Experimental and Conceptual Approaches to Studying Bet Hedging in Microorganisms

Colin Scott Maxwell

Department of Biology Duke University

Date: Approved:

Paul M. Magwene, Co-Supervisor

L. Ryan Baugh, Co-Supervisor

Nicolas E. Buchler

Katharina V. Koelle

Daniel J. Lew

Dissertation submitted in partial fulﬁllment of the requirements for the degree of Doctor of Philosophy in the Department of Biology in the Graduate School of Duke University 2016 Abstract Experimental and Conceptual Approaches to Studying Bet Hedging in Microorganisms

Colin Scott Maxwell

Department of Biology Duke University

Date: Approved:

Paul M. Magwene, Co-Supervisor

L. Ryan Baugh, Co-Supervisor

Nicolas E. Buchler

Katharina V. Koelle

Daniel J. Lew

An abstract of a dissertation submitted in partial fulﬁllment of the requirements for the degree of Doctor of Philosophy in the Department of Biology in the Graduate School of Duke University 2016 Copyright c 2016 by Colin Scott Maxwell All rights reserved except the rights granted by the Creative Commons Attribution-Noncommercial Licence Abstract

Bet-hedging strategies are used by organisms to survive in unpredictable environments. To pursue a bet-hedging strategy, an organism must produce multiple phenotypes from a single genotype. What molecular mechanisms allow this to happen? To address this question, I created a synthetic system that displays bet-hedging behavior, and developed a new technique called ‘TrackScar’ to measure the ﬁtness and stress-resistance of individual cells. I found that bet-hedging can be generated by actively sensing the environment, and that bet-hedging strategies based on active sensing need not be metabolically costly. These results suggest that to understand how bet-hedging strategies are produced, microorganisms must be examined in the actual environments that they come from.

iv Contents

Abstract iv

List of Tables ix

List of Figuresx

List of Abbreviations and Symbols xii

Acknowledgements xiii

1 Introduction1

2 Bet Hedging by Active Sensing Need not be Costly4

2.1 Introduction...... 4

2.2 Results...... 7

2.2.1 URA3 (counter-)selection mimics a classic model of bet-hedging7

2.2.2 Plasticity to estradiol can generate a phenotypic trade-oﬀ...9

2.2.3 Sensing an uninformative cue enables a bet-hedging strategy. 10

2.2.4 Bet hedging by plasticity need not carry a cost...... 14

2.2.5 Bet hedging by sensing an uniformative cue can be tuned by mutation...... 16

2.3 Discussion...... 18

2.3.1 Bet hedging by active sensing is plausible...... 19

2.3.2 Bet hedging by active sensing is evolvable...... 20

2.3.3 Bet hedging strategies based on phenotypic plasticity need not be costly...... 21

v 2.3.4 Sensing uninformative cues can serve as the basis of bet-hedging strategies...... 22

2.4 Methods...... 24

2.4.1 Strains and media...... 24

2.4.2 Plasmid construction...... 24

2.4.3 Strain construction...... 26

2.4.4 Flow cytometry and OD measurements...... 28

2.4.5 Fitness and competition experiments in liquid culture..... 28

2.4.6 Competiton experiments on agar plates...... 28

2.4.7 Patterns of growth, image processing, and statistics...... 29

2.5 Strains, primers, and plasmids used in this study...... 30

3 Known Unknowns: Sensing Uninformative Cues can Generate Bet- hedging Strategies 34

3.1 Introduction...... 34

3.2 Bet hedging in microbes and stochastic gene expression...... 36

3.3 Bet hedging can be generated by either environmental or physiological noise...... 37

3.4 Active sensing of an uninformative cue to can constitute a bet-hedging strategy...... 39

3.4.1 Uninformative cues can be reliable...... 43

3.5 Conclusions...... 44

4 The Quick and the Dead: Microbial Demography at the Yeast Ther- mal Limit 46

4.1 Introduction...... 46

4.2 Results...... 48

4.2.1 TrackScar measures the fecundity and mortality of single cells 48

4.2.2 Identifying diverse responses to heat stress...... 50

vi 4.2.3 Heat sensitivity is associated with the emergence of heterogeneous subpopulations...... 51

4.2.4 Increased mortality explains slow population growth for some heat sensitive strains...... 53

4.2.5 Fecundity can be positively or negatively associated with age during stress...... 56

4.2.6 Heat stress can cause premature senescence or early life mortality 57

4.2.7 Early life mortality is associated with a failure to inherit mitochondria...... 59

4.3 Discussion...... 62

4.3.1 TrackScar is a powerful method to measure fecundity and mortality...... 63

4.3.2 Asymmetries during cell division contribute to phenotypic diversity...... 63

4.3.3 Mortality or slower division rate can limit growth during heat stress...... 64

4.3.4 Fecundity and mortality can be age structured during heat stress 65

4.3.5 Diverse physiological factors delimit the niche of S. cerevisiae 66

4.4 Methods...... 67

A Supplementary Information for The Quick and the Dead 68

A.1 Introduction...... 68

A.2 Results...... 69

A.2.1 TrackScar minimally aﬀects cellular physiology...... 69

A.2.2 TrackScar provides a sensitive measure of diﬀerences in fecundity 70

A.2.3 Population Growth Rate and Mean Fecundity Are Well Cor- related...... 71

A.2.4 Six Hours of Recovery Is Suﬃcient To Distinguish Live from Dead Cells...... 71

A.2.5 Heat stress can alter the distribution of ages in a population. 72

vii A.3 Supplementary Methods...... 73

A.3.1 TrackScar Staining...... 73

A.3.2 Microscopy...... 74

A.3.3 Image processing and data analysis...... 75

A.3.4 Data collection and microscopy for time series...... 76

A.3.5 Growth curves...... 77

A.3.6 Strain construction...... 77

Bibliography 82

Biography 93

viii List of Tables

2.1 Strains used in Chapter2...... 31

2.2 Plasmids used in Chapter2...... 32

2.3 Primers used in Chapter2...... 33

A.1 Strains used in Chapter4...... 78

ix List of Figures

2.1 Ura3p mediates a phenotypic trade-oﬀ...... 8

2.2 Sensing estradiol can control a phenotypic trade-oﬀ...... 10

2.3 Sensing estradiol can actualize a bet-hedging strategy...... 13

2.4 Plasticity to estradiol does not always carry a cost...... 15

2.5 Fluorescent tags do not aﬀect the cost of plasticity...... 17

2.6 Mutation can tune bet-hedging strategies based on active sensing.. 19

2.7 Tuning Ura3p stability with degron tags...... 25

3.1 A schematic of the ﬁtness consequences of bet hedging and adaptive plasticity...... 38

3.2 A schematic of the developmental mechanism undelying bet hedging and active sensing...... 40

3.3 A schematic of a bet-hedging strategy based on active sensing.... 42

4.1 TrackScar measures the fecundity and viability of individual yeast... 50

4.2 Genetic diversity in thermotolerance among environmental yeast isolates 52

4.3 Subpopulations of cells with low fecundity in heat sensitive strains.. 54

4.4 The contribution of mortality to slow growth during heat stress... 55

4.5 Age structure of fecundity of S288C...... 57

4.6 The age structure of fecundity and mortality during heat stress... 58

4.7 Mitochondrial inheritance defects in a strain with early life mortality 60

4.8 Mitochondrial morphology during heat stress...... 62

x A.1 TrackScar is sensitive and minimally aﬀects cellular physiology.... 70

A.2 Correlation between TrackScar and growth curves...... 79

A.3 Heat stress can alter the distribution of ages in a population..... 80

A.4 Histograms of all sensitive and robust strains examined...... 81

xi List of Abbreviations and Symbols

Abbreviations

SC+5FOA Synthetic yeast media supplemented by uracil and 5-ﬂuororotic acid.

SC-ura Synthetic yeast media lacking uracil.

xii Acknowledgements

Thank-you to Paul Magwene for staying excited about my work; to Ryan Baugh for teaching me to write a paper; to Daniel Lew, Nicolas Buchler, Katia Koelle, Daniel McShea, Debra Murray, Daniel Skelly, and Helen Murphy for helpful conversations; to David McCandlish for a critically important citation; and to Kriti Sharma for her inspiration, determined optimism, and unﬂagging love. This work was supported in part by NIH (P50GM081883) and NSF (MCB1330545) awards.

xiii 1

Introduction

The expansion, persistence, decline, or extinction of populations is ultimately due to the rates of birth and death of individuals. However, since it is difficult to measure the fitness of individual cells, microorganisms are often treated as populations. A population-level view is appropriate when the phenotypes of individuals are similar to the average phenotype, but can obscure the underlying demographic and molecular mechanisms that lead to population-level phenotypes when there is pronounced variability between individuals. Populations made up of individuals with different phenotypes can behave very differently than populations made up of individuals with similar phenotypes. For example, genotypes that produce individuals with diverse phenotypes can persist in variable environments that would drive phenotypically homogenous genotypes to extinction (Cooper and Kaplan, 1982; Cohen, 1966). This strategy is called ‘bet hedging’. Bet hedging is an evolutionary strategy where a single genotype produces individuals with diverse phenotypes to ensure that some survive regardless of the state of the environment. Bet-hedging strategies are selected for in unpredictable and variable environments that impose strong phenotypic trade-offs (Donaldson-Matasci,

1 2008). Microorganisms are attractive systems with which to examine the molecular basis of bet hedging strategies because of their tractable genetics and short generation times. The intrinsic stochasticity involved in gene expression has emerged as a potent source of phenotypic diversification in microorganisms (Norman et al., 2014; Raj and van Oudenaarden, 2008; Elowitz, 2002). Since the transcripts of some genes are present at a only a few copies per cell, sampling error can cause large phenotypic differences between cells, which can generate bet hedging strategies. For example, antibiotic resistant persister cells form in extremely homogenous environments, strongly suggesting that they form due to molecular stochasticity (Balaban et al., 2004). Molecular stochasticity is also the mechanistic basis for some experimentally evolved bet hedging strategies (Gallie et al., 2015; Beaumont et al., 2009; Acar et al., 2008). Although stochastic gene expression is a potent source of phenotypic variability, bet hedging strategies could also be based on actively sensing the environment (Donaldson-Matasci et al., 2013; Simons, 2011; Cooper and Kaplan, 1982). However, this possibility has received less attention. In Chapter2 I construct a synthetic biological system based on active sensing, and show that bet hedging strategies based on active sensing need not be penalized due to costs of plasticity. One reason that research into the molecular mechanisms that underly bet hedging strategies has focused exclusively on molecular stochasticity seems to be a misunderstanding of the relationship between adaptive plasticity and bet hedging. In Chapter3, I examine the relationship between active sensing that leads to adaptive plasticity, and active sensing that leads to a bet hedging strategy. I show that although adaptive plasticity and bet hedging are evolutionary strategies with different consequences for the fitness of individuals, that they can nonetheless share a molecular basis. The possibility that bet-hedging strategies could be built on active sensing reveals

2 the need for new experimental systems to examine the fitness of microorganisms in the environments they live in. Since active sensing can lead to the variability needed for bet hedging strategies, to understand the mechanistic basis of bet hedging, it is not sufficient to examine phenotypic variability in homogenous laboratory microenvironments. Rather, microorganisms should be examined in situ so that the distribution of environmental cues that lead to phenotypic variability remain intact. However, it is challenging to gather information about the fitness of individual cells in the complex and often non-transparent environments that they inhabit in the wild. One possible solution to this difficulty is to develop tools that rely on marking and recapturing individuals, which would allow them to live in their natural environments. This sort of strategy has been widely to study larger organisms such as birds, whales, and deer, but not to microorganisms. In Chapter4, I develop a new method called ‘TrackScar’ to measure the fitness of individual budding yeast cells without the need for microscopic tracking. Using TrackSar, I investigate how rates of birth and death vary during heat stress in a genetically diverse panel of strains. By examining individual cells, I show that the inability of some populations of yeast to grow at high temperatures is due to either increased rates of mortality, decreased rates of fecundity, or a combination of the two. Furthermore, by examining the age structure of mortality, I uncover how these demographic rates are in turn caused by defects in the segregation of sub- cellular components–linking molecular mechanism to the fitness of individuals, and to population-level phenotypes. This approach confirms the power of analyzing individual cells to reveal the molecular basis of phenotypes, and lays the foundation for examining populations of budding yeast in realistic environments that are not well-suited to microscopic tracking of individuals.

3 2

Bet Hedging by Active Sensing Need not be Costly

2.1 Introduction

Organisms must contend with unpredictable and variable environments. To cope with uncertainty, some organisms produce offspring with diverse phenotypes so that some will survive regardless of the environment. This evolutionary strategy is called ‘bet-hedging’ or ‘adaptive coin flipping’ (Cohen, 1966; Cooper and Kaplan, 1982). A bet-hedging strategy is distinct from adaptive plasticity, although both are strategies to cope with variable environments (de Jong et al., 2011). In the case of adaptive plasticity, an organism senses a cue that provides information about the environment to adopt (or produce offspring with) a phenotype with high fitness in that environment. In contrast, an organism pursuing a bet-hedging strategy will produce offspring with diverse phenotypes regardless of the state of the environment (de Jong et al., 2011). Although bet-hedging has been well-studied theoretically, it is difficult to demonstrate that bet hedging is actually occurs in wild populations (Simons, 2011). Because of this difficulty, the mechanistic bases of bet-hedging are only beginning to be discovered.

4 Microbial models systems have emerged as important systems for exploring the mechanistic bases of bet hedging. Microbes have several advantages for mechanistic studies due to their tractable genetics and fast generation times. However, the interpretation of some studies that claim to investigate the molecular bases of bet- hedging strategies is controversial (de Jong et al., 2011; Grimbergen et al., 2015). Since little is known about the environments that individual microbes experience over evolutionarily relevant time-scales, most studies cannot conclude that the phenotypic variability that is observed in a population is due to bet hedging (Simons, 2011). One solution to this conundrum is to experimentally evolve (Beaumont et al., 2009; Fukami et al., 2007), or construct (Acar et al., 2008) bet-hedging strategies. Since the environment of these experimental systems is closely controlled, they allow the mechanistic basis of genuine bet-hedging strategies to be studied. One way in which a cellular bet-hedging strategy can be realized is to harness the statistical uncertainty associated with some proteins (such as transcriptional regu- lators) that are present at low levels in a cell (Norman et al., 2014). The molecular variation that arises from the statistics of small numbers can lead to stochastic phenotypic diﬀerences between individual cells (Rahn, 1932; Spudich and Koshland, 1976; Arkin et al., 1998). Molecular stochasticity likely contributes to the formation of antibiotic-resistant persister cells (Balaban et al., 2004), an example of bet-hedging in bacteria and yeast (Mulcahy et al., 2010; LaFleur et al., 2009; Lewis, 2010), and underlies phenotypic diversity in experimentally evolved (Gallie et al., 2015; Beaumont et al., 2009), and synthetically constructed (Acar et al., 2008) experimental models of bet-hedging. Other studies have shown that mutations can alter the amount of molecular noise associated with a particular locus (Ansel et al., 2008; Metzger et al., 2015), indicating that molecular stochasticity could be selected for by bet-hedging selection. In order to observe the eﬀects of molecular stochasticity most studies minimize the

5 effect of extracellular perturbations on cells or organisms by employing approximately constant environments such as chemostats or flow chambers (Veening et al., 2008; Norman et al., 2013; Levy et al., 2012; Acar et al., 2008). However, this strict focus on the contributions of intracellular processes to phenotypic heterogeneity ignores potential environmental contributions to the phenomenon of bet-hedging. By definition, pure bet-hedging strategies are those in which environmental cues have no predictive value in determining appropriate phenotypes to produce. How- ever, this does not preclude the possibility that organisms could exploit heterogeneity in environmental variables that are unrelated to organismal fitness as the basis for generating diverse phenotypes. Another way of saying this is that many cues are not correlated with the aspect of the environment that determines an organism’s fitness. By varying phenotypes in response to such a cue, an organism would not be pursuing a strategy of adaptive plasticity, but could be using the cue to hedge its bets (Donaldson-Matasci et al., 2013; Bull, 1987; Simons and Johnston, 1997; Cooper and Kaplan, 1982; Childs et al., 2010; Scheiner and Holt, 2012). For example, diversified germination timing is a bet-hedging strategy in the wild tobacco Lobelia inflata (Si- mons, 2009). The fitness of a seed that germinates at a particular time is determined by the amount of rainfall it experiences in the months after germinating. However, the reason that the seed germinates in the first place is determined by seemingly insignificant differences in the temperature of the microenvironment it finds itself in (Simons and Johnston, 2006). In this case, it seems unlikely that the microenvironmental temperature that the seed finds itself in acts as an informative cue (in the sense of providing information about what germination timing will have high fitness), but rather appears to act as a mechanism to actualize a bet-hedging strategy. Since natural environments show pervasive environmental structure that need not be well-correlated with aspects of the environment that affect fitness, this mechanism of bet hedging has potential to be common, yet remains largely unexplored (although

6 see (Donaldson-Matasci et al., 2013) and (Scheiner, 2014)), especially in microbial systems. In order to spur investigation of this mechanism of bet-hedging, we have constructed a simple ‘biological model’ of a bet-hedging strategy in yeast that is based on active sensing. Unicellular organisms are logical systems to examine this type of bet hedging due to extensive microenvironmental variability at small scales, and the ability to create reliably structured microenvironments in the laboratory. Our system captures the essential aspects of classic mathematical models of bet hedging. Using our system we formally show that sensing an uninformative cue can underlie a bet-hedging strategy. Furthermore, we show that a bet-hedging strategy based on sensing need not be selected against due to a cost of plasticity. Our results indicate that future studies should not rule out active sensing when studying the mechanisms that underlie bet-hedging strategies and other types of adaptive phenotypic diversity.

2.2 Results

2.2.1 URA3 (counter-)selection mimics a classic model of bet-hedging

We developed an experimental system to test whether active sensing of an uninformative environmental cue could serve as the mechanistic underpinning of a bet-hedging strategy. The classic model of bet hedging is that of a species living in an unpre- dictably variable environment that imposes a phenotypic trade-off (Cohen, 1966). The simplest version of such a model has three conditions: (1) the environment can be in two different states, (2) the organism can have two different phenotypes, and (3) when there is a mismatch between phenotype and environment, individual fitness is zero. In this scenario the most fit strategy is to produce offspring with diverse phenotypes with a frequency that matches the frequency at which the environmental state to which they are suited is experienced (Cohen, 1966). A simple way to translate this mathematical model into biological reality is to

7 URA3∆ura3

SC-ura

SC+5FOA

Figure 2.1: SC-ura media and SC+5FOA media impose a phenotypic tradeoﬀ based on the presence of Ura3p. Strains either containing (PMY1638) or lacking (CMY19) the URA3 gene were struck out on media lacking uracil (SC-ura) or containing 5- FOA (SC+5FOA) and incubated for 48hrs.

use a selectable/counter-selectable gene product, such as orotidine 5’-phosphate decarboxylase. This protein is encoded by the URA3 gene in the budding yeast Sac- charomyces cerevisiae, where it participates in the de novo synthesis of uracil. When Ura3p is present, cells are able to grow on synthetic media that lacks uracil (SC- ura). However, Ura3p converts the pro-toxin 5-ﬂuororotic acid (5FOA) to the toxin 5-ﬂuorouracil (5FU) (Boeke et al., 1984), leading to cell death. Therefore, Ura3p must be absent for cells to grow on media supplemented with uracil and 5FOA (SC+5FOA) (Fig. 2.1). If we consider the environment to be the media the yeast are growing on, then the environmental states are SC-ura media and SC+5FOA media. Since there are two phenotypes (presence of Ura3p and absence of Ura3p) that are each only suited to growth in one environmental state, the URA3 system provides an excellent experimental model for exploring scenarios that favor bet-hedging strategies.

