UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

A programmable droplet-based microfluidic device for multiparameter single-cell analysis Leung, Kaston 2013

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

Item Metadata


24-ubc_2013_spring_leung_kaston.pdf [ 53.73MB ]
JSON: 24-1.0073525.json
JSON-LD: 24-1.0073525-ld.json
RDF/XML (Pretty): 24-1.0073525-rdf.xml
RDF/JSON: 24-1.0073525-rdf.json
Turtle: 24-1.0073525-turtle.txt
N-Triples: 24-1.0073525-rdf-ntriples.txt
Original Record: 24-1.0073525-source.json
Full Text

Full Text

A PROGRAMMABLE DROPLET-BASED MICROFLUIDIC DEVICE FOR MULTIPARAMETER SINGLE-CELL ANALYSIS by Kaston Leung B.Sc., Queen’s University, 2001 M.Sc., The University of Alberta, 2005  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  Doctor of Philosophy in THE FACULTY OF GRADUATE STUDIES (Electrical and Computer Engineering)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)  January 2013  © Kaston Leung, 2013  Abstract The ability of microfluidic systems to perform biological analysis with greater sensitivity, lower cost, and higher throughput relative to conventional methods has now been widely demonstrated. Despite this transformative potential, application innovation and user adoption in biological research have lagged due to limited access to specialized fabrication facilities and expertise. In analogy to how the development of programmable integrated circuits has resulted in the ubiquity and utility of this technology among a broad community of developers and non-expert users, the advancement of programmable microfluidic devices stands to dramatically enhance the pervasiveness and impact of microfluidic systems. This thesis describes the development and application of a microfluidic device that combines the reconfigurable flow-routing capabilities of integrated microvalve technology with the sample compartmentalization inherent to mass transport in droplets to achieve programmable fluidhandling functionality. The device allows for the execution of user-defined multistep reaction protocols in an array of individually addressable nanolitre-volume storage chambers by consecutively merging programmable sequences of picolitre-volume droplets containing reagents or phenotypically sorted single cells. This functionality is enabled by “flow-controlled wetting,” a novel droplet docking and merging mechanism that exploits the physics of droplet flow through a channel to control the precise location of droplet wetting. The device also allows for automated cross-contamination-free recovery of reaction products from individual chambers for downstream analysis. The combined features of programmability, addressability, and selective recovery provide a general hardware platform that can be reprogrammed for multiple applications. This versatility is demonstrated by implementing multiple analyses on phenotypically sorted single cells including monoclonal culture, genomic PCR, whole genome amplification and whole transcriptome amplification.  These capabilities have been applied to a diverse range of  biological samples for applications ranging from the identification of microbial community members in environmental samples to the determination of mutation frequencies in human cancer at the single-cell level.  ii  Preface I designed and fabricated the microfluidic device that is described in this thesis, performed all on-chip experiments, developed all image analysis software for data acquisition from fluorescent images of the device, and analyzed data. Contributions from collaborators are noted below. My supervisor Carl Hansen provided suggestions at all stages of this work. Hans Zahn assisted in the design of the device elution nozzle and designed and built the 3-axis robot that allows for automated elution. Timothy Leaver designed an early version of the device on which the architecture of the current design is based. Bud Homsy provided mathematical formulae that elucidate the mechanism responsible for flow-controlled wetting. In Sections 4.1.1 and 4.1.2, bacterial strains and PCR assays designed by Nat Brown were used. In Section, Adam Quiring performed the majority of the DNA sequencing and sequencing data analysis. In Section 4.1.4, environmental samples obtained by Antoine P. Pagé and Steven J. Hallam were used. Kishori M. Konwar, Niels W. Hanson, Antoine P. Pagé, Chien-Chi Lo, Patrick S. Chain, and Steven J. Hallam performed analysis of sequencing data produced from all environmental samples. In Section 4.2, Jas Khattra prepared tumor nuclei samples for on-chip experiments and performed all off-chip processing and analysis of PCR amplicons generated on-chip. In Section 4.3, Kevin Heyries prepared whole transcriptome amplification reagents that were used on-chip and performed qPCR analysis of on-chip reaction products. Chapters 2 and 3, and Section 4.1 excluding have been published as Leung K, Zahn H, Leaver T, Konwar KM, Hanson NW, Pagé AP, Lo CC, Chain PS, Hallam SJ, Hansen CL (2012) A programmable droplet-based microfluidic device applied to multiparameter analysis of single microbes and microbial communities. Proc. Natl. Acad. Sci. USA 109:7665-7670. Relative contributions made by collaborators are listed above. Section is being prepared as a manuscript. iii  This work has resulted in filing of US provisional patent and a product based on this work is currently in development at Fluidigm Corporation.  iv  Table of Contents Abstract ........................................................................................................................................... ii	
   Preface ............................................................................................................................................ iii	
   Table of Contents ............................................................................................................................ v	
   List of Tables ................................................................................................................................. ix	
   List of Figures ................................................................................................................................. x	
   Acknowledgements ...................................................................................................................... xiii	
   Chapter 1 : Introduction .................................................................................................................. 1	
   Valve-based microfluidics ................................................................................................. 2	
   Droplet-based microfluidics ............................................................................................... 3	
   Programmable microfluidics .............................................................................................. 4	
   Single-cell isolation and analysis ....................................................................................... 6	
   Statement of research ......................................................................................................... 8	
   Chapter 2 : Device design and implementation .............................................................................. 9	
   Requirements and considerations for droplet-based flow in microfluidic channels .......... 9	
   General device architecture and operation ....................................................................... 11	
   Droplet formation and transport ....................................................................................... 15	
   Reagent metering ...................................................................................................... 15	
   Single-cell sorting ..................................................................................................... 15	
   Droplet transport .............................................................................................................. 17	
    v  2.5	
   Droplet merging and storage by “flow-controlled wetting” ............................................ 18	
   Velocity-dependent droplet wetting .......................................................................... 19	
   Storage element design and operation ...................................................................... 20	
   Sample recovery from selected storage chambers ........................................................... 28	
   Chapter 3 : Characterization of fluid-handling capabilities .......................................................... 33	
   Formulation accuracy and precision ................................................................................ 33	
   Prevention of protein adsorption onto droplet interfaces ................................................. 33	
   Quantification of cross-contamination during formulation and elution........................... 37	
   Elution efficiency ............................................................................................................. 40	
   Chapter 4 : Biological applications ............................................................................................... 41	
   Multi-parameter analysis of single microbes and microbial aggregates .......................... 41	
   Sorting and culture of single bacteria ....................................................................... 42	
   PCR-based genotyping of single sorted microbes .................................................... 45	
   Whole genome amplification of single cells ............................................................. 47	
   PCR-based whole genome amplification of single microbes ............................ 48	
   MDA-based whole genome amplification of single microbes........................... 51	
   Contamination ............................................................................................. 54	
   Exceptionally low representational bias in nanolitre MDA ........................ 57	
   Environmental genomics........................................................................................... 63	
   PCR-based genotyping of single human tumour cell nuclei ............................................ 67	
   Single-cell whole transcriptome amplification ................................................................ 74	
   vi  Chapter 5 : Conclusions ................................................................................................................ 81	
   Contributions to the state of the art .................................................................................. 81	
   Limitations and recommendations for technical improvements ...................................... 83	
   Future applications ........................................................................................................... 85	
   References ..................................................................................................................................... 87	
   Appendices .................................................................................................................................... 99	
   Appendix A : Microfluidic device fabrication and operation .................................................. 99	
   Appendix B : Methods for on-chip PCR, RT-PCR, and WTA ............................................... 100	
   Appendix C : Methods for on-chip bacterial culture ............................................................. 101	
   Appendix D : Methods for on-chip PCR-based genotyping of bacteria ................................ 102	
   Appendix E : Methods for PCR-based WGA ........................................................................ 103	
   Appendix F : Methods for MDA-based WGA....................................................................... 104	
   Appendix G : Preparation of environmental samples ............................................................ 105	
   Appendix H : Sequencing data analysis for environmental samples ..................................... 106	
   Appendix I : Kernal density of %GC content for individual ENV1 samples ........................ 109	
   Appendix J : Kernal density of %GC content for individual ENV2 samples ........................ 110	
   Appendix K : Kernal density of %GC content for individual ENV3 samples ...................... 111	
   Appendix L : Taxonomic assignments derived from sequencing data for individual ENV1 samples .................................................................................................................................... 112	
   Appendix M : Taxonomic assignments derived from sequencing data for individual ENV2 samples .................................................................................................................................... 113	
   vii  Appendix N : Taxonomic assignments derived from sequencing data for individual ENV3 samples .................................................................................................................................... 114	
    viii  List of Tables Table 1: Sequencing statistics for PCR-based WGA of single E. coli…………………………51 Table 2: Sequencing statistics for MDA-based WGA of single E. coli………………………..54 Table 3: Mutational frequencies obtained from sequencing reads for 7 single nuclei…………73  ix  List of Figures Figure 1: Programmable microfluidic reaction array…………………………………………… 12 Figure 2: Time course images of reagent metering……………………………………………. 16 Figure 3: Storage element geometry for droplet immobilization and coalescence by flowcontrolled wetting………………………………………………………………………………. 21 Figure 4: Finite element simulation of the flow velocity through storage element…………….. 22 Figure 5: Time course images of droplet merging and immobilization at the critical incoming velocity………………………………………………………………………………………….. 23 Figure 6: Time course images of droplet merging and immobilization below the critical incoming velocity……………………………………………………………………………….. 25 Figure 7: Time course images of droplet merging and immobilization above the critical incoming velocity………………………………………………………………………………………….. 27 Figure 8: Micrographs of stored droplets……………………………………………………….. 28 Figure 9: Stored droplet elution………………………………………………………………… 30 Figure 10: Photograph of 3-axis robotic chip-holder…………………………………………… 31 Figure 11: Photograph of microfuge tubes containing blue food dye eluted from a device…………………………………………………………………………………………… 32 Figure 12: Addressable and accurate formulation……………………………………………… 34 Figure 13: Fluorescent micrographs of stored droplets of FITC-labeled BSA using different aqueous and fluorous surfactants……………………………………………………………….. 36 Figure 14: Efficient on-chip qPCR with single molecule sensitivity……………………………38 Figure 15: Low cross-contamination during formulation and elution of stored droplets………. 39 Figure 16: On-chip culture of single sorted bacteria…………………………………………….44 x  Figure 17: 16S rRNA qPCR of single sorted S. typhimurium and E. coli……………………… 46 Figure 18: Strain-specific 16S rRNA qPCR of single sorted E. coli…………………………… 47 Figure 19: 16S rRNA copy number yielded from PCR-based WGA reactions…………………50 Figure 20: 16S rRNA copy number yielded from MDA-based WGA reactions……………….. 52 Figure 21: Copy number of 10 loci yielded from microfluidic single-cell MDA reactions performed without DTT………………………………………………………………………… 54 Figure 22: Read coverage of the Delftia acidovorans reference genome by sequencing data from two separate single - E. coli microfluidic MDA reactions……………………………………… 55 Figure 23: Read coverage of the E. coli reference genome for MDA-based WGA reactions….. 58 Figure 24: Overlaid normalized read coverage of the E. coli reference genome by sequencing data from two separate single-E. coli nanolitre MDA reactions………………………………... 59 Figure 25: Reference genome coverage versus mean sequencing coverage depth……………... 60 Figure 26: Histogram showing coverage depth of reference genome for each MDA reaction type at a mean coverage depth of 16x………………………………………………………………... 61 Figure 27: Summary of taxonomic profiles uncovered in metagenomes of 67 WGA samples originating from three distinct environments…………………………………………………… 65 Figure 28: Micrographs of primary tumour cell nuclei showing the morphological heterogeneity of the sample……………………………………………………………………………………. 69 Figure 29: Micrographs of a primary breast cancer pleural effusion cell nucleus in the cellsorting module and a stored droplet…………………………………………………………….. 70 Figure 30: RNase P qPCR of single sorted primary breast cancer pleural effusion cell nuclei… 70 Figure 31: qPCR curves for 6-plex PCR of single sorted primary breast cancer pleural effusion cell nuclei……………………………………………………………………………………….. 71  xi  Figure 32: Capillary electrophoresis plots of PCR amplicons for 5 somatic mutation loci from 4 on-chip single-nuclei multiplex PCR reactions………………………………………………… 72 Figure 33: Read coverage binned by chromosome for 5 loci amplicons from purified genomic DNA and on-chip multiplex PCR of a single primary breast cancer pleural effusion cell nucleus………………………………………………………………………………………….. 73 Figure 34: GAPDH qRT-PCR of dilutions of purified RNA……………………………………75 Figure 35: Quantification of GAPDH cDNA in WTA product by qPCR……………………… 76 Figure 36: Heatmap depicting CT values for 48 qPCR assays applied to WTA product………. 78 Figure 37: Standard curves of mean CT values from selected qPCR assays applied to on-chip WTA products…………………………………………………………………………………... 79 Figure 38: Comparison of gene abundances relative to 18S rRNA in on-chip and in-tube WTA product………………………………………………………………………………………….. 80  xii  Acknowledgements I would first like to thank my supervisor Carl Hansen. In addition to his guidance as an exceptional scientist and engineer, his enthusiasm, hyper-positive attitude, and standard of excellence have all pushed me to grow both within and outside of my graduate studies. As an exemplary mentor and teacher, Carl has always patiently answered my countless inquiries or discussed my ideas with me, regardless of how fully developed they were in my own mind. I am also most grateful for the numerous opportunities Carl has provided me, through both scientific and industrial collaborations, to apply my work to a diverse set of problems. In short, I could not have asked for a better supervisor. I have also had the opportunity to interact with a number of outstanding faculty members during my time at UBC.  In particular, I would like to thank Steven Hallam, whose lab I have  collaborated with extensively, for his unwavering support and for introducing me to the world of environmental microbial genomics. His enthusiasm and belief in this project, even in its earliest stages, has been a constant source of motivation. I also thank Bud Homsy, who was most helpful in providing a critical mathematical elucidation of flow-controlled wetting, and Sam Aparicio for his continued interest in applying this work to his study of cancer progression. Over the course of this project, I have had the privilege of working with and befriending the following excellent students, postdocs, and research engineers who all made significant contributions to this work:  Tim Leaver, Hans Zahn, Adam Quiring, Mike Vaninsberghe,  Antoine Pagé, Nat Brown, Jas Khattra, and Kevin Heyries. Special thanks go to Hans whose mechanical wizardry has undoubtedly spared me an intolerable amount of pipetting. In addition to those mentioned above, I would like to express my appreciation for my fellow Hansen lab members whose knowledge and ability have been invaluable resources, and whose kindness and friendship have made spending long days (and nights) in the lab a pleasure. In particular, I thank Anupam Singhal, Adam White, and Jens Huft with whom I have interacted most closely. I also thank Darek Sikorski, Veronique Lecault, Bertin Wong, Oleh Petriv, and Marketa Ricicova for their camaraderie. I thank my committee members Steven Hallam, Karen Cheung, and Andre Marziali for their critical reading of this thesis and for their useful suggestions. xiii  I am grateful to the Natural Sciences and Engineering Research Council of Canada and Genome BC for their scholarship and research funding support. Finally, I thank my Mom and Dad for their love and lifelong support, and for instilling in me an appreciation for science and education. I thank my close friends David Roth and Gianfranco Marino for helping me to get my mind off of chips, droplets, and biochemistry when I needed to. Most of all, I thank Michelle for her love, support, and belief in me. She has been my rock throughout this endeavour.  xiv  Chapter 1 : Introduction Microfluidic systems that control fluid flow at the nanolitre scale provide numerous advantages for biological analysis including automation, enhanced sensitivity and reaction efficiency in small volumes, favourable mass transport properties, and the potential for scalable and costeffective small volume assays (1-3). While the benefits of these advantages have now been widely demonstrated in virtually every branch of microbiology and molecular biology, application innovation and user adoption by the larger biological research community have lagged. With the exception of a handful of commercially available products (4, 5), the use of microfluidic devices has remained tethered to beta testers and engineering-focused academic laboratories (6). This is largely due to the prevailing paradigm in microfluidic research in which devices are “hardwired” for specific fluid handling tasks, necessitating a customized design for each application or change in protocol. Unfortunately, the development of these devices requires specialized fabrication facilities and technical expertise that are largely inaccessible to the typical biological research laboratory, precluding them from performing the iterative cycles of device design, fabrication, and testing that are necessary to produce a working prototype. Indeed, this presents a significant barrier to the adoption of microfluidic technology by those who stand to benefit the most from its use. In analogy to the microelectronic industry, the utility and ubiquity of integrated circuits is the result of programmable designs such as microprocessors.  In such devices, transistors are  organized in architectures that permit the execution of an unlimited number of algorithms based on instructions and data that are input by the user. This has allowed software and firmware developers to apply the complex control of electron flow, enabled by integrated circuits, to the task of computation, without requiring access to or understanding of integrated circuit design and fabrication. This has ultimately led to the widespread adoption of microelectronic technology by entire societies of users. Similarly, the advancement of programmable microfluidic devices that can be reconfigured by the end user to perform a wide variety of fluid-handling protocols stands to dramatically enhance the pervasiveness and impact of microfluidic systems in biological research. The work presented here endeavours to provide such a platform. This thesis describes the design and application of a multipurpose microfluidic device capable of executing user-defined 1  multistep reactions by programmable metering and combination of reagents in an array of nanolitre-volume chambers, followed by automated recovery of individual chamber contents. The device is also capable of sorting single cells into any chamber, enabling a variety of experiments for single-cell analyses, which are becoming increasingly important for the emerging understanding of cell-to-cell heterogeneity in all biological systems. The versatility of the device has been demonstrated by performing a variety of such single-cell experiments. The integration of these capabilities represents the highest degree of microfluidic fluid-handling functionality demonstrated to date and allows non-expert users to apply the intrinsic advantages of microfluidics to a wide variety of biological applications without the need to custom-design a device for each experiment. The development of this device is thus a significant step towards the goal of making microfluidic technology ubiquitous in the wider scientific research community. It is anticipated that the proliferation of such devices will democratize microfluidics research, providing a common hardware platform on which software and “wetware” will be developed and shared among a larger user community. Advances in microfluidic technology over the past decade have resulted in increasingly sophisticated functionality and the emergence of two dominant and orthogonal strategies for the high-throughput assembly and execution of biochemical reactions, based either on the use of integrated microvalves or the transport of microdroplets. The device described in this thesis exploits advantages of both strategies, which are thus reviewed here. 1.1  Valve-based microfluidics  The development of soft lithography (7) and the extension of this method to the fabrication of integrated elastomeric microvalves by multilayer soft lithography (8) has enabled the robust production of monolithic devices composed of poly(dimethylsiloxane) (PDMS) with thousands of active microvalves per cm2 (9). This high level of integration allows for higher order valvebased fluidic control elements such as peristaltic pumps, mixers (10), and multiplexing structures (9) that can route and modulate fluid flow through complex networks of interconnected channels and chambers using a modest number of control inputs. This enables the design of device architectures capable of executing hundreds to thousands of predefined multistep reactions in parallel where nanolitre-scale reagent volumes are lithographically defined by chamber geometries. Valves are used to control the timing of reagent flow through serially connected 2  chambers for multistep reactions where reagent mixing between chambers for consecutive protocol steps is performed by diffusion. The parallelization afforded by such valve-based devices has been used to perform a wide range of applications including protein structure (11) and interaction studies (12, 13), single-cell gene expression analysis (14, 15), single-cell genomics (5, 16-20), and single-molecule nucleic acid amplification (21, 22). 1.2  Droplet-based microfluidics  Another fluid-handling strategy that has developed alongside valve-based technology is the compartmentalization of reagents into aqueous droplets in an immiscible carrier fluid. In contrast with the parallel assembly of large numbers of reactions in valve-based devices, droplet-based microfluidics achieves high-throughput by serially assembling reactions at high speed. In this strategy, each droplet acts as a separate reaction vessel where the interface between the aqueous and carrier fluids functions as a diffusion-proof barrier that prevents cross-contamination and dilution. Highly monodisperse droplets with volumes ranging from picolitres to nanolitres can be generated at rates on the order of kHz by co-flowing aqueous and carrier fluids in T-junction (23) or flow-focusing (24) geometries. The aqueous phase is broken into droplets through the interplay of surface tension forces that act to reduce the interfacial area between the two phases, and viscous stresses that act to extend and drag the aqueous stream. Depending on the scheme used, droplet volumes and generation rates can be tuned by controlling the flow rates of the two phases and the dimensions of the orifice at which droplet breakup occurs. Surfactants must typically be used in order to stabilize droplets and prevent unwanted mixing of individual reaction vessels.  The choice of surfactant can be important as it must prevent droplet  coalescence but may also be required to have properties important to the chosen application such as deterrence of protein adsorption to the droplet interface (25) or compatibility with cell culture (26). For single-step reactions, multiple reagents and analytes can be encapsulated into the same droplet and their concentrations varied through control of the flow rate of each injected reagent (27). For multistep reactions, additional reagents must be introduced into droplets at a defined time after droplet creation. This can be done by merging droplets using microfluidic geometries that position them in close proximity and induce drainage of the carrier fluid that separates them (28-30), or by applying external forces using electric fields (31, 32) or localized heating (33) to 3  destabilize the interface. Electric fields can also be used to rupture the droplet interface in order to directly inject reagents from an aqueous stream (34). After addition of new reagents, droplet contents can be rapidly mixed by a process known as chaotic advection, which repeatedly folds and stretches droplet contents into fluid layers that become exponentially thinner to enable rapid mixing by diffusion (35). In combination with fluorescence-based detection, droplets containing analytes of interest can be sorted by pushing them into a separate stream using electric fields (36), surface acoustic waves (37), or deflectable membranes (38). The theoretically unlimited number of droplets that can be generated from a single device and the high speed at which the above functions can be executed affords droplet-based microfluidic systems unparalleled throughput. This has been exploited to perform high-throughput nucleic acid amplification (4, 39-45) and a variety of high-throughput screening applications including protein crystallization (46), drug screening (47-49), and directed evolution (50, 51). 