duel of the fates: the role of transcriptional circuits ... · duel of the fates: the role of...

Duel of the fates: the role of transcriptional circuits andnoise in CD4+ cellsDaniel Hebenstreit1, Andrew Deonarine1,M Madan Babu and Sarah A Teichmann

Available online at www.sciencedirect.com

CD4+ T cells play key roles in orchestrating adaptive immune

responses, and are a popular model for mammalian cell

differentiation. While immune regulation would seem to require

exactly adjusted mRNA and protein expression levels of key

factors, there is little evidence that this is strictly the case.

Stochastic gene expression and plasticity of cell types contrast

the apparent need for precision. Recent work has provided

insight into the magnitude of molecular noise, as well as the

relationship between noise, transcriptional circuits and

epigenetic modifications in a variety of cell types. These

processes and their interplay will also govern gene expression

patterns in the different CD4+ cell types, and the determination

of their cellular fates.

Address

MRC Laboratory of Molecular Biology, Hills Road, Cambridge CB2 0QH,

UK

Corresponding author: Teichmann, Sarah A ([email protected])1 These authors contributed equally to this work.

Current Opinion in Cell Biology 2012, 24:350–358

This review comes from a themed issue on

Nucleus and gene expression

Edited by Asifa Akhtar and Karla Neugebauer

Available online 11th April 2012

0955-0674/$ – see front matter

# 2012 Elsevier Ltd. All rights reserved.

DOI 10.1016/j.ceb.2012.03.007

IntroductionCells change phenotype when they specialize or adapt to

different environments. There are several well-known

systems that illustrate this phenomenon. Some examples

include the recently reported reprogramming of differ-

entiated cells into induced pluripotent stem cells (iPS) [1]

or the differentiation of CD4+ T cells [2]. These changes

or ‘switches’ in cellular phenotype are influenced by

changes in expression levels of key genes [3], and are

subject to molecular fluctuations [4��]. Such noise can be

classified as either intrinsic or extrinsic [5–8]. Intrinsic

noise is caused by gene-specific stochastic fluctuations in

abundances of molecules that are due to transcription,

translation, and related biochemical mechanisms, while

extrinsic noise originates from variations in the state or

concentration of other components in the cell [5]. As a


result, abundances of mRNAs and proteins of identical

genes can vary widely from cell to cell, constituting

‘noise,’ even within a homogenous cell population.

Noise propagation is strongly influenced by topologies of

regulatory networks [9,10], such as the transcriptional

circuits controlling cell fate. Examples include small

subnetworks, such as feedforward loops (FFLs) [11]

and feedback loops which can buffer or amplify noise,

depending on the parameters that influence the kinetics

of gene expression [12]. Recent theoretical studies have

tried to explore the structural requirements of network

controllability [13,14]. This question is important in the

context of ‘master regulators,’ which are essentially small

numbers or even single transcription factors (TFs) that

are able to ‘reprogram’ a cell type [15]. Besides the basic

TF–target gene interactions, other factors such as miRNA

[16], and epigenetic modifications [17,18] are part of

broader regulatory circuits and contribute to stabilization

or switching of phenotypes.

In this review we will focus on CD4+ T cells, which is a

good biological system for illustrating the relationships

between transcriptional circuits, expression level noise,

and epigenetics. CD4+ T cells form an important branch

of the adaptive immune system and function mainly by

secreting certain cytokines that regulate the behavior of

other cells, thus orchestrating the immune response [2].

Six major CD4+ T cell subtypes have been well charac-

terized to date: the naıve, T-helper 1 (Th1), Th2 and

Th17 cells, and induced and natural regulatory T cells

(iTreg and nTreg). Most of the (‘mature’-) subtypes are

derived from naıve CD4+ T cells, which enter a prolifer-

ation and differentiation process upon antigenic acti-

vation in the presence of cytokines (Figure 1a).

The phenotypic switches that CD4+ cells undergo upon

differentiation are linked to changes in epigenetic modi-

fications [19]. These switches are largely driven by TFs,

and illustrate the importance of master regulators, such as

Gata3 [20,21]. Recent studies on Th2 cells suggest that

phenotypic switches are mirrored by transcriptomic

switches [22�]. Protein expression levels of cytokines are

usually subject to large cell-to-cell variability, highlighting

the importance of noise in protein and mRNA expression.

