TACTICS for Bioimaging Informatics and Analysis of T Cells Raz Shimoni

TACTICS for Bioimaging Informatics and Analysis of T Cells

Raz Shimoni

Submitted in total fulfilment of the requirements of the degree of Doctor of Philosophy

Centre for Micro-Photonics (CMP)

Faculty of Science, Engineering and Technology

SWINBURNE UNIVERSITY OF TECHNOLOGY 2014

"Absence of evidence is not evidence of absence"

Dedicated to my princess Liliana and to my lovely wife Olga

You are my sunshine and my hope

iii

Abstract

T cells are a highly specialized type of white blood cell, with a critical role in cell-mediated immunity. The normal function of T cells is imperative for an efficient immune response, providing adaptable defence against new foreign bodies. Unfortunately, in some cases this system fails, leaving our body exposed to disease. Immunologists are now trying to dissect how T cells participate in the immune response, which might lead to the development of new drugs and treatments. One of the latest approaches to studying T cells is to utilize fluorescence microscopy imaging of live cells over time. Using this technique it is possible, for instance, to monitor how the selective recruitment of molecules to polarized regions within the cells can affect functionality and fate. Despite the rapid developments in fluorescence microscopy imaging and the wide range of available analytical software, the analysis of time-lapse data is not yet perfected.

The development of novel strategies to aid the interpretation of information collected by fluorescence time-lapse imaging, and how they can be used to improve our analysis of live cell imaging, are the main subjects of this thesis.

Herein, a modular and adaptable high-throughput toolbox named TACTICS has been developed. The TACTICS pipeline is first described, offering features not seen in other standard Bioimaging informatics software. Taking advantage of TACTICS as a computational platform, several challenges in the analysis of fluorescence microscopy data are addressed. In particular, the dependency of fluorescence measurements on experimental setup, the difficulty of automatically processing large data sets, and the effects of imaging settings on the accuracy of the analysis were explored. Next, these new tools were employed to measure protein polarization in T cells during migration and cell-cell interactions. Finally, multi- parametric analysis, integrating image cytometry and interactive lineage informatics analysis, was demonstrated to be extremely useful to study the fate of T cells. Together these advances have been used for the analysis of Asymmetric Cell Division (ACD) and polarity studies, providing an innovative toolbox to help elucidate key insights in the biology of T cells.

Acknowledgements

This work was conducted under the direct supervision of Prof. Sarah Russell, the head of the Immune Signalling Laboratory at Peter MacCallum Cancer Centre. Sarah, thank you for selecting me to carry out this project, thank you for your infinite patient, for the time you spent with me to look together into raw data, and for many ideas and suggestions to improve the analysis.

It has been a great honour for me to be a part of the Centre for micro- Photonics (CMP), supervised by Prof. Min Gu, the director of the CMP. Thank you Min for your support and for encouragement, it means a great deal for me.

I would like to express my sincere appreciation to my previous advisor, Dr. Zeev Bomzon, who initially suggested for the usability of image cytometry software dedicated to study polarity from time-lapse imaging. I would like to thank Dr. Daniel Day, who served as advisor in my first year of candidature, for our intriguing intellectual discussions, and for his valuable microfabrication training. Special thanks to my mentor Dr. Pavel Lobachevsky who represented me at the student committees, in addition to his research advices and support. I also would like to thank all members of my PhD committee, in particular the head of committee Dr. Ilia Voskoboinik and education officer Dr. Caroline Owens.

The interaction between programming and biology has been vital to the success of establishment new tools throughout this project. For exchanging ideas and opportunities, setting requirements, and provided their data for software development, I wish to thank my collaborators: Mohammed Yassin, Dr. Kim Pham, Dr. Jane Oliaro, Kelly Ramsbottom, Mandy Ludford-Menting, Dr. Edwin Hawkins and Dr. Kerrie-Ann McMahon. I have been benefited to work with you and to learn more about the biology aspect of your research, and enjoyed your friendly company at the lab. Additionally, I wish to thank TACTICS beta-testers Amelia Poetter, Emily O’Connor and Adam Poetter for helping to identify bugs and useful suggestions. Your work was essential to improve and validate the quality of the software and analyzed data. It was awesome to work with the next generation of scientists. I would like to the thank Dr. Sarah Ellis and Cameron Nowell for microscopy training. I would like to the thank Mandy Ludford-Menting for her PC2 and tissue culture training. I wish to thank Pierrette Michaux for clean room training. Additionally, I

vi wish to thank Ricardas Buividas and Dr. David MacDonald for their assistance with the laser setup for microfabrication experiments. I wish to thank all members of the CMP for their support and intriguing environment.

The contribution of the MATLAB open-code community has been imperative to the development of TACTICS. I wish to thank many developers who contributed their source code or gave free advice. I hope that my work will inspired others as you inspired me.

I would like to thank my supervisor Prof. Sarah Russell for her kind help with editing and correcting the chapters of my thesis. Selected sections were edited and corrected with additional assistance from Dr. Olga Shimoni, Dr. Jane Oliaro, Dr. Simon Partridge, and Cameron Nowell, Mohammed Yassin, and Mandy Ludford- Menting.

I am deeply grateful for the generous financial support I received for the last 3.5 years through Swinburne University Postgraduate Research Award (SUPRA). I also thank my Supervisor Dr. Sarah Russell who generously chose to further top up this scholarship. Further modest student budgets were kindly received from Faculty of Engineering and Industrial Science (FEIS) and Peter MacCallum Cancer Centre education department.

The submission of this thesis closes a long chapter in my life. First of all, I wish to thank my father, who always kept me in the frontier of science many years ago: my first computer Atari XL, life science books and kits, and even a first simple light microscope to explore new worlds beyond our sight. I would like to thank all my family and friends in Australia and in Israel, who supported and encouraged me to achieve my dream. In particularly, this thesis is dedicated to my beautiful daughter Liliana, who was born during the first year of my candidature. Special thanks to the wonderful staff of Goodstart Early Learning Centre in Altona, who provided me the security that my daughter is in their devoted care, allowing me spending the time required for this research. Last but not least, I dedicate this thesis to my wise wife Olga, who is a scientist herself and gave many useful advices during my research. Olga, thank you for your support, and that you always cheer me up. I have been truly privileged to have such a family.

vii

Declaration

This is to certify that:

1. The thesis contains no material which has been accepted for the award of any other degree or diploma, except where due reference is made.

2. To my best knowledge this thesis contains no material previously published or written by another person except where due reference is made in the text of the examinable outcome.

3. Where the work is based on joint research or publications, I disclose the relative contributions of the respective workers or authors.

Raz Shimoni

viii

Preface

In Chapter 3, the development of TACTICS version 2.0 was done mainly in collaboration with Dr Kim Pham, that her project required the development of TACTICS-ACD and TACTICS-Polarization Modules. The source code of TACTICS version 2.2 was published in co-authorship with Dr Kim Pham in [1]. The development of TACTICS version 3.0 was done mainly in collaboration with Mohammed Yassin who provided data images of CD8+ T cells and that his project required the development of TACTICS-Lineage and further improvements in the TACTICS-Tracking Modules. Particularly, the interactiveness of the TACTICS- Lineage Module, and novel ideas in the analysis of lineage trees reconstructions, gating, improved tools to the TACTICS-Tracking Module for interactive user corrections, and the revolutionary idea of selective tracking operator. Manual corrections were performed by Amelia Poetter, Emily O’Connor, and Mohammed Yassin. In Chapter 4, MLA cell line was provided by Mandy Ludford-Menting-. Mohammed Yassin assisted in establishing the time-lapse microscopy setup and validation of microwells. In addition, the development of the ACD module and new approaches to investigate ACD were done in collaboration with Dr Kim Pham, who investigated the hypothesis of the existence of asymmetric cell division in immune cells. In Chapter 5, time-lapse microscopy images of MLA and thymocytes undergoing migration and division were provided by Dr Kim Pham, who also conducted all supporting biological assays. Plasmid constructs of eGFP-Numb and the mutant variation eGFP-Numb-2a were fused by Mandy Menting-Ludford, using plasmid generated by C.J. McGlade. Microfabricated microwells were supplied by Daniel Day (Microsurfaces Pty Ltd). Cameron Nowell supplied the lif2tif journal to convert .lif files to categorized .tif files. Pavel Lubachevski supplied splitting algorithm in C code. Adam Poetter assisted me with beta-testing, manual corrections of segmentation, and tracking. My innovation of the major-minor normalization approach would be impossible without statistical assistance from Prof Terry Speed, who initially suggested looking on polarization ratios across other angles of polarity, and is based on previous normalization method formed by Dr Zeev Bomzon. This work has been published in co-authorship with Dr Kim Pham in [2].

Abbreviations

ACD: Asymmetric Cell Division AITP: Algorithmic Information Theoretic Prediction APC: Antigen Presenting Cell AOBS: Acoustic Optical Beam Splitter AOTF: Acoustic Optical Tuneable Filter APD: Avalanche photodiode aPKC: atypical Protein Kinase C (protein) CD: Cluster of Differentiation D: Dimensional DC: Dendritic Cell Dlg: Discs large protein (protein) DIC: Differential Interference Contrast DMEM: Dulbecco’s Modified Essential Medium DN: Double-negative DP: Double-positive eGFP: enhanced Green Fluorescent Protein EM-CCD: Electron Multiplying - Charge Coupled Device FACS: Fluorescence Activated Cell Sorting FFT: Fast Fourier Transform GFP: Green Fluorescent Protein GUI: Graphical User Interface HCA: High Content Analysis hrs: hours IPT: Image Processing Toolbox ITK: Insight Segmentation and Registration Toolkit Lgl: Lethal giant larvae (protein) LAT: Linker for activation of T cells (protein) LSC: Laser Scanning Cytometery MAT: Multitemporal Association Tracking MHC: Major Histocompatibility Complex min: minutes

x ms: miliseconds (1/1000 of a second) MSCV: Murine stem cell virus MSD: Mean Square Displacement MTOC: Microtubule orienting centre NaN: Not a Number nm: nanometer Par: Partitioning defective protein (protein) PDMS: Polydimethylsiloxane PIL: Python Image Processing PKC: Protein kinase C (protein) PMA: Phorbol 12-myristate 13-acetate PMT: Photomultiplier tube PSF: Point Spread Function RAM: Random-access memory RI: Refractive Index ROC: Receiver Operating Characteristic SCD: Symmetric Cell Division SNR: signal-to-noise ratio SP: single-positive t: time, index frame from series TCR: T cell receptors TIF: Tagged Image File TP: True Positive VTK: Visualization Toolkit WT: Wild Type

Table of Contents

Abstract ...... iv

Declaration ...... viii

Preface ...... ix

Abbreviations ...... x

Table of Contents ...... xii

Table of Figures ...... xvii

1. Chapter 1 ...... 1

1.1. Introduction ...... 2

1.2. Aims of thesis ...... 3

1.3. Thesis outline ...... 3

2. Chapter 2 ...... 6

2.1. Introduction ...... 7

2.2. The biology context and outstanding questions ...... 7

2.2.1. T cells: maturation, activation, expansion, and contraction ...... 7 2.2.2. Polarity is central in the biology of T cells ...... 9 2.2.3. ACD in T cells ...... 11

2.3. Principles in optical microscopy for single-cell studies ...... 13

2.3.1. Fluorescence microscopy ...... 13 2.3.2. Confocal Laser Scanning Microscopy ...... 14 2.3.3. Differential Interference Contrast Microscopy ...... 15 2.3.4. SP5 Multispectral Confocal Microscopy ...... 16 2.3.5. Spinning disk confocal laser microscopy ...... 17 2.3.6. Dimensions in microscopy imaging ...... 18

2.4. Microscopic imaging of immune cells ...... 19

xii

2.4.1. Live cell imaging as a tool to dissect functionality in T cells ...... 19 2.4.2. Fluorescence time-lapse imaging of T cells: advantages, challenges and solutions ...... 21 2.4.3. Micro-technologies for in vitro cell culture ...... 22

2.5. Bioimaging informatics and computational analysis ...... 23

2.5.1. Cell segmentation ...... 24 2.5.2. Tracking cells from time-lapse data ...... 25 2.5.3. Graphic visualization ...... 26 2.5.4. Lineage informatics ...... 27 2.5.5. High Content Analysis (HCA) ...... 28 2.5.6. Environment and resources for bioimaging software ...... 29 2.5.7. Open-code and commercial software ...... 30 2.5.8. Development of MATLAB based bioimaging applications ...... 31 3. Chapter 3 ...... 36

3.1. Introduction ...... 37

3.2. Methods ...... 39

3.2.1. Experimental methodology ...... 39 3.2.2. Imaging setup ...... 41 3.2.3. Image processing ...... 42

3.3. TACTICS pipeline: methods and features ...... 42

3.3.1. TACTICS Segmentation Module ...... 43 3.3.2. TACTICS Tracking Module ...... 45 3.3.2.1. Association (Figure 3.2.vi) ...... 46 3.3.2.2. Annotations (Figure 3.2.vii) ...... 50 3.3.2.3. Linking (Figure 3.2.viii) ...... 50 3.3.2.4. Manual corrections (Figure 3.2.ix) ...... 52 3.3.3. TACTICS Measurements Module ...... 53 3.3.4. End-point analysis (Figure 3.2. xi-xii) ...... 53

3.4. Results and discussion ...... 54

3.4.1. Exploratory tools for lineage informatics ...... 54

xiii

3.4.2. New tool for mapping the life histories of CD8+ T cells ...... 59

3.5. Summary ...... 63

4. Chapter 4 ...... 66

4.1. Introduction ...... 67

4.2. Methods ...... 69

4.2.1. Cell microwells and cell culture ...... 69 4.2.2. Time-lapse microscopy...... 69 4.2.3. Computational platform ...... 69 4.2.4. Processing of time-lapse data ...... 70 4.2.5. Calculation of Polarization Ratio (PR) ...... 70 4.2.6. Calculation of binarization ...... 71 4.2.7. Shape model of cell division ...... 72 4.2.8. Generation of synthetic images ...... 74

4.3. Results and discussion ...... 75

4.3.1. Polarization ratios are sensitive to image processing settings ...... 75 4.3.2. Simulations indicate that the degree of clustering of fluorescence alters the effect of thresholding ...... 79 4.3.3. Calculation of the polarity across the minor axis provides a useful control to assess noise and to normalize calculations of asymmetry ...... 81 4.3.4. Comparing PRmajor and PRminor demonstrates the effects of clustered fluorescence...... 84 4.3.5. Assessing the value of binarization in the analysis of polarity...... 84 4.3.6. Methods to incorporate PRminor in the assessment of polarity ...... 87 4.3.7. Sensitivity test from simulations ...... 90 4.3.8. Strategy for polarity analysis ...... 92 4.3.9. Description and utilization of the TACTICS ACD Module ...... 96

4.4. Summary ...... 98

5. Chapter 5 ...... 100

5.1. Introduction ...... 101

xiv

5.2. Methods ...... 103

5.2.1. Data input for TACTICS ...... 103 5.2.2. Constructs ...... 104 5.2.3. Imaging conditions ...... 104 5.2.4. Importation of file images to TACTICS ...... 105 5.2.5. TACTICS Pipeline for polarity measurements ...... 105 5.2.5.1. Cell segmentation (Figure 5.1.A.i-iii) ...... 105 5.2.5.2. Separation of touching cells (Figure 5.1.A.iv-vi) ...... 106 5.2.5.3. Geometrical descriptors for classifying function ...... 108 5.2.5.4. Cell tracking (Figure 5.1.B ) ...... 109 5.2.5.5. Creation of cell libraries (Figure 5.1.C) ...... 110 5.2.5.6. Spectral unmixing (Figure 5.1.D) ...... 110 5.2.5.7. 2-D image reconstruction and cell alignment (Figure 5.1.E.i) ...... 111 5.2.5.8. MTOC identification (Figure 5.1.E.ii) ...... 112 5.2.5.9. Quality control and manual inspection ...... 112 5.2.5.10. Measurements of polarity during migration ...... 113 5.2.5.11. Measurements of polarity during division ...... 113 5.2.6. Statistics and display ...... 114

5.3. Results and discussion ...... 114

5.3.1. Description of the TACTICS Polarization Module ...... 114 5.3.2. Polarization measurements of Numb using the major-minor axis ...... 117 5.3.3. Gating on the minor axis provides an objective comparison of polarization ...... 121 5.3.4. Measurements of polarization by alternative axis of polarity ...... 123 5.3.5. Measurements of polarization in migrating DN3 thymocytes ...... 125 5.3.6. Utilization of TACTICS to measure polarity in dividing DN3 Thymocytes ...... 130

5.4. Summary ...... 133

6. Chapter 6 ...... 134

6.1. Thesis outline ...... 135

6.2. Conclusions ...... 136

6.3. Future Work ...... 138

References ...... 140

xvi

Table of Figures

Figure 2.1. Schematic representation of the PDZ network protein distribution in three different systems...... 11

Figure 2.2. Asymmetric cell division (ACD) of T cells...... 12

Figure 2.3. Schematic representation of spinning disk confocal laser setup...... 18

Figure 3.1 Description of the experimental protocol for lineage study...... 41

Figure 3.2. Schematic illustration of TACTICS pipeline for lineage reconstruction...... 44

Figure 3.3. Screenshot of the Tracking Module...... 46

Figure 3.4. Illustration of the selective operator in the Tracking Module...... 48

Figure 3.5. Annotation procedures in the Tracking Module...... 49

Figure 3.6. Graphic visualization of lineage trees...... 55

Figure 3.7. Screenshot of the Lineage Module...... 56

Figure 3.8. Combining gating on lineage and scatter dot basis...... 58

Figure 3.9. Measurements of lifespan over several generations...... 60

Figure 3.10. Measurements of area over several generations...... 61

Figure 3.11. Lineage of activated founder cell...... 62

Figure 4.1. Axial subdivision for polarization analysis...... 71

Figure 4.2. Binary image generation of cell division...... 74

Figure 4.3. Simulations of cytokinesis...... 77

Figure 4.4. Ratio coefficients from experimental data are dependent on the threshold value...... 79

Figure 4.5. Ratio coefficients from synthetic data are dependent on the threshold value...... 80

Figure 4.6. PRmajor and PRminor are differentially affected by thresholding and clustering...... 82

Figure 4.7. Intensities histograms corresponding to the Figure 4.6...... 83

Figure 4.8.The effect of cluster number on the accuracy of PR measurements...... 85

xvii

Figure 4.9. Gating based on PRminor in simulated data...... 89

Figure 4.10. Sensitivity test based on simulations ...... 91

Figure 4.11. 100 cell division were simulated in increasing ratios from 1 to 1.5, giving 1100 divisions in total as represented in "jet" colors ...... 92

Figure 4.12. Suggested workflow for optimal analysis of ACD ...... 93

Figure 4.13. Screen shot of TACTICS ACD Module...... 97

Figure 5.1. Methods applied to process and structure data input for the Polarization Module ...... 107

Figure 5.2. Interactive analysis with TACTICS Polarization Module ...... 116

Figure 5.3. TACTICS analysis pipeline to quantify eGFP-Numb polarization in T cells ...... 118

Figure 5.4. Analysis of eGFP-Numb polarization in migrating T cells ...... 120

Figure 5.5. Utilization of TACTICS to compare between Numb WT and mutant in gated subpopulation based on major-minor plot ...... 122

Figure 5.6. Absolute PR as a function of cell intensity ...... 122

Figure 5.7. Comparison of fluorescent intensity between the base and tip of the uropod ...... 124

Figure 5.8. Numb and Numb2A are both polarized to the uropod in DN3 thymocytes...... 127

Figure 5.9. Gated fast migrating and elongated DN3 thymocytes on experiment base ...... 129

Figure 5.10. Exemplification of TACTICS to measure ACD in T cells...... 132

xviii

Table of Equations

Equation 1: ...... 25

Equation 2: ...... 70

Equation 3: ...... 71

Equation 4: ...... 72

Equation 5: ...... 72

Equation 6: ...... 73

Equation 7: ...... 73

Equation 8: ...... 75

Equation 9: ...... 110

xix

Introduction

1. Chapter 1

Raz Shimoni Chapter 1 Introduction

1.1. Introduction

T lymphocytes (also termed T cells), are highly specialized cells with a crucial role in the adaptive immune system within humans [3] and other vertebrates [4]. Across the body, T cells function as living sensors, searching continuously for signals that can alert to potential infection and eliminate foreign bodies [5].

The last decade has witnessed a great advance in our knowledge and understanding the multifaceted biology of T cells, yet many fundamental aspects in the biology of T cells are still unknown. One of the latest emerging questions is how the polarized recruitment of molecules during cell division can implicate the fate of the two daughter cells [6]. These and other questions require a deeper understanding of the molecular mechanisms that control cell polarity and fate determination in T cells [7]. One of the latest approaches to study T cell biology is by time-lapse fluorescence microscopy[8], which provides non-invasive quantitative measurements of protein localization over time[9]. Using time-lapse imaging, immunologists are trying to elucidate whether molecular mechanisms that control polarity in other type of cells, such as epithelial and stem cells[10], and other organisms such as Drosophila [11], are also conserved in T cells. These mechanisms might be involved in the development of leukaemia[12], similar to mechanisms observed in other cells[13].

Data analysis of imaged T cells is considered to be challenging due to the dynamic and elusive nature of these cells. In particular, the following aspects demand new computational and analytical tools[14]. Firstly, monitoring the fate of cells for tracking of T cells over several generations; secondly, localization measurements of fluorescence proteins require an objective approach for the analysis of polarity; and finally, the investigation of polarity in T cells during migration and division demands analytical tools that comprise accurate tracking and methods for polarization measurements. The development of a usable toolbox that contains those computational requirements, and the ability to adapt new assays, is the main subject of this thesis.

Raz Shimoni Chapter 1 Introduction

1.2. Aims of thesis

In order to realize a new toolset to analyze time-lapse microscopy images of T cells, three main objectives must be met; 1) to develop a versatile toolbox for cell tracking and quantification of cellular characteristics, 2) to utilize TACTICS to develop new strategies for image informatics, and 3) to integrate and utilize these methods to provide new insight into the biology of T cells in the context of fate determination and polarity.

1.3. Thesis outline

The thesis is structured as follows:

Chapter 2 reviews literature relevant to this thesis, divided into four fields - the biological context of this thesis, microscopy imaging of immune cells, the principles of optical microscopy, and the field of bioimaging informatics

Chapter 3 describes the development of TACTICS Toolbox as a versatile toolbox for High Content Analysis (HCA) that is generic yet multi-featured. Insight into different approaches used in TACTICS, including a description of different modules, is provided. Additionally, the utilization of the TACTICS for lineage informatics is described. In particularly, I describe novel methodologies to elucidate the initial parameters that can influence fate determination of T cells.

Chapter 4 describes a specialized polarization analysis that is based on intensity ratiometric measurements. I developed a method that provides more accurate quantification and statistical analysis of polarity by using a normalization procedure with internal ratios. To support this approach, computer simulations of fluorescently-tagged molecules are used and the ratio of symmetry under changeable settings is measured. Moreover, a new TACTICS module is developed that is dedicated to these approaches, and its utilization to analyze dividing cells is demonstrated.

Chapter 5 describes the utilization of TACTICS to extract robust measures of polarity and dissect the localization of wild type and mutant forms of Numb, an endocytic adaptor protein that exhibits intracellular polarity in many cell types. Additionally, the TACTICS Polarization Module, developed to deal with several

Raz Shimoni Chapter 1 Introduction problems in the quantification of polarized migrating cells, is introduced. This includes different approaches to measure polarization in high-throughput. Finally, I describe the utilization of TACTICS to investigate polarity during T cell division.

Chapter 6 summarizes this thesis and suggests future directions for this work.

Raz Shimoni Chapter 1 Introduction

Literature Review

2. Chapter 2

Raz Shimoni Chapter 2 Literature Review

2.1. Introduction

This chapter consists of four main topics: Section 2.2 introduces the biological themes relevant to this thesis. This includes an explanation of the different stages during T cell development and explains the importance of polarity and asymmetric cell division (ACD) in T cells. Section 2.3 explains the principles of optical microscopy techniques for single-cell studies and describes the fluorescence microscopes utilized throughout the thesis. Section 2.4 explains the principals of microscopy imaging, the current challenges with imaging immune cells and introduces microfabrication technology to address these issues. Section 2.5 provides an overview of the computational approaches to the analysis of the microscopy data, and reviews advance in the field of bioimaging informatics and HCA.

2.2. The biology context and outstanding questions

2.2.1. T cells: maturation, activation, expansion, and contraction

During T cell maturation, hematopoietic stem cells from the bone marrow migrate to the thymus where they mature into naïve T cells [15]. The exact process of T cell maturation is not fully understood, but the major steps during this process are known [16] and described as follows. T cell progenitors express no CD4, CD8, or T cell receptor (TCR), and are known as double-negative (DN) thymocytes (CD4- CD8-TCR-). The Cluster of Differentiation (CD) proteins are a group of glycoproteins expressed on the cellular surface, and are useful markers for phenotyping and distinguishing between different types of T cells (and other leukocytes). DN thymocytes progress through their development during four stages, and eventually become double-positive (DP) thymocytes (CD4+CD8+TCR+). This transition involves a series of differentiation steps that results in the generation of a large repertoire of antigen-specific T cells. The progression through the differentiation stages occurs in different locations within the thymus, and is mediated by stage-specific expression of chemokines receptors [17]. The DN3 stage is a pivotal point during the differentiation of thymocytes in the thymus, where several fate decisions are made following release from the pre-TCR checkpoint [18, 19]. The DN3 stage is further discussed in Chapter 4 of this thesis.

Raz Shimoni Chapter 2 Literature Review

Each T cell that completes the DN process receives a unique form of TCR that recognizes a specific antigen. This process involves DNA recombination, and can potentially create harmful autoreactive T cells that can attack ‘self’ antigens. Therefore, only thymocytes with low affinity to “self” antigen (‘negative selection’) and that express useful TCR (‘positive selection’) are induced for proliferation. Approximately 98% of thymocytes fail the selection process. The other 2% of thymocytes mature to immunocompetent single-positive (SP) T cells, which are either CD4+ (CD4+CD8-TCR+) or CD8+ (CD4-CD8+TCR+), and are released from the thymus into the peripheral tissues. Peripheral T cells are produced early in life and then can remain quiescent in the periphery for years, until they are activated and become effector cells. The activation of naive T cells is triggered when "foreign" antigens presented on the Major Histocompatibility Complex (MHC) of antigen presenting cells interact with the TCR expressed on T cells. One of the most abundant types of antigen presenting cells (APCs) is the Dendritic Cell (DC). When DCs recognize foreign antigens, they digest the protein, and present the peptides through their MHC. The DC then migrate to secondary lymphoid tissues such as the lymph nodes and spleen, where the DC interact with, and activate the naive T cells present that are specific for that antigen [20]. The molecular communication between the TCR on the T cell and MHC-antigen complex on the DC is called an immunological synapse and facilitates activation of the naïve T cells [21-24]. The interaction between T cells and APCs requires several hours, and can even be maintained through to cytokinesis [25].

Following activation, CD8+ naïve T cells differentiate into cytotoxic or effector T cells (usually referred to as cytotoxic T lymphocytes or CTLs). CTLs constantly seek to destroy cells infected with pathogens (particularly with viruses), tumour cells, or any cells that express atypical “self” antigen due to damage or infection by secretion of cytotoxic molecules such as perforin and Granzyme B [5]. In response to activation by antigen presentation, naïve CD8+ T cells enter into a rapid state of proliferation through repetitive cycles of mitotic divisions [26, 27]. Through 7 -20 rounds of mitotic divisions, approximately half a million descendants are generated. Most of the activated T cells become effector T cells, and destroy cells expressing their cognate antigen to eliminate infection and cancer, while others

Raz Shimoni Chapter 2 Literature Review become memory T cells, sustaining a small population that is ready to proliferate in response to future infections from their cognate antigen [28, 29]. A fundamental difference between memory and effector cells is that effector cells are destined to die after the infection is cleared, and usually live several weeks, but the memory cells can remain quiescent and live for years or decades, and then respond to a second exposure to the pathogen [30]. Within two weeks of an immune response, 95% of the cells die (the ‘contraction phase’). The surviving memory precursors remain alive for several years, providing a faster secondary response [5]. Identifying which cells survive the contraction and how naïve T cells are programmed to produce memory or effector cells presents a great challenge in immunology [31]. Furthermore, it is well known that a naïve CD8+ T cell can generate both memory and effector cells [32, 33], but it is not clear yet whether memory and effector cells evolve independently, or whether one lineage evolves from the other. The possibility that bifurcation occurs after the first cell division is supported by findings in our laboratory (the Immune Signalling Laboratory at Peter MacCallum Cancer Centre) and others, that the naïve CD8+ T cell divides asymmetrically, differentially distributing fate-determining molecules between the two daughters in a process known as Asymmetric Cell Division (ACD)[34-37]. However, before focusing on the different aspects of ACD in T cells, it is essential to build our understanding around the mechanism that controls T cell polarity.

2.2.2. Polarity is central in the biology of T cells

Cell polarity is defined as spatial differences in cellular structure and asymmetric distribution of various macromolecules, including proteins, membrane compartments, organelles, and RNA transcripts, at the two opposite sides of the cells across a defined axis [10, 38]. Polarity controls diverse biological activities such as migration and morphology, in T cells and other types of cells [7, 10, 12, 13, 38-56]. A well-known example is polarity in epithelial cells. Each epithelial cell is subdivided into an apical domain that faces the lumenal surface of the tissue and basolateral domain that contacts neighbouring cells and the underlying stroma [3].

Several years ago our laboratory identified that polarity in T cells is orchestrated by a network of proteins that contain PDZ motifs, hence termed the

Raz Shimoni Chapter 2 Literature Review

PDZ network [40](Figure 2.1). The PDZ network includes the Scribble, Crumbs and Par complexes [7, 23, 41-43] in polarized cells of solid tissues such as epithelial (Figure 2.1.A) and neuronal cells [40, 41, 44-48]. The members of these three complexes are referred to as ‘polarity proteins’. Members of the polarity protein family include the Scribble complex, which comprises Scribble, Discs large (Dlg), and Lethal giant larvae (Lgl); the Par complex, which comprises atypical protein kinase C (aPKC), Par3 and Par6 [52]; and the Crumbs complex, which comprises Crumbs, Patj and Pals [57]. Polarity proteins interact with molecules that are vital to coordinate cell shape and function such as Actin, microtubules, Myosin, Rho GTPases, ERM proteins, vesicles and cell surface receptor. Furthermore, polarity proteins are associated with development of various cancers [13, 49-51], suggesting their importance in immunology and in the development of diseases [12].

Polarity proteins participate in a broad range of cellular processes. For example, several members of the polarity protein group were shown to polarize preferentially to the proximal daughter cell during T cell ACD [34] (Figure 2.1.B). In addition, morphology of migrating T cells is characterized by the recruitment of several polarity proteins to the leading edge of the cell, and to the uropod (the trailing edge) at the rear [54, 58](Figure 2.1.C). The uropod allows cell-cell interactions, which tend to create large clusters [54]. The regulation of T cell shape by polarity proteins exemplifies the importance of polarity in T cell functionality, and will be further discussed in Chapter 4.

Raz Shimoni Chapter 2 Literature Review

Figure 2.1. Schematic representation of the PDZ network protein distributioon in three different systems. Polarity proteins participate in a broad range of cellular processes that involves polarity such as: (A) Apical-basolateral domains in epithelial cell as a classical example of cell polarity. (B) Immunological synapse formation during interaction between T cells with an antigen presenting cell. (C) The formation of the front–rear axis in migrating T cells. Colored lines correspond to the localization of each protein. Reprinted from Ludford-Menting and Oliaro et al., 2005 [40].

2.2.3. ACD in T cells

A facet of cell polarity that controls cell fate determination is ACD, a mechanism by which a dividing cell produces two daughter cells with different molecular composition, leading to the adoption of a different cellular fatte [55, 59]. This concept is demonstrated in Figure 2.2.A. The mechanisms that directt ACD can be predominantly classified as extrinsic or intrinsic [56]. The extrinsic mechanism refers to external cues that regulate distinct fates in cells that are initially identical. These cues can arise from the microenvironment and surrounding cells. The intrinsic mechanism includes the asymmetric segregation of fate determinants beetween two daughter cells, which induces diverse fates.

ACD was first identified twenty years ago, where asymmetry was observed in the disttribution of the Par (partitioning defective) protein in the C. elegans zygote (one-cell embryo) [60] (Figure 2.2.B). Another good example of intrinssic ACD is the asymmetric segregation of the fate regulator Numb between the two daughter

Raz Shimoni Chapter 2 Literature Review cells during cytokinesis of neuroblasts in the Drosophila larval brain [61]. The difference in the amount of Numb expressed in the two daughteerrs is translated into alternative differentiation pathways in the two daughter cells by inhibiting signal transduction through the Notch/Delta pathway [61].

Subsequently, ACD has been reported in a large variety of organisms and cell types. Many aspects of ACD are not fully understood, but a pictture of the role and mechanism of ACD is gradually emerging. It is believed that ACD is utilized in mutli-cellular organisms to alteer the fate of cells including stem cells, skin cancer cells and immune cells [12, 555, 62]. In addition, it has been suggested that ACD might play an important role in stem cells, where one daughter can remain a stem cell (and undergo self-renewal), while the second daughter cell can differentiate into a specialized cell type [62].

Figure 2.2. Asymmetric cell division (ACD) of T cells. (A) The mitotic spindle is aligned along the same axis as the fate determinants, which are recruited towards only one of the poles, thus directing two daughter cells into different fates (B) ACD in tthe C. elegans zygote is a striking example of the unequaal distribution of PAR-3 (red) and PAR-2 (green) during ACD. Scale bar: 5 μm. Images courtesy of Dorian Anderson (New York University School of Medicine). (C) ACD in T cells at 3 different stages of division during CD8+ T cell activation by dendritic cells. Scale bar: 10 μm. Reprinted from Oliaro ett al., 2010 [34].

As explained in Section 2.2.3, T cells express the mammalian homologues of the fate proteins involved in ACD, analogous to other systemms where ACD was established and confirmed, in simple multicellular organisms, such as Drosophila and C. elegans [12, 41, 63].

Raz Shimoni Chapter 2 Literature Review

2.3. Principles in optical microscopy for single-cell studies

The utilization of fluorescence microscopy to magnify, and to provide richness of data at the single-cell level, has been widely accepted as a standard tool in biology and the biomedical sciences [64-66]. Initially, fluorescence labelling for analysing the localization of proteins was predominantly based on antibody- conjugated fluorophores for immunohistochemistry. A significant milestone in the field of fluorescence microscopy was the discovery of the Green Fluorescent Protein (GFP) 20 years ago [67]. This discovery has revolutionized many imaging-based techniques, enabling non-invasive observation of protein localization inside living cells, and providing information about potential roles of the respective gene fused to the fluorescent protein [68]. Over the years, GFP-based variants with different spectral properties have been developed, and together with the latest organic fluorescent dyes, quantum dots and fluorescent nanodiamonds [69, 70], constitute a "fluorescence toolbox" for assessing protein localization and functionality [71]. Owing to the exceptional properties of fluorescence markers such as high spectral specificity and the ability to non-invasively tag molecules within both living and fixed cells, fluorescence techniques have emerged as a promising platform for a variety of imaging and sensing applications. Spectral specificity allows sensitive detection of multiple fluorescently labeled samples known as multispectral microscopy imaging, which was fundamental to the analysis performed in Chapters 4-6. The principles and applications of fluorescence microscopy are introduced in the next sections.

2.3.1. Fluorescence microscopy

The fundamentals behind any fluorescent-based technique are the illumination of the specimen with specific excitation wavelengths, and analysis of the emitted light from the specimen. Materials emitting fluorescence in response to incoming excitation are known as fluorophores. When a photon strikes a fluorophore molecule, there is a statistical probability that the electronic ground state of the molecule will be excited to a higher energy state. Since the high-energy state is unstable, it spontaneously returns to the ground state, followed by fluorescence emission of the energy that was absorbed by the molecule. This description is known

Raz Shimoni Chapter 2 Literature Review as the Jablonski diagram. The energy lost due to vibrations and heat transfer during the fluorescence process results in emission wavelengths longer than the excitation wavelengths. This shift in a fluorescence spectrum is known as the Stokes-shift. The Stokes-shift of a specific molecule is characteristic of that molecule, and is associated with the molecular structure of the molecule and its conformation [72, 73]. Fluorescence microscopy relies on the principle of Stokes-shift that allows spectral separation and the distinction between the emitted fluorescence and the excitation light source. Basic fluorescence microscopy setups such as wide-field microscopy requires a light source (e.g. xenon arc lamp, mercury-vapour lamp, or Light-emitting diode (LED)[74]) and (generally) an excitation filter, exploiting the property of fluorophores to emit fluorescence at specific spectral bands. A dichroic mirror can be used to reflect only light shorter than a specific wavelength, allowing the light from the main source to pass through the objective to the sample. The longer-wavelength emitted light can be directed towards an emission filter, which allow only specific wavelengths to be received by the detector [75, 76]. Extensions of conventional fluorescence microscopy techniques including wide-field deconvolution [77], multiphoton [78], confocal scanning laser microscopy [79] [80] and spinning disc confocal laser microscopy [81], all technologies that offer improved resolution [82], dimensions [51] or acquisition rate [83].

2.3.2. Confocal Laser Scanning Microscopy

Fluorescence confocal microscopy can provide non-invasive measurements of fluorescently labelled molecules within cells, with the advantage of depth selectivity. The principle of confocal microscopy was invented in the 1950s by Marvin Minsky [79, 80]. This microscope was based on light source that is focused into a small point, which is known as the focal point (also termed the focal volume), which is directed on the specimen, and only light within this volume is collected and translated to signal. Light outside the focal volume is rejected by a pinhole aperture, resulting sharper signal. Following Minsky's work, a later type of fluorescence microscope was developed in the late 60s by M. David Egger and Mojmir Petran, who implemented the principle of Nipgow disk [81] (see Section 2.3.5 below). In the early 70s the first point scanning system was introduced by Davidovits and Egger [84], and is the basis for modern confocal microscopes that are utilized these days.

Raz Shimoni Chapter 2 Literature Review

The principle is explained as follows. A laser beam is focused into a focal volume and scans a specimen to reconstruct a fluorescence image. The movement of the laser beam in relation to the specimen in the direction (x,y,z) is manipulated by moving mirrors or the stage. For instance, in the raster scanning mode, the laser beam scans the sample with parallel lines to cover a whole area in relation to the specimen. Fluorescence collected within the focal volume is confocal with the pinhole and consequently passes through the pinhole aperture. Hence, unlike conventional microscopy such as wide-field, most of the out-of-focus light is blocked, and high signal-to-noise ratio is achieved. Each photon from the emission that passes through the pinhole strikes a sensitive detector, usually a photomultiplier tube (PMT) or avalanche photodiode (APD), resulting in a statistical probability to produce a single photoelectron. These detectors rely on electrons emitted from material when they absorb energy from light. Notably, for his revolutionary work on this phenomenon, the so-called photoelectric effect, Einstein was awarded the Nobel Prize in 1921 [84]. The electronic signal is amplified by charge multiplication, and the current is then converted into an analogue electrical signal.

As the excitation beam scans the specimen, variations in the signal intensity illuminating the detector aperture correspond to variations in emission. The exact location of the focused laser beam is programmed and synchronized by a computer. The computer translates the electronic signals received from the detector into intensity discernments to reconstruct a 2-D grey level image, in which the intensity of each pixel is proportional to the fluorophore concentration. The exclusion of background caused by out-of-focus light enables sharp images without a blur resulting from background fluorescence [64]. Furthermore, it provides the capability to optically scan thick specimens by sequential optical slices, which can be reconstructed into a ‘z-stack’ that represents the 3-D fluorescence of the specimen [85].

2.3.3. Differential Interference Contrast Microscopy

Differential interference contrast (DIC), also known as Nomarski microscopy [86], is an optical microscopy illumination technique to enhance contrast of imaged transparent specimens by exploiting refractive index gradients. A polarized light

Raz Shimoni Chapter 2 Literature Review source is transmitted through the Nomarski-modified Wollaston prism and is split into two orthogonally polarized rays. The two rays are focused by the condenser to the specimen at two adjacent points in the sample. Each of the two rays travels through different optical path, whereas the ray that passes through material with higher refractive index (RI) or larger thickness moves slower. Thus, a phase shift is formed. The rays are focused through the objective lens through a second prism, which recombines the rays. Destructive and constructive interference (superposition) of the two displaced rays converts the phase shift into amplitude change of bright and dark shadows. Obtained images are characterized by a pseudo 3-D appearance, and with high contrast at the edges of biological structures. For this reason, DIC is commonly implemented for imaging unstained cells in media, and for facilitating automated algorithms for cell segmentation. DIC imaging was often applied throughout the research described in this thesis to image T cells, and is complimentary to imaging cells expressing fluorescent proteins such as GFP.

2.3.4. SP5 Multispectral Confocal Microscopy

The Leica TCS SP5 multispectral commercial confocal microscope (Leica Microsystems CMS GmbH, Germany) was utilized by Dr. Kim Pham to scan multiple sections in high resolution to record protein localization in motile and dividing T cells. It has a dedicated incubator to maintain optimal conditions as described in Section 2.2.2. The Leica SP5 has two novel components: An Acoustic Optical Tuneable Filter (AOTF) and an Acoustic Optical Beam Splitter (AOBS). The AOTF is an adjustable quartz crystal that is acoustic-optically modulated to select specific wavelengths of the excitation [87]. The AOBS is a novel beam splitter that was introduced in 2002 by Leica Microsystems Pty Ltd to replace the dichroic mirror and to improve the separation between the illumination and the detection paths. It works by exciting a piezoelectric crystal with a signal that is the sum of several specific frequencies, thereby enabling diffraction of several wavelengths simultaneously [88]. Thus, the AOBS allows simultaneous detection in multi- fluorescence channels, properties utilized for the acquisition of data images presented in Chapter 4.

Raz Shimoni Chapter 2 Literature Review

2.3.5. Spinning disk confocal laser microscopy

Spinning disk confocal laser microscopy is a technique that uses a rotating perforated disk to simultaneously scan a specimen with thousands of tiny points of light [89], and provides an alternative to standard confocal microscopy in term of acquisition time. Figure 2.2 illustrates how the specimen is imaged in a spinning disk setup. Collimated laser illumination passes through a lens disk consisting of an array of thousands of microlenses. The lens disk is aligned with a Nipkow disk, which contains thousands of pinholes. Each microlens is laterally coupled with a pinhole, and together, implement the confocal principle to reject out-of-focus light through the use of a pinhole. The two disks are connected to an electric motor, which drives the two disks at high speed. When the disks spin, an array of focused laser beams scan and excite fluorescent labels in the specimen. The fluorescence emission is collected through the pinhole. Only emission from the focal plane can pass through the pinholes, thus, scattered light is excluded. A dichroic mirror reflects the emission wavelengths, and separates the laser emission from any excitation light. Multiple narrow laser beams are collected and the Electron Multiplying - Charge Coupled Device (EM-CCD) camera reconstructs 2-D images. As a result, high signal-to-noise images can be acquired at a fast rate of 1000 frames per second [90]. Previously, a spinning disk microscope was used to allow fast imaging of dividing cells for the aim of lineage reconstruction [91]. Similarly, in this thesis, spinning disk microscopy was used to image T cells undergoing division, when faster image acquisition was required.

Raz Shimoni Chapter 2 Literature Review

Figure 2.3. Schematic representation of spinning disk confocal laser setup. Collimated Laser illumination passes through a two parallel and aligned disks. The top disk is actually glass plate containing thousands (~20k in latest Yokogawa systems) of Fresnel microlenses organized in a series of spirals. The bottom is Nipkow disk contains thousands of pinholes. Each microlens in the top plate is laterally coupled with a pinhole, and together, implement the confocal principle to reject out-of-focus light. A dichroic mirror reflects the emission collected from in approximately 1000 pinholes, and separates the laser emission from any excitation light. Multiple narrow laser beams are collected and the EM-CCD camera reconstructs 2-D images. As result, high signal-to-noise images can be acquired at fast rate. Figure was reprinted from [91].

2.3.6. Dimensions in microscopy imaging

Generally, fluorescence data from microscopy is saved as digital image files. Each image can be described as a 2-D discrete space with N rows and M columns. Each pixel at array a[m,n] has integer coordinates of [m,n] with [m=0,1,2,…,M–1] and [n=0,1,2,…,N–1][92]. The minimum pixel value is zero, and maximum pixel value is dependent on the number of bits (e.g. - 8 bits image pixel values range between 0 and 255). The value assigned to each pixel in this range is proportional to the intensity of the transformed light [75], and is usually measured in arbitrary units. Modern microscopes have the capacity for acquisition over higher dimensions. 3-D imaging consists of multiple focal planes (z stack), providing more representative view of subcellular structures and allows better quantification compared with 2-D. 4-

Raz Shimoni Chapter 2 Literature Review

D corresponds to 3-D imaging over time (m,n,z,t). Finally, 5-D corresponds to multichannel 3-D imaging over time (m,n,λ,z,t).

2.4. Microscopic imaging of immune cells

2.4.1. Live cell imaging as a tool to dissect functionality in T cells

Although in vivo (within living animals) and in situ (within organs) imaging are not the focus of this chapter, they play an important role in the study of immune cells, and are therefore discussed briefly below.

Multi-photon laser scanning microscopy has the capability to image deep in a tissue, [93] [94], thereby providing information such as polarization [95] and long- term fate tracking of T cells within the physical environment [96]. This capability has been exploited to image intact lymphoid organs and to study T cell activation by tracking the position of immune cells and measuring interactions with each other [97, 98]. In situ imaging by the Robey laboratory (Berkeley University) using two-photon microscopy has provided important information on the migratory behaviour of thymocytes within the thymic organ [97, 99-105].

