyte divisions adopt one of three outcomes. (Adapted from Marques-Pereira & Leblond, 1965). Radioautographs of rat oesophageal epithelium fixed 48 hours after injection of 3H-thymidine. Basal divisions are seen to divide with either (A) two basal daughters (B) a basal and a suprabasal daughter or (C) two suprabasal daughters. Basement membrane in red dashed line. G=gran- ular layer, S=suprabasal, B=basal. Magnification 875X. A B C 7 a “renewal time” of 16.9 days in the mouse epidermis (Storey & Leblond 1951). This long division time was due to the assumption that all cells in the basal layer were proliferating. Thus, all basal layer cells were called stem cells, whose rate of division was balanced by cell loss from the skin surface (Berenblum 1954). It was speculated that the number of stem cells in normal skin stays constant with half of their daughter cells persisting as stem cells and the other half “embarking on the process of maturation towards their ultimate conversion in to dead keratin” (Berenblum 1954). 1.3.3 Autoradiography and Cell Kinetics The passive observation of cell division in human skin was transformed with the advent of cell labelling techniques using DNA precursors and detection using autoradiography in the 1960s. Tritiated thymidine (3H-TdR) was the most widely used of these agents. After administration, it is incorporated in to the DNA of proliferating cells in S-phase. Short-term dilution of the dye with cell division allowed a limited degree of lineage tracing of labelled cells. Leblond and colleagues performed the initial studies on homeostasis in stratified squamous epithelia using the rat oesophagus (Marques-Pereira & Leblond 1965). Here, a 3H-TdR pulse was used to label cells in S-phase, and after a 12-hour chase period, pairs of labelled cells were seen in the basal layer. This demonstrated that divisions occurred in the basal plane. At 2 days after labelling cells in the labelled cell pairs was now seen to be one of 3 configurations: both cells basal (29%), both cells suprabasal (29%), or one of each (42%) (Figure 1.3). This was interpreted to support a model in which all proliferating basal cells were equivalent, and that each daughter cell made an independent choice to either remain in the basal layer (and undergo further cell division) or differentiate and stratify with equal probability (50:50). A variety of 3H-dTR protocols have been used to study the kinetics of mouse epidermal proliferation in vivo including pulse labelling, double labelling and continuous labelling. Biological and methodological variability with these methods is well recognised and widely reported (Halprin 1972; Camplejohn 1983). In the mouse epidermis, Potten used mathematical modeling of these labelling counts to infer heterogeneity of cell cycle times (Potten et al. 1982). However, there are potential problems with interpreting the data in this fashion. Firstly, assumptions about proliferating fractions and variability of grain counts makes accurate estimation of cell cycle length difficult. Secondly, the mathematical modeling makes assumptions that are difficult to justify (for example, cell cycle length lies between 100-200 hours) and 8Figure 1.4. Inconsistencies with the EPU model for skin homeostasis (Adapted from (A) Roshan & Jones,2012a; (B) Mackenzie, 1970; (C) Kaur, 2006; (D) Ghaziza- deh & Taichman 2001 & (E) Doupé et al 2010). (A) Typical epidermis from mouse ear, arrowhead indicates hexagonal cornified cell outlined by melanin pigment. Scale bar 0.5mm. (B) Cryostat section of mouse ear showed cornified stacks of cells. Mitotic figures in the outer quarters of the basal layer underlying these stacks were seen to be more common than those in the inner quarters (55.8% versus 44.2%). (C) This led to a model of a “functional” unit of the IFE maintained by a infrequently dividing “stem cell” at the centre depicted in yellow. Peripheral cells were actively GLYLGLQJ EOXH DQGJDYHULVHWRVWDFNVRIRUGHUO\VXSUDEDVDOFHOOV JUHHQ  ' շJDO expression in retrovirally-transduced clones of cells arising from the basal layer were seen in columns of cells. (E) Genetically induced labelling of single basal cells develop clones shapes at 1 year that do not respect the cornified stacking bounda- ries (DIC top panel). Clones span multiple proposed EPUs without fully occupying any (YFP clones bottom panels) Scale bar 20ۚm. V 44.2% 55.8% OuterMitosis: Inner A B C D E 9 tries to fit 21 independent variables. Thirdly, the cell cycle length of a proliferating stem cell population was inferred to differ from other dividing cells in the epidermis by a factor of 2 (180 hours versus 90 hours). Such a narrow difference in estimated cell cycle lengths would be difficult to distinguish functionally over the long term. 1.3.4 The Epidermal Proliferative Unit Histologically, the mouse ear epidermis is seen to be organised in stacks of cells with a hexagonal surface lying on a bed of approximately ten basal cells (Mackenzie 1970; Potten 1974) (Figure 1.4). This structure was hypothesised to function as an epidermal proliferative unit (EPU) with one putative stem cell at its centre. The number of mitoses occurring at the central half of this EPU was found to be marginally lower than those in the peripheral half on cryosections (44.2% versus 55.8% of 346 mitoses) (Mackenzie 1970). Additionally, the central cell was found to be less likely to be in S-phase compared to peripheral cells, and more likely to retain DNA precursors upon pulse-chase experiments (Potten 1974; Mackenzie & Bickenbach 1985). However, this rigid organisation of cells has been overturned by the finding that EPUs may frequently be polyclonal in origin, and that a single rate of division maintains the mouse ear epidermis (Schmidt et al. 1987; Doupé et al. 2010). Additionally, the shape of cornified cells means they self-organise, spontaneously packing into a regular array of columns regardless of the distribution of differentiating cells emerging from the basal layer (Menton 1976a; Menton 1976b; Honda & Oshibe 1984; Honda et al. 1996). 1.3.5 Molecular Evidence for Basal Cell Heterogeneity Epidermal keratinocyte heterogeneity was first recognised by meticulous experiments exploring the proliferative ability of single keratinocytes in culture (Barrandon & Green 1987b). This is considered in detail in Section 1.4.2. The molecular evidence of proliferative heterogeneity within the human epidermis in situ comes from fractionating cells based on their expression of a range of protein markers. The first report was on the basis of β1-integrin, where cells expressing a high level of functional integrin have high colony forming efficiency in vitro and successfully regenerate the architecture of the human epidermis in xenograft experiments, a finding that parallels the high expression of β1-integrins in mouse tail IFE stem cells (Jones et al. 1995; Mascré et al. 2012). Correspondingly, cells with low levels of β1-integrin have a markedly lower proliferative potential, generating small 10 11 colonies which terminally differentiate in culture and fail to regenerate epidermis in xenografts (Jones & Watt 1993; Jones et al. 1995). These results were interpreted as β1- integrin being a marker for stem cells, with low levels of expression indicating cells with little or no self-renewal ability. Staining of the human epidermis has shown that β1-integrin bright stem cells are clustered irregularly at the tips of dermal papillae (Jones et al. 1995; Jensen et al. 1999). High levels of cell surface proteins like Delta-like 1, Melanoma Chondroitin Sulphate Proteoglycan (MCSP) and leucine-rich repeats and immunoglobulin-like domains 1 (Lrig1), and low levels of Desmoglein 3 (Dsg3) all co- localise with the β1-integrin bright cell population (Estrach et al. 2007; Jensen & Watt 2006; Legg et al. 2003; Lowell & Watt 2001; Wan et al. 2003; Wan et al. 2007). A range of other markers have since been described using a combination of surrogate assays like colony forming assays, cell cycle status, co-localisation, cell sorting and cell size. Controversy persists to the degree of stem cell enrichment provided by these markers in isolation or in combination (Ghadially 2012). 1.3.6 Viral and Genetic Lineage Tracing A direct approach to understanding clonal contributions to homeostatic IFE is by tracing single cells and their progeny through time. This powerful technique gives data that nucleotide labelling or stem cell marker expression cannot. In early studies, infection of murine epidermal keratinocytes with a retrovirus expressing LacZ was done in vitro, and then transplanted back in vivo (Mackenzie 1997; Kamimura et al. 1997). Subsequent studies employed direct labelling in vivo, but required either stimulation of proliferation by dermabrasion in adult mice, or the labelling of highly proliferative epidermis in very young mice (Ghazizadeh & Taichman 2001; Kameda et al. 2003). In all these studies, columns of cells were seen, but ranged widely in size and shape. Colonies crossed EPU boundaries in disagreement with a classical EPU model. Similarly, data collected from lineage tracing of reconstituted human epidermis has been non-instructive as the clonal data is limited (Kolodka et al. 1998). However, some information may be gleaned from a human xenograft fate tracking experiment which is the closest to an in vivo lineage tracing experiment using human epidermis (Ghazizadeh & Taichman 2005) (Figure 1.5). Here, high doses of red and green fluorescent protein tagged lentiviruses were injected intradermally in to neonatal foreskin xenografts on Swiss nu/nu mice 6 weeks after transplantation. This resulted in confluent reporter adjacent to the injection site, but low frequency labelling more distantly. Xenograft sections were examined up to 28 weeks for labelled clones that varied widely in size and shape and 12 Figure 1.5. Distribution of labelled clones in human xenografts (Adapted from Ghazizadeh & Taichman, 2005). Human Xenografts were transduced in situ with a mixture of lentiviral vectors encoding GFP (green) and RFP (red). (A) Numerous clones were seen at 28 weeks after transducing seen here at low magnification. (B) Sections of skin transduced with GFP lentiviral vector alone showing clones ODEHOOHGDWWKHEDVH E VLGH V DQGWRS W RIDUHWHULGJH6FDOHEDU ۚP C–E more than 78% of EPU (n ¼ 88) in this region were between one and four cells in diameter, suggesting that at least 10% Distribution of fluorescently labeled epidermal proliferative unit (EPU ) in skin sections. Human skin was transduced vectors encoding enhanced green fluorescent protein (GFP) ( green ) and DS Red2 fluorescent protein (RFP) ( red) (A ). Tissue sections were analyzed at 28 wk by fluorescent microscopy. ( A) The overall distribution o!abeled EPU is shown in a low power ) EPU labeled by a single virus are evident as either green or red columns, whereas the EPU labeled with both vectors appears as ) EPU originating from relatively flat areas of the basal compartment contain two to six cells at the base. The arrows in ( transduced melanocytes ( red) in the basal layers of epidermis. Scale bar ¼ 100 mm, (A) and 40 mm, (B–E ). D A B 13 arose throughout the basal layer. The interpretation of this qualitative dataset requires caution, as it is not clear whether labelled areas originate from single cells. However, the capacity to regenerate long-lived clones is present throughout the basal layer as in the mouse. The wide distribution of clone sizes cannot be explained by an EPU. More recent attempts to analyse cell fate in mouse epidermis have used genetic labelling techniques (Ro & Rannala 2004; Ro & Rannala 2005). Here, transgenic mice were engineered to express a mutant form of green fluorescent protein (GFP) that cannot be translated owing to the presence of a stop codon in the EGFP-coding sequence. Topical application of a mutagen induces mutations that can remove the stop codon and restore expression of GFP. These sporadic mutations resulted in patches of GFP-positive cells within the IFE, with clones spanning multiple adjacent suprabasal cell columns. However, the mutagen treatment is likely to affect cell behaviour. A subsequent study by the same group relied on spontaneous mutations to generate labelled clones (Ro & Rannala 2005). Again, four out of ten observed clones spanned multiple cell columns inconsistent with the classical EPU. More recently, the advent of cre recombinase based inducible genetic labelling in transgenic mice has provided a direct way to track the behaviour of proliferating cells in vivo (Kretzschmar & Watt 2012; Alcolea & Jones 2013). Doubly transgenic mice are engineered to express a drug related form of cre and a reporter gene that is only expressed following cre-mediated excision of a “STOP” cassette which blocks reporter expression. By using low doses of inducing drugs it is possible to titrate reporter expression to label scattered single cells, which subsequently expand into clones with proliferation (Clayton et al. 2007). Lineage tracing can be combined with wholemount techniques in which pieces of IFE are removed and stained intact (Braun et al. 2003). The three dimensional reconstruction of confocal image stacks of wholemount IFE allows entire clones to be visualised at single cell resolution (Clayton et al. 2007; Doupé et al. 2010). In normal IFE, analysis of large numbers of clones can reveal the proliferation signatures of cell fate choice made across the population. In the IFE, three such experiments have been reported using different inducible cre lines (Clayton et al. 2007; Doupé et al. 2010; Mascré et al. 2012). Of these experiments, two were in the tail, while the third is from the more typical epidermis of the ear. Tail skin is a specialised epidermis, in which rectangular scales, alternate with clusters of HFs in a regular array (Roshan & Jones 2012a). The pattern of HFs demarcates rectangular regions of inter-scale IFE containing around 5000 basal cells. In the tail, the AhCre line under the CYP1A1 promoter used in one study labelled cells in the scale IFE, while the IVLcreERT under the IVL promoter labelled dividing cells across the 14 Figure 1.6. Lineage tracing and cell fate in mouse tail IFE (Adapted from Roshan & Jones 2012a). (A) Scale forming tail skin, with arrowheads indicating edges of a scale with clusters of hair follicles lying between the scales. Scale bar 0.5mm. (B) Proliferative compartments in the tail IFE. Slow cycling stem cells are predominant- ly localised to the interscale IFE, and sparse or absent in the scale IFE which is main- tained by progenitors. (C) Cell behaviour in tail IFE inferred from lineage tracing studies. Rare stem cells divide once every 10-12 weeks on average, generating either two stem cell daughters, two progenitor daughters or one of each in the proportions shown. Progenitors divide 12 times faster than stem cells, generating progenitor and differentiated daughters in the proportions shown. Stem+Progenitor Progenitor Stem Cell Divisions Progenitor Cell Divisions B 10 wks SC SC SC SC P P P 10% 10% 80% 1-2/wkP D P P P D D 10% 10% 80% SC Stem Cell P Progenitor Cell D Differentiated Cell V V V V A C 15 entire IFE (Clayton et al. 2007; Mascré et al. 2012) (Figure 1.6). In these studies, basal cells were induced to label single cells at low frequency across the IFE and cohorts analysed up to a year. The clone sizes increased as cells proliferated, but the number of clones declined as they were shed. There was no change in the marker expression comparing the cells within the clone and outside the clones indicating that the clones were representative of all cells in the IFE (Clayton et al. 2007; Doupé et al. 2010). There were observed similarities between the two strains across two body sites. (Clayton et al. 2007; Doupé et al. 2010; Mascré et al. 2012). All three studies observed that all keratinocyte divisions resulted in one of three outcomes: two dividing basal daughters, two differentiating suprabasal daughters or one of each type similar to previous 3H-thymidine studies in the rat oesophagus (Marques-Pereira & Leblond 1965). Importantly, the average size of the remaining clones at each time point scaled with time, indicating that a single growth rate occurred across all the labelled cells. An additional population of cells was identified in the interscale IFE labelled by the K14creERT under K14 promoter (Mascré et al. 2012). This identified a similar three way balanced fate, but with divisions at a 10x slower rate (4-6 times a year). This population did not contribute to homeostasis. This confirms that a self-maintaining progenitor population achieves the maintenance of homeostatic IFE. Although lineage data from the human IFE is limited, some lessons can be drawn from studies in aged, sun-exposed skin. The basal layer of sun-exposed human epidermis flattens, losing the undulating pattern of rete ridges and dermal papillae with its clustered staining of β1-integrin and MCSP (Giangreco et al. 2010). This argues against an obligatory role for stem cells to maintain epidermis. Analysis of the sizes of p53 mutant clones in sun-exposed human IFE is consistent with a model of stochastic fate, but with a slight bias towards producing dividing daughters resulting in p53 clone expansion at the expense of non-mutant clones (Jonason et al. 1996; Jensen et al. 1999; Klein et al. 2010; Roshan & Jones 2012b). The quantitative distribution of clones in sun-exposed human epidermis is the same as UV-exposed mouse epidermis (Zhang et al. 2001; Klein et al. 2010). This argues that in both the human and mouse epidermis, the effect of p53 mutation and UV exposure results in a small tilt in stochastic fate towards proliferation with sustained UV exposure. Similar biases to stochastic cell fate has been recently observed with common mutations (Apc loss, Kras activation or p53 mutation) offering context-dependent clonal survival advantage in mouse intestine, but neutral competition with wild-type cells during homeostasis (Vermeulen et al. 2013). 16 Figure 1.7. Clinical use of cultured human keratinocytes. (Adapted from (A) Green, 2008 and (B,C) Ronfard et al., 2000). (A) Colonies of human epidermal keratino- cytes cultured from single cells stained with rhodamine. If cultured for longer periods, colonies merge to form an epidermal sheet of cells that can be transplanted back to the human donor. (B) 3.5 year follow up from a 9-year old boy with cultured epithelia transplanted on fibrin matrix following 95% surface area burns. Skin showing pliability when pinched. (C) Histologi- cal appearance of skin from (B) showing rete ridges and neodermis with vascular arcades. 6FDOHEDU ۚP A B C 17 1.4 Lessons From Human Keratinocyte In Vitro Cultures 1.4.1 Development of Human Keratinocyte Cultures In 1898, Ljunggren reported that pieces of human skin from a child remained viable when cultured in ascites fluid at room temperature for 3 months, and that this skin could be grafted back on to wounds (Ljunggren 1898) (Figure 1.2B). During the first half of the twentieth century, as in vitro culture techniques developed, explant cultures using keratinocyte outgrowths from skin grafts were used for experiments (Fischer et al. 1980). The discovery that trypsinised sheets of keratinocytes from human skin grafts could be used both for primary transplantation and as cell cultures, opened new areas of research including the study of the immune responses to transplantation (Medawar 1941; Billingham & Medawar 1951). However, keratinocytes required high seeding densities and subculture was rarely successful (Green 1980). A major breakthrough came from a chance observation that human keratinocytes could be cultured at low clonal seeding density if supported by the presence of a feeder layer of lethally irradiated 3T3 mouse embryo cells (Rheinwald & Green 1975) (Figure 1.7A). This technique has been used to grow keratinocytes from a variety of stratified squamous epithelia including skin, oral cavity, oesophagus, exocervix, and conjunctiva (Navasaria et al. 1994). Large numbers of cells can be grown from small starting biopsies, allowing the technique to be applied for expanded culture autografts in extensive burns (O’Connor et al. 1981; Gallico et al. 1984). Long term studies on cultured autografts in burns patients have shown good integration with the host and characteristics of normal skin (Compton et al. 1989). Cultured epithelial autografts grown on detachable fibrin matrices have since shortened culture time and improved handling characteristics with good long-term results (Ronfard et al. 2000) (Figure 1.7B&C). The Rheinwald and Green culture with modifications has since been used in a variety of applications including autologous limbal cell grafts for corneal injury (Rama et al. 2010) and for gene therapy for junctional epidermolysis bullosa (Mavilio et al. 2006). This widespread use in a variety of applications without neoplastic transformation upon re-transplantation to human hosts gives confidence that such cultured cells are not radically transformed in culture. It is important to highlight that in pre-confluent cultures using the Rheinwald and Green method, differentiated cells remain with the growing colony of cells, and are not shed. Commitment to terminal differentiation can be recognised by the cessation of further divisions, and subsequent expression of keratins associated with suprabasal keratinocytes (K5/K14 to K1/K10) and late markers of differentiation (IVL). 18 Subclones of a Holoclone Holoclone Sub-culture Subclones of a Meroclone Meroclone Sub-culture Subclones of a Paraclone Paraclone Sub-culture Figure 1.8. Proliferative heterogeneity in human keratinocytes in vitro (Adapted from Barrandon & Green, 1987b and Barrandon et al., 2012). Three types of kerati- nocyte colonies are described based on the proliferative ability of their subclones. Holoclones are large, smooth edged, and always arise from holoclones. Paraclones are small and terminally differentiate becoming positive for late keratinocyte differ- entiation marker Involucrin. Subclones from a paraclone, if any, are always para- clones. Meroclones are usually of intermediate size, have wrinkled edges and are described as “subterminal paraclones”. Inset scale=5mm, lower petri dishes 35mm. Proliferative ability Single cell Single cell Single cell 7 d ay s c ul tu re 12 d ay s c ul tu re 19 1.4.2 Heterogeneity of Keratinocyte Colonies The pioneering work of Barrandon and Green demonstrated that human keratinocytes exhibit a range of proliferative potential in vitro (Barrandon & Green 1987b). In this study, single human keratinocytes were seeded in a 3T3 feeder layer, and the resultant colony of keratinocytes was classified based on the frequency of involucrin positive terminal colonies produced 12 days after sub-culturing (Figure 1.8). At 12 days, three colony types were recognised. Holoclones produced less than 5% of terminal colonies, Meroclones produced >5% terminal colonies, while Paraclones produced 100% terminal colonies, if any. Holoclones typically produced subclones that were large and smooth-edged reaching 10-30 mm2 in size by 12 days. Meroclones typically produced wrinkled, intermediate sized colonies. Paraclones usually resulted in small (<5mm2) highly irregular terminal colonies, if any subclones were seen. An order was seen to occur, with holoclones giving rise to meroclones, and meroclones giving rise to paraclones. Holoclones therefore were seen to have the greatest growth potential, and were thought likely to be stem cells. Paraclones had little or no self-renewal potential. However, the nature of meroclones was not fully resolved, as they could give rise to both smooth edged colonies, which in turn had high subcloning efficiency and small wrinkled colonies with limited subcloning potential. They were described as subterminal paraclones generated from holoclones in a process of clonal conversion (Barrandon & Green 1987b). Additionally, it was found that the limited growth potential in paraclones could be transformed by a recombinant retrovirus encoding adenovirus E1A (Barrandon et al. 1989). Transformed paraclones formed disorganised epidermis when transplanted subcutaneously to athymic mice. This suggests that genetic transformation can change proliferation potential in differentiating cells. 1.4.3 Live Imaging of Keratinocytes Direct observation of keratinocyte division has been previously attempted to obtain additional information about human skin behaviour. Two short reports exist looking at the in vitro division of colonies derived from single human keratinocytes (Kitano et al. 1983; Dover & Potten 1988). In the first study, primary human keratinocytes were placed in high density monolayer culture and small colonies were tracked from day 8- 20 post seeding for a 6-day period at 30 minute intervals (Kitano et al. 1983). Culture was without a feeder layer, in Eagle’s media without supplemented EGF, and no stratification was observed. Up to three generations of division were seen in six lineage trees derived from five humans of ages 5-10 and one of 55 years. Around 200 20 Figure 1.9. Selective human keratinocyte live tracking. (Adapted from Dover & Potten, 1988). A single keratinocyte clone of 11 cells at 3 days post seeding tracked (3 cells from the 11 illustrated) for up to 69 hrs. S denotes a “suprabasal migra- tion”, whereas L denotes a “lost” cell to migration or loss of focus. Note that indi- vidual cell divisions may result in two dividing daughters, two non-dividing daughters, or one of each type. 11 cell colony, selected at ~ 3 days L L L L S S S S S S S S S S S S 1 2 3“0” hrs +24 hrs +48 hrs +72 hrs “0” hrs +24 hrs +48 hrs +72 hrs 21 divisions were recorded, with an average cell division between 15.1-27.6 hours (range 8-68.5 hours). More than 99% of divisions took less than 48 hours. In the second study, 11 clones from neonatal foreskin primary culture at clonal density was examined with continuous recording (Dover & Potten 1988) (Figure 1.9). Culture was with a feeder layer, using the Rheinwald and Green culture media, but without supplemented EGF. The clones were pre-selected as reaching a “dividing size” at day 3, and tracked onward for up to 3 days at 5s intervals. Only part of one clone is presented as a lineage tree. The cell division rate is reported to be 15.6 hours (range 10.5-30.5 hours). These two studies are limited by pre-selection, different culture conditions and limited data presented making population-wide conclusions difficult. However, there are some qualitative observations one can draw. Firstly, cell cycle length does not correlate with proliferative ability of the daughter cells. Secondly, daughters within the same colony may choose different proliferation fates. Thirdly, cell cycle time varies but is less than 48hrs in nearly all cases and that cell cycle time does not differ between donor age groups. And lastly, that cells within a colony do not seem to mix between the periphery and the centre of colonies. Live imaging of a cohort of cells over a long period of time could resolve some of the issues around proliferative fractions and kinetics within the epidermis. There has been a recent study of live imaging in mouse HF for short periods (3-14h), but this is not feasible in a human (Rompolas et al. 2012; Rompolas et al. 2013). In vitro live imaging studies in other contexts, such as rat retinal cells and mouse mesodermal haemogenic cells, have revealed the developmental origin of cells, and given a basis for proliferative heterogeneity in these tissues (Gomes et al. 2011; He et al. 2012; Eilken et al. 2009; Schroeder 2013). The automated tools available to track population wide proliferative behaviour is currently limited to short timeframes, and is not possible in co-cultured conditions where white light recognition of different cell types is not currently possible (Schroeder 2011; Cohen et al. 2010). Manual tracking can be achieved, but limits sample sizes as it is labour intensive (He et al. 2012). 1.5 The Perturbation of Homeostasis: External Signalling Multiple external signals are recognised to influence keratinocyte proliferation (Sotiropoulou & Blanpain 2012). In my experiments, I use the Epidermal Growth Factor (EGF), and the Wnt agonist R-Spondin to see their effects on cell fate. In this section I will consider the effects of EGF and Wnt signalling on keratinocytes. 22 Figure 1.10. HER receptor dimers and their ligands. (Adapted from Okines et al, 2011). HER2 homodimers have no known ligand, and HER3 homodimers are inac- tive. Homodimers are less mitogenic that heterodimers, and HER2 is the preferred and most potent heterodimeric partner. 7 known ligands for EGFR are indicated, with Epiregulin and Epigen being weaker mitogens than others. HER=human epidermal growth factor receptor (EGFR), TGF-Ƚ=transforming growth factor-Ƚ. P P P P P P P P P P P P P EpiregulinEpiregulin Heparin-binding EGF-like ligand Heparin-binding EGF-like ligand EpigenEpigen Betacellulin BetacellulinAmphiregulin Heregulin 3 or 4 Heregulin 1 or 2EGF/TGF-Ƚ EGFR HER2 HER3 HER4 23 1.5.1 The EGF Signalling Pathway The term EGF was first coined in the early 1960s by Stanley Cohen to describe a polypeptide purified from mouse submaxillary glands that affects epidermal growth and differentiation in the newborn mouse (Cohen 1962; Cohen & Elliott 1963). Specifically, EGF causes premature eyelid opening and incisor eruption reflecting the accelerated differentiation of these structures during neonatal development. Shortly after its discovery, EGF was found to stimulate proliferation of epidermal keratinocytes leading to the characterisation of a cell surface receptor specific for EGF (Cohen 1965; Carpenter et al. 1975; Buhrow et al. 1982). During the last few decades, multiple EGF-related growth factors and at least four different EGF receptor-like receptors have been shown to affect development, regeneration, differentiation and transformation of cells derived from multiple tissues (Citri & Yarden 2006). The ERBB family of proteins (originally named because of their homology to the erythroblastoma viral gene product v-erbB) comprises of four receptors (HER for Human EGF Receptors) and 13 polypeptide extracellular ligands which contain a conserved EGF domain (Citri & Yarden 2006) (Figure 1.10). The HER proteins have been variously named as EGFR (HER1/ERBB1), HER2 (ERBB2/Neu), HER3 (ERBB3) and HER4 (ERBB4). Although the HER family is considered to be prototypical group of the receptor tyrosine kinase (RTK) family, an important defining feature of the HER network is that HER2 and HER3 are non-autonomous. HER2 lacks the capacity to interact with a growth factor ligand, while the kinase activity of HER3 is defective (Guy et al. 1994; Klapper et al. 1999). Despite this lack of autonomy, both HER2 and HER3 are capable of generating potent cellular signals. HER2 works as the preferred heterodimeric partner of the other three HER members (Tzahar et al. 1996). It causes potent mitogenic signals owing to simultaneous and prolonged recruitment of multiple signalling pathways (Pinkas-Kramarski et al. 1996). HER3 strongly activates PI3K, especially when in heterodimer form with HER2 (Wallasch et al. 1995). The autonomous receptors HER1 and HER4 share features, binding to multiple ligands and forming homodimers and heterodimers. Seven HER1-ligands have been identified so far, some of which bind to HER1 specifically, while others bind to both HER1 and HER4 (Schneider & Wolf 2009) (Figure 1.10). These are EGF, transforming growth factor-α (TGF-α), amphiregulin (AREG), heparin-binding EGF-like growth factor (HB-EGF), betacellulin (BTC), and two low-affinity ligands, epiregulin (EREG) and Epigen (EPGN). All HER1 ligands are synthesised as their membrane-anchored forms (proEGFR ligands), and are proteolytically cleaved in ectodomain shedding to become soluble forms. EGFR 24 Figure 1.11. The EGFR pathway (Adapted from Okines et al, 2011).Ligand binding to EGFR results in a conformational change of the receptor, allowing dimerisation. Dimerisation is essential for receptor function and precedes tyrosine kinase activa- tion via hospitalisation. This activates downstream proteins initiating the Ras/Raf/MEK/MAPK and PI3K/AKT pathways to initiate gene transcription. AKT=protein kinase B; BAD=bcl-2-associated death promoter; FOX=forkhead box; GSK-3B=glycogen synthase kinase 3Ⱦ; MAPK=mitogen-activated protein kinase; MEK=MAP kinase kinase; mTOR=mammalian target of rapamycin; NFɈB=nuclear factor kappa B; PDK1/2=3-phosphoinositide-dependent protein kinase 1/2; PI3K=phosphatidylinositol 3 kinase; PIP2=phosphatidylinositol 4,5-bisphosphate; PIP3=phosphatidylinositol 3,4,5-trisphosphate; PTEN=phosphatase and tensin homolog; Raf=GTPase Raf; Ras=GTPase Ras. Homodimerisation or heterodimerisationLigand EGFR P P PDK1/ PDK2 PIP2 PIP3 PI3K Glucose metabolism Cell-cycle progression, cell proliferation, survival and growth Phosphorylation PTEN AKT Ras Raf MEK MAPK Cyclin D1BADGSK-3B NFɈBmTOR FOX 25 undergoes rapid internalisation and degradation following ligand activation (Sorkin & Goh 2009; Lemmon & Schlessinger 2010) (Figure 1.12). At low physiological concentrations, EGF-induced EGFR internalisation is primarily by clathrin-mediated endocytosis. By contrast, with high EGF concentrations EGFR endocytosis is primarily internalised by clathrin-independent mechanisms. Although initial views were that EGFR endocytosis and degradation terminates the signal initiated by EGF, it is now known that signals can be transmitted from endosomes and that the spatial localisation plays an important role in control of signal specificity, duration and robustness (Miaczynska et al. 2004; Sigismund et al. 2008; Sousa et al. 2012). HER receptors can also be activated by transactivation by other signals including hormones, lymphokines and stress inducers (Yarden & Sliwkowski 2001). Examples include activation of G-protein coupled receptors (GPCRs), cytokine regulated tyrosine kinase Jak2 and Wnt signalling. These molecules stimulate HER tyrosine kinase activity either directly by phosphorylating the receptors or indirectly by inducing the cleavage of ligand precursors. The C terminus of HER1 contains a recognition site for the ubiquitin ligase Cbl, whereas no site is found that can directly recruit the lipid kinase phosphatidylinositol 3–kinase (PI3K) (Levkowitz et al. 1999). With the specificity of its docking sites, HER1 cannot directly activate the PI3K-AKT/protein kinase B (PKB) pathway, but it couples to the Ras-MAPK pathway, as well as to the Ras-PI3K-AKT/PKB pathway, forming the major downstream signalling routes for EGFR. Several tyrosine-based motifs recruit a number of signal transducers to the phosphorylated form of HER1 such as the adapter proteins growth-factor-receptor- bound-2 (GRB2) and Src-homology-2-containing (Shc), which are responsible for the recruitment of Ras and activation of the mitogen-activated protein kinase (MAPK) cascades (Schulze et al. 2005) (Figure 1.11). Ras mediates its effect on cell proliferation by triggering a cascade of phosphorylation events starting with the RAF protein (Hallberg et al. 1994). Activated RAF phosphorylates MAPK kinases (MEK1/2, previously known as ERK1/2 or extracellular signal regulated kinases) (Downward 2003). Substrates of ERK proteins can be found in the cytoplasm as well as the nucleus. One of the earliest recognised effects of MAPK activation is to activate mRNA translation via ribosomal protein S6 kinase (RSK) and in turn phosphorylation of ribosomal protein S6 (pS6) (Avruch et al. 2001; Pende et al. 2004). Upon activation, phosphorylated ERKs are translocated to the nucleus and trigger phosphorylation of several transcription factors including c-Fos, c-Jun and c-Myc leading to cell cycle progression (Johnson & Vaillancourt 1994; Downward 2003; Normanno et al. 2006). 26 Figure 1.12. Endocytosis and translocation of EGFR (Adapted from Citri & Yarden, 2006). Ligand binding to the EGFR receptors and their subsequent dimeri- sation induces receptor internalisation into endosomes. Here, autophosphorylated receptors might recruit the E3 ubiquitin ligase Cbl and undergo ubiquitylation. These appended ubiquitins then recruit adapters containing a ubiquitin interacting motif (UIM) and target internalised receptors to lysosomes for degradation. De-ubiquitylating enzymes (DUB) may abrogate this process and target EGFR mol- ecules to the default recycling pathway. Recycling Degradation in lysosomes P P P P Ub Ub Ub Cbl DUB EGFR UIM 27 Ras also activates the PI3K-AKT pathway (Rodriguez-Viciana et al. 1994) (Figure 1.11). A subgroup of the PI3K family of enzymes interacts directly with HER1 upon activation through the regulatory subunit (p85) or PI3K (Carpenter et al. 1993). Upon activation of PI3K phosphatidylinositol-3,4,5-trisphosphate (PIP3) is produced which binds to the major mediator of PI3K, AKT (Burgering & Coffer 1995). As a serine/threonine kinase and key mediator of signal transduction pathways, AKT is related to cell survival, growth and proliferation (Cantley 2002; Vivanco & Sawyers 2002; Testa & Tsichlis 2005). AKT mediates cell survival by inhibiting apoptosis by inhibiting pro-apoptotic players such as BAD (Bcl-2-associated death promoter) and caspase 9, inhibiting degradation of NF-κB, and enhancing degradation of p53 (Datta et al. 1997; Cardone et al. 1998; Romashkova & Makarov 1999; Mayo & Donner 2001). Cell growth effects of AKT are thought to be mediated by mTOR (mammalian target of rapamycin), a direct target which works as a molecular sensor regulating protein synthesis (Nave et al. 1999). The effects of AKT on proliferation are by inhibiting the degradation of cyclins required for cell cycle progression or by inhibiting the expression of cyclin dependent kinases (Diehl et al. 1998). 1.5.2 Keratinocyte Response to Supplemented EGF In Vitro Adding EGF to human keratinocyte cultures extends their lifetime (Rheinwald & Green 1977). Here, when EGF was supplemented to the neonatal foreskin keratinocyte cultures, lifespan tripled and colony stratification decreased while maintaining growth rate. The authors suggested that EGF inhibited terminal differentiation. Later studies have shown that the inhibition of EGFR activation in keratinocyte cultures triggers expression of early differentiation markers K1 and K10 (Peus et al. 1997). However, in contrast, in keratinocytes in suspension culture EGF promoted terminal differentiation, with the suggestion that EGFR activation in suprabasal cells may differ from basal cells (Wakita & Takigawa 1999). In 3D organotypic cultures on collagen, EGF treatment results in thinner, less well-organised constructs with impaired differentiation and frequent parakeratosis (Chen et al. 1995). Activation of EGFR signalling has been shown to induce the expansion of large human keratinocyte colonies by a lateral migration of peripheral cells (Barrandon & Green 1987a; Nanba et al. 2013) (Figure 1.13). However, the early effect (up to 3 hours) on small colonies is rapid shrinkage (Nanba et al. 2013). This difference in effect is due to a difference in actin filament organisation in each colony type, with inhibition by Rac1, a member of the Rho family of small guanosine triphosphatases (GTPases), resulting in re-organisation of filaments in a manner resembling smaller colonies. Rac1 28 Figure 1.13. Effect of EGF & TGF-Ƚ on large keratinocyte colonies in vitro (Adapt- ed from Barrandon & Green, 1987a). Single sister cells from a large 8 day old colony were placed in separate plates with feeder cells. After day 6, EGF 30ng/mL, equiva- lent TGF-Ƚ in a receptor binding assay, or no growth factors were supplemented every 3 days till 24 days of culture. (A) The radius of the growing colonies plotted against time for each of the three conditions show linear increase, with different slopes. (B) Rhodamine stained end colonies of EGF and control at 24 days, showing dramatic increase in final size. Scale=1cm. EGF TGF-Ƚ Control EGF Control A B 29 null mice fail to maintain the interfollicular epidermis, and although low levels of Rac1 does not show effects in the IFE, Rac1 inhibition or knockdown in vitro limits growth capacity of large human keratinocyte colonies (Benitah et al. 2005; Chrostek et al. 2006; Castilho et al. 2007; Nanba et al. 2013). This suggests that human keratinocytes with large colony forming potential are maintained in culture by the co- operation between EGFR signalling and actin filament dynamics. In keratinocytes in vitro, uneven distribution of EGFR between two daughter cells may also have a regulatory role in keratinocyte fate (Le Roy et al. 2010). In normal keratinocytes in culture, a small population of differentially expressed EGFR was reported, with the EGFR positive daughter behaving like a differentiated cell, while the EGFR negative cell possessing quiescence and ability to form large colonies. In studies using exogenous EGF, the contribution of endogenous EGFR ligands to keratinocyte proliferation and differentiation are not examined. Studies using EGFR knock-out mice have reported impaired epidermal stratification (Miettinen et al. 1995; Sibilia & Wagner 1995). However, grafting EGFR null skin on wild type mice results in a hyperproliferative epidermis compared to EGFR wild type grafts (Hansen et al. 1997). These studies highlight the complex nature of EGFR function in skin in vivo. 1.5.3 Supplementing EGF for the Clinical Treatment of Burns The investigation of exogenous EGF in acute wound healing was investigated as early as 1973, with hyperplasia seen after application to corneal wounds in vivo (Savage Jr & Cohen 1973). Quicker epithelial regeneration with less scar formation was reported in EGF-treated wounds on rabbit ears (Franklin & Lynch 1979). The use of massively expanded keratinocyte cultures for regenerating a functional epidermis in humans after autologous transplantation was found to require EGF signalling (Rheinwald & Green 1977; Barrandon & Green 1987a). As a consequence of this application, great promise was held by EGF therapies for large area burns. In human trials, initial reports using frequent high doses of EGF (5!g/cm2 every 12 hours) with sulfadiazine creams showed less time to heal than controls in the same patient with a 15-20% increase in area of re-epithelialisation (Brown et al. 1989). A similar increase in healing had already been reported with TGF-! (Schultz et al. 1987). With limited benefits reported from subsequent clinical trials, focus shifted to impaired wound healing in chronic ulcers. While there are marketed EGF- incorporated therapies available, effects have been limited (Falanga et al. 1992). Current research in the field concentrates on alternative drug delivery models in the 30 Figure 1.14. The Wnt/Ⱦ-catenin pathway (Adapted from Clevers & Nusse 2012). In the absence of Wnt ligand, the destruction complex lies in the cytoplasm, where it binds, phosphorylates and ubiquitinates Ⱦ-catenin by Ⱦ-TrCP. The proteasome recy- cles the complex by degrading the Ⱦ-catenin. In the presence of Wnt, the intact com- plex is associated with phosphorylated LRP. After binding to LRP, the destruction complex still captures and phosphorylates Ⱦ-catenin, but ubiquitination by Ⱦ-TrCP is blocked. Newly synthesized Ⱦ-catenin accumulates, and translocates to the nucleus to effect transcription. LRP=Lipoprotein receptor-related protein; Dvl=Di- shevelled; CK1=casein kinase 1; GSK3=glycogen synthase kinase 3; APC=adeno- matous polyposis coli gene product. LRPFrizzled Wnt Proteasome 31 chronic wound environment, but efficacy has yet to be determined. EGF monotherapy in acute burns on healthy humans seems to have a limited effect on keratinocyte re- epithelialisation limiting its widespread uptake as a clinical strategy. 