8 2.2.2 Plasticity to estradiol can generate a phenotypic trade-oﬀ

We sought to test whether response to a non-fitness related environmental cue could serve as the basis of a bet-hedging strategy. To do this, we created a system to modulate the expression of the URA3 gene using a cue orthologous to either the presence of 5FOA or the lack of uracil in the media. Z4EV is an engineered transcription factor that binds to DNA sequences not normally found in the yeast genome (McIsaac et al., 2013). The activity of Z4EV is controlled by the presence of estradiol. In the absence of estradiol, Z4EV is se- questered from the nucleus and thus cannot activate transcription. In the presence of estradiol, Z4EV is translocated to the nucleus, where it can activate gene expression. By replacing the promoter of a gene with the binding site for Z4EV, the expression level of a single gene can be controlled without affecting the expression of any other genes in the genome (McIsaac et al., 2013). We replaced the native URA3 promoter with the Z4EV promoter and added an N-terminal ubiquitin degron to the native URA3 gene to create a strain with the genotype of Z4EVpr:degron-URA3 at the URA3 locus (Fig. 2.2A). We found that it was necessary to lower the stability of Ura3p using the degron in order to allow conditional growth on SC-ura and SC+5FOA media (see methods for details). This strain is phenotypically plastic with respect to the presence or absence of estradiol. Bet-hedging strategies are favored when there are strong trade-offs between phenotypes that are fit in different selective environments (Donaldson-Matasci, 2008). The presence of these trade-offs can be visualized by plotting the fitness of an organism in one selective environment versus its fitness in a second environment for all possible phenotypes. A strong trade-off results in a plot with a concave curve, whereas weak trade-off results in a plot with a convex curve (Donaldson-Matasci, 2008). We plotted the fitness of yeast grown in SC+5FOA against the fitness of

9 those grown in SC-ura for twelve concentrations of estradiol (Fig. 2.2B). The resulting plot is concave across its full surface. This indicates that strong phenotypic trade-oﬀs exist and that a bet-hedging is predicted to be an evolutionarily stable strategy in an environment that alternates between SC-ura and SC+5FOA.

B ●● A 125 ●● 100nM ●● 25nM ●● URA3 100 ●●

75 ●● 6.2nM

25 0.1nM 1.6nM 0nM SC+5FOA SC+5FOA ●● 0.4nM ∆ura3 0 ●● ●● ● ●● ●●● ●●●● ●● SC-ura SC-ura fitness Fitness in SC-ura (cells/µl) Fitness 0 25 50 75 Fitness in SC+5FOA (cells/µl) Figure 2.2: Plasticity to estradiol produces a phenotypic trade-off between growth in SC-ura and SC+5FOA media. (A) A schematic of the system is shown. The promoter of the gene URA3 is replaced with a binding site for the artificial transcription factor Z4EV in the strain CMY97. In the presence of estradiol, the transcription factor translocates to the nucleus and binds to Z4EVpr, which catalyzes the transcription of URA3 (shown as a darkening of the cell). In the absence of estradiol the transcription factor is excluded from the nucleus and URA3 is not transcribed. Cells expressing URA3 are able to grow in the absence of uracil, but cannot grow in the presence of 5-FOA. Cells that do not express URA3 cannot grow in the absence of uracil, but can grow in the presence of 5-FOA. B The fitness of cells in SC+5FOA media is plotted against the fitness of cells in SC-ura media for twelve concentrations of estradiol. For comparison, the fitness of a ura3 (blue, CMY19) and a URA3 (orange, PMY1638) strain are shown. Errobars show the standard error of the mean for three independent biological replicates.

2.2.3 Sensing an uninformative cue enables a bet-hedging strategy

Given that plasticity to estradiol can mediate a trade-oﬀ between growth in SC-ura and SC+5FOA media, we hypothesized that this cue could be used to create a bet- hedging strategy in an environment that alternates between SC-ura and SC+5FOA.

In order to make estradiol an uninformative cue, we spotted small (12.5 µL) drops of 10 10 µM estradiol in the center of SC-ura and SC+5FOA agar plates and allowed the estradiol to diffuse for 24hr. This results in a gradient of estradiol (Fig. 2.3A). Subject to such an estradiol gradient, when grown on SC-ura plates, cells with the genotype Z4EVpr:degron-URA3, have non-zero fitness only near the center of the plates; conversely fitness is maximized at the periphery of plates containing SC+5FOA (Fig. 2.3B). Since the distribution of estradiol is identical in both environments, it provides no information to the organism about which phenotype to adopt. This scenario mimics microenvironmental heterogeneity, which is pervasive for microorganisms because of their small size and the high viscosity of water at small scales (Foster, 1988; Stewart and Franklin, 2008). We tested whether this microenvironmental heterogeneity would allow a plastic strain to outcompete other strategies to cope with a variable environment. The alternative strategies that an organism can employ to persist in an environment that alternates between different environmental states are: (1) to sense the selective environment and adopt a fit phenotype for that selective environment, (2) to adopt a generalist strategy that is fit in all environments, or (3) to choose a specialist phenotype that maximizes fitness despite being selected against in certain environmental states. We constructed our system so that sensing estradiol does not allow an organism to increase its overall fitness since there is always the same distribution of estradiol regardless of the selective media. Given that the expression of URA3 is only controlled by estradiol, this indicates that adaptive plasticity is not a viable strategy in our system. We previously showed that there is no generalist phenotype due to the strong fitness trade-off (Fig. 2.2A). In the absence of other options, the most fit strategy can be to specialize to the environmental state that allows the highest fitness (Donaldson-Matasci, 2008). Therefore, we tested whether the plastic strain could outcompete strains that were specialized to grow on either SC-ura or SC+5FOA media in an environment that alternated between SC+5FOA

11 and SC-ura. We constructed strains that either constitutively lacked or constitutively produced Ura3p by either deleting the URA3 coding sequence (ura3 ) or by retaining it under the control of its native promoter (URA3 ). Ura3p is produced constitutively under its native promoter and is absent in a deletion strain. We previously showed that these constitutive strains were specialized for growth on SC-ura and SC+5FOA media, respectively (Fig. 2.1). To allow the relative abundance of these strains to be tracked during competition, we labeled the specialist genotypes with a green fluorescent protein (GFP) and the plastic phenotype with a red fluorescent protein (mCherry). As a control, we also constructed a plastic strain labelled with GFP. We competed the constitutive and the plastic genotypes in a variable environment with microenvironmental heterogeneity of estradiol as follows (see also Fig. 2.3C): (1) an mCherry labelled plastic strain (Z4EVpr:degron-URA3 ) and either (A) a ura3 strain, (B) a URA3 strain, or (C) a GFP labelled plastic strain were mixed at the beginning of the experiment to equal proportions; (2) « 100, 000 cells total were applied to estradiol structured SC-ura or SC+5FOA plates and incubated for 24 hours, (3) cells were sampled by rolling beads over the surface of the agar plates and then transferred to non-selective media and grown for another 24 hours; (4) steps (2) and (3) were repeated until the experiment ended. After each selective episode, we recorded the number of cells of each genotype by measuring their concentration in a flow-cytometer. Whether, and the rate at which, a bet-hedging strategy outcompetes specialist strategies depends on the actual sequence of selective environments that occur. For the sake of experimental simplicity, we carried out our competition experiments under strict alternation of selective environments (SC-ura and SC+5FOA media). Regardless of whether we began with SC-ura or SC+5FOA media as the first selective

12 Figure 2.3: Plasticity to estradiol in a structured environment actualizes a bet hedging strategy. (A) A schematic of how a structured environment was created is shown. Estradiol was spotted in the center of a plate of either SC-ura or SC+5FOA media and allowed to diﬀuse. (B) Patterns of growth of a Z4EVpr:degron-URA3 strain (CMY121) on either SC-ura or SC+5FOA estradiol structured plates after 24hr are shown. (C) A schematic of the competition regime is shown. Fluores- cently tagged competitors are mixed at the beginning of the experiment. For each epoch, cells are spread on either SC-ura or SC+5FOA estradiol structured plates for 24hr. The cells are then randomly sampled from the plates and grown 24hr in non- selective media, then spread back on SC-ura or SC+5FOA structured plates. After each epoch, the number of mCherry cells was recorded in a ﬂow-cytometer.(D) The fraction of cells with the mCherry marker is shown as a function of epoch for competition experiments that started with either SC+5FOA selective media (top row) or SC-ura selective media (bottom row). For each of these orderings of the selective environment, three competitions are shown: Z4EVpr:degron-URA3 mCherry (CMY122) vs. Z4EVpr:degron-URA3 GFP (CMY121)(fuschia), Z4EVpr:degron-URA3 mCherry (CMY122) vs. ura3 GFP (CMY111)(purple), and Z4EVpr:degron-URA3 mCherry (CMY122) vs. URA3 GFP (CMY136)(green). Data for three independent biological replicates is shown.

13 state, the plastic strain drove the specialist genotypes to extinction after two epochs (Fig. 2.3D); thus the plastic genotype is the most ﬁt one in this variable environment. These experiments serve as a constructive proof that plasticity to microenvironmental heterogeneity can underlie a bet-hedging strategy.

2.2.4 Bet hedging by plasticity need not carry a cost

Given that either plasticity to microenvironmental heterogeneity or molecular stochasticity can underlie a bet-hedging strategy, what factors influence when one molecular basis would be favored over another? Since molecular stochasticity arises simply due to the expression level of a gene it doesn’t require dedicated sensing machinery. Since any system that senses the environment requires that its constituent proteins be produced, it could be that plasticity to the environment carries a metabolic cost of maintenance (Kussell and Leibler, 2005; Masel et al., 2007). Consistent with this idea, gratuitous gene expression has been shown to carry a cost (Lang et al., 2009). It has been shown that these metabolic costs can lead a bet-hedging strategy to outcompete strategies of adaptive plasticity (Kussell and Leibler, 2005). The same logic could favor molecular stochasticity as the basis of cellular bet-hedging strategies since it doesn’t require costly sensing systems. This scenario assumes that plasticity to environmental cues is gained by acquisi- tion of dedicated sensing systems. However, this is not the only way that a phenotype could come under the control of an external cue. For example, in our system, the expression of the URA3 gene became plastic to estradiol by replacing the native promoter of URA3 with the Z4EV promoter in a genetic background that already contained Z4EV. This sort of cis-regulatory variation is common in wild populations (Skelly et al., 2011; Zhang and Borevitz, 2009), and does not involve the wholesale gain of new sensing systems, but rather co-option of existing regulatory apparatus. Costs of plasticity can be identified by comparing the fitness of a specialist strain

14 A Environment A Environment B

cost Fitness Fitness

SpA SpB P SpA SpB P B SC+5FOA, 0nM estr. SC-ura, 100nM estr. competitor plastic strain ● ● ● ● 10 ● ●

● ●

Fitness (number of doublings) Fitness 0 URA3∆ura3 URA3∆ura3 Competitor

Figure 2.4: Plasticity to estradiol does not always carry a cost. (A) A schematic of how to measure a cost of plasticity is shown (adapted from (Murren et al., 2015)). The fitness for three genotypes is plotted for two environments. Genotypes SpA and SpB are specialists for environments A and B, respectively, and have high fitness in the environment for which they are specialized. Genotype P is a plastic genotype that has moderately high fitness in both environments. The difference between the fitness of genotype P and genotypes SpA and SpB in the environments they are specialized for is a measure of the cost of plasticity. A symmetric cost of plasticity is shown, but this need not be the case. (B) The fitness of the plastic strain Z4EVpr:degron- URA3 (CMY122) is shown compared to the fitness of the specialist strains URA3 (CMY136) and ura3 (CMY111) in estradiol concentrations that trigger adaptive plasticity in the plastic strain.

15 in the environment to which it is specialized to that of a plastic strain (Fig. 2.2A). Although the plastic strain in this case can sense the environment and therefore enjoys a relatively high fitness, any cost of plasticity should be reflected by a fitness decrease of the plastic genotype relative to the specialist genotype (Murren et al., 2015). We compared the fitness of the specialist strains to the plastic strain in SC+5FOA media that lacked estradiol as well as SC-ura media that contained estradiol (Fig. 2.4B). In SC+5FOA media, the plastic strain produced 1.8 (15%) fewer offspring on average than the constitutive ∆ura3 strain (paired one-sided t-test, p = 0.030), indicating a cost of plasticity. In contrast, in SC-ura media there was not a signficant difference between the fitness of the plastic strain and the SC-ura specialist strain (paired one-sided t-test, p= 0.11). We included a control experiment where the fitness of a plastic strain labelled with mCherry was compared to the fitness of the same strain labelled with GFP. However, the fluorescent marker did not significantly affect the fitness of the strains in either media (Fig. 2.5)(paired two-sided t-test, p > 0.26). Taken together, these results indicate that plasticity arising from a cis-regulatory mutation need not always penalize bet-hedging strategies based on active sensing of uninformative cues.

2.2.5 Bet hedging by sensing an uniformative cue can be tuned by mutation

In order for a bet-hedging strategy based on active sensing to evolve, it must be able to be tuned by mutation. For example, if the environment were to change from favoring a phenotype expressing Ura3p 30% of the time to one favoring it 50% of a time, would it be possible for a population composed of the bet-hedging genotype to evolve a new diversification strategy? One way this could occur would be via fixation of mutations that effect the uptake or export of estradiol from cells, thus changing the relative sensitivity of the system to external concentrations of the external cue. The

16 A SC+5FOA B SC-ura 0nM estradiol 100nM estradiol 13 GFP 13 mCherry ● ● ● 12 12 ● ●

11 ● 11 Fitness Fitness 10 ● 10

(number of doublings) plastic ∆ura3 (number of doublings) plastic URA3 Competitor Competitor

Figure 2.5: The ﬂuorescent tag used does not aﬀect the cost of plasticity. The number of doublings for plastic and constitutive strains is shown. Competition experiments between either plastic strains marked with mCherry and GFP or between the plastic strain marked with mCherry and (A) a ura3 strain marked with GFP in SC+5FOA media without estradiol or (B) a ura3 strain marked with GFP in SC-ura media with estradiol.

ABC transporters SNQ2 and PDR5 are known to pump estradiol out of cells (Mahé et al., 1996). We reasoned that by deleting these genes the effective concentration of estradiol within yeast cells would increase, leading to a different diversification strategy in an environment structured by estradiol (Fig. 2.6A). We deleted PDR5 and SNQ2 in the Z4EVpr:degron-URA3 genetic background. To test whether the diversification strategy of this genotype (Z4EVpr:degron-URA3, pdr5/snq2 ) was different from its ancestor (Z4EVpr:degron-URA3, PDR5/SNQ2 ), we grew these strains on SC-ura and SC+5FOA plates containing identical gradients of estradiol (spotted with 5 µL of estradiol). The pdr5/snq2 strain was able to grow further away from the center of the plate on SC-ura media, but its growth was inhibited further away from the center on SC+5FOA media (Fig. 2.6B). To quantify how much the diversification strategy had changed, we measured the point where the smoothed pixel intensity of a transect moving away from the center of the plate

17 reached its half maximum (Fig. 2.6C). We used this half-maximum point to estimate the relative proportion (area) of the environment that each strain was able to occupy.

A PDR5/SNQ2 strain occupies 24 ˘ 1.6% of the SC-ura environment and 69 ˘ 5.7% of the SC+5FOA environment. In contrast, a pdr5/snq2 strain occupies 46 ˘ 8% of the SC-ura environment and 49 ˘ 4% of the SC+5FOA environment (All results expressed as means ˘ 95% CI based on three biological replicates). These data show that mutations in PDR5 and SNQ2 can alter the diversiﬁcation strategy of an estradiol-cued bet-hedging genotype towards a strategy more ﬁt for an increased frequency of SC-ura environments. These result thus serve as a constructive proof of the evolvability of bet-hedging systems based on plasticity to microenvironmental heterogeneity.

2.3 Discussion

Previous studies have examined how stochastic molecular events can underlie bet- hedging strategies in microorganisms (Gallie et al., 2015; Acar et al., 2008). However, the role of phenotypic plasticity as the molecular underpinning of bet hedging has been largely ignored. In this study, we used a synthetic biological system of a variable environment and a plastic phenotypic trade-off to show that plasticity to an uninformative cue can underlie a bet-hedging strategy. Theoretical work has shown that plasticity and stochasticity are complemen- tary routes to creating phenotypic diversification (Donaldson-Matasci et al., 2013; Scheiner, 2014). Given that either stochasticity or plasticity can underlie a bet- hedging strategy, why would one mechanism be favored over another? Furthermore, is it plausible that the active sensing of a seemingly transient feature of the environment could lead to the reliable phenotypic diversification needed for a bet-hedging strategy?

18 Figure 2.6: Bet-hedging strategies based on sensing an uninformative cue can be tuned by mutation. (A) A schematic of how deleting the genes encoding the ABC transporters Pdr5p and Snq2p aﬀects the eﬀective concentration of estradiol inside a cell is shown. When PDR5 and SNQ2 are deleted, the resulting strain cannot pump out estradiol, leading to a higher internal concentration. (B) Pat- terns of growth of a strain containing the genotype Z4EVpr:degron-URA3 and either PDR5/SNQ2 (CMY121) or pdr5/snq2 (CMY115) on either SC-ura or SC+5FOA estradiol structured plates after 24hr are shown. (C) The number of cells as a function of distance from a spot of estradiol on both SC+5FOA and SC-ura plates is shown for the genotypes ura3 (CMY111)(blue), URA3 (CMY136)(orange), Z4EVpr:degron-URA3 PDR5/SNQ2 (CMY121)(black) and Z4EVpr:degron-URA3 pdr5/snq2 (CMY115)(maroon) is shown. Semi-transparent lines show the standard error of the mean for three independent biological replicates.

2.3.1 Bet hedging by active sensing is plausible

There is strong evidence that bet-hedging strategies explain variability in phenom- ena such as seed germination timing (Simons, 2009), seed dormancy (Venable, 2012), timing in copepod diapause (Hairston and Munns, 1984), and persister cell formation in bacteria and yeast (Mulcahy et al., 2010; LaFleur et al., 2009). Notably, each of these phenotypes (lack of germination, diapause, slow growth, persister cell

19 formation) are plastic to various cues in the environment (temperature, day length, moisture, nutrient availability) (Clauss and Venable, 2000; Simons and Johnston, 2006; Hairston Jr et al., 1990; Balaban et al., 2004; LaFleur et al., 2009), suggesting that variability in an uninformative cue could underlie each of these bet-hedging strategies. Consistent with this, small differences in temperature contribute to diversification in germination timing in the wild tobacco Lobelia inflata (Simons and Johnston, 2006). Many of these traits (seed germination timing, seed dormancy, diapause) can also act to adaptively alter the phenotype of an organism. This need not preclude sensing cues as a mechanism of bet hedging because in the presence of cues that convey imperfect information about the environment, organisms can adopt mixed strategies of adaptive plasticity and bet hedging (Donaldson-Matasci et al., 2013; Simons, 2014). Future studies will be needed to determine what fraction of plasticity in these systems (if any) is predictive and which fraction actually reflects a bet- hedging strategy.

2.3.2 Bet hedging by active sensing is evolvable

A ﬁt bet-hedging strategy is one in which individuals assume variable phenotypes with a frequency that matches the variability in the selective environment. In order to evolve, the mechanisms responsible for producing the diverse phenotypes must be subject to variation and selection. For example, a bet-hedging strategy may already exist that produces 50% of individuals with phenotype “A”. If the environment changes such that a strategy of producing 75% “A” is favored, what sorts of mutations are capable of ‘tuning’ the strategy to produce 75% “A”? One way to tune the frequency of a phenotype in a bet-hedging strategy based on molecular stochasticity is to alter how variable a gene’s expression level is across individuals. Several studies have shown that gene expression noise can be altered by

20 mutation (Ansel et al., 2008; Metzger et al., 2015; Jones et al., 2014; Carey et al., 2013), indicating that it is evolvable. Mutations can also tune the amount of phenotypic diversification in strategies based on sensing uninformative cues. Given a constant distribution of the cue, mutations that shift the norm of reaction to the cue to the ‘left’ or the ‘right’ (cf Fig. 2.6) can lead to different frequencies of a phenotype. In our synthetic system, loss of function mutations to the efflux pumps PDR5 and SNQ2 increased the effective internal concentration of estradiol in yeast because the cells can no longer remove it (Mahéet al., 1996). As a consequence, pdr5/snq2 mutants produced a diversification strategy more suited to environmental regimes with a greater frequency of SC-ura selection. However, mutations affecting other aspects of the system could also lead to the same effect. For example, mutations that affect the sensitivity of the estradiol binding domain in Z4EV to estradiol have been reported (Lindstrom and Gottschling, 2009), and the DNA binding domain of similar transcription factors can be tuned by mutation to have greater or less affinity for its targer (McIsaac et al., 2013). Taken together, this suggests that the mutational target size to tune bet-hedging strategies based on sensing uninformative cues is likely to be large.