T cell migration is crucial for the proper function of the immune system [106], as the recirculation of T cells between the blood and lymphoid organs and their migration into tissues following activation are all essential for an effective immune response [8]. The migration of T cells is directed by external cues such as chemokine gradients. By measuring different aspects of migration dynamics such as speed, acceleration, persistence, and directionality [94, 107,69], cellular dynamics can be correlated with functional outcomes such as maturation and activation. In some cases, control over the directionality of migration is subtle, and its detection compared to random migration requires sophisticated data analysis [108]. Two significant techniques utilized to distinguish between random and directed migration are Mean Square Displacement (MSD) plots and angle analysis. Directed migration results in a curved MSD and a mean angle that is lower than 90 degrees in 3-D cell migration [109]. For instance, angle analysis showed that naive CD8+ T cells move to sites where naive CD4+ T cells and dendritic cells interact [110]. Imaging thymocyte migration in the thymus revealed that DP thymocytes exhibit slow and

Raz Shimoni Chapter 2 Literature Review random walk migration in the outer cortex, whereas single-positive (SP) thymocytes exhibit directed cell migration at the central medulla [102]. This observation indicates that long range cues may be induced by positive selection. An example where in vivo imaging is extremely valuable is in the analysis of migration of T cells during DC-T cell interactions in the lymph node, which revealed that directed movement is induced in response to inflammatory chemokine production [110] [111].

Although the most physiologically relevant time-lapse imaging experiments are those performed in vivo, in many cases (such as high-throughput analysis of polarity and ACD that are presented in this thesis), in vitro imaging (imaging of cells in culture) is often more suitable, as explained below.

In vitro imaging of T cells allows the investigation of basic biological functions under simplified controlled conditions. In comparison to in vivo imaging, it does not require extraction of tissues for each experiment from laboratory animals, and allows for more consistent conditions and longer tracking. Since elucidating how one T cell can produce different subtypes requires imaging for long periods and physical isolation, in vitro imaging is preferable. In vitro imaging was utilized in Chapters 5 of this thesis to study if two daughter cells yield divergent fate through lineage tracing. Another significant advantage of in vitro imaging over imaging in vivo is the ability to collect large data sets in a high-throughput approach.

To prevent damage to the cells during in vitro imaging, conditions should be as similar as possible to the natural environment in the body. Moreover, fluorescence excitation at the time of imaging should be minimized as much as possible, as reactive oxygen species can lead to phototoxicity, cell damage, and mitotic arrest [112]. Notably, excitation of fluorophores with high illumination intensity increases the probability that fluorophores irreversibly lose their characteristic fluorescence, a phenomenon known as photobleaching [64]. During long experiments the number of photobleached molecules increases. This presents a limitation as the signal intensity is consequently reduced. Another consideration should be towards the selection of excitation. Since the energy delivered to the sample by longer excitation wavelengths (red and far red), it provides less stress on the cells compared to shorted UV lengths.

Raz Shimoni Chapter 2 Literature Review

A detailed guide for optimal conditions during time-lapse microscopy experiments was recently published by Coutu and Schroeder [113]. These conditions include molecular composition of media (i.e. supplements of antioxidants), sterility, and balanced pH. Typically, microscopes dedicated for the purpose of time-lapse imaging include an incubator that provides constant temperature (usually 37˚C), a constant concentration of CO2/O2 gas mixture, and sufficient humidity to avoid excessive evaporation of media culture.

2.4.2. Fluorescence time-lapse imaging of T cells: advantages, challenges and solutions

Techniques such as immunoblotting and biomolecular assays provide limited information and reflect a bulk population, while fluorescence imaging can potentially provide information down to the level of sub-cellular compartments. Fluorescence imaging can be classified as either static or dynamic. Static fluorescence imaging methods such as the use of high resolution microscopes to image fixed cells (e.g. by confocal or super-resolution microscopy) provides multiple images of a specific cell at a single time point. However, fixing cells can affect the proximity, angle and orientation of the imaged cells. Another emerging static method is the use of the automated Laser Scanning Cytometry (LSC) system [114] to capture single image from thousands of cells, but is limited to cells in suspension, and provides lower resolution in comparison to standard fluorescence microscopes. While static methods give only a snapshot of particular events, rather than capturing the stages in a dynamic process, time-lapse microscopy is based on interval observation of dynamic events by recording many images [115]. Long term imaging of the same cell is particularly useful to monitor spatiotemporal changes in cell activity and functionality; such as the regulation of cell death, mapping cellular signalling pathways, differentiation and more [116-118]. Moreover, time-lapse imaging allows us to link together records of the history of events such as cellular behaviour, interactions with other cells, functional activities, cellular properties, and external signals that the cell might receive upstream. This can be extremely valuable for the prediction of biological processes and cell fate [119]. Nevertheless, since there is a requirement to link data between many frames, the analysis of time-lapse data presents a higher degree of difficulty. Moreover, like most immune cells, T cells are

Raz Shimoni Chapter 2 Literature Review highly dynamic cells, and despite developments in imaging technology, the analysis of time-lapse experiments for these cells presents a significant challenge [120]. Obviously, images that are characterized by superior quality in terms of acquisition time, signal-to-noise ratio, and resolution, are more easily processed [121]. However in most imaging experiments, there is a trade-off between the optimal conditions for prolonged cell imaging and the quality of the images [118]. For instance, reducing exposure levels is important to avoid phototoxicity and to minimize photobleaching, but it also results in more crude images. Most immune cells are non-adherent and comparatively dynamic [122, 123]. T cells can crawl more rapidly than any other cell type in the body [8]. Naïve T cells migrate in uninfected lymph nodes at average 3-D speeds exceeding 18 μm/min [108]. Thus, by using conventional imaging methodology, T cells tend to move out of the field of view. Another problem arises from the fact that immune cells frequently interact with other cells [25, 124- 126].Therefore, when imaged over a long period of time (time scale of hours up to few days), it becomes increasingly difficult to efficiently track cells.

Currently, manual identification of cells is the most reliable technique [127], but it is also time-consuming and error-prone. Consequently, these challenges possess a substantial bottleneck and can give less consistent results. Additionally, many immunological events are rare, and therefore, the accurate and objective statistical analysis of such events requires a large number of repeats. In order to handle these challenges, two methods are presented in this thesis. Firstly, the development of analytical tools and automated processing pipelines. Secondly, the utilization of microfabricated wells that contain the cells within a field of view for the duration of the experiment can facilitate better conditions for in vitro imaging. The latter topic will be introduced in the next section.

2.4.3. Micro-technologies for in vitro cell culture

Microfabricated structures provide the means to contain cells being imaged in vitro within defined regions. Related to this thesis is the generation of microwells (a term utilized in Chapters 4-6). These microfabricated microwells were employed to prevent cells leaving/entering the field-of-view during imaging, enabling prolonged tracking of immune cells in our laboratory [2, 34, 40, 128, 129] and others [130]. It

Raz Shimoni Chapter 2 Literature Review provides the opportunity to image isolated cells in each chamber (for instance, one T cell and one antigen presenting cell, or one thymocyte and one stromal cell), which allows us to analyse large datasets.

Microfabricated devices are traditionally fabricated by soft lithography, which was originally adopted from the microelectronics industry [131]. Basically, soft lithography describes a collection of nanofabrication techniques for fabricating or replicating microstructures down to micro-, and even nanoscale resolution by using soft elastomeric materials. For example elastomeric stamps, replica molding, casting, and conformable photomasks [132] [133]}. The preparation of a master template can be achieved by using laser etching or UV-photolithography. UV- photolithography is based on the generation of desired microstructures using a photoresistant polymer (photo-curable material), such as SU-8. Exposure of the polymer to UV light through a photomask that contains printed features results in a polymerization of the exposed regions. The unexposed regions are dissolved and washed with developing reagent, leaving a positive structure [132]. Once the mold template is created, it can be used repeatedly to make many replicas of a secondary elastomeric material. Polydimethylsiloxane (PDMS) is one of most widespread elastomers used for soft lithography. It is a silicon-based organic polymer that is made up of repeating monomers (-SiO(CH3)2-). PDMS has some extraordinary physical properties: it is optically transparent above a wavelength of 230 nm, non- toxic, exhibits low expansion and shrinkage, and low surface energy that allows for easy release from the template [132]. In addition, PDMS is inexpensive and easy to polymerize. However, it is also important to note the disadvantages of PDMS, which are its sensitivity to deformation and structural collapse. In addition, because the PDMS is hydrophobic, in some cases fibronectin coating is necessary to allow attachment of adherent cells such as DCs to the surface [128].

2.5. Bioimaging informatics and computational analysis

Bioimaging informatics describes computational methodologies with which to process, produce, and analyse meaningful information from a set of images [134, 135]. These methodologies include: software for post-acquisition analysis, software for custom microscope setups [136], computational tools for image enhancement,

Raz Shimoni Chapter 2 Literature Review fluorescence measurements, cell tracking, measurements of cell morphology, image cytometry based cell-tracking, generic image processing functions and measurements, microscopy control, and more [137]. The next sections explain important aspects in bioimaging informatics and the requirement to extend the current limits of existing software.

2.5.1. Cell segmentation

Cell segmentation is one the most fundamental steps in processing imaged cells, usually applied to 2-D images. It describes the detection of pixels that represent part of a cell, and separation from its surrounding background. Many automated segmentation methods have been developed over the years with the aim to maximize the efficiency of detection with minimum computational effort. The quality of the segmentation depends on many parameters, such as types of images or microscopy used for the image acquisition. For instance, in bright field microscopy imaging the values of the pixels are inversely proportional to the optical density of the sample, which tend to be similar to the background pixels [138]. In contrast, fluorescence cells are represented by pixels brighter than the surrounding environment (the background), and automated segmentation of fluorescence microscopy is considered to be easier when the signal-to-noise ratio (SNR) is high.

An immense range of segmentation approaches have been developed over the years to distinguish cells based on fluorescence [139-141], and more comprehensive coverage of this topic can be found in other recent publications [75, 142, 143]. Two segmentation methods are relevant to this thesis. The first automated segmentation method is based on active-contour models (‘snakes’ [144]). By actively fitting a smooth spline (or set of points) around an object on 2-D surface, the minimum energy function defines the optimal cell boundaries. This method can be utilized to automatically segment or to split dividing cells, but is relatively slow because of its iterative nature [145-148]. A more robust way to segment cells is based on the histogram of pixel intensity, where the threshold value (called also brightness threshold) is defined to create a new binary array. All the pixels from the corresponding position in the intensity array with a value higher than the threshold value are labelled as 1, and the rest is labelled as 0 (the background).

Raz Shimoni Chapter 2 Literature Review

Hence, the binary array can be described as:[92]

Equation 1:

1,if a ( x , y ) T gxy(, ){ 0, else where a(x,y) is the input greyscale image, g(x,y) is the threshold binary image output , and T is the threshold value. The output is the label “object” (or foreground) or “background”, which is usually represented as a Boolean variable “1” or “0”. Frequently, the term ‘threshold level’ is used, and refers to the proportion from the maximum values of the intensity image and lies in the range [0, 1]. The threshold level can be adaptively determined by using an automated algorithm. Among a large selection of algorithms to find the optimal threshold level, the Otsu algorithm effectively minimizes the variance between pixels that are recognized as the background (pixels with value 0) or the object/cells (pixels with value 1)[149]. For the purposes of this thesis, segmentation therefore describes the separation between cells and background using thresholding. Once the grey-scale images are converted to binary images, mathematical operations are usually used to improve the quality of the segmentation. For example, the removal of isolated pixels or filling of holes using relatively straightforward approaches can clear the separation between cells and background. Standard mathematical operations on binary images are well- documented for MATLAB [142] (which is the environment for the development of TACTICS, as described in section 2.4.8).

2.5.2. Tracking cells from time-lapse data

An essential step in converting the individual frames of a time-lapse movie into useful biological data involves determining how cells in one frame relate to cells in another (i.e. which have the same identity). Long-term tracking of dividing cells is necessary to link cells and their ancestors and to reconstruct cell pedigrees. Migration is vital to the function of T cells, but the rapid movements of immune cells can make immune cell tracking challenging, requiring specialized tracking procedures. As reviewed in [107, 150-152], there are several automated cell tracking tools available, which involve either model evolution tracking or detection. In the tracking by model evolution approach: the evolved shape of each cell is located, and by combing both

Raz Shimoni Chapter 2 Literature Review segmentation and tracking simultaneously from frame to frame, the projection of the detected cell (area over time) in the entire image is created. The propagated 3-D tubular structure represents the spatial and temporal location of the cell. This approach is also able to handle topology changes resulting from cell divisions [153]. However, there are three main limitations to this approach. Firstly, accurate tracking requires the overlap of cell pixels between consecutive frames. Secondly, it is computationally demanding, and thirdly, it is less amenable to manual corrections. The second approach, which was applied through the course of this thesis, is based on the detection of cells and the use of algorithms to associate cells over subsequent frames. Cell trajectories are generated in two main steps. First, the distances between all objects in the subsequent frames are calculated. Then, optimization of the combinatorial choices is achieved by minimizing the total inter-object distance [154] [155] [100, 156, 157]. The exact utilization of tracking algorithms in TACTICS is explained in the next chapter.

2.5.3. Graphic visualization

Another increasingly important aspect of bioimaging informatics is data mining, which deals with methods to search for biological meaning in large data sets, and to manage complicated structured database or extracted libraries [158]. An example is the adaptive approach to manage large structured multi-parametric experimental data [159]. Another important aspect is visualization, which can involve simple plots or more complex graphical representations of data [160, 161]. Scatter plots are particularly useful diagrams to assess the degree of correlation between parameters. Since there are many numerical descriptors of cellular properties in image cytometry (either from microscopy imaging or LSC) by assigning different colors, shapes, and size to each data point, correlations between multiple parameter, sub-groups, or clusters, can be identified. When large numbers of data points are achieved (i.e. data from high content analysis [162]), 1-D histograms can be used to represent the overall distribution (or frequency) of a given parameter. Finally, graphic representation also provides the means to remove/ignore noisy data, using approaches such as gating, as will be illustrated in Chapters 4-6.

Raz Shimoni Chapter 2 Literature Review

2.5.4. Lineage informatics

Cellular lineage trees are the equivalent approach to the analysis of evolution and evolutionary taxonomy using methods that explore distances and symmetry [163, 164]. Historically, the origin of studies for cellular lineage trees was established using the invariant lineage of C. elegans as a model system for the investigation of developmental processes, such as mutant phenotypes, gene expression patterns, and the programming of cell death [165, 166]. This nematode features a transparent body, allowing insight into cellular division and differentiation under a light microscope. Through direct observation of C. elegans, the first and most comprehensive cell lineage map [167] was manually constructed, and a complete map of embryonic cell lineage from zygote to adult was generated [168] [169]. This study was performed by Sir John Edward Sulston, a winner of the Nobel Prize in Physiology or Medicine [166], who paved the way for cellular lineage informatics. In the 1990s, advances in microscopy facilitated the ability to record data from DIC imaging. By playing the acquired movies back and forth, it was easier to register and reference division events and to reconstruct lineage trees [170]. However, although being extremely useful, this approach was extremely tedious and does not capture important properties of shape and motion for the cells being analyzed. Based on that system, Schnabel et al. introduced ‘Simi Biocell’ [171], the first significant semi-automated software for cellular lineage trees. This software is utilized in dozens of publications to date.

In recent years, there has been much effort to achieve reliable automated identification and tracking of cells, which will lead to a further reduction in the time and labour required for the reconstruction of lineage trees. Significant endeavours to automate the tracking and lineage reconstruction of C. elegans was done using the open-source NucleiTracker4D software, which utilizes an automated tracking with a dedicated user interface for manual corrections [172]. While the programs above specifically aim for the lineage reconstruction of C. elegans, new dedicated software and computational approaches are required to deal with other biological systems such as migrating cells.

Raz Shimoni Chapter 2 Literature Review

A significant advance to automate reconstruction of lineage trees of migrating murine neural progenitor cells was introduced in 2006 by Al-Kofahi and colleagues, who automatically associated existing cell tracks to the next sequences of segmentations using a bipartite assignment method [173]. While this approach provide a high accuracy of detection (~85-90%), a significant drawback of this program was the lack of user interface, which is extremely important for manual corrections, and it prevented external users to operate the program. Based on this program, Winter et al. introduced the LEVER program for lineage editing and quantitative automated analysis for neural stem cells [174]. Their improved tracking algorithm, termed Multitemporal Association Tracking (MAT) is based on minimum-spanning tree optimization [175]. Additionally, LEVER is integrated with another outstanding program developed by Cohen et al. named Algorithmic Information Theoretic Prediction (AITP) [176]. AITP automatically allocates cell divisions by combing image sequence summarization [177] and semi-supervised spectral learning algorithms [178].

2.5.5. High Content Analysis (HCA)

Combining the power of modern microscopes to simultaneously image multiple positions with advanced micro-technologies has led to the requirement for high-performance computational tools. The further improvement in both biological and imaging conditions enables automated computational pipelines to process and analyze cells in a high-throughput manner [179] [180] [181]. This can be done, for instance, through the use of multi-well assays and totally automated acquisition of data [182], or by a program that can automatically focus the microscope in events classified as cell divisions [183]. In future, more sophisticated computerized tools and mathematical models will be required to handle multi-dimensional data sets in order to provide information about biologically relevant data in the most efficient approach. This will deliver a deeper understanding of the fundamental processes in a more systematic and effective way. For instance, automation of data processing with HCA based on the characteristics of different cells, positions or experiments [184] [185, 186].

Raz Shimoni Chapter 2 Literature Review

HCA involves several challenges. Firstly, there is a need to collect large data sets, and there is a limit to the data storage capability. Affordable data storage devices normally store only few terabytes, which was sufficient for the experiments described in this thesis work. In addition, computer memory, particularly the Random-access memory (RAM), is limited to several gigabytes. Therefore multiple experiments/positions/ wells are required to be processed in batch mode. The automation of this pipeline is a major task, in particular when there is large inconsistency in the data between different positions or within a movie. For this reason, standardization techniques should be utilized in robust analysis of cells, such as performed for expression measurements of certain molecules in response to a different treatment [187]. In addition, accessing extremely large data sets might be resource- demanding, can halt interactive exploration, and software might even "crash" due to memory limitations. Another challenge is that the quality control of large data sets is more demanding. HCA was one of the prerequisites during the development of TACTICS, as described in section 3.3.

2.5.6. Environment and resources for bioimaging software

Computer languages used for bioinformatics software programing are typically C++, JAVA, MATLAB, and Python. The developmental environment and the availability of computational resources such as image processing libraries, have an immense influence on the characteristics of the prospective bioinformatics software. For instance, Imaris® was written mostly in C++ and OpenGL, which facilitates high 3-D performance. Fiji [137], ICY [188] and ImageJ [189] were written in JAVA which provides an excellent interactive user interface to incorporate hundreds of accessible plugins and online connections [188], in addition to cross platform compatibility (e.g. Win, Mac, Linux).

An essential requirement for rapid development of bioimaging software is the availability of image processing resources and analysis functions. Scientific libraries and algorithms are a good example of such resources, such as the open source libraries, VTK (Visualization Toolkit) and ITK (Insight Segmentation and Registration Toolkit), provided by BioImageXD [184]. Image processing can be performed using most of the standard software for Bioimaging Informatics,

Raz Shimoni Chapter 2 Literature Review

MATLAB Image Processing Toolbox (IPT), or dedicated image processing libraries for C++, C, Python and Java languages such as the OpenCv library [190]. More examples are the Python Image Processing (PIL) and Matplotlib libraries for Python programmers [191], in addition to the powerful scientific libraries NumPy and SciPy [192]).

2.5.7. Open-code and commercial software

Software for cell tracking and analysis are predominantly classified by the accessibility sources of code [193]. Commercially software packages such as MetaMorph® (Molecular Devices), Imaris® (BitPlane) and ImagePro® (Media Cybernetics, Inc.) usually aim for a broad range of applications, and are characterized by higher finishing standards, excellent documentation, and supported extensively by a large community of users. In addition, programmers can share macro language, export data for additional analysis, or even use extra-customized modules. This provides some degree of flexibility that is often required in many applications.

Nevertheless, commercial software applications are typically compiled and their source code cannot be accessed. Bi-directional integration between image and numerical data is limited, and commonly only possible via extraction of data to third party software. Therefore, in many cases, these tools do not provide an adequate environment for customization and algorithm development. The second group of software packages are open-source, and provide accessibility for academic purposes such as Fiji [137], ICY [188] and ImageJ [189]. This group includes software that features a user-friendly development environment and multi-functional capabilities similar to the commercial software. Microscopy pipelines and modules are frequently updated by developers and users, who are encouraged to create and integrate code as plug-ins and macro-language tailored to their needs. For more specialized tasks, a higher degree of customization is required. One exceptional software package is CellProfiler [194] from the Carpenter lab (http://www.broadinstitute.org/~anne/). CellProfiler is a modular, high-throughput image analysis toolbox for image cytometry [195]. Other significant software designed for high-throughput cell imaging is CellCognition, that offers novel computer vision and machine learning

Raz Shimoni Chapter 2 Literature Review techniques based on classification of cell morphologies [196] and Cell-ID [197], that offers sophisticated automatic analyses using scripts input by the user. Extensive software that is dedicated for in vivo imaging of T cells named FARSIGHT [198] was developed for spectral unmixing, 3-D segmentation and tracking algorithms.

2.5.8. Development of MATLAB based bioimaging applications

Despite the dramatic increase in both the quality and variety of bioimaging informatics tools there is yet to be found the ultimate software that can perform all types of analysis, and each of the software packages have their strengths and weaknesses. Due to the lack of available software, researchers have looked to develop new tools for specific tasks. MATLAB is one of the most utilized and powerful environments for image processing and computational analysis, providing an excellent interactive environment for numerical computation, and offering a comprehensive set of built-in functions and many open-code libraries.

A good example is the MATLAB central file exchange1 which is an extensive library that offers outstanding resources for image computing algorithms and GUI tools, including image processing, algorithm implementation, and data visualization tools [199]. MATLAB has dozens of toolboxes (either commercial or free) for more specialized needs, such as the SBML-SAT and MATLAB SimBiology for sensitive analysis of systems biology and parameter estimation [200]. The most relevant toolbox for this thesis is the MATLAB IPT [201], which is commonly utilized to process microscopy images. MATLAB IPT is one of the most commonly used tools to program open code bioimaging software. Other examples include CellProfiler [202], DYNAMIK [203], CellAnimation [204] , DcellIQ [205], FARSIGHT [100], CellTracer [206], and more [107]. Frequently, MATLAB is chosen to develop prototypes. For instance, CellProfiler was initially programmed in MATLAB [202] to allow rapid development, and was later written in Python which provided a good compromise between speed and quality of the user interface [194].

Because of the advantages described above, MATLAB (with MATLAB IPT) was our environment of choice. However, MATLAB is missing several important

1 www.mathworks.com.au/matlabcentral/fileexchange/

Raz Shimoni Chapter 2 Literature Review capabilities, offering only basic functions, and requires additional development that can fill the gap between the lack of customization offered by most software, and the lack of elevated tools in MATLAB. In a recent publication by Kankaanpaa and colleagues (2012), a protocol for the comparison of several state-of-the-art bioimaging software (such as those described in Sections 2.5.7 - 2.5.8) and a detailed survey was performed [3]. Their survey included MATLAB (with IPT), which highlights the requirement for development new advanced tools. Therefore, I provide a new survey composed of three columns was generated (Table 1). In the first column are the ideal capabilities (such as those I aimed to achieve in TACTICS united with MATLAB, working as a full package), in the second column are the current MATLAB capabilities (with IPT), and in the third column are the standard capabilities of software extracted from the collection in the original comparison. Special features have been included for comparison, with some trivial parameters ignored.

Raz Shimoni Chapter 2 Literature Review

Ideal capabilities MATLAB + IPT Standards in the field

Availability of source Open-code2. Mixed (as explained in Section code. 2.4.7).

Customization of user Practically, there is no Most software applications have interface: modular, dedicated user interface, interactive and user friendly adaptable, interactive and but achievable through interface, but customization is user friendly. programming. limited or not straightforward.

Animation with tracks Practically, no. But Supported in most software. and annotations controlled achievable through from dedicated user programming. interface allowing sophisticated visualization.

Ability to track dividing Practically no. But Not trivial. cells and reconstruct achievable through lineage trees. programming.

Customization, MATLAB uses high-level Can rarely be extended, or adjustability and language, considered easy require complicated extendibility. to use. MATLAB-based programming and debugging. applications are more adjustable and extendible.

Possibility to process lists Practically it is unavailable, Trivial in most software. of commands in batch but achievable through mode and high- programming. throughput.

Dedicated tools for Practically it is unavailable, Not trivial. lineage informatics and but achievable through polarity analysis. programming.

Platform to develop Excellent. Not trivial. algorithms.

Availability of supporting Excellent. Not trivial. Usually involved toolboxes such as third party software (including allowing statistical MATLAB). analysis.

Table 1. Comparison between ideal requirements and current availability that demonstrates the demand for development of new MATLAB based tools.

2 Although MATLAB source code is closed, millions of users in the academy and industry validated it reliability.

Raz Shimoni Chapter 2 Literature Review

This survey shows that currently available tools lack the vital features that are required for the projects described in this thesis, thus necessitating the development of new approaches. Conversely, MATLAB is perfect environment for rapid development of image analysis algorithms, but lacks fundamental requirements and tools that require further programing. Consequently, the development of new tools that meets these requirements is the topic of the next chapter.

Raz Shimoni Chapter 2 Literature Review

TACTICS Toolbox for Lineage Informatics

3. Chapter 3

Raz Shimoni Chapter 3 Lineage Informatics

3.1. Introduction

To sustain an adaptable immune response that can be switched on and off as required, naïve T cells produce both effector and memory cells. One of the intriguing questions in the biology of T cells is whether memory and effector cells can evolve from each other, or are generated from independent lineages [207-210]. Previously, time-lapse imaging and analysis of lineages of hematopoietic stem cells have provided exclusive information about asymmetry and heterogeneity [211, 212]. Therefore, it can potentially be useful to elucidate whether the postulated transition between memory and effector cells is dictated through subtle but consistent differences in cellular properties that eventually affect downstream fate and differentiation, or if such alteration is triggered by a single division [34-37].

Recent in vivo studies showed that there is a vast diversity in the number of progeny and differentiation pattern from individual naïve CD8+ T cell [213-215]. However, considering the limitations of in vivo imaging (as described in Section 2.2), a comprehensive in vitro analysis is highly desired. To investigate how characteristics of the naïve CD8+ cell can orchestrate diversity in naïve CD8+ T cell after antigen stimulation, and to determinate whether naïve CD8+ T cells can adopt different fate through differentiation, our laboratory has developed a novel in vitro imaging methodology that includes microfabricated microwells and micropipetting for cell deposition. A single founder cell is activated; the two daughter cells are separated and deposited in different microwells and these daughters and their progenies were followed for several generations using live cell time-lapse imaging.

This methodology requires analytical software to track and analyse the obtained data. The major challenge in lineage analysis is the requirement to perfect the tracking of moving cells. Any single tracking error dramatically affects the entire pedigree. For instance, even a small error rate of only 0.1% will affect 10% of all cell lineages after only 100 time points [216]. Since perfect automated tracking is unavailable for low resolution and time-lapse data [157], and even less accurate for crowded dividing T cells along generations, manual inspection and correction of tracks is required. However, manual corrections are tedious and time consuming. In addition, human labour is also error-prone, and obviously more subjective.

Raz Shimoni Chapter 3 Lineage Informatics

Therefore, the ideal approach is to combine state-of-the-art automated tracking algorithms with tools to assist the person to inspect and correct segmentation, annotation, and tracking, are essential to speed the reconstruction of lineage trees in optimal accuracy. Such a software was previously successfully implemented by T. Schroeder et al., who developed a tool named Timm’s Tracking Tool (TTT) to study the existence of hemogenic endothelium [217] and demonstrated that hematopoietic cytokines can instruct lineage choice in hematopotic stem cells [218]. Another significant software for lineage tracking of B cells was recently introduced by Chakravorty and colleagues, who developed a customized JAVA based TrackAssist software to reconstruct lineage trees [219]. TrackAssist includes automated tracking algorithms and complimentary manual inspection tools, and is capable to correlate many aspects of cells history and fate more efficiently. Nevertheless, these tools, and other software such as introduced in Section 2.4, mainly focus in developing advanced tracking algorithms and prediction models, and lack several necessary properties such as sophisticated end-point analysis. Therefore, it was essential to develop a program that can perform efficient analysis of small and highly motile cells, such as T cells, with the capability to easily adopt new assays and types of analysis.

Herein, I present the development of TACTICS for interactive lineage informatics of T cells. The chapter is structured as follows. In Section 3.2 I describe the experimental methodology that has been developed in our laboratory to study fate decisions of CD8+ T cells. In Section 3.3 I provide insight into the architecture and programming approaches of TACTICS. This includes the design of the Segmentation Module, methods and implementations of the Tracking Module, and parameters measured by the Measurements Module. In addition, I emphasize the ability to expand this pipeline and to enable the adoption of TACTICS to other applications, such as shown in Chapters 4-5. Furthermore, TACTICS user-guide is attached as Appendix I, providing technical information relevant to the use of TACTICS, and describes additional venue of end-point-analysis selections. A list of TACTICS files is displayed Appendix II. . Several other open-code functions were utilized and are detailed in Appendix III with acknowledgment to its authors and links to its original source. In Section 3.4 I present the results of this chapter. First, I portray the general

Raz Shimoni Chapter 3 Lineage Informatics features of TACTICS Lineage Module, explain the advantage of gating within lineages, and show different visualization methods to explore lineage data. Then, I exemplify how tools for visualization and data exploration can promote the identification of trends in highly complex data, and demonstrate the capabilities of TACTICS to study lineages of activated CD8+ cells. Finally, I summarize the accomplished research in Section 3.5.

3.2. Methods

3.2.1. Experimental methodology

Experiments and data images were acquired by Mohammed Yassin, a PhD candidate in our lab. Briefly, naïve (not previously exposed to cognate antigen) CD8+ T cells from OT-1 mice (transgenic for a T cell receptor for the peptide of sequence, SIINFEKL) are incubated for 40 hours with peptide-pulsed bone marrow-derived Dendritic Cells (DC’s) (from CAG-DsRed fluorescent transgenic mice) to enable co- stimulation of inflammatory signals that have been demonstrated to be important for a full CD8 response [220]. The DCs are pre-adhered to microfabricated cell microwells as shown previously [34, 128]. The T cells express an enhanced Green Fluorescent Protein (eGFP) that allows tracking of individual cells and the DCs are detected in red channel (Figure 3.1.A). The DCs and T cells are cultured and confined within 125 μm2 microwells with 60 μm high walls to prevent the escape of cells and entrance of other T cells. For optimized survival and proliferation of activated CD8+ T cells, a low level of IL-2 (10u/ml) was added to the culturing media. The expression on the T cells of the memory differentiation marker CD62L (also called L-selectin) is measured by including low concentrations of Phycoerythrin (PE) labelled anti-CD62L in the medium during imaging. When concentrated on the surface of CD62L-expressing cells, fluorescence is clearly detectable. About half an hour before the first T cell division, the T cell stops moving and increases in size, cytokinesis takes approximately 20 minutes, after cytokinesis, each daughter is manually placed in a new microwell using a micropipette attached to the microscope (Figure 3.1.B). Multiple generations are then imaged every 30 seconds for 7-14 days, allowing tracking until 9th generation (8 rounds of divisions) (Figure 3.1.C) and further characterization on population base. In our latest

Raz Shimoni Chapter 3 Lineage Informatics experiments, the 2 daughter cells of each founder cell are imaged for approximately 4 generations, and then 4 or 8 T cells are secondarily transferred, and cultured and imaged in different microwells, while the remaining cells continue to be imaged. Each cell lineage includes approximately 11 positions that are automatically segmented, labelled, and tracked to generate subpedigrees, which requires about 10 hours of manual processing, depending how complicated the pedigrees are. To insure the reliability of the data, this process is inspected by a second person, which takes another several hours. The automated extraction of trajectories and montages requires overnight processing. Connection of the lineage and structuring as a file input into the Lineage Module takes several minutes with the Connecting Module. The Lineage Module is then used for lineage analysis (Figure 3.1.D).

Raz Shimoni Chapter 3 Lineage Informatics

Figure 3.1 Description of the experimental protocol for lineage study. (A) DC and naïve T cell are incubated together within the samme microwell until the first division of T cell occurs. (B) Micromanipulation system is utilized to transfer the two activated daaughter cells into different microwells. (C) T cells are imaged for more than a week throughout proliferation and death. (D) Obtained images are analyzed by TACTICS, which provides the ability to: 1) process multiple images and to track founder cells and their progeny through several rounds of divisions; 2) construct cellular pedigrees; 3) extract meaningful information from lineage trees, and 4) compare and correlate measured parameters between ancestors and siblings based on kinship.

3.2.2. Imaging setup

Time-lapse images were obtained with IX71 inverted microscope (Olympus, Tokyo, Japan) equipped with a Nipkow disk-type confocal unit (Yokogawa CSU22, Tokyo, Japan) and EM-CCD Andor camera (Model: iXon EM +885, Belfast, Northern Ireland). The objectives used were Olympus 20X and 40X. Mulltiple stage positions were captured (controlled by MetaMorph software version 7.7.11.0 ) with various sampling rate (e.g. 30 sec, 1 min, 2 min, or 10 min) for severall days, and

Raz Shimoni Chapter 3 Lineage Informatics were saved as 8-bit 2-D arrays (1002 × 1004 pixels). Cells were cultured in enriched RPMI media (GIBCO, Life Technologies, CA, USA) with 10 μl/ml of recombinant human IL-2 (R&D Systems, Abingdon, UK) in the presence of 1 μl/ml of PE conjugated CD62L antibody (LifeSpan Biosciences, Inc. Seattle, WA, USA). The imaging cultures were maintained in 37 degrees using The Box (Life Imaging Services GmbH, Basel, Switzerland) chamber around the microscope and the carbon dioxide level of 5% were achieved using The Cube (Life Imaging Services GmbH, Basel, Switzerland).

3.2.3. Image processing

Segmentation was based on fluorescence from the green channel (eGFP). Morphological close operation with a 3x3 disk structuring element, [0 1 0; 1 1 1; 0 1 0] was applied to each image to close spaces in the morphology of the cells (to smooth the edges)[142]. This step was followed by a background subtraction, which was applied by removing the mean intensity of the background from each pixel in the image. Cells were segmented according to Equation 1, where the threshold level was chosen by Otsu`s method[149]. Following the segmentation step, small segments (less than 100 pixels) were removed. Intensity pixels with zero values within the cells (‘holes’) were converted to values of one. To insure better segmentation, the process above (threshold, removal of isolated segments, filling “holes”) was repeated again for each cell. Finally, a splitting algorithm based on watershed was applied to automatically separate overlapping cells.

3.3. TACTICS pipeline: methods and features

TACTICS is structured from several central modules and tools, supporting flexible development and improvement. Each module has defined functions, with a network between the different modules enabling flow of information between modules for the production, assessment and presentation of the data. This simplifies the analysis scheme, and allows an unfamiliar user (the person who operates TACTICS) to study and follow the program. Furthermore, TACTICS achieves high- throughput automation with minimum human intervention by allowing batch processing and execution of inquiries list for many positions. TACTICS needs a

Raz Shimoni Chapter 3 Lineage Informatics specific file format and structure to interact between modules, a common requirement in the field [221]. Exportation of data from MATLAB to other formats such as Excel is also possible. The general use of TACTICS is shown in Figure 3.2, and is explained in the following sections.

3.3.1. TACTICS Segmentation Module

Commonly, microscopy images require some processing procedures prior to segmentation. For example, a noise-filtering algorithm based on a low-pass filters or filters to remove non-uniform illumination effects [92, 142, 222]. The Segmentation Module is employed to perform such image processing procedures and segmentation. The Segmentation Module consists of three main steps: firstly, an experiment file is uploaded (Figure 3.2.i). The raw images data are interactively visualized and the user can choose optimal settings of image processing (Figure 3.2.ii). In the second step, an interface for custom functions is utilized to improve the quality of the original raw files, by applying a pre-processing procedure, such as implementation of denoising and contrast enhancement algorithms (Figure 3.2.iii).

Raz Shimoni Chapter 3 Lineage Informatics

Figure 3.2. Schematic illustration of TACTICS pipeline for lineage reconstruction. TACTICS consists of several core modules that integrate automation, and tools for quality control and manual corrections. The aspects of validation, extraction, accumulation and organization of data are described as follows. The core of this pipeline is the experiment file that contains information about the images and the objects, which streams along the pipeline and the different modules. Therefore, the first step is the creation of new experimental data files for a single movie or data set that consists of multiple positions (i). To set up the experiment, the user inputs a set of .tif images into an experiment generator tool for preliminary inspection. Experimental folders are then created on the basis of position or optical section (for confocal microscopy), containing information about each processed channel. Next, the Segmentation Module is utilized to detect cells. In the Segmentation Module, the user interactively chooses the optimal settings (ii). Depended on user choice, each image is filtered to enhance the contrast between cell and background. Each binary image is saved as a new file, linked to its corresponding raw and filtered files (iii). The filtered grayscale images are binarized (iv) and morphological operations can be applied to improve the cell segmentation. The operations applied to the images are logged for further inspection on a frame-by-frame basis. Next, the Tracking Module is responsible to process the binarized data images. In the Tracking Module, cells are labeled and received an individual ID (v). The cells are associated with the cells in the following frames to construct the association matrix (vi). A user interface enables efficient control for quality, corrections, and annotations (i.e. marking cells that divide or die) (vii). A linking function receives the association between frames and constructs the trajectories matrix, which contain the cell trajectories over time (viii). Manual corrections on the tracking can be applied if required (ix). The information is saved in the experiment file and is imported into the Measurements Module. The Measurements Module is utilized to inscribe a list of parameters to be

Raz Shimoni Chapter 3 Lineage Informatics quantified on the basis of single cells, dividing cells, or the whole population (x). With a mouse click, the list is processed and cell libraries are generated and saved (xi). The libraries are then imported to the Analysis Module and processed by a dedicated tool that can merge lineage data (xii). Exported structured data is loaded to the Lineage Modules for end-point- analysis.

While some programs allow the creation of a procedure list task [184] or the customization of macros, this approach is inadequate in many cases. This is particularly important prior to automated segmentation when high-throughput is required. TACTICS implements an interactive pop-up menu, which allows automatic segmentation, including the generation of virtual channels, similar to other interactive approaches such as building blocks in ICY [223] and KNIME [224]. The interactive pop-up menu allows the user to explore different combinations of filtering, arithmetic operations and to customize a procedure list more efficiently. The user can export and import optimized combinations of settings for high- throughput analysis or applies the settings on a frame-by-frame basis. The third step is the parallel to the second step, but instead of using filters it applies segmentation procedures and morphological operations (Figure 3.2.iv). Routine segmentation methods used are Otsu MATLAB built-in function for fluorescent images, Level Set Evolution algorithm for DIC[225], and, where the images contain compounded intensity classes, K-means clustering algorithm is used. To separate touching cells, a watershed algorithm is used [226] in a similar approach to that described by Al- Kofahi et al.[173].

3.3.2. TACTICS Tracking Module

TACTICS employs robust tracking methods that require no prior information about directionality or turning angles. As an input, the Tracking Module accepts a single experimental file, and then registers and tracks cells in each channel. A customized GUI is dedicated to multiple tasks. A screenshot of the Tracking Module is shown in Figure 3.3. The user interface facilitates straightforward processing with stable and user-friendly dedicated tools, such as selection tools and interactive scrollbars [227], to follow cell divisions and to mark cell division and death. Since most immune cells, and in particular T cells, are typically characterized with a compact shape, the use of centroids to represent the location of the cells is a good approximation. Therefore, the presence of each cell within the binarized images is

Raz Shimoni Chapter 3 Lineage Informatics identified and labelled by its unique centroid (Figure 3.2.v). Quantitative measurement of cellular parameters, such as morphology, is a standard procedure in image cytometry [223], and the compilation of measures is also available in TACTICS. Thus, the labelling function outputs a new updated experimental file that contains structural information that is required for the subsequent steps of the pipeline.

Figure 3.3. Screenshot of the Tracking Module. (i) Main menu and icons bar providing a full set of options (see Appendix I. p. 17-19). (ii) Main figure axes and pop-up buttons to customize visualization options. In this given screenshot there are several tracked cells. The tracks are visualized in different colors. The tracks are shown for ch00 (which is the GFP in the shown screenshot). Visualization is shown for ch02 (which is DIC in the shown screenshot). The user easily controls basic operations, such as switching between channels, zoom, navigation and setting the type of visualization. (iii) Scrollbar to navigate through the movie. All cells were labelled, but only selected frames were associated (as exemplified as Figure 3.2.B.ii). (iv) Selection of several different modes that assist the user with handy inspection tools. (v) A secondary figure supports operations corresponding to the selected mode. These operations include manual annotation, validation and editing to correct for displacement, division, death, merging or splitting due to errors in the segmentation, association and tracking. In this particular example the area of the cells is shown for each tracked cell at each time point.

3.3.2.1. Association (Figure 3.2.vi)

The tracking in TACTICS is based on reconstruction of trajectories from centroids. Early versions of TACTICS (up to version 2.2, May 2012) relied on the MATLAB implementation of the ‘Crocker algorithm’ for tracking. The Crocker algorithm

Raz Shimoni Chapter 3 Lineage Informatics

was initially developed to track particles moving by Brownian motion, and is a well-known nearest-neighbour algorithm [154]. To reduce the complexity of this optimization problem, the Crocker algorithm only considers distances shorter than a particular length (dependent on the distance that the particles moved between frames). Although the Crocker algorithm is accurate and has the ability to recover cells that might disappear for several frames, it cannot automatically deal with multiple divisions and does not recognize one of the daughters as a new cell. Another major drawback of the Crocker algorithm is that increasing the complexity (i.e. large number of cells) results in exponentially more complicated combinatorics that can lead to a failure to track. Later versions (version 2.2 up to present) implement the ‘Hungarian algorithm’ (also known as the ‘Munkres-Kuhn algorithm’)[228-230] to associate cells by solving the ‘assignment problem’, (also known as the ‘maximum weighted bipartite matching problem’[155], which has been implemented in several other cell tracking software platforms [100, 156, 157]. The mathematical definition of the ‘assignment problem’ is to find the minimum distance that cells travelled in two consecutive frames. This algorithm allows finding the minimum weight matching of a bipartite graph. The edge weights are captured by a weight matrix. Cells at frame t are linked with the nearest cell to that position in frame (t+1). Each cell in frame t+1 is assigned an index number which corresponds to index of cell in frame t. If there is no match, either because the cell travelled a larger distance than expected or cells disappeared/appeared, the corresponding index will be −1. The main advantages of this approach are its computational efficiency in terms of speed, it can be applied even when cells move large distance between frames, it does not require prior knowledge and assumptions about directionality, and it supports convenient manual corrections (see Appendix I. p. 24). This method was timed to take only 6 seconds to associate 4126 frames (one cell divided into 16, in Dell Latitude E5540, 16 gigabytes RAM, Intel(R) Core(TM) i7-4600U CPU @ 2.10GHz). When very long tracking is required, such as the tracking of founder cells through 8 cycles of divisions (as will be shown in Chapter 6), skipping specific frames can bypass unnecessary manual labour for periods during the experiment where no division or death is taking place (for instance, association every 4 frames will save 75% of the labour). TACTICS has a unique dedicated selective operator with the capacity to

Raz Shimoni Chapter 3 Lineage Informatics

associate only selected frames, and to our best knowledge this ability does currently not exist in other programs. This method is explained in Figure 3.4. Firstly, an experiment is associated in low frequency (e.g. every 4 frames) (Figure 3.4.A). Next, the user selects to associate particular frames more frequently (e.g. to annotate cell divisions or when high proximity cells require closer inspection. Consequently, unification is applied between the new and existing association (Figure 3.4.B) and the new association is integrated. All associations are structured with the matrix of association (shown in Figure 3.5.A) that consists of 4

columns. The first two columns are the (xi,yi) coordinates of each cell, for instance cell 1 (163,429) in frame t-1. The third column represents the fate of the cell. For example, the fate of cell 2 in frame t is 0.1, which stands for division. The forth column represents the frame where the cell is pointing at (as explained above for un-selective versus selective operator).

Figure 3.4. Illustration of the selective operator in the Tracking Module. The association effectively links between the locations of the cells by optimizing the distance that cells moved over time (A). Cell associations based on every 4 frames. For example, frame 30 will be associated with frame 26, which is associated with frame 22, and so on (illustrated by blue colour numbers). In this mode the user can see the associated frames in the panel under TACTICS scrollbar, where green bands are displayed every 4 frames. (B.i) If the user decided to re-associate all frames between 20 to 25, a new association is created. Now, frame 25 is associated with frame 24 that is associated with frame 23, and so on for all 'red frames'.

Raz Shimoni Chapter 3 Lineage Informatics

(B.ii) Since the first frame (20) and the last frame (25) are not associated in the previous 4 frame sequence (A), the unification step is required. Therefore, the last 'blue frame' before the alignment (18) is associated with the first 'red frame' (20), and the first 'blue frame' after the alignment (26) is associated with the last 'red frame' (25). TACTICS displays green dense sequence (between 20 and 25) that represents more frequently associated region.

Figure 3.5. Annotation procedures in the Tracking Module. (A.i) The association matrix consists of 4 columns: the x and y coordinates, annotation number (i.e. 0.1 for cell division and 0.2 for cell death), and association value that refers to the location of the same cell in previous frame. (A.ii) When the user annotated cell 2 in frame t as a dividing cell (red x mark) the annotation number received a value of 0.1 (showed in red color). (A.iii) One daughter cell received association value of -1 (new cell emerging, showed in red), while the other daughter cell (the cell that was closer to the centroid of the parental) continues the association to the parental. (A.iv) When the user annotated cell 1 in frame t+1 as a dying cell (red + mark) the annotation number received value of 0.2 (showed in red color). The annotations are attached to the extracted data in the Measurements Module, and are utilized in later stages to reconstruct the lineage trees. (B)The splitting algorithm scans for annotation value of 0.1 and splits the matrix of trajectories in the next frame after the division was annotated. At the end of the tracking algorithm, the trajectories are organized in a matrix. (B.i) The matrix of trajectories before splitting. (B.ii) The matrix of trajectories after splitting. It can be seen that although the number of cells is constant, the number of columns in the split matrix is doubled. The different densities in the trajectories matrix are denoted by the different rate of association (as explained in Figure 3.4).

