The Epithelial-Mesenchymal Transition in Engineered Microenvironments by Susan E. Leggett B.S., University of Vermont, 2013 Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy of Pathobiology at Brown University Providence, Rhode Island May 2018 c Copyright by Susan E. Leggett, 2018. All rights reserved. This dissertation by Susan E. Leggett is accepted in its present form by the Pathobiology Graduate Program as satisfying the dissertation requirements for the degree of Doctor of Philosophy Date Ian Y. Wong, Ph.D., Advisor Recommended to the Graduate Council Date Agnes B. Kane, M.D., Ph.D., Chair Date Alexander S. Brodsky, Ph.D., Reader Date Eric M. Darling, Ph.D., Reader Date Andrea I. McClatchey, Ph.D., External Reader Approved by the Graduate Council Date Andrew Campbell, Ph.D., Dean of the Graduate School iii Curriculum Vitae Susan E. Leggett received a Bachelor of Science degree in Biochemistry, with a minor in Pure Mathematics from the University of Vermont in 2013. In the fall of 2013, she started graduate work in the Pathobiology Program at Brown University and joined Dr. Ian Wong’s lab in the School of Engineering to pursue her dissertation research titled, “The Epithelial-Mesenchymal Transition in Engineered Microenvironments.” As an undergraduate student at the University of Vermont, Susan received several notable academic and research awards. She received the Award for Excellence in Chemistry in 2010, the Summer Internship Research Grant in 2012, the John Thanassi Award in Biochemistry in 2013, and was nationally recognized as a distinguished undergraduate research scholar when she received the Barry M. Goldwater Scholarship in 2012. Susan joined Dr. Yvonne Janssen-Heininger’s lab in the Department of Environmental Pathology in 2011, where she ultimately conducted her Honors Thesis research titled, “The Role of S-Glutathionylation in the Epithelial to Mesenchymal Transition.” In 2013, she graduated Cum Laude and as an Honors College Scholar. As a graduate student at Brown University, Susan received the NIEHS NRSA T32 Pre-Doctoral Fellowship from 2014 to 2017, the Simper-Ronan Graduate Award in Cancer Research in 2016, and the Frederic Poole Gorham Pre-Doctoral Fellowship in 2017. Susan has also been nominated for the Sigma Xi Honor Society and will be inducted as a full member in 2018. iv Journal Publications † = these authors contributed equally M. Patel, S.E. Leggett, A.K. Landauer, I.Y. Wong, and C. Franck (2018). “Rapid, topology-based particle tracking for high-resolution measurement of large complex motion fields.” Scientific Reports. 8, 5581. DOI:10.1038/S41598-018-23488-Y T.M. Valentin, S.E. Leggett, P.-Y. Chen, J.K. Sodhi, L.H. Stephens, H.D. Mc- Clintock, J.Y. Sim, and I.Y. Wong (2017). “Stereolithographic Printing of Ionically- Crosslinked Alginate Hydrogels for Degradable Biomaterials and Microfluidics.” Lab on a Chip. 17(20), 3474-3488. DOI:10.1039/C7LC00694B S.E. Leggett, A. Khoo, and I.Y. Wong (2017). “Multicellular Tumor Invasion and Plasticity in Biomimetic Materials.” Biomaterials Science. 5(8), 1460-1479. DOI:10.1039/C7BM00272F S.E. Leggett and I.Y. Wong. (2017) “Nanomedicine: Catching tumour cells in the zone.” Nature Nanotechnology. 12, 191-193. DOI:10.1038/NNANO.2016.264 S.E. Leggett, J-Y. Sim, J.E. Rubins, Z.J. Neronha, E.K. Williams, and I.Y. Wong (2016). “Morphological Single Cell Profiling of the Epithelial-Mesenchymal Transi- tion.” Integrative Biology. 8(11), 1133-1144. DOI: 10.1039/C6IB00139D M. Gamboa Castro, S.E. Leggett, and I.Y. Wong (2016). “Clustering and Jam- ming in Epithelial-Mesenchymal Co-Cultures.” Soft Matter. 12(40), 8327-8337. DOI:10.1039/C6SM01287F Z. Wang,† D. Tonderys,† S.E. Leggett,† M.T. Kiani, E.K. Williams, I.Y. Wong, and R.H. Hurt (2016). “Wrinkled, Wavelength-Tunable Graphene Oxide Surface Topographies Direct Cell Alignment and Morphology.” Carbon. 97, 14-24 (Cover). DOI: 10.1016/J.CARBON.2015.03.040 J.L.J. van der Velden, A.S. Guala, S.E. Leggett, J. Sluimer, E.C.H.L. Badura. Y.M.W. Janssen-Heininger (2012). “Induction of a mesenchymal expression program in lung epithelial cells by wingless protein (Wnt)/-catenin requires the presence of c-Jun N-terminal kinase-1 (JNK1).” American Journal of Respiratory Cell and Molecular Biology. 47(3), 306-314. DOI:10.1165/RCMB.2011-0297OC v Conference Proceedings † = talk, ‡ = poster ‡ “Swarming Migration of Co-attracting Mesenchymal Cells into Fractal-like Epithe- lial Clusters.” Biophysical Society 2017 Meeting (San Francisco, CA. Feb 2017) † “Swarming Migration of Co-attracting Mesenchymal Cells into Fractal-like Epithe- lial Clusters.” Brown University Pathobiology Retreat (Providence, RI. Sept 2017) ‡ “Tracking the Epithelial-Mesenchymal Transition in 3D Multicellular Clusters.” Gordon Research Conference on Directed Cell Migration (Galveston, TX. Jan 2017) † “The Epithelial-Mesenchymal Transition in Engineered Microenvironments.” Ma- terials Research Conference Fall 2015 Meeting (Boston, MA. Dec 2015) ‡ “The Role of S-Glutathionylation in Epithelial to Mesenchymal Transition.” Stu- dent Research Conference, University of Vermont (Burlington, VT. May 2013) Teaching Experience Brown University Teaching Assistant, BIOL 1310/2310: Developmental Biology, (Fall 2017) Occasional Lecturer, ENGN 2910S: Cancer Nanotechnology (Spring 2015, Spring 2016) University of Vermont Teaching Assistant, CHEM 042: Organic Chemistry I (Fall 2011, Summer 2011) Additional Research Experience Brown University Advisor: John M. Sedivy, Ph.D. (Department of Molecular and Cellular Biology) Research Rotation (Jan 2014 - Mar 2014), “The Role of cMyc in Autophagy” Advisor: Olin Liang, Ph.D. (Department of Orthopedics, Rhode Island Hospital) Research Rotation (Oct 2013 - Jan 2014), “In Vitro Chondrogenesis of WT and SHIP-KO MSCs” vi Acknowledgements To Ian Wong, thank you for being an outstanding mentor over the past five years. Our many research discussions have been eye opening, and I can confidently say that your mentorship has allowed me to mature into a resilient academic scientist. Thank you for your continued feedback and support of my ideas, which has allowed us to pursue research avenues we did not anticipate. On behalf of all the minons I have overseen, thank you for going above and beyond for your students. It is very apparent that you genuinely care about helping your students meet their long-term goals, and that unique quality is something I aspire to adopt as an independent investigator one day. I look forward to following your work in the future and will always be grateful that you were my advisor. To my committee members- Agnes Kane, Eric Darling, Alexander Brodsky, and Andrea McClatchey. Thank you for all of your support and feedback throughout my graduate career. Your questions and our discussions have encouraged me to put my research into perspective and reflect on its meaning in a broader context. This way of thinking will continue to guide me throughout my career. To the wonderful faculty and staff members that have supported me over the years, thank you for all that you do. Notably, thank you Patricia Capece for your assistance with all our lab affairs, thank you Michele Welindt for making every Pathobiology event memorable, and Maria for all of your care and late night conversations in Arnold. Thank you to Richard Bennett and Jonathan Reichner for directing the Pathobiology program and encouraging all of your students. A special thank you to Jonathan Reichner for your constant support, research feedback, and concern for the well being of your students. To all lab mates, past and present, thank you for your friendship and scientific discussions. Thank you to all of my co-authors, who have provided invaluable support for my research projects- I appreciate all of your patience and hard work. A special vii thank you to Marielena Gamboa Castro, Evelyn Kendall Williams, and Jea Yun Sim- your dedication to the Wong lab surpassed that of most graduate students, even though you all started as undergraduates. As the first graduate student in the lab, I struggled to manage everything by myself. You all recognized this and went above and beyond to help the lab run smoothly, while also providing invaluable friendship. Lena- I remember when you first joined the lab you were following a pre-med track, but I recognized early on that your talents as a researcher would take you far. I am proud to have mentored you and am excited to see where your research will go- I look forward to our potential collaborations in the future. Thank you for being a great friend and colleague! To my family and friends, thank you for keeping me sane and providing a dis- traction from my research when needed. To my parents- thank you for your support and encouragement throughout my education. Although the sciences are outside of your wheelhouse, you taught me to be dedicated to my work and hold myself to a high standard, which has driven me to conduct research with the highest integrity. To my late grandfather, I greatly appreciate all of our past conversations and will keep your memory with me throughout my life. You have passed down a great deal of wisdom and a creative way of thinking, which I see in my father and now myself. To Michael Gagen, thank you for support and comfort over the past several years. I appreciate your patience for spending your weekends with me even though they are often intertwined with my work. Lastly, I would like to acknowledge all of the funding and resources that made my research possible. Thank you to D.A. Haber for the inducible MCF-10A cell lines, R. J. Giedt and R. Weissleder for the MDA-MB-231 GFP-H2B cell line. This work was supported by NIH Grants T32ES007272, P30GM110759, R21CA212932, and Start-Up Funds from Brown University. viii Contents Signature Page . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii CV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv 1 Introduction 1 1.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Multicellular Tumor Invasion and the ECM . . . . . . . . . . . . . . 3 1.2.1 From Basement Membrane to Interstitial Matrix . . . . . . . . 5 1.2.2 Integrins and other Cell-Matrix Adhesions . . . . . . . . . . . 7 1.3 Established Biomaterials for Tumor Invasion . . . . . . . . . . . . . . 10 1.4 Spreading Out: 3D Multicellular Spheroids on Planar Surfaces . . . . 19 1.5 Conspiring Across Borders: Co-Culture of Tumor and Stroma . . . . 32 1.6 Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 1.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 2 Wrinkled, wavelength-tunable graphene-based surface topographies for directing cell alignment and morphology 44 2.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 2.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 ix 2.3 Experimental . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 2.3.1 Fabrication of textured surfaces . . . . . . . . . . . . . . . . . 48 2.3.2 Morphology characterization . . . . . . . . . . . . . . . . . . . 48 2.3.3 Cell culture . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 2.3.4 Preparation of rGO-coated substrates for cell culture experiments 49 2.3.5 Cell viability . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 2.3.6 Immunostaining and fluorescence imaging . . . . . . . . . . . 50 2.3.7 Image processing for quantification of cell morphology . . . . . 51 2.3.8 Cell orientation and statistical analysis . . . . . . . . . . . . . 52 2.4 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 53 2.4.1 Film fabrication and structure . . . . . . . . . . . . . . . . . . 53 2.4.2 Thermal reduction for liquid phase stability . . . . . . . . . . 59 2.4.3 Cell alignment on flat and wrinkled rGO substrates . . . . . . 61 2.4.4 Cell morphology on flat and wrinkled rGO substrates. . . . . . 62 2.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 2.6 Supplementary Information . . . . . . . . . . . . . . . . . . . . . . . 68 3 Morphological Single Cell Profiling of the Epithelial-Mesenchymal Transition 74 3.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 3.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 3.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 3.3.1 Cell Culture . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 3.3.2 EMT Induction with Snail-1 or TGF-β1 . . . . . . . . . . . . 78 3.3.3 Density Dependent Induction of EMT . . . . . . . . . . . . . 79 3.3.4 Drug Treatment with Taxol for EMT Induction . . . . . . . . 79 3.3.5 Immunostaining and Fluorescent Imaging . . . . . . . . . . . . 79 3.3.6 Image Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 80 x 3.3.7 Phenotype Classification using a Gaussian Mixture Model (GMM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 3.4.1 Profiling Epithelial and Mesenchymal Phenotypes after Snail Induction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 3.4.2 Snail Induction Drives Rapid EMT over 72 h . . . . . . . . . . 86 3.4.3 TGF-β1 Induction Drives Gradual EMT over 72 h . . . . . . . 87 3.4.4 MCF-10A Cells Exhibit Plasticity and Undergo EMT in Sub- confluent Cultures . . . . . . . . . . . . . . . . . . . . . . . . 90 3.4.5 Sublethal Taxol Enhances EMT in Uninduced and Induced Snail Populations . . . . . . . . . . . . . . . . . . . . . . . . . 92 3.5 Discussion and Conclusions . . . . . . . . . . . . . . . . . . . . . . . 94 3.6 Supplementary Information . . . . . . . . . . . . . . . . . . . . . . . 102 3.6.1 Supplementary Methods . . . . . . . . . . . . . . . . . . . . . 102 3.6.2 Supplementary Figures . . . . . . . . . . . . . . . . . . . . . . 111 4 Motility-Limited Aggregation of Mammary Epithelial Cells into Fractal-like Clusters 134 4.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134 4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136 4.2.1 Cell Proliferation and Motility are Arrested by Reduced Growth Factor Conditions . . . . . . . . . . . . . . . . . . . . . . . . . 137 4.2.2 Dispersed Individuals Aggregate into Multicellular Clusters with Fractal-like Morphology . . . . . . . . . . . . . . . . . . 140 4.2.3 Transient Collective Migration Connects Clusters into Span- ning Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . 142 4.2.4 Clustering and Jamming are Governed by Local Cell Density and EGF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146 xi 4.3 Discussion and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . 147 4.4 Supplementary Information . . . . . . . . . . . . . . . . . . . . . . . 151 4.4.1 Supplementary Methods . . . . . . . . . . . . . . . . . . . . . 151 4.4.2 Supplementary Figures . . . . . . . . . . . . . . . . . . . . . . 153 5 3D Organoid Disorganization after the Epithelial-Mesenchymal Transition 154 5.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154 5.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155 5.3 Discussion and Future Directions . . . . . . . . . . . . . . . . . . . . 162 5.3.1 Chapter 2 Summary . . . . . . . . . . . . . . . . . . . . . . . 163 5.3.2 Chapter 3 Summary . . . . . . . . . . . . . . . . . . . . . . . 164 5.3.3 Chapter 4 Summary . . . . . . . . . . . . . . . . . . . . . . . 166 5.3.4 Chapter 5 Summary . . . . . . . . . . . . . . . . . . . . . . . 168 Bibliography 172 xii List of Tables 2.1 Data Summary for Fibroblast Morphology Characterization. . . . . . 73 3.1 Flow chart of CellProfiler analysis pipeline: Part 1-2 . . . . . . . . . . 112 3.2 Flow chart of CellProfiler analysis pipeline: Part 3-4. . . . . . . . . . 113 xiii List of Figures 1.1 Multicellular tumor invasion and the ECM. . . . . . . . . . . . . . . . 5 1.2 Cellular and molecular features of EMT. . . . . . . . . . . . . . . . . 8 1.3 2D Cell Monolayers in Microfabricated Geometries. . . . . . . . . . . 16 1.4 Multicellular spheroids spreading on planar surfaces. . . . . . . . . . 22 1.5 Multicellular aggregates spreading on collagen I substrates. . . . . . . 24 1.6 Epithelial morphogenesis and invasion in natural 3D hydrogels. . . . . 26 1.7 Epithelial morphogenesis and invasion in synthetic 3D hydrogels. . . . 30 1.8 Co-culture models of tumor cells and stromal cells. . . . . . . . . . . 34 2.1 Fabrication process and morphology characterization of wrinkled graphene-based multilayer films. . . . . . . . . . . . . . . . . . . . . . 55 2.2 Wavelength tunability through control of multilayer GO film and through substrate selection. . . . . . . . . . . . . . . . . . . . . . . . 58 2.3 Pre-strain provides effective control over wrinkle feature height. . . . 60 2.4 Fibroblast culture on wrinkled graphene materials results in highly aligned cells. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 2.5 Image analysis of fibroblasts on graphene-based materials demonstrates distinct morphological features for different cell types on wrinkled vs. flat substrates. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 xiv 2.6 Dependence of wavelength λ normalized by small-strain wavelength λ0 and amplitude A normalized by film thickness hf as a function of pre-strain. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 2.7 Tilt-view SEM images showing the average height of the GO wrinkles. 68 2.8 Top-down and tilt-view SEM images of wrinkled GO films. . . . . . . 69 2.9 FT-IR of GO film and stabilized GO films prepared by thermal treat- ment at 120◦ C overnight. . . . . . . . . . . . . . . . . . . . . . . . . . 69 2.10 SEM images of wrinkled s-GO. . . . . . . . . . . . . . . . . . . . . . 70 2.11 Thermal reduction preior to relaxtion and wrinkling yields less regular wrinkles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 2.12 NIH-3T3 cells cultured on s-GO materials are highly viable. . . . . . 71 2.13 NHF cells cultured on s-GO materials are highly viable. . . . . . . . . 72 3.1 Classification of epithelial and mesenchymal phenotypes after 72 h DMSO (control) and OHT (Induced Snail-1), followed by 72 h DMSO. 97 3.2 Time course measurements of cell morphology and biomarker expres- sion with Snail-1 induction (OHT). . . . . . . . . . . . . . . . . . . . 98 3.3 Time course measurements of cell morphology and biomarker expres- sion with TGF-β1 induction. . . . . . . . . . . . . . . . . . . . . . . . 99 3.4 Changes in cell morphology and biomarker expression with cell density. 100 3.5 Changes in cell morphology and biomarker expression with sublethal Taxol treatment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 3.6 Schematic of Experimental Conditions for GMM Training Set. . . . . 111 3.7 Histograms of all Nuclear, Vimentin, and Cytoplasmic metrics for Ep- ithelial and Mesenchymal Training Sets. . . . . . . . . . . . . . . . . 114 3.8 Histograms of all Nuclear, Vimentin, and Cytoplasmic metrics for Ep- ithelial and Mesenchymal Training Sets. . . . . . . . . . . . . . . . . 115 xv 3.9 Histograms of Vimentin Intensity metrics for Epithelial and Mesenchy- mal Training Sets. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 3.10 Boxplot of selected metrics used to distinguish Training and Test Ep- ithelial and Mesenchymal conditions. . . . . . . . . . . . . . . . . . . 117 3.11 Predicted vs. Actual Segmentation for GMM Training and Test sets. 118 3.12 Predicted vs. Actual Segmentation for GMM of T-47D and MDA- MB231 cells. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 3.13 Optimal combinations of Nuclear, Vimentin, and Cytoplasmic metrics for Cell Classification. . . . . . . . . . . . . . . . . . . . . . . . . . . 120 3.14 Schematic of experimental conditions and results for OHT induction. 121 3.15 Boxplot of selected metrics with duration of OHT induction. . . . . . 122 3.16 Posterior Probabilities of Gaussian Mixture Model for OHT induction. 123 3.17 Schematic of experimental conditions and results for TGF-β1 induction.124 3.18 Boxplot of selected metrics with duration of TGF-β1 induction. . . . 125 3.19 Posterior Probabilities of Gaussian Mixture Model for TGF-β1 induction.126 3.20 Schematic of experimental conditions and results for Density experi- ments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 3.21 Boxplot of selected metrics with Density experiments. . . . . . . . . . 128 3.22 Posterior Probabilities of Gaussian Mixture Model with Density. . . . 129 3.23 Schematic of experimental conditions and results for Taxol treatment. 130 3.24 Boxplot of selected metrics with Taxol Treatment. . . . . . . . . . . . 131 3.25 Live/Dead Staining and Percent Viability after Taxol treatment. . . . 132 3.26 Posterior Probabilities of Gaussian Mixture Model with Taxol treatment.133 4.1 Mammary epithelial cells cluster under reduced growth factor conditions.138 4.2 Dynamics of cluster aggregation in low EGF. . . . . . . . . . . . . . . 140 4.3 Emergence of collective migration behaviors in low EGF. . . . . . . . 145 xvi 4.4 Cell behavior as a function of EGF concentration reveals a jamming- like phase diagram. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148 4.5 Fluorescence microscopy images of cells clustering under low EGF con- ditions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 5.1 Preparation and Characterization of SF/COL Hydrogels for Cell Cul- ture in 3D. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156 5.2 3D Epithelial Organization is Disrupted with the Induction of the Epithelial-Mesenchymal Transition. Representative confocal z-slice im- ages of human mammary epithelial cells with cytoplasmic GFP and nuclear RFP expression. . . . . . . . . . . . . . . . . . . . . . . . . . 158 5.3 Morphological Profiling of EMT in 3D. . . . . . . . . . . . . . . . . . 159 5.4 Automated single cell tracking reveals dynamic migration behaviors during EMT. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 xvii Chapter 1 Introduction Chapter 1 has been previously published as a review article: S.E. Leggett, A.S. Khoo, and I.Y. Wong. “Multicellular Tumor Invasion and Plas- ticity in Biomimetic Materials.” Biomaterials Science. 5, 1460-1479, May 2017 1.1 Abstract Tumor invasion and metastasis occurs in the context of molecular and mechanical cues from the extracellular matrix (ECM), and ultimately results in over 90% of cancer- related fatalities.[1] In particular, the detachment and dissemination of individual cells from the periphery of carcinomas is reminiscent of the epithelial-mesenchymal transi- tion (EMT) in embryonic development and wound healing.[2] Moreover, tumor cells can exhibit multicellular collective invasion as loosely or tightly coordinated groups.[3] Such malignant tumor progression is accompanied by a dramatic remodeling and stiff- ening of the ECM (i.e. desmoplasia).[4] These microenvironmental changes can bias tumor cells towards individual or collective invasion, a phenomenon known as phe- notypic plasticity. This reciprocity between tumor invasion and ECM has previously 1 been investigated in cancer research using 2D monolayer culture, animal models, and patient histology. Biomimetic materials can complement these existing approaches by recapitulating important features of the tumor microenvironment.[5] Biomaterials can be fabricated or synthesized into microstructural architectures that mimic the ECM. For instance, semiconductor fabrication techniques have en- abled exquisitely detailed geometric features, comparable in size and spacing to matrix topography.[6] Moreover, hydrogels based on hydrated, crosslinked polymer networks can mimic the biochemical composition, mechanical stiffness and degradability of ECM in tissues and tumors.[7] Indeed, natural hydrogels have enabled key biolog- ical insights into EMT and tumor invasion. For instance, Hay and colleagues first observed what they termed an epithelial-to-mesenchymal transformation after em- bedding corneal epithelial tissue in a reconstituted matrix of fibrillar collagen I. [8] Subsequently, Bissell and colleagues showed that epithelial morphogenesis can be re- capitulated using reconstituted basement membrane in a 3D context, resulting in the self-organization of individual cells into gland-like structures with hollow lumens and differentiated cell-cell junctions (acini).[9, 10] The disorganization and dissemination of cells from these acini into the surrounding ECM recapitulates many key features of tumor progression.[11] Recent advances in biomaterials have enabled new physical insights into the tumor microenvironment[12] and may facilitate preclinical models of cancer with greater physiological relevance.[13] In this review, we highlight recent developments in cancer cell invasion and EMT enabled by new biomaterial platforms. We focus on multicellular tissues as in vitro and ex vivo models of cancer, with emphasis on results published within the last sev- eral years. In the following sections, we consider 1) microfabricated geometries that promote EMT from 2D monolayers, 2) the spreading of 3D multicellular aggregates onto planar surfaces, 3) epithelial morphogenesis and dissemination in 3D biomateri- 2 als, and 4) co-culture of tumor and stromal cells. We conclude with a discussion of future directions for the field. 1.2 Multicellular Tumor Invasion and the ECM Cancer can be defined as a disease in which “abnormal cells proliferate in an un- controlled fashion and spread throughout the body.”[14] Most human cancers are carcinomas derived from epithelial tissues, which line the walls of surfaces and cavi- ties of human organs. The most common types of cancers arise in the skin, prostate, breast, lung and colon.[15] Nevertheless, many types of skin cancer can be diagnosed early and treated. Thus, cancer-related fatalities are most frequently due to lung, colorectal, breast, prostate, and pancreatic cancers. [15] Tumor progression may be largely influenced by tissue-specific physiology. For instance, the functional unit of the breast is the lobule, which is lined by luminal and myoepithelial cells interfaced with intralobular stroma, further surrounded by interlobular stroma.[16] These stroma include extensive adipocytes (fat cells), as well as fibroblasts, immune cells, stem cells, and endothelial cells. Breast carcinomas are often classified as ER (estrogen receptor) positive, HER2 (human epidermal growth factor receptor 2) positive, or triple negative. In contrast, the functional unit of the lung is the alveolus, which is lined by alveolar epithelium that consists of flat- tened platelike type I pneumocytes and rounded type II pnuemocytes.[16] Alveolar macrophages are associated with these epithelial cells, as well as small numbers of fibroblast-like cells, smooth muscles cells, mast cells, and lymphocytes. Carcinomas of the lung are histologically classified as adenocarcinomas, squamous cell carcinomas, small cell carcinomas and large carcinomas. It should be noted that the lung is also the most common site for secondary metastases, which may be partially explained by hematogenous spread and arrest of cancer cells in the capillary bed of the lung.[17] 3 Finally, the skin includes squamous epithelial cells (keratinocytes), melanocytes, den- dritic cells, and lymphocytes. Common skin cancers include melanomas, sqamous cell carcinomas and basal cell carcinomas.[16] The details of tissue-specific pathophysiol- ogy are reviewed extensively elsewhere.[16] Benam et al. have also recently reveied tissue-specific in vitro cancer models.[18] In general, epithelial tissues transition to invasive carcinomas through multicel- lular disorganization and dissemination (Figure 1.1).[11] As this malignant progres- sion occurs, cells encounter distinct ECM with dramatically different biochemical and physical properties. Furthermore, tumor cells can actively deposit and remodel ECM, which can further alter the extracellular cues that govern molecular and mechanical phenotype.[4] In general, tumor cells display extraordinary phenotypic heterogeneity and plasticity in response to these dynamic microenvironmental conditions, enabling them to migrate with a variety of distinct collective and individual phenotypes.[1] In particular, various aberrant stimuli may trigger an EMT, resulting in a weakening of cell-cell adhesions, as well as enhanced motility and cell-matrix adhesion.[2] Overall, tumor cells are highly efficient in invasion processes for infiltration of the ECM, and subsequent intravasation to enter the bloodstream. However, successful extravasation, or exit from circulation to a secondary site, occurs with very low fre- quency due to the harsh nature of the bloodstream (e.g. mechanical shear stress) and susceptibility to immune cell surveillance.[19] Expression of the T-cell marker, CD44, on tumor cells may favor extravasation by enhancing the adhesive interac- tions between cancer cells and the endothelium. Subsequently, the colonization of secondary metastatic sites often depends on the permissiveness of the microenviron- ment. As a consequence, primary tumors originating from certain tissues often prefer- entially metastasize to other tissues, a phenomenon known as organ tropism.[17] For instance, primary breast tumors primarily metastasize to bone, lungs, liver, and brain, while primary lung tumors primarily metastasize to brain, bones, adrenal gland, and 4 liver. Indeed, the primary tumor may prime a metastatic niche for colonization by reprogramming stromal cells to structurally rearrange the ECM, recently reviewed elsewhere.[20] Although not a primary focus of this review, it should be noted that metastatic homing could be investigated by biomimetic materials that recapitulate the microenvironment of metastatic sites.[21] Fibrotic ECM (Aligned Collagen I) Epithelial Acini Partial EMT? Mesenchymal cell Basement Membrane Interstitial ECM (Collagen IV, Laminin) (Collagen I, Fibronectin, Proteoglycans, Tenasin C) Figure 1.1: Multicellular tumor invasion and the ECM. Epithelial tissues are enclosed by basement membrane. During tumor invasion and EMT, tissues disorganize and disseminate into the surrounding interstitial matrix. Tumor cells can further deposit and reorganize the matrix into a more fibrotic archi- tecture. 1.2.1 From Basement Membrane to Interstitial Matrix Under physiological conditions, the ECM presents mechanical and biochemical cues that are instructive for cellular function and the maintenance of tissue architecture.[4] The ECM consists of two major subtypes known as the basement membrane and inter- stitial matrix. Epithelium and endothelium that support biofluid adsorption and se- cretion are adherent to the basement membrane (Figure 1.1). [22] This thin, mesh-like network provides a structural framework for maintaining epithelial and endothelial tissue architecture, as well as a barrier to adjacent connective tissue. The basement membrane consists of specialized ECM proteins including collagen IV and glycopro- teins such as laminin, nidogen, and perlecan. In particular, laminin contains multiple 5 domains that mediate the binding interactions between basement membrane com- ponents and present sites for cell adhesion.[23] Importantly, laminin plays a crucial role in instructing epithelial phenotype.[24] Epithelial cells on the basement mem- brane typically exhibit a highly geometric, cobblestone-like morphology with tight junctions, which enables control over tissue barrier properties such as permeability and secretion. The adjacent connective tissue (stroma) contains interstitial matrix, which is a three-dimensional hydrogel composed of fibrillar collagens, fibronectin, tenascin-C, elastin, glycosaminoglycans, proteoglycans, and others. The exact composition of interstitial matrix varies with tissue type and often undergoes diverse changes during carcinoma progression (Figure 1.1).[25] The protein components that comprise the ECM have been recently characterized in terms of a “core matrisome,” which in- cludes the structural ECM components described previously, as well as “matrisome- associated” proteins, which include ECM-affiliated proteins (galectins, semaphorins, etc.), secreted proteins (growth factors, cytokines, etc.), and ECM regulators (matrix metalloproteinases, cross-linking enzymes), etc.[26] Together, these ECM constituents provide sites for cell adhesion and migration, and grant essential mechanical proper- ties to the tissue including tensile strength, elasticity, and resistance to compressive forces. Typically, brain tissue is one of the softest tissues (E ∼ 0.1 kPa), while breast and liver are slightly stiffer (E ∼ 0.4 − 1 kPa).[27] Cartilage and bone tend to be relatively stiff (E ∼ 10 − 20 kPa).[27] Interestingly, tissue stiffness scales roughly with collagen I concentration.[28] During tumor progression, the ECM is dynamically remodeled to a fibrotic-like state, which has analogies with wound healing processes, described as “a wound that does not heal.”[29] Stromal cells such as fibroblasts can synthesize, reorganize and de- grade the ECM, resulting in a macroscopic stiffening of the stroma (desmoplasia).[30] In particular, there is often increased deposition, cross-linking and alignment of fib- 6 rillar proteins such as collagen I, resulting in an aberrant matrix topography and biomechanical properties that affect cellular phenotype (Figure 1.1). These aligned, track-like structures can facilitate the directional migration of tumor cells toward lym- phatic and blood vessels for metastatic spread.[31] Additionally, excess ECM degra- dation can result in the release of sequestered growth factors and other bioactive molecules which are bound to proteoglycans.[32] Finally, altered ECM can enhance the recruitment and infiltration of immune cells as well as induce angiogenesis, which is reviewed elsewhere.[4] 1.2.2 Integrins and other Cell-Matrix Adhesions Cell adhesion and migration along the ECM are primarily mediated by integrins, a family of transmembrane glycoprotein receptors (Figure 1.2). Each integrin con- sists of an alpha and beta subunit, with 18 and 8 types respectively, yielding at least 24 distinct heterodimers.[33] These diverse integrin heterodimers permit some specificity in ligand binding to direct cell-matrix interactions in a context dependent manner, although there also exists some promiscuity. In particular, epithelial cells utilize integrins α6 β4 and α3 β1 to attach to collagen IV and laminin in the basement membrane (Figure 1.2B). During the initial stages of carcinoma progression, tumor cells can locally degrade the basement membrane through MMP (matrix metallo- proteinase) and ADAM (a disintegrin and metalloproteinase) protease families.[34]. As a consequence, tumor cells can interact with the surrounding interstitial matrix through integrins α5 β1 and α1/2 β1 , which bind to collagen I and fibronectin (Figure 1.2C). Furthermore, matrix degradation can coincide with additional ECM deposition and crosslinking. In particular, lysyl oxidase expression leads to further collagen I crosslinking and subsequent clustering of β1 integrin, which further drives cell survival via the PI3K pathway and invasion through focal adhesion kinases.[30] In addition to integrin binding, collagen also interacts with receptor Tyr kinases discoidin domain- 7 containing receptors 1 and 2 (DDR1/2), which have been shown to be critical for collective cell migration and stimulate the induction of EMT, respectively.[32] A Collagen IV Laminin Collagen I Fibronectin B C MMPs ,β) DDR rin (α 6 tin 5 1/2 E-cad 4 en 1 herin Integ m 5 Vi 3 1 1 D Epithelial Mesenchymal low motility, strong cell-cell high motility, cell-matrix adhesions, E-cadherin interactions, vimentin, MMP secretion Figure 1.2: Cellular and molecular features of EMT. (A) Epithelial and mesenchymal cells in ECM. (B) Epithelial cell polarity is governed by cell-cell junctions and integrins. (C) Mesenchymal cell polarity is governed by inte- grins and DDR, but modulated by matrix metalloproteinases (MMPs) and vimentin. (D) Phenotypic features of EMT. EMT Pathways Features of EMT have been observed in patient tumor histology and are often associated with poor prognosis.[35] These include a loss of tissue architecture with dedifferentiated phenotypes, diminished expression of E-cadherin, and invasive protrusions.[36] Cell-ECM interactions may lead to the expression of EMT regulatory genes, and reciprocally, several transcription factors that regulate the induction of EMT are also involved in ECM remodeling. For instance, ECM can participate in the initiation of signal transduction pathways by serving as a reservoir for soluble signals (e.g. growth factors). As an example, transforming growth factor beta (TGFβ) 8 binds to glycoproteins and proteoglycans in the ECM, fibrillin and heparin sulfate, respectively, and can be mobilized in response to cleavage of the complex[37] to drive downstream signaling including the induction of EMT.[38] Snail, Zeb, and Twist are prominent transcription factors, or master regulatory genes, involved in EMT induction.[39, 40] A common theme is the downregulation of E-cadherin to destabilize adherens junctions between cells. In addition, apical tight junctions and desmosomes are weakened by the repression of genes that en- code claudin and occludin, as well as desmoplakin and plakophilin. Moreover, these transcription factors suppress the synthesis of basement membrane components.[41] Overall, these changes in gene expression disrupt epithelial tissue architecture and barrier function, while impeding the formation of new junctions. An additional con- sequence of the loss of E-cadherin is the increased expression of N-cadherin, which is typically associated with mesenchymal cells. Further downstream signaling can occur via the crosstalk of TGF-β and matrix ligand adhesion, particularly PI3K-AKT, ERK MAPK, p38 and JNK pathways, which has recently been reviewed elsewhere.[42] At the cellular level, EMT is associated with dramatic alterations in the cytoskele- ton and morphology (Figure 1.2D).[43] Initially epithelial cells lose apicobasal polar- ization and gain front-back polarization, driven in part by Rho GTPases that regulate actin dynamics.[44] In particular, RAC1 and CDC42 activate actin polymerization and membrane protrusion formation. Simultaneously, RHOA displays a coordinated localization to the rear in order to facilitate actomyosin contractility for cell retraction. As a consequence, mesenchymal cell lines can display highly asymmetric tractions at the front and rear relative to epithelial cell lines.[45, 46] Moreover, EMT results in dramatic increases in actin stress fiber formation as well as actin-rich invadopodia that can direct local proteolytic cleavage of the matrix through the release of MMPs at the leading edge.[32] Finally, the composition of intermediate filaments is also switched from cytokeratin to vimentin, which enhances cell deformability.[47] 9 1.3 Established Biomaterials for Tumor Invasion Collagen I Collagen I is a fibrillar biopolymer constituted from natural sources (typically rodent tail or bovine dermis) and was first utilized as a transparent hydrogel substrate by Ehrmann and Gey.[48] Since collagen is maintained in its monomeric form in acidic solution and at low temperature,[49] controlled collagen polymerization into a hydro- gel can occur by increasing the pH and incubation in the presence of cells and culture media. Qualitatively, lower collagen concentrations result in lower fibril density and larger mesh size.[50] Moreover, lower polymerization temperatures can result in larger mesh sizes and larger fibril diameters.[51] Nevertheless, collagen hydrogels display a complicated dependence on mechanical stiffness and ligand density, making it difficult to decouple these mechanisms in epithelial cell processes. Typically, the elastic mod- uli for collagen gels can vary from 0.1 - 2 kPa for concentrations of rat-tail collagen from 2.0 - 4.0 mg/mL,[52] although these values are quite sensitive to preparation conditions.[53] Moreover, collagen hydrogels can be extensively degraded and remod- eled, which can be advantageous for studies of directed cell migration but may also limit long term stability.[54] Finally, collagen as a naturally-derived biomaterial can display some batch-to-batch variability. Overall, collagen I represents a reasonable mimic of the tumor ECM and has been utilized extensively to investigate EMT ever since the initial experiments by Greenburg and Hay.[8] Reconstituted Basement Membrane (Matrigel) Reconstituted Basement Membrane (rBM) was first extracted from mouse Engelbreth- Holm-Swarm (EHS) sarcoma tumors by Swarm and colleagues[55], which is commer- cially available as Matrigel. Major components of rBM typically include laminin, collagen IV, entactin, as well as proteoglycans such as heparan sulfate.[56] In 10 particular, the chief constituent of rBM, laminin, supports the differentiation and polarization of epithelial and endothelial cells and facilitates their attachment to the basement membrane. Furthermore, it is worth noting that there are 11 distinct chains that make up the laminin heterotrimer of α, β, and γ subunits, respectively. In particular, the specific isoform of laminin derived from EHS (laminin α1 β1 γ1 ) is critical in embryogenesis and development, however, adult epithelial tissues and carcinomas predominantly express laminin α3 and α5 isoforms.[23, 57, 58] rBM is liquid near freezing temperatures, but solidifies into a hydrogel at physiological tem- peratures. Bissell and coworkers pioneered the use of rBM to investigate mammary epithelial morphogenesis.[9, 10] By embedding cells within rBM, or seeding cells on top of rBM with a dilute overlay, epithelial cells self-organized into spherical acini with apicobasal polarity, hollow lumens, and tightly controlled growth and proliferation.[59, 11] Weaver et al. subsequently used rBM to investigate the role of integrins β1 and β4 in driving a polarized 3D epithelial architecture, as well as reversion of a malignant phenotype.[60, 61] Thin rBM coatings on a porous plastic membrane have also been utilized in a classical invasion assay[62] based on the Boyden chamber (Transwell assay)[63] This membrane partitions two fluidic com- partments, with migratory cells plated at the top and an optional chemoattractant solution underneath. Over time, migratory cells translocate across the membrane to the bottom, where they can be counted. Although established, this assay is limited in that it occurs as an endpoint and also biases towards individual cell migration, due to the relatively small pore size. Transitions to invasion have been observed by combining rBM with fibrillar col- lagen I. For instance, Paszek et al. showed that mammary acini initially seeded on relatively soft rBM ( ∼0.18 kPa) transitioned towards disorganized structures with destabilized cell-cell junctions when overlaid with collagen I with progressively increasing stiffness.[64] Similarly, Guzman et al. have demonstrated that blending 11 fibrillar collagen I into rBM can promote individual cell invasion.[65] It should be noted that rBM displays significant batch-to-batch variability in biochemical compo- sition and mechanical properties[56] and these biomaterial properties of rBM cannot be easily tuned for systematic studies. Nevertheless, rBM has been crucial in driving biological insights underlying epithelial morphogenesis and EMT, and remains widely used. Polyacrylamide Polyacrylamide (PA) is a versatile hydrogel with tunable mechanical and biochemical properties, first demonstrated as a biomaterial substrate by Pelham and Wang.[66] PA is often polymerized as a thin film coating by the reaction of acrylamide monomer and bis-acrylamide crosslinker.[67] The elastic modulus of this hydrogel can be systemat- ically varied from 0.1-100 kPa by changing the relative concentrations of monomer and cross-linker, permitting well-controlled mechanical properties.[68] It should be noted that PA lacks adhesion ligands and is largely bioinert, which deters cell attach- ment. To address this issue, ECM proteins such as collagen I, collagen IV, laminin, or fibronectin can be conjugated to PA through a bifunctional crosslinker such as sulfo-SANPAH.[68] The surface ligand density can be controlled via the ligand con- centration in solution, independently of mechanical properties. PA substrates have found great success in mechanobiology, particularly when fluorescent tracer particles are added for traction force microscopy.[69] Nevertheless, PA is limited in that the hydrogel precursors are highly cytotoxic, so it cannot be used to encapsulate cells in a fully 3D microenvironment. Moreover, PA cannot be degraded by cells. As a consequence, PA is primarily used as a “soft” 2D biomaterial substrate. Hybrid ap- proaches have also been demonstrated where cells adherent to PA are overlaid with a second hydrogel. 12 Polydimethylsiloxane Polydimethylsiloxane (PDMS) is a moderately soft elastomer that can be micro- molded with submicron fidelity, first demonstrated as a deformable cell substrate by Harris and coworkers.[70] PDMS is relatively straightforward to prepare, and displays ideal material properties for soft lithography approaches based on replica molding against a microfabricated silicon master.[71] In particular, PDMS is highly confor- mal and can adhere reversibly or irreversibly to other surfaces. PDMS can be used in microcontact printing as a topographically patterned “stamp” which can transfer molecular “ink” to a new surface.