To understand the interplay between these factors, sev-

eral questions need to be addressed: how do transcrip-

tional circuits mediate phenotypic switching? What are

www.sciencedirect.com

mailto:[email protected]

http://dx.doi.org/10.1016/j.ceb.2012.03.007

http://www.sciencedirect.com/science/journal/09550674

The role of transcriptional circuits and noise in CD4+ cells Hebenstreit et al. 351

Figure 1

(a)

Differentiation

Mature Cell ype

Master Regulator Gata3 Tbx21 Rorγt Foxp3

II17IfngII4II5

II13

Cd4

200

100

600

200

0

150

100

500

250

500

300

100

0

150

5060

2500

1000

0

4020

0

0 100 200

mRNAs/cell

mRNAs/cell

Fluorescence/cell (AU)

300 400 0 500 1000 1500 2000 2500 3000

0

0 100 200 300 400 0 10 20 30 40

0 200 400 600 800 0 10 20 30 40

II2

Tbx21

bivalent

II7r

Gata3

Freq

uenc

y

in T

h2 c

ells

Gata3 (Protein)

F = 83.9

F = 38.3

F = 28.4

F = 10.6

F = 4.00µ = 54.9

µ = 35.6

µ = 82.6

µ = 0.9

µ = 0.7

H3K4me3

H3K4me3

H3K4me3 H3K4me3

H3K27me3

Secreted Cytokines

(b)

Current Opinion in Cell Biology

Noise in CD4+ cells. (a) The CD4+ T cell system. Mature CD4+ subtypes are derived from naıve CD4+ cells through a differentiation process (except

nTreg, which mature inside the thymus). Master regulators and cytokines characteristic for the various subtypes are indicated. Plasticity is indicated by

arrows between the mature subtypes. (b) Skewed distributions of absolute mRNA numbers or protein expression (for Gata3 only, based on FACS

experiments) per single Th2 cells. Sample means (m) and Fano factors (F = s2/m) are given. The status of histone marks at genes was ascertained with

EpiChIP [52]. The data used were from Hebenstreit et al. [29��] (mRNA and protein) and Wei et al. [22�] (histone marks).

the roles of epigenetic modifications in this process? What

role does noise play: is it detrimental, tolerated, or even

beneficial? On the basis of the CD4+ T cell system, we

explore possible answers.


Phenotypic switches in CD4+ T cells aremediated by transcriptional circuitsThe difficulty of precisely defining cell types is well

illustrated by the distinct yet closely related subtypes


352 Nucleus and gene expression

of CD4+ T cells (Figure 1a). Though each of the six cell

types illustrated in Figure 1a is associated with a particu-

lar master regulator and defined by a specific cytokine

profile, there are many reports of intermediate cell types

as well as cell-type switching after differentiation (i.e.

plasticity), and reprogramming. For instance, mature Th2

cells can be induced to give rise to a mixed Th1/Th2

phenotype, Th17 cells can be transformed into Th1 cells,

or iTreg cells can be differentiated towards Th17 cells

[23] (Figure 1a).

The cytokines that are secreted by mature CD4+ T cells

have crucial, often opposing roles in regulating the immune

response. Thus, one would expect a tight regulation in

maintaining lineage identity. The plasticity in the differ-

entiation process as a whole is due to the complexity of the

transcriptional regulatory circuits, which generally include

many more components than just a single master regulator.

In Figure 2b, a small transcriptional circuit involving the

master regulator Gata3 is illustrated. The circuit consists of

overlapping FFLs which facilitate the differentiation of a

CD4+ naıve cell to the Th2 type [24], and represents a

small subnetwork in a much larger, more complex regu-

latory network involved in phenotypic switching and in

stabilizing the cellular state.

Differentiation of CD4+ cells is strongly linked to cellular

proliferation as many studies show a percentage increase

of cytokine-secreting T cells in successive cell gener-

ations [25,26]. Notably, the percentage of differentiated

cells added after each division is small and thus suggests

the involvement of a stochastic mechanism. Where does

this randomness come from and how can we reconcile it

with the precision of immune responses?

Sources of stochasticity and their relation tocircuit topologyT cell differentiation is regulated by TFs. It has become

clear in recent years that transcriptional regulatory net-

works are subject to intrinsic noise in the expression of

their components [27]. This has been shown by tech-

niques that focus on single cells rather than population

averages, such as single molecule RNA-FISH, which

measures absolute numbers of mRNAs per cell [8].

One of the surprising findings from these assays was that

the distributions of mRNA numbers per cell are usually

non-Poissonian, meaning that the variances are higher

than the mean levels of transcripts. This gives some

mechanistic insight into transcription by RNA polymer-

ase II (RNAPII) by hinting at a burst-like production of

transcripts, where a gene is transcribed multiple times in

short periods of time [28]. Our studies on Th2 cells [29��]confirm such highly skewed, ‘noisy’ distributions of tran-

script numbers per cell for five genes classically associated

with CD4+ cell biology, including Gata3 and CD4 itself

(Figure 1b). Skewed distributions are observed for


protein expression levels too (Figure 1b), suggesting that

either the mRNA expression noise is not buffered com-

pletely or downstream processes serve as additional

sources of noise. An interesting possibility is that cell-

to-cell variation in the expression levels of genes such as

Gata3 may simply reflect the heterogeneity in terms of

their recent signaling experience. Thus some genes could

be specifically programmed to be sensitive indicators of

different levels of signaling molecules in each cell.

Another contributor to the wide range of expression levels

between individuals in a cell population could be asym-

metric cell division. Polarized segregation of proteins has

been observed upon antigen-induced T cell division,

where activation and differentiation induced factors were

preferentially found in the daughter cell close to the

immunological synapse [30,31]. This included Tbx21

and may also affect other TFs and master regulators.

