Robustness and Tunability in Biological Networks
by
MASSACH USETTS INSTItUTE
OFT ECHNOLOGY
Shankar Mukherji
0
S.B., Physics
Massachusetts Institute of Technology, 2004
2 2010
LIE RARIES
S.B., Mathematics
Massachusetts Institute of Technology, 2004
Submitted to the Harvard-MIT Division of Health Sciences and Technology
in partial fulfillment of the requirements for the degree of
Doctor of Philosophy
ARCHIVES
at the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
May 2010
0 Massachusetts Institute of Technology 2010.
I
//I
Signature of author:
Division of Health Sciences and Technology
May 12, 2010
Certified by:
L;/ v
Alexander van Oudenaarden, Ph.D.
Professor of Physics and Biology
Thesis Supervisor
Accepted by:
Ram Sasisekharan, Ph.D.
Director, Harvard-MIT Division of Health Sciences and Technology
Edward Hood Taplin Professor of Health Sciences and Technology
and Biological Engineering
2
Robustness and Tunability in Biological Networks
by
Shankar Mukherji
Submitted to the Harvard-MIT Division of Health Science and Technology
on May 12, 2010 in partial fulfillment of
the requirements for the degree of
Doctor of Philosophy in Biomedical Engineering
Abstract
Cells face a core tension between studiously preventing change in certain properties from
extrinsic perturbations while allowing other properties to be tuned. One way cells have
resolved this tension is to utilize systems that are both robust and tunable. Systems can
achieve this through network design, which can contain submodules that are themselves
either robust or tunable, or through network components that are robust over only a
defined set of parameter ranges. This work examines these two categories with two
specific examples described below.
To explore how a simple network can be both robust and tunable, we make use of
the osmosensing pathway in the budding yeast Saccharomyces cerevisiae. The pathway
consists of two modules: a phosphorelay module that senses the osmotic shock signal that
feeds into a mitogen-activated protein kinase (MAPK) module. Using a combination of
systematic complementation experiments and computational sensitivity analysis, we
show that the phosphorelay module is robust to changes in the kinetic parameters
characterizing signal propagation through the module while signaling through the MAPK
module can be tuned by changing the rate constants. Furthermore, we show that pathway
robustness to rate constant changes has consequences for the evolvability of the
osmosensing cascade. Populations of yeast cells challenged to alter the input/output
relationship of the cascade saw their MAPK proteins preferentially targeted by natural
selection over their phosphorelay counterparts.
To explore how a simple regulatory element can be both robust and tunable, we
turn our attention to gene regulation by microRNA (miRNA). MiRNAs are short
regulatory RNA molecules that repress gene expression in a sequence-dependent manner.
By observing the strength of miRNA-mediated repression in individual cells, we show
that the strength of repression depends strongly on the relative abundance of the miRNA
and its target. Below a threshold level of target message miRNA robustly silences the
conversion of mRNA input into protein output, but above this threshold miRNAmediated repression generates an ultrasensitive response to mRNA input allowing the
strength of repression to be tuned over a wide variety of values.
Thesis Supervisor: Alexander van Oudenaarden
Title: Professor of Physics and Biology
4
Table of Contents
Preface ...............................................................................................
15
1 Introduction: Using synthetic circuits to uncover biological design ......... ........ 17
1.1 Summ ary .............................................................................
17
1.2 Understanding network design with synethetic parts ..........................
18
1.3 Toward a quantitative understanding of gene expression
.............. 20
1.3.1 Transcriptional regulation ..................................................
20
1.3.2 Promoter library studies ...................................................
21
1.3.3 Post-trascriptional and post-translational regulation .................. 24
1.3.4 Integrating transcriptional and post-transcriptional
control ......................................................................
25
1.4 Rewiring genetic and signaling pathways .........................................
27
1.4.1 The challenges of rewiring pathways ................................
27
1.4.2 Manipulating the sensors ..............................................
31
1.4.3 Manipulating sensor/transducer interactions
................ 32
1.4.4 Manipulating the intermediate transducers ..........................
34
1.4.5 Connecting pathway rewiring to evolvability ......................
34
1.5 Synthetic feedback networks .....................................................
1.5.1 O scillatory behaviors ......................................................
35
36
1.5.2 Using synthetic circuits as modeling benchmarks .................. 40
1.6 The ultimate goal: spatiotemporal control ........................................
1.6.1 Uncovering intra- and intercellular processes
41
............... 41
1.6.2 Modeling ecological interactions ........................................
1.7 Perspectives .........................................................................
43
45
2 Robust yet tunable regulatory networks: the case of the yeast
osmosensing patwahy ........................................................................
51
2.1 Sum m ary .............................................................................
51
2.2 Introduction ...........................................................................
52
2.3 HOG signaling displays varied sensitivity to ortholog substitutions ........... 54
2.3.1 Systematic complementation study ......................................
55
2.3.2 Focusing on network architecture: Pbs2 versus Ypdl ................ 56
2.4 Computational analysis of HOG pathway .........................................
59
2.4.1 Sensitivity analysis of a model of the HOG pathway ................ 60
2.4.2 Relating mutational robustness to local biochemistry ................. 63
2.5 Experimental evolution design ....................................................
66
2.6 Rapid adaptive evolution of yeast cells underexpressing YPD1 ................ 69
2.6.1 Restoration of basal wild-type HogI activity .......................
70
2.6.2 Trascriptional regulation of YPD1 is not upregulated
in the evolved strains ...................................................
73
2.6.3 PBS2 and SSK2 are preferentially mutated in independent
evolution experiments ...................................................
74
2.6.4 Mutations in PBS2 and SSK2 are mainly responsible for the
down-regulation of the hyperactive signaling and improved
fitness ...................................................................
. . 78
2.7 D iscussion ............................................................................
79
2 .8 M ethods .................................................................................
81
3 Robust yet tunable regulatory elements: the case of microRNA
............. 87
3.1 Sum m ary .............................................................................
87
3.2 microRNA background .............................................................
88
3.3 Two-color assay to measure regulation via microRNA ........................
88
3.3.1 Control experiments establishing eYFP as a transcriptional
reporter ...................................................................
89
3.4 microRNA mediated repression generates gene expression thresholds ...... 90
3.5 Generating thresholds without feedback ...........................................
92
3.5.1 Mathematical framework .................................................
93
3.5.2 Tuning the dissociation constant X ......................................
96
3.5.3 Tuning the threshold constant 0 ......................................
97
3.6 Experimentally tuning the ultrasensitive response ...............................
3.6.1 Increasing N in the mCherry 3'-UTR ................................
97
98
3.6.2 Calculating ratio transfer functions to measure fold repression ... 100
3.6.3 Changing [miR-20]total by transfecting mimic siRNA ............... 102
3.6.4 eYFP mRNA abundance at the threshold .............................
105
3.7 Observing ultrasensitivity in physiological contexts ...........................
106
3.7.1 Fusing natural 3'-UTRs to mCherry ...................................
107
3.7.2 Luciferase assays in mouse embryonic stem cells ................... 108
..........................................................................
110
3.9 M ethods ...............................................................................
111
3.8 Discussion
4 Conclusions and Perspectives ...............................................................
117
5 R eferen ces .......................................................................................
123
8
Table of Figures
Figure 1.3.1 Controlling the flow of information from DNA to protein using
synthetic elem ents ............................................................
23
Figure 1.3.2 An integrated transcriptional/post-transcriptional circuit to control
gene expression in mammalian systems ..................................
26
Figure 1.4.1 Rewiring signaling pathways ................................................
29
Figure 1.5.1 Building a robust, tunable oscillator in a living cell .....................
38
Figure 1.6.1 Building a synthetic predator-prey system from quorum sensing
components .......................................................................
44
Figure 1.6.2 Simpson's paradox observedin an engineered set of cell-cell
interactions ........................................................................
46
Figure 2.2.1 Schematic depiction of the Saccharomyces cerevisiaeHOG pathway ...53
Figure 2.3.1 Flow chamber system to control osmotic conditions in medium ........... 55
Figure 2.3.2 Hog1 nuclear enrichment dynamics in response to a 0.4M NaCl
osmotic shock .................................................................
56
Figure 2.3.3 Maximum Hog 1 nuclear enrichment of mutant strains with
orthologous YPD1 and PBS2 genes of varying degrees of "functional
scores" under a O.4M NaCl hyperosmotic shoch ............................
57
Figure 2.3.4 Maximum HogI nuclear enrichment of mutant strains with kinetically
characterized YPD1 alleles under a 0.4M NaCl hyperosmotic shock ..... 58
Figure 2.4.1 Surface representations of peak Hog1 phosphorylation as a function
of the rates associated with a given protein ................................
60
Figure 2.4.2 Local logarithmic gradients calculated for peak Hog1 phosphorylation
surfaces .............................................................................
61
Figure 2.4.3 Local logarithmic gradients calculated for initial Hog1 phosphorylation
61
rate surfaces .......................................................................
Figure 2.4.4 Calculating sensitivity metrics from surface plots using modified
standard deviation metric .....................................................
63
Figure 2.5.1 Artificially hyperactivating HOG pathway by underexpressing YPD1 .... 66
Figure 2.5.2 Growth rate of ancestor strain as a function of doxycycline ............... 67
Figure 2.5.3 HogI nuclear enrichment in ancestral strain with and without
doxycycline ......................................................................
68
Figure 2.5.4 Glycerol production in ancestor strain with and without doxycycline ..... 68
Figure 2.5.5 HogI nuclear enrichment in the ancestor strain as a function of
doxycycline ......................................................................
Figure 2.6.1 Time course of evolutionary dynamics .....................................
69
70
Figure 2.6.2 Restoration of growth rate levels to ancestral state .........................
71
Figure 2.6.3 Restoration of glycerol levels to ancestral state .............................
71
Figure 2.6.4 Restoration of pathway dynamics to near ancestral behavior ............... 72
Figure 2.6.5 Restoration of volume recovery dynamics for majority of the evolved
strains .........................................................................
. . .. 73
Figure 2.6.6 CFP data suggests that the evolved strains did not alter the properties of
rtTA in order to effect their recovery from pathway hyperactivation ...... 74
Figure 2.6.7 Mutations in evolved strains are predominantly in HOG pathway genes ..76
Figure 2.6.8 Distribution of genetic changes in evolved strains across 9 experiments ...77
Figure 2.6.9 Characterizing the spectrum of mutations according to target identity ...... 77
Figure 3.3.1 A synthetic two-color reporter construct for measuring miRNA
mediated gene regulation in single cells .........................................
89
Figure 3.3.2 Control experiments used to confirm idea that eYFP can act as a
faithful reporter of mCherry transcriptional activity in individual cells ..... 90
Figure 3.4.1 Arranging single cells according to eYFP expression level reveals gene
expression thresholding by miRNA ............................................
91
Figure 3.4.2 Transfer function relating eYFP to mCherry levels ............................
92
Figure 3.5.1 Biochemistry of the miRNA-mediated gene regulatory system .............. 95
Figure 3.5.2 miR-20 expression in Tet-On HeLa cells .......................................
95
Figure 3.5.3 Tuning the sharpness of the ultrasensitive switch by changing the rate
at which miRNA bind their target mRNA, k..................................96
Figure 3.5.4 Tuning both the placement and sharpness of the ultrasensitive switch
by titrating different total amounts of miRNA into the system ............... 97
Figure 3.6.1 Experimentally sharpening the ultrasensitive transition by engineering
different numbers of miR-20 binding sites into the 3'-UTR of mCherry ...99
Figure 3.6.2 Dye-swap control experiment ....................................................
100
Figure 3.6.3 Calculating the fold repression due to miRNA as a function of target
expression level ....................................................................
10 1
Figure 3.6.4 Bulk level measurements of miR-20 mediated repression ................... 102
Figure 3.6.5 Tuning the placement and the sharpness of the threshold by titrating
the amount of miR-20 available to the gene regulatory system ............ 103
Figure 3.6.6 miR-20 sponge experiments shift ultrasensitive regime to lower eYFP
levels as expected from the mathematical model .............................
104
Figure 3.6.7 Results from simultaneous fitting of model to experimental data .......... 105
Figure 3.6.8 Estimating the mRNA abundance at the threshold ...........................
106
Figure 3.7.1 Detecting ultrasensitive transitions with natural UTR's ..................... 107
Figure 3.7.2 Dual luciferase assay system used to measure miRNA mediated
repression in populations of mouse embryonic stem cells .................. 108
Figure 3.7.3 Fold repression increases as a function of miRNA abundance in mouse
109
embryonic stem cells ..............................................................
Figure 3.9.1 Binning procedure used to convert joint mCherry-eYFP single cell
distributions into transfer functions .............................................
112
12
Table of Tables
Table 2.6.1 Summary of sequencing depth and coverage for both the ancestral
strain and 5 evolved strains sent for Illumina sequencing ...................... 75
Table 2.6.2 Single nucleotide polymorphisms detected from whole genome
sequencing of evolved strains ...................................................
75
Table 2.6.3 Cataloguing the mutations in molecular detail ................................
78
14
Preface
Having spent my entire adult life at MIT, it is impossible to thank everyone who has
helped shape my education as a scientist. From midnight discussions in the East Campus
lounges to spirited group meetings in Building 68, MIT has been an extraordinary home
for me.
I would first like to thank my close collaborators in the work presented here: Margaret
Ebert for her tireless efforts in working out together how microRNA mediated regulation
works at the level of a single cell, Mei Lyn Ong for conducting the heroic experimental
evolution aspect of the yeast osmosensing pathway study, and Qiong Yang for helping
me with the complementation studies and teaching me about life in the lab in general. I
would be remiss if I did not also extend my deepest thanks to the members of the van
Oudenaarden lab. I cannot think of a more fun, intellectually stimulating place to spend
one's graduate years.
I have been very lucky to have a number of close mentors during my time at MIT. Leonid
Mirny oversaw my bachelor's thesis project and encouraged me to stay at MIT for
graduate school, during which he has been a constant source of good advice. The
members of my thesis committee, Roy Kishony and Phil Sharp, have been incredibly
helpful sounding boards for my sometimes crazy ideas. And of course there is no one I
can thank more than my advisor, Alexander van Oudenaarden. Alexander's creativity,
grace and good humor are all things I have had the good fortune to admire and enjoy over
the past years in the lab and hope to admire and enjoy even after I have flown the coop.
Lastly I would like to thank my parents, Sumantra and Lopamudra, my sisters, Aditi and
Amrita, and my wife, Emily, and all my family spread all over the world for putting up
with me and the demands of research life (my usual excuse for absent mindedness), but
also for forcing me to think deeply about how to talk about my ideas, which in turn
forced me to refine them even more. I hope you all enjoy reading this.
16
Chapter 1
Introduction: Using synthetic circuits
to uncover biological design
1.1 Summary'
The life of a cell is one with which many can sympathize: it must make decisions while
constantly buffeted by forces both from the external environment and from within. A
constant tension for the cell in the context of this struggle is whether to ignore these
forces and remain robust against the perturbations or whether to tune themselves in
response to these changes. An alternative to both of these possibilities, which is explored
in this thesis with two very specific examples, is that the cell can use either network
components or network designs that exhibit both robustness and tunability. For example,
a gene expression regulatory component can reject changes in mRNA concentrations
from affecting protein levels up until a critical value after which it allows the protein
1 See Mukherji, S and van Oudenaarden, A. Synthetic biology: understanding biological
design from synthetic circuits. Nature Reviews Genetics 10: 859-871 (2009)
level to be tuned by the mRNA level or perhaps a network can be broken into subparts
some of which can be used to tune the input/output relationship of the network while
others cannot. In order to test these ideas quantitatively, it is often useful to use highly
engineered regulatory pathways and interactions in order to cleanly isolate the
phenomena under study from other endogenous effects. Therefore, I will begin by
presenting a review of the literature highlighting the use of synthetic components and
networks used to understand natural biological effects.
1.2 Understanding network design with synthetic parts
An important aim of synthetic biology is to uncover the design principles of natural
biological systems through the rational design of gene and protein circuits. Here we
highlight how the process of engineering biological systems to the control of cell-cell interactions -
from synthetic promoters
has contributed to our understanding of how
endogenous systems are put together and function. Synthetic biological devices allow us
to intuitively grasp the ranges of behavior generated by simple biological circuits, such as
linear cascades and interlocking feedback loops, as well as to exert control over natural
processes such as gene expression and population dynamics.
One of the most astounding findings of the human genome project was that our genomes
contained as many genes as that of Drosophila melanogaster. This finding begged the
question: how do you get one organism to look like a fly and another like a human with
the same number of genes? One possibility is that the rich repertoire of non-protein
coding sequence found in the genomes of complex organisms adds many new parts with
which to generate complexity [Mattick 2004]. A decade of research has put forward the
rather different idea that instead of looking at the length of the parts list as the
determinant of organismal complexity, we should look at how those parts fit together
[Davidson 2006, Prud'homme 2007].
From this perspective, complexity arises from
novel combinations of pre-existing proteins and the ability to evolve new phenotypes
rests on the modularity of biological parts.
While natural examples have been found to illustrate this latter possibility [Prud'homme
2007], strong evidence to support this post-genomic view of biology has come from the
synthesis of new biological systems. Rational synthesis of biological systems can hint at
the natural history of how a particular system came to acquire its properties [Bridgham
2006, Rapp 2007]. More often, however, we use synthetic circuits to explore, in a handson fashion, the set of design principles that determine the structure and operation of
biological systems.
The core aim of synthetic biology is to develop and apply engineering tools to control
cellular behavior, using precisely characterized parts, such as cis-regulatory elements, to
achieve desired functions. An important direction, for example, has been to engineer cells
with an eye towards practical applications, such as bioremediation [Gilbert 2003],
biosensors [Rajendran 2008], biofuels [Steen 2008, Waks 2009], and even the beginnings
of clinical applications [Khosla 2003, Ro 2006, Anderson 2006]. In this Review,
however, we focus on how synthetic circuits help us to understand how natural biological
systems are genetically assembled and how they operate in organisms from microbes to
mammalian cells. In this light, synthetic circuits have been critical as simplified test-beds
in which to refine our ideas of how similarly structured natural networks function and
have served as tools to control natural networks. We highlight the contribution of
synthetic biology to putting together an increasingly quantitative description of gene
expression and signal transduction, in uncovering the diversity of behaviors that can arise
from positive and negative feedback systems, and progress in rationally controlling
spatial organization and cell-cell interactions. We pay particular attention to recent
progress in using synthetic systems to uncover novel aspects of cell biology, such as how
cells decide to undergo apoptosis and the molecular basis for communication between the
endoplasmic reticulum and mitochondrion. We aim to show that synthetic biological
approaches have given us a great deal of intuition on how the simple building blocks that
underlie complex natural systems work as well as basic tools to quantitatively
characterize natural phenomena, both of which are crucial for the field to progress into
the analysis and complete control of natural circuits.
1.3 Towards a quantitative understanding of gene expression
The first step in assembling a biological circuit is to gather the component parts. In the
cell, circuits are accomplished via gene expression, and so a great deal of effort in
synthetic biology has gone into investigating the rules surrounding gene expression,
particularly the processes of transcription and translation. The precise measurements
afforded by artificially constructed systems allows us to transform qualitative notions of
transcriptional repression and activation and post-transcriptional
regulation into
quantifiable effects such as how promoter architecture defines the rate of transcription
and the specific degradation rate specified by a given sequence motif.
1.3.1 Transcriptional regulation.
Among the earliest contributions of synthetic biology to understanding natural biological
processes include detailed, quantitative measurements of transcriptional regulation,
building on a foundation laid 50 years ago in the groundbreaking work of researchers
such as Jacob and Monod [Jacob 1961]. Synthetic constructs have been used to map out
the transfer function that relates the input concentration of transcription factor [Rosenfeld
2005, Pedraza 2005] and inducers [Setty 2003] to the output concentration of a reporter
gene [Rosenfeld 2005, Ozbudak 2002, Elowitz 2002], single mRNA molecules [Golding
2005, Raj 2006] or single proteins [Cai 2006]. Many of these same constructs were also
used to measure the mean output of the transcriptional process and also higher-order
moments (such as the variance) in organisms ranging from Escherichia coli and Bacillius
subtilis to mammalian cells. Single-molecule studies in these model organisms directly
established that mRNA and proteins are produced in bursts of activity [Raj 2008].
A key question in the study of transcriptional regulation is how promoter architecture
affects transcriptional activity. For example, below we describe several studies that have
informed how the number and genomic positions of transcription factor (TF) binding
sites affect transcriptional activity. Given the combinatorial control of gene expression, it
is also critical to study how multiple TFs interact with DNA and with each other to tune
mRNA production. Endogenous promoters use all of these parameters to specify either a
desired transcription rate or a boolean function, such as an AND gate that allows
transcription to occur only when all TF binding sites in the promoter are occupied.
1.3.2. Promoter library studies
The experimental breakthrough
that allowed quantitative measurements
of the
transcriptional power of different promoter architectures was the use of combinatorial
promoter libraries [Hammer 2006], which are shown in schematic form in Figure 1.3.1.
Libraries of promoters driving reporter proteins, such as luciferase or fluorescent
proteins, allow for an unbiased measurement of transcriptional activity over the space of
possible promoters - such an unbiased method can then be used to try and ascertain rules
that describe the responsiveness of a promoter to TFs. Earlier work used randomly
mutated promoters to draw inferences about the functional subparts of the promoter, such
as the TATA box; by contrast, the construction of combinatorial promoter libraries
involves identifying specific operator sites that bind TFs and randomly ligating them
together in a way that shuffles their relative positions and copy numbers. The studies
highlighted below have combined such promoter libraries and modeling to show that the
strength of a promoter is determined largely by the position of TF binding sites with
respect to key promoter elements such as the TATA box and with respect to each other..
The simplest case is to understand how the positioning of a single operator affects the
expression of a promoter. In prokaryotes, operators are classified as being in the core,
proximal, or distal regions of the promoter (Figure 1.3.1). Working in E. coli Cox et al.
[Cox 2007] and Kinkhabwala and Guet [Kinkhabwala 2008] independently observed that
repressors can effectively repress expression from all 3 promoter subregions, with Cox et
al. showing that the strength of repression is greatest when the repressor site is in the core
region of the promoter, less strong when in the proximal region, and weakest when in the
distal region. Activators, on the other hand, work only in the distal site; Cox et al. showed
that instead they have no effect in the core and proximal sites. Both studies go on to
develop simple models of promoter activity by taking into account binding reactions of
TF to DNA that are in thermodynamic equilibrium.
It was expected that the situation would be far more subtle in eukaryotes, where
chromatin structure can strongly influence expression levels [Segal 2009]. However, even
in Saccharomyces cerevisiae 49% of the variation in expression in the promoter library
could be explained by a simple thermodynamic model incorporating just TF-DNA and
TF-TF interactions [Gertz 2009], interactions that were also suggested in theoretical work
[Buchler 2003]. More surprisingly, Gertz et al. provided evidence that weak binding sites,
which are important for prokaryotic transcription, can also be important in eukaryotes.
Focusing on the TF multicopy inhibitor of GAL-] (Migi), Gertz et al. showed that
repression from one weak and one strong Mig1 binding site can be as effective as two
strong Migi binding sites. This is particularly crucial given that 24% of all yeast
promoters contain putative weak Migl binding sites.
The promoter library studies open the way to consider some general questions in
transcriptional control. The theoretical frameworks in the E. coli and yeast studies, for
example, differ slightly: the former studies make no use of TF-TF interactions and frame
the issue mostly in the language of Boolean logic, whereas the latter makes heavy use of
TF-TF interactions, particularly in the analysis of weak binding sites. Future singlemolecule studies of transcriptional control can help to resolve the relative importance of
TF-DNA and TF-TF interactions in generating transcriptional activity. Furthermore, the
fact that simple equilibrium binding explains much, but not all, of the effect of promoter
architecture on expression level suggests that the next goal should be to track down the
source of the remaining variation. Genomic location can be an important contributor to
expression and expression fluctuations [Becskei 2005], perhaps by affecting local
chromatin context. Knowing how to parcel out the variation due to these different effects
will be particularly helpful when these studies are extended to mammalian systems,
where there is considerably less control over where synthetic transgene constructs are
integrated into the genome.