Raz Shimoni Chapter 3 Lineage Informatics

3.3.2.2. Annotations (Figure 3.2.vii)

Although several tools for automated detection of mitosis and apoptosis exist [231], the accuracy of automated approaches is not perfect. In practice, it takes only few minutes to follow through a single movie and to annotate dying and dividing cells. It would take a similar time to manually inspect annotations that were allocated using automated algorithms to insure the validity of the data. From this reason the current version of TACTICS is based on manual annotation of death and division. However, algorithms to identify events such as division can be adopted or developed in the future if required.

The Tracking Module provides the user with the ability to manually annotate final fates such as mitosis and apoptosis. In the example shown in Figure 3.5, cell 2 is dividing in frame t into cells 2.1 and 2.2 in frame t+1. Dividing cells are manually annotated with red x mark, by a mouse click on the dividing/dying cell (Figure 3.5.A, see Appendix I. p.25). To assist the user during the step of annotation (i.e. to quickly find dividing cells), TACTICS provides graphs of characteristic phenotypes that cells exhibit prior to division or death. For example, dividing cells are characterized by an increase in area (which is proportional to the size of the cells), increased aspect ratio, and reduced movement [232]. These graphs are interactive, and by clicking on events (i.e. cells that show a sudden decrease in movement), direct the user to the selected frame. Such interactive tools assist the user to mark the divisions and follow the reconstruction of the lineage trees as the movie progresses (see Appendix I. p. 28- 29). Consequently, the user can reliably track further division cycles and achieve sequences that are more complex. These features are similar to the capabilities of the software TrackAssist.

3.3.2.3. Linking (Figure 3.2.viii)

Figure 3.5.B shows a representative image of the trajectories matrix before applying the splitting algorithm (explained below). The matrix has dimensions of [m,n], where rows represent time or frame, and n represents the index of tracked cells ordered by the time of appearance. To speed up the algorithm the X and Y trajectories are joined into one matrix, where each pair of columns within the

Raz Shimoni Chapter 3 Lineage Informatics

trajectories matrix contains x and y coordinates of the same cell (i.e. [X1 Y1 X2 Y2 X3

Y3 Xn Yn] is corresponding to [1:2 3:4 5:6 n:n+1]).

To link between associated cells in frame t+1 with those in frame t, a linker function propagates on the basis of cells and link the cells based on the association value (the 3rd column within the association matrix as illustrated in Figure 3.5.A) until it cannot find the associated cell in t+1, and the chain stops. To track dividing cells, a splitter algorithm associates the two closest cells with the parental cell one frame before divisions. The algorithm searches for the next three requirements: (1) A dividing cell (value 0.1 in the 3rd column (ai) within the association matrix) in frame t. (2). There is a new cell n (indexed with -1 in the 4th column) in frame t+1. (3) The emerging cell is close enough to the coordinates of the parental cell (the user can adjust the distance value).

If it finds the three conditions above it will edit the trajectories matrix as follows: (i) Split the trajectory matrix in column 2*n into two parts; (ii) Extend the left part right 1:(2*n)> 1:(2*n+2) with pair of columns with zeros values; (iii) Transfer the x and y trajectories of parental cell n from frame t+1 to m; (iv) Reconnect the two parts of the trajectory matrix. As a result, daughter 1 is the cell that is the closest to the dividing cell, and daughter 2 is the second closest cell.

This operation repeats itself for all dividing cells and takes only few seconds in a standard computer. By combining fast association algorithm [230], with TACTICS linker and splitter algorithms, the tracking takes only 10-20 seconds for a single movie (consisting of 4126 frames, one cell dividing into 16).

The outcome of the process above is the ability to track cells over several generations, which is the basis for lineage reconstruction. A dividing cell passes its number writing from left to right, following by a dot and the numbers 1 and 2, which corresponding to daughter 1 and daughter 2 , respectively. For instance, the two daughter cells of cell number 14 will be named 14.1 and 14.2, and the two daughters of cell 14.1 will be named 14.1.1and 14.1.2, and so on.

Raz Shimoni Chapter 3 Lineage Informatics

3.3.2.4. Manual corrections (Figure 3.2.ix)

Manual corrections can be a challenging task when there is a mass of data, many close-touching cells, or if the imaging quality is low. To aid the user, the Tracking Module includes interactive tools to reduce time from the procedure. The interface includes a graphic inscription placed on top of the raw image, which are used to display errors and to manually correct the automated segmentation, association, and tracking as described below:

Corrections of segmentation:

False detections, such as inadequately segmented cells, can be either improved or discarded, using the following tools:

1. The Tracking Module includes a built-in segmentation panel with comparable capabilities of the Segmentation Module (see Appendix I. p.22). 2. Overlapping between cells can be easily separated with a single mouse click. The separations approaches are based on an intensity splitting algorithm or watershed (see Appendix I. p. 21). 3. A drawing tool provides the ability to manually correct the border of faint cells based on DIC, split touching cells, and remove null pixels. (see Appendix I. p.26). 4. The selected region can be zoomed in and visualized in a separate window for further editing (see Appendix I. p. 20).

Corrections of association:

In some cases cells may suddenly move a large distance between imaging intervals. Another potential problem is incorrect matches in newborn daughter cells, which results from crowded movies. Editing an association in such cases is instant (see Appendix I. p. 29).

Corrections for tracks:

The reconstructed matrix of trajectories provides a good opportunity for different operations for manual correction of the tracks in the cases where two cells

Raz Shimoni Chapter 3 Lineage Informatics switch tracks or when two tracks need to be joined together. Operations for editing tracks include: rejection of invalid track points, definition of a new track point, track splitting, track flipping, and more (see Appendix I. p. 24). All tracks can be easily filtered out by optional parameters such as velocity and length. Importantly, modifications of the tracks do not affect the association. The applied corrections are logged and can be saved to the experiment file

3.3.3. TACTICS Measurements Module

The Measurements Module provides an efficient user interface to select a comprehensive list of cell properties, and extracts this data in high throughput for time-lapse data. The list is adaptable and more features can be easily added. Currently available parameters include cellular parameters such morphology, cell migration, fluorescence protein localization, and more (see Appendix I. p. 33.). These measurements can be taken on the basis of selected cell, parental-daughter, only dividing cells, and all population. An important ability of the Measurements Module is that intracellular protein localization can be measured by combining one channel for segmentation and tracking, a second channel for alignment and a third channel to quantify the protein of interest [233, 234]. Tracking is not feasible after a certain number of division cycles (resulting from a confluent cell population), therefore analytical measurements such as total cell number or total fluorescence can be taken without applying any cell tracking over the entire field of view.

The output data from the Measurements Module is structured and stored in libraries that are labelled by the type of experiment, positions, cell, or selected parameters (Figure 3.2.x), ready for the end-point-analysis as described in the next section.

3.3.4. End-point analysis (Figure 3.2. xi-xii)

At end of section 3.3.3, a database is created containing montages and trajectories of tracked cells. The technical procedure and instruction are located in Appendix I. p. 38-46. Firstly, the Analysis Module is utilized to read montages and trajectories data of multiple cells, and to integrate this data in a single lineage file. This procedure takes only several minutes. Additionally, the Analysis Module

Raz Shimoni Chapter 3 Lineage Informatics provides an interactive user interface to inspect each cell, and to remove selected cells from the single lineage file. Next, a dedicated connecting lineage tool is then utilized to read the lineage files and to label each image and position of each cell, and to structure different properties about cells in vectors. The recorded information includes data over multiple generations, such as division, differentiation, death, morphology, migration patterns, and polarity. There are two forms of parameters: the first form includes measures that are constant in time and described as scalars, such as the length of the cell life and generation index. The second form includes parameters that change over time and described as vectors, for instance, morphology and migration parameters. The vector parameters are exported into a designated input file for the Lineage Module. An important feature of this module is the ability for manual corrections of lineage trees, adapted from software AceTree [235]. Similarly, the connecting module has the capabilities to merge many separate lineages into one. Since spatial displacement of dividing cells (i.e. achieved by micropipetting) results in different movies, the connecting tool was designed for the reconstruction of separated lineages. The connection is possible at every point, and the number of connections is only limited by the computer memory. Within each connecting lineage, a spacer can be added to accommodate time an interval during which imaging was suspended. If the connected trees were not imaged at the same rate, the tree with the less frequent imaging can be diluted to match the other trees. For the dilution, instead of parameters Not-a-Number (NaN) values are inserted. Finally, lineage data files are imported into the Lineage Module for lineage informatics analysis.

3.4. Results and discussion

3.4.1. Exploratory tools for lineage informatics

Schematically, a lineage tree is based on the founder cell plotted first, and the subsequent progenies plotted next (Figure 3.6.A). Usually, the vertical lines correspond to individual tracks and the horizontal lines connect between the tracks of two daughter cells at the time of division. The life of cells starts at the end of the previous division, which is represented by a branch. The x-axis represents the relationship between the families, and y-axis represents the frame number or time.

Raz Shimoni Chapter 3 Lineage Informatics

Downstream generations expand exponentially with the progression of the lineage, which corresponds to the development state of the family tree. This graph allows identification of temporal and generational trends, so, for example, a liineage tree where cells within a generation have synchronized division cycles will look uniform [236]. Lineage data can also be displayeed as a "projjection tree", where montages of cell images are projected over time and a lineage tree is reconstructed by llinking the cell projection in a lineage style (Figure 3.1.D, Figure 3.6.B). Anothher type of lineage tree is a parametric lineage, where color-coding corresponds to the magnitude of measured parameters (Figuure 3.6.B). This display provides the means to track changes through progression of ancestry, such as differentiation in expression levels of fluorescently labelled molecules.

Figure 3.6. Graphic visualization of lineage trees. (A) A simple lineage tree that was manually created and plotted using Microsoft Excel (image courtesy of Mohammed Yassin). (B) TACTICS (current version) accomplishes automated reconstruction of lineage trees. In the example given here 2-D projections are displayed for each daughter cell. This approach can assist in analyzing how an internal localization of proteins changes over the course of its development (projections are useful technique to study polarity as shown in Chapter 4). (C) TACTICS can generate parametric trees within seconds. Selected parameters are ccolor-coded for visualization in the context of the cell pedigree. The Lineage Module allows the user to manually define selected pathways with a mouse click on the lineage branches and to track the proggression of corresponding parameters from selected cells. In addition, an efficient selection of cells caan be achieved by text formatting. For example, to choose all sibliinngs of cell 1.2.1, the user inputs ‘1.2.1*’, and to select only the two daughters of celll 1.2.1 the user inputs ‘1.2.1.?’. The corresponding parameters are pooled in a scattter plot for selected parameters (Appendix I. p. 45). Another useful ability of the Lineage Module is the comparison of cellular properties on the basis of kinshiip between

Raz Shimoni Chapter 3 Lineage Informatics siblings (Appendix I. p. 45), or on the basis of generation (Appendix I. p. 45). Such approaches were previously used to reveal a good correlation between cell size and the division destiny of progeny [237], and between cell size and the division time of daughter cells [238]. TACTICS provides the ability to generate multiple scatter plots of the same property, but based on any cell relationships, any chosen parameter, and with the ability to gate within sub-populations.

To handle a large number of founder cells, with the ability to select cells and their descendants with a button click, the TACTICS Lineage Module user interface was developed. The Lineage Module combines elaborate analysis that integrates lineage trees and scatter plots, with the ability to analyse multi-parametric data in approaches inspired by flow cytometry. A screen shot and description of the Lineage Module is shown in Figure 3.7.

Figure 3.7. Screenshot of the Lineage Module. (A) The master file is loaded and stored, containing structured lineage data. The Lineage Module provides a user friendly interface, allowing rapid selection and visualization of structured data. (i-iii) Scatter plots are displayed in the three main bottom axes, illustrating the relationship between selected data metrics. For example, in plot (i) four activated founder cells are plotted as a function of time. Different colors represent the four different founder cells. In the second plot (ii), the user has the ability to focus on parameters, such as area. The user can set the boundaries of the gate (either polygon or square) based on cytometric parameters to isolate and focus on a specific sub-population. For instance, in (iii), large cells from rapid expanding lineage are gated and the circularity of the cells can be displayed. Histograms (iv) display the overall distributed profile of parameters from x-axis and y-axis data. Multiple gating and sub-gating are supported (v), allowing study of sub-populations. The panel on the left side of the screen (vi) provides control over several visualization options, such as scatter dots colored according to different styles (experiment, cell, etc.).

Raz Shimoni Chapter 3 Lineage Informatics

The correlations between multiple parameters and their derivatives can be easily pooled and plotted. Each colour in the scatter plots and histograms corresponds to a different experiment, population, generation, parameter or cell, depending on the user selection. The user can click on a particular data point and receive many options, such as reference to raw movie, projection, table of properties, and more. The ability to visualize colored cells based on kinship and combination of different types of display provides the opportunity to search for trends in structured data. The data set in the x-, y-, and z-axes are chosen from the parameter list that contains descriptive metrics, such as cell properties, morphology features, expression levels, and fate. A list showing the extractable analytical features that are currently supported by TACTICS, in addition to different types of parameter builders, is shown in Appendix I, p. 47.

One of the most desirable features of the Lineage Module is its ability to perform multiple gating (Figure 3.8), which integrates gating on lineage trees, scatter plots, and operations. (Appendix I. p. 43-45). Data from large numbers of events is collected without bias and, depending on the parameters of interest, subsets of the population can be ‘gated on’ to compare sub-populations or to identify minority populations. As with flow cytometry analysis, this allows an informal exploration of the data that can lead to new hypotheses or new strategies for analysis. The application of this approach to time-lapse microscopy means that the history of the cells can be used in addition to the static parameters assessed by flow cytometry. For instance, DISC approach allows the linking of immune cell behaviour to cell phenotype in vivo [239]. Moogk and colleagues combined gating with lineage analysis, enabling identification of sub-groups or clusters [240]. The approach in TACTICS is similar, but is integrated with the additional analysis features for lineage-based analysis. An alternative approach is provided by a recently released plug-in, ‘Panorama’, for the image analysis program CellProfiler, where imaging data can be processed multiple times using different parameters and subsets of the data can be reprocessed as required [241]. However, this plug-in focuses less on the relationships between parameters and more on how changes in parameter values can influence the outcomes.

Raz Shimoni Chapter 3 Lineage Informatics

Figure 3.8. Combining gating on llineage and scatter dot basis. The purpose of gating is to isolate or exclude data for further analysis or identify sub-groups of cell distributions. Different colors correspond to different gates. Gates can be defined and displayed by both scatter plot (A) and lineage trees (B). The user can add gates by clickinng on the lineage trees and change the lines into a new color, for example dashed purples lines are shown in this example (C). All data points within the gate are shown in a scatter plot (D).

The commercially availaable Imaris Vantage (BitPlane Scientific Software) produces multi-dimensional scatterplots to determine relationships between different parameters, but to our knowledge does not facilitate gating. As highlighted in a review of immune cell imaging, a key requirement of high throughput image analysis approaches is to retain an intimate connection to the data [8]. Thee real-time approach utilized by TACTICS allows for more rapid and less data-inttensive analysis of possible relationships between different parameters, with the direct access to images and movies allowing quality coontrol of the analytical approach. Our approach is somewhat similar to the recently developed ImageRail, which stores fluorescence and localization data in semantically typed hypercubes (SDCubes) for interactive analysis [159]. ImageRail allows dynamic linking of extracted datta to source images, and allows gating for further analysis [159]. The commercially available MetaXpress/AcuityXpress® softtwware (Molecular Devices) also facilitates interactive analysis with gating, as does the strategy of importing data into tthe commonly used flow cytometry analysis softwaarre FlowJo® (TreeStar, OR) [242]. To conclude, the Lineage Module achieves similar performances as other cuttting-edge software. Together with the development of unique exploratory tools, and the adjustable and modular properties of TACTICS (as explained in Section 3.3), tthis provides a new and powerful tool for lineage informatics. The next section demonstrates the value of these attributes.

Raz Shimoni Chapter 3 Lineage Informatics

3.4.2. New tool for mapping the life histories of CD8+ T cells

The exact cellular parameters that can allow us to discriminate between memory and effector cells from time-lapse data are still under investigation, but there are several characteristics that can potentially indicate for differentiation. For example, effectors cells are characterized by a dynamic population that can expand upon activation and decay within several days, while a stable, small population of memory cells can live for years. So, by measuring the number of cells and monitoring cell death along generations, it might be possible to distinguish between the small and stable population (putative memory T cells) and fast expanding but ultimately eliminated by massive cell death (putative effector T cells). Although our system is currently limited to accurate tracking of eight cycles of divisions, the investigation of the first generations already provides interesting observations, presented below. Preliminary data based upon the pedigrees of four founder cells (from four different experiments) was gated to include only cells with known fate (division, death), whereas cells with unknown fate (lost tracks, imaging was aborted) were excluded from the analysis. The life span (time from birth to division/death) of sister-sister pairs was plotted (Figure 3.9.A) and the Pearson`s correlation coefficient was calculated for each generation (data is shown in the figure legend of Figure 3.9.A). It can be seen that high correlation (r > 0.98) is shown for all generations (but slightly lower to generation 5, r = 0.885). On the other hand, the cells exhibit a large range of lifespans, ranging between 6-10 hours, with 40 hours overall. The life span of each cell was plotted as function of time (Figure 3.9.B), demonstrating that the life span increases with time. In Figure 3.9.B.i colours represent different generations. It can be seen that the time of generations is synchronized in early generations, but a large diversity is developed along generations. In Figure 3.9.B.ii colours represent different founder cells. The life spans of cells from the same founder are clustered for each founder cell, but start to spread after several generations.

Raz Shimoni Chapter 3 Lineage Informatics

Figure 3.9. Measurements of lifespan over several generations. (A) Gating on cells with known fate and plotting the life span (time from birth to division or death) of one daughter as a function of the second revealed very high correlation. Data is shown from four different experiments. (B) Plotting the life span of the data from A as function of time reveals that the life span decreases significantly with time. (Bi) Color labelling is based on generation (same colours as in A). Life span for early generations (2-4) is clustered, but the spread increases greatly as function of generation. (Bii) Color labelling is based on founder. It can be seen that the lifespan of the daughters is strongly linked to the founder, but mixes after several generations.

Raz Shimoni Chapter 3 Lineage Informatics

The next parameter to be investigated is the size of cells. Figure 3.10 shows area (which is proportional to the size of the cells) plotted as function of time (data is from one representative founder cell). Colour represents different generations (Figure 3.10.i) or different cell index (Figure 3.10.ii). It can be seen that before the divisions the area of the cells increases and then drops sharply as the cell divides. Overall, the area of the cells decreases allong generations. Furthermore, similar to the observation of heterogeneity in lifespan, early generations are characcterized by similar cell size, but during expansion, cell size becomes more heterogeneous.

Figure 3.10. Measurements of area over several generations. (A) The area of cells increases after each division, but gradually decreases as a function of generation. Although the cells originate from the same founder cell a diverse population is already evident by generation 5 (day 4). Data is shown for a representative pedigree. (B) To distinguiish between cells from the same generation the data from Figure A is shown with colour labelling on a cell basis.

Raz Shimoni Chapter 3 Lineage Informatics

The next demonstration of TACTICS is the generation of parametric lineage trees, and its utilization to map the fluorescence intensity of the CD62L fate marker, which was shown to be implicatted in memory T cells [213]. A representative lineage is shown in Figure 3.11. It can bbe seen that there is a significant difference between the lineage of daughter 1.1 and daughter 1.2.

Figure 3.11. Lineage of activatted founder cell. Imaging with low concentrations of fluorescently tagged antibodies (non-blocking and non-activating) to surface epitope mark CD62L can be used to identify putative memory precursors. (A) Two daughters of an activated CD8+ T cell were transferred to new micro-wells for tracking over four generations, and then four cells ffrrom each well were transferred to new micro-wells for further tracking. To aid visualizatiion, the consequent lineages are shoown below the earlier generations (marked by question marks to indicate that the exact cell from which the lower pedigrees were derived is not known, although whether they came from daughter 1.1. or 1.2 is known). The crosses at the end of some branches mark cell death. Blue squares mark divisions. The LUT colours represent the average fluorescence intensitty of a CD62L within the cells at each time point. A rolling average of 100 time points was applied to remove noise. It can be seen that progeniess from daughters 1.1 and 1.2 start witth a similar pattern of proliferation, but after several geneerations the lineage becomes asymmetric. The offspring of daughter 1.1 proliferated more rapiidly in the final position, with higher rate of death and a lower expression of CD62L in commparison to daughter 1.2 (B) Cell ccounts show that the progeny of daughter 1.1 divided rapidly for eight to nine generatioonns, and then rapidly decreased in number (cell death), while the progeny of daughter 1.2 is more stable.

While the siblings of daughter 1.1 exhibit faster expannsion, with a larger number of dying cells, and lower CD62L intensity, the siblinnggs of daughter 1.2 exhibit the opposite. It is possible that slowly dividing non-dying cells with higher CD62L intensity are memory preecursors (daughter 1.2), while thee other daughter cell (daughter 1.1) yields effector cells lead by a massive expansion that is followed by death and contraction. If correct, and if the pattern is reproducible in many other pedigrees (preliminary data suggests that this is the case, Mohammed Yassin, personal communication), our findings suggest the possibilityy that memory cell

Raz Shimoni Chapter 3 Lineage Informatics precursors are characterized by several generations of rapid proliferation prior to their adopting of the ‘memory-like’ attributes of slow cell cycle, small size and CD62L expression. This observation demonstrates the ability of TACTICS to track the development of heterogeneity and differentiation from time-lapse data, suggests that the properties of the founder cells have great impact on the shape of its lineage, and that heterogeneity is increased and developed during generations. The observations shown here should be confirmed using larger data sets - more data is being accumulated presently - and analyzed with TACTICS in the near future.

3.5. Summary

The reliability and accuracy of immunological models relies on the quality of the raw data, and therefore the means to achieve and extract data need to be emphasized. In this chapter was developed, utilized, and validated a new toolbox named TACTICS that encompasses most features of standard bioimaging informatics software, with several exceptional novel features dedicated for lineage informatics, such as interactive features combined with exploration tools and the ability to access the raw data. The importance of these features was recently demonstrated by a technique named dynamic in situ cytometry (DISC), which combines live two- photon imaging and features of flow cytometry [239]. According the DISC approach, the acquired images are processed, cells are being tracked, and multi-parametric information is extracted from tracked cells and converted into an FCS file, readable by most flow cytometry analysis software. Here, I presented another version of that approach whereas TACTICS Toolbox already includes both the analysis pipeline and end-point analysis. Thus, the connection between end-point analysis and raw data is better allowing adjustments and correction of each step in the analysis. Furthermore, to my best knowledge, TACTICS is the only MATLAB-based toolbox that can perform the type of analysis described here. Programming TACTICS in MATLAB allows rapid development, modularity and usability. Thus, a good platform for further development of new tools was established, and is utilized in the following chapters.

The major application of these tools that is demonstrated in this chapter is the ability to analyse cellular descriptors over several generations, with the potential to

Raz Shimoni Chapter 3 Lineage Informatics be utilized in the study of differentiation pathway of activated CD8+ T cells of memory/effector cells. This work established a platform for undergoing study, which can potentially provide invaluable information about CD8+ cell fate decisions and more insight into the mechanism that regulates T cell effector and memory generation, as well as many other cell types. The presented results raise many important questions about the role of different parameters to discriminate between memory and effector cells, and the requirement for further investigation. This ongoing work is out of the scope of this thesis and currently under investigation in our laboratory. TACTICS will continue to be central tool for this work, which holds great potential, perhaps revolutionizing our understanding of the mechanisms controlling T cell fate decisions.

Raz Shimoni Chapter 3 Lineage Informatics

Normalization of Polarization Ratios for the Analysis of Cell Polarity

4. Chapter 4

Raz Shimoni Chapter 4 Normalization of PR

4.1. Introduction

One of the most puzzling questions in the biology of T cells is the exact role of ACD in development, functionality, and fate regulation[6]. While early studies of ACD in simple multicellular organisms such as C. elegans and Drosophila have been focused upon examples where the asymmetry between the two daughters is very obvious, more finely tuned quantification is required in T cells [14].

Asymmetry in molecular localization is generally measured by fluorescent labelling of molecules within intact cells followed by fluorescence microscopic imaging. Fluorescent labelling might involve tagging of exogenously expressed proteins with genetically encoded fluorophores, or labelling of endogenous protein with fluorescently tagged antibodies. There are several approaches to measure polarity, some of which compare the geometric centre of the cell with either the geometric centre of fluorescence or the brightest fluorescent pixel [95, 243]. An alternative approach, commonly used for measuring ACD, compares the total fluorescence from each half of the cell, often by deriving ratios of fluorescence in the two halves of the dividing cell [34, 35, 244, 245]. For this type of analysis, it is assumed that the ratios are proportional to the distribution of the molecules under investigation. The ratiometric approach has two advantages for ACD. First, the total fluorescence in each half is presumably more physiologically relevant than the other patterns of fluorescence within the cell, and should directly relate to the inheritance of those fluorescent molecules. Second, such measurements can be continued beyond the point of cell division in time-lapse imaging, making it more broadly useful for determining the functional consequence of ACD. Many variations of this approach have been implemented, such as comparing fluorescence along a line scan rather than using the total fluorescence, or measuring only nuclear asymmetry [246]. After deriving polarization measures in dividing cells, each event is then sometimes ascribed as Symmetric Cell Division (SCD) or ACD by arbitrarily assigning a cut-off value, with ratios above this arbitrary value considered asymmetric.

A ratiometric approach is only viable if the ratios that are derived from the fluorescent intensities are an accurate reflection of the ratios of protein in the two halves of the cell, and this has not previously been formally tested. Possible artefacts

Raz Shimoni Chapter 4 Normalization of PR that might lead to inaccurate ratios include: the acquisition settings (such as detector gain, fluorescence excitation power, scanning parameters, fluorophore properties, and more), and intensity variations contributed from instrumental precision limitations such as SNR [247, 248]. Additionally, post-acquisition image processing such as background subtraction, spectral unmixing, and averaging algorithms can directly influence the fluorescence measurements in a nonlinear fashion [75]. To assess the reliability of quantitative fluorescence analysis, biologists can use internal controls, such as the parallel imaging of a molecule that is known to divide symmetrically [249]. Such an approach estimates the noise contributed from imaging artefacts, such as uneven illumination or cell alignment (i.e. when the two halves of the cell are in different focal planes) [250]. However, because the fluorescence in the second channel is collected and processed differently to the channel of interest, this would not control for other acquisition and processing artefacts.

As described in this chapter, I developed a new normalization approach to control for the effect on polarization ratios of non-biological factors such as image settings and post-acquisition analysis. Section 4.2 describes the experimental and computational methods for this chapter. The results of this chapter are organized in Section 4.3 as the follows. To gain more insight into the effects of image processing on quantification of fluorescence ratios, and to develop new standardized approaches to correct for systematic differences that do not represent biological variation, I performed a comprehensive analysis of symmetric cell divisions under changeable threshold levels. Particularly, I utilized eGFP-expressing cells (section 4.3.1) and computer simulations of variations in fluorescence ratios (section 4.3.2) to demonstrate that image settings alter polarization measurements, and that clustered localization is more susceptible to artefacts than homogeneous localization. To accommodate these possible artefacts, I established a new approach that incorporates the PRminor in the assessment of polarity in section 4.3.3. In section 4.3.4 I investigated the effects of clustered fluorescence by comparing the PRmajor and PRminor, and assessed the value of binarization in the analysis of polarity in Section 4.3.5. In Section 4.3.6 I described the development of methods to incorporate the PRminor in gating approaches. In Section 4.3.7 sensitivity tests from simulations were performed, confirming the usability of our approach. Section 4.3.8 describes the

Raz Shimoni Chapter 4 Normalization of PR overall strategy for polarity analysis. In Section 4.3.9 I described the TACTICS ACD Module that implements the PRminor for better analysis of ACD from time-lapse data. Finally, Section 4.4 summarizes the chapter.

4.2. Methods

4.2.1. Cell microwells and cell culture

Cells were cultured in microfabricated microwells for cellular studies in vitro [128]. The cell microwells were made of PDMS with dimensions of 125 x 125 x 45 µm and with well-defined vertical sidewalls and a transparent base. Cell microwells were placed into a well of an 8 well chamber slide (LAB-TEK II, NUNC) sterilized with 100% EtOH and UV light, and rinsed with media prior to use. The gibbon ape leukaemia cell line, MLA-144[2], eGFP were cultured at 37C, 10% CO2 in Dulbecco’s Minimal Essential Medium (SECF) supplemented with 10% (v/v) fetal calf serum, L-glutamine (1mM) and 100ng/mL penicillin/streptomycin.

4.2.2. Time-lapse microscopy

Time-lapse images were obtained with IX71 inverted microscope (Olympus, Tokyo, Japan) equipped with a Nipkow disk-type confocal unit (Yokogawa CSU22, Tokyo, Japan) and EM-CCD Andor camera (Model: iXon EM +885, Belfast, Northern Ireland). Images were acquired in both DIC and green channels using a 20x air objective 0.45NA, which corresponded to a pixel size of 0.33 µm x 0. 33 µm. The working distance was 6.6-7.8 mm. Exposure time was 600 ms for green and 100 ms for DIC. Multiple stage positions were captured (controlled by MetaMorph software version 7.7.11.0 ) with a sampling rate of 1 minute for 1-24 hrs, and were saved as 8- bit 2-D arrays (1002 × 1004 pixels).

4.2.3. Computational platform

Synthetic images were generated and both synthetic and real images were analyzed using MATLAB® R2012b version 8.0 (the MathWorks, Inc., Natick, MA, USA) with the MATLAB IPT. Calculations were performed on an HP Z400 workstation equipped with a 3.3 GHz Intel Xeon W3580 Quad processor and 16 gigabytes of RAM working under a Windows7 64-bit operation system.

Raz Shimoni Chapter 4 Normalization of PR

4.2.4. Processing of time-lapse data

Divisions were split by minor and major axis. Pixels of the two corresponding sides of the cells were copied to reconstruct two new images, where each image represented one half of the cell. To define the major axis, a line-based Bresenham algorithm [251] was stretched across a defined axis to split the cell image through the centre of the cell into the two opposite pixels located on the cell perimeter. The minor axis was perpendicular to and passes through the midpoint of the major axis. When cells are close to each other it is impossible to accurately separate the pixels. Therefore, automated selection for splitting was applied by fitting the cell shape and area to a circle to find the angle of overlap . When  was smaller than 37 degrees, instead of trying to match from which daughter the overlapping pixels belong, the splitting was made so that an even number of pixels was contributed from each daughter. Since the use of subtraction techniques such as threshold has been shown to change the apparent size measurements [252], cell borders were kept constant and the initial detected borders in (T=0) were used when screening the effect of T value on PR.

4.2.5. Calculation of Polarization Ratio (PR)

Throughout this study the Polarization Ratio (PR) values were calculated as the fluorescence difference between the two halves of the cells (split by the major and minor axis as shown in Figure 4.1.A) using Equation 2:

D12 D PR  D12 D where D1 and D2 are the two halves of the cell, which can be either the mean or integrated pixel intensity, depending upon the experiment. To calculate the PRmajor the two halves are daughter 1 and daughter 2 (Figure 4.1.B.ii). To calculate the PRminor the two halves are the two sides across the minor axis (right and left in Figure 4.1.B.iii). Possible ranges vary between 0 (indicating maximum symmetry) and 1 (indicating maximum asymmetry).

Raz Shimoni Chapter 4 Normalization of PR

4.2.6. Calculation of binarization

The binarization of each division as symmetric versus asymmetric was achieved by counting only divisions that were polarized more along the major axis than along the minor axis:

Equation 3:

x1, if PRminor  PR major f ()x  {x0, else

 f (xi ) %Binarized  100 i n where f is result of specific binarized event, n is the number of diviisions, and %Binarized is the percentage of division identified as polarized.

Figure 4.1. Axial subdivision fof r polarization analysis. An approach for quantification of polarity. (A) Black arrows represent the major and its perpendicular minor axis. Splitting the image into two allows a direct comparison of fluorescence intensity across the minor or major axes. The major axis is derived from the longest diameter of an ellipse that overlaps the cell. The minor axis is defined as the perpendicular to the major axis. Blue and red colors show the areas from which pixel intensities were collected, and represent the two halves of the cell that would, if the cell divided, become daughter 1 and 2. The left and rigght sides of the cellss are bright and dim respectively. (B) Polarization ratios are extracted by integrating pixel intensities across the major or minor axis: (B.i) Ratio along the major axis (using segmentts divided by the minor axis). (B.ii) PRmajor : Normalized ratio along the major axis (across the minor axis). (B.iii) PRminor: Normalized ratio along the minor axis (across the major axis). (B.iv) Log ratio (major axis). (B..v) Log ratio (minor axis).

Raz Shimoni Chapter 4 Normalization of PR

4.2.7. Shape model of cell division

I developed a simplistic model that describes an imaged dividing cell as a circle and the two daughter cells as two overlapping ovals (Figure 4.2.A). To isolate the effect of asymmetric protein localization, the most important feature of the dividing cells is symmetrical morphology, but the exact shape of dividing cells is not important for the purpose of this study. Synthetic images of cell division were created by filling the perimeter of the simulated cells with pixels of value one, using the MATLAB function roipoly.m. To resemble real cell divisions the perimeter of each division was modelled as two overlapped polygons allocated around the same centre. The left (Equation 4) and right side (Equation 5) polygons were created in x and y plane using the parametric equations:

Equation 4:

p p xrcos( s12 ) s 11100 1 100 p p yrsin( s4 ) s 5 11100 1 100

0 1

2 2 

Equation 5:

pp xrcos( s12 ) s 21100 1 100 ppp ys()sin() 35 r  s4  s 22100 11100 100

0 2

1 2  where r is the equivalent radius of the polygons,  is the angular coordinate, s1=0.12 and s2= 1.1 are constant correction factors that were fitted into real data to give a slight stretch of the major axis along the x-axis (Figure 4.2.B). To generate various combinations of random shapes the model parameters included p1, p2, p3, p4, p5 random values from the standard uniform distribution on the open interval (0,1). The equivalent radius was estimated as function of the overlap between the two polygons([253]):

Raz Shimoni Chapter 4 Normalization of PR

Equation 6:

1/3 1 2 rR() 1 cos(3cos)  2

 0 2 where R is the radius of the parental cell before the creation of the furrow, θ is the angle between x-axis and the intersection point between the two circles.

In a real division, θ is a function of the division stage. When θ is equal to zero the two polygons completely overlap to represent the pre-divided parental cell, and as θ increased, the overlap decreased simulating a later division stage. All images were in size of 201 by 201 pixels, where the initial coordinates of the centre of each polygon were at (0,0). The shift of each cell from the centre was equal to half of the distance between the centroids of each daughter cell d, and was determined using a derivation of the Law of cosines:

Equation 7:

drr22222) rcos2 ( dr 2(1cos(2))2  

2 with correction factor: ds3 2 r (1 cos( 2 ))

 0 2 where the term (2)   denotes the angle contained between the two radii of the cells and opposite to d. Notably, this equation is valid only when the two radii of the cells are equal. The S3=1.1 is correction factor that fitted into real data in similar to s1 and s2.

Raz Shimoni Chapter 4 Normalization of PR

Figure 4.2. Binary image generation of cell division. (A) Diagram showing the geometry to calculate coordinates of the cells as function of θ. (B) Comparison of simulated shape and an image of a dividing cell. The magenta contour showing the goodness of the fit of the polygons equation compared with detected cell edges. Time points are 30 seconds between frames. θ was manually adjusted to give better fit. The R was 22 pixels.

4.2.8. Generation of synthetic images

Binary cell images were generated using a simplistic model of dividing cells as explained in Section 4.2.7. My simulations exemplify non-cllustered molecules, such as eGFP in addition to clustered molecules such as nanopparticles [246, 254], protein aggregates, [255] or mRNA [243] that are characterized by speckled density spots. The texture of non-clusttered, freely distributed fluorescence was generated using MATLAB implementation of Perlin texture created by Antti Lehmussola [256].

The MATLAB function spotmaker.m (written by Tristan Ursell, 20123) was used to generate clustered localization. Multiple clusters were simulated as diffraction-limited spots with noise. The initial position of the spots was randomly distributed, and for sequential frames the location was randomly mmoved in the x and

3 Downloaded from: http://www.mathworks.com/ MATLABcentral/fileexchange/36026-create-a- simulated-image-of-diffraction-limitedd-spots-with-noise/

Raz Shimoni Chapter 4 Normalization of PR y direction to simulate Brownian motion. To mimic the fact that fluorescent images are frequently derived from a projection of the 3-D volume of the cells, I used a 2-D radial intensity distribution with exponential slope over the polygon from its centre:

Equation 8:

r  Ir() e3 where r is the pixel coordinates around the origin (centroid of the cell). Finally, further imaging effects were added to give more realistic characteristics. To simulate the Gaussian point spread function (PSF) typical of confocal fluorescent images, the clustered images were filtered using the MATLAB function imfilter.m with a Gaussian filter.

To simulate a blurred effect for the non-clustered expression I used a disk filter. Both filters were generated by the MATLAB fspecial.m function. Realistic simulated data that contained some degree of noise was generated with skewed polarization ratios. For this aim, MATLAB imfilter.m function was utilized to convolve a Gaussian filter only to the part of cell that is out of the focal plane. To simulate white noise and detection response, Gaussian white was added using the MATLAB function imnoise.m.

Finally, each division was aligned along the major axis and the daughter cells categorized as daughter 1 for the top and 2 for the bottom cell, positive for right side and negative for left side. However, this labeling was only used to separate between the cells and had no effect on the analysis as the cells were randomly flipped to remove any potential systematic bias. The MATLAB source-code and simulated data are attached to this thesis, and were also uploaded online.

4.3. Results and discussion

4.3.1. Polarization ratios are sensitive to image processing settings

First, I tested whether polarization ratio measurements are settings-dependent using the MLA-144 T lymphocyte cell line expressing freely diffusing eGFP, which should always be distributed symmetrically. I focused on polarization during cell

Raz Shimoni Chapter 4 Normalization of PR division, where morphology allows for straightforward designation of the two polarity axes. Cells were tracked, and the frames in which the shape of the dividing cell most closely represented the telophase stage were extracted for analysis. The major axis was used to assess polarization along the mitotic spindle (Figure 4.1.A). Thresholding to remove background fluorescence is routinely used in image analysis, but from the analysis of cDNA microarrays it is known that intensity ratios are influenced by image processing [257] and the level of thresholding [258]. Therefore synthetic images were generated to explore the alterations in contrast that are frequently introduced during image acquisition or post acquisition processing, whereby pixels with intensities of a value less than the threshold value (between 0 and 90% of the value of the maximum pixel intensity of the cell) are either omitted [259] or set to 0 [260]. Similar effects can also be introduced experimentally during the image acquisition, for instance, by reducing the detection offset or the exposure time [261]. The threshold levels in the simulation vary between 0 to 90% of the value of the maximum pixel intensity of the cell (For additional data about the simulations see Sections 4.2.7-4.2.8 including Figure 4.2 and Figure 4.3).

Raz Shimoni Chapter 4 Normalization of PR

Figure 4.3. Simulations of cytokinesis. (A & B) Representative simulated imaages of cells (synthetic images) for clustered and non-clustered, symmetric and asymmetric as shown in 4 different division stages. ACD was simulated by introducing different mean of flfluorescence value (non-clustered) or number of clusters between the two daughter cellss. (C & D) Intensity projections along the major axis of each cell over 100 frames. Visualizattion of such heat maps can display qualitative differencces between the two daughter cells. (A) shows representative synthetic images of cells with non-clustered localization undergoiing division at four different developmental stages: 90, 64, 28, 0, degrees. Cells are aligned along the major axis of each division. The ratio of total pixel intensity in the two daughters was 1:1 in the symmetric and 1:1.5 in the asymmetric cell division. The magenta colored contours show how cells were subdivided and the border of the cells. Arbitrary units show projected sum intensity, and pixel intensity is color coded aaccording to the color bars. The parreental radius before division was 22 pixels. The average width of the clusters was 1.5 wiith standard deviation 0.02. For the aim of visualization, images were cropped to remove part of the image sides that do not include part of the cells. The increased intensity in the centre of the cells is due to the radial distribution function that was introduced to mimic vollume effect. (B) Shows representative synthetic images of a cell with clustered localization undergoing division presented in similar to (A). The numbers of clusters in the two daughters was 40 in the symmetric and 40 and 20 in the asymmettric cell division, which correspondss to ratios of 1:1 and 1:1.5 respectively. (C) Shows a 2-D plot showing the progression of asymmetry/symmetry fluorescence partitioning of the same cell shown in (A). Each column

Raz Shimoni Chapter 4 Normalization of PR represents the total fluorescence intensity of individual cells for a simulated division. The projections are composed of 100 sequence frames that represent the progression of the cell division, whereas θ varies from 0 to 90 degrees in even increments. The representative images in (A) (90, 64, 28, 0, degrees) that corresponding to frames 1, 30, 70, 100 in respectively. (D) Shows pixels with values lower than 140 (A.U.) were discarded to remove background and basal intensities from the projections.

I tested the effect of removal of background fluorescence on two ratios that have previously been used to assess polarity in hematopoietic cells: a simple proportion between the integrated fluorescence of the two halves of the cell [35](Figure 4.1.B.i) and a normalized Polarization Ratio (PR), calculated as the fluorescence difference between the two halves divided by the total fluorescence (Figure 4.1.B.ii-iii). In contrast to the ratio in Figure 4.1.B.i, the PR accounts for potential artefacts that might be introduced where cells have heterogeneous expression levels, and has previously been used to calculate polarization ratios in cell-cell interactions [233], dividing [34] and migrating T cells [2]. PRmajor describes polarization of relevance to ACD in these simulations (but could equally reflect polarization during migration), and PRminor describes polarization along the axis perpendicular to the major axis. Because eGFP is uniformly distributed in cells, it is expected to give low PR. However, although ratios calculated in both ways are low (compatible with SCD) at threshold values of up to 65%, the ratios increase dramatically at threshold values greater than 65% for many of the cells (Figure 4.4). In extreme threshold values of T=90%, some cells exhibited ratios larger than 1.5 (Figure 4.4.A), a cut-off value that has been used for categorization of ACD [35, 37]. Similar trends were observed using the PR (Figure 4.4.B). These data indicate that current approaches to quantification of polarization during division, which generally do not incorporate a systematic approach to background subtraction and contrast enhancement, run the risk of miscalculating polarization.

Raz Shimoni Chapter 4 Normalization of PR

Figure 4.4. Ratio coefficients from experimental data are dependent on the threshold value. A striking example is the large variance of calculated polarization ratios from a cell division that is known to be symmetric caused by altering the image threshold. Flluorescence intensity ratios were extracted from images of dividing MLA cells expressing eGFP, and for a random selection of 10 events, ratios plotted again threshold setting. Ratios werre calculated using Proportion (A) and PR (B). Different colors represent different eventts. The PR factorize for the total fluorescence in the denominator, which is important in particularly when comparing expression in heterogeneouus population. For instance, the MA plot visually represent DNA microarray gene expression, drawing on similar approach of log ratio versus the mean expression [262]. In addition, since the value range in PR is between -1 and 1 (or between 0 and 1 for absolute PR), the PR migght be more suitable ratio form for analysis than a simplee proportion.

4.3.2. SSimulations indicate that the deegree of clustering of fluorescence alters the effect of thresholdingn

In addition to the impact of acquisition and analysis settings ass discussed above, many intracellular molecules are not distributed as homogeneously as eGFP, which might lead to further artefacts. To assess this, I simulated cell diviisions with fluorescence in either clustered (representing protein aggregates, nanoparticles, or endosomal proteins [263, 264]) or non-clustered (representing diffusely distributed proteins such as eGFP) patterns, each with both symmetric and asymmetric distribution. PR values for 10 simulaated symmetric divisions of non-clustered

Raz Shimoni Chapter 4 Normalization of PR fluorescence (Figure 4.5.A) over different threshold values showwed a similar pattern to the eGFP-expressing cells (Figure 4.4.B), with threshold levels of greater than 80% (Figuure 4.5.A.iv), yielding high PR values that could be falsely interpreted as ACD. Simulations of symmetrically dividing cells with clustered fluorescence (Figure 4.5.B) showed a further increase in sensitivity to thresholding, with threshold levels over 40% yielding PR values of 0.1. At extreme levels of thresholding (of more than 80%, Figure 4.5.B.iv) all pixels that remained were from the original clusters. At this leevel of thresholding, PR values could exceed 0.3 (corresponding to an absolute raatio of 1:1.5), despite the fact thhat the division was symmetric. Thus, the simulations demonstrate an influence of image processing settings on the measured ratios that is similar to that observed with the eGFP- expressing cells, indicating theeir suitability for further analysis. Furthermore, the simulations indicate that cells with clustered fluorescence are stiill more susceptible to artefacts, further highlighting the need for a systematic method for image processing and background subtraction.

Figure 4.5. Ratio coefff icients from synthetic data are dependent on the threshold value. 10 images of cells simulated to have symmetric fluorescence partitioning with non-clustered (A) and clustered localization (B) were simulated to quantify the effect of background subtraction on ratio coefficients. The horizontal axis represents the Threshold value. The vertical axis represents the PR. The possible range varies between -1 to 1, where 0 is maximum symmetry and high absolute ratios indicate maximum asymmetry. Different colors represent different simulated divisions. Black crosses represent the ratio and T value of illustrative images below. Input for simulation: =0, R=22 pixels, number of clusters in both D1 and D2 was 20.