As a consequence of these sources of noise, there may not

be a simple, discrete transcriptional ‘switch.’ This may

vary from cell to cell depending on the number of copies

of a set of signaling molecules, including TFs. As a result,

statistical parameters such as the average burst size or the

average waiting time may be altered upon transcriptional

activation of RNAPII. Importantly, burst-like transcrip-

tion demonstrates that transcription is an intrinsically

noisy process per se, and in individual cells this phenom-

enon can result in large deviations in transcript numbers

from the population average. In principle, random fluctu-

ations give rise to different transcription rates in different

cells over time, even if the cells started under identical

conditions. For CD4+ cells, the large variance in the

distributions shown in Figure 1b suggests that this may

be relevant to phenotypic switching, as it means that a

certain fraction of cells will have high numbers of mRNAs

from genes that are lowly expressed, and vice versa.

The presence of large cell-to-cell variation may indicate

that noise is somewhat inevitable. This might be due to

the fact that noise suppression is a biologically expensive

process: a recent study showed that suppression only

occurs with the fourth root of the number of control

factors in a transcriptional circuit [32��]. However, many

investigations in other cell types or systems have shown

that stochastic transcriptional switches can even be

advantageous over deterministic ones. For instance, sto-

chastically switching phenotypes or noisy transcriptional

circuits provide efficient mechanisms for adapting to

fluctuating or unpredictable environments [33–35]. It

has also been shown that FFL circuits that incorporate

miRNA can dampen noise [36], while certain FFL archi-

tectures have specific noise profiles [37] and may be

selected for these profiles, thereby facilitating the

sampling of multiple phenotypes [38].

While it is uncertain whether such phenomena play a role

in T cell biology, it is clear that the transcriptional



Figure 2

Naive

0.15

0.10

% G

enes

0.05

0.00

-10 -5 0

Log(expression value)

5 10

Cel

l Fat

eTr

ansc

ripto

me

Net

wor

kE

pige

netic

sNoise

2σμ

Th2

ExpressionClasses:

NE - none

LE - low

HE - high

Intermediate

Intermediate(3030)

“Off”

TranscriptionalRegulation

(c)

(a)

(b)JunB

Dec2

Log(expression value)-10

0.00

0.05

% G

enes

0.10

0.15

-5 0 5 10

II4

II5Gata3

LE(2781)

NE(6783)

JunB

Gata3

Switch

H3K4me3

OpenBivalentClosed “On”

“Off”

“On”

(9413)HE

H3K27me3

FFL

2σμ

2σμ


Switches in cellular fate as reflected in the transcriptome, regulatory network, and epigenetic state of the cell, each of which is influenced by noise

(indicated by the Fano factor s2/m). (a) A density plot illustrating highly and lowly expressed transcripts (HE and LE, respectively), transcripts with no

detectable expression (NE) and transcripts with intermediate levels (numbers of genes are in brackets). (b) Transcriptional feedforward circuits involved

in the regulation of key cytokines that facilitate the ‘switch’ to Th2 from naıve cell [24]. Three feedforward loops are illustrated: Dec2-Gata3-Il4, Dec2-

Gata3-Il5, and Dec2-Jun2-Il4. (c) Histone modifications also play a role in mediating phenotypic switches, with activating histone medications (e.g.

H3K4me3) being associated with an open chromatin state and HE transcripts, while repressive modifications (e.g. H3K27me3) are associated with LE

transcripts. Bivalent chromatin is associated with transcripts that lie between HE and LE in expression level (intermediate).

www.sciencedirect.com Current Opinion in Cell Biology 2012, 24:350–358


regulatory networks of CD4+ differentiation are subject

to intrinsic noise and the fundamental limit of noise

suppression mentioned above.

The role of epigenetic switches in CD4+ cellsOne mechanism that could have evolved in order to permit

a cost-efficient control of such transcriptional noise could

be the system of epigenetic modifications. Genes that

become differentially expressed during Th differentiation

are accompanied by changes in the chromatin landscape

[19]. For instance, the chromatin at subtype-specific loci

becomes relaxed, as illustrated by DNaseI hypersensitive

regions. Other subtype-specific genes are marked with

repressive histone modifications in cell types not expres-

sing them [39]. Mariani et al. showed that in Th2 cells,

stochastic Il4 expression patterns can be explained com-

putationally using a two-state model, where promoter

activation/inactivation states are explained in terms of

chromatin dynamics (opening and closing due to activating

or suppressing histone modifications). Here, different

populations of cells are created stochastically over gener-

ations due to chromatin dynamics [6,40�]. This work

provides support for a link between noise in gene expres-

sion levels and histone modification levels.

