Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

The development of single-cell microfluidic technologies for the analysis of microRNA expression VanInsberghe, Michael 2018

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

Item Metadata

Download

Media
24-ubc_2018_may_vaninsberghe_michael.pdf [ 25.05MB ]
Metadata
JSON: 24-1.0365804.json
JSON-LD: 24-1.0365804-ld.json
RDF/XML (Pretty): 24-1.0365804-rdf.xml
RDF/JSON: 24-1.0365804-rdf.json
Turtle: 24-1.0365804-turtle.txt
N-Triples: 24-1.0365804-rdf-ntriples.txt
Original Record: 24-1.0365804-source.json
Full Text
24-1.0365804-fulltext.txt
Citation
24-1.0365804.ris

Full Text

The development of single-cell microfluidic technologiesfor the analysis of microRNA expressionbyMichael VanInsbergheB.Sc. Biophysics, The University of British Columbia, 2010A THESIS SUBMITTED IN PARTIAL FULFILLMENTOF THE REQUIREMENTS FOR THE DEGREE OFDoctor of PhilosophyinTHE FACULTY OF GRADUATE AND POSTDOCTORALSTUDIES(Genome Science and Technology)The University of British Columbia(Vancouver)April 2018c©Michael VanInsberghe, 2018AbstractNew technologies for single-cell RNA expression profiling have transformed ourunderstanding of cell biology. Despite several variations, the vast majority of avail-able methods are applicable only to messenger RNA expression measurements,leaving microRNAs and other classes of small RNAs largely unexplored. In thisdissertation I describe the creation and application of technology that allows for theefficient and precise analysis of microRNA expression in large numbers of singlecells.First, the foundational components necessary for microfluidic integration ofsingle-cell RT-qPCR analysis were developed. The resulting device executes andparallelizes all steps of cell capture, cell lysis, reverse transcription, and quantita-tive PCR on up to 300 cells per run. In comparison to standard benchtop assays,nanolitre-volume processing was found to increase measurement performance onsamples with limited template. The core functionality established in this part of thework provides a foundation for further microfluidic single-cell assays.The capabilities of this platform were next expanded to enable highly multi-plexed RT-qPCR analysis. By incorporating sophisticated microfluidic fabricationtechniques with an extended workflow that included a pre-amplification step, thenumber of assays per cell was increased from one or two to up to forty, whilemaintaining the same benefits to measurement performance that were previouslyobserved.Finally, small-volume analysis and carefully optimized molecular biology werecombined to develop a method to generate high-quality single-cell microRNA se-quencing libraries. This method was then applied to provide a comprehensive lookat microRNA expression dynamics across the human hematopoietic cell hierarchy.iiThe results indicated that the population structure derived from miRNA expressionprofiles supported a model of continuous, linear hematopoietic stem cell differen-tiation, in contrast to the prevailing model of stepwise, branched lineage commit-ment. An expanded set of miRNA markers that are highly expressed in HSCs,decrease gradually during differentiation, and are absent in mature cells were alsoidentified. Finally, an analysis of the relative expression of microRNA isoformswas performed, showing that they are a dynamic feature that varies between differ-ent microRNAs and cell types.The capabilities conferred by this suite of microfluidic devices will enable thecontinued, routine analysis of microRNA expression in single cells.iiiLay SummaryOrganisms are composed of many cell types, each of which fills a unique role inestablishing specialized tissues. Despite the acknowledged heterogeneity betweencells, our understanding of the molecular basis of life is largely based on measure-ments of these tissues, rather than the individual cells. Pooling cells together isrequired to overcome the sensitivity limits of common molecular profiling tech-niques, but masks the differences between individual cells. This dissertation de-velops methods to measure the expression of microRNAs, an important class ofregulatory molecule, in single cells. We created miniaturized integrated fluidiccircuits made up of networks of small tubes, valves, and reaction chambers to per-form these measurements. One of the demonstrations of our technology, measuringmicroRNA expression in the blood development system, discovered a set of newmarkers for hematopoietic stem cells, and helped to refine the model for how thesecells ultimately create all the different blood cell types.ivPrefaceThe work presented in this dissertation is part of a collaborative effort to developand apply techniques for single-cell RNA expression analysis, and has resulted inco-authored publications. The majority of the presented results derive from workthat I initiated, executed, and analyzed. Contributions from collaborators are notedbelow.Chapter 1Figure 1.1, Figure 1.2, Figure 1.4, Figure 1.5B, Figure 1.7B, Figure 1.8A, andFigure 1.9 were adapted or reproduced with permission.Chapter 2A version of Chapter 2 has been published: Adam K. White, Michael VanIns-berghe, Oleh I. Petriv, Mani Hamidi, Darek. Sikorski, Marco A. Marra, JamesPiret, Samuel. Aparicio, and Carl L. Hansen. High-throughput microfluidic single-cell RT-qPCR. Proceedings of the National Academy of Sciences, 108(34):13999-14004, 2011. Adam K. White and I are co-first authors on this publication.Work in this chapter is also part of a patent: Carl L. Hansen, Michael VanIns-berghe, Adam White, Oleh I. Petriv, Timothy Leaver, Anupam Singhal, WilliamBowden, Veronique Lecault, Daniel Da Costa, Leo Wu, Georgia Russell, andDarek Sikorski. Microfluidic Cell Trap and Assay Apparatus for High-ThroughputAnalysis. United States Patent 9,902,990.Adam designed the microfluidic device with my input and incorporating pre-liminary results from experiments that I designed and performed. Adam fabricatedthe master molds, and we both fabricated devices. The cell trap design emergedvout of a series of ideas generated in a collaborative brainstorming session in theHansen lab; I subsequently drew, fabricated, and tested these ideas to arrive at theone used in this chapter. I initiated and performed further systematic optimizationsto increase cell capture efficiency, and was later assisted in this endeavour by HansZahn.Adam and I worked very closely to design and execute all the microfluidic RT-qPCR experiments. I wrote image-analysis scripts and analyzed data. Darek Siko-rski performed OCT4 RNA-FISH expression measurements, prepared Figure 2.12,and differentiated CA1S hESC cells. Oleh Petriv and Mani Hamidi designed OCT4RT-qPCR expression assays and performed their off-chip validation. Oleh Petrivperformed off-chip single-cell RT-qPCR experiments and digital PCR analysis ofmiRNA expression in bulk cell lysates. Marco A. Marra, James Piret, SamuelAparicio, and Carl L. Hansen designed research.Adam, Carl, and I wrote the final manuscript, with final approval from all coau-thors.Chapter 3A version of Chapter 3 has been published: Michael VanInsberghe, Hans Zahn,Adam K. White, Oleh I. Petriv, and Carl L. Hansen. Highly multiplexed single-cell quantitative PCR. PLOS ONE, 13(1):118, 01 2018.Parts of this chapter are also part of the aforementioned patent application.Starting with my preliminary designs and with my close collaboration, HansZahn designed the microfluidic chip used in this chapter. Adam White providedadvice towards chip design. Hans and I both contributed towards lithographic pro-cess development to enable fabrication of master molds. I fabricated master moldsand microfluidic devices. Oleh Petriv provided advice towards protocol develop-ment for multiplexed microRNA expression measurements. I designed, performed,and analyzed all device validation and demonstration experiments. I wrote themanuscript with input and editing from Carl Hansen. All coauthors approved thefinal manuscript. Carl Hansen designed research.viChapter 4A version of Chapter 4 has been submitted for publication: Michael VanInsberghe,David J.H.F. Knapp, Michelle Moksa, Hans Zahn, Martin Hirst, Connie J. Eaves,Carl L. Hansen. Single-cell microRNA sequencing of the human hematopoieticcell hierarchy.I designed and fabricated all master molds and microfluidic devices. Hans Zahndesigned and machined a chip loader to pressurize the array of independent PCRindex brews. I designed, executed, sequenced, and analyzed all experiments tooptimize and validate the library construction protocol, with input from MichelleMoksa, Hans Zahn, and Martin Hirst. David Knapp prepared and sorted cord-blood samples and prepared Figure 4.1 and Figure 4.2. I performed, sequenced,and analyzed all hematopoietic cell libraries. I wrote the manuscript with inputand editing from Carl Hansen and Connie Eaves, and approval from all coauthors.Carl, Connie, and I designed research.viiTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiLay Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiiList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiiList of Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvAcknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xviii1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 MicroRNAs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.1.1 Animal microRNA biogenesis and activity . . . . . . . . 31.1.2 MicroRNA isoforms . . . . . . . . . . . . . . . . . . . . 61.2 Technologies for single-cell miRNA expression analysis . . . . . 91.2.1 In situ hybridization . . . . . . . . . . . . . . . . . . . . 121.2.2 Reporter assays . . . . . . . . . . . . . . . . . . . . . . . 141.2.3 Quantitative and digital PCR . . . . . . . . . . . . . . . . 171.2.4 RNA sequencing . . . . . . . . . . . . . . . . . . . . . . 201.3 Microfluidic technologies . . . . . . . . . . . . . . . . . . . . . . 23viii1.3.1 Single-cell manipulation . . . . . . . . . . . . . . . . . . 241.3.2 Microfluidic fabrication . . . . . . . . . . . . . . . . . . 251.4 Hematopoiesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291.4.1 MicroRNAs in hematopoiesis . . . . . . . . . . . . . . . 321.5 Research statement . . . . . . . . . . . . . . . . . . . . . . . . . 332 High-throughput microfluidic single-cell RT-qPCR . . . . . . . . . . 352.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352.2 Materials and methods . . . . . . . . . . . . . . . . . . . . . . . 372.3 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . 522.3.1 Device design . . . . . . . . . . . . . . . . . . . . . . . . 522.3.2 Validation of integrated single-cell RT-qPCR . . . . . . . 582.3.3 Application to measurement of single-cell microRNA ex-pression . . . . . . . . . . . . . . . . . . . . . . . . . . . 632.3.4 Co-regulation of miR-145 and OCT4 in single cells . . . . 672.3.5 Single nucleotide variant detection in primary cells . . . . 672.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 703 Highly multiplexed single-cell quantitative PCR . . . . . . . . . . . 723.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 723.2 Materials and methods . . . . . . . . . . . . . . . . . . . . . . . 763.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 853.3.1 Device design and operation . . . . . . . . . . . . . . . . 853.3.2 Device characterization . . . . . . . . . . . . . . . . . . . 903.3.3 Measurement of single-cell miRNA expression . . . . . . 983.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1034 Single-cell microRNA sequencing of the human hematopoietic cellhierarchy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1054.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1054.2 Materials and methods . . . . . . . . . . . . . . . . . . . . . . . 1074.2.1 Experimental . . . . . . . . . . . . . . . . . . . . . . . . 1074.2.2 Data analysis . . . . . . . . . . . . . . . . . . . . . . . . 1174.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125ix4.3.1 Library construction and performance . . . . . . . . . . . 1254.3.2 Comparison of performance to existing methods for miRNA-seq . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1324.3.3 Generation of single-cell miRNA profiles on 12 purifiedsubsets of human cord blood cells . . . . . . . . . . . . . 1354.3.4 MicroRNA isoforms are dynamically expressed during hem-atopoiesis . . . . . . . . . . . . . . . . . . . . . . . . . . 1454.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1565 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1595.1 Summary of contributions . . . . . . . . . . . . . . . . . . . . . 1595.1.1 Microfluidic architecture for single-cell expression analysis 1595.1.2 Library preparation procedure for single-cell miRNA-seq . 1615.1.3 Single-cell microRNA sequencing of the human hematopoi-etic cell hierarchy . . . . . . . . . . . . . . . . . . . . . . 1625.2 Future recommendations . . . . . . . . . . . . . . . . . . . . . . 1645.2.1 Technology development . . . . . . . . . . . . . . . . . . 1645.2.2 MicroRNAs in development . . . . . . . . . . . . . . . . 1665.3 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . 167Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169Appendix A Supplemental data tables . . . . . . . . . . . . . . . . . . 208xList of TablesTable 2.1 Protocol timing for performing heat lysis and two-step RT-qPCRin the microfluidic system . . . . . . . . . . . . . . . . . . . . 39Table 2.2 Protocol timing for performing chemical lysis and one-step RT-qPCR in the microfluidic system . . . . . . . . . . . . . . . . 42Table 2.3 Specifications for collecting qPCR images . . . . . . . . . . . 45Table 2.4 OCT4 probe sequences for mRNA FISH . . . . . . . . . . . . 47Table 3.1 Comparison of single-cell gene expression methods . . . . . . 74Table 3.2 Performance comparison of single-cell gene expression methods 75Table 3.3 Single-molecule dilution detection measurements . . . . . . . 95Table A.1 miRNA co-expression significance . . . . . . . . . . . . . . . 209Table A.2 Names and sequences for all oligonucleotides used in the single-cell miRNA-seq work. . . . . . . . . . . . . . . . . . . . . . . 216xiList of FiguresFigure 1.1 miRNA biogenesis . . . . . . . . . . . . . . . . . . . . . . . 5Figure 1.2 Mechanisms and consequences of miRNA isoforms . . . . . . 8Figure 1.3 Detection of miRNA isoforms . . . . . . . . . . . . . . . . . 11Figure 1.4 Reporter gene assays . . . . . . . . . . . . . . . . . . . . . . 16Figure 1.5 miRNA RT-qPCR methods . . . . . . . . . . . . . . . . . . . 19Figure 1.6 miRNA-seq library construction . . . . . . . . . . . . . . . . 21Figure 1.7 Soft lithography fabrication process . . . . . . . . . . . . . . 26Figure 1.8 Multi-layer soft lithography . . . . . . . . . . . . . . . . . . 28Figure 1.9 Classic model of hematopoietic development . . . . . . . . . 30Figure 2.1 Mixing by diffusion . . . . . . . . . . . . . . . . . . . . . . . 51Figure 2.2 Design and operation of the microfluidic device for single-cellgene expression analysis . . . . . . . . . . . . . . . . . . . . 54Figure 2.3 Precision and sensitivity of microfluidic RT-qPCR . . . . . . 56Figure 2.4 Histograms showing the size distribution of cultured and trappedcells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58Figure 2.5 On-chip cell washing . . . . . . . . . . . . . . . . . . . . . . 59Figure 2.6 Comparison of GAPDH measurements from K562 cell lysatewith RT performed in the microfluidic device (on-chip) or RTperformed in tubes (off-chip) prior to qPCR in the device. . . 61Figure 2.7 Single-cell loading and GAPDH expression measurements . . 62Figure 2.8 Single-cell miRNA measurements . . . . . . . . . . . . . . . 63Figure 2.9 Measurement of miR-16 in hESC cell aggregates . . . . . . . 64Figure 2.10 Single-cell miRNA measurements . . . . . . . . . . . . . . . 65xiiFigure 2.11 Optical multiplexing of single-cell RT-qPCR . . . . . . . . . 68Figure 2.12 mRNA-FISH of OCT4 in CA1S cells . . . . . . . . . . . . . 69Figure 3.1 Operation schematic for the two-step RT-qPCR workflow . . . 79Figure 3.2 Operation schematic for the one-step RT-qPCR workflow . . . 82Figure 3.3 Multiplexed RT-qPCR device schematic and operation . . . . 87Figure 3.4 Multiplexed RT-qPCR microfluidic device schematic . . . . . 89Figure 3.5 Device characterization and performance . . . . . . . . . . . 92Figure 3.6 Multiplexed RT-qPCR measurement variability . . . . . . . . 93Figure 3.7 Single-molecule cycle threshold cut-off . . . . . . . . . . . . 94Figure 3.8 Multiplexed single-cell mRNA expression . . . . . . . . . . . 96Figure 3.9 Variability of single-cell mRNA measurements . . . . . . . . 97Figure 3.10 Co-expression of endogenous control genes . . . . . . . . . . 98Figure 3.11 Multiplexed single-cell miRNA expression . . . . . . . . . . 100Figure 3.12 Differential miRNA expression . . . . . . . . . . . . . . . . . 101Figure 4.1 Example gating hierarchies for sorted progenitor populations . 108Figure 4.2 Example gating hierarchies for sorted mature populations . . . 109Figure 4.3 Microfluidic device schematic and operation . . . . . . . . . 114Figure 4.4 Custom MiSeq sequencing recipe . . . . . . . . . . . . . . . 118Figure 4.5 Method for single-cell miRNA-seq . . . . . . . . . . . . . . . 127Figure 4.6 Single-cell miRNA-seq method validation and performance . 129Figure 4.7 Correlogram denoting miRNA expression values from repli-cate K562 cell lysate samples . . . . . . . . . . . . . . . . . 130Figure 4.8 Measurement of conversion efficiency . . . . . . . . . . . . . 131Figure 4.9 Comparison of miRNA library quality metrics . . . . . . . . . 134Figure 4.10 Hematopoietic cell type organization . . . . . . . . . . . . . . 136Figure 4.11 t-SNE analysis of hematopoietic cells . . . . . . . . . . . . . 137Figure 4.12 Comparison of methods for determining pseudotime differen-tiation trajectories . . . . . . . . . . . . . . . . . . . . . . . . 138Figure 4.13 Comparison of pseudotime differentiation trajectories . . . . . 139Figure 4.14 miRNA expression in hematopoietic cells . . . . . . . . . . . 142Figure 4.15 Decrease in miRNA diversity and total expression . . . . . . . 143xiiiFigure 4.16 Highlighted miRNA expression . . . . . . . . . . . . . . . . 144Figure 4.17 Cumulative miRNA cut and non-templated base additions . . 146Figure 4.18 Proportional expression of miRNA isoforms during HSC de-velopment . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147Figure 4.19 miRNA isoform putative target repertoires . . . . . . . . . . . 148Figure 4.20 miRNAs with significantly different relative 5′ cut locations . 150Figure 4.21 miRNAs with significantly different relative 3′ cut locations (1of 3) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151Figure 4.22 miRNAs with significantly different relative 3′ cut locations (2of 3) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152Figure 4.23 miRNAs with significantly different relative 3′ cut locations (3of 3) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153Figure 4.24 miRNAs with significantly different relative 3′ non-templatedbase additions (1 of 2) . . . . . . . . . . . . . . . . . . . . . 154Figure 4.25 miRNAs with significantly different relative 3′ non-templatedbase additions (2 of 2) . . . . . . . . . . . . . . . . . . . . . 155xivList of AbbreviationsADAR Adenosine Deaminase, RNA SpecificAGO ArgonauteANK1 Ankyrin 1BaF3 Mouse pro-B-cell lineBCR-ABL Fusion gene between BCR (Breakpoint cluster region protein) andABL1 (ABL Proto-Oncogene 1, Non-Receptor Tyrosine Kinase)BSA Bovine serum albuminCAD Computer aided designCB Cord bloodCCD Charge-coupled devicecDNA Complementary DNACI Confidence intervalCLP Common lymphoid progenitorCML Chronic myeloid leukemiaCMP Common myeloid progenitorCT Cycle thresholdCV Coefficient of variationDGCR8 DiGeorge Syndrome Critical Region Gene 8DIG DigoxigeninDMEM Dulbecco’s modified eagle mediumDNA deoxyribonucleic aciddPCR Digital PCRdsDNA Double stranded DNAEDTA Ethylenediaminetetraacetic acidxvFACS Fluorescence-activated cell sortingFBS Fetal bovine serumFDR False discovery rateFISH Fluorescent in situ hybridizationGAPDH Glyceraldehyde-3-Phosphate DehydrogenaseGFP Green fluorescent proteinGGCX Gamma-Glutamyl CarboxylaseGMP Granulocyte/macrophage progenitorhESC Human embryonic stem cellsHOX HomeoboxHSC Hematopoietic stem cellHSPC Hematopoietic stem and progenitor cellICA Independent component analysisisomiR microRNA isoformJSD Jensen-Shannon distanceK562 BCR/ABL1 positive human cell line derived from a patient withchronic myeloid leukemia in blast crisisLNA Locked nucleic acidsLT-HSC Long-term reconstituting hematopoietic stem cellMEP Megakaryocyte/erythroid progenitormiRNA microRNAmiRNA-seq microRNA sequencingMLP Multilymphoid progenitorMPP Multipotent progenitormRNA Messenger ribonucleic acidmRNA FISH messenger RNA fluorescent in situ hybridizationMSL Multilayer soft lithographyNCC No cell controlNK Natural killerNTC No template controlOCT4 Octamer-bindingtranscription factor4, also known as POU5F1PBS Phosphate buffered salinePCR Polymerase chain reactionxviPDMS PolydimethylsiloxanePLA Proximity ligation assayPOU5F1 POU domain class 5 transcription factor 1pre-miRNA MicroRNA precursors, i.e. the microRNA hairpin structurepri-miRNA Primary microRNA, i.e. the primary microRNA transcriptQC Quality controlqPCR Quantitative polymerase chain reactionRISC RNA-induced silencing complexRNA Ribonucleic acidRNA-FISH RNA fluorescent in situ hybridizationRNAi RNA interferenceRNA-seq Ribonucleic acid sequencing, typically referring to messenger RNARPPH1 Ribonuclease P RNA Component H1RT Reverse transcriptionRT-qPCR Reverse transcription quantiative polymerase chain reactionSD Standard deviationSE Standard errorSNV Single nucleotide variantSP1 Sp1 Transcription FactorSSC Saline-sodium citrateST-HSC Short-term reconstituting hematopoietic stem cellTBP TATA-Box Binding ProteinTCGA The Cancer Genome AtlasTMCS chlorotrimethylsilaneTPM Transcripts per million mappedTRBP Transactivation response element RNA-binding proteint-SNE T-distributed stochastic neighbour embeddingUMI Unique molecular identifierUTR Untranslated regionUV UltravioletxviiAcknowledgmentsFirst and foremost I would like to thank my advisor Carl Hansen for his outstandingmentorship, steadfast support, and much-needed reminders to see the forest for thetrees. I am extremely grateful for the numerous and varied opportunities that youhave provided me; they have been fundamental in shaping my career.I wish to further thank my committee members Martin Hirst, Keith Humphries,and Gregg Morin for their guidance and helpful suggestions along the way. I amalso grateful to the Genome Science and Technology program and the generousfunding agencies who have supported me and my work, including The Universityof British Columbia, The Killam Trusts, and NSERC.To my colleagues and friends in the Hansen group – Daniel Da Costa, Carmende Hoog, Sherie Duncan, Didier Falconnet, Mani Hamidi, Kevin Heyries, JensHuft, Kevin Jepson, Anders Klaus, Tim Leaver, Ve´ronique Lecault, Kaston Leung,Nico Lingg, Kathleen Lisaingo, Georgia Mewis, Keith Mewis, Amanda Moreira,Oleh Petriv, Adam Quiring, Marke´ta R˘ic˘icova´, Darek Sikorski, Anupam Singhal,Marijn van Loenhout, David Walker, Adam White, Andre Wild, Bertin Wong, andHans Zahn – thank you for your insight, assistance, camaraderie, and for creat-ing an enjoyable work environment. Special thanks to fellow “assassins” – Adam,Hans, Oleh, Kevin, Georgia, Marijn, and Kevin – for your invaluable contribu-tions to this work; Carmen for keeping the lab running; Ve´ronique for getting mestarted in the lab; and Hans and Georgia for your encouragement. Thanks to ChrisSherwood and Jamie Piret for tissue culture support.Thank you to my collaborator David Knapp for his unwavering enthusiasm inpreparing samples and without whom the hematopoietic miRNA expression datacould never have been obtained. Also thank you to Michelle Moksa and membersxviiiof the Hirst lab for their sequencing support; you were an instrumental part of theproject.Thank you to the team members and fellow advisors on the UBC iGEM teamsfor making iGEM one of the highlights of my UBC experience. It has been inspir-ing to watch you develop.Thank you to Marek, Mark, Matt, Sherie, Mel, Dave, Kyla, Hans, Georgia,Adam, Sophie, Sibyl, Cam, and Anna for the adventures, support, and friendship.Finally, I would like to thank my parents, Laurie and Terry, and brother, David,for their constant love, support, and curiosity.xixChapter 1IntroductionTechniques for gene expression profiling have had profound impact on the biologi-cal sciences. Spurred by the technical innovations required for genome-sequencingprojects in the early 2000s, as well as the rich information these efforts created, thefield of functional genomics has now made tremendous progress towards under-standing how genes are expressed and regulated [1]. The ability to simultaneouslyanalyze feature sets ranging in size from tens of targets to the entire genome, hasenabled researchers to discover and annotate genes [2, 3], probe complex mecha-nisms [4, 5], and identify tissue types, functions, and processes [6]. In the clinic,gene expression profiles have been applied towards the molecular classification ofdisease types [7, 8], stratification of patients for different treatment regimens [9],and prediction of clinical outcomes [10].Despite these large strides in functional genomics, there remain many questionsthat are best answered by measuring single cells. For example: How many of theapproximately 40 trillion cells in the human body stem from distinct cell types?Is there molecular heterogeneity within cell populations that have been previouslyassumed to be homogeneous? What defines the functional state of a cell? Whatare the co-regulatory gene networks involved in establishing a certain function, andwhat are their constituents? What are the coordinated changes in expression duringa cell state transition? How can we design and optimize synthetic gene circuits totolerate or exploit inherent expression noise?The mainstay technologies behind expression studies in functional genomics1have been reverse-transcription quantitative PCR (RT-qPCR) [11, 12], DNA mi-croarrays [13, 14], and RNA-seq [15, 16]. A tremendous effort has recently goneinto reducing the minimum sample input requirements of these targeted and genome-wide methods so that they may be applied to single cells. Efforts to enable molec-ular analysis at single-cell sensitivity have been primarily focused on messengerRNA (mRNA), leaving an important technical gap in regards to small non-codingRNAs such as microRNAs. There is thus a need for sensitive, scalable technolo-gies that enable the measurement of microRNA expression in single cells. Thework presented in this dissertation addresses this need through the developmentand application of microfluidic and molecular biology methods for the targetedand genome-wide measurement of microRNA expression in single cells.1.1 MicroRNAsCoarsely characterized as coding and non-coding RNA, different classes of RNAtranscripts take on a variety of roles in central cellular processes. Coding RNA,commonly referred to as messenger RNA (mRNA), specifies a coding gene’s aminoacid sequence. Non-coding RNAs take on a much wider range of functions, in-cluding regulating mRNA transcription (e.g., long non-coding RNA) and matura-tion (e.g., small nuclear RNA), post-transcriptionally regulating mRNA expression(e.g., antisense RNA, small interfering RNA, short hairpin RNA, and microRNA),driving protein synthesis (e.g., ribosomal RNA and transfer RNA), and chemicallymodifying RNA nucleotides (e.g., small nucleolar RNA).MicroRNAs (miRNAs) are a class of small non-coding RNA involved in thepost-transcriptional modulation of gene expression. First discovered in 1993 aspart of a gene-regulatory network involved in controlling developmental timingin Caenorhabditis elegans [17, 18], the recognition of miRNAs as a novel classof regulatory RNAs did not occur for another seven years following the discov-ery of a second miRNA gene, let-7 [19, 20]. The high degree of let-7 sequenceconservation strongly suggested that there were other miRNA genes in animals.Indeed, thousands of miRNA genes have since been found in plants, animals, andtheir viruses [21]. A period of intense research interest followed, and significantprogress has been made towards understanding the mechanisms behind their bio-2genesis and function [22, 23], their biological effects [24], and their associationswith disease [25].1.1.1 Animal microRNA biogenesis and activityComplete miRNA biogenesis (Figure 1.1) involves numerous processing steps un-der tight regulatory control [22]. miRNA loci are found in a variety of genomiccontexts, including intergenic regions and the introns and exons of other genes.A considerable fraction (e.g., approximately 30 % in humans) are found in closeproximity to other miRNAs and are expressed as single polycistronic transcripts[26]. Transcriptional units harbouring miRNA sequences are transcribed by RNApolymerase II, creating a long primary miRNA (pri-miRNA) transcript that con-tains one or more local stem-loop structures. The Microprocessor protein complex,comprised of Drosha and DiGeorge Syndrome Critical Region Gene 8 (DGCR8),recognizes these structures and crops the pri-miRNAs at a precise distance from thebase of the stem to create ∼ 65-nucleotide hairpins referred to as a pre-miRNAs.The pre-miRNA then complexes with exportin 5 for translocation through the nu-clear pore. In the cytoplasm, the loop of the pre-miRNA hairpin is cleaved offby the Dicer complex, which contains Dicer and TRBP (transactivation responseelement RNA-binding protein), to release a double-stranded RNA duplex approxi-mately 22 base pairs long. Following Dicer processing, the RNA duplex is loadedinto an Argonaute protein and unwound; the “guide” strand remains bound andthe “passenger” strand is discarded and quickly degraded. Strand selection occursduring the loading process, and is primarily determined based on the relative ther-modynamic stability of the two ends of the RNA duplex.Once complexed with an Argonaute protein, microRNAs serve as the target-ing component of the RNA-induced silencing complex (RISC). Through base-pairinteractions with target messages, miRNAs guide RISC to degrade, destabilize, orrepress the translation of mRNAs [23]. miRNA binding sites are typically locatedwithin the 3′ untranslated regions of mRNAs [28]. The conformation of the guidestrand in the Argonaute protein is such that the region containing bases 2-8 fromthe 5′ end of the miRNA (the “seed” region) is the primary determinant of mRNAtarget recognition [29, 30]. Even though downstream nucleotides can also pair3RNA Pol IIStandard pathway Alternative pathwaym7GpppAAAA(n)DroshaPri-miRNAPasha orDGCR8 Pri-miRNAcroppingmRNA splicingBranched MirtronLariat debranchingNucleusm7GpppExportin 5Pre-miR-451CytoplasmLoqs orTRBPDicerPre-miRNAAGO2miRNA–miRNA*duplexHSC70 and HSP90 Pre-miRNAcleavage and2′ OH2′ OHmiRNA*AGO maturationPre-RISCRISCMature RISCm7GpppORF(n)AAAA4Figure 1.1 (previous page): miRNA biogenesis. Nuclear miRNA biogene-sis begins with RNA polymerase II transcribing a miRNA gene to create aprimary miRNA. The hairpin structure is then cropped from the primary tran-script by the Microprocessor complex (Drosha and DGCR8) to release a pre-miRNA. This short hairpin is then translocated through the nuclear pore byExportin 5. In the cytoplasm, the Dicer-TRBP complex removes the hairpinloop to create a double-stranded RNA duplex, which is subsequently loadedand unwound into an Argonaute (AGO) protein. The guide strand (red) re-mains bound, and the passenger strand (blue) is released and degraded. Oncecomplexed with AGO, miRNAs serve as the targeting component of the RNA-induced silencing complex (RISC), which acts to degrade, destabilize, or re-press translation. Adapted by permission from RightsLink/Springer Nature:Nature Reviews Molecular Cell Biology. Diversifying microRNA sequenceand function. Stefan L. Ameres and Phillip D. Zamore. c©2013. [27]with targets, extensive complementarity with the target message is rare. The rela-tively non-specific nature of these interactions results in the capacity for individualmiRNAs to modulate large cohorts of transcripts [31, 32].Identifying and characterizing the contributions that miRNAs make to gene-regulatory networks and how their dysregulation can result in disease is a bur-geoning area of research [24, 25, 28]. Efforts towards typifying miRNA-mediatedregulation networks have been complicated by their robust, redundant nature. Forexample: many miRNAs belong to multigene families that have identical seed se-quences and are predicted to have identical target repertoires; multiple differentmiRNAs have been seen to act together towards generating an integrated phenotype[33, 34]; and many miRNA:mRNA interactions are context specific with respectto UTR location, tissue type, developmental status, and spatial location. Further-more, as exemplified by the systematic deletion of individual miRNA genes in C.elegans revealing that most are not required for normal development, viability, orbehaviour [35], the phenotypic consequences of manipulation are typically chal-lenging to detect.In spite of these challenges, the functional effects of miRNA:mRNA interac-tions have been observed to range from the potent inhibition and fine-tuning ofindividual targets to the moderate, but coordinated regulation of large target reper-5toires. In general, miRNA regulation only results in a modest decrease to tar-get message and protein abundance [36, 37]. The importance of their influencehas been demonstrated in various processes including regulating gene expressionnoise [38], driving and reinforcing cell fate decisions [39], and affecting neoplastictransformation [40]. Furthermore, the majority of human protein-coding genes arepredicted to be under selective pressure to maintain miRNA pairing [41], suggest-ing that miRNAs modulate the expression of most human proteins. In addition tothese functional roles, classifiers based on miRNA expression profiles have alsobeen shown to outperform those based on mRNA expression [8], leading to theirdevelopment as an important class of prognostic and diagnostic biomarkers [42].1.1.2 MicroRNA isoformsIn general, miRNA gene names are sequentially numbered based on the species inwhich they are discovered with the prefix denoting the organism (e.g., hsa-mir-92from Homo sapiens) [21, 43]. Homologous loci in different species are assignedthe same number (e.g., mmu-mir-92 from Mus musculus). Paralogous miRNAsare assigned names with a numbered or lettered suffix, depending on whether thederived mature miRNA is identical in sequence (e.g., hsa-miR-92a-1-5p and hsa-miR-92a-2-5p) or contains differences (e.g., hsa-miR-92a-3p and hsa-miR-92b-3p), respectively. Precursor products, including the genomic locus, primary tran-script, or hairpin are signified by the lack of capitalization in the “mir”. Tradi-tionally, the most abundant mature products from a locus was denoted with a cap-italized “miR” (also called the guide strand), and the minor products (passengerstrand) with a * suffix (e.g., hsa-miR-92a*). However, as both strands were seen tobe functional [44, 45] and strand preference has been seen to change with cell type[46], this convention was altered to instead add a -5p or -3p suffix depending onwhether the mature miRNA originates from the 5′ or 3′ arm of the hairpin, respec-tively (e.g., hsa-miR-92a-3p). MicroRNA sequences are annotated and curated inthe miRBase database [21].Despite being catalogued as single sequences, small-RNA sequencing data hasrevealed that mature miRNA sequences are heterogeneous (Figure 1.2) [47–49].Structural and functional studies have further shown that these isoforms are capable65’ 3’+1 -1Seed5’ 3’5’ 3’-2 +2A 5’ miRNA isoformsB Internal miRNA isoformsC 3’ miRNA isoformsAUUTrimmingTailingAIAISeedSeedmiR-2-1miR-2-23’ tailed mirtronExosome?Branched mirtronBranched mirtronLariatdebranchingLariatdebranchingA-to-I editing5’ tailed mirtronmaturationmaturationmaturation24 nt 22 ntNibblermiRNA*Loqs orTRBPDicerMonouridylatedmiRNA duplexPre-miRNAUZCCH6, ZCCHC11or GLD-2UUUZCCHC11 DIS3L2LIN28UUUUDistributiveProcessive Pre-miRNA decay5’-to-3’ exonucleolytic5’-to-3’ exonucleolytic5’-to-3’ exonucleolyticParalogousmiRNAsAGO1ADARof differential interaction and regulation of mRNAs [50–52]. The production ofmiRNA isoforms (isomiRs) has been linked to variations in the precise precursorcropping and dicing steps, terminal trimming, and the addition of non-templatednucleotides [27, 51] (Figure 1.2). Isoforms are typically classified as 5′, internal,or 3′ based on the location of the modification with respect to the annotated maturesequence.5′ isomiRs occur as the result of modifications that change the position of the 5′cut relative to the annotated mature sequence (Figure 1.2A). This class of isomiRsis derived from differential processing of paralogous genes, imprecise cropping ordicing, and variable trimming of miRNAs produced via non-canonical maturationprocesses [53]. Precise processing at the 5′ end is thought to be under evolutionarypressure to minimize off-target effects [54]. Aligning with this prediction, as acumulative proportion of the expressed miRNAs, 5′ isoforms are typically rare,totalling less than 10 % of all sequence reads. However, some miRNAs are known7Figure 1.2 (previous page): Mechanisms and consequences of miRNA iso-forms. (A) While certain miRNA robustly yield 5′ isomiRs, they are, in gen-eral, rare. In addition to occurring as a result of alternate processing of thesame locus, 5′ isoforms can be produced via differential processing of par-alogous miRNA (e.g., the miR-2 family in Drosophila melanogaster), or theexonucleolytic processing of 5′ tailed mirtrons. 5′ isomiRs result in a miRNAswith different seed sequences, and thus different target repertoires. (B) Inter-nal isoforms are rare, and primarily generated as a result of ADAR-mediatedadenosine to inosine RNA editing. Edits that occur within the seed sequencewould result in changes to the target repertoire; the effect of edits occurringoutside this region is unknown. (C) The 3′ end is the most variable, withdifferential processing, exonucleolytic trimming, and non-templated base ad-ditions (tailing) all generating isoforms. Trimming is required for the mat-uration of all mirtrons, as well as many fly miRNAs. While 3′ isoformsare generally thought to alter the stability of mature miRNAs, modificationsto pre-miRNAs have been seen to regulate Dicer processing. For example,monouridylation of specific pre-miRNAs has been seen to enhance Dicer sub-strate regulation and processing. In contrast, the presence of LIN28 results inthe oligouridylation of pre-miRNAs containing a LIN28-binding motif, in-hibiting processing by Dicer and enhancing pre-miRNA decay. Adapted bypermission from RightsLink/Springer Nature: Nature Reviews Molecular CellBiology. Diversifying microRNA sequence and function. Stefan L. Ameresand Phillip D. Zamore. c©2013. [27]to reliably generate two or more 5′ isomiRs (e.g., miR-126-3p), and they can be thedominant miRNA produced from certain loci. Because the seed sequence is definedby the 5′ end, 5′ isoforms can profoundly change a miRNA’s target repertoire;target overlap between a 5′ isomiR and its canonical pair is typically limited.Changes to the internal sequence of miRNAs produce internal miRNA iso-forms (Figure 1.2B). Such modifications are almost always the result of adenosineto inosine RNA editing on the double-stranded precursors by ADAR (adenosinedeaminase acting on RNA). In most tissues, this process is rare. Consequently,the frequency of internal isomiRs is comparable to sequencing errors [53], makingthem difficult to detect. Edits that change the seed sequence would alter the targetrepertoire, whereas the functional consequences of edits outside of this region arestill uncertain.8The 3′ end of a mature miRNA is often different from the annotated sequence[27, 55]. The high frequency of 3′ modifications is thought to be as a result ofimprecise cropping or dicing, trimming [56], or non-templated base additions [57](Figure 1.2C). While the functional consequences of 5′ or internal modificationsthat change the targeting seed sequence are well-defined, the significance of thoseinvolving modification of the 3′ end are less clear. The effects of non-templatedbase addition and trimming of mature miRNAs are not well established, but theyare thought to affect the miRNA stability or extent of target repression [22, 27].Depending on the miRNA, the addition of non-templated bases to miRNA precur-sors (pre-miRNAs) has been associated with both positive and negative regulationof dicing, resulting in a downstream up- or down-regulation of the mature miRNA[58, 59]. In general, studies to determine the functional consequence of 3′ mod-ifications have been largely based on studies of individual miRNAs and have notbeen generalized. Further investigations are needed in order to reveal how trim-ming, tailing, and aberrant processing contribute to miRNA regulation.1.2 Technologies for single-cell miRNA expressionanalysisAt the start of this work, techniques for measuring RNA expression in singlecells were in their infancy. Most existing methods suffered from various techni-cal limitations, including only being demonstrated on cells with higher than typ-ical levels of RNA [60], not providing quantitative measurements [61], failing torobustly detect highly expressed endogenous reference genes [62], and having lim-ited throughput in terms of the number of cells analyzed per experiment [63]. Therehas since been intense interest in expanding on these early efforts. A variety of so-lutions have been proposed to address the main technical challenges associatedwith measuring RNA expression in single cells: capturing single cells, amplify-ing and quantifying the minute amounts of RNA, and scaling to representativecell numbers. Obtaining genome-wide mRNA expression measurements in hun-dreds to thousands of individual cells per experiment is now almost routine andis available in several commercially available solutions (e.g., Fluidigm C1, 10×Genomics Chromium, Illumina/Bio-Rad SureCell). In spite of these tremendous9advances, progress towards adapting methods for miRNA expression analysis forsingle-cell studies has lagged behind.miRNAs possess several unique features that present additional technical chal-lenges to their accurate detection and quantification [64, 65]. The 22-nucleotidelength characteristic of mature miRNAs is too short for analysis using traditionalRT-qPCR or other primer-based amplification methods. Additionally, their shortlength and inherent differences in GC content makes it difficult to design probesand standardize melting temperatures for hybridization-based methods such as mi-croarrays or fluorescent in situ hybridization (FISH). Furthermore, the lack of acommon sequence element, such as a poly-A tail, prevents the use of selective anduniversal enrichment strategies. Finally, there are multiple sources of moleculeswith either identical or very similar sequences that must be distinguished: pri- andpre-miRNA precursors and miRNA isoforms contain the mature sequence but arefunctionally different, and there are several examples of distinct mature miRNAsthat differ by a single nucleotide (Figure 1.3).In contrast to these technical challenges, several features make miRNAs at-tractive candidate biomolecules for single-cell expression profiling. First, eachmiRNA’s large pool of possible targets [66], compounded with a typically mod-est degree of target repression [36, 37], requires that miRNAs be expressed at arelatively high minimum copy number (∼ 10 to 100 copies per cell) for effectivegene regulation [67–69]. This minimum functional expression level reduces thedemands on assay sensitivity. Second, miRNA expression signatures may providea more reliable depiction of cell type and state than mRNA signatures. miRNAs aregenerally regarded as stable molecules, having an average half-life that is approxi-mately two times higher than the average mRNA [70, 71]. This increased stability,coupled with the extensive processing and association with protein complexes re-quired for them to be functional, may buffer the effects of transcriptional burstingthat are characteristic of mRNA expression measurements [72]. Furthermore, incontrast to mRNA, which are intermediate messages whose abundance only looselycorrelates with final protein amount [71], miRNA are effector molecules as the tar-geting component of the RNA induced silencing complex. Third, the combinedreduced gene family size and miRNA sequence length facilitates the scalable andcomprehensive sequence analysis on hundreds to thousands of single cells. For10  GAGGUAGUAGGUUGUAUAGUUU UGAGGUAGUAGGUUGUAUAGUU  AUGAGGUAGUAGGUUGUAUAGU   UGAGGUAGUAGGUUGUAUAGU   UGAGGUAGUAGGUUGUAUAG   UGAGGUAGUAGGUUGUAUAGUUA   UGAGGUAGUAGGUUGUAUAGUUUU   UGIGGUAGUAGGUUGUAUAGUU   UGAGGUAGUAGGUUGUIUAGUU  Internal isoforms3’ isoforms5’ IsoformsImprecise processing ortrimmingImprecise processing ornon-templated base additionsImprecise processingRNA editingAnnotated sequenceA B  DetectRT-qPCRHybridizationSequencingReporterC  DistinguishRT-qPCRSequencingHybridizationReporterFigure 1.3: Detection of miRNA isoforms. (A) Examples of the sequencechanges that result in 5′, internal, and 3′ isoforms. The seed sequence for eachmiRNA is highlighted in grey. (B) The ability of each class of miRNA expres-sion measurement assay to detect each isoform. As the hybridization, reporter,and RT-qPCR assays are typically designed using the annotated sequence, thedetection ability denoted here is determined based on these specific assays.Isoforms that are detected contribute to the overall expression measurementdetermined by a given assay. (C) The ability to design an assay to distinguishan isoform from its annotated form for each measurement class. For B and C,green denotes yes, red denotes no, and yellow can occur in specific situations.example, depending on the application, bulk-scale mRNA transcriptome analysisby high-throughput sequencing typically requires 5 to 100 million reads per repli-cate [73]. Depending on tissue type, comprehensive small-RNA profiling, on theother hand, can be achieved using 1-5 million mapped reads. As current single-cell mRNA sequencing methods reach library saturation at 2 million reads [74],it follows that comprehensive miRNA expression profiles could be economicallygenerated in a large number of cells in a single sequencing experiment.Expression profiling is an indispensable component of miRNA research. Assuch, a variety of targeted and genome-wide techniques have been developed toenable miRNA expression profiling (reviewed in [64, 65]). For tissue-level anal-11yses, Northern blots, reverse transcription quantitative polymerase chain reaction(RT-qPCR), microarrays, and RNA sequencing (RNA-seq) have emerged as themost widely used methods. Only a subset of these approaches, however, meets therequirements for single-cell analysis with respect to sample throughput, detectionsensitivity and specificity, and assay multiplexability. To date, miRNA expressionin single cells has been measured using single-molecule imaging, reporter assays,RT-qPCR, and RNA-seq.1.2.1 In situ hybridizationFluorescent in situ hybridization (FISH) is a molecular imaging technique usedto localize and quantify nucleic acids in their native contexts [75]. Compared toalternate methods for transcript quantification, FISH maintains the spatial localiza-tion of cells within tissues, and transcripts within cells. In this method, specificnucleic acid sequences can be detected by hybridization of labelled RNA or DNAprobes, followed by fluorescence microscopy. Initially developed to replace ra-dioactive labelling, improvements in labelling chemistries and imaging capabilitieshave transformed it from being restricted to detecting highly abundant targets to en-abling multiplexed single-molecule detection [76, 77]. These improvements havebeen recently further combined with multiple rounds of hybridization and imag-ing and error-correcting hybridization codes such that it is now possible to obtaintranscriptome-scale integer counts of mRNA copy numbers in tens of thousands ofindividual cells per experiment [78, 79].In order to perform single-molecule RNA-FISH, enough fluorescence must begenerated from each molecule such that it can be detected above the background.As this is not currently possible with single fluorogenic molecules, a variety ofamplification strategies have been developed. Commonly used strategies includelinking probes to enzymes that catalyze chromogenic or fluorogenic reactions [80],using heavily labelled probes (e.g., five 50-nt oligonucleotide probes each labelledwith five fluorogenic molecules) [81], or a large number of singly labelled probes(e.g., 30-50 20-nt probes) [77]. For mRNA detection, the last is the most widelyused due to its increased sensitivity, more uniform signal, and relative ease of probedesign, synthesis, and purification. The short length of miRNAs, however, com-12plicates the extension of these methods towards miRNA detection; they can onlyaccommodate the hybridization of one or two probes.Nucleic acid analogues with increased hybridization affinities have become anindispensable component of hybridization-based methods for analyzing miRNAs.By locking the conformation of a ribonucleotide through an O2′ to C4′ link, lockednucleic acids (LNA) demonstrate mismatch discrimination equal or superior to na-tive nucleic acids, but with significantly increased thermostabilities [82, 83]. Forexample, each modified monomer in an oligonucleotide raises the melting tem-perature against complementary RNA by 2 to 10 ◦C, and a fully modified 9-merhas a melting temperature of approximately 80 ◦C. LNA-modified probes againstmiRNAs are thus able to maintain the sequence specificities expected from longerDNA probes, while increasing the sensitivity of detection [84, 85].Several methods that use in situ hybridization to measure miRNA expressionhave been demonstrated [86–91]. Despite these improved hybridization propertiesof LNA probes, due to differences in the kinetics of probe hybridization, observedsignal intensities do not directly correlate with the miRNA abundance [89]. Thus,in the absence of integer single-molecule counts, it is generally only possible tomake intra-assay comparisons. Of these reports, only one had the resolution suf-ficient to visualize single miRNA molecules in single cells. Lu et al. stained fixedcells with LNA probes labelled with digoxigenin (DIG) at the 3′ end. Hybridizedprobes were then hybridized with an anti-DIG conjugated peroxidase and fluores-cence was generated using the ELF-97 (Enzyme-Labelled Fluorescence) substrate[87]. They demonstrated their approach by measuring the expression of miR-15 inHeLa (human cervical carcinoma) and MDA-MB-231 (human breast adenocarci-noma) cells, and miR-155-5p in MCF-7 (human breast adenocarcinoma) cells. Intheir analysis of 430 single cells, miRNAs were found to be uniformly distributedthroughout the cytoplasm, and within each cell line, the distribution of miRNAswas slightly skewed with a few cells having copy numbers much higher than themean. Individual miRNAs were directly countable at expression levels between 0and ∼ 1000 molecules per cell, at which point the individual spots were no longerdiscernable. Contrary to previous reports, the total fluorescence intensity per cellcorrelated with abundance, which was used to estimate copy number for highlyexpressed miRNA. They further observed a high concordance in the mean copy13number per cell measured with their method and with RT-qPCR. This method hasbeen adopted by several different groups and used as a tool to validate expressionin diverse biological systems.Instead of amplifying the fluorescent signal from each molecule, Koo et al.used atomic force microscopy (AFM) for detection [92]. After cell fixation, miRNAswere immobilized and hybridized with complementary DNA probes. AFM tipswere functionalized with the N-terminal domain of human RNase H1 that bindsRNA/DNA hybrids in a sequence independent manner. The adhesion force be-tween this domain and the miRNA/DNA hybrid was measured across the surface,thereby localizing individual miRNAs. While their demonstration was very limitedin scope (one miRNA measured over a 1 µm2 area of two single cells), it is possiblethat further development will enable larger analyses.In addition to imaging, in situ hybridization approaches have also been appliedto analyze miRNA expression using flow cytometry [93, 94]. The assay developedby Wu et al. used dual-DIG labelled LNA probes as a substrate for a proximityligation assay (PLA) that created a circular product, which was subsequently am-plified using rolling circle amplification (RCA). Fluorescently labelled detectionoligonucleotides were then hybridized to the RCA product, and cell fluorescencewas measured using a custom microfluidic device [94]. They measured the expres-sion of miR-155-5p in an activation model of Jurkat cells (T lymphocytes from anacute lymphoblastic leukemia). Porichis et al. used a branched DNA assay to mea-sure miR-155-5p and -21-5p in an activation model of primary T cells [93]. Thisassay uses a set of oligonucleotide probes that are sequentially hybridized to eachother in order to increase the binding area for the terminal fluorescently labelledprobe [95]. While expression analysis using flow cytometry loses the spatial infor-mation that is preserved in imaging approaches, it has the benefits of substantiallyhigher cell throughput. Despite their potential, neither of these methods has yetprogressed beyond proof-of-principle demonstrations.1.2.2 Reporter assaysReporter transgenes have been invaluable for visualizing expression patterns andstudying gene regulation. These genes code for proteins with an easily detectable14activity that has little effect on the host cell; commonly used examples includefluorescent proteins, luciferase, β -galactosidase, or antibiotic resistance markers.Similar to FISH, reporter genes can preserve spatiotemporal patterns in expressionmeasurements. However, unlike FISH, they are able to do so in live cells and invivo, but require genetic manipulation prior to analysis. For studies on proteinlocalization, the coding sequence for the reporter is either fused to the gene ofinterest, or to the promoter if it is known.Compared to other methods for measuring miRNA expression, adapting reporter-gene assays has been straightforward. Furthermore, these assays are unique in theirability to provide a read-out of miRNA activity, rather than simple expression.Early efforts to assay miRNA expression with reporter genes used promoter-fusionconstructs [96]. As this requires previous knowledge of promoter architecture