Raz Shimoni Chapter 4 Normalization of PR

4.3.3. Calculation of the polarity across the minor axis provides a useful control to assess noise and to normalize calculations of asymmetry

As described in Chapter 3, I have previously identified a unique characteristic of most polarized cells that provides an opportunity for improved analysis [2]. Namely, that true asymmetry should be evident across one axis, but should generally not occur across the perpendicular axis (Figure 4.1.A). In contrast, asymmetry that is caused by artefacts in the image acquisition or processing should be evident across both axes. The principle of my method is that settings applied during and after image acquisition will cause similar skewing of the ratio across both axes. The ratio across the minor axis can therefore be used to estimate the noise in the PR measurements (Figure 4.6 and Figure 4.7). To assess whether this approach might enable a systematic method to select image processing settings, and to determine the influence of background removal on minor and major ratios, 10 cells were simulated (representative cells are shown in Figure 4.6.A) for: (i) symmetric non-clustered; (ii) asymmetric non-clustered; (iii) symmetric clustered; (iv) asymmetric clustered cell, and the PRmajor and PRminor were compared (Figure 4.6.B). Symmetrically dividing cells exhibited low PR values for low threshold settings, and the PR increased slightly at higher settings. Importantly, the increase in PRmajor at high segmentation settings was similar to the increase in PRminor, indicating that these polarization measurements reflected artefacts in image processing rather than genuine asymmetry along the major axis. The simulations of asymmetrically dividing cells showed a strikingly different pattern, where the PRmajor and PRmajor were clearly different from each other. In non-clustered fluorescence (ii), PRminor was low (below 0.05) at low threshold settings, and increased slightly at higher threshold settings (to a maximum of 0.2). In contrast, the PRmajor hovered around 0.2 for low threshold values but increased dramatically at high threshold values, plateauing at 1.0 by 60% thresholding (where thresholding can lead to aberrant values, rendering the data meaningless). These data indicate that the pattern of PRminor response to threshold settings provides an opportunity to set the appropriate conditions for measurement of PR, and can be used to objectively ascribe an appropriate threshold value for analysis of the PRmajor. For instance, the appropriate threshold value for analysis could be

Raz Shimoni Chapter 4 Normalization of PR defined as the value just below that at which PRminor increases by 10% above the baseline (50-60% in this instance).

Figure 4.6. PRmajor and PRminnor are differentially affected by thresholding and clustering. Cells were simulated to be: (i) symmetric non-clustered; (ii) asymmetric non- clustered; (iii) symmetric clustered; (iv) asymmetric clustered. (A) Examples of simulated cells and approach to hemisphere separation. Blue and red lines describe the major and minor axes respectively, and the magenta contour shows the separation that gave an equal number of pixels to each hemisphere (slightly shifted from the major axis). (B) PRmajor and PRminor values for 10 simulated ceells were plotted against T value (C) PRmajor and PRminor values for 10 simulated cells were plotted against T value. Note that in the asymmetric cells, some fluorescence values were reduced to 0 for the higher threshold settings, causing misleading values of 1 in (B) and infinite (unplottable) values in (C). Input for simulations: θ=0, R=22 pixels, number of clusters in D1 and D2 was 20 in the symmetric and 20 and 30 in asymmetric. Intensities histograms corresponding to the representatiive data are shown in Figure 4.7.

Raz Shimoni Chapter 4 Normalization of PR

Simulated cells in which the fluorescence was clustered showedd a similar, although more noisy, trend, with low PRmajor and PRminoro values across the thresholding range for cells with symmetric distribution of clustered mmolecules. However, the PRmajor was clearly different from PRminor at higher threshold levels for asymmetric distribution of clustered molecules. In microarray analysis, the log ratio has proved to be as good as or better than the PR value for comparing pairs of fluorescent values, thus I determined whhether this approach might also be useful for polarity measurements by deriving log ratios across the major and minor axis (Figure 4.6.C). The log ratio also provided good discrimination betweenn symmetry and asymmetry, which again was highly dependent upon the thresholdinng. The log ratio appproach did not present any obvious advantages over the PR, and since the PR is more established in cell polarity measurements, I focused on PR for the remainder of this study. Together, these data confirm that the PRP minor could be used to assess the reliability of ratiometric measurements and to select an appropriate thresholding setting for analysis of PRmajor.

Figure 4.7. Intensities histograms corresponding to the Figure 4.6. Intensityy histograms of 4 simulated divisions that corresponding to the 4 cases illustrated in Figure 4.6 A-D. Pixel intensities for the two daughter cells (D1 and D2) and for the two subsections of tthe division (L and R) were set in intensity histograms starting from the maximum pixel intensities (255) to the minimum (0). The blue and red lines represent the two opposite sides.

Raz Shimoni Chapter 4 Normalization of PR

4.3.4. Comparing PRmajor and PRminor demonstrates the effects of clustered fluorescence

Having shown that a clustered pattern of fluorescence increases the PR (Figure 4.6), I used the differences between PRmajor and PRminor to assess the degree of clustering for which ACD could be reliably assessed. I simulated divisions of cells containing from 1 and 100 clusters (increments of 1), and with fold-differences in the number of clusters from 1 (symmetric along the major axis) to 2 (asymmetric along the major axis) with increments of 0.1 (Figure 4.8, two examples shown on left). PRmajor and PRminor were displayed in heat maps (Figure 4.8.A right hand side), calculated under 0%, 20%, 40%, 60%, and 80% threshold values. Again, increasing threshold levels resulted in an increase in both PRmajor and PRminor, but particularly promoted an increase in PRmajor. As expected, at high numbers of clusters, the data appeared similar to the unclustered analysis above: PRminor was aberrantly high at thresholds of 80%, but at T=60%, PRminor was reasonably low and PRmajor showed a good dynamic range. At T=60%, PRminor was reasonably low even as the cluster number diminished to approximately 8-10 clusters per cell. However, below 8-10 clusters per cell, PRminor was aberrantly high, as was the PRmajor for simulations in which the symmetry across the major axis was close to 1, which should not have yielded high values. These data suggest that clustering can still yield reliable polarity ratios, as long as the number of clusters exceeds 8-10. Evidently, different data sets might be more or less affected by cluster number, and I was conservative in simulating only weak polarity (2 fold differences across the major axis), so the analysis of more extensive polarity should be even more robust. Nonetheless, these data both demonstrate that clustered fluorescence can yield robust polarity analyzes if care is taken to ensure the best settings, and indicate the value of PRminor to assess the noise across a non-polarized axis and ensure appropriate quantification.

4.3.5. Assessing the value of binarization in the analysis of polarity

In many biological situations, a population of cells can comprise both symmetric and asymmetric divisions [265], in addition to divisions that were imaged or processed at wrong settings. In such cases, it is desirable to measure the degree of

Raz Shimoni Chapter 4 Normalization of PR asymmetry of both the population and of individual cells, and, in some instances, also to binarize such data and to designate each cell as symmetric or asymmetric.

Figure 4.8.The effect of cluster number on the accuracy of PR measurementss. Divisions were simulated to have increasing numbers of clusters ranged from 1 to 100 in inncrements of 1 for one of the daughter cells. The number of clusters in the second daughhtter was the number of clusters in daughter 1 multiplied with its corresponding ratio. Ratios vary from1 to 2 with increments of 0.1, and were calcculated under 0%, 20%, 40%, 60%, and 80% threshold. (A) The PRmajor and PRminor for eaach event are shown in heat maps, where the PR ranging from 0 to 1 are represented in "jet" colors (blue to red). The ratios were binarized using (B) cut-off value of 1.5 or (C) by Equation 2. Blue pixels represent events in which PRmajor was larger than PRminor; white pixels represent events in which PRmajor was smaller than PRminor.

Raz Shimoni Chapter 4 Normalization of PR

Binarization has previously been achieved by assigning a cut-off value and binning all events above the cut-off as polarized, and all events below the cut-off as non-polarized [34, 35]. Although convenient, such an approach clearly has the potential to introduce errors, as it does not account for the distribution of polarization ratios, the potential overlap of ratios amongst two subpopulations, or the errors in individual events. I utilized the results of Figure 4.8.A as ground truth data to formally assess the accuracy of the cut-off approach, and binarized as ACD or SCD using a cut-off of 1.5 (as has previously been used, [35-37]) (Figure 4.8.B). The number of events ascribed as ACD increased as the ratio increased (quantified in the histograms below each heat map), and that this was most evident at 60% thresholding (previously shown to be the most appropriate setting). However, the data was noisy, with many blue events (ACD) in the left columns (symmetry), and many white events (SCD) in the right columns (asymmetry). For this simulation, at the most appropriate threshold (60%) approximately 28% of the cells that were simulated to have a ratio of 1.0-1.3 were scored as ACD (i.e. 28% false positives), and 82 % of cells that were simulated to have a ratio of 1.7-2.0 were scored as ACD (i.e. 18% false negatives).

This analysis indicates that simply deriving a ratio and ascribing a cut-off can lead to error in the designation of divisions as ACD or SCD, and that there is no value that would effectively discriminate between high and low polarization ratios. I then assessed whether PRminor might enable an alternative approach to binarization of the data. The analyzes described above use PRminor to assess the noise in populations of cells and to optimize processing settings, but PRminor of individual cells also has potential value in the analysis of polarization of the population. I therefore used the data set in Figure 4.8.A to determine the value of binarizing by simply comparing PRmajor with PRminor. ACD was ascribed to cells in which PRmajor was greater than PRminor, and SCD was ascribed to cells in which PRmajor was less than or equal to PRminor (Figure 4.8.C). At all thresholding settings the number of events ascribed as ACD increased with increasing polarity, but as with the cut-off approach above, false positives (1:1 simulations ascribed as ACD) occurred even under optimal conditions (60% thresholding, high numbers of clusters). The number of false negatives (2:1 simulations ascribed as SCD) was low across all thresholding settings, and these

Raz Shimoni Chapter 4 Normalization of PR were only evident for simulations with low numbers of clusters. Compared with the cut-off approach (Figure 4.8.B), at the 60% thresholding levels, there were more false positives (54%), but fewer false negatives (5%) and the binarization was far less dependent upon thresholding than the cut-off approach. Interestingly, visual inspection of binarization plots indicated that the accuracy of binarization was much more dependent on clustering for the PRmajor versus PRminor approach than for the cut- off approach (presumably, because artifacts due to clustering can have twice the impact). These data indicate that binarization using either of these methods can be useful for comparing between two populations, that this approach is more appropriate for non-clustered data, and that events that are ascribed as SCD are most likely truly symmetric. Most importantly, these data indicate that binarization is not a reliable indicator of the number of truly asymmetric events, and the designation of an event as symmetric or asymmetric requires a more case-by-case approach. This analysis could potentially be further improved by calculating the confidence interval (e.g 95% confidence) for PRminor, which could be used to more rigorously define ACD by defining ACD as including only those events where PRmajor>PRminor + CI. The advantage of this approach is that it would increase the selectivity as it reduces the number of false positives (i.e. by 5%, or 1 % if 99% confidence interval is used) because of the noise in the calculation (which is reflected in the variance of PRminor). On the other hand, the disadvantage of this approach is that it reduces the sensitivity. In addition, in many experimental setups it is difficult to assess its quality due the lack of True Positives (TP). This leads to the next section, which focuses on methods to incorporate PRminor in the assessment of polarity.

4.3.6. Methods to incorporate PRminor in the assessment of polarity

The simulations in Figure 4.8.A indicate that, even after careful selection of thresholding values, PRmajor is still noisy, particularly for clustered fluorescence. To assess whether PRminor could be used to improve the analysis, I adopted a plot of PRmajor versus PRminor as used previously to explore the relationship between PR values [2]. Next, I assessed whether PRminor could also be used to remove aberrant events and so improve the quality of the analysis, by comparing the PRmajor vs. PRminor plots from five different thresholding values (Figure 4.9.A). These simulations incorporated ground truth "bad data" to assess the value of this approach:

Raz Shimoni Chapter 4 Normalization of PR

Each division was simulated to have the probability of p=0.025 that one daughter cell was out-of-focus, one side of the two cells was out-of-focus, one side of one of daughter was out-of-focus, or only one side of one of daughter was in focus. In total, 10% of the data is expected to contain an out-of-focus artifact. The simulations were designed to represent a mix of symmetric and asymmetric events, with minimal events of intermediate polarization. As in flow cytometric analysis, visual inspection of the PRmajor vs. PRminor plots provides much information: (i) a small number of outliers were evident (simulated as "bad data" such as poorly segmented or imaged cells), which can then be discarded from the analysis, (ii) a majority of events with low PRminor were clearly suitable for analysis, and (iii) a clear distribution into two populations of PRmajor, representing symmetric and asymmetric divisions could be seen. The "bad data" was removed by gating out the top 10% of PRminor values, and, PRmajor and PRminor were represented as histograms (Figure 4.9.B). In this instance, the distribution of the histogram allows clear discrimination between the two populations, making allocation into ACD and SCD populations both facile and valuable.

Raz Shimoni Chapter 4 Normalization of PR

Figure 4.9. Gating based on PRminor in simmulated data. Cell divisions were simulated with ratio 1 (blue color) or 1.5 (red color), 100 divisions for each, with 10% of out of focus "bad data" and PRmajor (y-axis) was plotted against PRminor (x-axis) (A) for non-clustered (i) and clustered (ii) fluorescence. The magenta linne show the gating border that to removes 10% outliers of the minor axis polarization, to remove "bad data". PRmajor of the gated eevents were plotted as a (B). Input for simulations:  wass chosen randomly varying from 0 to 90 degrees, distribution of parental radius and total intensity were selected randomly from real distribution from real data, number of clusters in one of the daughter cells was 20 to 100, and the number of clusters in the other daughter cell was multiplied in the simulated ratio giving possible range from 30 to 150.

In most real scenarios, there would probably not be a clear discrimination between the two subpopulations, and binning into ACD and SCD would likely require approaches such as those described above, with the accompanying issues of accuracy. These data illustrate that the comparison of PRmajor and PPRminor on individual events using a 2-D scatter plot provides and effective means too assess the quality of the data, gate out artefacts, and assess the distribution of polarity across the population. Conversion of the data to 1-D histograms allows for compariison across populations, and provides the basis for deecisions as to whether, and how, each event might be binned as ACD or SCD.

Raz Shimoni Chapter 4 Normalization of PR

4.3.7. Sensitivity test from simulations

The above analysis illustrates the difficulty of designation of events as ACD or SCD. An ideal experimental situation would contain known symmetric and asymmetric controls for comparisons with the test samples, and provide a basis for determining the best binarization approach, but such controls are not always available. The analysis performed in Figure 4.9 assumes that there are two populations that can be relatively easily separated to two clusters. However, in some cases the classification can be much harder. To study the value of our approach in samples without the clearly defined two populations described in Figure 4.9, I simulated a heterogeneous population of 1100 divisions in which the proportional ratio between the two daughter cells, varied from 1 to 1.5 fold with 100 cells representing each increment of 0.04. Scatter plots of PRmajor and PRminor for each simulated division at threshold values of 0%, 20%, 40%, 60%, and 80% (Figure 4.10.A and 4.10.B for non-clustered and clustered respectively) showed that 60% thresholding provided the most suitable dynamic range for PRmajor. The gate excluded all events in the top 10% of PRminor values (right of the pink line). For non- clustered simulations, color coding (each color represents simulations within a 0.05 window of ratios) indicates that the gated events generally yielded appropriate PRmajor values. These data indicate that gating removed some mid-range events that would otherwise have been allocated an inappropriate PRmajor (out-of-focus events were not simulated in this experiment, but should have also been reduced by this process). Gating out the top 10% of PRminor was particularly important for the analysis of clustered events, where PRmajor was appropriately distributed according to color for the low PRminor events, but inappropriately distributed in the high PRminor events (see histograms in Figure 4.11). Combined with the analysis of out-of-focus events in Figure 4.10, these data provide strong support for the value of plotting PRmajor against PRminor to exclude aberrant data. In addition, the finding that PR values distributed according to original input ratios (as illustrated by the color distribution in Figure 4.10), indicates that this approach can discriminate between increments of polarity, providing greater value than a mere allocation as asymmetric or symmetric.

Raz Shimoni Chapter 4 Normalization of PR

Figure 4.10. Sensitivity test based on siimulations. Cell divisions were siimulated in increasing ratios from 1 to 1.5 with even inncrements of 0.05, giving n=1100 divisions in total. PRmajor was plotted against PRminor for non-clustered (A) and clustered (B) data. Data is showed as major/minor plot, under a range of thresholds from T=0%, to T= 80% in increments of 20%. The magenta line showws gating exclusion of 10% of data with the highest PRminor. The gate shifts right as T value increase. The colours in the figure legend represent different ratios and corresponding to the ratio colour of each data dot. Input for simulations:  was chosen randomly varying from 0 to 90 degrees, distribution of parental radius and total intensity were selected randomly from a real distribution from real data, number of clusters in one of the daughter cells was 20 to 100, and the number of clusters in the other daughter cell was multiplied in the simulated ratio giving a possible range from 30 to 150.

Raz Shimoni Chapter 4 Normalization of PR

Figure 4.11. 100 cell division were simulated in increasing ratios from 1 to 1.5, giving 1100 divisions in total as represented in "jet" colors (blue to red). Data corresponds to Figure 7, non-clustered (A and B) and clustered (C and D). Data is showed counting histogram of PRminor (A and C) and PRmajor (B and D), under range of threshold from T=0%, to T= 80% in increments of 20%. Magenta line shows gating exclusion of 10% of data with high PRminor. Confidence intervals and the meddian distribution are color coded for each corresponding simulated ratio, and can be used to compare between the distribution of PRmajor and PRminor.

4.3.8. Strategy for polarity analysis

Normalization of polarizaation ratios against control fluorophores or the minor axis, allows for an objective assessment of the degree of asymmetry. This brings us closer to the ultimate goal of devveloping a means to determine whether ACD occurs in a population or in individual cells, even when ground truth data such as the degree of asymmetry in biologically relevant ACD events is not available. Although this approach was developed specifically for the analysis of ACD, it also provides a guide for the analysis of all forms of polarity in T cells and potentially in other cell types. A proposed workflow fofor one approach to the analysis of polarization is described in Figure 4.12.

Raz Shimoni Chapter 4 Normalization of PR

Figure 4.12. Suggested workflow for optimal analysis of ACD. A method of analysis that avoids some pitfalls of polarity measurement as illustrated in this study beginns with (1) extraction and segmentation of images, including demarcation of major and minor axes (either using the long axis as demonstrated here, or alternative strategies). (2) A randomly selected sample set of the data should be useed to plot PRmajor and PRminor againstt a range of processing settings, and used to (3) determine the optimal processing settings that avoid artificially high PR values (as indicated by PRminor analysis) but provide good dynamic range

Raz Shimoni Chapter 4 Normalization of PR of PRmajor). (4) The optimal processing settings are used to plot PRmajor against PRminor for the entire population. (5) PRmajor vs. PRminor is utilized for exploration of the quality of the data. Firstly, assuming that polarization occurs only along one axis, the two parameters should be independent of each other and this can be evaluated from the plot both visually (as is common in flow cytometric analysis where correlations can indicate errors in cross-spectral compensation) and by regression analysis. If the plots are still linked to the original data, any outliers can be readily examined to determine possible causes of error. For instance, using an interface such as provided by the TACTICS Toolbox [2], clicking on the dots can bring up the specific frame or movie associated with that data point and possible exclusion of aberrant data such as problems with the focus. Secondly, gating for cells with low PRminor values on the plots enables exclusion of noisy data and simultaneous assessment of the extent, range and variance of PRmajor. (6) The gated PRmajor can then be plotted as a histogram or scatter plot, enabling comparison with control data or between test populations. These plots represent an endpoint of the analysis, but can also be used to determine whether additional values such as mean or median PR, range (i.e. confidence intervals), variance or proportion in different PR values would be informative and could be extracted from the data. (7)(Optional) Depending upon the quality of the data and the goals of the analysis, binarization of the events into ACD and SCD could be achieved by either cut-off or comparison of PRmajor and PRminor values as described in Figure 4.8. In addition, a paired T- test or paired Mann Whitney test can be valuable to compare the PRmajor vs PRminor of each division for the gated population. The statistical inference is that the population proliferates by asymmetric division (on average) if p<0.05 as one rejects the null hypothesis which is symmetry (PRminor and PRmajor are equal).

In this approach, plots of PRmajor and PRminor are generated to compare any post-processing alterations under consideration (such as the variations in segmentation settings used in this study), and to select settings that do not increase the PRminor, but provide good dynamic range for PRmajor. Ideally, a random subset of the data would be analyzed to assess variation without biasing towards the final outcome would be used for this assessment. For instance, histograms plotting PRmajor and PRminor against segmentation for 10 randomly selected events. The same process can be undertaken for fluorescence of a control protein, if available, as different processing settings might be more appropriate. Having selected the most appropriate settings for each fluorescent color, all the data can be displayed in a 2-D scatter plot of PRmajor versus PRminor, and also PRmajor for the fluorescent molecule of interest plotted against PRmajor for the control protein. These plots can be scrutinized for indications of any possible problems with the data. For instance, if the two different fluorescent signals correlate, this might suggest that there is spectral bleed-through between the two fluorescent channels that needs to be corrected. Out-of-focus events might also be identified. Ideally, events that are inappropriately high for either

Raz Shimoni Chapter 4 Normalization of PR

PRminor or for the control protein PRmajor would be assessed to determine the cause of the issue, and depending upon the distribution of the data, and these events can be gated out for subsequent analysis of PRmajor of the control protein. This data can then be presented as a histogram (or 1-D scatter plot depending upon the sample size), and can be plotted alongside controls such as the PRminor of the protein of interest, and the PRmajor of the control protein. As with flow cytometry, the question being explored will then determine whether the histograms are used to comparison between populations, to derive mean or median polarization ratios, or to further subdivide the populations as polarized or not polarized. Additionally, it can be used to consider possible differences in other aspects and measures between two daughter cells such as size, migration patterns, and more. The TACTICS that I developed [1] provides one valuable means to follow this process, and conversion of microscopic data to a format readable by standard flow cytometry data as utilized in DISC [242] would presumably provide another. This strategy is applicable to studies of polarization during cell division as exemplified here (see Figure 4.12), cell migration (Chapter 5)[2], and presumably other forms of polarity such as immunological synapse formation and apicobasal or planar cell polarity [38, 39].

There are several ways in which quantitative data can be verified to ensure that it genuinely reflects the polarity of the cell, the approach suggested here provides one method, but each experiment will provide different opportunities for validation. For instance, in time-lapse microscopy, measuring fluorescence of the daughter cells would allow assessment of whether a polarity ratio correlates with asymmetric inheritance. Controls that are expected to alter polarity can also be informative regarding the validity of the measurements. The use of controls such as PRminor, and the application of gating to remove aberrant events, approximates more closely the rigour in analysis that is now standard in flow cytometry. Multiparametric comparisons, quality control and interactive probing of the data have proven extremely valuable for flow cytometry, and applying such approaches to the wealth of contextual information available in microscopy experiments should dramatically enrich our understanding of biology.

Raz Shimoni Chapter 4 Normalization of PR

4.3.9. Description and utilization of the TACTICS ACD Module

To facilitate the utilization of these approaches described above I developed the TACTICS ACD Module that includes interactive tools dedicated to the analysis of ACD. This includes visualizing localization of selected data points, and the integration of the major/minor normalization methods to improve the evaluation of polarization and asymmetry in dividing. As a paradigm for the utilization of TACTICS, MLA expressing eGFP were utilized (from the same data set shown in Figure 4.4).

A screenshot of the ACD Module is shown in Figure 4.13. The steps for the analysis were; (i) screening the PR (both major and minor) as a function of threshold values in incremental steps allowing systematic determination of the appropriate level (in similar to Figure 4.4). From this display it was observed that the patterns of PRmajor and PRminor are very similar. This can be explained by the eGFP dividing in a symmetric fashion, and shouldn`t be significant different for the PRmajor and the PRminor. Next, by shifting a scrollbar it was observed, interactively, that the same pattern is received from different time points of the divisions. Next, by clicking specific data points (i.e. cell 46), two types of tools were inspected - sequential montages (ii) and time projections of fluorescence intensity for selected division (iii). Projections provide qualitative information about polarity, and montages can be used to inspect the quality of segmentation and splitting applied for each data point. In the example screenshot, MLA cell expressing eGFP undergoes symmetric division in different stages of division; (a) the cell is elongated, (b) the creation of the cleavage furrow, (c) daughter cells dissociated. As can be seen in the screenshot, the segmentation and splitting were exquisitely applied, thus providing supporting evidence for the good quality of the data (iv). Next, the PRmajor vs. PRminor were plotted as demonstrated above, allowing removal of aberrant divisions with high PRminor. For the aim of visualization, and to give better separation, the axis limits were set in different values between [0 0.04]. Color-labelling of data points shows cells at different time points. In the example above, the cut-off was chosen to be 0.02. In addition to PR values, other parameters can be visualized. In this example, the cell intensity was plotted as a function of area, and the data points were color- labelled as a function of the development stage (v). The profile (vi) of the middle

Raz Shimoni Chapter 4 Normalization of PR scatter plot (v) is shown in a 1-D histogram that compares the occurrence of area before and after the disassociation of the two daughter cells (telophasic stage). It can be seen that in later stages of division the area is increased, as predicted. Scatter plot (vii) and (viii) histogram compare the PR values of gated data points.

Figure 4.13. Screen shot of TACTICS ACD Module. Screenshot and analysis using the TACTICS ACD Module, which offers an exclusive platform to measure polarity of dividing cells. Visual diagnostic tools for the quality of the data are shown (i). Clicking on each data point open new windows showing, for instance, montages (ii) and projections (iii). Sophisticated and multi-parametric analysis (iv) correlate polarization measurements with many more other parameters. The data is shown in scatter plots (iv,v,vii) or histograms for distribution analysis (vi,viii). Color labeling is shown based on different parameters (v). Similar to the simulation in Figure 4.9, TACTICS can be used to gate based on the PRminor (vi-vii before gating, vii-viii after gating).

As can be seen, a slight difference is exhibited between early and later stages of division, but not between PRmajor and PRminor. eGFP is expected to give a symmetric ratio, so this difference may be due to the loss of internal asymmetry in later stage cytokinesis (the change in morphology prior to the cell migration), or as a result of an out-of-focus effect (i.e. one of the daughter cells is imaged out-of-focus). Nevertheless, that the subtle differences in polarization ratio can be detected, regardless of the settings applied, highlights the value of this approach for assessing multiple frames to improve the accuracy of imaging and the decision of what time point during the division accurately characterizes the point of division. This principle is utilized in Section 5.3.6 to analyse polarity in dividing DN3 thymocytes.

Raz Shimoni Chapter 4 Normalization of PR

4.4. Summary

The quantification and analysis of polarity in living cells is increasingly important for elucidating biological pathways, and new methods are rapidly emerging. One approach to measuring polarity is by analysing the ratio of intensity in the two halves of the cell using fluorescence microscopy images. However, detection of fluorescence, and the ratiometric measurement, is likely to be sensitive to acquisition settings and image processing parameters. Using imaging of eGFP- expressing cells and computer simulations of variations in fluorescence ratios, the dependence of ratiometric measurements on processing parameters was characterized. This analysis showed that image settings alter polarization measurements, and that clustered localization is more susceptible to artefacts than homogeneous localization. Therefore, using a constant arbitrary value as cut-off for dynamic fluorescence ratios is inadequate. It can lead to inconsistency in the classification of divisions for being asymmetric or symmetric, and consequently, does not necessarily reflect any underlying biological difference between the two groups. As a result, the reproduction of constant results under different conditions is unfeasible. Until now, most systems of ACD studied have shown such strong asymmetry that sophisticated quantification methods were not required. Still, as more subtle forms of asymmetry are now studied in T cells, greater knowledge of the way that image acquisition and processing affects the ratiometric analysis of ACD is required. To correct for such inconsistencies, I developed and validated a method for choosing the most appropriate analysis settings, and for incorporating internal controls to ensure fidelity of polarity measurements. In comparison to other techniques such as the Receiver Operating Characteristic (ROC)[266], this approach doesn’t require a data sample that contains TPs, which in many cases (and in particularly the analysis of ACD) is lacking. The utilization of these approaches is included in the focus of the next chapter.

Raz Shimoni Chapter 4 Normalization of PR

Analysis of Polarity in T cells

5. Chapter 5

Raz Shimoni Chapter 5 Analysis of Polarity

5.1. Introduction

As explained previously in this thesis, the exact role of polarity in T cells is unclear, mainly because of the difficulty of generating robust, unbiased measures of polarity in immune cells. To promote the investigation of polarity in T cells, this chapter focuses in the utilization of TACTICS as a platform for the analysis of two important types of polarity in T cells - polarity during cell migration and polarity during division, namely ACD.

The analysis of polarity in T cell migration is much harder due to their highly dynamic nature, in particular the ability to rapidly change their morphology depending on their biological activity. For this reason, standard mathematical models [267, 268] and quantification methods to measure polarity [95] cannot resolve correct and consistent characterization of polarity during migration. Analysis of polarity during cell division was the subject of Chapter 4. As explained previously, ACD is relatively well understood in model systems such as Drosophila[11] and, in terms of developmental biology[269], its importance in lymphocyte development, function and disease requires further investigation [34-36, 270-272]. Some reports of ACD in immune cells were reliant on immunofluorescence of fixed cells, which provided a higher resolution [34, 35]. Others employed imaging flow cytometry that provided large data sets [270]. Unfortunately these methods only provided a snapshot and cannot be correlated with downstream events. Additionally, it is hard to identify telophasic cells, and two touching cells can mistakenly be identified as a dividing cell. As such, further investigation of ACD, in addition to a general investigation of cell polarity in T cells, requires a direct observation of parental cells and daughter cells, their tracking during migration, and reliable approaches to measure polarity in live cells. Ideally, quantitative analysis of fluorescence time-lapse microscopy data could provide the perfect means to track cells during migration, direct inspection of ACD and study statistical correlations between asymmetrical distribution of polarity proteins in the context of supporting cells and other polarity cues [273]. However, the elusive nature of T cells requires the development of new tools that provide a reliable quantification of polarity.

101

Raz Shimoni Chapter 5 Analysis of Polarity

To handle the challenges described above, TACTICS provides an efficient platform for algorithm development and enables robust and interactive analysis for many tracked cells. As a paradigm for the development of new approaches to analyse polarity, we study the cell fate determinant, Numb, which was shown to polarized during division in C. elegans blastocysts, Drosophila neuroblasts, and a variety of mammalian cells including T cells and thymocytes [35, 212, 263, 274-278]. Numb is a regulator of Notch signalling, and an endocytic adaptor protein that exhibits intracellular polarity in T cells [34, 35, 276, 279, 280]. In epithelial cells, Numb is phosphorylated, colocalized with, and controlled by the polarization of the polarity protein aPKC [281]. aPKC is polarized to the uropod (trailing edge) of migrating thymocytes [95], immunological synapse of human T cells [282], and is also polarized during ACD of T cells [34-36]. However, whether Numb is regulated by aPKC in T cells is yet to be determined. To investigate a possible functional association of aPKC with Numb, our laboratory has generated T cells and thymocyte cell lines expressing a mutant form of Numb that cannot be phosphorylated by aPKC at serine 7 and 295 (“Numb2A”). In epithelial cells Numb2A is excluded from the apical membrane and so does not polarize to the basolateral membrane [281]. Thus, determining whether there is a significant difference between the localization of Numb and Numb2A presented a major step towards better understanding in T cell polarity.

This chapter is organized as follows. Section 5.2 describes the experimental methodology and computational methods for this chapter. Section 5.3 presents the results of this chapter structured as the follows. Firstly, the Polarization Module is described in Section 5.3.1. The Polarization Module is dedicated to effective data visualization and with the capacity to display individual data points in the context of multiple parameters. This includes the ability to link between processed and raw data, gating, and quality control tools. In Section 5.3.2 I emphasize the technical details and the aspect of analysis, using the biological meaning of Numb as a paradigm for high-throughput quantification of polarity proteins in general. This approach was utilized in Section 5.3.3 to perform a comparison between the polarizations of Numb WT, and revealed that, in contrast to epithelial cell, the Numb mutant is polarized in T cells. In Section 5.3.4 I performed polarization

102

Raz Shimoni Chapter 5 Analysis of Polarity measurements within the uropod, and show that Numb is polarized to the tip of the uropod. Thus, confirming that the measured polarity in Sections 5.3.3-5.3.4 is not due to the location of the nucleus. Section 5.3.5 shows polarization measurements in DN3 thymocytes that use other parameters for gating, and attained similar results as in Section 5.3.3. In Section 5.3.6 I utilize TACTICS to measure polarity in dividing DN3 thymocytes. Section 5.4 summarizes the chapter.

5.2. Methods

5.2.1. Data input for TACTICS

Two types of cells were chosen to study how Numb polarized during migration and cell divisions. The first cell type was UCD-144-MLA (‘MLA’) cell line from a gibbon with spontaneous lymphosarcoma [283] that is highly polarized, relatively easy to culture, and, since it doesn’t require the presence of stromal cell, was used as control. The second cell type is thymocytes, T cell precursors in the DN3 stage of development. The DN3 is a stage within the DN in mammals. The rationale for this choice is that DN3 thymocytes cells are acutely dependent upon Notch signalling and thus more likely to be influenced by Numb [280, 284, 285]. In addition, thymocytes development represents a promising model system for elucidating the molecular events by which polarity orchestrates fate determination.

Data input for TACTICS was provided by Dr. Kim Pham. Briefly, hematopoietic stem cells were harvested from livers of C57/Bl6 mouse embryos (age E14.5 days). The cells were transfected with expression constructs encoding fluorescently tagged proteins. To investigate whether ACD occurs at the precursor stage of T cell development in DN3 thymocytes, an in vitro model system developed by Zuniga-Plucker and colleagues [286-288] was adopted: thymocytes cultured with OP9 stromal cells transfected with the Notch ligand, Dll1 (OP9-Dll1). This system has been extensively used and validated by many laboratories, and provides a robust and high fidelity replica of physiological thymocyte development [288-292]. After 9 days of culture, cells that expressed the fluorescent proteins and had progressed to the DN3 stage of thymocyte development were isolated using flow cytometry to select for cells that were negative for lineage markers (Mac1, CD3, RB6-8C5,

103

Raz Shimoni Chapter 5 Analysis of Polarity

NK1.1, Ter119, negative for CD4 and CD8 to remove mature T cells, and CD25+, CD44lo. Time-lapse imaging was based on co-culturing thymocytes with stromal cells (OP9-DL1) to support proliferation or differentiation and enable development into DP thymocytes. Since the DN3 thymocyte divisions are not synchronized, they provide a low yield of divisions per well of (less than one division per well).

5.2.2. Constructs

Numb and Numb2A constructs were tagged with fluorescent proteins, allowing non-invasive observation of Numb localization using fluorescence microscopy in migrating cells. Thymocytes expressing eGFP-Numb or the variant, eGFP-Numb-2A [281], and mCherry were generated using retroviral transfection by Dr. Kim Pham using the Murine stem cell virus (MSCV) retroviral expression system. MLA expressing eGFP-Numb and mCherry-tubulin were generated by Mandy Ludford-Menting. Immunostaining confirming that the localization of ectopically expressed Numb tagged with eGFP traffics similarly to endogenous Numb in T cells, in addition to complementary protein assays and flow cytometry supporting the biological results in this study were performed by Dr. Kim Pham [2].

5.2.3. Imaging conditions

Time-lapse confocal microscopy images of MLA cells or thymocytes and OP9-Dll1 undergo random migration and cell divisions were provided by Dr. Kim Pham. Briefly, 1x104 MLA cells or 1x104 thymocytes with 1x104 pre-adhered stromal cells, were plated in a 8-well IBIDI slide containing 250 μm x 250 μm x 60 μm microwells pre-washed with 100% EtOH followed by a warm media wash. Time-lapse data for experiments was acquired with a Leica TCS SP5 Confocal microscope (Leica Microsystems CMS GmbH, Germany) fitted with a temperature controlled chamber maintained at 37˚C and 5% CO2 (thymocytes) or 10% CO2 (MLA cells). Images were acquired using a HCXPLApo x63 glycerol immersion objective (NA 1.3). Multiple stage positions controlled by LAS AF v2.0 software interface were captured every 2 mins for 1-24hrs with line averaging of 2, and 5 z sections of range 7.5μm-15μm, corresponding to optical distance of 2.5-3.0 µm between each section. Where indicated, cells were treated with 0.5-1 μM Phorbol 12- myristate 13-acetate (PMA)(Sigma) for 20mins, 37C, 10% CO2 prior to imaging.

104

Raz Shimoni Chapter 5 Analysis of Polarity

5.2.4. Importation of file images to TACTICS

Sequential multiparametric images (x, y, z, time, and multi-channel of different fluorescent wavelengths), were captured for multiple positions and saved as .lif files (Leica LAS AF v2.0 support format). Each image within the stack was 512 x 512 pixels in size. Following acquisition, images were exported as individual 16-bit tifs and stacks using an automated memory handling journal “Lif to tif Adaptable with Projections” written by Cameron Nowell in MetaMorph Image Series 7.7 software (Version 7.7.4. Universal Imaging Corporation, USA). Images were imported into the TACTICS Toolbox for post-acquisition analysis. In total 80 MLA cell positions and 144 thymocyte positions were analysed and approximately 2 terabytes of data was processed. Calculations were performed on an HP Z400 workstation equipped with a 3.3 GHz Intel Xeon W3580 Quad processor and 16 gigabytes of RAM working under a Windows7 64bit operation system. Next, TACTICS version 2.0 was utilized (older version of TACTICS, but the same analysis is supported in all later versions), running in MATLAB R2008a version 7.6 and later versions using the MATLAB GUIDE, MATLB IPT, and the MATLAB Statistics Toolbox. The minimal system requirements are Windows® 32-bit or 64-bit operating system, Intel® Pentium® 4 processor or equivalent, and 1 gigabytes of RAM.

5.2.5. TACTICS Pipeline for polarity measurements

5.2.5.1. Cell segmentation (Figure 5.1.A.i-iii)

Each confocal stack imported into TACTICS Toolbox comprised 5 focal sections. For each stack, the 5 focal sections were projected into a 2D plane by averaging each pixel over its values in each stack. The new projected images were smoothed using a 3x3 2-D median filter followed by a morphological close-opening operation with a 3x3 disk structuring element, [0 1 0; 1 1 1; 0 1 0]. Objects in the 2-D projections were roughly segmented using a global threshold. Otsu’s method was used to automatically choose the threshold level that gives minimum interclass variance of the high and low fluorescence intensity [149]. False objects such as background noise, dust and dead cells were removed by applying object- discrimination geometry criteria. Objects with an area of less than 100 pixels or with

105

Raz Shimoni Chapter 5 Analysis of Polarity a circularity of less than 0.5 were discarded. Gaps within each object were filled using the MATLAB imfill.m function. The remaining objects within the binary images were labelled as individual cells. Since segmentation by fluorescence overestimates the perimeter of high intensity cells if a global threshold is applied, a second phase of segmentation was applied. New segmentation threshold values were obtained by applying Otsu’s method for the best section of each cell individually, with the best section automatically defined as the section with the largest grey-level variance. Because in some cases cells were oriented perpendicular to the imaging plane and could not be segmented based on only one section, a minimum threshold for fluorescence within a cell was set, and cells for which the fluorescence intensity was below the threshold were automatically discarded from the analysis.

5.2.5.2. Separation of touching cells (Figure 5.1.A.iv-vi)

In some cases the cell segmentation algorithm mentioned above identifies cells that are in contact with one another as a single cell. To apply an additional step to separate between the touching cells I used the following algorithm:

First, the MATLAB function treefit.m was used to create a classification tree. This classification tree used the morphological descriptors described in the next section to discriminate between objects containing one cell and objects containing two cells.

106

Raz Shimoni Chapter 5 Analysis of Polarity

Figure 5.1. Methods applied to process and structure data input for the Polarization Module. (A) Segmentation of raw images by automatic adjustment of the intensitty threshold to detect the border of each cell using control protein (mCherry-tubulin or mCherry).This is illustrated by: (A.i) Representative mean proojjected (raw) image. (A.ii) Representaative binary image of (i), after automated threshold was applied. (A.iii) Image of (i) including pink dots that represent the pixel perimeters detected frfrom the binary image (ii). (A.iv) Representative mean projected (raw) image of two touching thymocytes expressing mCherry. (A.v) Representative binary image of (iv), after automated threshold was applied based on mCherry. In this case, the two cells appeared as one segment. (A.vi) Image of (iv) including pink dots that represent the pixel perimeters detected after the TGI-Objects splitting algorithm was applied to automatically split the segment into two cell objects. (B) Once objects are detected, a particle tracking algorithm is applied. Each cell was tagged with a numerical ID for further analysis, and morphological features were extracted for each cell at each time point. This information is stored in experimental files for each pposition. A representative image of labelled tracked cells is shown. If a dividing cell was assigned, the

107

Raz Shimoni Chapter 5 Analysis of Polarity two daughter cells were labelled accordingly with a dot and numbers 1 and 2. (C) The TACTICS Measurements Module is used to extract each tracked cell from each position to automatically create cell libraries comprised of ‘seqTages’ (a sequence of cell images within bounding box over time represents in montage arrangement), generated for each z section and each channel. To remove bleedthrough between each pair of sections in the Cherry channel (C.i) and the eGFP channel (C.ii), a correction was applied. (D) A screenshot of the TACTICS module for spectral unmixing that automatically removes bleed through between each channel for each cell in each position. Examples of cells before (D.i) and after (D.ii) spectral unmixing are shown to the right for both elongated and rounded cells. (E) Each 3-D stack (E.i) is then mean projected to give the 2-D reconstructed image followed by automated cell alignment (E.ii). (F) Multiparametric data points extracted from different experiments for further analysis, each tagged by its ID to allowing linking to its original location. Schematic showing how the experimental database is structured and is ready as single file to be loaded to the TACTICS Polarization Module for interactive analysis.

To train the classifier, I used a training data base comprising 12 tracked dividing cells with a total of 1326 data points from which 174 data points were manually identified as touching cells. If an object was recognized as two touching cells, an TGI-objects splitting algorithm(written by Dr. Pavel Lobachevsky [293]) was implemented to split an object that potentially represents two-cells based on shape and spatial distribution of intensity. TGI-Objects works in the following manner: First, it finds the two closest pixels on opposite sections of the object boundary (a “bottleneck”). It then builds a path across the bottle-neck by connecting pixels with minimum intensity. To prevent frequent over-segmentation and allow automation, a feedback loop was incorporated that adjusted the minimum cell size and geometric criteria of roundness that could be recognized by the TGI-Objects function. A similar approach was demonstrated by[294].

5.2.5.3. Geometrical descriptors for classifying function

The following geometrical descriptors were used as input to the classifier described above:

 Area.  Perimeter.  Circularity: Derived as 4Area/Perimeter2. The circularity is equal to 1 for a circular disk. It is smaller than 1 for all other objects.

108

Raz Shimoni Chapter 5 Analysis of Polarity

 Major axis length: The length of the major axis of the segment is equal to the length of the major axis of the ellipse that has the same normalized second central moments as the segment.  Shape projection: The 1D horizontal shape projection is defined as

Ph(x) = Σyf(x,y), where f(x,y) is a binary function describing the object. (f(x,y)=1 if pixel (x,y) belongs to the shape, else f(x,y) = 0). The shape projection of a single cell will differ from the shape projection of two touching cells. While a cell with a disk shape will typically give one peak in the projection histogram, two touching cells will give two peaks. Hence this descriptor can be used to distinguish between an object comprising two touching cells, and an object comprising a single cell. To calculate the shape profile, the object is rotated so that its major axis is parallel to the x-axis. Then the object is projected according to the formula mentioned above.

 Contour connectivity score: To derive the contour connectivity score I applied level set active contours without edges algorithm and recorded how many iterations of the algorithm are required to split the object into two. To return the connectivity score I utilized the MATLAB function Level_Set_Evolution.m written by Chunming Li.4

 Number of detected disks: To count the number of multiple touching disks via the Hough Transform, I used the MATLAB function houghcircles.m written by Yuan-Liang Tang5. The input for minimum cell radius was set to 3.8 µm and the maximum cell radius was set to 9.6 µm.

5.2.5.4. Cell tracking (Figure 5.1.B )

Once cells had been identified, the MATLAB function regionprops.m was used to identify the centroid of each cell. To find the cell trajectories I applied a particle tracking algorithm[154] written by D. Blair (Georgetown University, Washington DC) and E. Dufresne6 (Yale University, New Haven, CT). I found that

4 Downloaded from: http://www.engr.uconn.edu/~cmli/ 5 Downloaded from: http://www.cyut.edu.tw/~yltang 6 Downloaded from: http://www.physics.emory.edu/~weeks/idl/

109

Raz Shimoni Chapter 5 Analysis of Polarity this algorithm was efficient and extremely useful, but often failed when large numbers of cells were within close proximity to one another. In these cases I applied a robust MATLAB implementation of the Hungarian algorithm to associate bipartite centroids (written by Yi Cao, Hungarian Algorithm for Linear Assignment Problems7), followed by my code to connect between consecutive frames. The tracking process does not require mathematical modelling of the cell motion.

5.2.5.5. Creation of cell libraries (Figure 5.1.C)

Once the cells are segmented and labelled, TACTICS automatically extracts and stores information from each identified cell in a data library, for further analysis and visualization of intracellular polarization. Each cell in each image represents a single data point, with which the software associates multiple parameters describing the cell. These parameters include: A) Morphological and geometric descriptors (circularity, major/minor axis length, perimeter), B) Parameters describing the motion of the cell including: apparent velocity, the MSD, and turning angle. C) A reconstructed cell image within a bounding box.

5.2.5.6. Spectral unmixing (Figure 5.1.D)

To gain good separation between the different fluorescent signals, a bleedthrough correction was applied. This was done by estimating that the total emission in each channel is nearly a linear sum of intensity contributed by each fluorescent protein. Hence, the signal between channels in double transfected cells was calculated using the estimation formulae:

Equation 9:

I 497 543nm  607 672nm x , y  I497 543 nm x , y  I 607 672 nm x , y I 607 672nm  497 543nm  x , y  I607 672 nm  x , y  I 497 543 nm x , y

where x and y are the pixel coordinates within the image, the left side of the equations shows the unmixed channel after the correction, I497-543nm is the intensity collected for the eGFP channel, I607-672nm is the intensity collected for the mCherry channel,  and  are approximation coefficients that are proportional to

7Downloaded from: http://csclab.murraystate.edu/bob.pilgrim/445/munkres.html/

110

Raz Shimoni Chapter 5 Analysis of Polarity the bleedthrough and were extracted in a separate experiment, by analyzing the bleedthrough from manually selected cells that express only one of the fluorescent proteins.

