Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Development and application of microfluidic single-cell polymerase chain reaction White, Adam 2015

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


24-ubc_2016_february_white_adam.pdf [ 8.58MB ]
JSON: 24-1.0220863.json
JSON-LD: 24-1.0220863-ld.json
RDF/XML (Pretty): 24-1.0220863-rdf.xml
RDF/JSON: 24-1.0220863-rdf.json
Turtle: 24-1.0220863-turtle.txt
N-Triples: 24-1.0220863-rdf-ntriples.txt
Original Record: 24-1.0220863-source.json
Full Text

Full Text

Development vny Vpplixvtion ofbixrouiyix hingleBCelleolymervse Chvin gevxtionbyAdam WhiteB.Sc. (Honours Physics), The University of British Columbia, 2007M.A.Sc. (Biomedical Engineering), The University of British Columbia, 2010A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDoctor of PhilosophyinThe Faculty of Graduate and Postdoctoral Studies(Genome Science & Technology)The University of British Columbia(Vancouver)December 2015© Adam White 2015VwstrvxtMethods for single-cell analysis are critical to revealing cell-to-cell variability in biolog-ical systems, such as during development or onset of disease, where the characteristicsof heterogeneity and minority cell populations are obscured by population-averagedmeasurements. Analysis of individual cells has been limited due to challenges associ-ated with small amounts of starting material, combined with the cost and throughputrequired to examine large numbers of cells. Microfluidic approaches are well suited tosingle-cell analysis, providing increased sensitivity, economy of scale, and automation.This thesis presents the development and application of microfluidic technologyfor single cell gene expression analysis. The foundational contribution of this workis an integrated microfluidic device capable of performing high-precision RT-qPCRmeasurements of gene expression from hundreds of single cells per run. This deviceexecutes all steps of single cell processing including cell capture, cell lysis, reversetranscription, and quantitative PCR. This device is further expanded upon by inte-grating the single cell and nucleic acid processing capabilities with final measurementof cDNA by high-density digital PCR. The direct quantification of single moleculesby digital PCR has advantages over RT-qPCR in the measurement of low abundancetranscripts, as well as obviating the need for relative abundance measurements orcalibration standards. This technology is demonstrated in over 5,000 individual cellmeasurements of mRNA, microRNA, and single nucleotide variant detection in a va-riety of cell types. Finally, this technology is applied to study the performance of lipidnanoparticles in delivery of RNA, and manipulation of gene expression in cells. Themicrofluidic integration of cell and nucleic acid processing established in this thesispermits analysis of hundreds of single cells in parallel, while improving work flow andreducing technical variation compared to samples prepared in microliter volumes. Ul-timately, this advances the tools available for precisely measuring transcripts in singlecells, and has application in research and clinical settings.iierefvxeThe work presented in this thesis is part of a collaborative effort to develop andapply microfluidic systems for single cell analysis, and has resulted in co-authoredpublications.A version of Chapter 2 has been published: Adam K. White, Michael VanIns-berghe, Oleh Petriv, Mani Hamidi, Darek Sikorski, Marco A. Marra, James M. Piret,Sam Aparicio, and Carl L. Hansen, Hizh-ghrouzhput Mivrouidiv finzlx-Vxll eg-qcVe, Proceedings of the National Academy of Sciences, 2011. AKW and MVcontributed equally. AKW and MV designed and fabricated microfluidic devices.AKW, MV, OP, and MH performed on-chip experiments and analyzed data. MVdeveloped image analysis code. OP and MH developed OCT4 assays and performedoff-chip experiments. DS performed hESC differentiation experiments and mRNAFISH measurements. CLH, SA, MM, and JP designed research. AKW, MV, andCLH wrote the manuscript.A version of Chapter 3 has been published: Adam K. White, Kevin A. Heyries,Colin Doolin, Michael VanInsberghe, and Carl L. Hansen, Hizh-ghrouzhput Mivrou-idiv finzlx-Vxll Dizittl cVe, Analytical Chemistry, 2013. AKW and KAH con-tributed equally. CLH designed research. AKW, KAH, CD and MV designed andfabricated devices. AKW, KAH, and CD performed experiments. AKW, KAH, CD,and MV analyzed data. AKW, KAH and CLH wrote the manuscript.A manuscript for publication based on Chapter 4 is in preparation: finzlx VxllTntlysis oy Lipid atnoptrtivlx eaT Dxlivxry. This work is a collaboration with Pre-cision Nanosystems, who provided lipid nanoparticles. I performed all experiments,and designed all experiments with input from Precision Nanosystems (specificallyAysha Ansari, David Zwaenepoel, Colin Walsh, Euan Ramsay and James Taylor)and Carl Hansen. Darek Sikorski and James Piret provided tissue culture support.iiiivwle of ContentsVwstrvxt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iierefvxe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiivwle of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivaist of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiiVxknofileygements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xF Introyuxtion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Why Measure Gene Expression? . . . . . . . . . . . . . . . . . . . . 11.3 Transcriptional Variability Between Single Cells . . . . . . . . . . . . 21.4 Types of Transcripts Defining Cellular State . . . . . . . . . . . . . . 31.5 Techniques for Single Cell Measurements of Gene Expression . . . . 51.5.1 Molecular Imaging . . . . . . . . . . . . . . . . . . . . . . . . 51.5.2 RNA-Sequencing . . . . . . . . . . . . . . . . . . . . . . . . . 61.5.3 Quantitative PCR Methods . . . . . . . . . . . . . . . . . . . 71.6 Integrated Microfluidic Technology for Single Cell Gene ExpressionAnalysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91.6.1 Multilayer Soft Lithography . . . . . . . . . . . . . . . . . . . 91.6.2 Review of Microfluidic Technology for Single Cell Analysis . . 111.7 Research Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171.7.1 Specific Aims . . . . . . . . . . . . . . . . . . . . . . . . . . . 17G HighBihroughput bixrouiyix hingleBCell giBqeCg . . . . . . . . 192.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19ivhuvly of Contynts2.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212.3.1 Device Fabrication and Operation . . . . . . . . . . . . . . . 212.3.2 Single Cell Transcript Measurements by Heat Lysis and 2-stepRT-qPCR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222.3.3 Single Cell Transcript Measurements by Chemical Lysis and1-step RT-qPCR . . . . . . . . . . . . . . . . . . . . . . . . . 242.3.4 Digital PCR Experiments . . . . . . . . . . . . . . . . . . . . 252.3.5 System for Real-Time PCR . . . . . . . . . . . . . . . . . . . 262.3.6 RT-qPCR Assays . . . . . . . . . . . . . . . . . . . . . . . . . 272.3.7 Image Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 282.3.8 mRNA FISH . . . . . . . . . . . . . . . . . . . . . . . . . . . 282.3.9 Cell Culture . . . . . . . . . . . . . . . . . . . . . . . . . . . 292.3.10 Transfer Efficiency Measurements . . . . . . . . . . . . . . . . 302.3.11 Cell Capture Measurements . . . . . . . . . . . . . . . . . . . 302.3.12 Mixing by Diffusion . . . . . . . . . . . . . . . . . . . . . . . 312.4 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 332.4.1 Device Design . . . . . . . . . . . . . . . . . . . . . . . . . . 332.4.2 Validation of Integrated Single Cell RT-qPCR . . . . . . . . . 402.4.3 Application to Measurement of Single Cell miRNA Expression 452.4.4 Co-regulation of miR-145 and OCT4 in Single Cells . . . . . 482.4.5 SNV Detection in Primary Cells . . . . . . . . . . . . . . . . 502.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51H HighBihroughput bixrouiyix hingleBCell Digitvl eCg . . . . . . 543.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 543.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 573.3.1 Device Fabrication . . . . . . . . . . . . . . . . . . . . . . . . 573.3.2 Device Operation . . . . . . . . . . . . . . . . . . . . . . . . 603.3.3 Cell Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . 603.3.4 Detecting mRNA with Chemical Lysis and RT-dPCR . . . . 613.3.5 Detecting miRNA with Heat Lysis and RT-dPCR . . . . . . . 613.3.6 Measurement of RNA Editing . . . . . . . . . . . . . . . . . . 643.3.7 Cell Culture and RNA Purification . . . . . . . . . . . . . . . 653.3.8 Device Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 65vhuvly of Contynts3.3.9 Transfer Efficiency Measurements . . . . . . . . . . . . . . . . 653.3.10 Ripleys K-Function . . . . . . . . . . . . . . . . . . . . . . . 663.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 663.4.1 Device Performance . . . . . . . . . . . . . . . . . . . . . . . 663.4.2 Single Cell Transcript Measurements . . . . . . . . . . . . . . 703.5 Discussion & Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . 73I hingleBCell Vnvlysis of aipiy cvnopvrtixle gcV Delivery . . . . . 784.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 784.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 794.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 824.3.1 Microfluidic Single-Cell Digital PCR . . . . . . . . . . . . . . 824.3.2 Cell Culture . . . . . . . . . . . . . . . . . . . . . . . . . . . 834.3.3 Lipid Nanoparticle Formulation . . . . . . . . . . . . . . . . . 834.3.4 Transfection . . . . . . . . . . . . . . . . . . . . . . . . . . . 834.3.5 Flow Cytometry . . . . . . . . . . . . . . . . . . . . . . . . . 854.3.6 Microscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 854.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 854.4.1 ApoE-Dependent LNP Uptake . . . . . . . . . . . . . . . . . 854.4.2 siRNA Knockdown . . . . . . . . . . . . . . . . . . . . . . . . 884.4.3 mRNA Delivery . . . . . . . . . . . . . . . . . . . . . . . . . 944.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1054.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1095 Conxlusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1115.1 Contribution to Knowledge . . . . . . . . . . . . . . . . . . . . . . . 1115.2 Future Recommendations . . . . . . . . . . . . . . . . . . . . . . . . 1125.2.1 Extending Microfluidic Single-Cell Analysis . . . . . . . . . . 1125.2.2 Further Experiments and Applications for Lipid NanoparticleDelivery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1155.3 Final Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117Wiwliogrvphy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118vihuvly of ContyntsVppenyixesV Design Consiyervtions . . . . . . . . . . . . . . . . . . . . . . . . . . . 141A.1 Single-Cell Digital PCR Prototype with Alternative Approach to Mix-ing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141W erotoxols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144B.1 Fabrication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144B.1.1 Flow Layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144B.1.2 Control and Blank Layer(s) . . . . . . . . . . . . . . . . . . . 144B.1.3 Bake Flow and Control Layers . . . . . . . . . . . . . . . . . 145B.1.4 Align Flow Layer to Control Layer . . . . . . . . . . . . . . . 145B.1.5 Ports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146B.1.6 Mounting Individual Devices . . . . . . . . . . . . . . . . . . 146B.1.7 General Considerations . . . . . . . . . . . . . . . . . . . . . 146B.2 Device Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147B.2.1 Cell Loading, Washing, and Heat Lysis . . . . . . . . . . . . . 147B.2.2 Reverse Transcription . . . . . . . . . . . . . . . . . . . . . . 148B.2.3 Real-Time Polymerase Chain Reaction . . . . . . . . . . . . . 148B.2.4 Chemical Lysis and One-Step RT-qPCR . . . . . . . . . . . . 149viiaist of Figures1.1 Valves in MSL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102.1 Mixing by diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322.2 Design and operation of the microfluidic device . . . . . . . . . . . . 332.3 Precision and sensitivity of microfluidic RT-qPCR . . . . . . . . . . . 352.4 Single cell loading and transcript measurements . . . . . . . . . . . . 382.5 The size distribution of cells isolated by microfluidic traps . . . . . . 392.6 Single-cell measurements . . . . . . . . . . . . . . . . . . . . . . . . . 412.7 On-chip cell washing . . . . . . . . . . . . . . . . . . . . . . . . . . . 422.8 Comparison of RT performed in the microfluidic device or in tubes . . 432.9 Number of cells is reflected in corresponding cycle threshold values . . 442.10 Single cell miRNA measurements . . . . . . . . . . . . . . . . . . . . 462.11 Optical multiplexing of single cell RT-qPCR . . . . . . . . . . . . . . 492.12 mRNA-FISH of OCT4 in CA1S cells . . . . . . . . . . . . . . . . . . 503.1 Microfluidic device design and operation . . . . . . . . . . . . . . . . 583.2 Characterization of the microfluidic device . . . . . . . . . . . . . . . 673.3 Digital PCR measurements . . . . . . . . . . . . . . . . . . . . . . . . 683.4 The spatial distribution of digital counts . . . . . . . . . . . . . . . . 693.5 Digital PCR measurements on K562 single cells . . . . . . . . . . . . 703.6 Measurement of single nucleotide RNA editing . . . . . . . . . . . . . 723.7 The digital array response curve . . . . . . . . . . . . . . . . . . . . . 744.1 ApoE-dependent LNP uptake in hESC . . . . . . . . . . . . . . . . . 864.2 Replicates for ApoE-dependent LNP uptake experiment in hESC . . 874.3 siNRA knockdown of HPRT in human embryonic stem cells . . . . . 894.4 siNRA knockdown of GAPDH in human embryonic stem cells . . . . 904.5 Low-end of GAPDH expression following 1.0 µg/mL siGAPDH for 24hrs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91viiiList of Figurys4.6 Single-cell dose-response curve for siGAPDH treatment in K562 cells 934.7 Single-cell flow cytometry measurement of LNP delivery and eGFPexpression in K562 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 944.8 Single-cell measurement of mRNA delivered by LNP in suspensionculture (K562) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 954.9 Effect of different gates on flow cytometry measurement of LNP deliv-ery and eGFP expression in K562 cells . . . . . . . . . . . . . . . . . 964.10 Single-cell flow cytometry measurement of LNP delivery and eGFPexpression in BJ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 984.11 Single-cell measurement of mRNA delivered by LNP in adherent cul-ture (BJ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 994.12 Inducing gene expression with mRNA delivery . . . . . . . . . . . . . 1004.13 Inducing gene expression with delivery of two different mRNAs . . . . 1014.14 Time-course measurement of mRNA-LNP performance with daily dosesfor two weeks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1024.15 Distribution of delivered mRNA in BJ cells with daily doses for twoweeks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1034.16 Distribution of delivered mRNA in cells under mock reprogrammingconditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1064.17 Comparison of cell numbers in different media conditions . . . . . . . 107A.1 Design schematic of an early prototype single cell digital PCR device 142ixVxknofileygementsI would first like to thank my supervisor, Dr. Carl Hansen, who has been a fantasticadvisor since the end of my undergraduate degree in Physics, and throughout mygraduate degrees in Biomedical Engineering and now Genome Science and Technol-ogy. It has been been a long journey! Carl, thank you for supporting my development,and sharing your ambition and passion for excellence. Working with you was a privi-lege, and I am forever grateful for the many opportunities you provided me to enrichmy experience.I also wish to thank my committee members, Drs. James Piret, Sam Aparicio,and Martin Hirst for their insight and support of my PhD work. Jamie, thank youfor your long-time support of my career. I have benefited greatly from your optimismand thoughtful perspectives on everything from experimental design to career and lifedecisions. I wish to thank Dr. Sam Aparicio for his enthusiasm to harness microfluidictechnology for single-cell genomics, and providing many ideas for how microfluidictechnology could be used for scientific inquiry. Sincere thanks to Dr. Martin Hirstfor his invaluable input since the beginning of this project, and for the scientific rigorhe brings to all our discussions. I would also like to thank my collaborator, Dr.Marco Marra, for being one of the visionaries to create the interdisciplinary GenomeScience and Technology program, and for his role in pushing microfluidics for single-cell genomics. I am very grateful to NSERC, MSFHR, Genome BC, and CIHR forsupporting my research endeavours.Special thanks to all my colleagues in the Hansen lab over the years for boththeir friendship and their helpful scientific discussions. In particular, I would like toacknowledge the exceptional team I was part of in developing the microfluidic systemspresented in this thesis: Michael VanInsberghe, Kevin Heyries, and Calum Doolin.We shared many failures, but ultimately achieved our goals. Their contribution tothis work is enormous. I would like to thank fellow “assassins” Hans Zahn andGeorgia Russell for sharing their enthusiasm for single-cell microfluidics. My sinceregratitude is extended Anupam Singhal, Jens Huft, Kaston Leung, Veronique Lecualt,xAwknowlyxgymyntsDarek Sikorski, and Tim Leaver for all their support of my research and life pursuits.We all started around the same time, and it was fun to share in the excitement ofbuilding novel devices to solve biological problems. Thank you Bertin Wong, AdamQuiring, Mani Hamidi, Marketa Ricicova, and Oleh Petriv for enriching life in theHansen lab. It would not have been the same without you. I have also greatlyappreciated all the advice I have received from Marijn van Loenhout on careers andnavigating academia. Thanks to Darek Sikorski for patiently teaching me how toculture embyronic stem cells, and for that Pearl Jam concert ticket. I would also liketo thank Navid Ghaffari, Blanche Lo, Judy Booth, and Chris Sherwood for alwaysmaking me feel welcome in the Piret lab. Chris, thank you for everything you do tokeep the tissue culture facilities working, and all your help in setting up different cellcultures.Thanks to the fantastic collaborators from Precision Nanosystems: Aysha Ansari,David Zwaenepoel, Colin Walsh, Euan Ramsay and James Taylor. Working withyou, I greatly enjoyed taking this project in new directions and learning more aboutbiotechnology entrepreneurship and gene therapy. I particularly want to thank AyshaAnsari for her patience with trying to get primary neurons to work in my microfluidicdevice.I would also like to thank Dr. Connie Eaves, Karen Lambie, and Gloria Shaw fromthe Stem Cell Assay Lab at the BC Cancer Agency for taking a chance on an inex-perienced physics co-op student back in 2003. That 8 month experience introducedme to lab work, and greatly contributed to launching me down this interdisciplinaryresearch path.Thanks to my dear friends Mike, Dana, Paul, Yichwin, Luke, Sabrina, Julia,Emilie, Martha, Ingrid and Sophie for your support despite the distance separatingus. Finally, I thank my family: their love and encouragement has been invaluable tomy success. I could not have done this without you.xiChvpter FIntroyuxtion1FCF IntroyuxtionThis dissertation describes the engineering and application of novel microfluidic de-vices for high-throughput single-cell gene expression analysis. In particular, work hasfocused on the microfluidic integration of components for single-cell handling, nucleicacid processing, and measurement of transcripts by performing reverse transcriptionfollowed by polymerase chain reaction (PCR). Considerable advances have been madein developing microfluidic technology for single-cell PCR, and in demonstrating thistechnology in biological applications benefiting from single-cell analysis. Here I re-view literature relevant to the development of single-cell gene expression analysiswith particular emphasis on microfluidic systems. The specific aims of this thesis areprovided at the end of the chapter.FCG lhy bevsure Gene EflpressionTThe cells of all organisms inherit genetic instructions that govern their behaviour.This complete set of genetic instructions (the genome) is composed of DNA and con-tains the code for genes. Cell differences arise in part as a result of different genesbeing actively (or not) expressed. There are also regulatory mechanisms that affectthe functionality of an expressed gene. Gene expression is the process by which ge-netic information is transcribed from DNA into RNA, and this RNA is translatedinto proteins (or processed into bioactive non-coding RNA). These proteins are chiefcomponents of the complicated molecular interactions inside a cell, performing tasksincluding catalysis of chemical reactions, intra- and extra-cellular signaling, and main-1ghis vhtpter is tn upwttew version oy the introwuvtory vhtpter yrom Microuidic gxchnologyyor High-ghroughput finglx Cxll Gxnx Exprxssion Tntlysis, uy Twtm White, hniversity oy UritishVolumuit, 2CDCA ghis version is upwttew with reyerenves tnw wisvussion oy revent twvtnvements inmivrouiwivs tnw single vell tntlysisA1EBGB hrunswriptionul juriuvilitfl Vytwyyn gingly Cyllstaining structural integrity. In order to maintain life, the type of proteins, abundance,and timing of gene expression is tightly regulated. A single gene can have profoundlydifferent functions depending on the timing, location, and amount of gene expres-sion [1]. Gene expression is regulated at the levels of transcription, RNA processing,translation, post-translational modifications, and degradation, allowing cells to re-spond to their environment. The transcriptome, encompassing all RNA transcriptsof the cell, represents all proteins that are actively being synthesized and thus pro-vides a unique signature of cell state. Thus, measuring transcripts allows for directlystudying cellular processes and variation.FCH irvnsxriptionvl kvrivwility Wetfieen hingleCellsMuch of our biomolecular knowledge of cells and cell tissue is the result of geneexpression measurements of transcription. However, transcription measurements aretraditionally performed on bulk samples of large numbers (thousands to millions) ofcells. The transcriptional variability between individual cells is obscured by ensembleaveraging. Heterogeneity is an ever present feature of biological systems, and cellularheterogeneity has been observed in cell types ranging from bacteria to mammals. Thesources of transcriptional variability between single cells include the stochastic natureof transcription, different stages of cell cycle, differentiation, and disease.Although heterogeneity between two individual cells can arise through differencesin the genome, even isogenetic cells exhibit variability in transcript expression. Thesedifferences may occur through several different mechanisms. The molecular kineticsinvolved in gene expression make it a stochastic process, subject to noise [2, 3]. Thisresults in bursts of expression and apparently random fluctuations that contribute tophenotypic variation through various feedback mechanisms [2, 3].Gene expression will also naturally vary between cells in different stages of the celldivision cycle. Cell division is initiated by external stimuli such as growth factors,and progression through the stages of division is governed by two classes of regulatorymolecules, cyclins, and cyclin-dependent kinases [4]. Cyclin-dependent kinases areconstitutively expressed in cells whereas cyclins are synthesised at specific stages ofthe cell cycle [5, 6]. The duration of time spent in different stages of the cell cyclevaries, and cell populations divide asynchronously [7]. This asynchronous division2EB4B hflpys of hrunswripts Dyning Cyllulur gtutyresults in transcriptional variability between cells. For this reason many studies oftengo to great effort to synchronize cell cultures, or arrest cells in a specific phase [5, 8].Transcriptional variability is also particularly significant in cells undergoing dif-ferentiation. Through differentiation, it is hypothesized that stem cells or progenitorcells asymmetrically divide in order to generate many different cell types. For ex-ample, a single embryonic stem cell develops into a multicellular organism includingmuscle cells, brain cells, and skin cells, all of which are characterized by very distincttranscriptional programs. Asymmetric division results in cells that express differ-ent genes, giving them different behaviour and diverging fates [9]. In many cases,the phenotype of stem cells and cells undergoing differentiation is not well defined.For example, current state-of-the-art enrichment strategies result in a population ofhematopoietic stem cells in which approximately 50% are capable of reconstitutingthe blood of mice, as determined by a functional assay [10]. Therefore, bulk analysisof hematopoietic stem cells obscures the relevant sub population.Transcription differences between cells can also arise due to the onset of disease.Diseases such as cancer often have their origin in a single cell [11]. Environmentalexposures, or the accumulation of genetic mutations through multiple cell lineagescan lead to pronounced heterogeneity in the tumor cells which is manifest as aberrantgene expression [12–14].FCI iypes of irvnsxripts Dening Cellulvr htvteThe two classes of RNA transcripts that determine cell state are messenger RNA(mRNA), which code for proteins, and non-coding RNA such as microRNA (miRNA)which act as regulators of gene expression.RNA is transcribed from DNA which is subsequently processed into mRNA, andgenetic information is encoded in the sequence of nucleotides. This nucleic acidis translated into protein according to codons, consisting of three bases each, whichencode for specific amino acids. mRNA consists of a 5’ cap, a coding region containingthe codons for translation, 5’ and 3’ untranslated regions, and a 3’ poly-adenine tail.A typical mammalian cell contains thousands of different types of proteins in varyingabundance. Proteins involved in metabolic functions, and structural integrity of thecell are generally found in high abundance, and are often referred to as housekeepinggenes. Although less abundant, mRNA transcripts also produce proteins involved3EB4B hflpys of hrunswripts Dyning Cyllulur gtutyin intra- and extracellular signaling, as well as transcription factors. Transcriptionfactors are proteins involved in the process of transcribing DNA into RNA, and playa significant role in regulating gene expression. Transcription factors have DNA-binding domains that give them the ability to bind to specific sequences of DNAcalled enhancer or promoter sequences. Some transcription factors bind to a DNApromoter sequence near the transcription start site and help form the transcriptioninitiation complex. Other transcription factors bind to regulatory sequences, such asenhancer sequences, and can either stimulate or repress transcription of the associatedgene.Non-coding RNAs such as transfer RNA (tRNA) and ribosomal RNA (rRNA) areknown to be essential in translating mRNA into protein. However, several varieties ofshort, non-coding RNAs such as small nucleolar RNA (snoRNA), and small interfer-ing RNA (siRNA) are increasingly being shown to play important roles in regulatinggene expression. In particular, microRNAs (miRNA) have been found to be regula-tors of gene expression and are drivers in development and cancer [15]. Discoveredin 1993, miRNA are a species of small (approximately 22 nucleotides) non-proteincoding RNA. Primary miRNA transcripts are transcribed as stemloop structures thatare then processed by a protein complex known as the Microprocessor complex (con-sisting of the nuclease Drosha and the double-stranded RNA binding protein Pasha)into shorter structures [16]. Further processing is performed in the cytoplasm bythe endonuclease Dicer, which cleaves the stemloop to form the mature miRNA [16].Interaction with Dicer initiates the formation of the RNA-induced silencing complex(RISC), responsible for the gene silencing observed due to miRNA expression andRNA interference [16]. This RISC-integrated miRNA strand regulates gene expres-sion by binding to complementary mRNA molecules and inhibiting translation orinducing degradation (by argonaute proteins of the RISC complex). Hundreds ofmiRNA species are known in humans, and their short sequence length of the critical5’ seed region makes them complementary to hundreds of mRNA transcripts that canpotentially be targeted for regulation [17]. Recent research has revealed tissue-specificdistributions of miRNAs appearing at different stages of mammalian development.In particular, Chen and colleagues demonstrated that overexpression of a select fewmiRNAs (e.g. miR-181a) can influence hematopoiesis [18], and Calin et al. providedevidence for miRNA involvement in cancer by determining that miR-15a and miR-16a are down regulated in over 68% of chronic lymphocytic leukemia patients [19].4EBIB hywhniquys for gingly Cyll ayusurymynts of Gyny EfipryssionThese findings suggest the potential application of using miRNA expression profilesto identify those miRNAs involved in human cancer development. Importantly forthe current work, single cell measurements of miRNA in highly purified cell popula-tions have been found to exhibit low cell-cell variability, suggesting that miRNA maybe a very useful biomarker of cellular state [20].FC5 iexhniques for hingle Cell bevsurements ofGene EflpressionEarly changes in cell state are first revealed in the transcriptome, where quantita-tive measurements with single molecule sensitivity are possible by both imaging andRT-PCR techniques. This section reviews the current state-of-the-art for single cellmeasurements of transcription.FC5CF bolexulvr ImvgingFluorescent in situ hybridization (FISH) is a technique for detecting specific DNA se-quences in fixed cells. FISH uses fluorescent microscopy to image fluorescently labeledprobes that bind to DNA with similar sequences. In 1998, Femino et al. modifiedFISH and digital imaging techniques in order to detect single RNA molecules [21].Specifically, multiple oligodeoxynucleotide probes were synthesized with five fluo-rochromes per molecule. The probes, each about 50 nucleotides long, are designed tohybridize to adjacent locations on the mRNA target such that their collective fluo-rescence becomes visible as a diffraction-limited spot. Single molecules are measuredby processing images acquired from a series of focal planes through a hybridizedcell. Combinations of these probes labeled with spectrally distinct colours have beenused to measure up to 11 genes simultaneously [22]. Raj et al. improved uponthis technique by probing each mRNA species with 48 or more short, singly labeledoligonucleotide probes [23], which improved the ability to resolve single transcripts.This technique has been applied in two studies looking at the effect of variable geneexpression on cell fate [24] and response [25]. By counting transcripts of the genes ina network in individual Vtxnorhtuditis xlxztns embryos (up to 200-cell stage), Rajet al. showed that the expression of an otherwise redundant gene (xnd-D) becomeshighly variable in skn-D mutants and that this variation is subjected to a threshold,5EBIB hywhniquys for gingly Cyll ayusurymynts of Gyny Efipryssionproducing an ON/OFF expression pattern of the master regulatory gene of intestinaldifferentiation [25]. Beyond quantifying mRNA in single cells, mRNA-FISH revealsthe location of the transcript inside fixed cells [23]. Also, the spatial organizationof gene expression among fixed cells can be assessed. Similar hybridization tech-niques have also been combined with rolling circle amplification for in situ detectionof mRNA [26–28].Although mRNA-FISH has been successfully applied [24, 25, 29], the system hasnot been widely adopted. One reason for this is the difficulty in synthesizing heavilylabeled oligonucleotides [23]. Additionally, mRNA-FISH requires a long protocol in-volving fixing cells, hybridizing probes, washing unbound probes, and taking stacksof images using fluorescent microscopy. This procedure requires highly specializedand expensive equipment and reagents. Processing the stack of focal plane images re-quires exhaustive deconvolution and is computationally intensive [23]. Furthermore, itis challenging to unambiguously identify all the fluorescent spots as mRNA moleculesas it is impossible to determine whether the detection of an individual probe arisesfrom legitimate binding to the target mRNA or from nonspecific binding [23]. Theuse of multiple probes bound to a single transcript also presents challenges in dis-tinguishing between closely related sequences. Small RNA species, such as miRNAs,are too short to accommodate multiple probes, making them refractory to analysisby FISH. Throughput of mRNA-FISH is limited by cost, labour intensive protocols,and imaging.FC5CG gcVBhequenxingThe deep coverage provided by next-generation sequencing has recently permitteda direct approach to single cell gene expression measurements by sequencing RNA,known as RNA-Seq or whole transcriptome shotgun sequencing (WTSS) [30]. Thereare many different approaches to RNA-Seq. In one example, mRNA is captured onpoly(T) coated magnetic beads prior to reverse transcription. The cDNA is then frag-mented, size selected, and sequenced [31]. The deep coverage allows expression levelsto be estimated based on the extent to which a sequence is detected [32]. RNA-Seqhas recently been applied to single cells of the inner cell mass from human embryonicstem cell development [33]. This study looked at expression dynamics of 385 genesin 74 single cells [33]. Further improvements in instrumentation, bioinformatics ap-proaches, and optimized reagent kits (such as the template-switching method from6EBIB hywhniquys for gingly Cyll ayusurymynts of Gyny EfipryssionClonetech) [34, 35] are rapidly establishing the coupling of whole transcriptome RNAamplification from single cells with high-throughput sequencing [36–38]. The primaryadvantage of RNA-Seq is that it permits analysis of much of the transcriptome, par-ticularly mRNAs. The major limitation to this approach is representation bias, whichmakes RNA-Seq poorly suited to diagnostics or other applications where the abun-dance of a given molecular species is in question [39]. Further bottlenecks in single celltranscriptome sequencing are cost (although sequencing costs are rapidly dropping),and sample preparation. In one of the early single cell RNA sequencing studies from2010, small numbers of single cells were laboriously isolated by mouth pipetting[33],limiting high-throughput application. However, droplet-based approaches and flowsorting are increasingly addressing this limitation.FC5CH fuvntitvtive eCg bethoysReverse transcription quantitative polymerase chain reaction (RT-qPCR) provides apowerful and sensitive method for quantitative analysis of transcript levels, and hasbeen extensively applied to single cell analysis [40–47]. RT-qPCR is based on thetraditional polymerase chain reaction (PCR), which is a method for specifically andexponentially amplifying DNA starting from as little as a single copy [48]. In RT-qPCR, the first strand of DNA is synthesized from a RNA template through a processcalled reverse transcription (RT). Oligonuclotides (primers) designed to be comple-mentary to the transcript of interest are used to specifically transcribe the RNA intoa complementary DNA (cDNA). After first strand synthesis, real-time quantitativePCR (qPCR) is performed similar to conventional PCR, however a fluorescent re-porter probe is added to the reaction. During the annealing stage of the PCR, bothprobe and primers anneal to the DNA target. A variety of molecular probes havebeen developed for RT-qPCR including intercalating dyes [49], molecular beacons[50], scorpion probes [51], and hydrolysis probes. Fluorescence (Forster) resonanceenergy transfer (FRET) probes are perhaps the most commonly used [52]. Theseprobes consist of a dual labeled DNA oligo, having a fluorophore and a quencherat the 5’ and 3’ ends respectively, that is complementary to an internal region ofthe amplicon. In close proximity, the quenching molecule prevents detection of thefluorescent molecule by absorbing energy from the reporter through a process calledForster resonance energy transfer (FRET) [53, 54]. Following annealing, the poly-merization of a new DNA strand is initiated from the PCR primers. Upon reaching7EBIB hywhniquys for gingly Cyll ayusurymynts of Gyny Efipryssionthe oligonuceotide of the probe, the exonuclease activity of the polymerase degradesthe probe, physically separating the fluorophore from the quenching moeity, and re-sulting in an increase in fluorescence. Fluorescence is detected through the use ofphotodetectors or a charge coupled device (CCD). Monitoring the fluorescent signalof the PCR reaction allows for quantitative measurements of transcript levels [55, 56].The specificity, sensitivity, dynamic range, and quantitative accuracy make RT-qPCR the most common technique for gene expression analysis [57]. RT-qPCR issensitive enough to detect transcripts at the level of single cells, and a number ofdifferent strategies for single cell RT-qPCR have been reported [40–43, 45–47, 58].Bengtsson et al. used RT-qPCR to reveal lognormal distributions of mRNA in singlecells of pancreatic islets of Langerhans [40]. RT-qPCR is also able to target smallRNAs, such as miRNAs, through the use of a stem-loop RT primer that yields a longercDNA strand for annealing qPCR primers and probes [59]. This stem loop primersystem has been used to perform highly multiplexed miRNA transcript measurementsin single embryonic stem cells [60, 61]. However, the application of RT-qPCR to largenumbers of single cells has been limited in part due to the high cost of probes andreagents. Addtionally, laborious techniques such as mouth pippetting, micropipet-ting, and FACS are used to isolate single cells for RT-qPCR reactions [45, 46]. Thelatter, although automated, requires careful optimization and calibration which makeit difficult or impossible to confirm single cell capture.Microfluidic lab-on-chip technology has enabled digital PCR (dPCR), wherebysingle DNA molecules are quantified by compartmentalizing a sample into thousandsof nano- or pico-liter PCR reactions. The sample is diluted such that each reactionchamber has a high probability of containing 1 or 0 molecules. After PCR in thepresence of a fluorescent probe, each reaction chamber in the array will be either flu-orescent if the PCR reaction was successful, or not fluorescent if the reaction did notoccur (i.e. no DNA template present). An end-point image of the array of reactionchambers can be used to detect DNA in a on/off (digital) format. Digital PCR hasadvantages over conventional quantitative PCR in that the measurement is absolute,and no reference gene or calibration curve is needed for comparison. This means PCRassays with different efficiencies can be directly quantified and compared. Further-more, by spatially isolating each PCR reaction, contributions from contaminant ornon-specific reactions are limited. This can reduce the chance of competing reactionsleading to a false-positive in the case of interrogating a single nucleotide variant. Dig-8EB6B Intygrutyx aiwrouixiw hywhnologfl for gingly Cyll Gyny Efipryssion Anulflsisital PCR has been applied to quantify expression of transcription factors in a limitednumber of single cells, following FACS isolation of single cells and RNA processing(lysis, RT reaction) in tubes [62]. More recently, the advent of commercial devices forperforming dPCR in (nanoliter) droplets is facilitating widespread adoption of thistechnique in a variety of applications [63, 64]. The small reaction volumes in dPCRprovide a 1000-fold reduction in reagent consumption cost, while providing singlemolecule sensitivity. However, the bottleneck in single cell dPCR remains laborioussingle cell isolation and sample preparation.FCK Integrvtey bixrouiyix iexhnology for hingleCell Gene Eflpression VnvlysisMicrofluidic systems offer a number of advantages for single cell analysis of geneexpression. A challenge in single cell RT-qPCR is the limited starting material [46].Microfluidics improve reaction sensitivity by reducing the volume of reactions, therebyincreasing the concentration of template. Reducing reaction volumes also decreasescostly reagent consumption. Microfluidic devices are automatable and highly scalable,permitting high-throughput and cost-effective application. Furthermore, the precisefluid handling capability of microfluidic systems is ideal for delicate manipulation ofsingle cells and the assembly of reactions with low technical variability.FCKCF bultilvyer hoft aithogrvphyThe microfluidic technology developed in this thesis is based on a fabrication tech-nique called multilayer soft lithography (MSL) [65]. In MSL, silicon wafers coveredin photoresist are exposed to UV light through a micro-patterned photomask. Thismask determines the pattern of features on the wafer after the resist is developed.For example, a typical negative resist such as SU8 consists of a non-photosensitivesubstrate, a photosensitive cross-linking agent, and a coating solvent. Crosslinks formwhen the photoresist is exposed to UV light, and the resist polymerises. This exposedphotoresist is now insoluble in a developer solution, while unexposed sections of thephotoresist are subsequently washed away by the developer. Using different pho-toresists (and coating spin speeds), wafers can be fabricated with features of varyingheights and shapes.9EB6B Intygrutyx aiwrouixiw hywhnologfl for gingly Cyll Gyny Efipryssion AnulflsisThese patterned wafers are used as replica molds for slabs of polydimethylsiloxane(PDMS) that are stacked on top of each other. Replica molding allows low costproduction of multiple chips from a single silicon master. Bonding between layers isachieved by complementary off-ratio stochiometric mixing of the potting prepolymercomponent (A) and hardener component (B) of the room temperature vulcanizingPDMS for each slab. For example, the normal stochiometric ratio of masses A:B is10:1. Bonding can be achieved between PDMS layers of A:B components, such as20:1 and 5:1.Figure 1.1: Valves in MSL. (A) A schematic profile of valve geometry in MSL devices.