2.3.3 Bet hedging strategies based on phenotypic plasticity need not be costly

Although adaptive phenotypic plasticity increases an organism’s ﬁtness when it is present, not all traits are plastic. This has led to the hypothesis that organisms with plastic traits may carry a cost relative to organisms with constitutive traits (Dewitt et al., 1998). If plasticity always carried a cost, bet-hedging strategies based on plasticity would always be disfavored relative to those based on molecular stochasticity. However, meta-analyses have shown no clear trend in detecting costs of plasticity (Van Buskirk and Steiner, 2009). The lack of a consistent cost of plasticity has led some researchers to hypothesize that there may only be negligible costs of plasticity

21 in many systems, notably microbes (Murren et al., 2015). Changes in gene expression are an important source of plasticity in microorganisms. In principle, gene expression plasticity can be lost by two routes with different effects on the fitness of the organism. Either (1) the transcription factor that regu- lates the expression of a gene could be lost (a trans-regulatory mutation) or (2) the composition of a gene’s promoter could be changed so that the transcription factor no longer binds (a cis-regulatory mutation). In the first case one might expect a fitness gain since the metabolic cost paid by a cell for producing the transcription factor would no longer be paid (Lang et al., 2009). However, in the second case a small change to the sequence of a non-coding region seems unlikely to change the fitness of a strain on its own. Using our synthetic system, we tested whether a strain that gained plasticity by a cis-regulatory mutation would pay a cost relative to non-plastic strains. In one of the two environments surveyed (SC-ura), we failed to detect a cost of plasticity. In the other environment (SC+5FOA), we did detect a cost. However, in the latter case the cost seems likely to be explained by the ‘leakiness’ of the Z4EV system in the absence of estradiol. Given that additional mutations could perhaps abolish this cost (Lindstrom and Gottschling, 2009), there is no reason to expect that a plastic strain will always pay a cost in SC+5FOA media. Our results indicate that the costs of plasticity need not prohibit bet-hedging strategies druven by sensing of uninformative cues and suggest that whether costs of plasticity exist will vary depending on the molecular details of the system being studied.

2.3.4 Sensing uninformative cues can serve as the basis of bet-hedging strategies

Compared to the intrinsic statistical properties of small numbers of molecules, microenvironmental heterogeneity may seem too unreliable to provide a basis for bet hedging. However, organisms have evolved to harness environmental heterogeneity in

22 other contexts to diversify phenotypes, including sex-determination in some reptiles (Janzen and Paukstis, 1991) and teleost ﬁsh (Bachtrog et al., 2014). Since plasticity is pervasive (West-Eberhard, 2003), our results indicate that bet hedging could be taking place in more systems than is currently recognized. Determining when bet hedging is taking place and what its molecular basis is will require greater focus on the details of the environments that microorganisms are found in. Bet hedging is a particularly interesting evolutionary strategy, but to understand it we must examine it in equally interesting environments.

23 2.4 Methods

2.4.1 Strains and media

All strains used in this study were derived from PMY1638 (DBY12397) or PMY1639 (DBY12398) which were kind gifts of Scott McIsaac and David Botstein. These strains are described in McIsaac and co-workers (McIsaac et al., 2013). The strains are derived from the S288c genetic background and contain the Z4EV artiﬁcial transcription factor driven by the constitutive ACT1pr integrated at the LEU2 locus

(leu2∆0 :: ACT 1pr ´ Z4EV ´ NatMX). Synthetic media (SC) and SC-ura was prepared as described in (Sherman, 2002). SC+5FOA media was prepared as described in (Boeke et al., 1984), except that 0.66mg/ml 5-FOA was used rather than 1mg/ml because we found that this concentration of 5FOA was required to allow for robust growth of Z4EVpr:degron-URA3 cells (Fig. 2.7).

2.4.2 Plasmid construction

The plasmids pCSM9, pCM10, and pCSM11 were used to replace the URA3 promoter with the KanMX-Z4EVpr and fuse an N-terminal degron to the URA3 gene were constructed using one step sequence and ligation independent clonging (Jeong et al., 2012). Briefly, the plasmid pMN10 was linearized using primers pMN10 F/R . Homology to the resulting linearized plasmid was added to the ubiquitin moiety and degron from the plasmids pCC17, pCC33, or pCC18, (kind gifts of Charlie Cooper and Nicolas Buchler) respectively using the primers pMN10 degron SLIC1/2. These pieces were assembled by digestion with the T7 polymerase as described in (Jeong et al., 2012) and transformed into NEB 5-alpha competent E. coli (NEB C2987I). Transformants were identified by plating on ampicillin, then screening for correctly assembled plasmids using a SpeI/XbaI digestion of plasmids purified from the result-

24 A Growth in SC + 50mg/L uracil 0nM estradiol 1000nM estradiol 0

-1 URA3 ura3 -2 Z4EVpr:URA3 -3 Z4EVpr:Ub−R−URA3 Z4EVpr:Ub−Q−URA3 log2 OD600 -4 Z4EVpr:Ub−I−URA3

-5 0 0.25 0.50 0.75 1.00 [5-FOA] (mg/ml) B Growth in SC−ura 0

URA3 ura3 -2 Z4EVpr:URA3 Z4EVpr:Ub−R−URA3

log2 OD600 Z4EVpr:Ub−Q−URA3 -4 Z4EVpr:Ub−I−URA3

0 10.1 10 100 1000 estradiol (nM)

Figure 2.7: The Z4EVpr is slightly leaky, which required the tuning of the Ura3p stability using N-terminal degron tags to allow growth on both SC-ura and SC+5FOA. The growth of cells without degron tags (blue), with various degron tags (purple, orange, and red) as well URA3 with its native promoter (black) and a URA3 knockout strain (grey) are shown for (A) various concentrations of 5-FOA in the presence and absence of estradiol and (B) the absence of uracil in the presence of diﬀerent concentrations of estradiol.

25 ing colonies. The degron sequence of each plasmid including the codon immediately following the ubiquitin moiety was verifying using Sanger sequencing at the Duke Sequencing Facility using the primers indicated in Table 2.3.

2.4.3 Strain construction

All yeast transformations were performed using the LiAc/SS DNA carrier/PEG method (Gietz and Schiestl, 2007). The strains with URA3 under the control of Z4EV and fused to arginine, glu- tamine, or isoleucine (CMY092, CMY094, CMY097, respectively) were made by PCR directed homologous recombination. The cassette KanMX-Z4EVpr-degron was ampli- ﬁed from pCSM9, pCSM10, or pCSM11, respectively, using the primers Z4EVpr URA3- F/R to add homology to the URA3 promoter and transformed into the strain DBY12398. Transformants were selected on YPD+G418 media and screened using colony PCR. The resulting strains were CMY92 (Z4EVpr:Ub-R-Ek-URA3), CMY94 (Z4EVpr:Ub-Q-Ek-URA3), and CMY97 (Z4EVpr:Ub-I-Ek-URA3). A ura3:HphMX strain was constructed using PCR-mediated knockout with the primers URA3 KO F/R and the plasmid pAG32 (Goldstein and McCusker, 1999) to yield CMY19. Using CMY19, we made sequential pdr5::hisG and snq2::hisG knock- outs using pMPY-ZAP (Schneider et al., 1996) and the primers PDR5 ZAP F/R and SNQ2 ZAP F/R, respectively, to ﬁrst knock-out each gene using PCR-mediated homologous (selecting for URA3 transformants), then selecting for pop-out of the URA3 cassette on SC+5FOA media. This yielded the strain CMY59 (pdr5::hisG, snq2::hisG, ura3::HygMX, Matα). This was then crossed with CMY97 (KanMX- Z4EVpr:Ub-I-Ek-URA3, Mata) and meiotic progeny were dissected. These progeny were screened using replica plating and colony PCR for HygS segregants with the snq2::hisG, pdr5::hisG genotype and the Mata mating type locus to yield CMY103 (snq2::hisG pdr5::hisG KanMX-Z4EVpr:Ub-I-Ek-URA3 Mata). Another meiotic

26 progeny was located that was HgS, SNQ2, PDR5, and Mata to yield CMY105 (KanMX-Z4EVpr:Ub-I-Ek-URA3 Mata). Fluorescently labelled Z4EVpr:degron-URA3 strains were created by crossing a strain expressing a fluorescent protein fused to Pgk1p to CMY103 or CMY105. Then screening the meiotic progeny of this cross for segregants with the PGK1- FP/Z4EVpr:degron-URA3 genotype and the appropriate genotype at PDR5 and SNQ2. To construct the parents expressing the fluorescent proteins, the primers PGK1-YRC-for/rev were used to add PGK1 homology to the GFP-KanMX cassette or the mCherry-KanMX cassette from the plasmids pFA6a-GFP-KanMX6 and pFA6a-mCherry-KanMX6, respectively (Longtine et al., 1998). The resulting PCR products were transformed into CMY19 ( ura3::HygMX ). Transformants were selected on YPD+G418 media and screened for fluorescence to yield CMY111 (ura3::HygMX, PGK1-GFP-KanMX, Matα) and CMY112 (ura3::HygMX, PGK1- mCherry-KanMX, Matα), respectively. A GFP tagged URA3 strain was constructed using the same primers to transform PMY1638 to yield CMY136 (URA3, PGK1- mCherry-KanMX, Matα). CMY111 and CMY112 were each mated to CMY105, and meiotic progeny from this cross were isolated that were both fluorescent, sensitive to hygromycin, and had the genotype Matα at the mating type locus to yield CMY121 (KanMX-Z4EVpr:Ub- I-Ek-URA3 PGK1-GFP-KanMX Matα) and CMY122 (KanMX-Z4EVpr:Ub-I-Ek- URA3 PGK1-mCherry-KanMX Matα). CMY111 was mated to CMY103, and a meiotic product from this cross were isolated that were both fluorescent, sensitive to hygromycin, and had the genotype Matα at the mating type locus, and the genotype pdr5:hisG and snq2:hisG at the loci PDR5 and SNQ2, respectively to yield CMY122 (snq2::hisG pdr5::hisG KanMX-Z4EVpr:Ub-I-Ek-URA3 PGK1-GFP- KanMX Matα).

27 2.4.4 Flow cytometry and OD measurements

All flow cytometry measurements were perfomed in a MACSQuant VYB flow cytometer. At least 10,000 events of each sample were collected for each measurement. Gating was done using the built-in MACSQuant software to: 1) identify cells based on FSC and SSC, 2) exclude doublets based on the ratio of FSC-H to FSC-A, 3) find GFP and mCherry positive cells using the channels B1 and Y1, respectively. The number of cells, and the percentage of cells in each gate were exported using the built-in software and coverted from the native HTML table format using custom software.

2.4.5 Fitness and competition experiments in liquid culture

To test the ﬁtness of strains for various concentrations of estradiol in liquid media (Fig. 2.2B), we prepared 13 concentrations of estradiol in a 1:2 serial dilution series from 100nM estradiol and a culture with 0nM estradiol in both SC-ura and SC+5FOA media. We washed overnight cultures of cells 2x in sH20, then counted in a hemocytometer and diluted to « 5x106cells/ml, then added 10 µL of these cells to each well to get 250,000cells/ml in the SC-ura or SC+5FOA media. Cells were then incubated in a plate shaker at 500rpm for 24hr and chilled to 4C until measurement. To measure, each culture was diluted 1:50 and measured in a ﬂow-cytometer. To compete plastic and specialist strains (Fig. 2.4), pools of competitors were washed 2x in sH20, then diluted into SC-ura or SC+5FOA media either supplemented with 100nM estradiol or 0nM estradiol as above and incubated on a plate shaker at 500rpm for 24hr and measured as above.

2.4.6 Competiton experiments on agar plates

Agar plates (2% agar) for the competition experiments were prepared as described in (Sherman, 2002), except 15ml rather than 20ml of each agar plate was poured.

28 Molten agar was poured into 100cm dishes, and the dishes were left to harden for

«20 minutes at room temperature. Agar plates were dried with the tops off for 30minutes in a sterile hood, then were stored in sleeves at 4C until use. To create plates with diffusion gradients of estradiol, we marked the center of the agar plates with a marker to indicate the center of the diffusion gradient. We then spotted either

5 µL (for Fig. 2.6) or 12.5 µL (for Fig. 2.3) of 10 µM estradiol above the mark and left to dry on the bench at room temperature for 24hr. To begin each experiment, competing strains were grown separately in YPD to get saturated overnight cultures. These strains were then counted in a hemocytometer and and diluted to get equal proportions of each strain and a ﬁnal concentration of 1x106cells/ml in sterile water. One hundred uL of these cells were then spread on agar plates using glass beads. Plates were incubated for 24hrs at 30 ˝C. After this incubation, cells were randomly sampled from the surface of the plate by shaking glass beads on the plates then transferring three beads to 2ml of SC media in a culture tube. These cells were then incubated overnight at 30 ˝C on a rotor drum. After this incubation, the next epoch was begun by diuting cells to 1x106cells/ml in sterile water, and the process of spreading and sampling was repeated for each epoch. Diluted cells were measured in the ﬂow cytometer after each epoch.

2.4.7 Patterns of growth, image processing, and statistics

To assess patterns of growth on gradients of estradiol, agar plates were prepared as above. Cells were diluted to 1e106, in sterile water, then spread onto the agar plates and incubated 24hr at 30C. Plates were scanned by removing the lids from the plates, and placing face-up on the bed of an Epson Expression 10000XL. The scanner was set for ‘positive ﬁlm’ with a focus of 2.5 and set to collect 16bit black-and-white 360dpi images. To create display images, the raw images were cropped in ImageJ, then aligned

29 manually in GIMP by superimposing the spot in the center of the plate where estradiol the was applied. Plates were cropped as appropriate, the layers were flattened, and the brightness and contrast was adjusted. To create plots showing the distribution of pixel intensity as a function of distance from the spot, 2.84cm x 1cm rectangles were cropped starting at the center of the diffusion gradient and moving out using ImageJ. The lookup table (LUT) of each rectangle was then inverted in ImageJ, and a stack was created from each rectangle. The background was subtracted from the stack by adjusting the levels so that plates with no growth (e.g. ∆ura3 strain on SC-ura media) appeared mostly black and ‘applying’ the LUT to each slice in the stack. Individual images from the stack were then read into R using the readTIFF function in the tiff package. The maximum pixel intensity for each pixel moving away from the center of the diffusion gradient was calculated, then a 51 pixel median smoother was applied. This smoothed version of the pixel intensity was used for the plots in Figure 2.6 and to compute the distance that each strain reached its half-maximum pixel intensity value from the center of the plate. All statistical tests were performed in the R computing environment .

2.5 Strains, primers, and plasmids used in this study

30 Table 2.1: Strains used in this study

Name Genotype PMY1638 leu2∆0::ACT1pr-Z4EV-NatMX, Matα PMY1639 leu2∆0::ACT1pr-Z4EV-NatMX, Mata CMY019 leu2∆0::ACT1pr-Z4EV-NatMX, ura3::hphMX, Matα CMY092 leu2∆0::ACT1pr-Z4EV-NatMX KanMX-Z4EVpr:Ub-R-Ek-URA3, Mata CMY094 leu2∆0::ACT1pr-Z4EV-NatMX KanMX-Z4EVpr:Ub-Q-Ek-URA3, Mata CMY097 leu2∆0::ACT1pr-Z4EV-NatMX KanMX-Z4EVpr:Ub-I-Ek-URA3, Mata CMY105 leu2∆0::ACT1pr-Z4EV-NatMX KanMX-Z4EVpr:Ub-I-Ek-URA3 Mata CMY111 leu2∆0::ACT1pr-Z4EV-NatMX ura3::HygMX, PGK1-GFP-KanMX, Matα CMY112 leu2∆0::ACT1pr-Z4EV-NatMX ura3::HygMX, PGK1-mCherry- KanMX, Matα CMY121 leu2∆0::ACT1pr-Z4EV-NatMX KanMX-Z4EVpr:Ub-I-Ek-URA3 PGK1-GFP-KanMX Matα CMY122 leu2∆0::ACT1pr-Z4EV-NatMX KanMX-Z4EVpr:Ub-I-Ek-URA3 PGK1-mCherry-KanMX Matα CMY136 leu2∆0::ACT1pr-Z4EV-NatMX, PGK1-GFP-KanMX, Matα CMY59 leu2∆0::ACT1pr-Z4EV-NatMX, pdr5::hisG, snq2::hisG, ura3::HygMX, Matα CMY103 snq2::hisG pdr5::hisG leu2∆0::ACT1pr-Z4EV-NatMX KanMX-Z4EVpr:Ub-I-Ek-URA3 Mata CMY115 snq2::hisG pdr5::hisG leu2∆0::ACT1pr-Z4EV-NatMX KanMX-Z4EVpr:Ub-I-Ek-URA3 PGK1-GFP-KanMX Matα

31 Table 2.2: Plasmids used in this study

Name Description Source PMB199 Plamid pMN10, carrying the Z4EVpr Botstein/McIsaac pCC17 Plasmid bearing the Ub-R-Ek degron Cooper/Buchler pCC18 Plasmid bearing the Ub-Q-Ek degron Cooper/Buchler pCC33 Plasmid bearing the Ub-Q-Ek degron Cooper/Buchler pCSM09 pMB10-Ub-R-Ek Constructed in house pCSM10 pMN10-Ub-Q-Ek Constructed in house pCSM11 pMN10-Ub-I-Ek Constructed in house PMB204 pFA6a-GFP-KanMX6 John McCusker PMB205 pBS34 Yeast resource center pMPY-ZAP pMPY-ZAP Duke model genome center PMB105 pAG32 John McCusker

32 Table 2.3: Primers used in this study

Name sequence pMN10 R GTTTTTTCTCCTTGACGTTAAAG pMN10 F TATAGGTACCACTAGTATGGACG pMN10 degron SLIC1 TACTTTAACGTCAAGGAGAAAAAACATGCAGATTTTCGTCAAG pMN10 degron SLIC2 GACGTCCATACTAGTGGTACCTATAGTTAATTAATGATCTTTGCATTAG Z4EVpr URA3 F TAATGTGGCTGTGGTTTCAGGGTCCATAAAGCTTTTCAATTCATCCGCAC TTAACTTCGCATCTG Z4EVpr URA3 R ACTAGGATGAGTAGCAGCACGTTCCTTATATGTAGCTTTCGACATTATAG TTTTTTCTCCTTGACG Z4EVpr Ub-R-Ek-URA3 R ACTAGGATGAGTAGCAGCACGTTCCTTATATGTAGCTTTCGACATGTTAA TTAATGATCTTTGCATTAGAGAATC Z4EVpr URA3 A AGGCTACTGCGCCAATTGAT 33 Z4EVpr URA3 D2 GGCTTAACTGTGCCCTCCAT URA3 KO F GCTACTCATCCTAGTCCTGTTGCTGCCAAGCTATTTAATATCATGCGTAC GCTGCAGGTCGAC URA3 KO R TTGTGAGTTTAGTATACATGCATTTACTTATAATACAGTTTTTTAATCGA TGAATTCGAGCTCG PGK1-YRC-for GGTAAGGAATTGCCAGGTGTTGCTTTCTTATCCGAAAAGAAAGGTCGACG GATCCCCGGG PGK1-YRC-rev GAAAAGAAAAAAATTGATCTATCGATTTCAATTCAATTCAATATCGATGA ATTCGAGCTCG PDR5 ZAP F TTAAGTTTTCGTATCCGCTCGTTCGAAAGACTTTAGACAAAAATGAGGGA ACAAAAGCTGG PDR5 ZAP R CATCTTGGTAAGTTTCTTTTCTTAACCAAATTCAAAATTCTATTACTATA GGGCGAATTGG SNQ2 ZAP F GGATAGAATAACACAGCTACCAAAATACGTAAAGAGAATTCAATGAGGGA ACAAAAGCTGG SNQ2 ZAP R GGCAGATGAATGCACAAAATGTTAAGTTATCTGAAGCCCACATTACTATA GGGCGAATTGG 3