5.2.5.7. 2-D image reconstruction and cell alignment (Figure 5.1.E.i)

The 2-D images were constructed using the following procedure: Firstly, although imaging was performed over five Z sections to capture cells as they moved slightly up and down, only four Z-sections were required to cover each cell, and the fifth contributed unwanted background fluorescence. Therefore, the best four optical sections for each tracked cell were automatically defined as the four sections with the largest grey-level variance. Next, intensity noise was removed from each section using a Fast Fourier Transform (FFT) low-pass denoising algorithm written by Peter Kovesi[295]8. Finally, spectral unmixing was applied to each section to distinguish between the eGFP and mCherry fluorescence (as detailed in section 5.2.8 above).

To align the 1-D projections of the cells along the axis of polarization, the following procedure was used: (a) The major and its perpendicular minor axis were determined by fitting an ellipse to the cell perimeter. The axes were found using the second moment of the cell shape. (b) The angle between the major axis and the y- axis of the image was found, and the images were rotated to align the major axis with the y-axis. The rotation was performed around the centroid of the cell. ‘Nearest neighbour’ interpolation was used for the rotation. (c) Once the cell was rotated, a new major and minor axis was found, and the minor axis was used as a border that split the cell into two (in this case not necessarily equal) parts.

The cells were then reoriented so that the uropod faced the bottom of the image (as shown in Figure 5.1.E.ii). The identification of the uropod was performed using the following procedure: (a) When the difference between the areas of the two parts of the cell was more than 1.5 fold, the smaller side was defined to be the uropod side and the image was flipped so the uropod was facing downwards (to create-90 degrees between the x-axis and the vector between the centroid and the uropod location). If the difference between the areas of the two sides was less than 1.5 fold,

8 Downloaded from: http//www.cs.uwa.edu.au/pub/robvis/papers/pk/denoise.ps.gz/

111

Raz Shimoni Chapter 5 Analysis of Polarity the orientation of the cells was found using as follows steps. The centre of the uropod extension was measured using the Euclidean distance to transform the centroid of the cell to the cell boundaries. Pixels on the boundary that were further from the centroid indicated a cell extension such as uropod formation. (b) To identify the centre of the uropod I created an image showing the distance from the centroid to the boundary pixels, and binned the cell pixels using MATLAB sliding-neighbourhood operations with maximum function. The centre of the bin that was furthest from the centroid was identified as being the centre of the uropod. (c) The cell image was rotated so the uropod faced the centre of the bottom of the image.

5.2.5.8. MTOC identification (Figure 5.1.E.ii)

The microtubule organizing centre (MTOC) locates in the rear side of migrating T cells, between the nucleus and the tip of the uropod. To allow identification of the MTOC mCherry-tubulin was used. The MTOC was determined as a bright spot in the rear of the MLA cells. To find the location of the MTOC I applied a robust spline smoothing algorithm to the images (MATLAB code was written by Damien Garcia9. This algorithm creates clear peaks in regions of high intensity. The MTOC coordinates were determined as the location of the peak (pixel with highest intensity).

5.2.5.9. Quality control and manual inspection

TACTICS includes a user interface that facilitates optimization of each step in the scheme. In addition, the interface enables manual corrections to be performed when algorithm performance is insufficient. For instance, TACTICS allows corrections to be made to cell segmentation, cell tracking, uropod selection, and cell orientation on the basis of control protein (mCherry-tubulin for MLA and mCherry for thymocytes). Using TACTICS, false detections of cells were removed with a mouse click. Poorly segmented cells were either discarded, or the segmentation manually improved. Repeating the image processing of specific images was applied in cases where real cells were not detected by the algorithm. In some cases where cells suddenly moved unusually large distances between consecutive images (cell may drift as a result of flow or when attached to stromal cells), cell trajectories were

9 Downloaded from: http://www.biomecardio.com/matlab/smoothn.html

112

Raz Shimoni Chapter 5 Analysis of Polarity corrected manually. These corrections included rejection of invalid track points, definition of new track points, and splitting and joining of tracks. For cells where other cell extensions were mistakenly identified as the uropod, or for cells where the uropod was missed by the software, manual corrections to uropod location were applied

5.2.5.10. Measurements of polarity during migration

To measure polarization ratios along a defined axis, a profile line was built across the image of the cell between two points- p1 and p2, located on opposite sides of the cell perimeter. The points were chosen to split the cell into two equal sections. To create the profile line, I employed the Bresenham line algorithm written by N. Chattrapiban10 to perform as follows: First, p1 and p2 were chosen in a way that the line between them was aligned with the major (or minor) axis of the cell (i.e. the centroid of the cell fell on the line between them). Drawing this line splits the cell into two parts. Second, if the number of pixels in each of the two sides of the cells is not equal, then p1 and p2 were shifted one pixel, the line between them redrawn, and the number of pixels in each section of the cell was recalculated. Third, this process is repeated until two parts with equal areas are found. Fourth, once the cell was split into two equal parts the polarization coefficient for a cell was determined as the ratio between the integrated pixel intensity of two cell halves of the cell (difference of integrated intensity over sum of integrated intensity).

5.2.5.11. Measurements of polarity during division

Data images of dividing thymocytes expressing eGFP-Numb and eGFP- Numb2A were taken from the same data set presented in previous section. Images of the parent cell and daughters were then utilized to create montages orientated along the major axis to enable qualitative assessment of eGFP-Numb and eGFP-Numb2A fluorescence distribution. Projections were generated by compressing and aligning images from sequential frames onto a line of single pixel width in similar to preformed for migrating cells. PR values (both major and minor) were calculated according Equation 2 for normalization of polarization ratios. For the aim of this

10 Downloaded from: http://www.mathworks.com/matlabcentral/fileexchange/12939-bresenhams- line/content/bresenham.mz/

113

Raz Shimoni Chapter 5 Analysis of Polarity analysis, the assumption is that it is impossible to discriminate between the two daughters (i.e.: by assigning the closer daughter cell as distal and the second daughter cell as proximal as shown by Oliaro et al. [34] ). Therefore, the PR were converted to absolutes to obtain values between 0 (completely symmetric) and 1 (completely asymmetric).

5.2.6. Statistics and display

A non-parametric Mann Whitney test (two-tailed) was used to compare median PR of gated thymocyte populations, minor versus major axis.

All PR were derived in TACTICS, but in some cases data was exported to GraphPad Prism (v5.04) software for plotting.

5.3. Results and discussion

5.3.1. Description of the TACTICS Polarization Module

I have developed the TACTICS Polarization Module to enable strategies similar those described in the Lineage Module (described in Chapter 3). Applying similar principles to measurements of polarity from migrating cells, the Polarization Module allows parameter optimization methods with which to intuitively and visually explore imaging data in the context of cell polarity such as the ability to focus on subpopulations of cells, displaying qualitative and quantitative parameters of interest, gating, and parametric generation. Similar to the principle of the Lineage Module, the Polarization Module has the ability to define and create new types parameters. In particular, different types of polarization measurements those are based on different axis of polarity and normalization formulations such as direction of movement, or based on localization of proteins in another channel. The next sections utilize the tools described above. In addition, the Polarization Module supplies an interactive user interface for data visualization and exploration (Figure 5.2.A), allowing for the history of the cells to be used in gating strategies and the images to be viewed directly from the data points. For instance, the user can easily derive various metrics of fluorescence and polarity with a button click. Good examples are time-projections (Figure 5.2.B.i), polarization ratios (Figure 5.2.B.ii),

114

Raz Shimoni Chapter 5 Analysis of Polarity and fluorescence weighted centre used to obtain an asymmetry magnitude for determining intracellular polarization (Figure 5.2.B.iii) as described in [95]. An advantage of such interactive analysis is the provision of an overview of patterns within the dataset, which can suggest the formulation of more useful questions with which to probe the data [296]. A precedent in biology is well demonstrated by the analysis of flow cytometry data, where the relationship between different parameters can be displayed with ease for thousands of events. This allows an initial, informal viewing of patterns which frequently prompts a new hypothesis that can be formally tested on the same dataset by gating strategies enabled by the same software. The next sections utilize the Polarization Module to measure polarization in migrating cells using axial subdivision, and to develop new tools for objective and controlled analysis.

115

Raz Shimoni Chapter 5 Analysis of Polarity

Figure 5.2. Interactive analysiis with TACTICS Polarization Module. TACTICS Polarization Module supports massive parallel analysis of given parraameter lists. (A) An example screenshot of the Polarization Module for quantitative exploration of population data. Scatter plots representing microscopic parameters or their derivattions such as polarity ratios allows visualization of data points in similar to flow cytometry. Such approach aids to illuminating of possible relationships between parameters within subpopulations of the images. Left hand panels indicatee the organization of the data and extracted parameters, middle top panels are 1-D projections for mCherry-tubulin (Ch-Tb), eGFP-Numb (G-Nb) and merged channels. Scatter plots (bottom left) compare parameters from the entire data set (in this case, velocity versus area and PR across major and minor axes. The red cross the upper scatter plot shows pointed cell in different parameters, where information about any cell is achieved with a mouse click on data point. In addition it can be linked to the corresponding image of that daatta point (top right), a montage of the tracked cell corresponding to that data point (middle bottom), and the cell tracks over time (top right). Histograms (bottom right) show PR of a subpopulation of cells gated from the scatter plot (gates not shown in this screenshott). Examples of data outputs that can be created from with the user interface includes (B.i) 1-D projection of cell fluorescence for a single cell tracked over time , (B.ii) Scatter plots of PR for minor versus major axiss. , and (C.iii) Polar coordinates (as shown previously by Melichar et al.[95].

116

Raz Shimoni Chapter 5 Analysis of Polarity

5.3.2. Polarization measurements of Numb using the major-minor axis

To develop methods by which a population of cells can be statistically tested for polarization along the major axis, and to assess whether the polarization across the minor axis could be used to determine the most appropriate settings for assessment of the degree of symmetry versus asymmetry, I next carried out a systematic investigation of several approaches that are based on the principle of using the internal symmetry as control. MLA cells expressing eGFP-Numb and mCherry-tubulin were used to demonstrate the utilization of TACTICS pipeline to measure polarization in migrating cells. MLA are highly polarized T cells, and therefore attractive for study of protein localization during the formation of the uropod while the cell migrate. For this analysis, each time frame of each cell was then treated as an independent data point, with a total of 5366 data points, which were extracted from 94 tracked cells.

For each data point a major axis and its perpendicular minor axis were derived for each data point (Figure 5.3.A) as explained in section 5.2.12. Each cell was aligned as explained in section 5.2.9, enabling TACTICS to display kymograph- like images in which the fluorescence of each T cell is projected onto the major or minor axis. The projected images of a subset of cells that had been manually selected for elongated cells with a clear uropod were displayed alongside each other to visually explore the data in qualitative manner (Figure 5.3.B). This series of images showed a pattern suggesting polarization of both mCherry-tubulin and eGFP-Numb along the major but not the minor axis. To quantify this, TACTICS automatically bisected the cells along major and minor axes, calculated total fluorescence in each hemisphere and derived the PR (Figure 5.3.C). To assess whether tubulin and Numb polarization were correlated, PR values of mCherry-tubulin (major axis) were plotted against PR of eGFP-Numb (major axis) (Figure 5.3.D).

117

Raz Shimoni Chapter 5 Analysis of Polarity

Figure 5.3. TACTICS analysis piipeline to quantify eGFP-Numb poolarization in T cells. (A) Further processing for analysis of cell polarization involved automated assignment of a major axis and minor axis, allowinng cell alignment for the projection of 1-D representations of the polarization along the major or minor axes. (B) Dotted line indiccates midpoint of the cell, and ratios representing the PR across either the major or minor axes (C). Possible values range between -1 and +1, or 0 and +1 when absolute values are used. The ratios derived as in (C) for the major axis of the entire population (5366 data points extraacted from 94 tracked cells) were represented as scatter plots comparing mCherry-tubulin pollarization with eGFP- Numb polarization for each data point (D). Data points related to individual cells tracked over time are each represented by a different colors. Clicking on the dots in the TACTICS interface allows for drilling down onto the cell images, the tracks, and movies or montages for each cell, with the two dots selected (black cross) linking to imagess of the selected cells, their tracks and associated time-lapse images (D.i-ii) Scatter plots coomparing polarization along the major versus minor axxis for mCherry-tubulin (E.i) allowed gating for highly polarized cells for further analysis of eGFP-Numb polarization in E.iii. (gated for all data points with greater than 0.2 or less than -0.2 PR along the major axis, and between -0.2 and 0.2 in the minor axis). A scatter pllot comparing polarization along the major versus minor axis for eGFP-Numb shows negligible polarization for ungated cells (grey crosses), but clear polarization for gated cells (black dots). The eGFP-Numb polarization ratios for events gated for high mCherry-tubulin polarization are also plotted as (E.iii) histograms comparing major and minor axis ratios for signed (left panel) and absolute (right panel) eGFP-Numb ratios.

118

Raz Shimoni Chapter 5 Analysis of Polarity

Polarization of eGFP-Numb and mCherry-tubulin were clearly correlated (Pearson’s coefficient R=0.87), indicating that at a population level, eGFP-Numb co- polarized with mCherry-tubulin. For this display, each data point was color-coded according to the tracked cell to which it was linked, and the spread of colors indicated that variation in PR did not correlate with cell-specific differences. TACTICS Polarization Module also facilitates links from individual data points to images or movies of the cells (Figure 5.3.B. & Figure 5.3.D). This allowed us to determine that data points with high PR (either positive or negative) were linked to images with clear polarization of both mCherry-tubulin and eGFP-Numb to the uropod (Figure 5.3.D.i), and were also linked to movies which indicated that the cells were migrating in a straight direction. In contrast, data points with PR around 0 connected to cells that were not well polarized, and were often stationary or turning (Figure 5.3.D.ii). These observations are compatible with the known localization of the MTOC to the base of the uropod in migrating T cells [63], and suggested that gating on a population of cells with high PR for mCherry-tubulin can be used to automatically select T cells that were migrating in a straight line, and so were most appropriate for accurate analysis of Numb polarization. Therefore major PR values of mCherry-tubulin were plotted against minor PR values, and TACTICS interface was used to select for all T cells with high polarization along the major axis, but low polarization along the minor axis (Figure 5.3.E.i), two black rectangles). A scatter plot of eGFP-Numb PR ((Figure 5.3.E.ii), major versus minor ratios) indicated that, although ungated cells showed an even spread of polarization along both axes (grey crosses), the gated cells were distinctly polarized along the major axis (black dots). The same data was displayed as relative frequency histograms (Figure 5.3.E.iii), either as signed (left panel) or absolute (right panel) PR. A comparison of absolute Numb PR from unselected (Figure 5.4.A), manually selected on the basis of a clear uropod (Figure 5.4.B), or automatically selected on the basis of tubulin polarization (Figure 5.4.C) indicated that both manual and automatic gating approaches yield clear polarization of Numb. For both manual and automatic gating, events were selected using only the mCherry-tubulin images to avoid bias. These data combined illustrate that measuring polarization along the major-minor axis from a control protein can provides objective assessment for the most appropriate settings and that interrogation of data from large populations (such as standardize on base of control

119

Raz Shimoni Chapter 5 Analysis of Polarity protein) provides a truly reliable technique to exclude effects such as the dynamic morphology of T cells on polarization measurements. In addition, it indicates that wild-type Numb is polarized to the uropod of migrating MLA cellls.

Figure 5.4. Analysis of eGFP-Numb polarization in migrating T cellls. (A) Comparison of Absolute PR of eGFP-Numb (majjor and minor axis) in all tracked T cells (ungated). (B) Manually selected uropod containinng T cells. (C) Gated on high mCherry-tubulin ratios on the major but not minor axis. The number N of events and the median Absolute PR are indicated.

120

Raz Shimoni Chapter 5 Analysis of Polarity

5.3.3. Gating on the minor axis provides an objective comparison of polarization

The principle of axial subdivision (as described in Section 5.4) can be used to compare between the polarization of eGFP-Numb and eGFP- Numb2A. The polarization of eGFP- Numb2A was quantified from time-lapse imaging in the same data processing pipeline that was applied to eGFP-Numb expressing cells. Fixed staining of eGFP-Numb2A in MLA cells indicated a broadly similar distribution to that of endogenous Numb and of eGFP-Numb (data shown in[2]).

Figure 5.5.A shows that Numb2A was highly polarized in MLA cells gated for mCherry-tubulin polarization (with median PR: 0.33 for major axis, 0.09 for minor axis), showing even higher polarization than eGFP-Numb (median PR: 0.20 for major axis, 0.05 for minor axis). TACTICS was used to determine whether hyper-activation of all isoforms by treatment with 12-myristate 13-acetate (PMA) impacted upon polarity. Previous studies in epithelial cells showed that eGFP-Numb was rapidly lost from the plasma membrane following the addition of PMA, but eGFP- Numb2A was insensitive to PMA treatment[281]. Interestingly, in T cells, treatment with PMA resulted in a decrease of polarization of both eGFP-Numb and eGFP-Numb-2A (Figure 5.5.B), with median polarity ratios 0.18 and 0.24 respectively).

These data show that the influence of these two phosphorylation sites of Numb differs between T cells and epithelial cells. To confirmed that differences in polarization are not due to different expression levels, the PR values were plotted against the integrated fluorescence intensity (Figure 5.6), and found little or no correlation between the fluorescence intensity and PR values over a greater than 10 fold range in fluorescence intensity (Pearson’s correlation coefficient: 0.53). Thus, expression levels were not a factor in polarization of Numb. Possible explanations for our findings that Numb2A in T cells is polarized and sensitive to PMA treatment include that aPKC does not phosphorylate Numb in T cells, or that different sites are phosphorylated by aPKC.

121

Raz Shimoni Chapter 5 Analysis of Polarity

Figure 5.5. Utilization of TACTICS to compare between Numb WWT and mutant in gated subpopulation based on major-minor plot. (A) PRs were derived from time-lapse imaging of migrating MLA cells transduced with mCherry-tubulin and either wild-type eGFP-Numb (red lines) or eGFP-Numb-2A (blue lines). Events were gated for polarized T cells based upon mCherry-tubulin fluorescence, and polarization along the major axis (solid lines) was compared with polarization along the minor axis (dotted llines). (B) PR values from mCherry-tubulin- polarized MLA cells in were compared witth ratios after PMA treatment (5 μM, added 30 minutes before imaging commenced). Only polarization along the major axis is shown, untreated cells are indicated by the solid line, and PPMA-treated cells by the dotted lines. Events analyzed: wild-type eGFP-Numb untreated n=1691, PMA treated n=2724; eGFP- Numb2A untreatedd- n=1186; PMA-treated n=2783.

Figure 5.6. Absolute PR as a function of cell intensity. Absolute majajor PR of MLA cells expressing eGFP-Numb plotted agaainst the integrated intensity for each cell. Each different color of the dots represents individual frames from a single tracked cell. For example, a cluster of data points (brown color) come from a very bright ceell (8-12x104 A.U.)

122

Raz Shimoni Chapter 5 Analysis of Polarity but with no extreme polarization (0.05-0.25), thus, demonstrates that high expression levels don’t necessarily affect the measured PR.

These data indicate that wild-type Numb is polarized to the uropod of migrating MLA cells, and Numb polarization is not abrogated by mutation of the aPKC target sites. Moreover, the ability to compare between Numb WT and mutant in gated subpopulations using the minor axis as a control demonstrates another successful application of the approaches that were developed in sections 5.4-5.6, and importantly, it shows the usability of TACTICS to solve biological questions.

5.3.4. Measurements of polarization by alternative axis of polarity

An advantage of TACTICS is its adjustability to measure other forms of polarity. For instance, TACTICS was easily applied to confirm that PR values are not caused by cytoplasmic Numb is being excluded from the cell body by the large nucleus in T cells (if eGFP- Numb2A had been delocalized from the cortex to the cytoplasm) by measuring the localization of Numb within the uropod.

A representative image of a MLA cell with extended uropod expressing eGFP-Numb2A (Figure 5.7.A) shows adaptation of the axial subdivision approach. It can be seen that Numb is predominantly cortical and appeared to be recruited to the tip of the uropod. To quantify this observation the location of the MTOC (highest intensity pixel of cherry fluorescence) was used to identify the region of the cell between the nucleus and the uropod tip, bisected this area, and then determined the ratio of eGFP fluorescence in the base versus the tip of the uropod (Figure 5.7.B).

This analysis avoided potential confounding effects from the nucleus, and still showed significant PR values for both wild-type and mutant Numb (Figure 5.7.C). The utilization of the Polarization Module to compare between the polarization of Numb WT and mutant in gated subpopulation based on the axis of uropod, exemplifies its ability to extend the axial subdivision approach to any given axis of polarity (as explained in Section 5.3), together with the utilization of its perpendicular minor axis as internal control.

123

Raz Shimoni Chapter 5 Analysis of Polarity

Figure 5.7. Comparison of fluoresscent intensity between the base and tip of the uropod. (A) Image of representative cell expressing eGFP- Numb2A and mCherry-tubulin showing cortical fluorescence at green channel, predominantly at the tip of thee uropod. The MTOC can be easily detected as the brighhttest spot in cherry channel, allowing the identification of the rear of the cell. (B) Shows the approach for bisecting the uropod region. (C) Shows Absolute PR of eGFP-Numb and eGFP- Numb2A (major and minor axis) calculated using only the region of the cell between tthe MTOC and the tip of the uropod..

124

Raz Shimoni Chapter 5 Analysis of Polarity

5.3.5. Measurements of polarization in migrating DN3 thymocytes

In this section I compare between the polarization of eGFP-Numb and eGFP- Numb2A in DN3 thymocytes during migration. DN3 thymocytes, cell that are physiologically more relevant than the MLA cell line. Similar to the T cell analysis above, thymocytes were cultured in microfabricated microwells, imaged by time- lapse confocal microscopy, and analyzed with each cell in each frame treated as a single data point (eGFP-Numb, n= 51349 data points, 1159 tracked cells; eGFP- Numb2A, n=77137 data points, 2731 tracked cells). In this instance, rather than mCherry-tubulin as a positive control for polarization, mCherry alone was used for tracking and segmentation, and to control for any possible artefacts of shape differences between the two hemispheres of the cell. Images of uropod-containing cells again suggested that both eGFP-Numb, and eGFP- Numb2A (but not mCherry) were polarized to the uropod (Figure 5.8.A).

Histograms describing PR for the entire thymocyte population showed little or no evidence of polarization of either protein (as would be indicated by a shift to the right for orientation along the major axis compared with orientation along the minor axis) (Figure 5.8.B). Instead of selecting for polarized cells on the basis of tubulin polarization as with the MLA cells, two alternative gating parameters were explored: the aspect ratio as parameter for elongation, and the distance that the cell travels as a quantitative characteristic of cell migration. Since round cells cannot be properly aligned by using the determination of the long and short axes of an ellipsoid, it is important to determine whether uneven Numb distributions could be averaged out, providing an aberrantly low PR value.

To formally test that the round cells were not polarized, I randomly chose 3 different representative round and still cells (Figure 5.8.C.i) and rotate each cell 360 degrees (every 30 degrees), where for each rotation the absolute PR was calculated. It can be seen from Figure 5.8.C.ii that the variations in PR are very small (0-0.08) in comparison to the elongated migrating cells (which can reach 0.5). Since even the highest PR in round and still cells is much lower than the migrating cells, this shows that Numb is indeed evenly distributed in round cells.

125

Raz Shimoni Chapter 5 Analysis of Polarity

The aspect ratio (ratio of length of major axis to length of minor axis) was plotted against the apparent speed (over frame -1 to frame +1, representing the frames before and after each data point) and gated for cells that were elongated (aspect ratio less than 0.6) and fast-moving (speed greater than 2 m/min) (Figure 5.8.D, green boxes). Relative frequency histograms of the PR indicated that elongated, fast-moving cells were clearly polarized for both eGFP-Numb (median PR 0.12) and eGFP- Numb2A (median PR 0.21) along the major axis as compared with the Cherry controls (median PR 0.05 and 0.06 respectively) (Figure 5.8.D.i). Notably, round cells don’t exhibit Numb polarization as shown in Figure 5.8.D.ii. Another ability of the Polarization Module is to inspect the data on the basis of individual groups within the dataset. This ability was applied to display the consistent difference in each of the 9 experiments from which these data were derived (Figure 5.9, P<0.0001). Interestingly, the polarity was again enhanced in eGFP- Numb2A compared with eGFP-Numb.

In contrast to elongated, fast-moving cells, gating for round cells (aspect ratio greater than 0.8) of all migration speeds, (red boxes) showed no polarization of either protein. This analysis shows that, as observed in the MLA cells, wild-type Numb was polarized to the uropod of migrating DN3 thymocytes, and Numb polarization was not abrogated by mutation of the aPKC target sites. Evidently, if the mechanisms of polarization in T cells were conserved, the polarization of Numb to the lymphocyte uropod would be expected to be abrogated by mutation of the two serine residues. However, this is not the case for either MLA cells or thymocytes [2]. The different polarization of Numb2A in T cells was not related to the lack of phosphorylation of Numb in T cells, because an antibody to phospho-Ser7 of Numb reacted with the wild-type but not the mutant form of the protein [2]. This, combined with our observation that the Numb2A mutant showed greater polarization in T cells than wild-type Numb, suggests the possibility that, instead of promoting polarization, aPKC might inhibit polarization in T cells.

126

Raz Shimoni Chapter 5 Analysis of Polarity

Figure 5.8. Numb and Numb2A are both polarized to the uropod in DN3 thymocytes. (A) DN3 thymocytes transduced with eGFP-Numb and eGFP- Numb2A were imaged by time-lapse fluorescence microscopy, and individual cell images were assembled into montage libraries using the TACTICS interface. Three representative examples of eGFP- Numb and eGFP- Numb2A are shown, in the DIC (grey), mCherry (red) and eGFP (green) spectral channels, automatically oriented with the major axis aligned vertically. Scale bar=10

127

Raz Shimoni Chapter 5 Analysis of Polarity

μm for all images. (B) All data points from imaged thymocytes expressing eGFP-Numb (left panels) or eGFP-Numb-2A (right panels) were plotted for aspect ratio (X-axis) against speed (Y-axis), and these plots were used to gate for polarized cells (green boxes). (C) To validate that eGFP- Numb is indeed symmetric in round cells, 3 representative cells were selected from the gated region of round and still cells. (i) images of GFP-Numb (left column) and GFP-Numb colocalized with Cherry (right column) are shown for each cell. (ii) Each cell was rotated in increments of 30 degrees and the PR was calculated from each angle. Color (green, red, blue) are corresponding to each of the cells (see legend in the left column). Maximum PR reached 0.08, which is substantially lower than the PR of the faster elongated cells (as shown below).(D) The PR values along the major axis for gated cells were plotted as a relative distribution histogram for eGFP-Numb (red lines, solid for eGFP-Numb expressing thymocytes (0.039) and dotted for eGFP- Numb2A expressing thymocytes (0.044)), and for mCherry (blue lines, solid for eGFP-Numb expressing thymocytes (0.069) and dotted for eGFP- Numb2A expressing thymocytes (0.059)). Events analyzed: eGFP- Numb n= 51349 data points, 1159 tracked cells; eGFP- Numb2A, n=77137 data points 2731 tracked cells.

Alternatively, aPKC might act on Numb via residues other than the two mutated in Numb2A, and presently up to 40 potential phosphorylation sites have been identified for Numb[281]. These data are also compatible with the notion that aPKC might act upon Numb indirectly, as has been shown in Drosophila neuroblasts where aPKC phosphorylates Lgl to orchestrate polarization of Numb[297]. These data clearly indicate that the mechanisms previously defined for Numb polarization in mammalian epithelial cells and Drosophila sensory organ precursors are not conserved in T cells.

To conclude, I show that residues of Numb that are phosphorylated by aPKC to mediate apicobasal polarity in epithelial cells are not required for polarization of Numb in T cells, suggesting that the role of aPKC is not conserved between T cells and epithelia.

128

Raz Shimoni Chapter 5 Analysis of Polarity

Figure 5.9. Gated fast migrating and elonngated DN3 thymocytes on experiiment base Numb and Numb2A are both polarized to the uropod in DN3 thymocytees. (A) Raw ungated thymocyte populations rotated along the major (blue shades) and minor axis (red shades) or (B) gated on migrating and elongated cells, also rotated along the major (blue shades) and minor axis (blue shades). (C) Coomparison of median PR over 9 experiments of raw or gated thymocyte populations. A non-parametric Mann Whitney test (two--tailed) was used to compare median PR of gated thymocyyte populations (minor: denoted X1-9 vs. major: denoted Y1-9), P<0.0001.

129

Raz Shimoni Chapter 5 Analysis of Polarity

5.3.6. Utilization of TACTICS to measure polarity in dividing DN3 Thymocytes

As explained in Chapter 4, arbitrary allocation of divisions into ACD and SCD can be misleading, and measurements of polarity during division require a careful analysis. Therefore, a new tool was developed to analyse ACD based on the axial subdivision approach as objective control and integrated in TACTICS. In sections 5.5-5.7 this approach was demonstrated as very effective even when the cells changed morphology during migration. It was shown that Numb is polarized in migrating MLA and DN3 thymocytes, and that Numb2A is even more polarized. The final step in this study leads to the utilization of TACTICS to analyze polarization of Numb and Numb2A in dividing DN3 thymocytes.

For this analysis, the same data set of migrating DN3 thymocytes was used with two alterations (as explained in section 5.2.14). Firstly, montages were generated based on parental cell and its corresponding two daughter cells. Secondly, PR ratios were ascribed to cell division, similarly to Chapter 4. Representative montages of eGFP-Numb (Figure 5.10.A.i) and Cherry control (Figure 5.10.A.ii) are displayed similarly to the montages of migrating T cells (Figure 5.1.c). Figure 5.10.A.i shows that Numb is mainly localized to the cortex of the cells, with a lower fluorescence originated from the nucleus and cytoplasm. Data images from the corresponding montages are compressed into a time projection (Figure 5.10.B). Evidently, while the projection of Cherry control (Figure 5.10.B.ii) looks symmetric and homogeneous, the projection of eGFP-Numb (Figure 5.10.B.ii) looks very heterogeneous and is not easy to resolve whether it is symmetric or asymmetric. This temporal analysis indicates that imaged cells can exhibit some degree of polarization over time, so averaging the PR over several time points provides more consistent analysis. Next, the PRmajor and PRminor where calculated, and averaged for the first 20 frames after the division from 30 dividing thymocytes expressing eGFP-Numb and eGFP-Numb2A. Since the ratios are measured in absolute values, averaging doesn’t critically influence the values, as long as the number of frames is not extreme. Over time, the polarization ratios can increase due to different location of the cells during imaging and this should be assessed carefully for each experiment. As demonstrated in Chapter 4, the combination of the axial subdivision approach and graphical

130

Raz Shimoni Chapter 5 Analysis of Polarity display is a reliable approach to select the level of threshold. Figure 4.10.C shows the dependency of the PR in the threshold level; (i) PRmajor and (ii) PRminor measured for eGFP-Numb and (iii) PRmajor and (iv) PRminor measured for eGFP-Numb2A. At the threshold of 15% the PRminor increases in polarization, and since the PRminor should be minimized, this threshold was set for quantitative comparison between Numb and Numb2A. To facilitate comparison between populations, and to discriminate between symmetric cell division (SCD) and ACD on the basis of PR values, the PRmajor versus PRminor were plotted on T=15% (Figure 4.10.D). Since most of the PRminor (the 90 percentile) in both Numb and Numb2A were smaller than PR<0.146, a PR value of 0.146 was set as a cut-off to gate on cells with PRminor that is smaller and PRmajor that is higher than that value. Within the region defined as asymmetric (upper pink square), 36.7% (11/30) of the Numb were assigned as ACD, compared with 16.7% (5/30) of the Numb2A assigned as ACD.

These results indicate that DN3 thymocytes can undergo both symmetric and asymmetric divisions for both Numb and Numb2A during migration. Surprisingly, in contrast to the data shown in Sections 5.5-5.7, where Numb2A was shown to be more polarized to the uropod, Numb2A gives a lower percentage of ACD. While ectopic expression of wild-type Numb does not grossly impact on DN3 fate decisions[2, 276], expression of a Numb2A (in which two serine substrates for phosphorylation by the polarity protein aPKC were mutated), showed defective progression from DN to DP [2]. Since aPKC regulates the polarization of Numb during ACD of many cell types including T cells [34], these results suggest that aPKC might also regulate Numb polarization in DN3 cells to control DN3 fate decisions. The difference in polarity during division between Numb and Numb2A suggests that Numb requires aPKC phosphorylation sites for polarity during DN3 divisions, and raises two important points. Firstly, it is still unclear whether the mode of DN3 division (asymmetric or symmetric) depends on the stage of differentiation. Secondly, quantification of the interaction between cells and its microenvironment (niche) is important to understand cellular functional states and to elucidate molecular mechanisms [9, 34, 298, 299]; for example, our laboratory demonstrated that polarity proteins interact with key proteins, downstream the TCR signals to dictate their localization and function, both at the immunological synapse and at the distal pole [34].

131

Raz Shimoni Chapter 5 Analysis of Polarity

Figure 5.10. Exemplification of TACTICS to measure ACD in T cells. (A) Representaative montage of dividing DN3 thymocytes expressing eGFP-Numb (i) and Cherry control (ii). Numb is majorly localized to the cortex, and therefore background fluorescence is required to be excluded from the analysis. (B) Time projections of eGFP-Numb (i) appeared to be very heterogeneous and polarized, while symmetriic and homogeneous projection is displayed for Cherry control (ii). (C) TACTICS provides polarity measurements based on normalizing the amount fluorescence across the major axis as internal control for symmetry. In this representative example, PR valuues were averaged for the first 20 frames after the division from 30 dividing thymocytes expressing, and were plotted as function of threshold value. (i)PRmajor and (ii)PRminor measured for eGFP-Numb minor minor and (iii) PRmajor and (iv)PR measured for eGFP-Numb2A. At T=15% the PR increases in polarization, and since the PRminor should be minimized, thee threshold was set for quantitative comparison between Numb and Numb2A. (D) PRmajor from (C) are plotted versus PRminor on T=15%. The 90 percentile of PRminor of both Numb annd Numb2A (0.146) is used as cut-off to gate on cells witth small PRminor and high PRmajor. The pink square defines this region. 36.7% (11/30) of the Nuumb within this region are assigned as ACD, in compared with 16.7% (5/30) of the Numb2A divisions. Importantly, a larger number of cells should be

132

Raz Shimoni Chapter 5 Analysis of Polarity analysed to reach statistical significance of the difference, and the complete study carried by Dr. Kim Pham can be found in [273].

Thirdly, complex morphology (such as multiple protrusions) of stromal cells prevent discrimination between proximal or distal T cells, and requires the development of innovative imaging techniques and improved segmentation that will potentially allow reliable measurement of stromal-thymocytes interactions from time-lapse images.

5.4. Summary

Presented in this chapter is the development and utilization of TACTICS to quantify cell polarity in migrating and dividing T cells. The interactive nature of TACTICS allows a deep interrogation of the raw data, and offers an unbiased means to exclude noise and extract robust measures of polarity. The importance of these new attributes is illustrated by our findings here, where analysis of the entire population of cells imaged by time-lapse microscopy shows a variable pattern of fluorescence that did not yield clear differences in the PR from the major and minor axes (Figure 5.4). This variability in PR was evidently caused by variation in activity (such as migration, division, apoptosis, interacting with microwell walls), as well as the orientation of movement, with both factors increasing noise in the data. But, by simply gating for cells of a particular shape that were moving at a reasonable speed, polarization patterns became clear. The interactive analysis and the biological results exemplify the applicability of TACTICS as a computational platform with a diverse range of applications. This is particularly highlighted in ICB Landmark 2014 [300]; these innovative tools offers the latest approach for objective standardization and truthful measurements of polarity in T cells. Future work is required to develop new assays to (a) monitor polarity in relation to multiple proteins, (b) monitor polarity in relation to the orientation of the thymocytes and the stromal cells, and (c) study polarity using in vivo imaging, which will take into consideration the true microenvironment of the imaged cells. The development of the approaches described above is currently ongoing in our laboratory, and TACTICS will continue to be at the core of this analysis.

133

Conclusions and Future Work

6. Chapter 6

Raz Shimoni Chapter 6 Conclusions

6.1. Thesis outline

This thesis demonstrates the development and utilization of the TACTICS Toolbox to investigate whether molecular mechanisms that were already established in other cells are also conserved in T cells.

Chapter 3 describes the methodological aspects and utilization of TACTICS Toolbox, such as multi-position reconstruction of lineage analysis. Specifically, it is demonstrated that TACTICS is a versatile and advanced computational platform that integrates high-content analysis and image cytometry based techniques. Although the biological results shown in this chapter are still at a preliminary stage, they highlight the ability of T cells to adopt different characteristics along several generations. This observation supports the model by which T cells can differentiate to provide various subsets, adjustable to the requirements of the body. Chapter 4 describes the performed simulations of dividing T cells that were undergoing symmetric or asymmetric cell division. TACTICS was utilized to demonstrate that the current routine approaches for the quantification of ACD can be misleading. Therefore, a new method was developed and validated based on axial subdivision, where the cell is divided into minor and major axes and the level of fluorescence is compared to an internal control. The ratio across the minor axis is employed to estimate the noise in the PR measurements and for choosing the most appropriate analysis settings. Thus, I demonstrated that the use of analytical approaches leads to obtaining more reliable measurements of polarity (Shimoni et al, 2014 in press). In Chapter 5, TACTICS was employed for the quantification of polarity proteins in migrating and dividing T cells. Firstly, the utilization of an axial subdivision approach enabled the exclusion of dependency of fluorescence ratiometric measurements on cell morphology and migration. By resolving that dependency, the polarity of Numb WT and mutant proteins could be compared. This comparison revealed that the role of aPKC in polarity is not conserved between T cells and epithelia during migration, as published[2]. Secondly, TACTICS was used to investigate the polarity of T cells during division. In contrast to the non-conserved role of aPKC in polarity during migration, the role of aPKC in polarity during division might be conserved, suggesting a role as fate protein during DN3 thymocyte division. Further work, including classical approaches to genetic modification, phenotyping and functional

135

Raz Shimoni Chapter 6 Conclusions analysis of cell populations, should be utilized to support these findings. This work is not in the scope of this thesis.

6.2. Conclusions

Overall, this thesis presents the development of a novel computational toolbox, namely TACTICS, and its verification, exemplified with real data. Specifically, TACTICS contains dedicated user interfaces for interactive exploration with respect to the type of biological question, providing the user with a resource that allows a detailed interpretation of biological patterns. An early version of TACTICS was published in [303], and an upgraded version is currently under preparation for publication. The advantageous capabilities of TACTICS are described below.

Since the number of combinations of parameters and different types of exploration tools is high, TACTICS is capable of addressing several biological parameters simultaneously allowing the user to probe the relationships between parameters. In addition, TACTICS alleviates user difficulties in the investigation of optimal image processing settings. With the ability of modern microscopes to capture multiple movies, TACTICS allows the robust extraction of multi-parameter information from many simultaneously imaged cells, originating from multiple- positions imaging techniques. Consequently, TACTICS pipeline facilitates high detection efficiency and tools for quality control and human intervention.

As the latest cutting-edge time-lapse assays set increasingly challenging analysis demands, the development of new computational approaches for systematic and reliable analysis cannot fall behind. Therefore, modular and flexible software applications should endorse the adaption of new methods. TACTICS has demonstrated the ability to be modularly extended upon this type of analysis, which makes TACTICS a good candidate for biological applications even further beyond those demonstrated in this thesis. The user interface of each TACTICS module can easily be updated, with new components, including future tracking and segmentation algorithms, able to be installed.

As highlighted by Carpenter et al. [301], the development of bioimaging informatics applications should have emphasis on software usability. One important

136

Raz Shimoni Chapter 6 Conclusions aspect of software usability is that scientific data analysis should not be ‘black box’, but open-code [184]. In that regard, the TACTICS code is completely available, and is easily validated, modified and tailored to answer specific biological questions. All programming operations, including the potential incorporation of multi-spectral intracellular protein localizations, cell segmentation, and tracking parameters, are described in detail and supported by the TACTICS website (www.tactics- toolbox.com). This website contains a user guide (Appendix I.) that includes comprehensive installation and operation instructions, data samples, video tutorials and raw data, including test images. This gives TACTICS users the flexibility to integrate other image processing and cell segmentation methods, to adopt new functions such as the tracking algorithm, and to integrate additional cell features to be analysed. Thus, TACTICS has the potential to be adapted by MATLAB developers in the future. While MATLAB experts can benefit from its flexibility for developers and incorporate additional bioinformatics processes and analysis, biologists with limited MATLAB programming may find these tools easier to use.

Since TACTICS runs within MATLAB, it can interact directly with other toolboxes such as the MATLAB IPT, and be compatible with various MATLAB- based applications, both commercial and open-code. An important aspect of TACTICS is the excellent interface that MATLAB provides to develop new algorithms and techniques for fluorescence-based image analysis. While third-party software is required to export data between software applications (e.g. Imaris can use the MATLAB IPT, ICY can import compiled MATLAB functions), TACTICS users stay within the environment of MATLAB that provides powerful tools such as for image analysis and statistics. I predict that MATLAB-based applications will continue to prevail in the future, with the ability to further upgrade upon demand. As a MATLAB based platform, TACTICS has the ability to incorporate new approaches quickly without the need to handle rigorous programming and debugging. Additionally, TACTICS can be easily customized, or integrated within other software, providing valuable tools for scientists.

137

Raz Shimoni Chapter 6 Conclusions

6.3. Future Work

Several other bioimaging informatics software applications have been made in MATLAB recently, and integration could bring the advantages from both. Immediate tasks include further optimization of the code, debugging where it is required, and improving the documentation of TACTICS user guide and comments within the code. Other tracking algorithms and alternative cell segmentation approaches should be incorporated to enable the potential value of TACTICS for other type of cells and imaging methods. For instance, algorithms that consider other migration patterns or cell parameters [139, 302] [140, 141], and bulb segmentation algorithm for confluent migrating cells using customized iterative mask-based cell segmentation [138], that were demonstrated to be extremely useful. Notably, memory is still an issue, as each lineage consists of ten to hundreds megabytes. Therefore, to enable higher numbers of founder cells and multiple gating approach future work should incorporate methods for improved memory management, for instance, Lazy evaluation as implemented by Endrov software [303]. In brief, Lazy evaluation is a strategy to increase performance and minimizing memory usage by avoiding processing and extraction of data until it is inquired. Middle range goals are to improve automated capabilities, such as automated recognition of cell divisions and to develop new TACTICS modules. Since TACTICS has a modular structure, it can be readily improved with regard to more customized assays, such as new implementations to provide insight into the molecular mechanism of stem cell differentiation, drug discovery and cancer immunotherapy. For example, similar to the microwell approach that was demonstrated in this thesis, screening individual cells isolated from cancer patients may reveal how cell behaviours, such as proliferation, apoptosis and differentiation contribute to tumour progression, or may be used to test therapeutic effects of drugs on cancer cells. Additionally, since TACTICS is close to reaching the maximum capabilities provided by a personal computer, stronger computers, further improvements of the programming approaches, and more internet control, including remote access to data are desirable. Although TACTICS can provide measurements and analysis based on z section, or averaging of 2-D images from 3-D data, the current version of TACTICS lacks the capacity for 3-D analysis. Therefore, another promising objective is to provide TACTICS with new capabilities that will allow a 3-D analysis similarly

138

Raz Shimoni Chapter 6 Conclusions demonstrated in this thesis with 2-D. Perhaps this requires the integration between MATLAB and another programing environment, such as C or C++ that are more suitable to 3-D analysis

Longer term possible directions might be the connection between TACTICS and hardware, either robotic micropipette or microfluidics. Single cell deposition with a robotic system, such as recently demonstrated [304], could potentially benefit from such integration. However, despite the success of the microwells that were utilized to prolong time-lapse imaging of T cells in this thesis work, open microfabricated devices suffer from several major drawbacks when image over a period longer than few days. In particular, the lacks of proper media exchange and over confluency are the major drawbacks. Compared with conventional cell culturing methods, "Lab-on-a-Chip" based microfluidics can supply a continuous perfusion to maintain better culturing conditions and support even longer time-lapse microscopy experiments. Moreover, the ability of live cell micromanipulation, deposition, and trapping in high-throughput is now scientific reality [305]. In comparison to micropipetting and optical methods (such as optical tweezers to capture individual cells [306]), microfluidics using hydrodynamic cell traps can give the ability to confine thousands of single-cells in high throughput manner. For these advantages and more, "Lab-on-a-Chip" has gained recognition as a fundamental tool in biological applications. A microfluidics device that can utilize the separation between daughter cells at the time of division might provide an efficient separation of two daughter cells, and clearly, a system comprised of dedicated software bundled with microfluidics device can be valuable to automate lineage trees. Expanding TACTICS to provide a platform for real-time signalling analysis will require an advanced real-time mode, but the basis for such development is already established. These new technologies will enable the exploration of new correlations between polarization and other cell properties and cell fate, and to further elucidate how cell fate is controlled by cell polarity. To conclude, this work opens a window of opportunities to study the interaction between protein localization and cell fate determination in many biological systems. Along with the arrival of new state-of-the- art microscopes to the field, new challenges and demands for quantitative image processing are emerging, and TACTICS will be useful for many other applications in the future.