andis unable to detect post-transcriptional regulation, an alternate approach was de-veloped to use the ability of miRNAs to cleave complementary mRNA targetsthrough RNA interference (RNAi) [97]. Instead of coupling reporter output tomiRNA regulatory elements, reporter genes are constitutively expressed and con-tain sequences complementary to a given miRNA in their 3′ untranslated region(UTR) [98, 99] (Figure 1.4A). Thus, the reporter gene is expressed in cells notcontaining the miRNA of interest, whereas its expression is absent or reduced inthose with robust miRNA expression. Recent updates to this method have aimedto control for reporter expression and expression noise by using two fluorescentreporters expressed using either identical constitutive promoters [100], or constitu-tive bidirectional promoters [38, 68, 101] (Figure 1.4B).For example, Gentner et al. assayed the activity of a panel of miRNAs in themajor classes of stem and progenitor stages of hematopoietic lineage differenti-ation using bidirectional reporter vectors [101]. These vectors contained desta-bilized green fluorescent protein (GFP) with miRNA-target sequences in the 3′UTR as the reporter, and a low-affinity nerve growth factor receptor (NGFR) asan internal normalizer. The activity of the miRNA was quantified using flow cy-tometry by calculating the fold repression of the miRNA reporter compared to thecontrol. By combining the standard hematopoietic purification markers with thiscytometric assay for miRNA expression, they were able to measure a cell type thatis typically too rare for other techniques. Ultimately, they were able to identify15ZsGreen mCherry 3’UTRpTRE-TightmRNAproteinmicroRNAAAAAAβ-gal onAAAAAβ-gal offA BFigure 1.4: Reporter gene assays. (A) Single-reporter sensor design. Sitescomplementary to the miRNA of interest are added to the 3′ UTR of a lacZreporter. β -galactosidase is constitutively expressed in cells lacking the com-plementary miRNA (top), resulting in a blue colour. In cells expressing themiRNA (red), the lacZ mRNA is targeted for degradation, resulting in an ab-sence of blue. Reprinted by permission from RightsLink/Springer Nature:Nature Genetics. microRNA-responsive ‘sensor’ transgenes uncover Hox-like and other developmentally regulated patterns of vertebrate microRNAexpression. Jennifer H. Mansfield, Brian D. Harfe, Robert Nissen, John Obe-nauer, Jalagani Srineel, Aadel Chaudhuri, Raphael Farzan-Kashani, MichaelZuker, Amy E. Pasquinelli, Gary Ruvkun, Phillip A. Sharp, Clifford J. Tabin,and Michael T. McManus, c©2004. [99]. (B) Dual-reporter sensor design.Two fluorescent proteins (e.g., ZsGreen and mCherry) are transcribed froma common bidirectional promoter. Sites complementary to the miRNA ofinterest are added to the 3′ UTR for one of the proteins (e.g. mCherry).miRNA activity is measured using the ratio of this protein (mCherry) to thenon-targeted one (ZsGreen). From Jo¨rn M Schmiedel, Sandy L. Klemm, Yan-nan Zheng, Apratim Sahay, Nils Blu¨thgen, Debora S. Marks, and Alexandervan Oudenaarden. microRNA control of protein expression noise. Science,348(6230):128-132, 2015. Reprinted with permission from AAAS. [38]a subset of miRNAs that are strongly expressed in hematopoietic stem cells anddown-regulated in the differentiated progeny cells.Mimicking the typical organization of miRNAs and their targets in intercon-nected gene regulatory networks, Xie et al. engineered a synthetic gene circuit tointegrate the expression of several miRNAs into a single output [102]. Using com-binations of activators and repressors, the circuit was designed such that when theexpression of six endogenous miRNAs matched a predetermined reference signa-ture containing highly or lowly expressed miRNAs, the cell classifier circuit wouldtrigger an output. This approach was demonstrated in the selective detection and16apoptosis of HeLa cells.1.2.3 Quantitative and digital PCRQuantitative PCR, or real-time PCR, is an extension of the polymerase chain reac-tion where the reaction progress is monitored at every cycle [11]. By comparinghow many cycles it takes each sample to accumulate a specified amount of ampli-con (the cycle threshold, CT), the relative starting abundances can be measured.This technique can be further extended to measure RNA abundance [12] throughthe simple addition of a reverse transcription (RT) step to convert RNA into com-plementary DNA (cDNA) prior to amplification. Owing to its high specificity,single-molecule sensitivity, large dynamic range, rapid assay time, ease of imple-mentation, and low cost, qPCR is widely regarded as the gold-standard in mRNAand DNA quantification.During the reaction, DNA abundance is measured using fluorogenic molecules,the most common of which are intercalating dyes and 5′ nuclease probes (reviewedin [103]). Intercalating dyes (e.g. SYBR Green and EvaGreen) fluoresce when as-sociated with double stranded DNA (dsDNA). Nuclease probes (also called hydrol-ysis or TaqMan probes) are oligonucleotides functionalized with a fluorophore anda non-fluorescent quencher. When in close proximity, energy is transferred fromthe fluorophore to the quencher, which disperses the energy as heat instead of light.The probes bind to the amplicon during the annealing stage, and as the DNA poly-merase progresses during the extension phase, the 5′ to 3′ exonuclease activity hy-drolyses the probe, separating the fluorophore and quencher, ultimately causing anincrease in fluorescence. On the one hand, intercalating dyes are simpler and lessexpensive, but on the other hand, hydrolysis probes are more specific, can detectsingle-nucleotide differences, and can be used to simultaneously monitor differentamplicons within the same reaction through the use of differing fluorophores.Despite the numerous benefits, qPCR does have important limitations: it doesnot provide absolute quantification of molecular abundance; measurement preci-sion is limited to approximately 10 %, even under optimal conditions; and thepractical limit of sensitivity is often determined by issues of contamination or non-specific amplification, both of which are aggravated by large reaction volumes.17Digital PCR (dPCR) is a simple extension of qPCR that provides a solution tothese drawbacks [104, 105]. In dPCR, a sample is first partitioned into multiplereactions at limiting dilution and then subjected to PCR amplification followed byend-point detection. The presence or absence of starting template molecules gener-ates a “digital” signal for each reaction, and the number of positive reactions can beused to determine the absolute concentration of starting template molecules with-out the need for a standard [106]. The recent development of microfluidic systemsthat simplify the sample partitioning and dramatically increase the number of reac-tions per sample [107, 108] have removed significant technical barriers, leading toits more widespread adoption.RT-qPCR chemistries designed to measure mRNA expression are not directlyapplicable to miRNA. Two strategies have been developed to extend a maturemiRNA such that it meets the length requirements of an RT-qPCR amplicon (Fig-ure 1.5). In the first approach, a poly-A tail is enzymatically added to the 3′ endof mature miRNAs, which is then used as a binding site for an anchored poly-Treverse transcription (RT) primer that contains a universal amplification sequence.qPCR amplification occurs in the presence of an intercalating reporter dye with oneprimer specific to the miRNA of interest, and the universal amplification sequenceas the other [109] (Figure 1.5A). The second technique uses an RT primer designedto fold into a stem-loop structure with a short 3′ overhang that specifically annealsto the 3′ end of a mature miRNA during cDNA synthesis. Quantitative PCR isthen performed using a miRNA-specific forward primer and fluorescently labelledprobe (e.g., TaqMan probe), and a reverse primer that binds to the stem-loop RTprimer [110] (Figure 1.5B). RT-qPCR assays to measure miRNA abundance haveseen widespread adoption, owing to their commercial availability, similar experi-mental workflow to standard qPCR assays, and compatibility with existing instru-mentation and analysis procedures. Of these approaches, miRNA stem-loop RT-qPCR is the most widely used, as it is amenable to highly multiplexed experiments,is able to discriminate single-nucleotide differences, and has single-cell sensitivity[63, 111–115].Tang et al. demonstrated the stem-loop RT-qPCR method for the highly mul-tiplexed (220-plex) profiling of microRNA expression in single embryonic stemcells [63]. Single cells were manually picked using a micro-capillary and lysed18Q FTaqMan probeReverseprimerForward primerStep 2:Real-time PCRStep 1:Stem-loop RTcDNAmiRNART primerBAAAAAAAAA(n)  AAAAAAAAA(n)VNTTTTTTTTTAnnealingPolyadenylationmiRNA  AAAAAAAAA(n)VNTTTTTTTTTReverse transcriptionDigestion of RNAsPCR or qPCRVNTTTTTTTTTVNTTTTTTTTTAFigure 1.5: miRNA RT-qPCR methods. (A) Poly-A method. Small RNAs(brown) are first polyadenylated using a poly(A) polymerase, then reversetranscribed using a long, anchored RT primer (blue). qPCR is then performedin the presence of an intercalating dye using a miRNA-specific forward primer(green) and universal reverse primer (blue) [109]. (B) Stem-loop RT-qPCR.Stem-loop RT primers (blue) first bind to the 3′ end of miRNAs (brown) andare used for cDNA synthesis (green). qPCR is performed using a miRNA-specific forward primer, a hydrolysis probe (TaqMan probe), and stem-loop-specific reverse primer. Panel B reproduced with permission from 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-loop RT-PCR.Nucleic Acids Research, 33(20):e179, 2005 [110].by heating. cDNA was next synthesized from the total cell lysate in a multiplexedreverse transcription reaction. In order to maintain assay sensitivity, a multiplexedPCR pre-amplification was then performed. Finally, samples were diluted, split,and analyzed in simplex against each specific target. A total of twenty single cellswere analyzed, five of which were assayed against all 220 miRNAs. This same ap-proach was later applied towards the miRNA profiling of twelve oocytes as part ofa study demonstrating that maternal miRNAs are essential during the early stagesof mouse development [113].19Petriv and Kuchenbauer et al. used the stem-loop RT-qPCR assay in theircomprehensive study of miRNA expression in the mouse hematopoietic hierar-chy [111]. In addition to the 41 independent samples that were each assayed induplicate against 288 miRNAs, they also measured the expression of 8 miRNAs in80 single cells. This dramatic increase in the sample and assay throughput wasachieved by implementing the final qPCR step in high-throughput microfluidicqPCR arrays (i.e., Fluidigm 48.48 Dynamic Arrays) [116]. While these microflu-idic qPCR arrays simplified the final detection step, they did not address either thelaborious step of manual single-cell isolation, or the imprecisions introduced byperforming nucleic acid molecular biology with low template concentrations.1.2.4 RNA sequencingRNA sequencing (RNA-seq) uses high-throughput DNA sequencing methods tomap and quantify the complete set of transcripts in a cell, commonly referred toas the transcriptome (reviewed in [16]). There are three main steps involved in atypical RNA-seq experiment. First, a library is constructed to convert the RNA intoa format that is compatible with a next-generation sequencer. In order to match theclass of RNA (e.g., mRNAs, non-coding RNAs, small RNAs, ribosomal RNAs) tothe experimental question being asked, enrichment or depletion steps are typicallyperformed prior to or during library construction. Next, the library is sequenced,generating a large number of sequence reads, each derived from a single moleculein the library. Finally, the resulting reads are bioinformatically analyzed in order toobtain the sequence and expression level for each gene. Expression is inferred byeither counting the number of reads or the number of unique molecular identifiers(UMIs) [117, 118] that align to a given feature. UMIs tag each single moleculein the library with a random sequence label. As they are added prior to amplifica-tion, UMI counts reduce the effects of amplification biases, greatly improving thequantitative accuracy of mRNA-seq.Compared to other methods for genome-wide transcriptome analysis, RNA-seqhas benefits of permitting de novo analyses, having a substantially expanded dy-namic range, and requiring less input sample material. It has therefore become themethod of choice for transcriptomics studies and those with discovery or profiling205’App ddC5’Phos 3’ OH3’ Ligation5’ LigationReverse transcriptionIndexing PCR+ T4 RNA ligase 2, adenylated 3’ adapter (DNA)+ ATP, T4 RNA Ligase 1, 5’ adapter (RNA)+ RT brew, RT primer+ PCR brew, reverse primer, sequencing-indexed primerPurification, (concatination, cloning), sequencingFigure 1.6: miRNA-seq library construction. Library construction typicallyproceeds through two single-stranded RNA ligations to attach 3′ (blue) and5′ (green) adapters to miRNA molecules (black). These adapters are used forsubsequent reverse transcription (purple) and PCR (grey-red-orange). PCR-amplified libraries are the purified to remove adapter dimers and sequencedon a next-generation sequencer, or concatenated, cloned, and sequenced usingcapillary sequencers.aims. Soon after RNA-seq analysis became commonplace, it was recognized thatthe combination of RNA amplification methods with whole-transcriptome analysishad tremendous potential for the study of single cells. The first single-cell tran-scriptomes were published shortly after the introduction of RNA-seq [60], whichkicked off a technological ‘arms race’ as researchers and companies developed amyriad of different single-cell mRNA-seq approaches (reviewed in [119, 120]).Unlike the numerous methods now available to sequence messenger RNA, themethods for construction of miRNA libraries have remained essentially unchangedsince the first demonstrations with capillary sequencing [121]. However, severalincremental protocol modifications have reduced the number of intermediate pu-rifications needed, increased the overall efficiency, and altered reaction conditions21to prevent chaining or circularizing adapters during the ligations [47, 122]. Libraryconstruction progresses through two consecutive single-stranded ligations that at-tach priming sites for subsequent cDNA synthesis and PCR (Figure 1.6).There are some important limitations of miRNA-seq to note. First, there arestructure-specific biases in ligation efficiency associated with the T4 RNA ligasesused to create these libraries [123, 124]. These biases make it difficult to compareresults prepared using different adapter sequences. Efforts to reduce these biasesthrough degenerate adapter sequences have been proposed [125], but have failedto gain traction. Second, despite the enrichment steps, small RNA sequencinglibraries can see significant contamination of adapter sequences and RNA degra-dation products. Enrichment for miRNAs partially occurs through the ligations, asRNA degradation intermediates do not typically have a 5′ phosphate and a 3′ hy-droxyl, and partially though a final size-selection step. Even though miRNAs aremore stable than mRNAs, the specific details in the methods of sample collection,RNA extraction, and library preparation can have a substantial impact on the finalresults. As with any method in genomics, proper quality control is an importantstep in collecting relevant data.The effects of the inefficiencies that accumulate at each step of small RNA li-brary construction have made these methods poorly suited for single-cell analysis.Recently, Faridani et al. described modifications to the library construction proto-col that allowed them to sequence small RNAs in single cells [126]. By combin-ing an increase in the ligation efficiency, an enzymatic degradation step to reduceadapter contamination, a substantial increase in the PCR amplification, and deepsequencing, they were able to recover miRNA expression profiles from 398 cellsfrom 7 disparate cell lines. Their analysis of the resulting miRNA expression pro-files showed that compared to mRNA and other classes of small RNAs, miRNAswere able to robustly distinguish cell types. Unfortunately, standard measures ofsample and library quality were not provided, making it difficult to confirm theirresults or to compare the performance of their method to established implementa-tions for bulk analysis.221.3 Microfluidic technologiesMiniaturization has been a powerful approach taken towards developing single-cell genetic assays. As previously mentioned, the main technical challenges facingsingle-cell analysis are scaling to a large number of reactions, increasing reactionsensitivities, and isolating individual cells. Microfluidic integration presents solu-tions to each of these challenges.By reducing the characteristic unit volume from microlitres down to nanolitres,microfluidics is able to reach higher levels of throughput without increasing the re-source requirement. The three main techniques used towards miniaturizing fluidicsystems for the genetic analysis of single cells are micro-droplets, microfluidicchips, and open microscale arrays.Droplet-based microfluidic systems generate highly monodisperse emulsionsby using simple channel geometries to bring two immiscible phases together ina controlled, predictable manner (reviewed in [127]). Thousands of droplets canbe generated per second, and systems have been developed to split, merge, andsort individual droplets at scale. In terms of sample throughput, droplet-based mi-crofluidic systems are currently unmatched; however, they are typically unable toperform the complicated multi-step procedures required for certain analysis proce-dures.Microfluidic chips contain enclosed networks of tubes, valves, and reactionchambers that have been configured to perform a specific assay (reviewed in [128]).The ability to easily integrate thousands of valves within one of these devices [129]facilitated the design of a wide array of functional elements that allow absolute con-trol over fluid flow. As a result, incredibly complex workflows can be miniaturizedand parallelized. On the other hand, this level integration comes at the expense ofdevice complexity and sample throughput as the device size scales with the numberof reactions and is constrained by practical fabrication requirements.Open microscale arrays are planar surfaces on which individual reactions areconstrained by either physical barriers [130] or surface-tension modifications [131].These systems are effectively miniaturized well-plates, with fluid control managedusing off-chip systems such as micro-capillaries or non-contact droplet spotters.Devices are thus drastically simplified, while providing the flexibility to imple-23ment different protocols, and in some cases, permitting higher reaction throughputthan with microfluidic chips. External fluid handling systems currently assemblereactions in serial, thus potentially adding significant delays to processing time thatmay alter cell states and introduce data artefacts. Extreme care must also be takento avoid external contaminants and sample cross-contamination during reaction as-sembly, incubation, and recovery [132].In addition to reducing reagent consumption and increasing reaction through-put, nanolitre reaction volumes generally outperform microlitre volumes when ini-tial template quantities are limited. The increase in template concentration achievedthrough the decreased volume brings the reaction conditions into closer alignmentwith their traditional, microlitre-scale implementation. This simple change hasbeen seen to reduce unwanted reaction side products and contamination [133], andincrease reaction efficiency [134], sensitivity [115], and reproducibility [131]. Itshould be noted that smaller reaction volumes are not necessarily always better.Cell lysate has a well-known inhibitory effect on common assays in molecular bi-ology; this is one of the reasons that benchtop protocols typically involve an RNAor DNA purification prior to further processing. Because such purifications are notgenerally amenable to microfluidic integration, care must be taken to ensure thatcell lysate is diluted enough to counteract these inhibitory effects [62, 115].1.3.1 Single-cell manipulationThe characteristic length scales used in microfluidic devices are ideally suited formanipulating single cells. Using standard fabrication procedures, microfluidic fea-ture heights range from roughly one micron to hundreds of microns, and lengthsvary from one micron up to several centimetres [135]. Similarly, micro-dropletsrange in size from five to hundreds of microns in diameter. The average diame-ter of a single eukaryotic cell ranges from approximately four to forty microns,nicely aligning with this size range. Consequently, several approaches have beendeveloped to trap and localize single cells (reviewed in [136]), of which the mostcommonly used ones include stochastic partitioning, active capture, and hydrody-namic focusing.In stochastic partitioning, the input concentration of cells is adjusted such that24once they are apportioned there is an average of less than one per reaction. Whilethis approach benefits from its simplicity and compatibility with a wide range ofcell sizes, the optimal cell loading concentration results in a large number of emptyreactions. It is therefore most commonly used in situations where an alternatetechnique does not exist (i.e., microscale arrays), or those that can easily scale toaccommodate the loss of reactions (i.e., micro-droplets [137]).Active capture involves manually inspecting a cell suspension while control-ling fluid flow using integrated valves, pumps, and demultiplexors [128, 129] toselect an individual cell and direct it to a specific reaction chamber. In additionto not having cell-size restrictions and ensuring complete reaction occupancy, thisapproach gives users complete control over which cells are sent for analysis, per-mitting the selection of specific attributes such as viability, morphology, or surfacephenotype. Due to its reliance on active microfluidic elements, active capture canonly be implemented in multilayer microfluidic chips [138–140].Hydrodynamic focusing exploits the well-defined properties of laminar fluidflow through microchannels to direct and keep single cells in physical trappingsites. Several different strategies have been used to this effect, including parallelfluid streams [141], filters [142], cups [143, 144], and balanced flow resistancesthrough paired traps and bypass channels [145, 146]. Of these approaches, theconfiguration of cell cups and some balanced-flow-resistance implementations al-lows their highly parallel integration with downstream cell-processing capabilities[115, 145]. By their very nature, the dimensions of all hydrodynamic-focusingapproaches must be tuned to match the approximate size of the cells being ana-lyzed. For example, capturing cells in a trap designed to work with larger cellshas been seen to result in a high proportion of traps containing more than one cell[147]. Furthermore, the small dimensions these features use can lead to cloggingif the cell population has particularly heterogeneous morphologies, or is prone toclumping.1.3.2 Microfluidic fabricationReplica moulding is the dominant fabrication technique used for prototyping mi-crofluidic devices. First introduced in the late 1990’s, the technique of soft lithogra-253. Expose photoresist-    coated substrate    to UV light2. Print mask1. CAD drawing4. Develop substrate6. Final master mouldA5. Repeat steps 2-4 with    different masks and    photoresist heightsB 12345Figure 1.7: Soft lithography fabrication process. (A) Master mould fabri-cation. Device designs are first drawn using CAD software. Each differentfeature height is printed onto a separate high-resolution mask. A planar sub-strate (i.e., a silicon wafer, grey) is next coated with photoresist (maroon), andexposed to UV light; exposed areas are polymerized. The substrate is thendeveloped, which washes away unexposed areas. These steps can be succes-sively repeated in order to build up different feature heights (blue, green) onthe final master mould. (B) Single-layer soft lithography. PDMS prepolymeris cast on the master mould and cured. The PDMS replica is then removed andsealed onto either a blank PDMS substrate or glass [148]. Access ports canalternatively be punched after removing the PDMS replica from the master(after step 4) instead of placing posts to create access ports (step 2). Panel Breprinted with permission from David C. Duffy, J. Cooper McDonald, OlivierJ. A. Schueller, and George M. Whitesides. Rapid prototyping of microfluidicsystems in poly(dimethylsiloxane). Analytical Chemistry, 70(23):4974-4984,1998. c©1998 American Chemical Society. [148]26phy quickly supplanted existing fabrication techniques involving micromachiningglass or silicon due its substantial improvements to fabrication ease, time, and cost[148]. There are two main steps involved in the soft lithography design and fab-rication process: fabricating a master mould that is subsequently used to cast themicrofluidic chip (Figure 1.7).Master moulds are fabricated using standard photolithography techniques (Fig-ure 1.7A). First, the design for a microfluidic device is drawn in 2D using computer-aided design (CAD) software. This design is then transferred to a high-resolutiontransparency called a mask, where, for example, clear areas define microfluidic fea-tures such as channels. Next, a planar substrate (typically a silicon wafer) is coatedin a light-sensitive material called photoresist. Ultraviolet (UV) light is then shonethrough the transparent areas of the mask and onto the photoresist, catalyzing itspolymerization. Unexposed areas are then washed away, leaving a patterned coat-ing on the substrate. The minimum feature size is determined by the resolutionof the lithographic process; for standard UV lithography, it is approximately onemicron. This process can be repeated several times in order to create features withdifferent heights.Microfluidic chips are next cast against these master moulds, most commonlyusing an elastomeric polymer called polydimethylsiloxane (PDMS) [148] (Fig-ure 1.7). PDMS’s properties of optical transparency, biocompatibility, gas per-meability, and mechanical flexibility have made it the most popular material formicrofluidic devices. PDMS components are mixed, poured on the mould, allowedto vulcanize, and peeled off. To enclose the channels, the PDMS replica is thenbound to glass or another layer of PDMS. Raised areas on the master mould trans-fer as depressions into the PDMS replica, creating the desired network of channelsand chambers. Several iterations of replica moulding can be performed from thesame master, thereby enabling the low-cost production of microfluidic devices.Shortly after the introduction of soft lithography, a method to fabricate inte-grated microfluidic valves was demonstrated [149]. This technique, called multi-layer soft lithography (MSL), involves stacking multiple PDMS layers that havebeen independently cast from separate master moulds (Figure 1.8A). Valves arecreated using two overlapping channels that are separated by a thin PDMS mem-brane. When pressure is applied to, for example, the bottom channel (Figure 1.8B),27FlowControlFlowControlA BCflatsubstratemoldFigure 1.8: Multi-layer soft lithography. (A) PDMS is cast on two separatemoulds: a “thin” layer (top left) is created by spin-coating PDMS prepolymer,and a “thick” layer (top right) is poured on the mould. After partial polymer-ization, the thick layer is removed from the mould and aligned to the thinlayer (middle). This combined slab is further polymerized, creating overlap-ping, orthogonal channels separated by a thin, flexible membrane. From MarcA. Unger, Hou-Pu Chou, Todd Thorsen, Axel Scherer, and Stephen R. Quake.Monolithic microfabricated valves and pumps by multilayer soft lithography.Science, 288(5463):113-116, 2000. Reprinted with permission from AAAS.[149] (B) Schematic of a cross-section of an un-actuated valve (left). The“thick” layer is in grey, the “thin” layer is green, and a glass slide used to sealthe chip is shown in blue. A top-down optical micrograph of an un-actuatedvalve is shown on the right. Scale bar 100 µm. (C) Applying pressure to thecontrol channel causes the thin membrane separating the thick and thin layersto deflect up into the channel above, pinching it off and creating a seal. Themiddle valve in the optical micrograph on the right is closed; the other twoare open. Scale bar 100 µm.28the membrane deflects up into the top channel, pinching it off (Figure 1.8C). Inorder to create a perfect seal, this top channel is rounded during master mouldfabrication such that its profile better matches that of the membrane deflection.The resulting valves can be repeatedly activated, have a low dead volume, and aresmall, with a standard size of 100 µm×100 µm. As they are fabricated in parallel,each device can contain thousands of valves [129]. Furthermore, multiple inde-pendently controlled valves can be used to build up more complicated secondaryfunctions such as pumps or (de)multiplexors. Since its introduction, the use ofMSL has shifted the focus in the microfluidic community from how to be able tomake devices to applying them to various problems ranging from biology to point-of-care diagnostics [135].1.4 HematopoiesisEvery day, more than 100 billion new blood cells are produced in the averageadult human. This enormous regenerative capacity is established and maintainedthroughout a lifetime by the hematopoietic system, which, through a series ofamplifying cell divisions, gives rise to the functionally diverse mature blood celltypes [150, 151]. The current understanding of cellular derivation and the molecu-lar mechanisms behind self-renewal, differentiation, and lineage choice have beendriven by decades of research based on methods to functionally analyze and purifysingle cells (reviewed in [150, 151]).Stem and progenitor cells in the hematopoietic hierarchy are functionally de-fined based on their capacities for self-renewal and potency. Seminal work in thefield demonstrated that a transplant of healthy mouse bone marrow could restorethe destroyed blood-forming system in recipients that had received a lethal doseof radiation [152]. A watershed moment towards identifying and quantifying thecells responsible for this regeneration was the observation that the visible coloniesthat appeared on the surface of the spleen of reconstituted irradiated hosts were de-rived from individual cells and that their abundance was proportional to the numberof transplanted cells [153, 154]. Subsequent characterization of the cells initiat-ing these colonies demonstrated that they were multipotent and able to self-renew[155, 156]. The heterogeneity of these behaviours, however, highlighted the im-29Haematopoietic stem cellLong term Short term MultipotentprogenitorB cellsT cellsNK cellsDendritic cellsGranulocytesMacrophagesPlateletsRed cellsPro-BPro-TPro-NKCLPCMPGMPMEPMkPErPFigure 1.9: In the classic model of hematopoietic development, hematopoi-etic stem cells lose self-renewal ability, then give rise to common lym-phoid progenitors (CLPs), which are precursors to all lymphoid cells, andcommon myeloid progenitors (CMPs), which are precursors to all myeloidcells. These oligopotent progenitors further give rise to lineage restrictedmegakaryocyte/erythroid (MEP), and granulocyte/macrophage (GMP) pro-genitors. MkP: megakaryocyte precursor, ErP: erythrocyte precursor, NK:natural killer. Adapted by permission from RightsLink/Springer Nature: Na-ture. Stem cells, cancer, and cancer stem cells. Tannishtha Reya, Sean J.Morrison, Michael F. Clarke & Irving L. Weissman, c©2001. [157]portance of clonogenic assays to detect and quantify cell types. Together, theseearly experiments paved the way towards establishing an operational definition ofhematopoietic stem cells (HSCs) based on their capacity to serially reconstitute theentire blood system of an irradiated host.30Characterization of the stem and intermediate progenitor states has been fur-ther enabled by methods to prospectively purify subsets of hematopoietic cells[158–162]. Cell separation strategies are based on the measurement of a panelof attributes including biochemical and physical characteristics and expression ofsurface antigens [150]. As these strategies do not separate cells to functional pu-rity (e.g., the current state of the art enriches mouse long-term repopulating cellsto a purity of ∼20 to 30 % [160], and human to ∼10 % [162]), the ability to quan-tify the functional outputs of single cells has been critical. This effort was largelysupported by the development of culture conditions that supported in vitro differen-tiation into a variety of mature blood cell types [163], enabling analogous quantita-tive assays for the capacity of restricted progenitor subsets. Initial careful analysisof the course of differentiation of these populations led to their organization in ahierarchical model.The classical model of hematopoiesis (Figure 1.9) places the HSC at the apex ofa series of stepwise, branching differentiation stages. First, as HSCs progressivelylose their capacity for self-renewal they turn from long-term reconstituting HSCs(LT-HSCs), to short-term reconstituting (ST-HSCs), and finally into multipotentprogenitors (MPPs). MPPs next bifurcate along either the myeloid or lymphoid lin-eages to form common myeloid (CMP) or common lymphoid (CLP) progenitors,respectively. CMPs further differentiate into the unipotent megakaryocyte/ery-throid (MEP), or granulocyte/macrophage (GMP) progenitors before yielding therespective mature effector cells. CLPs give rise to the T, B, or natural killer (NK)cells.Recent molecular and functional studies have, however, brought the classicparadigm of a stepwise differentiation hierarchy into question. These studies, basedon improved sorting strategies and in vitro assays for functional potential [159,164], and the application of recent innovations in single-cell RNA-seq [165–167],instead characterize differentiation as a continuous and graded process with nodistinct transitions between cell types. Uncommitted, multipotent cells only existin the stem cell compartment, and gradually acquire unipotent lineage specificationrather than through successive stages of multi- and bipotency. Future discoveryand updates to methods for the detailed molecular characterization of single cellsisolated from their unperturbed environment will continue to refine the model of31hematopoiesis and the molecular mechanisms behind cell states, choices, and fates.1.4.1 MicroRNAs in hematopoiesisExpression profiling, ablation, and gain- and loss-of-function studies have estab-lished miRNAs as an integral component of hematopoiesis. This combined workhas identified numerous miRNAs with functional roles impacting cell maintenance,differentiation and function (recently reviewed in [168–172]). Many miRNAs areseen to exert their influence over only specific lineages and cell types: the miR-125family has a pivotal role in the control of HSC self-renewal [33, 173–177]; miR-126-3p is an established functional marker of human HSCs, and governs the size ofthe HSC pool size [101, 178]; miRNAs -144, -451a and -486 are expressed specifi-cally during, and required for erythropoiesis [179–181]; miR-223-3p is specificallyexpressed within myeloid cells [182] and up-regulated during granulopoiesis [183];and the expression of miR-181 can drive the differentiation of B cells [182]. Incontrast, the expression of miR-150 inhibits the progression of B-cell development[184], but drives MEP differentiation into megakaryocytes [185]. Other miRNAsare seen to regulate specific functions irrespective of cell type. For example, miR-146a-5p, attenuates the inflammatory response in HSCs [186], myeloid cells [187],and classes of T cells [188].It is important to note, however, that there are numerous caveats to the afore-mentioned observations of miRNA involvement in hematopoiesis. First, due toinput requirements, rare HSPC subpopulations (even when measured in bulk) havethus far been inaccessible to genome-wide miRNA expression profiling techniques.As it is becoming clear that miRNAs can have different outcomes in different cel-lular contexts, integrated genome-wide data will be a valuable asset towards un-derstanding mechanistic relationships. Second, even phenotypically pure HSPCpopulations, while enriched for specific cell types, still exhibit functional hetero-geneity [189]. As a result, bulk analysis must be interpreted as a mixed molecu-lar signature [72]. Finally, the vast majority of this work has been performed inmice, and there has been poor success in transferring murine purification markersto humans [189]. While substantial progress has been made towards identifyingthe miRNAs involved in regulating hematopoietic development and function, new32single-cell approaches are needed to develop a better picture of the underlying net-work architecture.1.5 Research statementThis dissertation describes the development of three microfluidic tools that en-able the measurement of miRNA expression in hundreds of single cells per ex-periment. In addition to establishing solutions to long-sought technical milestonesin microfluidic and single-cell research, the application of these devices providedinsight into a variety of biological processes including intrinsic noise in gene ex-pression, transcript and miRNA co-regulation, and miRNA expression dynamicsduring hematopoietic stem cell differentiation.High-throughput microfluidic single-cell RT-qPCRAt the onset of this work, the scalable implementation of integrated single-cell geneexpression analysis had not been robustly demonstrated. Chapter 2 establishes thefoundational components required for incorporating single-cell manipulation withnucleic acid molecular biology in a microfluidic device. By integrating all of thesteps required to process and measure single cells using RT-qPCR, we demon-strated that nanolitre-volume processing increased measurement sensitivity and re-duced noise while increasing cell throughput and reducing reagent consumption.This technology was applied to measure the heterogeneity of single-cell miRNAexpression, the co-regulation of a miRNA and one of its target transcripts, anddetect single nucleotide variants in primary lobular breast cancer cells.Highly multiplexed single-cell quantitative PCRThere are inevitably many questions in single-cell analytics that can only be an-swered by measuring numerous features from each cell. Chapter 3 builds on thefoundational advances presented in Chapter 2 in order to increase the number ofassays from one or two per cell to up to forty. We demonstrated that the resultingdevice maintained the same benefits afforded through microfluidic integration thatwere seen in our earlier work, and then applied it towards measuring a panel oftwenty microRNAs in two cell types.33Single-cell microRNA sequencing of the human hematopoietic cell hierarchyRecent technological advancements enabling transcriptome-wide expression pro-filing on single cells have led to a paradigm shift in the approach taken to describingcomplex tissues. These efforts, however, have been almost exclusively restrictedto mRNA measurements, leaving a gap in regard to small non-coding RNAs. InChapter 4, we present and apply a method to create high-quality single-cell miRNAsequencing libraries. We first demonstrated that the libraries created using thismethod are of equivalent quality to those generated for consortium-level projectsusing traditional high-input, bulk methods. Next, we used this method to generatea high-resolution analysis of the microRNA expression profiles of primitive normalhuman cord blood cells at multiple different stages of hematopoiesis.34Chapter 2High-throughput microfluidicsingle-cell RT-qPCR12.1 IntroductionSingle cells represent a fundamental unit of biology; however, the vast majority ofbiological 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, intrin-sic noise in gene expression, and the origins of disease, that can be addressed onlyat the single-cell level. For example, single-cell analysis allows for the direct mea-surement of gene expression kinetics, or for the unambiguous identification of co-regulated genes, even in the presence of desynchronization and heterogeneity thatcould obscure population-averaged measurements. Similarly, single-cell methodsare vital in stem cell research and cancer biology, where isolated populations ofprimary cells are heterogeneous due to limitations in purification protocols, andit is often a minority cell population that is the most relevant. High-throughputsingle-cell measurement technologies are therefore of intense interest and have1A version of this chapter has been published: Adam K. White, Michael VanInsberghe, Oleh I.Petriv, Mani. Hamidi, Darek. Sikorski, Marco A. Marra, James Piret, Samuel. Aparicio, and Carl L.Hansen. High-throughput microfluidic single-cell RT-qPCR. Proceedings of the National Academyof Sciences, 108(34):13999-14004, 2011.35broad application in clinical and research settings.Existing methods for measuring transcript levels in single cells include RT-qPCR [190], single molecule counting using digital PCR [191] or hybridizationprobes [77, 192], and next generation sequencing [60]. Of these, single-cell RT-qPCR provides combined advantages of sensitivity, specificity, and dynamic range,but is limited by low throughput, high reagent cost, and difficulties in accuratelymeasuring low abundance transcripts [193].Microfluidic systems provide numerous advantages for single-cell analysis:economies of scale, parallelization and automation, and increased sensitivity andprecision that come from small volume reactions. Considerable effort over thelast decade has been directed toward developing integrated and scalable single-cellgenetic analysis on chip [194, 195]. Thus, many of the basic functionalities formicrofluidic single-cell gene expression analysis have been demonstrated in iso-lation, including cell manipulation and trapping [141, 144], RNA purification andcDNA synthesis [134, 138, 140], and microfluidic qPCR [196] following off-chipcell isolation, cDNA synthesis, and preamplification. In particular, microfluidicqPCR devices (Biomark Dynamic Array, Fluidigm) have recently been applied tosingle-cell studies [111, 197]. Although these systems provide a high-throughputqPCR readout, they do not address the front end sample preparation and requiresingle-cell isolation by FACS or micropipette followed by off-chip processing andpreamplification of starting template prior to analysis. The critical step of integrat-ing all steps of single-cell analysis into a robust system capable of performing mea-surements on large numbers of cells has yet to be reported. A single demonstrationof an integrated device for directly measuring gene expression in single cells wasdescribed by Toriello et al., combining all steps of RNA capture, PCR amplifica-tion, and end-point detection of amplicons using integrated capillary electrophore-sis [61]. Despite the engineering complexity of this system, throughput was limitedto four cells per run, cell capture required metabolic labeling of the cells, and theanalysis was not quantitative. Thus, there remains an unmet need for microfluidictechnologies capable of scalable and quantitative single-cell genetic analysis.Here we describe an integrated microfluidic device for high-throughput RT-qPCR analysis of mRNA and miRNA expression at a throughput of hundredsof single cells per experiment. We show that this technology provides a pow-36erful tool for scalable single-cell gene expression measurements with improvedperformance, reduced cost, and higher sensitivity as compared to analysis in mi-crolitre volumes. This technology represents the implementation of robust andhigh-throughput single-cell processing and amplification of nucleic acids on a chip,thereby achieving a major milestone in microfluidic biological analysis.2.2 Materials and methodsDevice fabrication and operationMicrofluidic devices were fabricated by multilayer soft lithography [129, 149].Planar silicon moulds were defined by photolithography using photomasks de-signed with CAD software (AutoCAD, Autodesk Inc.) and printed on transparencyfilms at a resolution of 20,000 dots per inch (CAD/Art services). The “control”mould was fabricated using SU8-2025 photoresist (Microchem) to deposit valvefeatures 24 µm in height. The “flow” mould was fabricated with three lithographicsteps. First, the channels for reagent injection and connections between chamberswere fabricated using 13-µm high SPR220-7 photoresist (Shipley). The SPR220-7 channels were rounded to facilitate valve closure by incubation at 115 ◦C for15 min. A hard bake at 190 ◦C for 2 h was used to prevent SPR photoresist erosionduring addition of subsequent layers. Second, the cell trap features were defined in14-µm SU8-2010 photoresist (Microchem). Finally, the large chambers and fluidicbus lines were constructed using 150-µm high SU8-100 photoresist. All photoresistprocessing was performed according to manufacturer specifications. All mouldswere fabricated on 4-inch silicon wafers (Silicon Quest International).Microfluidic devices were cast from these master moulds in polydimethylsilox-ane (PDMS, RTV615, General Electric). Each device consists of a three-layer elas-tomeric structure with a blank bottom layer, a middle control layer with channelsthat act as valves by pushing up and pinching channels closed in the above flowlayer. The moulds were first treated with chlorotrimethylsilane (TMCS, Aldrich)vapour for 2 min to prevent PDMS from bonding to the photoresist structures. Theflow layer was made by pouring a mixture of PDMS (5 parts RTV615A to 1 partRTV615B) onto the flow mould, degassing, and then baking for 60 min at 80 ◦C.37A thin control layer was made by spin coating the control mould with PDMS (20parts RTV615A:1 part RTV615B) at 1,800 rpm and baking for 45 min at 80 ◦C.After baking, the PDMS of the flow layer was peeled from the flow mould andaligned by hand to the control layer using integrated alignment features. Follow-ing a 60-min bake at 80 ◦C, the bonded two-layer structure was separated from thecontrol mould, 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 (2,000 rpm) and baking for 45 min at 80 ◦C. The bonded flowand control structure was mounted onto the blank layer and baked for 3 h at 80 ◦C.Finally, the three-layer bonded structure was removed from the blank mould, dicedinto individual devices, and these were each bonded to clean glass slides by bakingovernight at 80 ◦C.The device operation requires control of nine pneumatic valves and may beoperated using a simple manifold of manual valves. For the current study a semi-automated implementation was used in which microfluidic valves were controlledby solenoid actuators (Fluidigm Corp.) controlled through a digital input-outputcard (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.Microfluidic single-cell RT-qPCRThe device was designed to be compatible with commercially available RT-qPCRproducts. A protocol for heat lysis, followed by a two-step RT-qPCR was usedwith miRNA and OCT4 mRNA assays. Alternatively, a chemical lysis, followedby one-step RT-qPCR, was used for mRNA measurements of single nucleotidevariants (SNVs) and GAPDH expression.Single-cell transcript measurements by heat lysis and two-step RT-qPCRExperimental timing for two-step RT-qPCR is provided in Table 2.1.The device was primed by flowing PBS containing 0.5 mg/mL BSA and 0.5 U/µL38Table 2.1: Protocol timing for performing heat lysis and two-step RT-qPCRin the microfluidic systemStep Description Time1 Prime device with PBS 0.5 mg/mL BSA and 0.5 U/µL RNaseinhibitor1 min2 Inject cell suspension (passive cell trapping) 1 min3 On-chip cell washing with PBS containing 0.5 mg/mL BSAand 0.5 U/µL RNase inhibitor1 min4 Close valves partitioning cell loading channel and isolatingsingle cells30 s5 Count cells by visual inspection with microscope 7 min6 Heat lysis by placing device on flatbed thermocycler and heat-ing to 85 ◦C7 min7 Flush fluidic bus and reagent injection lines with reagent forRT2 min8 Inject RT reagent through the cell-capture chamber, dead-endfilling the 10-nL RT chamber1 min9 Close reagent injection valve, creating isolated reactors com-bining the cell-capture chamber and RT chamber30 s10 Perform reverse transcription (pulsed temperature protocol) byplacing device on flatbed thermocycler2.5 h11 Flush fluidic bus and reagent injection lines with reagent forPCR2 min12 Inject PCR reagent through combined cell-capture/RT cham-ber into 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 focuscamera5 min15 Run qPCR protocol (varies)39RNase inhibitor through all channels, while keeping the reverse transcription (RT)and PCR chambers empty and isolated by valves. The BSA helped prevent cellsfrom adhering to channel walls. After priming, but prior to cell loading, all valveswere closed. A single-cell suspension was injected into the device by applyingpressure (approximately 2-3 PSI) to microcapillary pipette tips plugged into thesample inlets. The sample inlets were first dead-end filled against an inlet valve toprevent air bubbles from entering the device. The sample inlet valves, cell chambervalves, and outlet valve were opened to allow the cell suspension to flow throughthe sample channels. Cells were loaded into the device suspended in culture media(directly from culture). Cell loading concentrations were kept between 5×105 and1×106 cells/mL, resulting in over 80 % single-cell occupancy of cell traps in 1 to2 min at a flow rate of approximately 20 nL/s. Lower concentrations were foundto require proportionately longer times to achieve high occupancy of trapped sin-gle cells. Concentrations greater than 2×106 cells/mL were found to occasionallyclog the inlet port or the channel at trap locations. A peristaltic pump was inte-grated into the device for controlling the flow rate; however, pressure-driven flowwas used for the current study.After injecting the cell suspension and trapping single cells, the cell-sampleinlet valve was closed and the cells were washed by flushing the line with thePBS-BSA solution. This removed un-trapped single cells, extracellular RNA, anddebris. Following on-chip washing, the cell chamber valves were closed to parti-tion the cell loading channel and isolate individual cell reactors. Visual inspectionof the cell-capture chambers under a microscope was used to confirm and countthe number of cells in each chamber. The cells were lysed by placing the microflu-idic device onto a flatbed thermocycler and heating it to 85 ◦C for 7 min and thencooling to 4 ◦C.RT was performed in the device by using the ABI High Capacity Reverse Tran-scription kit [110], with the addition of a surfactant to prevent adsorption of nu-cleic acids and proteins to PDMS surfaces (2 µL 10× Reverse Transcription Buffer,4 µL 5× RT stem-loop miRNA primer from ABI, 1 µL 100 mM dNTPs, 1.34 µL of50 U/µL Multiscribe Reverse Transcriptase, 0.26 µL of 20 U/µL RNase Inhibitor,2 µL 1 % Tween-20, 9.4 µL PCR-grade water). The RT mix was loaded into thedevice and flushed through the reagent injection channels. RT reagent was injected40into the reaction by opening the valve connecting the cell chamber to the RT cham-ber, and the valve connecting the cell chamber to the reagent injection line. The RTchamber was dead-end filled before closing the connection to the reagent injectionline. A pulsed temperature RT protocol was carried out by placing the microfluidicdevice on a flatbed thermocycler (2 min at 16 ◦C, followed by 60 cycles of 30 s at20 ◦C, 30 s at 42 ◦C, and 1 s at 50 ◦C). RT enzyme was inactivated at 85 ◦C for5 min, and then the device was cooled to 4 ◦C.The PCR reagent was prepared with 25 µL of 2× TaqMan Universal MasterMix (ABI), 2.5 µL 20× Real-Time miRNA assays (primers and probe, ABI), 5 µLof 1 % Tween-20, and 7.5 µL of PCR-grade water. The PCR reagent was flowedthrough the reagent injection channels to flush away the RT reagent. Valves wereopened and the PCR reagent was injected to dilute the RT product into the PCRreaction chamber. After completely filling the PCR reaction chamber, the valvesclosing the PCR chambers were actuated, and the device was transferred to anenclosure for real-time PCR (Prototype version of the Biomark Instrument, Flu-idigm). The real-time PCR enclosure consists of a custom flatbed thermocycler, axenon arc lamp and filter set, and a CCD imager with optics for fluorescent imagingof the entire device during PCR thermocycling (see description of real-time PCRInstrumentation 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.Single-cell transcript measurements by chemical lysis and one-step RT-qPCRExperimental timing the one-step procedure is provided in Table 2.2.Measurements of mRNA transcripts (SP1, GAPDH) were performed using theCells Direct kit (Invitrogen). Operation of the microfluidic device for chemicallysis and one-step RT-qPCR was similar to the methods described for heat lysisand two-step RT-qPCR with several distinctions. The device was primed and cellswere washed with PBS containing 0.5 mg/mL BSA. Additional RNase inhibitorwas omitted as the chemical lysis buffer (10 µL lysis resuspension buffer, 1 µL lysisenhancer solution, Invitrogen, USA) contained RNA stabilizing agents. Cell load-ing was the same as in the heat lysis and two-step RT-qPCR scenario. Single cells41Table 2.2: Protocol timing for performing chemical lysis and one-step RT-qPCR in the microfluidic systemStep Description Time1 Prime device with PBS 0.5 mg/mL BSA and 0.5 U/µL RNaseinhibitor1 min2 Inject cell suspension (passive cell trapping) 1 min3 On-chip cell washing with PBS containing 0.5 mg/mL BSAand 0.5 U/µL RNase inhibitor1 min4 Close valves partitioning cell loading channel and isolatingsingle cells30 s5 Count cells by visual inspection with microscope 7 min6 Flush fluidic bus and reagent injection lines with reagent forlysis2 min7 Inject lysis reagent through the cell-capture chamber, dead-endfilling the 10-nL chamber1 min8 Close reagent injection valve, creating isolated reactors com-bining the cell-capture chamber and lysis reservoir chamber30 s9 Perform lysis at room temperature and heat inactivation of thelysis reagent at 75 ◦C by placing device on flatbed thermocy-cler25 min10 Flush fluidic bus and reagent injection lines with reagent forRT-qPCR2 min11 Inject RT-qPCR reagent through combined cell-capture/lysischamber into 50-nL RT-qPCR chamber5 min12 Close valve to RT-qPCR chamber. Allow for mixing by diffu-sion40 min13 Load device into BioMark real-time PCR system and focuscamera5 min14 Run RT-qPCR protocol (varies)42were lysed by injecting a chemical lysis buffer through the cell-capture chamberand filling the 10-nL chamber (used for RT reagent injection in the two-step pro-tocol). The lysis reaction was incubated at room temperature for 10 min, followedby heat inactivation of the lysis reagent by placing the device on a flatbed thermo-cycler and incubating at 70 ◦C for 10 min. The one-step RT-qPCR mix [1 µL ofSuperScript III RT/Platinum Taq Mix, 25 µL of 2× Reaction Mix (with ROX refer-ence dye), 2.5 µL of 20× TaqMan Assay (primers and probes, ABI), 1 µL of 50 mMMgSO4, 5.5 µL of H2O, and 5 µL of 1 % Tween-20] was then combined with thecell lysate into the final 50-nL reaction chamber. The device was transferred tothe real-time PCR enclosure for temperature control and imaging of the one-stepRT-qPCR (20 min at 50 ◦C for RT, followed by a hot-start at 95 ◦C for 2 min, and50 cycles of 15 s at 95 ◦C and 30 s at 60 ◦C).Digital PCR experimentsFor the mRNA digital PCR analysis, cells were washed in tubes with PBS contain-ing 0.5 mg/mL BSA then lysed in Cells Direct lysis buffer. Reverse transcriptionwas then performed in tubes according to the protocol described above, and theresulting cDNA product was loaded into digital PCR arrays. For miRNA stud-ies, cells were lysed in PBS containing 0.5 mg/mL BSA and 0.5 U/µL RNase in-hibitor. Reverse transcription was then performed using miRNA stem-loop primers(Applied Biosystems) and the High Capacity cDNA Reverse Transcription kit (Ap-plied Biosystems) in 10-µL volumes. Prior to injection into microfluidic digitalPCR arrays, RT product was added to the PCR reagent as in the on-chip two-stepRT-qPCR protocol described above. Thermal cycling of digital PCR arrays wasalso performed using the same protocols as described above. PDMS digital PCRarrays consisting of 765 2-nL individual PCR chambers, of similar design to thosedescribed in Warren et al. [191] and Petriv et al. [198], were fabricated by multi-layer soft lithography. After thermal cycling, positive chambers were counted andactual molecule numbers were derived based on the binomial distribution.43RT-qPCR assaysMeasuring mRNA abundance in the presence of genomic DNA requires primersdesigned to specifically target mature mRNA sequences. In many cases, this canbe accomplished by designing intron-spanning primers. A specially designed stem-loop RT primer system (Applied Biosystems) is used for the specific targeting ofmature miRNAs.TaqMan assays for GAPDH (Applied Biosystems, Assay ID Hs99999905 m1)and miRNAs were obtained from Applied Biosystems. For GAPDH, a controlexperiment omitting the reverse transcriptase was performed off chip in microlitrevolumes with bulk 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 RTPrimerDB2 andsynthesized 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: TTGTGC ATA GTC GCT GCT TGA T. Measurement of OCT4 in single hESCs bymicrofluidic RT-qPCR without reverse transcriptase showed no amplification after40 cycles of PCR.BHQ-Plus probes with enhanced duplex stabilization (Biosearch TechnologiesInc.) were used for SNV detection to allow for shorter sequence lengths and in-creased specificity. The SNV location for the SP1 locus was selected from Table 2in Shah et al. [199]. Two hundred base pairs flanking this location on the hg18 se-quence were used for assay design using Primer3. The resulting primer and probesequences were as follows (the SNV is underlined):SP1 Mutant Probe: FAM-AGGCCAGCAAAAACAAGG-BHQ-1. 