[72] This gentle transfer is particularly useful for microscale patterning of soft substrates, such as the PA hydrogels discussed in the previous section.[73] The irreversible bonding of PDMS microstructures to a glass sub- strate is also useful to prepare confined geometries for manipulation of fluid flows[74] or cell migration.[75] Oxygen plasma treatment of PDMS can render it (temporar- ily) hydrophilic, which permits adequate protein physisorption for cell adhesion.[76] Although PDMS is advantageous for rapid prototyping, concerns have been raised over potentially adverse effects for long-term cell culture, including evaporation, ad- sorption of small hydrophobic molecules, and instability of surface treatments.[77] It should be noted that the formulation of PDMS typically utilized for rapid prototyping (Sylgard 184 at 10:1 base:curing agent) is still significantly stiffer than ECM or tissue (∼1 MPa). Instead, a higher ratio of base to curing agent (¿ 50:1) can result in softer PDMS (∼ 5 kPa), although there is some discrepancy in the reported values, likely due to the significant viscoelastic effects.[78, 79] Palchesko has recently demonstrated that a blend of two different PDMS types (Sylgard 184 and Sylgard 527) can result in a softer substrate of 5 kPa, mechanically tunable up to 1 MPa. [80] Overall, PDMS is highly effective for patterning microstructures with well controlled geometries, but may not mimic other physiochemical features of natural ECM. 13 On the Edge: 2D Cell Monolayers in Microfabri- cated Geometries Epithelial cell monolayers cultured on top of glass or polystyrene surfaces are widely used to study cancer cell biology. These substrate materials are six orders of mag- nitude stiffer than native ECM (∼GPa), presenting an asymmetrical mechanical cue that results in apicobasal polarization perpendicular to the surface. Interestingly, cells adjacent to empty space at the periphery of a monolayer can display a loss of apicobasal polarization that is analogous to EMT. This phenomenon is commonly used in “wound-healing” or scratch assays where an empty region is suddenly created by locally removing a part of the monolayer or some obstacle.[81] “Leader” cells can then establish a front-back polarization oriented towards the empty region (parallel to the surface). The activation of RhoGTPases RAC1 and CDC42 can trigger lamellipo- dial and filopodial protrusions at the leading edge through actin polymerization.[82] These leader cells typically retain some cell-cell junctions at the sides and rear, which can transmit mechanical signals that repolarize neighboring cells over longer length scales. Gilles et al. used a fluorescent reporter to show that mammary epithelial cells at the monolayer edge transiently express vimentin, consistent with an EMT.[83] Subsequently, vimentin is downregulated once the cells fully occupy the empty region and re-establish a continuous monolayer. Thus, cells at the periphery of an epithe- lial monolayer recapitulate some biological behaviors observed at the periphery of a tumor or tissue. Nelson et al. demonstrated the importance of the tissue periphery using a hy- brid fabrication approach that utilized both microfabrication and soft hydrogels.[84] Briefly, an elastomeric PDMS stamp was replicated from a silicon master and used to imprint microscale cavities of different shapes into a thin collagen hydrogel. Cells were seeded within the cavity, which was then capped with a second flat sheet of 14 collagen. For instance, epithelial cells in a square geometry treated with TGF-β dis- played vimentin and α-SMA only at the edges and corners, suggesting that geometric and mechanical cues mediated EMT induction.[85] It was further shown that EMT induction was mediated by intercellular transmission of cytoskeletal tension. Subse- quent work showed that a single tumor cell co-cultured with epithelial cells would display enhanced proliferation and invasion when located at the periphery relative to the interior.[86] These regions were associated with a geometric enhancement of me- chanical stress, which acts through RHOA and focal adhesion kinase (FAK) activity. Surface topography can also play a critical role in modulating cell shape and di- rected migration, a phenomenon known as contact guidance.[6] Curtis and Wilkinson demonstrated the early use of microfabrication techniques to pattern grooved sub- strates, along which fibroblasts would align and elongate.[90] In particular, the width of these grooves can be comparable to the aligned collagen fibrils in desmoplasmic ECM. Analogously, PDMS channels have been used to confine cells within a tube-like environment,[91] which mimics interstitial geometries or the aligned “microtracks” generated by invasive tumor cells. These approaches have primarily been used to investigate individual cell migration and have been recently reviewed elsewhere.[75] Overall, microfabrication techniques can thus permit excellent geometric control over the monolayer periphery, as well as to confine individual cells along defined tracks. Epithelial cell migration on microscale glass wires Yevick et al. characterized epithelial cell migration on microscale glass wires,[87] comparable in size to bundled collagen fibers.[6] Epithelial cells (MDCK) were seeded on a PDMS block, adjacent to protruding, fibronectin-coated glass wires with radii ranging from 2 µm to 85 µm (Figure 1.3A). Cells migrated collectively as a cohe- sive strand that encircled the wider glass wires (> 40µm). In contrast, single cells occasionally broke away individually on narrower wires (< 40µm). These individual 15 A B 10 µm PDMS Time (∆t = 1 h) glass wire 2 - 85 microns 50 µm C D Sox 2 Simulation PDMS stamp fibronectin 10 µm PA gel E F 12 h 18 h 24 h glass coverslip 600 μm 10 microns PDMS Vimentin, E-cadherin Figure 1.3: 2D Cell Monolayers in Microfabricated Geometries. (A) Microscale glass wires embedded in PDMS, (B) Individual cells can detach and scatter along glass wires from multicellular groups. Reproduced from [87] with per- mission from the National Academy of Sciences. (C) Microcontact printing using PDMS stamps to pattern fibronectin shapes on soft PA gels. (D) Cell monolayers display stem-like markers at the periphery and corners of these shapes. Reproduced from [88] with permission from Nature Publishing Group. (E) Cells undergoing con- fined migration within arrays of microscale PDMS pillars, with a glass ceiling. (F) Cells transition from collective to individual migration after EMT. Reproduced from [89] with permission from Nature Publishing Group. 16 cells displayed relatively rapid migration (up to 100 µm/h) with alternating elongated and rounded morphology, but frequently reversed course and rejoined the collectively migrating group (Figure 1.3B). Interestingly, for narrow wires, cells displayed actin stress fibers that were perpendicular to the wire axis and extended circumferentially around the wire, suggestive of mechanical connectivity. Moreover, the leading edge also exhibited a tensile actin cable, reminsicent of those utilized for closing epithelial wounds through a purse-string like mechanism.[92] An interesting possibility would be to examine whether these actin cables enhanced migration on a more compliant substrate. Moreover, these actin cables may exert an effective surface tension that restricts the detachment of individual cells along the wire. Cancer stem-like cells and multicellular geometries Lee et al. characterized the expression of stem-like biomarkers within multicellu- lar islands with varying geometric shape.[88] Soft lithography was used to transfer ECM proteins in specific geometric patterns onto hydrazine-modified PA hydrogels (Figure 1.3C). A murine melanoma cell line was cultured on these different shapes and immunostained for putative cancer stem-like cell markers (e.g. CD271, CD133, and ABCB5), as well as molecular markers of pluripotency and tumor initiation (e.g. Nestin, Nanog, Jarid1b, Oct4, Sox2, and Nanog). Strikingly, expression of these biomarkers occurred at the periphery, particularly at sharp corners and convex fea- tures with geometrically enhanced stress (Figure 1.3D). In order to maximize the number of cells located at an edge and minimize the number of cells at the interior, a narrow spiral shape was patterned. Gene expression profiling of cells on spirals rela- tive to uniform PA gels revealed that stem-like phenotypes were activated by integrin α5 β1 , MAPK and STAT signaling. A differentiated phenotype could be rescued by chemical inhibition of MAPK signaling (particularly through p38 and ERK), blocking integrin α5 β1 , as well as in- 17 hibition of JNK. These stem-like cells were also observed at the periphery when mul- ticellular aggregates were cultured in three-dimensional geometries, including PA mi- crowells, as well as encapsulation within PEG hydrogels. Subsequent wound-healing and Transwell assays confirmed the increased invasiveness of cells cultured on curved patterns. Moreover, stem-like cells from spirals displayed increased metastatic po- tential after tail vein injection into a murine model, with lung metastases and poor survival. Qualitatively similar expression of stem-like markers (e.g. CD133, CD144) was observed for various other murine and human cell lines. Overall, this work re- veals that the geometry of a curved edge can apply mechanical stimuli that affect cell polarization and activate a stem-like phenotype, consistent with previous work on wound healing assays.[83] It should be noted cells also experience asymmetric cell-cell interactions, which are strengthened towards the interior. Future work could examine the crosstalk between cell-cell and cell-matrix interactions in activating a stem-like phenotype and EMT. Transitions between collective and individual migration in confined micropillar arrays Wong et al. measured collective and individual migration phenotypes after EMT in confined micropillar arrays.[89] PDMS devices were fabricated with soft lithogra- phy, bonded to glass cover slips and coated with fibronectin. The resulting confined geometries consisted of a square array of micropillars with 10 µm pillar-to-pillar sep- aration, and 10 µm height, effectively restricting cells to single file migration with elongated phenotypes (Figure 1.3E). Remarkably, mammary epithelial cells that had undergone a controlled EMT through Snail induction migrated through the pillars led by individual cells, which broke away from a collectively advancing front (Figure 1.3F). Time-lapse fluorescence microscopy and automated tracking of cell nuclei re- vealed two distinct subpopulations with distinct migratory behaviors. In particular, 18 individually migrating cells traveled with faster and straighter trajectories relative to collectively migrating cells, which can be classified using a Gaussian mixture model. This reversion of induced mesenchymal cells to an epithelial phenotype is likely to occur due to differences in cell-cell contact within the micropillar array relative to the flat loading region behind it. This scenario has a physical analogy with the solidifi- cation of an undercooled binary mixture,[93] suggesting a competition of phenotypic interconversion and “sorting out” at the interfacial front.[94] One limiting case of this model occurs when the two species are effectively insoluble, and sort out with minimal interconversion, which is well known from classical “differential adhesion” experiments by Steinberg and others.[95] Indeed, differences in effective cell “sur- face tension” were proposed to arise from differences in cadherin expression, which is qualitatively consistent with EMT. More generally, these microfabrication tech- niques permit exquisite control of geometric features in order to define cell-cell and cell-matrix interactions. When combined with single cell tracking techniques, this approach enables the identification of rare cells and exceptional phenotypes that are associated with EMT. 1.4 Spreading Out: 3D Multicellular Spheroids on Planar Surfaces Multicellular spheroids can be prepared through the aggregation of single cells under low matrix adhesion conditions.[96] For instance, dispersed cells in solution can be continuously agitated,[97] sedimented onto low adhesion surfaces,[98] or confined within an air-liquid interface (hanging drops).[99] Essentially, these various approaches enhance cell-cell adhesion while minimizing cell-matrix adhesion. The formation of multicellular spheroids by aggregation represents a facile approach for preparing tissue-scale constructs consisting of hundreds to thousands of cells. More- 19 over, multicellular spheroids recapitulate many of the hallmarks of cancer observed in vivo, including heterogeneity, necrotic microregions, and differential responses to therapeutic treatment.[100] Remarkably, the aggregation of multiple dispersed cell types can display self- organizing behaviors analogous to tissue development.[94] Early work by Steinberg showed that mixtures of embryonic cells appeared to “sort out” based on differences in cell-cell adhesion, which he termed the “differential adhesion hypothesis.”[101] In particular, Steinberg proposed that cell sorting could be explained as the demixing of two liquids with different surface tensions. For instance, if two cell types had stronger homotypic adhesions than heterotypic adhesion, they were likely to actively self-segregate into subpopulations of like type. Instead, if these cell types had het- erotypic adhesions that were comparable or stronger than their homotypic adhesions, they would remain randomly dispersed. Finally, if the first cell type displayed stronger homotypic adhesions than the second, the first cell type would be segregated into the interior while the second type would be segregated to the periphery. Interestingly, these adhesions are dependent (in part) on cadherin expression, so that cells with high cadherin expression would be expected to sorted to the interior while cells with low cadherin expression would be sorted to the exterior.[102] This is qualitatively consis- tent with some tumor architectures, particularly since EMT at the periphery results in decreased cadherin expression.[36] Nevertheless, Pawlizak et al. have reported that spheroids consisting of a mixture of epithelial and mesenchymal cell lines under low adhesion conditions do not self-sort according to cadherin expression.[103] In con- trast, Carey et al report that mixed epithelial and mesenchymal spheroids embedded in collagen I display outward invasion led by mesenchymal cells,[104] suggesting that sorting emerges as a competition of cell-cell and cell-matrix interactions.[105] These physical analogies are intriguing since they may facilitate quantitative predictions of tumor progression. Nevertheless, biological systems are thermodynamically far-from- 20 equilibrium, which must be carefully addressed in these physical models. A natural extension of this physical analogy is to consider how droplets (tumor spheroids) in- teract with solid surfaces. Disorganizing Spheroids as a Droplet Wetting Transition Douezan et al. measured the disorganization of multicellular spheroids on flat sub- strates with varying stiffness.[106] Murine sarcoma cells were prepared as 3D aggre- gates under agitation and low-adhesion conditions, then deposited onto fibronectin- coated substrates with varying stiffness. Quantitatively, a spreading parameter can be defined S = WCS − WCC , where WCS and WCC represent the cell-surface and cell- cell adhesion energies, respectively. If S < 0, then the cell-cell adhesion is stronger than the cell-surface adhesion (i.e. WCS < WCC ), so that the cellular aggregate does not spread (e.g. partial wetting) (Figure 1.4Ai). Instead, when 0 < S, then the cell-surface adhesion is stronger than the cell-cell adhesion (i.e. WCC < WCS ), and the aggregate spreads out as a monolayer, either cohesively as a “liquid” or dispers- ing as a “gas” (Figure 1.4Aii). In particular, on very soft substrates (< 10 kPa), aggregates remained intact without any cell dispersal (Figure 1.4Bi). For substrates of intermediate stiffness (∼10 kPa), aggregates partially flattened into a spherical cap surrounded by dispersed single cells (Figure 1.4Bii). Finally, for relatively stiff glass or PDMS substrates (∼ MPa-GPa), aggregates spread cohesively as a continuous monolayer (Figure 1.4Biii). This scenario has physical analogies with a liquid droplet on a solid surface, which will stabilize with a characteristic contact angle dependent on the relative interfacial energies.[107] The spreading dynamics of the aggregate could be explained by an effective “fric- tion coefficient” for the “slippage” of the cell monolayer on the substrate.[106] This friction coefficient was negligible for very soft substrates, where cells do not adhere. Instead, the friction coefficient was maximum for intermediate substrate stiffness 21 Figure 1.4: Multicellular spheroids spreading on planar surfaces. (A) Partial or complete wetting is governed by the relative cell-surface and cell-cell adhesions. (B) Substrate stiffness governs the solid, liquid, or gas-like dispersion of cells. Reproduced from [106] with permission from the Royal Society of Chemistry. (Ec ∼ 8 kPa),. Finally, for stiffer substrates, the friction coefficient asymptotically saturated to smaler, constant value. This biphasic dependence of cell adhesion on substrate stiffness is qualitatively consistent with a motor-clutch model.[108] Ex- perimentally, as the substrate stiffness approaches Ec , the cell motility is observed to increase, enhancing spreading. Moreover, the cell-cell adhesion also weakens be- low Ec , resulting in individual cell dissemination. Further validation of this model occurred by varying substrate functionalization with fibronectin, effectively varying cell-substrate adhesion WCS .[109] Moreover, mixtures of cells with varying E-cadherin expression were used to vary average cell-cell adhesion WCC , resulting in a transition from collective to individual migration.[109] Overall, this proposed wetting model has analogies with the binary solidification model previously discussed,[89] as well as jamming-unjamming transitions.[110] The application of theoretical concepts in- 22 spired by soft matter physics may enable new fundamental insights into tumor inva- sion and EMT, but must be carefully applied in the context of biological signaling and adaptation.[111] Multicellular Aggregates Interact through Collagen Fibers Shi et al. analyzed the disorganization and dissemination of mammary epithelial acini on collagen I surfaces (Figure 1.5A).[112] Epithelial cells were cultured as acini in a rBM overlay assay for 8 days, then transferred onto a collagen I substrate. Over 40 h, the acini disorganized into collective and individually migrating cells, which traveled along the substrate. Motile cells at the periphery displayed increased vimentin, with decreased E-cadherin and β-cateinin, consistent with EMT (Figure 1.5B). Moreover, these cells displayed nuclear localization of the YAP/TAZ mechanotransducer, in- dicative of mechanical tension on their cytoskeleton. In contrast, cells that remained localized within the acini retained elevated E-cadherin and β-catenin, with minimal vimentin, indicative of an epithelial phenotype. These behaviors were attributed to the mechanical tension acting on the acini, so that tensile stress above some critical threshold resulted in an EMT-like transition to malignant phenotype. A non-invasive epithelial phenotype could be rescued by a rectangular cut around the acini that released the mechanical tension from its surroundings. Shi et al. further utilized a photoactivatable collagen binding protein to visu- alize local matrix remodeling and large scale deformation. Remarkably, each acini displayed aligned collagen bundles which emanated radially outward. An analysis of acini pairs interacting along a collagen line revealed that they disorganized rapidly, sending cells in the direction of the other, over length scales of millimeters. Guo et al. observed qualitatively consistent interactions using a different biomaterial architec- ture based on a rBM substrate with a collagen I overlay.[113] They showed that initial cell movements were necessary to locally bundle and align collagen fibers, which in 23 A Collagen I Zoom 2 Zoom 1 10 um 100 um 20 um Nucleus, Actin B Vimentin Leader cell on collagen line E-cadherin Figure 1.5: Multicellular aggregates spreading on collagen I substrates. (A) Long-range mechanical interactions occur through collagen I bundling and align- ment (B) Dissemination of leader cells with vimentin expression and polarized mor- phologies. Reproduced from [112] with permission from the National Academy of Sciences. turn directed cell migration along this path. Over longer time periods, multicellular strands migrated outwards along these collagen fibers before condensing into elon- gated tubules. In the work of both Shi et al. and Guo et al., the formation of aligned and bundled collagen permits relatively long range force transmission through the matrix (> 500 µm). Accounting for these nonlinear and discrete fiber effects in ECM will likely require revised constitutive equations coupled with careful experimental validation.[114] 24 Enter the Matrix: Epithelial Morphogenesis and Dissemination in 3D Biomaterials Multicellular invasion is impeded within fully 3D biomaterial environments relative to 2D substrates and microfabricated geometries. In particular, the matrix architecture can present a wide variety of topographies, rigidity and adhesiveness, and can spa- tially confine cells.[12] In response, tumor cells can undergo significant morphological changes as well as remodel the local matrix architecture.[34] Although there is increas- ing use of natural hydrogels such as rBM and collagen I to investigate tumor invasion and EMT, there remain many fundamental aspects of cell-matrix interactions that remain poorly understood. The development of synthetic hydrogel materials with improved control over mechanical properties, biochemistry and degradation serves to complement these pioneering studies. Transitions to Invasion in Collagen I Gels Nguyen-Ngoc et al. directly compared cell dissemination from epithelial tissue ex- plants in rBM and collagen I (Figure 1.6A).[115] Primary human or mouse breast tu- mors were mechanically dissociated into explants consisting of hundreds of cells, and then embedded in 3D matrix. Explants in rBM remained localized with a rounded tissue morphology (Figure 1.6B). Instead, explants in collagen I exhibited protrusions at the periphery with extensive local dissemination as mesenchymal, amoeboid and collective phenotypes (Figure 1.6B). Subsequently, explants were removed from 3D matrix and then re-embedded in either rBM or collagen I. Remarkably, the subsequent behavior of these explants was primarily determined by the new matrix conditions. For instance, rounded explants in rBM displayed local dissemination when trans- ferred to collagen I, and vice versa. In contrast, explants from normal mammary tissue displayed transient dissemination in collagen I, reverting back to branching 25 morphogenesis due to the deposition of basement membrane. Subsequent work from this group showed that these migration behaviors I depended sensitively on the pre- incubation time of collagen I at 4◦ C, which prepolymerized collagen fibrils and varied the matrix architecture.[116] Overall, these experiments represent a highly promising approach to translate natural ECM as a preclinical assay to characterize cell motility from human patient samples. A B Matrigel 30 hr 50 um 20 um Carcinoma 50 um 20 um Explant F-actin DAPI Collagen I C D MCF-10A HRas cell MCF-10A HRas ECM Collagen I matrigel Matrigel shell shell Figure 1.6: Epithelial morphogenesis and invasion in natural 3D hydrogels. (A, B) Mammary epithelial explants remain localized in rBM but invade in collagen I. Reproduced from [115] with permission from the National Academy of Sciences. (C) Multicellular spheroids can be encapsulated within a rBM shell, then embeded in collagen I. (D) Mammary epithelial cells display reduced invasion with a rBM shell. Reproduced from [117] with permission from Elsevier. Ahmadzadeh et al. computationally and experimentally investigated how spheroids assembled from a melanoma cell line invaded through collagen I matrix of varying initial concentration.[118] Their mechanochemical model incorporated a feedback between cell contractility (i.e. through Rho/ROCK) and collagen fiber alignment. A major prediction of this model was that a critical stiffness existed, 26 below which the spheroid was dominated by cell-cell adhesions and remained lo- calized, and above which cell-matrix adhesion would dominate, permitting cells to break away from the spheroid. Furthermore, above a certain stiffness (effectively, collagen I concentration), cell invasion was impeded by fibril density, independent of MMP degradation. Their results show impressive agreement between theory and experiment, particularly based on a continuum model that did not explicitly account for cellular heterogeneity. An interesting future direction could be to incorporate these cellular and molecular details through multiscale modeling. Carey et al. investigated the morphogeneis of single mammary epithelial cells embeded in mixtures of rBM and collagen.[119] Multicellular clusters displayed were primarily acini-like in rBM, with increasing fractions of protrusions and invasive mor- phologies with increasing collagen I, consistent with Nguyen-Ngoc’s resuls.[115] More- over, gene expression profiling of cells in collagen I relative to rBM indicated a char- acteristic EMT signature, including loss of E-cadherin, with a gain of Snail, vimentin, fibronectin and MT-MMP1. An epithelial phenotype could be rescued in collagen I by chemical inhibition or knockdown of MT-MMP1, as well as inhibitors of PI3K and Rac1. A controlled stiffening of collagen I by non-enzymatic glycation revealed that increasing matrix stiffness also promoted an invasive phenotype. Finally, Guzman et al. demonstrated an innovative technique to self-assemble tumor spheroids with a rBM shell in a collagen I matrix.[117] First, dispersed mam- mary epithelial cells and diluted rBM were seeded in a concave, low-adhesion well and centrifuged, causing them to self-assemble into a tumor spheroid with a rBM coating after 24 h (Figure 1.6C). Tumor spheroids were then embedded into collagen I. The thickness of the rBM shell could be tuned by its diluted concentration in precursor solution, which was able to contain non-transformed mammary epithelial (MCF-10A) cells. Nevertheless, transformed MCF-10A Ras cells were capable of breaching the basement membrane shell using collective and individual invasion phenotypes. More- 27 over, MCF-10A Ras spheroids with a rBM shell displayed a significant reduction in invasion after to MMP inhibition, relative to shell-free spheroids embedded in colla- gen I (Figure 1.6D). However, some multicellular streaming was observed, in contrast to complete inhibition of migration for shell-free spheroids embedded in a mixture of rBM and collagen I. Thus, tumors with rBM shells display some features of tumor pathophysiology that may not be recapitulated by matrix consisting of mixed rBM and collagen I. Epithelial Morphogenesis in PEG Hydrogels Gill et al. investigated the formation of acini from lung epithelial cells cultured in poly(ethylene glycol) (PEG).[120] PEG is a synthetic, hydrophilic polymer with non-fouling properties that renders it relatively “inert.” [121] The matrix backbone incorporated a proteolytically degradable PQ peptide flanked by PEG chains on either side. In addition, PEG chains were modified with a fibronectin-derived RGDS pep- tide to allow cell adhesion. By varying the composition of PEG-PQ and PEG-RGDS, respectively, the stiffness and biochemistry of the hydrogel could be independently tuned. Murine lung adenocarcinoma cells were encapsulated within photopolymer- ized PEG hydrogels, eventually forming epithelial acini with hollow lumens over 2 weeks. As hydrogel stiffness or ligand density was increased, the average acini di- ameter decreased due to reduced proliferation. Instead, a higher percentage of acini formed lumens due to increased apoptosis. Lumenized acini were then treated with TGF-β to induce EMT, which resulted in a loss of polarized organization and lumens, with random proliferative and apoptotic events. Molecular analysis using RT-PCR revealed a consistent downregulation of miR-200, resulting in a decrease in classical epithelial gene expression (CDH1, CRB3), and an increase in mesenchymal genes (CDH2, VIM, ZEB1) for varying matrix stiffness and ligand density. It should be 28 noted that TGF-β treatment did not result in MMP secretion, which may have im- peded cell outgrowth and dissemination. Enemchukwu also investigated the organization of epithelial acini in PEG hydro- gels with tunable ligand density, mechanics and proteolytic degradation.[122] Their approach utilizd a four-arm PEG macromer with maleimide groups at each terminus, which permitted better defined microstructure, ligand stoichiometry and cross-linking efficiency. Kidney epithelial cells (MDCK II) were encapsulated within PEG hydro- gels of varying concentration, revealing that the formation of lumenized acini with apicobasal polarity occurred within a relatively narrow window of stiffness ( E ∼ 4 kPa), RGD ligand density (∼250 µM), and degradable crosslinker density. Acini typically displayed defects in polarization below this window and failed to form above this window. Overall, this modular and well-controlled hydrogel represents a minimal system that can recapitulate many features of epithelial morphogenesis. EMT driven by Substrate Stiffness through TWIST1-G3BP2 Wei et al. investigated the role of matrix stiffness on EMT, but utilized a collagen I-functionalized polyacrylamide (PA) gel with an overlay of diluted (2%) rBM.[123] (Figure 1.6A) Non-transformed mammary epithelial cells was observed to form acini on soft PA gels (E ∼ 0.15 kPa), while displaying invasive outgrowths and partial EMT on stiffer PA gels (E ∼ 5.7 kPa) (Figure 1.6B). EMT induction by mechanical stiffness was found to require the nuclear localization of transcription factor TWIST1, which was found to be dependent on integrin β1 . Further investigation revealed that on softer matrix, TWIST1 is retained in the cytoplasm through its interaction with Ras GTPase binding protein (G3BP2). Knockdown of G3BP2 resulted in a downregulation of E-cadherin and disruption of basement membrane, indicative of an EMT, as well as cell invasion. TWIST1-G3BP2 was further demonstrated to promote metastasis in a xenograft model, as well as poor prognosis (with increasing collagen 29 A B 150 Pa 5700 Pa Diluted Matrigel MCF-10A cells PA E-cadherin Fibronectin Nuclei C D MCF-10A cells 30 Pa 310 Pa Calcium Alginate / Matrigel Figure 1.7: Epithelial morphogenesis and invasion in synthetic 3D hydrogels. (A) Multicellular clusters on PA substrates with variable stiffness, overlaid with rBM. (B) Stiffness-dependent epithelial disorganization and EMT. Reproduced from [123] with permission from Nature Publishing Group. (C) Multicellular clusters embedded within tunable alginate-rBM networks. (D) Stiffness-dependent epithelial prolifer- ation, disorganization and invasion. Reproduced from [124] with permission from Nature Publishing Group. organization) in breast cancer patients. Altogether, Wei et al. show that stiffer matrix causes TWIST1 to be released from its cytoplasmic anchor G3BP2, resulting in nuclear localization and EMT. Interestingly, cells appear to disseminate collectively on stiffer substrates, which is suggestive of a partial EMT phenotype. The lack of individual scattering may occur due to the uniform topography of the PA gel. 30 Malignant Transitions driven by ECM through Integrin β4 Chaudhuri et al. utilized an interpenetrating network of alginate and reconsti- tuted basement membrane to investigate the role of matrix stiffness in epithelial invasion.[124] Alginate is a flexible polysaccharide derived from seaweed, which does not display mammalian cell adhesive ligands.[125] Alginate can incorporate blocks of sequential guluronic acid resides (’G blocks’) that can be crosslinked with divalent cations (Ca2+ ). Remarkably, these ’G blocks’ form ‘egg box’ conformations that permit occupancy (binding) by varying numbers of divalent cations (Figure 1.7C). As a consequence, the crosslinker strength and mechanical stiffness of alginate hydrogels could be tuned by cation concentration, without affecting the corresponding pore size or hydrogel microstructure. Non-transformed mammary epithelial cells (MCF-10A) cultured in relatively soft matrix (E ∼ 0.1 kPa) organized into growth-arrested acini, consistent with previously reported results with rBM.[59] In contast, this same cell type in stiffer matrix (E ∼ 1 kPa) formed larger multicellular clusters with invasive outgrowths (Figure 1.7D), even though ligand density and pore size were consistent with soft matrix. This mechanotransduction and malignant invasion in stiffer matrix was found to occur through signaling of integrin β4 through PI3K and Rac1 activation. Nevertheless, a growth-arrested acini phenotype could be rescued at stiff matrix by increasing basement membrane density. To explain this stiffness and composition dependent response, Chaudhuri et al. proposed that integrin β4 bound to laminin on soft ECM can undergo relatively large lateral fluctuations along the cell membrane, allowing them to form hemidesmosomes. However, when integrin β4 is bound to laminin at comparable density on stiff IPN, they are more constrained, reducing integrin β4 clustering and the formation of hemidesmosomes. As a consequence, unsequestered integrin β4 is free to drive downstream activation of PI3K and Rac1, yielding the malignant phenotype observed in stiff matrix. Finally, increasing the laminin density will increase the corresponding integrin β4 density 31 on the cell membrane, making it more likely that they can associate and form hemidesmosomes. Overall, Chaudhuri et al. demonstrated an elegant biomaterial system for orthogonal control of ligand density, microstructure, and bulk rheology, which will likely have wide applications for mechanobiology. Moreover, this approach enables unique insights into the integrated transduction of chemical and mechanical signals by epithelial cells, particularly the role of integrin β4 . It should be noted that the alginate used in this study displays limited biodegradation, which may hinder cell invasion. Future work could potentially address this through the addition of MMP-cleavable groups.[126] 1.5 Conspiring Across Borders: Co-Culture of Tu- mor and Stroma The tumor microenvironment in vivo consists not only of aberrant ECM, but also a wide range of stromal cells.[4] For instance, tumor cells can interact with cancer- associated fibroblasts, macrophages, and other immune cells which drive tumor- promoting inflammation.[29] A first step towards recapitulating these heterotypic interactions is to co-culture tumor cells with other cell types. For instance, early work by Ronnov-Jessen et al. showed that co-culture of tumor cells with fibroblasts in 3D culture resulted in a transition towards an invasive phenotype.[127] Modern biomaterials permit more compartmentalized tumor-like architectures that replicate human pathophysiology, as well as improved imaging of multicellular interactions. Cancer Associated Fibroblasts lead Tumor Invasion Labernadie et al. investigated the role of fibroblasts in tumor invasion based on spheroid dissemination onto both 3D matrix and 2D substrates.[128] Multicellular spheroids consisting of vulval squamous cell carcinoma (SCC) cells (A431) were em- 32 bedded in a 3D mixture of collagen and rBM, which also included cancer associated fibroblasts. Remarkably, fibroblasts were observed to approach tumors from afar and attach to tumor cells through heterophilic E-cadherin/N-cadherin junctions (Figure 1.8A). Subsequently, these fibroblasts reversed directions and led multicellular tumor invasion into the matrix. These leader and follower behaviors also occurred when the spheroids were plated on a soft 2D substrate and surrounded by fibroblasts (Figure 1.8B). In this latter system, traction force microscopy was used to show that fibrob- last leaders exert pulling forces on cancer cell followers through these heterophilic cell-cell junctions. Further knockout of E-cadherin disrupted cell-cell junctions, pre- venting tumor cells from migrating collectively behind fibroblast leaders. It should be noted that these tumor cells were not defective at migration, but were simply unable to function as collective followers. Finally, these heterophilic junctions were observed for patient-derived cells cultured in fully 3D matrix. Overall, co-culture of tumor cells and fibroblasts revealed a cooperative mechanism whereby fibroblasts lead tumor cells into the matrix, while tumor cells sustain fibroblast polarization. It is also remarkable that heterophilic E-cadherin/N-cadherin junctions are mechanically active and com- parable in activity to homophilic junctions. These new results must be considered in the physical theories that incorporate differential-adhesion like mechanisms.[95] Sung et al. utilized a microfluidic device to pattern tumor cells and fibroblasts in adjacent compartments, revealing that soluble signals could trigger morphological changes, but that cell-cell contact is necessary for invasion (Figure 1.8C).[129, 132] Singh et al. encapsulated melanoma cells within a PEG hydrogel core, which was then enclosed with primary human dermal fibroblasts in a collagen I hydrogel. They found that co-culture suppressed epithelial cluster growth, but subsequently activated a program of collective or individual invasion.[133] Overall, the patterning of tumor and stromal cells into discrete compartments represents an important step towards reverse engineering the complex tumor architecture. 33 A A431 CAF D Fibroblast Collagen/Matrigel 1 Endothelial cells Endothelial cells Tumor cells CAF 2 Tumor cells 3 ECM B A431 CAF Fibrin gel ECM E PA Endothelial cells PA C MCF10-DCIS HMF Collagen I Tumor cells Endothelial cells collagen Tumor cells Figure 1.8: Co-culture models of tumor cells and stromal cells. (A). Fibroblasts are recruited to tumor spheroids and then lead collective invasion in 3D collagen-rBM hydrogels. (B) Fibroblasts display similar leader cell behavior on soft 2D PA substrates. Reproduced from [128] with permission from Nature Publish- ing Group. (C) Microfluidic patterning of fibroblasts and tumor cells into adjacent compartments. Reproduced from [129] with permission from the Royal Society of Chemistry. (D) Endothelial cells sprout into microvascular networks in fibrin gels, which permit intravasation-like behaviors by tumor cells. Reproduced from [130] with permission from the Royal Society of Chemistry. (E) Targeted EMT inhibitors can be screened in microfluidic devices with tumor spheroids and endothelial barriers. Reproduced from [131] with permission from the Royal Society of Chemistry. 34 Tumor Intravasation and Extravasation in Endothelial Vessels Ehsan et al. investigated the entry and migration of tumor cells within endothelial vessel networks in a 3D fibrin gel.[130] Fibrin is a natural polymer that arises from a cascade of enzymatic reactions during blood clotting.[134] Multicellular spheroids consisting of human tumor cells mixed with primary human endothelial cells were seeded in fibrin and surrounded by fibroblasts. Endothelial cells rapidly sprouted from these spheroids and formed highly branched networks within the gel over the course of a week (Figure 1.8D). Moreover, endothelial cells reorganized at the periphery of the spheroid, infiltrating into the spheroid in order to form an internally connected network. At the same time, individual tumor cells broke away from the spheroid and infiltrated the fibrin gel. Remarkably, a colon cancer cell line was observed to migrate along the lumens within the sprouting vascular network. This intravasation- like behavior was enhanced under hypoxic conditions, which was found to depend on the EMT transcription factor Slug. This complex intermixing of tumor cells and endothelial cells may have unexpected implications for abnormal angiogenesis in more advanced tumors in vivo. Aref et al. perturbed multicellular invasion using targeted EMT inhibitors in a microfluidic 3D culture system,[131] utilizing a compartmentalized microfluidic device where vertical endothelial monolayer could be cultured adjacent to a 3D hydrogel[135] Aref et al. utilized lung adenocarcinoma (A549) cells with some propensity to revert from a partial mesenchymal phenotype to a more epithelial phenotype after inhibi- tion of EMT-related pathways. Cells were self-assembled into 40-70 µm spheroids in a low-attachment dish, then seeded in a collagen matrix within the microfluidic device (Figure 1.8E). Interestingly, spheroids dispersed due to the presence of en- dothelial cells, but not in culture media supplemented with growth factors (FGF, EGF). Subsequent treatment with targeted inhibitors of EMT pathways (e.g. AkT, EGFR, IGF1, MEK, PDGFR¡ SRC and TGF-βR1) resulted in decreased prolifera- 35 tion and migration. Unexpectedly, the effective IC50 observed to inhibit migration in these 3D hydrogels was found to be roughly 5 to 10-fold lower than the corre- sponding IC50 needed to inhibit spreading of these spheroids onto a 2D substrate. The effective concentrations in 3D culture were reported to be comparable to plasma concentration that were effective in clinical trials. Overall, microfluidic devices may enable controlled application of molecular gradients, forces, and flows, which could recapitulate essential pathophysiological features.[136] These so-called ”organ-on-a- chip” microphysiological systems have great potential to complement existing animal models as predictive models of drug efficacy and toxicity.[137] Nevertheless, careful consideration of scaling will be necessary to compare drug pharmacokinetics with animal models or human patients.[138] 1.6 Future Directions What is the Matrix? Beyond Stiffness and Mesh Size Synthetic and natural biomaterials have typically been characterized in terms of bulk mechanical properties, such as elastic modulus (E) or shear modulus (G).[139] In the publications highlighted here, an overall trend is that epithelial cells cultured on stiffer biomaterials exhibit aberrant epithelial morphogenesis,[120, 122] enhanced invasiveness,[124] and EMT induction.[123] This transition to malignant behavior is qualitatively consistent with tumor progression in vivo, where ECM becomes dramat- ically stiffened with highly crosslinked and bundled collagen I.[30] It should be noted that synthetic hydrogels such as PA or PEG can be understood as flexible polymer networks, which can be engineered with a controlled stiffness and well-defined linear response.[140] Instead, natural hydrogels and physiological ECM exhibit additional physical complexity that is not captured solely by elastic modulus. In particular, natural hydrogels can be more fibrous with highly nonlinear rheology, requiring more 36 sophisticated physical descriptions based on semiflexible or rigid polymers.[141] Re- constituted collagen I or fibrin networks can display significant strain stiffening as a generic consequence of nonlinear force extension in semiflexible polymers.[142] This nonlinear behavior has important biological implications, since cells or tissues have been observed to mechanically interact over much longer distances than they would in a linear elastic material.[113, 112] Moreover, physiological ECM can exhibit signif- icant viscoelasticity. For instance, Chaudhuri et al. also showed that mesenchymal stem cells display faster cell spreading, proliferation, and osteogenic differentiation in hydrogels with faster stress relaxation times, even when the initial elastic modulus was held constant.[143] Further work is needed to characterize and fundamentally understand the role of nonlinear ECM rheology in tumor invasion and EMT. Cancer cell invasion is also restricted by the characteristic mesh size of the hy- drogel, with a limit of ∼7 µm2 imposed by nuclear deformability.[50] For synthetic hydrogels, the mesh size of flexible polymers is typically quite small, on the order of 1-100 nm,[139] requiring significant matrix degradation to facilitate cell migration. In contrast, natural hydrogels can exhibit larger mesh sizes on the order of microns, which may be more permissive for cell migration. Conceptually, natural hydrogels and pathological ECM may behave more as a discrete network of fibers rather than a continous porous architecture. Indeed, microstructural architectures may shape cells through contact-guidance like mechanisms, complementing the role of ligand density and bulk mechanical stiffness.