1.3  Programmable microfluidics  To place the work described in this thesis in context with the state of the art in programmable microfluidic functionality, a review is presented here. There are several examples of existing microfluidic platforms, based on different fluid handling strategies, which can be reconfigured in software to carry out different protocols. Perhaps the most prominent of these is the so-called “digital microfluidic” platform that manipulates droplets on arrays of electrodes coated with a dielectric layer. The application of electric potential across the dielectric can induce localized droplet wetting due to reduced interfacial tension (52), and can generate dielectrophoretic and electrostatic forces (53). These forces can be used to break off small droplets of reagent from a larger “reservoir” droplet, which can then be transported across multiple electrodes by sequentially applying voltage to contiguous electrodes in a desired path. Multiple droplets can be merged, the combined droplet can be mixed by repeatedly moving it in some pattern on the array to induce thinner fluid layers for rapid diffusive mixing, and combined droplets can later be split for further processing (54). Digital microfluidic devices are inherently programmable since each electrode can be randomly addressed for actuation, enabling droplet movement in arbitrary paths. These systems also allow for simultaneous control over the motion of multiple droplets, in theory allowing for a high level of parallelization. The manipulation of droplets on an open array onto which precipitants can sit has been shown to be well suited to applications requiring 4  extraction and rinsing of analytes from complex samples for downstream proteomic analysis (5557). However, despite the inherent programmability and potential for parallelization of these devices, a limited number of simultaneous reactions (<10) have been demonstrated to date, presumably due to control complexity or fabrication constraints. These systems also manipulate droplets that are typically on the order of 1 µL, approximately 100 times larger than volumes handled in channel-based devices, and thus may benefit less from the advantages of performing reactions in small volumes. Progress towards programmability has also been made in valve-based devices by using individually addressable valves to direct fluid flow through arbitrary paths in a two-dimensional array of interconnected nodes. PDMS membrane valves in a hybrid PDMS-glass device (58) were used to programmably route flow through an 8 by 8 square array of interconnected nodes in order to perform an enzymatic assay, mixing, and serial dilution (59). Improving upon this concept, MSL-based valves were used to control flow in all 4 possible directions from each node in a similar square array in order to achieve a higher degree of control over the connectivity between nodes (60).  This device was used to perform an immunoassay and culture and  stimulation of multiple yeast strains.  While these devices demonstrate the usefulness of  addressable valves for programmable flow control through a flexible fluidic architecture, the number of implemented nodes has thus far constrained the number of simultaneous reactions and the resolution of solution formulation, limiting their utility. The sample compartmentalization inherent to two-phase droplet-based flow has previously been exploited in valve-based devices to achieve higher levels of functionality. In an MSL-based device, a rotary mixer was used to mix multiple programmable proportions of aqueous buffers and the resulting solution was injected into a droplet for transport to one of several storage channels (61). This concept was extended and applied as a screening platform for a protein crystallization study in which programmed mixtures of buffers and protein were formulated in droplets and transported into large-cross section channels for incubation and modulation of osmotic strength (62). The combination of programmable mixing and droplet-based flow allow for flexible formulation of solutions in small volumes and transport to different locations on-chip without dilution or cross-contamination.  However, in these systems, once a solution is  encapsulated in a droplet, it is no longer accessible for further reagent addition, making multistep reactions impossible. 5  1.4  Single-cell isolation and analysis  The need for single-cell analysis is becoming increasingly clear as the importance of cell-to-cell heterogeneity at all levels of biology is brought to light. For example, the vast majority of microbial species present in the natural environment have yet to be successfully isolated and cultured in a laboratory setting (63). The isolation and analysis of single microbes directly from environmental samples is thus a direct and very attractive option for the study of individual species (64). Similarly, in mammalian systems, minority cell populations are often responsible for important biological functions or disease-causing aberrations (65), but detection or analysis of these cells by means of bulk measurements on large populations is confounded by the presence of other cell types that in many cases may be present at much higher abundances. Measurements on single cells circumvent these issues by allowing access to fundamentally pure cellular states. The advantages of performing biochemical reactions in microfluidic devices are particularly important in the execution of single-cell assays. While single cells can be isolated by manual pipetting or fluorescence-activated cell sorting (FACS), these methods deposit the cell into a conventional microlitre-scale reaction volume in which the target analyte is present at a very low concentration that can prevent reliable detection or amplification. These macroscale methods of cell isolation also suffer from increased potential for contamination that scales with the sorted volume (66). The detection of rare cells or clonal patterns in large cell populations requires measurements to be made on large numbers of single cells. Unfortunately, the reagent costs necessary to perform microlitre-scale assays on hundreds of single cells can be prohibitive. Microfluidic devices are ideally suited to addressing both of these issues. By confining the cell into a nanolitre-scale volume, its concentration is increased a thousand fold relative to conventional reaction volumes, proportionally reducing the inhibitory effect of competing species. The high-throughput execution of small-volume reactions enabled by microfluidic systems also makes large-scale single-cell studies economically feasible by reducing reagent consumption by orders of magnitude. Single-cell measurements of various types are thus becoming important and highly sought-after microfluidic applications (67, 68). The integration of single-cell handling into a programmable microfluidic system allows for multiple single-cell  6  assays and procedures to be performed using a single device. A review of the state of the art in microfluidic single-cell isolation techniques is thus presented here. In valve-based devices, single cells of interest that are identified by microscopic observation can be directed into chambers using valves and peristaltic pumps for assaying or processing. This strategy has been used to isolate and perform whole genome amplification on single microbes with a specific morphology (17) and chromosomes from single human metaphase cells (19). Optical trapping has also been used to isolate submicron-sized microbes into chambers (20) and to encapsulate single lymphocytes in droplets followed by photolysis and enzymatic assay (69). For applications in which the goal is to isolate and analyze each and every cell in a population, without the need to select individual cells that have a phenotype of interest, higher throughput methods can be used. Partitioning of a cell suspension at limiting dilution allows for the highthroughput isolation of single cells into reaction vessels with occupancies determined by the Poisson distribution. Valves have been used to compartmentalize single bacteria into chambers for the co-localization of multiple genes by PCR (5, 16). Limiting dilution has also been used to encapsulate single cells into droplets to perform single-cell genetic analysis (42, 45), single-cell protein assays (70, 71), single-cell drug toxicology screens (47, 48), and directed evolution (50). While dilution provides a simple method of partitioning a cell population into single-cell containers, Poisson statistics dictates that in order to limit the number of containers having more than one cell, most containers must hold zero cells. This constraint severely limits the number of useable containers for single-cell experiments. For example, only 15.6% of containers will hold one cell if no more than one in ten of the occupied containers can be allowed to hold two or more cells. To overcome this limitation in channel-based devices, hydrodynamic trapping of single cells can be performed using microstructures that initially draw in single cells by means of engineered flow patterns. Then, once a cell “trap” is occupied by a single cell, flow through the trap is reduced and the remaining cells in suspension travel around it to the next unoccupied trap. This strategy has been used to achieve high single-cell occupancy of chambers for single-cell gene expression analysis (15), single-cell enzyme kinetic measurements (72), and observation of single non-adherent mammalian cell division (73, 74). However, such approaches are not easily applied to microbial cells due to their small physical dimensions and constraints on microfabrication. The limitations of Poisson statistics for the encapsulation of single cells in 7  droplets have been overcome by using high flow rates to hydrodynamically induce selforganization of cells into an evenly spaced suspension that allows for single-cell droplet occupancy rates much higher than is stochastically achievable (75), although such approaches have not been widely adopted due to very narrow operating conditions and have yet to be applied in biological applications. 1.5  Statement of research  This thesis describes the achievement of the following specific research goals: 1. The development of a novel microfluidic device capable of: a. Programmable and precise metering of arbitrary volumes of reagents. b. Combination of these reagents with arbitrarily defined timing in an addressable array of nanolitre-volume storage chambers. c. Phenotype-based sorting of single cells into any chamber. d. Automated recovery of reaction products from selected chambers into standard microlitre-volume container formats. 2. Characterization of device fluid-handling capabilities necessary for this functionality. 3. Demonstration of the utility and versatility of the device by application to multiple experiment types for single-cell analysis.  8  Chapter 2 : Device design and implementation For a programmable liquid-handling system to be broadly applicable to biological research or processing, it must allow for metering of user-defined reagent volumes, transport of reagents to defined locations, combination and mixing of reagents, and random accessibility of any reaction for reagent addition or extraction of reaction products.  Indeed, these are the fundamental  capabilities provided by automated liquid-handling systems that are the workhorse of modern high-throughput laboratory automation in microlitre-volumes (76). The key aim of the present work is essentially to miniaturize these capabilities onto a microfluidic format to achieve programmability while preserving the benefits of performing reactions in nanolitre-volumes. In this chapter, the strategies and implementations for achieving this functionality is described. The device combines the reconfigurable flow-routing capabilities of integrated microvalve technology with the sample compartmentalization and dispersion-free transport that is inherent to the manipulation of droplets in a water-immiscible carrier fluid.  A discussion of the  advantages of mass transport in droplets and the factors that must be considered to achieve robust droplet transport in microfluidic channels is first presented. 2.1  Requirements and considerations for droplet-based flow in microfluidic channels  The transport of small volumes of solutions from one device location to another is a basic operation that is critical to achieving higher order programmable functionality, but is a fundamental problem for single-phase flows in microfluidic channels. For such flows, a no-slip boundary condition at the channel walls results in a velocity profile that is greatest in the centre of the channel cross section and is reduced to zero at the channel walls. This uneven velocity profile results in dispersion which, even in the absence of diffusion, spreads out a fluidic slug over the entire distance of traversed channel. Coupled with diffusion, this phenomenon is known as Taylor dispersion, and results in an initially well-defined impulse of solute being spread into a Gaussian profile as it is transported down a channel (77). This dispersion makes it impossible to transport a small volume of liquid from one part of a device to another without dilution or crosscontamination with solution that was previously passed through the same channel. As well, reagents may be adsorbed onto channel walls, resulting in depletion of reaction components and possible clogging of channels. These limitations are obviated by the compartmentalization of aqueous solutions into discrete droplets that are separated by an immiscible carrier fluid. This 9  imparts the important ability to sequester reagents to a small volume by introducing an interface around each droplet, which serves as a diffusion-proof barrier that eliminates dispersion. In order for the aqueous phase to be partitioned into droplets, the carrier fluid must preferentially wet channel walls and must therefore have a lower interfacial tension with the device material than the aqueous phase. As in most droplet-based microfluidic systems, the architecture of the device described here requires droplets containing different reagents to pass through the same channels. In these paths, wetting of the carrier fluid onto the channel walls must be consistent to prevent contact between the aqueous phase and the channel walls. If this condition is not met, flowing droplets may wet the channel walls, causing smaller droplets to break off and be left behind, which may merge with and contaminate subsequent droplets flowed through the same channel. To prevent such undesirable droplet wetting in regions of droplet transport, the aqueous phase must have a lower interfacial tension with the carrier fluid than with the channel walls to ensure that contact between the aqueous phase and carrier fluid is more energetically favourable than contact between the aqueous phase and channel walls (78). This can be accomplished by introducing a surfactant that lowers the tension at the aqueous phase-carrier fluid interface. The integrity of a moving droplet can be characterized by the dimensionless Capillary number, !" =  !" !  (1)  where µ is the viscosity of the carrier fluid [kg m-1 s-1], U is the one-dimensional flow velocity [m s-1], and γ is the interfacial tension [N m-1] between the two liquid phases. This number quantifies the competition between viscous forces that apply shear stresses to a droplet and tend to shear it apart, and the interfacial tension that minimizes its surface area and tends to keep a droplet intact by resisting deformation and shear. In order to prevent the breakup of droplets and the resulting cross-contamination, Ca must be kept low (79). Equation 1 states that this can be achieved by decreasing viscosity and velocity and increasing interfacial tension. If flow velocity is to be increased for faster droplet transport, the carrier fluid viscosity must be proportionally decreased to maintain the same value of Ca. Carrier fluids with the lowest viscosities should thus be used whenever possible. For a given carrier fluid and flow velocity, however, Equation 1 implies that for robust droplet transport, a balance must be struck between the low interfacial  10  tension between the aqueous and carrier phases required for consistent carrier fluid wetting onto channel walls, and the high interfacial tension required for droplet integrity. Hydrocarbon and silicone oils have very low interfacial tension with PDMS and thus have been used as carrier fluid in earlier droplet-based PDMS devices (23, 24, 61). However, these liquids are absorbed into PDMS, which can cause swelling of the bulk device material, resulting in unwanted alteration of device geometries (80) or even eventual closing of channels. Fluorocarbon liquids, which have similar molecular structure to hydrocarbons but substitute hydrogen atoms with fluorine atoms, are an optimal carrier fluid for a variety of reasons. Unlike hydrocarbons, they do not swell PDMS (80, 81) and their molecular properties also confer characteristics that make them ideal carrier fluids for droplet-based biochemical reagent transport (82).  The weak intermolecular or van der Waals forces between fluorocarbon  molecules render them highly hydrophobic and lipophobic, making them repel both water and lipids and giving them very low solubility for aqueous and biological reagents. This property makes fluorocarbon liquids excellent barriers to diffusion from droplets containing biochemical reactions. This weak intermolecular cohesion also creates gaps that can harbour gas molecules, giving fluorocarbons the highest gas solubility of all liquids, which is necessary for the viability of cells encapsulated in droplets (70, 83). Finally, the strength of the covalent fluorine-carbon bond, the strongest bond found in organic chemistry, results in a high degree of chemical and biological inertness. No enzymes are known to digest fluorocarbons and no organism is known to feed on them. 2.2  General device architecture and operation  The designed microfluidic device performs several functions that combine to achieve unprecedented microfluidic fluid-handling functionality. It is composed of a reagent-metering module capable of dispensing droplets with arbitrary programmed volumes of 9 reagents, a cellsorting module capable of microscopy-aided sorting of single cells into droplets, a twodimensional addressable array of storage chambers designed to merge an arbitrary number of reagent droplets, and an integrated nozzle that allows for automated recovery of storage chamber contents without cross-contamination. A schematic of the device is shown in Figure 1. It is composed of a reagent-metering module capable of dispensing droplets with arbitrary  11  Figure 1: Programmable microfluidic reaction array. (A) Device schematic showing the structure of an elution nozzle designed to interface with standard microfuge tubes during chamber elution. (B) Addressable array of 95 storage chambers organized in 19 rows and 5 columns. Control layer are shown in red. Actuation of row multiplexer and column valves creates a unique fluidic path (green arrow) flowing from high to lowpressure ports. (C) Cell-sorting module. (1) Peristaltic pumping is used to advance a single cell suspension through the vertical sorting channel. (2) Pumping is used to encapsulate the cell into a droplet for transport to the chamber array. (D) Reagentmetering module.  Pumping and valve actuation are used to precisely meter  programmable volumes of reagents into droplets. programmed volumes of 9 reagents, a cell-sorting module capable of microscopy-aided sorting of single cells The device can be divided into two sections based on the fluid being manipulated in each one: An “aqueous section” (shown in blue in Figure 1) containing continuous single-phase flows of aqueous reagents or cell suspensions, and an “oil section” (shown in grey in Figure 1) in which 12  the aqueous phase is dispersed into droplets in the carrier fluid. Prior to use, the oil section is primed with the carrier fluid by pressurizing and injecting it into the device while closing off all exit paths by actuating valves. In this process, referred to herein as “dead-end filling”, the gas permeability of PDMS is exploited and all air in the device is displaced by the pressurized liquid (11). Reagent channels are similarly primed with reagents by dead-end filling against valves. User-defined volumes of aqueous reagents or single cells are encapsulated into droplets in carrier fluid for dispersion-free delivery to any storage chamber in the array, which comprises 95 chambers arranged in 19 rows and 5 columns. A 20th row is reserved for a bypass channel that allows for flushing of reagents directly to the outlets without passing through the chamber array. To address a specific storage chamber in the two-dimensional array, a multiplexer is used to select the desired row (62) and a series of valves that control access to each column of chambers are used to select the desired column. Each chamber of the array can thus be individually accessed by a unique valve actuation pattern of row multiplexer and column access valves, which creates a fluidic path that passes from a high-pressure carrier fluid input port, past the reagent-metering and cell-sorting modules, to the selected storage chamber, and out to one of two low-pressure outlet ports (waste or elution) (Figure 1B). Each droplet is transported along this path in a pressure-driven stream of carrier fluid and is deposited in the storage chamber, which is designed to merge all incoming droplets with any previously stored droplets to formulate reagent mixtures. At any time, the stored droplet in any storage chamber can be accessed in this manner for addition of reagents or recovery from the device through an integrated elution nozzle designed to dispense directly into standard microliter-volume tube formats (Figure 1A). Both formulation and elution of the storage chamber array are fully automated by custom software that controls all on-chip valving, allowing the user to specify arbitrary formulation recipes and storage chamber elution sequences by inputted text files. The encapsulation of reagents in droplets not only allows for dispersion-free transport but also permits high-density storage of large numbers of individually addressable solutions. While valves can be used to enclose solutions and prevent diffusion in a single-phase, individual addressing in such a system requires one independently controllable valve per solution and thus quickly becomes impractical for large numbers. The dense two-dimensional solution storage and row/column addressing strategy used in this device are made possible by the presence of the carrier fluid. In the absence of this immiscible phase, each stored solution would diffuse out of 13  its chamber every time another chamber in the same column was addressed, since the valves that enclose chambers in the same column are connected to a single control channel. The carrier fluid prevents such diffusion even when the valves that enclose each chamber are opened, thus allowing for the individual addressing of large numbers of solutions using a relatively simple valving scheme. The ability to address any storage chamber for arbitrary addition of reagents or recovery of chamber contents is analogous to the write/read functionality of microelectronic random access memory and allows for a high degree of flexibility when running experiments that is rare in reported microfluidic systems. This allows, for example, the addition of a new reagent to a selected stored solution in response to some exhibited behaviour observed in realtime during an experiment. The array format also permits straightforward tracking and arbitrary monitoring of reactions by simple row and column spatial indexing. In previously reported droplet-based systems where droplets are processed in series, complex droplet indexing (84) or optical encoding (48, 49) schemes must be used for tracking, and optical interrogation, performed when a droplet passes by a detector, is restricted to a small number of predetermined time points. The addressable chamber array format is highly scalable as the row multiplexer used can address a number of rows that increases with the factorial of the number of control channels (85), providing an improvement over previously implemented architectures that achieve exponential scaling of addressable rows with the number of control channels (9).  With this factorial  multiplexer, 6 control channels address a modest 20 rows but doubling the number of control channels to 12 allows 924 rows to be addressed. The device has been designed such that the fluidic path that brings a droplet to an addressed chamber enters the array from one corner and exits the array at the diagonally opposite corner. This ensures that the paths passing through each storage chamber have equal fluidic resistance. The design and operation of each device function is described in further detail in the following sections. Details of device fabrication operation are given in Appendix A.  14  2.3 2.3.1  Droplet formation and transport Reagent metering  Programmable reagent dispensing, using a three-valve peristaltic pump, is used to deliver arbitrary volumes of reagents in discrete increments from eight separate reagent inlets by varying the number of pump cycles (Figure 1D). Each pump cycle advances a unit volume of fluid, herein referred to as a “pump increment”, that is determined by the volume displaced by the middle valve of the pump (86). In the work presented in this thesis, devices have been fabricated with pump increments of ~133 pL or 150 pL. Reagent droplets are dispensed directly into a flowing pressure-driven stream of the carrier fluid, where they break off through the combined effect of surface tension, shear stress, and valve actuation. All reagent inlet channels have been designed to have the same length to prevent differences in fluidic resistance from affecting the metering precision of different reagents. Time course images acquired from a video of the droplet dispensing process are shown in Figure 2. This valve and pump-based droplet formation has been shown to achieve complete control over droplet frequency, spacing, and volume, independent of fluid properties (62), simplifying the use of different reagents. In contrast, earlier valve-less strategies for droplet formation show a complex dependence of droplet formation on viscosity and Capillary number (78, 87). The cross-section of the channel into which droplets are dispensed has been designed to have a low aspect ratio and a sufficiently small area such that the minimum volume that can be metered (one pump increment) forms a droplet that occupies most of the channel’s cross-sectional area. The droplet thus has an axial length longer than its cross-sectional diameter and is separated from the channel walls by a thin film of carrier fluid while in transit. The “pancaked” droplet is thus also in an energetically unfavourable state as its surface area is not minimized, and this is exploited for the droplet storage strategy as will be discussed later. 2.3.2  Single-cell sorting  For single-cell applications, the phenotypic selection and isolation of single cells is achieved using a cell-sorting module (Figure 1C), which is similar to previous implementations (18, 19). A cell suspension is advanced at a programmable flow rate by peristaltic pumping while the channel intersection is monitored by microscopy in real-time.  When a cell of interest is  identified based on any visually discernible feature such as morphology or fluorescent reporter, it 15  Figure 2: Reagent metering. Time course images of a single pump increment of red food dye dispensed into a flowing carrier fluid stream for transport to the storage chamber array. A four-step peristaltic pump cycle advances the aqueous stream and valve actuation pinches off a droplet. Scale bar is 500 µm 16  is isolated by actuation of valves, and the aqueous volume containing the single cell is pumped into a droplet for delivery to the storage chamber array. Unlike stochastic or hydrodynamic trapbased single-cell isolation methods, this sorting method allows for the specific phenotype-based selection of single cells out of complex heterogeneous samples that may contain multiple cell types of different sizes or cellular or non-biological debris. In combination with addressable storage, this method allows for any user-defined multistep reaction protocol to be applied to single cells and also allows multiple single selected cells to be placed in the same nanolitrevolume. 2.4  Droplet transport  A fluorocarbon oil (FC-40, 3M) was initially used as the carrier fluid but droplets of water and food dye were found to intermittently wet the channel walls during transport, causing droplet breakup and cross-contamination. transport were considered.  