UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Systems biology of cellular signaling : quantitative experimentation and systems genetics approaches 2009

You don't seem to have a PDF reader installed, try download the pdf

Item Metadata

Download

Media
ubc_2009_spring_taylor_robert.pdf
ubc_2009_spring_taylor_robert.pdf [ 13.39MB ]
Metadata
JSON: 1.0067122.json
JSON-LD: 1.0067122+ld.json
RDF/XML (Pretty): 1.0067122.xml
RDF/JSON: 1.0067122+rdf.json
Turtle: 1.0067122+rdf-turtle.txt
N-Triples: 1.0067122+rdf-ntriples.txt
Citation
1.0067122.ris

Full Text

Systems Biology of Cellular Signaling Quantitative Experimentation and Systems Genetics Approaches by Robert James Taylor B.A.Sc., The University of British Columbia, 2002 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE STUDIES (Genetics) The University Of British Columbia (Vancouver) April 2009 c© Robert James Taylor Abstract Cellular regulation is governed by dense biomolecular networks consisting of pro- teins, nucleotides, lipids, and metabolites that dynamically coordinate cellular de- cision making in the face of complex and time-varying environmental stimuli. Ob- taining predictive models of these complex networks is a central goal of systems biology and requires sophisticated technologies for the acquisition and integration of many disparate data types. Recent genomic, proteomic and cellular imaging developments have greatly enabled systems-level studies, but further technologi- cal advances are needed. For instance, current high-throughput biochemical and cellular measurement techniques are generally limited to the analysis of cell pop- ulations, and the development of single-cell technologies are needed to advance predictive models of cellular networks. Large-scale genetic analyses are highly informative of the complex architecture of cellular networks but further compu- tational methods are required to manage data complexity. In this thesis I present the development of two technologies, a microfluidic single-cell experimental plat- form and a genetic-network computational analysis platform, to address these is- sues and apply them to the study of prototypical eukaryotic signaling systems in Saccharomyces cerevisiae. First I describe microfluidic technology for the high-throughput analysis of single-cells subject to complex environmental conditions. Using this platform, I studied cellular response of the mating pathway in Saccharomyces cerevisiae under a series of genetic and time-varying environmental perturbations. This analysis revealed dynamic phenotypes that are not observable under static conditions and allowed for the stratification of system components into distinct functional roles. In addition, I describe advances to this technology that allow for the tracking of ii individual cells over long experimental time frames. These developments enabled the investigation of sources of cell-to-cell variability not detectable otherwise. Second I describe a computational platform for analyzing complex genetic in- teraction networks. These networks describe functional relationships between gene systems and can be used to delineate information flows through complex cellular circuits. Genetic interactions networks are dense and information rich, and re- quire sophisticated computational methods for their analysis. In this work, I de- veloped network algorithms to identify biologically informative patterns within a multi-mode genetic interaction network to reveal functional sub-networks and information-hubs of the filamentation pathway in Saccharomyces cerevisiae. iii Table of Contents Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii Co-Authorship Statement . . . . . . . . . . . . . . . . . . . . . . . . . . xiv 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.1 General Introduction . . . . . . . . . . . . . . . . . . . . . . . . 2 1.2 Thesis Summary . . . . . . . . . . . . . . . . . . . . . . . . . . 4 1.3 Yeast Signaling . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1.3.1 Model Organisms . . . . . . . . . . . . . . . . . . . . . . 5 1.3.2 Signaling Pathways in Yeast . . . . . . . . . . . . . . . . 5 1.3.3 Mating and Filamentation/Invasion Pathways . . . . . . . 6 1.3.4 Complex Systems Models: Yeast . . . . . . . . . . . . . 8 1.4 Microfluidic Quantitative Single-cell Analysis . . . . . . . . . . . 9 1.4.1 Quantitative Single-Cell Studies . . . . . . . . . . . . . . 9 1.4.2 Single Cell Microfluidics . . . . . . . . . . . . . . . . . . 10 iv 1.5 Systems Genetics . . . . . . . . . . . . . . . . . . . . . . . . . . 13 1.5.1 Genetic Analysis of Signaling Pathways . . . . . . . . . . 13 1.5.2 Genetic Interaction Networks . . . . . . . . . . . . . . . 15 1.5.3 Network Analysis . . . . . . . . . . . . . . . . . . . . . . 17 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 2 Dynamic Analysis of MAPK Signaling Using a Microfluidic Live- Cell Imaging Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . 41 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 2.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 2.2.1 High-Throughput Microfluidic Live Cell Imaging Platform 45 2.2.2 Imaging Studies of Pheromone Response Pathway . . . . 49 2.2.3 Response Under Chemostatic Conditions . . . . . . . . . 50 2.2.4 Pathway Response Under Dynamic Stimulation . . . . . . 55 2.2.5 Response to Periodic Stimulation . . . . . . . . . . . . . 57 2.2.6 Mutants Implicated in Dynamic Phenotypes . . . . . . . . 60 2.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 2.4 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 2.4.1 Cell loading. . . . . . . . . . . . . . . . . . . . . . . . . 63 2.4.2 Microfluidic Control . . . . . . . . . . . . . . . . . . . . 64 2.4.3 Microfluidic Fabrication . . . . . . . . . . . . . . . . . . 65 2.4.4 Image Acquisition . . . . . . . . . . . . . . . . . . . . . 65 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 3 Network Motif Analysis of a Multi-Mode Genetic-Interaction Net- work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 3.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 3.2 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . 76 3.2.1 Multi-Mode Genetic-Interaction Network . . . . . . . . . 76 3.2.2 Genetic-Interaction Patterns Reflect the Underlying Molec- ular System . . . . . . . . . . . . . . . . . . . . . . . . . 77 3.2.3 Statistical Model of a Null Hypothesis . . . . . . . . . . . 79 v 3.2.4 Genetic-Interaction Network Motifs . . . . . . . . . . . . 80 3.2.5 Molecular Information and Genetic-Interaction Network Motifs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 3.2.6 Comparing Network Patterns in a Similar Genetic-Interaction Network . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 3.2.7 Open Source Software . . . . . . . . . . . . . . . . . . . 92 3.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 3.4 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . 94 3.4.1 Network Randomization . . . . . . . . . . . . . . . . . . 94 3.4.2 Motif Enumeration Techniques . . . . . . . . . . . . . . . 96 3.4.3 GoSlim Molecular Function Annotations . . . . . . . . . 97 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 4 Conclusions and Recommendations for Further Work . . . . . . . . 105 4.1 Summary of Thesis . . . . . . . . . . . . . . . . . . . . . . . . . 105 4.2 Dynamic Single Cell Analysis . . . . . . . . . . . . . . . . . . . 107 4.2.1 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . 107 4.2.2 Future Directions . . . . . . . . . . . . . . . . . . . . . . 109 4.3 Systems Genetics . . . . . . . . . . . . . . . . . . . . . . . . . . 112 4.3.1 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . 112 4.3.2 Future Directions . . . . . . . . . . . . . . . . . . . . . . 114 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 A Appendix to: Dynamic Analysis of MAPK Signaling Using a Mi- crofluidic Live-Cell Imaging Matrix . . . . . . . . . . . . . . . . . . 125 A.1 SI Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126 A.1.1 Fabrication Protocol . . . . . . . . . . . . . . . . . . . . 126 A.1.2 Microfluidic Control . . . . . . . . . . . . . . . . . . . . 126 A.1.3 Chemicals and Media . . . . . . . . . . . . . . . . . . . . 127 A.1.4 Cell Preparation . . . . . . . . . . . . . . . . . . . . . . 127 A.1.5 Chemical Mixing and Perfusion . . . . . . . . . . . . . . 127 A.1.6 Constant Stimulation Protocol . . . . . . . . . . . . . . . 128 vi A.1.7 Single Transient Pulse Protocol . . . . . . . . . . . . . . 131 A.1.8 Short Repeated Pulses Protocol . . . . . . . . . . . . . . 131 A.1.9 Biological Constructs . . . . . . . . . . . . . . . . . . . . 133 A.1.10 Image Analysis Pipeline Algorithms . . . . . . . . . . . . 133 A.2 Supporting Text . . . . . . . . . . . . . . . . . . . . . . . . . . . 140 A.2.1 Experimental Variability of Microfluidic Platform . . . . . 140 A.2.2 Morphology Classifications Under Constant Stimulation . 141 A.2.3 Single Pulse Analysis . . . . . . . . . . . . . . . . . . . . 144 A.2.4 Pulse-Width Dependent Growth Rate . . . . . . . . . . . 152 A.2.5 Calculation of d[GFP]/dt . . . . . . . . . . . . . . . . . . 155 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156 B A Microfluidic Platform for High-Throughput Single-Cell Tracking and Dynamic Environmental Control . . . . . . . . . . . . . . . . . 158 B.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158 B.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162 B.2.1 Microfluidic Chip Overview and Operation . . . . . . . . 162 B.2.2 High-Throughput Multimode Live-Cell Imaging . . . . . 166 B.2.3 Single-Cell Tracking Reveals Heterogeneous Decision Mak- ing in a Narrow Pheromone Concentration Range . . . . . 171 B.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174 B.4 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176 B.4.1 Chip Fabrication . . . . . . . . . . . . . . . . . . . . . . 176 B.4.2 Segmentation and Tracking Algorithm . . . . . . . . . . . 177 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 vii List of Tables A.1 Pumping Protocol for Creating 32 Different α-factor Concentrations.129 A.2 Strains Used in this Study . . . . . . . . . . . . . . . . . . . . . . 136 viii List of Figures 2.1 Schematic of the Microfluidic Device . . . . . . . . . . . . . . . 47 2.2 Mating Response to Persistent α-factor Stimulation . . . . . . . . 52 2.3 Morphological Response and Transient Stimulation Responses. . 55 2.4 Mating Response to Short Repeated Pulses of α-factor . . . . . . 60 3.1 Multi-Mode Genetic-Interaction Motifs and the Underlying Molec- ular System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 3.2 Motifs in the Yeast-Invasiveness Genetic-Interaction Network . . . 83 3.3 Motif Subnetworks . . . . . . . . . . . . . . . . . . . . . . . . . 85 3.4 Examples of Motifs Integrating Gene Annotations. . . . . . . . . 88 3.5 Annotation-Motif Subnetworks . . . . . . . . . . . . . . . . . . . 91 A.1 Sieve Valves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130 A.2 Stimulation protocols . . . . . . . . . . . . . . . . . . . . . . . . 132 A.3 Image Segmentation Algorithms . . . . . . . . . . . . . . . . . . 138 A.4 Morphological Response of the Yeast Strains to α-factor Concen- tration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144 A.5 Single Pulse Analysis . . . . . . . . . . . . . . . . . . . . . . . . 148 A.6 Response Variability within a Single Microfluidic Device. . . . . . 149 A.7 Reproducibility of Results. . . . . . . . . . . . . . . . . . . . . . 152 A.8 Image Focus over 12 hours . . . . . . . . . . . . . . . . . . . . . 155 B.1 Microfluidic Chip Design and Operation . . . . . . . . . . . . . . 164 B.2 On-Chip Reagents Mixing and Temporal Control of Chemical En- vironment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167 B.3 High-Throughput Gene Expression Measurements . . . . . . . . . 170 ix B.4 Lineage Tracking with Single-Cell Resolution; Identification of a Switch-Like Response . . . . . . . . . . . . . . . . . . . . . . . 174 x Glossary BSA Bovine serum albumin CV Coefficient of variability dSLAM Diploid-based synthetic lethality analysis on microarrays E-MAP Epistasis miniarray profile FACS Fluorescent-activated cell sorting FRET Fluorescence resonance energy transfer GFP Green fluorescent protein GPCR G-protein coupled receptor MAPK Mitogen activated protein kinase MSL Multilayer soft-lithography PAK P21-activated kinase PCR Polymerase chain reaction PDMS Poly(dimethylsiloxane) SCD Synthetic complete dextrose SGA Synthetic genetic analysis SGD Saccharomyces genome database (www.yeastgenome.org) xi Acknowledgments I would like to thank everyone who has helped in my research and my transition into the biological sciences. I particularly indebted to my supervisors Dr. Tim Galiski who taught me much about biology and research, and Dr. Phil Hieter who enabled me to conduct research from afar, at the Institute for Systems Biology. I would like to thank my supervisory committee, Dr. Anne Condon, Dr. Elizabeth Conibear, and Dr. Steven Jones, for guidance and direction. I am also grateful for those who supported my research, directly or indirectly at the University of British Columbia: Carl Hansen, Didier Falconnet, Shay Ben-Aroya, Kirk McManus, Jan Stoepel, Irene Barrett, and Dave Thomson, and at the Institute for Systems Biology: Greg Carter, Susanne Prinz, Iliana Avila-Campillo, Song Li, Ramsey Saleem, John Boyle, Jennifer Smith, James Spotts, and Alan Diercks. Finally, I am grateful to the Michael Smith Foundation for Health Research for my graduate scholarships. xii Dedication To my parents and sister who have supported me throughout my academic endeavors. xiii Co-Authorship Statement Chapters 2, 3, Appendix A, and Appendix B were co-authored work. In Chapter 2 and Appendix A, I designed the study, designed and fabricated the microfluidic platform used, constructed the biological strains used, performed many of the mi- crofluidic experiments, designed and coded the image analysis pipeline, conducted all analysis, and wrote the manuscript. D. Falconnet performed many of the mi- crofluidic experiments and assisted in preparation of the manuscript. A Niemisto designed and coded the image segmentation algorithms. S.A. Ramsey assisted with experimental design and analysis. S. Prinz assisted with experimental design. I. Shmulevich assisted with experimental design. T. Galitski assisted with study and experiment design and manuscript preparation. C.L. Hansen assisted with study and experiment design, microfluidic platform design, and manuscript preparation. In Chapter 3, I designed the study, designed and coded the analysis software, and prepared the manuscript. A.F. Siegel assisted in the statistical analysis of mo- tifs with respect to multiple hypothesis testing. T. Galitski assisted in study design, analysis design, and manuscript preparation. In Appendix B, I assisted in study design, microfluidic design, and designed and coded the image pipeline used. D. Falconnet lead the microfluidic design and xiv conducted all experiments and analysis. A. Niemisto designed and coded the im- age segmentation and tracking algorithms. T. Galiski assisted with study design. I. Shmulevich assisted with study design. C. Hansen assisted with study design, microfluidic design, and manuscript preparation. xv Chapter 1 Introduction ”The completion of the Human Genome Project fueled the expectation for the rapid translation of ”genes to drugs.” Unfortunately these hopes and dreams shattered with the realization that disease biology is much more complex than we had first realized. Despite all the sophisticated technology for drug discovery, the expected acceleration in innovative medical therapies reaching patients has not occurred. This is because deciphering the mechanisms of disease requires a deep knowledge of how signaling transduction pathways operate. This is why biology is undergoing a fundamental shift from a descriptive to a quantita- tive, predictive science. This transition is being driven by advances in genome sequencing, massive amounts of data, rapidly expanding computational resources, and the introduction of powerful new ana- lytical technologies. 1 Advancements in biology in the 21st century will be dominated by re- search at the convergence of biological, physical, and information sci- ences. Disease will be understood at a systems level, which promises predictive, preventive, and personalized medicine. Although systems biology is still in its infancy, we are at a turning point in our under- standing of what the future holds for biology and human medicine.” Biotech 2008. Life Sciences: A 20/20 Vision to 2020. Burrill & Com- pany. [14] 1.1 General Introduction Biological systems are irreducible: function and phenotype are emergent proper- ties of complex biomolecular networks that govern cellular regulation and physi- ology. Obtaining a predictive understanding of biological processes thus requires a global view of biomolecular systems and cannot be achieved by the independent study of individual genes or gene products. At the multicellular organism level, tissues are interdependent, communicating through dedicated endocrine, nervous, and circulatory systems. Individual organs cannot exist unaided outside of the organism, and perturbations occurring in one tissue cause measurable effects in other non-neighboring tissues [20]. At the single-cell level, complex systems of proteins, lipids, nucleotides, and metabolites interact together in dense biomolecu- lar networks to regulate phenotype. Analysis of complex single-cell and multicell systems thus requires sophisticated experimental and analysis tools. Systems biol- ogists are developing these tools and have begun analyzing biological response at the network and system level [63]. Systems biology is an emerging field, and con- 2 tinued efforts in the development of both experimental and analysis technologies are needed. Obtaining predictive models of biological systems is a primary goal of systems biology. This requires a quantitative understanding of the regulatory mechanisms of the cell such that response to genetic and chemical perturbations may be pre- dicted under a wide variety of environmental conditions [62]. Traditional biologi- cal methods rely on qualitative descriptive models that simplify biological process to allow for conceptual accessibility. Although successful in defining key biologi- cal themes, these methods are limited in their predictive power. For example, cell signaling systems were originally described as linear signaling cascades leading to the idea of a signaling ’pathway’. Such descriptions can predict qualitative phe- notypes, such as the loss of signaling upon removal of key system components, however are unable to predict quantitative phenotypes, such as the percentage of gene expression increase upon increased amount of stimuli. The pathway descrip- tion of signaling is now being replaced by a network view [27], and sophisticated network models of cellular regulation are leading to impressive predictive power of cellular response [10]. Although much progress has been made, constructing high-quality quantitative models of biological processes remains a challenge. Recent works have described methodologies for this task, with an emphasis on data driven reverse engineering techniques [10, 57]. The methods presented in these papers can be broadly classed into three major steps: 1) identify all parts involved in the process of interest; 2) delineate network topology by identifying and analyzing functional and physical 3 connections between parts; 3) iteratively describe the system quantitatively, exper- imentally test, and refine. Although stated simply, each task is challenging due to the complexity of cellular systems and of the environments for which cell live, and sophisticated experimental and analysis tools are required. In this thesis I present the development of such tools and their application to the study of cellular signal- ing. These works are summarized as follows. 1.2 Thesis Summary In Chapter 2 I describe an experimental platform that gives exquisite control of environmental conditions, allowing for the study of cellular response under time- varying environmental perturbations. I used this platform to screen for components of the mating pathway in Saccharomyces cerevisiae that are involved in regulating dynamic cellular phenotypes. In Chapter 3 I describe a computational method for extracting meaningful biological data from dense multi-mode genetic interac- tion networks. I used this method to identify paths of information flow within the filamentation response of Saccharomyces cerevisiae. Preceding these works is in- troductory background material describing the signaling pathways studied (Chap- ter 1.3), and gives an overview of relevant technical fields including single-cell analysis (Chapter 1.4) and the use of genetic interaction networks to delineate biological systems (Chapter 1.5). In Chapter 4 I present conclusions and future directions. 4 1.3 Yeast Signaling 1.3.1 Model Organisms The study of human biology is challenged by the complexity of the organism and the ethical implications of human experimentation. The majority of biolog- ical knowledge to date has been obtained through the study of model systems and the inference of understanding to humans. Models commonly used include complex multi-cellular organisms like rodents and non-human primates, simple multi-cellular model organisms like flies and worms, and single-cellular models like human cell lines, yeast, and bacteria. These models have provided a wealth of information translatable to human biology with homologous process occurring both at the single gene level and at the systems level. In this thesis, Saccharomyces cerevisiae is studied to obtain novel understandings relating to the basic principles of eukaryotic cellular signaling. 1.3.2 Signaling Pathways in Yeast Due to the obvious phenotype, facility of genetic manipulations, and the availabil- ity of genome-wide deletion and fusion libraries, Saccharomyces cerevisiae (yeast) has become a prototypical model of single-cell eukaryote biology. Yeast cell sig- naling has been particularly well studied, with many individual components char- acterized, including G-protein coupled receptors (GPCRs) [12, 50, 80], mitogen activated protein kinases (MAP kinases) [41], second messengers [76], and tran- scriptional regulators [29, 42]. In addition, common signaling network motifs have been well described, allowing for the construction of quantitative and predictive signaling models [22, 65, 100]. 5 One such highly studied motif is the MAP kinase signaling cascade [93, 121], a highly conserved signal transduction module used to relay external stimuli in- formation to the nucleus. MAP kinase cascades consist of two or three kinases that function through sequential phosphorylative activation, with the MAP kinase being the final active component. Five MAP kinases exist in S. cerevisiae, and are components in pathways involved in sensing environmental changes in: mating response, filamentation/invasion, high osmolarity growth, cell integrity, and spore wall assembly [45]. Years of genetic and biochemical analysis, and more recently high-throughput proteomic [36, 38] and microarray studies [119], have resulted in a detailed understanding of the core signaling circuitry of yeast response to these stimuli. Studies in this thesis focused on MAP kinase signaling in the mating and filamentation/invasion response. 1.3.3 Mating and Filamentation/Invasion Pathways The yeast mating system, which regulates the desire to mate between haploid yeast cells, is arguably the most well understood eukaryotic signaling pathway. Decades of research have identified all components involved in mating signal transduction, and many regulatory aspects of the network have been uncovered. Briefly, hap- loid yeast exist as one of two mating types: MATa and MATα . Haploid cells of opposite type can mate through a cellular fusion event to create a single diploid cell. Mating between haploid cells is initiated through the binding of pheromones (a-factor and α-factor) to mating type specific GPCRs (Ste2 for MATa or Ste3 for MATα) [12, 80] causing the activation of the signaling pathway leading to a large transcriptional response [119], mating morphology, and cell cycle arrest [116]. 6 Ste2/Ste3 activation causes the liberation of β γ subunits (Ste4 and Ste18) [50]. The liberated Ste4 and Ste18 complex recruits to the membrane the MAPK mod- ule, consisting of the kinases Ste11 [91], Ste7 [106], and Fus3 [41], and the scaffold protein Ste5 [87, 105]. Ste20 (a homolog of the mammalian P21-activated kinase, PAK) [31] phosphorylates Ste11 (a MEKK homolog) to initiate a kinase cascade in which Ste7 (a MEK homolog) and Fus3/Kss1 are sequentially phosphorylated. This activated MAP Kinase translocates to the nucleus where it phosphorylates the CKI/scaffold protein Far1 [110] and the master transcription factor Ste12 [29]. Ac- tivation of Far1 enforces G1 growth arrest and initiates a morphological transition of shmoo growth towards the mating partner [15, 116]. Activation of Ste12 ini- tiates a transcriptional program of pheromone response involving a suite of 200 genes [119]. The filamentous growth pathway (invasive growth in haploids and pseudohy- phal growth in diploids) is also a very well studied signaling system and is used to understand principles of cellular differentiation and signaling specificity. The filamentous response is induced when the cells are placed under certain starva- tion conditions: glucose depletion for haploids [28, 93] and nitrogen limitation for diploids [43]. The filamentation system shares many of the same components as the mating pathway including Ste20, Ste11, Ste7, Dig1, Dig2, and Ste12. Fil- amentation specific components include environmental receptors (which are not yet well defined), the lack of Ste5 scaffolding [3], cyclic AMP signaling [94, 95], and the use of the transcription factor Tec1 [42, 119]. In addition, upon pathway activation cells enter a filamentation specific cell cycle. In contrast to the the ax- ial budding (budding near bud scar and birth end) or bipolar budding (alternating 7 budding near and opposite bud scar and birth end) patterns of haploid and diploid vegetative yeast respectively, filamentous yeast elongate and bud in a unipolar pat- tern (daughter cells budding opposite of birth end). In addition, cells demonstrate increased adherence and invasion into the substratum, and increased cell-cell ad- hesion. These phenotypes give a colony of non-motile yeast a means to prospect for new territory when faced with nutrient limiting conditions. 1.3.4 Complex Systems Models: Yeast The richness of data and detail surrounding these pathways allow for new stud- ies of systems function to be placed upon a backdrop of molecular network data, and S. cerevisiae has become a key model for systems biology [57]. Research has progressed beyond studying the mechanisms of how information is propagated from the cellular membrane to the nucleus (signal transduction), to studying the mechanisms of information processing that allow cells to thrive in complex envi- ronments. For example, a long-standing question in cell biology is how signaling systems are able to translate multiple environmental signals into specific responses while using a shared set of core components. Work over the past decade has ex- plored this question using yeast as a model, and has found regulatory mechanisms at the scaffold [3, 37, 71], MAP kinase, and transcription factor level [5, 21, 35]. Recently, research has begun to focus on even more questions relating to the com- plexity of cellular regulation, including how signaling pathways communicate with other cellular modules, [92, 103], process cellular information in the face of bio- chemical noise [25], and regulate response in dynamically changing environments [8, 49, 78]. The collection of these studies reflect the complexity of cellular signal- ing and demonstrate that although well studied, much about these model systems 8 is still unknown. The work presented in this thesis focuses on the development of new technologies that help to manage this complexity, with specific application to studying the information processing mechanisms of the mating and filamentation signaling circuits in Saccharomyces cerevisiae. 1.4 Microfluidic Quantitative Single-cell Analysis 1.4.1 Quantitative Single-Cell Studies The construction and refinement of predictive biomolecular models requires an abundance of high quality quantitative data. Given the complexity of cellular regu- lation, these data need to be globally acquired under vast numbers of environmental conditions. Until recently, biomolecular analyses consisted of single measurements conducted serially, limiting early biomolecular models to be qualitative and over simplified. Recent technological advances have greatly improved the experimen- talist’s toolbox and high-throughput methods have enabled data driven methods for model building [10]. For example, advances in microarray and proteomic tech- nologies enabled the quantitative and global analysis of transcripts [69, 97] and proteins [38]. Yet, although these methods have greatly improved our ability to measure global response, they are not high-throughput in regards to the number of environmental conditions tested and are limited in their ability to study the re- sponse of individual cells. Further, such measurement tools have not improved our ability to test cellular response under complex environments such as multiple or sequential stimuli conditions. Advances are required for the high-throughput study of single cells under large numbers of well controlled conditions. This will require 9 the advancement of technologies that can manipulate and measure individual cells, such as microfluidics, for high-throughput analysis. 