(B) Applying pressure to fluid in a ‘control’ channel deflects the membrane with the‘flow’ channel to effectively valve the ‘flow’ line. These valves can be integrated intodevices as a peristaultic pump, shown in inverse microscope images with all valvesoff (C), and one valve on (D).A simple microfluidic device can be created from a ‘control’ wafer, and a ‘flow’wafer [65]. A thick slab of PDMS (with excess hardener) molded to the features of the‘flow’ wafer can be peeled from the ‘flow’ wafer and bonded to a thin layer of PDMS(with excess potting prepolymer) molded to the ‘control’ wafer. After bonding, thisdouble-slab of PDMS can be peeled from the ‘control’ wafer, punched with holes for10EB6B Intygrutyx aiwrouixiw hywhnologfl for gingly Cyll Gyny Efipryssion Anulflsisfluid inlets/outlets, and bonded to a blank layer of PDMS to close the bottom ofthe ‘control’ layer channels. Applying pressure (controlled off-chip by solenoids) onthe fluid in a control line can deflect the membrane between orthogonally crossingchannels in the adjacent ‘flow’ layer, effectively valving the flow channel (Figure 1.1).Microfluidic devices integrate thousands of these valves (100 extmum x 100 extmum)in order to partition channels, direct fluid flow, and build active structures such asperistaltic pumps and mixers [65, 66].FCKCG geviefi of bixrouiyix iexhnology for hingle CellVnvlysisThe past decade has seen a surge of efforts to manipulate and analyze individual cellsin controllable lab-on-a-chip devices [67, 68]. As microfluidic technology has matured,many of the functionalities required for single cell gene expression analysis, such ascell trapping or nucleic acid detection, have been demonstrated in forms ranging fromproof-of-concept to commercial products [29, 69]. Fitting these pieces of the puzzletogether into a single integrated microfluidic device for high throughput, cost effectivesingle cell gene expression analysis remains the current challenge.FCKCGCF Cell bvnipulvtionSingle cell manipulation and isolation is a task well suited to micro-scale devices,and a number of distinct strategies have been demonstrated. In particular, physicaltrapping, encapsulation in droplets, and dielectrophoresis trapping techniques showgreat potential for single cell analysis.Physical trapping of single cells in microfluidic devices has been accomplishedthrough integration of microwells [70], cups [71–75], weirs [76, 77], and active valving[78–80]. Wheeler et al. designed a microfluidic device capable of passively isolatingand trapping a single cell from a bulk suspension by positioning a square-cup “dock”with small drain channels at the stagnation point of a T-junction [75]. Similar strate-gies have explored different cup shapes [74, 77], and densely arraying the traps forlarge-scale experiments [72]. Skelley et al. developed a device for cell pairing thatfeatured 6,000 physical cell traps made of polydimethylsiloxane (PDMS) [81]. Thecell traps were densely arrayed (in an area of 8 mm by 4 mm) within a flow-throughchannel. Each cell trap consisted of a capture cup, and support pillars on either side11EB6B Intygrutyx aiwrouixiw hywhnologfl for gingly Cyll Gyny Efipryssion Anulflsisof the capture cup to allow flow into and under the trap. The pillar heights weredesigned to be slightly smaller than the cell diameter in order to trap a cell uponentering the capture cup. The obstruction provided by a trapped cell impedes fluidflow through the cell trap resulting in subsequent cells flowing past to be capturedby unoccupied cell traps. Skelley et al. observed that the trap spacing in the arraywas critical for efficient capture without clogging. With optimal column spacing of1-1.5 cell diameters (e20 µm), and a row spacing of 20-50 µm, Skelley et al. captured70 - 90% of cells entering the array [81]. This physical cell-trapping array is a highlyparallel and scalable technique.Microfluidic technology for generating monodisperse droplets of aqueous phasesolution inside an inert oil have been applied to the encapsulation of single cells.Once encapsulated inside a droplet, the droplet acts as an individual test tube, andreagents (other droplets) can be combined, and reactions can be carried out. Koster etal. developed a microfluidic device to encapsulate individual cells in picoliter aqueousdrops in a carrier fluid at rates of up to 250 Hz [82]. In addition to cells remainingviable for up to 6 hours incubating inside 33 pL droplets, the small volumes of thedrops enables the concentrations of secreted molecules such as antibodies to rapidlyattain detectable levels. One limitation to this system is variability in the numberof cells per drop due to stochastic cell loading. However, Edd et al. solved thisissue by designing a high aspect-ratio microchannel that hydrodynamically focusescells to be evenly spaced as they travel within the channel [83]. Thus, individual cellsenter the drop generator with the frequency of drop formation. Encapsulation of cellswithin picolitre-size monodisperse drops provides new means to perform large-scalequantitative biological studies on a single-cell basis.Microfluidic devices integrated with active electronics have been used manipulatecells with electric fields. This approach offers the advantage that the cells are notphysically contacted. In dielectrophoretic cell trapping, a non-uniform electric fieldis generated, and the force applied to the cell depends on the dipole induced withinthe cell. Voldman et al. used four monolithic pillars within a microfluidic channel aselectrodes to create a quadropole dielectrophoresis cell trap [84]. Dielectrophoresiscan be selective in only trapping particular cell types, such as selecting white bloodcells instead of erythroctyes [84]. In addition, the traps can be switched ‘on’ or ‘off’to facilitate cell recovery or subsequent manipulation. Alternatively, optical tweezershave also been combined with microfluidic devices for single cell manipulation [85–88].12EB6B Intygrutyx aiwrouixiw hywhnologfl for gingly Cyll Gyny Efipryssion AnulflsisFCKCGCG gcV eroxessingTechniques such as RT-qPCR and sequencing often require RNA manipulations in-cluding purification or reverse transcription into cDNA before further analysis. Singlecell capture, lysis, and reverse transcription have been implimented in a microfluidicrotary mixer that that may be injected with cell sample, reagents, and output RTproduct [78]. The throughput is limited to one cell, however. Bontoux et al. appliedthis system to neuronal progenitors, followed by template switching PCR in a tube,and reported the detection of 5000 genes in each cell (corresponding to the expectedtotal number of genes expressed) by microarray analysis [78]. However, due to thelow reported correspondence between different cells and the lack of data analysis it isunclear how much of this signal was specific. Interestingly, Bontoux et al. reportedthat the RT reaction was more efficient in nanoliter volumes inside the microfluidicdevice compared to microliter volume reactions in tubes [78].Zhong et al. reported a microfluidic device capable of purifying mRNA from20 single cells using oligo(dT) beads, followed by recovery for off-chip qPCR [89].Individual cells were stochastically isolated by partitioning a cell suspension betweenphysical microvalves. A chemical buffer was mixed with the sample fluid to lyse thecell. The cell lysate was pushed through a column of beads functionalized with shortsequences of deoxy-thymine nucleotides on the surface. The oligo(dT) strand bindsthe poly-A tail of mRNA transcripts, thereby capturing the mRNA from the cellwhile the remaining contents are washed away. By performing reverse transcriptionof the purified mRNA on the microfluidic device, Zhong et al. demonstrated a e4-fold increase in reverse transcription reaction efficiency (measured by cDNA yield)compared to performing the reaction in conventional tubes [89]. Measurement of3 transcripts in 54 single hESCs revealed a heterogeneous population [89], furtherunderscoring the need for discrete cell analysis. The throughput of this system islimited by stochastic cell loading, and challenges recovering the samples from thedevice for off-chip analysis.FCKCGCH Gene Eflpression VnvlysisMicrofluidic devices employing arrays of thousands of nanolitre micro-reactors havebeen used by researchers for highly multiplexed, as well as single molecule quantitativeanalysis of cDNA prepared from single cell samples. Digital PCR is performed in amicrofluidic “Digital Array” whereby a 7.5 µl sample is partitioned into 1,200 isolated13EB6B Intygrutyx aiwrouixiw hywhnologfl for gingly Cyll Gyny Efipryssion Anulflsisreaction chambers (“wells”), before PCR [62]. The sample is diluted such that eachreaction chamber has high probability of containing a single template molecule, orzero, allowing absolute quantification of single molecules by counting the numberof fluorescent reaction wells after PCR amplification. Warren et al. used FACS tosort 116 individual cells using hematopoietic differentiation markers to select cellsrepresenting hematopietic stem cells (HSC), common lymphoid progenitors (CLP),two sub-populations of common myeloid progenitors (CMP), and megakaryocyte-erythroid progenitors (MEP) [62]. Following off-chip reverse transcription, abundanceof transciption factor PU.1 was quantified using digital PCR. Warren et al. were ableto show differential expression of PU.1 between flk+ and flk- CMPs with single cell(and single molecule) resolution.Integrated fluidic circuits for digital PCR have been commercialized by Fluidigm,which also offers a Dynamic Array chip for quantitative analysis by highly multiplexedreal-time PCR. Following conventional (off-chip) sample preparation (including re-verse transcription and pre-amplification of cDNA), the Dynamic Array combines 48samples with 48 assays, to perform 2,304 real-time PCR reactions, each 10 nL involume (also available as 96 samples by 96 assays). In addition to fluid handlingadvantages, performing the equivalent multiplexing real-time PCR experiments inconventional microliter volumes quickly becomes cost prohibitive. This multiplexingallows for large-scale gene expression profiling, starting with small samples such assingle cells.The Biomark Dynamic Array (Fluidigm Corporation) technology has been lever-aged to study cellular development from zygote to blastocyst stage fertilized mouseembryos [90]. In 2010, Guo et al. investigated expression of 48 genes in a survey of 500single cells from 8, 16, 32, and 64 -cell stage embryos [90]. By tracking multiple ex-pression markers, Guo et al. revealed at least three distinct developmental expressionpatterns, and associate these with development of cells forming the trophectoderm(TE), the primitive endoderm (PE), and the epiblast (EPI) [90]. Furthermore, Id2and Sox2 were identified as the earliest markers of outer and inner cells, respectively.These results illustrate the power of single cell gene expression analysis to provideinsight into developmental mechanisms [91], and this technique is applicable to otherbiological systems [92]. The coupling of single cell isolation by flow cytometry, fol-lowed by RNA processing and final multiplexed qPCR in these microfluidic arrayshas since become well established for analyzing hundreds to low thousands of cells for14EB6B Intygrutyx aiwrouixiw hywhnologfl for gingly Cyll Gyny Efipryssion Anulflsislarge assay panels [93]. Importantly, these studies have uncovered clinically relevantcellular heterogeneity in diseases such as cancer. In 2011, Dalerba et al. used single-cell microfluidic RT-qPCR to identify distinct cell populations within colon cancertissues, and showed that the different gene expression signatures are predictive ofpatient survival and clinical outcomes [94]. Single-cell analysis has also uncoveredthat a subset of breast tumor cells exhibit increased expression of genes associatedwith reactive oxygen species scavenging, contributing to tumor radioresistance [95].FCKCGCI htvte of the VrtO Integrvtey hystems for Cell bvnipulvtion vnygcV VnvlysisThe above examples demonstrate the single cell handling, nucleic acid processing, andanalysis capabilities of microfluidic devices, however complete integration of all sam-ple processing and analysis into a single device remains a pursuit of active research.This was particularly true at the outset of the work presented in this thesis. At thattime, the only device to integrate all components of single cell isolation, RNA process-ing, and final measurement of gene expression was from Toriello et al., who developedan integrated microfluidic device for single cell gene expression analysis capable ofcapturing a single cell, cell lysis and reverse transcription of contained mRNA, fol-lowed by amplification and detection of product of interest [96]. The device features ananoliter metering pump, and DNA capture pads to catch functionalized single cells.An integrated heating element is used for cell lysis, followed by RT-PCR. The 200 nLPCR chamber is coupled to capillary electrophoresis for size-based measurement ofproducts. Each device is capable of measuring 4 single cells in parallel, and is usedto measure variable siRNA knockdown of the GAPDH gene in 8 individual Jurkatcells [96]. Other efforts towards microfluidic integration of single cell gene expressionanalysis in droplet [97] or micro-well [98] systems have suffered from lysate inhibitionof RT-qPCR reactions, reducing the sensitivity, precision, and robustness of thesemeasurements. These devices demonstrated the feasibility of a microfluidic approachto single cell expression analysis, however further development was still needed formicrofluidic based methods to become routine in single cell analysis.Here we present the first microfluidic device to achieve integration of all compo-nents for single cell gene expression analysis, such as single cell trapping, washing,lysis, reverse transcription and quantitative PCR measurements, at high-throughput(for the time) of hundreds of cells per run. In the course of my thesis work developing15EB6B Intygrutyx aiwrouixiw hywhnologfl for gingly Cyll Gyny Efipryssion Anulflsisand applying this technology, the field of microfluidics and single-cell gene expressionanalysis has advanced considerably. In particular, through publishing to disseminateknowledge, and licensing intelectual property, this work contributed to the devel-opment of a commercial microfluidic product (the C1, from Fluidigm Corp.) thatintegrates single cell trapping with nucleic acid processing with recovery of single-cellproducts for downstream assessment by sequencing or highly paralel PCR (such ason the Biomark Dynamic Array). The C1 currently has a throughput of up to 96cells per run. The C1 device has been gaining popularity [36, 99], enabling single-cellgenomics and transcriptomics studies in a variety of applications [38, 100, 101].In addition to integrated microfluidic devices using successive physical chambersfor reactions [102], micro-wells [103, 104] and droplet-based [105] approaches have alsomade advances. In a technique called Drop-seq, thousands of cells were separated intonano-liter sized aqueous droplets [106]. Each droplet was used to associate a differentmolecular identifier, or barcode, for transcripts originating from the encapsulated cell[107]. This allowed all of the cells to be sequenced together, while retaining tran-scripts’ cell of origin. Drop-seq has enabled the studies with over 44,000 single celltranscriptomes sequenced [106], the highest throughput reported to date. Molecularimaging techniques such as mRNA-FISH have been combined with in situ RNA se-quencing to look at a highly multiplexed number of genes while preserving spatialinformation of where the transcripts are located within the cell [108, 109]. Thesesingle cell gene expression analysis methods are beginning to deliver on the promiseof single cell genomics, but improving on these techniques remains an active area ofresearch.16EB7B fysyurwh cvjywtivyFCL gesevrxh dwjextiveTranscription measurements with single cell resolution are critical to understandingvariable responses in immunity, measuring stochastic noise in gene expression, andassessing the disease and developmental state of heterogeneous populations. Cur-rent methods for measuring transcript levels in single cells include reverse transcrip-tion followed by quantitative polymerase chain reaction (RT-qPCR), single moleculecounting using digital PCR or hybridization probes, and next generation sequencing.Widespread adoption of these techniques has been limited by challenges isolating sin-gle cells, high reagent costs, low sensitivity, and difficulties in accurately measuringlow abundance transcripts. Microfluidic systems offer a number of advantages forsingle cell analysis of gene expression by providing economy of scale, automation andparallelization, and increased sensitivity in small volume reactions. Furthermore, theprecise fluid handling capability of microfluidic systems is ideal for delicate manipu-lation of single cells.This thesis is focused on the development of new technologies for single cell anal-ysis with the following goals:1. Develop a microfluidic system for PCR-based transcription measurements insingle cells.2. Apply this technology to explore transcriptional heterogeneity in single cells.FCLCF hpexix Vims1. Development of an integrated microfluidic device for high-throughput single-cellRT-qPCR2. Development of an integrated microfluidic device for high-throughput single-celldigital PCR3. Application of single-cell digital RT-PCR to study the use of lipid nanoparticlesfor RNA delivery.Integrating single cell capture, lysis, and reverse transcription, with final mea-surement of transcripts by digital PCR will permit analysis of hundreds of single cellsin parallel, while improving work flow and reducing technical variation compared to17EB7B fysyurwh cvjywtivysamples prepared in microliter volumes. Thus, microfluidic single-cell digital PCRrepresents a significant advancement to the tools available for precisely measuringtranscripts in single cells, and has application in research and clinical settings.18Chvpter GHighBihroughput bixrouiyixhingleBCell giBqeCg1GCF dverviefiA long-sought milestone in microfluidics research has been the development of in-tegrated technology for scalable analysis of transcription in single cells. Here wepresent a fully integrated microfluidic device capable of performing high-precisionRT-qPCR measurements of gene expression from hundreds of single cells per run.Our device executes all steps of single cell processing including cell capture, cell ly-sis, reverse transcription, and quantitative PCR. In addition to higher throughputand reduced cost, we show that nanoliter volume processing reduced measurementnoise, increased sensitivity, and provided single nucleotide specificity. We apply thistechnology to 3300 single cell measurements of i) miRNA expression in K562 cells, ii)co-regulation of a miRNA and one of its target transcripts during differentiation inembryonic stem cells, and iii) single nucleotide variant detection in primary lobularbreast cancer cells. The core functionality established here provides the foundationfrom which a variety of on-chip single cell transcription analyses will be developed.GCG IntroyuxtionSingle cells represent the fundamental unit of biology; however, the vast majorityof biological knowledge has emerged as a consequence of studying cell populationsand not individual cells. Inevitably, there are fundamental and applied questions,such as those relating to transcriptional control of stem cell differentiation, intrinsic1T version oy this vhtpter hts ueen puulishewM Twtm White, Mivhtel itnInsuerghe, bleh cetriv,Mtni Htmiwi, Wtrek fikorski, Mtrvo TA Mtrrt, Jtmes MA ciret, ftm Tptrivio, tnw Vtrl LA Htnsen,High-ghroughput Microuidic finglx-Cxll eg-qcCe, croveewings oy the attiontl Tvtwemy oy fvi-enves, 2CDDA19FBFB Introxuwtionnoise in gene expression, and the origins of disease, that can only be addressed atthe single cell level. For example, single cell analysis allows for the direct measure-ment of gene expression kinetics, or for the unambiguous identification of co-regulatedgenes, even in the presence of de-synchronization and heterogeneity that could ob-scure population-averaged measurements. Similarly, single cell methods are vital instem cell research and cancer biology, where isolated populations of primary cells areheterogeneous due to limitations in purification protocols, and it is often a minoritycell population that is the most relevant. High-throughput single cell measurementtechnologies are therefore of intense interest and have broad application in clinicaland research settings.Existing methods for measuring transcript levels in single cells include RT-qPCR[46],single molecule counting using digital PCR[62] or hybridization probes[23, 26], andnext generation sequencing[30]. Of these, single cell RT-qPCR provides combined ad-vantages of sensitivity, specificity, and dynamic range, but is limited by low through-put, high reagent cost, and difficulties in accurately measuring low abundance transcripts[58].Microfluidic systems provide numerous advantages for single cell analysis: economiesof scale, parallelization and automation, and increased sensitivity and precision thatcomes from small volume reactions. Considerable effort over the last decade has beendirected towards developing integrated and scalable single cell genetic analysis onchip[67, 110]. Thus, many of the basic functionalities for microfluidic single cell geneexpression analysis have been demonstrated in isolation, including cell manipulationand trapping[75, 81], RNA purification and cDNA synthesis[78, 79, 89], and microflu-idic qPCR[90] following off-chip cell isolation cDNA synthesis and preamplification.In particular, microfluidic qPCR devices (Biomark Dynamic Array, Fluidigm) haverecently been applied to single cell studies[20, 95]. Although these systems provide ahigh-throughput qPCR readout, they do not address the front end sample preparationand require single cell isolation by FACS or micropipette followed by off-chip process-ing and pre-amplification of starting template prior to analysis. The critical step ofintegrating all steps of single cell analysis into a robust system capable of performingmeasurements on large numbers of cells has yet to be reported. A single demonstra-tion of an integrated device for directly measuring gene expression in single cells wasdescribed by Toriello et al., combining all steps of RNA capture, PCR amplification,and end-point detection of amplicons using integrated capillary electrophoresis[96].Despite the engineering complexity of this system, throughput was limited to four20FBGB aythoxscells per run, cell capture required metabolic labeling of the cells, and the analysis wasnot quantitative. Thus, there remains an unmet need for microfluidic technologiescapable of scalable and quantitative single cell genetic analysis.Here we describe an integrated microfluidic device for high-throughput RT-qPCRanalysis of mRNA and miRNA expression at a throughput of hundreds of single cellsper experiment. We show that this technology provides a powerful tool for scalablesingle cell gene expression measurements with improved performance, reduced cost,and higher sensitivity as compared to analysis in µL volumes. This technology rep-resents the first implementation of robust and high-throughput single cell processingand amplification of nucleic acids on a chip, thereby achieving a major milestone inmicrofluidic biological analysis.GCH bethoysGCHCF Devixe Fvwrixvtion vny dpervtionMicrofluidic devices were fabricated by multilayer soft lithography[65, 66]. Planarsilicon molds were defined by photolithography, using photomasks designed with CADsoftware (AutoCAD, Autodesk Inc.), and printed on transparency films at a resolutionof 20,000 dots per inch (CAD/Art services). The control mold was fabricated usingSU8-2025 photoresist (Microchem, USA) to deposit valve features 24 µm in height.The flow mold was fabricated with three lithographic steps. First, the channelsfor reagent injection, and connections between chambers were fabricated using 13µm high SPR220-7 photoresist (Shipley, USA). The SPR channels were rounded tofacilitate valve closure by incubation at 115 ◦C for 15 minutes. A hard bake at190 ◦C for 2 hours was used to prevent SPR photoresist erosion during addition ofsubsequent layers. Second, the cell trap features were defined in 14 µm SU8-2010photoresist (Microchem, USA). Finally, the large chambers and fluidic bus lines wereconstructed using 150 µm high SU8-100 photoresist. All photoresist processing wasperformed according to manufacturer specifications.Microfluidic devices were cast from these molds in polydimethylsiloxane (PDMS,RTV615, General Electric, USA). Each device consists of a three layer elastomericstructure with a blank bottom layer, a middle control layer with channels that actas valves by pushing up and pinching closed channels in the above flow layer. The21FBGB aythoxsmolds were first treated with chlorotrimethylsilane (TMCS, Aldrich) vapor for 2 minto prevent PDMS from bonding to the photoresist structures. The flow layer wasmade by pouring a mixture of PDMS (5 parts RTV615A : 1 part RTV615B) ontothe flow mold, degassing, and then baking for 60 min at 80 ◦C. A thin control layerwas made by spin coating the control mold with PDMS (20 parts RTV615A : 1 partRTV615B) at 1800 rpm and baking for 45 min at 80 ◦C. After baking, the PDMSof the flow layer was peeled from the flow mold and aligned to the control layer.Following a 60 min bake at 80 ◦C, the bonded two layer structure was separated fromthe control mold, and channel access holes were punched. A blank layer (withoutchannels) was prepared by spinning PDMS (20 parts RTV615A : 1 part RTV615B)on a blank wafer (2000 rpm) and baking 45 min at 80 ◦C. The bonded flow andcontrol structure was mounted on to the blank layer, and baked for 3 hours at 80 ◦C.Finally, the three layer bonded structure was removed from the blank mold, dicedinto individual devices, and these were each bonded to clean glass slides by bakingovernight at 80 ◦C.The device operation requires control of 9 pneumatic valves and may be operatedusing a simple manifold of manual valves. For the current study a semi-automatedimplementation was used in which microfluidic valves were controlled by solenoidactuators (Fluidigm Corp., San Francisco) controlled through a digital input out-put card (NI-DAQ, DIO-32H, National Instruments) operated using LabView drivers(National Instruments). Tygon tubing connected the solenoids to the microfluidicdevice by 20 gauge stainless steel pins (Small Parts Inc.) fitted into the control lineports. Krytox (DuPont) oil was used as the fluid in the control lines, and the valveswere actuated with 30 psi pressure.GCHCG hingle Cell irvnsxript bevsurements wy Hevt aysisvny GBstep giBqeCgThe device was primed by flowing PBS containing 0.5 mg/mL bovine serum albumin(BSA) and 0.5 U/µL RNase Inhibitor through all channels, while keeping the RT,and PCR chambers empty and isolated by valves. The BSA helped prevent cells fromadhering to channel walls. After priming, but prior to cell loading, all valves wereclosed. A single cell suspension was injected into the device by applying pressure (∼2-3 psi) to microcapillery pipette tips plugged into the sample inlets. The sample inlets22FBGB aythoxswere first dead-end filled against an inlet valve to prevent air bubbles from enteringthe device. The sample inlet valves, cell chamber valves and outlet valve were openedto allow the cell suspension to flow through the sample channels. Cells were loadedinto the device suspended in culture media (directly from culture). Cell loadingconcentrations were kept between 5×105 cells/mL and 1×106 cells/mL, resultingin over 80% occupancy of cell traps with single cells in 1-2 min at a flow rate ofapproximately 20 nL/s. Lower concentrations were found to require proportionatelylonger times to achieve high occupancy of trapped single cells. Concentrations greaterthan 2×106 cells/mL were found to occasionally clog the inlet port or the channel attrap locations. A peristaltic pump was integrated into the device for controlling theflow rate, however pressure driven flow was used for the current study.After injecting the cell suspension and trapping single cells the cell sample inletvalve was closed, and the cells were washed by flushing the line with the PBS solutionused to prime the device. This removed untrapped single cells, extracellular RNA, anddebris. Following on-chip washing, the cell chamber valves were closed to partitionthe cell loading channel and isolate individual cell reactors. Visual inspection of thecell capture chambers with a microscope was used to confirm and count the numberof cells in each chamber. The cells were lysed by placing the microfluidic device ontoa flatbed thermocycler and heating to 85 ◦C for 7 minutes (and then cooled to 4◦C).Reverse transcription (RT) was performed in the device by using the ABI HighCapacity Reverse Transcription kit[59], with the addition of a surfactant to preventadsorption of nucleic acids and proteins to PDMS surfaces (2 µL 10× Reverse Tran-scription Buffer, 4 µL 5X RT stem-loop miRNA primer from ABI, 1 µL 100mMdNTPs, 1.34 µL of 50 U/µL Multiscribe Reverse Transcriptase, 0.26 µL of 20 U/µLRNase Inhibitor, 2 µL 1% Tween 20, 9.4 µL PCR grade water). The RT mix wasloaded into the device, and flushed through the reagent injection channels. RT reagentwas injected into the reaction by opening the valve connecting the cell chamber to theRT chamber, and the valve connecting the cell chamber to the reagent injection line.The RT chamber was dead-end filled before closing the connection to the reagentinjection line. A pulsed temperature RT protocol was carried out by placing themicrofluidic device on a flatbed thermocycler (2 min at 16 ◦C, followed by 60 cyclesof 30 seconds at 20 ◦C, 30 seconds at 42 ◦C, and 1 second at 50 ◦C). RT enzyme wasinactivated at 85 ◦C (5 min), and then the device was cooled to 4 ◦C.The PCR reagent was prepared with 25 µL of 2× TaqMan Universal Master Mix23FBGB aythoxs(ABI), 2.5 µL 20× Real-Time miRNA assays (primers and probe, ABI), 5 µL of 1%Tween 20, and 7.5 µL of PCR grade water. The PCR reagent was flowed throughthe reagent injection channels to flush away the RT reagent. Valves were openedand the PCR reagent was injected to dilute the RT product into the PCR reactionchamber. After completely filling the PCR reaction chamber, the valves closing thePCR chambers were actuated, and the device was transferred to an enclosure for real-time PCR (Prototype version of Biomark Instrument, Fluidigm CA). The real-timePCR enclosure consists of a custom flatbed thermocycler, a xenon arc lamp and filterset, and a charged coupled device (CCD) imager with optics for fluorescent imaging ofthe entire device periodically during PCR thermocycling (see description of real-timePCR instrumentation below). PCRs were thermocycled with the following conditions:10 min at 95 ◦C, followed by 50 cycles of 15 s at 95 ◦C and 1 min at 60 ◦C. Imageswere acquired at 60 ◦C.GCHCH hingle Cell irvnsxript bevsurements wy Chemixvlaysis vny FBstep giBqeCgMeasurements of mRNA transcripts (SP1, GAPDH) were performed using the CellsDirect kit (Invitrogen, USA). Operation of the microfluidic device for chemical lysisand 1-step RT-qPCR was similar to the methods described for heat lysis and 2-stepRT-qPCR with several distinctions. The device was primed and cells were washedwith PBS containing 0.5 mg/mL BSA. Additional RNase Inhibitor was omitted asthe chemical lysis buffer (10 µL lysis resuspension buffer, 1 µL lysis enhancer solution,Invitrogen, USA) contained RNA stabilizing agents. Cell loading was the same as inthe heat lysis and 2-step RT-qPCR scenario. Single cells were lysed by injecting achemical lysis buffer through the cell capture chamber and filling the 10 nL chamber(used for RT reagent injection in the 2-step protocol). The lysis reaction was incu-bated at room temperature for 10 minutes, followed by heat inactivation of the lysisreagent by placing the device on a flatbed thermocycler and incubating at 70 ◦C for10 minutes. The one-step RT-qPCR mix (1 µL of SuperScript III RT/Platinum TaqMix, 25 µL of 2X Reaction Mix (with ROX reference dye), 2.5 µL of 20X TaqmanAssay (primers and probes, ABI), 1 µL of 50 mM MgSO4, 5.5 µL of H2O, and 5 µLof 1% Tween 20) was then combined with the cell lysate into the final 50 nL reactionchamber. The device was transferred to the real-time PCR enclosure for temperature24FBGB aythoxsTable 2.1: Heat Lysis and 2-Step RT-qPCR ProtocolStep Description Time1 Prime device with PBS 0.5 mg/mL BSA and 0.5 U/µL RNase In-hibitor1 min2 Inject cell suspension (passive cell trapping) 1 min3 On chip cell washing with PBS containing 0.5 mg/mL BSA and0.5 U/µL RNase Inhibitor1 min4 Close valves partitioning cell loading channel and isolating singlecells30 sec5 Count cells by visual inspection with microscope 7 mins6 Heat lysis by placing device on flatbed thermocycler and heatingto 85 ◦C7 min7 Flush fluidic bus and reagent injection lines with reagent for RT 2 min8 Inject RT reagent through the cell capture chamber, dead-end fill-ing the 10 nL RT chamber1 min9 Close reagent injection valve, creating isolated reactors combiningthe cell capture chamber and RT chamber30 sec10 Perform reverse transcription (pulsed temperature protocol) byplacing device on flatbed thermocycler2.5 hr11 Flush fluidic bus and reagent injection lines with reagent for PCR 2 min12 Inject PCR reagent through combined cell-capture/RT chamberinto 50 nL PCR chamber5 min13 Close valve to PCR chamber. Allow for mixing by diffusion 40 min14 Load device into BioMark real-time PCR system and focus camera 5 min15 Run qPCR protocol (varies)control and imaging of the 1-step RT-qPCR (20 min at 50 ◦C for RT, followed by ahot-start at 95 ◦C for 2 min, and 50 cycles of 15 s at 95 ◦C and 30 s at 60 ◦C).GCHCI Digitvl eCg EflperimentsFor mRNA digital PCR analysis cells were washed with PBS containing 0.5 mg/mLBSA, lysed in chemical lysis buffer, reverse transcription was performed in tubes ac-cording to the protocol described above, and the resulting cDNA product was loadedinto digital PCR arrays. For miRNA studies, cells were lysed in PBS containing 0.5mg/mL BSA and 0.5 U/µL RNase inhibitor. Reverse transcription was performedusing miRNA stem-loop primers (Applied Biosystems, USA) and the High CapacitycDNA Reverse Transcription kit (Applied Biosystems, USA) in 10 µL volumes. Prior25FBGB aythoxsTable 2.2: Chemical Lysis and 1-step RT-qPCR ProtocolStep Description Time1 Prime device with PBS 0.5 mg/mL BSA and 0.5 U/µL RNase In-hibitor1 min2 Inject cell suspension (passive cell trapping) 1 min3 On chip cell washing with PBS containing 0.5 mg/mL BSA and0.5 U/µL RNase Inhibitor1 min4 Close valves partitioning cell loading channel and isolating singlecells30 sec5 Count cells by visual inspection with microscope 7 mins6 Inject lysis reagent through the cell capture chamber, dead-endfilling the 10 nL chamber1 min7 Close reagent injection valve, creating isolated reactors combiningthe cell capture chamber and lysis reservoir chamber30 sec8 Perform lysis at room temperature and heat inactivation of thelysis reagent at 75 ◦C by placing device on flatbed thermocycler25 min9 Flush fluidic bus and reagent injection lines with reagent for RT-qPCR2 min10 Inject RT-qPCR reagent through combined cell-capture/lysischamber into 50 nL RT-qPCR chamber5 min11 Close valve to RT-qPCR chamber. Allow for mixing by diffusion 40 min12 Load device into BioMark real-time PCR system and focus camera 5 min13 Run RT-qPCR protocol (varies)to injection into microfluidic digital PCR arrays, RT product was added to the PCRreagent as in the on-chip 2-step RT-qPCR protocol described above. Thermal cy-cling of digital PCR arrays was also performed using the same protocols as describedabove. PDMS digital PCR arrays consisting of 765 2 nL individual PCR chambers,of similar design to those described in Warren et al.