1.5.4 Effects of Wnt/β-catenin Signalling on Keratinocytes In Vitro The Wnt family of secreted glycoproteins are one of the fundamental mechanisms regulating direct cell proliferation and cell fate determination (Clevers & Nusse 2012) (Figure 1.14). A critical and heavily studied Wnt pathway is the canonical Wnt pathway, which functions by regulating the amount of the transcriptional co-activator β -catenin, which controls key developmental gene expression programs. In the absence of Wnt, cytoplasmic β-catenin is constantly degraded by the action of the Axin complex, which is composed of the scaffolding protein Axin, the tumor suppressor adenomatous polyposis coli gene product (APC), casein kinase 1 (CK1), and glycogen synthase kinase 3 (GSK3). CK1 and GSK3 sequentially phosphorylate the amino terminal region of β-catenin, resulting in β-catenin recognition by β-Trcp, an E3 ubiquitin ligase subunit, and subsequent β -catenin ubiquitination and proteasomal degradation (He et al. 2004). This continual elimination of β-catenin prevents it from reaching the nucleus, and Wnt target genes are thereby repressed by the DNA-bound T cell factor/lymphoid enhancer factor (TCF/LEF) family of proteins. The Wnt/β-catenin pathway is activated when a Wnt ligand binds to the seven-pass transmembrane Frizzled (Fz or Fzd) receptor and its coreceptor, low-density lipoprotein receptor-related protein 6 (LRP6), or its close relative LRP5. The formation of a likely Wnt-Fz-LRP6 complex, together with the recruitment of the scaffolding protein Dishevelled (Dvl), results in LRP6 phosphorylation and activation and the recruitment of the Axin complex to the receptors. These events lead to inhibition of Axin-mediated β -catenin phosphorylation and thereby to the stabilisation of β - catenin, which accumulates and travels to the nucleus to form complexes with TCF/LEF and activates Wnt target gene expression. R-Spondin (R-Spo) is an unrelated protein to Wnts that acts through the Fz/LRP complex as a Wnt agonist (Clevers & Nusse 2012). Humans have four R-Spo proteins that are defined by two N-terminal furin domains and a thrombospondin domain. The first evidence that R-Spo proteins potently enhance Wnt/β-catenin signals came from Xenopus (Kazanskaya et al. 2004). R-Spo1 was subsequently found to feed into the canonical Wnt pathway, promoting intestinal crypt proliferation in vivo and in vitro (Kim et al. 2005; Sato et al. 2009). This supports a crucial role for R-Spos in Wnt/!β- 32 33 catenin signalling. Disruption of the gene RSPO1 is associated with palmoplantar hyperkeratosis and a predisposition to squamous cell carcinomas (Parma et al. 2006). Although the Wnt/β-catenin signalling is well explored in the HF, conflicting evidence exists regarding its involvement in the IFE (Dasgupta & Fuchs 1999). The constitutive attenuation of Wnt/β-catenin signalling in the epidermis either by loss-of- function mutations in β-catenin, LEF1 or Porcupine results in a normal IFE with loss of hair morphogenesis or cycling. However, many Wnt /β -catenin signalling mutations result in interfollicular phenotypes, for example Porcupine mutation characterises Focal Dermal Hypoplasia (Goltz syndrome) which has congenital streaks of markedly thin dermis (Grzeschik et al. 2007; Wang et al. 2007). This suggests some degree of β-catenin signalling is required for IFE maintenance. Experiments with human keratinocytes in vitro suggest that the Wnt/β-catenin pathway has a role in maintaining the IFE. Here, putative IFE stem cells showed an increase in colony-forming ability and expressed higher levels of noncadherin- associated cytoplasmic β-catenin than small colony forming cells (Zhu & Watt 1999). Retroviral transduction of constitutively stable N-terminally truncated β -catenin increased the proportion of putative stem cells to 90% of the proliferative subpopulation with no effect on differentiation or alteration to cell cycle kinetics. Conversely, induced expression of a dominant-negative form of β-catenin stimulated cells to exit from the stem cell compartment (Zhu & Watt 1999). These experiments provide evidence for a role of Wnt/β-catenin signalling in maintaining different proliferative abilities in keratinocyte cell pools, while also coordinating with other signals to influence subsequent decisions to differentiate. 1.6 The Perturbation of Homeostasis: Wounding Although the role of stem cells in maintaining homeostatic tissue is being challenged, they are still widely thought of as the only cell capable of responding to the high cell turnover required during wound repair. As non-stem cells are postulated to lack the ability to self-renew, they would be excluded from long-term contributions to healing tissue. In this section, I consider the evidence for different contributions of epithelial cells to re-epithelialisation following wounding. To study the effects of cellular proliferation in wounding, my experiments (Chapter 5) have used the mouse oesophageal epithelium. This particular epithelium is useful for such a study as it lacks appendages such as crypts or glands, which form stem cell 14 days 21 days 14 days 21 days K14creER IVLcreER 34 Figure 1.15. Stem cell contributions to wound healing in the mouse tail IFE (Adapted from Mascré et al., 2012). Clonal contributions to wound healing follow- ing punch biopsy was assessed by genetic lineage tracing in two mouse models. YFP immunostaining was performed on whole-mount of wounded tail epidermis 4 weeks after tamoxifen induction and analysed at timepoints indicated. (A) K14creER mice label a population of cells in the interscale epidermis that contribute large clones that persist at 35 days post wounding. (B) IVLcreER mice label a popu- lation of cells across the epidermis (both scale and interscale) that contributes to wound repair, but whose clones do not persist at 35 days post-wounding. Scale bars 200ۚm. 35 days 35 days10 days 10 days YF P YF P A B 35 niches in the skin (Solanas & Benitah 2013). As a simple tissue with 3-4 layers of keratinocytes, homeostasis is by a single population of basal progenitor cells (Doupé et al. 2012). 1.6.1 Mouse IFE is Repaired by Resident Stem and Progenitor Cells Most evidence for keratinocyte clonal dynamics in the IFE comes from atypical skin of the mouse tail (Gomez et al. 2013). Here, HFs are arranged in groups of three in staggered rows. The IFE adjacent to the HFs are known as the interscale IFE and undergoes orthokeratotic differentiation similar to the dorsal skin IFE. In contrast, the remaining scale IFE undergoes parakeratotic differentiation characterised by a lack of a granular layer and retention of nuclei in the cornified layers. Scales, like the HFs, are regularly spaced and arranged in rows around the tail. The infundibulum of the HF connects with the interscale IFE while the hair shafts overlie the scale. Label retention assays using BrdU have demonstrated the presence of slow cycling cells in the mouse tail epidermis, and confirmed by transgenic HGFP dilution (Braun et al. 2003; Mascré et al. 2012). These label retaining cells (LRCs) lie predominantly in the interscale IFE adjacent to the HF (Mascré et al. 2012) (Figure 1.15). Here, K14creERT mice induced with topical low-dose Tamoxifen were used to track clones in 2-day old mice when scale formation begins. This demonstrates that only 4% of clones at 1 week following induction, rising to 9% by 3 months, cross the scale- interscale boundary (Gomez et al. 2013). This would argue that during the rapid postnatal expansion of tail IFE, separate cell populations largely maintain the scale and interscale regions. In adult K14creERT mice treatment with low dose Tamoxifen preferentially labels slow cycling cells in a distribution similar to the H2BGFP retaining cells, enabling lineage tracing to be undertaken (Mascré et al. 2012). Here, labelled cells divide every 10-12 weeks in a three-way fate choice similar to progenitors, forming a balanced number of slow cycling stem cells or more rapidly dividing progenitors. The role of slow cycling interscale IFE stem cells changes dramatically upon wounding. Upon punch biopsy, the slow-cycling cells are recruited to repair the defect and generate long-lived clones in the healed epidermis at 35 days post wounding (Mascré et al. 2012). In contrast, scale IFE progenitor clones did not form long-lasting clones at 35 days, although they too contributed to the initial wound repair and underwent massive expansion in numbers. This is presumably due to their clones being lost to differentiation. Thus, although wounding initially heals with a contribution from both progenitor and stem cells, only stem cells produce persisting 36 Figure 1.16. Contributions from skin appendages to IFE wound repair (Adapted from (A) Ito et al., 2005 and (B) Rittié et al., 2013). (A) Lineage analysis of LacZ labelled bulge cells after punch biopsy of dorsal mouse skin. Gross appearance of wounds after full thickness excision at time points indicated. Bulge cells are seen to contribute to wound repair emerging from hair follicles and migrating linearly to close the wound by 8 days. The area covered by bulge-derived cells decreased with time from 26% at 8 days to 2% at 50 days. Scale bars 500ۚm. (B) 3D reconstruction from immunohistochemistry of whole skin biopsy samples in human volunteers obtained 3 days after wounding. Consecutive sections were cut parallel to the skin surface, and pilosebaceous units (arbitrary blue), eccrine sweat glands (arbitrary magenta), and new epidermis (arbitrary yellow) were marked. Computer recon- struction through stacking shows contributions to human wound healing from both the eccrine and pilosebaceous units. Gray mesh indicates sample contours. A B 2days 5days 8days 37 clones at 35 days after healing. This is in agreement with earlier studies in the mouse dorsal IFE, where mechanical abrasion or tape stripping revealed widespread cell recruitment and accelerated division arguing against a stem cell contribution alone (Morris & Argyris 1983; Potten et al. 2000). 1.6.2 Cells Outside the IFE Contribute to IFE Wound Repair Keratinocytes from outside the mouse IFE are also known to contribute to wound healing, including those from the HF, and other appendages including the sweat and sebaceous glands (Arwert et al. 2012). Hair follicle cells contribute to IFE wounding (Figure 1.16A). In the mouse back skin, lineage tracing indicates that clones from hair follicle stem cells in the bulge and upper follicle migrate and contribute to IFE repair, with some clones persisting long- term (Ito et al. 2005; Levy et al. 2005; Jaks et al. 2008; Snippert et al. 2010; Arwert et al. 2012; Page et al. 2013). In the mouse tail, where hair density is much lower, both HF and IFE stem cells are mobilised (Mascré et al. 2012). However, in the Edar mutant mouse, which lacks HF in the tail skin, wound healing is impaired but not absent arguing that HF contribution is non-essential (Langton et al. 2008). Although microdissection of human HF demonstrates evidence of varying ability to regenerate colonies in culture, with cells in the lower follicle below the bulge having the highest proliferative ability, it is not clear whether there is long term HF contribution to IFE re-epithelialisation in vivo in humans (Rochat et al. 1994; Hsu et al. 2011). Eccrine sweat glands are common across the epidermis in humans but limited to the footpad in mice (which is also not complicated by HFs). These glands show little or no signs of homeostatic change, giving rise to the possibility of harbouring a slow cycling stem cell population (Lu et al. 2012). To assess their response upon wounding, Sox9creER/RosaLacZ mice were induced labelling most cells in the sweat glands and ducts (Figure 1.16B). Their foot pads were then scraped, and cohorts followed up to 2 weeks post injury (Lu et al. 2012). This showed that duct cells, but not the gland contribute to wound repair. In human volunteers, a study using 3D reconstruction of human epidermis following CO2 laser generated partial thickness wounds found that the SG duct makes a substantial contribution to reepithelialisation, with coverage rates matching those from hair follicles (Rittié et al. 2013). Although one must be cautious when comparing data for cellular contributions in wound repair involving different wounding protocols and mouse models, these studies do indicate that multiple populations of cells outside the IFE contribute to IFE repair, both in the short- and long-term. The mouse IFE is maintained by progenitors, 38 39 but upon wounding can receive contributions from stem populations both resident and non-resident in the IFE. The observation that clones from non-IFE stem cells persist long-term suggest that they are not lineage-restricted, with the environment rather than cell ancestry determining cell behaviour in vivo. 1.6.3 Suprabasal Keratinocytes Contribute to IFE Wound Repair Tritiated thymidine labelling of tape stripped mouse epidermis reveals widespread suprabasal division, which is normally absent during homeostasis (Potten et al. 2000). Additionally, calcium induced differentiated mouse keratinocytes expressing involucrin when transplanted in vivo have been shown to regenerate the multiple lineages of skin epithelia (Mannik et al. 2010). Similarly, an observational study in human volunteers found that suprabasal cells adjacent to a punch biopsy wound expressed the proliferating marker Ki67, and expressed keratin 6, an inducible keratin associated with hyperproliferation (Patel et al. 2006). Taken together, these experiments provide indirect evidence that commitment to differentiation does not prohibit cells from re-entering the cell cycle, de-differentiating and contributing to regenerating epithelium. It is unclear whether contributions from such cells are long- lived. 1.6.4 Diverse Cell Types Contribute to Repair in Other Epithelia Lineage-specific amplification by defined stem cell populations is becoming recognised as a general mechanism for tissue replenishment in homeostasis. Examples from the mouse mammary gland and prostate show luminal and myoepithelial compartments are maintained independently throughout life (Ousset et al. 2012; Choi et al. 2012; Van Keymeulen et al. 2011). Similarly, the mouse skin sweat glands and upper pilosebaceous units are maintained independent of each other during homeostasis (Lu et al. 2012; Page et al. 2013). However, previous sections have discussed how this lineage specificity changes upon wounding in the skin allowing wider recruitment of cell types than those within that specific compartment. Here I consider the diverse cell types contributing to re-epithelialisation in the stomach, intestine and trachea indicating stem cell-independent contribution to repair in these tissues. The epithelium of the mouse gastric corpus is renewed almost exclusively by cells in the isthmus of the gland, with cells migrating bidirectionally to the pit and base (Karam & Leblond 1993). A subset of differentiated Chief cells at the base of the gland can, however, occasionally repopulate the entire gland (Stange et al. 2013). A subset of 40 41 these Chief cells expressing the stem cell marker Troy could however repopulate the entire gland, upon selective destruction of the proliferative isthmus cells by 5- fluoruracil. This reversion of a differentiated cell to regenerate tissue long term is indication that “reserve” populations of cells may take up regenerative function. Genetic lineage tracing of the airway basal cells have been shown to self-renew and differentiate in to multiple airway epithelial cell types (Rock et al. 2009). Differentiated luminal secretory club cells have both secretory function, and can dedifferentiate in to ciliated cells (Rawlins et al. 2009). It has recently been shown that labelling secretory cells en masse, and then specifically killing 80% of basal cells with inhaled doxycycline resulted in labelled secretory cells capable of producing both secretory and ciliated cells of the luminal lining (Tata et al. 2013). This suggests that the club cells had dedifferentiated to take this role. Although a high frequency induction schedule does not rule out some contribution from inadvertently labelled remnant basal cells, single secretory cells in culture were shown to dedifferentiate to form multiple lineages. Thus the differentiated club cells are seen to indirectly replenish multiple lineages of the trachea by regenerating basal stem cells. In the intestinal crypt, the Wnt target gene Lgr5 expressing cells generate long- lived clones containing all four epithelial cell lineages (Barker et al. 2007). Lgr5+ stem cells are restricted to the niche of the lower crypt requiring Wnt secreted by the adjacent Paneth cells (Sato et al. 2011). Efficient transgenic deletion of the Lgr5+ cells results in only subtle changes including doubling of enteroendocrine lineage cells and migration of Paneth cells out of the crypt (Tian et al. 2011). This Lgr5+ independent repopulation included Bmi1+ cells, which has been suggested as a stem cell marker at the +4 position, but is expressed in a wider lineage including differentiating progenitors generating secretory cells including enteroendocrine cells, and differentiated Paneth cells after injury (Sangiorgi & Capecchi 2008; van Es et al. 2012; Roth et al. 2012). Thus, regeneration following wounding in the intestine too recruits multiple lineages for repair. The ability to track the fate of cells in epithelia using lineage tracing is providing new insights in to the contributions provided by different cell types. These contributions appear to be different in diverse mouse epithelia. It remains to be seen if humans exhibit similar principles of lineage plasticity in wound healing. 42 43 1.7 Objectives of Research Presented The aim of my research was to use the power of live imaging to resolve cell fate choices at a single cell level in human keratinocytes in vitro. I began my work looking at live imaging and lineage tree reconstructions across an unselected sample of human neonatal and adult keratinocytes (Chapter 3). I used directly observed cell fate choices to identify behaviour. I then analysed data to observe links between cell fate choices and location within the growing colony of cells. Additionally, cell fate choices of related pairs of cells were analysed to address the question if individual cell division affected neighbouring cells. Using novel mathematical approaches, I then looked at colony size distributions at 12 days of culture to understand the basis of the three clonal morphologies seen in the literature. Using larger cohorts of dye-labelled and unlabelled cells, I then describe and confirm the dynamics seen from live imaging. The identification of early molecular markers for the different observed behaviours was the next key goal. Here, using transcriptional microarrays of 8-cell colonies (data from Dr. Kasumi Murai), I investigated the heterogeneity of expression within early keratinocyte colonies. The next challenge was to identify behavioural changes that occurred following perturbation with growth factors (Chapter 4). I decided to study keratinocyte growth in the absence of supplemented EGF to contrast with those in chapter 2 which had 10ng/mL of supplemented EGF. I used live imaging and colony distribution to contrast effects on growth. I also investigated the effects of EGF on colony area to gain additional insights. Finally, using increasing levels of EGF, and R-spondin at 50ng/mL, I looked at the effects on different population behaviours. Finally, I investigated the cellular response to perturbations in homeostasis with localized wound healing in vivo (Chapter 5). Previous research has shown that the mouse oesophageal epithelium is maintained in homeostasis without the need for a stem cell contribution. However, wound healing is widely thought to recruit stem cell contributions in epithelial tissue. To identify such behaviour in the mouse oesophageal epithelium I designed a novel microendoscopic technique, to investigate the role, if any, that progenitors played in wound healing. Collectively, my experiments aimed to define the cell fate choices made by keratinocytes in standard culture, and in response to perturbation with growth factors in vitro or wound healing in vivo. 44 45 CHAPTER 2 MATERIALS AND METHODS In this chapter, I have described details of all experimental methods or materials used in this dissertation. Section 2.1 is concerned with cell culture methods and primary human material. Section 2.2 describes immunostaining, imaging and cell lysate preparation from these cell cultures. Data analysis, including simulations, is detailed in Section 2.3. To analyse the data from my experiments, I developed novel combinatorial statistical methods, the background theory for which is detailed in Section 2.4. Finally, in section 2.5, I outline the mouse micro-endoscopy technique I developed to address the question of cellular contributions in oesophageal wound healing. 2.1 Cell Culture 2.1.1 Fibroblast Culture and Feeder Layer Preparation 3T3 J2 fibroblasts (Jones laboratory stock, originally gift from J. Rheinwald, Harvard University, USA) (Todaro & Green 1963; Rheinwald & Green 1975) were maintained in Dulbecco’s Modified Eagle Medium (D-MEM) (Invitrogen™ 12320-032) supplemented with 10% donor bovine serum (Invitrogen™ 16030-074) at 37°C in an atmosphere of 5% CO2. Cultures were passaged twice per week by 5 minute treatment with 1X 0.25% trypsin - 3.4mM Ethylene-diamine-tetra acetic acid (EDTA, diluted from Sigma-Aldrich® T4174) to detach cells. Trypsin-EDTA solution was neutralised with culture medium and cells were seeded to new flasks at a dilution of 1 in 6. Cultures were maintained until passage 15-16 before a lower passage was thawed. Feeder layers were prepared by 3 hour incubation with 4!g/ml mitomycin C (Sigma-Aldrich® M4287), followed by washing with phosphate buffered solution (PBS: 10mM Na2HPO4, 1.8mM KH2PO4, 137mM NaCl, 2.7mM KCl at pH 7.4). J2 feeders were then detached by trypsinisation as above and seeded at densities of 2 x 106 cells into 75cm2 flasks (~25,000 cells/cm2) to form a feeder layer prior to addition of keratinocytes. 2.1.2 Human Neonatal Keratinocyte Culture Human neonatal foreskin keratinocytes (NFSKs) (ATCC™ PCS-200-010) were maintained in complete FAD medium (50:50 high glucose D-MEM (Invitrogen™ 46 Table 2.1. Donor details for primary human keratinocytes. All clinical material was collected under NHS ethics committee approval LREC 08/H0308/128. Sample ID Site Donor age (years) Gender TB11.583 Abdomen 24.1 Male TB11.701 Abdomen 45.2 Female TB11.1516 Abdomen 37.3 Female 47 11971-025): D-MEM F12 (Invitrogen™ 31330-038), supplemented with 5!g/ml insulin (Sigma-Aldrich® I5500), 1.8x10-4M adenine (Sigma-Aldrich® A3159), 0.5!g/ml hydrocortisone (Calbiochem® 386698), 1x10-10M cholera toxin (Sigma-Aldrich® C8052), 10ng/ml Epidermal Growth Factor (EGF, PeproTech EC Ltd 100-15) and 5% foetal calf serum (PAA Laboratories A15-041)) with a J2 fibroblast feeder layer as described above (Rheinwald & Green 1975; Rheinwald & Green 1977; Green 1978). Cultures were maintained at 37°C in 5% CO2. Cells were passaged at around 80% confluence and split 1 in 20 following detachment by incubation for 15 minutes in 1X trypsin- EDTA (~8000 cells/cm2). Cells were used between passage 3-7. 2.1.3 Primary Adult Human Keratinocyte Harvest and Culture All primary human tissue was collected from Addenbrooke’s Hospital under the Jones’ group ethical approval 08/H0308/128 from the NHS Cambridgeshire 2 Research Ethics Committee entitled “Modelling cell behaviour in normal and malignant tissues”. Relevant anonymised clinical details of donors are summarised in Table 2.1. Excess human skin from abdominoplasty surgery was harvested as 350-400!m thickness skin grafts with a handheld modified Watson-Braithwaite knife. Skin was decontaminated and washed in PBS thrice. Pieces of skin, approximately 5x5mm, were treated with 5mM EDTA (BDH® 100935V) in PBS for 90-120 minutes to form epidermal wholemounts (Jensen et al., 2010). Wholemounts were physically minced with a scalpel till a fine uniform state. The epidermal mince was treated with 1X trypsin– EDTA for periods up to 30 minutes on a magnetic stirrer at 37°C. Treated cells were passed through a 70µm strainer, before counting on a haemocytometer. Cells were seeded onto a J2 feeder layer and maintained as above for primary human epidermal keratinocytes. 2.1.4 Culture at Clonal Density Mitomycin treated J2 feeder layers were prepared as above at 25,000 cells/cm2. Neonatal or adult keratinocytes were seeded at 25 cells/cm2 and maintained as above for clonal density cultures. For some experiments, the thymidine analogue 5-ethynyl- 2’deoxyuridine (EdU) was added at 10!M concentration for 24 hours prior to fixation to label cells with active DNA synthesis (Invitrogen™ A10044) (Salic & Mitchison 2008). At time points indicated in the text, feeder cells were removed by gentle pipetting, and plates were fixed with 4% Paraformaldehyde (PFA, Fisher Scientific® W0443L) for 5 minutes. The plates were then washed thrice with PBS, and stored in 48 Figure 2.1. Labelling of keratinocytes with PKH26. 2x107 keratinocytes were labelled in a 2µM PKH26 solution at room temperature. 1 2 15 2 x 107 keratinocytes washed in 2 mL media without serum 1 2 15 2 x 107 keratinocytes in 1mL Diluent C Centrifuge 400g x 5 minutes Resuspend pellet in 1mL Diluent C 1 2 15 2 mL of 2µM PKH26 + 2x107 keratinocytes Incubate for 5 minutes room temperature Add to 4µM PKH26 dye solution (1mL) Stop reaction with serum Centrifuge and wash x3 before use Labelled keratinocytes for clonal experiments 49 PBS at 4°C until further analysis. Results presented are in a minimum of triplicated dishes at each timepoint, although often substantially more. Macroscopic colonies at 12 days post seeding were detached from culture plates by 5 minute incubation in 5mM EDTA at 37°C. Colonies were fixed in 4% PFA for 30 minutes and stored in PBS at 4°C until further analysis. 2.1.5 Clonal Cultures for Live Imaging Clonal cultures for live imaging were done using a J2 feeder layer at 15,000 cells/cm2, and keratinocytes at 400 cells/cm2. This was because the imaging was of a fixed central area of the culture plate (3.7% of the total well). A lower density of J2s allowed better visualisation of small keratinocyte colonies that would otherwise be difficult to visualise. The increased keratinocyte density did not lead to much colony fusion, but since the colonies were directly visualised, no errors were introduced. Cultures were performed in ImageLock™ 24 multiwell plates (Fisher Scientific® NC0443043). These plates have fiducial markers at the bottom of the plate which allows accurate referencing of imaging fields for timelapse capture. The markers are on the outer surface of the well, and do not interfere with culture. Unlike other experiments where fresh mitomycin-treated J2 cells were added once a week, only conditioned media changes were done twice a week. Parallel co-cultures of J2 fibroblasts and NFSKs at 2 days of culture provided the conditioned media which was filtered through a 0.22!m membrane filter (Millipore GPWP02500) and diluted 1:1 in complete keratinocyte media prior to use. As before, for some experiments 10!M EdU was added for 24 hours prior to fixation to label cells with active DNA synthesis. At indicated timepoints, feeders were removed and fixation was with 4% PFA as above. 2.1.6 PKH26 Dye Labelling of Keratinocytes The PKH26 Cell Linker Kit (Sigma-Aldrich® PKH26GL) stably incorporates a yellow-orange fluorescent dye with long aliphatic tails (PKH26) into lipid regions of the cell membrane, making it widely used for cell tracking studies in normal and neoplastic tissues (Cicalese et al. 2009; Pece et al. 2010; Wang et al. 2012). Keratinocytes were labelled according to the manufacturer’s instructions (Figure 2.1). A suspension containing 2x106 keratinocytes was washed in PBS to remove serum and resuspended in 1mL of Diluent C (labelling vehicle, iso-osmotic aqueous solution). This cell suspension was added to a PKH26 dye solution to achieve a final dye concentration of 2!M. The mixed suspension was incubated for 5 minutes at room temperature with periodic mixing. Adding serum and further incubation for a minute stopped the 50 Table 2.2. Primary antibodies used for immunoflourescence. Target Supplier Catalogue Clone Host Dilution Active -Catenin Millipore 05-665 8E7 Mouse 1:500 HP1 Cell Signalling #2619 Polyclonal Rabbit 1:100 IRF6 Cell Signalling #6948 Polyclonal Rabbit 1:500 Involucrin Santa Cruz sc-15225 K-19 Goat 1:500 Keratin 1 Covance PRB-149P AF-87 Rabbit 1:500 Keratin 10 Covance PRB-159P Polyclonal Rabbit 1:500 Keratin 14 Covance PRB-155P AF-64 Rabbit 1:500 Keratin 14 Covance SIG-3476 Polyclonal Chicken 1:500 Pancytokeratin Dako Z0622 Polyclonal Rabbit 1:500 SUZ12 Cell Signalling #3737 D39F6 Rabbit 1:100 YY1 Santa Cruz sc-281 C-20 Rabbit 1:200 51 labelling reaction. The cell suspension was then centrifuged and washed thrice in media. Labelled keratinocytes were seeded onto a J2 feeder layer in 50mm plates at clonal density (25 cells/cm2) for tracking experiments. As before, for some experiments 10!M EdU was added for 24 hours prior to fixation to label cells with active DNA synthesis. At indicated timepoints, feeders were removed and fixation was with 4% PFA as above. For clonal isolation experiments, a single clone was seeded onto a J2 feeder layer in half area flat-bottomed 96 well plates (6 cells/cm2) (Corning® 3696). 2.1.7 Cultures with altered EGF and R-Spondin Some keratinocyte cultures as indicated in the text were performed at 0ng/mL, 5ngl/mL, or 20 ng/mL of supplemented EGF (EGF0, EGF5, and EGF20 respectively). No other alterations to the media were made. Recombinant human R-Spondin 1 (R&D Systems® 4645-RS-025) was reconstituted at 100!g/mL in PBS containing 0.1% BSA (Calbiochem 126575). Further dilutions to final experimental concentrations were done in complete media as indicated in the main text. 2.2 Culture Immunostaining, Imaging and Lysates 2.2.1 Immunofluorescence Staining of Culture Plates Plates were permeabilised and blocked with 0.5% Triton X100 (BDH® 28817.295), 0.25% Fish Skin Gelatin (FSG, Sigma-Aldrich® G7765), 1% BSA, and 10% Donkey or Goat serum (Sigma-Aldrich® D9663 or G9023 respectively) in PBS. Primary antibodies (Table 2.2) were incubated in 0.5% Triton, 0.25% FSG and 1% BSA in PBS overnight at 4°C. Plates were then washed three times with PBS for 5 minutes each. Secondary antibodies (Invitrogen™ Molecular Probes®) were used at 1 in 500 in 0.5% Triton, 0.25% FSG and 1% BSA in PBS for 3 hours. Plates were washed as previously, and either mounted in Vectashield® Hardset™ Mounting Medium with 4’,6-diamidino-2- phenylindole (DAPI, Vector labs H-1500) or counterstained with 1:1000 DAPI (Molecular Probes® D1306). 2.2.2 Detection of EdU Incorporated EdU was detected using the Click-iT® EdU Alexa Fluor® Imaging Kits (Invitrogen™ Molecular Probes® 10337-10340) according to the manufacturer’s 52 53 instructions. Plates were permeabilised in 0.5% Triton PBS for 1 hour and washed three times in 3% BSA. The pates were then incubated in the Click-iT® reaction cocktail (containing buffer, CuSO4, Alex Fluor® azide, and buffer additive) for 1 hour at room temperature. Samples were then washed three times for 5 minutes each in 3% BSA. When used in conjunction with standard immunofluorescence as in 2.2.1, the sequence of labelling was: block, primary antibody incubation, washes, post fix 4% PFA 30 minutes, washes, EdU detection protocol, secondary antibody incubation (without Triton), washes (PBS only) and finally mounting. 2.2.3 Microscopy Immunofluorescent staining for two dimensional cultures was imaged on a Zeiss Axio Observer D1 microscope with Zeiss AxioVision® Software. Three dimensional images were acquired with a Zeiss Axioplan 2 confocal microscope with Zeiss LSM510 software, which allowed differential interference contrast (DIC) imaging. 2.2.4 Image Acquisition, Processing and Figure Preparation Images were initially analysed using Zeiss LSM Viewer software. For 3 dimensional analysis confocal z stacks were rendered using Volocity® Visualisation software (Improvision). Figures were prepared using Adobe® Photoshop® and Adobe® Illustrator®. Colony area was measured by the Zeiss AxioVision® software. Clonal counting for large colonies was automated using the ImageJ (1.40g NIH) cell counter plugin. 2.2.5 Cell Lysate Preparation for Western Blots At appropriate times during experimental protocols, whole cell lysates were prepared after washing cell plates twice with PBS at 4°C. Lysis buffer was prepared and added to plates with a constitution of 20mM Hepes-NaOH pH 7.9, Glycerol 10%, NaCl 0.4M, NP-40 0.5%, EDTA 0.2mM, Dithiothreitol (DTT) 1mM, Protease inhibitor, and Phosphatase inhibitor. Cell lysates were scraped from plates, and transferred to pre-chilled 1.5mL tubes and centrifuged at 13000rpm 4°C for 10 minutes. Projected protein concentrations were calculated using standard Bradford protein assays. A dilution series of standard dye reagent and bovine serum albumin (BioRAD QuickSTART™ Bradford Dye Reagents 500-0202) was made, and 1!L of protein extract from experimental or control samples were similarly diluted to the same volume. An Eppendorf® BioPhotometer™ Plus was used to measure absorbance at 600nM Figure 2.2. Deriving lineage trees from timelapse images. Timelapse images were used to reconstruct lineage trees. (A) Single cells were identified at early timepoints. (B) First division recorded in time from seeding. The first division time was not included in any cell cycle analysis. (C) Any subsequent divisions were recorded with parentage noted. (D) Any cell which did not divide within 48 hours were classified as a non-dividing cell. Dividing cells were recorded as green nodes, and non-dividing cells were coloured red. Time bar in hours. 54 32.83 ; 32.83 69.83 ; 37 0 24 48 72 96 120 144 168 day 0; 05:10 day 1; 09:30 day 2; 22:10 day 6; 18:00 3 cell colony lineage tree A B C D 55 wavelength. The absorbance of the standards was plotted against their concentration to obtain the extinction co-efficient. This was used to calculate the concentration of the unknown samples. Remaining cell lysate was mixed 1:1 with loading buffer (2X concentration of Tris-HCl pH6.8, 4% SDS, 20% Glycerol, Bromophenol blue and 0.2% β-mercaptethanol). Samples were heated at 95°C for 5 minutes before storage in pre- chilled 1.5mL tubes at -20°C. Western Blot analysis was done by Dr Kasumi Murai. 2.3 Data Analysis 2.3.1 Comparisons of Populations For all statistical analysis, I have used non-parametric tests not assuming Gaussian distributions. Confidence intervals were computed for 95% confidence and statistical significance was defined using an alpha of 0.05. All computations were done using GraphPad Prism® 6 for Mac. To compare the entire distributions of populations, I have used the Kolmogorov- Smirnov test, which is a nonparametric test to compare unpaired groups of data. It compares the cumulative distribution of the entire dataset, and is therefore sensitive to any differences in the shape, spread or median, unlike the more commonly used Mann-Whitney test which is sensitive to the median alone. To compare paired observations, such as 2-, 3- and 4-cell colonies in various culture conditions, I have used the Wilcoxon matched-pairs signed-rank test To compare unpaired median observations, such as average colony sizes in large keratinocyte colonies, I have used the Mann-Whitney unpaired test. To compare multiple unmatched groups, such as EdU distribution groups within keratinocyte populations, I have used the Kruskal-Wallis test. 2.3.2 Lineage Trees Videos from the Incucyte were initially traced working backwards to identify keratinocyte colonies. Then, working forwards, starting from a single keratinocyte, all divisions were noted in time from the seeding as the separation of daughter cells after mitosis (Figure 2.2). Cells that divided were marked green, while cells which did not divide after a minimum of 48 hours of observation were marked red (non-dividing). When cells did not divide within the observed 48 hours (typically cells in the last 48 Figure 2.3. Lineage trees can be drawn from larger colonies. (A) Lineage trees were drawn from larger colonies in the same manner as Figure 2.4. Here, all cells were traced till ~96 hours. After this time point, it was more practical to sample a few cells (up to 6) to draw sub-colonies within the colony. Cells not followed for >48 hours are shaded grey, and not used in the outcome analysis. (B) Still images at the bottom showing progress of the colony through time. Trees scaled to time were subsequently plotted (not shown here). 56 22.17; 22.17 33.83; 11.67 34.5; 12.33 56.67; 22.83 45.17; 11.33 49.17; 14.67 48.67; 14.17 71.83 15.17 74.83 18.17 60.33 15.17 74.83 29.67 69.17 20 66.17 17 66 17.33 62 13.33 ԇ ӹ ԇ ӹ ԇ ӹ ԇ ӹ ԇ ӹ ԇ ӹ ԇ ӹ ԇ ӹ ԇ ӹ ԇ ӹ ԇ ӹ ԇ ӹ ԇ ӹ ԇ ӹ ԇ 11 4ӹ    ԇ    ӹ    ԇ 11 1.6 7ӹ    ԇ 11 2.6 7ӹ    ԇ 11 1.5 ӹ   ԇ 11 5.8 3ӹ    ԇ 11 3.1 7ӹ  ԇ   ӹ    ԇ 11 4.3 3ӹ    ԇ    ӹ   ԇ   ӹ  ԇ 11 1.5 ӹ  ԇ 11 4.3 3ӹ    ԇ    ӹ    ԇ 11 6.3 3ӹ    ԇ 11 8.3 3ӹ    ԇ    ӹ  ԇ    ӹ    ԇ 11 3.3 3ӹ  ԇ 11 5.1 7ӹ    Time TimeA B 57 hours of recording), they were marked as unknown (grey), and not included in the analysis (Figure 2.3). Lineage trees were then scaled to time according to the observations. Relationships of the dividing cell pairs were also noted. For large colonies between D7-D12, cells were additionally given locational co-ordinates within the colony (Figure 3.7). At any cell division in such data, the distance of the cell from the centre of the colony (bisecting maximal dimensions), and its distance from the edge along the same axis were noted. Cells within the inner third of the colony were classified as “inner” cell divisions. Similarly, “outer” divisions occurred in the outer third of the colony, with the remaining colonies being in the “middle” third. Lineage trees were drawn in Adobe® Illustrator® CS6 for Mac. Groups of division types and relationships were initially built in in Microsoft® Excel® 2011 for Mac, and analysed in GraphPad Prism® 6. 2.3.3 Monte Carlo Simulations Monte Carlo simulations were run in Excel® using a sequential random number generator. Repeated sequences of divisions were repeated up to 21 rounds of divisions in up to 210 divisions per simulated colony. Triplicated sets of 10,000 colonies were used for each dataset analysed. The master set of observed divisions in the relevant Incucyte cohorts were used for sampling. Resulting distributions were plotted in Prism® 6. 2.3.4 Growth Curve Fitting Growth curve fits were done by a least squares approach using Prism® 6. Curves were constrained at starting points, but otherwise unaltered. Goodness of fit was calculated using R2 analysis in Prism® 6. 2.3.5 Figure Preparation All graphs were made in Prism® 6. Publishing layout was finalised in Illustrator® CS6. At cell division Stratification a b c Dividing cell Differentiated cell Stratified cell a = c Figure 2.4. Cell division as a branching birth-death process. Cell division is a birth-death process with three possible outcomes based on the proliferative ability of its daughters. A dividing cell (P) may divide to two dividing daughters (PP), a dividing and differentiated daughter (PD), or two differentiated daughters (DD) in proportions a, b and c respectively. Differentiated cells are lost through stratifica- tion. In tissues with high turnover, the number of new dividing cells is equal to the number lost through stratification (a=c). 58 59 2.4 Combinatorial Statistics Theory In Chapters 3 and 4, I describe a combinatorial approach to understanding population-wide size distribution of colony sizes achieved by single human keratinocytes ex vivo. The conclusions of my studies hinge on the quantitative insights that can be drawn from the analysis of the colony size distribution from single cells. In this section I expand on the basis of my statistical rationale and its application, with further details of applications in Appendix 2. 2.4.1 Assumptions and Parameters Adult mammalian epidermis has a single layer of dividing cells on a basement membrane. As dividing progenitor cells (P) divide, they produce two daughter cells either of which may be a progenitor cell or a differentiated cell (D) resulting in the combinations (PP), (PD) or (DD). We assume the probabilities of these occurring are a, b, and c, respectively represented in Figure 2.4. Assumption 1: I will assume for the moment that cell division is a Markovian process, where the outcome probabilities are independent in time, holding the same value at any observation point. Biologically, this implies that individual cell division is independent, and memory of previous cell divisions is not retained. This assumption is borne out by experimental data in Section 3.4. Assumption 2: I will also assume for the purposes of calculations, that the population averages of a, b, and c are constant under the same growth conditions, but may respond dynamically with changes in external environment or signalling. That is to say that the population likelihood of any particular cell division outcome is the same given the same growth conditions. This is confirmed in the experiments described in Section 3.2 and Section 3.3. Assumption 3: The cell cycle distribution of the keratinocytes must follow a narrow and defined distribution. This allows accurate definition of an average cell cycle to be used in the modelling process. The large cohort of directly observed cell cycles bears this out in Section 3.2. The general approach to calculations of clone size distributions is usually as a function dependent on the time of observation. To remove time from the equations, I take a different approach. The observed frequency of any given clone size at a time point that is far distant from the time taken to reach that size is constant. This implies that at late time-points the number of small clones is constant. In larger clones with persistent progenitor cells, division or differentiation will continue depending on the number of Figure 2.5. Motzkin paths are analogous to cell division choices. On an infinite chessboard, if the king were to be started off at the origin, and have only “forward” moves in the x-axis, the resulting set of moves lies within a defined area. In such a game, the player would be able to move “up”, “flat” forward, or “down”, but never horizontally up or down, or “backward” on the x-axis. These three moves are akin to PP, PD and DD divisions respectively. Note that the player cannot cross the diag- onal, and that if the piece falls below the horizontal, the game ends in “death”. The resulting triangle of possible moves is known in combinatorial statistics as the Motzkin Triangle, and is a description of the probabilities of reaching any available position on the board after a known set of moves. 60 Death Up Flat Down Up Flat Down X-axis -Y -a xis Motzkin Triangle Forward moves 61 dividing cells within the clone. This basic realisation allows me to use the most frequent clone size distribution, which is that of small clone sizes, to build a robust picture of cell fate outcome. Small clone distribution data is the most statistically robust due to the high frequency of observation. 2.4.2 Colony Size Distribution as Motzkin Paths For simplicity, we assume that we start with a single dividing cell. We will denote the total number of cells in a colony as n, with a total number of dividing progenitor cells as p. Thus, any specific colony can be described as (n,p). With each division, irrespective of outcome, the colony size increases by 1 cell, forming a colony of n+1 cells. If the cell division results in two progenitor daughters (PP), the number of dividing cells p increases to p+1. If the cell division results in a progenitor cell and a differentiated cell (PD), the number of dividing cells p stays the same. The production of two differentiated daughters (DD) results in a loss of dividing cells to p-1. Thus, With each cell division, !!!!"#$%&!!! ! + 1, ! + 1 ! "#ℎ!!"#$%$&!"#$!! !, ! !!!!"#$%&!!! ! + 1, ! + 0 ! "#ℎ!!"#$%$&'&()!! !!!!"#$%&!!! ! + 1, ! − 1 ! "#ℎ!!"