1.4.2 Single Cell Microfluidics Microfluidics allows for fluidic manipulation at the scales of individual cells and is a promising technology class for single cell studies. Microfluidic devices ex- ist in glass, silicon, and polymers, and devices from all material types have made much progress in single cell analysis in recent years [34]. Recent studies have demonstrated on-chip live cell assays such as cell capture and culturing [4, 17, 26, 44, 56, 86, 104, 118], cell sorting [39, 40, 52, 55, 113], exquisite control of the microenvironment both in terms of chemical composition [1, 56, 59, 83] and chemical sequence [8, 49, 78], along with many biochemical assays such as micro- assay PCR [68, 82], DNA/RNA analysis [9, 13, 51, 74, 114], and protein analysis [16, 54, 61, 96, 117]. Recent studies have combined single cell microfluidic technologies with sys- tems biology studies of yeast signaling. These works demonstrate the power of applying advanced microfabrication techniques to analyze single cells. For exam- ple, in Paliwal et al. [83], microfluidic devices were fabricated to create precise concentration gradients without flow, allowing for sensitive studies of the concen- tration dependent pheromone response at the single cell level. Three recent studies developed microfluidic devices to rapidly exchange the media surrounding cells, allowing for the study of the frequency dependent response of cells under oscillat- ing stimuli. In these studies Hersen et al. [49] and Mettetal et al [78] measured response in the osmolarity system while Bennett et al. [8] measure response in the 10 galactose metabolic pathway. Lastly, Charven et al. [19] used a simple microflu- idic device to track single cell lineages while analyzing properties of the inducible MET3 promoter. Although these studies have identified novel biological under- standings, the platforms used lack scalability both in terms of the numbers and complexity of assays completed, ultimately limiting the breath of biological analy- ses possible. This limitation is technological and due to the microfluidic techniques involved. A recent technological advance called multilayer soft-lithography (MSL) promises to solve microfluidic issues of scalability and integration. MSL’s predecessor, soft- lithography (SL), is a microfluidic fabrication technique [32, 115] that casts micro- channels into a soft elastomer, usually poly(dimethylsiloxane) (PDMS), using a hard pre-fabricated master mold [32]. PDMS is an inert silicone elastomer with excellent biocompatibility and optical properties [102] making it an ideal material for biological studies. Soft lithography allows for the inexpensive and rapid fab- rication of devices, but is limited in its ability to manipulate fluids due to the lack of micro-mechanical control elements such as valves and pumps. MSL remedies this issue by stacking planar fluidic layers separated by a thin flexible PDMS mem- brane. In this way, orthogonal crossing channels can be used to create monolithic soft elastomer valves [112]. In contrast to mechanical valves fabricated in glass and silicon devices, MSL valves are extremely robust and fabricated with high yield, allowing for valve densities of thousands per cm2 to be routinely achieved. As a fundamental element, multiple MSL valves can be used to create higher or- der fluidic control elements. Demonstrated components include on chip: i) pumps to allow for high-precision fluidic control and metering without the need to tightly 11 regulate fluid pressures [112], ii) multiplexers that enable fluidic addressing using a reduced number of pin-out control connections [53, 107], iii) cell sorting junctions to allow for rapid sorting of fluorescently labeled cells, and iv) cell culturing mi- crochamber traps to capture non-adherent cell types [8, 19, 26]. These components can be easily integrated into the same device allowing for previously unattainable levels of on chip fluid handling, component integration, and assay parallization [47, 68, 74, 77, 107]. High-throughput MSL microfluidic technologies have recently been developed for various biochemical assays including protein crystallization [46, 47], PCR [75, 82, 114], mRNA analysis [74, 75, 114], and protein-DNA interaction analysis [33, 70]. Cellular analyses are following suit, and a small number of cell culture technologies have been presented [4, 44]. Further development is required how- ever, before cell based microfluidic assays fulfill the promise of high-throughput analysis of live single cells under complex environments. One area of needed development is in the ability to conduct single cell studies in time-varying stim- uli in a high-throughput manner. In chapter 2 I present such technology that, for the first time, combines the microfluidic ability to rapidly exchange environmen- tal conditions with MSL parallelization. We applied this technology to the high- throughput dynamic study of the mating response pathway in yeast under complex time-varying environmental conditions. In addition, in Appendix B I describe an extension of this technology that our group has developed for the tracking of indi- vidual cells over time. 12 1.5 Systems Genetics 1.5.1 Genetic Analysis of Signaling Pathways With the advent of genome sequencing came the ability to obtain a global ’parts lists’ of an organism [66]. Coupling this to high-throughput phenotypic screens, collections of genes involved in a system of interest can be efficiently obtained [81]. Understanding how these genes function as networks to process cellular informa- tion is now the more challenging aspect of cell biology and methods to generate physical and functional networks are needed. High-throughput strategies for test- ing physical connections between biomolecules have recently emerged and include technologies like yeast two hybrid [58, 111], mass spectrometry [36], ChIP-on- chip [90], and protein arrays [88]. These techniques have generated dense physical interaction maps and provide possible paths of information flow through a bio- logical system of interest. Although topologically informative, these maps do not however detail how cellular networks actually process information. Further tech- niques are needed to delineate the functional relationships between components of these physical interaction maps. One such technique is the use of genetic interactions. Genetic interactions describe the phenotypic outcome of dual gene perturbations, and can be used to functionally relate gene pairs. Combined with an understanding of the underlying molecular network, these functional relationships can dissect flows of information through complex biological systems [18]. A genetic interaction comprises phe- notype measurements from four genotypes: the reference genotype (WT), a single gene perturbation (A), a second single gene perturbation (B), and the dual gene per- 13 turbation (AB). The relative ordering of these four measurements determines the type of genetic interaction. For example, a commonly studied interaction called ’synthetic’, describes two genes that when perturbed simultaneously results in a strong phenotype not seen if either gene is perturbed in isolation. It has been pro- posed that this interaction can indicate that the genes are involved in the same process but in parallel redundant pathways, indicating gene buffering [11]. A sec- ond important example is the epistatic genetic interaction, which can be defined in two different ways. In the first definition, the term epistasis refers to a statistical deviation of the double mutant from what would be expected given that the two genes act in independent processes. In this way an epistatic interaction describes one of three functional digenic relationships in that the double mutant phenotype is measured to be: i) more severe than expected (aggregating interaction), ii) less severe than expected (alleviating interaction), and iii) as expected (no interaction). It has been shown that these different epistatic types can inform information flow between cellular processes [99]. In the second definition, the term epistasis refers to the masking of one genetic perturbation by a second (the double perturbation has the same outcome as the masking single perturbation). In this classic genetic de- scription, this relationship indicates that the gene products act in a common process with one ’upstream’ of the second. In this way an epistatic interaction is directional and indicates information flow. Cell biologists have used this second definition for decades in focused low-througput analyses [48]. 14 1.5.2 Genetic Interaction Networks In recent work, high-throughput techniques have been used to generate networks of genetic interactions. The nodes of these networks indicate genes of interest and edges indicate functional relationships between gene pairs. Comprehensive anal- ysis of genetic interaction relationships within a biomolecular system promises to deliniate the functional organization of biomolecules and pathways. Whereas single genetic interactions describe functional relationships between gene pairs, networks of genetic interactions describe functional relationships between gene systems. Genetic network analysis requires the high-throughput acquisition of gene in- teraction data, and many approaches have been developed. These methods vary in the type and precision of the interaction, resulting in the construction of var- ious genetic network types. For example, in Tong et al. [108] a method called the synthetic genetic array (SGA) approach was described for systematically test- ing gene pairs for synthetic interactions. SGA uses a series of mating and meiotic recombination steps to take an input set of single gene perturbations to generate the full set of double perturbations. Synthetic interactions were recorded when double mutant fitness was much reduced as compared to the two single mutant counterparts. This method can be fully automated using fluid handling robotics, and in a follow up study [109] a large ≈1000 node ≈4000 edge synthetic lethal interaction network was constructed. Analysis of this network allowed for the al- gorithmic grouping of genes into functionally related modules and gave insights to the global structure of genetic interaction networks. Diploid-based synthetic 15 lethality analysis on microarrays (dSLAM) is a related method [85] where instead of growing yeast in spots on plates the entire set of mutants are grown in bulk in liquid. To demultiplex individual mutants from the bulk culture, genome spe- cific barcodes are used and growth rates are quantified with genomic microarrays. Pan et al. used this method to construct a large synthetic lethal DNA integrity network in Saccharomyces cerevisiae, allowing for the stratification of genes into distinct functional modules [84]. In a more quantitative approach, Schuldiner et al. constructed epistatic miniarray profiles, (E-MAPS), using growth rates as a quantitative phenotype [98]. This method allowed for the identification of func- tional sub-modules of the early secretory [24, 98] and chromosome manipulation [23] systems in yeast. And Segre et al used similar methods to identify functional modules in yeast metabolism [99]. Lastly Drees et al., demonstrated a more com- plex description of genetic interactions, defining nine discrete types, four of which were directional [30]. This methodology was applied to the filamentation signal- ing pathway in Saccharomyces cerevisiae, and by identifying mutually informative patterns of genetic interaction between perturbations, genes could be placed in a high-resolution mapping of filamentation signaling. Taken together, these studies demonstrate the ability to rapidly construct dense networks of functional gene relationships, and initial analyses demonstrate the richness of information that they contain. Due to the complexity of these net- works, experimentation techniques need to be coupled with computational algo- rithms for the extraction of meaningful biological information. Preliminary analy- ses used simple techniques to study network topology and cluster functional mod- ules [23, 24, 30, 98, 99, 109]. Although informative, much opportunity exists to 16 develop sophisticated computation methods to further extract meaningful biolog- ical data from these information rich genetic interaction networks. One area of considerable potential for methods comes from the broad field of network analysis. 1.5.3 Network Analysis The creation of large networks of biological data has encouraged parallel advance- ments in complex network analysis. For example, pivotal studies analyzing net- works of physical interaction data have uncovered organizational structures of biomolecular systems [6, 7, 89, 92]. In these studies, Barabasi and colleagues iden- tified that biological networks exhibit a scale free architecture [60], similar to that seen in the internet and social networks. Although these studies do not give direct insight into molecular functions of specific molecules or processes, they provide a foundation for further complex systems analysis and uncover general properties of biological networks including modularity [89, 92] and robustness [2, 64]. Further work by Alon and colleagues provided tools to identify network sub-structures that represent the basic building blocks of complex networks, called network motifs. Network motifs are small network structures that occur more often than expected by random, implying relevant function [79, 101]. The functions of these network substructures are directly testable [73] and network motifs have been found in bac- teria [101], yeast [67], and worms [79]. Quantitative modeling and experimentation has revealed specific functions of these motifs and highlighted the importance of their use in biological systems. For example, in the context of protein-DNA net- works, the coherent feed-forward motif uses multiple time scales to filter incoming high-frequency transient signals[72, 73]. Further, recent studies have found certain 17 network motifs dysregulated in human disease [27]. The tools developed to analyze networks of physical interaction data can be ap- plied to genetic interaction networks. For example, in Zhang et al [120], the idea of network motifs was used to integrate genetic interaction data with other interaction types including protein-protein, protein-DNA, sequence homology, and expression correlation interactions. This method allowed a single edge-type (synthetic) ge- netic interaction network to be placed in context of the the underlying molecular system, and higher-order network structures were identified. Further network stud- ies of more complex genetic interaction network types holds the potential to de- lineate high resolution maps of information flows through local regulatory systems [18]. In Chapter 3, I present work using network motifs to identify biologically in- formative patterns of genetic interaction in the dense multi-mode genetic network of Drees et al. [30]. This analysis identified network structures that are biologi- cally informative, and helped to find paths of information flow through the circuitry regulating filamentation signaling. 18 Bibliography [1] Vinay V Abhyankar, Mary A Lokuta, Anna Huttenlocher, and David J Beebe. Characterization of a membrane-based gradient generator for use in cell-signaling studies. Lab on a chip, 6(3):389–93, Mar 2006. [2] R Albert, H Jeong, and A Barabasi. Error and attack tolerance of complex networks. Nature, 406(6794):378–82, Jul 2000. [3] Jessica Andersson, David M Simpson, Maosong Qi, Yunmei Wang, and Elaine A Elion. Differential input by ste5 scaffold and msg5 phosphatase route a mapk cascade to multiple outcomes. EMBO J, 23(13):2564–76, Jul 2004. [4] Frederick K Balagaddé, Lingchong You, Carl L Hansen, Frances H Arnold, and Stephen R Quake. Long-term monitoring of bacteria undergoing pro- grammed population control in a microchemostat. Science, 309(5731):137– 40, Jul 2005. [5] Marie Z Bao, Monica A Schwartz, Greg T Cantin, John R Yates, and Hiten D Madhani. Pheromone-dependent destruction of the tec1 transcription factor 19 is required for map kinase signaling specificity in yeast. Cell, 119(7):991– 1000, Dec 2004. [6] A Barabasi and R Albert. Emergence of scaling in random networks. Sci- ence, 286(5439):509–12, Oct 1999. [7] Albert-László Barabási and Zoltán N Oltvai. Network biology: understand- ing the cell’s functional organization. Nat Rev Genet, 5(2):101–13, Feb 2004. [8] M Bennett, W Pang, N Ostroff, B Baumgartner, S Nayak, L Tsimring, and J Hasty. Metabolic gene regulation in a dynamically changing environment. Nature, 454:1119–1122, Jul 2008. [9] Robert G Blazej, Palani Kumaresan, and Richard A Mathies. Microfab- ricated bioprocessor for integrated nanoliter-scale sanger dna sequencing. Proc Natl Acad Sci USA, 103(19):7240–5, May 2006. [10] Richard Bonneau, Marc T Facciotti, David J Reiss, Amy K Schmid, Min Pan, Amardeep Kaur, Vesteinn Thorsson, Paul Shannon, Michael H John- son, J. Christopher Bare, William Longabaugh, Madhavi Vuthoori, Kenia Whitehead, Aviv Madar, Lena Suzuki, Tetsuya Mori, Dong-Eun Chang, Jo- celyne DiRuggiero, Carl H Johnson, Leroy Hood, and Nitin S Baliga. A predictive model for transcriptional control of physiology in a free living cell. Cell, 131:1354–1365, Jan 2007. [11] Charles Boone, Howard Bussey, and Brenda J Andrews. Exploring ge- netic interactions and networks with yeast. Nat Rev Genet, 8(6):437–49, Jun 2007. 20 [12] A C Burkholder and L H Hartwell. The yeast alpha-factor receptor: struc- tural properties deduced from the sequence of the ste2 gene. Nucleic Acids Res, 13(23):8463–75, Dec 1985. [13] M A Burns, B N Johnson, S N Brahmasandra, K Handique, J R Webster, M Krishnan, T S Sammarco, P M Man, D Jones, D Heldsinger, C H Mas- trangelo, and D T Burke. An integrated nanoliter dna analysis device. Sci- ence, 282(5388):484–7, Oct 1998. [14] G Steven Burrill. BiotechBook2008. Apr 2008. [15] A C Butty, P M Pryciak, L S Huang, I Herskowitz, and M Peter. The role of far1p in linking the heterotrimeric g protein to polarity establishment pro- teins during yeast mating. Science, 282(5393):1511–6, Nov 1998. [16] Long Cai, Nir Friedman, and X Sunney Xie. Stochastic protein expression in individual cells at the single molecule level. Nature, 440(7082):358–62, Mar 2006. [17] Dino Di Carlo, Liz Y Wu, and Luke P Lee. Dynamic single cell culture array. Lab on a chip, 6(11):1445–9, Nov 2006. [18] Gregory W Carter, Susanne Prinz, Christine Neou, J Patrick Shelby, Bruz Marzolf, Vesteinn Thorsson, and Timothy Galitski. Prediction of phenotype and gene expression for combinations of mutations. Mol Syst Biol, 3:96, Jan 2007. [19] Gilles Charvin, Frederick R Cross, and Eric D Siggia. A microfluidic device 21 for temporally controlled gene expression and long-term fluorescent imag- ing in unperturbed dividing yeast cells. PLoS ONE, 3(1):e1468, Jan 2008. [20] Yanqing Chen, Jun Zhu, Pek Yee Lum, Xia Yang, Shirly Pinto, Douglas J MacNeil, Chunsheng Zhang, John Lamb, Stephen Edwards, Solveig K Sieberts, Amy Leonardson, Lawrence W Castellini, Susanna Wang, Marie- France Champy, Bin Zhang, Valur Emilsson, Sudheer Doss, Anatole Ghaz- alpour, Steve Horvath, Thomas A Drake, Aldons J Lusis, and Eric E Schadt. Variations in dna elucidate molecular networks that cause disease. Nature, 452(7186):429–35, Mar 2008. [21] Song Chou, Lan Huang, and Haoping Liu. Fus3-regulated tec1 degrada- tion through scfcdc4 determines mapk signaling specificity during mating in yeast. Cell, 119(7):981–90, Dec 2004. [22] Song Chou, Su Zhao, You Song, Haoping Liu, and Qing Nie. Fus3- triggered tec1 degradation modulates mating transcriptional output during the pheromone response. Mol Syst Biol, 4:212, Jan 2008. [23] Sean R Collins, Kyle M Miller, Nancy L Maas, Assen Roguev, Jeffrey Fill- ingham, Clement S Chu, Maya Schuldiner, Marinella Gebbia, Judith Recht, Michael Shales, Huiming Ding, Hong Xu, Junhong Han, Kristin Ingvars- dottir, Benjamin Cheng, Brenda Andrews, Charles Boone, Shelley L Berger, Phil Hieter, Zhiguo Zhang, Grant W Brown, C James Ingles, Andrew Emili, C David Allis, David P Toczyski, Jonathan S Weissman, Jack F Greenblatt, and Nevan J Krogan. Functional dissection of protein complexes involved 22 in yeast chromosome biology using a genetic interaction map. Nature, 446(7137):806–10, Apr 2007. [24] Sean R Collins, Maya Schuldiner, Nevan J Krogan, and Jonathan S Weiss- man. A strategy for extracting and analyzing large-scale quantitative epistatic interaction data. Genome Biol, 7(7):R63, Jan 2006. [25] Alejandro Colman-Lerner, Andrew Gordon, Eduard Serra, Tina Chin, Orna Resnekov, Drew Endy, C Gustavo Pesce, and Roger Brent. Regulated cell- to-cell variation in a cell-fate decision system. Nature, 437(7059):699–706, Sep 2005. [26] Scott Cookson, Natalie Ostroff, Wyming Lee Pang, Dmitri Volfson, and Jeff Hasty. Monitoring dynamics of single-cell gene expression over multiple cell cycles. Mol Syst Biol, 1:2005.0024, Jan 2005. [27] Qinghua Cui, Yun Ma, Maria Jaramillo, Hamza Bari, Arif Awan, Song Yang, Simo Zhang, Lixue Liu, Meng Lu, Maureen O’Connor-McCourt, En- rico O Purisima, and Edwin Wang. A map of human cancer signaling. Mol Syst Biol, 3:152, Jan 2007. [28] P J Cullen and G F Sprague. Glucose depletion causes haploid invasive growth in yeast. Proc Natl Acad Sci USA, 97(25):13619–24, Dec 2000. [29] J W Dolan, C Kirkman, and S Fields. The yeast ste12 protein binds to the dna sequence mediating pheromone induction. Proc Natl Acad Sci USA, 86(15):5703–7, Aug 1989. [30] Becky L Drees, Vesteinn Thorsson, Gregory W Carter, Alexander W Rives, 23 Marisa Z Raymond, Iliana Avila-Campillo, Paul Shannon, and Timothy Gal- itski. Derivation of genetic interaction networks from quantitative phenotype data. Genome Biol, 6(4):R38, Jan 2005. [31] F Drogen, S M O’Rourke, V M Stucke, M Jaquenoud, A M Neiman, and M Peter. Phosphorylation of the mekk ste11p by the pak-like kinase ste20p is required for map kinase signaling in vivo. Curr Biol, 10(11):630–9, Jun 2000. [32] DC Duffy, JC McDonald, OJA Schueller, and GM Whitesides. Rapid pro- totyping of microfluidic systems in poly(dimethylsiloxane). Anal Chem, 70:4974–4984, Jan 1998. [33] S Einav, D Gerber, P Bryson, E Sklan, M Elazar, S Maerkl, J Glenn, and S Quake. Discovery of a hepatitis c target and its pharmacological inhibitors by microfluidic affinity analysis. Nat Biotechnol, 26:1019–1027, Aug 2008. [34] Jamil El-Ali, Peter K Sorger, and Klavs F Jensen. Cells on chips. Nature, 442(7101):403–11, Jul 2006. [35] Elaine A Elion, Maosong Qi, and Weidong Chen. Signal transduction. sig- naling specificity in yeast. Science, 307(5710):687–8, Feb 2005. [36] Rob M Ewing, Peter Chu, Fred Elisma, Hongyan Li, Paul Taylor, Shane Climie, Linda McBroom-Cerajewski, Mark D Robinson, Liam O’Connor, Michael Li, Rod Taylor, Moyez Dharsee, Yuen Ho, Adrian Heilbut, Lynda Moore, Shudong Zhang, Olga Ornatsky, Yury V Bukhman, Martin Ethier, Yinglun Sheng, Julian Vasilescu, Mohamed Abu-Farha, Jean-Philippe Lam- bert, Henry S Duewel, Ian I Stewart, Bonnie Kuehl, Kelly Hogue, Karen 24 Colwill, Katharine Gladwish, Brenda Muskat, Robert Kinach, Sally-Lin Adams, Michael F Moran, Gregg B Morin, Thodoros Topaloglou, and Daniel Figeys. Large-scale mapping of human protein-protein interactions by mass spectrometry. Mol Syst Biol, 3:89, Jan 2007. [37] Laura J Flatauer, Sheena F Zadeh, and Lee Bardwell. Mitogen-activated protein kinases with distinct requirements for ste5 scaffolding influence sig- naling specificity in saccharomyces cerevisiae. Mol Cell Biol, 25(5):1793– 803, Mar 2005. [38] Mark R Flory, Hookeun Lee, Richard Bonneau, Parag Mallick, Kyle Serikawa, David R Morris, and Ruedi Aebersold. Quantitative proteomic analysis of the budding yeast cell cycle using acid-cleavable isotope-coded affinity tag reagents. Proteomics, 6(23):6146–57, Dec 2006. [39] A Y Fu, C Spence, A Scherer, F H Arnold, and S R Quake. A microfab- ricated fluorescence-activated cell sorter. Nat Biotechnol, 17(11):1109–11, Nov 1999. [40] Anne Y Fu, Hou-Pu Chou, Charles Spence, Frances H Arnold, and Stephen R Quake. An integrated microfabricated cell sorter. Anal Chem, 74(11):2451–7, Jun 2002. [41] A Gartner, K Nasmyth, and G Ammerer. Signal transduction in saccha- romyces cerevisiae requires tyrosine and threonine phosphorylation of fus3 and kss1. Genes Dev, 6(7):1280–92, Jul 1992. [42] V Gavrias, A Andrianopoulos, C J Gimeno, and W E Timberlake. Saccha- 25 romyces cerevisiae tec1 is required for pseudohyphal growth. Mol Micro- biol, 19(6):1255–63, Mar 1996. [43] C J Gimeno, P O Ljungdahl, C A Styles, and G R Fink. Unipolar cell divisions in the yeast s. cerevisiae lead to filamentous growth: regulation by starvation and ras. Cell, 68(6):1077–90, Mar 1992. [44] Rafael Gómez-Sjöberg, Anne A Leyrat, Dana M Pirone, Christopher S Chen, and Stephen R Quake. Versatile, fully automated, microfluidic cell culture system. Anal Chem, 79(22):8557–63, Nov 2007. [45] M C Gustin, J Albertyn, M Alexander, and K Davenport. Map kinase pathways in the yeast saccharomyces cerevisiae. Microbiol Mol Biol Rev, 62(4):1264–300, Dec 1998. [46] Carl L Hansen, Emmanuel Skordalakes, James M Berger, and Stephen R Quake. A robust and scalable microfluidic metering method that allows protein crystal growth by free interface diffusion. Proc Natl Acad Sci USA, 99(26):16531–6, Dec 2002. [47] Carl L Hansen, Morten O A Sommer, and Stephen R Quake. Systematic investigation of protein phase behavior with a microfluidic formulator. Proc Natl Acad Sci USA, 101(40):14431–6, Oct 2004. [48] L H Hartwell, J Culotti, J R Pringle, and B J Reid. Genetic control of the cell division cycle in yeast. Science, 183(120):46–51, Jan 1974. [49] Pascal Hersen, Megan N McClean, L Mahadevan, and Sharad Ramanathan. 26 Signal processing by the hog map kinase pathway. Proc Natl Acad Sci USA, 105(20):7165–70, May 2008. [50] J E Hirschman, G S De Zutter, W F Simonds, and D D Jenness. The g beta gamma complex of the yeast pheromone response pathway. subcellular fractionation and protein-protein interactions. J Biol Chem, 272(1):240–8, Jan 1997. [51] Jong Wook Hong, Vincent Studer, Giao Hang, W French Anderson, and Stephen R Quake. A nanoliter-scale nucleic acid processor with parallel architecture. Nat Biotechnol, 22(4):435–9, Apr 2004. [52] Xiaoyuan Hu, Paul H Bessette, Jiangrong Qian, Carl D Meinhart, Patrick S Daugherty, and Hyongsok T Soh. Marker-specific sorting of rare cells using dielectrophoresis. Proc Natl Acad Sci USA, 102(44):15757–61, Nov 2005. [53] Zhishan Hua, Yongmei Xia, Onnop Srivannavit, Jean-Marie Rouillard, Xi- aochuan Zhou, Xiaolian Gao, and Erdogan Gulari. A versatile microreac- tor platform featuring a chemical-resistant microvalve array for addressable multiplex syntheses and assays. J Micromech Microeng, 16:1433–1443, Jan 2006. [54] Bo Huang, Hongkai Wu, Devaki Bhaya, Arthur Grossman, Sebastien Granier, Brian K Kobilka, and Richard N Zare. Counting low-copy num- ber proteins in a single cell. Science, 315(5808):81–4, Jan 2007. [55] Ying Huang, Sunghae Joo, Melanie Duhon, Michael Heller, Bruce Wallace, and Xiao Xu. Dielectrophoretic cell separation and gene expression profiling on microelectronic chip arrays. Anal Chem, 74(14):3362–71, Jul 2002. 27 [56] Paul J Hung, Philip J Lee, Poorya Sabounchi, Robert Lin, and Luke P Lee. Continuous perfusion microfluidic cell culture array for high-throughput cell-based assays. Biotechnol Bioeng, 89(1):1–8, Jan 2005. [57] T Ideker, V Thorsson, J A Ranish, R Christmas, J Buhler, J K Eng, R Bum- garner, D R Goodlett, R Aebersold, and L Hood. Integrated genomic and proteomic analyses of a systematically perturbed metabolic network. Sci- ence, 292(5518):929–34, May 2001. [58] T Ito, T Chiba, R Ozawa, M Yoshida, M Hattori, and Y Sakaki. A compre- hensive two-hybrid analysis to explore the yeast protein interactome. Proc Natl Acad Sci USA, 98(8):4569–74, Apr 2001. [59] Noo Li Jeon, Harihara Baskaran, Stephan K W Dertinger, George M White- sides, Livingston Van de Water, and Mehmet Toner. Neutrophil chemotaxis in linear and complex gradients of interleukin-8 formed in a microfabricated device. Nat Biotechnol, 20(8):826–30, Aug 2002. [60] H Jeong, B Tombor, R Albert, Z N Oltvai, and A L Barabási. The large-scale organization of metabolic networks. Nature, 407(6804):651–4, Oct 2000. [61] Emil P Kartalov, Jiang F Zhong, Axel Scherer, Stephen R Quake, Clive R Taylor, and W French Anderson. High-throughput multi-antigen microflu- idic fluorescence immunoassays. Biotechniques, 40(1):85–90, Jan 2006. [62] Hiroaki Kitano. Computational systems biology. Nature, 420(6912):206– 10, Nov 2002. 28 [63] Hiroaki Kitano. Systems biology: a brief overview. Science, 295(5560):1662–4, Mar 2002. [64] Hiroaki Kitano. Biological robustness. Nat Rev Genet, 5(11):826–37, Nov 2004. [65] Bente Kofahl and Edda Klipp. Modelling the dynamics of the yeast pheromone pathway. Yeast, 21(10):831–50, Jul 2004. [66] E S Lander, L M Linton, B Birren, C Nusbaum, M C Zody, J Baldwin, K Devon, K Dewar, M Doyle, W FitzHugh, R Funke, D Gage, K Harris, A Heaford, J Howland, L Kann, J Lehoczky, R LeVine, P McEwan, K McK- ernan, J Meldrim, J P Mesirov, C Miranda, W Morris, J Naylor, C Raymond, M Rosetti, R Santos, A Sheridan, C Sougnez, N Stange-Thomann, N Sto- janovic, A Subramanian, D Wyman, J Rogers, J Sulston, R Ainscough, S Beck, D Bentley, J Burton, C Clee, N Carter, A Coulson, R Deadman, P Deloukas, A Dunham, I Dunham, R Durbin, L French, D Grafham, S Gre- gory, T Hubbard, S Humphray, A Hunt, M Jones, C Lloyd, A McMur- ray, L Matthews, S Mercer, S Milne, J C Mullikin, A Mungall, R Plumb, M Ross, R Shownkeen, S Sims, R H Waterston, R K Wilson, L W Hillier, J D McPherson, M A Marra, E R Mardis, L A Fulton, A T Chinwalla, K H Pepin, W R Gish, S L Chissoe, M C Wendl, K D Delehaunty, T L Miner, A Delehaunty, J B Kramer, L L Cook, R S Fulton, D L Johnson, P J Minx, S W Clifton, T Hawkins, E Branscomb, P Predki, P Richardson, S Wen- ning, T Slezak, N Doggett, J F Cheng, A Olsen, S Lucas, C Elkin, E Uber- bacher, M Frazier, R A Gibbs, D M Muzny, S E Scherer, J B Bouck, E J Sodergren, K C Worley, C M Rives, J H Gorrell, M L Metzker, S L Nay- 29 lor, R S Kucherlapati, D L Nelson, G M Weinstock, Y Sakaki, A Fujiyama, M Hattori, T Yada, A Toyoda, T Itoh, C Kawagoe, H Watanabe, Y Totoki, T Taylor, J Weissenbach, R Heilig, W Saurin, F Artiguenave, P Brottier, T Bruls, E Pelletier, C Robert, P Wincker, D R Smith, L Doucette-Stamm, M Rubenfield, K Weinstock, H M Lee, J Dubois, A Rosenthal, M Platzer, G Nyakatura, S Taudien, A Rump, H Yang, J Yu, J Wang, G Huang, J Gu, L Hood, L Rowen, A Madan, S Qin, R W Davis, N A Federspiel, A P Abola, M J Proctor, R M Myers, J Schmutz, M Dickson, J Grimwood, D R Cox, M V Olson, R Kaul, C Raymond, N Shimizu, K Kawasaki, S Minoshima, G A Evans, M Athanasiou, R Schultz, B A Roe, F Chen, H Pan, J Ramser, H Lehrach, R Reinhardt, W R McCombie, M de la Bastide, N Dedhia, H Blöcker, K Hornischer, G Nordsiek, R Agarwala, L Aravind, J A Bai- ley, A Bateman, S Batzoglou, E Birney, P Bork, D G Brown, C B Burge, L Cerutti, H C Chen, D Church, M Clamp, R R Copley, T Doerks, S R Eddy, E E Eichler, T S Furey, J Galagan, J G Gilbert, C Harmon, Y Hayashizaki, D Haussler, H Hermjakob, K Hokamp, W Jang, L S Johnson, T A Jones, S Kasif, A Kaspryzk, S Kennedy, W J Kent, P Kitts, E V Koonin, I Korf, D Kulp, D Lancet, T M Lowe, A McLysaght, T Mikkelsen, J V Moran, N Mulder, V J Pollara, C P Ponting, G Schuler, J Schultz, G Slater, A F Smit, E Stupka, J Szustakowski, D Thierry-Mieg, J Thierry-Mieg, L Wag- ner, J Wallis, R Wheeler, A Williams, Y I Wolf, K H Wolfe, S P Yang, R F Yeh, F Collins, M S Guyer, J Peterson, A Felsenfeld, K A Wetterstrand, A Patrinos, M J Morgan, P de Jong, J J Catanese, K Osoegawa, H Shizuya, S Choi, Y J Chen, J Szustakowki, and International Human Genome Se- quencing Consortium. Initial sequencing and analysis of the human genome. 30 Nature, 409(6822):860–921, Feb 2001. [67] Tong Ihn Lee, Nicola J Rinaldi, François Robert, Duncan T Odom, Ziv Bar-Joseph, Georg K Gerber, Nancy M Hannett, Christopher T Harbi- son, Craig M Thompson, Itamar Simon, Julia Zeitlinger, Ezra G Jennings, Heather L Murray, D Benjamin Gordon, Bing Ren, John J Wyrick, Jean- Bosco Tagne, Thomas L Volkert, Ernest Fraenkel, David K Gifford, and Richard A Young. Transcriptional regulatory networks in saccharomyces cerevisiae. Science, 298(5594):799–804, Oct 2002. [68] Jian Liu, Carl Hansen, and Stephen R Quake. Solving the ”world-to-chip” interface problem with a microfluidic matrix. Anal Chem, 75(18):4718–23, Sep 2003. [69] D J Lockhart, H Dong, M C Byrne, M T Follettie, M V Gallo, M S Chee, M Mittmann, C Wang, M Kobayashi, H Horton, and E L Brown. Expres- sion monitoring by hybridization to high-density oligonucleotide arrays. Nat Biotechnol, 14(13):1675–80, Dec 1996. [70] Sebastian J Maerkl and Stephen R Quake. A systems approach to mea- suring the binding energy landscapes of transcription factors. Science, 315(5809):233–7, Jan 2007. [71] Seth Maleri, Qingyuan Ge, Elizabeth A Hackett, Yuqi Wang, Henrik G Dohlman, and Beverly Errede. Persistent activation by constitutive ste7 pro- motes kss1-mediated invasive growth but fails to support fus3-dependent mating in yeast. Mol Cell Biol, 24(20):9221–38, Oct 2004. 