5′ Modifica-tion: FAM, 3′ Modification: BHQ-1 Plus. Tm = 62.7 ◦CSP1 WT probe: Cal Fluor-CAGGCCAGCAAAAAGAA-BHQ-1. 5′ Modifi-cation: CAL Fluor Orange 560, 3′ Modification: BHQ-1 plus. Tm = 62.1 ◦CSP1 Forward Primer: CCAGACATCTGGAGGCTCATTG, Tm = 65.8 ◦CSP1 Reverse Primer: TGAACTAGCTGAGGCTGGATA, Tm = 66.0 ◦CControl experiments without reverse transcriptase showed positive amplifica-2http://www.rtprimerdb.org44Table 2.3: Specifications for collecting qPCR imagesImage resolution and bit depth: 4 Megapixel, 16 bitFilters: FAM: Ex 485±20, Em 525±25CAL: Ex 530±20, Em 570±30ROX: Ex 580±25, Em 610±15QUASAR: Ex 580±25, Em 680±25Light source: 175-W xenon arc bulbtion. Therefore the measurement of SP1 mutant and wild-type abundance in singlecells by RT-qPCR does not discriminate between mature mRNA transcripts andgenomic DNA.System for real-time PCRThe 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 thermocyclerinside the enclosure, permitting the use of custom microfluidic devices in additionto the intended commercial IFCs. Fundamental specifications for data collectionare included in Table 2.3.Image analysisFluorescence images of the entire device were taken in at least two different colours(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 restof the image using the first image of the passive reference dye. The image wasmanually rotated so that all of the reaction chambers were square with the edges ofthe image. Next, the average image intensities across each row and column werecalculated and a threshold was manually set to differentiate bright areas from back-ground. Regions containing both bright rows and bright columns were assigned tothe reaction chambers.All subsequent images were automatically aligned to this initial image by min-45imizing the absolute distance between the average row and column intensities ofthe initial image, and the one being analyzed. For each image, the intensities ofthe reporter and passive dyes were recorded for each reaction chamber. Real-timeamplification curves were generated by normalizing the intensity of each reporterdye to that of the passive dye. Linear components were removed from these curvesby fitting the equation of a line to the pre-exponential region and extrapolating andsubtracting the result from the entire curve. The threshold for determining CT val-ues was automatically determined as the median normalized fluorescence value atthe maximum second derivative of all amplification curves.mRNA FISHCells were grown on LABTEK chambered cover glass and were washed with PBS,fixed in 4 % formaldehyde for 10 min at room temperature and permeabilized in70 % ethanol at 40 ◦C overnight. The next day, cells were rinsed with wash buffer(15 % formamide in 2× SSC) and then hybridized with the appropriate dilution ofmRNA-FISH probes specific to OCT4 (Table 2.4) in hybridization solution [dex-tran sulfate, yeast tRNA, vanadyl ribonucleoside complex (New England Biolabs),BSA, 15 % formamide in 2× SSC] overnight at 30 ◦C. The next morning, theOCT4 hybridization solution was aspirated and cells were sequentially rinsed andincubated with wash buffer at 30 ◦C for 30 min, then washed with 2× SSC. Onedrop (25 µL) of Slowfade GOLD antifade reagent with DAPI was then added to thecells, covered immediately with a coverslip, 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 andTexas-red filter spectra.Fluorescent spots corresponding to individual mRNA molecules in each imagestack were evaluated manually because automatic thresholding using previouslyreported algorithms was found to be unreliable. Difficulty in automating this pro-cess was attributed to inconsistent signal to noise using reported protocols andmay be related to the thickness of hESC cells (approximately 15 µm). In addition,manual intervention was needed to ascertain the boundaries of adjacent cells. Tooptimize the signal to noise we systematically varied the probe concentration, in-46cubation time, incubation temperature, as well as the formamide concentration inthe hybridization buffer solution.Table 2.4: OCT4 probe sequences for mRNA FISHProbe sequence Name Probe # Position % GCtgaaatgagggcttgcgaag OCT4 1 1 2 50aaatccgaagccaggtgtcc OCT4 2 2 61 55atcacctccaccacctggag OCT4 3 3 95 60aggtccgaggatcaacccag OCT4 4 4 138 60aggagggccttggaagctta OCT4 5 5 161 55aatcccccacacctcagagc OCT4 6 6 215 60atccccccacagaactcata OCT4 7 7 253 50actagccccactccaacctg OCT4 8 8 289 60tcaggctgagaggtctccaa OCT4 9 9 322 55agttgctctccaccccgact OCT4 10 10 354 60ttctccttctccagcttcac OCT4 11 11 418 50ctcctccgggttttgctcca OCT4 12 12 440 60ttctgcagagctttgatgtc OCT4 13 13 466 45cttggcaaattgctcgagtt OCT4 14 14 488 45tgatcctcttctgcttcagg OCT4 15 15 510 50atcggcctgtgtatatccca OCT4 16 16 533 50aaatagaacccccagggtga OCT4 17 17 560 50tcgtttggctgaataccttc OCT4 18 18 582 45taagctgcagagcctcaaag OCT4 19 19 612 50gcagcttacacatgttcttg OCT4 20 20 636 45tccacccacttctgcagcaa OCT4 21 21 661 55gattttcattgttgtcagct OCT4 22 22 684 35tctgctttgcatatctcctg OCT4 23 23 706 45actggttcgctttctctttc OCT4 24 24 743 45ttgcctctcactcggttctc OCT4 25 25 766 55ctgcaggaacaaattctcca OCT4 26 26 788 45atctgctgcagtgtgggttt OCT4 27 27 814 50atccttctcgagcccaagct OCT4 28 28 851 55ttacagaaccacactcggac OCT4 29 29 874 50tagtcgctgcttgatcgctt OCT4 30 30 910 50ctcaaaatcctctcgttgtg OCT4 31 31 932 45ctgagaaaggagacccagca OCT4 32 32 954 55agaggaaaggacactggtcc OCT4 33 33 976 55atagcctggggtaccaaaat OCT4 34 34 1010 45agtacagtgcagtgaagtga OCT4 35 35 1038 45ttccccctcagggaaaggga OCT4 36 36 1064 60tgacggagacagggggaaag OCT4 37 37 1086 60agtttgaatgcatgggagag OCT4 38 38 1116 4547Probe sequence Name Probe # Position % GCattcctagaagggcaggcac OCT4 39 39 1139 55ttttctttccctagctcctc OCT4 40 40 1176 45aaaaaccctggcacaaactc OCT4 41 41 1200 45ccttagtgaatgaagaactt OCT4 42 42 1226 35accctttgtgttcccaattc OCT4 43 43 1249 45aaccagttgccccaaactcc OCT4 44 44 1278 55cattgaacttcaccttccct OCT4 45 45 1300 45gtgggattaaaatcaagagc OCT4 46 46 1322 40ccaggcttctttatttaaga OCT4 47 47 1359 35aagtgtgtctatctactgtg OCT4 48 48 1381 40Cell culture and RNA purificationK562 cells were cultured in DMEM (Gibco) supplemented with 10 % FBS (Gibco).Purified RNA was extracted from K562 cells using RNA MiniPrep (Qiagen).CA1S hESCs [200, 201] were propagated in mTeSR [202] basal medium (STEM-CELL Technologies, Inc.) supplemented with antibiotic-antimycotic mix (100 U/mLpenicillin, 100 mg/mL streptomycin, and 0.25 mg/mL amphotericin B) (Invitro-gen). Upon passaging, hESCs were washed with PBS prior to incubating withTrypLE Express (Invitrogen) at 37 ◦C for 10 min to detach single hESCs from 4-to 8-day-old cultures, depending on confluency. TrypLE Express was neutralizedwith mTeSR supplemented with antibiotic-antimycotic mix and suspensions werethen transferred into new tissue culture dishes containing a pre-coated layer of 1:30diluted Matrigel (Becton Dickinson) and mTeSR supplemented with antibiotic-antimycotic mix. For differentiation, mTeSR was replaced with DMEM with 10 %FBS 1 day after plating cells.When harvesting hESCs for RT-qPCR, cells were incubated with TrypLE Ex-press (Invitrogen) at 37 ◦C for 20 min in order to produce a more uniform single-cell suspension from 4- to 8-day-old cultures.Cryovials 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 tofresh culture medium and incubated for 2 days before analyzing in the microfluidicdevice.48Transfer efficiency measurementsA solution containing 10 µM FAM-labeled 40-base long poly-A oligonucleotides(Integrated DNA Technologies), 0.1 % Tween-20, and ROX passive reference dye(from CellsDirect kit, Invitrogen, P/N 54880) diluted 100× was loaded into thecell-capture chambers and sequentially pushed into the 10-nL and 50-nL chamberswith water containing 0.1 % Tween-20, and ROX reference dye diluted 100×. Flu-orescence images acquired of FAM and ROX were used to measure the transfer ofoligonucleotides from one chamber to the next. The transfer efficiency for eachchamber was calculated as (Initial Signal - Final Signal)/(Initial Signal), whereSignal = (FAM Intensity FAM Background) / (ROX Intensity ROX Background).A conservative estimate of the lower bound of transfer efficiency was taken to beone standard deviation from the mean measurement of transfer efficiency.Cell-capture measurementsA 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(Hamamatsu ORCA-ER). The device was primed with 0.05 % BSA (Gibco) inPBS (Gibco). Prior to loading in the device, cells were washed twice in freshculture media [DMEM (Gibco) supplemented with 10 % FBS (Gibco)]. After thefinal wash cells were resuspended at a concentration of 1 million per millilitre. 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 usinga downstream microfluidic peristaltic pump at a rate of approximately 1 nL/s, andthe number of cells that bypassed each trap before a successful trapping event wasrecorded. These counts were fit using a maximum-likelihood estimator for a ge-ometric 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 successfulcapture for each cell.To measure cell viability after loading, cells were loaded into the array usingpressure driven flow as described above until the majority (> 80 %) of the cell49traps were occupied. 0.2 % trypan blue (Gibco) in PBS was then flowed over thetrapped cells for approximately 2 min; the cells were then rinsed with PBS andcounted. Viability was calculated as the number of unstained cells divided by thetotal number of cells.Cell diameter was measured from Cedex images and images of cells trappedin the microfluidic device using ImageJ (version 1.43u). A two-sample t-test wasused to test the hypothesis that the resulting size distributions were significantlydifferent. The assumption of equal variance was tested using an F-test. For opti-mized cell trap geometries the cell trapping efficiency was improved to 87 % bybringing the cup within one cell diameter of the focuser and by including a smallbypass shunt through the cup, similar to the cup geometry presented in Skelley etal. [144].Mixing by diffusionMixing of solutions by diffusion was characterized in the microfluidic device byloading fluorescently labeled 40-base poly-A oligonucleotides into the 10-nL cham-bers, and pushing the contents of the chamber into the adjacent 50-nL cham-bers. Time-lapse imaging was used to measure the evolution of the distributionof fluorescently labeled oligonucleotides in the PCR chambers over time (Fig-ure 2.1). The standard deviation of the pixel intensities in each chamber throughtime was used as a metric of mixing. The resulting curves of all analyzed chambers(N = 200) were each fit to a decaying exponential using least squares regression todetermine the characteristic mixing time constant. This resulted in a mean mixingtime of 15.2±1.0 min.Using the Stokes-Einstein relation and assuming a random coil we estimate thediffusion constant of a 40-base oligonucleotide to beD =KBT6piηRh, (2.1)where KBT is the thermal energy (4.1 pNnm), η is the fluid viscosity (approx-imately 0.001 kgs/m), and Rh is the coil hydrodynamic radius (10). The hydrody-namic radius is proportional to the radius of gyration Rg, and is given by5010 20 30 40 50 60 70 80 900.010.020.030.040.050.060.07Time [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 pixelintensity values for a chamber as a function of time following the transfer ofa solution of fluorescently labeled 40 base poly-A oligonucleotide from theRT chamber (10 nL) to the PCR chamber (50 nL) by flushing with buffer. Anexponential fit to the data to each of 200 chambers yields a mean mixing timeconstant of 15.2±1.0 min. A representative time-lapse series of images fromone chamber is shown (Right).Rh ≈ 0.5Rg ≈ 0.5(Lp/3) 12 , (2.2)where L is the contour length of single-stranded DNA (40 bases× 4.3 A˚/base)and p is the persistence length (approximately 40 A˚) [203]. This yields a diffusionvalue of approximately 9.0×10−11 m2/s, which is comparable to the diffusionconstant of polymerase, the largest molecule in the PCR mix. Since the templatesolution constitutes only 1/5 of the final PCR reaction it must diffuse the longestdistance to equilibrate across the chamber. Therefore, the measured diffusion timeof 15.2 min represents an upper bound to the time constant for complete mixing ofall components.512.3 Results and discussion2.3.1 Device 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 2.2A. To facilitate the precise comparison of different samplesand cell types, our prototype consists of six independent sample-loading lanes,each containing 50 cell-processing units. We resolved previously limiting technicalpitfalls by the inclusion of design elements to (i) allow for efficient distribution ofsingle cells without mechanical damage, (ii) minimize background signal arisingfrom free RNA or cell debris in the medium, and (iii) avoid reaction inhibition bycell lysates in nanolitre volumes.In order to reduce device complexity and obviate the need for RNA purifica-tion, we optimized our device to be compatible with commercially available as-says that use “one-pot” RT-qPCR protocols requiring only the sequential additionof reagents into a single reaction vessel. Each cell-processing unit consists of acompound chamber, formed by a cell capture chamber connected sequentially totwo larger chambers for RT and qPCR (Figure 2.2B). This simple fluidic architec-ture allows the implementation of either heat lysis followed by two-step RT-qPCR(Figure 2.2 D-I), or chemical lysis followed by one-step RT-qPCR. A detailed de-scription of device operation for each of these protocols is provided in sectionSection 2.2. All lanes are connected to a common feed channel that, followingthe completion of each reaction step, is used to inject the next reaction master mixthrough the upstream chambers, thereby diluting the intermediate product (celllysate or cDNA) and assembling the next reaction mixture. This parallelizationof reaction assembly in a microfluidic format ensures equal timing of all reactionsteps and greatly reduces technical variability associated with pipetting and mixingsteps in microlitre volumes. Fluorescence measurements were performed to ensurethe efficient and reproducible transfer of reactants at each step, showing that lossesin sample transfer are below 5 %. To minimize device expense and complexity,temperature control and fluorescence detection were performed using peripheralhardware including a CCD detector mounted above a flatbed thermocycler plate.52ACell 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 mixingWe designed our chamber volumes to ensure sufficient dilution between eachprocessing step to avoid reaction inhibition while at the same time maintaining hightemplate concentrations and assay sensitivity. Initial attempts to perform RT-qPCRin low nanolitre volumes were found to produce highly variable results, includingnonspecific amplification and inconsistent detection of abundant transcripts [62].Cell lysate dilutions showed that reaction inhibition becomes significant at con-centrations in excess of 0.2 cells/nL, or 10 cells per 50-nL reaction (Figure 2.3D).On the other hand, RT-qPCR measurement noise has been shown to become thedominant source of variability when starting at concentrations below one copy per53Figure 2.2 (previous page): Design and operation of the microfluidic devicefor single-cell gene expression analysis. (A) Schematic of microfluidic device.Scale bar: 4 mm. The device features 6 sample input channels, each dividedinto 50 compound reaction chambers for a total of 300 RT-qPCR reactionsusing approximately 20 µL of reagents. The rectangular box indicates theregion depicted in B. (B) Optical micrograph of array unit. For visualization,the fluid paths and control channels have been loaded with 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 reverse transcription(RT) chamber, and (iv) a 50-nL PCR chamber. Scale bar: 400 µm. (C) Opticalmicrograph of two cell capture chambers with trapped single cells indicatedby black arrows. Each trap includes upstream deflectors to direct cells into thecapture region. Scale bar: 400 µm. (D-I) Device operation. (D) A single-cellsuspension is injected into the device. (E) Cell traps isolate single cells fromthe fluid stream and permit washing of cells to remove extracellular RNA.(F) Actuation of pneumatic valves results in single-cell isolation prior to heatlysis. (G) Injection of reagent (green) for RT reaction (10 nL). (H) Reagentinjection line is flushed with subsequent reagent (blue) for PCR. (I) Reagentfor qPCR (blue) is combined with RT product in 50-nL qPCR chamber. Scalebar for D-I: 400 µm.100 nL [193], illustrating that minimizing reaction volumes is critical for precisemeasurements on limited template. Finally, experiments in tubes were performedto determine that a dilution ratio of at least 5:1 (PCR mix:RT product) is optimumfor PCR efficiency. We therefore designed our combined reactors to have an aggre-gate total volume of 60.6 nL, consisting of a 0.6-nL cell capture chamber, a 10-nLRT chamber, and a 50-nL qPCR chamber. These volumes allow for the reliableamplification of single molecules (Figure 2.3A) and result in a final template con-centration of 330 ng/mL when starting from a single-cell equivalent of RNA (20 pgfor K562 cells). The use of larger volume RT and PCR chambers has the addedadvantage of reducing their surface-to-volume ratio, thereby minimizing reagentevaporation through the gas permeable device material (polydimethylsiloxane).54A0 10 20 30 40 5000.20.40.60.811.2CycleΔ 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 5000.20.40.60.81Cycle 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 = 0.9983E55Figure 2.3 (previous page): Precision and sensitivity of microfluidic RT-qPCR. (A) Fluorescence image of entire device showing 300 reactions in 6lanes. Image is taken after 40 cycles of PCR from dilution series of purifiedtotal RNA from K562 cells. (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 amplification at limiting dilutionresults in a digital amplification pattern for 10- and 78-fg lanes. No ampli-fication is observed in NTC lane (N = 50). (B) Three hundred real-time am-plification curves generated from processing sequences of images similar toA. 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 an 8×serial dilution of purified total RNA shows improved sensitivity in nanolitrevolume reactions. In the microfluidic system, CT values for the 10-fg samplecorrespond to single molecule amplifications detected in 19 of 50 chambers.The mean and standard deviation from single-cell measurements is shown ingreen for both on- and off-chip analysis. CT values obtained on chip corre-spond to a mean of 20 pg of RNA per cell. Off-chip measurements of singleK562 cells washed twice in PBS and isolated by glass capillary exhibit arti-ficially increased levels due to residual signal from debris and free RNA inthe supernatant (red). Cells were transferred in approximately 2 µL of super-natant, which was measured to contain approximately 20 pg of extracellularRNA. Error bars represent standard deviation of measured CT values for allamplified reactions. (D) Real-time amplification curves of GAPDH in K562cell lysate dilutions. Inhibition of RT-PCR occurs at cell lysate concentrationsbeyond 10 cell equivalents per 50-nL reaction. (E) Measured CT values forGAPDH in dilution series of cell lysate. No inhibition occurs for single-celllysates.Another critical step toward integration was to efficiently distribute single cellsinto each location on the array without mechanical damage. To achieve repro-ducible and deterministic loading of single cells into each array element, we en-gineered a hydrodynamic single-cell trap within each capture chamber. Cell trapsconsisting of a single cup structure [143] were found to be highly inefficient, cap-turing less than 0.1 % of cells passing in close proximity to the centre of the chan-nel structure. To improve capture efficiency, we incorporated upstream deflectors,located 22.5 µm from the trap, to focus cells into the central streamlines where cap-56ture is most efficient (Figure 2.2C). Using these structures we were able to achievehigh single-cell occupancy of array locations (Figure 2.7 A and B). Over eight sep-arate experiments, a loading protocol of approximately 60 s (106 cells/mL, 20 nL/sper lane) resulted in the successful isolation of single cells in 1,518/1,700 cham-bers (89.3 %), with a cell capture efficiency of 5.0±0.5 %. Staining with trypanblue was used to assess the viability of cells after loading and was determined to beequivalent to the viability of the input sample (97.4 % viability vs. input 96.8 %).Finally, measurements of the distribution of cell diameters prior to and after loadingindicated that cell trapping did not introduce significant bias (p = 0.67, two-samplet test) in selecting cells of different sizes (Figure 2.4). This cell trap geometry andloading protocol were used in all qPCR measurements presented in this chapter.Further improvement of trap and deflector geometries were found to achieve fillfactors of > 99 % (100 single cells captured out of 100 traps analyzed) and cellcapture 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 ttest), making this method applicable to the analysis of limited quantity samplessuch as rare 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(Figure 2.5 A and B). The efficiency of chamber washing, determined by loadingpurified RNA template (36.5 ng/µL), followed by washing and RT-qPCR analysis,was greater than 99.99 % (1.1× 104 copies measured without wash, 0 copies de-tected after washing) (Figure 2.5C). In addition, RT-qPCR measurements testingdifferent cell loading and washing protocols demonstrated that on-chip washingallows for loading directly from culture medium with low background as com-pared to off-chip wash steps followed by analysis in microlitre volumes (Fig-ure 2.3C). Importantly, on-chip washing allows for lysis within seconds of wash-ing, thereby minimizing spurious transcriptional responses that may arise fromsequential medium exchange and spin steps.5710 15 20 25 30 35051015200102030Percent of PopulationDiameter [μm]AB10 15 20 25 30 35Diameter [μm]Percent of PopulationFigure 2.4: Histograms showing the size distribution of cells in original sam-ple as measured by Cedex (A) are consistent with the size distribution ofcells isolated by microfluidic traps (B). Under the assumption of sphericalcell shape the distribution of diameters of trapped cells corresponds to a meanvolume of 4.2 pL with a standard deviation of 2.0 pL.2.3.2 Validation of integrated single-cell RT-qPCRWe 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 totalRNA, ranging from 40 pg (approximately 2 cell equivalents) to 10 fg (approxi-mately 1/2,000 cell equivalents). RNA was purified from K562 cells, a BCR/ABL1positive human cell line derived from a patient with chronic myeloid leukemia[204] (Figure 2.3A-C). The efficiency of amplification was determined over thefour highest template concentrations (40 pg, 5 pg, 625 fg, and 78.125 fg) as theslope from a linear least squares fit of log2(concentration) vs. cycle threshold(CT) and was found to be 0.988± 0.055. The standard deviation of CT val-ues was less than 0.15 at the three highest concentrations (SD = 0.08, 0.10, and0.14 for the 40 pg, 5 pg, and 625 fg samples, respectively), indicating uniformamplification across the array and technical error of less than 10 % in absoluteconcentration, near the limit of qPCR precision. The highest measurement vari-58Sup. 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 50−0.200.20.40.60.811.2CycleΔ RNNo WashWashCFigure 2.5: On-chip cell washing. (A) Measurements of GAPDH in cellswashed in PBS off-chip prior to injection into microfluidic device, withoutan on-chip wash contain background signal from template in supernatant.Without on-chip washing, untrapped cells remain in the capture chambers,resulting in fewer single-cell measurements (histogram inlayed). Detection ofresidual RNA after washing is dramatically reduced by comparison to off-chipresults (Figure 2.8) due to small-volume processing. (B) On-chip washing wasfound to reduce the background signal from free RNA in the supernatant anddramatically increased the number of single cells analyzed. (C) Comparisonof GAPDH measurements from loading purified RNA and washing, or notwashing, the cell-capture chambers.59ability was observed in the 78-fg sample, where shot noise (Poisson samplingnoise) is most pronounced and accounts for approximately 50 % of the measure-ment variance. Template amounts below 625 fg resulted in a digital pattern char-acteristic of single molecule amplification (49/50 for 78 fg and 19/50 for 10 fg)and consistent with the expected occupancy of chambers as determined by a bi-nomial distribution [191]. Based on the frequency of single molecule detectionin the 10-fg sample, we measured the average copy number of GAPDH to be979± 240 transcript copies per single-cell equivalent (20 pg) (Figure 2.3). Thismeasurement is comparable to previous reports [140] and is in close agreementwith an independent estimate based on normalizing the dilution series to CT val-ues obtained for single molecules (copies/20 pg = 1/2× copies/40 pg = 1/2× (1+efficiency)(CT(40 pg)−CT(single molecule)) = 1,407± 153 copies/20 pg). It should benoted that these estimates represent a lower bound because they do not accountfor RT efficiency; the RT efficiency of GAPDH has been previously estimated tobe approximately 50 % [134] but is dependent on transcript secondary structureand assay design. A comparison of CT values obtained from on-chip qPCR fromcDNA synthesized off-chip demonstrated that on-chip RT efficiency is equal to thatobtained off-chip when working from the same RNA concentrations (Figure 2.6).Finally, comparison of the same dilution series of RNA, assayed for GAPDH bothon-chip and in tubes (20-µL volume) (Figure 2.3C), showed that on-chip analysisprovides improved sensitivity.We next evaluated the efficiency and reliability of on-chip cell processing bycomparing our GAPDH measurements of purified RNA to measurements performeddirectly from single K562 cells (Figure 2.3C and Figure 2.7C). K562 cells wereloaded directly from culture medium followed by washing and analysis using achemical lysis and one-step RT-qPCR protocol (Cells DirectTM, Invitrogen). Us-ing a CT threshold of 31.5, corresponding to the mean CT of a single moleculeof GAPDH (CT = 30.5) plus two standard deviations (SD = 0.5), we observedsuccessful amplification in 100 % of single cells (N = 233) (Figure 2.7 A and B).Adjacent chambers that did not contain a cell were clearly separated from single-cell measurements with an average ∆ CT value of 5.7 (five empty chambers, threeof which amplified) (Figure 2.7 A and B and Figure 2.8). Consistent with pre-vious reports [205], we observed a log-normal distribution of GAPDH in single600 5 10 15 20 25 30 35 40 45 50−0.200.20.40.60.811.21.4CycleΔRNOn−ChipOff−ChipOn Chip Off Chip19.819.92020.120.2CTFigure 2.6: Comparison of GAPDH measurements from K562 cell lysatewith RT performed in the microfluidic device (on-chip) or RT performed intubes (off-chip) prior to qPCR in the device. Obtained CT values (Inset) showno significant difference in efficiency.cells with mean CT values of 20.3 (SD = 0.8) and an average of 1,761 (SD = 648)copies per cell (Figure 2.7C). These expression levels are consistent with previ-ous estimates in single cells [140]. Additionally, the mean CT of 20.3 observedfor single cells matches measurements of single-cell equivalent lysate (CT = 20.2,Figure 2.3D). Using digital PCR on cDNA prepared from K562 cell lysate, wemeasured an average of 1,229±72 GAPDH molecules per single-cell equivalent.We conclude that the relative efficiency of on-chip single-cell lysis and mRNA ex-traction/accessibility is equal to that achieved when working from RNA purifiedfrom large numbers of cells. Finally, as expected, RT-qPCR measurements fromchambers loaded with more than one cell show reduced variability and lower CTvalues for both measurements of mRNA and miRNA (Figure 2.5A and Figure 2.9,respectively). Taken together, these results establish the precise measurement ofmRNA abundance with single molecule sensitivity and the dynamic range neededfor single-cell analysis.611718192021222324252627Ct0 1000 2000 3000 4000051015202530Copy NumberNumber of Single Cells0 1 2 3 4 51820222426CtNumber of Cells0 1 2 3 4 5020406080100Percent1628ACBFigure 2.7: Single-cell loading and GAPDH expression measurements. (A)The locations of cells in each chamber along all six lanes of a device, asdetermined by brightfield microscopy, are represented as white circles andoverlaid on a heat map of CT values obtained from RT-qPCR measurements ofGAPDH in K562 cells. Red circles indicate NTC. (B) Scatter plot showing CTmeasurements for experiment shown in A. Histogram (Inset) shows 93.2 %single-cell occupancy. (C) Distribution of the number of GAPDH transcriptsmeasured in single K562 cells (N = 233).6222.8 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 23.422.423.320.723.023.023.023.023.019202122232425262728A BFigure 2.8: Single-cell miRNA measurements. (A) The locations of cellsin each chamber along all six lanes of a device, as determined by brightfieldmicroscopy, are represented as white circles and overlaid on a heat map ofcycle threshold (CT) values obtained from RT-qPCR measurements of miR-27a in K562 cells. Red circles indicate NTC. (B) Fluorescence image of entiredevice, corresponding to experiment in A after 30 PCR cycles. Cell corpsesremain after heat lysis and are visible as punctuate fluorescent spots adjacentto reaction chambers.2.3.3 Application to measurement of single-cell microRNAexpressionWe next applied our technology to the study of single-cell miRNA expression.miRNAs are thought to provide a unique signature of cellular state and are centralplayers in orchestrating development and oncogenesis, making them a promisingclass of biomarker for single-cell analysis [8, 111, 206]. Importantly, the shortlength of miRNAs (approximately 22 nucleotides) makes them difficult to detect631 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 miR-16 in hESC cell aggregates demonstratesthat the number of cells is reflected in corresponding CT values.by hybridization approaches, so that RT-qPCR is the dominant quantification strat-egy. To demonstrate the robustness and throughput of our technology, we per-formed a total of 1,672 single-cell measurements to examine single-cell variabilityin the expression of nine miRNAs spanning a wide range of abundance (> 16,000copies per cell to < 0.2 average copies per cell). K562 cells were again chosen asa heterogeneous population for this study because they are known to exhibit mixedcharacteristics of erythrocytes, granulocytes, and monocytes [204, 207]. We firstmeasured the expression of miR-16, a highly expressed microRNA that is foundin many tissue types [208] and has been suggested as a suitable internal standardfor normalization [63]. We found that miR-16 was log-normally distributed acrossK562 cells, but with slightly lower expression and notably tighter regulation thanGAPDH, having an average of 804 (SD = 261) copies per cell and a standard de-viation of 30 % (mean CT = 21.4, SD = 0.4). This strikingly low variability iswithin our estimates of cell volume differences (Figure 2.4). Matched experimentson single cells, isolated by micropipette into 20-µL volume tubes displayed an in-crease in measurement variability to approximately 90 % (mean CT = 29.5, SD =64C 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 measure-ments of miR-16 expression in K562 cells and hESCs. Measurements ofsingle K562 cells isolated using a microcapillary and assayed in 20-µL vol-umes are shown for comparison of technical variability. The observed shiftin mean CT values between on- and off-chip measurements is due to lowertemplate concentrations, and hence increased required PCR cycles, in the off-chip samples. (B) Differential expression of miR-223 between K562 cells andhESCs. Right-most bar indicates cells for which miR-223 was not detected(ND). (C) Mean single-cell miRNA copy numbers measured by RT-qPCR inthe microfluidic device compared to digital PCR measurements from bulk celllysate. Error bars represent standard deviation of single-cell measurements foreach miRNA. (D) One thousand five hundred and sixty-one single-cell mea-surements of the expression of 9 miRNAs in K562 cells. Reflected histogramsrepresent the expression distributions for each miRNA.0.9), demonstrating the improved precision of parallel microfluidic cell processingin nanolitre volumes (Figure 2.10A). Microlitre volume experiments also showed apronounced increase in measured CT values that results from the low concentrationof template and the large number of required PCR cycles.65To 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 [182, 207]. In contrast to miR-16, K562 cell miR-223 ex-pression was found to be highly variable (mean CT = 22.2, SD = 1.6, copy number= 513, SD = 406) and was not log-normally distributed (Figure 2.10B), consistentwith the known functional heterogeneity of K562 cells. These measurements high-light the utility of single-cell miRNA expression analysis for assessing the hetero-geneity of cell populations and for identifying miRNAs that are useful biomarkersof cellular state. To further explore this possibility, we measured the expression ofan additional seven miRNAs (nine total) and plotted the patterns of single-cell ex-pression in K562 populations (Figure 2.10D). Following the procedure describedabove, we used single molecule CT values, obtained by digital PCR, to translatemeasured CT values to absolute copy number. Assuming 100 % efficient amplifi-cation, we observed that the copy number, calculated as 2(CT(single cell)−CT(single molecule)),was well correlated (coefficient of 0.9932) with the average copy number obtainedby digital PCR of cDNA prepared from bulk lysates (Figure 2.10C). Single-cellmeasurements revealed distinct patterns of miRNA expression, with miR-16, miR-92, and miR-17-5p each exhibiting unimodal and tightly regulated distributions,whereas miR-223, miR-196a, and miR-145 showed multimodal distributions anda high level of cellular heterogeneity. Notably, for the lowest abundance miRNA,miR-200a, we detected expression in only a small fraction of cells and at levels be-low approximately five copies per cell. The average miR-200a copy number overall cells was within a factor of two of that obtained by digital PCR (0.2 copies percell). In contrast, miR-92 was found to be the most abundant miRNA and waspresent at approximately 16,000 copies per cell. These measurements establishedmiRNA quantification in single cells with a dynamic range of greater than 104 andat single molecule sensitivity.Finally, to illustrate the utility of single-cell measurements in precisely assess-ing differences in both the average expression and the heterogeneity between twodifferent cell populations, the expression levels of miR-16 and miR-223 in K562cells were compared to those in CA1S cells [200, 201], a human embryonic stemcell line (hESC). Although miR-16 was found to be expressed in hESC at similarlevels to K562 (∆CT = 0.6), we observed approximately a twofold greater variabil-66ity in expression (mean CT = 22.0, SD = 0.7) (Figure 2.10A). In contrast, whencompared to K562, single CA1S cell measurements of miR-223 showed strongdown-regulation, with miR-223 detected in only 3.6 % of cells. The absence ofsignificant miR-223 expression in hESC is expected due to the role of miR-223 asa differentiation-specific miRNA [182, 207].2.3.4 Co-regulation of miR-145 and OCT4 in single cellsThe measurement of multiple transcripts in single cells allows for quantitative mea-surements of gene coregulation that would otherwise be masked by cellular hetero-geneity [196]. To demonstrate this capability we designed an optically multiplexedassay to study the coregulation of miR-145 and OCT4, a known target of miR-145 [209], during the differentiation of hESCs (Figure 2.11 A-C). A total of 1,094single-cell measurements were performed at 0, 4, 6, and 8 days of differentia-tion. Cell distributions at each time point were used to map out the evolution ofthese transcripts and showed that average miR-145 levels increased approximately20-fold (copy numbers: D0: mean = 18.9, SD = 25.5, D8: mean = 380.3, SD =259.4) over 8 days. Increases in miR-145 were accompanied by progressive down-regulation of OCT4, ultimately reaching a 30-fold average decrease after 8 days(copy numbers: D0: mean = 755.7, SD = 306.4, D8: mean = 27.8, SD = 124.5).This decrease in OCT4 was independently verified by mRNA-FISH (Figure 2.12).Notably, single-cell analysis at day 6 showed a bimodal distribution in both OCT4and miR-145, revealing a transition of cellular state [209] that likely reflects thespontaneous differentiation of a subpopulation of cells. The observed single-celldynamics of miR-145 and OCT4 coregulation are not apparent in population mea-surements, highlighting the use of scalable single-cell transcriptional analysis incorrelating molecular signatures to cellular decision making [196].2.3.5 Single nucleotide variant detection in primary cellsFinally, to establish the specificity of our method we used multiplexed measure-ments of mRNA single nucleotide variants (SNVs) to assess the genomic hetero-geneity within a primary tumour sample. A total of 117 single cells isolated froma plural effusion of a metastatic breast cancer were assayed for the expression of a67 Day 0Day 4Day 6Day 8ND 0 1 2 3ND01234log   OCT4 Copy Number10log   miR-145 Copy Number102224262830ND2224262830NDMutant Expression (CT)Wildtype Expression (CT)94.6%72.6%5.4%7.7%0.0%0.9%0.0%18.8%K562 CellsPrimary CellsA BCDFigure 2.11: Optical multiplexing of single-cell RT-qPCR. (A) Multiplexedanalysis of the co-expression of OCT4 and miR-145 in differentiating hESC.Points are colour-coded to represent single-cell measurements (N = 547) foreach time point. Crosses represent population mean copy number. (B-C) His-tograms showing the distribution of each transcript are projected on the axeswith the mean copy number indicated by a dashed line. (D) Co-expressionmeasurements of SP1 wild-type and SNV mutant transcripts in primary cellsisolated from a lobular breast cancer sample. Mutant SP1 is detected in 23 of117 primary cells, and undetected in K562 cells (N = 37).68BAFigure 2.12: mRNA-FISH of OCT4 (red) counterstained with DAPI (blue) inCA1S cells. (A) Representative image of mRNA-FISH of OCT4 in a CA1Scell after 7 d of FBS differentiation. Estimate of average copy number ofOCT4 mRNA as determined by manual inspection of image stacks is 42 (SD= 41, N = 6). (B) Representative image of undifferentiated CA1S cells. Es-timate of average copy number as determined by manual inspection of imagestacks is 988 (SD = 368, N = 6). Scale bar, 10 µm.SNV mutant of the transcription factor SP1, previously identified by deep sequenc-ing [199] (Figure 2.11D). Primers were designed using sequences flanking the SNVlocation and do not discriminate between the genomic DNA and mRNA transcript.Of the 117 primary cells analyzed, 22 (18.8 %) were heterozygous for the mutantand wild-type allele, 85 (72.6 %) were homozygous wild type, 1 (0.9 %) was ho-mozygous mutant, and the transcripts were undetected in 9 (7.7 %). We did notdetect the SP1 mutation in 37 control K562 cells and failed to detect the wild-typetranscript in only 2 of these cells. In the absence of copy number alterations in theprimary sample, these observed frequencies would suggest a mutant to wild-typeSP1 ratio of 11.2 % (18.8×1+0.9×2 = 20.6 mutant to 18.8×1+72.6×2 = 164wild type). However, using digital PCR on purified DNA from the primary sam-ple, we found the ratio of mutant to wild-type SP1 alleles to be 18.7±2.3 %, inagreement with the previously reported ratio of 21.9 %, obtained by deep sequenc-ing [199]. The lower frequency of cells expressing the mutant SP1 allele may be69due to allelic expression bias or an amplification of the SP1 mutant allele, both ofwhich are supported by Shah et al. [199]. Regardless, given that the frequency oftumour cells within the original sample was approximately 89 % [199], both DNAmolecule counting and single-cell RNA expression measurements show that themetastasis of this tumour is derived from multiple cancer cell lineages.2.4 ConclusionHere we have demonstrated the implementation of scalable and quantitative single-cell gene expression measurements on an integrated microfluidic system. The pre-sented device performs 300 high-precision single-cell RT-qPCR measurements perrun, surpassing the throughput of previous microfluidic systems by a factor of ap-proximately 100. Further scaling the throughput to over 1,000 measurements on adevice with an area of one square inch is straightforward as each array element oc-cupies an area of 0.6 mm2. In terms of performance, we have established a dynamicrange of at least 104, measurement precision of better than 10 %, single moleculesensitivity, and specificity capable of discriminating the relative abundance of alle-les differing by a single nucleotide. Compared to tube-based single-cell RT-qPCR,microfluidic processing provides improved reproducibility, precision, and sensitiv-ity, all of which may be critical in identifying subtle differences in cell populations.Nanolitre volume processing also results in a 1,000-fold reduction in reagent con-sumption, thereby enabling cost-effective analysis of large numbers of single cells.In over 3,300 single-cell experiments, using adherent and suspension cell linesas well as clinical samples, we have shown that microfluidic RT-qPCR is well-suited to the quantitative analysis of miRNA expression and SNV detection, bothof which are difficult or inaccessible by alternative hybridization methods. No-tably, our device allowed for precise comparison of the distributions of GAPDHand miR-16 expression. miR-16 was found to be exquisitely regulated in K562cells, a finding that is striking given the known functional heterogeneity of this pop-ulation and the high variability in the expression of other measured miRNAs. Wepostulate that higher variability of GAPDH expression reflects the fundamentallystochastic process of transcriptional bursts followed by mRNA degradation. Incor-poration of miRNAs into the RNA-induced silencing complex (RISC) is known to70provide enhanced stability so that miRNAs are inherently less subject to temporalfluctuations; miRNAs may thus be particularly suited as biomarkers for assess-ing single-cell state and population heterogeneity. We anticipate that scalable andprecise single-cell miRNA analysis will become an invaluable tool in stratifyingpopulations of mixed differentiation state [111].In this chapter we have established the critical element of combining all single-cell-processing steps into an integrated platform. As there is limited utility in beingable to measure the expression of only one or two targets per cell, the primary aimof this section was to build a solid foundation on which increasingly advancedmicrofluidic single-cell RNA expression analysis tools may be built. This corefunctionality is expanded upon in chapters 3 and 4 to create microfluidic devicesfor the targeted, multiplexed measurement of RNA expression, and the genome-wide measurement of miRNA expression.71Chapter 3Highly multiplexed single-cellquantitative PCR13.1 IntroductionSingle-cell analysis preserves a wealth of information that is lost when measure-ments are instead taken by averaging cells together. While the importance of main-taining this resolution is well appreciated, techniques with the requisite sensitivityand scalability for single-cell molecular analysis have only recently been avail-able. Perhaps the most significant advancement in this field is the development oftechnologies for measuring the variations in and expression of nucleic acids, themain thrust of which has been measurements of mRNA expression levels. Thisrapid advancement of measurement technologies with increasing throughput, ro-bustness, and sensitivity has, in turn, sparked the development of new single-cellanalytics that meet the unique challenges associated with interpreting large single-cell data sets [210, 211]. As a result, single-cell RNA expression profiling nowprovides new avenues for the classification of cell types [212], the identificationof gene regulatory networks [213], and the high-resolution reconstruction of statetransitions [214].1A version of this chapter has been published: Michael VanInsberghe, Hans Zahn, Adam K.White, Oleh I. Petriv, and Carl L. Hansen. Highly multiplexed single-cell quantitative PCR. PLOSONE, 13(1):118, 01 2018.72Single-cell measurements of transcription can generally be categorized as (i)untargeted methods suitable for genome-wide discovery and profiling, or (ii) tar-geted methods for assessing the expression of a panel of genes. A variety ofgenome-wide techniques have emerged over the past five years, each based ondifferent methods of whole-transcriptome amplification and library construction,followed by high-throughput sequencing and bioinformatics-based quantificationusing read counts or binning of unique molecular identifiers [60, 137, 145, 215–217]. Targeted transcript measurements are best suited to situations where a finitenumber of biomarkers of interest have already been established, and are generallyfaster, more sensitive, and less expensive than global studies. Methods for tar-geted analysis include reverse transcription polymerase chain reaction (RT-qPCR)[115, 190, 193, 196] and single molecule counting using digital PCR [114, 191]or imaging of hybridization probes [77, 78]. Of these approaches, RT-qPCR is themost widely used and offers combined advantages of single-molecule sensitivity,speed to complete measurements in hours, and ease of assay design. The use of mi-crofluidic arrays for the final qPCR step has been particularly useful for single-cellanalysis, providing a means to greatly increase throughput and reduce costs [74,111, 116, 218]. This workflow, however, requires separate pre-processing steps forcell isolation, cDNA synthesis, and multiplexed pre-amplification to achieve theconcentrations of cDNA that are needed for nanolitre-volume detection (Table 3.1and Table 3.2).In this chapter, building on my work in Chapter 2, I report the development ofa microfluidic device that integrates the complete workflow for performing highlymultiplexed RT-qPCR on up to 200 single cells. The technical performance isfirst established on dilutions of purified RNA and then on-chip cell processing isvalidated by comparing multiplexed measurements of widely used housekeepinggenes to those made using existing single-cell qPCR solutions. This method isthen applied to measuring miRNA expression in heterogeneous cell populations.73Table 3.1: Comparison of single-cell gene expression methodsPlatformSingle-cell Detection Number Assayed Assay Cells per Assays Specializedmanipulation method of devices molecule length experiment per cell equipmentThis chapter Cell traps qPCR 1 cDNA 6 h 200 20-40Microfluidicfabrication, devicecontrol, and qPCRSingleCell trapsqPCR,1 cDNA 2.5-4 h 200-300 1-2Microfluidicmicrofluidic dPCR fabrication, devicesystems control, and qPCRor array scannerMultipleCell traps qPCR 2 cDNA 11 h 96 48-96Microfluidicmicrofluidic chips, devicesystems control, and qPCRBenchtop cell FACS, qPCR,1 cDNA 11 h+ 48-96 48-96FACS, microfluidicprocessing with capillary dPCR chips, devicemicrofluidic mouth pipet control, and qPCRquantificationBenchtop cell FACS,qPCR 0 cDNA 11 h+ ∼10-100 1-3 FACSprocessing and capillaryquantification mouth pipetRNA-FISH N/AMoleculeN/A RNA∼1 h∼1001-5 to 20 High-resolutioncounting to fluorescence microscope,overnight probe setsN/A N/A RNA ∼1.5 daysHigh-resolutionMultiplexed Molecule ∼100- 100- fluorescence microscope,RNA-FISH counting 40,000 1,000 fluid controllers,complex probe sets74Table 3.2: Performance comparison of single-cell gene expression methods. Dynamic range reported as orders ofmagnitude.Platform Pre-amplification Sensitivity Measurement Dynamic Exampleprecision range referencesThis chapter 12 PCR cycles Single-molecule +++ >4 Chapter 3, [219]Single None Single-molecule +++ >4 [114, 115]microfluidic systemsMultiple 18-22 PCR cycles Not investigated. ++ 5.5 [74]microfluidic systems Theoretical single moleculeBenchtop cell processing18-22 PCR cyclesNot investigated.