#$%$&'&()!! Each colony has therefore a defined number of steps it can take at given time that is independent of the time of observation. Such defined steps is analogous to a combinatorial statistics problem, called a Motzkin path (Weisstein 2012b; Motzkin 1948; Aigner 1998; Donaghey & Shapiro 1977). To visualise such a system, imagine an infinite chessboard, where a king-piece was placed on the origin of the lattice (Figure 2.5). With each move, such a piece is allowed only to move one square “forward” along the x-axis in one of three moves – an “up” move diagonally upwards the y-axis, a “flat” move horizontally, or a “down” move diagonally downwards the y-axis. Any move that results in falling below the horizontal is “dead” and ends the game. In our analogy, the chess piece is a dividing cell, and the number of total cells n and the number of dividing cells p are the x- y-axis respectively. A “up” move is a PP division resulting in adding a P cell to the colony, a “flat” move maintains the number of P cell, while a “down” move reduces the number of P cells in colony, and in effect changes the proliferative potential of the colony overall. In effect, the progress of the colony is limited between the upwards diagonal and the x-axis, resulting in a triangle of paths Figure 2.6. Cell proliferation as a combinatorial branching process. (A) Cell divi- sion in progenitor cells is a branching process with three possible outcomes (PP, PD or DD, Figure 2.4). The expansion of a single cell to form a colony of cells is thus a combinatorial process, where any outcome of total colony size n and proliferating cells within it p occurs along fixed paths of a Motzkin triangle. Colonies that fall below the horizontal are dead. (B) Example showing the 9 paths that a single prolif- erating cell can take to reach a clone of n=5 and p=2. The first three routes have three PP divisions and one DD division, while the remaining six routes involve two each of PP and PD divisions (cumulative probability =3a3c+6a2b2). 62 b c a b c a b c a b c a b c a b c a b c a b c a b c a b c a b c a b c a b c a b c a b c a N um be r o f d iv id in g p ro ge ni to rs, p 4 3 2 1 5 Number of total cells, n 432 5 A B N um be r o f d iv id in g p ro ge ni to rs, p Number of total cells, n 3 3 3 5 5 5 Proliferating progenitor cell Non-proliferating differentiated cell 63 called the Motzkin triangle (Kociemba 2004). Defined in terms of probability, if !(!,!) represents the probability of reaching any point on the clone size of n with dividing progenitors of p, then it follows from the previous equation: ! !,! != !! !!!,!!! !+ !! !!!,! !+ !! !!!,!!! where, n≥2 and n≥p≥1 This implies that the number of ways in which a cell can reach any given position on the triangle is a sum of the number of ways the piece can reach the three preceding possible positions. Any position on the Motzkin triangle can also be defined as a combination of all the possible paths taken to reach that position (example illustration in Figure 2.6), and is given by the equation (Kociemba 2004): !(!,!) = !!!!!!!!!(!!!)/!!!! !!!!!!! − !!!!!!!!! For any given path to reach a colony of total n cells and p dividing cells, the following hold: ∆!! = ! − 1 = !! − !! ∆! = ! − 1 = !! + !! + !! where ∆!=change in number of dividing cells, ∆!=change in total number of cells, !!=total number of up steps, !!=total number of flat steps, and !!=total number of down steps. Thus, for any given path, if !! = !, !! = ! + ! − 1, and !! = ! − ! − 2! Thus it follows that the weighted probabilities of any colony is given by: !(!,!) = !!!!!!!!!(!!!)/!!!! !!!!!!! − !!!!!!!!! !!!!!!!!!!!!!!! This implies that for the first few colony sizes, the probability is given by: !(!,!) = !! = ! ! !,! = !! = !" !(!,!) = !! = !!! + !!! !(!,!) = !! = 3!"!! + !!! … Figure 2.7. Small clone sizes alone can define proliferative behaviour. We have two proliferative cell populations with progeny outcomes a1/b1/c1 and a2/b2/c2 respectively. These are mixed in the ratio x:(1-x), where x>>(1-x). If a2>>c2, the contribution of population 2 to the small colony sizes in the mixed population is negligible. The small clone size distribution in the mixed population therefore is related to a1/b1/c1 and x. Although x is unknown, a1/b1/c1 can be inferred from the relative ratios of the small clones in the mixed population. If C2=clones of size 2, and C3=clones of size 3, b1=C3/C2. Similar derivations for a1 and c1 can be made as a1+b1+c1=1. 64 C2=x ∙ (c1) C3=x ∙ (b1c1) C4=x ∙ (a1c12+b12c1) a1=c1 Proportion x Cell Population 1 a2>>c2 Proportion (1-x) Cell Population 2 Mixed Cell Population 65 We consider a few special cases of biological significance: Special Case 1 For the case where there are no dividing cells remaining in the colony, the colony must transit through a stage with only one dividing cell remaining, and undergo a enforced final DD division, the probability of which can be given by: ! !!!,! = ! ∙ ! !,! where, !(!,!) = !!!!!(!!!)/!!!! !!! − !!!!! !!!!!!!!!!! Special Case 2 For the case where the probabilities a, b, and c are equal, implying that the process has no weightage, the probability ℰ(!,!)!simplifies to: ℰ(!,!) = !!!!!!!!!(!!!)/!!!! !!!!!!! − !!!!!!!!! !! (!!!) Special Case 3 For the case where all cells in the colony remain dividing (n=p), the probability !(!,!) reduces to: ! !,! = ! !!! In a cell with no death process (a theoretical stem cell) a=1, this would indicate absolute probability of continued division. 2.4.3 Small Colony Sizes Can Define Proliferative Behaviour If we were to have two cell populations mixed together, we have an additional variable to consider, the proportion in which the two populations are mixed (Figure 2.7). If this proportion in which the two populations are mixed is unknown we need additional insight to compute the proliferative characteristics a, b, and c. Thus, Population 1 Population 2 Relative Proportion (Proportion x) (Proportion (1-x) 2 cell colonies !! ∙ ! !! ∙ (1 − !) 3 cell colonies !!!! ∙ ! !!!! ∙ (1 − !) 4 cell colonies !!!!! + !!!!! ∙ ! !!!!! + !!!!! ∙ (1 − !) Figure 2.8. Gambler’s ruin predicts the proportion of “everlasting” clones. If weighted Motzkin paths continue indefinitely, the relative weights can inform us of the clone size distribution. (A) If a clone has the relative probability of a=1, all divi- sions occur on the diagonal with exponential increase in cell numbers. (B) If the clone has a relative probability of c=1, the clone is destined to be ruined and form a 2 cell differentiated colony. (C) If a=c, the colony will be of finite size (n-1)/2 and contain a maximum of either 1 dividing cell (if n=even) or no dividing cell (if n=odd). There will be no “everlasting” clones. (D) If ac, there will be a proportion of cells given by 1-(c/a) that are everlasting. 66 N um be r o f d iv id in g c ell s, p Total number of cells, n a=1 c=1 a=cac A B CD E 67 If population 2 is rare, that is (1 − !) is small, and it is biased towards proliferation, that is !! is small, the contribution that population 2 makes to small colony sizes is negligible. We can thus assume that all the very small colonies (<5 cells) resulting from a mixed ex vivo culture occur through population 1. We still have to resolve the proportion x. We can use the ratios between the proportions of small colonies to eliminate this unknown variable x. Since the values !!,!!,!!… are known, we can compute a, b, and c. Thus, !! = !! ∙ !!! = !! ∙ !! !! = !!!! !! = !!!!!! + !!!!!! = !! ∙ !! + !!!! ∙ !! and, recalling that !! + !! + !! = 1 !! = !!!! + !!! ∙ !!! !!!! = !!!! + !!! !!!! = !! ∙ 1 − !! − !!!! + !!!! ! !!! + !!!! − 1 !! + !!!! − !!!! ! = 0 !! = −(!!!! − 1) ± (!!!! − 1)! − 4 !!!! − !!!! ! 2 ! 2.4.4 A Game of Gambler’s Ruin Additional insight can be gained from the comparison with another statistical problem known as the Gambler’s Ruin (Weisstein 2012a; Lengyel 2009; Kraitchik 1942). If a gambler is playing to win, draw or loose on each bet, one can work out the probability of winning overall in n games if prior odds of winning, drawing and loosing are known and constant. If the chances of winning an individual game are equal to the chances of loosing the game, on average the player will not make any gain. If the chances of winning are higher than the chances of loosing, the gambler will 68 69 be better off. If however, the chances of loosing are higher than the chances of winning, the player will loose the initial investment and be “ruined”. In our analogy, if for any given clone size n, knowing the relationships between a, b, and c can tell us the chances of proliferative success. If the chances of winning are higher (! > !), the some colonies will continue to divide indefinitely. If the chances of differentiation are higher (! < !), there can be no indefinitely dividing colonies, and all the colonies will eventually differentiate. If exact odds are maintained (! = !), colonies of even number will be left with a single dividing cell (no gain or loss), while colonies of odd number will be completely differentiated (Figure 2.8). 2.5 Mouse Oesophagus Experiments All experiments were conducted according to the UK Home Office Project Licence 80/2282 entitled “Stem and Progenitor cell fate regulation in vivo”. 2.5.1 Mouse Strains Adult mice doubly transgenic for the inducible cre allele AhcreERT and the conditional reporter allele of EYFP targeted to the Rosa 26 locus (AhcreERTR26flEYFP/wt) were generated for lineage tracing in wounded oesophagus (Srinivas et al. 2001; Kemp et al. 2004). For label retaining studies, Rosa26MrtTA/TetO-HGFP mice doubly transgenic for a reverse tetracycline-controlled transactivator (rtTA-M2) targeted to the Rosa 26 locus and a HIST1H2BJ/EGFP fusion protein (HGFP) expressed from a tetracycline promoter element was used (Kanda et al. 1998; Tumbar et al. 2004; Hochedlinger et al. 2005). For short term cell proliferation assays using EdU, wild type C57BL/6 mice were used. 2.5.2 Induction of Cre Mediated Recombinase In AhcreERTR26flEYFP/wt mice, transcription of the cre mutant oestrogen receptor fusion protein (creERT) is induced by β-napthoflavone (β-NF). Tamoxifen is also required for creERT to gain access to the nucleus and excise the loxP flanked “STOP” cassette resulting in EYFP expression. EYFP expression was induced in animals aged 8-12 weeks by an intraperitoneal dose of 80mg/kg β-NF and 1mg Tamoxifen (Kemp et al. 2004; Clayton et al. 2007; Doupé et al. 2012). β-NF (Sigma N3633) was dissolved in Figure 2.9. Mouse oesophageal endoscopic biopsies. (A) Microendoscope with biopsy forceps emerging from the instrument channel and jaws open to show relative size. (B) Endoscopic image of oesophageal inlet showing 1.5 cm mark on biopsy forceps for mid oesophageal wounding. (C) Stereomicroscopic image show- ing wound at 24 hours. (D) Healed wound at 5 days. Scale bar 0.5 mm. 70 A B C D Oesophageal inlet Vocal chords 1.5 cm biopsy gauge 24 hours 5 days 71 autoclaved sunflower oil at a final concentration of 8mg/mL. Tamoxifen (MP Biomedicals 02156738) was dissolved in autoclaved corn oil at 10mg/ml. 2.5.3 Induction of HGFP Expression In Rosa26MrtTA/TetO-HGFP mice, HGFP expression was induced by treatment with 2mg/ml doxycycline (Sigma-Aldrich D9891) in drinking water sweetened with sucrose (Sigma-Aldrich S-9378) for 4 weeks, after which doxycycline was withdrawn for up to 4 weeks to chase HGFP dilution. 2.5.4 Labelling of S-Phase with EdU EdU given as a single 0.1mg (100µl of 1mg/ml) dose administered by intraperitoneal injection at times indicated in the main text. 2.5.5 Eosophageal Microendoscopy To perform the endoscopy on mice anaesthesia was given using a combination of 100mg/kg Ketamine (Pfizer Animal Health) and 10mg/kg Xylazine (Bayer HealthCare AG) administered intraperitoneally. I used a 9.5 Fr 30° forward oblique diagnostic miniature endoscope with a 3 Fr instrument channel in conjunction with an AIDA® COM II image capture system for high definition video recording (Karl Storz GmBH) (Figure 2.9). To cannulate the oesophageal inlet, I placed the mouse on its dorsal surface, and extended the neck to allow straight line passage of the endoscope under direct visualisation up to the vocal chords. Endoscopy of the mouse differed significantly from the human in the flexibility of the neck and voicebox, and therefore direct visualisation was very important to ensure atraumatic cannulation. I then used a 3Fr diameter biopsy forceps with double action jaws (Karl Storz GmBH) to create single superficial wound in the middle third of the mouse oesophagus, approximately 1.5cm from the vocal chords, of between 0-4-0.9mm diameter. For post-procedure analgesia, 0.1mg/kg of Buprenorphine (Alstoe Animal Health) was given subcutaneously, before reversal of anaesthesia using Atipamezole (Pfizer Animal Health) given at 1mg/kg subcutaneously at least 20 minutes after induction. Post- procedure, animals were maintained on a soft mashed diet and euthanised at appropriate timepoints mentioned in the text. 72 73 2.5.6 Wholemount Preparation Wholemounts of the oesophagus were prepared by harvesting the entire oesophagus and making an anterior slit to avoid the wound on the posterior surface. The entire oesophagus was incubated in 5mM EDTA for 2-3 hours at 37°C. The epithelium was then carefully peeled away from the underlying tissue with fine forceps and fixed in 4% PFA for 15-25 minutes. Wounded oesophageal wholemount images were taken with a Leica M205 stereomicroscope. Wholemount immunofluorescence staining and imaging were done by Dr Maria Alcolea. 2.5.7 Wholemount Imaging To investigate the intensity of HGFP fluorescence in wounded oesophageal epithelium from Rosa26M2rtTA/TetO-HGFP mice, samples were analysed at the following time points after 4 weeks of Doxycycline treatment: 0 hrs, 3 hrs, 6 hrs, 12 hrs, 24 hrs, 48 hrs, 72 hrs, 96 hrs, 168 hrs, 10 days and 14 days. Unstained wholemounts of the entire oesophagus were imaged on a Leica SP5 II confocal microscope using identical settings for all samples. Single 2!m slice images of the basal layer were analysed for quantitative assessment of HGFP fluorescence in each independent cell using Volocity 5 (Improvision). 74 Figure 3.1 Live tracking of clonal density neonatal keratinocytes. (A-F) Single keratinocytes can be tracked as they divide in co-culture with 3T3 J2 fibroblasts using the IncucyteTM live imaging system. Daily snapshots are shown at 24 hour intervals showing auto-focussed high definition images with single cell resolution. (G) Population colony sizes showing no change in growth distribution with differ- ent culture environment (standard incubator vs. IncucyteTM). Plot as a 1-99% box and whisker plot, p=0.15 Kolmogorov-Smirnov test, IncucyteTM n=81, Bulk=1487. (H) Directly observed cell cycle distribution shows a median cell cycle of 15.67 hrs (range 4.67-100.17 hrs, n=2127). 99.2% of all cell divisions occur within 48 hrs. (I) DD divisions are longer than PP or PD divisions. (p=0.007, Kolmogorov-Smirnov test, PP n=1109, PD n=338, DD n=330, PP vs. PD, p=0.28). A B C D E F 1 4 16 64 256 1024 Co lo ny Si ze s Incucyte tracking does not affect colony size distributions Cultured Neonatal Keratinocytes BulkIncucyte p = ns G Almost all cell divisions occur within 48 hours 24 48 72 96 Av era ge 0 4 8 12 Time (hrs) Pe rce nt ag e H 0 5 10 15 20 Division outcome Tim e (h ou rs) PP PD DD I p = ns p = ** DD divisions are significantly longer 75 CHAPTER 3 HUMAN KERATINOCYTES HAVE TWO GROWTH DYNAMICS IN VITRO 3.1 Chapter Overview Growth of individual human keratinocytes in culture have been used to gain insight in to their proliferative capacity (Barrandon & Green 1987b; Jones & Watt 1993). These cultures contain a mixture of proliferating and post-mitotic differentiating keratinocytes. The description of three morphological colony types, with a wide range of colony sizes, suggests a diversity of proliferative behaviour. In this chapter I set out experiments conducted to examine human keratinocyte divisions in detail to investigate these appearances. Section 3.2 describes proliferative behaviour from live tracking of human neonatal foreskin keratinocyte cultures that identifies two distinct modes of cell division. Section 3.3 uses the same methodology to study primary adult keratinocytes, and finds the same two modes of division. Section 3.4 analyses lineage data from previous two sections to argue that cell division outcomes are independent of neighbouring cells. Section 3.5 makes use of these insights in to the outcome of individual cell divisions to demonstrate that two modes of cell division are sufficient to achieve the heterogeneity of colony sizes seen in culture. To verify these findings, I present data from two additional datasets. Section 3.6 tracks dye labelled keratinocytes to confirm the existence of two modes of growth in culture. Section 3.7 uses the distribution of small colonies to deduce the proliferation characteristics of one of the two modes of growth. Section 3.8 looks at the distribution of differentiated and dividing cells at (60 hours) to confirm proliferative behaviour. Section 3.9 investigates transcriptional signatures for these two behaviours, validating microarrays done by Dr Kasumi Murai. Finally, in Section 3.10 I discuss a few key implications of my findings in this chapter. 3.2 Live Imaging of Cultured Human Neonatal Keratinocytes To identify the types of proliferative behaviour in human keratinocytes, I performed live imaging of fourth to sixth passage single unlabelled neonatal foreskin keratinocytes (NFSKs) co-cultured at clonal density with mitotically inactivated 3T3 J2 fibroblasts (Rheinwald & Green 1975). High definition brightfield images were taken 76 Figure 3.2 Type 1 colonies in neonatal keratinocytes. Type 1 colonies (all data in Figure 3.3 overleaf) are characterised by frequent and evenly distrib- uted DD divisions. The proportions of PP: DD are similar, with an overlap- ping range. PP n=353, PD n=259, DD n=318, 95% confidence interval in brackets. Type 1 colony growth dynamics a ǀc 0.34 (0.31-0.37) 0.28 (0.25-0.31) 0.38 (0.35-0.41) Proportion (95% CI) 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 Figure 3.3. Lineage trees of neonatal keratinocyte Type 1 clonal growth. Single keratinocytes colonies tracked up to 168 hours. Green nodes are dividing cells. Red nodes do not divide within at least 48 hours of birth. Note similar red and green numbers. Grey nodes are undivided within observation period less than 48 hours. Divisions n=1052, colonies n=70. 77-78 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 Figure 3.5. Lineage trees of neonatal keratinocyte Type 2 clonal growth. Single keratinocytes colonies tracked up to 168 hours. Colour code as in Fig 3.3. Note overwhelming green cells, and only PP divisions within first three rounds of division. Divisions n=1156, colonies n=11. 80-81 79 Figure 3.4 Type 2 colonies in neonatal keratinocytes. Type 2 colonies (all data in Figure 3.5 overleaf) are characterised by predominant PP divisions. PP n=819, PD n=91, DD n=18, 95% confidence interval in brackets. Type 2 colony growth dynamics a >> c 0.02 (0.01-0.03) 0.10 (0.08-0.12) 0.88 (0.86-0.90) Proportion (95% CI) 82 Figure 3.6. Similar cell cycles across division types. (A) Cell cycle distribution of keratinocyte clones does not differ between Type 1 and Type 2 growth patterns for the same division outcome. ( PP divisions p=0.1 n=301 & n=808, PD divisions p=0.6 n=247 & n=91, DD divisions p=0.3 p=312 & n=18; Kolmogorov-Smirnov test, box & whisker plot with 1% outliers). (B&C) Immunofluorescence images of colonies at 168 hours. 25-cell colony typically produced through Group 1 growth dynamics composed of non-dividing cells (Keratin 1 differentiation marker, EdU prolifera- tion marker), while an illustrative 1196 cell colony formed by Type 2 dynamics had predominantly EdU positive cells, with a few central Keratin 1 positive cells. Scale bar=50µm. Cell cycle distribution is similar in Type 1 & Type 2 Type 1 Type 1 Type 1Type 2 Type 2 Type 2 PP Divisions PD Divisions DD Divisions 0 24 48 72 Ti m e ( ho ur s) p = ns p = ns p = ns A B K14 DAPI K1 EdU Type 1 25 cells K1 EdUK14 DAPI Type 2 1196 cells C 83 at 10-minute intervals for at least 168 hours (and up to 278 hours) using the Incucyte™ Kinetic Imaging System (Essen BioScience Inc.). Small colonies at time points longer than 14 days are lost from the culture plate to the media (Jones & Watt 1993). Static images were stitched to form videos from which keratinocyte lineage trees were reconstructed (Figure 3.1A-F). The average exposure time per frame was 1.07 ms, resulting in a cumulative exposure time of 1.08 s over 168 hours, thus minimising light toxicity. A fixed area at the centre of the culture well totaling 3.7% of the total seeded area was recorded, and all clones that remained in the field of vision throughout the observation period were included in the analysis. There was no difference in clone distribution comparing the imaged cells with those cultures in a standard incubator (Figure 3.1G). At the end of the experiment, colonies were fixed and stained for proliferation and differentiation markers. 2208 complete cell cycles were observed in 81 colonies. Excluding the first cell cycle, to account for any plating lag, the median cell cycle length was 15.67 hrs (range 4.67-100.17 hrs, n=2127, Figure 3.1H). As this experiment was done on live cells, I needed criteria to identify post-mitotic differentiating cells. I observed that 99.2% of all keratinocyte divisions occurred within 48 hours. All cells that were observed for a minimum of 48 hours and did not divide were hence classified as post-mitotic. Cells observed for time periods less than 48 hours, and were classified as “Unknown” and were not included in the analysis (exclusions n=228). Colonies did move within the field of observation, but rarely moved more than 400!m from their original seeding point. Any colony that migrated outside the field of view was not included in the analysis. Rare colony fragmentation was observed (2 colonies, 2.46%), as was colony fusion (2 cases, 2.46%). These were late events, occurring at time-points >144 hours. It was still possible to resolve individual cell outcomes within such colonies. These events were therefore included in the data. Three types of cell division outcomes were noted based on the proliferative ability of the two daughter cells. A proliferating cell (P) could produce two dividing daughters (a PP outcome), a dividing and a non-dividing daughter (a PD outcome), or two non-dividing daughters (a DD outcome). The median cell cycle time for each of these types of division were 15.17 hrs (range 4.67-68 hrs, n=1109), 15.00 hrs (range 6.16-60.83 hrs, n=338), and 15.33 hrs (range 5.34-77.16 hrs, n=330) respectively. There was no statistical difference between the cell cycle time for PP or PD outcomes (non- parametric unpaired comparisons of cumulative distribution using Kolmogorov- Smirnov test p=0.28) (Figure 3.1I). However, the DD cell cycle is significantly longer (Kolmogorov-Smirnov test p=0.007) (Figure 3.1I). 84 Figure 3.7 Cell fate changes in the middle of large colonies. (A) Cell divisions within large colonies with type 2 cell fate were tracked between 7-12 days. Cells were classified as being within the inner third or outer two thirds based on area. Example clone shown on 9th day with maximum diameter and perpendicular determining centre of the colony. Scale=500µm. (B) Divisions within the inner third of the colony have type 1 cell fate with widespread DD outcomes, unlike divisions in the outer two thirds which have type 2 cell fate, consisting of predominantly PP outcomes. Inner third n=77 divisions in 17 sub colo- nies, Outer two thirds n=104 in 4 sub colonies. 168 192 216 240 264 288 168 192 216 240 264 288 168 192 216 240 264 288 168 192 216 240 264 288 Inner third Outer third 9d 09:40 Sub-colonies within the inner third Sub-colonies within the outer two thirds A B 85 Lineage trees were drawn for all colonies, which ranged in cell number from 2 cell clones to 722 cells (Figure 3.3 and 3.5). In large colonies, it was not possible to manually track all dividing cells at late time points. In these cases, I tracked all cells up to ~96 hours, and then randomly chose up to 6 cells from with in the clone to track to 168 hours. Up to 20 cell generations were tracked in this manner. Based on the frequencies of the three division types, PP:PD:DD, two distinct types of colonies emerged (Figure 3.2 and 3.4). In type 1 colonies (86.4%, n=70) the frequency of PP divisions approximately equals DD divisions (! ≈ !; 0.38 and 0.34 respectively, n=353 and n=318, respectively). These colonies are usually small in final size, and in the rare cases that the total colony size reaches 10s of cells, the majority of the colony consists of non-dividing cells. In contrast, Type 2 colonies (13.5%, n=11) divide exclusively by PP divisions for the first three rounds of division and overall most divisions produce PP progeny (a=0.88, n=819), with occasional PD divisions (b=0.10, n=91) and rare DD divisions (c=0.02, n=18) (Figure 3.4). There is no difference in cell cycle length between the two types of colonies (Figure 3.6A). Type 2 division dynamics formed colonies typically in the hundreds of cells at 168 hours, although rare colonies may remain less that this size. Immunofluorescence of colonies fixed and stained after 168 hours showed that type 1 colonies had a majority of differentiated cells (Keratin 1 positive, EdU negative), while type 2 colonies have widespread EdU uptake and patchy Keratin 1 positivity in stratified cells (Figure 3.6B&C). It was not possible to accurately match each photographed cell to immunofluorescence image due to frame shift algorithms within the Incucyte™ software. To further investigate whether the near exponential growth rate in Type 2 colonies is maintained, I looked at the videos of such colonies between 7 days to 12 days. As it was not possible to track all the cells within such a large colony, a sample of consecutively observed divisions was chosen. Lineage trees were created in a similar fashion to previous colonies, with additional data on the location within the clone of the cell that was tracked. At any observation point the colony was divided in to thirds based on area. Keratinocytes were assigned to belong to the “inner”, “middle” or “outer” third of the colony (Figure 3.7A). 21 sub-colonies were reconstructed in this manner with 181 observed cell divisions (Figure 3.7B). A clear difference was noted between sub-colonies in the inner third of the colony versus the outer two thirds. In the inner third of the colonies cell fate appeared similar to that in type 1 colonies with 29% of divisions resulting in a DD outcome. On the other hand, divisions in the outer two thirds of the colonies continued with the type 2 cell fate resulting in PP divisions 86 Figure 3.8. Live tracking of clonal density adult keratinocytes. (A-D) Divisions in colonies from single primary human keratinocytes are recorded in the same manner as Section 3.2. (A) Entire population colony sizes showing no change in growth distribution with different culture environment (standard incubator vs. IncucyteTM). Plot as a 1-99% box and whisker plot, p=0.052 Kolmogorov-Smirnov test, IncucyteTM n=56, Bulk=627. (B) Directly observed cell cycle distribution shows a median cell cycle of 15.0 hrs (range 5.5-62.33 hrs, n=1100). 99.6% of all cell divi- sions occur within 48 hrs. (C) DD divisions are longer than PP or PD divisions (Kol- mogorov-Smirnov test p<0.0001). PD divisions are longer than PP divisions (Kol- mogorov-Smirnov test p=0.024). PP n=724, PD n=207, DD n=169. (D) Cell cycle distribution of keratinocyte clones does not differ between Type 1 and Type 2 growth patterns for the same division outcome. (PP divisions p=0.2 n=187 & n=537, PD divisions p=0.15 n=155 & n=52, DD divisions p=0.3 p=155 & n=14; Kolmogor- ov-Smirnov test, box & whisker plot with 1% outliers). 1 4 16 64 256 1024 Co lo ny Si ze s Incucyte tracking does not affect colony size distributions Primary Adult Keratinocytes BulkIncucyte p = ns Almost all cell divisions occur within 48 hours 24 48 72 96 Av era ge 0 4 8 12 Time (hours) Pe rce nt ag e A B C DDD divisions are significantly longer PP PD DD 0 5 10 15 20 Division outcome Tim e (h ou rs) p = * p = **** 0 24 48 72 Tim e ( ho ur s) Cell cycle distribution is similar in Group 1 & Group 2 p = ns p = ns p = ns Gro up 1 Gro up 1 Gro up 1 Gro up 2 Gro up 2 Gro up 2 PP Divisions PD Divisions DD Divisions 87 in 95% of divisions. This demonstrates that cell fate changes within the central area of large colonies to form increasing numbers of differentiated cells, while outer rim cells continue proliferating in a near-exponential manner. 3.3 Live Imaging of Primary Human Adult Keratinocytes To see whether first passage human keratinocytes exhibited the same growth dynamics, I performed the live tracking experiment with identical methodology as the previous section on independent primary keratinocyte cultures from three adults ranging in age from 24-45 years (clinical details of donors in Chapter 2). There was no difference to colony size distribution due to culturing within the Incucyte™ machine versus a standard incubator (Figure 3.8A). Additionally, no colony fragmentation or fusion was seen. Here, I followed 1156 keratinocyte cell divisions over 56 colonies. Excluding the first cell cycle, the median cell cycle length was 15.0 hrs (range 5.5-62.33 hrs, Figure 3.8B). I used a 48 hour cut-off to identify post-mitotic cells, as in the previous section. 99.6% of all divisions occurred within this 48 hour window. With the benefit of experience from section 3.2, all cells were followed for a minimum of 48 hours, and therefore there are no cells of “unknown” growth potential, with the video capture lasting till 216 hours. All three proliferation outcomes observed with NFSKs were seen (PP, PD, and DD). The median cell cycle times for each of these divisions were 14.5 hrs (range 5.5- 33.5 hrs, n=724), 15.67 hrs (range 7.83-56.17 hrs, n=207), and 17.33 hrs (range 7.5-62.33 hrs, n=169) respectively. Unlike neonatal foreskin, there was a progressive statistically significant lengthening of cell cycle between division types. Divisions with PD outcomes were longer than those with a PP outcome (p=0.024, Kolmogorov-Smirnov test) (Figure 3.8C), and divisions with DD outcome were significantly longer than both PD and DD outcomes (p<0.0001, Kolmogorov-Smirnov test) (Figure 3.8C). There was no difference in cell cycle distribution between the two modes of growth (Figure 3.8D). Lineage trees were drawn in a similar fashion to Section 3.2 (Figure 3.10 and 3.12). The colony sizes varied from 2 cells to 876 cells. Similar dynamics are seen to that in neonatal foreskin cultured keratinocytes, with the distinct emergence of two types of growth dynamics. Type 1 colonies (89.3%, n=50) have ! ≈ ! (0.38 and 0.31 respectively, n=202 and n=173) while Type 2 colonies have predominantly PP 88 Figure 3.9 Type 1 colonies in adult keratinocytes. Type 1 colonies (all data in Figure 3.9 overleaf) are characterised by frequent and evenly distributed DD divisions. The proportions of PP: DD are similar, with an overlapping range. PP n=202, PD n=172, DD n=173, 95% confidence interval in brackets. Type 1 colony growth dynamics a ǀc 0.32 (0.28-0.36) 0.31 (0.28-0.36) 0.37 (0.33-0.41) Proportion (95% CI) 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 Figure 3.10. Lineage trees of adult keratinocyte Type 1 colonies. Single keratinocytes colonies tracked up to 168 hours. Green nodes are dividing cells. Red nodes do not divide within at least 48 hours of birth. Note similar red and green numbers. Divisions n=547, colonies n=50. 89-90 024 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 Figure 3.12. Lineage trees of adult keratinocyte Type 2 colonies. Single keratinocytes colonies tracked up to 168 hours. Colour code as in Fig 3.9. Note overwhelming green cells, and only PP divisions within first three rounds of division. Divisions n=609, colonies n=6. 92-93 91 Figure 3.11 Type 2 colonies in adult keratinocytes. Type 2 colonies (all data in Figure 3.11 overleaf) are characterised by predominant PP divisions. PP n=543, PD n=52, DD n=14, 95% confidence interval in brackets. Type 2 colony growth dynamics a >> c 0.02 (0.01-0.04) 0.09 (0.06-0.11) 0.89 (0.86-0.92) Proportion (95% CI) 94 Figure 3.13 Closely related keratinocytes in a colony stay together. (A) Schematic lineage tree showing indirect lineage relations. In order of degree of kinship, they are sister (first degree), niece (second degree), cousin (third degree), and second cousin (fifth degree). (B) Pairs of related cells stay together. On average, sister cell pairs have no cells separating them in 97.7% of cases, cousins have a single cell separating in 92.3%, and second cousins have 2 cells separating in 91.8% of cases (sisters n=349, cousins n=456, second cousins n=390). (C) Serial photographs show- ing that descendants stay together within a clone. In this colony, by 12 hours a second cousin has divided, and the index and sister divide by 24 hours. Descend- ants of these founding cells form areas within the colony that largely stays together over time. Scale bar = 100µm. A B Index cell Sister Cousin Second cousin Niece Closely related cells stay closer Sister Cousin Second Cousin 0.0 0.5 1.0 1.5 2.0 2.5 Relationship M ax im um n um be r of ce lls se pa ra tin g +12h +24h +48h +96h C IndexSister Cousin 2nd cousin 95 divisions (a=0.89, b=0.09 and c=0.02; n=543, n=52 and n=14 respectively) (Figure 3.9 and 3.11). The pattern of immunofluorescence imaging in fixed and stained colonies at 168 hours have features similar to NFSKs. In effect, the pattern of proliferation seen in primary human culture is not distinguishable from neonatal foreskin keratinocytes in established culture. 3.4 Cell Division Outcomes are Independent Apart from the cell cycle data that emerged from tracing keratinocytes, the second aspect of lineage trees was the detailed relationships between cells within a single colony. As the origin of individual cells, and their progeny outcome was known, it was possible to investigate whether related cells had the same fate. In other words, was cell fate dependent on a cell’s past, or was it independent in the future? For any given colony, pairs of cells were recorded based on their degree of relationship. Similar terminology to that used in family genealogy trees will be applied to the rest of this section. First-degree relations were comprised of sister or daughter cell (Figure 3.13A). Second-degree relations were niece or granddaughter cell. Third degree relations were a cousin or a great-grand-daughter. Some relationships are direct descendent relatives (“lineal kinship” - daughter, grand- daughter, great-grand-daughter, and so on), while others are indirect descendent relatives (“collateral kinship” - sister, niece, cousin, et cetera). I have used collateral relatives to study any causative relationships. It is not possible to use direct descendants to study division outcomes as any PD or DD division biases the number of subsequent descendants. Also direct descendants are completely separated from each other in time, making the changes in neighbouring cells a potential source of bias. The first question was whether related cells stayed together within a colony. This was important in order to address whether the microenvironment within a colony influenced cell division outcome. I looked at the maximum number of cells between two related pairs of cells. For this, the related cells needed to exist at the same time of observation. It is thus logistically more practical to look at sister pairs, cousin pairs, and second cousin pairs (first, third and fifth degree relatives, Figure 3.13C). I found that sister cells stay together throughout their life cycle in 97.7% of cases (n=349 cell pairs, range 0-1 cells). Cousins had 1 cell between them in 92.3% of cases, while 91.8% of second cousins had 2 cells separating them (Cousin pairs n=456 range=0-3, second 96 Figure 3.14 Closely related keratinocytes make independent outcome choices. Pairs of related cells were observed for fate outcome. (A) Sister pairs showed no difference in either colony type (Type 1 n=305, Type 2 n=415). For each possible outcome of the index cell, the sister cell has a similar chance of the same outcome. None of the matched pairs had a statistically different outcome. (B) The same result was seen in aunt-niece (Type 1 n=617, Type 2 n=782) and cousin pairs (Type 1 n=413, Type 2 n=689). Population chances of PP, PD and DD division types are plot- ted as background filled bars. Proportions of individual cell pairs with the same outcome are plotted as foreground symbols with 95% confidence interval. Large confidence intervals are seen with the PD-PD and DD-DD pairs in Type 2 colonies due to small numbers. A B Index cell Sister cell Type 1 colony 36.4 ± 5.7% 24.6 ± 5.2% 39.0 ± 5.5% 38.6 ± 8.9% 23.0 ± 10.2% 38.4 ± 11.1% Type 2 colony 88.4 ± 2.9% 9.2 ± 3.2% 2.4 ± 2.0% 88.8 ± 3.1% 11.9 ± 13.7% 0 ± 45.9% Both sisters PP Both sisters PD Both sisters DD Index & niece PP Index & niece PD Index & niece DD Both cousins PP Both cousins PD Both cousins DD PP PD DD0.0 0.2 0.4 0.6 0.8 1.0 Pr op or tio n Related pairs in Type 1 colonies PP PD DD0.0 0.2 0.4 0.6 0.8 1.0 Pr op or tio n Related pairs in Type 2 colonies 97 cousin pairs n=390 range 2-4, Figure 3.13B). Thus, descendants of related cells stayed together within a clone (Figure 3.13C). As related cells stayed together, it was possible to explore division outcomes in more depth without the potential confounder of a different microenvironment. Division outcomes of related sister cell pairs were observed in both type 1 (n=305) and type 2 colonies (n=415). For any of three possible outcomes of division for an index cell, there was no increase or decrease in the likelihood of the sister cell having the same outcome in either of the colony types (Figure 3.14A). To look at more distant relations, division outcomes of aunt-niece pairs (Type 1 n=617, Type 2 n=782) and cousin pairs (Type 1 n=413 and Type 2 n=689) were investigated (Figure 3.14B). Here too, for any given division outcome in the index cell, the likelihood of having the same outcome was not increased or decreased, and the likelihood of having a different outcome is the same as the general population. The proportions are shown in Figure 3.14B. Taken together, the analysis of cell division pairs showed that related cells stayed together within a keratinocyte colony, but all related cells had independent division outcomes. 3.5 Two Growth Dynamics Explain the Entire Population Distribution From the insights of the live imaging data, predictions could be made for validation. Since individual cell divisions retained no memory of previous outcomes, colonies formed as a sequence of cell divisions with one of three outcomes, selected at random but with the likelihood of each outcome dependent on whether the founder cell had type 1 or type 2 behaviour. I was able to create a Monte Carlo simulation of the colony size distribution by repeatedly randomly sampling from the observations (paired cell cycle time and division outcomes), in a large number of simulated colonies. I first designed an algorithm to randomly choose, in the observed proportions of a, b and c, from the Type 1 distribution of PP, PD and DD divisions. This was repeated 10,000 times in three separate runs. For any given end observation time point, I could then find the colony size and the number of dividing cells within the virtual colony. I ran this simulation for 168 hours (7 days) and 288 hours (12 days). Similarly, I ran a simulation for the Type 2 colony growth dynamics for 7 days. I did not run a 98 Figure 3.15 Growth behaviour can predict colony size distributions. Monte Carlo simulations of keratinocyte colony sizes (total cells and dividing cells) using Type 1 and Type 2 live imaging data. (A) At 168 hours, colonies arising from type 1 divi- sions overlap in cell number with those from type 2 divisions. This overlap (46-148 cells) can be resolved by the number of dividing cells within the clone. (B) At 288 hours, colonies arising from type 1 divisions can reach several 100s of cells in size. (C) At 168 hours, type 1 colonies have <50% dividing cells, while type 2 colonies have >50% dividing cells within clones of 43-148 cells. (D) Colony area is not linear- ly related to cell number. Smaller clones are larger than expected from a linear fit, while large colonies are smaller than expected. For the same area (e.g. 0.1 mm2), a five fold difference in cell number is seen. Inset 261 cell colony of 0.057mm2, Scale EDUۚP A B Type 2 colonies ”FHOOV Type 1 colonies •FHOOV 0.0 0.5 1.0 0 100 200 300 400 5000 10000 15000 0.0001 0.001 0.01 0.1 1 Colony size (cells) Pr op or tio n Monte Carlo simulation of keratinocyte colonies at 168 hours Type 2 colonies total Type 1 colonies alive Type 1 colonies total 0 100 200 300 400 5000 10000 15000 0.0001 0.001 0.01 0.1 1 Colony size (cells) Pr op or tio n Monte Carlo simulation of Type 1 keratinocyte colonies at 288 hours 0 250 500 750 1000 1250100 0 0.05 0.10 0.15 0.20 0.25 Colony size (cells) Ar ea (m m 2 ) Type 1 colonies alive Type 1 colonies total Pr op or tio n o f d ivi di ng ce lls w ith in co lon y Type 2 colonies have more dividing cells Colony area is not directly related to cell numberC D Overlap 46-148 cells 99 simulation for type 2 colonies for 12 days, as the change in cell fate observed is dependent on location and difficult to simulate. At 168 hours, the distribution of the simulated type 1 and 2 colonies overlapped. By maintaining type 1 divisions, the colony sizes ranged between 2-148 cells, while the size range reached by clones running on type 2 dynamics was 46-2697 cells (Figure 15A). Thus, there was a predicted overlap at this time point with 3.69% of type 1 clones having reached sizes ≥46 cells, and 0.84% of type 2 clones remaining ≤148 cells. However, as may be expected, a prediction of the type of clone could be made based on the number of dividing cells within a clone. Thus, in a type 1 clone above the size of 46 cells, the proportion of dividing cells remaining ranged between 0-49.4%, while in a type 2 colony smaller than 143 cells, the proportion of dividing cells ranged between 70.2-99.1% (Figure 15C). This would reflect the fact that type 1 dynamics would continue to form non-dividing colonies, while type 2 colonies persisted in forming ever larger colonies. This distribution at 168 hours also brought a prediction that could be tested. Small colonies (<46 cells) are formed by type 1 behaviour, while large colonies (>148 cells) are a result of type 2 proliferation. Intermediate sized colonies (46-148 cells) with ≤50% dividing cells were predicted to have type 1 behaviour, while those in this range with >50% had type 2 behaviour. If 50% of cells were dividing in a colony, and the mean cell division time was known, it would be expected that this colony would increase in total size by ~50% in a 24 hour period. Or taken another way, the ratio of colony size at 168 hours to the colony size at 144 hours would be 1.5. This D7/D6 ratio for intermediate colonies was utilised further in Section 3.6. Although, intermediate colony cell number alone at the 168 hour time-point did not reveal the behaviour of the founder cell, it would be expected that this would become apparent at later time points (288 hrs, 12 days) as the gap between type 1 and type 2 colonies will get broader over time. I used my algorithm to predict the distribution of colony sizes for Type 1 proliferative behaviour at this time point. Triplicated sets of 10,000 Type 1 colonies alone were simulated as before (Figure 15B). At 288 hours, type 1 cell fate produced colonies in the size range 2-373 cells with 0- 34.6% proliferating cells remaining within them. As the number of dividing cells within the Type 2 colonies is much higher at 168 hours, there is likely to be a clear separation of these two types based on cell number by 288 hours. Colonies formed from type 1 dynamics, that still contained a small fraction of dividing cells, continued to proliferate reaching cell numbers in the hundreds, and are predicted to form K14 12.51mm2D K10 EdU DAPI K10 EdU DAPI K14 3.29mm2 K14 0.04mm2 K10 EdU DAPI S M LA 100 Figure 3.16 Microscopic characteristics of day 12 keratinocyte colonies. (A) 12 day keratinocyte clones show three macroscopic morphologies (L=Large, M=Medium, S=Small). (B) Small colonies have 10s of keratinocytes, have no EdU positive cells, and express the differentiation marker Keratin10. (C) Wrinkled edge medium colonies have 100s of keratinocytes, have occasional EdU positive cells at the edge (inset), and widely express Keratin10. (D) Smooth edged large colonies have 1000s of keratinocytes, and uniformly uptake EdU with increased uptake in the outer rim (inset from middle and periphery of clone), and express Keratin 10 in suprabasal layers. (Scale bars: main=0.5mm, inset=0.1mm) B C 101 irregular margins due to the number of differentiated cells accumulating at the edge of colonies (Barrandon & Green 1987a; Klein et al. 2011). It was important to note here that cell number did not equate to colony size. Individual keratinocytes become larger in size upon differentiation, to the extent that some researchers have used cell size as a surrogate marker for differentiation (Watt & Green 1981). Thus for colonies with large proportion of dividing cells, the overall colony would appear smaller than a colony with the same number of cells but more of which were differentiated. To confirm this point, I measured the microscopic area of colonies of keratinocytes against the cell number in a random sample of 139 colonies (Figure 15D) of total cell number ranging between 2-1204 cells. If cell size remained constant, it would be expected to see a linear increase in colony size compared to cell number. However, I found that for the best linear fit curve, small colonies were bigger, while larger colonies were smaller than expected. To highlight this further, for the same colony size of 0.1mm2, there was a 5-fold difference in cell number (~100 cells versus ~500 cells). Colony area alone is a thus a poor predictor of proliferative behaviour. Figure 3.16 shows the macroscopic morphologies and the immunofluorescence characteristics of keratinocyte colonies at 12 days of culture. These colonies were cultured in the presence of the nucleoside analogue EdU for the last 24 hours of growth. Cells in small clones do not take up and EdU, while a small proportion of cells in intermediate sized colonies do. In the largest colonies, where there is significant stratification, a proportion of the basal layer cells in the centre, and nearly all the cells at the edge take up EdU. Correspondingly, all cells in small colonies were positive for the differentiation marker Keratin 10, most cells in intermediate sized colonies were positive for Keratin 10, while in the largest colonies Keratin 10 was expressed only in the large stratified keratinocytes. 