[12, 6] Thus, some care is necessary when uti- lizing continuum metrics such as an elastic modulus, or average properties such as a mesh size. Instead, more systematic characterization of network geometry[144] and pore size distribution[145] may yield new physical insights. Indeed, an intriguing approach is to quantify the connectivity of a given matrix architecture (i.e. a perco- lation threshold for a nucleus or cell-sized object),[146] and compare this with actual migration trajectories. These characterization tools may be further combined with 37 recent advances in 3D traction force microscopy to visualize localized cell-generated forces.[147, 148, 149] Overall, understanding the interplay between invasion, matrix architecture and mechanical properties will also be facilitated by improved materials characterization techniques at cellular and subcellular length scales. Follow That Cell: Tracking Phenotypic Heterogeneity Cells that disseminate from the tumor periphery represent uncommon phenotypes, which are likely to differ from the cells that remain within the tumor. One advan- tage of engineered biomaterial platforms is their experimental accessibility relative to animal models, which is particularly useful for optical microscopy. In the publica- tions highlighted here, fluorescence microscopy was utilized to reveal cell migration at the periphery,[87, 89, 109, 150, 115, 118, 117] distinct biomarker expression,[88] disorganization of epithelial architecture,[112, 113, 119, 120, 122, 123, 124] as well as coordination between tumor and stromal cells.[129, 133, 128, 130, 131] However, many of these publications utilized manual analysis of selected cells or tissues, which sam- ples a limited subset of the population with low throughput. Instead, computer vision and machine learning represent a powerful approach for phenotypic profiling.[151] In general, fluorescently labeled cells must be individually identified and distinguished from the image background. Next, cell shape, texture, and biomarker expression can be analyzed as a set of quantitative features. Finally, the most biologically relevant parameters are determined based on feature selection or dimensionality reduction. Single cell features can then be profiled through unsupervised approaches that cluster distinct phenotypic subpopulations, or supervised approaches for classification based on linear or nonlinear decision boundaries. We have recently demonstrated auto- mated profiling of EMT in 2D monolayer culture based on single cell morphology and biomarker expression,[152] as well as cell morphology on nanotextured materials.[153] It should be noted that even with automated image analysis, migratory cells may be 38 detected in limited numbers, which are insufficient to define a phenotypic cluster. Unsupervised outlier detection may be useful to profile cells which are highly dissim- ilar from the others, but are biologically significant. Moreover, new algorithms (i.e. SPADE) can downsample abundant cells in high-dimensional feature space in order to facilitate the clustering of rarer cells.[154] Continuous longitudinal tracking of single cell behaviors represents the next step beyond single cell profiling based on discrete “snapshots.”[155] These high-resolution measurements are crucial to capture rare and transient events, such as the detachment of cells from the tumor periphery. Nevertheless, these experiments are associated with significant technical challenges, including maintaining consistent cell-friendly condi- tions (i.e. temperature, CO2, humidity, and medium osmolality), minimizing photo- toxicity and photobleaching, as well as optimization of live cell fluorescent markers and reporters. Moreover, image analysis requires not only accurate cell segmentation (as previously described), but also linking cell positions into continuous trajectories across time-lapse images. Nevertheless, this approach permits unique insights into the cell dissemination. In particular, we have shown that distinct migration phenotypes can be classified based on quantitative metrics such as the speed and straightness of cell migration trajectories, as well as lifetime averaged nearest neighbors.[89] Inter- estingly, the percentage of cells classified as individual was relatively small (∼ 16%), and these cells could be further subdivided into subpopulations that were mostly collective with some individual migration, as well as mostly individual with some col- lective migration. An ongoing challenge is to keep track of cells as they proliferate, which results in increasing cell density with enhanced cell contact or overlap. We and others have tracked cell nuclei, since they remain relatively well dispersed with a relatively well defined shape. Our recent results have utilized these approaches to track the migration and self-organization of heterogeneous mixtures of epithelial and mesenchymal tumor cells.[156] 39 Increasing use of fully 3D hydrogel microenvironments will permit increased phys- iological relevance but will also result in new experimental challenges. First, three- dimensional fluorescence imaging requires the use of spinning disk confocal micro- scopes, multiphoton microscopy, or other advanced superresolution techniques.[157] The imaging of cell and matrix volumes, rather than areas, will result in exponentially larger datasets, which must be stored, visualized and analyzed differently.[158] In ad- dition, immunofluorescent staining of cells will be complicated by the slow transport and non-specific adsorption of reagents through the hydrogel, requiring additional experimental optimization. Next, integration of single cell imaging with molecular analyses remains complicated. In particular, measurements of single cell gene expres- sion, epigenetics and protein content are typically destructive and must be conducted at an endpoint. Moreover, cells must be isolated from the biomaterial, which is challenging to achieve with 100% extraction efficiency and loses spatial information. Overall, considerable challenges remain in elucidating multicellular invasion in bio- materials, particularly a mechanistic understanding of how microenvironmental cues are transduced to downstream molecular signaling, resulting in phenotypic decisions. Scaling Up: Standardization using Multiwell Plates One advantage of engineered biomaterials is that they can be utilized with multi- well plates, which enables higher throughput screens of experimental conditions in a massively parallel format. In practice, much of the research described here has been implemented in some type of multiwell plate in order to improve optical imaging conditions and limit aqueous solutions from leaking. For instance, rBM or collagen I can be manually dispensed,[115] or PDMS devices can be bonded to glass bottom 24-well plates.[89] It has also been shown that standardized PA gel substrates can be cast in a 96 well plate format.[159] Nevertheless, the transition to increased labora- tory automation, including plate handling robotics, high precision fluid dispensing, 40 and automated high-content microscopy remains somewhat limited. Clevers and col- leagues have demonstrated that tumor and normal organoids can be derived from colorectal carcinoma patients in a 384 well plate format and screened against a panel of 83 drugs.[160] Lutolf and coworkers have developed modular approaches to screen synthetic PEG hydrogels with well-defined mechanical and biochemical properties. For instance, a robotic liquid-dispensing platform was demonstrated that enabled in- dependent control of cell density, hydrogel stiffness, MMP sensitivity, extracellular matrix components, cell-cell interaction components, and soluble factors.[161] This screen of 3D culture conditions revealed that physical confinement of induced pluripo- tent stem cells could enhance reprogramming through an accelerated mesenchymal- epithelial transition and epigenetic modifications.[162] As the well size becomes pro- gressively smaller, some consideration of scaling effects will likely be necessary, par- ticularly the depletion of biochemical factors in solution. Overall, the use of stan- dardized biomaterials in an automated format should enable preclinical therapeutic testing, rational drug design, as well as predictive and prognostic assays of human samples. Biological Questions: EMT and Interpatient Heterogeneity? Tumors display a variety of collective and individual invasion phenotypes, a phe- nomenon that remains poorly understood.[1] In particular, the role of EMT remains controversial, given the reliance on 2D monolayer culture and animal models in the field.[35] Recent reports by Fischer et al. and Zheng et al. using mouse models suggest that EMT is not required for invasion and metastasis, but play a role in therapeu- tic resistance.[163, 164] Interestingly, circulating tumor cells from human patients can exhibit a co-expression of epithelial and mesenchymal biomarkers in an inter- mediate or partial EMT,[165] which is not observed for classical EMT in embryonic development.[2] The interconversion between epithelial and mesenchymal phenotypes 41 remains similarly controversial, due in part to the analysis at endpoints, as well as the lack of tumor-specific mesenchymal biomarkers. The technologies and future di- rections discussed here may enable improved experimental control of the ECM, as well as enhanced spatiotemporal resolution to address these ongoing questions. We envision that these approaches are more generally applicable beyond EMT to pro- file phenotypic heterogeneity and plasticity in tumors, which represents a subtle and complex problem. Our review of multicellular behaviors in biomaterials is closely related to recent developments in organoid culture, which have been broadly defined. In particular, Shamir and Ewald suggest that: “...in the field of mammary gland biology, the term organoids refers to primary explants of epithelial ducts into 3D extracellular matrix (ECM) gels. Conversely, in studies of intestinal biology, organoids can refer to clonal derivatives of primary epithelial stem cells that are grown without mesenchyme or can refer to epithelial-mesenchymal co-cultures that are derived from embryonic stem cells or induced pluripotent stem cells”[166, 167] Recent successes have been based on extensive use of rBM or collagen I scaffolds with optimized media formulations, but Gjorevski et al. have demonstrated a tunable and modular PEG-based scaffold for organoid culture.[168] Interestingly, normal intestinal stem cells required a soft matrix and laminin-based adhesion, and could be differentiated by gradual softening through hydrolytic degradation. In contrast, stiffer and MMP-degradable matrix resulted in tissue disorganization and the upregulation of stress and inflammatory gene expression, representing a pathological state. These approaches could be used to investigate how patient-derived cells respond to alterations in the microenvironment, as well as to maintain them more effectively within certain phenotypes. Overall, the potential derivation of organoids from human patients could potentially bridge the gap between transformed cancer cell lines and patient-derived xenograft models.[13] 42 An exciting prospect is to utilize organoids for biomarker and drug discovery, as well as preclinical screens for personalized therapies. 1.7 Conclusions Natural and artificial biomaterials recapitulate important physical or biochemical features of the ECM, which can be used to elucidate tumor disorganization and dis- semination. In this review, we highlight a selection of recent results that address multicellular invasion and EMT within biomimetic microenvironments. First, micro- fabrication techniques can precisely define interfacial geometries, revealing the detach- ment and directed migration of individual cells, as well as the emergence of stem-like phenotypes. Second, multicellular aggregates spread on planar surfaces as a bal- ance of cell-cell and cell-matrix adhesions, which has physical analogies with droplet wetting. Moreover, when plated on fibrillar collagen I substrates, these aggregates mechanically interact and reorganize collagen into aligned bundles. Third, epithelial cells embedded in 3D hydrogels can organize into well-defined glandular architectures in reconstituted basement membrane, but undergo a transition to invasion in fibrillar collagen I or stiffer biomaterials. Fourth, co-culture of tumor cells with stromal cells such as fibroblasts or endothelial cells can drive heterotypic soluble or mechanical signaling. Interestingly, fibroblasts can act as leader cells to enhance multicellular invasion, while tumor cells can intravasate into endothelial networks. Finally, we consider future directions for the field, including improved physical understanding of ECM, profiling of single cell heterogeneity, standardization of biomaterials in a multi- well plate format, as well as the use of organoid culture. Altogether, we envision that the technologies reviewed here will facilitate fundamental insights into reciprocal in- teractions between tumor progression and the ECM, as well as enable patient-specific biomarker discovery and drug testing for precision medicine. 43 Chapter 2 Wrinkled, wavelength-tunable graphene-based surface topographies for directing cell alignment and morphology Chapter 2 has been previously published as a primary article: Z. Wang,† D. Tonderys,† S.E. Leggett,† († co-first authors) E.K. Williams, M.T. Kiani, R.S. Steinberg, Y. Qiu, I.Y. Wong, and R.H. Hurt. “Wrinkled, wavelength- tunable graphene-based surface topographies for directing cell alignment and morphology.” Carbon. 97, 14-24, Feb 2016 44 2.1 Abstract Textured surfaces with periodic topographical features and long-range order are highly attractive for directing cell-material interactions. They mimic physiological environments more accurately than planar surfaces and can fundamentally alter cell alignment, shape, gene expression, and cellular assembly into superstructures or microtissues. Here we demonstrate for the first time that wrinkled graphene-based surfaces are suitable as textured cell attachment substrates, and that engineered wrinkling can dramatically alter cell alignment and morphology. The wrinkled sur- faces are fabricated by graphene oxide wet deposition onto pre-stretched elastomers followed by relaxation and mild thermal treatment to stabilize the films in cell culture medium. Multilayer graphene oxide films form periodic, delaminated buckle textures whose wavelengths and amplitudes can be systematically tuned by variation in the wet deposition process. Human and murine fibroblasts attach to these textured films and remain viable, while developing pronounced alignment and elongation relative to those on planar graphene controls. Compared to lithographic patterning of nanogratings, this method has advantages in the simplicity and scalability of fabrication, as well as the opportunity to couple the use of topographic cues with the unique conductive, adsorptive, or barrier properties of graphene materials for functional biomedical devices. 2.2 Introduction Patterning of surface topography is a powerful technique for controlling interfacial interactions between a material and its environment [169]. Topographical patterns can be created by etching or molding the surface of a single-component material, or through creation of heterostructures consisting of a substrate and a surface film with engineered texture. An emerging method for surface texturing is the creation of wrin- 45 kle patterns by controlled shrinkage of a stiff coating on a softer, compliant substrate [170, 171, 172, 173]. This approach has been experimentally implemented in a variety of polymeric and inorganic material systems [174], and the wrinkle morphologies have been the subject of theoretical treatments of buckling instability [175, 176, 177]. An exciting new approach to the creation of these textured surface films is the growth or deposition of two-dimensional, sheet-like nanomaterials, such as graphene, whose atomically thin nature enables the creation of the ultrathin flexible films suitable for controlled wrinkling. Topographically patterned graphene has found numerous ap- plications in optical and electronic devices, energy storage, and functional coatings [178, 179, 180, 181, 182, 183, 184, 185, 186, 187]. Here we demonstrate another application area for graphene surfaces with engi- neered wrinkle structures: as functional substrates for cell and tissue engineering. Planar graphene and graphene oxide (GO) have already been explored as substrates for biological cells and tissues [188, 189, 190], and remarkably, mesenchymal stem cells, myoblasts and fibroblasts appear to display enhanced viability compared to conven- tional tissue culture substrates [191, 192, 193, 194, 195]. A limitation of conventional flat, uniform 2D cell culture substrates, however, is that they lack the complexity of structural architectures found in the extracellular matrix in living tissue. On planar 2D surfaces, cells adopt strongly flattened morphologies, and the resulting cellular behavior can deviate from the natural behavior observed in a physiological 3D con- text. Modern nanopatterning approaches strive to create biomimetic features that are comparable in size and geometry with molecular elements of the natural microen- vironment [196, 197, 198, 6]. In particular, interstitial collagen in the extracellular matrix are bundled together with diameters from tens to hundreds of nanometers, as well as pore sizes or gaps on the order of 5-20 µm [12]. These anisotropic topogra- phies can affect cell morphology and orientation, a phenomenon known as contact 46 guidance [199]. Cell-substrate interactions that mimic this anisotropy using aligned grooves have been previously investigated on polymeric materials using controlled buckling [200, 201, 202] and micropatterning [203, 204, 205, 206, 207], revealing al- tered migration dynamics, proliferation, gene expression and differentiation. In this context, graphene substrates incorporating microscale topography are intriguing as functional substrates for cell and tissue engineering, but have not, to our knowledge, been previously examined. Overall, the design of biomaterial interfaces based on graphene represents an exciting approach for understanding fundamental cell biol- ogy, nanostructured scaffolds for tissue engineering and regenerative medicine as well as to promote biocompatibility and biointegration of functional medical implants in neuronal, cardiovascular or epidermal tissues. In this article, we demonstrate the formation of wrinkled multilayer graphene surfaces using GO solution phase deposition on pre-stretched elastomeric substrates followed by relaxation and thermal stabilization. We find that the topography of these reduced graphene oxide (rGO) surfaces is maintained during thermal reduction, and displays exceptionally sharp features, whose spatial periodicity can be systematically tuned by simple variation of the GO concentration in the deposition suspension. We examine the effect of these wrinkled rGO architectures on human and murine fibroblast cells, which attach and remain viable, and cause important changes in cell orientation, alignment, and morphology relative to cells on planar rGO. This work establishes the feasibility of graphene wrinkle engineering for the fabrication of textured substrates for cell and tissue engineering and potential applications in biomedical implants. 47 2.3 Experimental 2.3.1 Fabrication of textured surfaces GO suspensions were prepared by a modified Hummers method and purified and characterized as described previously [208]. These GO sheets are primarily in mono- layer form in aqueous suspension and are 1-5 µm in lateral dimension with a C/O atomic ratio of approximately 1.8 [208]. The elastomeric substrates were silicone rub- ber sheets (McMaster-Carr) of 1/16 thickness and either 50A or 20A hardness. The elastic modulus of the substrates was measured using an Instron 5882. The elastomer films were cut into 4x2 cm pieces, washed with deionized water, and fixed at one end to glass slides. The films were stretched uniaxially to a desired pre-strain (1.5%-50%) and secured with a fastener. A specific volume of GO aqueous suspension (typically 150 µL of 0.2 g/L GO) was pipetted onto the substrate, dried overnight, and relaxed to form the wrinkle patterns (Fig. 2.1A). For cell studies the films were rendered stable to re-dissolution in cell culture medium by mild thermal reduction through annealing overnight at 120◦ C in air. 2.3.2 Morphology characterization Morphologies of the textured surfaces were characterized by optical microscopy, atomic force microscopy (Asylum MFP-3D Origin AFM) in pulse mode, and field emission SEM (LEO 1530 VP) in the variable operating pressure mode without coating the samples, in top view (for wavelength determination) or high tilt (for transverse profiles). The characteristic wavelength of the wrinkle features were quantified by sampling grey-scale line profiles from the micrographs using Gatan DigitalMicrograph, followed by Fast Fourier Transformation in MATLAB (Math- works). 48 2.3.3 Cell culture NIH-3T3 mouse fibroblasts were a gift from Dr. Agnes Kane (Brown University), while normal human fibroblasts (NHF cells) derived from neonatal foreskin were a gift from Dr. Jeffrey Morgan (Brown University). Cells were cultured in high glucose, pyruvate Dulbeccos Modified Eagles Medium (DMEM, Life Technologies # 11995) supplemented with 10% Fetal Bovine Serum (FBS), 100 units/mL penicillin, and 100 µg/mL streptomycin. Both cell lines were carried out under standard culture conditions: incubation at 37◦ C in a 5% CO2 gaseous environment, which pairs with the DMEMs sodium bicarbonate buffer system to maintain physiological pH, cells were cultured in T-25 flasks, and passaged at 70-80% confluence. Only low passage numbers were used in experiments (P5-P12 for NIH-3T3s and P4-P10 for NHFs). 2.3.4 Preparation of rGO-coated substrates for cell culture experiments Paired wrinkled and flat rGO substrates were prepared as described previously. For cell culture, substrates were further processed by immersion and rinsing in DI water (Milli-Q, 18.2 MΩ) for at least 12 hours to remove debris and any contaminants. Substrates were then sterilized by UV exposure with a 30W lamp for 1 hour before transferring to a 24 well plate. Once the rGO materials were prepared and sterilized, cells cultured in T25 flasks were incubated with HyClone 0.05% Trypsin for detachment. Once cells were lifted off the flask, the suspension was transferred to a conical tube containing growth medium and centrifuged at 1000RPM for 5 minutes. The cell pellet was then re-suspended in culture medium, and cells were subsequently counted using an automated cell viability counter (Nexcelom Cellometer Auto 1000). Next, a 100 µL droplet with a concentration of 40,000 cells/mL was deposited on the materials, allowed to settle for 49 ∼1 hour, and then the well was filled with 1 mL of culture media. After 48 h, the cell culture media was replenished. 2.3.5 Cell viability After culturing cells on polystyrene, flat rGO, and wrinkled rGO substrates for 96 hours, culture medium was aspirated and replenished with media containing Hoechst (bis-Benzimide), diluted 1:2000. Cells were incubated for 30 minutes at 37C in 5% CO2 to label cell nuclei, the Hoechst-media was aspirated, and cells were rinsed with 1X phosphate buffered saline (PBS). To label dead cells, DRAQ7 (BioStatus), diluted 1:100 in culture medium, was added to each well. Cells were incubated for an additional 10 minutes, and then imaged using an inverted epifluorescence microscope (Nikon TiE). Images were acquired using a 10X Plan Fluor objective (NA 0.3, long working distance) with fluorescence illumination was provided using a light-guide coupled Lumen Dynamics X-Cite 120 LED system. All images were acquired with 14-bit resolution using a sCMOS camera (Andor Neo). Care was taken to ensure that all images were recorded with identical acquisition parameters (exposure time, camera gain/gamma control and microscope aperture settings). To determine the percentage of viable cells in each condition, the total cell nuclei (Hoechst label) and dead cells (DRAQ7 label) were counted using CellProfiler (version 2.1, Broad Institute). Cell viability was defined by the ratio of live cells (Total nuclei count Dead cell count) to the total number of cells (Total nuclei count); the results were plotted as bar graphs, with the use of MATLAB (Mathworks). 2.3.6 Immunostaining and fluorescence imaging After 96 h of culture on wrinkled and flat rGO substrates, cells were fixed using 4% paraformaldehyde in 1X PBS. Cells were then permeabilized with 0.1% Triton- X 100 in 1X PBS. For F-actin microfilament detection, cells were immunostained 50 with a conjugated antibody, Alexa Fluor 647 Phalloidin (Invitrogen), diluted 1:80 in 1% nonfat dry milk in 1X PBS. Cell nuclei were also labeled by counterstaining with Hoechst (bis-Benzimide), diluted 1:5000 in 1X PBS. 1X PBS was added to the wells, and immunostained samples were placed upside down in a new 24 well plate for imaging using an inverted epifluorescence microscope (Nikon TiE). Images were acquired using a 10X Plan Fluor objective (NA 0.3, long working distance) or a 20X Super Plan Fluor objective (NA 0.45, extra long working distance). Fluorescence illumination was provided using a light-guide coupled Lumencore Sola white light excitation system. All images were acquired with 14-bit resolution using a sCMOS camera (Andor Neo). Care was taken to ensure that all images were recorded with identical acquisition parameters (exposure time, camera gain/gamma control and microscope aperture settings). 2.3.7 Image processing for quantification of cell morphology CellProfiler (version 2.1, Broad Institute) was utilized for cell segmentation and analysis[209], with manual verification afterwards. First, fluorescently labeled nuclei were segmented as primary objects using maximum correlation thresholding (MCT) and clumped objects were resolved using shape and intensity. Next, based on the location of the nuclei, fluorescently labeled F-actin in the cell body was segmented as a secondary object. For wrinkled surfaces, secondary objects were best segmented by a robust background threshold and adaptive threshold strategy. Instead, for flat controls, secondary objects were best segmented using the Ostu thresholding method and global thresholding strategy. Finally, the MeasureObjectSizeShape module was employed to extract the following shape descriptors for the detected objects: Area: The actual number of pixels in the region. Eccentricity: The eccentricity of the ellipse that has the same second-moments as the region. The eccentricity is the ratio of the distance between the foci of the ellipse 51 and its major axis length. The value is between 0 and 1. (0 and 1 are degenerate cases; an ellipse whose eccentricity is 0 is actually a circle, while an ellipse whose eccentricity is 1 is a line segment.) Solidity: The proportion of the pixels in the convex hull that are also in the object, i.e. ObjectArea/ConvexHullArea. Equals 1 for a solid object (i.e., one with no holes or has a concave boundary), or < 1 for an object with holes or possessing a convex/irregular boundary. Orientation: The angle (in degrees ranging from -90 to 90 degrees) between the x-axis and the major axis of the ellipse that has the same second-moments as the region. Compactness: The variance of the radial distance of the object’s pixels from the centroid divided by the area. MinorAxisLength: The length (in pixels) of the minor axis of the ellipse that has the same normalized second central moments as the region. MaxFeretDiameter: The Feret diameter is the distance between two parallel lines tangent on either side of the object (imagine taking a caliper and measuring the object at various angles). The maximum Feret diameter is the largest possible diameter, rotating the calipers along all possible angles. We found that width of a cell was poorly approximated by the minor axis of an ellipse with equivalent second central moments, resulting in an overestimation of cell width. Instead, we defined an alternate metric: Average Cell Width = Area/MaxFeretDiameter, which is based on the actual area of the cell. 2.3.8 Cell orientation and statistical analysis A variety of circular statistical tests were applied using Oriana (Kovach Computing Services) to test whether cellular orientations were uniformly distributed (one-sample 52 tests) or to compare the similarity of two distributions (multiple sample tests) [210]. Rao’s Spacing Test was used to evaluate the null hypothesis that a given circular dataset is uniformly distributed by checking if the differences between successive measurements are comparable to 180◦ /N , where N is the number of measurements. In addition, the Mardia-Watson-Wheeler test was used to evaluate the null hypothesis that two samples have identical distributions as a circular analogue of the two sample t-test. It should be noted that, for simplicity, the coordinate system for wrinkled materials was defined along the length of the wrinkles. Circular statistics were plotted in MATLAB (Mathworks) using ROSE. For all other metrics of cell morphology, the statistical distributions were com- pared using the two-sample Kolmonogorov-Smirnov test in MATLAB (Mathworks). Statistical significance was determined by rejecting the null hypothesis at p ≤ 0.05 (5% significance level). Plots were also generated in MATLAB, as a combination of BOXPLOT, where the dividing line corresponds to the median, 25th and 75th per- centiles are indicated by the box edges and whiskers correspond to 99.3% coverage, as well as PLOTSPREAD (MATLAB File Exchange), which displays data points representing individual cell measurements. 2.4 Results and discussion 2.4.1 Film fabrication and structure The main objective of this research was to explore wrinkled GO surfaces as anisotropic cell attachment substrates for control of cell alignment and morphology in tissue engineering applications. We chose a fabrication route based on GO wet deposition and mild thermal reduction, which is a potentially practical and scalable method that is an attractive alternative to large-area coverage by pristine CVD graphene. Figure 2.1 shows the basic morphology of the wrinkled GO surfaces prior to reduction created 53 by relaxation of substrates with 50% uniaxial pre-strain and covered by 200 nm thick GO multilayer films. Relaxation produces a series of nearly parallel ridges with high- curvature crests separated by broader valleys (Fig. 2.1A and C). The wrinkle patterns can be removed by re-stretching to the initial state (Fig. 2.1B), and the process is reversible over multiple cycles. These features are similar to those observed previously for pristine few-layer graphene coatings undergoing similar uniaxial compression [184, 186]. The present films show longitudinal cracking (Fig. 2.1D left) during relaxation, which we suspected was due to Poisson expansion in the transverse direction. The cracks here did not appear to affect the periodic wrinkle textures, but do reveal the underlying substrate, which would be an undesirable feature in some applications. We therefore sought a solution, and found that the cracks could be suppressed by fixing the sides of the film during the pre-stretch to mechanically constrain its Poisson contraction, which then also suppresses its subsequent Poisson expansion during the relaxation and wrinkling process (Fig. 2.1D right). The longitudinal cracks in the films fabricated in our early attempts without transverse pinning were undesired, but did make it possible to directly view the cross- section wrinkle profiles in detail (Fig. 2.1E). The wrinkles are delaminated buckled structures with nearly straight slopes and with curvatures concentrated at the ridge tops. The ridge tops have characteristic curvatures on the order of the film thickness, and are similar to the sharp ridges seen in crumpled graphene particles formed during isotropic compression [211, 212], which are 3D structures having both ridges and vertices. The films remain on the substrate through adhesion at the valley floor regions, and the local delamination produces cavities with triangular cross-sections between the substrate and film. This delamination buckling in multilayer GO has also been observed in pristine few-layer or multi-layer graphene films [184, 186], while amorphous carbon films fabricated by ion beam deposition from hydrocarbons do not show such local delamination when they undergo spontaneous buckling associated 54 Figure 2.1: Fabrication process and morphology characterization of wrinkled graphene-based multilayer films. (A) Illustration of the fabrication process for wrinkled GO multilayer films. (B) Demonstration of reversibility by re-stretching under an optical microscope, scale bar is 100 µm. (C) AFM image showing the basic morphology of parallel ridges with sharp crests (light) and broader valleys (dark). Film thickness is 20 nm and scale bar is 20 µm. (D) SEM showing longitudinal microcracks that form when simple uniaxial relaxation occurs (left image). These cracks can be suppressed (right image) if the substrate is constrained in the transverse direction during relaxation to prevent the expansion associated with the Poisson effect, scale bar 100 µm. (E) High-tilt SEM image of transverse profiles, which shows a delaminated buckle structure with nearly straight slopes (arrow 1) and curvature concentrated at the ridge tops (arrow 2). These wrinkles are high-amplitude buckled ridges in which some primary peaks have collapsed on neighbors to create double-ridges (arrow 3). Scale bar 20 µm. All samples in Fig. 2.1 unless noted use 50% pre-stretch and 200 µg/mL GO suspension that gives a nominal GO film thickness of 200 nm. 55 with growth-induced film compression [187]. The difference may reflect very different film-substrate bonding in the graphene-based films and the ion-beam deposited films. For manipulation of cell alignment and morphology, it is desirable to control the amplitude and wavelength of the wrinkle patterns in our multilayer graphene films. Approaches to this control are readily suggested by the significant literature on the theoretical and experimental mechanics of stiff thin films on pre-stretched compliant substrates [183, 213, 214, 215]. The critical buckling strain is reported to be c =  2/3 1 3Es (1−νf2 ) 4 Ef (1−νs2 ) , where Ef and Es are the Young’s modulus of film and substrate, respectively, and νs and νf are the Poisson’s ratio of substrate and film, respectively. When this strain is exceeded, wrinkles develop with a wavelength: Ef (1 − νs2 ) " # λ0 = 2f (2.1) 3Es (1 − νf2 ) where hf is the film thickness. This theory is valid in the limit of small deformation. When the deformation is large, the wavelength becomes strain-dependent [216], and consequently the corresponding wavelength is given by [215]: λ0 λ= (2.2) (1 + pre )(1 + ξ)1/3 where λ0 is the wavelength based on small deformation theory, pre is the pre-strain and a pre-stretch factor of (1 + pre )(1 + ξ)1/3 is included with ξ = 5pre (1 + pre )/32. When pre is constant, the superficial wavelength λ only depends on the thickness of the film. Equation 2.1 suggests the use of film thickness or substrate stiffness to tune wave- length. The former is an attractive option for multilayer GO films, whose thickness can be easily and systematically altered by changing the concentration in the deposi- tion fluid. The results of both methods for wavelength tuning are illustrated in Fig. 2.2. For a given substrate, wavelengths can be changed by an order of magnitude (e.g. 56 3 to 30 µm for the softer substrate) by varying GO film thickness from 20-200 nm through increasing GO suspension concentration from 0.02-0.2 mg/ml. Based on the work of Kunz et al. [183], we anticipate the ability to extend to nanoscale features by using monolayer or few-layer rGO, though it becomes increasingly difficult to fully cover the substrate with such ultrathin films made by tiling microscale GO sheets. The linear relationship between the wavelength and film thickness is consistent with Eq. 2.1, and can be used to estimate the Youngs modulus of the multilayer GO film. Taking pre = 50%, νf = 0.197 [217], νs = 0.48, we obtain from Eqs 2.1 and 2.2: !3 λ Ef = 0.057Es (2.3) hf giving an estimate of the film modulus of 8.6-13.3 GPa, which is similar to published experimental results obtained by other methods [218, 219]. The wavelength of our wrinkle features decreases with increasing pre-stretch in agreement with Eq. 2 (Supp. Info, Fig. 2.6). In dynamic experiments under the optical microscope, we observe that the ridges do not all appear at once during relaxation, but rather in pairs separated by a dis- tance corresponding to the characteristic wavelength. Upon further relaxation the ridge pairs migrate closer together in response to the substrate shrinkage and the characteristic periodic wrinkle structure becomes uniform over large areas. The am- plitude of the sinusoidal wrinkles that appear at small deformation is predicted by q pre theory to be A0 = hf c − 1, while for large defomation, Song et al. proposed [215]: A0 A= √ (2.4) 1 + pre (1 + ξ)1/3 This suggests that both film thickness and pre-stretch can be used to control feature height. Though these relations are for adherent and elastically deformed 57 Figure 2.2: Wavelength tunability through control of multilayer GO film and through substrate selection. (A) SEM images of wrinkled GO films of thickness from 10 to 200 nm on the softer elastomer (modulus 5 · 104 Pa) and (B) their Fourier transform spectra with the dominant spatial frequency shaded. (C) Thickness-dependent wavelengths on softer (black squares) and stiffer elastomer (modulus 6 · 105 Pa; red circles) with linear regression slopes λ/hf shown. The scale bars in (A) represent 50 nm. 58 films, they can be used as an estimate for the delaminated buckle structures observed here. The height and periodicity of these buckled structures was quantified through the use of atomic force microscopy (AFM). A representative scan of a buckled structure formed after a small pre-strain (1.5%) is shown in Fig. 2.3A. The corresponding height profile for these structures shows a characteristic ridge height of 2 µm (Fig. 2.3B). At moderate pre-strains (8.5% or 25%), increased ridge heights of approximately 7 and 12 um were observed, respectively (Supplementary Information, Fig. 2.7). Finally, for large pre-strains (50%), the ridge heights became large enough that characterization with high-tilt SEM was more appropriate than AFM. These ridge heights were 24 µm (Fig. 2.3C), with no correction for foreshortening needed to estimate these values at the high tilt angle used. Overall, pre-strain affects ridge amplitude much more than it affects wavelength (Supplementary Information, Fig. 2.6), so it is an effective and facile means to control feature height. 2.4.2 Thermal reduction for liquid phase stability The data in Figs. 2.1-2.3 are for multilayer GO films in their as-produced, fully- oxidized state. We observed that these films were poorly suited for biological appli- cations due to gradual re-dissolution of GO sheets in cell culture medium as well as instability in the feature heights in response to wetting and drying cycles. Infiltration of aqueous solutions into the delamination cavities between the film and substrate leads to peak collapse by capillary forces when the solution is driven out during the drying process (Supplementary Information, Fig. 2.8). We hypothesize this effect is due to the hydrophilicity of GO, and could be managed by reductive deoxygenation. We used mild thermal reduction by overnight heating at 120◦ C in air, and found the reduced graphene oxide (rGO) films showed good retention of the wrinkle features in GO and also good stability following water immersion and drying (Supplementary 59 Figure 2.3: Pre-strain provides effective control over wrinkle feature height. (A) AFM image of multilayer graphene oxide film after release of 1.5% pre-strain on the softer elastomer substrate. (B) Height profile along line marked in (A). (C) High-tilt SEM image of wrinkled graphene on the soft elastomer substrate with 50% pre-strain. Information, Fig. 2.9). These rGO films were therefore used exclusively in the cell studies below. We were also interested in whether the wrinkled rGO films could be made if the thermal reduction was carried out prior to the relaxation step. We found that relaxing pre-reduced (rGO) films led to a less regular wrinkle pattern and a larger number of small cracks (Supplementary Information, Fig. 2.10). This may reflect stronger adhesion of rGO to the substrate, but was not further investigated, and the higher quality films made by post-wrinkling reduction were used in all of the cell studies. 60 2.4.3 Cell alignment on flat and wrinkled rGO substrates Based on previous studies of cellular interactions with anisotropic grooved substrates, we hypothesized that wrinkled rGO substrates with wavelength λ ∼ 25µm would be sufficient for cell confinement and alignment through contact guidance mechanisms [196, 197, 198, 6]. As a model system, we chose to study murine and human fibrob- lasts on matched samples of 25 µm wavelength and flat rGO substrates. Cell viability was measured on wrinkled and flat rGO substrates compared with conventional tissue culture plastic (polystyrene) using DRAQ7 (Biostatus), a far-red fluorescent DNA dye that only stains dead and membrane compromised cells [220]. These measurements showed that cell viability was consistently over 95% on wrinkled and flat rGO sub- strates (Supplementary Figs. 2.12 and 2.13), indicating excellent biocompatibility in vitro. The angular orientation of NIH-3T3 mouse fibroblasts was then compared on wrinkled (∼25 µm wavelength) and flat rGO substrates after culture for 96 h. To better visualize the cell morphology, both the nuclei and F-actin cytoskeleton were immunostained with fluorescent markers (Fig. 2.4A,B). Qualitatively, NIH-3T3 fi- broblasts appeared highly aligned on the wrinkled substrates (Fig. 2.4A) and ran- domly oriented on the flat substrates (Fig. 2.4B). To quantify this alignment, the cell morphology was automatically extracted using CellProfiler (Broad Institute) based on F-actin immunostaining. This software then determined an orientation based on the major axis of a best-fit ellipse to the cell morphology, which was manually verified. A circular histogram of these results are plotted as a rose plot for wrinkled (n = 164) and flat substrates (n = 137). The coordinate system for the wrinkled rGO was defined so that 0◦ points along the wrinkle patterns (perpendicular to the axis of uniaxial stretch). NIH-3T3 fibroblasts on wrinkled substrates displayed a mean orientation angle of 0◦ with a standard deviation of 15◦ . In contrast, NIH-3T3 fibroblasts on flat substrates displayed a mean orientation angle of 13◦ and a standard deviation of 57◦ , 61 which approximates a random orientation (with 0◦ mean, 52◦ standard deviation). To corroborate these results, primary human fibroblasts derived from neonatal foreskin (Normal Human Fibroblasts, NHF cells) were also cultured on these substrates. NHF cells also displayed qualitatively similar alignment on wrinkled substrates relative to flat substrates (Fig. 2.4C,D). Quantitatively, NHF cells on wrinkled substrates dis- played a mean orientation angle of 0◦ with a standard deviation of 18◦ . In comparison, NHF cells on flat rGO substrates displayed a mean orientation angle of 16◦ with a standard deviation of 53◦ , also approximating a random orientation distribution. The statistical significance of these results was assessed using two metrics. First, Rao’s Spacing Test was used to evaluate the null hypothesis that a given circular dataset is uniformly distributed (Fig. 2.4E). For both NIH-3T3 and NHF cells, the angular orientation on wrinkled substrates was non-uniform (p < 0.01), while the an- gular orientation on flat substrates was uniform (p > 0.10). Next, the Mardia-Watson- Wheeler test was used to evaluate the null hypothesis that two circular datasets have identical distributions. For both NIH-3T3 and NHF cells, the distribution of angular orientations of wrinkled and flat substrates were significantly different (p < 10−12 ). 2.4.4 Cell morphology on flat and wrinkled rGO substrates. The morphology of NIH-3T3 and NHF cells was analyzed by extracting cell shape based on F-actin immunostaining. Briefly, CellProfiler was used to segment fluores- cent nuclei in the blue fluorescent channel, which served as a starting point to segment the surrounding cell body in the red fluorescent channel (Fig. 2.5A). The morphology was then quantified using a number of geometric parameters. First, an average cell width was measured for NIH-3T3 and NHF cells on wrinkled and flat substrates. NIH-3T3 cells on wrinkled substrates displayed an average cell width of 9 µm and standard deviation of 3.3 µm, while cells on flat substrates had an average cell width of 13 µm and a standard deviation of 8.7 µm (Fig. 2.5B). In 62 Figure 2.4: Fibroblast culture on wrinkled graphene materials results in highly aligned cells. (A-D) Fluorescence images of Actin-phalloidin (red) and nuclei (blue) for NIH-3T3 and NHF cells cultured on 200 nm wrinkled s-GO substrates (∼25 µm period) for 96 h (A and C) and on paired flat graphene controls (B and D). Circular histograms of fibroblast orientation, ranging from -90◦ to 90◦ , for cells on wrinkled and flat materials are displayed as orientation rose plots to the right of each corresponding fluorescence image (A-D). (E) Table of statistics summarizing circular standard deviation, and statistical significance for uniformity (one-sample test) and distribution (two-sample test) using Rao’s Spacing Test and Mardia-Watson-Wheeler test, respectively. 