Basple of matwal
Gene mpreslon
subpracess
SyntiwiC bhp
v
Transciption
inducible promoter
* Stochastic gene apression
"Gene regulation function
Pronoter
. Genomic pcsitionng of
TF sites
"Weak TF-DNA Interactions
"TF-TF interactions
p--na enaanalyzed
ary
Core
Distal
PrcKrna
00
=1 AjMff
u
cri
I1M1uffM~-7
(l
AUl
fAraC * LadM
*LuR
Post-transcriptior
translation
RBS accass
ty
aBS
RBRacssaby
Aptarner
Translation
I
inducible protease
*TtR
Stocastic gene
exprsion
Cal cycle progression
Enzyme kinetics
Figure 1.3.1 Controlling the flow of information from DNA to protein using synthetic
elements. The diagram shows the transcriptional and post-transcriptional processes in
gene expression that can be manipulated by synthetic biology tools, and some example
applications. TF, transcription factor; RBS, ribosome binding site.
Promoter library diagram from Cox RS 3rd, Surette MG, Elowitz MB. Programming
gene expression with combinatorial promoters. Mol. Syst. Bio. 3: 145 (2007). RBS
accessibility diagram from Isaacs FJ, Dwyer DJ, Ding C, Pervouchine DD, Cantor CR,
Collins JJ. Engineered riboregulators enable post-transcriptional control of gene
expression. Nat Biotech 22, 841-847 (2004) Aptamer diagram from Grate and Wilson
Grate D, Wilson C. Inducible regulation of the S. cerevisiae cell cycle mediated by an
RNA aptamer-ligand complex. Bioorg Med Chem 9: 2565-2570 (2001)
1.3.3 Post-transcriptional and post-translational regulation
Although much of the early work in synthetic biology focused on transcriptional
regulation significant progress has also been made in incorporating post-transcriptional
effects into synthetic circuits, affecting both RNA and protein. At the RNA level, for
example, mutagenesis screens based on synthetic constructs have been used to determine
the sequences that are recognized by RNA editing enzymes to change adenine into
inosine [Pokharel 2006]. Furthermore, as regulatory RNAs have been increasingly
appreciated as important drivers of gene expression, synthetic circuits have included
elements from the RNA interference pathway [Beisel 2008], aptamers [Win 2008,
Werstruck 1998, Grate 2001], and riboswitches [Suess 2004, Desai 2004] to control the
flow of genetic information [Davidson 2007].
Synthetic circuits involving enzymatic RNAs have mostly been developed as platforms to
tune gene expression, but many of these platforms can easily be extended to understand
natural biological phenomena. In the study of Grate and Wilson, for example, an aptamer
is used to control the expression of cyclinB-2 (Clb2), a key regulator of the cell cycle, in a
tetramethylrosamine (TMR)-dependent manner [Grate 2001]. The authors slowed the
speed of the cell cycle by adding TMR; this method can be useful in measuring how the
level of Clb2 protein affects the speed at which the cell cycle progresses while keeping
all transcriptional feedback constant.
Synthetic studies have also directly tweaked how mRNA is translated into protein and
how long proteins persist before being degraded. Several experiments in prokaryotic
systems, especially those studying the stochastic nature of gene expression, have altered
the translation rate by mutating ribosomal binding sites (RBS) [Ozbudak 2002, Issacs
2004]. Apart from demonstrating another possible layer of quantitative regulation of gene
expression, studies involving RBS variants provided early evidence that E.coli cells could
tune the stochasticity in the expression level of a given gene independently of its mean.
Lastly, Grilly et al. have developed a circuit that controls the degradation of a target
protein using the well-known ClpXP protease machinery from E.coli [Grilly 2007].
Typically, models of gene expression treat protein degradation as an exponential decay
process, with the decay being due to growth of cell volume over time. Regulated
proteolysis, however, can depend on the formation of enzyme-substrate complexes as
intermediates on the way to degradation. In finding that the degradation follows
Michaelis-Menten kinetics, Grilly et al. completed one of the few quantitative
comparisons of specific protease activity to models of enzyme kinetics.
Taken together, these results point to some interesting similarities between transcription
and translation - both are inherently noisy processes that can be quantitatively modulated
by specific sequence elements, such as RBS and protease recognition sites. Future studies
can use the ideas and methods from the study of transcription, such as combinatorial
library approaches, to more systematically explore the process of translation.
1.3.4 Integrating transcriptional and post-transcriptional control.
The two approaches of engineering specific promoter architectures or using a natural
inducible promoter to tune transcriptional activity, and using specific sequence sites to
tune translational yield can combined to achieve precise yet flexible control over gene
expression [Ozbudak 2002, Beisel 2008]. A nice example of using these two ingredients
to study natural processes in mammalian cells can be found in recent work, outlined in
Figure 1.3.2, in which a Tet and Lac controlled regulation was adapted and combined
with RNAi for use in HeLa cells [Deans 2007].
.............
....
....
......
..
...............................................
..........................
a
PTG
.Transcptioral
-.--. Post-transcriptional
b 30 -0-
-
Induced apoptotic cells
Induced dead cells
--
Nontransfected cels
-s-
Nontransfected cels
2.5-
2.5
0
1.000
IPTG (pM)
Figure 1.3.2 An integrated transcription/translation circuit to control gene
expression in mammalian systems. A) Deans et al. created a genetic switch whose state is
read out by a GFP reporter or a gene of interest; here, the gene of interest that we focus
on is Bax, a pro-apoptotic gene. Bax is under the transcriptional control of the Lac
repressor (Lad) and under the translational control of a short hairpin (sh)RNA, which
itself is under transcriptional control of the TetR repressor. In the "OFF" state, Lac
inhibits transcription of Bax. Additionally, Lac inhibits transcription of the TetR
repressor; this allows the transcription of the shRNA, which goes on to inhibit translation
of Bax by cleaving its mRNA. The result of this dual-layered repression is the creation of
a truly off "OFF" state; whereas in the initial characterization each mode of repression
alone was able to reduce reporter levels by about 80%, leaving a basal expression of
20%, the combination resulted in greater than 99% repression. The circuit can then be
tunably activated by adding varying amounts of IPTG, which blocks the effects of LacI.
B) The fraction of cells that undergo apoptosis is determined by Bax expression levels.
Data obtained by tuning Bax with IPTG, as described above, offer some tantalizing clues
into the fundamental molecular biology underlying the apoptosis pathway. In particular
the data are consistent with the idea that the decision to undergo apoptosis (assessed by
retention of propidium iodide (PI) dye relative to PI retention due to the transfection
protocol alone) is determined by reaching a threshold level of Bax. Although the Bax
threshold data are not conclusive, the result demonstrates the power of a technique that
allows one to rationally tune the level of any gene of interest and examine the
consequences.
Panel B is reproduced with permission from Figure 6b of Deans TL, Cantor CR, Collins
JJ. A tunable genetic switch based on RNAi and repressor proteins for regulating gene
expression in mammalian cells. Cell 130, 363-372 (2007)
As synthetic biology begins to recapitulate more realistic systems, which contain many
moving parts, demand will increase for circuits that control every step of the process that
turns DNA sequence into protein. Such layered circuits can help illuminate why certain
regulatory schemes are employed to control gene expression over others in a given
context. For example, gene expression in natural systems can be attenuated by epigenetic
silencing, transcriptional repressors or post-transcriptional regulators such as microRNAs
(either alone or in concert with other molecules); this begs the questions of why a system
uses one system rather than the other and to what extent different layers of regulation
generate collective effects that no one layer can accomplish. One area that will be
increasingly under study, and that may help unravel the issues surrounding layered
circuits, is the dynamics of the different steps that contribute to expression; the studies
highlighted above almost exclusively focus their attention on steady state behavior. While
intuition tells us that transcription factors act slowly compared to post-transcriptional
players such as regulatory RNAs, as the latter presumably do not have to be transported
back to the nucleus and then locate a specific genomic locus, there is currently a lack of
data that would enable us to turn these intuitive notions into quantitative facts.
1.4 Rewiring genetic and signaling pathways.
The act of engineering cellular pathways has allowed insight into two key properties:
precise measurement and control of the input-output relationship of a pathway, and the
functional architecture of the pathway constituents themselves. In the case of signaling
and metabolic pathways, the latter has meant insights into the functional significance of
specific protein sequences and structures, such as being able to pinpoint exactly which
protein domains and which amino acid residues are responsible for mediating specific
interactions along the pathway.
1.4.1 The challenges of rewiring pathways.
Initially, pathway engineering was primarily explored in the context of metabolism
[Martin 2009]. Metabolic engineering typically involved the use of genetic screens and
directed evolution to maximize targeted metabolic fluxes. Synthetic efforts in boosting
metabolic fluxes have begun to pay off, as is exemplified in a recent study in which a
synthetic protein scaffold was used to draw metabolic enzymes spatially closer to each
other [Dueber 2009] -
however, it should be noted that this study does not involved any
pathway rewiring. By contrast, rational rewiring of pathways involves specific
manipulations to the components of the system to achieve a desired outcome. The most
crucial aspect of protein and gene structure that synthetic biologists use to rewire
pathways is the inherent modularity of many proteins [Janin 1985] (signalling proteins,
for example, typically have dedicated domains for recognizing binding partners that act
independently of other functional
...........................
........
b Changing
pathway output with swappdsd"d
mott~vation
1
hYWlldtype
Ew
eqec
TRIRLATEWSEQDG ------ YLAlSINKD
-AERL1
VRLRYRLEM6DWI.SA ---
Example: Ught-drivn Omppathway
dstrm~iftnmidm
D
EnvZ/RstB chimWc sequence
moo
090Q90
Position ingradient
C Chang pathway output with
5timulus
gneered adapter protein
input Forskolin
n
60-
mu20
prti
enieee
stigtue)cnb
:Rewiredmrtewton
-1,Endogenous interaction
effectors
C
output
S40-
R
togn hvseor(genadcnb
F.IOPOCM
(microspikes)
aet
E
0N
0
0
2
B
6
4
forskobnl WjM)
0
Figure 1.4.1 Rewiring signaling pathways. As the central cartoon shows, membrane
proteins (light blue) can be engineered to have sensors (green), and can be made to
interact with adapters which can in turn can be made to interact with other adapters (dark
blue). More formally the input/output relationship can be controlled in two ways: by
changing the stimulus that a receptor is triggered by (shown schematically in panel A), or
by changing the transducing molecules that the receptor uses to pass the information from
the environment to the cellular interior (panels B and C).
Chimeric photoreceptors illustrate the first type of change. Although chimeric receptors
have been used previously [Kwon 2003], photoreceptors allow for much higher
sensitivity measurements and avoid crosstalk effects. In the case of E coli, the rewiring is
accomplished by transcriptionally fusing the cytosolic signal transduction domain of the
pathway sensor, the histidine kinase domain of EnvZ, to cyanobacterialphytochrome 1
(Cphl), thereby resulting in a system in which EnvZ's response regulator, OmpR, can be
triggered by light [Levskaya 2005]. Pathway activity is read out by placing the lacZ gene,
whose product creates a black compound, under control of the OmpR-depdendent ompC
promoter.
The response to a light gradient input serves as a very precise measurement of the
transfer function of the pathway (panel A, lower subpanel). The transfer curve seems to
indicate that the pathway operates in a threshold linear manner, though whether that is
due to the phytochrome sensor itself rather than the pathway needs to be explored. Such
thresholding could serve to protect the cell from overreacting to small signals.
Shimizu-Sato et al. operated on similar principles in yeast, but instead fused a Gal4
binding site domain (GBD) to the red-light absorbing phytochrome form Pr and a Gal
activating domain (GAD) to the phytochrome's binding partner phytochrome interacting
factor 3 (PIF3), thus bypassing the galactose signaling cascade [Shimizu-Sato 2002]. Any
gene of interest can thus be controlled by placing it under the control the gall promoter
and simply exposing the cells to red light instead of galactose.
Once activated, the signal from the sensor must be specifically transduced to affect
specific downstream processes. By studying covariance among residues from interacting
proteins, one can use statistical scores such as mutual information to predict which
residues determine the specificity of the interaction. As shown in panel B, when
specificity-determining residues from the protein RstB (shown in bold) were substituted
into EnvZ, resulting in the chimeric protein Chim1, phosphotransfer occurred between
EnvZ and RstA rather than EnvZ's normal partner OmpR [Skerker 2008].
Finally, a great deal of signal processing takes place in between the triggering of a sensor
by the environment and the output of the pathway, especially in eukaryotes. One major
intermediate in eukaryotes is the class of proteins known as guanine exchange factors
(GEFs), which control morphological pathways. Yeh et al. swapped wildtype GEFs
controlling formation of filopodia and lamellipodia for synthetic GEFs that can be
induced by the small molecule forskolin and that generate novel morphological outputs67 .
Specifically, GEFs contain an autoinhibitory domain that Yeh et al. substitute with a
PKA-responsive inhibitory domain, PDZ. Placing an endogenous pathway under tunable
control allows us to characterize crucial aspects of cell biology in quantitative detail.
Interestingly, Yeh et al. find that the morphological output is only manifest
probabilistically - it is the fraction of cells that display either filopodia (shown here) or
lamellopodia (not shown) that increases with increasing forskolin.
(A) lower subpanel is taken from Levskaya et al.,
Levskaya A, Chevlier AA, Tabor JJ, Simpson ZB, Lavery LA, Levy M, Davidson EA,
Scouras A, Ellington AD, Marcotte EM, Voigt CA. Engineering bacteria to see light.
Nature 438: 441-442 (2005)
(B) is taken from Skerker et al.
Skerker JM, Perchuk BS, Siryaporn A, Lubin EA, Ashenberg 0, Goulian M, Laub MT.
Rewiring the specificity of two component signal transduction systems. Cell 133: 10431054 (2008)
(C) is reproduced from Yeh et al.
Yeh BJ, Rutigliano RJ, Deb A, Bar-Sagi D, Lim WA. Rewiring cellular morphology
pathways with synthetic guanine nucleotide exchange factors. Nature 447: 596-600
(2007)
domains); most rewiring studies therefore focus on signal transduction and genetic
cascades, which are highlighted in Figure 1.4.1. There are fewer examples of achieving
metabolic
control
through
specifically
designed
changes
in
protein
sequence
[Rothlisberger 2008, Kaplan 2004]. Changes in the structure of an allosteric site in a
metabolic enzyme are more prone to alter the active site of the enzyme than is the case
with signaling proteins [Bhattacharyya 2006]. This property allows for regulation of
metabolic fluxes by effects such as allostery, but the relative lack of modularity also
makes it difficult to forward engineer new behaviors by altering one domain but holding
all others constant.
Even within signaling systems, however, researchers are presented with severe
challenges. Among the major limitations in understanding the signal propagation
characteristics of many pathways is confusion over what cue triggers the cascade and
whether the cue affects other processes taking place in the cell. Take for example the case
of osmotic shock. While many organisms have dedicated signaling systems to relay
information about an osmotic shock to the cell, the presence of abundant osmolyte will
affect numerous processes besides signaling, such as global transcription-factor binding
[Proft 2004]. The examples described below illustrate how techniques that both
specifically and sensitively activate a selected cascade allow one to focus on pathway
behavior independently of such off-target effects.
1.4.2 Manipulating the sensors.
One of the most direct ways of rewiring the input-output relationship of a pathway is by
directly changing the cue that the pathway sensor responds to. If the cue is chosen such
that its level can be directly modulated, then one can measure pathway transfer functions
much as was described above for promoters. Armbruster et al., for example, generated a
G protein coupled receptor (GPCR) that responded to a pharmacologically inert
compound that could then be titrated in to measure pathway response [Armbruster 2007],
while Anderson et al. engineered sensors that can detect changes in tumour-related
microenvironments [Anderson 2006]. Alternatively, one can manipulate the ligands that
drive pathway activity, as was done by Cironi et al. when they linked together epidermal
growth factor (EGF) and mutated forms of interferon a-2a (IFNa-2a) such that the only
cells that could correctly respond to the IFNa-2a signal were those that coexpressed the
EGF receptor [Cironi 2008].
A particularly striking example of how sensor rewiring can shed light on the operation of
a cascade in vivo in a sensitive and specific manner can be found in the use of chimeric
photoreceptors, shown in Figure 1.4.1 A. Two studies used light itself as the cue to drive a
signaling system [Levskaya 2005, Shimizu-Sato 2002]; this approach is unlike traditional
implementations of light-driven systems [Cruz 2000, Cambridge 2006, Dugave 2003],
such as those that use light to activate a small molecule that then activates a desired
biological process [Young 2007]. Levskaya et al. engineered the Escherichia coli
EnvZ/OmpR two-component system to respond to light, while Shimizu-Sato et al.
focused on the Saccharomyces cerevisiae galactose utilization pathway, by fusing a
phytochrome and its binding partner to selected pathway proteins. Armed with this
engineered cascade, Levskaya et al. proceeded to map out the input-output transfer
function with very high precision by exposing a lawn of rewired bacteria to a light
gradient. The transfer function measured in Levskaya et al. suggests that a threshold level
of the environmental cue is needed before triggering pathway activity. While careful
titration of an osmolyte would have allowed precision measurement of the transfer
function, such as through the use of microfluidic devices [Taylor 2009], matching the
sensitivity of a simple light gradient will be difficult to accomplish. Furthermore,
matching the specificity of using light to drive pathway activity is probably impossible.
Given the ease with which we can deliver precisely controlled light signals to cells
compared to delivering chemical signals, the Levskaya et al. and Shimizu-Sato et al.
studies can be easily extended to perform tasks such as measuring Bode plots, as was
recently done for the yeast osmoresponse system [Mettetal 2008, Hersen 2008].
1.4.3 Manipulating sensor/transducer interactions.
Whereas swapping the sensor in a signaling pathway is a way to engineer the input side
of the input-output relationship, changing the identity of the molecules that carry the
signal from sensor to downstream effectors can affect the output side. In fact, given the
high degree of sequence homology between many sensor/transducer pairs, there is great
interest in developing a detailed description of sensor/transducer
interactions to
understand the multiple ways in which pathways prevent crosstalk [Ubersax 2007] - for
example, by using scaffold proteins [Harris 2001], mutual inhibition [McClean 2007],
and kinetic insulation [Behar 2007].
This is the basic strategy that was followed by Skerker et al to rewire the EnvZ/OmpR
system [Skerker 2008]. This study made heavy use of the large amount of sequence data
available for two-component systems to computationally detect individual amino acid
residues that covary between cognate pairs. Specifically, they calculated the mutual
information between all possible pairs of residues from sensors and response regulators
and found the pair that maximized mutual information. These pairs were hypothesized to
be the specificity-determining residues. Remarkably they then substituted a given
sensor's specificity-determining residues for a different sensor's specificity-determining
residues, keeping all other residues intact, and thereby activated the latter sensor's
pathway with the former sensor's trigger. Furthermore -they perform the same rewiring
feat by substituting specificity-determining residues in the response regulator; their
results are highlighted in Figure 1.4.1 B.
For now, the relative paucity of sequence data precludes the use of this technique for
other systems, such as eukaryotic homologues of two-component systems. Nevertheless,
this study provides a framework in which one can go beyond crude domain-level protein
engineering all the way to molecular details. . A particularly enticing possibility, which is
explored in Skerker et al., is to unite the bioinformatically guided rewiring approach with
data on crystal structure, especially structures of protein-protein complexes. Using a
crystal structure of a complex made up of proteins similar to EnvZ and OmpR, Skerker et
al. show that the specificity-determining residues for both sensor kinase and response
regulator probably occur at the interface of the two proteins, suggesting that the
coevolving residues interact physically rather than allosterically. Combining structural
and rewired pathway data can indicate how to explore further the numerous systems in
which docking site interactions have been identified [Tatebayashi 2003, Remenyi 2006].
Synthetic pathways and crystallography together can be key in unraveling fundamental
biophysical interactions underlying signal transduction.
1.4.4 Manipulating the intermediate transducers.
Altering the way in which a sensor interacts with its environmental cues and its
immediate downstream signaling partner represents the most obvious way to manipulate
signal transduction. The next most obvious idea is to follow the signal and tackle the
intermediate transducers in the pathway. Howard et al., for example, took the proapoptotic Fadd death domain and fused it to Grb2 and ShcA, members of the receptor'
tyrosine kinase (RTK) pathway; as a result, RTK-triggered signals could be used to drive
apoptosis [Howard 2003].
At the adapter level, one key target for pathway engineering is the family of guanine
nucleotide exchange factors (GEF) that regulate the actin cytoskeleton through the Rho
family of GTPases [Yeh 2007]. Yeh et al. exploited the presence of an autoinhibitory
domain in GEFs that can be swapped for an inhibitory domain that itself is under the
control of a small molecule. Yeh et al., shown in Figure 1.4.1C, swap wildtype GEFs
controlling formation of filopodia and lamellipodia for synthetic GEFs that can be
induced by the small molecule forskolin. In this study, Yeh et al. daisy-chained two GEFs
in series and show that the combined, and thus longer, GEF system is both more sensitive
to inducer and displays a sharper separation between ON and OFF states. These results
are exactly what one would expect from previous synthetic studies examining the
sensitivity and sharpness of transcriptional cascades as the cascade length is varied
[Hooshangi 2005]. As seen above in the case of apoptosis in the RNAi switch, placing an
endogenous pathway - morphological in this case -
under tunable control allows us to
characterize crucial aspects of cell biology in quantitative detail.
1.4.5 Connecting pathway rewiring to evolvability
Another interesting and complementary theme that emerges from rewiring studies is how
differently rewired circuits can yield the same output. The library of combinatorially
synthesized gene networks constructed by Guet et al. contains instances of systems that
have different connectivity properties but the same Boolean truth table and those that
have the same connectivities but different boolean truth tables [Guet 2002]. Along these
same lines, Isalan et al. show that randomly rewiring the transcriptional network of E.
coli results in growth defects in only 5% of the rewirings, a level of tolerance difficult for
manmade systems to replicate [Isalan 2008]. The idea of rewiring a circuit but
maintaining its logic seems also to have been employed in the evolution of the mating
type switch in yeast, where Candida albicans a genes activate the a mating type while
Saccharomyces cerevisiae alpha genes represses the a mating type [Tsong 2006].
Theoretical studies on the evolvability of biochemical networks suggests that networks
that are wired differently but produce the same output constitute a 'neutral space' that
allows flexibility in the design of networks to perform some function and thus eases the
way for phenotypic changes to take place [Wagner 2007, Gerhart 2007]. Continuing in
the theme of using rewired pathways to highlight system flexibility, Antunes et al.
transplant a bacterial two-component system into the eukaryotic plant Arabidopsis
thaliana. Thee prokaryotic transcriptional activator manages to cross into the nucleus to
drive gene expression, fueling speculation that pathway evolution can even be driven by
horizontal gene transfer between organisms from different kingdoms of life [Antunes
2009].
1.5 Synthetic feedback networks.
Synthesis has uncovered several rules governing how DNA is turned into proteins and
then how proteins interact to generate diverse phenotypes without the need for a
combinatorial explosion in the number of genes. In the examples considered above,
however, the flow of information was largely an ordered sequence of events: diverse
outcomes in these systems resulted from combinatorial rearrangements of modular parts.
The complexity of naturally occurring cellular networks, however, is often dominated by
feedback and feedforward loops. By incorporating these features, synthetic circuits have
also taught us about the dynamics and systems-level function of more complex molecular
interactions.
Initial work in this area primarily focused on the identification [Alon 2007] and
experimental characterization of simple motifs that occur frequently in genetic and
signaling networks. In this first generation of synthetic biology, studies mimicked natural
systems and confirmed theoretical expectations that positive feedback systems can be
bistable [Maeda 2006, Becskei 2001, Ozbudak 2004, Isaacs 2003], negative feedback
systems are noise resistant [Becskei 2000] and can speed up circuit dynamics [Rosenfeld
2002]. More recently, engineered feedback loops have been extended to signaling and
metabolic systems by generating novel protein-protein and genetic interactions to explore
how signaling pathways set their sensitivity to input and how they tune their kinetics
[Bashor 2008, Fung 2005]. One concrete way, highlighted in section 1.5.2, in which
synthetic circuits are helping us approach more complicated interaction networks is by
serving as benchmarks against which theoretical and computational tools can be tested
[Cantone 2009, Ellis 2009].