Similarly, our recent work relates gene expression with

histone modification levels, confirming such a two-state

model. In this study, we identified groups of lowly

expressed (LE) and highly expressed (HE) genes in

Th2 cells based on the global distribution of gene expres-

sion levels measured by RNA-sequencing (RNA-seq)

[29��], which is illustrated in Figure 2a. Here, we annotated

another subset of genes (derived from the LE group) called

not expressed (NE). NE genes are defined as genes with no

reads mapping to that particular transcript, which is de-

pendent on sequencing depth. This concept is similar to

the three expression classes of mRNA previously identified

by Hastie and Bishop [41]. While the HE genes featured

the activating histone mark H3K9/14 acetylation at their

promoters, the LE genes did not. The existence of the

lowly expressed, tissue-specific LE group of genes

suggests that the cell robustly tolerates a certain amount

of background expression, representing a type of ‘tran-

scriptional noise.’ This is often referred to as leaky expres-

sion and might be due to a tradeoff between the cost of

(nearly) completely repressed transcription and the

adverse consequences of stochastic expression [42,43].

The leaky expression reflects the inactive promoter state

in the two-state model, while HE transcripts are more

representative of the active promoter state.

In Figure 3, we analyzed the chromosomal organization of

NE, LE, and HE genes. As illustrated in Figure 3a, the

physical distances between adjacent HE and LE genes

were found to be less than between HE and NE genes.

This suggests that expression in LE genes may be due to

their proximity to HE genes. Similarly, as depicted in


Figure 3b, both HE and NE genes tend to form hom-

ogenous multigene clusters more often than expected by

chance. In contrast, LE genes usually occur alone. These

findings support a genomic organization illustrated in

Figure 3c, with LE genes close to clusters of HE genes,

and clusters of NE genes further away from HE gene

clusters. In this model, expression from neighboring HE

genes leads to leaky expression of neighboring LE genes

via some form of chromatin state spreading.

An interesting aspect of histone modifications is the

existence of ‘bivalent’ genes, which feature both activat-

ing and repressive marks. In mammalian cells, there is

increasing evidence from sequential ChIP and other

experiments that these marks are present simultaneously

at the same allele [44–48]. In Drosophila and Xenopus, it is

unclear whether this is the case [49–51]. In mouse

T-helper cells, Wei et al. observed such bivalent marks

at master regulator loci [22]. The authors argued that such

bivalent marks could poise them for expression. This

could explain the plasticity of these cell types as well

as why perfectly ‘pure’ cultures of single CD4+ subtypes

are difficult to obtain. Using CD4+ microarray expression

and H3K4me3/H3K27me3 histone modification data

(analyzed with Epichip [52]) from Wei et al., we found

that genes with bivalent marks have a higher probability

of differential expression upon maturation than genes

with the H3K27me3 mark or no modification (see

Figure 3d). They are also enriched in TFs relative to

the genomic background. Whether these bivalent genes

also have a specific noise profile is yet to be elucidated.

The functional relevance of noise intranscriptional regulatory networksThe question remains concerning how much of the

observed stochasticity in CD4+ gene expression is toler-

ated and how much is beneficial. With regard to the

beneficial aspects of stochastic expression, Guo et al.argued that probabilistic expression of the Th2 cytokine

Il4 by a fraction of cells is a more flexible mechanism of

immune regulation than fine-tuning of the exact Il4

amounts secreted by a homogenous cell population

[40�,53].

Another recently published study on CD8+ cells, closely

related to CD4+ cells, demonstrates how a certain amount

of ‘controlled’ randomness in protein expression gener-

ates useful variation in the antigen responsiveness [54�] of

the cell. This system is based on a combination of positive

and negative feedback loops and demonstrates that the

architecture of the regulatory circuits can have a profound

impact on the existence of stable states and/or extent of

noise. In line with this, a recent analytical study showed

that feedback systems provided by cytokine secretion and

stimulation, including downstream transcriptional inter-

actions, are sufficient for yielding mathematically stable

states that resemble cell types among CD4+ cells. The



Figure 3

(a) 0.2 80

70

60

50

40

% o

f gen

es

30

20

10

01 2 3 4 5 6 7 8 9

Group Size10 11 12 13 14 15

0.15

0.1

Den

sity

0.05

0

3’ 5’

5 10 15 20

20

Bivalent

HELENE

% HE

% Random HE

% Random LE

% Random NE

% LE

% NE

NE-HE

NE-HE

NE-HE

LE-HE

LE-HE

LE-HE mean

NE-HE mean

NE-HE median

LE-HE median

LE-HE

H3K27me3

H3K4me3

None

40 60% of genes

80

% of all mouse genes with mark innaïve cells

% of mouse genes with mark that aredifferentially expressed from naïve toanother cell type (excluding nTreg)

(# TranscriptionFactors)

(234)(162)

(302)(145)

(517)(410)

(114)(75)

Log2(Physical Distance)

(b)

(c)

(d)