[62], were fabricated by multilayersoft lithography. After thermal cycling, positive chambers were counted and actualmolecule numbers were derived based on the binomial distribution.GCHC5 hystem for gevlBiime eCgThe BioMark™ Reader is a commercially available real-time PCR instrument devel-oped by Fluidigm and designed to run Fluidigm Integrated Fluidic Circuits (IFCs).The prototype version of this system allowed access to the flatbed thermocycler in-side the enclosure, permitting the use of custom microfluidic devices in addition to26FBGB aythoxsthe intended commercial IFCs. Image Resolution and bit depth: 4 Megapixel, 16bit; Filters: FAM (Ex 485/20, Em 525/25), VIC (Ex 530/20, Em 570/30), ROX (Ex580/25, Em 610/15), QAS (Ex 580/25, Em 680/25); Light Source: 175 W xenon arcbulb.GCHCK giBqeCg VssvysMeasuring mRNA in the presence of genomic DNA requires primers designed tospecifically target mature mRNA sequences. In many cases, this can be accomplishedby designing intron-spanning primers. A specially designed stem-loop RT primersystem (Applied Biosystems) is used for the specific targeting of mature miRNAs.TaqMan assays for GAPDH (Applied Biosystems, Assay ID Hs99999905 m1) andmiRNAs were obtained from Applied Biosystems. For GAPDH, a control experimentomitting the reverse transcriptase was performed off-chip, in microliter volumes withbulk cell lysate (at equivalent concentration of a single cell on-chip, 105 cells/mL),and showed no amplification after 40 cycles of PCR.OCT4 (POU5F1) primer sequences were obtained from RTPrimerDB and syn-thesized by Biosearch Technologies Inc; Forward primer: ACC CAC ACT GCAGCA GAT CA, Reverse primer: CAC ACT CGG ACC ACA TCC TTC T, Probe:Quasar670-CCA CAT CGC CCA GCA GCT TGG-BHQ-2, RT primer: TTG TGCATA GTC GCT GCT TGA T. Measurement of OCT4 in single hESCs by microflu-idic RT-qPCR without reverse transcriptase showed no amplification after 40 cyclesof PCR.BHQ-Plus probes with enhanced duplex stabilization (Biosearch Technologies Inc)were used for SNV detection to allow for shorter sequence lengths and increased speci-ficity. The SNV location for the SP1 locus was selected from Table 2 in Shah et al.[32].Two hundred bp flanking this location on the hg18 sequence were used for assay de-sign using Primer3. The resulting primer and probe sequences are as follows (theSNV is in bold). SP1 Mutant Probe: FAM-AGGCCAGCAAAAACAAGG-BHQ-1, 5’ Modification: FAM, 3’ Modification: BHQ-1 Plus, Tm = 62.7 ◦C. SP1 WTprobe: Cal Fluor-CAGGCCAGCAAAAAGAA-BHQ-1, 5’ Modification: CAL FluorOrange 560, 3’ Modification: BHQ-1 plus, Tm = 62.1 ◦C. SP1 Forward Primer:CCAGACATCTGGAGGCTCATTG, Tm = 65.8 ◦C. SP1 Reverse Primer: TGAAC-TAGCTGAGGCTGGATA, Tm = 66.0 ◦C.Control experiments without reverse transcriptase showed positive amplification.27FBGB aythoxsTherefore the measurement of SP1 mutant and wilde-type abundance in single cellsby RT-qPCR does not discriminate between mature mRNA transcript and genomicDNA.GCHCL Imvge VnvlysisFluorescence images of the entire device were taken in at least two different colors(one passive reference dye and one or more reporter dyes) after each PCR cycle andwere analyzed using custom scripts written in MATLAB (MathWorks) to generatereal-time amplification curves. Reaction chambers were segmented from the rest ofthe image using the first image of the passive reference dye. The image was manuallyrotated so that all of the reaction chambers were square with the edges of the im-age. Next, the average image intensities across each row and column were calculatedand a threshold was manually set to differentiate bright areas from background. Re-gions containing both bright rows and bright columns were assigned to the reactionchambers.All subsequent images were automatically aligned to this initial image by minimiz-ing the absolute distance between the average row and column intensities of the initialimage, and the one being analyzed. For each image, the intensities of the reporterand passive dyes were recorded for each reaction chamber. Real time amplificationcurves were generated by normalizing the intensity of each reporter dye to that ofthe passive dye. Linear components were removed from these curves by fitting theequation of a line to the pre-exponential region and extrapolating and subtractingthe result from the entire curve. The threshold for determining CT values was auto-matically determined as the median normalized fluorescence value at the maximumsecond derivative of all amplification curves.GCHCM mgcV FIhHCells grown on LABTEK chambered cover-glass were washed with PBS, fixed in 4%formaldehyde for 10 min at room temperature and permeabilized in 70% EtOH at 40◦C overnight. The next day cells were rinsed with wash buffer (15% Formamide in2× SSC) and then hybridized with the appropriate dilution of mRNA FISH probesspecific to OCT4 (see table) in hybridization solution (dextran sulfate, Yeast tRNA,NEB, BSA, 15% Formamide in 2 SSC) overnight at 30 ◦C. The next morning the28FBGB aythoxsOCT4 hybridization solution was aspirated and cells were sequentially rinsed andincubated with wash buffer at 30 ◦C for 30 minutes then washed with 2× SSC. Onedrop (25 µL) of Slowfade GOLD antifade reagent with DAPI was then added to thecells, covered immediately with a cover slip, and imaged. Stacks of 32-64 mRNAhybridization images (spaced by 0.5 µm) were acquired for each cell using a LeicaDMI 6000B inverted microscope with a 100 objective (N.A. 1.3) in DAPI and TexasRed filter spectra.Fluorescent spots corresponding to individual mRNA molecules in each imagestack were evaluated manually since automatic thresholding using previously reportedalgorithms were found to be unreliable. Difficulty in automating this process was at-tributed to inconsistent signal to noise using reported protocols and may be related tothe thickness of hESC cells (∼15 µm). In addition, manual intervention was neededto ascertain the boundaries of adjacent cells. To optimize the signal to noise we sys-tematically varied the probe concentration, incubation time, incubation temperatureas well as the formamide concentration in the hybridization buffer solution.GCHCN Cell CultureK562 cells were cultured in Dulbecos Modified Eagle Medium (DMEM) (Gibco) sup-plemented with 10% fetal bovine serum (FBS) (Gibco). Purified RNA was extractedfrom K562 cells using RNA MiniPrep (Qiagen, USA).CA1S hESCs[111, 112] were propagated in mTeSR[113] basal medium (STEM-CELL Technologies, Inc., Vancouver, BC, Canada), additionally supplemented withantibiotic-antimycotic (100 U/mL penicillin, 100 mg/mL streptomycin and 0.25 mg/mLamphotericin B) (Invitrogen, Carlsbad, CA, USA). Upon passaging, hESCs werewashed with phosphate-buffered saline (PBS) prior to incubating with TrypLE Ex-press (Invitrogen, Carlsbad, CA, USA) at 37 ◦C for 10 minutes to detach singlehESCs from 4-8 day-old cultures depending on confluency. TrypLE Express was neu-tralized with mTeSR supplemented with antibiotic-antimycotic and suspensions werethen transferred into new tissue culture dishes containing a precoated layer of 1:30diluted Matrigel (Becton Dickinson, San Jose, CA, USA) and mTeSR supplementedwith antibiotic-antimycotic. For differentiation, mTeSR was replaced with Dulbeccosmodified eagle medium with 10% fetal bovine serum (FBS) 1 day after plating cells.When harvesting hESCs for qRT-PCR, cells were incubated with TrypLE Express(Invitrogen, Carlsbad, CA, USA) at 37 ◦C for 20 minutes in order to produce a more29FBGB aythoxsuniform single cell suspension from 4-8 day-old cultures.Cryo-vials of primary cells isolated from a lobular breast cancer metastasis wereprovided by the BC Cancer Agency in accordance with ethical guidelines of theUniversity of British Columbia. To increase viability, cells were transferred to freshculture medium and incubated for 2 days before analyzing in the microfluidic device.GCHCFE irvnsfer Exienxy bevsurementsA solution containing 10 µM FAM-labeled 40-mer poly-A oligonucleotides (IDT,USA), 0.1% Tween 20, and ROX passive reference dye (from CellsDirect kit, In-vitrogen, P/N 54880) diluted 100× was loaded into the cell capture chambers andsequentially pushed into the 10 nL and 50 nL chambers with water containing 0.1%Tween 20, and ROX reference dye diluted 100×. Fluorescence images acquired ofFAM and ROX were used to measure the transfer of oligonucleotides from one cham-ber to the next. The transfer efficiency for each chamber was calculated as (InitialSignal Final Signal)/(Initial Signal), where Signal = (FAM Intensity FAM Back-ground)/(ROX Intensity ROX Background). A conservative estimate of the lowerbound of transfer efficiency was taken to be one standard deviation from the meanmeasurement of transfer efficiency.GCHCFF Cell Cvpture bevsurementsA custom microfluidic device with a linear array of cell trap geometries was fab-ricated using protocols described above. The device was mounted on an invertedmicroscope (Leica DM IRE2) and imaged in bright field using a CCD camera (Hama-matsu ORCA-ER). The device was primed with 0.05% bovine serum albumen (BSA)(Gibco) in phosphate-buffered saline (PBS) (Gibco). Prior to loading in the device,cells were washed twice in fresh culture media (Dulbecos Modified Eagle Medium(DMEM) (Gibco) supplemented with 10% Fetal Bovine Serum (Gibco)). After thefinal wash cells were resuspended to be at a concentration of 1 million per mL. In-put sample viability was measured with the Cedex Automated Cell Counter (Rocheinnovatis AG).To measure the capture efficiency, cells were pumped through the array using adownstream microfluidic peristaltic pump at a rate of approximately 1 nL/secondand the number of cells that bypassed each trap before a successful trapping event30FBGB aythoxswas recorded. These counts were fit using a maximum-likelihood estimator for ageometric distribution with the fitdistr function (MASS package version 7.3-6) in R(version 2.11.1). Efficiencies are reported as the probability of a successful capturefor each cell.To measure cell viability after loading, cells were loaded into the array usingpressure driven flow as described above until high trap occupancy was observed.0.2% Trypan Blue (Gibco) in PBS was then flowed over the trapped cells. Viabilitywas calculated as the number of unstained cells divided by the total number of cells.Cell diameter was measured from Cedex images and images of cells trapped inthe microfluidic device using ImageJ (version 1.43u). A two sample t-test was usedto test the hypothesis that the resulting size distributions were significantly different.The assumption of equal variance was tested using an F test. For optimized celltrap geometries the cell trapping efficiency was improved to 87% by bringing the cupwithin one cell diameter of the focuser and by including a small bypass shunt throughthe cup, similar to the cup geometry presented in Skelley et al.[81].GCHCFG bifling wy DiusionMixing of solutions by diffusion was characterized in the microfluidic device by loadingfluorescently labeled 40 base poly-A oligonucleotides into the 10 nL chambers, andpushing the contents of the chamber into the adjacent 50 nL chambers. Time-lapseimaging was used to measure the evolution of the distribution of fluorescently labeledoligonuceotides in the PCR chambers over time (Figure 2.1). The standard deviationof the pixel intensities in each chamber through time was used as a metric of mixing.The resulting curves of all analyzed chambers (N = 200) were each fit to a decayingexponential using least squares regression to determine the characteristic mixing timeconstant. This resulted in a mean mixing time of 15.2 ± 1 minutes.Using the Stokes-Einstein relation and assuming a random coil we estimate thediffusion constant of a 40 base oligonucleotide to be:D =KBT6Rh; (2.1)where KBT is the thermal energy (4.1 pN·nm),  is the fluid viscosity (∼ 0.001kg/m·s), and Rh is the coil hydrodynamic radius (10). The hydrodynamic radius is31FBGB aythoxs10 20 30 40 50 60 70 80 900. [min]σ  DataAe   + Cabcdefghijklmabcdefghi j k l m-t/τR² = 0.9999A = 0.12τ = 15.9 minC = 0.014 Figure 2.1: Mixing by diffusion. Plot shows the standard deviation of pixel intensityvalues for a chamber as a function of time following the transfer of a solution offluorescently labelled 40 base pair poly-A oligonucleotide from the RT chamber (10nL) to the PCR chamber (50 nL) by flushing with buffer. An exponential fit tothe data to each of 200 chambers yields a mean mixing time constant of 15.2 ± 1.0minutes. A representative time-lapse series of images from one chamber is shown onthe right.proportional to the radius of gyration Rg, and is given byRh ≈ 0:5Rg ≈ 0:5(LpL3) 12 ; (2.2)where L is the contour length of single stranded DNA (40 bases× 4.3 Angstroms/base)and p is the persistence length (∼ 40 Angstroms)[114]. This yields a diffusion valueof approximately 9.0×10−11 m2s−1, which is comparable to the diffusion constantof polymerase, the largest molecule in the PCR mix. Since the template solutionconstitutes only 1/5 of the final PCR reaction it must diffuse the longest distanceto equilibrate across the chamber. Therefore, the measured diffusion time of 15.2minutes represents an upper bound to the time constant for complete mixing of allcomponents.32FB4B fysults unx DiswussionGCI gesults vny DisxussionGCICF Devixe DesignAn integrated microfluidic device that performs 300 parallel RT-qPCR assays andexecutes all steps of single-cell capture, lysis, reverse transcription, and qPCR isshown in Figure 1A. To facilitate the precise comparison of different samples and celltypes, our prototype consists of 6 independent sample-loading lanes, each containing50 cell-processing units. We resolved previously limiting technical pitfalls by theinclusion of design elements to 1) allow for efficient distribution of single cells withoutmechanical damage, 2) minimize background signal arising from free RNA or celldebris in the medium, and 3) avoid reaction inhibition by cell lysates in nL volumes.In order to reduce device complexity and obviate the need for RNA purification,we optimized our device to be compatible with commercially available assays that useone-pot RT-qPCR protocols requiring only the sequential addition of reagents intoa single reaction vessel. Each cell-processing unit consists of a compound chamber,formed by a cell capture chamber connected sequentially to two larger chambers forRT and qPCR (Figure 2.2B). This simple fluidic architecture allows the implementa-tion of either heat lysis followed by two-step RT-qPCR (Figure 2.2D-I), or chemicalFigure 2.2 (following pugy): Design and operation of the microfluidic device forsingle cell gene expression analysis. (A) Schematic of microfluidic device. Scale bar:4 mm. The device features 6 sample input channels, each divided into 50 compoundreaction chambers for a total of 300 RT-qPCR reactions using approximately 20 µL ofreagents. Rectangular box indicates the region depicted in B. (B) Optical micrographof array unit. For visualization, the fluid paths and control channels have been loadedwith blue and red dyes, respectively. Each unit consists of (i) a reagent injection line,(ii) a 0.6 nL cell capture chamber with integrated cell traps, (iii) a 10 nL RT chamber,and (iv) a 50 nL PCR chamber. Scale bar: 400 µm. (C) Optical micrograph of twocell capture chambers with trapped single cells indicated by black arrows. Each trapincludes upstream deflectors to direct cells into the capture region. Scale bar: 400µm. (D-I) Device operation. (D) A single cell suspension is injected into the device.(E) Cell traps isolate single cells from the fluid stream and permit washing of cellsto remove extracellular RNA. (F) Actuation of pneumatic valves results in single cellisolation prior to heat lysis. (G) Injection of reagent (green) for reverse transcription(RT) reaction (10 nL). (H) Reagent injection line is flushed with subsequent reagent(blue) for PCR. (I) Reagent for qPCR (blue) is combined with RT product in 50 nLqPCR chamber. Scale bar for D-I: 400 µm.33FB4B fysults unx DiswussionACell Suspension Wash Buffer RT Mix PCR Mix Closed Valve Open ValveEmptyBiiiiiiviCCell loading and captureDIECell washGRT loading and mixingFCell isolation and heat lysisHPCR reagent primingPCR loading and mixing34FB4B fysults unx Diswussionlysis followed by one-step RT-qPCR. A detailed description of device operation foreach of these protocols is provided in the methods section. All lanes are connected toa common feed channel which, following the completion of each reaction step, is usedto inject the next reaction master mix through the upstream chambers, thereby dilut-ing the intermediate product (cell lysate or cDNA) and assembling the next reactionmixture. This parallelization of reaction assembly in a microfluidic format ensuresequal timing of all reaction steps and greatly reduces technical variability associatedwith pipetting and mixing steps in µL volumes. Fluorescence measurements wereperformed to ensure the efficient and reproducible transfer of reactants at each step,showing that losses in sample transfer are below 5%. To minimize device expense andcomplexity, temperature control and fluorescence detection were performed using pe-ripheral hardware including a CCD detector mounted above a flatbed thermocyclerplate.Figure 2.3 (following pugy): Precision and sensitivity of microfluidic RT-qPCR. (A)Fluorescence image of entire device showing 300 reactions in 6 lanes. Image is takenafter 40 cycles of PCR from dilution series of purified total RNA from K562 cells.From left to right the samples are 40 pg/chamber, 5 pg/chamber, 625 fg/chamber,78 fg/chamber, 10 fg/chamber, and no-template control (NTC). Single molecule am-plification at limiting dilution results in a digital amplification pattern for 10 fg and78 fg lanes. No amplification is observed in NTC lane (N = 50). (B) 300 real timeamplification curves generated from processing sequences of images similar to (A).The threshold for determining CT values is indicated by the dashed line. (C) On-chip (black) and off-chip (blue) RT-qPCR for GAPDH from a 8× serial dilutionof purified total RNA shows improved sensitivity in nL volume reactions. In themicrofluidic system, CT values for the 10 fg sample correspond to single moleculeamplifications detected in 19 of 50 chambers. The mean and standard deviation fromsingle cell measurements is shown in green for both on and off-chip analysis. CTvalues obtained on chip correspond to a mean of 20 pg of RNA per cell. Off-chipmeasurements of single K562 cells washed twice in PBS and isolated by glass cap-illary exhibit artificially increased levels due to residual signal from debris and freeRNA in the supernatant (red). Cells were transferred in approximately 2 µL of su-pernatant, which was measured to contain ∼20 pg of extracellular RNA. Error barsrepresent standard deviation of measured CT values for all amplified reactions. (D)Real-time amplification curves of GAPDH in K562 cell lysate dilutions. Inhibitionof RT-PCR occurs at cell lysate concentrations beyond 10 cell equivalents per 50 nLreaction. (E) Measured CT values for GAPDH in dilution series of cell lysate. Noinhibition occurs for single cell lysates.35FB4B fysults unx DiswussionA0 10 20 30 40 5000.Δ RN  40 pg5 pg0.625 pg0.0781 pg0.0098 pgNTCB −8 −6 −4 −2 0 2 4 6 81520253035404550 y = −0.9878x + 24.46 2R  = 0.9997Single Cells20 μL tubeMicrofluidicSupernatanty = −1.35x + 38.23 2R  = 1.00CCtlog (RNA) [pg]2 0 10 20 30 40 5000. 40 cells10 cells2.5 cells0.625 cells0.1563 cellsNTCDΔ RN−2 −1 0 1 2 3 4 516182022log2(Cell Equivalent Lysate)Cty = −0.8962x + 20.24 R2 = 0.9983E36FB4B fysults unx DiswussionWe designed our chamber volumes to ensure sufficient dilution between each pro-cessing step to avoid reaction inhibition while at the same time maintaining hightemplate concentrations and assay sensitivity. Initial attempts to perform RT-qPCRin low nL volumes were found to produce highly variable results, including nonspe-cific amplification and inconsistent detection of abundant transcripts[98]. Cell lysatedilutions showed that reaction inhibition becomes significant at concentrations inexcess of 0.2 cells/nL, or 10 cells per 50 nL reaction (Figure 2.3D). On the otherhand, RT-qPCR measurement noise has been shown to become the dominant sourceof variability when starting at concentrations below 1 copy per 100 nL[58], illustrat-ing that minimizing reaction volumes is critical for precise measurements on limitedtemplate. Finally, experiments in tubes were performed to determine that a dilu-tion ratio of at least 5:1 (PCR mix:RT product) is optimum for PCR efficiency. Wetherefore designed our combined reactors to have an aggregate total volume of 60.6nL, consisting of a 0.6 nL cell capture chamber, a 10 nL RT chamber, and a 50 nLqPCR chamber. These volumes allow for the reliable amplification of single molecules(Figure 2.3A), and result in a final template concentration of 330 ng/mL when start-ing from a single cell equivalent of RNA (20 pg). The use of larger volume RT andPCR chambers has the added advantage of reducing their surface to volume ratio,thereby minimizing reagent evaporation through the gas permeable device material(polydimethylsiloxane).Another critical step towards integration was to efficiently distribute single cellsinto each location on the array without mechanical damage. To achieve reproducibleand deterministic loading of single cells into each array element we engineered ahydrodynamic single cell trap within each capture chamber. Cell traps consisting ofa single cup structure[71] were found to be highly inefficient, capturing less than 0.1%of cells passing in close proximity to the center of the channel structure. To improvecapture efficiency, we incorporated upstream deflectors, located 22.5 µm from thetrap, to focus cells into the central streamlines where capture is most efficient (Figure2.4C). Using these structures we were able to achieve high single cell occupancy ofarray locations (Figure 2.4A-B). Over 8 separate experiments, a loading protocol of∼60 seconds (106 cells/mL, 20 nL/s per lane) resulted in the successful isolation ofsingle cells in 1518/1700 chambers (89.3%), with a cell capture efficiency of 5.0±0.5%.Staining with Trypan Blue™ was used to assess the viability of cells after loading andwas determined to be equivalent to the viability of the input sample (97.4% viability37FB4B fysults unx Diswussion1718192021222324252627Ct0 1000 2000 3000 4000051015202530Copy NumberNumber of Single Cells0 1 2 3 4 51820222426CtNumber of Cells0 1 2 3 4 5020406080100Percent1628ACBFigure 2.4: Single cell loading and transcript measurements. (A) The locations of cellsin each chamber along all 6 lanes of a device, as determined by brightfield microscopy,are represented as white circles and overlaid on a heat map of CT values obtainedfrom RT-qPCR measurements of GAPDH in K562 cells. Red circles indicate NTC.(B) Scatter plot showing CT measurements for experiment shown in (A). Histogram(inset) shows 93.2% single cell occupancy. (C) Distribution of the number of GAPDHtranscripts measured in single K562 cells (N=233).38FB4B fysults unx Diswussion10 15 20 25 30 35051015200102030Percent of PopulationDiameter [μm]AB10 15 20 25 30 35Diameter [μm]Percent of PopulationFigure 2.5: Histograms showing the size distribution of cells in original sample asmeasured by Cedex (A) are consistent with the size distribution of cells isolated bymicrofluidic traps (B). Under the assumption of spherical cell shape the distributionof diameters of trapped cells corresponds to a mean volume of 4.2 pL with a standarddeviation of 2.0 pL.39FB4B fysults unx Diswussionvs. input 96.8%). Finally, measurements of the distribution of cell diameters priorto and after loading indicated that cell trapping did not introduce significant bias(p=0.67, two sample t-test) in selecting cells of different sizes (Figure 2.5). This celltrap geometry and loading protocol were used in all subsequent qPCR measurementspresented below. Further improvement of trap and deflector geometries were foundto achieve fill factors of >99% (100 single cells captured out of 100 traps analyzed)and cell capture efficiencies of 87.0±4.5%, with cell viability again matching the inputsample (>98%) and not significantly biasing cell sizes (p=0.35, two-sample t-test),making this method applicable to the analysis of limited quantity samples such asrare stem cells or clinical samples.The immobilization of cells in traps was also used for on-chip washing of cellsprior to lysis to remove free RNA, cellular debris, and untrapped cells that wouldotherwise give rise to background signal or result in low single cell occupancy (Figure2.7A-B). The efficiency of chamber washing, determined by loading purified RNAtemplate (36.5 ng/µL), followed by washing and RT-qPCR analysis, was >99.99%(1.1×104 copies measured without wash, 0 copies detected after washing) (FigureS2C). In addition, RT-qPCR measurements testing different cell loading and washingprotocols demonstrated that on-chip washing allows for loading directly from culturemedium with low background as compared to off-chip wash steps followed by analysisin µL volumes (Figure 2.3C). Importantly, on-chip washing allows for lysis withinseconds of washing, thereby minimizing spurious transcriptional responses that mayarise from sequential medium exchange and spin steps.GCICG kvliyvtion of Integrvtey hingle Cell giBqeCgWe first tested the sensitivity and precision of RT-qPCR in our device by perform-ing measurements of GAPDH expression over an 8-fold dilution series of total RNA,ranging from 40 pg (∼2 cell equivalents) to 10 fg (∼1/2000 cell equivalents). RNAwas purified from K562 cells, a BCR-ABL positive human cell line derived from apatient with chronic myeloid leukemia (20) (Figure 2.3A-C). The efficiency of am-plification was determined over the four highest template concentrations (40 pg, 5pg, 625 fg, 78.125 fg) as the slope from a linear least squares fit of log2(C) vs. CT,and was found to be 0.988±0.055. The standard deviation of CT values was lessthan 0.15 at the three highest concentrations (s.d.=0.08, 0.10, 0.14 for the 40 pg,5 pg, and 625 fg samples respectively), indicating uniform amplification across the40FB4B fysults unx Diswussion22.8 21.6 28.226.7 22.5 22.0 22.422.9 23.3 21.922.7 22.2 22.3 22.1 22.421.2 22.3 23.1 21.1 21.321.9 23.4 22.8 21.2 21.320.7 21.3 22.9 23.2 22.121.9 23.4 22.6 23.1 22.122.2 23.6 23.3 22.3 25.723.3 23.3 23.2 21.9 22.022.2 22.7 22.7 21.2 22.322.1 23.4 22.1 22.8 22.422.5 23.4 22.8 23.3 22.623.1 22.5 23.022.9 22.5 21.6 23.7 23.122.8 23.8 23.0 22.5 22.222.0 22.9 22.8 22.1 21.822.3 22.4 22.0 21.5 22.521.5 21.5 21.4 22.4 22.621.9 23.0 22.6 22.8 22.621.4 23.6 20.3 22.7 21.321.3 22.6 22.6 22.1 22.122.5 23.4 24.8 23.2 22.019.6 21.7 22.2 22.6 21.122.7 23.3 21.8 22.1 22.422.0 22.5 21.622.2 22.3 21.0 21.922.2 21.1 23.2 21.722.4 22.5 21.3 23.2 22.922.4 22.6 22.4 23.4 21.322.4 22.4 21.8 22.322.5 24.1 22.5 23.2 21.622.5 23.9 22.1 22.9 21.421.6 22.5 21.5 22.0 22.521.8 23.4 23.0 20.1 22.222.4 23.0 23.6 22.5 23.222.9 23.0 23.7 23.6 22.123.2 22.7 23.1 23.322.2 23.2 22.7 22.8 22.322.5 23.5 22.1 22.3 22.623.1 22.2 22.7 22.023.6 22.6 22.0 22.7 23.223.0 22.3 22.3 22.6 23.122.4 23.3 23.4 23.222.7 22.2 22.1 25.8 27.923.2 23.1 22.5 22.822.1 20.6 22.2 22.3 22.322.3 23.4 22.0 23.622.2 21.2 21.8 23.3 23.122.9 22.9 23.5 22.8 23.422.423.320.723. BFigure 2.6: Single-cell miRNA measurements. (A) The locations of cells in eachchamber along all 6 lanes of a device, as determined by brightfield microscopy, arerepresented as white circles and overlaid on a heat map of CT values obtained fromRT-qPCR measurements of miR27a in K562 cells. Red circles indicate NTC. (B)Fluorescence image of entire device, corresponding to experiment in (A) after 30 PCRcycles. Cell corpses remain after heat lysis and are visible as punctuate fluorescentspots adjacent to reaction chambers.41FB4B fysults unx DiswussionSup. 1 2 3 4 5 6 7 8 91617181920212223242526CtNumber of Cells1 2 3 4 5 6 7 8 9010203040PercentNumber of Cells0 1 2 3 4 516182022242628CtNumber of Cells0 1 2 3 4 5050100PercentNumber of CellsA B0 5 10 15 20 25 30 35 40 45 50−Δ RNNo WashWashCFigure 2.7: On-chip cell washing. (A) Measurements of GAPDH in cells washedin PBS off-chip prior to injection into microfluidic device, without an on-chip washcontain background signal from template in supernatant. Without on-chip washing,untrapped cells remain in the capture chambers, resulting in fewer single cell measure-ments (histogram inlayed). Detection of residual RNA after washing is dramaticallyreduced by comparison to off-chip results (Figure 2.6) due to small volume processing.(B) On-chip washing was found to reduce the background signal from free RNA inthe supernatant, and dramatically increased the number of single cells analyzed. (C)Comparison of GAPDH measurements from loading purified RNA and washing, ornot washing, the cell capture chambers.42FB4B fysults unx Diswussion0 5 10 15 20 25 30 35 40 45 50−Δ RN  On−ChipOff−ChipOn Chip Off Chip19.819.92020.120.2CtFigure 2.8: Comparison of GAPDH measurements from K562 cell lysate with RTperformed in the microfluidic device or RT performed in tubes prior to qPCR in thedevice. Obtained CT values (inset) show no significant difference in efficiency.array and technical error of less than 10% in absolute concentration, near the limitof qPCR precision. The highest measurement variability was observed in the 78 fgsample, where shot noise (Poisson sampling noise) is most pronounced and accountsfor approximately 50% of the measurement variance. Template amounts below 625fg resulted in a digital pattern characteristic of single molecule amplification (49/50for 78 fg, 19/50 for 10 fg) and consistent with the expected occupancy of chambers asdetermined by a binomial distribution[62]. Based on the frequency of single moleculedetection in the 10 fg sample, we measured the average copy number of GAPDH tobe 979 ± 240 transcript copies per single cell equivalent (20 pg) (Figure 2.3). Thismeasurement is comparable to previous reports[79] and is in close agreement with anindependent estimate based on normalizing the dilution series to CT values obtainedfor single molecules (copies/20 pg = 0.5×copies/40 pg = 0.5×(1+efficiency)(CT(40pg) - CT(single molecule)) = 1407 ± 153 copies/20 pg). It should be noted that43FB4B fysults unx Diswussionthese estimates represent a lower bound since they do not account for RT efficiency;the RT efficiency of GAPDH has been previously estimated to be ∼50%[89] but isdependent on transcript secondary structure and assay design. A comparison of CTvalues obtained from on-chip qPCR from cDNA synthesized off-chip demonstratedthat on-chip RT efficiency is equal to that obtained off-chip when working from thesame RNA concentrations (Figure 2.8). Finally, comparison of the same dilution se-ries of RNA, assayed for GAPDH both on-chip and in tubes (20 µL volume) (Figure2.3C), showed that on-chip analysis provides improved sensitivity.1 2 3 4 5 6 7 8 9 1019202122232425Number of CellsCtN=56 N=22 N=36 N=7 N=10 N=7 N=2 N=1 N=1Figure 2.9: Measurement of miR16 in hESC cell aggregates demonstrates that thenumber of cells is reflected in corresponding cycle threshold (CT) values.We next evaluated the efficiency and reliability of on-chip cell processing by com-paring our GAPDH measurements of purified RNA to measurements performed di-rectly from single K562 cells (Figure 2.3C, Figure 2.4C). K562 cells were loadeddirectly from culture medium followed by washing and analysis using a chemical lysisand one-step RT-qPCR protocol (Cells Direct™, Invitrogen). Using a CT threshold of31.5, corresponding to the mean CT of a single molecule of GAPDH (CT = 30.5) plustwo standard deviations (s.d. = 0.5), we observed successful amplification in 100%of single cells (N = 233) (Figure 2.4A-B). Adjacent chambers that did not contain acell were clearly separated from single cell measurements with an average delta CT44FB4B fysults unx Diswussionvalue of 5.7 (5 empty chambers, 3 of which amplified) (Figure 2.4A-B). Consistentwith previous reports[40], we observed a log-normal distribution of GAPDH in singlecells with mean CT values of 20.3 (s.d. = 0.8) and an average of 1761 (s.d. = 648)copies per cell (Figure 2.4C). These expression levels are consistent with previous es-timates in single cells[79]. Additionally, the mean CT of 20.3 observed for single cellsmatches measurements of single cell equivalent lysate (CT = 20.2, Figure 2.3D). Us-ing digital PCR on cDNA prepared from K562 cell lysate, we measured an average of1229±72 GAPDH molecules per single cell equivalent. We conclude that the relativeefficiency of on-chip single cell lysis and mRNA extraction/accessibility is equal tothat achieved when working from RNA purified from large numbers of cells. Finally,as expected, RT-qPCR measurements from chambers loaded with more than one cellshow reduced variability and lower CT values (Figure 2.7A, Figure 2.9). Taken to-gether, these results establish the precise measurement of mRNA abundance withsingle molecule sensitivity and the dynamic range needed for single cell analysis.GCICH Vpplixvtion to bevsurement of hingle Cell migcVEflpressionWe next applied our technology to the study of single cell miRNA expression. miR-NAs are thought to provide a unique signature of cellular state and are central play-ers in orchestrating development and oncogenesis, making them a promising class ofbiomarker for single cell analysis[16, 20, 115]. Importantly, the short length of miR-NAs (∼22 nucleotides) makes them difficult to detect by hybridization approaches,so that RT-qPCR is the dominant quantification strategy. To demonstrate the ro-bustness and throughput of our technology, we performed a total of 1672 single cellmeasurements to examine single-cell variability in the expression of 9 miRNAs span-ning a wide range of abundance (>16000 copies per cell to K0.2 average copies percell). K562 cells were again chosen as a heterogeneous population for this study sincethey are known to exhibit mixed characteristics of erythrocytes, granulocytes, andmonocytes[116, 117]. We first measured the expression of miR-16, a highly expressedmicroRNA that is found in many tissue types[118] and has been suggested as a suit-able internal standard for normalization[60]. We found that miR-16 was log-normallydistributed across K562 cells, but with slightly lower expression and notably tighterregulation than GAPDH, having an average of 804 (s.d. = 261) copies per cell and45FB4B fysults unx DiswussionC t20 24 28 3201020304050Percent of PopulationOff-Chip K562K562hESC−2 0 2 4 6−2024r = 0.9932log   Digital PCR Copy Number10log   qPCR Copy Number10ND0123410log   Copy NumberN=σ=1100.352300.232280.212020.39740.47940.132350.172240.20miRNA: 181a 196a 27a 145 223 16 17-5p 92200a1640.19A B CDPercent of populationC tK562hESC20 24 28 ND0102090Figure 2.10: Single cell miRNA measurements. (A) Single cell measurements of miR-16 expression in K562 cells and hESCs. Measurements of single K562 cells isolatedusing a microcapillary and assayed in 20 µL volumes are shown for comparison oftechnical variability. The observed shift in mean CT values between on and off-chipmeasurements is due to lower template concentrations, and hence increased requiredPCR cycles, in the off-chip samples. (B) Differential expression of miR-223 betweenK562 cells and hESCs. Right-most bar indicates cells for which miR-223 was notdetected (ND). (C) Mean single cell miRNA copy numbers measured by RT-qPCRin the microfluidic device compared to digital PCR measurements from bulk celllysate. Error bars represent standard deviation of single cell measurements for eachmiRNA. (D) 2072 single cell measurements of the expression of 9 miRNA in K562cells. Reflected histograms represent the expression distributions for each miRNA.46FB4B fysults unx Diswussiona standard deviation of 30% (mean CT = 21.4, s.d. = 0.4). This strikingly lowvariability is within our estimates of cell volume differences (Figure 2.5). Matchedexperiments on single cells, isolated by micropipette into 20 µL volume tubes dis-played an increase in measurement variability to ∼90% (mean CT = 29.5, s.d. =0.9), demonstrating the improved precision of parallel microfluidic cell processing innL volumes (Figure 2.10A). Microliter volume experiments also showed a pronouncedincrease in measured CT values that results from the low concentration of templateand the large number of required PCR cycles.To demonstrate the utility of our device for measuring differential expression insingle cells we next measured the expression of miR-223, a miRNA implicated inmyeloid differentiation[18, 117]. In contrast to miR-16, K562 cell miR-223 expressionwas found to be highly variable (mean CT = 22.2, s.d. = 1.6, copy number = 513,s.d. = 406) and was not log-normally distributed (Figure 2.10B), consistent withthe known functional heterogeneity of K562 cells. These measurements highlight theutility of single cell miRNA expression analysis for assessing the heterogeneity of cellpopulations and for identifying miRNAs that are useful biomarkers of cellular state.To further explore this possibility we measured the expression of an additional 7 miR-NAs (9 total) and plotted the patterns of single cell expression in K562 populations(Figure 2.10C). Following the procedure described above, we used single molecule CTvalues, obtained by digital PCR, to translate measured CT values to absolute copynumber. Assuming 100% efficient amplification, we observed that the copy number,calculated as 2(CT(single cell) CT(single molecule)), was well correlated (coefficientof 0.9932) with the average copy number obtained by digital PCR of cDNA preparedfrom bulk lysates (Figure 2.10D). Single