Known Unknowns: Sensing Uninformative Cues can Generate Bet-hedging Strategies

3.1 Introduction

Bet hedging is an evolutionary strategy to cope with unpredictable and variable environments. In a bet-hedging strategy, an organism produces diverse phenotypes independent of the state of the environment so as to ensure that some offspring will always survive (Cooper and Kaplan, 1982; Cohen, 1966). In order to pursue a bet- hedging strategy, a single genotype must ‘randomly’ produce offspring with different phenotypes. What molecular mechanisms allow this phenotypic diversification? The theoretical conditions that favor bet hedging are well-described (Cooper and Kaplan, 1982; Cohen, 1966; Donaldson-Matasci, 2008; Leibler and Kussell, 2010). However, it is difficult to identify organisms pursuing bet-hedging strategies in the wild (Simons, 2011). To show that a bet-hedging strategy is occuring, the fitness of individuals must be tracked for an extended period of time and a quantitative fit must be found between the degree of phenotypic diversification and the variability in the environment. Although it is challenging to document that all of these criteria for

34 bet-hedging are met, there are a number of convincing examples across the tree of life, of bet-hedging organisms (summarized in (Simons, 2011)). Although the mechanistic bases these examples is generally unknown, a parallel literature has examined the consequences of phenotypic variability in microbes (Avery, 2006; Norman et al., 2014). In this literature, bet hedging is often invoked as a potential evolutionary explanation for phenotypic variability within populations (Norman et al., 2014). Microbes are attractive systems to unravel the mechanisms of bet hedging because of their tractable genetics, rapid generation time, and the ease of generating clonal populations. Furthermore, there is strong evidence that the formation of antibiotic- resistant persister cells is a bet-hedging strategy (Lewis, 2010; LaFleur et al., 2009; Mulcahy et al., 2010). However, there is little or no quantitative information about the selective environments that microbes (and other model organisms) experience in wild settings. This makes it impossible to assess whether the amount of phenotypic diversity that is observed in a microbial population is consistent with a bet-hedging strategy, except in cases of experimental evolution (Graham et al., 2014; Beaumont et al., 2009; Gallie et al., 2015; Acar et al., 2008). This lack of information means that bet hedging is usually, but not always, invoked mistakenly to explain phenotypic variability in microbes (de Jong et al., 2011; Grimbergen et al., 2015) (for examples of mistaken attribution in the last year see: (Cenens et al., 2015; Finkelshtein et al., 2015; Weinberger, 2015)). In this article we will consider another conceptual pitfall related to studying bet hedging in microbes. Studies of the mechanistic basis of adaptive phenotypic diversification—including bet hedging—have been conducted in extremely homogenous environments designed to minimize the effect of environmental variability on cells. This common experimental is appropriate for estimating the amount of phenotypic diversification that is independent of the environment. However, bet-hedging strategies need not be based only on mechanisms inside the cell. Since most natural

35 environments are not at all homogenous, this design underestimates the amount of phenotypic variability coming from active sensing, and makes the interpretation of the variability observed in populations of microorganisms diﬃcult.

3.2 Bet hedging in microbes and stochastic gene expression

Variability in gene expression between individual cells is an important cause of phenotypic variability in microbes (Norman et al., 2014), and bet-hedging strategies are frequently invoked to explain why such variability exists (Richard and Yvert, 2014; Weinberger, 2015). Even cells grown in extremely homogenous conditions can display significant cellular individuality as a result of the stochasticity associated with gene expression (Raj and van Oudenaarden, 2008). The detailed mechanistic basis of stochastic gene expression is complex (Jones et al., 2014; Ansel et al., 2008), but is in part due to the sampling error that arises from cellular processes catalyzed by small numbers of molecules (Spudich and Koshland, 1976). The amount of stochasticity (or ‘noise’) associated with a protein (defined as the variability divided by the mean across a population of cells) is affected by both the sequence of the promoter of the gene encoding the protein (‘intrinsic’ factors) as well as factors ‘extrinsic’ to the promoter including differences between cells in their cell cycle, age, or differences in microenvironments (Elowitz, 2002; Hilfinger and Paulsson, 2011). Most studies of phenotypic variability in microbes take pains to minimize the effect of extrinsic factors that originate outside of cells (e.g. microenvironmental variability) in order to measure ‘intrinsic’ variability or at least variability that is caused by cyclic factors such as cell cycle and aging. To exclude microenvironmental variability, cells are frequently grown at low density in their media or in microfluidics devices (Veening et al., 2008; Levy et al., 2012; Norman et al., 2013) and controls are included to verify that differences between cells are not due differences in the environments between cells (Levy et al., 2012; Balaban et al., 2004; Norman et al., 2013).

36 The rationale for this approach is to determine “how much of a cell’s behaviour can be attributed to the environment and how much to the internal program, that is, the behaviour the [cell] would implement were the environment fixed” (Norman et al., 2013). Following this logic, the best way to determine how a complex behavior—such as bet hedging—is produced mechanistically is to isolate the cell from all external influences. Some researchers have even gone so far as to assert that to prove that a bet-hedging strategy is operating “cell lineages must be demonstrated to intercon- vert between phenotypic states reversibly and independently of the prevailing environment” (our emphasis)(Levy et al., 2012). Although isolating cells from external variability allows the effects of ‘intrinsic’ factors on the phenotype of an organism to be observed, it is not not true that stochastic gene expression is the only possible mechanistic basis for bet hedging.

3.3 Bet hedging can be generated by either environmental or physiological noise

Many organisms respond to cues in their environments to increase their fitness. For example, microbes can sense sugars and induce the expression of enzymes needed to metabolize them. This response is adaptive because microbes that do not induce the expression of the genes cannot grow on the sugar, whereas those that always express the genes pay the metabolic cost of gratuitously producing the enzymes. As in a bet- hedging strategy, an organism that is capable of adaptive plasticity produces multiple phenotypes from a single genotype. However, this strategy of adaptive plasticity is distinct from bet hedging because the cells are responding to a cue in order to adopt a phenotype that is appropriate to the environment (Fig. 3.1)(de Jong et al., 2011). In contrast, an organism pursuing a bet hedging strategy will adopt (or produce offspring with) diverse phenotypes without regard to whether those phenotypes are fit or not in the current environment. In fact, by definition, some individuals in an

37 organism pursuing a bet-hedging strategy have low ﬁtness in any given state of the environment.

adaptive plasticity bet hedging

1 2 3 1 2 3 epoch epoch

Figure 3.1: A schematic of the fitness consequences of bet hedging and adaptive plasticity is shown. The environment alternates between epochs of different environmental states, shown as black and white. Organisms are able to adopt two phenotypes “A” (white) and “B” (black) that are suited to the environmental states shown as the same color. In the event of a match between the phenotype and environment, the organism produces two offspring, whereas it produces zero offspring in the event of a mismatch. An organism pursuing a strategy of adaptive plasticity can sense the environment, so each individual adopts a fit phenotype. In contrast, an organism pursuing a strategy of bet hedging does not sense the environment, and produces a mixture of fit and unfit individuals.

Although phenotypic variability has different fitness consequences in organisms pursuing strategies of bet hedging versus adaptive plasticity, the mechanistic basis of these strategies can be the same. Bet hedging can be generated by organisms sensing external cues provided that sensing the cue does not always increase the fitness of each individual. Noisy external cues have been proposed to be the mechanistic basis of bet hedging (Donaldson-Matasci et al., 2013; Bull, 1987; Simons and Johnston, 1997; Cooper and Kaplan, 1982; Childs et al., 2010; Scheiner and Holt, 2012). An example of external noise underlying a bet-hedging strategy is seed germination timing

38 in Lobelia inflata. These plants pursue a bet hedging strategy by diversifying the germination time of their offspring (Simons, 2009), and this diversification is caused by microenvironmental variability in temperature (Simons and Johnston, 2006). How- ever, the microenvironment that the seed finds itself does not predict the aspects of the environment that determine its fitness (e.g. the total rainfall throughout the growing season). This suggests that response to microenvironmental temperature is a bet-hedging strategy and not an example of adaptive plasticity.

3.4 Active sensing of an uninformative cue to can constitute a bet- hedging strategy

To analyze the sources of variability that can be harnessed by bet hedging, we will outline an abbreviated version of the schematic used by Donaldson-Matasci and co-workers (Donaldson-Matasci et al., 2013)(Fig. 3.2). For the purposes of our discussion, we will assume an environment that randomly assumes two states with equal odds and an organism with two diﬀerent phenotypes (x “‘A’ or ‘B’) that are specialized to the two states (cf Fig. 3.1). We will also assume that the ﬁtness of the organism is zero for a mismatch between x and e. We will assume some develop-

mental cue c is present in the environment that the organism senses with P pc | eq. The cue affects the phenotype of the organism with P px | cq. If the cue does not predict (is not correlated with) the environment, then P pc|eq “ P pcq and the cue is uninformative since it does not allow the organism to increase its fitness (Fig. 3.2). If it is impossible for an organism to effectively sense the state of e, the only way to persist is a bet hedging strategy, and the most fit strategy in our scenario is to produce phenotypes ‘A’ and ‘B’ in equal proportion. Finally, we will assume that the fitness of the organism depends only on the environment and the phenotype that

it assumed fpx, eq To mechanistically examine how this optimal strategy of producing 50% ‘A’ and

39 informative cue un-informative cue P(c|e) ≠ P(c) P(c|e) = P(c) e c x f xce f

Figure 3.2: A schematic of the developmental mechanism underlying a strategy of adaptive plasticity and a strategy of bet hedging by active sensing is shown. The environment e can take two states (shown as black and white), and the organism’s phenotype x can take two states (shown as black and white) that are only able to produce offspring (f ą0) in the environment to which they are matched. In this figure, the phenotype of the organism is solely determined by a developmental cue c ( P px | cq “ 0 or 1). When an informative cue is present, P pc | eq approaches 0 or 1, and there is a high correlation between the state of the environment and the cue. This allows an organism whose development is determined by the cue to adopt a fit phenotype in each environment, which is a strategy of adaptive plasticity. In contrast, when P pc|eq “ P pcq, then there is no correlation between the environment and the developmental cue. When an uninformative cue is present, this allows the diversification of phenotypes needed for a bet-hedging strategy.

50% ‘B’ is generated, we can examine the quantity P px | cq. If P px | cq is close to one or zero, then the phenotype of the organism is determined only by the developmental cue. In contrast, if it is close to 0.5, then the cue provides no information about the phenotype of the organism (since the odds of the organism producing ‘A’ or ‘B’ are 50:50). Because of this, Donaldson-Matasci argue that the information contained in c about the phenotype x is a measure of the ‘developmental sensitivity’ and the uncertainty about x after accounting for c is a measure of ‘developmental stochasticity’ (Donaldson-Matasci et al., 2013).

40 Given this framework, we can examine how a bet-hedging strategy can be actualized. Donaldson-Matasci and co-workers show that an optimal developmental strategy will produce offspring with the diverse phenotypes needed for a bet-hedging strategy using different amounts of developmental sensitivity and developmental stochasticity. In fact, developmental stochasticity and sensitivity can be substi- tuted for one another; in the presence of greater individual variability in the P pcq, the optimal developmental strategy is produced by greater developmental sensitivity (Donaldson-Matasci et al., 2013). Intutively, this result makes sense because in the absence of variability in a developmental cue, developmental stochasticity must be harnessed to produce the diverse phenotypes needed for a bet-hedging strategy. Similar results were found using an different theoretical framework by Scheiner and co-workers (Scheiner, 2014). Bet-hedging strategies that rely on active sensing of an uninformative cue (uninformative in the sense that it does not correlate with the state of the environment e) are simple to produce mechanistically. To illustrate this, we have recently constructed a synthetic system to formally show that bet hedging can be actualized by active sensing. We used the expression of the selectable/counter-selectable gene product Ura3p in the budding yeast Saccharomyces cerevisiae to model a phenotype that is specialized to two different environmental states. Yeast cells that do or do not produce Ura3p can be selected for using two different growth media (analogous to e in the model). We placed URA3 under the control of an engineered transcription factor that is activated by directly binding the hormone estradiol (Fig. 3.3). The hormone itself has not effect on the fitness of S. cerevisiae (McIsaac et al., 2013), so the fitness of the cells depends only on the match between the expression of URA3 and the growth media. To generate a bet-hedging strategy using this system, we created structured environments with homogenous concentrations of the selective media, but inhomogenous concentrations of estradiol (see Chapter2). Genotypes

41 Figure 3.3: Sensing an uninformative cue can actualize a bet-hedging strategy. (A) A schematic of a synthetic system imposing a plastic phenotypic trade-oﬀ is shown. In this system, the expression of the gene URA3 is controlled by an engineered transcription factor. The activity of the transcription factor is controlled by estradiol (shown in blue). In the presence of estradiol, the transcription factor translocates to the nucleus and catalyzes the production of Ura3p (shown as a darkening of the cytoplasm). In the presence of Ura3p, cells can grow on SC-ura, but not SC+5FOA selective media and the converse is true for cells withouth Ura3p. (B) A schematic of how estradiol acts as an uninformative cue is shown. A small volume of concentrated estradiol in the center of both SC+5FOA and SC-ura agar plates. This results in a diﬀusion gradient of estradiol (shown in blue). Since the distribution of estradiol is identical for both SC+5FOA and SC-ura plates, the concentration of estradiol contains no information about the selective environment. (C) Patterns of growth for the strain shown in (A) on SC+5FOA and SC-ura spotted plates are shown. Approximately 100,000 cells were spread on each plate and were incubated at 30C for 24hr. (D) A cartoon of a selective regime favoring a bet-hedging strategy is shown. Epochs alternated between growth on SC-ura and SC+FOA. In these conditions, only a strain plastic to estradiol was able to survive by hedging its bets.

42 that were capable of sensing estradiol diversiﬁed their phenotypes and outcompeted other strategies when e varies (Fig. 3.3), providing a constructive proof of the possibility that bet hedging can be generated by active sensing of uninformative cues.

3.4.1 Uninformative cues can be reliable

The idea that sensing uninformative cues can provide a mechanism for bet hedging may appear counterintuitive because chance external factors may not seem reliable enough to provide the consistent patterns of phenotypic diversification needed for a bet hedging strategy. In constrast, since gene expression noise is a consequence of the sampling statistics of small numbers of molecules in each cell, this noise may appear to be a firmer foundation for a bet-hedging strategy. This mirrors controversies about what is responsible for the development of organisms. A common conceptual model of development is that the genes of an organism provide a recipe or a program that pre-exists and is executed during development (Pigliucci, 2010). A contrasting view suggests that the orderly progression of development comes about as a result of a large number of equally explanatory factors (Oyama, 2000). This latter view suggests that development be described by considering all factors that are reliably present (including genes, but also external factors) during development rather than privileging those inside the nucleus. Similarly, we suggest that bet hedging is best thought of as a strategy that arises from conditions that are reliably present inside but also outside of a cell. A critical test of this view is whether there are external factors that reliably produce adaptive phenotypic diversification. Analogously to bet hedging, many organisms produce consistent ratios of sexes or mating types without adaptively sensing the ratio of sexes that would maximize individual fitness (although some organisms do plastically switch sexes (FISHELSON, 1970)). As humans, we are most familiar with mechanisms of sex determination that rely on random factors internal to the cell–the segregation of sex chromosomes dur-

43 ing meiosis. However, mechanisms of sex determination that rely on environmental factors such as temperature have evolved multiple times in reptiles and teleost ﬁsh (Bachtrog et al., 2014). Niche construction allows these organisms to maintain stable ratios of sexes, with mothers seeking out nest sites with certain temperatures, even across large environmental clines (Ewert et al., 2005; Doody et al., 2006). Although patterns of environmental variability can be very reliable, patterns of gene expression noise can be highly sensitive to the environment. Extrinsic variability dominates gene expression noise in yeast cells (Stewart-Ornstein et al., 2012). Consistent with this, one study of natural variability in gene expression noise in yeast mapped the variability to alleles in several amino acid transporters, suggesting that they could alter the sensitivity of the yeast to microenvironmental variation (Fehrmann et al., 2013b). Furthermore, since gene expression noise correlates strongly with the expression level of a gene (Newman et al., 2006), any plasticity in the gene can alter the statistical properties of the noise associated with that gene. Taken together, these factors suggest that gene expression noise will not always reliably produce diversity in phenotypes.

3.5 Conclusions

Microbes have evolved multiple ways to ensure phenotypic diversification to cope with uncertain environments. Some of these, such as phase switching, involve changes to DNA (Moxon et al., 1994), whereas others, such as persister cell formation, do not (Lewis, 2010). Since the generation of diversity plays a critical role in the fitness of microbes, determining the mechanisms by which this diversity is produced is important to undestanding their ecology as well as developing treatments to combat pathogens. The last decade has seen an explosion of interest in bet hedging as a mechanism that explains the evolution of strategies of phenotypic diversification, but determining the molecular underpinnings of these strategies has been hampered

44 by a lack of information about how the actual environments that microbes live in impact their physiology and fitness. Since microenvironmental variability can generate bet-hedging strategies, studies of phenotypic variability in homogenous environments cannot elucidate the mechanisms underlying bet-hedging in general. To understand the molecular mechanisms by which microbes produce bet-hedging strategies, studies must determine what causes variability in fitness in individual cells in realistic models of the actual environments that these organisms inhabit. Although this may seem technically daunting given the complexity of natural environments and the difficulty in tracking individual cells, some environments such as biofilms have already been replicated in the laboratory, and the physiological heterogeneity of microbes in these environments has been studied (Boles, 2004; Smukalla et al., 2008). Since active sensing is a valid mechanism of bet hedging, in another sense the experimental burden is lowered since studies seeking to simply measure the total amount of variability in these complex environments need not strictly distinguish between intrinsic and extrinsic sources of variability, which greatly simplifies experimental design. Bet hedging as an evolutionary strategy is related to the problem of decision making. Sometimes the best strategy we can come up with is to randomly decide on some course of action. In these cases, we sometimes rely on a ‘known unknown’ to ensure that our choices are truly random. We know that we don’t know what face of a coin will show up after a flip, which allows us to act randomly. As pointed out by Cooper and Kaplan (Cooper and Kaplan, 1982), this is an example of using active sensing of the environment to actualize a bet-hedging strategy. Given that variability in the cues that are sensed by organisms are both ubiquitous and reliable, this mechanism of bet hedging is likely common not just in humans, but in other organisms as well.