63 Figure 2.5: Image analysis of fibroblasts on graphene-based materials demonstrates distinct morphological features for different cell types on wrinkled vs. flat substrates. (A) Process flow for image analysis/cell segmentation: fluorescence images are split into separate channels for each color, CellProfiler outlines cells by distinguishing be- tween pixels in the background and foreground, automated object shape features are calculated. (B-D) Boxplots of quantitative shape features for the fibroblasts were plotted with MATLAB. 25th and 75th percentiles are indicated by the box edges, the median value is displayed as the dividing line of the box, data not considered outliers are contained within the boxplot whiskers, and data points representing individual cell measurements are overlaid as plotted dots (Red boxplots = cells on wrinkled sub- strates, blue boxplots = cells on flat control substrates). (B) Comparison of average cell width for fibroblasts on wrinkled and flat substrates (as in Fig. 2.4) for NIH-3T3 and NHF cells (left and right, respectively). (C) Comparison of eccentricity values for cells on wrinkled and flat substrates for NIH-3T3 and NHF cells (left and right, respectively). (D) Solidity values for NIH-3T3 cells on wrinkled and flat substrates (D, left) and compactness values for NHF cells on wrinkled and flat substrates (D, right). Statistical significance is indicated as follows: p < 0.05; p < 0.01; p < 0.001. 64 addition, NHF cells on wrinkled substrates display an average cell width of 16 µm with a standard deviation of 8.0 µm, while cells on flat substrates had an average cell width of 29 µm with a standard deviation of 10.3 µm (Fig. 2.5B). Overall, both fibroblast lines displayed narrower morphologies with smaller standard deviations on wrinkled substrates compared to flat substrates, with highly significant differences (p < 10−3 ). These narrower cell widths were comparable to the wrinkle spacing of ∼25 µm, suggesting that the elongation reflects some degree of cell micro-confinement in the inter-ridge valley spaces. Next, the relative elongation of these cells was assessed in terms of the eccentricity of the best-fit ellipse to the cell morphology. An eccentricity of 0 corresponds to a circle, while an eccentricity of 1 corresponds to a line. NIH-3T3 cells on wrinkled sub- strates displayed an average eccentricity of 0.921, compared to cells on flat substrates with an average eccentricity of 0.907 (Fig. 2.5C). These values, although similar, dis- play a statistically significant difference (p < 0.05). NHF cells on wrinkled substrates displayed an average eccentricity of 0.947, compared to cells on flat substrates with an average eccentricity of 0.872. Interestingly, NHF cells on flat substrates display con- siderable variability in eccentricity, with a standard deviation of 0.112, compared to 0.069 for NHF on wrinkled substrates. Altogether, fibroblasts are intrinsically quite elongated on flat substrates, but are even more elongated on our wrinkled graphene surfaces, where they have eccentricity values close to 1, and the effect of the wrinkle texture is statistically significant. In order to further elucidate differences across these cell lines and conditions, several additional metrics were considered. Qualitatively, NIH-3T3 fibroblasts may display multipolar, star-like morphologies, particularly on flat substrates. This shape was quantified in terms of the cell solidity metric, which describes an object shape based on the regularity of its boundaries. For example, a solid object with no holes has an area equal to the convex hull and thus has a solidity of 1, while an object with a very irregular boundary has a solidity 65 that approaches 0. NIH-3T3 cells cultured on wrinkled substrates have an average solidity of 0.80 and a standard deviation of 0.12, while cells on flat substrates have an average solidity of 0.69 with a standard deviation of 0.16 (Fig. 2.5D). This data indicates that NIH-3T3 fibroblasts are biased towards bipolar morphologies on wrin- kled substrates, compared to multipolar morphologies with more irregular boundaries on flat substrates. NHF cells (Supplemental Table 2.1) display similar solidities on both wrinkled and flat substrates, which may occur due to a more dominant bipolar phenotype. Finally, NHF cells displayed the greatest variability in radial distance from the object center, known as compactness. NHF cells on wrinkled substrates displayed a mean compactness of 4.8, with a standard deviation of 3.4, and cells on flat substrates have a mean compactness of 2.8, with a standard deviation of 1.7 (Fig. 2.5D, right). On the other hand, NIH-3T3 cells did not display a significant difference for cell compactness between wrinkled and flat substrates. These differences likely arise from differences in cell size between NHF and NIH-3T3 cells. Since NHF cells are larger, they are more confined and elongated on wrinkled substrates but can be more spread out on flat substrates, resulting in larger differences in compactness. In summary, NIH-3T3 and NHF fibroblasts on wrinkled substrates display reduced cell widths and increased eccentricity relative to cells on flat substrates. Furthermore, NIH-3T3 cells on wrinkled substrates display increased solidity, while NHF cells dis- play increased compactness. These results indicate that the topography of these wrinkled graphene substrates can strongly influence cell morphology (summarized in Supplemental Table 2.1). 66 2.5 Conclusions This study demonstrates the feasibility of wrinkled graphene as a surface texturing agent to direct cell alignment and morphology in tissue engineering. Wet deposi- tion of graphene oxide multilayer tiled films on pre-stretched elastomers followed by relaxation and mild thermal reduction produces cell-culture-compatible textured sub- strates with long-range periodic topography in the form of parallel ridges and valleys. Both the spatial period and the wrinkle amplitude can by systematically tuned from 1-25 µm by selection of the substrate stiffness, by variation of pre-strain, and by al- teration of the GO precursor concentration in the feed suspension. Fibroblasts are observed to attach to these textured rGO surfaces and remain viable, while the wrin- kles induce statistically significant cell alignment and elongation relative to flat rGO substrates in a manner consistent with contact guidance on lithographically fabricated nanograting architectures. Together the results suggest that rGO wrinkle engineering is a promising new approach for creation of functional surfaces and scaffolds in tissue engineering. In comparison with competing methods of patterning, wrinkled rGO offers potential advantages in the ease, cost, and scale-up of fabrication over large areas. We envision a modular graphene-based platform for the application of orthog- onal topographic stimuli, chemical stimuli (through functionalization or adsorption), electrical stimuli (enabled by conductive graphene), as well as dynamic mechanical actuation of topographic confinement. Furthermore, we believe these biologically in- spired architectures will be widely applicable to other cell types, including neurons, skeletal, smooth muscle, cardiomyocytes and stem cells. These wrinkled rGO archi- tectures could potentially be applied to soft, stretchable implantable devices such as neural prostheses, cardiac assist devices, catheters or epidermal electronics in order to enhance biocompatibility and biointegration [221, 222]. 67 2.6 Supplementary Information Figure 2.6: Dependence of wavelength λ normalized by small-strain wavelength λ0 and amplitude A normalized by film thickness hf as a function of pre-strain. Lines give behavior predicted by Eq. (2) and (4), which is compared to measurements from this study. Squares: amplitude; triangles: wavelength. Figure 2.7: Tilt-view SEM images showing the average height of the GO wrinkles. Induced by 8.5% pre-strain (A) and 25% pre-strain (B). GO film thickness: 200 nm. 68 Figure 2.8: Top-down and tilt-view SEM images of wrinkled GO films. SEM Images for wrinkled GO films (200 nm thickness and 50% pre-strain) after deposition and drying of one water droplet, showing capillary collapse and reduced feature height compared to 20 µm height in Fig. 2.3C under the same conditions. Figure 2.9: FT-IR of GO film and stabilized GO films prepared by thermal treatment at 120◦ C overnight. 69 Figure 2.10: SEM images of wrinkled s-GO. SEM images of wrinkled s-GO (100 nm thickness and 50% pre-strain) after deposition and drying of one water droplet, showing good stability against capillary forces during water evaporation. Figure 2.11: Thermal reduction preior to relaxtion and wrinkling yields less regular wrinkles. SEM image of less regular wrinkled s-GO (200 nm thickness and 50% pre-strain) that results from carrying out the thermal reduction stop prior to relaxation and wrinkling. 70 Figure 2.12: NIH-3T3 cells cultured on s-GO materials are highly viable. (A) Fluorescence images of nuclei (blue, Hoechst) and dead cells (red, DRAQ7) for NIH-3T3 cells cultured on polystyrene, flat graphene, or wrinkled graphene substrates (∼25 µm period) for 96 hours. White arrows point to representative dead cells. (B) Bar graph of the percentage of viable NIH-3T3 cells on the indicated substrates. The total number of cells counted per condition is indicated by (n); the number of replicates per condition is indicated by (r). Scale bar = 50 µm. 71 Figure 2.13: NHF cells cultured on s-GO materials are highly viable. (A) Fluorescence images of nuclei (blue, Hoechst) and dead cells (red, DRAQ7) for NHF cells cultured on polystyrene, flat graphene, or wrinkled graphene substrates (∼25 µm period) for 96 hours. White arrows point to representative dead cells. (B) Bar graph of the percentage of viable NHF cells on the indicated substrates. The total number of cells counted per condition is indicated by (n); the number of replicates per condition is indicated by (r). Scale bar = 50 µm. 72 Table 2.1: Data Summary for Fibroblast Morphology Characterization. Legend: * p < 0.05; ** p < 0.01; *** p < 0.001; NS Not Significant 73 Chapter 3 Morphological Single Cell Profiling of the Epithelial-Mesenchymal Transition Chapter 3 has been previously published as a primary article: S.E. Leggett, J.Y. Sim, J.E. Rubins, Z.J. Neronha, E.K. Williams, and I.Y. Wong. “Morphological Single Cell Profiling of the Epithelial-Mesenchymal Transition.” Integrative Biology. 11, 1133-1144, Nov 2016 3.1 Abstract Single cells respond heterogeneously to biochemical treatments, which can complicate the analysis of in vitro and in vivo experiments. In particular, stressful perturbations may induce the epithelial-mesenchymal transition (EMT), a transformation through which compact, sensitive cells adopt an elongated, resistant phenotype. However, classical biochemical measurements based on population averages over large num- 74 bers cannot resolve single cell heterogeneity and plasticity. Here, we use high con- tent imaging of single cell morphology to classify distinct phenotypic subpopulations after EMT. We first characterize a well-defined EMT induction through the mas- ter regulator Snail in mammary epithelial cells over 72 h. We find that EMT is associated with increased vimentin area as well as elongation of the nucleus and cytoplasm. These morphological features were integrated into a Gaussian mixture model that classified epithelial and mesenchymal phenotypes with > 92% accuracy. We then applied this analysis to heterogeneous populations generated from less con- trolled EMT-inducing stimuli, including growth factors (TGF-β1), cell density, and chemotherapeutics (Taxol). Our quantitative, single cell approach has the potential to screen large heterogeneous cell populations for many types of phenotypic variabil- ity, and may thus provide a predictive assay for the preclinical assessment of targeted therapeutics. 3.2 Introduction The epithelial-mesenchymal transition (EMT) transforms compact, adherent cells into an elongated, motile phenotype and has been associated with tumor dissemination and drug resistance [223, 224]. During EMT, cell-cell contacts are downregulated (e.g. E-cadherin) and cytoskeletal proteins are upregulated (e.g. vimentin), which promote scattering and local dissemination from the tumor surface [225]. EMT has also been associated with resistance to apoptosis, particularly in the context of con- ventional chemotherapies [226]. One master regulator of EMT is the Snail family of zinc-finger transcription factors, which directly repress E-cadherin transcription [39] and are associated with poor clinical outcome [227]. More generally, aberrant extracellular stimuli from the microenvironment such as growth factors (e.g. TGF- β1) [228] or sublethal stresses (e.g. cytotoxic treatments) [229] can trigger various 75 EMT pathways and associated programs of phenotypic changes [230]. Such pheno- typic plasticity is challenging to measure using classical biochemical assays based on population averages at endpoints. Instead, single cell measurements are essential to reveal distinct subpopulations or rare and exceptional phenotypes [231, 232]. These deeper insights into heterogeneity may improve the predictive capability of preclinical drug testing and guide the design of targeted therapies [233]. High-content screening (HCS) is a promising approach to quantify heterogeneous single cell morphology through optical fluorescence microscopy [234]. HCS incorpo- rates automated segmentation of cellular and subcellular features from digital im- ages, enabling a complete readout of all single cell behaviors within a population, including outliers and infrequent events [235]. This approach can elucidate the in- trinsic heterogeneity in a given cell population and reveal the effects of controlled perturbations, based on immunostaining of fixed cells or fluorescent live cell reporters [236, 237, 238, 239, 240, 241, 242, 243, 244, 245, 246]. Moreover, HCS has been applied to screen candidate anticancer compounds against cell lines in a multiwell plate format [239, 244, 245, 247, 248]. Several studies have recently characterized the effect of inhibitors on epithelial or mesenchymal cell lines [244], and EMT has been used to calibrate HCS analyses [246, 249]. However, HCS has not been applied for longitudinal studies of EMT in response to defined stimuli such as growth factor or chemical stimulation. Such quantitative measurements of single cell heterogeneity are crucial given previous observations of phenotypic plasticity that occur spontaneously [250] and may be amplified in response to sublethal drug treatment [251]. Indeed, drug response can vary with cell plating density [252], which may be a consequence of EMT. For instance, basal epithelial cells such as MCF-10A have been observed to display mesenchymal-like phenotypes in low-density cultures, measured through transcriptional profiling or fluorescent reporters [253, 254, 83]. At increasing densi- ties, the cell population tends to be more epithelial, suggesting some cells may revert 76 back from a mesenchymal phenotype. Single cell profiling could elucidate these den- sity dependent phenotypic changes, which may occur in physiological or pathological contexts. In this article, we measure single cell morphology and biomarker expression dur- ing EMT induction. First, we apply a highly controlled EMT stimulus through an inducible Snail-1 construct in human mammary epithelial cells, which results in a dra- matic elongation of subcellular nuclear, vimentin, and cytoplasmic features. These quantitative changes in shape features define a reference model to classify the com- bined population of control and induced EMT cells as a mixture of two phenotypic subpopulations. Next, this reference model is applied to classify heterogeneous pop- ulations undergoing EMT at varying time points after Snail-1 induction, treatment with TGF-β1, and at varying densities. Finally, we show that treatment with the microtubule inhibitor Taxol can exacerbate EMT. This overall approach establishes a general framework for reconstructing EMT and single cell heterogeneity in the context of biochemical perturbations. 3.3 Methods 3.3.1 Cell Culture Human mammary epithelial cells stably transfected with an ER-Snail-16SA retrovi- ral construct (MCF-10A Snail) as well as the parental cells (MCF-10A) were a gift from D.A. Haber (Massachusetts General Hospital) [255]. The Snail-16SA variant is refractory to phosphorylation and is thus stably expressed and localized in the nucleus, where it initiates EMT induction [256]. Both MCF-10A variants were cul- tured following Brugge and coworkers [257]. Cells were plated at low density (500 cells/well, unless specified) in 96-half-well high content imaging microplates (Fisher Scientific) coated with 5 µg/cm2 fibronectin (Sigma-Aldrich). After settling and ad- 77 hering overnight, cells were subjected to exogenous stimuli or drugs as indicated. Special care was taken to ensure that cells remained subconfluent at the conclusion of the experiments, unless noted. T-47D breast carcinoma cells were acquired from the Developmental Therapeutics Program (DTP) of the National Cancer Institute, cultured in RPMI 1640 (ATCC) supplemented with 10% Fetal Bovine Serum (Fisher Scientific), 0.2U/mL bovine insulin (Sigma-Aldrich), 100 units/mL penicillin, and 100 µg/mL streptomycin, and were passaged following standard ATCC protocol. MDA- MB-231 cells were purchased from ATCC and were cultured in DMEM containing L-Glutamine, L-glucose, and sodium pyruvate (Fisher Scientific), and supplemented with 10% FBS (Fisher Scientific), 100 units/mL penicillin, and 100 µg/mL strepto- mycin. T-47D and MDA-MB-231 cells were cultured for 96 h in high content imaging plates as described and plated with 3,000 and 1,500 cells/well, respectively. These higher seeding densities with respect to MCF-10A experiments were experimentally determined, and compensate for differences in the characteristic doubling times of each cell line. 3.3.2 EMT Induction with Snail-1 or TGF-β1 Snail-1 expression was induced in MCF-10A Snail cells through the addition of 4- hydroxytamoxifen (OHT; Sigma-Aldrich) resuspended in DMSO and used at a final concentration of 500 nM in media. To induce EMT to varying extents at the popu- lation level, a time course of Snail-1 induction was prepared with four conditions, all over a course of 72 h. The durations of OHT exposure included 0, 24, 48, and 72 h, where DMSO (0.05% final) was applied during the remainder of the 72 h time course. Separately, EMT was induced in MCF-10A cells through treatment with 5 ng/mL recombinant human TGF-β1 (R&D Systems) in growth media. A comparable time course was conducted with four conditions over 72 h. All conditions were initially treated with growth media and TGF-β1 was added to different wells for durations of 78 0, 24, 48 and 72 h. For both EMT induction experiments, media was replenished at the 48 h time point, to prevent nutrient deprivation and growth factor depletion. 3.3.3 Density Dependent Induction of EMT MCF-10A cells were plated in high content imaging plates, as described, where 375 cells/well was chosen as the lowest seeding density. This density was first validated during routine culture (equivalent to 1:20 split) as it maintained the ability to pro- duce characteristic epithelial monolayers several days after passaging cells. Thus, to generate a range of terminal densities, cells were plated at either 375 or 500 cells/well (the typical density for all other experiments) and subsequently cultured for either 72 or 96 h before fixation and immunostaining. 3.3.4 Drug Treatment with Taxol for EMT Induction MCF-10A Snail cells were cultured in media with OHT for 72 hours to induce a mesenchymal phenotype (preinduced) or with DMSO to maintain an epithelial phe- notype (uninduced). Both preinduced and uninduced cells were seeded at a density of 750 cells/well, allowed to adhere to fibronectin-coated multiwell plates overnight as described, and the growth media containing OHT or DMSO was removed. Both cell types were then treated with either DMSO (0.05%, control) or 4 nM paclitaxel (Taxol) for 48 hours, and subsequently fixed and immunostained. 3.3.5 Immunostaining and Fluorescent Imaging At the conclusion of time course experiments, cells were fixed for 20 minutes at 4C with 4% paraformaldehyde in 1X PBS (all solutions are in 1X PBS unless specified). Cells were permeabilized with 0.1% Triton X-100, washed, and blocked with 10% goat serum. Cells were washed with sodium acetate buffer for 15 minutes and then 79 rinsed several times with 1% nonfat dry milk. Cells were incubated overnight at 4C with primary antibodies: 250 µg/mL E-cadherin (Fisher Scientific) diluted at 1:500 and vimentin (Cell Signaling Technology) diluted at 1:200 in 1% milk. Cells were then washed with 1% milk and incubated in the dark at room temperature with suitably matched secondary antibodies: 2 mg/mL Alexa Fluor 488 and Alexa Fluor 555 (Invitrogen) diluted at 1:500 in 1% milk. Lastly, cells were washed with 1X PBS, incubated with 2 µg/mL Hoechst Pentahydrate (Invitrogen) and 2 µg/mL HCS CellMask Deep Red (ThermoFisher Scientific) for 30 minutes at room temperature, and washed again. Immunostained cells were imaged using an inverted epifluorescence microscope (Nikon TiE). Images were acquired with a 20X Super Plan Fluor objective (NA = 0.45, extra long working distance) and with a 14-bit resolution sCMOS camera (Andor Neo). Fluorescence illumination was provided by a light-guide coupled Lumencore Sola white light excitation system. Care was taken to ensure all images were acquired with identical acquisition parameters (exposure time, camera gain/gamma control, and microscope aperature settings). Post-acquisition, fluorescence thresholds for each experiment were set for qualitative comparison across conditions using built-in NIS Elements AR settings to eliminate background noise and reduce pixel saturation. Detailed settings are described in Supporting Information. 3.3.6 Image Analysis Cell detection and shape feature analysis were performed using CellProfiler 2.1 (Broad Institute) [209]. First, the uneven illumination was corrected across channels and image sets. Next, fluorescently labeled nuclei (DAPI channel) were segmented as primary objects. Based on the nuclei positions, fluorescently labeled vimentin and cytoplasm were segmented as separate secondary objects. These fluorescent objects were manually verified and corrected as needed. Detailed segmentation parameters 80 are described in the Supplementary Information (Table 3.1, 3.2). Finally, shape measurements were extracted from the segmented objects, particularly nuclear max radius, vimentin area, cytoplasm form factor, and cytoplasm max feret diameter. 3.3.7 Phenotype Classification using a Gaussian Mixture Model (GMM) An initial training set was developed by segmenting cells treated either with DMSO or OHT for 72 h, corresponding to putative epithelial and mesenchymal phenotypes. For each condition, CellProfiler was used to segment >100 cells for 15 nuclear, vimentin, and cytoplasmic shape metrics, which were rescaled between 0 and 1 for consistent comparison (see Supplementary Information for additional details, Fig. 3.6-3.9, Ta- bles 3.1-3.2). Next, the single cell features from both conditions were combined into a single dataset and an expectation maximization (EM) algorithm was used to de- termine maximum likelihood estimates of the parameters for a GMM [258] assuming two subpopulations (gmdistribution.fit, MATLAB R2013b). Based on the 15 metrics, all possible combinations of metrics (up to 5) were used to train the GMM. This classifier was then tested against a second dataset of control and induced cells pre- pared under consistent experimental conditions. A third dataset based on epithelial (T-47D) and mesenchymal (MDA-MB-231) cell lines were also tested. An optimized 4 metric GMM distribution was used to partition all subsequent datasets into ep- ithelial and mesenchymal subpopulations, respectively. For visualization purposes, the phenotypic distributions are displayed by normalized cytoplasmic maximum feret diameter, which is highly predictive. 81 3.4 Results 3.4.1 Profiling Epithelial and Mesenchymal Phenotypes after Snail Induction To establish a training set of epithelial and mesenchymal phenotypes, a well-defined EMT stimulus was applied through the master regulator Snail. Human mammary epithelial cells (MCF-10A Snail) transfected with a constitutively active variant of Snail-1[256] fused to an estrogen receptor response element were exposed to 4- hydroxytamoxifen (OHT) for 72 h. Previous work using this inducible Snail construct has demonstrated that 72 h treatment with OHT results in nearly complete EMT in the population based on cell morphology and biomarker expression levels [255]. Based on these results, cells were treated with either DMSO or OHT for 72 h, followed by an additional 72 h of DMSO treatment to generate pronounced differences in control and induced populations, respectively (Supplementary Information, Fig. 3.9). Cells from both conditions were then fixed for immunofluorescent staining of the nucleus (Hoechst), epithelial (E-cadherin) and mesenchymal (vimentin) biomarkers, and cytoplasmic morphology (CellMask). Subsequently, widefield fluorescence mi- croscopy revealed that cells in the control condition (DMSO) displayed a cobblestone morphology with relatively compact nuclei, vimentin, and cytoplasm, as well as enhanced E-cadherin expression at the edges (Fig. 3.1A). In contrast, cells in the induced condition (OHT) displayed a spindle-like morphology with more elongated nuclei, vimentin, and cytoplasmic features, as well as nearly nonexistent E-cadherin expression (Fig. 3.1A). These dramatic differences in cell morphology and biomarker expression are consistent with an epithelial phenotype for the control condition (DMSO) and a mesenchymal phenotype for the induced condition (OHT). These qualitative changes in morphology and biomarker expression were then quantified at the single cell level through automated object segmentation and mea- 82 surement (Fig. 3.1B, Tables 3.1-3.2, and Supporting Information). CellProfiler was used to identify single cells based on their nuclei, which were then used to locate the corresponding vimentin and cytoplasmic features. Based on these segmented nuclear, vimentin, and cytoplasmic features, a total of 44 metrics were extracted, including 15 notable shape features (Supplementary Information,Fig. 3.7A, 3.7B). These met- rics included fluorescence intensities, which were somewhat variable across repeated experiments, perhaps due to antibodies [259] and slight variations in imaging con- ditions [260, 261]. For instance, cells displayed substantial differences in integrated vimentin intensity for control (DMSO) and induced EMT (OHT) conditions, but this was largely due to changes in area, since the mean and median vimentin intensities were relatively similar (Supplementary Information, Fig. 3.7C). Similarly, E-cadherin quantification proved to be challenging due to its inconsistent or minimal expression after EMT, so it was only used qualitatively for confirmation of cell phenotype. Thus, we focused primarily on shape metrics, which are standardized and should be broadly applicable across experimental conditions. Well over 100 cells were analyzed per con- dition, allowing good sampling of the statistical distribution associated with each phenotype. Gaussian mixture models (GMM) were fit to different training datasets based on all possible combinations of single cell metrics (up to five) combined from the con- trol (DMSO) and induced (OHT) conditions. An optimal combination of 4 metrics was selected: nuclear maximum radius, vimentin area, cytoplasmic form factor, and cytoplasmic maximum feret diameter (Supplementary Information, Fig. 3.10). This classifier was then tested on a separate dataset with comparable experimental con- ditions. For the control (DMSO) condition, 92% of cells were classified as epithelial. Instead, for the induced (OHT) condition, 96% of cells were classified as mesenchymal (Fig. 1C, Supplementary Information Fig. 3.5A, 3.5B). This discrepancy between the experimental condition and phenotype classification may accurately represent the het- 83 erogeneity in each experimental condition. For instance, MCF-10A cells can undergo a spontaneous but transient EMT [253, 254], which could explain why the control (DMSO) condition had a small percentage of cells which were classified as mesenchy- mal (Supplementary Information, Fig. 3.11C). In addition, we observed that some cells in the induced (OHT) condition displayed smaller nuclei and vimentin features due to recent division, which could explain why these cells were classified as epithelial (Supplementary Information, Fig. 3.11D). The classifier was then tested against cell lines with established epithelial or mes- enchymal phenotypes. In particular, T-47D is a human breast carcinoma line with a strong luminal epithelial phenotype, and a biomarker profile distinct from basal and fibroblast phenotypes [241]. T-47D cells displayed compact, rounded morphologies, and formed epithelial clusters with prominent E-cadherin expression and lack of vi- mentin expression (Supplementary Information, Fig. 3.12A). Instead, MDA-MB-231 is a highly metastatic human breast adenocarcinoma line with a highly mesenchymal phenotype. MDA-MB-231 cells were dispersed in culture with elongated, spindle- like morphologies, as well as strong vimentin expression, and absence of E-cadherin expression (Supplementary Information, Fig. 3.12A). For the same combination of four metrics, the GMM classifier displayed 100% agreement for actual and predicted T-47D, as well as 95% for actual and predicted MDA-MB-231 (Supplementary Infor- mation, Fig. 3.12B, 3.12C). This improved accuracy is consistent with the phenotypic homogeneity that would be expected from these two cell lines. Since T-47D did not exhibit noticeable vimentin expression, the accuracy of a classifier based on only nuclear maximum radius, cytoplasmic form factor, and cytoplasmic maximum feret diameter was also evaluated. This reduced set of 3 metrics was highly accurate for T-47D cells, with 99% agreement (Supplementary Information, Fig. 3.12D). How- ever, the absence of vimentin area in this reduced set was significantly worse for classifying MDA-MB-231 cells, with only 63% agreement. Further reduction to two 84 metrics for classification revealed notable combinations including vimentin area and cytoplasmic max feret diameter, which resulted in slightly increased accuracy for ep- ithelial phenotypes (MCF-10A DMSO, T-47D), but slightly decreased accuracy for mesenchymal phenotypes (MCF-10A OHT, MDA-MB-231) (Supplementary Informa- tion, Fig. 3.13). In general, combinations of vimentin and cytoplasmic metrics were good predictors of EMT phenotype, with 93-100% accuracy for each epithelial and mesenchymal phenotype. However combinations of nuclear and vimentin or nuclear and cytoplasmic were slightly worse predictors, with 90-100% and 87-94% accuracy per condition, respec- tively. Further, nuclear, vimentin, or cytoplasmic metrics alone were even worse predictors with accuracy ranges including 61-88%, 87-97%, and 86-95%, respectively. Thus, through a process of combinatorial optimization, a set of 4 metrics (nuclear max radius, vimentin area, cytoplasm form factor, cytoplasm max feret diameter) was chosen for evaluating subsequent experiments and demonstrates the applicability of GMM for cell classification across cell lines. The high accuracies achieved for dis- tinguishing between epithelial and mesenchymal subpopulations reflect the dramatic differences in the chosen metrics across phenotypes. For instance, between the control and induced MCF-10A cells, there was a four-fold increase in mean vimentin area and a two-fold increase in mean cytoplasmic max feret diameter. Similarly, between the epithelial T-47D and mesenchymal MDA-MB-231 cells, there was a five-fold increase in mean cytoplasmic max feret diameter, as well as a 33% decrease in cytoplasmic form factor. Thus, this single cell profiling can detect subtle variations within a popula- tion, but these do not affect the accuracy of probabilistic assignment to epithelial and mesenchymal classes. Overall, this analysis establishes a robust classification scheme for epithelial and mesenchymal phenotypes based on biomarker and morphology. 85 3.4.2 Snail Induction Drives Rapid EMT over 72 h Next, this classification scheme was applied to reconstruct single cell phenotypes during a time course of Snail-mediated EMT induction over 72 h. MCF-10A Snail cells were cultured in four different conditions with varying OHT treatment times over 72 h. For clarity, each condition is denoted by the duration of OHT exposure, e.g. 0 h (control), 24 h, 48 h and 72 h, where the remainder of the 72 h experiment corresponds to treatment with DMSO (Supplementary Information, Fig. 3.14A). Between 0 h and 24 h, a noticeable loss of E-cadherin was visible, although the nuclei and vimentin remained relatively similar in size and shape (Fig. 3.2A). From 24 h to 48 h, the cytoplasmic and vimentin features became dramatically elongated with brighter vimentin intensities and there was some subtle elongation of the nuclei. This trend continued between 48 h and 72 h, although the differences in morphology were less dramatic. Overall, these trends are consistent with previous bulk measure- ments of Snail-mediated EMT induction, which showed an initial loss of E-cadherin and other epithelial markers, followed by a rapid gain of vimentin and mesenchymal markers [255]. At the single cell level, the quantitative changes in segmented nuclear, vimentin, and cytoplasmic shape features were consistent with the qualitative trends. In particular, vimentin and cytoplasmic associated features (vimentin area and cy- toplasm max feret diameter) remained relatively consistent between 0 h and 24 h, increased dramatically between 24 h and 48 h, then remained relatively consistent between 48 h and 72 h. Instead, nuclear maximum radius displayed an increase as early as 24 h, but did not continue to change appreciably at later time points. In addition, cytoplasmic form factor (object roundness) remained unchanged between 0 h and 24 h, but gradually decreased from 24 to 72 h (Supplementary Information, Fig. 3.15). Next, these morphological features were used to classify cells from each condi- tion into epithelial (E) and mesenchymal phenotypes (M). Initially at 0 h and 24 86 h, the population was almost entirely epithelial (92%) (Fig. 3.2B, Supplementary Information Fig. 3.14B). At 48 h, there was a dramatic shift in the distribution of phenotypes, such that the population was now primarily mesenchymal (83%). At 72 h, this distribution remained comparable and was dominated by the mesenchymal phenotype (87%) (Fig. 3.2B, Supplementary Information Fig. 3.14B). Altogether, these results suggest that Snail-mediated EMT induction is rapid with relatively con- sistent kinetics across single cells. Further, 72 h OHT treatment followed by 72 h withdrawal of the stimulus (training and test conditions) yielded a 96% mesenchymal population. This result suggests that 72 h of OHT treatment is sufficient to induce a stable mesenchymal phenotype, likely sustained through epigenetic modifications (Supplementary Information, Fig. 3.14C, 3.14D) [255]. Further examination of the posterior probability distribution revealed that 93-96% of the population classified as either epithelial or mesenchymal phenotype with high probability (Supplementary Information, Fig. 3.16). This serves as an effective validation for this GMM clas- sification scheme, which will subsequently be applied to quantify EMT induction in response to less defined biochemical and environmental stimuli. 3.4.3 TGF-β1 Induction Drives Gradual EMT over 72 h The cytokine transforming growth factor TGF-β1 is well-known to induce EMT, al- though the cellular response can vary in context and remains poorly understood [228]. Since the kinetics of EMT induction through TGF-β1 are likely to be less consistent from cell to cell, the use of single cell profiling may yield new insights that cannot be ascertained from population-averaged measurements. To characterize TGF-β1 induced EMT, wild type MCF-10A cells were cultured in four different conditions with varying treatment times (5 ng/mL TGF-β1), again ranging from 0 h to 72 h (Supplementary Information, Fig.3.17A). 87 Between 0 h and 24 h, there was a noticeable, but incomplete loss of E-cadherin at the cell edges with some elongation and overexpression of vimentin (Fig. 3.3A). From 24 h through 72 h, the E-cadherin at the cell-cell junctions was largely internalized, while vimentin was brighter and more elongated. In comparison to the previous Snail induction, this TGF-β1 induction resulted in a widened rectangular shape, with less elongated vimentin and cytoplasmic features (Fig. 3.2A, 3.3A). Overall, these quali- tative trends indicate that TGF-β1 induction results in more variable kinetics at the single cell level relative to Snail induction through OHT. The quantitative single cell metrics also showed more gradual changes from 0 h to 72 h with TGF-β1, instead of the rapid EMT observed with Snail-induction from 24 h to 48 h. In particular, the vimentin area and cytoplasm max feret diameter displayed a gradual increase as early as 24 h, which continued to rise over 48 h and 72 h (Supplementary Information, Fig. 3.18). Additionally, there was an obvious decrease in cytoplasmic form factor from 0 to 24 h, which remained low across 48 and 72 h, relative to the 0 h condition. On the other hand, nuclear max radius did not change appreciably across TGF-β1 induction conditions, thus representing a subtle distinction between the morphologic changes that take place for Snail vs. TGF-β1 mediated EMT induction (Supplementary Infor- mation, Fig. 3.18). At the single cell level, it should be noted that the morphological changes are subtler with TGF-β1 induction relative to Snail induction, so that the overall distributions of vimentin, and cytoplasmic associated features display less of a shift through 72 h. The smaller shift of the statistical distribution may arise from the less elongated and more teardrop shaped vimentin features at the single cell level. The classification of these single cell morphological features into epithelial and mesenchymal phenotypes further corroborates with these trends. At 0 h, the pop- ulation was primarily epithelial (97%) (Fig. 3.3B, Supplementary Information, Fig. 3.17B). However, at 24 h, some of the population underwent EMT, resulting in a di- minished epithelial subpopulation (68%) and increased mesenchymal subpopulation 88 (31%). This trend continues at 48 h, with a further decrease in epithelial subpop- ulation (34%) and increase in mesenchymal subpopulation (65%). By 72 h, there is an additional shift in the population, with the fewest epithelial cells (23%) and a mostly mesenchymal subpopulation (76%) (Fig. 3.3B, Supplementary Information, Fig. 3.17B). This incomplete EMT at the population level after 72 h is consistent with flow cytometry measurements reported elsewhere, which show that complete loss of epithelial biomarkers and gain of mesenchymal markers requires 7-15 days, the kinetics of which are more rapid with increasing dosage [262]. Thus, to examine whether or not an initial TGF-β1 exposure stimulates long term EMT induction, TGF-β1 was pulsed for 72 h, then withdrawn for an additional 72 h (Supplemen- tary Information, Fig. 3.17C). Interestingly, 72 h withdrawal of TGF-β1 after 72 h treatment yielded a primarily mesenchymal population (97%) (Supplementary In- formation, Fig. 3.17D). These results are in agreement with OHT pulse/withdrawal experiments, suggesting that 72 h exposure to either stimulus (Snail induction via OHT or TGF-β1) is sufficient to initiate an EMT program that is self-sustained over a 72 h period of withdrawal from the stimulus (Supplementary Information, Fig. 3.14C, 3.14D, 3.17C, 3.17D) [255]. Further examination of the posterior probability distribution revealed that 90% of the treated populations classified as either epithe- lial or mesenchymal with high probability, but there was an appreciable 10% where classification was less definitive (Supplementary Information, Fig. 3.19). Thus, the GMM used here was able to reveal distinct differences in EMT induction between the potent Snail stimulus, displaying rapid EMT kinetics, and the exogenous application of the growth factor TGF-β1, displaying a more gradual EMT induction. 