The role of highly and lowly expressed transcripts. (a) The physical distances (defined as the shortest genomic distance between genes in nucleotides)

between adjacent pairs of highly expressed (HE) and not expressed (NE)/lowly expressed (LE) genes in mouse Th2 cells. RNA-seq derived gene

expression levels from Hebenstreit et al. [29��] were classified as LE or HE; Tophat [63] was also used for junction mapping. In general, LE genes are

found closer to HE genes than NE genes. The difference between the LE–HE versus NE–HE distributions is statistically significant (Wilcoxon rank sum

test, P < 10�4). This observation may support the concept of leaky expression, or the ripple effect [42,43]. (b) The percentage of all NE, LE, and HE

genes (with intermediate genes between LE and HE being classified as HE) in the genome that occur in a particular cluster size, compared to the

cluster sizes obtained by averaging values across 10 000 randomized sequences of NE, LE, and HE genes. HE and NE genes are significantly more

likely to form larger clusters than expected by chance (P < 10�4 estimated from simulation). (c) A model describing the organization of NE, LE, and HE

genes according to their distances and clustering. (d) Data from Wei et al. [22�] were normalized using MAS5 [64], and differential expression between

naıve and Th1, Th2, Th17, and iTreg was calculated using Limma [65] (an adjusted P-value <0.05 was used to determine significance). More than half

of the genes with bivalent histone marks are differentially expressed upon CD4+ cell maturation. As well, genes with bivalent modifications were

enriched in transcription factors (annotated by DBD [66]) relative to the whole mouse genome (Fisher’s exact test, P < 10�4).

model also allows for plasticity and gives rise to inter-

mediate cell types [55].

Experimental studies have indeed identified various

circuits involved in CD4+ cell differentiation, which


probably have important roles in regulating transcrip-

tional and thus phenotypic noise. Djuretic et al. noted

that in Th1 cells, the TF Tbx21 induces another TF

Runx3, and together these molecules repress Il4 by

binding to its silencer [56] and activate Ifng by binding



to its promoter. Hence, Tbx21 and Runx3 form FFLs

with Il4 and Ifng. In Treg cells, it has been shown by

Bruno et al. that the TF Runx can promote the expression

of the master regulator Foxp3 [57], and both of these TFs

regulate target cytokines such as Il2 thereby forming a

coherent FFL.

FFL circuits also play an important role in phenotype

determination in biological systems other than CD4+

cells. Foxp3 is known to suppress the oncogene Satb1.

Additionally, it was recently found that Foxp3 also forms

coherent FFLs involving miRNAs that bind to the 30-UTR of Satb1 (miR-7 and miR-155) [58] in breast tissue.

Satb1 also plays an important role in CD4+ cell biology,

particularly cytokine regulation. As noted previously, the

Th2-specific cytokines Il4 and Il5 are regulated by FFL

circuits (see Figure 2) [24]. At the epigenetic level, Satb1

remodels chromatin to create a 200 kb transcriptionally

active region associated with Stat6 and Gata3. These TFs

physically bridge this region with the Th2 ‘cytokine

locus,’ which contains the genes Il4, Il5, and Il13. Cai

et al. showed that this Satb1 mediated remodeling, in

addition to the presence of Stat6 and Gata3, is essential to

cytokine induction [59].

Finally, the regulation of cytokine receptors plays an

important role in the circuits that stabilize Th lineages.

This is supported by the aforementioned modeling

approach [55]. A well-known biological example includes

the IL-4 receptor a-chain (Il4ra), which is upregulated by

IL-4 via Stat6 [60,61] and thus constitutes a Th2-reinfor-

cing positive feedback loop.

ConclusionsStudies on CD4+ cells and other systems have high-

lighted the importance of transcriptional circuits and

other mechanisms in regulating noise and mediating

phenotypic switches. The strongly skewed mRNA

distributions suggest that the sharpness of lineage iden-

tities is subject to fundamental limitations. Some

examples indicate that there can be benefits from tol-

erating a certain amount of stochasticity, but in general

the cell employs multiple mechanisms to control noise.

These include the topologies of transcriptional networks

and circuits, and epigenetic mechanisms, as exemplified

by the discovery of the LE/HE transcript expression

classes. Future studies will provide insight into the role

of transcriptional circuits and epigenetics in broader

regulatory networks, how these networks affect noise,

and the probabilistic nature of phenotypic switches

and lineage stabilities. This may eventually lead to

more accurate, probabilistic definitions of these switches

and stabilities, paving the way for new pharmaceutical

strategies [62], and improving how we confront the

‘dueling fates’ of pathological versus normal cellular

phenotypes.


AcknowledgementsWe thank Guilhem Chalancon, Jesper Nissen and Valentina Proserpio forhelpful suggestions concerning this manuscript. The authors acknowledgethe Lister Institute (UK), the Medical Research Council (UK) grantreference nos. U105161047 and U105185859, and the Clinician InvestigatorProgram at the University of British Columbia (Canada) for their support.

References and recommended readingPapers of particular interest, published within the period of review,have been highlighted as:

� of special interest

�� of outstanding interest

1. Takahashi K, Yamanaka S: Induction of pluripotent stem cellsfrom mouse embryonic and adult fibroblast cultures bydefined factors. Cell 2006, 126:663-676.