cell measurements revealed distinct patternsof miRNA expression, with miR-16, miR-92, and miR-17-5p each exhibiting unimodaland tightly regulated distributions, while miR-223, miR-196a, and miR-145 showedmulti-modal distributions and a high level of cellular heterogeneity. Notably, for thelowest abundance miRNA, miR-200a, we detected expression in only a small fractionof cells and at levels below ∼5 copies per cell. The average miR-200a copy numberover all cells was within factor of two of that obtained by digital PCR (0.2 copiesper cell). In contrast, miR-92 was found to be the most abundant miRNA and waspresent at approximately 60,000 copies per cell. These measurements establishedmiRNA quantification in single cells with a dynamic range of greater than 104 andat single molecule sensitivity.47FB4B fysults unx DiswussionFinally, to illustrate the utility of single cell measurements in precisely assessingdifferences in both the average expression and the heterogeneity between two differentcell populations, the expression levels of miR-16 and miR-223 in K562 cells werecompared to those in CA1S cells[111, 112], a human embryonic stem cell line (hESC).Although miR-16 was found to be expressed in hESC at similar levels to K562 (∆CT= 0.6), we observed approximately a two-fold greater variability in expression (meanCT = 22.0, s.d. = 0.7) (Figure 2.10A). In contrast, when compared to K562, singleCA1S cell measurements of miR-223 showed strong down-regulation, with miR-223detected in only 3.6% of cells. The absence of significant miR-223 expression in hESCis expected due to the role of miR-223 as a differentiation-specific miRNA[18, 117].GCICI CoBregulvtion of migBFI5 vny dCiI in hingle CellsThe measurement of multiple transcripts in individual cells allows for quantitativemeasurements of gene co-regulation that would otherwise be masked by cellularheterogeneity[90]. To demonstrate this capability we designed an optically multi-plexed assay to study the co-regulation of miR-145 and OCT4, a known target ofmiR-145[119], during the differentiation of hESCs (Figure 2.11A-C). A total of 1094single cell measurements were performed at 0, 4, 6, and 8 days of differentiation.Cell distributions at each time point were used to map out the evolution of thesetranscripts and showed that average miR-145 levels increased approximately 20 fold(copy numbers: D0: mean = 18.9, s.d. = 25.5, D8: mean = 380.3, s.d. = 259.4)over 8 days. Increases in miR-145 were accompanied by progressive down-regulationof OCT4, ultimately reaching an average of 30-fold suppression (copy numbers: D0:mean = 755.7, s.d. = 306.4, D8: mean = 27.8, s.d. = 124.5) after 8 days (indepen-dently verified by mRNA-FISH) (Figure 2.12). Notably, single cell analysis at day 6showed a bimodal distribution in both OCT4 and miR-145, revealing a transition ofcellular state[119] that likely reflects the spontaneous differentiation of a subpopula-tion of cells. The observed single cell dynamics of miR-145 and OCT4 co-regulationare not apparent in population measurements, highlighting the use of scalable singlecell transcriptional analysis in correlating molecular signatures to cellular decisionmaking[90].48FB4B fysults unx Diswussion Day 0Day 4Day 6Day 8ND 0 1 2 3ND01234log   Oct4 Copy Number10log   miRNA145 Copy Number102224262830ND2224262830NDMutant Expression (C  )t Wildtype Expression (C ) t  94.6%72.6%5.4%7.7%0.0%0.9%0.0%18.8%K562 CellsPrimary CellsA BC DFigure 2.11: Optical multiplexing of single cell RT-qPCR. (A) Multiplexed analysisof the co-expression of OCT4 and miR145 in differentiating hESC. Points are color-coded to represent single cell measurements (N = 547) for each time point. Crossesrepresent population mean copy number. (B,C) Histograms showing the distributionof each transcript are projected on the axes with the mean copy number indicated bya dashed line. (D) Co-expression measurements of SP1 wild-type and SNV mutanttranscripts in primary cells isolated from a lobular breast cancer sample. Mutant SP1is detected in 23 of 117 primary cells, and undetected in K562 cells (N=37).49FB4B fysults unx DiswussionBAFigure 2.12: mRNA-FISH of OCT4 (red) counterstained with DAPI (Blue) in CA1Scells (A) Representative image of mRNA-FISH of OCT4 in a CA1S cell after 7 days ofFBS differentiation. Estimate of average copy number of OCT4 mRNA as determinedby manual inspection of image stacks is 42 (s.d. = 41, N=6). (B) Representativeimage of undifferentiated CA1S cells. Estimate of average copy number as determinedby manual inspection of image stacks is 988 (s.d. = 368, N = 6). Scale bar = 10 µm.GCIC5 hck Detextion in erimvry CellsFinally, to establish the specificity of our method we used multiplexed measurementsof mRNA single nucleotide variants (SNV) to assess the genomic heterogeneity withina primary tumor sample. A total of 117 single cells isolated from a plural effusionof a metastatic breast cancer were assayed for the expression of a SNV mutant ofthe transcription factor SP1, previously identified by deep sequencing[32] (Figure2.11D). Primers were designed using sequences flanking the SNV location and do notdiscriminate between the genomic DNA and mRNA transcript. Of the 117 primarycells analyzed, 22 (18.8%) were heterozygous for the mutant and wildtype allele,85 (72.6%) were homozygous wildtype, 1 (0.9%) was homozygous mutant and thetranscripts were undetected in 9 (7.7%). We did not detect the SP1 mutation in37 control K562 cells and failed to detect the wild-type transcript in only 2 of thesecells. In the absence of copy number alterations in the primary sample, these observedfrequencies would suggest a mutant to wild-type SP1 ratio of 11.2% (18.8×1 + 0.9×2= 20.6 mutant to 18.8×1 + 72.6×2 = 164 wild-type). However, using digital PCRon purified DNA from the primary sample, we found the ratio of mutant to wild-type50FBIB ConwlusionSP1 alleles to be 18.7 ± 2.3%, in agreement with the previously reported ratio of21.9%, obtained by deep sequencing[32]. The lower frequency of cells expressing themutant SP1 allele may be due to allelic expression bias or an amplification of the SP1mutant allele, both of which are supported by Shah et al.[32]. Regardless, given thatthe frequency of tumor cells within the original sample was approximately 89%[32],both DNA molecule counting and single cell RNA expression measurements showthat the metastasis of this tumor is derived from multiple cancer cell lineages.GC5 ConxlusionHere we have demonstrated the first implementation of scalable and quantitativesingle cell gene expression measurements on an integrated microfluidic system. Thepresented device performs 300 high-precision single cell RT-qPCR measurements perrun, surpassing previous microfluidic systems by a factor of ∼100 in throughput.Further scaling the throughput to over 1000 measurements on a device with an areaof one square inch is straightforward as each array element occupies an area of 0.6mm2. In terms of performance, we have established a dynamic range of at least 104,measurement precision of better than 10%, single molecule sensitivity, and specificitycapable of discriminating the relative abundance of alleles differing by a single nu-cleotide. Compared to tube-based single cell RT-qPCR, microfluidic processing pro-vides improved reproducibility, precision, and sensitivity, all of which may be criticalin identifying subtle differences in cell populations. Nanoliter volume also results in a1000-fold reduction in reagent consumption, thereby enabling cost effective analysisof large numbers of single cells.In over 3300 single cell experiments, using adherent and suspension cell lines aswell as clinical samples, we have shown that microfluidic RT-qPCR is well-suited tothe quantitative analysis of miRNA expression and SNV detection, both of which aredifficult or inaccessible by alternative hybridization methods. Notably, our device al-lowed for precise comparison of the distributions of GAPDH and miR-16 expression.miR-16 was found to be exquisitely regulated in K562 cells, a finding that is strikinggiven the known functional heterogeneity of this population and the high observedvariability in the expression of other measured miRNAs. We postulate that higher ob-served variability of GAPDH expression reflects the fundamentally stochastic processof transcriptional bursts followed by mRNA degradation. Incorporation of miRNA51FBIB Conwlusioninto the RISC complex is known to provide enhanced stability so that miRNA areinherently less subject to temporal fluctuations; miRNA are thus particularly suitedas biomarkers for assessing single cell state and population heterogeneity. We antici-pate that scalable and precise single cell miRNA analysis will become an invaluabletool in stratifying populations of mixed differentiation state[20].Here we have established the critical element of combining all single-cell process-ing steps into an integrated platform. This functionality provides a solid foundationupon which increasingly advanced microfluidic single cell transcription analysis maybe built. We anticipate that more complex fluid routing[120], to distribute cell con-tents across multiple chambers, will allow for the multiplexed measurements of tensof targets across hundreds of cells, and for combining this technology with singlemolecule detection by digital PCR. Alternatively, the microfluidic system describedhere could be used for single cell processing and pre-amplification, with recovered re-action products analyzed by high-throughput microfluidic qPCR or sequencing. Wecontend that the simplicity of device operation will soon allow for the robust and au-tomated implementation of single cell RT-qPCR, leading to its widespread adoptionin research applications and opening the prospect of diagnostic tests based on singlecell analysis.52FBIB ConwlusionTable 2.3: mRNA FISH ProbesProbe Sequence (5’ - 3’) Probe Name Position GC content (%)tgaaatgagggcttgcgaag OCT4 1 2 50aaatccgaagccaggtgtcc OCT4 2 61 55atcacctccaccacctggag OCT4 3 95 60aggtccgaggatcaacccag OCT4 4 138 60aggagggccttggaagctta OCT4 5 161 55aatcccccacacctcagagc OCT4 6 215 60atccccccacagaactcata OCT4 7 253 50actagccccactccaacctg OCT4 8 289 60tcaggctgagaggtctccaa OCT4 9 322 55agttgctctccaccccgact OCT4 10 354 60ttctccttctccagcttcac OCT4 11 418 50ctcctccgggttttgctcca OCT4 12 440 60ttctgcagagctttgatgtc OCT4 13 466 45cttggcaaattgctcgagtt OCT4 14 488 45tgatcctcttctgcttcagg OCT4 15 510 50atcggcctgtgtatatccca OCT4 16 533 50aaatagaacccccagggtga OCT4 17 560 50tcgtttggctgaataccttc OCT4 18 582 45taagctgcagagcctcaaag OCT4 19 612 50gcagcttacacatgttcttg OCT4 20 636 45tccacccacttctgcagcaa OCT4 21 661 55gattttcattgttgtcagct OCT4 22 684 35tctgctttgcatatctcctg OCT4 23 706 45actggttcgctttctctttc OCT4 24 743 45ttgcctctcactcggttctc OCT4 25 766 55ctgcaggaacaaattctcca OCT4 26 788 45atctgctgcagtgtgggttt OCT4 27 814 50atccttctcgagcccaagct OCT4 28 851 55ttacagaaccacactcggac OCT4 29 874 50tagtcgctgcttgatcgctt OCT4 30 910 50ctcaaaatcctctcgttgtg OCT4 31 932 45ctgagaaaggagacccagca OCT4 32 954 55agaggaaaggacactggtcc OCT4 33 976 55atagcctggggtaccaaaat OCT4 34 1010 45agtacagtgcagtgaagtga OCT4 35 1038 45ttccccctcagggaaaggga OCT4 36 1064 60tgacggagacagggggaaag OCT4 37 1086 60agtttgaatgcatgggagag OCT4 38 1116 45attcctagaagggcaggcac OCT4 39 1139 5553Chvpter HHighBihroughput bixrouiyixhingleBCell Digitvl eCg1HCF dverviefiHere we present an integrated microfluidic device for the high-throughput digital PCR(dPCR) analysis of single cells. This device allows for the parallel processing of singlecells and executes all steps of analysis including cell capture, washing, lysis, reversetranscription, and dPCR analysis. The cDNA from each single cell is distributed intoa dedicated dPCR array consisting of 1020 chambers, each having a volume of 25pL, using surface tension-based sample partitioning. The high density of this dPCRformat (118,900 chambers per cm2) allows the analysis of 200 single cells per run, fora total of 204,000 PCR reactions using a device footprint of 10 cm2. Experimentsusing RNA dilutions show this device achieves the shot-noise limited performancein quantifying single molecules, and with a dynamic range of 104. We performedover 1200 single cell measurements, demonstrating the use of this platform in theabsolute quantification of both high and low abundance mRNA transcripts, as wellas microRNA that are not easily measured using alternative hybridization methods.We further apply the specificity and sensitivity of single cell dPCR to performingmeasurements of RNA editing events in single cells. High-throughput dPCR providesa new tool in the arsenal of single cell analysis methods, with a unique combination ofspeed, precision, sensitivity and specificity. We anticipate this approach will enablenew studies where high-performance single cell measurements are essential, includingthe analysis of transcriptional noise, allelic imbalance, and RNA processing.1T version oy this vhtpter hts ueen puulishewM Twtm White, Kevin Heyries, Volin Woolin,Mivhtel itnInsuerghe, tnw Vtrl LA Htnsen, High-ghroughput Microuidic finglx-Cxll Digittl cCe,Tntlytivtl Vhemistry, 2CD3A54GBFB IntroxuwtionHCG IntroyuxtionCells are the fundamental unit of biology. Despite this the vast majority of geneexpression measurements have been performed using bulk samples of RNA extractedfrom large populations of cells having undefined composition and heterogeneity. Un-derlying the interpretation of such data is the assumption that all cells are similar,and that the ensemble average of many individuals accurately captures the biology.Unfortunately this assumption is often false. Significant cellular heterogeneity existsin most samples and is manifest at multiple levels including epigenetic[121, 122] andtranscriptional states[123], protein expression[124] and post-translational modifica-tions, growth characteristics[125], and functional responses[126]. Population mea-surements generally obscure this heterogeneity and muddy the biological interpre-tation: existing protocols for the isolation of rare stem cell populations typicallyprovide functional purities between 1% and 50%[10, 127, 128], resulting in significantand often overwhelming contamination from undefined subpopulations; the analysisof gene expression responses can be blurred by cellular asynchrony[129]; and manyfundamental biological questions, including the degree to which cells regulate mRNAexpression and processing, require measurements at the single cell level.The development of methods to measure and understand cellular variability hasreached the point of mission critical and stands to impact a wide array of fundamentaland applied fields ranging from immunology to regenerative medicine to microbiol-ogy. In response to this need, an expanding array of single cell genomics methodsare being advanced, each appropriate to different levels of inquiry. The coupling ofwhole transcriptome RNA amplification with high-throughput sequencing is now anestablished method for global analysis of single cell expression profiles[130]. Althoughthere still remain issues of representational bias, technical noise, and sequencing cost,the continued development of improved instrumentation, bioinformatics approaches,and optimized reagent kits[34] is likely to bring single cell WTA analysis into themainstream. RT-qPCR, either in conventional[40, 58] or microfluidic[90, 94] formats,is perhaps the most versatile method for single cell gene expression analysis[131].Although well-suited to identifying and monitoring cellular subpopulations using es-tablished panels of genes[20, 60, 132], RT-qPCR does not provide absolute measure-ments and has limited precision and specificity when working with low abundancetemplates. These limitations make RT-qPCR suboptimal for applications that re-quire high-performance measurements of a small number of targets, including studies55GBFB Introxuwtionof transcriptional noise or allelic imbalance, single cell genotyping, and the absolutequantitative analysis of low copy transcripts.For such demanding applications two single molecule counting methods haveemerged: mRNA fluorescence in situ hybridization (mRNA FISH)[23, 26] and digitalPCR (dPCR)[62]. mRNA FISH has the important advantage of preserving spatialinformation regarding the location of cells within a tissue and has been applied towhole organisms. However, this method requires a lengthy sample preparation pro-tocol, complex probe design, and the need for sophisticated image acquisition andanalysis steps. In addition, although mRNA FISH can in principal provide absolutetranscript numbers, in practice this is often limited by difficulty in resolving closelylocalized fluorescent spots, interference from background fluorescence, ambiguity indefining cellular boundaries, and a broad distribution of intensities from hybridiza-tion events. More fundamentally, the need for multiple hybridization probes makesmRNA FISH poorly suited to measurements of small RNA species or discriminationof transcripts with high sequence homology[26].The alternative approach, dPCR, uses compartmentalization of single moleculesat limiting dilution followed by PCR amplification and end-point detection to enableprecise and highly specific quantification of transcripts. Although this approach hasbeen used to make precise measurements of single cell transcription[62], throughput istypically restricted by the labour and cost of cell isolation and processing in conven-tional microliter volumes, as well as limitations in the throughput of dPCR analysis.Microfluidic systems can address these issues by providing economy of scale, automa-tion and parallelization, as well as increased reproducibility and sensitivity in smallvolume reactions[68]. We previously developed a microfluidic device that implementsintegrated RT-qPCR at a throughput of 300 single cells per run[129]. This device per-forms steps of cell lysis, RT, and PCR by sequentially transferring reagents throughthree chambers, ending with PCR amplification of each cell product in single 50 nLchambers. A related device, featuring additional cell processing chambers and sam-ple elution capabilities, has recently been released as a commercial product (FluidigmC1™). In separate work we have also developed a dPCR format that uses surface ten-sion partitioning to achieve planar dPCR densities up to 400,000 chambers/cm2[133].Here we combine the advantages of integrated single cell processing[129] and high-density dPCR[133] to enable high-throughput single cell dPCR. Our device is capableof processing 200 single cells per run, each of which is analyzed in an array of 1,02056GBGB aythoxsPCR reactions, each having a volume of 25 pL, for a total of 204,000 PCR reactionsper experiment. We establish the technical performance of this system using RNAdilutions and demonstrate its use in making absolute single cell measurements of theexpression of high- (GAPDH) and low-abundance (BCR-ABL) transcripts, as wellas the abundance of a mature microRNA (miR-16). Finally we apply our systemto the measurement and single nucleotide discrimination of RNA editing of EEF2Ktranscripts in single cells.HCH bethoysThe microfluidic device presented here consists of 200 identical modules (Figure 3.1A)divided into 4 linear arrays. Each array contains 50 modules (Figure 3.1B) with 3chambers connected in serial, followed by a dPCR array. This architecture allowsfor the implementation of a three-step protocol that requires sequential additions ofreagents without purification, followed by dPCR analysis of the resulting products.Examples of two such protocols, one for mRNA quantification and one for miRNAquantification, can be performed as shown in Figure 3.1C (see below for protocoldetails). Cells are first trapped in the cell capture chamber (0.8 nL) using hydrody-namic traps[129] and are then washed with fresh PBS. Reagents are then introducedthrough the cell capture chamber to fill the subsequent cell lysis chamber (10 nL)and the RT chamber (50 nL). Finally, half of the product from each RT chamberis loaded into an independent dPCR array having 1020 chambers, sufficient to ac-curately measure the mRNA and miRNA targets ranging from approximately 3 to5,000 copies per cell. This section describes device fabrication and operation, as wellas experimental protocol details.HCHCF Devixe FvwrixvtionMicrofluidic devices were fabricated using multilayer soft lithography[65], followinga similar procedure to those previously described[79, 89, 129]. The microfluidic de-vice was designed using in-house developed software to generate a digital drawing ofthe device in the Caltech Intermediate Form (CIF) file format. Further processingwas done with CleWin (Phoenix Software) to array and arrange the device to besuitable for fabrication on 4-inch silicon wafers. Five separate 5-inch photomasks(Microchrome) for flow and control layers were printed at 2 µm resolution using a57GBGB aythoxsCelltrapLysis chamberRT chamber Digital PCR array150 µm 15 µm Closed valve30 µm 10 µmA Open valveB CReversetranscriptionOil partitioningPCR solutionPDMS10 nL50 nLCellcaptureCelllysisDigitalPCRReagents InletsValves ValvesValves ValvesHydrationHydrationOutletOutletOutletOutletFigure 3.1: Microfluidic device design and operation. (A) Layout of the microfluidicmodules for single cell digital PCR analysis. The cells are trapped, lysed and thetranscript target is reverse transcribed before being injected into a digital PCR array.The respective height dimensions for the chambers and channels are indicated. (B)Complete microfluidic device. Four main panels contain 50 identical modules capableof parallel processing of up to 200 single cells. The device also contains a networkof hydration channels to prevent water loss during thermocycling. (C) Workflowschematic for single cell digital PCR analysis of mRNA. Depending on the protocolused for mRNA or miRNA analysis, cells are either chemically or heat lysed afterbeing trapped (insert, K562 cells trapped; the scale bar is 100 µm). Transcriptsare then reverse-transcribed to cDNA by mixing RT reagents using diffusion. PCRreagents are then injected into the device, mixed by diffusion with cDNA productsand injected in to the digital PCR array using dead end filling. A fluorinated oil(FC-40) is then used to displace the remaining PCR solution in the channels andcompartmentalize individual PCR digital chambers.58GBGB aythoxsLaserWriter (Microtech). The flow layer refers to the layer of channels used to handlesamples and reagents, while the control layer refers to the layer of channels below theflow layer which are used to valve flow channelsThe flow mold was manufactured by the subsequent layering and lithographyof four different photoresists. Fabrication using photoresists was done according tomanufacturer’s specifications. Blank silicon wafers (5 inches, Silicon Quest) weredehydrated at 190 ◦C for 1 hour and exposed to hexamethyldisilazane (Sigma) for5 minutes prior any use. First, 16 µm thick valvable channels were fabricated byspinning SPR220-7 (Rohm and Haas). After exposure and development the entirewafer was baked at 190 ◦C for one hour. Next, a layer of 12 µm thick SU8-3010(MicroChem) was used to create the cell traps and channels into the digital array. A50 µm layer of SU8-3025 (MicroChem) was then used to make the 25 pL reactors ofthe digital array as well as the microfluidic channels responsible for reagent transferinto and out of the device. Finally 150 µm high SU8-100 (MicroChem) was used toconstruct the large 10 nL and 50 nL chambers. To fabricate the control mold a singlelayer of 25 µm high SU8-3025 was used. After fabrication, all the molds were coatedwith an ∼100 nm layer of poly(paraxylylene) (parylene C) to prevent adhesion bypolymerized poly(dimethylsiloxane) (PDMS) and increase mold durability[134].The microfluidic devices were assembled through the subsequent bonding of threelayers of replica molded PDMS on top of a 50×75mm glass slide. Approximately 60g of 5:1 (part A:part B) PDMS (General Electric, RTV 615) was degassed, poured onto the flow mold and then baked at 80 ◦C to polymerize for 60 minutes. Meanwhile20:1 PDMS was spun at 1800 rpm for 60 s on the control molds and was left tocure at 80 ◦C for 30 minutes. PDMS casted onto the flow mold was then carefullypeeled off, trimmed to device size and reagent inlets and outlets were punched usinga 20 gauge coring tool (Technical Innovations). The PDMS flow layer was manuallyaligned on top the PDMS coated control molds and cured for a further 30 minutesat 80 ◦C. While this is happening 20:1 PDMS was spun at 1800 rpm for 60 s ontocleaned 50 x 75 mm glass slide and baked at 80 ◦C for 30 minutes. Following bonding,the combined two layer PDMS structure was peeled from the mold and holes werepunched to allow fluid input to the control layer. The PDMS slab was then croppedinto individual devices and deposited on the PDMS coated glass slides to completethe devices and baked overnight at 80 ◦C. Finally, the devices were kept in a sealedcontainer with 99% humidity for at least 24 hours prior to any use.59GBGB aythoxsHCHCG Devixe dpervtionThe device was operated through control of 12 pneumatic valves, which apply pressure(30 psi) to the Krytox (DuPont) oil used as fluid in the control lines. The device maybe controlled using a manifold of manual valves, however for current study valveswere controlled by solenoid actuators (Fluidigm Corp.) in a semi-automated fashionthrough a digital input-output card (NI-DAQ, DIO-32H, National Instruments) usinga LabView program (National Instruments). The solenoids are connected to themicrofluidic device by Tygon tubing ending in stainless steel pins (20-gauge, SmallParts Inc.) fitted into the control line ports. Reagents were injected into the devicethrough pipette tips (Xcluda Style G Aerosol Barrier Pipet Tips, BioRad) pluggedinto device ports and applying pressure though a manifold of manual valves.The microfluidic device was designed to be compatible with commercially availablelysis and RT-qPCR reagents. Following a common procedure for cell loading, themicrofluidic device was used to perform either chemical lysis followed by RT anddigital PCR, or heat lysis followed by RT and digital PCR. The microfluidic devicewas designed to implement one-pot chemistries, by which reagents are sequentiallyadded to the reaction without intermediate product purification steps.HCHCH Cell aovyingEach cell-loading lane was primed by flowing PBS with 0.5% (m/v) bovine serumalbumin through the channel for 30 s. This treatment helped prevent cells fromadhering to the PDMS walls of the microfluidic channels. Cells were taken directlyfrom suspension culture and injected into the cell loading lanes. Typically 5 µL to 20µL of cells in suspension at concentrations between 5 × 105 and 1 × 106 cells/mL wereinjected into the device with a flow rate of approximately 20 nL/s. Concentrationshigher than 2 × 106 cells/mL occasionally caused clogging at the inlet port or inthe channel at trap locations. Following loading, 2 µL of the PBS solution was runthrough each of the four lanes of cell traps in order to displace the culture mediumand remove any free nucleic acids that may be present. Finally valves were actuatedto isolate each of the cell traps. Manual inspection of each of the 200 cell traps wasthen performed to confirm and record the number of cells present in each trap.60GBGB aythoxsHCHCI Detexting mgcV fiith Chemixvl aysis vny giByeCgFor measurements of mRNA, we used the CellsDirect™ One-Step qRT-PCR kit (In-vitrogen) to perform chemical lysis followed by RT and digital PCR. Lysis solution(15 µL) was prepared according to the manufacturer and loaded into the 4 reagentinjection lanes. Lysis is performed by opening valves to the cell trap chambers andinjecting the lysis reagent through the cell trap chambers and dead-end filling the 10nL chamber. The lysis reaction is incubated at room temperature for 10 min, fol-lowed by heat-inactivation of the lysis reagent at 70◦C for for 10 min. Temperatureis controlled by placing the device on a flatbed thermocycler.The one-step RT-qPCR mix [prepared as 1 µL of SuperScript III RT/PlatinumTaq Mix, 25 µL of 2× Reaction Mix (with ROX reference dye), 2.5 µL of 20× TaqManAssay (primers and probes, Life Technologies), 1µL of 50 mM MgSO4, 5.5 µL ofwater, and 5 µL of 1% Tween-20] was then combined with the cell lysate into the50 nL reaction chamber, and allowed to mix by diffusion for 30 minutes. Reversetranscription was performed by incubating the device on a flatbed thermocycler at 50◦C for 20 minutes. The RT product and PCR reaction mix was then pushed into thedigital PCR array. After completely dead-end filling all of the chambers, fluorinatedoil (FC-40, Sigma) was used to partition the array of digital PCR chambers[133]. Acontinuous flow of this oil was maintained throughout the thermocycling reaction toensure reaction segregation. Light mineral oil, applied between the glass substrateand the thermocycler, was used to improve the thermal contact during thermoclying(30 second hot-start at 95 ◦C, followed by 30 cycles of 95 ◦C for 3 seconds and 60 ◦Cfor 30 seconds). Commercially available primers and probes were used to measureGAPDH (Hs 02758991 g1, Life Technologies) and BCR-ABL (Hs 03024784 ft, LifeTechnologies). The procedure is summarized in the table below.HCHC5 Detexting migcV fiith Hevt aysis vny giByeCgThe protocol for detecting microRNAs is similar to the protocol used for measuringmRNA transcripts. Cells were trapped, washed, and isolated between valves as de-scribe above. However, in this protocol, heat lysis (85 ◦C for 7 minutes) was usedin the place of chemical lysis and was followed by a 2-step RT and PCR reactionusing the High Capacity cDNA Reverse Transcription kit (Applied Biosystems) andthe TaqMan® Fast Universal Master Mix. The RT reaction was performed in the61GBGB aythoxsTable 3.1: Chemical Lysis and 1-step RT-dPCR ProtocolStep Description Time1 Prime device with PBS 0.5 mg/mL BSA and 0.5 U/µL RNase In-hibitor1 min2 Inject cell suspension (passive cell trapping) 1 min3 On chip cell washing with PBS containing 0.5 mg/mL BSA and0.5 U/µL RNase Inhibitor1 min4 Close valves partitioning cell loading channel and isolating singlecells30 sec5 Count cells by visual inspection with microscope 7 mins6 Inject lysis reagent through the cell capture chamber, dead-endfilling the 10 nL chamber1 min7 Close reagent injection valve, creating isolated reactors combiningthe cell capture chamber and lysis reservoir chamber30 sec8 Perform lysis at room temperature and heat inactivation of thelysis reagent at 75 ◦C by placing device on flatbed thermocycler25 min9 Flush reagent injection lines with reagent for RT-qPCR 30 sec10 Inject RT-qPCR reagent through combined cell-capture/lysischamber into 50 nL RT chamber5 min11 Close valve to RT chamber. Allow for mixing by diffusion (30mins) before RT (20 mins)50 min12 Open valves to digital PCR chamber array, and inject RT product(continuing to push with RT-qPCR reagent). Close all valves afterdead-end fill3 min13 Prime the oil fluidic bus lines, before opening valves to allow theoil to flow through the reactors and to the output port. Place onflatbed thermocycler with light mineral oil and tape down for ther-mal contact8 min14 Run RT-qPCR protocol (varies)62GBGB aythoxsTable 3.2: Heat Lysis and 2-Step RT-dPCR ProtocolStep Description Time1 Prime device with PBS 0.5 mg/mL BSA and 0.5 U/µL RNase In-hibitor1 min2 Inject cell suspension (passive cell trapping) 1 min3 On chip cell washing with PBS containing 0.5 mg/mL BSA and0.5 U/µL RNase Inhibitor1 min4 Close valves partitioning cell loading channel and isolating singlecells30 sec5 Count cells by visual inspection with microscope 7 mins6 Heat lysis by placing device on flatbed thermocycler and heatingto 85 ◦C7 min7 Flush reagent injection lines with reagent for RT 2 min8 Inject RT reagent through the cell capture chamber, dead-end fill-ing the 10 nL RT chamber1 min9 Close reagent injection valve, creating isolated reactors combiningthe cell capture chamber and RT chamber30 sec10 Perform reverse transcription (pulsed temperature protocol) byplacing device on flatbed thermocycler2.5 hr11 Flush reagent injection lines with reagent for PCR 2 min12 Inject PCR reagent through combined cell-capture/RT chamberinto 50 nL mixing chamber5 min13 Close valve to mixing chamber. Allow for mixing by diffusion 40 min14 Push RT product and PCR mix into digital PCR array, and parti-tion array with oil8 min15 Run digital PCR protocol (varies)10 nL chamber, and the 50 nL chamber was simply used to mix the RT product withthe PCR reagent prior to injection into the dPCR array. A pulsed temperature RTprotocol was carried out by placing the microfluidic device on a flatbed thermocycler(2 min at 16 ◦C, followed by 60 cycles of 30 seconds at 20 ◦C, 30 seconds at 42 ◦C,and 1 second at 50 ◦C). Following RT thermocycling the RT enzyme was inactivatedat 85 ◦C (5 min), and then the device was cooled to 4 ◦C. After injection into thedPCR array, fast PCR thermocycling was performed as described in the protocol formeasuring mRNA, but with temperatures according to manufacturer specifications(95 ◦C for 30 seconds, followed by 30 cycles of 95 ◦C for 3 seconds and 60 ◦C for 30seconds). Commercially available primers and probes were used to measure miR-16(hsa-miR-16, Applied Biosystems). The procedure is summarized in the table below.63GBGB aythoxsHCHCK bevsurement of gcV EyitingThe EEF2K RNA edit location (hg18 chr16:22204361, hg19 chr16:22296860) wasinitially identified and validated by Shah et al. through RNA-seq on an estrogen-receptor--positive metastatic lobular breast cancer[32]. This editing location wasnot predicted by Park et al. in their RNA editing analysis of the K562 ENCODERNA-seq data[135]. However, our inspection of their mapped RNA-seq data (GEOGSM958729) revealed that replicates 1 and 2 had 20/57 and 15/48 mapped readswith A to G variants, respectively, indicating that it may also be a candidate edit inK562 cells as well.A two-colour MGB TaqMan qPCR assay was designed to distinguish betweenwildtype (A) and variant (G) (F: CCC TCC TCA AAG TGC TGA GAT TAC, R:TTC AAT GGA ATT CAG CTC TCA CAT, WT: VIC- AGATGCTVGGTGCG,Edit: 6FAM- ATGCTGGGTGCG). This assay was initially validated against DNAand total RNA purified from K562 cells. After 40 cycles of PCR, the predicted editwas not detected in either the DNA samples or the no-RT controls on the RNAsamples, but was present in all reverse-transcribed RNA samples. The cDNA prod-uct, gDNA and no-RT RNA additionally all contained only a single amplicon of theexpected length (108 bp). Probe specificity for the single nucleotide variant was con-firmed by performing microfluidic digital PCR in custom chips consisting of 765 2-nLarrays; the majority of all positive chambers (69 only variant, 406 only wildtype, 93variant and wildtype) showed amplification in only one colour. The primer specificityin purified total RNA and DNA, probe specificity, and detection of the variant in onlyreverse-transcribed RNA strongly suggest that we are measuring true RNA editingevents.cDNA was synthesized using the High Capacity cDNA kit (Applied Biosystems).Final reaction brews contained 1× RT buffer, 1× RT Random Primers, 5 mM dNTPs,0.1 % Tween-20, 3.3 U/µL MultiScribe Reverse Transcriptase and were incubated at25 ◦C for 10 min, 37 ◦C for 1 hour, 85 ◦C for 5 minutes and held at 10 ◦C untilPCR cycling. PCR reaction brews contained 1× CellsDirect Buffer (Invitrogen), 0.1% Tween-20, 900 nM each forward and reverse primer, 250 nM each probe and 0.04U/µL Platinum Taq (Invitrogen) and were cycled for 2 min at 95 ◦C, followed by 40cycles of 95 ◦C for 15 s, 61 ◦C for 1 minute. All conditions for single-cell RT-qPCRwere the same, except the PCR cycling times were reduced (30 second hot-start at95 ◦C, followed by 30 cycles of 95 ◦C for 3 seconds and 60 ◦C for 45 seconds).64GBGB aythoxsHCHCL Cell Culture vny gcV eurixvtionK562 cells were obtained from ATCC (CCL243) and were cultured in DMEM (Gibco)supplemented with 10% FBS (Gibco). Purified RNA was extracted from K562 cellsusing the mirVana RNA isolation kit (Ambion) according to manufacturer directions.HCHCM Devixe VnvlysisMicrofluidic devices were scanned at 2 µm resolution using a custom built confocalscanner (Huron Technologies International Inc.). Automated analysis of the digitalimages was performed with in-house software written in C. 50 rows of 1020 chambersof the digital PCR array (51,000 total) were analyzed simultaneously by specifyingmanually control points (top left, top right and bottom left) of the region to beanalyzed. The chambers were automatically located by detecting a pre-mixed pas-sive dye (ROX) using a previously described algorithm6. A fluorescence intensitymeasurement was then calculated for each chamber by summing the pixel values thefluorophore of interest in a 21×21 pixels square region centered at each chamber andnormalized to the overall fluorescence of the ROX. The fluorescence intensity values ofevery chamber in each of the 50 sample lanes were then plotted on a single histogram,which was repeatedly smoothened using a gaussian kernel until only one point on thehistogram remained with both neighbouring points larger than it. This minimumpoint was then used as the thresholding value to distinguish positive chambers fromnegative chambers and the number of positive chambers in each of the sample laneswas written to a data file.This data file was then further processed using SigmaPlot® software to determinethe number of transcripts present in each cell. Figures and plots were generatedusing the same software. Log normal distribution curve fitting used log normal, 3parameters with default settings.HCHCN irvnsfer Exienxy bevsurementsFluorescence images were used to measure the sequential transfer of oligonucleotidesfrom the cell capture chamber into the 10 nL and 50 nL chambers. The cell capturechamber was loaded with a solution of 10 µM FAM-labeled 40-mer poly-A oligonu-cleotides (IDT), 0.1% Tween-20, and ROX passive reference dye (from CellsDirect kit,Invitrogen, P/N 54880) diluted 100×, and was pushed into the subsequent chambers65GB4B fysultswith water containing 0.1% Tween 20 and diluted ROX reference dye. The trans-fer efficiency for each chamber was calculated as (Initial Signal Final Signal)(InitialSignal), where Signal = (FAM Intensity FAM Background) (ROX Intensity ROXBackground).HCHCFE gipleys KBFunxtionRipleys K-function with the rectangular edge correction conditions described by Gore-aud et al.