Separating Growth and Production Phases Using Synthetic Quorum Sensing in Saccharomyces cerevisiae Thomas Carl Williams BSc, MSc

A thesis submitted for the degree of Doctor of Philosophy at The University of Queensland in 2014 Australian Institute for Bioengineering and Nanotechnology

Abstract The goal of the field of metabolic engineering is to manipulate metabolic networks to produce valuable metabolites such as fuels, chemicals, and pharmaceuticals. This goal can be extremely challenging because it often conflicts with the evolved ‘goals’ of genome replication and survival that are facilitated by naturally occurring metabolic networks. This conflict arises because engineered metabolic pathways consume the same ATP, redox potential, and carbon sources which are required for normal growth and survival, and produce products which are toxic and can interfere with native regulation mechanisms. This is a problem because in order to achieve the high product titers which are required for commercial success, a producing population of microbes must be able to grow to a high density – thus providing are a large number of production-units/catalysts. An obvious solution is to engineer separate growth and production phases so that a population can first grow to a high density, then ‘switch’ to a production mode. In this work, the concept of engineering separate growth and production phases was explored in the industrial microbe Saccharomyces cerevisiae (yeast). Separating growth and production phases requires that the genes which are critical for controlling flux through a production pathway be dynamically regulated in such a way that they are switched on at the appropriate time. However, in yeast there are a limited number of mechanisms for implementing dynamic regulation at the genetic level, and no way to prevent the drive towards growth during the production phase. To address these problems, the growth-arrest associated with the yeast mating phenotype has been investigated as a production phase. The metabolism underlying the pheromone response was highly active even though cells had arrested growth in the G1 cell-cycle phase. While these results are exciting at the lab scale, the addition of synthetic mating pheromone to an industrial scale reactor to separate growth from production is not feasible. The native pheromone communication system was therefore converted into a synthetic autocrine quorum sensing circuit where cells could produce and sense their own mating pheromone. The concentration of extracellular pheromone then acted as a proxy for population density. By altering the regulation and strength of the promoter which controls pheromone production, quorum sensing circuit dynamics could be fine-tuned to switch at different population densities. The activation dynamics of gene expression were also highly tuneable, ranging from completely graded autoinduction to highly switch-like gene expression. The best circuit topology was applied to dynamic regulation of gene knockouts that were predicted to dramatically increase PHBA yield, but significantly decrease biomass formation at the same time. Several of these genes were essential, making them perfect candidates to test the novel quorum sensing system. By coupling an RNA interference module to the quorum sensing circuit, two essential genes (CDC19, and ARO7) could be strongly repressed (~30 fold) after a growth phase. Along with the expression of relevant production pathway genes, this strategy resulted in a 37 fold improvement in PHBA titer, and effectively demonstrated the potential of dynamically regulating a metabolic pathway. By separating growth from production using the synthetic quorum sensing circuit and the growth-arrest phenotype, the fundamental conflict between the evolved network’s drive towards growth and the engineered pathway modifications was alleviated. Declaration by author This thesis is composed of my original work, and contains no material previously published or written by another person except where due reference has been made in the text. I have clearly stated the contribution by others to jointly-authored works that I have included in my thesis.

I have clearly stated the contribution of others to my thesis as a whole, including statistical assistance, survey design, data analysis, significant technical procedures, professional editorial advice, and any other original research work used or reported in my thesis. The content of my thesis is the result of work I have carried out since the commencement of my research higher degree candidature and does not include a substantial part of work that has been submitted to qualify for the award of any other degree or diploma in any university or other tertiary institution. I have clearly stated which parts of my thesis, if any, have been submitted to qualify for another award.

I acknowledge that an electronic copy of my thesis must be lodged with the University Library and, subject to the policy and procedures of The University of Queensland, the thesis be made available for research and study in accordance with the Copyright Act 1968 unless a period of embargo has been approved by the Dean of the Graduate School.

I acknowledge that copyright of all material contained in my thesis resides with the copyright holder(s) of that material. Where appropriate I have obtained copyright permission from the copyright holder to reproduce material in this thesis.

Publications during candidature

Journal Articles Williams, T. C., Nielsen, L. K. & Vickers, C. E. Engineered Quorum Sensing Using Pheromone- Mediated Cell-to-Cell Communication in Saccharomyces cerevisiae. ACS Synthetic Biology 2, 136- 149, doi:10.1021/sb

Williams, T. C., Averesch, N.J.H., Winter, G., Vickers, C. E., Nielsen, L. K., Krömer, J.O. Quorum-Sensing Linked RNA interference for Dynamic Metabolic Pathway Control in Saccharomyces cerevisiae. Accepted. Metabolic Engineering.

Brennan, T. C. R., Williams, T. C., Schulz, B. L., Palfreyman, R. W., Krömer, J.O., Nielsen, L. K. Truncated Tcb3p improves tolerance towards monoterpenes in yeast. Accepted. Applied and Environmental Microbiology.

Williams, T. C., Espinosa, M, I., Nielsen, L, K., Vickers, C. E. Dynamic Regulation of Gene Expression Using Sucrose Responsive Promoters and RNA interference in Saccharomyces cerevisiae. Under review. Microbial Cell Factories.

Conference Proceedings Williams, T. C., Averesch, N.J.H., Winter, G., Vickers, C. E., Nielsen, L. K., Krömer, J.O. (2014). Quorum-Sensing Linked RNA interference for Dynamic Metabolic Pathway Control in Saccharomyces cerevisiae, Poster and rapid-fire oral presentation: Metabolic Engineering X. Vancouver, Canada. 2nd place poster award.

Williams, T. C. (2013). Engineered Quorum Sensing in Yeast. Invited speaker. Synthetic Biology and Bioengineering Workshop. Macquarie University, Sydney, Australia.

Williams, T. C., Nielsen, L. K., and Vickers, C. E. (2012). Engineered Pheromone Communication for Nutrient and Quorum Sensing in S. cerevisiae, Poster Presentation, 1st International Conference on BioNano Innovation (ICBNI). Brisbane, Australia. 1st place poster award.

Williams, T. C., Nielsen, L. K., and Vickers, C. E. (2012). Engineered Pheromone Communication for Nutrient and Quorum Sensing in S. cerevisiae, Poster Presentation: Metabolic Engineering IX. Biarritz, France.

Publications included in this thesis

Williams, T. C., Nielsen, L. K. & Vickers, C. E. Engineered Quorum Sensing Using Pheromone- Mediated Cell-to-Cell Communication in Saccharomyces cerevisiae. ACS Synthetic Biology 2, 136- 149, doi:10.1021/sb

The above publication has been incorporated as Chapter 3 in the thesis with the following contributions from each author. Contributor Statement of contribution Williams, T. C. (Candidate) Designed experiments (90%) Performed experiments (100%) Wrote the paper (80%) Nielsen, L. K. Designed experiments (5%) Wrote and edited paper (5%) Vickers, C. E. Designed experiments (5%) Wrote and edited paper (15%)

Williams, T. C., Averesch, N.J.H., Winter, G., Plan, M.R., Vickers, C. E., Nielsen, L. K., Krömer, J.O. Quorum-Sensing Linked RNA interference for Dynamic Metabolic Pathway Control in Saccharomyces cerevisiae. Under review. Metabolic Engineering.

The above publication was submitted to ‘Metabolic Engineering’ on the 22.08.14 and is currently under peer review. The manuscript has been incorporated as Chapter 4 in the thesis with the following contributions from each author. Contributor Statement of contribution Williams, T. C. (Candidate) Designed experiments (70%) Performed molecular biology and growth experiments (95%) Wrote the paper (85%) Averesch, N.J.H. Elementary flux mode analysis (100%) Molecular biology (5%) Winter, G. Designed experiments (5%) Plan, M.R. Developed and performed HPLC methods (100%) Vickers, C. E. Wrote and edited paper (5%), advised candidate. Nielsen, L. K. Wrote and edited paper (5%), advised candidate. Krömer, J.O. Designed experiments (25%) Wrote and edited paper (5%) Contributions by others to the thesis Nils Averesch performed the elementary flux mode analysis included in chapter 4. Manuel Plan developed the HPLC method described for PHBA analysis in chapter 4, and processed samples for PHBA analysis. Jens Krömer assisted in the experimental design of the work included in chapter 4, and helped revise and edit the manuscript. Lars Nielsen and Claudia Vickers provided guidance and technical assistance for all work.

Statement of parts of the thesis submitted to qualify for the award of another degree None

Acknowledgements I am highly grateful to my supervisors Claudia Vickers, Jens Krömer, and Lars Nielsen for providing me with the opportunity to pursue a career in research. In particular I appreciate the guidance in molecular biology and scientific writing that Claudia provided. Students in the Nielsen group are allowed to pursue their own curiosities and ideas and I am grateful to have been afforded this freedom. Jens Krömer and Nils Averesch were my collaborators for the work carried out in Chapter 4. It was a pleasure to work with them and it was refreshing to gain different perspectives on metabolic engineering.

To my friends in the Nielsen group, I am thankful for all of the coffee, advice, beer, and laughter that we shared (usually in that order). The people are one of the most important parts of any working environment and it was a great experience to be involved with this group.

I am thankful to my Mum and Dad for encouraging me to pursue my passions in life and to not worry too much about materialistic goals. Having these qualities may be the only way to be content with a career in science! I’m also grateful for the many ways in which they supported me throughout my lengthy education. I owe a lot to Richard Edge for helping to foster my keen interest in science from a young age.

Finally to Jessie, thank you for being so understanding of the pressures and trials which come hand- in-hand with doing a PhD. Thank you for keeping me grounded during the highs, and supporting me during the lows. Thank you for everything, you mean the world to me. Keywords quorum sensing, RNA interference, yeast, pheromone, mating, para-hydroxybenzoic acid, shikimate, dynamic regulation, elementary mode analysis.

Australian and New Zealand Standard Research Classifications (ANZSRC) ANZSRC code: 060113 Synthetic Biology 100%

Fields of Research (FoR) Classification FoR code: 0601 Biochemistry and Cell Biology 100%

Table of Contents

1 Introduction ...... 1

1.1 Synthetic Biology and Metabolic Engineering ...... 1

1.2 Saccharomyces cerevisiae is an important industrial microorganism...... 4

1.3 Metabolic Engineering conflicts with evolved function ...... 5

1.3.1 Dynamic regulation of gene expression and quorum sensing ...... 6

1.4 Engineered Quorum sensing ...... 7

1.4.1 Pheromone communication as a quorum sensing mechanism ...... 8

1.4.2 Intracellular MAPK mating response ...... 11

1.4.3 Potential for synthetic MAPK circuit design ...... 12

1.5 The pheromone response is a distinct state of the S. cerevisiae life-cycle ...... 15

1.6 Aims of this thesis ...... 15

2 The Saccharomyces cerevisiae Pheromone-Response Phenotype is a Metabolically-Active Phase ...... 17

2.1 Abstract ...... 17

2.2 Introduction ...... 18

2.3 Materials and Methods ...... 20

2.3.1 Media ...... 20

2.3.2 Strains and plasmids...... 20

2.3.3 Growth conditions ...... 25

2.3.4 Analytics ...... 25

2.3.5 Transcriptome analysis ...... 25

2.4 Results and Discussion ...... 26

2.4.1 Gene expression capacity and growth characteristics of the pheromone response phenotype ...... 26

2.4.2 Transcriptome Analysis ...... 29

2.4.3 Engineered pathway productivity during the pheromone response ...... 41

2.5 Conclusions ...... 43 3 Engineered Quorum-Sensing Using Pheromone-Mediated Cell-to-Cell Communication in Saccharomyces cerevisiae ...... 45

3.1 Abstract ...... 45

3.2 Introduction ...... 46

3.3 Materials and methods...... 49

3.3.1 Media ...... 49

3.3.2 Strains and Plasmids ...... 49

3.3.3 Sender-Receiver assays ...... 55

3.3.4 Flow cytometry ...... 55

3.3.5 Fermentation conditions ...... 56

3.3.6 Statistical analysis ...... 56

3.4 Results and Discussion ...... 58

3.4.1 Circuit topology ...... 58

3.4.2 Secreted α-pheromone elicits a population-density-dependent cell-to-cell communication response ...... 62

3.4.3 Positive feedback quorum sensing ...... 65

3.4.4 Aromatic amino acid induced quorum sensing ...... 73

3.4.5 Summary and Conclusions...... 76

4 Quorum Sensing Linked RNAi for Dynamic Pathway Control in Saccharomyces cerevisiae . 79

4.1 Abstract ...... 79

4.2 Introduction ...... 80

4.3 Materials and Methods ...... 81

4.3.1 Elementary flux mode analysis ...... 81

4.3.2 Media ...... 82

4.3.3 Growth conditions ...... 82

4.3.4 Strains and plasmids...... 83

4.3.5 Analytics ...... 91

4.3.6 Quantitative Real-Time PCR ...... 91

4.3.7 Enzyme Assays ...... 92 4.3.8 Statistical Analysis ...... 92

4.4 Results and Discussion ...... 92

4.4.1 Strain Design ...... 92

4.4.2 Validation of the QS response ...... 95

4.4.3 Stepwise engineering for improved PHBA production ...... 96

4.4.4 Validation of RNAi knock-down ...... 99

4.4.5 Conclusions ...... 102

5 General Discussion ...... 103

6 Bibliography...... 111

List of Figures Figure 1-1. The design, build, test cycle of metabolic engineering ...... 4 Figure 1-2. Mating of wild haploid yeast ...... 10 Figure 1-3. MAPK phosphorylation cascade ...... 12 Figure 2-1. Pheromone-response GFP expression, extracellular metabolites, and growth characteristics ...... 27 Figure 2-2. Transcriptional changes in central carbon metabolism in response to mating pheromone ...... 40 Figure 2-3. Engineered PHBA production pathway ...... 42 Figure 2-4. PHBA production in pheromone treated cultures ...... 43 Figure 3-1. Pheromone communication circuit design ...... 60 Figure 3-2. Sender-receiver pheromone communication...... 63 Figure 3-3. Positive feedback quorum sensing ...... 67 Figure 3-4. Model of graded signalling as a consequence of overlayed positive and negative feedback ...... 71 Figure 3-5. Graded autoinduced GFP expression ...... 72 Figure 3-6. Aromatic amino acid induced quorum sensing ...... 75 Figure 4-1. PHBA production from the shikimate pathway with pheromone quorum sensing ...... 93 Figure 4-2. Elementary flux mode analysis of the pyruvate kinase reaction knock-out ...... 95 Figure 4-3. Quorum sensing mediated PHBA production ...... 96 Figure 4-4. PHBA titers using quorum sensing mediated modulation of shikimate pathway genes. 98 Figure 4-5. Relative mRNA levels of RNAi targets ...... 100 Figure 4-6. Enzyme activity assays ...... 101 Figure 5-1. Summary of circuit-dependent quorum sensing gene expression dynamics from Chapter 3 ...... 106

List of Tables Table 2-1. Primers ...... 22 Table 2-2. Plasmids ...... 23 Table 2-3. S. cerevisiae strains ...... 24 Table 2-4. Summary of external metabolite flux rates...... 29 Table 2-5. Up-regulated genes and GO terms...... 31 Table 2-6. Down regulated genes and GO terms ...... 34 Table 3-1. Primers used in this study ...... 50 Table 3-2. Plasmids used in this study ...... 51 Table 3-3. Yeast strains used in this study ...... 53 Table 3-4. Quorum sensing circuit properties ...... 68 Table 4-1. Oligonucleotides used in this study ...... 85 Table 4-2. Plasmids ...... 87 Table 4-3. Saccharomyces cerevisiae strains used in this study ...... 89

1 Introduction

Modern civilisation is dependent upon oil for energy, fuels, and chemicals. However, due to the non-renewable, and environmentally non-sustainable nature of oil, alternative routes to these resources are required. Microbial metabolic processes are extremely diverse in nature and are capable of producing compounds (or precursors) which can replace all major products currently produced via petrochemical feedstocks (Nikolau et al., 2008, Nielsen, 2011) as well as many plant derived natural products (Xu et al., 2013). Microbial production of fuels and chemicals is an attractive alternative because these processes rely on renewable resources and are far more environmentally friendly in terms of net greenhouse gas emissions.

The field of metabolic engineering aims to manipulate metabolic processes such that target biochemicals are produced at commercially relevant yields, rates, and titers using economically viable processes. This is often extremely difficult to achieve because even the simplest of microorganisms have evolved a high level of complexity to enable their survival and proliferation. The disruption of highly complex metabolic and regulatory systems via genetic manipulations such as gene deletion or over-expression can often cause severe growth defects or result in non-viability in the host organism. This is due to the competition for carbon, energy, and redox cofactors between production pathways and endogenous processes which facilitate growth, and is problematic because for high product titers to be achieved, a high population density of producing cells is required.

A promising solution to alleviate the conflict between growth and production in metabolic engineering is to use dynamic regulation. Using this approach, relevant pathway modifications could be controlled to ‘switch on’ or even ‘switch off’ after the completion of a growth phase, so that the network modifications do not interfere with biomass accumulation. To achieve this sort of dynamic regulation of metabolism, the principles and components of synthetic biology need to be integrated with those of metabolic engineering.

1.1 Synthetic Biology and Metabolic Engineering

Synthetic biology is a rapidly advancing field with the potential to revolutionise traditional industrial biotechnology. Synthetic biology aims to make the engineering of biology easier, and allow for greater control and diversity of cellular function (Andrianantoandro et al., 2006). The term ‘synthetic biology’ encapsulates a wide area of research which is broadly defined as the design of new-, and the re-design of existing- biological parts, pathways, and organisms using engineering 1 principles such as modularisation, hierarchical organisation and abstraction, and computer aided design (Endy, 2005). Synthetic biology at the level of parts and devices has existed since the dawn of molecular biology in the 1970s. These basic parts include: promoter and terminator sequences, ribosome binding sites, protein coding sequences, replication origins, auxotrophic and prototrophic markers, and antibiotic resistance markers. Parts are typically assembled into plasmid vectors which enable the expression of genes of interest once transformed into host cells. Synthetic biology at the pathway level has traditionally been facilitated either by groups of plasmid vectors, or genomic integration of heterologous genes.

Early efforts in synthetic biology included the construction of synthetic regulatory circuits which enabled the ‘oscillation’ or ‘toggling’ of gene expression (Elowitz & Leibler, 2000, Gardner et al., 2000). Circuits such as these provided valuable insights into the fundamentals of engineering genetic regulation, but were not constructed with any particular application in mind. Since these early examples, the applications and reach of synthetic biology have become truly diverse. For example; synthetic pathways have been implemented for the production of various valuable compounds (Farmer & Liao, 2000, Ro et al., 2006, Gunawardena et al., 2010, Whited et al., 2010, Yim et al., 2011), cell-cell communication and quorum sensing (Chen & Weiss, 2005), control of gene expression using RNA parts (Win et al., 2009), the creation of logic gates and environmental biosensors enabling E. coli to detect and destroy cancer (Anderson et al., 2006) or the human pathogen Pseudomonas aeruginosa (Saeidi et al., 2011), protein scaffolding (Dueber et al., 2009), and substrate degradation (Matsushika et al., 2009). Synthetic biology was recently expanded into the realm of whole genome engineering when the Mycoplasma genitalium genome was synthesised (Gibson et al., 2008) and successfully transplanted into the cell of a closely related bacteria (Gibson et al., 2010) making the world’s first synthetic life form. This technology offers the synthetic biologist the opportunity to design an entire genome on a computer, with the possibility of choosing the exact combination of genes to be present in the genome. However, due to the overwhelming complexity of biological systems and the current state of our understanding of them, the likelihood that designer chimeric genomes or de-novo gene combinations engineered on this scale will function reliably is small. Until a minimal genome is defined, and suitable ‘chassis’ cells are developed, designer genome engineering will not proceed beyond small deviations from well understood systems.

Industrial chemical production is the first milestone for applied synthetic biology, and can be considered a ‘low hanging fruit’ in the context of the full potential of the technology. The core

2 concepts of synthetic biology are highly applicable to the metabolic engineering design cycle (Figure 1-1), which is notoriously slow due to the number of iterations and the difficulty of the techniques which are required to achieve engineering objectives. The cycle typically begins by using in silico stoichiometric models of metabolism to suggest genetic engineering targets such as gene knock-outs and over-expressions to direct carbon via a pathway of interest towards a product. After these genetic modifications have been implemented in vivo, strains are tested for performance criteria and assessed using systems biology techniques. These data are used to understand and overcome the limitations of pathway productivity in the next design cycle. Taking a synthetic biology approach to metabolic engineering problems will speed-up the design cycle by offering new engineering tools which enable more accurate and efficient genetic manipulations (Keasling, 2012). The design cycle highlights another important aspect of synthetic biology. That is the potential to gain an understanding of biology via the synthesis of biological modules and components. The field of systems biology aims to understand biological systems in their entirety and is highly intertwined with synthetic biology. The intersection of these two fields is especially evident in metabolic engineering where the tools of systems biology are frequently used to assess the performance of engineered cells, and suggests routes to further improving them (Figure 1-1). As our ability to assemble large synthetic DNA constructs continues to improve, synthetic biology is likely to complement systems biology by enabling the testing of complex hypotheses, accelerating our understanding of biological systems and therefore our ability to rationally engineer them.

3

Figure 1-1. The design, build, test cycle of metabolic engineering

The metabolic engineering process typically involves a synthesis component whereby genetic changes are designed in a ‘dry’ lab using genome scale modelling of metabolism before the build phase in the ‘wet’ lab involving genetic engineering and strain testing. System wide analysis of proteins, RNA, fluxes, and metabolites can then be used to identify pathway bottlenecks, or other potentially limiting factors which are considered as targets for improvement in the next cycle. Image taken directly from http://web.aibn.uq.edu.au/cssb/aboutUs.html, courtesy of Lars Nielsen.

1.2 Saccharomyces cerevisiae is an important industrial microorganism

S. cerevisiae is the oldest domesticated microorganism, being used in bread, beer, and wine production for millennia (Nevoigt, 2008). Aside from being integral in the production of these pillars of civilisation, S.cerevisiae is now being widely used to produce bioethanol from corn or sugar cane at industrial scale. However, the suitability of food-crop derived biomass for feedstock as well as the use of ethanol as a fuel are being challenged on the grounds that the practice is not truly sustainable due to competition with the human food chain (Hedegaard et al., 2008) and the poor fuel quality of ethanol relative to fossil fuels in existing petrol engines (Hansen et al., 2005). Given the depth of characterisation of S. cerevisiae and the ease of genetic manipulation, there has 4 been considerable interest in both engineering alternative catabolic pathways to enable the fermentation of non-food feedstocks (such as lignocellulose), as well as the engineering of anabolic pathways which lead to the production of biofuels with superior fuel properties, as well as sustainable chemicals (Nevoigt, 2008). For example, the mevalonate pathway in yeast has been engineered for the production of the “drop in” biofuel farnesene, but can also be manipulated for the production of various other valuable non-fuel isoprenoids such as the anti-malarial drug precursor artemisinic acid, the synthetic rubber precursor isoprene, and the chemotherapy drug taxol (Chandran et al., 2011). These examples illustrate the great potential for industrial bioengineering in S. cerevisiae beyond simple biofuels. However, such large scale pathway engineering is not without consequences for normal host strain physiology.

1.3 Metabolic Engineering conflicts with evolved function

From the first chemical information systems in Earth’s pre-biotic soup to today’s highly complex modern genomes, life has been governed by the imperatives of replication and survival (Dawkins, 2006). All rational metabolic engineering efforts are fundamentally limited by the fact that a given organism has a genome which has evolved to execute survival and proliferation functions. The very best metabolic engineering efforts have rendered a target compound the primary by-product of normal growth based metabolism. The ideal goal of metabolic engineering should be to render target compound production the primary life-function of an organism. Presently this is not possible due to our fundamental lack of understanding of evolved biological complexity, and the small scale at which genetic modifications are implemented.

Since its inception over 20 years ago, the field of metabolic engineering has made significant progress in the directed manipulation of life for the sustainable production of chemicals, fuels, and pharmaceuticals (Stephanopoulos, 2012). Initial attempts at over-producing target compounds involved the overexpression of the relevant pathway genes and the elimination of enzymes which compete for pathway substrate. However, these strategies are usually only effective at converting a small fraction of the available carbon source into the desired product since most is required for normal growth. In order to truly convert cells into chemical factories, the evolutionary drive towards resource consumption, growth, and survival must be replaced with the objective of compound production. This would allow target compound yields to approach 100% of their theoretical maximums. One way to approach this problem is to separate growth and production phases so that engineered pathway productivity does not compete with growth based metabolism.

5

The simplest way to separate growth and production phases is to engineer dynamic regulation of the expression of relevant metabolic pathway genes. In silico modelling has shown that dynamic regulation of engineered pathways can increase desired compound yields (Gadkar et al., 2005, Anesiadis et al., 2008). However, the current methods for implementing dynamic regulation are not economically feasible in an industrial context because they rely on expensive chemical inducers/repressors of gene expression (Nevoigt et al., 2007, Keasling, 2010). For example, popular inducers include the Tet ON/Tet OFF induction systems which require the addition of the complex hormone β-estradiol to yeast cultures (McIsaac et al., 2011), as well as copper induction from the CUP1 promoter (Etcheverry, 1990), and galactose induction (Westfall et al., 2012). Determining the feasibility of using such inducer compounds in industrial settings requires complex techno- economic analysis of specific process conditions and profit margins. However, it is a certainty that they add at least some cost to a bioprocess, making any autonomous system far-superior to genetic systems which require inducers.

1.3.1 Dynamic regulation of gene expression and quorum sensing

One of the main objectives of synthetic biology is to link environmental inputs to the precise control of cellular output (Endy, 2005). This typically involves the use of transcriptional logic gates to facilitate synthetic gene network construction. Such logic gates require a diverse array of biological mechanisms for activating transcription and much progress has been made in this area, in particular for bacterial systems (Purnick & Weiss, 2009). Given the greater level of complexity inherent in eukaryotic cells such as yeast, there has been a greater challenge in the construction of dynamic control circuits. However, the additional levels of regulatory ‘currency’ present in eukaryotic cells may mean that there potentially are greater opportunities for synthetic circuit design (Haynes & Silver, 2009).

Expression of heterologous genes or pathways in yeast has traditionally relied upon a small cohort of well characterised promoter and terminator sequences. Constitutive promoters such as TEF1 and ADH1, inducible promoters such as GAL1, and repressible promoters such as MET3 have all been utilised extensively in the past (Blount et al., 2012). The use of induction/repression systems based on GAL or MET3 promoters is limited by the need to add expensive compounds such as galactose or methionine to fermentations. In addition, the popular galactose induction system is too leaky to be used for the dynamic control of sensitive processes such as RNA interference in yeast (Drinnenberg et al., 2009). Other emerging options for the dynamic control of gene expression in yeast lies in the use of secondary mRNA structures such as aptamers and ribozymes in the 5’ or 3’ 6 untranslated regions (UTR) of genes of interest (Win & Smolke, 2007, Win & Smolke, 2008, Beisel & Smolke, 2009, Culler et al., 2010, Babiskin & Smolke, 2011, Chang et al., 2012). These systems work via the binding of an intracellular metabolite to an RNA aptamer, thereby altering the RNA conformation and activity of a self-cleaving ribozyme or ribosome binding site on an mRNA molecule. RNA aptamers can be evolved in vitro for almost any compound using SELEX (Stoltenburg et al., 2007) and have huge potential for use as designer control modules for intracellular metabolites of interest. However, because aptamers which can function in vivo are only available for a small range of compounds these systems currently lack suitability in most industrial metabolic engineering contexts due to the need to add these compounds as inducers to fermentation medium. Furthermore, the current suite of ribo-regulators suffer from significant leakiness in expression and poor dynamic ranges. An attractive alternative to chemically inducible systems lies in the pervasive phenomena of microbial cell-to-cell communication called quorum sensing.

Quorum sensing is a form of cell-cell communication used by microorganisms to detect their population density, and modulate gene expression to exhibit relevant cellular-, and population-level phenotypes. Quorum sensing has been discovered in a variety of microorganisms, and is used for the control of a diverse and fascinating array of cell phenotypes (Williams et al., 2007). It has great potential for application in biotechnology, and in particular, metabolic engineering (March & Bentley, 2004, Choudhary & Schmidt-Dannert, 2010, Tsao et al., 2010, Song et al., 2011, Carter et al., 2012). Engineered quorum sensing provides the potential to mitigate the negative effects of product toxicity and metabolic burden, while facilitating ‘ON’ or ‘OFF’ states of gene expression according to population growth phase. Gene expression could not only be autonomous, but coordinated at the level of a whole population rather than only within single cells.

1.4 Engineered Quorum sensing

In order for biotechnology to realise its potential, engineered genetic regulatory systems require a greater level of complexity and efficacy (Holtz & Keasling, 2010) so that they more closely resemble natural systems. Quorum sensing is one such regulatory mechanism which has previously been engineered, mainly at the ‘proof of concept’ stage, but also with the prospect of more immediate applications. Engineered quorum sensing also offers the potential for gaining a greater scientific understanding of the functional principles and social-microbiological significance of these systems.

7

Engineered quorum sensing has taken inspiration from many of the systems observed in nature in terms of both mechanism and function. Microbial pattern and biofilm formation, regulated cell death, chemotaxis and motility, and even cellular memory have all been successfully engineered using quorum sensing architectures (Carter et al.). As mentioned previously, quorum sensing has the potential to mitigate the negative effects of metabolic burden and product toxicity in industrial microorganisms. One specific example of this includes the minimal rewiring of the native AI-2 quorum sensing network in E.coli for recombinant protein overexpression (Tsao et al., 2010). It has long been documented that the overexpression of recombinant proteins places a metabolic burden on their host producers (Bentley et al., 2009), but by controlling expression using quorum sensing the metabolic burden was evenly distributed across the entire population after autonomous induction (Tsao et al., 2010). A particularly interesting example of the potential for engineered quorum sensing was demonstrated in the creation of an E.coli strain which could sense and kill P. aeruginosa, a human pathogen (Saeidi et al., 2011). P. aeruginosa controls virulence behaviour using a quorum sensing system, and in this study a non-pathogenic E.coli was engineered to detect the P. aeruginosa quorum sensing signal and respond by a) producing a compound which is toxic to P. aeruginosa and b) undergoing cell autolysis once a critical quorum sensing signal threshold is reached, thereby releasing the toxic compound from within the E.coli cell and eliminating P. aeruginosa (Saeidi et al., 2011). A similar system has been engineered in E.coli for the detection and destruction of hypoxic cancer cells (Anderson et al., 2006). The E.coli strain was engineered to conditionally invade and destroy cancer-derived human cells when a) E.coli population density was high, as indicated by an integrated V. fischeri quorum sensing circuit and b) low oxygen conditions were detected via a hypoxia responsive fdhF promoter (Anderson et al., 2006). Bacterial quorum sensing systems have a wide potential for engineering application in biotechnology.