89 3.4.4 MCF-10A Cells Exhibit Plasticity and Undergo EMT in Subconfluent Cultures Cell plating density has been observed to affect epithelial phenotype, as well as prolif- eration rate and drug sensitivity. Density-dependent effects are challenging to resolve using conventional bulk assays, but may be evaluated using single cell profiling. To characterize the role of cell density on EMT, wild type MCF-10A cells were cul- tured under four different conditions to yield varying levels of confluency ranging from 15% to 95% (Supplementary Information, Fig. 3.20A, 3.20B). Cells at high confluency (95%) formed a monolayer with the characteristic epithelial phenotype including strong E-cadherin expression at cell-cell junctions, low levels of vimentin, and cuboidal shaped cells (Fig. 3.4A). As cell density decreased, cells became elon- gated with a gradual loss of E-cadherin at cell junctions, an increase in internalized E-cadherin, and a noticeable upregulation of vimentin expression (Fig. 3.4A). Fur- thermore, single cell segmentation revealed that vimentin area, cytoplasm maximum feret diameter, and nuclear maximum radius all increased considerably from 95% to 70% confluence (Supplementary Information, Fig. 3.21). At the population level, these vimentin and cytoplasm morphologic features were relatively consistent across the subconfluent densities (70% and 50%), while few cells at 50% confluence had even larger vimentin areas, suggesting a slight shift towards more EMT at the lower density. Finally, the sparse confluency condition represented the most extreme in- crease in vimentin area, cytoplasm maximum feret diameter, and nuclear maximum radius relative to higher density conditions (Supplementary Information, Fig. 3.21). Surprisingly, there was no clear trend for cytoplasmic form factor, which may result from the more boxy morphologies. Classification of epithelial or mesenchymal phenotypes by GMM revealed that the high confluency condition ( 95%) was classified as primarily epithelial (92%) (Fig. 3.4B). However, at less confluent densities, an appreciable number of cells were clas- 90 sified as mesenchymal. In particular, cells at 70% confluency were classified as 36% mesenchymal. Similarly, cells at 50% confluency were classified as 41% mesenchymal. Finally, at sparse coverage ( 15%), approximately 43% of the population was mes- enchymal. This comparable percentage of mesenchymal cells may result from local aggregation, so that the effective local density is similar despite differences in overall density. It is also noteworthy that the statistical distribution of these two subpop- ulations remained relatively narrow across these conditions, particularly in terms of their cytoplasmic maximum feret diameter. Nevertheless, at sparse coverage (15%), the mesenchymal subpopulation displayed significant kurtosis with a heavy right tail, indicative of more extreme mesenchymal phenotypes. This result was corroborated by the distribution of posterior probabilities, since 11-14% of the population was less definitively classified, relative to the 86-89% of the population classified as epithelial or mesenchymal with high probability (Supplementary Information, Fig. 3.22). Over- all, these results indicate that MCF-10A cells can display less epithelial phenotypes at subconfluent densities. Thus, during the course of cell culture experiments, as MCF-10A cells progress from sparse to subconfluent to confluent densities, the statis- tical distribution of phenotypes will initially be more mesenchymal before reverting to more epithelial. Thus, EMT can occur transiently in culture even in the absence of other exogenous stimuli. Such a transient EMT is likely to be advantageous for epithelial wound healing, permitting a rapid migratory response from cells at high density to damaged areas with diminished cell density [83]. These results are thus likely to be relevant for investigations of directed epithelial cell migration both in vitro and in vivo. 91 3.4.5 Sublethal Taxol Enhances EMT in Uninduced and In- duced Snail Populations More generally, EMT activation is associated with stressful microenvironmental con- ditions, including exposure to sublethal doses of cytotoxic drugs [229]. In particular, it has been demonstrated that the conventional chemotherapeutic paclitaxel (Taxol) is capable of inducing EMT and enhancing metastatic potential [263]. Further, even short term, 24 h exposures to Taxol have been demonstrated to induce EMT and stem-like properties via Snail-1 in various breast cancer cell lines [264]. Such a per- turbation is likely to result in a highly heterogeneous response at the single cell level, which can be sensitively profiled with this technique. To measure the effect of sublethal chemotherapeutics on morphological phenotype, MCF-10A Snail cells were treated for 72 h with DMSO or OHT, followed by treatment for 48 h with DMSO or 4 nM Taxol (Supplementary Information, Fig. 3.23A). For control epithelial cells (72 h DMSO), subsequent treatment with sublethal Taxol resulted in a visible loss of E-cadherin and a dramatic increase in vimentin area with brighter intensity for a subset of cells in the population (Fig. 3.5A, Supple- mentary Information Fig. 3.25). The nuclei now displayed a wide range of aberrant and distorted morphologies, while the vimentin organization displayed characteristic bundling and tangling morphologies (Fig. 3.5A). The corresponding single cell metrics showed a pronounced shift in cytoplasm-associated features, especially a significant increase in cytoplasm max feret diameter and decrease in cytoplasm form factor for a subset of cells in the population. The average nuclear max radius decreased slightly with Taxol treatment, but the variability increased noticeably. In addition, the vi- mentin area increased markedly with Taxol treatment, and displayed a much wider distribution than the control (Supplementary Information, Fig. 3.24). These trends are in agreement with the morphological classification by GMM. In particular, the uninduced control condition (DMSO/DMSO) as primarily epithelial (97%), while the 92 uninduced Taxol condition was largely mesenchymal (72%) with a smaller epithelial (27%) subset (Fig. 3.5B, Supplementary Information Fig. 3.23B). For preinduced EMT cells (72 h OHT), subsequent treatment with sublethal Taxol further exacerbated the mesenchymal phenotype, with highly elongated vimentin and cytoplasmic features (Fig. 3.5C). The corresponding single cell metrics showed further substantial increases in vimentin area and cytoplasm max feret diameter, as well as a significant decrease in cytoplasm form factor. The average nuclear max radius did not change appreciably with the addition of taxol, however, the distribution widened, indicating the nuclear variability that results from taxol treatment (Supplementary Information, Fig. 3.24). These induced Taxol conditions correspond to single cell morphologies that are comparable to, or even more pronounced than the 72 h OHT induction of Snail (Supplementary Information, Fig. 3.15). Classification of epithelial or mesenchymal phenotype by GMM showed that the preinduced control condition (OHT/DMSO) was distributed as 12% epithelial and 87% mesenchymal (Fig. 3.5D). Similarly, the preinduced Taxol condition (OHT/Taxol) was primarily mesenchymal (86%) with some residual epithelial phenotype (13%). Indeed, the statistical distri- bution of the mesenchymal subpopulation in the induced Taxol condition displays a relatively wide variance (Fig. 3.5D, Supplementary Information Fig. 3.24), likely resulting from an overall increase in the vimentin and cytoplasm-associated features. Biologically, sublethal Taxol treatment is likely to stress cells to activate EMT, confer- ring drug resistance, but also to enhance the stability of polymerized vimentin [263]. It should be noted that the overall cell numbers are noticeably diminished relative to untreated controls (Fig. 3.5A,B). Although 4 nM Taxol is sublethal, there may be some anti-proliferative effects on a fraction of the subpopulation (Supplementary Information, Fig. 3.25). Moreover, this treatment may be driving some phenotypic selection. As a consequence, the variability of the nuclear, vimentin, and cytoplasmic morphologies results in a very wide distribution of mesenchymal phenotypes relative 93 to previous induction techniques. This result was corroborated by the distribution of posterior probabilities, where 94-98% of the population classified as epithelial or mes- enchymal with high probability (Supplementary Information, Fig. 3.26). Altogether, these results indicate that Taxol treatment is sufficient to induce EMT and further enhance EMT when applied in combination with endogenous stimuli. 3.5 Discussion and Conclusions EMT has been hypothesized to drive phenotypic heterogeneity and plasticity in can- cer cell populations, which can be challenging to measure using population-averaged assays at endpoints. Nevertheless, cancer cells often display characteristic patterns of functional phenotypes, suggesting that a reductionist approach could be sufficient to capture tumor complexity. For instance, the overall population could be approx- imated as a weighted mixture of subpopulations with distinct phenotypes [231]. In principle, the statistical distributions of these subpopulations could be inferred by measuring large numbers of single cell phenotypes at discrete snapshots in time [233]. Our probabilistic approach classifies a heterogeneous population with a Gaussian mixture model based on two phenotypes: epithelial and mesenchymal. These two extreme states were based on well-defined training phenotypes from controlled Snail induction. In contrast, EMT induction through TGF-β1, varying plating density and Taxol resulted in greater variability in activation kinetics and cell morphology. Indeed, EMT induction with TGF-β1 and plating density resulted in less elongated morphologies, while Taxol treatment resulted in highly aberrant morphology. Nev- ertheless, further examination of the posterior probability distribution revealed that at least 90% of cells could be classified as epithelial or mesenchymal with very high probability, even with sample sizes of hundreds of cells. Interestingly, a small sub- population was classified less definitively, reaching 10% for TGF-β1 treatment and 94 subconfluent density. It has been recently been demonstrated that partial EMT phe- notypes can occur with partial loss of both E-cadherin and vimentin after treatment with moderate concentrations of TGF-β1 [262]. In addition, circulating tumor cells in breast cancer can display a spectrum of epithelial to mesenchymal phenotypes, based on the coexpression of epithelial and mesenchymal transcripts (from RNA-in situ hybridization) [165]. We interpret our results conservatively to avoid overfitting with too many phenotypes, but it is conceivable that this subpopulation represents a partial EMT phenotype. Further work is needed to define a partial EMT pheno- type as well as experimental conditions where it may occur. Overall, we envision our approach based on morphology can be widely utilized by researchers with different antibodies and instrumentation, as well as complementing absolute measurements of molecular biomarkers. More generally, improved analyses for single cell heterogeneity will help to in- terpret preclinical drug testing and predict patient therapies. Currently, the use of large panels of cell lines to systematically identify biomarkers of drug sensitivity is performed at the population level [265]. Nevertheless, sublethal dosing can actually exacerbate drug resistance, perhaps through EMT [229]. The use of HCS to measure single cell and subpopulation behaviors may yield new insights into the emergence of heterogeneity and the efficacy of combinatorial or sequential therapeutic regimens. A complementary approach is to reconstruct phenotypic heterogeneity and plasticity by lineage tracing of single cells over extended periods [266]. Live cell imaging using fluorescent proteins and reporters could more sensitively resolve the kinetics of EMT, as well as the associated consequences for migration and proliferation [89]. Finally, qualitative analysis of cell morphology is already a standard practice to interpret tumor histology. Cells with elongated morphology are frequently observed in malig- nant tumor sections from human patients, although they cannot be unambiguously distinguished from stromal cells with mesenchymal biomarkers (e.g. fibroblasts) [35]. 95 Understanding the extraordinary intratumoral diversity that occurs in vivo due to genomic instability and the aberrant tumor microenvironment represents a funda- mentally important challenge [267]. Future work will examine the applicability of our classifiers to analyze live cell imaging, 3D culture as well as tissue histology. In conclusion, we have demonstrated a probabilistic classification scheme for EMT based on single cell morphology and biomarkers. This approach was validated with controlled Snail induction and identified distinct epithelial and mesenchymal pheno- types associated with dramatic elongation and overexpression of vimentin, as well as some elongation of the nuclei. This analysis was applied to longitudinal measure- ments of Snail induction, showing that it was relatively rapid and switch-like. In contrast, TGF-β1 induction kinetics were more variable at the single cell level and occurred more gradually. Even without exogenous stimuli, EMT was observed at low cell densities. Finally, sublethal treatment with the chemotherapeutic Taxol exacer- bated EMT for both uninduced and induced Snail populations, resulting in highly elongated vimentin and aberrant nuclear morphologies. Overall, these single cell ap- proaches offer unique biological insights into phenotypic heterogeneity and plasticity that would be overlooked in higher-throughput assays based on population averages and endpoints. 96 Figure 3.1: Classification of epithelial and mesenchymal phenotypes after 72 h DMSO (control) and OHT (Induced Snail-1), followed by 72 h DMSO. (A) Immunofluorescent staining of nuclei, vimentin, cytoplasm and E-cadherin (blue, red, gray, and green, respectively). Scale = 25 µm. (B) Process flow of single cell seg- mentation and feature extraction. (C) Performance classifier of actual and predicted subpopulations with the number of cells and percent ( ) of cells in each group. 97 Figure 3.2: Time course measurements of cell morphology and biomarker expression with Snail-1 induction (OHT). (A) Control cells (DMSO) show characteristic epithelial features, including compact, cobblestone like morphologies with high E-cadherin at the cell edges and low vimentin expression. 24 h OHT treatment results in decreased E-cadherin at the edges. 48 h and 72 h OHT treatment result in vimentin and cytoplasmic features with increased area and elongation. Scale = 50 µm. (B) GMM phenotypic classification shows that 0 h and 24 h are primarily epithelial (green bars and %), with rapid transition to mesenchymal phenotype at 48 h and 72 h (red bars and %), relative to the com- plete population (black line). N indicates the number of individual cells pooled per condition. 98 Figure 3.3: Time course measurements of cell morphology and biomarker expression with TGF-β1 induction. (A) Control cells (0 h) also show characteristic epithelial features with high E-cadherin at the cell edges and low vimentin expression. 24 h TGF-β1 treatment results in loss of E-cadherin for some cells, compared with increased vimentin expression. 48 h and 72 h OHT treatment result in vimentin and cytoplasmic features with increased area and elongation. Interestingly, TGF-β1 results in a more boxy shape than with Snail-1 induction. Scale = 50 µm. (B) GMM phenotypic classification shows that 0 h is primarily epithelial (green bars and %), with a gradual increase in mesenchymal phenotypes (red bars and %) over 72 h, relative to the complete population (black line). N indicates the number of individual cells pooled per condition. 99 Figure 3.4: Changes in cell morphology and biomarker expression with cell density. (A) Cells at high density (95%) form a confluent monolayer with cobblestone mor- phology, tight E-cadherin junctions and low vimentin. Cells at lower, subconfluent densities (70%, 50%) display some loss of E-cadherin, with some elongation. Sparse cells (15%) display no E-cadherin junctions, with vimentin and cytoplasmic features that display increased area and elongation. Scale = 50 µm. (B) GMM phenotypic classification shows that 95% density is primarily epithelial (green bars and %), with a gradual increase in mesenchymal (red bars and %) phenotypes up to 51% at 15% density, relative to the complete population (black line). N indicates the number of individual cells pooled per condition. 100 Figure 3.5: Changes in cell morphology and biomarker expression with sublethal Taxol treatment. (A) Uninduced control cells (DMSO/DMSO) show epithelial features, but uninduced treated cells (DMSO/Taxol) show increased vimentin expression and elongation with aberrant nuclear morphologies. Scale = 50 µm. (B) GMM phenotypic classification shows an increase in mesenchymal (red bars and %) phenotype after Taxol treat- ment compared to the untreated control, which is primarily epithelial (green bars and %), relative to the complete population (black line). (C) Induced control cells (OHT/DMSO) show more mesenchymal features. Induced treated cells (OHT/Taxol) show strong mesenchymal features, with aberrant nuclei and extremely elongated mor- phologies. Scale = 50 µm. (D) GMM phenotypic classification shows a strong shift towards mesenchymal phenotype after Taxol treatment. N indicates the number of individual cells pooled per condition. 101 3.6 Supplementary Information 3.6.1 Supplementary Methods Cell Culture Human mammary epithelial cells stably transfected with an ER-Snail-16SA retrovi- ral construct (MCF-10A Snail) as well as the parental cells (MCF-10A) were a gift from D.A. Haber (Massachusetts General Hospital).[255] The Snail-16SA variant is refractory to phosphorylation and is thus stably expressed and localized in the nu- cleus, where it initiates EMT induction.[256] Both MCF-10A variants were cultured following Brugge and coworkers.[257] Corning 96-well half area high content imaging microplates (Fisher Scientific, Cat. No. 15-100-170) were used for all imaging exper- iments. First, microplates were coated with fibronectin to aid cell adhesion; Briefly, fibronectin from human plasma (Sigma-Aldrich, Cat. No. F2006) was diluted with 1X phosphate buffered saline (PBS) to a final concentration sufficient to coat the well bottom with 5 µg/cm2 of protein. Cells were seeded into the microplate wells at a concentration of 500 cells/well (unless otherwise specified), allowed to settle and adhere overnight, and were then subjected to exogenous application of stimuli and/or drugs as indicated. Special care was taken to ensure that cells remained subconfluent at the conclusion of the experiments. T47D breast carcinoma cells were purchased from the DTP, DCTD Repository (Charles River Labs, Inc.), cultured in RPMI 1640 (ATCC Cat. No. 30-2001) supplemented with 10% Fetal Bovine Serum (Fisher Sci- entific, Cat. No. SH3007103), 0.2U/mL bovine insulin (Sigma-Aldrich, Cat. No. I1882), 100 units/mL penicillin, and 100 µg/mL streptomycin, and were passaged following standard ATCC protocol. MDA-MB-231 cells were purchased from ATCC and were cultured in DMEM containing L-Glutamine, L-glucose, and sodium pyru- vate (Fisher Scientific, Cat. No. MT-10-013-CV) and supplemented with 10% FBS, 100 units/mL penicillin, and 100 µg/mL streptomycin. T47D and MDA-MB-231 cells 102 were cultured for 96 h in high content imaging plates as described and plated with 3,000 and 1,500 cells/well, respectively. These higher seeding densities with respect to MCF-10A experiments were experimentally determined, and compensate for dif- ferences in the characteristic doubling times of each cell line. All cell lines used were cultured in T25 flasks and passaged at 70-80% confluence; only low passage numbers were used (P<15). Cultures were maintained under standard conditions including humid incubation at 37◦ C, 5% CO2. EMT Induction with Snail-1 or TGF-β1 Snail-1 expression was induced in MCF-10A Snail cells through the addition of 4- hydroxytamoxifen (OHT; Sigma-Aldrich, Cat. No. H7904) resuspended in DMSO and used at a final concentration of 500 nM in media. To induce EMT to varying extents at the population level, a time course of Snail-1 induction was prepared with four conditions, all over a course of 72 h. The durations of OHT exposure included 0, 24, 48, and 72 h, where DMSO (0.05% final concentration, consistent with OHT dose) was applied during the remainder of the 72 h time course. Separately, EMT was induced in MCF-10A cells through treatment with 5 ng/mL recombinant human TGF-β1 (R&D Systems, Cat. No. 240-B) in growth media. TGF-β1 was prepared at 20 µg/mL in 4mM HCl containing 1mg/mL bovine serum albumin and the final media contained 0.025% of stock solution. In the absence of TGF-β1, cells received regular growth media. A comparable time course was conducted with four conditions over 72 h. All conditions were initially treated with growth media and TGF-β1 was added to different wells for durations of 0, 24, 48 and 72 h. For both EMT induction experiments, media was replenished at the 48 h time point, to prevent nutrient deprivation and growth factor depletion. To examine whether or not the 72 h time course of OHT treatment was sufficient to induce EMT and yield a stable mesenchymal population, a 72 h washout of OHT 103 was performed. Briefly, MCF-10A Snail cells were preinduced with OHT for 72 h, washed with media, and then media containing 0.05% DMSO was added for an additional 72 h (Fig. S8). Similarly, a TGF-β1 washout experiment was performed where TGF-β1 was applied to MCF-10A cells for 72 h, cells were washed with media, and supplemented with media only for an additional 72 h (Fig. S10). At the end of time courses, cells were fixed, immunostained, segmented using CellProfiler, and classified into epithelial and mesenchymal populations using GMM. Density Dependent Induction of EMT MCF-10A cells were plated in high content imaging plates, as described, where 375 cells/well was chosen as the lowest seeding density. This density was first validated during routine culture (equivalent to 1:20 split) as it maintained the ability to pro- duce characteristic epithelial monolayers several days after passaging cells. Thus, to generate a range of terminal densities, cells were plated at either 375 or 500 cells/well (the typical density for all other experiments) and subsequently cultured for either 72 or 96 h before fixation and immunostaining. The resulting density for each seg- mented (CellProfiler) and analyzed (GMM) condition was calculated by first counting the number of cells per 20x field of view of the acquired images, which was then di- vided by the area of the 20x field of view to yield: Cells/Surface Area. For ease of comparison between conditions, the overall confluency of cells in the entire 20x field was estimated by eye. Thus, both qualitative and quantitative density measures were determined in order to effectively evaluate density dependent induction of EMT (Fig. S12). For the sparse density condition, multiple experimental replicates were pooled to obtain >100 cells for GMM and thus, the density was determined as an average estimate of 15% confluency across replicates. 104 Drug Treatment with Taxol for EMT Induction MCF-10A Snail cells were cultured in media with OHT for 72 hours to induce a mesenchymal phenotype (preinduced) or with DMSO to maintain an epithelial phe- notype (uninduced). Both preinduced and uninduced cells were seeded at a density of 750 cells/well. This elevated seeding density relative to OHT and TGF-β1 ex- periments was used due to the shortened duration of the time course with Taxol treatment (48 h Taxol vs. 72 h OHT and TGF-β1). Cells were then allowed to adhere to fibronectin-coated multiwell plates overnight as described, and the growth media containing OHT or DMSO was removed. Both cell types were then treated with either DMSO (0.05%, control) or 4 nM paclitaxel (Taxol) for 48 hours, and subsequently fixed and immunostained. Cell Viability: The 4nM Taxol dose was determined to be sublethal by running two different viability assays. At the completion of the experiment, dead cells were labeled with DRAQ7 stain (Abcam, Cat. No. ab109202). Briefly, DRAQ7 (0.3mM) was added to each well at a final dilution of 1:200 and cells were incubated at 37◦ C, 5% CO2, for 10 minutes. Next, cells were imaged as described, using a 10X Plan Fluor objective (NA 0.3, long working distance); images were acquired using Phase contrast to capture images of live and dead cells, while Cy5 was used to capture images of DRAQ7 positive, dead cells. The indicated drug dose resulted in cell death comparable to untreated cells and cells treated with DMSO (Fig. S19A-C). To quantify the degree of cell death for each condition, cells were plated in 6-well tissue culture plates (Genesee scientific), allowed to settle and adhere overnight as in imaging experiments, then treated with media (control), 0.05% DMSO, or 4nM Taxol for 48 h. At the end of the 48 h treatment period, cells were washed with 1X PBS, and Accumax was added to obtain a single cell suspension. Once fully lifted, cells were resuspended in MCF-10A resuspension media, centrifuged, resupended in 200 µL growth media, and then a small aliquot (20 µL) of cell suspension was mixed at a 105 1:1 ratio with 0.2% Trypan Blue. Next, 20 µL of the mixed suspension was added to a Cellometer Counting Chamber (Nexcelom Bioscience, Cat. No. SD100) and percent viability was determined using an automated imaging and analysis program: Trypan Blue Viability Assay with Cellometer Auto 1000 Cell Viability Counter (Nexcelom Bioscience) (Fig. S16D). Percent viability was determined for 3 replicates of each condition. Immunostaining and Fluorescent Imaging At the conclusion of time course experiments, cells were fixed for 20 minutes at 4C with 4% paraformaldehyde in 1X PBS (all solutions are in 1X PBS unless otherwise specified) containing 2mM calcium chloride and 2mM magnesium chloride. Cells were subsequently immunostained for the epithelial and mesenchymal biomarkers E- cadherin and vimentin, respectively. Cells were permeabilized with 0.1% Triton X-100 for ¡5 minutes, washed, and blocked with 10% goat serum. Cells were washed with sodium acetate buffer (7.5mM, pH 7.4, in 1X PBS) for 15 minutes and then rinsed several times with 1% nonfat dry milk. Cells were incubated overnight at 4C with primary antibodies: 250 µg/mL E-cadherin (Fisher Scientific, Cat. No. BDB610181) diluted at 1:500 and vimentin (Cell Signaling Technology, Cat. No. 5741S) diluted at 1:200 in 1% milk. Cells were then washed with 1% milk and incubated in the dark at room temperature with suitably matched secondary antibodies: 2mg/mL Alexa Fluor 488 and Alexa Fluor 555 (ThermoFisher Scientific, Cat. No. A-11001 and A-21428, respectively) diluted at 1:500 in 1% milk. Lastly, cells were washed with 1X PBS, incubated with 2 µg/mL Hoechst Pentahydrate and 2 µg/mL HCS CellMask Deep Red (ThermoFisher Scientific, Cat. No. 33258 & H32721, respectively) for 30 minutes at room temperature, and washed again. Immunostained cells were imaged using an inverted epifluorescence microscope (Nikon TiE). Images were acquired with a 20X Super Plan Fluor objective (NA 106 = 0.45, extra long working distance) and with a 14-bit resolution sCMOS camera (Andor Neo). Fluorescence illumination was provided by a light-guide coupled Lu- mencore Sola white light excitation system. Care was taken to ensure all images were acquired with identical acquisition parameters (exposure time, camera gain/gamma control, and microscope aperature settings). Post-acquisition, fluorescence thresh- olds for each experiment were set for qualitative comparison across conditions using built-in NIS Elements AR settings, specifically 10% Low (under-exposed) to elimi- nate background, and 0.5% High (overexposed) to reduce pixel saturation. Thresholds were set on a channel-by-channel basis, with respect to the condition with the highest level of biomarker expression. Lastly, gamma was adjusted to 0.5 for all images of cytoplasmic features (Deep Red stain, Cy5 channel) to accentuate dim features of cells for visualization purposes only (Figures); gamma was kept at 1 for all other images presented in the paper and for the segmentation of all images. Image Analysis Cell detection and shape feature analysis were performed using automated cell seg- mentation with CellProfiler 2.1 (Broad Institute) (Fig. S2).[ref 38]?? First, the Cor- rect Illumination Calculate and Apply modules were applied to correct for uneven illumination across channels and image sets. Second, fluorescently labeled nuclei (DAPI channel) were segmented as primary objects, which then served as seed ob- jects for the detection of fluorescently labeled vimentin (TRITC channel), a biomarker of EMT, and fluorescently labeled cytoplasm (Cy5 channel), which became the sec- ondary objects for the cell body. These fluorescent objects were manually verified and corrected as needed using the EditObjectsManually module. Detailed segmentation parameters are described in the Supporting Information (Fig. S2). Finally, shape measurements were extracted from the segmented objects, particularly nuclear max radius, vimentin area, cytoplasm form factor, and cytoplasm max feret diameter, us- 107 ing the MeasureObjectSizeShape module. Fluorescence intensity metrics were also extracted using the MeasureObjectIntensity module, however, subsequent analysis revealed that shape metrics were more valuable for consistent cell classification. Phenotype Classification using a Gaussian Mixture Model (GMM) An initial training set was developed by segmenting cells treated either with DMSO or 4-OHT for 72 h, followed by 72 h treatment with DMSO, corresponding to putative epithelial and mesenchymal phenotypes. For each condition, CellProfiler was used to segment 100-500 cells for 15 shape features (Figure S3). These single cell features from both conditions were then combined into a single dataset. For consistent comparison, shape features were rescaled between 0 and 1 by dividing by the maximum value of each shape feature for the combined dataset. Notable shape features used to distinguish between epithelial and mesenchymal phenotypes included the following (as defined by CellProfiler 2.1): Area: The actual number of pixels in the region. FormFactor: Calculated as 4∗π∗Area/Perimeter2 . Equals 1 for a perfectly circular object. MaxFeretDiameter: The Feret diameter is the distance between two parallel lines tangent on either side of the object (imagine taking a caliper and measuring the object at various angles). The maximum Feret diameter is the largest possible diameter, rotating the calipers along all possible angles. MaxRadius: The maximum distance of any pixel in the object to the closest pixel outside of the object. For skinny objects, this is 1/2 of the maximum width of the object. Cell intensity metrics were also extracted, where the most noteworthy differences between epithelial and mesenchymal phenotypes were observed for vimentin metrics including mean, median and integrated intensity: 108 IntegratedIntensity: The sum of pixel intensities within an object. Based on the 15 shape parameters and vimentin intensity metrics, all possible combinations of features (up to 5) were screened for their classification accuracy. An expectation maximization (EM) algorithm was used to determine maximum likelihood estimates of the parameters for a Gaussian Mixture Model[258] (gmdistribution.fit, MATLAB R2013b). This analysis specifies two subpopulations and assumes that the total population can be reconstructed as a weighted mixture of two phenotypes defined by a Gaussian distribution around some distinct phenotype with some overall probability. Based on this reference subpopulation model, the heterogeneity of a particular cell population can be estimated. For each cell, the posterior probability that it is either epithelial or mesenchymal can be computed from the reference model using Bayes rule (posterior, MATLAB R2013b). This classifier was then tested against separate datasets with known phenotypes, which were partitioned into two subpopulations and then evaluated for accuracy (clus- ter, MATLAB R2013b). First, a test set was constructed using MCF-10A cells, again treated either with DMSO or 4-OHT for 72 h, followed by 72 h treatment with DMSO. Second, a renormalized test set was constructed using epithelial (T-47D) and mesenchymal (MDA-MB-231) cells. The accuracy of the partitioning was as- sessed by comparing the actual and predicted values of control and induced conditions (classperf, MATLAB 2013b). Overall, an optimized four metric GMM distribution was selected based on vi- mentin area, cytoplasmic maximum feret diameter and form factor, as well as nuclear max radius. Vimentin area, cytoplasmic max feret diameter, and nuclear max radius all showed a consistent increase when comparing the control condition (DMSO) with induced EMT (OHT), while cytoplasmic form factor displayed a consistent decrease across these conditions. Overall, this classifier showed the highest accuracy for ep- ithelial and mesenchymal phenotypes in both training sets. Based on this four metric 109 GMM reference model, all subsequent datasets were partitioned into epithelial and mesenchymal subpopulations, respectively (cluster, MATLAB R2013b). Statistical Analysis To assess statistical significance between the distributions of data for cell morphologic features between Epithelial (DMSO) and Mesenchymal (OHT) train and test condi- tions, the two-sample Kolmogorov-Smirnov test in MATLAB (Mathworks) was used. The null hypothesis was rejected at p ≤ 0.05% (5% significance level) and only highly statistically different morphologic features were used for cell classification of epithe- lial and mesenchymal subpopulations. Boxplots displaying these cell shape features were generated in MATLAB (Mathworks) using the BOXPLOT function, in which the dividing line of the box represents the median, box edges signify the 25th and 75th percentiles, and whiskers indicate 99.3% coverage of the data. Notched boxplots were plotted with extremes at 1.57(q3 − q1) q2 ± √ , (3.1) n where q2 is the median (50th percentile), q1 and q3 are the 25th and 75th per- centiles, respectively, and n is the number of observations. Two medians will be significantly different at the 5% significance level if their notches do not overlap. The PLOTSPREAD function (Mathworks File Exchange) was additionally used to overlay data points corresponding to individual cell metrics, which were appropriately col- ored according to cell classification with Gaussian mixture modeling, where epithelial = green and mesenchymal = red. This function offsets data points horizontally to aid visualization of the statistical distribution; this offset should not be interpreted as meaningful. The percentage of cells classified as epithelial and mesenchymal for 110 each experiment were plotted as histograms and bar graphs using the HIST and BAR functions, respectively, in MATLAB (Mathworks). 3.6.2 Supplementary Figures Figure 3.6: Schematic of Experimental Conditions for GMM Training Set. Overview of conditions that generate epithelial (E) and mesenchymal (M) populations used to train the Gaussian mixture model. MCF-10A Snail cells were seeded at t=0 (DMSO control, white bar), while OHT (500nM final concentration, red bar) was added to cells for a total duration of 72 h to induce a mesenchymal phenotype. From 72 h to 144 h, OHT was replaced with DMSO, to allow for the further progression of E and M phenotypes. At the completion of the time course, cells were fixed and subsequently immunostained for epithelial and mesenchymal biomarkers. 111 Table 3.1: Flow chart of CellProfiler analysis pipeline: Part 1-2 112 Table 3.2: Flow chart of CellProfiler analysis pipeline: Part 3-4. Flow chart of CellProfiler analysis pipeline. General process overview for image anal- ysis using CellProfiler 2.1 (Broad Institute). NUC, VIM, and CYT represent the original TIFs acquired for nuclear, vimentin, and cytoplasm channels, respectively, after fluorescence imaging of immunostained samples. CellProfiler modules were uti- lized in order as indicated, with the given input and output images. Parameters and comments provide key information for image enhancement, object segmentation, automated measurements, and data export used in the analysis. 113 Figure 3.7: Histograms of all Nuclear, Vimentin, and Cytoplasmic metrics for Ep- ithelial and Mesenchymal Training Sets. Histograms of normalized distributions from 0 to 1 of all nuclear, vimentin, and cytoplasm shape metrics for the epithelial (DMSO) group (green), and mesenchymal (72 h OHT, 72 h DMSO) group (red). Counts (y-axis) represent the number of individual cells with the indicated normalized value (xaxis) for each metric. 114 Figure 3.8: Histograms of all Nuclear, Vimentin, and Cytoplasmic metrics for Ep- ithelial and Mesenchymal Training Sets. Histograms of normalized distributions from 0 to 1 of all nuclear, vimentin, and cytoplasm shape metrics for the epithelial (DMSO) group (green), and mesenchymal (72 h OHT, 72 h DMSO) group (red). Counts (y-axis) represent the number of individual cells with the indicated normalized value (xaxis) for each metric. 115 Figure 3.9: Histograms of Vimentin Intensity metrics for Epithelial and Mesenchymal Training Sets. Histograms of normalized distributions from 0 to 1 of all vimentin intensity metrics including integrated intensity, mean intensity, and median intensity for the epithelial (DMSO) group (green), and mesenchymal (72 h OHT, 72 h DMSO) group (red). Vi- mentin area is also displayed for comparison, which demonstrates the overlap between area and integrated intensity distributions. Counts (y-axis) represent the number of individual cells with the indicated normalized value (x-axis) for each metric. 116 Figure 3.10: Boxplot of selected metrics used to distinguish Training and Test Ep- ithelial and Mesenchymal conditions. Boxplots displaying individual cell raw values for each of the selected shape metrics of epithelial cells (E; DMSO condition, green) and mesenchymal cells (M; OHT con- dition, red) in the GMM training and test sets. For each of these selected metrics, the two-sample Kolmogorov-Smirnov test (kstest2 in MATLAB) demonstrates that the data for E and M groups compromise different, continuous distributions. 117 Figure 3.11: Predicted vs. Actual Segmentation for GMM Training and Test sets. (A) Histograms of normalized distributions from 0 to 1 of selected metrics for the epithelial (DMSO condition, green) and mesenchymal (OHT condition, red) groups for training and test sets, displaying the GMM classification with predicted DMSO (small dashes), and predicted OHT (large dashes). Counts (y-axis) represent the number of individual cells with the indicated normalized value (x-axis). (B) Con- fusion matrix displaying the number and percentage of cells in each group. (C-D) Immunofluorescent staining of nuclei (blue), E-cadherin (green), vimentin (red), and cytoplasm (gray). Scale = 25 µm. (C) Representative IF of the test set epithelial condition (72 h DMSO); Arrow indicates a misclassified cell predicted to be in the OHT group with GMM, likely due to phenotypic plasticity. (D) Representative IF of the test set mesenchymal condition (72 h OHT, 72 h DMSO); Arrow indicates a cell undergoing a division event that was misclassified by GMM (predicted DMSO). 118 Figure 3.12: Predicted vs. Actual Segmentation for GMM of T-47D and MDA-MB231 cells. (A) Immunofluorescent staining of nuclei (blue), E-cadherin (green), vimentin (red), and cytoplasm (gray) of epithelial and mesenchymal cell lines, T47-D and MDAMB- 231, respectively. Scale = 25 µm. (B) Histograms of normalized distributions from 0 to 1 of selected metrics for T-47D (green) and MDA-MB-231 (red) groups, displaying the GMM classification with predicted T-47D (small dashes), and predicted MDA- MB231 (large dashes). Counts (y-axis) represent the number of individual cells with the indicated normalized value (x-axis). (C-D) Confusion matrix displaying the per- centage of cells in each group by classification using all 4 metrics (C) and a reduced set of 3 metrics with only nuclear and cytoplasm features (D). 119 Figure 3.13: Optimal combinations of Nuclear, Vimentin, and Cytoplasmic metrics for Cell Classification. Parallel coordinates plot displaying the % accuracy (y-axis) for GMM classification of various epithelial and mesenchymal conditions (x-axis labels) with each of the following combinations of classification metrics (right). The 4-metric combination (solid, dark blue line) represents the best overall set of metrics for classifying both epithelial and mesenchymal conditions, and has the highest Total % accuracy (x-axis, far left). 120 Figure 3.14: Schematic of experimental conditions and results for OHT induction. (A) Overview for the timeline of Snail-1 induction with OHT. All cells were seeded at t=0 (DMSO control, white bar), while OHT (500nM final concentration, red bar) was added to cells at the indicated time points to yield 0, 24, 48, and 72 h OHT conditions. At the completion of the time course, cells were fixed and subsequently immunostained for epithelial and mesenchymal biomarkers. (B) Barplot of epithelial and mesenchymal fractions with duration of OHT induction. The percentage of cells classified as epithelial (green) largely decreases from 24 to 48 hours of OHT treatment, while the percentage of cells classified as mesenchymal (red) greatly increases between these conditions. (C-D) Overview of the timeline for the test condition (from Fig.1D, S4, S5) of OHT pulse (72 h OHT), withdrawal (72 h DMSO) (C), which yields a primarily mesenchymal population as represented by the barplot of epithelial and mesenchymal fractions (D). 121 Figure 3.15: Boxplot of selected metrics with duration of OHT induction. Boxplots overlaid with data points representing individual cell values for each of the selected shape metrics of DMSO (control cells) and OHT (induced cells) for 0, 24, 48, and 72 h conditions. Individual cells were classified as epithelial (green) or mesenchymal (red) by GMM. 122 Figure 3.16: Posterior Probabilities of Gaussian Mixture Model for OHT induction. Posterior probability distributions depicting the conditional likelihood that a cell is mesenchymal (p=1). Percentage of cells classified as epithelial (green), mesenchymal (red) and total number of cells analyzed per conditioned (N) are displayed above graphs (right). For p = 0 to 0.25 and 0.75 to 1, there is a high probability that an individual cell is epithelial (green) and mesenchymal (red), respectively, with relative percentages displayed within graphs. The middle percentage where p = 0.25 to 0.75 represents the fraction of cells with less definitive classification. (A) For the time course of OHT induction and (B) for the OHT pulse, withdrawal condition. 123 Figure 3.17: Schematic of experimental conditions and results for TGF-β1 induction. (A) Overview for the timeline of TGF-β1 induction. All cells were seeded at t=0 h (control, white bar), while TGF-β1 (5 ng/mL final concentration, red bar) was added to cells at the indicated time points to yield 0, 24, 48, and 72 h TGF-β1 conditions. At the completion of the time course, cells were fixed and subsequently immunostained for epithelial and mesenchymal biomarkers. (B) The percentage of cells classified as epithelial (green) gradually decreases from 0 to 72 hours of TGF-β1 treatment, while the percentage of cells classified as mesenchymal (red) increases for these conditions. (C-D) Overview of the timeline for the TGF-β1 pulse (72 h TGF-β1), withdrawal (72 h media) experiment (C), which yields a primarily mesenchymal population as represented by the barplot of epithelial and mesenchymal fractions (D). 124 Figure 3.18: Boxplot of selected metrics with duration of TGF-β1 induction. Boxplots overlaid with data points representing individual cell values for each of the selected shape metrics for untreated cells (control) and TGF-β1 treated cells for the indicated durations. Individual cells were classified as epithelial (green) or mesenchymal (red) by GMM. 125 Figure 3.19: Posterior Probabilities of Gaussian Mixture Model for TGF-β1 induc- tion. Posterior probability distributions depicting the conditional likelihood that a cell is mesenchymal (p=1). Percentage of cells classified as epithelial (green), mesenchymal (red) and total number of cells analyzed per conditioned (N) are displayed above graphs. For p = 0 to 0.25 and 0.75 to 1, there is a high probability that an individual cell is epithelial (green) and mesenchymal (red), respectively, with relative percentages displayed within graphs. The middle percentage where p = 0.25 to 0.75 represents the fraction of cells with less definitive classification. (A) For the time course of TGF-β1 induction and (B) for the TGF-β1 pulse, withdrawal condition. 126 Figure 3.20: Schematic of experimental conditions and results for Density experi- ments. (A) Overview for the timeline of cell density experiments. All cells were seeded at t=0 h, while cells were plated at two different densities, either 375 or 500 cells/well, and were subsequently cultured for both 72 h and 96 h durations to yield conditions with a wide range of endpoint densities (high to low confluency: yellow to blue bars). At the completion of the time course, cells were fixed and subsequently immunostained for epithelial and mesenchymal biomarkers. (B) Table of quantified values for the number of cells per image, cells per surface area (cm2 ), description of density, and estimated confluency of cells at the experimental endpoint; resulting densities ranged from 95 to 15% confluence. (C) Barplot of epithelial and mesenchymal fractions that result from the indicated confluency condition. The percentage of cells classified as epithelial (green) gradually decreases from 95 to 15% confluence, while the percentage of cells classified as mesenchymal (red) gradually increases between these conditions, with the lowest density having the largest portion of mesenchymal cells at approximately 43% of the population. 127 Figure 3.21: Boxplot of selected metrics with Density experiments. Boxplots overlaid with data points representing individual cell values for each of the selected shape metrics for cells at a range of densities (95 to 15% confluence). Individual cells were classified as epithelial (green) or mesenchymal (red) by GMM. 128 Figure 3.22: Posterior Probabilities of Gaussian Mixture Model with Density. Posterior probability distributions depicting the conditional likelihood that a cell is mesenchymal (p=1) for the indicated confluencies. Percentage of cells classified as epithelial (green), mesenchymal (red) and total number of cells analyzed per con- ditioned (N) are displayed above graphs. For p = 0 to 0.25 and 0.75 to 1, there is a high probability that an individual cell is epithelial (green) and mesenchymal (red), respectively, with relative percentages displayed within graphs. The middle percentage where p = 0.25 to 0.75 represents the fraction of cells with less definitive classification. 129 Figure 3.23: Schematic of experimental conditions and results for Taxol treatment. (A) Overview for the timeline of Taxol experiments. All cells were first treated with DMSO (uninduced; A, B) or 500nM OHT (preinduced; C, D) for 72 h, followed by treatment with DMSO (uninduced and preinduced controls; A & C, respectively) or 4nM Taxol (uninduced and preinduced Taxol; B & D, respectively) for 48 h. Cells were subsequently fixed and immunostained for epithelial and mesenchymal biomark- ers. (B) Barplot of epithelial and mesenchymal fractions with the indicated conditions A-D. The percentage of cells classified as epithelial (green) largely decreases in unin- duced cells treated with Taxol (A vs. B), while the percentage of cells classified as mesenchymal (red) greatly increases between these conditions, and is thus very sim- ilar to the preinduced OHT condition (C). At the population level, Taxol treatment in preinduced cells appears to maintain the mesenchymal phenotype (C vs. D). 