139

References

1. Shimoni, R., et al., TACTICS, an interactive platform for customized high-content bioimaging analysis. Bioinformatics, 2013. 29(6): p. 817-8. 2. Pham, K., et al., Divergent lymphocyte signalling revealed by a powerful new tool for analysis of time-lapse microscopy. Immunol Cell Biol, 2013. 91(1): p. 70-81. 3. Liston, A., A. Enders, and O.M. Siggs, Unravelling the association of partial T-cell immunodeficiency and immune dysregulation. Nat Rev Immunol, 2008. 8(7): p. 545- 558. 4. Muller, L., T. Fulop, and G. Pawelec, Immunosenescence in vertebrates and invertebrates. Immun Ageing, 2013. 10(1): p. 12. 5. Zhang, N. and M.J. Bevan, CD8(+) T cells: foot soldiers of the immune system. Immunity, 2011. 35(2): p. 161-8. 6. Chang, J.T., Polarity and lymphocyte fate determination. Current opinion in cell biology, 2012. 24(4): p. 526-533. 7. Russell, S., How polarity shapes the destiny of T cells. J Cell Sci, 2008. 121(Pt 2): p. 131-6. 8. Germain, R.N., E.A. Robey, and M.D. Cahalan, A decade of imaging cellular motility and interaction dynamics in the immune system. Science, 2012. 336(6089): p. 1676-81. 9. Muzzey, D. and A. van Oudenaarden, Quantitative time-lapse fluorescence microscopy in single cells. Annu Rev Cell Dev Biol, 2009. 25: p. 301-27. 10. Martin-Belmonte, F. and M. Perez-Moreno, Epithelial cell polarity, stem cells and cancer. Nat Rev Cancer, 2012. 12(1): p. 23-38. 11. Januschke, J. and C. Gonzalez, Drosophila asymmetric division, polarity and cancer. Oncogene, 2008. 27(55): p. 6994-7002. 12. Hawkins, E.D. and S.M. Russell, Upsides and downsides to polarity and asymmetric cell division in leukemia. Oncogene, 2008. 27(55): p. 7003-17. 13. Humbert, P.O., et al., Control of tumourigenesis by the Scribble/Dlg/Lgl polarity module. Oncogene, 2008. 27(55): p. 6888-907. 14. Pham, K., F. Sacirbegovic, and S.M. Russell, Polarized Cells, Polarized Views: Asymmetric Cell Division in Hematopoietic Cells. Front Immunol, 2014. 5: p. 26. 15. Miller, J.F., Immunological function of the thymus. Lancet, 1961. 2(7205): p. 748-9. 16. Robey, E. and B.J. Fowlkes, Selective events in T cell development. Annu Rev Immunol, 1994. 12: p. 675-705. 17. Petrie, H.T. and J.C. Zúñiga-Pflücker, Zoned out: functional mapping of stromal signaling microenvironments in the thymus. Annu. Rev. Immunol., 2007. 25: p. 649- 679. 18. Trampont, P.C., et al., CXCR4 acts as a costimulator during thymic beta-selection. Nat Immunol, 2010. 11(2): p. 162-70. 19. Janas, M.L., et al., Thymic development beyond beta-selection requires phosphatidylinositol 3-kinase activation by CXCR4. J Exp Med, 2010. 207(1): p. 247-61, S1-2. 20. Banchereau, J. and R.M. Steinman, Dendritic cells and the control of immunity. Nature, 1998. 392(6673): p. 245-252. 21. Blanchard, N. and C. Hivroz, The immunological synapse: the more you look the less you know. Biol Cell, 2002. 94(6): p. 345-54. 22. Grakoui, A., et al., The immunological synapse: a molecular machine controlling T cell activation. Science, 1999. 285(5425): p. 221-7. 23. Krummel, M.F. and M.D. Cahalan, The immunological synapse: a dynamic platform for local signaling. J Clin Immunol, 2010. 30(3): p. 364-72.

140

24. Kupfer, A., Signaling in the immunological synapse: defining the optimal size. Immunity, 2006. 25(1): p. 11-3. 25. Stoll, S., et al., Dynamic imaging of T cell-dendritic cell interactions in lymph nodes. Science, 2002. 296(5574): p. 1873-6. 26. Jameson, S.C. and D. Masopust, Diversity in T cell memory: an embarrassment of riches. Immunity, 2009. 31(6): p. 859-71. 27. Zhou, L., M.M. Chong, and D.R. Littman, Plasticity of CD4+ T cell lineage differentiation. Immunity, 2009. 30(5): p. 646-55. 28. Williams, M.A. and M.J. Bevan, Effector and memory CTL differentiation. Annu Rev Immunol, 2007. 25: p. 171-92. 29. Lanzavecchia, A. and F. Sallusto, Understanding the generation and function of memory T cell subsets. Curr Opin Immunol, 2005. 17(3): p. 326-32. 30. Joshi, N.S. and S.M. Kaech, Effector CD8 T cell development: a balancing act between memory cell potential and terminal differentiation. J Immunol, 2008. 180(3): p. 1309-15. 31. Restifo, N.P. and L. Gattinoni, Lineage relationship of effector and memory T cells. Curr Opin Immunol, 2013. 25(5): p. 556-63. 32. Gerlach, C., et al., One naive T cell, multiple fates in CD8+ T cell differentiation. J Exp Med, 2010. 207(6): p. 1235-46. 33. Stemberger, C., et al., A Single Naive CD8+ T Cell Precursor Can Develop into Diverse Effector and Memory Subsets. Immunity, 2007. 27(6): p. 985-997. 34. Oliaro, J., et al., Asymmetric cell division of T cells upon antigen presentation uses multiple conserved mechanisms. J Immunol, 2010. 185(1): p. 367-75. 35. Chang, J.T., et al., Asymmetric T lymphocyte division in the initiation of adaptive immune responses. Science, 2007. 315(5819): p. 1687-91. 36. Chang, J.T., et al., Asymmetric proteasome segregation as a mechanism for unequal partitioning of the transcription factor T-bet during T lymphocyte division. Immunity, 2011. 34(4): p. 492-504. 37. King, C.G., et al., T cell affinity regulates asymmetric division, effector cell differentiation, and tissue pathology. Immunity, 2012. 37(4): p. 709-20. 38. St Johnston, D. and J. Ahringer, Cell polarity in eggs and epithelia: parallels and diversity. Cell, 2010. 141(5): p. 757-74. 39. Roybal, K.T., et al., The actin-driven spatiotemporal organization of T-cell signaling at the system scale. Immunol Rev, 2013. 256(1): p. 133-47. 40. Ludford-Menting, M.J., et al., A network of PDZ-containing proteins regulates T cell polarity and morphology during migration and immunological synapse formation. Immunity, 2005. 22(6): p. 737-48. 41. Krummel, M.F. and I. Macara, Maintenance and modulation of T cell polarity. Nat Immunol, 2006. 7(11): p. 1143-9. 42. Dustin, M.L., A.K. Chakraborty, and A.S. Shaw, Understanding the structure and function of the immunological synapse. Cold Spring Harb Perspect Biol, 2010. 2(10): p. a002311. 43. Fung, I., S. Russell, M, , and J. Oliaro, Interplay of polarity proteins and GTPases in T lymphocyte function. . Clinical and Developmental Immunology, 2012. In press (2012). 44. Lin, J., et al., The polarity protein Par1b/EMK/MARK2 regulates T cell receptor- induced microtubule-organizing center polarization. J Immunol, 2009. 183(2): p. 1215-21. 45. Pike, K.A., S. Kulkarni, and T. Pawson, Immature T-cell clustering and efficient differentiation require the polarity protein Scribble. Proc Natl Acad Sci U S A, 2010. 108(3): p. 1116-21. 46. Round, J.L., et al., Scaffold protein Dlgh1 coordinates alternative p38 kinase activation, directing T cell receptor signals toward NFAT but not NF-kappaB transcription factors. Nat Immunol, 2007. 8(2): p. 154-61.

141

47. Round, J.L., et al., Dlgh1 coordinates actin polymerization, synaptic T cell receptor and lipid raft aggregation, and effector function in T cells. J Exp Med, 2005. 201(3): p. 419-30. 48. Zanin-Zhorov, A., et al., Scaffold protein Disc large homolog 1 is required for T-cell receptor-induced activation of regulatory T-cell function. Proc Natl Acad Sci U S A, 2012. 109(5): p. 1625-30. 49. Bilder, D., M. Li, and N. Perrimon, Cooperative regulation of cell polarity and growth by Drosophila tumor suppressors. Science, 2000. 289(5476): p. 113-6. 50. Dow, L.E., et al., hScrib is a functional homologue of the Drosophila tumour suppressor Scribble. Oncogene, 2003. 22(58): p. 9225-30. 51. Tonikian, R., et al., A specificity map for the PDZ domain family. PLoS Biol, 2008. 6(9): p. e239. 52. Nance, J. and J.A. Zallen, Elaborating polarity: PAR proteins and the cytoskeleton. Development, 2011. 138(5): p. 799-809. 53. Etienne-Manneville, S., Polarity proteins in migration and invasion. Oncogene, 2008. 27(55): p. 6970-80. 54. Sanchez-Madrid, F. and J.M. Serrador, Bringing up the rear: defining the roles of the uropod. Nat Rev Mol Cell Biol, 2009. 10(5): p. 353-9. 55. Knoblich, J.A., Asymmetric cell division: recent developments and their implications for tumour biology. Nat Rev Mol Cell Biol, 2010. 11(12): p. 849-60. 56. Neumuller, R.A. and J.A. Knoblich, Dividing cellular asymmetry: asymmetric cell division and its implications for stem cells and cancer. Genes Dev, 2009. 23(23): p. 2675-99. 57. Assémat, E., et al., Polarity complex proteins. Biochimica et Biophysica Acta (BBA) - Biomembranes, 2008. 1778(3): p. 614-630. 58. Ridley, A.J., et al., Cell Migration: Integrating Signals from Front to Back. Science, 2003. 302(5651): p. 1704-1709. 59. Morrison, S.J. and J. Kimble, Asymmetric and symmetric stem-cell divisions in development and cancer. Nature, 2006. 441(7097): p. 1068-74. 60. Rose, L.S. and K.J. Kemphues, Early patterning of the C. elegans embryo. Annu Rev Genet, 1998. 32: p. 521-45. 61. Rhyu, M.S., L.Y. Jan, and Y.N. Jan, Asymmetric distribution of numb protein during division of the sensory organ precursor cell confers distinct fates to daughter cells. Cell, 1994. 76(3): p. 477-91. 62. Morrison, S.J. and J. Kimble, Asymmetric and symmetric stem-cell divisions in development and cancer. Nature, 2006. 441(7097): p. 1068-1074. 63. Humbert, P.O., L.E. Dow, and S.M. Russell, The Scribble and Par complexes in polarity and migration: friends or foes? Trends Cell Biol, 2006. 16(12): p. 622-30. 64. Conchello, J.-A. and J.W. Lichtman, Optical sectioning microscopy. Nat Meth, 2005. 2(12): p. 920-931. 65. Yuste, R., Fluorescence microscopy today. Nat Meth, 2005. 2(12): p. 902-904. 66. Miyawaki, A., Visualization of the Spatial and Temporal Dynamics of Intracellular Signaling. Developmental cell, 2003. 4(3): p. 295-305. 67. Chalfie, M., et al., Green fluorescent protein as a marker for gene expression. Science, 1994. 263(5148): p. 802-5. 68. Miyawaki, A., Green fluorescent protein glows gold. Cell, 2008. 135(6): p. 987-90. 69. McGuinness, L.P., et al., Quantum measurement and orientation tracking of fluorescent nanodiamonds inside living cells. Nat Nano, 2011. 6(6): p. 358-363. 70. Chang, H.-C., Development and Use of Fluorescent Nanodiamonds as Cellular Markers, in Nanodiamonds. 2010, Springer. p. 127-150. 71. Giepmans, B.N., et al., The fluorescent toolbox for assessing protein location and function. Science, 2006. 312(5771): p. 217-24. 72. Valeur, B., Molecular fluorescence: principles and applications. 2013: John Wiley & Sons.

142

73. Lakowicz, J.R., Principles of fluorescence spectroscopy. 2009: Springer. 74. Petty, H.R., Fluorescence microscopy: established and emerging methods, experimental strategies, and applications in immunology. Microsc Res Tech, 2007. 70(8): p. 687-709. 75. Pawley, J.B., Handbook of biological confocal microscopy. 3rd ed. 2006, New York, NY: Springer. xxviii, 985 p. 76. Cox, G., Optical imaging techniques in cell biology. 2012: CRC Press. 77. Agard, D.A., Optical sectioning microscopy: cellular architecture in three dimensions. Annu Rev Biophys Bioeng, 1984. 13: p. 191-219. 78. Denk, W., J.H. Strickler, and W.W. Webb, Two-photon laser scanning fluorescence microscopy. Science, 1990. 248(4951): p. 73-6. 79. Minsky, M., US patent 3013467. 1961. 80. Minsky, M., Memoir on inventing the confocal scanning microscope. Scanning, 1988. 10(4): p. 128-138. 81. PETRÁŇ, M., et al., Tandem-scanning reflected-light microscope. JOSA, 1968. 58(5): p. 661-664. 82. Huang, B., M. Bates, and X. Zhuang, Super-resolution fluorescence microscopy. Annu Rev Biochem, 2009. 78: p. 993-1016. 83. Fan, G.Y., et al., Video-Rate Scanning Two-Photon Excitation Fluorescence Microscopy and Ratio Imaging with Cameleons. Biophysical Journal, 1999. 76(5): p. 2412-2420. 84. Davidovits, P. and M. Egger, Scanning laser microscope for biological investigations. Applied Optics, 1971. 10(7): p. 1615-1619. 85. Gu, M., Principles of three-dimensional imaging in confocal microscopes. 1996: World Scientific. 86. Nomarski, G., Microinterféromètre différentiel à ondes polarisées. J. phys. radium, 1955. 16(7): p. S9-S13. 87. Harris, S. and R. Wallace, Acousto-optic tunable filter. JOSA, 1969. 59(6): p. 744- 747. 88. Borlinghaus, R. and L. Kuschel, Spectral fluorescence lifetime imaging microscopy: new dimensions with Leica TCS SP5. Nature methods, 2006. 3(10). 89. Toomre, D. and J.B. Pawley, Disk-scanning confocal microscopy, in Handbook of biological confocal microscopy. 2006, Springer. p. 221-238. 90. Nakano, A., Spinning-disk confocal microscopy -- a cutting-edge tool for imaging of membrane traffic. Cell structure and function, 2002. 27(5): p. 349-55. 91. Yasuda, T., et al., Towards automatic construction of cell-lineage of C. elegans from Nomarski DIC microscope images. GENOME INFORMATICS SERIES, 1999: p. 144-154. 92. Young, I.T., et al., Fundamentals of image processing. 1998: Delft University of Technology The Netherlands. 93. Germain, R.N., et al., Dynamic imaging of the immune system: progress, pitfalls and promise. Nature Reviews Immunology, 2006. 6(7): p. 497-507. 94. Cahalan, M.D. and I. Parker, Choreography of cell motility and interaction dynamics imaged by two-photon microscopy in lymphoid organs. Annual review of immunology, 2008. 26: p. 585. 95. Melichar, H.J., et al., Quantifying subcellular distribution of fluorescent fusion proteins in cells migrating within tissues. Immunol Cell Biol, 2011. 89(4): p. 549-57. 96. Henrickson, S.E., et al., Antigen availability determines CD8(+) T cell-dendritic cell interaction kinetics and memory fate decisions. Immunity, 2013. 39(3): p. 496-507. 97. Bousso, P., et al., Dynamics of thymocyte-stromal cell interactions visualized by two-photon microscopy. Science, 2002. 296(5574): p. 1876-80. 98. Miller, M.J., et al., Two-photon imaging of lymphocyte motility and antigen response in intact lymph node. Science, 2002. 296(5574): p. 1869-73.

143

99. Le Borgne, M., et al., The impact of negative selection on thymocyte migration in the medulla. Nat Immunol, 2009. 10(8): p. 823-30. 100. Chen, Y., et al., Automated 5-D analysis of cell migration and interaction in the thymic cortex from time-lapse sequences of 3-D multi-channel multi-photon images. J Immunol Methods, 2009. 340(1): p. 65-80. 101. Ladi, E., et al., Thymocyte-dendritic cell interactions near sources of CCR7 ligands in the thymic cortex. J Immunol, 2008. 181(10): p. 7014-23. 102. Witt, C.M., et al., Directed migration of positively selected thymocytes visualized in real time. PLoS Biol, 2005. 3(6): p. e160. 103. Ladi, E., et al., Thymic microenvironments for T cell differentiation and selection. Nat Immunol, 2006. 7(4): p. 338-43. 104. Witt, C.M. and E.A. Robey, Thymopoiesis in 4 dimensions. Semin Immunol, 2005. 17(1): p. 95-102. 105. Azar, G.A., et al., Subcellular dynamics of T cell immunological synapses and kinapses in lymph nodes. Proc Natl Acad Sci U S A. 106. von Andrian, U.H. and C.R. Mackay, T-cell function and migration. Two sides of the same coin. N Engl J Med, 2000. 343(14): p. 1020-34. 107. Meijering, E., O. Dzyubachyk, and I. Smal, Methods for cell and particle tracking. Methods Enzymol, 2012. 504: p. 183-200. 108. Textor, J., et al., Defining the quantitative limits of intravital two-photon lymphocyte tracking. Proc Natl Acad Sci U S A, 2011. 108(30): p. 12401-6. 109. Beltman, J.B., A.F. Maree, and R.J. de Boer, Analysing immune cell migration. Nat Rev Immunol, 2009. 9(11): p. 789-98. 110. Castellino, F., et al., Chemokines enhance immunity by guiding naive CD8+ T cells to sites of CD4+ T cell-dendritic cell interaction. Nature, 2006. 440(7086): p. 890-5. 111. Hugues, S., et al., Dynamic imaging of chemokine-dependent CD8+ T cell help for CD8+ T cell responses. Nat Immunol, 2007. 8(9): p. 921-30. 112. Dixit, R. and R. Cyr, Cell damage and reactive oxygen species production induced by fluorescence microscopy: effect on mitosis and guidelines for non-invasive fluorescence microscopy. The Plant Journal, 2003. 36(2): p. 280-290. 113. Coutu, D.L. and T. Schroeder, Probing cellular processes by long-term live imaging--historic problems and current solutions. J Cell Sci, 2013. 126(Pt 17): p. 3805-15. 114. Harnett, M.M., Laser scanning cytometry: understanding the immune system in situ. Nat Rev Immunol, 2007. 7(11): p. 897-904. 115. Stephens, D.J. and V.J. Allan, Light Microscopy Techniques for Live Cell Imaging. Science, 2003. 300(5616): p. 82-86. 116. Rieder, C.L. and A. Khodjakov, Mitosis Through the Microscope: Advances in Seeing Inside Live Dividing Cells. Science, 2003. 300(5616): p. 91-96. 117. Neumann, B., et al., Phenotypic profiling of the human genome by time-lapse microscopy reveals cell division genes. Nature, 2010. 464(7289): p. 721-727. 118. Schroeder, T., Long-term single-cell imaging of mammalian stem cells. Nat Methods, 2011. 8(4 Suppl): p. S30-5. 119. Schroeder, T., The electronic crystal ball: predicting cell fate from time-lapse data. Nat Methods, 2010. 7(3): p. 190-1. 120. Beltman, J.B., A.F. Marée, and R.J. de Boer, Analysing immune cell migration. Nature Reviews Immunology, 2009. 9(11): p. 789-798. 121. Dorn, J.F., G. Danuser, and G. Yang, Computational Processing and Analysis of Dynamic Fluorescence Image Data, in Methods in Cell Biology, F.S. Kevin, Editor. 2008, Academic Press. p. 497-538. 122. Dustin, M.L., Stop and go traffic to tune T cell responses. Immunity, 2004. 21(3): p. 305-14. 123. von Andrian, U.H. and T.R. Mempel, Homing and cellular traffic in lymph nodes. Nat Rev Immunol, 2003. 3(11): p. 867-78.

144

124. Celli, S., F. Lemaitre, and P. Bousso, Real-time manipulation of T cell-dendritic cell interactions in vivo reveals the importance of prolonged contacts for CD4+ T cell activation. Immunity, 2007. 27(4): p. 625-34. 125. Gunzer, M., et al., Antigen presentation in extracellular matrix: interactions of T cells with dendritic cells are dynamic, short lived, and sequential. Immunity, 2000. 13(3): p. 323-32. 126. Beltman, J.B., A.F. Maree, and R.J. de Boer, Spatial modelling of brief and long interactions between T cells and dendritic cells. Immunol Cell Biol, 2007. 85(4): p. 306-14. 127. Rapoport, D.H., et al., A novel validation algorithm allows for automated cell tracking and the extraction of biologically meaningful parameters. PLoS One, 2011. 6(11): p. e27315. 128. Day, D., et al., A method for prolonged imaging of motile lymphocytes. Immunol Cell Biol, 2009. 87(2): p. 154-8. 129. Oliaro, J., et al., Ligation of the cell surface receptor, CD46, alters T cell polarity and response to antigen presentation. Proc Natl Acad Sci U S A, 2006. 103(49): p. 18685-90. 130. Zaretsky, I., et al., Monitoring the dynamics of primary T cell activation and differentiation using long term live cell imaging in microwell arrays. Lab Chip, 2012. 12(23): p. 5007-15. 131. Whitesides, G.M., et al., Soft lithography in biology and biochemistry. Annu Rev Biomed Eng, 2001. 3: p. 335-73. 132. Sia, S.K. and G.M. Whitesides, Microfluidic devices fabricated in poly(dimethylsiloxane) for biological studies. Electrophoresis, 2003. 24(21): p. 3563-76. 133. Xia, Y. and G.M. Whitesides, Soft Lithography. Angewandte Chemie International Edition, 1998. 37(5): p. 550-575. 134. Peng, H., Bioimage informatics: a new area of engineering biology. Bioinformatics, 2008. 24(17): p. 1827-1836. 135. Li, F., et al., Bioimage Informatics for Systems Pharmacology. PLoS computational biology, 2013. 9(4): p. e1003043. 136. Chi, K.R., Imaging and detection: focusing on software. Nature methods, 2008. 5(7): p. 651-658. 137. Schindelin, J., et al., Fiji: an open-source platform for biological-image analysis. Nat Methods, 2012. 9(7): p. 676-82. 138. Wang, Q., et al., Image segmentation and dynamic lineage analysis in single-cell fluorescence microscopy. Cytometry A, 2010. 77(1): p. 101-10. 139. Beaumont, L.M.A., J. Wakefield, and J.A. Noble. Spatiotemporal Bayesian cell population tracking and analysis with lineage construction. in Biomedical Imaging: From Nano to Macro, 2008. ISBI 2008. 5th IEEE International Symposium on. 2008. 140. House, D., et al. Tracking of cell populations to understand their spatio-temporal behavior in response to physical stimuli. in Computer Vision and Pattern Recognition Workshops, 2009. CVPR Workshops 2009. IEEE Computer Society Conference on. 2009. 141. Rajaram, S., et al., PhenoRipper: software for rapidly profiling microscopy images. Nat Meth, 2012. 9(7): p. 635-637. 142. Gonzalez, R.C., R.E. Woods, and S.L. Eddins, Digital Image processing using MATLAB. 2nd ed. 2009, Tennessee: Gatesmark Publishing. xviii, 826 p. 143. Burger, W. and M. Burge, Principles of digital image processing : fundamental techniques. Undergraduate topics in computer science,. 2009, London: Springer. xiv, 260 p. 144. Kass, M., A. Witkin, and D. Terzopoulos, Snakes: Active contour models. International journal of computer vision, 1988. 1(4): p. 321-331.

145

145. Chan, T.F. and L.A. Vese, Active contours without edges. Image Processing, IEEE Transactions on, 2001. 10(2): p. 266-277. 146. Zimmer, C. and J.C. Olivo-Marin, Coupled parametric active contours. IEEE Trans Pattern Anal Mach Intell, 2005. 27(11): p. 1838-42. 147. Yonggang, S. and W.C. Karl. Real-time tracking using level sets. in Computer Vision and Pattern Recognition, 2005. CVPR 2005. IEEE Computer Society Conference on. 2005. 148. Dzyubachyk, O., et al., Advanced level-set-based cell tracking in time-lapse fluorescence microscopy. IEEE Trans Med Imaging, 2010. 29(3): p. 852-67. 149. Otsu, N., A Threshold Selection Method from Gray-Level Histograms. IEEE Transactions on Systems, Man, and Cybernetics, 1979. 9(1): p. 62-66. 150. Zimmer, C., et al., On the digital trail of mobile cells. Signal Processing Magazine, IEEE, 2006. 23(3): p. 54-62. 151. Li, K., et al., Cell population tracking and lineage construction with spatiotemporal context. Med Image Anal, 2008. 12(5): p. 546-66. 152. Idema, T., A new way of tracking motion, shape, and divisions. Eur Biophys J, 2013. 42(8): p. 647-54. 153. Dufour, A., et al., Segmenting and tracking fluorescent cells in dynamic 3-D microscopy with coupled active surfaces. IEEE Trans Image Process, 2005. 14(9): p. 1396-410. 154. Crocker, J.C. and D.G. Grier, Methods of digital video microscopy for colloidal studies. J. Colloid Interface Sci., 1996. 179: p. 298-310. 155. Asratian, A.S., Bipartite graphs and their applications. 1998: Cambridge University Press. 156. Kachouie, N.N. and P.W. Fieguth, Extended-Hungarian-JPDA: Exact Single-Frame Stem Cell Tracking. Biomedical Engineering, IEEE Transactions on, 2007. 54(11): p. 2011-2019. 157. Downey, M.J., et al., Extracting fluorescent reporter time courses of cell lineages from high-throughput microscopy at low temporal resolution. PLoS One, 2011. 6(12): p. e27886. 158. Pepperkok, R. and J. Ellenberg, High-throughput fluorescence microscopy for systems biology. Nat Rev Mol Cell Biol, 2006. 7(9): p. 690-696. 159. Millard, B.L., et al., Adaptive informatics for multifactorial and high-content biological data. Nat Methods, 2011. 8(6): p. 487-93. 160. Walter, T., et al., Visualization of image data from cells to organisms. Nat Methods, 2010. 7(3 Suppl): p. S26-41. 161. Gehlenborg, N. and B. Wong, Points of view: Heat maps. Nat Meth, 2012. 9(3): p. 213-213. 162. Buchser, W., et al., Assay Development Guidelines for Image-Based High Content Screening, High Content Analysis and High Content Imaging. 2004. 163. Ramunas, J., et al., True monolayer cell culture in a confined 3D microenvironment enables lineage informatics. Cytometry A, 2006. 69(12): p. 1202-11. 164. Wiley, E.O. and B.S. Lieberman, Phylogenetics: theory and practice of phylogenetic systematics. 2011: John Wiley & Sons. 165. Brenner, S., The genetics of Caenorhabditis elegans. Genetics, 1974. 77(1): p. 71- 94. 166. Check, E., Worm cast in starring role for Nobel prize. Nature, 2002. 419(6907): p. 548-549. 167. Ankeny, R.A., The natural history of Caenorhabditis elegans research. Nat Rev Genet, 2001. 2(6): p. 474-9. 168. Sulston, J.E. and H.R. Horvitz, Post-embryonic cell lineages of the nematode, Caenorhabditis elegans. Dev Biol, 1977. 56(1): p. 110-56. 169. Sulston, J.E., et al., The embryonic cell lineage of the nematode Caenorhabditis elegans. Dev Biol, 1983. 100(1): p. 64-119.

146

170. Hird, S.N. and J.G. White, Cortical and cytoplasmic flow polarity in early embryonic cells of Caenorhabditis elegans. J Cell Biol, 1993. 121(6): p. 1343-55. 171. Schnabel, R., et al., Assessing Normal Embryogenesis inCaenorhabditis elegansUsing a 4D Microscope: Variability of Development and Regional Specification. Developmental Biology, 1997. 184(2): p. 234-265. 172. Giurumescu, C.A., et al., Quantitative semi-automated analysis of morphogenesis with single-cell resolution in complex embryos. Development, 2012. 139(22): p. 4271-4279. 173. Al-Kofahi, O., et al., Automated cell lineage construction: a rapid method to analyze clonal development established with murine neural progenitor cells. Cell Cycle, 2006. 5(3): p. 327-35. 174. Winter, M., et al., Vertebrate neural stem cell segmentation, tracking and lineaging with validation and editing. Nat Protoc, 2011. 6(12): p. 1942-52. 175. Papadimitriou, C.H. and K. Steiglitz, Combinatorial optimization : algorithms and complexity. 1998, Mineola, N.Y.: Dover Publications. xvi, 496 p. 176. Cohen, A.R., et al., Computational prediction of neural progenitor cell fates. Nat Methods, 2010. 7(3): p. 213-8. 177. Cohen, A.R., et al., Automatic summarization of changes in biological image sequences using algorithmic information theory. IEEE Trans Pattern Anal Mach Intell, 2009. 31(8): p. 1386-403. 178. Kamvar, S.D., D. Klein, and C.D. Manning, Spectral learning, in Proceedings of the 18th international joint conference on Artificial intelligence. 2003, Morgan Kaufmann Publishers Inc.: Acapulco, Mexico. p. 561-566. 179. Kobel, S. and M. Lutolf, High-throughput methods to define complex stem cell niches. Biotechniques, 2010. 48(4): p. ix-xxii. 180. Jain, S. and P. Heutink, From single genes to gene networks: high-throughput-high- content screening for neurological disease. Neuron, 2010. 68(2): p. 207-17. 181. Abraham, V.C., D.L. Taylor, and J.R. Haskins, High content screening applied to large-scale cell biology. Trends in biotechnology, 2004. 22(1): p. 15-22. 182. Conrad, C., et al., Micropilot: automation of fluorescence microscopy-based imaging for systems biology. Nat Methods, 2011. 8(3): p. 246-9. 183. Neumann, B., et al., Phenotypic profiling of the human genome by time-lapse microscopy reveals cell division genes. Nature, 2010. 464(7289): p. 721-7. 184. Kankaanpaa, P., et al., BioImageXD: an open, general-purpose and high-throughput image-processing platform. Nat Methods, 2012. 9(7): p. 683-9. 185. Conrad, C. and D.W. Gerlich, Automated microscopy for high-content RNAi screening. J Cell Biol, 2010. 188(4): p. 453-61. 186. Loo, L.H., L.F. Wu, and S.J. Altschuler, Image-based multivariate profiling of drug responses from single cells. Nat Methods, 2007. 4(5): p. 445-53. 187. Cohen, A.A., et al., Dynamic proteomics of individual cancer cells in response to a drug. Science, 2008. 322(5907): p. 1511-6. 188. de Chaumont, F., et al., Icy: an open bioimage informatics platform for extended reproducible research. Nat Methods, 2012. 9(7): p. 690-6. 189. Schneider, C.A., W.S. Rasband, and K.W. Eliceiri, NIH Image to ImageJ: 25 years of image analysis. Nat Meth, 2012. 9(7): p. 671-675. 190. Bradski, G., The OpenCV Library Dr. Dobb’s J. Software Tools, 2000. 25 (120 ): p. 122–125. 191. Hunter, J.D., Matplotlib: A 2D Graphics Environment. Computing in Science & Engineering, 2007. 9(3): p. 90-95. 192. Oliphant, T.E., Python for Scientific Computing. Computing in Science & Engineering, 2007. 9(3): p. 10-20. 193. Cardona, A. and P. Tomancak, Current challenges in open-source bioimage informatics. Nat Methods, 2012. 9(7): p. 661-5.

147

194. Kamentsky, L., et al., Improved structure, function and compatibility for CellProfiler: modular high-throughput image analysis software. Bioinformatics, 2011. 27(8): p. 1179-80. 195. Jones, T.R., et al., Scoring diverse cellular morphologies in image-based screens with iterative feedback and machine learning. Proc Natl Acad Sci U S A, 2009. 106(6): p. 1826-31. 196. Held, M., et al., CellCognition: time-resolved phenotype annotation in high- throughput live cell imaging. Nat Meth, 2010. 7(9): p. 747-754. 197. Gordon, A., et al., Single-cell quantification of molecules and rates using open- source microscope-based cytometry. Nat Methods, 2007. 4(2): p. 175-81. 198. Bjornsson, C.S., et al., Associative image analysis: a method for automated quantification of 3D multi-parameter images of brain tissue. J Neurosci Methods, 2008. 170(1): p. 165-78. 199. Schmidt, H. and M. Jirstrand, Systems Biology Toolbox for MATLAB: a computational platform for research in systems biology. Bioinformatics, 2006. 22(4): p. 514-5. 200. Ghosh, S., et al., Software for systems biology: from tools to integrated platforms. Nat Rev Genet, 2011. 12(12): p. 821-32. 201. Rajaram, S., et al., SimuCell: a flexible framework for creating synthetic microscopy images. Nat Meth, 2012. 9(7): p. 634-635. 202. Carpenter, A.E., et al., CellProfiler: image analysis software for identifying and quantifying cell phenotypes. Genome Biol, 2006. 7(10): p. R100. 203. Jaeger, S., Q. Song, and S.-S. Chen, DYNAMIK: a software environment for cell DYNAmics, Motility, and Information tracKing, with an application to Ras pathways. Bioinformatics, 2009. 25(18): p. 2383-2388. 204. Georgescu, W., J.P. Wikswo, and V. Quaranta, CellAnimation: an open source MATLAB framework for microscopy assays. Bioinformatics, 2012. 28(1): p. 138-9. 205. Li, F., et al., Multiple nuclei tracking using integer programming for quantitative cancer cell cycle analysis. Medical Imaging, IEEE Transactions on, 2010. 29(1): p. 96-105. 206. Wang, Q., et al., Image segmentation and dynamic lineage analysis in single-cell fluorescence microscopy. Cytometry Part A, 2010. 77A(1): p. 101-110. 207. Kaech, S.M. and W. Cui, Transcriptional control of effector and memory CD8+ T cell differentiation. Nat Rev Immunol, 2012. 12(11): p. 749-61. 208. Restifo, N.P. and L. Gattinoni, Lineage relationship of effector and memory T cells. Current opinion in immunology, 2013. 25(5): p. 556-563. 209. Arens, R. and S.P. Schoenberger, Plasticity in programming of effector and memory CD8+ T‐cell formation. Immunol Rev, 2010. 235(1): p. 190-205. 210. Farber, D.L., N.A. Yudanin, and N.P. Restifo, Human memory T cells: generation, compartmentalization and homeostasis. Nature Reviews Immunology, 2014. 14(1): p. 24-35. 211. Hermann, A., Analysis of asymmetric division of hematopoietic stem cells by continuous single cell observation. 2010, lmu. 212. Shen, Q., et al., Asymmetric Numb distribution is critical for asymmetric cell division of mouse cerebral cortical stem cells and neuroblasts. Development, 2002. 129(20): p. 4843-4853. 213. Gerlach, C., et al., Heterogeneous differentiation patterns of individual CD8+ T cells. Science, 2013. 340(6132): p. 635-9. 214. Buchholz, V.R., et al., Disparate individual fates compose robust CD8+ T cell immunity. Science, 2013. 340(6132): p. 630-5. 215. Plumlee, C.R., et al., Environmental cues dictate the fate of individual CD8+ T cells responding to infection. Immunity, 2013. 39(2): p. 347-56.

148

216. Amat, F. and P.J. Keller, Towards comprehensive cell lineage reconstructions in complex organisms using light-sheet microscopy. Dev Growth Differ, 2013. 55(4): p. 563-78. 217. Eilken, H.M., S. Nishikawa, and T. Schroeder, Continuous single-cell imaging of blood generation from haemogenic endothelium. Nature, 2009. 457(7231): p. 896- 900. 218. Rieger, M.A., et al., Hematopoietic Cytokines Can Instruct Lineage Choice. Science, 2009. 325(5937): p. 217-218. 219. Chakravorty, R., et al., Labour-efficient in vitro lymphocyte population tracking and fate prediction using automation and manual review. PLoS One, 2014. 9(1): p. e83251. 220. Arens, R. and S.P. Schoenberger, Plasticity in programming of effector and memory CD8 T-cell formation. Immunol Rev, 2010. 235(1): p. 190-205. 221. Goldberg, I.G., et al., The Open Microscopy Environment (OME) Data Model and XML file: open tools for informatics and quantitative analysis in biological imaging. Genome biology, 2005. 6(5): p. R47. 222. Kovesi, P.D., MATLAB and Octave Functions for Computer Vision and Image Processing, download from: www.csse.uwa.edu.au/~pk/Research/MatlabFns/index.html. 223. de Chaumont, F., et al., Icy: an open bioimage informatics platform for extended reproducible research. Nature methods, 2012. 9(7): p. 690-696. 224. Stoter, M., et al., CellProfiler and KNIME: open source tools for high content screening. Methods Mol Biol, 2013. 986: p. 105-22. 225. Li, C., et al., Distance regularized level set evolution and its application to image segmentation. Image Processing, IEEE Transactions on, 2010. 19(12): p. 3243-3254. 226. Vincent, L. and P. Soille, Watersheds in digital spaces: an efficient algorithm based on immersion simulations. Pattern Analysis and Machine Intelligence, IEEE Transactions on, 1991. 13(6): p. 583-598. 227. Altman, Y.M., Undocumented secrets of MATLAB-Java programming. 2012, Boca Raton, FL: CRC Press. xxi, 663 p., 16 p. of plates. 228. Kuhn, H.W., The Hungarian method for the assignment problem. Naval research logistics quarterly, 1955. 2(1‐2): p. 83-97. 229. Munkres, J., Algorithms for the assignment and transportation problems. Journal of the Society for Industrial & Applied Mathematics, 1957. 5(1): p. 32-38. 230. Cao, A.M.a.Y., The Hungarian algorithm, http://www.mathworks.com/MATLABcentral/fileexchange/20328. 2012. 231. Kanade, T., et al. Cell image analysis: Algorithms, system and applications. in Applications of Computer Vision (WACV), 2011 IEEE Workshop on. 2011. 232. Baye, L.M. and B.A. Link, Interkinetic nuclear migration and the selection of neurogenic cell divisions during vertebrate retinogenesis. J Neurosci, 2007. 27(38): p. 10143-52. 233. Ramsbottom, K.M., et al., Cutting Edge: DNAX Accessory Molecule 1-Deficient CD8+ T Cells Display Immunological Synapse Defects That Impair Antitumor Immunity. J Immunol, 2014. 192(2): p. 553-7. 234. Ramsbottom, K.M., et al., Cutting Edge: DNAX Accessory Molecule 1-Deficient CD8+ T Cells Display Immunological Synapse Defects That Impair Antitumor Immunity. J Immunol, 2013. 235. Murray, J.I., et al., The lineaging of fluorescently-labeled Caenorhabditis elegans embryos with StarryNite and AceTree. Nat Protoc, 2006. 1(3): p. 1468-76. 236. Glauche, I., et al., A novel view on stem cell development: analysing the shape of cellular genealogies. Cell Prolif, 2009. 42(2): p. 248-63. 237. Hawkins, E.D., et al., A single-cell pedigree analysis of alternative stochastic lymphocyte fates. Proc Natl Acad Sci U S A, 2009. 106(32): p. 13457-62.

149

238. Tyson, J.J. and O. Diekmann, Sloppy size control of the cell division cycle. Journal of theoretical biology, 1986. 118(4): p. 405-426. 239. Moreau, H.D., et al., Dynamic in situ cytometry uncovers T cell receptor signaling during immunological synapses and kinapses in vivo. Immunity, 2012. 37(2): p. 351- 63. 240. Moogk, D.R., Live cell imaging, cell tracking and lineage analysis as a tool to investigate dynamic culture processes in heterogeneous cell systems. 2009, University of Waterloo. 241. Pretorius, A.J., et al., Visualization of parameter space for image analysis. IEEE Trans Vis Comput Graph, 2011. 17(12): p. 2402-11. 242. Moreau, H.D., et al., Dynamic In Situ Cytometry Uncovers T Cell Receptor Signaling during Immunological Synapses and Kinapses In Vivo. Immunity, 2012. 243. Skamagki, M., et al., Asymmetric Localization of Cdx2 mRNA during the First Cell- Fate Decision in Early Mouse Development. Cell Rep, 2013. 3(2): p. 442-57. 244. Habib, S.J., et al., A localized Wnt signal orients asymmetric stem cell division in vitro. Science, 2013. 339(6126): p. 1445-8. 245. Dalton, C.M. and J. Carroll, Biased inheritance of mitochondria during asymmetric cell division in the mouse oocyte. J Cell Sci, 2013. 126(Pt 13): p. 2955-64. 246. Yan, Y., et al., Particles on the move: intracellular trafficking and asymmetric mitotic partitioning of nanoporous polymer particles. ACS Nano, 2013. 7(6): p. 5558-67. 247. Frigault, M.M., et al., Live-cell microscopy - tips and tools. J Cell Sci, 2009. 122(Pt 6): p. 753-67. 248. Waters, J.C., Accuracy and precision in quantitative fluorescence microscopy. J Cell Biol, 2009. 185(7): p. 1135-48. 249. Wu, M., et al., Imaging hematopoietic precursor division in real time. Cell stem cell, 2007. 1(5): p. 541-54. 250. Filby, A., et al., An imaging flow cytometric method for measuring cell division history and molecular symmetry during mitosis. Cytometry A, 2011. 79(7): p. 496- 506. 251. Bresenham, J.E., Algorithm for computer control of a digital plotter. IBM Systems journal, 1965. 4(1): p. 25-30. 252. Guberman, J.M., et al., PSICIC: noise and asymmetry in bacterial division revealed by computational image analysis at sub-pixel resolution. PLoS Comput Biol, 2008. 4(11): p. e1000233. 253. Effler, J.C., P.A. Iglesias, and D.N. Robinson, Regulating cell shape during cytokinesis, in Control of Uncertain Systems: Modelling, Approximation, and Design. 2006, Springer. p. 203-224. 254. Summers, H.D., et al., Statistical analysis of nanoparticle dosing in a dynamic cellular system. Nat Nanotechnol, 2011. 6(3): p. 170-4. 255. Rujano, M.A., et al., Polarised asymmetric inheritance of accumulated protein damage in higher eukaryotes. PLoS Biol, 2006. 4(12): p. e417. 256. Lehmussola, A., et al., Computational framework for simulating fluorescence microscope images with cell populations. IEEE Trans Med Imaging, 2007. 26(7): p. 1010-6. 257. Yang, Y.H., M.J. Buckley, and T.P. Speed, Analysis of cDNA microarray images. Brief Bioinform, 2001. 2(4): p. 341-9. 258. Quackenbush, J., Microarray data normalization and transformation. Nat Genet, 2002. 32 Suppl: p. 496-501. 259. Dunn, K.W., M.M. Kamocka, and J.H. McDonald, A practical guide to evaluating colocalization in biological microscopy. Am J Physiol Cell Physiol, 2011. 300(4): p. C723-42.

150

260. Villalta, J.I., et al., New algorithm to determine true colocalization in combination with image restoration and time-lapse confocal microscopy to MAP kinases in mitochondria. PLoS One, 2011. 6(4): p. e19031. 261. Pang, Z., N.E. Laplante, and R.J. Filkins, Dark pixel intensity determination and its applications in normalizing different exposure time and autofluorescence removal. J Microsc, 2012. 246(1): p. 1-10. 262. Yang, Y.H., et al., Normalization for cDNA microarray data: a robust composite method addressing single and multiple slide systematic variation. Nucleic Acids Res, 2002. 30(4): p. e15. 263. Couturier, L., K. Mazouni, and F. Schweisguth, Numb localizes at endosomes and controls the endosomal sorting of notch after asymmetric division in Drosophila. Curr Biol, 2013. 23(7): p. 588-93. 264. Ting, S.B., et al., Asymmetric segregation and self-renewal of hematopoietic stem and progenitor cells with endocytic Ap2a2. Blood, 2012. 119(11): p. 2510-22. 265. Huh, D. and J. Paulsson, Random partitioning of molecules at cell division. Proc Natl Acad Sci U S A, 2011. 108(36): p. 15004-9. 266. Metz, C.E. Basic principles of ROC analysis. in Seminars in nuclear medicine. 1978: Elsevier. 267. Morelli, L.G., et al., Computational approaches to developmental patterning. Science, 2012. 336(6078): p. 187-91. 268. Mogilner, A. and J. Zhu, Cell polarity: tension quenches the rear. Curr Biol, 2012. 22(2): p. R48-51. 269. Knoblich, J.A., Mechanisms of asymmetric stem cell division. Cell, 2008. 132(4): p. 583-97. 270. Hawkins, E.D., et al., Regulation of asymmetric cell division and polarity by Scribble is not required for humoral immunity. Nat Commun, 2013. 4: p. 1801. 271. Restifo, N.P. and L. Gattinoni, Lineage relationship of effector and memory T cells. Curr Opin Immunol, 2013. 272. Pham, K., F. Sacirbegovic, and S.M. Russell, Polarised cells, polarised views: Asymmetric cell division in hematopoietic cells. Name: Frontiers in Immunology, 2014. 5: p. 26. 273. Pham, K., (2013). Polarity during migration & asymmetric cell division in fate determination of thymocytes. PhD. Thesis: Swinburne University of Technology, Australia. 274. Zhong, W., et al., Asymmetric localization of a mammalian numb homolog during mouse cortical neurogenesis. Neuron, 1996. 17(1): p. 43-53. 275. Gonczy, P., Mechanisms of asymmetric cell division: flies and worms pave the way. Nat Rev Mol Cell Biol, 2008. 9(5): p. 355-66. 276. Aguado, R., et al., The endocytic adaptor Numb regulates thymus size by modulating pre-TCR signaling during asymmetric division. Blood, 2010. 116(10): p. 1705-14. 277. Lee, C.Y., et al., Drosophila Aurora-A kinase inhibits neuroblast self-renewal by regulating aPKC/Numb cortical polarity and spindle orientation. Genes Dev, 2006. 20(24): p. 3464-74. 278. Wirtz-Peitz, F., T. Nishimura, and J.A. Knoblich, Linking cell cycle to asymmetric division: Aurora-A phosphorylates the Par complex to regulate Numb localization. Cell, 2008. 135(1): p. 161-73. 279. Luty, W.H., et al., Antiparallel segregation of notch components in the immunological synapse directs reciprocal signaling in allogeneic Th:DC conjugates. J Immunol, 2007. 179(2): p. 819-29. 280. Anderson, A.C., et al., The Notch regulator Numb links the Notch and TCR signaling pathways. J Immunol, 2005. 174(2): p. 890-7. 281. Smith, C.A., et al., aPKC-mediated phosphorylation regulates asymmetric membrane localization of the cell fate determinant Numb. Embo J, 2007. 26(2): p. 468-80.