45 4

The Quick and the Dead: Microbial Demography at the Yeast Thermal Limit

4.1 Introduction

Population growth rate is arguably the most measured variable in all of microbiology. It is used to study the effects of genetic mutations (Breslow et al., 2008), as an indicator of physiological stress (Warringer et al., 2003), as a read-out of selection (Vasi et al., 1994), and as a proxy for fitness (Joseph and Hall, 2004). However, net population growth rate is itself a composite function of two more fundamental demographic variables — the birth and death rates of individuals — and measuring only the population growth rate confounds them. Furthermore, birth and death rates frequently change with the age. Because analyses of population growth play such a critical role in studies of microbes, disentangling the effects of birth, death, and aging on population growth should lead to a better understanding of microbial physiology, ecology, and evolution. In ecological terms, the range of environmental conditions where the population growth rate of a species is positive corresponds to what the ecologist G. E. Hutchin-

46 son called a species’ fundamental niche (Hutchinson, 1957). The limits of a species’ fundamental niche are determined by a combination of physiological, genetic, and demographic factors. One strategy for identifying speciﬁc factors that limit the environments where an organism can live is to dissect the demographic mechanisms that drive decreases in population growth rate when a population is exposed to conditions near those limits. Is a decrease in population growth rate driven by a population wide decline in reproductive output as environmental conditions approach their limit? Al- ternately, are there increases in mortality, perhaps in an age-structured manner, that limit growth? How do such environmentally induced shifts in demographic param- eters vary with genetic background? Questions like these can only be addressed by characterizing the fecundity, mortality, and age of individuals in populations. Such individual level characterization is particularly challenging in microbes due to their small size, large populations, and the consequent challenges of tracking or recovering individual cells. Temperature is an environmental variable that plays a critical role in limiting the fundamental niches of most microorganisms. Although the total temperature range

over which microbial life is found is quite large — at least ´5 ˝C to 113 ˝C(Pikuta et al., 2008) — most species are confined to a relatively modest range of temperatures that support robust growth. The ability to maintain a relatively high growth rate at high temperatures is thought to be critical to the ecological success of the budding yeast Saccharomyces cerevisiae. S. cerevisiae is able to grow at higher temperatures relative to closely related species (Salvadóet al., 2011; Warringer et al., 2011), and the production of heat during fermentation enables it to dominate initially diverse communities of microbes during wine-making (Goddard, 2008). Thermo-tolerance also varies widely between strains of S. cerevisiae (Liti et al., 2009) and has a complex genetic basis (Steinmetz et al., 2002; Sinha et al., 2006, 2008; Parts et al., 2011). This genetic diversity for growth at high temperatures may also affect human health; a

47 high thermal limit has been hypothesized to contribute to the opportunistic invasion of human hosts by S. cerevisiae (McCusker et al., 1994). Using a newly developed method, called TrackScar, that allows us to simulta- neously measure the birth rate, viability, and age of individual cells, we analyzed a genetically diverse panel of yeast strains during growth near their upper thermal limits. We find that disparate demographic and physiological responses limit population growth of different strains during heat stress. Either fecundity or mortality can be necessary and sufficient to explain slow population growth, depending on genetic background. Furthermore, the age-structure of mortality can differ with genetic background, leading to both “premature senescence” as well as “early life mortality.” We show that in one genetic background early life mortality likely results from a temperature sensitive mutation affecting mitochondrial inheritance. By measuring the demographic rates of individuals, we elucidate the diverse physiological and demographic factors that limit the fundamental niche of budding yeast.

4.2 Results

4.2.1 TrackScar measures the fecundity and mortality of single cells

During the process of asexual mitotic reproduction (budding) in S. cerevisiae and related yeasts, ring-shaped scars of chitin are formed on the cell wall at the site of cytokinesis in both mother and daughter cells. These chitinous rings, known as “bud scars”, provide a cellular record of the number of daughters an individual cell has produced. Bud scars can be visualized using ﬂuorescently labeled compounds that preferentially bind to chitin (Pringle, 1991). Wheat germ agglutinin (WGA), a lectin, is a particularly suitable bud scar stain because it can be can be conjugated to a variety of ﬂuorophores and has low background staining of the cell wall. We developed an experimental method called “TrackScar” for measuring the reproductive output (fecundity) of individual yeast cells based on the sequential stain-

48 ing of bud scars (Fig. 4.1A,B). The principle behind TrackScar is simple: 1) label a population of yeast cells at two time points (t0, t1), using different fluorophores at each time (e.g. cyan at t0, yellow at t1); 2) image cells that were stained at both time points; and 3) quantify the number of cell divisions individual cells have un- dergone in the given time interval by counting the difference in the number of bud scars produced (e.g. no. of yellow scars ´ no. of cyan scars). The number of daughter cells produced in a given time interval is a discrete and unambiguous measure of the rate of fecundity at the single cell level. Detailed experimental procedures describing TrackScar staining, microscopy, and image analysis are provided in the supplementary information. Exposing cells to stressful conditions is often accompanied by a decrease in fecundity, with extreme stress leading to cell death. Since two-color TrackScar only measures the number of daughters that a cell produces in one interval, it is possible that cells with low observed fecundity could have died before the first stain (WGA stains both live and dead cells), or in the interval between stains. In order to distinguish between slowly dividing but still viable cells and cells that are dead, the TrackScar method can be extended to include a “recovery phase” (in this case six hours) under permissive growth conditions (30 ˝C), followed by staining with a third fluorophore. Assuming the recovery phase is sufficiently long, cells that show at least one reproductive event during the recovery phase are still viable; cells that show no growth during recovery are either dead or severely growth arrested (Fig. 4.1C,D). Three-color TrackScar provides more information about individual cells, but data collection is more difficult because mother cells are further diluted during the recovery phase and because more bud scars must be counted. We performed control experiments to demonstrate that: 1) TrackScar can detect subtle differences in fecundity between individuals; 2) that staining cells using WGA does not affect their division rate; 3) and that six hours of recovery is sufficient to

49 A B t growtht count 0 1 3 6 7 * * * * * * * * * * * * * * * * * *

C live D high temp. recovery live dead live **

** dead

** Figure 4.1: TrackScar measures the fecundity and viability of individual yeast. A) A schematic two-color TrackScar is shown. To measure the reproductive output of individuals, cells are: 1) stained with wheat germ agglutinin (WGA) conjugated to Alexa488 (cyan) at t0; 2) grown in the absence of the stain; 3) stained with WGA conjugated to tetramethylrhodamine (yellow) at t1. New buds (labeled with asterixes) are only stained by the second dye. B) Examples of haploid S288C cells that produced diﬀerent numbers of daughters during 6 hr of growth at 30 ˝C in YPD are shown. C) A schematic of three-color TrackScar is shown. Two colors of WGA (cyan and yellow) are used to measure the fecundity of cells during heat stress. To determine viability, the cells are transferred to permissive conditions and then stained with a third color of WGA (magenta). Viable cells produce new buds that are only stained by the third dye. D) Examples of diploid YJM693 cells that were either viable or inviable after exposure to growth at 35.5 ˝C are shown. Arrows highlight one of the bud scars produced by live cells in permissive conditions. All micrographs are maximum intensity projections of z-stacks. Scale bar shows ﬁve microns.

distinguish live from dead cells. These results are described in ChapterA.

4.2.2 Identifying diverse responses to heat stress

The fundamental niche of a species is limited by the range of temperatures (‘thermal limits’) in which it can survive. Microorganisms such as yeast display a characteristic pattern of population growth across this range, with growth rate falling rapidly as temperatures reach their upper thermal limit (Ratkowsky et al., 1982). Within a species the thermal limit can vary substantially, reﬂecting genetic diversity. To disentangle the demographic basis of heat sensitivity we applied TrackScar to a

50 genetically diverse panel of S. cerevisiae strains growing in a range of thermal regimes.

Identifying slow growing yeast populations

We estimated the maximum population growth rates for a panel of 93 sequenced yeast strains (Strope et al., 2015), across a range of temperatures from 21 ˝C to 40 ˝C, by measuring the time-varying optical density of liquid cultures grown in a spectrophotometer (Fig. 4.2A). Consistent with previous studies, the population growth rate of individual strains gradually increased with temperature until reaching a maximum whereupon they decreased (Ratkowsky et al., 1982; Salvadóet al., 2011). At low temperatures, we observed only small differences between genetic backgrounds in their population growth rates. However at 35.5 ˝C and above there is substantial strain-to-strain variation in maximal population growth rate. To identify heat sensitive versus robust strains, we calculated the ratio of maximum growth rates at 35.5 ˝C, a potentially stressful temperature with high between- strain variance, relative to 30 ˝C, a standard permissive temperature. Strains with a thermal growth ratio near 1.0 were considered robust to thermal stress, while those with ratios significantly lower than 1.0 were classified as heat sensitive (Fig. 4.2B). While the population growth rate provides important physiological information, TrackScar provides much richer insights into how population fecundity changes across environments. For example, Figure 4.2C illustrates how the population distributions of fecundity change over a temperature range of 30 ˝C to 40 ˝C, for diploid S288c cells. S288c grows well at both 30 ˝C and 35.5 ˝C. By contrast, at higher temperatures the distribution of fecundities shifts left and there is a noticeable increase in the variance.

4.2.3 Heat sensitivity is associated with the emergence of heterogeneous subpopulations

At 35.5 ˝C, populations of cells from heat sensitive strains grow more slowly than do those of robust strains. Two alternative explanations for slower growth are: 1)

51 A 0.5 0.4 0.3 0.2 S288C Growth rate √ Growth 0.1 All others

25 30 35 40 B ˚Celsius S288C 10 YJM996 YJM693 5

0 0.250.50 0.75 1.00 1.25 Number of strains Growth at 35.5˚C / Growth at 30˚C C 60% 30.0˚C 35.5˚C 37.0˚C 40% 38.5˚C 40.0˚C 20% Frequency 0% 0 1 2 3 4 5 6 7 8 Daughters in 6hr

Figure 4.2: There is extensive genetic diversity in thermotolerance among environmental yeast isolates. A) The square-root of the maximum growth of each of 93 yeast strains is shown as a function of temperature. The strain S288C is shown in red. B) A histogram of the growth rate of each of 93 yeast strains at 35.5 ˝C normalized to their growth rate at 30 ˝C is shown. C) Histograms of the number of daughters produced by cells in the strain S288C during 6 hr of growth at 30 ˝C (brown circles), 35.5 ˝C (gold triangles), 37 ˝C (green squares), 38.5 ˝C (black crosses), and 40 ˝C (red crossed boxes) are shown. Error bars are the standard error of the mean for three biological replicates. the population as a whole experiences a uniform shift to lower fecundity; or 2) some cells continue to reproduce at the unstressed rate, while other cells in the population experience a more drastic decrease in fecundity. These alternative explanations are unresolvable based on standard bulk population measurements but are easily distinguished using TrackScar.

52 Figure 4.3 shows the distributions of fecundity at 30 ˝C and 35.5 ˝C for four robust (Fig. 4.3A) and four heat sensitive (Fig. 4.3B) strains. At 30 ˝C most strains exhibit unimodal and approximately symmetric fecundity distributions, although the mean and variances of fecundity diﬀer between backgrounds. At 35.5 ˝C robust strains continue to exhibit symmetrical fecundity distributions, with only modest diﬀerences, if any, in mean fecundity. By contrast, many sensitive strains (Fig. 4.3B) have asymmetric fecundity distributions, with a substantial increase in slow growing or non-growing cells. In many cases the distributions of fecundity of heat sensitive strains exposed to heat stress are bimodal, suggesting the existence of two subpopulations (Fig. 4.3; Fig. A.4). The one heat sensitive strain that did not exhibit an asymmetric fecundity distribution at higher temperatures, YJM1478, showed a uniform decrease of approximately two divisions per six-hour interval.

4.2.4 Increased mortality explains slow population growth for some heat sensitive strains

One mechanism that could lead to asymmetric and bimodal fecundity distributions we observe during heat stress is if some cells died before or during the interval in which we observe their fecundity. To explore this possibility, we used the three- color TrackScar assay to test whether cells with low fecundity during heat stress were viable when the stress is removed. We focused our analyses on three strains — S288C, YJM693, and YJM996 — that have distinct fecundity distributions near their thermal maxima (Fig. 4.2C, Fig. 4.3). Since S288C does not exhibit a growth rate decrease at 35.5 ˝C, we assayed this strain at 40 ˝C, a temperature where it exhibits a distinct decrease in fecundity and a large number of non-dividing cells. The other two strains were assayed at 35.5 ˝C. We measured fecundity for six hours during heat stress, incubated the cells overnight at 4 ˝C, then transferred the cells to 30 ˝C and measured the number of divisions during a subsequent six-hour period.

53 A YJM1399 YJM195 30.0˚C 40% 35.5˚C 20% 0% YJM975 YJM470

Frequency 40% 20% 0% 0 2 4 6 8 0 2 4 6 8 Daughters in 6hr B YJM693 YJM996 30.0˚C 40% 35.5˚C 20% 0% YJM1248 YJM1478

Frequency 40% 20% 0% 0 2 4 6 8 0 2 4 6 8 Daughters in 6hr

Figure 4.3: Strains that are sensitive to heat stress frequently have subpopulations with very low fecundity. Plots showing histograms of the number of daughters produced in 6hr at 30 ˝C (brown circles) and 35.5 ˝C (gold triangles) are shown for A) robust and B) sensitive strains are shown. A single representative replicate is shown for each temperature for each strain.

We used three-color TrackScar to determine whether increased mortality explained the low observed fecundity of cells during heat stress. We compared the number of daughters produced during heat stress of cells that either survived or died during heat stress with the fecundity of unstressed cells. S288C cells with low fecundity during heat stress are frequently alive at the end of the period of stress, but have much lower fecundity during recovery than unstressed cells (Fig. 4.4). In contrast, viable YJM996 cells frequently produce as many daughters as unstressed cells, but with a broad distribution. Remarkably, and in contrast with the other

54 two strains we examined, the distribution of fecundity among YJM693 cells that survived heat stress was identical to that of unstressed cells (Kolmogorov-Smirnov test; p = 0.99). This indicates that heat stress causes YJM693 cells to split into two sub-populations — one population of quickly dividing viable cells and another population of cells that dies during the stress. Taken together, these results show that in some genetic backgrounds mortality can be the sole cause of slow population growth during heat stress, while in other strains a combination of increased mortality and reduced individual fecundity occurs.

Unstressed all S288C 60% live Heat stress dead 40%

20%

0% YJM996 60%

40%

20%

0% YJM693 60% Within temperature frequency temperature Within 40%

20%

0% 0 1 2 3 4 5 6 7 8 Fecundity during heat stress

Figure 4.4: Cells were heat stressed at either 35.5 ˝C or 40 ˝C, then assayed for viability at 30 ˝C. The histograms plot the fecundity during heat stress for cells that were subsequently classiﬁed as viable (magenta triangles) and inviable (black circles) from strains S288C, YJM996, and YJM693. The fecundity of unstressed cells is plotted as a dashed line with green crosses. Error bars are the standard error of the mean for three biological replicates.

55 4.2.5 Fecundity can be positively or negatively associated with age during stress

Yeast mother cells faithfully segregate sub-cellular components to their daughter cells, but also asymmetrically retain aging factors such as damaged proteins and mitochondria (Lai et al., 2002; McFaline-Figueroa et al., 2011; Aguilaniu et al., 2003). Genetic variation could lead to temperature sensitive mutations that affect either of these processes, which could lead to lower fitness in either younger or older cells. Individual yeast cells produce a finite number of daughters over their “replicative lifespan” before senescing (Mortimer and Johnston, 1959). Since TrackScar works by measuring the replicative age of a cell at the beginning and end of an experiment, the age structure of fecundity emerges naturally during analysis. We will call a group of cells with the same replicative age at the beginning of the experiment a “cohort.” To gather sufficient data about older cohorts, we sought out and imaged old cells rather than imaging random cells (see Section A.3.2). We measured the average fecundity of cohorts of cells in the strains YJM693, YJM996, and S288C. To test whether fecundity was affected by age, we excluded the fecundity of daughter cells because the fecundity of daughters is lower than mothers due to an extended G1 phase of the cell cycle (Hartwell and Unger, 1977)(Fig. A.1B). At 30 ˝C, neither the average fecundity of cohorts YJM693 nor YJM996 was significantly affected by age (Fig. 4.6A)(linear model, p ą 0.1). The average fecundity of a cohort of S288C was not affected by its age at any temperature (Fig. 4.5). In contrast, at 35.5 ˝C, replicative age significantly affects the average fecundity of a cohort in the strains YJM693 and YJM996 (linear model, p “ 0.012 and p “ 6.024 ˆ 10´8, respectively). Interestingly, while YJM693 cells produce an average of 0.17 fewer daughters in six hours per cohort when heat stressed, YJM996 cells produce an average of 0.37 more daughters in six hours per cohort (Fig. 4.6A).

56 6

2 30.0˚C Fecundity +/- SEM Fecundity 35.5˚C 37.0˚C 38.5˚C 40.0˚C

1 2 3 4 5 6 7 8 9 Cohort

Figure 4.5: The fecundity of S288C cells is not age structured at any temperature. The average fecundity of cohorts of S288C cells during 6hr during growth at 30 ˝C (brown circles), 35.5 ˝C (gold triangles), 37 ˝C (green squares), 38.5 ˝C (black crosses), and 40 ˝C (red crossed boxes). Error bars are the standard error of the mean for at least two biological replicates. Only cohorts with at least three cells in each of the biological replicates are shown.

4.2.6 Heat stress can cause premature senescence or early life mortality

The fecundity of YJM693 cells is lower in older cells (Fig. 4.6A). Since we found that the lower fecundity of YJM693 cells at 35.5 ˝C is due to cell death (Fig. 4.4), we tested if there was increased mortality in older cohorts of YJM693. YJM693 cells that divided less than four times during a six-hour period of heat stress were dead by the end of the experiment (Fig. 4.4). Consistent with our hypothesis, we found that the older a cohort of cells was, the more cells in that cohort failed to divide at least four times in six hours, indicating that cell death was more common in older cohorts (Fig. 4.6B). Since by deﬁnition each cell in a cohort had survived

57 at least long enough to enter the cohort, this indicates that the likelihood of death increases with replicative age at 35.5 ˝C in YJM693 and cannot be explained by a model of constant survival risk. Using logistic regression, we estimate that there is a

4.9%˘1.2% increase in the probability of death for each additional unit of replicative

A 30.0˚C 35.5˚C 6

Fecundity YJM693 YJM996 0 1 2 3 4 5 6 1 2 3 4 5 6 7 8 910 Cohort B YJM693 YJM996

75%

50%

Mortality 25%

0% 1 2 3 4 5 6 7 1 2 3 4 5 6 7 8 910 Cohort

Figure 4.6: Genetic background determines the age structure of fecundity and mortality during heat stress. A) The average fecundity of a cohort during 6hr growth at either 30 ˝C or 35.5 ˝C for the strains YJM693 (blue circles) and YJM996 (black triangles). Error bars are the standard error of the mean for at least three biological replicates. Only cohorts with at least three cells in each of the three biological replicates are shown. B) The mortality of cohorts of YJM693 and YJM996 cells is shown. YJM693 cells were considered “dead” if they divided four times or fewer growth at 35.5 ˝C, whereas YJM996 cells were considered “dead” if they did not divide at all during six hours of growth at 35.5 ˝C. Error bars show the standard error of the mean for at least three biological replicates. Only cohorts with at least ﬁve cells in each of the three biological replicates are shown.

58 age in this strain during heat stress. These results indicate that the probability of dying increases substantially with each additional replicative event during heat stress in YJM693, suggesting that thermal stress aﬀects an asymmetrically inherited aging factor in this genetic background. In contrast to our ﬁndings for YJM693, the fecundity of YJM996 cells is higher in older cells (Fig. 4.6A). Although not all cells with low fecundity in YJM996 died during the exposure to stress, cells that produced no daughters during the six hours of growth at 35.5 ˝C were viable less than 5% of the time (Fig. 4.4). Therefore, we considered these cells dead for the purpose of estimating age-structured mortality. Younger cohorts in the YJM996 genetic background have substantially higher mortality than older cohorts. More than 80% of the youngest cells did not produce a daughter cell during 6 hours of recovery at 30 ˝C, which strongly suggests that they had died (Fig. 4.6B). Cohorts that had divided only one or two times also had substantially elevated mortality. In contrast, cells in cohorts six and above showed substantially reduced mortality (Fig. 4.6B). YJM996 thus exhibits an early life mortality phenotype in response to heat stress. Interestingly, the high mortality of young cells leads older cells to accumulate in the population (Fig. A.3). This pattern of mortality in young cells in consistent with heat stress impacting the segregation of a critical cellular component.

4.2.7 Early life mortality is associated with a failure to inherit mitochondria

The early life mortality phenotype of YJM996, revealed by heat stress, suggests that this strain may have a temperature-sensitive defect in the segregation of some cellular component to daughter cells. Failure to properly segregate mitochondria (e.g. in an ∆mmr1 strain) leads to sub-populations of short- and long-lived cells (McFaline- Figueroa et al., 2011), which could generate mortality patterns similar to the ones in YJM996. We examined the morphology of mitochondria at both 30 ˝C and 35.5 ˝C

59 A B C YJM996 30˚C YJM996 35.5˚C S288C 35.5˚C

t0 t1 DAPI mito

D 30C 37C 100% ● Threads Clumps 75% Threads & Clumps No mito.