130 Figure 3.24: Boxplot of selected metrics with Taxol Treatment. Boxplots overlaid with data points representing individual cell values for each of the selected shape metrics for the indicated Taxol conditions (uninduced control, unin- duced Taxol, preinduced control, and preinduced Taxol, from left to right). Individual cells were classified as epithelial (green) or mesenchymal (red) by GMM. 131 Figure 3.25: Live/Dead Staining and Percent Viability after Taxol treatment. (A-C) Merged images for phase contrast and DRAQ7 (dead cell stain) of MCF-10A cells after 48 h of the following treatments: untreated (control, A), 0.05% DMSO (DMSO control, B), and Taxol treated (4nM Taxol, C). Scale = 100 µm. Arrows in- dicate DRAQ7 positive staining cells (red). (D) Percent viability was also determined for triplicates of each condition using an automated program: Trypan Blue Viabil- ity Assay (Cellometer 1000 Auto, Nexcelom Bioscience). The percentage viability is comparable for each of the conditions. 132 Figure 3.26: Posterior Probabilities of Gaussian Mixture Model with Taxol treatment. Posterior probability distributions depicting the conditional likelihood that a cell is mesenchymal (p=1). Percentage of cells classified as epithelial (green), mesenchymal (red) and total number of cells analyzed per conditioned (N) are displayed above graphs. For p = 0 to 0.25 and 0.75 to 1, there is a high probability that an individual cell is epithelial (green) and mesenchymal (red), respectively, with relative percentages displayed within graphs. The middle percentage where p = 0.25 to 0.75 represents the fraction of cells with less definitive classification. (A) For uninduced cells treated with DMSO (top) or Taxol (bottom) and (B) for preinduced (OHT) cells treated with DMSO (top) or Taxol (bottom). 133 Chapter 4 Motility-Limited Aggregation of Mammary Epithelial Cells into Fractal-like Clusters Chapter 4 is in preparation as a primary article: S.E. Leggett, Z.J. Neronha, D. Bhaskar, J.Y. Sim, T.M. Perdikaris, and I.Y. Wong. “Motility-Limited Aggregation of Mammary Epithelial Cells into Fractal-like Clus- ters.” 4.1 Abstract Transitions between individual and collective cell migration are associated with tis- sue morphogenesis, wound healing, and tumor progression [268]. In particular, the epithelial-mesenchymal transition (EMT) occurs when tightly adherent epithelial cells detach and disseminate as individual mesenchymal cells [2]. For instance, clusters of epithelial cells in vitro can “scatter” individually upon addition of epidermal growth 134 factor (EGF) [269, 270] or hepatocyte growth factor (HGF) [271, 272, 273]. Similarly, a “partial” EMT has been associated with leader cells with enhanced motility, which retain some cell-cell contacts to mechanically guide their followers [274, 275, 276]. Instead, a reverse mesenchymal-epithelial transition (MET) can occur when mes- enchymal cells condense and differentiate into a compact epithelial tissue, associated with skeletal development in vivo [277]. Analogous “swarming” behaviors have been observed during neural crest development [278], neutrophil recruitment [279], and Dictyostelium aggregation [280], although these cells do not acquire strong cell-cell junctions in the process. Overall, this emergence of complex spatial structure from collective cellular motion and cell-cell adhesion remains poorly understood. Soft matter systems such as colloidal particles in a fluid medium also exhibit collec- tive phase behaviors due to local interactions [281]. For instance, dispersed colloids that diffuse randomly but adhere irreversibly can aggregate into highly branched, connected clusters with fractal-like architectures (i.e. diffusion-limited aggregation) [282, 283, 284]. This local arrest of particle dynamics at low densities corresponds to a macroscopic transition from a fluid-like solution to solid-like gel, which can be mapped to a “jamming” phase diagram controlled by particle density, the interparticle adhesion, and shear stress [285]. Furthermore, in the limit of higher packing densities, colloidal systems can exhibit a “jamming” transition as they approach random close packing, analogous to a glass transition [286]. Interestingly, closely packed epithelial monolayers also exhibit arrested dynamics as they approach confluent cell densities [287, 288, 289, 290, 291, 292, 156, 293]. Based on these results, a jamming-like phase diagram for cell monolayers has been proposed based on cell density, cell-cell adhesion, and cell speed [294], but the importance of cell density remains unresolved [291, 292]. Since both colloidal particles and living cells exhibit a glass-like jamming transition at high densities, and colloidal particles exhibit a gelation-driven jamming transi- 135 tion at low densities, an intriguing possibility is that living cells could also exhibit aggregation and arrest at low (subconfluent) densities. Here, we show that mammary epithelial cells aggregate into multicellular clusters with branched, fractal-like architectures when cultured in reduced growth factor con- ditions. Cluster formation was only observed with mitogen-dependent epithelial cells, which exhibited diminished proliferation and motility. Single cell tracking revealed that migratory individuals adhered irreversibly to multicellular clusters, significantly arresting their motion. Treatment with tamoxifen to induce Snail expression resulted in the formation of leader cells at the cluster periphery, guiding collective migration outwards to connect clusters into spanning networks. We constructed a phase dia- gram for cluster formation and arrested migration, which was determined by local cell coordination rather than overall density. We further developed a minimal physical model that recapitulated cluster formation and leader cell formation. These biophys- ical phenomena are unanticipated by existing models of cellular jamming and may yield insights into the formation of branching network architectures in living tissues. 4.2 Results Human mammary epithelial cells (MCF-10A) were cultured in high EGF media (con- taining ∼20 ng/mL EGF, 5% horse serum) and low EGF media (containing 2% horse serum, no added EGF). Since MCF-10A cells are non-transformed, they are mitogen- dependent and require EGF for proliferation [295]. Cells were plated at varying densities (500-3000 cells) in a high-content 96-well plate coated with collagen I. Cells were allowed to adhere for ∼4 h, after which fresh media containing DMSO, tamox- ifen (OHT), and/or drug was added. We estimate that some residual EGF (∼0.075 ng/mL) remains in assay media, since the resuspended cells are maintained in growth media until seeding. Cells were then transferred to a fluorescence microscope with 136 automated stage and environmental control (37◦ C and 5% CO2 ) for time-lapse imag- ing (every 15 min). To facilitate image analysis, this MCF-10A cell line was stably transfected with red fluorescent protein in the nucleus (H2B mCherry) and green flu- orescent protein (GFP) in the cytoplasm. These cells were also transfected with an inducible ER-Snail-16SA construct to undergo EMT after treatment with tamoxifen [255, 89]. 4.2.1 Cell Proliferation and Motility are Arrested by Re- duced Growth Factor Conditions Single cell dynamics at high and low EGF concentrations were visualized from the nucleus as well as the cytoplasm, which each overexpressed different fluorescent pro- teins. In high EGF media containing OHT to induce Snail and EMT [255], dispersed cells exhibited relatively fast proliferation and formed continuous monolayers that oc- cupied almost all the available area by 60 h (Fig. 4.1A). These cells were also highly motile and constantly rearranged themselves, even at near-confluent cell densities. These behaviors were qualitatively consistent in DMSO controls and over a range of initial cell densities (Supporting Information, Fig. 4.5C). In contrast, cells in reduced EGF assay media displayed significantly reduced proliferation with considerable areas left unoccupied after 60 h (Fig. 4.1B), even from higher starting densities (Support- ing Information, Fig. 4.5B, 4.5D). Despite these subconfluent cell densities, initially dispersed cells became associated with a multicellular cluster by 24 h. The overall morphology of these clusters remained roughly consistent through 60 h, suggesting that individual cell motility was arrested and rearrangements were limited. Inter- estingly, the OHT-treatment resulted in highly branched clusters which merged into spanning networks over time (Fig. 4.1B). In contrast, clusters in DMSO controls ap- peared more compact and rounded, and exhibited reduced connectivity at longer times relative to OHT treatment (Supporting Information, Fig. 4.5B, 4.5D). Qualitatively 137 similar clusters were observed after treatment with an EGF receptor (EGFR) inhibitor (gefitinib), even in high EGF media (Supporting Information, Fig. 4.5E, 4.5F). A sec- ond mitogen-dependent mammary epithelial cell line, hTERT-HME1, also exhibited clustering in reduced growth factor conditions across a range of densities (500, 1000 cells) and after gefitinib treatment (data not shown). In contrast, clustering was not observed with the highly metastatic breast adenocarcinoma cell line, MDA-MB-231, which is transformed and not mitogen dependent (data not shown). These experi- ments suggest that clustering only occurs for mitogen-dependent mammary epithelial cells and is driven by decreased EGFR signaling, either through reduced exogeneous EGF or by chemical inhibition. Figure 4.1: Mammary epithelial cells cluster under reduced growth factor conditions. Mammary epithelial cells form confluent monolayers over 60 h in high EGF (A), but branching clusters in low EGF (B). Cell proliferation is decreased in low EGF (C), and cell migration is also significantly arrested (D). Cells exhibit statistically significant spatial clustering in low EGF according to Ripley’s H function (E). 138 Proliferation rates in high and low EGF media were quantified by detecting the number of fluorescent nuclei present over the duration of the experiment. In high EGF media, MCF-10A cells exhibited exponential growth with a ten-fold increase in cell density over 60 h (Fig. 4.1C). In low EGF media, cells exhibited logistic growth where cell density reached an inflection point at 30 h and plateaued at a four-fold increase in cell density by 40 h. Next, individual cell speeds were determined by comprehensively tracking nuclear motion of all cells over time. In high EGF media, MCF-10A cells initially exhibited a population averaged speed of ∼25 µm/h, which gradually decreased to ∼7 µm/h by 60 h (Fig. 4.1D). In low EGF media, cells started out with average speeds of ∼12 µm/h, gradually slowing to ∼2 µm/h by 60 h. This decrease in average speed occurred with comparable kinetics across all initial cell densities, suggesting that arrested motion was not dependent on overall cell density (Fig. 4.1D, inset). Although overall cell density was consistently sparser in low EGF media relative to high EGF media, the cells appeared to be locally clustered together rather than uniformly dispersed (Fig. 4.1B). The non-uniformity of this spatial distribution was evaluated using the Ripley H-function, which compares the number of cells located within a certain distance with the number expected from a random Poisson distribu- tion [296]. Essentially, positive values of the H-function indicate that cells tend to be located closer together than random at a given length scale, while negative values indicate cells are farther apart. In high EGF media, cells were randomly distributed and were neither significantly closer nor farther apart than 99% of randomly gener- ated distributions with the equivalent number of cells (Fig. 4.1E). Instead, in low EGF media, cells were significantly closer together at radial distances through 300 µm relative to 99% of random distributions. Note that this analysis only evaluates the spatial proximity of nuclei, although it is consistent with the appearance of physi- cally connected clusters. Overall, these combined observations indicate that low EGF 139 media results in the formation of multicellular clusters and spanning networks despite diminished cell proliferation and motility. Figure 4.2: Dynamics of cluster aggregation in low EGF. Dispersed individual cells exhibit random migration but arrest into clusters (A); Scale = 50 µm, cell tracks over 5 h shown (red line). The fraction of individual cells decreases over time (B). The number of clusters decreases over time (C), and the average cluster size increase (D). The radius of gyration scales as cluster size as a power law with fractal dimension (E). 4.2.2 Dispersed Individuals Aggregate into Multicellular Clusters with Fractal-like Morphology Clustering dynamics were then elucidated from comprehensive single cell tracking over the duration of the experiment. Initially dispersed cells migrated randomly, but their motion was arrested upon encountering other cells (Fig. 4.2A). Based on visual 140 inspection, individual cells were defined as cells whose nuclei which were located at least 75 µm from any other nuclei. Based on this definition, roughly 10-40% of individuals were typically observed at the start of the experiment, which decreased to a minimum of 10% or less over ∼24 h (Fig. 4.2B). For these individuals, the ensemble averaged mean square displacement h∆r2 (τ )i scaled linearly in time from τ = 1 h to τ = 10 h, indicating that their migration could be treated as an unbiased random walk over these timescales (Fig. 4.2B, inset). Multicellular clusters were then defined by computationally linking nuclei located less than 75 µm apart, which was verified by segmenting cluster morphologies in the cytoplasmic channel (data not shown). This relatively high size cutoff was chosen to correctly associate elongated cells with a cluster, but more compact cells within a cluster were typically located less than 50 µm apart [156]. For each field of view, ∼10 clusters were typically observed, a number which remained relatively stable over ∼24 h, indicating that cells aggregated irreversibly and that clusters did not dissociate (Fig. 4.2C). Nevertheless, the number averaged cluster size hM i increased steadily over time, reaching 10-100 cells depending on initial density (Fig. 4.2D). This increase could be attributed both to migratory individuals being “captured” at the cluster periphery, as well as limited proliferation. At later times, large clusters (> 10 cells) exhibited a dendritic, non-compact morphology with geometrically “rough” features at the periphery (Fig. 4.2A). In particular, cells at the interior were tightly connected with many neighbors, while cells at the periphery were less connected with fewer neighbors, indicating a decrease in local cell density with increasing radial distance. Both the cluster morphology and this radially decreasing density were strikingly reminiscent of fractal-like structures associated with the diffusion-limited aggregation of colloids [282, 283, 284]. A quan- titative signature of fractals is the fractal dimension Df , which scales the radius of gyration Rg with the cluster size M as a power law: Rg = M 1/Df . For these multi- 141 cellular clusters, the number of cells per cluster scales with the radius of gyration as Df = 1.73, for all clusters of at least 4 cells over varying densities and early times (Fig. 4.2E). This fractal dimension was also evaluated independently using a box- counting analysis of the cluster morphology over varying length scales, which yielded a similar Df = 1.7. It should be noted that Df = 1.7 is also observed experimentally and computationally for diffusion-limited aggregation of colloids, which exhibit ran- dom diffusive motion but adhere irreversibly on contact [282, 283, 284]. Overall, this motility-driven aggregation of living cells into stationary fractal-like clusters exhibits striking analogies with the diffusion-limited aggregation of (non-living) colloids. 4.2.3 Transient Collective Migration Connects Clusters into Spanning Networks Within stationary clusters, small groups of cells often exhibited transient but highly correlated motions (Fig. 4.3A). These spatially heterogeneous dynamics were quan- tified by classifying cells as “mobile” (“immobile”) based on whether they traveled more (less) than 1 nuclear diameter (10 µm) over some time interval τ , corresponding to 0% (100%) overlap, respectively. This cutoff distance is comparable to the 15% of a cell diameter used elsewhere for jamming in epithelial monolayers [292]. The self-overlap function hQ(τ )i was calculated by averaging these overlap values for all cells within a time frame, then ensemble averaging over a range of start times [297]. Essentially, hQ(τ )i represents the fraction of immobile cells after some time duration τ has elapsed. For instance, hQ(τ )i = 1 at τ = 0 h but decreases to to hQ(τ )i = 0.5 at τ = 2 h at early times (< 12 h), indicating that roughly half of the cells were mobile (non-overlapping) after 2 h (Fig. 4.3B). At later times (> 12h), hQ(τ )i in- creased, showing that a larger fraction of cells was immobile and that cell motility was more arrested over time. It should be noted that hQ(τ )i does not account for tem- poral correlations between mobile cells. To address this, the four-point susceptibility 142 χ4 (τ ) = N [hQ(τ )2 i − hQ(τ )i2 ] can be calculated from the moments of hQ(τ )i and the total number of cells N . At early times (< 12 h), the peak χ∗4 (τ ) occurs at τ ∼ 1.5 h, which represents to a characteristic lifetime of correlated motion (Fig. 4.3B). At later times, χ∗4 (τ ) shifts slightly to τ ∼ 2 h. The peak height χ∗4 (τ ) decreases from 5 cells down to 2 cells, which likely reflects the increase in cell number N over time. To examine the motility of these collective groups more carefully, spatial correlations were calculated explicitly from the cell tracks based on this timescale τ ∼ 1.5 h. At each time, mobile cells were identified if they traveled more than one nuclear diameter within 1.5 h (7 µm/h), then linked together based on spatial proximity of their nuclei. Typically, mobile subclusters were identified ranging from 5-30 cells or more. Correlations in directional motion were determined from the scalar product of the normalized velocity vectors Cvv (rij , t) = hvi (t) · vj (t)ii6=j , where brackets denote an ensemble average over all cell pairs i, j separated by a distance rij . Cells within mobile subclusters were typically more correlated in direction, with Cvv ∼ 0.6 at small separations rij ∼ 20 µm relative to Cvv ∼ 0.4 for all cells (Fig. 4.3C). These velocity correlations were roughly exponential and became longer ranged over time, with a characteristic timescale ξ ∼60 µm for the mobile subclusters relative to ξ ∼30 µm for all cells at 60 h. These increasing correlations in velocity could be attributed to maturation of cell-cell junctions [291]. At the completion of the experiment, cells were immunostained for E-cadherin, which was strongly localized at cell-cell junctions (Fig. 4.3C, inset). It is worth noting that for OHT treated cells in high EGF media, there is strong expression of vimentin, while E-cadherin junctions are completely absent (data not shown). Further, DMSO treated cells in high EGF media express strong E-cadherin junctions and low levels of vimentin, while cell junctions in low EGF are even more exaggerated with an intercalated phenotype, which is consistent in OHT. Vimentin expression is similarly increased for DMSO treated cells in low EGF, particularly for cells at the periphery of clusters (data not shown). Overall, 143 subclusters in low EGF media encompass an intriguing partial EMT phenotype, which may drive their ability to coordinate multicellular motion. Mobile subclusters also appeared to exhibit non-compact, but connected mor- phologies, analogous to the fractal-like structure of the overall clusters. Thus, the scaling of the radius of gyration Rg,s and subcluster size Ms were again fit to a power law, revealing a fractal dimension Df ∼ 1.62 (Fig. 4.3D), comparable to the previous value for clusters. It should be noted that this scaling only spans over 1.5 decades, so some caution is necessary in its interpretation. Nevertheless, one explanation is that a spatially connected subset of cells located on an underlying fractal geometry will also acquire a fractal-like structure. Some mobile subclusters were observed to connect isolated clusters together to form a large, continuous cluster or spanning network. Upon careful examination of single cell behaviors within mobile subclusters, it was observed that individual leader cells frequently emerged at the cluster periphery, which drove the highly directional motion of follower cells (Fig. 4.3E). Further, adjacent mobile subclusters were ob- served to merge together, which often coincided with a leader cell from each cluster that directed the motion of followers towards each other (Fig. 4.3F). The emergence of leader cells occurred with ∼2x greater frequency for OHT conditions compared to DMSO, which explains the enhanced formation of spanning networks for OHT treated cells in low EGF (Supporting Infromation, Fig. 4.5B, 4.5D). Overall, cells within clusters exhibited transient bouts of collective motion guided by leader cells, which ultimately resulted in the convergence of local clusters into a highly spanning network. 144 Figure 4.3: Emergence of collective migration behaviors in low EGF. Transient collective migration events occur within otherwise stationary clusters (A); Scale = 100 µm, red spots represent nuclei positions, cell tracks over 5 h shown in inset (red line). The dynamic four-point susceptibility function identifies a characteristic timescale for collective migration (B). Mobile cells exhibit highly correlated migration relative to the bulk population (C); Scale = 50 µm. Mobile cells also exhibit fractal- like scaling (D). Leader cells drive outwards connections between clusters (E, F); Scale = 25 µm, red arrows indicate displacement over 5 h (E), actual migration paths for leaders (red) and followers (pale orange) over the indicated time intervals (F). 145 4.2.4 Clustering and Jamming are Governed by Local Cell Density and EGF Cells were observed to accumulate greater numbers of neighbors over time during the transitions from dispersed individuals to aggregated clusters. As a measure of local cell density, the number of neighbors (“bonds”) hB(t)i was determined by counting all nuclei located within 75 µm of a given cell, then ensemble averaging over all cells within a field of view. For the lowest initial density condition (500 cells), the transition from dispersed individuals to aggregated clusters occurred at hBi ∼ 4 when t ∼ 24 h (Fig. 4.4A). Similarly, for a higher density condition (1000 cells), this transition to aggregated clusters occurred at hBi ∼ 4 when t ∼ 20 h, while the subsequent formation of spanning networks occurred hBi ∼ 7 when t ∼ 40 h. Higher density conditions formed clusters even more rapidly (by 20 h), which also connected together soon afterwards (by 30 h). It should be noted that cells typically pack together more closely than 75 µm within epithelial monolayers, so that this calculation likely overestimates the number of cells in direct proximity at higher densities. It should be noted that hBi ∼ 4 is consistent with Maxwell’s isostatic condition for rigidity in a two-dimensional network. Nevertheless, hBi ∼ 7 will likely depend on the architecture of the spanning network, particularly since the connections between clusters can be several cells wide. Overall, hB(t)i captures the transitions to clusters and spanning networks across several experimental conditions. The experimental regime where clustering and arrested motility occurred was characterized by systematically varying the initial EGF concentration from 0 ng/mL up to the 20 ng/mL typically used to culture MCF-10A cells. For low [EGF] < 0.5 ng/mL, cells were observed to form clusters (hBi < 4) and spanning networks (hBi < 7) (Fig. 4.4B), with proliferation arrested at subconfluent cell densities (data not shown). Moreover, for [EGF] ≤ 0.075 ng/mL, cell motility was largely arrested for both clusters and spanning networks (< 7 µm/h), while at [EGF] = 0.1ng/mL, cell 146 motility was only arrested in spanning networks. At high [EGF] ≥ 0.5 ng/mL and above, cells typically migrated much more rapidly and did not form stable clusters (Fig. 4.4B), but proliferated exponentially. Motility was consistently arrested at (hBi ∼ 13), which correlated to the formation of a confluent, space-filling monolayer. It should be noted that these EMT-induced cells did not express strong E-cadherin at cell-cell junctions for high concentrations of EGF, suggesting that the slowdown in motility may occur through transient cell-cell contacts, perhaps due to contact inhibi- tion of migration. Overall, these results indicate that increasing EGF concentration enhances both motility and proliferation, and that cluster formation and arrested motility occur for [EGF] ≤ 0.1 ng/mL and hBi < 14. 4.3 Discussion and Conclusion This work reveals a previously unobserved, two-step process of multicellular pattern formation governed by cell-cell adhesion and collective migration. First, we find that dispersed mammary epithelial cells cultured in reduced EGF exhibit random individ- ual migration that is arrested upon cell-cell contact. As a consequence, individual cells aggregated into multicellular clusters with a dendritic morphology, which can be quantified with a fractal dimension Df ∼ 1.7. This scaling has been previously observed for non-living colloidal particles that strongly attract, resulting in diffusion- limited aggregation [282, 283, 284]. Previous computational studies have suggested that such non-compact clusters can only form when particles travel via a random walk (i.e. diffusion). Instead, more persistent (i.e. ballistic) motion will result in compact clusters with Df ∼ 2.0 [298]. Similarly, if particles exhibit weaker attrac- tion, they will reversibly sample multiple attachment points and aggregate into a compact cluster (e.g. reaction-limited aggregation) [299]. Experimentally, as EGF concentrations are increased, mammary epithelial cells migrate faster with longer 147 Figure 4.4: Cell behavior as a function of EGF concentration reveals a jamming-like phase diagram. Local neighbor density is a readout of individual, clustered and spanning phases (A). A jamming-like phase diagram can be defined based on local neighbor density and EGF concentration (B). persistence times, which hinders the formation of strong cell-cell adhesions and thus non-compact clusters. It is tempting to speculate on the potential biological impli- cations of multicellular cluster formation under these conditions. For instance, the formation of cell-cell adhesions in epithelial clusters may compensate for anoikis and provide a survival benefit in stressful microenvironments [300]. Analogous behav- iors also occur during developmental processes, when cells chemotax into compact 148 clusters in vivo. In particular, mesenchymal stem cells undergo “condensation” with locally arrested cell migration and increased cell density during the formation of bone and cartilage [277]. Moreover, Dictyostelium cells under starvation conditions will “swarm” into multicellular mounds that differentiate into fruiting bodies for sporula- tion [301]. These similarities may reflect a conserved program of molecular signaling driven by biophysical interactions between cells. Second, small groups of cells exhibit transient collective migration within other- wise stationary clusters. These motile groups were often located at the cluster pe- riphery and migrated outwards to connect isolated clusters into spanning networks. Treatment with OHT to activate Snail resulted in increased numbers of “leader cells,” which exhibited elongated morphologies and ruffled lamellipodia at the leading edge, which are consistent with a “partial” EMT phenotype [274, 302, 275]. These leader cells were mechanically coupled with a group of follower cells, and exhibited highly correlated directional motion. These transient behaviors are not observed during cluster-cluster aggregation in colloidal particles, where an entire cluster will diffuse as a whole to approach another [303]. Instead, cooperative rearrangements of motile particles are a signature of dynamic heterogeneity in soft matter systems such as su- percooled liquids, colloidal glasses, and granular materials [286]. Interestingly, motile particles are often observed as quasi-one-dimensional “strings” [297], which are rem- iniscent of the single file “strands” of cells observed here. This phenotypic plasticity may provide insight into the morphogenesis of vascular or neuronal networks, which often feature leader cells forming multicellular connections in the context of on at- tractive and repulsive cues [304]. An exciting prospect is to manipulate these cellular aggregation processes using mechanical and biochemical cues to engineer complex tissue architectures that recapitulate form and function in vivo. Finally, we show that epithelial cell migration can be arrested at relatively sparse cell densities at reduced EGF concentrations, which differs from previous studies 149 based on fully confluent monolayers cultured under standard serum conditions [288, 289, 290, 291, 292, 156, 293]. It has been hypothesized that these behaviors can be captured by a jamming-like glass transition governed by motility, cell-cell adhesion, and density [294], analogous to soft matter systems governed by temperature, shear, and density [285]. Our experimental findings are qualitatively consistent with this hypothesis, since decreasing EGF results in slower (average) motility, which permits jamming at lower overall densities. Nevertheless, the lower cell density regime investi- gated here is more reminiscent of gelation, since cells organize into tenuous networks that are tightly packed locally [281]. These experiments may exhibit further “aging”- like behaviors based on the gradual depletion of EGF, cell proliferation [288], and the maturation of cell-cell junctions [291]. In this context, the similarities and differences between “active” living cells and “passive” non-living particles represent an exciting direction for future work, particularly the role of non-equilibrium fluctuations in self- propelled entities. Given the phenotypic heterogeneity of single cells, it is remarkable that their emergent behaviors can be captured using these minimal physical analogies. In summary, we demonstrate that dispersed mammary epithelial cells cultured un- der low EGF conditions migrate into aggregated clusters with fractal-like morpholo- gies. Subsequently, groups of cells within the cluster exhibit transient and collective migration to link clusters together into spanning networks. These linkage events are often led by elongated leader cells, which occur more frequently after tamoxifen treatment to activate Snail and EMT. These behaviors exhibit unexpected physical analogies with diffusion-limited cluster aggregation in non-living colloidal particles. Indeed, EGF is an experimental control parameter to regulate cell speed and den- sity, which tunes a jamming-like transition at subconfluent densities. Overall, these results link active, non-equilibrium systems with gelation and the glass transition in soft matter, with potential biological implications for epithelial morphogenesis in development, wound-healing, and cancer progression. 150 4.4 Supplementary Information 4.4.1 Supplementary Methods Cell Culture The bulk of experiments, cell tracking, and image analysis were conducted using human mammary epithelial cells (MCF-10A). Two variants of MCF-10A cells with inducible expression of Snail1 were used: one unlabeled cell line for immunostaining (MCF10A SS) and one with stable expression of fluorescent proteins in the cytoplasm (GFP) and nucleus (mCherry-H2B) for cell tracking and image analysis (MCF-10A XMAS SS). Both cell lines were generously gifted by G. Smolen and D. Haber (Mas- sachusetts General Hospital) [255, 89]. MCF-10A cells were routinely cultured in growth media consisting of DMEM/F12 HEPES buffer (Fisher 11330057) supple- mented with 5% horse serum (Fisher 16050122), 20 ng/ml Animal-Free Recombinant Human Epidermal Growth Factor (EGF; PeproTech AF-100-15), 0.5 mg/mL hydro- cortisone (Sigma H0888), 100 ng/mL cholera toxin (Sigma C8052), 10g/mL Insulin from bovine pancreas (Sigma I1882), and 1% Penicillin-Streptomycin (Fisher MT- 30-002-CI) [257]. For low EGF conditions, media was identical except for a lowered serum content (2% horse serum) and EGF level (0.075ng/mL, or as indicated). Stable expression of Snail1 (Snail6SA variant) was induced via an estrogen receptor response element through the addition of 4-hydroxytamoxifen (OHT, 500nM) to initiate an EMT program [255, 89]. Gefitinib was used at 500nM in DMSO to block EGFR activity and DMSO content was kept at 0.05% across all conditions. Results obtained from the MCF-10A cell line were corroborated by conduct- ing analogous experiments with human mammary epithelial cells immortalized with hTERT (hTERT-HME1, ATTC CRL-4010), which are similarly mitogen-dependent and routinely cultured in EGF containing media. Briefly, hTERT-HME1 cells were grown in Mammary Epithelial Cell Growth medium (MEGM), which is composed 151 of Mammary Epithelial Basal Medium (Lonza CC-3151) supplemented with a Bul- letKit (Lonza CC-4136) containing bovine pituitary extract, human EGF, hydrocorti- sone, insulin, and gentamicin/amphotericin-B. Conversely, mitogen-independent and highly metastatic breast adenocarcinoma cells (MDA-MB-231) were used as a nega- tive control. MDA-MB-231 cells expressing green fluorescent protein in the nucleus (GFP-H2B) were a generous gift from R.J. Giedt and R. Weissleder. MDA-MB-231 cells were grown in Dublecco’s Modified Eagle’s Medium containing high glucose and glutamine (Fisher MT-10-013-CV), which was supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin. All cell types were maintained under humid- ified incubation (37◦ C, 5% CO2) for routine culture and during time-lapse imaging experiments. 152 4.4.2 Supplementary Figures Figure 4.5: Fluorescence microscopy images of cells clustering under low EGF condi- tions. Cells (GFP, false-colored gray), and nuclei (RFP, red) for cells seeded at a range of densities (500-3000 cells per well) for high and low EGF, with and without OHT and Gefitinib, as indiciated. Scale = 100 µm. 153 Chapter 5 3D Organoid Disorganization after the Epithelial-Mesenchymal Transition Chapter 5 is ongoing work with contributing authors: S.E. Leggett, Thomas M. Valentin, Marielena Gamboa Castro, Zachary J. Neronha, and I.Y. Wong. 5.1 Abstract Multicellular disorganization and dissemination in 3D matrix can mimic tissue mor- phogenesis in development and disease. In particular, the epithelial-mesenchymal transition (EMT) is associated with a loss of cell-cell adhesions and a gain in cell motility, resulting in collective or individual invasion phenotypes. However, EMT has historically been studied using population averages and endpoints, which may be inadequate to resolve single cell heterogeneity and phenotypic plasticity. Here, 154 we quantitatively profile heterogeneous epithelial and mesenchymal phenotypes for mammary epithelial cells embedded in engineered 3D silk-collagen hydrogels. Using live cell tracking, we demonstrate that induction of the EMT master regulator Snail leads to the dissolution of compact epithelial organoids, coinciding with a shift from collective to individual cell migration. Further, comprehensive profiling of the dy- namic shape changes of Snail induction revealed a shift from spherical to elliptical morphologies, which were associated with apparent differences in cell migration dy- namics. Notably, epithelial cells exhibited characteristic periodic motion, while EMT induction leads to more stochastic motion with straighter trajectories. Overall, this approach enables the direct measurement of heterogeneous cell populations, including rare and exceptional phenotypes. We envision that our 3D culture assay may serve as a powerful tool for preclinical screening of anticancer chemotherapeutics and a platform for testing patient samples for personalized medicine. 5.2 Results Our lab is currently developing a 3D cell culture system using silk fibroin / colla- gen 1 (SF/COL) composite hydrogels with modifiable mechanical and biochemical properties for the study of cancer cell invasion and drug resistance. Silk fibroin is extracted from Bombyx mori cocoons using an approach developed by Rockwood et al [305]. Briefly, silkworm cocoons are degummed by boiling in sodium carbonate so- lution, rinsed with ultrapure water, dried, and the degummed silk is then dissolved in lithium bromide solution at 60◦ C. Lastly, the silk fibroin solution is dialyzed against ultrapure water over a 2 day period with 6 water changes at increasing time intervals. The purified silk fibroin solution is then stored at 4 ◦ C until ready for use. Compos- ite SF/COL hydrogels are made by mixing SF solution with appropriate cell culture medium, sonicating the SF-media solution to speed gelation, mixing in collagen I, 155 adding in the desired number of cells from a single cell suspension, and the resulting mixture is pipetted into a 96-well plate and subsequently incubated at 37◦ C under humidified conditions with 5% CO2 (Figure 5.1A). Silk fibroin only hydrogels form large local aggregates due to rapid β-sheet formation and lack binding sites for cell adhesion. Collagen I hydrogels form highly fibrillar matrices with many sites for cell adhesion, however, these matrices bias epithelial cells toward a mesenchymal pheno- type [8]. In contrast, composite SF/COL hydrogels form a homogenous matrix with Figure 5.1: Preparation and Characterization of SF/COL Hydrogels for Cell Culture in 3D. (A) Illustration depicting the general process for engineering composite Silk Fibroin (SF) and collagen I (Col I) hydrogels for 3D cell culture. SF is extracted from silk cocoons and purified, which will form characteristic β-sheets upon hydrogel polymer- ization. Collagen I naturally forms triple helices following gelation, which interact with SF β-sheets to form a homogenous matrix through hydrogen bonding interac- tions, and provides sites for cell-matrix adhesions. (B) Confocal reflectance image (max intensity projection) of Silk Fibroin (1%), Collagen 1 (0.1%), and composite SF/COL hydrogels (1%/0.1%), from left to right; scale =10 µm. more highly dispersed SF and smaller collagen I fibrils compared to either parent gel alone (Figure 5.1B). Uniquely, the stiffness of composite SF/COL hydrogels can be independently tuned by changing the concentration of SF while keeping the collagen I concentration constant. In contrast, changing the concentration of collagen I in colla- gen only hydrogels results in a shift in many mechanical and biochemical features such 156 as matrix stiffness, poresize, ligand density, etc. Thus, the composite SF/COL hydro- gels presented here are a desirable candidate for 3D culture due to the precise control over biochemical and mechanical properties, which removes confounds imposed by various other matrices (collagen I, matrigel, etc). Here, we present preliminary data to demonstrate the utility of composite SF/COL hydrogels for longitudinal studies of EMT induction using live cell microscopy and subsequent comprehensive analysis of cell morphology and migration dynamics. Human mammary epithelial cells with inducible expression of Snail1 (MCF-10A- Snail16SA ) and stable expression of fluorescent proteins in the cytoplasm and nucleus (green and red, respectively) were embedded in SF/COL hydrogels from a single cell suspension. For long term culture studies, cells were imaged every 3 days using con- focal microscopy and media was replenished following imaging. Cells treated with DMSO form organized, acini-like structures after several days, which lumenize over time and form glandular-like structures by day 15 (Figure 5.2A). In contrast, cells treated with 4-hydroxytamoxifen (OHT) to induce Snail1 expression at the time of embedding in 3D form more elongated, branched clusters over several days, which ex- hibit outgrowth and single cell invasion by day 15 (Figure 5.2B). Further, cells treated with OHT for 72 hours prior to embedding in SF/COL hydrogels are clearly elon- gated and more highly dispersed, with extensive protrusion formation within small clusters (Figure 5.2C). These structures are reminiscent of epithelial organization for DMSO treated cells, compared to a more mesenchymal phenotype for OHT treated cells. To confirm this, paired experiments were conducted with uncolored (MCF-10A- Snail16SA ) cells, which were fixed and immunostained for the mesenchymal biomarker, vimentin, as well as a nuclear (Hoechst) and cell (CellMask) markers. DMSO treated cells displayed compact nuclei with low levels of perinuclear vimentin, consistent with an epithelial phenotype (Figure 5.2D). Cells treated with OHT at the time of em- bedding displayed elongated nuclei with high levels of vimentin throughout the cell, 157 with notably high expression for the most peripheral cells with an invasive phenotype (Figure 5.2E). This result was analogous to that of cells treated with OHT prior to embedding (Figure 5.2F), both consistent with a mesenchymal phenotype. Further, we aimed to convert these qualitative observations of EMT phenotype to quantitative readouts by tracking cell morphology. Figure 5.2: 3D Epithelial Organization is Disrupted with the Induction of the Epithelial-Mesenchymal Transition. Representative confocal z-slice images of human mammary epithelial cells with cytoplasmic GFP and nuclear RFP expression. (A) Epithelial cells form compact, acini-like structures in 3D silk-collagen hydrogels (B) Epithelial cells treated with OHT to induce Snail expression at the time of 3D embedding form branched structures that dissociate over time. (C) Epithelial cells that were induced to undergo EMT in 2D (3 days prior to embedding) exhibit elon- gated, protrusive morphologies in 3D. Scale = 20 µm, GFP cytoplasm and RFP nucleus merged expression are shown at the indicated times after embedding in 3D silk-collagen. (D-F) Confocal max intensity projection for immunostained cells after 10 days in silk-collagen for epithelial cells (D), induced EMT cells (E), and mes- enchyal cells (F), corresponding with treatment groups in (A-C), respectively. Cell cytoplasm labeled with CellMask (FarRed, false colored green), nuclei labeled with hoechst (blue), and vimentin expression (red), Scale = 20 µm. Morphology of cell clusters in SF/COL hydrogels was analyzed by thresholding and segmenting the cytoplasmic channel (GFP) for 3D confocal stacks across time using the Surfaces module in Bitplane Imaris (Figure 5.3A). At day 0 after embed- 158 Figure 5.3: Morphological Profiling of EMT in 3D. (A) 3D segmentation of multicellular clusters in Imaris for epithelial (DMSO), induced EMT (OHT, day 0), and mesenchymal (OHT, day - 3) cells at days 0, 3, and 6 after embedding in silk-collagen. Cell sphericity is shown as a spectrum from 0.25 to 1, as a best fit for the range of observed values; Scale = 50 µm. (B, C) Parallel coordinate plots displaying mean cell shape features for cell sphericity (B) and prolate ellipticity (C) for each condition across time as depicted in (A). ding in 3D SF/COL, individual cells exhibited a highly spherical morphology with sphericity >0.9 across all conditions. Sphericity remained high for DMSO treated cells over time, while cells treated with OHT prior to embedding displayed a rapid decrease in sphericity from day 0 to day 3, which remained low through day 6. In contrast, cells treated with OHT at the time of embedding exhibited a more grad- 159 ual decrease in sphericity from day 0 to day 3, with a rapid drop across day 3 to day 6 (Figure 5.3B). On the other hand, the ellipticity, or elongated nature of cell clusters, displayed an inverse trend with sphericity. DMSO treated cells displayed a gradual decrease in ellipticity over time, while cells treated with OHT prior to em- bedding underwent rapid increases in ellipticity over days 0 to 3, and cells treated with OHT at the time of embedding displayed a more marginal increase over days 3 to 6 (Figure 5.3C). Overall, this morphological analysis revealed kinetic differences in the adoption of a mesenchymal morphology (relative low sphericity, high ellipticity) across OHT conditions, and confirmed the maintenance of an epithelial morphology (relative high sphericity, low ellipticity) for DMSO treated cells. Next, we tracked single cell migration to determine if these changes in EMT biomarker expression and morphology within cell clusters were associated with differential motility behaviors across conditions. To visualize single cell migration behaviors, confocal images of cells embedded in 3D SF/COL hydrogels were taken every 30 minutes over a one day period. Imaging was initiated 24 hours after embedding in order to allow for strong cell-matrix adhe- sions to form. Next, cell nuclear positions (RFP channel) were detected and linked into migration tracks using the Spots module in Bitplane Imaris (Figure 5.4A). The nuclear position data was exported and analyzed using custom MATLAB code to plot X,Y,Z positions as a function of time in order to visualize migration patterns (Figure 5.4B). Strikingly, epithelial cells treated with DMSO exhibited a highly regu- lar orbiting motion within two cell clusters. This smooth migration pattern could be well modeled by a sinusoidal wave function for X and Y positions over time, which yielded a characteristic periodicity of 5 hours (Figure 5.4C). On the other hand, cells treated with OHT at the time of embedding exhibited irregular motion with respect to each other, with more jagged fluctuations in X and Y position (Figure 5.4D). Lastly, cells induced to undergo EMT with OHT prior to embedding displayed highly 160 Figure 5.4: Automated single cell tracking reveals dynamic migration behaviors dur- ing EMT. (A) Confocal maximum intensity projection of representative epithelial cells with 2h tracks as detected by nuclear RFP location, to show the relative change in position over the indicated times (t); scale = 10 µm. (B) XYZ plot of the tracked positions (20 minute interval) for 2 different cells, indicated by unique colors that darken with time. (C-E) Plots depicting cell position in X and Y (left), time vs. X position (middle), and Y position vs. time (right ) for (C) epithelial (DMSO), (d) induced EMT (OHT at time of embedding, day 0), and (e) mesenchymal (OHT at day -3) over a 6h time span. An interval of 20 minutes was used for time lapse imaging, which began 24h after embedding to allow for adequate cell-matrix adhesions to form. Individual cells are represented by uniquely colored tracks, which gradually darken over time as a visualization aid. 161 straightened trajectories with uncorrelated motion with respect to adjacent cells in the same cluster (Figure 5.4E). Overall, this analysis reveals a migration signature associated with epithelial and mesenchymal cell motility, which can be described as a transition from regular, periodic motion to irregular, linear motion. An intriguing direction of future work would be to combine the analyses described here, includ- ing EMT biomarker expression, morphology, and migration dynamics to quantify epithelial and mesenchymal phenotype across time as a predictive platform for drug screening and personalized medicine. 