31 [72] S Mangan and U Alon. Structure and function of the feed-forward loop network motif. Proc Natl Acad Sci USA, 100(21):11980–5, Oct 2003. [73] S Mangan, A Zaslaver, and U Alon. The coherent feedforward loop serves as a sign-sensitive delay element in transcription networks. J Mol Biol, 334(2):197–204, Nov 2003. [74] Joshua S Marcus, W French Anderson, and Stephen R Quake. Microfluidic single-cell mrna isolation and analysis. Anal Chem, 78(9):3084–9, May 2006. [75] Joshua S Marcus, W French Anderson, and Stephen R Quake. Parallel pico- liter rt-pcr assays using microfluidics. Anal Chem, 78(3):956–8, Feb 2006. [76] K Matsumoto, I Uno, Y Oshima, and T Ishikawa. Isolation and characteri- zation of yeast mutants deficient in adenylate cyclase and camp-dependent protein kinase. Proc Natl Acad Sci USA, 79(7):2355–9, Apr 1982. [77] Jessica Melin and Stephen R Quake. Microfluidic large-scale integration: the evolution of design rules for biological automation. Annual review of biophysics and biomolecular structure, 36:213–31, Jan 2007. [78] Jerome T Mettetal, Dale Muzzey, Carlos Gómez-Uribe, and Alexander van Oudenaarden. The frequency dependence of osmo-adaptation in saccha- romyces cerevisiae. Science, 319(5862):482–4, Jan 2008. [79] R Milo, S Shen-Orr, S Itzkovitz, N Kashtan, D Chklovskii, and U Alon. Network motifs: simple building blocks of complex networks. Science, 298(5594):824–7, Oct 2002. 32 [80] Fred Naider and Jeffrey M Becker. The alpha-factor mating pheromone of saccharomyces cerevisiae: a model for studying the interaction of peptide hormones and g protein-coupled receptors. Peptides, 25(9):1441–63, Sep 2004. [81] Rammohan Narayanaswamy, Wei Niu, Alexander D Scouras, G Traver Hart, Jonathan Davies, Andrew D Ellington, Vishwanath R Iyer, and Edward M Marcotte. Systematic profiling of cellular phenotypes with spotted cell mi- croarrays reveals mating-pheromone response genes. Genome Biol, 7(1):R6, Jan 2006. [82] Elizabeth A Ottesen, Jong Wook Hong, Stephen R Quake, and Jared R Lead- better. Microfluidic digital pcr enables multigene analysis of individual en- vironmental bacteria. Science, 314(5804):1464–7, Dec 2006. [83] Saurabh Paliwal, Pablo A Iglesias, Kyle Campbell, Zoe Hilioti, Alex Grois- man, and Andre Levchenko. Mapk-mediated bimodal gene expression and adaptive gradient sensing in yeast. Nature, 446(7131):46–51, Mar 2007. [84] Xuewen Pan, Ping Ye, Daniel S Yuan, Xiaoling Wang, Joel S Bader, and Jef D Boeke. A dna integrity network in the yeast saccharomyces cerevisiae. Cell, 124(5):1069–81, Mar 2006. [85] Xuewen Pan, Daniel S Yuan, Dong Xiang, Xiaoling Wang, Sharon Sookhai- Mahadeo, Joel S Bader, Philip Hieter, Forrest Spencer, and Jef D Boeke. A robust toolkit for functional profiling of the yeast genome. Mol Cell, 16(3):487–96, Nov 2004. 33 [86] MJ Powers, K Domansky, MR Kaazempur-Mofrad, A Kalezi, A Capitano, A Upadhyaya, P Kurzawski, KE Wack, DB Stolz, R Kamm, and LG Griffith. A microfabricated array bioreactor for perfused 3d liver culture. Biotechnol Bioeng, 78:257–269, Jan 2002. [87] J A Printen and G F Sprague. Protein-protein interactions in the yeast pheromone response pathway: Ste5p interacts with all members of the map kinase cascade. Genetics, 138(3):609–19, Nov 1994. [88] Jason Ptacek, Geeta Devgan, Gregory Michaud, Heng Zhu, Xiaowei Zhu, Joseph Fasolo, Hong Guo, Ghil Jona, Ashton Breitkreutz, Richelle Sopko, Rhonda R McCartney, Martin C Schmidt, Najma Rachidi, Soo-Jung Lee, Angie S Mah, Lihao Meng, Michael J R Stark, David F Stern, Claudio De Virgilio, Mike Tyers, Brenda Andrews, Mark Gerstein, Barry Schweitzer, Paul F Predki, and Michael Snyder. Global analysis of protein phosphory- lation in yeast. Nature, 438(7068):679–84, Dec 2005. [89] E Ravasz, A L Somera, D A Mongru, Z N Oltvai, and A L Barabási. Hierarchical organization of modularity in metabolic networks. Science, 297(5586):1551–5, Aug 2002. [90] B Ren, F Robert, J J Wyrick, O Aparicio, E G Jennings, I Simon, J Zeitlinger, J Schreiber, N Hannett, E Kanin, T L Volkert, C J Wilson, S P Bell, and R A Young. Genome-wide location and function of dna binding proteins. Science, 290(5500):2306–9, Dec 2000. [91] N Rhodes, L Connell, and B Errede. Ste11 is a protein kinase required for 34 cell-type-specific transcription and signal transduction in yeast. Genes Dev, 4(11):1862–74, Nov 1990. [92] Alexander W Rives and Timothy Galitski. Modular organization of cellular networks. Proc Natl Acad Sci USA, 100(3):1128–33, Feb 2003. [93] R L Roberts and G R Fink. Elements of a single map kinase cascade in saccharomyces cerevisiae mediate two developmental programs in the same cell type: mating and invasive growth. Genes Dev, 8(24):2974–85, Dec 1994. [94] L S Robertson and G R Fink. The three yeast a kinases have specific signaling functions in pseudohyphal growth. Proc Natl Acad Sci USA, 95(23):13783–7, Nov 1998. [95] S Rupp, E Summers, H J Lo, H Madhani, and G Fink. Map kinase and camp filamentation signaling pathways converge on the unusually large promoter of the yeast flo11 gene. EMBO J, 18(5):1257–69, Mar 1999. [96] Kiichi Sato, Maho Yamanaka, Hiroko Takahashi, Manabu Tokeshi, Hiroko Kimura, and Takehiko Kitamori. Microchip-based immunoassay system with branching multichannels for simultaneous determination of interferon- gamma. Electrophoresis, 23(5):734–9, Mar 2002. [97] M Schena, D Shalon, R W Davis, and P O Brown. Quantitative monitoring of gene expression patterns with a complementary dna microarray. Science, 270(5235):467–70, Oct 1995. [98] Maya Schuldiner, Sean R Collins, Natalie J Thompson, Vladimir Denic, 35 Arunashree Bhamidipati, Thanuja Punna, Jan Ihmels, Brenda Andrews, Charles Boone, Jack F Greenblatt, Jonathan S Weissman, and Nevan J Kro- gan. Exploration of the function and organization of the yeast early secretory pathway through an epistatic miniarray profile. Cell, 123(3):507–19, Nov 2005. [99] Daniel Segrè, Alexander Deluna, George M Church, and Roy Kishony. Modular epistasis in yeast metabolism. Nat Genet, 37(1):77–83, Jan 2005. [100] Danying Shao, Wen Zheng, Wenjun Qiu, Qi Ouyang, and Chao Tang. Dy- namic studies of scaffold-dependent mating pathway in yeast. Biophys J, 91(11):3986–4001, Dec 2006. [101] Shai S Shen-Orr, Ron Milo, Shmoolik Mangan, and Uri Alon. Network motifs in the transcriptional regulation network of escherichia coli. Nat Genet, 31(1):64–8, May 2002. [102] Samuel K Sia and George M Whitesides. Microfluidic devices fabri- cated in poly(dimethylsiloxane) for biological studies. Electrophoresis, 24(21):3563–76, Nov 2003. [103] Shelly C Strickfaden, Matthew J Winters, Giora Ben-Ari, Rachel E Lamson, Mike Tyers, and Peter M Pryciak. A mechanism for cell-cycle regulation of map kinase signaling in a yeast differentiation pathway. Cell, 128(3):519– 31, Feb 2007. [104] Nicolas Szita, Paolo Boccazzi, Zhiyu Zhang, Patrick Boyle, Anthony J Sinskey, and Klavs F Jensen. Development of a multiplexed microbioreac- 36 tor system for high-throughput bioprocessing. Lab on a chip, 5(8):819–26, Aug 2005. [105] Satoe Takahashi and Peter M Pryciak. Membrane localization of scaffold proteins promotes graded signaling in the yeast map kinase cascade. Curr Biol, 18(16):1184–91, Aug 2008. [106] M A Teague, D T Chaleff, and B Errede. Nucleotide sequence of the yeast regulatory gene ste7 predicts a protein homologous to protein kinases. Proc Natl Acad Sci USA, 83(19):7371–5, Oct 1986. [107] Todd Thorsen, Sebastian J Maerkl, and Stephen R Quake. Microfluidic large-scale integration. Science, 298(5593):580–4, Oct 2002. [108] A H Tong, M Evangelista, A B Parsons, H Xu, G D Bader, N Pagé, M Robin- son, S Raghibizadeh, C W Hogue, H Bussey, B Andrews, M Tyers, and C Boone. Systematic genetic analysis with ordered arrays of yeast deletion mutants. Science, 294(5550):2364–8, Dec 2001. [109] Amy Hin Yan Tong, Guillaume Lesage, Gary D Bader, Huiming Ding, Hong Xu, Xiaofeng Xin, James Young, Gabriel F Berriz, Renee L Brost, Michael Chang, YiQun Chen, Xin Cheng, Gordon Chua, Helena Friesen, Debra S Goldberg, Jennifer Haynes, Christine Humphries, Grace He, Shamiza Hus- sein, Lizhu Ke, Nevan Krogan, Zhijian Li, Joshua N Levinson, Hong Lu, Patrice Ménard, Christella Munyana, Ainslie B Parsons, Owen Ryan, Raffi Tonikian, Tania Roberts, Anne-Marie Sdicu, Jesse Shapiro, Bilal Sheikh, Bernhard Suter, Sharyl L Wong, Lan V Zhang, Hongwei Zhu, Christopher G Burd, Sean Munro, Chris Sander, Jasper Rine, Jack Greenblatt, Matthias 37 Peter, Anthony Bretscher, Graham Bell, Frederick P Roth, Grant W Brown, Brenda Andrews, Howard Bussey, and Charles Boone. Global mapping of the yeast genetic interaction network. Science, 303(5659):808–13, Feb 2004. [110] M Tyers and B Futcher. Far1 and fus3 link the mating pheromone signal transduction pathway to three g1-phase cdc28 kinase complexes. Mol Cell Biol, 13(9):5659–69, Sep 1993. [111] P Uetz, L Giot, G Cagney, T A Mansfield, R S Judson, J R Knight, D Lock- shon, V Narayan, M Srinivasan, P Pochart, A Qureshi-Emili, Y Li, B God- win, D Conover, T Kalbfleisch, G Vijayadamodar, M Yang, M Johnston, S Fields, and J M Rothberg. A comprehensive analysis of protein-protein interactions in saccharomyces cerevisiae. Nature, 403(6770):623–7, Feb 2000. [112] M A Unger, H P Chou, T Thorsen, A Scherer, and S R Quake. Monolithic microfabricated valves and pumps by multilayer soft lithography. Science, 288(5463):113–6, Apr 2000. [113] Mark M Wang, Eugene Tu, Daniel E Raymond, Joon Mo Yang, Haichuan Zhang, Norbert Hagen, Bob Dees, Elinore M Mercer, Anita H Forster, Ilona Kariv, Philippe J Marchand, and William F Butler. Microfluidic sorting of mammalian cells by optical force switching. Nat Biotechnol, 23(1):83–7, Jan 2005. [114] Luigi Warren, David Bryder, Irving L Weissman, and Stephen R Quake. 38 Transcription factor profiling in individual hematopoietic progenitors by digital rt-pcr. Proc Natl Acad Sci USA, 103(47):17807–12, Nov 2006. [115] G M Whitesides, E Ostuni, S Takayama, X Jiang, and D E Ingber. Soft lithography in biology and biochemistry. Annual review of biomedical engi- neering, 3:335–73, Jan 2001. [116] C Wittenberg and S I Reed. Plugging it in: signaling circuits and the yeast cell cycle. Curr Opin Cell Biol, 8(2):223–30, Apr 1996. [117] Hongkai Wu, Aaron Wheeler, and Richard N Zare. Chemical cytometry on a picoliter-scale integrated microfluidic chip. Proc Natl Acad Sci USA, 101(35):12809–13, Aug 2004. [118] Mengsu Yang, Cheuk-Wing Li, and Jun Yang. Cell docking and on-chip monitoring of cellular reactions with a controlled concentration gradient on a microfluidic device. Anal Chem, 74(16):3991–4001, Aug 2002. [119] Julia Zeitlinger, Itamar Simon, Christopher T Harbison, Nancy M Hannett, Thomas L Volkert, Gerald R Fink, and Richard A Young. Program-specific distribution of a transcription factor dependent on partner transcription fac- tor and mapk signaling. Cell, 113(3):395–404, May 2003. [120] Lan V Zhang, Oliver D King, Sharyl L Wong, Debra S Goldberg, Amy H Y Tong, Guillaume Lesage, Brenda Andrews, Howard Bussey, Charles Boone, and Frederick P Roth. Motifs, themes and thematic maps of an integrated saccharomyces cerevisiae interaction network. J Biol, 4(2):6, Jan 2005. [121] Z Zhou, A Gartner, R Cade, G Ammerer, and B Errede. Pheromone- 39 induced signal transduction in saccharomyces cerevisiae requires the se- quential function of three protein kinases. Mol Cell Biol, 13(4):2069–80, Apr 1993. 40 Chapter 2 Dynamic Analysis of MAPK Signaling Using a Microfluidic Live-Cell Imaging Matrix 1 2.1 Introduction Cellular processes are governed by complex protein signaling networks that func- tion as robust and dynamic control systems, ensuring appropriate responses to sus- tained and transient stimuli. These networks feature emergent properties including bistability, adaptation, and memory that make their behavior inherently dependent on previous stimulation and current cell states. As examples: system bistability provides a selective advantage by allowing populations of cells to test the responses 1A version of this chapter has been accepted for publication. R. J. Taylor, D. Falconnet, A. Niemist, S. A. Ramsey, S. Prinz, I. Shmulevich, T. Galitski and C. L. Hansen. Dynamic Analysis of MAPK Signaling Using a Microfluidic Live-Cell Imaging Matrix. Proceedings of the National Academy of Sciences. 41 of alternative states to a given condition [22, 28]; network adaptation to a sustained change in stimulant concentration limits the metabolic cost of a sustained response [9, 41]; and network memory allows more rapid accommodation of recurrent stim- ulations [1, 34]. Due to experimental tractability, these emergent properties were first studied in model systems and have recently been uncovered in key mammalian regulatory networks, include those dysregulated in disease [37]. Owing to the facility of genetic manipulations and availability of reporters, yeast has emerged as the prototypical model of cell signaling. In particular, the pheromone response pathway in Saccharomyces cerevisiae is arguably the best characterized mitogen activated protein kinase (MAPK) signaling network, and has been a particularly fruitful model of eukaryotic signaling. MAPK signaling is of central importance to a wide range of cellular decision making processes, respond- ing to a staggering range of stimuli including growth factors, cytokines, hormones, cellular adhesion, stress, and nutrient conditions [5]. Regulated signaling governs cellular growth and differentiation while deviations from normal MAPK regulation are implicated in the onset of disease including cancer [10]. The yeast pheromone response is initiated by the binding of a mating-peptide, either α-factor or a-factor, to a membrane-localized G-protein-coupled receptor, either Ste2 or Ste3 on MATa or MATα cells respectively. Pheromone signaling is communicated through a MAPK signaling cascade that ultimately results in the phosphorylation of key substrates including the cyclin-dependent-kinase inhibitor Far1, which initiates growth arrest, and the transcription factor Ste12, which ac- tivates a program of gene expression involving over 200 genes [5, 7, 24, 32, 40]. Despite the wealth of detailed information gleaned from 30 years of biochemi- cal and genetic studies, the vast majority of these data are qualitative and derived 42 from measurements of cellular response under conditions of constant α-factor lev- els. More recently genome-wide analysis of transcription, protein expression, and protein interactions have been applied to systems-level studies of the pheromone response, delineating the tapestry of protein-protein interactions that mediate sig- naling [13, 31]. However, these studies are limited by poor temporal and spatial resolution, making it difficult to probe the dynamics of network function. Per- haps most importantly, these methods require the study of large populations of cells and are completely blind to cell differences that arise from a combination of de-synchronization, bistability, and stochastic variation in gene expression. The combination of fluorescent reporters and quantitative microscopy has re- cently been used to address cell-cell variability in the yeast mating response under constant environmental conditions [8, 28]. Such techniques are scalable to high- throughput formats in multi-well plates but provide only crude control over the microenvironment and are poorly suited to the study of response dynamics or his- tory effects. Indeed, little is known regarding cellular regulation in dynamically changing environments. This dearth of understanding is largely due to the techno- logical challenges involved in precisely controlling time-varying conditions, and limitations in throughput. Achieving a quantitative understanding of protein net- work function requires new tools for high-throughput studies under a large number of genetic perturbations and changing chemical environments, and with single-cell resolution [39]. In particular, microfluidics offers the combined advantages of precision fluid control necessary for exchange of media conditions around cells, and scalability for parallel analysis of multiple conditions on a single device. In yeast the pre- cise microfluidic control of conditions has been applied to investigations of modest 43 number of genotypes or chemical conditions under both constant [16, 28] or chang- ing [2, 20, 26] media conditions. More scalable devices have been applied to the culture and analysis of mammalian cells on-chip [21, 23] although these have to date be focused primarily on adherent cell types [6, 15, 23]. In particular, mi- crofluidic large-scale integration [35, 36] of devices having hundreds to thousands of valves has proven a powerful technique for simultaneously realizing the advan- tages of temporal control over media conditions and scalability of culture. Here we further extend the throughput and functionality of this approach in the devel- opment of a microfluidic high-throughput single cell analysis platform optimized for live-cell imaging studies of yeast. This system features a throughput of 256 simultaneous perfusion experiments with non-adherent yeast, integrated on-chip mixing and control software for programmable control of media conditions, and image processing algorithms and computational infrastructure for large-scale data analysis. We use our platform to investigate the role of signaling genes in network mem- ory and the filtering of transient stimulation. Recent studies have demonstrated the capacity for memory in cellular circuits, including the pheromone pathway [25] but were limited to long time scales [1] or in very few conditions [25, 34]. Using our microfluidic platform we examined cellular memory by performing over 3000 experiments investigating the combined effect of gene deletions and changing stimulant conditions on the mating response. These studies show that the mating system depends strongly on the frequency of stimulation and identifies genes that play a dominant role in regulating memory. 44 2.2 Results 2.2.1 High-Throughput Microfluidic Live Cell Imaging Platform To test the combined effect of genetic perturbations and chemical sequences on mating response we developed a microfluidic live cell imaging matrix in which 8 yeast strains are tested against a total of 32 stimulant-concentration sequences for a total of 256 simultaneous experiments (Fig. 2.1A). Unique genetic and chemical conditions are created along the matrix columns and rows, respectively. During cell loading all columns are isolated by actuation of row-valves (top to bottom) (Fig. 2.1B top), confining each yeast strain to a unique column without cross- contamination. At each vertex of the matrix the loaded cells were trapped in a perfusion chamber formed by a series of cell traps designed to immobilize yeast while allowing for the rapid and complete exchange of surrounding media (Fig. 2.1B bottom, Fig. 2.1C). Each cell trap consists of a partially closed valve that, when actuated, creates a cell filter to allow media exchange while retaining the cells (Fig. A.1). Cells are loaded through each column unimpeded and then trapped stochastically upon hydraulic actuation of the trap valves at approximately 120 kPa pressure. At each of the 256 perfusion chambers 5 sieve valves are actuated over a doubled flow line to create 8 cell traps (2048 per device). The size of these cham- bers is easily adjusted and in the current experiments was designed to accommodate approximately 600 cells before reaching confluence. 45 46 Figure 2.1: Schematic of the Microfluidic Device (A) Layout of microflu- idic live-cell imaging matrix. Device features two layers of channels including a flow structure (Blue) in which cells and reagents are intro- duced, and a control structure (Red) for pneumatic valves. Regions of the device are indicated including 1) Cell loading ports, 2) Experiment matrix, 3) Chemical inputs and control, 4) Peristaltic pump, 5) Fluidic multiplexer, And 6) Waste outlet. Each column of imaging matrix cor- responds to a single yeast genotype. Each row corresponds to a single experimental condition. (B) Control architecture for cell loading and perfusion. Dark red and orange or transparent red and orange lines in- dicate actuated and non-actuated control lines, respectively. Valve actu- ation isolated columns to direct flow through matrix during cell loading (top). During perfusion (middle) rows are isolated and flow is directed horizontally across the matrix perfusion chambers (bottom) formed by arrays of pneumatically actuated cell-traps (dark orange). (C) DIC im- age of cell yeast cells in perfusion chambers. (D) Stimulation condi- tions: 1) Step function: the cells are stimulated with constant α-factor solutions, with different pulse heights (∆h) (α-factor concentrations). 2) Pulse function: the cells are stimulated with a transient α-factor solu- tion, with different pulse heights (∆h) and pulse widths (∆w) (duration of stimulus) analyzed. 3) Short Repeated Pulses: the cells are stimu- lated with short repeated pulses of α-factor with different pulse heights (∆h) and different delays between pulses (∆d). (E) Pheromone pathway. Grey nodes indicate genes deleted in this study. 47 Perfusion of immobilized yeast allows for studies under well-defined and time- varying chemical conditions. Our device features fluidic elements for the periodic programmable mixing and delivery of chemical formulations to each row of the matrix to generate arbitrary chemical sequences of nutrient and stimulant concen- trations in time. On-chip generation of programmable chemical conditions is ac- complished by a peristaltic pump that precisely meters varying proportions of 8 stock reagents to enable accurate and continuous control of stimulant concentra- tion. Sequences of varying numbers of 120 pL aliquots of the input reagents are mixed in line by Taylor dispersion as they are transported from the mixing element to the array (Fig. 2.2E). A single mixing element controls all rows of the matrix using a time-division multiplexing strategy in which each row is sequentially ad- dressed using a fluidic multiplexer [35]. Between sequential perfusions the entire fluidic path connecting the mixer and matrix is purged through wash channels lo- cated between every row, thereby eliminating cross-contamination. Experiments using a fluorescent tracer show that contamination between rows is less than 1 part in 10,000 which was the detection limit of our detector. Automated perfu- sion of each row is performed periodically during experiments at approximately 100 s intervals to maintain nutrient levels, remove metabolites, and change condi- tions, thereby enabling long-term study of response under constant (chemostatic) or changing (chemodynamic) conditions. The measured generation time for yeast grown in SCD media at room temperature was 220 min and was found to be inde- pendent of seeding density below 200 cells per chamber, resulting in approximately a 10x increase in cell number over a 12 h experiment. This growth rate is consistent with off-chip measurements in bulk culture (240 min). Throughout each experiment the cells are confined in the vertical direction by 48 3.5 µm height of the perfusion chambers, restricting them to a monolayer of cells in a single focal plane and allowing for long-term imaging over multiple genera- tions. In each experiment high-resolution brightfield and fluorescence images of all 256 chambers were taken with 15 min time resolution over the entire length of each experiment (12.5 h). Two fields of view are required for complete imaging of each chamber so that a single experimental run generates over 50,000 images capturing millions of single-cell measurements. To handle the volume of raw image data we developed an image analysis pipeline to record single-cell data including cell num- ber, cell size, cell morphology, and concentration of a fluorescent gene-expression reporter molecule (GFP). 2.2.2 Imaging Studies of Pheromone Response Pathway Microfluidic parallelization allows for the simultaneous collection of unified data sets in a single experiment, thereby allowing for sensitive comparisons of wild- type with multiple mutant responses under a wide array of changing chemical conditions. We investigated the signaling response of wild type cells and a panel of 11 mutants having deletions of mating signaling genes (DIG2, RGA1, RGA2, SLT2, MSG5, PTP2, FUS3, KSS1, STE50, FAR1, BEM3) that are reported to have subtle or complex mutant phenotypes under constant α-factor stimulation [3, 27, 30, 32, 38, 40]. Mutant response was screened against a wide range of static and time-varying (Fig. 2.1D) conditions including: 1) constant stimulation under finely-varied concentrations to measure dose response of pathway activation and morphological variability; 2) transient pulses of varying concentration and du- ration to measure pathway deactivation and adaptation; 3) repeated short pulses of varying concentration and frequency to measure cellular memory of transient stim- 49 ulation. Mating-specific gene expression was reported using an enhanced green fluorescent protein (GFP) gene under the control of a minimal promoter including the tandem pheromone-response elements of the PRM1 promoter [19]. The BAR1 gene, encoding a secreted α-factor protease, was deleted from all strains to focus on the roles of intracellular elements. Details of strain construction are included in Appendix A. 2.2.3 Response Under Chemostatic Conditions Frequent media exchange allows for precise control of chemical conditions over long times to perform highly resolved studies of the dose response of signaling output. Using this control we validated our platform in the high-throughput anal- ysis of all 12 genotypes under static conditions of finely varied α-factor concen- trations. Using 5 identical devices we tested 8 strains per device with at least 3 replicates for each of the 12 strains. This analysis rapidly and faithfully repro- duced a broad range of observations collated from previous studies [8, 28] and further extended these results in terms of the number of chemical conditions, range of genetic perturbations, and temporal resolution. Signaling response of all strains was measured across 32 exponentially distributed α-factor concentrations, ranging from 1 to 100 nM. Gene expression was detectable over the full range of concen- trations and showed a 15-fold increase at saturating concentrations of 30-100 nM α-factor [8]. A representative data set for one of the 256 experiments, showing the distribution of single cell GFP expression and growth rate for wild type cells under 20 nM α-factor stimulation, is shown in Fig. 2.2A-B. Signaling response was mapped for each mutant as high-resolution GFP expression surfaces, show- ing the interplay between stimulation strength, time, and GFP concentration (Fig. 50 2.2C for wild type). The simultaneous testing of identical stimulation conditions in multiple strains allows for precise comparative analysis by normalization of expres- sion to wild type response (Fig. 2.2D). Under constant stimulation we identified hyper-responders (kss1∆, msg5∆, ptp2∆), wild-type-like responders (fus3∆, slt2∆, dig2∆, rga1∆, rga2∆, bem3) and hypo-responders (far1∆, ste50∆) (Fig. 2.2D). Generally, the degree of differential expression was found to be concentration- dependent with hyper- and hypo-responding phenotypes exhibited most strongly at low non-saturating α-factor concentrations, highlighting the context-specific effect of non-essential genetic perturbations to network output [8, 28]. 51 Figure 2.2: Mating response to persistent α-factor stimulation. (A,B) WT time course data showing mean and variation of response to constant stimulation with 20 nM α-factor. (A) Measured GFP concentration, re- porting mating specific gene expression, of each cell with mean of pop- ulation indicated in red. (B) Number of cells in the chamber as a func- tion of time showing arrest under alpha-factor stimulation. (C) Time course and dose response of mean GFP concentration in WT cells for all α-factor concentrations. (D) Strain comparison of signaling under constant stimulation. Initial dGFP/dt for all strains at the given concen- trations relative to WT (see SI text). Initial dGFP/dt is calculated as the slope of a line fitted to the population averaged GFP concentrations between 30-180min. (E) Performance of on-chip chemical formulation. Fluorescent measurements of 32 concentrations generated on-chip as detailed in SI methods and Table S1. 52 High-throughput imaging allows for the direct comparison of morphological transitions across varying genetic backgrounds as a function of alpha factor dose. Gene expression, cell-cycle arrest, and cell morphology changes are tightly cou- pled in the mating response. However, in contrast to mating-pathway-dependent gene expression, which increased continuously with increasing α-factor concen- tration, cell-cycle arrest and morphological transitions were found to exhibit de- fined thresholds. Analysis of single-cell morphology after 6 hours under varying α-factor concentrations reveals three distinct cell types: proliferating ovoid cells at very low concentrations (<4 nM), highly elongated cells at intermediate concen- trations (4-20 nM), and cells with shmoos (mating projections) at high concentra- tions (>20nM). At intermediate α-factor concentrations (4-20 nM) we find the co- existence of all three morphological types [11, 12, 18, 28] with characteristic levels of transcriptional output, a phenomenon that has been attributed previously to net- work bistability[28]. Fig. 2.3A depicts average wild-type (WT) gene expression in each morphological cluster after 6 h. Interestingly, some mutant strains were found to undergo morphological transitions at different thresholds of α-factor con- centration and to support the coexistence of phenotypes over differing concentra- tion ranges (Fig. A.4). For example, the morphological switch in msg5∆ mutants is more sensitive, exhibiting elongated morphologies at lower α-factor concentra- tions than WT (Fig. 2.3B). In contrast, ste50∆ presents no elongated morphology at any concentrations tested at 6 hours (Fig. 2.3C). Interestingly, fus3∆ displayed a delayed morphological response, with no observable elongation or shmooing until 10 h (data not shown). 53 54 Figure 2.3: Morphological response and transient stimulation responses. Morphological response of (A) WT, (B) msg5∆, (C) ste50∆ across all α-factor concentrations under constant stimulation. Color code for mor- phology shows representative images of yeast-form cells (YF) at 3.2nM (red); hyperelongated cells (HE) at 17 nM (blue), and shmoo cells (S) at 90 nM (green). Figure indicates mean GFP concentration, reporting mating specific gene expression, as a function of α-factor concentra- tion for each classification with dot opacity indicating the fraction of cells with that morphology. Error bars represent standard deviation of measured GFP expression for cells of each morphology. Measurements were taken 360 min after exposure to pheromone. (D,E) WT time course response to a 180 min duration 50 nM α-factor pulse. Cells are stim- ulated with α-factor at t=0; shading indicates the presence of α-factor. (D) GFP concentration, reporting mating specific gene expression, per cell with mean of population indicated in red with nearest neighbor time point averaging used to smooth the mean curve. (E) Total number of cells vs. time showing transient growth arrest during α-factor pulse. Each blue diamond is the total number of cells in the microchamber ar- ray at a given time. (F) Representative data of population mean GFP concentration in response to transient α-factor pulse of varying dura- tion. Data is shown for 20 nM α-factor condition. 