50%

● 25% ● ● ●

Fraction of population Fraction ● ● 0% ● ● 0 2 4 6 0 2 4 6 Daughters in 6hr

Figure 4.7: A defect in mitochondrial inheritance is associated with early life mortality in the YJM996 genetic background. Representative images are shown of TrackScar staining at zero hours (cyan) and six hours (yellow), DAPI staining (blue), and mitochondria-localized GFP (magenta) for a ∆ura3 derivative of the strain YJM996 at A) 30 ˝C and B) 35.5 ˝C, and a ∆ura3 S288C derivative at C) 35.5 ˝C. Each budding cell is staged in the process of nuclear division to facilitate comparison of mitochondrial inheritance. D) The fecundity of cells with thread-like mitochondrial morphology, clumpy morphology, a mixture of clumps and threads, and cells that lack mitochondria is shown for cells in the YJM996 genetic background at both 30 ˝C and 35.5 ˝C. Error bars show the standard error of the mean for three biological replicates.

60 in strains YJM693, YJM996, and S288C using a mitochondrially localized GFP expressed from a low copy number plasmid. Since this plasmid was uracil selectable, we used ∆ura3 derivatives of these strains and grew them in synthetic media lacking uracil. The strain YJM996 has a temperature sensitive defect in mitochondrial morphology and inheritance. In S288C and YJM693, at both 30 ˝C and 35.5 ˝C, the mitochondria form a branching, thread-like network. In contrast, the mitochondria of YJM996 grown at 35.5 ˝C formed large globular clumps (Fig. 4.7a). This clumpy morphology was also present with low (˜10%) penetrance at 30 ˝C (Fig. 4.8). We noticed that the buds of YJM996 cells often lacked mitochondria even near the end of mitosis (Fig. 4.7a), suggesting that the mitochondria of this strain may be inherited ineﬃciently. Consistent with this, at 30 ˝C and 35.5 ˝C, approximately 10% and 75%, respectively, of YJM996 cells lack mitochondria entirely (Fig. 4.8). We observed cells lacking mitochondria to contain nuclei (Fig. 4.7a), and vacuoles (data not shown), suggesting that the inheritance of other organelles is not aﬀected. Since mitochondria cannot be created de novo and must be inherited from mother cells, this indicates that mitochondrial inheritance is inhibited by high temperatures in YJM996. We hypothesized that this mitochondrial inheritance defect was the cause of the the early life mortality phenotype in YJM996. Consistent with this, we found that YJM996 cells that cells that lack mitochondria divide fewer times than cells that contain mitochondria at both 30 ˝C and 35.5 ˝C (Fig. 4.7). Clumpy mitochondria were found with low frequency at 30 ˝C, and we hypothesized that these mitochondria would be associated with slower growth. However, cells with clumpy mitochondria produced the same number of daughters as cells with wild-type mitochondria, whereas cells without mitochondria produced few or no daughters at either temperature (Fig. 4.7). This indicates that mortality in this strain is not caused by a defect

61 in mitochondrial morphology per se, although the altered morphology of the mitochondria could contribute to the defect in inheritance. Taken together, these results strongly suggest that YJM693 grows slowly at elevated temperatures due to a defect in the segregation of mitochondria, leading to high mortality in young cells.

YJM996 YJM693 S288C ● clumps 60% clumps and threads no mitochondria 40% ● 20% ● ● 0% ● ● ● 30 35.5 30 35.5 30 35.5 Percent of population of Percent Temperature

Figure 4.8: Heat stress changes the frequency of cells with abnormal mitochondrial morphology in the strain YJM996. The percentage of the population with abnormal mitochondria at 30C and 35.5C is shown for the strains CMY177 (YJM693 genetic background), CMY178 (YJM996 genetic background), and CMY182 (S288C genetic background). Mitochondria were grouped into the classes: clumpy (red circles), clumpy with some threads (orange triangles), and no mitochondria (blue squares).

4.3 Discussion

We have dissected the demographic mechanisms that delimit the fundamental niche of genetically diverse S. cerevisiae strains. We find that different strains placed at temperatures near the upper limit of their niche grow slowly due to the impact of heat on distinct demographic rates and through distinct physiological mechanisms. Slow population growth can be a result of higher mortality, lower fecundity, or a combination of the two. We find that the risk of dying during chronic heat stress can either increase or decrease with age, suggesting the effect of genetic variation on the

62 segregation of sub-cellular components during asymmetric cell division. Our results highlight the importance of taking into account the behavior of individual cells in order to understand the factors that determine the range of environments in which microbes can thrive.

4.3.1 TrackScar is a powerful method to measure fecundity and mortality

The growth rate of a population of microbes in the laboratory at a particular temperature can accurately predict the actual environments in which a microbe can thrive (Madigan et al., 2015). The growth rate is, in turn, the sum of the rates of birth and death, and the eﬀect of temperature on growth rate could be due to a change in either. TrackScar allows one to measure age-structured rates of birth and death in populations of budding yeast without the need for trapping or microscopic tracking of cells. This technique is analogous to “mark-release-recapture” methods used to study larger organisms, greatly simplifying experimental design and opening the door to studies of microbial demography in environments inaccessible to microscopic tracking of individuals.

4.3.2 Asymmetries during cell division contribute to phenotypic diversity

The rates of birth and death of organisms are ultimately due to the integration of the underlying molecular processes that make up their physiology. Single-celled organisms rarely behave identically, even if they share the same genome and the same environment. Random events like stochastic gene expression can cause these diﬀer- ences(Spudich and Koshland, 1976; Maamar et al., 2007; Bumgarner et al., 2012). However, predictable events such as cell cycle and aging (Avery, 2006; Levy et al., 2012), and epigenetic inheritance of growth rate (Ziv et al., 2013; Levy et al., 2012) can also contribute. Genetic variation can impact cellular individuality, indicating that it could be selected for during evolution (Ziv et al., 2013; Fehrmann et al.,

63 2013b; Holland et al., 2013). This individuality makes the average a poor predictor of the behavior of individuals, and can obscure the physiological basis of population level phenotypes. During heat stress, changes in the mean population growth rate obscure the behavior of individual yeast cells, as shown by the asymmetric and frequently bimodal distributions of the fecundity of individuals. We found age to be an important predictor of fecundity and mortality during heat stress with either young or old cells disproportionately dying as a result of heat stress depending on the genetic background. In at least one case, this age structure is a consequence of the failure of cells to appropriately segregate sub-cellular components (see below). The ability to easily distinguish mother and daughter cells in budding yeast make these patterns particularly apparent, but asymmetries also lead to aging in ostensibly symmetrically dividing organisms such as Escherichia coli of Schizosaccharomyces pombe (Barker and Walmsley, 1999). We speculate that genetic variation in cellular mechanisms that establish or normalize asymmetries between mother and daughter cells may be an important cause of phenotypic variability in other organisms.

4.3.3 Mortality or slower division rate can limit growth during heat stress

Slow growth during heat stress could be due to increased mortality, a slower division rate, or some combination of the two. A previous study of Caulobacter crescentus in a microfluidic device found that heat stress causes both slower division rates as well as an increase in mortality, but only examined a single strain (Iyer-Biswas et al., 2014). We found that genetic variation affects the relative contribution of mortality and fecundity to stress induced changes population growth rate. For example, increased mortality was both necessary and sufficient to explain the slow growth of the clinical isolate YJM693. Cells that survived chronic heat stress in this strain had fecundity identical to unstressed cells, whereas nearly all cells with low fecundity were dead.

64 In contrast, cells with low fecundity during heat stress in the clinical isolate YJM996 or the genomic reference strain S288C are sometimes alive. Our results indicate that changes in population growth rate should be interpreted cautiously, since they can reﬂect the eﬀects of distinct demographic mechanisms.

4.3.4 Fecundity and mortality can be age structured during heat stress

Budding yeast reproduces by asymmetric division; mother cell and daughter cells differ both in size and molecular make-up. Active processes ensure that sufficient quantities of organelles and other sub-cellular components are segregated to daughter cells (Chernyakov et al., 2013; Marston, 2014) while at the same time, damaged proteins, organelles and other aging factors are asymmetrically retained in the mother cell, eventually leading to senescence (Lai et al., 2002; McFaline-Figueroa et al., 2011; Aguilaniu et al., 2003; Sinclair and Guarente, 1997; Dillin et al., 2014). Mutations affecting these two processes could lead to different demographic outcomes. Aber- rant production or distribution of an asymmetrically distributed factor could lead to premature aging, whereas the failure to produce or distribute a critical sub-cellular component could lead to newly born cells dying.

Elevated mortality can aﬀect either young or old cells

Older cohorts in the strain YJM693 produce fewer daughters on average than younger cohorts and contain a greater percentage of dead cells. Constant mortality with replicative age cannot explain this pattern because by deﬁnition a cell must have survived long enough to enter a cohort, indicating there is an increased likelihood of dying in older cohorts in this strain. Senescence is deﬁned as a increased likelihood of dying in older cells, indicating that YJM693 cells senesce prematurely at elevated temperatures. Whatever causes senescence in budding yeast must be asymmetrically retained in

65 mother cells (Kennedy et al., 1994). A number of molecular mechanisms are associated with senescence in good conditions in yeast including autonomously replicating circularized rDNA, damaged vacuoles and mitochondria, and the aggregation of damaged proteins (Lai et al., 2002; McFaline-Figueroa et al., 2011; Aguilaniu et al., 2003; Sinclair and Guarente, 1997; Dillin et al., 2014), although some of these remain controversial (Fehrmann et al., 2013a). We speculate that the strain YJM693 contains a temperature sensitive mutation in a gene required for normal replicative lifespan. In contrast to the strain YJM693, older cells in the strain YJM996 were more fecund than younger cells. The low fecundity of young cohorts in these strains is partly, but not entirely, due to very high mortality in young cohorts with around 75% of the youngest cells failing to ever produce a daughter cell. In contrast, the mortality of older cohorts is low, leading them to accumulate in the population. The force of natural selection on fecundity of a cohort scales with the relative survivorship of that cohort (Hamilton, 1966). This implies that natural selection on the fecundity of older cells is much stronger in this strain, but only during heat stress. We found that the early life mortality of YJM996 is linked to a defect in mitochondrial inheritance. YJM996 cells contain mitochondria with an aberrant globular morphology during heat stress, and daughter cells in this strain frequently fail to inherit mitochondria. Although mtDNA is dispensable in S. cerevisiae, the mitochondria themselves are essential (Kispal et al., 2005; Chernyakov et al., 2013; Frederick et al., 2008). This strongly suggests that the failure to inherit mitochondria causes the high mortality in young cells in this strain, although we cannot rule out that other essential components may fail to be segregated as well.

4.3.5 Diverse physiological factors delimit the niche of S. cerevisiae

The ability to grow at high temperatures in the presence of alcohol is a key innovation critical to the ecological success of Saccharomyces cerevisiae (Goddard, 2008). In

66 this study, we examined the standing genetic variation present within S. cerevisiae for the ability to grow at high temperatures. The ability of cells to grow at high temperatures has been hypothesized to be a trait that enables pathogenesis in yeast, but we show that clinical isolates can also be sensitive to heat. Thermal tolerance in yeast is a genetically complex trait (Steinmetz et al., 2002; Sinha et al., 2006, 2008; Parts et al., 2011), and consistent with this we found that heat can operate on distinct demographic functions to lower growth rate, reflecting effects on different cellular processes involved in asymmetric division in budding yeast. Our results connect physiology to demography, and shed light on how genetic variation affects the fundamental niche of budding yeast.

4.4 Methods

All of our experiments were conducted using clonal populations grown in YPD media. Before staining, cells were propagated for at least 16 hours in log-phase growth at the temperature being surveyed. Cells were first stained with Alexa488-conjugated wheat germ agglutinin (WGA) for 15 min. Cells were washed once in YPD and diluted to 0.33ˆ106 cells/ml, then grown for six hours. Cells were then fixed in PBS + 8% formaldehyde, and stored until imaging. Before imaging, cells were stained as above with tetramethylrhodamine (TMR)-conjugated WGA. Widefield Z-stacks were acquired, deconvolved, and then processed using a maximum pixel intensity projection. The numbers of buds stained by each dye were counted by manually from these projections. Growth curves were collected on a Tecan Sunrise over a 48hr period with measurements every 15 minutes. See Section A.3 for detailed methods including for three-color TrackScar.

67 Appendix A

Supplementary Information for The Quick and the Dead

A.1 Introduction

Measuring fecundity and mortality in microorganisms is challenging because it requires microscopic observation of individuals. Individuals can be followed on agar pads (C D Kelly, 1932; Burns, 1956; Hartwell and Unger, 1977) or in microﬂuidic devices (Wang et al., 2010; Lee et al., 2012; Fehrmann et al., 2013a) using timelapse microscopy. However, observation of cells on agar pads is limited to a few generations because of crowding during micro-colony formation, and microﬂuidics, while powerful, can be technically demanding. Furthermore, each technique requires isolating cells from the context of interactions with other individuals; such interactions are an important mediator of physiological and developmental processes (Shapiro, 1998; Gore et al., 2009). In this paper, we present a new method to measure the fecundity and mortality of individual yeast that we call “TrackScar”. TrackScar has several advantages over other techniques. Rather than relying on continuous observation, individuals

68 are “marked and released” between applications of fluorescent stains, simplifying the experimental design and creating the possibility of observing individuals in any environment or community. Furthermore, since the number of bud scars represents the replicative age of yeast cells, the age structure of fecundity and mortality emerge naturally during the analysis of the data generated by TrackScar experiments. In this supplement, we present data that demonstrate the sensitivity and accuracy of TrackScar, evidence that TrackScar minimally affects the physiology of cells, and analyze the effect of heat stress on the distribution of ages in a population.

A.2 Results

A.2.1 TrackScar minimally aﬀects cellular physiology

A potential concern with any vital-staining method is the possibility that it could change the cell physiological processes of interest. For example, another bud scar stain, calcofluor white, increases cell mortality by interfering with cell wall synthesis (Roncero and Durán, 1985). To explore this possibility, we compared division time estimates based on TrackScar to results from previous studies using timelapse microscopy. Using TrackScar we estimated the average division time to be 73.9 minutes for haploid cells of the genomic reference strain S288c grown in rich-media conditions. This is very similar to an estimate of 72.6 minutes derived previously for this same strain using timelapse microscopy (Lord and Wheals, 1981). To further explore the physiological impact of WGA staining on cellular division rates, we carried out a TrackScar experiment in which the time interval between the first and second stain was varied between one and six hours. We reasoned that if WGA staining is a significant cellular stress that the reproductive rates immediately following the first stain should be systematically lower than in later time points, because extended incubation would allow cells to recover from any physiological perturbations caused by staining. We carried out this experiment for nine genetically diverse S. cerevisiae

69 A B Daughters 1 bud old 1.0 40% mothers

0.5 20% Frequency

0.0 0% Repro. rate (buds/hr) 1 2 3 4 5 6 2 3 4 5 6 7 8 Hours after first stain Daughters produced in 6hr

Figure A.1: TrackScar is sensitive and minimally affects cellular physiology. A) The average number of daughters produced in a one hour interval by cells in nine genetically diverse yeast strains is shown as a function of time since the first WGA stain. A linear regression line is plotted as a blue dashed line. B) Histograms of the number of daughters produced by daughter cells (orange circles) and mother cells that had divided once (blue triangles) in a 6hr interval for are shown. Error bars show the standard error of the mean for three biological replicates. strains. We found no evidence that reproductive rates at earlier time points were any lower than later time points (Fig. A.1A). Indeed, our data show that cells at time points immediately following the first stain produce slightly more daughters than those at later time points (linear model; p = 0.03). Taken together, these two sets of comparisons suggest that TrackScar does not significantly perturb the division rate of S. cerevisiae cells.

A.2.2 TrackScar provides a sensitive measure of diﬀerences in fecundity

To determine the sensitivity of TrackScar, we tested whether this method could detect the increased cell cycle time of daughter cells. S. cerevisiae reproduces by asymmetrical division, and smaller daughter cells must grow longer until they reach a critical size threshold before dividing for the ﬁrst time (Hartwell and Unger, 1977; Johnston et al., 1979; Lord and Wheals, 1981). Thus, daughter cells should have lower fecundity than mother cells that had already divided at least once. Consistent

70 with this expectation, daughter cells of haploid strain S288C produced an average of 4.9 daughters in a six-hour period, whereas mother cells produced an average of 5.4 daughters (Fig. A.1B). This difference is significant (Paired t-test; n=3; p = 0.030). These results indicate that TrackScar can accurately detect small differences in fecundity between individual cells.

A.2.3 Population Growth Rate and Mean Fecundity Are Well Correlated

The diﬀerent growth rates of sensitive and robust strains provided another oppor- tunity to test the accuracy of TrackScar. We used two-color TrackScar to estimate fecundity at 30 ˝C and 35.5 ˝C for 8 sensitive and 13 robust strains. We compared these estimates of average fecundity to the estimates of growth rate obtained from growth curves. The average fecundity of cells measured using TrackScar and the maximum population growth rate measured by optical density at 35.5 ˝C are well-

correlated (r2 “ 0.62; Fig. A.2).

A.2.4 Six Hours of Recovery Is Suﬃcient To Distinguish Live from Dead Cells

Three-Color TrackScar relies on a sufficiently long recovery phase to distinguish between live and dead cells. We tested whether six hours is sufficient to distinguish live from dead cells by determining if YJM693 cells that did not divide after six hours of recovery would subsequently form a colony on an agar plate. We grew YJM693 to early-log phase at 35.5 ˝C as above and stained the cells with Ax488-WGA. After washing off the stain, we spread the cells on a plate of YPD and randomly selected cells from the patch of cells on the plate. Cells were immediately singled and subsequently moved into a grid on the YPD plate. If a cell divided in between the time that the cell was singled and when it was moved to the grid, both the mother cell and its progeny were placed in the same space on the grid. The plate was incubated for 6hr at 30 ˝C and colonies were inspected for new daughter cells

71 using a microscope. After 48hr, the number of colonies formed by founder cells was recorded. Out of 24 cells that did not divide after six hours of recovery at 30, only one of these these cells subsequently divided in the next 48 hours. This indicates that six hours of recovery in good conditions is adequate to distinguish live cells from dead cells.

A.2.5 Heat stress can alter the distribution of ages in a population

In an exponentially growing yeast population, the frequency of cells of age n in the

population would be approximately 1{2n if age had no inﬂuence on fecundity or mortality (the asymmetric division of yeast leads to a more complicated actual dis-

tribution of ages that approximates 1{2n when the doubling time of the population is short (Lord and Wheals, 1980)). Therefore, without specifically enriching for old cells, it should be quite rare to observe cells that have budded more than about ten times as they should make up less than 1 in 1,000 cells in the population. However, as described above, heat stress can induce changes in age related patterns of fecundity and mortality. One would predict therefore that some stress related changes, particularly those that deplete young cells, could be sufficient to alter population demography broadly. We found that in the absence of heat stress, the distribution of ages in the populations of all three strains approximately matched the 1{2n expectation, consistent with no influence of age. Furthermore, neither YJM693 nor S288C showed significantly different age distributions at 30 ˝C and 35.5 ˝C (Kolmogorov-Smirnov test, p ą 0.5). However, YJM996 had a significantly different distribution of ages during growth at 35.5 ˝C (Kolmogorov-Smirnov test, p “ 1.50ˆ10´9). This difference is due to a strong enrichment of cells older than five buds and a depletion of cells between 2 and 4 buds (Fig. 4.6). Consequently, we frequently found cells with more than 15 buds in YJM996 populations during growth at 35.5 ˝C. This is consistent with

72 the low fecundity and increased mortality we observed in young cells and indicates that heat stress can substantially alter the age distribution of a population of cells by aﬀecting their demographic rates.

