ava i l ab l e a t www.sc i enced i rec t . com

www.e l sev i e r. com/ loca te / sc r

Stem Cell Research (2008) 1, 157–168

REVIEW

Toward a Rosetta stone for the stem cell genome:Stochastic gene expression, network architecture,and external influencesJulianne D. Halley a,⁎, David A. Winkler a,b, Frank R. Burden a,b,c

a CSIRO Molecular and Health Technologies, Private Bag 10, Clayton South MDC 3169, Australiab School of Chemistry, Monash University, VIC 3800, Australiac SciMetrics, Fitzroy, VIC 3065, Australia

Received 16 January 2008; received in revised form 17 March 2008; accepted 21 March 2008

Abstract We review literature relating to three types of factors known to influence stem cell behavior. These factors arestochastic gene expression, regulatory network architecture, and the influence of external signals, such as thoseemanating from the niche. Although these factors are considered separately, their shared evolutionary history necessitatesintegration. Stochastic gene expression pervades network components; network architecture controls, modulates, or exploitsthis noise while performing additional computation; and such complexity also interplays with factors external to cells.Adequate understanding of each of these components, and how they interact, will lead to a conceptual model of the stem cellregulatory system that can be used to drive hypothesis-driven research and facilitate interpretation of experimental data.

© 2008 Elsevier B.V. All rights reserved.

Contents

Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157Stochastic gene expression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

Architecture of stem cell regulatory networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160Key regulatory genes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160Binary vs complex lineage specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162External factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165

158

⁎ Corresponding author. Fax: +61 3 9545 2561.E-mail address: julianne.halley@csiro.au (J.D. Halley).

1873-5061/$ – see front matter © 2008 Elsevier B.V. All rights reserved.doi:10.1016/j.scr.2008.03.001

158 J.D. Halley et al.

Introduction

One of the major challenges in representing biologicalcomplexity is to characterize key functional relationshipsin a transparent manner (Hu et al., 2007). Models should helpus to understand, in an intuitive way, the properties ofbiological systems (Tomlin and Axelrod, 2007) and facilitaterapid exploration of the properties of a theory (Laforgeet al., 2005). Successful models are those that help to answerspecific questions about biological processes, such as, is themodel a plausible representation of the true system and, ifnot, how should it be changed to make it plausible? (Tomlinand Axelrod, 2007). As the study of information processing incells shifts from that of single components or signalingpathways to increasingly complex networks, mathematicalmodels become indispensable tools (Bornholdt, 2005). Thedaunting challenge of developing useful conceptual modelsfor how biological systems work is also drawing interest(Lauffenburger, 2000). Although detailed predictive modelsof biological networks are not yet within reach, with timesuch models have the potential to revolutionize our under-standing of cell biology and complex diseases (Bornholdt,2005), because complex functions, traits, and pathologies arerarely caused by single genes, but are instead context-dependent entities to which single genes make only partialcontributions (Hartwell et al., 1999; Qu et al., 2007; Weath-erall, 2001).

Despite extensive experimental work that has attemptedto unravel the complex networks governing stem cell lineagespecification, the general principles underlying these reg-ulatory mechanisms remain unclear (Roeder and Glauche,2006). Although it is clear that lineage specification isinfluenced by multiple transcription factors and their proteinproducts, it is unclear whether low level expression of sometranscription factors is an artifact of leaky transcriptionmachinery or an active mechanism of lineage specification(Akashi, 2005; Akashi et al., 2003; Cantor and Orkin, 2001,2002; Cross and Enver, 1997; Cross et al., 1994; Davey andZandstra, 2004; Enver and Greaves, 1998; Hu et al., 1997;Huang et al., 2007; Orkin, 2000; Shivdasani and Orkin, 1996).The low-level expression of transcription factors in stemcells is sometimes referred to as “priming” or “multilineagepriming.” During lineage specification, the balance of thetranscription factors is upset, leading to an up-regulation ofsome transcription factors specific to a particular lineageand a down-regulation of others (Orkin, 2000; Roeder andGlauche, 2006). Hence, the transcription factor networkcontrolling stem cell behavior appears capable of switch-likebehavior, changing from apparently nonspecific coexpressionof several transcription factors to a specific gene expressionpattern associated with a differentiated cell type (Roederand Glauche, 2006). In other words, lineage commitmentinvolves consolidation of a specific gene expression programwith the concurrent silencing of alternatives, a resolution ofcomplexity to relative simplicity (Bruno et al., 2004; Crosset al., 1997).