1.5.1 Oscillatory behaviors.
To make the lessons concrete, we focus on how biological parts can be arranged to create
a biologically relevant dynamical system: an oscillator. Cells display a range of
oscillatory behaviors. Some oscillators have tunable periods, such as the dependence of
the cell cycle period on nutrient levels available, whereas others are more robust to
changes in parameters, such as the circadian oscillator. Examples include oscillatory
signaling from nuclear factor kappa B (NFkB), which governs its control over gene
expression [Nelson 2004], and p53-murine double minute 2 (Mdm2), which oscillates to
drive the DNA damage response [Geva-Zatorsky 2006]. How can one construct a robust
yet tunable oscillator in a living cell? The construction of in vivo oscillators provides a
particularly beautiful example of how the interplay of analysis of naturally occurring
systems, modeling, and construction of synthetic systems can yield deep insights into
biological phenomena. The story here begins with the observation that the simplest
oscillator design, a delayed negative feedback, cannot sustain oscillations beyond a small
number of periods when operating in a cell. Instead, as highlighted in Figure 1.5.1,
naturally occurring oscillators hinted at the crucial role of interlocking positive feedback
in maintaining a robust oscillator, which was employed in the genetic oscillators recently
syntheisized by Stricker et al [Stricker 2008] and Tigges et al [Tigges 2009].
As the studies in Figure 1.5.1 show, oscillators, in addition to being fun to watch, are
among the simplest in vivo systems that can be used to understand interactions between
different types of feedback loop. While systems biologists are increasingly comfortable
with our understanding of simple motifs, we cannot say the same about interactions of
those motifs. It is worth considering, for example, that even for interlocking positive and
negative feedback loops multiple behaviors are possible as one varies the parameters of
the system and includes stochastic effects. For example, in the yeast galactose utilization
pathway, the negative feedback effectively counteracts the positive feedback and limits
the parameter space over which the system is bistable [Acar 2005]. Beyond two or three
loops, however, we are usually at a loss to describe the system - especially a natural one
that may contain even more interactions than is being accounted for. Synthetic circuits
are helping us systematically understand how motifs interact to generate ever-richer
behavior.
..
.......................
o
a
Ana~fse natural
circuit to
improve on initial
120-
synthetic design
S100-
so -
b
Cycn B
Cdc25.
60-
A Cdc2O
DKl-CydinB
Myt (C
20-
0
02
0
100
200
0Mitotic
300
400
Time (mn)
500
phosphorylations
600
implement findings from dual feedback
natural network in a synthetic sy stem
C
Microbial
Mammalian
10-
12.
2.0-
.A
12
6-
20
01.0-
.0-
0
60
120
Time (min)
180
0
200
400
600
80
Time (mn)
1,000
1.200
Figure 1.5.1 Building a robust, tunable oscillator in a living cell. The simplest way to
achieve oscillation is through use of a delayed negative feedback loop [Conrad 2006].
Imagine that you construct a system with two genes, A and B, and that protein A activates
the transcription of B whereas protein B inhibits the transcription of A. Turning on gene A
leads to build up of the protein A, but also of protein B. After some time, enough protein
B builds up to cause protein A levels to decrease - this then results in a decrease in
protein B levels, which allows protein A levels to rise, and so on.
However, when one builds a simple negative feedback circuit as described above, the
oscillations are in general not robust. In the repressilator of Elowitz and Leibler [Elowitz
2000], which consists of a cycle of 3 transcriptional repressors and a fluorescent protein
readout (panel A), the oscillators fall out of phase and damp out following a small
number of cycles. Swinburne et al. engineered an autoinhibitory circuit in which the
delay timescale in the negative feedback was set by the length of an intron engineered
into the construct; they also find that even for a given intron length the oscillation period
varies widely from cell to cell [Swinburne 2008]. The source of the damping in both
cases can be found in the stochastic nature of gene expression: random amounts of
protein produced at random times result in uncoordinated behavior that causes the
components making up the oscillator to fall out of phase. The synthetic genetic oscillator
was missing a key ingredient.
A strong hint as to the identity of that key ingredient was provided by the analysis of
naturally occurring oscillators. In particular, the cell cycle oscillator contained
interlocking positive feedback loops in addition to the core negative feedback loop that
was generally assumed to generate the oscillations (panel B). Experiments in the cell
cycle of frog embryos along with computational simulations suggested that the positive
feedback loops could stabilize two states that the system would cycle between via the
negative feedback loop [Pomerening 2003, Pomerening 2005, Tsai 2008], creating a
relaxation oscillator. Could something as simple as positive feedback be responsible for
robustness in genetic oscillators in organisms as diverse as bacteria to mammals? And
can positive feedback enable cells to independently tune the amplitude and frequency of
the oscillations?
Two recent studies, in agreement with earlier work [Atkinson 2003], indicate that
coupling positive and negative feedback is indeed sufficient to ensure stable oscillations.
Stricker et al. implemented a transcriptional circuit in E. coli that integrates a positive and
negative feedback loop in a common inducible promoter [Stricker 2008] (panel c), while
Tigges et al., working in mammalian cells, used transcriptional positive feedback and
negative feedback mediated by transcription of an antisense RNA [Tigges 2009].
Experimentally, Stricker et al. observe that the dual feedback oscillator is robust to a
number of perturbations, including changes in inducer level and temperature; these
features could not be adequately described by their initial modeling of this circuit [Hasty
2002]. It was only through the addition of various biological steps in the negative
feedback, such as TF-DNA binding and multimerization, that the model could reproduce
the robustness of the oscillator to parameter changes. The authors conclude that from the
point of view of the oscillator's operation, what matters is not the details of what
processes make up the negative feedback but instead that the negative feedback includes
a delay; by contrast, the positive feedback only ensures robustness and tunability.
The system built by Tigges et al. shares many of these details, with the delay in the
negative feedback coming from post-transcriptional repression of the circuit's
transcriptional activator, but the system itself is sensitive to molecular details such as the
relative ratios of the circuit components - for some ratios of circuit components
oscillations are abolished.
(A) is taken from Elowitz MB, Leibler S. A synthetic oscillatory network of
transcriptional regulators. Nature 403(6767): 335-338 (2000) (B) is reproduced from Tsai
TY, Choi YS, Ma W, Pomerening JR, Tang C, Ferrell JE. Robust, tunable biological
oscillations from interlinked positive and negative feedback loops. Science 321(5885):
126-129 (2008) (C) is taken from Stricker J, Cookson S, Bennett MR, Mather WH,
Tsimring LS, Hasty J A fast, robust and tunable synthetic gene oscillator. Nature
456(7221): 516-519 (2008) and Tigges M, Marquez-Lago TT, Stelling J, Fussenegger M.
A tunable synthetic mammalian oscillator. Nature 457(7227): 309-312 (2009)
1.5.2 Using synthetic circuits as modeling benchmarks
One of the most important functions that synthetic circuits have served has been their use
in building and refining analytic and computational models of biological systems. When
modeling a gene or protein circuit, one must make a series of choices. The first choice
has to do with how fine a scale one wishes to model the input/output relationship typically this choice boils down to whether one wants to view the system as a Boolean
logic operator or a dynamical system. The dynamical system framework can be further
broken down along 2 dimensions, depending on whether spatial or stochastic effects need
to be taken into account. Spatial effects can usually be ignored when the biochemical
reactions that make up the system occur on timescales slower than the time it takes to mix
the reactants by diffusion. Stochastic effects can usually be ignored if the dynamical
variables of the system can be represented as continuous rather than discrete entities; that
is, when we are interested in the concentrations of a molecule rather than the number of
molecules. Synthetic circuits have been used to explore all of these issues in some detail.
Until recently, the choice of modeling methodology was based on one's best guess for
which effects were important to include, along with post-hoc comparison of the model
with data. Detailed comparisons of different modeling paradigms have been lacking.
Cantone et al. [Cantone 2009] and Ellis et al. [Ellis 2009] have offered the field some
guidance through the introduction of benchmark networks -
that is, a network that has a
defined topology that interacts only minimally with endogenous systems, against which
to test proposed modeling methods. In particular, Cantone et al. create a relatively
sophisticated synthetic transcriptional network of 5 genes that serves as an oracle that is
queried by different perturbations, such as overexpression of the network genes and
induction by transcriptional inducers. Finally they test methods based on ordinary
differential equations, bayesian inference, and information theory to uncover the
connectivity of the network; they find that differential equations and Bayesian inference
were better at uncovering the functional relationships than the information theory-based
approach, as expected for such a small network. Cantone et al. thus provide an example
of how synthetic circuits can be helpful in refining our understanding of large-scale
biological systems by improving the algorithms we use to analyse genomic and
proteomic datasets.
1.6 The ultimate goal: spatiotemporal control.
If there is one context in which all of the various biological processes tackled by
synthetic biologists come together it is in the engineering of spatiotemporal interactions,
both intracellular and intercellular. Engineering cell-cell interactions in a rational manner
requires us to master rational manipulation of communication devices (signaling
pathways), using promoters to specify desired transcriptional responses to a given signal
strength, and arrange these elements in a circuit architecture that robustly encodes the
function we are trying to implement. If we hope to systematically build up our
understanding of functional compartments of the cell, development, and ecology then it is
imperative that we integrate lessons learned from diverse areas of synthetic biology.
1.6.1 Uncovering intra- and intercellular processes
Perhaps the most striking feature of the eukaryotic cell is its organization into functional
subcompartments: the nucleus for genetic material, mitochondria for respiration,
endoplasmic reticulum (ER) for protein production, etc. For the eukaryotic cell to
accomplish its tasks, the behavior of these compartments must be coordinated in space
and time. A recent study in S. cerevisiae from has yielded new insight on how the
mitochondrion and ER communicate, by using a genetic screen coupled with a synthetic
construct that is designed to specifically tether the two organelles [Kornmann 2009].
Kornmann et al. find that the synthetic tether complements mutations in maintainance of
mitochondrial morphology 1 (Mmml), mitochondrial distribution and morphology 10
(Mdm1O), 12 (Mdml2), and 34 (Mdm34), thus identifying these 4 proteins as
constituents of a complex that ties the organelles together and allows the exchange of
phospholipids (needed by the mitochondrial membranes) and calcium (which acts as a
signaling molecule between the two).
Two properties that we still cannot reliably engineer are the dynamics of a circuit and
spatial control. Both these behaviours have one major biological process in common:
development. In anticipation of one day tackling developmental processes and other
intercellular pathways, some groups have designed circuits to spatiotemporally control
gene expression. Using a network mimicking naturally occurring feedforward circuits, for
example, Basu et al. have designed cells that can respond to the signal acyl-homoserine
lactone (AHL) from nearby cells but ignore equal concentrations of this signal from
faraway cells [Basu 2004]. This feat is accomplished by a key property of the
feedforward network in the signal receiving cells - it responds not only to the
concentration of the signal but also to the rate of increase of that concentration. Signal
sending cells nearby signal receiving cells increase the rate of AHL concentration more
rapidly than distant sending cells. Basu et al. built on this work to create a circuit that
could respond to only a narrow range of AHL signal, much like a band filter, thereby
exhibiting another feature of developmental processes [Basu 2005].
The exquisite coordination that is a hallmark of development also almost certainly
requires the use of networks that can act as genetic timers and counters. Friedland et al.
have provided a design for a network that constitutively pumps out GFP mRNA
transcripts that are translationally inhibited but whose inhibition can be lifted by a transactivating RNA (taRNA) [Friedland 2009]; the transcription of the taRNA is inducible by
arabinose and so the network output, in the form of discrete amounts of GFP, represents
pulses of arabinose. Finally, Isalan et al. have gone as far as building a mock-up of a
realistic D. melanogaster embryo, modeling the syncytium as a collection of
paramagnetic beads coated with DNA, in which genetic networks analogous to the gap
gene system can be placed [Isalan 2005]. Interestingly, this 'minimal embryo' leads the
authors to suggest that pattern formation in the real embryo requires activator molecules
to propagate faster than inhibitors, implying that the gap system is a reaction-diffusion
system that uses a mechanism quite unlike Turing instabilities to lay down patterns. As
the authors point out, this is hardly surprising given that the gap system uses
nonhomogeneous initial conditions in the form of spatially localized components
deposited in the insect egg and as the activator is not autocatalytic. Whether these lessons
carry over to their natural settings remains to be seen.
1.6.2 Modelling ecological interactions.
As is the case with the band filter circuits described above, most synthetic circuits
involved in cell-cell communication make use of the quorum sensing pathway; one such
circuit is highlighted in Figure 1.6.1 [Tanouchi 2009]. These circuits usually borrow
components from organisms like Vibriofischeri, although attempts at incorporating other
systems have also been successful [Bulter 2004, Chen 2005]. Examples of using such
systems to study natural phenomena are more limited. Balagadde et al., by adapting an
earlier design [You 2004], used the quorum sensing proteins to drive expression of an
antibiotic to create a synthetic predator-prey system [Balagadde 2008], while Brenner et
al. used a similar system to study the ability of cells to signal in the context of a biofilm
[Brenner 2007].
Synthetk predator-prey systm
1..20
10-
0
0
40
80
120
Time (hr)
160
200
Figure 1.6.1 Building a synthetic predator-prey system from quorum sensing
components: using synthetic circuits to engineer cell-cell interactions. Studies of ecology
and evolution are often dependent on carefully characterizing the interactions of different
organisms. In a natural setting, however, such data collection often proves to be noisy at
best and impossible at worst. At the same time, mathematical models in theoretical
ecology and evolutionary biology are among the most sophisticated in all of the life
sciences. Laboratory-scale experiments on cellular interactions could quantitatively test
some of the remarkable predictions and open the way to new theory. Among the most
elementary interactions in nature is the predator-prey interaction. The prey in this case
produces the quorum sensing pathway protein LuxI, which is engineered to drive a
transcriptional cascade in the predator that produces CedA, which inhibits the DNA
replication inhibitor CedB thereby allowing the predators to replicate. Meanwhile the
predator produces the quorum sensing pathway protein LasI which activates CcdB in a
LasR-dependent manner in the prey. CcdB expression in the prey prevents it from
replicating. The cyclic dynamic is very similar in style to genetic oscillators: high levels
of prey leads to low levels of CedB and thus high levels of predator; high levels of
predator leads to high levels of CcdB and thus low levels of prey, which subsequently
leads to high levels of CcdB in predators, etc. As shown in Balagadde et al., predatorprey interactions can thus result in limit cycle oscillations about an unstable fixed point of
the dynamics, most commonly studied in the framework of the Lotka-Volterra model.
Reproduced with permission from Balagadde FK, Song H, Ozaki J, Collins CH, Barnet
M, Arnold FH, Quake SR, You L. A synthetic Escherichia coli predator-prey ecosystem.
Mol Syst Biol 4: 187 (2008)
Chuang et al. recently have used engineered circuits shown in Figure 1.6.2 for cell-cell
interactions to study the evolutionary phenomenon of Simpson's paradox, in which the
cells that provide a useful product to the population make up a diminishing fraction of the
population but nevertheless increase in absolute number by promoting population growth
[Chuang 2009]. Gore et al. provide another example of synthetic ecology in their study of
the evolutionary game dynamics underlying sucrose metabolism in yeast [Gore 2009].
The study establishes that sucrose metabolism can be thought of as a snowdrift game, in
which both cells that metabolize sucrose (cooperators) and those that do not (cheaters)
stably coexist in a population, thereby opening an avenue to show how competition
between different alleles can actually promote diversity in a population.
............
.
........................
.
...................
...................
.
Smpson paradOK inongin..rmd int3ractions
*
RN autoinducer
P.t
ecat
10-weE
pools
Vector legend
Vector legend.
P
Pf
a= predicted p'
068
E
43
0.4
-10-
0
-~0.2-
I
I
I1
1
0.2
0.4
0.6
0.8
I
1
0.2
0.4
0.6
0.8
Initial pJ and final (p'=p + Ap)
Initial (p)and final (p'=p + Ap)
producer proportions
global producer proportions
1
Figure 1.6.2 Simpson's paradox observed in an engineered set of cell-cell interactions.
Simpson's paradox is a statistical phenomenon that captures the fact that even if the
producer of a common good grows at a slower rate in all given subpopulations than a
nonproducer, it can nevertheless make up an increasing fraction of the population as a
whole. While Simpson's paradox usually arises as a result of misinterpretation of data,
natural populations can in fact display heterogeneities in sample size that often underlie
the paradox. The particular implementation in Chuang et al. casts bacteria that generate
the autoinducer Rhl as the producer. Both producer and nonproducers use this Rhl that is
rewired to activate synthesis of a chloramphenicol resistance gene catL VA. As shown in
the middle panel, in each subpopulation the fraction of producers decreases, but as the
bottom panel shows in the global population the fraction of producers actually increases,
thus satisfying Simpson's paradox. Reproduced from Chuang JS, Rivoire 0, Leibler S.
Simpson's paradox in a synthetic microbial system. Science 323(5911): 272-275 (2009).
.
..............
Studies such as these on fundamental aspects of ecology and evolution are difficult to
carry out in natural environments due to the multiplicity of confounding factors, but
synthetically engineered populations provide a way to cleanly separate different effects.
More generally, studies on engineered populations not only highlight the ability to
connect the molecular details of a network to population level effects but also the utility
of abstracting away from such details and focusing on the interactions between cells
Taking sucrose metabolism from Gore et al. as an example, it was possible to predict
population level responses to changes in the cost of cooperation just on the basis of the
game theoretic characterization of the interaction between cheaters and cooperators, with
no direct knowledge of the molecular details. Indeed, this approach of constructing
synthetic systems dedicated to characterize how cells interact can be very useful in cases
such as cancer dynamics, where the underlying molecular details are either poorly
understood or exceedingly complicated but population level measurements are both
feasible and relevant to understanding the phenomenon.
1.7 Perspectives
The synthetic biology community has made great strides in working out some of the most
basic features of regulatory networks and cellular pathways. We are exerting greater
control over the process of gene expression, and we have a wealth of information
regarding the effects of network topology on system function. Topological details such as
connectivity, cascade length, feedback structure have been explored. But there is much
work yet to do before we can treat biological circuits like we treat electronic ones.
In the future, we can expect to see that the synthetic circuits deployed in cells will be of
growing complexity, and should increasingly integrate diverse processes, as has been
done for genetic regulation and metabolism [Fung 2005]. We should also expect to see
increasing contact with large-scale cell biology, such as through the creation of synthetic
organelles, whose in vivo construction will be guided by synthetic regulatory networks.
Progress along these fronts is limited by many of the same obstacles found across the sub
disciplines of biology: we are still in need of more ways to specifically modulate the
expression level of genes of interest, the activity state of pathways of interest, and we
require more sensitive techniques (ideally at single-molecule resolution) to measure the
abundance of mRNAs, proteins and specifically modified proteins in live cells.
One of the main ways in which methodological advances will be useful is in tightly
constraining models of biological networks. Obstacles to rapidly moving synthetic
circuits from the blackboard to the cell can often be traced to the fact that the system
under study does not behave as initial modeling indicates. This, in turn, is usually due to
the fact that the systems are underdetermined, meaning that many different models can
usually describe the circuit data. Higher resolution data, both in terms of abundances of
the relevant molecules and as a function of time, will constrain the space of possible
models significantly and should allow for more rational, predictable design processes.
Assuming these technical obstacles are overcome, in a future where man-made circuits
increasingly look like their byzantine natural counterparts, it is not unreasonable to
expect nearly synthetic or fully synthetic cells to make their appearance. At these extreme
levels of complexity, it may prove difficult or even unhelpful to mechanistically model
the relevant systems. It is likely, however, to prove useful to compare the performance of
natural and synthetic circuits and cells in a rigorous fashion - perhaps through the
formulation of a Turing test for synthetic biology - as differences in performance can
point to possible design principles.
Looking back on the various examples of circuits and processes that synthetic biologists
have examined, we can see that the utility of synthetic circuits can be measured along 3
different dimensions. First, synthetic circuits can serve as easily manipulable toy models
that we can characterize in exacting quantitative detail and thereby build intuition for
how similarly structured natural networks operate. Second, synthetic circuits can be used
to allow us control over natural networks and so make discoveries about the molecular
and cell biology underlying important physiological processes. Third, on a more
conceptual level, synthetic systems provide clear evidence that one can generate
complexity by rearranging even well-known parts, thus bolstering claims on the
evolvability of natural systems.
While we are still very far from rationally assembling a living organism from scratch, and
far from understanding all the design principles according to which biological networks
operate, the first generation of synthetically designed systems have offered us a glimpse
at the need to weave our tools from disparate processes from transcriptional regulation to
signal transduction in order to approach fundamental questions in modem biology.
50
Chapter 2
Robust yet tunable regulatory networks:
the case of the yeast osmosensing pathway
2.1 Summary
Genetic variation underlies much of the phenotypic diversity observed in nature.
However, the functional robustness of cellular networks to coding sequence variations of
its component genes is often difficult to quantify. Here, we challenged the osmosensing
signaling pathway in the budding yeast Saccharomyces cerevisiae by systematically
swapping each component gene, except for its terminal MAPK gene HOG1, with its
orthologs from various yeast species, and measured their abilities to recapitulate wildtype signaling. We found that signaling was significantly altered by sequence variation in
the downstream MAPK cascade genes, but remained relatively robust to changes in the
upstream phosphorelay components. These experimental findings are consistent with a
computational sensitivity analysis that predicts that HOG signaling is most sensitive to
kinetic parameter changes involving the MAPK cascade genes. We then performed
evolution experiments on yeast cells with hyperactive HOG signaling, and found that
they rapidly adapted and restored wild-type fitness and signaling predominantly due to
point mutations in the MAPK genes. Our results suggest that the skewed sensitivities of
signaling dynamics to underlying component variations is a direct consequence of its
biochemical circuitry, and might impact the evolvability of this network.
2.2 Introduction
Remarkably, organisms can exhibit phenotypic robustness against a diverse array of
stochastic, environmental and genetic variation. Genetic robustness in particular, can
endow organisms with the ability to maintain phenotypic stability against genetic
perturbations [de Visser et al., 2003], thus making them less vulnerable to mutations.
Despite its apparent prevalence in nature, understanding of genetic robustness and its
consequence for evolvability remain elusive. Mathematical modeling has played a pivotal
role in advancing the study of robustness [Barkai 1997, Alon 1999, von Dassow 2000,
Kitano 2004]. Using quantitative modeling, genetic variation can be simulated by varying
the kinetic parameters in the dynamical model, and the system's robustness to genetic
perturbations can be predicted. Despite the widespread utility of such computational
approaches, there has been a dearth of experimental studies that either directly or
comprehensively test these predictions.
Here, we take a three-pronged strategy combining experimental, computational and
evolutionary approaches to investigate the robustness of cellular signaling to genetic
perturbations of its underlying molecular network. We employ the high osmolarity
glycerol (HOG) pathway in the budding yeast Saccharomyces cerevisiae, which forms a
core module of the hyperosmotic shock response [Hohmann 2002]. This pathway is
especially well suited for robustness analysis because its molecular components and
interactions have been well characterized [Brewster 1993, Maeda 1994, Posas 1996,
Krantz 2009]. Moreover, its network input (extracellular osmolyte concentration) and
output (Hog 1 activity) can be quantitatively measured and manipulated.
The HOG pathway consists of a phosphorelay chain of proteins (Slnl, Ypdl and Sskl)
that acts on a downstream MAP kinase cascade (Ssk2, Pbs2 and Hogl) to ultimately
modulate HogI activity, as shown in schematic form in Figure 2.2.1. When the cell faces
a hyperosmotic shock, the turgor pressure on the cell membrane drops [Posas 1996],
reducing the autophosphorylation of Slnl. Lack of phosphorylated Slnl limits the
phosphate current in the direction of Sskl, ultimately resulting in build-up of
dephosphorylated Sskl. Dephosphorylated Sskl catalyzes the phosphorylation of the
MAPKKK Ssk2 and results in activation of the remainder of the MAPK pathway. Upon
activation, Hogi translocates into the nucleus [Ferrigno 1998] to initiate transcriptional
changes in response to the osmotic shock [O'Rourke 2004].