++ 5.5 [111, 191, 196, 218]with microfluidic Theoretical limit dependsquantification on specific implementation.Benchtop cell processing None ∼10-100 molecules + ∼4 [190, 193]and quantificationRNA-FISH None Single-molecule +++ >4 [76, 77, 220, 221]Multiplexed RNA-FISH None Single-molecule +++ >4 [78, 79]753.2 Materials and methodsDevice fabricationMicrofluidic devices were fabricated using multilayer soft lithography [149]. De-vices were designed using CAD software (AutoCAD, AutoDesk Inc.) and printedto 5-inch chrome photomasks (HTA Enterprises) using a LW405 mask writer (Mi-crotech). All masks were printed with 8-µm resolution, except for the layer con-taining the cell traps, which was printed at 1-µm resolution. Photomasks wereprocessed according to manufacturer directions.The set of master moulds comprising one “flow” mould and one “control”mould were fabricated on 4-inch silicon wafers (Silicon Quest International). Thecontrol mould containing the valves and assay delivery channels was fabricatedusing a single layer of 25-µm high SU8-2025 (MicroChem). The flow mouldwas fabricated using six lithographic steps. First, an 8-µm high layer of Ship-ley’s Photoresist 220 (SPR220-7.0, DOW) was deposited to form the channels con-necting the sample-splitting chambers to the detection chambers. These channelswere rounded to enable valve closure by placing the processed wafer at 135 ◦C for10 min, then ramping to 190 ◦C and holding for 2 h. Next, 10-µm and 15-µm highlayers of SU8-2010 (MicroChem) were successively deposited to form the sample-splitting and cell-capture chambers, respectively. A 15-µm high layer of AZ-50XT(AZ Electronic Materials) was then added to form the remaining valveable chan-nels (between cell-capture sites, reagent delivery and general device control). ThisAZ layer was rounded by ramping the processed wafer from 65 ◦C to 190 ◦C in aconvection oven, holding for 4 h and slowly ramping to 25 ◦C; the channel heightafter rounding was 13 µm. Finally, the reagent injection channels were formedusing a 60-µm high layer of SU8-50 (MicroChem), followed by a 300-µm highlayer of SU8-2150 (MicroChem) to form the chambers for reverse transcription,pre-amplification, and detection. All resist processing was performed accordingto manufacturer specifications. Master moulds were coated with Parylene-C afterfabrication [222] to facilitate subsequent elastomer release.Microfluidic devices consisted of three layers of polydimethylsiloxane (PDMSRTV615, Momentive) elastomer, with a blank bottom layer, a middle layer con-76taining the channels on the control mould, and then the flow layer on top. The flowlayer was made by casting PDMS mixed at a 5:1 ratio (5 parts RTV615A to 1 partRTV615B) on the flow mould and baking at 80 ◦C for 30 min. The thin controllayer was made by spin-coating PDMS mixed at a 20:1 ratio at 1,800 rpm on to thecontrol mould and immediately baking at 80 ◦C for 20 min. After baking, both theflow and control layers were allowed to cool to room temperature. The flow layerwas then peeled off of the mould and aligned to the spin-coated, baked controllayer. Alignment was performed by hand under a dissection microscope progress-ing from one corner of the device to the other. As the feature array was too denseto allow for alignment marks, layer registration was performed using functionalfeatures. This structure was then baked at 80 ◦C for 30 min to allow the two layersto bond. After bonding, the combined slab was peeled off of the control mould andaccess ports were punched using a coring tool (CR0350255N20R4, Syneo). Inter-layer connections (called vias) were then laser ablated using a custom instrument[223]. The blank layer was prepared by spin-coating 20:1 PDMS at 1,800 rpm ona blank wafer and baking this at 80 ◦C for 20 min. The bonded, punched, and ab-lated control-flow slab was placed onto the baked blank layer and allowed to bondat 80 ◦C overnight. Finally, the combined 3-layer slab was peeled off the blankwafer, diced, and then bonded to glass slides using oxygen plasma (PDC-32G &PDC-FMG, Harrick Plasma). These mounted, finished chips were further cured at80 ◦C overnight prior to use.Some experiments presented herein were done with an early prototype thatcontained 52 cell-processing units per subarray, for a total of 208 cells per device.Two cell processing units per subarray were removed from the final version in orderto better accommodate the imaging area.Device operationMicrofluidic devices were operated semi-automatically using 21 solenoid actua-tors (Fluidigm Corp.) connected to a digital input-output card (NI PCI-DIO-32HS,NI PCI-6512, SCB-100, National Instruments) controlled using custom LabView(National Instruments) software. Teflon tubing fastened to 20-gauge hollow stain-less steel pins (Small Parts Inc.), or gel-loading pipet tips were used to connect the77chip to external pressure sources. Valves were operated at a pressure of 45 PSI,reagents were injected using 5 PSI, and cells were loaded at 1 to 2 PSI. Krytox 102(DuPont) oil was used in the control lines isolating the qPCR detection chambersand water was used in the remainder.Lysis, reverse transcription, and pre-amplification thermal incubations wereperformed on a flatbed thermocycler block (Bio-Rad DNA Engine PTC-200; MJResearch/Bio-Rad) with light mineral oil (Fisher Scientific) added between theblock and the glass slide.Microfluidic quantitative PCR was performed on a prototype version of theBioMark Instrument, (Fluidigm) [115] with the following fluorescence imagingcapabilities listed in Table 2.3.Cell LoadingCell-loading channels were first primed with 10 mg/mL bovine serum albumin(BSA, Gibco) in phosphate buffered saline (PBS, Gibco) [138]. Cells were thenloaded in the device by applying 1 to 2 PSI to a gel-loading pipet tip containing asingle-cell suspension (1×105 to 1×106 cells/mL) in complete media. As cellsflowed down the channel, they were captured at the start of each cell-processingunit by lithographically defined cell traps. After sufficient trap occupancy was ob-tained, cell wash buffer (1 mg/mL BSA in PBS) was flowed down the channel toremove untrapped cells, extracellular RNA, and debris. The cell isolation valveswere then closed to partition the cell loading channel and isolate individual cellprocessing units. Each cell trap was then visually inspected using brightfield mi-croscopy, and the status of each trap (single cell, multiple cells, debris, or empty)was recorded.Two-step RT-qPCRTwo-step RT-qPCR was used for all measurements made on microRNA. The work-flow was based on those designed for multiplexed miRNA profiling on 100-1,000sof pooled cells [111] and single-plex measurements on single cells [115]. A chip-operation schematic for the two-step workflow is presented in Figure 3.1.Following visual inspection of the cell-capture chambers, the cells were lysed781EmptyCell suspensionCell wash bufferRT mixPre-amplification mixWash solutionOpen valveClosed valveqPCR mixes23456789Load single-cell suspensionWash trapped cellsInspect and isolate trapped cellsLoad reverse transcription brewWash excess brew and isolate RT chamberFlush reagent injection line with pre-amplification mixLoad pre-amplifica-tion brewLyse cellsReverse transcriptionDiffusively mix RT brew with lysate10121314151617Wash excess brew and isolate pre-amplification chamberPush pre-amp product into sample-metering chambersLoad independent qPCR brews into the shared assay delivery chambers through the interlayer connectionPush pre-amp product into detection chambers with qPCR mixIsolate detection chambersAnalyze fluorescence images11PCR pre-amplificationDiffusively mix pre-amplification brew with cDNAqPCR amplificationDiffusively mix qPCR reagents with pre-amp productFigure 3.1: Operation schematic for the two-step RT-qPCR workflow79by placing the device onto a flatbed thermocycler and heating to 85 ◦C for 7 minand then cooling to 4 ◦C.Complementary DNA (cDNA) synthesis was next performed using the HighCapacity Reverse Transcription Kit (ABI). Gene-specific reverse transcription (RT)assays were first pooled to create a 5× working solution [2 µL of each of twenty5× assays combined with 60 µL of TE buffer pH 8.0 (Ambion)]. Reverse tran-scription brew [1 µL 10× RT buffer, 2 µL 5× RT assay pool, 0.125 µL 100 mMdNTPs, 1.625 µL 50 U/µL MultiScribe reverse transcriptase, 0.0625 µL 20 U/µLRNase inhibitor, 0.2 µL 5 % Tween-20 (Sigma Aldrich), brought up to 9.2 µL withUltraPure water (Gibco)] was flowed from the common reagent delivery bus overthe cell traps and into the 10-nL reverse transcription chamber. Excess RT brewwas washed out of the device by flowing 100 µL of reagent wash solution [0.1 %Tween-20 in UltraPure water] through the reagent injection lines. The device wasthen placed on a flatbed thermocycler for a 2-min incubation at 16 ◦C, followed by60 cycles of 2 min at 20 ◦C, 30 s at 42 ◦C, 1 s at 50 ◦C, then finishing with a 5-minincubation at 85 ◦C, and cooled to 4 ◦C.Pre-amplification was next performed by adding pre-amplification brew [12.5 µLof 2× Universal Master Mix (Applied Biosystems), 2 µL 25 mM MgCl2 (AppliedBiosystems), 1 µL 100 mM dNTP (Applied Biosystems), 1.25 µL 5 U/µL Ampli-Taq Gold (Applied Biosystems), 1.25 µL 20× pre-amplification primer pool, 0.5 µL5 % Tween-20, brought up to 20 µL with UltraPure water] through the commonreagent injection line, diluting the cDNA product into the 50-nL pre-amplificationchamber. After completely filling this chamber, the excess pre-amplification brewwas then washed out of the device by flowing 100 µL of wash solution throughthe reagent injection lines. The device was transferred to a thermocycler and in-cubated at 25 ◦C for 15 min to ensure complete mixing of cDNA product with thepre-amplification brew. Next, the device was incubated at 95 ◦C for 10 min, 55 ◦Cfor 2 min, then 6 cycles of 72 ◦C for 2 min, 95 ◦C for 15 s, and 60 ◦C for 4 min. Thedevice was then incubated at 99 ◦C for 10 min and then cooled to 25 ◦C. The 20×pre-amplification primer pool contained 3 µL of each 20× assay brought to a finalvolume of 100 µL with TE pH 8.Following pre-amplification, the reagent injection lines were again washedwith 100 µL of wash solution, which was then used to push the pre-amplification80product into the 0.15-nL sample splitting chambers. Valves were then actuatedto isolate the sample-splitting chambers. The independent qPCR brews [12.5 µL2× Universal Master Mix no UNG (Applied Biosystems), 1.25 µL 20× TaqManqPCR assay, 2.5 µL 1 % Tween-20, 0.25 µL 40× ROX Reference Dye (Invitrogen),brought up to 24.42 µL with UltraPure water] were then used to push the meteredpre-amplification product into the 6.4-nL detection chambers. The device was thentransferred to an instrument for microfluidic real-time PCR and thermocycled withthe following conditions: 95 ◦C for 10 min, then 40 cycles of 95 ◦C for 15 s, 60 ◦Cfor 1 min and fluorescence imaging at 60 ◦C.One-step RT-qPCROne-step RT-qPCR was used for all measurements made on mRNA and was per-formed using the CellsDirect kit (Invitrogen). Processing steps are similar to thoseused for the two-step workflow, with the main distinctions being that the “RT cham-ber” was instead used for cell lysis, and the “pre-amplification chamber” was usedfor both cDNA synthesis and pre-amplification steps. A chip-operation schematicfor the one-step workflow is presented in Figure 3.2.Following visual inspection of the cell-capture chambers, the cells were lysedby flowing lysis buffer [10 µL Lysis Resuspension Buffer (Invitrogen), 1 µL LysisEnhancer Solution (Invitrogen)] over the cell traps into the 10-nL “RT chamber”.Excess lysis buffer was washed out of the reagent injection line with 100 µL ofwash solution. The device was then incubated at 25 ◦C for 10 min, then 70 ◦C for10 min and cooled to 4 ◦C.cDNA synthesis and pre-amplification were then performed by pushing thelysate into the 50-nL pre-amplification chamber with the one-step RT-PCR brew(12.5 µL 2× Reaction Mix with ROX, 0.5 µL 50 mM MgSO4, 1.67 µL 15× pooledprimer mix, 0.5 µL Superscript III/Platinum Taq enzyme mix, 2.5 µL 1 % Tween-20, brought to a final volume of 19.62 µL with UltraPure water). The 15× pooledprimer mix contained 1.5 µL of each 20× TaqMan assay, brought to a final volumeof 40 µL with TE pH 8.0. The excess brew was then washed out of the devicewith 100 µL of wash solution. The device was then transferred to a thermocyclerand incubated at 25 ◦C for 15 min for diffusive mixing, then at 50 ◦C for 20 min for81EmptyCell suspensionCell wash bufferCell lysis bufferRT/pre-amplification mixWash solutionOpen valveClosed valveqPCR mixesLoad single-cell suspensionWash trapped cellsInspect and isolate trapped cellsFlush reagent injection line with one-step RT/pre-amplification mixLoad RT/pre-amplifi-cation brew12346785Load lysis bufferWash excess buffer and isolate lysis chamberWash excess brew and isolate RT/pre-amplification chamberPush pre-amp product into sample-metering chambersLoad independent qPCR brews into the shared assay delivery chambers through the interlayer connectionPush pre-amp product into detection chambers with qPCR mixIsolate detection chambersAnalyze fluorescence images911121314151610Reverse transcription and PCR pre-amplificationDiffusively mix RT/pre-amp brew with lysateqPCR amplificationDiffusively mix qPCR reagents with pre-amp productLyse cellsFigure 3.2: Operation schematic for the one-step RT-qPCR workflow82cDNA synthesis, followed immediately by 95 ◦C for 2 min, then 12 cycles of 95 ◦Cfor 15 s, 60 ◦C for 4 min. The device was then incubated at 99 ◦C for 10 min andthen cooled to 25 ◦C.Following cDNA synthesis and pre-amplification, the reagent injection lineswere again washed with 100 µL of wash solution, and this was then used to pushthe pre-amplification product into the 0.15-nL sample splitting chambers. Valveswere then actuated to isolate the sample splitting chambers. The independent qPCRbrews [12.5 µL 2× Universal Master Mix no UNG (Applied Biosystems), 1.25 µL20× TaqMan assay, 2.5 µL 1 % Tween-20] were used to push the pre-amplificationproduct into the 6.4-nL detection chambers. Quantitative PCR was then performedwith the following conditions: 95 ◦C for 10 min, then 40 cycles of 95 ◦C for 15 s,60 ◦C for 1 min and fluorescence imaging at 60 ◦C.The following assays were used for mRNA detection: GAPDH Hs02758991 g1(ABI); TBP Hs00427620 m1 (ABI); GGCX Hs00168139 m1 (ABI); BCR-ABLHs01036528 m1 (ABI); and RPPH1 forward GAGGTCAGACTGGGCAGGAG, re-verse CCTCACCTCAGCCATTGAACTC, probe FAM-TGCCGTGGACCCCGCCCTT-CG -BHQ1 (Integrated DNA Technologies) [107].Cell culture and RNA purificationK562 cells were obtained from ATCC (CCL243) and were cultured in DMEM(Gibco) supplemented with 10 % FBS (Gibco) and 1× GlutaMAX (Gibco). IL-3independent BaF3 cells were a generous gift from Dr. James Piret at the Uni-versity of British Columbia. They were cultured in RPMI-1640 (Gibco) supple-mented with 10 % FBS and 1× GlutaMAX. K562 cells had an average diameter of16.8±1.3 µm, and BaF3 cells had an average diameter of 13.5±1.9 µm.Purified total RNA was extracted from K562 cells using the mirVana RNAisolation kit (Ambion) according to manufacturer directions, omitting the optionalsize-enrichment step.Data analysis and plottingData are reported as mean ± standard deviation (SD). Fluorescence images wereanalyzed using scripts written in MATLAB (MathWorks, version 2014b) to ex-83tract real-time PCR curves and algorithmically calculate cycle-threshold (CT) val-ues [115]. All subsequent data analysis was performed using R (version 3.4.0)with plyr (v1.8.4) and reshape (v0.8.6) packages. Plots were generated using gg-plot2 (v2.2.1), ggbeeswarm (v0.6.0), cowplot (v0.7.0), gplots (v3.0.1), corrgram(v1.12), and RColorBrewer (v1.1-2). Values for boxplots were calculated usingthe default settings in geom boxplot (ggplot2): the middle line represents the me-dian, the lower and upper hinges correspond to the first and third quartiles, theupper whisker extends from the upper hinge to the largest value no further than 1.5times the interquartile range, the lower whisker extends from the lower hinge to thesmallest value no further than 1.5 times the interquartile range, and data beyond thewhiskers are deemed outliers and plotted individually. Data have been depositedin the NCBI Gene Expression Omnibus under accession GSE102734. Details forspecific sections are reported below.Single-molecule sensitivityTotal RNA at limiting dilution was analyzed using the one-step RT-qPCR work-flow with 12 cycles of pre-amplification and assayed against GAPDH. A single-molecule CT cut-off was calculated to contain 99 % of the distribution of CT valuesfrom cell-processing units that generated a signal in all twenty detection chambers(mean CT of 21.5 added to 2.5758 multiplied by the standard deviation of 0.67).Cell processing units were defined to have a successful single-molecule amplifica-tion event if greater than 75 % of the detection chambers had a CT value lower thanthis cut-off.The digital array response curve [106] for 52 chambers was used to convert thenumber of positive cell-processing units per array to the expected number of inputmolecules and corresponding 95 % confidence intervals (Table 3.3).Expression analysisAmplification efficiency was measured based on the slope of a linear least-squaresfit of the log of the total RNA concentration versus cycle threshold values. Datawere pooled from three separate device runs, and the inverse of the standard de-viations of the measurements at each concentration were used as the fit weights.84Efficiency was calculated as 100 %× (1−10−1/slope). Efficiency uncertainties arereported as a range based on the standard error (SE) from the least-squares fit.Cycle threshold values were converted to copy number based on our measure-ments of the GAPDH single-molecule CT and amplification efficiency, namelyCopy number = 2CTsingle molecule− CTmeasured . For miRNA analysis, five cycles wereadded to the GAPDH cut-off to account for the difference in pre-amplificationPCR.Differential miRNA expression between K562 and BaF3 cells was assessedusing a Wilcoxon rank-sum test. P-value adjustments for multiple testing werecomputed using the Benjamini-Hochberg correction, and significance was deter-mined at a false discovery rate (FDR) of less than 0.01. Hierarchical clusteringwas performed on log-scaled expression values using (1-pearson correlation)/2 asa distance metric and ward.D2 linkages.Measurement precision was calculated as the standard deviation σ divided bythe mean µ for the copy-number normalized measurements on purified RNA at thespecified concentration; i.e., Precision = 100% × σ200 pgµ200 pg .Correlation analysisExpression correlations between transcripts were computed using Spearman’s rho.The significance of a correlation coefficient was assessed based on a null distri-bution derived by recalculating the correlation coefficient for randomly permutedsingle-cell expression measurements. Correction for multiple testing across multi-ple pairs was controlled using a Benjamini-Hochberg correction. Significance wasassessed at an FDR of 0.01.3.3 Results3.3.1 Device design and operationWe designed a microfluidic device that integrates and parallelizes the processingsteps necessary for multiplexed gene expression analysis, from cell capture throughto qPCR. The resulting architecture uses several design elements from our previ-ous systems for simplex quantitative and digital PCR [114, 115], including cell85traps and sequential reaction chambers. In contrast to these implementations, how-ever, the reaction products from each cell are split across twenty independentlyaddressable reaction chambers for the final qPCR readout. An important consid-eration during the design process was to maintain measurement sensitivity duringthis multiplexed analysis. Even under ideal conditions, splitting the contents of asingle cell across several different reactions severely limits the detection sensitivity.Therefore, a key element of our design addresses this potential shortcoming withthe inclusion of fluidics that allow for a low-cycle multiplexed pre-amplificationstep prior to simplex qPCR.The device features four linear arrays of 50 cell-processing units for a totalcapacity of up to 200 cells analyzed per device (Figure 3.4). Each cell process-ing unit (Figure 3.3A) is comprised of (i) a reagent injection channel, (ii) a 0.3-nLcell capture chamber, (iii) a 10-nL reverse transcription (RT) chamber, (iv) a 50-nLPCR pre-amplification chamber, (v) twenty 0.15-nL sample splitting chambers,(vi) twenty shared assay loading chambers, and (vii) twenty 6.4-nL qPCR detec-tion chambers. During operation, reactions are assembled in parallel, with reac-tion brews transferring and mixing with previously generated intermediates in thenext reaction chamber. Sequential reagent metering is achieved by dead-end fillinglithographically defined volumes, causing trapped air to be expelled into the gas-permeable PDMS elastomer from which the device is made. The reaction chamberswere configured to provide sufficient dilution between each processing step so asto avoid reaction inhibition, and in such a way to accommodate a variety of assaytypes, including one- and two-step RT-PCR workflows.In performing a two-step RT-PCR workflow (Figure 3.1), device operation firstbegins with passivation of the cell-capture chambers to avoid cell adhesion dur-ing loading. Next, a single-cell suspension is injected into the device and flowsthrough channels that have lithographically defined cell traps designed to isolateand capture cells at the front of each cell processing unit (Figure 3.3B and D).Following cell trapping, the cell-loading channel is flushed with cell wash bufferto expel any untrapped cells and extracellular RNA. The trapped cells are thenisolated by actuation of separation valves, and the device is placed on a thermo-cycler to perform heat lysis. Reverse transcription brew is then injected throughthe cell-capture chamber into the neighbouring RT chamber, and allowed to diffu-86Control layer:Flow layer:8 µm SPR15 µm AZ10 µm SU815 µm SU860 µm SU8300 µm SU8Control valvesAssay deliveryInterlayer connectioni   Reagent injectionii   Cell trapv   20 sample-splitting chambers vi  20 shared assay-loading chambers vii  20 detection chambers0.5 mmiii  Reverse transcription chamberiv  Pre-amplification chamberAB CD Cell loading E Reverse transcription F PCR pre-amplification G Sample splitting and qPCREmptyCell suspensionCell wash bufferRT mixPre-amplification mixWash solutionOpen valveClosed valveqPCR mixesFigure 3.3: Multiplexed RT-qPCR device schematic and operation. (A)Two cell-processing units. Components include (i) a reagent injection bus,(ii) a 0.3-nL cell capture chamber, (iii) a 10-nL reverse transcription (RT)chamber, (iv) a 50-nL pre-amplification chamber, (v) twenty 0.15-nL sample-splitting chambers, (vi) twenty shared assay-delivery chambers, and (vii)twenty 6.4-nL detection chambers. The “flow” layer is made up of featuresof six different heights. Control valves and assay-delivery channels are on thesame 25-µm high SU8 “control” layer. Assay delivery from the control layerto the flow layer occurs through laser-ablated interlayer connections. Scalebar 0.5 mm. (B) Optical micrograph of a single K562 cell (indicated by ablack arrow) caught in a cell trap. Scale bar 50 µm. (C) Optical micrograph ofa subsection of the detection array. Control valves are coloured red. Scale bar0.5 mm. (D-G) Schematic illustration of device operation. (D) Single cellsare first loaded into the device, then washed, isolated, and lysed in situ. (E)Reverse transcription brew is then injected into the RT chamber, mixed withthe lysate, and then the device is thermocycled. (F) Similarly, multiplexedpre-amplification mix is injected into the device, mixed with cDNA, and thenthe device is again thermocycled. (G) Finally, the pre-amplification productis split between detection chambers for qPCR.87Cell inlet valveCell wash inletReagent inletsCell inletHydrationqPCR isolationvalveCell wash inletCell inletCell wash inletCell inletCell wash inletCell inletRT-to-PCR valveReagent inlet valveReagent inlet valveCell wash valveCell inlet valveCell inlet valveCell inletvalveCell isolationvalveCell-to-RTvalveReagent injectionvalveReagent outlet valveCell outlet valveqPCR inlet valveCell outlet valveCell outlet valveReagent outlet valveOutletHydrationSample splitting valveCell outlet valveReagent outlet valveqPCR assay 1qPCR assay 2qPCR assay 3qPCR assay 4qPCR assay 5qPCR assay 6qPCR assay 7qPCR assay 8qPCR assay 9qPCR assay 10qPCR assay 11qPCR assay 12qPCR assay 13qPCR assay 14qPCR assay 15qPCR assay 16qPCR assay 17qPCR assay 18qPCR assay 19qPCR assay 205 mm88Figure 3.4 (previous page): Multiplexed RT-qPCR microfluidic deviceschematic. Schematic of microfluidic device for performing 200 20-plexsingle-cell RT-qPCR reactions. Features on the “flow” layer are indicatedin blue, those on the “control” layer are indicated in red and interlayer con-nections are shown in yellow. Scale bar 5 mm.sively mix with cell lysate (Figure 3.3E). The RT chamber is then isolated, and thedevice is again placed on a thermocycler for temperature control during cDNA syn-thesis. Next, the reagent injection channel is flushed with pre-amplification brew,and injected through the cell-capture and RT chambers, thereby pushing cDNAinto the pre-amplification chamber (Figure 3.3F). The pre-amplification chamberis then isolated, and the device is transferred to a thermocycler for low-cycle PCRpre-amplification of all targets using multiplexed primer sets. On completion ofthe pre-amplification, ∼20 % of the resulting product is pushed into the sample-splitting chambers. Each individual gene-specific qPCR assay brew is then injectedinto a shared assay-loading chamber, through each isolated sample-metering cham-ber, and into each of the qPCR detection chambers (Figure 3.3G). These detectionchambers are then isolated and the device is placed on a thermocycler with fluo-rescence imaging capabilities to perform qPCR. Finally, custom image processingsoftware is used to extract the fluorescence signal in each chamber during PCR cy-cling, generate characteristic amplification curves for qPCR-based quantification,and calculate cycle threshold (CT) values.Device operation for performing one-step RT-PCR (Figure 3.2) closely followsthe two-step protocol, except that a non-denaturing lysis buffer is flowed over thetrapped, washed, and isolated cells into the “RT chamber”, and one-step RT-PCRbrew is then used to push the cell lysate into the “pre-amplification chamber” toperform both reverse transcription and PCR pre-amplification.A single device run consists of 4,000 qPCR reactions, can easily be completedwithin a day, and, with optical multiplexing, can produce upwards of 8,000 single-cell measurements. We note that implementing multistep processing at this scalerequired the fabrication of devices with very dense fluidic integration; each deviceintegrates roughly 15,000 valves and 2,800 layer interconnects (vias) within a total89footprint of 3.5×5 cm (Figure 3.4). To achieve this reaction density we combinedseveral advancements: high aspect ratio lithography (> 5:1) to increase reactor vol-umes using minimal device area, a 64 % reduction in the standard valve size that isused in multilayer soft lithography (MSL) [149], and the use of three-dimensionalmicrofluidic fabrication techniques [223] to support complex fluid routing.3.3.2 Device characterizationTo benchmark the performance of our device we assessed assay linearity, precision,and sensitivity under different workflows and using assays of varying complex-ity. In order to remove the contributions of cell-to-cell variability, these measure-ments were done using dilution series of purified total RNA obtained from K562cells, a human BCR/ABL1 positive cell line from a patient with chronic myeloidleukemia [204]. Input RNA concentrations were varied from 200 to 0.2 pg percell-processing unit, corresponding to approximately 10 to 1/100 cell-equivalents,assuming 20 pg of total RNA per cell [115].We first tested assay linearity and amplification uniformity using a simplex as-say for the expression of GAPDH, a widely expressed gene that is often used asan endogenous control. A GAPDH hydrolysis probe assay and 12 cycles of spe-cific target pre-amplification was used with the one-step workflow for RT-qPCR(Figure 3.5A). The efficiency of amplification, as determined by the slope of thelinear least-squares fit of log total RNA concentration versus CT, was measuredto be 100 % (SE: 96 to 105 %) with an R2 of 0.998 (Figure 3.5B). The CT val-ues obtained across all replicates ranged from 9.57± 0.22 to 19.49± 0.58, indi-cating uniform amplification across the array (Figure 3.5C), and a measurementprecision at the higher concentrations of approximately 14.5 %, approaching thelimit of qPCR. While the technical variability increased at lower concentrations,this noise can be attributed to stochastic sampling in the initial RNA partitioning,as the variability within each cell processing unit was significantly lower than thevariability between cell processing units (Figure 3.5B-C, Figure 3.6) (e.g., standarddeviation of the CTs from the lowest concentration was 0.58± 0.05, whereas theaverage standard deviation of CTs from each cell processing unit was significantlydifferent at 0.18±0.07, p = 0.0034, Wilcoxon rank-sum test).90200 pg20 pg2 pg0.2 pg0.00.40.81.210 20 30 40CycleΔRN101520log10 RNA [pg]CTmiR-17-5p0 1 2GAPDHSlope: -3.32 (SE 0.11)Efficiency: 100% (SE 96-105%)R2 = 0.998Slope: -3.31 (SE 0.20)Efficiency: 100% (SE 93-109%)R2 = 0.99310 15 20CT20 21 22 23CTA BC DWe next assessed these same performance metrics under demanding multi-plexed assay conditions. A two-step workflow was used with a pool of 20 commer-cially available stem-loop microRNA assays [110] for gene-specific cDNA synthe-sis and 7 cycles of pre-amplification (let-7b-5p, let-7c-5p, let-7d-5p, miR-10a-5p,150-5p, 155-5p, 16-5p, 17-5p, 17-3p, 181a-5p, 200c-3p, 20a-5p, 221-3p, 223-3p,24-3p, 27a-3p, 29b-3p, 451, 93-5p and 184-3p). In order to measure amplifica-tion uniformity across the device, these multiplexed reaction intermediates wereall assayed against one of the highly expressed microRNA from this pool (miR-17-5p). Based on three independent device replicates, the amplification efficiency wasagain measured to be 100 % (SE: 93 to 109 %) with an R2 of 0.993 (Figure 3.5B).The CT values from all replicates ranged from 12.5±0.6 to 22.6±1.2, correspond-ing to a precision at the higher concentrations of approximately 31.5 %.We next characterized the detection sensitivity by measuring RNA at limiting91Figure 3.5 (previous page): Device characterization and performance. (A)4,160 real-time amplification curves generated from one experiment on a 10×dilution series of purified total K562 RNA. Each curve is generated from pro-cessing fluorescence images of the entire device taken after each cycle ofPCR. The threshold for determining CT values is indicated by the dashed line.(B) Standard curves derived from three replicate 10× dilution series of K562RNA that validate the one- and two-step workflows measured using GAPDH(blue) and miR-17-5p (green), respectively. Points represent the mean ±standard deviation of all 1,040 detection chambers per concentration fromall three replicates. Input RNA ranged from 200 pg (10 cell equivalent) to0.2 pg (1/100 cell equivalent) per cell-processing unit. (C) Heatmap represen-tation of the CT values extracted from the real-time curves shown in A. Valuesare organized according to the device layout; the leftmost panel correspondsto the lowest concentration, the rightmost panel corresponds to the highestconcentration and each row within each panel originates from the same cell-processing unit. The three rightmost panels (200 pg/cell to 2 pg/cell process-ing unit) indicate uniform amplification across the device. The variabilitywithin the leftmost panel (0.2 pg/cell-processing unit) is much higher thanthat within each cell-processing unit (row), illustrating the increased contri-bution of stochastic sampling towards the variability of this measurement. (D)Heatmap of CT values for GAPDH measured on a 5× dilution series at lim-iting RNA concentrations. Values are organized according to device layout,with the highest concentration seen on the left. Black areas were not detected.After sufficient PCR pre-amplification, cell-processing units that initially con-tained a single cDNA molecule were amplified to levels such that signal isproduced in an average of 19.8/20 detection chambers.dilution. In the absence of amplification prior to detection, single cDNA moleculeswould only be sporadically seen in individual detection chambers of positive cell-processing units. However, with pre-amplification, ideally enough product is gen-erated from a single cDNA molecule to result in the above-background qPCR am-plification in all twenty detection chambers. Four concentrations of purified K562RNA at limiting dilution were mixed with one-step RT-qPCR brew, each loadedinto one of the linear arrays, processed according to the one-step workflow, andassayed against GAPDH. The resulting pattern of CT values across the device hadseveral characteristics indicative of single-molecule amplification (Figure 3.5D,920.2 pg2 pg20 pg200 pg01020300102030010203001020300.0 0.2 0.4 0.6Standard deviationNumber of cell processing unitsReplicate123Figure 3.6: Multiplexed RT-qPCR measurement variability. Standard de-viations from the measurements derived from the cell-processing units (his-togram) or the entire subarray (vertical lines) for each of three experimentreplicates. There is a slight, but significant (mean SD = 0.125 to 0.179 for 200and 0.2 pg/unit, respectively; p = 9.3×10−18, Kruskal-Wallis rank-sum test)shift in the distributions derived from the cell-processing units, with thosefrom lower RNA input amounts seeing higher variability. This shift, how-ever, is much smaller than that seen between the variabilities calculated fromthe entire subarray (i.e., between vertical lines in figure) (mean SD = 0.156to 0.579 for 200 and 0.2 pg/unit, respectively; p = 0.0273, Kruskal-Wallisrank-sum test). Furthermore, the difference in variability between the cell-processing units and the full array is only significant between the two low-est concentrations (200 pg: p = 0.333, 20 pg: p = 0.264, 2 pg: p = 0.0105,0.2 pg: p = 0.0105; Wilcoxon rank-sum test, Benjamini-Hochberg correc-tion). We attribute this difference to the effects of stochastic sampling duringRNA partitioning and initiation of cDNA synthesis.93202428CTA020406020 24 28CTNumber of detection chambersBFigure 3.7: Single-molecule cycle threshold cut-off. (A) Heatmap of unpro-cessed CT values used to calculate a cut-off cycle threshold value for a singlecDNA molecule. (B) Histogram of unprocessed CT values with the calculatedcut-off shown in red.Figure 3.7). First, there was clear separation in the chip-wide distribution of CTvalues between single cDNA molecules and non-specific background (Figure 3.7).The average single-molecule CT was found to be 21.4± 0.6, corresponding to acut-off CT (capturing 99 % of the distribution) of 23.2. Second, using this cut-off, positive detection chambers were organized based on their respective cell-processing unit (Figure 3.5D, amplification is seen across rows). In these posi-tive cell-processing units, an average of 19.8 out of 20 detection chambers showedamplification, whereas only 0.1 out of 20 amplified in negative units. Finally, theobserved dilutions between successive linear arrays were found to be in agreementwith shot-noise resulting from a Poisson distribution of starting template molecules(Table 3.3). Together, these results demonstrate a lower sensitivity limit of a singlecDNA molecule. We note that successful detection, however, depends on conver-sion from RNA to cDNA, a process that has been previously seen to be variablebetween templates, assays, and enzymes [224].We next evaluated the performance of on-chip cell processing by measuringgene expression in 175 single K562 cells. These tests were performed using anassay panel designed to interrogate genes that spanned a wide range of abundanceand included (i) three genes typically used as endogenous controls in bulk gene ex-94Table 3.3: Single-molecule dilution detection measurements. Expected num-ber of molecules and 95 % confidence intervals (CI) based on the digital arrayresponse curve for a 52-chamber array. Cell-processing units were counted aspositive if more than 15 of the 20 detection chambers (75 %) had a CT valueless than the cut-off.Array DilutionNumber of Expected Lower 95 % Upper 95 %positive units molecules CI CI1 1 36 60.7 48 792 1/5 5 5.3 4 73 1/25 1 1.0 1 14 1/125 1 1.0 1 1pression experiments (human GAPDH, RPPH1, and TBP), (ii) a fusion transcriptassociated with CML (BCR-ABL), iii) a heterogeneously expressed gene (GGCX),and iv) a negative control (murine Gapdh). Using the previously derived single-molecule CT cut-off, and requiring positive amplification in all replicates for eachcell, we detected both RPPH1 and GAPDH in 100 % (N = 175), TBP in 92 %,BCR-ABL in 92 %, and GGCX in 31.4 % of single cells (Figure 3.8A). As expected,murine Gapdh was not detected in the vast majority of reactions, with only 4 outof 525 detection chambers generating signal, and these hits were spread across dif-ferent cell processing units. While we initially expected both TBP and BCR-ABLto be ubiquitously expressed, six of the cells (3.4 %) failed to generate signal in allreplicates for each gene. This repeated absence suggests that these cells are truenegatives, which, consistent with the low copy-number of these transcripts, mightbe explained by stochastic expression in some cells. Adjacent processing units thatdid not contain a cell (no cell controls, NCC) were clearly separated from single-cell measurements (Figure 3.8A). In these control samples, TBP, BCR-ABL, andGGCX all showed no amplification. RPPH1 and GAPDH had an average differ-ence of 154 and 116 fold, respectively, showing that cross-contamination arisingbetween cells or due to free nucleic acid contamination is less than 1 part in 100.Finally, similar to the effects of stochastic sampling seen between measurementsof low concentrations of purified RNA, the measurement variability between repli-cates was substantially smaller than the expression variability seen between single95A B1 2 3 4log10 ExpressionRPPH1GAPDHTBPBCR-ABLGGCXGapdh175 single K562 cells 4 NCCRPPH1 GAPDH TBP BCR-ABL GGCX01234log 10 ExpressionSingle cellWhite et al. [115]White et al. [114]Taniguchi et al. [190]Verma et al. [225]Figure 3.8: Multiplexed single-cell mRNA expression. (A) Heatmap of ex-pression values of six mRNA on 175 single K562 cells and 4 no-cell con-trols (NCC). Each column represents a cell, and each row represents an assay.Replicate measurements are grouped according to assay. As expected, murineGapdh was not detected, only spuriously generating signal in 4/525 detectionchambers. No-cell controls are clearly distinguishable from those from singlecells, with an average difference in magnitude greater than 100×. The vari-ability between replicate measurements for each gene is much smaller thanthe variance in expression seen between different cells. (B) Distribution ofthe copy number of each gene measured in each single cell, along with themean expression values obtained from previously published single-cell qPCRmeasurements [114, 115, 190, 225]. Error bars represent the mean± standarddeviation.cells (Figure 3.9).As expected, these genes were expressed with a wide array of abundances span-ning approximately four orders of magnitude (Figure 3.8B). The average numberof cDNA molecules per cell were 7,734 (SD = 3,284) for RPPH1, 1,488 (SD = 815)for GAPDH, 7.2 (SD = 5.5) for TBP, 17.7 (SD = 16.2) for BCR-ABL and 1.4 (SD= 2.4) for GGCX. For GAPDH, these results closely match our earlier validationexperiments on purified K562 RNA, which had an average of 1,730 (SD = 387)cDNA molecules per single-cell equivalent (20 pg) (cells: mean CT 12.91, SD =0.84, purified RNA: mean CT 12.50, SD = 0.32) and are also consistent with ourpreviously published estimates of GAPDH copy-number in single K562 cells [979(SD = 240) to 1,761 (SD = 649) copies per cell, N = 233 and 1,421 (SD = 599) to96-6-4-20log10 Expressionlog 10 CVRPPH1GAPDHTBPBCR-ABLGGCXSingle cellPopulation0 1 2 3 4Figure 3.9: Variability of single-cell mRNA measurements. While not fullyindependent, replicate qPCR measurements (N = 3 for RPPH1, GAPDH andBCR-ABL, N = 4 for TBP and GGCX) from each single cell (circles) give ameasure of the qPCR variability within each cell-processing unit. In all cases,this variability is smaller than the expression variability between single cells(triangles).1,741 (SD = 573) copies per cell, N = 273 for references [114, 115], respectively].Similarly, our results on BCR-ABL agree with previous measurements done by us[33 (SD = 18.9) copies per K562 cell, N = 242, ref. [114]] and Verma et al. [49.9(SD = 57.2) copies per K562 cell, N = 69, ref. [225]], and those on TBP corre-spond with an independent measurement of 13.7 (SD = 7.9) copies per HCT116cell, N = 14 [190]. We note that there was only a modest correlation in expres-sion between the endogenous control genes (average spearman correlation 0.51,SD = 0.17, Figure 3.10), suggesting that their expression is not simply a reflectionof cell size and indicates the presence of additional sources of variability. Thisobservation further reiterates previous recommendations against using traditionalreference genes in normalizing single-cell expression data [114, 191, 226].Together, these multiplexed RT-qPCR gene expression measurements on singlecells are all in close agreement with both the quantities and variabilities obtainedin previous studies. Combined with our initial validation experiments on purifiedRNA, they demonstrate the requisite precision, sensitivity, and specificity over the97rs = 0.702***rs = 0.409***01234GAPDHRPPH1rs = 0.413***01234GAPDHTBP01234RPPH10 4321 0 43210 4321TBPFigure 3.10: Co-expression of endogenous control genes. Only modest co-expression is observed (average spearman correlation coefficient of 0.51, SD= 0.17). Spearman correlation coefficients, rs and the corresponding co-expression significance are denoted; *** p < 0.001.dynamic range necessary for single-cell analysis.3.3.3 Measurement of single-cell miRNA expressionMicroRNAs are a class of small (∼ 22 nt) non-coding RNA that direct the RNA-induced silencing complex to post-transcriptionally degrade, destabilize, or repressmRNAs [23]. They are involved in a vast array of biological processes includingdevelopment [39] and oncogenesis [40], and cancer-type classifiers based on theirexpression profiles have been shown to outperform those based on mRNA expres-sion [8]. Owing to these characteristics, single-cell measurements of miRNAs mayoffer high potential for dissecting cell type and origin within heterogeneous popu-lations.To evaluate our device for performing single-cell miRNA profiling, we mea-sured the expression of a panel of miRNAs on two distinct hematopoietic cell lines,K562 and BaF3. K562 is an undifferentiated human erythroleukemic cell line [204]98that has been used extensively as a benchmark for single-cell genomics, and BaF3is a mouse pro-B-cell line [227]. The motivation for using cell lines of relatedlineage, but from two different species, was to assess potential evolutionary con-servation of miRNA regulation pathways. Furthermore, as these cell lines are bothderived from hematopoietic progenitor cells with different lineage potential andare known to spontaneously differentiate, they also provide inherently heteroge-neous populations. Profiling was performed against twenty conserved miRNAs(let-7b-5p, miR-10a-5p, 150-5p, 155-5p, 16-5p, 17-5p, 17-3p, 181a-5p, 200c-3p,20a-5p, 221-3p, 223-3p, 23a-3p, 24-3p, 27a-3p, 29b-3p, 451, 92a-3p, 93-5p, andsnoRNA-142) that were selected to capture this extrinsic and intrinsic heterogene-ity and include miRNAs previously associated with hematopoietic development[111, 168, 172].Figure 3.11A shows results from a single device run in which half of the arraywas loaded with K562 cells and half with BaF3 cells. Loading resulted in a totalof 95 single K562 cells and 81 single BaF3 cells, as determined by bright-field mi-croscopy, for a total single-cell occupancy of 85 %. MicroRNAs were found to beexpressed across a wide dynamic range, with miR-17-5p detected in 100 % of theanalyzed cells at an average of 15,700 (SD = 9,900) copies per cell, and miR-181a-5p detected in 5.7 % of the cells at an average of 1.5 (SD = 6.5) copies per cell.As expected, a substantial proportion (15/20, 75 %) of the analyzed miRNAs werefound to be significantly differentially expressed between the two cell populations(Figure 3.12). However, it is notable that, despite the obvious biological differ-ences between these two cell types, only a minority of the miRNAs (7/20, 35 %)exhibited strict cell-type expression specificity, and that these specific miRNAswere generally those that were weakly expressed (mean expression 17.2, SD 57.8copies per cell).Unsupervised hierarchical clustering of the miRNA expression profiles (Fig-ure 3.11A) unambiguously separated the single cells into their original populations,consistent with the observed differential expression. However, these clustered ex-pression profiles further revealed significant cell-to-cell heterogeneity within eachcell line. For example, miR-221-3p expression was found to be bimodal in K562cells, coinciding with both previous reports that miR-221-3p is down-regulatedduring erythropoiesis [228], and the propensity of K562 cells to undergo sponta-99024024log 10 Expression024024log10 ExpressionmiR-17-5p miR-20a-5p miR-92a-3p miR-223-3pmiR-24-3pmiR-223-3pmiR-92a-3pmiR-20a-5p0 42 0 42 0 42 0 420.95***0.87***0.73***0.64***0.78***0.68***0.38**0.28*0.47***0.26ns0.53***0.11ns0.47***0.62***0.52***0.60***0.40***0.43***0.26ns0.20nsmiR-181a-5pmiR-150-5pmiR-24-3pmiR-155-5plet-7b-5psno142miR-221-3pmiR-29b-3pmiR-23a-3pmiR-200c-3pmiR-451miR-17-3pmiR-27a-3pmiR-16-5pmiR-10a-5pmiR-223-3pmiR-93-5pmiR-92a-3pmiR-20a-5pmiR-17-5pBaF3K562A B0 1 2 3 4log10 ExpressionFigure 3.11: Multiplexed single-cell miRNA expression. (A) Heatmap ofexpression values of 20 miRNAs in each of 95 K562 cells (green) and 81BaF3 cells (purple). Unsupervised hierarchical clustering correctly groupsthe cell types based on their miRNA expression signatures. (B) Scatterplotsof co-expressed miRNAs. miRNAs 17-5p, 20a-5p and 92a-3p are all stronglyco-expressed as they are members of the miR-17∼92 polycistronic cluster.The co-expression of miRNA 24-3p with these three miRNAs demonstrateshow single-cell resolution identifies their positive regulation where bulk mea-surements would predict their negative regulation. Spearman correlation co-efficients and their co-expression significance are denoted for each pair; *p < 0.05, ** p < 0.01, *** p < 0.001, ns not significant.neous erythroid differentiation in culture [229]. Similarly, the expression of miR-155-5p was bimodal in BaF3 cells, and this miRNA has been shown to play a keyrole in B lymphocyte maturation [230]. In contrast, the expression of miR-16-5p was not significantly different between the two populations (absolute medianchange 1.12×, p = 0.074, Wilcoxon rank-sum test), but was very heterogeneouswithin each population (K562: 0 to 28,622 copies per cell, CV of 342 %; BaF3:43 to 2,102 copies per cell, CV = 92 %). miR-16-5p has been shown to be widelyexpressed amongst different tissue types from both humans and rodents [208], and100BaF3 K56201234012340123401234BaF3 K562 BaF3 K562 BaF3 K562 BaF3 K562let-7b-5pp=7.9 × 10-32miR-92a-3pp=5.3 × 10-28sno142p=2.2 × 10-26miR-221-3pp=4.8 × 10-26miR-20a-5pp=1.5 × 10-24miR-223-3pp=1.5 × 10-20miR-17-5pp=3.7 × 10-20miR-24-3pp=9.8 × 10-17miR-155-5pp=3.5 × 10-14miR-93-5pp=5.2 × 10-11miR-150-5pp=3.4 × 10-7miR-10a-5pp=6.9 × 10-7miR-17-3pp=1.1 × 10-5miR-451p=4.8 × 10-4miR-181a-5pp=6.0 × 10-4miR-16-5pp=7.4 × 10-2miR-23a-3pp=2.9 × 10-1miR-200c-3pp=3.0 × 10-1miR-29b-3pp=7.1 × 10-1miR-27a-3pp=9.1 × 10-1log 10 ExpressionFigure 3.12: Differential miRNA expression. Boxplots show differentialmiRNA expression between K562 and BaF3 cells. Plots are sorted in or-der of decreasing significance, from top left to bottom right. Those in thebottom row were not significantly differentially expressed between the twopopulations. P-values were calculated using the Wilcoxon rank-sum test andBenjamini-Hochberg corrected.101is known to regulate genes involved in cell cycle progression [231].Collecting multiple measurements from the same single cell allowed us to usethe observed variability to investigate miRNA co-expression networks. We identi-fied 22 and 35 miRNA pairs that were significantly co-expressed in K562 and BaF3cells, respectively (Table A.1; example miRNA pairs denoted in Figure 3.11B). Asubset of these pairs (14 of 43 unique significant pairs) was found in both popula-tions, suggesting that they share a conserved regulatory framework. The three mosthighly co-expressed miRNA pairs in both populations occurred between miRNAs17-5p, 20a-5p, and 92a-3p, with an average spearman correlation of 0.78± 0.12.These three miRNAs are all members of the miR-17∼92 polycistronic cluster fromwhich they are expressed as a single transcript prior to post-transcriptional pro-cessing into mature miRNAs [232, 233]. Despite their origins as a single primarytranscript, we observed biased expression between individual miR-17∼92 clus-ter members. For example, miR-17-5p was expressed at an average of 7.3 (SD= 1.2) and 10.3 (SD = 2.5) fold higher abundance than miR-20a-5p in K562 andBaF3 cells, respectively, an observation that implies a strong post-transcriptionalregulatory mechanism. Indeed, the establishment of precise ratios between ma-ture members of the miR-17∼92 polycistron has been shown to be due to intrinsiccharacteristics conferred by secondary and tertiary structures of the primary 17∼92transcript [232].Our single-cell expression measurements illustrate the danger in determininggene co-expression using bulk measurements. As described by Simpson’s paradox[234], an apparent trend can often be reversed or even disappear depending on howdata are grouped. For example, if we naively remove the cell-type classificationfrom our measurements of miR-24-3p and miR-92a-3p (Figure 3.11B, lower row),the genes are significantly negatively co-expressed (spearman’s correlation -0.37,p = 7.1×10−5), a conclusion that would also be made using bulk measurements.Exploiting the inherent cell-to-cell heterogeneity within a sample to identify trulyco-regulated genes has been proposed as a solution to this problem [211], and mul-tiplexed single-cell measurements are ideally suited to this task. Indeed, once thecells are correctly grouped by type, these two miRNAs are found to be signifi-cantly positively regulated (K562: spearman’s correlation 0.40, p = 6.1× 10−4;BaF3: spearman’s correlation 0.44, p = 6.0×10−4).102We believe that streamlined technologies for highly multiplexed single cellRT-qPCR are particularly well suited to the investigation of cellular heterogene-ity using miRNA signatures. Our measurements of twenty miRNAs in 176 singlecells from two distinct cell types show clear evidence of well-known cell-to-cellheterogeneity while also recovering the established, evolutionarily conserved co-expression of the miR-17∼92 polycistronic cluster. The broader investigation ofmiRNA expression at the single-cell level is anticipated to lead to exciting devel-opments in miRNA and cell biology.3.4 DiscussionIn this chapter, I presented a microfluidic device for performing highly multi-plexed quantitative PCR on hundreds of single cells. In terms of performance,we have established single-molecule sensitivity, a dynamic range of at least fourorders of magnitude, a measurement precision of 15 %, and high single-cell oc-cupancy rates. Compared to existing multiplexed single-cell RT-qPCR solutions,this device maintains the same advantages seen in integrated devices such as theone presented in Chapter 2, including increased throughput, streamlined workflow,and reduced sample-processing variations and reagent consumption. This multi-plexed device, however, adds the ability to collect dozens of measurements percell, thereby enabling cell profiling studies to dissect subpopulations or identifygene co-expression networks.Implementing this level of integration came at the expense of device complex-ity. In order to maintain cell throughput within the standard device footprint, asignificant increase in feature density was required. Substantial lithographic pro-cess development was required in order to create the high-aspect, high-resolutionfeatures used here. Similarly, successful accommodation of the interlayer connec-tions, reduced valve size, and small tolerances during hand alignment of the flowand control layers all required considerable know-how and ultimately reduced thetotal device yield.The capabilities provided by this device fill an unmet need in the current reper-toire of analytical methods for single-cell analysis by providing rapid measure-ments on panels of genes in hundreds of cells per run. Here we have demonstrated103the quantitative measurement of mRNA and miRNA, but further assay develop-ment has the potential to expand into genotyping, epigenotyping, and combinationsof assay classes. There are two important technical limitations to note. First, whilethe capacity to measure up to 20-40 targets is suitable for profiling well-studiedsystems, it is not enough for discovery purposes. Second, the lack of fluidics torecover reaction products restricts assay types to qPCR or similar fluorescent read-outs.There are important caveats to measuring miRNA expression with quantitativePCR. Based on their design, RT-qPCR assays for miRNA expression do not typ-ically detect mature miRNAs whose 3′ ends differ from the annotated sequence[48] (Figure 1.3). These 3′ end differences are either due to incorrect annotationof the most common form [47, 49] or common end modifications. As an extremeexample, due to its non-canonical biogenesis, the annotated form of miR-451a istypically present at a frequency of less than 10 % of the total reads mapping to thatlocus. While annotations are continually being improved, 3′ variations have beenseen to vary with tissue type [48], and so differences in expression as measured byRT-qPCR may be simply due to differential processing or modification.In the following chapter, I aim to round out the capabilities available for per-forming single-cell miRNA measurements by developing and applying a methodfor the genome-wide sequence-based analysis of miRNA expression in populationsof single cells.104Chapter 4Single-cell microRNA sequencingof the human hematopoietic cellhierarchy14.1 IntroductionNewly available tools for sensitive and scalable single-cell transcriptomics haveenabled an unprecedented description of complex tissues. In addition to identify-ing and characterizing novel and rare cell types [235], measurements of expres-sion heterogeneity at the level of individual cells have been used to infer high-resolution differentiation trajectories [214] and to elucidate co-regulatory gene net-works [236]. To date, these efforts have focused mainly on mRNA profiling, leav-ing associated patterns of small non-coding RNA expression largely unexploredand generally outside of the domain of genome-wide single-cell analyses.MicroRNAs (miRNAs) are a class of small (∼22 nt) non-coding RNA moleculesinvolved in the post-transcriptional modulation of gene expression [23]. The im-portance of their influence has been demonstrated in various processes includingregulating gene expression noise [38], driving and reinforcing cell fate decisions1A version of this chapter has been submitted for publication: Michael VanInsberghe, DavidJ.H.F. Knapp, Michelle Moksa, Hans Zahn, Martin Hirst, Connie J. Eaves, Carl L. Hansen. Single-cell microRNA sequencing of the human hematopoietic cell hierarchy. 