5.3 Discussion and Future Directions Cells at the tissue-matrix interface encounter mechanical signals and forces that do not occur in 2D cell culture. Further, cancer cell populations exhibit extensive het- erogeneity at the single cell level, including differences in biomarker expression, mor- phology, and migratory behaviors, which are misrepresented with classical 2D assays yielding population-averaged readouts. In this thesis, we aim to address these issues by (1) engineering 2.5D (Chapter 2) and 3D (Chapter 5) microenvironments that recapitulate physiologically relevant features of the ECM for the study of cell-matrix biology and (2) developing quantitative single cell analyses using morphological pro- filing (Chapter 3) and tracking cell motion (Chapter 4) to characterize individual cell phenotype within heterogeneous populations. Ultimately, we envision that engineered microenvironments incorporating ECM features will advance technology development for biomedical engineering and provide a more physiogical platform for drug discovery and personalized medicine. By further applying our quantitiatve approaches to cells cultured in complex microenvironments (preliminary work, Chapter 5), we hope to reveal new insights into the mechanical and biochemical mechanisms of EMT with single cell resolution. 162 5.3.1 Chapter 2 Summary An intricate interplay exists between biological cells, which act as responders, and the microenvironment, which provides signals to regulate cell function and phenotype. Notably, the composition and structure of the extracellular matrix (ECM) can largely influence the maintenance of tissue architecture by providing spatial and mechanical cues that regulate cell differentiation and polarity, proliferation, adhesion, and cell behavior [306]. Extraordinarily, topographic features of the ECM exist at submicron length scales, but exert powerful control over multicellular tissues, including organ morphogenesis, function, and homeostasis [307]. Furthermore, aberrant remodeling of the tissue microenvironment plays a critical role in the development of pathologies, particularly in the context of cancer and fibrosis. For instance, matrix remodeling in these diseased states often leads to the formation of dense collagen I fibrils, which re- sults in a highly aligned and stiffened ECM [308]. These topographic cues are poorly represented by traditional 2D monolayer culture, which biases cell form and function through uniform mechanical and geometric cues, and thus does not provide a phys- iologically relevant microenvironment for the study of cell-material interactions. On the other hand, it has been shown that cells cultured on highly aligned nanoarchi- tectures and fibrous structures exhibit a phenomenon known as contact guidance, in which cells elongate and align in the direction of these topographies[199]. However, conventional methods for the fabrication of nanotopographies, including lithographic patterning and etching, have major limitations due to cost, complexity, and scala- bility [309]. To bridge this gap, in Chapter 2 we describe a simple method for the generation of wrinkled-graphene surfaces to direct cell-material interactions. Our fabrication method involves the wet deposition of graphene oxide (GO) onto pre-streched elastomers followed by relaxation and thermal treatment to stabilize the multilayered wrinkled films. Remarkably, the width and amplitude of the wrinkled substrates can be easily and independently tuned by modifying the GO film thick- 163 ness, or the elastomer pre-strain, respectively. This allows for the generation of a wide array of biocompatible topographies with the unique material properties of graphene, which may be widely applicable to researchers across disciplines. Further, we demon- strated that human and mouse fibroblasts cultured on these wrinkled topographies were highly viable and exhibited strong alignment and elongation along the direction of wrinkles. Thus, we envision the utility of these substrates for directing cell-material interactions for a variety of cell types. As previously discussed, increasing ECM density is associated with fibrosis and matrix remodeling in cancer. As such, an interesting direction of future work could be to examine how cells respond to decreasing wrinkle width, which may mimic the topographic cues associated with these ECM changes. For instance, it has been shown that EMT can be triggered by confinement of epithelial cell clusters [310], however, the method described here could be used to elucidate this process at the single cell level. Further, a unique application of this method could be to apply controlled mechanical actuation to cells by stretching and relaxing the underlying wrinkled sub- strates. Such experiments may provide novel insights into cell mechanotransduction pathways. Since the initial publication of this work, our lab has demonstrated the ability to generate more complex topographies [311] with desirable chemical, elec- trical, [312] and actuation properties [313]. Altogether, it would be intriguing to expand upon this materials research and investigate the cellular responses to these varied graphene substrates in combination with applied orthogonal stimuli. 5.3.2 Chapter 3 Summary Cancer cell populations often respond heterogeneously to potent chemotherapeutics and targeted therapies, exacerbating malignant tumor progression. One adaptive re- sponse to drug treatment and potential driver of heterogeneity may be the induction of EMT, which transforms compact, sensitive cells into an elongated, resistant phe- 164 notype. In this context, EMT leads to cancer cell invasion, metastasis, and amplifies drug resistance [314]. The identification of target compounds that inhibit or reverse EMT may thus be critical for improved patient survival. As such, it is essential to consider cancer cell plasticity and tumor heterogeneity in the drug discovery pro- cess. However, classical biochemical assays based on population averages over large numbers cannot resolve single cell heterogeneity and may mask the existence of rare cell subpopulations that influence therapeutic response. To address this technological gap, in Chapter 3 we developed a quantitative, single cell assay that has the potential to screen large heterogeneous cell populations for phenotypic variability. Specifically, we developed a guassian mixture model to distinguish between epithelial and mes- enchymal cells with high accuracy based on the quantification of cell and nuclear morphologies, as well as the mesenchymal biomarker vimentin. We demonstrated the applicability of this platform to resolve the kinetics of EMT induction in response to the master regulator Snail, transforming growth factor−β, and in response to the chemotherapeutic compound paclitaxel. Overall, we envision this approach may serve as a high-throughput predictive biological assay for the evaluation of novel therapeu- tics [231]. It is exciting to imagine the utility of this platform as a prognostic tool for guiding personalized medicine. Aberrant expression of EMT-transcription factors has been associated with worsening survival in patients with many carcinomas including lung, esophogeal, oral, ovarian, cervical, and breast cancer [315]. Such poor prognosis may be attributed to the ability of EMT to promote metastatic spread by supporting an enhanced invasive capacity, drug resistance, and cancer cell stemness [314]. As such, the precise quantification of the contribution of epithelial and mesenchymal-like cancer cells from a patient biopsy may provide critical insight for the cancer staging process and development of individual treatment plans. However, it is important to consider the diverse cell types found within the tumor microenvironment. 165 In addition to the classical epithelial and mesenchymal cells that can be detected with our model, tumors are often highly heterogeneous and contain a variety of ad- ditional stromal cell types such as immune cells (e.g. macrophages, T-cells, etc), cancer-associated fibroblasts (CAFs), mesenchymal stem cells, endothelial cells, and pericytes. Further, cancer cells that have undergone an EMT are strikingly similar to the native fibroblasts derived from EMT during embryogenesis and wound healing, and thus a common underlying biochemical and phenotypic signature exists [316]. As a result, the analysis of epithelial and mesenchymal subpopulations from tissue biopsies may be skewed towards a larger mesenchymal fraction due to the presence of native fibroblasts and CAFs. However, activated fibroblasts in the tumor stroma can elicit cancer-promoting effects and an increased density of CAFs is associated with poorer prognosis for several human cancers including breast, lung, liver, and esophogeal cancer [317],[318]. As such, CAFs may serve as an additional prognostic marker and potential therapeutic target [319],[318]. Thus, an intriguing future di- rection is to apply the epithelial/mesenchymal classification scheme described here to quantify the contribution of EMT and aberrant stroma in patient samples, which may improve prognostic power and refine cancer treatment decisions. 5.3.3 Chapter 4 Summary Transitions between collective and individual migration are prominent events in tis- sue morphogenesis, wound healing, and cancer invasion and metastasis. In particular, the mode of cell migration may be largely influenced by both cell phenotype and mi- croenvironmental stimuli. For instance, cells can adopt various modes of single cell migration (mesenchymal, amoeboid) and collective migration (leader cell driven, mul- ticellular streaming, strand-like invasion, etc.), while also demonstrating plasticity of invasion [1]. The epithelial-mesenchymal transition (EMT) encompasses a classical transition of collectively migrating epithelial cells to individually scattering mesenchy- 166 mal cells [89]. Nonetheless, EMT is not a binary, switch-like process, but rather gives rise to many intermediate phenotypes, in which cells express both epithelial and mes- enchymal biomarkers that can exhibit collective motility [320]. This “partial” EMT may be particularly important for cancer invasion, whereby partial-mesenchymal leader cells drive collective migration of partial-epitheial followers. However, such partial EMT phenotypes are notably difficult to observe due to their questionable stability and strong EMT stimuli studied in vitro. In Chapter 4, we demonstrate that in low epidermal growth factor conditions, the induction of EMT in mammary epithelial cells yields a unique partial EMT phenotype that drives the self organization of cells into a spanning network. Mitogen dependent mammary epithelial cells are typically cultured in high levels of growth factors to sustain exponential proliferation in 2D culture. However, such saturating levels of growth factors may obscure cell signaling and hinder collective behaviors. Our research demonstrated that human mammary epithelial cells treated with DMSO, or OHT to induce EMT, exhibit a random walk with little cell-cell interactions at subconfluent densities in media with high levels of epidermal growth factor (20ng/mL EGF). In contrast, the same perturbations (DMSO or OHT) gave rise to a strikingly different phenotype in media with low levels of EGF (<0.1ng/mL), in which cells rapidly formed clusters that aggregated together over time. Surprisingly, while vimentin expression was increased in the OHT group, E-cadherin junctions were also maintained with the induction of Snail1 expresssion, which is a strong inducer of EMT and inhibitor of E-cadherin [255]. Thus, there appears to be some antagonism between EMT induction through Snail1 and reduced EGF signaling, giving rise to a partial EMT-like state. Uniquely, this partial EMT phenotype was associated with the emergence of increased numbers of leaders cells that drove the collective migration of clusters toward each other. As a consequence, cell clusters were highly interconnected even at very subconfluent densities (∼50% confluence), which could be described by 167 a jamming phase diagram of EGF concentration, number of cell neighbors, and cell velocity. Further, it is appealing to speculate on the potential biological implications of multicellular cluster formation. A provocative idea is that multicellular clustering may provide a survival bene- fit under stressful microenvironmental conditions, such as growth factor restriction demonstrated here. In particular, the formation of cell-cell adhesions may compen- sate for anoikis and provide a survival benefit in stressful conditions [300], which has been suggested for circulating tumor cell clusters traversing the bloodstream [321]. As such, an intriguing direction of future work is the identification of a cluster-promoting factor and its inducers. Further, the ECM consists of a complex array of secreted proteins and structural elements that provide adhesive molecules for cell attachment and mechanical support for the maintenance of tissue architecture. As such, the ECM can serve as a reservoir of sequestered growth factors that can participate in cell signaling pathways as migrating cells encounter them, and ligands can further be released upon degradation of the ECM [32]. Thus, an exciting direction of future work would be to culture cells in a 3D matrix where growth factors are tethered to the ECM, which may mimic growth factor signaling in the in vivo state. Overall, this works reveals a previously unobserved model of epithelial self organization, which may provide critical inisght into collective migration processes involved in embryogenesis, wound healing, and cancer progression. 5.3.4 Chapter 5 Summary The conversion of epithelial cells to a mesenchymal phenotype (EMT) is a fundamen- tal event in embryonic development, wound healing, and is implicated in organ fibrosis and metastatic cancer [223]. Epithelial tissues are composed of a continuous sheet of cells that rest on a basement membrane. During EMT, epithelial cell integrity is lost via the activation of a transcription program that downregulates epithelial genes and 168 induces the expression of mesenchymal genes [6]. EMT encompasses cytoskeletal re- arrangement, dissolution of tight junctions, degradation of the basement membrane, and results in a transition from apical-basal polarity to front-back polarity. Con- sequently, cells that undergo EMT have enhanced migratory capacity and reduced susceptibility to apoptosis [308]. Historically, EMT has been challenging to study due to the limitations of existing cell culture models. For instance, EMT occurs in vivo at the 3D interface between epithelial tissues and the surrounding matrix, which is inadequately modeled by the rigid, 2D surfaces used in cell culture. Further, matrix remodeling and aberrant composition of the ECM, which are prominent in fibrosis and cancer, may directly trigger EMT through interactions of epithelial cells with stiff collagen I bundles, as opposed to the basal lamina, which helps to maintain epithelial tissue integrity [4],[230],[42]. Although it is widely recognized that degra- dation of the basement membrane and matrix remodeling through deposition of ECM proteins can contribute to EMT and disease progression, the field of EMT research has been primarily focused on soluble signals, growth factors, and transcription fac- tor regulation of EMT, while the role of ECM topography on EMT induction is not fully understood and remains to be elucidated [42]. In Chapter 5, we engineer silk fibroin/collagen I composite hydrogels as a 3D cell culture platform to study EMT in a more physiologically relevant context than traditional 2D models. In this chapter, we show that silk fibroin/collagen I composite hydrogels can be used as a 3D matrix to support mammary epithelial cell culture over extended periods of time (> 2 weeks). Induction of EMT through the master regulator Snail1 led to the increased expression of the mesenchymal biomarker vimentin and vast changes in the morphology of multicellular clusters. We demonstrate that ellipticity and sphericity gradually shift over time in response to EMT induction, suggesting that these measures may serve as a foundation for identifying epithelial and mesenchymal phenotypes in 3D culture. We further show that epithelial cells cultured in 3D hy- 169 drogels move in periodic orbits, whereas EMT induction initially results in a loss of this coordinated motion, which further progresses to highly straightened trajectories for fully mesenchymal cells. Together, these findings reveal that tracking morphology and migration dynamics may provide insight into the profound cell shape changes and epithelial plasticity that are observed with EMT induction. Upon further characteri- zation of these dynamic phenotypes, we envision that this approach may identify rare and exceptional phenotypes within heterogeneous populations, as well as predict their preclinical response to targeted inhibitors. As such, the study of EMT in engineered microenvironments with 3D ECM cues may provide a more physiologically relevant model for the study of chemotherapeutic efficacy, drug resistance, and the identifi- cation of EMT inhibitors, which could have profound clinical impacts. Further, a major advantage of the hydrogels presented here is their biochemical and mechanical tunabilty, which could be further utilized to study how microenvironmental stimuli impact EMT phenotype. The physical properties of the ECM can greatly alter cell mechanics and migra- tion [322], including the induction of EMT. Understanding how these mechanical cues evoke a cellular response may provide crucial insight into the development and progression of fibrosis and cancer. An interesting direction of future work would be to tune the stiffness of SF/COL hydrogels and quantify the kinetics of EMT induc- tion across these conditions. Similarly, additional ligands could be added to these composite hydrogels (e.g. laminin, collagen IV, etc.) to examine their impact on cell morphology, migration, and EMT induction. Further, it has been proposed that targeting the tumor microenvironment could provide a therapeutic benefit to cancer patients [323]. Thus, the tunable matrices described here could be further used to evaluate therapeutic response to microenvironmental targeting compounds and their impact on cancer cell growth and invasion. Future incorporation of patient samples into 3D composite hydrogels could provide a powerful platform for longitudinal stud- 170 ies of initial drug response, resistance, and relapse. Overall, the technologies and quantitative methods described in this thesis are a stepping stone to understanding how heterogeneity contributes to cancer progression, which may provide crucial in- sight for prognostics and drug discovery. I am hopeful that future advances in in vitro culture systems, quantitative methods for predictive biology, and platforms for personalized medicine will speed the translation of basic research to impactful clinical outcomes. 171 Bibliography [1] Peter Friedl and Stephanie Alexander. Cancer invasion and the microenviron- ment: plasticity and reciprocity. Cell, 147(5):992–1009, November 2011. [2] M Angela Nieto, Ruby Yun-Ju Huang, Rebecca A Jackson, and Jean Paul Thiery. EMT: 2016. Cell, 166(1):21–45, June 2016. [3] Pernille Rørth. Collective cell migration. Annu. Rev. Cell Dev. Biol., 25:407– 429, 2009. [4] Pengfei Lu, Valerie M Weaver, and Zena Werb. The extracellular matrix: A dynamic niche in cancer progression. J Cell Biol, 196(4):395–406, February 2012. [5] Luo Gu and David J Mooney. Biomaterials and emerging anticancer therapeu- tics: engineering the microenvironment. Nat. Rev. Mater, 16(1):56–66, January 2016. [6] Joshua Z Gasiorowski, Christopher J Murphy, and Paul F Nealey. Biophysical cues and cell behavior: the big impact of little things. Annu. Rev. Biomed. Eng., 15:155–176, 2013. [7] Steven R Caliari and Jason A Burdick. A practical guide to hydrogels for cell culture. Nat Meth, 13(5):405–414, April 2016. [8] G Greenburg and E D Hay. Epithelia suspended in collagen gels can lose po- larity and express characteristics of migrating mesenchymal cells. J. Cell Biol., 95(1):333–339, 1982. [9] M H Barcellos-Hoff, J Aggeler, T G Ram, and M J Bissell. Functional differ- entiation and alveolar morphogenesis of primary mammary cultures on recon- stituted basement membrane. Development, 105(2):223–235, February 1989. [10] O W Petersen, L Rønnov-Jessen, A R Howlett, and M J Bissell. Interaction with basement membrane serves to rapidly distinguish growth and differentiation pattern of normal and malignant human breast epithelial cells. Proc Natl Acad Sci USA, 89(19):9064–9068, October 1992. [11] Jayanta Debnath and Joan S Brugge. Modelling glandular epithelial cancers in three-dimensional cultures. Nat Rev Cancer, 5(9):675–688, September 2005. 172 [12] Guillaume Charras and Erik Sahai. Physical influences of the extracellular en- vironment on cell migration. Nat Rev Mol Cell Biol, 15(12):813–824, December 2014. [13] Norman Sachs and Hans Clevers. Organoid cultures for the analysis of cancer phenotypes. Curr Opin Genet Dev, 24:68–73, February 2014. [14] Lewis J. Kleinsmith. Principles of Cancer Biology. Pearson Education Limited, 2013. [15] Rebecca L Siegel, Kimberly D Miller, and Ahmedin Jemal. Cancer Statistics, 2017. CA: Cancer J Clin, 67(1):7–30, January 2017. [16] Vinay Kumar, Abul K. Abbas, and Jon C. Aster. Robbins & Cotran Pathologic Basis of Disease (Robbins Pathology). Saunders, 2014. [17] Don X Nguyen, Paula D Bos, and Joan Massagu´e. Metastasis: from dissemina- tion to organ-specific colonization. Nat Rev Cancer, 9(4):274–284, April 2009. [18] Kambez H. Benam, Stephanie Dauth, Bryan Hassell, Anna Herland, Abhishek Jain, Kyung-Jin Jang, Katia Karalis, Hyun Jung Kim, Luke MacQueen, Roza Mahmoodian, Samira Musah, Yu suke Torisawa, Andries D. van der Meer, Remi Villenave, Moran Yadid, Kevin K. Parker, and Donald E. Ingber. Engineered in vitro disease models. Annu Rev Pathol, 10(1):195–262, 2015. PMID: 25621660. [19] Joan Massagu´e and Anna C Obenauf. Metastatic colonization by circulating tumour cells. Nature, 529(7586):298–306, January 2016. [20] Anette M Høye and Janine T Erler. Structural ECM components in the premetastatic and metastatic niche. AJP: Cell Physiology, 310(11):C955–67, June 2016. [21] L E Barney, L E Jansen, S R Polio, S Galarza, M E Lynch, and S R Peyton. The Predictive Link between Matrix and Metastasis. Curr Opin Chem Eng, 11:85–93, February 2016. [22] Valerie S LeBleu, Brian MacDonald, and Raghu Kalluri. Structure and Function of Basement Membranes. Exp. Biol. Med, 232(9):1121–1129, 2007. [23] K Beck, I Hunter, and J Engel. Structure and function of laminin: anatomy of a multidomain glycoprotein. FASEB J, 4(2):148–60, 1990. [24] C H Streuli, C Schmidhauser, N Bailey, P Yurchenco, A P Skubitz, C Roskelley, and M J Bissell. Laminin mediates tissue-specific gene expression in mammary epithelia. J Cell Biol, 129(3):591–603, May 1995. [25] Janna K Mouw, Guanqing Ou, and Valerie M Weaver. Extracellular matrix assembly: a multiscale deconstruction. Nat Rev Mol Cell Biol, 15(12):771–785, November 2014. 173 [26] Alexandra Naba, Karl R. Clauser, Huiming Ding, Charles A. Whittaker, Steven A. Carr, and Richard O. Hynes. The extracellular matrix: Tools and insights for the omics era. Mat Biol, 49:10 – 24, 2016. [27] Darci T Butcher, Tamara Alliston, and Valerie M Weaver. A tense situation: forcing tumour progression. Nat Rev Cancer, 9(2):108–122, February 2009. [28] Joe Swift, Irena L Ivanovska, Amnon Buxboim, Takamasa Harada, P C Dave P Dingal, Joel Pinter, J David Pajerowski, Kyle R Spinler, Jae-Won Shin, Manorama Tewari, Florian Rehfeldt, David W Speicher, and Dennis E Dis- cher. Nuclear lamin-A scales with tissue stiffness and enhances matrix-directed differentiation. Science, 341(6149):1240104, August 2013. [29] Ruchi Malik, Peter I Lelkes, and Edna Cukierman. Biomechanical and biochem- ical remodeling of stromal extracellular matrix in cancer. Trends Biotechnol, 33(4):230–236, April 2015. [30] Kandice R Levental, Hongmei Yu, Laura Kass, Johnathon N Lakins, Mikala Egeblad, Janine T Erler, Sheri F T Fong, Katalin Csiszar, Amato Giaccia, Wolfgang Weninger, Mitsuo Yamauchi, David L Gasser, and Valerie M Weaver. Matrix crosslinking forces tumor progression by enhancing integrin signaling. Cell, 139(5):891–906, November 2009. [31] Mikala Egeblad, Morten G Rasch, and Valerie M Weaver. Dynamic inter- play between the collagen scaffold and tumor evolution. Curr Opin Cell Biol, 22(5):697–706, October 2010. [32] Caroline Bonnans, Jonathan Chou, and Zena Werb. Remodelling the extracellu- lar matrix in development and disease. Nat Rev Mol Cell Biol, 15(12):786–801, December 2014. [33] Richard O Hynes. Integrins: bidirectional, allosteric signaling machines. Cell, 110(6):673–687, September 2002. [34] Kai Kessenbrock, Vicki Plaks, and Zena Werb. Matrix Metalloproteinases: Regulators of the Tumor Microenvironment. Cell, 141(1):52–67, April 2010. [35] Ruben Bill and Gerhard Christofori. The relevance of EMT in breast cancer metastasis: Correlation or causality? FEBS Lett, 589(14):1577–1587, June 2015. [36] Thomas Brabletz. To differentiate or not–routes towards metastasis. Nat Rev Cancer, 12(6):425–436, June 2012. [37] Patrick M¨uller and Alexander F Schier. Extracellular movement of signaling molecules. Dev Cell, 21(1):145–158, July 2011. [38] Jian Xu, Samy Lamouille, and Rik Derynck. TGF-β-induced epithelial to mes- enchymal transition. Cell Res, 19(2):156–172, February 2009. 174 [39] H´ector Peinado, David Olmeda, and Amparo Cano. Snail, Zeb and bHLH factors in tumour progression: an alliance against the epithelial phenotype? Nat Rev Cancer, 7(6):415–428, June 2007. [40] Jing Yang, Sendurai A Mani, and Robert A Weinberg. Exploring a new twist on tumor metastasis. Cancer Res, 66(9):4549–4552, May 2006. [41] Alain Puisieux, Thomas Brabletz, and Julie Caramel. Oncogenic roles of EMT- inducing transcription factors. Nature Cell Biology, 16(6):488–494. [42] Samy Lamouille, Jian Xu, and Rik Derynck. Molecular mechanisms of epithe- lial– mesenchymal transition. Nature Rev Mol Cell Biol, 15(3):178–196, March 2014. [43] Mahmut Yilmaz and Gerhard Christofori. EMT, the cytoskeleton, and cancer cell invasion. Cancer Metastasis Rev., 28(1-2):15–33, June 2009. [44] Richard G Hodge and Anne J Ridley. Regulating Rho GTPases and their regulators. Nat Rev Mol Cell Biol, 17(8):496–510, June 2016. [45] Thorsten M Koch, Stefan M¨unster, Navid Bonakdar, James P Butler, and Ben Fabry. 3D Traction Forces in Cancer Cell Invasion. PLoS ONE, 7(3):e33476, 2012. [46] Casey M Kraning-Rush, Joseph P Califano, and Cynthia A Reinhart-King. Cellular traction stresses increase with increasing metastatic potential. PLoS ONE, 7(2):e32572, 2012. [47] Jason Lowery, Edward R Kuczmarski, Harald Herrmann, and Robert D Gold- man. Intermediate Filaments Play a Pivotal Role in Regulating Cell Architec- ture and Function. J. Biol. Chem., 290(28):17145–17153, July 2015. [48] R L Ehrmann and G O Gey. The growth of cells on a transparent gel of reconstituted rat-tail collagen. J. Natl. Cancer Inst., 16(6):1375–1403, June 1956. [49] B D Walters and J P Stegemann. Strategies for directing the structure and function of three-dimensional collagen biomaterials across length scales. Acta Biomater, 10(4):1488–1501, April 2014. [50] Katarina Wolf, Mariska Te Lindert, Marina Krause, Stephanie Alexander, Joost Te Riet, Amanda L Willis, Robert M Hoffman, Carl G Figdor, Stephen J Weiss, and Peter Friedl. Physical limits of cell migration: control by ECM space and nuclear deformation and tuning by proteolysis and traction force. J Cell Biol, 201(7):1069–1084, June 2013. [51] Ya-li Yang, St´ephanie Motte, and Laura J Kaufman. Pore size variable type I collagen gels and their interaction with glioma cells. Biomaterials, 31(21):5678– 5688, July 2010. 175 [52] Ya li Yang, Lindsay M. Leone, and Laura J. Kaufman. Elastic moduli of collagen gels can be predicted from two-dimensional confocal microscopy. Biophys J, 97(7):2051 – 2060, 2009. [53] J L Bailey, P J Critser, C Whittington, J L Kuske, M C Yoder, and S L Voytik- Harbin. Collagen oligomers modulate physical and biological properties of three- dimensional self-assembled matrices. Biopolymers, 95(2):77–93, February 2011. [54] Frederick Grinnell and W Matthew Petroll. Cell motility and mechanics in three-dimensional collagen matrices. Annu Rev Cell Dev Biol, 26:335–361, November 2010. [55] R W Orkin, P Gehron, E B McGoodwin, G R Martin, T Valentine, and R Swarm. A murine tumor producing a matrix of basement membrane. J Exp Med, 145(1):204–220, January 1977. [56] Gabriel Benton, Irina Arnaoutova, Jay George, Hynda K Kleinman, and Jen- nifer Koblinski. Matrigel: from discovery and ECM mimicry to assays and models for cancer research. Adv. Drug Deliv. Rev., 79-80:3–18, December 2014. [57] Susanne Sch´eele, Alexander Nystr¨om, Madeleine Durbeej, Jan F. Talts, Marja Ekblom, and Peter Ekblom. Laminin isoforms in development and disease. J Mol Med, 85(8):825–836, 2007. [58] Monique Aumailley. The laminin family. Cell Adh Migr, 7(1):48–55, Jan 2013. 2012CAM0080R[PII]. [59] Celeste M Nelson and Mina J Bissell. Modeling dynamic reciprocity: engi- neering three-dimensional culture models of breast architecture, function, and neoplastic transformation. Semin. Cancer Biol., 15(5):342–352, October 2005. [60] V M Weaver, O W Petersen, F Wang, C A Larabell, P Briand, C Damsky, and M J Bissell. Reversion of the malignant phenotype of human breast cells in three-dimensional culture and in vivo by integrin blocking antibodies. J Cell Biol, 137(1):231–245, April 1997. [61] Valerie M Weaver, Sophie Leli`evre, Johnathon N Lakins, Micah A Chrenek, Jonathan C R Jones, Filippo Giancotti, Zena Werb, and Mina J Bissell. beta4 integrin-dependent formation of polarized three-dimensional architecture con- fers resistance to apoptosis in normal and malignant mammary epithelium. Cancer Cell, 2(3):205–216, September 2002. [62] A Albini, Y Iwamoto, H K Kleinman, G R Martin, S A Aaronson, J M Ko- zlowski, and R N McEwan. A rapid in vitro assay for quantitating the invasive potential of tumor cells. Cancer Res, 47(12):3239–3245, June 1987. [63] S Boyden. The chemotactic effect of mixtures of antibody and antigen on polymorphonuclear leucocytes. J Exp Med, 115:453–466, March 1962. 176 [64] Matthew J Paszek, Nastaran Zahir, Kandice R Johnson, Johnathon N Lakins, Gabriela I Rozenberg, Amit Gefen, Cynthia A Reinhart-King, Susan S Mar- gulies, Micah Dembo, David Boettiger, Daniel A Hammer, and Valerie M Weaver. Tensional homeostasis and the malignant phenotype. Cancer Cell, 8(3):241–254, September 2005. [65] Asja Guzman, Michelle J Ziperstein, and Laura J Kaufman. The effect of fibrillar matrix architecture on tumor cell invasion of physically challenging environments. Biomaterials, 35(25):6954–6963, August 2014. [66] R J Pelham and Y l Wang. Cell locomotion and focal adhesions are regulated by substrate flexibility. Proc. Natl. Acad. Sci. U.S.A., 94(25):13661–13665, December 1997. [67] Casey E Kandow, Penelope C Georges, Paul A Janmey, and Karen A Beningo. Polyacrylamide hydrogels for cell mechanics: steps toward optimization and alternative uses. Methods Cell Biol., 83:29–46, 2007. [68] Justin R Tse and Adam J Engler. Preparation of hydrogel substrates with tunable mechanical properties. Curr Protoc Cell Biol, Chapter 10:Unit 10.16, June 2010. [69] Robert W Style, Rostislav Boltyanskiy, Guy K German, Callen Hyland, Christo- pher W MacMinn, Aaron F Mertz, Larry A Wilen, Ye Xu, and Eric R Dufresne. Traction force microscopy in physics and biology. Soft Matter, 10(23):4047– 4055, June 2014. [70] A K Harris, P Wild, and D Stopak. Silicone rubber substrata: a new wrinkle in the study of cell locomotion. Science, 208(4440):177–179, April 1980. [71] Albert Folch. Introduction to BioMEMS. CRC Press, 2016. [72] R S Kane, S Takayama, E Ostuni, D E Ingber, and G M Whitesides. Patterning proteins and cells using soft lithography. Biomaterials, 20(23-24):2363–2376, December 1999. [73] Ning Wang, Emanuele Ostuni, George M Whitesides, and Donald E Ingber. Mi- cropatterning tractional forces in living cells. Cell Motil. Cytoskeleton, 52(2):97– 106, June 2002. [74] Samuel K Sia and George M Whitesides. Microfluidic devices fabricated in poly(dimethylsiloxane) for biological studies. Electrophoresis, 24(21):3563–3576, November 2003. [75] Colin D Paul, Wei-Chien Hung, Denis Wirtz, and Konstantinos Konstantopou- los. Engineered Models of Confined Cell Migration. Annu. Rev. Biomed. Eng., 18:159–180, July 2016. 177 [76] Jessamine Ng Lee, Xingyu Jiang, Declan Ryan, and George M Whitesides. Com- patibility of mammalian cells on surfaces of poly(dimethylsiloxane). Langmuir, 20(26):11684–11691, December 2004. [77] Erwin Berthier, Edmond W K Young, and David Beebe. Engineers are from PDMS-land, Biologists are from Polystyrenia. Lab on a Chip, 12(7):1224–1237, April 2012. [78] Xin Q Brown, Keiko Ookawa, and Joyce Y Wong. Evaluation of polydimethyl- siloxane scaffolds with physiologically-relevant elastic moduli: interplay of sub- strate mechanics and surface chemistry effects on vascular smooth muscle cell response. Biomaterials, 26(16):3123–3129, June 2005. [79] Mirjam Ochsner, Marc R Dusseiller, H Michelle Grandin, Sheila Luna-Morris, Marcus Textor, Viola Vogel, and Michael L Smith. Micro-well arrays for 3D shape control and high resolution analysis of single cells. Lab Chip, 7(8):1074– 1077, August 2007. [80] Rachelle N Palchesko, Ling Zhang, Yan Sun, and Adam W Feinberg. Develop- ment of polydimethylsiloxane substrates with tunable elastic modulus to study cell mechanobiology in muscle and nerve. PLOS ONE, 7(12):e51499, 2012. [81] William J Ashby and Andries Zijlstra. Established and novel methods of inter- rogating two-dimensional cell migration. Integr Biol (Camb), 4(11):1338–1350, November 2012. [82] Roberto Mayor and Sandrine Etienne-Manneville. The front and rear of collec- tive cell migration. Nat Rev Mol Cell Biol, 17(2):97–109, February 2016. [83] C Gilles, M Polette, J M Zahm, J M Tournier, L Volders, J M Foidart, and P Birembaut. Vimentin contributes to human mammary epithelial cell migra- tion. J Cell Sci, 112 ( Pt 24):4615–4625, December 1999. [84] Celeste M Nelson, Martijn M Vanduijn, Jamie L Inman, Daniel A Fletcher, and Mina J Bissell. Tissue geometry determines sites of mammary branching morphogenesis in organotypic cultures. Science, 314(5797):298–300, October 2006. [85] Esther W Gomez, Qike K Chen, Nikolce Gjorevski, and Celeste M Nelson. Tissue geometry patterns epithelial-mesenchymal transition via intercellular mechanotransduction. J. Cell. Biochem., 110(1):44–51, May 2010. [86] Eline Boghaert, Jason P Gleghorn, KangAe Lee, Nikolce Gjorevski, Derek C Radisky, and Celeste M Nelson. Host epithelial geometry regulates breast cancer cell invasiveness. Proc Natl Acad Sci USA, 109(48):19632–19637, November 2012. 178 [87] Hannah G Yevick, Guillaume Duclos, Isabelle Bonnet, and Pascal Silberzan. Architecture and migration of an epithelium on a cylindrical wire. Proc Natl Acad Sci USA, 112(19):5944–5949, May 2015. [88] Junmin Lee, Amr A Abdeen, Kathryn L Wycislo, Timothy M Fan, and Kristo- pher A Kilian. Interfacial geometry dictates cancer cell tumorigenicity. Nature Mat, 15(8):856–862, August 2016. [89] Ian Y Wong, Sarah Javaid, Elisabeth A Wong, Sinem Perk, Daniel A Haber, Mehmet Toner, and Daniel Irimia. Collective and individual migration fol- lowing the epithelial-mesenchymal transition. Nature Mat, 13(11):1063–1071, November 2014. [90] P Clark, P Connolly, A S Curtis, J A Dow, and C D Wilkinson. Topograph- ical control of cell behaviour: II. Multiple grooved substrata. Development, 108(4):635–644, April 1990. [91] Daniel Irimia and Mehmet Toner. Spontaneous migration of cancer cells un- der conditions of mechanical confinement. Integr Biol (Camb), 1(8-9):506–512, September 2009. [92] Andrea Ravasio, Ibrahim Cheddadi, Tianchi Chen, Telmo Pereira, Hui Ting Ong, Cristina Bertocchi, Agusti Brugues, Antonio Jacinto, Alexandre J Kabla, Yusuke Toyama, Xavier Trepat, Nir Gov, Lu´ıs Neves de Almeida, and Benoit Ladoux. Gap geometry dictates epithelial closure efficiency. Nat Commun, 6:7683, July 2015. [93] Stephen H. Davis. Theory of Solidification. Cambridge University Press, 2001. [94] Madeline A Lancaster and Juergen A Knoblich. Organogenesis in a dish: modeling development and disease using organoid technologies. Science, 345(6194):1247125, July 2014. [95] Malcolm S Steinberg. Differential adhesion in morphogenesis: a modern view. Curr Opin Genet Dev, 17(4):281–286, August 2007. [96] Franziska Hirschhaeuser, Heike Menne, Claudia Dittfeld, Jonathan West, Wolfgang Mueller-Klieser, and Leoni A Kunz-Schughart. Multicellular tu- mor spheroids: an underestimated tool is catching up again. J. Biotechnol., 148(1):3–15, July 2010. [97] R M Sutherland, J A McCredie, and W R Inch. Growth of multicell spheroids in tissue culture as a model of nodular carcinomas. J. Natl. Cancer Inst., 46(1):113–120, January 1971. [98] R M McAllister, G Reed, and R J Huebner. Colonial growth in agar of cells de- rived from adenovirus-induced hamster tumors. J. Natl. Cancer Inst., 39(1):43– 53, July 1967. 179 [99] Jens M Kelm, Nicholas E Timmins, Catherine J Brown, Martin Fussenegger, and Lars K Nielsen. Method for generation of homogeneous multicellular tu- mor spheroids applicable to a wide variety of cell types. Biotechnol Bioeng, 83(2):173–180, July 2003. [100] R M Sutherland. Cell and environment interactions in tumor microregions: the multicell spheroid model. Science, 240(4849):177–184, April 1988. [101] M S Steinberg. Reconstruction of tissues by dissociated cells. Some morpho- genetic tissue movements and the sorting out of embryonic cells may have a common explanation. Science, 141(3579):401–408, August 1963. [102] M S Steinberg and M Takeichi. Experimental specification of cell sorting, tissue spreading, and specific spatial patterning by quantitative differences in cadherin expression. Proc Natl Acad Sci USA, 91(1):206–209, January 1994. [103] Steve Pawlizak, Anatol W Fritsch, Steffen Grosser, Dave Ahrens, Tobias Thal- heim, Stefanie Riedel, Tobias R Kießling, Linda Oswald, Mareike Zink, M Lisa Manning, and Josef A K¨as. Testing the differential adhesion hypothesis across the epithelial mesenchymal transition. New J Phys, 17(8):1–17, August 2015. [104] Shawn P Carey, Alina Starchenko, Alexandra L McGregor, and Cynthia A Reinhart-King. Leading malignant cells initiate collective epithelial cell invasion in a three-dimensional heterotypic tumor spheroid model. Clin Exp Metastasis, 30(5):615–630, January 2013. [105] Alec E Cerchiari, James C Garbe, Noel Y Jee, Michael E Todhunter, Kyle E Broaders, Donna M Peehl, Tejal A Desai, Mark A LaBarge, Matthew Thomson, and Zev J Gartner. A strategy for tissue self-organization that is robust to cellular heterogeneity and plasticity. Proc Natl Acad Sci USA, 112(7):2287– 2292, February 2015. [106] St´ephane Douezan, Julien Dumond, and Fran¸coise Brochard-Wyart. Wetting transitions of cellular aggregates induced by substrate rigidity. Soft Matter, 8(17):4578–4583, 2012. [107] P. G. de Gennes. Wetting: statics and dynamics. Rev. Mod. Phys., 57:827–863, Jul 1985. [108] Benjamin L Bangasser, Steven S Rosenfeld, and David J Odde. Determinants of maximal force transmission in a motor-clutch model of cell traction in a compliant microenvironment. Biophys J, 105(3):581–592, August 2013. [109] St´ephane Douezan, Karine Guevorkian, Randa Naouar, Sylvie Dufour, Damien Cuvelier, and Fran¸coise Brochard-Wyart. Spreading dynamics and wetting tran- sition of cellular aggregates. Proc Natl Acad Sci USA, 108(18):7315–7320, May 2011. 180 [110] Jin-Ah Park, Lior Atia, Jennifer A Mitchel, Jeffrey J Fredberg, and James P Butler. Collective migration and cell jamming in asthma, cancer and develop- ment. J Cell Sci, 129(18):3375–3383, September 2016. [111] David Gonzalez-Rodriguez, Karine Guevorkian, St´ephane Douezan, and Fran¸coise Brochard-Wyart. Soft matter models of developing tissues and tu- mors. Science, 338(6109):910–917, November 2012. [112] Quanming Shi, Rajarshi P Ghosh, Hanna Engelke, Chris H Rycroft, Luke Cassereau, James A Sethian, Valerie M Weaver, and Jan T Liphardt. Rapid disorganization of mechanically interacting systems of mammary acini. Proc Natl Acad Sci USA, 111(2):658–663, January 2014. [113] Chin-Lin Guo, Mingxing Ouyang, Jiun-Yann Yu, Jordan Maslov, Andrew Price, and Chih-Yu Shen. Long-range mechanical force enables self-assembly of ep- ithelial tubular patterns. Proc Natl Acad Sci USA, 109(15):5576–5582, April 2012. [114] Hailong Wang, A S Abhilash, Christopher S Chen, Rebecca G Wells, and Vivek B Shenoy. Long-range force transmission in fibrous matrices enabled by tension-driven alignment of fibers. Biophys J, 107(11):2592–2603, December 2014. [115] Kim-Vy Nguyen-Ngoc, Kevin J Cheung, Audrey Brenot, Eliah R Shamir, Ryan S Gray, William C Hines, Paul Yaswen, Zena Werb, and Andrew J Ewald. ECM microenvironment regulates collective migration and local dis- semination in normal and malignant mammary epithelium. Proc Natl Acad Sci USA, 109(39):E2595–604, September 2012. [116] K V Nguyen-Ngoc and A J Ewald. Mammary ductal elongation and myoep- ithelial migration are regulated by the composition of the extracellular matrix. Journal of Microscopy, 251(3):212–223, September 2013. [117] Asja Guzman, V´ıctor S´anchez Alemany, Yen Nguyen, Catherine Ruiqi Zhang, and Laura J Kaufman. A novel 3D in vitro metastasis model elucidates dif- ferential invasive strategies during and after breaching basement membrane. Biomaterials, 115:19–29, January 2017. [118] Hossein Ahmadzadeh, Marie R Webster, Reeti Behera, Angela M Jimenez Va- lencia, Denis Wirtz, Ashani T Weeraratna, and Vivek B Shenoy. Modeling the two-way feedback between contractility and matrix realignment reveals a non- linear mode of cancer cell invasion. Proc Natl Acad Sci USA, page 201617037, February 2017. [119] Shawn P Carey, Karen E Martin, and Cynthia A Reinhart-King. Three- dimensional collagen matrix induces a mechanosensitive invasive epithelial phe- notype. Sci Rep, 7:42088, February 2017. 