1.4.1 Pheromone communication as a quorum sensing mechanism

The potential for engineered quorum sensing is obvious in a variety of metabolic engineering contexts. A robust, tuneable, engineered quorum sensing mechanism has not previously been reported in S. cerevisiae. There is one existing QS circuit in yeast which relies on detection of plant cytokinins (Chen & Weiss, 2005). However, this circuit is problematic because of the engineering required to integrate the sender signal with endogenous yeast metabolism. This interface created the need to avert a lethal phenotype and resulted in a system dependent on a galactose inducible promoter. Aside from this burden, the system was only able to ‘switch’ at relatively low population densities, precluding its use for high titer production of commercial compounds. A potential solution to both of these problems is to adapt a native cell-cell communication mechanism, since an 8 interface with endogenous metabolism already exists for signal transduction to proceed through. The native quorum sensing system of S. cerevisiae is involved in the coordination of stationary phase entry (Chen & Fink, 2006), and is not an ideal system for use in an industrial context due to the low gene expression capacity and metabolic activity in the stationary phase. The native pheromone-based cell-cell communication system in S. cerevisiae is more promising due to its depth of characterisation, and its potential to induce a state of cellular growth arrest which may not have the stress responses and low productivity associated with stationary phase.

Most laboratory yeast strains used in metabolic engineering are haploids which are not able to reproduce sexually. An understanding of the wild yeast mating process is therefore necessary for the implementation of an artificial quorum sensing system in a haploid asexual laboratory strain. Wild yeast exist in a diploid sixteen chromosome state and grow asexually most of the time. When cells are starved of nitrogen they undergo meiotic division to produce four haploid spores. Two of the spores are mating type ‘a’ (MATa) and two are mating type ‘α’ (MATα). Wild haploid yeast can switch their mating type by copying mating genes from a silent chromosome locus to the MAT locus. This gene transposition is facilitated by an endonuclease coded by the homothallic (HO) gene, which causes a double strand break at the MAT locus. Haploid strains are much easier to manipulate than diploid (due to only one allele being present for each gene, not two) and laboratory strains are therefore all ho mutants, allowing them to exist as stable haploids which are incapable of mating type switching and subsequent sexual reproduction.

When haploid yeast (including ho mutants) detect the pheromone from a nearby cell of the opposite mating type (Figure 1-2) they begin to produce their own mating pheromone, so that the other cell can detect their presence and mating type and respond accordingly. α-Pheromone is a 13 amino acid peptide (sequence WHWLQLKPGQPMY), while a-pheromone is a 12 amino acid peptide with a farnesyl group (sequence YIIKGVFWDPAC). At a threshold concentration of ~30 µM, cells undergo a morphological transition where the cellular cytoskeleton forms a pear-shaped projection called a ‘shmoo’ (Malleshaiah et al., 2010) (Figure 1-2). The nucleus of each cell moves to the shmoo tip, cell division arrests in the G1 phase, cell membranes and nuclei fuse, and DNA is combined to form a MATa/MATα diploid (Figure 1-2). Cells are able to detect pheromone gradients via a graded transcriptional response, although the shmoo morphology is exhibited in a bimodal fashion (Paliwal et al., 2007, Malleshaiah et al., 2010). This allows cells to ‘choose’ the most suitable mating partner if more than one is available (Paliwal et al., 2007, Malleshaiah et al., 2010). Once a diploid has been formed, mating-specific genes are suppressed and cells do not

9 respond to either type of mating pheromone until they have once again undergone meiosis to become haploid (Herskowitz, 1988).

Figure 1-2. Mating of wild haploid yeast

Haploid MATa and MATα yeast cells detect the mating pheromone of the opposite mating type (i), exhibit polarised growth towards the mating partner via ‘shmoo’ projections (ii), and fuse cell membranes and nuclei to become diploid (iii).

The native pheromone communication system has great potential for conversion into a quorum sensing mechanism. Enabling an ‘a’ type cell to produce its own α-pheromone (or vice versa) would in theory convert the hormone response to an endocrine response, and render the concentration of pheromone outside of the cell as a proxy for population density. Genes of interest

10 could be regulated by pheromone responsive promoters, and the strength of pheromone expression could then be adjusted to tune the timing of gene expression activation. The successful engineering of any biological system requires a thorough consideration of the underlying components and an awareness of any engineering which has previously been achieved. The mitogen activated protein kinase (MAPK) is one such module in yeast that has been a test-bed for many synthetic biology devices and applications.

1.4.2 Intracellular MAPK mating response

The intracellular mating response utilises a mitogen activated protein kinase (MAPK) signalling pathway, which has been recently reviewed (Bardwell, 2005). Figure 1-3 summarises the key aspects of the MAPK pathway. Upon binding of pheromone to the appropriate membrane G-protein coupled receptor (ste2 for α-pheromone and ste3 for a-pheromone) the Gα subunit exchanges GDP for GTP, causing the release of the Gβγ heterodimer (Figure 1-3) (Bardwell, 2005). Membrane bound Gβγ is then able to induce phosphorylation of the MAPK cascade via association with the Ste11/Ste5 complex (Bardwell, 2005).

The Ste5 protein has no catalytic activity but serves as a scaffold for the localisation and phosphorylation of Ste11p, Ste7p, and Fus3p MAPKs. Phosphorylated Fus3p dissociates from the Ste5p scaffold and together with phosphorylated Kss1p, de-represses the Ste12p transcription factor by interacting with Dig1p and Dig2p (Malleshaiah et al., 2010) (Figure 1-3). Fus3p also activates Far1p, which causes cell cycle arrest association with Cdc28p, and polarised growth towards a mating partner with Cdc24p (Elion et al., 1993). Far1p is also thought to be responsible for the repression of about 100 genes which are involved in cell cycle regulation, and plays a strong part in the maintenance of the G1 growth arrest state (Roberts et al., 2000). The Ste12p transcription factor up-regulates the expression of approximately 200 genes (Roberts et al., 2000), of which 100 are induced by direct Ste12p DNA binding (Zeitlinger et al., 2003). Together Ste12p and Far1p work to coordinate the mating response, which is a cellular state separated from normal growth metabolism but also distinct from the nutrient limitation induced stationary phase G1 arrest (Thevelein, 1994).

11

Figure 1-3. MAPK phosphorylation cascade

The binding of mating pheromone to a membrane receptor initiates a mitogen activated protein kinase phosphorylation cascade which results in the transcription of genes required for the mating response.

1.4.3 Potential for synthetic MAPK circuit design

The MAPK pathway has been studied in yeast since the dawn of molecular biology in the 1970’s and is subsequently one of the best understood in eukaryotic biochemistry (Bardwell, 2005). This depth of characterisation makes the MAPK pathway ideal for synthetic biology, and this has been demonstrated recently via several recent reports.

Ingolia and Murray (Ingolia & Murray, 2007) were able to tune the MAPK pheromone response into a bimodal, rather than a graded transcriptional output by engineering various intracellular positive feedback loops. These positive feedback loops not only resulted in transcriptional bimodality, but also in a sustained MAPK response even after the removal of pheromone. A sustained transcriptional response in the absence of pheromone was achieved by decoupling the intracellular circuitry from the G-protein coupled receptor Ste2p, which is the main point of

12 regulated desensitisation to pheromone via the Sst2p suppressor system (Dohlman et al., 1996). This decoupling was achieved by engineering positive feedback expression (via the FUS1 promoter) of STE4, STE11, and STE5 genes respectively. Even after the removal of pheromone, the expression of a yeast enhanced yellow fluorescent protein (yEGFP) from the pheromone responsive promoter pFUS1 continued due to the establishment of an intracellular positive feedback loop which was independent from the Ste2p alpha pheromone membrane receptor. The other consequence of the intracellular positive feedback loops was transcriptional bimodality, the existence of ‘on’ and ‘off’ states of pheromone response rather than a graded continuum. Interestingly, bimodality was also a product of gene copy number, where in some cases a higher copy number was required, as in the case of STE11 where three copies were required for the induction of a persistent and stable active state. The basal and induced expression levels from the FUS1 promoter were also manipulated to tune the pheromone response. Screening of a FUS1 promoter library revealed sequence variants which resulted in decreased basal expression (or noise) which was indistinguishable from wild type fluorescence levels as well as increased induction levels which were higher than those observed with the native promoter.

Bashor et al. (Bashor et al., 2008) were also able to demonstrate that intracellular feedback loops can be engineered in the MAPK response using competitive protein scaffold interactions between various MAPK components. For example when the MSG5 gene (a phosphatase which negatively regulates the MAPK response) was placed under positive feedback (via the FUS1 promoter) and cells were treated with pheromone, a negative feedback loop was established which resulted in a damped response (GFP expression). Similarly when STE50 (positively modulates the MAPK response by promoting interaction between Ste20p and Ste11p) expression was placed under the control of the FUS1 promoter in addition to the native copy, increased positive feedback was observed in the MAPK response. When Msg5p and Ste50p respectively were physically tethered to the Ste5 scaffold complex via leucine ‘zippers’ their respective effects on MAPK response were even more pronounced. The strength of the synthetic scaffold recruitment, as well as the strength of the feedback promoter were used to tune the MAPK response towards a variety of desired outputs. This study effectively demonstrated the utility of engineering gene expression outputs using unnatural protein-protein interactions within a signalling network.

While there has been significant progress in engineering aspects of the intracellular MAPK response, there have also been reports of manipulating the pheromone based communication which triggers this response. Groß et al. (Groß et al., 2011) engineered both constitutive and positive

13 feedback expression of the alpha pheromone in separate ‘sender’ MATa cell types and co-cultured them in different ratios with receiver MATa cells which produced GFP in response to pheromone. Regot et al. (2011) took this theme a step further by creating MATα cells which only produced alpha pheromone in response to certain combinations of environmental stimuli such as doxycycline, sodium chloride, galactose, and glucose. In order to interpret the signal, MATa cells were engineered not only to express GFP in response to alpha pheromone received from the MATα cells, but also to express essential components of the MAPK network (such as fus3p) only in the presence or absence of combinations of the environmental stimuli mentioned above. This complex network of signal inputs was used to create various biological-computation functions such as; AND, OR, NOR, and NAND. The complexity of these computational outputs was distributed between different ‘sender’ and ‘receiver’ cell types (a and α). This biological computation is distinguished from traditional efforts at engineering cellular circuitry in that it distributes the metabolic load of what would be an extremely complex circuit within a single cell, throughout a population of different cell types which are able to communicate with one another.

A common theme between these reports is that they either require the addition of synthetic alpha- peptide to growth cultures, or the co-culture of MATa with MATα cells. An alternative is to re-wire pheromone production and response so that pheromone producing cells also respond to their own intercellular pheromone, a phenomenon termed autocrine activation. An autocrine activation of the MAPK pheromone response was first achieved by Nakayama et al. (1987) who expressed the alpha factor receptor gene STE2 in a MATα, alpha-pheromone producing cell. STE2 expression was controlled by a galactose inducible promoter so that when grown on galactose, cells responded to their endogenously produced alpha-pheromone as part of a positive feedback loop, and arrested growth as they would in the presence of a mating partner. This study identified that although MATa and MATα cells have distinct pheromone receptors, they share a common intracellular signal response mechanism, which we now know as the MAPK pathway in yeast. The expression of the alpha-pheromone producing mfα1 gene in an alpha-pheromone MATa cell population also resulted in an autocrine response of growth arrest, but only when both BAR1 and SST2 genes were non- functional (Whiteway et al., 1988). The MATALPHA2 gene encodes a protein that represses the transcription of mating specific genes in diploids (Siliciano & Tatchell, 1986), and when this gene is deleted in haploids, both mating pheromones and both pheromone receptors are expressed, resulting in an autocrine growth-arrest response (Rivers & Sprague, 2003). The autocrine response exhibited upon the repression of MATALPHA2 expression (via switching to glucose in a MATALPHA2- strain with galactose promoter driven expression of MATALPHA2) was found to

14 be attenuated by Sst2p and Asg7p within 3 hours (Rivers & Sprague, 2003). Collectively, these results indicate that the yeast pheromone response can be re-wired for autocrine activation, using the high strength and inducible GAL promoter. These studies demonstrate that it may be possible to engineer autonomous population density dependent signalling using the MAPK pheromone response in yeast.

1.5 The pheromone response is a distinct state of the S. cerevisiae life-cycle

An interesting property of pheromone-induced cell-cycle arrest in yeast is the fact that it is a morphologically distinct phase of existence in the cell cycle (Bardwell, 2005). Together Ste12p and Far1p work to coordinate the mating response, which is a cellular state separated from normal growth metabolism. It is possible that the pheromone arrest phenotype may serve as a mechanism to decouple product formation from growth, allowing the implementation of genetic manipulations which would otherwise compete with biomass formation. Very little is known about the metabolic state of cells responding to pheromone, and it is not clear if this phase is suitable for metabolic engineering strategies. To date there has been some systems analysis of the pheromone response (Serikawa et al., 2003, Gruhler et al., 2005) as well as a transcriptomics study on a FAR1 overexpressing strain (Busti et al., 2012). However, none of these reports were relevant to the specific question of ‘metabolic activity’ in the pheromone response. For an engineered pheromone quorum sensing circuit to be applied to control flux through a production pathway it is imperative that the pheromone response phenotype is first assessed for its potential as a production phase.

1.6 Aims of this thesis

The overarching goal of this work is to decouple growth and production phases. This would enable the implementation of growth-limiting engineering strategies after the completion of a population growth phase. Achieving the separation of growth and production phases requires the construction of dynamic regulatory systems with spatiotemporal control of not only gene expression, but of an entire metabolic pathway. These goals have been approached by re-engineering the S. cerevisiae pheromone response system, and can be divided into three specific components which comprise the three main chapters of this thesis.

1. Identify metabolic changes associated with the S. cerevisiae mating phenotype, and determine the level of suitability as a metabolic engineering production phase target.

15

2. Engineering of the native pheromone communication system into a tuneable quorum sensing module.

3. Application of the engineered quorum sensing module to dynamically control a metabolic pathway for the production of a target compound.

16

2 The Saccharomyces cerevisiae Pheromone-Response Phenotype is a Metabolically-Active Phase

2.1 Abstract

The growth characteristics and underlying metabolism of microbial production hosts such as Saccharomyces cerevisiae are critical to the productivity of metabolically engineered pathways. The growth arrest phenotype associated with the Saccharomyces cerevisiae pheromone response is potentially an attractive production phase because it offers the possibility of decoupling production from population growth. However, little is known about the metabolic phenotype associated with the pheromone response, and it is not clear if it is a suitable production phase. Analysis of extracellular metabolite fluxes of pheromone-treated and non-treated populations showed that although growth was significantly reduced, glucose uptake and metabolic end-product formation rates were as high as in the exponential phase. Furthermore, re-analysis of previously reported transcriptomic data indicated that there were significant changes in expression of a variety of key genes in central carbon metabolism. Together these observations demonstrate that a highly active and distinct metabolism underlies the pheromone response phenotype. Finally, pheromone treated populations were tested for changes in production of the heterologous product para-hydroxybenzoic acid (PHBA) from a minimally engineered shikimate pathway. There was no significant difference in PHBA titer between pheromone treated and exponentially growing populations. These results indicated that the pheromone response phenotype is suitable for use as a production phase which is separate from normal growth based metabolism.

17

2.2 Introduction

Metabolic productivity is dependent not only on the engineered features of a production pathway, but also on the general capacity of the global network for metabolic and gene expression activity. As with most industrial hosts, when grown in liquid culture Saccharomyces cerevisiae populations undergo an exponential growth phase where carbon and nitrogen resources are rapidly consumed until they limit biomass production. Carbon or nitrogen limited populations cease rapid growth and enter a ‘stationary phase’ which is characterized by the induction of stress survival mechanisms and a drastic reduction in the overall rate of protein synthesis relative to the exponential phase (Werner-Washburne et al., 1993). While many secondary metabolic pathways are activated in this phase, it is clearly not an ideal production phase due to the reduced carbon flux through the network and lower gene expression capacity.

During exponential growth about 90 % of cellular energy is directed towards ribosome biogenesis (Warner et al., 2001). This allows high expression levels of heterologous enzymes and together with the high glycolytic fluxes of the growth phase, enables high yields of products which are coupled to this phase. Similarly to the mating phenotype, entry into the stationary phase also involves cell-cycle arrest at G1, but is accompanied by the induction of stress resistance genes due to nutrient limitation, and an overall reduction in the rate of protein synthesis (Werner-Washburne et al., 1993). The pheromone response phenotype represents a unique phase in the life-cycle of S. cerevisiae, which might potentially be useful as a production phase for metabolic engineering purposes. Pheromone treatment results in polarized growth, remodelling of cellular morphology and global transcription patterns, and arrest of growth in the G1 phase of the cell-cycle (Bardwell, 2005).

Pheromone treatment could result in a number of different scenarios with respect to a metabolic engineering outcome for a specific product. These include; an unproductive phenotype similar to the G1 arrest of the stationary phase, higher productivity due to the limitation of carbon flux towards biomass, or no overall effect on cellular productivity. In addition to considerations of general metabolic productivity, it is also important to identify any fundamental differences in metabolism, as they can help to decide which heterologous products will be favoured by the natural fluxes in the network. For example, specific anabolic pathways could be up-regulated in response to pheromone, suggesting that industrial products 18 which are derived from these pathways would have higher yields during the pheromone response.

The concept of limiting biomass formation to enhance cellular productivity has received some attention in the field of therapeutic protein production in mammalian cell cultures (Kumar et al., 2007). In particular, the manipulation of the Eukaryotic cell cycle to induce a growth arrest phenotype has been successfully used to improve heterologous protein production. For example, the over-expression of the cyclin dependent kinase inhibitor p21and its inducer C/EBPα in a Chinese Hamster Ovary cell line resulted in stable cell-cycle arrest in the G1 phase and a 10-15 fold higher protein productivity per cell (Fussenegger et al., 1998). Similarly, the overexpression of the p21 cyclin inhibitor in an NS0 mouse myeloma cell line increased protein productivity ~4 fold (Watanabe et al., 2002). The increased productivity due to p21 mediated cell-cycle arrest has been attributed to higher mitochondrial membrane potential providing more ATP for peptide bond formation, and increased ribosomal biogenesis (Bi et al., 2004, Khoo & Al-Rubeai, 2009). It is possible that the cell cycle arrest phenotype of the S. cerevisiae pheromone response could result in similar productivity improvements.

In this work, the pheromone response phenotype in S. cerevisiae was investigated as a growth arrest phase for metabolic engineering applications. Fundamental metabolic differences in pheromone treated populations were identified by comparing external metabolite fluxes, metabolic and global gene expression patterns, and the production capacity of a heterologous compound of industrial importance, para-hydroxybenzoic acid (PHBA) (Krömer et al., 2012).

19

2.3 Materials and Methods

2.3.1 Media

Strains were grown experimentally in chemically defined liquid medium with 5 g/L ammonium sulfate, 20 g/L glucose, vitamins and trace elements as described previously (Verduyn et al., 1992). During strain construction purified amino acids (Sigma) were used to complement appropriate auxotrophies in agar plates (same composition as chemically defined media above) while YPD or YPG supplemented with appropriate antibiotics was used during gene deletion procedures. E. coli DH5α was used for plasmid propagation/storage and was grown in LB medium with appropriate antibiotics.

2.3.2 Strains and plasmids

Primers, plasmids, and strains used in this study are shown in Table 2-1, Table 2-2, and Table 2-3 respectively. DNA manipulation and propagation were carried out using standard techniques (Sambrook & Russell, 2001) unless stated otherwise. All S. cerevisiae transformations were carried out using the lithium acetate method (Gietz & Schiestl, 2007). Deletion of the FUS1 gene was performed with the reusable LoxP-KanMX-LoxP cassette as described previously (Güldener et al., 1996) using the FUS1KOF and FUS1KOR primers to amplify the phleomycin deletion cassette from pUG66 and the FUS1DCF and FUS1DCR primers to check the chromosomal locus for deletion and marker removal. Strains transformed with yeast integrating plasmids were screened for correct integration using PCR as previously described (Stansfield & Stark, 2007). The BAR1 ORF was replaced with the bleomycin resistance cassette from pUG66 using the same method as for FUS1 except with the BAR1 primers.

The pTEF1-yEGFPCLN2PEST-416 plasmid was made by inserting the TEF1 promoter amplified from pSF019 using primers 9/10 into the yEGFPCLN2PEST-pRS406 with XhoI/EcoRI. The gene-CYC1 terminators for UBiC, ARO4, and TKL1 were amplified from the pTCW022 plasmid (see chapter 4) using primers 13/14, 17/18, and 20/21. TEF1 promoters with 40 bp homologous overlap junctions for each gene-CYC1 terminator cassette were amplified from pSF019 using primers 11/12 (UBiC), 15/16 (ARO4), and 19/20 (ARO4). Overlap extension PCR (Horton et al., 1989) was used to assemble the TEF1

20 promoters 5’ of their respective ORF-CYC1 terminators. The pTEF1-UBiC-CYC1t cassette from the overlap assembly process was then PCR amplified using primer pair 23/24 and inserted into pRS406 using XhoI and EcoRI to make plasmid PHBA01. Similarly the ARO4 expression cassette was amplified (primers 25/26) and inserted into pHB01 using EcoRI and NotI to make PHBA02. The TKL1 expression cassette generated by overlap assembly was amplified with primer pair 27/28 and inserted into PHBA02 using NotI cut sites at both ends to make PHBA03.

21

Table 2-1. Primers

Primer name Primer 5 to 3 sequence number FUS1KOF 1 TTTCCTTTAAGAGCAGGATATAAGCCATCAAGTTTCTGA AAATCAAAATGCCAGCTGAAGCTTCGTACG FUS1KOR 2 TACAGAATTATAGGTATAGATTAAATGCGAACGTCAATA TTATTTTCATCACTATAGGGAGACCGGCAG FUS1DCF 3 GCGCTGTCTCATTTTGGTGC FUS1DCR 4 TGCATTCCCTAGTTTCGCGG BAR1 KOF 5 CGCCTAAAATCATACCAAAATAAAAAGAGTGTCTAGAA GGGTCATATACCAGCTGAAGCTTCGTACG BAR1KOR 6 CTATATATTTGATATTTATATGCTATAAAGAAATTGTACT CCAGATTTCTCACTATAGGGAGACCGGCAG BAR1DCF 7 AGAGATGCGTTGTCCCTGTT BAR1DCR 8 ATGGTCAGAATGGGCGCTTG XhoI-pTEF1F 9 TATTATCTCGAGGCACACACCATAG EcoRI-pTEF1R 10 TATTATGAATTCTTGTAATTAAAACTTAGATTAGATTG pRS5’ pTEF1F 11 CGATAAGCTTGATATCGAATTCCTGCAGCCCCTCGAGGC ACACACCATAG UBiC5' pTEF1R 12 GTGTCAAAGCTGGATGGGACATTTGTAATTAAAACTTAG ATTAGATTG pTEF13' UBICF 13 CAATCTAATCTAAGTTTTAATTACAAATGTCCCATCCAG CTTTGACAC pTEF1Fb5' 14 AAGGAGTAGAAACATTTTGAAGCACGATGAGAGTGTAA CYC1tR ACTGC UBiC3' pTEF1Fb 15 GCAGTTTACACTCTCATCGTGCTTCAAAATGTTTCTACTC CTT ARO45' pTEF1R 16 GAACATTGGAGATTCACTCATTTGTAATTAAAACTTAGA TTAGATTG pTEF1F3' ARO4F 17 CAATCTAATCTAAGTTTTAATTACAAATGAGTGAATCTC CAATGTTC pTEF1F5' CYC1tR 18 CTATGGTGTGTGCCTCGAGACGATGAGAGTGTAAACTGC ARO43' pTEF1F 19 GCAGTTTACACTCTCATCGTCTCGAGGCACACACCATAG TKL15' pTEF1R 20 CAATGTCAGTGAATTGAGTCATTTGTAATTAAAACTTAG ATTAGATTG

22 pTEF1F3' TKL1F 21 CAATCTAATCTAAGTTTTAATTACAAATGACTCAATTCA CTGACATTG pRS3' CYC1tR 22 GGCGGCCGCTCTAGAACTAGTGGATCCCCCACGATGAG AGTGTAAACTGC UBiC- pTEF1F 23 TATTATGTCGACCTCGAGGCACACACCATAG UBiC-CYC1tR 24 TATTATGAATTCACGATGAGAGTGTAAACTGC ARO4- pTEF1F 25 TATTATGAATTCCTCGAGGCACACACCATAG ARO4-CYC1tR 26 TATTATGCGGCCGCACGATGAGAGTGTAAACTGC TKL1-AfeI- 27 GTTTACACTCTCATCGTGCGGCCGCCACCGCTCGAGGCA pTEF1F CACACCATAG TKL1-AfeI- 28 CTAAAGGGAACAAAAGCTGGAGCTCCACCGACGATGAG CYC1tR AGTGTAAACTGC

Table 2-2. Plasmids

Name Details Origin pRS406 URA3 integrating vector (Sikorski & Hieter, 1989), Euroscarf yEGFPCLN2PEST- Destabilized GFP gene in pRS406 (Williams et al., 2012) pRS406 pSF019 pTEF1 driven lacZ expression Unpublished study pTEF1- TEF1 promoter driven GFP expression This Study yEGFPCLN2PEST pUG6 Contains geneticin resistance marker gene (Güldener et al., 1996), Euroscarf pUG66 Contains phleomycin resistance marker gene (Güldener et al., 1996), Euroscarf pTCW022 pFUS1J2-UBiC-CYC1t-pFUS1J2-ARO4- Chapter 4 CYC1t-pFUS1J2-TKL1-CYC1t-pRS406 PHBA01 pTEF1-UBiC-CYC1t-pRS406 This study PHBA02 pTEF1-UBiC-CYC1t-pTEF1-ARO4-CYC1t- This study pRS406 PHBA03 pTEF1-UBiC-CYC1t-pTEF1-ARO4-CYC1t- This study pTEF1-TKL1-CYC1t-pRS406

23

Table 2-3. S. cerevisiae strains

Name Genotype Notes Origin

CEN.PK113- MATa; ura3-52; MAL2-8C; SUC2 Haploid MATa lab strain Euroscarf 5D

PSP01 CEN.PK113-5D, bar1::phleo BAR1 gene deleted This study

PSP02 CEN.PK113-5D, bar1::phleo, fus1::KanMX BAR1 and FUS1 deleted This study

PSP03 CEN.PK113-5D, bar1::phleo, fus1::KanMX, ura3-52::pTEF1-GFPCLN2PEST-ADH1t- PSP02 + constitutive This study pRS406 destabilized GFP expression

PSP04 CEN.PK113-5D, bar1::phleo, fus1::KanMX, ura3-52::pRS406 Prototrophic control strain This study

PSP05 CEN.PK113-5D, bar1::phleo, fus1::KanMX, ura3-52::pTEF1-UBiC-CYC1t-pTEF1- para-hydroxybenzoic acid This study ARO4-CYC1t-pTEF1-TKL1-CYC1t-pRS406 producing strain

24

2.3.3 Growth conditions

Shake-flask fermentations were carried out at 30 ˚C, 200 rpm with aluminium foil used to cover flask tops and media making up 10% of the baffled flask volume. Single colonies from solid media were used to inoculate 10 mL of CBS media. After 24 h growth cells were passaged into a second pre-culture (25 mL) and grown to mid-log phase (OD660nm of 1-5) prior to inoculation of the experimental culture (50 mL) at an OD660nm of 0.4. Synthetic alpha-pheromone (Genscript) was added to flasks at a final concentration of 1 µM at indicated time points. Samples for analysis of extracellular metabolites were obtained by spinning 1 mL of culture at 13000g for 7 minutes at 4 ˚C and transferring the supernatant for storage at -20 ˚C until analysis. Population density was measured using absorbance at 660nm

(OD660nm) on a spectrophotometer (LibraS4, Biochrom UK). OD660nm values were converted to biomass using a previously determined conversion factor of 0.243 g dry cell weight per 1 OD.

2.3.4 Analytics

Extracellular glucose, ethanol, glycerol, and acetate concentrations were determined using HPLC as previously described (Dietmair et al., 2010). Metabolite concentration were normalised to carbon moles along with biomass, and these values were used to calculate CO2 production and O2 consumption as described in (Stephanopoulos et al., 1998). pHBA concentrations were measured in extracellular supernatants as described in chapter 4, section 4.3.5.

2.3.5 Transcriptome analysis

The log normalised fold changes in gene expression and corresponding p values for populations treated with 50 nM alpha pheromone for 2 hours were obtained from a previous study (Roberts et al., 2000). The gene names were assigned to GO terms using the SGD GO slim mapper, and any genes under the categories of ‘sexual reproduction’ and/or ‘conjugation’ were removed from the data. Genes with p values ≤ 0.01 and fold changes ≥ 2 were then used for GO term analysis. For central carbon metabolic gene analysis transcripts with p values ≤ 0.05 and fold changes ≤ -1.5 and ≥ +1.5 were considered.

25

2.4 Results and Discussion

2.4.1 Gene expression capacity and growth characteristics of the pheromone response phenotype

An ideal production phase for metabolic pathway engineering has a high general capacity for gene expression such that heterologous enzymes can be expressed to a high level. A production phase also needs to have high metabolic activity through central carbon metabolism. These factors are particularly important to assess for the pheromone response phenotype because since all of the other known types of growth arrest in yeast result in a ‘stationary phase’ which is characterised by low gene expression and metabolic activity (Werner-Washburne et al., 1993).

In order to address the question of gene expression capacity of the cell-cycle arrest associated with the pheromone response phenotype GFP expression levels between strains treated with and without synthetic α-pheromone were compared. The TEF1 promoter was used to control GFP expression because TEF1 mRNA levels are unaffected by pheromone treatment (Roberts et al., 2000), and the promoter is constitutively active (Da Silva & Srikrishnan, 2012). Consequently, the level of GFP should reflect the overall gene expression capacity of the populations. To induce the mating response, alpha-pheromone was added to one set of cultures at 4.5 hours. A rapid reduction in growth rate was observed within two hours relative to the non-pheromone-treated control, consistent with the cell cycle arrest phenotype of the pheromone response (Bardwell, 2005) (Figure 2-1a). The rate of GFP expression was calculated for each population using the LINEST equation in Microsoft Excel (± standard error, r2 > 0.9). After the addition of pheromone the GFP expression rate (-782 ± 104 au/cell/hr) was not significantly different from the control strain (-934 ± 54 au/cell/hr) over the remainder of the cultivation time following initiation of culture growth arrest. This suggests that gene expression capacity remains as active during G1 phase growth arrest as it is in exponentially growing populations, thus demonstrating the potential for this phase to be useful for production.

26

Figure 2-1. Pheromone-response GFP expression, extracellular metabolites, and growth characteristics

A strain with constitutive expression of GFP (PSP03, a), and a strain with no GFP expression (PSP04 b-e) were grown for 15 hours in shake-flasks with (squares, blue lines) and without (triangles, black lines) 1 µM alpha pheromone added at 4.5 hours. GFP expression per cell (dashed lines), and population density (solid lines) were measured (a) along with extracellular glucose and organic acid concentrations (b-e). Relative carbon concentrations were calculated as described in (Stephanopoulos et al., 1998) after 15 hours of growth are shown f), where an S. cerevisiae carbon content of 24.6 g/C-mol with 7% ash (Stephanopoulos et al., 1998)was used along with a previously-determined conversion factor of 0.243 to convert OD660nm to gram dry cell weight. All measurements were carried out in biological triplicate with error bars representing ± 1 standard deviation.