3.6 PKH26 Labelled Keratinocytes Confirm 2 Rates of Growth To confirm the observation of two growth rates from Section 3.2, I used a different approach to measure this directly. I labelled cultured neonatal foreskin keratinocytes at single cell clonal density after labelling with the inert membrane lipid dye PKH26 for visualisation. These labelled keratinocytes were tracked at 24 hour intervals for 7 days under a microscope to follow the growth of each plated cells. At the end of the observation period, the colonies were fixed and stained for the proliferative marker EdU administered in the last 24 hours of culture. This approach allowed a larger 102 Figure 3.17 PKH26 labelled keratinocytes show two growth patterns. Keratino- cytes colonies labelled with the inert lipid membrane dye PKH26 were tracked at 24 hour intervals for 168 hours (n=339). (A) A 23 cell clone is tracked, with only 2 cells remaining EdU positive. (B) A 406 cell clone with most cells EdU positive. (C & D) Growth curves for all colonies tracked. Colonies of size <46 cells or between 46-148 cells with <50% EdU positive cells are Type 1, while remaining are Type 2. Individu- al colonies plotted as background lines, while linear and exponential fits with 95% CI in foreground. Linear fit to Type 1 colonies with R2=0.96, and exponential fit to Type 2 colonies with R2 6FDOHEDUۚP PKH26 labelled live tracking of a 23 cell colony PKH26 D1 PKH26 PKH26PKH26PKH26 PKH26PKH26 D2 D3 D4 D5 D6 D7 D7 PanCK EdU 1 cell 3 cell 5 cell 11 cell 14 cell 16 cell 23 cell PKH26 labelled live tracking of a 406 cell colony PKH26 D1 PKH26 PKH26PKH26PKH26 PKH26PKH26 D2 D3 D4 D5 D6 D7 D7 PanCK EdU 4 cell 12 cell 29 cell 63 cell 129 184 406 EdU positivity Pr op or tio n Time (hours) Co lon y c ell nu m be r Proportion of EdU positive cells in colonies of size 46-148 cells Type 1 colonies Type 2 colonies Type 1: Linear fit Cycle time = 15.71h r2=0.96 0 24 48 72 96 120 144 168 50 100 150 150 300 450 600 Type 2: Exponential fit Cycle time = 20.8h r2=0.98 PKH26 labelled keratinocytes daily growth curve EdU uptake in colonies of 46-148 cells A B C D Type 1 colony Type 2 colony Linear fit (95%CI) Exponential fit (95%CI) 0.0 0.5 1.0 0.00 0.05 0.10 0.15 103 number of colonies to be indirectly investigated, although at lower resolution than live imaging. 934 single labelled keratinocytes were followed through time in this manner, of which 339 divided to form colonies. An example of a 23 cell colony with 2 EdU positive cells, and a 406 cell colony with 302 EdU positive cells for comparison is illustrated (Figure 17A and 17B). Based on my Monte Carlo simulations, I expected colonies less than 46 cells in number to follow type 1 growth and those more than 148 cells to follow type 2 growth. Colonies in the intermediate size range between 46-148 cells could be assigned to either type 1 or type 2 colonies based on the proportion of EdU positive cells. Intermediate clones in this range with less than 50% of EdU positive cells were classified as type 1, while those with EdU positivity in more than 50% of cells were classified as type 2. Based on my previous experiments, colonies undergoing type 1 growth where a≈c should demonstrate an average linear rate of increase. Colonies undergoing Type 2 divisions, would approximate exponential growth. This prediction could be tested from the growth curves measured in this experiment. In my cohort of 339 labelled colonies, I found 298 (87.9%) were smaller than 46 cells, 21 (6.2%) were larger than 148 cells, and the remaining 20 (5.9%) were intermediate in size. Of the intermediate sized colonies, 11 had <50% EdU positive cells (range 0.08-0.29), while 9 had >50% EdU positive cells (range 0.51-0.84). These were classified as type 1 and type 2 respectively. Additionally, type 1 colonies classified in this manner increased by less than 50% in cell number in the last 24 hours of growth (D7/D6 ratio <1.5), while those classified as Type 2 had ratios>1.5. For these two groups of cells, the best linear fit to the average of the type 1 colonies and the best exponential fit to the average of the type 2 colonies was performed using a least squares fit approach (Figure 17C & D). The type 1 colonies had good agreement (R2=0.96) with a linear fit, while type 2 colonies too had a good agreement (R2=0.98) with the exponential fit. This analysis confirms that the entire population of keratinocytes at 7 days of culture can be described in terms of two growth behaviours: one resulting in a linear expansion of colony size, and another approaching exponential growth. In the colonies classified as type 1 based on a combination of final cell size, and EdU positive cells, a linear growth process was the most likely outcome. Similarly, in the colonies classified as type 2, a near-exponential single curve was the best description for the entire dataset. 104 Figure 3.18 Day 7 keratinocyte colony distribution. (A) Colony distribution was determined in independent datasets showing good agreement (n=1460, 321 and 305). (B) Colony size was fractionated according to the proportion of positive EdU cells, with each point representing a colony. The average of each group increases with EdU positivity (Kruskal-Wallis test p<0.001). However, the dispersion of distribution makes this an imperfect correlation (R2=0.40). (C) Small colony sizes (up to 4 cells) can be used to determine type 1 growth proportions. (n=259, 72 and 53 for 2-, 3- and 4-cell colonies respectively). B Colony size distribution fractionated by %EdU positive cells Ed U 10% 10< Ed U 20% 20< Ed U 30% 30< Ed U 40% 40< Ed U 50% 50< Ed U 60% 60< Ed U 70% 70< Ed U 80% 80< Ed U 90% 90< Ed U 100 % 2 8 32 128 512 2048 Cl on e S ize Clone Size 2 8 32 128 512 2048 C Relative proportions of the smallest colonies gives type 1 growth dynamics 2 3 4 0.00 0.05 0.10 0.15 0.20 Clone Size Pr op or tio n 0.31 0.28 0.41 Proportion A Observed keratinocyte colony size distribution at 168 hours 0 5 10 15 20 2 3-4 5-8 9-1 6 17- 32 33- 64 65- 128 129 -25 6 257 -51 2 513 -10 24 102 5-2 048 Pr op or tio n Colony size 105 3.7 Small Colonies Confirm Type 1 Proliferative Dynamics Further experimental evidence of type 1 clonal growth could be gleaned from distribution of small colonies in large keratinocyte populations. The distribution of the smallest colonies composed of 2 cells, 3 cells and 4 cells could be used to calculate the proportions of PP, PD and DD divisions in type 1 colonies (Section 2.4, statistics methods). Since these colonies are the most frequent in any clonal population distribution, it is predicted to give a statistical derivation for a, b and c. As previously observed, this is independent of the proportion of type 1 colonies in the overall population as such small colonies are formed by type 1 divisions alone at late time points. 1460 keratinocyte colonies seeded at clonal density were counted at 168 hours. Two smaller replicates of 305 and 321 colonies were also done to confirm the distribution (Figure 18A). 26.3% of all clones were 2-4 cells in size. The proportion of EdU positive cells was counted within each colony and grouped in increasing proportions. Here I found that increasing EdU positivity was broadly related to a larger average clone size, but that this was imperfect for individual clones (Figure 18B, Kruskal-Wallis test p<0.0001 for mean clone size in each group, but R2=0.40). I observed 259 2-cell colonies (!!), 72 3-cell colonies (!!), and 53 4-cell colonies (!!), resulting in the proportions !! = 0.177 ± 0.01, !! = 0.049 ± 0.006 and !! = 0.036 ±0.005 (Figure 18C). The derived values are thus, ! = 0.415, ! = 0.278 ± 0.036 and ! = 0.307. The value of b is consistent with my observation in the Incucyte™ data (0.28 ±!0.03). The values of a and c are correlated when calculated in this manner, and therefore confidence intervals cannot be calculated from standard error propagation methods. 3.8 Proliferating Cells Within Colonies at Early Time Points Keratinocyte colony size distributions were informative for confirming type 1 clonal dynamics at late time points. At early time points, however, quantitative predictions using a sequential combinatorial approach were difficult to make due to two main factors. Firstly, colony size distributions from the two types of cell division overlap at early stages, making prediction only possible if the relative ratio of the two types were known. Secondly, multiple dividing cells within a colony divided in parallel making sequential models difficult to apply. However, combinatorial Figure 3.19 Keratinocyte colonies at 72 hours post seeding. 2-, 4- and 8-cell colo- QLHVZHUHIL[HGDWKRXUVSRVWVHHGLQJZLWKKRXUVRISULRUۚ0(G8WUHDW- PHQW $ FHOOFRORQLHVZLWKRUFHOOV(G8SRVLWLYH WRS RU.HUDWLQSRVLWLYH ERWWRP  % FHOOFRORQLHVZLWKRU(G8SRVLWLYHFHOOV WRS RU.HUDWLQ SRVLWLYH ERWWRP  & FHOOFRORQLHVZLWKFHOOV(G8SRVLWLYH6FDOHEDU ۚP 106 A B & . (G8 (G8 (G8 (G8 (G8 (G8 (G8 (G8 (G8 (G8 (G8 (G8 (G8 (G8 (G8 (G8 (G8 (G8 2 c ell co lo ni es 4 c ell co lo ni es 8 c ell co lo ni es . .UW . . . . . . . . 107 approaches did suggest heterogeneity in proliferating cells within colonies that could be explored qualitatively. I set out to determine qualitatively at early time points if colonies can demonstrate the heterogeneous distribution of dividing and non-dividing cells predicted. I seeded unlabelled keratinocyte colonies at clonal density, and added 10!M EdU in the last 24 hours of culture. At 3 days, feeders were removed and plates were fixed and stained for EdU as a proliferation marker and Keratin 1 as a differentiation marker (Figure 3.19). I found that in 2 cell colonies, all three predicted colony types were observed: colonies with 2 EdU positive cells, 1 EdU positive cell, or no EdU positive cells. Similarly Keratin 1 positivity was seen in 0, 1 or 2 cells. The heterogeneous distribution of EdU positive and Keratin1 expressing cells were also seen in 4-cell and 8-cell colonies. 3.9 Colony Behaviours Have Distinct Transcriptional Profiles I then set out to examine if colonies at early time points could be identified based on their transcriptional signatures. With the absence of direct markers for behaviour type, prospective identification is difficult. At early time points, colony size is not helpful either, as demonstrated by my PKH26 data. I therefore first examined if within a fixed colony size (8 cell colonies), the two division behaviours would lead to overlapping characteristics at early timepoints. 8 cell colonies should have enough RNA for reliable amplification, thus allowing identification of differentially expressed genes. Using my Monte Carlo algorithm, I ran triplicated sets of 10,000 simulated clones for each of the behaviour types, for hourly intervals between 48-72 hours (Figure 3.20A). This time period would cover 3-5 rounds of division on average based on the median 15.1 hour cell cycle time from the live imaging data. In simulated colonies dividing on Type 1 cell fate, the proportion of 8-cell clones gradually increased from 1.4% to 6.0% of the total colonies. In these 8-cell colonies, the average number of dividing cells decreased in this timeframe from 4.3 cells to 2.1 cells. Therefore, this represents that following type 1 cell fate results in the accumulation of non-dividing 8- cell clones progressively throughout the 48-72 hour period. By contrast, colonies dividing by type 2 division dynamics show a peak of 8-cell colonies by 54-60 hours at 15.7-16.8% of all type 2 colonies. After this peak, numbers of 8-cell Type 2 colonies decrease as further cell divisions increase the total number of cells. In the Type 2 8-cell colonies, the average number of dividing cells in the 48-72 hour period decreases only Figure 3.20 Studying 8-cell keratinocyte colonies at 2.5 days post seeding. (A) Monte Carlo simulation of Type 1 and Type 2 dynamics distinguishes 8 cell clones formed by either dynamic. Type 1 8-cell colonies have 3-4 dividing cells on average, while those formed by type 2 divisions have 6-7 dividing cells. (B) Experimental schematic for micro-arrays of 8-cell colonies (from Dr. Kasumi Murai). (C) Hierar- chical clustering showing three groups of 8-cell colonies. The two most different groups showed low fold change of the differentiation genes FABP5 and the S100 family in the larger cohort, suggesting this may be associated with type 1 cell fate. 108 48 52 56 60 64 68 72 0 2 4 6 8 0 5 10 15 20 25 48 52 56 60 64 68 72 0 2 4 6 8 0 5 10 15 20 25 8 cell clones as % of all Type 1 clones 8 cell clones as % of all Type 2 clones Dividing cells in 8 cell Type 1 clones Dividing cells in 8 cell Type 2 clones Time (hours) Nu m be r o f di vid in g c ell s Time (hours) Percentage of all clones 7 9 11 13 8 cell NFSK colonies n=11 RNA extraction Whole transcriptome amplification & purification Fragmentation & labelling Affymetrix Human Gene 1.0 ST Array A B C FABP5, S100 family fc 1.3-1.8x 109 marginally from 7.3 cells to 6.4 cells per colony. This indicates that type 2 divisions result in increasing number of 8 cell colonies capable of further divisions, and with time have fewer 8-cell colonies as they carry on dividing. If transcriptional differences reflected the difference in behaviour at this early timepoint, it would be expected that two groups of colonies could be identified, but with low fold differences as the number of dividing cells differs two-fold between the two groups. To investigate the transcriptional differences between the two groups of cells, I drew from experiments conducted by Dr Kasumi Murai (Figure 3.20B). Here, single colonies were grown per well of a half-area 96 well plate. Colonies of 8 cells were identified at 2½ days (n=11). From these colonies, RNA was extracted and the whole transcriptome was amplified and purified. The Affymetrix Human Gene 1.0 ST array was used for gene expression profiling for whole-transcript coverage. On hierarchical clustering, three groups of colonies emerged – a group of 4 colonies that were distinct from the other two groups of 2 and 5 colonies (Figure 3.20C). Comparing the two most different groups of 8-cell colonies, the differentiation markers FABP5 and S100 family were seen at low fold changes compared with the other (fold changes FABP5 1.67, S100A10 1.59, S100A11 1.83 and S100A2 1.31). This suggests that this group had a higher proportion of differentiated cells, and is likely to be following type 1 dynamics. A gene list was drawn up based on fold change, and number of probes detecting changes. To validate the differences seen in the array data, I performed immunofluorescence imaging of fixed and stained colonies of 2½ day old keratinocyte colonies to see if heterogeneity of protein expression could be detected. I found that for the proteins for which commercially available and validated antibodies were available, two higher expressed proteins, and one lower expressed protein did indeed show the heterogeneity of cell-to-cell expression expected within colonies (Figure 3.21). This indicated that at the protein expression level, there were distinct signatures of the two behaviours predicted from live imaging. While these proteins were unlikely to be regulators of the behaviour, they were likely to be effectors of the fate decisions taken by the cell. Taken together, the two types of observed cell fate in keratinocytes have distinguishable transcriptional differences seen through cell-to-cell heterogeneity within colonies. Figure 3.21 Protein heterogeneity in keratinocyte colonies. Transcriptional differences identified from microarrays (Figure 3.20) were validated using immunofluorescence of keratinocytes at day 7 and day 2.5. Day 7 colonies show heterogeneity of expression in intermediate sized colonies. 4 cell colo- nies at day 2.5 show all combinations of cell expression (0-4 cells). Scale bar =ۚP 110 A B Protein X Protein Y Protein ZC D7 D7 D7 X PanCK DAPI X D2.5 D2.5 D2.5 0/4 1/4 2/4 3/4 4/4 0/4 1/4 2/4 3/4 4/4 0/4 1/4 2/4 3/4 4/4 Y PanCK DAPI Y Z PanCK DAPI Z 111 3.10 Discussion Keratinocyte colony size distributions may vary based on either cell cycle time or cell division outcome. Live imaging directly demonstrates that cell cycle time does not significantly vary between colony sizes. A broad distribution of colony sizes is, however, generated by two, and only two, modes of cell fate. Such fate choices are independent in neighbouring cells within a colony. Fate choices are not fixed, with cells at the centre of large colonies switching their fate after initial symmetric growth indicating that such switches can occur, at least unidirectionally. Traditional ways of describing a cell’s growth potential in vitro have been to use combinations of surrogate markers based on prior proliferative ability (clone size) along with its future colony forming ability (sub-clonal assays). Using the proliferative ability in the present live imaging study coupled with population-wide analysis in the form of 7 day colony size distributions, we see that the “continuum” of colony sizes can be achieved by proliferating cells with one of two modes of cell division. This is consistent with evidence from the literature, where quantitative observations of keratinocyte growth in vitro shows three morphologies of colonies, but I demonstrate this is produced by only two types of divisions (Barrandon & Green 1987b). It is an important question to ask whether the two observed modes of proliferation indicate a fixed cell phenotype, or if these are interchangeable cell states. A fixed phenotype is not the case within the centre of large colonies, where cells undergo change from type 2 cell fate to type 1 fate. However, it is not proven from my results whether this is unidirectional, and if type 1 colonies can change fate to type 2 outcomes. Evidence from the literature suggests that small keratinocyte colonies have the potential for forming large colonies through viral oncogene transduction (Barrandon et al. 1989), and that large colonies undergo clonal conversion to small colonies with continuous Rac1 inhibition (Nanba et al. 2013). In the context of wound healing, dedifferentiation of suprabasal differentiated cells evidenced by changes in keratin expression are seen to contribute to wound re-epithelialisation (Usui et al. 2005; Patel et al. 2006). Additionally, differentiated cells have been shown to take up stem cell function following the depletion of resident stem cells in the stomach or lung in recent in vivo studies (Stange et al. 2013; Tata et al. 2013). Thus, it is conceivable that an alteration in epigenetic programs regulating type 1 dynamics may be able to change cells from type 1 behaviour in to type 2 (Mulder et al. 2012). While exponential growth of keratinocytes has been extensively studied, less is known about the contribution of type 1 fate outcomes to normal IFE. There is strong 112 113 evidence that the IFE is maintained separately from the follicular epidermis, although some contribution is made by the hair follicle and other appendages following wound healing (Rochat et al. 1994; Claudinot et al. 2005; Ito et al. 2005; Levy et al. 2005; Lu et al. 2012; Rittié et al. 2013). Within the separately maintained mouse inter-follicular epidermis, there is evidence for the existence of type 1 keratinocyte behaviour from in vivo mouse lineage tracing experiments (Clayton et al. 2007; Doupé et al. 2010; Mascré et al. 2012). There is also indirect evidence that biased type 1 dynamics exist in p53 mutant clones in sun-damaged IFE in the presence of persistent UV irradiation (Klein et al. 2010). My findings show that balanced cell division occurs within human keratinocytes in vitro. The existence of such behaviour in vitro indicates that this is robust, and not generated by external signals within homeostatic tissue. The wide range of colony sizes seen in human xenograft colony tracing experiments is also consistent with stochastic outcome choices in human keratinocytes (Ghazizadeh & Taichman 2005). The distribution of such colonies throughout the epidermis in these experiments may indicate that this is the dominant mode of cell division in the human epidermis in vivo. Since the proliferative capacity of large colony forming cells seems to be reduced with age, type 1 divisions forming the bulk of homeostatic divisions provides a mechanism for skin maintenance in the face of diminishing proliferative reserve (Barrandon & Green 1987b; Giangreco et al. 2010). These findings have implications for regenerative medicine. Switching the more numerous keratinocytes with stochastic cell fate to a proliferative cell fate may be an appealing complimentary strategy for regeneration and wound healing. Taken together, my results argue that human keratinocytes demonstrate two robust cell intrinsic modes of proliferation in vitro. Despite the histological differences between mouse and human epidermis, keratinocytes from both sources share the same proliferative organisation. Thus, fate choices in dividing cells may be a conserved feature of homeostasis in the epidermis, and its manipulation may lead to renewed regenerative strategies. 114 Figure 4.1 Culture conditions influence keratinocyte protein expression. (A) Western Blot analysis of the effects of EGF levels on keratinocytes in culture. Culture and cell lysate preparation by AR, Western from Dr. Kasumi Murai. Cells grown with no added EGF(0), 10ng/mL EGF (10) or 20ng/mL EGF (20) at indicated timepoints. Phosphorylated EGF was lower at early timepoints in EGF0, with corre- sponding downstream decrease in phospho-ERK. The effects were similar by 72 hours. By 16 hours, total EGF is higher, and ligand induced degradation products (arrow) are absent in cells at EGF0. (B) Immunofluorescence of keratinocyte colo- nies grown in the presence or absence of R-Spondin 150ng/mL. Widespread nucle- ar շFDWHQLQLVVHHQZLWK5VSRQGLQ6FDOHEDUۚP A B 0 10 20 0 10 20 15 min 1 h Time: EGF (ng/ml): P-EGFR (Y1068) pan EGFR -tubulin -tubulin P-ERK1/2 P-AKT P-p90RSK P-S6 Pan ERK 0 10 20 0 10 20 16 h 72 h P-ERK1/2 Time: 15min 1 h 16 h 72 h 0 10 20EGF (ng/mL): 0 10 20 0 10 20 0 10 20 - ( 1068) P Ƚ P-p90RSK -AKT ERK1/2 S6 P-ERK1/2 an ERK Ƚ-tubulin 0 10 20 0 10 20 15 min 1 h Time: EGF (ng/ l) P-EGFR (Y1068) pan EGFR -tubulin -tubulin 2 P 0 10 20 0 10 20 16 h 72 h - /2 PanCK $FWշFDWHQLQ PanCK $FWշFDWHQLQ +R-Spo 50ng/mL Control 115 CHAPTER 4 EFFECTS OF EGF AND R-SPONDIN ON KERATINOCYTE GROWTH 4.1 Chapter Overview The supplementation of Epidermal Growth Factor (EGF) in culture media increases culture lifetime, improves sub-culture efficiency making it an important component of standard keratinocyte culture conditions (Rheinwald & Green 1977; Navasaria et al. 1994). When grown in the absence of supplemented EGF, keratinocytes appear to grow at the same rate in early culture (up to 7 days), but at slower rates at late timepoints (up to 24 days) (Rheinwald & Green 1977; Barrandon & Green 1987). In this chapter I set out to examine the effect supplemented EGF on the growth pattern of cultured keratinocytes using the methodologies used in Chapter 3. Additionally, I also studied the effect of R-Spondin, a secreted agonist of the canonical Wnt/!-catenin signalling pathway known to have mitogenic effects on epithelial cell proliferation (Parma et al. 2006; Kim et al. 2005). Section 4.2 demonstrates that altering the concentration of added EGF in culture has downstream signal effects determined by western blots. Additionally, the effect on individual colonies of adding R-Spondin to culture can be detected by immunofluorescence. Section 4.3 uses live imaging of keratinocytes in the absence of supplemented EGF to define cell fate in this culture condition. In Section 4.4, I show mathematically that exponential growth of a fixed rim of cells in a colony will result in linear rates of radial growth in agreement with published literature. Section 4.5 describes the effect of EGF withdrawal on keratinocyte colony area. Section 4.6 analyses the distribution of colonies to demonstrate that EGF and R-Spondin change the population distribution without affecting Type 1 colonies. Finally, Section 4.7 discusses a few key implications of my findings. 4.2 Changes in Culture Media Affects Protein Expression I first investigated if varying the levels of supplemented EGF has a detectable molecular effect on keratinocytes. I cultured keratinocytes at 1000 cells/cm2 with varying levels of supplemented EGF: no EGF (EGF0), 10ng/mL (EGF10) or 20ng/mL (EGF20). These culture conditions were maintained from seeding up to harvest. I collected protein extracts from technical triplicates at 15 minutes, 1 hour, 16 hours and 116 Figure 4.2. Cell cycle distribution in EGF0 is similar to EGF10. (A) The median cell cycle length in EGF0 was 15.17hrs (range 5.83-54.33hrs, n=805). (B) Cell cycle length was not statistically different for different outcomes (PP n=410 , PD n=192 , DD n=203). (C) 1-99% Box and whisker plot show- ing that cell cycle distribution did not differ between division types (Type 1 PP n=112, PD n=94, DD n=135; Type 2 PP n=335, PD n=105, DD n=75). (D) 1-99% Box and whisker plot showing cell cycle distribution does not differ between divisions in EGF0 and EGF10 (EGF0 n=805, EGF10 n=1777). Type 1 Type 1 Type 1Type 2 Type 2 Type 2 PP Divisions PD Divisions DD Divisions Ti m e ( ho ur s) p = ns 0 24 48 72 p = ns p = ns A Individual cell division time 24 48 72 96 Me dia n0.00 0.04 0.08 0.12 Time (hours) Pr op or tio n 0 5 10 15 20 PP PD DD Division outcome Ti m e ( ho ur s) p=ns p = ns 0 24 48 72 Ti m e ( ho ur s) p = ns EGF0 EGF10 B C D 117 72 hours after seeding and quantified the levels of protein. Electrophoresis was on a 7.5% SDS polyacrylamide gel for EGFR and 10% SDS polyacrylamide gel for other proteins. Western blots were performed by Dr Kasumi Murai. At early time points up to 16 hours, there were decreased levels of phosphorylated EGF receptor in EGF0 cultures, which returned to comparable levels at 72 hours compared with EGF10 and EGF20 (Figure 4.1). In contrast, levels of the total EGF receptor was higher in EGF0 cultures throughout the 72 hour time course with minimal ligand induced degradation products as reported in the literature (Beguinot et al. 1984; Sousa et al. 2012). Together, this reflects decreased EGFR degradation and increased phosphorylation of EGFR in EGF0 culture. However, only modest decreases in downstream p-p90RSK and p-ERK1/2 in EGF0 culture was seen. P-AKT and p-S6 did not appear markedly different between culture conditions. To examine if there was a cellular effect on keratinocytes cultured with the Wnt agonist R-Spondin, I cultured clonal density keratinocytes at clonal density in serial dilutions between 1-100 ng/mL, changing the media on day 3. At 7 days, feeders were removed, and plates were fixed and stained for activated !-catenin. Increased nuclear staining was seen in concentrations between 10-100ng/mL compared to controls. A typical colony grown with 50ng/mL R-Spondin1 is illustrated in Figure 4.1B. This demonstrates that R-Spondin can have effects on Wnt signalling in individual keratinocytes at this concentration. 4.3 Live Imaging of Cultured Keratinocytes Without Added EGF To identify the types of proliferative behaviour in the absence of supplemented EGF, I performed live imaging of fourth to sixth passage single unlabelled keratinocytes at clonal density in the Incucyte™ imaging system. Images, and videos were acquired in the same manner as before up to 216 hours. At the end of the experiment, colonies were fixed and stained for proliferation and differentiation markers. 856 complete cell cycles were observed in 51 colonies. Excluding the first cell cycle to account to for any plating lag, the median cell cycle length was 15.17 hrs (range 5.83-54.33 hrs, n=805, Figure 4.2A). 99.8% of all divisions occurred within 48 hours. There were no colony fragmentation or fusion events. The colonies were similar to before, but had an increased cell density making tracking challenging, especially in large colonies at later time points. 118 Figure 4.3 Cell fate in keratinocytes without supplemented EGF. (A) Type 1 divisions (all data in Figure 4.3 overleaf) are characterised by frequent and evenly distributed DD divisions similar to Type 1 divisions in the presence of EGF10. The proportions of PP: DD are similar, with an overlapping range. PP n=112, PD n=94, DD n=135. (B) Type 2 division in the absence of EGF have sustained proliferation (a>>c) early in their lineage (up to 96 hours), but then switch to balanced division (a=c) in their inner 2/3rds, with continued sustained proliferation in an outer rim of cells. Early PP n=186, PD n=31, DD n=8; Late outer rim PP n=84, PD n=11, DD n=3; Late inner 2/3rds PP n=65, PD n=63, DD n=64. Type 1 colony growth dynamics a ǀc 0.40 (0.33-0.45) 0.28 (0.23-0.33) 0.33 (0.28-0.38) Proportion (95% CI) Type 2 colony growth dynamics a >> c 0.04 (0.02-0.07) 0.14 (0.10-0.19) 0.83 (0.77-0.87) Proportion (95% CI) 0.03 (0.01-0.09) 0.11 (0.06-0.19) 0.86 (0.77-0.92) Outer rim, a >> c Early (<96 hours) Late (>96 hours) Proportion (95% CI) 0.33 (0.27-0.40) 0.33 (0.26-0.40) 0.34 (0.27-0.41) Inner 2/3rds, a ǀc 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 0 24 48 72 96 120 144 168 Figure 4.4. Lineage trees of keratinocytes without supplemented EGF. Single keratinocytes colonies tracked up to 168 hours. Green nodes are dividing cells. Red nodes do not divide within at least 48 hours of birth. Divisions in Type 1 are similar to growth in EGF10. Divisions in large colonies followed similar fate to those in EGF10 early in culture (up to 4 days), but changed dynam- ics subsequently. Dividing cells at the outer rim (empty green nodes) of the type 2 clones continue with near symmetric growth. Divisions in Type 1 n=341, colonies n=45; Divisions in Type 2 n=515, colonies n=6. 119-120 Type 1 cell fate Type 2 cell fate 121 Cell divisions were classified as PP, PD or DD based on the proliferative ability of the daughters. The median cell cycle time for each type of division was 15.17 hrs (range 6.17-54.33 hrs, n=410), 14.83 hrs (range 5.83-34.00 hrs, n=192), and 15.67 hrs (range 6.33-39.83 hrs, n=203) respectively. There was no statistical difference between the cell cycle time for PP or PD outcomes (Kolmogorov-Smirnov test p=0.35) or between PD and DD outcomes (Kolmogorov-Smirnov test p=0.08) (Figure 4.2B). Cell cycle distribution did not differ between the two modes of division (Kolmogorov- Smirnov Type 1 PP vs. Type 2 PP p=0.34, Type 1 PD vs. Type 2 PD p=0.51, Type 1 DD vs. Type 2 DD p=0.30) (Figure 4.2C). Finally, there was no statistically significant difference in cell cycle distribution between the cohorts grown at EGF10 and at EGF0 (Kolmogorov-Smirnov EGF10 vs. EGF0 p=0.06; EGF0 n=805, EGF10 n=1777) (Figure 4.2D). Lineage trees were drawn for all colonies (Figure 4.4). As before, in large colonies all cells were tracked up till ~96 hours, but subsequently 4-6 consecutively observed cells were tracked up to the end of the video. As before, two colony types emerged based on the frequency of observed cell fate outcomes (Figure 4.3). In type 1 colonies (88.2%, n=45), the frequency of PP divisions approximately equaled DD divisions (a=0.33 (95%CI 0.28-0.38) and c=0.40 (95% CI 0.23-0.33) respectively; n=112 and n= 135 respectively). These colonies remained small in size. Compared to the type 1 colonies in EGF10, the c is 10% higher than a, but this difference is not statistically significant. In type 2 colonies (11.8%, n=6), the dynamics are different from that seen in EGF10. Up till 96 hours, cells undergoing type 2 divisions operate in a manner similar to keratinocytes in EGF10. Most progeny are PP (a=0.83, n=186), while the remaining are PD (b=0.14, n=31) or DD (c=0.04, n=8). However, divisions after 96 hours show different dynamics depending on the location of the cell, similar to that seen in colonies growing in EGF10 after 7 days. Thus, cells in the inner two thirds of the colony have division proportions similar to those seen in type 1 divisions (a=0.34, b=0.33, c=0.33; n=65, 63 and 64 respectively). However, divisions occurring at the outer rim of these colonies continue dividing in the same manner as early culture (a=0.85, b=0.11, c=0.03; n=84, 11 and 3 respectively). Thus, the colonies with EGF0 show growth dynamic change in the inner part of large colonies at an earlier stage (96 hours onwards) compared to those at EGF10 (168 hours onwards), and a much smaller rim of cells continue to divide almost exponentially. Immunofluorescence of the fixed and stained colonies was then performed at day 3 and day 7 (Figure 4.5). At day 3, 2-cell colonies were seen with 2-cells, 1-cell or no cell Figure 4.5 Keratinocyte colonies in EGF0 culture media. (A&B) 2-cell and 4-cell colonies at 3 days showing EdU distribution of all combinations. (C) Colony at 7 days showing 8 cell colony with no EdU positive cells likely to form from type 1 dynamics. (D) Large colony at 7 days with central stratified keratin 1 positive cells and EdU positive cells mostly as an outer rim, typical of type 2 dynamics. Scale EDU ۚP 122 A B C K14 EdU 0/2 EdU+ 1/2 EdU+ 2/2 EdU+ 0/4 EdU+ 1/4 EdU+ 2/4 EdU+ 3/4 EdU+ 4/4 EdU+ Da y 3 co lo ni es K14 DAPI K1 EdU Type 1 8 cells K1 EdUK14 DAPI Type 2 792 cells D Da y 7 co lo ni es 123 positive for EdU. Similarly, 4-cell colonies had the entire range of EdU positivity. At day 7, small colonies were largely positive for the differentiation marker Keratin 1, with occasional EdU positive cells. Large colonies at day 7 were more densely packed than those at EGF10, and showed a rim of EdU positive cells with central stratification and Keratin 1 positivity. This is consistent with the difference in dynamics in the central area of the colonies. 4.4 A Dividing Rim of Cells Drives Linear Radial Colony Growth According to my live imaging data, after 96 hours large colonies grow in size mainly through a rim of exponentially dividing cells. Here I explore mathematically the effects of such sustained growth at the rim on the growth rates of the entire colony. Let us assume a colony of radius !!, and a fixed rim of dividing cells of constant width a. After one average cell cycle, the new radius of the colony is !!, and the central core of non-dividing cells remains constant at (!! − !). As the dividing rim of cells almost doubles in number over the cell cycle, it holds that the increase in area is twice the area of the initial rim. Or, !"#$%&'%!!"!!"#! = ! ∙ !"#!!!"!!"#"!"$%!!"# !!!! − ! !! − ! ! = ! ∙ !!!! − ! !! − ! ! !!! − !! − ! ! = ! ∙ !!! − !! − ! ! !!! = ! ∙ !!! − !! − ! ! !!! = !!! + !"!! − !! !! = (!!! + !"!! − !!) This expression resulted in a linear rate of radial growth, with the slope only dependent on the width of the fixed rim of dividing cells. A wider rim of dividing cells will result in a faster rate, but would be still linear (Figure 4.6). This is in agreement with published growth rates of colonies at up to 24 days, where the presence of EGF results in a faster linear rate of growth compared to no supplemented EGF (Barrandon & Green 1987). It is worthwhile noting that the increase in area with such growth is quadratic, and not exponential. 124 Figure 4.6. An exponentially dividing rim gives linear colony growth. (A) A colony of keratinocytes with a small (colony 1) or large (colony 2) fixed outer rim of cells dividing exponentially. With progressive rounds of divi- sion, the colony increases in size with an expanding central core of non-di- viding cells. (B) Colony radius plotted against time with such divisions result in a linear increase of the colony radius. The rate of growth is depend- ent on the thickness of the dividing rim. 0 2 4 6 8 10 0 20 40 60 Rounds of division Ra diu s o f c olo ny (m m ) A fixed dividing rim follows linear growth Colony 1 Colony 2 Colony 1 Colony 2 Fixed rim of dividing cells Expanding non-dividing core Expanding non-dividing core A B Fast linear colony growth Slow linear colony growth 125 4.5 Changes in Colony Size With Supplemented EGF One of the most striking differences to colonies grown in the absence of EGF compared to those grown with supplemented EGF is their smaller size (Rheinwald & Green 1977). Changes to the size of large colonies is rapid in the presence of the transforming growth factor-α (TGF-α) occurring within a couple of hours (Barrandon & Green 1987). In this section I set out to explore if the presence or absence of EGF in the culture media affected all types of colony sizes equally. To study the effects of EGF on colony size, I quantified the colony area against cell number for four culture conditions (Figure 4.7A&B). Firstly, I quantified 7-day old colonies cultured in EGF10 or EGF0, which had been previously cultured in EGF10 (EGF10→EGF10 or EGF10→EGF0 respectively). From the cultures in EGF10, parallel 7 day cultures were done in either EGF10 or EGF0 (EGF0→EGF10 or EGF0→EGF0 respectively). A comparable number of colonies across the entire range of colony sizes were quantified (EGF10→EGF10 n=133, EGF0→EGF10 n=117, EGF0→EGF0 n=109, EGF10→EGF0 n=136). There was no difference in the population distribution of colony cell number between an EGF10 bulk population and the quantified colonies grown in EGF10 (Kolmogorov-Smirnov test, p=0.16). Similarly, there was no difference in the population distribution between an EGF0 bulk population, and the quantified colonies grown in EGF0 (Kolmogorov-Smirnov test, p=0.23) (Figure 4.7C). I then plotted the area for each of the four culture conditions against colony cell number (Figure 4.7D). Here I found that small colonies (<50 cells) in EGF0 and EGF10 had comparable areas with no difference (Wilcoxon matched-pairs signed-rank test EGF0 versus EGF10 p=0.53). However, for the larger colonies, there was a significant difference based on the culture condition. Colonies larger than 50 cells grown at EGF10 were significantly larger that those grown at EGF0 (Mann-Whitney unpaired test EGF0 versus EGF10 p<0.0001). Additionally, colonies initially grown at EGF0, when grown back in EGF10 reverted to larger colony sizes, while those re-plated in EGF0 persisted with small sizes but with no further reduction. Note that an unpaired analysis was done for large colonies as matched pairs of the same cell numbers in this distribution was not possible. From the experiments in this section I concluded that small colonies did not show any sustained size changes associated with varying levels of culture EGF. However, EGF readily affected large colonies in terms of colony area. Additionally, this change of colony size with EGF was reversible with re-supplementing withdrawn EGF. EGF 261 cells 56861.12ۚP2 212 cells 35010.76ۚP2 16 cells 10908.06ۚP2 18 cells 9430.76ۚP2 Figure 4.7 Effect of EGF on area of keratinocyte colonies. (A) Area of 18 cell colony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ۚP 126 & ' &oPSarison of SoSXlaWLons 4 16 64 256 1024 4096 & lo ne sL ]H L RJ 2) EGF10 EGF10 EGF0 EGF10 %XON EGF10 EGF0 EGF0 EGF0 %XON EGF10 EGF0 EffecW of EGF on NeraWinocyWe colony area 2 8 32 128 512 204850 0 50000 100000 150000 200000 &lone 6i]e Ar ea (ۚ P 2 ) EGF10 n 133) EGF0 n 117) EGF10nJ/PL No aGGeG EGF EGF0 n 109) EGF10 n 136) A %EGF0 EGF0EGF10 EGF10 127 withdrawal in successive re-plating did not have an incremental effect of decreased colony size. 4.6 EGF and R-Spondin Do Not Affect Type 1 Behaviour To quantify the effect of EGF and R-Spondin on the population distribution of keratinocyte colonies, I counted cohorts of keratinocytes at clonal density in EGF0, EGF5, EGF20 and R-Spondin at 50ng/mL to compare with my previous cohort cultured at EGF10 in Chapter 3. The total number of colonies observed in each of the five conditions were as follows: EGF0 (n=1682, range 2-1123 cells), EGF5 (n=723, range 2-1320 cells), EGF10 (Chapter 2 cohort n=1460, range 2-1812), EGF20 (n=563, range 2-1254), and R-Spondin (n=882, range 2-793). Each distribution had replicates with smaller distributions to determine the confidence intervals for proportions. The distribution for EGF0 with 95% confidence intervals is shown in Figure 4.8A. To study if the culture conditions changed the distribution of keratinocyte colonies, I compared each distribution against EGF10 using the Kolmogorov-Smirnov test (Figure 4.8B). Each of the populations thus compared were significantly different from the reference EGF10 population (EGF10 versus EGF0 p=0.0015, EGF10 versus EGF5 p<0.0001, EGF10 versus EGF20 p<0.0001, and EGF10 versus R-Spondin p<0.0001). As expected, this confirmed that the culture conditions had an effect on the population distribution. I then compared the relative proportions of 2-cell, 3-cell and 4-cell colonies to see if changes in type 1 dynamics contributed significantly to this difference in population distribution (Figure 4.8C). To do this, I compared the proportions of 2-, 3- and 4-cell colonies in each of the cohorts and compared them to the EGF10 reference cohort. Wilcoxon matched pair analysis was then done to see if the distributions were different. In all the conditions, I found that the distribution of 2-, 3- and 4-cell colonies was not different from the reference population (EGF10 versus EGF0 p=0.75, EGF10 versus EGF5 p=0.25, EGF10 versus EGF20 p=0.25, and EGF10 versus R-Spondin p=0.25). As the relative distribution of 2-, 3- and 4-cell colonies were the determinants for a, b, and c in type 1 cell fate, they were not significantly changed by rising EGF, or R-Spondin at 50ng/mL. From these observations it follows that the change in colony distribution seen with changes in EGF concentration is due to a change in type 2 dynamics. R-Spondin too, Figure 4.8 Effect of EGF and R-spondin on keratinocyte population distributions. (A) Population distribution of keratinocytes in EGF0 at day 7 (n=1682, 787 and 689). (B) Comparison of entire day 7 population distribution in EGF0, EGF5, EGF20 and R-Spondin compared with EGF10. All distributions were different from EGF10 (Kolmogorov-Smirnov vs. EGF0 p=0.0015, vs. EGF5 p<0.0001, vs. EGF20 p<0.0001, vs. R-Spondin p<0.0001). (C) Proportions of 2-, 3- , and 4-cell colonies were com- pared in the same populations versus similar sized colonies in EGF10. No signifi- cant difference was seen. (Wilcoxon matched pairs vs. EGF0 p=0.75, vs. EGF5 p=0.25, vs. EGF20 p=0.25, vs. R-Spondin p=0.25). 