A.3 Supplementary Methods

A.3.1 TrackScar Staining

All of our experiments were conducted using recently clonal cell lines in early- to mid-log-phase growth, grown in YPD media (Sherman 2002), except for experiments to visualize mitochondrial morphology, which where conducted in SC-ura media. To begin each experiment, frozen stocks were streaked onto YPD or SC-ura plates to obtain single colonies and incubated at 30 ˝C. After 24 to 48 hrs, single colonies were inoculated into 2ml of YPD or SC-ura overnight at 30 ˝C. To obtain log-phase cells, overnight cultures were diluted 1:100 and then in a 1:4 serial dilution in the columns of a 96 well plate. The ﬁnal volume in each well of the plate was 150 µL. The diluted cells were then placed in an incubator and shaken at 500rpm at the desired temperature for at least 16 hr. Dilutions containing putative early log-phase cultures were located by examining the optical density of cells by eye and were then counted in a hemocytometer. A dilution whose concentration was between 0.7ˆ106 and 7ˆ106 cells/ml was chosen. Log-phase cells were gently spun out of solution by centrifugation at 500xg for 1.5 minutes. All centrifuge steps on live cells were done at this speed. Cells were stained in 60 µL of fresh media with 33 µg mL´1 wheat germ agglutinin (WGA) conjugated to Alexa488 (Invitrogen, W11261) at 30 ˝C for 15 min on a rotor. Cells were washed once in 200 µL of fresh media then diluted to 0.33ˆ106 cells/ml in one column of a 96 well plate. Cells were then placed back in the incubator for 6 hr. To ﬁx the cells for subsequent imaging, cells were spun out of solution at 5,000xg for 2 minutes. All subsequent centrifugation steps were done at this speed. Cells

73 were resuspended in 500 µL of sterile PBS. Cells were ﬁxed by adding 100 µL of 40% formaldehyde to the PBS and incubating at room temperature for at least 15 minutes. Cells were washed once in sterile PBS then resuspended in sterile PBS and stored at 4 ˝C in the dark until imaging. Cells were generally imaged less than a week after collection, but the staining was stable for at least three months. Immediately prior to imaging, cells were spun out of PBS and resuspended in

60 µL of fresh media with WGA conjugated with tetramethylrhodamine (Invitrogen, W849) for 15 minutes at 30 ˝C on a rotor. Cells were washed 1x in 500 µL of sterile PBS, then sonicated in six short bursts using a Branson Sonifier 150 at setting 4 to break up any clumps. For experiments examining mitochondrial morphology, TrackScar staining was performed as above, except the first WGA stain was WGA-TMR and the second WGA stain six hours later was WGA-640CF (Biotium). For the “Three-Color TrackScar” experiments used to determine if cells with low fecundity during heat stress would divide subsequently after incubation at 30 ˝C, cells were: 1) stained as above with WGA-Ax488 at the beginning of the experiment; 2) stained as above with WGA-640CF after 6hr of incubation at either 35.5 ˝C or 40 ˝C; 3) stored overnight at 4 ˝C in YPD (for experimental convenience); 4) incubated in YPD for 6hr at 30 ˝C on a rotor drum; 5) fixed with formaldehyde as above and stored, and; 6) stained as above with WGA-TMR.

A.3.2 Microscopy

To count the stained bud scars for all experiments except for the time series in Fig. S1 (see below), cells were imaged on a DeltaVision wide-field fluorescent microscope using a 100x lens. Since cells that do not have the first stain (WGA-Ax488) do not have any information about the number of daughters produced by the cell, we randomly imaged cells that were stained with the first stain. To find cells to image,

74 we moved in transects along the slide and imaged each cell that had staining in the Ax488 channel. The exception to this was when we sought out old cells specifically. In this case, we counted the number of buds that were visible in the Ax488 channel and ignored young cells. For each field of view containing a cell, ten- to twelve-micron wide z-stacks with 0.2 to 0.4 micron slices were collected at every point with a 0.1 second exposure. The width of the z-stacks was chosen to ensure that both sides of every cell were imaged. To visualize cell outlines, z-stacks of Nomarski bright-field images were collected with a 0.025 second exposure. Either 768x768 or 1024x1024 images were collected. Z-stacks were deconvolved using the default settings of the DeltaVision software. The deconvolved z-stacks were cropped to remove deconvolution artifacts. Each channel was projected onto a single slice using a maximum pixel intensity projection. At least 150 images (containing at least one cell) for each sample were collected. To generate images for display items in the figures, the brightness and contrast of each cell and each channel was adjusted independently so that the first stain would be visible. The gamma of the brightfield channel was sometimes adjusted in order to remove saturated pixels that made display difficult.

A.3.3 Image processing and data analysis

The projected images of each z-stack were assembled into TIFF stacks using a custom Python script and ImageMagick. Individual cells with the first TrackScar stain (either Ax488 or TMR) were cropped from the wider images using custom ImageJ macros. To count the bud scars, a false-color image of the individual cell was dis- played. To aid in the counting, we wrote ImageJ macros that allowed the rapid toggling of different channels. Each bud scar in each channel was counted using the following rules: 1) the birth scar is always counted as the first scar; 2) if a bud scar has any staining of the previous stain that is visible at a normal brightness and con-

75 trast level, then that bud scar is counted as being stained by that stain; 3) bud scars that correspond to daughter cells still attached to the mother cell are not counted unless the daughter cell has divided (has a bud scar besides its birth scar). Mitochondrial morphology for Fig. 4.8 was scored by examining random cells in an Axio Imager and using a 100x lens. The presence of clumps of mitochondria and mitochondrial threads was scored. Mitochondrial morphology for Fig. 4.6 was scored by examining maximum projections of Z-stacks. The presence of clumps of mitochondria and mitochondrial threads was scored before examining the TrackScar staining for each cell. All statistical tests were performed using R (R Core Team 2015). All graphs were generated using the R package ggplot2 (Wickham 2009). To estimate the division rate of haploid S288C cells, we fit a linear model to a plot of buds produced as a function of time. To test for the effect of age on the average fecundity of a cohort, we computed the average fecundity of the cohort of each strain for three biological replicates and fit a linear model to those averages.

A.3.4 Data collection and microscopy for time series

To collect the time series in Fig. A.1, we used the strains diploid YJM1529, YJM1549, YJM1573, YJM195, YJM554, YJM555, YJM693, and S288C as well as haploid S288C cells. TrackScar staining was done as above except the time between the two stains was varied between one and six hours. To count the number of daughters produced by cells in the time series in Fig. A.1, the number of buds on at least ﬁfteen random cells stained by each stain were counted were counted using an Axio Imager and a 100x lens in “real time” (without acquiring images). To see buds on both sides of the cell, the plane of focus was moved up and down.

76 A.3.5 Growth curves

To measure the maximum population growth rate of strains in the 100 genomes collection, 93 strains were grown overnight in a 96 well plate with 100 µL of YPD per well at the desired temperature on a plate shaker at 280rpm. Cultures were mixed and diluted 1:5000 into a 96 well plate with 100 µL of YPD per well. Growth curves were measured in a Tecan Sunrise plate reader. Measurements were collected every 15min for 48hr with 30s of shaking at the highest setting before each measurement. Four biological replicates were collected for the temperatures 30 ˝C and 35.5 ˝C and at least two replicates were collected for all other temperatures. The maximum growth rate of a culture was calculated using the “cellGrowth” package from Bioconductor (Gagneur and Neudecker 2012). To identify strains sensitive to heat stress, we looked for strains whose maximum population growth rate at 35.5 ˝C was 93% or lower than their growth rate at 30 ˝C. This cutoﬀ was chosen because S288C populations grew 94% as quickly at 35.5 ˝C at 30 ˝C.

A.3.6 Strain construction

To generate the uracil auxotrophs CMY145 and CMY146 in the YJM693 and YJM996 genetic background, the URA3 gene was disrupted by a KanMX3 cassette. The plasmid M3927 (Voth et al., 2003)(aka PMB161) was digested using BamHI, then transformed into YJM693 and YJM996 using LiAc/PEG transformation (Gietz and Schiestl, 2007). G418 resistant colonies were identiﬁed using colony PCR and were sporulated. Tetrads were dissected and G418 resistant, uracil auxotrophs were iden- tiﬁed by replica plating. To visualize mitochondrial morphology, the plasmid p416- GPD-mito-roGFP1 (aka PMB273) (McFaline-Figueroa et al., 2011)(a kind gift from Liza Pon) was transformed into CMY145, CMY146, and PMY044 (a uracil auxotroph in the S288C background) to yield CMY177, CMY178, and CMY182, respectively.

77 Table A.1: Strains used in this study

Name PMY No. Genotype YJM996 PMY1523 Mata/Matα YJM693 PMY1513 Mata/Matα S288C PMY1590 ho/ho, Mata/Matα CMY177 PMY2170 ∆ura3::kanMX3/∆ura3::kanMX3, Mata/Matα CMY178 PMY2171 ∆ura3::kanMX3/∆ura3::kanMX3, Mata/Matα CMY182 PMY2172 his3∆1/his3∆1 leu2∆1/leu2∆1 lys2∆0/LYS2 MET15/met15∆0 ura3∆0/ura3∆0, Mata/Matα

78 0.20

0.15

0.10

0.05 Max population growth rate (OD/hour) rate growth Max population

0.00 0.25 0.50 0.75 1.00 Mean division rate (daughters per hour)

Figure A.2: Estimates of the growth rate of a population made using TrackScar and growth curves are well-correlated. A scatter plot of the max growth of a population inferred by measuring its rate of increase in a spectrophotometer and the average fecundity of a population measured using TrackScar is shown for a diverse panel of yeast strains grown at 35.5 ˝C. A linear regression line is shown in blue.

79 30.0˚C 35.5˚C 0 S288C −1 YJM693 −2 YJM996 −3 −4 −5 frequency

2 −6 −7 log −8 n −9 1/(2 ) 1 3 5 7 9 11 13 1 3 5 7 9 11 13 Cohort

Figure A.3: Heat stress can alter the distribution of ages in a population. The frequency of cells in a cohort in a sample of randomly chosen cells at 30 ˝C and 35.5 ˝C is shown for the strains S288C (red crosses), YJM693 (blue circles), and YJM996 (black triangles) is shown. The grey dotted line shows 1{p2nq. Each point represents the pooled observations across at least three biological replicates. Only cohorts with at least ﬁve individual cells are shown.

80 robust, YJM1341 robust, YJM1381 robust, YJM1385 robust, YJM1387 robust, YJM1399 0.75 30C 37C 0.50 0.25 0.00 robust, YJM1477 robust, YJM1479 robust, YJM1552 robust, YJM195 robust, YJM271 0.75 0.50 0.25 0.00 robust, YJM470 robust, YJM975 robust, YJM993 sensitive, YJM1248 sensitive, YJM1433 0.75 0.50 0.25

Frequency 0.00 sensitive, YJM1478 sensitive, YJM693 sensitive, YJM969 sensitive, YJM972 sensitive, YJM987 0.75 0.50 81 0.25 0.00 sensitive, YJM996 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 0.75 Daughters in 6hr 0.50 0.25 0.00 1 2 3 4 5 6 7 8 9 Daughters in 6hr

Figure A.4: Sensitive strains frequently show sub-populations of cells with very low fecundity. Plots showing histograms of the number of daughters produced in 6hr at 30 ˝C (brown) and 35.5 ˝C (gold) are shown for all strains that were examined using TrackScar. Bibliography

Acar, M., Mettetal, J. T., and van Oudenaarden, A. (2008), “Stochastic switching as a survival strategy in ﬂuctuating environments,” Nature Genetics, 40, 471–475.

Aguilaniu, H., Gustafsson, L., Rigoulet, M., and Nystr¨om,T. (2003), “Asymmet- ric inheritance of oxidatively damaged proteins during cytokinesis.” Science, 299, 1751–1753.

Ansel, J., Bottin, H., Rodriguez-Beltran, C., Damon, C., Nagarajan, M., Fehrmann, S., Fran¸cois,J., and Yvert, G. (2008), “Cell-to-cell stochastic variation in gene expression is a complex genetic trait.” PLoS Genet, 4.

Arkin, A., Ross, J., and McAdams, H. H. (1998), “Stochastic kinetic analysis of developmental pathway bifurcation in phage lambda-infected Escherichia coli cells.” Genetics, 149, 1633–1648.

Avery, S. V. (2006), “Microbial cell individuality and the underlying sources of heterogeneity,” Nature Reviews Microbiology, 4, 577–587.

Bachtrog, D., Mank, J. E., Peichel, C. L., Kirkpatrick, M., Otto, S. P., Ashman, T.-L., Hahn, M. W., Kitano, J., Mayrose, I., Ming, R., Perrin, N., Ross, L., Valen- zuela, N., Vamosi, J. C., and Consortium, T. T. o. S. (2014), “Sex Determination: Why So Many Ways of Doing It?” PLoS Biol, 12, e1001899.

Balaban, N. Q., Merrin, J., Chait, R., Kowalik, L., and Leibler, S. (2004), “Bacterial Persistence as a Phenotypic Switch,” Science, 305, 1622–1625.

Barker, M. G. and Walmsley, R. M. (1999), “Replicative ageing in the ﬁssion yeastSchizosaccharomyces pombe,” Yeast, 15, 1511–1518.

Beaumont, H. J. E., Gallie, J., Kost, C., Ferguson, G. C., and Rainey, P. B. (2009), “Experimental evolution of bet hedging,” Nature, 462, 90–93.

Boeke, J. D., La Croute, F., and Fink, G. R. (1984), “A positive selection for mutants lacking orotidine-5’-phosphate decarboxylase activity in yeast: 5-ﬂuoro-orotic acid resistance,” Molecular and General Genetics MGG, 197, 345–346.

82 Boles, B. R. (2004), “Self-generated diversity produces ”insurance eﬀects” in bioﬁlm communities,” Proc Natl Acad Sci USA, 101, 16630–16635.

Breslow, D. K., Cameron, D. M., Collins, S. R., Schuldiner, M., Stewart-Ornstein, J., Newman, H. W., Braun, S., Madhani, H. D., Krogan, N. J., and Weissman, J. S. (2008), “A comprehensive strategy enabling high-resolution functional analysis of the yeast genome,” Nat Meth, 5, 711–718.

Bull, J. J. (1987), “Evolution of Phenotypic Variance,” Evolution, 41, 303.

Bumgarner, S. L., Neuert, G., Voight, B. F., Symbor-Nagrabska, A., Grisaﬁ, P., van Oudenaarden, A., and Fink, G. R. (2012), “Single-Cell Analysis Reveals that Noncoding RNAs Contribute to Clonal Heterogeneity by Modulating Transcription Factor Recruitment,” in Molecular cell, pp. 470–482, Cell Press.

Burns, V. W. (1956), “Temporal studies of cell division. I. The inﬂuence of ploidy and temperature on cell division in s. cerevisial,” Journal of Cellular and Comparative Physiology, 47, 357–375.

C D Kelly, O. R. (1932), “The Growth Rate of Individual Bacterial Cells,” Journal of Bacteriology, 23, 147.

Carey, L. B., van Dijk, D., Sloot, P. M. A., Kaandorp, J. A., and Segal, E. (2013), “Promoter Sequence Determines the Relationship between Expression Level and Noise,” PLoS Biol, 11, e1001528.

Cenens, W., Makumi, A., Govers, S. K., Lavigne, R., and Aertsen, A. (2015), “Viral Transmission Dynamics at Single-Cell Resolution Reveal Transiently Immune Sub- populations Caused by a Carrier State Association.” PLoS Genet, 11, e1005770.

Chernyakov, I., Santiago-Tirado, F., and Bretscher, A. (2013), “Active segregation of yeast mitochondria by Myo2 is essential and mediated by Mmr1 and Ypt11.” Current biology : CB, 23, 1818–1824.

Childs, D. Z., Metcalf, C. J. E., and Rees, M. (2010), “Evolutionary bet-hedging in the real world: empirical evidence and challenges revealed by plants,” Proceedings of the Royal Society B: Biological Sciences, 277, 3055–3064.

Clauss, M. J. and Venable, D. L. (2000), “Seed Germination in Desert Annuals: An Empirical Test of Adaptive Bet Hedging,” The American Naturalist, 155, 168–186.

Cohen, D. (1966), “Optimizing reproduction in a randomly varying environment,” Journal of Theoretical Biology, 12, 119–129.

Cooper, W. S. and Kaplan, R. H. (1982), “Adaptive “coin-ﬂipping”: a decision- theoretic examination of natural selection for random individual variation,” Jour- nal of Theoretical Biology.

83 de Jong, I. G., Haccou, P., and Kuipers, O. P. (2011), “Bet hedging or not? A guide to proper classiﬁcation of microbial survival strategies,” BioEssays, 33, 215–223.

Dewitt, T. J., Sih, A., and Wilson, D. S. (1998), “Costs and limits of phenotypic plasticity.” Trends in Ecology & Evolution, 13, 77–81.

Dillin, A., Gottschling, D. E., and Nystr¨om,T. (2014), “The good and the bad of being connected: the integrons of aging.” Current Opinion in Cell Biology, 26, 107–112.

Donaldson-Matasci, M. C. (2008), “Phenotypic diversity as an adaptation to environmental uncertainty,” Evolutionary Ecology Research, 10, 493–515.

Donaldson-Matasci, M. C., Bergstrom, C. T., and Lachmann, M. (2013), “When Unreliable Cues Are Good Enough ,” The American Naturalist, 182.

Doody, J. S., Guarino, E., Georges, A., Corey, B., Murray, G., and Ewert, M. (2006), “Nest site choice compensates for climate eﬀects on sex ratios in a lizard with environmental sex determination,” Evolutionary Ecology, 20, 307–330.

Elowitz, M. B. (2002), “Stochastic Gene Expression in a Single Cell,” Science.

Ewert, M. A., Lang, J. W., and Nelson, C. E. (2005), “Geographic variation in the pattern of temperature-dependent sex determination in the American snapping turtle (Chelydra serpentina) ,” Journal of Zoology, 265, 81–95.

Fehrmann, S., Paoletti, C., Goulev, Y., Ungureanu, A., Aguilaniu, H., and Charvin, G. (2013a), “Aging yeast cells undergo a sharp entry into senescence unrelated to the loss of mitochondrial membrane potential.” Cell Reports, 5, 1589–1599.

Fehrmann, S., Bottin-Duplus, H., Leonidou, A., Mollereau, E., Barthelaix, A., Wei, W., Steinmetz, L. M., and Yvert, G. (2013b), “Natural sequence variants of yeast environmental sensors confer cell-to-cell expression variability.” Molecular Systems Biology, 9, 695.

Finkelshtein, A., Roth, D., Ben-Jacob, E., and Ingham, C. J. (2015), “Bacterial swarms recruit cargo bacteria to pave the way in toxic environments.” mBio, 6, e00074–15.

FISHELSON, L. (1970), “Protogynous Sex Reversal in the Fish Anthias squamipin- nis (Teleostei, Anthiidae) regulated by the Presence or Absence of a Male Fish,” Nature, 227, 90–91.

Foster, R. C. (1988), “Microenvironments of soil microorganisms,” Biology and fer- tility of soils, 6, 189–203.

84 Frederick, R. L., Okamoto, K., and Shaw, J. M. (2008), “Multiple pathways inﬂuence mitochondrial inheritance in budding yeast.” Genetics, 178, 825–837.

Fukami, T., Beaumont, H. J. E., Zhang, X.-X., and Rainey, P. B. (2007), “Immigra- tion history controls diversiﬁcation in experimental adaptive radiation.” Nature, 446, 436–439.

Gallie, J., Libby, E., Bertels, F., Remigi, P., Jendresen, C. B., Ferguson, G. C., De- sprat, N., Buﬃng, M. F., Sauer, U., Beaumont, H. J. E., Martinussen, J., Kilstrup, M., and Rainey, P. B. (2015), “Bistability in a Metabolic Network Underpins the De Novo Evolution of Colony Switching in Pseudomonas ﬂuorescens,” PLoS Biol, 13, e1002109.

Gietz, R. D. and Schiestl, R. H. (2007), “High-eﬃciency yeast transformation using the LiAc/SS carrier DNA/PEG method.” Nature Protocols, 2, 31–34.