27

In addition to transcriptional activity, another principle requirement for metabolic engineering applications is to have an active central carbon metabolic network. This can be investigated generally by examining biomass accumulation in parallel with carbon source consumption and concentrations of the predominant extracellular metabolites produced by S. cerevisiae. Parallel measurements of glucose uptake, along with biomass, acetate, ethanol, and glycerol production throughout 15 hours of shake-flask growth clearly demonstrated that populations treated with pheromone were at least as metabolically active as the non-treated control populations (Figure 2-1 b-h). As with the GFP expression strains, pheromone addition at 4.5 hours resulted in a characteristic decrease in growth rate (Figure 2-1 b), but surprisingly a very similar glucose consumption profile (Figure 2-1 c) with rates of 16.95 ± 2.18 mmol g-1 h-1 and 13.79 ± 1.88 mmol g-1 h-1 for pheromone treated and non-treated respectively (Table 2-4). While the rate of ethanol production was not significantly different in the pheromone treated populations (Table 2-4), glycerol and acetate production were markedly increased (Figure 2-1, Table 2-4).

When summarized with end-point metabolite concentration values converted to carbon moles (Figure 2-1 f) it is clear that the reduced carbon present in biomass is directed towards side- product formation in the form of glycerol, acetate, and CO2 (calculated as 296 and 273 carbon moles for pheromone treated and non-treated respectively). When carbon balance values were used to infer CO2 and O2 production and consumption rates, the respiratory quotients obtained (Table 2-4) strongly suggest that the pheromone response entails a shift towards a more respiratory metabolism compared to the fermentative metabolism of the control cultures. The current understanding of the shift between fermentative and respiratory metabolism in yeast is that a reduced glucose uptake rate results in de-repression of TCA cycle enzymes and a corresponding increase in TCA cycle flux, oxidative phosphorylation, and a decrease in fermentative flux towards ethanol (Dijken et al., 1993, Blank & Sauer, 2004, Heyland et al., 2009). The reduced ethanol production rate of pheromone treated populations is consistent with this understanding, but it is interesting that the specific glucose uptake rates are not significantly different between the groups. The most likely explanation is that the differences in respiratory activity were too subtle to be reflected significantly in the glucose uptake rates.

28

Table 2-4. Summary of external metabolite flux rates

Growth and extracellular product secretion rates are shown for cultures treated with alpha-pheromone (treated) and without (control). All rates were calculated using data points after the addition of pheromone at 4.5 hrs. Mean values of biological triplicates are presented with errors representing ± standard error. * denotes significant difference between control and pheromone treated groups (two sided students t-test with equal variance). µ (h-1) * Glucose Ethanol Glycerol * Acetate * Respiratory (mmol g-1 h-1) (mmol g-1 h-1) (mmol g-1 h-1) (mmol g-1 h-1) quotient Control 0.29 ± 0.03 13.79 ± 1.88 22.76 ± 2.97 0.96 ± 0.22 0.32 ± 0.06 4 Treated 0.17 ± 0.03 16.95 ± 2.18 19.05 ± 2.46 2.12 ± 0.25 0.87 ± 0.1 2

2.4.2 Transcriptome Analysis

Global changes in S. cerevisiae gene expression in response to pheromone treatment have previously been reported (Roberts et al., 2000). This study elegantly demonstrated the complexity of the pheromone response pathway and the degree to which its signalling components are related to other MAPK modules. However, the data analysis did not include transcriptional changes outside of the signalling and effector components of the pheromone response and related MAPK modules. To gain insight into other changes of relevance to metabolic engineering, the data were re-analysed after excluding any genes primarily involved in the pheromone response (GO terms ‘sexual reproduction’ and ‘conjugation’). A 99 % confidence interval and minimum 2-fold change were used as selection criteria to identify up-regulated (Table 2-5 ) and down-regulated (Table 2-6) genes.

Structural processes, including cell wall and cytoskeletal organization, were up-regulated in pheromone-treated populations. This likely relates to the characteristic shmoo cell morphology of the mating phenotype. Many up-regulated genes were assigned to categories associated with control of the cell cycle, mitosis, budding, and cytokinesis. These genes regulate the characteristic G1 phase cell-cycle arrest of the mating phenotype. The same categories also featured in the down-regulated gene GO terms (Table 2-6), highlighting the complexity of regulation required to arrest cell division.

29

The GO term ‘Transposition’ refers to the movement of DNA between non-homologous sites and includes many retrotransposon genes. It has previously been shown that Ty3 retrotransposons are up-regulated in mating populations (Kinsey & Sandmeyer, 1995), and it was interesting to see that genes more generally involved in transposition were significantly up-regulated (Table 2-5). Transposition during the pheromone response phenotype might provide a mechanism to increase genetic variation in the population prior to mating and, given that the process appears to be regulated by the host, could serve as an example of symbiotic retrotransposition.

The most notable down-regulated genes were involved in ribosomal RNA biogenesis and processing. This is consistent with the fact that much of the cellular resources of an exponentially growing population are directed towards ribosome synthesis (Warner et al., 2001). Many down-regulated genes were involved in DNA replication and repair, and chromosome segregation, again reflecting the arrested state of cell division. It is interesting to note that these processes are also typical of the starvation responses associated with stress induced stationary phases (Wu et al., 2004), but that the yeast which were used for the pheromone response transcriptomics were cultured in rich YPD media, and were not starving (Roberts et al., 2000). This observation highlights the unique nature of the pheromone mediated growth-arrest, and validates the idea of attempting to use it as a production phase where flux towards biomass is limited while nutrients are still abundant.

30

Table 2-5. Up-regulated genes and GO terms

GO term Gene(s) CYS3,CIT2,SLM5,IDP1,GCN2,ARO10,ARG5,6,SER3,TRP2,MET6,TRP5,STR3,GTF1,ADE3,ARG4,THR1,PUT2,LY cellular amino acid S1,GTT1,MET3,BNA2,MET14,SRY1,ECM4,GTT2,AAT2,PDC5,PUT1,CAR2,ARG7,ALD3,ALD2,GAD1,ADH2,IDH metabolic process 1,LEU4,FPR1,LAP3,CIT1,PLB3,ORT1,IDH2,HIS3,CPA1,CAR1 CDC19,TPS1,ARA1,MAL33,GLK1,CIT2,RBK1,DLD1,NTH1,TPS2,HKR1,GLC3,UBC8,PCM1,HXK1,PYC1,AMS1, carbohydrate metabolic GND2,GRE3,IMP2',GUT2,TDH1,INO1,FBP26,PGU1,UGP1,GFA1,PGM1,KTR2,PDC5,TSL1,PGM2,NDE1,MSS11,C process AT8,ZWF1,MNT4,GPD2,KRE6,GPH1,RHO1,GDB1 PRX1,ZTA1,MXR2,TRX3,YDL124W,NRG1,CAD1,DIG2,FRD1,PTC2,SPI1,ACT1,UFD1,CTT1,SLI1,VMR1,SOD2,S response to chemical LT2,GRE3,SCH9,DOT5,MET30,HYR1,YJL144W,SOD1,HAP1,ORM2,VPS36,ECM5,GAD1,CAT8,SIS1,NCE103,BO stimulus P3,MLF3,LAP3,ZWF1,YNR064C,SNF8 YAR009C,YAR010C,YBL005W-A,YBL005W-B,YBL100W-A,YBL100W-B,YBR012W-A,YBR012W- B,TEC1,YCL019W,YCL020W,YCL074W,YCL075W,YDR170W-A,YER138C,YER160C,YFL002W- A,YIL082W,YJL113W,YJR026W,YJR027W,YJR028W,YJR029W,YLR035C- transposition A,YML039W,YML040W,YML045W,YMR045C,YMR046C,YMR050C,YMR051C generation of precursor CDC19,ACS1,GLK1,DLD1,SDH4,GLC3,HXK1,LSC2,TDH1,SOD1,UGP1,PGM1,SDH1,PET10,PDC5,HAP1,ACO1, metabolites and energy MIC17,PGM2,NDE1,PAH1,IDH1,CIT1,IDH2,PET20,FUM1,GPH1,GDB1,QCR2 cell wall organization FLC2,RCR1,SED1,EXG2,SSD1,HKR1,PCM1,ACT1,SLT2,YPS6,SOD1,GFA1,MYO3,PIR3,UTH1,KTR2,YPS1,YPS3, or biogenesis YLR194C,CCW14,MYO5,DFG5,WSC2,KRE1,HPF1,FLC1,KRE6,RHO1 cytoskeleton AKL1,AIM3,CDC28,ARC40,SYP1,ABP1,PCL2,RDI1,CDC37,GIC2,ACT1,TWF1,LSB1,YSC84,GVP36,PAN1,BBC1,

31 organization ARP3,ENT3,VRP1,CRN1,DSK2,MSB3,SHE4,SCD5,PIN3,RHO1 CDC28,EHT1,GPI8,ISC1,GPI16,LSB6,OPI3,OSH6,YSR3,TGL4,YEH1,ICT1,ORM2,CSR1,PLB1,MOT3,PAH1,LAP2, lipid metabolic process INP52,ERG24,PLB3,GRE2,ERG10,GDE1,FAS2,PDH1 transcription from RNA polymerase II CLN3,TEC1,CDC28,NRG1,RPT3,CAD1,DIG2,BUR6,RPH1,ACT1,PTI1,SCH9,MET30,HAP1,VPS36,MOT3,MSS11, promoter CAT8,LAP3,SFL1,WTM1,SNF8,RLM1,CUP9,HAL1 nucleobase-containing small molecule NFS1,GPD1,STF1,GUD1,SWA2,APA2,PNC1,ADE3,GND2,CAB2,GUT2,BNA1,BNA2,UGP1,PGM1,PGM2,NDE1,A metabolic process DH2,YMR315W,YNL200C,ZWF1,MSB3,GPD2,NPT1,ALD4 cofactor metabolic ACS1,GPD1,COQ4,HEM1,PNC1,ADE3,LSC2,GND2,CAB2,GUT2,BNA1,BNA2,ISA1,HMX1,NDE1,ADH2,YMR315 process W,YNL200C,ZWF1,CIT1,GPD2,NPT1,ALD4 proteolysis involved in cellular protein UBX7,SAF1,STP22,MGR1,GGA1,RPT3,UBC8,CDC26,UFD1,MET30,RPN2,DAS1,CPS1,VPS36,UBX4,ADD37,DSK catabolic process 2,SIS1,ATG19,MCA1,SNF8,PEP4 FLC2,SUL1,PHO89,RHB1,ENA5,ENA1,PIC2,AVT6,MEP1,FMP43,YKE4,KCH1,GAP1,CSR1,PRM6,FRE7,ORT1,SI ion transport A1,TPO4,MCH5,FLC1 COS111,RDI1,CDC37,HKR1,PTC2,GPG1,HOS2,MDS3,PDE1,SLT2,KSP1,TPK1,RCN1,TFS1,WSC2,CMK2,WSC3,P signalling TP2,RLM1,RHO1 protein complex CYC3,ARC40,CDC37,ATP22,RPT3,PPM1,AFG3,MMS2,TWF1,LSB1,TRS65,RPN2,ARP3,CRN1,SPG5,DSK2,RCF2, biogenesis ATG34,PIN3 protein CLN3,AKL1,SLI15,CDC28,PCL2,KIN1,GCN2,PTC2,DBF2,SLT2,KSP1,SCH9,TPK1,PTK2,TDA1,CMK2,PTP2,HRK 32 phosphorylation 1 response to oxidative PRX1,ZTA1,MXR2,TRX3,YDL124W,ACT1,CTT1,SOD2,GRE3,SCH9,DOT5,HYR1,SOD1,HMX1,ECM5,GAD1,NC stress E103,ZWF1 ECM21,AKL1,SDS24,RVS167,ACT1,ROG3,RIM8,YSC84,GVP36,PAN1,ENT3,MYO3,OSH6,VRP1,MYO5,MSB3,S endocytosis CD5,RHO1 cellular respiration DLD1,SDH4,LSC2,SOD1,SDH1,PET10,HAP1,ACO1,MIC17,PAH1,IDH1,CIT1,IDH2,PET20,FUM1,QCR2 transmembrane transport FLC2,SSA3,YBR241C,SUL1,PHO89,ENA5,ENA1,PIC2,AVT6,PEX8,KCH1,MDL1,PRM6,FRE7,YOL162W regulation of cell cycle CLN3,CKS1,SLI15,CDC28,PCL2,CDC37,GCN2,GIC2,PTC2,HOS2,SPO12,FPR3,NIS1,NDJ1 regulation of organelle organization CDC28,ARC40,RDI1,GYP7,GIC2,ATP22,TWF1,LSB1,SPO12,ARP3,CRN1,NIS1,PIN3,RHO1 protein targeting PEP1,SSA3,ATG8,STP22,GGA1,PEX8,APL6,SLT2,VTH2,DID2,VPS36,ATG19,WTM1,SNF8 cytokinesis SLI15,SDS24,HKR1,ACT1,DBF2,SLT2,PAN1,MYO3,HYM1,VRP1,MYO5,CHS1,AFI1,AXL1 response to osmotic stress GPD1,ENA1,NRG1,CDC37,HKR1,ISC1,PTC2,SCH9,MYO3,MOT3,MYO5,PTP2,HAL1 mitochondrion organization NGR1,NFS1,SLM5,SED1,UPS3,ATP22,MRP20,ACT1,THI4,ARP3,ACO1,RCF2,MAM3 meiotic cell cycle CDC28,RAD51,MSH4,RIM8,IME4,HOS2,SPO11,SPO12,FPR3,MSC1,NDJ1,WTM1 mitotic cell cycle CLN3,CKS1,CDC28,PCL2,GIC2,ACT1,SWC4,SPO12,MET30,PTK2,NIS1,ULP1 pseudohyphal growth TEC1,NRG1,TMN2,PAM1,DIG2,IME4,KSP1,PGU1,MSS11,DFG5,DIA1,SFL1

33

Golgi vesicle transport ATG8,TRS23,GGA1,TRS65,APL6,VTH1,VTH2,ENT3,CSR1,VTI1,PEP12,GYP5 response to DNA damage stimulus CDC28,TDP1,GCN2,PTC2,ACT1,MMS2,SWC4,SHU1,IMP2',RAD33,DDR48 cell budding CDC28,ACT1,PAN1,MYO3,HYM1,TGL4,VRP1,MYO5,DFG5,AXL1,RHO1 cellular ion homeostasis NFS1,TMN2,DBF2,PTK2,SOD1,HMX1,PGM2,MAM3,HRK1,FIT3,ISU1 chromatin organization ACS1,CKS1,CDC28,RPH1,ACT1,CDC26,HOS2,SWC4,UTH1,FPR1 DNA recombination CDC28,RAD51,MSH4,YRF1-3,SHU1,YRF1-4,MSC1,YRF1-6,NDJ1,YRF1-7 vacuole organization CLN3,ATG8,RDI1,GYP7,ACT1,GVP36,PAH1,VTI1,PEP12,RHO1

Table 2-6. Down regulated genes and GO terms

GO term Gene(s) POP8,KRR1,RPL35B,NOP14,SAS10,FAP7,RPL35A,RRP8,BFR2,UTP4,FCF1,UTP6,NSA2,DBP3,ROK1,POP6,NOP7, UTP8,ENP2,LRP1,RPF1,UTP9,RIX1,FAF1,UTP25,MTR4,MRT4,EBP2,RPS21A,RPF2,DRS1,RPS31,NOP56,UTP13,T rRNA processing SR2,RPS1A,RPS1B,RRP5,RLP7,LSM7,IPI3,NOP12,PNO1,PUS7,RRP36,NOP58,RPS6A,NOG1,NOP53,MRD1 NUP170,POL30,MCD1,PCL9,FOB1,SWI5,SCC2,RSC3,YHP1,GIN4,CIN8,SWI4,SMC1,ALK1,RME1,CLB1,CLB6,C BF2,YHR127W,SWE1,CDC6,BUD2,TOF2,SPC34,MSA2,SIC1,CDC45,YCS4,HOF1,PDS5,CTF18,TOP2,SPC98,NRM mitotic cell cycle 1,CTR9,BUB3,SFG1,ACM1,CLB2 ribosomal small KRR1,NOP14,SAS10,FAP7,UTP4,FCF1,RPS17B,UTP6,ROK1,NOP7,UTP8,ENP2,UTP9,FAF1,UTP25,RPS21A,RPS3

34 subunit biogenesis 1,UTP13,TSR2,RPS1A,RPS17A,RPS1B,RRP5,LSM7,PNO1,RRP36,NOP58,RPS6A,MRD1 TRM7,RPL21A,RPS29B,RPL35B,RPL35A,RPS17B,RPL22B,RPL29,RPL26B,RPL42B,RPS22A,RPS21A,RPL8B,RPS 31,RPS28B,RPL38,RPS29A,RPS1A,RPS17A,RPS1B,RPL9B,RPL42A,RPS28A,RPS10A,RPL20B,RPL21B,RPS6A,RP cytoplasmic translation L1A,RPL11A response to DNA FUN30,RFA1,HTA2,POL30,MCD1,MSH6,SCC2,HTB1,HTA1,SMC1,RPB9,SNF6,RTT107,MPH1,SRS2,RFA3,RFC2, damage stimulus RAD27,RAD5,CDC45,TOS4,SMC6,OGG1,HOF1,PDS5,CTF18,POL1,CTR9 FUN30,RFA1,HTA2,POL30,MCD1,MSH6,SCC2,HTB1,HTA1,SMC1,RPB9,SNF6,RTT107,MPH1,SRS2,RFA3,RFC2, DNA repair RAD27,RAD5,CDC45,SMC6,OGG1,PDS5,CTF18,POL1,CTR9 mitochondrion MBA1,YDR115W,MRPS28,TIM9,TIM10,MRPL6,MRPL49,PAM16,RSM26,MRPL20,COX17,PAM18,MMR1,MRPL organization 15,MDM1,MRPS8,MRPS17,SUN4,TOM7,NAM9,MRPL50,PET494,MDM20,TOM6,MGE1,MRP2 transmembrane TAT1,SEC66,VCX1,YPQ2,TIM9,SIT1,FCY2,FCY21,FTR1,PMA1,FLC3,TIM10,PAM16,PHO90,PAM18,ZRT2,PHO8 transport 4,FET3,AQR1,TOM7,YPQ1,AUS1,TOM6,MGE1,SEC63 CST26,IFA38,PER1,SUR2,GPI11,LAG1,APQ12,GWT1,PHS1,ELO1,AUR1,LAC1,TGL1,YEH2,GAA1,SUR4,GAB1,P lipid metabolic process LB2,TCB2,CYB5,PDR16,HST3,ALE1,PIS1 ribosomal large SYO1,RPL35B,RPL35A,MAK21,RRP8,PUF6,NSA2,DBP3,NOP7,CIC1,RPF1,RIX1,MRT4,RPF2,DRS1,RRP5,RLP7, subunit biogenesis NOP15,IPI3,NOP12,BRX1,NOG1,NOP53,RPL11A transcription from RNA polymerase II RPB5,PCL9,SWI5,TAF10,HTB1,HTA1,RSC3,SRB7,YHP1,SRB4,SWI4,MIG1,RPB9,TAF6,RME1,SNF6,RPI1,MSA2, promoter MED11,SPT21,NRM1,CTR9,RPB10 TAT1,PHO88,PMP1,VCX1,PMP3,YPQ2,PMP2,FTR1,PMA1,FLC3,PHO90,OAC1,COX17,ZRT2,DIC1,PHO84,FET3, ion transport ZRC1,AQR1,PDR16,MID1,YPQ1,DIP5

35

MAK21,RPS17B,GIN4,LSM4,CBF2,RPF1,RIX1,MRT4,RPF2,DRS1,RPS31,RPS17A,CDC5,BNI5,IPI3,BRX1,NOC2, organelle assembly CAF20,RPL11A,MRD1 cytoskeleton GIN4,GEA2,CIN8,CLB1,CBF2,MYO1,YHR127W,SPC97,SWE1,GEA1,SPC34,CDC5,SPC24,NOP15,SPC98,BNI5,M organization DM20,BBP1,CLB2 RFA1,HEK2,RFC5,POL30,FOB1,RSC3,CLB6,RIX1,MPH1,RFA3,CDC6,RFC2,RAD27,CDC45,ORC1,CTF18,TOP2,P DNA replication OL1,IPI3 CHS2,SUP45,BUD3,MNN10,BUD9,MYO1,DSE2,SWE1,BUD19,BUD4,BUD2,RAX2,HOF1,NOP15,BNI5,BNI4,EGT cytokinesis 2,DSE4,AIM44 ECM1,NUP170,BCP1,KAP123,SRM1,UTP8,RPF1,NMD3,RIX1,APQ12,SRP40,RPS28B,NUP188,RPS28A,NOC2,RP nuclear transport S10A,NOG1,NOP53 protein complex NUP170,MBA1,GIN4,GIM4,VMA21,VOA1,CBF2,COX17,RAD5,CDC5,BNI5,PET494,CTR9,TOM6,PNO1,POC4,SR biogenesis P72,BBP1 chromosome NUP170,RFC5,POL30,MCD1,FOB1,SCC2,CIN8,SMC1,CBF2,DSN1,RFC2,TOF2,YCS4,SMC6,CDC5,PDS5,CTF18,S segregation PC24 carbohydrate metabolic process PSK1,PMT5,MNN10,ERD1,UTR2,MNN1,SNF6,OST1,PMT4,SEC59,SWP1,RKI1,KTR1,GNT1,MNN9,ALG5 NUP170,SEC66,TIM9,KAP123,WSC4,TIM10,SPL2,PAM16,SRP102,PAM18,NUP188,TOM7,TOM6,MGE1,SEC63,S protein targeting RP72 organelle fission NUP170,POL30,MCD1,FOB1,SWI5,SCC2,CIN8,SMC1,ALK1,BUD2,EBP2,TOF2,YCS4,PDS5,CTF18,BUB3 regulation of cell cycle NUP170,RSC3,GIN4,RME1,CLB1,CLB6,SWE1,RFC2,BUD2,SIC1,CDC5,CLN1,TOP2,BUB3,CLB2 chromatin organization FUN30,HTB2,HTA2,FOB1,TAF10,NGG1,HTB1,HTA1,RSC3,TAF6,SNF6,SAS2,TOP2,CTR9,HST3 36 ribosome assembly MAK21,RPS17B,RPF1,RIX1,MRT4,RPF2,DRS1,RPS31,RPS17A,IPI3,BRX1,NOC2,RPL11A,MRD1 mitochondrial MBA1,YDR115W,MRPS28,MRPL6,MRPL49,RSM26,MRPL20,MRPL15,MRPS8,MRPS17,NAM9,MRPL50,PET494, translation MRP2 regulation of organelle organization NUP170,MBA1,CLB1,SWE1,BUD2,SPC34,YPT7,CDC5,PET494,CTR9,BUB3,CAF20,CLB2 protein glycosylation PMT5,MNN10,ERD1,MNN1,OST1,PMT4,SEC59,SWP1,KTR1,GNT1,MNN9,ALG5 protein phosphorylation PSK1,GIN4,ALK1,CLB1,CLB6,SWE1,SIC1,CDC5,CLN1,CTR9,CLB2,NCE102 DNA recombination RFA1,FOB1,MPH1,IRC8,SRS2,RFA3,RAD27,CDC45,SMC6,CDC5,CTF18,TOP2 Golgi vesicle transport RER1,SHR3,GEA2,GET1,EMP24,YIP1,GEA1,SUR4,YPT7,GOT1,ERP4,RET3 mRNA processing POP8,LSM2,FIR1,LSM4,LSM5,SMX2,POP6,SMD1,RNA14,LSM7,CTR9 cellular ion homeostasis GDT1,PER1,VCX1,IZH1,SIT1,FTR1,EMP70,FET3,ZRC1,IZH4,YVC1 meiotic cell cycle RFA1,POL30,MSH6,RME1,CLB1,RFA3,SWE1,YCS4,CDC5,PDS5,TOP2 cell wall organization or biogenesis LRE1,MNN10,HLR1,UTR2,RPI1,PIR1,YEH2,GAS1,ZEO1,KTR1,SRL1 regulation of DNA metabolic process FOB1,SWI5,RSC3,RIX1,MPH1,SRS2,CDC6,TOP2,IPI3,CTR9 nucleobase-containing compound transport FCY2,SRM1,FLC3,VRG4,UTP8,APQ12,RPS28B,NUP188,RPS28A,RPS10A

37

Changes in the expression levels of metabolic genes are of direct relevance to the metabolic component of the pheromone response phenotype. Therefore in addition to analysing global transcriptional changes, central carbon-metabolism specific changes were analysed by mapping the expression levels of significantly changed (95 % confidence intervals) metabolic genes along with the reactions they encode (Figure 2-2). Significant increases in the transcript levels of a multitude of metabolic enzymes were observed, further suggesting that the mating phenotype has a distinct and active metabolism. In particular, expression levels of genes involved trehalose and glycerol synthesis, the TCA cycle, and the pentose phosphate pathway were significantly up-regulated (Figure 2-2). The high expression levels observed for multiple central carbon metabolic genes, along with the active secretion of metabolic side- products (Figure 2-1) act as a strong indication that central carbon metabolism is sufficiently active during the growth-arrest phenotype. There were also a number of interesting trends in regards to specific metabolic processes which are worth speculating on.

Trehalose acts as a major storage carbohydrate in yeast, and increased synthesis has been associated with exposure to thermal, osmotic, and ethanol stress (Pereira et al., 2001). Given that these experiments were carried out at 30 ˚C (Roberts et al., 2000) and considering that the hyper osmolarity glycerol (HOG) response to osmotic stress and the pheromone response are insulated from one another (O’Rourke & Herskowitz, 1998) it is possible that ethanol stress may be linked to the observed up-regulation of trehalose synthesis genes. It has previously been reported that trehalose synthesis is required to enable endocytosis at relatively low ethanol concentrations (Lucero et al., 2000). Endocytosis is a process where cells internalise their plasma membrane proteins from the extracellular environment, and is integral to the pheromone response in yeast where pheromone bound membrane receptor proteins are internalised (Marsh et al., 1991). Consistent with this idea, genes involved in endocytosis were highly up-regulated in response to pheromone (Table 2-5). It is therefore possible that the up-regulation of trehalose biosynthetic genes in response to pheromone evolved as a mechanism to protect cells from ethanol stress during pheromone bound receptor endocytosis. An alternative explanation is that there is actually a low level of osmotic stress associated with the pheromone response, which is responsible for the up- regulation of storage carbohydrate synthesis.

38

Up-regulation of glycerol synthesis genes (Figure 2-2) is completely consistent with the much higher levels of glycerol accumulation observed in pheromone-treated cells (Figure 2-1). It has been proposed that mating yeast cells require a precise osmotic balance prior to cell wall degradation and membrane fusion, and that this balance is achieved with the export of glycerol from the cell via the FPS1 transporter (Philips & Herskowitz, 1997). Recent work has further demonstrated the capacity for yeast responding to pheromone to excrete glycerol, demonstrating that the HOG pathway is actually partially activated by the pheromone response (Baltanas et al., 2013). The results presented here strongly support these findings, and support the role of trehalose synthesis as an osmoprotectant rather than solely to enable endocytosis under ethanol stress.

Respiratory metabolism requires a greater flux through the TCA cycle for the generation of reducing power to drive oxidative phosphorylation through the electron transport chain. The strong up-regulation of TCA cycle genes suggests that pheromone treatment results in a more highly respiratory metabolism. The extracellular metabolite accumulation rates and respiratory quotients independently support this finding (Table 2-4). The oxidative branch of the pentose phosphate pathway (PPP) is initiated by glucose-6-phosphate dehydrogenase (ZWF1) in an irreversible step (Nogae & Johnston, 1990). The PPP is responsible for producing NADPH, a critical source of reduction potential required by many anabolic pathways (Minard et al., 1998). The PPP also plays a major role in mitigating the effects of oxidative stress by supplying NADPH to glutathione- and thioredoxin-dependent enzymes (Slekar et al., 1996). The strong up-regulation of PPP genes (ZWF1, GND2, RPE1) in response to pheromone (Figure 2-2) could occur as a consequence of the increased rate of respiration initiated by pheromone (Table 2-4) and the subsequent increase in oxygen radicals. In concordance with this idea was the up-regulation of a multitude of genes involved in both oxidative stress and respiration (Table 2-5).

It is important to note that the cultures which were used for RNA extraction in the original transcriptomics study (Roberts et al., 2000) were carried out under different conditions than the cultures used here to calculate extracellular flux rates (rich YPD rather than minimal medium, Figure 2-1, Table 2-4). Therefore the trends in metabolic gene expression levels which correspond to the flux data should be considered as independent, but consistent observations.

39

Figure 2-2. Transcriptional changes in central carbon metabolism in response to mating pheromone

Transcript levels for genes encoding central metabolic enzymes are shown with up-regulation in red and down- regulation in green. Metabolites are connected by arrows representing reactions catalysed by enzymes (gene names in boxes). Single headed arrows represent one way reactions and double headed denote reversible reactions. The metabolic map and abbreviations were adapted from Oliveira et al. (2012).

40

2.4.3 Engineered pathway productivity during the pheromone response

To directly test the hypothesis that the pheromone response phenotype is suitable for metabolic engineering applications, heterologous compound titers from an engineered pathway in identical strains treated with and without pheromone were compared. The analysis of PHBA productivity under pheromone exposure here was used to assess the suitability of the mating phenotype as a production phase. PHBA is an important industrial chemical used in liquid crystal polymers (Krömer et al., 2012) and was also chosen to specifically validate the potential for its production using the pheromone response in preparation for the experiments described in Chapter 4. A minimally engineered para- hydroxybenzoic acid producing strain was constructed as in Figure 2-3 using manipulations previously shown to be effective at increasing shikimate pathway flux (Curran et al., 2013). The constitutive TEF1 promoter was used to drive heterologous (UBiC) and over-expressed (TKL1, ARO4) genes in the engineered pathway; this ensured that gene expression levels would be consistent between treated and untreated populations.

41

Figure 2-3. Engineered PHBA production pathway

Gene expression constructs (a) and the role of the expressed enzymes (in red) are shown for PHBA production from the shikimate pathway (b).

A PHBA producing strain (PHBA03) was grown with and without 1 µM alpha pheromone treatment at early log phase (OD660nm ~1, 4 hrs). As expected, growth rate slowed significantly after -pheromone treatment, indicating that the pheromone response had been initiated (Figure 2-4 a). After 48 hours, population densities were significantly different, with the control group at an OD660nm of 11.48 ± 0.28, and the pheromone treated group at 6.46 ± 0.88. Absolute PHBA concentrations were measured after 48 hours of fermentation (Figure 2-4 b), with no significant difference between the means (control = 0.29 ± 0.03 mM, alpha = 0.22 ± 0.02 mM). However, after accounting for the difference in population density, there was a significant difference in the specific PHBA production (p = 0.047) (Figure 2-4 c). The

42 control population-density specific PHBA concentration was 2.2 ± 0.5 µM/OD660nm, while the pheromone treated population-density specific PHBA was 5.2 ± 1.8 µM/OD660nm.