151

282. Bertrand, F., et al., Activation of the ancestral polarity regulator protein kinase C zeta at the immunological synapse drives polarization of Th cell secretory machinery toward APCs. J Immunol, 2011. 185(5): p. 2887-94. 283. Rabin, H., et al., Spontaneous release of a factor with properties of T cell growth factor from a continuous line of primate tumor T cells. J Immunol, 1981. 127(5): p. 1852-6. 284. Ciofani, M., et al., Obligatory role for cooperative signaling by pre-TCR and Notch during thymocyte differentiation. J Immunol, 2004. 172(9): p. 5230-9. 285. Ciofani, M. and J.C. Zuniga-Pflucker, Notch promotes survival of pre-T cells at the beta-selection checkpoint by regulating cellular metabolism. Nat Immunol, 2005. 6(9): p. 881-8. 286. Schmitt, T.M. and J.C. Zuniga-Pflucker, Induction of T cell development from hematopoietic progenitor cells by delta-like-1 in vitro. Immunity, 2002. 17(6): p. 749-56. 287. Schmitt, T.M. and J.C. Zuniga-Pflucker, T-cell development, doing it in a dish. Immunol Rev, 2006. 209: p. 95-102. 288. Schmitt, T.M., et al., Induction of T cell development and establishment of T cell competence from embryonic stem cells differentiated in vitro. Nat Immunol, 2004. 5(4): p. 410-7. 289. Holmes, R. and J.C. Zuniga-Pflucker, The OP9-DL1 system: generation of T- lymphocytes from embryonic or hematopoietic stem cells in vitro. CSH Protoc, 2009. 2009(2): p. pdb prot5156. 290. de Pooter, R. and J.C. Zuniga-Pflucker, T-cell potential and development in vitro: the OP9-DL1 approach. Curr Opin Immunol, 2007. 19(2): p. 163-8. 291. de Pooter, R.F., T.M. Schmitt, and J.C. Zuniga-Pflucker, In vitro generation of T lymphocytes from embryonic stem cells. Methods Mol Biol, 2006. 330: p. 113-21. 292. La Motte-Mohs, R.N., E. Herer, and J.C. Zuniga-Pflucker, Induction of T-cell development from human cord blood hematopoietic stem cells by Delta-like 1 in vitro. Blood, 2005. 105(4): p. 1431-9. 293. Ivashkevich, A.N., et al., gammaH2AX foci as a measure of DNA damage: a computational approach to automatic analysis. Mutat Res, 2011. 711(1-2): p. 49-60. 294. Wurflinger, T., et al., Automated segmentation and tracking for large-scale analysis of focal adhesion dynamics. J Microsc, 2011. 241(1): p. 37-53. 295. Kovesi, P., Phase Preserving Denoising of Images. The Australian Pattern Recognition Society Conference: DICTA'99, 1999. 1: p. 212-217. 296. Tweedie, L. and R. Spence, The Attribute Explorer: information synthesis via exploration. Interacting with computers, 1998. 11: p. 137-146. 297. Betschinger, J., K. Mechtler, and J.A. Knoblich, The Par complex directs asymmetric cell division by phosphorylating the cytoskeletal protein Lgl. Nature, 2003. 422(6929): p. 326-30. 298. Sumen, C., et al., Intravital microscopy: visualizing immunity in context. Immunity, 2004. 21(3): p. 315-329. 299. Bousso, P., T-cell activation by dendritic cells in the lymph node: lessons from the movies. Nat Rev Immunol, 2008. 8(9): p. 675-684. 300. Parish, C.R., Immunology and Cell Biology turns 90. Immunology and cell biology, 2013. 301. Carpenter, A.E., L. Kamentsky, and K.W. Eliceiri, A call for bioimaging software usability. Nat Methods, 2012. 9(7): p. 666-70. 302. Georgescu, W., J.P. Wikswo, and V. Quaranta, CellAnimation: an open source MATLAB framework for microscopy assays. Bioinformatics, 2012. 28(1): p. 138- 139. 303. Henriksson, J., et al., Endrov: an integrated platform for image analysis. Nat Methods, 2013. 10(6): p. 454-6.

152

304. Lu, Z., et al., Single cell deposition and patterning with a robotic system. PLoS One, 2010. 5(10): p. e13542. 305. Nilsson, J., et al., Review of cell and particle trapping in microfluidic systems. Anal Chim Acta, 2009. 649(2): p. 141-57. 306. Zhang, H. and K.K. Liu, Optical tweezers for single cells. J R Soc Interface, 2008. 5(24): p. 671-90.

153

Appendix I

TACTICS Toolbox v3.x

Interactive MATLAB Platform For Bioimaging informatics User Guide Appendix I

TACTICS v3.x by R.S. 2010-2014 Appendix I

∑ Content

Overview 3 TACTICS Version 3.x License Agreement & Contributors 5 Installation of TACTICS Toolbox 6 TACTICS workflow 8 Input data for TACTICS 9 Setting new experiment 11 Segmentation Module 15 Cell Tracking Module 17 Measurements Module 31 Robust Module 36 Analysis Module 37 Lineage Module 38 Polarization Module 47 ACD Module 48 About Figure in MATLAB 50 Troubleshooting 51 TACTICS for developers 52

∑ Attached video tutorials

Video 1 – Installation and generation of experiment file 14 Video 2 – Segmentation module 15 Video 3 – Cell Tracking module 17 Video 4 – Cell Tracking module-track correction 24 Video 5 – Measurements module 31 Video 6 – Robust Module 36 Video 7 – Analysis module- a 37 Video 8 – Analysis module- b 37 Video 9 – Polarization module 47

TACTICS v3.x by R.S. 2010-2014 -156- Appendix I

Overview

TACTICS is a versatile MATLAB toolbox for High Content Analysis (HCA) of microscopy data, in particular, TACTICS aim for customized time lapse microscopy analysis of non-adherent cells such as hematopoietic cells. TACTICS includes a dedicated user interface and customized algorithms to analyse multiple time lapse microscopy experiments with a mouse click. For instance, TACTICS can create scatter plots on a population basis, measure morphology features, measure ratios of fluorescence intensity both within and between cells, and analyse migration features of cells. TACTICS integrates a collection of valuable open code tools and original tools, into one package. All data can be easily exported in MATLAB format, .XLS , .XML or .CSV.

This user guide is supplied to help the unfamiliar user get started with TACTICS, and refers to several online videos for further support. Current applications are quantification and analysis of cell tracking, trafficking fluorescently- tagged protein within tracked immune cells, and multi-generational analysis of cell that extracts information from tracked parental lymphocytes and its sibling over several generations.

TACTICS v3.x by R.S. 2010-2014 -157- Appendix I

TACTICS V3.x : New features, better performances

• Drawing tool for image segmentation for multiple channels. •Speed up performances with improved tracking linker. • GUI to connect founder and daughter cells including time factorization and spacers. • New modules: Lineage and ACD Modules. • Tracking skipping and selective operator

High- High Content resolution Analysis (HCA)

Use Micro- Customized fabrication Migration Data mining analysis (i.e. grids)

Time-lapse Microscopy TACTICS v3.x

visualization Computational of sibling cells platform for (niche) algorithms

Applications

Cell polarity Asymmetric Cell Division Tracking life history of T cells

TACTICS v3.x by R.S. 2010-2014 -158- Appendix I

TACTICS Version 3.x License and copyrights

TACTICS V3.x is available, in executable form, free for academic research use and distributed under the BSD License (see the Open Source Initiative site- http://opensource.org/licenses/bsd- license.php) as the following: Copyright (c) 2010-2014, Raz Shimoni All rights reserved. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: •Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. •Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. ------NOTES- •TACTICS Toolbox has been written by Raz Shimoni1,2 as a sole and exclusive programmer under the direct supervision and financial support by Prof. Sarah Russell1,2,3 1Centre for Micro-Photonics, Faculty of Engineering and Industrial Sciences, Swinburne University of Technology, Melbourne, Australia, http://www.swinburne.edu.au/engineering/cmp 2Immune Signalling Laboratory, Peter MacCallum Cancer Centre, Melbourne, Australia. 3The Sir Peter MacCallum Department of Oncology, The University of Melbourne, Melbourne, Australia • Copyrights to this TACTICS user-guide is held by Raz Shimoni and appears as appendix to his PhD thesis. •TACTICS is supported through Matlab file exchange forums. http://www.mathworks.com.au/matlabcentral/fileexchange/39920-tactics-toolbox And through TACTICS home page: http://tactics-toolbox.com/ •If other programmer modify or add new capabilities to TACTICS, only the new prospective code should credited to the corresponding programmer. To sharp the points above, if a developer made X contribution to TACTICS (where X stands for a specific function, modification, new code, new module, new tool, etc), a statement such as the following shall be provided: "Using TACTICS as a platform, I programmed/wrote/coded/etc X in order to…..." Thus, the credit shall be towards the X contribution but not towards TACTICS. • Support and feedbacks Updates are expected regularly. Suggestions to improve the automation capability and the quality of the analysis are welcomed. New versions are expected on yearly base. •Special thanks to the contributors of open-code functions used in TACTICS and biologists who provided data input. To see a list of all current contributors Go to: >Acknowledgments at the Main Manu. • Since TACTICS uses third party open-code library (i.e. tracking algorithms, filters, GUI tools, etc), MATLAB developers who wish to modify TACTICS to create a new program should be aware of and acknowledge the relevant contribution. • TACTICS use requires a MATLAB commercial license. •TACTICS users are not expected to share co-authorship for any use of TACTICS. However, when TACTICS is used for publications, please refer to TACTICS publications in Bioinformatics Journal. New components customized by the requirements can be easily set up (not necessarily in exchange for co-authorship).

TACTICS v3.x by R.S. 2010-2014 -159- Appendix I

Installation of TACTICS Toolbox- 1

Requirements and compatibility For MATLAB 2012b(The MathWorks, Inc.) and above, TACTICS can be installed as a MATLAB App (instruction A below) otherwise, at least MATLAB R2008a version 7.6 is required (instruction B below). The system requirements depend upon the data size and exact task, and the minimal requirements are: Windows® 32-bit/ or 64-bit operating system Intel® Pentium® 4 processor, or equivalent 1 GB of RAM

A. To Install TACTICS toolbox as MATLAB App

1.a) Go to MATLAB APPS

1.a) Select Install APPS

3) Select Install

2) Select TACTICSv3.x app installation file

4) Installation was successful

TACTICS v3.x by R.S. 2010-2014 -160- Appendix I

Installation of TACTICS Toolbox- 2

B. To Install TACTICS toolbox (not as MATLAB App) 1. Download the TAC.rar from MATLAB File Exchange. 2. Decompress the TAC.rar file and create new directory (does not have to be placed on specific location). 3. Run MATLAB. 4. First time running - calibration may be required for each GUI. Enter in MATLAB command window the command ‘GUIDE’ and press Enter. Select the necessary GUIs (.fig extension). Adjust the GUI window to the full size of the screen and launch the GUI (green play button). 5. To run TACTICS change the current MATLAB directory to where TACTICS is located, Input ‘run’ in the command window of MATLAB, and press Enter to execute. TACTICS main interface will appear.

MATLAB command window

Screen capture of the TACTICS starter menu interface

Selection buttons to open the different modules from TACTICS starter menu

TACTICS v3.x by R.S. 2010-2014 -161- Appendix I

TACTICS workflow

• Philosophy

TACTICS is structured from several main collaborative and 1 modifiable modules. This simplifies the analysis scheme, allowing unfamiliar users to study and follow the program. The workflow follows the following paradigm:

1.Creation of new experimental data files (for a single movie or experiment file data set that is consisted of multiple positions). The experimental files can be imported and exported between the Segmentation [2], Cell Tracking [3], and Measurements modules [4] for a single experiment file or in batch mode using the Robust Module. 2 2.Filtering and segmentation with TACTICS Segmentation Module.

3.Cell labeling and tracking with TACTICS Cell-Tracking Module. Version 3.x containing several imperative improvements that increase the accuracy and efficiency of long tracking. 3 Multi position processing position Multi 4.Multichannel quantification with TACTICS Measurements Module. The cell libraries generated by the Measurement Modules (.fig files) can be imported to the Analysis Module [5].

5.The Analysis Module can export structured data for the more dedicated modules [6-8].

6.The Polarization Module allows gating and visualization of 4 many aspects of cell polarization (i.e. different normalization, alignment, directionality, selection of data).

7. The Lineage Module (new for ver. 3.x) allows combining image cytometry analysis with pedigree analysis for many Cell generations of tracked cells. libraries (.fig) 8.The ACD Module (new for ver. 3.x) allows incorporation of our major/minor approach for the analysis of Asymmetric Cell Division (ACD).

Exploration tools for HCA

Polarization Module Lineage Module ACD Module

6 7 8 5 Analysis Module

TACTICS v3.x by R.S. 2010-2014 -162- Appendix I

Input data for TACTICS - 1

TACTICS experiment file

TACTICS was designed to deal with a large number of frames. For example, a data set consisted of 2000 time points, 3 channels (therefore 3 images for each time point), 1024 x 1024 pixels (~1Mb for each tif file) takes approximately ~6Gb of space. To prevent memory issues when dealing with large data sets, TACTICS works in a virtual stack mode as default, while selected images can be uploaded to the computer RAM. The location of the directory that contains single images is saved in experiment data file, and each frame is loaded in its turn. The experiment data file contains information about each processed channel in the following order: 1.File names 2.Pathnames 3.Channel type 4.Cell ID and features of tracked cells 5.Cell trajectories 6.Information about the format of the image files 7.Flags (i.e. what channels were labeled or tracked) 8.Information about applied image processing procedures 9.Number of bits 10.Experiment file name

Folders for multi-positional analysis

TACTICS analysis can be done to a single experiment or multi-positional batch processing, where images from each position are saved in different subfolders. For example, see below a screenshot of the folder that contains several Pos## located in c:\ap2a2 directory. ## is the index number of each position.

• STRUCTURE – To start a new experiment in multi- positional batch processing the subfolders name should be Pos##, whereas ## is the position number index.

TACTICS v3.x by R.S. 2010-2014 -163- Appendix I

Input data for TACTICS - 2

• File format – The files accessible by TACTICS are tif (.tiff and .TIF are also common tags for TIFF files), high-quality graphics files that contain the raw data in a lossless compression format. In each subfolder the microscopy images must be named in the format: “ &&_Pos##_t@@_ch**.tif”, where && is the experiment name, ## position index, @@ time index, and ** is the channel index. An example (“100310_T24-02_GFP-Ch-AP2A2 DN3 on cless DL1_01_Pos01_t002_ch01”) indicates a time series with three channels arranged as shown below:

•If the data contains z-sections, each subfolder is expected to contain the microscopy images in the format: “ &&_Pos##_t@@_z$$_ch**.tif”, whereas $$ is the section index.

File exporting/renaming – since different microscopes export the images in different file formats, conversion might be required. Several conversion tools are available: 1.To convert .lif to .tif : use ImageJ macro tool or MetaMorph journal that convert .lif to .tif (written by Cameron Nowell and available upon request). 2.If the files are already in .tif format, but renaming is required the user can use renaming tool supplied with TACTICS (located in TAC\supporting tools folder): Olympus2TACTICS and MM2TACTICS. 3.If the images are in z section format, use the dedicated file projector Z_projector (located in TAC\supporting tools folder) to project the section. This will be shown in the next pages.

TACTICS v3.x by R.S. 2010-2014 -164- Appendix I

Setting a new experiment - 1

Batch projection tool (only required for experiments with Z-sections)

I M P O R T A N T: backup the data (the raw .tif images) before starting! From TACTICS Main Menu: Go to >’Tools’>>’Z_projector’ Z_projector– to create the mean projections from multiple z sections and to transfer each .tif to its z location (see next page for screenshot).

1.Select the main folder where all the pos folders are located -For fast search within specific folder: Go to >’Load pos folders’>> ‘choose fast search’ -For slower search within subfolders: Go to >’Load pos folders’>> ‘sensitive search’ Wait a few seconds. All the position folders in the main folder will appear in the listbox. Since the search is not selective, the user can remove unnecessary folders from the listbox by selecting the unwanted folder from the list and clicking on ‘Remove selected folder from list’ button.

2. Define user options: a.Remove last frames (because usually the last frame/s can be corrupted when user stops the image acquisition during the movies). b. Choose position automatically- when user wants to convert from selected index. c. Z selection and channel selections- to define the sections and channels to convert according the movie. d. Backbone name- the string that appear in the .tif filenames. e. Build z folders- create a new directory that contain the original .tifs converted into TACTICS format, organized by z sections. This option needs be on. f. Project by mean is the only projection that is working at the moment. Other types of projections can be easily added by demand.

3. Click the green ‘start’ button. A timebar will appear giving the estimate time to complete the conversion. The measured time in Acer TravelMate 6593 2.6GHz VPro with 4 GB RAM is ~12 min for a position with 700 time points and 5 sections. I M P O R T A N T: do not use MATLAB while the projection is running, and generally it is better to completely leave the computer alone.

4. To separate between the original un-projected and the newly projected images, the un- projected images have to be deleted. Therefore, once the projection process is completed press the red button (‘Delete section files after constructing their projections’). For caution, after pushing the red button a message box will ask the user to continue or not, followed by whether to abort or not. Click ‘Yes’ in the first message box and ‘No’ in the second message box to remove all the original z sectional .tif images from each position folder (but not the newly-created z subfolder).

I M P O R T A N T: do not press the red button before the green otherwise you delete the original files before creating the new z subfolder! 5. Go to >exit in main panel to go back to main TACTICS menu.

TACTICS v3.x by R.S. 2010-2014 -165- Appendix I

Starting a new experiment - 2

Close-up of the z-projection tool interface

1.Load positions

2. Define user options

3. Start projections

4. Discard un-projected images

TACTICS v3.x by R.S. 2010-2014 -166- Appendix I

Starting a new experiment - 3

As explained previously, the first step in TACTICS analysis is to allocate the microscopy images by generating an experimental file that contains information about the microscopy images and the tracked cells.

Instructions: 1. To run the Experiment generator from TACTICS starter : Go to>File>> Experiment_generator.

2. Load files. a. For single experiment mode: Load a sequential series of 2-D images of the sample and to create a new data file by: Go to: >Create New>> Batch file>>>Load .tif>>>> choose .tif or folder containing .tif Wait a few seconds. The positions/tifs will be loaded into the listbox. Click delete button to remove unnecessary positions or .tifs from the listbox. I M P O R T A N T: if one channel is deleted from a specific time point, all the .tifs from this time point must be deleted. An error message will appear if missing files are found. b. For batch mode: -For fast search within specific folder: Go to:> Create New>> Batch file>>>Load pos folders>>>> choose fast search -For slower search within subfolders: Go to:> Create New>> Batch file>>>Load pos folders>>>> choose sensitive search

3. Define the channels according your experiment.

4. Choose a tag name for your experiment that will appear in each data experiment file. Modifying experiment name will prevent overwriting the pos data file when setting new data file for the same pos folder (this allows multiple experiments data for the same dataset).

5. Optional- Define section area to be process. (otherwise all the image will be processed as default).

6. Optional- Define path of the associated .tifs (both processed and unprocessed). However, it is recommended to leaving it in as the default path. The path can be also defined again once the .dat file is created.

7. Launch. This can take several minutes. A new experiment data will be created (for single or each position in batch mode) in the format ‘TACTICS_EXP_##.dat’, whereas ## is the tag name. In addition, new subfolders for each corresponding channel for the filtered and segmented images will be created.

TACTICS v3.x by R.S. 2010-2014 -167- Appendix I

Starting a new experiment - 4

Create new experiment file Video tutorial 1 1.Load positions/tif`s 5.Save and Generate experiment files 4. Define section

Screen capture of the experiment generator interface 2. Define channels 3. Define tag name

6. New folders will appear after processing is completed.

TACTICS v3.x by R.S. 2010-2014 -168- Appendix I

Video tutorial Segmentation Module -1 2 The Segmentation Module is used for image processing-based filtering functions and segmentation. The GUI is comprised of three listboxes, which contain the raw, filtered, and segmented images. From the main menu, go to >Segmentation Module To load and segment a movie: 1.Load an experiment data file to TACTICS by clicking on the open file icon . Navigate through to your experimental folder. This folder will contain a series of .tif files and other files and folders. Within the folder choose the experiment .dat file.

2.The left listbox will show all the raw .tif file images that are associated with the experimental data file. The user can navigate through the images using functional options that define the channel and section, and changing the current frame using an interactive scrollbar. Both filtering and segmentation can be performed for each z section or for the mean projected image.

3.The images must be filtered to remove noise prior to the cell segmentation. The left scrollbar panel allows the user to set up a combination of MATLAB built-in functions or customized functions using the interactive scrollbar, which maximizes the efficiency of manual customization of specific settings for each experiment. This step can be applied for selected frames without saving, with saving, or in batch mode for selected frames. An image of the filtered file will appear.

4.All saved frames are added into the middle listbox, dedicated to the filtered images. The next step is to segment the cells to convert to binary images. The default for segmentation is threshold-based fluorescence intensity. The user can define the threshold value manually or to apply automated global image threshold using Otsu's method (Otsu, N., "A Threshold Selection Method from Gray-Level Histograms," IEEE Transactions on Systems, Man, and Cybernetics, Vol. 9, No. 1, 1979, pp. 62-66.). Briefly, Otsu`s method gives the threshold value that gives minimum interclass variance of the cells and the background. Operations on the binary image such as dilation, erosion, and removal of segments by size can be applied to remove false objects. MATLAB users can easily adopt other methods. Screen capture of the Segmentation Module interface

Raw Image Filtered Image Segmentation Image

Navigation tools

Segmentation panel ‘ Filtering panel

TACTICS v3.x by R.S. 2010-2014 -169- Appendix I

Segmentation Module - 2

Channel selection (also available in the Cell Tracking Module) You can view the cells in different channels (representing different florescent wavelengths or transmitted light images) and also choose whether to view the data in raw, filtered, segmented, F&S (filtered AND segmented for background elimination), and R&S (raw images AND segmented for background elimination).

Instructions for changing channels:

1.To choose how you would like to view the data in both the Primary screen and Secondary Screen, you will use the pull down menu.

2.To change the channels, each screen has its own (linked) pull down menu.

•Virtual channels The user can create a virtual channel (for instance, by selecting a new destination channel. The current number of supported channels is 14, but can be extended. •Save/Load optimal settings The settings for each frame will be saved to the RAM memory of the computer. After processing the raw .tif, the logged information about the image processing procedure must be saved as a new/overwritten experiment file. Otherwise, this information will be lost and the operation will have to be repeated again. In addition, once a suitable setting is defined, the user can load and save optimal settings in log file that contain information about the settings. Go to >Optimal setting>>Save optimal setting .The log file format is TACTICS_SEG_###.dat, where ### is any name chosen by the user. To retrieve settings for specific processed frame, select a frame using the raw listbox (the left hand listbox), then Go to >Optimal setting>>Setting for selected file

•Supported in the icon bar-

Load experimental data file. Save experimental data file. Create new experimental data file (run the Experiment Generator GUI). Zoom in. Zoom out. Pan.

Open new MATLAB figure showing the data in the left axes. The data is stored in the ‘userdata’ property of the figure.

TACTICS v3.x by R.S. 2010-2014 -170- Appendix I

Video Cell Tracking Module – 1 (user interface) tutorial 3 Once the images were successfully segmented, the next step in the processing scheme, is the cell tracking and manual correction interface: From the main menu, go to >Cell Tracking Module The Cell Tracking Module is used to correct segmentation, label cells and tracking. The GUI has several selection methods for manual image segmentation that allow the user to inspect and correct cell segmentation manually, and allows for multiple objects tracking correction.

Screen capture of the Cell Tracking Module interface Secondary Screen

Editing tracks panel Trajectory panel Primary Screen Segmentation panel

Navigation panel

Functions supported in the icon bar-

Load experimental data file Mark cell death Save experimental data file Mark cell Drag zoom rectangle division Capture ROI to secondary screen Zoom in Zoom 100% Zoom out Pan Error Zoom (temporary solution for Zoom issues)

Open new MATLAB figure 1. Run the Linker 2. Split cells after divisions 3. Save manual correction of cell associations 4. Show lineage tree 5. Mark cell division 6. Mark cell death

TACTICS v3.x by R.S. 2010-2014 -171- Appendix I

Cell Tracking Module – 1 (user interface)

Screen capture of the Cell Tracking Module interface

Channel and display mode of the secondary screen

Change the channel Read and upload and the display images within mode (raw, filtered selected range to etc.) of the Primary RAM (dependent on screen using this the computer pull down menu memory)

Keyboard shortcuts: Uparrow - Next frame D - Dead cell mode Downarrow - Previous frame E - Next frame Space - label frame again R - Remove segment mode Enter - associate frame overpassing selective operator U - uncheck/check F - Format painting P - Paint Tool mode S - Segmentation mode M - Mark division mode E - Next frame T - Trajectory mode A - Association mode

TACTICS v3.x by R.S. 2010-2014 -172- Appendix I

Cell Tracking Module – 2 (cell labeling and tracking)

Instructions for cell labeling and tracking: 1.Load an experiment data file to TACTICS . 2.To label the cells- Go to >’Label cells’>> ‘Label cells’ . A labelling function will label each segment from n to m (n and m are user input). Reference channel can be used to help to interpret normalized pixel intensities in the quantified channel and allow to track two types of cells at the same time. After running the Label Cells function, the cells will be detected, identified and labeled including feature extraction. A bounding box will mark each segment, and all labeled frames will be marked in red tab under the scrollbar:

3. To connect between the cells over time- Go to >’Label cells’>> ‘Label association‘. An input box will ask for the range of frames to be associated and maximum distance that the cells can travel (a new cell will be defined if the distance to a closer point in time (t-1) is above this criteria. The input units are currently in pixels. Once finished, all associated frames will be marked in green tab under the scrollbar (note that the first frame cannot be associated):

4.To track the cells, click on the ‘tracking’ icon or Go to >Track cells>> Hungarian algorithm>>>Tracking linker. The associated points will be connected to create the cell`s tracks. The user must input a cut off value that sets the possible minimum track length (a value of 1 is frequently acceptable). ------As an alternative to label association followed by Hungarian algorithm, the user can utilize the Crocker algorithm for tracking multiple objects Go to >Track cells>> Crocker algorithm. The advantage is the ability to recover for missing points. However, the algorithm is slower and fails when the number of cells or time points is large (dozens of cells for 1000 frames will fail the algorithm). ------•Once cells are tracked, trajectories of cell over time can be visualized. The location of the cells in different time points is marked with a colored point, and the size of the point is proportional to the timeframe of the track. •To mark dividing cells, go to cell division mode and click on the cell at the point of division (one time point before the appearance of two separated cells) with the middle mouse button. X sign will be added/removed correspondingly. •To mark dying cell, go to cell death mode and click on the cell at the last point of appearance with the middle mouse button. + sign will be added/removed correspondingly.

IMPORTANT. Save the experiment file (overwrite or create new file) before exiting TACTICS (or before loading new experiment file). Otherwise, any operation will be lost.

TACTICS v3.x by R.S. 2010-2014 -173- Appendix I

Cell Tracking Module – 3 (cell labeling corrections)

A. Improving segmentation of a single frame using the secondary screen view 1. Click on the ‘Capture ROI to secondary screen’ icon . Once you have clicked on the icon a window will come up with an image of the screen you have been working on. Hold down mouse (left click) anywhere on the new screen. This will create a second figure from the right side of the screen containing a magnified image of the ROI within the images that is chosen for zoom. 2. While still holding down left click use the arrow keys to form a box around the cells that you would like to work with in the secondary screen (Make sure the aspect ratio is 1:1 or you will not be able to see the cells clearly or correct them with good accuracy). Once you are happy with the size and placement of your box, double left click to confirm. The selected region will be magnified to the secondary screen view and the borders of this region will be appeared in the primary screen view. 3. When using the mouse-wheel up and down, the ROI will be shown in the primary screen view. User can use the Drag zoom rectangle icon to move the ROI keeping the current dimensions. 4. Go to Segmentation mode, and select ‘Segment current frame and save to HD’ option in the popup menu. 5. To load the segmentation settings go to the File menu > import > (import segmentation settings OR segmentation settings for current frame). The user can then adjust those settings to improve segmentation of the current frame by altering values on the bottom right frame.. 6. The user can remove individual pixels by clicking with middle mouse button (or mouse-wheel) while in the ‘segmentation mode’.

Screen capture of the Cell Tracking Module interface Secondary Screen showing the zoomed-in ROI

Box containing the ROI

Primary screen view Secondary screen view

TACTICS v3.x by R.S. 2010-2014 -174- Appendix I

Cell Tracking Module – 3 (cell labeling corrections)

Separating merged cells. TACTICS provides multiple options for manually separating merged cells, including: 1. Splitting a single object using watershed or intensity shell algorithms. 2. Splitting a single object by drawing a line through it or deleting individual pixels. 3. Improving segmentation of a single frame (as per previous page).

1. Splitting objects using watershed or intensity shell algorithms. If a labeled object is clearly two or more adjoined cells, move the cursor over the box surrounding the object, and left mouse click (watershed) or right mouse click (intensity shell) in the primary screen view. If the algorithm succeeds, the bounding box should split into two boxes. If the algorithm fails, the alternative algorithm can be attempted. After separating all the adjoined cells in a frame scroll forward or back through the movie to save the changes.

2. Splitting objects by drawing a line or deleting pixels. a. To separate between cells by drawing a line of deleting pixels use the Secondary Screen (See the previous page for instructions on how to access the secondary screen using the capture ROI icon ). b. Draw a line where cells need to be divided. To do this, left click at the first point of where your line should be and a yellow dot will appear. Left click again at the same point and drag the blue line to divide your cells. An intensity line will go between the cells to give two labelled cells. c. Double left click on the blue line to accept changes. d. Click middle button (wheel) or CTRL+A to save changes. e. The properties of each segment that are saved in the .dat experiment file (described in Appendix 1), and bounding box will appear for each segment (in the Primary Screen). In the secondary screen, outlining of the cells by different colors indicates successful segmentation.

First point identified Final product (two defined cells) Blue line drawn Initial problem cell f. Run tracking algorithm again . This is required to incorporate the newly segmented cells, and can be performed after resegmenting one or several frames.

I M P O R T A N T: Once this data file is created and loaded into Cell Tracking Module, any modification in the data file will be logged in the loaded data file. Therefore it is necessary to save the data file once processing is applied.

TACTICS v3.x by R.S. 2010-2014 -175- Appendix I

Cell Tracking Module – 3 (cell labeling corrections)

3. Splitting objects by altering segmentation parameters. Resegmenting the entire frame can also help with splitting two cells, although care must be taken that further segmentation problems do not occur in other areas of the frame. Below is an example where two joined cells were separated by increasing the threshold value as per instructions in A (p 18).

Changed threshold value

Optional. Changing the LUT is a standard procedure that helps to distinguish between pixel intensities. An example shown below shows that pixel intensities in the cell perimeter have low value, causing over-segmentation by Otsu’s method.

TACTICS v3.x by R.S. 2010-2014 -176- Appendix I

Cell Tracking Module – 3 (cell labeling corrections)

Removal of segmented pixels that are falsely identified as cells: Sometimes a new object appears during segmentation correction that is incorrectly labeled as a cell. An example of this is shown below:

Cell Tracking Module incorrectly recognizes this as a cell

Instructions for removal:

1. Zoom-in on the problem area in the secondary screen until each pixel is clearly visible. 2. Right click on one of the pixels to remove the colored dots indicating the segment edge. 3. You can now remove each unwanted pixel by right clicking on it. To replace a pixel that you have deleted right click in the same place. Similarly, right clicking on a black pixel converts it from 0 to1, and so can be used to link separated cells.

Shot of secondary After pixels had been screen with pixels After one right click removed. clearly visible

TACTICS v3.x by R.S. 2010-2014 -177- Appendix I

Cell Tracking Module – 4 (cell tracking corrections)

The tracking accuracy depends on the quality of the raw data, the imaging system, the biological system, and the computational approach to analyze the data. Common problems with tracking are when a cell suddenly moves a large distance. Tracks can be changed manually using several approaches, that either involve moving track vectors on the primary screen, or use a pull down menu in which cells can be connected by numbers.

To correct cell association tracks by shifting connecting vectors: Video tutorial 1.Choose (n-1) in the secondary axes. 4 2.Click association button. 3.Drag the lines between the middle f the cell bounding-box (n-1) to its origin in n. 4.Click on the green icon in the icon panel to accept changes. Cell without match between n and (n-1) marked by blue centroid. Cells with match are marked in red and id number. Watch the video tutorial for more information.

Instructions for manual corrections for tracking:

1.Go to edit tracks mode. 2.Point with the mouse curser on required cell and click on the middle button. A selection window will give several different options.

•IMPORTANT: the manual corrections are applied on the matrix of trajectories. Therefore, when applying the linker again, the manual corrections are deleted. For this reason, it is always better to apply manual corrections for cell association.

TACTICS v3.x by R.S. 2010-2014 -178- Appendix I

Cell Tracking Module –5 (marking cell division and death)

•Marking cell division

•Go to the (Mark cell division) mode •Go to the last frame before division, where only one cells in visible •In the primary screen, wheel click on the dividing cell •Click on the (tracking splitter) icon

•Parental cells will have name ##-**, whereas ## is the name chosen by the user and ** is a number by order of appearance. •Daughter cells of parental cells will have name ##-**.1 and ##-**.2, whereas ##-** is the name of the parental, and 1 and 2 symbolise daughter 1 and 2. •For each additional cell division a dot and number 1 or 2 will be added. For instance: ##-**.1.2.1.2.2 Screen capture of the Cell Tracking Module interface

4-click tracking splitter

1- Cell division mode

Frame n Frame n+1 2- cells are about to divide

5- (next frame) two cells identified as two related 3- wheel click on Frame n Frame n+1 daughter cells and the cell, information is saved in Adding red x the experimental file annotate the cell as dividing cell

Frame n : Final frame before division

TACTICS v3.x by R.S. 2010-2014 -179- Appendix I

Cell Tracking Module – 6 (drawing tool)

The painting tool allows manually to draw, fill, and remove pixels on any selected channels. This can be very useful to correct segmentation when automated segmentation fails.

To apply the drawing tool: 1.Go to drawing tool mode. 2. Select show segmentation to “on”. 3.Select channel to apply segmentation, channel for visualization, and channel for tracking. Zooming-in is recommended, but not a recruitment. The blue net will appear for each existing pixel in the binary matrix. 4. Select the size of painting point. 5. Press mouse left click. Hold it and move the mouse curser to fill new pixels. 6. To remove pixels- Press mouse right click. Hold it and move the mouse curser. 7. To accept: click mouse wheel button. This will save the modified image as segmented one (overwrite) and will label again (relabeling should be saved to experimental file otherwise this will be lost). 8. To cancel (before acceptance): just use mouse wheel to refresh frame (go to the next or previous frames) 9. To cancel (after acceptance): currently TACTICS doesn't have this feature (is that useful?). Therefore need to segment the frame again.

4- select point size 1- drawing tool mode

3- select channels

2- click show segment on

TACTICS v3.x by R.S. 2010-2014 -180- Appendix I

Cell Tracking Module – 7 (Parameter selection)

The parameter selection window ddisplay parameters such as intensity, and velocity for a particular selected cell, which allow to improve the cell tracking.

To activate the parameter selection window : 1. Select cell. 2. Choose- ‘plot data for selected cell’ option. 3. New Manu will popup. Choose parameter to be plotted. (more parameters can be easily added). 4. When using the mouse scroll wheel, the plot will be shown in the secondary figure view for the relevant time points.

4- parameter is plotted 3- select parameters in popup menu

2- select mode

1- select cell

Display number of objects in the current frame

TACTICS v3.x by R.S. 2010-2014 -181- Appendix I

Cell Tracking Module – 8 (corrections of lineage trees)

Complicated switch in tracking can accrues when cells drift and tracking is incorrect for the following tracks. This can strongly influence the lineage tree and its analysis.

To correct tracks for lineage trees: 1. From main Go to >’Tools’>> ‘Generate Lineage tree’. A popup window shows the lineage for any selected cell. In this particular example it can be seen that a problem accrues in the tracking. Click on the lineage window interactively goes to the frame of selection in the Cell Tracking Module. 2. In some cases other supporting tools can help to allocate events suspected as wrong. For instance the cell tracks window. From main Go to >’Tools’>> ‘open cell tracks window’. Other tools can be useful Go to >’ Data inspection’ , such as show number of objects over time, show trajectories matrix and search for undefined association. 3. Adjust the current settings showing for frame n and (n-1). 4. To visualize the association between cells at particular frame click on Centy button. In this particular example, because (n-x) is set to -1, the cells in DIC in frame 382 but the tracks and bounding boxes are actually shown for frame 381. This provides the locations of the cells in the previous frame. The Centy button shows that a new cell was defined (blue), and that the second daughter cell was wrongly labelled as cell 1.1.2.2.

2- (optional) open cell tracks window 1- Generate Lineage tree or use other inspection tools

3.a.- input frame to show to be -1

3.b.- input channels to 4- Click on centy visualize frame n-1 3.c.- input channels to track and visualize frame n

TACTICS v3.x by R.S. 2010-2014 -182- Appendix I

Cell Tracking Module – 8 (corrections of lineage trees)

cell association tracks by shifting connecting vectors

5. The user can move the pink lines, so the pink star * touch the destination location in frame 383. In this case a line was stretched from cell number 7 to 1.1.2.2, while the line that start. In this example, from the cell between 1.1.2.2 and 1.1.2 was pointed back on itself. This mean that this cell doesn't Have a much and will therefore recognized as a new cell. To accept click the cell association correction button.

6. After the correction, run the linker again. The lineage tree is now corrected. If the user regrets or want to repeat the process, labelling the frame again goes to initial association.

5- drag lines to link between cells

6- lineage tree is corrected

Two-dimensional graphic visualization of lineage tree. Bifurcations in the tree represent cell divisions. X- axis represents the cell index. Y-axis represents the frame (or time) that each cell appear.

TACTICS v3.x by R.S. 2010-2014 -183- Appendix I

Cell Tracking Module – 9 (selective operator)

Selective operator is method that allows the user to select what frames will be associate or skipped. The user can select by frequency to be skipped (for instance every 3rd frame) or/and certain frames in the movie. The associated frames appear in green bars under the movie scrollbar. Only the frames between following green bar coded frames are associated.

Selective operator Show frames Show tracks Show tracks Modes annotations

Non-Selective All frames Only in selected frames Only in selected frames

Selective All frames Only in selected frames Only in selected frames

Only Selective Skip to selected frames Skip to selected frames Skip to selected frames

Partly Selective All frames All frames Only in selected frames

Trajectories matrix shows association between every 10 frames

Green bars shows the associated frames

Number of cells in frame in selective mode

Selective operator

TACTICS v3.x by R.S. 2010-2014 -184- Appendix I

Measurement Module – 1 (user interface) Video tutorial 5 The Measurement Module is used to extract different aspects of the cell morphology, cell migration, protein distribution of the POI (Protein Of Interest) and interactions between different types of cells as long as they are fluorescently labelled. The output of the Measurements Module is data libraries, including an image gallery, histograms and scatter plots of morphology or migration features, projections and more. The GUI is divided into several regions (with drop down menus). The upper selection tool enables the user to create a task list and to run it in batch mode. This provides the means to call for data collection on cell or population base, and for parental and daughter cells from multi spectral data. For each task a new active visualization figure will be displayed, plotting the required information. The user can choose how to visualize, track, and quantify the data. For instance, a reference channel can help to interpret normalized pixel intensities in the quantified channel and allow to track two types of cells at the same time. The right bottom panel is the stack control for changing the current ‘tif’ file from stack, and is used to define the overlay of images from different z sections and channels. Visualization includes the raw, filtered, and segmented images for each z section and colocalization of up to 3 channels plus DIC/bright field channel. In addition, tactical plot showing cell index, trajectories, and proximity vectors between cells in different channels can be shown

Close-up of Measurements Module interface

Select channel (Merging up to 3 channels and DIC)

Plot list

•The Main Menu includes additional tools. For example, a tool to convert a group of single-image .tif files into one multiple images .tif file (a stack file). •Since each image is loaded to the computer’s RAM only by demand, there is a lag of a fraction of second for MATLAB to read and to image the .tif file. The user can load the movie to the RAM for visualization purposes if the memory of the computer is sufficient (depended on the number of frames and computer RAM).

TACTICS v3.x by R.S. 2010-2014 -185- Appendix I

Measurement Module – 2 (control panel)

3. Choose parental cell and its daughter for 2. Define analysis options Or- 4. Choose channel for tracking, quantification and alignment (can be 5. Define data the same channel) type for analysis 3. Select cell for analysis

6. Add to list 7. Plot the list

8. (optional) Save generated data to defined 1. Choose library channel, and automatically section

Close-up of Measurements Module user selections panel Define options:

° Keep/ignore missing points- TACTICS gives the option to set an active and passive regions. While the cells are tracked in both active and passive regions, they will be quantified only from the active region. The user can define collect all data points including the “empty points” which are Not a Number (NaN) values, or to extract data points only from the active regions ° Parental off/on- remove or keep the cell before division from the plot. ° Use modified segmentation: ° Consider get 8 option- this function is to quantify cell properties during the cell division. ° Merge channels- plot montage or projections based on multi-channels.

TACTICS v3.x by R.S. 2010-2014 -186- Appendix I

Measurement Module – 3 (parameters)

Instructions to create functional analysis list: 1. Choose for quantification of single cell, on population base, on for a dividing cell and its two daughter cells. 2. A list with 40 different functions will be appear in the middle selection bar. Choose one feature. 3. If the feature is type1 (I.e. 2D-projection), optional selection will appear for alignment. If the feature is type2 (I.e. Area), optional selection will appear for plot versus any selected parameter (current options are shown below, and can be easily include with another analysis function.

1. Analysis for selected cell, on population base, or for dividing cell

2. Select function for analysis

3. Use alignment option (type 1) or plot versus another function (type 2)

LIST OF MEASUREMENTS

TACTICS v3.x by R.S. 2010-2014 -187- Appendix I

Measurement Module – 4 (multi-channel analysis)

Extraction of data on cell or population base from multi spectral data. Reference channel can help to interpret normalized pixel intensities in the quantified channel and allow to track two types of cells at the same time.

Channel 1: Cherry Control Channel 2: GFP-Numb

Multi-Channel stack alignment by double-transfected cells over space and time

Brightest pixel Time

θ Centroid

“speckles” of POI

Cell perimeter segmented by fluorescence intensity of control protein One channel for tracking, Second channel for alignment, Third channel for quantification

TACTICS v3.x by R.S. 2010-2014 -188- Appendix I

Measurement Module – 5 (quantification within ROI) Quantification within changeable ROI- While cells are partly leaving/entering the field-of-view between imaging interval they can be still tracked, however will resulted misleading quantification of only part of the cell. The Measurements Module gives the option to define a border whereas quantification will be considered only when the cells inside this borders, and data points where cell outside this limit are skipped. The Segmentation Module can generate up to 99 virtual channels based on user input (i.e. borders, rectangle and disk shapes) or based on another channel. The creation of virtual channel around segmented moving objects (i.e. beads, cells, microfabricated structure) allows quantification in/out borders of segments. For example, it can be utilized to study of statistical correlations between cells activity in relation to polarity cues from second type of cells. Alternatively, the Tracking Module can utilized to define the ROI. Empty points returned as NaN`s, or compressed (by removal all NaN`s).

Tracked only Tracked & quantified

Control box Select cell for analysis

choose if to show points where cell is out/in the ROI

Use NETO option for quantification within changeable ROI change

Merge channel for montage and projections Use DIC for montage Remove parental cell from the analysis

TACTICS v3.x by R.S. 2010-2014 -189- Appendix I

Video Robust Module tutorial 6

Robust Module is used to achieve automation of multi-positions analysis, whereas each module can be called on repeatedly. Instructions: 1. Load multiple experiment data files by selecting the folder that contains the position folders. 2. Define number of steps to run. 3. Choose the module. 4. The GUI is divided into several modular regions that will appear according the requested modules, whereas each module requires different settings. Processing can apply based on z section or mean projection. • To set it to automatically to send notes about status within the loop (via mail) please contact us for further instructions.

Pos01 Pos02 Pos03 Pos04 Pos05 Pos06 Pos07 Pos08 Pos09 Posx1 Posx2 Posx3 Posx4 Posx5 Posx6 Posx7 Posx8 Posx9

3. Choose module

1. Load positions. 4. Define settings

5. Run

2. Define number of steps to run

Screen capture of TACTICS Robust Module interface

TACTICS v3.x by R.S. 2010-2014 -190- Appendix I

Video tutorials Analysis Module 7 and 8

The GUI is composed of six axes that gives the ability to look in different point of view and to compare data. The listbox on the left shows the files to be used. The icons at the top screen are for fast use of common operations. Its gives the ability to analyse population of cells, and an efficient way to screen and plot all the data generated by the Measurements Module.

Instructions: 1. Import Data Files: Input files for Analysis Module has to be .Fig files format from the data libraries generated by Measurements Module. Individual .fig files can also be imported directly into the collection box. 2. Once the files selected, the list will appeared in the list box. 3. To explore a stack file quickly, the user has to select the specific file from the files list and to load it to the computer memory. The file can be controlled by the control buttons, and can be modified by using filters and is ready for analyzing.

Screen capture of TACTICS Analysis Module interface trajectories montage

Scatter plot projections

TACTICS v3.x by R.S. 2010-2014 -191- Appendix I

Lineage Module 1 (create data structure file) To create the file for ACD Module : 1.Use the Measurement Module to create measurement list, as save the list. 2.Use Robust Module to automatically run iterative process of generating folders of many .fig files. Each .fig corresponding to cell from division (parental, daughter 1, daughter 2) from the time lapse movies. This step is time consuming from the computational side. Once the user trigger this process, it can take few hours depending on the number of positions. 3.Use the Analysis Module to load the data libraries. 4. From main Go to >’Export’>> ‘Generate ACD file’. This will process the data stored in the boxes according the scheme shown in p. ?? (previous one), export and save.

1. Create measurement list

2. Create data libraries

3. Load to Analysis Module (box 1: montage control, box 2: montage POI, box 3: trajectories)

4. Export

TACTICS v3.x by R.S. 2010-2014 -192- Appendix I

Lineage Module 2 (generating Lineage data file)

To create Lineage data file follow the next instructions.

1- From the Measurements Module: create measurements file as explained in p. 33-34.