Sini
HoI
Figure 2.2.1 Schematic depiction of the Saccharomyces cerevisiae
HOG pathway. The regulatory arrows depicted in grey indicate
The study outlined in this chapter consists of three stages. First, we systematically
generate mutant strains in which each component gene upstream of HOG1 is replaced by
its orthologs from various yeast species, and then we measure their corresponding
signaling dynamics. We find that while signaling is tolerant of coding sequence variation
in the upstream phosphorelay genes, it is significantly less robust against changes in the
MAPK cascade components. Second, we show that the experimental results are
consistent with a computational robustness analysis of the pathway, which predicts that
HOG signaling is most sensitive to kinetic parameter changes involving the MAPK
cascade genes. Third, we underexpress the YPD1 gene in the HOG pathway to induce
hyperactive signaling, and we evolve nine independent lines of the yeast strain
underexpressing YPD1 using turbidostats [Acar 2008]. We find strikingly that mutations
in the MAPK cascade genes i.e. PBS2 and SSK2 significantly dominate the genetic
changes among the pathway genes across the majority of the independently adapted
populations consistent with the computational robustness analysis. Importantly, we show
that these mutations are largely responsible for the down-regulation of the hyperactive
signaling and the improved fitness in the evolved strains.
2.3 HOG signaling displays varied sensitivity to ortholog
substitutions
To characterize the effects of sequence variations in the genes of the HOG pathway on
signaling, we utilized the natural variation in the HOG pathway genes across different
yeast species and systematically generated mutant strains in which each pathway gene
except HOG] was replaced with its ortholog from two evolutionarily diverged yeast
species i.e. Candidaglabrataand Candidaalbicans. Then, we measured their abilities to
recapitulate wild-type signal propagation under a hyperosmotic shock. By using
presumably functional orthologs rather than randomly mutated sequences, we more
efficiently searched the space of sequences that had a reasonable chance of
complementing wild-type behavior. Compared with S. cerevisiae, all C. glabrata
pathway proteins had sequence ClustalW similarity scores between 50 and 60, except
Ssk1 which scored 37. C. albicans, being evolutionarily more distant from S. cerevisiae
than C. glabrata, displayed lower sequence conservation for all the pathway proteins,
ranging from 22 to 46. To estimate the degree of protein functional changes manifested
by the sequence divergence of the orthologs, we computed for each ortholog the
percentage of amino acid changes at highly conserved residues ("functional score")
identified from comparative genomic analyses of the HOG pathway proteins across
various fungi species (Supplemental Data, Krantz et al., 2006). We calculated "functional
score" as the percentage of amino acid changes in the orthologous sequence compared to
that of the S. cerevisiae sequence at conserved residues identified through multiple
sequence alignment of orthologous genes from twenty fungal species (Krantz et al.,
2006). Here, we consider a residue as being conserved if either all the residues at that
position are identical across all sequences in the alignment, or if conserved or semiconserved substitutions are observed.
...........................
In order to assay for pathway activity, we made use of the fact that pathway activation
leads to increased localization of the Hog1 protein to the nucleus. To this end we fused
the yellow fluorescent protein (YFP) to the C-terminal domain of Hog1 to track its
subcellular localization, and we labeled the nucleus of each single cell by using strains
that contained a fusion of the nuclear pore protein Nrdl to the red fluorescent protein
(RFP). To drive pathway activity, we exposed the cells to standard dropout media
containing O.4M sodium chloride (NaCl); the media used to culture the cells was
controlled by means of a fluid cell device which allowed for computer-controlled
switching between media with and without NaCl as shown in Figure 2.2.2.
Top View
Slide Gasket
Computer
control
Inlet
Cells
Coverslip
with ConA
Side View
Figure 2.3.1 Schematic diagram of flow-cell setup. Reproduced from
[Mettetal 2008].
2.3.1 Systematic complementation study
We found in Figure 2.3.2 that upon a hyperosmotic shock, the Hogl nuclear enrichment
dynamics of the Slnl- and Ypdl-ortholog hybrid pathways (from both yeast species)
were indistinguishable from that of the wild-type response despite their high functional
scores. By contrast, the majority of Ssk2- and Pbs2-ortholog hybrid pathways displayed
grossly defective signaling. While the hybrid pathways comprising the C. glabrataand C.
albicans phosphotransfer module proteins exhibited almost identical initial Hog1
phosphorylation rate, peak Hog1 nuclear enrichment, and adaptation time compared to
wild-type, their protein kinase counterparts displayed significant signaling changes such
as decreased initial Hog1 phosphorylation rate and peak Hog1 nuclear enrichment.
Importantly, the ability of the hybrid pathways to approach wild-type signaling did not
correlate in any simple way to sequence conservation. For example, C. albicans Ssk2 and
Sln1 have similar functional scores indicating that each protein has a similar fraction of
.. ...
...
.......
..
....
..........
....
..
highly conserved amino acid residues changed, but clearly C. albicans Sln1 can
complement its S. cerevisiae counterpart, while Ssk2 cannot.
SLN1, WT
YPD1, WT
SSK2, WT
.
PBS2, WT
0.3
4*
60.2
E
0.2
0
9%
22
0.3
0
10
2*/
12
2 38% 06142
31
2%17%.
-~0.2
14*/
3%
22
03172
.
0
10
1I
35
27%.
0
20
0
0
-14
2
9%
L~ST
0
10
20
0
10
20
0
10
20
0
10
10
20
Time [minutes]
Figure 2.3.2 Hogl nuclear enrichment dynamics in response to a 0.4 M NaCl
hyperosmotic shock measured in the wild-type strain (in gray) and mutant
strains with each of the pathway proteins (denoted by the different colors), except
Hogl, replaced with its orthologs from C. glabrataand C. albicans. Shown in
the upper right corner of each plot is the ClustalW score of the ortholog when
aligned to the S. cerevisiae sequence. Right below the ClustalW score is the
"functional score" which for each ortholog, represents the percentage of
amino acid changes at highly conserved residues identified via comparative
genomics. The traces show the average response, obtained by taking the
average of population averages from independent experiments (n = 3) ± SEM.
Systematic complementation study suggests that HOG pathway can tolerate
large-scale protein sequence variations in phosphotransfer proteins and
maintain endogenous pathway dynamics, while changes in MAPK protein
sequences tune the pathway dynamics to the point of eliminating activity
altogether in the case of C. albicans Ssk2.
2.3.2 Focusing on network architecture: Pbs2 versus Ypdl
To further substantiate this finding, and in order to examine the role of network context in
producing the observed sensitivities we focused on the proteins Ypdl and Pbs2 which
belong to the phosphorelay and MAPK modules respectively. Ypdl is a phosphotransfer
protein sandwiched between phosphotransfer proteins, whie Pbs2 is a kinase protein
sandwiched between kinase proteins; the significance of this difference will be
_
............
highlighted in section 2.4.2. We generated strains with PBS2 and YPD1 ortholog
substitutions from three evolutionarily more distant yeast species including Neurospora
crassa, Debaryomyces hansenii, and Kluyverimyces lactis. Despite higher sequence
divergence and functional scores of these Ypdl proteins, the data shown in Figure 2.3.3
shows that all of them still fully mimicked wild-type Hogi signaling. In contrast,
signaling performance decreased with increasing sequence divergence and functional
scores of the Pbs2 protein.
1.5
-
-
V
E
1
PBS2
---1
ITorthologs
m YPD1
orthologs
EIE 0.5
E
0
0
10
20
30
% of amino acid changes at well-conserved residues
Figure 2.3.3 Maximum Hogl nuclear enrichment of mutant strains
with orthologous YPD1 and PBS2 genes of varying degrees of "functional
scores" under a 0.4 M NaCl hyperosmotic shock normalized against the
wild-type response. Data point at 0 percentage change represents the
wild-type response. Data depicts mean (n = 3) ± SEM. A broad range
of orthologous versions of Ypdl are able to fully complement the
endogenous version of Ypdl, while pathway activity decays as the
substituted ortholog of Pbs2 becomes increasingly dissimilar from the
native Pbs2.
The absence of observed changes in Hogi dynamics for the Slnl- and Ypdl- ortholog
hybrid pathways, however, did not preclude the possibility that they possessed similar
kinetic constants as that of the S. cerevisiae Ypdl protein. Therefore, we took advantage
of a previous in vitro study that had characterized Ypdl mutants with drastic changes in
either the phosphotransfer rate ksnzjpsyd1 or binding constant Kdln1pypd1 between
phosphorylated Slnl and Ypdl (Janiak-Spens et al., 2005). We transformed these Ypdl
alleles into our wild-type strain. We then measured their signaling abilities under the
same hyperosmotic shock. None of the strains with kinetically defective alleles displayed
significant changes in Hog1 signaling dynamics compared to wild-type, even in the case
where kslalp-ypdj was reduced by 17-fold; this is evident by the flat curve in Figure 2.3.4.
Using these kinetically defective Ypd I alleles, we were able to directly rule out the
possibility that, in the case of Ypdl, the ortholog complementation tests were leading us
to a false conclusion that Ypd1 rates seem not to affect cascade dynamics. Taken
together, these complementation experiments led us to hypothesize that HOG signaling is
likely to be more robust to variations in parameters affecting the upstream
phosphotransfer relay than the downstream MAPK cascade.
1.5
mutants
E o~: 0.5
0
-20
-10
0
10
ksn1-yPdl (s-1) or KdSIMP-Ypdl (pM) fold change
(relative to WT-YPDI)
Figure 2.3.4 Maximum Hog1 nuclear enrichment of mutant strains with
characterized YPD1 alleles under a 0.4 M NaCl hyperosmotic shock normalized
against the wild-type response. Two of the alleles exhibit a three- and
seventeen-fold reduction in the Slnl-to-Ypdl phosphotransfer rate ksljP-ypd1,
while another has a three-fold increase in the binding constant Kdsn1paydl
compared to wild-type Ypdl [Janiak-Spens 2004]. Data point at 0 fold
change represents the wild-type response. Data depicts mean (n = 3) ± SEM.
Kinetic mutants of Ypdl, much like orthologous versions of Ypdl, were readily
able to complement the wildtype Ypdl in terms of pathway dynamics,
demonstrating that the HOG pathway is robust to variations in the kinetic rates
associated with the internal details of the phosphotransfer module.
2.4 Computational analysis of HOG pathway
To computationally investigate the effects of coding sequence variations in the HOG
pathway genes on signaling dynamics, we performed sensitivity analyses on key
dynamical properties of the signaling module, i.e. the peak Hogi phosphorylation level
MHog1
and the initial Hog1 phosphorylation rate rHgi, using a simplified biochemical
network model [Klipp 2005]. Similar to an approach used to study the segmentation
polarity network in fruit flies [Dassow 2000] we modeled changes in coding sequences as
changes in the kinetic rate constants parametrizing the dynamical model.
The effects of coding sequence variations in each pathway gene, except HOG], were
simulated by simultaneously varying all three rate constants associated with the protein
over two orders of magnitude about wild-type levels and computing the corresponding
signaling outputs i.e. MHogJ and rHogi. Figure 2.4.1 illustrates how MHogJ changes as two
of the three rate constants associated with either Pbs2 (Figure 2.4.1, left panel) or Ypdl
(Figure 2.4.1, right panel) are varied about wild-type parameters. We observed a
strikingly flat surface for Ypdl-associated parameter changes, indicating that
MHogJ
remains almost unchanged over a wide range of parameter space. In contrast, Pbs2associated parameter changes significantly altered the MHogJ landscape. To systematically
compare the effects of parameter variations across individual pathway proteins on
signaling, we computed the local logarithmic gradient of the landscape evaluated at wildtype levels and defined this metric as our sensitivity measure.
Phosphorelay intermediate protein Ypdl
MAPKK Pbs2
C
0 0.16
.~0.1'
10.0
0.06
4
3C
0
0
20
a.
-2
4
-3
.4 -2
110
2
22
kypdjj.ajni
0
0
"o
10
kAn
ypdj
-1
0910 k=.,s,
Figure 2.4.1 Surface representations of peak Hogi phosphorylation as a function
of the rates associated with a given protein. While YpdI exhibits a very flat
surface, indicating that the peak HogI phosphorylated level is relatively constant
over a wide range of Ypd1-associated rates, the Pbs2 surface exhibits greater
curvature. In subsequent analyses, we summarize these surfaces by plotting the
magnitude of their gradients evaluated at given base points, such as the estimated
wildtype parameter set.
2.4.1 Sensitivity analysis of a model of the HOG pathway
We implemented the following steps: i) model changes in sequence as changes in kinetic
rate constants, ii) define a sensitivity metric that captures how HOG signaling changes as
kinetic rate constants are varied using the model shown below in Equation 2.8. We
examined several methods to execute step (ii). The first analysis involved computing the
magnitude of the local logarithmic gradient about the wild-type parameter set from the
model outputs namely initial Hogi phosphorylation rate and the steady state Hogi
phosphorylation level. To directly compare the different model outputs, we utilized
logarithmic gradient calculations to render our analysis dimensionless:
i-1
( Alnk
ln
)
2[2.1]
i wt
where # is the model output whose sensitivity we are computing, and the k's represent
the rate constants that are being varied. The wild-type parameters are obtained from
Klipp et al. although similar results are obtained in a model with wild-type rate constants
--
------------------
set equal to one another. The results of this analysis for # are summarized in Figures
2.4.2 and 2.4.3.
Sensitivity of peak Hogi phosphorylation level
0.08
0.06
0.04
0.02
AMM-
Sin1
* Sin1
Ypd1
U
Ypdl
Ssk1
Ski
Ssk2
Pbs2
Ssk2 mPbs2
Figure 2.4.2 Local logarithmic gradients calculated for peak Hog1
Phosphorylation surfaces for sensitivity analysis of each intermediate
pathway protein. As expected from the complementation tests, the
MAPK proteins Ssk2 and Pbs2 show the highest curvature, implying
the greatest sensitivities.
Sensitivity of initial Hogi phosphorylation rate
0.60
0.45
0.30
0.15
0.00
SIn1
Ypd1.
Ssk1
Ssk2
Pbs2
Figure 2.4.3. Local logarithmic gradients calculated for initial Hog1
Phosphorylation rate surfaces for sensitivity analysis of each intermediate
pathway protein. Just as for Figure 2.3.2, the complementation studies
qualitatively match the trend of the results of the computational analysis
here with the MAPK pathway proteins Pbs2 and Ssk2 displaying much
higher curvatures, i.e., higher sensitivities, than the phosphorelay proteins
Slnl, Ypdl, and Sskl.
To overcome the uncertainty in the wild-type parameters used in Equation 2.1, we
formulated a 2 "dmetric that is less dependent on the choice of the particular wild-type
parameters. This method utilizes the full distribution of$, instead of the only region
around the wild-type level, and measures the relative spread of this distribution to
determine the effects of variations in rate constants on $. Using the same model outputs,
we computed the following modified deviation metric:
-
((k) - #widtype)2
[2.2]
where V is the phase space volume over which the parameters are swept. Similar to the
local logarithmic gradient, large values of the modified standard deviation indicate
greater sensitivity to parameter variations, while smaller values indicate greater
robustness to parameter variations. The results of this analysis are shown in Figure 2.4.4.
In summary, both analyses highlighted above yielded the same qualitative answer i.e.
HOG signaling is most affected by changes in the rate constants of the downstream
MAPK proteins and least by the upstream phosphorelay proteins.
Figures 2.4.2 and 2.4.3 summarize the sensitivities of MHogJ and rHgi respectively for all
pathway genes upstream of HOG1. Consistent with the complementation results, both
MHogl
and rHogi are most sensitive to kinetic rate constant changes involving the MAPK
cascade genes, and are least affected by variations in the phosphorelay components. This
theoretical prediction is qualitatively reproduced in multiple analyses that utilize different
sensitivity measures (Figure 2.4.4).
.
......
...............
Sensitivity of steady state Hog1 phoshorylation
Sensitivity of initial Hogi phoshorylation rate
0.7level
0.6
2.0
0.5
1.6
0.4
1.2
0.3
0
0
Sini
Ypd1
Sski
Sk2
0.40
Pbs2
SinI
Ypd1
Seki
Sk2
Pbs2
Figure 2.4.4 Calculating sensitivity metrics from surface plots using modified
standard deviation metric shown in Equation 2.2.
2.4.2 Relating mutational robustness to local biochemistry
One possible mechanism that could explain the pattern of mutational robustness we
observe experimentally is that the biochemical circuitry of the phosphorelay network
renders the terminal phosphorelay protein insensitive to changes in kinetic parameters of
its upstream pathway components. To mathematically determine the contribution of this
effect, consider a chain of signaling proteins where the steady state phosphorylation level
of any cascade protein consists of a basal phosphorylation level independent of pathway
activity, and an additional
component that is inducible by the steady state
phosphorylation level of its immediate upstream activator:
x =x 2+ - (
Here, x1 and
x2
[2.3]
x)
8x1
are the basal phosphorylation levels of the 1 "t and
2 nd
proteins in the
cascade, and primed symbols represent the total protein phosphorylation levels, while the
partial derivative denotes the amount of phosphorylated
2 nd
proteins derived from every
phosphorylated l' protein. Extending these equations for the 3 rd protein in the cascade
yields:
x' =x+
X3
3 (x -x 2 )
x 2
2
2
[2.4]
Substituting [2.3] into [2.4] we obtain:
3x 2
x2 (xx+ x_)
d
8x1
[2.5]
Extending the analysis for the jth protein in the cascade, we obtain:
x'= x
(x,- x 1 )
i=2
[2.6]
x_
From [2.6] it is clear that the biochemical details of signal transmission are buried
mathematically in the chain of derivatives i.e. they represent how the activity of the
cascade protein furthest upstream is transduced into changing the activity of the jth
cascade protein. For example, the contribution of the kth protein to the chain of
derivatives arises from two factors i.e. the effect of the (k- 1 )th protein on the activation of
the kth protein and the effect of the k protein on the activation of the (k+1)* protein:
axk+ axk_
a9Xk a9Xk-
where we term
k
= Xk+l =
k
[2.7]
ak-1
the steady state throughput of the kth protein.
The central claim of the throughput analysis is that the sensitivity of
k
to changes in
parameters describing the k* protein can predict to what extent sequence changes in the
kth protein will be tolerated by the system. An important corollary to this claim is that if
k
is invariant under parameter variations, then sequence changes in the kth protein will
not affect signaling unless the sequence changes completely inactivate the protein
altogether. To put this analysis into effect, we used a simplified model of the HOG
pathway [Klipp 2005]:
d[Sln1P] = k,( Ut) 2 [Sln1] + k-2[YpdlP][Slnl] - k 2 [Ypdl][SIn1P]
dt
O(t)
d[YpdlP] = k 2 [Ypdl][Sln1P]
k2[YpdlP][Slnl] - k3 [YpdlP][Sskl]
-
dt
d[SsklP]
-
k [Ypd1P][Sskl]
k[pl[sk
-
k_[SskP]
k[sl
dt
d[Ssk2P]
=k 4 [Ssk2][Sskl] - k 4 [Ssk2P]
d
dt
d[Pbs2P] = k5[Pbs2][Ssk2P] - k_,[Pbs2P]
dt
d[Hog1P] = k6[Hogl][Pbs2P] - k6[Hog1P]
dt
[2.8]
[.
Since the signaling dynamics are fast relative to the osmotic pressure variable, separation
of timescales allows one to treat the signaling system as if it were in steady state at every
moment in time (the signaling pathway adiabatically follows the osmotic pressure
dynamics, readjusting itself to the osmotic pressure variable at every point in time). To
determine the effect of local biochemistry on gk, we examined the two most
architecturally distinct proteins i.e. Pbs2 and Ypdl whereby Pbs2 is a kinase sandwiched
between similar kinase proteins, while Ypdl is a phosphotransfer protein sandwiched
between similar phosphotransfer proteins.
ad[HoglP]
k5 k_,kok-6 Pbs2T Hog1T
b Ssk2P] 2[k5k6[Ssk2P] Pbs2T+ k-6(k5-Ssk2P]+k
From this expression, we observe that
4Pbs2
[2.9]
5)
2
'
depends on Pbs2 interaction parameters i.e.
phosphorylation rate of Pbs2 and Hogl etc. Changes in Pbs2 sequence can alter these
rates and affect the steady state throughput, and can impact Hog1 phosphorylation levels.
On the other hand, the throughput of Ypdl is:
k,( H(t)
d[Sskl]
Ypd
[Slnl]
_
0
(t)
2
[2.10]
............... III
Remarkably,
4Ypd1
.IN!
is independent of Ypdl parameters. This implies that, as a direct
consequence of the local architecture of the network of biochemical reactions, Hogl
phosphorylation is shielded from potential changes in Ypdl rate constants.
2.5 Experimental evolution design
To test our theoretical result, we devised an experimental evolution strategy where we
harnessed naturally occurring genetic variation, and imposed a strong selective pressure
on HOG signaling. Then, we compared the adaptive genetic variants found in the
evolution experiments with our theoretical predictions.
doxycycline
ab
MM O2
rtTA
EE
E
HogPathwaymadce
I
WTgrawth
Consftbte Hog
I
PTETO7
P
YFP
adon
NRD1
.
HOG1
!Wow
goth
Figure 2.5.1 Artificially hyperactivating HOG pathway by
underexpressing YPD1. a) In media containing no doxycycline, Ypdl
is greatly underexpressed, leading to constitutive activation of the
pathway and thus a greatly reduced growth rate. b) By placing the
endogenous Ypdl gene under the transcriptional control of the Tet
promoter, we could tune its expression level.
To design our selection experiment, we took advantage of the knowledge that deletion of
the YPD1 gene in the HOG pathway leads to hyperactivation of the pathway and
subsequent cell lethality [Posas 1996]; the logic of the experimental intervention is shown
in Figure 2.5. lA. Thus, shown in Figure 2.5. lB, we placed YPD1 under the control of a
TetO7 promoter where we could induce its expression with doxycycline and control the
I
. .......
....................................
degree of activation of the pathway. In addition, a cyan fluorescent protein was placed
under a second TetO7 promoter to serve as an indirect readout for YPD1 expression.
From growth rate measurements at different doxycycline concentrations, we found that
the cells suffered a severe growth defect at low doxcycline concentrations, where YPD1
expression was repressed (Figure 2.5.2). To determine if the HOG pathway was activated
under YPD1 underexpression, we imaged the cells under the microscope to measure their
Hogl-YFP and Nrdl-RFP signals. We found that Hogi was predominantly localized in
the nucleus, therefore confirming that the pathway was indeed hyperactivated under
YPD1 underexpression (Figure 2.5.3). In contrast, Hog1 was uniformly distributed
throughout the cytoplasm in cells with high YPD1 expression and in wild-type cells.
0.4-
0.3-
0.2-
0.1
I
-4
-3
-2
-1
0
log 10(doxycycline) [pg/mL]
Figure 2.5.2 Growth rate of ancestor strain as a function of doxycycline present
in the media. The experiment shows that as long as the the doxycycline
concentration remains below 0.01 pg/mL, the strain exhibits a severe growth rate
deficit of roughly 70%.
Because Hogl activation induces the expression of GPD1 and GPP2, which encode
proteins responsible
for glycerol synthesis
[Albertyn
1994],
we assessed the
transcriptional readout of the signaling activity by measuring intracellular glycerol. We
found that cells underexpressing YPD1 had at least two-fold higher intracellular glycerol
concentration than cells with high YPD1 expression (Figure 2.5.4), which was consistent
with our observation that the pathway was hyperactivated under YPD1 underexpression.
Ancestor with doxycycline
Nrdl-RFP
Ancestor without doxycycline
Hog1-YFP
Nrdl-RFP
Hogi-YFP
Figure 2.5.3 HogI nuclear enrichment in ancestral strain with and without
doxycycline. Fluorescence microscopy confirms that, consistent with the idea
that underexpression of Ypdl leads to constitutive pathway activation, in
our strains grown in the absence of doxycycline there is a notable
basal enrichment of nuclear Hog 1.