2.2.4 Pathway Response Under Dynamic Stimulation Single Pulse Experiments: Microfluidics offers unique opportunities for measuring cellular response to precisely controlled time-varying stimulation and with high 55 temporal resolution [2, 4, 20]. We used this temporal control to investigate differ- ences in network memory between mutants. In particular, the propagation of signal through the MAPK network results in changes in the dynamic state of the system, including alterations in protein expression, phosphorylation, localization, and com- plex assembly. These changes modulate the network’s ability to transmit signals (network capacity), giving rise to memory effects in which the cellular response is history-dependent [1]. We first measured cellular response to transient α-factor, and tested whether cellular recovery depends on duration of stimulation. We stimulated yeast with single transient pulses of α-factor across a broad range of both stimulation strength and pulse duration: all combinations of four α-factor concentration (5, 10, 20, and 50 nM), and eight pulse widths (20, 40, 60, 90, 120, 150, 180, 210 min) were tested. Consistent with experiments under static conditions, we observed no threshold of response and measured expression in all conditions (Fig. 2.3F). Release from stimulation resulted in a characteristic decay time of 3.6 hours, beginning approximately 30 minutes after release, which was independent of pulse duration and the maximum level of GFP. This is consistent with reported GFP mat- uration times and dilution of GFP during cell growth, suggesting that the rapid deactivation of signaling output is independent of input dose (Fig. 2.3F). In con- trast to the case of periodic stimulation (described below), single-pulse stimulation revealed no new differences between mutants, suggesting that any changes in net- work dynamics arise through transients with fast characteristic time scales or adap- tation occurring at very long time scales. Similarly, analysis of cell cycle response (Fig. A.5C) indicates that cell growth quickly resumes upon α-factor removal (Fig. 2.3E). No morphological changes were observed in any cells for pulses shorter than 56 90 min even at saturating α-factor concentrations indicating that the emergence of a full mating response requires sustained stimulation. Directly probing signaling at faster time scales using single-pulse experiments is limited by low expression and the long maturation time of GFP, and will require future studies with faster reporters such as those using fluorescence resonance energy transfer (FRET), pho- toactivatable GFP [29], or mRNA tagging[14]. 2.2.5 Response to Periodic Stimulation Under constant stimulation different deletion mutants may exhibit phenotypes that are indistinguishable, thus making it difficult to assign unique functions to these genes. These genes may nonetheless have distinct roles in controlling short time- scale network dynamics and thus may be separated by analysis under varying stim- ulation. Although these differences are difficult to detect with fluorescent protein reporters, the monitoring of response under periodic stimulation allows for integra- tion of the GFP output to amplify subtle differences across conditions and mutant genotypes. We used this strategy to investigate mutant variations in pathway mem- ory by measuring transcriptional output to repeated 10 min pulses of pheromone of varying frequency. All strains were tested under repeated pulse conditions of varying concentrations (5, 10, 20 and 50 nM) and delay times (15, 40, 65, and 140 min) between pulses (Fig. 2.4A). Although the wild-type response was found qualitatively to increase with total time-averaged alpha-factor dose, there were no- table deviations from this trend that suggest a more complicated dependence on the frequency response of signaling. Cells were found to respond comparatively more strongly to repeated intermittent pulses of low pheromone than would be expected under a model of response to simple time-averaged concentration. For instance, 57 conditions of repeated 10 minute pulses of 5 nM pheromone every 25 minutes, corresponding to a time averaged dose of 2 nM, resulted in a gene expression re- sponse similar to conditions of 50 nM and 20 nM stimulation with pulse delays of 40 minutes (corresponding to time-averaged dose of 10 nM and 4 nM respectively). Similarly, 10 minute pulses of both 5 and 50 nM pheromone every 75 minutes, cor- responding to 0.67 nM and 6.7 nM time-averaged doses respectively, show very similar response. Taken together these observations under periodic stimulation in- dicate that pathway output depends strongly on cell history and the frequency of signal input. 58 59 Figure 2.4: Mating response to short repeated pulses of α-factor. (A) Pop- ulation mean GFP concentration, reporting mating specific gene expres- sion, of wild-type cells for the 50, 20, 10, and 5 nM α-factor concen- trations across four pulse delay lengths (15, 40, 60, 140 min). In each condition, 10 minute pulses of α-factor are used. The green rectangles depict the 4 pulse patterns of α-factor stimulation over time; values in- dicated delay time in min. (B) Sensitivity of kss1∆, fus3∆, msg5∆, and ptp2∆ mutants under periodic stimulation under conditions of varying α-factor concentration (columns) and pulse delays (row). Mean popu- lation GFP concentration over three experimental replicates are shown normalized to WT. Data are taken at t=600 min. 2.2.6 Mutants Implicated in Dynamic Phenotypes Analysis under periodic stimulation allows for the classification of mutants on the basis of differing dynamic response. To test this idea we compared the dynamic responses of all mutants (Fig. A.7B) and found distinct patterns of hypersensitivity for mutants lacking two highly related kinases, Fus3 and Kss1[24], and two phos- phatases, Msg5 and Ptp2 [40]. All four of these mutants exhibit WT-like mating phenotypes, are indistinguishable under saturating pheromone concentrations (50 nM) [8], and were found to be increasingly hypersensitive at low α-factor concen- trations (Fig. 2.4B). This divergence from WT behaviour was greatly amplified under low frequency periodic stimulation across all concentrations tested. More- over, the frequency-response of mating-pathway-dependent gene expression under varying pheromone concentrations was unique to each mutant. The ability to unambiguously stratify mutants on the basis of response to time- 60 varying stimulation provides a stringent test for the development and testing of quantitative network models and suggests new regulatory roles of for signaling pro- teins. Across all conditions, kss1∆ mutants exhibited the greatest divergence from WT. In addition to previously reported hypersensitivity at low pheromone con- centrations, kss1∆ mutants display hypersensitivity under intermittent pheromone stimulation. This effect was evident for all transient conditions, even when pulses are delayed by as little as 15min, and was most pronounced for low frequency stimulation with high pheromone concentrations. By comparison, mutant fus3∆ cells, which show similar pathway output to WT under all constant stimulation conditions, exhibit hypersensitivity to transient stimulation only for pulse delays of 40 min or more. Also, whereas the degree of hypersensitivity for fus3∆ cells was found to depend primarily on the frequency of stimulation, the sensitivity of kss1∆ mutants exhibits both concentration and frequency dependence, being most pronounced for transient pulses of high concentration. Taken together these re- sults implicate Kss1 in a regulatory mechanism which acts to filter both weak and intermittent α-factor stimulation while Fus3 appears primarily to filter transient signals (Fig. 2.4B). Analysis of the phosphatase mutants reveals a similar trend in which the hypersensitivity of msg5∆ mutants is largely determined by frequency, whereas ptp2∆mutants exhibit hypersensitivity depending on both pulse frequency and pheromone concentration. Similarities in behavior were also noted between kinase and phosphatase deletion mutants. The sensitivity trend for fus3∆ mutants under dynamic stimulation, having a frequency threshold with little concentration dependence, is similar to that of msg5∆ mutants at pulse delays longer than 15 minutes. At short pulse delays (15 minutes) kss1∆ mutants exhibit hypersensitivity similar to that of msg5∆ mutants. 61 2.3 Discussion Here we have presented a new system for high-throughput measurements of single- cell response over a broad array of precisely controlled and time-varying condi- tions. Further improvements will allow for both increased genotypic throughput and more refined analysis of single cells in time. Straightforward device modifica- tions will allow for the parallel analysis of 40 strains, allowing for comprehensive network-scale analysis on a single device run. Additionally, an important compo- nent of single-cell analysis is the ability to track single-cell response and lineage through time, something that is difficult to automate in the current format due to the motion of cells within the traps during perfusion sequences. We are currently refin- ing cell immobilization techniques, image processing algorithms, and data analysis methods to enable automated high-throughput tracking of tens of thousands of sin- gle cells, spanning multiple genotypes and stimulation conditions, through time (Appendix B). Microfluidics provides a powerful method for high-throughput imaging anal- ysis with programmable control over the chemical environment, offering a new temporal dimension to live-cell imaging studies. Our present study demonstrates that the analysis of cellular networks under static conditions or with coarse chem- ical resolution is insufficient to reveal the function of genes in regulating network response. Indeed, dynamic analysis of mutants compromised for genes known to be key players in the pheromone response, including Kss1, Fus3, Msg5, and Ptp2, reveals unique properties of network response that are invisible under con- stant stimulation and that suggest possible mechanisms of network regulation. For instance the similarity in frequency threshold for the hypersensitivity of kss1∆ and 62 msg5∆mutants may be due to increased Fus3 activity in both of these mutants. The Kss1 kinase competes with Fus3[33] and Msg5 phosphatase acts on Fus3 [40]. Usually, the response to transient stimulus is discussed in the context of signal filtering to prevent unproductive response. Conversely, we speculate that frequency- dependent signal memory enables productive responses in natural environments. For instance, one might expect that the paracrine induction and auto-repression of the mating response, including the secretion of pheromone and pheromone- degrading enzymes, coupled with hydrodynamics, varying cell density, and cell motion, creates spatiotemporal variations in pheromone concentration and inter- mittent opportunities for successful mating. Pathway mechanisms selected to filter or remember stimulations over appropriate time-scales could act to prime cells for more rapid response, thus increasing mating success. Testing of such hypotheses will ultimately require combined approaches based on quantitative modeling and experiment. We contend that high-throughput single-cell measurements of net- work dynamics will provide a stringent test for in silico models and are essential for ultimately developing a quantitative and predictive understanding of cellular decision-making. 2.4 Methods 2.4.1 Cell loading. For each experiment yeast cells were grown with aeration overnight in YPD (30C), diluted and grown to log-phase in SCD on and loaded into the microfluidic device at 3-15 p.s.i. (20.7 103.4 kPa). Prior to loading the cells were sonicated for 5s, concentrated to an OD600 of 3 and manually aspirated into tygon tubes (internal 63 diameter 0.020 in) with a 1ml syringe. The chip was primed with SCD (synthetic complete media with 2% dextrose) medium containing 20mg/ml of bovine serum albumin (BSA) for approximately 3 h prior to cell loading. Adsorption of BSA on PDMS channel walls allowed performing reproducible experiments by preventing cell adhesion and significant non-specific binding of alpha factor to the channel walls. Chemical Mixing and Perfusion. Between every exchange or refreshing of medium, the following refresh proto- col was used: 1) Prime and wash multiplexer: Perfuse adjacent waste row with chemical solution for 20s. 2) Refresh microchambers: Perfuse experiment row with chemical solution for 70s. 3) Wait: A wait of 3s is used to dissipate pressure build up that occurs across the high-impedance microchamber traps. This protocol allowed us to fully refresh an experimental row approximately every 100s. 2.4.2 Microfluidic Control Microfluidic operation was fully computer controlled, excluding cell loading and trapping. Custom software was designed and executed in LabVIEW (National In- struments, Austin, TX). Microfluidic valves were actuated using high-speed com- puter controlled solenoid micropump manifolds (Fluidigm Inc., San Francisco, CA) via a PCI-6533 digital input/output (DIO) card (National Instruments). A single LabVIEW program operated all experiment types, with user designed ex- periments inputted as parsed text files. Scheduling algorithms were included to maximize the frequency of experiment refresh of all 32 chemical sequences. 64 2.4.3 Microfluidic Fabrication Fabrication of the microfluidic device was accomplished using multi-layer soft lithography [17, 36]. Our chips used a 3-layer design: The top layer was a ’flow layer’, containing the cells and the chemical channels. The middle layer was a ’control layer’, containing channels used for pneumatic valves. The bottom layer was a ’blank layer’, used to tightly seal the control channels to the glass slide. All devices were made from polydimethylsiloxan (RTV615 General Electric, Fairfield, CT). Briefly, photolithography masks were designed using AutoCAD software (Au- todesk, Inc., San Rafael, CA) and used to generate high-resolution (20,000 dpi) transparency masks (CAD/Art Services). Molds were fabricated by photolithogra- phy on 4 inch silicon wafers (Silicon Quest International, Santa Clara, CA). The flow layer consisted of two different channel profiles: 3.5 µm high rectangular trapping channels allowing for sieve valving, and 12 µm high rounded channels used for standard flow. The 3.5 µm layer was made with SU8-5 negative pho- toresist (Microchem Corp., Newton, MA) and the 12 µm rounded layer was made with SPR220-7 positive photoresist (Microchem Corp.). The control master was a single layer mold consisting of 25 µm high squared features made with SU8- 2025 negative photoresist (Microchem Corp.). Resist processing was performed according to the manufacturer’s specifications. 2.4.4 Image Acquisition Microfluidic devices were mounted onto a Leica DMIRE2 fluorescent microscope modified with a custom LED brightfield source to increase acquisition speed. Cells were imaged with a 40x air objective (HCX, long working distance, FLUOTAR PL 65 with correction collar, NA=0.6) and 1344x1024 pixel cooled CCD camera (ORCA- ER, Hamamatsu Photonics, Hamamatsu Japan). Z focus was determined indepen- dently for each acquisition position. Differential interference contrast (DIC) and fluorescent images were captured with a 100 ms and 250 ms exposure respectively. Each set of microfluidic cell traps was acquired in two fields of view, giving a total of 256 experiment traps x 2 = 512 differential interference contrast / fluorescent image pairs acquired per time point. We acquired a full set of images every 15min, which was slightly longer than the time it took to iterate over and acquire the 512 image pairs. Over a 12.5 h experiment this resulted in over 50,000 images ( 120 Gb of image data). To process data a customized image analysis pipeline was devel- oped using MATLAB software (Mathworks, Natick, MA). Detailed descriptions of algorithms for cell segmentation and GFP concentration calculation are provided in the SI methods. 66 Bibliography [1] Murat Acar, Attila Becskei, and Alexander van Oudenaarden. Enhance- ment of cellular memory by reducing stochastic transitions. Nature, 435(7039):228–32, May 2005. [2] M Bennett, W Pang, N Ostroff, B Baumgartner, S Nayak, L Tsimring, and J Hasty. Metabolic gene regulation in a dynamically changing environment. Nature, 454:1119–1122, Jul 2008. [3] A Breitkreutz, L Boucher, and M Tyers. Mapk specificity in the yeast pheromone response independent of transcriptional activation. Curr Biol, 11(16):1266–71, Aug 2001. [4] Gilles Charvin, Frederick R Cross, and Eric D Siggia. A microfluidic device for temporally controlled gene expression and long-term fluorescent imaging in unperturbed dividing yeast cells. PLoS ONE, 3(1):e1468, Jan 2008. [5] Raymond E Chen and Jeremy Thorner. Function and regulation in mapk signaling pathways: lessons learned from the yeast saccharomyces cerevisiae. Biochim Biophys Acta, 1773(8):1311–40, Aug 2007. 67 [6] R Cheong, C Wang, and A Levchenko. High-content cell screening in a microfluidic device. Mol Cell Proteomics, Oct 2008. [7] K Y Choi, B Satterberg, D M Lyons, and E A Elion. Ste5 tethers multiple protein kinases in the map kinase cascade required for mating in s. cerevisiae. Cell, 78(3):499–512, Aug 1994. [8] Alejandro Colman-Lerner, Andrew Gordon, Eduard Serra, Tina Chin, Orna Resnekov, Drew Endy, C Gustavo Pesce, and Roger Brent. Regulated cell-to- cell variation in a cell-fate decision system. Nature, 437(7059):699–706, Sep 2005. [9] K Doi, A Gartner, G Ammerer, B Errede, H Shinkawa, K Sugimoto, and K Matsumoto. Msg5, a novel protein phosphatase promotes adaptation to pheromone response in s. cerevisiae. EMBO J, 13(1):61–70, Jan 1994. [10] Julian Downward. Targeting ras signalling pathways in cancer therapy. Nat Rev Cancer, 3(1):11–22, Jan 2003. [11] S Erdman and M Snyder. A filamentous growth response mediated by the yeast mating pathway. Genetics, 159(3):919–28, Nov 2001. [12] F W Farley, B Satterberg, E J Goldsmith, and E A Elion. Relative depen- dence of different outputs of the saccharomyces cerevisiae pheromone re- sponse pathway on the map kinase fus3p. Genetics, 151(4):1425–44, Apr 1999. [13] Mark R Flory, Hookeun Lee, Richard Bonneau, Parag Mallick, Kyle Serikawa, David R Morris, and Ruedi Aebersold. Quantitative proteomic 68 analysis of the budding yeast cell cycle using acid-cleavable isotope-coded affinity tag reagents. Proteomics, 6(23):6146–57, Dec 2006. [14] Ido Golding, Johan Paulsson, Scott M Zawilski, and Edward C Cox. Real- time kinetics of gene activity in individual bacteria. Cell, 123(6):1025–36, Dec 2005. [15] Rafael Gómez-Sjöberg, Anne A Leyrat, Dana M Pirone, Christopher S Chen, and Stephen R Quake. Versatile, fully automated, microfluidic cell culture system. Anal Chem, 79(22):8557–63, Nov 2007. [16] Alex Groisman, Caroline Lobo, HoJung Cho, J Kyle Campbell, Yann S Du- four, Ann M Stevens, and Andre Levchenko. A microfluidic chemostat for experiments with bacterial and yeast cells. Nat Methods, 2(9):685–9, Sep 2005. [17] Carl L Hansen, Morten O A Sommer, and Stephen R Quake. Systematic investigation of protein phase behavior with a microfluidic formulator. Proc Natl Acad Sci USA, 101(40):14431–6, Oct 2004. [18] Nan Hao, Sujata Nayak, Marcelo Behar, Ryan H Shanks, Michal J Nagiec, Beverly Errede, Jeffrey Hasty, Timothy C Elston, and Henrik G Dohlman. Regulation of cell signaling dynamics by the protein kinase-scaffold ste5. Mol Cell, 30(5):649–56, Jun 2008. [19] M G Heiman and P Walter. Prm1p, a pheromone-regulated multispanning membrane protein, facilitates plasma membrane fusion during yeast mating. J Cell Biol, 151(3):719–30, Oct 2000. 69 [20] Pascal Hersen, Megan N McClean, L Mahadevan, and Sharad Ramanathan. Signal processing by the hog map kinase pathway. Proc Natl Acad Sci USA, 105(20):7165–70, May 2008. [21] Paul J Hung, Philip J Lee, Poorya Sabounchi, Robert Lin, and Luke P Lee. Continuous perfusion microfluidic cell culture array for high-throughput cell- based assays. Biotechnol Bioeng, 89(1):1–8, Jan 2005. [22] Nicholas T Ingolia and Andrew W Murray. Positive-feedback loops as a flexible biological module. Curr Biol, 17(8):668–77, Apr 2007. [23] Kevin R King, Sihong Wang, Daniel Irimia, Arul Jayaraman, Mehmet Toner, and Martin L Yarmush. A high-throughput microfluidic real-time gene ex- pression living cell array. Lab on a chip, 7(1):77–85, Jan 2007. [24] H D Madhani, C A Styles, and G R Fink. Map kinases with distinct in- hibitory functions impart signaling specificity during yeast differentiation. Cell, 91(5):673–84, Nov 1997. [25] Megan N McClean, Areez Mody, James R Broach, and Sharad Ramanathan. Cross-talk and decision making in map kinase pathways. Nat Genet, 39(3):409–14, Mar 2007. [26] Jerome T Mettetal, Dale Muzzey, Juan M Pedraza, Ertugrul M Ozbudak, and Alexander van Oudenaarden. Predicting stochastic gene expression dynamics in single cells. Proc Natl Acad Sci USA, 103(19):7304–9, May 2006. [27] Rammohan Narayanaswamy, Wei Niu, Alexander D Scouras, G Traver Hart, Jonathan Davies, Andrew D Ellington, Vishwanath R Iyer, and Edward M 70 Marcotte. Systematic profiling of cellular phenotypes with spotted cell mi- croarrays reveals mating-pheromone response genes. Genome Biol, 7(1):R6, Jan 2006. [28] Saurabh Paliwal, Pablo A Iglesias, Kyle Campbell, Zoe Hilioti, Alex Grois- man, and Andre Levchenko. Mapk-mediated bimodal gene expression and adaptive gradient sensing in yeast. Nature, 446(7131):46–51, Mar 2007. [29] George H Patterson and Jennifer Lippincott-Schwartz. A photoactivatable gfp for selective photolabeling of proteins and cells. Science, 297(5588):1873–7, Sep 2002. [30] M R Rad, G Xu, and C P Hollenberg. Ste50, a novel gene required for acti- vation of conjugation at an early step in mating in saccharomyces cerevisiae. Mol Gen Genet, 236(1):145–54, Dec 1992. [31] C J Roberts, B Nelson, M J Marton, R Stoughton, M R Meyer, H A Bennett, Y D He, H Dai, W L Walker, T R Hughes, M Tyers, C Boone, and S H Friend. Signaling and circuitry of multiple mapk pathways revealed by a matrix of global gene expression profiles. Science, 287(5454):873–80, Feb 2000. [32] W Sabbagh, L J Flatauer, A J Bardwell, and L Bardwell. Specificity of map kinase signaling in yeast differentiation involves transient versus sustained mapk activation. Mol Cell, 8(3):683–91, Sep 2001. [33] Monica A Schwartz and Hiten D Madhani. Principles of map kinase signaling specificity in saccharomyces cerevisiae. Annu Rev Genet, 38:725–48, Jan 2004. 71 [34] Alex Sigal, Ron Milo, Ariel Cohen, Naama Geva-Zatorsky, Yael Klein, Yuvalal Liron, Nitzan Rosenfeld, Tamar Danon, Natalie Perzov, and Uri Alon. Variability and memory of protein levels in human cells. Nature, 444(7119):643–6, Nov 2006. [35] Todd Thorsen, Sebastian J Maerkl, and Stephen R Quake. Microfluidic large- scale integration. Science, 298(5593):580–4, Oct 2002. [36] M A Unger, H P Chou, T Thorsen, A Scherer, and S R Quake. Monolithic microfabricated valves and pumps by multilayer soft lithography. Science, 288(5463):113–6, Apr 2000. [37] Guang Yao, Tae Jun Lee, Seiichi Mori, Joseph R Nevins, and Lingchong You. A bistable rb-e2f switch underlies the restriction point. Nat Cell Biol, 10(4):476–82, Apr 2008. [38] Richard C Yu, C Gustavo Pesce, Alejandro Colman-Lerner, Larry Lok, David Pincus, Eduard Serra, Mark Holl, Kirsten Benjamin, Andrew Gordon, and Roger Brent. Negative feedback that improves information transmission in yeast signalling. Nature, 456(7223):755–61, Dec 2008. [39] Eli Zamir and Philippe I H Bastiaens. Reverse engineering intracellular bio- chemical networks. Nat Chem Biol, 4(11):643–7, Nov 2008. [40] X L Zhan, R J Deschenes, and K L Guan. Differential regulation of fus3 map kinase by tyrosine-specific phosphatases ptp2/ptp3 and dual-specificity phosphatase msg5 in saccharomyces cerevisiae. Genes Dev, 11(13):1690– 702, Jul 1997. 72 [41] J Zhou, M Arora, and D E Stone. The yeast pheromone-responsive g alpha protein stimulates recovery from chronic pheromone treatment by two mech- anisms that are activated at distinct levels of stimulus. Cell Biochem Biophys, 30(2):193–212, Jan 1999. 73 Chapter 3 Network Motif Analysis of a Multi-Mode Genetic-Interaction Network 1 3.1 Background The cell is an elaborate network of biomolecular and environmental interactions that together bring about complex phenotypes. Understanding the functional con- sequences of molecular interactions is fundamental to understanding phenotypes. A highly successful approach is the use of genetic interactions. Genetic interac- tions describe the phenotypic consequences of combinations of genetic perturba- tions. Genetic interactions combined with molecular interaction data can delin- eate information flows through complex biochemical systems. The concept of the 1A version of this chapter has been published. Network Motif Analysis of a Multi-Mode Genetic- Interaction Network. R. J. Taylor, A. F. Siegel, and T. Galitski. Genome Biology. 8R:160. 2007 74 molecular signaling pathway owes much to this approach. A genetic interaction comprises phenotype measurements of four genotypes: the reference genotype (wild type (WT)); a single gene perturbation A; a pertur- bation B of a different gene; and the double perturbation AB. By themselves, the single perturbations link individual genes to specific phenotypes and biological processes. Studying a double perturbation defines functional relationships between the perturbed genes. The relative ordering of the four phenotype measurements de- fines different genetic-interaction modes [7]. Genetic-interaction modes indicate one or more possible molecular relationships, for example, upstream/downstream. Networks of genetic interaction, and the molecular wiring, constrain these possi- bilities. In this way, genetic-interaction modes are a reflection of the underlying biochemical system. Geneticists have formalized collections of genetic interactions into genetic- interaction networks of perturbed-gene nodes and genetic-interaction edges. Tong et al. [31] created a network consisting of edges representing a single type of ge- netic interaction, synthetic lethal. Zhang et al. [35] integrated this network with disparate data types, including protein-protein and protein-DNA interactions, se- quence homologies, and expression correlations. In this study, network patterns were used to reduce the overall system into a thematic map of biological rela- tionships. The E-MAP method [4, 24] creates high-density genetic-interaction networks consisting of aggravating or alleviating edge types. This method has been fruitful for identifying both system-level and protein-complex-level func- tional modularity. Further work has generated networks of multiple genetic-interaction modes (edge types). In Drees et al. [7], all possible genetic interactions were classified 75 into nine modes, of which four are asymmetric (directed edges). A multi-mode genetic-interaction network was derived from a large set of quantitative phenotype data. This work revealed local and global genetic-interaction patterns suggesting the prevalence of information contained in the structure and distribution of genetic interactions within the network. Further network information can be extracted from such complex networks by identifying significantly repeated genetic-interaction patterns, network motifs [14, 16, 28]. In this study, we report a network-motif analysis of the dense multi-mode genetic-interaction network of Drees et al. [7]. 3.2 Results and Discussion 3.2.1 Multi-Mode Genetic-Interaction Network In the network of Drees et al. [7], there are 1,760 genetic interactions among 128 perturbed genes controlling the agar-invasion phenotype of diploid budding yeast. The perturbations included gene deletions as well as overexpressers and domi- nant alleles. This yeast-invasiveness network contains all nine possible genetic- interaction modes, including noninteracting, epistatic, synthetic, suppressive, ad- ditive, conditional, asynthetic, nonmonotonic, and double-nonmonotonic interac- tion. Four of these modes (epistatic, suppressive, conditional, and nonmonotonic) are directional, giving thirteen possible edges between any pair of nodes. Note that the genetic-interaction modes discussed in this paper refer to those defined in Drees et al. [7], and that there are semantic differences between the Drees definitions and other genetic-interaction classifications. Example interactions for each mode are shown in Additional data file 22 [20]. 76 3.2.2 Genetic-Interaction Patterns Reflect the Underlying Molecular System Prior to rigorous statistical motif analysis, we inspected the yeast-invasiveness net- work to discern possible patterns of genetic interactions reflecting the underlying molecular system. Fig. 3.1 shows genetic interactions among components of three main signaling pathways controlling yeast invasiveness [2, 3, 5, 8, 15, 17–19, 21– 23, 29, 30, 34]. Subsequently, we investigated our preliminary observations (de- scribed below) quantitatively and globally in the network. We initially observed that there are local patterns incorporating both edge type and network topology. For example, consider the interactions between the overex- pressers of CDC42 and GLN3 and the deletions of DIG2 and TPK2. Both CDC42 and GLN3 interact asynthetically with DIG2 and nonmonotonically with TPK2, creating a two-mode bi-fan interaction pattern. Also, we observed that patterns of genetic interaction can reflect the direc- tion of information flow through the molecular network. For instance, epistatic interactions involving the STE12 overexpresser originate from upstream signaling components. Also, many genetic interaction modes occur repeatedly between par- allel information paths. For instance, the HOG1 deletion interacts synthetically with deleted components of the cAMP pathway and additively with overexpressed components of the filamentation/invasion MAP-kinase (fMAPK) pathway. 77 78 Figure 3.1: Multi-mode genetic-interaction motifs and the underlying molecular system. Genetic-interaction edges are superimposed onto a diagram of the cAMP, fMAPK, and HogMAPK signaling pathways. Gene perturbations are marked: hc, high copy overexpresser; ∆ , dele- tion. 3.2.3 Statistical Model of a Null Hypothesis Biologically relevant genetic-interaction patterns can be identified by finding those occurring more frequently in the genetic network than expected at random. This can be done by comparing the number of times a given pattern occurs in the genetic network to the number of times it occurs in a set of properly randomized networks. The randomized networks represent a statistical null hypothesis and effectively model the level of pattern noise in the network [16] [24]. In this way, significance can be assigned to each identified pattern. In this study we highlight those patterns with a significance level of p < 0.05/n, using the Bonferroni multiple-hypothesis- testing correction, where n is the number of patterns tested in each analysis. Al- gorithms were developed to create the set of randomized networks modeling a null hypothesis. The yeast-invasiveness network contains nine edge types of which four are directed. Randomized networks were generated by a Monte Carlo method it- eratively selecting a pair of edges at random and swapping their edge types. See Materials and methods for details. Randomizations were subject to specific constraints to preclude the introduc- tion of biases to the results. Each edge represents the results of a given experiment (repeated measurement of the phenotypes of WT, A, B, and AB). Every genetic experiment creates a resulting genetic edge, with noninteracting edge types used 79 in the cases of genetically noninteracting loci. This causes the topology of the net- work (the simple presence or absence of an edge of any type linking each pair of nodes) to be determined by experimental design (the set of experiments performed or not performed), not by genetics. Thus, for proper randomization the network topology is held constant. The results could also be biased by the selection of mu- tant alleles included in the experiments. As described in Additional data file 22 [20], the data for a genetic interaction consist of the ordering of four phenotypes: WT, A, B, and AB. The single-mutant phenotypes could be biased by the selection of mutant alleles. To preclude this allele-selection bias, in our Monte Carlo switch- ing we restricted edge-type swaps to those in which the two edges have the same relative ordering of A, B, and WT. Lastly, in some of the analyses below, molecular data are mapped onto the genetic network. In these cases the genetic-interaction edge types are randomized under the above constraints, while the molecular data are held constant. Note that our randomization methods are strictly conservative and restrict the number of significant motifs. Such methods are necessary to en- sure that the calculated significance is due to biological significance rather than experimental design. 