Multiple options for eliminating droplet wetting during  The treatment of channel walls with coatings that enhance  fluorocarbon wetting has been reported in the literature. hydrophobic  and  fluorophilic  by  flowing  Channels have been rendered  (tridecafluoro-1,1,2,2,-tetrahydrooctyl)-1-  trichlorosilane vapor (25), a commercial hydrophobic surface coating agent (Aquapel, PPG) (88), or Teflon amorphous fluoropolymer (89) through devices. However, the application of all of these coatings requires subsequent flow-through of another fluid to remove excess coating material and control the thickness of the coating (89). This was demonstrated in relatively simple device geometries essentially consisting of a single flow path with constant cross-section, but it would be much more difficult to ensure a uniform coating thickness or prevent the coating material from clogging channels and accumulating in storage chambers in the much more complex channel and chamber network of the present device. The use of surfactants is in principle undesirable in the present device as it reduces interfacial tension, making droplets more prone to breakup, and also tends to prevent droplet coalescence which is required for the proposed solution formulation strategy. However, it was determined that the use of a surfactant was the simplest and most robust method of preventing unwanted droplet wetting to channel walls. Surfactants prevent such wetting by reducing the interfacial tension between the aqueous and carrier phases relative to that between the aqueous phase and the channel walls, making the former interface energetically favourable (78).  The 17  fluorosurfactant 1H,1H,2H,2H-perfluoro-1-octanol (PFO) is composed of a fluorocarbon chain connected to a hydrophilic hydroxyl group and thus adsorbs to fluorous/aqueous interfaces, thereby decreasing interfacial tension. This fluorosurfactant has been shown to reduce the interfacial tension between a fluorinated carrier phase and an aqueous phase from 55 mN/m to 14 mN/m (78), well below the aqueous/PDMS interfacial tension of 38 mN/m (90), and has thus been used in the carrier fluids of numerous droplet-based applications to prevent droplet wetting and breakup (27, 51, 91-93). In the present device, a carrier fluid composed of a 5:1 (v/v) mixture of fluorocarbon oil FC-40 and PFO was observed to eliminate droplet wetting during transport to the storage chamber array. Using the viscosity of FC-40 (0.034 kg m-1 s-1), and the maximum flow velocities used in the device (0.01 m/s), Ca is found to be ~0.02, comparable to the value of Ca (~0.01) shown to prevent droplet breakup and cross-contamination in other studies (79). Droplet wetting at low velocities, however, was not prevented by the fluorosurfactant, and this was exploited for droplet immobilization at storage chambers as will be discussed below. As well, although the inclusion of the fluorosurfactant did increase the time required for coalescence between droplets in contact, it did not prevent coalescence entirely (26, 92), thus allowing for formulation of solutions by coalescence of different reagent droplets. 2.5  Droplet merging and storage by “flow-controlled wetting”  The programmable functionality of this device, which enables the user-defined formulation of solutions composed of reagent droplets dispensed from the reagent-metering module, requires a method of both immobilizing and merging of a sequence of droplets, arbitrary in number and timing, at each addressable storage location. The integration of these two tasks is non-trivial. Immobilization of droplets has been previously accomplished by exploiting surface tension (92, 94) and by hydrodynamic trapping (95, 96). Merging has been demonstrated by a variety of mechanisms as previously discussed (28-33). However, these methods require droplets to be constantly in motion and only merge a small (2 or 3) and predetermined number of droplets. In one report, both droplet trapping and merging was demonstrated by pushing droplets out of a flowing stream into chambers by dielectrophoresis (97), but the merging of only two droplets was shown. In this device, the properties of two-phase hydrodynamic flow in confining channels are exploited to implement a simple and robust method that prevents droplet wetting during 18  transport, which can result in droplet cross-contamination as discussed above, but induces droplet wetting of channel walls at precisely defined storage locations in order to immobilize and merge an arbitrary sequence of droplets. 2.5.1  Velocity-dependent droplet wetting  A droplet flowing down a channel filled with an immiscible fluid is separated from the channel walls by a thin lubricating film (98), the thickness of which is related to droplet velocity by the following equation: 3!" ! = 0.643! !  ! !  (2)  where b is the film thickness, r is half the height of the droplet, µ is the viscosity of the carrier phase, U is the droplet velocity, and γ is the interfacial tension between the carrier fluid and aqueous droplet. If the film thickness is reduced to a critical value, an instability arises in which intermolecular forces between the droplet and the channel surface cause the film to spontaneously rupture (99), causing the droplet to wet the surface. This critical thickness is given by the following equation adapted from equation 30 in reference (99): !!! ℎ! = !!"# !  ! !  (3)  where h0 is the critical film thickness, A is the Hamaker constant describing the intermolecular interaction between two bodies within close (on the order of nm) contact, R is the radius of the approximated disc of carrier fluid separating the droplet from the channel wall, and ξmax is a numerical constant associated with the most unstable mode of a perturbation to the uniform film. Selective wetting may therefore be achieved, without modification of surface properties (30), by engineering the device geometry such that droplet velocity remains above this critical value until arrival at the storage area. For a carrier fluid composed of a 5:1 (v/v) mixture of FC-40 and PFO, γ is 14 mN/m (obtained from data in Figure 4 of reference (25)). Carrier fluid viscosity is taken to be that of FC-40, 19  which is 0.034 kg m-1 s-1. The height and width of the channels through which droplets are transported are 10 µm and 100 µm respectively, so r is 5 µm and the radius of the disc of carrier fluid separating the droplet from the wall can be approximated as 50 µm. Hamaker constant A between PDMS and water is 10-26 J (100) and the smallest value of ξmax (~1.7) is used as it leads to the greatest instability of the film (99). Substituting these values into Equation 3, the critical film thickness for droplet wetting h0 is found to be ~8 nm. This value can then be substituted for film thickness b in Equation 2 to find the droplet velocity U at which the film thickness is equal to h0 and spontaneous wetting of the droplet to channel surfaces occurs. This velocity is herein referred to as the “droplet wetting velocity” and is found to be ~170 µm/s. 2.5.2  Storage element design and operation  In order to immobilize droplets at desired locations, microfluidic geometries were designed to decelerate incoming droplets to the droplet wetting velocity by diverting carrier fluid flow through bypass channels (20). A schematic of the droplet storage geometry is shown in Figure 3. Each storage element consists of a large cross-section cylindrical storage chamber that is fed by an inlet channel connected to a series of small side channels. The side channels connect the inlet channel to bypass channels that flow around each side of the storage chamber. When an incoming droplet first arrives at the inlet channel of an addressed storage element, a thin film of the carrier fluid prevents wetting of the channel walls (Figure 3, step 1). When the droplet reaches the side channels, the fluidic resistance to carrier fluid flow through the side and bypass channels is less than that of the channel leading into the storage chamber, whose cross-section is occupied by the droplet. Carrier fluid flow is thus diverted through the side channels, causing the droplets to slow (Figure 3, step 2). The droplets themselves, however, do not pass through the side channels, as the deformation required to do so would result in an increase in surface area that is energetically unfavourable. A storage element design with an inlet channel having 18 side channels (30 µm long x 10 µm wide x 5 µm high) along each side, decelerates droplets to the droplet wetting velocity in the inlet channel for incoming mean flow velocities as high as 3.9 mm/s measured at the storage element inlet. This flow velocity is herein referred to as the “critical incoming velocity”. Flow velocities were measured by observing the advancement of the carrier fluid meniscus in a length of tubing with known cross-sectional area connected to the device outlet while sending droplets 20  Figure 3: Storage element geometry for droplet immobilization and coalescence by flow-controlled wetting.  (1) During transport to an addressed storage element, a  lubricating thin film of oil prevents wetting of channel walls. (2) Side channels create a bypass for the oil (green arrows), reducing droplet velocity. (3i) Below the critical flow velocity, wetting occurs and the droplet is positioned at the cylindrical chamber entrance. (3ii) Above the critical flow velocity, the droplet does not wet at the entrance but travels into the chamber and docks at the chamber ceiling. to an addressed chamber. From analysis of videos of droplet deceleration in the storage element, the lowest velocity achievable before droplets wet to channel walls is estimated to be 100 ± 50 µm/s, comparable to the calculated droplet wetting velocity of 170 µm/s. When a droplet is sent to a storage element with a mean flow velocity equal to the critical incoming velocity, it wets the inlet channel immediately upstream of the storage chamber. Relative to a droplet that is surrounded by a film of carrier fluid, a droplet that is wet to the channel wall is effectively immobilized. However, the droplet in its wetted state is still able to move slowly along the channel surface under the influence of carrier fluid flow due to poorly understood mechanisms that allow for apparent slipping at the liquid-solid interface, resulting in a moving three-phase contact line (101). As the leading edge of a droplet enters the chamber, it is pulled in by surface tension due to the volumetric expansion provided by the chamber, which allows the droplet, whose cross-section is constrained in the channel, to reduce its surface area by adopting a more spherical shape (92, 94). 21  Once in the chamber, the droplet wets the chamber’s sidewall, precisely positioning it immediately above the entrance (Figure 3, step 3i). Finite element simulation of fluid flow through the storage element geometry was performed using COMSOL v4.0a (COMSOL) using a constant flow rate that results in a mean flow velocity of 3.9 mm/s at the inlet (the critical incoming velocity), and setting the fluid viscosity to that of FC-40. A heat plot of the flow velocity magnitude at a height of 2.5 µm (half of height of side channels) and on the vertical plane through the center of the storage element is shown in Figure 4 and indicates that the location of droplet immobilization is sequestered from high carrier fluid shear flows. Contact line pinning forces are sufficient to resist shear forces and the droplet remains immobilized indefinitely for tested mean flow velocities of up to 50 mm/s through the storage element inlet. It should be noted that surface tension forces between the droplet and the carrier fluid do not contribute to retention of the droplet at the chamber entrance, since advancement of the droplet further into the chamber would not increase the droplet’s interfacial area. Each subsequent droplet sent to the storage element repeats this process, is immobilized in approximately the same position, and held in contact with the stored droplet indefinitely, thereby ensuring sufficient time for coalescence even when partially stabilizing surfactants are used. Time course images acquired from a video of droplets of water sent to a chamber at a mean flow velocity of 3.9 mm/s through the storage element inlet, equal to the critical incoming velocity, are shown in Figure 5. When a droplet is sent to a storage element with a flow velocity below the critical incoming velocity, the droplet is decelerated to the droplet wetting velocity further upstream of the storage chamber. In this case, other droplets must first coalesce with it in order  Figure 4: Finite element simulation of the flow velocity through storage element at a height of 2.5 µm (half of height of side channels) (left) and on the vertical plane through the center of the storage element (right). 22  Figure 5: Time course images of droplet merging and immobilization at the critical incoming velocity. Droplets are decelerated to the droplet wetting velocity immediately before reaching the storage chamber and are pulled into it by surface tension. Scale bar is 100 µm. 23  for the merged droplet to reach the edge of the storage chamber, at which point it is pulled in by surface tension. Still images acquired from a video of droplets of water sent to a chamber at a mean flow velocity of 2.9 mm/s through the storage element inlet (less than the critical incoming velocity) are shown in Figure 6. This method allows for robust droplet merging regardless of the time between arrivals of multiple droplets at the storage element. Droplets have been merged even after waiting several days between sending droplets to a storage element. Stored droplets also remain at the storage chamber entrance after prolonged heating of the device, permitting additional reagent droplets to be merged with the stored solution after extended heating steps required for many molecular biology protocols. The volume of the storage chamber defines only an upper limit on the volume of the stored droplet, but the storage element design allows for the formulation and storage of a solution with any volume less than or equal to this limit, allowing for programmable control over the final solution volume. Once the stored droplet reaches this limit, further droplet additions result in the ejection of droplets into the carrier fluid stream as it exits the chamber. Due to the surface tension of the stored droplet, its maximum volume is ~60% of the volume of the cylindrical storage chamber.  In the work presented this thesis, devices have been fabricated with a  maximum stored droplet volume of ~40 nL and a pump increment of ~133 pL allowing for a formulation resolution of 1 in 300 (0.33%), or a maximum storded droplet volume of ~60 nL and pump increment of ~150 pL allowing for a formulation resolution of 1 in 400 (0.25%). If droplets enter the storage element with a mean flow velocity above the critical incoming velocity, they are not sufficiently decelerated by the side channels and enter the chamber without wetting the channel walls. In this case, the free droplets follow an upward trajectory determined by a combination of laminar flow and buoyancy, coming to rest at the chamber ceiling where they wet and are immobilized (Figure 3, step 3ii). Regardless of the velocity at which droplets enter the storage element, they are sent to the same location. In the absence of a surfactant in the aqueous phase, the droplets coalesce shortly after making contact, thereby allowing for reliable droplet merging even at mean flow velocities above the critical incoming velocity. Time course images acquired from a video of droplets of water sent to a chamber at a mean flow velocity of 7.2 mm/s through the storage element inlet (greater than the critical incoming velocity) are 24  Figure 6: Time course images of droplet merging and immobilization below the critical incoming velocity. Droplets are decelerated to the droplet wetting velocity upstream of the storage chamber. Multiple droplets merge before the combined droplet reaches the chamber and is pulled into it by surface tension. Scale bar is 100 µm. 25  shown in Figure 7. In the last image, the focus has been shifted to the top of the chamber to show that the droplet is positioned at the chamber roof. However, in addition to surfactants in the carrier phase, it is often desirable to include a surfactant in the aqueous phase to reduce the adsorption of analytes to channel walls in the aqueous section of the device, or to the droplet interface (25). The inclusion of such surfactants (0.1% Tween 20) has been observed to partially stabilize droplets, significantly increasing the time required for coalescence. Thus, when the droplet contents include a surfactant, a droplet sent into the chamber above the critical incoming velocity has only transient contact with a previously stored droplet, “bouncing” off of it and coming to rest at a different location on the chamber ceiling. Thus, when using aqueous surfactants, the robust merging of each droplet sent to a storage element requires that they be held in contact for an extended time, which can only be ensured by operating at or below the critical incoming velocity to ensure droplet immobilization by wetting. Operating the device using these flow velocities, we routinely observe 100% coalescence in 1000 events (20 droplets x 50 chambers) both with and without surfactant in the aqueous phase (0.1% Tween 20). Stored droplets with and without surfactant are shown in Figure 8. The aqueous surfactant appears to enhance wetting of droplets onto PDMS surfaces as can be seen by the reduced contact angle in Figure 8B. Once the total volume of droplets sent to a storage chamber is sufficient to occupy enough of the storage chamber volume (~25%), all subsequently sent droplets will merge with previously stored droplets simply due to space constraints in the chamber. Thus, if the final stored droplet volume is sufficiently large and the sequence of individual droplet merging is unimportant, flow velocities much higher than the critical incoming droplet velocity can be used to achieve faster formulation.  Operating in this regime, a storage chamber can be filled with 100 pump  increments in approximately 5 seconds. While the encapsulation of reagents into droplets eliminates unwanted diffusion, it also facilitates improved mixing within the droplet. The spherical shape of stored droplets has a diameter smaller than the distances over which reagents must diffuse to achieve complete mixing in typical single-phase microfluidic systems that use a series of interconnected chambers to perform multistep reactions (15, 18). The time required for complete mixing of a stored droplet by diffusion alone is thus significantly shorter, as the diffusion time of an analyte has a quadratic dependence on the diffusion distance. In addition, the shear stress imparted by the flow of 26  Figure 7: Time course images of droplet merging and immobilization above the critical incoming velocity. Droplets are not decelerated to the droplet wetting velocity and freely flow to the storage chamber ceiling where they merge. Focus has been shifted up vertically in the last image to show the merged droplet positioned at the ceiling. Scale bar is 100 µm. 27  Figure 8: Micrographs of 2.7 nL stored droplets of (A) water and (B) 0.1% Tween 20 surfactant in water. The surfactant enhances wetting onto the device surface, resulting in a reduced contact angle relative to water alone. carrier fluid against the stored droplet while transporting droplets to a storage element results in recirculating flows that advectively mix droplet contents (102), further decreasing mixing times. These factors also allow for small reagent volumes to be added and rapidly mixed with a much larger stored droplet, which is problematic in typical single-phase systems, as the reagents from a small chamber can take a long time to diffusively mix completely into an adjacent chamber with much larger dimensions. 2.6  Sample recovery from selected storage chambers  In many microfluidic applications, nanolitre-volume reaction products must be extracted out of the device for further “off-chip” analysis or processing. For example, in genomics applications, it is often necessary to sequence nucleic acids amplified on-chip. While methods have been devised to distribute microliter-volumes of reagents into multiple individual microfluidic reaction chambers in order to solve the “world-to-chip” problem (103), the corresponding “chipto-world” problem of recovering microfluidic reaction products into a format that can be handled by conventional means has received less attention. In single-phase systems, this has been accomplished by simply flowing reaction products from each individual chamber to a separate access port into which a pipette tip can be plugged for sample collection (18, 19). While this simple method allows for elution of multiple chambers in parallel, it is not scalable, as the on-chip space required for ports, the routing of channels that feed them, and the manual handling of pipette tips becomes impractical for more than a few tens of reactions. 28  In previously reported droplet-based systems, due to the difficulty in handling the small volumes of microfluidic droplets by conventional means, droplets recovered from a device are typically pooled together and the emulsion is then broken (4, 50) but this obviously leads to mixing of individual  reaction  products  and  negates  the  advantages  of  on-chip  droplet  compartmentalization, making off-chip analysis of individual reaction products impossible. Solid phase support such as beads can be used to distinguish and sort reaction products by the different fluorophores associated with them (45), but such optical encoding strategies become impractically complex for larger numbers of reactions. The selective merging of individual sorted droplets with a continuous aqueous stream for further on-chip processing has been demonstrated (104), but this has not been extended to off-chip recovery. In the present device, the addressability of the storage chamber array and the compartmentalization of two-phase flow are exploited to achieve arbitrary elution of individual stored droplets directly into standard microliter-volume tubes in a fully automated manner and with negligible cross-contamination between storage chambers.  Elution is performed by  flushing an addressed storage chamber with a continuous carrier fluid-sheathed stream of aqueous buffer. This stream, formed by applying equal pressures to the aqueous and carrier phases that join at a T-junction at the reagent-metering module (105), coalesces with the stored droplet until it exceeds the chamber capacity. At this point, a carrier fluid-sheathed aqueous stream containing the stored droplet’s contents is ejected from the storage chamber and directed to the elution channel. Just as the carrier fluid prevents cross-contamination of reagent droplets during formulation by preferentially wetting channel walls, it also surrounds the eluted stream, preventing cross-contamination with the next eluted chamber. Elution of a stored droplet of blue food dye is shown in Figure 9. The separation of the eluted aqueous stream from the channel walls of the storage element outlet can be clearly seen. The total aqueous volume used for elution can be controlled by programming the time for which the aqueous phase is allowed to flow into the T-junction. For a perfectly mixed chamber, each chamber volume that is flushed through the chamber expels half of its initial contents and lowers concentration two-fold. Thus, the concentration of the chamber contents during elution can be modeled by the following equation:  29  ! = !!  1 2  !  (4)  where C is the concentration at any time during elution, C0 is the concentration prior to elution, and x is the number of chamber volumes flushed through the chamber. According to this model, ~9.966 times the chamber volume must be flushed through the chamber to reduce the concentration of chamber contents to 0.1% of its initial value. However, during elution of the current device, the incoming buffer is not well-mixed with chamber contents. This can be seen in Figure 9 where the dye concentration is clearly variable in different chamber locations. This lack of complete mixing during elution results in the above model underestimating the volume required for elution.  In light of this, elution is typically performed using ~5 µL of buffer, which  is equivalent to 125 times the maximum stored droplet volume. This excessive elution volume effectively ensures complete flushing of the sample from the chamber. This elution method comes at the expense of sample dilution. However, this dilution is unavoidable by any method due to practical limitations on volumes that can be handled off-chip (> 1 µL). Eluted reaction products are infact typically diluted into at least 10 µL in order to allow for redistribution into downstream reactions by pipetting. The elution channel has no dead volume and its exit is located at the tip of an elution nozzle built  Figure 9: Stored droplet elution. Micrograph showing an oil-sheathed stream of water flowing into a storage element to perform elution. Separation of the aqueous phase (containing blue dye) from the channel walls is visible at the storage element outlet. 30  into the device (Figure 1A), which is designed to fit into standard microfuge tubes or microwell plates. During elution, the chip is mounted via a vacuum chuck to a custom 3-axis robot built from three interconnected precision stages (Figure 10), which allows for automated control of the exact position of the elution nozzle. Custom software is used to calculate the position of each well in any two-dimensional grid based on the position of three corners of the grid that are defined by the user, and enables automated insertion of the elution nozzle into selected wells during elution. Each well is prefilled with light mineral oil, and the tip of the elution nozzle is completely immersed before elution begins. As light mineral oil has a lower interfacial tension with PDMS than both the aqueous and fluorocarbon phases, it coats the opening of the elution channel and the outside surfaces of the elution arm, preventing any of the eluted aqueous phase from adhering to the exterior nozzle surfaces and contaminating the next well that the nozzle is lowered into. As light mineral oil also has a lower density than water, aqueous droplets expelled from the elution channel sink to the bottom of the well to minimize the chance of aqueous adhesion onto the nozzle. Elution with a volume of ~5 µL is sufficient to ensure that any aqueous droplets that may remain on the nozzle exterior at the end of the elution process contain  Figure 10:  Photograph of 3-axis robotic chip-holder, which allows for computer-  controlled positioning of the elution nozzle into microfuge tubes for automated elution. 31  a negligible amount of stored droplet contents. After elution of an addressed storage chamber is complete, channels of the array that are in the elution path of other storage chambers are similarly flushed with an aqueous stream to ensure that any possible contaminant droplets are expelled. As a final precaution, after elution of each storage element, the nozzle is rinsed in an isopropanol bath to wash away any aqueous droplets that may remain attached to the nozzle’s exterior and can lead to sample carry-over. Isopropanol is chosen because it dissolves light mineral oil, thus allowing aqueous droplets on the nozzle exterior, which may be encased in light mineral oil, to be washed away. A photograph of tubes containing eluted samples from a device is shown in Figure 11. Blue food dye was used as the elution buffer for easy visualization of the eluted volume. By visual inspection, eluted volumes are estimated to vary by no more than ~10%. After elution, if desired, additional aqueous buffer can be added to the tubes to obtain a larger volume for handling by pipette, and the tubes centrifuged to ensure coalescence of all aqueous components.  Figure 11: Photograph of microfuge tubes containing blue food dye eluted from a device. 32  Chapter 3 : Characterization of fluid-handling capabilities In this chapter, the fluid-handling capabilities of the device necessary for programmable execution of various molecular biology protocols are characterized. Fluorescent measurements and quantitative PCR have been used to gauge formulation accuracy, extent of protein adsorption onto droplet interfaces, cross-contamination between storage chambers during loading and elution of reagents, and elution efficiency. 3.1  Formulation accuracy and precision  To determine the metering precision and formulation accuracy of the device, a series of stored droplets were formulated, having 10 different concentrations of Quasar 670 fluorescent dye ranging from 100 nM to 1 µM. Three replicate stored droplets of each concentration were formulated by metering 200 pump increments, each ~133 pL in volume, of 1 µM dye or a diluting buffer (Tris EDTA) for a total volume of 26.6 nL. The resulting dye concentrations were determined by acquiring fluorescent confocal images of the droplet array and using Image J software to find the mean fluorescent intensity of each droplet, which was in excellent agreement with target values over the full range (R2=0.999) as shown in Figure 12A. Formulation was also highly reproducible, as the mean coefficient of variation for all formulated concentrations was 1.4%.  These results indicate that the device is capable of highly accurate and precise  formulation. In another demonstration of addressable formulation, we applied the device as a programmable display by formulating a two-fold dilution series of three colors of food dye in water to write three letters on the storage chamber array (Figure 12B). Each stored droplet was composed of 300 pump increments (39.9 nL) of either dye or water. 