2018.105[39] and affecting neoplastic transformation [40]. In addition to these functionalroles, miRNA expression classifiers have also been shown to outperform thosebased on mRNA expression [8], leading to their development as an important classof prognostic and diagnostic biomarkers [42]. It may therefore be expected thatmiRNA analyses of single cells would make a powerful contribution to decipheringthe cellular constituents of complex tissues and their altered properties in diseasestates.miRNA expression has previously been measured in single cells using sev-eral targeted approaches (reviewed in Section 1.2). These include quantitativeRT-PCR [63, 111, 114, 115], hybridization techniques [94], and reporter assays[67, 101, 237]. Such targeted approaches benefit from increased detection sensi-tivity and scalability when appropriate markers are known, but are poorly suitedfor global analysis and discovery. Further, these methods face technical challengesfor detecting or distinguishing common sequence variants known as miRNA iso-forms [55, 238]. Because such isoforms have been previously observed to havedistinct target repertoires and also be regulated in a cell-type specific manner, theirinclusion is a vital aspect of miRNA expression profiling [51]. High-throughputsequencing of miRNAs offers the potential to provide unbiased, genome-widemeasurements at single-base resolution. However, until now, the inefficiencies insmall-RNA library construction methods have limited miRNA-seq to cases wherehundreds of nanograms of purified RNA template could be obtained [64].New methods for the construction of high-quality miRNA sequencing librariesfrom picogram quantities of template without the need for purification are requiredto bring miRNA-seq into the realm of routine single-cell analysis. An importantcontribution towards this goal was recently provided by Faridani et al. [126] withthe description of an improved protocol suitable for analyzing the small-RNA con-tent in 398 single cells. However, in order to obtain representative single-cellmiRNA expression profiles, a very large sequencing effort was required due to li-brary construction inefficiencies, with less than 0.5 % of reads aligned to miRNAs,and libraries dominated by non-specific by-products and RNA degradation inter-mediates (Figure 4.9, Section 4.3.2).In this chapter, I describe miRLin, a robust method to generate small-RNAsequencing libraries from hundreds to thousands of single cells. We first charac-106terized and benchmarked miRLin using K562 cells to demonstrate that the qualityof our single-cell libraries is comparable to those obtained using high-input, bulkmethods. We then applied this technology to enable the production of the firstsingle-cell miRNA profiles of multiple different phenotypically defined hematopoi-etic cell subpopulations present in normal human cord blood and functionally char-acterized in parallel.4.2 Materials and methods4.2.1 ExperimentalHuman cord bloodAnonymized heparinized human umbilical cord blood was obtained from normalfull-term deliveries in accordance with procedures approved by the Research EthicsBoard of the University of British Columbia. Bulk low-density (<1.077 g/mL)cells were obtained by centrifugation using Lymphoprep (STEMCELL Technolo-gies). These cells were viably cryopreserved with DMSO (STEMCELL Tech-nologies) for subsequent pooling. Cells were thawed and pooled from > 500 in-dividual cords to allow multiple experiments to be performed on the same con-sistent cell input and once again cryopreserved. On a given sort day cells werethawed by dropwise addition of Iscove’s Modified Dulbecco’s Medium (STEM-CELL Technologies) supplemented with fetal bovine serum (FBS) and 10 µg/mLDNase I (Sigma-Aldrich). Enrichment for CD34 was then performed using anEasySepTMCD34 positive selection kit. CD34+ cells were then processed for pro-genitor sorts (Figure 4.1). The remaining CD34- cells were retained for sortingmature cell populations (Granulocytes, Monocytes, B cells, T cells, and Erythrob-lasts, Figure 4.2).Cell sortingCells were blocked in Hank’s Balanced Salt Solution (HSBS) supplemented with5 % human serum and 1.5 µg/µL anti-human CD32 antibody (Clone IV.3, STEM-CELL Technologies). Following blocking, CD34 enriched cells were stained with107Pre-B/NKMEPGMPCMPMLPMPPHSCFigure 4.1: Example gating hierarchy for progenitor populations.Sorted populations include Megakaryocyte/Erythroid Progenitors (MEP,CD34+CD38+CD10-CD7- CD45RA-CD135-), Common Myeloid Pro-genitors (CMP, CD34+CD38+CD10-CD7-CD45RA+CD135-), Gran-ulocyte/Macrophage Progenitors (GMP, CD34+CD38+CD10-CD7-CD45RA+CD135+), Multi-Lymphoid Progenitor (MLP; CD34+CD38-CD45RA-CD10+CD7+/-), Multi-Potent Progenitors (MPP, CD34+CD38-CD45RA-CD90-CD49f-), and hematopoietic stem cells (HSC/49f,CD34+CD38-CD45RA-CD90+CD49f+).108BT ErythroidMonocyteGranulocyteFigure 4.2: Example gating hierarchy for mature cell populations. Sortedpopulations include total CD3+ T cells, CD19+ B cells, Erythroblasts(CD3-CD19-CD33-GPA+), Monocytes (CD3-CD19-CD33+CD14+CD15-),and Granulocytes (CD3-CD19-CD33+CD14-CD15+).CD34 Alexa Fluor-700 (581), CD90 PECy7 (5E10), CD10 (HI10a), CD7 FITC(M-T701; all from BD Biosciences), CD49f eF450 (GoH3), CD38 eF650 (HB7),CD45RA eF605NC (HI100), CD135 PE (BV10A4H2; all from eBiosciences) for1 h in the dark on ice. CD34- cells were stained with CD33 APC (WM53), CD3eF605 or FITC (OKT3), CD19 PE (SJ25C1, all from eBiosciences), CD34 AlexaFluor-700 (581, BD Biosciences) or PerCP-eF710 (4H11, eBiosciences), CD15Horizon V500 (HI98, BD Biosciences), CD14 PECy7 (MoP9, STEMCELL Tech-nologies), Glycophorin A (GPA) (HI264, Biolegend) again for 1 h in the dark onice. Following staining, cells were washed and re-suspended in HBSS supple-mented with 2 % FBS and 1 µg/mL propidium iodide. Cells were then sorted 4ways on a BD FACSAria III or Fusion cell sorter on purity mode. Gates for progen-itor and CD34- populations are shown in Figure 4.1 and Figure 4.2, respectively.109Cell culture and RNA purificationK562 cells were obtained from ATCC (CCL243) and were cultured in DMEM(Gibco) supplemented with 10 % FBS (Gibco) and 1× GlutaMAX (Gibco).Purified total RNA was extracted from K562 cells using the mirVana RNAisolation kit (Ambion) according to manufacturer directions, omitting the optionalsize-enrichment step.Device fabricationMicrofluidic devices were fabricated using multilayer soft lithography [149]. De-vices were designed using CAD software (AutoCAD, AutoDesk Inc.), and printedto 5-inch chrome photomasks (HTA Enterprises) using a LW405 mask writer (Micro-tech) with 1-µm (cell trap layers) or 8-µm (all other layers) resolution. Photomaskswere processed according to manufacturer directions. The “flow” moulds were fab-ricated in three photolithographic steps. Two versions of this mould were fabricated(“large” and “small”) with different cell trap geometries designed to accommodatea variety of cell sizes. Small cell-trap features were fabricated using a 7-µm SU-85 (MicroChem) layer and large cell trap features were fabricated using a 15-µmSU-8 2010 (MicroChem) layer. Next, an 11-µm layer of AZ 50XT (AZ ElectronicMaterials) was deposited to form the channels connecting chambers and reagentdelivery and recovery channels. These channels were rounded to facilitate valveclosure by ramping the processed wafer from 65 ◦C to 190 ◦C, holding for 4 h, andslowly ramping to 25 ◦C; the channel height after rounding was 13 µm. Finally, thechamber and reagent delivery and recovery channels were fabricated using a 60-µmlayer of SU-8 50 (MicroChem). The “control” mould was fabricated using a singlelayer of 25-µm high SU8 2025. All photoresist processing was performed accord-ing to manufacturer specifications. All moulds were fabricated on 4-inch siliconwafers (Silicon Quest International). Parylene-C was deposited on all moulds afterfabrication to facilitate elastomer release [222].Microfluidic devices consisted of three layers of polydimethylsiloxane (PDMS,RTV615, Momentive), with a blank bottom layer, and a middle layer containing thecontrol channels that act as valves by deflecting up into the above flow layer. Theflow layer was made by casting PDMS mixed at a 5:1 ratio (5 parts RTV615A to1101 part RTV615B) on the “flow” mould and baking at 80 ◦C for 30 min. The thincontrol layer was made by spin-coating PDMS mixed at a 20:1 ratio at 1800 rpmon to the “control” mould and immediately baking at 80 ◦C for 20 min. After bak-ing, both the flow and control layers were allowed to cool to room temperature.The flow layer was then peeled off of the mould and aligned to the spin-coated,baked control layer. Next, this structure was baked at 80 ◦C for 30 min to allow thetwo layers to bond. After bonding, the combined slab was peeled off of the controlmould and access ports were punched using a coring tool (CR0350255N20R4 forstandard access ports or CR0890735N1354 for large ports, Syneo). Interlayer con-nections (vias) were then laser ablated using a custom instrument [223]. The blanklayer was prepared by spin-coating 20:1 PDMS at 1800 rpm on a blank waferand baking this at 80 ◦C for 20 min. The bonded, punched, and ablated control-flow slab was then placed onto the baked blank layer and allowed to bond at 80 ◦Covernight. Finally, the combined 3-layer slab was peeled off the blank wafer, diced,and bonded to glass slides using oxygen plasma. These mounted, finished chipswere cured at 80 ◦C overnight prior to use.Device operationMicrofluidic valves were actuated using 32 pneumatic valves that were semi-auto-matically controlled using solenoid actuators (Fluidigm Corp. and Festo) con-nected to a digital input-output card (NI PCI-DIO-32HS or NI PCI-6512) con-trolled using custom LabView (National Instruments) software. Tygon and Teflontubing connected the solenoids to the microfluidic device through 20-gauge hol-low stainless steel pins (Small Parts Inc.) fitted into the control ports. Krytox102 (DuPont) oil was used as the fluid in the control lines and Fluorinert FC-40(3M) was used in the pump control lines. Valves were operated at a pressure of45 PSI, reagents were injected using 5 PSI, and cells were loaded at 1 to 2 PSI. Themicrofluidic peristaltic pump was operated with a 550 ms period. All incubationswere performed on a flatbed thermocycler block (Bio-Rad DNA Engine PTC-200;MJ Research/Bio-Rad) with light mineral oil (Fisher Scientific) added between theblock and the glass slide.Device operation began with priming the cell-loading channel to help prevent111cell adhesion. Cells were then loaded, washed and isolated (Figure 4.3C). A lysisbuffer was then flowed over the cells into the 3-nL cell lysis chamber (Figure 4.3D).Following lysis, the 3′ ligation brew was used to push the cell lysate into the adja-cent 3′ ligation chamber (Figure 4.3E) and was actively mixed with the lysate bypumping it around the reaction chamber using the peristaltic pump (Figure 4.3F).The entire device was then placed on a flatbed thermocycler for temperature con-trol during the incubation. The 5′ ligation brew was then loaded into the compoundreaction chamber (Figure 4.3G) and again actively mixed and placed on a thermo-cycler. These same general steps were repeated for cDNA synthesis (Figure 4.3H)and PCR (Figure 4.3I), except that a unique PCR brew is delivered to each reaction.Following PCR, reaction products from each cell-processing unit were pooled andeluted (Figure 4.3J) for off-chip quality control, size selection, quantification, andsequencing.Spike-in controlSpike-in sequences were selected such that they did not align to either the human(GRCh37) or mouse (GRCm38) reference genomes. Random 22-base long se-quences were generated and discarded if they either had homopolymer regionslonger than 2 bases, aligned to the reference genomes with up to a maximumedit distance of 4 (bwa aln -n 4, BWA version 0.7.5a-r405) or produced a hitwhen in a nucleotide-nucleotide BLAST against the genome (blastn -taskblastn-short, blastn version 2.2.27+). Ten sequences meeting these crite-ria were randomly selected and synthesized as 5′ phosphorylated RNA oligonu-cleotides by Integrated DNA Technologies (IDT). Sequences were resuspended to100 µM in TE pH 7.0 (Ambion) based on the manufacturer quantification, and di-luted to the appropriate final concentration in TE pH 7.0, 0.1 % Tween-20 beforepooling and aliquoting. All experiments were done using single-use aliquots fromthe same pool that had been stored at −80 ◦C. Each aliquot was serially-diluted to1/10,000× into the lysis buffer. The final concentration of each spike-in was 5000,500, 50, 5, 0.5, 1000, 100, 10, 1, 0.1 molecules/reaction for RmiR01 to RmiR10,respectively (Table A.2 for sequences).112ABCell lysis3’ Ligation brew injection5’ Ligation brew injection3’ Ligation brew active mixing with lysateRT brew injectionIndependent PCR brew injectionPCR product pooling and recoveryCell capture and washCDEFGHIJWash Buffer Lysis Buffer 3’ Ligation BrewClosed ValveOpen ValveEmptyPCR BrewRT Brew5’ Ligation Brew5 mmLysis 3’ Ligation 5’ Ligation RT PCR Elution500 μm3 nL lysis 7 nL 3’10 nL total4 nL 5’14 nL total16 nL RT30 nL total70 nL PCR100 nL total113Figure 4.3 (previous page): Microfluidic device schematic and operation. (A)Schematic of microfluidic device for generating 96 single cell small-RNA se-quencing libraries. Channels on the “flow” layer are shown in blue, those onthe “control” layer are denoted in red, and laser-ablated layer interconnectsare shown as orange circles. Scale bar 5 mm. Black box denotes a singlecell-processing unit magnified in B. (B) One cell-processing unit. Compo-nents include a reagent injection channel; a 0.64-nL cell capture chamber; acompound reaction chamber comprising a 3-nL lysis, 7-nL 3′ ligation, 4-nL5′ ligation, 16-nL reverse transcription, and 70-nL PCR chambers; an inde-pendent PCR inlet; and recovery channels. (C-J) Schematic illustration ofdevice operation. (C) A cell suspension is flowed down the cell-loading chan-nel, isolating single cells from the carrier fluid in cell traps. The isolated cellsare then washed, removing untrapped cells and extracellular debris and RNA.(D) Lysis buffer is flowed through the reagent injection channel, over eachcell trap and into the 3-nL lysis chamber. The entire device is then placed ona flatbed thermocycler for temperature control. (E) 3′ ligation brew is thenused to push the cell lysate into the adjacent 3′ ligation chamber, and (F) anintegrated microfluidic peristaltic pump is used to actively mix the lysate withthe reaction brew prior to device incubation on a flatbed thermocycler. Thesesame steps of reagent injection, mixing, and incubation are repeated for (G) 5′ligation, (H) cDNA synthesis, and (I) indexing PCR. After PCR, the librariesare (J) pooled and eluted from the device.Library constructionLibrary construction began with priming the cell-loading channel with 10 mg/mLUltraPure bovine serum albumin (BSA) (Ambion) in phosphate buffered saline(PBS, Gibco). Cells were then loaded in the device by applying 1 to 2 PSI to a gelloading pipet tip containing a single cell suspension (105 to 106 cells/mL) in media.Cells flowed down the cell loading channel, where lithographically defined celltraps isolated and positioned cells. After injecting the cell suspension, the cell washbuffer (1 mg/mL BSA in PBS) was then flowed down the cell-loading channel toremove untrapped cells, extracellular RNA and debris. The cell-isolation valveswere then closed to partition the cell-loading channel and isolate individual cellreaction chambers. Each cell trap was then visually inspected and the number oftrapped cells was recorded.114Lysis buffer [1.67 µM miRNA3PE 3A (IDT), 533 mM guanidium thiocyanite(GuSCN, Sigma Aldrich), 0.2 % Tween-20 (Sigma Aldrich), 2 U/µL RNaseINPlus (Promega), 1×10−5 dilution of spike-in control, in UltraPure water (Gibco)]was then flowed from the common reagent delivery bus over the cell traps into the3-nL lysis chamber. Excess lysis buffer was washed out of the device by flow-ing 100 µL reagent wash solution (0.1 % Tween-20 in UltraPure water) though thereagent injection lines. Next, the device was incubated at 25 ◦C for 2 min, then70 ◦C for 2 min, and held at 4 ◦C. Reagent injection lines were again washed with100 µL of reagent wash solution then dried with compressed air.The 3′ ligation was next performed by adding the 3′ ligation brew [1× T4 RNALigase Reaction Buffer (New England BioLabs), 20 % PEG-8000 (New EnglandBioLabs), 0.07 % Tween-20 (Sigma Aldrich), 1.72 U/µL RNaseIN Plus (Promega),30 U/µL T4 RNA Ligase 2 Truncated R55K K227Q (T4RNL2 tr KQ, New Eng-land BioLabs)] to the compound reaction chamber, dead-end filling the 7-nL 3′ligation chamber. Excess 3′ ligation brew was washed out of the device by flowing100 µL of reagent wash solution through the reagent injection lines. The devicewas then incubated at 4 ◦C for 18 h, during which the reaction brew was activelymixed with lysate using 6000 cycles of the microfluidic peristaltic pump (approxi-mately 1 h). After incubation, the reagent injection lines were washed with 100 µLof reagent wash solution then dried with compressed air.The 5′ ligation was next performed by adding the 5′ ligation brew [19.5 %PEG-8000 (New England BioLabs), 0.1 % Tween-20 (Sigma Aldrich), 500 nMmiRNA3PE 5A (IDT. 5′ adapter was denatured at 70 ◦C for 2 min then immedi-ately placed on ice for 2 min before adding to the reaction brew), 2.5 mM ATP(New England BioLabs), 1.25 U/µL T4 RNA Ligase 1 (Ambion)] to the compoundreaction chamber, dead-end filling the 4-nL 5′ ligation chamber. Excess reactionbrew was washed out of the device by flowing 100 µL of reagent wash solutionthrough the reagent injection lines. The device was then incubated at 10 ◦C for 1 h,during which the reaction brew was actively mixed with lysate using 6000 cyclesof the microfluidic peristaltic pump. After reagent mixing, the device was incu-bated at 37 ◦C for 2 h, then 65 ◦C for 10 min, then held at 4 ◦C. After incubations,the reagent injection lines were again washed with 100 µL of reagent wash solutionthen dried with compressed air.115cDNA synthesis was next performed by adding the reverse transcription brew[1.875×Maxima RT Buffer (ThermoFisher Scientific), 0.188 % Tween-20 (SigmaAldrich), 2.5 µM miRNA3PE P2 (IDT), 9.375 µM miRNA3PE RTBlock (Exiqon),1.25 mM dNTPs (New England BioLabs), 1.875 U/µL RNaseIN Plus (Promega),9.375 U/µL Maxima H Minus Reverse Transcriptase (ThermoFisher Scientific)] tothe compound reaction chamber, dead-end filling the 16-nL RT chamber. Excessreaction brew was washed out of the device by flowing 100 µL reagent wash so-lution through the reagent injection lines. The device was then incubated at 10 ◦Cfor 1 h, during which the reaction brew was actively mixed with lysate using 6000cycles of the microfluidic peristaltic pump. After reagent mixing, the device wasincubated at 50 ◦C for 1 h, then 85 ◦C for 5 min, then held at 4 ◦C. After these incu-bations, the reagent injection lines were again washed with 100 µL of reagent washsolution.Indexing PCR reaction brews, each with a unique index, [1.43× Phusion HFBuffer (New England BioLabs), 0.14 % Tween-20 (Sigma Aldrich), 0.36 mM dNTPs(New England BioLabs), 4.29 % DMSO (Sigma Aldrich), 1.79 µM miRNA3P3 P2(IDT), 1.79 µM miRNA3PE iPCRNN (IDT, NN denotes index number) 0.03 U/µLPhusion Hot Start Flex DNA Polymerase (New England BioLabs)] were pipettedinto the appropriate PCR reagent inlet and injected to the reaction chamber usinga custom-built chip loader. After dead-end filling the PCR reaction chambers, thedevice was incubated at 25 ◦C for 2 h and 50 min while reagents were mixed with18,000 cycles of the microfluidic peristaltic pump. The device was then incubatedat 95 ◦C for 5 min, followed by 16 cycles of 98 ◦C for 20 s, 62 ◦C for 30 s, 72 ◦C for15 s, and then a final 5 min incubation at 72 ◦C and finally holding at 4 ◦C.Following PCR amplification, reaction products from each cell-processing unitwere simultaneously pooled and eluted by flowing recovery buffer [TE buffer pH8.0 (Ambion) with 0.1 % Tween-20] through the elution channels. Recovered prod-ucts were brought up to a final 20-µL volume (per half-chip) with recovery bufferprior to quality-control using the High-Sensitivity DNA Bioanalyzer (Agilent) as-say.Library pools were then run on a 12 % polyacrylamide gel at 200 V for 6 h, andthe 10-base-pair region from 110 bp to 120 bp was excised. The band was thencrushed and soaked in elution buffer [5:1 LoTE (3 mM Tris-HCl pH 8.0, 0.2 mM116EDTA pH 8.0):7.5 M ammonium acetate (Sigma Aldrich)] overnight at 4 ◦C, thenfor 1 h at 65 ◦C. Eluate was then precipitated in ethanol.Library concentration was quantified using custom digital PCR chips contain-ing 12 arrays of 765 2.1-nL reaction chambers [239]. Libraries were seriallydiluted in EBT buffer [buffer EB (Qiagen), 0.1 % Tween-20] to total dilutionsof 1/400,000×, 1/2,000,000×, and 1/10,000,000×. PCR reactions were pre-pared in triplicate consisting of 1× SsoFast EvaGreen Supermix (Bio-Rad), 0.1 %Tween-20 (Sigma Aldrich), 500 nM of each miRNA3PE QF and miRNA3PE QR,and 0.375× of each library dilution (e.g. 3.75 µL diluted library into a 10 µL finalvolume). Reaction brews were dead-end filled into the device at 5 PSI, then sepa-ration valves filled with Krytox 102 (DuPont) were closed at 50 PSI. Devices werethermocycled and imaged on a prototype version of the BioMark (Fluidigm) for95 ◦C for 30 s, then 35 cycles of 95 ◦C for 15 s, 60 ◦C for 30 s. Images were takenafter every cycle, and those from cycle 35 were manually counted and the digitalarray response curve [106] was used to calculate a final concentration.Oligonucleotide sequences used for library construction are listed in Table A.2.Library sequencingLibraries were sequenced using V2 chemistry on a MiSeq (Illumina), with 36cycles for read 1 and 7 cycles for the index read. A custom sequencing primermiRNA3PE SP (Exiqon) was used for read 1 and the following custom sequenc-ing recipe (Figure 4.4) was used to read the single index off of the P5 end.4.2.2 Data analysisSequence read processingThe read trimming and annotation pipeline developed to analyze miRNA sequencedata for The Cancer Genome Atlas [122] was used to trim, annotate, and quan-tify sequencing data. Briefly, adapter sequences were first trimmed, and any readsless than 15 bases were discarded. Next, trimmed reads were aligned with bwa(version 0.7.12-r1039) [240] aln and samse to build 38 of the human genome.Aligned reads were then annotated using the miRBase v21a [21] (2014-06-25) and117<?xml version="1.0"?><Protocol Version="Fraise 0.7.2 Amplicon v1.1"><ChemistryRef ChemistryName="OnBoardClusterGeneration" /><ChemistryRef ChemistryName="FirstReadPreparation" /><ReadRef ReadName="FirstRead" /><ChemistryRef ChemistryName="EndDeblock" /><ChemistryRef ChemistryName="Deprotection" /><ChemistryRef ChemistryName="Index2FirstBaseDark" /><ChemistryRef ChemistryName="Index2CompleteCycleDark2" /><ChemistryRef ChemistryName="Index2CompleteCycleDark2" /><ChemistryRef ChemistryName="Index2CompleteCycleDark2" /><ChemistryRef ChemistryName="Index2CompleteCycleDark2" /><ChemistryRef ChemistryName="Index2CompleteCycleDark2" /><ChemistryRef ChemistryName="Index2CompleteCycleDark2" /><ReadRef ReadName="IndexRead2" /><ChemistryRef ChemistryName="EndDeblock" /><ChemistryRef ChemistryName="TemplateRinse" /><ChemistryRef ChemistryName="End" /></Protocol>Figure 4.4: Custom MiSeq sequencing recipe to sequence miRLin miRNAlibraries.UCSC refGene (2015-04-20), knownGene (2014-05-18), kgX (2014-05-18), andrmsk (2014-03-06) tables. Finally, reads were again aligned to the spike-in con-trols. The read annotations were subsequently used to output alignment statisticsand read count expression matrices.The annotation pipeline was modified in order to extract the 5′ and 3′ relativecut locations and any non-templated base additions. The location of the first, orlast, base that matched the reference sequence was taken as the cut location atthe 5′ and 3′ end, respectively. Any additional nucleotides were recorded as non-templated base additions. The quantification output was modified to output readcounts for mature miRNAs instead of precursor miRNAs.Data have been made available in the Gene Expression Omnibus under theaccession number GSE104433.Statistics and visualizationData are reported as mean± standard deviation. Manipulations and statistical test-ing on annotation and expression outputs were performed using R version 3.4.0,with Rcpp (v0.12.10), plyr (v1.8.4) and reshape (v0.8.6). P-value corrections formultiple testing were computed using the Bonferroni correction where relevant.Plots were generated using ggplot2 (v2.2.1), cowplot (v0.7.0), gplots (v3.0.1), cor-118rgram (v1.12), and RColorBrewer (v1.1-2). Values for boxplots were calculatedusing the default settings in geom boxplot (ggplot2): the middle line representsthe median, the lower and upper hinges correspond to the first and third quartiles,the upper whisker extends from the upper hinge to the largest value no further than1.5 times the interquartile range, the lower whisker extends from the lower hingeto the smallest value no further than 1.5 times the interquartile range, and data be-yond the whiskers are deemed outliers and plotted individually. Details for specificsections are reported below.Alignment metricsAnnotation classes were combined to succinctly summarize data. Total miRNA in-cluded mature, precursor, stemloop and crossmapped miRNA annotations. Othersmall RNA included small nuclear RNA (snRNA), transfer RNA (tRNA), signalrecognition particle RNA (srpRNA), small nucleolar RNA (snoRNA) and smallconditional RNA (scRNA). RNA included coding exon, 5′ UTR, 3′ UTR, NoCDS and intron classes. Repeats comprised RepeatMasker unknown, simple re-peat, DNA, low complexity, other, and RNA, and LINE, SINE, LTR, and Satelliteclasses.Library quality controlWe observed that one of the cells, denoted with an asterisk in Figure 4.6B, sig-nificantly underperformed in library quality metrics; we believe that the cell wasdead or had poor viability that was not detected during inspection by bright fieldmicroscopy. This sample, and similar clear outliers in subsequent libraries, wereexcluded from further analyses.Library diversityPreviously published miRNA-seq datasets generated from K562 cells (SRR039-192 [241], ENCFF001RDL and ENCFF001RDM [1], SRR907615 [242]) weredownloaded, adapter-trimmed to a maximum read length of 36 bases, then runthrough the alignment and annotation pipeline.The single-cell library diversity was compared to these bulk libraries that were119down-sampled to match sequencing efforts. SAM files were randomly down-sampled with samtools view -s (samtools version 1.2) to match the totalreads minus the number mapping to the spike-ins for each single-cell K562 library(N = 87).The diversity of these previously published libraries was also compared to a cu-mulative, bulk-equivalent K562 library containing the single cells passing qualitycontrol. This cumulative library was down sampled for comparisons to ENCFF0-01RDL and SRR039192. SRR907615 was down sampled for comparisons to ourbulk equivalent library.The average number of miRNAs with a coverage of at least 10 reads was re-ported.Conversion efficiencyFour libraries were generated using 1.66× 10−19 moles (100,000 copies per 3 nLlysis volume) of a 22-base synthetic RNA randomer (oligonucleotides RmiRN)and the spike-in control as the starting template. These libraries were sequencedto an average depth of 4.7× 105± 5.6× 104 reads per sample. Adapter-trimmedreads were first aligned to the spike-in control. Reads not aligning to the spike-inwere clustered using CrunchClust (version 43) [243] using a Levenshtein distancecut-off of 2 and counting all differences and end gaps (--diff 2 --d all--endgaps). Consensus sequences for each cluster were then generated us-ing Clustal Omega (version 1.2.0) [244]. The consensus sequence and numberof reads from each cluster was output for further analyses. Consensus sequenceswere counted as detected if they were between 17 and 27 bases in length and hadbetween 5 and 150 supporting reads. A histogram of the read counts per sequenceis presented in Figure 4.8A. A no-template control (NTC) prepared in parallel con-tained an average of 1.8 % of the unmapped reads as those containing the randomsequence, demonstrating a low background contamination.We recovered an average of 6498± 979 unique sequences per library, corre-sponding to a conversion efficiency per single molecule of 6.50±0.98 %.The Poisson distribution was used to calculate the detection sensitivity for a120given efficiency. The probability of a reaction being positive is given byp = 1− e−xE (4.1)where p is the detection probability, x is the number of molecules, and E is theconversion efficiency. At a detection confidence of 95 %, our efficiency of 6.5 %corresponds to a detection sensitivity of 46 molecules.Expression analysisRead counts were normalized and converted to a copy number per cell by dividingthe expression count by the library-averaged read counts per molecule for RmiR-01, RmiR-02, and RmiR-06. These normalized expression values were used for allsubsequent analysis except in determining significant changes in isoform ratios. T-distributed stochastic neighbour embedding (t-SNE) [245] was computed with theRtsne (version 0.13) R package on the set of genes containing up to the first withthe following parameters: perplexity=15, pca=FALSE, theta=0.0,max iter=1000. Spatial density clustering was performed with the dbscan (ver-sion 1.1-1) package with an epsilon neighbourhood of 8 and a minimum of 8 pointsper cluster.miRNAs that were significantly differentially expressed between the sorted cellpopulations were identified using a Kruskal-Wallis rank sum test. miRNAs thatwere expressed higher than 20 copies per cell in at least 20 cells, and had a meanexpression value in expressing cells greater than 20 copies were initially selectedfor analysis. Corrected p-values less than 0.01 were deemed significant. Hierar-chical clustering was performed on log-scaled expression values using (1-pearsoncorrelation)/2 as a distance metric and ward.D2 linkages. Clusters were determinedusing dynamicTreeCut (v1.63-1, with the hybrid method; method="hybrid",deepSplit=4, pamRespectsDendro=TRUE, minClusterSize=4).Independent component analysis (ICA) was performed in order to identify themiRNAs most influential in cell-type discrimination. ICA decomposition was per-formed using the information-maximization approach [246] from the ica package(version 1.0-1) with the following parameters: center=TRUE, maxit=1e5,tol=1E-15, alg="gradient", rate=0.1. Eight components were ex-121tracted, as this was the minimum number of components that yielded distinct sep-arations for each mature cell type.Pseudotime orderingHematopoietic stem and progenitor cells were ordered along a pseudotime differ-entiation trajectory according to their position along the major axis of the tSNE-extracted progenitor cluster (Figure 4.12A). To ensure that this pseudotime order-ing was representative, we also ordered the progenitor cells using five differentstrategies utilizing four different techniques for dimensionality reduction (ICA, dif-fusion map, Wanderlust, and local linear embedding) with four different trajectorymodelling steps (linear projection, minimum-spanning tree, principal curve fitting,and Wanderlust) (Figure 4.12, Figure 4.13).Independent components were extracted as described above, and cell positionwas determined by projection onto the HSC component, and using the minimumspanning tree approach from Monocle [214].A diffusion map was calculated using destiny (version 1.0.0 with Euclideandistance, 3 eigenvectors/values and a sigma of 3.758; sigma = 3.758374,distance = "euclidean", n eigs = 3, n local = 3). A princi-pal curve was then fit to the first and third components using princurve (version1.1-12 with a threshold of 1×10−6 and up to 1000 iterations; thresh = 1E-6,maxit=1000). The start of the trajectory was determined manually based on theenrichment of sort phenotypes of the cells.As Wanderlust [247] does not support branched paths, we computed a separateordering for the progenitors alone, and the progenitors with each of a single maturecell type (e.g., progenitors and monocytes, etc.). As there was a high correlation be-tween the progenitor cell’s locations along each trajectory, they were each normal-ized between 0 and 1 and then averaged to produce the final ordering. Wanderlust(CYT version 2) was run in MATLAB (version 2014b, MathWorks) with no inter-nal normalization and the following parameters: metric=’cosine’, k=5,l=30, num graphs=50, num landmarks=20, verbose=TRUE. One ofthe 49f cells was randomly selected as the starting cell for each graph.Local linear embedding dimensionality reduction was computed using the lle122package (version 1.1 with two components and 10 neighbours; m=2, k=10). Aprincipal curve was then fit to the resulting components as described above.Isoform analysisBased on a preliminary examination of modification frequencies (Figure 4.17A),we restricted downstream analyses to only consider deviations of up to two basesup or downstream of the annotated cut site, and 3′ non-templated additions of singleadenines, uridines, double adenines, and an adenine and a uridine.Global modification profiles were calculated based on the cumulative propor-tion of each relative modification for each cell; miRNA-specific modification pro-files were calculated based on the proportion of each modification for each miRNAin each cell. In order to identify isoforms at functionally relevant expression levels,miRNAs with potential isoforms were only considered for further analyses if theyhad at least one isoform with a cumulative modification frequency greater than 5 %,a mean expression greater than 10 molecules in more than 10 cells, and a minimumcumulative expression greater than 1000 molecules.The significance of changes to the relative isoform abundance was assessedbased on the approach used in Cuffdiff v2.2.1 [248]. Briefly, the Jensen-Shannondistance (JSD; the square root of the Jensen-Shannon divergence) between the av-erage modification profiles from two populations of cells was calculated. Next, anempirical null distribution was generated for each population by randomly sam-pling, with replacement, the cell-specific modification profiles used to calculate apopulation average and subsequent JSD. The resulting JSD distributions were com-bined and a p-value was estimated based on the one-sided tail probability. This pro-cedure was repeated for each miRNA for all 66 pairwise population combinations,except in cases where a given miRNA was expressed at less than 10 copies/cell inless than 10 cells from a population. miRNAs with at least one significant pairwisedifference were shown in Figure 4.20, Figure 4.21, and Figure 4.24. Significancewas assessed at a corrected p-value less than 0.01.123Target PredictionMicroRNA targets for the 5′ isoforms were predicted using both TargetScan [249]and miRanda [250]. TargetScan (version 7.0) was run using the default parametersagainst the UTR sequences from the 84-way alignments provided with TargetScan.Transcript hits were only considered if they were from human transcripts and ei-ther the 7mer-1a, 7mer-m8, or 8mer-1a classes. The degree of conservation wasnot considered, as the complete scripts required to assess conservation were notavailable. miRanda (version 3.3) was run using the default parameters against thesame human UTR sequences used for TargetScan predictions. The degree of targetoverlap between 5′ isoforms and their official sequence was compared to overlapdistributions generated using the reshuffled pairs of 5′ isoforms and their annotatedform, and to 104 randomly selected pairs from genome-wide predictions.Comparison of performance to existing methods for miRNA-seqSequencing and alignment metrics from five data sets were used to benchmarkthe performance of miRLin. The libraries presented in this work were segregatedinto three categories: 1) 87 single K562 cells to represent a cell line (referredto as “K562”), 2) 2057 single primary hematopoietic cells isolated from twelvedifferent phenotypic fractions (referred to as “hema”), and 3) 211 of the primaryhematopoietic cells that had the library fraction greater than∼28 bp analyzed (i.e.,the fraction longer than the 18-28 base region used to enrich for miRNAs, referredto as “hema L”). These sets were compared to the 398 single-cell small-RNA li-braries generated by Faridani et al. (referred to as Faridani or smRNA-seq) [126],and the 69 human miRNA-seq libraries generated by the ENCODE consortiumfrom bulk purified total RNA [1]. In order to enable meaningful comparisons, allreads mapping to our spike-in controls were not used in the discard and alignmentanalyses.Raw sequence reads and UMI expression counts from Faridani et al. weredownloaded from the GEO repository (GSE81287). The first 10 bases of eachread, which contained the UMI sequence, were first removed, and the adapterswere trimmed, aligned, annotated, and quantified using our analysis pipeline. Li-braries were filtered to include only those that were from single cells and were124generated using the smRNA-seq protocol.Raw sequence reads were downloaded from the ENCODE repository (acces-sions: ENCFF035NXR, ENCFF051JIL, ENCFF073RGX, ENCFF079FVD, ENC-FF087EUE, ENCFF088PTQ, ENCFF093BVP, ENCFF094PWC, ENCFF096DFL,ENCFF102CYC, ENCFF122GDK, ENCFF130ZBC, ENCFF149UPH, ENCFF1-51VUG, ENCFF247RWG, ENCFF262GGJ, ENCFF288GGM, ENCFF327IXJ, E-NCFF344XLP, ENCFF349IAF, ENCFF372LLN, ENCFF391UNV, ENCFF437V-TZ, ENCFF456WPB, ENCFF468NDE, ENCFF470IXT, ENCFF470NHA, ENC-FF486DIS, ENCFF504YKU, ENCFF505MBD, ENCFF515ENF, ENCFF519OSJ,ENCFF522BMH, ENCFF571MWI, ENCFF586YSA, ENCFF588AZB, ENCFF6-28ABP, ENCFF629XUN, ENCFF632LTU, ENCFF668WDZ, ENCFF681CKF, E-NCFF683ZDJ, ENCFF689CZI, ENCFF692VQN, ENCFF704RNF, ENCFF707G-SP, ENCFF712HVP, ENCFF731SJZ, ENCFF753VAQ, ENCFF764UDR, ENC-FF777XPZ, ENCFF779GVO, ENCFF785RWT, ENCFF799SRR, ENCFF802BSL,ENCFF820BWI, ENCFF824HJD, ENCFF852VBU, ENCFF881FED, ENCFF8-95NVP, ENCFF902FJY, ENCFF925KAH, ENCFF986EDF, ENCFF987SHZ, ENC-FF988BHN, ENCFF001RDJ, ENCFF001RDK, ENCFF001RDL, ENCFF001RD-M). For reads from all libraries except ENCFF001RDJ, ENCFF001RDK, ENC-FF001RDL, and ENCFF001RDM, the first 6 bases were removed as they con-tained adapter sequences. Reads were then adapter-trimmed, aligned, annotated,and quantified using our analysis pipeline. Libraries ENCFF111WIC and ENC-FF703USS were not included in further analyses as the majority of reads did notalign.4.3 Results4.3.1 Library construction and performanceA microfluidic device was designed to integrate single-cell isolation, library con-struction, and recovery. Each device was configured with four independent sam-ple inlets, each connected to 24 cell processing units enabling a total of 96 re-actions per chip (Figure 4.3A). The core fluidic architecture is based off of thatpresented in Chapter 2, and is similar to previously published and commercially125available designs [115, 145]. Each unit (Figure 4.5A, Figure 4.3B) contains a cell-capture chamber connected to a compound reaction chamber, enabling sequentialvolume expansions and efficient mixing after each step (Figure 4.3C-J). The cell-processing units are connected to both a shared feed channel to deliver commonmaster mixes, and an independent inlet to deliver PCR brews containing uniqueindexing primers. After the molecular steps are completed, the valves are actuatedto serially connect the cell-processing units on each half of the chip to enable pool-ing and elution of reaction products for off-chip size selection and sequencing. Adetailed description of device operation is provided in Section 4.2.1 and Figure 4.3.Small-RNA sequencing libraries were constructed using a one-pot, strand-specific protocol based on that developed to sequence miRNAs for The CancerGenome Atlas [122]. As the sample input requirements of this and similar proto-cols are substantially different than what is attainable from single cells, we exten-sively optimized reaction conditions in order to tolerate both total cell lysate andsignificantly reduced input amounts. Library construction progressed through celllysis, then two successive single-stranded adapter ligations that added priming sitesfor subsequent cDNA synthesis and indexing PCR amplification (Figure 4.5B). Af-ter amplification, samples were pooled and size selected to enrich for ∼22-bp in-sert lengths characteristic of mature miRNAs, and reduce adapter-dimers and otherclasses of small RNAs.In order to assess the performance of miRLin, we generated and sequencedlibraries from individual K562 cells, a human BCR/ABL-positive cell line derivedfrom a patient with chronic myeloid leukemia in blast crisis [204]. After loadingthe cells into a device, 88 of 96 cell traps were visually confirmed to contain asingle cell using bright field microscopy, 4 contained debris, and the remaining 4were no-cell controls (NCC).A variety of sequencing metrics were assessed in order to determine libraryquality. Libraries prepared from single cells generated an average of 19× 104±5.1× 104 reads per cell, and the NCCs produced 2.5× 104± 1.1× 104. Averag-ing across all libraries, 1.4±2.7 % of all reads were adapter-dimers and 97±1 %aligned to either the human genome or the spike-in control. As expected, a higherproportion of the aligned and filtered reads from the single-cell libraries mappedto the genome (80±5 %) than those from the NCC libraries (3.0±0.3 %); the126Lysis3 nL3’ Ligation+ 7 nL5’ Ligation+ 4 nLcDNA synthesis+ 16 nLIndexing PCR+ 70 nLElution+ 10 μL500 μmA5’App ddC5’Phos 3’ OHB3’ Ligation5’ LigationReverse transcriptionIndexing PCR+ T4 RNA ligase 2 Tr KQ, adenylated 3’ adapter (DNA)+ ATP, T4 RNA Ligase 1, 5’ adapter (RNA)+ RT brew, RT primer, adapter-adapter block (LNA)+ PCR brew, reverse primer, sequencing-indexed primerFigure 4.5: Method for single-cell miRNA-seq. (A) Schematic of one cell-processing unit from the microfluidic device used for library construction.Each unit contains a cell capture chamber and a compound reaction chamberto successively assemble small-RNA libraries. (B) The library constructionstrategy. Single cells were first lysed to release RNA. A pre-adenylated 3′adapter (blue) was then ligated to the 3′ end of any small RNAs, followed bythe ligation of the 5′ adapter (green) to the 5′ ends. Complementary DNA wasnext created in the presence of an LNA-modified adapter-adapter block [251](dark blue) to help remove adapter-dimers. Finally, sequencing adapters andindexes were added during PCR amplification. Libraries were then simulta-neously pooled and recovered from the device for subsequent quality-control,size-selection and sequencing.127reads from the NCC libraries predominantly stemmed from the spike-in control(92±1 %). These statistics demonstrate specificity for, and uniform read cover-age across single cells, as well as a low contamination of unwanted by-productsof library construction. We found a tight distribution of read lengths for a repre-sentative library (Figure 4.6A), which, combined with the read annotations (Fig-ure 4.6B) showed that RNA integrity was maintained prior to and during libraryconstruction. Libraries showed both a high percentage of miRNA tags (61±7 %),and a low contribution of the most abundant RNA degradation products (10±3 %,cumulative ribosomal RNA, exonic and intronic RNA). We also observed highmiRNA diversity, with the average single-cell library having 145± 28 miRNAsper cell with at least 10 reads. This measurement was better than or comparable topreviously published bulk K562 microRNA-seq data that were down-sampled tomatch sequencing efforts; i.e., 57±7, 155±17, and 78±9 miRNAs with at least10 reads for references [1, 241, 242], respectively. Diversities achieved by poolingsingle cells to match the sequencing effort of these bulk data sets were also superioror comparable; i.e., 213 %, 106 %, and 149 % of the total diversities, respectively.We next evaluated the quantitative performance and sensitivity of miRLin us-ing a spike-in control that was included in the lysis buffer and consisted of tensynthetic RNA species at known concentrations. By averaging the reads per mil-lion mapped for each species across the 87 K562 cells that passed QC, we ob-served the measured abundance to increase linearly (R2 = 0.95) with input concen-tration (Figure 4.6C). Technical noise was assessed by generating libraries fromsingle-cell equivalents of bulk K562 cell lysate. The coefficient of variation in thelysate samples and the spike-in controls ranged from approximately 13 % for thehighly expressed miRNAs to 62 % for those approaching the sensitivity limit, andwas below the variation observed between single cells at all concentrations (Fig-ure 4.6). The average spearman correlation between lysate samples processed ondifferent chips and days was 0.76±0.02 (Figure 4.7), and that of the average lysateto average single cell was 0.78, demonstrating a high reproducibility both withinand between samples (Figure 4.6E, Figure 4.7). Single-cell miRNA expressiondistributions closely resembled those derived from bulk libraries, with the top 10miRNAs comprising 73±4 % of the entire library, followed by a long, gradual tail(Figure 4.6F). After converting read counts to copy-number using the spike-in con-128miRNA per cell × 1042 3 4 50105Number of cells0 50 100 200150Ranklog 10 Expression01243Single cellMedianlog10 Average celllog 10 Average lysate4-220-2420miRNASpike-inrs=0.77Cell: miRNACell: Spike-inLysate: miRNALysate: Spike-in-2log10 Mean expressionlog 10 CV20 2 4-2-1010log10 molecules/reaction1 2 3log 10 TPM024Single cellMean% of aligned reads025507510088 single K562 cells 4 NCCTotal miRNAOther small RNArRNARNARepeatsUnknownSpike-inUnaligned0Insert Size [nt]10 20 30% of reads01020305040A B CFEDFigure 4.6: Single-cell miRNA-seq method validation and performance. (A)Insert size distribution and alignment tags from a representative single-celllibrary. The insert lengths were centered on 22-bases, the characteristiclength of miRNAs. (B) Alignment statistics for 88 single K562 cells and 4no-cell controls (NCC). With one exception, there was uniform representa-tion of the tag classes across single cells, with the majority of aligned reads(61.4±6.9 %) annotated as miRNAs. The outlier cell (denoted with an as-terisk) contained a disproportionate amount of common RNA degradationproducts and is thus likely not viable. The vast majority (92±1 %) of thereads from the NCC stemmed from the spike-in control. (C) Linearity of thespike-in control. Log transcripts per million mapped (TPM) at the known con-centration for each of the 87 single-cell K562 libraries. Linear regression wasperformed using the mean TPM values. (D) Quantification of technical andbiological variation. The technical noise as described by libraries generatedon single-cell equivalents of cell lysate (blue) and the spike-in controls (greenand purple) is below the biological variation seen between single cells (red).(E) Assay reproducibility. The average normalized expression of the single-cell libraries closely correlates with the average of the single-cell equivalentlysate libraries (spearman correlation rs = 0.77). (F) Expression distributions.The single-cell miRNA expression distributions were characterized by a fewmiRNAs dominating the library followed by a long, gradual tail. Inset: TotalmiRNAs per K562 cell as estimated based on our capture efficiency.1290.77 0.75 0.78 0.75 0.74 0.74 0.74 0.78 0.75 0.74 0.770.77 0.79 0.79 0.79 0.79 0.74 0.77 0.81 0.77 0.790.74 0.75 0.74 0.74 0.76 0.74 0.75 0.79 0.770.75 0.76 0.76 0.74 0.75 0.75 0.75 0.790.76 0.78 0.71 0.76 0.79 0.76 0.810.74 0.74 0.74 0.75 0.73 0.750.74 0.75 0.75 0.74 0.780.75 0.75 0.72 0.750.79 0.74 0.770.78 0.770.77Experiment replicateDay 1 Day 2 Day 3LysateD1.R1LysateD1.R2LysateD1.R3LysateD1.R4LysateD2.R1LysateD2.R2LysateD2.R3LysateD2.R4LysateD3.R1LysateD3.R2LysateD3.R3LysateD3.R4Figure 4.7: Correlogram denoting miRNA expression values derived fromfour replicate K562 lysate samples at concentrations equivalent to that froma single cell, from each of three independent experiment days (red, yellow,and blue). Spearman correlations for each pairwise comparison are indicatedin the upper right half. The high correlation between all samples indicatesassay reproducibility between samples both within and between independentexperiments.trol, we estimated there to be an average of 31×103±7×103 miRNAs per singleK562 cell (Figure 4.6F inset), roughly agreeing with estimates based on fractionalmiRNA abundance (∼ 1×105 miRNAs/cell) [115, 252].Finally, we estimated our conversion efficiency to be 6.5±1.0 % by generating130Rand.1_S16Rand.2_S17Rand.3_S18Rand.4_S19Rand.NTC_S208006004002000Count252015105Percent of pairs0Levenshtein distance2 4 6 8 10 12 14 16 18 20 22 24Number of reads per unique random sequence0 50 100 150A BFigure 4.8: Measurement of conversion efficiency. (A) The number ofreads from each random sequence recovered from four replicates and oneno-template control (NTC). There was an average of 56.9± 23.9 reads perunique randomer. (B) Histogram of the pairwise Levenshtein distance be-tween the consensus randomer sequences. The large average distance be-tween consensus sequences suggests that clusters originated from true singlemolecules rather than as artefacts of library construction or sequencing errors.libraries from a known concentration of a synthetic 22-base randomer and countingthe number of unique recovered sequences per library (Figure 4.8). At a 95 %detection confidence, this efficiency corresponds to a sensitivity of 46 moleculesper cell, which matches the intracellular concentration at which miRNAs have beenreported to cause functional effects [67, 68]. We therefore believe that we arecapturing the majority of the functional miRNAs within each single cell analyzed.These overall statistics demonstrate that miRLin generates high-quality miRNAsequencing libraries with quality metrics comparable to, or exceeding, those ob-tained using traditional, high-input, bulk library preparation methods and with asensitivity sufficient to capture the functional miRNAs within a single cell.1314.3.2 Comparison of performance to existing methods formiRNA-seqWe next compared general sequencing metrics generated using our method to bulkmiRNA-seq libraries generated for the ENCODE consortium and to the single-cellsmall-RNA libraries prepared by Faridani et al. [126] (Figure 4.9A).On average, miRLin libraries were sequenced to an average depth of 16.0×104±6.3×104 reads per cell, smRNA-seq libraries to 4.4×106±3.5×106 readsper cell, and ENCODE to 8.6× 106± 8.4× 106 reads per sample. smRNA-seqsingle-cell libraries thus had an average of 27.2× more sequencing effort than themiRLin libraries. The contribution of adapter-dimers was fairly similar acrossmethods, with miRLin libraries producing an average of 9.5±8.2 %, smRNA-seq 15.0±13.8 %, and ENCODE 3.1±3.1 %. Libraries prepared using smRNA-seq, however, contained a substantial proportion of reads whose insert size wasless than our minimum cut-off of 15 bases (miRLin: 0.6±0.4 %, smRNA-seq:32.3±10.2 %, ENCODE: 1.3±1.2 %). Furthermore, a substantial fraction of thesize-filtered reads from these libraries did not map to the human genome (miRLin:5.7±2.4 %, smRNA-seq: 55.2±21.9 %, ENCODE: 10.3±7.4 % unaligned). Cu-mulatively, these by-products of library construction resulted in a high percentageof discarded reads in the smRNA-seq libraries (miRLin: 15.1±9.1 %, smRNA-seq: 76.5±13.0 %, ENCODE: 14.0±9.2 % discarded).The insert-size distribution from a representative library of each set were nextexamined (Figure 4.6A, Figure 4.9B). In general, the distributions from the EN-CODE (Figure 4.9B right) and miRLin (K562: Figure 4.6A, hema: Figure 4.9Bleft) sets were characteristic of miRNA libraries, with the insert size centered on 22nucleotides and the majority of reads aligning to known miRNAs. While the sizefraction that contained long fragments (insert size approximately > 28 bp, referredto as “hema L”) was not fully sequenced with the 61-base read length, reads fromthese samples predominantly mapped to other known classes of small RNAs (Fig-ure 4.9B, middle left). In contrast, the majority of the inserts from the smRNA-seqlibrary (Figure 4.9B, middle right) were either less than 15 nucleotides or greaterthan the 41-base read length, and only a minority mapped to known miRNAs andother small RNAs. Furthermore, as the smRNA-seq library construction protocoldoes not include a size selection, the tag insert-size distribution between 15 and 30132K562hemahema_LFaridaniENCODEReads per library 1×1071×105 % adapter dimersK562hemahema_LFaridaniENCODE0255010075% < 15 bp% size-filtered reads aligned to hg38K562hemahema_LFaridaniENCODE0255010075K562hemahema_LFaridaniENCODE0255010075K562hemahema_LFaridaniENCODE0255010075% of reads discardedAB% of reads020400204001020300102051525Insert size [nt]hema_LInsert size [nt]hema0 10 20 30 0 20 40 60Insert size [nt]Faridani0 10 20 30 40Insert size [nt]ENCODE0 10 20 30 40 50Total miRNAOther small RNArRNARNARepeatsUnknownSpike-inUnalignedInsert size [nt]15 20 25 301.20.90.60.30.0K562hemahema_LFaridaniENCODE0255075100K562hemahema_LFaridaniENCODE0255075100K562hemahema_LFaridaniENCODE0255075100K562hemahema_LFaridaniENCODE0255075100K562hemahema_LFaridaniENCODE0255075100K562hemahema_LFaridaniENCODE0255075100% Total miRNA% Other small RNA% rRNA% RNA% Repeats% UnknownClog 10 UMI (Faridani pipeline)0123log10 TPM (TCGA pipeline)0 2 4log 10 number of miRNA with 10  ≥ reads0123log10 Total reads5 6 7K562hemaFaridaniENCODED Elog10 count0 321133Figure 4.9 (previous page): Comparison of miRNA library quality met-rics for miRNA/small-RNA libraries generated using miRLin (K562, hema,hema L), smRNA-seq (Faridani et al. [126]), and the human ENCODE li-braries. (A) Total reads per library (left) with the cumulative (right) andbreakdown percentages of reads discarded due to adapter-dimers, minimumsize filters, and genome alignment. (B) Insert size distributions with align-ment classes from a representative single-cell library from the hema (li-brary 49F-C01 S2), hema L (library L49F-C01 S2), smRNA-seq (librarySRR3495777, 51 HEK 1) and ENCODE (ENCFF001RDL, K562) sets. (C)Percent of hg38-aligned reads for each alignment classification. miRLinmiRNA-enriched libraries (K562 and hema) predominantly map to miRNAsand the long fraction (hema L, insert size greater than approximately 30 bp) ischiefly other classes of small-RNAs. smRNA-seq libraries, on the other hand,contain a substantial fraction of reads stemming from other classes of RNA.