[136] was used to assess the spatial distribution of the hits in the array.Variance-stabilized K-function estimates for each array were given a z-score for eachdistance s based on the mean and standard deviation derived from 500 randomlygenerated arrays with the same fill factor.HCI gesultsHCICF Devixe eerformvnxeWe first characterized the efficiency, and uniformity of mixing and sample transfer inour microfluidic device using FAM labelled 40 bp oligonucleotides. Sample transferbetween the lysis chamber and RT chamber, as measured by fluorescence quantifica-tion, was found to be better than 99%; the amount of oligonucleotide remaining in the10 nL chamber was below 1%, which is determined as the limit of detection for thismeasurement based on 2 standard deviations from measured background fluorescence(Figure 3.2A). We next assessed the mixing of reagents in the 50 nL chamber by tak-ing time lapse images of fluorescence and calculating the variability in oligonucleotideconcentration across the chambers. Regression of the resulting data to an exponen-tial decay curve yielded a mixing time constant of approximately 11 minutes (Figure3.2D). Based on this analysis we implemented a 30 minute wait time in our protocolto allow for sufficient reaction mixing. Finally, we measured the reproducibility oftransfer from the RT chamber to the digital PCR array. Fluorescence measurementsof the 50 nL chamber before and after injection into the dPCR array showed thatthe fraction of solution that was injected into the digital PCR array was 50.38% (s.d.=1.31%) (Figure 3.2E).We next performed a series of experiments to assess the signal to noise and mea-surement sensitivity of our system using total RNA template purified from K56266GB4B fysultsTransfer from the RT chamber to the dPCR arrayTime (s)0 500 1000 1500 2000Coefficient of variation of fluorescence00. Diffusion mixing in the RT chambery0 = 0.19a = 0.24t = 625 s R2 = 0.99 Chamber number (top to bottom)0 10 20 30 40 50% fluorescence transferred030405060mean = 50.38%s.d. = 1.31Lysis Chamber RT chamber dPCR arrayABCDEFigure 3.2: Characterization of the microfluidic device. (A) A solution containing 40base oligonucleotides labelled with FAM was injected in the lysis chamber followingthe normal procedure. A second solution (not fluorescent) was then used to sequen-tially push the first solution into the RT chamber (B), and then the dPCR array(C). (D) Mixing by diffusion in the RT chamber (B) was monitored by measuring thestandard deviation of the fluorescent signal over time. As expected, the signal followsan exponential decay with an averaged time constant (t) of 625 seconds (R2=0.9999).(N=50) (E) To measure the amount of the cell sampled by the dPCR array, the flu-orescent solution in the RT chamber was pushed into the digital PCR array using anon-fluorescent solution (C). The transfer efficiency was evaluated for 50 chambersand found to be 50.38% (s.d.=1.31). ±3 from the mean lines are also displayed.67GB4B fysultsPurified RNA from K562 cells (pg)0.1 1 10 100 1000Measured GAPDH transcripts110100100010,000640480320160Relative Fluorescence Intensity (a.u.)0 1Mean  = 0.84s.d. = 0.06n = 51,000Number of chambersBCD010 1000 1100 1200 1300Relative position (pixels)Fluorescence intensity (a.u.)Ab = 1.2619a = 0.9618 pg-1R2 = 0.99Linear regression: y = ax + b 2 4 6 8 10-3-2-10123sL(s)−s z−scoreFigure 3.3: (A) End point fluorescence signal from the entire microfluidic device afterdPCR targeting GAPDH. RNA dilutions have been loaded into 4 different visiblepanels and reverse-transcribed on chip. The resulting cDNA is quantified by digitalPCR where digital PCR positive hits (green) in 30 pL chambers have a high signal tonoise ratio (>30). A passive dye (red) is visible in all the chambers. (B) Histogramof the distribution fluorescence intensity for 51,000 digital PCR chambers with 20 pgRNA as starting material. Normalized mean and standard deviation of the bright,or positive, chambers are indicated. (C) Randomness of the distribution of detectedtranscripts within a distance s in the dPCR array assessed by using z-scores of avariance stabilized Ripley K-function (L-function) within a 95% confidence interval.(D) Dynamic range of the digital PCR arrays using 5-fold serial dilutions of total RNApurified from K562 cells, looking at GAPDH mRNA. Linear regression analysis usingthe experimental data was performed (parameters displayed with 95% confidenceboundaries).68GB4B fysultsFigure 3.4: Z-scores of variance stabilized Ripleys K-function estimates for the spatialdistribution of digital counts at distance s from single cells in high (A, GAPDH,average 60% full), medium (B, miR-16, average 30% full) and low (C, BCR-ABL,average 2.3% full) fill factors. The 95% confidence intervals (1.96 standard deviations)are shown in black.cells. Using hydrolysis probes (TaqMan®), the signal to noise ratio obtained usingthe 4 different assays allowed a clear and unambiguous discrimination between thepositive and negative chambers (Figure 3A). Figure 3B shows a histogram of fluores-cent intensities, with a signal to noise ratio in excess of 30, obtained across 51,000fluorescent chambers in one lane of the device measuring GAPDH transcripts (Figure3.3A insert and Figure 3.3B).Digital PCR requires template molecules to be randomly distributed among reac-tion chambers in order to accurately infer the number of molecules in the array fromthe number of reaction chambers with PCR amplification. Our integrated single-celldPCR device mixes PCR solution with template by diffusion in 50 nL chambers priorto injection into the array of 1020 × 25 pL dPCR chambers, having an aggregatearray volume of approximately 25 nL. We assessed the randomness of the result-ing distribution of cDNA molecules across the dPCR array by plotting the z-scoreof the variance stabilized Ripleys k-function[133, 137] for 50 sub-arrays in a singlelane of the device loaded with purified RNA at a concentration of ∼18 pg per array(single cell equivalent) (Figure 3.3C), with the RNA loaded through the cell process-ing chambers, and subsequent processing performed in the same sequence as duringsingle cell analysis. This analysis shows that the distributions are not significantlydifferent from random and lie within a 95% confidence interval constructed around69GB4B fysultsHigh-Throughput Microfluidic Single-Cell Digital PCR Adam K. White*, Kevin Heyries*, Callum Doolin, Michael VanInsberghe, and Carl Hansen Centre for High-Throughput Biology, University of British Columbia, Vancouver, British Columbia, Canada Introduction: Here we present an integrated microfluidic device for the high-throughput digital PCR (dPCR) analysis of single cells. This device allows for the parallel processing of single cells and executes all steps of analysis including cell capture, washing, lysis, reverse transcription, and dPCR analysis.  Methods: The cDNA from each single cell is distributed into a dedicated dPCR array consisting of 1020 chambers, each having a volume of 25 pL, using surface tension-based sample partitioning.  The high density of this dPCR format (118,900 chambers per cm-2) allows the analysis of 200 single cells per run, for a total of 204,000 PCR reactions using a device footprint of 10 cm2.   Results: Experiments using RNA dilutions show this device achieves the shot-noise limited performance in quantifying single molecules, and with a dynamic range of 104.  We performed over 1000 single cell measurements, demonstrating the use of this platform in the absolute quantification of both high and low abundance mRNA transcripts, as well as microRNA that are not easily measured using alternative hybridization methods.   Conclusion: High-throughput dPCR provides a new tool in the arsenal of single cell analysis methods, with a unique combination of speed, precision, sensitivity and specificity.  We anticipate this approach will enable new studies where high-performance single cell measurements are essential, including the analysis of transcriptional noise, allelic imbalance, and RNA processing.   Abstract Microfluidic Device Design and Operation Digital PCR Performance Single Cell Transcript Measurements Device Characterization (A) A solution containing 40 base oligonucleotides labeled with FAM was injected in the lysis chamber. A second solution (not fluorescent) was then used to sequentially push the first solution into the RT chamber (B), and then the dPCR array (C).   (D) Mixing by diffusion in the RT chamber (B) (N=50).  (E) Transfer from RT do dPCR array was found to be 50.38% (s.d.=1.31). ±3σ from the mean lines are also displayed.  A.  End-point fluorescence of digital PCR positive hits (green) have a high signal to noise ratio (>30). A passive dye (red) is in all chambers.  B.  Fluorescence intensity for 51,000 digital PCR chambers (20 pg RNA starting material), using TaqMan qPCR probes (Life Technologies). C.  Randomness of the distribution of detected transcripts in the dPCR array assessed by the Ripley k function within 95% confidence.  D.  Dynamic range of the digital PCR arrays using 5-fold serial dilutions of total RNA purified from K562 cells, looking at GAPDH mRNA.  A.  Layout of the microfluidic modules for single cell digital PCR analysis. The cells are trapped, lysed and the transcript target is reverse transcribed before being injected into a digital PCR array. The respective height dimensions for the chambers and channels are indicated.  B.  Complete microfluidic device. Four main panels contain 50 identical modules capable of parallel processing of up to 200 single cells.  C.  Workflow schematic:  i.  Cells are either chemically or heat lysed after being trapped (insert, K562 cell trapped; the scale bar is 100 µm).  ii.  Reagents are sequentially injected and mixed by diffusion to perform RT and PCR iii.  A fluorinated oil (FC-40) is then used to displace the remaining PCR solution in the channels and compartmentalize individual PCR chambers.  Microfluidic Integration of Single Cell Processing and Digital PCR Single-Cell RT-qPCR ‘Megapixel’ Digital PCR Single-Cell Digital PCR •  Microfluidics well suited to manipulating single cells, and small (nL, pL) volumes increase effectivae concentration of single molecules •  Digital PCR provides single molecule resolution and improves precision over qPCR techniques   •  Microfluidic integration enables scalable combination of processing many cells (100’s) and performing many dPCR reactions (1000’s) White AK, et al., PNAS 2011 Heyries K, et al. Nature Methods 2011 High-Abundance mRNA (GAPDH) Low-Abundance mRNA (BCR-ABL) MicroRNA (miR-16) 0 1 2 30 35 (16%) 10 2 0 12 (5%)1 42 16 7 02 20 12 4 03 21 7 3 04 10 8 3 15 7 1 0 06 0 7 0 07+ 3 0 2 0103 (47%) 71 (32%)RNA Editing in Single Cells •  Measurement and discrimination of single nucleotide RNA edit •  Adenosine to Inosine in EEF2K mRNA  •  RNA edit identified by ENCODE •  Optical multiplexing with 2-colour TaqMan probes •  N = 221 K562 cells  EEF2K wt mRNA (copies) EEF2K Edited mRNA (copies) Purified RNA from K562 cells (pg)0.1 1 10 100 1000Measured GAPDH transcripts110100100010,000b = 1.2619a = 0.9618 pg-1R2 = 0.99Linear regression: y = ax + b -2-1.5-1-0.500.511.521 2 3 4 5 6SL(S) - S640480320160Relative Fluorescence Intensity (a.u.)0 1Mean  = 0.84s.d. = 0.06n = 51,000Number of chambersBCD010 1000 1100 1200 1300Relative position (pixels)Fluorescence intensity (a.u.)ACelltrapLysis chamberRT chamber Digital PCR array150 µm 15 µm Closed valve30 µm 10 µmA Open valveB CReversetranscriptionOil partitioningPCR solutionPDMS10 nL50 nLCellcaptureCelllysisDigitalPCRReagents InletsValves ValvesValves ValvesHydrationHydrationOutletOutletOutletOutletFigure 3.5: Digital PCR measurements on K562 single cells. From left to right:(A) GAPDH mRNA transcripts abundance measurements over 3 independent ex-periments. The transcript abundance frequency displayed on the x-axis is binnedevery 200 transcripts. (B) BCR-ABL fusion gene mRNA transcript measurementover 3 independent experiments. The transcript abundance frequency displayed onthe x-axis is binned every 4 transcripts. (C) MicroRNA miR-16 measurements over3 independent experiments. The transcript abundance frequency displayed on thex-axis is binned every 100 transcripts.the mean of 500 simulated data sets, based on a Poisson distribution of templatemolecules in each chamber[137, 138]. This statistical analysis was also repeated forsingle cell measurements and demonstrates that mRNA distributions are also within95% confidence intervals under this scenario (Figure 3.4).Finally, to verify the response and precision of the device, we measured the abun-dance of GAPDH mRNA from serial dilutions of total RNA purified from K562 cells(Figure 3.3D). Measured concentrations from 5-fold RNA dilutions were in excellentagreement with the expected template abundance and dilution factor (R2=0.9989),with tested values spanning a chamber occupancy rate between 0.02% and 48% (Fig-ure 3.3A), corresponding to a range of 80 to 0.64 pg total RNA template per reaction.HCICG hingle Cell irvnsxript bevsurementsWe then conducted a series of single cell experiments to establish the combined ca-pabilities of high-throughput cell processing and dPCR analysis. As a first test weperformed measurements of GAPDH transcripts from single K562 cells (Figure 3.5A).GAPDH encodes for glyceraldehyde 3-phosphate dehydrogenase and is a commonlyused high-copy number endogenous control for RT-qPCR experiments[62, 129, 139].Over a total of 3 device runs we observed successful amplification for 100% of isolatedsingle cells (N = 288). We note that the total number of cells analyzed was lower70GB4B fysultsthan the theoretical device capacity due to the use of some device lanes for controlsand sub-optimal cell trapping efficiency; these experiments were performed with anun-optimized cell trap geometry that resulted in approximately 60% single cell occu-pancy. Further trap optimization has resulted in 96% single cell occupancy. However,reaction chambers that did not capture a cell provide stringent no-cell-controls fromwhich to evaluate background signal in analysis. In these no-cell controls, we ob-served a very low frequency of mRNA detection events, with an average of 1.6 hitsper no cell (N = 106 chamber), corresponding to background mRNA contaminationlevels below 0.1%. No correlation was observed between background signal and thecopy-number measured on single cells in adjacent wells.The measured number of transcripts in single cells revealed a log-normal distribu-tion (0.94 K R2 K 0.98) of GAPDH with an average copy number of 1,563 transcriptsper single cell. In three replicate measurements, performed on separate devices anddifferent cultures, the mean GAPDH expression was very reproducible, ranging from1,412 to 1,741 copies per cell. These values are in good agreement with estimatesof 1,761 (SD = 648) copies we have previously made by normalizing single cell mi-crofluidic RT-qPCR measurements to a standard curve calibrated by dPCR[129]. Allreplicates revealed a large variability in GAPDH expression between cells, with mea-sured values spanning approximately ten-fold in absolute copy number (∼400 to 4000copies). This variability in GAPDH levels was also consistent between runs with anaverage standard deviation of 601±24 copies per cell.To demonstrate the use of high-throughput digital PCR in accurate quantificationof low abundance transcripts, we next measured the expression of BCR-ABL tran-scripts, a fusion gene resulting from a reciprocal translocation of chromosomes 9 and22 that is associated with chronic myelogenous leukemia (CML) and is also presentin K562 cells. Over 3 runs we reliably detected the BCR-ABL fusion transcript inevery cell analyzed (N = 242), with a mean expression of 33 copies per cell, a rangespanning 25 fold in relative expression (4 to 100 copies per cell), and a standard devi-ation of 18.9 copies per cell (Figure 3.5B). These measurements were consistent withan independent study based on RT-qPCR[140] which measured BCR-ABL mRNAin single K562 cells ranging from 2 to 262 copies per cell with the majority of cellscontaining approximately 40 copies per cell.Using an alternative two-step RT-PCR protocol we next demonstrated the useof our technology for absolute measurements of microRNA (miRNA) levels in single71GB4B fysults0 1 2 30 35 (16%) 10 2 0 12 (5%)1 42 16 7 02 20 12 4 03 21 7 3 04 10 8 3 15 7 1 0 06 0 7 0 07+ 3 0 2 0103 (47%) 71 (32%)Edited EEF2K mRNA (copies) WT EEF2K mRNA (copies) Figure 3.6: Measurement of single nucleotide RNA editing of EEF2K in single K562cells. Single cells are binned according to abundance of wildtype (WT) or editedEEF2K transcript. Quadrants containing homozygous wildtype, homozygous edit,heterozygous, and not-detected EEF2K expression are indicated in colour.cells. miRNAs are a large family of conserved short (∼22 base pairs) non-codingRNAs that associate into the RNA induced silencing complex (RISC) and functionas post-transcriptional regulators via targeted degradation of complementary tran-scripts or by repressing translation[141]. miRNA are important regulators in manybiological processes, including development, oncogenesis[142], and immunity, makingthem of high interest as biomarkers of disease[143]. Due their short length miRNA arenot easily amenable to single molecule analysis by RNA FISH[144] and although hy-bridization techniques have been used to visualize miRNAs[145, 146], single moleculequantification of miRNA copy numbers and variability in single cells by FISH has notbeen achieved. Using a commercially available RT-qPCR detection strategy, basedon miRNA-specific stem-loop reverse transcription primers[60], we measured the ex-pression and variability of miR-16, a microRNA expressed at medium levels across abroad range of tissues, in K562 cells. Over 3 replicates, K562 cells were found to havea mean copy number of 675 per cell which is in close agreement with dPCR measure-ments performed on bulk samples[129]. The distribution of miR-16 expression wasconsistent across runs, with a standard deviation of 262±44 (Figure 3.5C).Finally, we applied our single cell dPCR device to measure the extent of singlenucleotide RNA editing. Editing of RNA molecules by nucleoside deaminases hasrecently gained attention as a mechanism by which transcripts can be modified away72GBIB Diswussion & Conwlusionfrom the genomic code, with potential implications for transcript stability, alternativesplicing, translation efficiency, and protein sequence. Next generation sequencingtechniques have provided strong evidence for large numbers of RNA editing eventsacross different tissues. These events are typically incomplete, representing a fractionof total RNA. However, little is known regarding the variability of editing betweenindividual cells. Is there a subset of cells that edit all transcripts? Is the frequencyof editing similar across different cells? To investigate these questions we used thespecificity of dPCR to measure adenosine to inosine editing of position chr16:22296860(hg19) in the mRNA coding for EEF2K in single cells. This edit was initially identifiedin a lobular breast cancer[32] and found to be edited at a frequency of ∼0.33 in RNA-seq data from ENCODE[147]. Wildtype and edited versions of EEF2K, differingby a single nucleotide substitution, were simultaneously measured on single cellsusing a two-color TaqMan SNV assay based on minor grove binding probes (LifeTechnologies). We measured EEF2K in 221 single K562 cells and found 71/221(32%) cells expressed both wildtype and edited EEF2K transcripts, 12/221 (5%)cells expressed only edited transcripts, 103/221 (47%) cells expressed only wildtypetranscripts, and 35/221 (16%) cells in which neither form of EEF2K transcript wasdetected (Figure 3.6). The population-averaged editing frequency was found to be∼0.19 from single cell measurements which is consistent with our measurements ondilutions of purified RNA, producing an edit frequency of ∼0.18. We observed nonon-specific amplification in any no-cell reactions. A control experiment omitting theRT enzyme detected genomic wildtype EEF2K in 10 out of 228 single cell reactions(∼4.4%) and no observed positives for the edited transcript.HC5 Disxussion & ConxlusionSingle cell dPCR is particularly well suited to measurements that require high pre-cision and sensitivity. We note that the half-sampling approach implemented here(where we inject half of the single-cell RT product, i.e. cDNA, into the digital PCRarray) introduces additional stochastic variation in measured cDNA copy numberthat is particularly important for low abundance transcripts. A theoretical analysisof the measurement precision of digital PCR has previously been presented[148]. Fol-lowing this treatment we calculated the expectation value, confidence intervals, andcoefficient of variance for our 1020 chamber dPCR array as a function of the total73GBIB Diswussion & ConwlusionFigure 3.7: The digital array response curve. (A) Average values of the startingmolecule number m (black) are shown for the combined half-sampling and digitalquantification bounded by 95% confidence interval in grey. Values were derived usinga Monte Carlo method to simulate both half sampling and digital PCR array quan-tification. (B) Coefficient of variance for the quantification with (black) and without(blue) half-sampling.74GBIB Diswussion & Conwlusionnumber of molecules that are loaded into the array (assuming no losses for sampling).In order to generate confidence intervals that also include the effect of half-samplingwe implemented a Monte Carlo method. Briefly, a starting population of m moleculeswere uniformly and randomly assigned to two groups, n (sampled in the array) and o(not sampled in the array). Each n molecules were then uniformly and randomly as-signed a detection chamber, allowing for the possibility of multiple molecules assignedto the same chamber. The number of chambers containing at least one molecule (k),was then counted and stored in association with the starting m. This process wasrepeated for 1 ≤ m ≤ 20400 a total of 10000 times each. These results were thensearched for each value of k (1 ≤ k ≤ 1020) to generate distributions for the numberof starting m molecules which led to k hits. Sample average, standard deviation and95% confidence intervals were then calculated from these distributions. Based on theresults, summarized in Figure 3.7, half sampling is the dominant source of measure-ment error over the entire dynamic range of measurement (approximately 5 logs).Nevertheless, the aggregate precision is still excellent with 93 % of the potential 5-logdynamic range of the device has a coefficient of variance (CV) less than 0.1, and75 % has a CV less than 0.05. We further note that incomplete sampling of singlecell products is also a feature of previously reported single cell dPCR[62, 149]. Thissampling variability could be reduced through improvements in device architectureand/or modified protocols including pre-amplification prior to loading the dPCR ar-ray, pushing the sample into the array with an immiscible fluid to avoid dilution,or using sequential load and mix steps to assay dilutions of a single cell product inmultiple arrays. Finally, we note that dPCR experiments should be interpreted asmeasurements of absolute cDNA molecule copy number, with the direct correlationto mRNA abundance dependent on RT efficiency which may vary between differentassays.Here we have demonstrated the application of our system in making absolutemeasurements of cDNA derived from mRNA and miRNA across hundreds of singlecells. Measurements of GAPDH transcript levels, commonly used as an endogenouscontrol in qPCR analysis, show that population-averaged expression measurementsare consistent across different cultures, but are widely variable (CV ∼ 40%) at thesingle cell level. This highlights how the common practice of normalizing single cellRT-qPCR expression measurements to a control gene is not advisable and is likely tointroduce additional noise. Our measurements also show that miR-16, widely used75GBIB Diswussion & Conwlusionas an endogenous reference for miRNA analysis[60], exhibits a similar coefficient ofvariation (CV ∼ 40% ) at the single cell level, although at approximately 2 fold lowerexpression. As a measure of variability we calculate the Fanno factor, defined as F =2/, for GAPDH and miR-16 to be 231 and 102 respectively, showing that miR-16 ismore tightly regulated in these cells. K562 cells are an inherently heterogeneous sam-ple and much of the measured variability in both GAPDH and miR-16 may representdifferences in differentiation state, cell cycle stage, and cell volume differences.In addition to the quantification of short transcripts such as miRNA, dPCR isuniquely suited to the detection and quantification of transcripts having high se-quence homology. Here we have demonstrated this capability in studying the singlecell distribution of RNA editing events that give rise to a single nucleotide variantin EEF2K. Given the low copy number of this transcript it is difficult to concludethat the noise in EEF2K measurements represents a significant difference in editingbetween cells; we note that EEF2K was chosen as it was also found to be edited inother tissues (see methods), but it is not the most abundant edited transcript in K562cells. Further testing of differences in editing activity between cells may benefit fromoverexpression of editing substrates. In an analogous application, dPCR may alsobe used to accurately measure the abundance of transcripts derived from differentalleles having heterozygous SNPs, thereby allowing for the assessment of differentialallelic imbalance and/or epigenetic silencing across many single cells. Alternatively,analysis of genomic DNA for SNVs or copy number variants may be achieved throughimproved lysis and nuclei digestion, coupled with pre-amplification.Our measurements of the BCR-ABL fusion transcript also demonstrate the reli-able detection and quantification of rare targets across large numbers of single cellswith no false-positives observed in no-cell chambers. This sensitivity and specificitymay allow for expanding the range of useful single cell biomarkers, including low-copy transcription factors, and enables the identification and discrimination of mi-nority sub-populations in complex tissues, or the detection of rare cells harbouringmutations for early detection in the monitoring of minimum residual disease[150]. Al-though increased dPCR density may be used to extend the current throughput of thisdevice from 200 to approximately 1000 cells per run, practical limitations in fabrica-tion are likely to make further scaling difficult. However we believe the throughputpresented here is well-suited for the analysis and quantification of ultra-rare cells,such as circulating tumour cells, following enrichment by FACS or immunocapture.76GBIB Diswussion & ConwlusionIntegrated single cell dPCR provides unique capabilities in terms of combinedthroughput, precision, sensitivity, and specificity[129, 151]. These capabilities arecomplementary to the expanding array of single cell analysis tools being developedand applied. Here we have demonstrated how these capabilities may be used tomeasure a variety of transcriptional features including mRNA expression, miRNAexpression, low copy fusion transcripts and SNV discrimination. Moving forward, weanticipate that this single cell platform may be adapted to an expanding range ofdigital single cell analyses, including analysis of single cell copy number variations oraneuploidy, SNV genotyping, and digital protein quantification[152].77Chvpter IhingleBCell Vnvlysis of aipiycvnopvrtixle gcV Delivery1ICF dverviefiTechniques to manipulate gene expression are fundamental to probing the function ofproteins in cellular processes, and hold tremendous potential to enable gene therapiesfor a wide variety of disorders. Here we use single cell analysis to investigate a methodusing lipid nanoparticles (LNPs) to deliver small interfering RNA (siRNA) and mes-senger RNA (mRNA) to efficiently knock down, or knock in, gene expression in celllines including human embryonic stem cells, fibroblasts, and erythroleukemia-typeK562. LNPs were prepared using the NanoAssemblr™ microfluidic-based nanoparti-cle manufacturing platform, and mimic the neutral cholesterol containing structure oflow-density-lipoproteins (LDL). Uptake of siRNA-LNP in embryonic stem cells wasenhanced by the presence of apolipoprotein E4 (ApoE4), resulting in efficient uptakeafter 24 hours with little apparent toxicity. We characterized the siRNA-LNP bymeasuring the dependence of particle uptake as a function of ApoE, siRNA concen-tration and incubation time. A microfluidic device for performing high-throughputsingle-cell digital PCR was used to precisely quantify knockdown of gene expressionin hundreds of individual cells. Flow cytometry measurements of LNP uptake andeGFP translation were compared to single-cell digital PCR measurements of mRNAdelivery to characterize transfection performance. By examining the cell-to-cell vari-ability, kinetics, and efficiency of using this lipid nanoparticle technology for nucleicacid delivery, we provide quantitative measurements on both expression levels anddistributions which should be useful to direct the use of this technology in a varietyof applications.1ghe work in this vhtpter is in volltuorttion with crevision atnosystems InvA784BFB IntroxuwtionICG IntroyuxtionIn this chapter we sought to apply our newly developed single-cell transcription anal-ysis technologies in characterizing the use of nanoparticles for manipulation of geneexpression. Lipid nanoparticles (LNPs) are currently one of the leading deliverysystems for conveyance of RNA (e.g. mRNA, siRNA) to cell cytoplasm. However,as most studies of LNP performance have relied on bulk measurements of RNA, thecell-to-cell variability in delivered RNA abundance and efficacy remains unknown. Byusing single-cell digital PCR, we were able to address this knowledge gap by preciselymeasuring the abundance and distribution of mRNA delivered to cells. Similarly,were able to assess the efficacy of siRNA delivery to perform targeted knockdownof specific genes by measuring the amount of the targeted mRNA in single cells.These single-cell digital PCR measurements provide important transcript informa-tion that complements measurements of translated fluorescent proteins or delivereddye by flow cytometry or microscopy. In this study, we combine the single-cell mea-surement techniques of microfluidic digital PCR and flow cytometry to characterizethe performance of a novel LNP reagent for mediating gene expression with single cellresolution in order to inform the development and use of nanoparticles for scientific,and potentially clinical applications.Delivery of nucleic acids, such as mRNA and small interfering RNA (siRNA),to cells is an important tool for altering gene expression with applications in re-search and therapeutics[153, 154]. In research settings, RNA interference (RNAi) isnow widely used for inhibition of gene expression in dominant hereditary diseases(e.g. Huntington’s)[155], cancer[156], and infectious diseases, as well as fundamen-tal studies of the effect of suppressing expression of target genes[157]. Delivery ofmRNA is less established than siRNA, however has found utility in the generationof vaccines[158–160], and the reprogramming of cells (e.g. to pluripotency, or di-rected differentiation)[161, 162]. The major obstacle to RNA-based gene therapy isefficient cytosolic delivery. Recent developments in delivery vehicles, such as the useof lipid nanoparticles (LNPs) to encapsulate siRNA, show promise in addressing thisissue, and have now advanced to clinical trials. The past decade has seen intensecommercial interest in the development of RNA-LNPs to mediate gene expression.In 2014, improvements in Tekmira’s small nucleic acid LNPs enabled Alnylam toachieve 100- to 1000-fold improvements in therapeutic index with siRNA[163]. Thesame year, Tekmira’s siRNA-LNPs were tested in the treatment of Ebola in non-794BFB Introxuwtionhuman primates[164]. Pharmaceutical companies such as Merk and AstraZenecahave invested hundreds of millions of dollars into Moderna Therapeutics, a companythat is developing mRNA-based therapies. Delivery of mRNA enables the cell tomake a wide variety of proteins itself, bypassing much of the expense and difficulty ofmanufacturing protein drugs. Some of the most successful drugs today are proteinssuch as Genentech’s Herceptin for breast cancer, and the therapeutic expectation ofmRNA has helped Moderna raise over $1 billion from investors and partners in thepast five years[165]. Development and application of LNPs for research and genetherapy would greatly benefit from understanding the amount of mRNA deliveredin individual cells, and the cell-to-cell variability in delivery (or efficiency of siRNAknockdown of targeted mRNA). Given this commercial and therapeutic interest, thesingle-cell measurements reported here are useful and timely information for the de-velopment of RNA delivery technology.Messenger RNA (mRNA) delivery has several advantages over DNA for mediatinggene expression, including the lack of any requirement for nuclear localization or tran-scription and the low likelihood of genomic integration[161]. Delivery of mRNA forcell reprogramming has the advantage that the timing and dose is transient and canbe controlled. However, the labile nature of mRNA and its capacity to elicit innateimmune responses are important limitations to its application. Double- and single-stranded RNAs interact with certain pattern-recognition receptors (PRRs), includingToll-like receptors (TLRs)[166], which detect pathogen-associated molecular patternsas a first-line defense against microbial invasion, and activate cellular and inflamma-tory reactions as part of an antiviral response. Endogenous RNA molecules are distin-guished from those of invading microbes by nucleotide modifications that affect PRRengagement[167]. Recent studies have shown that by incorporating these modifica-tions into delivered mRNA, the immune response was substantially reduced[159, 160],and almost negligible for small doses[161]. In particular, nucleotide analogs of pseu-douridine (or 2-thiouridine) and 5-methylcytidine triphosphates have been substi-tuted during in vitro transcription[161, 168].These advancements have assuaged a major obstacle in RNA therapeutics, how-ever the major barrier to widespread adoption of RNA delivery in therapeutics andresearch remains the need for safe and effective drug delivery vehicles[153, 169, 170].While ‘naked’, or chemically modified mRNA has shown efficacy in certain physiologi-cal settings such as in vitro cell reprogramming[161], and in vivo delivery to liver[168],804BFB Introxuwtionthere are many tissues that require an additional delivery system to facilitate transfec-tion. This is because naked mRNA is subject to degradation by endogenous enzymes,and is too large and too negatively charged to cross cellular membranes[160, 161].Strategies to improve mRNA delivery include chemical modifications such as conju-gation of polymers or viral elements[171–174], or encapsulation within lipid nanopar-ticles (LNPs)[153, 175]. LNPs are made from a variety of lipid compositions anddifferent methods to yield different sizes and structures[170, 176, 177]. Formulationof RNA encapsulated LNPs is one of the most widely used strategies for in vivo deliv-ery of siRNA, and has been successfully used to silence therapeutically relevant genesin primates[175, 178, 179] and are being evaluated in clinical trials. This chapterdescribes single cell measurements aimed a characterizing a new class of LNPs thatare fabricated using a microfluidic process that was developed in a collaboration be-tween Dr. Carl Hansen’s laboratory and Dr. Pieter Cullis’ laboratory. This processhas since been commercialized by Precision Nanosystems Inc., a University of BritishColumbia startup company. The NanoAssemblr™ from Precision Nanosystems usesa microfluidic herringbone structure[180] to enable rapid mixing of lipids in ethanolwith an aqueous phase containing RNA as described elsewhere[177, 181, 182]. Thisresults in the precipitation of lipid nanoparticles ∼50-60 nm in diameter. These LNPsmay be formulated at ‘limit size’ defined by the lipid composition, and exhibit lowpolydispersity[177, 181].Quantitative measurement techniques are fundamental to evaluating the efficacyof RNA delivery and altered gene expression. To date, studies quantifying RNAdelivery tend to use PCR or molecular imaging strategies. Quantitative PCR mea-surements have been useful in measuring dose-response and temporal relationshipsin bulk samples[183, 184], but would benefit from measurements of individual cellsin order to assess potential heterogeneity in either mRNA delivery or altered geneexpression. For example, if RT-qPCR from a bulk-sample (e.g. from hundreds tothousands of cells) shows ∼50% knockdown of a target gene, it remains unknownwhether all cells exhibit 50% knockdown, or whether half the cells have complete(100%) knockdown and the other half have no (0%) knockdown (resulting in the same50% knockdown upon averaging). Single-cell analysis can answer this question andelucidate cell heterogeneity in terms of delivery or response. Imaging techniques havethe advantage of analyzing single cells and preserving spatial information, howeverare generally limited in throughput of cells due to field of view and laborious proto-814BGB aythoxscols. Furthermore, measurement of fluorescently labeled oligonucleotides by imaginggenerally lacks the precision of qPCR or especially dPCR. This is particularly truefor measurement of siRNAs, as the short length makes labeling with multiple flu-orescent probes challenging[185]. However, microscopy has been useful to examineLNP uptake and intracellular transport[186, 187]. Considering the potential utilityof using LNPs to deliver RNA to cells and manipulate gene expression, evaluationof the efficiency of delivery would benefit from precise quantification of RNA withinsingle cells.The goal of this project is to characterize the delivery of mRNAs by LNPs throughprecisely measuring the distribution of mRNA within single cells. We combine flow cy-tometry with microfluidic single-cell digital PCR to perform high-throughput single-cell measurements. In addition to measuring the abundance and distribution of de-livered mRNA, this study also looks at conditions affecting LNP uptake, kinetics oftransfection and protein abundance, siRNA knockdown of gene expression, and cellviability. These parameters are explored in suspension and adherent cell cultures,with cell lines including human eurythroleukemic cells, hESCs, and fibroblasts. Thisproject is motivated by the potential applications of manipulating gene expressionwith exogenous RNA, and in particular by the potential advantages that mRNAdosing may provide to reprogramming somatic cells (i.e. fibroblasts) into inducedpluripotent stem cells. By characterizing the performance of LNP transfection, thisstudy aims to provide an informed basis for their use in a broad range of applicationsrequiring the manipulation of gene expression.ICH bethoysICHCF bixrouiyix hingleBCell Digitvl eCgMicrofluidic single-cell dPCR was performed using the CellsDirect kit protocol formRNA developed and described in detail in Chapter 3. A TaqMan Gene ExpressionAssay (primers and probe) for eGFP (Mr04329676 mr, FAM, MGB) was orderedfrom Life Technologies (Cat # 4351372). The GAPDH assay (Hs02758991 g1) wasalso from Life Technologies (Cat # 4331182).824BGB aythoxsICHCG Cell CultureThe human embryonic stem cell line, CA1S, was used for LNP uptake and siRNAknockdown experiments. CA1S were normally cultured in mTESR (STEMCELLTechnologies, Cat # 05850), a feeder-free maintenance medium. Wells were coatedwith Matrigel (Corning, Product #354277). Medium was exchanged daily. Cells werepassaged with TrypLE (Life Technologies, Cat # 12604013).The human newborn foreskin fibroblast cell line, BJ (ATCC CRL-2522) was usedfor mRNA transfection experiments. BJ cells were cultured in EMEM (Gibco) sup-plemented with 10% FBS. Cells were passaged with TrypLE (Life Technologies, Cat# 12604013).The human erythroleukemia cell line, K562 (ATCC CCL-243) was used as a modelcell line for suspension cultures. K562 cells were cultured in DMEM (Gibco), with10% FBS. All cell cultures were incubated at 37◦C and 5% CO2. Cell counts andviability staining were performed by the Cedex cell counter.ICHCH aipiy cvnopvrtixle FormulvtionLNP were prepared using the NanoAssemblr microfluidic-based nanoparticle manu-facturing platform[177]. LNPs prepared using the NanoAssemblr™ instrument mimicthe neutral, cholesterol containing structure of low-density-lipoproteins (LDL), whichare taken up by cells though the LDL-receptor (LDLR) in presence of ApolipoproteinE4 (ApoE4). LNPs were labeled with a red lipophilic dye (Ex/Em = 549/565 nm) toenable fluorescence imaging of particle uptake in cells. eGFP mRNA was purchasedfrom TriLink Biotechnologies (996 nucleotides, 5-methylcytidine, pseudouridine, L-6101). The siRNA duplex targeting GAPDH was purchased from IDT with the se-quence 5’-UGG CCA AGG UCA UCC AUG AdTdT-3’ (sense), and 3’-dTdTA CCGGUU CCA GUA GGU ACU-5’ (antisense), where the dT represents a DNA base forthymine (the rest are RNA).ICHCI irvnsfextionTo measure the ApoE dependence of LNP uptake, CA1S cells were plated at 400,000cells/well (6-well plates). Twenty-four hours after plating, siRNA-LNPs and ApoEwas added to the cultures (24 hours before flow cytometry). Four hours before anal-ysis by flow cytometry, siRNA-LNPs and were added to the 4-hour cohort. Each834BGB aythoxsexperiment condition was performed in duplicate.In the siRNA knockdown of HPRT in CA1S, cells were seeded at 150,000 cells/well(6-well plate) in 3 mL mTESR. Twenty-four hours later, media was exchanged, andApoE was added to the media for a final concentration of 1.0 µg/mL. siRNA wasadded to give final concentrations for a ’low’ dose of 0.1 µg/mL siHPRT, and a’high’ dose of 1.0 µg/mL siHPRT. Medium was exchanged again 24 hours later, withmTESR supplemented with the same amount of ApoE and siRNA as previously. 