2. Zhu J, Yamane H, Paul WE: Differentiation of effector CD4 T cellpopulations (*). Annu Rev Immunol 2010, 28:445-489.

3. Kalmar T, Lim C, Hayward P, Munoz-Descalzo S, Nichols J,Garcia-Ojalvo J, Martinez Arias A: Regulated fluctuations innanog expression mediate cell fate decisions in embryonicstem cells. PLoS Biol 2009, 7:e1000149.

4.��

Trott J, Surani A, Babu MM, Martinez-Arias A: Dissectingensemble networks in ES cell populations revealsmicro-heterogeneity underlying pluripotency. Mol BioSyst2012, 8:744-752.

In this work, it was found that individual mouse ES cells maintainedpluripotency by utilizing different transcription factor subnetworks. Thiswas discovered using single-cell gene expression profiling.

5. Swain PS, Elowitz MB, Siggia ED: Intrinsic and extrinsiccontributions to stochasticity in gene expression. Proc NatlAcad Sci U S A 2002, 99:12795-12800.

6. Miller-Jensen K, Dey SS, Schaffer DV, Arkin AP: Varyingvirulence: epigenetic control of expression noise and diseaseprocesses. Trends Biotechnol 2011, 29:517-525.

7. Silva-Rocha R, de Lorenzo V: Noise and robustness inprokaryotic regulatory networks. Annu Rev Microbiol 2010,64:257-275.

8. Raj A, van Oudenaarden A: Single-molecule approaches tostochastic gene expression. Annu Rev Biophys 2009,38:255-270.

9. Bruggeman FJ, Bluthgen N, Westerhoff HV: Noise managementby molecular networks. PLoS Comput Biol 2009, 5:e1000506.

10. Chalancon G, Ravarani C, Balaji S, Martinez Arias A, Aravind L,Jothi R, Babu MM: Interplay between gene expression noiseand regulatory network architecture. Trends Genet 2012 doi:10.1016/j.tig.2012.01.006.

11. Mangan S, Alon U: Structure and function of the feed-forwardloop network motif. Proc Natl Acad Sci U S A 2003,100:11980-11985.

12. Rodrigo G, Elena SF: Structural discrimination of robustness intranscriptional feedforward loops for pattern formation. PLoSOne 2011, 6:e16904.

13. Liu YY, Slotine JJ, Barabasi AL: Controllability of complexnetworks. Nature 2011, 473:167-173.

14. Muller FJ, Schuppert A: Few inputs can reprogram biologicalnetworks. Nature 2011, 478:E4 (discussion E4–5).

15. Mullen AC, Orlando DA, Newman JJ, Loven J, Kumar RM,Bilodeau S, Reddy J, Guenther MG, Dekoter RP, Young RA:Master transcription factors determine cell-type-specificresponses to TGF-beta signaling. Cell 2011, 147:565-576.

16. Bartel DP: MicroRNAs: target recognition and regulatoryfunctions. Cell 2009, 136:215-233.

17. Goldberg AD, Allis CD, Bernstein E: Epigenetics: a landscapetakes shape. Cell 2007, 128:635-638.


http://dx.doi.org/10.1016/j.tig.2012.01.006

http://dx.doi.org/10.1016/j.tig.2012.01.006


18. Natoli G: Maintaining cell identity through global control ofgenomic organization. Immunity 2010, 33:12-24.

19. Wilson CB, Rowell E, Sekimata M: Epigenetic control ofT-helper-cell differentiation. Nat Rev Immunol 2009,9:91-105.

20. Zhang DH, Cohn L, Ray P, Bottomly K, Ray A: Transcriptionfactor GATA-3 is differentially expressed in murine Th1 andTh2 cells and controls Th2-specific expression of theinterleukin-5 gene. J Biol Chem 1997, 272:21597-21603.

21. Zheng W, Flavell RA: The transcription factor GATA-3 isnecessary and sufficient for Th2 cytokine gene expression inCD4 T cells. Cell 1997, 89:587-596.

22.�

Wei G, Wei L, Zhu J, Zang C, Hu-Li J, Yao Z, Cui K, Kanno Y,Roh TY, Watford WT et al.: Global mapping of H3K4me3 andH3K27me3 reveals specificity and plasticity in lineage fatedetermination of differentiating CD4+ T cells. Immunity 2009,30:155-167.

This article provides ChIP-sequencing data for two histone modificationsin all major CD4+ cell types.

23. Murphy KM, Stockinger B: Effector T cell plasticity: flexibility inthe face of changing circumstances. Nat Immunol 2010,11:674-680.

24. Yang XO, Angkasekwinai P, Zhu J, Peng J, Liu Z, Nurieva R, Liu X,Chung Y, Chang SH, Sun B et al.: Requirement for the basichelix-loop-helix transcription factor Dec2 in initial TH2 lineagecommitment. Nat Immunol 2009, 10:1260-1266.

25. Bird JJ, Brown DR, Mullen AC, Moskowitz NH, Mahowald MA,Sider JR, Gajewski TF, Wang CR, Reiner SL: Helper T celldifferentiation is controlled by the cell cycle. Immunity 1998,9:229-237.