3.2.4 Genetic-Interaction Network Motifs To identify genetic-interaction network patterns that reflect biological relationships such as those illustrated in Fig. 3.1, we identified network motifs. Network motifs are small repeatedly occurring multi-element components of a network, where the repetition suggests functional significance. Such methods have been successful in extracting information from various other network types [14, 16, 28, 32, 33], as well as identifying general themes in the evolved organization of molecular sys- 80 tems [35]. The simplest network patterns containing information about the genetic-interaction modes and their system-level organization are 3-node motifs (3n-motifs). Using the null hypothesis method described above, we enumerated all 3n patterns in the yeast invasiveness network and tested each one for biological significance. We found 27 significant motifs among the 489 different patterns observed in the network (5.5%). Many of these motifs occur hundreds or thousands of times in the yeast-invasion network. Examples are shown in Fig. 3.2a. The full set is found in Additional data file 1 [20]. Homogeneous-edge-type motifs were found frequently, with 9 of the 13 possible homogeneous 2-edge patterns being significant (3n-motifs 1, 4, 5, 6, 9, 10, 11, 23, 27). Examples of such motifs occur in Fig. 3.1. Their global frequency may reflect the tendency of gene perturbations to show ’monochromatic’ interac- tion [7, 26]. Many heterogeneous motifs also were found (3n-motifs 2, 3, 7, 8, 12, and so on), as were various fully connected motifs (for example, 3n-motifs 22, 24, 25, 26, and so on). 81 82 Figure 3.2: Motifs in the yeast-invasiveness genetic-interaction network. (a) Examples of significant 3-node motifs. The number of instances of each motif is indicated as is the p value. A statistical cutoff of p = 0.05/489 = 1.02 × 10−4 was used to define significant patterns. (b) Examples of significant 4-node motifs. The number of occurrences is shown as the percentage of the full number of patterns sampled. P val- ues are shown and a statistical cutoff of p = 0.05/1,505 = 3.32 × 10−5 was used to define significant patterns. Edge colours indicate genetic interaction mode as indicated in Fig. 3.1. The full collection of motifs is in Additional data files 1 and 4 [20]. We also identified significant 4-node patterns (4n-motifs). Because the num- ber of pattern instances contained in a network scales combinatorially with local network density and pattern order (number of nodes in the pattern), the full enu- meration of 4n pattern instances was computationally infeasible. Thus, a sampling algorithm (Materials and methods) [28] was employed. Of the 1,505 4n patterns sampled from the original network, 190 (12.6%) were repeated significantly. The full list of 4n-motifs can be found in Additional data file 4 [20]. Fig. 3.2b shows examples. We found 4n-motifs exhibiting the edge-type homogeneity detected among 3n-motifs, as well as mixed-edge-type motifs. We noted that specific nodes (gene perturbations) often appear repeatedly among the numerous instances of a specific motif. This suggested that the instances of mo- tifs are connected structural units of larger single-motif subnetworks. Such subnet- works can highlight the main perturbations contributing to a motif, and show the large scale organization of instances of the motif. Fig. 3.3 shows an example of single-motif subnetworks, and additional examples are in Additional data file 23 83 [20]. In Fig. 3.3 is the incoming epistatic motif network of 3n-motif 9. In an epistatic interaction, the phenotype of the double mutant is the same as one of the two gene perturbations, and depending on the allele type (hypermorphic or hypo- morphic), orders the epistatic gene upstream or downstream (see mode definitions in Drees et al. [7]). In this way, epistatic interactions have been commonly used to help identify and delineate directed information flows in biochemical systems. As shown in Fig. 3.3, the epistatic motif network is organized around six main gene perturbation hubs: the overexpressions of STE20, STE12, CDC42 and GLN3, and the deletions of IPK1 and HSL1. Extending the concept of single epistatic inter- actions, these repeated interactions suggest critical hubs of information flow, and genes whose influences are likely to flow through them. 84 Figure 3.3: Motif subnetworks. An example of a motif subnetwork. A mo- tif subnetwork is the union of all instances of a specific motif. Shown here is the subnetwork of 3n-motif 9. The gene perturbations compris- ing the genetic interactions are marked with the suffixes: hc, high copy overexpresser; ∆, deletion. 85 3.2.5 Molecular Information and Genetic-Interaction Network Motifs Fig. 3.1 illustrates genetic-interaction patterns describing specific functional rela- tionships within and between the signaling pathways. To identify significant rela- tionships between genetic interactions and molecular-function data, we integrated these data types [7, 9, 11, 12, 24, 25]. Patterns from such integrated networks can be tested for statistical significance allowing for the identification of signifi- cant network motifs. In our case, these motifs are genetic-interaction patterns that exhibit significance in the context of the molecular system [31]. Filamentation/invasion signaling is a directed system that can be characterized loosely by the molecular functions of the system components. Plasma-membrane receptors transfer information to cytoplasmic signaling components that then regu- late nuclear transcription factors. These molecular functions capture a first approx- imation of the directionality of the system. By mapping the GoSlim [10] ’molec- ular function’ annotations onto the nodes of the yeast-invasiveness network, we identified genetic-interaction network motifs involving these loosely directed rela- tionships. Fig. 3.4a,b shows examples of the significant 2-node and 3-node motifs for the molecular-function annotations, respectively. The full sets are found in Addi- tional data files 7 and 10 [20], respectively. Of the 575 observed 2-node GoSlim molecular function patterns in the original network, 6 (1.0%) were found signifi- cant (2nGO-motifs). Of the 23,286 observed 3-node molecular-function patterns, 116 (<0.5%) were found significant (3nGO-motifs). These significant patterns illustrate a correspondence between the genetic-interaction modes and the under- lying biochemical system. For example, 2nGO-motif 1 (Fig. 3.4a) shows additive 86 interactions between perturbations of protein-binding proteins and transcriptional regulators. Among the instances of this motif are additive interactions of a deletion of DIG2 with overexpression of FLO8 and deletion of SFL1. The Dig2 protein binds and inhibits the Ste12 protein, a transcriptional activator of the filamenta- tion/invasion MAP-kinase (fMAPK) pathway. DIG2 deletion interacts additively with perturbations of FLO8 and SFL1, encoding transcription factors of a different filamentation/invasion-promoting pathway, the cyclic-AMP pathway. The addi- tive interaction reflects the separate contributions of these pathways. As another example, 3nGO-motif 166 (Fig. 3.4b) shows perturbations of protein kinase/trans- ferase activity proteins interacting supressively to transcriptional regulator proteins and to hydrolase activity proteins. In the context of filamentation signaling, envi- ronmental signals are transmitted through hydrolase (for example, GTPase) and kinase activity proteins to transcriptional regulators. In a suppressive genetic in- teraction, a suppressor gene perturbation ameliorates the effects of the suppressed perturbation, indicating the suppressor perturbation reverses or short-circuits the suppressed perturbation. A specific instance of this is that a deletion of the cAMP- dependent protein kinase subunit Tpk3 abrogates the effects of overexpression of both the membrane localized hydrolase Cdc42 and the transcriptional regulator Ste12. Cdc42 is an upstream activator of the fMAPK signaling pathway, and Ste12 is a downstream transcription factor of the same pathway [1–3, 6]. This motif in- stance suggests that loss of TPK3 activity in the parallel cAMP pathway offsets the effects of overexpression of CDC42 or STE12 activity in the fMAPK pathway. 87 Fi gu re 3. 4: E xa m pl es of m ot ifs in te gr at in g ge ne an no ta tio ns . E xa m pl es of si gn ifi ca nt (a ) 2- no de an d (b ) 3- no de m ot if s in vo lv e ge ne tic -i nt er ac tio n ed ge s an d G O Sl im m ol ec ul ar -f un ct io n ge ne -a nn ot at io n no de s. T he nu m be r of in st an ce s an d ca lc ul at ed p va lu e of ea ch m ot if is in di ca te d. Fo r th e 2n G O -m ot if s a st at is tic al cu to ff of p = 0. 05 /5 75 = 8. 7 × 10 −5 w as us ed .F or th e 3n G O -m ot if s a st at is tic al cu to ff of p = 0. 05 /2 3, 28 6 = 2. 14 × 10 −6 w as us ed .T he fu ll co lle ct io n of m ot if s is in A dd iti on al da ta fil es 7 an d 10 [2 0] . 88 To investigate the distribution of these motif examples within the full network, motif subnetworks were generated. Fig. 3.5a,b shows the motif subnetworks for 2nGO-motif 1 and 3nGo-motif 166, respectively. The 2nGo-motif 1 network is organized around the transcription factor tri-hub MSN1, PHD1, and FLO8, and the two separate single transcription factor hubs, SFL1 and GLN3. This network exhibits a high degree of mutually informative genetic interactions. Each of the eight protein binding proteins that interact with the tri-hub (AGA1, BMH1, LIN1, SSA4, MSN5, URE2, DIG2, and ENT1) interacts with each tri-hub member. This suggests overlapping pathway functionality within the set of protein binding pro- teins and within the set of transcription factors. This motif-instance organization contrasts with that of 3nGo-motif 166. The 3nGo-motif 166 subnetwork centers on the single protein kinase/transferase hubs TPK3, PBS2, HOG1, and HSL1. These kinases are information flow constriction points in their respective signaling path- ways: TPK3 in the cAMP pathway, PBS2 and HOG1 in the osmolarity sensing pathway, and HSL1 in the morphogenic checkpoint pathway. In contrast to the 2nGo-motif network, these single hubs primarily act independently of each other, with two hubs having at most only two nodes in common. This likely reflects the differing roles these pathways play in the invasion phenotype. Interestingly, the osmolarity sensing kinases Pbs2 and Hog1 show differing interaction patterns, although they are implicated in the same pathway. This possibly reflects subtly differing roles of the two kinases. These examples illustrate how the aggregation of motif information in motif subnetworks highlights biological information not present in individual motif instances. 89 90 Figure 3.5: Annotation-motif subnetworks. (a) The union of all instances of 2nGO-motif 1, which comprises perturbations of protein binding pro- teins and transcriptional regulators acting additively. (b) The union of all instances of 3nGO-motif 166, which comprises perturbations of pro- tein kinase/transferase activity proteins interacting supressively to tran- scriptional regulator proteins and to hydrolase activity proteins. Gene perturbations are marked: hc, high copy overexpresser; ∆, deletion. 3.2.6 Comparing Network Patterns in a Similar Genetic-Interaction Network The diversity of networks that can be formed from 13 edge types and large numbers of nodes is enormous. Thus, the yeast-invasiveness genetic-interaction network probably contains a sample of biologically relevant genetic-interaction motifs. To gauge the scope of our analysis we made a comparison of motifs in the yeast inva- siveness network (derived from yeast diploid strains) to a similar network, a yeast diploid agar-adhesion network. The adhesion network was created in parallel to the invasion network reported in Drees et al. [7] (data not shown), and although the two phenotypes are related, many genetic interactions differed between the two (652 of 1,751 (37.2%)). To compare the networks, we enumerated their 3-node motifs. For consistency, we pruned the networks such that they had exactly the same topological set of nodes (128) and edges (1,751). We found 27 motifs in both the invasion network and the adhesion network out of 419 and 414 candidate pat- terns (6.4% and 6.5%, respectively). Of these 27 motifs, 20 (74%) were common to both. This indicates that although common genetic-interaction motifs exist in the two networks, each genetic network also contains a unique subset. The fact 91 that these are related phenotypes underscores this observation. To further understand the different motif sample spaces of the two networks, we compared the null hypotheses generated by the invasion and adhesion networks. Using the 378 3n patterns common to both networks, we compared the mean num- ber of times each pattern occurred in the adhesion randomized network set to that of the invasion randomized network set. By making this comparison across all pat- terns, an understanding of how similar the global null hypotheses are is obtained [24]. The comparison was accomplished by calculating the correlation coefficient between the mean number of occurrences of the 378 network patterns in the adhe- sion and invasion randomized network, obtaining a value of 0.974. A completely correlated null hypothesis would have given a correlation coefficient close to 1, while a completely uncorrelated null hypothesis will give a value close to 0 (due to randomization). This shows that though the networks contain different motif sets, they display similar null hypotheses. These observations demonstrate the sig- nificance of the network comparison and suggest that there is no universal set of genetic-interaction motifs that will apply uniformly to all genetic-interaction net- works. Rather, analyses of each network will be necessary. 3.2.7 Open Source Software To facilitate the application of the analyses used in this study to other networks, we developed an open source software package entitled Network Motif Finder. Network Motif Finder was designed to identify motifs in any network type, and to include any number of edge and node types. Network Motif Finder acts as a plugin to the network analysis platform Cytoscape [27], and identifies significant multi-mode genetic interaction patterns. In addition, Network Motif Finder has the 92 functionality of extracting motif sub-networks as shown in Fig. 3.3 and 3.5. The plugin is available as open source, with a user manual, at [37]. 3.3 Conclusion In this study we develop methods to address the challenges of analyzing complex genetic-interaction networks. Specifically, we use statistical techniques to identify biologically significant multi-mode genetic interaction network patterns, network motifs. Utilizing randomized null hypotheses of the genetic network, those pat- terns that occur more frequently than randomly expected can be identified. These motifs highlight biologically informative network patterns of the genetic network. Further, the union of all instances of a motif forms a motif subnetwork. These subnetworks illustrate the distribution of the motif instances within the full genetic network. This allows for the identification of all genes involved in such a motif and can highlight those genes that dominate the motif’s occurrence. In this way, motif subnetworks extract the biological information that was identified by motif analysis. We also identified network motifs that reflect the underlying biochemical net- work. This was done by integrating our genetic network with gene-annotation data. In this way, we describe an unbiased approach to understand how genetic interac- tions reflect the biological properties of the underlying system. Lastly, this analy- sis has been developed into an open source plugin to the network analysis software Cytoscape, allowing users to analyze their own multi-mode genetic-interaction net- work datasets. 93 3.4 Materials and Methods 3.4.1 Network Randomization Statistical significance of each network pattern was calculated by comparing the number of times the pattern occurred in the observed genetic-interaction network, to a set of randomized networks. The randomized networks represent the null hy- pothesis. To ensure that pattern significance was due solely to the genetics of the system and not experimental design, we constrained our randomizations in the fol- lowing way. First, as described in the text, the topology of the genetic interaction network defines which genetic interaction experiments were conducted, while the interaction types describe the genetic results. Thus, in all our randomizations, the topology of the network is held constant and the genetic interaction types (edge colors) are switched. Second, as described in Drees et al. [7] and Additional data file 22 [20], each genetic interaction consists of the four phenotypes: φWT, φA, φB, φAB. These quantitative phenotypes are ordered into 1 of 75 possible genetic interaction inequalities, and the inequalities are grouped into 9 possible genetic interaction types. As the phenotypes of the single genetic perturbations (φA, φB) are dependent on experimental allele selection, it is necessary to avoid randomizing these single-gene phenotypes to prevent allele-selection bias in the results. Thus, in our Monte Carlo switching we strictly maintain the ordering of each edge’s single- perturbation and wild-type phenotypes (φWT, φA, φB). In all randomizations we uniformly chose a random pair of ordered edges and exchanged their genetic in- teraction types only if the inequality relationship of φWT, φA, and φB (regardless of φAB) was identical for both edges. In the case of nonidentical inequality rela- tionships, we retested after swapping the positions of φA and φB in the inequality 94 of the second edge of the pair and exchanged only if the resulting edge inequality relationship of φWT, φA, and φB was identical. These methods conserve the to- tal number of each genetic interaction edge type in all randomizations and ensure that statistical significance does not depend on initial experimental design or allele selection. We employed a Monte Carlo method of genetic-interaction edge-type switch- ing for the randomization algorithm. Each edge was switched in the Monte Carlo algorithm at least ten times per randomization. This level of switching has been shown to provide good mixing [16]. A sample size of 1,000 randomized networks to represent the null hypothesis was used for each analysis unless specified below. Modifications to this scheme were employed for the motifs involving annotation data and are described below. All algorithms are implemented in our open-source software package, Network Motif Finder. In the motif analyses including GOSlim annotations, the positions of the GOSlim node annotations were held constant, and only the genetic interaction types were randomized as described above. This ensures that the underlying molecular struc- ture of the system remains constant, while only the resulting genetic relationships are randomized. As well, we identified both 2-node and 3-node motifs. In the enumeration of 3-node network pattern instances the total number of 2-node net- work pattern instances was held constant. This ensures that the significance of a 3-node pattern is due to its 3-node architecture and not because it contained a sig- nificant 2-node pattern. Edge directions are conserved in this restriction. Also, the relationships between node annotations and the single gene perturbation data were maintained. Due to the extra calculations that are made during these random- izations this algorithm was much slower, particularly for the 3-node analysis. To 95 compensate, we reduced the sample size representing the null hypothesis in the 3-node analysis from 1,000 to 500. This null hypothesis reduction was conducted for the dual invasion/adhesion network comparison as well. Lastly, to avoid significance due to multiple testing, we corrected our signifi- cance threshold by applying the conservative Bonferroni correction. Specifically, a statistical threshold of p < 0.05/n was used, where n is the total number of patterns tested for significance in each analysis. For the 3n-motifs, 4n-motifs, 2nGO-motifs, and 3nGo-motifs, n was 489, 1,505, 575, and 23,286, respectively. To obtain a p value resolution greater than what is possible empirically (p < 1 × 10−3 for a 1,000 randomized network set), we parametrically fit the null hypothesis network pattern distributions to Gaussian (or Poisson when the pattern’s mean count was <3). Please see Additional data files 3, 6, 9, 20 and 21 [20] for the network pattern distributions and parametric fits. 3.4.2 Motif Enumeration Techniques In all analyses except those containing 4-node patterns, a full enumeration of the network pattern instances was conducted. However, this was not computationally feasible for the 4-node patterns, and a sampling algorithm was employed [28]. There are >3 ×106 individual 4-node network pattern instances in our analyzed network; we sampled 100,000 without replacement. This sample rate is compara- ble to those used in other sampling studies [13]. In enumerating network patterns involving GoSlim annotations, we needed to account for genes having multiple annotations. For instance, a particular GoSlim molecular function gene may be annotated as both a transferase and a protein ki- nase. In enumerating a specific network pattern, we allowed genes sharing a sin- 96 gle common annotation to be considered equal. For instance, consider the set of 1-node patterns annotated transferase, transferase/protein kinase, and protein ki- nase, respectively. In our scheme, we would have three patterns (transferase, trans- ferase/protein kinase, and protein kinase), containing two, three, and two instances, respectively. In the general motif analysis we identified motifs containing purely noninter- acting edge types. It is possible that these motifs occur due to gene perturbations irrelevant to the filamentation phenotype. In our analyses using GoSlim annota- tions, we included such motifs when stating the percentage of significant patterns, but removed them from the Additional data files to avoid highlighting relatively uninformative patterns. 3.4.3 GoSlim Molecular Function Annotations The GoSlim molecular function annotations were downloaded on 5 June 2006 from the Saccharomyces Genome Database [39]. 97 Bibliography [1] L Bardwell, J G Cook, D Voora, D M Baggott, A R Martinez, and J Thorner. Repression of yeast ste12 transcription factor by direct binding of unphos- phorylated kss1 mapk and its regulation by the ste7 mek. Genes Dev, 12(18):2887–98, Sep 1998. [2] L Bardwell, J G Cook, J X Zhu-Shimoni, D Voora, and J Thorner. Differen- tial regulation of transcription: repression by unactivated mitogen-activated protein kinase kss1 requires the dig1 and dig2 proteins. Proc Natl Acad Sci USA, 95(26):15400–5, Dec 1998. [3] Song Chou, Shelley Lane, and Haoping Liu. Regulation of mating and fil- amentation genes by two distinct ste12 complexes in saccharomyces cere- visiae. Mol Cell Biol, 26(13):4794–805, Jul 2006. [4] Sean R Collins, Kyle M Miller, Nancy L Maas, Assen Roguev, Jeffrey Fill- ingham, Clement S Chu, Maya Schuldiner, Marinella Gebbia, Judith Recht, Michael Shales, Huiming Ding, Hong Xu, Junhong Han, Kristin Ingvars- dottir, Benjamin Cheng, Brenda Andrews, Charles Boone, Shelley L Berger, Phil Hieter, Zhiguo Zhang, Grant W Brown, C James Ingles, Andrew Emili, 98 C David Allis, David P Toczyski, Jonathan S Weissman, Jack F Green- blatt, and Nevan J Krogan. Functional dissection of protein complexes in- volved in yeast chromosome biology using a genetic interaction map. Nature, 446(7137):806–10, Apr 2007. [5] K D Davenport, K E Williams, B D Ullmann, and M C Gustin. Activation of the saccharomyces cerevisiae filamentation/invasion pathway by osmotic stress in high-osmolarity glycogen pathway mutants. Genetics, 153(3):1091– 103, Nov 1999. [6] J W Dolan, C Kirkman, and S Fields. The yeast ste12 protein binds to the dna sequence mediating pheromone induction. Proc Natl Acad Sci USA, 86(15):5703–7, Aug 1989. [7] Becky L Drees, Vesteinn Thorsson, Gregory W Carter, Alexander W Rives, Marisa Z Raymond, Iliana Avila-Campillo, Paul Shannon, and Timothy Gal- itski. Derivation of genetic interaction networks from quantitative phenotype data. Genome Biol, 6(4):R38, Jan 2005. [8] V Gavrias, A Andrianopoulos, C J Gimeno, and W E Timberlake. Saccha- romyces cerevisiae tec1 is required for pseudohyphal growth. Mol Microbiol, 19(6):1255–63, Mar 1996. [9] L Giot, J S Bader, C Brouwer, A Chaudhuri, B Kuang, Y Li, Y L Hao, C E Ooi, B Godwin, E Vitols, G Vijayadamodar, P Pochart, H Machineni, M Welsh, Y Kong, B Zerhusen, R Malcolm, Z Varrone, A Collis, M Minto, S Burgess, L McDaniel, E Stimpson, F Spriggs, J Williams, K Neurath, N Ioime, M Agee, E Voss, K Furtak, R Renzulli, N Aanensen, S Carrolla, 99 E Bickelhaupt, Y Lazovatsky, A DaSilva, J Zhong, C A Stanyon, R L Fin- ley, K P White, M Braverman, T Jarvie, S Gold, M Leach, J Knight, R A Shimkets, M P McKenna, J Chant, and J M Rothberg. A protein interaction map of drosophila melanogaster. Science, 302(5651):1727–36, Dec 2003. [10] M A Harris, J Clark, A Ireland, J Lomax, M Ashburner, R Foulger, K Eilbeck, S Lewis, B Marshall, C Mungall, J Richter, G M Rubin, J A Blake, C Bult, M Dolan, H Drabkin, J T Eppig, D P Hill, L Ni, M Ringwald, R Balakrishnan, J M Cherry, K R Christie, M C Costanzo, S S Dwight, S Engel, D G Fisk, J E Hirschman, E L Hong, R S Nash, A Sethuraman, C L Theesfeld, D Botstein, K Dolinski, B Feierbach, T Berardini, S Mundodi, S Y Rhee, R Apweiler, D Barrell, E Camon, E Dimmer, V Lee, R Chisholm, P Gaudet, W Kibbe, R Kishore, E M Schwarz, P Sternberg, M Gwinn, L Hannick, J Wortman, M Berriman, V Wood, N de la Cruz, P Tonellato, P Jaiswal, T Seigfried, R White, and Gene Ontology Consortium. The gene ontology (go) database and informatics resource. Nucleic Acids Res, 32(Database issue):D258–61, Jan 2004. [11] Won-Ki Huh, James V Falvo, Luke C Gerke, Adam S Carroll, Russell W Howson, Jonathan S Weissman, and Erin K O’Shea. Global analysis of pro- tein localization in budding yeast. Nature, 425(6959):686–91, Oct 2003. [12] T Ito, T Chiba, R Ozawa, M Yoshida, M Hattori, and Y Sakaki. A compre- hensive two-hybrid analysis to explore the yeast protein interactome. Proc Natl Acad Sci USA, 98(8):4569–74, Apr 2001. [13] N Kashtan, S Itzkovitz, R Milo, and U Alon. Efficient sampling algorithm 100 for estimating subgraph concentrations and detecting network motifs. Bioin- formatics, 20(11):1746–58, Jul 2004. [14] Tong Ihn Lee, Nicola J Rinaldi, François Robert, Duncan T Odom, Ziv Bar- Joseph, Georg K Gerber, Nancy M Hannett, Christopher T Harbison, Craig M Thompson, Itamar Simon, Julia Zeitlinger, Ezra G Jennings, Heather L Mur- ray, D Benjamin Gordon, Bing Ren, John J Wyrick, Jean-Bosco Tagne, Thomas L Volkert, Ernest Fraenkel, David K Gifford, and Richard A Young. Transcriptional regulatory networks in saccharomyces cerevisiae. Science, 298(5594):799–804, Oct 2002. [15] H D Madhani, C A Styles, and G R Fink. Map kinases with distinct in- hibitory functions impart signaling specificity during yeast differentiation. Cell, 91(5):673–84, Nov 1997. [16] R Milo, S Shen-Orr, S Itzkovitz, N Kashtan, D Chklovskii, and U Alon. Network motifs: simple building blocks of complex networks. Science, 298(5594):824–7, Oct 2002. [17] H U Mösch, E Kübler, S Krappmann, G R Fink, and G H Braus. Crosstalk be- tween the ras2p-controlled mitogen-activated protein kinase and camp path- ways during invasive growth of saccharomyces cerevisiae. Mol Biol Cell, 10(5):1325–35, May 1999. [18] S M O’Rourke and I Herskowitz. The hog1 mapk prevents cross talk be- tween the hog and pheromone response mapk pathways in saccharomyces cerevisiae. Genes Dev, 12(18):2874–86, Sep 1998. 101 [19] F Posas and H Saito. Osmotic activation of the hog mapk pathway via ste11p mapkkk: scaffold role of pbs2p mapkk. Science, 276(5319):1702–5, Jun 1997. [20] T Galitski R J Taylor, A F Siegel. Network motif analysis of a multi-mode genetic interaction network; supplemental materials (http://genomebiology.com/2007/8/8/r160). [21] D C Raitt, F Posas, and H Saito. Yeast cdc42 gtpase and ste20 pak-like kinase regulate sho1-dependent activation of the hog1 mapk pathway. EMBO J, 19(17):4623–31, Sep 2000. [22] Massoud Ramezani-Rad. The role of adaptor protein ste50-dependent reg- ulation of the mapkkk ste11 in multiple signalling pathways of yeast. Curr Genet, 43(3):161–70, Jun 2003. [23] S Rupp, E Summers, H J Lo, H Madhani, and G Fink. Map kinase and camp filamentation signaling pathways converge on the unusually large promoter of the yeast flo11 gene. EMBO J, 18(5):1257–69, Mar 1999. [24] Maya Schuldiner, Sean R Collins, Natalie J Thompson, Vladimir Denic, Arunashree Bhamidipati, Thanuja Punna, Jan Ihmels, Brenda Andrews, Charles Boone, Jack F Greenblatt, Jonathan S Weissman, and Nevan J Kro- gan. Exploration of the function and organization of the yeast early secretory pathway through an epistatic miniarray profile. Cell, 123(3):507–19, Nov 2005. [25] B Schwikowski, P Uetz, and S Fields. A network of protein-protein interac- tions in yeast. Nat Biotechnol, 18(12):1257–61, Dec 2000. 102 [26] Daniel Segrè, Alexander Deluna, George M Church, and Roy Kishony. Mod- ular epistasis in yeast metabolism. Nat Genet, 37(1):77–83, Jan 2005. [27] Paul Shannon, Andrew Markiel, Owen Ozier, Nitin S Baliga, Jonathan T Wang, Daniel Ramage, Nada Amin, Benno Schwikowski, and Trey Ideker. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res, 13(11):2498–504, Nov 2003. [28] Shai S Shen-Orr, Ron Milo, Shmoolik Mangan, and Uri Alon. Network mo- tifs in the transcriptional regulation network of escherichia coli. Nat Genet, 31(1):64–8, May 2002. [29] Kazuo Tatebayashi, Mutsuhiro Takekawa, and Haruo Saito. A docking site determining specificity of pbs2 mapkk for ssk2/ssk22 mapkkks in the yeast hog pathway. EMBO J, 22(14):3624–34, Jul 2003. [30] K Tedford, S Kim, D Sa, K Stevens, and M Tyers. Regulation of the mat- ing pheromone and invasive growth responses in yeast by two map kinase substrates. Curr Biol, 7(4):228–38, Apr 1997. [31] Amy Hin Yan Tong, Guillaume Lesage, Gary D Bader, Huiming Ding, Hong Xu, Xiaofeng Xin, James Young, Gabriel F Berriz, Renee L Brost, Michael Chang, YiQun Chen, Xin Cheng, Gordon Chua, Helena Friesen, Debra S Goldberg, Jennifer Haynes, Christine Humphries, Grace He, Shamiza Hus- sein, Lizhu Ke, Nevan Krogan, Zhijian Li, Joshua N Levinson, Hong Lu, Patrice Ménard, Christella Munyana, Ainslie B Parsons, Owen Ryan, Raffi Tonikian, Tania Roberts, Anne-Marie Sdicu, Jesse Shapiro, Bilal Sheikh, Bernhard Suter, Sharyl L Wong, Lan V Zhang, Hongwei Zhu, Christopher G 103 Burd, Sean Munro, Chris Sander, Jasper Rine, Jack Greenblatt, Matthias Pe- ter, Anthony Bretscher, Graham Bell, Frederick P Roth, Grant W Brown, Brenda Andrews, Howard Bussey, and Charles Boone. Global mapping of the yeast genetic interaction network. Science, 303(5659):808–13, Feb 2004. [32] Sharyl L Wong, Lan V Zhang, Amy H Y Tong, Zhijian Li, Debra S Gold- berg, Oliver D King, Guillaume Lesage, Marc Vidal, Brenda Andrews, Howard Bussey, Charles Boone, and Frederick P Roth. Combining bio- logical networks to predict genetic interactions. Proc Natl Acad Sci USA, 101(44):15682–7, Nov 2004. [33] Esti Yeger-Lotem, Shmuel Sattath, Nadav Kashtan, Shalev Itzkovitz, Ron Milo, Ron Y Pinter, Uri Alon, and Hanah Margalit. Network motifs in inte- grated cellular networks of transcription-regulation and protein-protein inter- action. Proc Natl Acad Sci USA, 101(16):5934–9, Apr 2004. [34] Julia Zeitlinger, Itamar Simon, Christopher T Harbison, Nancy M Hannett, Thomas L Volkert, Gerald R Fink, and Richard A Young. Program-specific distribution of a transcription factor dependent on partner transcription factor and mapk signaling. Cell, 113(3):395–404, May 2003. [35] Lan V Zhang, Oliver D King, Sharyl L Wong, Debra S Goldberg, Amy H Y Tong, Guillaume Lesage, Brenda Andrews, Howard Bussey, Charles Boone, and Frederick P Roth. Motifs, themes and thematic maps of an integrated saccharomyces cerevisiae interaction network. J Biol, 4(2):6, Jan 2005. 