48hours after transfection, cells were collected for analysis by single-cell digital PCR(final cell number was ∼2×106 per well).For siRNA knockdown of GAPDH, CA1S cells were plated at 350,000 cells/well(6-well plates). Twenty-four hours later, ApoE (1.0 µg/mL final concentration) andsiRNA (1.0 µg/mL and 0.1 µg/mL final siRNA concentrations) were added duringmedia change, and incubated for 24 hours before collection for analysis. Cells werecultured in 3 mL mTESR medium. Non-Targeting siRNA was included as a parallelcontrol to show that GAPDH knockdown was specific to siGAPDH.K562 cells were transfected with siGAPDH in 96-well plates, seeded at ∼5×105cells/mL (200 µL culture volume). ApoE was added to a final concentration of 1.0µg/mL, and siGAPDH was added to different wells for each concentration tested.Cells were incubated for 24 hours before measurement.In mRNA transfections of BJ and K562 cells, cells were seeded at 400,000 cells/well(6-well format), and transfected for 24 hours with 100 ng/mL and 500 ng/mL eGFPmRNA. ApoE was added to the media at a final concentration of 1.0 µg/mL.For the proof-of-concept imaging of eGFP and mCherry in BJ cells, 500 ng/mLfor each eGFP and mCherry mRNA was added to 6-well plates containing ∼4×105cells/well (3 mL total media), and incubated for 24 hours.To test repeated high-dosing, a transfection protocol was adapted from a com-mercial iPS reprogramming kit (Stemgent), and a modification to this protocol fromWarren et al.[188]. BJ cells were seeded on day 0 at low-density (50,000 cells/well;6-well plate). Medium (EMEM) with 10% FBS was supplemented with 1.0 µg/mLApoE. Cells were transfected by adding LNPs to the culture after each daily mediachange for two weeks. Cells were split 1/6th during passage on day 7 before becom-ing confluent. Over the first four days, the dosage was increasingly ramped up from25%, 50%, 75%, to 100% of the final dose of RNA (400 ng/mL, 2 mL) by increasingthe volume of LNP-media added. This same protocol was performed in parallel on844B4B fysultscells cultured with Pluriton medium (Stemgent) supplemented with B19R interferoninhibitor (200 ng/mL), in wells coated with Matrigel (Corning).ICHC5 Flofi CytometryFlow cytometry was performed on a BD FACSCalibur (BD Biosciences), collecting50,000 events, and analyzed using FlowJo (FlowJo Data Analysis Software, LLC).Only 15,000 events were collected during the reprogramming time-course experiment.ICHCK bixrosxopyFluorescence microscopy was used to visually inspect transfection (LNP deliveredfluorescent dye) and eGFP in BJ cells. Cells were imaged at different locations inthe 6-well plate, in 90 minute intervals for 90 hours. The 6-well plate was on anautomated stage inside a custom build environmental enclosure for the microscope,with the temperature at 37◦C, 5% CO2, 70% humidity. ImageJ was used to combinebrightfield, eGFP and DiI fluorescence images. Due to background fluorescence andsignal-to-noise issues, these images were only used for qualitative observations of cellmorphology, fluorescence signal location within cells, and time duration to see eGFP.ICI gesultsICICF VpoEBDepenyent ace jptvkeWe sought to measure the uptake of lipid nanoparticles in a variety of cell tissues(e.g. blood, hESC), and types (e.g. primary, suspension or adherent). In particular,human embryonic stem cells are traditionally hard to transfect, and we wanted totest the transfection performance of our LNPs in this system. In order to explorethe potential of using these LNPs in embryonic stem cell applications, we used flowcytometry to measure the uptake of LNPs in human embryonic stem cells adaptedto in vitro culture (CA1S)[189]. The uptake of lipid nanoparticles is inferred bymeasuring the intensity of a fluorescent reporter dye (DiI), which is encapsulated inthe LNPs, and can be used to trace the LNP cargo delivery in cells (Figure 4.1). Thisflow cytometry measurement facilitates high-throughput single-cell analysis, howeverit is poorly suited to properly assessing LNP uptake and release as cells may appear854B4B fysultsFigure 4.1: ApoE-dependent LNP in hESC. Flow cytometry measurements offorward-scatter (FSC) vs. DiI (FL2), where DiI fluorescence was used as a reporterfor cellular uptake of LNPs. From left to right: siRNA-LNP dose was 1.0 µg/mL;siRNA-LNP dose was 0.1 µg/mL; untreated control. From top to bottom: 24 hourtransfection with 1.0 µg/mL ApoE; 4 hour transfection with 1.0 µg/mL ApoE; 24 hourtransfection without ApoE. Transfection for 4 hours without ApoE showed negligibleuptake and is not shown. Experimental replicates yielded similar results (Figure 4.2).864B4B fysults2"Single-cell measurement of LNP uptake by flow cytometry •  Measure the distribution of LNP uptake in single cells •  Measure the kinetics of uptake and Apolipoprotein E mediated uptake •  6-well plate format •  Untreated, 0.1 ug/mL, 1.0 ug/mL siRNA •  1.0 ug/mL ApoE Figure 4.2: Replicates for ApoE-dependent LNP uptake experiment in hESC. His-togram displays flow cytometry measurements of DiI (FL2), where DiI fluorescencewas used as a surrogate for cellular uptake of LNPs (same experiment as Figure 4.1).Untreated cells are shown in dark green; cells receiving a siRNA-LNP dose of 0.1µg/mL shown in orange and light green; cells receiving a siRNA-LNP dose of 1.0µg/mL shown in blue and red.874B4B fysultsfluorescent if LNPs are bound to the cell membrane, or if contained within endosomes.The flow cytometry detection of this dye should therefore be seen as a metric of LNPassociation with the cell, and may overrepresent the number of transfected cells.Previous studies in the brain[190] and liver[175] have shown that LNP uptakeis facilitated by adsorption of endogenously produced apolipoprotein E (ApoE) tothe LNPs, which can then be recognized by scavenging receptors and low-densitylipoprotein receptors. This takes advantage of the property that these LNPs mimicnatural cholesterol. We hypothesized that adding exogenous ApoE to hESC cultureswould similarly increase the transfection efficiency. We measured the uptake of LNPsafter 4 hours and 24 hours of transfection, measuring approximately 15% of hESCscontaining the delivered dye after a dose of 1.0 µg/mL for 24 hours (Figure 4.1). Thismeasurement represents a lower bound on positive uptake, as the threshold for callinga cell positive is set to exclude all of the untransfected negative control cell population.Furthermore, sufficient dye must be delivered to be detected by flow cytometry. Theaddition of 1.0 µg/mL ApoE increased the transfection efficiency, to 67.3%, a ∼4.4Xincrease (Figure 4.1). In the presence of ApoE, cells positive for DiI continued toincrease from 32.9% at 4 hours of transfection to 67.3% at 24 hours. However, itis notable that almost half of the transfection occurs within the first 4 hours. Thisexperiment delivered non-targeting siRNA (siNT) (i.e. not complementary to anymRNA), and viability for all treated conditions matched the untreated control. Thesetransfection results encourage future testing of LNP transfection in primary hESC.ICICG sigcV KnoxkyofinWe next assessed our ability to knock down gene expression by LNP delivery ofsiRNA. Using the previously developed microfluidic device for performing single-celldigital PCR measurements[191] (from Chapter 3), we measured levels of the tar-geted transcript within single-cells undergoing a treatment of siRNA-LNP. We treatedhESC cultures for 48 hours with a low dose (0.1 µg/mL), and high dose (1.0 µg/mL)of siRNA targeting HPRT. For a low-abundance transcript, HPRT, we observed adistribution of transcript abundance (Figure 4.3) ranging from 0 to 25 per individualcell, with a mean of 8.8 copies per cell (s.d. 5.6 copies, N=35). Upon treatment,we witnessed the HPRT distribution shift towards lower copy number, with less than10% of cells containing more than 10 transcripts, and a small percentage containingno detected HPRT copies (mean 6.6 copies per cell, s.d. 5.0 copies, N=48). Increasing884B4B fysults0 3 6 9 12 15 18 21 2400. mRNA copies per cellFraction of cell population (N=35)CA1S hESC Untreated0 3 6 9 12 15 18 21 2400. mRNA copies per cellFraction of cell population (N=48)CA1S hESC with 0.1 µg/mL siHPRT0 3 6 9 12 15 18 21 2400. mRNA copies per cellFraction of cell population (N=99)CA1S hESC with 1.0 µg/mL siHPRTFigure 4.3: siNRA knockdown of HPRT in human embryonic stem cells. Histogramsshowing the distribution of measured HPRT transcripts measured in single CA1ShESC undergoing 48 hours of transfection at dose of 1.0 µg/mL (bottom), 0.1 µg/mL(middle), and an untreated control (top).894B4B fysults500 1000 1500 2000 2500 3000 3500 4000 4500 500000. mRNA copies per cellFraction of cell population (N=44)siNT (0.1 µg/mL, 24 hrs)500 1000 1500 2000 2500 3000 3500 4000 4500 500000. mRNA copies per cellFraction of cell population (N=48)siNT (1.0 µg/mL, 24 hrs)500 1000 1500 2000 2500 3000 3500 4000 4500 500000. mRNA copies per cellFraction of cell population (N=48)siGAPDH (0.1 µg/mL, 24 hrs)500 1000 1500 2000 2500 3000 3500 4000 4500 500000. mRNA copies per cellFraction of cell population (N=45)siGAPDH (1.0 µg/mL, 24 hrs)Figure 4.4: siNRA knockdown of GAPDH in human embryonic stem cells. His-tograms showing the distribution of measured GAPDH transcripts measured in singleCA1S hESC undergoing 24 hours of transfection at dose of 1.0 µg/mL, 0.1 µg/mL,and non-targeting siRNA (siNT) control.904B4B fysults0 100 200 300 400 500 600 700 80000. mRNA copies per cellFraction of cell population (N=45)Lowïend of GAPDH expression following 1.0 µg/mL siGAPDH for 24 hrsFigure 4.5: Low-end of GAPDH expression following 1.0 µg/mL siGAPDH for 24 hrs.This is a detailed view of the histogram information from Figure 4.4.the siRNA dose to 1.0 µg/mL further shifted the transcript abundance to a distribu-tion between 0 and 10 copies per single cell, with over 80% of the cells containing 0to 4 copies (mean 2.3 copies per cell, s.d. 2.6 copies, N=99). Trypan blue stainingshowed similar viability between untreated and treated cultures (∼97%). Followinga similar dosing treatment, but only transfecting over 24 hours, we measured knock-down of a highly expressed gene, GAPDH (Figure 4.4). Here we observed GAPDHconstitutively expressed in all cells (mean 2,430 copies per cell, s.d. 781 copes) .Similar to knockdown of HPRT, GAPDH expression exhibited partial knockdown at0.1 µg/mL siGAPDH. With a 24-hour dose of 1.0 µg/mL siGAPDH, we measuredsignificant knockdown of GAPDH (mean 362 copies per cell, s.d. 693 copies), with∼89% (40/45) of cells containing less than 750 copies (see Figure 4.5 for detailedhistogram information). Remarkably, the same dose that is unable to completelyeliminate the tens of HPRT transcripts is able to knock down hundreds to thousandsof GAPDH transcripts in each cell (but still not completely eliminate it, with thelowest measure of 8 copies per cell). The difficulty in completely eliminating such‘housekeeping’ genes may reflect the speed at which such genes are transcribed com-914B4B fysultspared to interacting with siRNA. Furthermore, nascent mRNA residing in the nucleusis inaccessible to siRNA, and it is possible that fragments of the degraded transcriptsare still detectable RT-qPCR. Treated cultures remained viable, however significantlyhigher doses targeting GAPDH will lead to toxicity, as GAPDH is needed in glycol-ysis. Increasing a dose of non-targeting siRNA from 0.1 µg/mL to 1.0 µg/mL didnot significantly alter the abundance of GAPDH, indicating the reduction of GAPDHunder targeted siRNA was the result of RNA interference and not due to the particlesby themselves. The abundance of GAPDH in cells treated with non-targeting siRNAfit a log-normal distribution similar to untreated cells.In addition to the adherent CA1S hESCs, we measured the dose-response re-lationship in a suspension cell line, K562 (Figure 4.6A). K562 cells are a humanerythroleukemia cell line we have previously characterized GAPDH with single cellRT-qPCR and dPCR. Using single-cell dPCR, we measured a mean of 1,453 (s.d.519) GAPDH transcripts 20 hours after a dose of 6 ng/mL (see Methods section fordetails). Increasing the dose to 60 ng/mL did not change expression significantly(mean 1,511 copies, s.d. 509 copies), but dramatic reduction in transcript levels(mean 582 copies, s.d. 421 copies) occurs with a dose of 600 ng/mL, suggesting apossible threshold of LNP delivery (i.e. siRNA abundance) before knockdown occurs(Figure 4.6A). With a siRNA dose of 1.2 µg/mL, GAPDH levels were reduced toa mean of 153 copies (s.d. 132 copies), a ∼90% knockdown of constitutive levels.Notably however, GAPDH was still detected in all cells, with the minimum detectedbeing 1 copy. In a separate experiment repeating some of these doses, and looking ata higher dose of 2.5 µg/mL siRNA, results showed a greater knockdown effect (Fig-ure 4.6B). For this transfection, cells were incubated with siRNA-LNPs for 24 hours.This extra time may have contributed to the greater knockdown, but we previouslyobserved the majority of uptake occurs within 4 hours. It should also be noted thatthis difference may arise from using a different manufacturing batch of siRNA-LNPs,or from a different thawed vial (and passage number) of K562 cells. In this experi-ment, knockdown of GAPDH was observed with a siRNA dose of 60 ng/mL, yieldinga mean of 1,063 copies of GAPDH per cell (s.d. 504, N=48). Increasing the siRNAdose to 600 ng/mL resulted in significant knockdown, with a mean average of 132GAPDH transcripts measured per single cell (s.d. 206, N=46). The minimum num-ber of GAPDH transcripts detected was 5. With a treatment of 1.2 µg/mL siRNA,GAPDH levels dropped to an average of 61 copies per cell (s.d. 64), with 1/47 cells924B4B fysults10ï2 10ï1 100 10105001000150020002500siRNA dose (µg/mL)GAPDH mRNA copies per cell10ï3 10ï2 10ï1 100 10105001000150020002500siGAPDH dose (µg/mL)GAPDH mRNA copies per cellA B Figure 4.6: Single-cell dose-response curve for siGAPDH treatment in K562 cells.Coloured marks represent single-cell digital PCR measurements of GAPDH mRNAcopies. Black circles indicate mean. Experiments in A and B show variability inresponse to siRNA treatment over biological replicates (and different LNP batches).934B4B fysultsFigure 4.7: Single-cell flow cytometry measurement of LNP delivery and eGFP ex-pression in K562. Flow cytometry profiles show eGFP signal vs. DiI, where eGFPfluorescence was used as a reporter for translation from delivered mRNA, and DiIwas a reporter of LNP uptake. From left to right: untreated control sample; 24 hourtransfection of 100 ng/mL eGFP mRNA-LNP; 24 hour transfection of 500 ng/mLeGFP mRNA-LNP. The fluorescence signal threshold to call positive eGFP and DiIdetection was set such that it excluded signal from untreated cells.without detected GAPDH. Increasing this dose to 2.5 µg/mL siRNA resulted in 3/49cells without detected GAPDH. For this final dose, the mean (78 copies/cell) andstandard deviation (124 copies/cell) are increased due to the presence of seemingoutliers (Figure 4.6B). By measuring gene expression knockdown in individual cells,these results reveal the variability in expression knockdown, and show that there isa distribution of knockdown levels rather than a binary response with some cellsexhibiting complete knockdown while others are unaffected.ICICH mgcV DeliveryHaving demonstrated the ability to knock down gene expression with siRNA delivery,we sought to expand our capabilities of manipulating gene expression to includemediating, or knocking-in, gene expression. We chose to test our transfection systemin suspension and adherent cell culture conditions, using K562 as our suspensioncondition. As proof-of-concept, we delivered mRNA encoding for green fluorescentprotein (eGFP), the detection of which was useful for reporting successful translationof delivered mRNA. Using flow cytometry, we were able to measure LNP transfection(or at least cellular localization) by detection of delivered dye fluorescence (Figure4.7). We were then able to compare the single-cell flow cytometry measurements oftransfection and protein abundance, to the delivered mRNA levels in single cells as944B4B fysults0 1 10 20 30 40 50 60 70 80 90 100 110 120 13000. mRNA copies per cellFraction of K562 cellpopulation (N = 81)100 ng/mL eGFP mRNA LNP 24 hour incubation0 1 100 200 300 400 500 600 700 80000. mRNA copies per cellFraction of K562 cellpopulation (N = 73)500 ng/mL eGFP mRNA LNP 24 hour incubationFigure 4.8: Single-cell measurement of mRNA delivered by LNP in suspension culture(K562). (Top) Histogram showing the distribution of eGFP mRNA measured bysingle-cell digital PCR in K562 cells administered 100 ng/mL for 24 hours. (Bottom)Histogram showing the distribution of eGFP mRNA measured by single-cell digitalPCR in K562 cells administered 500 ng/mL for 24 hours.954B4B fysultsmeasured by microfluidic digital PCR (Figure 4.8). In K562, a dose of 500 ng/mLeGFP mRNA-LNP for 24 hours resulted in ∼94% of cells reporting eGFP positive,and over 99% of cells were positive for the particle dye (Figure 4.7). A dose of 100ng/mL for 24 hours showed reduced penetrance, with ∼91% of cells transfected butonly 35% of cells positive for eGFP. Using microfluidic single-cell dPCR, we confirmedthe presence of eGFP mRNA in 100% (N=73) of the K562 cells transfected withthe 500 ng/mL dose (Figure 4.8), with a mean abundance of 201 copies (s.d. 150copies) per cell. Interestingly, reducing the mRNA-LNP dose 5X, to 100 ng/mL,resulted in more than a 5X reduction in eGFP transcripts measured in single-cells(mean 23 copies, s.d. 23 copies). This supra-linear relationship between dose andmeasured mRNA levels may be due the speed at which eGFP mRNA is degraded inthe cell. Inspection with microscopy revealed that LNP uptake is rapid within thefirst 4 hours, and the low-dose culture condition may rapidly become LNP-depletedcompared to the high-dose condition. Compared to the flow cytometry measurementof 91% cells transfected at the dose of 100 ng/mL, we detected eGFP mRNA in 98%(79/81) of cells, albeit at low levels. Measurement after 24 hours was chosen as such‘overnight’ treatments are common in cell culture experiments, and also to give timefor eGFP translation. The doubling time for K562 cells is approximately 24 hours,which may result in dilution of eGFP mRNA or protein. Analysis of flow cytometrymeasurements using different cell population gates revealed no significant correlationbetween cell size and LNP dye or eGFP expression (Figure 4.9).For adherent culture conditions, we transfected a human neonatal foreskin fi-Figure 4.9 (following pugy): Effect of different population gates on flow cytometrymeasurement of LNP delivery and eGFP expression in K562 cells. (A) A generousgate is displayed in the flow cytometry profile of K562 cells showing granularity (sidescatter, SSC) and size (forward scatter, FSC). (B,C) The corresponding measure-ments of eGFP (FL1) vs. LNP dye (FL2) for 100 ng/mL and 500 ng/mL RNAdosages, as in Figure 4.7. (D) A relatively small gate at the densest region of theK562 cell population and the measured eGFP and LNP dye for 100 ng/mL (E) and500 ng/mL (F) eGFP mRNA doses. (G) The smallest and largest cells are gated andresultant eGFP vs. dye profiles are shown in H and I respectively (500 ng/mL RNAdose). (J) Reverse-gating of cells with the highest intensity of eGFP and dye yieldsa subpopulation (K) with size and granularity similar to the generous gate in A. (L)Plot of eGFP (FL1) vs. size (FSC) shows that eGFP intensity is not proportionalcell size.964B4B fysults974B4B fysultsFigure 4.10: Single-cell flow cytometry measurement of LNP delivery and eGFPexpression in BJ. Flow cytometry profiles show eGFP signal vs. DiI, where eGFPfluorescence was used as a reporter for translation from delivered mRNA, and DiIwas a reporter of LNP uptake. From left to right: untreated control sample; 24 hourtransfection of 100 ng/mL eGFP mRNA-LNP; 24h hour transfection of 500 ng/mLeGFP mRNA-LNP. The fluorescence signal threshold to call positive eGFP and DiIdetection was set such that it excluded signal from untreated cells.broblast cell line, BJ. BJ cells were selected as they have been extensively used intransfection studies to produce iPS cells, and could provide a basis for comparisonwith our LNP transfection technique. In contrast to K562 cultures, BJ cells trans-fected with the same dose showed markedly less transfection and eGFP. Cells dosedat 100 ng/mL were ∼71% positive for delivered dye, but no eGFP was detected(Figure 4.10). This corresponded to 76% (42/55) of cells containing eGFP mRNA(Figure 4.11), but RNA levels were low, with most cells containing between 1 and 20copies (mean 6 copies, s.d. 7.5 copies). Increasing the dose to 500 ng/mL resultedin 90% of cells positive for fluorescent dye, but only 29% positive for eGFP (Figure4.10). Single-cell RT-dPCR revealed this population to contain eGFP mRNA in allcells (N=86), with abundance ranging from tens to hundreds of transcripts (mean213 copies, s.d. 220 copies). Uptake and transcript abundance is somewhat higherin K562 cells, and this may simply be due to geometric constraints in adherent cellculture, where only some of the cell surface is exposed to LNPs, as compared to cellsin suspension. The differences in protein abundance suggest a physiological differencebetween K562 and BJ cells, and further experiments are needed to elucidate if thisdifference is due to mRNA accessibility (e.g. endosome trafficking of LNP cargo), ordifferences in translation kinetics. Taken together, these results could be interpretedas suggesting a minimum of ∼20 eGFP transcripts per cell are required to detect984B4B fysults0 1 10 20 30 40 5000. mRNA copies per cellFraction of BJ cellpopulation (N = 55)100 ng/mL eGFP mRNA LNP 24 hour incubation0 1 100 200 300 400 500 600 700 800 900 1,00000. mRNA copies per cellFraction of BJ cellpopulation (N = 86)500 ng/mL eGFP mRNA LNP 24 hour incubationFigure 4.11: Single-cell measurement of mRNA delivered by LNP in adherent culture(BJ). (Top) Histogram showing the distribution of eGFP mRNA measured by single-cell digital PCR in BJ cells administered 100 ng/mL eGFP mRNA-LNP for 24 hours.(Bottom) Histogram showing the distribution of eGFP mRNA measured by single-celldigital PCR in BJ cells administered 500 ng/mL mRNA-LNP for 24 hours.994B4B fysultsInducing Gene Expression with mRNA Delivery  20 SUB9KITS™ Inducing Gene Expression with mRNA Delivery GFP (green), Dye from LNPs (red), BJ fibroblasts Dosed at 0 Hours with 500 ng/mL GFP mRNA LNPs (3mL culture; 6-well format) Figure 4.12: Inducing gene expression with mRNA delivery. Time-lapse microscopyimaging of DiI (red) and eGFP (green) fluorescence in BJ cells dosed with 500 ng/mLeGFP mRNA (at 0 hours). Images suffered from high background fluorescence andlow signal-to-noise, and were used gauge the onset of eGFP signal, while flow cytom-etry was used to quantify eGFP and DiI in cells.eGFP at this time point. One concern with adapting this LNP system to mRNAdelivery is that the long mRNA strands (relative to the short siRNA duplexes, ∼22bp) may be more susceptible loss of function due to a number of factors includingfragmentation from shear stress during formulation, secondary structure, or endo-somal trafficking. Also, we do not know what fraction of synthesized RNA is fulllength. The result showing 94% of K562 cells positive for eGFP indicates functionaltranscripts, however further verification of the state of delivered mRNA is needed.Visual inspection of cells undergoing transfection (Figure 4.12) showed rapid andwidespread uptake of LNPs during the first 4 hours, as reported by DiI fluorescencelocalized within cells. eGFP was visible after ∼10 hours, and remained in cells forover 72 hours. In contrast to the diffuse eGFP signal throughout the cytoplasm,DiI signal was often punctuate, suggesting the dye was contained within intracellularmembranes such as endosomes. Notably, DiI particularly accumulated in two oppositepoints on either side of the nucleus. Cells exhibiting DiI uptake in endosomes, butwithout eGFP expression, may reflect endosomal recycling or sequestering of mRNAas a major factor limiting functional transfection efficiencies[186].1004B4B fysultsFigure 4.13: Inducing gene expression with delivery two different mRNAs. Proof-of-concept demonstration of delivering multiple RNAs, showing false-colour overlay offluorscence from eGFP (green) and mCherry (red), resulting in diffuse orange signalin cytoplasm.1014B4B fysultsDay 2 Day 5 Day 8 Day 11 Day 14ControlEMEMPluritonFigure 4.14: Time-course measurement of mRNA-LNP performance in BJ cells withdaily doses for two weeks. Flow cytometry plots showing eGFP signal vs. DiI asa measure of protein abundance and particle uptake (respectively). The ‘Control’series refers to untreated (i.e. not dosed with mRNA-LNP) BJ cells in EMEM (10%FBS). The ‘EMEM’ series refers to BJ cells treated daily with mRNA-LNP, in normalculture conditions of EMEM with 10% FBS (and ApoE). The ‘Pluriton’ series refersto BJ cells treated daily with mRNA-LNP in reprogramming culture conditions ofPluriton medium and matrigel (no serum, but with ApoE).1024B4B fysults0 1 50 100 150 200 250 300 350 400 450 50000. ofcells (N=39)0 1 500 1,000 1,500 2,000 2,500 3,00000. ofcells (N=76)0 1 500 1,000 1,500 2,000 2,500 3,00000. ofcells (N=92)0 1 500 1,000 1,500 2,000 2,500 3,00000. ofcells (N=29)0 1 500 1,000 1,500 2,000 2,500 3,00000.51eGFP mRNA per cellFraction ofcells (N=84)Figure 4.15: Distribution of delivered mRNA in BJ cells with daily doses for twoweeks. Histograms display results from single-cell digital PCR measurement of eGFPtranscripts from cells undergoing daily transfections in EMEM with 10% FBS. Thetime-course is represented from top-to-bottom: day 2, day 5, day 8, day 11, day 14,and corresponds to measurements on cells from the ‘EMEM’ series in Figure 4.14.1034B4B fysultsOne of the applications of mRNA delivery of keen interest is the reprogramming ofcells, in particular the creation of iPS cells that can then be directed in differentiationto desired tissue types (e.g. neurons). Reprogramming fibroblasts to pluripotencywas recently demonstrated with the transfection synthetic mRNA encoding the tran-scription factors Klf4, c-Myc, Oct4, and Sox2 (originally achieved by Yamanaka andcolleagues by enforced expression)[188]. As demonstration of delivering more thanone mRNA at the same time, we simultaneously delivered two different mRNAs, en-coding for eGFP and mCherry, and observed fluorescence from both proteins after adose of 500 ng/mL each (Figure 4.13). Reprogramming protocols with RNA trans-fections typically involve repeated high doses over multiple weeks, and we sought toevaluate the use of LNPs for repeated mRNA delivery as required for reprogram-ming. For this experiment, a transfection protocol was adapted from a commercialiPS reprogramming kit (Stemgent). Further modifications by Warren et al.[188] toperform reprogramming in xeno-free conditions, and without feeder-cells, were alsoincorporated. BJ cells were transfected by adding LNPs to the culture after eachdaily media change for two weeks (details in Methods section). Cells were seededin 6-well plates at low-density, 50,000 cells per well, and were passaged on day 6at ∼80% confluence. Over the first four days, the dosage was increasingly rampedup from 25%, 50%, 75%, to 100% of the final RNA dose (400 ng/mL, 2 mL), so asnot to overdose the low cell concentration[188]. Cells were analyzed on day 2, 5, 8,11, and 14 by measuring LNP uptake (dye) and eGFP with flow cytometry (Figure4.14), and mRNA abundance with microfluidic single-cell RT-dPCR (Figure 4.15).Trypan blue staining showed all culture conditions supported over 95% viable cells.Although we previously showed incomplete penetrance of eGFP in BJ cells after 24hours of transfection, this experiment demonstrated that nearly complete penetrancewas possible after longer time with repeated doses. Figure 4.14 shows ∼98% of cellspositive for LNP dye by day 2, and over 99% from day 5 to 14. The fraction of cellsreporting eGFP increased from 82.5% of cells on day 2, to ∼98% by day 5. Thefraction of eGFP-positive BJ cells was maintained above 95% from day 5 to day 14.Single-cell RT-dPCR measurements in Figure 4.15 show mean abundance of deliveredeGFP mRNA to be 196 copies (s.d. 212) on day 2, rising to upwards of thousandsof copies per cell for days 5 through 11 (day 11 mean 1,002 copies, s.d. 803). Day14 shows reduced mRNA levels compared to previous measurements, perhaps due toincreased cell numbers. This result demonstrated the performance of these LNPs in1044BIB Diswussiona long-term, high-dose, transfection protocol.Interestingly, substitution of our normal BJ medium (EMEM, 10% FBS) formedium designed for mRNA reprogramming (Pluriton, Stemgent) resulted in dra-matically less transfection and translation. This reprogramming condition is xeno-free, absent of any serum additions, and replaces the feeder-cells normally used inStemgents protocol with matrigel to provide a substrate and extracellular matrix.Both media conditions were supplemented with 1 µg/mL apoE to facilitate uptake,although the FBS also contains lipoproteins. The reprogramming culture conditionsresulted in ∼87% of cells containing DiI on day 2, increasing to ∼99% for day 5through 14. Despite a high-percentage of cells reporting positive particle uptake,only ∼20% of cells were eGFP-positive (with day 11 being an outlier at 43%). Instark contrast to the hundreds or even thousands of mRNA found by day 5 in thenormal culture, cells in the reprogramming conditions continued to have only tensto low hundreds of mRNA present throughout the experiment. Notably, the growthrate of cells in the reprogramming condition matched normal culture conditions (withand without transfection) for the first 7 days (Figure 4.17), but showed dramaticallyreduced proliferation after cell passage (while viability remained over 95%). It ispossible the difference in plating or growth after cell passage may be ameliorated bythe addition of ROCK inhibitor (e.g. Y27632, Stemgent), but it is interesting to notethat cells in EMEM/FBS conditions did not require this. The difference in transfec-tion and translation efficiencies between these two conditions underscores the needto characterize (and optimize) LNP performance in application-specific conditions.Further experiments are needed to determine whether the differences are due to themedia, lack of serum, or the matrigel.IC5 DisxussionThe study of gene expression requires the ability to both mxtsurx gene expression,and pxrturu the system. LNPs, particularly the recently developed particles used inthis study, represent an effective means for manipulating gene expression through thedelivery of RNA to the cytoplasm. Combining this ability to perturb gene expressionwith precise transcript measurements by high-throughput single-cell digital PCR pro-vides an opportunity to study gene expression at the single cell level. To apply thistechnology properly, LNP transfection behaviour must first be characterized in order1054BIB Diswussion0 1 50 100 150 200 250 300 350 400 450 50000.51Fraction ofcells (N=42)0 1 50 100 150 200 250 300 350 400 450 50000. ofcells (N=72)0 1 50 100 150 200 250 300 350 400 450 50000. ofcells (N=82)0 1 50 100 150 200 250 300 350 400 450 50000. ofcells (N=54)0 1 50 100 150 200 250 300 350 400 450 50000. mRNA copies per cellFraction ofcells (N=68)Figure 4.16: Distribution of delivered mRNA in cells under mock reprogrammingconditions. Histograms display results from single-cell digital PCR measurement ofeGFP transcripts from cells undergoing daily transfections in Pluriton medium witha substrate coated with matrigel (no serum). The time-course is represented fromtop-to-bottom: day 2, day 5, day 8, day 11, day 14, and corresponds to measurementson cells from the ’Pluriton’ series in Figure 4.14.1064BIB Diswussion0 2 4 6 8 10 12 14 16 18051015Time (days)Total number of cells per well [× 105 ]Figure 4.17: Comparison of BJ cell numbers in different media conditions undergoingdaily transfections. Cells were passaged by splitting 1/6, on day 7. Red circlesrepresent cells undergoing daily transfections in EMEM with 10% FBS; Blue trianglesrepresent cells undergoing daily transfections in Pluriton with matrigel; Black squaresrepresent untreated cells in EMEM with 10% FBS.1074BIB Diswussionto provide an informed basis for protocol development and optimization. In this studywe have begun the important work of characterizing the use of LNP RNA deliveryto manipulate gene expression. Through the use of single-cell analysis techniquessuch as microfluidic digital PCR and flow cytometry, we investigated the kinetics,precision, variability and single cell distribution of LNP transfection performance.Our results indicate that transfection begins rapidly after addition of LNPs toculture media, with uptake efficiency greatly enhanced in the presence of ApoE. Thehigh transfection efficiency seen through DiI measurements and siRNA knockdowneffects show that 1.0 µg/mL ApoE supplement is sufficient to boost LNP uptake,resulting in a 4X increase in our serum-free hESC conditions. Further optimizationmay be possible by increasing ApoE concentrations, as seen in neuronal cultures[190].ApoE facilitation is convenient as ApoE is endogenously synthesized by astrocytesin the brain or neuronal in vitro cutlure conditions, and lipoproteins are also highlyabundant in common culture conditions with medium containing fetal bovine serum.Furthermore, the results presented in this study show that siRNA-LNPs efficientlyknockdown expression of targeted genes, and mRNA-LNPs effectively give rise tofunctional proteins, at dosage levels that do not lead to any observed toxicity.One of the questions motivating single cell analysis of RNA delivery was whenbulk PCR shows partial knockdown, say 50%, does that reflect all cells with par-tial knockdown, or some of the cells with 100% knockdown and some with 0%? ThesiRNA knockdown experiments presented here showed a shifting distribution of targetmRNA with increasing siRNA dose, rather than an all-or-nothing change of expres-sion. This study demonstrated the potency of siRNA-LNPs, showing that a doseof 1.0 µg/mL siRNA was sufficient to reduce GAPDH mRNA levels by ∼10-fold,eliminating over one thousand transcripts per cell when measured 24 hours after ad-ministration. Complete elimination of such an actively expressed gene is challengingas newly transcribed mRNA is inaccessible in the nucleus, and RT-PCR may detectRNA fragments that no longer give rise to functional protein.Single-cell analysis is particularly well suited to elucidating the relationship ofco-occurrences that would otherwise be masked in the averaging of bulk analysis.In this study we use flow cytometry to measure the relationship between LNP up-take (by DiI) and eGFP protein abundance. We compare this to single-cell digitalPCR measurements of the delivered RNA, using cells from the same sample. Thistranscript information revealed that different cell types (K562 compared to BJ) with1084B6B Conwlusionsimilar amounts of mRNA, gave rise to different amounts of translated protein. Thisunderscores the fact that although LNPs effectively deliver mRNA to cells, physio-logical differences in translation activity between different cell types can still resultin different protein levels.LNPs are an attractive transfection technology for using RNA to reprogram cellsinto induced pluripotent stem cells. As opposed to conventional techniques which relyon viruses and genomic inserts, RNA reprogramming is non-integrative and ‘footprintfree’, requiring no clean-up phase compared to DNA vectors or RNA viruses. Fur-thermore, RNA delivery allows for direct control of the timing and abundance ofgene expression. Here we show that