Ancestor with doxycycline
MPAncestor without doxycycline
0
0.01
002
0.03
0.04
005
006
[glyceroissyjcell
(OD54OD )
Figure 2.5.4 Glycerol production in ancestor strain with and without doxycycline.
In the "ancestor" strain (which contains the Tet07-Ypdl construct), intracellular
glycerol levels show a nearly 3-fold increase when doxycycline is withheld from
the media. This is consistent with the idea that withholding doxycycline
hyperactivates the HOG pathway, which is known to upregulate glycerol
production in the cell upon activation.
Finally, as shown in Figure 2.5.5, by measuring Hogi nuclear enrichment at different
doxycycline levels, we further established that the growth rate was inversely correlated
with Hogi nuclear accumulation.
1.6C
0
E
C
Z
1.5
Z
1.3
-4
-2
-1
-3
logl [doxcycline] Ipg/mI
0
1
Figure 2.5.5 Hogl nuclear enrichment as a function of doxycycline in the
"ancestor" strain.
2.6 Rapid adaptive evolution of yeast cells underexpressing YPD1
We evolved nine independent lines of the yeast strain underexpressing YPDJ each with a
population size on the order of 107 cells, and monitored their mean population growth
rates using turbidostats [Acar 2008]. Rapid adaptation occurred after merely five days,
shown in Figure 2.6.1, and qualitatively similar adaptation dynamics were observed in
the nine experiments. The dynamics revealed three distinct regimes. During the first 14
hours (phase I), we observed a transient decrease in the growth rate as a consequence of
the dilution of the Ypd1 proteins due to cell division and degradation. Phase II exhibited
the lowest growth rate (-0.20 ± 0.05 hr-), and this quasi-steady state lasted for about 26
hours. The growth rate rapidly recovered within the next 36 hours (phase III) before
eventually reaching a steady-state level similar to that of the ancestor under the
unstressed condition. At the end of the evolution experiments, a small aliquote of the
turbidostat culture was plated on selective plates and five randomly selected single
colonies were isolated from each of the nine adapted populations for further analyses.
Importantly, the evolved populations maintained their growth advantage when placed
under the same selective pressure in media without doxcycline (transferred from media
without doxycycline to with doxcycline and subsequently to without doxycycline again),
indicating that their phenotypic changes were stable (Figure 2.6.2).
0.1'
0
1
2
3
4
Time after doxycycline removal [Day]
5
Figure 2.6.1 Time course of evolutionary dynamics.
2.6.1 Restoration of basal wild-type Hog1 activity
To determine if the hyperactivation of the HOG pathway had been resolved, we measured
the evolved strains' Hogi nuclear enrichment and intracellular glycerol in two randomly
selected colonies out of five from each of the nine adapted populations. In 17 out of 18
evolved strains, both Hog1 nuclear enrichment and intracellular glycerol content had
restored to levels comparable with the ancestor in the unstressed condition (Figures 2.6.3
and 2.6.4). Thus, we established that the hyperactivation of the HOG pathway had been
alleviated in almost all evolved strains.
.
k
* EvlIved straios
- 0oxycycIn4
a
a
-- 4
AtIcestor +
doxcycline
A cestor dcxcycIIne
-4
0
0.1
0.2
0.4
0.3
0.5
0.6
0.7
Growth rate (hr4)
Figure 2.6.2 Restoration of growth rate levels to ancestral state. As can be seen
by comparing the pink bars to the green bar and yellow bar, the evolved cells
match the growth rate in the ancestral strain in the presence of doxycycline,
well above the rate in the absence of doxycycline.
* Evolved strains without doxycycline
Ancestor with doxcycline
* Ancestor without doxcycline
-ammes--I
I
-
Imm
INUMNEF---i
0.01
0.02
0.03
0.04
0.05
[glyceroljt,j/cell
(OD/ODa00 )
Figure 2.6.3 Restoration of glycerol levels to ancestral state. Consistent with
the hypothesis that growth rate restoration followed downregulation of HOG
pathway activity, we see a drop in amount of intracellucular glycerol.
To further assess if the evolved strains were capable of mediating hyperosmotic shock
recovery, we simultaneously measured both the dynamics of their Hogi nuclear
enrichment and cellular volume, a proxy for turgor pressure in response to hyperosmotic
shock after adding 0.6 M NaCl. Interestingly, for a majority of the evolved strains, the
amplitude and dynamic changes in Hogl nuclear accumulation, shown in Figure 2.6.4,
and cellular volume, shown in Figure 2.6.5, were similar to those of a wild-type response.
0
10
20
30
40
Time [minutes]
50
60
Figure 2.6.4 Restoration of pathway dynamics to near ancestral behavior. With
the exception of a single trace from an evolved strain, all evolved strains (data
shown in light pink) exhibit step response trajectories that are very similar
to the ancestral trajectory in the presence of doxycycline.
In contrast, 4 of the evolved strains displayed drastically different dynamical Hogi
signaling behaviors. The amplitude and activation rate of Hogl in these strains were
significantly reduced compared to wild-type (Figure 2.6.4). Despite these gross
differences in signaling dynamics, however, we still observed restoration of turgor
pressure and the rate of volume recovery was not significantly different from wild-type
(Figure 2.6.5). These data together indicated that both signal transduction and Hogl-
mediated transcriptional regulation of glycerol-producing factors responsible for turgor
pressure recovery were at least partially functional in most of the evolved strains.
0.
L
t
E
>.
-0.2-
-0.4
0
1'0
20
30
40
50
60
Time [minutes]
Figure 2.6.5 Restoration of volume recovery dynamics for the majority
of the evolved strains, further bolstering the data from Figure 2.6.4.
2.6.2 Transcritpional regulation of YPD1 is not upregulated in the evolved
strains
Mutations in the synthetic transcriptional activator rtTA have been found to be
responsible for the rapid adaptation of a synthetic gene circuit (Pando and van
Oudenaarden, unpublished results). To determine if YPD1 expression had increased in
the evolved strains, we measured CFP levels which served as a proxy for Ypd1 protein
levels. All eighteen evolved strains showed low CFP intensities similar to that of the
ancestor under no doxycycline conditions (Figure 2.6.6), thus indicating that the adaptive
molecular changes most likely did not occur in the rtTA gene.
0
100
200
300
400
Mean CFP intensity (-Ypdl)
(Au)
Figure 2.6.6 CFP data suggests that the evolved strains did not alter the
properties of rtTA in order to effect their recovery from pathway
hyperactivation.
2.6.3 PBS2 and SSK2 are preferentially mutated in independent evolution
experiments
To identify the candidate molecular changes that led to the evolved phenotypes, we
performed a genome-wide screen for point mutations in the evolved strains. Full genomic
sequences of the ancestral strain and five of the evolved strains were obtained using the
Illumina sequencing platform. Data on the quality of the sequencing results, such as the
number of reads and the genome coverage, can be found in Table 2.6.1. A total of 517
single-nucleotide polymorphisms (SNPs) were found in the ancestral strain compared
with the published Saccharomyces cerevisiae genome. 508 of these SNPs (-98%)
appeared in all 5 evolved strains and were thus likely to be present in the ancestral strain
(data not shown). At most three statistically significant mutations were identified per
evolved strain (p-value < 0.05, )? test), which are highlighted in Table 2.6.2. We found
that 5 of the confirmed mutations mapped to the coding sequence of the genes in the
MAPK cascade, i.e. PBS2 and SSK2.
Total number of
filtered reads 106 (a)
8.5
% of filtered reads aligning
to the genome (b)
95.6
Mean genome
coverage (c)
24
% of genome coverage
(< 5 reads)
6
9
91.2
E2
8.4
90.64
27
22
4
3
E3
8.2
85.2
20
8
E4
7.7
E5
8.9
79.1
86
21
27
5
6
Strain
Ancestor
El
(a) Reads that passed the Illumina ELAND pipeline filter.
(b) A maximum of two mismatches per read was set in the alignment process using the MAQ
software.
(c) The average number of reads aligned to each nucleotide position in the genome.
Table 2.6.1 Summary of sequencing depth and coverage for both the ancestral
strain and 5 evolved strains sent for Illumina sequencing.
Whole-genome
sequenced evolved strain
Chro
Genome
position;
Ancestral
nucleotide
Evolved
nucleotide
ORF
Impact (a)
2
533725;1
T
C
-
-
4
247140; -1
C
A
A158E
10
179973; -1
T
G
YDL119
C
PBS2
E3
10
3
12
178832; -1
49444; -1
162383; -1
G
G
A
E4
E5
14
14
681259; -1
681040; -1
C
C
A
A
T
T
T
PBS2
PDIl
SSK1
SSK2
SSK2
G423D
G260E
1504F
P1393L
P1466L
El
E2
Y43D
(a) Impact represents either synonymous or non-synonymous mutation. Notation A158E indicates that
amino acid A at gene position 158 in the ancestral strain has been changed to E in the
corresponding evolved strain.
Table 2.6.2 Single nucleotide polymorphisms detected from whole genome
sequencing of evolved strains. In virtually all cases, the detected SNP was found
to affect a HOG pathway gene.
.....................
To determine how prevalent mutations in HOG pathway genes were in the evolved
strains, we sequenced all six genes in the pathway including their promoter regions for
the remaining 40
strains. Unexpectedly, all of the 45 strains except 5 contained a single point mutation in
one of the genes in the pathway (Figure 2.6.7).
5
a No. of evolved strains
with mutations in the
Hog network
No. of evolved strains
with no mutations in
the Hog network
Figure 2.6.7 Mutations in evolved strains are predominantly in HOG pathway
genes. Remarkably, 40 out of 45 evolved strains contained mutations within
the HOG pathway proteins themselves.
Evolved strains isolated from the same population harbored different genetic changes,
thereby confirming that the populations at the end of the evolution experiments were not
isogenic (Figure 2.6.8). We found that, among the pathway genes, PBS2 and SSK2
mutations were the most highly represented across a majority of the nine independently
adapted populations, which were largely consistent with the computational robustness
analysis. In addition, we observed two incidences in which identical PBS2 mutations
were found in independent experiments. To account for the different mutational target
sizes across pathway genes, we normalized for every gene, the number of unique
mutations against gene length. The trend was robustly reproduced (Figures 2.6.9) with
the MAPK cascade genes PBS2 and SSK2 being the most frequently mutated.
.........................
. ...
................
......
...........
................................
--.........................
..
- Number inside bar indicates fraction of unique gene mutations
-* and " represent identical mutations inindependent experiments
&,
C
.C
u
1
0.8
1/2
Unknown
SLN1
SSKI
YPD1
SSK2
HOGI
*PBS2
2/3
0.6
C0.4
C
0
2/2
0.2
C
0
2
3
4
5
6
7
8
9
Evolution experiment
Figure 2.6.8 Distribution of genetic changes in evolved strains corresponding to
(A) across the nine experiments. Indicated inside bars are the fractions of unique
gene mutations observed in individual experiments. Notations * and * * represent
two particular mutations which were found in independent experiments.
0.005
0.004
0.003
0.002
M~ 0.001
Th
A
SLN1
YPD1
SSK1
SSK2
PBS2
HOG1
Figure 2.6.9 Characterizing the spectrum of mutations according to target identity.
Remarkably we find that the MAPK proteins Ssk2 and Pbs2 are the most readily
targeted proteins in the pathway, precisely as expected from the systematic
complementation and computational sensitivity analyses.
We identified a total of 25 unique mutations and all except one were missense mutations,
and more than half of them were in the protein kinase domains, which are highly
conserved. The mutations are catalogued in Table 2.6.3. When examined more closely,
the mutations typically occur in protein domains relevant to signaling. For example, 4 of
the 7 mutations in the SSK2 gene occur in the protein kinase domain, while 2 of the
remaining mutations take place near or in the domain that binds the upstream activator
Sskl. Taken together, the above findings revealed that PBS2 and SSK2 were
preferentially and repeatedly mutated in independent evolution experiments, and
suggested that the observed phenotypic changes most likely arose from genetic changes
within the HOG pathway.
A
SSKI
SSK2
PBS2
B
1504F
Close to response regulator receiver domain
V402A
Ssk1 binding domain, essential for Ssk2 activation
W427C
Essential for Ssk2 activation
C1172Y
Unknown
P1393L
Kinase domain
P1466L
Kinase domain
G1471V
Kinase domain
W1557C
Kinase domain
Y43D
Docking site for Ssk2
R61L
Docking site for Ssk2
G423D
Kinase domain
G509S
Kinase domain
M526R
Kinase domain
R640 to STOP
NLS?
0.5
A =Evolved strain
B =Ancestral strain with endogenous gene replaced with mutant allele
0.2
Table 2.6.3 Cataloguing the mutations in molecular detail, including the domains
the mutations occur in.
2.6.4 Mutations in PBS2 and SSK2 are mainly responsible for the down-
regulation of the hyperactive signaling and improved fitness
To test whether PBS2 and SSK2 mutations account for the adaptive phenotype, the
endogenous gene in the ancestral strain was replaced with 13 of the unique mutant alleles
("transformed strains"), and their growth dynamics were compared to those of the
ancestor, and either a PBS2A or a SSK2A strain. These 13 mutant alleles were randomly
selected to broadly represent mutations across various protein domains. Unlike the
ancestral allele, almost all the mutations conferred a significant growth advantage when
the cells were transferred from media with doxycycline to without doxycycline (Table
2.6.3). The growth increase conferred by the single mutations matched the fitness
advantage of most of the evolved strains, confirming that PBS2 and SSK2 mutations were
primarily responsible for the improved fitness.
For a majority of the transformed strains, the growth rates were similar to that of their
respective gene deletion strain i.e. PBS2A or SSK2A under no doxycycline conditions i.e.
(0.38 ± 0.07) hr' and (0.42 ± 0.05) hr'. Since HOG signaling was not completely
abolished in the evolved strains, these data further supported that the mutations cause a
partial loss-of-function of PBS2 and SSK2, thereby mitigating signaling hyperactivation.
2.6 Discussion
Studies on the robustness of cellular phenotypes to gene expression changes have been
greatly facilitated by experimental techniques which allow quantitative manipulation of
gene expression [Batchelor 2003] and [Moriya 2006]. In contrast, there is no simple
experimental strategy to comprehensively assess the robustness of network function to
coding sequence variation of its component genes. For instance, to explore amino acid
substitutions at only a few positions of a protein would involve generating thousands of
variant proteins, and measuring their effects on system outputs. Alternatively, a
comparative method utilizes existing natural genetic variation, and infers genetic
robustness by comparing the structure and function of cellular networks across closely
related species [Tanay 2005]. But this approach is limited by the inexact knowledge of
the environments to which the organisms adapted.
Here, we show how a quantitative understanding of genetic robustness can be achieved
using a combination of theoretical and experimental approaches. In particular, our work
demonstrates the feasibility and promise of applying experimental evolution to the study
of genetic robustness. By manipulating either the environmental conditions or the
genotype of the organisms, specific hypotheses can be tested based on the evolutionary
outcomes. Through a computational analysis, we find that signaling is most affected by
kinetic parameter changes in the MAPK cascade genes, and yet, it is highly robust to
changes in the upstream phosphorelay components.
To test this, we induce hyperactive signaling in yeast cells, and harness the combined
forces of evolution and natural selection to sieve out adaptive genetic variants that can
significantly affect signaling, and restore fitness. The model predicts that mutations in the
upstream phosphorelay genes have a minimal effect on network behavior, thereby
suggesting that genetic changes in these genes would be effectively neutral. The
theoretical results are largely consistent with the evolutionary outcomes, where none of
the evolved strains out of a total of forty from nine independent experiments had
mutations in the phosphorelay genes SLN and YPD1, and instead, almost all the
mutations were in either PBS2 or SSK2.
As with many studies attempting to predict the outcome of evolutionary experiments, due
to the inherently contingent nature of changing allele frequencies via natural selection,
some important caveats apply in assessing these results. Although our results have shown
that the HOG pathway is less sensitive to kinetic rate changes in the phosphorelay system
than the MAPK system, and although they have also shown that the MAPK system is
preferentially targeted by natural selection to downregulate the activity of the pathway
when a population of budding yeast cells is challenged to do so, it is possible that these
are unrelated. The MAPK system, being positive regulators of Hog1 translocation to the
nucleus, needs to suffer loss-of-function mutations to downregulate pathway activity,
which intuitively are thought to be easier to obtain than the gain-of-function mutations
that the phosphorelay system proteins would need to similarly downregulate the pathway.
It should be noted, however, that there is a loss-of-function path for the phosphorelay
system to downregulate Hogl translocation: Sln1 could fail to accept phosphate groups
from Ypdl, which would increase the phosphorylated form of Sskl and lower pathway
activity - this is a mutation we do not observe.
Furthermore, mutations in HOG1 can significantly affect signaling, however, these were
rarely found. Because Hogi genetically and physically interacts with a large number of
genes and proteins, it is quite likely that the pleiotropic costs outweigh the benefits in
signaling changes, thereby rendering these "solutions" much less probable. Notably, we
observe that rapid restoration of signaling and fitness can be achieved solely via a singlenucleotide mutation. And interestingly, all the mutations were found in the protein coding
regions. [mention that this might be because of the large selective pressure in this system]
Because cis-regulatory regions tend to exhibit greater plasticity than coding sequences
[Borneman 2007] and [Odom 2005], it appears more likely that a single point mutation
would affect protein function than drastically alter gene expression.
Finally, our results show how a relatively simple eukaryotic signal transduction pathway
could have evolved its biochemical circuitry to allow for genetic robustness and
evolvability. One can rationalize the theoretical result by recognizing that the MAPK
cascade and the phosphorelay chain operate via distinctively different modes of signaling.
The MAPK cascade
utilizes
a catalytic
signaling mechanism,
that is every
phosphorylated molecule goes on to phosphorylate multiple downstream molecules, and
thus the cascade output (amount of phosphorylated Hogl) is ultimately dependent on all
the MAPK cascade component parameters. By contrast, the phosphorelay chain operates
via stochiometric signaling where the output of the chain is determined entirely by the
influx and efflux of phosphate groups [Shinar 2007], rather than the rate constants of
components within the phosphotransfer relay. Thus, changing those parameters should
not affect signaling, consistent with what we have observed experimentally. Our results
suggest that the nature of biochemical interactions within a network can significantly
shape the space of targets that natural selection can act upon.
2.7 Methods
Strain background and construction
Our haploid ancestor strain (DMY028) was derived from the DMY017 strain [Muzzey
2009], the only difference being that it contained a plasmid bearing two TetO7
promoters, one of which drives the expression of CFP, while the other controls YPD1
expression. The mutant strains referred to in this study were similarly derived from the
DMY017 strain, except that the endogenous genes in the Sln1 branch of the HOG
pathway were singly knocked out and replaced with its corresponding orthologs from
various yeast species. Firstly, the endogenous genes were singly knocked out and
replaced with the Candida albicans URA3 gene using the pAG60 plasmid (Euroscarf).
SLN1 and YPD1 gene deletions are lethal due to the hyperactivation of the pathway. To
circumvent this, we knocked out these genes using a cassette containing both the C.
albicans URA3 gene and the Hog1 phosphatase PTP2 placed under the control of the
ADH1 promoter. The orthologous genes from various yeast species were stitched to the
500-bp S. cerevisiae upstream and downstream gene flanking sequences using overlap
extension PCR. These final constructs were then transformed into the endogenous gene
knockout strains described earlier, and single colonies were selected for the absence of
URA3 expression on 5-FOA plates. All integrations were subsequently confirmed by
sequencing.
Growth and media conditions
Unless otherwise stated, all experiments were performed on exponentially growing cell
cultures in synthetic dropout media with the appropriate amino acid supplements at 30
'C. The ancestral and evolved strains were grown consistently in 0.4 M NaCl for all
experiments, except when their signaling abilities were analyzed upon a hyperosmotic
shock of 1 M NaCl. In addition, all experiments involving the evolved strains were
performed in the absence of doxycycline. Prior to the evolution experiment, the ancestral
strain was grown overnight with doxycycline and the culture media was replaced with
media without doxycycline before propagating them in the turbidostat [Acar 2008]. In
experiments where cells were treated with doxycycline, a 5 pg/ml concentration was
used.
Glycerol assays
Intracellular glycerol levels were measured using the Free Glycerol Reagent Kit (Sigma)
as described [Muzzey 2009]. For details regarding the method and cell preparations, see
the Supplemental Data.
Fluorescence microscopy and image analysis
Cell preparation and immobilization, and image acquisition and segmentation were
performed as described [Mettetal 2008]. For our signaling experiments involving mutant
strains with the orthologous pathway proteins, we corrected for any possible effects from
outside the HOG pathway by measuring signaling in the respective pathway gene
knockout strains in response to the same hyperosmotic shock ("basal signal"), and we
subtracted this basal signal from that of the mutant strain's mean Hogl trace. In addition,
the reported Hogl nuclear enrichment here represents the measured signal subtracted by
the nuclear enrichment level prior to hyperosmotic shock.
Whole-genome sequencing
Genomic DNA (gDNA) was extracted from 10 ml of stationary phase cultures using a
standard protocol with a few modifications noted below [Hoffman 1987]. Three
consecutive phenol/chloroform/isoamyl-alcohol extractions were performed to reduce
protein contamination. RNA contamination was reduced by treating the gDNA samples
with 70 ng/pl affinity-purified RNAse A (Ambion) for 1 hour at 37 "C. A final
chloroform extraction was performed to remove phenol contamination prior to ethanol
precipitation. The final gDNA yield was quantified using a ND-1000 spectrophotometer
(NanoDrop). Genomic libraries for whole-genome sequencing were prepared as directed
(Illumina). Image analysis, base calling and sequence alignment were performed
according to the Illumina Genome Analyzer pipeline. The Illumina pipeline tool "eland"
was used to uniquely align 36 bp reads to the S288c yeast reference genome [Cherry
1998] allowing a maximum of 2 mis-matches per read.
Subsequent
data
analysis
was
carried
out
using
the
MAQ
software
(http://maq.sourceforge.net/), where the filtered reads were mapped to the reference
genome and assembled to create the consensus genomic sequence, and to detect SNPs.
The same procedure was carried out for the synthetic constructs i.e. PMro2 -rtTA and
PTeto 7 YPD1.
The MAQ-generated SNPs were further filtered upon analyzing criteria
such as average chromosomal depth and repetitiveness using custom-written scripts.
Finally, we determined the statistical significance of the SNPs by performing a x test on
the distributions of nucleotide bases of the reads obtained at each SNP position for the
ancestral and the evolved strain. A randomization test for goodness-of-fit was carried out
in cases where there were fewer than five reads.
We obtained between seven to nine million Illumina-pipeline filtered reads for each
sequencing attempt, 80-95 % of which aligned to the S288c reference genome. Within
the non-repetitive genomic regions, an overall mean depth of around 20 reads per
nucleotide position was obtained, with less than 10 % of nucleotide positions having
fewer than five reads (Table 2.6.1).
85
86
Chapter 3
Robust yet tunable regulatory elements:
the case of microRNA
3.1 Summary
MicroRNAs (miRNAs) are short, highly conserved non-coding RNA molecules that
repress gene expression in a sequence-dependent manner. Each miRNA is predicted to
target hundreds of genes [Lewis 2005, Selbach 2008, Baek 2008, Friedman 2009], and a
majority of protein-coding genes are predicted to be miRNA targets [Friedman 2009,
John 2004]. Bulk measurements on populations of cells have indicated that, although
pervasive, repression due to miRNAs is on average quite modest (-2-fold) [Selbach
2008, Baek 2008, Bartel 2004]. Information on the magnitude of repression in single
cells, however, has been lacking. Here we perform single-cell measurements using
quantitative fluorescence microscopy and flow cytometry to monitor a target gene's
protein expression in the presence and absence of regulation by miRNA. We find that
while the average level of repression is modest and in agreement with previous
population-based
measurements,
the
repression
among
individual
cells varies
dramatically. In particular, we show that regulation by miRNAs establishes a threshold
level of target mRNA below which protein production is highly repressed. Beyond this
threshold, there is a regime in which expression responds ultrasensitively to target mRNA
input until reaching high enough mRNA levels to almost escape repression by miRNA.