128 C A B Colony distribution at day 7 in EGF0 2 3-4 5-8 9-1 6 17- 32 33- 64 65- 128 129 -25 6 257 -51 2 513 -10 24 102 5-2 048 0 5 10 15 20 25 EGF10 EGF0 0.0 0.1 0.2 0.3 p=ns Pr op or tio n EGF10 EGF5 0.0 0.1 0.2 0.3 p=ns Pr op or tio n EGF10 EGF200.0 0.1 0.2 0.3 p=ns Pr op or tio n EGF10 R-Spo 0.0 0.1 0.2 0.3 p=ns Pr op or tio n EGF10 EGF0 2 16 128 1024 p=** Cl on e s ize (L og 2) EGF10 EGF5 Cl on e s ize (L og 2) p=**** 2 16 128 1024 EGF10 EGF20 p=**** 2 16 128 1024 Cl on e s ize (L og 2) EGF10 R-Spo p=**** 2 16 128 1024 Cl on e s ize (L og 2) Comparison of entire populations Comparison of 2-, 3-, and 4-cell colonies EGF10 vs. EGF0 EGF10 vs.EGF5 EGF10 vs.EGF20 EGF10 vs.R-Spondin Colony size Pr op or tio n 129 changes colony size distribution of keratinocyte populations without changing the proportions of 2-, 3- and 4-cell colonies. 4.7 Discussion In this chapter I have shown that keratinocyte populations respond to changes in EGF by changing their cell fate, and not their cell cycle time. Direct observation of the keratinocyte colonies in the absence of EGF demonstrates that there is a negligible effect on the type 1 colony dynamics. In contrast, type 2 colonies showed an early switch to stochastic fate within their centres, and developed a narrower rim of exponentially dividing cells. Finally, the signatures of type 1 growth in the distribution of 2-, 3- and 4- cell colonies did not show any change for varying levels of EGF, and R-Spondin. In trying to perturb the dynamics of cell fate choice in culture using EGF, I have found that type 1 dynamics is relatively insensitive to varying levels of EGF or to R- Spondin. A recent paper has shown that small keratinocyte clones respond to higher levels of EGF (30ng/mL) given late in culture by shrinking in the first few hours after administration (Nanba et al. 2013). I find that at later time points, this is not sustained. The robustness of Type 1 dynamics argues that it may have a mechanistic role to play in epidermal homeostasis. Type 2 colonies, on the other hand, has been recently shown to undergo clonal conversion with continuous Rac1 inhibition suggesting that this is much more sensitive to external environmental cues (Nanba et al. 2013). While balanced type 1 dynamics is seen in murine inter-follicular epidermis, the ratios are different to those seen in human in vitro cultures (a:b:c=10:80:10 versus a:b:c of 30:40:30 respectively) (Clayton et al. 2007; Mascré et al. 2012). It is not clear why this is the case. A larger proportion of divisions resulting in asymmetric divisions in type 1 colonies will result in a larger maximum colony size. Thus 80% PD divisions as seen in vivo, are likely to maintain larger maximum colony sizes than can be reached by cells dividing in a balanced stochastic manner with 40% PD divisions as seen in vitro. This indicates that in homeostatic epidermis in vivo larger colonies are likely to be maintained than those in culture conditions. This is an avenue that warrants further exploration. Where cell replacement in excess of homeostasis is required, as in the case of re- epithelialisation following major burns, type 1 cell fate alone cannot provide the increase in cell division. Type 2 dynamics on the other hand, has the ability to increase cell division exponentially. The ability of EGF to increase the keratinocyte colony area, 130 131 and recruit larger rims of dividing cells within colonies reconciles the linear radial growth rate of such colonies with cellular fate dynamics (Barrandon & Green 1987). Blister fluid from both superficial wounds and superficial burns have increased levels of heparin-binding EGF (HB-EGF), platelet derived growth factor (PDGF) and TGF-α, although not EGF (Marikovsky et al. 1993; Ono et al. 1994; Ono et al. 1995). Both HP- EGF and TGF-α function as autocrine growth factors in normal human keratinocytes culture (Coffey Jr et al. 1987; Hashimoto et al. 1994; Shirakata et al. 2005). TGF-α has been seen to have effects on human keratinocytes in a manner similar to EGF, albeit with greater potency (Barrandon & Green 1987). This suggests that such growth factors in a large wound healing environment may contribute to the wound healing response by increasing recruitment of exponentially dividing cells. As a consequence of the effects of EGF on keratinocyte growth, great promise was held by EGF therapies in burns. While there is a 15-20% increase in re-epithelialisation rates seen with high doses of EGF administered through silver sulfadiazine creams (10!g/mL), the benefits are not large enough for routine use clinical burns treatment (Brown et al. 1989). Similar results have been seen with TGF-α (Schultz et al. 1987). Limited benefits from trials of EGF in chronic wounds too, have resulted in limited use, although some improvement is likely to be made by improved drug delivery (Falanga et al. 1992). This suggests that the recruitment of type 2 cell fate has limits. While some cells may retain the capacity to switch to type 2 cell fate, not all cells are likely to have this capacity, thereby limiting the clinical effects seen. The combined used of live imaging and combinatorial statistics in this chapter also highlights its power to understand the effects of growth factors on keratinocytes. Understanding changes in cell fate in culture could be an effective screening strategy prior to expensive in vivo testing. Improved ability to automatically trace live imaged cell monocultures have led to advances in understanding of fate in certain cell types, although for shorter time frames (Eilken et al. 2009; Gomes et al. 2011; He et al. 2012; Schroeder 2013). Computer algorithms recognising co-cultured cells, and being able to track for longer periods would pave the way for wider use of this versatile technique. Figure 5.1 Cell proliferation in wounded oesophageal epithelium. (A) Timeline of experimental protocol. Wounded wild type C57/Bl6 mice were culled at 2 and 5 days post wounding. Animals were treated with EdU 24 hours prior to timepoints. (B) At 48 hours after wounding, wounds were still open, completely closing by 5 days (inset higher magnification). (C) Immunostaining of epithelial wholemounts showing widespread and uniform EdU uptake at 2 days in the proliferating zone (PZ) around the wound and little uptake in the migratory front (MF). (D) High power view showing cell pairs labelled by EdU in the controls at 48 hours, uptake of EdU in the PZ of wounded oesophagi, and return to control levels of EdU adja- cent to the healed wounds at 5 days. Images C-D from Dr. Maria Alcolea. Scale bars (B & C) 500ۚm, (D) 50ۚm. 132 C A B Wild type C57/Bl6 0 1 5d Wound 2 4 EdU EdU EdU EdU DAPI EdUDAPI EdU 48h, PZControl 5d, Healed MF PZ PZ D 48 hours 5 days 48 hours 5 days 133 CHAPTER 5 CELL PROLIFERATION IN INJURED MOUSE OESOPHAGEAL EPITHELIUM 5.1 Chapter Overview In the murine oesophagus, quantitative cell fate analysis across the epithelium has shown that homeostasis is maintained by dividing cells with balanced stochastic fate (Doupé et al. 2012). Although there is no contribution from a slow cycling population in homeostasis, such quiescent cells may be recruited upon injury. In this chapter I set out experiments that looked at the re-epithelialisation of murine oesophagus following superficial microendoscopic biopsy. Some figures have been published (Appendix 1). For this chapter, experiments were done in collaboration with Dr. Maria Alcolea, who performed the wholemount immunostaining and imaging, apart from those in Figure 5.3. Section 5.2 uses EdU uptake in wounded epithelium to detect dividing cells. Section 5.3 studied the contribution of cell populations across the epithelium following wound repair compared to controls via dilution of HGFP. Section 5.4 explored the contribution of clonally labelled keratinocytes that were adjacent to the biopsy wound. Finally, Section 5.5 discusses a few key implications of my findings. 5.2 Cell Proliferation in Wounded Oesophagus To study patterns of active cells following wounding, I used wild type C57/Bl6 mice of 3-6 months in age. Parallel cohorts of experimental and control animals were wounded using my endoscopic technique (Figure 5.1A). Animals were culled at 2 days and 5 days following wounding, after being injected with EdU 24 hours prior to culling. Oesophageal wholemounts were harvested and stained for DAPI and EdU. After an initial weight loss following the procedure, all animals regained their pre- procedure weight and remained alert and active until harvest. The oesophageal wounds showed a typical pattern of wound repair (Gurtner et al. 2008). Macroscopically, wounds were still open at 48 hours, and closed completely by 5 days (Figure 5.1B). At 48 hours, microscopy revealed extensive recruitment of cells in to cycle in an area closely surrounding, but not adjacent to the wound edges (Figure 5.1C). This has been described in the literature as a proliferating zone (PZ) of cells that Figure 5.2 Uniform contribution of cells to oesophageal wound repair. (A) Time- line of experimental protocol. R26M2rtTA/TetO-HGFP mice expressed HGFP upon doxycycline induction in all cells. Cohorts were culled at 2 and 5 days post wound- ing. (B) Low magnification showing retained HGFP in the migrating front, and uniform dilution in the proliferating zone. A similar pattern, albeit with further dilution was seen at 5 days. (C&D) High magnification images taken with the same exposure settings. Control oesophagi had uniform retention of HGFP indicating slow turnover. PZ showed uniform dilution at 2 days, that subsequently progressed by 5 days. Inset of 5 day PZ field showing a Langerin positive HGFP bright cell. Images from Dr. Maria Alcolea. Scale bars (B) 500ۚm, (C) 50ۚm. 134 C A B R26M2rtTA/TetO-HGFP -28d 0d 5d Wound 2d Doxycycline induction V HGFP 2d, PZ2d, Control 5d, PZ MF PZ 48 hours 5 days HGFP MF PZ HGFP Langerin 5d, ControlD 135 contribute to wound repair, while an area adjacent to the wound with infrequent cell divisions constitutes a migratory front (MF) (Potten & Allen 1975). This recruitment of cells in to cycle in the PZ was much higher compared with unwounded controls at 48 hours (Figure 5.1D). In the controls, scattered labelling of cell pairs indicated that uniform proliferation at slow rates occurred with single labelled cells having divided once during the 24 hour period of EdU administration. By 5 days, in contrast, EdU uptake in cells adjacent to the healed wound returned to levels comparable to the unwounded controls. Pairs of EdU positive cells were seen throughout the epithelium, comparable to homeostatic epithelium (Figure 5.1D) Taken together, this indicates that there was a widespread contribution to the cellular proliferation response following wounding. This occurred predominantly in a proliferating zone of cells separated from the wound edge by a relatively quiet migratory front. Proliferative recruitment returned to control levels following early stages of re-epithelialisation. 5.3 Dilution of HGFP in Wounded Mouse Oesophagus To further investigate if cell contribution to wound healing was uniform across the epithelium, or if some cells contributed more than others, I took a complementary approach studying HGFP dilution. I first labelled all the cells of the basal layer of the oesophageal epithelium using doxycycline induction of Rosa26M2rtTA/TetO-HGFP mice. Withdrawal of the doxycycline results in halting HGFP transcription, after which time the protein is diluted by division in proliferating cells but retained in slow cycling cells. Cohorts of induced mice underwent oesophageal biopsy, and samples were harvested at 2 days and 5 days as before (Figure 5.2A). In contrast to the EdU experiment, dividing cells dilute label resulting in an inverse relationship between label retention and cell division. With successive cell division, HGFP was progressively diluted in daughters of dividing cells, while remaining undiluted in non-dividing cell populations. At 2 days after biopsy, I found that HGFP was substantially and evenly diluted within the proliferating zone of the wounded epithelium compared with unwounded controls (Figure 5.2 B&C). The migrating front retained HGFP indicating that substantially fewer divisions occurred in this zone. HGFP fluorescence in areas distant from the wound (>10mm) was similar to unwounded controls suggesting that these areas are not recruited to turnover cells at a higher rate. The wound edges in the Figure 5.3 Rapid and uniform dilution of HGFP during wound repair. (A) H2BGFP was induced in Rosa26M2rtTA/TetO-HGFP mice similar to Figure 5.2, and timepoints were taken up to 14 days. Entire oesophageal imaging was done using a standardised exposure. Figure shows composite entire oesophagus at 12hours and 7 days with inset showing wound. Overexposed 7d oesophagus to show orientation (inset) (B) Migrating front shows slow HGFP dilution, compared with unwounded controls. (C) Proliferating zone dilution of HGFP is rapid with changes as soon as 12 hours post wounding. Overexposure shows dilution is more severe than unwounded controls at 7 days. (D) Unwounded controls showed uniform dilution of HGFP for comparison. Scale bars (A) 1mm and (B-D) 250ۚm. 136 D A C B Proliferating Zone dilution 3h 6h 12h 24h 48h 72h 4d 7d 5x exposure Std exposure Control dilution 3h 6h 12h 24h 48h 72h 4d 7d Migrating front dilution 3h 6h 12h 24h 48h 72h 4d 7d 5x ex po su re St d ex po su re 5x exposure Std exposure 12h 7d 14d 137 HGFP sample illustrated in the figure have already just opposed by 2 days (Figure 5.2B) At 5 days after biopsy, dilution had progressed becoming marked in the proliferating zone compared to either the migratory front (which retained HGFP), and distant areas (which diluted HGFP at rates comparable with unwounded controls (Figure 5.2 B&C). Dilution within the proliferating zone was uniform, apart from retention in non-keratinocyte lineages indicating that all keratinocytes divided at a uniform rate in this area contributing to re-epithelialisation. As an example of this, a Langerhans cell is shown with label retention and CD207 (Langerin) positivity in the inset of Figure 5.2C. To further elucidate the time course of cellular recruitment to proliferation following wounding, I performed a more detailed time course of dilution (Figure 5.3 A-D). Here Rosa26M2rtTA/TetO-HGFP mice were induced to express HGFP by 4- week treatment with doxycycline. Cohorts of animals underwent anaesthesia and oesophageal biopsy, with samples taken at 3 hours, 6 hours, 12 hours, 24 hours, 2 days, 3 days, 4 days and 7 days. Control cohorts were induced, and underwent anaesthesia, but no procedure, and were taken at the same timepoints, but with additional samples at 10 days and 14 days. Here, I found that the proliferating zone underwent rapid recruitment of cell division, with HGFP dilution being noticeable at 12 hours, and marked by 24 hours (Figure 5.3 C&D). By 48 hours, HGFP dilution was comparable to 4-day dilution in unwounded controls. By 4 days in the proliferating zone, dilution was comparable to 14-day dilution in unwounded controls (Figure 5.3 C&D). This rate of dilution is equivalent to three extra rounds of division compared to normal dilution at homoeostatic rates of cell division. Although there was further dilution between 4 days and 7 days in wounded oesophagi, this was not as marked as prior to 4 days (Figure 5.3 C&D). In contrast, HGFP in the migratory front remained brighter than matched controls through out the 7 day time-course, although there was a slower rate of dilution (Figure 5.3 B&D). Taken together, this suggested that rapid widespread cell division occurred in the proliferating zone between 12 hours-4 days after wounding at 2-3 times normal cell turnover. By 7 days, the proliferative recruitment had waned. 5.4 Wounding of Clonally Labelled Oesophagus In Vivo To investigate the contribution of oesophageal progenitor cells to wound healing, I made use of clonal genetic labelling in AhcreERTR26flEYFP/wt mice. Low frequency clonal Figure 5.4 Clonal contribution to wound repair in oesophageal epithelium. (A) Timeline of experimental protocol. AhcreERTR26flEYFP/wt mice label individual cells and their subsequent progeny upon induction at clonal density. Cohorts were culled at 1 and 10 days post wounding, having received EdU 1 hour prior to culling. (B) Low magnification image at 1 day showing migrating colonies fragmented by movement in the proliferating zone (PZ). High power view showing fragmented colony. Arrows showing direction of migration. (C) Low magnification image at 10 days showing fragmented colonies with a Christmas tree appearance due to cell divisions. Inset showing labelled colony. (D) Unwounded controls showing single and double cell YFP labelling at day 1, and small clone sizes at day 10 characteristic of homeostatic epithelium. Images from Dr. Maria Alcolea, Scale bars 50ۚm. 138 C A B AhcreERTR26flEYFP/wt -7d 0d 10d Wound 1d Label K14 YFP EdU EdUEdU 1 day 10 days 10 days Control 10 days Control 1 day wound PZ MF D inset 1 day PZ MF inset 139 labelling was induced 1 week before endoscopic biopsy. Whole oesophagi were harvested at timepoints of 1 day and 10 days following biopsy, with a pulse of EdU given 1 hour prior to harvest (Figure 5.4A). At 24 hours following wounding, fragmented trails of clones were seen aligned in the direction of the wound edge, and spanning the junction of the proliferating zone and the migratory front (Figure 5.4B). These clones, highlighted at higher magnification were a largely a single cell across, and several cells in length, fragmented by gaps of a about a cell (Figure 5.4B inset). This indicated that rapid recruitment, and rapid increase in cell turnover was coupled with migration of daughter cells towards closing the epithelial defect. The proliferative zone indicated by the area showing high EdU positivity was apparent even with the short 1 hour pulse given. At 10 days following wounding, the wounds had completely healed, but could be easily identified by the disrupted pattern of clustered cells marking the restoration of epithelial integrity (Figure 5.4C). The YFP labelled colonies of cells spanning this area showed a “Christmas tree” appearance. This was due to the fragmented linear column of migrating cells dividing laterally once the wound edges had been opposed (Figure 5.4C inset). The orientation of these colonies, as before indicated the direction of travel in relation to the wound edge. EdU uptake was much less pronounced than before, with levels similar to unwounded controls. Control homeostatic unwounded oesophageal epithelium showed characteristic single or 2-cell colonies at 24 hours post induction, and a wide distribution of small colonies at 10 days. EdU uptake in these wholemounts was scattered throughout, and remained unchanged at the two time points (Figure 5.4D). As EYFP labelled cells in this system had been previously shown to be proliferating progenitors, I concluded that these progenitor cells participate in tissue regeneration after wounding (Doupé et al. 2012). 5.5 Discussion In this chapter I have shown with three complementary cell labelling strategies that dividing progenitors participate in wound regeneration, and that participation of cells across the proliferating zone is rapid and uniform. For repair of biopsy wounds, no recourse is needed to a specialised subset of cells that have different rates of proliferation compared to the surrounding cells. 140 141 Epithelial tissues are continually subject to trauma ranging from minor friction to major lacerations. To heal these insults, epithelial populations must have the ability for the transient generation of large numbers of cells above that normally required for homeostasis. This role for transient massive expansion is usually attributed to activation of a resident slow cycling stem cell, although nature of the cells involved in this regeneration, and their relative contributions to different types of injury has not previously been resolved. In the mouse oesophageal epithelium, BrdU label retention assays have been reported to identify a subpopulation of cells that reconstitute the epithelium in organotypic cultures (Kalabis et al. 2008). In my wounding studies in vivo, no label retaining population seems to contribute to the repair of the esophageal epithelium. Additionally, label retention in the basal layer was seen to be a feature of non- keratinocyte populations predominantly expressing CD45 (Doupé et al. 2012). Multiple recent studies across different epithelia have shown that a variety of cell populations are involved in epithelial repair. In the skin, both interfollicular slow- cycling cells and stem cells from epidermal appendages contribute to wound repair (Tumbar et al. 2004; Ito et al. 2005; Levy et al. 2007; Snippert et al. 2010; Mascré et al. 2012; Lu et al. 2012). Additionally, differentiated suprabasal keratinocytes have been widely reported to contribute to wound repair, at least in early stages (Potten et al. 2000; Usui et al. 2005). In the oesophageal epithelium, experiments in this chapter show that epithelia can be maintained without the presence of a reserve cell with different proliferating dynamics (Doupé et al. 2012). In the intestine, early differentiating precursor cells can revert to regenerate the epithelium following radiation induced injury (Buczacki et al. 2013). Accelerated dedifferentiation of Troy+ chief cells in response to injury reveals plasticity of the wound response in the stomach epithelium (Stange et al. 2013). A recent report highlights that the dedifferentiation of airway secretory Clara cells can renew both multi-ciliated and secretory cell types following injury (Tata et al. 2013). Taken together, these studies highlight that cellular response to wounding in epithelia is undertaken by multiple cell types with diverse functions in normal homeostasis. 142 143 CHAPTER 6 CONCLUSIONS AND OUTLOOK 6.1 Key Lessons In this dissertation, I have demonstrated the power of using live imaging combined with novel statistical methodology to unravel cell fate choices in human keratinocytes in vitro. I have found that human keratinocytes have a narrow cell cycle time distribution that does not change with changes in the supplemented EGF or R-Spondin. With a narrow cell cycle distribution, any variations in colony size are due to cell division outcome choices. Analysis of cell divisions across the entire population reveals there are two, and only two, modes of cell fate. These are responsible for the wide range of colony sizes observed in human keratinocyte cultures. Additionally, cell fate is not fixed, and may change as seen in the centre of large keratinocyte colonies. Type 1 cell fate choices are seen to be robust and do not change with varying levels of EGF or R- spondin. The demonstration that Type 1 cell divisions occur in human keratinocytes parallels observations in mouse IFE in vivo, p53 mutant cell dynamics in human sun- damaged IFE, and viral clone labelling studies in human xenograft experiments. I have also resolved the cell contributions made by keratinocytes to wound healing in the mouse esophageal epithelium in vivo. Here, I have developed a technically challenging technique to take a superficial biopsy of the mouse oesophageal epithelium alone, allowing studies of repair to be carried out. The mouse oesophageal epithelium does not contain keratinocyte LRCs, is maintained by stochastically dividing progenitors, and is devoid of stem cell niches. In this tissue, I find repair is taken on by actively dividing progenitors without the need for recruitment of a specialised quiescent stem cell. 6.2 Outlook: Open Questions and Future Experiments Several exciting questions remain for exploration. 6.2.1 Molecular Mechanisms Underlying Cell Fate Cell fate is clearly defined by direct visualisation. However, the identification of such behaviour does not reveal the molecular mechanisms by which cell fate decisions are made and adjusted in response to external cues. As type 1 stochastic behavior is 144 145 relatively resistant to external growth factors, it is of interest to understand the molecular mechanisms underlying such fate. Such mechanisms are likely to be conserved as this type of behaviour is seen in both murine and human keratinocytes, although ratios of balance (10:80:10) versus (35:30:35) may reveal further insights. One possible mechanism for such differences may lie in the inherent stochasticity of transcription, as demonstrated by the discovery that key regulatory genes oscillate in normal cells. The fate outcomes may vary depending on the level of gene oscillation at the time of cell division (Shimojo et al. 2008; Hoffmann et al. 2008). 6.2.2 Lineage Tracing in Human Xenografts: Steady state Ex Vivo Genetic lineage tracing in mouse has revealed much about cell fate in vivo. To be able to conduct such studies in vivo in man has major practical obstacles. To circumvent this, ex vivo introduction of inducible viral labelling in single keratinocytes within human epidermal sheets, and subsequent xenografting to mice may provide insights. A period of time between grafting and inducing the label would allow wound responses to settle, before tracking individual clone contributions to xenograft maintenance. 6.2.3 Cell Fate Dynamics in Mutant Clones and Carcinogenesis Additional resources should be put to investigate cell fate choices in mutant clones in human epidermis. My work demonstrates that a population of cells with stochastic type 1 fate is seen in normal human keratinocytes. With work in human sun-damaged p53 mutant clone distributions demonstrating that such stochastic fate becomes biased with ongoing sun-exposure, it would be interesting to investigate the underlying mechanisms for this. A next generation sequencing approach could allow determining the genetic differences in pre-neoplastic clones that allow such a bias to occur. This has important medical applications. 6.2.4 Automated Long Term Imaging In Vivo In vitro studies are ideally placed to utilise the advances in live cell imaging of cells. Two problems remain that are starting to be addressed. Firstly, automated methods of cell identification, especially in co-cultured cells for long periods allow a scale that manual methods do not afford. Secondly, practical limitations such as length of anaesthesia currently limit in vivo applications to a few cell cycles at best. Addressing both these two issues would lead to resolution of many outstanding questions of cell fate choices during homeostasis and perturbation. 146 147 REFERENCES Adami JG. (1901) The causation of cancerous and other new growths. Brit Med J 1(2098), 621–8. Aigner M. (1998) Motzkin Numbers. Eur J Comb 19(6), 663–675. Alcolea MP & Jones PH. (2013) Tracking cells in their native habitat: lineage tracing in epithelial neoplasia. Nat Rev Cancer 13(3), 161–71. Arwert EN, Hoste E & Watt FM. (2012) Epithelial stem cells, wound healing and cancer. Nat Rev Cancer 12(3), 170–180. Avruch J, Khokhlatchev A, Kyriakis JM, Luo Z, Tzivion G, Vavvas D & Zhang X-F. (2001) Ras activation of the Raf kinase: Tyrosine kinase recruitment of the MAP kinase cascade. Recent Prog Horm Res 56, 127–155. Barker N, van Es JH, Kuipers J, Kujala P, van den Born M, Cozijnsen M, Haegebarth A, Korving J, Begthel H, Peters PJ & Clevers H. (2007) Identification of stem cells in small intestine and colon by marker gene Lgr5. Nature 449(7165), 1003–1007. Barrandon Y, Grasset N, Zaffalon A, Gorostidi F, Claudinot S, Droz-Georget SL, Nanba D & Rochat A. (2012) Capturing epidermal stemness for regenerative medicine. Semin Cell Dev Biol 23, 937–944. Barrandon Y & Green H. (1987a) Cell migration is essential for sustained growth of keratinocyte colonies!: The roles of Transforming Growth Factor-! and Epidermal Growth Factor. Cell 50(7), 1131–7. Barrandon Y & Green H. (1987b) Three clonal types of keratinocyte with different capacities for multiplication. Proc Natl Acad Sci U S A 84(8), 2302–6. Barrandon Y, Morgan JR, Mulligan RC & Green H. (1989) Restoration of growth potential in paraclones of human keratinocytes by a viral oncogene. Proc Natl Acad Sci U S A 86(11), 4102–6. Beguinot L, Lyall RM, Willingham MC & Pastan I. (1984) Down-regulation of the epidermal growth factor receptor in KB cells is due to receptor internalization and subsequent degradation in lysosomes. Proc Natl Acad Sci U S A 81(8), 2384–8. 148 Benitah SA, Frye M, Glogauer M & Watt FM. (2005) Stem cell depletion through epidermal deletion of Rac1. Science 309(5736), 933–5. Berenblum I. (1954) A speculative review: the probable nature of promoting action and its significance in the understanding of the mechanism of carcinogenesis. Cancer Res 14(7), 471–477. Billingham RE & Medawar PB. (1951) The technique of free skin grafting in mammals. J Exp Biol 28(3), 385–402. Blanpain C & Simons BD. (2013) Unravelling stem cell dynamics by lineage tracing. Nat Rev Mol Cell Biol 14(8), 489–502. Branca A. (1912) Le tégument externe et ses dérivés. In Traité d’Anat Humaine. Paris: Masson et Cie, pp. 102–103. Braun KM, Niemann C, Jensen UB, Sundberg JP, Silva-Vargas V & Watt FM. (2003) Manipulation of stem cell proliferation and lineage commitment: visualisation of label-retaining cells in wholemounts of mouse epidermis. Development 130(21), 5241–55. Brown GL, Nanney LB, Griffen J, Cramer AB, Yancey JM, Curtsinger LJ, Holtzin L, Schultz GB, Jurkiewicz MJ & Lynch JB. (1989) Enhancement of wound healing by tropical treatment with epidermal growth factor. New Engl J Med 321(2), 76–79. Buczacki SJA, Zecchini HI, Nicholson AM, Russell R, Vermeulen L, Kemp R & Winton DJ. (2013) Intestinal label-retaining cells are secretory precursors expressing Lgr5. Nature 495(7439), 65–69. Buhrow SA, Cohen S & Staros J V. (1982) Affinity labeling of the protein kinase associated with the epidermal growth factor. J Biol Chem 257(8), 4019–4022. Burgering BMT & Coffer PJ. (1995) Protein kinase B (c-Akt) in phosphatidylinositol-3- OH kinase signal transduction. Nature 376(599-602). Camplejohn RS. (1983) Flash and double labelling experiments in human epidermis. In N. A. Wright & R. S. Camplejohn, eds. Psoriasis: Cell Proliferation. London: Chuchill Livingstone, pp. 13–23. Candi E, Schmidt R & Melino G. (2005) The cornified envelope: a model of cell death in the skin. Nat Rev Mol Cell Biol 6(4), 328–40. Cantley LC. (2002) The phosphoinositide 3-kinase pathway. Science 296(5573), 1655– 1657. 149 Cardone MH, Roy N, Stennicke HR, Salvesen GS, Franke TF, Stanbridge S, Frisch S & Reed JC. (1998) Regulation of cell death protease Caspase-9 by phosphorylation. Science 282, 1318–1321. Carpenter CL, Augerll KR, Chanudhuri M, Yoakim M, Schaffhausen B, Shoelsonll S & Cantley LC. (1993) Phosphoinositide 3-Kinase is activated by phosphopeptides that bind to the SH2 domains of the 85-kDa subunit. J Biol Chem 268(13), 9478– 9483. Carpenter G, Lembach KJ, Morrison MM & Cohen S. (1975) Characterization of the binding of 125-I-labeled epidermal growth factor to human fibroblasts. J Biol Chem 250, 4297–4304. Castilho RM, Squarize CH, Patel V, Millar SE, Zheng Y, Molinolo a & Gutkind JS. (2007) Requirement of Rac1 distinguishes follicular from interfollicular epithelial stem cells. Oncogene 26(35), 5078–85. Chen CS, Lavker RM, Rodeck U, Risse B & Jensen PJ. (1995) Use of a serum-free epidermal culture model to show deleterious effects of epidermal growth factor on morphogenesis and differentiation. J Invest Dermatol 104(1), 107–112. Choi N, Zhang B, Zhang L, Ittmann M & Xin L. (2012) Adult murine prostate basal and luminal cells are self-sustained lineages that can both serve as targets for prostate cancer initiation. Cancer Cell 21(2), 253–65. Chrostek A, Wu X, Quondamatteo F, Hu R, Sanecka A, Niemann C, Langbein L, Haase I & Brakebusch C. (2006) Rac1 is crucial for hair follicle integrity but is not essential for maintenance of the epidermis. Mol Cell Biol 26(18), 6957–6970. Cicalese A, Bonizzi G, Pasi CE, Faretta M, Ronzoni S, Giulini B, Brisken C, Minucci S, Di Fiore PP & Pelicci PG. (2009) The tumor suppressor p53 regulates polarity of self-renewing divisions in mammary stem cells. Cell 138(6), 1083–95. Citri A & Yarden Y. (2006) EGF-ERBB signalling: towards the systems level. Nat Rev Mol Cell Biol 7(7), 505–516. Claudinot S, Nicolas M, Oshima H, Rochat A & Barrandon Y. (2005) Long-term renewal of hair follicles from clonogenic multipotent stem cells. Proc Natl Acad Sci U S A 102(41), 14677–82. Clayton E, Doupé DP, Klein AM, Winton DJ, Simons BD & Jones PH. (2007) A single type of progenitor cell maintains normal epidermis. Nature 446(7132), 185–9. Clevers H & Nusse R. (2012) Wnt/"-catenin signaling and disease. Cell 149(6), 1192– 205. 150 Coffey Jr RJ, Derynck R, Wilcox JN, Bringman TS, Goustin AS, Moses HL & Pittelkow MR. (1987) Production and auto-induction of transforming growth factor-! in human keratinocytes. Nature 328(6133), 817–820. Cohen AR, Gomes FLAF, Roysam B & Cayouette M. (2010) Computational prediction of neural progenitor cell fates. Nat Methods 7(3), 213–8. Cohen S. (1962) Isolation of a mouse submaxillary gland protein accelerating incisor eruption and eyelid opening in the new-born animal. J Biol Chem 237, 1555–1562. Cohen S. (1965) The stimulation of epidermal proliferation by a specific protein (EGF). Dev Biol 12(3), 394–407. Cohen S & Elliott GA. (1963) The stimulation of epidermal keratinization by a protein isolated from the submaxillary gland of the mouse. J Invest Dermatol 40, 1–5. Compton CC, Gill JM, Bradford DA, Regauer S, Gallico GG & O’Connor NE. (1989) Skin regenerated from cultured epithelial autografts on full-thickness burn wounds from 6 days to 5 years after grafting. A light, electron microscope and immunohistochemical study. Lab Investig 69(5), 600–12. Dasgupta R & Fuchs E. (1999) Multiple roles for activated LEF/TCF transcription complexes during hair follicle development and differentiation. Development 126, 4557–4568. Datta SR, Dudek H, Tao X, Masters S, Fu H, Gotoh Y & Greenberg ME. (1997) Akt phosphorylation of BAD couples survival signals to the cell-intrinsic death machinery. Cell 91, 231–241. Diehl JA, Cheng M & Roussel MF. (1998) Glycogen synthase kinase-3" regulates cyclin D1 proteolysis and subcellular localization. Genes Dev 12(22), 3499–3511. Donaghey R & Shapiro LW. (1977) Motzkin Numbers. J Comb Theory, Ser A 23(3), 291– 301. Doupé DP, Alcolea MP, Roshan A, Zhang G, Klein AM, Simons BD & Jones PH. (2012) A single progenitor population switches behavior to maintain and repair esophageal epithelium. Science 337(6098), 1091–1093. Doupé DP, Klein AM, Simons BD & Jones PH. (2010) The ordered architecture of murine ear epidermis is maintained by progenitor cells with random fate. Dev Cell 18(2), 317–23. Dover R & Potten CS. (1988) Heterogeneity and cell cycle analyses from time-lapse studies of human keratinocytes in vitro. J Cell Sci 89, 359–364. 151 Downward J. (2003) Targeting Ras signalling pathways in cancer therapy. Nat Rev Cancer 3, 11–22. Eckert RL, Broome A-M, Ruse M, Robinson N, Ryan D & Lee K. (2004) S100 proteins in the epidermis. J Invest Dermatol 123(1), 23–33. Eilken HM, Nishikawa S-I & Schroeder T. (2009) Continuous single-cell imaging of blood generation from haemogenic endothelium. Nature 457(7231), 896–900. Van Es JH, Sato T, van de Wetering M, Lyubimova A, Nee ANY, Gregorieff A, Sasaki N, Zeinstra L, van den Born M, Korving J, Martens ACM, Barker N, van Oudenaarden A & Clevers H. (2012) Dll1+ secretory progenitor cells revert to stem cells upon crypt damage. Nat Cell Biol 14(10), 1099–104. Estrach S, Legg J & Watt FM. (2007) Syntenin mediates Delta1-induced cohesiveness of epidermal stem cells in culture. J Cell Sci 120(16), 2944–2952. Ewing JS. (1909) Cancer Problems. In E. O. Jordan, ed. The Harvey Lectures 1907-1908. Philadelphia: J B Lippincott Co, pp. 34–88. Falanga V, Eaglstein WH, Bucalo B, Katz MH, Harris B & Carson P. (1992) Topical use of human recombinant epidermal growth factor (h-EGF) in venous ulcers. J Dermatol Surg Oncol 18(7), 604–6. Fischer SM, Viaje A, Mills GD & Slaga TJ. (1980) Explant methods for epidermal cell culture. Methods Cell Biol 21, 207–227. Flemming W. (1882) Zellsubstanz, kern and zelltheilung, Leipzig: F C W Vogel. Franklin JD & Lynch JB. (1979) Effects of topical application of epidermal growth factor on wound healing: Experimental study on rabbit ears. Plast Reconstr Surg 64(6), 766–770. Fuchs E. (2007) Scratching the surface of skin development. Nature 445(7130), 834–42. Fuchs E & Green H. (1980) Changes in keratin gene expression during terminal differentiation of the keratinocyte. Cell 19, 1033–1042. Gallico GG, O’Connor NE, Compton CC, Kehinde O & Green H. (1984) Permanant coverage of large burn wounds with autologous cultured human epithelium. New Engl J Med 311(7), 448–451. Ghadially R. (2012) 25 Years of Epidermal Stem Cell Research. J Invest Dermatol 132(3), 797–810. 152 Ghazizadeh S & Taichman LB. (2001) Multiple classes of stem cells in cutaneous epithelium: a lineage analysis of adult mouse skin. EMBO J 20(6), 1215–22. Ghazizadeh S & Taichman LB. (2005) Organization of stem cells and their progeny in human epidermis. J Invest Dermatol 124(2), 367–72. Giangreco A, Goldie SJ, Failla V, Saintigny G & Watt FM. (2010) Human skin aging is associated with reduced expression of the stem cell markers "1 integrin and MCSP. J Invest Dermatol 130(2), 604–8. Gomes FLAF, Zhang G, Carbonell F, Correa JA, Harris WA, Simons BD & Cayouette M. (2011) Reconstruction of rat retinal progenitor cell lineages in vitro reveals a surprising degree of stochasticity in cell fate decisions. Development 138(2), 227– 35. Gomez C, Chua W, Miremadi A, Quist S, Headon DJ & Watt FM. (2013) The interfollicular epidermis of adult mouse tail comprises of two distinct cell lineages that are differentially regulated by Wnt, Edaradd and Lrig1. Stem Cell Reports 1, 19–27. Green H. (1978) Cyclic AMP in relation to proliferation of the epidermal cell!: a new view. Cell 15(3), 801–811. Green H. (2008) The birth of therapy with cultured cells. BioEssays 30(9), 897–903. Green H. (1980) The keratinocyte as differentiated cell type. In The Harvey Lectures, Series 74. New York: Academic Press, pp. 101–139. Grzeschik K-H, Bornholdt D, Oeffner F, König A, del Carmen Boente M, Enders H, Fritz B, Hertl M, Grasshoff U, Höfling K, Oji V, Paradisi M, Schuchardt C, Szalai Z, Tadini G, Traupe H & Happle R. (2007) Deficiency of PORCN, a regulator of Wnt signaling, is associated with focal dermal hypoplasia. Nat Genet 39(7), 833–5. Gudjonsson JE, Johnston A, Dyson M, Valdimarsson H & Elder JT. (2007) Mouse models of psoriasis. J Invest Dermatol 127(6), 1292–308. Gurtner GC, Werner S, Barrandon Y & Longaker MT. (2008) Wound repair and regeneration. Nature 453(7193), 314–21. Guy PM, Platko J V, Cantley LC, Cerione RA & Carraway III KL. (1994) Insect cell- expressed p180erbB3 possesses an impaired tyrosine kinase activity. Proc Natl Acad Sci U S A 91, 8132–8136. Hallberg B, Rayter SI & Downward J. (1994) Interaction of Ras and Raf in intact mammalian cells upon extracellular stimulation. J Biol Chem 269(6), 3913–3916. 153 Halprin KM. (1972) Epidermal “turnover time”: A re-examination. Brit J Dermatol 86(1), 14–19. Hansen LA, Alexander N, Hogan ME, Sundberg JP, Dlugosz A, Threadgill DW, Magnuson T & Yuspa SH. (1997) Genetically null mice reveal a central role for Epidermal Growth Factor Receptor in the differentiation of the hair follicle and normal hair development. Am J Pathol 150(6), 1959–1975. Hashimoto K, Higashiyama S, Asada H, Hashimura E, Kobayashi T, Sudo K, Nakagawa T, Damm D, Yoshikawa K & Taniguchi N. (1994) Heparin-binding epidermal growth factor-like growth factor is an autocrine growth factor for human keratinocytes. J Biol Chem 269(31), 20060–6. He J, Zhang G, Almeida AD, Cayouette M, Simons BD & Harris WA. (2012) How variable clones build an invariant retina. Neuron 75(5), 786–798. He X, Semenov M, Tamai K & Zeng X. (2004) LDL receptor-related proteins 5 and 6 in Wnt/"-catenin signaling: arrows point the way. Development 131(8), 1663–77. Hochedlinger K, Yamada Y, Beard C & Jaenisch R. (2005) Ectopic expression of Oct-4 blocks progenitor-cell differentiation and causes dysplasia in epithelial tissues. Cell 121(3), 465–77. Hoffmann M, Chang HH, Huang S, Ingber DE, Loeffler M & Galle J. (2008) Noise- driven stem cell and progenitor population dynamics. PLoS One 3(8), e2922. Honda H & Oshibe S. (1984) A computer simulation of cell stacking for even thickness in mammalian epidermis. J Theor Biol 111(4), 625–33. Honda H, Tanemura M & Imayama S. (1996) Spontaneous architectural organization of mammalian epidermis from random cell packing. J Invest Dermatol 106(2), 312– 5. Hsu Y-C, Pasolli HA & Fuchs E. (2011) Dynamics between stem cells, niche, and progeny in the hair follicle. Cell 144(1), 92–105. Ito M, Liu Y, Yang Z, Nguyen J, Liang F, Morris RJ & Cotsarelis G. (2005) Stem cells in the hair follicle bulge contribute to wound repair but not to homeostasis of the epidermis. Nat Med 11(12), 1351–4. Jaks V, Barker N, Kasper M, van Es JH, Snippert HJ, Clevers H & Toftgård R. (2008) Lgr5 marks cycling, yet long-lived, hair follicle stem cells. Nat Genet 40(11), 1291– 1299. 154 Jensen KB, Driskell RR & Watt FM. (2010) Assaying proliferation and differentiation capacity of stem cells using disaggregated adult mouse epidermis. Nat Protoc 5(5), 898–911. Jensen KB & Watt FM. (2006) Single-cell expression profiling of human epidermal stem and transit-amplifying cells: Lrig1 is a regulator of stem cell quiescence. Proc Natl Acad Sci U S A 103(32), 11958–63. Jensen UB, Lowell S & Watt FM. (1999) The spatial relationship between stem cells and their progeny in the basal layer of human epidermis: a new view based on whole-mount labelling and lineage analysis. Development 126(11), 2409–18. Johnson GL & Vaillancourt RR. (1994) Sequential protein kinase reactions controlling cell growth and differentiation. Curr Opin Cell Biol 6(2), 230–238. Jonason AS, Kunala S, Price GJ, Restifo RJ, Spinelli HM, Persing JA, Leffell DJ, Tarone RE & Brash DE. (1996) Frequent clones of p53-mutated keratinocytes in normal human skin. Proc Natl Acad Sci U S A 93(24), 14025–9. Jones PH, Harper S & Watt FM. (1995) Stem cell patterning and fate in human epidermis. Cell 80(1), 83–93. Jones PH & Simons BD. (2008) Epidermal homeostasis: do committed progenitors work while stem cells sleep? Nat Rev Mol Cell Biol 9(1), 82–8. Jones PH, Simons BD & Watt FM. (2007) Sic transit gloria: farewell to the epidermal transit amplifying cell? Cell Stem Cell 1(4), 371–81. Jones PH & Watt FM. (1993) Separation of human epidermal stem cells from transit amplifying cells on the basis of differences in integrin function and expression. Cell 73(4), 713–24. Kalabis J, Oyama K, Okawa T, Nakagawa H, Michaylira CZ, Stairs DB, Figueiredo J, Mahmood U, Diehl JA, Herlyn M & Rustgi AK. (2008) A subpoulation of mouse esophageal basal cells has properties of stem cells with the capacity for self- renewal and lineage specification. J Clin Invest 118(12), 3860–3869. Kameda T, Nakata A, Mizutani T, Terada K, Iba H & Sugiyama T. (2003) Analysis of the cellular heterogeneity in the basal layer of mouse ear epidermis: an approach from partial decomposition in vitro and retroviral cell marking in vivo. Exp Cell Res 283(2), 167–183. Kamimura J, Lee D, Baden HP, Brissette J & Dotto G. (1997) Primary mouse keratinocyte cultures contain hair follicle progenitor cells with multiple differentiation potential. J Invest Dermatol 109(4), 534–40. 155 Kanda T, Sullivan KF & Wahl GM. (1998) Histone-GFP fusion protein enables sensitive analysis of chromosome dynamics in living mammalian cells. Curr Biol 8(7), 377–85. Karam SM & Leblond CP. (1993) Dynamics of epithelial cells in the corpus of the mouse stomach. I. Identification of proliferative cell types and pinpointing of the stem cell. Anat Rec 236(2), 259–279. Kaur P. (2006) Interfollicular epidermal stem cells: identification, challenges, potential. J Invest Dermatol 126(7), 1450–8. Kazanskaya O, Glinka A, Barrantes B, Stannek P, Niehrs C & Wu W. (2004) R- Spondin2 Is a secreted activator of Wnt/"-catenin signaling and is required for Xenopus myogenesis. Dev Cell 7, 525–534. Kemp R, Ireland H, Clayton E, Houghton C, Howard L & Winton DJ. (2004) Elimination of background recombination: somatic induction of Cre by combined transcriptional regulation and hormone binding affinity. Nucleic Acids Res 32(11), e92. Van Keymeulen A, Rocha AS, Ousset M, Beck B, Bouvencourt G, Rock J, Sharma N, Dekoninck S & Blanpain C. (2011) Distinct stem cells contribute to mammary gland development and maintenance. Nature 479(7372), 189–193. Kim K-A, Kakitani M, Zhao J, Oshima T, Tang T, Binnerts M, Liu Y, Boyle B, Park E, Emtage P, Funk WD & Tomizuka K. (2005) Mitogenic influence of human R- spondin1 on the intestinal epithelium. Science 309(5738), 1256–9. Kitano Y, Nagase N, Okada N & Okano M. (1983) Cinemicrographic study of cell proliferation pattern and interdivision times of human keratinocytes in primary culture. In M. Seiji & I. A. Bernstein, eds. Current Problems in Dermatology Vol 11: Normal and Abnormal Epidermal Differentiation. Tokyo: Univeristy of Tokyo Press, pp. 97–108. Klapper LN, Glathe S, Vaisman N, Hynes NE, Andrews GC, Sela M & Yarden Y. (1999) The ErbB-2/HER2 oncoprotein of human carcinomas may function solely as a shared coreceptor for multiple stroma-derived growth factors. Proc Natl Acad Sci U S A 96, 4995–5000. Klein AM, Brash DE, Jones PH & Simons BD. (2010) Stochastic fate of p53-mutant epidermal progenitor cells is tilted toward proliferation by UV B during preneoplasia. Proc Natl Acad Sci U S A 107(1), 270–5. 