(D) miRNA library diversity versus the number of reads for each library. (E)Comparison of the UMI-based expression quantification calculated by Fari-dani et al. to the TPM values calculated by the TCGA pipeline [122] usedhere.nucleotides was also examined (Figure 4.9B, middle right inset). While the readsmapping to known miRNAs were centered at 22 nucleotides, reads within this sizerange were primarily from other annotation classes.Extending the read annotations across each complete set of libraries (Figure 4.9C)further highlighted the differences in library content. As expected, the miRLinand the ENCODE miRNA-seq library sets were predominantly miRNA (miRLin:76.5±11.7 %, ENCODE: 75.4±11.5 %), and the hema L libraries were primarilyother classes of small RNA (65.5±7.7 % total small RNA, 62.1 % snoRNA, 1.7 %tRNA, 1.2 % scRNA, 0.5 % snRNA, and 0.03 % srpRNA). These libraries fur-ther had a low contribution of common RNA degradation intermediates (miRLin:5.2±4.3 % rRNA, 6.6±5.9 % RNA; smRNA-seq: 4.6±2.8 % rRNA, 45.0±14.6 %RNA; ENCODE: 4.9±2.8 % rRNA, 6.3±4.7 % RNA). In contrast, the smRNA-seq libraries primarily contained other types of RNA, with the majority of thesestemming from 5.8S and 45S ribosomal RNA. Only 2.0±1.1 % and 18.1±8.9 %of the aligned reads mapped to known miRNAs and other classes of small RNAs,respectively (12.4 % snoRNA, 5.45 % tRNA, 0.0823 % scRNA, 0.109 % snRNA,134and 0.0249 % srpRNA).Together, these insert-size distributions and read annotations suggest that RNAdegradation intermediates are a substantial component of the smRNA-seq libraries.The majority constituents cDNA libraries prepared from small RNAs are typicallymiRNAs and other classes of small silencing RNAs [19, 253]. While it is likelythat some of the observed differences may be attributed to the different cell typesmeasured with each method, they cannot account for all of them as the ENCODElibraries span a wide variety of tissue types and include some of those that wereused by Faridani et al. We also note that without a “no-template control” includedwith the smRNA-seq data, background signal cannot be determined. We thus be-lieve that the performance provided by smRNA-seq presents considerable barri-ers preventing its adoption for routine single-cell miRNA profiling. Aside fromthe practical concerns associated with recovering 0.5±0.3 % miRNA reads per li-brary, certain applications, such as miRNA discovery, require that RNA integrity ismaintained [254, 255].4.3.3 Generation of single-cell miRNA profiles on 12 purified subsetsof human cord blood cellsWe then applied this technology to analyze the changes in miRNA expression thatoccur as normal human hematopoietic stem cells (HSCs) differentiate. Accord-ingly, twelve populations (Figure 4.10A) were isolated by fluorescent activatedcell sorting (FACS) from a pool of normal cord blood (CB) cells (Figure 4.1,Figure 4.2). In addition to total CD34+ and CD34+38-, these populations in-cluded three subsets of CD34+38- cells: CD34+CD38-CD45RA-CD90+CD49f+cells (49f cells), a population highly enriched for long-term repopulating hematopoi-etic stem cells (LT-HSCs, ∼10 % cells with 30-week repopulation ability in xeno-transplant models) [162, 256] as well as downstream cells referred to as MPP andderivative MLPs (Figure 4.10A). From the more mature CD34+38+ (progenitor)population, we isolated three subtypes referred to as CMPs, MEPs, and GMPs,and from the CD34- compartment we isolated four types of mature blood cells:monocytes, T cells, B cells, and erythroid cells. A total of 2057 single-cell librariespassing QC were analyzed (444 CD34+, 276 CD34+38-, 181 49f, 215 MPP, 164MLP, 157 CMP, 151 MEP, 80 GMP, 82 Monocytes, 44 Erythroid, 160 B, 103 T;135TBNKDendriticMonocytesGranulocytesErythrocytesMKCCD34+ CD38-CD34+MLPMEP GMP49fMPPCMP tSNE2-40400-60 -30 0 30 60tSNE1Cumulative percent10075Pseudotime differentiation axis502501007550250A B CFigure 4.10: Hematopoietic cell type organization. (A) Classic hematopoi-etic developmental hierarchy (adapted from [157]). Single cells were isolatedfrom the flow-sorted populations that have been coloured (Figure 4.1, Fig-ure 4.2). (B) t-SNE analysis of hematopoietic cells. The colour of each pointcorresponds to the phenotypic population from which it originated. Cells areseparated into five main clusters, one containing the stem and progenitor cellsand the other four each containing a mature cell type (Figure 4.11A). (C)Empirical cumulative distribution function of the HSC and progenitor pop-ulations along a pseudotime differentiation axis. Cells were ordered basedon their location along the major axis of the progenitor cluster derived fromthe t-SNE analysis (Figure 4.12, Figure 4.13). The most primitive cell types(49f, MPP) are highly enriched towards the start of the ordering, whereas thecommitted progenitors (MEP, GMP) appear towards the end.total 2057) from thirty chip runs, generating an average of 15.4×104±5.2×104reads per library with 82±7 % aligning to the genome or spike-in, and 88±7 %of the genome-aligned reads coming from miRNAs. We estimated there to bean average of 12.3× 103± 6.1× 103 miRNAs per single primary hematopoieticcell, in close agreement with a previous estimate of 11,587 miRNAs per humanCD34+133- cell [252].We first asked how the miRNA expression profiles segregated the differentphenotypes of cells examined. Analysis of the data using t-distributed stochasticneighbour embedding (t-SNE) dimensionality reduction [245] and density-basedspatial clustering [257] recovered five distinct clusters (Figure 4.10B, Figure 4.11)consisting of a large group of progenitor cells and the four mature cell types an-alyzed. Within the progenitor cluster, the cell-type distribution was loosely asso-13640-40300-60 -30 60tSNE1300-60 -30 60tSNE1tSNE2 0400-40tSNE212345Cluster CD34+CD34+38-49fMPPMLPCMPGMPMEPMonocytesErythroidBTA BFigure 4.11: t-SNE analysis of hematopoietic cells. (A) Clusters identifiedbased on spatial density [257], with cluster 1 (red) containing the HSC andprogenitor cells, cluster 2 (olive) containing the monocytes, cluster 3 (green)the T-cells, cluster 4 (blue) the B-cells and cluster 5 (purple) the erythroidcells. (B) Cell-types used for downstream analysis as reclassified using theclusters described in (A) to correct the 16 obvious phenotypic sorting errorsseen in Figure 4.10B.ciated with surface phenotypes. The 49f cells localized primarily at the top of thecluster, followed by the MPPs in the middle, with MLPs, CMPs, GMPs, and MEPstowards the bottom. As expected, total CD34+ cells were distributed throughout.Notably, the MLPs grouped most closely with the GMPs. Despite this rough sep-aration, substantial overlap between different phenotypes was prominent and therewas no evidence of sharp transitions in cell states.To examine how miRNA expression evolves during differentiation in more de-tail, we next ordered cells from all of the analyzed populations from the HSCs (49fcells) to the mature subsets by projecting their location along the major axis of theprogenitor cluster ellipse (Figure 4.10C). The resulting positioning of cells alongthis pseudotime differentiation axis was further confirmed using independent meth-ods for reconstructing differentiation trajectories [214, 247, 258] (average spear-man correlation 0.83, Figure 4.12, Figure 4.13). Using this marker-independentmeasure of cell differentiation, we then performed hierarchical clustering of the137-0.05 0.00 0.05 0.200.150.10-0.050.000.05DM3DM1-2 0 2 4 6-2-101LLE1LLE2-3 -2 -1 0 1 2-3-2-121060300-30-60-40040tSNE2IC1. BIC2. HSCtSNE1A BC DFigure 4.12: Dimensionality reduction and pseudotime ordering computedusing (A) t-distributed stochastic neighbour embedding (tSNE) followed byprojection along the major axis of the progenitor cluster, (B) independentcomponent analysis decomposition using information-maximization [246]followed by minimum-spanning tree ordering [214], (C) diffusion map de-composition [258] and fitting a principal curve to components 1 and 3, and(D) local linear embedding followed by principal curve ordering.differentially expressed miRNAs to reveal the coordinated changes in expressionassociated with the progression of human hematopoiesis from the HSC-enrichedsubset, through the intermediate progenitors, and finally to the mature cell types(Figure 4.14A).138-2-101230501001500.000.250.500.751.000.250.500.751.000.000.250.500.751.000500100015000 500 1000 15000500100015000 500 1000 15000500100015000 500 1000 15000500100015000 500 1000 15000500100015000 500 1000 1500-30 0 30 -2 -1 0 1 2 3 0 50 100 150 0.00 0.50 1.00 0.25 0.50 0.75 1.00A: tSNEPCA orderedB: ICAAxis orderedB: ICAMonocleorderedC: Diffusion MapPrincipal curveorderedWanderlustD: LLEPrincipal curveorderedFigure 4.13: Pairwise scatterplots comparing pseudotime orderings in rawmethod-specific coordinates (lower half), and ranked coordinates (upper half).The pseudotime differentiation metric used in our analysis was computed us-ing tSNE dimensionality reduction and HSPC ordering along the major axis(Figure 4.11A). The high concordance between the results derived from thesevaried methods supports this pseudotime ordering.139Two groups of miRNAs showed cell-type specificity based on this hierarchicalclustering exercise. The first contained four miRNAs – miRNAs 451a, 144-5p,144-3p and 486-5p – exclusively expressed within the erythroid cells (erythrob-lasts) (Figure 4.14A and B, group ii). This observation corroborates previous workdemonstrating both the importance and expression specificity of miR-451a and 144[179, 180], and 486 [181] in erythropoiesis. The second group contained twelvemiRNAs – miR-10a-5p, 125a-5p, 99b-5p, 125b-5p, let-7c-5p, 99a-5p, 126-5p,126-3p, 127-3p, 181c-5p, 181d-5p, and 196b-5p – that were primarily expressedwithin the progenitor compartment (Figure 4.14A and B, group i). Expressionand functional studies on several members of this group have implicated them inbalancing HSC self-renewal and quiescence with differentiation and proliferation[33, 101, 111, 173, 177, 178, 259–261]. Notably, Gentner et al. identified miR-126-3p as a biomarker for the prospective purification of human HSCs [101]. Here,the expression of miR-126-3p was also seen to reach its apex within the primi-tive HSCs and steadily decrease during differentiation (Figure 4.16A), consistentwith our predicted developmental ordering. This expression trend was maintainedacross all members of this miRNA group (average spearman correlation to 126-3p of 0.74±0.09), with miR-10a-5p being the most highly expressed and sharplydownregulated. These observations identify a set of miRNAs that are preferentiallyexpressed within HSCs and that may play a key role in maintaining HSC function.Even though the majority of detected miRNAs were differentially expressedbetween the sorted cell populations, their expression was generally not restrictedto a specific cell type. Independent component analysis (ICA) was performed toidentify the miRNA driving cell type delineation. The first five (56 % of the totalvariance) of the eight extracted components separated the B, HSC, erythroid, T,and monocyte cells, respectively (Figure 4.14B-C).We observed substantial over-lap amongst the miRNAs that defined these components (Figure 4.14B); rather thancontaining sets of “marker” miRNAs, they were instead defined by subtle combi-natorial differences. This overlap between expression signatures arose largely as aconsequence of the primitive HSCs exhibiting broad and robust expression of manyof the miRNAs that were most highly differentially expressed between mature celltypes (e.g., miRs 146ba-5p, 223-3p, 150-5p).Cell projections onto the five components delineating the principal cell identi-140HSCMonocytesErythroidB T IC8IC6 IC7miR-144-3pmiR-144-5pmiR-451amiR-486-5pmiR-181a-3pmiR-181b-5pmiR-181a-5pmiR-22-3pmiR-140-3pmiR-191-5pmiR-425-5pmiR-15b-5pmiR-186-5pmiR-30b-5pmiR-374b-5pmiR-374a-5plet-7i-5plet-7f-5plet-7a-5pmiR-26b-5pmiR-29c-3pmiR-29a-3pmiR-26a-5pmiR-342-3pmiR-150-5plet-7g-5pmiR-142-5pmiR-142-3pmiR-21-3pmiR-21-5pmiR-181a-2-3pmiR-374a-3pmiR-28-3pmiR-30e-3pmiR-30e-5pmiR-101-3pmiR-30d-5pmiR-25-3pmiR-92a-3pmiR-106b-5pmiR-20b-5pmiR-130a-3pmiR-17-5pmiR-20a-5pmiR-335-3pmiR-30c-5pmiR-19a-3pmiR-19b-3pmiR-103a-3pmiR-93-5pmiR-223-3pmiR-542-3pmiR-199b-3pmiR-199a-3pmiR-27b-3pmiR-23b-3pmiR-27a-3pmiR-24-3pmiR-23a-3pmiR-320amiR-221-3pmiR-361-5pmiR-155-5pmiR-146a-5pmiR-146b-5pmiR-125b-5pmiR-127-3pmiR-181d-5plet-7c-5pmiR-181c-5pmiR-99b-5pmiR-125a-5pmiR-196b-5pmiR-99a-5pmiR-10a-5pmiR-126-5pmiR-126-3pIC1: BIC2: HSCIC3: ErythroidIC8IC7IC6IC5: MonocytesIC4: TBCiiiAlog10 miR-486-5p21-60 -30 0 30 60tSNE1tSNE2-40040Diversity100500 50 100Pseudotime differentiation axisM E B T0 4log10 expression2-5ICA projection0 50-0.4 0.4Mixing coefficientCell populationCD34+CD34+38-MPPMLPCMP49fGMPMEPMonocytesBTErythroidD E141Figure 4.14 (previous page): miRNA expression in hematopoietic cells. (A)Heat map of expression (log10 copies per cell) of detected miRNAs. Pro-genitor cells are ordered based on their position along the pseudotime differ-entiation axis; the top colour bars indicate cell population. 12 HSC-specificmiRNAs are denoted in group i; 4 erythroid-specific miRNAs in group ii.(B) Heat map of miRNA mixing coefficients derived from ICA analysis. Themagnitude of a miRNA’s coefficient is a measure of its contribution to definingthat component. (C) Heat map of cell projections onto the eight independentcomponents defined in B. While there is sufficient discriminatory power in themiRNA expression profiles to separate the cell types, the distinguishing fea-tures rely on combinatorial differences in expression rather than sets of markermiRNAs. (D) HSC differentiation is generally characterized by a decrease inboth miRNA expression and diversity. (E) Expression of miR-486-5p is in-creased in erythroid cells and a mature subset of progenitor cells, suggestingthat these progenitor cells have committed towards the erythroid lineage.ties exhibited a gradual downregulation of the HSC component during progenitorderivation, followed by a stark transition to an expression signature characteristicof a mature cell type (Figure 4.14C). HSC differentiation was further character-ized by a decrease in both miRNA diversity and total miRNA (Figure 4.14D, Fig-ure 4.15). As endogenous mRNA content was previously shown to increase duringmurine HSC differentiation [166], this miRNA decrease is unlikely to simply be acompensatory response to changes in total cellular RNA amounts [262].Although the HSCs did not display “mixing” of the progenitor signature withthose of the mature cells, the last three independent components (11 % of the totalvariance) captured heterogeneity within the progenitor cells. IC6 contained numer-ous miRNAs with similar levels of influence, and thus was likely mainly describingcell-to-cell variability. IC7 primarily captured the up-regulation of members of themiR-181a family during progenitor maturation. While this family has been previ-ously associated with modulating the differentiation of B cells [182], we observedit to be universally up-regulated during HSC differentiation (Figure 4.16C) andco-expressed with miR-223-3p (Figure 4.16B,D), a miRNA previously reported tobe both specifically expressed within myeloid cells [182] and up-regulated duringgranulopoiesis [183]. It is therefore doubtful that this increased miR-181a expres-1421230Pseudotime differentiation axisM E B T0 10050Pseudotime differentiation axisM E B T0 10050Aligned reads ×105240Total reads ×105M E B T0 10050 M E B T0 10050120Total expression/spike-in 45310050DiversityA BDCCell populationMLPCMPGMPMEPMonocytesErythroidBTCD34+CD34+38-49fMPPFigure 4.15: Decrease in (A) miRNA diversity and (B) total miRNA expres-sion during HSPC differentiation. In contrast, while there is a slight drop inboth the number of (C) aligned and (D) total reads during HSPC differentia-tion, they remain relatively constant.143log 10 miR-223-3p expressionlog 10 miR-181a-5p expressionlog10 miR-223-3p expression320 1 43201Pseudotime differentiation axis0 50 100 M E B Tlog 10 miR-181a-5p expressionlog 10 miR-126-3p expressionPseudotime differentiation axis0 50 100 M E B TPseudotime differentiation axis0 50 100 M E B T3201320132014Cell populationMLPCMPGMPMEPMonocytesErythroidBTCD34+CD34+38-49fMPPA BDCFigure 4.16: Highlighted miRNA expression. (A) miR-126-3p expressionpeaked in the HSCs, dropped during differentiation and was absent in themature cells. (B) miR-223-3p was constitutively expressed in the progenitorcells, upregulated in the monocytes, down-regulated in B-cells and heteroge-neously expressed in T-cells. (C) miR-181a-5p was universally up-regulatedin cells from the committed progenitors. (D) miR-181a-5p and miR-223-3pwere both co-expressed in all HSC and progenitor cells analyzed.144sion was indicative of lymphoid commitment. In contrast, IC8 identified a subsetof the most differentiated HSPCs that were expressing miR-486-5p (Figure 4.14E).As miR-486 is located within the last intron of ANK1, an integral membrane proteinthat is selectively expressed in erythroid cells [263] and required for erythroid de-velopment [264], it is likely that the increased expression in these cells is evidenceof progenitor commitment towards the erythroid lineage. With this one exception,the miRNA expression signatures did not provide clear evidence of lineage com-mitment within the progenitor cells.4.3.4 MicroRNA isoforms are dynamically expressed duringhematopoiesisVariations in the precise processing steps necessary for miRNA maturation can re-sult in the production of miRNA isoforms (isomiRs) [27]. As isomiR abundancehas been previously seen to change with cell type, we examined the abundance ofthree different classes of miRNA isoforms during hematopoiesis. As expected,the 5′ end was observed to be, cumulatively, more precisely cut and exhibitedreduced non-templated base additions (NTA) compared to the 3′ end (95.2 % vs60.6 % of molecules cut a the annotated 5′ and 3′ locations; 99.7 % vs 74.0 % ofthe molecules were unmodified at the 5′ and 3′ ends; Figure 4.17). On a per-celllevel, the proportions of the total miRNAs at 5′ cut sites relative to the annotationwere found to be stable (Figure 4.18A), whereas those at the 3′ relative cut sites(Figure 4.18B) and 3′ non-templated base additions (Figure 4.18C) were foundto change both during HSPC differentiation and between mature cell types. Col-lectively, these global processing profiles demonstrate that miRNA isoforms are asubstantial and differentially regulated component of miRNA expression in hema-topoiesis.Using aggregate data, we identified ten miRNAs yielding isomiRs with an al-ternate 5′ cut location. These isoforms altered the miRNA seed sequence, poten-tially transforming the target repertoire. As targets for only four of these isomiRshad been previously annotated [249], we used TargetScan [249] and miRanda [250]to predict target transcripts for both the canonical and alternate 5′ isomiRs. Wefound that despite there being a significantly higher overlap in putative targets be-tween an isoform and its annotated form than between random pairs, the predicted145-3 3001007550253’ cut positionw.r.t. annotation5’ cut positionw.r.t. annotation3’ non-templated base additionAUNA A U AA G UU TA C-3 30% of cumulative expressionNA A5’ non-templated base additionA BFigure 4.17: Cumulative miRNA cut and non-templated base additions for(A) the 5′ and (B) 3′ ends. NA corresponds to no addition. Only non-templated base additions at a frequency greater than 0.1 % for each end areshown.target repertoires were still mostly unique (Figure 4.19) (TargetScan: mean overlapof 15.9 % vs. 8.4 % seen in randomly shuffled pairs, p = 0.0016; miRanda: meanoverlap of 19.5 % vs. 9.7 %, p = 8.5× 10−6; Wilcoxon rank sum test). Addi-tionally, while the cumulative contribution of these isoforms to the combined totalexpression was low (2.4 %), their fractional expression from each specific maturelocus was substantially higher (5.2 to 98.0 %, mean 31.7 %). As many of theseloci were highly expressed, several isoforms thus had an average expression levelexceeding that of other distinct miRNAs (Figure 4.18D). Finally, the fractionalexpression from each mature locus was maintained across cell types; only three5′ isomiRs exhibited statistically significant differences between cell types, andthese observed differences were modest (miR-101-3p: mean change 30 molecules/-cell; miR-142-5p: mean change 0.6 molecules/cell; miR-142-3p: mean change 53molecules/cell; Figure 4.20). Together, these data indicate that 5′ isoforms havedistinct targeting repertoires, are expressed at functionally relevant levels, and aregenerated in a miRNA-intrinsic manner.As we previously observed substantial overlap in the miRNAs that were differ-1463’ non-templated base additions NA A U AAAU3’ cut position-2 -1 0 1 25’ cut position-2 -1 0 1 21007550250Percent of expressed10075502501007550250Percent of expressedPercent of expressedCanonical miRNAUnannotated 5’ isoformAnnotated 5’ isoformlog 10 mean expression0321miR-10a-5p +1miR-101-3p -1miR-451a +1miR-126-3p +1miR-142-3p +1miR-181c-3p +1miR-140-3p +1miR-142-5p -2miR-29a-3p -1miR-542-3p -1Rank0 50 100100500100500100500100500100500Percent of expressedmiR-30e-5p 3’ cutlet-7a-5p 3’ cutmiR-101-3p 3’ cutmiR-92a-3p 3’ cutmiR-92a-3p 3’ non-templated base additionsABCDEIHGFCell populationMPPMLPCMPGMPMEPMonocytesCD34+CD34+38-49fErythroidBTFigure 4.18: Proportional expression of miRNA isoforms during HSC devel-opment. The proportion of the total expression from each single cell at eachof the (A) relative 5′ cut locations, (B) relative 3′ cut locations, and (C) 3′non-templated base additions is colour coded for each isomiR class (greento pink). HSPC cells are ordered according to their pseudotime differentia-tion state (top colour bar). Compared to the 3′ modification profiles, those ofboth the relative 5′ cut locations and 3′ non-templated base additions are rel-atively stable during HSC differentiation and between mature cell types. (D)Ranked average expression of canonical mature miRNAs (light blue), anno-tated 5′ isomiRs (dark blue), and previously unannotated 5′ isomiRs (pink).Black horizontal lines connect the expression of a canonical miRNA with its5′ isoform, and the isomiR names and change in cut location are indicatedabove. Many of these isoforms are expressed at functionally relevant lev-els, and exceed those of other, distinct miRNAs. (E-I) Selected examples ofmiRNAs whose 3′ cut location and non-templated base addition proportionprofiles change with respect to cell type. There was substantial variability be-tween miRNAs in terms of both how these profiles changed with cell type andthe magnitude of these changes.1470 25 1007550miR-101-3pmiR-10a-5pmiR-126-3pmiR-140-3pmiR-142-3pmiR-142-5pmiR-181c-3pmiR-29a-3pmiR-451amiR-542-3p0 25 1007550% of TargetScan hits % of miRanda hitsCanonicalIsoformBothA0 10 3020% Overlap of TargetScan hits0 10 3020% Overlap of miRanda hits010302005001000010302040013202507505001250100010Number of pairsTargetScanAll Human miRNA, 10000 random pairsAnnotated and isoforms, all mismatched pairsmiRandaAll Human miRNA, 10000 random pairsAnnotated and isoforms, all mismatched pairsAnnotated and isoforms, all matched pairs Annotated and isoforms, all matched pairsBFigure 4.19: (A) Overlap in the putative target repertoires between canon-ical miRNAs and their 5′ isoforms. (B) Percent putative target overlap for(top) 10,000 random pairs selected from all canonical human miRNAs, (mid-dle) all (N = 180) mismatched pairs from the set of 10 canonical miRNAsand their 5′ isoforms (e.g., miR-101-3p vs. miR-10a-5p), and (bottom) all(N = 10) matched pairs from the same set (e.g., miR-101-3p and its isoform).There was a significantly higher overlap in putative targets between the set ofcanonical miRNAs and their isoforms compared to the global set (TargetScan:p = 0.040, miRanda: p = 8.9× 10−5, Wilcox rank sum test) and the mis-matched set (TargetScan: p = 0.0016, miRanda: p = 8.5×10−6, Wilcox ranksum test).148entially between the cell populations, it was unlikely that the observed changes tothe global 3′ modification profiles were simply due to this differential expression.We therefore next analyzed changes to the relative isomiR abundance from eachmature locus in order to identify miRNAs with varied 3′ isomiR proportions. Thefrequency of miRNAs with 3′ modifications was higher than those with variable 5′ends, with the majority of robustly expressed miRNAs having at least one 3′ iso-form expressed above our detection thresholds (3′ cut location: 78/133; 3′ NTA:111/133). In contrast to the stable global and miRNA-specific 5′ cut-location pro-files, 51 and 28 miRNAs had 3′ cut (Figure 4.21, Figure 4.22) and non-templatedbase addition (Figure 4.24, Figure 4.25) proportion profiles that varied significantlywith cell type, respectively. The magnitudes of these changes were also larger thanthose seen at the 5′ end. A marked increase in miRNAs that were cut upstreamof the annotated 3′ cut site in B- and T-cells emerged as a common trend in themodification profiles; however, this shift was not universal (e.g., miR-101-3p pro-portions remained relatively constant, Figure 4.18G). Instead, there was substantialvariability between the modification profiles for each miRNA, with differences inthe modification levels (e.g., miR-30e-5p, vs. -101-3p and -92a-3p 3′ cut, Fig-ure 4.18E, G, H), their cell-to-cell variability (e.g., let-7a-5p vs. miR-92a-3p 3′cut, Figure 4.18F, H), and how they changed with cell type (e.g., miR-30e-5p vs.let-7a-5p vs. miR-92a-3p 3′ cut and NTA, Figure 4.18E, F, H). Of note, the frac-tion of adenylated miR-92a-3p was seen to decrease in the erythroid cells and thesubset of progenitors found to be expressing miR-486-5p (Figure 4.18I).Overall, these results depict 3′ modifications as a dynamic feature of the miRNAexpression landscape, whose abundance can vary substantially between differentmiRNAs and cell types. While their functional consequences are not clear, bothtrimming of and non-templated base additions to mature miRNAs are thought toaffect stability and extent of target repression, and modifications to pre-miRNAshave been associated with positive and negative regulation of processing by Dicer[27]. These differences could thus be integral to the coordinated changes in geneand miRNA expression required for differentiation to occur.149hsa-miR-101-3p hsa-miR-142-5phsa-miR-142-3p100750255010075025501007502550Percent of expressedPercent of expressedCell populationMLPCMPGMPMEPMonocytesErythroidBTCD34+CD34+38-49fMPP-2 -1 0 1 25’ cut positionNot expressedFigure 4.20: miRNAs with significantly different relative 5′ cut locations.Stacked bar charts depict the proportion of each isoform expressed in eachcell. Changes in cut location resulting in an upstream shift are denoted ingreen, and those resulting in a downstream shift are denoted in pink. Progen-itor cells are ordered based on their position along the pseudotime differenti-ation axis; the top colour bars indicate cell population.150hsa-let-7a-5p hsa-let-7f-5phsa-let-7g-5p hsa-miR-101-3phsa-miR-103a-3p hsa-miR-106b-3phsa-miR-125a-5p hsa-miR-140-3phsa-miR-142-5p hsa-miR-142-3phsa-miR-146b-5p hsa-miR-155-5phsa-miR-15a-5p hsa-miR-15b-5phsa-miR-17-5p hsa-miR-181a-3phsa-miR-181a-5p hsa-miR-181a-2-3p100750255010075025501007502550100750255010075025501007502550100750255010075025501007502550100750255010075025501007502550100750255010075025501007502550100750255010075025501007502550Percent of expressedPercent of expressedCell populationMLPCMPGMPMEPMonocytesErythroidBTCD34+CD34+38-49fMPP-2 -1 0 1 23’ cut positionNot expressedFigure 4.21: miRNAs with significantly different relative 3′ cut locations (1of 3). Stacked bar charts depict the proportion of each isoform expressedin each cell. Changes in cut location resulting in an upstream shift are de-noted in green, and those resulting in a downstream shift are denoted in pink.Progenitor cells are ordered based on their position along the pseudotime dif-ferentiation axis; the top colour bars indicate cell population.151hsa-miR-221-3p hsa-miR-223-3phsa-miR-23a-3p hsa-miR-23b-3phsa-miR-24-3p hsa-miR-25-3phsa-miR-26a-5p hsa-miR-26b-5phsa-miR-27a-3p hsa-miR-29a-3p1007502550100750255010075025501007502550100750255010075025501007502550100750255010075025501007502550Percent of expressedPercent of expressedCell populationMLPCMPGMPMEPMonocytesErythroidBTCD34+CD34+38-49fMPP-2 -1 0 1 23’ cut positionNot expressedhsa-miR-191-5p hsa-miR-196b-5phsa-miR-19a-3p hsa-miR-19b-3phsa-miR-20a-5p hsa-miR-21-5p100750255010075025501007502550100750255010075025501007502550hsa-miR-181b-5p hsa-miR-186-5p10075025501007502550Figure 4.22: miRNAs with significantly different relative 3′ cut locations (2of 3). Stacked bar charts depict the proportion of each isoform expressedin each cell. Changes in cut location resulting in an upstream shift are de-noted in green, and those resulting in a downstream shift are denoted in pink.Progenitor cells are ordered based on their position along the pseudotime dif-ferentiation axis; the top colour bars indicate cell population.152hsa-miR-30e-3p hsa-miR-30e-5phsa-miR-320a hsa-miR-338-3phsa-miR-342-3p hsa-miR-361-5phsa-miR-374a-5p hsa-miR-374a-3phsa-miR-425-5p hsa-miR-92a-3phsa-miR-93-5p10075025501007502550100750255010075025501007502550100750255010075025501007502550100750255010075025501007502550Percent of expressedPercent of expressedCell populationMLPCMPGMPMEPMonocytesErythroidBTCD34+CD34+38-49fMPP-2 -1 0 1 23’ cut positionNot expressedhsa-miR-29b-3p hsa-miR-29c-3phsa-miR-30c-5p hsa-miR-30d-5p1007502550100750255010075025501007502550Figure 4.23: miRNAs with significantly different relative 3′ cut locations (3of 3). Stacked bar charts depict the proportion of each isoform expressedin each cell. Changes in cut location resulting in an upstream shift are de-noted in green, and those resulting in a downstream shift are denoted in pink.Progenitor cells are ordered based on their position along the pseudotime dif-ferentiation axis; the top colour bars indicate cell population.153hsa-miR-101-3p hsa-miR-103a-3phsa-miR-140-3p hsa-miR-16-5phsa-miR-17-5p hsa-miR-181a-3phsa-miR-181a-5p hsa-miR-181a-2-3phsa-miR-196b-5p hsa-miR-197-3phsa-miR-19b-3p hsa-miR-221-3phsa-miR-223-3p hsa-miR-23a-3phsa-miR-23b-3p hsa-miR-24-3phsa-miR-25-3p hsa-miR-26a-5p100750255010075025501007502550100750255010075025501007502550100750255010075025501007502550100750255010075025501007502550100750255010075025501007502550100750255010075025501007502550Percent of expressedPercent of expressedCell populationMLPCMPGMPMEPMonocytesErythroidBTCD34+CD34+38-49fMPP3’ non-templated base additionsNA A U AAAUNot expressedFigure 4.24: miRNAs with significantly different 3′ non-templated base ad-ditions (1 of 2). Stacked bar charts depict the proportion of each isoformexpressed in each cell. Single-base additions are denoted in green and two-base additions in pink. Progenitor cells are ordered based on their positionalong the pseudotime differentiation axis; the top colour bars indicate cellpopulation.154hsa-miR-29c-3p hsa-miR-30d-5phsa-miR-30e-3p hsa-miR-30e-5phsa-miR-361-3p hsa-miR-361-5phsa-miR-92a-3p hsa-miR-93-5p10075025501007502550100750255010075025501007502550100750255010075025501007502550Percent of expressedPercent of expressedCell populationMLPCMPGMPMEPMonocytesErythroidBTCD34+CD34+38-49fMPP3’ non-templated base additionsNA A U AAAUNot expressedhsa-miR-27a-3p hsa-miR-29b-3p10075025501007502550Figure 4.25: miRNAs with significantly different 3′ non-templated base ad-ditions (2 of 2). Stacked bar charts depict the proportion of each isoformexpressed in each cell. Single-base additions are denoted in green and two-base additions in pink. Progenitor cells are ordered based on their positionalong the pseudotime differentiation axis; the top colour bars indicate cellpopulation.1554.4 DiscussionIn this chapter I demonstrate a robust and sensitive method that enables miRNA se-quencing libraries to be generated reproducibly from large numbers of single cells.A key component was the development of reaction conditions that substantiallyincrease capture efficiency while maintaining RNA integrity. These advances areimmediately transferable to the plate-based analysis of samples containing hun-dreds of cells, and we anticipate that incorporating strategies to further reduce by-products of library construction will permit single-cell sensitivity. Slight modifi-cations to the construct design could also allow adaptation to other small-volumeformats such as droplet microfluidics [137] or microscale arrays [130]. Althoughwe chose to focus on the miRNA fraction of small-RNAs, we demonstrated thatother classes such as snoRNAs can be interrogated through trivial, concomitantadjustments to the size enrichment and sequencing length used (Figure 4.9B-C).We then used our method to provide the first comprehensive analysis of miRNAexpression in single cells isolated from the most primitive type of human hematopoi-etic cell thus far identified through to four types of mature blood cells. The miRNAdiversity seen was substantially smaller (5-10×) than what is typically observed inmRNA measurements, yet the expression profiles contained sufficient power to dis-criminate and structure major cell types. We observed that MLPs were out of placeas would have been predicted based on their gene expression profile [265] anddevelopmental potential [159]. Instead of localizing with the multipotent progeni-tors downstream of the MPPs, they were instead closest to the GMPs, in alignmentwith recent findings suggesting that GMPs are more closely related to MLPs than toerythroid progenitors [266]. Importantly, an unsupervised analysis of the miRNAexpression profiles revealed a continuous process of HSC change with no evidenceof discrete co-ordinated transitions, in agreement with molecular [165, 166] andfunctional [159, 164] studies that have brought the paradigm of a strict differentia-tion hierarchy into recent question.Our unsupervised analysis also described a dynamic expression landscape start-ing with pervasive miRNA expression within the HSCs, a global down-regulationduring differentiation, and an abrupt commitment to an expression signature char-acteristic of each effector cell type comprised of combinatorial expression dif-156ferences. These characteristics, combined with a surprising lack of evidence ofan early lineage commitment step may explain previous seemingly contradictoryobservations of miRNAs influencing multiple hematopoietic lineages. They alsosuggest that networks of miRNAs act closely with specific transcription factorstowards effecting ultimate cell fate decisions.While there is generally a limited understanding of miRNA-gene interactions,the set of HSC-enriched miRNAs we identified contained several examples of in-teractions with Homeobox (HOX) genes known to regulate hematopoietic devel-opment [267]. These include miR-10a and miR-196 that are embedded withinthe HOX gene clusters in both human and mouse genomes and preferentially tar-get HOX mRNAs [268, 269]; the tricistronic cluster containing miR-125b, let-7cand miR-99a is activated by HOXA10 [33]; miR-126 has been shown to targetmurine Hoxa9 [270]; and numerous additional predicted interactions have yet tobe tested. These observations underscore the importance of further delineation ofHOX-miRNA interaction networks within hematopoiesis [101].Our characterization of existing and unannotated 5′ isoforms as highly ex-pressed and locus-intrinsic proposes an additional feature of miRNA-target inter-action networks. Precise processing at the 5′ end has been previously explainedto be under evolutionary pressure to minimize off-target effects [54]. However, asa substantial proportion of miRNAs occur in polycistronic clusters in the genome[26], and the miRNAs from these clusters have been seen to work together towardsgenerating an integrated phenotype [33, 34], the potential regulatory and evolu-tionary advantages to generating multiple, tightly co-expressed miRNAs from thesame locus should not be overlooked. It is tempting to suggest that miRNAs thatgenerate multiple 5′ isomiRs may share this feature. Similarly, the widespreadand varied miRNA-specific 3′ modification profiles seen here reiterates the impor-tance of future work towards determining the functional consequences of this classof miRNA isoforms. Evidence of multiple independent processes for generatingtightly co-expressed miRNAs with different targeting repertoires and effective de-grees of stability and target repression adds an interesting layer to the regulationcapabilities of miRNAs, especially when considered at a systems-level view. De-lineating these nuances in miRNA-mRNA interaction networks is likely to lead toa refined understanding of how miRNAs are involved in disease development or in157regulating cell fate decisions.Our measurements of miRNA expression during human hematopoiesis providean ideal basis on which to both integrate previously disparate miRNA-gene inter-action observations and identify promising leads for further interrogation. Fur-thermore, the miRNA expression intricacy seen here suggests that a more nuancedapproach to experimental design may need to be taken in order to elucidate thefunctional effects of specific miRNAs and the need for systems-level analyses infuture investigations of the role miRNAs play in normal and deviant tissue be-haviour.Genome-wide, single-cell expression analysis techniques are ideally suited tothe unbiased classification of cell-types, characterization of transition states andidentification of gene regulatory modules. There is a growing appreciation forthe critical role that miRNAs play in determining the topology of these systems.miRLin enables the collection of high-quality genome-wide single-cell miRNAexpression measurements.158Chapter 5Conclusion5.1 Summary of contributionsIn this dissertation, I set out to address a need for scalable techniques to measuremiRNA expression profiles in single cells. Towards this goal, a suite of microflu-idic tools was developed whose capabilities range from the highly precise mea-surement of one to two assays in up to 300 cells, to the genome-wide measurementof miRNA expression in up to 96 cells per experiment. The main contributionscan be summarized in terms of technical advancements in microfluidic analysis ofsingle-cells, library construction methods for single-cell miRNA-seq, and analysisof miRNA expression dynamics during human hematopoiesis.5.1.1 Microfluidic architecture for single-cell expression analysisThe complete integration of single-cell manipulation with downstream analyticalprocedures is a critical step towards scalable, precise analysis. While several ofthe individual pieces of these multi-step methods had been previously presented,including cell isolation and microfluidic PCR, their complete integration had not.A microfluidic architecture that integrated single-cell capture with different work-flows for measuring transcript abundance was developed. The cell trap was a criti-cal component of this successful integration.Cell trap designs were extensively optimized to increase capture efficiency.Initial designs consisting of a single cup structure [143] in the middle of a chan-159nel were extremely inefficient, resulting in cell occupancy much less than couldhave been achieved using simple stochastic partitioning. The simple addition ofupstream deflectors to position cells into the appropriate streamlines substantiallyincreased the occupancy (Figure 2.2). The capture efficiency, defined as the chancethat a given cell is captured when it encounters a trap, however, was too low forapplications with limited cell inputs. Furthermore, the geometry completely failedto capture the smaller cells in primary hematopoietic populations (∼ 8 µm diame-ter). The cup shape and size, deflector size, and distance between the deflector andcup were all subsequently optimized for single-cell capture efficiency and compat-ibility with smaller cells. The optimized traps use a larger deflector section, have asmaller cup, and the spacing between the deflector and the cup is adjusted to matchthe cell size; this geometry is used in Chapter 3 (Figure 3.3) and Chapter 4 (Fig-ure 4.5). In practice, we found that a “large” (∼ 12 to 18 µm diameter) and “small”(∼ 6 to 10 µm diameter) trap were sufficient to cover the range of cell sizes thatwere encountered.While the effect of cell shape was not explicitly examined, it is not believed tohave a significant influence on cell capture. As use of these microfluidic devicesnecessitates generating a single-cell suspension prior to cell loading, cells typically“ball up” and have roughly the same shape. Internally, in addition to the cell typespresented in this dissertation, we have used a variety of adherent (CA1S hESCs,184-hTERT-L2 breast epithelial, BJ fibroblast) and suspension (GM18507 lym-phoblastoid) cell lines, and primary cells (breast cancer xenograft, plural effusionof a metastatic breast cancer).One of the largest obstacles to consistent cell loading, especially for adher-ent and primary cells, is managing debris and cell clumps. Because the cell trapsare arrayed in serial and contain narrow constrictions typically on the order of thediameter of the cells being loaded, they can easily clog and foul the cell-loadingchannels. Several precautions were routinely taken in order to reduce these clogs.During chip fabrication, access ports for cell-loading channels were punched withfresh, sharp bits to decrease the amount of PDMS debris. During sample prepa-ration, adherent cells were passed through a cell strainer after trypsinization, andsorted cells were deposited into sterile-filtered medium. Finally, some devices in-tegrate on-chip cell filters upstream of the cell traps.160Similar to previous reports, microfluidic integration resulted in improved reac-tion performance at single-cell quantities. In the RT-qPCR implementations pre-sented in Chapter 2 and Chapter 3, measurement precision approached the limit forqPCR and sensitivity was at the limit of a single molecule; these metrics outper-form those reported for benchtop single-cell RT-qPCR (Figure 2.3) performed byus and others (Table 3.1, Table 3.2). Furthermore, the quality metrics for the mi-crofluidic single-cell miRNA-seq libraries generated in Chapter 4 vastly surpassedthose from the only other reported method for single-cell miRNA-seq (Figure 4.9).While there were several differences between these two protocols, the reduction inreaction side products (e.g., adapter-dimers) is a direct and necessary consequenceof decreased reaction volumes. The work presented in this dissertation providesfurther evidence demonstrating that nanolitre volumes can enhance single-cell as-say performance.The core functionality established in Chapter 2 provided the foundation fromwhich a variety of microfluidic devices for single-cell analysis were developed byus [114, 115, 219, 271] and others [145, 272–275]. Once the set of rules around de-signing microfluidic systems for single-cell nucleic acid analysis were established,they proved to be robust to different assays. For example, the cell-processing unitdesign for the multiplexed RT-qPCR and miRNA-seq devices presented in Chap-ter 3 and Chapter 4 remained relatively unchanged since their initial conception.Each device went through only four minor design iterations to address fluid-routingissues. Furthermore, the simplicity of device operation allowed for the basic mi-crofluidic architecture to be integrated as part of the automated Fluidigm C1 in-strument.5.1.2 Library preparation procedure for single-cell miRNA-seqExisting methods for small-RNA library preparation were insufficient for single-cell analysis. The number of successive enzymatic steps required results in a com-pounding of even modest inefficiencies, thereby reducing the overall sensitivity. Inparticular, the initial steps of single-stranded RNA ligations are notoriously inef-ficient. As a result, standard miRNA-seq library preparation protocols typicallyrequire an input of 100 to 1000 ng of purified RNA; this is 5,000 to 50,000 times161more than is available in a typical single cell. In response, we performed extensiveoptimization experiments, systematically testing lysis conditions, titrating adapterconcentrations, adding molecular crowding agents, testing different RNA ligases,reverse transcriptases, and RNase inhibitors, and introducing strategies to reduceadapter-dimers.Proper cell lysis conditions were found to be critical in achieving high li-brary qualities. Early tests of those commonly used in single-cell mRNA-seqmethods yielded library size profiles and alignment metrics characteristic of ex-tensive RNA degradation. It is likely that similar effects are occurring in mRNA li-braries, but they are masked by the inability to perform the quality control steps thatare typically used prior to constructing bulk transcriptome libraries. As the fieldof single-cell genomics matures and moves towards multi-institution consortium-level projects [276], I believe that the importance of cell preparation [277] andRNA degradation will be increasingly appreciated. The lysis conditions developedas part of this work are likely transferable to single-cell mRNA-seq libraries.Our single-cell miRNA-seq protocol substantially increased the yield on lim-ited samples and removed the requirement to use purified RNA as an input material.In our validation experiments, we showed that the library qualities met or exceededthose prepared using high amounts of input RNA, that measurements were quanti-tative and reproducible, and that the sensitivity limit was sufficient to obtain mean-ingful single-cell expression profiles. I believe that the technology developed herewill enable the routine analysis of miRNA expression in single cells.5.1.3 Single-cell microRNA sequencing of the human hematopoieticcell hierarchyMolecular characterization of the populations that make up the classic hematopoi-etic development hierarchy has discovered several miRNAs involved in cell func-tion, self-renewal, and lineage choice. However, these data are incomplete, asmany of the most primitive rare populations have been inaccessible to genome-wide techniques. Furthermore, there is increasing evidence in support of alternatemodels of differentiation, potentially confounding the measured molecular signa-tures with data from mixed populations. We thus applied our method for gener-ating single-cell miRNA-seq libraries to provide a comprehensive look at miRNA162expression in human hematopoiesis.We measured 2057 single cells isolated from 12 phenotypically sorted pop-ulations encompassing the gamut of HSPC cells and the major mature lineages.An unbiased analysis of the population structure as determined by the miRNA ex-pression profiles indicated a smooth transition as HSCs differentiate into lineage-restricted progenitors and showed substantial mixing between the phenotypic pop-ulations. These observations run contrary to the classic model of stepwise progres-sion through increasingly restricted progenitor intermediates.Recent functional and molecular studies proposed a model of differentiationwhere HSCs undergo smooth transitions during which unilineage potential is grad-ually acquired. Our miRNA-seq data supports this continuous linear progressionfrom HSCs to committed progenitors. However, in contrast to single-cell RNA-seq data [167], we did not observe widespread evidence of lineage commitmentin our single-cell miRNA-seq profiles from the committed progenitor populations.The only clear evidence of lineage commitment was the expression of miR-486-5p, which is located in an intron of a gene required for erythropoiesis (ANK1);these seemingly contradictory observations may suggest that miRNAs may playmore of a role in reinforcing hematopoietic cell types, rather than driving cell-fatedecisions. Methods to simultaneously measure both mRNA and miRNA expres-sion within the same single cell would enable high-resolution measurements of therelative expression dynamics between the mRNA and miRNA required for thesedevelopmental processes, providing insight into their respective roles.We also analyzed coordinated changes in miRNA expression. Hierarchicalclustering of the miRNA expression profiles identified a set of miRNAs whose ex-pression was highest within the HSCs and decreased during differentiation. Manymembers of this set had been previously implicated in regulating HSC self-renewaland quiescence, including miR-126-3p, the expression of which had been shown toprospectively enrich for HSCs. In addition to integrating these previously disparatestudies, we identified additional miRNAs that share this expression pattern. Simi-larly, several of the miRNAs that we found to be differentially expressed betweenthe mature populations had also been corroborated by previous studies. For exam-ple, miR-451a was seen to be exclusively expressed within the erythroid cells, andmiRs-150-5p and 223-3p were upregulated in the mature lymphoid and myeloid163populations, respectively. Collecting genome-wide expression data throughout theprogenitor maturation process, however, showed that the expression of the majorityof these miRNAs is not restricted to these specific cell types, nor are they sufficientto define them.Finally, we characterized the expression of miRNA isoforms. During differen-tiation, the expression of 5′ isomiRs was proportional to the total expression fromtheir host miRNA, suggesting that they are generated intrinsically. This stable gen-eration of multiple miRNAs from the same locus is reminiscent of polycistronicclusters, of which there is accumulating evidence suggesting that their members ex-ert concerted effort towards establishing a phenotype. In contrast, the proportionalexpression of the 3′ isomiRs was seen to change during HSPC differentiation andbetween mature cell types. Furthermore, the expression trends were not the samebetween different miRNAs, suggesting that they may be dynamically regulated. As3′ modifications are thought to affect stability, these widespread modifications maybe integral to coordinating changes in miRNA expression.In general, our findings establish the ability of this technology to obtain high-quality single-cell miRNA-seq data from complex tissues, forecasting its routineuse in single-cell analysis. Our measurements of miRNA expression during hemato-poiesis provide an integral resource on which to build miRNA-gene interactionnetworks and identify promising leads for further investigation.5.2 Future recommendations5.2.1 Technology developmentWhile microfluidic chips have played an important part in the development ofminiaturized systems for single-cell analysis, they have seen modest adoption out-side of specialized academic groups. One reason for this could be that designing,fabricating, and testing these devices requires a relatively rare skillset and accessto specialized fabrication equipment. Efforts have gone into “democratizing” thesesystems by creating programmable chips [139, 271], providing commercially avail-able solutions, and encouraging the development of user-defined protocols [145],but these have done little to expand their acceptance. This may change as protocols164for single-cell genomics continue to develop, increase in complexity, and requiremore sophisticated solutions. Another reason for the limited adoption could bethat the number of analyzed cells per device is much lower than what the single-cell genomics community is currently demanding. The Human Cell Atlas, a projectto catalogue the different cell types and states in the body, has a goal to profile atleast 10 billion cells [276]; it is unlikely that this number will be ever be reachedat less than 100 cells per experiment.Adapting the single-cell miRNA-seq protocol developed as part of this work toother microfluidic platforms is likely to be beneficial towards increasing its adop-tion. The protocol presented here could be directly transferred to open microscalearrays with micro-capillary dispensers, or with slight modifications to systems thatuse piezoelectric droplet dispensers. Implementation in the former could enable thedirect sorting and profiling of ultra-rare cell types such as circulating tumour cells,whereas the latter would result in thousands of cells to be analyzed per run. Sim-ilarly, modifying the adapter sequence to barcode cells during the initial ligationstep could enable the protocol’s implementation in droplet-based methods similarto drop-seq [137], drastically increasing the cell throughput. Finally, there couldalso be benefits to implementing the method in standard well-plates. As the de factostandard for molecular assays, adaptation to the well-plate would allow seamlessinterfacing with other techniques such as index sorting [278], laser-capture mi-crodissection, or tomo-seq [279]. The overwhelming abundance of adapter-dimersis likely the largest hurdle towards scaling the reaction volumes up; additional tech-niques for preventing adapter-dimers or removing them from the reaction will beneeded.miRNA expression profiling would be a valuable part of multi-omics meth-ods for single-cell analysis. As miRNAs exert their effects through wide-reachingregulatory networks, their expression profiles could accurately describe the typeand state of a cell; this discriminatory power has been recognized in classifyingcancer and cell types [8, 126]. Combing molecular phenotype classifications withwhole-genome sequence information could, for example, enable the identificationof the subpopulations that are ultimately responsible for cancer maintenance andresistance. Similar to existing methods for combined genome and transcriptomesequencing [280, 281], miRNAs could be first captured on beads using a ligation165step, then physically separated from the genome prior to parallel library construc-tion on the DNA and miRNA. Alternatively, the specificities of the Tn5 transposaseand RNA ligase could be exploited to perform library construction within the sametube as was done with the recent Simul-seq method [282]. Collecting miRNAand mRNA expression profiles from the same cell could, along with appropriatebioinformatics analysis techniques, begin to assemble miRNA:mRNA interactionnetworks. The most straightforward way to collect these complete transcriptomeswould be to introduce an RNA fragmentation and end repair step prior to ligation,and a ribosomal RNA depletion step prior to sequencing. Alternatively, similarto a technique for 5′ RACE [283], mRNA decapping could be used, followed byadapter ligation and poly-T priming, to specifically capture the mRNA fraction.5.2.2 MicroRNAs in developmentOur analysis of the miRNA expression dynamics during hematopoiesis raised sev-eral potential areas for future research.First, the HOX-miRNA interaction networks and their influence on hematopoi-esis warrant further investigation. The set of HSC-enriched miRNAs that we iden-tified contained several examples of known and putative interactions with HOXgenes. For example, miR-196b is located upstream of HoxA9 and preferentiallytargets HOX genes; the miR-125b∼99a tricistron is activated by HoxA10; miR-126 targets HoxA9; and miR-10a is located upstream of and is co-expressed withHoxB4, and has been shown to preferentially target other HOX genes [268]. It isworth noting that we found that miR-10a-5p was the most highly expressed andsharply downregulated member of the HSC-enriched set, and that it also robustlygenerated a highly expressed seed-shifted isoform. As the Hox gene family hasbeen linked to HSC self-renewal and leukemic transformation [267], it is likelythat miRNA interactions play an important role in establishing and regulating thesefunctions.Second, more work needs to be done to elucidate the mechanisms responsi-ble for and the function of 3′ modifications on miRNAs. Previous studies havefound that imprecise processing, trimming, or non-templated base additions affectthe miRNA stability [22, 27]. These studies, however, are largely based on ex-166amination of specific miRNAs. The widespread and miRNA-specific modificationpatterns seen here suggest that caution be taken in extrapolating these studies tothe prediction of both the functional outcomes of, and the mechanisms responsiblefor, a specific modification. Further investigations will be needed to reveal howtrimming, tailing, and aberrant processing contribute to miRNA regulation.Finally, there are opportunities to better describe the role that miRNAs play inregulating cell-state transitions. For example, in contrast to the prevailing modelof lineage-specifying miRNAs, strong evidence of lineage commitment was notobserved in the miRNA expression profiles obtained during HSC differentiation.However, coordinated decreases in expression across several miRNAs were ob-served. In combination with recent work that described miRNA’s control of proteinexpression noise [38] and the impact of expression noise during lineage decisions[284], a model could be proposed where miRNAs indirectly alter lineage choiceby modulating stochastically expressed regulators.5.3 Concluding remarksDespite single cells being a fundamental unit of living organisms, most of what isknown about their molecular function is derived by studying the average of largeensembles of many cells. This is almost entirely due to technical constraints im-posed by genome-wide profiling techniques. The field of single-cell genomics hasmade enormous progress in recent years towards establishing methods to profilethe genomes, epigenomes, and transcriptomes in hundreds to thousands of singlecells. As these methods mature and are increasingly adopted, there will be an ac-celeration of efforts in cataloguing cell types in tissue atlases, examining dynamicsystems, and identifying gene regulatory networks.This dissertation presents important contributions to both of these themes. Thefoundational microfluidic components demonstrated in Chapter 2 have spurred thedevelopment of a variety of academic and commercially available solutions forsingle-cell analysis. This integrated small-volume processing facilitated the cre-ation of one of the first methods to generate high-quality miRNA sequencing li-braries from single cells. Finally, these unique capabilities enabled the first high-resolution single-cell analysis of miRNA expression dynamics as human HSCs167differentiate. These resources will continue to be of interest to researchers inthe rapidly growing miRNA, single-cell genomics, and developmental biology re-search communities.168Bibliography[1] ENCODE Project Consortium. An integrated encyclopedia of DNAelements in the human genome. Nature, 489(7414):57–74, 2012.doi:10.1038/nature11247. → pages 1, 119, 124, 128[2] M. Schena, D. Shalon, R. Heller, A. Chai, P. O. Brown, and R. W. Davis.Parallel human genome analysis: microarray-based expression monitoringof 1000 genes. Proceedings of the National Academy of Sciences, 93(20):10614–10619, 1996. → page 1[3] D. D. Shoemaker, E. E. Schadt, C. D. Armour, Y. D. He, P. Garrett-Engele,P. D. McDonagh, P. M. Loerch, A. Leonardson, P. Y. Lum, G. Cavet, L. F.Wu, S. J. Altschuler, S. Edwards, J. King, J. S. Tsang, G. Schimmack, J. M.Schelter, J. Koch, M. Ziman, M. J. Marton, B. Li, P. Cundiff, T. Ward,J. Castle, M. Krolewski, M. R. Meyer, M. Mao, J. Burchard, M. J. Kidd,H. Dai, J. W. Phillips, P. S. Linsley, R. Stoughton, S. Scherer, and M. S.Boguski. Experimental annotation of the human genome using microarraytechnology. Nature, 409(6822):922–927, 2001. doi:10.1038/35057141. →page 1[4] Aravind Subramanian, Pablo Tamayo, Vamsi K. Mootha, SayanMukherjee, Benjamin L. Ebert, Michael A. Gillette, Amanda Paulovich,Scott L. Pomeroy, Todd R. Golub, Eric S. Lander, and Jill P. Mesirov. Geneset enrichment analysis: A knowledge-based approach for interpretinggenome-wide expression profiles. Proceedings of the National Academy ofSciences, 102(43):15545–15550, 2005. doi:10.1073/pnas.0506580102. →page 1[5] B. Zhang and S. Horvath. A general framework for weighted geneco-expression network analysis. Statistical Applications in Genetics andMolecular Biology, 4(1):Article 17, 2005. doi:10.2202/1544-6115.1128.