Figure 2-4. PHBA production in pheromone treated cultures

A strain minimally engineered to produce PHBA (strain PHBA03) was grown with (blue lines/bars) and without (black lines/bars) 1 µM alpha pheromone treatment at early exponential phase (4 hours). (a) Population density

(OD660nm) was measured for shake-flask growth with and without pheromone over 48 hours in biological triplicate. Absolute PHBA titer (b) and PHBA titer per OD660nm at 48 hrs (c) were measured for biological triplicate cultures. Using a two sided t-test with equal variance, ns= not significant, * = p < 0.05.

2.5 Conclusions

The S. cerevisiae pheromone response leads to a distinct and active metabolic phenotype which is suitable for the production of PHBA, and for metabolic engineering in general. The alternative scenarios of low or high metabolic productivity can be ruled out because the

43

PHBA titers of pheromone-arrested populations were similar to those of populations undergoing normal growth conditions. The key metabolic differences which result from pheromone treatment were identified as high glucose uptake rate (considering that cells were undergoing growth arrest), up-regulation of storage carbohydrate synthesis genes associated with osmolarity and oxidative stress responses, and increased respiratory activity. It is conceivable that the pheromone response phenotype could be used to increase flux towards metabolites of interest which are associated with the metabolic changes that underlie the phenotype.

Although it is possible to use the mating phenotype as a production phase, adding purified pheromone to a large scale fermentation is not practical. A more feasible situation would be for cells to produce and respond to their own pheromone by arresting growth and expressing pathway modifications. In theory such an autonomous regulatory circuit would enable the separation of growth from production. These concepts are explored in the next two chapters.

44

3 Engineered Quorum-Sensing Using Pheromone- Mediated Cell-to-Cell Communication in Saccharomyces cerevisiae

This chapter incorporates the following publication:

Williams, T. C., Nielsen, L. K. & Vickers, C. E. Engineered Quorum Sensing Using Pheromone-Mediated Cell-to-Cell Communication in Saccharomyces cerevisiae. ACS Synthetic Biology 2, 136-149, doi:10.10

3.1 Abstract

Population-density dependent control of gene expression, or quorum sensing, is widespread in nature and is used to coordinate complex population-wide phenotypes through space and time. We have engineered quorum sensing in S. cerevisiae by re-wiring the native pheromone communication system which is normally used by haploid cells to detect potential mating partners. In our system, populations consisting of only mating type ‘a’ cells produce and respond to extracellular α-type pheromone by arresting growth and expressing GFP in a population-density dependent manner. Positive-feedback quorum sensing dynamics were tuned by varying α-pheromone production levels using different versions of the pheromone responsive FUS1 promoter as well as different versions of pheromone genes (mfα1 or mfα2). In a second system, pheromone communication was rendered conditional upon the presence of aromatic amino acids in the growth medium by controlling α-pheromone expression with the aromatic amino acid responsive ARO9 promoter. In these circuits, pheromone communication and response could be fine-tuned according to aromatic amino acid type and concentration. The genetic control programs developed here are responsive to dynamic spatiotemporal and chemical cellular environments, resulting in up-regulation of gene expression. These programs could be used to control biochemical pathways for the production of fuels and chemicals which are toxic or place a heavy metabolic burden on cell growth.

45

3.2 Introduction

In nature, gene expression and cellular metabolism are subject to multi-level regulation which is responsive to both intra- and extra-cellular environmental cues throughout space and time. Synthetic biology aims to engineer modular genetic control programs which resemble those found in nature in terms of their capacity to respond to a dynamic extracellular environment (Purnick & Weiss, 2009, Holtz & Keasling, 2010, Nandagopal & Elowitz, 2011, Smolke & Silver, 2011). These goals are particularly relevant to the engineering of microorganisms for the production of fuels, chemicals, and pharmaceuticals. However, traditional metabolic engineering has relied upon the constitutive and static expression of enzymatic pathways to produce valuable end-products (Nevoigt, 2008) without regard for the dynamic spatiotemporal and chemical environment of the cell (Holtz & Keasling, 2010). Despite the fact that several methods are now emerging which afford dynamic, environmentally regulated control of gene expression (Holtz & Keasling, 2010, Khalil & Collins, 2010, Chang et al., 2012) there is still a greater need for ‘next generation’ control systems which act autonomously to regulate pathway flux towards desired products (Lu et al., 2009).

There are several advantages gained by engineering a greater level of dynamic control over gene expression. Engineered pathways often place a metabolic burden on their host cell, as they draw cellular resources away from normal growth-based metabolism (Lee et al., 2008). It has been demonstrated in silico that dynamic and biphasic use of engineered pathways can be advantageous (Gadkar et al., 2005, Anesiadis et al., 2008). This is because higher population densities are achieved before the growth-limiting metabolic burden of compound production is initiated, resulting in higher volumetric productivity. The merits of separating growth from production are also pronounced when pathway products are toxic to the host cell (Keasling, 2008, Lee et al., 2008). Both metabolic burden and product toxicity can be overcome by controlling gene expression at a desirable point during fermentation rather than using constitutive pathway expression. The delayed induction of such pathways is desirable, but is not currently feasible in an industrial context due to the cost of chemical inducers/repressors and lack of suitability of inducible/repressible promoters (Nevoigt et al., 2007, Keasling, 2010).

In nature, microorganisms sense population density by emitting inter-cellular signalling molecules which cause a population-wide cellular response once a critical signal 46 concentration is reached (‘quorum sensing’) (Williams et al., 2007). This enables a population to delay the induction of various genetic programs which are more effective if coordinated at the population level. Our approach to implementing dynamic control of gene expression is to engineer cell-cell chemical communication for the population-density dependent modulation of gene expression (quorum sensing). Engineered quorum sensing has been implemented previously for a variety of applications and has great potential to serve as a metabolic pathway control mechanism (Carter et al., 2012). S. cerevisiae is one of the most widely utilized industrial micro-organisms due to its depth of characterisation and genetic tractability. However, there is a distinct lack of synthetic gene circuits available for S. cerevisiae relative to E. coli and even mammalian cells (Blount et al., 2012).

The only other synthetic quorum sensing module available in yeast was engineered by using an Arabidopsis thaliana hormone as an intercellular signalling molecule as part of a positive feedback loop (Chen & Weiss, 2005). This was the first example of transferring a heterologous population-density dependent signalling system into S. cerevisiae. However, there are key aspects of this system arising from the interface between the heterologous A. thaliana components and endogenous host metabolism which make it inappropriate for application in a metabolic engineering scenario. The integration of the heterologous plant hormone membrane receptor with endogenous yeast metabolism required that the high osmolarity glycerol (HOG) response be constitutively repressed in order to avoid non- specific HOG induction, leading to cell death. This was achieved by expressing a repressor (PTP2p) of the HOG1 protein using a galactose inducible promoter. This requirement is undesirable in an industrial fermentation setting; furthermore, galactose is an unsuitable carbon source for industrial applications due to its high cost (Westfall et al., 2012) and the low specific growth rate of yeast on galactose. This approach to the implementation of quorum sensing in yeast represents a common theme in synthetic biology where heterologous components are imported into a desired host organism (Nandagopal & Elowitz, 2011), and highlights the challenges associated with interfacing these components with endogenous host metabolism.

An alternative to importing heterologous systems is to ‘rewire’ existing components to achieve the desired output. S. cerevisiae has a cell-to-cell communication system which is used by haploid cells to detect and respond to mating partners. Briefly, haploids of opposite

47 mating type (a or ) sense nearby potential mating partners by emitting mating type-specific small peptide pheromones (Bardwell, 2005). Upon pheromone binding to mating type- specific membrane receptors at a threshold concentration, intracellular mitogen activated protein kinase (MAPK) signalling effects the mating phenotype (Bardwell, 2005). During this process, cell cycle arrest occurs in the G1 phase, and mating-specific genes are up-regulated by approximately 100-fold. This coordinated modulation of gene expression results in the formation of membrane projections (called shmoos) to which nuclei localize prior to shmoo tip fusion between mating partners, followed by DNA exchange and the formation of diploids (Bardwell, 2005).

The characteristics of this native communication system are highly amenable to an integrated (non-heterologous) synthetic biology approach. The interface between extracellular signalling molecules and intracellular metabolism already exists in the form of the native membrane pheromone receptor, Ste2p, which rapidly enacts changes in transcription of mating specific genes via the MAPK phosphorylation cascade and the Ste12p transcription factor. This abrogates the need to further manipulate the regulatory network to facilitate the expression of genes of interest. The pheromone responsive MAPK network has proven to be highly plastic, with several synthetic gene networks engineered recently (Bhattacharyya et al., 2006, Ingolia & Murray, 2007, Bashor et al., 2008, Regot et al., 2011, Wei et al., 2012). It has previously been demonstrated that mating type a (MATa) haploid yeast is capable of producing α- pheromone (Whiteway et al., 1988), and that cells can respond to endogenously-produced pheromone when both types of pheromone membrane receptor protein (Ste2p and Ste3p) are expressed (Nakayama et al., 1987) and when both types of pheromone and receptor are expressed (Rivers & Sprague, 2003). However, these studies did not explore the potential of pheromone communication in yeast as a tuneable quorum sensing module.

In this work we have engineered the native pheromone cell-to-cell communication system in yeast to function as an autonomous quorum sensing module by enabling α-pheromone expression in α-pheromone sensitive MATa haploids. This quorum sensing module affords dynamic control of gene expression which may be useful for the separation of growth from toxic biochemical production.

48

3.3 Materials and methods

3.3.1 Media

Strains were grown in chemically defined liquid medium with 5 g/L ammonium sulfate and 2

% glucose (w/v). Chemically defined media contained: 3 g/L KH2PO4, 0.5 g/L MgSO4.7H2O,

4.5 mg/L ZnSO4.7H2O, 0.3 mg/L CoCl2.6H2O, 1 mg/L MnCl2.4H2O, 0.3 mg/L CuSO4.5H2O,

4.5 mg/L CaCl2.2H2O, 3 mg/L FeSO4.7H2O, 0.4 mg/L Na2MoO4.7H2O, 1 mg/L H3BO4, 0.1 mg/L KI, 15 mg/L EDTA, 50 µL biotin, 1 mg/L calcium panthothenate, 1mg/L nicotinic acid, 25 mg/L myo-inositol, 1 mg/L thiamine HCl, 1 mg/L pyridoxal HCl, 0.2 mg/L para- aminobenzoic acid. For solid medium, 15 g/L agar was added. During strain construction Sigma dropout solution (-Ura, -Trp, -His, -Leu) was used to complement appropriate auxotrophies in agar plates (same composition as chemically defined media above) while YPD or YPG supplemented with appropriate antibiotics was used during gene deletion procedures. E. coli DH5α strains were grown in LB medium with kanamycin.

3.3.2 Strains and Plasmids

DNA manipulation and propagation were carried out using standard techniques (Sambrook & Russell, 2001). All S. cerevisiae transformations were carried out using the lithium acetate method (Gietz & Schiestl, 2007). All S. cerevisiae genes used were amplified from CEN.PK2-1c genomic DNA extracted using a MoBio microbial DNA isolation kit. Gene deletions were performed with the reusable LoxP-KanMX-LoxP cassette as described previously (Güldener et al., 1996). All KanMX markers were removed via expression of Cre recombinase from pSH65; pSH65 was subsequently removed by growing strains without phleomycin for 48 h and patch plating colonies on YPD plates with and without phleomycin. For knock-out of BAR1, primers BAR1KOF and were used to amplify the LoxP-KanMX- LoxP cassette from the pUG6 plasmid (Table 3-2). Deletion was confirmed by PCR using primers BAR1DCF and BAR1DCR which flank the BAR1 ORF location.

Primers used in this study are shown in (Table 3-1) and plasmids are shown in Table 3-2. All plasmid constructs were transformed into and maintained in E. coli DH5α. All promoter regions were amplified using primers which create a 5 XhoI cut site and a 3 EcoRI cut site. Similarly, all coding regions were amplified with 5 EcoRI sites and 3 BamHI sites. Plasmid 49 pTCW001 was constructed by insertion of the yEGFP-CLN2PEST- ADH1t region of the P30419 plasmid (amplified using primers yEGFPDSF and yEGFPDSR) into pRS406 at the EcoRI and BamHI sites. A 700 bp region upstream from the FUS1 start codon was amplified from genomic S. cerevisiae DNA using primers pFUS1F and pFUS1R and inserted into the XhoI and EcoRI sites of pTCW001 to construct pTCW002 (pFUS1- yEGFP-CLN2PEST- ADH1t expression cassette). pTCW003 was constructed by inserting the coding region and 300 bp downstream of the stop codon of the mfα1 gene (amplified using mfα1F and mfα1R primers) into pRS413 at the EcoRI and BamHI sites. pTCW004 was made the same way except the insert was the mfα2 gene and 3 region (amplified using mfα2F and mfα2R primers). pTCW005 and pTCW006 were made via insertion of the pFUS1 promoter (amplified using pFUS1F and pFUS1R primers) 5 of the mfα1 and mfα2 genes in pTCW003 and pTCW004 respectively to create pFUS1- mfα1- mfα1t and pFUS1- mfα2- mfα2t expression cassettes. pTCW007 and pTCW008 were made by inserting the 500bp FUS1J2 promoter amplified from pNTI144 (Ingolia & Murray, 2007) (using pFUS1J2F and pFUS1J2R primers) 5’ of the mfα1 and mfα2 genes in pTCW003 and pTCW004 to make pFUS1J2- mfα1- mfα1t and pFUS1J2- mfα2- mfα2t expression cassettes. Similarly, pTCW009 and pTCW010 was made by inserting the ARO9 promoter region (amplified with pARO9F and pARO9R) 5 of the mfα1and mfα2 genes in pTCW003 and pTCW004 resulting in pARO9- mfα1- mfα1t and pARO9- mfα2- mfα2t expression cassettes. All plasmids were sequenced to check for PCR-mediated mutations introduced during the cloning steps.

Table 3-1. Primers used in this study

Primer 5 to 3 sequence name BAR1KOF CGCCTAAAATCATACCAAAATAAAAAGAGTGTCTAGAAGGGTCATATAC CAGCTGAAGCTTCGTACG BAR1KOR CTATATATTTGATATTTATATGCTATAAAGAAATTGTACTCCAGATTTCT CACTATAGGGAGACCGGCAG BAR1DCF AGAGATGCGTTGTCCCTGTT BAR1DCR ATGGTCAGAATGGGCGCTTG yEGFPDSF TATTTCGAATTCATGTCTAAAGGTGAAGAATTATTC yEGFPDSR TATTATGGATCCGATCTGCCGGTAGAGGTGTG mfα1F ATAATAGAATTCATGAGATTTCCTTCAATTTTTAC

50 mfα1R AATAATGGATCCGATTCGATTCACATTCATC mfα2F TACGAATTCATGAAATTCATTTCTACCTTTCTCAC mfα2R ATAATAGGATCCAGAGCTCCAACCATAGTGAAC pFUS1F TATTATCTCGAGATCAACAACAGGGTCAGCAG pFUS1R TATTATGAATTCTTTGATTTTCAGAAACTTGATGG pFUS1J2F GAGCTCCTCGAGCCCTCCTTCAATTTTTCTG pFUS1J2R ATCGATGAATTCTTTGATTTTCAGAAACTTGTTGG pARO9F TATTATCTCGAGTTGCCGCGTGGAGACATCTG pARO9R TATTATGAATTCTGAGTCGATGAGAGAGTGTAATTG

Table 3-2. Plasmids used in this study

Name Details Origin pRS404 TRP1 integrating vector (Sikorski & Hieter, 1989), Euroscarf pRS405 LEU2 integrating vector (Sikorski & Hieter, 1989), Euroscarf pRS406 URA3 integrating vector (Sikorski & Hieter, 1989), Euroscarf p30419 pFA6a-yEGFP3-CLN2PEST- (Van Driessche et al., 2005), natMX6 Euroscarf pTCW001 pRS406- yEGFP-CLN2PEST- This study ADH1t pTCW002 pRS406-pFUS1-yEGFP- This study CLN2PEST-ADH1t pRS413 URA3 low copy number vector (Sikorski & Hieter, 1989), Euroscarf pTCW003 pRS413-mfα1 This study pTCW004 pRS413-mfα2 This study pTCW005 pRS413-pFUS1-mfα1- mfα1t This study pTCW006 pRS413-pFUS1-mfα2- mfα2t This study pTCW007 pRS413-pFUS1J2-mfα1- mfα1t This study pTCW008 pRS413-pFUS1J2-mfα2- mfα2t This study pTCW009 pRS413-pARO9-mfα1- mfα1t This study pTCW010 pRS413-pARO9-mfα2- mfα2t This Study

51 pNTI144 PFUS1J2-STE4-TADH1 (Ingolia & Murray, 2007) pUG6 LoxP-KanMX-LoxP cassette (Güldener et al., 1996), Euroscarf pSH65 Galactose inducible cre- (Güldener et al., 1996), Euroscarf recombinase

Strains (Table 3-3) were constructed by transforming the plasmid components from

Table 3-2 successively into the relevant deletion strain and selecting on appropriate dropout and/or antibiotic containing media. Yeast integrating plasmid transformants were screened for correct genomic integration as described previously (Stansfield & Stark, 2007). Plasmids containing genes which would potentially result in pheromone production were always integrated into the relevant strain background as the last strain construction step prior to analysis. All quorum sensing strains and the control strain had all auxotrophies repaired so that a) only the genetic differences of interest were compared and b) only the amino acid of interest needed to be present in the growth media of aromatic amino acid responsive quorum sensing strains. Each biological replicate was derived from an individual transformant colony.

52

Table 3-3. Yeast strains used in this study

Name Genotype Notes Origin

CEN.PK2-1c MATa; ura3-52; trp1-289; leu2-3,112; his3Δ 1; MAL2-8C; SUC2 Haploid MATa lab strain Euroscarf

JRS01 CEN.PK2-1c: ura3::pRS406-pFUS1-yEGFP-CLN2PEST Expresses GFP in response to This study α-pheromone JRS02 CEN.PK2-1c : bar1Δ, ura3::pRS406-pFUS1-yEGFP-CLN2PEST Same as JRS01 except the α- This study pheromone protease BAR1 gene is deleted JRS03 CEN.PK2-1c: bar1Δ, ura3::pRS406-pFUS1-yEGFP-CLN2PEST, Base strain This study trp1::pRS404, leu2::pRS405 JRS05 JRS03 + pRS413-pFUS1-mfα1 Positive feedback quorum This study sensing with mfα1 gene, native FUS1 promoter JRS06 JRS03 + pRS413-pFUS1-mfα2 Positive feedback quorum This study sensing with mfα2 gene, native FUS1 promoter JRS07 JRS03 + pRS413-pFUS1J2-mfα1 Positive feedback quorum This study sensing with mfα1 gene, pFUS1J2 promoter JRS08 JRS03 + pRS413-pFUS1J2-mfα2 Positive feedback quorum This study sensing with mfα2 gene,

53

pFUS1J2 promoter

JRS09 JRS03 +pRS413 Control strain with This study auxotrophies matching ‘test’ strains JRS10 JRS03 + pRS413-pARO9-mfα1 Aromatic amino acid induced This study quorum sensing with mfα1 gene JRS11 JRS03 +pRS413-pARO9-mfα2 Aromatic amino acid induced This study quorum sensing with mfα2 gene

54

3.3.3 Sender-Receiver assays

Halo assays similar to those used to determine mating type and pheromone sensitivity (Chan & Otte, 1982) were carried out where MATa cells engineered to produce α-pheromone were used as ‘senders’ and non pheromone producing MATa cells were used as ‘receivers’. Pheromone sensitive bar1Δ cells (JRS09) were grown for three days in chemically defined liquid media and spread uniformly onto a chemically defined agar plate to act as α-pheromone receivers. After incubating the plate for half an hour at room temperature to allow for absorbance into the solid medium approximately 1 cm2 patches of sender strains were spread on top of the bar1Δ receiver lawn and the plate was incubated at 30 C for 48 h. Clearing in the receiver lawn around the sender strain patches relative to the control strain (non pheromone producing, JRS09) was indicative of α- pheromone production by the sender strains causing growth arrest in the receiver lawn.

For ‘conditioned media’ experiments, pheromone producing sender strains were grown to either low population density, non-autoinduced (OD660 ≤ 0.2 for pARO9 strains and ≤ 0.07 for all others,

GFP fluorescence per cell ≤ 1500 au) or high population density, autoinduced (OD660 ≥ 2, GFP ≥ 5000 au) conditions. Aromatic amino acid responsive sender strains were grown with 100 µg/mL tryptophan. Pheromone producing sender and non-pheromone producing receiver cells were removed from media by centrifugation at 13000g for 5 min. Receiver strains were then incubated at a starting OD660 of 1 for 3.5 hours in media conditioned by sender strain growth prior to measuring GFP fluorescence as below.

3.3.4 Flow cytometry

GFP fluorescence measurements were carried out on an Accuri C6 flow cytometer (BDbiosciences). Culture samples (0.1-2 mL as required) were sonicated for 30 s (to break cell clumps and buds) prior to the recording of 5,000 events per sample. GFP fluorescence intensity was measured using 488nm excitation with a 533 ±30 nm emission filter. Populations were gated between 5  105 and 5  106 by forward scatter to exclude debris (below 5  105) and cell clumps (above 5  106). Median GFP intensity was recorded for each population. Mean GFP intensity and standard deviation were calculated from individual sample medians and reported for each strain in biological triplicate. All raw data were normalised by subtracting the autofluorescence observed from the CEN.PK2-1c base strain, which has no GFP gene. Overlay histograms were produced using Flowing Software version 2.5.

55

3.3.5 Fermentation conditions

All time-course growth experiments were carried out in biological triplicate for each experimental strain or condition in baffled shake flasks covered in aluminium foil at 30 ˚C, 200 rpm, with media comprising 10% of the total shake flask volume. Positive feedback strains produced pheromone constitutively and were therefore grown from single transformant plate colonies directly into main cultures in order to avoid pre-culture conditions affecting quorum sensing behaviour. Individual colonies which did not display quorum sensing behaviour (about 1 in 6) were disregarded as background transformants not containing pheromone producing plasmid. Amino acid inducible and other non positive-feedback strains were able to be grown in pre-culture without any pheromone quorum sensing. Pre-cultures were grown to mid log phase (OD660 nm of 1-3) before being diluted into the final culture volume to a starting OD660 of 0.2 (or as stated otherwise). Population density was measured spectrophotometrically using absorbance at 660 nm wavelength.

For dose-response experiments, pre-cultures of indicated strains were grown as above before being divided into 200 µL aliquots in 96 well round bottom plates (Greiner 650161) and treated with the relevant pheromone (Genscript)/amino acid/conditioned-media in triplicate for each concentration. Aliquots were incubated at 30 C, 200 rpm, for 3.5 – 4 hours before being analysed on an Accuri C6 flow cytometer as specified above.

3.3.6 Statistical analysis

For each time point or dose response concentration, GFP fluorescence and OD660 are reported as the mean of biological triplicate cultures with error bars representing ± 1 standard deviation. Population growth rates were calculated using linear regression of log normalised average OD660 values from triplicate fermentations for each strain (all R2 ≥ 0.99). Hill coefficients for time-course GFP profiles were determined (similarly to previously described (Bhattacharyya et al., 2006)) using a four parameter variable slope dose response model in GraphPad Prism 6 using the equation: 푌 = 퐺퐹푃푚푖푛 + ((퐺퐹푃푚푎푥 − 퐺퐹푃푚푖푛) ÷ (1 + (10ˆ log 퐺퐹푃50 − 푋) × 푛퐻)) where Y = GFP response, X = Ln OD660, GFPmin = the non-autoinduced GFP per cell, GFPmax = fully autoinduced GFP, GFP50 = the half-maximal GFP, and nH = Hill coefficient. GFPmax was constrained to the maximum value observed for each strain in order to allow accurate fitting of the Hill equation. Hill coefficients were reported with ± standard error values generated by the curve fitting algorithm. Significant differences between conditioned media treated ‘receiver’ GFP-reporter levels were

56 assessed in biological triplicate for each ‘sender’ strain using a two-tailed two-sample t-test with unequal variance in Microsoft Excel.

57

3.4 Results and Discussion

3.4.1 Circuit topology

Two pheromone-mediated communication circuits with fundamentally different topologies were constructed in S. cerevisiae. The first circuit is defined by a positive feedback loop where pheromone expression is up-regulated in the presence of pheromone. Positive feedback loops are widely dispersed in nature and are particularly prevalent in implementing cellular decision making where switch-like changes in gene expression are required (Shah & Sarkar, 2011). All cells are able to respond to changes in their extracellular environment by altering gene expression, making such systems highly amenable for the engineering of dynamic control of gene expression. The second circuit we constructed therefore controls pheromone production in response to the aromatic amino acid content of the extracellular environment and has no positive feedback component under the growth conditions which were used. All circuits were constructed in the haploid MATa CEN.PK2- 1c strain background which is unable to switch mating type due to mutation at the HO locus (van Dijken et al., 2000).

The first circuit is a positive feedback loop controlled by the pheromone-responsive FUS1 promoter (Figure 3-1 a). Fus1p is a membrane protein involved in cell fusion during mating (Trueheart et al., 1987); the FUS1 promoter, pFUS1, is highly induced by -pheromone (Hagen et al., 1991). pFUS1 was linked to α-pheromone production to construct a positive feedback loop. Two different versions of the circuit were constructed by linking pFUS1 to two different mating factor  (mf) genes. The mfα1 gene encodes a precursor peptide with four repeating units of the 13 amino acid α-pheromone peptide, while the mfα2 gene encodes a precursor peptide with only two α-pheromone units (Singh et al., 1983). Binding of pheromone to the transmembrane G-protein coupled receptor Ste2p triggers a mitogen-activated protein kinase (MAPK) phosphorylation signal cascade which results in the de-repression of the Ste12p transcription factor (Kemp & Sprague, 2003). Ste12p activates the Far1p cyclin-dependent kinase, which is responsible for cell cycle arrest in the G1 phase and is involved in polarized growth and mating projection formation (Bardwell, 2005). Because of this cell cycle arrest phenotype, we initially planned to delete FAR1. However, FAR1 deletion mutants did not exhibit a sustained quorum sensing response; this is consistent with previous findings, which have shown that Far1p is essential for maintaining the pheromone response phenotype (Kemp & Sprague, 2003). The mating phenotype is controlled by the coordinated activity of Ste12p (which up-regulates approximately 200 genes) and FAR1 (which represses a similar number of

58 genes that are not active in the G1 phase of the cell cycle) (Roberts et al., 2000). Wild type yeast strains have a secreted -pheromone protease gene (BAR1) (Sprague & Herskowitz, 1981); deletion of this gene is required to avoid signal degradation (Chasse et al., 2006). The circuits were therefore constructed in a bar1 genetic background. A destabilized green fluorescent protein (GFP) gene codon-optimised for yeast expression (yEGFP) (Mateus & Avery, 2000) under the control of the FUS1 promoter was integrated into the genome as a reporter of pFUS1-driven expression. The destabilized yEGFP has a half-life of approximately 30 minutes and is consistently unstable throughout the cell-cycle, making it suitable for the detection of dynamic changes in gene expression.

59

Figure 3-1. Pheromone communication circuit design

(a) Positive feedback quorum sensing. Pheromone production is auto-induced with positive feedback via the MAPK signal cascade in a population-density dependent manner. The FUS1 promoter (pFUS1) controls expression of two or

60 four α-pheromone peptides using either the mfα2 (2 peptides) or the mfα1 (4 peptides) gene. α-pheromone secreted outside the cell binds to the Ste2 membrane receptor which activates an intracellular mitogen activated protein kinase (MAPK) phosphorylation cascade. The Ste12 transcription factor is de-repressed upon MAPK activation and promotes transcription from the pheromone responsive FUS1 promoter (pFUS1). The pheromone response includes FAR1 mediated cell cycle arrest (Chang & Herskowitz, 1990), and expression of a destabilised green fluorescent protein. (b) Aromatic amino acid responsive quorum sensing. ARO9 is up-regulated in the presence of aromatic amino acids (Iraqui et al., 1999). When α-pheromone expression is controlled by the ARO9 promoter (pARO9), induction of pheromone communication can be controlled by aromatic amino acid concentrations in the growth media. As in the positive feedback quorum sensing circuit, the pheromone response is tracked via pFUS1-GFP expression in response to MAPK activation. (c) Pheromone quorum sensing concept. (i) Pheromone secreted by cells at a low population is not concentrated enough to induce GFP expression. (ii) As the population grows in a confined space (shake-flask) the extracellular pheromone concentration increases, causing some of the population to switch on GFP-reporter expression (represented by green yeast cells). (iii) At a high population density, pheromone concentration is high enough to induce GFP-reporter expression in most members of the population.

In the second circuit (Figure 3-1 b), pheromone expression is controlled by the ARO9 promoter (Iraqui et al., 1998) which is up-regulated when aromatic amino acids are present in the growth medium (Iraqui et al., 1999). By controlling α-pheromone production with the ARO9 promoter, the pheromone response can be tuned according to the aromatic amino acid concentration. This circuit renders quorum sensing behaviour conditional upon the presence and concentration of aromatic amino acids, meaning that quorum sensing behaviour can be activated when desired. Similar to the positive feedback pFUS1 system (Figure 3-1 a), the circuit could be tuned by expressing α- pheromone using either the mfα1 or the mfα2 gene, and MAPK activated yEGFP was used as a reporter. This circuit was therefore designed to respond to both extracellular aromatic amino acids and population density over time. It should be noted that the ARO9 promoter is involved in the native yeast quorum-sensing response through positive feedback production of tryptophol as a communication molecule (Chen & Fink, 2006). Tryptophol is an aromatic alcohol which is produced downstream of the aromatic amino acid-degrading ARO9 gene. In late stationary phase, tryptophol accumulates extracellularly as a communication molecule and up-regulates transcription from the ARO9 promoter via the Aro80p transcription factor. This regulation is only observed under nitrogen starvation and very high population density conditions (neither of which were encountered in this study).

These circuits were designed so that at low population density, α-pheromone concentrations are insufficient to trigger a strong response via the intracellular MAPK signalling cascade (Figure 3-1 c, panel i). At a high population density, α-pheromone is more concentrated, and coordinates

61 population-wide MAPK activation and expression of genes of interest (Figure 3-1 c, panel iii). Circuits producing pheromone via mf1 are expected to elicit a stronger/faster response than circuits producing pheromone via mf2. An alternate version of the positive feedback circuit was constructed in which the mfα1 and mfα2 genes are regulated by an engineered version of the FUS1 promoter, pFUS1J2 (Ingolia & Murray, 2007). pFUS1J2 has an almost undetectable level of basal expression, and a higher level of induced expression relative to the native FUS1 promoter. Circuits controlled via pFUS1J2 were expected to require a higher population density than pFUS1 to initiate the positive feedback expression of α-pheromone.