Goddard, M. R. (2008), “Quantifying the Complexities of Saccharomyces cerevisiae’s Ecosystem Engineering via Fermentation,” Ecology, 89, 2077–2082.

Goldstein, A. L. and McCusker, J. H. (1999), “Three new dominant drug resistance cassettes for gene disruption in Saccharomyces cerevisiae,” Yeast, 15, 1541–1553.

Gore, J., Youk, H., and van Oudenaarden, A. (2009), “Snowdrift game dynamics and facultative cheating in yeast,” Nature, 459, 253–256.

Graham, J. K., Smith, M. L., and Simons, A. M. (2014), “Experimental evolution of bet hedging under manipulated environmental uncertainty in Neurospora crassa,” Proceedings of the Royal Society B: Biological Sciences, 281, 20140706.

Grimbergen, A. J., Siebring, J., Solopova, A., and Kuipers, O. P. (2015), “Microbial bet-hedging: the power of being diﬀerent,” Current Opinion in Microbiology, 25, 67–72.

Hairston, N. G. and Munns, W. R. (1984), “The timing of copepod diauapause as an evolutionarily stable strategy,” The American Naturalist, 123, 733–751.

Hairston Jr, N. G., Dillon, T. A., and De Stasio Jr, B. T. (1990), “A Field Test for the Cues of Disapause in a Freshwater Copepod,” Ecology, pp. 2218–2223.

Hamilton, W. D. (1966), “The moulding of senescence by natural selection,” Journal of Theoretical Biology, 12, 12–45.

Hartwell, L. and Unger, M. W. (1977), “Unequal division in Saccharomyces cerevisiae and its implications for the control of cell division.” The Journal of Cell Biology, 75, 422–435.

85 Hilﬁnger, A. and Paulsson, J. (2011), “Separating intrinsic from extrinsic ﬂuctuations in dynamic biological systems.” Proceedings of the National Academy of Sciences of the United States of America, 108, 12167–12172.

Holland, S. L., Reader, T., Dyer, P. S., and Avery, S. V. (2013), “Phenotypic heterogeneity is a selected trait in natural yeast populations subject to environmental stress.” Environmental Microbiology.

Hutchinson, G. E. (1957), “Concluding remarks,” Cold Spring Harbor Symposia on Quantitative Biology, pp. 415–427.

Iyer-Biswas, S., Wright, C. S., Henry, J. T., Lo, K., Burov, S., Lin, Y., Crooks, G. E., Crosson, S., Dinner, A. R., and Scherer, N. F. (2014), “Scaling laws governing stochastic growth and division of single bacterial cells.” Proc Natl Acad Sci USA, 111, 15912–15917.

Janzen, F. J. and Paukstis, G. L. (1991), “Environmental sex determination in reptiles: ecology, evolution, and experimental design.” The Quarterly review of biology, 66, 149–179.

Jeong, J.-Y., Yim, H.-S., Ryu, J.-Y., Lee, H. S., Lee, J.-H., Seen, D.-S., and Kang, S. G. (2012), “One-Step Sequence- and Ligation-Independent Cloning as a Rapid and Versatile Cloning Method for Functional Genomics Studies,” Applied and En- vironmental Microbiology, 78, 5440–5443.

Johnston, G., Ehrhardt, C. W., Lorincz, A., and Carter, B. L. (1979), “Regulation of cell size in the yeast Saccharomyces cerevisiae.” Journal of Bacteriology, 137, 1–5.

Jones, D. L., Brewster, R. C., and Phillips, R. (2014), “Promoter architecture dic- tates cell-to-cell variability in gene expression,” Science, 346, 1533–1536.

Joseph, S. B. and Hall, D. W. (2004), “Spontaneous Mutations in Diploid Saccha- romyces cerevisiae More Beneﬁcial Than Expected,” Genetics, 168, 1817–1825.

Kennedy, B. K., Austriaco, N. R., and Guarente, L. (1994), “Daughter cells of Sac- charomyces cerevisiae from old mothers display a reduced life span.” The Journal of Cell Biology, 127, 1985–1993.

Kispal, G., Sipos, K., Lange, H., Fekete, Z., Bedekovics, T., Jan´aky, T., Bassler, J., Netz, D. J. A., Balk, J., Rotte, C., and Lill, R. (2005), “Biogenesis of cytosolic ribosomes requires the essential iron–sulphur protein Rli1p and mitochondria,” The EMBO Journal, 24, 589–598.

Kussell, E. and Leibler, S. (2005), “Phenotypic Diversity, Population Growth, and Information in Fluctuating Environments,” Science, 309, 2075–2078.

86 LaFleur, M. D., Qi, Q., and Lewis, K. (2009), “Patients with Long-Term Oral Car- riage Harbor High-Persister Mutants of Candida albicans,” Antimicrobial agents and chemotherapy, 54, 39–44.

Lai, C.-Y., Jaruga, E., Borghouts, C., and Jazwinski, S. M. (2002), “A mutation in the ATP2 gene abrogates the age asymmetry between mother and daughter cells of the yeast Saccharomyces cerevisiae.” Genetics, 162, 73–87.

Lang, G. I., Murray, A. W., and Botstein, D. (2009), “The cost of gene expression underlies a ﬁtness trade-oﬀ in yeast,” Proc Natl Acad Sci USA, 106, 5755–5760.

Lee, S. S., Avalos Vizcarra, I., Huberts, D. H. E. W., Lee, L. P., and Heinemann, M. (2012), “Whole lifespan microscopic observation of budding yeast aging through a microﬂuidic dissection platform ,” Proc Natl Acad Sci USA.

Leibler, S. and Kussell, E. (2010), “Individual histories and selection in heterogeneous populations,” Proc Natl Acad Sci USA, 107, 13183–13188.

Levy, S. F., Ziv, N., and Siegal, M. L. (2012), “Bet Hedging in Yeast by Heteroge- neous, Age-Correlated Expression of a Stress Protectant,” PLoS Biol, 10, e1001325 EP –.

Lewis, K. (2010), “Persister cells,” Annual review of microbiology, 64, 357–372.

Lindstrom, D. L. and Gottschling, D. E. (2009), “The mother enrichment program: a genetic system for facile replicative life span analysis in Saccharomyces cerevisiae.” Genetics, 183, 413–13SI.

Liti, G., Carter, D. M., Moses, A. M., Warringer, J., Parts, L., James, S. A., Davey, R. P., Roberts, I. N., Burt, A., Koufopanou, V., Tsai, I. J., Bergman, C. M., Bensasson, D., O’Kelly, M. J. T., van Oudenaarden, A., Barton, D. B. H., Bailes, E., Nguyen, A. N., Jones, M., Quail, M. A., Goodhead, I., Sims, S., Smith, F., Blomberg, A., Durbin, R., and Louis, E. J. (2009), “Population genomics of do- mestic and wild yeasts,” Nature, 458.

Longtine, M. S., McKenzie III, A., Demarini, D. J., and Shah, N. G. (1998), “Addi- tional modules for versatile and economical PCR-based gene deletion and modiﬁ- cation in Saccharomyces cerevisiae,” Yeast.

Lord, P. G. and Wheals, A. E. (1980), “Asymmetrical division of Saccharomyces cerevisiae.” J. Bacteriol., 142, 808–818.

Lord, P. G. and Wheals, A. E. (1981), “Variability in individual cell cycles of Sac- charomyces cerevisiae ,” Journal of cell science, 50, 361.

Maamar, H., Raj, A., and Dubnau, D. (2007), “Noise in gene expression determines cell fate in Bacillus subtilis.” Science, 317, 526–529.

87 Madigan, M. T., Martinko, J. M., Bender, K. S., Buckley, D. H., and Stahl, D. A. (2015), Brock Biology of Microorganisms, vol. 4, Pearson, Boston, 14 edn. Mah´e,Y., Lemoine, Y., and Kuchler, K. (1996), “The ATP binding cassette transporters Pdr5 and Snq2 of Saccharomyces cerevisiae can mediate transport of steroids in vivo.” The Journal of biological chemistry, 271, 25167–25172. Marston, A. L. (2014), “Chromosome segregation in budding yeast: sister chromatid cohesion and related mechanisms.” Genetics, 196, 31–63. Masel, J., King, O. D., and Maughan, H. (2007), “The loss of adaptive plasticity during long periods of environmental stasis,” The American Naturalist, 169, 38– 46. McCusker, J. H., Clemons, K. V., Stevens, D. A., and Davis, R. W. (1994), “Genetic characterization of pathogenic Saccharomyces cerevisiae isolates.” Genetics, 136, 1261–1269. McFaline-Figueroa, J. R., Vevea, J., Swayne, T. C., Zhou, C., Liu, C., Leung, G., Boldogh, I. R., and Pon, L. A. (2011), “Mitochondrial quality control during inheritance is associated with lifespan and mother-daughter age asymmetry in budding yeast.” Aging cell, 10, 885–895. McIsaac, R. S., Oakes, B. L., Wang, X., Dummit, K. A., Botstein, D., and Noyes, M. B. (2013), “Synthetic gene expression perturbation systems with rapid, tunable, single-gene speciﬁcity in yeast ,” Nucleic Acids Research, 41, e57. Metzger, B. P. H., Yuan, D. C., Gruber, J. D., Duveau, F., and Wittkopp, P. J. (2015), “Selection on noise constrains variation in a eukaryotic promoter,” Nature, 521, 344–347. Mortimer, R. K. and Johnston, J. R. (1959), “Life Span of Individual Yeast Cells,” Nature, 183, 1751–1752. Moxon, E. R., Rainey, P. B., Nowak, M. A., and Lenski, R. E. (1994), “Adaptive evolution of highly mutable loci in pathogenic bacteria.” Current biology : CB, 4, 24–33. Mulcahy, L. R., Burns, J. L., Lory, S., and Lewis, K. (2010), “Emergence of Pseu- domonas aeruginosa Strains Producing High Levels of Persister Cells in Patients with Cystic Fibrosis,” Journal of Bacteriology, 192, 6191. Murren, C. J., Auld, J. R., Callahan, H., Ghalambor, C. K., Handelsman, C. A., Heskel, M. A., Kingsolver, J. G., Maclean, H. J., Masel, J., Maughan, H., Pfennig, D. W., Relyea, R. A., Seiter, S., Snell-Rood, E., Steiner, U. K., and Schlichting, C. D. (2015), “Constraints on the evolution of phenotypic plasticity: limits and costs of phenotype and plasticity,” Heredity, 115, 293–301.

88 Newman, J. R. S., Ghaemmaghami, S., Ihmels, J., Breslow, D. K., Noble, M., DeRisi, J. L., and Weissman, J. S. (2006), “Single-cell proteomic analysis of S. cerevisiae reveals the architecture of biological noise,” Nature, 441, 840–846.

Norman, T. M., Lord, N. D., Paulsson, J., and Losick, R. (2013), “Memory and modularity in cell-fate decision making.” Nature, 503, 481–486.

Norman, T. M., Lord, N. D., Paulsson, J., and Losick, R. (2014), “Stochastic Switch- ing of Cell Fate in Microbes,” Annual Review of Microbiology, 69, 381–403.

Oyama, S. (2000), The ontogeny of information: Developmental systems and evolution, Duke University Press.

Parts, L., Cubillos, F., Warringer, J., Jain, K., Salinas, F., Bumpstead, S. J., Molin, M., Zia, A., Simpson, J. T., Quail, M. A., Moses, A. M., Louis, E. J., Durbin, R., and Liti, G. (2011), “Revealing the genetic structure of a trait by sequencing a population under selection,” Genome Research, 21, gr.116731.110–1138.

Pigliucci, M. (2010), “Genotype–phenotype mapping and the end of the ‘genes as blueprint’ metaphor,” Philosophical Transactions of the Royal Society of London B: Biological Sciences, 365, 557–566.

Pikuta, E. V., Hoover, R. B., and Tang, J. (2008), “Microbial Extremophiles at the Limits of Life,” Critical reviews in microbiology, 33, 183–209.

Pringle, J. (1991), “Staining of bud scars and other cell wall chitin with calcoﬂuor.” Methods in enzymology, 194, 732–735.

Rahn, O. (1932), “A chemical explanation of the variability of the growth rate,” The Journal of general physiology, 15, 257–277.

Raj, A. and van Oudenaarden, A. (2008), “Nature, Nurture, or Chance: Stochastic Gene Expression and Its Consequences,” Cell, 135, 216–226.

Ratkowsky, D. A., Olley, J., McMeekin, T. A., and Ball, A. (1982), “Relationship between temperature and growth rate of bacterial cultures.” Journal of Bacteriology, 149, 1–5.

Richard, M. and Yvert, G. (2014), “How does evolution tune biological noise?” Fron- tiers in genetics, 5, 374.

Roncero, C. and Durán,A. (1985), “Effect of Calcofluor white and Congo red on fun- gal cell wall morphogenesis: in vivo activation of chitin polymerization.” Journal of Bacteriology, 163, 1180–1185.

89 Salvadó,Z., Arroyo-López, F. N., Guillamón,J. M., Salazar, G., Querol, A., and Barrio, E. (2011), “Temperature adaptation markedly determines evolution within the genus Saccharomyces.” Applied and Environmental Microbiology, 77, 2292– 2302.

Scheiner, S. M. (2014), “The genetics of phenotypic plasticity. XIII. Interactions with developmental instability,” Ecology and Evolution, 4, 1347–1360.

Scheiner, S. M. and Holt, R. D. (2012), “The genetics of phenotypic plasticity. X. Variation versus uncertainty,” Ecology and Evolution, 2, 751–767.

Schneider, B. L., Steiner, B., Seufert, W., and Futcher, A. B. (1996), “pMPY-ZAP: a reusable polymerase chain reaction-directed gene disruption cassette for Saccha- romyces cerevisiae.” Yeast, 12, 129–134.

Shapiro, J. A. (1998), “Thinking about bacterial populations as multicellular organisms.” Annual review of microbiology, 52, 81–104.

Sherman, F. (2002), “Getting started with yeast.” Methods in enzymology, 350, 3–41.

Simons, A. M. (2009), “Fluctuating natural selection accounts for the evolution of di- versiﬁcation bet hedging,” Proceedings of the Royal Society B: Biological Sciences, 276, 1987–1992.

Simons, A. M. (2011), “Modes of response to environmental change and the elusive empirical evidence for bet hedging,” Proceedings of the Royal Society B: Biological Sciences, 278, 1601–1609.

Simons, A. M. (2014), “Playing smart vs. playing safe: the joint expression of phenotypic plasticity and potential bet hedging across and within thermal environments,” Journal of Evolutionary Biology, 27, 1047–1056.

Simons, A. M. and Johnston, M. O. (1997), “Developmental Instability as a Bet- Hedging Strategy,” Oikos, 80, 401.

Simons, A. M. and Johnston, M. O. (2006), “Environmental and genetic sources of diversiﬁcation in the timing of seed germination: implications for the evolution of bet hedging.” Evolution, 60, 2280–2292.

Sinclair, D. A. and Guarente, L. (1997), “Extrachromosomal rDNA Circles A Cause of Aging in Yeast,” in Cell, pp. 1033–1042, Cell Press.

Sinha, H., Nicholson, B. P., Steinmetz, L. M., and McCusker, J. H. (2006), “Complex Genetic Interactions in a Quantitative Trait Locus,” PLoS Genet, 2, e13.

90 Sinha, H., David, L., Pascon, R. C., Clauder-Münster,S., Krishnakumar, S., Nguyen, M., Shi, G., Dean, J., Davis, R. W., Oefner, P. J., McCusker, J. H., and Stein- metz, L. M. (2008), “Sequential elimination of major-effect contributors identifies additional quantitative trait loci conditioning high-temperature growth in yeast.” Genetics, 180, 1661–1670. Skelly, D. A., Johansson, M., Madeoy, J., Wakefield, J., and Akey, J. M. (2011), “A powerful and flexible statistical framework for testing hypotheses of allele-specific gene expression from RNA-seq data,” Genome Research, 21, 1728–1737. Smukalla, S., Caldara, M., Pochet, N., Beauvais, A., Guadagnini, S., Yan, C., Vinces, M. D., Jansen, A., Prevost, M. C., Latgé,J.-P., Fink, G. R., Foster, K. R., and Verstrepen, K. J. (2008), “FLO1 Is a Variable Green Beard Gene that Drives Biofilm-like Cooperation in Budding Yeast,” in Cell, pp. 726–737, Cell Press. Spudich, J. L. and Koshland, D. E. (1976), “Non-genetic individuality: chance in the single cell.” Nature, 262, 467–471. Steinmetz, L. M., Sinha, H., Richards, D. R., Spiegelman, J. I., Oefner, P. J., Mc- Cusker, J. H., and Davis, R. W. (2002), “Dissecting the architecture of a quantitative trait locus in yeast.” Nature, 416, 326–330. Stewart, P. S. and Franklin, M. J. (2008), “Physiological heterogeneity in biofilms,” Nat Rev Micro, 6, 199–210. Stewart-Ornstein, J., Weissman, J. S., and El-Samad, H. (2012), “Cellular Noise Regulons Underlie Fluctuations in Saccharomyces cerevisiae,” in Molecular cell, pp. 483–493, Cell Press. Strope, P. K., Skelly, D. A., Kozmin, S. G., Mahadevan, G., Stone, E. A., Magwene, P. M., Dietrich, F. S., and McCusker, J. H. (2015), “The 100-genomes strains, an S. cerevisiae resource that illuminates its natural phenotypic and genotypic variation and emergence as an opportunistic pathogen.” Genome Research, 25, 762–774. Van Buskirk, J. and Steiner, U. K. (2009), “The fitness costs of developmental canal- ization and plasticity.” Journal of Evolutionary Biology, 22, 852–860. Vasi, F., Travisano, M., and Lenski, R. E. (1994), “Long-Term Experimental Evolu- tion in Escherichia coli. II. Changes in Life-History Traits During Adaptation to a Seasonal Environment,” The American Naturalist, 144, 432–456. Veening, J., Stewart, E. J., Berngruber, T. W., Taddei, F., Kuipers, O. P., and Hamoen, L. W. (2008), “Bet-hedging and epigenetic inheritance in bacterial cell development,” Proc Natl Acad Sci USA, 105, 4393–4398. Venable, D. L. (2012), “Bet Hedging in a Guild of Desert Annuals,” Ecology, 88, 1086–1090.

91 Voth, W. P., Wei Jiang, Y., and Stillman, D. J. (2003), “New ‘marker swap’ plasmids for converting selectable markers on budding yeast gene disruptions and plasmids,” Yeast, 20, 985–993.

Wang, P., Robert, L., Pelletier, J., Dang, W. L., Taddei, F., Wright, A., and Jun, S. (2010), “Robust Growth of Escherichia coli,” Current Biology, 20, 1099–1103.

Warringer, J., Ericson, E., Fernandez, L., Nerman, O., and Blomberg, A. (2003), “High-resolution yeast phenomics resolves diﬀerent physiological features in the saline response,” Proc Natl Acad Sci USA, 100, 15724–15729.

Warringer, J., Zörgö,E., Cubillos, F. A., Zia, A., Gjuvsland, A., Simpson, J. T., Forsmark, A., Durbin, R., Omholt, S. W., Louis, E. J., Liti, G., Moses, A., and Blomberg, A. (2011), “Trait variation in yeast is defined by population history.” PLoS Genet, 7, e1002111.

Weinberger, L. S. (2015), “A minimal fate-selection switch,” Current Opinion in Cell Biology, 37, 111–118.

West-Eberhard, M. J. (2003), Developmental plasticity and evolution, Oxford Uni- versity Press.

Zhang, X. and Borevitz, J. O. (2009), “Global Analysis of Allele-Speciﬁc Expression in Arabidopsis thaliana,” Genetics, 182, 943–954.

Ziv, N., Siegal, M. L., and Gresham, D. (2013), “Genetic and non-genetic determi- nants of cell-growth variation assessed by high-throughput microscopy.” Molecular Biology and Evolution.

92 Biography

Colin Scott Maxwell was born in Bonners Ferry, Idaho in the United States of Amer- ica on April 19th 1985. He earned his Bachelor of Science from the University of Oregon in 2006.