26. Gett AV, Hodgkin PD: Cell division regulates the T cell cytokinerepertoire, revealing a mechanism underlying immune classregulation. Proc Natl Acad Sci U S A 1998, 95:9488-9493.

27. Balazsi G, van Oudenaarden A, Collins JJ: Cellular decisionmaking and biological noise: from microbes to mammals. Cell2011, 144:910-925.

28. Suter DM, Molina N, Gatfield D, Schneider K, Schibler U, Naef F:Mammalian genes are transcribed with widely differentbursting kinetics. Science 2011, 332:472-474.

29.��

Hebenstreit D, Fang M, Gu M, Charoensawan V, vanOudenaarden A, Teichmann SA: RNA sequencing reveals twomajor classes of gene expression levels in metazoan cells. MolSyst Biol 2011, 7:497.

This article is the first to contain RNA-seq and single molecule RNA-FISHdata for mouse Th2 cells. It describes two classes of mRNA expressionlevels which are distinguished by histone marks at the promoters of thegenes.

30. Chang J, Palanivel V, Kinjyo I, Schambach F, Intlekofer A,Banerjee A, Longworth S, Vinup K, Mrass P, Oliaro J et al.:Asymmetric T lymphocyte division in the initiation of adaptiveimmune responses. Science 2007, 315:1687-1691.

31. Chang J, Ciocca M, Kinjyo I, Palanivel V, McClurkin C, Dejong C,Mooney E, Kim J, Steinel N, Oliaro J et al.: Asymmetricproteasome segregation as a mechanism for unequalpartitioning of the transcription factor T-bet during Tlymphocyte division. Immunity 2011, 34:492-504.

32.��

Lestas I, Vinnicombe G, Paulsson J: Fundamental limits onthe suppression of molecular fluctuations. Nature 2010,467:174-178.

This is an elegant theoretical study providing mathematical proofs for theextremely high costs of limiting expression noise in the cell.

33. Acar M, Mettetal JT, van Oudenaarden A: Stochastic switchingas a survival strategy in fluctuating environments. Nat Genet2008, 40:471-475.

34. Cagatay T, Turcotte M, Elowitz MB, Garcia-Ojalvo J, Suel GM:Architecture-dependent noise discriminates functionallyanalogous differentiation circuits. Cell 2009, 139:512-522.

35. Jothi R, Balaji S, Wuster A, Grochow JA, Gsponer J, Przytycka TM,Aravind L, Babu MM: Genomic analysis reveals a tight link


between transcription factor dynamics and regulatorynetwork architecture. Mol Syst Biol 2009, 5:294.

36. Osella M, Bosia C, Cora D, Caselle M: The role of incoherentmicroRNA-mediated feedforward loops in noise buffering.PLoS Comput Biol 2011, 7:e1001101.

37. Ghosh B, Karmakar R, Bose I: Noise characteristics of feedforward loops. Phys Biol 2005, 2:36-45.

38. Kittisopikul M, Suel GM: Biological role of noise encoded in agenetic network motif. Proc Natl Acad Sci U S A 2010,107:13300-13305.

39. Hirahara K, Vahedi G, Ghoreschi K, Yang XP, Nakayamada S,Kanno Y, O’Shea JJ, Laurence A: Helper T-cell differentiationand plasticity: insights from epigenetics. Immunology 2011,134:235-245.

40.�

Mariani L, Schulz EG, Lexberg MH, Helmstetter C, Radbruch A,Lohning M, Hofer T: Short-term memory in gene inductionreveals the regulatory principle behind stochastic IL-4expression. Mol Syst Biol 2010, 6:359.

In this study, a combination of experimental and theoretical analysesreveals how stochastic mechanisms acting at the chromatin level can giverise to population heterogeneities of cytokine expression.

41. Hastie ND, Bishop JO: The expression of three abundanceclasses of messenger RNA in mouse tissues. Cell 1976,9:761-774.

42. Ebisuya M, Yamamoto T, Nakajima M, Nishida E: Ripples fromneighbouring transcription. Nat Cell Biol 2008, 10:1106-1113.

43. De S, Babu MM: Genomic neighbourhood and the regulation ofgene expression. Curr Opin Cell Biol 2010, 22:326-333.

44. Bernstein B, Mikkelsen T, Xie X, Kamal M, Huebert D, Cuff J, Fry B,Meissner A, Wernig M, Plath K et al.: A bivalent chromatinstructure marks key developmental genes in embryonic stemcells. Cell 2006, 125:315-326.

45. Roh T, Cuddapah S, Cui K, Zhao K: The genomic landscape ofhistone modifications in human T cells. Proc Natl Acad Sci U S A2006, 103:15782-15787.

46. Pan G, Tian S, Nie J, Yang C, Ruotti V, Wei H, Jonsdottir GA,Stewart R, Thomson JA: Whole-genome analysis of histone H3lysine 4 and lysine 27 methylation in human embryonic stemcells. Cell Stem Cell 2007, 1:299-312.