3.4.2 Secreted α-pheromone elicits a population-density-dependent cell-to-cell communication response

In order for the -pheromone to behave as a population density reporting molecule, it is necessary for cells to both produce the pheromone and to export it into the extracellular medium. To determine whether -pheromone accumulates in the extracellular medium at levels significant enough to effect a biological response, sender-receiver growth arrest assays were performed (similar to the commonly used halo assay (Chan & Otte, 1982)). When cells engineered to produce -pheromone were grown on solid media which had been spread with an -pheromone sensitive but non-producing MATa strain (JRS09; see Table 3-3), growth inhibition was observed in the area surrounding test strains but not the control strain (Figure 3-2 a, b). This indicates that the ‘sender’ test strains are indeed producing -pheromone, and that the -pheromone is diffusing into the medium and causing the ‘receiver’ cells to arrest growth. One might question why, given that ‘sender’ strains are also sensitive to α-pheromone, they do not arrest growth under these conditions. Previous results have shown that α-pheromone producing MATa cells do not undergo a population- wide autoinduced growth-arrest unless the SST2 gene (responsible for desensitisation to pheromone (Dohlman et al., 1996)) is disrupted (Whiteway et al., 1988). The fact that the ‘sender’ strains secrete α-pheromone and still grow is consistent with these findings.

62

Figure 3-2. Sender-receiver pheromone communication.

Pheromone producing pFUS1-mfα2, pFUS1-mfα1, pFUS1J2-mfα2, pFUS1J2-mfα1, pARO9-mfα2, pARO9-mfα1, and a vector only control strain were grown on a lawn of pheromone sensitive MATa cells (JRS09) without tryptophan (a) or with 50 µg/mL tryptophan (b). Zones of growth inhibition around test strains indicate sender-receiver communication between pheromone producing strains and the pheromone sensitive lawn which arrests growth upon receiving the signal. The assay is only semi-quantitative, since test strains were derived from single colonies of different 63 sizes on solid media. Figure is representative of replicate experiments using different colonies of the same strain. (c)

Strains JRS01 (BAR1) and JRS02 (bar1Δ) were grown to an OD660 of 1 prior to treatment with the indicated concentration of synthetic α-pheromone. Populations were incubated for 3.5 hours prior to GFP fluorescence measurement. (d) A non pheromone-producing ‘receiver’ strain (JRS09) engineered to produce GFP in response to pheromone (via the FUS1 promoter) was treated with media conditioned by pheromone producing ‘sender’ strains which had been grown to low population density, un-autoinduced (OD660 ≤ 0.2 for pARO9 strains and ≤ 0.07 for all others, GFP fluorescence per cell ≤ 1500 au) or high population density, autoinduced (OD660 ≥ 2, GFP ≥5000 au) conditions which were defined according to strain behaviour observed in quorum sensing experiments (Figure 3-3, Figure 3-6 b, c). Aromatic amino acid responsive sender strains were grown with 100 µg/mL tryptophan. Receiver strains were incubated with conditioned media at a starting OD660 of 1 for 3.5 hours prior to measuring GFP. GFP fluorescence in the control strain reflects basal activity of the FUS1 promoter. Differences in GFP fluorescence levels of conditioned media treated ‘receiver’ populations were analysed in biological triplicate within each ‘sender’ strain type (p ≤ 0.05, two-tailed two-sample t-tests with unequal variance).

The size of the zone of clearing around the positive feedback strains (driven by pFUS1 or pFUS1J2) was similar regardless of the presence or absence of aromatic amino acids (Figure 3-2 a, b), consistent with circuit topology (Figure 3-1 a). There was no obvious difference in the size of the clearing zone between pFUS1 and pFUS1J2, or between the two different -pheromone genes (encoding 2 or 4 peptide units); the semi-quantitative assay is clearly not sensitive enough to distinguish between responses driven by the two promoters or the two genes. When strains carrying the aromatic amino acid responsive circuits (driven by pARO9) were grown on solid media without tryptophan (Figure 3-2 a), very small zones of growth inhibition were observed in the receiver growth lawn. In contrast, when the same strains were grown with tryptophan, strong growth inhibition was observed in the receiver strain (Figure 3-2 b). These data are consistent with previous findings that demonstrated very low levels of gene expression from the ARO9 promoter in the absence of aromatic amino acids (Iraqui et al., 1999). Expression of α-pheromone from the ARO9 promoter in the presence of tryptophan is not subject to the same feedback regulation as the FUS1 promoter, which may explain the slightly larger zones of growth inhibition in the ‘receiver’ lawn in these strains relative to the positive feedback strains (Figure 3-2 b).

A key requirement of the engineered circuits is that they respond in a dose-dependent fashion to the appropriate inducer. This was examined using a pheromone-sensitive receiver strain which has been engineered to produce GFP in response to pheromone by placing the GFP gene under the control of the FUS1 promoter. Previously, it has been shown that the FUS1 promoter responds to α- pheromone in a dose-dependent manner (Poritz et al., 2001). This was confirmed by treating the engineered receiver strain (which does not produce α-pheromone) with varying concentrations of 64 synthetic α-pheromone (Figure 3-2 c). We also confirmed that deletion of the BAR1 gene in MATa haploids makes responding populations significantly more sensitive to α-pheromone, consistent with previous results (Chasse et al., 2006). Peak GFP responses were observed about 2 orders of magnitude apart: the GFP response peaked at ~1 M α-pheromone in the BAR1 strain, whereas it peaked at ~ 0.01 M in the bar1 strain.

To confirm that the observed response elicited by the sender strains is due to a mobile molecule released into the medium, and that the response is proportional to population density, the pheromone-sensitive ‘receiver’ strains were treated with media which had been conditioned by pheromone-producing strains. According to the circuit design, the engineered strains are in a non- autoinduced state at low population density, and in a maximally autoinduced state at high population density. Media was conditioned by growing pheromone-producing cells to either low population density or high population density; the cells were then removed and the conditioned media used to treat the pheromone-sensitive receiver strain which has been engineered to produce GFP in response to pheromone (driven by the FUS1 promoter; JRS09 – see Table 3-3). When media conditioned with pheromone-producing strains growing at low population density was used to treat the receiver strain, GFP levels were slightly greater than the basal level from the FUS1 promoter (Table 3-1 d). This is consistent with low levels of pheromone being produced and released into the medium under these conditions. This effect was more pronounced in the aromatic amino acid responsive strains due to the presence of tryptophan in the media highly up-regulating pARO9-mediated pheromone expression even at a low population density. When media conditioned by pheromone-producing strains at high population density was used to treat the receiver strain, significant increases in GFP were observed, consistent with much higher levels of pheromone being produced under these conditions (p ≤ 0.05, two-tailed two sample t-tests with unequal variance). This demonstrated that strains carrying mfα1 or mfα2 genes as shown in Figure 3-1 produce functional α-pheromone as an extracellular communication molecule, which increases in concentration with increasing population density. This is the hallmark feature of an intercellular quorum sensing molecule (Waters & Bassler, 2005).

3.4.3 Positive feedback quorum sensing

We have engineered mating type ‘a’ strains which produce and respond to exported α-pheromone by expressing GFP and arresting growth in a population-density dependent manner. A positive- feedback loop was created by controlling α-pheromone expression with a promoter (FUS1) which

65 has a low basal activity, and is up-regulated when cells respond to pheromone (Figure 3-1 a). We hypothesised that these strains would dynamically and autonomously control gene expression by quorum sensing with the α-pheromone.

Strains engineered with positive-feedback circuits (Figure 3-1 a) were grown in liquid media from single colonies in order to avoid quorum sensing behaviour occurring in pre-cultures affecting experimental observations. As expected, GFP reporter expression increased in parallel with increasing population density (Figure 3-3 a). The growth curves of the mfα2- and mfα1-expressing strains were identical; however, the GFP expression activity profile of the mfα2 strain (JRS06) was slightly damped relative to that of the mfα1 strain (JRS05), with GFP accumulation starting ~2 h later and reaching maximal expression at higher population density (Figure 3-3 a; Table 3-4). This is consistent with expected behaviour due to the number of -pheromone peptides encoded by each gene. The population densities of the pheromone-producing mfα1 and mfα2 strains reached a -1 -1 modest OD660 of ~1.3 and strains had exponential growth rates of 0.22 h and 0.20 h respectively, -1 while the control strain reached a final OD660 of 6.4 and had an exponential growth rate of 0.30 h . These differences in population growth dynamics can be explained by the fact that cells responding to pheromone arrest growth in the G1 phase of the cell cycle due to the action of the FAR1 protein (Chang & Herskowitz, 1990). A whole population growth arrest was not observed in the experimental timeframes of this study, although the growth arrest of population subsets can be inferred by comparing the population growth curves of pheromone producing strains with those of the control strain. Whole population growth arrest has been observed in α-pheromone autoinduced populations which had the SST2 gene (which is central to the recovery from α-pheromone growth arrest (Chasse et al., 2006)) deleted (Whiteway et al., 1988). It is therefore likely that the continued growth of quorum sensing populations showing maximum GFP-reporter expression was due to the recovery from G1 arrest of a subset of the population via SST2 dependent mechanisms. Deletion of the SST2 gene was not practical in the design of the quorum sensing circuits created here due to the higher basal and non-specific activation of the FUS1 promoter in SST2 deleted strains (Siekhaus & Drubin, 2003).

66

Figure 3-3. Positive feedback quorum sensing

(a) Pheromone-producing strains pFUS1-mfα1 (JRS05) and pFUS1-mfα2 (JRS06), as well as non pheromone- producing control strain JRS09 were assayed for quorum sensing behaviour by following population density (OD660) and pheromone responsive gene expression (pFUS1-GFP) over time. Single colonies from transformation plates were used to inoculate liquid media without pre-culturing. (b) Strains pFUS1J2-mfα1 (JRS07) and pFUS1J2-mfα2 (JRS08) were grown as in (a), with behaviour being observed from an OD660 of ~0.07 after an initial growth phase of 14 hours where only basal GFP levels were observed. Markers and error bars represent mean and standard deviation of biological triplicates.

The native FUS1 promoter circuits initiated quorum sensing at very low population densities (Figure 3-3 a). For industrial application, it is desirable to reach higher population density so that sufficient biomass is available for production of biochemicals before pathway expression is triggered. Alternate positive-feedback quorum sensing strains were engineered by using the pFUS1J2 promoter (Ingolia & Murray, 2007) in place of the native FUS1promoter for the control of α-pheromone expression. The FUS1J2 promoter has nearly undetectable basal levels of expression, and higher levels of pheromone induced expression relative to the native FUS1 promoter (Ingolia & Murray, 2007). It was therefore expected that pFUS1J2 strains would initiate quorum sensing behaviour at higher population densities than the strains engineered with the native FUS1 promoter. As for the pFUS1 circuit, reporter GFP expression (from the native FUS1 promoter) increased with increasing population density, and the growth rate was reduced relative to the control strain (Figure 3-3 b). As expected, the FUS1J2 promoter resulted in maximal autoinduced GFP reporter expression being delayed until higher population densities were reached relative to the native FUS1 promoter carrying strains (Table 3-4). Similarly to the pFUS1 promoter circuits, the pFUS1J2-mfα2 circuit initiated quorum sensing at a higher population density than the pFUS1J2-mfα1 circuit (Table 3-4). In addition to tuning the quorum sensing ‘threshold’ towards a higher population density response, the use of the FUS1J2 promoter also resulted in a much more ‘switch-like’ 67 transition between non-autoinduced and autoinduced GFP-reporter expression. The Hill coefficient, which is commonly used to describe the response dynamics of engineered regulatory systems (Shah & Sarkar, 2011), can be used to define response cooperativity of GFP activity here. The native FUS1 promoter resulted in approximately linear GFP expression dynamics with Hill coefficients ˂1, while the engineered FUS1J2 promoter resulted in a much steeper, sigmoidal increase from uninduced to induced GFP expression (Hill coefficients ˃1) (Table 3-4). The FUS1J2 promoter was originally designed to convert the graded MAPK transcriptional response into a switch-like response (Ingolia & Murray, 2007) and has been used here successfully to the same effect.

Table 3-4. Quorum sensing circuit properties

Quorum sensing initiation was defined as the population density at which GFP per cell first exceeds 1800 au. Quorum sensing saturation point was defined as the population density at which GFP per cell plateaus or decreases (above 4000 au). Error bars represent ± 1 standard deviation of biological triplicates. Hill coefficients were determined using the four parameter variable slope dose response model in GraphPad Prism 6, with average log normalised OD660 values plotted against GFP fluorescence readings in triplicate for each strain. Errors for Hill coefficients are ± standard error. Response time is defined as the time between quorum sensing initiation and quorum sensing saturation. Dynamic range was determined by dividing the mean maximum GFP by the mean minimum GFP from triplicate experiments for a given strain. “Induced” refers to the addition of 100 µg/mL of tryptophan to the media at the population density indicated.

68

Quorum sensing initiation Quorum sensing Hill coefficient Response time (h) Dynamic

Circuit/condition (OD660nm ) saturation (OD660nm) range pFUS1-mfα1 0.009 ± 0.001 0.550 ± 0.070 0.63 ± 0.09 24 7.6 pFUS1-mfα2 0.016 ± 0.001 0.800 ± 0.090 0.97 ± 0.12 22 6.4 pFUS1J2-mfα1 0.497 ± 0.036 1.577 ± 0.057 3.90 ± 0.60 6 7.0 pFUS1J2-mfα2 0.768 ± 0.184 3.817 ± 0.332 2.40 ± 0.27 8 4.4

pARO9-α1 Induced at OD660nm of 0.183 ± 0.007 1.967 ± 0.210 0.56 ± 0.12 8 4.9 0.183 pARO9-α2 Induced at OD660nm 0.185 ± 0.003 1.847 ± 0.056 0.50 ± 0.10 8 5.2 of 0.185 pARO9-α2 Induced at OD660nm 0.021 ± 0.000 2.497 ± 0.150 0.59 ± 0.12 16 4.0 of 0.021 pARO9-α2 Induced at OD660nm 1.400 ± 0.061 4.017 ± 0.142 3.12 ± 0.79 4 5.0 of 1.400

69

A key feature of quorum sensing behaviour is the requirement for a high population density to activate gene expression. However, cells are able to exhibit intermediate levels of FUS1 mediated expression in response to intermediate levels of pheromone (Poritz et al., 2001) (Figure 3-3). Therefore, if a population initiates quorum sensing behaviour at a relatively low population density, with pheromone concentration not high enough to elicit a maximal pFUS1-GFP response, then the population will have a gradual increase in quorum sensing regulated gene expression. Another important feature is that a subset of cells in a population responding to pheromone will arrest cell division in the G1 phase, thereby slowing the transition from low to high population density, contributing to the graded nature of the response. Alternatively, if the initiation of pheromone communication is delayed until a higher population density then the effective pheromone concentration will be high enough to elicit a rapid population-wide GFP response.

These important differences in circuit output can be thought of as overlaid positive (α- pheromone expression) and negative (α-pheromone diffusion at a low population density, and pheromone induced growth arrest) feedback and are graphically represented in Figure 3-4. This concept explains the differences in GFP-reporter expression observed between the native FUS1 and FUS1J2 promoter strains (Figure 3-3). The higher level of basal expression from the native FUS1 promoter resulted in initiation of positive feedback pheromone expression at a lower population density relative to the FUS1J2 promoter, so that quorum sensing behaviour was self limiting and the transition from low to high autoinduced GFP- reporter expression was extremely gradual (Figure 3-3 a, Table 3-4). This explains the negative cooperativity observed for GFP expression dynamics of the native FUS1 promoter positive feedback strains (Hill coefficients of ˂1; Table 3-4). The more ‘switch-like’ increase in GFP expression observed with the FUS1J2 promoter (Figure 3-3 b, Table 3-4) is consistent with the fact that the level of basal expression is much lower than that seen with the native FUS1 promoter and the level of induced expression is higher (Ingolia & Murray, 2007). This allowed a higher population density to be reached before the positive feedback expression of α-pheromone was initiated, meaning that maximal GFP-reporter expression could be reached without quorum sensing behaviour being significantly limited by α-pheromone diffusion and the growth-arrest response. The result is positive cooperativity (Hill coefficients of >1; Table 3-4). The interaction of engineered circuit activation with host organism physiology and 70 growth dynamics has previously been noted for yielding non-intuitive circuit outputs (Tan et al., 2009) and is evident here also.

Figure 3-4. Model of graded signalling as a consequence of overlayed positive and negative feedback

1. As populations transition from relatively low to high population density, α-pheromone concentration increases. 2. Increased pheromone concentration induces the expression of the α-pheromone gene from the FUS1 or FUS1J2 promoter (Figure 3-1 a) as part of a positive feedback loop. 3. Cells responding to pheromone arrest growth in the G1 phase of the cell cycle. 4. As a proportion of a pheromone producing population undergoes cell cycle arrest, the rate of population growth decreases, delaying the attainment of high population density, and therefore a higher pheromone concentration. The output of this model is either a graded, linear increase in GFP reporter expression in the case of the native FUS1 promoter (5a) (Figure 3-3 a, Table 3-4), or a switch-like sigmoidal increase with the FUS1J2 promoter (5b) (Figure 3-3 b, Table 3-4).

This distinction is dependent on the population density at which positive feedback pheromone expression is initiated. The native FUS1 promoter has a significantly higher level of basal expression compared to the FUS1J2 promoter (Ingolia & Murray, 2007), meaning that expression of α-pheromone is noisier at a low population density, initiating positive feedback, α-pheromone diffusion, and growth-arrest before population density and pheromone concentration are sufficient to elicit maximal GFP-reporter expression. With a lower level of basal expression the FUS1J2 promoter requires a higher population density (and pheromone concentration) to initiate positive feedback pheromone expression, meaning that the growth

71 arrest response and pheromone diffusion do not limit the attainment of a high population density and maximal GFP-reporter response.

Positive feedback is often associated with bimodality in gene expression, where instead of a normal distribution of expression levels amongst cells in a population, there are two separate expression profiles (Mitrophanov & Groisman, 2008). Although the circuits engineered here were defined by the positive feedback regulation of α-pheromone expression, and exhibited ‘switch-like’ responses in the case of the FUS1J2 promoter, bimodality in expression of the GFP reporter was never observed (GFP populations always showed unimodal, approximately normal distributions) (Figure 3-5). These results are consistent with previous studies which demonstrated the unimodal, graded nature of pFUS1 regulated transcription during the pheromone response (Poritz et al., 2001, Paliwal et al., 2007). There was a slight shoulder observed in the ‘uninduced’ GFP expression histogram from the pFUS1J2 quorum sensing strain (Figure 3-5 b). This most likely represents a sub-population of activated cells which can be assumed to be present in all populations without being as pronounced as in this particular population at that time point.

Figure 3-5. Graded autoinduced GFP expression

Overlayed GFP fluorescence intensity histograms from representative ‘uninduced’ (GFP ≤ 1800 au), ‘intermediate’ (GFP 3000-4000 au), and ‘autoinduced’ (GFP ≥ 4500 au) time-points in fermentations of a: strain JRS05 which produces α-pheromone using the native FUS1 promoter and the mfα1 gene and b: strain JRS07 which produces α-pheromone using the engineered pFUS1J2 promoter (Ingolia & Murray, 2007) and the mfα1 gene and c: strain JRS10 which controls mfα1 expression with the ARO9 promoter. Strains expressing alpha pheromone with the mfα2 gene showed the same trend. The x axes (FL1-A) represent logGFP fluorescence and y axes show cell count for each fluorescence value.

72

3.4.4 Aromatic amino acid induced quorum sensing

The quorum sensing strains which were engineered with positive feedback α-pheromone production allowed dynamic control of gene expression in response to population density. Although these responses were fine-tuneable according to circuit topology, they were constitutively active. This meant that normal cultivation techniques (including pre-culturing regimens) could not be followed, as quorum-sensing behaviour and subsequent cell-cycle arrest would occur during the pre-culture period. In order to analyse these strains, they had to be grown from single colonies on solid media (transformation plates) and inoculated straight into experimental shake-flasks with no pre-culture. This situation is not useful for industrial growth of S. cerevisiae, as pre-culturing to high population densities prior to bioreactor inoculation is necessary to minimise process times, maximise product yields and maintain fermentation reproducibility. Furthermore, quorum sensing was still triggered at relatively low population densities even in the improved pFUS1J2-driven circuit.

In order to address these problems we implemented a greater level of user control by making pheromone expression conditional on the presence of aromatic amino acids in the growth media. The amino acid responsive quorum sensing circuits (Figure 3-1 b) were designed so that strains could be grown in pre-culture without any aromatic amino acids and therefore without any quorum sensing behaviour. Furthermore they were designed to allow the fine tuning of α-pheromone communication in response to aromatic amino acid concentration and type (e.g. tryptophan, phenylalanine, or tyrosine). Population behaviour could then be altered via simple modifications to the growth medium.

Amino acid concentration was expected to influence the response (Iraqui et al., 1999) and potentially initiate cell-cycle arrest at different population-densities. Therefore, in order to understand the pheromone communication and response dynamics of these circuits, strains were grown in the absence of the inducer to a population density at which pheromone communication would elicit a response (OD660 of 1, as determined from the FUS1 circuits) prior to the addition of different types and concentrations of aromatic amino acids (Figure 3-6 a). In growth medium without amino acids, only basal levels of GFP-reporter were observed. This indicates that α-pheromone production is very low under these conditions, consistent with the ‘sender’- ‘receiver’ growth arrest assay (Figure 3-2 a). When quorum sensing strains

73 were incubated with increasing concentrations of aromatic amino acids, pheromone responsive GFP-reporter expression increased in a dose-dependent manner (Figure 3-6 a). The dose-response dynamics were also tuneable according to the type of aromatic amino acid used: GFP expression was initiated at lower concentrations of tryptophan (Trp; 0.5 g/mL) than of phenylalanine (Phe) and tyrosine (Tyr) (5 g/mL for both). Consistent with the background levels and pheromone dose-responses observed for the previous circuits, using mfα2 resulted in lower basal GFP expression and a slightly damped response dynamic relative to mfα1. Phe treatment resulted in the sharpest increase in GFP-reporter expression with a stable plateau of maximal expression. In contrast, Trp and Tyr treatment resulted in more gradual GFP-reporter increases with expression declining slightly after maximal expression being reached (25 µg/mL for Trp and ~250 µg/mL for Tyr), and maximal GFP- reporter expression being lower in Tyr-treated populations (Figure 3-6 a). Our results with respect to the response level to different aromatic amino acids are in accordance with previous results, which showed that the ARO9 system responded more strongly to Trp than to Phe and Tyr (Iraqui et al., 1999). These circuits render the pheromone response inducible by aromatic amino acids, with a fine level of control over the dynamic response according to the concentration and type of aromatic amino acid.

74

Figure 3-6. Aromatic amino acid induced quorum sensing

(a) Strains carrying pARO9-mfα1 or pARO9-mfα2 constructs (JRS10 and JRS11) were grown to an OD660 of 1 in amino acid free media (shake-flask) before being treated with the indicated concentrations of tryptophan (Trp), phenylalanine (Phe), or tyrosine (Tyr) for 4 hours (200 µL aliquots, 96 well plate) prior to GFP fluorescence measurement. (b) Strains pARO9-mfα1, pARO9-mfα2 and the non-pheromone producing control strain (JRS09) were incubated with 100 µg/mL tryptophan (black arrow) after two pre-culture passages without any amino acids present in the media, and inoculation into the main culture at an OD660 of 0.18 (shake-flasks).

(c) Control and pARO9-mfα2 strains were inoculated into main culture at an OD660 of 0.02. The control strain and one lot of pARO9- mfα2 replicates were treated with 100 µg/mL tryptophan upon inoculation (‘Low’, blue arrow) while another set of pARO9- mfα2 replicates were treated with 100 µg/mL tryptophan after the population had grown to an OD660 of 1.4 (‘High’, purple arrow). GFP fluorescence and OD660 were used to measure the pheromone response and population density respectively for all strains. Markers and error bars represent mean and standard deviation of biological triplicates.

The amino acid dose-response experiments were carried out in populations of the same density (OD660 of 1) and therefore represent a static response system. The aromatic amino acid responsive circuits (Figure 3-1 b) were also used to demonstrate quorum sensing behaviour in a dynamic system by addition of aromatic amino acids at low population density (Figure 3-6 b). Although pheromone production was initiated with Trp (100 µg/mL) at low population density (OD660nm 0.185), maximal autoinduced GFP-reporter expression was not observed until an OD660 of 1.8 was reached, 8 hours later (Figure 3-6 b). As with the dose- response assay (Figure 3-6 a), pheromone expression using the mfα2 gene resulted in a slightly damped GFP-reporter expression profile relative to the mfα1 strain (Figure 3-6 b). The quorum sensing strains showed reduced growth rates and final population densities (0.27 -1 -1 h , OD660 of ~4) relative to the control strain (0.34 h , OD660 of ~8) (Figure 3-6 b) due to the cell-cycle arrest phenotype of cells responding to α-pheromone (Chang & Herskowitz, 1990). A whole-population growth arrest phenotype was not observed, consistent with results from the positive feedback quorum sensing strains (Figure 3-3).

The pARO9-α2 strain was also used to test the effect of the population density at which quorum sensing is initiated on circuit output. When quorum sensing was initiated (with 100

µg/mL Trp) at an extremely low OD660 of 0.02 (Figure 3-6 c) a graded, non-cooperative GFP response profile (Hill coefficient 0.59 ± 0.12) was obtained. In a separate fermentation, quorum sensing was initiated at a 70 fold higher OD660 of 1.4 (Figure 3-6 c), resulting in the GFP expression profile being positively cooperative and switch-like (Hill coefficient of 3.12

75

± 0.79). Such switch-like dynamics are commonly attainable via the use of chemical inducers such as β-estradiol (McIsaac et al., 2011) which act directly on the output gene. In contrast, our tryptophan induction system acts to induce pheromone communication (rather than GFP expression directly) and consequently requires that cells be in close proximity. This mechanism represents a new inducible expression system in S. cerevisiae which is distinct from other commonly used systems (Da Silva & Srikrishnan, 2012) due to the population- density dependent nature of the output and the multitude of cellular changes which occur as part of the pheromone response (Bardwell, 2005).

These results (Figure 3-6 c) are consistent with the conceptual model which explains the differences between the FUS1 and FUS1J2 promoter strains (Figure 3-4) where initiation of pheromone communication at a low population density results in a graded GFP-reporter response while initiation at a high population density induces a switch-like response. Although in this instance the α-pheromone diffusion effect appeared to be the driving influence behind the graded response as only a minor decrease in pARO9-mfα2 population growth rate was observed when pheromone communication was induced at a low population density. These results are also consistent with a population level ‘AND’ logic gate where both high cell density and high aromatic amino acid concentration are required to maximally induce the pheromone response. The AND gate reinforces the population-density dependent nature of the GFP response and exemplifies the tuneability of the pARO9-α2 strain quorum sensing behaviour. Similar to the positive feedback circuits, the aromatic amino acid responsive circuits always resulted in unimodal GFP population distributions (Figure 3-5 c).

3.4.5 Summary and Conclusions

The quorum sensing circuits developed in this study serve as modular genetic control programs which enable autonomous and dynamic control of gene expression. Furthermore, these programs incorporated the well-characterised MAPK-mediated pheromone response in yeast, and have the potential to be interfaced with other synthetic MAPK circuits (Ingolia & Murray, 2007, Bashor et al., 2008, Groß et al., 2011, Regot et al., 2011) via the incorporation of α-pheromone expression constructs. The only other quorum sensing circuit that has been engineered in S. cerevisiae incorporated positive feedback production of a plant hormone as an intercellular signalling molecule (Chen & Weiss, 2005), and resulted in autoinduction being triggered at similar population densities to those observed in the native FUS1 promoter 76 positive feedback loops created in this study. Although the plant hormone signalling system was highly orthologous to native yeast processes, interfacing it with endogenous metabolism required growth on galactose (an expensive and therefore industrially irrelevant carbon source) for strain viability. The quorum sensing circuits developed here were minimally adapted from endogenous genetic architecture in S. cerevisiae, and reflect an emerging theme in synthetic biology which favours systems which can be integrated more seamlessly with endogenous host organism physiology (Nandagopal & Elowitz, 2011).

All of the circuits showed modest dynamic ranges of GFP expression with the best being about 7-fold (Table 3-4). This was in part due to the fact that the destabilized version of GFP which we used underestimates induction levels (Mateus & Avery, 2000). The yeast enhanced green fluorescent protein (yEGFP) has an approximate maturation time of 1 hour, and a half- life of ~7 hours making it unsuitable for measuring dynamic changes in gene expression (Mateus & Avery, 2000). For these reasons we chose to use a destabilized version of yEGFP which has a half-life of 30 minutes and never reaches full fluorophore maturation (on average) making it suitable for detecting rapid and transient changes in gene expression (Mateus & Avery, 2000). Even considering the effects of the destabilized GFP, the dynamic ranges which we observed are consistent with the only other synthetic quorum sensing network in yeast (~10 fold using the highly stable yEGFP (Chen & Weiss, 2005)). A greater dynamic range would be desirable for application of these circuits in a metabolic engineering context because it would afford a greater separation of growth and production phases. The dynamic ranges of these circuits could potentially be improved for future applications by using the FUS1J2 promoter to control circuit output in place of the noisier native version of the FUS1 promoter which we used (Figure 3-1). While the dynamic ranges observed in the positive feedback circuits were slightly better than those obtained with the aromatic amino acid responsive circuits, the latter are far more suitable for industrial application due to the fine level of control over circuit output which is definable by environmental conditions (Table 3-4). Strains carrying these circuits can be grown in pre-culture, allowing high biomass levels to be achieved prior to inoculation of main fermentation volumes. The constitutive activation of the positive feedback circuits (there is no way to “turn them off”) preclude their application in industrial fermentations and make them difficult to handle.

77

The application of aromatic amino acid responsive pheromone quorum sensing for the dynamic control of metabolite production in an industrial setting would require specific re- tuning in each case. Therefore the quorum sensing behaviour observed for the pARO9 strains (Figure 3-6 b, c) serve only as examples of how they might be used. In theory it should be possible to scale the use of these circuits up in a bioreactor setting so that a significantly higher population density can be reached before tryptophan is used to initiate quorum sensing. However, the scale up of synthetic gene networks has proven to be a non-trivial problem with correct circuit function being degraded in some cases (Moser et al., 2012). The addition of purified amino acids to an industrial fermentation could be cost prohibitive (depending on product value), this work therefore represents a ‘proof of concept’ for the conditional control of pheromone quorum sensing. Although quorum sensing populations showed significantly reduced growth rates due to the growth arrest phenotype of cells responding to pheromone (Chang & Herskowitz, 1990) this may serve as a mechanism to decouple the population growth phase from the compound production phase. Future work will explore the potential of pheromone quorum sensing for the dynamic control of metabolic pathways which impose metabolic burden and/or toxicity limitations on population growth phases. Specifically, this work will involve applying the aromatic amino acid responsive quorum sensing circuits to the bioreactor scale production of an industrially relevant fine chemical.

78

4 Quorum Sensing Linked RNAi for Dynamic Pathway Control in Saccharomyces cerevisiae

This chapter is incorporates the following manuscript, which has been submitted to ‘Metabolic Engineering’:

Williams, T. C., Averesch, N.J.H., Winter, G., Plan, M.R., Vickers, C. E., Nielsen, L. K., Krömer, J.O. Quorum-Sensing Linked RNA interference for Dynamic Metabolic Pathway Control in Saccharomyces cerevisiae. Under review. Metabolic Engineering.