104 Chapter 4 Conclusions and Recommendations for Further Work 4.1 Summary of Thesis In this thesis I describe novel experimental and computational methods for analyz- ing complex biomolecular systems. These methods extend the toolset available to cell biologists, and allow for the analyses of questions not previously answerable. Each technology was applied to the study of cellular signaling in Saccharomyces cerevisiae and novel understandings were achieved. In Chapter 2 I present a novel microfluidic platform for high-throughput sin- gle cell analysis of signaling pathways under complex environmental conditions. 105 Our platform leveraged recent advances in microfluidic technologies to allow for the design of highly integrated fluidic handling systems. Using a parallel assay architecture we conducted over 3,000 live cell experiments across three major ex- periment types. These included the analysis of signaling activation using constant stimulation conditions, analysis of signaling deactivation using transient stimula- tion conditions, and analysis of signaling memory using periodic stimulation con- ditions. For these experiment types we mapped cellular response across a broad range of stimuli parameters including stimuli strength, duration, and frequency. We used our technology to study the mating response in Saccharomyces cerevisiae under dynamic stimuli conditions, comparing a set 11 gene deletions to WT to im- plicate genes involved in the dynamic regulation of mating response. Our results uncovered cellular regulation not observable under static conditions, emphasizing the need to study cellular systems in dynamic environments. In Chapter 3 I present computational work to extract biologically relevant in- formation from a complex multi-mode genetic interaction network. Here I identify network patterns that occur more often than expected by random, implying biolog- ical relevance. The identified patterns revealed common regulatory themes within the system including that gene perturbations often interact similarly with a broad class second perturbations. By compiling all motif instances of a specific type, I was able to identify motif sub-networks that delineated information flow through the filamentation system. For example, by extracting all instances of a significantly occurring 3-node epistatic network, genes representing key information hubs were identified. Finally, through analysis of a second similar genetic interaction net- work, we determined that a universal set of significant genetic interaction patterns 106 is unlikely. 4.2 Dynamic Single Cell Analysis 4.2.1 Discussion Current methods for the analysis of many single cells rely on either fluorescent- activated cell sorting (FACS) [13] or high-throughput multi-well plate imaging[4]. Both methods excel in certain experiment types but do not provide complete sin- gle cell solutions. For instance, FACS has the ability to rapidly acquire fluores- cent reporter expression data from tens of thousands of individual cells creating robust distributions of population response. Yet due to its fluorescent-only and flow-through methods this technique is unable to reliably obtain morphology data or measure the same cell multiple times. Multi-well high-throughput imaging has the ability to measure morphology repeatedly on the same set of cells, however it is extremely difficult to control the microenvironment and studies in time-varying conditions are not realistic. As demonstrated in Chapter 2 the ability to measure the combination of morphology data, the same set of cells repeatedly, and cell re- sponse to time-varying environments is critical in the analysis of dynamic cellular networks. The issues of studying single cells in complex dynamic environments has been approached in other recent studies. For example, Bennett et al. [11] ,Hersen et al. [29] and Mettetal et al. [41] have demonstrated the utility of studying signaling systems under dynamic conditions. These works however, were limited by their use of serial low-throughput microfluidic technologies and as a result each study 107 focused only a small number of environmental conditions and genotypes. Other studies have developed devices for parallel cell culture, but lack either precise on- chip fluidic control for high-throughput media exchange [28, 30, 33] or the ability to trap non-adherent cell types [27, 30, 33]. These limitations inhibit the ability to test the vast number of environmental conditions required to reverse-engineer large protein networks[51], as well as the ability to test important cell types including many single cell organisms like yeast and bacteria. The technology we present in Chapter 2 combine microfluidic large scale integration, novel cell-trapping meth- ods, and high-throughput image analysis to enable hundreds of simultaneous live cell-imaging experiments under programmable time-varying conditions. We applied our technology to the study of dynamic signaling in the mating response in Saccharomyces cerevisiae. This was, to the best of our knowledge, the most comprehensive quantitative analysis of dynamic cellular signaling to date and enabled us to uncover many novel regulatory processes of the mating pathway. For instance, we uncovered that the mating system has the ability to remember prior stimulation and that this memory dissipates over time. Previous studies have investigated memory in other cellular networks, including in the yeast galactose induction system [1], . These studies analyzed cellular circuits over long periods of stimulation [1] and in very limited number of conditions [40, 44]. Utilizing our microfluidic platform we were able to examine cellular memory to brief stimula- tion under many different conditions and genotypes. We first stimulated wild type cells periodically under many different stimulation strengths and frequencies. We compared our measured results to an in silico model simulating a null hypothesis of a memoryless response, and found that as stimuli frequency increased, cellu- lar response became greater than our model predicted. This indicated that as time 108 between stimulation decreased, cells were influenced by their previous response. We speculate that this frequency dependent memory is due to the relaxation time needed for the system to return to a pre-stimulated state upon stimuli removal. In this scenario, if a deactivating system is stimulated again before it has a chance to fully recover, the system may respond greater than if it was stimulated from a fully deactivated state. By stimulating the system with different frequencies we were able to obtain a characteristic time for pathway recovery. We found that for 10 min stimulation pulses, the responses of wild type yeast differed greatly between short (15 min, 40 min) versus long (65 min, 140 min) deactivation times indicat- ing the cells likely need between 40 - 65 min to fully return to a pre-stimulated state. Further analysis of mutants comprised of deletions of genes involved in the mating system implicated network components involved in this memory response. Interestingly, mutant analysis in these dynamic environments uncovered hypersen- sitivity phenotypes that were not-observable under static conditions, supporting the notion that cellular systems contain important circuitry for regulating time-varying stimuli. This has broad implications for cell signaling research, particularly that it demonstrates that much cell circuitry will not be discoverable by studying systems in static conditions, the primary technique used today. 4.2.2 Future Directions Our platform can be applied broadly enabling many areas of future study, includ- ing further analysis of yeast signaling networks and application of the technology to study mammalian cells. In the continuing study of yeast signaling, our abil- ity to generate defined complex environments should be used to ask important yet difficult questions in cell biology, for example how cells respond to dynamically 109 changing stimuli [11, 29, 41] and how signaling systems are able to integrate mul- tiple types of information using a shared set of components [3, 6, 17, 22, 24, 39]. In Chapter 2 we lay the foundation for the former, and further work should be conducted to obtain a circuit-level understanding of how the mating system regu- lates dynamically changing environments. Our initial analysis implicated the key MAP kinases, Kss1 and Fus3, and the key phosphatases, Msg5 and Ptp2, in this circuitry: these mutants each displayed a hypersensitive response under period stimulation in a frequency dependent manner that is not observable under con- stant stimulation conditions. Quantitative analysis revealed that the hypersensi- tivity profiles between homologous kinases or between homologous phosphatases were different, indicating that these components have unique dynamic regulatory roles. Similar to our initial genetic screen, the reverse engineering of this circuitry will require a combination of microfluidics, genetic perturbations, and molecular reporters. However, whereas we initially used full gene knockouts to implicate genes in regulatory processes, more sophisticated genetic perturbations will be re- quired to investigate circuit dynamics. For instance, Fus3 and Kss1 are highly ho- mologous MAP kinases [9] and under certain conditions, the loss of one kinase is can be masked by the other ([38], Chapter 2). Previous studies have stratified func- tions specific to Fus3 or Kss1 by using a collection of mutant alleles that allow for the disruption of protein function without full deletion. These include alleles that are catalytically inactive (removing kinase function)[37, 48], un-phosphorylatable [8] (removing kinase function and affecting inhibitory protein binding activity), or disrupt protein-binding interactions [16] . It is currently unclear if dynamic reg- ulation relies primarily on the kinase or protein-protein binding activities of Fus3 and Kss1 (or both), and testing response to dynamic stimuli using a collection of 110 these alleles can begin to tease these roles apart. Further, techniques using highly specific small-molecule kinase inhibitors for Fus3 and Kss in combination with site directed mutagensis enable real-time perturbation of kinase activity [12, 50]. Such techniques allow for dynamic network perturbations by modulating the lo- cal chemical environment (adding or subtracting the inhibitor to growth media), which is ideally suited to our microfluidic platform. Using these tools, further insights into the memory response can be obtained. For example, testing if the system’s frequency-dependent hypersensitivity is reversible or not could be easily tested by inhibiting kinase activity for a duration and then release by washing out the inhibitor. Lastly, Fus3 has recently been implicated in a negative feedback loop responsible for down regulating signal response following stimulation [50]. Due to their dynamic nature, feedback loops are particularly relevant time-varying cir- cuit connections and disruption of such interactions may uncover specific dynamic regulatory mechanisms. Lastly, computational models of the pathway should be employed to compare experimental results to quantitatively described hypothe- ses. Many computational model classes exist including those that successfully model circuitry dynamics (differential equation models) and cell-to-cell variability (stochastic models). A second important area of future development will be in the transfer of our technology to mammalian cell culture. Simple modifications can be made to our design to allow for mammalian cells to grow and be analyzed in chip [27]. Such work has implications in both mammalian cellular research as well as in drug dis- covery. In particular, our work analyzing yeast cellular response to complex envi- ronmental conditions is directly translatable to mammalian systems. Mammalian signaling networks are presumed larger and integrate more environmental cues, 111 and thus the need for sophisticated experimental tools is even greater. In addition, our platform acts as a live-cell high-content screening platform for measuring com- bined chemical and genetic interactions. Systems biology theory predicts that drug therapy of disease networks likely requires chemical perturbations of networks at multiple critical foci [2, 7, 34]. This indicates that drug combinations will be re- quired to target disease to achieve desired efficacy and to avoid drug resistance, with drug combinations being applied in parallel or in sequence[35]. The use of high-throughput microfluidic technoloqies has already found success in the field of drug discovery [21], and the immediate application of our platform into the areas of oncology and infectious disease are realistic. 4.3 Systems Genetics 4.3.1 Discussion The work in Chapter 3 revealed that complex genetic networks contain emergent information that is not observable from the set of interactions alone: biologically meaningful patterns can be identified once interactions are abstracted into a net- work format. Extraction of such information first requires the construction of dense genetic networks followed by network analyses with sophisticated computational algorithms. In this thesis I focused on the computational aspects of genetic analy- sis, leveraging recent advancements in the field of complex systems. Complex system analysis has rapidly matured, largely due to the excitement generated by the recent flux of new large biomolecular network data sets [23, 26, 32, 36, 46, 47]. One such advance is in the identification of network motifs, sig- 112 nificant network patterns that are biologically informative. Network motif analysis has been primarily applied to physical interactions networks, for example networks of protein-to-DNA or protein-to-protein interactions. Our work(Chapter 3, [45] ), and that of Zang et al [52], were the first examples to apply network motif tech- niques to study genetic interaction network data. In Zang et al [52] network motifs were identified in a single edge-type (’synthetic lethal’) genetic interaction net- work. They used network motif analysis to globally integrate their genetic network with other network types like physical interaction networks and gene co-expression networks. This method was successful in identifying global thematic maps of cel- lular regulation, however their use of single genetic interaction edge-type funda- mentally limits the resolution of the functional maps. In our study (Chapter 3) we analyzed a dense multi-color interaction network that comprised of 9 genetic interaction types. Our analysis identified significant patterns of genetic interac- tion that often consisted of multiple genetic types, demonstrating the additional information obtained by studying multi-color genetic interaction networks. We compiled all instances of a significant pattern into a motif sub-networks to identify paths of information flow through the underlying biological system. For example, the aggregation all instances of a significant epistatic interaction pattern uncovered network nodes that are key points of information processing, and by aggregating all instances of certain multi-color pattern integrated with functional ontology data uncovered overlapping pathway functionality between a set of protein binding pro- teins and a set of transcription factors. The richness of the multi-colored genetic interaction network allowed us interrogate our specific signaling system at high- resolution, a technique that has gained favor in recent years [18, 19, 43]. 113 4.3.2 Future Directions Our methods can be applied broadly and future analyses should be extended to a diverse set of genetic interaction data sets. An obvious direction will be to ap- ply our methods to other pathways and organisms. For example, with the advent of RNA interference gene knock-downs[42], a rush of genetic studies are being completed in higher organisms, for example in worms[10, 49] and flies[5]. This will allow for similar analyses to ours to be applied to these data sets, helping to identify information flows through multicellular regulatory systems. In addition, in Chapter 3 we demonstrated that the overlapping set of signification genetic net- work patterns using two similar phenotypes, the filamentation system in diploid yeast and the invasion system in hapliod yeast, did not product a universal set of significant genetic patterns. It would be interesting to see if in the analysis of many genetic networks a universal set would emerge. A second future direction will be to apply our methods to the analysis of net- works relating gene variation to disease susceptibility. Advances in genotyping has enabled researchers to begin exploring how allelic variation leads to complex phe- notypes in humans. Genome-wide association studies, which rely on population- based samples to associate disease with common genetic traits, has become very popular of late [20, 25]. These studies primarily focus on independently associating single gene variation with disease, and although many significant associations have been identified, these associations have often shown weak penetrance[31]. Efforts to expand single gene associations to multi-genetic variations have begun. These efforts lead to network based approaches[15] and are amenable complex system based analyses like those presented in this thesis. As such, the future increase in 114 human genetic networks will demand the further development sophisticated com- putational algorithms. 115 Bibliography [1] Murat Acar, Attila Becskei, and Alexander van Oudenaarden. Enhance- ment of cellular memory by reducing stochastic transitions. Nature, 435(7039):228–32, May 2005. [2] R Albert, H Jeong, and A Barabasi. Error and attack tolerance of complex networks. Nature, 406(6794):378–82, Jul 2000. [3] Jessica Andersson, David M Simpson, Maosong Qi, Yunmei Wang, and Elaine A Elion. Differential input by ste5 scaffold and msg5 phosphatase route a mapk cascade to multiple outcomes. EMBO J, 23(13):2564–76, Jul 2004. [4] C Bakal, J Aach, G Church, and N Perrimon. Quantitative morphological sig- natures define local signaling networks regulating cell morphology. Science, 316(5832):1753–1756, Jan 2007. [5] Chris Bakal, Rune Linding, Flora Llense, Elleard Heffern, Enrique Martin- Blanco, Tony Pawson, and Norbert Perrimon. Phosphorylation networks reg- ulating jnk activity in diverse genetic backgrounds. Science, 322(5900):453– 6, Oct 2008. 116 [6] Marie Z Bao, Monica A Schwartz, Greg T Cantin, John R Yates, and Hiten D Madhani. Pheromone-dependent destruction of the tec1 transcription factor is required for map kinase signaling specificity in yeast. Cell, 119(7):991–1000, Dec 2004. [7] Albert-László Barabási and Zoltán N Oltvai. Network biology: understanding the cell’s functional organization. Nat Rev Genet, 5(2):101–13, Feb 2004. [8] L Bardwell, J G Cook, D Voora, D M Baggott, A R Martinez, and J Thorner. Repression of yeast ste12 transcription factor by direct binding of unphos- phorylated kss1 mapk and its regulation by the ste7 mek. Genes Dev, 12(18):2887–98, Sep 1998. [9] Lee Bardwell. A walk-through of the yeast mating pheromone response path- way. Peptides, 25(9):1465–76, Sep 2004. [10] L Ryan Baugh, Joanne C Wen, Andrew A Hill, Donna K Slonim, Eugene L Brown, and Craig P Hunter. Synthetic lethal analysis of caenorhabditis ele- gans posterior embryonic patterning genes identifies conserved genetic inter- actions. Genome Biol, 6(5):R45, Jan 2005. [11] M Bennett, W Pang, N Ostroff, B Baumgartner, S Nayak, L Tsimring, and J Hasty. Metabolic gene regulation in a dynamically changing environment. Nature, 454:1119–1122, Jul 2008. [12] A C Bishop, J A Ubersax, D T Petsch, D P Matheos, N S Gray, J Blethrow, E Shimizu, J Z Tsien, P G Schultz, M D Rose, J L Wood, D O Morgan, and K M Shokat. A chemical switch for inhibitor-sensitive alleles of any protein kinase. Nature, 407(6802):395–401, Sep 2000. 117 [13] W Bonner, H Hulett, R Sweet, and L Herzenberg. Fluorescence activated cell sorting. Review of Scientific Instruments, Jan 1972. [14] J M. Bower and H Bolouri. Computational modeling of genetic and biochem- ical networks. page 356, Jan 2004. [15] Yanqing Chen, Jun Zhu, Pek Yee Lum, Xia Yang, Shirly Pinto, Dou- glas J MacNeil, Chunsheng Zhang, John Lamb, Stephen Edwards, Solveig K Sieberts, Amy Leonardson, Lawrence W Castellini, Susanna Wang, Marie- France Champy, Bin Zhang, Valur Emilsson, Sudheer Doss, Anatole Ghaz- alpour, Steve Horvath, Thomas A Drake, Aldons J Lusis, and Eric E Schadt. Variations in dna elucidate molecular networks that cause disease. Nature, 452(7186):429–35, Mar 2008. [16] K Y Choi, B Satterberg, D M Lyons, and E A Elion. Ste5 tethers multiple protein kinases in the map kinase cascade required for mating in s. cerevisiae. Cell, 78(3):499–512, Aug 1994. [17] Song Chou, Lan Huang, and Haoping Liu. Fus3-regulated tec1 degrada- tion through scfcdc4 determines mapk signaling specificity during mating in yeast. Cell, 119(7):981–90, Dec 2004. [18] Sean R Collins, Kyle M Miller, Nancy L Maas, Assen Roguev, Jeffrey Fill- ingham, Clement S Chu, Maya Schuldiner, Marinella Gebbia, Judith Recht, Michael Shales, Huiming Ding, Hong Xu, Junhong Han, Kristin Ingvars- dottir, Benjamin Cheng, Brenda Andrews, Charles Boone, Shelley L Berger, Phil Hieter, Zhiguo Zhang, Grant W Brown, C James Ingles, Andrew Emili, C David Allis, David P Toczyski, Jonathan S Weissman, Jack F Green- 118 blatt, and Nevan J Krogan. Functional dissection of protein complexes in- volved in yeast chromosome biology using a genetic interaction map. Nature, 446(7137):806–10, Apr 2007. [19] Sean R Collins, Maya Schuldiner, Nevan J Krogan, and Jonathan S Weiss- man. A strategy for extracting and analyzing large-scale quantitative epistatic interaction data. Genome Biol, 7(7):R63, Jan 2006. [20] Wellcome Trust Case Control Consortium. Genome-wide association study of 14,000 cases of seven common diseases and 3,000 shared controls. Nature, 447(7145):661–78, Jun 2007. [21] S Einav, D Gerber, P Bryson, E Sklan, M Elazar, S Maerkl, J Glenn, and S Quake. Discovery of a hepatitis c target and its pharmacological inhibitors by microfluidic affinity analysis. Nat Biotechnol, 26:1019–1027, Aug 2008. [22] Elaine A Elion, Maosong Qi, and Weidong Chen. Signal transduction. sig- naling specificity in yeast. Science, 307(5710):687–8, Feb 2005. [23] Rob M Ewing, Peter Chu, Fred Elisma, Hongyan Li, Paul Taylor, Shane Climie, Linda McBroom-Cerajewski, Mark D Robinson, Liam O’Connor, Michael Li, Rod Taylor, Moyez Dharsee, Yuen Ho, Adrian Heilbut, Lynda Moore, Shudong Zhang, Olga Ornatsky, Yury V Bukhman, Martin Ethier, Yinglun Sheng, Julian Vasilescu, Mohamed Abu-Farha, Jean-Philippe Lam- bert, Henry S Duewel, Ian I Stewart, Bonnie Kuehl, Kelly Hogue, Karen Col- will, Katharine Gladwish, Brenda Muskat, Robert Kinach, Sally-Lin Adams, Michael F Moran, Gregg B Morin, Thodoros Topaloglou, and Daniel Figeys. 119 Large-scale mapping of human protein-protein interactions by mass spec- trometry. Mol Syst Biol, 3:89, Jan 2007. [24] Laura J Flatauer, Sheena F Zadeh, and Lee Bardwell. Mitogen-activated pro- tein kinases with distinct requirements for ste5 scaffolding influence signaling specificity in saccharomyces cerevisiae. Mol Cell Biol, 25(5):1793–803, Mar 2005. [25] K Frazer, D Ballinger, D Cox, and D Hinds. A second generation human haplotype map of over 3.1 million snps. Nature, 449:851–861, Jan 2007. [26] L Giot, J S Bader, C Brouwer, A Chaudhuri, B Kuang, Y Li, Y L Hao, C E Ooi, B Godwin, E Vitols, G Vijayadamodar, P Pochart, H Machineni, M Welsh, Y Kong, B Zerhusen, R Malcolm, Z Varrone, A Collis, M Minto, S Burgess, L McDaniel, E Stimpson, F Spriggs, J Williams, K Neurath, N Ioime, M Agee, E Voss, K Furtak, R Renzulli, N Aanensen, S Carrolla, E Bickelhaupt, Y Lazovatsky, A DaSilva, J Zhong, C A Stanyon, R L Fin- ley, K P White, M Braverman, T Jarvie, S Gold, M Leach, J Knight, R A Shimkets, M P McKenna, J Chant, and J M Rothberg. A protein interaction map of drosophila melanogaster. Science, 302(5651):1727–36, Dec 2003. [27] Rafael Gómez-Sjöberg, Anne A Leyrat, Dana M Pirone, Christopher S Chen, and Stephen R Quake. Versatile, fully automated, microfluidic cell culture system. Anal Chem, 79(22):8557–63, Nov 2007. [28] Alex Groisman, Caroline Lobo, HoJung Cho, J Kyle Campbell, Yann S Du- four, Ann M Stevens, and Andre Levchenko. A microfluidic chemostat for 120 experiments with bacterial and yeast cells. Nat Methods, 2(9):685–9, Sep 2005. [29] Pascal Hersen, Megan N McClean, L Mahadevan, and Sharad Ramanathan. Signal processing by the hog map kinase pathway. Proc Natl Acad Sci USA, 105(20):7165–70, May 2008. [30] Paul J Hung, Philip J Lee, Poorya Sabounchi, Robert Lin, and Luke P Lee. Continuous perfusion microfluidic cell culture array for high-throughput cell- based assays. Biotechnol Bioeng, 89(1):1–8, Jan 2005. [31] Mark M Iles. What can genome-wide association studies tell us about the genetics of common disease? Plos Genet, 4(2):e33, Feb 2008. [32] T Ito, T Chiba, R Ozawa, M Yoshida, M Hattori, and Y Sakaki. A compre- hensive two-hybrid analysis to explore the yeast protein interactome. Proc Natl Acad Sci USA, 98(8):4569–74, Apr 2001. [33] Kevin R King, Sihong Wang, Daniel Irimia, Arul Jayaraman, Mehmet Toner, and Martin L Yarmush. A high-throughput microfluidic real-time gene ex- pression living cell array. Lab on a chip, 7(1):77–85, Jan 2007. [34] Hiroaki Kitano. Biological robustness. Nat Rev Genet, 5(11):826–37, Nov 2004. [35] Hiroaki Kitano. A robustness-based approach to systems-oriented drug de- sign. Nature reviews Drug discovery, 6(3):202–10, Mar 2007. [36] Tong Ihn Lee, Nicola J Rinaldi, François Robert, Duncan T Odom, Ziv Bar- Joseph, Georg K Gerber, Nancy M Hannett, Christopher T Harbison, Craig M 121 Thompson, Itamar Simon, Julia Zeitlinger, Ezra G Jennings, Heather L Mur- ray, D Benjamin Gordon, Bing Ren, John J Wyrick, Jean-Bosco Tagne, Thomas L Volkert, Ernest Fraenkel, David K Gifford, and Richard A Young. Transcriptional regulatory networks in saccharomyces cerevisiae. Science, 298(5594):799–804, Oct 2002. [37] D Ma, J G Cook, and J Thorner. Phosphorylation and localization of kss1, a map kinase of the saccharomyces cerevisiae pheromone response pathway. Mol Biol Cell, 6(7):889–909, Jul 1995. [38] H D Madhani and G R Fink. The riddle of map kinase signaling specificity. Trends Genet, 14(4):151–5, Apr 1998. [39] Seth Maleri, Qingyuan Ge, Elizabeth A Hackett, Yuqi Wang, Henrik G Dohlman, and Beverly Errede. Persistent activation by constitutive ste7 pro- motes kss1-mediated invasive growth but fails to support fus3-dependent mat- ing in yeast. Mol Cell Biol, 24(20):9221–38, Oct 2004. [40] Megan N McClean, Areez Mody, James R Broach, and Sharad Ramanathan. Cross-talk and decision making in map kinase pathways. Nat Genet, 39(3):409–14, Mar 2007. [41] Jerome T Mettetal, Dale Muzzey, Carlos Gómez-Uribe, and Alexander van Oudenaarden. The frequency dependence of osmo-adaptation in saccha- romyces cerevisiae. Science, 319(5862):482–4, Jan 2008. [42] Jason Moffat and David M Sabatini. Building mammalian signalling path- ways with rnai screens. Nat Rev Mol Cell Biol, 7(3):177–87, Mar 2006. 122 [43] Maya Schuldiner, Sean R Collins, Natalie J Thompson, Vladimir Denic, Arunashree Bhamidipati, Thanuja Punna, Jan Ihmels, Brenda Andrews, Charles Boone, Jack F Greenblatt, Jonathan S Weissman, and Nevan J Kro- gan. Exploration of the function and organization of the yeast early secretory pathway through an epistatic miniarray profile. Cell, 123(3):507–19, Nov 2005. [44] Alex Sigal, Ron Milo, Ariel Cohen, Naama Geva-Zatorsky, Yael Klein, Yu- valal Liron, Nitzan Rosenfeld, Tamar Danon, Natalie Perzov, and Uri Alon. Variability and memory of protein levels in human cells. Nature, 444:643– 646, Jan 2006. [45] R James Taylor, Andrew F Siegel, and Timothy Galitski. Network motif anal- ysis of a multi-mode genetic-interaction network. Genome Biol, 8(8):R160, Jan 2007. [46] P Uetz, L Giot, G Cagney, T A Mansfield, R S Judson, J R Knight, D Lock- shon, V Narayan, M Srinivasan, P Pochart, A Qureshi-Emili, Y Li, B God- win, D Conover, T Kalbfleisch, G Vijayadamodar, M Yang, M Johnston, S Fields, and J M Rothberg. A comprehensive analysis of protein-protein in- teractions in saccharomyces cerevisiae. Nature, 403(6770):623–7, Feb 2000. [47] P Uetz and R E Hughes. Systematic and large-scale two-hybrid screens. Curr Opin Microbiol, 3(3):303–8, Jun 2000. [48] F van Drogen, V M Stucke, G Jorritsma, and M Peter. Map kinase dynamics in response to pheromones in budding yeast. Nat Cell Biol, 3(12):1051–9, Dec 2001. 123 [49] Gijs van Haaften, Nadine L Vastenhouw, Ellen A A Nollen, Ronald H A Plas- terk, and Marcel Tijsterman. Gene interactions in the dna damage-response pathway identified by genome-wide rna-interference analysis of synthetic lethality. Proc Natl Acad Sci USA, 101(35):12992–6, Aug 2004. [50] Richard C Yu, C Gustavo Pesce, Alejandro Colman-Lerner, Larry Lok, David Pincus, Eduard Serra, Mark Holl, Kirsten Benjamin, Andrew Gordon, and Roger Brent. Negative feedback that improves information transmission in yeast signalling. Nature, 456(7223):755–61, Dec 2008. [51] Eli Zamir and Philippe I H Bastiaens. Reverse engineering intracellular bio- chemical networks. Nat Chem Biol, 4(11):643–7, Nov 2008. [52] Lan V Zhang, Oliver D King, Sharyl L Wong, Debra S Goldberg, Amy H Y Tong, Guillaume Lesage, Brenda Andrews, Howard Bussey, Charles Boone, and Frederick P Roth. Motifs, themes and thematic maps of an integrated saccharomyces cerevisiae interaction network. J Biol, 4(2):6, Jan 2005. 124 Appendix A Appendix to: Dynamic Analysis of MAPK Signaling Using a Microfluidic Live-Cell Imaging Matrix 1 1A version of this chapter has been accepted for publication. R. J. Taylor, D. Falconnet, A. Niemist, S. A. Ramsey, S. Prinz, I. Shmulevich, T. Galitski and C. L. Hansen. Dynamic Analysis of MAPK Signaling Using a Microfluidic Live-Cell Imaging Matrix. Proceedings of the National Academy of Sciences. 