→ page 1169[6] Amir Ben-Dor, Laurakay Bruhn, Nir Friedman, Iftach Nachman, Miche`lSchummer, and Zohar Yakhini. Tissue classification with gene expressionprofiles. Journal of Computational Biology, 7(3-4):559–83, 2000.doi:10.1089/106652700750050943. → page 1[7] Ash A. Alizadeh, Michael B. Eisen, R. Eric Davis, Chi Ma, Izidore S.Lossos, Andreas Rosenwald, Jennifer C. Boldrick, Hajeer Sabet, Truc Tran,Xin Yu, John I. Powell, Liming Yang, Gerald E. Marti, Troy Moore, JamesHudson, Lisheng Lu, David B. Lewis, Robert Tibshirani, Gavin Sherlock,Wing C. Chan, Timothy C. Greiner, Dennis D. Weisenburger, James O.Armitage, Roger Warnke, Ronald Levy, Wyndham Wilson, Michael R.Grever, John C. Byrd, David Botstein, Patrick O. Brown, and Louis M.Staudt. Distinct types of diffuse large B-cell lymphoma identified by geneexpression profiling. Nature, 403(6769):503–511, 2000.doi:10.1038/35000501. → page 1[8] Jun Lu, Gad Getz, Eric A. Miska, Ezequiel Alvarez-Saavedra, Justin Lamb,David Peck, Alejandro Sweet-Cordero, Benjamin L. Ebert, Raymond H.Mak, Adolfo A. Ferrando, James R. Downing, Tyler Jacks, H. RobertHorvitz, and Todd R. Golub. microRNA expression profiles classify humancancers. Nature, 435(7043):834–838, 2005. doi:10.1038/nature03702. →pages 1, 6, 63, 98, 106, 165[9] Joseph A. Sparano, Robert J. Gray, Della F. Makower, Kathleen I.Pritchard, Kathy S. Albain, Daniel F. Hayes, Charles E. Jr. Geyer,Elizabeth C. Dees, Edith A. Perez, John A. Jr. Olson, JoAnne Zujewski,Tracy Lively, Sunil S. Badve, Thomas J. Saphner, Lynne I. Wagner,Timothy J. Whelan, Matthew J. Ellis, Soonmyung Paik, William C. Wood,Peter Ravdin, Maccon M. Keane, Henry L. Gomez Moreno, Pavan S.Reddy, Timothy F. Goggins, Ingrid A. Mayer, Adam M. Brufsky,Deborah L. Toppmeyer, Virginia G. Kaklamani, James N. Atkins, Jeffrey L.Berenberg, and George W. Sledge. Prospective validation of a 21-geneexpression assay in breast cancer. New England Journal of Medicine, 373(21):2005–2014, 2015. doi:10.1056/NEJMoa1510764. → page 1[10] Laura J. van ’t Veer, Hongyue Dai, Marc J. van de Vijver, Yudong D. He,Augustinus A. M. Hart, Mao Mao, Hans L. Peterse, Karin van der Kooy,Matthew J. Marton, Anke T. Witteveen, George J. Schreiber, Ron M.Kerkhoven, Chris Roberts, Peter S. Linsley, Rene Bernards, and Stephen H.Friend. Gene expression profiling predicts clinical outcome of breast170cancer. Nature, 415(6871):530–536, 2002. doi:10.1038/415530a. → page1[11] C. A. Heid, J. Stevens, K. J. Livak, and P. M. Williams. Real timequantitative PCR. Genome Research, 6(10):986–994, 1996.doi:10.1101/gr.6.10.986. → pages 2, 17[12] Marisa L. Wong and Juan F. Medrano. Real-time PCR for mRNAquantitation. Biotechniques, 39(1):75–88, 2005. → pages 2, 17[13] Jorg D. Hoheisel. Microarray technology: beyond transcript profiling andgenotype analysis. Nature Reviews Genetics, 7(3):200–210, 2006.doi:10.1038/nrg1809. → page 2[14] Mark Schena, Dari Shalon, Ronald W. Davis, and Patrick O. Brown.Quantitative monitoring of gene expression patterns with a complementaryDNA microarray. Science, 270(5235):467–470, 1995.doi:10.1126/science.270.5235.467. → page 2[15] Ali Mortazavi, Brian A. Williams, Kenneth McCue, Lorian Schaeffer, andBarbara Wold. Mapping and quantifying mammalian transcriptomes byRNA-seq. Nature Methods, 5(7):621–628, 2008. doi:10.1038/nmeth.1226.→ page 2[16] Zhong Wang, Mark Gerstein, and Michael Snyder. RNA-seq: arevolutionary tool for transcriptomics. Nature Reviews Genetics, 10(1):57–63, 2009. doi:10.1038/nrg2484. → pages 2, 20[17] Rosalind C. Lee, Rhonda L. Feinbaum, and Victor Ambros. The C. elegansheterochronic gene lin-4 encodes small RNAs with antisensecomplementarity to lin-14. Cell, 75(5):843–854, 1993.doi:10.1016/0092-8674(93)90529-Y. → page 2[18] Bruce Wightman, Ilho Ha, and Gary Ruvkun. Posttranscriptionalregulation of the heterochronic gene lin-14 by lin-4 mediates temporalpattern formation in C. elegans. Cell, 75(5):855–862, 1993.doi:10.1016/0092-8674(93)90530-4. → page 2[19] Victor Ambros. The evolution of our thinking about microRNAs. NatureMedicine, 14(10):1036–1040, 2008. doi:10.1038/nm1008-1036. → pages2, 135171[20] Brenda J. Reinhart, Frank J. Slack, Michael Basson, Amy E. Pasquinelli,Jill C. Bettinger, Ann E. Rougvie, H. Robert Horvitz, and Gary Ruvkun.The 21-nucleotide let-7 RNA regulates developmental timing inCaenorhabditis elegans. Nature, 403(6772):901–906, 2000.doi:10.1038/35002607. → page 2[21] Ana Kozomara and Sam Griffiths-Jones. miRBase: annotating highconfidence microRNAs using deep sequencing data. Nucleic AcidsResearch, 42(D1):D68–D73, 2014. doi:10.1093/nar/gkt1181. → pages2, 6, 117[22] Minju Ha and V. Narry Kim. Regulation of microRNA biogenesis. NatureReviews Molecular Cell Biology, 15(8):509–524, 2014.doi:10.1038/nrm3838. → pages 3, 9, 166[23] Stefanie Jonas and Elisa Izaurralde. Towards a molecular understanding ofmicroRNA-mediated gene silencing. Nature Reviews Genetics, 16(7):421–433, 2015. doi:10.1038/nrg3965. → pages 3, 98, 105[24] Alex S. Flynt and Eric C. Lai. Biological principles of microRNA-mediatedregulation: shared themes amid diversity. Nature Reviews Genetics, 9(11):831–842, 2008. doi:10.1038/nrg2455. → pages 3, 5[25] Wigard P. Kloosterman and Ronald H. A. Plasterk. The diverse functionsof microRNAs in animal development and disease. Developmental Cell, 11(4):441–450, 2006. doi:10.1016/j.devcel.2006.09.009. → pages 3, 5[26] Yael Altuvia, Pablo Landgraf, Gila Lithwick, Naama Elefant, Se´bastienPfeffer, Alexei Aravin, Michael J. Brownstein, Thomas Tuschl, and HanahMargalit. Clustering and conservation patterns of human microRNAs.Nucleic Acids Research, 33(8):2697–2706, 2005. doi:10.1093/nar/gki567.→ pages 3, 157[27] Stefan L. Ameres and Phillip D. Zamore. Diversifying microRNAsequence and function. Nature Reviews Molecular Cell Biology, 14(8):475–488, 2013. doi:10.1038/nrm3611. → pages 5, 7, 8, 9, 145, 149, 166[28] David P. Bartel. microRNAs: Target recognition and regulatory functions.Cell, 136(2):215–233, 2009. doi:10.1016/j.cell.2009.01.002. → pages 3, 5[29] Benjamin P. Lewis, I. Hung Shih, Matthew W. Jones-Rhoades, David P.Bartel, and Christopher B. Burge. Prediction of mammalian microRNAtargets. Cell, 115(7):787–798, 2003.doi:10.1016/S0092-8674(03)01018-3. → page 3172[30] Nicole T. Schirle, Jessica Sheu-Gruttadauria, and Ian J. MacRae. Structuralbasis for microRNA targeting. Science, 346(6209):608–613, 2014.doi:10.1126/science.1258040. → page 3[31] Cameron P. Bracken, Hamish S. Scott, and Gregory J. Goodall. Anetwork-biology perspective of microRNA function and dysfunction incancer. Nature Reviews Genetics, 17(12):719–732, 2016.doi:10.1038/nrg.2016.134. → page 5[32] Benjamin P. Lewis, Christopher B. Burge, and David P. Bartel. Conservedseed pairing, often flanked by adenosines, indicates that thousands ofhuman genes are microRNA targets. Cell, 120(1):15–20, 2005.doi:10.1016/j.cell.2004.12.035. → page 5[33] Stephan Emmrich, Mareike Rasche, Jennifer Scho¨ning, Christina Reimer,Sarva Keihani, Aliaksandra Maroz, Ying Xie, Zhe Li, Axel Schambach,Dirk Reinhardt, and Jan-Henning Klusmann. miR-99a/100∼125btricistrons regulate hematopoietic stem and progenitor cell homeostasis byshifting the balance between TGFβ and Wnt signaling. Genes &Development, 28(8):858–874, 2014. doi:10.1101/gad.233791.113. →pages 5, 32, 140, 157[34] Andrea Ventura, Amanda G. Young, Monte M. Winslow, Laura Lintault,Alex Meissner, Stefan J. Erkeland, Jamie Newman, Roderick T. Bronson,Denise Crowley, James R. Stone, Rudolf Jaenisch, Phillip A. Sharp, andTyler Jacks. Targeted deletion reveals essential and overlapping functionsof the miR-17∼92 family of miRNA clusters. Cell, 132(5):875–886, 2008.doi:10.1016/j.cell.2008.02.019. → pages 5, 157[35] Eric A. Miska, Ezequiel Alvarez-Saavedra, Allison L. Abbott, Nelson C.Lau, Andrew B. Hellman, Shannon M. McGonagle, David P. Bartel,Victor R. Ambros, and H. Robert Horvitz. Most Caenorhabditis elegansmicroRNAs are individually not essential for development or viability.PLOS Genetics, 3(12):e215, 2007. doi:10.1371/journal.pgen.0030215. →page 5[36] Daehyun Baek, Judit Ville´n, Chanseok Shin, Fernando D. Camargo,Steven P. Gygi, and David P. Bartel. The impact of microRNAs on proteinoutput. Nature, 455(7209):64–71, 2008. doi:10.1038/nature07242. →pages 6, 10[37] Matthias Selbach, Bjo¨rn Schwanha¨usser, Nadine Thierfelder, Zhuo Fang,Raya Khanin, and Nikolaus Rajewsky. Widespread changes in protein173synthesis induced by microRNAs. Nature, 455(7209):58–63, 2008.doi:10.1038/nature07228. → pages 6, 10[38] Jo¨rn M. Schmiedel, Sandy L. Klemm, Yannan Zheng, Apratim Sahay, NilsBlu¨thgen, Debora S. Marks, and Alexander van Oudenaarden. microRNAcontrol of protein expression noise. Science, 348(6230):128–132, 2015.doi:10.1126/science.aaa1738. → pages 6, 15, 16, 105, 167[39] Kathryn N. Ivey and Deepak Srivastava. microRNAs as regulators ofdifferentiation and cell fate decisions. Cell Stem Cell, 7(1):36–41, 2010.doi:10.1016/j.stem.2010.06.012. → pages 6, 98, 106[40] Madhu S. Kumar, Jun Lu, Kim L. Mercer, Todd R. Golub, and Tyler Jacks.Impaired microRNA processing enhances cellular transformation andtumorigenesis. Nature Genetics, 39(5):673–677, 2007.doi:10.1038/ng2003. → pages 6, 98, 106[41] Robin C. Friedman, Kyle Kai-How Farh, Christopher B. Burge, andDavid P. Bartel. Most mammalian mRNAs are conserved targets ofmicroRNAs. Genome Research, 19(1):92–105, 2009.doi:10.1101/gr.082701.108. → page 6[42] Xi Chen, Yi Ba, Lijia Ma, Xing Cai, Yuan Yin, Kehui Wang, Jigang Guo,Yujing Zhang, Jiangning Chen, Xing Guo, Qibin Li, Xiaoying Li, WenjingWang, Yan Zhang, Jin Wang, Xueyuan Jiang, Yang Xiang, Chen Xu,Pingping Zheng, Juanbin Zhang, Ruiqiang Li, Hongjie Zhang, XiaobinShang, Ting Gong, Guang Ning, Jun Wang, Ke Zen, Junfeng Zhang, andChen-Yu Zhang. Characterization of microRNAs in serum: a novel class ofbiomarkers for diagnosis of cancer and other diseases. Cell Research, 18(10):997–1006, 2008. doi:10.1038/cr.2008.282. → pages 6, 106[43] Victor Ambros, Bonnie Bartel, David P. Bartel, Christopher B. Burge,James C. Carrington, Xuemi Chen, Gideon Dreyfuss, Sean R. Eddy, SamGriffiths-Jones, Mhairi Marshall, Marjori Matzke, Gary Ruvkun, andThomas Tuschl. A uniform system for microRNA annotation. RNA, 9(3):277–279, 2003. doi:10.1261/rna.2183803. → page 6[44] Katsutomo Okamura, Michael D. Phillips, David M. Tyler, Hong Duan,Yu-ting Chou, and Eric C. Lai. The regulatory activity of microRNA*species has substantial influence on microrna and 3′ UTR evolution. NatureStructural and Molecular Biology, 15:354, 2008. doi:10.1038/nsmb.1409.→ page 6174[45] Jr-Shiuan Yang, Michael D. Phillips, Doron Betel, Ping Mu, AndreaVentura, Adam C. Siepel, Kevin C. Chen, and Eric C. Lai. Widespreadregulatory activity of vertebrate microRNA* species. RNA, 17(2):312–326,2011. doi:10.1261/rna.2537911. → page 6[46] Sam Griffiths-Jones, Jerome H.L. Hui, Antonio Marco, and MatthewRonshaugen. MicroRNA evolution by arm switching. EMBO reports, 12(2):172–177, 2011. doi:10.1038/embor.2010.191. → page 6[47] Ryan D. Morin, Michael D. O’Connor, Malachi Griffith, FlorianKuchenbauer, Allen Delaney, Anna-Liisa Prabhu, Yongjun Zhao, HelenMcDonald, Thomas Zeng, Martin Hirst, Connie J. Eaves, and Marco A.Marra. Application of massively parallel sequencing to microRNAprofiling and discovery in human embryonic stem cells. Genome Research,18(4):610–621, 2008. doi:10.1101/gr.7179508. → pages 6, 22, 104[48] Lik Wee Lee, Shile Zhang, Alton Etheridge, Li Ma, Dan Martin, DavidGalas, and Kai Wang. Complexity of the microRNA repertoire revealed bynext-generation sequencing. RNA, 16(11):2170–2180, 2010.doi:10.1261/rna.2225110. → page 104[49] Florian Kuchenbauer, Ryan D. Morin, Bob Argiropoulos, Oleh I. Petriv,Malachi Griffith, Michael Heuser, Eric Yung, Jessica Piper, Allen Delaney,Anna-Liisa Prabhu, Yongjun Zhao, Helen McDonald, Thomas Zeng,Martin Hirst, Carl L. Hansen, Marco A. Marra, and R. Keith Humphries.In-depth characterization of the microRNA transcriptome in a leukemiaprogression model. Genome Research, 18(11):1787–1797, 2008.doi:10.1101/gr.077578.108. → pages 6, 104[50] Nicole Cloonan, Shivangi Wani, Qinying Xu, Jian Gu, Kristi Lea, SheilaHeater, Catalin Barbacioru, Anita L. Steptoe, Hilary C. Martin, EhsanNourbakhsh, Keerthana Krishnan, Brooke Gardiner, Xiaohui Wang, KatiaNones, Jason A. Steen, Nicholas A. Matigian, David L. Wood, Karin S.Kassahn, Nic Waddell, Jill Shepherd, Clarence Lee, Jeff Ichikawa, KevinMcKernan, Kelli Bramlett, Scott Kuersten, and Sean M. Grimmond.microRNAs and their isomiRs function cooperatively to target commonbiological pathways. Genome Biology, 12(12):R126, 2011.doi:10.1186/gb-2011-12-12-r126. → page 7[51] Corine T. Neilsen, Gregory J. Goodall, and Cameron P. Bracken. IsomiRs –the overlooked repertoire in the dynamic microRNAome. Trends in175Genetics, 28(11):544–549, 2012. doi:10.1016/j.tig.2012.07.005. → pages7, 106[52] Geok Chin Tan, Elcie Chan, Attila Molnar, Rupa Sarkar, Diana Alexieva,Ihsan Mad Isa, Sophie Robinson, Shuchen Zhang, Peter Ellis, Cordelia F.Langford, Pascale V. Guillot, Anil Chandrashekran, Nick M. Fisk, LeandroCastellano, Gunter Meister, Robert M. Winston, Wei Cui, DavidBaulcombe, and Nick J. Dibb. 5′ isomiR variation is of functional andevolutionary importance. Nucleic Acids Research, 42(14):9424–9435,2014. doi:10.1093/nar/gku656. → page 7[53] H. Rosaria Chiang, Lori W. Schoenfeld, J. Graham Ruby, Vincent C.Auyeung, Noah Spies, Daehyun Baek, Wendy K. Johnston, Carsten Russ,Shujun Luo, Joshua E. Babiarz, Robert Blelloch, Gary P. Schroth, ChadNusbaum, and David P. Bartel. Mammalian microRNAs: experimentalevaluation of novel and previously annotated genes. Genes & Development,24(10):992–1009, 2010. doi:10.1101/gad.1884710. → pages 7, 8[54] Herve´ Seitz, Megha Ghildiyal, and Phillip D. Zamore. Argonaute loadingimproves the 5′ precision of both microRNAs and their miRNA* strands inflies. Current Biology, 18(2):147–151, 2007.doi:10.1016/j.cub.2007.12.049. → pages 7, 157[55] Haoquan Wu, Joel R. Neilson, Priti Kumar, Monika Manocha, PremlataShankar, Phillip A. Sharp, and N. Manjunath. miRNA profiling of naı¨ve,effector and memory CD8 T cells. PLOS ONE, 2(10):e1020, 2007.doi:10.1371/journal.pone.0001020. → pages 9, 106[56] Bo W. Han, Jui-Hung Hung, Zhiping Weng, Phillip D. Zamore, andStefan L. Ameres. The 3′-to-5′ exoribonuclease Nibbler shapes the 3′ endsof microRNAs bound to Drosophila Argonaute1. Current Biology, 21(22):1878–1887, 2011. doi:10.1016/j.cub.2011.09.034. → page 9[57] Olivia S. Rissland, Andrea Mikulasova, and Chris J. Norbury. EfficientRNA polyuridylation by noncanonical poly(A) polymerases. Molecularand Cellular Biology, 27(10):3612–3624, 2007.doi:10.1128/mcb.02209-06. → page 9[58] Hao-Ming Chang, Robinson Triboulet, James E. Thornton, and Richard I.Gregory. A role for the Perlman syndrome exonuclease Dis3l2 in theLin28-let-7 pathway. Nature, 497(7448):244–248, 2013.doi:10.1038/nature12119. → page 9176[59] Inha Heo, Minju Ha, Jaechul Lim, Mi-Jeong Yoon, Jong-Eun Park, S. ChulKwon, Hyeshik Chang, and V. Narry Kim. Mono-uridylation ofpre-microRNA as a key step in the biogenesis of group II let-7 microRNAs.Cell, 151(3):521–532, 2012. doi:10.1016/j.cell.2012.09.022. → page 9[60] Fuchou Tang, Catalin Barbacioru, Yangzhou Wang, Ellen Nordman,Clarence Lee, Nanlan Xu, Xiaohui Wang, John Bodeau, Brian B. Tuch,Asim Siddiqui, Kaiqin Lao, and M. Azim Surani. mRNA-seqwhole-transcriptome analysis of a single cell. Nature Methods, 6(5):377–382, 2009. doi:10.1038/nmeth.1315. → pages 9, 21, 36, 73[61] Nicholas M. Toriello, Erik S. Douglas, Numrin Thaitrong, Sonny C. Hsiao,Matthew B. Francis, Carolyn R. Bertozzi, and Richard A. Mathies.Integrated microfluidic bioprocessor for single-cell gene expressionanalysis. Proceedings of the National Academy of Sciences, 105(51):20173–20178, 2008. doi:10.1073/pnas.0806355106. → pages 9, 36[62] Yuan Gong, Adebola O. Ogunniyi, and J. Christopher Love. Massivelyparallel detection of gene expression in single cells using subnanolitrewells. Lab on a Chip, 10(18):2334–2337, 2010. doi:10.1039/C004847J.→ pages 9, 24, 53[63] Fuchou Tang, Petra Hajkova, Sheila C. Barton, Kaiqin Lao, and M. AzimSurani. microRNA expression profiling of single whole embryonic stemcells. Nucleic Acids Research, 34(2):e9, 2006. doi:10.1093/nar/gnj009. →pages 9, 18, 64, 106[64] Colin C. Pritchard, Heather H. Cheng, and Muneesh Tewari. MicroRNAprofiling: approaches and considerations. Nature Reviews Genetics, 13(5):358–369, 2012. doi:10.1038/nrg3198. → pages 10, 11, 106[65] Alastair W. Wark, Hye Jin Lee, and Robert M. Corn. Multiplexed detectionmethods for profiling microRNA expression in biological samples.Angewandte Chemie International Edition, 47(4):644–652, 2008.doi:10.1002/anie.200702450. → pages 10, 11[66] Andrew D. Bosson, Jesse R. Zamudio, and Phillip A. Sharp. EndogenousmiRNA and target concentrations determine susceptibility to potentialceRNA competition. Molecular Cell, 56(3):347–359, 2014.doi:10.1016/j.molcel.2014.09.018. → page 10[67] Brian D. Brown, Bernhard Gentner, Alessio Cantore, Silvia Colleoni,Mario Amendola, Anna Zingale, Alessia Baccarini, Giovanna Lazzari,177Cesare Galli, and Luigi Naldini. Endogenous microRNA can be broadlyexploited to regulate transgene expression according to tissue, lineage anddifferentiation state. Nature Biotechnology, 25(12):1457–1467, 2007.doi:10.1038/nbt1372. → pages 10, 106, 131[68] Shankar Mukherji, Margaret S. Ebert, Grace X. Y. Zheng, John S. Tsang,Phillip A. Sharp, and Alexander van Oudenaarden. MicroRNAs cangenerate thresholds in target gene expression. Nature Genetics, 43(9):854–859, 2011. doi:10.1038/ng.905. → pages 15, 131[69] Liang Meng Wee, C. Fabia´n Flores-Jasso, William E. Salomon, andPhillip D. Zamore. Argonaute divides its RNA guide into domains withdistinct functions and RNA-binding properties. Cell, 151(5):1055–1067,2012. doi:10.1016/j.cell.2012.10.036. → page 10[70] Matteo J. Marzi, Francesco Ghini, Benedetta Cerruti, Stefano de Pretis,Paola Bonetti, Chiara Giacomelli, Marcin M. Gorski, Theresia Kress,Mattia Pelizzola, Heiko Muller, Bruno Amati, and Francesco Nicassio.Degradation dynamics of microRNAs revealed by a novel pulse-chaseapproach. Genome Research, 26(4):554–565, 2016.doi:10.1101/gr.198788.115. → page 10[71] Bjo¨rn Schwanha¨usser, Dorothea Busse, Na Li, Gunnar Dittmar, JohannesSchuchhardt, Jana Wolf, Wei Chen, and Matthias Selbach. Globalquantification of mammalian gene expression control. Nature, 473(7347):337–342, 2011. doi:10.1038/nature10098. → page 10[72] Arjun Raj and Alexander van Oudenaarden. Nature, nurture, or chance:Stochastic gene expression and its consequences. Cell, 135(2):216–226,2008. doi:10.1016/j.cell.2008.09.050. → pages 10, 32[73] Ana Conesa, Pedro Madrigal, Sonia Tarazona, David Gomez-Cabrero,Alejandra Cervera, Andrew McPherson, Michał Wojciech Szczes´niak,Daniel J. Gaffney, Laura L. Elo, Xuegong Zhang, and Ali Mortazavi. Asurvey of best practices for RNA-seq data analysis. Genome Biology, 17(1):13, 2016. doi:10.1186/s13059-016-0881-8. → page 11[74] Angela R. Wu, Norma F. Neff, Tomer Kalisky, Piero Dalerba, BarbaraTreutlein, Michael E. Rothenberg, Francis M. Mburu, Gary L. Mantalas,Sopheak Sim, Michael F. Clarke, and Stephen R. Quake. Quantitativeassessment of single-cell RNA-sequencing methods. Nature Methods, 11(1):41–46, 2014. doi:10.1038/nmeth.2694. → pages 11, 73, 75178[75] Jeffrey M. Levsky and Robert H. Singer. Fluorescence in situhybridization: past, present and future. Journal of Cell Science, 116(14):2833–2838, 2003. doi:10.1242/jcs.00633. → page 12[76] Jeffrey M. Levsky, Shailesh M. Shenoy, Rossanna C. Pezo, and Robert H.Singer. Single-cell gene expression profiling. Science, 297(5582):836–840,2002. doi:10.1126/science.1072241. → pages 12, 75[77] Arjun Raj, Patrick van den Bogaard, Scott A. Rifkin, Alexander vanOudenaarden, and Sanjay Tyagi. Imaging individual mRNA moleculesusing multiple singly labeled probes. Nature Methods, 5(10):877–879,2008. doi:10.1038/nmeth.1253. → pages 12, 36, 73, 75[78] Kok Hao Chen, Alistair N. Boettiger, Jeffrey R. Moffitt, Siyuan Wang, andXiaowei Zhuang. Spatially resolved, highly multiplexed RNA profiling insingle cells. Science, 348(6233), 2015. doi:10.1126/science.aaa6090. →pages 12, 73, 75[79] Jeffrey R. Moffitt, Junjie Hao, Guiping Wang, Kok Hao Chen, Hazen P.Babcock, and Xiaowei Zhuang. High-throughput single-cellgene-expression profiling with multiplexed error-robust fluorescence in situhybridization. Proceedings of the National Academy of Sciences, 113(39):11046–11051, 2016. doi:10.1073/pnas.1612826113. → pages 12, 75[80] Violette B. Paragas, Yu-Zhong Zhang, Richard P. Haugland, and Victoria L.Singer. The ELF-97 alkaline phosphatase substrate provides a bright,photostable, fluorescent signal amplification method for FISH. Journal ofHistochemistry & Cytochemistry, 45(3):345–357, 1997.doi:10.1177/002215549704500302. → page 12[81] Andrea M. Femino, Fredric S. Fay, Kevin Fogarty, and Robert H. Singer.Visualization of single RNA transcripts in situ. Science, 280(5363):585–590, 1998. doi:10.1126/science.280.5363.585. → page 12[82] M. Petersen and J. Wengel. LNA: a versatile tool for therapeutics andgenomics. Trends in Biotechnology, 21(2):74–81, 2003.doi:10.1016/S0167-7799(02)00038-0. → page 13[83] Sanjay K. Singh, Alexei A. Koshkin, Jesper Wengel, and Poul Nielsen.LNA (locked nucleic acids): synthesis and high-affinity nucleic acidrecognition. Chemical Communications, (4):455–456, 1998.doi:10.1039/A708608C. → page 13179[84] Mirco Castoldi, Sabine Schmidt, Vladimir Benes, Mikkel Noerholm,Andreas E. Kulozik, Matthias W. Hentze, and Martina U. Muckenthaler. Asensitive array for microRNA expression profiling (miChip) based onlocked nucleic acids (LNA). RNA, 12(5):913–920, 2006.doi:10.1261/rna.2332406. → page 13[85] Anna Va´lo´czi, Csaba Hornyik, No´ra Varga, Jo´zsef Burgya´n, SakariKauppinen, and Zolta´n Havelda. Sensitive and specific detection ofmicroRNAs by northern blot analysis using LNA-modified oligonucleotideprobes. Nucleic Acids Research, 32(22):e175–e175, 2004.doi:10.1093/nar/gnh171. → page 13[86] Wigard P. Kloosterman, Erno Wienholds, Ewart de Bruijn, SakariKauppinen, and Ronald H. A. Plasterk. In situ detection of miRNAs inanimal embryos using LNA-modified oligonucleotide probes. NatureMethods, 3:27, 2005. doi:10.1038/nmeth843. → page 13[87] Jing Lu and Andrew Tsourkas. Imaging individual microRNAs in singlemammalian cells in situ. Nucleic Acids Research, 37(14):e100–e100, 2009.doi:10.1093/nar/gkp482. → page 13[88] Peter T. Nelson, Don A. Baldwin, Wigard P. Kloosterman, SakariKauppinen, Ronald H.A. Plasterk, and Zissimos Mourelatos. RAKE andLNA-ISH reveal microRNA expression and localization in archival humanbrain. RNA, 12(2):187–191, 2006. doi:10.1261/rna.2258506.[89] John T. G. Pena, Cherin Sohn-Lee, Sara H. Rouhanifard, Janos Ludwig,Markus Hafner, Aleksandra Mihailovic, Cindy Lim, Daniel Holoch,Philipp Berninger, Mihaela Zavolan, and Thomas Tuschl. miRNA in situhybridization in formaldehyde and EDC-fixed tissues. Nature Methods, 6:139, 2009. doi:10.1038/nmeth.1294. → page 13[90] Asli N. Silahtaroglu, Dorrit Nolting, Lars Dyrskjøt, Eugene Berezikov,Morten Møller, Niels Tommerup, and Sakari Kauppinen. Detection ofmicroRNAs in frozen tissue sections by fluorescence in situ hybridizationusing locked nucleic acid probes and tyramide signal amplification. NatureProtocols, 2:2520, 2007. doi:10.1038/nprot.2007.313.[91] Erno Wienholds, Wigard P. Kloosterman, Eric Miska, EzequielAlvarez-Saavedra, Eugene Berezikov, Ewart de Bruijn, H. Robert Horvitz,Sakari Kauppinen, and Ronald H. A. Plasterk. MicroRNA expression inzebrafish embryonic development. Science, 309(5732):310–311, 2005.doi:10.1126/science.1114519. → page 13180[92] Hyunseo Koo, Ikbum Park, Yoonhee Lee, Hyun Jin Kim, Jung Hoon Jung,Joo Han Lee, Youngkyu Kim, Joung-Hun Kim, and Joon Won Park.Visualization and quantification of microRNA in a single cell using atomicforce microscopy. Journal of the American Chemical Society, 138(36):11664–11671, 2016. doi:10.1021/jacs.6b05048. → page 14[93] Filippos Porichis, Meghan G. Hart, Morgane Griesbeck, Holly L. Everett,Muska Hassan, Amy E. Baxter, Madelene Lindqvist, Sara M. Miller,Damien Z. Soghoian, Daniel G. Kavanagh, Susan Reynolds, Brett Norris,Scott K. Mordecai, Quan Nguyen, Chunfai Lai, and Daniel E. Kaufmann.High-throughput detection of miRNAs and gene-specific mRNA at thesingle-cell level by flow cytometry. Nature Communications, 5:5641, 2014.doi:10.1038/ncomms6641. → page 14[94] Meiye Wu, Matthew Piccini, Chung-Yan Koh, Kit S. Lam, and Anup K.Singh. Single cell microRNA analysis using microfluidic flow cytometry.PLOS ONE, 8(1):e55044, 2013. doi:10.1371/journal.pone.0055044. →pages 14, 106[95] Audrey N. Player, Lu-Ping Shen, Daryn Kenny, Vincent P. Antao, andJanice A. Kolberg. Single-copy gene detection using branched DNA(bDNA) in situ hybridization. Journal of Histochemistry & Cytochemistry,49(5):603–611, 2001. doi:10.1177/002215540104900507. → page 14[96] Steven M. Johnson, Shin-Yi Lin, and Frank J. Slack. The time ofappearance of the C. elegans let-7 microRNA is transcriptionally controlledutilizing a temporal regulatory element in its promoter. DevelopmentalBiology, 259(2):364–379, 2003. doi:10.1016/S0012-1606(03)00202-1. →page 15[97] Gyo¨rgy Hutva´gner and Phillip D. Zamore. A microRNA in amultiple-turnover RNAi enzyme complex. Science, 297(5589):2056–2060,2002. doi:10.1126/science.1073827. → page 15[98] Julius Brennecke, David R. Hipfner, Alexander Stark, Robert B. Russell,and Stephen M. Cohen. bantam encodes a developmentally regulatedmicroRNA that controls cell proliferation and regulates the proapoptoticgene hid in Drosophila. Cell, 113(1):25–36, 2003.doi:10.1016/S0092-8674(03)00231-9. → page 15[99] Jennifer H. Mansfield, Brian D. Harfe, Robert Nissen, John Obenauer,Jalagani Srineel, Aadel Chaudhuri, Raphael Farzan-Kashani, Michael181Zuker, Amy E. Pasquinelli, Gary Ruvkun, Phillip A. Sharp, Clifford J.Tabin, and Michael T. McManus. MicroRNA-responsive ‘sensor’transgenes uncover Hox-like and other developmentally regulated patternsof vertebrate microRNA expression. Nature Genetics, 36:1079, 2004.doi:10.1038/ng1421. → pages 15, 16[100] D. De Pietri Tonelli, F. Calegari, J. F. Fei, T. Nomura, N. Osumi, C. P.Heisenberg, and W. B. Huttner. Single-cell detection of microRNAs indeveloping vertebrate embryos after acute administration of adual-fluorescence reporter/sensor plasmid. Biotechniques, 41(6):727–32,2006. → page 15[101] Bernhard Gentner, Ilaria Visigalli, Hidefumi Hiramatsu, Eric Lechman,Silvia Ungari, Alice Giustacchini, Giulia Schira, Mario Amendola, AngeloQuattrini, Sabata Martino, Aldo Orlacchio, John E. Dick, Alessandra Biffi,and Luigi Naldini. Identification of hematopoietic stem cell-specificmiRNAs enables gene therapy of globoid cell leukodystrophy. ScienceTranslational Medicine, 2(58):58ra84–58ra84, 2010.doi:10.1126/scitranslmed.3001522. → pages 15, 32, 106, 140, 157[102] Zhen Xie, Liliana Wroblewska, Laura Prochazka, Ron Weiss, and YaakovBenenson. Multi-input RNAi-based logic circuit for identification ofspecific cancer cells. Science, 333(6047):1307–1311, 2011.doi:10.1126/science.1205527. → page 16[103] Ian M. Mackay, Katherine E. Arden, and Andreas Nitsche. Real-time PCRin virology. Nucleic Acids Research, 30(6):1292–1305, 2002.doi:10.1093/nar/30.6.1292. → page 17[104] P. J. Sykes, S. H. Neoh, M. J. Brisco, E. Hughes, J. Condon, and A. A.Morley. Quantitation of targets for PCR by use of limiting dilution.Biotechniques, 13(3):444–9, 1992. → page 18[105] Bert Vogelstein and Kenneth W. Kinzler. Digital PCR. Proceedings of theNational Academy of Sciences, 96(16):9236–9241, 1999.doi:10.1073/pnas.96.16.9236. → page 18[106] Luigi A. Warren, Joshua A. Weinstein, and Stephen R. Quake. The digitalarray response curve, 2007. → pages 18, 84, 117[107] Kevin A. Heyries, Carolina Tropini, Michael VanInsberghe, CallumDoolin, Oleh I. Petriv, Anupam Singhal, Kaston Leung, Curtis B.182Hughesman, and Carl L. Hansen. Megapixel digital PCR. Nature Methods,8(8):649–651, 2011. doi:10.1038/nmeth.1640. → pages 18, 83[108] Benjamin J. Hindson, Kevin D. Ness, Donald A. Masquelier, PhillipBelgrader, Nicholas J. Heredia, Anthony J. Makarewicz, Isaac J. Bright,Michael Y. Lucero, Amy L. Hiddessen, Tina C. Legler, Tyler K. Kitano,Michael R. Hodel, Jonathan F. Petersen, Paul W. Wyatt, Erin R.Steenblock, Pallavi H. Shah, Luc J. Bousse, Camille B. Troup, Jeffrey C.Mellen, Dean K. Wittmann, Nicholas G. Erndt, Thomas H. Cauley, Ryan T.Koehler, Austin P. So, Simant Dube, Klint A. Rose, Luz Montesclaros,Shenglong Wang, David P. Stumbo, Shawn P. Hodges, Steven Romine,Fred P. Milanovich, Helen E. White, John F. Regan, George A.Karlin-Neumann, Christopher M. Hindson, Serge Saxonov, and Bill W.Colston. High-throughput droplet digital PCR system for absolutequantitation of DNA copy number. Analytical Chemistry, 83(22):8604–8610, 2011. doi:10.1021/ac202028g. → page 18[109] Seungil Ro, Chanjae Park, Jingling Jin, Kenton M. Sanders, and Wei Yan.A PCR-based method for detection and quantification of small RNAs.Biochemical and Biophysical Research Communications, 351(3):756–763,2006. doi:10.1016/j.bbrc.2006.10.105. → pages 18, 19[110] 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 bystem-loop RT-PCR. Nucleic Acids Research, 33(20):e179, 2005.doi:10.1093/nar/gni178. → pages 18, 19, 40, 91[111] 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,C. Benz, E. Kuroda, V. Ho, F. Antignano, T. Halim, V. Giambra,G. Krystal, C. J. F. Takei, A. P. Weng, J. Piret, C. Eaves, M. A. Marra, R. K.Humphries, and C. L. Hansen. Comprehensive microRNA expressionprofiling of the hematopoietic hierarchy. Proceedings of the NationalAcademy of Sciences, 107(35):15443–15448, 2010.doi:10.1073/pnas.1009320107. → pages18, 20, 36, 63, 71, 73, 75, 78, 99, 106, 140[112] Fuchou Tang, Petra Hajkova, Sheila C. Barton, Do´nal O’Carroll, CarolineLee, Kaiqin Lao, and M. Azim Surani. 220-plex microRNA expressionprofile of a single cell. Nature Protocols, 1:1154, 2006.doi:10.1038/nprot.2006.161.183[113] Fuchou Tang, Masahiro Kaneda, Do´nal O’Carroll, Petra Hajkova, Sheila C.Barton, Y. Andrew Sun, Caroline Lee, Alexander Tarakhovsky, Kaiqin Lao,and M. Azim Surani. Maternal microRNAs are essential for mouse zygoticdevelopment. Genes & Development, 21(6):644–648, 2007.doi:10.1101/gad.418707. → page 19[114] Adam K. White, Kevin A. Heyries, Callum Doolin, Michael VanInsberghe,and Carl L. Hansen. High-throughput microfluidic single-cell digitalpolymerase chain reaction. Analytical Chemistry, 85(15):7182–7190, 2013.doi:10.1021/ac400896j. → pages 73, 75, 85, 96, 97, 106, 161[115] Adam K. White, Michael VanInsberghe, Oleh I. Petriv, Mani. Hamidi,Darek. Sikorski, Marco A. Marra, James Piret, Samuel. Aparicio, andCarl L. Hansen. High-throughput microfluidic single-cell RT-qPCR.Proceedings of the National Academy of Sciences, 108(34):13999–14004,2011. doi:10.1073/pnas.1019446108. → pages18, 24, 25, 73, 75, 78, 84, 85, 90, 96, 97, 106, 126, 130, 161[116] Sandra L. Spurgeon, Robert C. Jones, and Ramesh Ramakrishnan. Highthroughput gene expression measurement with real time PCR in amicrofluidic dynamic array. PLOS ONE, 3(2):e1662, 2008.doi:10.1371/journal.pone.0001662. → pages 20, 73[117] Hubert Hug and Rainer Schuler. Measurement of the number of moleculesof a single mRNA species in a complex mRNA preparation. Journal ofTheoretical Biology, 221(4):615–624, 2003. doi:10.1006/jtbi.2003.3211.→ page 20[118] Teemu Kivioja, Anna Va¨ha¨rautio, Kasper Karlsson, Martin Bonke, MartinEnge, Sten Linnarsson, and Jussi Taipale. Counting absolute numbers ofmolecules using unique molecular identifiers. Nature Methods, 9:72, 2011.doi:10.1038/nmeth.1778. → page 20[119] Aleksandra A. Kolodziejczyk, Jong Kyoung Kim, Valentine Svensson,John C. Marioni, and Sarah A. Teichmann. The technology and biology ofsingle-cell RNA sequencing. Molecular Cell, 58(4):610–620, 2015.doi:10.1016/j.molcel.2015.04.005. → page 21[120] Ehud Shapiro, Tamir Biezuner, and Sten Linnarsson. Single-cellsequencing-based technologies will revolutionize whole-organism science.Nature Reviews Genetics, 14:618, 2013. doi:10.1038/nrg3542. → page 21184[121] Sayda M. Elbashir, Winfried Lendeckel, and Thomas Tuschl. RNAinterference is mediated by 21- and 22-nucleotide RNAs. Genes &Development, 15(2):188–200, 2001. doi:10.1101/gad.862301. → page 21[122] A. Chu, G. Robertson, D. Brooks, A. J. Mungall, I. Birol, R. Coope, Y. Ma,S. Jones, and M. A. Marra. Large-scale profiling of microRNAs for TheCancer Genome Atlas. Nucleic Acids Research, 44(1):e3, 2016.doi:10.1093/nar/gkv808. → pages 22, 117, 126, 134[123] Markus Hafner, Neil Renwick, Miguel Brown, Aleksandra Mihailovic´,Daniel Holoch, Carolina Lin, John T.G. Pena, Jeffrey D. Nusbaum, PavelMorozov, Janos Ludwig, Tolulope Ojo, Shujun Luo, Gary Schroth, andThomas Tuschl. RNA-ligase-dependent biases in miRNA representation indeep-sequenced small RNA cDNA libraries. RNA, 17(9):1697–1712, 2011.doi:10.1261/rna.2799511. → page 22[124] Fanglei Zhuang, Ryan T. Fuchs, Zhiyi Sun, Yu Zheng, and G. Brett Robb.Structural bias in T4 RNA ligase-mediated 3′-adapter ligation. NucleicAcids Research, 40(7):e54–e54, 2012. doi:10.1093/nar/gkr1263. → page22[125] Jeanette Baran-Gale, C. Lisa Kurtz, Michael R. Erdos, Christina Sison,Alice Young, Emily E. Fannin, Peter S. Chines, and Praveen Sethupathy.Addressing bias in small RNA library preparation for sequencing: A newprotocol recovers microRNAs that evade capture by current methods.Frontiers in Genetics, 6(352), 2015. doi:10.3389/fgene.2015.00352. →page 22[126] Omid R. Faridani, Ilgar Abdullayev, Michael Hagemann-Jensen, John P.Schell, Fredrik Lanner, and Rickard Sandberg. Single-cell sequencing ofthe small-RNA transcriptome. Nature Biotechnology, 34(12):1264–1266,2016. doi:10.1038/nbt.3701. → pages 22, 106, 124, 132, 134, 165[127] Mira T. Guo, Assaf Rotem, John A. Heyman, and David A. Weitz. Dropletmicrofluidics for high-throughput biological assays. Lab on a Chip, 12(12):2146–2155, 2012. doi:10.1039/C2LC21147E. → page 23[128] Jessica Melin and Stephen R. Quake. Microfluidic large-scale integration:The evolution of design rules for biological automation. Annual Review ofBiophysics and Biomolecular Structure, 36(1):213–231, 2007.doi:10.1146/annurev.biophys.36.040306.132646. → pages 23, 25185[129] Todd Thorsen, Sebastian J. Maerkl, and Stephen R. Quake. Microfluidiclarge-scale integration. Science, 298(5593):580–584, 2002.doi:10.1126/science.1076996. → pages 23, 25, 29, 37[130] Liang Wu, Xiaolong Zhang, Zhikun Zhao, Ling Wang, Bo Li, Guibo Li,Michael Dean, Qichao Yu, Yanhui Wang, Xinxin Lin, Weijian Rao,Zhanlong Mei, Yang Li, Runze Jiang, Huan Yang, Fuqiang Li, GuoyunXie, Liqin Xu, Kui Wu, Jie Zhang, Jianghao Chen, Ting Wang, KarstenKristiansen, Xiuqing Zhang, Yingrui Li, Huanming Yang, Jian Wang, YongHou, and Xun Xu. Full-length single-cell RNA-seq applied to a viralhuman cancer: applications to HPV expression and splicing analysis inHeLa S3 cells. GigaScience, 4(1):51, 2015.doi:10.1186/s13742-015-0091-4. → pages 23, 156[131] Kaston Leung, Anders Klaus, Bill K. Lin, Emma Laks, Justina Biele,Daniel Lai, Ali Bashashati, Yi-Fei Huang, Radhouane Aniba, MichelleMoksa, Adi Steif, Anne-Marie Mes-Masson, Martin Hirst, Sohrab P. Shah,Samuel Aparicio, and Carl L. Hansen. Robust high-performancenanoliter-volume single-cell multiple displacement amplification on planarsubstrates. Proceedings of the National Academy of Sciences, 113(30):8484–8489, 2016. doi:10.1073/pnas.1520964113. → pages 23, 24[132] Jeff Gole, Athurva Gore, Andrew Richards, Yu-Jui Chiu, Ho-Lim Fung,Diane Bushman, Hsin-I. Chiang, Jerold Chun, Yu-Hwa Lo, and KunZhang. Massively parallel polymerase cloning and genome sequencing ofsingle cells using nanoliter microwells. Nature Biotechnology, 31:1126,2013. doi:10.1038/nbt.2720. → page 24[133] Charles F. A. de Bourcy, Iwijn De Vlaminck, Jad N. Kanbar, Jianbin Wang,Charles Gawad, and Stephen R. Quake. A quantitative comparison ofsingle-cell whole genome amplification methods. PLOS ONE, 9(8):e105585, 2014. doi:10.1371/journal.pone.0105585. → page 24[134] Jiang F. Zhong, Yan Chen, Joshua S. Marcus, Axel Scherer, Stephen R.Quake, Clive R. Taylor, and Leslie P. Weiner. A microfluidic processor forgene expression profiling of single human embryonic stem cells. Lab on aChip, 8(1):68–74, 2008. doi:10.1039/B712116D. → pages 24, 36, 60[135] George M. Whitesides. The origins and the future of microfluidics. Nature,442:368, 2006. doi:10.1038/nature05058. → pages 24, 29186[136] J. Nilsson, M. Evander, B. Hammarstro¨m, and T. Laurell. Review of celland particle trapping in microfluidic systems. Analytica Chimica Acta, 649(2):141–157, 2009. doi:10.1016/j.aca.2009.07.017. → page 24[137] Evan Z. Macosko, Anindita Basu, Rahul Satija, James Nemesh, KarthikShekhar, Melissa Goldman, Itay Tirosh, Allison R. Bialas, Nolan Kamitaki,Emily M. Martersteck, John J. Trombetta, David A. Weitz, Joshua R.Sanes, Alex K. Shalek, Aviv Regev, and Steven A. McCarroll. Highlyparallel genome-wide expression profiling of individual cells usingnanoliter droplets. Cell, 161(5):1202–1214, 2015.doi:10.1016/j.cell.2015.05.002. → pages 25, 73, 156, 165[138] 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. Lab on a Chip, 8(3):443–450, 2008.doi:10.1039/B716543A. → pages 25, 36, 78[139] Kaston Leung, Hans Zahn, Timothy Leaver, Kishori M. Konwar, Niels W.Hanson, Antoine P. Page´, Chien-Chi Lo, Patrick S. Chain, Steven J.Hallam, and Carl L. Hansen. A programmable droplet-based microfluidicdevice applied to multiparameter analysis of single microbes and microbialcommunities. Proceedings of the National Academy of Sciences, 109(20):7665–7670, 2012. doi:10.1073/pnas.1106752109. → page 164[140] Joshua S. Marcus, W. French Anderson, and Stephen R. Quake.Microfluidic single-cell mRNA isolation and analysis. AnalyticalChemistry, 78(9):3084–3089, 2006. doi:10.1021/ac0519460. → pages25, 36, 60, 61[141] Aaron R. Wheeler, William R. Throndset, Rebecca J. Whelan, Andrew M.Leach, Richard N. Zare, Yish Hann Liao, Kevin Farrell, Ian D. Manger, andAntoine Daridon. Microfluidic device for single-cell analysis. AnalyticalChemistry, 75(14):3581–3586, 2003. doi:10.1021/ac0340758. → pages25, 36[142] Anupam Singhal, Charles A. Haynes, and Carl L. Hansen. Microfluidicmeasurement of antibody-antigen binding kinetics from low-abundancesamples and single cells. Analytical Chemistry, 82(20):8671–8679, 2010.doi:10.1021/ac101956e. → page 25[143] Dino Di Carlo, Nima Aghdam, and Luke P. Lee. Single-cell enzymeconcentrations, kinetics, and inhibition analysis using high-density187hydrodynamic cell isolation arrays. Analytical Chemistry, 78(14):4925–4930, 2006. doi:10.1021/ac060541s. → pages 25, 56, 159[144] Alison M. Skelley, Oktay Kirak, Heikyung Suh, Rudolf Jaenisch, and JoelVoldman. Microfluidic control of cell pairing and fusion. Nature Methods,6(2):147–152, 2009. doi:10.1038/nmeth.1290. → pages 25, 36, 50[145] A. A. Pollen, T. J. Nowakowski, J. Shuga, X. Wang, A. A. Leyrat, J. H. Lui,N. Li, L. Szpankowski, B. Fowler, P. Chen, N. Ramalingam, G. Sun,M. Thu, M. Norris, R. Lebofsky, D. Toppani, 2nd Kemp, D. W., M. Wong,B. Clerkson, B. N. Jones, S. Wu, L. Knutsson, B. Alvarado, J. Wang, L. S.Weaver, A. P. May, R. C. Jones, M. A. Unger, A. R. Kriegstein, and J. A.West. Low-coverage single-cell mRNA sequencing reveals cellularheterogeneity and activated signaling pathways in developing cerebralcortex. Nature Biotechnology, 32(10):1053–8, 2014.doi:10.1038/nbt.2967. → pages 25, 73, 126, 161, 164[146] Wei-Heong Tan and Shoji Takeuchi. A trap-and-release integratedmicrofluidic system for dynamic microarray applications. Proceedings ofthe National Academy of Sciences, 104(4):1146–1151, 2007.doi:10.1073/pnas.0606625104. → page 25[147] Fluidigm. Redesign of C1 Medium-Cell 96 and HT IFCs ImprovesSingle-Cell Capture Efficiency [Whitepaper]. 2016. → page 25[148] David C. Duffy, J. Cooper McDonald, Olivier J. A. Schueller, andGeorge M. Whitesides. Rapid prototyping of microfluidic systems inpoly(dimethylsiloxane). Analytical Chemistry, 70(23):4974–4984, 1998.doi:10.1021/ac980656z. → pages 26, 27[149] Marc A. Unger, Hou-Pu Chou, Todd Thorsen, Axel Scherer, andStephen R. Quake. Monolithic microfabricated valves and pumps bymultilayer soft lithography. Science, 288(5463):113–116, 2000.doi:10.1126/science.288.5463.113. → pages 27, 28, 37, 76, 90, 110[150] Connie J. Eaves. Hematopoietic stem cells: concepts, definitions, and thenew reality. Blood, 125(17):2605–2613, 2015.doi:10.1182/blood-2014-12-570200. → pages 29, 31[151] Stuart H. Orkin and Leonard I. Zon. Hematopoiesis: An evolving paradigmfor stem cell biology. Cell, 132(4):631–644, 2008.doi:10.1016/j.cell.2008.01.025. → page 29188[152] L. O. Jacobson, E. L. Simmons, E. K. Marks, and J. H. Eldredge. Recoveryfrom radiation injury. Science, 113(2940):510–511, 1951.doi:10.1126/science.113.2940.510. → page 29[153] Andrew J. Becker, Ernest A. McCulloch, and James E. Till. Cytologicaldemonstration of the clonal nature of spleen colonies derived fromtransplanted mouse marrow cells. Nature, 197:452, 1963.doi:10.1038/197452a0. → page 29[154] J. E. Till and E. A. McCulloch. A direct measurement of the radiationsensitivity of normal mouse bone marrow cells. Radiation Research, 14(2):213–222, 1961. doi:10.2307/3570892. → page 29[155] L. Siminovitch, E. A. McCulloch, and J. E. Till. The distribution ofcolony-forming cells among spleen colonies. Journal of Cellular andComparative Physiology, 62(3):327–336, 1963.doi:10.1002/jcp.1030620313. → page 29[156] A. M. Wu, J. E. Till, L. Siminovitch, and E. A. McCulloch. A cytologicalstudy of the capacity for differentiation of normal hemopoieticcolony-forming cells. Journal of Cellular Physiology, 69(2):177–184,1967. doi:10.1002/jcp.1040690208. → page 29[157] Tannishtha Reya, Sean J. Morrison, Michael F. Clarke, and Irving L.Weissman. Stem cells, cancer, and cancer stem cells. Nature, 414(6859):105–111, 2001. doi:10.1038/35102167. → pages 30, 136[158] Koichi Akashi, David Traver, Toshihiro Miyamoto, and Irving L.Weissman. A clonogenic common myeloid progenitor that gives rise to allmyeloid lineages. Nature, 404:193, 2000. doi:10.1038/35004599. → page31[159] Sergei Doulatov, Faiyaz Notta, Kolja Eppert, Linh T. Nguyen, Pamela S.Ohashi, and John E. Dick. Revised map of the human progenitor hierarchyshows the origin of macrophages and dendritic cells in early lymphoiddevelopment. Nature Immunology, 11(7):585–593, 2010.doi:10.1038/ni.1889. → pages 31, 156[160] David G. Kent, Michael R. Copley, Claudia Benz, Stefan Wo¨hrer, Brad J.Dykstra, Elaine Ma, John Cheyne, Yongjun Zhao, Michelle B. Bowie, YunZhao, Maura Gasparetto, Allen Delaney, Clayton Smith, Marco Marra, andConnie J. Eaves. Prospective isolation and molecular characterization of189hematopoietic stem cells with durable self-renewal potential. Blood, 113(25):6342–6350, 2009. doi:10.1182/blood-2008-12-192054. → page 31[161] Motonari Kondo, Irving L. Weissman, and Koichi Akashi. Identification ofclonogenic common lymphoid progenitors in mouse bone marrow. Cell, 91(5):661–672, 1997. doi:10.1016/S0092-8674(00)80453-5.[162] Faiyaz Notta, Sergei Doulatov, Elisa Laurenti, Armando Poeppl, IgorJurisica, and John E. Dick. Isolation of single human hematopoietic stemcells capable of long-term multilineage engraftment. Science, 333(6039):218–221, 2011. doi:10.1126/science.1201219. → pages 31, 135[163] T. R. Bradley and D. Metcalf. The growth of mouse bone marrow cells invitro. Australian Journal Of Experimental Biology And Medical Science,44:287, 1966. doi:10.1038/icb.1966.28. → page 31[164] Faiyaz Notta, Sasan Zandi, Naoya Takayama, Stephanie Dobson, Olga I.Gan, Gavin Wilson, Kerstin B. Kaufmann, Jessica McLeod, Elisa Laurenti,Cyrille F. Dunant, John D. McPherson, Lincoln D. Stein, Yigal Dror, andJohn E. Dick. Distinct routes of lineage development reshape the humanblood hierarchy across ontogeny. Science, 351(6269), 2016.doi:10.1126/science.aab2116. → pages 31, 156[165] Iain C. Macaulay, Valentine Svensson, Charlotte Labalette, LaurenFerreira, Fiona Hamey, Thierry Voet, Sarah A. Teichmann, and Ana Cvejic.Single-cell RNA-sequencing reveals a continuous spectrum ofdifferentiation in hematopoietic cells. Cell Reports, 14(4):966–977, 2016.doi:10.1016/j.celrep.2015.12.082. → pages 31, 156[166] Sonia Nestorowa, Fiona K. Hamey, Blanca Pijuan Sala, EvangeliaDiamanti, Mairi Shepherd, Elisa Laurenti, Nicola K. Wilson, David G.Kent, and Berthold Go¨ttgens. A single-cell resolution map of mousehematopoietic stem and progenitor cell differentiation. Blood, 128(8):e20–e31, 2016. doi:10.1182/blood-2016-05-716480. → pages 142, 156[167] Lars Velten, Simon F. Haas, Simon Raffel, Sandra Blaszkiewicz, SaifulIslam, Bianca P. Hennig, Christoph Hirche, Christoph Lutz, Eike C. Buss,Daniel Nowak, Tobias Boch, Wolf-Karsten Hofmann, Anthony D. Ho,Wolfgang Huber, Andreas Trumpp, Marieke A. G. Essers, and Lars MSteinmetz. Human haematopoietic stem cell lineage commitment is acontinuous process. Nature Cell Biology, 19:271, 2017.doi:10.1038/ncb3493. → pages 31, 163190[168] Arnav Mehta and David Baltimore. MicroRNAs as regulatory elements inimmune system logic. Nature Reviews Immunology, 16(5):279–294, 2016.doi:10.1038/nri.2016.40. → pages 32, 99[169] Sara Montagner, Lorenzo Deho´, and Silvia Monticelli. MicroRNAs inhematopoietic development. BMC Immunology, 15(1):14, 2014.doi:10.1186/1471-2172-15-14.[170] Francisco Navarro and Judy Lieberman. Small RNAs guide hematopoieticcell differentiation and function. The Journal of Immunology, 184(11):5939–5947, 2010. doi:10.4049/jimmunol.0902567.[171] R. M. O’Connell, D. S. Rao, and D. Baltimore. MicroRNA regulation ofinflammatory responses. Annual Review of Immunology, 30:295–312,2012. doi:10.1146/annurev-immunol-020711-075013.[172] Ryan M. O’Connell, Dinesh S. Rao, Aadel A. Chaudhuri, and DavidBaltimore. Physiological and pathological roles for microRNAs in theimmune system. Nature Reviews Immunology, 10(2):111–122, 2010.doi:10.1038/nri2708. → pages 32, 99[173] Shangqin Guo, Jun Lu, Rita Schlanger, Hao Zhang, Judy Y. Wang,Michelle C. Fox, Louise E. Purton, Heather H. Fleming, Bradley Cobb,Matthias Merkenschlager, Todd R. Golub, and David T. Scadden.MicroRNA miR-125a controls hematopoietic stem cell number.Proceedings of the National Academy of Sciences, 107(32):14229–14234,2010. doi:10.1073/pnas.0913574107. → pages 32, 140[174] Shangqin Guo and David T. Scadden. A microRNA regulating adulthematopoietic stem cells. Cell Cycle, 9(18):3637–3638, 2010.doi:10.4161/cc.9.18.13174.[175] A. G. Lisa Ooi, Debashis Sahoo, Maddalena Adorno, Yulei Wang, Irving L.Weissman, and Christopher Y. Park. MicroRNA-125b expandshematopoietic stem cells and enriches for the lymphoid-balanced andlymphoid-biased subsets. Proceedings of the National Academy ofSciences, 107(50):21505–21510, 2010. doi:10.1073/pnas.1016218107.