181 [120] Bartley J Gill, Don L Gibbons, Laila C Roudsari, Jennifer E Saik, Zain H Rizvi, Jonathon D Roybal, Jonathan M Kurie, and Jennifer L West. A Synthetic Matrix with Independently Tunable Biochemistry and Mechanical Properties to Study Epithelial Morphogenesis and EMT in a Lung Adenocarcinoma Model. Cancer Res, 72(22):6013–6023, November 2012. [121] Chien-Chi Lin and Kristi S Anseth. PEG hydrogels for the controlled release of biomolecules in regenerative medicine. Pharm Res, 26(3):631–643, March 2009. [122] Nduka O Enemchukwu, Ricardo Cruz-Acu˜ na, Tom Bongiorno, Christopher T Johnson, Jos´e R Garc´ıa, Todd Sulchek, and Andr´es J Garc´ıa. Synthetic matrices reveal contributions of ECM biophysical and biochemical properties to epithelial morphogenesis. J Cell Biol, 212(1):113–124, January 2016. [123] Spencer C Wei, Laurent Fattet, Jeff H Tsai, Yurong Guo, Vincent H Pai, Han- nah E Majeski, Albert C Chen, Robert L Sah, Susan S Taylor, Adam J Engler, and Jing Yang. Matrix stiffness drives epithelial-mesenchymal transition and tumour metastasis through a TWIST1-G3BP2 mechanotransduction pathway. Nat Cell Biol, 17(5):678–688, May 2015. [124] Ovijit Chaudhuri, Sandeep T Koshy, Cristiana Branco da Cunha, Jae-Won Shin, Catia S Verbeke, Kimberly H Allison, and David J Mooney. Extracellular matrix stiffness and composition jointly regulate the induction of malignant phenotypes in mammary epithelium. Nature Mat, 13(10):970–978, October 2014. [125] Alexander D Augst, Hyun Joon Kong, and David J Mooney. Alginate hydrogels as biomaterials. Macromol. Biosci., 6(8):623–633, August 2006. [126] Keila B Fonseca, S´ılvia J Bidarra, Maria J Oliveira, Pedro L Granja, and Cristina C Barrias. Molecularly designed alginate hydrogels susceptible to local proteolysis as three-dimensional cellular microenvironments. Acta Biomater, 7(4):1674–1682, April 2011. [127] L Rønnov-Jessen, O W Petersen, V E Koteliansky, and M J Bissell. The origin of the myofibroblasts in breast cancer. Recapitulation of tumor environment in culture unravels diversity and implicates converted fibroblasts and recruited smooth muscle cells. J. Clin. Invest., 95(2):859–873, February 1995. [128] Anna Labernadie, Takuya Kato, Agust´ı Brugu´es, Xavier Serra-Picamal, Ste- fanie Derzsi, Esther Arwert, Anne Weston, Victor Gonz´alez-Tarrag´o, Alberto Elosegui-Artola, Lorenzo Albertazzi, Jordi Alcaraz, Pere Roca-Cusachs, Erik Sahai, and Xavier Trepat. A mechanically active heterotypic E-cadherin/N- cadherin adhesion enables fibroblasts to drive cancer cell invasion. Nat Cell Biol, 19(3):224–237, March 2017. 182 [129] Kyung Eun Sung, Ning Yang, Carolyn Pehlke, Patricia J Keely, Kevin W Eli- ceiri, Andreas Friedl, and David J Beebe. Transition to invasion in breast can- cer: a microfluidic in vitro model enables examination of spatial and temporal effects. Integr Biol (Camb), 3(4):439–450, April 2011. [130] Seema M Ehsan, Katrina M Welch-Reardon, Marian L Waterman, Christopher C W Hughes, and Steven C George. A three-dimensional in vitro model of tumor cell intravasation. Integr Biol (Camb), 6(6):603–610, June 2014. [131] Amir R Aref, Ruby Yun-Ju Huang, Weimiao Yu, Kian-Ngiap Chua, Wei Sun, Ting-Yuan Tu, Jing Bai, Wen-Jing Sim, Ioannis K Zervantonakis, Jean Paul Thiery, and Roger D Kamm. Screening therapeutic EMT blocking agents in a three-dimensional microenvironment. Integrative Biology, 5(2):381–389, Febru- ary 2013. [132] Kyung Eun Sung, Xiaojing Su, Erwin Berthier, Carolyn Pehlke, Andreas Friedl, and David J Beebe. Understanding the impact of 2D and 3D fibroblast cultures on in vitro breast cancer models. PLoS ONE, 8(10):e76373, 2013. [133] S P Singh, M P Schwartz, E Y Tokuda, Y Luo, R E Rogers, M Fujita, N G Ahn, and K S Anseth. A synthetic modular approach for modeling the role of the 3D microenvironment in tumor progression. Sci Rep, 5:17814, December 2015. [134] John W Weisel and Rustem I Litvinov. Mechanisms of fibrin polymerization and clinical implications. Blood, 121(10):1712–1719, March 2013. [135] Seok Chung, Ryo Sudo, Peter J Mack, Chen-Rei Wan, Vernella Vickerman, and Roger D Kamm. Cell migration into scaffolds under co-culture conditions in a microfluidic platform. Lab Chip, 9(2):269–275, January 2009. [136] Linda G Griffith and Melody A Swartz. Capturing complex 3D tissue physiology in vitro. Nat Rev Mol Cell Biol, 7(3):211–224, March 2006. [137] Sangeeta N Bhatia and Donald E Ingber. Microfluidic organs-on-chips. Nat Biotechnol, 32(8):760–772, August 2014. [138] Hasan Erbil Abaci and Michael L Shuler. Human-on-a-chip design strategies and principles for physiologically based pharmacokinetics/pharmacodynamics modeling. Integr Biol (Camb), 7(4):383–391, April 2015. [139] M L Oyen. Mechanical characterisation of hydrogel materials. Int Mater Rev, 59(1):44–59, December 2013. [140] Anne S van Oosten, Peter A Galie, and Paul A Janmey. Mechanical properties of hydrogels. In Gels Handbook: Fundamentals, Properties and Applications, chapter 4, pages 67–79. World Scientific, 2016. 183 [141] C P Broedersz and F C MacKintosh. Modeling semiflexible polymer networks. Rev Mod Phys, 86(3):995–1036, July 2014. [142] Cornelis Storm, Jennifer J Pastore, F C MacKintosh, T C Lubensky, and Paul A Janmey. Nonlinear elasticity in biological gels. Nature, 435(7039):191–194, May 2005. [143] Ovijit Chaudhuri, Luo Gu, Darinka Klumpers, Max Darnell, Sidi A Bencherif, James C Weaver, Nathaniel Huebsch, Hong-pyo Lee, Evi Lippens, Georg N Duda, and David J Mooney. Hydrogels with tunable stress relaxation regulate stem cell fate and activity. Nature Mat, 15(3):326–334, March 2016. [144] Andrew M Stein, David A Vader, Louise M Jawerth, David A Weitz, and Leonard M Sander. An algorithm for extracting the network geometry of three- dimensional collagen gels. J Microsc, 232(3):463–475, December 2008. [145] Walter Mickel, Stefan M¨ unster, Louise M Jawerth, David A Vader, David A Weitz, Adrian P Sheppard, Klaus Mecke, Ben Fabry, and Gerd E Schr¨oder- Turk. Robust pore size analysis of filamentous networks from three-dimensional confocal microscopy. Biophys J, 95(12):6072–6080, December 2008. [146] Jennifer C Ashworth, Marco Mehr, Paul G Buxton, Serena M Best, and Ruth E Cameron. Cell Invasion in Collagen Scaffold Architectures Characterized by Percolation Theory. Adv Healthc Mater, 4(9):1317–1321, June 2015. [147] Julian Steinwachs, Claus Metzner, Kai Skodzek, Nadine Lang, Ingo Thievessen, Christoph Mark, Stefan M¨ unster, Katerina E Aifantis, and Ben Fabry. Three- dimensional force microscopy of cells in biopolymer networks. Nat Meth, pages 1–9, December 2015. [148] David A Stout, Eyal Bar-Kochba, Jonathan B Estrada, Jennet Toyjanova, Ha- neesh Kesari, Jonathan S Reichner, and Christian Franck. Mean deformation metrics for quantifying 3D cell-matrix interactions without requiring informa- tion about matrix material properties. Proc Natl Acad Sci USA, 113(11):2898– 2903, March 2016. [149] Matthew S Hall, Farid Alisafaei, Ehsan Ban, Xinzeng Feng, Chung-Yuen Hui, Vivek B Shenoy, and Mingming Wu. Fibrous nonlinear elasticity enables pos- itive mechanical feedback between cells and ECMs. Proc Natl Acad Sci USA, 113(49):14043–14048, December 2016. [150] Gr´egory Beaune, Tomita Vasilica Stirbat, Nada Khalifat, Olivier Cochet- Escartin, Simon Garcia, Vasily Val´er¨ı´evitch Gurchenkov, Michael P Murrell, Sylvie Dufour, Damien Cuvelier, and Fran¸coise Brochard-Wyart. How cells flow in the spreading of cellular aggregates. Proc Natl Acad Sci USA, 111(22):8055– 8060, May 2014. 184 [151] Ben T Grys, Dara S Lo, Nil Sahin, Oren Z Kraus, Quaid Morris, Charles Boone, and Brenda J Andrews. Machine learning and computer vision approaches for phenotypic profiling. J. Cell Biol., 216(1):65–71, January 2017. [152] Susan E Leggett, Jea Yun Sim, Jonathan E Rubins, Zachary J Neronha, Eve- lyn Kendall Williams, and Ian Y Wong. Morphological single cell profiling of the epithelial-mesenchymal transition. Integr Biol (Camb), 8(11):1133–1144, November 2016. [153] Zhongying Wang, Daniel Tonderys, Susan E Leggett, Evelyn Kendall Williams, Mehrdad T Kiani, Ruben Spitz Steinberg, Yang Qiu, Ian Y Wong, and Robert H Hurt. Wrinkled, wavelength-tunable graphene-based surface topographies for directing cell alignment and morphology. Carbon, 97:14–24, February 2016. [154] Peng Qiu, Erin F Simonds, Sean C Bendall, Kenneth D Gibbs, Robert V Brug- gner, Michael D Linderman, Karen Sachs, Garry P Nolan, and Sylvia K Plevri- tis. Extracting a cellular hierarchy from high-dimensional cytometry data with SPADE. Nat Biotechnol, 29(10):886–891, October 2011. [155] Stavroula Skylaki, Oliver Hilsenbeck, and Timm Schroeder. Challenges in long-term imaging and quantification of single-cell dynamics. Nat Biotechnol, 34(11):1137–1144, November 2016. [156] Marielena Gamboa Castro, Susan E Leggett, and Ian Y Wong. Clustering and jamming in epithelial-mesenchymal co-cultures. Soft Matter, 12(40):8327–8337, October 2016. [157] Francesco Pampaloni, Emmanuel G Reynaud, and Ernst H K Stelzer. The third dimension bridges the gap between cell culture and live tissue. Nat Rev Mol Cell Biol, 8(10):839–845, September 2007. [158] Meghan K Driscoll and Gaudenz Danuser. Quantifying Modes of 3D Cell Mi- gration. Trends Cell Biol, 25(12):749–759, December 2015. [159] Justin D Mih, Asma S Sharif, Fei Liu, Aleksandar Marinkovi´c, Matthew M Symer, and Daniel J Tschumperlin. A multiwell platform for studying stiffness- dependent cell biology. PLoS ONE, 6(5):e19929, 2011. [160] Marc van de Wetering, Hayley E Francies, Joshua M Francis, Gergana Bounova, Francesco Iorio, Apollo Pronk, Winan van Houdt, Joost van Gorp, Amaro Taylor-Weiner, Lennart Kester, Anne McLaren-Douglas, Joyce Blokker, Sridevi Jaksani, Sina Bartfeld, Richard Volckman, Peter van Sluis, Vivian S W Li, Sara Seepo, Chandra Sekhar Pedamallu, Kristian Cibulskis, Scott L Carter, Aaron McKenna, Michael S Lawrence, Lee Lichtenstein, Chip Stewart, Jan Koster, Rogier Versteeg, Alexander van Oudenaarden, Julio Saez-Rodriguez, Robert G J Vries, Gad Getz, Lodewyk Wessels, Michael R Stratton, Ultan McDer- mott, Matthew Meyerson, Mathew J Garnett, and Hans Clevers. Prospec- tive derivation of a living organoid biobank of colorectal cancer patients. Cell, 161(4):933–945, May 2015. 185 [161] Adrian Ranga, Mehmet Girgin, Andrea Meinhardt, Dominic Eberle, Massi- miliano Caiazzo, Elly M Tanaka, and Matthias P Lutolf. Neural tube mor- phogenesis in synthetic 3D microenvironments. Proc Natl Acad Sci USA, 113(44):E6831–E6839, November 2016. [162] Massimiliano Caiazzo, Yuya Okawa, Adrian Ranga, Alessandra Piersigilli, Yoji Tabata, and Matthias P Lutolf. Defined three-dimensional microenvironments boost induction of pluripotency. Nature Mat, 15(3):344–352, March 2016. [163] Kari R Fischer, Anna Durrans, Sharrell Lee, Jianting Sheng, Fuhai Li, Stephen T C Wong, Hyejin Choi, Tina El Rayes, Seongho Ryu, Juliane Troeger, Robert F Schwabe, Linda T Vahdat, Nasser K Altorki, Vivek Mittal, and Dingcheng Gao. Epithelial-to-mesenchymal transition is not required for lung metastasis but contributes to chemoresistance. Nature, 527(7579):472–476, November 2015. [164] Xiaofeng Zheng, Julienne L Carstens, Jiha Kim, Matthew Scheible, Judith Kaye, Hikaru Sugimoto, Chia-Chin Wu, Valerie S LeBleu, and Raghu Kalluri. Epithelial-to-mesenchymal transition is dispensable for metastasis but induces chemoresistance in pancreatic cancer. Nature, 527(7579):525–530, November 2015. [165] Min Yu, Aditya Bardia, Ben S Wittner, Shannon L Stott, Malgorzata E Smas, David T Ting, Steven J Isakoff, Jordan C Ciciliano, Marissa N Wells, Ajay M Shah, Kyle F Concannon, Maria C Donaldson, Lecia V Sequist, Elena Brachtel, Dennis Sgroi, Jose Baselga, Sridhar Ramaswamy, Mehmet Toner, Daniel A Haber, and Shyamala Maheswaran. Circulating Breast Tumor Cells Exhibit Dynamic Changes in Epithelial and Mesenchymal Composition. Sci- ence, 339(6119):580–584, February 2013. [166] Eliah R Shamir and Andrew J Ewald. Three-dimensional organotypic culture: experimental models of mammalian biology and disease. Nat Rev Mol Cell Biol, 15(10):647–664, October 2014. [167] Marina Simian and Mina J Bissell. Organoids: A historical perspective of thinking in three dimensions. J Cell Biol, 216(1):31–40, January 2017. [168] Nikolce Gjorevski, Norman Sachs, Andrea Manfrin, Sonja Giger, Maiia E Brag- ina, Paloma Ord´on ˜ez-Mor´an, Hans Clevers, and Matthias P Lutolf. Designer matrices for intestinal stem cell and organoid culture. Nature, 539(7630):560– 564, November 2016. [169] Younan Xia, John A Rogers, Kateri E Paul, and George M Whitesides. Uncon- ventional Methods for Fabricating and Patterning Nanostructures. Chem Rev, 99(7):1823–1848, July 1999. 186 [170] Bo Li, Yan-Ping Cao, Xi-Qiao Feng, and Huajian Gao. Mechanics of mor- phological instabilities and surface wrinkling in soft materials: a review. Soft Matter, 8(21):5728–5745, 2012. [171] Jan Genzer and Jan Groenewold. Soft matter with hard skin: From skin wrin- kles to templating and material characterization. Soft Matter, 2(4):310–323, 2006. [172] Srikanth Singamaneni and Vladimir V Tsukruk. Buckling instabilities in peri- odic composite polymeric materials. Soft Matter, 6(22):5681–5692, 2010. [173] Shu Yang, Krishnacharya Khare, and Pei-Chun Lin. Harnessing Surface Wrinkle Patterns in Soft Matter. Adv. Funct. Mater., 20(16):2550–2564, July 2010. [174] N Bowden, S Brittain, A G Evans, and J W Hutchinson. Spontaneous forma- tion of ordered structures in thin films of metals supported on an elastomeric polymer. Nature, 1998. [175] M A Biot. Folding instability of a layered viscoelastic medium under compres- sion. In P Roy Soc A - Math Phy, 1957. [176] J Y Sun, S Xia, M W Moon, K H Oh, and K S Kim. Folding wrinkles of a thin stiff layer on a soft substrate. P Roy Soc A - Math Phy, 468(2140):932–953, February 2012. [177] M Diab and K S Kim. Ruga-formation instabilities of a graded stiffness bound- ary layer in a neo-Hookean solid. P Roy Soc A - Math Phy, 470(2168):20140218– 20140218, May 2014. [178] Keun Soo Kim, Yue Zhao, Houk Jang, Sang Yoon Lee, Jong Min Kim, Kwang S Kim, Jong-Hyun Ahn, Philip Kim, Jae-Young Choi, and Byung Hee Hong. Large-scale pattern growth of graphene films for stretchable transparent elec- trodes. Nature, 457(7230):706–710, February 2009. [179] Wenzhong Bao, Feng Miao, Zhen Chen, Hang Zhang, Wanyoung Jang, Chris Dames, and Chun Ning Lau. Controlled ripple texturing of suspended graphene and ultrathin graphite membranes. Nat Nanotechnol, 4(9):562–566, September 2009. [180] Qiyuan He, Herry Gunadi Sudibya, Zongyou Yin, Shixin Wu, Hai Li, Freddy Boey, Wei Huang, Peng Chen, and Hua Zhang. Centimeter-long and large- scale micropatterns of reduced graphene oxide films: fabrication and sensing applications. ACS Nano, 4(6):3201–3208, June 2010. [181] Scott Scharfenberg, D Z Rocklin, Cesar Chialvo, Richard L Weaver, Paul M Goldbart, and Nadya Mason. Probing the mechanical properties of graphene using a corrugated elastic substrate. Appl. Phys. Lett., 98(9):091908–3, 2011. 187 [182] Lotta E Delle, Ruben Lanche, Jessica Ka-Yan Law, Maryam Weil, Xuan Thang Vu, Patrick Wagner, and Sven Ingebrandt. Reduced graphene oxide micropat- terns as an interface for adherent cells. Phys. Status Solidi A, 210(5):975–982, April 2013. [183] Daniel A Kunz, Patrick Feicht, Sebastian G¨odrich, Herbert Thurn, Georg Pa- pastavrou, Andreas Fery, and Josef Breu. Space-Resolved In-Plane Moduli of Graphene Oxide and Chemically Derived Graphene Applying a Simple Wrin- kling Procedure. Adv. Mater., 25(9):1337–1341, December 2012. [184] Jianfeng Zang, Seunghwa Ryu, Nicola Pugno, Qiming Wang, Qing Tu, Markus J Buehler, and Xuanhe Zhao. Multifunctionality and control of the crumpling and unfolding of large-area graphene. Nat Mater, 12(4):321–325, April 2013. [185] Jianfeng Zang, Changyong Cao, Yaying Feng, Jie Liu, and Xuanhe Zhao. Stretchable and High-Performance Supercapacitors with Crumpled Graphene Papers. Sci Rep, 4:6492, October 2014. [186] Ping Xu, Junmo Kang, Jae-Boong Choi, Jonghwan Suhr, Jianyong Yu, Faxue Li, Joon-Hyung Byun, Byung-Sun Kim, and Tsu-Wei Chou. Laminated Ultra- thin Chemical Vapor Deposition Graphene Films Based Stretchable and Trans- parent High-Rate Supercapacitor. ACS nano, 8(9):9437–9445, September 2014. [187] Sk Faruque Ahmed, So Nagashima, Ji Yeong Lee, Kwang-Ryeol Lee, Kyung-Suk Kim, and Myoung-Woon Moon. Self-assembled folding of a biaxially compressed film on a compliant substrate. Carbon, 76(C):105–112, September 2014. [188] Hong Ying Mao, Sophie Laurent, Wei Chen, Omid Akhavan, Mohammad Imani, Ali Akbar Ashkarran, and Morteza Mahmoudi. Graphene: Promises, Facts, Opportunities, and Challenges in Nanomedicine. Chem Rev, 113(5):3407–3424, May 2013. [189] Sumit Goenka, Vinayak Sant, and Shilpa Sant. Graphene-based nanomaterials for drug delivery and tissue engineering. J Control Release, 173:75–88, January 2014. [190] Dimitrios Bitounis, Hanene Ali-Boucetta, Byung Hee Hong, Dal-Hee Min, and Kostas Kostarelos. Prospects and Challenges of Graphene in Biomedical Ap- plications. Adv. Mater., 25(16):2258–2268, March 2013. [191] Wong Cheng Lee, Candy Haley Y X Lim, Hui Shi, Lena A L Tang, Yu Wang, Chwee Teck Lim, and Kian Ping Loh. Origin of enhanced stem cell growth and differentiation on graphene and graphene oxide. ACS Nano, 5(9):7334–7341, September 2011. [192] G Y Chen, D W P Pang, S M Hwang, H Y Tuan, and Y C Hu. A graphene- based platform for induced pluripotent stem cells culture and differentiation. Biomaterials, 33(2):418–427, January 2012. 188 [193] Sook Hee Ku and Chan Beum Park. Myoblast differentiation on graphene oxide. Biomaterials, 34(8):2017–2023, March 2013. [194] Tapas R Nayak, Henrik Andersen, Venkata S Makam, Clement Khaw, Sukang Bae, Xiangfan Xu, Pui-Lai R Ee, Jong-Hyun Ahn, Byung Hee Hong, Gior- ¨ gia Pastorin, and Barbaros Ozyilmaz. Graphene for controlled and acceler- ated osteogenic differentiation of human mesenchymal stem cells. ACS Nano, 5(6):4670–4678, June 2011. [195] Soo-Ryoon Ryoo, Young-Kwan Kim, Mi-Hee Kim, and Dal-Hee Min. Behaviors of NIH-3T3 fibroblasts on graphene/carbon nanotubes: proliferation, focal ad- hesion, and gene transfection studies. ACS Nano, 4(11):6587–6598, November 2010. [196] Christopher J Bettinger, Robert Langer, and Jeffrey T Borenstein. Engineer- ing substrate topography at the micro- and nanoscale to control cell function. Angew Chem Int Ed Engl, 48(30):5406–5415, 2009. [197] Deok-Ho Kim, Paolo P Provenzano, Chris L Smith, and Andre Levchenko. Matrix nanotopography as a regulator of cell function. J Cell Biol, 197(3):351– 360, April 2012. [198] A Curtis and C Wilkinson. Topographical control of cells. Biomaterials, 18(24):1573–1583, December 1997. [199] P Weiss. Cellular dynamics, volume 31. Rev Mod Phys, 1959. [200] Jesus Isaac Luna, Jesus Ciriza, Marcos E Garcia-Ojeda, Marco Kong, Anthony Herren, Deborah K Lieu, Ronald A Li, Charless C Fowlkes, Michelle Khine, and Kara E McCloskey. Multiscale biomimetic topography for the alignment of neonatal and embryonic stem cell-derived heart cells. Tissue Eng Part C Methods, 17(5):579–588, May 2011. [201] Jiaxian Wang, Aaron Chen, Deborah K Lieu, Ioannis Karakikes, Gaopeng Chen, Wendy Keung, Camie W Chan, Roger J Hajjar, Kevin D Costa, Michelle Khine, and Ronald A Li. Effect of engineered anisotropy on the susceptibility of hu- man pluripotent stem cell-derived ventricular cardiomyocytes to arrhythmias. Biomaterials, 34(35):8878–8886, November 2013. [202] Changyong Cao, Hon Fai Chan, Jianfeng Zang, Kam W Leong, and Xuanhe Zhao. Harnessing Localized Ridges for High-Aspect-Ratio Hierarchical Patterns with Dynamic Tunability and Multifunctionality. Adv. Mater., 26(11):1763– 1770, December 2013. [203] Kristian Kolind, Kam W Leong, Flemming Besenbacher, and Morten Foss. Guidance of stem cell fate on 2D patterned surfaces. Biomaterials, 33(28):6626– 6633, October 2012. 189 [204] Evelyn K F Yim, Eric M Darling, Karina Kulangara, Farshid Guilak, and Kam W Leong. Nanotopography-induced changes in focal adhesions, cytoskele- tal organization, and mechanical properties of human mesenchymal stem cells. Biomaterials, 31(6):1299–1306, February 2010. [205] Matthew J Dalby, Mathis O Riehle, Stephen J Yarwood, Chris D W Wilkinson, and Adam S G Curtis. Nucleus alignment and cell signaling in fibroblasts: response to a micro-grooved topography. Exp Cell Res, 284(2):274–282, April 2003. [206] Ana I Teixeira, George A Abrams, Paul J Bertics, Christopher J Murphy, and Paul F Nealey. Epithelial contact guidance on well-defined micro- and nanos- tructured substrates. J Cell Sci, 116(Pt 10):1881–1892, May 2003. [207] Deok-Ho Kim, Elizabeth A Lipke, Pilnam Kim, Raymond Cheong, Susan Thompson, Michael Delannoy, Kahp-Yang Suh, Leslie Tung, and Andre Levchenko. Nanoscale cues regulate the structure and function of macroscopic cardiac tissue constructs. Proc Natl Acad Sci USA, 107(2):565–570, January 2010. [208] Yang Qiu, Zhongying Wang, Alisa C E Owens, Indrek Kulaots, Yantao Chen, Agnes B Kane, and Robert H Hurt. Antioxidant chemistry of graphene- based materials and its role in oxidation protection technology. Nanoscale, 6(20):11744–11755, October 2014. [209] Lee Kamentsky, Thouis R Jones, Adam Fraser, Mark-Anthony Bray, David J Logan, Katherine L Madden, Vebjorn Ljosa, Curtis Rueden, Kevin W Eliceiri, and Anne E Carpenter. Improved structure, function and compatibility for CellProfiler: modular high-throughput image analysis software. Bioinformatics, 27(8):1179–1180, April 2011. [210] Grace NgaYin Li and Diane Hoffman-Kim. Evaluation of neurite outgrowth anisotropy using a novel application of circular analysis. J Neurosci Meth, 174(2):202–214, September 2008. [211] Jiayan Luo, Hee Dong Jang, Tao Sun, Li Xiao, Zhen He, Alexandros P Kat- soulidis, Mercouri G Kanatzidis, J Murray Gibson, and Jiaxing Huang. Com- pression and aggregation-resistant particles of crumpled soft sheets. ACS nano, 5(11):8943–8949, November 2011. [212] Yantao Chen, Fei Guo, Ashish Jachak, Sang-Pil Kim, Dibakar Datta, Jingyu Liu, Indrek Kulaots, Charles Vaslet, Hee Dong Jang, Jiaxing Huang, Agnes Kane, Vivek B Shenoy, and Robert H Hurt. Aerosol Synthesis of Cargo-Filled Graphene Nanosacks. Nano Lett, 12(4):1996–2002, April 2012. [213] X Chen and John W Hutchinson. Herringbone Buckling Patterns of Compressed Thin Films on Compliant Substrates. J. Appl. Mech., 71(5):597, 2004. 190 [214] Z Y Huang, W Hong, and Z Suo. Nonlinear analyses of wrinkles in a film bonded to a compliant substrate. J Mech Phys Solids, 53(9):2101–2118, September 2005. [215] J Song, H Jiang, Z J Liu, D Y Khang, Y Huang, J A Rogers, C Lu, and C G Koh. Buckling of a stiff thin film on a compliant substrate in large deformation. Int J Solids Struct, 45(10):3107–3121, May 2008. [216] A L Volynskii, S Bazhenov, O V Lebedeva, and N F Bakeev. Mechanical buckling instability of thin coatings deposited on soft polymer substrates. J Mater Sci, 35(3):547–554, 2000. [217] Qing Peng and Suvranu De. Mechanical properties and instabilities of ordered graphene oxide C6O monolayers. RSC Adv., 3(46):24337–8, 2013. [218] Sungjin Park, Kyoung-Seok Lee, Gulay Bozoklu, Weiwei Cai, SonBinh T Nguyen, and Rodney S Ruoff. Graphene oxide papers modified by divalent ions-enhancing mechanical properties via chemical cross-linking. ACS nano, 2(3):572–578, March 2008. [219] Chengmeng Chen, Quan-Hong Yang, Yonggang Yang, Wei Lv, Yuefang Wen, Peng-Xiang Hou, Maozhang Wang, and Hui-Ming Cheng. Self-Assembled Free- Standing Graphite Oxide Membrane. Adv. Mater., 21(29):3007–3011, August 2009. [220] Jin Akagi, Magdalena Kordon, Hong Zhao, Anna Matuszek, Jurek Dobrucki, Rachel Errington, Paul J Smith, Kazuo Takeda, Zbigniew Darzynkiewicz, and Donald Wlodkowic. Real-time cell viability assays using a new anthracycline derivative DRAQ7 .R Cyto Part A, 83(2):227–234, February 2013. [221] Dae-Hyeong Kim, Roozbeh Ghaffari, Nanshu Lu, and John A Rogers. Flexible and stretchable electronics for biointegrated devices. Annu Rev Biomed Eng, 14:113–128, 2012. [222] Ivan R Minev, Pavel Musienko, Arthur Hirsch, Quentin Barraud, Nikolaus Wenger, Eduardo Martin Moraud, J´erˆome Gandar, Marco Capogrosso, Tomis- lav Milekovic, L´eonie Asboth, Rafael Fajardo Torres, Nicolas Vachicouras, Qi- han Liu, Natalia Pavlova, Simone Duis, Alexandre Larmagnac, Janos V¨or¨os, Silvestro Micera, Zhigang Suo, Gr´egoire Courtine, and St´ephanie P Lacour. Biomaterials. Electronic dura mater for long-term multimodal neural interfaces. Science, 347(6218):159–163, January 2015. [223] Jean Paul Thiery, Herv´e Acloque, Ruby Y J Huang, and M Angela Ni- eto. Epithelial-mesenchymal transitions in development and disease. Cell, 139(5):871–890, November 2009. [224] Raghu Kalluri and Robert A Weinberg. The basics of epithelial-mesenchymal transition. J. Clin. Invest., 119(6):1420–1428, June 2009. 191 [225] Gerhard Christofori. New signals from the invasive front. Nature, 441(7092):444–450, May 2006. [226] A Singh and J Settleman. EMT, cancer stem cells and drug resistance: an emerging axis of evil in the war on cancer. Oncogene, 29(34):4741–4751, August 2010. [227] Maria J Blanco, Gema Moreno-Bueno, David Sarri´o, Annamaria Locascio, Am- paro Cano, Jos´e Palacios, and M Angela Nieto. Correlation of Snail expression with histological grade and lymph node status in breast carcinomas. 2002. [228] Joan Massagu´e. TGFβ signalling in context. Nat Rev Mol Cell Biol, 13(10):616– 630, October 2012. [229] Caitriona Holohan, Sandra Van Schaeybroeck, Daniel B Longley, and Patrick G Johnston. Cancer drug resistance: an evolving paradigm. Nat Rev Cancer, 13(10):714–726, October 2013. [230] David M Gonzalez and Damian Medici. Signaling mechanisms of the epithelial- mesenchymal transition. Sci Signal, 7(344):re8–re8, 2014. [231] Steven J Altschuler and Lani F Wu. Cellular heterogeneity: do differences make a difference? Cell, 141(4):559–563, May 2010. [232] Sarah E Kolitz and Douglas A Lauffenburger. Measurement and modeling of signaling at the single-cell level. Biochemistry, 51(38):7433–7443, September 2012. [233] R Yang, M Niepel, T K Mitchison, and P K Sorger. Dissecting variability in responses to cancer chemotherapy through systems pharmacology. Clin. Pharmacol. Ther., 88(1):34–38, July 2010. [234] Fabian Zanella, James B Lorens, and Wolfgang Link. High content screening: seeing is believing. Trends in Biotechnology, 28(5):237–245, May 2010. [235] Gaudenz Danuser. Computer Vision in Cell Biology. Cell, 147(5):973–978, November 2011. [236] Zachary E Perlman, Michael D Slack, Yan Feng, Timothy J Mitchison, Lani F Wu, and Steven J Altschuler. Multidimensional drug profiling by automated microscopy. Science, 306(5699):1194–1198, November 2004. [237] Michael D Slack, Elisabeth D Martinez, Lani F Wu, and Steven J Altschuler. Characterizing heterogeneous cellular responses to perturbations. Proc Natl Acad Sci USA, 105(49):19306–19311, December 2008. [238] A A Cohen, N Geva-Zatorsky, E Eden, M Frenkel-Morgenstern, I Issaeva, A Si- gal, R Milo, C Cohen-Saidon, Y Liron, Z Kam, L Cohen, T Danon, N Perzov, and U Alon. Dynamic proteomics of individual cancer cells in response to a drug. Science, 322(5907):1511–1516, December 2008. 192 [239] Karen E Gascoigne and Stephen S Taylor. Cancer cells display profound intra- and interline variation following prolonged exposure to antimitotic drugs. Can- cer Cell, 14(2):111–122, August 2008. [240] Sabrina L Spencer, Suzanne Gaudet, John G Albeck, John M Burke, and Pe- ter K Sorger. Non-genetic origins of cell-to-cell variability in TRAIL-induced apoptosis. Nature, 459(7245):428–432, May 2009. [241] Thouis R Jones, Anne E Carpenter, Michael R Lamprecht, Jason Moffat, Ser- ena J Silver, Jennifer K Grenier, Adam B Castoreno, Ulrike S Eggert, David E Root, Polina Golland, and David M Sabatini. Scoring diverse cellular mor- phologies in image-based screens with iterative feedback and machine learning. Proc Natl Acad Sci USA, 106(6):1826–1831, February 2009. [242] Pawel Paszek, Sheila Ryan, Louise Ashall, Kate Sillitoe, Claire V Harper, David G Spiller, David A Rand, and Michael R H White. Population robustness arising from cellular heterogeneity. Proc Natl Acad Sci USA, 107(25):11644– 11649, June 2010. [243] Zheng Yin, Amine Sadok, Heba Sailem, Afshan Mccarthy, Xiaofeng Xia, Fuhai Li, Mar Arias Garcia, Louise Evans, Alexis R Barr, Norbert Perrimon, Christo- pher J Marshall, Stephen T C Wong, and Chris Bakal. A screen for morpholog- ical complexity identifies regulators of switch-like transitions between discrete cell shapes. Nat Cell Biol, 15(7):860–871, July 2013. [244] J E Sero, H Z Sailem, R C Ardy, H Almuttaqi, T Zhang, and C Bakal. Cell shape and the microenvironment regulate nuclear translocation of NF- B in breast epithelial and tumor cells. Molecular Systems Biology, 11(3):790–790, March 2015. [245] Manuela Quintavalle, Leonardo Elia, Jeffrey H Price, Susanne Heynen-Genel, and Sara A Courtneidge. A cell-based high-content screening assay reveals activators and inhibitors of cancer cell invasion. Sci Signal, 4(183):ra49–ra49, 2011. [246] Robert J Steininger, Satwik Rajaram, Luc Girard, John D Minna, Lani F Wu, and Steven J Altschuler. On comparing heterogeneity across biomarkers. Cy- tometry A, 87(6):558–567, June 2015. [247] Jonathan Low, Shuguang Huang, Wayne Blosser, Michele Dowless, John Burch, Blake Neubauer, and Louis Stancato. High-content imaging characterization of cell cycle therapeutics through in vitro and in vivo subpopulation analysis. Molecular Cancer Therapeutics, 7(8):2455–2463, August 2008. [248] Albert H Gough, Ning Chen, Tong Ying Shun, Timothy R Lezon, Robert C Boltz, Celeste E Reese, Jacob Wagner, Lawrence A Vernetti, Jennifer R Gran- dis, Adrian V Lee, Andrew M Stern, Mark E Schurdak, and D Lansing Taylor. 193 Identifying and Quantifying Heterogeneity in High Content Analysis: Applica- tion of Heterogeneity Indices to Drug Discovery. PLOS ONE, 9(7):e102678–16, July 2014. [249] Peter D Caie, Rebecca E Walls, Alexandra Ingleston-Orme, Sandeep Daya, Tom Houslay, Rob Eagle, Mark E Roberts, and Neil O Carragher. High-content phenotypic profiling of drug response signatures across distinct cancer cells. Molecular Cancer Therapeutics, 9(6):1913–1926, June 2010. [250] Piyush B Gupta, Christine M Fillmore, Guozhi Jiang, Sagi D Shapira, Kai Tao, Charlotte Kuperwasser, and Eric S Lander. Stochastic state transitions give rise to phenotypic equilibrium in populations of cancer cells. Cell, 146(4):633–644, August 2011. [251] Sreenath V Sharma, Diana Y Lee, Bihua Li, Margaret P Quinlan, Fumiyuki Takahashi, Shyamala Maheswaran, Ultan McDermott, Nancy Azizian, Lee Zou, Michael A Fischbach, Kwok-Kin Wong, Kathleyn Brandstetter, Ben Wittner, Sridhar Ramaswamy, Marie Classon, and Jeff Settleman. A chromatin-mediated reversible drug-tolerant state in cancer cell subpopulations. Cell, 141(1):69–80, April 2010. [252] Marc Hafner, Mario Niepel, Mirra Chung, and Peter K Sorger. Growth rate inhibition metrics correct for confounders in measuring sensitivity to cancer drugs. Nature Methods, 13(6):521–527, June 2016. [253] David Sarri´o, Socorro Mar´ıa Rodriguez-Pinilla, David Hardisson, Amparo Cano, Gema Moreno-Bueno, and Jos´e Palacios. Epithelial-mesenchymal transi- tion in breast cancer relates to the basal-like phenotype. Cancer Res, 68(4):989– 997, February 2008. [254] Magdalena A Cichon, Celeste M Nelson, Derek C Radisky, and D Radisky. Regulation of epithelial-mesenchymal transition in breast cancer cells by cell contact and adhesion. Cancer Inform, 14(Suppl 3):1–13, 2015. [255] Sarah Javaid, Jianmin Zhang, Endre Anderssen, Josh C Black, Ben S Wit- tner, Ken Tajima, David T Ting, Gromoslaw A Smolen, Matthew Zubrowski, Rushil Desai, Shyamala Maheswaran, Sridhar Ramaswamy, Johnathan R Whet- stine, and Daniel A Haber. Dynamic chromatin modification sustains epithelial- mesenchymal transition following inducible expression of Snail-1. Cell Rep, 5(6):1679–1689, December 2013. [256] Binhua P Zhou, Jiong Deng, Weiya Xia, Jihong Xu, Yan M Li, Mehmet Gun- duz, and Mien-Chie Hung. Dual regulation of Snail by GSK-3beta-mediated phosphorylation in control of epithelial-mesenchymal transition. Nat Cell Biol, 6(10):931–940, October 2004. 194 [257] Jayanta Debnath, Senthil K Muthuswamy, and Joan S Brugge. Morphogen- esis and oncogenesis of MCF-10A mammary epithelial acini grown in three- dimensional basement membrane cultures. Methods, 30(3):256–268, July 2003. [258] A. P. Dempster, N. M. Laird, and D. B. Rubin. Maximum likelihood from incomplete data via the em algorithm. JOURNAL OF THE ROYAL STATIS- TICAL SOCIETY, SERIES B, 39(1):1–38, 1977. [259] A. Bradbury and A. Plckthun. [260] Valerie C Coffman and Jian-Qiu Wu. Counting protein molecules using quanti- tative fluorescence microscopy. Trends in Biochemical Sciences, 37(11):499–506, November 2012. [261] Jennifer C Waters and Torsten Wittmann. Concepts in quantitative fluorescence microscopy, volume 123. Elsevier Inc., 1 edition, 2014. [262] Jingyu Zhang, Xiao-Jun Tian, Hang Zhang, Yue Teng, Ruoyan Li, Fan Bai, Subbiah Elankumaran, and Jianhua Xing. TGF-β-induced epithelial-to- mesenchymal transition proceeds through stepwise activation of multiple feed- back loops. Sci Signal, 7(345):ra91–ra91, 2014. [263] Hiroaki Kajiyama, Kiyosumi Shibata, Mikio Terauchi, Mamoru Yamashita, Kazuhiko Ino, Akihiro Nawa, and Fumitaka Kikkawa. Chemoresistance to paclitaxel induces epithelial-mesenchymal transition and enhances metastatic potential for epithelial ovarian carcinoma cells. Int. J. Oncol., 31(2):277–283, August 2007. [264] So-Yeon Park, Min-Jin Kim, Sang-A Park, Jung-Shin Kim, Kyung-Nan Min, Dae-Kee Kim, Woosung Lim, Jeong-Seok Nam, and Yhun Yhong Sheen. Combinatorial TGF-β attenuation with paclitaxel inhibits the epithelial- to-mesenchymal transition and breast cancer stem-like cells. Oncotarget, 6(35):37526–37543, November 2015. [265] Sreenath V Sharma, Daniel A Haber, and Jeff Settleman. Cell line-based plat- forms to evaluate the therapeutic efficacy of candidate anticancer agents. Nat Rev Cancer, 10(4):241–253, April 2010. [266] Maria P Alcolea and Philip H Jones. Tracking cells in their native habitat: lineage tracing in epithelial neoplasia. 13:1–11, February 2013. [267] Vanessa Almendro, Andriy Marusyk, and Kornelia Polyak. Cellular hetero- geneity and molecular evolution in cancer. Annu. Rev. Pathol. Mech. Dis., 8:277–302, January 2013. [268] El˝od M´ehes and Tam´as Vicsek. Collective motion of cells: from experiments to models. Integr Biol (Camb), 6(9):831–854, September 2014. 195 [269] MD Pope, NA Graham, BK Huang, and AR Asthagiri. Automated quantitative analysis of epithelial cell scatter. Cell Adh Migr, 2(2):1–7, October 2008. [270] Raghvendra Singh, Pedro Lei, and Stelios T Andreadis. PKC-δ binds to E-cadherin and mediates EGF-induced cell scattering. Exp Cell Res, 315(17):2899–2913, October 2009. [271] Johan de Rooij, Andre Kerstens, Gaudenz Danuser, Martin A Schwartz, and Clare M Waterman-Storer. Integrin-dependent actomyosin contraction regu- lates epithelial cell scattering. J Cell Biol, 171(1):153–164, October 2005. [272] Dinah Loerke, Quint le Duc, Iris Blonk, Andre Kerstens, Emma Spanjaard, Matthias Machacek, Gaudenz Danuser, and Johan de Rooij. Quantitative imag- ing of epithelial cell scattering identifies specific inhibitors of cell motility and cell-cell dissociation. Sci Signal, 5(231):rs5, 2012. [273] Venkat Maruthamuthu and Margaret L Gardel. Protrusive activity guides changes in cell-cell tension during epithelial cell scattering. Biophys J, 107(3):555–563, August 2014. [274] T Omelchenko, J M Vasiliev, I M Gelfand, H H Feder, and E M Bonder. Rho- dependent formation of epithelial ”leader” cells during wound healing. Proc Natl Acad Sci USA, 100(19):10788–10793, September 2003. [275] M Poujade, E Grasland-Mongrain, A Hertzog, J Jouanneau, P Chavrier, B Ladoux, A Buguin, and P Silberzan. Collective migration of an epithe- lial monolayer in response to a model wound. Proc Natl Acad Sci USA, 104(41):15988–15993, October 2007. [276] M Reffay, M C Parrini, O Cochet-Escartin, B Ladoux, A Buguin, S Coscoy, F Amblard, J Camonis, and P Silberzan. Interplay of RhoA and mechanical forces in collective cell migration driven by leader cells. Nat Cell Biol, 16(3):217– 223, March 2014. [277] B K Hall and T Miyake. All for one and one for all: condensations and the initiation of skeletal development. Bioessays, 22(2):138–147, February 2000. [278] Carlos Carmona-Fontaine, Eric Theveneau, Apostolia Tzekou, Masazumi Tada, Mae Woods, Karen M Page, Maddy Parsons, John D Lambris, and Roberto Mayor. Complement fragment C3a controls mutual cell attraction during col- lective cell migration. Dev Cell, 21(6):1026–1037, December 2011. [279] Philippe V Afonso, Mirkka Janka-Junttila, Young Jong Lee, Colin P McCann, Charlotte M Oliver, Khaled A Aamer, Wolfgang Losert, Marcus T Cicerone, and Carole A Parent. LTB4 is a signal-relay molecule during neutrophil chemotaxis. Dev Cell, 22(5):1079–1091, May 2012. 196 [280] Colin P McCann, Paul W Kriebel, Carole A Parent, and Wolfgang Losert. Cell speed, persistence and information transmission during signal relay and collective migration. J Cell Sci, 123(Pt 10):1724–1731, May 2010. [281] Peter J Lu and David A Weitz. Colloidal Particles: Crystals, Glasses, and Gels. Annu. Rev. Condens. Matter Phys., 4(1):217–233, April 2013. [282] T A Witten and L M Sander. Diffusion-Limited Aggregation, a Kinetic Critical Phenomenon. Phys Rev Lett, 47(19):1400–1403, November 1981. [283] Paul Meakin. Formation of Fractal Clusters and Networks by Irreversible Diffusion-Limited Aggregation. Phys Rev Lett, 51(13):1119–1122, September 1983. [284] D A Weitz and M Oliveria. Fractal Structures Formed by Kinetic Aggregation of Aqueous Gold Colloids. Phys Rev Lett, 52(16):1433–1436, April 1984. [285] V Trappe, V Prasad, L Cipelletti, P N Segre, and D A Weitz. Jamming phase diagram for attractive particles. Nature, 411(6839):772–775, June 2001. [286] Ludovic Berthier and Giulio Biroli. Theoretical perspective on the glass tran- sition and amorphous materials. Rev. Mod. Phys., 83:587–645, Jun 2011. [287] B Szab´o, G Sz¨oll¨osi, B G¨onci, Zs Jur´anyi, D Selmeczi, and Tam´as Vicsek. Phase transition in the collective migration of tissue cells: Experiment and model. Phys. Rev. E, 74(6), December 2006. [288] Thomas E Angelini, Edouard Hannezo, Xavier Trepat, Manuel Marquez, Jef- frey J Fredberg, and David A Weitz. Glass-like dynamics of collective cell migration. Proc Natl Acad Sci USA, 108(12):4714–4719, March 2011. [289] Dhananjay T Tambe, C Corey Hardin, Thomas E Angelini, Kavitha Rajen- dran, Chan Young Park, Xavier Serra-Picamal, Enhua H Zhou, Muhammad H Zaman, James P Butler, David A Weitz, Jeffrey J Fredberg, and Xavier Trepat. Collective cell guidance by cooperative intercellular forces. Nat Mater, 10(6):469–475, June 2011. [290] Kenechukwu David Nnetu, Melanie Knorr, Steve Pawlizak, Thomas Fuhs, and Josef A K¨as. Slow and anomalous dynamics of an MCF-10A epithelial cell monolayer. Soft Matter, 9(39):9335–7, 2013. [291] Simon Garcia, Edouard Hannezo, Jens Elgeti, Jean-Fran¸cois Joanny, Pascal Silberzan, and Nir S Gov. Physics of active jamming during collective cellu- lar motion in a monolayer. Proc Natl Acad Sci USA, 112(50):15314–15319, December 2015. [292] Jin-Ah Park, Jae Hun Kim, Dapeng Bi, Jennifer A Mitchel, Nader Taheri Qazvini, Kelan Tantisira, Chan Young Park, Maureen McGill, Sae-Hoon Kim, 197 Bomi Gweon, Jacob Notbohm, Robert Steward, Stephanie Burger, Scott H Ran- dell, Alvin T Kho, Dhananjay T Tambe, Corey Hardin, Stephanie A Shore, El- liot Israel, David A Weitz, Daniel J Tschumperlin, Elizabeth P Henske, Scott T Weiss, M Lisa Manning, James P Butler, Jeffrey M Drazen, and Jeffrey J Fred- berg. Unjamming and cell shape in the asthmatic airway epithelium. Nature Mat, 14(10):1040–1048, October 2015. [293] Thuan Beng Saw, Amin Doostmohammadi, Vincent Nier, Leyla Kocgozlu, Sumesh Thampi, Yusuke Toyama, Philippe Marcq, Chwee Teck Lim, Julia M Yeomans, and Benoit Ladoux. Topological defects in epithelia govern cell death and extrusion. Nature, 544(7649):212–216, April 2017. [294] Monirosadat Sadati, Nader Taheri Qazvini, Ramaswamy Krishnan, Chan Young Park, and Jeffrey J Fredberg. Collective migration and cell jamming. Differen- tiation, 86(3):121–125, October 2013. [295] H D Soule, T M Maloney, S R Wolman, W D Peterson, R Brenz, C M McGrath, J Russo, R J Pauley, R F Jones, and S C Brooks. Isolation and characterization of a spontaneously immortalized human breast epithelial cell line, MCF-10. Cancer Res, 50(18):6075–6086, September 1990. [296] B. D. Ripley. Modelling spatial patterns. J Roy Stat Soc Series B Stat Methodol, 39(2):172–212, 1977. [297] Aaron S Keys, Adam R Abate, Sharon C Glotzer, and Douglas J Durian. Mea- surement of growing dynamical length scales and prediction of the jamming transition in a granular material. Nature Phys, 3(4):260–264, March 2007. [298] D Bensimon, E Domany, and A Aharony. Crossover of Fractal Dimension in Diffusion-Limited Aggregates. Phys Rev Lett, 51(15):1394–1394, October 1983. [299] M. Y. Lin, H. M. Lindsay, D. A. Weitz, R. C. Ball, R. Klein, and P. Meakin. Universal reaction-limited colloid aggregation. Phys. Rev. A, 41:2005–2020, Feb 1990. [300] M L Day, X Zhao, J Vallorosi, M Putzi, T Powell, C Lin, and KC Day. E- cadherin Mediates Aggregation-dependent Survival of Prostate and Mammary Epithelial Cells through the Retinoblastoma Cell Cycle Control Pathway*. J Biol Chem, 274(14):1–10, March 1999. [301] Rex L Chisholm and Richard A Firtel. Insights into morphogenesis from a simple developmental system. Nat Rev Mol Cell Biol, 5(7):531–541, July 2004. [302] Hisashi Haga, Chikako Irahara, Ryo Kobayashi, Toshiyuki Nakagaki, and Kazushige Kawabata. Collective movement of epithelial cells on a collagen gel substrate. Biophys J, 88(3):2250–2256, March 2005. [303] D. A. Weitz, J. S. Huang, M. Y. Lin, and J. Sung. Dynamics of diffusion-limited kinetic aggregation. Phys. Rev. Lett., 53:1657–1660, Oct 1984. 198 [304] Peter Carmeliet and Marc Tessier-Lavigne. Common mechanisms of nerve and blood vessel wiring. Nature, 436(7048):193–200, July 2005. [305] Danielle N Rockwood, Rucsanda C Preda, Tuna Y¨ ucel, Xiaoqin Wang, Michael L Lovett, and David L Kaplan. Materials fabrication from Bombyx mori silk fibroin. Nature Protocols, 6(10):1612–1631, September 2011. [306] Mina J Bissell and Mark A LaBarge. Context, tissue plasticity, and cancer. Cancer Cell, 7(1):17–23, January 2005. [307] Celeste M Nelson and Mina J Bissell. Of Extracellular Matrix, Scaffolds, and Signaling: Tissue Architecture Regulates Development, Homeostasis, and Can- cer. Annu Rev Cell Dev Biol, 22(1):287–309, November 2006. [308] Raghu Kalluri and Eric G Neilson. Epithelial-mesenchymal transition and its implications for fibrosis. J. Clin. Invest., 112(12):1776–1784, December 2003. [309] Zhongying Wang, Daniel Tonderys, Susan E Leggett, Evelyn Kendall Williams, Mehrdad T Kiani, Ruben Spitz Steinberg, Yang Qiu, Ian Y Wong, and Robert H Hurt. Wrinkled, wavelength-tunable graphene-based surface topographies for directing cell alignment and morphology. CARBON, 97:14–24, 2016. [310] Samila Nasrollahi and Amit Pathak. Topographic confinement of epithelial clus- ters induces epithelial-to-mesenchymal transition in compliant matrices. Scien- tific Reports, pages 1–12, December 2015. [311] Po-Yen Chen, Jaskiranjeet Sodhi, Yang Qiu, Thomas M Valentin, Ruben Spitz Steinberg, Zhongying Wang, Robert H Hurt, and Ian Y Wong. Multiscale Graphene Topographies Programmed by Sequential Mechanical Deformation. Advanced Materials, 28(18):3564–3571, March 2016. [312] Po-Yen Chen, Muchun Liu, Thomas M Valentin, Zhongying Wang, Ruben Spitz Steinberg, Jaskiranjeet Sodhi, Ian Y Wong, and Robert H Hurt. Hierar- chical Metal Oxide Topographies Replicated from Highly Textured Graphene Oxide by Intercalation Templating. ACS Nano, 10(12):10869–10879, November 2016. [313] Po-Yen Chen, Mengke Zhang, Muchun Liu, Ian Y Wong, and Robert H Hurt. Ultrastretchable Graphene-Based Molecular Barriers for Chemical Protection, Detection, and Actuation. ACS Nano, 12(1):234–244, January 2018. [314] A Singh and J Settleman. EMT, cancer stem cells and drug resistance: an emerging axis of evil in the war on cancer. Oncogene, 29(34):4741–4751, June 2010. [315] Saber Imani, Hossein Hosseinifard, Jingliang Cheng, Chunli Wei, and Junjiang Fu. Prognostic Value of EMT-inducing Transcription Factors (EMT-TFs) in Metastatic Breast Cancer: A Systematic Review and Meta- analysis. Scientific Reports, pages 1–10, June 2016. 199 [316] Michael Zeisberg and Eric G Neilson. Biomarkers for epithelial-mesenchymal transitions. J. Clin. Invest., 119(6):1429–1437, June 2009. [317] Tian Tian Wang, Chang-Chang Jia, Ying-Fen Hong, Dan-Yun Ruan, Dong-hao Wu, Xing Li, and Xiang-yuan Wu. The density of cancer-associated fibrob- lasts as a prognostic indicator for hepatocellular carcinoma. Journal of Clinical Oncology, 34(15):e15606, May 2016. [318] Sang Yun Ha, So-Young Yeo, Yan-hiua Xuan, and Seok-Hyung Kim. The Prog- nostic Significance of Cancer-Associated Fibroblasts in Esophageal Squamous Cell Carcinoma. PLoS ONE, 9(6):e99955–9, June 2014. [319] Leilei Tao, Guichun Huang, Haizhu Song, Yitian Chen, and Longbang Chen. Cancer associated fibroblasts: An essential role in the tumor microenvironment. Oncology Letters, 14(3):2611–2620, June 2017. [320] Kyra Campbell and Jordi Casanova. A common framework for EMT and col- lective cell migration. Development, 143(23):4291–4300, November 2016. [321] Nicola Aceto, Mehmet Toner, Shyamala Maheswaran, and Daniel A Haber. En Route to Metastasis: Circulating Tumor Cell Clusters and Epithelial-to- Mesenchymal Transition. Trends Cancer, 1(1):44–52, September 2015. [322] A Pathak and S Kumar. Independent regulation of tumor cell migration by ma- trix stiffness and confinement. Proceedings of the National Academy of Sciences, 109(26):10334–10339, June 2012. [323] N E Sounni and A No¨el. Targeting the Tumor Microenvironment for Cancer Therapy. Clinical Chemistry, 59(1):85–93, January 2013. 200