4.1 Abstract

Some of the most productive metabolic engineering strategies involve genetic modifications that cause severe metabolic burden on the host cell. Growth-limiting genetic modifications can be more effective if they are ‘switched on’ after a population growth phase has been completed. To address this problem we have engineered dynamic regulation using a previously developed synthetic quorum sensing circuit in Saccharomyces cerevisiae. The circuit autonomously triggers gene expression at a high population density, and was linked with an RNA interference (RNAi) module to enable target gene silencing. As a demonstration the circuit was used to control flux through the shikimate pathway for the production of para- hydroxybenzoic acid (PHBA). Dynamic RNA repression allowed gene knock-downs which were identified by elementary flux mode analysis as highly productive but with low biomass formation to be implemented after a population growth phase, resulting in the highest published PHBA titer in yeast (148 mg/L).

79

4.2 Introduction

The metabolic engineering which is required to over-produce many commercially valuable bio-products can impose a severe metabolic burden on native cellular processes (Lee et al., 2008), limiting population growth and therefore product titer. This is because engineered pathways compete with native metabolism for carbon flux, ATP, and redox cofactors, and can result in the accumulation of toxic compounds (Keasling, 2008, Lee et al., 2008). Furthermore, many of the most productive gene deletion or overexpression strategies identified by in silico modelling are not feasible in vivo due to the limitations they impose on cell growth or survival. We demonstrate that this problem can be overcome by separating growth and production phases using an autonomous genetic regulatory circuit which enables dynamic modulation of gene expression.

There are only a few well characterized genetic induction systems in S. cerevisiae. They include the GAL promoters and the heterologous doxycycline responsive Tet promoters (Blount et al., 2012). These systems are inherently limited by the requirement for the addition of expensive inducers to the media, and in the case of the GAL promoters by the repression of gene expression on glucose. Furthermore, they do not truly separate growth from production since cells still grow normally when gene expression is induced. Here we have investigated the use of a recently developed synthetic quorum sensing circuit (Williams et al., 2012) for the dynamic control of metabolism and the separation of growth and production phases in yeast via a cell-cycle arrest phenotype.

In nature, many microorganisms are able to coordinate population-wide gene expression according to population density via cell-to-cell communication with extracellular signalling molecules. This phenomenon, termed ‘quorum sensing’ (QS) holds promise as a mechanism for separating growth from production in industrial microorganisms for the production of growth-limiting compounds (Choudhary & Schmidt-Dannert, 2010) and QS has been explored for application in biotechnology (March & Bentley, 2004, Tsao et al., 2010, Song et al., 2011). Our previously developed QS circuit enables the spatiotemporal control of gene expression, and is activated using an ‘AND’ gate system requiring both high population density and the presence of low concentrations of aromatic amino acids in the media. One limitation of the pheromone QS circuit is the lack of a mechanism for repression or silencing of gene expression. To enable the dynamic repression of gene expression we interfaced the 80

QS circuit with an RNA interference (RNAi) module (Drinnenberg et al., 2009). Unlike most Eukaryotes, S. cerevisiae does not have a native RNAi system; it was lost recently in evolutionary history due to selective pressure in favour of an endogenous RNA virus (Drinnenberg et al., 2011). When Argonaute (AGO1) and Dicer (DCR1) genes from the related yeast S. castelii are imported into S. cerevisiae, mRNAs can be targeted for degradation via the transcription of double stranded hairpin RNA with complementarity to the target (Drinnenberg et al., 2009). This RNAi module has the potential to repress the translation of any mRNA via the transcription of simple hairpin RNA constructs which target mRNAs for degradation according to simple base-pair complementarity.

To test the pheromone QS system as a mechanism for dynamic pathway control and product formation, it was applied to the production of para-hydroxybenzoic acid (PHBA) from the shikimate pathway in S. cerevisiae. Petroleum-derived PHBA is used in liquid crystal polymers and has an estimated market value of ~$150 million per annum (Krömer et al., 2012). The potential for biological PHBA production has been explored previously in plants (Siebert et al., 1996, Köhle et al., 2003, McQualter et al., 2005) and bacteria (Müller et al., 1995, Barker & Frost, 2001, Verhoef et al., 2010, Meijnen et al., 2011), although these processes are not at commercially viable levels. S. cerevisiae is a promising host for PHBA production because of its superior stress tolerance and fermentation properties, ease of genetic manipulation, and high theoretical yield of PHBA (Krömer et al., 2012).

4.3 Materials and Methods

4.3.1 Elementary flux mode analysis

EFMs were calculated using EFMTool 4.7.1 (Terzer & Stelling, 2008) freely available from http://www.csb.ethz.ch/tools/efmtool in MATLAB R2014a (The MathWorks, Natick, USA) on a desktop computer (Intel core i7-4770 CPU @ 3.4 GHz and 32 GB RAM). The maximum carbon yield of the different elementary modes was calculated in MATLAB by drawing carbon balances around the transport reactions into and out of the network. The stoichiometric network of Saccharomyces cerevisiae was compiled based on literature (Krömer et al., 2012) and metabolic pathway databases (Cherry et al., 2012, Caspi et al., 2014) and can be found in the supplementary material. It comprises the central carbon

81 metabolism as well as the aromatic amino acid biosynthesis pathways and a chorismate lyase (UbiC) reaction that catalyses PHBA formation from chorismate. For determination of knock-out targets the method of constrained minimal cut sets was used, as described in (Hadicke & Klamt, 2011). Since we wanted first to implement a minimal yield constraint, only one knock-out was allowed for the cut sets in a strain background that already had phenylalanine, tyrosine and tryptophan production eliminated.

4.3.2 Media

Strains were grown experimentally in chemically defined liquid medium (CBS) with 5 g/L ammonium sulfate, 20 g/L glucose, vitamins and trace elements as described previously (Verduyn et al., 1992). During strain construction purified amino acids (Sigma) were used to complement appropriate auxotrophies in agar plates (same composition as chemically defined media above) while YPD or YPG (galactose) supplemented with appropriate antibiotics was used during gene deletion procedures. E. coli DH5α was used for plasmid propagation/storage and was grown in LB medium with kanamycin. For glucose spike experiments, CBS medium containing 100 g/L glucose was diluted 1-in-10 into growth cultures at the indicated time points.

4.3.3 Growth conditions

Shake-flask fermentations were carried out at 30 C, 200 rpm with aluminium foil used to cover tops of baffled flasks, and with medium making up 10% of the flask volume. Single colonies from solid medium were used to inoculate 10 mL of liquid CBS medium. After 24 h of growth cells were passaged into a second pre-culture and grown to mid-log phase (OD660nm of 1-5) prior to inoculation of the experimental culture at an OD660nm of 0.4. Quorum sensing was initiated via the addition of 100 µg/mL tryptophan, 2 h post inoculation, at an OD660nm of

~0.8 with QS mediated gene expression expected to be fully activated above OD660nm of 5 based on previous findings (Williams et al., 2012). Eight hours after inoculation, cultures were spiked with 10 g/L glucose by diluting a 100 g/L glucose CBS medium 1:10 into the main culture. All experiments were carried out in biological triplicate where individual colonies from solid media were used for pre-cultures and subsequent inoculation of main cultures.

82

4.3.4 Strains and plasmids

Primers, plasmids, and strains used in this study are shown in Table 4-1, Table 4-2, and Table 4-3 respectively. Annotated plasmid sequences are available in the supplementary material. DNA manipulation and propagation were carried out using standard techniques (Sambrook & Russell, 2001) unless stated otherwise. All S. cerevisiae transformations were carried out using the lithium acetate method (Gietz & Schiestl, 2007). Deletion of the FUS1 gene was performed with the reusable LoxP-KanMX-LoxP cassette as described previously (Güldener et al., 1996) using the FUS1KOF and FUS1KOR primers to amplify the KanMX deletion cassette from pUG6 and the FUS1DCF and FUS1DCR primers to check the chromosomal locus for deletion and marker removal. Unless otherwise stated genes, promoters, and terminators were PCR amplified from CEN.PK 2-1c genomic DNA. Strains transformed with yeast integrating plasmids were screened for correct single copy integration using PCR as previously described (Stansfield & Stark, 2007), except for AGO1 and DCR1 integration plasmids which were later confirmed to have each integrated twice at the TRP1 locus. Multiple integration events were confirmed by PCR amplifying through the integrating vector TRP1 and LEU2 loci with pRSIntF/pRSIntR primers and via read depth comparison with surrounding chromosome using whole genome sequencing.

Several plasmids were generated to enable the expression of shikimate pathway enzymes and silencing constructs in response to pheromone. Plasmid pTCW011 was created by inserting the 500 bp FUS1J2 promoter (amplified from pNTI144 using pFUS1J2F and pFUS1J2R primers) into pRS406 using XhoI and EcoRI sites. pTCW012 was then made by inserting 300 bp of the CYC1t terminator region (PCR-amplified using CYC1tF and CYC1tR) 3 of the FUS1J2 promoter in pTCW011 using XbaI and SacI sites. pTCW013 was made by amplifying a codon-optimised UBiC gene synthesised by GeneArt using UBiCF and UBiCR primers and inserting it in between the FUS1J2 promoter and the CYC1t terminator in pTCW012 using EcoRI and XmaI cut sites. pTCW014 was made in the same way except a feedback-resistant ARO4 gene (ARO4K229L) from the ARO4K229LpCV3 plasmid was used in place of the UBiC gene. ARO4K229L contains point mutation at the 229th codon (from “AAC” to “AAA”) of the ARO4 gene of S288c resulting in a change from lysine (K) to leucine (L) resulting in a tyrosine feedback inhibition insensitive enzyme (Luttik et al., 2008). The point mutation was introduced using overlap extension PCR (Ho et al., 83

1989). Four primers were designed, two of them targeting the ends of the gene (ARO4 NotI forw, ARO4 BamHI rev) with amplifying direction towards each other and two primers incorporating the mutation at their 5' end while amplifying in opposing directions (ARO4 mut. rev, ARO4 mut. forw). The primers ‘ARO4 NotI forw’ and ‘ARO4 BamHI rev’ were also used for cloning into the pCV3 vector to create pCV3-ARO4K229L. pTCW015 was made by inserting PCR-amplified PFUS1J2-ARO4-CYC1t 3 (primers cyc1-FUS1J2F and CYC1GR) 3 of the CYC1t region in Eco53kI-linearized pTCW013 using the Gibson Assembly technique (Gibson et al., 2009)

Hairpin fragments (250 bp sense – rad9 linker – 250 bp antisense) were synthesised by GeneArt with the exception of the ZWF1 and CDC19 constructs, which was synthesised by DNA 2.0, and SST2 which was made by PCR and Gibson Assembly. The rad9 linker is a 73 bp intron from Schizosaccharomyces pombe which serves as a loop which separates the two stems of hairpin constructs as in (Drinnenberg et al., 2009). The 250 bp sequences were arbitrarily chosen from the centre of each open reading frame, except for the CDC19 hairpin, which used 250 bp of sequence which had high homology (80 %) to its paralog, PYK2. This homologous region was chosen with the intention of suppressing the expression of PYK2 along with CDC19 using a single hairpin expression construct.

A hairpin expression construct for the SST2 gene was made by Gibson Assembly where five parts were simultaneously assembled to create an expression construct which transcribes (in response to pheromone) an RNA hairpin. The FUS1J2 promoter was amplified from pNTI144 using pFUS1J2Fc and pFUS1J2Rc, SST2sense, SST2antisense, and CYC1t were amplified using SST2senseF and SST2senseR, SST2antiF and SST2antiR, CYC1GF and CYC1GR primers respectively. PCR products were cleaned and Gibson-assembled into Eco53kI-linearized pRS413 to create pTCW016. pTCW017 was made by amplifying the FUS1J2-SST2sense cassette (primers pFUS1J2Fc and SST2senseR) and SST2antisense- CYC1t (primers SST2antiF and CYC1GR) and Gibson-assembling with NotI linearized pTCW010. pTCW017 was cut with ClaI and XmaI to remove the SST2 hairpin sequence. The linearized PARO9-mfα2-PFUS1J2-CYC1t-pRS413 fragment was gel-purified and used as a vector to insert a ClaI/ XmaI-digested ARO7 hairpin expression construct, generating pTCW018.

84 pTCW019 was made by PCR-amplifying the TKL1 gene (primers TKL1F and TKL1R) and cloning it between the FUS1J2 promoter and CYC1t terminator in pTCW012 using EcoRI and XbaI restriction sites. pTCW020 was made by amplifying the PFUS1J2-TKL1-CYC1t cassette from pTCW019 (primers 406X-J2F and 406X-CYC1tR2), digesting with AvrII and AflII, and ligating with PCR-linearized pRS406 (pRS406F and pRS406R primers). These primers had 5 extensions to provide BspEI and NotI restriction sites 5 of the PFUS1J2- TKL1-CYC1t cassette, and NruI and SalI sites 3 of the cassette. PFUS1J2-ARO4-CYC1t was amplified from pTCW014 (primers M13F2 and pRS406R) and inserted into pTCW020 using XhoI and AvrII sites to make pTCW021. PFUS1J2-UBiC-CYC1t was amplified from pTCW013 using BssHII-pFUS1J2F and XhoI-CYC1t primers and inserted into pTCW021 to create pTCW022.

The PFUS1J2-ZWF1sense-rad9 linker-ZWF1antisense-ADH1t construct was inserted 5 of the TKL1 expression cassette in pTCW022 using BspEI and NotI to create pTCW023. Similarly, the CDC19sense-rad9 linker-CDC19antisense-CYC1t construct was inserted 3 of the TKL1 expression cassette in pTCW025 using NruI and SalI to create pTCW024. All strains were created by transformation and selection for the yeast integrating or centromeric plasmids listed in Table 4-2.

Table 4-1. Oligonucleotides used in this study

Oligo name 5 to 3 sequence PFUS1J2F GAGCTCCTCGAGCCCTCCTTCAATTTTTCTG PFUS1J2R ATCGATGAATTCTTTGATTTTCAGAAACTTGTTGG FUS1KOF TTTCCTTTAAGAGCAGGATATAAGCCATCAAGTTTCTGAAAATCAAAATGCCA GCTGAAGCTTCGTACG FUS1KOR TACAGAATTATAGGTATAGATTAAATGCGAACGTCAATATTATTTTCATCACT ATAGGGAGACCGGCAG FUS1DCF GCGCTGTCTCATTTTGGTGC FUS1DCR TGCATTCCCTAGTTTCGCGG UBiCF TTAATTGAATTCATGTCCCATCCAGCTTTGACAC UBiCR TATTATCCCGGGTCAGTATAATGGGGAAGCTGGC ARO4F TTAATTGAATTCATGAGTGAATCTCCAATGTTC ARO4R TATTATCCCGGGCTATTTCTTGTTAACTTCTC

85

ARO4 NotI F AGTCGCGGCCGCAAAAAAATGAGTGAATCTCCAATGTTCG ARO4 BamHI R CTGAGGATCCCTATTTCTTGTTAACTTCTCTTCTTTGTCTGAC ARO4 mut. R CGAAGGAGACCAAATCAGCCAAGTATTTAGGAGAAATGG ARO4 mut. F GGTTCTGAAATGCTTGATACCATTTCTCCTAAATACTTGGCTG CYC1tF TATAATTCTAGAACAGGCCCCTTTTCCTTTGT CYC1tR TATTATGAGCTCACGATGAGAGTGTAAACTGC cyc1-pFUS1J2F TTAATTTGCAAGCTTCGCAGTTTACACTCTCATCGTCGAGCCCTCCTTCAATTT TTCTG PFUS1J2Fc CACTAGTTCTAGAGCGGCCGCCACCGCGGTGGAGCTCCCCTCCTTCAATTTTT CTG PFUS1J2Rc CATGCATAATATCAGTACAGTATCGATTTTGATTTTCAGAAACTTGTTGG SST2senseF CCAACAAGTTTCTGAAAATCAAAATCGATACTGTACTGATATTATGCATG SST2senseR GAATTCCTCTATCTTTACCAATTAGTTTCAATGTTTAGTAAGGTTTGAAAAAA GTTCCAACACACCTGATTTTCTCGAGATGACCCCTTAATGTGAAATC SST2antiF CTCGAGAAAATCAGGTGTGTTGGAACTTTTTTCAAACCTTACTAAACATTGAA ACTAATTGGTAAAGATAGAGGAATTCATGACCCCTTAATGTGAAATC SST2antiR ACAAAGGAAAAGGGGCCTGTCCCGGGACTGTACTGATATTATGCATG SST2hpJ2F ATAAATTCCCGTTTTAAGAGCTTGGTGAGCCCCTCCTTCAATTTTTCTG SST2hp- TTCAAACGATACCTGGCAGTGACTCCTAGCACGATGAGAGTGTAAACTGC CYC1tR CYC1GF CATGCATAATATCAGTACAGTCCCGGGACAGGCCCCTTTTCCTTTGT CYC1GR GCGCGCAATTAACCCTCACTAAAGGGAACAAAAGCTGGAGACGATGAGAGT GTAAACTGC TKL1F TATTATGAATTCATGACTCAATTCACTGACATTG TKL1R TATTATTCTAGATTAGAAAGCTTTTTTCAAAGGAG 406X-J2F TATTATCCTAGGTCCGGATATTATGCGGCCGCCCCTCCTTCAATTTTTCTG 406X-CYC1tR2 TATTATCTTAAGGTCGACTATTATTCGCGAACGATGAGAGTGTAAACTGC PRS406F TATTATCTTAAGCATGGTCATAGCTGTTTCC PRS406R TATTATCCTAGGAAAGGGAACAAAAGCTGGAG BssHII- TATTATGCGCGCCCCTCCTTCAATTTTTCTG pFUS1J2F XhoI-CYC1tR TATTATCTCGAGACGATGAGAGTGTAAACTGC M13F2 CGCCAGGGTTTTCCCAGTCACGAC TRP1A AGAGACCAATCAGTAAAAATCAACG TRP1D GCGAAAAGACGATAAATACAAGAAA pRSIntF ACCATTATTATCATGACATTAAC

86 pRSIntR TTCTCCTTACGCATCTGTGC ARO7 qRTF TCCCGCTATACACAAAGCTGATCA ARO7 qRTR CGTTGGTAGGGTCCACACCA CDC19 qRTF GCTGTCGCTGCTGTTTTCGA CDC19 qRTR AGCAGCTCTTGGGCATCTGG zwf1 qRTF TAAGCCCGCCTACGTGGATG zwf1 qRTR CATCATGATGGGGACGCCCT ALG9F CACGGATAGTGGCTTTGGTGAACAATTAC ALG9R TATGATTATCTGGCAGCAGGAAAGAACTTGGG

Table 4-2. Plasmids

Name Details Origin pRS406 URA3 integrating vector (Sikorski & Hieter, 1989), Euroscarf pRS404-PTEF -Ago1 TRP1 integrating vector with constitutive (Drinnenberg et al., Argonaute expression 2009) pRS405-PTEF-Dcr1 LEU2 integrating vector with constitutive (Drinnenberg et al., Dicer expression 2009) pNTI144 PFUS1J2-STE4-ADH1t (Ingolia & Murray, 2007) pRS413 URA3 low copy number vector (Sikorski & Hieter, 1989), Euroscarf pCV3 Cloning vector (Krömer et al., 2012) pCV3-ARO4K229L DNA source for feedback resistant ARO4 This study gene pTCW010 PARO9-mfα2- mfα2t- pRS413 (Williams et al., 2012) pTCW011 PFUS1J2-pRS406 This study pTCW012 PFUS1J2-CYC1t-pRS406 This study pTCW013 PFUS1J2-UBiC-CYC1t-pRS406 This study pTCW014 PFUS1J2-ARO4-CYC1t-pRS406 This study pTCW015 PFUS1J2-UBiC-CYC1t- pFUS1J2- This study ARO4-CYC1t-pRS406 pTCW016 PFUS1J2-SST2sense-rad9linker- This study SST2antisense-CYC1t-pRS413

87 pTCW017 PARO9-mfα2- mfα2t-PFUS1J2- This study SST2sense-rad9linker-SST2antisense- CYC1t-pRS413 pTCW018 PARO9-mfα2- mfα2t-PFUS1J2- This study ARO7sense-rad9linker-ARO7antisense- CYC1t-pRS413 pTCW019 PFUS1J2-TKL1-CYC1t-pRS406 This study pTCW020 PFUS1J2-TKL1-CYC1t-pRS406 (with This study BspEI, NotI, SalI, NruI cut sites) pTCW021 PFUS1J2-ARO4-CYC1t- PFUS1J2- This study TKL1-CYC1t -pRS406 pTCW022 PFUS1J2-UBiC-CYC1t- PFUS1J2- This study ARO4-CYC1t- pFUS1J2-TKL1-CYC1t - pRS406 pTCW023 PFUS1J2-UBiC-CYC1t- PFUS1J2- This study ARO4-CYC1t- pFUS1J2- ZWF1sense- rad9linker-ZWF1antisense-ADH1t - PFUS1J2-TKL1-CYC1t -pRS406 pTCW024 PFUS1J2-UBiC-CYC1t- PFUS1J2- This study ARO4-CYC1t- PFUS1J2- ZWF1sense- rad9linker-ZWF1antisense-ADH1t - PFUS1J2-TKL1-CYC1t - PFUS1J2- CDC19sense-rad9linker-CDC19antisense- CYC1t- pRS406 pUG6 LoxP-KanMX-LoxP cassette (Güldener et al., 1996), Euroscarf pSH65 Galactose inducible cre-recombinase (Güldener et al., 1996), Euroscarf

88

Table 4-3. Saccharomyces cerevisiae strains used in this study

Name Genotype Notes Origin

CEN.PK2-1c MATa; ura3-52; trp1-289; leu2-3,112; his3Δ 1; MAL2-8C; SUC2 Haploid auxotrophic MATa lab Euroscarf strain. PHB01 CEN.PK2-1c, bar1Δ, trp1::pRS404-PTEF-Ago1(x2)-pRS405-PTEF- Argonaute and Dicer integrated base This study Dcr1(x2) strain. PHB02 CEN.PK2-1c, bar1Δ, fus1Δ, trp1::pRS404-PTEF-Ago1(x2)-pRS405- PHB01 + FUS1 gene deletion. This study PTEF-Dcr1 (x2) PHB03 PHB02, ura3::pRS406, pTCW010 QS control strain This study

PHB04 PHB01, ura3::pTCW013, pTCW010 Pheromone QS with UBiC This study expression PHB05 PHB02, ura3:: pTCW013, pTCW010 Pheromone QS with UBiC This study expression, FUS1 gene deleted. PHB06 PHB02, ura3::pTCW015, pTCW010 Pheromone QS with UBiC and This study feedback resistant ARO4 expression, FUS1 gene deleted. PHB07 PHB02, ura3::pTCW015, pTCW018 pHB06 with RNAi mediated ARO7 This study repression. PHB08 PHB02, ura3::pTCW022, pTCW018 PHB07 with TKL1 expression and This study RNAi mediated ARO7 repression. PHB09 PHB02, ura3::pTCW023, pTCW018 PHB08 with ZWF1 hairpin This study construct. 89

PHB10 PHB02, ura3::pTCW024, pTCW018 PHB09 with RNAi mediated CDC19 This study repression. PHB11 PHB02, ura3::pTCW022, pTCW010 QS strain with no hairpin expression This study

90

4.3.5 Analytics

Extracellular PHBA concentrations were determined by taking 0.5 mL of culture supernatant at nominated times. Cells were removed from the supernatant by centrifugation at 4 C, for 7 min. at 13000 × g. Clarified supernatant was stored in a fresh microfuge tube at -20 C. PHBA concentrations were quantified by RP-HPLC using an Agilent 1200 HPLC system. In brief, samples were kept at 4 C in a high-performance auto-sampler (Agilent HiP-ALS, G1367B). Sample (30 µL) was injected into a C18 column (Zorbax Extend C18, 3.5 um, 4.6 x 150 mm, Agilent PN: 763953-902) with a guard column (SecurityGuard Gemini C18, Phenomenex PN: AJO-7597). Column temperature was kept at 35 C in a thermostatted column compartment (Agilent TCC, G1316B), and analytes were eluted using gradient chromatography at 1 mL/min flow rate. The gradient method was as follows: 5-35 % B from 0-5 min, 35-40 % B from 5-14 min, 100 % B from 15-18 min, and 5 % B from 18.1-20 min - using a binary pump (Agilent Bin Pump, G1312A). Solvent B was 0.045 % TFA in 80 % acetonitrile in water, while Solvent A was 0.045 % TFA in high purity water (18.2 MΩ⋅cm). Solvents were degassed using an in-line degasser (Agilent Degasser, G1379B). Analytes of interest were monitored using a diode array detector (Agilent DAD SL, G1315C) at 206 and 254 nm wavelengths. Spectral scans were also performed on each eluting peaks from 190-400 nm to confirm their spectral fingerprint and purity. Chromatograms were integrated using ChemStation (Rev B.03.02[341]). Extracellular glucose, and ethanol concentrations were measured as previously described (Dietmair et al., 2010) in cell free supernatants collected as mentioned earlier.

4.3.6 Quantitative Real-Time PCR

To quantify relative mRNA repression in genes targeted by RNAi hairpin constructs, the final QS strain PHB10 was compared with an identical strain which lacked any hairpin expression constructs (PHB11). Biological triplicates were pre-cultured and grown as described earlier, except that the main culture was inoculated at an OD660nm of 0.1. Two hours after inoculation (at an OD660nm of ~0.2), each set of cultures was dosed with 100 µg/mL tryptophan to initiate quorum sensing. Samples (1-2 mL) were taken immediately and centrifuged to pellet cells; cell pellets were resuspended in RNAlater (Ambion) and stored at -20 ˚C until extraction. RNA was extracted using the RiboPure Yeast RNA kit (Life Technologies). RNA (100-1000 ng) was reverse transcribed using OligoDT18 primers and SuperScriptIII reverse transcriptase (Life Technologies) according to the manufacturer’s instructions. ARO7, ZWF1, and CDC19 mRNAs were quantified 91 using the ‘qPCR’ primers listed in Table 4-1 by normalizing to the housekeeping gene ALG9 (Teste et al., 2009) using the ΔCq method for each biological replicate individually. Errors were reported as the percentage errors of the standard deviation of these biological triplicates, with mean expression values of the control normalised to 1. qPCR was carried out in a RotorGene 6, using Platinum SYBR Green qPCR supermix-UDG (Life Technologies).

4.3.7 Enzyme Assays

Cell pellets were collected from the fermentation described in the previous section at 48 hours and kept at -80 °C. Crude protein extracts were generated using the Y-PER Yeast Protein Extraction Reagent as per the manufacturer’s instructions (Thermo Scientific) and total protein was quantified using a 2D quant kit (GE Healthcare Life Sciences). Chorismate mutase activity was assayed as previously described (Luttik et al., 2008). Pyruvate kinase activity was measured using a colorimetric Pyruvate Kinase Activity Assay (Sigma Aldrich kit catalogue number MAK072) according to the manufacturer’s instructions.

4.3.8 Statistical Analysis

Mean values from biological triplicate experiments (cultures derived from individual colonies on solid media) were compared for PHBA titers, mRNA expression levels and enzyme activities from different strains/conditions. Pairwise differences were tested for statistical significance using a two- sided student’s t-test with equal variance in the GraphPad Prism software package. An ordinary one-way ANOVA was used in GraphPad Prism to test for differences between the multiple PHBA titers which were measured for the different strains.

4.4 Results and Discussion

4.4.1 Strain Design

Interfacing PHBA production with the QS regulatory network required that expression of the relevant genes be controlled by the pheromone responsive FUS1J2 promoter (Ingolia & Murray, 2007) (Figure 4-1 a, b). When an aromatic amino acid such as tryptophan is present in the media the α-pheromone peptide is expressed from the ARO9 promoter and secreted from the cell (Williams et al., 2012) (Figure 4-1 b). At a high population density the extracellular pheromone triggers gene

92 expression from the FUS1J2 promoter and causes cells to arrest growth in the G1 phase of the cell cycle. This mechanism serves to decouple population growth and production phases, allowing growth limiting engineering strategies to be implemented (Figure 4-1 c). During the pheromone- autoinduced growth arrest, PHBA production is switched on via the expression of the relevant pathway genes and RNAi repression constructs from the FUS1J2 promoter (Figure 4-1 a, b). Specifically, several pathway enzymes were dynamically expressed along with the knock-down targets. Previous work has identified TKL1p and a feedback-resistant ARO4p as important enzymes for the production of shikimate derived compounds (Luttik et al., 2008, Curran et al., 2013). Finally, UbiCp, the enzyme that catalyses PHBA production from chorismate (Siebert et al.) was also expressed under QS regulation to dynamically control production.

Figure 4-1. PHBA production from the shikimate pathway with pheromone quorum sensing

(a) The shikimate pathway feeds aromatic amino acid biosynthesis using phosphoenolpyruvate and erythrose-4- phosphate. We optimised the supply of these precursors by overexpressing TKL1 and targeting CDC19 for repression using RNA interference. ARO3 and ARO4 encode 3-deoxy-D-arabino-heptulosonate-7-phosphate (DAHP) synthase isozymes which are strongly feedback inhibited by phenylalanine and tyrosine respectively. In order to enhance flux through the shikimate pathway in response to pheromone, a feedback resistant version of ARO4 (K229L) was placed under the control of the FUS1J2 promoter. S. cerevisiae has no known native enzyme for PHBA formation; we 93 therefore incorporated a codon optimised version of the E. coli chorismate pyruvate lyase (UBiC) under the control of the FUS1J2 promoter to catalyse PHBA production from chorismate. Chorismate availability was increased via the expression of an RNAi repression construct for ARO7. (b) The circuit topology of the final quorum sensing PHBA producing strain. Aromatic amino acids present in the growth media initiate the production of α-pheromone and quorum sensing behaviour. At a high population density the extracellular concentration of α-pheromone is high, and once bound to the membrane-bound Ste2p pheromone receptor, triggers the mitogen activated protein kinase phosphorylation cascade. The activated Ste12p transcription factor then up-regulates transcription from the FUS1J2 promoter (Ingolia & Murray, 2007) as well as native pheromone responsive promoters in the genome. FAR1p regulates cell-cycle arrest in the G1 phase. PFUS1J2 mediates expression of the UBiC, ARO4K229L, and TKL1 genes as well as hairpin RNA structures which target complementary mRNAs for degradation via S. castelii Dicer and Argonaute proteins. (c) As a quorum sensing population grows over time the concentration of α-pheromone increases. At a high population density a critical α-pheromone concentration is reached which triggers gene expression from pheromone responsive promoters. Cells switch on the production of PHBA and arrest growth as part of a coordinated production phase.