156 Klein AM, Nikolaidou-Neokosmidou V, Doupé DP, Jones PH & Simons BD. (2011) Patterning as a signature of human epidermal stem cell regulation. J R Soc Interface 8(65), 1815–24. Kociemba H. (2004) Motzkin Triangle, Sequence A026300. From: Online Encyclopedia of Integer Sequences. At: http://oeis.org/A026300 Kolodka TM, Garlick JA & Taichman LB. (1998) Evidence for keratinocyte stem cells in vitro: long term engraftment and persistence of transgene expression from retrovirus-transduced keratinocytes. Proc Natl Acad Sci U S A 95(8), 4356–61. Kraitchik M. (1942) The Gambler’s Ruin. In Mathematical Recreations. New York: W. W. Norton, p. 140. Kretzschmar K & Watt FM. (2012) Lineage tracing. Cell 148(1-2), 33–45. Lajtha LG. (1979) Stem cell concepts. Differentiation 14, 23–34. Langton AK, Herrick SE & Headon DJ. (2008) An extended epidermal response heals cutaneous wounds in the absence of a hair follicle stem cell contribution. J Invest Dermatol 128(5), 1311–8. Leblond CP & Walker BE. (1956) Renewal of cell populations. Physiol Rev 36(2), 255– 76. Legg J, Jensen UB, Broad S, Leigh I & Watt FM. (2003) Role of melanoma chondroitin sulphate proteoglycan in patterning stem cells in human interfollicular epidermis. Development 130(24), 6049–63. Lemmon MA & Schlessinger J. (2010) Cell signaling by receptor tyrosine kinases. Cell 141(7), 1117–34. Lengyel T. (2009) Gambler’s ruin and winning a series by m games. Ann Inst Stat Math 63(1), 181–195. Levkowitz G, Waterman H, Ettenberg SA, Katz M, Tsygankov AY, Alroy I, Lavi S, Iwai K, Reiss Y, Ciechanover A, Lipkowitz S & Yarden Y. (1999) Ubiquitin ligase activity and tyrosine phosphorylation underlie suppression of growth factor signaling by c-Cbl/Sli-1. Mol Cell 4, 1029–1040. Levy V, Lindon C, Harfe BD & Morgan BA. (2005) Distinct stem cell populations regenerate the follicle and interfollicular epidermis. Dev Cell 9(6), 855–61. Levy V, Lindon C, Zheng Y, Harfe BD & Morgan BA. (2007) Epidermal stem cells arise from the hair follicle after wounding. FASEB J 21(7), 1358–66. 157 Ljunggren C. (1898) Von der Fähigkeit des Hautepithels, ausserhalb des Organismus sein Leben zu behalten, mit Berücksichtigung der Transplantation. Dtsch Zeitschrift für Chir 5-6, 608–615. Lowell S & Watt FM. (2001) Delta regulates keratinocyte spreading and motility independently of differentiation. Mech Dev 107(1-2), 133–40. Lu CP, Polak L, Rocha AS, Pasolli HA, Chen S, Sharma N & Blanpain C. (2012) Identification of stem cell populations in sweat glands and ducts reveals roles in homeostasis and wound repair. Cell 150, 136–150. Mackenzie IC. (1970) Relationship between mitosis and the ordered structure of the stratum corneum in mouse epidermis. Nature 226(May 16), 653–655. Mackenzie IC. (1997) Retroviral transduction of murine epidermal stem cells demmonstrates clonal units of epidermal structure. J Invest Dermatol 109(3), 377– 383. Mackenzie IC & Bickenbach JR. (1985) Label-retaining keratinocytes and Langerhans cells in mouse epithelia. Cell Tissue Res 242(3), 551–6. Mannik J, Alzayady K & Ghazizadeh S. (2010) Regeneration of multilineage skin epithelia by differentiated keratinocytes. J Invest Dermatol 130(2), 388–97. Marikovsky M, Breuing K, Liu PY, Eriksson E, Higashiyama S, Farber P, Abraham J & Klagsbrun M. (1993) Appearance of heparin-binding EGF-like growth factor in wound fluid as a response to injury. Proc Natl Acad Sci U S A 90(9), 3889–93. Marques-Pereira JP & Leblond CP. (1965) Mitosis and differentiation in the stratified squamous epithelium of the rat oesophagus. Am J Anat 117, 73–90. Mascré G, Dekoninck S, Drogat B, Youssef KK, Brohée S, Sotiropoulou PA, Simons BD & Blanpain C. (2012) Distinct contribution of stem and progenitor cells to epidermal maintenance. Nature 489(7415), 257–262. Mavilio F, Pellegrini G, Ferrari S, Di Nunzio F, Di Iorio E, Recchia A, Maruggi G, Ferrari G, Provasi E, Bonini C, Capurro S, Conti A, Magnoni C, Giannetti A & De Luca M. (2006) Correction of junctional epidermolysis bullosa by transplantation of genetically modified epidermal stem cells. Nat Med 12(12), 1397–402. Mayo LD & Donner DB. (2001) A phosphatidylinositol 3-kinase/Akt pathway promotes translocation of Mdm2 from the cytoplasm to the nucleus. Proc Natl Acad Sci U S A 98(20), 11598–11603. 158 Medawar PB. (1941) Sheets of pure epidermal epithelium from human skin. Nature 148(3765), 783. Menton DN. (1976a) A liquid film model of tetrakiadecahedral packing to account for the establishment of epidermal cell columns. J Invest Dermatol 66(5), 283–291. Menton DN. (1976b) A minimum-surface mechanism to account for the organization of cells into columns in the mammalian epidermis. Am J Anat 145(1), 1–22. Miaczynska M, Pelkmans L & Zerial M. (2004) Not just a sink: endosomes in control of signal transduction. Curr Opin Cell Biol 16(4), 400–6. Miettinen PJ, Berger JE, Meneses J, Phung Y, Penderson RA, Werb Z & Derynck R. (1995) Epithelial immaturity and multiorgan failure in mice lacking epidermal growth factor receptor. Nature 376, 337–341. Morris R & Argyris TS. (1983) Epidermal cell cycle and transit times during hyperplastic growth induced by abrasion or treatment with 12-O- Tetradecanoylphorbol-13-acetate. Cancer Res 43, 4935–4942. Motzkin T. (1948) Relations between hypersurface cross ratios and a combinatorial formula for partitions of a polygon, for permanent preponderance, and for non- associative products. Bull Am Math Soc 54(4), 352–360. Mulder KW, Wang X, Escriu C, Ito Y, Schwarz RF, Gillis J, Sirokmány G, Donati G, Uribe-Lewis S, Pavlidis P, Murrell A, Markowetz F & Watt FM. (2012) Diverse epigenetic strategies interact to control epidermal differentiation. Nat Cell Biol 14(7), 753–763. Nanba D, Toki F, Matsushita N, Matsushita S, Higashiyama S & Barrandon Y. (2013) Actin filament dynamics impacts keratinocyte stem cell maintenance. EMBO Mol Med 5(4), 640–53. Navasaria H, Sexton C, Bouvard V & Leigh IM. (1994) Human epidermal keratinocytes - Growth of keratinocytes with a 3T3 feeder layer: basic techniques. In I. M. Leigh & F. M. Watt, eds. Keratinocyte Methods. Cambridge: Cambridge University Press, pp. 5–12. Nave BT, Ouwens DM, Withers DJ, Alessi DR & Shepherd PR. (1999) Mammalian target of rapamycin is a direct target for protein kinase B!: identification of a convergence point for opposing effects of insulin and amino-acid deficiency on protein translation. Biochem J 344, 427–431. Nguyen L V, Vanner R, Dirks P & Eaves CJ. (2012) Cancer stem cells: an evolving concept. Nat Rev Cancer 12, 133–143. 159 Normanno N, De Luca A, Bianco C, Strizzi L, Mancino M, Maiello MR, Carotenuto A, De Feo G, Caponigro F & Salomon DS. (2006) Epidermal growth factor receptor (EGFR) signaling in cancer. Gene 366, 2–16. O’Connor NE, Mulliken JB, Banks-Schlegel S, Kehinde O & Green H. (1981) Grafting of burns with cultured epithelium from autologous epidermal cells. Lancet 317(8211), 75–78. Okines A, Cunningham D & Chau I. (2011) Targeting the human EGFR family in esophagogastric cancer. Nat Rev Clin Oncol 8(8), 492–503. Ono I, Gunji H, Suda K, Iwatsuki K & Kaneko F. (1994) Evaluation of cytokines in donor site wound fluids. Scand J Plast Reconstr Hand Surg 28(4), 269–273. Ono I, Gunji H, Zhang J-Z, Maruyama K & Kaneko F. (1995) A study of cytokines in burn blister fluid related to wound healing. Burns 21(5), 352–355. Ousset M, Keymeulen A Van, Bouvencourt G, Sharma N, Achouri Y, Simons BD & Blanpain C. (2012) Multipotent and unipotent progenitors contribute to prostate postnatal development. Nat Cell Biol 14(11), 1131–1138. Page ME, Lombard P, Ng F, Göttgens B & Jensen KB. (2013) The epidermis comprises autonomous compartments maintained by distinct stem cell populations. Cell Stem Cell 13(4), 471–482. Parma P, Radi O, Vidal V, Chaboissier MC, Dellambra E, Valentini S, Guerra L, Schedl A & Camerino G. (2006) R-spondin1 is essential in sex determination, skin differentiation and malignancy. Nat Genet 38(11), 1304–9. Patel GK, Wilson CH, Harding KG, Finlay AY & Bowden PE. (2006) Numerous keratinocyte subtypes involved in wound re-epithelialization. J Invest Dermatol 126(2), 497–502. Pece S, Tosoni D, Confalonieri S, Mazzarol G, Vecchi M, Ronzoni S, Bernard L, Viale G, Pelicci PG & Di Fiore PP. (2010) Biological and molecular heterogeneity of breast cancers correlates with their cancer stem cell content. Cell 140(1), 62–73. Pende M, Um SH, Mieulet V, Goss VL, Mestan J, Mueller M, Fumagalli S, Kozma SC & Thomas G. (2004) S6K1-/-/S6K2-/- mice exhibit perinatal lethality and rapamycin-sensitive 5’-terminal oligopyrimidine mRNA translation and reveal a mitogen-activated protein kinase-dependent S6 kinase pathway. Mol Cell Biol 24(8), 3112–3124. 160 Peus D, Hamacher L & Pittelkow MR. (1997) EGF-receptor tyrosine kinase inhibition induces keratinocyte growth arrest and terminal differentiation. J Invest Dermatol 109, 751–756. Pinkas-Kramarski R, Soussan L, Waterman H, Levkowitz G, Alroy I, Klapper L, Lavi S, Seger R, Ratzkin BJ, Sela M & Yarden Y. (1996) Diversification of Neu differentiation factor and epidermal growth factor signaling by combinatorial receptor interactions. EMBO J 15(10), 2452–2467. Potten CS. (1974) The Epidermal Proliferation Unit: The possible role of the central basal unit. Cell Tissue Kinet 7, 77–88. Potten CS & Allen TD. (1975) The fine structure and cell kinetics of mouse epidermis after wounding. J Cell Sci 17, 413–447. Potten CS, Barthel D, Li YQ, Ohlrich R, Matthé B & Loeffler M. (2000) Proliferation in murine epidermis after minor mechanical stimulation. Part 1. Sustained increase in keratinocyte production and migration. Cell Prolif 33(4), 231–46. Potten CS & Loeffler M. (1990) Stem cells: attributes, cycles, spirals, pitfalls and uncertainties. Lessons for and from the crypt. Development 110(4), 1001–20. Potten CS, Wichmann HE, Loeffler M, Dobek K & Major D. (1982) Evidence for discrete cell kinetic subpopulations in mouse epidermis based on mathematical analysis. Cell Tissue Kinet 15, 305–329. Rama P, Matuska S, Paganoni G, Spinelli A, De Luca M & Pellegrini G. (2010) Limbal stem-cell therapy and long-term corneal regeneration. New Engl J Med 363(2), 147–55. Rawlins EL, Okubo T, Xue Y, Brass DM, Auten RL, Hasegawa H, Wang F & Hogan BLM. (2009) The role of Scgb1a1+ Clara cells in the long-term maintenance and repair of lung airway, but not alveolar, epithelium. Cell Stem Cell 4(6), 525–534. Rheinwald JG & Green H. (1977) Epidermal growth factor and the multiplication of cultured human epidermal keratinocytes. Nature 265, 421–424. Rheinwald JG & Green H. (1975) Serial cultivation of strains of human epidermal keratinocytes: the formation keratinizing colonies from single cells. Cell 6(3), 331– 343. Rittié L, Sachs DL, Orringer JS, Voorhees JJ & Fisher GJ. (2013) Eccrine sweat glands are major contributors to reepithelialization of human wounds. Am J Pathol 182(1), 163–71. 161 Ro S & Rannala B. (2004) A stop-EGFP transgenic mouse to detect clonal cell lineages generated by mutation. EMBO Rep 5(9), 914–20. Ro S & Rannala B. (2005) Evidence from the stop-EGFP mouse supports a niche- sharing model of epidermal proliferative units. Exp Dermatol 14(11), 838–43. Rochat A, Kobayashi K & Barrandon Y. (1994) Location of stem cells of human hair follicles by clonal analysis. Cell 76(6), 1063–73. Rock JR, Onaitis MW, Rawlins EL, Lu Y, Clark CP, Xue Y, Randell SH & Hogan BLM. (2009) Basal cells as stem cells of the mouse trachea and human airway epithelium. Proc Natl Acad Sci U S A 106(31), 12771–12775. Rodriguez-Viciana P, Warne PH, Dhand R, Vanhaesebroeck B, Gout I, Fry MJ, Waterfield MD & Downward J. (1994) Phosphatidylinositol-3-OH kinase as a direct target of Ras. Nature 370, 527–532. Romashkova JA & Makarov SS. (1999) NF-#B is a target of AKT in anti-apoptotic PDGF signalling. Nature 401, 86–90. Rompolas P, Deschene ER, Zito G, Gonzalez DG, Saotome I, Haberman AM & Greco V. (2012) Live imaging of stem cell and progeny behaviour in physiological hair- follicle regeneration. Nature 487(7408), 496–9. Rompolas P, Mesa KR & Greco V. (2013) Spatial organization within a niche as a determinant of stem-cell fate. Nature 502(7472), 513–8. Ronfard V, Rives JM, Neveux Y, Carsin H & Barrandon Y. (2000) Long-term regeneration of human epidermis on third-degree burns transplanted with autologous cultured epithelium grown on a fibrin matrix. Transplantation 70(11), 1588–1598. Roshan A & Jones PH. (2012a) Act your age: Tuning cell behavior to tissue requirements in interfollicular epidermis. Semin Cell Dev Biol 23(8), 884–889. Roshan A & Jones PH. (2012b) Chronic low dose UV exposure and p53 mutation: tilting the odds in early epidermal preneoplasia. Int J Radiat Biol 88(10), 682–687. Roth S, Franken P, Sacchetti A, Kremer A, Anderson K, Sansom O & Fodde R. (2012) Paneth cells in intestinal homeostasis and tissue injury. PLoS One 7(6), e38965. Le Roy H, Zuliani T, Wolowczuk I, Faivre N, Jouy N, Masselot B, Kerkaert J-P, Formstecher P & Polakowska R. (2010) Asymmetric distribution of epidermal growth factor receptor directs the fate of normal and cancer keratinocytes in vitro. Stem Cells Dev 19(2), 209–20. 162 Ruse M, Broome A-M & Eckert RL. (2003) S100A7 (psoriasin) interacts with epidermal fatty acid binding protein and localizes in focal adhesion-like structures in cultured keratinocytes. J Invest Dermatol 121(1), 132–41. Ruse M, Lambert a, Robinson N, Ryan D, Shon KJ & Eckert RL. (2001) S100A7, S100A10, and S100A11 are transglutaminase substrates. Biochemistry 40(10), 3167– 73. Salic A & Mitchison TJ. (2008) A chemical method for fast and sensitive detection of DNA synthesis in vivo. Proc Natl Acad Sci U S A 105(7), 2415–20. Sangiorgi E & Capecchi MR. (2008) Bmi1 is expressed in vivo in intestinal stem cells. Nat Genet 40(7), 915–20. Sato F & Sato K. (2000) cAMP-dependent Cl(-) channel protein (CFTR) and its mRNA are expressed in the secretory portion of human eccrine sweat gland. J Histochem Cytochem 48(3), 345–353. Sato T, van Es JH, Snippert HJ, Stange DE, Vries RG, van den Born M, Barker N, Shroyer NF, van de Wetering M & Clevers H. (2011) Paneth cells constitute the niche for Lgr5 stem cells in intestinal crypts. Nature 469(7330), 415–8. Sato T, Vries RG, Snippert HJ, van de Wetering M, Barker N, Stange DE, van Es JH, Abo A, Kujala P, Peters PJ & Clevers H. (2009) Single Lgr5 stem cells build crypt- villus structures in vitro without a mesenchymal niche. Nature 459(7244), 262–5. Savage Jr CR & Cohen S. (1973) Proliferation of corneal epithelium induced by epidermal growth factor. Exp Eye Res 15(3), 361–366. Schmidt GH, Blount MA & Ponder BA. (1987) Immunochemical demonstration of the clonal organization of chimaeric mouse epidermis. Development 100(3), 535–41. Schneider MR & Wolf E. (2009) The epidermal growth factor receptor ligands at a glance. J Cell Physiol 218(3), 460–466. Schroeder T. (2013) Heterogeneity of sister cell fates. Nat Rev Mol Cell Biol 14(6), 327– 327. Schroeder T. (2011) Long-term single-cell imaging of mammalian stem cells. Nat Methods 8(4s), S30–S35. Schultz GS, White M, Mitchell R, Brown G, Twardzik DR & Todaro GJ. (1987) Epithelial wound healing enhanced by transforming growth factor-! and vaccinia growth factor. Science 235(4786), 350–352. 163 Schulze WX, Deng L & Mann M. (2005) Phosphotyrosine interactome of the ErbB- receptor kinase family. Mol Syst Biol 1, 42–54. Shimojo H, Ohtsuka T & Kageyama R. (2008) Oscillations in notch signaling regulate maintenance of neural progenitors. Neuron 58(1), 52–64. Shirakata Y, Kimura R, Nanba D, Iwamoto R, Tokumaru S, Morimoto C, Yokota K, Nakamura M, Sayama K, Mekada E, Higashiyama S & Hashimoto K. (2005) Heparin-binding EGF-like growth factor accelerates keratinocyte migration and skin wound healing. J Cell Sci 118(11), 2363–70. Sibilia M & Wagner EF. (1995) Strain-dependent epithelial defects in mice lacking the EGF receptor. Science 269(5221), 234–238. Sigismund S, Argenzio E, Tosoni D, Cavallaro E, Polo S & Di Fiore PP. (2008) Clathrin- mediated internalization is essential for sustained EGFR signaling but dispensable for degradation. Dev Cell 15(2), 209–19. Snippert HJ, Haegebarth A, Kasper M, Jaks V, van Es JH, Barker N, van De Wetering M, van Den Born M, Begthel H, Vries RG, Stange DE, Toftgard R & Clevers H. (2010) Lrg6 marks stem cells in the hair follicle that generate all cell lineages of the skin. Science 327(5971), 1385–1389. Solanas G & Benitah SA. (2013) Regenerating the skin: a task for the heterogeneous stem cell pool and surrounding niche. Nat Rev Mol Cell Biol 14(11), 737–748. Sorkin A & Goh L. (2009) Endocytosis and intracellular trafficking of ErbBs. Exp Cell Res 315(4), 683–696. Sotiropoulou PA & Blanpain C. (2012) Development and homeostasis of the skin epidermis. Cold Spring Harb Perspect Biol 4(7), a008383. Sousa LP, Lax I, Shen H, Ferguson SM, De Camilli P & Schlessinger J. (2012) Suppression of EGFR endocytosis by dynamin depletion reveals that EGFR signaling occurs primarily at the plasma membrane. Proc Natl Acad Sci U S A 109(12), 4419–24. Srinivas S, Watanabe T, Lin CS, William CM, Tanabe Y, Jessell TM & Costantini F. (2001) Cre reporter strains produced by targeted insertion of EYFP and ECFP into the ROSA26 locus. BMC Dev Biol 1, 4. Stange DE, Koo B-K, Huch M, Sibbel G, Basak O, Lyubimova A, Kujala P, Bartfeld S, Koster J, Geahlen JH, Peters PJ, van Es JH, van de Wetering M, Mills JC & Clevers H. (2013) Differentiated Troy(+) chief cells act as reserve stem cells to generate all lineages of the stomach epithelium. Cell 155(2), 357–68. 164 Storey WF & Leblond CP. (1951) Measurement of the rate of proliferation of epidermis and associated structures. Ann N Y Acad Sci 53(3), 537–545. Tata PR, Mou H, Pardo-Saganta A, Zhao R, Prabhu M, Law BM, Vinarsky V, Cho JL, Breton S, Sahay A, Medoff BD & Rajagopal J. (2013) Dedifferentiation of committed epithelial cells into stem cells in vivo. Nature 503, 218–223. Testa JR & Tsichlis PN. (2005) AKT signaling in normal and malignant cells. Oncogene 24(50), 7391–7393. Tian H, Biehs B, Warming S, Leong KG, Rangell L, Klein OD & de Sauvage FJ. (2011) A reserve stem cell population in small intestine renders Lgr5-positive cells dispensable. Nature 478(7368), 255–259. Todaro GJ & Green H. (1963) Quantitative studies of the growth of the mouse embryo cells in culture and their development into established lines. J Cell Biol (6), 299– 313. Tumbar T, Guasch G, Greco V, Blanpain C, Lowry WE, Rendl M & Fuchs E. (2004) Defining the epithelial stem cell niche in skin. Science 303(5656), 359–63. Tzahar E, Waterman H, Chen X, Levkowitz G, Karunagaran D, Ratzkin BJ & Yarden Y. (1996) A hierarchical network of interreceptor interactions determines signal transduction by Neu differentiation factor/neuregulin and epidermal growth factor. Mol Cell Biol 16(10), 5276–5287. Usui ML, Underwood RA, Mansbridge JN, Muffley LA, Carter WG & Olerud JE. (2005) Morphological evidence for the role of suprabasal keratinocytes in wound reepithelialisation. Wound Repair Regen 13(5), 468–479. Vermeulen L, Morrissey E, Heijden M van der, Nicholson AM, Sottoriva A, Buczaki S, Kemp R, Tavare S & Winton DJ. (2013) Defining stem cell dynamics in models of intestinal tumor initiation. Science 342(6161), 995–998. Vivanco I & Sawyers CL. (2002) The Phosphatidylinositol 3-kinase-Akt pathway in human cancer. Nat Rev Cancer 2, 489–501. Wakita H & Takigawa M. (1999) Activation of Epidermal Growth Factor Receptor promotes late terminal differentiation of cell-matrix interaction-disrupted keratinocytes. J Biol Chem 274(52), 37285–37291. Wallasch C, Wei FU, Niederfellnerl G, Jallall B, Issing W & Ullrich A. (1995) Heregulin-dependent regulation of HER2/neu oncogenic signaling by heterodimerization with. EMBO J 14(17), 4267–4275. 165 Wan H, Stone MG, Simpson C, Reynolds LE, Marshall JF, Hart IR, Hodivala-Dilke KM & Eady RAJ. (2003) Desmosomal proteins, including desmoglein 3, serve as novel negative markers for epidermal stem cell-containing population of keratinocytes. J Cell Sci 116(20), 4239–4248. Wan H, Yuan M, Simpson C, Allen K, Gavins FNE, Ikram MS, Basu S, Baksh N, O’Toole EA & Hart IR. (2007) Stem/progenitor cell-like properties of desmoglein 3dim cells in primary and immortalized keratinocyte lines. Stem Cells 25(5), 1286– 97. Wang X, Reid Sutton V, Omar Peraza-Llanes J, Yu Z, Rosetta R, Kou Y-C, Eble TN, Patel A, Thaller C, Fang P & Van den Veyver IB. (2007) Mutations in X-linked PORCN, a putative regulator of Wnt signaling, cause focal dermal hypoplasia. Nat Genet 39(7), 836–8. Wang Z, Wang Y, Farhangfar F, Zimmer M & Zhang Y. (2012) Enhanced keratinocyte proliferation and migration in co-culture with fibroblasts. PLoS One 7(7), e40951. Watt FM & Green H. (1981) Involucrin synthesis is correlated with cell size in human epidermal cultures. J Cell Biol 90(3), 738–742. Watt FM & Hogan BLM. (2000) Out of Eden: Stem cells and their niches. Science 287(5457), 1427–1430. Weisstein EW. (2012a) Gambler’s Ruin. From: MathWorld-A Wolfram Web Resource. At: http://mathworld.wolfram.com/GamblersRuin.html Weisstein EW. (2012b) Motzkin Number. From: MathWorld-A Wolfram Web Resource. At: http://mathworld.wolfram.com/MotzkinNumber.html Yarden Y & Sliwkowski MX. (2001) Untangling the ErbB signalling network. Nat Rev Mol Cell Biol 2, 127–137. Zhang W, Remenyik E, Zelterman D, Brash DE & Wikonkal NM. (2001) Escaping the stem cell compartment!: Sustained UVB exposure allows p53-mutant keratinocytes to colonize adjacent epidermal proliferating units without incurring additional mutations. Proc Natl Acad Sci U S A 98(24), 13948–13953. Zhu AJ & Watt FM. (1999) "-Catenin signalling modulates proliferative potential of human epidermal keratinocytes independently of intercellular adhesion. Development 126(10), 2285–98. A-1 APPENDIX 1: PUBLISHED ARTICLE Doupé DP, Alcolea MP, Roshan A, Zhang G, Klein AM, Simons BD, & Jones PH. (2012) A single progenitor population switches behaviour to maintain and repair esophageal epithelium. Science 337(6098), 1091-1093. 1 A-2 DOI: 10.1126/science.1218835 , 1091 (2012);337 Science et al.David P. Doupé Repair Esophageal Epithelium A Single Progenitor Population Switches Behavior to Maintain and This copy is for your personal, non-commercial use only. clicking here.colleagues, clients, or customers by , you can order high-quality copies for yourIf you wish to distribute this article to others here.following the guidelines can be obtained byPermission to republish or repurpose articles or portions of articles ): September 3, 2012 www.sciencemag.org (this information is current as of The following resources related to this article are available online at http://www.sciencemag.org/content/337/6098/1091.full.html version of this article at: including high-resolution figures, can be found in the onlineUpdated information and services, http://www.sciencemag.org/content/suppl/2012/07/19/science.1218835.DC1.html can be found at: Supporting Online Material http://www.sciencemag.org/content/337/6098/1091.full.html#related found at: can berelated to this article A list of selected additional articles on the Science Web sites http://www.sciencemag.org/content/337/6098/1091.full.html#ref-list-1 , 7 of which can be accessed free:cites 30 articlesThis article http://www.sciencemag.org/content/337/6098/1091.full.html#related-urls 1 articles hosted by HighWire Press; see:cited by This article has been http://www.sciencemag.org/cgi/collection/development Development subject collections:This article appears in the following registered trademark of AAAS. is aScience2012 by the American Association for the Advancement of Science; all rights reserved. The title CopyrightAmerican Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005. (print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by theScience o n Se pt em be r 3 , 2 01 2 ww w. sc ien ce m ag .o rg Do wn loa de d fro m and the Kavli Institute. S.J.G., J.R.P., and L.M. designed the study. S.J.G., J.R.P., A.G.M., and L.M. conducted the research. S.J.G. and J.R.P. performed the biological and biophysical experiments. S.J.G., A.G.M., and L.M. handled biophysical theory. S.J.G., J.R.P., A.G.M., and L.M. contributed analytical tools and reagents and analyzed data. S.J.G., J.R.P., and L.M. wrote the paper. Harvard University has filed a patent application relating to a tunable, twistless overwinding spring based on the results of this study. Supplementary Materials www.sciencemag.org/cgi/content/full/337/6098/1087/DC1 Materials and Methods Supplementary Text Figs. S1 to S5 References (26–29) Movies S1 to S11 13 April 2012; accepted 19 June 2012 10.1126/science.1223304 A Single Progenitor Population Switches Behavior to Maintain and Repair Esophageal Epithelium David P. Doupé,1,4* Maria P. Alcolea,1* Amit Roshan,1 Gen Zhang,2 Allon M. Klein,2,3 Benjamin D. Simons,2,4 Philip H. Jones1† Diseases of the esophageal epithelium (EE), such as reflux esophagitis and cancer, are rising in incidence. Despite this, the cellular behaviors underlying EE homeostasis and repair remain controversial. Here, we show that in mice, EE is maintained by a single population of cells that divide stochastically to generate proliferating and differentiating daughters with equal probability. In response to challenge with all-trans retinoic acid (atRA), the balance of daughter cell fate is unaltered, but the rate of cell division increases. However, after wounding, cells reversibly switch to producing an excess of proliferating daughters until the wound has closed. Such fate-switching enables a single progenitor population to both maintain and repair tissue without the need for a “reserve” slow-cycling stem cell pool. Murine esophageal epithelium (EE) con-sists of layers of keratinocytes. This tis-sue lacks structures such as crypts or glands that form stem cell niches in other epithe- lia (Fig. 1, A and B) (1–5). Proliferation is con- fined to cells in the basal layer (6). On commitment to terminal differentiation, basal cells exit the cell cycle and subsequently migrate to the tissue sur- face from which they are shed. Early studies sug- gested that all proliferating cells were functionally equivalent, but recent reports propose that a dis- crete population of slow-cycling stem cells is re- sponsible for bothmaintenance andwound healing (7–11). This controversy and the importance of EE in disease motivated us to resolve the prolif- erative cell behavior in homeostatic EE and in tissue challenged by systemic treatment with the vitaminAmetabolite all-trans retinoic acid (atRA) or acute local wounding (12, 13). To investigate cell division rates in EE, we used a transgenic label-retaining cell (LRC) assay (Fig. 1C) (1, 14, 15). Doxycycline (DOX) induc- tion of the fusion protein Histone-2B enhanced green fluorescent protein (HGFP) expression in Rosa26M2rtTA/TetO-HGFP mice resulted in nu- clear fluorescent labeling throughout the EE (Fig. 1D and fig. S1A). When DOX is withdrawn, HGFP is diluted by cell division, leaving 0.4% basal layer cells (561 out of 140,000) retaining label after a 4-week chase (Fig. 1E and fig. S1B). Three-dimensional imaging showed that these LRCs had smaller nuclei than the surrounding keratinocytes and did not stain for the basal ke- ratinocyte marker Keratin14 (0 out of 561 LRCs) (fig. S1, C and D). The stem cell markers CD34 and Lgr5were also undetectable in LRCs or other cells (figs. S2 and S3) (2, 4, 10, 16). However, 99.9% (2457 out of 2459) of LRCs were positive for the pan leukocyte marker CD45 (Fig. 1E, in- set), comprising a mixture of Langerhans cells and lymphocytes (fig. S1, E and F). These find- ings lead to the unexpected conclusion that, unlike tissues such as the epidermis, there are no slow- cycling or quiescent epithelial stem cells in EE (1, 17). Indeed, HGFP dilution in basal cells was strikingly homogeneous, suggesting that all cells divide at a similar rate of about twice per week (fig. S1G). Although epithelial cells have the same rate of division, they may still differ in their ability to generate cycling and differentiated progeny. We therefore used inducible cre-lox–based genetic marking to investigate whether the proliferating cell population is heterogeneous and to quantify cell behavior (18, 19). The fate of single-cell- derived clones was tracked in cohorts of adult AhcreERT R26flEYFP/wt mice at multiple time points over a year after induction, during which period EE was homeostatic (Fig. 2A and fig. S4). Crucially, analysis of the composition of clones at 1 year showed that they were repre- sentative of unlabeled cells (fig. S5). Over the time course, clone number decreased through differentiation, whereas the size of the remaining clones progressively increased (Fig. 2, B and C). Although variation in labeling efficiency limits the accuracywith which the proportion of labeled cells can be estimated, within statistical error, 1Medical Research Council (MRC) Cancer Cell Unit, Hutchison- MRC Research Centre, Cambridge CB2 0XZ, UK. 2Cavendish Laboratory, Department of Physics, J. J. Thomson Avenue, University of Cambridge, Cambridge CB3 0HE, UK. 3Depart- ment of Systems Biology, HarvardMedical School, 200 Longwood Avenue, Boston, MA 02115, USA. 4The Wellcome Trust/Cancer Research UK Gurdon Institute, University of Cambridge, Tennis Court Road, Cambridge CB2 1QN, UK. *These authors contributed equally to this work. †To whom correspondence should be addressed. E-mail: phj20@cam.ac.uk Fig. 1. Esophageal epithelium contains no slow- cycling epithelial cells. (A) Microendoscopy show- ing esophageal lumen; scale bar, ~500 mm. (B) Section of epithelium, basal layer (b), suprabasal layers (sb), and lumen (l); scale bar, 10 mm. (C) Protocol: Adult Rosa26M2rtTA/TetO-HGFPmice treated with doxycycline (DOX) express HGFP (green). Af- ter DOX withdrawal, HGFP is diluted upon cell division, except in slow-cycling cells. (D and E) Rendered confocal z stacks, showing HGFP (green) at time 0 (D) and after 4-week chase (E). Scale bar, 10 mm. Dashed line indicates basement membrane. Inset shows CD45 (red) staining in HGFP-retaining cell at 4 weeks. 4 ,´6-diamidino-2-phenylindole (DAPI), blue; scale bar, 5 mm. www.sciencemag.org SCIENCE VOL 337 31 AUGUST 2012 1091 REPORTS o n Se pt em be r 3 , 2 01 2 ww w. sc ien ce m ag .o rg Do wn loa de d fro m this proportion remains constant, which is con- sistent with the labeled population being in ho- meostasis (Fig. 2D). Notably, the average size of persisting clones increased linearly with time, and their size dis- tribution acquired long-term scaling behavior, a hallmark of a single functionally equivalent population of cells dividing at the same rate (Fig. 2C and fig. S6, A, B, and E) (18–21). Studies of interfollicular epidermis (IFE) revealed that this pattern of clonal evolution was consistent with progenitors dividing stochastically to generate dif- ferentiated and cycling daughters with equal prob- ability (18, 19). By implementing a Bayesian inference analysis, we showed that the entire data set conforms to the IFE paradigm (Fig. 2E; fig. S6, C to E; and supplementary theory). We con- clude that esophageal progenitors (EP) are func- tionally equivalent. The observation of similar progenitor be- havior in EE and epidermis, derived from en- doderm and ectoderm, respectively, argues that squamous epithelia share a common mecha- nism of homeostasis irrespective of their devel- opmental origin. However, EP behavior differs from that of crypt stem cells in the endoderm-derived in- testinal epithelium, where stochastic fate is a result of competition for limited niche space (16, 22). Unlike progenitors in other tissues, such as the epidermis, EP are not supported by a discrete slow-cycling stem cell population (1). This raises the intriguing question of how the tissue responds to stress or injury. To investigate this issue, we subjected EE to a tissue-wide challenge in the form of atRA treatment and to acute local exci- sional wounding. To determine the effects of atRA, we selected a dose that induced a “hyperproliferative” re- sponse (fig. S7A) and then used quantitative lineage tracing to define the changes in cell be- havior (23, 24). Mice were treated for 9 days, clonal labeling was induced, and treatment then continued for a further 21 days, when clone size was scored (Fig. 3A). Bayesian analysis revealed that the rates of EP proliferation and differen- tiated cell stratification had approximately dou- bled. There was a small but statistically significant decrease in the proportion of proliferative cells but, critically, no significant change in the pro- portions of symmetric and asymmetric divisions, which indicated that the treated tissue was ho- meostatic (Fig. 3, B to D). To evaluate this find- ing, we used a second experimental schedule in which clonal labeling was induced before atRA treatment. The values of parameters determined in the first experiment accurately predicted the number of basal cells per clone on completion of the second protocol (fig. S7, B to D). We con- clude that during atRA treatment, EP establish a new homeostatic state. To investigate the repair of EP after wound- ing, we developed microendoscopic biopsy of mouse esophagus (fig. S8F). Biopsy produced a typical epithelial wound response (25, 26). Cells immediately next to the defect formed a migrat- ing front (mf) in which there was minimal prolif- eration, surrounded by a proliferative zone (pz) in which cell division was dramatically increased (Fig. 4B and fig. S8D). We used three different protocols to analyze cell behavior (Fig. 4). First, we examined clonally labeled EP in AhcreERT R26flEYFP/wt mice induced 1 week before biopsy Fig. 2. Proliferating cell fate in esophageal epithelium. (A) Protocol: Clonal labeling was induced in AhcreERTR26flEYFP/wt mice and analyzed at intervals from 3 days to 1 year (triangles). Images are rendered confocal z stacks of the basal layer showing typical clones at times indicated. Enhanced yellow fluorescent protein (EYFP), yellow; DAPI, blue. Scale bars, 10 mm. (B to D) Clone quan- tification. (B and C) Clone density and average clone size (basal cells). Observed values (orange) with error bars (mean T SEM); green curves show predictions of model (E). (D) Average percentage of labeled basal cells at indicated time points (orange); error bars indicate mean T SEM. Green line and shading show average and SEM across all time points. (E) Cell fate in EE. Basal layer comprises 65% functionally equivalent EP (green, dividing at a rate of 1.9/week, consistent with the rate of dilution of HGFP) (fig. S1G) and 35% postmitotic cells (pink), which stratify (arrow) at a rate of 3.5/week. Ten percent of EP divisions generate two EP daughters, 10% two differentiated daughters, and 80% one of each fate. Values are optimal fits with 95% plausible intervals. Fig. 3. All-trans retinoic acid (atRA) treatment of EE. (A) Protocol (see text). (B and C) Size distribution of multicellular clones containing at least one basal cell in control [(B), 307 clones] and atRA-treated [(C), 300 clones] EE. Green bars indicate 95% plausible fit to models in Figs. 2E (control) and3D (atRA). (D) Optimal fit during atRA treatment; proliferation and differen- tiation rates (red) increase compared with control. 31 AUGUST 2012 VOL 337 SCIENCE www.sciencemag.org1092 REPORTS o n Se pt em be r 3 , 2 01 2 ww w. sc ien ce m ag .o rg Do wn loa de d fro m (Fig. 4, A to C). Twenty-four hours after wound- ing, fragmented clones of labeled cells were seen, aligned toward the wound and spanning the pz and mf (Fig. 4, A and B, and fig. S8D). By 10 days, clones were evident in and around the re- paired defect (Fig. 4C and fig. S8, A and E). These findings indicate that EP participate in tissue re- generation after wounding, a behavior recapitu- lated in explant culture, suggesting that active recruitment of immune cells is not essential for the switch inEP fate (fig. S9) (27, 28). To investigate the proportion of EP that participate in regeneration, we biopsied DOX-treated Rosa26M2rtTA/TetO- HGFP mice (Fig. 4, D to F). HGFP was substan- tially and evenly diluted within the pz at 2 and 5 days after biopsy comparedwith controls but was retained outside the mf (Fig. 4, D to F; fig. S8B; and fig. S10, A and B). We conclude that there is widespread mobilization of EP within the pz and that the recruited cells proliferate to a similar ex- tent. In a complementary experiment, animalswere injected with EdU (5-ethynyl-2′-deoxyuridine) 24 hours before culling, revealing extensive recruit- ment of cells into cycle in the pz at 2 days, which reverted to control levels at 5 days when the epi- thelial defect had closed (Fig. 4, G to I; fig. S8C; and fig. S10, C to F). This indicates that the switch in EP fate after wounding is reversible. In summary, these results show that EE is bothmaintained and repaired by a single progenitor cell population capable of reversibly switching between homeostatic and regenerative behavior in response to injury. These findings may be reconciled with the reported proliferative hetero- geneity of EE cells in vitro if only some EP cells switch into “wound mode”when placed into cul- ture (10). The widespread participation of pro- genitors in tissue repair provides a rapid and robust mechanism of wound healing without an underpinning stem cell pool. References and Notes 1. T. Tumbar et al., Science 303, 359 (2004). 2. N. Barker et al., Cell Stem Cell 6, 25 (2010). 3. N. Barker et al., Nature 449, 1003 (2007). 4. V. Jaks et al., Nat. Genet. 40, 1291 (2008). 5. E. Goetsch, Am. J. Anat. 10, 1 (1910). 6. B. Messier, C. P. Leblond, Am. J. Anat. 106, 247 (1960). 7. J. P. Marques-Pereira, C. P. Leblond, Am. J. Anat. 117, 73 (1965). 8. J. P. Seery, F. M. Watt, Curr. Biol. 10, 1447 (2000). 9. D. Croagh, W. A. Phillips, R. Redvers, R. J. Thomas, P. Kaur, Stem Cells 25, 313 (2007). 10. J. Kalabis et al., J. Clin. Invest. 118, 3860 (2008). 11. D. Croagh, R. J. Thomas, W. A. Phillips, P. Kaur, Stem Cell Rev. 4, 261 (2008). 12. J. Dent, H. B. El-Serag, M. A. Wallander, S. Johansson, Gut 54, 710 (2005). 13. A. Jemal et al., CA Cancer J. Clin. 61, 69 (2011). 14. T. Kanda, K. F. Sullivan, G. M. Wahl, Curr. Biol. 8, 377 (1998). 15. K. Hochedlinger, Y. Yamada, C. Beard, R. Jaenisch, Cell 121, 465 (2005). 16. H. J. Snippert et al., Cell 143, 134 (2010). 17. K. M. Braun et al., Development 130, 5241 (2003). 18. E. Clayton et al., Nature 446, 185 (2007). 19. D. P. Doupé, A. M. Klein, B. D. Simons, P. H. Jones, Dev. Cell 18, 317 (2010). 20. A. M. Klein, D. P. Doupé, P. H. Jones, B. D. Simons, Phys. Rev. E Stat. Nonlin. Soft Matter Phys. 76, 021910 (2007). 21. A. M. Klein, B. D. Simons, Development 138, 3103 (2011). 22. C. Lopez-Garcia, A. M. Klein, B. D. Simons, D. J. Winton, Science 330, 822 (2010). 23. B. Chapellier et al., EMBO J. 21, 3402 (2002). 24. C. A. Collins, F. M. Watt, Dev. Biol. 324, 55 (2008). 25. C. S. Potten, T. D. Allen, J. Cell Sci. 17, 413 (1975). 26. G. C. Gurtner, S. Werner, Y. Barrandon, M. T. Longaker, Nature 453, 314 (2008). 27. P. Martin et al., Curr. Biol. 13, 1122 (2003). 28. A. Jacinto, A. Martinez-Arias, P. Martin, Nat. Cell Biol. 3, E117 (2001). Acknowledgments: We thank E. Choolun and the staff at ARES and CBS Cambridge for technical assistance; D. Winton and D. Adams (Cambridge) for mice; and M. Gonzalez (London) for the Geminin antibody. We acknowledge the support of the MRC, EPSRC (Engineering and Physical Sciences Research Council), the NC3Rs (National Centre for the Replacement, Refinement and Reduction of Animals in Research), the Wellcome Trust, Sidney Sussex College, Cambridge (D.P.D.), European Union Marie Curie Fellowship PIEF-LIF-2007-220016 (M.P.A.), the Royal College of Surgeons of England (A.R.), and Cambridge Cancer Centre (A.R.). This work uses methods included in the patent WO2009010725 (A2), a method of detecting altered behavior in a population of cells; inventors were P.H.J., B.D.S., and A.M.K. Supplementary Materials www.sciencemag.org/cgi/content/full/science.1218835/DC1 Materials and Methods Figs. S1 to S13 Table S1 References (29, 30) 6 January 2012; accepted 10 July 2012 Published online 19 July 2012; 10.1126/science.1218835 Fig. 4. Response of EP to wounding. Cartoons show protocols; blue triangles indicate sampling. (A to C) Wounding of clonally labeled mice. Confocal z stacks, 1 (B) or 10 (C) days after biopsy. Solid line shows pz-mf boundary; dashed line shows wound margin. (A) Day 1 unwounded control. EYFP is yellow, keratin 14 (Krt14) red, and EdU grayscale. Scale bars, 50 mm. (D to F) Dilution of HGFP. Confocal z stacks from unwounded control day 2 (D) and wounded mice at 2 (E) and 5 (F) days after biopsy, showing HGFP (green). Arrow indicates HGFP bright cell (overexposed to reveal remaining cells); such cells stain for CD45 (red, inset). Scale bars, 10 mm. (G to I) Cell proliferation. Confocal z stacks from unwounded control at 2 days (G) and experimental mice at 2 (H) and 5 (I) days after biopsy stained for EdU (grayscale). Scale bars, 10 mm. www.sciencemag.org SCIENCE VOL 337 31 AUGUST 2012 1093 REPORTS o n Se pt em be r 3 , 2 01 2 ww w. sc ien ce m ag .o rg Do wn loa de d fro m Originally published 19 July 2012; corrected 22 August 2012 www.sciencemag.org/cgi/content/full/science.1218835/DC1 Supplementary Materials for A Single Progenitor Population Switches Behavior to Maintain and Repair Esophageal Epithelium David P. Doupé, Maria P. Alcolea, Amit Roshan, Gen Zhang, Allon M. Klein, Benjamin D. Simons, Philip H. Jones* *To whom correspondence should be addressed. E-mail: phj20@cam.ac.uk Published 19 July 2012 on Science Express DOI: 10.1126/science.1218835 This PDF file includes: Materials and Methods Figures S1 to S13 Table S1 References Supplementary Methods: Theory Correction: The “Supplementary Methods: Theory” section was added. 2 Methods Animals Adult mice doubly transgenic for the inducible cre allele AhcreERT and the conditional reporter allele of EYFP targeted to the Rosa 26 locus (AhcreERTR26flEYFP/wt) were generated as described (12, 13). In these mice, transcription of the cre mutant estrogen receptor fusion protein (creERT) is induced by ß-napthoflavone. Tamoxifen is also required for creERT to gain access to the nucleus and excise the loxP flanked “STOP” cassette resulting in EYFP expressionEYFP expression was induced in animals aged 8-12 weeks by an intraperitoneal dose of 80mg/kg -napthoflavone and 1mg Tamoxifen (18). For label retaining studies, Rosa26M2rtTA/TetO-HGFP mice doubly transgenic for a reverse tetracycline-controlled transactivator (rtTA-M2) targeted to the Rosa 26 locus and a HIST1H2BJ/EGFP fusion protein (HGFP) expressed from a tetracycline promoter element were used (1,14 and 15). HGFP expression was induced by treatment with doxycycline (DOX, 2mg/ml in drinking water sweetened with sucrose) for 4 weeks, after which DOX was withdrawn for two and four weeks in order to chase HGFP dilution (1,14 and 15) . Cohorts of three animals per time point were culled and esophagus and tail epidermis taken for analysis. Visualization of Lgr5 transcription was achieved by using the Lgr5EGFP-IRES-CreERT2/+ reporter mouse, in which EGFP was targeted to the 3’ untranslated region of the Lgr5 gene (3). All experiments were conducted according to Home Office project licenses PPL80/2056 and PPL22/2282. Immunostaining Wholemounts from the middle third of esophageal epithelium and tail skin were prepared by cutting tissue into rectangular pieces of approximately 5 by 8mm and incubating for 2-3 hours in 5mM EDTA at 37ºC. The epithelium was then carefully peeled away from underlying tissue with fine forceps and fixed in 4% paraformaldehyde (PFA) in PBS for 15- 25 minutes. For staining, wholemounts were blocked for 1 hour in staining buffer (0.5% Bovine Serum Albumin, 0.25% Fish skin gelatin, and 0.5% Triton X-100 in PBS) with 10% donkey or goat serum, according to the secondary antibody used. Primary and secondary antibodies were incubated in staining buffer overnight, followed by washing for 2 hours with 0.2% Tween-20 in PBS. Cryosections of 10µm thickness were fixed with 4% PFA for 5 minutes. EdU incorporation was detected with a Click-iT® chemistry kit according to the manufacturer’s instructions (Invitrogen). When EdU staining was combined with immunofluorescent staining, the Click-labeling reaction was performed between primary and secondary antibody incubation. Please refer to Table S1 for a list of antibodies used in immunofluorescence. Confocal images were acquired on Zeiss LSM 510 META, Nikon ECLIPSE TE2000-U and Leica TCS SP5 II confocal microscopes and reconstructed using Volocity 5 image processing software (Improvision). Esophageal endoscopy and biopsy Mice undergoing endoscopy had anesthesia using a 100mg/kg Ketamine (Pfizer Animal Health) and 10mg/kg Xylazine (Bayer HealthCare AG) combination administered intraperitoneally. A 9.5Fr 30° forward oblique diagnostic miniature endoscope with a 3Fr instrument channel was used in conjunction with an AIDA COM II image capture system for high definition video recording (Karl Storz GmBH). Cannulation of the mouse esophagus was done under direct visualization up to a distance of 1.5-2cm from the vocal chords. 