125 A.1 SI Methods A.1.1 Fabrication Protocol Devices were made using Multilayer Soft Lithography in which consecutive replica molding and bonding steps are used to realized monolithic multilayer devices. Pho- tolithography masks were designed using AutoCAD software (Autodesk, Inc., San Rafael, CA) and used to generate high-resolution (20,000 dpi) transparency masks (CAD/Art Services). Molds were fabricated by photolithography on 4 inch silicon wafers (Silicon Quest International, Santa Clara, CA). The flow layer consisted of two different channel profiles: 3.5 µm high rectangular trapping channels allowing for sieve valving, and 12 µm high rounded channels used for standard flow. The 3.5 µm layer was made with SU8-5 negative photoresist (Microchem Corp., Newton, MA) and the 12 µm rounded layer was made with SPR220-7 positive photoresist (Microchem Corp.). The control master was a single layer mold consisting of 25 µm high squared features made with SU8-2025 negative photoresist (Microchem Corp.). Resist processing was performed according to the manufacturer’s specifi- cations. A.1.2 Microfluidic Control Microfluidic operation was fully computer controlled, excluding cell loading and trapping. Custom software was designed and executed in LabVIEW (National In- struments, Austin, TX). Microfluidic valves were actuated using high-speed com- puter controlled solenoid micropump manifolds (Fluidigm Inc., San Francisco, CA) via a PCI-6533 digital input/output (DIO) card (National Instruments). A single LabVIEW program operated all experiment types, with user designed ex- 126 periments inputted as parsed text files. Scheduling algorithms were included to maximize the frequency of experiment refresh of all 32 chemical sequences. A.1.3 Chemicals and Media Yeast cells were grown with aeration overnight in YPD (30C), diluted and grown to log-phase in SCD on the day of the experiment. BSA (20 mg/mL) was added to all SCD solutions to act as an anti-fouling agent. We found that this helped avoiding adherence of yeast cells to PDMS walls as well as reducing non-specific binding of the α-factor to the PDMS or tubing walls[2]. α-factor was purchased from ZymoResearch (Orange, CA). A.1.4 Cell Preparation Cells were diluted from overnight culture into fresh SCD and grown to log-phase. To improve uniform cell loading, the cell solution was sonicated at low power causing the dissociation of cell clumps. To achieve ideal seeding density, cells were concentrated to an OD600 of 3 immediately before loading. This allowed for a seeding density of 25 cells per microchamber, and with 8 microchambers per experiment, 20 - 30 initial cells per experiment. Once loaded onto the microfluidic device, the cells were perfused with fresh SCD for at least 2 h prior to initial α- factor stimulation. Image acquisition was started at least one hour before α-factor stimulation in order to record basal fluorescence level as well as initial cell number. Cells were grown and stimulated at room temperature. A.1.5 Chemical Mixing and Perfusion Between every exchange or refreshing of medium, the following refresh proto- col was used: 1) Prime and wash multiplexer: Perfuse adjacent waste row with 127 chemical solution for 20s. 2) Refresh microchambers: Perfuse experiment row with chemical solution for 70s. 3) Wait: A wait of 3s is used to dissipate pressure build up that occurs across the high-impedance microchamber traps. This protocol allowed us to fully refresh an experimental row approximately every 100s. A.1.6 Constant Stimulation Protocol Yeast strains were continuously stimulated with 32 exponentially distributed α- factor concentrations ranging from 0 - 100 nM beginning at t = 0 s. α-factor con- centrations were calculated as: α-factor concentration = 1.16i nM; where i = row number. The calculated concentrations were then rounded to account for discrete ratio mixing. The first row (row index 0) was used as a negative control (α-factor = 0 nM). All 32 α-factor concentrations were created on chip using ratio mixing enabled by the peristaltic pump. Mixing protocols are found in Table A.1. Each chemical mixture protocol was based on a 10 pump cycle period. To continually administer a particular mixture, the 10 pump cycle was repeated. 128 i 1.16i Used (nM) 0 nM 1 nM 10 nM 100 nM 0 1 0 10 0 0 0 1 1.16 1 0 10 0 0 2 1.35 1.3 6 3 1 0 3 1.56 1.5 4 5 1 0 4 1.81 1.8 1 8 1 0 5 2.10 2 8 0 2 0 6 2.43 2.4 4 4 2 0 7 2.82 2.8 0 8 2 0 8 3.27 3.2 5 2 3 0 9 3.80 4 6 0 4 0 10 4.41 4.4 2 4 4 0 11 5.12 5 5 0 5 0 12 5.94 6 4 0 6 0 13 6.89 7 3 0 7 0 14 7.99 8 2 0 8 0 15 9.27 9 1 0 9 0 16 10.75 10 0 0 10 0 17 12.46 12.5 2 5 2 1 18 14.46 14.5 0 5 4 1 19 16.78 17 2 0 7 1 20 19.46 20 8 0 0 2 21 22.57 22.5 1 5 2 2 22 26.19 26 2 0 6 2 23 30.38 30 7 0 0 3 24 35.24 35 2 0 5 3 25 40.87 41 5 0 1 4 26 47.41 46 0 0 6 4 27 55.00 50 5 0 0 5 28 63.80 64 0 0 4 6 29 74.01 73 0 0 3 7 30 85.85 90 1 0 0 9 31 99.60 100 0 0 0 10 Table A.1: Pumping protocol for creating 32 different α-factor concentra- tions. The first column is the row index. The second column is the exponentially calculated α-factor concentration. The third column is the actual α-factor concentration used. The final four columns are in units of number of pumps for a 10 pump cycle. 129 Mixing of pumped solutions occurred through mechanisms of Taylor dispersion[8]. We tested the mixing in our chip by creating a concentration gradient of fluorescent dye (fluorescein). Effective concentration was measured by taking a fluorescent in- tensity measurement of the dye as it passed the entrance of an experiment row. A plot of measured fluorescence intensity vs. relative concentrations is shown in Fig. 2.2E. The results show good agreement, indicating our solutions are well mixed, and our concentration profiles are precise. Figure A.1: Sieve Valves. (A) Standard MSL valves deflect the elastomer membrane into a rounded flow channel, causing complete closure. (B) Sieve valves deflect the elastomer membrane into a rectangular flow channel, causing incomplete closure. 130 A.1.7 Single Transient Pulse Protocol Scheduling of the single pulse experiment is shown in Fig. A.2A Protocols of the same concentration were grouped together to allow for the simultaneous refreshing of multiple experiment rows. Concentration groups were staggered by 2.5min. Approximately 16 p.s.i. (110.3 kPa) was applied to the different chemical lines. A.1.8 Short Repeated Pulses Protocol Scheduling of the repeated pulse experiment is shown in Fig. A.2B. Concentration groups were staggered by 2.5min. Approximately 16 p.s.i. (110.3 kPa) was applied to the different chemical lines. 131 Fi gu re A .2 :S tim ul at io n pr ot oc ol s. (A ) si ng le pu ls e ex pe ri m en ts . C ol or in di ca te s th e ad m in is te re d α -f ac to r co nc en - tr at io n. T he ro w s in di ca te d th e ex pe ri m en ta lr ow s in th e m ic ro flu id ic m at ri x. H or iz on ta la xi s is el ap se d tim e. (B ) R ep ea te d pu ls e ex pe ri m en ts . C ol or in di ca te s th e ad m in is te re d α -f ac to r co nc en tr at io n. T he ro w s in di ca te d th e ex pe ri m en ta lr ow s in th e m ic ro flu id ic m at ri x. H or iz on ta la xi s is el ap se d tim e. 132 A.1.9 Biological Constructs The full list of strains is given in Table A.2. Deletion mutants were obtained from the Open Biosystems Yeast Knockout (YKO) collection, as was the wild type (WT) BY4714 strain. All strains are derived from the parental S288C strain and have the following base genotype: MATa his3∆1 leu2∆0 met15∆0 ura3∆0. Deletion mu- tants in the YKO collection were derived using a polymerase chain reaction (PCR) based strategy to replace the open reading frame with a KanMX deletion cassette. We confirmed each deletion using PCR methods. To report on mating-pathway- dependent gene expression we transformed the strains with the gene coding for enhanced green fluorescent protein (GFP) under control of a mating-specific pro- moter. The promoter consisted of 3 consecutive pheromone response elements (PREs) from the PRM1 promoter, plus 3 flanking PRM1 promoter bases on each side, plus a CYC1 core promoter placed just prior to the ATG start site. The PRE- promoter-GFP construct (PRE-GFP) was placed adjacent to the HIS3 selection ele- ment using the Longtine et al. cassettes[5] and integrated at the his3∆1 locus using flanking sequences introduced by long-oligo PCR. The integration was verified by PCR. In each strain the pheromone protease gene, BAR1, was deleted to preclude complications resulting from genotype-dependent Bar1 activity. The BAR1 cod- ing sequence was deleted using PCR methods with the pFA6a-hphNTI hygromycin cassette as described in Janke et al [4]. A.1.10 Image Analysis Pipeline Algorithms Each set of microfluidic cell traps was acquired in two fields of view, giving a total of 256 experiment traps x 2 = 512 differential interference contrast / fluorescent image pairs acquired per time point. We acquired a full set of images every 15min, 133 which was slightly longer than the time it took to iterate over and acquire the 512 image pairs. Over a 12.5 h experiment this resulted in over 50,000 images ( 120 Gb of image data). To process data a customized image analysis pipeline was developed using MATLAB software (Mathworks, Natick, MA). Central to our image analysis pipeline were algorithms for automated cell seg- mentation and enhanced green fluorescence protein (GFP) concentration calcula- tion. These algorithms consisted of the three major steps described below. Identification of Microfluidic Region of Interest Prior to cell segmentation the images were cropped to exclude the non-yeast con- taining regions outside of the cell flow channels. This reduced computational time as well as false positives due to out-of-channel segmentation. To detect the channel boundaries, the detection algorithms took advantage of 1) the horizontal orienta- tion of the yeast flow channels and 2) the dark illumination of channel edges. The horizontal projection of the image, calculated by summing each image column, re- sulted in a one-dimensional signal in which four sharp local minima represented the four channel edges. The projected values were then normalized to the range [0, 1] and the local minima were detected by thresholding. The edge detection algorithm was an iterative procedure that used the a priori knowledge that there should be two channels whose approximate width and sepa- ration from each other (in pixels) was known. The threshold was at first set to a user-defined initial value. If thresholding found two local minima whose distance from each other was less than 50 pixels, only the one whose value was smaller was kept. If this procedure detected four local minima, these local minima were kept as the channel edges. If fewer than four local minima were found, the procedure was 134 repeated with the threshold incremented by 0.01 (a.u.); and if more than four local minima were found, the procedure was repeated with the threshold decremented by 0.01. If the procedure failed to converge to four local minima, channel detection failed, and cell segmentation was performed on the whole image. Fig. A.3B shows the detected channel edges overlaid in red on the DIC image. 135 Id en tifi er G en ot yp e W T M AT a le u2 ∆0 m et 15 ∆0 ur a3 ∆0 ba r1 ∆: :H ph N T 2 hi s3 ::P R E -G FP -H IS 3 be m 3∆ M AT a le u2 ∆0 m et 15 ∆0 ur a3 ∆0 ba r1 ∆: :H ph N T 2 hi s3 ::P R E -G FP -H IS 3 be m 3∆ ::K an M X 4 di g2 ∆ M AT a le u2 ∆0 m et 15 ∆0 ur a3 ∆0 ba r1 ∆: :H ph N T 2 hi s3 ::P R E -G FP -H IS 3 di g2 ∆: :K an M X 4 fa r1 ∆ M AT a le u2 ∆0 m et 15 ∆0 ur a3 ∆0 ba r1 ∆: :H ph N T 2 hi s3 ::P R E -G FP -H IS 3 fa r1 ∆: :K an M X 4 fu s3 ∆ M AT a le u2 ∆0 m et 15 ∆0 ur a3 ∆0 ba r1 ∆: :H ph N T 2 hi s3 ::P R E -G FP -H IS 3 fu s3 ∆: :K an M X 4 ks s1 ∆ M AT a le u2 ∆0 m et 15 ∆0 ur a3 ∆0 ba r1 ∆: :H ph N T 2 hi s3 ::P R E -G FP -H IS 3 ks s1 ∆: :K an M X 4 m sg 5∆ M AT a le u2 ∆0 m et 15 ∆0 ur a3 ∆0 ba r1 ∆: :H ph N T 2 hi s3 ::P R E -G FP -H IS 3 m sg 5∆ ::K an M X 4 pt p2 ∆ M AT a le u2 ∆0 m et 15 ∆0 ur a3 ∆0 ba r1 ∆: :H ph N T 2 hi s3 ::P R E -G FP -H IS 3 pt p2 ∆: :K an M X 4 rg a1 ∆ M AT a le u2 ∆0 m et 15 ∆0 ur a3 ∆0 ba r1 ∆: :H ph N T 2 hi s3 ::P R E -G FP -H IS 3 rg a1 ∆: :K an M X 4 rg a2 ∆ M AT a le u2 ∆0 m et 15 ∆0 ur a3 ∆0 ba r1 ∆: :H ph N T 2 hi s3 ::P R E -G FP -H IS 3 rg a2 ∆: :K an M X 4 sl t2 ∆ M AT a le u2 ∆0 m et 15 ∆0 ur a3 ∆0 ba r1 ∆: :H ph N T 2 hi s3 ::P R E -G FP -H IS 3 sl t2 ∆: :K an M X 4 st e5 0∆ M AT a le u2 ∆0 m et 15 ∆0 ur a3 ∆0 ba r1 ∆: :H ph N T 2 hi s3 ::P R E -G FP -H IS 3 st e5 0∆ ::K an M X 4 Ta bl e A .2 :S tr ai ns us ed in th is st ud y. 136 137 Figure A.3: Image Segmentation Algorithms. (A) Original cell image. (B) Microfluidic channel detection. (C) Enhanced bright-field image. (D) Yeast cell wall segmentation. (E) Cell wall segmentation after removal of channel edges. (F) Cell area mask. (G) Cell segmentation before post-processing steps. (H) Final cell segmentation. Cell Segmentation Following channel detection, the DIC image was enhanced using background sub- traction. The background was estimated by spatially averaging the image using a 21 × 21 mean filter. The resultant image is shown in Fig. A.3C. The first step in cell segmentation identified the cell walls. Yeast cell walls were clearly visible as continuous borders that were darker than the background, giving two useful properties: the local mean was low and the local variance was high. Cell wall pixels were marked as those pixels whose local mean was below a threshold Tm and whose variance was above a threshold Tv[6]. A 5 × 5 neighborhood was used to calculate the local mean and variance. The threshold Tm was set to µm - (1/2) σm, where the µm and σm were the global mean and standard deviation of the local mean image respectively. Similarly, the threshold Tv was set to µv + (1/3) σv, where the µv and σv were the global mean and standard deviation of the local variance image respectively. The cell wall segmentation result is shown in Fig. A.3D. This segmentation result has both false negatives and false positives (apparent discontinuities in cell walls and non-cell wall pixels detected as cell wall pixels respectively), and further processing was required. First, small holes inside the de- tected cell walls were removed by a morphological closing with a 5× 5 structuring 138 element. Second, detected channel edges were removed by assigning to zero each row of pixels five pixels above and below the detected channel edges followed by hysteresis thresholding[1]. The result is shown in Fig. A.3E. Typically some parts of the channel edges were still falsely detected as cell wall pixels, but these pixels were removed in the subsequent processing steps. The image with the detected cell walls was used to obtain a mask of cell ar- eas. First, the image was dilated by a circular structuring element inside a 11 × 11 square. Following, holes inside the objects were filled and the image was sub- sequently eroded by the same circular structuring element. The result is shown in Fig. A.3F. The detected cell walls were then removed from the mask (Fig. A.3F). Due to noise in cell wall detection, some cells were incorrectly grouped together and recognized as a single cell. These were separated from each other with the watershed of the Euclidean distance function of the complement image[7]. The h-maxima transformation was used in order to prevent over-segmentation. Finally, objects that were smaller than 80 pixels in size were excluded. The final cell seg- mentation result is shown in Fig. A.3H. Measurement of GFP Concentration GFP concentration values were calculated from the fluorescent channel images. Average background florescence, as calculated by the mean fluorescence of values outside of the segmented cells, was first subtracted from each pixel value. Total fluorescence of each cell was then summed from the image pixel values bounded by the segmented regions identified from the segmentation algorithms. For each cell, the GFP concentration was obtained by dividing the total cellular fluorescence intensity by a volume estimate for the cell. The volume estimate was based on the 139 cross-sectional cell area that was obtained from the cell segmentation result and calculated using the conical method as described previously[3]. A.2 Supporting Text A.2.1 Experimental Variability of Microfluidic Platform Experimental variability is due to condition differences between experiments within a single device and between experiments taken on different devices on different days. Sources of in-chip variability include precision limits of chemical mixing, consistency of media conditions across experimental positions, and variations in z-position. To maximize chemical mixture precision we used on-chip pumping and mixing of α-factor stock solutions to create precise final stimulant concen- trations (demonstrated in Fig. 2.3E and discussed in paper text). To ensure re- producibility across experimental positions (different columns of the microfluidic device) we used FITC and food dye tracers to determine the perfusion time needed to reliably replace the well-mixed media solutions across a row of microchambers (30 s). During experimentation we used a much longer refresh time (70s) to ro- bustly exchange the media. We empirically characterized in-chip reproducibility by stimulating eight identical yeast genotypes (WT) with a series of constant α- factor concentrations (Fig. A.6). We find variability is minimal (<10 - 20%) and not dependent of column position. To reduce variability due to image focus, we manually defined focus positions for all images prior to image acquisition and pro- grammed our microscope to return to these positions at each time-point. We found this method was robust and long time courses could be obtained with minimal fo- cus drift (Fig. A.8). In addition, we used a 40x long distance objective (NA=0.6) 140 with a depth of field (estimated 1.5 µM) similar to the diameter of yeast cells. This increased the robustness of the fluorescent measurement, which was advantageous over a higher magnification and numerical aperture objectives. Variability across experiments taken with different chips on different days is primarily due to precision limitations of stock α-factor solutions and day-to-day fluctuations in fluorescent excitation intensity. Quantitative comparisons between genotypes were possible by internally controlling all measurement by normalizing response to a WT control. In each experiment we reserved one column per-chip (chosen randomly) for WT and mutant responses were analyzed after normaliza- tion to WT. Lastly, we found that variability in population averaged statistics (e.g. mean response) increased when initial seeding density was low, due to sampling error of the biological response. An initial density of 20-30 cells (2.5 - 3.75 cells per micro chamber) greatly reduced this source of error and we conducted our experiments in this regime. Increases in chamber size will further address issue, and technical modifications to the trapping scheme are in development (Appendix B). A.2.2 Morphology Classifications Under Constant Stimulation Throughout our experiments we found α-factor concentration-dependent morpho- logical responses. This was most apparent when the cells were subjected to con- tinuous stimulation for long times under chemostatic conditions. In our study, we classified the morphologies after 6 hours of constant α-factor stimulation using three general morphology types: 1. Budding Yeast. Yeast cells maintained their vegetative rounded shape seen for exponentially growing cells. At the lowest concentrations these cells did 141 not cell cycle arrest (<5nM). 2. Elongated. Elongated cells demonstrated rod shaped phenotypes. The de- gree of elongation varied and at long time scales (>14 h) we observed cells with major axes > 5× minor axis. Elongated morphologies were observed at intermediate α-factor concentrations (4nM - 20nM). 3. Shmooing. Shmooing cells were rounded with small sharp protrusions and cell cycle arrest. These cells were mostly observed at higher α-factor con- centrations (>20nM). Fig. A.4 demonstrates morphology analysis of each strain across all α-factor concentrations taken at 6 h after initial stimulation. The concentration range for each morphology varied drastically among some mutants. Some mutants like far1∆ only showed the budding yeast morphology and lacked any other type. Others like msg5∆ and ptp2∆ displayed an extended shmoo concentration range (extending to below 4nM) as compared to WT. 142 143 Figure A.4: Morphological response of the yeast strains to α-factor con- centration. Color dots represent morphologically stratified population mean GFP response, with dot opacity indicating the percentage of cells with that morphology for the specific experiment. Red = yeast form; blue = hyperelongated; green = shmoo. Error bars represent standard deviation of response. Measurements were taken from the t = 360 min time point for all strains except fus3∆ which was taken at t = 600 min. Mating morphologies in fus3∆ we not observable until this time. A.2.3 Single Pulse Analysis To investigate the pulse-width-dependence of the rate of pathway deactivation, we quantitatively examined the kinetics of pathway shut down following α-factor re- lease in multiple ways. First, we measured the delay between α-factor release and time to reach maximal GFP concentration across all pulse-widths (Fig. A.5A). We find that this time was independent of pulse-width, occurring 30min after release from α-factor stimulation for all conditions. Second, we measured GFP decay rates upon α-factor release by fitting the post-maximum GFP time-course data to a model of exponential decay. Across replicates (n=3) and across all conditions giv- ing significant GFP expression (all combinations of 50 and 20 nM x 60, 90, 120, 150, 180, and 210 min), we found a condition-independent decay rate of 0.0046 ± 0.0014 min−1. Third, we compared the integrated GFP response to total integrated input dose. As we know that the activation rate for a given α-factor concentration is constant, as is the time to initiate deactivation upon release, deactivation rates that are independent of pulse-width should map linearly to input dosage. We in fact observed this relationship (Fig. A.5B). Taken together, these results indicate 144 that, for the pulse-widths tested, we do not measure network adaptation effects resulting in increased or decreased rate of pathway shutdown. 145 146 147 Figure A.5: Single Pulse Analysis. (A) Time to reach maximal GFP con- centration following α-factor release. The green bar indicates α-factor stimulation. The orange line indicates approximate time of GFP con- centration maximum. A higher image acquisition rate of every 7.5min was used to increase the accuracy of peak finding. (B) Total integrated GFP out vs. total integrated α-factor input for wild-type yeast. Dot color represents different input α-factor concentration: black = 50 nM; blue = 20 nM. (C) Growth curves of WT cells stimulated with a sin- gle 50nM α-factor pulse. Pulse duration: 20 min; 40 min; 60 min; 90 min; 120 min; 150 min: 180 min; 210 min. α-factor stimulation was initiated at t = 0. Yeast cell proliferation arrests upon stimulation with α-factor and resumes when the α-factor is washed out. 148 Figure A.6: Response variability within a single microfluidic device. Each data point is the steady-state population averaged GFP concentration normalized by the basal fluorescence (y-axis) for a given α-factor con- centration (x-axis). Each microfluidic column contains the same WT genotype, with each dot color representing a different column. Red dots and lines represent the mean and standard deviation response of all columns. Cases where an experimental position contained zero cells were removed from this analysis. 149 150 151 Figure A.7: Reproducibility of results. (A) dGFP/dt measurements. Each bar plot gives the mutant initial dGFP/dt over WT initial dGFP/dt across all α-factor concentrations between 5nM and 100nM. Initial dGFP/dt is calculated as the slope of a line fitted to the population averaged GFP concentrations between 30-180min. Error bars give standard deviation across experiments (n = 2 to 5). (B) Periodic α-factor stimulation for all mutants. Heat plots and bar plots of mutant GFP concentrations vs. WT are given for all tested strains. Heat plot scales are in log2(mutant / WT) and the bar plots are in (mutant / WT). In both cases, response is taken at t=600min and averaged across n experiments (n is indicated for each strain). Error bars on the bar plot represent the standard de- viation of the measurement across experimental replicates. Number of replicates is given for each strain. If replicates from the same chip were considered, the total number of replicates from different chips is indi- cated. Each colored rectangle of the heat plots corresponds to a specific α-factor concentration (column) and delay between successive pulses (row). Experimental conditions of the bars of the bar plot are indicated by the colored base bar (dark blue = 5nM; orange = 10nM; yellow = 20nM; light blue = 50nM) and colored dot (dark blue = constant stim- ulation; yellow = 15 min delay; green = 40 min delay; light blue = 65 min delay; grey = 140 min delay). A.2.4 Pulse-Width Dependent Growth Rate Manual counts of numbers of cells across time were used to obtain exact cell growth curves. Fig. A.5C gives the growth curves for wild-type (WT) cells across 152 all single pulse widths of the 50 nM concentration. Cell-cycle arrest is observed approximately 75 min after stimulation and approximately persists for a duration equivalent to the pulse width. We were unable to detect any arrest for the <20 nM conditions. 153 154 Figure A.8: Image focus over 12 hours. Representative DIC images yeast cells taken over a 12 hr time-course. Using an automated XYZ posi- tioning, images were acquired at each position of the microfluidic de- vice (256 microchambers x 2 fields of view per microchamber = 512 positions) every 15 min for 12 hrs (52 images). Shown here are the 2 hr time-points for a single position. A.2.5 Calculation of d[GFP]/dt Initial rate of GFP molecule production (d[GFP]/dt) was calculated as the slope of the line fitted to the population averaged GFP response for t = 30 min to t = 180 min. This time interval was chosen to allow for GFP maturation and to end before response saturation. We found that yeast cells demonstrated an α-factor concentration-dependent rate of GFP expression. 155 Bibliography [1] J Canny. A computational approach to edge detection. IEEE Transactions on Pattern Analysis and Machine . . . , 8(6):679–698, Nov 1986. [2] Alejandro Colman-Lerner, Andrew Gordon, Eduard Serra, Tina Chin, Orna Resnekov, Drew Endy, C Gustavo Pesce, and Roger Brent. Regulated cell-to- cell variation in a cell-fate decision system. Nature, 437(7059):699–706, Sep 2005. [3] A Gordon, A Colman-Lerner, T Chin, and K Benjamin. Single-cell quantifi- cation of molecules and rates using open-source microscope-based cytometry. Nat Methods, 4:175–181, Jan 2007. [4] Carsten Janke, Maria M Magiera, Nicole Rathfelder, Christof Taxis, Simone Reber, Hiromi Maekawa, Alexandra Moreno-Borchart, Georg Doenges, Eti- enne Schwob, Elmar Schiebel, and Michael Knop. A versatile toolbox for pcr-based tagging of yeast genes: new fluorescent proteins, more markers and promoter substitution cassettes. Yeast, 21(11):947–62, Aug 2004. [5] M S Longtine, A McKenzie, D J Demarini, N G Shah, A Wach, A Brachat, P Philippsen, and J R Pringle. Additional modules for versatile and economical 156 pcr-based gene deletion and modification in saccharomyces cerevisiae. Yeast, 14(10):953–61, Jul 1998. [6] Antti Niemistö, Matti Nykter, Tommi Aho, Henna Jalovaara, Kalle Marjanen, Miika Ahdesmäki, Pekka Ruusuvuori, Mikko Tiainen, Marja-Leena Linne, and Olli Yli-Harja. Computational methods for estimation of cell cycle phase distributions of yeast cells. EURASIP journal on bioinformatics & systems biology, page 46150, Jan 2007. [7] P Soille. Morphological image analysis: Principles and applications. page 391, Jan 2003. [8] TM Squires and SR Quake. Microfluidics: Fluid physics at the nanoliter scale. Rev Mod Phys, 77:977–1026, Jan 2005. 157 Appendix B A Microfluidic Platform for High-Throughput Single-Cell Tracking and Dynamic Environmental Control 1 B.1 Introduction Systems biology studies based on genome wide analysis of transcript and protein expression have greatly advanced our understanding of the molecular interactions that govern complex cellular behaviours[28]. However, such studies are typically 1Figures B.1, B.2, B.3, and B.4 of this chapter are taken from Didier Falconnet, Antti Niemisto, James Taylor, Tim Galitski, Ilya Shmulevich, and Carl Hansen. A microfluidic platform for high- throughput single-cell tracking and dynamic environmental control. In Preparation. The lead author conducted the bulk of the work in the said manuscript. My role in this study was in supporting the technology development of the microfluidics and image analysis pipeline. 