3.2  Prevention of protein adsorption onto droplet interfaces  Proteins are known to adsorb onto liquid interfaces and consequently denature (106), which can result in the removal of enzymes that perform important functions or targeted proteins from a reaction volume. In microfluidic devices, the ratio of surface area to volume is much higher than in conventional microlitre-volume reactions, making such surface effects commensurately more important. PDMS in particular is prone to adsorption of proteins and hydrophobic molecules (107) and thus PDMS microfluidic channel surfaces must be treated or surfactants must be used to prevent such adsorption. Commercially available surfactants derived from poly(ethylene 33  Figure 12: Addressable and accurate formulation. (A) Mean fluorescent intensity and standard deviation of fluorescent measurements of formulated 26.6 nL droplets, composed of 200 pump increments of 1 µM dye or diluting buffer.  Inset shows  corresponding fluorescent confocal image of the array of stored droplets. Scale bar is 1 mm. (B) Microfluidic display showing addressable and programmable formulation. Stored droplets are composed of 300 pump increments arranged in letters with a twofold dilution series of dye from top to bottom of each letter. Scale bar is 2 mm. oxide) such as Pluronic and Tween have been shown to prevent protein adsorption onto PDMS surfaces (108), and are thus added to solutions containing enzymes or proteins prior to use in PDMS microfluidic devices. Tween 20 is added to aqueous reagents to prevent such adsorption to channel walls in the aqueous section of the present device but, since reactions are ultimately assembled and performed in droplets, it is also necessary to prevent adsorption to droplet interfaces.  In one study, fluorosurfactants capped with different functional groups were  evaluated for their capacity to deter non-specific protein adsorption to microfluidic droplet 34  interfaces, and an oligoethylene glycol (OEG) functional group presented to the aqueous phase by a fluorosurfactant was found to resist protein adsorption most effectively (25). However, the inclusion of surfactants in the aqueous phase was not tested. To determine the extent to which protein adsorption to droplet interfaces occurs in the present device, stored droplets of fluorescein isothiocyanate (FITC)-labeled bovine serum albumin (BSA) and Alexa 488-labeled fibrinogen in phosphate buffered saline (PBS) were fluorescently imaged using different surfactants added to the fluorous and aqueous phases.  BSA and  fibrinogen are known to adsorb to a wide variety of surfaces and are often used as test proteins in studies of protein adsorption. Four surfactant combinations were tested: PFO mixed with FC-40 (1:5 v/v ratio) as the carrier phase with and without 0.1% Tween 20 surfactant added to the aqueous phase, and the OEG-capped fluorosurfactant used in (25) mixed with FC-40 (1:4 v/v ratio) as the carrier phase with and without 0.1% Tween 20 added to the aqueous phase. The latter fluorosurfactant was extracted from Zonyl FSO-100 as described in the supporting information of (25). Fluorescent images of stored droplets of FITC-labeled BSA are shown in Figure 13. The PFO/Tween 20 combination was tested first with a 100 nM solution of BSA (Figure 13A). No apparent adsorption to the droplet interface was observed, as fluorescent intensity fades towards the edge of the droplet. Next, the BSA solution without Tween 20 using the same carrier fluid was imaged, and fluorescence was virtually undetectable using identical camera exposure and gain settings. A 10x increase in BSA concentration to 1 µM was required to obtain comparable fluorescent intensity (Figure 13B), providing evidence of BSA adsorption to PDMS channel walls before encapsulation into droplets. Moreover, adsorption to the droplet interface is clearly visible in Figure 13B as a ring of increased fluorescent intensity at the edge of the droplet. The contrast between Figure 13A and B indicate that inclusion of Tween 20 prevents protein adsorption to droplet interfaces. Using another device, the OEG-capped fluorosurfactant/Tween 20 combination was tested, again using a 1 µM BSA solution (Figure 13C). No apparent adsorption to the droplet interface was observed. However, this fluorosurfactant was noticeably less effective than PFO at preventing droplet adhesion onto PDMS walls during both transport and storage. During transport, this resulted in the occasional breakoff of satellite droplets. The BSA solution without Tween 20 was then imaged (Figure 13D), and fluorescent intensity of the droplet decreased relative to the solution with Tween 20, again confirming that Tween 20 35  Figure 13: Fluorescent micrographs of stored droplets of FITC-labeled BSA using different aqueous and fluorous surfactants. (A) 100 nM BSA + 0.1% Tween 20, FC-40 + PFO, (B) 1 uM BSA, FC-40 + PFO, (C) 1 uM BSA + 0.1% Tween 20, FC-40 + OEG fluorosurfactant, (D) 1 uM BSA, FC-40 + OEG fluorosurfactant. prevents adsorption to PDMS channel walls. In contrast with the test case using PFO, no BSA adsorption to the droplet interface is seen, corroborating the previous report stating that OEGcapped fluorosurfactant prevents BSA adsorption to droplet interfaces while PFO (alone) does not (25). Similar results for the above surfactant combinations were obtained with Alexa 488-labeled fibrinogen (data not shown). Although only BSA and fibrinogen were tested, these results suggest that the inclusion of 0.1% Tween 20 in the aqueous phase prevents protein adsorption to both PDMS surfaces and fluorous/aqueous droplet interfaces for both fluorosurfactants tested. Considering that PFO was observed to be superior to OEG-capped fluorosurfactant at preventing unwanted droplet adhesion to channel walls, it was concluded that the addition of PFO to the carrier phase and Tween 20 to the aqueous phase prevents unwanted protein adsorption while maintaining desired droplet-based fluid-handling functionality. 36  3.3  Quantification of cross-contamination during formulation and elution  On-chip quantitative polymerase chain reaction (qPCR) was used as a sensitive assay to quantify the cross-contamination between storage chambers during both formulation and elution. In addition to providing an accurate method of assessing cross-contamination, on-chip nucleic acid amplification is necessary for the genetic and gene expression analysis of cells, which are target applications of this device. Thus, the sensitivity and efficiency of on-chip qPCR in stored droplets was important to establish. Exploiting the on-chip formulation capabilities of the device, 90 stored droplets of varying template concentration were formulated by combining 100 pump increments (~13.3 nL) of either human genomic DNA (gDNA) or water to achieve concentrations of 44.33 (133 pg, N=4), 11.08 (33.25 pg, N=4), 2.66 (7.98 pg, N=39), 0.89 (2.66 pg, N=39), and 0 (N = 4) haploid genome copies per reaction, assuming each haploid genome has a mass of 3 pg. qPCR reactions were then assembled by dispensing equal volumes of PCR master mix to each storage chamber, including primers and a hydrolysis probe designed for the detection of the RNase P gene which is found at a single copy per haploid genome. Following reaction assembly, the device was thermocycled on a prototype microfluidic qPCR instrument (Biomark, Fluidigm) for 40 cycles, and fluorescent images of the droplet array were acquired at each cycle.  The endpoint  fluorescent image acquired after the last cycle is shown in Figure 14A. Fluorescent images were analyzed using custom software in order to generate a qPCR curve for each stored droplet (Figure 14B) and cycle threshold (CT) values were extracted from these curves (Figure 14C). For the two most concentrated template dilutions, CT values were 22.56 (SD = 0.12) and 24.49 (SD = 0.14) respectively. The mean absolute precision in concentration measurement, calculated as 2!" − 1, is thus 9.4% which is near the limit of qPCR.  The difference in CT (ΔCT)  corresponding to the 4-fold concentration difference between the two most concentrated template dilutions was found to be 1.93 ± 0.18. PCR efficiency, calculated as −1 + 10!!  !  where x is the slope of the CT vs. log10(concentration) curve, is 105%. At the two lowest template concentrations, we observed digital patterns of amplification, with 37/39 and 18/39 positive reactions for template concentrations of 2.66 and 0.89 haploid genome copies per chamber respectively.  These frequencies fall within symmetric 95% binomial confidence  intervals constructed around the expected concentrations: 31/39 to 38/39 for 2.66 copies per chamber, and 16/32 to 29/32 for 0.89 copies per chamber. Taken together, these results indicate 37  Figure 14:  Efficient on-chip qPCR with single molecule sensitivity. (A) Endpoint  fluorescent image of the chamber array following 40 cycles of PCR. Storage chambers were loaded with varying amounts of human genomic DNA template, indicated as expected haploid equivalents per droplet.  Blue rectangles denote control reactions  mixed off-chip. Scale bar is 1 mm. Digital patterns of amplification are observed for the two lowest concentrations. (B) qPCR curves for each stored droplet. (C) Mean CT values and standard deviations from B at each template dilution. that highly efficient nucleic acid amplification with single molecule sensitivity can be performed in this device. Details of PCR reagents, thermocycling protocols, qPCR instrumentation, and image analysis software algorithms are provided in Appendix B. After establishing the sensitivity of this PCR assay, it was used to quantify cross-contamination between consecutively loaded storage chambers. 50 storage chambers were alternately loaded in 38  a checkerboard pattern, each receiving 100 pump increments (~13.3 nL) of PCR reagents premixed with either gDNA (~1476 haploid genome copies) or no template, and the device was again thermocycled on the qPCR instrument. This checkerboard pattern maximizes the shared fluidic path length to storage chambers with different contents, thus making this test of crosscontamination as stringent as possible. The endpoint fluorescent image after PCR shows that all stored droplets containing template were successfully amplified while none of the no-template control (NTC) chambers amplified, indicating that no detectable cross-contamination occurred during loading (Figure 15A). Based on the previously demonstrated ability to detect a single copy of the target gene, the upper bound on cross-contamination was determined to be 1 in 1476. It is highly likely that this upper bound is infact much higher, but this measurement was constrained by the maximum template concentration commercially available. This PCR assay was also used to measure cross-contamination between storage chambers during elution. 47 storage chambers were first loaded with 13.3 nL of water and another 47 chambers were then loaded with an equal volume of qPCR solution containing 18 haploid genome copies of human gDNA in a checkerboard pattern of alternating water and PCR droplets. Following 40  Figure 15: Low cross-contamination during formulation and elution of stored droplets. (A) Endpoint fluorescent image following 40 cycles of PCR of chambers loaded with either template (1476 genome copies per droplet) or buffer (NTC) in a checkerboard pattern. The image indicates no detectable cross-contamination between positive (white) and NTC (dark grey) droplets during loading of storage array. Scale bar is 1 mm. (B) Fold concentration difference of template in eluted pairs of droplets containing amplified template and water. Red line denotes mean. 39  cycles of on-chip PCR amplification, pairs of PCR product and water droplets were alternately eluted from the device into separate microfuge tubes and conventional off-chip qPCR was used to measure the degree of amplicon carry-over between tubes. The fold concentration difference between pairs of tubes containing eluted stored droplets of PCR product and water, calculated as 2ΔCT, is plotted in Figure 15B. The mean fold concentration difference was 4.84  ×  10!   with a standard deviation of 19.8. This level of cross-contamination is acceptable for the most stringent downstream analyses including DNA sequencing, cloning, or gene expression profiling. 3.4  Elution efficiency  Elution of storage chambers was tested by acquiring fluorescent images of a 40 nL stored droplet of 5 µM fluorescein-labeled 40-mer oligonucleotides before and after elution with ~500 nL of water. A separate chamber filled with water to an equal volume was then imaged for comparison. The mean fluorescent intensity, measured by Image J software, of the water-filled chamber was subtracted from that of the eluted chamber in order to account for background fluorescence, and the result was found to be 0.16% of the oligonucleotide-filled chamber before elution, indicating 99.84% sample recovery. Three other water-filled chambers of equal volume were also imaged to measure the noise of the imaging measurement. The coefficient of variation was found to be 1.7%. Inspection of the entire elution path, including the elution channel, by fluorescence microscopy indicated that none of the oligonucleotide solution remained in the device after elution. In the latest version of the elution protocol, at least 5 µL of water is used to elute each chamber, which is a 10-fold increase relative to the volume used for the elution test described above. This excessive elution volume also ensures that storage chamber contents do not remain attached to the elution nozzle exterior but sink to the bottom of the light mineral oilfilled microfuge tube.  40  Chapter 4 : Biological applications In this chapter, the versatility and unique fluid-handling capabilities of the device are demonstrated by its application to different types of single-cell analyses on both microbial and mammalian cells.  Sorting, culture, and genetic analysis by PCR and whole genome  amplification are first validated on single bacterial cells from laboratory strains. To illustrate its utility in environmental genomic studies, the device is then applied to the genomic analysis of single microbes and microbial aggregates isolated from environmental samples. Next, PCRbased genotyping of single human primary cancer cell nuclei is performed and initial results of a clinical study are presented that demonstrate how the device can be used to determine mutational evolution in human disease at single-cell resolution. Finally, results from whole transcriptome amplification experiments are presented that demonstrate how the device may be used to perform quantitative analysis of the transcriptional state of single cells. 4.1  Multi-parameter analysis of single microbes and microbial aggregates  Microorganisms represent the majority of all life forms on the planet (109), play a critical role in natural ecosystems (110) and human health (111), and harbour a plethora of useful enzymes with applications in biotechnology and bioenergy.  Understanding and exploiting this diversity  requires the study of the constituent species and dynamic interactions that are characteristic of microbial communities.  Unfortunately, it is estimated, based on analysis of molecular  phylogenetic markers in small-subunit (SSU or 16S) ribosomal RNA (rRNA), that the vast majority of microorganisms have not been isolated and cultured (63, 112) and are thus not amenable to conventional genomic and proteomic analyses requiring significantly more template material than is present in a single cell. Metagenomic analysis, involving the extraction, cloning, and analysis of microbial DNA directly from the environment, provides a culture-independent method of examining the genetic diversity, population structure, and metabolic potential of entire communities of microorganisms (113). However, as such analyses are performed on mixed pools of genomic DNA derived from large numbers of microorganisms (114), the ability to associate functional and phylogenetic marker genes to the same organism are lost, precluding the reconstruction of enzymatic pathways and the identification of species that perform ecological activities of interest.  41  The isolation and analysis of single microbes circumvents these limitations by excluding the “noise” contributed by other organisms, and is an ideal application for microfluidic systems. Micrometre-sized droplets are particularly well suited to the high-throughput isolation and assaying of single microbes, which, due to their small size, are difficult to manipulate by alternative hydrodynamic trapping mechanisms. Encapsulation of single microbes at limiting dilution has been used for PCR-based genetic analysis of single-bacteria (42, 45), single-cell enzyme assays (50, 71), and single-cell drug toxicology screens (47). Valve-based flow control in combination with microscopy, on the other hand, allows for phenotype-based sorting of microbes and this has been applied to the isolation and whole genome amplification of single bacteria with a specific morphology from a human oral sample (17). The present device combines these two strategies to perform phenotype-based sorting of single microbes at higher throughput than previously demonstrated, and in combination with programmable fluid-handling, is capable of executing a variety of single-microbe experiments. This versatility has been demonstrated by performing multiple user-defined single-cell applications using laboratory bacterial strains: phenotype-based sorting of bacteria out of a mixed sample followed by clonal analysis of growth rates, taxonomic identification of single bacteria by small subunit ribosomal RNA gene quantitative qPCR and sequencing, and highthroughput single-microbe whole genome amplification (WGA) and sequencing. The device was then applied to the genomic analysis of single cells and microbial consortia in environmental samples to demonstrate how it may be used to examine relationships between microbial community members. 4.1.1  Sorting and culture of single bacteria  As a first demonstration of phenotype-based sorting and isolation of single cells from a mixed population, a series of cell culture experiments were performed in which single Salmonella typhimurium bacteria, selected from a mixture of two strains expressing either green or red fluorescent protein (GFP or RFP), were isolated and grown in microdroplet reactors. The strains are genetically identical with the exception of the encoded fluorescent protein. For single-cell sorting, a cell suspension of S. typhimurium consisting of a 1:1 ratio of GFP- to RFP-expressing cells in cell culture media was prepared. Fluorescence-based cell sorting was performed as described in Section 2.3.2 and shown in Figure 1C using a fluorescent microscope equipped with 42  excitation and emission filters for GFP and RFP channels. The cell concentration was adjusted to ~1 cell/10 nL prior to use on-chip to ensure that no more than one cell was present at any given time in the channel intersection, which has a volume of ~300 pL. While manually monitoring the intersection by microscopy in one of the two fluorescent channels, the suspension was advanced down the sorting channel at a defined flow rate using peristaltic pumping. When a single cell was identified in the intersection, the filter set was switched to ensure no cells expressing the other fluorescent protein were present, and the cell was isolated and pumped into a droplet for delivery to a selected storage chamber. A combined fluorescent and brightfield image of a single RFP-expressing cell isolated in a stored droplet is shown in Figure 16A. Using this method, storage chambers were seeded with single GFP- or RFP-expressing cells to initiate monoclonal cultures (N=20 for each) and co-cultures containing one cell of each strain (N=20). In addition, suspensions of each strain at varying concentrations were loaded into reagent inlets in order to seed other chambers with 100 GFP- or RFP-expressing cells (N=5 for each), and co-cultures containing ~10 (N=5), ~100 (N=5), and ~1,000 (N=5) cells of each strain. A total of 85 cell culture experiments were seeded, each in a separate storage chamber, and then filled with growth media to a final volume of 40 nL. The device was then incubated at 25 °C and fluorescent images in both channels were acquired every 10 min for 23.3 hours. Growth curves were generated for each culture by analyzing fluorescent images using custom software to determine the total GFP and RFP expression in each microdroplet reactor at each timepoint (Figure 16B and C). Details of bacterial strain preparation, culture media, and image analysis software algorithms are provided in Appendix C. End-point confocal fluorescent images of the droplet array in both fluorescent channels acquired after incubation show that no GFP fluorescence was detected in the RFP-expressing monoclonal cultures and vice versa, indicating contamination-free single cell sorting (Figure 16E). Comparable plating efficiency was observed for both the GFP and RFP-expressing strains, with colony formation observed in 17 of 20 (85%) and 16 of 20 (80%) of the monoclonal GFP and RFP-expressing cultures respectively. Successful monoclonal cultures exhibited heterogeneous growth curves, showing that differences in the proliferative capacity of single microbes can be significant even in isogenetic populations. These differences resulted in stochastic variability in the final composition of co-cultures loaded with equal but varying numbers of cells (1, 10, 100,  43  Figure 16: On-chip culture of single sorted bacteria. (A) Combined brightfield and fluorescent micrograph of a single RFP-expressing cell in a stored droplet. Growth curves of each on-chip culture seeded with (B) GFP-expressing and (C) RFP-expressing cells. (D) Scatter plot of normalized endpoint fluorescence intensity in GFP and RFP channels for mixed cultures seeded with different numbers of both strains. (E) Overlaid GFP and RFP-channel confocal images of all cultures in the stored droplet array after incubation. Cultures were seeded with (1) single cells (dark parts of the array are unsuccessful cultures), (2) a single cell of each strain, (3) ~ 1000 cells of each strain, (4) ~ 100 cells of each strain, (5) ~ 10 cells of each strain, (6) ~ 100 GFP-expressing cells, and (7) ~ 100 RFP-expressing cells. 1,000) from each strain (Figure 16D). Variability was largest when starting from single-cell cultures and was progressively reduced as the size of the starting populations increased. As a test of single-microbe manipulation, this simple experiment validates the device’s ability to perform phenotype-based sorting of single bacteria at higher throughput than previously demonstrated. While stochastic confinement of single bacteria has been reported for the study of 44  variability in pathway activation (115), the droplet-based functionality of this device can be used to expose single microbes to any combination of stimulants allowing for a wide variety of singlecell experiments.  In addition, this device is the first to enable confinement of multiple  specifically selected single microbes into the same nanolitre-volume. The simple co-culture experiment enabled by this capability illustrates how stochastic differences between individual cells can lead to large disparities in the proliferation of small numbers of two microbial species populating a microenvironment, even in the case of equal fitness for large starting numbers of cells. 4.1.2  PCR-based genotyping of single sorted microbes  As a second demonstration of single-cell analysis, genotyping experiments based on PCR amplification and sequencing of the 16S rRNA gene were performed on single bacteria sorted from a mixed population of Escherichia coli and RFP-expressing S. typhimurium. E. coli cells were stained with fluorescent SYTO9 DNA stain, which fluoresces in the GFP channel, in order to distinguish them from RFP-expressing S. typhimurium by fluorescence microscopy as described above. Storage chambers were loaded with single S. typhimurium (N=30), single E. coli (N=29), ~50 S. typhimurium (N=5), and ~50 E. coli (N=5), and mixed with PCR reagents containing an intercalating dye and primers targeting a 144-bp segment of the 16S rRNA gene. The sequence of this segment differs by four single base pair mismatches between the two species. On-chip qPCR was performed, with an initial 3-minute heating step at 95 C included to perform heat lysis of bacteria, and qPCR curves for all reactions were constructed from acquired images (Figure 17A). The target sequence was amplified in 16 of 30 (53%) single S. typhimurium, and 25 of 29 (86%) single E. coli, as determined by qPCR curves for each reaction. The difference in mean CT between single and ~50-cell reactions was 1.96 and 7.24 for S. typhimurium and E. coli respectively, indicating sub-optimal PCR efficiencies of 71.7% and 636% respectively. Following PCR, the amplicons from each reaction were eluted and six successful single-cell reactions from each species were chosen at random for further off-chip amplification and capillary sequencing. Based on the sequence data at the four mismatched positions of the 144-bp amplicon, all six single E. coli cells and five of six single S. typhimurium cells were correctly  45  Figure 17: 16S rRNA qPCR of single sorted S. typhimurium and E. coli and multiple cells of each species. (A) qPCR curves for all reactions. (B) Mean CT values and standard deviation for single and multiple cell reactions. identified. The single S. typhimurium amplicon that could not be identified also did not match the expected sequence for E. coli. Overall, the success rate of PCR amplification from single cells was 41 of 59 (69%), which is comparable to previous reports (5, 42, 45). To determine whether reaction failures were specific to the assay used, additional experiments were run in which a strain-specific fragment of the E. coli 16S rRNA gene (present at 7 copies per genome) was amplified in single E. coli cells using an optimized primer set (116). A total of 77 reactions were formulated and amplified containing either single cells (N=62), ~100 cells (N=5), or cell suspension fluid containing no cells as determined during cell sorting (N=10). qPCR curves showed that the target sequence was successfully amplified in 60 of 62 (97%) single cells, 4 of 5 (80%) 100-cell reactions, and none of the no-cell control reactions (Figure 18A). The difference in mean CT between single and 100-cell reactions (Figure 18B) was found to be 6.52, indicating an excellent assay efficiency of 102.7%. Amplicons of all on-chip reactions were again eluted and capillary sequencing of 10 randomly selected successful single-cell reactions was performed following an additional round of off-chip amplification. All 10 sequenced amplicons were confirmed to have the expected sequence.  Details of PCR reagents, thermocycling protocols, image analysis software  algorithms, DNA sequencing, and sequencing data analysis are provided in Appendix D. 46  Figure 18: Strain-specific 16S rRNA qPCR of single sorted E. coli, multiple cells, and no-cell controls. (A) qPCR curves and (B) mean CT values and standard deviation for single and multiple cell reactions. The discrepancy between the number of successful single-E. coli and single-S. typhimurium amplifications in the first experiment may indicate that E. coli genome is more easily made accessible by heat lysis than that of S. typhimurium, or that the assay primers bind to the target sequence in E. coli with higher efficiency than in S. typhimurium. The results of the second experiment indicate that with an optimized assay and efficient cell lysis, robust PCR amplification can be achieved from single microbes. These experiments demonstrate that this device could be used to taxonomically identify any single microbe in a given sample by amplification and sequencing of phylogenetic marker genes. This could find use in a variety of applications in which the microbial constituents in some sample of interest must be identified. For example, the identification of bacterial pathogens in a clinical setting could be performed without the need to perform culture and DNA extraction (117). 