We constructed a mathematical model describing repression of target gene expression by
both non-catalytic and catalytic activity of miRNA. The model predicted, and
experiments confirmed, that the ultrasensitive regime could be shifted to higher target
mRNA levels by transfecting additional miRNA or by increasing the number of miRNA
binding sites in the 3' UTR of the target mRNA. The ultrasensitive transition is not
observed when the miRNA targets a perfect complementary site that can undergo
catalytic cleavage. These results demonstrate that even a single species of miRNA can act
as a switch to effectively silence gene expression and as a fine-tuner of gene expression.
3.2 microRNA background
MicroRNAs regulate protein synthesis in the cell cytoplasm by promoting target
mRNAs' degradation or inhibiting their translation. Their importance is suggested by
their abundance, with some miRNAs expressed as high as 50,000 copies per cell [Lim
2003]; by their sequence conservation, with some miRNAs conserved from sea urchins to
humans [Grimson 2008]; and by their number of targets, the majority of protein-coding
genes [John 2004]. miRNAs can regulate a large variety of cellular processes, from
differentiation and proliferation to apoptosis [Chen 2004, Yi 2008, Sluijter 2010, Esau
2004, Le 2009, Cimmino 2005, Song 2009, Sood 2006, Li 2005, Makeyev 2007,
Bernstein 2003]. Further, miRNAs also confer robustness to systems by stabilizing gene
expression during stress and in developmental transitions [Li 2009, Li 2006].
3.3 Two-color assay to measure regulation via microRNA
Despite the evidence for the importance of gene regulation by miRNAs, the typical
magnitude of observed repression by miRNAs is relatively small [Friedman 2009], with
some notable exceptions such as the switch-like transitions due to miRNAs lin-4 and let7 targeting the heterochronic genes lin-14 and lin-41 respectively in Caenorhabditis
elegans [Bagga 2005]. Importantly however, most of the previous studies of regulation
by miRNAs in mammalian cells have measured population averages which often obscure
how individual cells respond to signals [Raj 2008].To assay for miRNA activity in single
..
.......
..............
.
mammalian cells, we cloned a two-color fluorescent reporter construct that permits
simultaneous monitoring of protein levels in the presence and absence of regulation by
miRNA, depicted in Figure 3.3.1. The construct consists of a bidirectional Tet-inducible
promoter driving two genes expressing the fluorescent proteins mCherry and eYFP
tagged with nuclear localization sequences. The 3' UTR of mCherry is engineered to
contain N binding sites for miRNA regulation. In the initial experiments, the inserted
sites are recognized by miR-20, which is expressed endogenously in Hela cells along
with its seed family members miR-17-5p and miR-106b. The 3' UTR of eYFP is left
unchanged so that it can serve as a reporter of the transcriptional activity in a single cell.
pTRE-Tight
I NLS-eYFP
3'-UTR
T
N miR-20
binding site(s)
,
(TACCTGCACTCGCGCACTTTA)N
3'-UTR
Figure 3.3.1 A synthetic two-color reporter construct for measuring miRNA
mediated gene regulation in single cells. The construct consists of a
bidirectional tetracycline-responsive promoter that drives the transcription of
two fluorescent reporter proteins: eYFP and mCherry. Nuclear localization
signals (NLS) are fused to the fluorescent reporters to facilitate image
processing. We fuse N miR-20 binding sites to the 3'-UTR of mCherry to
measure the effects of miRNA-mediated regulation.
3.3.1 Control experiments establishing eYFP as a transcriptional readout
In order to confirm that the bidirectional Tet promoter was indeed driving transcription of
both fluorescent reporters symmetrically, we performed a series of control measurements
with the N = 0 construct. First, as shown in Figure 3.3.2a, we performed quantitative RTPCR to measure the mRNA levels of mCherry and eYFP in bulk populations and
observed that the two reporter transcript abundances were roughly the same. However
this method does suffer from the drawback that it could in principle obscure the picture
for an individual cell: in principle, it could have been possible that although on average
the two reporters are expressed at the same level, any given cell either expresses eYFP or
mCherry but not both. In this case eYFP would not be a faithful reporter of mCherry
transcript levels - indeed it could be the opposite. Thus secondly, as shown in Figure
3.3.2b, we examined the raw joint mCherry-eYFP distributions for N= 0 to ensure that
the single cells clustered around a line of slope
a
=
1.
b
1
fluoemne data
400
C)
300
.
NO0.4I~~~i
P
200
-
~0.2
0
Ox eYFP
0
Ox mCherry
100
200
300
eYFP (arb. units)
400
Figure 3.3.2 Control experiments used to confirm idea that eYFP can act as a faithful
reporter of mCherry transcriptional activity in individual cells. a) RT-PCR signals
from N = 0 eYFP and mCherry samples are normalized to the N = 0 eYFP value,
showing that the mCherry transcript level is very similar to the eYFP transcript level.
b) The joint mCherry-eYFP single cell distribution agrees with the result of panel a for
the case of individual cells as well.
3.4 microRNA mediated repression generates gene expression
thresholds
We constructed cell lines that stably expressed the fluorescent reporter construct with
either a single bulged miR-20 binding site or no site in the mCherry 3' UTR. The levels
of
eYFP and mCherry protein were measured for single cells using quantitative
fluorescence microscopy. Arranging individual cells according to their eYFP expression
level, as shown in Figure 3.4.1, we observed that cells whose mCherry 3' UTR lacks
miRNA binding sites had a concomitant increase in mCherry expression. This indicates
I
. .................
that in the absence of miRNA targeting of the mCherry mRNA, the level of expression of
eYFP is directly related to the level of expression of mCherry.
eYFP
C|
mCherry
0||
1.3
0.8
1.2
1.2
1.7
2.0
1.1
1.0
1.9
ratio
eYFP
mCherry
0.0
0.0
0.0
0.1
0.1
0.1
0.6
1.0
0.9
ratio
Figure 3.4.1 Arranging single cells according to eYFP expression level
reveals gene expression thresholding by miRNA. Cell outlines are shown
in yellow. Below each representative single cell is the ratio of the mean
pixel intensity in the mCherry channel to the eYFP channel.
However, in cells with a miR-20 site in the mCherry 3' UTR, the eYFP fluorescence
initially increases with no corresponding increase in mCherry expression level, seen in
Figure 3.4.1 (lower panel). To capture this behavior quantitatively, we measured joint
distributions of mCherry and eYFP levels in single cells and binned the single cell data
according to their eYFP levels (see Figure 3.9.1 for a more detailed outline of the binning
procedure). Within each eYFP bin, we calculated the mean mCherry level; this process is
outlined in Figure 3.4.2. We refer to the binned joint distribution as the transfer function.
As suggested by the representative single cells shown in Figure 3.4.1, the transfer
function in Figure 3.4.3 shows a threshold-linear behavior in which the mCherry level,
which represents the target protein production, does not appreciably rise until the curve
reaches a threshold level of eYFP.
100
A N = 0
A
N
=A1
EN=
AAAA
25 Ao
0A
AAAA
AAA
A
&A&&&AA
A
AA A
A
0
AAAAA A
4"*A1AAzAA
0
25
50
75
100
eYFP (a.u.)
Figure 3.4.2 Transfer function relating eYFP to mCherry levels. As expected
from representative single cells shown in Figure 3.4. 1, the Tet promoter must
operate above a threshold transcriptional activity in order to escape from
robust silencing due to miRNA-mediated repression.
3.5 Generating thresholds without feedback
We developed a simple mathematical model of miRNA-mediated regulation that could
reproduce the nonlinearity in the above transfer function. This model, depicted in cartoon
form in Figure 3.5.1, is similar to previous models [Elf 2003] used to describe proteinprotein titration [Buchler 2008] and small RNA (sRNA) regulation in bacterial systems
[Levine 2007]. It describes the concentration of free target mRNA (r)
subject to
regulation by miRNA (m). We assume that only r can be translated into protein.
Experimentally, r corresponds to the mCherry signal, while runargetedcorresponds to the
eYFP signal. The core of the model involves the binding of r to m to form a mRNAmiRNA complex and the release of m from the complex back into the pool of active
miRNA molecules either with or without the accompanying destruction of r. We assume
that the total amount of miRNA is fixed; experimentally we observe no decrease in the
miR-20 level beyond experimental uncertainly as a function of eYFP (Figure 3.5.2).
3.5.1 Mathematical framework
In order to describe our data, we devised a simple mathematical model of the
biochemistry of miRNA-mediated gene regulation. The model is largely similar to
models of protein-protein interactions proposed by Buchler and Louis as well as models
of sRNA regulation of expression proposed by Levine et al. The model describes the time
evolution of the target mRNA free of miRNA (r) and the target mRNA bound by miRNA
(r*) and assumes that the turnover of miRNA is slow compared to the timescale of gene
expression so that it can be held constant. The model consists of the following set of
coupled, first-order, ordinary differential equations and the conservation relation for
miRNA:
dr
-=k
dt
dr*
-=
R
-konr
kr
[miRNA] + koffr
[miRNA]-kffr
Yr
YRr
*3.11
r
[3.2]
dt
[miRNAIT
=
[miRNA] + r*
[3.31
For the sake of simplicity, we assume that no translation can occur from the miRNAbound target mRNA such that for the purposes of protein production it is sufficient to
track only the free target mRNA (r). Solving for the steady-state level of r yields:
r=
[r.,ag,
where:
- A -6
+ 4untargeted ]
[3.4]
runtargeted = kR
YR
YR'
+ kof
ko
6=
0al
YR.[miRNA]
YR
Just as in the Buchler and Louis and Levine et al. cases, when the dissociation constant
(here denoted by X) is small - meaning that the interaction strength is high between the
miRNA and its target - it is possible to achieve a threshold-linear relationship between
the free target mRNA and the total amount of mRNA (denoted by
runtargeted,
which in the
experiments is reported by the eYFP signal). In our case, because we allow recycling of
the miRNA following destruction of its bound target mRNA, the titration effect only
becomes apparent when the rate at which free miRNAs are removed from the system
(kon) is much larger than the rate at which they reappear in the system, which itself
consists of two parts: unbinding of the miRNA from its target (kff) and destruction of the
target (YR*). In the most extreme case, for example, where ko n>> ko + YR* such that X -
0 one obtains:
-61(
[3.5]
1
2
untargeted
{0
1 untargeted
untargeted
if
runtargeted
< 6
1
*)2
[3.71
untargeted
In this limit, we see that the constant 0 sets the level of expression at which the threshold
takes place.
k
| gene | -- * P
fYR
translation
free mRNA (r)
-
koff
Go
on
miRNA
miRNA-mRNA complex (r*)
YR*
Figure 3.5.1 Biochemistry of the miRNA-mediated gene regulatory system.
copies/cell
30Ont
miR-20a
li
I
A
tRNAo
0
Figure 3.5.2 miR-20 expression in Tet-On HeLa cells. a. Absolute miR-20
expression measured by northern blot. Total RNA from Tet-On HeLa cells
transfected with various reporter constructs was probed for miR-20 expression
compared to a standard curve of miR-20 mimic spiked into yeast RNA.
tRNAgln serves as a loading control. b. Relative miR-20 expression above and
below the threshold measured by RT-PCR. Cells transfected with the N = 7
target reporter or the N = 0 control reporter were sorted into low and high
fractions. Total RNA was assayed for miR-20 and normalized to miR-31 as a
loading control. Bar height and error bars represent the average relative
normalized miR-20 value in the high fraction compared to the low fraction
and the s.e.m. of three RT-PCR assays.
3.5.2 Tuning the dissociation constant X
The qualitative shape of the transfer functions generated by the model depends on two
key lumped parameters. The dissociation constant k governs the sharpness of the
threshold, as seen in Figure 3.5.3. On a log-log plot relating r to runtargeted as seen in
Figure 3.5.3 (right panel), the increased sharpness manifests itself as a line with slope
(which we refer to as the logarithmic gain) greater than 1, marking an ultrasensitive
transition connecting the branches of the transfer function of slope 1 that indicate little
protein expression (below the ultrasensitive transition) and nearly maximal protein
production (above the ultrasensitive transition). k is inversely proportional to the rate at
which miRNA binds the target mRNA (kon); as kon increases at a constant kog, k
decreases and thus sharpens the transition.
low ko
low k
C,)
0
high kon
high kon
runtargeted
|O g (runtargeted)
Figure 3.5.3 Tuning the sharpness of the ultrasensitive switch by changing
the rate at which miRNA bind their target mRNA, kon. As shown especially
strikingly in the right panel, increasing kon has a dramatic effect on the
strength of repression below the transition to escape from miRNA repression,
thus sharpening the transition, but does not change the transcript level needed
to encounter the transition.
3.5.3 Tuning the threshold constant 0
The threshold constant 0 plays a role in the placement of the threshold and also in the
sharpness of the transition between the threshold and escape regimes, as seen in Figure
3.5.4. 0 is proportional to the concentration of free miRNA available within the cell; as
the total concentration of free miRNAs increases, 0 increases and pushes the
ultrasensitive transition to higher values of runtrgetedasdepicted in Figure 3.5.4b.
O
low [miRNA]
low [miRNA]
0.
high [miRNA]
high [miRNA]
runtargeted
|og g(runtargeted)
Figure 3.5.4 Tuning both the placement and sharpness of the ultrasensitive
transition by titrating different total amounts of miRNA into the system. As
with Figure 3.5.3, the key features of the effects of adding miRNA into the
system are best shown in the log-log transfer function (right panel). As with
increasing ko., increasing [miRNA] increases the strength of repression
below the ultrasensitive transition. But unlike with kn changes, increasing
the [miRNA] can also increase the threshold transcript level needed to escape
beyond maximum fold-repression.
3.6 Experimentally tuning the ultrasensitive transitions
The mathematical model thus suggests experiments that could be performed to modulate
the ultrasensitive transitions generated by miRNA-mediated regulation. As our stable cell
lines could not achieve high enough levels of reporter expression to capture the complete
ultrasensitive transition to escape from miRNA-mediated repression, we carried out the
remainder of our experiments by transiently transfecting HeLa cells with reporter
constructs and measuring fluorescence via flow cytometry to increase the number of cells
in our datasets.
3.6.1 Increasing N in the mCherry 3'-UTR
To sharpen the transitions by increasing kon we engineered the 3' UTR of mCherry to
increase N, the number of miRNA target sites. The maximum logarithmic gain increases
smoothly from ~1 when N=l to 1.8 when N= 7, shown in Figure 3.6.1; as expected from
the model, the effect is stronger going from 1 to 4 binding sites than from 4 to 7 sites. We
were also able to recapitulate the transfer function with N=7 in the 3' UTR of eYFP,
shown in Figure 3.6.2, thus isolating the effect to miR-20 mediated regulation rather than
any property intrinsic to the mCherry reporter. Interestingly, unlike with previous
experiments using bacterial sRNA [Levine 2007], we can also directly test the
importance of titration to generate thresholds by using miR-20 binding sites that are
perfectly complementary to the endogenous miR-20, thus converting the interaction
between target and miRNA into a strongly catalytic, RNAi-type repression. We observe
in Figure 3.6.1 (grey points) that when the miR-20 bulged binding sites are replaced by a
perfectly complementary binding site that yields the same maximum repression as N=7
bulge sites and, the ultrasensitive transition is abolished altogether.
1 perfect
.N
4
04e
0
3 *.0
3
4
log 10(eYFP)
5
Figure 3.6.1 Experimentally sharpening the ultrasensitive transition by
engineering differing numbers of miR-20 binding sites into the 3'-UTR
of mCherry. The angular symbols are meant to denote the derivative of
the transfer function at the location indicated; this derivative is referred
to as the logarithmic gain, which is a key system parameter characterizing
the regulatory interaction. Additionally, we can abolish the ultrasensitive
response by using a miR-20 binding site that is perfectly complementary
to miR-20, as shown in the dark grey points.
.......
....
....
......
-slope=
+
1
N=7
loglo(eYFP)
Figure 3.6.2 Dye-swap control experiment. We observe a quantitatively similar
logarithmic transfer function with a N=- 7 construct engineered to contain the
miR-20 binding sites in the 3'-UTR of eYFP rather than mCherry, except that
the curve is reflected about the y-x line as expected. This suggests that the
thresholding with ultrasensitivity can be attributed to miR-20 mediated regulation,
not to any property intrinsic to mCherry.
3.6.2 Calculating ratio transfer functions to measure fold repression
To measure the fold repression as a function of target expression level, we measure the
transfer function in the absence of miR-20 binding sites and calculate the ratio of this
control transfer function to transfer functions in the presence of 1, 4, and 7 miR-20 sites;
the results are plotted in Figure 3.6.3. As expected from Figure 3.6. 1, increasing the
number of binding sites both increases the fold repression at lower eYFP levels, from just
over 2-fold repression with a single miR-20 site to ~10-fold repression with seven miR20 sites, while not significantly changing the fold repression at high eYFP (Figure 3.6.3).
Seen this way, we demonstrate that rather than being only a subtle effect as suggested by
population-based averages, which in this case results in at most 2.5-fold repression with
seven binding sites (see Figure 3.6.4), regulation by miR-20 can exert very strong
repression of protein production at low target transcript levels. Moreover the boundary of
100
...........
:::::::::
........
............
M
...................
....
.. . ..
....
. ......
.......
the regime of strongest repression is marked by the ultrasensitive transition, so shifting
this transition to lower or higher target mRNA levels can be of functional significance.
*N=1
ON=4
-
10
*N 7
0
*0
40
41
0
x 104
1
2
3
eYFP
4
Figure 3.6.3 Calculating the fold repression due to miRNA as a
function of target expression level. We obtain this plot by taking
the ratio of the N = 0 curve to the N = 1, 4, and 7 curves. The
magnitude of the fold repression stands in stark constrast to
estimates from bulk measurements as in Figure 3.6.3. The
potential functional significance of the ultrasensitive transition
is also highlighted in this plot as it is precisely this transition
that allows the regulatory system to tune through virtually all
magnitudes of fold-repression just by regulating the target
expression level.
101
0
C,)
10
N=1
N=4
N=7
Figure 3.6.4 Bulk level measurements of miR-20 mediated repression. The
levels of repression observed using our fluorescent reporter system is on
average similar to those previous observed. This data was calculated from
flow cytometry: we compute the ratio of the mean eYFP level to the mean
mCherry level for N = 1, 4, and 7. We then normalize this ratio by the mean
eYFP to mean mCherry ratio for N= 0; we refer to this normalized ratio as the
fold repression. Error bars are estimated by bootstrapping from the single cell
flow cytometry data.
3.6.3 Changing [miR-20]otai by transfecting mimic siRNA
Consistent with the model, the ultrasensitive transition can be shifted to either higher or
lower eYFP levels by transfecting either miR-20 mimic oligonucleotides (siRNAs) or
miRNA sponges that inhibit miR-20 activity [Ebert 2007] (Figure 3.6.5; Figure 3.6.6).
Increasing the level of miRNA increased the fold-repression below the threshold; the
threshold mRNA level needed for protein expression; and the sharpness of the transition.
In the extreme case of 7 miR-20 binding sites with 30 nM miR-20 mimic transfected
(Figure 3.6.5, right panel), miRNA-mediated repression can achieve -40-fold repression
compared to a target with no miRNA binding site; the threshold is shifted to a 10-fold
higher eYFP level; and the transition between repressed and unrepressed expression is
quite sharp with a maximum logarithmic gain of ~5.4 (Figure 3.6.5, right panel),
compared to -1.8 without the transfected miR-20 mimic, i.e. endogenous levels (Figure
3.6.1).
102
.............
::
...........
SN=0
+ N
5
*N=7
mim
+ 30nM mimic
-model
A + UnM mimic
+ 9nM mimic
5
30nM mimic
model
+
-
3.2
S
5.4
&
e
4C
4
E
E
3
4-
3
0
4-
2.
3
4
Iog1 (eYFP)
5
3
4
log 1O(eYFP)
5
Figure 3.6.5 Tuning the placement and the sharpness of the threshold by titrating
the amount of miR-20 available to the gene regulatory system. As expected from
the theoretical results, increasing the amount of miR-20 molecules in the system
both greatly increases the maximal fold repression (which reaches 40-fold in the
case of N= 7 with 3OnM miR-20 added) as well as the sharpness with which the
transfer function snaps onto the unrepressed regime (with a maximal logarithmic
gain of 5.4 in the case of N = 7).
103
a
+ control sponge
*
+ miR-20 sponge
*
+ control sponge
+ miR-20 sponge
+ control sponge
*
+ miR-20 sponge
5
5D
E
4
E
0
4
S3
4
1og 1 (eYFP)
5
log 1 (eYFP)
logl,(eYFP)
+ control sponge
+ control sponge
+ miR-20 sponge
*
+ miR-20 sponge
5
E
0
3
4
4
3
3
5
log,,(eYFP)
4
5
logl,(eYFP)
Figure 3.6.6 miR-20 sponge experiments shift ultrasensitive regime to lower
eYFP levels as expected from the mathematical model. a-e) Transfer functions
resulting from cotransfection of indicated reporter system (N = 0, 1, 1 perfect,
4, and 7 respectively) with indicated sponge construct.
To quantitatively compare the data to the model, we simultaneously fit all the datasets
holding k constant across the fits to particular N and 0 constants for a particular amount
of transfected siRNA mimic. Interestingly, we see that the fit parameter 0 increases with
increasing siRNA mimic (Figure 3.6.7, left panel), but in a saturable fashion, while 1/k
increases linearly with N (Figure 3.6.7, right panel). This suggests that the amount of
transfected miRNA entering functional complexes is limited by entry into the cytoplasm
or availability of miRNP components.
104
.
15
10
Z
_410
8
-6
45
CD
2
X 104
.
...
0
10
OX 10-5
20
[miRNA]transfected (nM)
1
30
4
N
7
Figure 3.6.7 Results from simultaneous fitting of model to experimental data.
The fitting results suggest that our manipulations to tune the regulatory system
behaved as we expected: titrating in miR-20 mimicking siRNA increased the
threshold constant, while increasing N increased 1/k. It is possible that the
deviation of fitted 1I/X from the straight line is a signature of cooperative binding
of miR-20 molecules to target mRNAs, but further study is needed to explore
this.
3.6.4 eYFP mRNA abundance at the threshold
In order to get a feel for whether or not the transitions occurred at target mRNA levels at
all physiologically relevant, we sought to measure the eYFP transcript abundance at
which the ultrasensitive transition began to occur. Using cell sorting via flow cytometry,
we could isolate cellular subpopulations that exhibit only below-threshold and only
above-threshold gene expression, as seen in Figure 3.6.8. We can then perform RT-PCR
on eYFP mRNA transcripts from these two subpopulations to estimate the threshold
mRNA abundance. From these estimates, the data suggests that the threshold transition
occurs at approximately 100 target mRNAs per cell with seven typical sites in the 3'
UTR at an endogenous level of approximately 2,000 miR-20 per cell, shown in Figure
3.6.8 (lower panel).
105
(G1: RI & R2)
104-r---
R--- -R4
(Gl
-1R8
R1 & R2)
10
-Z
N
R1q_--
. .
A7
R7
j.
10:1
'-:1
10
R6
100
10DO
R
1
101
100
YFP Log
103
10p
10
104
10
10'
10Lo
YFP Log
1.2
Fraction
mRNAs per cell
N=0 low
56+/-36
1
0.0
S0.6
N=0high
Og
1066 +/-472
N=0 low
N=7 low
N=0 high
N=7 high
Figure 3.6.8 Estimating the mRNA abundance at the threshold generated by
miR-20 mediated regulation.
3.7 Observing ultrasensitivity in physiological contexts
In order to test the generality of these findings, that the strength of repression of a
miRNA target depends strongly on the relative amounts of the miRNA and its target, we
sought to recapitulate the results in more physiological settings.