LNPs provide efficient delivery to a variety ofcell types, and that LNPs potentially offer a convenient, self-contained approach toreprogramming somatic cells to pluripotency. Previous studies have delivered mRNAencoding for the Yamanaka factors (Oct3/4, Sox2, Klf4, c-Myc) and shown conversionefficiencies of 1-4%[188]. Single-cell digital RT-PCR measurements of delivered RNAcan provide insight into this low efficiency by revealing the abundance and stochiome-try of the different mRNAs at different points during transfection for reprogramming.Our results showing repeated delivery of mRNA to sustain eGFP in over 95% of cellsfor 2 weeks with no apparent toxicity open many possibilites for maintaining highlevels of ectopic proteins in cells, with scientific and therapeutic applications.ICK ConxlusionHere we have shown the utility of single-cell analysis in characterizing the performanceof LNP RNA delivery in a variety of in vitro conditions. LNP uptake was facilitated bythe presence of ApoE, and transfection was particularly efficient in situations whereculture medium contains lipoprotein-rich serum. LNPs demonstrated potent siRNAknockdown of a highly expressed gene, and mediated functional protein expressionby mRNA delivery. We have shown the efficacy of these particles in the contextof adherent and suspension cell cultures, and under different transfection protocols.Single-cell digital PCR measurement of RNA abundance complements microscopyand flow cytometry techniques, and provides insight into the RNA levels requiredto see functional effects. Importantly, measuring the variability in the number oftranscripts delivered to each cell elucidates the heterogeneity and extent to whichone can control transfection. By examining the cell-to-cell variability, kinetics, and1094B6B Conwlusionefficiency of using this lipid nanoparticle technology for nucleic acid delivery, weprovide an informed basis for optimizing the in vitro manipulation of gene expressionin cells. This single-cell approach may be extended to wide variety of delivery systems,contributing to the development of research tools and therapeutic strategies.110Chvpter 5ConxlusionIn this thesis I set out to address the need for techniques enabling high-throughputsingle-cell gene expression analysis. Through developing scalable microfluidic tech-nology integrating components to perform cell isolation, lysis, reverse-transcription,and final measurement by real-time or digital PCR, this research has contributed tothe tools available for precisely measuring transcripts in single cells. The utility ofthis technology is demonstrated in measuring the cell-to-cell variability of a variety oftranscripts, and miRNAs, in tissues ranging from cell lines to stem cells and primarybreast cancer samples. Additionally, this measurement tool is used to assess anothertool - a transfection reagent - in order to help understand and improve its use anddevelopment.5CF Contriwution to KnofileygeThis research has made an impact in three main areas: engineering technology, tech-nology transfer, and biological findings.Although microfluidic technology is a relatively young field, a long-sought goal inmicrofluidics research has been the development of integrated technology for scalableanalysis of transcription in single cells. This project achieved this goal through devel-opment of the first integrated system for cell isolation and RNA analysis in a scalablefashion. The core functionality established here provides the foundation from whicha variety of on-chip single-cell analyses can be developed. In particular, this standsto impact the burgeoning field of single-cell DNA and RNA sequencing. Throughcombining scalable cell processing with high-density digital PCR, this project fur-ther advanced the state-of-the-art in terms of combined throughput and precisionfor single-cell transcript analysis. It was also interesting to see that this device waswell suited to characterizing the performance of other technologies, informing thedevelopment and use of lipid nanoparticles for RNA delivery.In addition to dissemination of this research through publications and conference111IBFB Futury fywommynxutionspresentations, aspects of the technology presented in this thesis will be commercializedto facilitate its widespread adoption in research (and potentially clinical) settings.The microfluidic systems for single-cell trapping followed by RT-qPCR or dPCR arethe subject of a patent filing (PCT/CA2011/000612) which has been licensed toFluidigm Corporation. Fluidigm has since released a commercial product related tothis technology for single-cell trapping, lysis, reverse-transcription, and recovery fordownstream analysis. This product represents the first major commercial foray intointegrated microfluidic single-cell devices, and has enabled single-cell studies for userswithout microfluidic experience.Finally, although this project is focused on technology development, significantbiological insights have been made through single-cell transcription measurements.At the time of their publication, the measurements reported in this thesis were someof the first large number (hundreds) of single-cell transcript measurements explicitlyreported in copy number (as opposed to relative vales such as ∆Ct). In addition to ob-serving how tightly regulated certain miRNAs are, or revealing a bi-stable switch-likebehaviour in the co-regulation of a miRNA and its target mRNA during differentia-tion, we precisely quantify and characterize the variability that exists even in ‘house-keeping’ genes (i.e. GAPDH). These measurements of GAPDH provide an excellentpoint of comparison for other quantification technologies (e.g. sequencing). Thedemonstration of single-cell measurement of SNVs in a primary breast cancer sampleserves as an example of the sort of assay that may reveal important heterogeneitywithin clinical cell populations (e.g. the distribution and co-occurrence of muta-tions). The quantification of the oncogenic fusion transcript BCR-ABL demonstratesthe potential utility of using digital PCR in the monitoring of minimum residual dis-ease. Finally, the evaluation of mRNA and siRNA delivery efficacy may lead to newreagents or methods for research, potentially contributing to improved therapeuticstrategies.5CG Future gexommenyvtions5CGCF Efltenying bixrouiyix hingleBCell VnvlysisAs stated before, the core functionality established here provides the foundation fromwhich a variety of on-chip single-cell and molecular biology protocols can be devel-112IBFB Futury fywommynxutionsoped.Colleagues in Dr. Carl Hansen’s lab are currently pursuing such applicationsas DNA and RNA sequencing from libraries prepared on integrated single-cell mi-crofluidic devices. Fluidic architecture for recovery of single-cell reaction products isreadily incorporable, and represents the logical extension of our microfluidic systemsto enable further down-stream analysis. Such functionality has been demonstratedin the C1 device from Fluidigm, which is increasingly being used for single-cell anal-ysis by DNA and RNA sequencing, as well as epigenetic analysis. In the context ofthis thesis, single-cell whole-transcriptome analysis would be useful to look at globalexpression changes (e.g. off-target effects) in consequence to siRNA treatment. Per-forming whole genome amplification or whole transcriptome amplification in limitingdilution ‘digital’ arrays has also been shown to reduce bias and contributions fromcontaminating nucleic acids[192].Although the microfluidic devices presented in this thesis (and the C1) makestrides in addressing the bottle-neck between cell isolation and nucleic acid processing,issues with cell isolation remain one of the greatest limitations with this technology.More routine still are studies coupling FACS with RT-qPCR arrays (e.g. FluidigmBiomark). The challenge in microfluidic isolation of single-cells with physical celltraps is that many cell-types of interest do not readily form single-cell suspensions.This can result in clumping and clogging. Primary cells, thawed samples, or tissuesrequiring aggressive dissociation protocols, often suffer from viability and clumpingissues. Damaged cell membranes may lead to RNA contents leaking and alteringdetectable abundance. Treating cell suspensions with DNase (to digest DNA clump-ing), or suspending them in different solutions (e.g in presence of EDTA, or Ficoll)may ameliorate clumping of cells. Physical cell traps may be tailored to capturecells of different size ranges, if a universal cell trap is difficult to engineer. A pos-sible approach could be to encapsulate single cells in uniform-sized droplets or gelsbefore isolating with uniform traps. Upstream manipulation of cells could also in-clude nucleus or cytoplasm isolation (or fixing cells). Ultimately, for many tissues,researchers and clinicians would like to able to perform single-cell analysis while pre-serving spatial information and molecular imaging or techniques such as laser-capturemicrodisection[193] or Tomo-Seq[194] remain advantageous over our microfluidic ap-proach to single-cell isolation[195]. Having a linear array of cell traps also increasesthe risk of a cell clog as the number of traps is scaled from tens to hundreds and po-113IBFB Futury fywommynxutionstentially thousands, suggesting the need for improved cell bypass lanes or a parallelloading strategy.The device designs presented for single cell RT-qPCR or digital PCR are limitedin the number of different transcripts that can be simultaneously measured to thenumber of fluorescent probes that can be spectrally distinguished. This optical mul-tiplexing is generally limited to 3 or 4 colours (and a passive reference dye) whenusing commercially available hydrolysis probes (e.g. TaqMan) and common filters.This limitation can be overcome by splitting the template reaction into separatedassay chambers (that can be of the same probe colour), thereby spatially multiplex-ing the reaction. Pre-amplification of the pooled cDNA may be required to preserverepresentation when partitioning into multiple assay chambers.There are a number of further types of analyses the microfluidic single-cell digitalPCR device could be adapted to accomodate. Future work could look at the ability tomeasure genomic copy number alterations, potentially with a pre-amplification step.In the course of this work I explored the potential use of microfluidic digital PCR forprotein counting, by using a proximity ligation assay[196]. In the proximity ligationassay, two different antibodies bind to a target protein of interest. Each antibody isattached to a short DNA strand, and when the two antibodies are in close proximity(as in when bound to the same protein) the DNA strands can be ligated together.This newly combined DNA strand then forms the starting template for PCR amplifi-cation, and could be quantified with digital PCR. Performing the proximity ligationassay in a microfluidic single-cell digital PCR device may be a direct way to combineand correlate mRNA measurements with protein abundance[152]. The current mi-crofluidic device for single-cell digital PCR could be easily modified with additionalchambers to perform antibody binding and ligation prior to dPCR. One of the limita-tions of the proximity ligation assay is the likelyhood of false-positive PCR reactionsdue to ligation of DNA from antibodies in close proximity but not bound to thesame protein. This probability is reduced by diluting the bound antibodies prior toligation, but false-positives remain. Although single-cell digital PCR does not elimi-nate this issue, it does offer a chance to quantify the base level false-positives, whichcan then be subtracted from single cell measurements. Ultimately, this microfluidicapproach to single-cell protein quantification lacks the throughput and multiplexingcapabilities of recently developed single-cell mass cytometry techniques[197].Beyond genetic and transcript measurements, enzymatic assays (e.g. measuring114IBFB Futury fywommynxutionstelomerase activity) could be incorporated into a similar microfluidic architecture.Cell trapping upstream of analysis could be further utilized to briefly expose the cellsto stimuli, before measuring responses in the transcriptome. A more thorough imageanalysis of isolated cells prior to lysis could also be used to compare morphologieswith intracellular contents. The trap and assay architecture of the devices couldbe used to expose cells to chemical stimuli (e.g. drugs, siRNA, antibodies) prior torapidly lysing and measuring gene expression (or other molecules of interest). Themicrofluidic architecture for single-cell molecular biology protocols could also be inte-grated downstream of cell culture arrays, enabling automated genomic or transcriptanalysis after testing different culture conditions and treatments.5CGCG Further Eflperiments vny Vpplixvtions for aipiycvnopvrtixle DeliveryThe use of LNPs to mediate gene expression should be studied further. The ApoEfacilitation of LNP transfection warrants further investigation into the mechanism ofreceptor mediated transport. What is the receptor density of different cells? Cancells be cholesterol starved to stimulate production of more LDL receptors, and doesthis increase LNP uptake? Future experiments could investigate the effect of in-cubating the LNPs with ApoE prior to transfection. One potential concern is theaddition of ApoE may cause unwanted environmental stimuli in cases where cells aredelicately sensitive to culture conditions (e.g. ESC, iPS conversion). In such situa-tions it may be worth exploring alternative protocols such as transfection during cellpassaging (adherent cells could be temporarily incubated in suspension with a shake-flask), or pre-mixing LNPs with the plate coating substrate (e.g. Matrigel). Themicrofluidic single-cell digital PCR device is well suited to investigate the time-scalesinvolved in lipid nanoparticle uptake. For example, single-cells may be isolated in celltraps within the microfluidic device prior to briefly washing them with a solution ofmRNA-containing lipid nanoparticles. Cells can then be processed within minutes (orhours) to examine the kinetics of transfection. Using fluorescence microscopy, futureexperiments could also measure DiI and eGFP in cells trapped in the microfluidicdevice and directly correlate dye and protein abundance with measured transcriptabundance within each cell. Protein abundance could be further quantified withsingle-cell proximity ligation assay[152].115IBFB Futury fywommynxutionsIt would be useful to measure both the delivered siRNA as well as the targetedmRNA in each cell to properly assess the dose-response relationship. This can bemeasured in the microfluidic single-cell digital PCR device by designing a stem-loopRT primer and TaqMan qPCR assay[59] allowing optical multiplexing[129, 191]. Fur-ther experiments testing increased doses and/or transfection times should investigatethe feasibility of completely suppressing a highly expressed gene such as GAPDHby siRNA, as RNAi machinery may not be able to keep up with constitutively ac-tive transcription. This also raises the question of at what point does siRNA po-tency become RISC abundance-limited. siRNA is already known to cause off-targetdown-regulation, and by overwhelming the RNAi machinery exogenous siRNA mayfurther disrupt gene expression regulation by out-competing natural microRNAs forincorporation in RISC[198]. Single-cell RNA sequencing would be useful to asses thegenome-wide effect of siRNA transfection. RISC-incorporated RNA could be assessedfollowing antibody pull-down of RISC, and compared with free miRNAs and siRNAs.In characterizing the performance of LNPs, we sought to understand how wellwe could control transfection efficiency and how precisely we could control the abun-dance of delivered RNA to each cell. We have started to explore this with differentdoses of RNA, and different transfection protocols. However future pulse-chase typeexperiments are needed to explore the effect of different dosing regimes, for example,the affect of short vs. long transfection times (pulse-width), and high- compared tolow-dose (pulse amplitude) on the precision of delivered RNA. Pulse-chase experi-ments will also be useful in measuring the time between transfection and translation(e.g. eGFP), and the duration of sustained effect. For example, is it possible tosustain low levels of transcripts delivered to the entire cell population?In investigating how well delivered RNA abundance can be controlled, it would beuseful to test delivery of two different mRNAs at different ratios of abundance, andsee if this ratio is preserved in delivery. While different RNAs could be combined andencapsulated at different relative abundance during LNP formulation, keeping themindividually packaged allows for more control in how they are combined and used. Bytransfecting with equal amounts of mRNA encoding for two different colours, but witha similar length and sequence (e.g. eGFP and YFP), one could look at the variationin delivery compared to variation in functionality (i.e. translation to protein). Itmay also be possible to attenuate the noise, or long tails of low-abundance (or high-abundance) proteins by co-delivering microRNA or siRNA targeting the delivered116IBGB Finul fymurksmRNA.Chapter 4 focused on the PCR measurement of RNA however a similar analysiscould be extended to characterize the delivery or affect of other cargo, such as DNA,proteins, or small molecules. In particular, LNPs could be harnessed to deliver guideDNA strands for CRISPR gene editing. This can be used for applications rangingfrom editing mutations in genes related to cancer, to antiviral treatments. For exam-ple, the CRISPR/Cas9 system can be adapted for antiviral treatment in human cellsby specifically targeting the genomes of viral infections[199, 200]. Measurement ofdelivered DNA is not subject to reverse transcription efficiencies, and could provide auseful surrogate measurement of RNA (or other species) delivery as it could be con-tinually used without re-designing new primers each time the LNP cargo is changed.This study focused on characterizing the performance of a particular LNP reagent,however the single-cell analysis methodologies employed here, namely flow cytometryand single-cell digital PCR, could be used to asses the performance of other trans-fection methods. Ultimately, the LNP system used in this study should be comparedto alternative transfection reagents such as the commonly used Lipofectamine (LifeTechnologies).5CH Finvl gemvrksRecent years have seen rapid development of single-cell and fluidic technologies, andthe work in this thesis may ultimately be integrated into larger work-flows and de-vices. In particular, micro-wells, integrated electronic components and sensors inmicrofluidic devices, droplet dispensers for printing arrays of reagents and cells areall exciting technologies pushing the capabilities for single-cell analysis. Advancesin 3D printing may be used to construct microfluidic devices (either directly, or asmoulds), potentially democratizing microfluidic prototyping.The development and demonstration of microfluidic single-cell analysis reportedhere, and in other applications ranging from human haplotyping and drug discovery,to stem cell development and cancer progression, should lead to widespread adoptionof this technology.117Wiwliogrvphy[1] Y. Sasai and E. M. De Robertis. Ectodermal patterning in vertebrate embryos.Dxv Uiol, 182(1):5–20, 1997.[2] M. B. Elowitz, A. J. Levine, E. D. Siggia, and P. S. Swain. Stochastic geneexpression in a single cell. fvixnvx, 297(5584):1183–6, 2002.[3] A. Raj, C. S. Peskin, D. Tranchina, D. Y. Vargas, and S. Tyagi. Stochasticmrna synthesis in mammalian cells. cLof Uiol, 4(10):e309, 2006.[4] M. J. Ravitz and C. E. Wenner. Cyclin-dependent kinase regulation during g1phase and cell cycle regulation by tgf-beta. Tdv Vtnvxr exs, 71:165–207, 1997.[5] M. Anger, W. A. Kues, J. Klima, M. Mielenz, M. Kubelka, J. Motlik, M. Esner,P. Dvorak, J. W. Carnwath, and H. Niemann. Cell cycle dependent expressionof plk1 in synchronized porcine fetal fibroblasts. Mol exprod Dxv, 65(3):245–53,2003.[6] J. Lukas, C. Lukas, and J. Bartek. Mammalian cell cycle checkpoints: signallingpathways and their organization in space and time. DaT exptir (Tmst), 3(8-9):997–1007, 2004.[7] J. A. Smith and L. Martin. Do cells cycle? crov attl Tvtd fvi h f T,70(4):1263–7, 1973.[8] W. A. Kues, M. Anger, J. W. Carnwath, D. Paul, J. Motlik, and H. Niemann.Cell cycle synchronization of porcine fetal fibroblasts: effects of serum depriva-tion and reversible cell cycle inhibitors. Uiol exprod, 62(2):412–9, 2000.[9] M. S. Rhyu, L. Y. Jan, and Y. N. Jan. Asymmetric distribution of numbprotein during division of the sensory organ precursor cell confers distinct fatesto daughter cells. Vxll, 76(3):477–91, 1994.118Vivliogruphfl[10] O. H. Yilmaz, M. J. Kiel, and S. J. Morrison. Slam family markers are conservedamong hematopoietic stem cells from old and reconstituted mice and markedlyincrease their purity. Ulood, 107(3):924–30, 2006.[11] D. Douer, A. M. Levin, R. S. Sparkes, I. Fabian, M. Sparkes, M. J. Cline, andH. P. Koeffler. Chronic myelocytic leukaemia: a pluripotent haemopoietic cellis involved in the malignant clone. Ur J Htxmttol, 49(4):615–9, 1981.[12] M. R. Stratton, P. J. Campbell, and P. A. Futreal. The cancer genome. atturx,458(7239):719–24, 2009.[13] L. Fink, G. Kwapiszewska, J. Wilhelm, and R. M. Bohle. Laser-microdissectionfor cell type- and compartment-specific analyses on genomic and proteomiclevel. Exp goxivol ctthol, 57 Suppl 2:25–9, 2006.[14] L. Zhang, W. Zhou, V. E. Velculescu, S. E. Kern, R. H. Hruban, S. R. Hamilton,B. Vogelstein, and K. W. Kinzler. Gene expression profiles in normal and cancercells. fvixnvx, 276(5316):1268–72, 1997.[15] K. Ruan, X. Fang, and G. Ouyang. Micrornas: novel regulators in the hallmarksof human cancer. Vtnvxr Lxtt, 285(2):116–26, 2009.[16] D. P. Bartel. Micrornas: genomics, biogenesis, mechanism, and function. Vxll,116(2):281–97, 2004.[17] B. Zhang, X. Pan, G. P. Cobb, and T. A. Anderson. micrornas as oncogenesand tumor suppressors. Dxv Uiol, 302(1):1–12, 2007.[18] C. Z. Chen, L. Li, H. F. Lodish, and D. P. Bartel. Micrornas modulatehematopoietic lineage differentiation. fvixnvx, 303(5654):83–6, 2004.[19] G. A. Calin, C. G. Liu, C. Sevignani, M. Ferracin, N. Felli, C. D. Dumitru,M. Shimizu, A. Cimmino, S. Zupo, M. Dono, M. L. Dell’Aquila, H. Alder,L. Rassenti, T. J. Kipps, F. Bullrich, M. Negrini, and C. M. Croce. Micrornaprofiling reveals distinct signatures in b cell chronic lymphocytic leukemias.crov attl Tvtd fvi h f T, 101(32):11755–60, 2004.[20] O. I. Petriv, F. Kuchenbauer, A. D. Delaney, V. Lecault, A. White, D. Kent,L. Marmolejo, M. Heuser, T. Berg, M. Copley, J. Ruschmann, S. Sekulovic,119VivliogruphflC. Benz, E. Kuroda, V. Ho, F. Antignano, T. Halim, V. Giambra, G. Krystal,C. J. Takei, A. P. Weng, J. Piret, C. Eaves, M. A. Marra, R. K. Humphries, andC. L. Hansen. Comprehensive microrna expression profiling of the hematopoi-etic hierarchy. crov attl Tvtd fvi h f T, 107(35):15443–8, 2010.[21] A. M. Femino, F. S. Fay, K. Fogarty, and R. H. Singer. Visualization of singlerna transcripts in situ. fvixnvx, 280(5363):585–90, 1998.[22] J. M. Levsky, S. M. Shenoy, R. C. Pezo, and R. H. Singer. Single-cell geneexpression profiling. fvixnvx, 297(5582):836–40, 2002.[23] A. Raj, P. van den Bogaard, S. A. Rifkin, A. van Oudenaarden, and S. Tyagi.Imaging individual mrna molecules using multiple singly labeled probes. attMxthods, 5(10):877–9, 2008.[24] H. Maamar, A. Raj, and D. Dubnau. Noise in gene expression determines cellfate in bacillus subtilis. fvixnvx, 317(5837):526–9, 2007.[25] A. Raj, S. A. Rifkin, E. Andersen, and A. van Oudenaarden. Variability in geneexpression underlies incomplete penetrance. atturx, 463(7283):913–8, 2010.[26] C. Larsson, I. Grundberg, O. Soderberg, and M. Nilsson. In situ detection andgenotyping of individual mrna molecules. att Mxthods, 7(5):395–7, 2010.[27] C. Larsson, J. Koch, A. Nygren, G. Janssen, A. K. Raap, U. Landegren, andM. Nilsson. In situ genotyping individual dna molecules by target-primedrolling-circle amplification of padlock probes. att Mxthods, 1(3):227–32, 2004.[28] A. Lagunavicius, E. Merkiene, Z. Kiveryte, A. Savaneviciute, V. Zimbaite-Ruskuliene, T. Radzvilavicius, and A. Janulaitis. Novel application of phi29dna polymerase: Rna detection and analysis in vitro and in situ by targetrna-primed rca. eaT, 15(5):765–71, 2009.[29] Brian Munsky, Gregor Neuert, and Alexander van Oudenaarden. Using geneexpression noise to understand gene regulation. fvixnvx, 336:183–7, 2012.[30] F. Tang, C. Barbacioru, Y. Wang, E. Nordman, C. Lee, N. Xu, X. Wang,J. Bodeau, B. B. Tuch, A. Siddiqui, K. Lao, and M. A. Surani. mrna-seqwhole-transcriptome analysis of a single cell. att Mxthods, 6(5):377–82, 2009.120Vivliogruphfl[31] R. Morin, M. Bainbridge, A. Fejes, M. Hirst, M. Krzywinski, T. Pugh, H. Mc-Donald, R. Varhol, S. Jones, and M. Marra. Profiling the hela s3 transcrip-tome using randomly primed cdna and massively parallel short-read sequencing.Uiotxvhniquxs, 45(1):81–94, 2008.[32] S. P. Shah, R. D. Morin, J. Khattra, L. Prentice, T. Pugh, A. Burleigh,A. Delaney, K. Gelmon, R. Guliany, J. Senz, C. Steidl, R. A. Holt, S. Jones,M. Sun, G. Leung, R. Moore, T. Severson, G. A. Taylor, A. E. Teschendorff,K. Tse, G. Turashvili, R. Varhol, R. L. Warren, P. Watson, Y. Zhao, C. Caldas,D. Huntsman, M. Hirst, M. A. Marra, and S. Aparicio. Mutational evolutionin a lobular breast tumour profiled at single nucleotide resolution. atturx,461(7265):809–13, 2009.[33] F. Tang, C. Barbacioru, S. Bao, C. Lee, E. Nordman, X. Wang, K. Lao, andM. A. Surani. Tracing the derivation of embryonic stem cells from the innercell mass by single-cell rna-seq analysis. Vxll ftxm Vxll, 6(5):468–78, 2010.[34] D. Ramskold, S. Luo, Y. C. Wang, R. Li, Q. Deng, O. R. Faridani, G. A. Daniels,I. Khrebtukova, J. F. Loring, L. C. Laurent, G. P. Schroth, and R. Sandberg.Full-length mrna-seq from single-cell levels of rna and individual circulatingtumor cells. att Uiotxvhnol, 30(8):777–82, 2012.[35] S. Islam, U. Kjallquist, A. Moliner, P. Zajac, J. B. Fan, P. Lonnerberg, andS. Linnarsson. Highly multiplexed and strand-specific single-cell rna 5’ endsequencing. att crotov, 7(5):813–28, 2012.[36] S. Islam, U. Kjallquist, A. Moliner, P. Zajac, J. B. Fan, P. Lonnerberg, andS. Linnarsson. Characterization of the single-cell transcriptional landscape byhighly multiplex rna-seq. Gxnomx exs, 21(7):1160–7, 2011.[37] X. Adiconis, D. Borges-Rivera, R. Satija, D. S. DeLuca, M. A. Busby, A. M.Berlin, A. Sivachenko, D. A. Thompson, A. Wysoker, T. Fennell, A. Gnirke,N. Pochet, A. Regev, and J. Z. Levin. Comparative analysis of rna sequencingmethods for degraded or low-input samples. att Mxthods, 10(7):623–9, 2013.[38] A. R. Wu, N. F. Neff, T. Kalisky, P. Dalerba, B. Treutlein, M. E. Rothen-berg, F. M. Mburu, G. L. Mantalas, S. Sim, M. F. Clarke, and S. R. Quake.121VivliogruphflQuantitative assessment of single-cell rna-sequencing methods. att Mxthods,11(1):41–6, 2014.[39] D. Hebenstreit. Methods, challenges and potentials of single cell rna-seq. Uiol-ozy (Utsxl), 1(3):658–67, 2012.[40] M. Bengtsson, A. Stahlberg, P. Rorsman, and M. Kubista. Gene expressionprofiling in single cells from the pancreatic islets of langerhans reveals lognormaldistribution of mrna levels. Gxnomx exs, 15(10):1388–92, 2005.[41] M. K. Chiang and D. A. Melton. Single-cell transcript analysis of pancreasdevelopment. Dxv Vxll, 4(3):383–93, 2003.[42] N. Faumont, T. Al Saati, P. Brousset, C. Offer, G. Delsol, and F. Meggetto.Demonstration by single-cell pcr that reed–sternberg cells and bystander b lym-phocytes are infected by different epstein–barr virus strains in hodgkin’s disease.J Gxn iirol, 82(Pt 5):1169–74, 2001.[43] S. Hahn, X. Y. Zhong, C. Troeger, R. Burgemeister, K. Gloning, and W. Holz-greve. Current applications of single-cell pcr. Vxll Mol Liyx fvi, 57(1):96–105,2000.[44] K. M. Keays, G. P. Owens, A. M. Ritchie, D. H. Gilden, and M. P. Burgoon.Laser capture microdissection and single-cell rt-pcr without rna purification. JImmunol Mxthods, 302(1-2):90–8, 2005.[45] A. Stahlberg and M. Bengtsson. Single-cell gene expression profiling usingreverse transcription quantitative real-time pcr. Mxthods, 2010.[46] K. Taniguchi, T. Kajiyama, and H. Kambara. Quantitative analysis of geneexpression in a single cell by qpcr. att Mxthods, 6(7):503–6, 2009.[47] I. Tietjen, J. M. Rihel, Y. Cao, G. Koentges, L. Zakhary, and C. Dulac. Single-cell transcriptional analysis of neuronal progenitors. axuron, 38(2):161–75,2003.[48] R. K. Saiki, D. H. Gelfand, S. Stoffel, S. J. Scharf, R. Higuchi, G. T. Horn,K. B. Mullis, and H. A. Erlich. Primer-directed enzymatic amplification of dnawith a thermostable dna polymerase. fvixnvx, 239(4839):487–91, 1988.122Vivliogruphfl[49] T. B. Morrison, J. J. Weis, and C. T. Wittwer. Quantification of low-copytranscripts by continuous sybr green i monitoring during amplification. Uiotxvh-niquxs, 24(6):954–8, 960, 962, 1998.[50] S. Tyagi and F. R. Kramer. Molecular beacons: probes that fluoresce uponhybridization. att Uiotxvhnol, 14(3):303–8, 1996.[51] D. Whitcombe, J. Theaker, S. P. Guy, T. Brown, and S. Little. Detection ofpcr products using self-probing amplicons and fluorescence. att Uiotxvhnol,17(8):804–7, 1999.[52] M. C. Vicens, A. Sen, A. Vanderlaan, T. J. Drake, and W. Tan. Investigationof molecular beacon aptamer-based bioassay for platelet-derived growth factordetection. Vhxmuiovhxm, 6(5):900–7, 2005.[53] C. A. Heid, J. Stevens, K. J. Livak, and P. M. Williams. Real time quantitativepcr. Gxnomx exs, 6(10):986–94, 1996.[54] K J Livak, S J Flood, J Marmaro, W Giusti, and K Deetz. Oligonucleotideswith fluorescent dyes at opposite ends provide a quenched probe system usefulfor detecting pcr product and nucleic acid hybridization. cVe Mxthods Tppl,4:357–362, 1995.[55] S. A. Bustin. Quantification of mrna using real-time reverse transcription pcr(rt-pcr): trends and problems. J Mol Endovrinol, 29(1):23–39, 2002.[56] M. W. Pfaffl. A new mathematical model for relative quantification in real-timert-pcr. auvlxiv Tvids exs, 29(9):e45, 2001.[57] H. D. VanGuilder, K. E. Vrana, and W. M. Freeman. Twenty-five years ofquantitative pcr for gene expression analysis. Uiotxvhniquxs, 44(5):619–26, 2008.[58] M. Bengtsson, M. Hemberg, P. Rorsman, and A. Stahlberg. Quantification ofmrna in single cells and modelling of rt-qpcr induced noise. UMV Mol Uiol,9:63, 2008.[59] C. Chen, D. A. Ridzon, A. J. Broomer, Z. Zhou, D. H. Lee, J. T. Nguyen,M. Barbisin, N. L. Xu, V. R. Mahuvakar, M. R. Andersen, K. Q. Lao, K. J.Livak, and K. J. Guegler. Real-time quantification of micrornas by stem-looprt-pcr. auvlxiv Tvids exs, 33(20):e179, 2005.123Vivliogruphfl[60] F. Tang, P. Hajkova, S. C. Barton, K. Lao, and M. A. Surani. Microrna expres-sion profiling of single whole embryonic stem cells. auvlxiv Tvids exs, 34(2):e9,2006.[61] F. Tang, P. Hajkova, S. C. Barton, D. O’Carroll, C. Lee, K. Lao, and M. A.Surani. 220-plex microrna expression profile of a single cell. att crotov,1(3):1154–9, 2006.[62] L. Warren, D. Bryder, I. L. Weissman, and S. R. Quake. Transcription factorprofiling in individual hematopoietic progenitors by digital rt-pcr. crov attlTvtd fvi h f T, 103(47):17807–12, 2006.[63] B. J. Hindson, K. D. Ness, D. A. Masquelier, P. Belgrader, N. J. Heredia, A. J.Makarewicz, I. J. Bright, M. Y. Lucero, A. L. Hiddessen, T. C. Legler, T. K.Kitano, M. R. Hodel, J. F. Petersen, P. W. Wyatt, E. R. Steenblock, P. H.Shah, L. J. Bousse, C. B. Troup, J. C. Mellen, D. K. Wittmann, N. G. Erndt,T. H. Cauley, R. T. Koehler, A. P. So, S. Dube, K. A. Rose, L. Montesclaros,S. Wang, D. P. Stumbo, S. P. Hodges, S. Romine, F. P. Milanovich, H. E. White,J. F. Regan, G. A. Karlin-Neumann, C. M. Hindson, S. Saxonov, and B. W.Colston. High-throughput droplet digital pcr system for absolute quantitationof dna copy number. Tntl Vhxm, 83(22):8604–10, 2011.[64] L. B. Pinheiro, V. A. Coleman, C. M. Hindson, J. Herrmann, B. J. Hindson,S. Bhat, and K. R. Emslie. Evaluation of a droplet digital polymerase chainreaction format for dna copy number quantification. Tntl Vhxm, 84(2):1003–11,2012.[65] M. A. Unger, H. P. Chou, T. Thorsen, A. Scherer, and S. R. Quake. Mono-lithic microfabricated valves and pumps by multilayer soft lithography. fvixnvx,288(5463):113–6, 2000.[66] T. Thorsen, S. J. Maerkl, and S. R. Quake. Microfluidic large-scale integration.fvixnvx, 298(5593):580–4, 2002.[67] C. E. Sims and N. L. Allbritton. Analysis of single mammalian cells on-chip.Ltu Vhip, 7(4):423–40, 2007.124Vivliogruphfl[68] V. Lecault, A. K. White, A. Singhal, and C. L. Hansen. Microfluidic singlecell analysis: from promise to practice. Vurr bpin Vhxm Uiol, 16(3-4):381–90,2012.[69] J. S. Marcus, W. F. Anderson, and S. R. Quake. Parallel picoliter rt-pcr assaysusing microfluidics. Tntl Vhxm, 78(3):956–8, 2006.[70] J R Rettig and A Folch. Large-scale single-cell trapping and imaging usingmicrowell arrays. Tntl Vhxm, 77:5628–5634, 2005.[71] D Di Carlo, N Aghdam, and L P Lee. Single-cell enzyme concentrations, ki-netics, and inhibition analysis using high-density hydrodynamic cell isolationarrays. Tntl Vhxm, 78:4925–4930, 2006.[72] D. Di Carlo, L. Y. Wu, and L. P. Lee. Dynamic single cell culture array. LtuVhip, 6(11):1445–9, 2006.[73] P. J. Lee, P. J. Hung, V. M. Rao, and L. P. Lee. Nanoliter scale microbioreactorarray for quantitative cell biology. Uiotxvhnol Uioxnz, 94(1):5–14, 2006.[74] X. Li and P. C. Li. Microfluidic selection and retention of a single cardiac my-ocyte, on-chip dye loading, cell contraction by chemical stimulation, and quanti-tative fluorescent analysis of intracellular calcium. Tntl Vhxm, 77(14):4315–22,2005.[75] A. R. Wheeler, W. R. Throndset, R. J. Whelan, A. M. Leach, R. N. Zare,Y. H. Liao, K. Farrell, I. D. Manger, and A. Daridon. Microfluidic device forsingle-cell analysis. Tntl Vhxm, 75(14):3581–6, 2003.[76] S. Kobel, A. Valero, J. Latt, P. Renaud, and M. Lutolf. Optimization ofmicrofluidic single cell trapping for long-term on-chip culture. Ltu Vhip,10(7):857–63, 2010.[77] P. C. Li, L. de Camprieu, J. Cai, and M. Sangar. Transport, retention andfluorescent measurement of single biological cells studied in microfluidic chips.Ltu Vhip, 4(3):174–80, 2004.[78] N. Bontoux, L. Dauphinot, T. Vitalis, V. Studer, Y. Chen, J. Rossier, andM. C. Potier. Integrating whole transcriptome assays on a lab-on-a-chip forsingle cell gene profiling. Ltu Vhip, 8(3):443–50, 2008.125Vivliogruphfl[79] J. S. Marcus, W. F. Anderson, and S. R. Quake. Microfluidic single-cell mrnaisolation and analysis. Tntl Vhxm, 78(9):3084–9, 2006.[80] H. Wu, A. Wheeler, and R. N. Zare. Chemical cytometry on a picoliter-scaleintegrated microfluidic chip. crov attl Tvtd fvi h f T, 101(35):12809–13, 2004.[81] A. M. Skelley, O. Kirak, H. Suh, R. Jaenisch, and J. Voldman. Microfluidiccontrol of cell pairing and fusion. att Mxthods, 6(2):147–52, 2009.[82] S. Koster, F. E. Angile, H. Duan, J. J. Agresti, A. Wintner, C. Schmitz, A. C.Rowat, C. A. Merten, D. Pisignano, A. D. Griffiths, and D. A. Weitz. Drop-based microfluidic devices for encapsulation of single cells. Ltu Vhip, 8(7):1110–5, 2008.[83] J. F. Edd, D. Di Carlo, K. J. Humphry, S. Koster, D. Irimia, D. A. Weitz, andM. Toner. Controlled encapsulation of single-cells into monodisperse picolitredrops. Ltu Vhip, 8(8):1262–4, 2008.[84] J. Voldman, M. L. Gray, M. Toner, and M. A. Schmidt. A microfabrication-based dynamic array cytometer. Tntl Vhxm, 74(16):3984–90, 2002.[85] A Ashkin. The study of cells by optical trapping and manipulation of livingcells using infrared laser beams. TfGfU Uull, 4:133–146, 1991.[86] A Ashkin, J M Dziedzic, and T Yamane. Optical trapping and manipulationof single cells using infrared laser beams. atturx, 330:769–771, 1987.[87] H. Zhang and K. K. Liu. Optical tweezers for single cells. J e fov Intxrytvx,5(24):671–90, 2008.[88] Z. C. Landry, S. J. Giovanonni, S. R. Quake, and P. C. Blainey. Optofluidiccell selection from complex microbial communities for single-genome analysis.Mxthods Enflymol, 531:61–90, 2013.[89] J. F. Zhong, Y. Chen, J. S. Marcus, A. Scherer, S. R. Quake, C. R. Taylor, andL. P. Weiner. A microfluidic processor for gene expression profiling of singlehuman embryonic stem cells. Ltu Vhip, 8(1):68–74, 2008.126Vivliogruphfl[90] G. Guo, M. Huss, G. Q. Tong, C. Wang, L. Li Sun, N. D. Clarke, and P. Robson.Resolution of cell fate decisions revealed by single-cell gene expression analysisfrom zygote to blastocyst. Dxv Vxll, 18(4):675–85, 2010.[91] K.H. Kazim H Narsinh, Ning Sun, Veronica Sanchez-Freire, Andrew S A.S.Lee, Patricia Almeida, Shijun Hu, Taha Jan, Kitchener D K.D. Wilson, DeniseLeong, Jarrett Rosenberg, others, Mylene Yao, Robert C Robbins, and Joseph CWu. Single cell transcriptional profiling reveals heterogeneity of human inducedpluripotent stem cells. Gxnx Exprxssion, 121:1217, 2011.[92] L. Flatz, R. Roychoudhuri, M. Honda, A. Filali-Mouhim, J. P. Goulet,N. Kettaf, M. Lin, M. Roederer, E. K. Haddad, R. P. Sekaly, and G. J.Nabel. Single-cell gene-expression profiling reveals qualitatively distinct cd8t cells elicited by different gene-based vaccines. crov attl Tvtd fvi h f T,108(14):5724–9, 2011.[93] K. J. Livak, Q. F. Wills, A. J. Tipping, K. Datta, R. Mittal, A. J. Goldson,D. W. Sexton, and C. C. Holmes. Methods for qpcr gene expression profilingapplied to 1440 lymphoblastoid single cells. Mxthods, 59(1):71–9, 2013.[94] P. Dalerba, T. Kalisky, D. Sahoo, P. S. Rajendran, M. E. Rothenberg, A. A.Leyrat, S. Sim, J. Okamoto, D. M. Johnston, D. Qian, M. Zabala, J. Bueno,N. F. Neff, J. Wang, A. A. Shelton, B. Visser, S. Hisamori, Y. Shimono,M. van de Wetering, H. Clevers, M. F. Clarke, and S. R. Quake. Single-cell dis-section of transcriptional heterogeneity in human colon tumors. att Uiotxvhnol,29(12):1120–7, 2011.[95] M. Diehn, R. W. Cho, N. A. Lobo, T. Kalisky, M. J. Dorie, A. N. Kulp, D. Qian,J. S. Lam, L. E. Ailles, M. Wong, B. Joshua, M. J. Kaplan, I. Wapnir, F. M.Dirbas, G. Somlo, C. Garberoglio, B. Paz, J. Shen, S. K. Lau, S. R. Quake,J. M. Brown, I. L. Weissman, and M. F. Clarke. Association of reactive oxygenspecies levels and radioresistance in cancer stem cells. atturx, 458(7239):780–3,2009.[96] N. M. Toriello, E. S. Douglas, N. Thaitrong, S. C. Hsiao, M. B. Francis, C. R.Bertozzi, and R. A. Mathies. Integrated microfluidic bioprocessor for single-cellgene expression analysis. crov attl Tvtd fvi h f T, 105(51):20173–8, 2008.127Vivliogruphfl[97] P. Mary, L. Dauphinot, N. Bois, M. C. Potier, V. Studer, and P. Tabeling.Analysis of gene expression at the single-cell level using microdroplet-basedmicrofluidic technology. Uiomivrouidivs, 5(2):24109, 2011.