4.1.3  Whole genome amplification of single cells  While amplification and sequencing of the 16S rRNA gene in single microbes allows for their taxonomic identification, the complete characterization of an organism requires knowledge of its complete genome sequence. Although the genome sequences of unculturable microbes that happen to dominate a given environment may be obtained by direct sequencing of extracted DNA (118), the genomes of individual community members in highly diverse environments may not be obtainable with such an approach, or may only be obtained at the high cost of sequencing 47  to very high depth. In such cases, individual genome sequences can only be obtained by sequencing single cells. To this end, there has been much effort devoted to whole genome amplification (WGA) of single microbes from environmental samples to generate sufficient DNA quantities for sequencing (64). The two most common methods of WGA are PCR-based amplification, in which a library of DNA fragments with universal priming sites are derived from genomic DNA and are then PCRamplified by universal primers (119, 120), and multiple displacement amplification (MDA), in which highly processive phi29 DNA polymerase and random primers perform isothermal amplification of the genome based on strand-displacement synthesis (121). Single microbes have previously been physically isolated for WGA by micromanipulation (122), FACS (123127), and microfluidic devices (17, 18, 20). The most significant challenge in obtaining useful WGA product is the minimization of representational bias, which can compromise downstream PCR-based genotyping efforts and significantly increases the sequencing effort and cost required to obtain useful genome coverage. Unfortunately, both PCR and MDA-based WGA methods exhibit significant bias (66, 128). An enzymatic library normalization method that degrades overly abundant sequences has been used to reduce bias in WGA product (124). Bias in individual single-cell reactions has also been mitigated by combining sequencing datasets from multiple clonal single-cell WGA reactions that each amplify different genomic regions (20), although such clonal populations may not be accessible in many cases. There is evidence suggesting that reduced reaction volumes result in lowered MDA representational bias although the mechanism remains unclear (18). As a third demonstration of single-microbe analysis, the device was used to perform single-cell WGA followed by product recovery and shotgun sequencing. Both PCR and MDA-based chemistries were evaluated.  PCR-based whole genome amplification of single microbes  A commercially available PCR-based WGA protocol that has not previously been applied in microfluidic devices (Picoplex, Rubicon Genomics) was first evaluated. While the exact details of the methods implemented by this protocol are proprietary and thus not publicly available, it is believed that the protocol consists of a cell lysis step at an elevated temperature by means of a 48  thermostable proteinase, synthesis of a fragment library of genomic DNA initiated by multiple cycles of priming by random primers including a universal adapter sequence, and PCRamplification of the fragment library using primers that bind to the universal adapters. Using two devices, this multistep protocol was performed on single E. coli cells (N=127), no-cell control reactions containing only cell suspension fluid (N=20), ~10 cells (N=10), and ~1000 cells (N=10). Off-chip qPCR using an assay targeting a strain-specific fragment of the 16S rRNA gene (used in Section 4.1.2) was used to quantify amplified copy number of this gene (present at 7 copies per E. coli genome) in eluted WGA product from different starting cell numbers by comparison of CT values against a standard curve generated from qPCR reactions on dilutions of purified E. coli gDNA with known copy number (Figure 19). Reactions containing no cells, single cells, ~10 cells, and ~1000 cells produced mean copy numbers of 220, 2.5 × 106, 3.2 × 106, and 4.3 × 106 respectively. The coefficient of variation of copy number in all single-cell reactions was 507%, and 72 of 127 (57%) single-cell reactions resulted in at least a 100-fold amplification of the 16S rRNA gene relative to the 7 copies present in a single cell. Product from six successful single-cell reactions, two no-cell control reactions, and one 1,000cell reaction were chosen for sequencing, along with a bulk sample of unamplified E. coli gDNA as a positive control, using an Illumina Genome Analyzer 2 sequencing instrument. Sequencing libraries for each single cell were constructed both from reaction product eluted directly from the chip and from samples that had been subjected to a second round of WGA off-chip in order to increase yield. Sequence data was aligned to the E. coli reference genome to generate coverage statistics for each sample, which are summarized in Table 1.  Details of WGA reagents,  thermocycling protocols, quantification of WGA product, DNA sequencing, and sequencing data analysis are provided in Appendix E. Reference genome coverage of the single-cell reactions from on-chip WGA product alone ranged from 15.2% to 64.6% for mean coverage depths (total bp of sequence data / 4.6 Mbp E. coli genome length) ranging from 77x to 220x.  After a second round of off-chip WGA,  reference genome coverage ranged from 24.5% to 62.8% for mean coverage depths ranging from 36x to 81x. No-cell control reactions showed no significant alignment to the reference genome. To the author’s knowledge, this is the first demonstration of significant genome sequence 49  Figure 19: 16S rRNA copy number yielded from each PCR-based WGA reaction. Red circles denote mean. recovery from a single cell using nanolitre-volumes of reagents alone, without a second amplification step in a microlitre-volume. The ability to robustly perform single-cell WGA in such small volumes would allow for large-scale single-cell sequencing studies that are currently intractable due to prohibitively high reagent costs. However, the representational bias of these single-cell reactions was much higher than in previously reported microfluidic single-cell WGA using the same organism (18), and this limited the reference genome coverage to below 42% in five of six single-cell reactions. Indeed, PCR-based WGA amplification is known to exhibit large bias (128). This bias is likely responsible for the relatively low fraction of single-cell reactions that resulted in a 100-fold amplification of the 16S rRNA gene, as it is likely that genomic regions other than the one targeted by the qPCR assay were preferentially amplified. The fact that the highest reference genome coverage obtained from a single-cell reaction was comparable to that of the 1000-cell reaction indicates that representational bias of the WGA reaction, and not starting template quantity, is the limiting factor for reference genome coverage.  50  75 bp reads no cell control 1 no cell control 2 single cell 1 single cell 2 single cell 3 single cell 4 single cell 5 single cell 6 ~1000 cells unamplified gDNA  On-chip WGA % of % of genome reads covered aligned at ≥ 1x  % of genome covered at ≥ 10x  On-chip and off-chip WGA % of % of % of genome genome 50 bp reads covered covered reads aligned at ≥ 1x at ≥ 10x  5663384  1.0  7.7  0.1  4639808  1.1  5.3  0.1  4341480 13483184 9784130 4708954 5738682 10268078 10644760  0.2 55.3 47.2 5.7 1.0 48.4 34.5  8.7 64.6 40.6 15.2 18.6 28.2 30.3  0.1 43.1 22.5 4.1 1.6 13.8 14.0  6677002 7458890 6941568 3304200 7354470 6085898 5373842 8074402  0.1 79.8 80.2 52.2 68.4 56.6 42.6 78.8  5.4 62.8 41.9 24.5 42.3 26.3 27.5 61.6  0.0 40.0 22.9 8.5 24.8 12.1 12.0 34.5  62513866  90.9  99.8  99.7  Table 1: Sequencing statistics for PCR-based WGA of single E. coli.  MDA-based whole genome amplification of single microbes  A commercially available MDA-based WGA protocol (Repli-G, Qiagen) was also evaluated using the same E. coli. strain. Initially, the protocol recommended by the manufacturer was followed, which lyses the cell and denatures the genomic DNA using an alkaline lysis buffer containing dithiothreitol (DTT), followed by addition of a neutralization buffer, and phi29 DNA polymerase and random primers for the MDA reaction. However, single-cell reactions were unsuccessful as determined by qPCR of a strain-specific fragment of the 16s rRNA gene, as described above. Modifications to the recommended protocol were tested and it was discovered that the omission of DTT in the alkaline lysis buffer was critical for successful single-cell MDA. In order to directly compare reactions performed with and without DTT, a total of 90 reactions were performed on a single device using either a lysis buffer including DTT or another in which DTT was replaced with water. For each lysis buffer, MDA reactions were performed on single cells (N=30), ~400 cells (N=5), and ~4000 cells (N=5). Ten no-cell control reactions were also performed using the lysis buffer without DTT. After completion of the MDA reaction, products were eluted and amplified 16S rRNA gene copy number in each reaction was quantified (Figure 20).  51  Figure 20: 16S rRNA copy number yielded from each microfluidic MDA reaction. Red circles denote mean. MDA reactions using DTT in the lysis buffer resulted in variable 16S rRNA gene amplification dependent on the starting template quantity. Mean copy number yielded by the single-cell reactions was comparable to that of the no-cell control reactions (61 and 95 respectively), while the 400-cell and 4000-cell reactions yielded mean copy numbers of 1.5 × 104 and 5.5 × 107 respectively.  In contrast, MDA reactions performed without DTT resulted in comparable  amplified copy number for all starting template quantities with single-cell, 400-cell, and 4000cell reactions having means of 2.2 × 107, 6.9 × 107, and 4.5 × 107 respectively. As the DNA yield of MDA reactions should be independent of the amount of starting material (129), these results thus suggest that, in the present device, DTT has an inhibitory effect on the MDA reaction that is dependent on the starting template quantity. Interestingly, such inhibition was not observed in a previous report in which single-cell microfluidic MDA reactions were performed using the same MDA chemistry and organism (18). A mechanism for this inhibition is as of yet undetermined. 52  The mean 16S rRNA copy number resulting from single-cell MDA reactions performed without DTT was 8.8 times that of the single-cell PCR-based WGA reactions, and the coefficient of variation was 205%, 2.5 times less than that of the single-cell PCR-based WGA reactions. These results indicate that MDA is a more robust protocol for single-cell WGA than the PCR-based protocol previously used. In light of this improvement in 16S rRNA gene amplification by MDA, qPCR assays were used to quantify the amplified copy number of 10 single-copy loci across the E. coli genome (130) in 2 no-cell control reactions and all 30 single-cell reactions performed without DTT in the lysis buffer (Figure 21) as an initial gauge of representational bias in the MDA reactions. For all 30 single-cell reactions, the coefficient of variation of the mean copy number for all 10 loci was 84%, lower than what was reported in (18). To more completely assess representational bias, sequencing of product from one of these reactions was performed, this time using an Ion Torrent PGM sequencing instrument. Conventional microlitre-volume MDA reactions were also performed and their products sequenced in order to compare their performance with microfluidic reactions. Sequencing was performed on a nanolitre-volume microfluidic single-cell reaction, a second microlitre-volume MDA reaction performed on the product of this microfluidic reaction, a microfluidic no-cell control reaction, a conventional microlitre-volume MDA reaction on a single FACS-sorted cell, and unamplified purified E. coli genomic DNA as a positive control. Sequencing reads and assembled contigs were aligned to the E. coli reference genome to generate coverage statistics for each sample, which are summarized in Table 2. Details of WGA reagents and protocols, quantification of WGA product, DNA sequencing, and sequencing data analysis are provided in Appendix F.  53  Figure 21: Copy number of 10 loci yielded from microfluidic single-cell MDA reactions performed without DTT.  Sequencing effort (Mbp) Fraction of data aligned to reference Fraction of reference covered Mean length of assembled contigs (kbp) Total length of assembled contigs (Mbp) Fraction of reference covered by contigs  Unamplified gDNA 91.4  nL MDA  nL/µL MDA  µL MDA  528  225  223  nL No-cell MDA 559  86.5%  16.5%  73.3%  84.8%  86.8%  99.5%  99.4%  99.4%  99.0%  3.28%  1.92  2.04  7.79  2.62  2.15  4.12  4.23  4.55  4.31  0.133  85.6%  87.6%  94.6%  87.4%  2.73%  Table 2: Sequencing statistics for MDA-based WGA of single E. coli.  Contamination  As can be seen in Table 2, relative to the other sequenced samples, the nanolitre-volume MDA product contained a very small fraction of sequencing data that aligned to the expected reference 54  genome, with only 16.5% of data aligning to E. coli. To discover the source of the remaining unaligned data, it was compared to a database containing the reference genomes of all known organisms  using  the  Basic  Local  Alignment  Search  Tool  (BLAST,  http://blast.ncbi.nlm.nih.gov/). It was discovered that 68% of the data largely aligned to a 30 kbp contiguous section between positions 6.54 and 6.57 Mbp of the reference genome of the bacteria Delftia acidovorans (Figure 22A). Initially, it was thought that this was due to the presence of a contaminant amplicon in the sequencing library preparation reagents. Product of another singleE. coli microfluidic MDA reaction was thus selected for a second-round microlitre-volume MDA reaction, and product of both the microfluidic reaction and the microlitre-volume reaction were sequenced using new reagents. BLAST analysis showed that 75% of data from the microfluidic reaction again aligned to the D. acidovorans genome, but this time to a different 5kbp contiguous section between positions 4.293 and 4.298 Mbp (Figure 22B). BLAST analysis was also performed on the microfluidic no-cell control reaction, which indicated that only 0.01% of reads aligned to D. acidovorans. These results indicate the presence of different single fragments of contaminant D. acidovorans DNA in the two single-E. coli microfluidic reactions and the lack of such contamination in the microfluidic no-cell reaction containing cell suspension fluid but no cells. This could only be attributed to the presence of contaminant fragments at limiting dilution.  Figure 22: Read coverage of the Delftia acidovorans reference genome by sequencing data from two separate single - E. coli microfluidic MDA reactions. 55  Similarly high fractions of sequencing data aligning to D. acidovorans from single-cell MDA reactions using reagents from the same manufacturer have previously been reported and attributed to MDA reagent contamination (20, 122, 131). Using microfluidic digital MDA, the level of contamination in MDA reagents from multiple manufacturers has been found to be on the order of hundreds of fragments per microlitre (22). Based on sequencing results and these prior reports, it is most likely that the cause of the observed contamination is the presence of D. acidovorans DNA fragments in one or more of the MDA reagents used. UV treatment of MDA reagents has been shown to effectively suppress amplification of contaminant DNA (66, 131), and will be performed in all future MDA experiments. Only 0.3% of the reads from the microlitre-volume single-cell MDA reaction aligned to D. acidovorans, and it is believed that this is due to the use of a different, and relatively contamination-free, batch of MDA reagents than was used for the on-chip reactions. In the two microfluidic single-cell MDA reaction products sequenced, the fraction of reads aligned to D. acidovorans greatly exceeded the fraction of reads aligned to E. coli (68% versus 17% and 75% versus 7% in the first and second reactions sequenced respectively). Thus, if the proposed explanation for contamination is correct, single contaminant DNA fragments with lengths on the order of kbp must have been amplified preferentially over the much larger E. coli genome. It is not completely clear why this might have happened, but one explanation is that cellular debris may have remained attached to genomic DNA after cell lysis, making the genome less accessible to MDA reagents than a bare contaminant DNA fragment. It is also possible that the contaminating fragments may be circular integrons, which amplify more efficiently during the MDA reaction. Neither of these hypotheses has been tested to date. A more puzzling phenomenon is the decrease in the fraction of reads aligned to D. acidovorans from the nanolitre-volume reaction to the second-round microlitre-scale reaction. This was observed for two sequenced single-cell reactions where this fraction dropped from 68% to 14% and 75% to 20% in the first and second single-cell reactions sequenced respectively. One would expect that the fraction of total template represented by the more abundant species would be increased if not kept constant after a second-round microlitre reaction, but not significantly decreased as observed. 86.8% of the sequencing data from the microfluidic no-cell control reaction aligned to 3.28% or ~151 kbp of the E. coli reference genome, suggesting the presence of E. coli DNA in either the 56  cell suspension fluid, presumably derived from the sorted E. coli cells, or in the MDA reagents. However, if the MDA reagents contained E. coli DNA, one would expect that such contamination would have also been present in at least one of the sequenced microfluidic singlecell reactions that were also contaminated with D. acidovorans DNA. If this occurred, the contaminant D. acidovorans DNA would have had to be amplified with high preference over contaminant E. coli DNA in order to produce the observed dominant fraction of sequencing data aligned to D. acidovorans in these reactions. As such preferential amplification is unlikely, it is believed that the source of the E. coli DNA in the no-cell control reaction is the cell suspension fluid.  Exceptionally low representational bias in nanolitre MDA  To compare the representational bias of all MDA reactions, the sequencing data aligned to E. coli from each reaction type was first randomly subsampled at mean coverage depths ranging from 1x to 16x in order to compare equal quantities of data for each reaction. Coverage maps for each reaction type displaying the number of sequencing reads covering each position of the E. coli reference genome at 16x mean coverage depth are shown in Figure 23. From these coverage maps, it can be qualitatively seen that of the single-cell MDA reactions, the nanolitrevolume reaction has the least variation in coverage followed by the combined nanolitre/microlitre reaction and the microlitre reaction in order of increasing variation. Overlaid normalized coverage maps, showing minimum to maximum coverage, for the two nanolitre single-cell reactions sequenced are shown in Figure 24. It can be qualitatively seen that many regions of peak coverage are shared between the two reactions, perhaps suggesting sequence-based bias that leads to preferential amplification of these regions. To more quantitatively assess the bias of each MDA reaction type, the fraction of the reference covered at mean coverage depths ranging from 1x to 16x were found as shown in Figure 25. The ideal coverage that would be obtained from a perfectly unbiased sample, as predicted by Poisson statistics, is also shown. The single-cell nanolitre and combined nanolitre/microlitre reactions have very similar reference coverage to that of the unamplified genomic DNA at all mean coverage depths, while that of the single-cell microlitre reaction is significantly lower, confirming that bias is greatly reduced by performing MDA in a nanolitre volume. The nanolitre reaction in fact has slightly higher reference coverage than that of the combined 57  Figure 23: Read coverage of the E. coli reference genome by sequencing data from (A) unamplified genomic DNA, (B) nanolitre MDA, (C) combined nanolitre/microlitre MDA, (D) microlitre MDA, (E) and nanolitre no-cell control MDA. 58  nanolitre/microlitre reaction for all mean coverage depths, suggesting that a microfluidic reaction alone can achieve equivalent or slightly reduced representational bias relative to the combined nanolitre/microlitre reaction with a thousand times lower reagent consumption. As might be expected, this difference is most pronounced at lower mean coverage depths and decreases at higher depths. At a mean coverage depth of 2x, the unamplified genomic DNA, nanolitre reaction, combined nanolitre/microlitre reaction, and microlitre reaction have reference coverage of 84.4%, 83.1%, 81%, and 72.2% respectively. At a mean coverage depth of 8x, the single-cell nanolitre reaction covers 99.1% of the reference. These results, to the author’s knowledge, represent the highest reference coverage and lowest representational bias obtained from a singlecell WGA reaction reported to date. To further depict the bias of each reaction type, the fraction of the reference genome covered at various depths for a mean coverage depth of 16x was plotted in a histogram (Figure 26). The  Figure 24: Overlaid normalized read coverage of the E. coli reference genome by sequencing data from two separate single- E. coli nanolitre MDA reactions (red and cyan). Overlapping regions are in dark cyan. 59  result that would be obtained for an ideally unbiased sample, as predicted by Poisson statistics, is also shown. A numerical measure of representational bias is the coefficient of variation (CV) of the coverage of each position of the reference. CV values for an ideal sample, the unamplified genomic DNA, the nanolitre single-cell MDA reaction, the combined nanolitre/microlitre singlecell MDA reaction, and the microlitre single-cell MDA reaction are 25%, 36%, 45%, 57%, and 90% respectively, again illustrating that the nanolitre MDA reaction has the lowest representational bias of the three single-cell reactions. The exact mechanism responsible for this exceptionally low bias is currently unknown. It has been suggested that reduced bias in nanolitre volumes is due to the increased concentration of the target template relative to contaminants, which results in more DNA polymerase molecules per target template (18). While this should result in an increase in amplified DNA originating from the target template, it does not provide a mechanism for reduced bias in the amplification of this template. Indeed, the different quantities of contamination that happened to be present in  Figure 25: Reference genome coverage versus mean sequencing coverage depth for each sample. 60  Figure 26:  Histogram showing coverage depth of reference genome for each MDA  reaction type at a mean coverage depth of 16x.  For each sequenced sample, the  coefficient of variation (CV) of the coverage for each position of the reference genome is shown. the different single-cell reactions described above provides evidence contrary to this theory, as the single-cell nanolitre-volume reaction in which the most contamination was observed resulted in the least bias while the single-cell microlitre-volume reaction in which the least contamination was observed resulted in the most bias. One explanation for the reduction in bias with reaction volume involves the difference between the volume occupied by the genome and the volume of the reaction. Consider the genome modeled as a sphere that is packed with DNA, where each part of the sphere consumes WGA reagents and thus acts as a reagent sink. At the beginning of the reaction, all parts of the sphere see the same local concentration of reagents, resulting in uniform amplification of all parts of the 61  genome.  As the reaction progresses, the reagents inside the sphere are depleted and new  reagents must be drawn in from other parts of the reaction volume outside the sphere. However, since all parts of the sphere consume reagents and new reagents can only arrive from outside the sphere, a gradient of local reagent concentration within the sphere will be established, with the highest concentration at the sphere’s exterior and the lowest concentration at the sphere’s centre. This gradient thus results in preferential amplification of DNA sequences that happen to be closer to the exterior of the sphere, which have access to higher local concentrations of reagents. The gradient persists until all reagents in the reaction volume are depleted. In fact, as regions on the outside are preferentially amplified they generate increased template and further restrict transport to the interior of the sphere. The larger the reaction volume relative to the sphere of DNA, the longer this gradient remains and the longer the biased amplification is allowed to occur. This mechanism would explain the observed reduction in bias with smaller reaction volume and also explains why a subsequent microliter volume amplification, in which the genome has already been amplified and is well-mixed in the reaction, does not introduce a large amount of additional bias. In the case where the size of the sphere of DNA is equal to the size of the reaction volume, there is no part of the reaction volume outside of the sphere from which reagents can arrive, and the gradient is thus never established. The reaction proceeds with all parts of the genome seeing equal local reagent concentrations until reagents are depleted from the reaction volume in all regions simultaneously. This scenario would result in the least biased amplification. If the genome is modeled as a random coil composed of a chain of monomers with a persistence length of 50 nm, then the root mean square diameter of the coil is !! where N is the number of monomers and L is the persistence length. For the 4.6 Mbp E. coli genome, this diameter is ~9 µm. The diameter of the stored droplet in which the microfluidic MDA reaction is performed is 660 µm. Thus, although the diameter of the microfluidic reaction volume does not match the diameter of the modeled genome, the discrepancy is much smaller than in a microlitre reaction. This hypothesis could thus be tested by comparison of the bias in two single-cell MDA reactions in microlitre-volumes: one in which the genome is fragmented and distributed homogenously throughout the reaction volume and another in which the genome is not fragmented.  A  microlitre-volume would allow for a large difference in spatial confinement of the genomic 62  DNA between these two cases. If the hypothesis is correct, the reaction in which the genome is fragmented and allowed to occupy the entire reaction volume should have no gradient in local reagent concentration and should thus result in less bias. At the time of this thesis submission these experiments are ongoing. The demonstrated ability to perform single-cell WGA at high throughput with the lowest reported representational bias to date using nanolitres of reagent per reaction has significant implications for future single-cell genomic studies. Besides the obvious reduction in WGA reagent costs, reduced representational bias allows for genome coverage with reduced sequencing effort, thus also reducing sequencing costs. This capability thus has the potential to enable currently intractable genomic studies of large numbers of single cells. This low level of representational bias has, however, thus far only been demonstrated on E. coli. Further work is thus required to show that similar results can be obtained with cell types of higher biological interest such as microbes from environmental samples or single human cells of medical relevance. These experiments are ongoing and it is anticipated that results similar to those obtained with E. coli should be achievable provided that the genome is made accessible by adequate cell lysis. 4.1.4  Environmental genomics  Immediately following the PCR-based WGA experiments on E. coli described in Section, this chemistry was applied to the WGA and sequencing of microbes in environmental samples to explore genomic relationships within natural microbial communities. The PCR-based protocol was used because, at the time, the MDA-based protocol had not yet been shown to yield successful single-cell amplifications on the model organism E. coli. Samples were selected from three environments representing varying levels of structural complexity. Environment 1 (ENV1) was a bacterial enrichment culture from seawater chosen to represent a low-complexity environment. Environment 2 (ENV2) was a human oral biofilm chosen to represent a high-complexity microenvironment. Environment 3 (ENV3) was a 3-8 µm fraction from deep-sea sediments associated with methane seepage. ENV1 and ENV3 were obtained from Steven Hallam’s laboratory in the Department of Immunology and Microbiology. Based on the complexity and aggregation state of each environment, alternative on-chip sorting strategies were used. Details of environmental sample preparation are provided in Appendix G. 63  Single cells were isolated from ENV1, individual extended filamentous aggregates were isolated from ENV2, and individual spherical aggregates were isolated from ENV3. A total of 203 onchip WGA reactions using the previously described PCR-based protocol were performed (50 in ENV1, 60 in ENV2, 93 in ENV3) including 5 no-cell controls consisting of equal volumes of cell suspension fluid containing no visible cells. A total of 74 samples representing each of the environments were randomly selected for a subsequent round of off-chip amplification in a microlitre-volume and sequencing library construction, resulting in 72 successful libraries: 24 single cells from ENV1, 22 filamentous aggregates from ENV2, 23 spherical aggregates from ENV3, and 3 no cell control samples. The two remaining samples were excluded due to suspected contamination or mislabeling during library preparation. Samples were indexed, pooled and sequenced on a single lane of an Illumina Genome Analyzer II instrument, generating a total of 4.8 billion bases in 64 million reads. Assemblies were performed for each sample and contigs greater than 200 bp in length were used for further analysis. The number of contigs for each sample varied between environments with ENV1 assemblies yielding the highest average number per sample (mean of 1,998 contigs covering 70% of reads), followed by ENV2 (mean of 659 covering 76% of reads) and ENV3 (mean of 431 contigs covering 70% of reads). This correlated with contig length differences between samples with mean contig lengths of 471, 424, and 324 bp for ENV1, ENV2, and ENV3 respectively. It should be noted that individual assemblies were limited by sequencing depth and that the higher number of contigs in ENV1 is likely due to reduced sample complexity. No-cell controls resulted in 7 – 20 contigs per sample, which covered less than 30% of reads. The genomic complexity of the indexed samples was first analyzed by plotting kernal density functions of GC composition. All ENV1 samples exhibited a single characteristic peak, consistent with targeted amplification of closely related donor genotypes (Figure 27A, Appendix I). By comparison, the GC content exhibited by ENV2 samples was a mixture of unimodal and multimodal curves consistent with targeted amplification of both single-cell genomes and mixtures of adhering cells (Figure 27A, Appendix J). Finally, ENV3 samples also exhibited multimodal curves and single spreading peaks consistent with amplification of multicellular aggregates (Figure 27A, Appendix K). The taxonomic structure of each sample was then determined using a tripartite binning approach. A stringent binning criteria was initially adopted based on 40 conserved phylogenomic markers mapped onto the tree of life using MLTreeMap 64  Figure 27: Summary of taxonomic profiles uncovered in metagenomes of 67 WGA samples originating from three distinct environments. A) Superimposed GC kernel density plot for all contigs generated from assemblies of individual metagenomic datasets. B) Hierarchical cluster analysis of sample-specific taxonomic profiles generated through a MEGAN analysis of blastx sequence comparisons against the RefSeq proteomic database. C) Taxonomic profiles of three environment-representative metagenomes, as generated through three distinct procedures (MLTreeMap, blastx against egg NOG, blastx against RefSeq proteomic). 