The most promising targets for gene knockout to enhance flux towards PHBA carry a severe growth penalty. In earlier studies the knockout of ARO7 (encoding chorismate mutase) and TRP3 (encoding anthranilate synthase) has been described as a way to increase chorismate availability for PHBA formation (Krömer et al., 2012). However, this strategy also causes phenylalanine, tyrosine and tryptophan auxotrophy and the associated growth defect, even when cells are fed these amino acids (Winter et al., 2014). To avoid the auxotrophies associated with gene deletion during the growth phase, these mutations were selected for conditional knockdown using our novel QS circuit (Williams et al., 2012) in combination with RNAi. Using a stoichiometric model of central carbon metabolism, elementary flux modes of PHBA production were subjected to determination of constrained minimal cut-sets, with the aim to identify an intervention strategy that would confer a minimum yield constraint to ensure PHBA production. Therefore a network that already had the chorismate mutase and anthranilate synthase knockouts implemented was analysed. Each point on the graph (Figure 4-2) represents a unique combination of metabolic reactions (an elementary mode) and the relative contribution that this mode makes towards pHBA production (x axis) and biomass formation (y axis). A cut-set with a single target reaction that would confer a minimum yield constraint was found: the knockout of the pyruvate kinase-catalysed reaction (CDC19 and PYK2 genes) (Figure 4-2). CDC19 mutants are unable to grow on glucose (Sprague, 1977); consequently, traditional gene deletion is not possible for glucose-based fermentations, which are favoured at large scale due to feedstock cost and availability. It was therefore essential that pyruvate kinase repression be implemented dynamically along with chorismate mutase (ARO7) and anthranilate synthase (TRP2, TRP3) repression. Given that the QS circuit is activated by the presence of tryptophan in the growth medium, and that tryptophan causes strong feedback inhibition

94 of anthranilate synthase (Miozzari et al., 1978, Graf et al., 1993), RNAi-targeting of anthranilate synthase was not necessary.

A B

]

]

mol

mol

-

-

c

c

/ /

/ /

mol

mol

-

-

c

c

[

[

Biomassyield Biomassyield

pHBA yield [c-mol / c-mol] pHBA yield [c-mol / c-mol] Figure 4-2. Elementary flux mode analysis of the pyruvate kinase reaction knock-out

Elementary modes are shown with relative contributions to biomass (y-axes) and product formation (x-axes) for a network of S.cerevisiae with chorismate mutase and anthranilate synthase knockouts (A) and a network of S.cerevisiae with pyruvate kinase also removed (B).

4.4.2 Validation of the QS response

To verify that the QS strains were behaving appropriately, each strain was grown with and without tryptophan (example shown in Figure 4-3 a). In accordance with induction of QS, PHBA production was initiated (Figure 4-3 c). This resulted in a concurrent reduction in growth rate relative to untreated cultures (Figure 4-3 a) due to cell cycle arrest and the “shmoo” mating phenotype of the pheromone response (Figure 4-3 d). Conversely, without quorum sensing activated the same strains grew to a higher population density due to the absence of the growth arrest/shmoo phenotype (Figure 4-3 a, d). While PHBA was produced in non-induced cultures as a result of basal expression from the FUS1J2 promoter, titers were significantly lower than in induced cultures (Figure 4-4). This is consistent with our previous data, which demonstrated that the QS circuit functions as an AND gate: gene expression is not fully activated unless both an aromatic amino acid is present in the medium AND a high population density is reached (Williams et al., 2012). This mode of functionality is further supported by the observed sharp increase in PHBA concentration between 8 and 12 hours, when population density (Figure 4-3 a) nears its maximum in the quorum sensing culture.

95

Figure 4-3. Quorum sensing mediated PHBA production

The final production strain (PHB10) was grown with (black lines, ‘QS’) or without (grey lines, ‘Control’) tryptophan- mediated activation of quorum sensing via tryptophan addition 2 hr post-inoculation (OD660nm ~0.8). Both treatments were spiked with 10 g/L of glucose after 8 h of fermentation, when approximately half of the expected maximum shake flask population density was reached. Population density (a), extracellular glucose and ethanol concentrations (b), and PHBA titres (c) are shown. n = 3  1 SD. (d) Light microscopy (1000 × magnification) images of the PHB09 strain throughout fermentation carried out as described for the PHB10 strain. The images are representative of the other strains and replicates under similar conditions.

4.4.3 Stepwise engineering for improved PHBA production

PHBA titers were measured with and without QS activated in the same strain after the introduction of each genetic modification (Figure 4-4). In every case PHBA production was significantly lower

96 without the activation of the QS circuit, demonstrating the dynamic functionality and range of the engineered regulation.

QS-mediated expression of a codon-optimised E.coli chorismate pyruvate lyase (UBiC) in strain PHB04 resulted in PHBA production reaching 46 µM. This was a 77 % increase over the tryptophan-treated control strain (PHB03), which contains only the QS and RNAi modules and produces endogenous levels of PHBA (Figure 4-4). Deletion of the FUS1 gene (involved in facilitating membrane fusion of mating partners) has been shown to reduce cell death from 30 % of a population to 5 % in response to pheromone (Zhang et al., 2006). We therefore expected increased PHBA production in the FUS1 pheromone quorum sensing strain expressing QS-linked UBiC (PHB05). Interestingly, accumulation was not significantly different from PHB04. This suggests that either cell death in response to pheromone is not a limiting factor at this level of production, or that PHBA productivity was more significantly influenced by endogenous feedback inhibition in the shikimate pathway. The additional QS-mediated expression of a feedback resistant 3-deoxy-D-arabino-heptulosonate-7-phosphate (DAHP) synthase enzyme (ARO4K229L) increased PHBA titer a further 5.7-fold to 297 µM (Figure 4-4; strain PHB06). This is consistent with previous data showing that over-expression of ARO4K229L enhances flux through the shikimate pathway 4-5 fold (Luttik et al., 2008). All strains carried an RNAi module (in the form of Argonaute and Dicer genes). When an ARO7-targeted RNAi hairpin was expressed, PHBA titer increased a further 80 % to 534 µM (Figure 4-4; strain PHB07). This suggests that conditional repression of Aro7p during the production phase, increases chorismate availability for the UBiCp enzyme (Figure 4-2 a).

It has previously been demonstrated that the deletion of ZWF1 and overexpression of TKL1 re- wires pentose phosphate pathway flux to increase erythrose-4-phosphate supply to the shikimate pathway (Curran et al., 2013). When pheromone-responsive TKL1 expression was introduced, PHBA titer increased by 49 % to 795 µM (strain PHB08). Expression of TKL1 also increased the PHBA titer in the non-QS-activated control cultures, although this effect was not as prominent in subsequent strains. This effect can be attributed to the combinatorial leaky expression of pathway components from the FUS1J2 promoter. When a ZWF1-targeted hairpin construct was introduced, there was no significant difference in PHBA production (760 µM, Figure 4-4; strain PHB09). This was because there was no detectable level of repression of ZWF1 mRNA (see section 4.4.4). In order to fully exploit the use of dynamic regulation, we conditionally repressed the CDC19 gene. CDC19 encodes a pyruvate kinase enzyme; CDC19 strains are not viable for growth on glucose 97

(Giaever et al., 2002). QS-linked RNAi-mediated repression of CDC19 resulted in a further 41 % increase in PHBA titer over the previous strain (PHB09), to 1.07 mM (Figure 4-4; strain PHB10). This final strain represented a 37-fold increase in titer over the base strain (PHB03), and a yield of

4.93 mg/gglucose.

Figure 4-4. PHBA titers using quorum sensing mediated modulation of shikimate pathway genes.

PHBA titers after 72 hours of shake-flask fermentation with (black bars) and without (grey bars) quorum sensing activated. Blue squares represent the absence of the listed modification while grey squares indicate the presence. Strain names are listed below each column. The control strain PHBA03 has a QS circuit and RNAi module, and produces endogenous levels of PHBA. n= 3 ± 1 SD.

98

4.4.4 Validation of RNAi knock-down

RNAi module functionality was examined by measuring the level of target gene mRNA using RT- qPCR. ARO7, CDC19, and ZWF1 mRNA levels were compared between the final production strain (PHB10) and a strain with no hairpin RNA expression constructs but which is otherwise identical (PHB11; Table 4-3). Even before quorum sensing was initiated, ARO7 mRNA levels were lower (25.2 ± 2.7% of the reference strain) in PHB10 (Figure 4-5 a). This suggests the presence of ‘leaky’ expression of the ARO7 RNAi construct, leading to repression of ARO7. This was surprising given that the FUS1J2 promoter is known to have very low expression levels in the absence of pheromone (Ingolia & Murray, 2007) and reflects the sensitivity of the RNAi system, which was previously noted for its extreme leakiness when under the control of the galactose induction system (Drinnenberg et al., 2009). This leaky expression was clearly insufficient to facilitate repression of CDC19 before QS activation, as CDC19 mRNA levels were not significantly different compared to the control. This difference can be attributed to the fact that CDC19 mRNA levels were 132 times higher than those observed for ARO7 prior to the activation of QS at 2 hours. The leaky hairpin expression was evidently insufficient to decrease the level of a highly abundant transcript such as CDC19.

The quorum sensing response was fully activated in both strains at 24 hours post-inoculation, as determined by the shmoo cell morphology (Figure 4-3 d). At this point, there was significant further repression of ARO7 mRNA (to 3.0 ± 0.3% of reference strain), and repression of CDC19 mRNA (to 12.2 ± 1.4% of reference strain) was evident (Figure 4-5 b). These repression levels represent dynamic ranges of 7.4 (ARO7) and 7.2 (CDC19), reflecting the dynamic range of the FUS1J2 promoter which drives the hairpin construct. ARO7 and CDC19 mRNA levels were still significantly lower than in the control strain at 48 hours, although not as strongly repressed as at 24 hours (Figure 4-5 c). This was most likely due to weaker hairpin expression from the FUS1J2 promoter as cells become gradually desensitized to pheromone over time (Bardwell, 2005). There was no significant difference in ZWF1 mRNA levels at any time point (data not shown). This indicates that the ZWF1-targeted RNAi construct was ineffective at decreasing mRNA levels. Recent work on the heterologous RNAi module in S. cerevisiae has demonstrated that the secondary structure of the mRNA region which is targeted for repression is an important aspect of hairpin design (Si et al., 2014). The lack of repression observed for the ZWF1 gene in our system is most likely due to the fact that the secondary structure of the target mRNA region was not considered during the design.

99

Figure 4-5. Relative mRNA levels of RNAi targets

RNAi target gene mRNA levels were compared between a strain with (black bars, strain PHB10) and without (grey bars, strain PHB11) hairpin constructs using RT-qPCR with internal normalisation to the ALG9 housekeeping gene. Relative mRNA levels were measured at 2 hours post inoculation before quorum sensing activation (a), 24 hours after inoculation, when the quorum sensing phenotype had been fully activated (b), and 48 hours post inoculation (c). Means of biological triplicate fermentations and RNA extractions are shown as percentages normalised to control strain mRNA levels set at 1, with error bars representing percentage errors of ± 1 standard deviation. * denotes statistically significant differences.

Protein abundance and mRNA levels are often not well correlated in yeast (Greenbaum et al., 2003). We therefore investigated whether the engineering approach successfully reduced protein levels. Enzyme activity assays were used as a proxy to assess whether the RNAi system was effectively decreasing protein levels for the target genes. Chorismate mutase (ARO7) activity was significantly reduced (17.6 % of that observed in the control strain) while pyruvate kinase activity (the result of both PYK2p and CDC19p activity) was reduced to 11.1 % of the control at 48 hr post- inoculation (Figure 4-6). 100

Figure 4-6. Enzyme activity assays

Chorismate mutase (a) and pyruvate kinase (b) enzyme activities were assayed using protein extracts from strains with (black bars) and without (grey bars) RNAi hairpin constructs after 48 hours of quorum sensing fermentation. Error bars represent one standard deviation of biological triplicate measurements. Significant differences between means are indicated by *.

Although the dynamic repression of the essential CDC19 gene resulted in a significant increase (41 %) in PHBA yield, the increase is nowhere near the ≥ 30 % carbon yield predicted by elementary mode analysis of the pyruvate kinase knockout (Figure 4-2) (carbon yield was 0.75 % c-moles/c- moles of glucose). This can be attributed to several factors, the first of which is the fact that the dynamic RNAi approach did not completely remove target mRNA from the system, even though there was a dramatic repression effect (Figure 4-5 b). Although CDC19 was effectively repressed by the hairpin construct, there was still residual pyruvate kinase activity (Figure 4-6 b), allowing non-productive flux modes to operate. This may be attributed to the expression of the CDC19 paralog PYK2 which is transcriptionally de-repressed under low glucose conditions (Boles et al., 1997). The other explanations for the discrepancy between theoretical and actual yields regard the assumptions taken by the modelling approach. Namely, that there is steady-state growth with glucose as the sole carbon source and that enzyme kinetics and product inhibition are not considered. These assumptions are violated by the dynamic nature of the system, the shift from glucose to ethanol consumption, and the fact that little is known about any metabolic perturbations associated with the unique pheromone response phenotype and G1 cell cycle arrest.

101

4.4.5 Conclusions

For metabolites which are not natural products of a host metabolism, there is a fundamental conflict between growth and production. The dynamic regulation of metabolism is a promising approach for separating growth from production (Anesiadis et al., 2008), which can be used to implement genetic modifications that cannot be implemented constitutively. The principal barrier to the realisation of dynamic metabolic control in yeast has been the lack of suitable mechanisms for both activating and repressing gene expression. We have demonstrated a system that is capable of achieving these objectives to effectively implement dynamic control.

The QS-linked RNAi circuit facilitates the separation of growth and production phases, allowing both the expression and/or repression of genes that would normally prevent biomass accumulation (and hence product titer if controlled constitutively). Using this system, populations maintain normal growth prior to the activation of quorum sensing and the cell-cycle arrested production phase. When shikimate pathway modifications were successively coupled to quorum sensing, a titer of 1.07 mM (148 mg/L) was achieved, representing a 37 fold improvement over the base strain and the highest recorded PHBA level in yeast.

102

5 General Discussion

The goal of metabolic engineering is to manipulate a metabolic network to over-produce a single metabolite of commercial interest. This goal is difficult to achieve because naturally occurring metabolic networks have evolved to facilitate the growth and survival of their genome. Effective growth and survival requires that carbon sources be used to produce a variety of metabolites which serve as the building blocks for biomass, energy carriers, and redox cofactors. This is part of an intricate balance that is coordinated at genetic, transcriptional, and posttranscriptional levels through a myriad of regulatory processes. Unless the metabolite of interest is a natural growth- associated by-product (e.g. ethanol in S. cerevisiae), then there is a fundamental conflict between the natural function of a metabolic network and the objectives of metabolic engineering. Traditional approaches in metabolic engineering involve the static deletion or overexpression of genes relevant to a production pathway of interest in order to channel carbon flux towards product. Many of these genetic modifications reduce the growth rate of the host microbe because they interfere with growth-based metabolism. In addition, many of the most productive modifications which are predicted by in silico modelling are not possible to implement in vivo because they completely eliminate growth. This is a major impediment to the realisation of the potential of metabolic engineering because the ability of a production strain to grow to a high population density is critical to achieving high compound titers, yet so is the ability to effectively divert carbon flux towards product and away from biomass.

This fundamental conflict between growth and production can be alleviated by separating the two modes of metabolism in time and space so that a producing population can first grow to a high density, then switch to production mode. This strategy would allow for genetic modifications with a high metabolic burden to be implemented, and has long been considered a desirable option for bioprocesses (Carter et al., 2012). The problem is that there are very few mechanisms for implementing the dynamic control of gene expression which is necessary to engineer separate growth and production phases, and fewer still which are appropriate in an industrial setting. For example, the most widely used method for conditional gene expression in yeast uses galactose inducible promoters (Lohr et al., 1995). These promoters are not only very leaky but require the presence of prohibitively expensive galactose and the absence of glucose in growth media for activation. The other commonly used induction systems in yeast are equally inappropriate for the separation of growth and production. An ideal dynamic regulatory system would be autonomous and independent of expensive inducers, be easily tuneable to activate gene expression at desired

103 point in a fermentation, and have switch-like activation dynamics. In addition to these properties, the ideal system would enable the separation of growth from production by limiting the flux towards biomass while maintaining an active metabolism which is suitable for metabolite production.

In this thesis the S. cerevisiae pheromone response phenotype was explored as a system which can be used for the dynamic regulation of gene expression and metabolism, and the separation of growth from production. Given that little is known about the metabolic nature of the pheromone response phenotype, and the fact that the characteristic growth arrest could serve as a mechanism for separating metabolic activity from growth, it was first necessary to determine the suitability of the phenotype as a production phase. Chapter 2 accomplished this using extracellular flux analysis in association with transcriptomics which specifically considered transcripts peripheral to the canonical mating phenotype. Somewhat surprisingly, even when cells arrested growth in response to pheromone, the rates of glucose uptake and metabolic end-product secretion were at least as high as those observed in a normal exponentially growing population. This alone strongly suggested that the metabolic phenotype associated with the pheromone response was active, and suitable for metabolic engineering applications. The mapping of transcriptomics data to central carbon metabolism revealed that the pheromone response phenotype has a metabolism which is not only active, but distinct from exponential growth. Specific changes included an up-regulation of storage carbohydrate and glycerol synthesis as well as TCA cycle enzymes, which coincided with the decrease in respiratory quotient calculated from growth kinetics. In order to truly test the pheromone response phenotype for suitability as a production phase for metabolic engineering, production of para-hydroxybenzoic acid was measured in cultures treated with and without pheromone. Although pheromone treated cultures reached a significantly lower population density, absolute PHBA titer was not different. When PHBA production was normalised to population density, it was clear that pheromone treatment resulted in slightly higher PHBA productivity per cell. Together these findings indicated that pheromone treatment results in a distinct metabolic phenotype which is suitable for metabolic engineering applications, with the potential to effectively separate growth from production.

In Chapter 2 the pheromone response was induced and investigated by adding purified mating peptide to cultures. In order to utilise this system effectively as a dynamic regulatory system, it was necessary to engineer an autoinduction mechanism. In this way cultures could make and respond to their own pheromone, and use pheromone concentration as a proxy for population density so that

104 growth and production could be separated. Chapter 3 details how this was achieved by endowing MAT a type cells with the capacity to produce the MATα type pheromone, which they already have the capacity to detect via the MAPK-activating Ste2p membrane receptor. By controlling α- pheromone expression with the pheromone responsive FUS1 promoter, it was possible to control gene expression according to population density through a positive feedback loop. Although the positive feedback design was fully autonomous, and was used to successfully convert pheromone communication into a synthetic quorum sensing module, it had several major drawbacks. Firstly, the positive feedback topology resulted in a highly sensitive system where the quorum sensing response was fully activated even at relatively low population densities. Although this situation was partially resolved by using a more tightly regulated version of the pheromone responsive promoter (pFUS1J2), the system was still unsuitable for application to control compound production due to the low population density at which gene expression was activated. Secondly, the circuit was always active, even in pre-cultures. As mentioned previously, part of the pheromone response is to arrest growth, and the fact that this response was being activated early in the growth phase meant that the growth rate was extremely slow. This situation arose because the system required growth to fully activate the quorum sensing response, and a sub-set of the population was arresting growth at a relatively low population density, thereby slowing the attainment of a high population density. The result of this situation was a negative feedback loop which overlaid the engineered positive feedback, causing graded GFP expression dynamics instead of the desirable switch-like response which was originally envisioned. Furthermore, if the circuit were used to control pathway modifications which carried a severe metabolic burden instead of the innocuous GFP gene, then their activation in pre-culture would render large scale production impossible. The solution to these problems was to make pheromone expression conditional on another environmental factor. This was achieved by using the ARO9 promoter to control pheromone expression so that quorum sensing would not occur unless an aromatic amino acid was present in the growth media. This effectively created an AND logic gate where gene expression was not fully activated unless both aromatic amino acids AND high population density conditions were encountered. Using this system, populations could be precultured without the quorum sensing circuit activated, and the concentration and type of aromatic amino acid could be used to fine-tune the timing and dynamics of gene expression in main cultures. This allowed development of tailored response patterns, from slow and graded to sharp and switch-like (Figure 5-1).

105

Figure 5-1. Summary of circuit-dependent quorum sensing gene expression dynamics from Chapter 3

The gene expression activation dynamics of the positive feedback (pFUS1-alpha) and aromatic amino acid regulated (pARO9-alpha) quorum sensing circuits are shown with log normalised population density (x-axis) plotted against GFP expression per cell (y-axis). The bottom left graph shows the GFP expression dynamics when quorum sensing is initiated at a low population density of 0.02

OD660nm, while the bottom right graph shows the same strain with quorum sensing activated at a high population density of OD660nm 1.5.

Having established the suitability of the pheromone response for metabolic engineering, and developed pheromone communication into an effective quorum sensing circuit, the system was applied to achieve dynamic control of a metabolic pathway (Chapter 4). PHBA production from the shikimate pathway was chosen as a proof of concept system for dynamic regulation for two reasons. The pheromone response phenotype had already proven effective at producing PHBA (Chapter 2),

106 and the pathway provided opportunities for implementing genetic modifications which severely limited growth, after a growth phase. Previous research had shown that deleting the ARO7 gene greatly improved PHBA titer, but resulted in a growth rate half that of the wild-type even when the metabolites downstream of ARO7 were added to the media (Krömer et al., 2012). Furthermore, elementary flux mode analysis predicted that the deletion of the genes encoding the pyruvate kinase reaction (CDC19 and PYK2) would dramatically improve the yield of PHBA. However, CDC19 is an essential gene under normal conditions and deletion completely eliminates growth on glucose (Sprague, 1977), and PYK2 is repressed when glucose is abundant (Boles et al., 1997). These features of the pathway made it an ideal candidate to demonstrate the merits of using dynamic regulation. The synthetic quorum sensing circuits developed in Chapter 3 were capable of switching on gene expression, but had no mechanism for switching off genes of interest. To provide a dynamic alternative to traditional gene deletion, a heterologous RNA interference module was linked to the quorum sensing circuit. This system was used to repress ARO7 and CDC19 expression levels after a growth phase. Along with the expression of other pathway enzymes using the quorum sensing circuit this approach resulted in a dramatic improvement in PHBA titer (providing the maximum titres currently reported in yeast) and enabled dynamic control of flux through the shikimate pathway.

An important issue facing the field of synthetic biology is the stability of genetic circuits throughout long-term growth of engineered microbes. The retention of synthetic DNA over long periods will be critical to the effective use of engineered organisms in the environment, and for the sustained productivity of microbial cell factories which are used in industrial biotechnology (Moser et al., 2012). Genetic stability of engineered functions can be an issue during clonal expansion of a population if there is selection pressure against the functions encoded by the synthetic DNA, allowing cells which lose the DNA to outcompete their engineered counterparts. The situation is exacerbated in yeast if there are repetitive sequences used, since this greatly increases the potential for homologous recombination between DNA segments with high similarity and subsequent loss of the intervening sequences. The ‘loop-out’ of DNA between repeated sequences with 100 bp or more of homology occurs in approximately 1 per 100,000 cells in a population (Santos-Rosa & Aguilera, 1994). Repeated sequences are very common in synthetic biology due to the re-use of identical promoter or terminator parts for different genes. The work in this thesis involved the construction of large synthetic DNA clusters with high repetition of FUS1 promoters and CYC1 terminators. Given that the transcriptional units that these repetitive segments were regulating resulted in a high metabolic burden (i.e. repression of CDC19), it there was potential for negative selection pressure

107 to result in the expansion of populations which had looped-out these segments. However, the dynamic regulation of the metabolic burden, and the fact that the growth arrest phenotype occurs during the production phase mean that significant circuit instability is very unlikely with the pheromone QS system. Although this phenomenon was not investigated here, it is an interesting and potentially very valuable aspect of dynamic regulation and the separation of growth from production that could be explored in the future.

There are a limited number of molecular tools available for implementing dynamic regulation in yeast, with most being inadequate for industrial application. The commonly used galactose induction system (Lohr et al., 1995) has a noise level which is unacceptable for some applications such as controlling RNA interference (Drinnenberg et al., 2009). Furthermore, galactose is prohibitively expensive for use at an industrial scale (Westfall et al., 2012). Previous efforts to implement dynamic regulation under industrially relevant conditions required the knockout of galactose utilization genes so that a small amount of galactose can be added to fermentations to gratuitously induce gene expression (Westfall et al., 2012). High concentrations of fermentable carbon sources repress gene expression from galactose promoters (GAL) via a carbon catabolite repression mechanism (Gancedo, 1998). This means that cheap, fermentable sugars such as glucose or sucrose cannot be used during a production phase when the GAL promoters are used to achieve dynamic regulation. To circumvent this limitation ethanol has been used as a carbon source for fed- batch cultures with the GAL promoters (Westfall et al., 2012). However, this is also problematic because ethanol is a carbon source which is more expensive than commonly used sugar feedstocks, and is in fact a common commercial product of industrial yeast fermentations. Other systems such as the doxycycline inducible promoters have the advantage of being completely orthologous to native yeast regulation (Wishart et al., 2005), but are also too expensive to be employed at a large scale. There is therefore a significant requirement for mechanisms which enable dynamic regulation and have suitable properties for industrial use of cell factories such as yeast. For example, an ideal promoter would have low noise, high dynamic range, switch-like activation, high absolute expression levels, sustained induction, and high activity on carbon sources which support high metabolic flux such as glucose or sucrose.

There are many aspects of the quorum sensing system which could be improved upon in the future. The most serious drawback is the requirement of an aromatic amino acid in the growth medium to activate quorum sensing. Most industrial fermentations do not have aromatic amino acids at the concentrations required to activate the quorum sensing circuits presented here. For these systems to

108 be applied at a larger scale a cheaper activation mechanism would need to be explored. Ongoing work will focus on characterising promoters which are activated in the presence of sucrose and repressed in the presence of glucose. Sucrose is a preferred and cost-effective feedstock for industrial scale fermentations for a variety of reasons (Bruschi et al., 2012, Sabri et al., 2013), and promoters that can be used during a sucrose-fed production phase are therefore highly desirable. These promoters would in theory allow for the tuning of gene expression dynamics according to the relative concentrations of glucose and sucrose in the growth medium. Literature suggests that the ARO9 promoter, used here to activate pheromone production, is induced at high cell densities (Chen & Fink, 2006). An alternative approach is to investigate the possibility of high cell density activation of the ARO9 promoter as an induction mechanism. Future work will also explore the possibility of improving the dynamic range of the system. At present there is a significant level of leaky repression of RNAi-targeted genes, which occurs even without quorum sensing activated. A way to circumvent this would be to change the expression of the RNAi enzymes Argonaute and Dicer from strong-constitutive (currently controlled by the TEF1 promoter) to dynamic by placing them under the control of the pheromone responsive FUS1 promoter.

Despite the imperfect nature of the synthetic circuits which are presented in this work, the dynamic control of PHBA production using the synthetic quorum sensing circuit represents a proof-of- concept solution to some of the biggest problems which face the field of metabolic engineering. The conflict between the evolved function of metabolic networks and the objectives of metabolic engineering are epitomised by reaction knockouts which are predicted to enhance productivity, but which aren’t possible because they require the deletion of essential genes. Only through the dynamic control of gene expression and the separation of growth from production can these essential genes be repressed and the corresponding productive modes of metabolism accessed. The dynamic control systems developed in this thesis represent the most effective systems available for dynamic gene silencing in yeast, and the only example of a system where high metabolic activity is decoupled from normal cell growth and division. The application of quorum sensing to metabolic engineering also has other novel features which are hugely advantageous to industrial bioprocesses. For example, there is a natural phenotypic heterogeneity even in isogenic microbial populations which along with the variable environmental conditions in a large bioreactor can contribute to vast variations in cellular performance across a population (Lara et al., 2006). This means that some cells in a population are producing as intended, while others have relatively low production levels, and due to lack of metabolic burden can grow more rapidly and take over the population. In contrast, a quorum sensing population reaches a critical population density and then via cell-to-cell

109 communication coordinates the activation of a population-wide production phase. This mass- coordination of metabolism can ensure that there are no ‘cheaters’ in the population and that the conversion of sugar to product is maximised during the production phase.

The ultimate synergy between synthetic biology and metabolic engineering would be the design and synthesis of tailored, streamlined whole genomes with the goal of over-producing a single metabolite (Vickers et al., 2010). This approach would require careful balancing of carbon, energy carriers, and redox cofactors to facilitate high flux through the pathway of interest. Such an approach would couple metabolite production to growth so that very high yields can be obtained as is the case for ethanol production in yeast. However, there would still be an optimal biomass concentration at which it would be advantageous to cease growth and direct a greater proportion of carbon towards product. Therefore dynamic regulation, not only of metabolism but of growth associated processes such as ribosome biogenesis and cell division would need to be a principal feature of whole genome design. Even in such a ‘futuristic’ scenario it is not difficult to envision the utility of a simple regulatory motif such as quorum sensing being the mechanism for implementing such dynamic regulation. Quorum sensing is a seemingly ubiquitous microbiological phenomenon which has conveyed and maintained selective advantages to each of the three domains of life (Williams, 2007). The evolutionary success of quorum sensing suggests that it will find great utility in synthetic biological systems of the future.

110

6 Bibliography

Anderson JC, Clarke EJ, Arkin AP & Voigt CA (2006) Environmentally Controlled Invasion of Cancer Cells by Engineered Bacteria. Journal of Molecular Biology 355: 619-627.

Andrianantoandro E, Basu S, Karig DK & Weiss R (2006) Synthetic biology: new engineering rules for an emerging discipline. Mol Syst Biol 2.

Anesiadis N, Cluett WR & Mahadevan R (2008) Dynamic metabolic engineering for increasing bioprocess productivity. Metabolic Engineering 10: 255-266.

Babiskin AH & Smolke CD (2011) Engineering ligand-responsive RNA controllers in yeast through the assembly of RNase III tuning modules. Nucleic Acids Research.

Baltanas R, Bush A, Couto A, Durrieu L, Hohmann S & Colman-Lerner A (2013) Pheromone- Induced Morphogenesis Improves Osmoadaptation Capacity by Activating the HOG MAPK Pathway. Sci Signal 6: ra26-.

Bardwell L (2005) A walk-through of the yeast mating pheromone response pathway. Peptides 26: 339-350.

Barker JL & Frost JW (2001) Microbial synthesis of p-hydroxybenzoic acid from glucose. Biotechnology and Bioengineering 76: 376-390.

Bashor CJ, Helman NC, Yan S & Lim WA (2008) Using Engineered Scaffold Interactions to Reshape MAP Kinase Pathway Signaling Dynamics. Science 319: 1539-1543.

Beisel CL & Smolke CD (2009) Design Principles for Riboswitch Function. Plos Computational Biology 5.

Bentley WE, Mirjalili N, Andersen DC, Davis RH, Kompala DS & William E. Bentley Ib (2009) Plasmid-encoded protein: The principal factor in the “metabolic burden” associated with recombinant bacteria. Biotechnology and Bioengineering 102: 1283-1297.

Bhattacharyya RP, Reményi A, Good MC, Bashor CJ, Falick AM & Lim WA (2006) The Ste5 Scaffold Allosterically Modulates Signaling Output of the Yeast Mating Pathway. Science 311: 822-826.

Bi J-X, Shuttleworth J & Al-Rubeai M (2004) Uncoupling of cell growth and proliferation results in enhancement of productivity in p21CIP1-arrested CHO cells. Biotechnology and Bioengineering 85: 741-749.

Blank LM & Sauer U (2004) TCA cycle activity in Saccharomyces cerevisiae is a function of the environmentally determined specific growth and glucose uptake rates. Microbiology 150: 1085- 1093.