158 limited by poor temporal and spatial resolution. In addition, these methods only provide an average population response and are hence blind to cell differences that arise from a combination of de-synchronization[9], bistability[17, 26], and stochastic variations in expression[10, 20]. Technical advances in microscopy, fluorescent reporters and live-cell imaging setups have allowed the visualization and quantification of gene expression as well as intra-cellular trafficking in real time within single cells [13, 29]. Studies on cel- lular behavior with single-cell resolution have open new experimental alleys and pioneered work about heterogeneous response and its benefits for a population to explore simultaneously different phenotypes[2, 26], signaling network adaptation to limit the metabolic cost of sustained response[33], and network memory to allow more rapid accommodation of recurrent stimuli[1, 5]. Such studies are traditionally performed by imaging cells in multi-well plates (eg, high content cell screenings) or by confining them between a microscope slide and an agar pad[13]. However, both approaches lack the ability to finely control and modulate the chemical envi- ronment and suffer accumulation of metabolites in the culture medium. The ten- dency of the cells to grow out of the focal plane further limits the experiment du- ration. Flow cytometry is a high-throughput alternative that allows measuring the response of thousands of single cells and provides a snapshot of the cell states[2] however; it doesnt allow repeatedly interrogating individual cells and thus can- not provide time course data with a single-cell resolution nor provide sub-cellular information. Systems level studies aiming at unraveling complex signaling networks re- quire the quantitative time resolved analysis of single cells under a range of well- defined chemical conditions and genetic perturbations[32]. Microfluidic devices 159 combined with state-of-the art microscopy and image analysis have emerged in the past decade as powerful tools to study cells in a controlled environment and in a multiplexed, high-throughput and high-content manner [3, 12, 14, 18, 21, 26, 30] (Chapter 2). This technology allows integrating at a large scale micro- valves, pumps and reaction chambers with astonishing reliability and low cost. The de- vices are easily automated with commercial interfacing software and are thus now broadly adopted to conduct novel studies on bacteria, yeast or mammalian cells under chemostatic [3, 14, 18], or chemodynamic conditions [5, 12] (Chapter 2). Despite of the relative maturity of this technology existing microfluidic sys- tems for live-cell imaging still suffer from major limitations: the vast majority of microfluidic systems have been developed for studying adherent cells[8, 12, 18] and their fluidic wiring makes them unsuitable for non-adherent cells such as yeast or hematopoeitic lineages as they would be washed away by the fluid flow. Con- sidering the tremendous importance of yeast as a model organism and the prolific research on hematopoeitic stem cells there is an obvious need for a new generation of microfluidic devices enabling the study of non-adherent cells with a systems biology approach. A few devices including a commercial one have proposed ways to immobilize non-adherent cells under flow conditions. The proposed immobilization strategies consist of seeding cells on a surface and clamping them with a soft permeable membranes embedded in a single fluidic chamber [6]. The second approach[19] exploits the elasticity of the of the device material to inflate the chambers dur- ing cell loading, upon pressure release the cells are physically trapped between the ceiling and the floor of the chamber (http://www.cellasic.com). These devices have proven extremely useful for studies on a limited number of cell types and chemical 160 inputs they are however lacking the throughput needed to perform systems level studies on living cells across multiple mutants and chemical conditions. Moreover, the later approach doesnt exclude the loss of newly budded cells as these are po- tentially smaller than the trap heights. This often overlooked aspect also applies to adherent mammalian cells since they round up during division and are thus more likely to be washed away with the flow. To address the lack of current throughput and non-optimal trapping strategies a new type of device is required. Here we present a microfluidic live-cell imaging device capable of running simultaneously and in an unattended manner 128 different experiments on tens of thousands of immobilized non-adherent single yeast cells exposed to various pre-programmed chemical perfusion schemes. The cells are trapped in an agarose network within the chambers by a straight forward in situ gelling procedure which allows us to track each individual cell in time via a custom built algorithm and thus generate multidimensional time plots of gene expression, morphological properties and cellular growth throughout 8 different cell types and 16 chemical conditions. Expanding the experimental dimensions (number of strains, chemical conditions and time points) also results in generating large data sets often in the order of hundreds of gigabytes. To accompany such developed we built powerful custom algorithms to analyze the images and present the data in useful ways. We demonstrate our microfluidic technology by studying the pheromone in- duced mitogen-activated protein kinase (MAPK) signaling pathway in Saccha- romyces cerevisiae[4]. MAPK signaling is of central importance to a wide range of cellular decision making processes, responding to a staggering range of stim- uli including growth factors, cytokines, hormones, cellular adhesion, stress, and nutrient conditions[27]. Yeast is an ideal model organism to study the highly con- 161 served MAPK signaling cascade as it undergoes rapid division (a few hours) and genetic manipulations are straight forward. Yeast cells of mating type a can readily be induced by soluble α-factor pheromones via the membrane-localized G-protein- coupled receptor Ste2 (Ste3 for MATα cells). The pheromone peptide when bound to its receptor triggers a MAPK cascade activating expression of about 200 genes and culminating in cellular growth arrest and formation of a pointed extension (a shmoo) towards the pheromone gradient[4]. To report on mating-pathway activity we transformed the strains with the gene coding for enhanced green fluorescent protein (EGFP) under control of a mating-specific promoter (Supplementary meth- ods). B.2 Results B.2.1 Microfluidic Chip Overview and Operation Our microfluidic device is composed by an array of 128 chambers (Fig. B.1A, region 1) located at the intersections of 8 columns and 16 rows (Fig. B.1B). Each column (Fig. B.1A, region 2) is loaded with a different mutant (here single gene deletions) while the rows permit perfusion of 16 different chemical formulations (here different α-factor concentrations) that are mixed on chip automatically from 3 stock solutions. The innovative fluidic and valving circuitry was designed so that liquid gel suspensions of yeast cells could be flown into the columns and cham- bers of the chip at room temperature and subsequently flushed away in areas that are not part of the 128 experimental cell-chambers. The cells were then gently immobilized in the chambers by cooling down the chip and consequently forming a gel (agarose). The low density gel prevents any fluid flow from the perfusion 162 channels into the chambers and thus do not displace the cells. Diffusion conduits connect the chambers to the perfusion channels to deliver nutrients (pheromones or any molecule of interest) to the cells while simultaneously remove metabolites (Fig. B.1B). The fluidic device is equipped with 8 chemical inlet ports (Fig. B.1A, region 3, Fig. B.2A) for stock solutions prepared off-chip. They can either be perfused to the cells as prepared or they can be further mixed on-chip. Each row can be individually perfused or a combination of rows can be opened simultaneously to allow multiple row perfusions with an identical solution (Fig. B.2B). On-chip mix- ing is achieved by sequentially opening the appropriate chemical input valves for a predefined number of pump cycles. For example a 23 nM solution is prepared by running the two first pump cycle with 100 nM stock solution, then three cycles with the 10 nM valve open and finally completing with 5 cycles of 0 nM (pure media) to reach a fluidic slug of 23 nM with a total of 10 pump cycles. The pro- cess is repeated for the time required to replace the whole volume of a perfusion channel (approx. 35 s). Mixing between the small volumes of stock solutions is achieved by Taylor dispersion and solutions were fully mixed when they entered a perfusion row. To characterize the efficiency and accuracy of our mixing protocol we generated onchip 16 different fluorescein concentrations from 3 stock solutions (1x, 10x, 100x) and measured the fluorescence intensity in each chamber. Figure B.2C is a plot of the fluorescence intensity in function of the relative concentra- tions. Y-error bars are standard deviations across all eight chambers of each row. An almost perfect linear fit (R2 = 0.9996) demonstrates the accuracy of the on-chip formulator. Besides being fast and reproducible on chip mixing alleviates human error. Nutrients as well as any molecules used to induce a response in the cells 163 will diffuse from the perfusion channels to the cells. Figure B.2D demonstrates the kinetics of fluorescein diffusion under continuous flow. The chambers reach equilibrium with the feeding channel within 9 min (for non-continuous flow see supplementary notes). Microfluidics offer unique capabilities for precise handling of fluids including switching between different chemicals as illustrated by the arbi- trary sequence in Figure B.2D alternating between fluorescein solution and media. These switching speeds were adequate for the biological experiments conducted in this work. If faster solution exchange is needed, faster switching can easily be achieved by simply reducing the chamber width or widening or shortening the diffusion channels connecting the chambers to the perfusion channels. 164 Figure B.1: Microfluidic chip design and operation.. (a) The central part of the chip (1) is composed of an array of 128 chambers (8 columns × 16 rows). Each column (2) is loaded with a different strain at the desired density. The rows are fed by 8 chemical inlets (3) which controlled by independent valves. Single or multiple rows can be perfused simulta- neously by combined actuation of the multiplexer valves (5). Replaced fluids are collected into a 1 ml bottle connected to the chip by outlet ports (4). Fluids are moved into the channels by applying positive pres- sure to the chemical inlets or by using the on-chip integrated peristaltic pump (6). (b) Detailed chip architecture; each chamber (light blue) is connected to the perfusion channels (dark blue) by diffusion channels (dark blue). All diffusion channels can be closed by actuating the dif- fusion valves (yellow). Each of the 16 rows has its own waste channel (dark blue above each chamber) in order to flush out the binary tree between placed between the cell array (1) and the chemical inlets (3). The chambers are partitioned during the experiment by the side-valves (pink). The loading valves (red) are actuated when the cells are loaded into the chip in order to avoid cross-contamination between strains. (c) A chamber with cells trapped in the agarose gel and growing in media at a density 3.5×109 cells/ml. Note that each chamber is imaged with 2 fields of views. The chambers are 684×260×4.4 µm3 filling a vol- ume of 0.7 nl. The 6 squares are posts providing mechanical stability. (d) Frequent media refreshing allows the culturing cells in chemostatic conditions and thus achieving high densities as shown by the growth curve. 165 B.2.2 High-Throughput Multimode Live-Cell Imaging Each chamber can be repeatedly imaged with a temporal resolution as low as 10 min while the perfusion protocols run fully unattended for multiple days. The tem- poral resolution is limited by two factors; the speed of the automated stage moving to each chamber and the number of chambers to image. The chamber dimensions are 684×260×4.4 µm3 filling a volume of 0.7 nl only. Each chamber can host over 6,000 cells corresponding to 8.5 109 cells/ml. For comparison a yeast culture reaches stationary phase at about 2-3×108 cells/ml. Exceptionally high cell densi- ties can be achieved in the chip by frequently refreshing the entire volume of the perfusion channels and thus providing nutrients and removing metabolites. Note that often such high cell numbers are not necessary and may result in unnecessar- ily image analysis time. The height of the chambers (4.4 µm) are similar to the cells diameter and thus confine them to a single focal plane allowing for accurate quantitative measurements on each cell repeatedly. 166 Figure B.2: On-chip reagents mixing and temporal control of chemical environment. (a) 8 chemical inlet ports shown with food dyes. The 3 valves on the right constitute the on-chip peristaltic pump. (b) Vi- sual illustration using different food dyes perfusing separate rows. (c) On-chip formulation of 16 different concentrations of fluorescein from 3 stock solution of fluorescein (10×, 1×, 0×). The intensity was recorded in each row and plotted in function of relative concentrations. The error bars are standard deviations across the 8 chambers on each row. (d) Less than 9 min are sufficient to reach concentration equilib- rium within the experimental chambers. Temporal modulation of the chemical environment is demonstrated by arbitrary pulses alternating between media and fluorescein. 167 We monitored the response of over 60,000 individual yeast cells from 8 mutants exposed to 16 different α-factor concentrations (Supplementary Table 2) using a 20 min sampling period. During a typical 24h experiment over 40,000 images (bright field and fluorescence) are recorded holding 4 million cell measurements. We built a custom image analysis pipeline in Matlab (Mathworks, Inc) to rapidly process the images, the algorithms performed the following tasks: (i) segment and label the cells in each image, (ii) calculate total fluorescence in each cell, (iii) calculate statistics across all cells and experimental conditions and, (iv) present time-course and steady-state data in human-readable formats like figures, movies, and tables. Figure 3a is a typical time course plot of the GFP concentration (GFP normalized to the cell volume) within each wild-type cell. The cells were initially perfused with medium and at time 0 the media was switched to a 10 nM α-factor solu- tion. After approximately 300 min of pheromone induction the population reached steady-state and displayed a 12 fold increase in GFP expression compared to the basal fluorescent level. We observed a spread of response across the population which highlights the heterogeneity of cellular behavior also commonly referred to as noise[9, 22]. Coefficients of variability (CV = standard deviation/mean) of 0.3 were typically measured for such populations. Movies were automatically gener- ated for each chamber to allow rapid visualization of the evolution of cellular re- sponse under each tested condition. This allows to efficiently screening the whole chip for interesting or unexpected behaviors and can thus help focusing the study on more targeted experiments. Figure B.3 (wild-type) represents the mean wild-type response for the 16 pheromone concentrations over approximately 600 min (this corresponds to a complete column of the chip). Cells show a subtle response at α-factor concentrations as low as 1nM 168 with response saturating in a range of 22-30 nM. Below saturation, the response is monotonically graded with pheromone concentration. This device allows study- ing simultaneously the response of 7 mutants in addition to wild-type. Figure B.3 also displays the response of each of these mutants; in brief we observe that ptp2∆, msg5∆ and kss1∆ are hyper-sensitive (in increasing order) while ste50∆, far1∆, slt2∆ and fus3∆ are hypo-sensitive (in decreasing order) with respect to wild-type. These observations are supported by our previous high-throughput study (Chap- ter 2) and by others [7, 9, 26] and thus validate our platform. In addition, different response saturation thresholds are found amongst mutants. For example, slt2∆ sat- urates at α-factor concentrations as low as 10 nM while ste50∆ does not reach steady state within the time of the experiment. We are currently tackling the re- sponse of the different mutant with a high-degree of detail and will thus not be discussed further as it is outside of the scope of this method article. 169 Fi gu re B .3 :H ig h- th ro ug hp ut ge ne ex pr es si on m ea su re m en ts . (a ) 3D pl ot s sh ow in g th e ch am be r- av er ag ed G FP co nc en tr at io ns in fu nc tio n of tim e an d α -f ac to rc on ce nt ra tio n. (b )G FP co nc en tr at io n of w ild -t yp e ce lls ex po se d to 10 nM α -f ac to rm ea su re d ev er y 20 m in ov er 9 h. E ac h da ta po in ti s a ce ll w ith in th is sp ec ifi c ch am be r. 170 B.2.3 Single-Cell Tracking Reveals Heterogeneous Decision Making in a Narrow Pheromone Concentration Range Yeast cells are known to adopt various morphologies depending on their surround- ing pheromone concentration [11, 16] (Chapter 2). We observed that between 0-3 nM wild-type cells displayed a budding yeast morphology while between 4 and 8 nM an increasing fraction of the population showed an elongated morphology. At 8 nM the totality of the cells transitioned from budding yeast to an elongated phenotype accompanied by growth arrest. When the cells were exposed with 30 nM or higher they adopted the more common shmooing phenotype characterized by one or more short and pointed projections. Fig. B.4 is a representative example of wild-type cells exposed to 5 nM α-factor. Fig. B.4B is close up with a time sequence of 2 small, a priori, isogenic populations exposed to the same chemical environment (5 nM α-factor). Interestingly, the population circled in red undergoes typical growth arrest and cellular volume increase while the green-circled popula- tion keeps dividing and is apparently insensitive to the presence of pheromones. Furthermore, the GFP intensity reporting on the pheromone pathway activity pro- vides complementary information to the bright field images; we visually observe a direct correlation between growth arrest accompanied by elongation with higher GFP expression. To accurately quantify the GFP expression in each cell and over time we built a cell tracking algorithm which labels each cell (Fig. B.4C) and al- lows plotting their respective time-lapse gene expression profile as shown in Fig. B.4D. Cells undergoing growth arrest (red) display a strong GFP signal while di- viding cells (green) only show a moderate GFP response. The strongest responder displays a 12 fold increase over basal fluorescence level while the strongest re- sponding cells in the green population only have a 5 fold increase. Due to the 171 relatively dim signal in the weakly responding cells and their lack of growth arrest simple visual observation could lead to erroneously concluding that these cells ig- nore the pheromone stimulus. By quantifying accurately their GFP concentration we show that their pheromone pathway becomes activated but to a milder level. This suggests that the activity of the signaling transduction cascade is graded but that there might be a threshold above which the cells commit to growth arrest and morphological transition. Our tracking algorithm keeps track of each cell and attributes new labels to newborn cells however, it is unable to track the lineages (relate the daughter cell to its mother). By analyzing time courses we manually reconstructed the lineages of the 2 populations as shown in Fig. B.4E. Specific markers such as budneck stains for automating lineage tracking has recently been demonstrate successfully[6]. What causes one population significantly less sensitive to pheromones than its neighbor remains an open question which will drive further studies. Particularly interesting would be to decipher the potential lineage-dependent behavior; in other words if the decision to growth arrest and elongate is coordinated within the mem- ber of the populations; in the present example, 3 out of 3 cells in the red circle undergo growth arrest and elongation, while the cells in the green population keep budding. The cells within these small populations are close relatives (separated by 1 to 3 generations at most) and maybe the cause of the concerted response. This points to the question of potential epigenetic (non-genetic) inheritance of compo- nents playing a role in the pheromone response. 172 173 Figure B.4: Lineage tracking with single-cell resolution; identification of a switch-like response. (A) A snapshot of an isogenic colony of bar1∆ cells exposed to 5nM α-factor pheromones after 40min. (B) A time- lapse sequence to illustrate the variability of pheromone response of genetically identical cells. The red population undergoes growth ar- rest and displays an elongated phenotype while the green keeps bud- ding over multiple generations, seemingly ignoring the pheromone in- put. (C) The tracking algorithm attributes a unique label to each cell, allowing for correlation of lineage, morphology, budding pattern and gene expression. (D) EGFP concentration in each cells reporting the pheromone pathway activity. Note that plotted gene expression is nor- malized for calculated cell volume and therefore represents EGFP con- centration. B.3 Discussion We designed and optimized a microfluidic chip that combines multiple impor- tant features: non-adherent single cells can be measured repeatedly under a pre- programmed perfusion protocol sequences with on-chip chemical input switching and mixing. The device was designed as a tool for systems biology studies where multiple cell types (strains) were exposed to a range of stimuli over multiple gen- erations and in an unattended manner. We developed custom software to run the chips in an automated manner as well as sophisticated image analysis algorithms to efficiently and accurately extract relevant information from million single cell data. In this study we report multiple observations that raise questions such as the source 174 for the heterogeneous response in an isogenic population or the potential epigenetic inheritance of components implicated in the pheromone response. We argue that our device combined with our software is an ideal tool to tackle such fundamen- tal and current questions with an efficient and precise approach. The accurate and rapid fluid handling capability combined with the minute volumes required in mi- crofluidics have imposed this technology as a standard and cost-effective method in the areas of analytical biology, drug discovery and diagnostics [21, 23]. It is likely that in future the same will be seen for cellular in vitro assays such as the one presented in this work. The spreading of this technology to biology labs has been refrained by the need for special equipment and know-how however, access to custom microfluidic chips has recently become widely available: recognizing the impact of microfluidics, several institutions have created fabrication facilities that can provide any researcher with the desired chips. The devices are fully cus- tomizable and easily modified as the channels are initially drawn on CAD software. This provides virtually unlimited flexibility for adapting the chips to specific ap- plications and for studying other cells types such as bacteria or mammalian cells. Particularly appealing is the rapid turn over time between designing a new chip and performing the first experiment. This typically required less than 4 days while the chip fabrication step (with molds in hand) was routinely performed in 1 day and produced 8 functional chips allowing to perform 1024 independent experiments. The footprint of our chip is approximately 4×3 cm2 and can thus be used on any regular microscope equipped with a motorized stage. Sources of compressed air required to run a chip are widely available either from in house lines or through mobile gas cylinders. The hardware for controlling the valves is commercially available (see Methods). CAD files and software can be obtained upon request. 175 This work presents the-state-of-the-art microfluidic sophistication for studying non-adherent yeast cells under well control chemical conditions and at a through- put adequate for systems level studies with single cell resolution and track capa- bility. Further, our design is scalable and the number of genotypes measured can be greatly increased to enable the possibility of genome-wide studies using a mod- est number of devices. Finally, we believe that such large scale data sets should provide stringent tests for in silico models of signaling networks and lead to more accurate and ultimately predictive understanding of complex cellular decision mak- ing. B.4 Methods B.4.1 Chip Fabrication Fabrication of the microfluidic device was accomplished using standard multi-layer soft lithography techniques[15, 21, 31]. We used a 2-layer design: The top layer is a ’control layer’, containing channels used for pneumatic valving and the bottom layer was a ’flow layer’, containing the cells and the chemical channels. The chip was covalently bonded to the 0.7 mm low autofluorescent glass slides (borofloat, S.I. Howard Glass Co., Inc. MA) by a 10 s O2-plasma surface treatment. The assembled and punched poly-dimethylsiloxane chips (PDMS, RTV615 manufac- tured by General Electric, CT) were baked for at least 15h at 80C. Chip design was completed using AutoCAD software (Autodesk, Inc., San Rafael, CA). Mas- ter negative molds were fabricated by standard photolithography techniques on 4 inch (101.6 mm) silicon wafers (Silicon Quest International, Santa Clara, CA). High resolution transparency masks (20,000 dpi) were printed by CAD/Art Ser- 176 vices. The flow layer consisted of two feature types: 4.4 µm high rectangular cell microchambers, and 9 µm high rounded flow channels. The rounded channels cross-section was obtained with by placing the wafer on a 130C for 30 min. Each cell microchamber had a volume of 0.71 nL with dimensions 684x260x4.4 µm3. The 4.4 µm layer was made with SU8-5 negative photoresist (Microchem Corp., Newton, MA) and the 9 µm rounded layer was made with SPR220-7 positive pho- toresist (Microchem Corp.). The control master was a single layer mold consisting of 25 µm high squared features made with SU8-2025 negative photoresist (Mi- crochem Corp.). Resist processing was performed according to the manufacturer’s specifications. B.4.2 Segmentation and Tracking Algorithm Cell segmentation was exclusively performed on bright-field images in order to be independent of any fluorescent reporter for this task. Yeast cell walls were clearly visible as continuous borders that were darker than the background. The local mean and variance were calculated for each pixel of the image using a small local neighborhood, and those pixels for which the local mean was below a threshold and the local variance was above a threshold were marked as cell wall pixels[24]. Then, a mask of cell areas was obtained by thresholding the local variance image with a threshold found with the help of Otsu’s method[25], followed by a series of operations based on mathematical morphology. Then, the initially recognized cell walls were removed from the mask. In the resulting image some cells were incor- rectly grouped together, and those were separated from each other by a watershed method. The cell tracking problem was posed in terms of combinatorial optimiza- tion as the classic assignment problem. Cells from the image frame at time point t 177 acted as the agents, cells from the image frame at time point t- acted as tasks, and the cost of assigning a task to an agent was equal to the Euclidean distance between the respective cells from time points t and t-1. If the number of cells at each time point was the same, the assignment problem would be correctly formulated, and the solution with the smallest total cost (sum of distances between cells) could be readily found. 178 Bibliography [1] Murat Acar, Attila Becskei, and Alexander van Oudenaarden. Enhance- ment of cellular memory by reducing stochastic transitions. Nature, 435(7039):228–32, May 2005. [2] Murat Acar, Jerome T Mettetal, and Alexander van Oudenaarden. Stochas- tic switching as a survival strategy in fluctuating environments. Nat Genet, 40(4):471–5, Apr 2008. [3] Frederick K Balagaddé, Lingchong You, Carl L Hansen, Frances H Arnold, and Stephen R Quake. Long-term monitoring of bacteria undergoing pro- grammed population control in a microchemostat. Science, 309(5731):137– 40, Jul 2005. [4] Lee Bardwell. A walk-through of the yeast mating pheromone response path- way. Peptides, 25(9):1465–76, Sep 2004. [5] M Bennett, W Pang, N Ostroff, B Baumgartner, S Nayak, L Tsimring, and J Hasty. Metabolic gene regulation in a dynamically changing environment. Nature, 454(1119-1122), Aug 2008. [6] Gilles Charvin, Frederick R Cross, and Eric D Siggia. A microfluidic device 179 for temporally controlled gene expression and long-term fluorescent imaging in unperturbed dividing yeast cells. PLoS ONE, 3(1):e1468, Jan 2008. [7] Raymond E Chen and Jeremy Thorner. Function and regulation in mapk signaling pathways: lessons learned from the yeast saccharomyces cerevisiae. Biochim Biophys Acta, 1773(8):1311–40, Aug 2007. [8] R Cheong, C Wang, and A Levchenko. High-content cell screening in a microfluidic device. Mol Cell Proteomics, Oct 2008. [9] Alejandro Colman-Lerner, Andrew Gordon, Eduard Serra, Tina Chin, Orna Resnekov, Drew Endy, C Gustavo Pesce, and Roger Brent. Regulated cell-to- cell variation in a cell-fate decision system. Nature, 437(7059):699–706, Sep 2005. [10] Michael B Elowitz, Arnold J Levine, Eric D Siggia, and Peter S Swain. Stochastic gene expression in a single cell. Science, 297(5584):1183–6, Aug 2002. [11] S Erdman and M Snyder. A filamentous growth response mediated by the yeast mating pathway. Genetics, 159(3):919–28, Nov 2001. [12] Rafael Gómez-Sjöberg, Anne A Leyrat, Dana M Pirone, Christopher S Chen, and Stephen R Quake. Versatile, fully automated, microfluidic cell culture system. Anal Chem, 79(22):8557–63, Nov 2007. [13] Andrew Gordon, Alejandro Colman-Lerner, Tina E Chin, Kirsten R Ben- jamin, Richard C Yu, and Roger Brent. Single-cell quantification of 180 molecules and rates using open-source microscope-based cytometry. Nat Methods, 4(2):175–81, Feb 2007. [14] Alex Groisman, Caroline Lobo, HoJung Cho, J Kyle Campbell, Yann S Du- four, Ann M Stevens, and Andre Levchenko. A microfluidic chemostat for experiments with bacterial and yeast cells. Nat Methods, 2(9):685–9, Sep 2005. [15] Carl L Hansen, Morten O A Sommer, and Stephen R Quake. Systematic investigation of protein phase behavior with a microfluidic formulator. Proc Natl Acad Sci USA, 101(40):14431–6, Oct 2004. [16] Nan Hao, Sujata Nayak, Marcelo Behar, Ryan H Shanks, Michal J Nagiec, Beverly Errede, Jeffrey Hasty, Timothy C Elston, and Henrik G Dohlman. Regulation of cell signaling dynamics by the protein kinase-scaffold ste5. Mol Cell, 30(5):649–56, Jun 2008. [17] Benjamin B Kaufmann, Qiong Yang, Jerome T Mettetal, and Alexander van Oudenaarden. Heritable stochastic switching revealed by single-cell geneal- ogy. PLoS Biol, 5(9):e239, Sep 2007. [18] Kevin R King, Sihong Wang, Daniel Irimia, Arul Jayaraman, Mehmet Toner, and Martin L Yarmush. A high-throughput microfluidic real-time gene ex- pression living cell array. Lab on a chip, 7(1):77–85, Jan 2007. [19] Philip J Lee, Noah C Helman, Wendell A Lim, and Paul J Hung. A microflu- idic system for dynamic yeast cell imaging. Biotechniques, 44(1):91–5, Jan 2008. 181 [20] Narendra Maheshri and Erin K O’Shea. Living with noisy genes: how cells function reliably with inherent variability in gene expression. Annual review of biophysics and biomolecular structure, 36:413–34, Jan 2007. [21] Jessica Melin and Stephen R Quake. Microfluidic large-scale integration: the evolution of design rules for biological automation. Annual review of biophysics and biomolecular structure, 36:213–31, Jan 2007. [22] Jerome T Mettetal, Dale Muzzey, Juan M Pedraza, Ertugrul M Ozbudak, and Alexander van Oudenaarden. Predicting stochastic gene expression dynamics in single cells. Proc Natl Acad Sci USA, 103(19):7304–9, May 2006. [23] Sunitha Nagrath, Lecia V Sequist, Shyamala Maheswaran, Daphne W Bell, Daniel Irimia, Lindsey Ulkus, Matthew R Smith, Eunice L Kwak, Subba Digumarthy, Alona Muzikansky, Paula Ryan, Ulysses J Balis, Ronald G Tompkins, Daniel A Haber, and Mehmet Toner. Isolation of rare circu- lating tumour cells in cancer patients by microchip technology. Nature, 450(7173):1235–9, Dec 2007. [24] Antti Niemistö, Matti Nykter, Tommi Aho, Henna Jalovaara, Kalle Marjanen, Miika Ahdesmäki, Pekka Ruusuvuori, Mikko Tiainen, Marja-Leena Linne, and Olli Yli-Harja. Computational methods for estimation of cell cycle phase distributions of yeast cells. EURASIP journal on bioinformatics & systems biology, page 46150, Jan 2007. [25] N. Otsu. Threshold selection method from gray-level histograms. IEEE Transactions on Systems Man and Cybernetics, 9:62–66, 1979. 182 [26] Saurabh Paliwal, Pablo A Iglesias, Kyle Campbell, Zoe Hilioti, Alex Grois- man, and Andre Levchenko. Mapk-mediated bimodal gene expression and adaptive gradient sensing in yeast. Nature, 446(7131):46–51, Mar 2007. [27] G Pearson, F Robinson, T Beers Gibson, and B Xu. Mitogen-activated protein (map) kinase pathways: Regulation and physiological functions 1. Endocrine Reviews, 22(2):153–183, Jan 2001. [28] C J Roberts, B Nelson, M J Marton, R Stoughton, M R Meyer, H A Bennett, Y D He, H Dai, W L Walker, T R Hughes, M Tyers, C Boone, and S H Friend. Signaling and circuitry of multiple mapk pathways revealed by a matrix of global gene expression profiles. Science, 287(5454):873–80, Feb 2000. [29] Yaron Shav-Tal, Robert H Singer, and Xavier Darzacq. Imaging gene expres- sion in single living cells. Nat Rev Mol Cell Biol, 5(10):855–61, Oct 2004. [30] Todd Thorsen, Sebastian J Maerkl, and Stephen R Quake. Microfluidic large- scale integration. Science, 298(5593):580–4, Oct 2002. [31] M A Unger, H P Chou, T Thorsen, A Scherer, and S R Quake. Monolithic microfabricated valves and pumps by multilayer soft lithography. Science, 288(5463):113–6, Apr 2000. [32] Eli Zamir and Philippe I H Bastiaens. Reverse engineering intracellular bio- chemical networks. Nat Chem Biol, 4(11):643–7, Nov 2008. [33] J Zhou, M Arora, and D E Stone. The yeast pheromone-responsive g alpha protein stimulates recovery from chronic pheromone treatment by two mech- 183 anisms that are activated at distinct levels of stimulus. Cell Biochem Biophys, 30(2):193–212, Jan 1999. 184

Cite

Citation Scheme:

    

Usage Statistics

Country Views Downloads
United States 2 0
United Kingdom 1 0
Germany 1 134
China 1 0
City Views Downloads
Beijing 1 0
Unknown 1 134
Dunfermline 1 0
Mountain View 1 0
Ashburn 1 0

{[{ mDataHeader[type] }]} {[{ month[type] }]} {[{ tData[type] }]}

Share

Share to:

Comment

Related Items