[98] Y. Gong, A. O. Ogunniyi, and J. C. Love. Massively parallel detection of geneexpression in single cells using subnanolitre wells. Ltu Vhip, 10(18):2334–7,2010.[99] R. Sandberg. Entering the era of single-cell transcriptomics in biology andmedicine. att Mxthods, 11(1):22–4, 2014.[100] B. Treutlein, D. G. Brownfield, A. R. Wu, N. F. Neff, G. L. Mantalas, F. H.Espinoza, T. J. Desai, M. A. Krasnow, and S. R. Quake. Reconstructing lin-eage hierarchies of the distal lung epithelium using single-cell rna-seq. atturx,509(7500):371–5, 2014.[101] A. Zeisel, A. B. Munoz-Manchado, S. Codeluppi, P. Lonnerberg, G. La Manno,A. Jureus, S. Marques, H. Munguba, L. He, C. Betsholtz, C. Rolny, G. Castelo-Branco, J. Hjerling-Leffler, and S. Linnarsson. Brain structure. cell types inthe mouse cortex and hippocampus revealed by single-cell rna-seq. fvixnvx,347(6226):1138–42, 2015.[102] A. M. Streets, X. Zhang, C. Cao, Y. Pang, X. Wu, L. Xiong, L. Yang, Y. Fu,L. Zhao, F. Tang, and Y. Huang. Microfluidic single-cell whole-transcriptomesequencing. crov attl Tvtd fvi h f T, 111(19):7048–53, 2014.[103] 2nd Wadsworth, M. H., T. K. Hughes, and A. K. Shalek. Marrying microfluidicsand microwells for parallel, high-throughput single-cell genomics. Gxnomx Uiol,16(1):129, 2015.[104] S. Bose, Z. Wan, A. Carr, A. H. Rizvi, G. Vieira, D. Pe’er, and P. A. Sims.Scalable microfluidics for single-cell rna printing and sequencing. Gxnomx Uiol,16(1):120, 2015.[105] M. T. Guo, A. Rotem, J. A. Heyman, and D. A. Weitz. Droplet microfluidicsfor high-throughput biological assays. Ltu Vhip, 12(12):2146–55, 2012.[106] E. Z. Macosko, A. Basu, R. Satija, J. Nemesh, K. Shekhar, M. Goldman,I. Tirosh, A. R. Bialas, N. Kamitaki, E. M. Martersteck, J. J. Trombetta, D. A.128VivliogruphflWeitz, J. R. Sanes, A. K. Shalek, A. Regev, and S. A. McCarroll. Highly paral-lel genome-wide expression profiling of individual cells using nanoliter droplets.Vxll, 161(5):1202–14, 2015.[107] A. M. Klein, L. Mazutis, I. Akartuna, N. Tallapragada, A. Veres, V. Li,L. Peshkin, D. A. Weitz, and M. W. Kirschner. Droplet barcoding for single-celltranscriptomics applied to embryonic stem cells. Vxll, 161(5):1187–201, 2015.[108] J. H. Lee, E. R. Daugharthy, J. Scheiman, R. Kalhor, J. L. Yang, T. C. Ferrante,R. Terry, S. S. Jeanty, C. Li, R. Amamoto, D. T. Peters, B. M. Turczyk,A. H. Marblestone, S. A. Inverso, A. Bernard, P. Mali, X. Rios, J. Aach, andG. M. Church. Highly multiplexed subcellular rna sequencing in situ. fvixnvx,343(6177):1360–3, 2014.[109] J. H. Lee, E. R. Daugharthy, J. Scheiman, R. Kalhor, T. C. Ferrante, R. Terry,B. M. Turczyk, J. L. Yang, H. S. Lee, J. Aach, K. Zhang, and G. M. Church.Fluorescent in situ sequencing (fisseq) of rna for gene expression profiling inintact cells and tissues. att crotov, 10(3):442–58, 2015.[110] R. N. Zare and S. Kim. Microfluidic platforms for single-cell analysis. Tnnuexv Uiomxd Enz, 12:187–201, 2010.[111] H G Schulze, S O Konorov, N J Caron, J M Piret, M W Blades, and R F Turner.Assessing differentiation status of human embryonic stem cells noninvasivelyusing raman microspectroscopy. Tntl Vhxm, 82:5020–5027, 2010.[112] O. Adewumi, B. Aflatoonian, L. Ahrlund-Richter, M. Amit, P. W. Andrews,G. Beighton, P. A. Bello, N. Benvenisty, L. S. Berry, S. Bevan, B. Blum,J. Brooking, K. G. Chen, A. B. Choo, G. A. Churchill, M. Corbel, I. Dam-janov, J. S. Draper, P. Dvorak, K. Emanuelsson, R. A. Fleck, A. Ford, K. Ger-tow, M. Gertsenstein, P. J. Gokhale, R. S. Hamilton, A. Hampl, L. E. Healy,O. Hovatta, J. Hyllner, M. P. Imreh, J. Itskovitz-Eldor, J. Jackson, J. L. John-son, M. Jones, K. Kee, B. L. King, B. B. Knowles, M. Lako, F. Lebrin, B. S.Mallon, D. Manning, Y. Mayshar, R. D. McKay, A. E. Michalska, M. Mikkola,M. Mileikovsky, S. L. Minger, H. D. Moore, C. L. Mummery, A. Nagy, N. Nakat-suji, C. M. O’Brien, S. K. Oh, C. Olsson, T. Otonkoski, K. Y. Park, R. Passier,H. Patel, M. Patel, R. Pedersen, M. F. Pera, M. S. Piekarczyk, R. A. Pera, B. E.129VivliogruphflReubinoff, A. J. Robins, J. Rossant, P. Rugg-Gunn, T. C. Schulz, H. Semb, E. S.Sherrer, H. Siemen, G. N. Stacey, M. Stojkovic, H. Suemori, J. Szatkiewicz,T. Turetsky, T. Tuuri, S. van den Brink, K. Vintersten, S. Vuoristo, D. Ward,T. A. Weaver, L. A. Young, and W. Zhang. Characterization of human em-bryonic stem cell lines by the international stem cell initiative. att Uiotxvhnol,25(7):803–16, 2007.[113] T. E. Ludwig, V. Bergendahl, M. E. Levenstein, J. Yu, M. D. Probasco, andJ. A. Thomson. Feeder-independent culture of human embryonic stem cells.att Mxthods, 3(8):637–46, 2006.[114] B. Tinland, A. Pluen, J. Sturm, and G. Weill. Persistence length of single-stranded dna. Mtvromolxvulxs, 30(19):5763–5765, 1997.[115] J. Lu, G. Getz, E. A. Miska, E. Alvarez-Saavedra, J. Lamb, D. Peck, A. Sweet-Cordero, B. L. Ebert, R. H. Mak, A. A. Ferrando, J. R. Downing, T. Jacks,H. R. Horvitz, and T. R. Golub. Microrna expression profiles classify humancancers. atturx, 435(7043):834–8, 2005.[116] B. B. Lozzio, C. B. Lozzio, E. G. Bamberger, and A. S. Feliu. A multipotentialleukemia cell line (k-562) of human origin. crov fov Exp Uiol Mxd, 166(4):546–50, 1981.[117] J. Y. Yuan, F. Wang, J. Yu, G. H. Yang, X. L. Liu, and J. W. Zhang. Microrna-223 reversibly regulates erythroid and megakaryocytic differentiation of k562cells. J Vxll Mol Mxd, 13(11-12):4551–9, 2009.[118] P. Landgraf, M. Rusu, R. Sheridan, A. Sewer, N. Iovino, A. Aravin, S. Pfeffer,A. Rice, A. O. Kamphorst, M. Landthaler, C. Lin, N. D. Socci, L. Hermida,V. Fulci, S. Chiaretti, R. Foa, J. Schliwka, U. Fuchs, A. Novosel, R. U. Muller,B. Schermer, U. Bissels, J. Inman, Q. Phan, M. Chien, D. B. Weir, R. Choksi,G. De Vita, D. Frezzetti, H. I. Trompeter, V. Hornung, G. Teng, G. Hartmann,M. Palkovits, R. Di Lauro, P. Wernet, G. Macino, C. E. Rogler, J. W. Nagle,J. Ju, F. N. Papavasiliou, T. Benzing, P. Lichter, W. Tam, M. J. Brownstein,A. Bosio, A. Borkhardt, J. J. Russo, C. Sander, M. Zavolan, and T. Tuschl. Amammalian microrna expression atlas based on small rna library sequencing.Vxll, 129(7):1401–14, 2007.130Vivliogruphfl[119] N. Xu, T. Papagiannakopoulos, G. Pan, J. A. Thomson, and K. S. Kosik.Microrna-145 regulates oct4, sox2, and klf4 and represses pluripotency in humanembryonic stem cells. Vxll, 137(4):647–58, 2009.[120] J. Huft, D. J. Da Costa, D. Walker, and C. L. Hansen. Three-dimensionallarge-scale microfluidic integration by laser ablation of interlayer connections.Ltu Vhip, 10(18):2358–65, 2010.[121] P. J. Murphy, B. R. Cipriany, C. B. Wallin, C. Y. Ju, K. Szeto, J. A. Hagarman,J. J. Benitez, H. G. Craighead, and P. D. Soloway. Single-molecule analysis ofcombinatorial epigenomic states in normal and tumor cells. crov attl Tvtd fvih f T, 110(19):7772–7, 2013.[122] M. Kantlehner, R. Kirchner, P. Hartmann, J. W. Ellwart, M. Alunni-Fabbroni,and A. Schumacher. A high-throughput dna methylation analysis of a singlecell. auvlxiv Tvids exs, 39(7):e44, 2011.[123] J. Tischler and M. A. Surani. Investigating transcriptional states at single-cell-resolution. Vurr bpin Uiotxvhnol, 24(1):69–78, 2013.[124] Y. Taniguchi, P. J. Choi, G. W. Li, H. Chen, M. Babu, J. Hearn, A. Emili, andX. S. Xie. Quantifying e. coli proteome and transcriptome with single-moleculesensitivity in single cells. fvixnvx, 329(5991):533–8, 2010.[125] V. Lecault, M. Vaninsberghe, S. Sekulovic, D. J. Knapp, S. Wohrer, W. Bowden,F. Viel, T. McLaughlin, A. Jarandehei, M. Miller, D. Falconnet, A. K. White,D. G. Kent, M. R. Copley, F. Taghipour, C. J. Eaves, R. K. Humphries, J. M.Piret, and C. L. Hansen. High-throughput analysis of single hematopoietic stemcell proliferation in microfluidic cell culture arrays. att Mxthods, 8(7):581–6,2011.[126] D. Falconnet, A. Niemisto, R. J. Taylor, M. Ricicova, T. Galitski, I. Shmule-vich, and C. L. Hansen. High-throughput tracking of single yeast cells in amicrofluidic imaging matrix. Ltu Vhip, 11(3):466–73, 2011.[127] F. Notta, S. Doulatov, E. Laurenti, A. Poeppl, I. Jurisica, and J. E. Dick. Isola-tion of single human hematopoietic stem cells capable of long-term multilineageengraftment. fvixnvx, 333(6039):218–21, 2011.131Vivliogruphfl[128] D. G. Kent, M. R. Copley, C. Benz, S. Wohrer, B. J. Dykstra, E. Ma, J. Cheyne,Y. Zhao, M. B. Bowie, Y. Zhao, M. Gasparetto, A. Delaney, C. Smith,M. Marra, and C. J. Eaves. Prospective isolation and molecular characteri-zation of hematopoietic stem cells with durable self-renewal potential. Ulood,113(25):6342–50, 2009.[129] A. K. White, M. VanInsberghe, O. I. Petriv, M. Hamidi, D. Sikorski, M. A.Marra, J. Piret, S. Aparicio, and C. L. Hansen. High-throughput microfluidicsingle-cell rt-qpcr. crov attl Tvtd fvi h f T, 108(34):13999–4004, 2011.[130] F. Tang, K. Lao, and M. A. Surani. Development and applications of single-celltranscriptome analysis. att Mxthods, 8(4 Suppl):S6–11, 2011.[131] P. Mestdagh, T. Feys, N. Bernard, S. Guenther, C. Chen, F. Speleman, andJ. Vandesompele. High-throughput stem-loop rt-qpcr mirna expression profilingusing minute amounts of input rna. auvlxiv Tvids exs, 36(21):e143, 2008.[132] A. Diercks, H. Kostner, and A. Ozinsky. Resolving cell population hetero-geneity: real-time pcr for simultaneous multiplexed gene detection in multiplesingle-cell samples. cLof bnx, 4(7):e6326, 2009.[133] K. A. Heyries, C. Tropini, M. Vaninsberghe, C. Doolin, O. I. Petriv, A. Singhal,K. Leung, C. B. Hughesman, and C. L. Hansen. Megapixel digital pcr. attMxthods, 8(8):649–51, 2011.[134] K. A. Heyries and C. L. Hansen. Parylene c coating for high-performance replicamolding. Ltu Vhip, 11(23):4122–5, 2011.[135] E. Park, B. Williams, B. J. Wold, and A. Mortazavi. Rna editing in the humanencode rna-seq data. Gxnomx exs, 22(9):1626–33, 2012.[136] F Goreaud and R Plissier. On explicit formulas of edge effect correction forripley’s kfunction. Journtl oy ixzxtttion fvixnvx, 10:433–438, 1999.[137] S. Bhat, J. Herrmann, P. Armishaw, P. Corbisier, and K. R. Emslie. Singlemolecule detection in nanofluidic digital array enables accurate measurementof dna copy number. Tntl Uiotntl Vhxm, 394(2):457–67, 2009.132Vivliogruphfl[138] S. Dube, J. Qin, and R. Ramakrishnan. Mathematical analysis of copy numbervariation in a dna sample using digital pcr on a nanofluidic device. cLof bnx,3(8):e2876, 2008.[139] D. Ling and P. M. Salvaterra. Robust rt-qpcr data normalization: validationand selection of internal reference genes during post-experimental data analysis.cLof bnx, 6(3):e17762, 2011.[140] M. Verma, E. G. Karimiani, R. J. Byers, S. Rehman, H. V. Westerhoff, and P. J.Day. Mathematical modelling of mirna mediated bcr.abl protein regulation inchronic myeloid leukaemia vis-a-vis therapeutic strategies. Intxzr Uiol (Vtmu),5(3):543–54, 2013.[141] D. Sayed and M. Abdellatif. Micrornas in development and disease. chysiolexv, 91(3):827–87, 2011.[142] D. P. Bartel. Micrornas: target recognition and regulatory functions. Vxll,136(2):215–33, 2009.[143] C. C. Pritchard, H. H. Cheng, and M. Tewari. Microrna profiling: approachesand considerations. att exv Gxnxt, 13(5):358–69, 2012.[144] N. R. Christoffersen, A. Silahtaroglu, U. A. Orom, S. Kauppinen, and A. H.Lund. mir-200b mediates post-transcriptional repression of zfhx1b. eaT,13(8):1172–8, 2007.[145] E. Barrey, G. Saint-Auret, B. Bonnamy, D. Damas, O. Boyer, and X. Gidrol.Pre-microrna and mature microrna in human mitochondria. cLof bnx,6(5):e20220, 2011.[146] M. Wu, M. Piccini, C. Y. Koh, K. S. Lam, and A. K. Singh. Single cell micrornaanalysis using microfluidic flow cytometry. cLof bnx, 8(1):e55044, 2013.[147] Encode Project Consortium. An integrated encyclopedia of dna elements in thehuman genome. atturx, 489(7414):57–74, 2012.[148] LA Luigi A Warren, JA Joshua A Weinstein, and Stephen R SR Quake. Thedigital array response curve, 2007.133Vivliogruphfl[149] N. Farago, A. K. Kocsis, S. Lovas, G. Molnar, E. Boldog, M. Rozsa, V. Sze-menyei, E. Vamos, L. I. Nagy, G. Tamas, and L. G. Puskas. Digital pcr to deter-mine the number of transcripts from single neurons after patch-clamp recording.Uiotxvhniquxs, 54(6):327–36, 2013.[150] A. Kumari, C. Brendel, A. Hochhaus, A. Neubauer, and A. Burchert. Lowbcr-abl expression levels in hematopoietic precursor cells enable persistence ofchronic myeloid leukemia under imatinib. Ulood, 119(2):530–9, 2012.[151] Y. Chen, B. Zhang, H. Feng, W. Shu, G. Y. Chen, and J. F. Zhong. An au-tomated microfluidic device for assessment of mammalian cell genetic stability.Ltu Vhip, 12(20):3930–5, 2012.[152] A. Stahlberg, C. Thomsen, D. Ruff, and P. Aman. Quantitative pcr analysisof dna, rnas, and proteins in the same single cell. Vlin Vhxm, 58(12):1682–91,2012.[153] K. A. Whitehead, R. Langer, and D. G. Anderson. Knocking down barriers:advances in sirna delivery. att exv Druz Disvov, 8(2):129–38, 2009.[154] B. L. Davidson and Jr. McCray, P. B. Current prospects for rna interference-based therapies. att exv Gxnxt, 12(5):329–40, 2011.[155] M. DiFiglia, M. Sena-Esteves, K. Chase, E. Sapp, E. Pfister, M. Sass, J. Yoder,P. Reeves, R. K. Pandey, K. G. Rajeev, M. Manoharan, D. W. Sah, P. D.Zamore, and N. Aronin. Therapeutic silencing of mutant huntingtin with sirnaattenuates striatal and cortical neuropathology and behavioral deficits. crovattl Tvtd fvi h f T, 104(43):17204–9, 2007.[156] S. D. Li, Y. C. Chen, M. J. Hackett, and L. Huang. Tumor-targeted deliveryof sirna by self-assembled nanoparticles. Mol ghxr, 16(1):163–9, 2008.[157] G. J. Hannon and J. J. Rossi. Unlocking the potential of the human genomewith rna interference. atturx, 431(7006):371–8, 2004.[158] X. Su, J. Fricke, D. G. Kavanagh, and D. J. Irvine. In vitro and in vivomrna delivery using lipid-enveloped ph-responsive polymer nanoparticles. Molchtrm, 8(3):774–87, 2011.134Vivliogruphfl[159] G. Cannon and D. Weissman. Rna based vaccines. DaT Vxll Uiol, 21(12):953–61, 2002.[160] A. J. Geall, A. Verma, G. R. Otten, C. A. Shaw, A. Hekele, K. Banerjee, Y. Cu,C. W. Beard, L. A. Brito, T. Krucker, D. T. O’Hagan, M. Singh, P. W. Mason,N. M. Valiante, P. R. Dormitzer, S. W. Barnett, R. Rappuoli, J. B. Ulmer, andC. W. Mandl. Nonviral delivery of self-amplifying rna vaccines. crov attl Tvtdfvi h f T, 109(36):14604–9, 2012.[161] L. Warren, P. D. Manos, T. Ahfeldt, Y. H. Loh, H. Li, F. Lau, W. Ebina, P. K.Mandal, Z. D. Smith, A. Meissner, G. Q. Daley, A. S. Brack, J. J. Collins,C. Cowan, T. M. Schlaeger, and D. J. Rossi. Highly efficient reprogramming topluripotency and directed differentiation of human cells with synthetic modifiedmrna. Vxll ftxm Vxll, 7(5):618–30, 2010.[162] J. R. Plews, J. Li, M. Jones, H. D. Moore, C. Mason, P. W. Andrews, andJ. Na. Activation of pluripotency genes in human fibroblast cells by a novelmrna based approach. cLof bnx, 5(12):e14397, 2010.[163] Anonymous. Is this really the rnaissance? att Uiotxvhnol, 32(3):201, 2014.[164] E. P. Thi, C. E. Mire, A. C. Lee, J. B. Geisbert, J. Z. Zhou, K. N. Agans, N. M.Snead, D. J. Deer, T. R. Barnard, K. A. Fenton, I. MacLachlan, and T. W.Geisbert. Lipid nanoparticle sirna treatment of ebola-virus-makona-infectednonhuman primates. atturx, 521(7552):362–5, 2015.[165] L. Timmerman. Merck joins messenger rna frenzy, betting $100m on modernatherapeutics. Foruxs, (September 7, 2015), 2015.[166] T. Kawai and S. Akira. The role of pattern-recognition receptors in innateimmunity: update on toll-like receptors. att Immunol, 11(5):373–84, 2010.[167] K. Kariko, M. Buckstein, H. Ni, and D. Weissman. Suppression of rna recog-nition by toll-like receptors: the impact of nucleoside modification and theevolutionary origin of rna. Immunity, 23(2):165–75, 2005.[168] M. S. Kormann, G. Hasenpusch, M. K. Aneja, G. Nica, A. W. Flemmer,S. Herber-Jonat, M. Huppmann, L. E. Mays, M. Illenyi, A. Schams, M. Griese,135VivliogruphflI. Bittmann, R. Handgretinger, D. Hartl, J. Rosenecker, and C. Rudolph. Ex-pression of therapeutic proteins after delivery of chemically modified mrna inmice. att Uiotxvhnol, 29(2):154–7, 2011.[169] N. Desai. Challenges in development of nanoparticle-based therapeutics. TTcfJ, 14(2):282–95, 2012.[170] S. C. Semple, A. Akinc, J. Chen, A. P. Sandhu, B. L. Mui, C. K. Cho, D. W.Sah, D. Stebbing, E. J. Crosley, E. Yaworski, I. M. Hafez, J. R. Dorkin, J. Qin,K. Lam, K. G. Rajeev, K. F. Wong, L. B. Jeffs, L. Nechev, M. L. Eisen-hardt, M. Jayaraman, M. Kazem, M. A. Maier, M. Srinivasulu, M. J. Wein-stein, Q. Chen, R. Alvarez, S. A. Barros, S. De, S. K. Klimuk, T. Borland,V. Kosovrasti, W. L. Cantley, Y. K. Tam, M. Manoharan, M. A. Ciufolini,M. A. Tracy, A. de Fougerolles, I. MacLachlan, P. R. Cullis, T. D. Madden,and M. J. Hope. Rational design of cationic lipids for sirna delivery. attUiotxvhnol, 28(2):172–6, 2010.[171] A. Sridharan, C. Patel, and J. Muthuswamy. Voltage preconditioning allowsmodulated gene expression in neurons using pei-complexed sirna. Mol ghxrauvlxiv Tvids, 2:e82, 2013.[172] B. Urban-Klein, S. Werth, S. Abuharbeid, F. Czubayko, and A. Aigner.Rnai-mediated gene-targeting through systemic application of polyethylenimine(pei)-complexed sirna in vivo. Gxnx ghxr, 12(5):461–6, 2005.[173] P. Kumar, H. Wu, J. L. McBride, K. E. Jung, M. H. Kim, B. L. Davidson, S. K.Lee, P. Shankar, and N. Manjunath. Transvascular delivery of small interferingrna to the central nervous system. atturx, 448(7149):39–43, 2007.[174] E. Song, P. Zhu, S. K. Lee, D. Chowdhury, S. Kussman, D. M. Dykxhoorn,Y. Feng, D. Palliser, D. B. Weiner, P. Shankar, W. A. Marasco, and J. Lieber-man. Antibody mediated in vivo delivery of small interfering rnas via cell-surface receptors. att Uiotxvhnol, 23(6):709–17, 2005.[175] A. Akinc, W. Querbes, S. De, J. Qin, M. Frank-Kamenetsky, K. N. Jayaprakash,M. Jayaraman, K. G. Rajeev, W. L. Cantley, J. R. Dorkin, J. S. Butler, L. Qin,T. Racie, A. Sprague, E. Fava, A. Zeigerer, M. J. Hope, M. Zerial, D. W. Sah,K. Fitzgerald, M. A. Tracy, M. Manoharan, V. Koteliansky, Ad Fougerolles,136Vivliogruphfland M. A. Maier. Targeted delivery of rnai therapeutics with endogenous andexogenous ligand-based mechanisms. Mol ghxr, 18(7):1357–64, 2010.[176] J. Sparks, G. Slobodkin, M. Matar, R. Congo, D. Ulkoski, A. Rea-Ramsey,C. Pence, J. Rice, D. McClure, K. J. Polach, E. Brunhoeber, L. Wilkinson,K. Wallace, K. Anwer, and J. G. Fewell. Versatile cationic lipids for sirnadelivery. J Vontrol exlxtsx, 158(2):269–76, 2012.[177] N. M. Belliveau, J. Huft, P. J. Lin, S. Chen, A. K. Leung, T. J. Leaver, A. W.Wild, J. B. Lee, R. J. Taylor, Y. K. Tam, C. L. Hansen, and P. R. Cullis.Microfluidic synthesis of highly potent limit-size lipid nanoparticles for in vivodelivery of sirna. Mol ghxr auvlxiv Tvids, 1:e37, 2012.[178] T. S. Zimmermann, A. C. Lee, A. Akinc, B. Bramlage, D. Bumcrot, M. N.Fedoruk, J. Harborth, J. A. Heyes, L. B. Jeffs, M. John, A. D. Judge, K. Lam,K. McClintock, L. V. Nechev, L. R. Palmer, T. Racie, I. Rohl, S. Seiffert,S. Shanmugam, V. Sood, J. Soutschek, I. Toudjarska, A. J. Wheat, E. Ya-worski, W. Zedalis, V. Koteliansky, M. Manoharan, H. P. Vornlocher, andI. MacLachlan. Rnai-mediated gene silencing in non-human primates. atturx,441(7089):111–4, 2006.[179] M. Frank-Kamenetsky, A. Grefhorst, N. N. Anderson, T. S. Racie, B. Bramlage,A. Akinc, D. Butler, K. Charisse, R. Dorkin, Y. Fan, C. Gamba-Vitalo, P. Had-wiger, M. Jayaraman, M. John, K. N. Jayaprakash, M. Maier, L. Nechev, K. G.Rajeev, T. Read, I. Rohl, J. Soutschek, P. Tan, J. Wong, G. Wang, T. Zim-mermann, A. de Fougerolles, H. P. Vornlocher, R. Langer, D. G. Anderson,M. Manoharan, V. Koteliansky, J. D. Horton, and K. Fitzgerald. Therapeu-tic rnai targeting pcsk9 acutely lowers plasma cholesterol in rodents and ldlcholesterol in nonhuman primates. crov attl Tvtd fvi h f T, 105(33):11915–20, 2008.[180] A. D. Stroock, S. K. Dertinger, A. Ajdari, I. Mezic, H. A. Stone, and G. M.Whitesides. Chaotic mixer for microchannels. fvixnvx, 295(5555):647–51, 2002.[181] I. V. Zhigaltsev, N. Belliveau, I. Hafez, A. K. Leung, J. Huft, C. Hansen, andP. R. Cullis. Bottom-up design and synthesis of limit size lipid nanoparticle sys-tems with aqueous and triglyceride cores using millisecond microfluidic mixing.Ltnzmuir, 28(7):3633–40, 2012.137Vivliogruphfl[182] A. K. Leung, I. M. Hafez, S. Baoukina, N. M. Belliveau, I. V. Zhigaltsev,E. Afshinmanesh, D. P. Tieleman, C. L. Hansen, M. J. Hope, and P. R. Cullis.Lipid nanoparticles containing sirna synthesized by microfluidic mixing exhibitan electron-dense nanostructured core. J chys Vhxm V atnomttxr Intxrytvxs,116(34):18440–18450, 2012.[183] Y. Pei, P. J. Hancock, H. Zhang, R. Bartz, C. Cherrin, N. Innocent, C. J.Pomerantz, J. Seitzer, M. L. Koser, M. T. Abrams, Y. Xu, N. A. Kuklin,P. A. Burke, A. B. Sachs, L. Sepp-Lorenzino, and S. F. Barnett. Quantitativeevaluation of sirna delivery in vivo. eaT, 16(12):2553–63, 2010.[184] M. T. Abrams, M. L. Koser, J. Seitzer, S. C. Williams, M. A. DiPietro,W. Wang, A. W. Shaw, X. Mao, V. Jadhav, J. P. Davide, P. A. Burke, A. B.Sachs, S. M. Stirdivant, and L. Sepp-Lorenzino. Evaluation of efficacy, biodis-tribution, and inflammation for a potent sirna nanoparticle: effect of dexam-ethasone co-treatment. Mol ghxr, 18(1):171–80, 2010.[185] S. Mukherji, M. S. Ebert, G. X. Zheng, J. S. Tsang, P. A. Sharp, and A. vanOudenaarden. Micrornas can generate thresholds in target gene expression. attGxnxt, 43(9):854–9, 2011.[186] J. Gilleron, W. Querbes, A. Zeigerer, A. Borodovsky, G. Marsico, U. Schubert,K. Manygoats, S. Seifert, C. Andree, M. Stoter, H. Epstein-Barash, L. Zhang,V. Koteliansky, K. Fitzgerald, E. Fava, M. Bickle, Y. Kalaidzidis, A. Akinc,M. Maier, and M. Zerial. Image-based analysis of lipid nanoparticle-mediatedsirna delivery, intracellular trafficking and endosomal escape. att Uiotxvhnol,31(7):638–46, 2013.[187] G. Sahay, W. Querbes, C. Alabi, A. Eltoukhy, S. Sarkar, C. Zurenko, E. Kara-giannis, K. Love, D. Chen, R. Zoncu, Y. Buganim, A. Schroeder, R. Langer,and D. G. Anderson. Efficiency of sirna delivery by lipid nanoparticles is limitedby endocytic recycling. att Uiotxvhnol, 31(7):653–8, 2013.[188] L. Warren, Y. Ni, J. Wang, and X. Guo. Feeder-free derivation of humaninduced pluripotent stem cells with messenger rna. fvi exp, 2:657, 2012.[189] N. J. Caron, B. K. Gage, M. D. O’Connor, C. J. Eaves, T. J. Kieffer, and138VivliogruphflJ. M. Piret. A human embryonic stem cell line adapted for high throughputscreening. Uiotxvhnol Uioxnz, 110(10):2706–16, 2013.[190] R. L. Rungta, H. B. Choi, P. J. Lin, R. W. Ko, D. Ashby, J. Nair, M. Manoha-ran, P. R. Cullis, and B. A. Macvicar. Lipid nanoparticle delivery of sirna tosilence neuronal gene expression in the brain. Mol ghxr auvlxiv Tvids, 2:e136,2013.[191] A. K. White, K. A. Heyries, C. Doolin, M. Vaninsberghe, and C. L. Hansen.High-throughput microfluidic single-cell digital polymerase chain reaction. TntlVhxm, 85(15):7182–90, 2013.[192] P. C. Blainey and S. R. Quake. Digital mda for enumeration of total nucleicacid contamination. auvlxiv Tvids exs, 39(4):e19, 2011.[193] K. A. Janes, C. C. Wang, K. J. Holmberg, K. Cabral, and J. S. Brugge. Iden-tifying single-cell molecular programs by stochastic profiling. att Mxthods,7(4):311–7, 2010.[194] J. P. Junker, E. S. Noel, V. Guryev, K. A. Peterson, G. Shah, J. Huisken, A. P.McMahon, E. Berezikov, J. Bakkers, and A. van Oudenaarden. Genome-widerna tomography in the zebrafish embryo. Vxll, 159(3):662–75, 2014.[195] N. Crosetto, M. Bienko, and A. van Oudenaarden. Spatially resolved transcrip-tomics and beyond. att exv Gxnxt, 16(1):57–66, 2015.[196] S. Fredriksson, M. Gullberg, J. Jarvius, C. Olsson, K. Pietras, S. M. Gustafsdot-tir, A. Ostman, and U. Landegren. Protein detection using proximity-dependentdna ligation assays. att Uiotxvhnol, 20(5):473–7, 2002.[197] S. C. Bendall, E. F. Simonds, P. Qiu, A. D. Amir el, P. O. Krutzik, R. Finck,R. V. Bruggner, R. Melamed, A. Trejo, O. I. Ornatsky, R. S. Balderas, S. K.Plevritis, K. Sachs, D. Pe’er, S. D. Tanner, and G. P. Nolan. Single-cell mass cy-tometry of differential immune and drug responses across a human hematopoi-etic continuum. fvixnvx, 332(6030):687–96, 2011.[198] A. A. Khan, D. Betel, M. L. Miller, C. Sander, C. S. Leslie, and D. S. Marks.Transfection of small rnas globally perturbs gene regulation by endogenousmicrornas. att Uiotxvhnol, 27(6):549–55, 2009.139[199] J. Wang and S. R. Quake. Rna-guided endonuclease provides a therapeuticstrategy to cure latent herpesviridae infection. crov attl Tvtd fvi h f T,111(36):13157–62, 2014.[200] M. K. White, W. Hu, and K. Khalili. The crispr/cas9 genome editing method-ology as a weapon against human viruses. Disvov Mxd, 19(105):255–62, 2015.140Vppenyifl VDesign ConsiyervtionsVCF hingleBCell Digitvl eCg erototype fiithVlternvtive Vpprovxh to biflingDigital PCR requires template molecules to be randomly distributed among reactionchambers in order to accurately infer the number of molecules in the array from thenumber of reaction chambers with PCR amplification. If the number of moleculesin the array is significantly less than the number of chambers compartmentalizingthe solution, then most of the chambers will contain either one or zero templatemolecules. In this case, the number of positive wells (with amplified product) atthe end of PCR gives a count of the number of molecules in the sample to a closeapproximation. However, if the number of template molecules is not small comparedto the number of chambers, many chambers will contain more than one templatemolecule. This results in the number of positive reaction chambers being significantlyless than the number of molecules in the sample. However, as long as the array is notfully saturated, template abundance can still be estimated based on the statisticalrelationship between the number of molecules and the expected number of positivewells.In integrating digital PCR with the cell processing previously established, one ofthe challenges is ensuring the sample is thoroughly mixed prior to dPCR compart-mentalization. In conventional dPCR, a solution that is already completely mixedis simply injected into a dPCR array, or partitioned into droplets. In our single-cellRT-qPCR device, chemical reactions are assembled on-chip by diluting a sample intoa larger chamber with the required reagent. Some mixing of solutions occurs throughadvection, however the majority of the combined solution is mixed by diffusion. Aninitial prototype attempted to solve this problem by pushing template solution withPCR reagent into a dPCR array consisting of a bifurcating structure of channelsthat can partitioned with valves into dPCR chambers. However, prior to compart-141ABEB ginglyACyll Digitul dCf drototflpy with Altyrnutivy Approuwh to aifiingPrototype: Integrated Microfluidic Device for Single      Cell Digital PCR … Reagent Injection Cell Capture Lysis (12 nL) Reverse Transcription (60 nL) Digital PCR Array (2000 × 1 nL) Peristaltic Pump Similar to RT-qPCR device but with digital PCR array as final chamber. Figure A.1: Design schematic of an early prototype single cell digital PCR device.This design is based on physical valve partitioning of channels into digital PCR cham-bers, and requires a much larger footprint compared to the surface tension partitioningwith oil strategy presented in Chapter 3. This architecture incorporates a peristalticpump to create a rotary mixer in the array to completely mix the solution beforepartitioning. Valve (or ‘control’) channels are indicated in red.mentalization, the injected PCR and template solution can be mixed by using valvesto pump the solution around the ring of dPCR channels (Figure A.1). This rotarymixing reduces diffusion times by wrapping the template and PCR solutions withineach other to decrease characteristic diffusion distances. In practice, this approachstill required long mixing times and was further limited in dPCR chamber density asit used mechanical valves to partition the channels into dPCR chambers.The high-density planar emulsion dPCR arrays previously developed by Heyries etal. overcome the large footprint of conventional valve-based dPCR arrays. However,this approach requires a pre-mixed PCR/template solution to be injected as the dPCRarray is dead-end filled, preventing any rotary mixing and imposing long diffusion dis-tances between far chambers of the array. In order to incorporate high-density planaremulsion dPCR arrays, the prototype single-cell dPCR device mixes PCR solutionwith template by diffusion in a 50 nL chamber (see Chapter 3). This mixed PCR solu-tion is then pushed (with unmixed PCR reagent) into an array of 1020 × 25 pL dPCRchambers which accommodates approximately half of the mixed PCR/template so-lution. This strategy of sampling half of the pre-mixed PCR/template solution, andthe architecture of the mixing chamber, accomplishes injection of a completely mixed142ABEB ginglyACyll Digitul dCf drototflpy with Altyrnutivy Approuwh to aifiingPCR solution into the dPCR array. The disadvantage of this approach is the entirecontent of the cell is not analyzed, limiting the precision and sensitivity of the device,particularly in cases where transcript abundance is less than 6 copies per cell. Poten-tially, the mixed PCR/template solution may be pushed with an immiscible phasefluid such as oil or a gas in order to facilitate fluid transfer between chambers withoutdilution or an unmixed front of solution.143Vppenyifl Werotoxols1WCF FvwrixvtionAfter fabricating the molds described in the methods section of chapter 2, mul-tilayer soft lithography is used to create microfluidic devices. This protocol usespolydimethylsiloxane (PDMS): specifically RTV615. The device uses a ‘push-up’ ge-ometry, with the valve/control channels in the layer underneath the flow channels(for sample and reagents). This information is contained in Chapter 2, however ispresented here in expanded form with greater detail.Before coating with PDMS, treat wafers with TMCS (trichloromethylsilande).Seal wafers in a box containing a beaker of TMCS. TMCS is volatile and will coatwafer to prevent PDMS from sticking to the photoresist (freatures).WCFCF Flofi avyer1. Mix Flow layer (thick) (5:1 A:B). Use 50 g A and 10 g B per wafer.2. Pour 5:1 mixture on the top layer wafer in foiled dish.3. Use pipette tips with bigger side down to centre flow wafer and push it downto release bubbles from under wafer4. Degas the mixture/wafer by placing in vacuum for minimum 1 hour.WCFCG Control vny Wlvnk avyer(s)1. While degassing, mix Control and blank layers (thin) (20:1 A:B). Use 20g Aand 1g B per wafer (40 g A and 2 g B for both control and blank if being used)2. Place control wafer on spinner and pour 20:1 mixture on about 2/3 of wafer.1ctrts oy this protovol tre reprowuvew yrom Microuidic gxchnology yor High-ghroughput finglxCxll Gxnx Exprxssion Tntlysis, uy Twtm White, hniversity oy Uritish Volumuit, 2CDCA144VBEB Fuvriwution3. Spin cycle:• Ramp 10s to 500 rpm, hold 5s• Ramp 5s to 1800 rpm, hold 60s• Ramp 5s to 0 rpm4. When finished spinning, place in a dish and close lid (can put wafer on lidbecause its flatter)5. Repeat for blank(s) and then set aside blank layer.WCFCH Wvke Flofi vny Control avyers1. When there are only a few bubbles left on flow layer, remove from vacuumchamber, use pipette tip to drag bubbles off the features, and push wafer tobottom.2. Bake Flow Layer 80◦C, 60 Min.3. Wait 15 min and then bake Control Layer 80◦C, 45 Min (this way you canremove both layers from oven at the same time)WCFCI Vlign Flofi avyer to Control avyer1. Remove both layers from oven and allow to cool for a couple minutes2. Cut on the inside of the flow wafer for a clean lift off. Cut multiple times toensure there is no debris that might fold under flow layer and prevent properbonding to control layer. Peel flow layer off of wafer.3. Immediately after peeling flow layer, place it on top of control layer to minimizechance of debris getting between the 2 layers.4. Align flow layer to control layer (still on wafer).5. Use tape (scotch) to clean off top of the chip6. Bake Flow and Control layers for 1 hour at 80◦C.145VBEB FuvriwutionWCFC5 eorts1. Peel combined flow and control layers off of wafer. Make multiple cuts to ensurea clean lift off.2. Punch holes in devices (bonded Control and Flow layers).3. When you think you are 45 min away from finishing punching holes, bake blanklayer for 45 min at 80◦C4. Clean bottom of bonded layers vigorously with tape. Remove blank layer fromoven and place bonded layers control and flow layers onto the blank layer. Besure the control (thin) layer of the bonded layers is down. Check for collapsedvalves and use syringe to suck them out if necessary. Try to avoid bubbles.Clean top surface with tape.5. Cook overnight at 80◦C. Mimimum 3 hours.WCFCK bounting Inyiviyuvl Devixes1. Dice Chips2. Place chips on glass slides. Be sure to clean slides with water and IPA first.Clean chips with tape (bottom before fastening to slide, top after fastening).Make sure plug holes are up (blank and thin on the bottom, thick on top)3. Bake slides with chips at 80◦C over night. Be sure to label the slides withinformation about chip, fabrication date, etc.WCFCL Genervl Consiyervtions1. RTV stands for room temperature vulcanization. Do not let RTC A:B sit fortoo long ( greater than 4 hours).2. RTV A : RTV B, 10:1 is the stoichiometrically equivalent ratio.3. Use Nitrile gloves since Latex gloves contain sulfur that may react with Ptcatalyst in RTV.4. Check that layers have baked properly before alignment (touch edge of waferswith a tweezer.146VBFB Dyviwy cpyrutionWCG Devixe dpervtionThe first device operation protocol presented is for a heat lysis, followed by a 2-stepRT-qPCR with seperate reverse transcription (RT) and qPCR steps. This protocolwas used with miRNA and mRNA assays. Alternatively, a chemical lysis followedby 1-step RT-qPCR protocol is also presented. This protocol is faster, however didnot work for miRNA assays, and was applied to measurements of GAPDH and SNVmeasurements. Krytox oil is used as the fluid in the control lines.WCGCF Cell aovyingA lvshingA vny Hevt aysis1. Device is primed with PBS containing 0.5 mg/mL BSA and 0.5 U/L RNaseInhibitor. The bovine serum albumin (BSA) prevents cells from sticking tochannel walls.2. Cells loaded into device suspended in culture media (directly from culture).Optional off-chip wash. Cell suspensions may be drawn into microcapillerypipette tips, and plugged into the sample inlet ports. The pipette tip is releasedfrom the pipette and air pressure is applied to opening. We used PDMS plugsto seal the pipette tips around the applied air pressure line. Alternatively, thecell suspension may be drawn into tygon tubing with a steel pin on the end ofit, which is in turn plugged into the microfluidic device. Cell loading works bestat concentrations between 5x105 and 1x106 cells/mL. Lower concentrations willwork but it will take longer to achieve high occupancy of trapped single cells.Higher concentrations may lead to clogging in the inlet port or at the traps.Load cells at approximately 2 psi. Optionally, the peristaltic pump may be usedfor gentler and controlled cell loading.3. After loading cells, perform on-chip wash to remove untrapped cells and extra-cellular RNA. Cells are washed with the same solution that primes the device.4. Close valves to partition cell loading channel in order to isolate cells in capturechambers.5. Using microscope, confirm and count which chambers contain cells (enter intospreadsheet).147VBFB Dyviwy cpyrution6. Acutate valves to isolate cell capture chamber. Place device on flatbed thermo-cycler for heat lysis at 85◦C for 7 minutes; followed by 4◦C hold.WCGCG geverse irvnsxription1. Using the ABI High Capacity Reverse Transcription kit, the RT solution pre-pared as below (modified from ABI protocol).• 10x RT Buffer: 2.00 L• 5x RT primer: 4.00 L• dNTPs: 1.00 L• Multiscribe RT Enzyme: 1.34 L• RNase Inhibitor: 0.26 L• Tween 20 (1%): 2.0 L• H20: 9.4 L2. Reverse transcription mix is loaded into the device and flushed through thereagent injection channels.3. RT reagent is injected into the reaction by opening the valve connecting thecell chamber to the RT chamber, and the valve connecting the cell chamber tothe reagent injection line. RT chamber is dead-end filled, and then the reagentthe connection to the reagent injection line is closed.4. The device is placed on a flatbed thermocycler for a pulsed temperature RTprotocol.• 16◦C x 2 min• 60 cycles of (20◦C x 30 s, 42◦C x 30 s, 50◦C x 1 s)• 85◦C x 5 min• Hold 4◦CWCGCH gevlBiime eolymervse Chvin gevxtion1. The reagent mix for the PCR reaction is prepared using the ABI Taqman Uni-versal Master Mix.148VBFB Dyviwy cpyrution• 2x Taqman Universal Master Mix (ABI): 25.0 L• 20x Real-Time Primer/Probe: 2.50 L• Tween 20 (1%): 5.0 L• Water: 7.5 L• RT product (already in device): 10.0 L2. Flush reagent injection lines prior to injecting PCR reagent into reaction cham-bers (similar to RT injection). Input pressure may be increased to increasedead-end filling rate. Pressure should not be decreased as this may result inback-flow from the reaction chambers, and could lead to cross-contamination.3. One the PCR reaction chamber is filled, the valves closing the PCR chambers areactuated. The rest of the control lines may be removed (cut away or unplugged)to facilitate placing the device into the custom flatbed thermocycler apparatusfor imaging the qPCR reaction.4. The thermocycler controls termperature cycling for the PCR protocol• 95◦C Hot Start for 10 mins• 50 cycles of 95◦C x 15 s (denature) and 60◦C x 1 min (anneal/extend)WCGCI Chemixvl aysis vny dneBhtep giBqeCgThe chemical lysis followed by one-step RT-qPCR protocol is based on the IvitrogenCellsDirect One-Step RT-qPCR Kit.1. For device priming and cell loading, follow steps 1-5 from the heat lysis protocol.2. The lysis buffer is prepared following the CellsDirect kit instructions.• Resuspension buffer: 30 L• Lysis enhancer solution: 3 L3. The lysis buffer is injected into the 10 nL chambers used for reverse transcriptionin the other protocol. This follows the same procedure of flushing the reagentinjection lines prior to opening valves to permit dead-end filling.4. Incubate lysis reaction for 10 minutes at room temperature.149VBFB Dyviwy cpyrution5. Place device on flatbed thermocycler for heat inactivation of lysis reagent, 10minutes at 75◦C.6. Prepare RT-qPCR reagent as instructed in CellsDirect kit, with addition ofTween 20 surfactant.• SuperScript III RT/Platinum Taq Mix: 1.0 L• 2X Reaction Mix (with ROX Reference Dye): 25 L• 20X Taqman Primers/Probes 2.5 L• Magnesium Sulphate: 1.0 L• Tween (1%): 5.0 L• Water: 5.5 L• Lysate (already in device): 10 L (equivalent)7. RT-qPCR reagent is injected into device similar to final PCR steps in the heatlysis protocol presented above, and the device is placed in the custom flatbedthermocycler and imaging apparatus.8. A one-step RT-qPCR reaction is performed by RT followed by qPCR withoutinterruption or addition of reagents.• 50◦C for 15 minutes• 95◦C for 2 minutes• 40-50 cycles of 95◦C x 15 seconds and 60◦C x 30 seconds150


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items