65  (132). However, due to low sequencing depth only a handful of these markers were identified. To increase taxonomic resolution, the eggNOG (133) and NCBI ref_seq databases were queried using open reading frames (ORFs) predicted on contigs from each indexed sample. Results from the ref_seq search were then mapped onto the NCBI taxonomic hierarchy using Metagenome Analyzer (MEGAN) to define the most probable ancestor for each query sequence (134). Open reading frames assigned to taxonomic nodes by MEGAN were normalized by the fraction within each sample and hierarchically clustered, resulting in three distinct clusters for the ENV1, ENV2 and ENV3 samples. Branch lengths within each of the three clusters were consistent with increasing levels of genomic complexity with ENV1 samples exhibiting the least complexity followed by ENV3 and ENV2 (Figure 27B). The taxonomic origins of ORFs predicted in ENV1 samples were primarily affiliated with the genus Pseudoalteromonas within the Gammaproteobacteria (Figure 27C, Appendix L). Based on hierarchical clustering results two genotypic variants were resolved, consistent with the presence of closely related subpopulations within the enrichment culture. ORFs from ENV2 samples were dominated by known human oral microbiome constituents including Capnocytophaga and Flavobacterium within the Bacteroidetes, Corynebacterium, Rothia, Kocuria and Actinomyces within the Actinobacteria, Fusobacterium within the Fusobacteria, and Clostridium and Streptococcus within the Firmicutes (Figure 27C, Appendix M). Low-level representation of the candidate division TM7 was also observed. Different samples contained overlapping but not identical subsets of these taxonomic groups, with Streptococcus, Corynebacterium and Capnocytophaga being the most common overlapping taxa. Many of the taxonomic configurations observed in ENV2 samples have been previously described in the context of coaggregation and biofilm formation within the oral cavity (135-138), and several have been directly visualized using combinatorial labeling and spectral imaging techniques (139). ORFs from ENV3 samples were dominated by sulfate reducing bacteria (SRB) affiliated with Desulfatibacillum, Desulfobacterium and Desulfococcus within the Deltaproteobacteria (Figure 27C, Appendix N). Intermediate levels of representation were observed for unaffiliated Gammaproteobacteria, and Betaproteobacteria in addition to methanogenic archaea. Low-level representation of other taxa was observed in specific ENV3 samples, including ORFs affiliated with Alphaproteobacteria, Bacteroidetes, Firmicutes, Chloroflexi and Clostridia. Given the low sequence coverage for each sample and limited database representation of reference genomes for relevant sediment bacteria and archaea, it remains to be determined to what extent these 66  configurations represent known or novel modes of structural integration (140, 141). Further experiments to directly test these alternatives using more in-depth sequencing and hybridization approaches are ongoing. Details of sequencing data analysis are provided in Appendix H. Here, it has been demonstrated how phenotype-based sorting and droplet-based WGA followed by sequencing can be used to identify single microbes and members of microbial aggregates with a particular morphology. In the latter case, this ideally allows for the inspection of entire genomes of constituent members within aggregates, going beyond mere co-localization of small numbers of genes (5, 16), and enabling the analysis of metabolic pathways that can more precisely characterize potential symbiotic relationships within physical aggregates. The high representational bias of the PCR-based WGA protocol used for these experiments likely limited genomic coverage and thus did not permit such detailed genomic analysis. However, sequencing of amplified samples at greater depth is planned in hopes of acquiring broader coverage. Future experiments on environmental samples will use MDA-based WGA, which has been shown above to result in much lower bias and should thus permit higher genomic coverage with less sequencing effort. 4.2  PCR-based genotyping of single human tumour cell nuclei  Cellular heterogeneity is increasingly being shown to be a characteristic of human disease that has implications for both diagnosis and treatment. For example, in cancer, genetic analysis of different spatial regions within individual tumours have revealed branching patterns of tumour “evolution”, resulting in distinct subpopulations that can be grouped based on genetic aberrations such as genomic loci copy number variation, allelic imbalance, and mutations that are putative disease “drivers” (142, 143). This implies that specimens obtained from single biopsies may only reveal a subset of the aberrations of the whole tumour and may not identify those that are ubiquitous and thus important targets for therapy. It has also been shown that genetic clonal diversity can predict progression from a premalignant condition to a cancer (144), suggesting that increased diversity provides a wider base upon which natural selection can act to produce a tumour. In order to study cancer progression at higher spatial resolution, clonality can be analyzed at the single-cell level. Fluorescent in-situ hybridization (FISH) has been used to enumerate copy number variations of 8 genetic loci in individual cells, enabling inference of evolutionary trees 67  based on the frequencies of these variations (145). PCR-based WGA and sequencing of FACSsorted single-cell tumour nuclei was used to analyze loci copy number variation across the entire genome in 200 cells, derived from two separate cancers, to determine that the tumours progressed in “punctuated” clonal expansions that yielded distinct tumour subpopulations each distant from their root (128). Similarly, MDA-based WGA and sequencing of 25 single cells from a single tumour, isolated by manual micromanipulation, indicated the tumour likely did not result from mutations typical of that cancer and that, in contrast to the above study, there were no distinct clonal subpopulations (146). Collaborators at the British Columbia Cancer Research Centre have estimated clonal frequencies of somatic mutations in breast cancers by sequencing PCR amplicons from bulk DNA derived from tumours (147, 148). In order to more exactly determine the distribution of mutations within a tumour, however, the loci of interest must be amplified and sequenced in single tumour cells. This can be accomplished in the present microfluidic device by single-cell PCR-based genotyping as demonstrated in Section 4.1.2 on single bacteria, and work towards this goal has commenced. Ultimately, the genotyping of hundreds of single cells from primary lobular breast cancer pleural effusion and primary triple-negative breast cancers (TNBCs) used in the above studies (147, 148) will be performed. TNBCs are a cancer type defined by lack of oestrogen receptor, progesterone receptor, and ERBB2 gene amplification, and are thus of particular interest because they are unresponsive to the most successful currently available treatments for breast cancer, which target these receptors. As it is difficult to derive single-cell suspensions from solid tumour tissue, cell nuclei are extracted from the tumour samples (128). However, the resulting nuclei samples are highly heterogeneous in morphology, as the nuclei themselves can vary in size and the extraction process leaves a variety of cell debris in the sample (Figure 28).  The ability to perform  morphology-based sorting of nuclei using the present device is thus a significant advantage. As a first stringent test of genomic PCR on primary breast cancer pleural effusion cell nuclei, onchip qPCR targeting the RNase P gene, present at one copy per haploid genome (2 per cell), was performed on single nuclei (N=80), ~50 haploid genome copies of purified human gDNA (N=5), and suspension fluid containing no nuclei as determined by microscopy-based sorting (N=5). Micrographs of a cell nucleus in the cell-sorting module and in a stored droplet are shown in 68  Figure 28:  Micrographs of primary tumour cell nuclei showing the morphological  heterogeneity of the sample (provided by Jas Khattra). Figure 29. The 3-minute PCR hot start at 95C was used to lyse the nuclei. qPCR curves indicated that the target sequence was successfully amplified in 78 of 80 (98%) single nuclei, all 5 gDNA reactions, and 2 of 5 no-nuclei control reactions (Figure 30A). The latter result is most likely due to free gDNA from the nuclei sample in the suspension fluid. The difference in mean CT between reactions containing single nuclei (containing 2 gene copies) and 50 haploid genome copies (Figure 30B) was found to be 4.59 cycles, indicating an assay efficiency of 101.6%. This nearly ideal assay efficiency indicates that the gDNA within the nuclei is made accessible to PCR reagents by the protocol used, and, as in the PCR experiments on single bacteria, show that with an optimized assay and efficient lysis, robust PCR amplification can be achieved from single human cell nuclei. Having established that the gDNA of cell nuclei could be accessed for on-chip PCR, primer pairs targeting 6 genomic loci were then tested in multiplex qPCR reactions, including an intercalating dye for realtime reaction monitoring, on single nuclei (N=63) and no-nuclei controls (N=10). Five of the 6 loci contain somatic mutations of interest: FGA, GOLGA4, KIAA1468, KIF1C, and MORC1 (148) and the sixth locus was a multi-copy germline control NOTCH2NL. qPCR curves for all reactions are shown in Figure 31. While these qPCR curves provide some indication of amplification, it should be noted that they are less indicative of reaction progress than qPCR curves in single-plex PCR reactions due to interaction between primer pairs for different assays. 69  CT values were observed to be quite late for all single-nuclei reactions. Nevertheless, these values were used as a guide to select a subset of single-nuclei reactions for further analysis. Following on-chip 6-plex PCR, reaction products were eluted and off-chip single-plex PCR of each of the 5 somatic mutation loci was performed on the product of 7 of the on-chip single nuclei reactions with the lowest on-chip qPCR CT values as well as 2 of the no-nuclei control reactions. Each amplicon of these single-plex PCR reactions were then visualized by capillary electrophoresis.  Plots  for  4  of  the  single  nuclei  reactions  are  Figure 29: Micrographs of a primary breast cancer pleural effusion cell nucleus (A) in the cell-sorting module and (B) in a stored droplet.  Figure 30: RNase P qPCR of single sorted primary breast cancer pleural effusion cell nuclei (A) qPCR curves for all reactions. (B) Mean CT values and standard deviation for all reaction types. 70  Figure 31: qPCR curves for 6-plex PCR of single sorted primary breast cancer pleural effusion cell nuclei. shown in Figure 32. In total, 33 of 35 (94%) possible amplicons from the 7 single-nuclei reactions analyzed were successfully amplified as determined by the presence of a band with expected size. Bands were also seen for 3 amplicons in the no-nuclei controls. All 45 single-plex PCR amplicons (7 single nuclei and 2 no-nuclei controls with 5 loci each) were further analyzed by sequencing on an Ion Torrent PGM instrument. The amplicons from each nucleus and all amplicons from both no-nuclei controls were pooled and barcoded for sequencing. For comparison, 20 ng of bulk gDNA extracted from millions of cells was also subjected to the same protocol of multiplex PCR followed by single-plex PCR, but in conventional microlitre-volumes at a template concentration approximately 10 times greater than in on-chip single-nuclei reactions. Amplicon sequencing data binned by chromosome coverage from a representative single nucleus and bulk gDNA (Figure 33) indicates that the on-chip multiplex PCR amplifies target loci with similar representational bias to the microlitre-scale reaction performed on bulk gDNA. The higher coverage of chromosome 3 is due to the fact that two of the loci are located on that chromosome. The number of reads obtained from each amplicon, the fraction of reads matching mutations reported in (148), the means and coefficients of variation of these statistics for all single nuclei, and the mutational frequencies obtained from analysis of bulk DNA in (148) are shown in Table 3. 71  Figure 32: Capillary electrophoresis plots of PCR amplicons for 5 somatic mutation loci (left to right: FGA, GOLGA4, KIAA1468, KIF1C, and MORC1) from 4 on-chip singlenuclei multiplex PCR reactions. The mutational frequencies observed in single nuclei are relatively variable for all loci, with coefficients of variation above 0.4 with the exception of MORC1, suggesting a heterogeneous population. For the most part, frequencies are close to the expected theoretical ratios of 0, 0.5, and 1, corresponding to nuclei that are homozygous for a non-mutant variant, heterozygous, and homozygous for the mutation respectively. Departures from these ratios could be explained by loci copy number variations that result in more than 2 alleles (148). While the sample size of this dataset (N=7) is too small to effectively compare mean mutational frequencies of single nuclei with the bulk gDNA measurements from (148), these frequencies should theoretically converge as more single nuclei are analyzed. However, the mutational frequencies obtained from the repeated experiment on bulk gDNA also did not match the previously reported values 72  (Table 3). This variation could be explained by tumour heterogeneity which may have resulted in sampling genetically different cells from the same cancer (142), or the presence of noncancerous cells in the analysis performed in (148), which would artificially decrease mutational frequencies. Of note, is the presence of loci-specific reads in the pooled no-nuclei control reaction products in 4 of the 5 loci. The exact source of this contamination is unknown and its discovery would require careful dissection of the entire workflow.  However, systemic  Figure 33: Read coverage binned by chromosome for 5 loci amplicons from (A) purified genomic DNA and (B) on-chip multiplex PCR of a single primary breast cancer pleural effusion cell nucleus.  Table 3: Mutational frequencies obtained from sequencing reads for 7 single nuclei. 73  contamination of on-chip samples and reagents can be practically ruled out for loci in which the mutation frequency varies greatly among on-chip reactions, since one would expect such contamination to result in similar frequencies in all reactions. Work on scaling up both the number of targeted loci and the number of single nuclei, and on processing nuclei prepared from solid TNBC tumours is ongoing. The eventual workflow will likely consist of on-chip multiplex PCR on single nuclei followed by high-throughput singleplex PCR in a commercial microfluidic system (Access array, Fluidigm) in conjunction with primers that will include adapters for sequencing. This application illustrates how the single-cell genomic analyses, previously demonstrated on microbes, are equally applicable to eukaryotic cells. This work will allow for the exact determination of the clonal frequency of mutations in an unprecedented number of single cancer cells, which will ultimately enable the examination of clonal evolution with unparalleled resolution and scale. 4.3  Single-cell whole transcriptome amplification  While genetic aberrations typical of diseases such as cancer are sources of cellular heterogeneity, even cells of a healthy organism, which essentially share the same genome, clearly exhibit phenotypic diversity that allows for a plethora of physiological functions. These differences are due to cell-to-cell variations in the transcriptome, the set of all RNA molecules that comprises the functional output of the genome.  It is generally thought that persistent variation in  genetically identical cells is caused by the stochastic nature of gene expression, due to small copy numbers of genes, and the presence of multiple metastable transcriptional states (149-151). In order to fully understand the transcriptional mechanisms responsible for this cell-to-cell heterogeneity, vital to the determination of cell fate (152), or to identify minority cell populations based on transcriptional state (153), it is necessary to analyze the transcriptomes of single cells. The combination of new methods for the amplification of RNA quantities present in a single cell and the high throughput of modern sequencing instruments now offers the possibility of sequencing the entire transcriptome (RNA-seq) of many single cells (154-156). Importantly, sequencing of the transcriptome allows for the identification and discovery of posttranscriptional modifications to RNA molecules that may alter proteins coded by the genome, which may play a role in disease (148, 157).  74  As in whole genome amplification, the minimization of representational bias is crucial in whole transcriptome amplification (WTA) for RNA-seq in order to both minimize sequencing effort and allow for accurate measurement of the relative abundances of RNA molecules. In addition to the obvious advantage of lowered reagent costs, single-cell WTA in nanolitre-volumes may benefit from lowered representational bias relative to microlitre-volumes, as has been previously shown with single-cell multiple displacement amplification in Section In a final application of the present microfluidic device, work towards performing single-cell WTA is described here. As a first test of RNA quantification, quantitative reverse-transcription PCR (qRT-PCR) targeting GAPDH mRNA was performed on purified RNA derived from a human k562 cell line. qRT-PCR reactions were assembled in stored droplets containing 0.2 pg (N=32), 2 pg (N=15), 20 pg (N=9), 200 pg (N=9), and 0 pg (N=4) of purified RNA and RT-PCR master mix including primers and a hydrolysis probe for the detection of the GAPDH gene. The amount of total RNA present in a single mammalian cell is ~20 pg. Following reaction assembly, on-chip qRT-PCR was performed as previously described for 50 cycles, with fluorescent images of the droplet array being acquired at each cycle. Fluorescent images were analyzed using custom software in order to generate a qPCR curve for each stored droplet (Figure 34A) and CT values were extracted from these curves (Figure 34B). The PCR efficiency, determined by the slope of the  Figure 34: GAPDH qRT-PCR of dilutions of purified RNA (A) qRT-PCR curves for all reactions. (B) Mean CT values fitted to a line and standard deviation for all reactions.  75  fitted line through the CT vs. log10(template quantity) datapoints, is 98.6%, indicating that quantitative measurements of RNA abundance can be performed in the device. Next, a commercially available WTA protocol (Omniplex, Sigma) was tested on purified RNA. The multistep protocol consists of priming of all RNA by primers composed of random hexamers and a universal sequence, followed by reverse transcription which produces a library of cDNA fragments flanked by the universal sequence, and PCR amplification of the fragment library using universal primers. The protocol was tested on 0.2 pg (N=10), 2 pg (N=10), 20 pg (N=10), 200 pg (N=10), and 0 pg (N=5) of purified RNA derived from k562 cells in order to span the range of RNA quantities expected in a single mammalian cell. Following on-chip WTA, reaction products from all stored droplets were eluted and cDNA abundances were analyzed by qPCR. GAPDH cDNA, a common reference gene, was quantified by conventional microlitre-volume qPCR in the WTA product. qPCR curves on WTA product from each starting RNA quantity are shown in Figure 35A. qPCR curves for the 2, 20, and 200 pg WTA reaction products are tightly clustered, whereas those of the 0.2 pg WTA reaction products have large spread and have CT values comparable to those of the NTC WTA reactions. WTA on 0.2 pg of RNA was thus considered unreliable. Mean CT values and standard deviation of the 2, 20, and 200 pg WTA reaction products are plotted in Figure 35B. The exceptionally low standard deviation for all starting RNA quantities highlights the high reproducibility of both the on-chip WTA reactions and the elution process. The amplification efficiency, determined by the slope of the fitted line  Figure 35: Quantification of GAPDH cDNA in WTA product by qPCR. (A) qPCR curves for all reactions, (B) Mean CT values fitted to a line and standard deviation for all reactions. 76  through CT vs. log2(template quantity) datapoints, is 97.6%, indicating highly quantitative onchip WTA on RNA quantities spanning an order of magnitude above and below that expected in a single mammalian cell. In order to further assess WTA performance, a panel of 48 qPCR assays targeting endogenous control genes was applied to WTA product from each on-chip reaction using a commercial microfluidic qPCR device (48.48 Dynamic Array, Fluidigm) that allows for the application of up to 48 assays against 48 samples in nanolitre-volume reactions. A heatmap depicting CT values for all reactions is shown in Figure 36. For each gene, CT values for were plotted against log2(template quantity) and PCR efficiencies were calculated. These plots are shown in Figure 37 for genes which exhibited PCR efficiencies between 85% and 115%. The poor efficiencies of the remaining genes outside this range may be due to either poor assay efficiency or non-linear amplification of these genes by WTA over the range of RNA quantities tested. The efficiencies of these assays can only be determined by generating a standard curve from qPCR reactions on known quantities of target cDNA, and this work is ongoing. However, based on these results, it can be concluded that on-chip WTA is quantitative for at least the genes shown in Figure 37. To compare the WTA performance in microfluidic and conventional formats, WTA was also performed on 100 ng of purified RNA in microlitre-volume reactions and products were again quantified by qPCR in a Dynamic Array device. Abundances of the above 10 genes relative to the 18S rRNA gene, which was found to have the lowest CT of all genes quantified, were determined by comparison of mean CT values for all starting RNA quantities and calculated as 2ΔCT in both on-chip and conventional WTA reaction products (Figure 38). The similarity of gene abundances in the microfluidic and conventional formats offers further evidence that the on-chip WTA protocol is quantitative. A number of future experiments have been planned in order to further characterize and apply WTA in the present device. To determine the representational bias of WTA performed in nanolitre and microlitre-volumes, in order to determine whether or not the microfluidic format offers an advantage in this regard, the relative abundances of RNA species in the starting template must be known.  While this can be obtained by sequencing or qPCR, potential  differences in reverse transcription efficiency and bias may confound these measurements. The use of various RNA species mixed in known ratios as starting template, however, both obviates the need for such measurements and removes reverse transcription as a confounding variable. 77  Future WTA experiments using known input RNA abundances have thus been planned. To more accurately quantify cDNA abundances generated by WTA, sequencing of WTA product (155) will be performed, which has the added benefit of not requiring qPCR assay validation. The flexibility of programmable droplet-based reaction assembly will also be exploited to test other commercially available WTA protocols that use template-switching chemistries to amplify full-length RNA molecules and that use MDA-based cDNA amplification. Once all of the above have been performed to achieve and validate a microfluidic WTA protocol with lowrepresentational bias, it will be applied to single cells. Ultimately, this tool will be used for the transcriptional profiling of hundreds of cells in a biological system of interest, a currently intractable proposition in conventional formats, for such applications as the elucidation of transcriptional mechanisms responsible for stem cell differentiation and renewal or the discovery of post-transcriptional modifications that play a role in human disease.  Figure 36: Heatmap depicting CT values for 48 qPCR assays applied to WTA product.  78  Figure 37: Mean CT values from selected qPCR assays applied to on-chip WTA products.  79  Figure 38: Comparison of gene abundances relative to 18S rRNA in on-chip and in-tube WTA product.  80  Chapter 5 : Conclusions 5.1  Contributions to the state of the art  The technological contribution of the work described in this thesis is the development of a new type of microfluidic device with an unprecedented degree of fluid-handling functionality and programmability. This has been achieved through the development of novel capabilities as well as the integration of valve-based and droplet-based microfluidic fluid handling methods. The key novel technological advancement in this work, which enables programmable formulation in the device, is the development of a method for merging and immobilizing an arbitrary number of droplets. This work is the first to demonstrate this capability, which renders the formulation of reagent mixtures independent of the volume of the container. The flowcontrolled wetting mechanism by which this is accomplished is distinct from previously described techniques based on surface tension (29, 92, 94) or hydrodynamic trapping (95). It should be noted that, depending on the choice of surfactant and carrier phase, surface wetting might also play a role in other reported droplet storage designs, although this has not been previously recognized. In combination with valve-based fluid metering and single-cell sorting, this droplet merging technique allows for the precise formulation of solutions of multiple reagents and single cells in an array of nanolitre-volume chambers with user-defined proportions, timing, and volume.  This functionality forms the basis for programmable  microfluidic biological analysis, allowing a single device to perform numerous experiment types. It is the belief of the author that such programmability is ultimately necessary for the widespread adoption of microfluidic technology by the greater biological research community, just as programmable integrated circuits have facilitated the ubiquitous use of electron manipulation for computation by non-expert users. The combination of addressability afforded by valve-based flow control and the sample compartmentalization of droplet-based fluid handling also enables recovery of the contents of any selected storage chamber with negligible cross-contamination. This work is the first to demonstrate recovery of a single droplet selected out of many in a closed microfluidic device and demonstrates the first scalable technique for cross-contamination-free recovery of individual reaction products from any microfluidic device. This functionality is crucially important, as in many applications the products of reactions that can be performed at high-throughput in 81  microfluidic devices must ultimately be analyzed by conventional off-chip methods such as DNA sequencing. Valve-based flow control for the sorting of single cells and encapsulation of isolated cells in droplets allows for the high-throughput sorting, isolation, and transport of single cells selected from highly heterogeneous samples based on any phenotype discernible by microscopy. In combination with programmable formulation, this powerful ability can be used in the device to apply any reaction protocol to selected single cells that are otherwise difficult to isolate and transport. This work is the first to enable multiple experiment types on single selected cells in a microfluidic device.  