3Fr diameter biopsy forceps with double action jaws (Karl Storz GmBH) were used to create 3 superficial wounds in the middle third of the mouse esophagus of between 0.4 – 0.9mm diameter. Anesthesia was reversed using Atipamezole (Pfizer Animal Health) given at 1mg/kg subcutaneously at least 20 minutes after induction. Post-procedure, animals were maintained on a soft mashed diet and euthanized at appropriate timepoints mentioned in the text. Dilution of Histone 2B-EGFP (HGFP) protein in esophageal epithelium (EE) in Rosa26M2rtTA/TetO-HGFP mice To quantify the intensity of HGFP fluorescence in EE from Rosa26M2rtTA/ TetO-HGFP mice, samples were analyzed at time 0 (immediately after 4 weeks DOX treatment) and two weeks later. Unstained wholemounts from the middle third of the esophagus were imaged on a Leica TCS SP5 II confocal microscope using identical settings for all samples. Single 2 m slice images of the basal layer were analyzed for quantitative assessment of HGFP fluorescence in each independent cell by using Volocity 5 (Improvision). Flow cytometric analysis Esophagi from 3-5 adult mice (C57/Bl6; 2-3 month old) per sample were pooled, opened longitudinally, washed in PBS and treated with 0.5mg/ml thermolysin (Sigma) for 30 minutes at 37°C. The epithelium was then carefully peeled away from underlying tissue with fine forceps and washed in PBS to remove any contaminants. Epithelial sheets were then thoroughly minced and single-cell suspension obtained by non-enzymatic tissue dissociation using gentleMACS™ Dissociator (Miltenyi Biotec) followed by filtration through a 30µm cell strainer. Cells were centrifuged and resuspended in 1% fetal bovine serum in PBS. Staining with primary antibodies and/or isotype controls (5µg/ml) was performed for 30 minutes at 4°C. The following immunoglobulins were used for FACS at (1:100): FITC Rat anti-Mouse CD34 (RAM34; #553733) and FITC Rat IgG2a, isotype control (#553929) both from BD Pharmingen; Alexa Fluor® 647 Hamster anti-Mouse/Rat CD29 (HMȕ1-1; #102214) and Alexa Fluor® 647 Hamster IgG isotype control (#400924) both from Biolegend. Incorporation of 7AAD (2µg/ml) was used to determine the cell viability. Samples were analyzed on a MoFlow cell sorter (Dako Cytomation) and FlowJo FACS software (v7/9; Tree Star, Inc). 45,000 viable single cells were acquired for analysis. Isotype controls were used to exclude unspecific immunoglobulin binding. Surface expression of CD29 was used to identify basal cells. A population of doubly positive cells for CD34/CD29 could not be identified when compared to fluorescence minus one control for FITC (CD34/7AAD/CD29 was compared to FITC isotype/7AAD/CD29). Results were reproduced in each of 3 independent experiments. Activated Caspase-3 staining Wild type mice esophagi were cut into 10 µm cryosections. Immunostaining was performed as described. Fluorescent images were collected on a Zeiss Observer D1 microscope. 20 sections were analyzed for each mouse (3 in total), containing a total of 30000 basal cells. A positive control for the staining was obtained by irradiating a fresh esophageal sample with UVC and maintaining it in explant culture for 24 hours prior to sectioning (18). 4 Parameters measured to determine homeostasis Wholemounts from the middle third of the oesophagus were stained with DAPI and the basal cell marker keratin14, and Z stack images acquired by confocal microscopy. Images were viewed in Volocity software (Improvision). DAPI staining was used to score the number of basal cells per unit area and mitotic index, and Ki67 positive basal cells were scored from confocal images: 10 fields containing at least 2000 cells were scored for each mouse. The basal:suprabasal cell ratio was determined from Z stacks encompassing the basal layer and all nucleated suprabasal cells. The ratio of keratin 14 positive basal cells to nucleated, DAPI positive, suprabasal cells was scored using Volocity in at least three Z stacks per mouse. Representativeness of labeling Z stack images of clones one year post-induction were visualized by anti-GFP immunostaining. To assess proliferation in clones and surrounding unlabelled cells, we co- stained for either Ki-67 or Geminin, a protein expressed solely from S phase to early mitosis, using an anti-Geminin antibody that gives no staining in Geminin null embryos (29). The basal: nucleated suprabasal cell ratio and the number of basal cells/unit area were scored in 138 clones and compared with the ratio in unlabelled cells measured as above. Explant cultures In order to study esophageal epithelial regrowth without the influence of the immune system, we developed a novel explant technique. Esophageal submucosa from wild type C57/Bl6 mice is dissected away from both the mucosa and muscle layers after 2 hour incubation with 5mM EDTA at 37ºC. The obtained submucosa is then cut in 5mm x 10 mm pieces and placed in transparent tissue culture inserts of 0.4 microns pore size (Greiner Bio- one). Immediately afterwards, two strips of esophageal epithelium (5 x 1 mm) from 7 days induced AhcreERTR26flEYFP/wt mice were placed on top of the submucosa. The mounted explants were left for 5 minutes to settle at 37ºC, and then covered with minimal medium containing: 1 part DMEM, 1 part Ham’s F12, 10% fetal calf serum, 0.18 mM adenine, 5 µg/ml transferrin, 100 U/ml penicillin, 100 µg/ml streptomycin, 1.25ng/ml amphoteracin. The epithelium was cultured for 15 days at 37ºC and 5% CO2, replacing the medium on alternate days. During this time new epithelium grows over the submucosa developing a new epithelial sheet. The new epithelium was either cryosectioned or wholemounted by peeling it away from the submucosa in a similar manner to the esophageal samples described above. Statistical AnalysisThe methods used to analyze the cell lineage tracing experiments in AhcreERTR26flEYFP/wt mice are set out in supplementary methods: theory. 5 Fig. S1 Dilution of Histone 2B-EGFP (HGFP) protein in esophageal epithelium (EE) in Rosa26M2rtTA/TetO-HGFP mice and immunostaining of HGFP label retaining cells. HGFP dilution in EE. Rosa26M2rtTA/TetO-HGFP mice were treated with doxycycline (DOX) for 4 weeks, after which DOX was withdrawn and animals culled immediately (time 0) and 4 weeks later (4w) (Fig. 1C). A, B: Wholemounts of EE were immunostained for GFP (green). At time 0, HGFP is expressed throughout the epithelium, whilst scattered HGFP retaining cells are seen at 4w. Scale bars: Main panels 500 m, insets10 m. C - F: Rendered confocal z stacks of EE basal layer showing label retaining cells (green) at 4w are negative for Keratin 14 (red, C, D), but positive for CD45 (Fig. 1E inset), comprising Langerin positive Langerhan’s cells (E) and CD3 positive lymphocytes (F). Scale bars 5 m. G: Quantification of HGFP dilution. Box plots show median (central line), 25th and 75th percentile (box), 2.5th and 97.5th percentile (whiskers) and outliers (dots) for three mice at times indicated. AU, arbitrary units. 6 Fig. S2 Histone 2B-EGFP (HGFP) retaining cells in EE do not express CD34 or Lgr5. A: CD34 immunostaining in HGFP retaining cells in Rosa26M2rtTA/TetO-HGFP mice treated with doxycycline (DOX) for 4 weeks, after which DOX was withdrawn and animals culled 4 weeks later (4w, Fig. 1C). Wholemounts of EE and epidermis were immunostained for HGFP (green), CD34 (red), the epithelial cell marker E Cadherin (purple) and DAPI (blue). CD34 is expressed by HGFP retaining cells in the hair follicle bulge (1), but was undetectable in the HGFP retaining cells in the EE from the same animals. Scale bar 10 m. B: Lgr5 expression in EE. Esophageal and epidermal wholemounts were prepared from Lgr5EGFP-IRES-CreERT2/+ mice in which EGFP, targeted to the 3’ untranslated region of the Lgr5 gene, reports Lgr5 transcription (3). Cells in the hair follicle express EGFP, as previously reported, but no EGFP expression is detected in EE (5). Lgr5 transcription is revealed by GFP staining, green, Keratin14 (Krt14) is red and Dapi is blue. Scale bars: Esophagus 10 m, Epidermis 50 m. 7 Fig. S3 Esophageal epithelium does not contain a CD34 positive cell population. Esophageal keratinocytes from C57/Bl6 mice were freshly isolated and stained for FACS analysis with 7AAD (viability marker), CD29-Alexa Fluor® 647 (basal keratinocyte marker) and, CD34-FITC or the FITC isotype control (A and B, respectively). 45000 events from viable single cells were acquired per sample. Comparison of the FITC-CD34 stained sample with the FITC-isotype control reveals no specific staining for CD34. Results shown are typical of three independent experiments. 8 Fig. S4 Homeostasis in EE over one year A-D: Measurements of mitotic index (A), % Ki67 positive basal cells (B), basal cell density (cells/100 m2) (C), and the ratio of basal cells: nucleated suprabasal cells (D) were performed, from at least 3 mice per time, dashed line shows the mean of all values at all time points, error bars indicate s.e.m. E: Activated Caspase 3 staining in EE. Left panel shows section of normal EE, right panel shows a UV irradiated explant culture as a positive control. Blue is Dapi, arrowheads indicate activated Caspase 3 positive cells (green), dashed line indicates basement membrane, scale bar 10 m. None of 30,000 basal cells from normal esophagus were positive for activated Caspase 3. Fig. S5 Representativeness of clonally labeled cells A-D: Data from confocal imaging of labeled cells within clones and surrounding non labeled cells in EE wholemounts one year post-labeling. A, % Geminin positive basal cells; B, % Ki67 positive basal cells; C, basal cell density (cells/100• •m2); D, the ratio basal to nucleated suprabasal cells. Error bars indicate s.e.m. There was no statistically significant difference (P>0.2) when comparing labeled and unlabeled cells for any parameter by 2 tailed Student's t- test. A-15 10 Fig. S6 Clonal data and fit to model A: Clone quantification. Clone size distribution (basal cells/clone) in a total of 1784 clones. Each point represents one clone, histograms are normalized so modal values are of equal width at each time. B: Cumulative clone size distribution shows convergence onto scaling behavior at late times. The raw basal layer clone size distribution in panel A is reproduced here as the cumulative clone size distribution, Cn(t), plotted as a function of n divided by the average clone size for each timepoint. The cumulative distribution Cn(t) represents the probability of finding a surviving clone with a basal layer size larger than n. At times of 3 months or more post- induction, the data sets converge onto each other, a manifestation of long-term scaling behavior. Combined with the observed homeostatic nature of the turnover, such scaling behavior shows that the self-renewing progenitor cells function as a functionally equivalent population. The black curve denotes an exponential cumulative clone size distribution; the long-term behavior predicted for any strategy involving population asymmetric self-renewal including the model discussed here (1-2). 11 Fig S6 continued C, D: Detailed clone size distribution at 30 (C) and 84 (D) days post-induction. Graph shows the joint distribution of basal and suprabasal cells in a total of 292 (C) and 489 (D) clones containing at least one basal layer cell, e.g. bar in red box in C indicates clones with 1 basal cell (b) and 2 suprabasal cells (sb), as shown in cartoon. The blue crosses show the prediction of the model represented in Fig. 2E with the 95% plausible interval indicated by vertical blue bars (see Supplementary Discussion and Results). E: Cumulative basal layer clone size distribution, Cn(t), as a function of size (number of cells) rescaled by the average for each time point, n/ as in A, i.e. Cn(t) denotes the probability of finding a clone with a size equal to, or larger than, n/. Here we have presented the cumulative probability distribution on a logarithmic scale and, for clarity, we have separated consecutive time points by one decade. The points denote experimental data from panel A, and the lines represent the corresponding model predictions following the fit to the data with the parameters shown in Fig. 2E and 2F. The shaded regions represent estimates of the stochastic error due to finite sample size, indicating approximately one standard deviation (68%). In the long time limit, the model predicts that the distribution should tend to a simple exponential, exp(-n/), (black line). The model shows good agreement with the experimental data at both short and long time points. 12 Fig S7 Response to atRA and second treatment protocol A: Sections from control and treated mice, 24 hours after a single dose of atRA, stained for retinoid target genes CRABPII (red) or FABP5 (green), and DAPI (blue). Dotted lines enclose EE, note thickening in atRA samples. Scale bar 20 m. Graph shows % Ki67 positive basal cells during atRA treatment. B: To validate the model of the effects of atRA on EE shown in Fig. 3D, labeling was induced in AhcreERTR26flEYFP/wt mice, which were then left untreated for 21 days after which they were treated with atRA on alternate days for 9 days. EE was then collected. C: Number of basal cells per clone in EE treated as in B: orange triangles indicate experimental data, blue line indicates prediction of model shown in Figs 2F and 3D with 95% plausible interval. A total of 316 clones were scored in 3 mice. D: Detailed clone size distribution of the same 316 clones, in orange. The blue crosses show the prediction of the model represented in Figs 2F and 3D with the 95% plausible interval indicated by vertical blue bars (see Supplementary text). 13 Fig. S8 Controls for esophageal biopsy experiments and low power views of wound sites in biopsied mice A-C Controls were anesthetized but did not undergo endoscopic biopsy, (see Fig. 4). A: Clonal labeling control at 10 day post-wounding (compare with Fig. 4D). AhcreERTR26flEYFP/wt mice were treated as shown in Fig. 4, column 1. EdU was given 1h before culling the animals and wholemounts were stained for YFP (yellow), EdU (white) and Keratin14 (Krt14, red). Scale bar 50 m. B: HGFP dilution at 5 days post-wounding (compare with Fig. 4F). Rosa26M2rtTA/TetO-HGFP mice were treated with doxycycline (DOX) for 4 weeks pre-biopsy and sampled as shown in Fig. 4, column 2. Wholemounts were imaged by confocal microscopy to detect endogenous HGFP fluorescence (green). Scale bar 10 m. C: 24 hour EdU control at 5 day post-wounding (compare with Fig. 4I). Wild type animals were treated as shown in Fig. 4, column 3. EdU was given 24h before culling and wholemounts stained for EdU (greyscale). Scale bar 10 m. Clone appearances, EdU staining and HGFP levels distant from the wound site in biopsied animals were similar to those in the unwounded controls. D, E: Low power views of the wounds shown in Figs 4B and 4C respectively (regions imaged in Fig 4 indicated by white box), wholemounts were stained for YFP (yellow), EdU (white) and Keratin14 (red), scale bars 50 m. F: Stereoscopic image of wound 2 days post-biopsy, scale bar 500 m. 14 Fig. S9 Progenitor fate switching in esophageal explant cultures. A: Protocol for explant culture on submucosa: clonal labeling was induced in AhcreERTR26flEYFP/wt mice 7 days prior tissue harvesting. Parallel strips of EE, with underlying submucosa, were laid on submucosa from unlabeled mice and cultured for a period of 15 days, over which new epithelium forms between the strips (see supplementary methods). B-C: cryosections of cultured epithelium, stained for hematoxylin and eosin (B) or Cyclin D1 (green) and DAPI (blue, C). Dashed line indicates basement membrane, scale bar 10 m. D: Rendered confocal Z stack of wholemount of cultured epithelium stained for the basal layer marker keratin 14 (red) and the suprabasal marker keratin 4 (purple), dotted line indicates basement membrane. Scale bar 10 m. E: Confocal z stack of epithelial wholemount preparation of explant culture. Yellow is YFP, blue is DAPI. Dashed lines indicate boundary of new epithelium formed in culture and the original strips of EE. Scale bar 200 m. Note labeled cells make a substantial contribution to the new epithelium, consistent with in vivo behavior (Fig. 4). 15 Fig. S10 Epithelial proliferation at wound site. A-B: Low power views of wounded EE in Rosa26M2rtTA/TetO-HGFP mice treated with doxycycline (DOX) for 4 weeks pre-biopsy and sampled as shown in Fig. 4 column 2, at 2 days (A) and 5 days (B) post-biopsy, HGFP is green, note the epithelial defect has closed by 2 days in A. C-F: Low power view of wounded EE in wild type mice, treated with EdU 24 hours before culling, at 2 days (C, E) or 5 days (D, F) post-biopsy (Fig. 4, column 3). EdU is grayscale, dapi blue, dashed line in C and E indicates epithelial injury. Note HGFP retaining cells, encircled in dotted line in A and B, and EdU negative cells adjacent to the wound margin (C-F) correspond to the migratory front as indicated in Fig. 4B. Scale bars 100 m. 16 Fig. S11 Estimation of from the basal clone size distribution of individual mice (data from Fig. S6, see supplementary theory). From the data from each mouse, we can separately estimate the combined parameter Here we plot log with 68% (1 ) credible intervals for 28 mice, along with the combined average (and 1 credible interval) in green from treating all mice as exactly identical. 17 Fig. S12 Estimation of from the detailed (basal and suprabasal) clone size distributions of individual mice (see Supplementary Theory). log with 68% (1 ) credible intervals is plotted for 11 mice, with the combined average in purple. The combined average from basal data in fig. S11 is reproduced in green. 18 Fig. S13 Relative frequency of clones with one basal cell and no suprabasal cells (see supplementary theory). There is a very slow decay of the relative frequency of clones consisting of a single basal cell. Nevertheless, this is well accounted for by the model (Fig. 2F). Here we plot the observed relative frequency in orange, and in blue the model prediction (with 95% likelihood intervals arising from finite number of clones counted). 19 Table S1: Antibodies used for immunofluorescence Antigen Clone Company Cat No Species Dilution GFP - Invitrogen A11039 Chicken 1/500 K14 - Covance PRB-155P Rabbit 1/1000 K4 6B10 Vector VP-C399 Mouse 1/1000 CD45 30-F11 Biolegend 103102 Rat 1/200 CD3 17A2 eBiosciences 16-0032-82 Rat 1/100 CD34 RAM34 BD Biosciences 553731 Rat 1/100 Langerin - Santa Cruz sc-22620 Goat 1/100 E-cadherin 24E10 Cell Signaling 3195 Rabbit 1/100 CyclinD1 SP4 Thermo Scientific RM-9104 Rabbit 1/100 Active Caspase3 - Abcam Ab2302 Rabbit 1/100 Geminin - Gift Dr. MA Gonzalez 3988065 Rabbit 1/300 CRABPII - Proteintech 10225-1-AP Rabbit 1/50* FABP5 - R&D systems AF1476 Goat 1/50* *Acetone/methanol fixation: all other samples were fixed with paraformaldehyde. 20 References 1. T. Tumbar et al., Defining the epithelial stem cell niche in skin. Science 303, 359 (2004). doi:10.1126/science.1092436 Medline 2. N. Barker et al., Lgr5+ve stem cells drive self-renewal in the stomach and build long-lived gastric units in vitro. Cell Stem Cell 6, 25 (2010). doi:10.1016/j.stem.2009.11.013 Medline 3. N. Barker et al., Identification of stem cells in small intestine and colon by marker gene Lgr5. Nature 449, 1003 (2007). doi:10.1038/nature06196 Medline 4. V. Jaks et al., Lgr5 marks cycling, yet long-lived, hair follicle stem cells. Nat. Genet. 40, 1291 (2008). doi:10.1038/ng.239 Medline 5. E. Goetsch, The structure of the mammalian oesophagus. Am. J. Anat. 10, 1 (1910). doi:10.1002/aja.1000100102 6. B. Messier, C. P. Leblond, Cell proliferation and migration as revealed by radioautography after injection of thymidine-H3 into male rats and mice. Am. J. Anat. 106, 247 (1960). doi:10.1002/aja.1001060305 Medline 7. J. P. Marques-Pereira, C. P. Leblond, Mitosis and differentiation in the stratified squamous epithelium of the rat esophagus. Am. J. Anat. 117, 73 (1965). doi:10.1002/aja.1001170106 Medline 8. J. P. Seery, F. M. Watt, Asymmetric stem-cell divisions define the architecture of human oesophageal epithelium. Curr. Biol. 10, 1447 (2000). doi:10.1016/S0960-9822(00)00803- 4 Medline 9. D. Croagh, W. A. Phillips, R. Redvers, R. J. Thomas, P. Kaur, Identification of candidate murine esophageal stem cells using a combination of cell kinetic studies and cell surface markers. Stem Cells 25, 313 (2007). doi:10.1634/stemcells.2006-0421 Medline 10. J. Kalabis et al., A subpopulation of mouse esophageal basal cells has properties of stem cells with the capacity for self-renewal and lineage specification. J. Clin. Invest. 118, 3860 (2008). Medline 11. D. Croagh, R. J. Thomas, W. A. Phillips, P. Kaur, Esophageal stem cells: A review of their identification and characterization. Stem Cell Rev. 4, 261 (2008). doi:10.1007/s12015- 008-9031-3 Medline 12. J. Dent, H. B. El-Serag, M. A. Wallander, S. Johansson, Epidemiology of gastro-oesophageal reflux disease: A systematic review. Gut 54, 710 (2005). Medline 13. A. Jemal et al., Global cancer statistics. CA Cancer J. Clin. 61, 69 (2011). doi:10.3322/caac.20107 Medline 14. T. Kanda, K. F. Sullivan, G. M. Wahl, Histone-GFP fusion protein enables sensitive analysis of chromosome dynamics in living mammalian cells. Curr. Biol. 8, 377 (1998). doi:10.1016/S0960-9822(98)70156-3 Medline 15. K. Hochedlinger, Y. Yamada, C. Beard, R. Jaenisch, Ectopic expression of Oct-4 blocks progenitor-cell differentiation and causes dysplasia in epithelial tissues. Cell 121, 465 (2005). doi:10.1016/j.cell.2005.02.018 Medline 20a 16. H. J. Snippert et al., Intestinal crypt homeostasis results from neutral competition between symmetrically dividing Lgr5 stem cells. Cell 143, 134 (2010). doi:10.1016/j.cell.2010.09.016 Medline 17. K. M. Braun et al., Manipulation of stem cell proliferation and lineage commitment: Visualisation of label-retaining cells in wholemounts of mouse epidermis. Development 130, 5241 (2003). doi:10.1242/dev.00703 Medline 18. E. Clayton et al., A single type of progenitor cell maintains normal epidermis. Nature 446, 185 (2007). doi:10.1038/nature05574 Medline 19. D. P. Doupé, A. M. Klein, B. D. Simons, P. H. Jones, The ordered architecture of murine ear epidermis is maintained by progenitor cells with random fate. Dev. Cell 18, 317 (2010). doi:10.1016/j.devcel.2009.12.016 Medline 20. A. M. Klein, D. P. Doupé, P. H. Jones, B. D. Simons, Kinetics of cell division in epidermal maintenance. Phys. Rev. E Stat. Nonlin. Soft Matter Phys. 76, 021910 (2007). doi:10.1103/PhysRevE.76.021910 Medline 21. A. M. Klein, B. D. Simons, Universal patterns of stem cell fate in cycling adult tissues. Development 138, 3103 (2011). doi:10.1242/dev.060103 Medline 22. C. Lopez-Garcia, A. M. Klein, B. D. Simons, D. J. Winton, Intestinal stem cell replacement follows a pattern of neutral drift. Science 330, 822 (2010). doi:10.1126/science.1196236 Medline 23. B. Chapellier et al., Physiological and retinoid-induced proliferations of epidermis basal keratinocytes are differently controlled. EMBO J. 21, 3402 (2002). doi:10.1093/emboj/cdf331 Medline 24. C. A. Collins, F. M. Watt, Dynamic regulation of retinoic acid-binding proteins in developing, adult and neoplastic skin reveals roles for beta-catenin and Notch signalling. Dev. Biol. 324, 55 (2008). doi:10.1016/j.ydbio.2008.08.034 Medline 25. C. S. Potten, T. D. Allen, The fine structure and cell kinetics of mouse epidermis after wounding. J. Cell Sci. 17, 413 (1975). Medline 26. G. C. Gurtner, S. Werner, Y. Barrandon, M. T. Longaker, Wound repair and regeneration. Nature 453, 314 (2008). doi:10.1038/nature07039 Medline 27. P. Martin et al., Wound healing in the PU.1 null mouse: Tissue repair is not dependent on inflammatory cells. Curr. Biol. 13, 1122 (2003). doi:10.1016/S0960-9822(03)00396-8 Medline 28. A. Jacinto, A. Martinez-Arias, P. Martin, Mechanisms of epithelial fusion and repair. Nat. Cell Biol. 3, E117 (2001). doi:10.1038/35074643 Medline 29. M. A. Gonzalez et al., Geminin is essential to prevent endoreduplication and to form pluripotent cells during mammalian development. Genes Dev. 20, 1880 (2006). doi:10.1101/gad.379706 Medline 30. A. M. Klein, D. P. Doupé, P. H. Jones, B. D. Simons, Mechanism of murine epidermal maintenance: Cell division and the voter model. Phys. Rev. E Stat. Nonlin. Soft Matter Phys. 77, 031907 (2008). doi:10.1103/PhysRevE.77.031907 Medline Supplementary methods: theory In the main text, we describe a lineage tracing experiment using inducible AhcreERTR26flEYFP/wt mice. The conclusions of our study hinge on the quantitative insights that can be drawn from the analysis of the clones of labeled cells in this experiment. The aim of this supplementary methods theory section is to expand on the basis of this analysis and its application. In doing so, we will set out in detail the experimental evidence in support of the esophageal progenitor (EP) cell model, how the parameters of this model are constrained by the experimental data, and how the model can be challenged experimentally by drug treatment. In section 1, we will discuss the additional observations that underpin the quantitative analysis of the clonal fate data – the homeostatic nature of esophageal epithelium (EE), and the representativeness of labeling. In section 2 we will describe how the experimental data identifies the dynamics of a single pro- genitor cell population allowing us to formulate a simple model of EE turnover. In section 3 we will use the clonal fate data to fit the parameters of the modeling scheme. Finally, in section 4, we will challenge the model by considering the effects of drug delivery on tissue. 1 Clonal analysis: controls 1.1 Homeostasis of tissue turnover in normal EE To prepare for the quantitative analysis of the clone fate data, it is first necessary to establish whether the EE is in homeostasis over the year long timecourse of the experiment. We found that there was no significant difference in the mitotic index in the basal layer, the proportion of 21 cells expressing the proliferation associated antigen Ki-67, the number of basal cells per unit area or the ratio of basal cells to differentiated, nucleated suprabasal cells over the year of the experiment (Figs S4A–D). On this basis, we conclude that the tissue was indeed homeostatic over this period. 1.2 Apoptosis We also investigated whether apoptosis occurred in normal EE by staining for activated Caspase 3. Positive cells were readily detected in control sections of EE which had been UV irradiated and cultured for 24 hours, but undetectable in 30,000 basal cells in sections of normal EE (Fig. S4E).We therefore conclude that apoptosis is negligible in normal EE. 1.3 Representativeness of the labeled cell population Although the analysis of the Rosa26M2rtTA/TetO-HGFP transgenic mice confirms the absence of slow-cycling or quiescent epithelial cells in EE, this does not rule out heterogeneity in the proliferative potential of cycling cells. To reliably interpret the results of the clonal labeling experiment, it is therefore important to confirm that the labeled cell population is representative of the total cell population and therefore capable of revealing proliferative heterogeneity should it exist. To this end, we first compared the cell composition within clones at the one-year timepoint with that of the surrounding unlabeled epithelium. We found that there was no statistically sig- nificant difference in the proportion of cycling cells between the labeled and unlabeled tissue, as determined by immunostaining for the S-M phase marker Geminin as well as the prolifer- ation associated antigen Ki-67 (Figs S5 A, B). Similarly there was no statistically significant difference in the density of basal cells (cell number per unit area), or the ratio of basal cells to nucleated suprabasal cells when comparing cells within and outside labeled clones (Figs S5 C, 22 D). Although the cell composition and activity of the clones within the unlabeled tissue at the one-year timepoint confirm that the labeled cell population is representative of the total pop- ulation, it does not immediately follow that dividing and non-dividing cells are induced in a representative manner, or that there is not a proliferative hierarchy comprising a self renewing stem cell population which supports a population of non self-renewing progenitor cells. Pro- viding the induction protocol includes the self-renewing population, over time, the persistent clones will converge to a stationary and representative ensemble, while clones derived from a non self renewing progenitor cells that were primed or committed to terminal differentiation on induction are transient and will become lost over time. However, since we find no statisti- cally significant variation in the percentage of labeled basal cells over the entire experimental timecourse, from 1 month to 1 year (Fig. 2Bc), we can conclude that either (a) clones derived from a putative non self-renewing progenitor compartment are very rapidly lost (within the first month post-labeling), or (b) that such cells do not exist - i.e. there is only one progenitor cell population - or (c) that basal layer cell types are induced in the exact proportions in which they are present in tissue. Analysis of the detailed clonal fate data below will confirm the veracity of option (b). 2 Clonal analysis: defining the model of EP cell dynamics 2.1 Scaling and equipotency To isolate and quantify the behavior of proliferating cells, we scored the number of basal cells in clones containing two or more cells (i.e. clones in which a proliferative cell was labeled at the start of the experiment) from 3 days to one year post-induction, by confocal imaging of esophageal wholemounts (Fig. 2A). Previously, it has been shown that in a similar homeostatic stratified squamous epithelium, interfollicular epidermis, a simple statistical characterization 23 can be used to both establish the equipotency of a self-renewing cell population, and elucidate the pattern of cell fate (18, 19). In particular, if self-renewal involves the stochastic loss and replacement of such progenitors, clones derived from these cells will undergo a process of “neutral drift” in which ongoing clonal loss is compensated by the expansion of adjacent clones. However complex is the underlying dynamics, such processes lead inexorably to the long-term scaling of the surviving clone size distribution, in which the chance of finding a clone with a size larger than some multiple of the average remains constant over time. Applied to the current dataset, the convergence of the basal layer clone size distribution onto a scaling form (Fig. S6A) both confirms the functional equivalence of the self-renewing progenitor cell population, and shows that the balance between their proliferation and differen- tiation is achieved on a population basis, and not at the level of individual cell divisions, i.e., in the course of turnover, some cells may undergo terminal division leading to loss while others may undergo symmetric duplication. Indeed, such behavior is consistent with the observed het- erogeneity in the clonal composition at short times, which reveals that, soon after induction, all possible permutations of two-cell clones can be found (i.e. two basal cells, one basal and one suprabasal, and two suprabasal cells). The same degree of heterogeneity is apparent in larger clones (Figs S6B, C). Taken together, the results above suggest that, as in interfollicular epidermis, EE is main- tained by a single cycling progenitor cell population in which cell division can lead to all three possible fate outcomes; two daughters that go on to divide, two cells that exit cycle and then stratify out of the basal layer, or one daughter that goes on to divide and one that differentiates (18, 19). To assess the validity of this model, and to quantify the respective rates and proba- bilities, we must turn now to a more detailed analysis of the short-time data, prior to scaling, where signatures of the detailed dynamics can be elucidated. To prepare for this analysis, we must embed the progenitor cell dynamics into a parameterization that includes the stratification 24 and loss of terminally differentiated cells. 2.2 Parameterizing EP dynamics To be concrete, we will suppose that the basal layer comprises a single population of cycling esophageal progenitor cells (with proportion ⇢), and their differentiating progeny (with propor- tion 1⇢), which remain in the basal layer without dividing until they stratify. If we assume that the progenitors divide with an average rate , and the differentiating cells stratify at an average rate , to achieve homeostasis, it follows that ⇢ = (1⇢), i.e. the rate at which differentiated cells are generated in the basal layer must be perfectly matched by the rate at which they stratify into the suprabasal cell layer. To further specify the model, we must also define the distribution of cell cycle and strat- ification times. Here, for simplicity, we will suppose that division and stratification are both uncorrelated between successive events leading, in both cases, to an exponential (Poisson) time distribution, with averages set by the rates and , respectively. While this assumption is surely unsafe (after all, cell division must be accompanied by a small refractory period before further division is possible), any degree of synchrony in the cell cycle or stratification timings will be rapidly erased from the clonal record over the timecourse. In particular, we expect features due to potential synchrony to be lost within experimental error bars after ca. 1–2 rounds of division (20). Alongside the division and stratification rates, we must further specify the probabilities for the respective fate outcomes following division. Since both the labeled cell number and their composition is found to be conserved over the time course, any process of cell division must be consistent with homeostatic turnover. Moreover, since, through scaling, EPs are seen to function as an equipotent cell population, we will therefore suppose that proliferation and differentiation is finely balanced so that, with probability, r, EP division results in two dividing cells, with 25 probability, 1 2r, one dividing and one non-dividing basal later cell, and with probability, r, two non-dividing basal layer cells. Although we cannot rule out a small contribution arising from perpendicular cell divisions resulting in the placement of one of the daughters directly into the suprabasal layer, given such divisions are observed to be rare, the effect of such “orienta- tionally asymmetric” divisions would again be impossible to resolve. Similarly, we will neglect apoptosis, which was found to be negligible in the basal layer (Fig. S4E). In summary, the time-evolution of the basal layer population, along with the clonal fate data, is therefore fully characterized by three adjustable parameters, the division rate, , the stratification rate, , and the fraction of divisions that lead to symmetrical fate, r. The progenitor cell fraction is then fixed by the rates ⇢ = /( + ). To fully characterize the behavior of the total clone size, including suprabasal as well as basal cells, we must include a further parameter, µ, which defines the average rate at which suprabasal cells are shed from the tissue (once again, we will suppose for simplicity that the corresponding distribution of shedding times is Poisson). Fortunately, this additional parameter can be related directly to the division and stratification rates through the ratio of nucleated suprabasal to basal layer cells, m, which can be measured directly. Then, since the “flux” of differentiated cells stratifying from the basal layer must be perfectly compensated by the flux of differentiated cells that are shed, we must have mµ = ⇢ = /( + ), thereby providing a relation linking µ with and . Taken together, the EP cell model dynamics can be cast as a critical (i.e. balanced) continu- ous time Markovian branching process, EP ! 8<: EP + EP Pr. rEP + TB Pr. 1 2r TB + TB Pr. r , TB ! TS, TS µ! loss, where EP represents the progenitor, TB/S differentiated cells in the basal/suprabasal layer, and the rates , , and µ are defined above. As discussed above, we suppose that EP cell fate follows a pattern of intrinsic or cell-autonomous regulation in which the fate outcome following division 26 is uncorrelated with the behavior of neighboring cells. In previous studies, we have shown that, in the two-dimensional geometry pertinent to EE, the long-term clonal dynamics is largely insensitive to this choice (30). This completes the specification of the cellular dynamics as a generalized branching-type process. In principle, we could further refine the modeling scheme to include aspects of the spatial regulation. However, our aim here is to establish and challenge the simplest model which is consistent with the observed range of clonal fate data. Indeed, we will find that, by itself, this model provides a faithful description of the cellular dynamics over the timecourse of the experiment. To assess the integrity of the model, and to fit the three adjustable parameters, we will make use of a Bayesian approach. 3 Clonal analysis: quantitative modeling 3.1 Scaling According to the EP cell paradigm, following induction, a differentiated basal or suprabasal cell would progressively stratify, lose its cell nucleus and be shed, providing only a transient contri- bution to the clonal dynamics. By contrast, a clone derived from an EP cell would progressively undergo chance expansion or contraction according to the fate choice of the constituent cells. As tissue turns over, the gradual extinction of some clones due to chance commitment to ter- minal differentiation will be perfectly compensated by the expansion of other clones. Analysis of the branching process shows that, over time, neutral drift dynamics leads to a progressive (linear) increase in the average size of the surviving clones, while the surviving fraction falls proportionally such that the total density of labeled cells remains constant (Fig. 2B) (18). In this regime, the basal layer clone size distribution (Fig. S6A) acquires a scaling behavior (Fig. S6A) described above. In particular, the chance of finding a clone with more than n basal layer cells, takes the form of an exponential exp[n/hn(t)i] where, according to the parameters of 27 the model, the average basal layer clone size (i.e. the average size of the clone “footprint” on the basal layer) is given by hn(t)i ' t/⌧ with 1/⌧ = ⇢/r. Although the long-term scaling behavior provides the means to extract the characteristic timescale ⌧ (see below), the rapid convergence to this regime does not allow the individual parameters ⇢, r, and to be inferred reliably from the basal layer clone size distribution alone. Therefore, to properly validate the model, and effect a reliable fit of the parameters, we consider the full range of clonal fate data, short-term and long-term, basal and suprabasal. 3.2 Parameter estimation The data itself is acquired in the form of a set of frequencies D = {fbn(t)} describing the number of observations of clones with b basal cells and n suprabasal cells at time t; an example is Fig. S6B. The stochastic nature of our model will predict probabilities pbn(t) for a single cell to turn into a clone at time twith b basal and n suprabasal cells, based on the parameters ⇢, r and . One might approach the fit with a least-squares method to optimize the parameters. However, as a significant portion of our data lies in the tails with large b and n, where the counts are small, we would have to employ complex methods such as ad-hoc binning or continuity corrections. As such, we have elected to use a more Bayesian approach, which allows us to analyze the experiment with no manipulation of the data. To fit the model to the range of experimental data, we implement a basic algorithm involving Bayes’ theorem for updating prior probabilities in the presence of data. More precisely, the probability that the observed clonal fate data D is described by the model is specified by the proportionality, P (, r, ⇢|D) / P (D|, r, ⇢)P (, r, ⇢) , where the posterior distribution P (D|, r, ⇢) denotes the probability of obtaining the data D given the parameters , r, and ⇢, multiplied by the likelihood that the parameters are given by 28 those same values. The constant of proportionality may be calculated a posteriori by imposing the normalisation, R1 0 d R 1 0 d⇢ R 1 2 0 dr P (, r, ⇢|D) = 1. In our case, the posterior distributions will turn out to be approximately Gaussian, so we may characterize the distribution by its first two moments and treat them as a point estimate for the parameters, and a credible interval covering approximately 68% (using one standard deviation) or 95% (using two). From Bayes’ equation above, it is apparent that we need to specify a prior distribution P (, r, ⇢) and a likelihood function P (D|, r, ⇢). We chose the maximum entropy prior as we have no further cogent information, which corresponds to ⇢ uniform on the interval [0, 1], r uniform on the interval ⇥ 0, 12 ⇤ , and log-uniform. Crucially, since we have sufficient data, the likelihood function is sharply peaked and only significant on a small support, and thus the posterior distribution is dominated by the likelihood function, and so any reasonable prior (one which does not fluctuate strongly over that narrow support) would give quantitatively similar results. Turning to the likelihood function, since each observed clone is independent, it is a multi- nomial distribution with a countable number of outcomes: P (D|, r, ⇢) = Y t [ P bn fbn(t)]!Q bn [fbn(t)!] Y bn [pbn(t)] fbn(t) , where pbn(t) denote the expected probabilities that a clone contains b basal cells and n suprabasal cells at time t post-induction. In the following, we will suppress the time index t for notational concision where it is considered unambiguous. To eliminate potential ambiguities due to uncer- tain induction frequencies of the two cell types, we conditioned the data on having at least two cells, which removes the events where a differentiated cell is induced. Moreover, to focus on a defined population, we further consider only clones which retain at least one basal layer cell. Thus we consider the probabilities, pbn 7! p˜bn = ( 0 b = 0 or b+ n < 2, pbn 1p10Pn p0n otherwise. 