111

Blount BA, Weenink T & Ellis T (2012) Construction of synthetic regulatory networks in yeast. FEBS letters 586: 2112-2121.

Boles E, Schulte F, Miosga T, Freidel K, Schlüter E, Zimmermann FK, Hollenberg CP & Heinisch JJ (1997) Characterization of a glucose-repressed pyruvate kinase (Pyk2p) in Saccharomyces cerevisiae that is catalytically insensitive to fructose-1,6-bisphosphate. Journal of Bacteriology 179: 2987-2993.

Bruschi M, Boyes SJ, Sugiarto H, Nielsen LK & Vickers CE (2012) A transferable sucrose utilization approach for non-sucrose-utilizing Escherichia coli strains. Biotechnology Advances 30: 1001-1010.

Busti S, Gotti L, Balestrieri C, Querin L, Drovandi G, Felici G, Mavelli G, Bertolazzi P, Alberghina L & Vanoni M (2012) Overexpression of Far1, a cyclin-dependent kinase inhibitor, induces a large transcriptional reprogramming in which RNA synthesis senses Far1 in a Sfp1-mediated way. Biotechnology Advances 30: 185-201.

Carter KK, Valdes JJ & Bentley WE (2012) Pathway engineering via quorum sensing and sRNA riboregulators—Interconnected networks and controllers. Metabolic Engineering.

Caspi R, Altman T, Billington R, et al. (2014) The MetaCyc database of metabolic pathways and enzymes and the BioCyc collection of Pathway/Genome Databases. Nucleic Acids Res 42: D459- 471.

Chan RK & Otte CA (1982) Physiological characterization of Saccharomyces cerevisiae mutants supersensitive to G1 arrest by a factor and alpha factor pheromones. Mol Cell Biol 2: 21-29.

Chandran SS, Kealey JT & Reeves CD (2011) Microbial production of isoprenoids. Process Biochemistry 46: 1703-1710.

Chang AL, Wolf JJ & Smolke CD (2012) Synthetic RNA switches as a tool for temporal and spatial control over gene expression. Current Opinion in Biotechnology.

Chang F & Herskowitz I (1990) Identification of a gene necessary for cell cycle arrest by a negative growth factor of yeast: FAR1 is an inhibitor of a G1 cyclin, CLN2. Cell 63: 999-1011.

Chasse SA, Flanary P, Parnell SC, Hao N, Cha JY, Siderovski DP & Dohlman HG (2006) Genome- Scale Analysis Reveals Sst2 as the Principal Regulator of Mating Pheromone Signaling in the Yeast Saccharomyces cerevisiae. Eukaryotic Cell 5: 330-346.

Chen H & Fink GR (2006) Feedback control of morphogenesis in fungi by aromatic alcohols. Genes & Development 20: 1150-1161.

Chen M-T & Weiss R (2005) Artificial cell-cell communication in yeast Saccharomyces cerevisiae using signaling elements from Arabidopsis thaliana. Nat Biotech 23: 1551-1555.

112

Cherry JM, Hong EL, Amundsen C, et al. (2012) Saccharomyces Genome Database: the genomics resource of budding yeast. Nucleic Acids Res 40: D700-705.

Choudhary S & Schmidt-Dannert C (2010) Applications of quorum sensing in biotechnology. Applied Microbiology and Biotechnology 86: 1267-1279.

Culler SJ, Hoff KG & Smolke CD (2010) Reprogramming Cellular Behavior with RNA Controllers Responsive to Endogenous Proteins. Science 330: 1251-1255.

Curran KA, Leavitt JM, Karim AS & Alper HS (2013) Metabolic engineering of muconic acid production in Saccharomyces cerevisiae. Metabolic Engineering 15: 55-66.

Da Silva NA & Srikrishnan S (2012) Introduction and expression of genes for metabolic engineering applications in Saccharomyces cerevisiae. FEMS Yeast Research 12: 197-214.

Dawkins R (2006) The selfish gene. OUP Oxford.

Dietmair S, Timmins NE, Gray PP, Nielsen LK & Krömer JO (2010) Towards quantitative metabolomics of mammalian cells: Development of a metabolite extraction protocol. Analytical Biochemistry 404: 155-164.

Dijken J, Weusthuis R & Pronk J (1993) Kinetics of growth and sugar consumption in yeasts. Antonie van Leeuwenhoek 63: 343-352.

Dohlman HG, Song J, Ma D, Courchesne WE & Thorner J (1996) Sst2, a negative regulator of pheromone signaling in the yeast Saccharomyces cerevisiae: expression, localization, and genetic interaction and physical association with Gpa1 (the G-protein alpha subunit). Molecular and Cellular Biology 16: 5194-5209.

Drinnenberg IA, Fink GR & Bartel DP (2011) Compatibility with Killer Explains the Rise of RNAi-Deficient Fungi. Science 333: 1592.

Drinnenberg IA, Weinberg DE, Xie KT, Mower JP, Wolfe KH, Fink GR & Bartel DP (2009) RNAi in Budding Yeast. Science 326: 544-550.

Dueber JE, Wu GC, Malmirchegini GR, Moon TS, Petzold CJ, Ullal AV, Prather KLJ & Keasling JD (2009) Synthetic protein scaffolds provide modular control over metabolic flux. Nat Biotech 27: 753-759.

Elion EA, Satterberg B & Kranz JE (1993) FUS3 phosphorylates multiple components of the mating signal transduction cascade: evidence for STE12 and FAR1. Molecular Biology of the Cell 4: 495-510.

Elowitz MB & Leibler S (2000) A synthetic oscillatory network of transcriptional regulators. Nature 403: 335-338.

Endy D (2005) Foundations for engineering biology. Nature 438: 449-453. 113

Etcheverry T (1990) Induced expression using yeast copper metallothionein promoter. Methods in Enzymology, Vol. Volume 185 (David VG, ed.) p.^pp. 319-329. Academic Press.

Farmer WR & Liao JC (2000) Improving lycopene production in Escherichia coli by engineering metabolic control. Nat Biotech 18: 533-537.

Fussenegger M, Schlatter S, Datwyler D, Mazur X & Bailey JE (1998) Controlled proliferation by multigene metabolic engineering enhances the productivity of Chinese hamster ovary cells. Nat Biotechnol 16: 468-472.

Gadkar KG, Doyle Iii FJ, Edwards JS & Mahadevan R (2005) Estimating optimal profiles of genetic alterations using constraint-based models. Biotechnology and Bioengineering 89: 243-251.

Gancedo JM (1998) Yeast Carbon Catabolite Repression. Microbiology and Molecular Biology Reviews 62: 334-361.

Gardner TS, Cantor CR & Collins JJ (2000) Construction of a genetic toggle switch in Escherichia coli. Nature 403: 339-342.

Giaever G, Chu A, Ni L, et al. (2002) Functional profiling of the Saccharomyces cerevisiae genome. Nature 418: 387-391.

Gibson DG, Young L, Chuang R-Y, Venter JC, Hutchison CA & Smith HO (2009) Enzymatic assembly of DNA molecules up to several hundred kilobases. Nat Meth 6: 343-345.

Gibson DG, Benders GA, Andrews-Pfannkoch C, et al. (2008) Complete Chemical Synthesis, Assembly, and Cloning of a Mycoplasma genitalium Genome. Science 319: 1215-1220.

Gibson DG, Glass JI, Lartigue C, et al. (2010) Creation of a Bacterial Cell Controlled by a Chemically Synthesized Genome. Science 329: 52-56.

Gietz RD & Schiestl RH (2007) Quick and easy yeast transformation using the LiAc/SS carrier DNA/PEG method. Nat Protocols 2: 35-37.

Graf R, Mehmann B & Braus GH (1993) Analysis of feedback-resistant anthranilate synthases from Saccharomyces cerevisiae. Journal of Bacteriology 175: 1061-1068.

Greenbaum D, Colangelo C, Williams K & Gerstein M (2003) Comparing protein abundance and mRNA expression levels on a genomic scale. Genome Biology 4: 117.

Groß A, Rödel G & Ostermann K (2011) Application of the yeast pheromone system for controlled cell–cell communication and signal amplification. Letters in Applied Microbiology 52: 521-526.

Gruhler A, Olsen JV, Mohammed S, Mortensen P, Færgeman NJ, Mann M & Jensen ON (2005) Quantitative Phosphoproteomics Applied to the Yeast Pheromone Signaling Pathway. Molecular & Cellular Proteomics 4: 310-327.

114

Güldener U, Heck S, Fiedler T, Beinhauer J & Hegemann JH (1996) A New Efficient Gene Disruption Cassette for Repeated Use in Budding Yeast. Nucleic Acids Research 24: 2519-2524.

Gunawardena U, Meinhold P, Peters MW, Urano J & Feldman RRM (2010) BUTANOL PRODUCTION BY METABOLICALLY ENGINEERED YEAST. Vol. 20100062505 p.^pp. United States Gevo, Inc. (Pasadena, CA, US), U.S.A.

Hadicke O & Klamt S (2011) Computing complex metabolic intervention strategies using constrained minimal cut sets. Metab Eng 13: 204-213.

Hagen DC, McCaffrey G & Sprague GF, Jr (1991) Pheromone response elements are necessary and sufficient for basal and pheromone-induced transcription of the FUS1 gene of Saccharomyces cerevisiae. Mol Cell Biol 11: 2952-2961.

Hansen AC, Zhang Q & Lyne PWL (2005) Ethanol–diesel fuel blends –– a review. Bioresource Technology 96: 277-285.

Haynes KA & Silver PA (2009) Eukaryotic systems broaden the scope of synthetic biology. Journal of Cell Biology 187: 589-596.

Hedegaard K, Thyø KA & Wenzel H (2008) Life Cycle Assessment of an Advanced Bioethanol Technology in the Perspective of Constrained Biomass Availability. Environmental Science & Technology 42: 7992-7999.

Herskowitz I (1988) Life cycle of the budding yeast Saccharomyces cerevisiae. Microbiological Reviews 52: 536-553.

Heyland J, Fu J & Blank LM (2009) Correlation between TCA cycle flux and glucose uptake rate during respiro-fermentative growth of Saccharomyces cerevisiae. Microbiology 155: 3827-3837.

Ho SN, Hunt HD, Horton RM, Pullen JK & Pease LR (1989) Site-directed mutagenesis by overlap extension using the polymerase chain reaction. Gene 77: 51-59.

Holtz WJ & Keasling JD (2010) Engineering Static and Dynamic Control of Synthetic Pathways. Cell 140: 19-23.

Horton RM, Hunt HD, Ho SN, Pullen JK & Pease LR (1989) Engineering hybrid genes without the use of restriction enzymes: gene splicing by overlap extension. Gene 77: 61-68.

Ingolia NT & Murray AW (2007) Positive-Feedback Loops as a Flexible Biological Module. Current Biology 17: 668-677.

Iraqui I, Vissers S, Cartiaux M & Urrestarazu A (1998) Characterisation of Saccharomyces cerevisiae ARO8 and ARO9 genes encoding aromatic aminotransferases I and II reveals a new aminotransferase subfamily. Molecular and General Genetics MGG 257: 238-248.

115

Iraqui I, Vissers S, Andre B & Urrestarazu A (1999) Transcriptional Induction by Aromatic Amino Acids in Saccharomyces cerevisiae. Mol Cell Biol 19: 3360-3371.

Keasling JD (2008) Synthetic Biology for Synthetic Chemistry. ACS Chemical Biology 3: 64-76.

Keasling JD (2010) Manufacturing Molecules Through Metabolic Engineering. Science 330: 1355- 1358.

Keasling JD (2012) Synthetic biology and the development of tools for metabolic engineering. Metabolic Engineering 14: 189-195.

Kemp HA & Sprague J, George F. (2003) Far3 and Five Interacting Proteins Prevent Premature Recovery from Pheromone Arrest in the Budding Yeast Saccharomyces cerevisiae. Molecular and Cellular Biology 23: 1750-1763.

Khalil AS & Collins JJ (2010) Synthetic biology: applications come of age. Nat Rev Genet 11: 367- 379.

Khoo SHG & Al-Rubeai M (2009) Metabolic characterization of a hyper-productive state in an antibody producing NS0 myeloma cell line. Metabolic Engineering 11: 199-211.

Kinsey PT & Sandmeyer SB (1995) TY3 TRANSPOSES IN MATING POPULATIONS OF YEAST - A NOVEL TRANSPOSITION ASSAY FOR TY3. Genetics 139: 81-94.

Köhle A, Sommer S, Li S-M, Schilde-Rentschler L, Ninnemann H & Heide L (2003) Secondary metabolites in transgenic tobacco and potato: high accumulation of 4-hydroxybenzoic acid glucosides results from high expression of the bacterial gene ubiC. Molecular Breeding 11: 15-24.

Krömer JO, Nunez-Bernal D, Averesch NJH, Hampe J, Varela J & Varela C (2012) Production of aromatics in Saccharomyces cerevisiae—A feasibility study. Journal of Biotechnology.

Kumar N, Gammell P & Clynes M (2007) Proliferation control strategies to improve productivity and survival during CHO based production culture. Cytotechnology 53: 33-46.

Lara A, Galindo E, Ramírez O & Palomares L (2006) Living with heterogeneities in bioreactors. Molecular Biotechnology 34: 355-381.

Lee SK, Chou H, Ham TS, Lee TS & Keasling JD (2008) Metabolic engineering of microorganisms for biofuels production: from bugs to synthetic biology to fuels. Current Opinion in Biotechnology 19: 556-563.

Lohr D, Venkov P & Zlatanova J (1995) Transcriptional regulation in the yeast GAL gene family: a complex genetic network. The FASEB Journal 9: 777-787.

Lu TK, Khalil AS & Collins JJ (2009) Next-generation synthetic gene networks. Nat Biotech 27: 1139-1150.

116

Lucero P, Peñalver E, Moreno E & Lagunas R (2000) Internal Trehalose Protects Endocytosis from Inhibition by Ethanol in Saccharomyces cerevisiae. Applied and Environmental Microbiology 66: 4456-4461.

Luttik MAH, Vuralhan Z, Suir E, Braus GH, Pronk JT & Daran JM (2008) Alleviation of feedback inhibition in Saccharomyces cerevisiae aromatic amino acid biosynthesis: Quantification of metabolic impact. Metabolic Engineering 10: 141-153.

Malleshaiah MK, Shahrezaei V, Swain PS & Michnick SW (2010) The scaffold protein Ste5 directly controls a switch-like mating decision in yeast. Nature 465: 101-105.

March JC & Bentley WE (2004) Quorum sensing and bacterial cross-talk in biotechnology. Current Opinion in Biotechnology 15: 495-502.

Marsh L, Neiman AM & Herskowitz I (1991) Signal Transduction During Pheromone Response in Yeast. Annual Review of Cell Biology 7: 699-728.

Mateus C & Avery SV (2000) Destabilized green fluorescent protein for monitoring dynamic changes in yeast gene expression with flow cytometry. Yeast 16: 1313-1323.

Matsushika A, Inoue H, Kodaki T & Sawayama S (2009) Ethanol production from xylose in engineered <i>Saccharomyces cerevisiae</i> strains: current state and perspectives. Applied Microbiology and Biotechnology 84: 37-53.

McIsaac RS, Silverman SJ, McClean MN, Gibney PA, Macinskas J, Hickman MJ, Petti AA & Botstein D (2011) Fast-acting and nearly gratuitous induction of gene expression and protein depletion in Saccharomyces cerevisiae. Molecular Biology of the Cell 22: 4447-4459.

McQualter RB, Chong BF, Meyer K, Van Dyk DE, O'Shea MG, Walton NJ, Viitanen PV & Brumbley SM (2005) Initial evaluation of sugarcane as a production platform for p-hydroxybenzoic acid. Plant Biotechnology Journal 3: 29-41.

Meijnen J-P, Verhoef S, Briedjlal A, Winde J & Ruijssenaars H (2011) Improved p- hydroxybenzoate production by engineered Pseudomonas putida S12 by using a mixed-substrate feeding strategy. Applied Microbiology and Biotechnology 90: 885-893.

Minard KI, Jennings GT, Loftus TM, Xuan D & McAlister-Henn L (1998) Sources of NADPH and Expression of Mammalian NADP+-specific Isocitrate Dehydrogenases inSaccharomyces cerevisiae. Journal of Biological Chemistry 273: 31486-31493.

Miozzari G, Niederberger P & Hütter R (1978) Tryptophan biosynthesis in Saccharomyces cerevisiae: control of the flux through the pathway. Journal of Bacteriology 134: 48-59.

Mitrophanov AY & Groisman EA (2008) Positive feedback in cellular control systems. BioEssays 30: 542-555.

117

Moser F, Broers NJ, Hartmans S, Tamsir A, Kerkman R, Roubos JA, Bovenberg R & Voigt CA (2012) Genetic Circuit Performance under Conditions Relevant for Industrial Bioreactors. ACS Synthetic Biology 1: 555-564.

Müller R, Wagener A, Schmidt K & Leistner E (1995) Microbial production of specifically ring- 13C-labelled 4-hydroxybenzoic acid. Applied Microbiology and Biotechnology 43: 985-988.

Nakayama N, Miyajima A & Arai K (1987) Common signal transduction system shared by STE2 and STE3 in haploid cells of Saccharomyces cerevisiae: autocrine cell-cycle arrest results from forced expression of STE2. EMBO J 6: 249-254.

Nandagopal N & Elowitz MB (2011) Synthetic Biology: Integrated Gene Circuits. Science 333: 1244-1248.

Nevoigt E (2008) Progress in Metabolic Engineering of Saccharomyces cerevisiae. Microbiol Mol Biol Rev 72: 379-412.

Nevoigt E, Fischer C, Mucha O, Matthäus F, Stahl U & Stephanopoulos G (2007) Engineering promoter regulation. Biotechnology and Bioengineering 96: 550-558.

Nielsen LK (2011) Metabolic engineering: From retrofitting to green field. Nat Chem Biol 7: 408- 409.

Nikolau BJ, Perera MADN, Brachova L & Shanks B (2008) Platform biochemicals for a biorenewable chemical industry. The Plant Journal 54: 536-545.

Nogae I & Johnston M (1990) Isolation and characterization of the ZWF1 gene of Saccharomyces cerevisiae, encoding glucose-6-phosphate dehydrogenase. Gene 96: 161-169.

O’Rourke SM & Herskowitz I (1998) The Hog1 MAPK prevents cross talk between the HOG and pheromone response MAPK pathways in Saccharomyces cerevisiae. Genes & Development 12: 2874-2886.

Oliveira AP, Ludwig C, Picotti P, Kogadeeva M, Aebersold R & Sauer U (2012) Regulation of yeast central metabolism by enzyme phosphorylation. Mol Syst Biol 8: 623.

Paliwal S, Iglesias PA, Campbell K, Hilioti Z, Groisman A & Levchenko A (2007) MAPK- mediated bimodal gene expression and adaptive gradient sensing in yeast. Nature 446: 46-51.

Pereira M, Eleutherio E & Panek A (2001) Acquisition of tolerance against oxidative damage in Saccharomyces cerevisiae. BMC Microbiology 1: 11.

Philips J & Herskowitz I (1997) Osmotic Balance Regulates Cell Fusion during Mating in Saccharomyces cerevisiae. The Journal of Cell Biology 138: 961-974.

118

Poritz MA, Malmstrom S, Kim MKH, Rossmeissl PJ & Kamb A (2001) Graded mode of transcriptional induction in yeast pheromone signalling revealed by single-cell analysis. Yeast 18: 1331-1338.

Purnick PEM & Weiss R (2009) The second wave of synthetic biology: from modules to systems. Nat Rev Mol Cell Biol 10: 410-422.

Regot S, Macia J, Conde N, Furukawa K, Kjellen J, Peeters T, Hohmann S, de Nadal E, Posas F & Sole R (2011) Distributed biological computation with multicellular engineered networks. Nature 469: 207-211.

Rivers DM & Sprague GF (2003) Autocrine activation of the pheromone response pathway in mata2 - cells is attenuated by SST2- and ASG7-dependent mechanisms. Molecular Genetics and Genomics 270: 225-233.

Ro D-K, Paradise EM, Ouellet M, et al. (2006) Production of the antimalarial drug precursor artemisinic acid in engineered yeast. Nature 440: 940-943.

Roberts CJ, Nelson B, Marton MJ, et al. (2000) Signaling and Circuitry of Multiple MAPK Pathways Revealed by a Matrix of Global Gene Expression Profiles. Science 287: 873-880.

Sabri S, Nielsen LK & Vickers CE (2013) Molecular Control of Sucrose Utilization in Escherichia coli W, an Efficient Sucrose-Utilizing Strain. Applied and Environmental Microbiology 79: 478- 487.

Saeidi N, Wong CK, Lo T-M, Nguyen HX, Ling H, Leong SSJ, Poh CL & Chang MW (2011) Engineering microbes to sense and eradicate Pseudomonas aeruginosa, a human pathogen. Mol Syst Biol 7.

Sambrook J & Russell DW (2001) Molecular Cloning: A Laboratory Manual. Cold Spring Harbor Laboratory Press.

Santos-Rosa H & Aguilera A (1994) Increase in incidence of chromosome instability and non- conservative recombination between repeats in Saccharomyces cerevisiae hpr1Δ strains. Molecular and General Genetics MGG 245: 224-236.

Serikawa KA, Xu XL, MacKay VL, Law GL, Zong Q, Zhao LP, Bumgarner R & Morris DR (2003) The Transcriptome and Its Translation during Recovery from Cell Cycle Arrest in Saccharomyces cerevisiae. Molecular & Cellular Proteomics 2: 191-204.

Shah NA & Sarkar CA (2011) Robust Network Topologies for Generating Switch-Like Cellular Responses. PLoS Comput Biol 7: e1002085.

Si T, Luo Y, Bao Z & Zhao H (2014) RNAi-Assisted Genome Evolution in Saccharomyces cerevisiae for Complex Phenotype Engineering. ACS Synthetic Biology.

119

Siebert M, Severin K Fau - Heide L & Heide L Formation of 4-hydroxybenzoate in Escherichia coli: characterization of the ubiC gene and its encoded enzyme chorismate pyruvate-lyase.

Siebert M, Sommer S, Li S, Wang Z, Severin K & Heide L (1996) Genetic Engineering of Plant Secondary Metabolism (Accumulation of 4-Hydroxybenzoate Glucosides as a Result of the Expression of the Bacterial ubiC Gene in Tobacco). Plant Physiology 112: 811-819.

Siekhaus DE & Drubin DG (2003) Spontaneous receptor-independent heterotrimeric G-protein signalling in an RGS mutant. Nat Cell Biol 5: 231-235.

Sikorski RS & Hieter P (1989) A System of Shuttle Vectors and Yeast Host Strains Designed for Efficient Manipulation of DNA in Saccharomyces cerevisiae. Genetics 122: 19-27.

Siliciano PG & Tatchell K (1986) Identification of the DNA sequences controlling the expression of the MAT alpha locus of yeast. Proceedings of the National Academy of Sciences 83: 2320-2324.

Singh A, Chen EY, Lugovoy JM, Chang CN, Hitzeman RA & Seeburg PH (1983) Saccharomyces cerevisiae contains two discrete genes coding for the alpha-factor pheromone. Nucleic Acids Research 11: 4049-4063.

Slekar KH, Kosman DJ & Culotta VC (1996) The Yeast Copper/Zinc Superoxide Dismutase and the Pentose Phosphate Pathway Play Overlapping Roles in Oxidative Stress Protection. Journal of Biological Chemistry 271: 28831-28836.

Smolke CD & Silver PA (2011) Informing Biological Design by Integration of Systems and Synthetic Biology. Cell 144: 855-859.

Song H, Payne S, Tan C & You L (2011) Programming microbial population dynamics by engineered cell–cell communication. Biotechnology Journal 6: 837-849.

Sprague GF & Herskowitz I (1981) Control of yeast cell type by the mating type locus : I. Identification and control of expression of the a-specific gene BAR1. Journal of Molecular Biology 153: 305-321.

Sprague GF, Jr. (1977) Isolation and characterization of a Saccharomyces cerevisiae mutant deficient in pyruvate kinase activity. J Bacteriol 130: 232-241.

Stansfield I & Stark MJ (2007) Yeast gene analysis second edition. Elsevier.

Stephanopoulos G (2012) Synthetic Biology and Metabolic Engineering. ACS Synthetic Biology 1: 514-525.

Stephanopoulos G, Aristidou AA & Nielsen J (1998) Metabolic Engineering: Principles and Methodologies. Elsevier Science.

Stoltenburg R, Reinemann C & Strehlitz B (2007) SELEX--A (r)evolutionary method to generate high-affinity nucleic acid ligands. Biomolecular Engineering 24: 381-403. 120

Tan C, Marguet P & You L (2009) Emergent bistability by a growth-modulating positive feedback circuit. Nat Chem Biol 5: 842-848.

Terzer M & Stelling J (2008) Large-scale computation of elementary flux modes with bit pattern trees. Bioinformatics 24: 2229-2235.

Teste M-A, Duquenne M, Francois J & Parrou J-L (2009) Validation of reference genes for quantitative expression analysis by real-time RT-PCR in Saccharomyces cerevisiae. BMC Molecular Biology 10: 99.

Thevelein JM (1994) Signal transduction in yeast. Yeast 10: 1753-1790.

Trueheart J, Boeke J & Fink G (1987) Two genes required for cell fusion during yeast conjugation: evidence for a pheromone-induced surface protein. Molecular and Cellular Biology 7: 2316-2328.

Tsao C-Y, Hooshangi S, Wu H-C, Valdes JJ & Bentley WE (2010) Autonomous induction of recombinant proteins by minimally rewiring native quorum sensing regulon of E. coli. Metabolic Engineering 12: 291-297. van Dijken JP, Bauer J, Brambilla L, et al. (2000) An interlaboratory comparison of physiological and genetic properties of four Saccharomyces cerevisiae strains. Enzyme and Microbial Technology 26: 706-714.

Van Driessche B, Tafforeau L, Hentges P, Carr AM & Vandenhaute J (2005) Additional vectors for PCR-based gene tagging in Saccharomyces cerevisiae and Schizosaccharomyces pombe using nourseothricin resistance. Yeast 22: 1061-1068.

Verduyn C, Postma E, Scheffers WA & Van Dijken JP (1992) Effect of benzoic acid on metabolic fluxes in yeasts: A continuous-culture study on the regulation of respiration and alcoholic fermentation. Yeast 8: 501-517.

Verhoef S, Ballerstedt H, Volkers RM, Winde J & Ruijssenaars H (2010) Comparative transcriptomics and proteomics of p-hydroxybenzoate producing Pseudomonas putida S12: novel responses and implications for strain improvement. Applied Microbiology and Biotechnology 87: 679-690.

Vickers CE, Blank LM & Kromer JO (2010) Grand Challenge Commentary: Chassis cells for industrial biochemical production. Nat Chem Biol 6: 875-877.

Warner JR, Vilardell J & Sohn JH (2001) Economics of Ribosome Biosynthesis. Cold Spring Harbor Symposia on Quantitative Biology 66: 567-574.

Watanabe S, Shuttleworth J & Al-Rubeai M (2002) Regulation of cell cycle and productivity in NS0 cells by the over-expression of p21CIP1. Biotechnology and Bioengineering 77: 1-7.

Waters CM & Bassler BL (2005) QUORUM SENSING: Cell-to-Cell Communication in Bacteria. Annual Review of Cell and Developmental Biology 21: 319-346. 121

Wei P, Wong WW, Park JS, Corcoran EE, Peisajovich SG, Onuffer JJ, Weiss A & Lim WA (2012) Bacterial virulence proteins as tools to rewire kinase pathways in yeast and immune cells. Nature 488: 384-388.

Werner-Washburne M, Braun E, Johnston GC & Singer RA (1993) Stationary phase in the yeast Saccharomyces cerevisiae. Microbiol Mol Biol Rev 57: 383-401.

Westfall PJ, Pitera DJ, Lenihan JR, et al. (2012) Production of amorphadiene in yeast, and its conversion to dihydroartemisinic acid, precursor to the antimalarial agent artemisinin. Proceedings of the National Academy of Sciences.

Whited GM, Feher FJ, Benko DA, Cervin MA, Chotani GK, McAuliffe JC, LaDuca RJ, Ben- Shoshan EA & Sanford KJ (2010) TECHNOLOGY UPDATE: Development of a gas-phase bioprocess for isoprene-monomer production using metabolic pathway engineering. Industrial Biotechnology 6: 152-163.

Whiteway M, Hougan L & Thomas DY (1988) Expression of MFα1 in MATa cells supersensitive to α-factor leads to self-arrest. Molecular and General Genetics MGG 214: 85-88.

Williams P (2007) Quorum sensing, communication and cross-kingdom signalling in the bacterial world. Microbiology 153: 3923-3938.

Williams P, Winzer K, Chan WC & Cámara M (2007) Look who's talking: communication and quorum sensing in the bacterial world. Philosophical Transactions of the Royal Society B: Biological Sciences 362: 1119-1134.

Williams TC, Nielsen LK & Vickers CE (2012) Engineered Quorum Sensing Using Pheromone- Mediated Cell-to-Cell Communication in Saccharomyces cerevisiae. ACS Synthetic Biology 2: 136- 149.

Win MN & Smolke CD (2007) A modular and extensible RNA-based gene-regulatory platform for engineering cellular function. Proceedings of the National Academy of Sciences 104: 14283-14288.

Win MN & Smolke CD (2008) Higher-order cellular information processing with synthetic RNA devices. Science 322: 456-460.

Win MN, Liang JC & Smolke CD (2009) Frameworks for Programming Biological Function through RNA Parts and Devices. Chemistry & Biology 16: 298-310.

Winter G, Averesch NJH, Nunez-Bernal D & Krömer JO (2014) In vivo instability of chorismate causes substrate loss during fermentative production of aromatics. Yeast n/a-n/a.

Wishart JA, Hayes A, Wardleworth L, Zhang N & Oliver SG (2005) Doxycycline, the drug used to control the tet-regulatable promoter system, has no effect on global gene expression in Saccharomyces cerevisiae. Yeast 22: 565-569.

122

Wu J, Zhang N, Hayes A, Panoutsopoulou K & Oliver SG (2004) Global analysis of nutrient control of gene expression in Saccharomyces cerevisiae during growth and starvation. Proceedings of the National Academy of Sciences of the United States of America 101: 3148-3153.

Xu P, Bhan N & Koffas MAG (2013) Engineering plant metabolism into microbes: from systems biology to synthetic biology. Current Opinion in Biotechnology 24: 291-299.

Yim H, Haselbeck R, Niu W, et al. (2011) Metabolic engineering of Escherichia coli for direct production of 1,4-butanediol. Nat Chem Biol 7: 445-452.

Zeitlinger J, Simon I, Harbison CT, Hannett NM, Volkert TL, Fink GR & Young RA (2003) Program-Specific Distribution of a Transcription Factor Dependent on Partner Transcription Factor and MAPK Signaling. Cell 113: 395-404.

Zhang N-N, Dudgeon DD, Paliwal S, Levchenko A, Grote E & Cunningham KW (2006) Multiple Signaling Pathways Regulate Yeast Cell Death during the Response to Mating Pheromones. Molecular Biology of the Cell 17: 3409-3422.

123