The versatility of this cell-handling method is evidenced by its  demonstrated application to various cell types ranging from single microbes and microbial aggregates isolated from complex environmental samples to primary human cancer cell nuclei isolated from samples cluttered by cellular debris. The unique capabilities of this device have also allowed for the isolation of two phenotypically selected single cells into the same nanolitrevolume, which is the first such demonstration to the author’s knowledge. The functionality of the device has thus far primarily been applied to a variety of genetic analyses of single cells or cellular aggregates. To this end, the programmability of the device has been exploited to execute various molecular biology techniques, each requiring a different liquid-handling protocol, without the need to design, fabricate, and test a custom device for each one. This has undoubtedly resulted in timesavings on the order of months, and the completion of all of the applications presented in this thesis in the allotted time would have been impossible without this flexibility. The practical benefit of programmable microfluidic biological analysis has thus already been realized and it is the hope of the author that this flexibility will continue to be exploited in many diverse future applications. In the application of this device to various experiments, significant contributions to biological research have been made. The genetic analysis of microbial aggregates isolated from deep sea sediment samples, enabled by on-chip sorting and WGA, has led to the discovery of potentially novel microbial relationships relevant to important biogeochemical cycling of sulfates and methane. More in-depth sequencing to further investigate these relationships is ongoing. MDA performed in nanolitre-volume droplets has been shown to amplify the genomes of single cells with unprecedented uniformity, which has significant implications for this increasingly 82  important method for the understanding of cellular heterogeneity in all biological systems. Finally, the genotyping of single primary human breast cancer cells using this device has begun to elucidate the clonal structure of tumours at single-cell resolution. Results from this study will enable the reconstruction of mutational evolution of cancer and will experimentally test predictions of clonal mutation frequencies based on computational models (147, 148). The low cost at which this analysis can be performed in the present device will enable such single-cell genomic analyses of human tissue on a scale that is impractical by conventional means. 5.2  Limitations and recommendations for technical improvements  The programmable functionality of this device necessitates serial formulation and elution of each reaction. This results in much slower assembly of multiple reactions relative to microfluidic devices that perform reactions in parallel (15) and high-speed droplet-based systems that do not rely on peristaltic pumps and valves for reagent metering and addition (31, 34), which places a practical limit on throughput. Increased throughput can be achieved by simply adding additional storage chambers to the array, which can be easily scaled up while maintaining addressability due to the factorial row multiplexer design (85). Due to the serial nature of the device, however, a significant increase in the number of storage chambers would require a commensurate increase in droplet transport velocity to maintain reasonable device operation speeds. Flow-controlled wetting, though, is only able to reliably merge incoming droplets, containing an aqueous surfactant, traveling at a velocity less than the critical incoming velocity. Thus, to increase the throughput of the current device design, this critical incoming velocity must be increased, which is achievable by reducing the fluidic resistance of the side and bypass channels that decelerate incoming droplets. An alternative storage chamber design has also been considered. A chamber with the shape of a dome would cause all incoming droplets to come to rest and merge in the centre of the chamber ceiling regardless of incoming velocity, provided that droplets do not wet the ceiling at their first point of contact, which may not be in the centre, and a carrier phase denser than water is used. In contrast to the current design, for this droplet merging mechanism, wetting would ideally be prevented for all conditions, and device surface treatments or a modified carrier phase might be necessary to ensure this. Such a geometry would eliminate the need for the side and bypass channels in the current design and would thus have a much smaller footprint and allow for higher 83  chamber density. This geometry was proposed at early stages of the device design but its microfabrication was considered too complex. Recent efforts, however, have proven successful (personal communication with Hans Zahn) and testing of these chambers is currently in progress. While peristaltic pumping has been shown to be a precise method of metering aqueous reagents for delivery to the storage chamber array, for larger volumes this method can be timeconsuming. To dispense reagents more quickly, a stream of aqueous phase could be formed at a T-junction at the reagent-metering module (as currently done during elution), where the total volume dispensed is controlled by the time that the aqueous phase is allowed to flow out of a reagent inlet, which can be controlled by gating of a valve. However, as the volume dispensed in this scheme may not be linearly dependent on this time, especially for short periods, and is dependent on capillary number (105), additional experiments would need to be performed to determine the precision and robustness of this metering method. In all of the single-cell experiments described in this thesis, cell sorting has been performed by manual microscopic inspection of samples. This process is fairly labor-intensive and would be impractical for a significant increase in the number of storage chambers. Real-time software image analysis of micrographs of the cell-sorting channel junction could be used to automate the cell-sorting process. For samples where a significant amount of non-biological or cellular debris is present, automated image analysis alone may be ineffective at robustly identifying target cells. Additional fluorescent staining or fluorescent in-situ hybridization probes that specifically label target cells could be used to improve this. FACS-sorting of samples prior to on-chip processing could also be used to ensure that a high percentage of the particles in the sample are cells of interest. This would significantly simplify cell-sorting by automated image analysis. Serial elution of each individual chamber would also be impractical for a significantly larger number of chambers. This could be addressed by using alternative device architectures that allow for parallel elution of multiple chambers at once. For example, this could be achieved through the use of a bifurcating channel network that allows for a single stream of buffer to be evenly divided into multiple streams that are used to elute chambers in the same column but multiple rows at once. Parallel deposition of eluted products into multiple microfuge tubes could be achieved through the use of multiple elution nozzles with a pitch that matches that of the tubes. 84  While this device is capable of executing multistep reaction protocols in which reagents are consecutively added, it lacks the ability to perform purification-type procedures involving the retention of one analyte while others are flushed away. This could be addressed by the use of functionalized beads to which selected analytes could be bound. Gratings, sized to prevent beads from passing through them, could be built into the storage chamber exit, or a magnetic field could be used to keep magnetic beads in the storage chamber while flushing all other contents out by overfilling it, as is done during elution. 5.3  Future applications  The versatility and novel fluid-handling capabilities of the microfluidic device presented in this work makes it well suited to a number of important applications that have the potential to make significant contributions in many diverse fields of biological research. The ability to rapidly apply different protocols to sorted single cells of various types will continue to be exploited. The exceptionally low representational bias of single-cell MDA reactions performed in nanolitre-volume droplets is perhaps the most exciting biological result presented here and, combined with the ever increasing throughput of DNA sequencing technology (158), offers the possibility of fully exploring the genomes of large numbers of single cells at an unmatched pergenome cost. This has significant implications for the study of biological systems in which cellto-cell heterogeneity plays a crucial role.  Microfluidic MDA of single microbes from  environmental samples will be optimized with the goal of shedding light on the immense dearth in the catalogue of known microbial genomes. As the role of microbial diversity and metabolism in human health becomes more clear (159), genomic analysis of constituents of the human microbiome may be an important future medical application. Although the genotyping of single cancer cell nuclei presented here is based on PCR amplification of known mutation sites, WGA would enable genome-wide analysis and will thus be tested. Building on the positive results obtained from WTA of single-cell equivalent quantities of purified RNA, WTA of single cells will be optimized and applied to systems where knowledge of the entire transcriptional state in single cells stands to yield new biological insight. Considering the low representational bias seen in on-chip single-cell MDA, MDA-based WTA protocols may prove to be equally effective at uniformly amplifying RNA.  Successful  85  demonstration of both single-cell WGA and WTA would beg the combination of these two capabilities to enable the analysis of both the genome and transcriptome of a single cell. The ability to place multiple selected single cells in the same nanolitre-volume was exploited in order to seed single-cell bacterial co-cultures. However, this capability also offers opportunities for studying more interesting intercellular interactions, such as host-pathogen or predator-prey dynamics, at the single-cell level. Beyond single-cell analysis, which is the focus of the work described here, this device could be useful in other applications that more fully exploit the ability to vary the contents of each storage chamber such as enzyme characterization and the optimization of molecular biology protocols. While the current design does not have the necessary throughput for some screening applications, scale up of the device to an array of 1000 addressable chambers would be easily achievable and the reduced complexity of row-column sample tracking relative to the optical encoding schemes, required by more conventional serial droplet-based systems, and the ability to recover selected chamber contents may be seen as acceptable trade-offs for reduction in throughput in some cases.  86  References 1.  Whitesides GM (2006) The origins and the future of microfluidics. Nature 442:368-373.  2.  El-Ali J, Sorger PK, & Jensen KF (2006) Cells on chips. Nature 442:403-411.  3.  deMello AJ (2006) Control and detection of chemical reactions in microfluidic systems. Nature 442:394-402.  4.  Tewhey R, et al. (2009) Microdroplet-based PCR enrichment for large-scale targeted sequencing. Nat Biotech 27:1025-1031.  5.  Ottesen EA, Hong JW, Quake SR, & Leadbetter JR (2006) Microfluidic Digital PCR Enables Multigene Analysis of Individual Environmental Bacteria. Science 314:14641467.  6.  Blow N (2009) Microfluidics: the great divide. Nat Meth 6:683-686.  7.  Xia YN & Whitesides GM (1998) Soft lithography. Annu. Rev. Mater. Sci. 28:153-184.  8.  Unger MA, Chou H-P, Thorsen T, Scherer A, & Quake SR (2000) Monolithic Microfabricated Valves and Pumps by Multilayer Soft Lithography. Science 288:113116.  9.  Thorsen T, Maerkl SJ, & Quake SR (2002) Microfluidic Large-Scale Integration. Science 298:580-584.  10.  Chou H-P, Unger MA, & Quake SR (2001) A Microfabricated Rotary Pump. Biomed. Microdevices 3:323-330.  11.  Hansen CL, Skordalakes E, Berger JM, & Quake SR (2002) A robust and scalable microfluidic metering method that allows protein crystal growth by free interface diffusion. Proc. Natl. Acad. Sci. USA 99:16531-16536.  12.  Maerkl SJ & Quake SR (2007) A Systems Approach to Measuring the Binding Energy Landscapes of Transcription Factors. Science 315:233-237.  13.  Singhal A, Haynes CA, & Hansen CL (2010) Microfluidic Measurement of AntibodyAntigen Binding Kinetics from Low-Abundance Samples and Single Cells. Anal. Chem. 82:8671-8679.  14.  Marcus JS, Anderson WF, & Quake SR (2006) Microfluidic Single-Cell mRNA Isolation and Analysis. Anal. Chem. 78:3084-3089.  15.  White AK, et al. (2011) High-throughput microfluidic single-cell RT-qPCR. Proc. Natl. Acad. Sci. USA 108:13999-14004. 87  16.  Tadmor AD, Ottesen EA, Leadbetter JR, & Phillips R (2011) Probing Individual Environmental Bacteria for Viruses by Using Microfluidic Digital PCR. Science 333:5862.  17.  Marcy Y, et al. (2007) Dissecting biological "dark matter" with single-cell genetic analysis of rare and uncultivated TM7 microbes from the human mouth. Proc. Natl. Acad. Sci. USA 104:11889-11894.  18.  Marcy Y, et al. (2007) Nanoliter reactors improve multiple displacement amplification of genomes from single cells. PLoS Genet. 3:1702-1708.  19.  Fan HC, Wang J, Potanina A, & Quake SR (2011) Whole-genome molecular haplotyping of single cells. Nat Biotech 29:51-57.  20.  Blainey PC, Mosier AC, Potanina A, Francis CA, & Quake SR (2011) Genome of a Low-Salinity  Ammonia-Oxidizing  Archaeon  Determined  by  Single-Cell  and  Metagenomic Analysis. PLoS ONE 6:e16626. 21.  Fan HC & Quake SR (2007) Detection of Aneuploidy with Digital Polymerase Chain Reaction. Anal. Chem. 79:7576-7579.  22.  Blainey PC & Quake SR (2011) Digital MDA for enumeration of total nucleic acid contamination. Nucleic Acids Res. 39:e19.  23.  Thorsen T, Roberts RW, Arnold FH, & Quake SR (2001) Dynamic Pattern Formation in a Vesicle-Generating Microfluidic Device. Phys. Rev. Lett. 86:4163-4166.  24.  Anna SL, Bontoux N, & Stone HA (2003) Formation of dispersions using ``flow focusing'' in microchannels. Applied Physics Letters 82:364-366.  25.  Roach LS, Song H, & Ismagilov RF (2005) Controlling Nonspecific Protein Adsorption in a Plug-Based Microfluidic System by Controlling Interfacial Chemistry Using Fluorous-Phase Surfactants. Anal. Chem. 77:785-796.  26.  Holtze C, et al. (2008) Biocompatible surfactants for water-in-fluorocarbon emulsions. Lab Chip 8:1632-1639.  27.  Song H & Ismagilov RF (2003) Millisecond Kinetics on a Microfluidic Chip Using Nanoliters of Reagents. J. Am. Chem. Soc. 125:14613-14619.  28.  Tan Y-C, Fisher JS, Lee AI, Cristini V, & Lee AP (2004) Design of microfluidic channel geometries for the control of droplet volume, chemical concentration, and sorting. Lab Chip 4:292-298.  88  29.  Niu X, Gulati S, Edel JB, & deMello AJ (2008) Pillar-induced droplet merging in microfluidic circuits. Lab Chip 8:1837-1841.  30.  Fidalgo LM, Abell C, & Huck WTS (2007) Surface-induced droplet fusion in microfluidic devices. Lab Chip 7:984-986.  31.  Ahn K, Agresti J, Chong H, Marquez M, & Weitz DA (2006) Electrocoalescence of drops synchronized by size-dependent flow in microfluidic channels. Applied Physics Letters 88:264105-264103.  32.  Link DR, et al. (2006) Electric Control of Droplets in Microfluidic Devices. Angewandte Chemie International Edition 45:2556-2560.  33.  Baroud CN, Robert de Saint Vincent M, & Delville J-P (2007) An optical toolbox for total control of droplet microfluidics. Lab Chip 7:1029-1033.  34.  Abate AR, Hung T, Mary P, Agresti JJ, & Weitz DA (2010) High-throughput injection with microfluidics using picoinjectors. Proc. Natl. Acad. Sci. USA 107:19163-19166.  35.  Song H, Tice JD, & Ismagilov RF (2003) A Microfluidic System for Controlling Reaction Networks in Time. Angewandte Chemie International Edition 42:768-772.  36.  Ahn K, et al. (2006) Dielectrophoretic manipulation of drops for high-speed microfluidic sorting devices. Applied Physics Letters 88:024104.  37.  Franke T, Abate AR, Weitz DA, & Wixforth A (2009) Surface acoustic wave (SAW) directed droplet flow in microfluidics for PDMS devices. Lab Chip 9:2625-2627.  38.  Abate AR, Agresti JJ, & Weitz DA (2010) Microfluidic sorting with high-speed singlelayer membrane valves. Applied Physics Letters 96:203509-203503.  39.  Beer NR, et al. (2007) On-Chip, Real-Time, Single-Copy Polymerase Chain Reaction in Picoliter Droplets. Anal. Chem. 79:8471-8475.  40.  Beer NR, et al. (2008) On-Chip Single-Copy Real-Time Reverse-Transcription PCR in Isolated Picoliter Droplets. Anal. Chem. 80:1854-1858.  41.  Kiss MM, et al. (2008) High-Throughput Quantitative Polymerase Chain Reaction in Picoliter Droplets. Anal. Chem. 80:8975-8981.  42.  Kumaresan P, Yang CJ, Cronier SA, Blazej RG, & Mathies RA (2008) High-Throughput Single Copy DNA Amplification and Cell Analysis in Engineered Nanoliter Droplets. Anal. Chem. 80:3522-3529.  89  43.  Mazutis L, et al. (2009) Droplet-Based Microfluidic Systems for High-Throughput Single DNA Molecule Isothermal Amplification and Analysis. Anal. Chem. 81:48134821.  44.  Zhong Q, et al. (2011) Multiplex digital PCR: breaking the one target per color barrier of quantitative PCR. Lab Chip.  45.  Zeng Y, Novak R, Shuga J, Smith MT, & Mathies RA (2010) High-Performance Single Cell Genetic Analysis Using Microfluidic Emulsion Generator Arrays. Anal. Chem. 82:3183-3190.  46.  Zheng B, Roach LS, & Ismagilov RF (2003) Screening of Protein Crystallization Conditions on a Microfluidic Chip Using Nanoliter-Size Droplets. J. Am. Chem. Soc. 125:11170-11171.  47.  Boedicker JQ, Li L, Kline TR, & Ismagilov RF (2008) Detecting bacteria and determining their susceptibility to antibiotics by stochastic confinement in nanoliter droplets using plug-based microfluidics. Lab Chip 8:1265-1272.  48.  Brouzes E, et al. (2009) Droplet microfluidic technology for single-cell high-throughput screening. Proc. Natl. Acad. Sci. USA 106:14195-14200.  49.  Miller OJ, et al. (2012) High-resolution dose-response screening using droplet-based microfluidics. Proc. Natl. Acad. Sci. USA 109:378-383.  50.  Agresti JJ, et al. (2010) Ultrahigh-throughput screening in drop-based microfluidics for directed evolution. Proc. Natl. Acad. Sci. USA 107:4004-4009.  51.  Fallah-Araghi A, Baret J-C, Ryckelynck M, & Griffiths AD (2012) A completely in vitro ultrahigh-throughput droplet-based microfluidic screening system for protein engineering and directed evolution. Lab Chip 12:882-891.  52.  Pollack MG, Fair RB, & Shenderov AD (2000) Electrowetting-based actuation of liquid droplets for microfluidic applications. Applied Physics Letters 77:1725-1726.  53.  Chatterjee D, Hetayothin B, Wheeler AR, King DJ, & Garrell RL (2006) Droplet-based microfluidics with nonaqueous solvents and solutions. Lab Chip 6:199-206.  54.  Fair R (2007) Digital microfluidics: is a true lab-on-a-chip possible? Microfluid. Nanofluid. 3:245-281.  55.  Luk VN & Wheeler AR (2009) A Digital Microfluidic Approach to Proteomic Sample Processing. Anal. Chem. 81:4524-4530.  90  56.  Shih SCC, et al. (2012) Dried Blood Spot Analysis by Digital Microfluidics Coupled to Nanoelectrospray Ionization Mass Spectrometry. Anal. Chem. 84:3731-3738.  57.  Mousa NA, et al. (2009) Droplet-Scale Estrogen Assays in Breast Tissue, Blood, and Serum. Science Translational Medicine 1:1ra2-1ra2.  58.  Grover WH, Skelley AM, Liu CN, Lagally ET, & Mathies RA (2003) Monolithic membrane valves and diaphragm pumps for practical large-scale integration into glass microfluidic devices. Sensors Actuators B: Chem. 89:315-323.  59.  Jensen EC, Bhat BP, & Mathies RA (2010) A digital microfluidic platform for the automation of quantitative biomolecular assays. Lab Chip 10:685-691.  60.  Fidalgo LM & Maerkl SJ (2011) A software-programmable microfluidic device for automated biology. Lab Chip 11:1612-1619.  61.  Urbanski JP, Thies W, Rhodes C, Amarasinghe S, & Thorsen T (2006) Digital microfluidics using soft lithography. Lab Chip 6:96-104.  62.  Lau BTC, Baitz CA, Dong XP, & Hansen CL (2007) A Complete Microfluidic Screening Platform for Rational Protein Crystallization. J. Am. Chem. Soc. 129:454-455.  63.  Rappe MS & Giovannoni SJ (2003) THE UNCULTURED MICROBIAL MAJORITY. Annu. Rev. Microbiol. 57:369-394.  64.  Ishoey T, Woyke T, Stepanauskas R, Novotny M, & Lasken RS (2008) Genomic sequencing of single microbial cells from environmental samples. Curr. Opin. Microbiol. 11:198-204.  65.  McClellan J & King M-C (2010) Genetic Heterogeneity in Human Disease. Cell 141:210-217.  66.  Zhang K, et al. (2006) Sequencing genomes from single cells by polymerase cloning. Nat. Biotechnol. 24:680-686.  67.  Zare RN & Kim S (2010) Microfluidic Platforms for Single-Cell Analysis. Annu. Rev. Biomed. Eng. 12:187-201.  68.  Lecault V, White AK, Singhal A, & Hansen CL (2012) Microfluidic single cell analysis: from promise to practice. Curr. Opin. Chem. Biol.  69.  He M, et al. (2005) Selective Encapsulation of Single Cells and Subcellular Organelles into Picoliter- and Femtoliter-Volume Droplets. Anal. Chem. 77:1539-1544.  91  70.  Clausell-Tormos J, et al. (2008) Droplet-Based Microfluidic Platforms for the Encapsulation and Screening of Mammalian Cells and Multicellular Organisms. Chem. Biol. 15:427-437.  71.  Huebner A, et al. (2008) Development of Quantitative Cell-Based Enzyme Assays in Microdroplets. Anal. Chem. 80:3890-3896.  72.  Di Carlo D, Aghdam N, & Lee LP (2006) Single-Cell Enzyme Concentrations, Kinetics, and Inhibition Analysis Using High-Density Hydrodynamic Cell Isolation Arrays. Anal. Chem. 78:4925-4930.  73.  Kobel S, Valero A, Latt J, Renaud P, & Lutolf M (2010) Optimization of microfluidic single cell trapping for long-term on-chip culture. Lab Chip 10:857-863.  74.  Faley SL, et al. (2009) Microfluidic single cell arrays to interrogate signalling dynamics of individual, patient-derived hematopoietic stem cells. Lab Chip 9:2659-2664.  75.  Edd JF, et al. (2008) Controlled encapsulation of single-cells into monodisperse picolitre drops. Lab Chip 8:1262-1264.  76.  Felton MJ (2003) Product Review: Liquid handling: Dispensing reliability. Anal. Chem. 75:397 A-399 A.  77.  Taylor G (1953) Dispersion of Soluble Matter in Solvent Flowing Slowly through a Tube. Proceedings of the Royal Society of London. Series A. Mathematical and Physical Sciences 219:186-203.  78.  Tice JD, Song H, Lyon AD, & Ismagilov RF (2003) Formation of Droplets and Mixing in Multiphase Microfluidics at Low Values of the Reynolds and the Capillary Numbers. Langmuir 19:9127-9133.  79.  Adamson DN, Mustafi D, Zhang JXJ, Zheng B, & Ismagilov RF (2006) Production of arrays of chemically distinct nanolitre plugs via repeated splitting in microfluidic devices. Lab Chip 6:1178-1186.  80.  Dangla R, Gallaire F, & Baroud CN (2010) Microchannel deformations due to solventinduced PDMS swelling. Lab Chip 10:2972-2978.  81.  Lee  JN,  Park  C,  &  Whitesides  GM  (2003)  Solvent  Compatibility  of  Poly(dimethylsiloxane)-Based Microfluidic Devices. Anal. Chem. 75:6544-6554. 82.  Riess JG (2001) Oxygen Carriers ("Blood Substitutes") Raison d'Etre, Chemistry, and Some Physiology Blut ist ein ganz besondrer Saft. Chemical Reviews 101:2797-2920.  92  83.  Koster S, et al. (2008) Drop-based microfluidic devices for encapsulation of single cells. Lab Chip 8:1110-1115.  84.  Zheng B, Tice JD, & Ismagilov RF (2004) Formation of Droplets of Alternating Composition in Microfluidic Channels and Applications to Indexing of Concentrations in Droplet-Based Assays. Anal. Chem. 76:4977-4982.  85.  Hua Z, et al. (2006) A versatile microreactor platform featuring a chemical-resistant microvalve array for addressable multiplex syntheses and assays. J. Micromech. Microeng. 16:1433-1443.  86.  Hansen CL, Sommer MOA, & Quake SR (2004) Systematic investigation of protein phase behavior with a microfluidic formulator. Proc. Natl. Acad. Sci. USA 101:1443114436.  87.  Tice JD, Lyon AD, & Ismagilov RF (2004) Effects of viscosity on droplet formation and mixing in microfluidic channels. Anal. Chim. Acta 507:73-77.  88.  Abate AR, et al. (2009) Impact of inlet channel geometry on microfluidic drop formation. Physical Review E 80:026310.  89.  Sung Hwan C, Godin J, & Yu-Hwa L (2009) Optofluidic Waveguides in Teflon AFCoated PDMS Microfluidic Channels. Photonics Technology Letters, IEEE 21:10571059.  90.  Reiter G, Khanna R, & Sharma A (2000) Enhanced Instability in Thin Liquid Films by Improved Compatibility. Phys. Rev. Lett. 85:1432-1435.  91.  Shestopalov I, Tice JD, & Ismagilov RF (2004) Multi-step synthesis of nanoparticles performed on millisecond time scale in a microfluidic droplet-based system. Lab Chip 4:316-321.  92.  Shim J-u, et al. (2007) Control and Measurement of the Phase Behavior of Aqueous Solutions Using Microfluidics. J. Am. Chem. Soc. 129:8825-8835.  93.  Kline TR, Runyon MK, Pothiawala M, & Ismagilov RF (2008) ABO, D Blood Typing and Subtyping Using Plug-Based Microfluidics. Anal. Chem. 80:6190-6197.  94.  Schmitz CHJ, Rowat AC, Koster S, & Weitz DA (2009) Dropspots: a picoliter array in a microfluidic device. Lab Chip 9:44-49.  95.  Huebner A, et al. (2009) Static microdroplet arrays: a microfluidic device for droplet trapping, incubation and release for enzymatic and cell-based assays. Lab Chip 9:692698. 93  96.  Boukellal H, Selimovic S, Jia Y, Cristobal G, & Fraden S (2009) Simple, robust storage of drops and fluids in a microfluidic device. Lab Chip 9:331-338.  97.  Wang W, Yang C, & Li CM (2009) On-demand microfluidic droplet trapping and fusion for on-chip static droplet assays. Lab Chip 9:1504-1506.  98.  Bretherton FP (1961) The motion of long bubbles in tubes. J. Fluid Mech. 10:166-188.  99.  Baldessari F, Homsy GM, & Leal LG (2007) Linear stability of a draining film squeezed between two approaching droplets. J. Colloid Interface Sci. 307:188-202.  100.  Steinhaus B, Spicer PT, & Shen AQ (2006) Droplet Size Effects on Film Drainage between Droplet and Substrate. Langmuir 22:5308-5313.  101.  Greenspan HP (1978) On the motion of a small viscous droplet that wets a surface. J. Fluid Mech. 84:125-143.  102.  Baroud CN, Gallaire F, & Dangla R (2010) Dynamics of microfluidic droplets. Lab Chip 10:2032-2045.  103.  Liu J, Hansen C, & Quake SR (2003) Solving the "World-to-Chip" Interface Problem with a Microfluidic Matrix. Anal. Chem. 75:4718-4723.  104.  Fidalgo LM, et al. (2008) From Microdroplets to Microfluidics: Selective Emulsion Separation in Microfluidic Devices. Angewandte Chemie International Edition 47:20422045.  105.  Utada AS, Fernandez-Nieves A, Stone HA, & Weitz DA (2007) Dripping to Jetting Transitions in Coflowing Liquid Streams. Phys. Rev. Lett. 99:094502.  106.  Graham DE & Phillips MC (1979) Proteins at liquid interfaces: I. Kinetics of adsorption and surface denaturation. J. Colloid Interface Sci. 70:403-414.  107.  Anderson JM, et al. (1996) Protein adsorption and macrophage activation on polydimethylsiloxane and silicone rubber. Journal of Biomaterials Science, Polymer Edition 7:159-169.  108.  Boxshall K, et al. (2006) Simple surface treatments to modify protein adsorption and cell attachment properties within a poly(dimethylsiloxane) micro-bioreactor. Surface and Interface Analysis 38:198-201.  109.  Whitman WB, Coleman DC, & Wiebe WJ (1998) Prokaryotes: The unseen majority. Proc. Natl. Acad. Sci. USA 95:6578-6583.  110.  Falkowski PG, Fenchel T, & Delong EF (2008) The Microbial Engines That Drive Earth's Biogeochemical Cycles. Science 320:1034-1039. 94  111.  Cho I & Blaser MJ (2012) The human microbiome: at the interface of health and disease. Nat Rev Genet 13:260-270.  112.  Amann RI, Ludwig W, & Schleifer KH (1995) Phylogenetic identification and in situ detection of individual microbial cells without cultivation. Microbiological Reviews 59:143-169.  113.  Riesenfeld CS, Schloss PD, & Handelsman J (2004) METAGENOMICS: Genomic Analysis of Microbial Communities. Annu. Rev. Genet. 38:525-552.  114.  Venter JC, et al. (2004) Environmental Genome Shotgun Sequencing of the Sargasso Sea. Science 304:66-74.  115.  Boedicker JQ, Vincent ME, & Ismagilov RF (2009) Microfluidic Confinement of Single Cells of Bacteria in Small Volumes Initiates High-Density Behavior of Quorum Sensing and Growth and Reveals Its Variability. Angewandte Chemie International Edition 48:5908-5911.  116.  Lee C, Lee S, Shin S, & Hwang S (2007) Real-time PCR determination of rRNA gene copy number: absolute and relative quantification assays with Escherichia coli. Appl. Microbiol. Biotechnol. 78:371-376.  117.  Luna RA, et al. (2007) DNA Pyrosequencing-Based Bacterial Pathogen Identification in a Pediatric Hospital Setting. J. Clin. Microbiol. 