47. Stock J, Giadrossi S, Casanova M, Brookes E, Vidal M, Koseki H,Brockdorff N, Fisher A, Pombo A: Ring1-mediated ubiquitinationof H2A restrains poised RNA polymerase II at bivalent genes inmouse ES cells. Nat Cell Biol 2007, 9:1428-1435.

48. Brookes E, de Santiago I, Hebenstreit D, Morris K, Carroll T, Xie S,Stock J, Heidemann M, Eick D, Nozaki N et al.: Polycombassociates genome-wide with a specific RNA polymerase IIvariant, and regulates metabolic genes in ESCs. Cell Stem Cell2012, 10:157-170.

49. Schuettengruber B, Ganapathi M, Leblanc B, Portoso M,Jaschek R, Tolhuis B, van Lohuizen M, Tanay AGC: Functionalanatomy of polycomb and trithorax chromatin landscapes inDrosophila embryos. PLoS Biol 2009, 7:e13.

50. Gan Q, Schones D, Ho Eun S, Wei G, Cui K, Zhao K, Chen X:Monovalent and unpoised status of most genes inundifferentiated cell-enriched Drosophila testis. Genome Biol2010, 11:R42.

51. Akkers R, van Herringen S, Jacobi U, Janssen-Megens E,Francoijs K, Stunnenberg H, Vennstra G: A hierarchy of H3K4me3and H3K27me3 acquisition in spatial gene regulation inXenopus embryos. Dev Cell 2009, 17:425-434.

52. Hebenstreit D, Gu M, Haider S, Turner DJ, Lio P, Teichmann SA:EpiChIP: gene-by-gene quantification of epigeneticmodification levels. Nucleic Acids Res 2011, 39:e27.

53. Guo L, Hu-Li J, Paul WE: Probabilistic regulation of IL-4production in Th2 cells: accessibility at the Il4 locus. Immunity2004, 20:193-203.

54.�

Feinerman O, Veiga J, Dorfman JR, Germain RN, Altan-Bonnet G:Variability and robustness in T cell activation from



regulated heterogeneity in protein levels. Science 2008,321:1081-1084.

This article shows how ‘controlled stochasticity’ can benefit the cell bygenerating useful variation during T cell activation.

55. Naldi A, Carneiro J, Chaouiya C, Thieffry D: Diversity andplasticity of Th cell types predicted from regulatory networkmodelling. PLoS Comput Biol 2010, 6:e1000912.

56. Djuretic IM, Levanon D, Negreanu V, Groner Y, Rao A, Ansel KM:Transcription factors T-bet and Runx3 cooperate to activateIfng and silence Il4 in T helper type 1 cells. Nat Immunol 2007,8:145-153.

57. Bruno L, Mazzarella L, Hoogenkamp M, Hertweck A, Cobb BS,Sauer S, Hadjur S, Leleu M, Naoe Y, Telfer JC et al.: Runx proteinsregulate Foxp3 expression. J Exp Med 2009, 206:2329-2337.

58. McInnes N, Sadlon TJ, Brown CY, Pederson S, Beyer M,Schultze JL, McColl S, Goodall GJ, Barry SC: FOXP3 and FOXP3-regulated microRNAs suppress SATB1 in breast cancer cells.Oncogene 2011:1-10.

59. Cai S, Lee CC, Kohwi-Shigematsu T: SATB1 packages denselylooped, transcriptionally active chromatin for coordinatedexpression of cytokine genes. Nat Genet 2006, 38:1278-1288.


60. Wei L, Vahedi G, Sun H, Watford W, Takatori H, Ramos H,Takahashi H, Liang J, Gutierrez-Cruz G, Zang C et al.: Discreteroles of STAT4 and STAT6 transcription factors in tuningepigenetic modifications and transcription during T helper celldifferentiation. Immunity 2010, 32:840-851.

61. Kotanides H, Reich N: Interleukin-4-induced STAT6 recognizesand activates a target site in the promoter of the interleukin-4receptor gene. J Biol Chem 1996, 271:25555-25561.

62. Amit I, Regev A, Hacohen N: Strategies to discover regulatorycircuits of the mammalian immune system. Nat Rev Immunol2011, 11:873-880.

63. Trapnell C, Pachter L, Salzberg SL: TopHat: discovering splicejunctions with RNA-Seq. Bioinformatics 2009, 25:1105-1111.

64. Hubbell E, Liu WM, Mei R: Robust estimators for expressionanalysis. Bioinformatics 2002, 18:1585-1592.

65. Smyth GK: Linear models and empirical Bayes methods forassessing differential expression in microarray experiments.Stat Appl Genet Mol Biol 2004, 3 Article3.

66. Kummerfeld SK, Teichmann SA: DBD: a transcription factorprediction database. Nucleic Acids Res 2006, 34:D74-D81.


duel of the fates: the role of transcriptional circuits ... · duel of the fates: the role of...

Documents