106
3.7.1 Fusing natural 3'-UTRs to mCherry
First, we tested whether similar ultrasensitive transitions would be observed when the
reporter construct incorporated naturally occurring miRNA binding sequences by fusing
the 3' UTRs of the oncogene HMGA2 and the major GABA transporter gene SLC6A1 to
the mCherry reporter and performing dual-color FACS. The HMGA2 3' UTR contains
seven binding sites for the miRNA family let-7, which is abundant in HeLa cells, while
SLC6A1 contains three binding sites for the neuronal miRNA miR-218, which we
supplied exogenously. The experiments, whose results are depicted in Figures 3.7.1,
showed that we could indeed observe ultrasensitive transitions with these constructs and
for HMGA2, we increased the ultrasensitive threshold incrementally by transfecting
higher doses of let-7 siRNA mimic (Figure 3.7.1).
a
b
43
445
2
3
2
3
3
1
2.5
3.5
4
_5
4.5
log 10(eYFP)
iog 10(eYt-v)
* Mutant HMGA2 3 UTR
HMGA2 3'UTR
let-7 mimic
HMGA2 3' UTR + 10knM
*
SLC6A1 3 UTR
+30nM miR-218 imnic
* HMGA2 3' UTR + 31nM let-7 mimic
* HMGA2 3 UTR +10nM let-7 mimic
Figure 3.7.1 Detecting ultrasensitive transitions with natural UTR's. a) Using
the 3'-UTR from the gene HMGA2, which contains 7 binding sites for the
miRNA let-7, we observe that we can tune the system to display a clear
signature of ultrasensitivity (the slope of the log-log transfer function exceeds
the slope = 1 guide to the eye shown in grey) when adding increasing amounts
of let-7 mimic siRNA. b) The 3'-UTR from the neuronal gene SLC6A1 also
shows a clear ultrasensitive transition when we add its targeting miRNA miR-218.
107
3.7.2 Luciferase assays in mouse embryonic stem cells
Finally we used a standard dual-luciferase assay, shown in Figure 3.7.2 in schematic
form, to measure target expression in mouse embryonic stem cells (ES cells) using only
their endogenous pool of miRNA to retain physiological relevance. Furthermore, we
measured a transfer function complementary to that in the experiments with Hela cells:
the mRNA target level remained fixed while the miRNA concentration varied. To test
varying miRNA concentrations we exploited the fact that different miRNA species are
present at different abundances in ES cells. Finally, to gauge the strength of miRNA
repression, target expression in wild-type ES cells was normalized to target expression in
ES cells that lack the enzyme Dicer and thus contain no miRNAs.
N=2 sites or
CXCR4 control
3'UTR
o
F-uferase
Transfect R-luc with 2
bulged miRNA sites or
CXCR4 control sites;
F-luc is the loading control
Dcr +/+)
Measure expression of
construct with miRNA
sites relative to construct
with CXCR4 control sites
relative expression in
fold repression
=
relative expression in
relative expression in
relative expression in
(r/
Figure 3.7.2 Dual luciferase assay system used to measure miRNA mediated
repression in populations of mouse embryonic stem cells.
We observe a similar threshold-linear curve in Figure 3.7.3 except that it reflected the
level of miRNAs: at high miRNA abundances, fold-repression is 5-fold but decreases
with miRNA abundance until at the lowest miRNA abundances target expression in wildtype cells is virtually indistinguishable from that in the miRNA-free Dcr~ cells.
108
...................
6
C
.0 5
3
.2
*
I
*e*
0 2 4 6 8 10 12 14
[miRNA]
x 103 per cell
Figure 3.7.3 Fold repression increases as a function of miRNA abundance
in mouse embryonic stem cells. The miRNA abundance is estimated from
semi-quantitative Northern blots, while the fold repression is measured using
the dual luciferase assay detailed in Figure 3.7.2. See section 3.9 for more
detailed methods. Data courtesy of Grace Zheng.
The threshold in regulation by miRNA is determined by the level of the miRNA, and the
number and affinity of the target sites. Many of these miRNAs as miRNP complexes
could be bound to the endogenous miR-20 target mRNAs in the cell, leaving a limited
pool for binding to the reporter mRNAs. Since these experiments are done at steady state
conditions, this suggests that the miRNA system has very limited capacity to
accommodate increases in target populations. These results are consistent with previous
observations using "sponges" to suppress the activities of a family of miRNAs. Here
expression of high levels of miR-20 target sites from an exogenously added sponge
construct strongly suppressed miR-20 regulation of the target reporter; if endogenous
miR-20 target mRNA production were to increase substantially, some escape from miR20 repression would also be expected (as seen in Figure 3.6.6). The sponge phenomenon
has been observed in multiple mammalian and non-mammalian organisms indicating the
general nature of this threshold behavior for miRNA regulation.
109
3.8 Discussion
Our analysis of miRNA-mediated gene regulation at high target expression levels is
consistent with previous bulk results, but measuring single cells offers a level of detail
inaccessible to population-based assays. The detailed picture, which revealed the
ultrasensitive response bounded by a high degree of repression at low target mRNA
levels and little repression at high levels of target mRNA, may have important
implications for miRNA-mediated regulation. There has been disparity between the
concept of miRNAs as switches, exemplified by the lin-14 developmental switch in
Caenorhabditiselegans where there is a high degree of repression by the miRNA Lin-4,
versus many observations of miRNA-mediated regulation in mammalian cells where they
are best considered as fine-tuners of gene expression. These results show that for any
given miRNA-target interaction, the miRNA behaves both as a switch, in the target
expression regime below the threshold, and as a fine-tuner, in the ultrasensitive transition
between the threshold and the minimal repression regime at high mRNA levels. This
model is consistent with miRNAs providing robustness to systems. Target mRNAs that
were transcribed at low levels and/or only transiently would be strongly repressed but
then upon increased and sustained expression the system could produce a rapid and
irreversible transition to stable state expressing high levels of the target protein.
The target expression thresholds generated by miRNAs could be important in
development. Ultrasensitivity characterizes developmental switches such as cell fate
decisions. To maintain their identity, differentiated cells must be able to distinguish
between leaky and legitimate transcripts. Consistent with this, miRNAs are known to
participate in feedback and feed-forward networks [Tsang 2007] and miRNA-mediated
feedback networks have been implicated in imparting robustness in developing embryos
[Stark 2005]. Molecular titration by tissue-specific miRNAs could set a threshold below
which transcripts would be treated as leaky. Such a phenomenon is consistent with the
observed tendency of mammalian miRNAs induced upon differentiation to target
mRNAs that were highly expressed in the previous developmental stage [Farh 2005], and
with the reported tendency of Drosophila miRNAs to target mRNAs that are highly
expressed in neighboring tissues derived from a common progenitor [Stark 2005]. The
110
ultrasensitive transition would minimize the range of uncertainty between leaky and
legitimate. Decisive on-off regulation of gene expression is necessary in differentiation
and in the continual reinforcement of cell/tissue identity throughout the life of the animal.
3.9 Methods
Fluorescent reporters were cloned into pTRE-Tight-BI (Clontech). NLS sequences were
appended to the N-terminus of the eYFP and mCherry ORFs by PCR. The NLS-eYFP
was inserted with EcoRI and NdeI. The NLS-mCherry was inserted with BamHI and
Clal. Regulatory elements were placed into the eYFP 3' UTR with NdeI and XbaI; they
were placed into the mCherry 3' UTR with Clal and EcoRV. 4x and 7x miR-20 sites
were PCR-amplified from miR-20 sponge constructs. All constructs were sequenceconfirmed. HMGA2 w.t. and seed-mutant UTRs were a gift from Christine Mayr, David
Bartel lab. The SLC6A1 3' UTR fragment was PCR-amplified from human genomic
DNA.
Generation of stable lines
Reporter plasmids were linearized with AseI and cotransfected at 20:1 ratio with linear
puromycin marker (Clontech). Transfected cells were selected in 2.5 ug/ml puromycin
with 200 ug/ml G418. Individual eYFP-positive colonies were isolated, grown, and
sorted for eYFP-positivity upon dox induction (MoFlo instrument).
Fluorescence microscopy
Cells were plated on glass-bottomed Nunc chambers (#1), induced with dox for 4 days,
imaged in a Nikon TEI-2000 inverted fluorescence microscope with a Princeton
Instruments Pixis back-cooled CCD camera. Images were processed using custom
software in MATLAB. Briefly, following subtraction of camera background and any
cellular autofluorescence,
pixel values
in both eYFP
and mCherry
channels
corresponding to cells expressing the construct were extracted. The single-cell data were
then binned along the eYFP axis. Figure ld reports the result of this binning procedure;
the error bars are the standard errors of the mean within its corresponding bin.
111
subtract
autofluorescent
background
log,,(eYFP)
log,,(eYFP)
I'.
calculate mean
mCherry ineach
eYFP bin
7
I.
-
-
/
-
-.
I-
1og0 O(eYFP)
Figure 3.9.1 Binning procedure used to convert joint mCherry-eYFP single
cell distributions into transfer functions. The example showed here is from
a typical flow cytometry experiment. The raw data in the first panel is
first subjected to background subtraction. Then all the cells between two
particular values of eYFP, as the schematic red column indicates in the
second panel, are analyzed for their mCherry intensity value. The mean
mCherry intensity is then plotted as a point in the third panel; the collection
of all such points makes up the full transfer function depicted in the third
panel.
Transient transfection
Tet-On HeLa cells (Clontech) below passage 10 were plated in G418 (Gibco) 200 ug/ml
and doxycycline (Sigma) 1 ug/ml media in 12-well dishes the day before transfection.
Reporter plasmids were diluted 1/50 in pUC18b carrier plasmid (Qiagen HiSpeed
maxipreps) and mixed with DreamFect Gold (Oz Biosciences) at 4:1 ul reagent: ug DNA.
miR-20a, let-7b, and miR-218 mimics (Dharmacon) were cotransfected at the indicated
concentrations. For U6 sponge assays, reporter plasmids were diluted 1/50 in sponge
112
plasmid. Media was changed 24hr post-transfection. Assays were performed 48hr posttransfection.
Flow cytometry
Cells were run on LSRII analyzer (Becton Dickinson) with FACSDiva software. The raw
FACS data were analyzed with FlowJo to gate cells according to their forward (FSC-A)
and side (SSC-A) scatter profiles; specifically we chose cells near the peak of the (FSCA, SSC-A) distribution. Untransfected cells were used to characterize the cellular
autofluorescence in the LSRII analyzer from which we obtain the mean and standard
deviation of the autofluorescence
distribution. Each cell's eYFP and mCherry
fluorescence values were subtracted by the mean autofluorescence plus twice the
standard deviation. Following background subtraction, cells with eYFP fluorescence
levels less than 0 (i.e. indistinguishable from background) were excluded from further
analysis and mCherry fluorescence levels less than 0 were set equal to 0. The single-cell
data were then binned in the same manner as described above.
Fluorescence-activated cell sorting
Cells transfected with the N=0 or N=7 reporter were sorted 48hr post-transfection into
low and high fractions using a MoFlo high-speed sorting instrument (DAKOCytomation). Cell pellets were washed and snap-frozen before RNA isolation.
RT-PCR
Total RNA was harvested using RNeasy Micro Plus kit with the protocol modified for
inclusion of small RNAs (Qiagen). RNA was treated with DNaseI (Ambion) and reversetranscribed with oligo-dT primer using MMLV RTase (Ambion). qPCR for mCherry and
eYFP was performed in triplicate reactions using SYBRGreen mix (Applied Biosystems),
run on Applied Biosystems 7500 Real-Time PCR instrument. Single-stranded DNA
standards spiked into untransfected cell cDNAs were used for estimation of mCherry
mRNAs per cell. miR-20 was measured with miScript RT-PCR assay (Qiagen) in
quadruplicate reactions using miR-31 and snoRNA as controls.
113
Small RNA Northern blot
24 ug of total RNA from transfected cells was run on 12% polyacrylamide gel (UreaGel
system, National Diagnostics), with miR-20 mimic as a standard, spiked into yeast
sheared total RNA (Ambion). The blot was probed for miR-20a and tRNAgn as loading
control.
mES cell luciferase assays
Reporters were constructed by insertion of two bulged binding sites into the 3' UTR of
CMV Renilla luciferase. Cells were transfected in triplicate in 24-well plates with 2ul
Lipofectamine 2000 (Invitrogen), 0.0lug of CMV-Renilla plasmid, 0.lug of pGL3
(Promega), and 0.69ug of pWS (carrier plasmid). Cells were lysed and assayed 24hr posttransfection by Dual Luciferase reporter assay (Promega) using a on Glomax 20/20
luminometer (Promega).
114
115
116
Chapter 4
Conclusion and Perspectives
In this Thesis, I have attempted to address a simple question that confronts any living
cell: how does it resolve the tension between maintaining itself unchanged in the face of
the random forces constantly confronting it but then changing when circumstances
demand? The tension exists because either extremal behavior would lead to suboptimal
outcomes for any given challenge the cell faces. If the cell rapidly changed with every
fluctuating input, then it is likely it could devote excessive resources to responding to
minor challenges or worse, especially in the context of the development of a multicellular
organism for example, switch on a gene expression program in response to an aberrant
signal generated by stochastic fluctuations. On the other hand, if the cell robustly
maintained its dynamical behavior despite real changes in its environment that cannot be
averaged out and must instead be responded to then its existence could be threatened.
Clearly the cell must be able to balance these conflicting goals.
The case of microRNA mediated gene regulation provides a simple, intuitive example of
how the cell can use simple regulatory components - specifically the nonlinearities in the
biochemistry of these components - to balance these goals. By examining how single
cells employ microRNA mediated regulation, we were able to observe that microRNAs
establish gene expression thresholds: transcriptional activity must reach a critical
threshold level before appreciable protein product is generated for the cell. The
117
thresholding phenomenon thus establishes two regimes in the gene expression system:
below the threshold expression is robustly switched off, while above the threshold the
system can explore every possible degree of repression until escaping from microRNA
mediated regulation altogether. In this case, the balance between robustness and
tunability is achieved directly by biochemical details: the affinity of the microRNA for its
target and the relative abundance of the microRNA and its target determine the strength
of the nonlinearities in the interaction that in turn mark the decision boundary between
robust and tunable expression regimes. Very strong binding and very abundant levels of
microRNA compared to target mRNA levels most clearly delineate when the cell will
robustly keep expression silenced and when it will allow proteins to be produced.
While microRNA mediated regulation provided an example of how to be robust yet
tunable in the face of biochemical fluctuations, the case of the yeast osmosensing
pathway provides a concrete example of how simple networks can achieve this balance
against genetic perturbations. By conducting a systematic complementation study in
which we substituted orthologs for endogenous HOG pathway genes, we observed that
the large-scale genetic changes introduced into the pathway were only readily observable
in the downstream MAPK module rather than the upstream phosphorelay module. In this
case the balance between robustness and tunability is achieved by decomposition: in the
context of the HOG pathway, the input/output relationship of the phosphorelay Module is
relatively robust to changes in the sequence of its constituent genes while that of the
MAPK module can be tuned genetically. Similar to the microRNA system, our modeling
suggests, the decision boundary in this case is due to biochemistry. According to our
modeling, stoichiometric signaling as exemplified by the phosphorelay module, because
its steady states are entirely determined the inflow and outflow of the signaling molecule
(phosphate in our case) with all internal details suppressed, is more robust to changes in
kinetic rates than catalytic signaling. Catalytic signaling as exemplified by the MAPK
module, with its potential for signal amplification, can be tuned by each rate constant in
the reaction scheme and thus by any change in protein sequence that can affect any of
those rates.
118
There are clearly many other routes toward achieving this balance, some of which have
been explored but some of which are in need of attention. Arguably the best explored of
these other routes, at least in the systems biology literature, is positive feedback,
especially when it generates bistability. Taking an analogy from statistical mechanics,
which is commonly done in visualizing bistable systems, imagine the system has two
states separated by a potential energy barrier. The most natural way to construct a robust
yet tunable system from this picture is for the high state to have a relatively flat profile:
output (e.g. gene expression)
In such a system, the potential energy barrier marks the decision between robustly
keeping the output of the system in its low state, while the flat landscape above the
barrier allows the system to be tuned to a variety of values in the high state. Positive
feedback can even be used in conjunction with other regulatory schemes, such as delayed
negative feedback, to help make the dominant dynamics of those other regulatory
schemes, such as oscillation, more robust and tunable (see Section 1.5 for examples).
Another way to achieve this balance is to exploit the potential for combinatorial control
of a given process. Downstream gene expression systems can ignore the presence of a
particular signaling molecule unless sensitized by the presence of a separate signaling
molecule, after which the amount of the first signaling molecule can dictate the
magnitude of the transcriptional response. Such a principle could underlie the concept of
"danger" signals in the immune system, which is very clearly an example of a system that
must both be robustly silenced when its functions are not needed for the organism but
then must be tuned to have an appropriate response, neither too vigorous nor too weak.
119
One aspect of this balance between robustness and tunability that has been clearly
underexplored experimentally is the case of spatially structured systems. These issues
have been begun to be explored in some studies in developmental biology [Gregor 2007,
Ben-Zvi 2010]. For example, spatial averaging could play a role in determining the
robustness of pattern formation systems to spatial fluctuations: the number of cells over
which a particular molecular abundance is averaged can determine the length scale over
which a spatial fluctuation will persist. Robust averaging below this length scale can
smooth out unwanted fluctuations but still allow patterns to be sculpted at longer length
scales.
The tension between robustly ignoring perturbations and tunably responding to them
when necessary is among the core tensions in the life of the cell. The studies presented
here add two new examples of how cells exhibit both robust and tunable behaviors. We
hope this and the work of our colleagues will inspire future studies addressing these two
aspects of biology.
120
121
122
References
Acar M, Becskei A, van Oudenaarden A. Enhancement of- cellular memory by
reducing stochastic transitions. Nature 435(7039): 228-232 (2005)
Acar M, Mettetal JT, van Oudenaarden A. Stochastic switching as a survival strategy
in fluctuating environments. Nat. Genet. 40: 471-475 (2008)
Albertyn J, Homann S, Thevelein JM, Prior BA. GPD1, which encodes glycerol-3phosphate dehydrogenase,
is essential for growth under osmotic
stress in
Saccharomyces cerevisiae, and its expression is regulated by the high-osmolarity
glycerol response pathway. Mol Cell Biol. 14: 4135-4144 (1994)
Alon U, Surette MG, Barkai N and Leibler S. Robustness in bacteria chemotaxis.
Nature 397: 168-171 (1999)
Alon U. Network motifs: theory and experimental approaches. Nat Rev Genet 8: 450461 (2007)
Anderson JC, Clarke EJ, Arkin AP, Voigt CA. Environmentally controlled invasion
of cancer cells by engineered bacteria. J Mol Biol. 355, 619-627 (2006)
Antunes MS, Morey KJ, Tewari-Singh N, Bowen TA, Smith JJ, Webb CT, Hellinga
HW, Medford JI. Engineering key components in a synthetic eukaryotic signal
transduction pathway. Mol Syst Bio 5: 270 (2009)
Armbruster BN, Li X, Pausch MH, Herlitze S, Roth BL. Evolving the lock to fit the
key to create a family of G protein-coupled receptors potently activated by an inert
ligand. PNAS 104, 5163-5168 (2007)
Atkinson MR, Savageau MA, Myers JT, Ninfa AJ. Development of genetic circuitry
exhibiting toggle switch or oscillatory behavior in Escherichiacoli. Cell 113(5): 597607 (2003)
123
Baek D, Villen J, Shin C, Camargo FD, Gygi SP, Bartel DP. The impact of
microRNAs on protein output. Nature 455: 64-71 (2008)
Bagga S, Bracht J, Hunter S, Massirer K, Holtz J, Eachus R, Pasquinelli AE.
Regulation by let-7 and lin-4 miRNAs results in target mRNA degradation. Cell 122:
553-563 (2005)
Balagadde FK, Song H, Ozaki J, Collins CH, Barnet M, Arnold FH, Quake SR, You
L. A synthetic Escherichia coli predator-prety ecosystem. Mol Syst Biol 4: 187
(2008)
Barkai N, Leibler S. Robustness in simple biochemical networks. Nature 387, 913917 (1997)
Bartel DP, Chen CZ. Micromanagers of gene expression: the potentially widespread
influence of metazoan microRNAs. Nat Rev Genet. 5(5): 396-400 (2004)
Bashor CJ, Helman NC, Yan S, Lim WA. Using engineered scaffold interactions to
reshape MAP kinase pathway signaling dynamics. Science 319(5869): 1539-1543
(2008)
Basu S, Mehreja R, Thiberge S, Chen MT, Weiss R. Spatiotemporal control of gene
expression with pulse-generating networks. PNAS 101(17): 6355-6360 (2004)
Basu S, Gerchman Y, Collins CH, Arnold FH, Weiss R. A synthetic multicellular
system for programmed pattern formation. Nature 434(7037): 1130-1134 (2005)
Batchelor E, Goulian M. Robustness and the cycle of phosphorylation and
dephosphorylation in a two-component regulatory system. Proc. Natl. Acad. Sci.
USA 100: 691-696 (2002)
Becskei A and Serrano L. Engineering stability in gene networks by autoregulation.
Nature 405: 590-593 (2000)
124
Becskei A and Serrano L. Positive feedback in eukaryotic gene networks: cell
differentiation by graded to binary response conversion. EMBO J. 20(10): 2528-2535
(2001)
Becskei A, Kaufmann BB, van Oudenaarden A. Contributions of low molecule
number and chromosomal positioning to stochastic gene expression. Nat. Genet. 37,
937-944 (2005)
Behar M, Dohlman HG, Elston TC. Kinetic insulation as an effective mechanism for
achieving pathway specificity in intracellular signaling networks. PNAS 104, 1614616151 (2007)
Beisel CL, Bayer TS, Hoff KG, Smolke CD. Model-guided design of ligandregulated RNAi for programmable control of gene expression. Mol Syst Biol 4: 224
(2008)
Ben-Zvi D, Barkai N. Scaling of morphogen gradients by an expansion-repression
integral feedback control. Proc Natl Acad Sci 107(15): 6924 - 6929 (2010)
Bernstein E, Kim SY, Carmell MA, Murchison EP, Alcorn H, Li MZ, Mills AA,
Elledge SJ, Anderson KV, Hannon GJ. Dicer is essential for mouse development. Nat
Genet 35(3): 215-217 (2003)
Bhattacharyya RP, Remenyi A, Yeh BJ, Lim WA. Domains, motifs, and scaffolds:
the role of modular interactions in the evolution and wiring of cell signaling circuits.
Ann Rev Biochem 75: 655-680 (2006)
Borneman AR, Gianoulis TA, Zhang ZD, Yu H, Rozowsky J, Seringhaus MR, Wang
LY, Gerstein M, Snyder M. Divergence of transcription factor binding sites across
related yeast species. Science 317: 815-819 (2007)
Brenner K, Karig DK, Weiss R, Arnold FH. Engineered bidirectional communication
mediates a consensus in a microbial biofilm consortium. PNAS 104(44):17300-17304
(2007)
125
Brewster JL, de Valoir T, Dwyer ND, Winter E, Gustin MC An osmosensing signal
transduction pathway in yeast. Science 259: 1760-1763 (1993)
Bridgham JT, Carroll SM, Thornton JW. Evolution of hormone-receptor complexity
by molecular exploitation. Science 312, 97-101 (2006)
Buchler NE, Gerland U, Hwa T. On schemes of combinatorial transcription logic.
PNAS 100, 5136-5141 (2003)
Buchler N and Louis M. Molecular titration and ultrasensitivity in regulatory
networks. J Mol Biol. 384(5): 1106-1119 (2008)
Bulter, T. et al. From the Cover: Design of artificial cell-cell communication using
gene and metabolic networks. Proc. Natl. Acad. Sci. U. S. A. 101, 2299-2304 (2004)
Cai L, Friedman N, Xie XS. Stochastic protein expression in individual cells at the
single molecule level. Nature 440, 358-362 (2006)
Cambridge SB, Geissler D, Keller S, Curten B. A caged doxycycline analogue for
photoactivatable gene expression. Angew Chem Int Ed. 45: 2229-2231 (2006)
Cantone I, Marucci L, Iorio F, Ricci MA, Belcastro V, Bansal M, Santini S, di
Bernardo M, di Bernardo D, Cosma MP. A yeast synthetic network for in vivo
assessment of reverse-engineering and modeling approaches. Cell 137(1): 172-181
(2009)
Chen C-Z, Li L, Lodish HF, Bartel DP. MicroRNAs modulate hematopoietic lineage
differentiation. Science 303(5654): 83-86 (2004)
Chen M-T, Weiss R. Artificial cell-cell communication in yeast Saccharomyces
cerevisiae using signaling elements from Arabidopsis thaliana.Nat.Biotech. 23:15511555 (2005)
126
Cherry JM, Adler C, Ball C, Chervitz SA, Dwight SS, Hester ET, Jia Y, Juvik G, Roe
T, Schroeder M, et al. SGD: Saccharomyces Genome Database. Nucleic Acids
Research 26: 73-79 (1998)
Chuang JS, Rivoire 0, Leibler S. Simpson's paradox in a synthetic microbial system.