45:2985-2992.  118.  Chivian D, et al. (2008) Environmental Genomics Reveals a Single-Species Ecosystem Deep Within Earth. Science 322:275-278.  119.  Gribble S, Ng B, Prigmore E, Burford D, & Carter N (2004) Chromosome paints from single copies of chromosomes. Chromosome Research 12:143-151.  120.  Hornak M, et al. (2011) Frequency of Aneuploidy Related to Age in Porcine Oocytes. PLoS ONE 6:e18892.  121.  Dean FB, Nelson JR, Giesler TL, & Lasken RS (2001) Rapid amplification of plasmid and phage DNA using phi29 DNA polymerase and multiply-primed rolling circle amplification. Genome Res. 11:1095-1099.  122.  Woyke T, et al. (2010) One Bacterial Cell, One Complete Genome. PLoS ONE 5:e10314.  123.  Stepanauskas R & Sieracki ME (2007) Matching phylogeny and metabolism in the uncultured marine bacteria, one cell at a time. Proc. Natl. Acad. Sci. USA 104:9052-9057.  124.  Rodrigue S, et al. (2009) Whole Genome Amplification and De novo Assembly of Single Bacterial Cells. Plos One 4:e6864. 95  125.  Woyke T, et al. (2009) Assembling the Marine Metagenome, One Cell at a Time. Plos One 4:e5299.  126.  Swan BK, et al. (2011) Potential for Chemolithoautotrophy Among Ubiquitous Bacteria Lineages in the Dark Ocean. Science 333:1296-1300.  127.  Yoon HS, et al. (2011) Single-Cell Genomics Reveals Organismal Interactions in Uncultivated Marine Protists. Science 332:714-717.  128.  Navin N, et al. (2011) Tumour evolution inferred by single-cell sequencing. Nature 472:90-94.  129.  Dean FB, et al. (2002) Comprehensive human genome amplification using multiple displacement amplification. Proc. Natl. Acad. Sci. USA 99:5261-5266.  130.  Raghunathan A, et al. (2005) Genomic DNA amplification from a single bacterium. Appl. Environ. Microbiol. 71:3342-3347.  131.  Woyke T, et al. (2011) Decontamination of MDA Reagents for Single Cell Whole Genome Amplification. PLoS ONE 6:e26161.  132.  Stark M, Berger S, Stamatakis A, & von Mering C (2010) MLTreeMap - accurate Maximum Likelihood placement of environmental DNA sequences into taxonomic and functional reference phylogenies. BMC Genomics 11:461.  133.  Powell S, et al. (2011) eggNOG v3.0: orthologous groups covering 1133 organisms at 41 different taxonomic ranges. Nucleic Acids Res.  134.  Huson DH, Mitra S, Ruscheweyh H-J, Weber N, & Schuster SC (2011) Integrative analysis of environmental sequences using MEGAN4. Genome Res. 21:1552-1560.  135.  Lens PNL (2003) Biofilms in medicine, industry and environmental biotechnology: characteristics, analysis and control (IWA Publishing).  136.  Kolenbrander PE (1988) Intergeneric Coaggregation Among Human Oral Bacteria and Ecology of Dental Plaque. Annu. Rev. Microbiol. 42:627-656.  137.  Kolenbrander PE (1989) Surface Recognition among Oral Bacteria: Multigeneric Coaggregations and their Mediators. Crit. Rev. Microbiol. 17:137-159.  138.  Lancy P, Jr., Appelbaum B, Holt SC, & Rosan B (1980) Quantitative In Vitro Assay for "Corncob" Formation. Infect. Immun. 29:663-670.  139.  Valm AM, et al. (2011) Systems-level analysis of microbial community organization through combinatorial labeling and spectral imaging. Proc. Natl. Acad. Sci. USA.  96  140.  Pernthaler A, et al. (2008) Diverse syntrophic partnerships from deep-sea methane vents revealed by direct cell capture and metagenomics. Proc. Natl. Acad. Sci. USA 105:70527057.  141.  Victoria J O (2009) Methods for unveiling cryptic microbial partnerships in nature. Curr. Opin. Microbiol. 12:231-237.  142.  Gerlinger M, et al. (2012) Intratumor Heterogeneity and Branched Evolution Revealed by Multiregion Sequencing. New Engl. J. Med. 366:883-892.  143.  Navin N, et al. (2010) Inferring tumor progression from genomic heterogeneity. Genome Res. 20:68-80.  144.  Maley CC, et al. (2006) Genetic clonal diversity predicts progression to esophageal adenocarcinoma. Nat Genet 38:468-473.  145.  Anderson K, et al. (2011) Genetic variegation of clonal architecture and propagating cells in leukaemia. Nature 469:356-361.  146.  Xu X, et al. (2012) Single-Cell Exome Sequencing Reveals Single-Nucleotide Mutation Characteristics of a Kidney Tumor. Cell 148:886-895.  147.  Shah SP, et al. (2012) The clonal and mutational evolution spectrum of primary triplenegative breast cancers. Nature advance online publication.  148.  Shah SP, et al. (2009) Mutational evolution in a lobular breast tumour profiled at single nucleotide resolution. Nature 461:809-813.  149.  Losick R & Desplan C (2008) Stochasticity and Cell Fate. Science 320:65-68.  150.  Chang HH, Hemberg M, Barahona M, Ingber DE, & Huang S (2008) Transcriptomewide noise controls lineage choice in mammalian progenitor cells. Nature 453:544-547.  151.  Choi PJ, Cai L, Frieda K, & Xie XS (2008) A Stochastic Single-Molecule Event Triggers Phenotype Switching of a Bacterial Cell. Science 322:442-446.  152.  Yamanaka S & Blau HM (2010) Nuclear reprogramming to a pluripotent state by three approaches. Nature 465:704-712.  153.  Petriv OI, et al. (2010) Comprehensive microRNA expression profiling of the hematopoietic hierarchy. Proc. Natl. Acad. Sci. USA 107:15443-15448.  154.  Tang F, et al. (2010) RNA-Seq analysis to capture the transcriptome landscape of a single cell. Nat. Protocols 5:516-535.  155.  Tang F, et al. (2009) mRNA-Seq whole-transcriptome analysis of a single cell. Nat Meth 6:377-382. 97  156.  Tang F, et al. (2010) Tracing the Derivation of Embryonic Stem Cells from the Inner Cell Mass by Single-Cell RNA-Seq Analysis. Cell Stem Cell 6:468-478.  157.  He C (2010) Grand Challenge Commentary: RNA epigenetics? Nat Chem Biol 6:863865.  158.  Metzker ML (2010) Sequencing technologies - the next generation. Nat Rev Genet 11:3146.  159.  Consortium THMP (2012) Structure, function and diversity of the healthy human microbiome. Nature 486:207-214.  160.  Li H & Durbin R (2010) Fast and accurate long-read alignment with Burrows-Wheeler transform. Bioinformatics 26:589-595.  161.  Rothberg JM, et al. (2011) An integrated semiconductor device enabling non-optical genome sequencing. Nature 475:348-352.  162.  Li H, et al. (2009) The Sequence Alignment/Map format and SAMtools. Bioinformatics 25:2078-2079.  163.  Quinlan AR & Hall IM (2010) BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics 26:841-842.  164.  Zerbino DR & Birney E (2008) Velvet: Algorithms for de novo short read assembly using de Bruijn graphs. Genome Res. 18:821-829.  165.  Chaisson MJ & Pevzner PA (2008) Short read fragment assembly of bacterial genomes. Genome Res. 18:324-330.  166.  Sommer D, Delcher A, Salzberg S, & Pop M (2007) Minimus: a fast, lightweight genome assembler. BMC Bioinformatics 8:64.  167.  Muller J, et al. (2009) eggNOG v2.0: extending the evolutionary genealogy of genes with enhanced non-supervised orthologous groups, species and functional annotations. Nucleic Acids Res. 38:D190-D195.  168.  Huson DH, Auch AF, Qi J, & Schuster SC (2007) MEGAN analysis of metagenomic data. Genome Res. 17:377-386.  98  Appendices Appendix A: Microfluidic device fabrication and operation Devices were fabricated using multilayer soft lithography (8). The device uses a three-layer design: The top layer is a “flow layer,” containing channels for reagent manipulation. The middle layer is a “control layer,” containing channels used for pneumatic valves. The bottom layer is a “blank layer,” to which control channels are sealed. All devices were made from polydimethylsiloxane (RTV615; General Electric). Devices were bonded to glass slides after plasma treatment of the bottom of the device and the slide (Harrick Plasma). Photolithography masks were designed by using AutoCAD software (Autodesk) and used to generate highresolution (20,000 dpi) transparency masks (CAD/Art Services). Molds were fabricated by photolithography on 10.2 cm silicon wafers (Silicon Quest International).  The flow layer  consisted of three different profiles: 5 µm-high rectangular frits, 12 µm-high rounded channels, and 200 µm-high cylindrical storage chambers. The 5 µm layer was made with SU8-5 negative photoresist (Microchem Corp.), the 12 µm rounded layer was made with SPR220-7 positive photoresist (Microchem Corp.), and the 200 µm layer was made with SU8-100 negative photoresist (Microchem Corp.). The control layer consisted of two different profiles: 25 µmhigh rectangular channels used for valves and 5 µm-high features used for sections of control lines passing under flow channels where valving was unwanted. The 5 µm layer was made with SU8-5 negative photoresist and the 25 µm layer was made with SU8-2025 negative photoresist (Microchem Corp.).  Resist processing was performed according to the manufacturer’s  specifications. Masters machined in Poly(methyl methacrylate) were used to mold the elution nozzle on the chip. Device operation was automated using custom software written in LabVIEW (National Instruments).  On-chip valve actuation was controlled using pneumatic solenoid actuators  (Fluidigm) connected to a PCI-6533 digital input/output card (National Instruments). A single LabVIEW program was used to execute user-designed formulation scripts inputted as text files. Compressed air (5 psi – 20 psi) was used to push reagents into the device. Prior to experiments, devices were dead-end filled with carrier fluid. A 5:1 mixture (v/v) of FC-40, FC-72, or FC3283 (Sigma Aldrich) and 1H,1H,2H,2H-perfluorooctanol (PFO) (Sigma Aldrich) was used as the carrier fluid. The 3-axis robot that enables automated sample recovery is built using three 99  interconnected precision stages T-LSM025A, T-LSR300D, T-LSR160D (Zaber).  During  elution, the device is vacuum-sealed to the lowering arm using a vacuum pump and LabVIEW software is used to coordinate stage control and on-chip valve actuation to automate insertion of the elution nozzle of the device into selected microfuge tubes. Microfluidic devices were mounted onto a DMIRE2 fluorescent microscope (Leica) or a SMZ1500 stereoscope (Nikon) for imaging. Leica L5 and TX2 filter cubes were used to image GFP and RFP fluorescence respectively. Still images of the device were acquired using CCD cameras (Q imaging Retiga 4000R and Canon 50D). Videos were made using an IV-CCAM2 CCD camera (Industrial Vision Source).  A confocal scanner (Wellscope, Biomedical  Photometrics) was used to acquire confocal fluorescent scans of the device. Appendix B: Methods for on-chip PCR, RT-PCR, and WTA For all on-chip experiments, all aqueous solutions were supplemented with 0.1% Tween 20 surfactant to avoid reagent adsorption onto PDMS surfaces of the reagent inlets and droplet interfaces, and recommended reagent proportions were used. PCR reactions on human gDNA template (Biochain) were performed using the RNAse P FAM detection kit (Biorad) and Universal Fast PCR Mix (Biorad), which includes a passive ROX fluorescent dye. RT-PCR was performed using the CellsDirect kit (Life Technologies).  WTA was performed using the  TransPlex Complete Whole Transcriptome Amplification Kit (Sigma). On-chip qPCR was performed using a prototype version of the Biomark microfluidic qPCR instrument (Fluidigm), consisting of a flatbed thermocycler equipped with a CCD camera, fluorescent illumination, and filters.  Off-chip qPCR was performed using a Chromo 4  thermocycler (Biorad) and data was analyzed using Opticon Monitor 3 software (Biorad). The thermocycling protocol for RNAse P PCR consisted of an initial hotstart at 95C for 20s, followed by 40 cycles of 95C for 1s and 60C for 30s. The thermocycling protocol for PCR reactions on all bacteria consisted of an initial hotstart at 95C for 3 min. which was also used to lyse cells, followed by 40 cycles of 95C for 10s, 60C for 30s, and 72C for 30s. All custom image analysis software described herein was written in MATLAB (Mathworks) and used functions from the Image Processing Toolbox. Volumes of stored droplets in storage chambers were computed assuming a spherical droplet geometry and using custom software to 100  segment and determine the radius of stored droplets from microscopy images. Custom software was written to analyze all on-chip qPCR images. For each cycle, droplets were first segmented using the passive ROX dye images. This dye was lncluded in the PCR reaction mix for all onchip reactions. Segmentation after each cycle is necessary since the high temperatures that the chip is heated to during PCR cause the positions of the droplets to shift slightly in the storage chambers from cycle to cycle. A pixelwise division of the FAM probe or LC green intercalating dye image by the passive ROX dye image was used to normalize data for variations in illumination across the droplet array and to account for increase in signal due to evaporation. For each droplet, an amplification curve was generated by subtracting the median normalized pixel intensity for each cycle was from that of the first cycle, and removing linear components extracted from the pre-exponential phase. Manual thresholding of the amplification curves in the exponential phase was performed to determine the cycle threshold (CT) of each droplet. For the RNAse P qPCR experiment, any reactions with a CT greater than 2 standard deviations above the mean CT corresponding to a single molecule were determined to be nonspecific amplifications and were classified as not detected. Appendix C: Methods for on-chip bacterial culture For on-chip bacterial culture, Salmonella typhimurium SL1344 transformed with plasmids encoding an ampicillin resistance gene and G/RFP were each first aerobically cultured in 2 mL of LB broth (Sigma Aldrich) with 100 µg/mL ampicillin for ~18 hrs at 37C to reach stationary growth phase (~109 cells/mL). For each strain, 2 mL of fresh culture media was then inoculated with 6 µL of cell culture and incubated for another 2 hrs to produce exponential growth phase cultures, which were mixed in a 1:1 ratio and diluted to a concentration of ~1 cell/10 nL for single-cell sorting. This was the cell concentration used for all on-chip single-cell sorting experiments. The concentration of the stationary growth phase cultures of both strains were measured by absorbance (OD 600) to be equivalent, and 10x dilutions of these cultures in media were then used to seed the on-chip multiple cell cultures. Custom software was written to analyze fluorescent images acquired from on-chip culture of GFP and RFP-expressing bacteria. As the culture media used was slightly fluorescent in the GFP channel, the first GFP image was used to segment each droplet. The boundary of each droplet was then slightly dilated to generate a new boundary, which was used to identify droplets 101  in all subsequent images. Since the incubation of the chip was performed at a relatively low temperature (25C), the droplets did not shift position significantly during the time interval between image acquisitions and this method was able to identify all droplets for all images. To generate a growth curve for each stored droplet, fluorescence intensity was first integrated over each droplet for each image in both GFP and RFP channels and a moving average filter with a window width of 3 was applied to all datapoints between the 3rd and final images for each droplet in order to remove noise. When comparing endpoint GFP and RFP fluorescence in each two-strain co-culture, normalization was performed by dividing the integrated fluorescence intensity in each channel from the final image by that of the culture with the highest endpoint integrated fluorescence intensity in each group of co-cultures seeded with the same number of cells. Appendix D: Methods for on-chip PCR-based genotyping of bacteria Suspensions of K12 Escherichia coli bacteria (ATCC 10798) were cultured and prepared as above, but without ampicillin in the media, and were stained with SYTO 9 DNA stain (Invitrogen) to aid in visualization prior to use on-chip.  For all on-chip PCR and WGA  experiments, bacterial cultures were resuspended 3 times in PBS to remove free DNA from the suspension fluid prior to use on-chip. On-chip qPCR was performed as described above. Reactions amplifying a fragment of the 16S gene in Escherichia coli and Salmonella typhimurium bacteria were performed using LC green intercalating dye (Idaho Technology Inc.), Itaq Supermix (Biorad), which includes a passive ROX fluorescent dye, and the following primers: 5’-TCGTGTTGTGAAATGTTGGGTT-3’, 5’TAAGGGCCATGATGACTTGAC-3’ (500 nM each). On-chip PCR reactions amplifying a fragment of the 16S gene gene specific to E. coli were performed as above but with primer sequences from (116). All off-chip PCR reactions on bacterial DNA were performed using the same primers and primer concentrations as on-chip and iQ SYBR Green Supermix (Biorad). 16S gene PCR amplicon from on-chip single-cell reactions to be sequenced was first eluted and diluted into 20 µL of water, 2 µL of which was used as template in a second off-chip PCR reaction to increase DNA mass for sequencing. This amplicon was then run on an agarose gel, the band was cut out, and DNA extracted using a Qiagen Qiaguick Gel Extraction kit. DNA was then sequenced using an Applied Biosystems 3730S 48-capillary DNA Analyzer with POP-7 102  BigDye® Terminator v3.1 sequencing chemistry. Sequencing data was analyzed using CLC Bio Main Workbench software. The expected sequences of the fragment amplified by the 16S gene assay in E. Coli and S. typhimurium (respectively) are: TCGTGTTGTGAAATGTTGGGTTAAGTCCCGCAACGAGCGCAACCCTTATCCTTTGTT GCCAGCGGTCCGGCCGGGAACTCAAAGGAGACTGCCAGTGATAAACTGGAGGAAG GTGGGGATGACGTCAAGTCATCATGGCCCTTA and TCGTGTTGTGAAATGTTGGGTTAAGTCCCGCAACGAGCGCAACCCTTATCCTTTGTT GCCAGCGGTTAGGCCGGGAACTCAAAGGAGACTGCCAGTGATAAACTGGAGGAAG GTGGGGATGACGTCAAGTCATCATGGCCCTTA Both amplicons are 144 bp long with mismatches between the 2 sequences at positions 51, 67, 68, and 87. Appendix E: Methods for PCR-based WGA Reactions were performed using the Picoplex WGA Kit for Single Cells (Rubicon Genomics). Additional amplification of eluted on-chip WGA product was performed using only a portion of the full protocol as recommended by the manufacturer of the kit. Thermocycling steps for on-chip WGA were performed by placing the device on a flatbed thermocycler and taping the device to the heating surface to ensure good thermal contact. Recommended thermocycling protocols were used. On-chip WGA reaction product submitted to a second round of WGA in tubes was eluted off-chip and diluted into 30 or 40 µL of water. 5 µL of this was used as template in the second round of WGA. To quantify on-chip WGA-amplified E. coli DNA, eluted sample was diluted into 20 µL of water, 2 µL of which was used in an off-chip qPCR reaction using a K12 E. coli-specific 16S gene qPCR assay as above. CT values were compared to those from a standard curve generated from qPCR reactions on dilutions of purified E. coli gDNA (ATCC) with known 16S gene copy number (7 per genome).  103  All sequencing of WGA product was performed on an Illumina Genome Analyzer IIx after library preparation performed according to recommended protocols.  E. coli samples were  sequenced using 75 and 50 bp paired end reads for 1 and 2 rounds of WGA amplfication respectively. E.coli sequence data was aligned to NCBI reference genome NC_000913 (E.coli substrain MG1655) by the BWA program (160) Appendix F: Methods for MDA-based WGA FACS-sorted E. coli cells were prepared as described above and adjusted to a concentration of ~106/mL. FACS sorting was performed using a BD Influx cell sorter using a 70 µm nozzle at 30 psi with a 488 nm excitation laser and 530/30 filter to detect the SYTO9 stain. All MDA reactions were performed using the REPLI-g MIDI kit (Qiagen). On-chip reactions were performed using the recommended protocol for amplification of cells with the following modifications.  Two versions of the lysis buffer were used.  Thermocycling steps were  performed by placing the device on a flatbed thermocycler and taping the device to the heating surface to ensure good thermal contact. On-chip MDA products were eluted from the device into a final volume of 40 µL. The second microliter-volume MDA on the single-cell on-chip MDA product was performed following the protocol recommended by the manufacturer for genomic DNA using 5 µL of the eluted on-chip product. The microliter-volume MDA on the FACS-sorted single cell was performed using the recommended protocol. Quantification of WGA product by E. coli strain-specific qPCR was performed as described above. Prior to the construction of sequencing libraries, all samples (with the exception of the unamplified gDNA) were bead-purified with the Agencourt AMPure XP kit (Beckman Coulter) and quantified using a Nanodrop ND-1000 spectrophotometer (NanoDrop Products). Using the manufacturer’s protocol, libraries were constructed for sequencing on the Ion Torrent PGM, starting with 1 µg of DNA from the combined nanolitre/microliter sample and 100 ng from the unamplified gDNA (ATCC) and microliter samples. The nanolitre sample and the no-cell control were both below the detection limit of the NanoDrop, so library construction for these samples began with less than 100 ng. All libraries were then sequenced on the Ion Torrent PGM platform (Life Technologies) (161), using two Ion 316 chips to the nanolitre sample and one Ion 316 chip to each of the other samples. 104  De novo assembly was performed using CLC Genomics Workbench software (CLC bio) and the assembled contigs were mapped to NCBI reference genome NC_000913 (E.coli substrain MG1655) using the BWA-SW algorithm (160), while the raw sequencing reads were mapped to the same reference using Torrent MAP software. The sequence alignment data were reformatted using SAMtools (162) and subsampled with a custom bash script. Reference coverage statistics were calculated with SAMtools and BEDtools (163) and figures were generated with a custom MATLAB script. Appendix G: Preparation of environmental samples Environment 1 – Seawater enrichment culture. A modular medium buffered with 10mM MOPS pH 7.2 containing 1X seawater base amended with salts (10mM NH4Cl, 1.5 mM KH2PO4), cofactors (12X Vitamins, B12, and trace elements) was used. The carbon source supplied to the medium was 2 mM Na2HCO3. Electron donors included 8 mM Na2S or 30 mM NaS2O3 with 20mM NaNO3 as alternative electron acceptor. Primary enrichment cultures were generated at 26°C using modular medium mixed with varying dilutions of Saanich Inlet 135 meter inoculum. 10 µL of the enrichment culture was diluted in 1 mL of PBS prior to on-chip use. Environment 2 – Human oral swab. The oral biofilm sample was obtained from the mouth of a 31-year-old male by scraping a tooth with a sterile pipette tip and resuspending the accumulated biofilm in 1 ml of PBS for on-chip use. Environment 3 – Marine sediments. Aggregates of microbial cells were extracted from marine sediments collected at a depth of 836 m in the Santa Monica Basin (LAT: 33°47’99’’N, LONG: 118°38’83’’W). Once thawed from their storage conditions (-80°C), 10 g (wet weight) of a core’s 3-6 cm (bsf) layer were suspended in 14 ml of 1X PBS. The slurry was vortexed and sonicated on ice at 180 Watts for 20 sec with a Sonicator Ultrasonic Processor XL 2020. Percoll density gradients were generated by centrifuging 30 ml of percoll/PBS 1X solution (1:1 vol/vol) at 38325 ×g for 30 min (4°C). The entire sediment slurry was gently dispensed on top of 8 percoll gradients (approx.. 3 ml per tube), which were then centrifuged at 10300 ×g for 15 min (4°C). A cell suspension was gathered by removing 20 ml from the top of each percoll gradient, and cell aggregates ranging from 3-8 µm in diameter were enriched by filtration through two 8 µm pore size polycarbonate membranes and onto a 3 µm pore size membrane. Cells were 105  collected from the 3 µm pore size membrane with 1 ml of PBS 1X/ethanol (1:1 vol/vol), and stored at -20°C until processed. Cells were washed and further enriched prior to sorting on-chip by centrifuging 200 µl of the PBS/ethanol suspension at 5900 ×g for 5 min (4°C). The supernatant was removed and the pellet resuspended in 200 µl of 1X PBS, before the suspension was again centrifuged at 5900 ×g for 10 min (4°C). The final pellet was resuspended in 10 µl 1X PBS prior to use on-chip. All environmental samples were stained with SYTO9 DNA stain prior to use on-chip. Appendix H: Sequencing data analysis for environmental samples For environmental samples, raw Illumina reads (65,067,118 paired-end reads, 75 bp in length) were first de-multiplexed into 74 fastq files. Each set of reads was trimmed for low quality from both  ends  and  were  assembled  using  Velvet  (164)  (also  see  http://www.ebi.ac.uk/~zerbino/velvet/) at a range of Kmers (51,55,59,63,67,71) using all the trimmed reads and the default Velvet settings (flags: -exp_cov auto). Contigs generated by each assembly (6 total contig sets), were merged using a combination of in-house Perl scripts. Contigs were then sorted into two pools based on length. In an effort to join minimally overlapping contigs, those smaller than 1800 bp were assembled together using Newbler (165) in attempt to generate larger contigs (flags: -tr, -rip, -mi 98, -ml 80). All assembled contigs larger than 1800 bp, as well as the contigs generated from the final Newbler run were combined together using minimus2 (166) (also see http://sourceforge.net/projects/amos) using an overlap length of 80 bp, a minimum overlap identity cutoff of 98%, and a consensus error of 0.06 for merging contigs. The BWA program was used to map reads back to the final contigs for contig verification and in order to establish contig fold coverage and percent read assembly statistics. The taxonomic evenness of each sequenced WGA sample was evaluated by plotting GC content distribution using a kernel density (Gaussian) as implemented by R package lattice. Data from all metagenomes of each of the three environments was summarized in Figure 27. The taxonomic profile of each metagenomic dataset was first evaluated using a phylogenomic approach, as implemented by MLTreeMap (132). Briefly, this procedure is based on the detection and phylogenetic identification of the 40 most taxonomically informative genes (i.e. COGs). For each metagenome, the overall taxonomic identity of all instances of these genes was 106  compiled using the probabilistic distribution of query sequences among the reference sequences of the 40 COG summary ‘tree of life’. Reference sequences with less than 5% of the query sequence assignment probabilities were removed for noise reduction, and taxonomic profiles were compiled by summing the taxonomy of all remaining reference sequences at the phylum or subphylum level and normalizing against the total number of assigned genes. Taxonomic profiles were then further defined using a direct sequence comparison approach. Contigs assembled from each metagenome were searched against the database eggNOG (167) (also  see  http://eggnog.embl.de/)  and  against  NCBI’s  RefSeq  proteomic  database  (http://www.ncbi.nlm.nih.gov/RefSeq/) using blastx with an E value cutoff of 1e-8. For the search against eggNOG, contigs were taxonomically assigned to the phylum or subphylum of the reference sequence producing the best alignment (as determined by bitscores). Dataset-specific taxonomic profiles were then compiled by normalizing against the total number of assigned contigs. For the search against RefSeq, contigs were taxonomically assigned from the blast outputs using MEGAN (168) (also see http://ab.inf.uni-tuebingen.de/software/megan/) with default settings. Uninformative categories of MEGAN taxonomic assignment (e.g. not hits, not assigned, unicellular organisms) were removed, and dataset-specific taxonomic profiles were again compiled by normalizing against the total number of assigned contigs. Information gathered using MLTreeMap, eggNOG, and RefSeq is presented in Appendix L to Appendix N for all metagenomes, and for selected datasets from each environment in Figure 27C. To compare the taxonomic profiles characterizing the 69 WGA samples of all three environments (no NTCs), the proportional representation of all informative taxonomic categories identified by MEGAN in the blast outputs against RefSeq were compiled in a matrix (samples 21 and 93 were removed due to lack of sufficient information). Samples were then submitted to a hierarchical cluster analysis using R package hclust (Figure 27B). As mentioned in the main text, two of the 72 submitted samples failed to produce libraries that passed QC. These two libraries were prepared from samples 69 and 70, which were expected to be no cell controls for ENV1. Neither of these was found to have any significant similarity to single cell samples from ENV1 but did produce significant alignment to unrelated taxa. We 107  believe that these samples were either contaminated during library preparation or mislabeled and thus excluded them from any subsequent analysis.  108  Appendix I: Kernal density of %GC content for individual ENV1 samples  109  Appendix J: Kernal density of %GC content for individual ENV2 samples  110  Appendix K: Kernal density of %GC content for individual ENV3 samples  111  Appendix L: Taxonomic assignments derived from sequencing data for individual ENV1 samples  112  Appendix M: Taxonomic assignments derived from sequencing data for individual ENV2 samples  113  Appendix N: Taxonomic assignments derived from sequencing data for individual ENV3 samples  114  


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



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


Related Items