29 To determine the probabilities pbn, we consider the more fundamental distribution plmn de- scribing the (unconditioned) probability of finding a clone with l progenitors, m differentiated basal layer cells and n suprabasal cells. From these probabilities we can determine pbn through the relation, pbn = P l+m=b plmn. The probabilities plmn obey a continuous time Markov pro- cess, with the time-evolution given by the Master equation, d dt plmn = ⇥ r(l 1)p(l1)mn + (1 2r)lpl(m1)n + r(l + 1)p(l+1)(m2)n lplmn ⇤ + ⇥ (m+ 1)pl(m+1)(n1) mplmn ⇤ µnplmn. with the initial conditions plmn = l1m0n0. Although this is a straightforward set of coupled differential equations, direct numerical solution is prohibitively computationally expensive because each element is coupled to all the others: attempts to truncate the lmn-space causes errors that are hard to control. Instead, to develop the computational scheme, it is helpful to package the equations into the form of a gen- erating function, G(x, y, z, t) = P lmn plmn(t)x lymzn, which obeys the differential equation, @G @t = ⇥ rx2 + (1 2r)xy + ry2 x⇤ @G @x + (z y) @G @y + µ(1 z)@G @z , = [fx(x, y, z) x] @G @x + [fy(x, y, z) y] @G @y + µ [fz(x, y, z) z] @G @z . with G(x, y, z, t = 0) = x and fx(x, y, z) = rx 2 + (1 2r)xy + ry2, fy(x, y, z) = z, fz(x, y, z) = 1 which can be recognized as the probability generating functions for individual divisions of each cell type. The partial differential equation for G may be solved by the method of characteristics. In- troducing the auxiliary equations, @ @t Fv(v, t) = av [fv (Fx, Fy, Fz) v] 30 where v = (x, y, z), v ranges over the set {x, y, z}, av ranges over {, , µ} and Fv(v, t = 0) ranges over {x, y, z}, we can write the solution to G as G(x, y, z, t) = Fx(x, y, z, t). In fact, we can recognize the auxiliary function Fv as the probability generating functions for the clone size distribution starting with a cell of type v. Finally, the coefficients plmn may be extracted from G by complex analysis. Indeed, it is possible to directly extract pbn by considering G(x, x, z, t). Defining H(x, y, t) = G(x, x, y, t) = X bn pbn(t)x byn, we see that H is analytic in both x and y on the complex unit disc |x|, |y|  1, and so there are no poles within it. Defining C to be an anticlockwise contour around the unit circle, we can recover pbn by considering residues at zero: pbn(t) = ✓ 1 2⇡i ◆2 I C dx I C dy H(x, y, t) xb+1yn+1 . Note that, up to a minus sign, it is just a multi-dimensional inverse Z-transform. The integral can be approached by a sum pbn(t) = lim M!1 MX j=1 MX k=1 H e2⇡ij/M , e2⇡ik/M , t e2⇡ijb/Me2⇡ikn/M the truncation of which at finite M approximates the integral and is in the form of a two- dimensional discrete Fourier transform. This allows the use of industrial Fast Fourier Trans- forms to evaluate all pbn forM being a power of two simultaneously. We use adaptive oversam- pling to estimate the error in the truncation, and use it to bound the errors. This gives a robust black box algorithm which gives point estimates and credible intervals for arbitrary clonal data sets. Before finally carrying out the parameter estimation, it is important to confirm that the degree of inter-mouse variation is sufficiently small to justify collating data from different mice. 31 For this purpose, we made use of the long-term scaling behavior of the clone size distribution. Although, generically, the dynamics of the basal layer cells depends independently on the three adjustable parameters, ⇢, and r, in the scaling limit, the basal layer clone size depends only on the only on the combination, 1/⌧ = r/⇢, a measure of the rate of progenitor cell loss and replacement. As a result, this parameter can be estimated with a smaller uncertainty, allowing it to be inferred on a mouse by mouse basis for each timepoint. Inferring this parameter from the basal layer clone size distribution, as well as the total clone size at shorter times, the comparison shown in Figs S11 and S12 reveal that, although there is some degree of inter-mouse variation, the amalgamation of data from different mice at the same timepoint can be justified. Using the full clone data (Figs 2B, 3B, S6B and C) we computed a posterior distribution on ⇢, and r. This distribution is narrow and Gaussian-like, so we can describe it using its first two moments. Specifically, the mean is a point estimate for the parameters, and twice the standard deviation is an estimate for the 95% credible interval; the results are shown in Fig. 2C. We show a comparison between the fitted parameters and the detailed basal and suprabasal cell number distribution in Figs 3B, S6B and C. We used the point estimates to computed a set of probabilities, with which we can compute 95% likelihood intervals for fbn knowing the total number of clones observed (which we have called plausible intervals due to a lack of standard nomenclature); these are plotted as the error bars in the model, and the graphs form a visual significance test with the null hypothesis being the model with the parameters as estimated. We draw attention to the fact that these do not contain the uncertainties in the parameters themselves. Furthermore, Fig. S6D contains a detailed comparison with the much larger basal clone size data set, which shows the expected deviations from scaling at shorter times; again the error regions refer to the sampling error only. Lastly, in Fig. 2Ba and 2Bb we show the comparisons with density and average size of basal clones; the error bars in are the mouse to mouse variation. Given that we can fit the entire distributions at all times the average 32 size also fits (unsurprisingly). However it should be noted that we can make a prediction of the density which fits within the experimental error, with only one parameter, the induction efficiency, estimated by the labeling frequency at time zero (Fig. 2Ba). 4 Clonal analysis: challenging the model 4.1 Effects of atRA treatment It is straightforward to implement the same inference algorithm to address the atRA treated tissue. Wit the assumption that atRA pre-treatment establishes a new steady-state EP cell dy- namics, the resulting fit of the model to the experimental data is shown in Fig. 3D, while the resulting model predictions are shown alongside the experimental data in Fig. 3C. Once again, the fits reveal a close agreement of the model with the experimental data. With the parameters for normal and atRA treated tissue in hand, it is then possible to pre- dict the outcome when atRA is applied after induction. In particular, if we assume that, fol- lowing atRA treatment, the division and stratification rates immediately adjust from their nor- mal to atRA treated steady-state values, we can predict the resulting clone dynamics. Indeed, when compared to the measured basal clone distribution, we find that the predictions offer a favourable fit (Fig. S7C). By contrast, comparison with the joint suprabasal and basal distri- bution shows significant deviations (Fig. S7D). This departure can be easily understood as the effect of atRA treatment on cells which have already stratified and left the basal layer is more complex, and there may be some time-lag before the entire tissue changes to the new steady- state behavior. Nevertheless, these results suggests that, as far as the basal layer is concerned, the application of atRA treatment can be well-approximated as an instantaneous change of the parameters of the EP cell dynamics. 33 4.2 Single cell clones Finally, the development of the EP model relied upon the observation of long-term scaling behavior of the basal layer clone size distribution which suggested that tissue maintenance in- volves only a single equipotent progenitor cell population. However, by focusing on clones with a total size greater than one, we eliminated the potential signature of a second very slow-cycling or quiescent cell population. Although the existence of such a population in EE was ruled out by the H2B-GFP assay, with the predictions of the EP model, we can do back and question whether the dynamics of the single-cell clones are consistent with the ansatz of the modeling scheme. Since, for times in greatly in excess of the cell stratification time, 1/, differentiated cells labeled at induction will have been lost from the basal layer, we can use the long-term evolution of the single-cell basal clones to challenge the predicted model dynamics. Once again, the results, shown in Fig. S13, reveal an excellent agreement between theory and experiment over the year long timecourse. In particular, we can conclude that the persistence of single-cell clones merely reflects the set of clones which, by chance, have expanded and the sister cells have been shed leaving behind a single labeled cell. 5 Conclusion In summary, motivated by parallel studies of tissue turnover in interfollicular epidermis, we have shown that the range of clonal fate data is consistent with a simple model dynamics in which a single population of epidermal progenitors give rise to differentiated cells that stratify and become shed at the surface of tissue. In this model, EP cell division results in balanced stochastic fate where, on average, following division, one cell remains in cycle and the other commits to differentiation. The model is fully-defined by just four adjustable parameters: the 34 EP cell division rate , the ratio of symmetric to asymmetric cell divisions controlled by the parameter r, the stratification rate , and the loss rate of cells from tissue µ. Indeed, the latter parameter is constrained and therefore fixed by the ratio of basal to suprabasal cells, which is easily and accurately measured. In the long-term scaling regime, the dynamics is fully-specified by a single parameter reflecting the ratio of the remaining three adjustable parameters. By adopting a Bayesian approach, we have shown that this simple model provides an excel- lent agreement with the range of complex clonal fate data (covering both basal and suprabasal cells) engaging more than 140 independent data points (Figs 3B, S6B, C and D). Moreover, we have shown that this model can capture the effect of atRA treatment through a straightforward adjustment of the average EP cell division rate. The predictive power of the model has also been tested both through the clone survival curve (Fig. 2B) as well as a second schedule of atRA treatment. 35 A-43 APPENDIX 2: PRE-PUBLISHED ARTICLE Roshan A, Jones PH & Greenman CD. (2013) An exact time-independent approach to clone size distributions in normal and mutated cells. arXiv:1311.5769 [q-bio.QM] 2 A-44 An Exact, Time-Independent Approach to Clone Size Distributions in Normal and Mutated Cells Roshan A1, Jones PH1, and Greenman CD§2 1 MRC Cancer Cell Unit, Hutchison-MRC Research Centre, Cambridge, CB2 2XZ, UK 2School of Computing Sciences, University of East Anglia, Norwich, NR4 7TJ, UK Abstract Biological tools such as genetic lineage tracing, 3D confocal microscopy and next gen- eration DNA sequencing are providing new ways to quantify the distribution of clones of normal and mutated cells. Population-wide clone size distributions in vivo are complicated by multiple cell types, and overlapping birth and death processes. This has led to the in- creased need for mathematically informed models to understand their biological significance. Standard approaches usually require knowledge of clonal age. We show that modelling on clone size independent of time is an alternative method that o↵ers certain analytical advan- tages; it can help parameterize these models, and obtain distributions for counts of mutated or proliferating cells, for example. When applied to a general birth-death process common in epithelial progenitors this takes the form of a gambler’s ruin problem, the solution of which relates to counting Motzkin lattice paths. Applying this approach to mutational pro- cesses, an alternative, exact, formulation of the classic Luria-Delbru¨ck problem emerges. This approach can be extended beyond neutral models of mutant clonal evolution, and also describe some distributions relating to sub-clones within a tumour. The approaches above are generally applicable to any Markovian branching process where the dynamics of di↵erent ‘coloured’ daughter branches are of interest. Key Words: Clone Size Distribution; Dyck Paths; Motzkin Triangle; Luria-Delbru¨ck; Mathematical Modeling 1 Introduction One approach to understanding the cellular hierarchy in multicellular organized tissue has been tracking the fate of individual cells either labeled in vivo or isolated ex vivo [1]-[6]. Improved techniques including genetic lineage tracing and 3D imaging by confocal microscopy have helped further investigate this basic area of research [7]–[9]. Typically, a cell type of interest is labeled with an identifier and the distribution of its progeny at later time points are observed. Clone distribution data can then be used to decipher division dynamics across the population of cells with great resolution. However, the current methods use population averaging, and are time- dependent posing analytical challenges. There is a need for alternative statistical approaches that may be complementary. §Corresponding Author 1 arX iv: 13 11 .57 69 v1 [q -bi o.Q M ] 2 2 N ov 20 13 B At cell division C a’ b’ c’ a b c a = c a’ > c’ D Mutagen Mutagen ȝ1 ȝ0 Dividing cell Differentiated cell Mutant cell Krt14 EdU 0/2 EdU+ 1/2 EdU+ 2/2 EdU+A ȝ1 ȝ0 Figure 1: Colony formation in normal and mutated cells. (A) Immunofluorescence images of 2-cell clones with the keratinocyte marker Keratin14, and the proliferation marker EdU, showing three possible outcomes of division: two non-proliferating daughters (0/2 EdU+), a non-proliferating and a proliferating daughter (1/2 EdU+), or two proliferating daughters (2/2 EdU+). Scale bar 50µ. (B) Cell division is a birth-death process with three possible outcomes based on the proliferative ability of its daughters. As above, a dividing cell (P) may divide to two dividing daughters (PP), a dividing and di↵erentiated daughter (PD), or two di↵erentiated daughters (DD) in proportions a, b and c respectively. In tissues with high turnover, the number of new dividing cells is equal to the number of non-dividing cells (a = c). (C) In the presence of mutagens like UV radiation, this process is imbalanced in p53 mutant clones in favour of proliferation (a0 > c0). This gives a survival advantage to mutant clones. (D) Mutant cell formation itself is a birth process that can follow one of three possibilities. The first is cell division independent and can occur with background exposure. The second and third possibilities occur following cell division, producing one or two mutant cells out of two daughter cells with probability µ1 = 1 µ0. Adult mammalian epithelium has a high rate of cell division during steady state. Despite this rapid rate of proliferation, the tissue remains in homeostasis as new cells are being generated at the same rate as loss of di↵erentiated cells in a birth-death process (a = c in Figure 1B). A sim- ple illustration of this is in the inter-follicular epidermis, where cell division occurs in the basal layer of a multi-layered epithelium. Cell division here can produce proliferating daughters, that remain in the basal layer, or non-dividing daughters, which are shed to the supra-basal layers, and eventually lost in a process of di↵erentiation. When these keratinocytes are grown in culture, a typical cell division can result in two dividing daughters, one dividing daughter or no divid- ing daughter out of two total daughters as seen through the uptake of the proliferation marker 5-ethynyl-2’-deoxyuridine (EdU, Figure 1A). Genetic lineage tracing in basal keratinocytes has allowed conditional expression of fluorescent proteins, with all subsequent daughter cells retain- ing the label, and thus being highlighted as a clone. Clone size distributions thus observed shows maintenance through a population asymmetry of fate outcome in dividing progenitors, with re- serve stem cells contributing to wound healing [1],[3],[10]. Additionally, this balance is disturbed in chronic UV irradiation, where p53 mutant keratinocyte clones gain a survival advantage over non-mutant clones mediated through increased proportions of proliferative daughters [11] (a > c in Figure 1C). The recent technical advance of live-imaging in epithelia may provide additional information to these models, such as the distribution of cell cycle times [12]. There is also an increasing body of work investigating the growth dynamics of pre-neoplastic and neoplastic tissue [13]–[16]. A growing colony of cells can be modeled as a branching process. 2 Luria and Delbru¨ck were the first to produce an analytical examination of the distribution of the number of mutant cells in growing bacterial colonies [17]. They used this to show that mutations arise randomly rather than in response to the environment. Their argument was partly deterministic and Lea and Coulson [18] and Bartlett [19], [20] derived approaches with greater stochastic rigour. These methods generally consider the problem of how many mutants are present after a fixed amount of time. An unpublished combinatorial method by Haldane also exists [21] where all cells divide simultaneously. These distributions generally assign genes the binary status of mutated or non-mutated. They do not consider the number of mutations in a gene, or the number of di↵erent combinations of mutations a subclone of cells may contain. Modern sequencing techniques mean greater resolution of mutations is now possible and there is increased interest in considering distributions associated with combinations of mutations [22]. As Kendall observed [23], [24] there are broadly three models for mutation formulation (Fig- ure 1D). The first formulation would indicate a single cell converts to mutated status at any time independent of the cell division process. This may be the case for continuous exposure to mutagens, such as UV light [25]. The second formulation is the most common formulation where mutations occur in one of the two daughter cells during the cell division process. This is likely to be the case for many mutational processes, where nucleotide errors occur on one of the two DNA strands [26]. DNA repair machinery then erroneously corrects this during checkpoints in the cell cycle, resulting in one mutant daughter cell. The third formulation assumes that both daughter cells are mutant. This is also a valid model, and is likely to arise when double stranded breaks occur. When double stranded repair incorrectly repairs the damage, rearrangements result and both daughter cells will be mutant. Some processes such as breakage-fusion-bridge cycles will even result in two mutant daughter cells with distinct rearrangements [27], [28]. For analytical purposes in this paper, we will assume the most common second formulation. Additionally, we assume that a mutation does not increase the chance of cell loss through apoptosis. In this work, we consider a di↵erent statistical approach to clonal distributions. A standard technique to analyzing a branching process involving two classes of objects, such as mutant/non- mutant, or progenitor/di↵erentiated, is to write down a Chapman-Kolomogorov equation for Pm,n(t); the probability of having m and n cells of the two types, at time t, and obtain a solution [29]. Instead, we determine the distribution of the number of di↵erent types of cells that are present when a fixed number of cells have accumulated, rather than the time that has passed. With this approach, we will see that treating cell di↵erentiation or mutation as time-independent results in exact analytic forms for the distributions of interest. In the next section we obtain the distribution for the number of dividing cells in an epithelial population. We then obtain distributions for the number of mutant cells in a clone undergoing a pure birth process. 2 Distribution of Colony Sizes in Homeostatic Tissue Tissue homeostasis is balanced by two types of cells; progenitor (dividing) cells (P ), and di↵er- entiated (non-dividing) cells (D). As progenitor cells (P) divide, they produce two daughter cells which may be either a progenitor cell or a di↵erentiated cell (D) resulting in the combinations (PP), (PD) or (DD). We assume the probabilities of these occurring are a, b, and c, respectively represented in Figure 1B. Across a population, these probabilities are assumed to be constant, holding the same values for any cell division that takes place at steady state. There is the possibility that apoptosis may form an additional component of this process. Whilst one could incorporate this as an additional branch in the process of Figure 1A, it is assumed negligible in the following analysis. 3 For simplicity, we assume that we start with a single dividing cell. We also assume the number of descendant cells can be observed, but that (P) and (D) cells can not be distinguished. There are two problems we would like to consider. Firstly, if we trace the lineage of a single cell, we wish to determine the distribution of the number of progenitor (P) cells present. Secondly, the physical similarity between (P) and (D) cells without any protein markers make the parameters a, b and c dicult to directly measure. Thus, we would like a method to estimate them. Time Frequency a a b c t1 t5t4t3t2 tn 1 6 5 4 3 2 0 0 A No. Cell Divisions Frequency 1 5432 n 1 6 5 4 3 2 0 0 B Total No. CellsNo. Proliferating Cells c t6 a 6 Proliferating Cell Differentiated Cell Figure 2: A branching process of di↵erentiated and pro- liferating cells. A single dividing cell is followed in time with the height of the solid line indicating total number of cells, and the height of the dashed line indicating num- ber of dividing cells. In (A), plotted against time, we see the rate of cell division is dependent upon the number of proliferating cells. In (B), plotted against number of cell divisions, we see the number of proliferating cells only depends upon the nature and number of cell divisions, not their timing. Now, our approach is based on the size of the clone (rather than time passed). Now, with each cell division, irrespective of out- come, the colony size n increases by 1 forming a clone of n+1 cells. If the cell division results in two progenitor daughters (PP), the number of dividing cells k increases to k+1. If the cell division results in a progenitor cell and a dif- ferentiated cell (PD), the number of dividing cells k stays the same. The production of two di↵erentiated daughters (DD) results in a loss of dividing cells to k 1. We can thus model the number of P cells as a discrete random walk that can move up, remain flat, or move down with probabilities a, b and c, where we have one forward step to take at every cell division as in Figure 2A,B. Note that if the colony becomes fully di↵erentiated, k = 0, we have no dividing cells and our process stops. We note that the timing of these divisions does not relate to the count of proliferating cells. In Figure 2A we see the time dependent process, with a division rate that will be pro- portional to the number of proliferating cells. In Figure 2B we see the same information in- dexed by the number of cell divisions; the tim- ing is not important. Such a problem is closely related to count- ing Motzkin lattice paths [30]. Lattice paths are paths connecting positions with integer co- ordinates and can take a variety of forms [31], [32]. In particular, Motzkin paths start from the origin (0, 0) on a 2-d integer lattice and allow movement with an up (1, 1) step, a flat (1, 0) step, or a down (1,1) step such that we never move below the horizontal axis. There are several path counting techniques for such conditions [30], [33], [34], which have also seen applications to paths similar to the ones we describe [35] [36]. These have been studied for a range of combinatorial problems [37], including some problems with weighted edges [38]. These paths can be utilised to represent our problem. The position (n, k) corresponds to the total number of cells, n, and the number of dividing cells, k, respectively. The PP, PD or DD divisions correspond to the up, flat and down steps, respectively. There are three di↵erences to Motzkin paths to note. Firstly, we start with one (P) cell, represented by position (1, 1). Secondly, we stop if we touch the horizontal axis, because no dividing (P) cells remain (k = 0). Lastly, we have probabilities a, b and c associated with each step. Now, we would like to find the probability Pn,k of finding k dividing cells in a clone of size n. This probability then corresponds 4 b c a b c a b c a b c a b c a b c a b c a b c a b c a b c a b c a b c a b c a b c a b c a N um be r o f d iv id in g pr og en ito rs , k 4 3 2 1 5 Number of total cells, n 432 5 A B N um be r o f d iv id in g pr og en ito rs , k Number of total cells, n 3 3 3 5 5 5 Proliferating progenitor cell Non-proliferating differentiated cell C 1 5432 1 5 4 3 2 0 0 6 6 D 1 5432 1 5 4 3 2 0 0 6 6 E 1 5432 1 5 4 3 2 0 0 6 6 Figure 3: Cell proliferation as a combinatorial branching process with predictable paths. (A) Cell division in progenitor cells is a branching process with three possible outcomes (PP, PD or DD, See Figure 1). The expansion of a single cell to form a clone of cells is thus a combinatorial process, where any outcome of total clone size n and proliferating cells within it k occurs along fixed paths of a Motzkin like triangle. Clones that reach the horizontal axis have only non-dividing cells, and therefore do not progress further. (B) Example showing the 9 paths that a single proliferating cell can take to reach a clone of n = 5 and k = 2. The first three routes have three a divisions and one c division, while the remaining six routes involve two each of a and b divisions (cumulative probability = 3a3c + 6a2b2). (C) is a Dyck path, which moves up and down and not below the horizontal axis. (D) is a Motzkin path, which also includes horizontal moves. (E) is a gamblers ruin problem, which starts from height 1 rather than the origin, representing the formation of a fully di↵erentiated clone from a single dividing cell. to a weighted sum of Motzkin paths from (1, 1) to (n, k), where Motzkin paths in this context do not touch the horizontal axis. 2.1 Motzkin Paths Describe the Entire Distribution of Colony Sizes We have the following distribution for the number of progenitor (P) cells in a colony. Theorem 2.1. If we seed a single dividing cell, then the probability of having k(> 1) dividing cells when the colony is of size n is given by: Pn,k = bnk2 cP i=0 n1 k+2i1 ( k+2i1 i k+2i1i1 )ak+i1bnk2ici Proof. We start with Dyck paths; paths from (0, 0) to (0, 2n) that do not go below the horizontal axis involving steps of type up, (1, 1) or down, (1,1), such as portrayed in Figure 3C. The number of such paths is known to be counted by the Catalan numbers Cn = 1 n+1 2n n [39]. A Dyck triangle is the collection of paths from (0, 0) to (n, k) that do not go below the horizontal 5 axis and involve up and down steps. Note that n and k must have the same parity. If Dn,k count these paths then conditioning over one step we find Dn,k = Dn1,k1 + Dn1,k+1. It is straightforward to show by substitution that Dn,k = k+1 n+1 n+1 1 2 (nk) satisfies this recurrence, along with boundary condition D2n,0 = Cn. This formula di↵ers to other counts involving Dyck triangles because this lattice formulation of the triangle is rotated through ⇡4 to the usual presentation [40]. We now turn to Motzkin paths, which are the same as Dyck paths except we now allow an additional horizontal step (1, 0). Now any Motzkin path from (0, 0) to (n, k) can be partitioned into a Dyck path from (0, 0) to (k + 2i, k) involving k + i up steps and i down steps, along with n k 2i horizontal steps, where i 2 0, 1, ..., bnk2 c. For any i, the probability of such a path arising is ak+ibnk2ici. Then noting that we have n k+2i permutations of the horizontal steps with the Dyck path steps, we sum across the possibilities to get the following probability. mn,k = bnk2 cP i=0 Dk+2i,k n k+2i ak+ibnk2ici = bnk2 cP i=0 n k+2i ( k+2i i k+2ii1 )ak+ibnk2ici Finally we note that we are going from position (1, 1) to (n, k) without touching the horizontal axis, so substituting n! n 1 and k ! k 1 gives the required result; Pn,k = mn1,k1. This result allows us to look at the case where all n cells in the colony are fully di↵erentiated (all are (D) cells) and there is not further potential for growth. In our Motzkin triangle analogy, this would be a Motzkin path (with an additional final down step) from (1, 1) to (n, 0), such as the path in Figure 3E. All colonies that have a corresponding path touching the horizontal axis thus have no proliferating cells. We have an absorbing barrier, also known as the gambler’s ruin problem. Corollary 2.1. The probability Pn,0 is given by weighted Motzkin numbers: Pn,0 = bn22 cP i=0 n2 2i ( 2i i 2ii1)aibn22ici+1 = bn22 cP i=0 n2 2i C2iaibn22ici+1 Proof. For the case where there are no dividing cells remaining in the colony, the colony must transit through a penultimate stage (n1, 1) with only one dividing cell remaining, and undergo an enforced final (DD) division. Multiplying the formula for Pn1,1 by c gives the required result. Both of these results have corresponding generating functions as described in the following result: Theorem 2.2. The generating function F (x, t) = 1P n=0 nP k=0 Pn,kxktn is given by: F (x, t) = 1bt p (bt1)24act2 2a + x(2tax1+bt+ p ((bt1)24act2)) 2a(xtctbxtax2) Proof. First we construct a weighted generating function for paths in a standard Motzkin triangle, m(x, t) = 1P n=0 nP k=0 mn,kxktn, where mn,k are the Motzkin numbers weighted by the elements a, b and c associated with each path from (0, 0) to (n, k). Now conditioning over a single step gives the following recurrence; mn+1,k = cmn,k+1 + bmn,k + amn,k1. Then substituting this into the generating function yields the following: m(x, t) = 1 + 1P n=1 nP k=0 mn,kxktn = 1 + 1P n0=0 n0+1P k=0 mn0+1,kxktn 0+1 = 1 + 1P n0=0 n0+1P k=0 (cmn0,k+1 + bmn0,k + amn0,k1)xktn 0+1 = 1 + tcx (mm(0, t)) + tbm+ taxm 6 Which gives us: m(x, t) = xtcm(0,t)xtctbxtax2 . To find m(0, t) we note that a Motzkin path from (0, 0) to (n, 0) involves either a first horizontal step and a weighted Motzkin path one step smaller, or an up step, a motzkin path, a down step, and a Motzkin path. This is summarised as mn+1 = bmn+ ac n1P k=0 mkmn1k, where mn is the weighted sum of these paths. Now substituting this recurrence into the generating function m(0, t) = 1P k=0 mktk = 1+ t 1P k=0 mk+1tk yields m(0, t) = 1+ btm(0, t)+ t2acm(0, t)2. The solution satisfying m(0, 0) = 1 is then: m(0, t) = 1bt p (bt1)24act2 2act2 Substituting this above then yields the general form: m(x, t) = 2tax1+bt+ p (bt1)24act2 2at(xtctbctax2) The result is obtained by noting that the generating function for Pn,k corresponds to paths from (1, 1) to (n, k). Furthermore, a path from (1, 1) to (n, 0) involves a weighted Motzkin path of length n 2, followed by a down step, and we find that: F (x, t) = 1P n=0 Pn,0tn + P n,k1 Pn,kxktn = t2c 1P n=0 mn,0tn + xt P n,k0 mn,kxktn = t2cm(0, t) + xtm(x, t). Substituting the weighted Motzkin generating functions results in the desired form. 2.2 Gambler’s Ruin We are now in a position to describe the probability of ruin, or equivalently the probability of a fully di↵erentiated clone, where we have the following result: Corollary 2.2. The generating function G(t) = 1P n=0 Pn,0tn is given by: G(t) = 1bt p (bt1)24act2 2a This results in an alternative expression for the probability Pn,0 that a clone of size n is fully di↵erentiated: Pn,0 = ( 12 )n 2a Pn r=0 2(nr) nr 2r r (b+2pac)nr(b2pac)r (2(nr)1)(2r1) Furthermore, we find that the probability P0 that a single proliferating cell will become fully di↵erentiated is given by: P0 = ⇢ 1 a > c a c a < c Proof. To get the generating function G(t), we simply substitute x = 0 into F (x, t) from Theorem 2.2. To get the alternative expression for the probabilities Pn,0 note that we can write G(t) as: G(t) = 12a [1 bt (1 (b+ 2 p ac)t) 1 2 (1 (b 2pac)t) 12 ] A double binomial expansion gives us: G(t) = 12a [1 bt P1 j=0 P1 k=0 2j j 2k k ( b+2pac2 )j( b2pac2 )k (2j1)(2k1) t j+k] The constant and linear terms cancel and a reordering of the summation to collect powers of t leaves us with the required expression. Lastly we note that G(1) = 1P n=0 Pn,0 and so substituting t = 1 into the generating function gives us: G(1) = 12a (1 b p (b 1)2 4ac = 12a (a+ c |a c|) 7 where we have used 1 b = a+ c. Separately considering the cases a > c and a < c gives the required results. 2.3 Estimating Di↵erentiation Probabilities Time Frequency ȝ1 ȝ0 ȝ1 ȝ1 t1 t5t4t3t2 tn 1 6 5 4 3 2 0 0 A No. Divisions Frequency 1 5432 n 1 6 5 4 3 2 0 0 B Total No. Cells No. Mutant Cells Non-Mutant Cell Mutant Cell Figure 4: A branching process of non-mutated and mu- tated cells. A single dividing cell is followed in time with the height of the solid line indicating total number of cells, and the height of the dashed line indicating number of mutant cells. In (A), plotted against time, we see the rate of cell division is proportional to the total number of cells, resulting in exponential growth. In (B), plotted against the number of divisions, we see the number of mutant cells only depends upon the number of mutant cell divisions, not their timing. We are now in a position to estimate the probabilities a, b and c of getting the dif- ferent daughter cell combinations of (PP), (PD) or (DD), even when (P) and (D) cells are visually indistinguishable. Clone size dis- tributions in a range of homeostatic epithe- lia demonstrate that dividing progenitor cells have (PP) outcomes in similar proportions to (DD) outcomes (or a=c) [11]. Colonies aris- ing from such populations will eventually be- come fully di↵erentiated and stop growing, as represented in the bottom row of Figure 3A. Therefore, at late time points of obser- vation all colonies of cells with few cell num- bers will be formed exclusively of non-dividing cells, as any colonies with dividing cells will continue to expand in cell number. Thus repeated measurements of small clone sizes, nc, of fully di↵erentiated non-dividing colonies of size n can readily be counted. We can then compare these counts to the probabili- ties {c, bc, c(b2 + ac), ...} = {Pn,0}n of either Corollary 2.2 or 2.1, and hence determine a, b and c using maximum likelihood. Small clone sizes form the bulk of clones seen in popula- tion distributions, and therefore can provide robust quantifiable results. It is also important to highlight, that this is not a↵ected by the presence of additional cell populations which have a branching birth process alone (putative stem cell populations). The clones formed by such populations will be much larger, continuing to expand with time, so can be readily identified and excluded. 2.4 Stochastic Processes Approach Finally, we remark that a lot of the derivations using Motzkin paths can also be replaced with approaches from stochastic processes. We highlight this with an alternative derivation of the gambler’s ruin generating function of Corollary 2.2 in the Appendix. 8 3 An Exact Luria Delbru¨ck Distribution We now investigate the mutation process of a growing clone of cells. Here, we assume no death process is involved, and initially that the mutation provides no additional survival advantage. In all that follows k = m+ n is the number of cells, where m and n count the number of mutants and non-mutants, respectively. Again we start with a single dividing cell. An example of this can be seen in Figure 4A. The cells are dividing randomly at a rate according to the following Markovian branching (Yule-Furry) process. When any non-mutant cell divides we assume a mutant cell arises with probability µ1, such as the first division of Figure 4A at time t1. Conversely, we may obtain two non-mutants with probability µ0 = 1 µ1, such as in the second division portayed at time t2. Finally, any dividing mutant produces two mutant daughters with probability 1, as displayed at times t3 and t4. We ignore any back mutation or loss of mutation. As the colony grows, the rate of division, k, increases in proportion to the number of cells present, k. If tk is the time of the kth division, the mean time intervals tk+1 tk correspondingly decrease as we get exponential growth. Note that at time tk the colony increases in size (by one cell) to k+1 cells. It is this single dividing cell that has the opportunity to e↵ect the number of mutations at this point; this is independent of the either the time tk at which this takes place, or the time tk+1 tk between divisions. In Figure 4B we see the mutation process as a discrete process on the number of divisions that have taken place. We assume for the moment that mutant and non-mutant cells divide at the same rate in a Markovian manner. All cells are thus equally likely to divide at any point in time. If we have n non-mutant cells and m mutant cells, we then find that a mutant will divide with probability mm+n resulting in m + 1 mutants and m + n + 1 cells. Conversely, a non-mutant divides with probability nm+n resulting in m + n + 1 cells. This non-mutant will mutate with probability µ1 resulting in m+ 1 mutants, otherwise we will still have m mutants, with probability µ0. This observation leads to the following correspondence. Theorem 3.1. If p(k)m denotes the probability of having m mutant cells present when the popu- lation is of size k, we have the following recurrence, which is initialized with p(1)0 = 1. p(k)m = (m1k1 + km k1 µ1)p (k1) m1 + ( k1m k1 µ0)p (k1) m Proof. This result is a statement of conditional probability. There are two ways we can obtain m mutants amongst k cells, depending upon the mutation status of the k 1 cells prior to the previous cell division. If we have m 1 mutant cells out of k 1 cells in total, then to obtain m mutants in k cells we either select a mutant to divide with probability m1k1 , or pick a non-mutant to divide and generate a new mutation with probability kmk1 µ1. Alternatively, if we already have m mutants from k 1 cells, then we require that the next division be a non-mutant cell that doesn’t mutate on division, with probability k1mk1 µ0. Note that we have reduced the mutation process to a discrete heterogeneous Markovian random walk starting from (1, 0) where we have either a horizontal step (1, 0) with probability km k µ0, or the step (1, 1) with probability m k + km k µ1. We have the following general form for the k division distribution of mutants, p(k)m . Theorem 3.2. p(k)m = µ k1 0 P {1i1 Tn) = R1 0 R tm 0 1 mm emmtm 1mme mmtmdtndtm = mm mm+nn = ⇢m⇢m+n Thus we just have to weight the mutant count by the relative increase in division rate. In particular, if we have m mutant cells and n1 non-mutant cells, the probability that we have m mutants and n non-mutants after the next cell division requires a non-mutant to divide without a new mutation forming. This occurs with probability n1n1+⇢mµ0. Similarly, if we have m 1 mutant cells and n non-mutant cells, the probability that we have m mutants and n non-mutants after the next cell division requires a mutant to divide, or a non-mutant to divide with a new mutation forming. This occurs with probability ⇢(m1)⇢(m1)+n + n n+⇢(m1)µ1. The recurrence is a statement of conditional probability connecting these two observations. The recurrence can be used to derive formulae for p(k)m and the moments. The method is identical to that of Theorem 3.2 and the details are left to the reader. An example of the distribution can be seen in Figure 5C, where we have mutation rate µ1 = 0.05 and relative fitness ⇢ = 2. This gave a comparable distribution to Figure 5B, where the mutation rate is µ1 = 0.20 with neutral relative fitness ⇢ = 1, although the variance is notably higher in Figure 5C. 5 Distributions of Sub-Clones in Mutated Colonies In the questions considered so far, we just have the binary status of mutated or non-mutated. This is generally the status of a gene, or a portion of a chromosome that may be of interest, but could also be the status of a single nucleotide of DNA, which number in the billions. DNA sequencing techniques now mean that individual mutations can now be distinguished by their position in the genome. For example in Figure 6A we see that five of six cells are mutant, arising 11 AChromosomal Coordinate Number of cells 3 2 1 0 B Mutants } } } } Time Clone 1; 1 Mutation; 2 Cells Clone 2; 3 Mutations; 1 Cell Clone 3; 1 Mutation; 2 Cells Clone 4; 0 Mutation; 1 Cell † † † Figure 6: (A) A representation of clonal mutant growth; six cells result from five cell divisions, three of which produce four mutations(†), which cluster into four clones. (B) represents the cellular count for each mutation against illustrative chromosomal co-ordinates. from four mutations produced during three cell divisions (†), that combine into into four distinct clones. In Figure 6B we have the distribution of the number of cells for each mutation. This is a symbolic representation of the mutation and sequencing depth information obtained from modern experiments and points to other avenues of investigation. Firstly we would like to know the number of cells containing a randomly selected mutation. Secondly, we would like to know the number of clones. Thirdly, we would like to know the number of cells. Finally, we would like to know the number of distinct mutations in a randomly selected clone. We have the following results. 5.1 The Number of Cells Containing a Specific Mutation Theorem 5.1. If p(k)r is the probability that a randomly selected mutation exists in r cells in a colony of k cells, we have: p(k)r = kr+1P j=1 j1 (k1)2 r1Q m=1 (1 j2km1 ) This di↵ers slightly to the original problem considered by Luria and Delbru¨ck in that instead of asking how many cells contain a mutation in a specific gene (or region), which may involve many di↵erent mutation events, we randomly sample a mutation from all mutations found in that region, and count the corresponding number of cells containing that mutation. We assume each mutation arises only once, which may not be true for large colonies or small genomes. Proof. Now there are k 1 divisions that take place to give a sample size of k. Now if we randomly select a mutation it can arise during any of these divisions with equal probability. We let q(j,k)r denote the probability that if a mutation forms when there are j cells, it is present in r cells when the cell population is k j. Then if p(k)r is the probability a randomly selected mutation is in r cells when the population is of size k, we have: p(k)r = 1k1 kr+1P j=1 q(j,k)r If the mutation arises when the population has size j, then this mutation may be present in any of 1 to k j +1 cells, depending on whether the cells containing the mutation divide. Thus j  k r + 1. Furthermore, following a population size of k 1, we either had r 1 copies of the mutation and the next cell division duplicates a copy, or we have r mutant cells, and the dividing cell does not contain the mutation of interest. This gives us the recurrence: 12 q(j,k)r = q (j,k1) r (1 rk1 ) + q(j,k1)r1 ( r1k1 ) Now if we start with the initial value p(j,j)1 = 1, so that initially one of j cells carries the mutation, then we can show by substitution that this recurrence and initial condition is satisfied by the following expression: q(j,k)r = (j 1) (kj)r1(k1)r where (a)b = a(a 1) . . . (a (b 1)) is the Pochhammer symbol. Substituting into the expression above then gives: p(k)r = kr+1P j=1 j1 k1 (kj)r1 (k1)r This is equivalent to the expression in the theorem. 5.2 Distribution of the Number of Clones The second problem requires the distribution of the number of clones. Every time a new mutation occurs, it will occur in a single cell that belongs to some clone already present. That cell will divide into two daughters, one of which will contain the new mutation. That cell will have a new combination of mutations and a new clone is born. We thus trivially observe that the number of clones is always one more than the number of cells divisions that produce new mutations. Now, mutations can arise during a cell division. For a colony of size k we have k 1 independent cell divisions in total, each of which may generate new mutations with probability µ1. We thus find that: Theorem 5.2. If C represents the number of colonies, we find that for a colony of size k, C 1 has Binomial distribution Bin(k 1, µ1). ⇤ 5.3 Size Distribution of Mutant Clones The third question concerns the size of the clones. For example, in Figure 6A we note that clone 2 was formed in the 3rd cell division, and contains a single cell. The associated distribution for the size of a random clone is described in the following result. Theorem 5.3. If p(k)n represents the probability a randomly selected clone from a population of size k contains n cells, and p(i,k)n is the corresponding probability for a clone formed in the ith cell division, then p(k)n = 1k1 knP i=1 p(i,k)n where: p(i,k)n = P {i