[176] Edyta E. Wojtowicz, Marta A. Walasek, Mathilde J. C. Broekhuis, EllenWeersing, Martha Ritsema, Albertina Ausema, Leonid V. Bystrykh, andGerald de Haan. MicroRNA-125 family members exert a similar role in theregulation of murine hematopoiesis. Experimental Hematology, 42(10):909–918.e1, 2014. doi:10.1016/j.exphem.2014.06.010.191[177] Edyta E. Wojtowicz, Eric R. Lechman, Karin G. Hermans, Erwin M.Schoof, Erno Wienholds, Ruth Isserlin, Peter A. van Veelen, Mathilde J. C.Broekhuis, George M. C. Janssen, Aaron Trotman-Grant, Stephanie M.Dobson, Gabriela Krivdova, Jantje Elzinga, James Kennedy, Olga I. Gan,Ankit Sinha, Vladimir Ignatchenko, Thomas Kislinger, BertienDethmers-Ausema, Ellen Weersing, Mir Farshid Alemdehy, Hans W. J.de Looper, Gary D. Bader, Martha Ritsema, Stefan J. Erkeland, Leonid V.Bystrykh, John E. Dick, and Gerald de Haan. Ectopic miR-125a expressioninduces long-term repopulating stem cell capacity in mouse and humanhematopoietic progenitors. Cell Stem Cell, 19(3):383–396, 2016.doi:10.1016/j.stem.2016.06.008. → pages 32, 140[178] Eric R. Lechman, Bernhard Gentner, Peter van Galen, Alice Giustacchini,Massimo Saini, Francesco E. Boccalatte, Hidefumi Hiramatsu, UmbertoRestuccia, Angela Bachi, Veronique Voisin, Gary D. Bader, John E. Dick,and Luigi Naldini. Attenuation of miR-126 activity expands HSC in vivowithout exhaustion. Cell Stem Cell, 11(6):799–811, 2012.doi:10.1016/j.stem.2012.09.001. → pages 32, 140[179] Louis C. Dore, Julio D. Amigo, Camila O. dos Santos, Zhe Zhang, XiaowuGai, John W. Tobias, Duonan Yu, Alyssa M. Klein, Christine Dorman,Weisheng Wu, Ross C. Hardison, Barry H. Paw, and Mitchell J. Weiss. AGATA-1-regulated microRNA locus essential for erythropoiesis.Proceedings of the National Academy of Sciences, 105(9):3333–3338,2008. doi:10.1073/pnas.0712312105. → pages 32, 140[180] Kasper D. Rasmussen, Salvatore Simmini, Cei Abreu-Goodger, NenadBartonicek, Monica Di Giacomo, Daniel Bilbao-Cortes, Rastislav Horos,Marieke Von Lindern, Anton J. Enright, and Do´nal O’Carroll. ThemiR-144/451 locus is required for erythroid homeostasis. The Journal ofExperimental Medicine, 207(7):1351–1358, 2010.doi:10.1084/jem.20100458. → page 140[181] Li-Sheng Wang, Ling Li, Liang Li, Su Chu, Keh-Dong Shiang, Min Li,Hui-Yan Sun, Jun Xu, Feng-Jun Xiao, Guihua Sun, John J. Rossi, YinWeiHo, and Ravi Bhatia. MicroRNA-486 regulates normal erythropoiesis andenhances growth and modulates drug response in CML progenitors. Blood,125(8):1302–1313, 2015. doi:10.1182/blood-2014-06-581926. → pages32, 140[182] Chang-Zheng Chen, Ling Li, Harvey F. Lodish, and David P. Bartel.MicroRNAs modulate hematopoietic lineage differentiation. Science, 303192(5654):83–86, 2004. doi:10.1126/science.1091903. → pages32, 66, 67, 142[183] Jonathan B. Johnnidis, Marian H. Harris, Robert T. Wheeler, SandraStehling-Sun, Michael H. Lam, Oktay Kirak, Thijn R. Brummelkamp,Mark D. Fleming, and Fernando D. Camargo. Regulation of progenitor cellproliferation and granulocyte function by microRNA-223. Nature, 451(7182):1125–1129, 2008. doi:10.1038/nature06607. → pages 32, 142[184] Changchun Xiao, Dinis Pedro Calado, Gunther Galler, To-Ha Thai,Heide Christine Patterson, Jing Wang, Nikolaus Rajewsky, Timothy P.Bender, and Klaus Rajewsky. miR-150 controls B cell differentiation bytargeting the transcription factor c-Myb. Cell, 131(1):146–159, 2007.doi:10.1016/j.cell.2007.07.021. → page 32[185] Jun Lu, Shangqin Guo, Benjamin L. Ebert, Hao Zhang, Xiao Peng, JocelynBosco, Jennifer Pretz, Rita Schlanger, Judy Y. Wang, Raymond H. Mak,David M. Dombkowski, Frederic I. Preffer, David T. Scadden, and Todd R.Golub. microRNA-mediated control of cell fate inmegakaryocyte-erythrocyte progenitors. Developmental Cell, 14(6):843–853, 2008. doi:10.1016/j.devcel.2008.03.012. → page 32[186] Jimmy L. Zhao, Dinesh S. Rao, Ryan M. O’Connell, Yvette Garcia-Flores,and David Baltimore. MicroRNA-146a acts as a guardian of the qualityand longevity of hematopoietic stem cells in mice. eLife, 2:e00537, 2013.doi:10.7554/eLife.00537. → page 32[187] Konstantin D. Taganov, Mark P. Boldin, Kuang-Jung Chang, and DavidBaltimore. NF-κB-dependent induction of microRNA mir-146, aninhibitor targeted to signaling proteins of innate immune responses.Proceedings of the National Academy of Sciences, 103(33):12481–12486,2006. doi:10.1073/pnas.0605298103. → page 32[188] Li-Fan Lu, Mark P. Boldin, Ashutosh Chaudhry, Ling-Li Lin,Konstantin D. Taganov, Toshikatsu Hanada, Akihiko Yoshimura, DavidBaltimore, and Alexander Y. Rudensky. Function of miR-146a incontrolling Treg cell-mediated regulation of Th1 responses. Cell, 142(6):914–929, 2010. doi:10.1016/j.cell.2010.08.012. → page 32[189] Sergei Doulatov, Faiyaz Notta, Elisa Laurenti, and John E. Dick.Hematopoiesis: A human perspective. Cell Stem Cell, 10(2):120–136,2012. doi:10.1016/j.stem.2012.01.006. → page 32193[190] Kiyomi Taniguchi, Tomoharu Kajiyama, and Hideki Kambara. Quantitativeanalysis of gene expression in a single cell by qPCR. Nature Methods, 6(7):503–506, 2009. doi:10.1038/nmeth.1338. → pages 36, 73, 75, 96, 97[191] Luigi Warren, David Bryder, Irving L. Weissman, and Stephen R. Quake.Transcription factor profiling in individual hematopoietic progenitors bydigital RT-PCR. Proceedings of the National Academy of Sciences, 103(47):17807–17812, 2006. doi:10.1073/pnas.0608512103. → pages36, 43, 60, 73, 75, 97[192] Chatarina Larsson, Ida Grundberg, Ola Soderberg, and Mats Nilsson. Insitu detection and genotyping of individual mRNA molecules. NatureMethods, 7(5):395–397, 2010. doi:10.1038/nmeth.1448. → page 36[193] Martin Bengtsson, Martin Hemberg, Patrik Rorsman, and AndersSta˚hlberg. Quantification of mRNA in single cells and modelling ofRT-qPCR induced noise. BMC Molecular Biology, 9(1):63, 2008.doi:10.1186/1471-2199-9-63. → pages 36, 54, 73, 75[194] Christopher E. Sims and Nancy L. Allbritton. Analysis of singlemammalian cells on-chip. Lab on a Chip, 7(4):423–440, 2007.doi:10.1039/B615235J. → page 36[195] Richard N. Zare and Samuel Kim. Microfluidic platforms for single-cellanalysis. Annual Review of Biomedical Engineering, 12(1):187–201, 2010.doi:10.1146/annurev-bioeng-070909-105238. → page 36[196] Guoji Guo, Mikael Huss, Guo Qing Tong, Chaoyang Wang, Li Li Sun,Neil D. Clarke, and Paul Robson. Resolution of cell fate decisions revealedby single-cell gene expression analysis from zygote to blastocyst.Developmental Cell, 18(4):675–685, 2010.doi:10.1016/j.devcel.2010.02.012. → pages 36, 67, 73, 75[197] Maximilian Diehn, Robert W. Cho, Neethan A. Lobo, Tomer Kalisky,Mary Jo Dorie, Angela N. Kulp, Dalong Qian, Jessica S. Lam, Laurie E.Ailles, Manzhi Wong, Benzion Joshua, Michael J. Kaplan, Irene Wapnir,Frederick M. Dirbas, George Somlo, Carlos Garberoglio, Benjamin Paz,Jeannie Shen, Sean K. Lau, Stephen R. Quake, J. Martin Brown, Irving L.Weissman, and Michael F. Clarke. Association of reactive oxygen specieslevels and radioresistance in cancer stem cells. Nature, 458(7239):780–783, 2009. doi:10.1038/nature07733. → page 36194[198] Oleh I. Petriv, Kevin A. Heyries, Michael VanInsberghe, David Walker, andCarl L. Hansen. Methods for multiplex template sampling in digital PCRassays. PLOS ONE, 9(5):e98341, 2014.doi:10.1371/journal.pone.0098341. → page 43[199] Sohrab P. Shah, Ryan D. Morin, Jaswinder Khattra, Leah Prentice, TrevorPugh, Angela Burleigh, Allen Delaney, Karen Gelmon, Ryan Guliany,Janine Senz, Christian Steidl, Robert A. Holt, Steven Jones, Mark Sun,Gillian Leung, Richard Moore, Tesa Severson, Greg A. Taylor, Andrew E.Teschendorff, Kane Tse, Gulisa Turashvili, Richard Varhol, Rene L.Warren, Peter Watson, Yongjun Zhao, Carlos Caldas, David Huntsman,Martin Hirst, Marco A. Marra, and Samuel Aparicio. Mutational evolutionin a lobular breast tumour profiled at single nucleotide resolution. Nature,461(7265):809–813, 2009. doi:10.1038/nature08489. → pages 44, 69, 70[200] Oluseun Adewumi, Behrouz Aflatoonian, Lars Ahrlund-Richter, MichalAmit, Peter W. Andrews, Gemma Beighton, Paul A. Bello, NissimBenvenisty, Lorraine S. Berry, Simon Bevan, Barak Blum, Justin Brooking,Kevin G. Chen, Andre B.H. Choo, Gary A. Churchill, Marie Corbel, IvanDamjanov, Jon S. Draper, Petr Dvorak, Katarina Emanuelsson, Roland A.Fleck, Angela Ford, Karin Gertow, Marina Gertsenstein, Paul J. Gokhale,Rebecca S. Hamilton, Ales Hampl, Lyn E. Healy, Outi Hovatta, JohanHyllner, Marta P. Imreh, Joseph Itskovitz-Eldor, Jamie Jackson,Jacqueline L. Johnson, Mark Jones, Kehkooi Kee, Benjamin L. King,Barbara B. Knowles, Majlinda Lako, Franck Lebrin, Barbara S. Mallon,Daisy Manning, Yoav Mayshar, Ronald D.G. Mckay, Anna E. Michalska,Milla Mikkola, Masha Mileikovsky, Stephen L. Minger, Harry D. Moore,Christine L. Mummery, Andras Nagy, Norio Nakatsuji, Carmel M.O’Brien, Steve K.W. Oh, Cia Olsson, Timo Otonkoski, Kye-Yoon Park,Robert Passier, Hema Patel, Minal Patel, Roger Pedersen, Martin F. Pera,Marian S. Piekarczyk, Renee A. Reijo Pera, Benjamin E. Reubinoff,Allan J. Robins, Janet Rossant, Peter Rugg-Gunn, Thomas C. Schulz,Henrik Semb, Eric S. Sherrer, Henrike Siemen, Glyn N. Stacey, MiodragStojkovic, Hirofumi Suemori, Jin Szatkiewicz, Tikva Turetsky, Timo Tuuri,Steineke van den Brink, Kristina Vintersten, Sanna Vuoristo, Dorien Ward,Thomas A. Weaver, Lesley A. Young, and Weidong Zhang.Characterization of human embryonic stem cell lines by the internationalstem cell initiative. Nature Biotechnology, 25(7):803–816, 2007.doi:10.1038/nbt1318. → pages 48, 66[201] H. Georg Schulze, Stanislav O. Konorov, Nicolas J. Caron, James M. Piret,195Michael W. Blades, and Robin F. B. Turner. Assessing differentiationstatus of human embryonic stem cells noninvasively using ramanmicrospectroscopy. Analytical Chemistry, 82(12):5020–5027, 2010.doi:10.1021/ac902697q. → pages 48, 66[202] Tenneille E. Ludwig, Veit Bergendahl, Mark E. Levenstein, Junying Yu,Mitchell D. Probasco, and James A. Thomson. Feeder-independent cultureof human embryonic stem cells. Nature Methods, 3(8):637–646, 2006.doi:10.1038/nmeth902. → page 48[203] Bernard Tinland, Alain Pluen, Jean Sturm, and Gilbert Weill. Persistencelength of single-stranded DNA. Macromolecules, 30(19):5763–5765, 1997.doi:10.1021/ma970381+. → page 51[204] B. B. Lozzio, C. B. Lozzio, E. G. Bamberger, and A. S. Feliu. Amultipotential leukemia cell line (K-562) of human origin. Proceedings ofthe Society for Experimental Biology and Medicine, 166(4):546–50, 1981.→ pages 58, 64, 90, 98, 126[205] Martin Bengtsson, Anders Sta˚hlberg, Patrik Rorsman, and Mikael Kubista.Gene expression profiling in single cells from the pancreatic islets ofLangerhans reveals lognormal distribution of mRNA levels. GenomeResearch, 15(10):1388–1392, 2005. doi:10.1101/gr.3820805. → page 60[206] David P. Bartel. MicroRNAs: Genomics, biogenesis, mechanism, andfunction. Cell, 116(2):281–297, 2004.doi:10.1016/S0092-8674(04)00045-5. → page 63[207] Jin-Yun Yuan, Fang Wang, Jia Yu, Gui-Hua Yang, Xiao-Ling Liu, andJun-Wu Zhang. MicroRNA-223 reversibly regulates erythroid andmegakaryocytic differentiation of K562 cells. Journal of Cellular andMolecular Medicine, 13(11-12):4551–4559, 2009.doi:10.1111/j.1582-4934.2008.00585.x. → pages 64, 66, 67[208] Pablo Landgraf, Mirabela Rusu, Robert Sheridan, Alain Sewer, NicolaIovino, Alexei Aravin, Se´bastien Pfeffer, Amanda Rice, Alice O.Kamphorst, Markus Landthaler, Carolina Lin, Nicholas D. Socci, LeandroHermida, Valerio Fulci, Sabina Chiaretti, Robin Foa`, Julia Schliwka, UtaFuchs, Astrid Novosel, Roman-Ulrich Mu¨ller, Bernhard Schermer, UteBissels, Jason Inman, Quang Phan, Minchen Chien, David B. Weir, RuchiChoksi, Gabriella De Vita, Daniela Frezzetti, Hans-Ingo Trompeter, VeitHornung, Grace Teng, Gunther Hartmann, Miklos Palkovits, Roberto196Di Lauro, Peter Wernet, Giuseppe Macino, Charles E. Rogler, James W.Nagle, Jingyue Ju, F. Nina Papavasiliou, Thomas Benzing, Peter Lichter,Wayne Tam, Michael J. Brownstein, Andreas Bosio, Arndt Borkhardt,James J. Russo, Chris Sander, Mihaela Zavolan, and Thomas Tuschl. Amammalian microRNA expression atlas based on small RNA librarysequencing. Cell, 129(7):1401–1414, 2007.doi:10.1016/j.cell.2007.04.040. → pages 64, 100[209] Na Xu, Thales Papagiannakopoulos, Guangjin Pan, James A. Thomson,and Kenneth S. Kosik. MicroRNA-145 regulates OCT4, SOX2, and KLF4and represses pluripotency in human embryonic stem cells. Cell, 137(4):647–658, 2009. doi:10.1016/j.cell.2009.02.038. → page 67[210] Oliver Stegle, Sarah A. Teichmann, and John C. Marioni. Computationaland analytical challenges in single-cell transcriptomics. Nature ReviewsGenetics, 16(3):133–145, 2015. doi:10.1038/nrg3833. → page 72[211] Cole Trapnell. Defining cell types and states with single-cell genomics.Genome Research, 25(10):1491–1498, 2015. doi:10.1101/gr.190595.115.→ pages 72, 102[212] Naomi Habib, Yinqing Li, Matthias Heidenreich, Lukasz Swiech, InbalAvraham-Davidi, John J. Trombetta, Cynthia Hession, Feng Zhang, andAviv Regev. Div-Seq: Single-nucleus RNA-seq reveals dynamics of rareadult newborn neurons. Science, 2016. doi:10.1126/science.aad7038. →page 72[213] Zhigang Xue, Kevin Huang, Chaochao Cai, Lingbo Cai, Chun-yan Jiang,Yun Feng, Zhenshan Liu, Qiao Zeng, Liming Cheng, Yi E. Sun, Jia-yinLiu, Steve Horvath, and Guoping Fan. Genetic programs in human andmouse early embryos revealed by single-cell RNAsequencing. Nature, 500:593, 2013. doi:10.1038/nature12364. → page 72[214] Cole Trapnell, Davide Cacchiarelli, Jonna Grimsby, Prapti Pokharel,Shuqiang Li, Michael Morse, Niall J. Lennon, Kenneth J. Livak, Tarjei S.Mikkelsen, and John L. Rinn. The dynamics and regulators of cell fatedecisions are revealed by pseudotemporal ordering of single cells. NatureBiotechnology, 32(4):381–386, 2014. doi:10.1038/nbt.2859. → pages72, 105, 122, 137, 138[215] Siddharth S. Dey, Lennart Kester, Bastiaan Spanjaard, Magda Bienko, andAlexander van Oudenaarden. Integrated genome and transcriptome197sequencing of the same cell. Nature Biotechnology, 33(3):285–289, 2015.doi:10.1038/nbt.3129. → page 73[216] Tamar Hashimshony, Florian Wagner, Noa Sher, and Itai Yanai. CEL-seq:Single-cell RNA-seq by multiplexed linear amplification. Cell Reports, 2(3):666–673, 2012. doi:10.1016/j.celrep.2012.08.003.[217] Daniel Ramskold, Shujun Luo, Yu-Chieh Wang, Robin Li, Qiaolin Deng,Omid R. Faridani, Gregory A. Daniels, Irina Khrebtukova, Jeanne F.Loring, Louise C. Laurent, Gary P. Schroth, and Rickard Sandberg.Full-length mRNA-seq from single-cell levels of RNA and individualcirculating tumor cells. Nature Biotechnology, 30(8):777–782, 2012.doi:10.1038/nbt.2282. → page 73[218] Veronica Sanchez-Freire, Antje D. Ebert, Tomer Kalisky, Stephen R.Quake, and Joseph C. Wu. Microfluidic single-cell real-time PCR forcomparative analysis of gene expression patterns. Nature Protocols, 7(5):829–838, 2012. doi:10.1038/nprot.2012.021. → pages 73, 75[219] Michael VanInsberghe, Hans Zahn, Adam K. White, Oleh I. Petriv, andCarl L. Hansen. Highly multiplexed single-cell quantitative PCR. PLOSONE, 13(1):1–18, 01 2018. doi:10.1371/journal.pone.0191601. → pages75, 161[220] Marshall J. Levesque and Arjun Raj. Single-chromosome transcriptionalprofiling reveals chromosomal gene expression regulation. NatureMethods, 10:246, 2013. doi:10.1038/nmeth.2372. → page 75[221] Sydney M. Shaffer, Min-Tzu Wu, Marshall J. Levesque, and Arjun Raj.Turbo FISH: A method for rapid single molecule RNA FISH. PLOS ONE,8(9):e75120, 2013. doi:10.1371/journal.pone.0075120. → page 75[222] Kevin A. Heyries and Carl L. Hansen. Parylene C coating forhigh-performance replica molding. Lab on a Chip, 11(23):4122–4125,2011. doi:10.1039/C1LC20623K. → pages 76, 110[223] Jens Huft, Daniel J. Da Costa, David Walker, and Carl L. Hansen.Three-dimensional large-scale microfluidic integration by laser ablation ofinterlayer connections. Lab on a Chip, 10(18):2358–2365, 2010.doi:10.1039/C004051G. → pages 77, 90, 111[224] Anders Sta˚hlberg, Joakim Ha˚kansson, Xiaojie Xian, Henrik Semb, andMikael Kubista. Properties of the reverse transcription reaction in mRNA198quantification. Clinical Chemistry, 50(3):509–515, 2004.doi:10.1373/clinchem.2003.026161. → page 94[225] Malkhey Verma, Ehsan Ghayoor Karimiani, Richard J. Byers, SamrinaRehman, Hans V. Westerhoff, and Philip J. R. Day. Mathematicalmodelling of miRNA mediated BCR.ABL protein regulation in chronicmyeloid leukaemia vis-a-vis therapeutic strategies. Integrative Biology, 5(3):543–554, 2013. doi:10.1039/C3IB20230E. → pages 96, 97[226] Andrew McDavid, Greg Finak, and Raphael Gottardo. The contribution ofcell cycle to heterogeneity in single-cell RNA-seq data. NatureBiotechnology, 34(6):591–593, 2016. doi:10.1038/nbt.3498. → page 97[227] Ronald Palacios and Michael Steinmetz. IL3-dependent mouse clones thatexpress B-220 surface antigen, contain Ig genes in germ-line configuration,and generate B lymphocytes in vivo. Cell, 41(3):727–734, 1985.doi:10.1016/S0092-8674(85)80053-2. → page 99[228] N. Felli, L. Fontana, E. Pelosi, R. Botta, D. Bonci, F. Facchiano, F. Liuzzi,V. Lulli, O. Morsilli, S. Santoro, M. Valtieri, G. A. Calin, C. G. Liu,A. Sorrentino, C. M. Croce, and C. Peschle. MicroRNAs 221 and 222inhibit normal erythropoiesis and erythroleukemic cell growth via kitreceptor down-modulation. Proceedings of the National Academy ofSciences, 102(50):18081–6, 2005. doi:10.1073/pnas.0506216102. → page99[229] A. R. Green, E. DeLuca, and C. G. Begley. Antisense SCL suppressesself-renewal and enhances spontaneous erythroid differentiation of thehuman leukaemic cell line K562. The EMBO Journal, 10(13):4153–8,1991. doi:10.1002/j.1460-2075.1991.tb04993.x. → page 100[230] Elena Vigorito, Susan Kohlhaas, Dong Lu, and Rebecca Leyland.miR-155: an ancient regulator of the immune system. ImmunologicalReviews, 253(1):146–157, 2013. doi:10.1111/imr.12057. → page 100[231] Peter S. Linsley, Janell Schelter, Julja Burchard, Miho Kibukawa,Melissa M. Martin, Steven R. Bartz, Jason M. Johnson, Jordan M.Cummins, Christopher K. Raymond, Hongyue Dai, Nelson Chau, MicheleCleary, Aimee L. Jackson, Michael Carleton, and Lee Lim. Transcriptstargeted by the microRNA-16 family cooperatively regulate cell cycleprogression. Molecular and Cellular Biology, 27(6):2240–2252, 2007.doi:10.1128/mcb.02005-06. → page 102199[232] Cesar Seigi Fuziwara and Edna Teruko Kimura. Insights into regulation ofthe miR-17-92 cluster of miRNAs in cancer. Frontiers in Medicine, 2(64),2015. doi:10.3389/fmed.2015.00064. → page 102[233] Virginie Olive, Iris Jiang, and Lin He. miR-17-92, a cluster of miRNAs inthe midst of the cancer network. The International Journal of Biochemistry& Cell Biology, 42(8):1348–1354, 2010. doi:10.1016/j.biocel.2010.03.004.→ page 102[234] E. H. Simpson. The interpretation of interaction in contingency tables.Journal of the Royal Statistical Society. Series B (Methodological), 13(2):238–241, 1951. → page 102[235] Allon Wagner, Aviv Regev, and Nir Yosef. Revealing the vectors of cellularidentity with single-cell genomics. Nature Biotechnology, 34(11):1145–1160, 2016. doi:10.1038/nbt.3711. → page 105[236] Victoria Moignard, Steven Woodhouse, Laleh Haghverdi, Andrew J. Lilly,Yosuke Tanaka, Adam C. Wilkinson, Florian Buettner, Iain C. Macaulay,Wajid Jawaid, Evangelia Diamanti, Shin-Ichi Nishikawa, Nir Piterman,Valerie Kouskoff, Fabian J. Theis, Jasmin Fisher, and Berthold Go¨ttgens.Decoding the regulatory network of early blood development fromsingle-cell gene expression measurements. Nature Biotechnology, 33(3):269–276, 2015. doi:10.1038/nbt.3154. → page 105[237] Brian D. Brown, Mary Anna Venneri, Anna Zingale, Lucia Sergi Sergi, andLuigi Naldini. Endogenous microRNA regulation suppresses transgeneexpression in hematopoietic lineages and enables stable gene transfer.Nature Medicine, 12(5):585–591, 2006. doi:10.1038/nm1398. → page 106[238] Anita Schamberger and Tama´s I. Orba´n. 3′ isomiR species and DNAcontamination influence reliable quantification of microRNAs by stem-loopquantitative PCR. PLOS ONE, 9(8):e106315, 2014.doi:10.1371/journal.pone.0106315. → page 106[239] Sarah Chang, Robert J. Johnston, and Oliver Hobert. A transcriptionalregulatory cascade that controls left/right asymmetry in chemosensoryneurons of C. elegans. Genes & Development, 17(17):2123–2137, 2003.doi:10.1101/gad.1117903. → page 117[240] Heng Li and Richard Durbin. Fast and accurate short read alignment withBurrows-Wheeler transform. Bioinformatics, 25(14):1754–1760, 2009.doi:10.1093/bioinformatics/btp324. → page 117200[241] C. Vaz, H. M. Ahmad, P. Sharma, R. Gupta, L. Kumar, R. Kulshreshtha,and A. Bhattacharya. Analysis of microRNA transcriptome by deepsequencing of small RNA libraries of peripheral blood. BMC Genomics,11:288, 2010. doi:10.1186/1471-2164-11-288. → pages 119, 128[242] Q. Xiong, Y. Yang, H. Wang, J. Li, S. Wang, Y. Li, Y. Yang, K. Cai,X. Ruan, J. Yan, S. Hu, and X. Fang. Characterization of miRNomes inacute and chronic myeloid leukemia cell lines. Genomics ProteomicsBioinformatics, 12(2):79–91, 2014. doi:10.1016/j.gpb.2014.02.001. →pages 119, 128[243] Martin Hartmann, Charles G. Howes, David VanInsberghe, Hang Yu,Dipankar Bachar, Richard Christen, Rolf Henrik Nilsson, Steven J. Hallam,and William W. Mohn. Significant and persistent impact of timberharvesting on soil microbial communities in northern coniferous forests.The ISME Journal, 6(12):2199–2218, 2012. doi:10.1038/ismej.2012.84.→ page 120[244] Fabian Sievers, Andreas Wilm, David Dineen, Toby J Gibson, KevinKarplus, Weizhong Li, Rodrigo Lopez, Hamish McWilliam, MichaelRemmert, Johannes So¨ding, Julie D Thompson, and Desmond G Higgins.Fast, scalable generation of high-quality protein multiple sequencealignments using Clustal Omega. Molecular Systems Biology, 7(1), 2011.doi:10.1038/msb.2011.75. → page 120[245] Laurens van der Maaten and Geoffrey Hinton. Visualizing data usingt-SNE. Journal of Machine Learning Research, 9(Nov):2579–2605, 2008.→ pages 121, 136[246] Anthony J. Bell and Terrence J. Sejnowski. An information-maximizationapproach to blind separation and blind deconvolution. NeuralComputation, 7(6):1129–59, 1995. → pages 121, 138[247] Sean C. Bendall, Kara L. Davis, El-ad David Amir, Michelle D. Tadmor,Erin F. Simonds, Tiffany J. Chen, Daniel K. Shenfeld, Garry P. Nolan, andDana Pe’er. Single-cell trajectory detection uncovers progression andregulatory coordination in human B cell development. Cell, 157(3):714–725, 2014. doi:10.1016/j.cell.2014.04.005. → pages 122, 137[248] Cole Trapnell, David G. Hendrickson, Martin Sauvageau, Loyal Goff,John L. Rinn, and Lior Pachter. Differential analysis of gene regulation attranscript resolution with RNA-seq. Nature Biotechnology, 31(1):46–53,2013. doi:10.1038/nbt.2450. → page 123201[249] Vikram Agarwal, George W. Bell, Jin-Wu Nam, and David P. Bartel.Predicting effective microRNA target sites in mammalian mRNAs. eLife,4:e05005, 2015. doi:10.7554/eLife.05005. → pages 124, 145[250] Anton J. Enright, Bino John, Ulrike Gaul, Thomas Tuschl, Chris Sander,and Debora S. Marks. MicroRNA targets in Drosophila. Genome Biology,5(1):R1, 2003. doi:10.1186/gb-2003-5-1-r1. → pages 124, 145[251] M. Kawano, C. Kawazu, M. Lizio, H. Kawaji, P. Carninci, H. Suzuki, andY. Hayashizaki. Reduction of non-insert sequence reads by dimereliminator LNA oligonucleotide for small RNA deep sequencing.Biotechniques, 49(4):751–5, 2010. doi:10.2144/000113516. → page 127[252] Ute Bissels, Stefan Wild, Stefan Tomiuk, Angela Holste, Markus Hafner,Thomas Tuschl, and Andreas Bosio. Absolute quantification ofmicroRNAs by using a universal reference. RNA, 15(12):2375–2384, 2009.doi:10.1261/rna.1754109. → pages 130, 136[253] Victor Ambros, Rosalind C. Lee, Ann Lavanway, Peter T. Williams, andDavid Jewell. MicroRNAs and other tiny endogenous RNAs in C. elegans.Current Biology, 13(10):807–818, 2003.doi:10.1016/S0960-9822(03)00287-2. → page 135[254] Marc R. Friedla¨nder, Sebastian D. Mackowiak, Na Li, Wei Chen, andNikolaus Rajewsky. miRDeep2 accurately identifies known and hundredsof novel microRNA genes in seven animal clades. Nucleic Acids Research,40(1):37–52, 2012. doi:10.1093/nar/gkr688. → page 135[255] Dimitrios M. Vitsios, Elissavet Kentepozidou, Leonor Quintais, EliaBenito-Gutie´rrez, Stijn van Dongen, Matthew P. Davis, and Anton J.Enright. Mirnovo: genome-free prediction of microRNAs from small RNAsequencing data and single-cells using decision forests. Nucleic AcidsResearch, pages gkx836–gkx836, 2017. doi:10.1093/nar/gkx836. → page135[256] David J. H. F. Knapp, Colin A. Hammond, Paul H. Miller, Gabrielle M.Rabu, Philip A. Beer, Marketa Ricicova, Ve´ronique Lecault, DanielDa Costa, Michael VanInsberghe, Alice M. Cheung, Davide Pellacani,James Piret, Carl Hansen, and Connie J. Eaves. Dissociation of survival,proliferation, and state control in human hematopoietic stem cells. StemCell Reports, 8(1):152–162, 2017. doi:10.1016/j.stemcr.2016.12.003. →page 135202[257] Martin Ester, Hans-Peter Kriegel, Jo¨rg Sander, and Xiaowei Xu. Adensity-based algorithm for discovering clusters a density-based algorithmfor discovering clusters in large spatial databases with noise. InProceedings of the Second International Conference on KnowledgeDiscovery and Data Mining, KDD’96, pages 226–231. AAAI Press, 1996.→ pages 136, 137[258] Philipp Angerer, Laleh Haghverdi, Maren Bu¨ttner, Fabian J. Theis, CarstenMarr, and Florian Buettner. destiny: diffusion maps for large-scalesingle-cell data in R. Bioinformatics, 32(8):1241–1243, 2015.doi:10.1093/bioinformatics/btv715. → pages 137, 138[259] Ping Jin, Ena Wang, Jiaqiang Ren, Richard Childs, Jeong Won Shin, HanhKhuu, Francesco M. Marincola, and David F. Stroncek. Differentiation oftwo types of mobilized peripheral blood stem cells by microRNA andcDNA expression analysis. Journal of Translational Medicine, 6(1):39,2008. doi:10.1186/1479-5876-6-39. → page 140[260] Eric R. Lechman, Bernhard Gentner, Stanley W. K. Ng, Erwin M. Schoof,Peter van Galen, James A. Kennedy, Silvia Nucera, Fabio Ciceri, Kerstin B.Kaufmann, Naoya Takayama, Stephanie M Dobson, Aaron Trotman-Grant,Gabriela Krivdova, Janneke Elzinga, Amanda Mitchell, Bjo¨rn Nilsson,Karin G. Hermans, Kolja Eppert, Rene Marke, Ruth Isserlin, VeroniqueVoisin, Gary D. Bader, Peter W. Zandstra, Todd R. Golub, Benjamin L.Ebert, Jun Lu, Mark Minden, Jean C. Y. Wang, Luigi Naldini, and John E.Dick. miR-126 regulates distinct self-renewal outcomes in normal andmalignant hematopoietic stem cells. Cancer Cell, 29(2):214–228, 2016.doi:10.1016/j.ccell.2015.12.011.[261] Ryan M. O’Connell, Aadel A. Chaudhuri, Dinesh S. Rao, William S. J.Gibson, Alejandro B. Balazs, and David Baltimore. MicroRNAs enrichedin hematopoietic stem cells differentially regulate long-term hematopoieticoutput. Proceedings of the National Academy of Sciences, 107(32):14235–14240, 2010. doi:10.1073/pnas.1009798107. → page 140[262] Joel R. Neilson, Grace X.Y. Zheng, Christopher B. Burge, and Phillip A.Sharp. Dynamic regulation of miRNA expression in ordered stages ofcellular development. Genes & Development, 21(5):578–589, 2007.doi:10.1101/gad.1522907. → page 142[263] Patrick G. Gallagher, Marc Romana, William T. Tse, Samuel E. Lux, andBernard G. Forget. The human ankyrin-1 gene is selectively transcribed in203erythroid cell lines despite the presence of a housekeeping-like promoter.Blood, 96(3):1136–1143, 2000. → page 145[264] Gerhard Rank, Rosemary Sutton, Vikki Marshall, Rachel J. Lundie, JacintaCaddy, Tony Romeo, Kate Fernandez, Matthew P. McCormack, Brian M.Cooke, Simon J. Foote, Brendan S. Crabb, David J. Curtis, Douglas J.Hilton, Benjamin T. Kile, and Stephen M. Jane. Novel roles for erythroidAnkyrin-1 revealed through an ENU-induced null mouse mutant. Blood,113(14):3352–3362, 2009. doi:10.1182/blood-2008-08-172841. → page145[265] Elisa Laurenti, Sergei Doulatov, Sasan Zandi, Ian Plumb, Jing Chen, CraigApril, Jian-Bing Fan, and John E. Dick. The transcriptional architecture ofearly human hematopoiesis identifies multilevel control of lymphoidcommitment. Nature Immunology, 14(7):756–763, 2013.doi:10.1038/ni.2615. → page 156[266] Andre´ Go¨rgens, Stefan Radtke, Michael Mo¨llmann, Michael Cross, JanDu¨rig, Peter A Horn, and Bernd Giebel. Revision of the humanhematopoietic tree: Granulocyte subtypes derive from distincthematopoietic lineages. Cell Reports, 3(5):1539–1552, 2013.doi:10.1016/j.celrep.2013.04.025. → page 156[267] Bob Argiropoulos and R. Keith Humphries. Hox genes in hematopoiesisand leukemogenesis. Oncogene, 26(47):6766–6776, 2007.doi:10.1038/sj.onc.1210760. → pages 157, 166[268] A. H. Lund. miR-10 in development and cancer. Cell Death &Differentiation, 17(2):209–214, 2009. doi:10.1038/cdd.2009.58. → pages157, 166[269] Soraya Yekta, Clifford J. Tabin, and David P. Bartel. MicroRNAs in theHox network: an apparent link to posterior prevalence. Nature ReviewsGenetics, 9(10):789–796, 2008. doi:10.1038/nrg2400. → page 157[270] Wei-Fang Shen, Yu-Long Hu, Lalita Uttarwar, Emmanuelle Passegue, andCorey Largman. MicroRNA-126 regulates HOXA9 by binding to thehomeobox. Molecular and Cellular Biology, 28(14):4609–4619, 2008.doi:10.1128/mcb.01652-07. → page 157[271] Hans Zahn, Adi Steif, Emma Laks, Peter Eirew, Michael VanInsberghe,Sohrab P. Shah, Samuel Aparicio, and Carl L. Hansen. Scalablewhole-genome single-cell library preparation without preamplification.204Nature Methods, 14(2):167–173, 2017. doi:10.1038/nmeth.4140. → pages161, 164[272] Florian Buettner, Kedar N. Natarajan, F. Paolo Casale, Valentina Proserpio,Antonio Scialdone, Fabian J. Theis, Sarah A. Teichmann, John C. Marioni,and Oliver Stegle. Computational analysis of cell-to-cell heterogeneity insingle-cell RNA-sequencing data reveals hidden subpopulations of cells.Nature Biotechnology, 33(2):155–160, 2015. doi:10.1038/nbt.3102. →page 161[273] Aaron M. Streets, Xiannian Zhang, Chen Cao, Yuhong Pang, XinglongWu, Liang Xiong, Lu Yang, Yusi Fu, Liang Zhao, Fuchou Tang, and YanyiHuang. Microfluidic single-cell whole-transcriptome sequencing.Proceedings of the National Academy of Sciences, 111(19):7048–7053,2014. doi:10.1073/pnas.1402030111.[274] Matthew T. Walsh, Alexander P. Hsiao, Ho Suk Lee, Zhixia Liu, andXiaohua Huang. Capture and enumeration of mRNA transcripts fromsingle cells using a microfluidic device. Lab on a Chip, 15(14):2968–2980,2015. doi:10.1039/C5LC00445D.[275] Zhilong Yu, Sijia Lu, and Yanyi Huang. Microfluidic whole genomeamplification device for single cell sequencing. Analytical Chemistry, 86(19):9386–9390, 2014. doi:10.1021/ac5032176. → page 161[276] Aviv Regev, Sarah A. Teichmann, Eric S. Lander, Ido Amit, ChristopheBenoist, Ewan Birney, Bernd Bodenmiller, Peter J. Campbell, PieroCarninci, Menna Clatworthy, Hans Clevers, Bart Deplancke, Ian Dunham,James Eberwine, Roland Eils, Wolfgang Enard, Andrew Farmer, LarsFugger, Berthold Go¨ttgens, Nir Hacohen, Muzlifah Haniffa, MartinHemberg, Seung K. Kim, Paul Klenerman, Arnold Kriegstein, Ed Lein,Sten Linnarsson, Emma Lundberg, Joakim Lundeberg, Partha Majumder,John C. Marioni, Miriam Merad, Musa Mhlanga, Martijn Nawijn, MihaiNetea, Garry Nolan, Dana Pe’er, Anthony Phillipakis, Chris P. Ponting,Stephen R. Quake, Wolf Reik, Orit Rozenblatt-Rosen, Joshua R. Sanes,Rahul Satija, Ton N. Schumacher, Alex K. Shalek, Ehud Shapiro,Padmanee Sharma, Jay W. Shin, Oliver Stegle, Michael R. Stratton,Michael J. T. Stubbington, Fabian J. Theis, Matthias Uhlen, Alexander vanOudenaarden, Allon Wagner, Fiona M. Watt, Jonathan S. Weissman,Barbara J. Wold, Ramnik J. Xavier, Nir Yosef, and Participants Human CellAtlas Meeting. Science forum: The human cell atlas. eLife, 6:e27041,2017. doi:10.7554/eLife.27041. → pages 162, 165205[277] Susanne C. van den Brink, Fanny Sage, A´bel Ve´rtesy, Bastiaan Spanjaard,Josi Peterson-Maduro, Chloe´ S. Baron, Catherine Robin, and Alexandervan Oudenaarden. Single-cell sequencing reveals dissociation-inducedgene expression in tissue subpopulations. Nature Methods, 14:935, 2017.doi:10.1038/nmeth.4437. → page 162[278] Nicola K. Wilson, David G. Kent, Florian Buettner, Mona Shehata, Iain C.Macaulay, Fernando J. Calero-Nieto, Manuel Sa´nchez Castillo, Caroline A.Oedekoven, Evangelia Diamanti, Reiner Schulte, Chris P. Ponting, ThierryVoet, Carlos Caldas, John Stingl, Anthony R. Green, Fabian J. Theis, andBerthold Go¨ttgens. Combined single-cell functional and gene expressionanalysis resolves heterogeneity within stem cell populations. Cell StemCell, 16(6):712–724, 2015. doi:10.1016/j.stem.2015.04.004. → page 165[279] Jan Philipp Junker, Emily S. Noe¨l, Victor Guryev, Kevin A. Peterson, GopiShah, Jan Huisken, Andrew P. McMahon, Eugene Berezikov, JeroenBakkers, and Alexander van Oudenaarden. Genome-wide RNAtomography in the zebrafish embryo. Cell, 159(3):662–675, 2014.doi:10.1016/j.cell.2014.09.038. → page 165[280] Iain C. Macaulay, Wilfried Haerty, Parveen Kumar, Yang I. Li,Tim Xiaoming Hu, Mabel J. Teng, Mubeen Goolam, Nathalie Saurat, PaulCoupland, Lesley M. Shirley, Miriam Smith, Niels Van der Aa, RubyBanerjee, Peter D. Ellis, Michael A. Quail, Harold P. Swerdlow, MagdalenaZernicka-Goetz, Frederick J. Livesey, Chris P. Ponting, and Thierry Voet.G&T-seq: parallel sequencing of single-cell genomes and transcriptomes.Nature Methods, 12:519, 2015. doi:10.1038/nmeth.3370. → page 165[281] Florian Mertes, Bjo¨rn Lichtner, Heiner Kuhl, Mirjam Blattner, Jo¨rg Otte,Wasco Wruck, Bernd Timmermann, Hans Lehrach, and James Adjaye.Combined ultra-low input mRNA and whole-genome sequencing of humanembryonic stem cells. BMC Genomics, 16(1):925, 2015.doi:10.1186/s12864-015-2025-z. → page 165[282] Jason A. Reuter, Damek V. Spacek, Reetesh K. Pai, and Michael P. Snyder.Simul-seq: combined DNA and RNA sequencing for whole-genome andtranscriptome profiling. Nature Methods, 13:953, 2016.doi:10.1038/nmeth.4028. → page 166[283] Micheline Fromont-Racine, Edouard Bertrand, Raymond Pictet, andThierry Grange. A highly sensitive method for mapping the 5′ termini ofmRNAs. Nucleic Acids Research, 21(7):1683–1684, 1993. → page 166206[284] Ga´bor Bala´zsi, Alexander van Oudenaarden, and James J. Collins. Cellulardecision making and biological noise: From microbes to mammals. Cell,144(6):910–925, 2011. doi:10.1016/j.cell.2011.01.030. → page 167207Appendix ASupplemental data tables208Table A.1: miRNA co-expression significance. Spearman correlation coefficients rS, raw, and Benjamini-Hochbergcorrected p-values for each pairwise comparison for the K562 and BaF3 cells. Pairs in which either cell population didnot express both miRNAs are denoted with NA. * p < 0.05, ** p < 0.01, *** p < 0.001.miRNA 1 miRNA 2K562 Cells BaF3 CellsrS P-value FDR Sig. rS P-value FDR Sig.let-7b-5p miR-10a-5p -0.01834 0.86805 1 NA NA NAlet-7b-5p miR-150-5p -0.02162 1 1 0.01397 0.90567 0.92173let-7b-5p miR-155-5p NA NA NA 0.19843 0.07464 0.16221let-7b-5p miR-16-5p 0.02896 0.78751 1 0.30699 0.00462 0.01698 *let-7b-5p miR-17-5p -0.07362 0.48866 0.83071 0.55957 0.00002 0.00021 ***let-7b-5p miR-17-3p 0.10012 0.41132 0.75821 NA NA NAlet-7b-5p miR-181a-5p NA NA NA 0.15334 0.17358 0.31476let-7b-5p miR-200c-3p 0.01165 0.83881 1 0.20433 0.0672 0.14982let-7b-5p miR-20a-5p 0.04425 0.67545 0.96695 0.60022 0.00002 0.00021 ***let-7b-5p miR-221-3p -0.07542 0.50329 0.8462 0.36864 0.00094 0.00399 **let-7b-5p miR-223-3p 0.15689 0.13466 0.39621 0.16816 0.13116 0.25482let-7b-5p miR-23a-3p 0.14741 0.44952 0.79053 -0.12206 0.27998 0.43767let-7b-5p miR-24-3p 0.01509 0.89001 1 0.42286 0.00016 0.00104 **let-7b-5p miR-27a-3p 0.08092 0.44366 0.78929 0.26677 0.01588 0.05022let-7b-5p miR-29b-3p -0.09722 0.89601 1 0.13736 0.22278 0.35772let-7b-5p miR-451 -0.08677 1 1 NA NA NAlet-7b-5p miR-92a-3p 0.03464 0.74979 0.9805 0.36186 0.00092 0.00399 **let-7b-5p miR-93-5p 0.02406 0.82205 1 0.33551 0.00208 0.00857 **let-7b-5p sno142 -0.07174 1 1 0.1452 0.19556 0.3391miR-10a-5p miR-150-5p -0.06225 1 1 NA NA NAmiR-10a-5p miR-155-5p NA NA NA NA NA NAmiR-10a-5p miR-16-5p 0.0818 0.43046 0.77799 NA NA NAmiR-10a-5p miR-17-5p 0.19636 0.05602 0.21909 NA NA NAmiR-10a-5p miR-17-3p 0.06717 0.51027 0.84861 NA NA NAmiR-10a-5p miR-181a-5p NA NA NA NA NA NA209miRNA 1 miRNA 2K562 Cells BaF3 CellsrS P-value FDR Sig. rS P-value FDR Sig.miR-10a-5p miR-200c-3p 0.06245 0.53911 0.85629 NA NA NAmiR-10a-5p miR-20a-5p 0.16168 0.11808 0.35424 NA NA NAmiR-10a-5p miR-221-3p 0.12775 0.21826 0.55656 NA NA NAmiR-10a-5p miR-223-3p 0.20937 0.04226 0.18473 NA NA NAmiR-10a-5p miR-23a-3p 0.0481 0.57803 0.86819 NA NA NAmiR-10a-5p miR-24-3p 0.00699 0.94219 1 NA NA NAmiR-10a-5p miR-27a-3p -0.04767 0.64825 0.9446 NA NA NAmiR-10a-5p miR-29b-3p 0.02647 0.77547 0.99704 NA NA NAmiR-10a-5p miR-451 -0.17825 0.08226 0.26863 NA NA NAmiR-10a-5p miR-92a-3p 0.19535 0.0571 0.21909 NA NA NAmiR-10a-5p miR-93-5p 0.35915 0.00032 0.00245 ** NA NA NAmiR-10a-5p sno142 0.18964 0.08252 0.26863 NA NA NAmiR-150-5p miR-155-5p NA NA NA 0.14494 0.19744 0.3391miR-150-5p miR-16-5p 0.05642 0.70447 0.96695 0.04861 0.67023 0.81667miR-150-5p miR-17-5p 0 1 1 0.25077 0.02372 0.07169miR-150-5p miR-17-3p -0.05438 1 1 NA NA NAmiR-150-5p miR-181a-5p NA NA NA 0.0156 0.85881 0.91043miR-150-5p miR-200c-3p -0.05598 1 1 -0.01757 0.90607 0.92173miR-150-5p miR-20a-5p 0.01504 0.93379 1 0.22576 0.04288 0.11663miR-150-5p miR-221-3p 0.02361 0.88435 1 0.37269 0.00082 0.00385 **miR-150-5p miR-223-3p 0.02257 0.88441 1 0.08411 0.45994 0.62551miR-150-5p miR-23a-3p -0.02676 1 1 -0.0864 0.43492 0.61613miR-150-5p miR-24-3p 0.03385 0.82333 1 0.09028 0.42196 0.60406miR-150-5p miR-27a-3p 0.01185 0.94945 1 0.17925 0.10998 0.21996miR-150-5p miR-29b-3p -0.04784 1 1 -0.05971 0.60981 0.76812miR-150-5p miR-451 -0.0427 1 1 NA NA NAmiR-150-5p miR-92a-3p -0.11284 0.37548 0.73651 0.18247 0.10336 0.2098miR-150-5p miR-93-5p 0.03009 0.84717 1 0.27706 0.01166 0.03964 *miR-150-5p sno142 -0.03531 1 1 0.00197 0.98679 0.98679miR-155-5p miR-16-5p NA NA NA 0.31171 0.00462 0.01698 *210miRNA 1 miRNA 2K562 Cells BaF3 CellsrS P-value FDR Sig. rS P-value FDR Sig.miR-155-5p miR-17-5p NA NA NA 0.37523 0.00054 0.00262 **miR-155-5p miR-17-3p NA NA NA NA NA NAmiR-155-5p miR-181a-5p NA NA NA 0.08989 0.42022 0.60406miR-155-5p miR-200c-3p NA NA NA 0.02682 0.80021 0.87052miR-155-5p miR-20a-5p NA NA NA 0.42127 0.00016 0.00104 **miR-155-5p miR-221-3p NA NA NA 0.39789 0.00016 0.00104 **miR-155-5p miR-223-3p NA NA NA 0.21932 0.04892 0.12794miR-155-5p miR-23a-3p NA NA NA -0.20096 0.06476 0.14679miR-155-5p miR-24-3p NA NA NA 0.44425 0.00006 0.00051 ***miR-155-5p miR-27a-3p NA NA NA 0.19263 0.08388 0.17284miR-155-5p miR-29b-3p NA NA NA -0.01809 0.87755 0.91805miR-155-5p miR-451 NA NA NA NA NA NAmiR-155-5p miR-92a-3p NA NA NA 0.14856 0.1839 0.32481miR-155-5p miR-93-5p NA NA NA 0.1579 0.15552 0.28973miR-155-5p sno142 NA NA NA 0.03356 0.76785 0.85597miR-16-5p miR-17-5p 0.48779 0.00002 0.00025 *** 0.53415 0.00002 0.00021 ***miR-16-5p miR-17-3p 0.13404 0.19692 0.52857 NA NA NAmiR-16-5p miR-181a-5p NA NA NA 0.03758 0.74213 0.84108miR-16-5p miR-200c-3p 0.06043 0.55929 0.86436 0.19186 0.08378 0.17284miR-16-5p miR-20a-5p 0.46294 0.00002 0.00025 *** 0.47148 0.00002 0.00021 ***miR-16-5p miR-221-3p 0.11763 0.2536 0.58646 0.50524 0.00002 0.00021 ***miR-16-5p miR-223-3p 0.13775 0.183 0.49998 0.1568 0.16304 0.29964miR-16-5p miR-23a-3p -0.04206 0.68971 0.96695 0.05389 0.63651 0.79418miR-16-5p miR-24-3p 0.22285 0.0298 0.14248 0.39584 0.00026 0.00141 **miR-16-5p miR-27a-3p 0.27958 0.006 0.034 * 0.06824 0.54113 0.7009miR-16-5p miR-29b-3p -0.102 0.3238 0.68518 0.06968 0.54077 0.7009miR-16-5p miR-451 0.10584 0.31192 0.67216 NA NA NAmiR-16-5p miR-92a-3p 0.34285 0.00058 0.00423 ** 0.33119 0.00254 0.00987 **miR-16-5p miR-93-5p 0.26764 0.00848 0.04634 * 0.20998 0.05922 0.14497miR-16-5p sno142 0.11918 0.25042 0.58646 0.35775 0.0009 0.00399 **211miRNA 1 miRNA 2K562 Cells BaF3 CellsrS P-value FDR Sig. rS P-value FDR Sig.miR-17-5p miR-17-3p 0.19116 0.0634 0.23095 NA NA NAmiR-17-5p miR-181a-5p NA NA NA 0.11998 0.2842 0.43921miR-17-5p miR-200c-3p 0.3165 0.0017 0.01131 * 0.26741 0.015 0.04857 *miR-17-5p miR-20a-5p 0.95062 0.00002 0.00025 *** 0.87468 0.00002 0.00021 ***miR-17-5p miR-221-3p 0.14161 0.16802 0.4674 0.56847 0.00002 0.00021 ***miR-17-5p miR-223-3p 0.38338 0.00016 0.00144 ** 0.28268 0.01054 0.03675 *miR-17-5p miR-23a-3p 0.08977 0.3952 0.74648 0.13653 0.22358 0.35772miR-17-5p miR-24-3p 0.47306 0.00002 0.00025 *** 0.62441 0.00002 0.00021 ***miR-17-5p miR-27a-3p 0.36827 0.00022 0.00187 ** 0.38369 0.00054 0.00262 **miR-17-5p miR-29b-3p -0.04109 0.69341 0.96695 -0.01344 0.90367 0.92173miR-17-5p miR-451 0.19122 0.06202 0.23095 NA NA NAmiR-17-5p miR-92a-3p 0.73 0.00002 0.00025 *** 0.6381 0.00002 0.00021 ***miR-17-5p miR-93-5p 0.6028 0.00002 0.00025 *** 0.40305 0.00018 0.00111 **miR-17-5p sno142 0.10144 0.32692 0.68518 0.23187 0.03604 0.10428miR-17-3p miR-181a-5p NA NA NA NA NA NAmiR-17-3p miR-200c-3p 0.12929 0.22226 0.55747 NA NA NAmiR-17-3p miR-20a-5p 0.18571 0.07354 0.25572 NA NA NAmiR-17-3p miR-221-3p 0.11973 0.2465 0.58646 NA NA NAmiR-17-3p miR-223-3p 0.06105 0.55721 0.86436 NA NA NAmiR-17-3p miR-23a-3p 0.04677 0.72725 0.9805 NA NA NAmiR-17-3p miR-24-3p 0.06449 0.53267 0.85629 NA NA NAmiR-17-3p miR-27a-3p 0.16929 0.10304 0.3153 NA NA NAmiR-17-3p miR-29b-3p 0.05779 0.57283 0.86819 NA NA NAmiR-17-3p miR-451 0.12959 0.23004 0.56767 NA NA NAmiR-17-3p miR-92a-3p 0.03073 0.76673 0.99415 NA NA NAmiR-17-3p miR-93-5p 0.01661 0.87157 1 NA NA NAmiR-17-3p sno142 -0.1114 0.2159 0.55656 NA NA NAmiR-181a-5p miR-200c-3p NA NA NA -0.05822 0.72255 0.83275miR-181a-5p miR-20a-5p NA NA NA 0.2072 0.06186 0.14505miR-181a-5p miR-221-3p NA NA NA 0.01921 0.86357 0.91043212miRNA 1 miRNA 2K562 Cells BaF3 CellsrS P-value FDR Sig. rS P-value FDR Sig.miR-181a-5p miR-223-3p NA NA NA -0.01059 0.93283 0.93974miR-181a-5p miR-23a-3p NA NA NA 0.0662 0.60997 0.76812miR-181a-5p miR-24-3p NA NA NA 0.20815 0.06076 0.14497miR-181a-5p miR-27a-3p NA NA NA -0.08294 0.47122 0.63451miR-181a-5p miR-29b-3p NA NA NA 0.03603 0.69751 0.83174miR-181a-5p miR-451 NA NA NA NA NA NAmiR-181a-5p miR-92a-3p NA NA NA 0.0792 0.48218 0.6429miR-181a-5p miR-93-5p NA NA NA 0.01944 0.86213 0.91043miR-181a-5p sno142 NA NA NA 0.03095 0.78361 0.86643miR-200c-3p miR-20a-5p 0.34053 0.00096 0.00668 ** 0.21967 0.04826 0.12794miR-200c-3p miR-221-3p 0.11907 0.24962 0.58646 0.16396 0.14146 0.2672miR-200c-3p miR-223-3p 0.2116 0.04092 0.18414 0.03848 0.72865 0.83275miR-200c-3p miR-23a-3p -0.14081 0.39218 0.74648 0.16131 0.20446 0.3391miR-200c-3p miR-24-3p 0.2897 0.00442 0.02705 * 0.1653 0.13882 0.26591miR-200c-3p miR-27a-3p 0.20208 0.05152 0.21896 -0.04491 0.70331 0.83174miR-200c-3p miR-29b-3p -0.00642 0.95991 1 0.09545 0.419 0.60406miR-200c-3p miR-451 0.20693 0.05728 0.21909 NA NA NAmiR-200c-3p miR-92a-3p 0.30582 0.00302 0.01925 * -0.02836 0.80417 0.87052miR-200c-3p miR-93-5p 0.23122 0.02552 0.12595 0.08503 0.45016 0.62087miR-200c-3p sno142 -0.04786 0.60285 0.8955 0.04369 0.69891 0.83174miR-20a-5p miR-221-3p 0.09767 0.34412 0.702 0.52884 0.00002 0.00021 ***miR-20a-5p miR-223-3p 0.46828 0.00002 0.00025 *** 0.26382 0.0179 0.05533miR-20a-5p miR-23a-3p 0.11032 0.29386 0.6516 -0.05083 0.65515 0.81001miR-20a-5p miR-24-3p 0.52844 0.00002 0.00025 *** 0.6047 0.00002 0.00021 ***miR-20a-5p miR-27a-3p 0.37375 0.00028 0.00225 ** 0.39508 0.00028 0.00146 **miR-20a-5p miR-29b-3p -0.01187 0.90959 1 0.04838 0.67255 0.81667miR-20a-5p miR-451 0.18 0.0809 0.26863 NA NA NAmiR-20a-5p miR-92a-3p 0.77801 0.00002 0.00025 *** 0.6792 0.00002 0.00021 ***miR-20a-5p miR-93-5p 0.62247 0.00002 0.00025 *** 0.44941 0.00004 0.00036 ***miR-20a-5p sno142 0.05235 0.61079 0.89857 0.2761 0.01252 0.04153 *213miRNA 1 miRNA 2K562 Cells BaF3 CellsrS P-value FDR Sig. rS P-value FDR Sig.miR-221-3p miR-223-3p 0.00785 0.93117 1 0.4082 0.00022 0.00125 **miR-221-3p miR-23a-3p -0.00254 0.99277 1 -0.01256 0.90817 0.92173miR-221-3p miR-24-3p 0.07433 0.47602 0.81832 0.56305 0.00002 0.00021 ***miR-221-3p miR-27a-3p 0.03323 0.74431 0.9805 0.33882 0.00246 0.00984 **miR-221-3p miR-29b-3p -0.17177 0.09156 0.29184 0.02739 0.80651 0.87052miR-221-3p miR-451 0.09393 0.36492 0.72509 NA NA NAmiR-221-3p miR-92a-3p 0.07381 0.47406 0.81832 0.41215 0.00008 0.0006 ***miR-221-3p miR-93-5p 0.14955 0.14708 0.42459 0.46631 0.00004 0.00036 ***miR-221-3p sno142 -0.06498 0.54287 0.85629 0.20564 0.06416 0.14679miR-223-3p miR-23a-3p 0.09571 0.36474 0.72509 0.03469 0.75739 0.85128miR-223-3p miR-24-3p 0.2643 0.0102 0.05381 0.19837 0.07514 0.16221miR-223-3p miR-27a-3p 0.1849 0.07324 0.25572 0.19444 0.08062 0.17132miR-223-3p miR-29b-3p 0.17192 0.09666 0.30181 -0.14975 0.17844 0.31931miR-223-3p miR-451 -0.40328 0.00012 0.00115 ** NA NA NAmiR-223-3p miR-92a-3p 0.52651 0.00002 0.00025 *** 0.11303 0.3172 0.48471miR-223-3p miR-93-5p 0.45857 0.00004 0.00047 *** 0.12206 0.27706 0.43767miR-223-3p sno142 0.05763 0.57879 0.86819 0.24233 0.02938 0.08686miR-23a-3p miR-24-3p -0.03503 0.74081 0.9805 -0.11074 0.32408 0.48972miR-23a-3p miR-27a-3p 0.20295 0.05642 0.21909 -0.0983 0.389 0.57504miR-23a-3p miR-29b-3p 0.00457 0.73773 0.9805 -0.18014 0.2112 0.34606miR-23a-3p miR-451 0.12869 0.31066 0.67216 NA NA NAmiR-23a-3p miR-92a-3p -0.00764 0.94705 1 0.1106 0.32884 0.49145miR-23a-3p miR-93-5p -0.01809 0.86429 1 -0.08622 0.45196 0.62087miR-23a-3p sno142 0.05472 0.54239 0.85629 0.04029 0.71951 0.83275miR-24-3p miR-27a-3p 0.41548 0.00006 0.00061 *** 0.23076 0.03906 0.1098miR-24-3p miR-29b-3p 0.02557 0.80363 1 0.14395 0.19994 0.3391miR-24-3p miR-451 0.12961 0.2127 0.55656 NA NA NAmiR-24-3p miR-92a-3p 0.39972 0.00006 0.00061 *** 0.43469 0.00008 0.0006 ***miR-24-3p miR-93-5p 0.28457 0.00518 0.03048 * 0.28645 0.0098 0.03507 *miR-24-3p sno142 0.11033 0.29036 0.6516 0.1699 0.12898 0.25422214miRNA 1 miRNA 2K562 Cells BaF3 CellsrS P-value FDR Sig. rS P-value FDR Sig.miR-27a-3p miR-29b-3p -0.01557 0.88877 1 -0.14287 0.20282 0.3391miR-27a-3p miR-451 0.14515 0.16522 0.4674 NA NA NAmiR-27a-3p miR-92a-3p 0.21242 0.0391 0.18128 0.22972 0.03956 0.1098miR-27a-3p miR-93-5p 0.25211 0.01394 0.07109 0.21688 0.0528 0.13368miR-27a-3p sno142 0.08917 0.38522 0.74605 0.07279 0.51789 0.68382miR-29b-3p miR-451 -0.11784 0.25682 0.58646 NA NA NAmiR-29b-3p miR-92a-3p 0.03874 0.70783 0.96695 0.085 0.45036 0.62087miR-29b-3p miR-93-5p -0.01489 0.89227 1 -0.06665 0.55659 0.71412miR-29b-3p sno142 0.10708 0.33284 0.68816 0.04208 0.71323 0.83275miR-451 miR-92a-3p 0.0415 0.68645 0.96695 NA NA NAmiR-451 miR-93-5p 0.08681 0.40012 0.74656 NA NA NAmiR-451 sno142 -0.04761 0.70321 0.96695 NA NA NAmiR-92a-3p miR-93-5p 0.6429 0.00002 0.00025 *** 0.39878 0.00022 0.00125 **miR-92a-3p sno142 0.08152 0.43222 0.77799 0.20935 0.06018 0.14497miR-93-5p sno142 0.0652 0.53219 0.85629 0.21443 0.05308 0.13368215Table A.2: Names and sequences for all oligonucleotides used in the single-cell miRNA-seq work.Name Description SequencemiRNA3PE 3A 3’ Adapter for miRNA3 /5rApp/ ATCTCGTATGCCGTCTTCTGCTTGT /3ddC/miRNA3PE 5A 5’ Adapter for miRNA3rGrUrUrCrArGrArGrUrUrCrUrArCrArGrUrCrCrGrArCrGrArUrCrUrGrGrUrCrArAmiRNA3PE P2 Reverse PCR primer CAAGCAGAAGACGGCATACGAmiRNA3PE RTBlock LNA Adapter-adapter block +TA+CG+AG+AT+TTGAC+CA+GA+TC+GT+CmiRNA3PE iPCR01 Forward PCR Primer barcode 1 AATGATACGGCGACCACCGAGATCTACACCGTGATGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR02 Forward PCR Primer barcode 2 AATGATACGGCGACCACCGAGATCTACACCTGATCGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR03 Forward PCR Primer barcode 3 AATGATACGGCGACCACCGAGATCTACACGGGGTTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR04 Forward PCR Primer barcode 4 AATGATACGGCGACCACCGAGATCTACACCTGGGTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR05 Forward PCR Primer barcode 5 AATGATACGGCGACCACCGAGATCTACACAGCGCTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR06 Forward PCR Primer barcode 6 AATGATACGGCGACCACCGAGATCTACACCTTTTGGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR07 Forward PCR Primer barcode 7 AATGATACGGCGACCACCGAGATCTACACTGTTGGGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR08 Forward PCR Primer barcode 8 AATGATACGGCGACCACCGAGATCTACACAGCTAGGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR09 Forward PCR Primer barcode 9 AATGATACGGCGACCACCGAGATCTACACAGCATCGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR10 Forward PCR Primer barcode 10 AATGATACGGCGACCACCGAGATCTACACCGATTAGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR11 Forward PCR Primer barcode 11 AATGATACGGCGACCACCGAGATCTACACCATTCAGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR12 Forward PCR Primer barcode 12 AATGATACGGCGACCACCGAGATCTACACGGAACTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR13 Forward PCR Primer barcode 13 AATGATACGGCGACCACCGAGATCTACACACATCGGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR14 Forward PCR Primer barcode 14 AATGATACGGCGACCACCGAGATCTACACAAGCTAGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR15 Forward PCR Primer barcode 15 AATGATACGGCGACCACCGAGATCTACACCAAGTTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR16 Forward PCR Primer barcode 16 AATGATACGGCGACCACCGAGATCTACACGCCGGTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR17 Forward PCR Primer barcode 17 AATGATACGGCGACCACCGAGATCTACACCGGCCTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR18 Forward PCR Primer barcode 18 AATGATACGGCGACCACCGAGATCTACACTAGTTGGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR19 Forward PCR Primer barcode 19 AATGATACGGCGACCACCGAGATCTACACGCGTGGGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR20 Forward PCR Primer barcode 20 AATGATACGGCGACCACCGAGATCTACACGTATAGGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR21 Forward PCR Primer barcode 21 AATGATACGGCGACCACCGAGATCTACACCCTTGCGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR22 Forward PCR Primer barcode 22 AATGATACGGCGACCACCGAGATCTACACGCTGTAGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR23 Forward PCR Primer barcode 23 AATGATACGGCGACCACCGAGATCTACACATGGCAGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR24 Forward PCR Primer barcode 24 AATGATACGGCGACCACCGAGATCTACACTGACATGTTCAGAGTTCTACAGTCCGA216Name Description SequencemiRNA3PE iPCR25 Forward PCR Primer barcode 25 AATGATACGGCGACCACCGAGATCTACACGCCTAAGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR26 Forward PCR Primer barcode 26 AATGATACGGCGACCACCGAGATCTACACGTAGCCGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR27 Forward PCR Primer barcode 27 AATGATACGGCGACCACCGAGATCTACACAGTCTTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR28 Forward PCR Primer barcode 28 AATGATACGGCGACCACCGAGATCTACACTATCGTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR29 Forward PCR Primer barcode 29 AATGATACGGCGACCACCGAGATCTACACAATTATGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR30 Forward PCR Primer barcode 30 AATGATACGGCGACCACCGAGATCTACACCCGGTGGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR31 Forward PCR Primer barcode 31 AATGATACGGCGACCACCGAGATCTACACCATGGGGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR32 Forward PCR Primer barcode 32 AATGATACGGCGACCACCGAGATCTACACTCTGAGGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR33 Forward PCR Primer barcode 33 AATGATACGGCGACCACCGAGATCTACACAAGTGCGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR34 Forward PCR Primer barcode 34 AATGATACGGCGACCACCGAGATCTACACATTATAGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR35 Forward PCR Primer barcode 35 AATGATACGGCGACCACCGAGATCTACACCCAGCAGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR36 Forward PCR Primer barcode 36 AATGATACGGCGACCACCGAGATCTACACGGACGGGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR37 Forward PCR Primer barcode 37 AATGATACGGCGACCACCGAGATCTACACTGGTCAGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR38 Forward PCR Primer barcode 38 AATGATACGGCGACCACCGAGATCTACACTACAAGGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR39 Forward PCR Primer barcode 39 AATGATACGGCGACCACCGAGATCTACACTCGCTTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR40 Forward PCR Primer barcode 40 AATGATACGGCGACCACCGAGATCTACACGAGAGTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR41 Forward PCR Primer barcode 41 AATGATACGGCGACCACCGAGATCTACACCCGTATGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR42 Forward PCR Primer barcode 42 AATGATACGGCGACCACCGAGATCTACACATCGTGGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR43 Forward PCR Primer barcode 43 AATGATACGGCGACCACCGAGATCTACACCCACTCGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR44 Forward PCR Primer barcode 44 AATGATACGGCGACCACCGAGATCTACACCAGCAGGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR45 Forward PCR Primer barcode 45 AATGATACGGCGACCACCGAGATCTACACCGCGGCGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR46 Forward PCR Primer barcode 46 AATGATACGGCGACCACCGAGATCTACACGAATGAGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR47 Forward PCR Primer barcode 47 AATGATACGGCGACCACCGAGATCTACACGCGCCAGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR48 Forward PCR Primer barcode 48 AATGATACGGCGACCACCGAGATCTACACCTCTACGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR49 Forward PCR Primer barcode 49 AATGATACGGCGACCACCGAGATCTACACCACTGTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR50 Forward PCR Primer barcode 50 AATGATACGGCGACCACCGAGATCTACACATGTTTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR51 Forward PCR Primer barcode 51 AATGATACGGCGACCACCGAGATCTACACGTCCTTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR52 Forward PCR Primer barcode 52 AATGATACGGCGACCACCGAGATCTACACATCAGTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR53 Forward PCR Primer barcode 53 AATGATACGGCGACCACCGAGATCTACACTAGGATGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR54 Forward PCR Primer barcode 54 AATGATACGGCGACCACCGAGATCTACACTGAGTGGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR55 Forward PCR Primer barcode 55 AATGATACGGCGACCACCGAGATCTACACTTGCGGGTTCAGAGTTCTACAGTCCGA217Name Description SequencemiRNA3PE iPCR56 Forward PCR Primer barcode 56 AATGATACGGCGACCACCGAGATCTACACGGTTTCGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR57 Forward PCR Primer barcode 57 AATGATACGGCGACCACCGAGATCTACACTAAGGCGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR58 Forward PCR Primer barcode 58 AATGATACGGCGACCACCGAGATCTACACTCGGGAGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR59 Forward PCR Primer barcode 59 AATGATACGGCGACCACCGAGATCTACACTTCGAAGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR60 Forward PCR Primer barcode 60 AATGATACGGCGACCACCGAGATCTACACGCGGACGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR61 Forward PCR Primer barcode 61 AATGATACGGCGACCACCGAGATCTACACATTGGCGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR62 Forward PCR Primer barcode 62 AATGATACGGCGACCACCGAGATCTACACTGCTTTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR63 Forward PCR Primer barcode 63 AATGATACGGCGACCACCGAGATCTACACCCTATTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR64 Forward PCR Primer barcode 64 AATGATACGGCGACCACCGAGATCTACACTCTTCTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR65 Forward PCR Primer barcode 65 AATGATACGGCGACCACCGAGATCTACACATAGATGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR66 Forward PCR Primer barcode 66 AATGATACGGCGACCACCGAGATCTACACCGCCTGGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR67 Forward PCR Primer barcode 67 AATGATACGGCGACCACCGAGATCTACACCTAAGGGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR68 Forward PCR Primer barcode 68 AATGATACGGCGACCACCGAGATCTACACTTATTCGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR69 Forward PCR Primer barcode 69 AATGATACGGCGACCACCGAGATCTACACTGGAGCGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR70 Forward PCR Primer barcode 70 AATGATACGGCGACCACCGAGATCTACACCTTCGAGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR71 Forward PCR Primer barcode 71 AATGATACGGCGACCACCGAGATCTACACGGAGAAGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR72 Forward PCR Primer barcode 72 AATGATACGGCGACCACCGAGATCTACACTTTCACGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR73 Forward PCR Primer barcode 73 AATGATACGGCGACCACCGAGATCTACACGATCTGGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR74 Forward PCR Primer barcode 74 AATGATACGGCGACCACCGAGATCTACACGCATTTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR75 Forward PCR Primer barcode 75 AATGATACGGCGACCACCGAGATCTACACGTTTGTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR76 Forward PCR Primer barcode 76 AATGATACGGCGACCACCGAGATCTACACCTATCTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR77 Forward PCR Primer barcode 77 AATGATACGGCGACCACCGAGATCTACACGCTCATGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR78 Forward PCR Primer barcode 78 AATGATACGGCGACCACCGAGATCTACACGCCATGGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR79 Forward PCR Primer barcode 79 AATGATACGGCGACCACCGAGATCTACACTTCTCGGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR80 Forward PCR Primer barcode 80 AATGATACGGCGACCACCGAGATCTACACTCCGTCGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR81 Forward PCR Primer barcode 81 AATGATACGGCGACCACCGAGATCTACACTGTGCCGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR82 Forward PCR Primer barcode 82 AATGATACGGCGACCACCGAGATCTACACTGCCGAGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR83 Forward PCR Primer barcode 83 AATGATACGGCGACCACCGAGATCTACACAAACCTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR84 Forward PCR Primer barcode 84 AATGATACGGCGACCACCGAGATCTACACGGCCACGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR85 Forward PCR Primer barcode 85 AATGATACGGCGACCACCGAGATCTACACTCAAGTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR86 Forward PCR Primer barcode 86 AATGATACGGCGACCACCGAGATCTACACCGTACGGTTCAGAGTTCTACAGTCCGA218Name Description SequencemiRNA3PE iPCR87 Forward PCR Primer barcode 87 AATGATACGGCGACCACCGAGATCTACACAGATGTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR88 Forward PCR Primer barcode 88 AATGATACGGCGACCACCGAGATCTACACGATGCTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR89 Forward PCR Primer barcode 89 AATGATACGGCGACCACCGAGATCTACACAGGAATGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR90 Forward PCR Primer barcode 90 AATGATACGGCGACCACCGAGATCTACACAAAATGGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR91 Forward PCR Primer barcode 91 AATGATACGGCGACCACCGAGATCTACACATTCCGGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR92 Forward PCR Primer barcode 92 AATGATACGGCGACCACCGAGATCTACACTATATCGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR93 Forward PCR Primer barcode 93 AATGATACGGCGACCACCGAGATCTACACCAGGCCGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR94 Forward PCR Primer barcode 94 AATGATACGGCGACCACCGAGATCTACACGGTAGAGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR95 Forward PCR Primer barcode 95 AATGATACGGCGACCACCGAGATCTACACTTGACTGTTCAGAGTTCTACAGTCCGAmiRNA3PE iPCR96 Forward PCR Primer barcode 96 AATGATACGGCGACCACCGAGATCTACACCGAAACGTTCAGAGTTCTACAGTCCGARmiR-01 Spike-in seq24416 /5Phos/rCrCrGrArCrCrGrUrArGrUrCrUrGrCrGrUrArCrArUrARmiR-02 Spike-in seq25110 /5Phos/rUrUrCrCrGrGrCrGrUrCrGrArCrGrArArUrCrGrArArURmiR-03 Spike-in seq28644 /5Phos/rGrUrCrCrGrCrUrGrUrUrGrUrCrGrCrUrArArCrGrArURmiR-04 Spike-in seq28745 /5Phos/rCrArUrArArCrCrUrCrGrArCrGrGrUrGrArUrCrGrCrURmiR-05 Spike-in seq36602 /5Phos/rUrArUrArUrArCrGrCrUrCrArGrCrGrCrArArGrCrCrGRmiR-06 Spike-in seq47497 /5Phos/rGrCrGrArUrUrCrGrUrArGrCrGrArArUrArGrArGrUrGRmiR-07 Spike-in seq47735 /5Phos/rGrUrUrArArCrCrGrCrGrUrArUrArGrGrCrGrUrArUrCRmiR-08 Spike-in seq93492 /5Phos/rArUrCrUrArUrGrArUrCrGrArCrGrArGrArCrGrCrGrURmiR-09 Spike-in seq95411 /5Phos/rArGrUrCrCrGrUrArCrCrGrCrUrArUrCrGrGrUrArGrARmiR-10 Spike-in seq98395 /5Phos/rArArCrCrArCrArCrGrCrCrGrArCrGrArArUrArArUrC/5Phos/r(N1:25252525)r(N1)r(N1)r(N1)r(N1)r(N1)r(N1)r(N1)RmiRN 22-mer RNA randomer r(N1)r(N1)r(N1)r(N1)r(N1)r(N1)r(N1)r(N1)r(N1)r(N1)r(N1)r(N1)r(N1)r(N1)miRNA3PE QF F Primer for library quant AATGATACGGCGACCACCGAmiRNA3PE QR R Primer for library quant CAAGCAGAAGACGGCATACGA219

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

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"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0365804/manifest

Comment

Related Items