These measurements files are provided as example in the next folder: TACTICS\supporting data\Lineage measurements files

2- From the Robust Module: create libraries as shown in p. 39. This an take few hours

The measurement files are located within defined folder

TACTICS v3.x by R.S. 2010-2014 -193- Appendix I

Lineage Module 2 (generating Lineage data file)

3- From the Analysis Module: upload data boxes 1,2,3 (see instructions to load data into the Lineage Module in p. 38)

montages ch01 montages ch02 trajectories 4- Export from Manu: Export \Export 5- To connect the lineages data: Run the connecting lineages lineage data for multiple positions. module and follow steps 1,2,3. This will automatically create lineage Step 1 to load lineage of parents A, step 2 is to load lineage of data for each position: (takes several daughters B, and step 3 is to connect B to selected cell from A. minutes)

Input spacers as relative time where there was no imaging (i.e. time took to transfer cells from one position to another). Dilute lineage if time rate is different. NaN`s values will fill temporal gaps.

Alternative to 5 - Robust connecting is faster when many lineages required to be connected. Select the folder where the lineages data located.

TACTICS v3.x by R.S. 2010-2014 -194- Appendix I

Lineage Module 2 (generating Lineage data file)

8- Input parameters for each position or randomly assign positions (steps 8a-8b)

Random assignment is important for experiments where connection is not reliable, but yet lineage can be connected for other types of analysis such as analysis that is based on division cycles. Blue represent reliable connection. Green represent random connection.

6- Select founder, daughters, input 8a- To randomly assign positions spacer and interval time. Click the 7- Select number of choose ‘Rand’ for the selected pink step 2 button. The lineages of positions (lineages to position, and click on ‘Choose daughter 1 and 2 will be connected to connect) to be random’ button. This will open a the founder. connected new window

8a- select positions and potential cells, and click finish after the selection. This will close the window and assign 9- Click on the red button to the positions to cells. finish. This will automatically 10. Final step: Go to >Label lineage to export the lineage connect all the lineages in the data. list

TACTICS v3.x by R.S. 2010-2014 -195- Appendix I

Lineage Module 3 (adding parameters to lineage)

1- Load lineage data file 3- Choose parameter (currently only number of cells is available) and click on the ‘Create parametric tree’. The parametric tree (before the connection) will be displayed in the right axes

2- Load parameter vector (.fig file). Optional: apply moving average (MA) and dilution

4- Connect to any required cell. After connection the vector continue the parameter of the linked cell

5. export the lineage data.

TACTICS v3.x by R.S. 2010-2014 -196- Appendix I

Lineage Module 4 (multiple gating mode)

To use multiple gating mode 1.Go to multiple gating mode. 2.Set the right panel axes1 to create temporal gates (remove all existing gate from axes1 if required). 3.Click on one of the gating toggle buttons with mouse right click. A new popup window will come up. 4.Histograms of x-axis and y-axis gated data are shown, matching to defined gates. 5.There are three different options: i. Transfer gate to axes 1: The location of the selected gate will be transferred to axes1,enableing to change the location of existing gate. ii. Edit Gate: will allow to redefine name and color , but keeping existing location of the gate. iii. Create new Gate: to define completely new gate (including to define gate name, color, and the mode of the gate) using the location in axe 1 and . 6.Alternative to steps 3-4, user can define gate from existing gates, point on the designated toggle button, hold left mouse button, and pressing keyboard ‘space’ will open gating operations window. 7.Clicking left mouse button on each toggle button of defined gate will activate/deactivate it.

7- left mouse button on each gate will activate/deactivate

3- right mouse click to call the gate window

1- select multiple gating mode

2- remove temporal gate before setting new gate and define temporal gate

5.i.

5.iii. 5.ii.

4- histogram matching to defined gates

6- define new gate using logical operations

TACTICS v3.x by R.S. 2010-2014 -197- Appendix I

Lineage Module 4 (multiple gating mode)

To use logical operations between gating :

1. Go to >Gating>> Gate on lineage tree. New window will come up, displaying the lineage tree and color tagged each cell. Selection can be either base on inclusive gates (select data that falls within the limits set) and exclusive gates (select data that falls outside the limits). 2. Define name, mode, and color for the new gate. 3.Choose cells to gate on the lineage tree by clicking on the cell lines or by kinship. For instance, 1.2.* will mark all ancestors of cell 1.2. 4.Save and update. 5.The new gate will appear in the Lineage Module panel. 6. Point with the mouse arrow on specific gate panel and click Ctrl key. 7.Currently available simple logical operations (such as And, Or, and Xor), but more can be easily added.

1- open the gating on lineage tree window

3- choose cells

2- set parameters of new gate

5- gating panel

4- new gate is active

TACTICS v3.x by R.S. 2010-2014 -198- Appendix I

Lineage Module 5 (exploration tools)

Lineage trees based analysis is ready from axes1,axes2 or axes3 (gated or ungated). Currently there are three different types of analysis: 1. Go to >Options>>Cell Measurements for selected progeny

This tool provides plot different parameters of selected progeny or selected cells from progeny. The parameters are y1 vs. x1 from the Lineage Module. To plot y1 vs. time choose x1 as internal counter. The lineage tree appears in red lines. To select cells, user click on the arms to convert to blue lines (clicking again will convert back again to blue for unselecting the cells). Once wanted cells were selected, the data is shown in subplot for chosen parameter.

2. Go to >Options>>Measurements for selected progeny vs. each other

3. Go to >Options>>Measurements on generation base

TACTICS v3.x by R.S. 2010-2014 -199- Appendix I

Lineage Module 6 (measured parameters)

‹ Currently available list: 1. D1/D2: daughter 1 or 2 after first division (value 0 or 1) 2. Generation number: generation index (i.e.: for 1.2.2.1 n=4) 3. Lifespan: arm length(time take to divide or dye) 4. Time at birth(starting time point) 5. Time at death/division(ending time point) 6. Class (0 for parental, 1 for daughter 1, 2 for daughter 2) 7.Eccentricity 8.aspect_ratio 9.Perimeter 10.MinorAxisLength 11.Area 12. MajorAxisLength 13. EquivDiameter 14. Orientation 15. Circularity 16. Total intensity Ch01 17. Total intensity Ch02 18. Fate (0.1 for division, 0.2 for death) 19. Under development 20. Turning angle (angle between trajectories in C(t-1) and C(t), C is the centroid) 21. Angle between x-axis and C(t+1 22. Angle between x-axis and C(t-1) 23. Distance travelled (C(t-1) into C(t+1)) 24. Distance travelled (C(t) into C(t+1)) 25. Distance travelled (C(t-1) into C(t)) 26. Time from birth 27. Cell index (Cell ID) 28. Position: position based on microwell, experiment, or founder 29. Total counter: Time from the beginning of the lineage 30. Cell counts: Number of cells at a given time point 31.Normalized cell counts

‹ The parameter builder menu can be utilized to generate a large range of new parameters.

TACTICS v3.x by R.S. 2010-2014 -200- Appendix I

Video Polarization Module tutorial 9 Cell polarity is described as asymmetric geometry and unequal distribution of components within a cell. TACTICS can be applied to study regulation of cell polarity by supplying data about individual cells such as cell morphology, localization and movement. The Polarization Module offers an efficient way to screen and plot all the data (i.e.- intensity distribution of fluorescent gene of interest, speed, shape, size) from microscopy in the same way as flow cytometry programs (i.e.-FlowJo, FCS express). Read- Divergent lymphocyte signalling revealed by a powerful new tool for analysis of time-lapse microscopy (Pham et al., ICB, 2012). Histograms of back gated data (For Y and X) Information about selected (pointed) cell

Gating panel (polygon or square)

Projections provide a qualitative info

Image of selected cell including marker of centroid, location of brightest Back-gated data points pixel in control and POI channels, and trajectories. Screen capture of TACTICS Polarization Module interface

1. Choose normalization: 2. Choose data type: 3. Choose the type of polarization:

•Manually rotated cells Absolute POI raw •Random rotation Absolute control raw •Major axis by elongation log2(U/L) signed POI raw •Minor axis by elongation (U-L)/(U+L) Signed control raw •Y angle y axis and C(t+1) (Ui^2-Li^2)/(Ui^2+Li)^2 Absolute POI raw •X angle x axis and C(t+1) log2(Ui/Li) Absolute control raw •Y angle y axis and C(t-1) signed POI raw •X angle x axis and C(t-1) Signed control raw •Y angle y axis and brightest pixels in POI •X angle x axis and brightest pixels in POI •Y angle y axis and brightest pixels in control •X angle : x axis and brightest pixels in control

TACTICS v3.x by R.S. 2010-2014 -201- Appendix I

ACD Module 1 (the interface)

Asymmetrical Cell Division (ACD) is asymmetrical property between two daughters cells. TACTICS has the ACD Module , which is dedicated interface to investigate and characterize ACD at the single-cell level for all bulk population. The ACD Module offers several attributes that are necessary to alleviated quantification for ACD based on fluorescence time lapse data: 1. Axis choices (i.e. PRmajor/minor, CH1/2, morphology, velocity, etc) and the ability to gate on parameters, specific cells or division frame, and to extract data on selected events. In addition: a. Focusing in particular data points to inspect, explore and pool additional data. b. The possibility to link between multi-parameter data and the original movies (in similar to the Polarization Module). c. Linked plots of PRmajor vs. minor for Cl1 (e.g. control protein), exclude cells with high PRmajor by gating on that plot, and then observed the refined population on an adjacent plot of PRmajor vs. minor for Channel 2 (e.g. protein of interest), as shown in (cite the PLOSC) d. The user has the possibility of selecting individual divisions for longitudinal analysis, for instance PR over time (major/minor/Ch1/Ch2). e. To robustly evaluate correlation between multiple parameters. 2. Data output (e.g. PRmajor) is on selected range of coupled daughter cells base or parental. 3. The ability to choose the most appropriate thresholding settings in an objective fashion using PR minor and either all or a subset of the divisions. Moving the scrollbar interactively apply thresholding setting to compare PRmajor and minor for protein of interest and/or control protein. 4. Heat-maps or a 2D dot plots of the PR s function of threshold for all cells for PRmajor/minor of control and POI, which allowing the user to select the optimal threshold setting. 5. Selecting different cells to display (aligned, sub-divided, different channel, colocalizes ,etc.). This includes the ability to export the data/images for records/publication, and/or look at the data to choose the thresholding value. 7. Visualization of divisions (i.e. montages, projections) 3. Interactive scrollbar to adjust the T level

4. Heat-maps visualizing the effect of T level

2. range of coupled daughter cells

1. Visualization of data points and gating

5. Select cell/division to view

TACTICS v3.x by R.S. 2010-2014 -202- Appendix I

ACD Module 2 (data structure)

To use the ACD Module the user needs to create structured data file that contains information about the cells, whereas each data point is given multiple parameters in the following organization: 4 5 6 7 8 1 2 3 D1 P x D2

Division 1 9 10 17 11 12 13 14 15 16 D1 P x D2 Division 2 18 And so on, structuring the next format:

where n is the number of parameters

•The input can be from a single or several experiments, and from multiple positions. •The order of the parameters is not important. However the order should be the same when exporting from the Analysis Module and importing to the ACD Module.

TACTICS v3.x by R.S. 2010-2014 -203- Appendix I

About Figure in MATLAB

Data visualization in TACTICS is displayed in a MATLAB Figure, which provides a complete environment and framework for data visualization. The figure can include additional tools such as data statistics, basic fitting, zoom, colorbar, legend, and more. The figure can be saved and load as it is, or exported in other format such as *.EPS (Encapsulated Post Script file for Latex or Illustrator).

One of the advantages of TACTICS is the ability to export data (i.e.- numeric arrays, strings, or structures) to MATLAB workplace for further analysis. This data is stored in the ‘userdata’ property of the MATLAB figures, and can be copied into the workplace by using the MATLAB command ‘data=get(gcf,’userdata’)’, whereas data is the name of the destination variable. The data can be then stored and loaded trough ‘.mat’ files format using the MATLAB ‘save’ and ‘load’ function of the variable that contain the data.

Users with basic knowledge about MATLAB windows, basic commands, and MATLAB figure environment can easily and quickly visualize the data in a set of display command functions such as imagesc, surf,mesh, plot,bar and hist. Each figure properties can be easily chanced. A control panel gives the option to visual the original, filtered, and segmented images including tactical plot showing cell index, trajectories, and proximity vectors between cells in different channels.

For more details go to MATLAB Help navigator.

TACTICS v3.x by R.S. 2010-2014 -204- Appendix I

Troubleshooting

• Z_projector If a red error appeared this can be because: 1. You started from position 41 but forgot to edit the Starting from position value .Please edit this value and press RUN again. 2. You started to project from the wrong section or channel. One common mistake is to project from z00 but the Leica software always starts its z sections from z01. 3. You already deleted the non-projected files from the pos ** folder. Delete the all position/s and restart the process.

• Cell Tracking Module Memory issue is expected in Cell Tracking Module if the number of segments exceeds reasonable number of segments as a result of wrong segmentation settings.

TACTICS v3.x by R.S. 2010-20141 Appendix I

Adding filtering option- editing Filter File MATLAB user can easily add more filter options upon demand. Instructions are as the following: 1.Open Filter_file.m located in supporting functions library with MATLAB editor. 2. Add indexed case. For instance, if there are 20 cases, add case 21. 3.For this case add the next format: case 21 if nargin==0 matrix(11)={ ‘user given name for the operation'} else matrix(:,:,1)=operation added by the user end %Example: %case 11 % if nargin==0 % matrix(11)={ 'Canny edge detection'} % else % matrix(:,:,1)=Canny(matrix(:,:,1)); % end Whereas nargin==0 is used to give the function name, matrix(:,:,1) is the returned output (filtered image), 11 is the case index Canny is the function used on input matrix matrix can be in 3-D format 4. Save and exit the Filter_file.m . 5. Open TAC_Segmentation_Module.m with MATLAB editor. 6. Add to the indexed case under F_popup_function slider setting. This settings are depended on the maximum input value. i.e.: eval( strcat ('set(handles.F_edit_',num2str(ii), ',','''Visible''', ',', '''on''', ')')); eval( strcat ('set(handles.F_edit_',num2str(ii), ',','''String''', ',0)')); eval( strcat ('set(handles.F_Slider_',num2str(ii), ',','''Visible''', ',', '''on''', ')')); eval( strcat ('set(handles.F_Slider_',num2str(ii), ',','''Max''', ',100)')); eval( strcat ('set(handles.F_Slider_',num2str(ii), ',','''Min''', ',0)')); eval( strcat ('set(handles.F_Slider_',num2str(ii), ',','''Value''', ',0)')); eval( strcat ('set(handles.F_Slider_',num2str(ii), ',','''Sliderstep''', ',[0.005 0.1])')); 7. Save and exit TAC_Segmentation_Module.m .

TACTICS for developers 1 (expending filters)

Different cell shape and size may require installation of other image processing procedures. This can be easily installed as the following:

TACTICS v3.x by R.S. 2010-20141 Appendix I

Adding segmentation operations- editing Segmentation Module File MATLAB user can easily add more segmentation and binary operations options upon demand. Instructions are as the following: 1.Open Segmantation_file.m located in supporting functions library with MATLAB editor. 2. Add indexed case. For instance, if there are 10 cases, add case 11. 3.For this case add the next format: case 11 if nargin==0 matrix(11)={ ‘user given name for the operation'} else matrix=operation added by the user end %Example: %case 7 % if nargin==0 % matrix(7)={ 'imfill holes'} % else % matrix=imfill(matrix,'holes'); % end

Whereas nargin==0 is used to give the function name, matrix is the input and returned output image 7 is the case index imfill is the function used on input matrix matrix can be only in 2-D format 4. Save and exit the Segmantation_file.m . 5. Open TAC_Segmentation_Module.m with MATLAB editor. 6. Add to the indexed case under T_popup_function slider setting. This settings are depended on the maximum input value. i.e.: eval( strcat ('set(handles.T_edit_',num2str(ii), ',','''Visible''', ',', '''on''', ')')); eval( strcat ('set(handles.T_edit_',num2str(ii), ',','''String''', ',0)')); eval( strcat ('set(handles.T_Slider_',num2str(ii), ',','''Visible''', ',', '''on''', ')')); eval( strcat ('set(handles.T_Slider_',num2str(ii), ',','''Max''', ',1)')); eval( strcat ('set(handles.T_Slider_',num2str(ii), ',','''Min''', ',0)')); eval( strcat ('set(handles.T_Slider_',num2str(ii), ',','''Value''', ',0)')); eval( strcat ('set(handles.T_Slider_',num2str(ii), ',','''Sliderstep''', ',[0.005 0.05])')) 7. Save and exit TAC_Segmentation_Module.m .

TACTICS for developers 2 (expending segmentation)

TACTICS v3.x by R.S. 2010-20141 Appendix I

TACTICS for developers 3 (expending measurements)

Adding more properties to be measured by the Measurements_Module: 1. The function: for this example, a function that gives the polarization of the protein distribution: Get_Cell_Polarisation. 2. The type : the type will be type 2 for this function : can be plotted vs. other parameter. 3. add 'Polarisation' to define the plot list. Apply modifications in the next functions: > function popup1_Callback(hObject, eventdata, handles), >function popup2_Callback(hObject, eventdata, handles) :popup2_str >function popup3_Callback(hObject, eventdata, handles): popup3_str >If the type is to dividing pairs : parental_num_Callback(hObject, eventdata, handles) I>f the type is on cell base function Div_Cells_Callback(hObject, eventdata, handles)

Set_TAC_Measurments_Module_Settings(hObject, eventdata, handles,TAC_Measurments_Module_Settings) no….

4. to add the function called from cell, population, dividing at function Go_Callback(hObject, eventdata, handles) because function is type 2: if findstr(str,'Dividing- Polarisation')==1 18 Dividing_plot_function(handles,n,D1,D2,Vs,'Polarisation') ; end if findstr(str,'Population- Polarisation')==1 19 Population_plot_function(handles,'Polarisation',Vs,n) ; end if findstr(str,'Cell- Polarisation')==1 19 Cell_plot_function(handles,n,'Polarisation',Vs) ; end 5. add the sub-call: at Cell_plot_function(handles,n,str_in,Vs) : if strfind(Vs,'Polarisation')>1 X=Get_Cell_Polarisation(handles,n,Vs) ; Vs_X='Polarisation'; end at Population_plot_function(handles,'Polarisation',Vs,n) ; : if findstr(Vs,'Polarisation') X=Get_Cell_Polarisation(handles,jj,Vs) ; end at Dividing_plot_function(handles,n,D1,D2,Vs,'Polarisation'): if findstr(Vs,'Polarisation') X=Get_Cell_Polarisation(handles,start_frame,Vs) ; end 6. create Get_Cell_Polarisation(handles,start_frame,Vs) ; function [Data_out]=Get_Cell_Polarisation(handles,n,Vs) Data=Get_Cell_stack(handles,n,Vs) ; Data_out=[]; for ii=1:size(Data,2) temp =Data(ii).cdata; temp=temp(temp>0); Data_out(jj)=(max(temp)-min(temp))/median(temp) ; end

TACTICS v3.x by R.S. 2010-2014

Appendix II

List of TACTICS files

Raz Shimoni Appendix II TACTICS files

The next table summeries the files of TACTICS.

The first column describes the names of the files, whilst the second column identifies the location of each file. The third column and the forth columns define the complexity and the size of each file, respectively. The complexity and size of each file are classified as Low (L), Medium (M), and High (H). Selection and conversion tools, very simple and short algorithms, and measurements or labelling functions are majorly scalssfied as L. Pipeline tools, simple medium length algorithms, and visualization tools are majorly scalssfied as M. Modules and complex algorithms are scalssfied as H. Functions with less than 500 lines of code are classified as L. Functions with less than 2000 and more than 500 lines of code are classified as M , while functions with more than 2000 lines of code are classified as L.

Raz Shimoni Appendix II TACTICS files

File name Library Functionality complexity size

Main TACTICS user TACTICS TAC\modules M H interface

TAC_Cell_Tracking_ TAC\modules\Tracking The main Tracking H H Module Module module interface

different versions of first F_associate TAC\modules\Tracking selective operator M L (1_n,1_r2,r1_n,r1_r2) Module\add association association algorithm

different versions of S_associate TAC\modules\Tracking selective operator M L (1_n,1_r2,r1_n,r1_r2) Module\add association association algorithm

different versions of O_associate (i.e. no TAC\modules\Tracking selective operator M L selective, n selective) Module\add association association algorithm for relabeling

TAC\modules\Tracking re-association for reassociation M L Module\add association selective operation mode

different versions of TAC\modules\Tracking O_unassociate (i.e. no selective operator for un- Module\remove M L selective, n selective) association algorithm for association relabeling

TAC\modules\Tracking un-association for unassociation Module\remove M L selective operation mode association

TAC\modules\Tracking function to split tracks of break_track Module\supporting M L dividing cells functions

TAC\modules\Tracking plotting parametric plot_lineage_param Module\supporting lineage tree to navigate M L functions the Tracking Module

plotting parametric TAC\modules\Tracking plot_lineage_param_c lineage tree to navigate Module\supporting M L ompressed the Tracking Module functions (Nan`s removed)

plotting parametric TAC\modules\Tracking plot_lineage_param_re lineage tree to navigate Module\supporting M L covered the Tracking Module functions (Nan`s recovered)

L L Association_settings_ TAC\modules\Tracking interface to define

Raz Shimoni Appendix II TACTICS files

GUI Module\supporting tools association settings

interface to navigate the TAC\modules\Tracking cell_lineage_window Tracking Module based M L Module\supporting tools on lineage trees

interface to navigate the cell_lineage_window_ TAC\modules\Tracking Tracking Module based M L param Module\supporting tools on parametric lineage trees

interface to navigate the TAC\modules\Tracking cell_window Tracking Module based M L Module\supporting tools on cells life span

interface to navigate the TAC\modules\Tracking cell_window_param Tracking Module based M L Module\supporting tools on cellular parameters

interface to define TAC\modules\Tracking Crocker_params parameters for Crocker L L Module\supporting tools tracking algorithm

TAC\modules\Tracking edit_tracks tool to edit tracks M L Module\supporting tools

interface to define TAC\modules\Tracking Paint_Tool_GUI parameters for the L L Module\supporting tools drawing tool

TAC\modules\Tracking Tracking_option_GUI tool to edit tracks M L Module\supporting tools

interface to define Tracking_settings_GU TAC\modules\Tracking parameters for the L L I Module\supporting tools Hungarian algorithm for cell association

TAC_Segmentation_ TAC\modules\Segmantio The main Segmentation H H Module n Module module interface

TAC\modules\Robust The main Robust module TAC_Robust_Module H H Module interface

TAC\modules\ACD The main ACD module TAC_ACD_Module H H Module interface

TAC_\Analysis_Modu TAC\modules\Analysis The main Analysis H H le Module module interface

TAC_Polarfization_M TAC\modules\Polarizatio The main Polarization H H odule n Module module interface

TAC\modules\Polarizatio tool to add new add_polarization_vect L L n Module\supporting formulations of

Raz Shimoni Appendix II TACTICS files

or functions polarization ratios

TAC_Connecting_Mo TAC\modules\Connectin The main Connecting H H dule g Module module interface

complimentary Connecting module TAC_Connecting_Mo TAC\modules\Connectin interface to connect H H dule2 g Module cellular parameters to lineages

TAC\modules\Connectin rand_connecting_linea interface to randomize g Module\supporting L L ge connection of lineages tools

TAC\modules\Connectin algorithm to connect to connect_AB g Module\supporting M L lineages functions

TAC\modules\Connectin function that connect connect_AB_matrix g Module\supporting vector of parameters to L L functions selected cell from lineage

TAC\modules\Connectin function that structure create_data2 g Module\supporting L L lineage data functions

TAC\modules\Connectin function that dilute dilute_function g Module\supporting lineage data by inserting M L functions NaN`s

label data from the TAC\modules\Connectin Connecting Module and label_parameters g Module\supporting M L structure in readable functions format to Lineage Module

TAC\modules\Connectin plot parametric lineage plot_Cell_Lineage_ma g Module\supporting that was converted to L L trix functions matrix

TAC\modules\Connectin upload_cell g Module\supporting upload selected cell L L functions

TAC_Cell_Lineage_ TAC\modules\Lineage The main Lineage module H H Module Module interface

Add_new_parameter TAC\modules\Tracking collection of interfaces to (e.g. Events, M H Module\supporting tools build new parameters MA_per_cell) gate_cell_lineage_GU TAC\modules\Tracking interface to gate on L L I Module\supporting tools lineages

Raz Shimoni Appendix II TACTICS files

interface to perform Lineage_Module_logi TAC\modules\Tracking logical operations on M L cal_gate Module\supporting tools lineages

interface to project TAC\modules\Tracking Project_param_GUI parameters from lineage M L Module\supporting tools trees

collection of interfaces to select_cell_lineage_G TAC\modules\Tracking select, compare and M H UI Module\supporting tools correlate cells based on different criteria

TAC_Cell_Measurem TAC\modules\Measurem The main Measurements H H ents_Module ents Module module interface

calculator of different TAC\modules\Measurem normalization Consider_8_GUI ents Module\supporting formulations between L L tools different sections of a dividing cell

TAC\modules\Measurem Cell_2D_Projection_f ents Measurements functions L L unction Module\Measurements_li b

TAC\modules\Measurem Cell_3D_Projection_f ents Measurements functions L L unction Module\Measurements_li b

TAC\modules\Measurem Cell_Angle_MaxP_pr ents Measurements functions L L oximity_function Module\Measurements_li b

TAC\modules\Measurem Cell_brightest_Pixel_ ents Measurements functions L L X_function Module\Measurements_li b

TAC\modules\Measurem Cell_brightest_Pixel_ ents Measurements functions L L Y_function Module\Measurements_li b

TAC\modules\Measurem ents Cell_data_function Measurements functions L L Module\Measurements_li b

Cell_distance_from_or TAC\modules\Measurem Measurements functions L L igion_function ents

Raz Shimoni Appendix II TACTICS files

Module\Measurements_li b

TAC\modules\Measurem ents Cell_Edge_function Measurements functions L L Module\Measurements_li b

TAC\modules\Measurem Cell_Maximum_pixel ents Measurements functions L L _function Module\Measurements_li b

TAC\modules\Measurem Cell_Montage_distanc ents Measurements functions L L e_transform_function Module\Measurements_li b

TAC\modules\Measurem Cell_Montage_functio ents Measurements functions L L n Module\Measurements_li b

TAC\modules\Measurem Cell_Montage_waters ents Measurements functions L L hed_function Module\Measurements_li b

TAC\modules\Measurem ents Cell_Movie_function Measurements functions L L Module\Measurements_li b

TAC\modules\Measurem ents Cell_MSD_function Measurements functions L L Module\Measurements_li b

TAC\modules\Measurem ents Cell_plot_function Measurements functions L L Module\Measurements_li b

TAC\modules\Measurem Cell_Montage_waters ents Measurements functions L L hed_function Module\Measurements_li b

TAC\modules\Measurem ents Cell_Movie_function Measurements functions L L Module\Measurements_li b

Cell_MSD_function TAC\modules\Measurem Measurements functions L L ents

Raz Shimoni Appendix II TACTICS files

Module\Measurements_li b

TAC\modules\Measurem ents Cell_plot_function Measurements functions L L Module\Measurements_li b

TAC\modules\Measurem Cell_projection_water ents Measurements functions L L shed_function Module\Measurements_li b

TAC\modules\Measurem Cell_Proximity_vector ents Measurements functions L L _function Module\Measurements_li b

TAC\modules\Measurem Cell_Shell_Projection ents Measurements functions L L _function Module\Measurements_li b

TAC\modules\Measurem Cell_Trajectories_func ents Measurements functions L L tion Module\Measurements_li b

TAC\modules\Measurem Cell_turning_angle_fu ents Measurements functions L L nction Module\Measurements_li b

TAC\modules\Measurem Data_for_images_Con ents Measurements functions L L fluency_function Module\Measurements_li b

TAC\modules\Measurem Data_for_images_Mea ents n_Area_for_object_fu Measurements functions L L Module\Measurements_li nction b

TAC\modules\Measurem Data_for_images_Mea ents n_Image_Intensity_fu Measurements functions L L Module\Measurements_li nction b

TAC\modules\Measurem Data_for_images_Mea ents n_Intensity_for_object Measurements functions L L Module\Measurements_li _function b

Data_for_images_Nu TAC\modules\Measurem Measurements functions L L mber_of_objects_func ents

Raz Shimoni Appendix II TACTICS files

tion Module\Measurements_li b

TAC\modules\Measurem Data_for_images_Nu ents mber_of_Pixels_functi Measurements functions L L Module\Measurements_li on b

TAC\modules\Measurem Data_for_images_Ots ents Measurements functions L L u_value_function Module\Measurements_li b

TAC\modules\Measurem Data_for_images_Tota ents l_Image_Intensity_fun Measurements functions L L Module\Measurements_li ction b

TAC\modules\Measurem Dividing_2D_Projecti ents Measurements functions L L on_function Module\Measurements_li b

TAC\modules\Measurem Dividing_3D_Projecti ents Measurements functions L L on_function Module\Measurements_li b

TAC\modules\Measurem Dividing_Edge_functi ents Measurements functions L L on Module\Measurements_li b

TAC\modules\Measurem Dividing_Montage_fu ents Measurements functions L L nction Module\Measurements_li b

TAC\modules\Measurem Dividing_Movie_funct ents Measurements functions L L ion Module\Measurements_li b

TAC\modules\Measurem Dividing_MSD_functi ents Measurements functions L L on Module\Measurements_li b

TAC\modules\Measurem Dividing_plot_functio ents Measurements functions L L n Module\Measurements_li b

TAC\modules\Measurem Dividing_SEQtage_fu Measurements functions L L ents

Raz Shimoni Appendix II TACTICS files

nction Module\Measurements_li b

TAC\modules\Measurem Dividing_Trajectories ents Measurements functions L L _function Module\Measurements_li b

TAC\modules\Measurem dscatter2_TACWrapp ents Measurements functions L L er Module\Measurements_li b

TAC\modules\Measurem ents Extendable features lib Measurements functions L L Module\Measurements_li b

TAC\modules\Measurem ents Get_Cell_Area Measurements functions L L Module\Measurements_li b

TAC\modules\Measurem ents Get_Cell_Circularity Measurements functions L L Module\Measurements_li b

TAC\modules\Measurem ents Get_Cell_Eccentricity Measurements functions L L Module\Measurements_li b

TAC\modules\Measurem ents Get_Cell_Ellipticity Measurements functions L L Module\Measurements_li b

TAC\modules\Measurem Get_Cell_EquivDiame ents Measurements functions L L ter Module\Measurements_li b

TAC\modules\Measurem ents Get_Cell_Extent Measurements functions L L Module\Measurements_li b

TAC\modules\Measurem Get_Cell_graycoprops ents Measurements functions L L _Contrast Module\Measurements_li b

TAC\modules\Measurem Get_Cell_graycoprops Measurements functions L L ents

Raz Shimoni Appendix II TACTICS files

_Correlation Module\Measurements_li b

TAC\modules\Measurem Get_Cell_graycoprops ents Measurements functions L L _Energy Module\Measurements_li b

TAC\modules\Measurem Get_Cell_graycoprops ents Measurements functions L L _Homogeneity Module\Measurements_li b

TAC\modules\Measurem Get_Cell_graycoprops ents Measurements functions L L _std2 Module\Measurements_li b

TAC\modules\Measurem ents Get_Cell_I_per_A Measurements functions L L Module\Measurements_li b

TAC\modules\Measurem ents Get_Cell_Intensity Measurements functions L L Module\Measurements_li b

TAC\modules\Measurem ents Get_Cell_limstack Measurements functions L L Module\Measurements_li b

TAC\modules\Measurem Get_Cell_number_of_ ents Measurements functions L L disks Module\Measurements_li b

TAC\modules\Measurem Get_Cell_number_of_ ents Measurements functions L L peaks_X Module\Measurements_li b

TAC\modules\Measurem Get_Cell_number_of_ ents Measurements functions L L peaks_Y Module\Measurements_li b

TAC\modules\Measurem ents Get_Cell_Orientation Measurements functions L L Module\Measurements_li b

Get_Cell_Perimeter TAC\modules\Measurem Measurements functions L L ents

Raz Shimoni Appendix II TACTICS files

Module\Measurements_li b

TAC\modules\Measurem ents Get_Cell_Polarisation Measurements functions L L Module\Measurements_li b

TAC\modules\Measurem ents Get_Cell_Solidity Measurements functions L L Module\Measurements_li b

TAC\modules\Measurem ents Get_Cell_stack Measurements functions L L Module\Measurements_li b

TAC\modules\Measurem ents Get_Cell_Velocity Measurements functions L L Module\Measurements_li b

TAC\modules\Measurem ents Get_Dividing_stack Measurements functions L L Module\Measurements_li b

TAC\modules\Measurem Get_Dividing_Velocit ents Measurements functions L L y Module\Measurements_li b

TAC\modules\Measurem Population_2D_Projec ents Measurements functions L L tion_function Module\Measurements_li b

TAC\modules\Measurem Population_Area_func ents Measurements functions L L tion Module\Measurements_li b

TAC\modules\Measurem Population_Cell_Num ents Measurements functions L L ber_function Module\Measurements_li b

TAC\modules\Measurem Population_Eccentricit ents Measurements functions L L y_function Module\Measurements_li b

TAC\modules\Measurem Population_MSD_fun Measurements functions L L ents

Raz Shimoni Appendix II TACTICS files

ction Module\Measurements_li b

TAC\modules\Measurem Population_Orientatio ents Measurements functions L L n_function Module\Measurements_li b

TAC\modules\Measurem Population_plot_functi ents Measurements functions L L on Module\Measurements_li b

TAC\modules\Measurem Population_Trajectorie ents Measurements functions L L s_function Module\Measurements_li b

TAC\modules\Measurem Population_Velocity_f ents Measurements functions L L unction Module\Measurements_li b

TAC\modules\Measurem ents set_popup2_str Measurements functions L L Module\Measurements_li b

TAC\modules\Measurem ents set_popup3_str Measurements functions L L Module\Measurements_li b

lunch TACTICS and run_TAC TAC\ define path of TACTICS L L folders

Wrapper for MATLAB boxplot_TACWrapper TAC\MATLAB wrapper function boxplot that L L allows selection of colors

Wrapper for MATLAB boxplotCsub TAC\MATLAB wrapper function boxplot that L L allows selection of colors

Wrapper for MATLAB function bwlabel that bwlabel_max TAC\MATLAB wrapper L L allows to limit number of labeled cells

Wrpper for MATLAB function close that allows Close TAC\MATLAB wrapper L L to close all figures but to keep GUIs

Raz Shimoni Appendix II TACTICS files

Example for MATLAB properties to send emails my_notifier TAC\MATLAB wrapper L L (i.e. notify status of batch processing)

Wrapper for MATLAB nanmean_resize TAC\MATLAB wrapper function resize but L L enabled for NaN`s

Wrapper for MATLAB nlfilter_TACWrapper TAC\MATLAB wrapper L L function nfilter

Function for spline smoothing of a 2-D input spaps_smooth TAC\MATLAB wrapper L L matrix using MATLAB spaps

counting time to wait (use wait_pause TAC\MATLAB wrapper to let the memory to set L L before continuing)

compress parameter compress_vector TAC\supporting functions M L vectors

replace NaN`s by cover_nan_function TAC\supporting functions M L neighboring values

return the trajectories of create_vector TAC\supporting functions M L selected cell as vector cut2four_monochanne TAC\supporting functions channel M L l

subdivide cell in multiple cut2four_multichannel TAC\supporting functions M L channel into 4 equal parts

function to split one 2-D object into two halves cutter_function2 TAC\supporting functions M L with the same area by a defined axis

decompress parameter decompress vector TAC\supporting functions M L vectors

Link filtering files (such Filter_file TAC\supporting functions as from MATLAB IPT) L L and TACTICS

Id cells by allocating Find_Centroids TAC\supporting functions L L centroids

Label cells by Find_L TAC\supporting functions L L regionprops function

Raz Shimoni Appendix II TACTICS files

Tracking algorithm linker Find_Tracks TAC\supporting functions H L based on frames

Link the Crocker tracking Find_Tracks_Crocker TAC\supporting functions L L algorithm to TACTICS

Find_Tracks_vectorize Fast tracking algorithm TAC\supporting functions H L d linker based on cells

Fast tracking algorithm Find_Tracks_vectorize TAC\supporting functions linker based on cells with H L d_selective selective operator

customized function to Get_division_8 TAC\supporting functions M L detect divisions

return last frame of get_last_cell_index TAC\supporting functions L L selected cell

algorithm to split cells by I_split_Xaxis TAC\supporting functions intensity gradient using M L intensity_split_function

function that uses I_split_Xaxis_2nd_ste I_split_Xaxis to separate TAC\supporting functions M L p_use_sections overlapping cells from best section

function that uses I_split_Xaxis_2nd_ste I_split_Xaxis to separate TAC\supporting functions M L p_without_sections overlapping cells from one

algorithm that moves intensity_split_functio TAC\supporting functions pointer according Ler M L n pixels intensity

add pixels to segments to make2round TAC\supporting functions return a more circular M L morphology

merge_lineage TAC\supporting functions merge to lineages two one M L

merged 3 channels Merged_image TAC\supporting functions L L (R,G,B) into one

reduce linear my_unmix TAC\supporting functions bleedthrough between L L fluorescence channels

use second threshold Otsu_2nd_step_use_se (applied only to already TAC\supporting functions M L ctions segmented cells) to split overlapping cells

Raz Shimoni Appendix II TACTICS files

Otsu_2nd_step_use_secti Otsu_2nd_step_witho TAC\supporting functions ons but based on one H L ut_sections section

plot_Cell_Lineage different versions of (1,2,22,3,33,gen,parm TAC\supporting functions M L plotting lineage trees _rec)

functions to measure ratio (X,Y, long axis, TAC\supporting functions polarization ratios using M L manual) different axis of polarity

read tif images (i.e. based read_image3 TAC\supporting functions L L on time, channel, section)

rotate image based on Rotate_by_Tubulin TAC\supporting functions M L brightest pixel intensity

split dividing cell to 4 SCD_SIM_SPLIT TAC\supporting functions M L sections with equal area

Link files or post- segmentation operation segmentation_file TAC\supporting functions L L (such as from MATLAB IPT) and TACTICS

function that call a saved set_new_filename TAC\supporting functions image from the L L Segmentation Module

function that works with South_North_function TAC\supporting functions SCD_SIM_SPLIT to label L L the for different section

project stack of 2-D stack2projection TAC\supporting functions M L images by x axis

project stack of 2-D stack2Yprojection TAC\supporting functions L L images by y axis

swap TAC\supporting functions swap two L L

similar to Threshold_second_seg Otsu_2nd_step_without_s TAC\supporting functions L L ment_without_split ections but without splitting overlapping cells

algorithm that associate founder cells and its TRYME TAC\supporting functions M L daughters based on proximity

change_LUT TAC\supporting tools Interface to selects LUT L L by different channels (e.g.

Raz Shimoni Appendix II TACTICS files

GFP, Cherry, etc.)

connect_D12 TAC\supporting tools Connects two cells L L

Interface to create new Experiment_Generator TAC\supporting tools L L experiment file

Interface to convert files MM2TACTICS TAC\supporting tools L L to TACTICS format

Interface to convert files Olympus2TACTICS TAC\supporting tools L L to TACTICS format

Interface to plot lineage plot_lineage_projectio TAC\supporting tools and all selection of L L n_GUI parameters

Interface to project Z_projector TAC\supporting tools multiple sections to on 2- L L D image

select_channel TAC\supporting tools Selection tool L L select_label_parameter TAC\supporting tools Selection tool L L

select_parameter TAC\supporting tools Selection tool L L

select_tag_gui TAC\supporting tools Selection tool L L set_origion_of_merge TAC\supporting tools Selection tool L L d_lineage

interface to change path Change_path TAC\supporting tools L L of experiment file

Appendix III

List of open-code files

Raz Shimoni Appendix III Source of Open-code Files

Open-code files are located in: TAC\open_code_lib\ with '_TACWrapper' extension (without the apostrophe).

______

1. File: hungarianlinker.m

Description: Simple multiple particle tracker with gap closing

Downloaded from- http://www.mathworks.com/matlabcentral/fileexchange/34040-simpletracker/content/SimpleTracker/hungarianlinker.m

______

2. File: kmeans.m

Description: kmeans image segmentation

Downloaded from - http://www.mathworks.com.au/matlabcentral/fileexchange/8379- kmeans-image-segmentation

______

3. File: Munkres.m

Description: Munkres' Assignment Algorithm, Modified for Rectangular Matrices

Reference: http://csclab.murraystate.edu/bob.pilgrim/445/munkres.html

Downloaded from- http://www.mathworks.com/matlabcentral/fileexchange/20652- hungarian-algorithm-for-linear-assignment-problems-v2-3

______

4. File: ANGLE_DEG_2D.m

Description: Matlab function that returns the angle swept out between two rays in 2D

Downloaded from- http://people.sc.fsu.edu/~jburkardt/f_src/dutch/dutch.f90

______

Raz Shimoni Appendix III Source of Open-code Files

5. File:vol3d.m

Description: 3-d volume (voxel) rendering

Downloaded from- http://www.mathworks.com/matlabcentral/fileexchange/22940-vol3d- v2/content/vol3d.m

______

6. File: track_crocker (based on track.m)

Description: Constructs n-dimensional trajectories from a scrambled list of particle coordinates determined at discrete times

Downloaded from- http://glinda.lrsm.upenn.edu/~weeks/idl

______

7. File: timebar.m

Description: Progress bar with estimated time remaining

Downloaded from- http://www.mathworks.com/matlabcentral/fileexchange/1255-timebar

______

8. File: speaker.m (based on tts.m)

Description: text-to-speech, speech synthesis, tts, let Matlab speak

Downloaded from- http://www.mathworks.com/matlabcentral/fileexchange/18091-text-to- speech)

______

9. File: Noisecomp.m

Description: Function for denoising an image using FFT

Reference: Peter Kovesi, "Phase Preserving Denoising of Images".

The Australian Pattern Recognition Society Conference: DICTA'99.

Raz Shimoni Appendix III Source of Open-code Files

December 1999. Perth WA. pp 212-217

Copyright: 1998-2000 Peter Kovesi, School of Computer Science & Software Engineering, The University of Western Australia

E-mail: http://www.csse.uwa.edu.au/

Downloaded from- http://www.cs.uwa.edu.au/pub/robvis/papers/pk/denoise.ps.gz.

______

10. File: Canny.m

Description: Canny edge detection

E-mail: http://www.csse.uwa.edu.au/

Downloaded from- http://www.csse.uwa.edu.au/~pk/research/matlabfns/Spatial/canny.m

______

11. File: nhist.m

Description: featured plot histograms

Downloaded from- http://www.mathworks.com/matlabcentral/fileexchange/27388-plot-and- compare-nice-histograms-by-default

______

12. File: Magnify.m (based on magnifyrecttofig.m)

Description: Opened new magnified image

Downloaded from- http://www.mathworks.com/matlabcentral/fileexchange/7286- magnifyrecttofig

______

13. File: Level_Set_Evolution.m

Description: level set method for object segmentation

Raz Shimoni Appendix III Source of Open-code Files

Reference: "Level Set Evolution Without Re-initialization: A New Variational Formulation", in Proceedings of CVPR'05, vol. 1, pp. 430-436.

Downloaded from- URL: http://www.engr.uconn.edu/~cmli/

______

14-16. File: smoothn.m

Description: Robust spline smoothing for 1-D to N-D data

Reference: Garcia D, Robust smoothing of gridded data in one and higher dimensions with missing values. Computational Statistics & Data Analysis, 2010

Downloaded from- http://www.biomecardio.com/matlab/smoothn.html

Supporting files:

IDCTN.m, N-D inverse discrete cosine transform.

DCTN.m, N-D discrete cosine transform.

Reference for supporting files: Narasimha M. et al, On the computation of the discrete cosine

transform, IEEE Trans Comm, 26, 6, 1978, pp 934-936.

______

17. File: houghcircles.m

Description: detects multiple disks (coins) in an image using Hough Transform.

Department of Information Management, Chaoyang University of Technology, Taichung, Taiwan

Downloaded from- http://www.mathworks.com/matlabcentral/fileexchange/22543-detects- multiple-disks-coins-in-an-image-using-hough-transform

______

18. File: getjframe.m

Description: retrieves the current figure (gcf)'s underlying Java frame,

Raz Shimoni Appendix III Source of Open-code Files

E-mail: [email protected]

Downloaded from - http://www.mathworks.com/matlabcentral/fileexchange/15830- getjframe-retrieves-a-figures-underlying-java-frame

______

19. File: bresenham.m

Description: Generate a line profile of a 2d image using Bresenham's algorithm

Reference: http://en.wikipedia.org/wiki/Bresenham's_line_algorithm

Downloaded from - http://www.mathworks.com/matlabcentral/fileexchange/12939- bresenhams-line

______

20. File: distinguishable_colors.m

Description: generates a set of colors which are distinguishable by reference to the "Lab" color space, which more closely matches human color perception than RGB

Downloaded from - http://www.mathworks.com/matlabcentral/fileexchange/29702-generate- maximally-perceptually-distinct-color

______

21. File: Imrotate2.m (based on IMROTATE.m)

Description: Rotate image, allowing for a non-black FILL.

______

22. File: dscatter.m

Description: creates a scatter plot coloured by density

Reference: Paul H. C. Eilers and Jelle J. Goeman, Enhancing scatterplots with smoothed densities, Bioinformatics, 004; 20: 623 - 628.

Raz Shimoni Appendix III Source of Open-code Files

Downloaded from - http://www.mathworks.com/matlabcentral/fileexchange/8430-flow- cytometry-data-reader-and-visualization

______

23. File: drlse_edge.m

Description: edge-based active contour

Reference: C. Li, C. Xu, C. Gui, M. D. Fox, "Distance Regularized Level Set Evolution and Its Application to Image Segmentation",

IEEE Trans. Image Processing, vol. 19 (12), pp.3243-3254, 2010.

Mutat Res, 2011, 711(1-2):49-60.

Downloaded from - http://www.engr.uconn.edu/~cmli/

______

24. File: TGI_Objects.mexw32, TGI_Objects.mexw64

Description: run a path of low intensity pixels to separate between cells where they contact

Reference: gammaH2AX foci as a measure of DNA damage: a computational approach to automatic analysis, Ivashkevich A.N., Martin O.A., Smith A.J., Redon C.E., Bonner W.M., Martin R.F., Lobachevsky P.N.,

Mutat Res, 2011, 711(1-2):49-60.