Science 323(5911): 272-275 (2009)
Cimmino A, Calin GA, Fabbri M, Iorio MV, Ferracin M, Shimizu M, et al. miR-15
and miR-16 induce apoptosis by targeting Bcl2. Proc Natl Acad Sci. 102(39): 1394413949 (2005)
Cironi P, Swinburne IA, Silver PA. Enhancement of cell type specificity by
quantitative modulation of a chimeric ligand. J. Biol. Chem. 283: 8469-8476 (2008)
Conrad ED, Tyson JT. Modeling molecular interaction networks with nonlinear
ordinary differential equations. System Modeling in Cellular Biology, ed. Z. Szallasi,
J. Stelling and V. Periwal: 116-118, MIT Press (2006)
Cox RS 3rd, Surette MG, Elowitz MB. Programming gene expression with
combinatorial promoters. Mol. Syst. Bio. 3: 145 (2007)
Cruz FG, Koh JT, Link KH. Light activated gene expression. JACS 122: 8777-8778
(2000)
Davidson EA. The regulatory genome: gene regulatory networks in development and
evolution. Academic Press. (2006)
Davidson EA, Ellington AD. Synthetic RNA circuits. Nat Chem Biol 3, 23-28 (2007)
Deans TL, Cantor CR, Collins JJ. A tunable genetic switch based on RNAi and
repressor proteins for regulating gene expression in mammalian cells. Cell 130, 363372 (2007)
Desai SK, Gallivan JP. Genetic screens and selections for small molecules based on a
synthetic riboswitch that activates protein translation. JACS 126: 13247-13254 (2004)
127
de Visser JAGM, Hermisson J, Wagner GP, Ancel Meyers L, Bagheri-Chaichian, H,
Blanchard JL, Chao L, Cheverud JM, Elena SF, Fontana W, et al. Perspective:
Evolution and detection of genetic robustness. Evolution 57: 1959-1972 (2003)
Dueber JE, et al. Synthetic protein scaffolds provide modular control over metabolic
flux. Nature Biotechnology 27: 753-759 (2009)
Dugave C and Demange L. Cis-trans isomerization of organic molecules and
biomolecules: implications and applications. Chem Rev 103: 2475-2532 (2003)
Ebert MS, Neilson JR, Sharp PA. MicroRNA sponges: competitive inhibitors of small
RNAs in mammalian cells. Nature Methods 4(9): 721-726 (2007)
Elf J, Paulsson J, Berg OG, Ehrenberg M. Near-critical phenomena in intracellular
metabolite pools. Biophys J 84: 154-170 (2003)
Ellis T, Wang X, Collins JJ. Diversity-based, model-guided construction of synthetic
gene networks with predicted functions. Nat Biotech 27(5): 465-471 (2009)
Elowitz MB, Leibler S. A synthetic oscillatory network of transcriptional regulators.
Nature 403(6767): 335-338 (2000)
Elowitz MB, Levine AJ, Siggia ED, Swain PS. Science 297(5584): 1183-1186 (2002)
Esau C, Kang X, Peralta E, Hanson E, Marcusson EG, et al. MicroRNA-143 regulates
adipocte differentiation. J. Biol Chem. 279: 52361-52365 (2004)
Farh KK, Grimson A, Jan C, Lewis BP, Johnston WK, Lim LP, Burge CB, Bartel DP.
The widespread impact of mammalian microRNAs on mRNA repression and
evolution. Science. 310(5755):1817-21 (2005)
Ferrigno P, Posas F, Koepp D, Saito H, Silver P. Regulated nucleo/cytoplasmic
exchange of HOGI MAPK requires the importin beta homologs NMD5 and XPO1.
EMBO J. 17: 5606-5614 (1998)
128
Friedland A, Lu TK, Wang X, Shi D, Church G, Collins JJ. Synthetic gene networks
that count. Science 324 (5931): 1199-1202 (2009)
Friedman RC, Farh KK, Burge CB, Bartel DP. Most Mammalian mRNAs are
conserved targets of microRNAs. Genome Research 19: 92-105 (2009)
Fung E, Wong WW, Suen JK, Bulter T, Lee S, Liao JC. A synthetic gene-metabolic
oscillator. Nature 435: 118-122 (2005)
Gerhart J, Kirschner M. The theory of facilitated variation. PNAS 104 Suppl 1: 85828589 (2007)
Gertz J, Siggia ED, Cohen BA. Analysis of combinatorial cis-regulation in synthetic
and genomic promoters. Nature 457, 215-218 (2009)
Geva-Zatorsky N, Rosenfeld N, Itzkovitz S, Milo R, Sigal A, Dekel E, Yarnitzky T,
Liron Y, Polak P, Lahav G, Alon U. Oscillations and variability in the p53 system.
Mol Syst Biol 2:2006.0033 (2006)
Gilbert ES, Walker AW, Keasling JD. A constructed microbial consortium for
biodegradation of the organophosphorus insecticide parathion. Appl. Microbiol.
Biotechnol. 61:77-81. (2003)
Golding I, Paulsson J, Zawilski SM, Cox EC. Real-time kinetics of gene activity in
individual bacteria. Cell 123, 1025-1036 (2005)
Gore J, Youk H, van Oudenaarden A. Snowdrift game dynamics and facultative
cheating in yeast. Nature 459(7244): 253-256 (2009)
Grate D, Wilson C. Inducible regulation of the S. cerevisiae cell cycle mediated by an
RNA aptamer-ligand complex. Bioorg Med Chem 9: 2565-2570 (2001)
Gregor T, Wieschaus EF, McGregor AP, Bialek W, Tank DW. Stability and nuclear
dynamics of the bicoid morphogen gradient. Cell 130(1): 141-152 (2007)
129
Grilly C, Stricker J, Pang WL, Bennett MR, Hasty J. A synthetic gene network for
tuning protein degradation in Saccharomyces cerevisiae. Mol Syst Bio 3: 127 (2007)
Grimson, A., M. Srivasatva, B. Fahey, B.J. Woodcroft, H.R. Chiang, N. King, B.M.
Degnan, D.S. Rokhsar, and D.P. Bartel. Early origins and evolution of microRNAs
and Piwi-interacting RNAs in animals. Nature 455: 1193-1197 (2008)
Guet CC, Elowitz MB, Hsing W, Leibler S. Combinatorial synthesis of genetic
networks. Science 296, 1466-1470 (2002)
Hammer K, Mijakovic I, Jensen PR. Synthetic promoter libraries - tuning of gene
expression. Trends Biotechnol. 24, 53-55 (2006)
Harris K, Lamson RE, Nelson B., Hughes TR, Marton MJ, Roberts CJ, Boone C,
Pryciak PM, Role of scaffolds in MAP kinase pathway specificity revealed by custom
design of pathway-dedicated signaling proteins. Curr. Biol. 11, 1815-1824 (2001)
Hasty J, Dolnik M, Rottschafer V, Collins JJ. Synthetic gene network for entraining
and amplifying cellular oscillations. Phys Rev Lett 88(14): 148101 (2002)
Hersen P, McClean MN, Mahadevan L, Ramanathan S. Signal processing by the
HOG MAP kinase pathway. PNAS 105, 7165-7170 (2008)
Hoffman CS UNIT 13.11 Preparation of Yeast DNA. In Current Protocols in
Molecular Biology: Ausubel FM, Brent R, Kingston RE, Moore DD, Seidman JG,
Smith JA, and Struhl K, eds. John Wiley and Sons (2003)
Hohmann, S. Osmotic stress signaling and osmoadaptation in yeasts. Microbiol. Mol.
Biol. Rev. 66: 300-372 (2002)
Hooshangi S, Thiberge S, Weiss R. Ultrasensitivity and noise propagation in a
synthetic transcriptional cascade. PNAS 102(10): 3581-3586 (2005)
130
Howard PL, Chia MC, Del Rizzo S, Liu FF, Pawson T. Redirecting tyrosine kinase
signaling to an apoptotic caspase pathway through chimeric adaptor proteins, PNAS
100:11267-11272(2003)
Isaacs FJ, Hasty J, Collins JJ. Prediction and measurement of an autoregulatory
genetic module. PNAS 100(13): 7714-7719 (2003)
Isaacs FJ, Dwyer DJ, Ding C, Pervouchine DD, Cantor CR, Collins JJ. Engineered
riboregulators enable post-transcriptional control of gene expression. Nat Biotech 22,
841-847 (2004)
Isalan M, Lemerle C, Serrano L. Engineering gene networks to emulate Drosophila
embryonic pattern formation. PLoS Biol. 3, e64 (2005)
Isalan M, Lemerle C, Michalodimitrakis K, Horn C, Beltrao P, Raineri E, GarrigaCanut M, Serrano L. Evolvability and hierarchy in rewired bacterial gene networks.
Nature 452, 840-845 (2008)
Jacob F and Monod J. Genetic regulatory mechanisms in the synthesis of proteins. J
Mol Biol 3: 318-356 (1961)
Janiak-Spens F, Cook PF, West AH. Kinetic analysis of YPD1-dependent
phosphotransfer reactions in the yeast osmoregulatory phosphorelay
system.
Biochemistry (44): 377-386 (2005)
Janin J, Chothia C. Domains in proteins: definitions, location, and structural
principles. Methods Enzymol 115: 420-30 (1985)
John B, Enright AJ, Aravin A, Tuschl T, Sander C, Marks DS. Human microRNA
targets. PLoS Biology 2(11): e363 (2004)
Kaplan J and DeGrado WF. De novo design of catalytic proteins. PNAS 101, 1156611570 (2004)
131
Khosla C, Keasling JD. Metabolic engineering for drug discovery and development.
Nature Reviews Drug Discovery 2, 1019-1025 (2003)
Kinhabwala A and Guet CC. Uncovering cis regulatory codes using synthetic
promoter shuffling. PLoS One 3, e2030 (2008)
Kitano H. Biological robustness. Nat Rev Genet 5: 826-837 (2004)
Klipp E, Nordlander B, Kruger R, Gennemark P, Hohmann S. Integrative model of
the response of yeast to osmotic shock. Nat. Biotechnol. 23: 975-982 (2005)
Kornmann B, Currie E, Collins SR, Schuldiner M, Nunnari J, Weissman JS, Walter P.
An ER-mitochondria tethering complex revealed by a synthetic biology screen.
Science 325(5939): 477-481 (2009)
Krantz M, Becit E, Hohmann S. Comparative genomics of the HOG-signaling system
in fungi. Curr. Genet. 49: 137-151 (2006)
Kwon 0, Georgellis D, Lin ECC. Rotational On-off switching of a hybrid membrane
sensor kinase Tar-ArcB in Escherichia coli. Journal of Biological Chemistry.
278(15): 13192-13195 (2003)
Le MTN, Xie H, et al. MicroRNA-125b promotes neuronal differentiation in human
cells by repressing multiple targets. Mol and Cell Biol 29(19): 5290-5305 (2009)
Levine E, Zhang Z, Kuhlman T, Hwa T. Quantitative characteristics of gene
regulation by small RNA. PLoS Biol 5(9): e229 (2007)
Levskaya A, Chevlier AA, Tabor JJ, Simpson ZB, Lavery LA, Levy M, Davidson
EA, Scouras A, Ellington AD, Marcotte EM, Voigt CA. Engineering bacteria to see
light. Nature 438: 441-442 (2005)
132
Lewis BP, Burge CB, Bartel DP. Conserved seed pairing, often flanked by
adenosines, indicates that thousands of human genes are microRNA targets. Cell 120:
15-20 (2005)
Li X, Carthew RW. A microRNA mediates EGF receptor signaling and promotes
photoreceptor differentiation in the Drosophila eye. Cell 123(7): 1267-1277 (2005)
Li X, Cassidy JJ, Reinke CA, Fischboeck S, Carthew RW. A microRNA imparts
robustness against environmental fluctuation during development. Cell 137(2): 273282 (2009)
Li Y, Wang F, Lee JA, Gao FB. MicroRNA-9a ensures the precise specification of
sensory organ precursors in Drosophila. Genes Dev. 20(20): 2793-2805 (2006)
Lim LP, Lau NC, Weinstein EG, Abdelhakim A, Yekta S, Rhoades MW, Burge CB,
Bartel DP. The microRNAs of Caenorhabditiselegans. Genes Dev 17(8): 991-1008
(2003)
Maeda T, Wurgler-Murphy SM, and Saito H. A two-component system that regulates
an osmosensing MAP kinase cascade in yeast. Nature 369: 242-245 (1994)
Maeda YT, Sano M. Regulatory dynamics of synthetic gene networks with positive
feedback. J Mol Biol 359(4): 1107-1124 (2006)
Makeyev EV, Zhang J, Carrasco MA, Maniatis T. The microRNA miR-124 promotes
neuronal differentiation by triggering brain-specific alternative-pre-mRNA splicing.
Mol. Cell 27(3): 435-448 (2007)
Martin CH, Nielsen DR, Solomon KV, Prather KL. Synthetic metabolism:
engineering biology at the protein and pathway scales. Chemistry and Biology 16,
277-286 (2009)
Mattick JS. RNA regulation: a new genetics? Nat Rev Genet. 5, 316-323 (2004)
133
McClean MN, Mody A, Broach JR, Ramanathan S. Cross-talk and decision making
in MAP kinase pathways. Nat Genet 39, 409-414 (2007)
Mettetal JT, Muzzey D, Gomez-Uribe C, van Oudenaarden A. The frequency
dependence of osmo-adaptation in Saccharomyces cerevisiae. Science 319, 482-484
(2008)
Moriya H, Shimizu-Yoshida Y, Kitano H. In vivo robustness analysis of cell division
cycle
genes
in
Saccharomyces
cerevisiae.
PLoS
Genet.
2(7):
el 11.
10.1371/journal.pgen.0020111(2006)
Muzzey D, Gomez-Uribe C, Mettetal J, van Oudenaarden A. (2009). A systems-level
analysis of perfect adaptation in yeast osmoregulation. Cell 138: 160-171 (2009)
Nelson DE, et al. Oscillations in NF-kB signaling control the dynamics of gene
expression. Science 306(5696): 704-708 (2004)
Odom DT, Dowell RD, Jacobsen ES, Gordon W, Danford TW, MacIsaac KD, Rolfe
PA, Conboy CM, Gifford DK, Fraenkel E. Tissue-specific transcriptional regulation
has diverged significantly between human and mouse. Nat. Genet. 39: 730-732
(2007)
O'Rourke S, Herskowitz I. Unique and redundant roles for HOG MAPK pathway
components as revealed by whole-genome expression analysis. Mol. Biol. Cell 15:
532-542 (2004)
Ozbudak EM, Thattai M, Kurtser I, Grossman AD, van Oudenaarden A. Regulation
of noise in the expression of a single gene. Nature Genetics 31, 69-73 (2002)
Ozbudak E, Thattai M, Lim HN, Shraiman BI, van Oudenaarden A. Multistability in
the lactose utilization network of Escherichia coli. Nature 427(6976): 737-740.
(2004)
134
Pedraza J, van Oudenaarden A. Noise propagation in gene networks. Science 307,
1965-1968 (2005)
Pokharel S and Beal PA. High-throughput screening for function adenosine to inosine
RNA editing systems. JACS Chem Biol 1, 761-765 (2006)
Pomerening JR, Sontag ED, Ferrell JE. Building a cell cycle oscillator: hysterisis and
bistability in the activation of Cdc2. Nat Cell Biol 5(4): 346-351 (2003)
Pomerening JR, Kim SY, Ferrell JE. Systems-level dissection of the cell-cycle
oscillator: bypassing positive feedback produces damped oscillations. Cell 122(4):
565-578 (2005)
Posas F, Wurgler-Murphy SM, Witten EA, Thai TC and Saito H. Yeast HOGi MAP
kinase cascade is regulated by a multistep phosphorelay mechanism in the SLN1YPD1-SSK1 "two-component" osmosensor. Cell 86: 865-875. (1996)
Proft M, Struhl K. MAP Kinase-mediated stress relief that precedes and regulates the
timing of transcriptional induction. Cell 118, 351-361 (2004)
Prud'homme BP, Gompel N, Carroll SB. Emerging principles of regulatory
evolution. Proc.NatL. Acad Sci. USA, 104 Suppl 1:8605-86012. (2007)
Raj A, Peskin CS, Tranchina D, Vargas DY, Tyagi S. Stochastic mRNA synthesis in
mammalian cells. PLoS Biology 4, e309 (2006)
Raj A and van Oudenaarden A. Nature, nurture, or chance: stochastic gene expression
and its consequences. Cell 135: 216-226 (2008)
Rajendran M, Ellington AD. Selection of fluorescent aptamer beacons that light up in
the presence of zinc. Anal. Bioanal. Chem. 390, 1067-1075 (2008).
Rapp M, Seppala S, Granseth E, von Heijne G. Emulating membrane protein
evolution by rational design. Science 315, 1282-1284 (2007)
135
Remenyi A, Good MC, Lim WA. Docking interactions in protein kinase and
phosphatase networks. Curr Opin in Struct Biol 16: 676-685 (2006)
Ro D, et al. Production of the antimalarial drug precursor artemisinc acid in
engineered yeast. Nature 440, 940-943 (2006)
Rosenfeld N, Elowitz MB, Alon U. Negative autoregulation speeds the response
times of transcription networks. J Mol Biol 323: 785-793 (2002)
Rosenfeld N, Young JW, Alon U, Swain PS, Elowitz MB. Gene regulation at the
single-cell level. Science 307(5717): 1962-1965 (2005)
Rothlisberger D, Khersonsky 0, Wollacott AM, Jian L, DeChancie J, Betker J,
Gallaher JL, Althoff EA, Zanghellini A, Dym 0, Albeck S, Houk KN, Tawfik DS,
Baker D. Kemp elimination catalysts by computational enzyme design. Nature 453,
190-195 (2008)
Segal E, Widom J. From DNA sequence to transcriptional behaviour: a quantitative
approach. Nat. Rev. Genetics 10, 443-456 (2009)
Selbach M, Schwanhausser B, Thierfelder N, Fang Z, Khanin R, Rajewsky N.
Widespread changes in protein synthesis induced by microRNAs. Nature 455: 58-63
(2008)
Setty Y, Mayo AE, Surette MG, Alon U. Detailed map of a cis-regluatory input
function. PNAS 100, 7702-7707 (2003)
Shimizu-Sato S, Huq E, Tepperman JM, Quail PH. A light-switchable gene promoter
system. Nat Biotech 20: 1041-1044 (2002)
Shinar G, Milo R, Martinez R, Alon U. Input-output robustness in simple bacteria
signaling systems. Proc. Natl. Acad. Sci. USA 104: 19931-19935 (2007)
136
Skerker JM, Perchuk BS, Siryaporn A, Lubin EA, Ashenberg 0, Goulian M, Laub
MT. Rewiring the specificity of two component signal transduction systems. Cell
133: 1043-1054 (2008)
Sluijter JPG, van Mil A, van Vliet P, Metz CGH, Liu J, Doevendans PA, Goumans
M-J. MicroRNA-1 and -499 regulate differentiation and proliferation in humanderived cardiomyocyte progenitor cells. Arteriosclerosis, Thrombosis, and Vascular
Biology (2010)
Song G, Zhang Y, Wang L. MicroRNA-206 targets notch3, activates apoptosis, and
inhibits tumor cell migration and focus formation. J Biol. Chem. 284: 31921-31927
(2009)
Sood P, Krek A, Zavolan M, Macino G, Rajewsky N. Cell-type-specific signatures of
microRNAs on target mRNA expression. Proc Natl Acad Sci 103(8): 2746-2751
(2006)
Stark A, Brennecke J, Bushati N, Russell RB, Cohen SM. Animal microRNAs confer
robustness to gene expression and have a significant impact on 3' UTR evolution.
Cell. 123(6):1133-46 (2005)
Steen EJ, Chan R, Prasad N, Myers S, Petzold CJ, Redding A, Ouellet M, Keasling
JD. Metabolic engineering of Saccharomyces cerevisiae for the production of nbutanol. Microb Cell Fact. 7:36 (2008)
Stricker J, Cookson S, Bennett MR, Mather WH, Tsimring LS, Hasty J A fast, robust
and tunable synthetic gene oscillator. Nature 456(7221): 516-519 (2008)
Suess B, Fink B, Berens C, Stentz R, Hillen WA. A theophylline responsive ribswitch
based on helix slipping controls gene expression in vivo. Nuc Acids Res 32: 16101614 (2004)
137
Swinbume IA, Miguez DG, Landgraf D, Silver PA. Intron length increases
oscillatory periods of gene expression in animal cells. Genes Dev 22(17): 2342-2346
(2008)
Tanay A, Regev A, Shamir R. Conservation and evolvability in regulatory network:
The evolution of ribosomal regulation in yeast. Proc. Natl. Acad. Sci. USA 102:
7203-7208 (2005)
Tanouchi Y, Pai A, You L. Decoding biological principles using gene circuits. Mol.
BioSyst. 5: 695-703 (2009)
Tatebayashi K, Takekawa M, Saito H. A docking site determining specificity of Pbs2
MAPKK for Ssk2/Ssk22 MAPKKKs in the yeast HOG pathway. EMBO J 22, 36243634 (2003)
Taylor RJ, Falconnet D, Niemisto A, Ramsey SA, Prinz S, Shmulevich I, Galitski T,
Hansen CL. Dynamic analysis of MAPK signaling using a high-throughput
microfluidic single-cell imaging platform. PNAS 106, 3758-3763 (2009)
Tigges M, Marquez-Lago TT, Stelling J, Fussenegger M. A tunable synthetic
mammalian oscillator. Nature 457(7227): 309-312 (2009)
Tsai TY, Choi YS, Ma W, Pomerening JR, Tang C, Ferrell JE. Robust, tunable
biological oscillations from interlinked positive and negative feedback loops. Science
321(5885): 126-129 (2008)
Tsang J, Zhu J, van Oudenaarden A. MicroRNA-mediated feedback and feedforward
loops are recurrent network motifs in mammals. Mol Cell 26(5): 753-767 (2007)
Tsong AE, Tuch BB, Li H, Johnson AD. Evolution of alternative transcriptional
circuits with identical logic. Nature 443, 415-420 (2006)
138
Ubersax JA, Ferrell JE. Mechanisms of specificity in protein phosphorylation. Nature
Reviews Molecular Cell Biology 8: 530-541 (2007)
von Dassow, G., Munro, E. M., Odell, G. M. The segment polarity network is a
robust developmental module. Nature 406:188-192 (2000)
Wagner A. Robustness and evolvability in living systems. Princeton University Press.
(2007).
Waks, Z. and Silver, PA. Engineering a synthetic dual organism system for hydrogen
production. Applied and Environmental Microbiology. 75:1867-75 (2009)
Werstruck G, Green MR. Controlling gene expression in living cells through small
molecule-RNA interactions. Science 282: 296-298 (1998)
Win MN, Smolke CD. Higher-order cellular information processing with synthetic
RNA devices. Science 322, 456-460 (2008)
Yeh BJ, Rutigliano RJ, Deb A, Bar-Sagi D, Lim WA. Rewiring cellular morphology
pathways with synthetic guanine nucleotide exchange factors. Nature 447: 596-600
(2007)
Yi R, Poy MN, Stoffel M, Fuchs E. A skin microRNA promotes differentiation by
repressing 'stemness'. Nature 452: 225-229 (2008).
You L, Cox RS III, Weiss R, Arnold FH. Programmed population control by cell-cell
communication and regulated killing. Nature 428(6985): 868-871 (2004)
Young DD and Deiters A. Photochemical control of biological processes. Org Biomol
Chem 5: 999-1005 (2007)
139