Here, we review three prominent classes of factorsthought to impact stem cell decision-making processes.Although the review cannot be exhaustive (the literature onstem cells grows at a staggering rate indeed), it is hoped thatour distillation of this knowledge to build a conceptual modelwill lead to a computational model of stem cell decision

making that can be fruitfully compared with stem cellbiochemical experiments. Our particular approach to thedifficult problems that stem cell biologists face arises froman interdisciplinary background in molecular science, animalsocial behavior, and the emergence and evolution ofcollective intelligence in distributed systems, and fromneural networks as tools for nonlinear problem solving. Thisbackground in complex systems science illuminates problemsfrom a different perspective and has already led to newinsights and different ways of looking at stem cell complex-ity. As the dynamics of regulatory systems are not readilyextrapolated from the behavior of individual genes or evengenemotifs (Swiers et al., 2006), higher level conceptual andcomputational approaches are required. The three topics wereview here are: (1) stochastic gene expression, (2)regulatory network architecture, and (3) external influenceson stem cell fate. A synthesis of this review in the form of ahigher-level conceptual framework and a computationalmodel is the focus of a subsequent paper.

Stochastic gene expression

It has become clear that stochastic interactions play animportant role in both gene expression and cellular differ-entiation (Laforge et al., 2005). In small cells like bacteria orbudding yeast, intracellular regulators are often present insmall copy numbers, meaning that some cells will possess no,or very few, molecules of particular mRNAs (Ghaemmaghamiet al., 2003). McAdams and Arkin (1999) review cell-to-cellvariations in regulatory molecule concentrations that arisefrom internal cellular processes. Such variation includes theinevitable statistical variation in the partitioning of smallnumbers of proteins between daughter cells when cellsdivide. In bacterial cells, for example, some molecules arepresent at concentrations of only a few tens per cell, butthese can still perform functions crucial to proper cellfunctioning (Guptasarma, 1995). Stochastic gene expressionhas been observed directly in eukaryotic cells and appearserratic and bursty just as in prokaryotes, but with longeraverage intervals from one burst to the next (Chelly et al.,1989; Ko, 1992; McAdams and Arkin, 1999; Ross et al., 1994;Zlokarnik et al., 1998).

Even in apparently homogeneous populations of cells, theunavoidable consequences of molecular-level interactionsproduce stochasticity and fluctuations, deeply rooted in thestatistical mechanical behavior of nanoscale chemicalsystems in which concentrations of reactants can be low(Arkin et al., 1998; Elowitz et al., 2002; Guptasarma, 1995;McAdams and Arkin, 1999). How cells function and processinformation despite the stochastic nature of molecularevents remains an open question (Arkin et al., 1998;McAdams and Arkin, 1999; Rao et al., 2002; Ross et al.,1994). Development in some organisms (for example, Cae-norhabditis elegans) is so regular and invariant thatdifferentiation pathways for almost every cell are traceable.Such invariance stems partly from highly reproducibleintercellular signaling processes and the robustness ofregulatory networks (Sternberg and Félix, 1997; Sulstonand Horvitz, 1977). Another part of the answer is thatproteins generally remain in cells longer than mRNAs.Preexisting pools of proteins receive periodic supplementsas a consequence of transient bursts of mRNA synthesis.

159The stem cell genome and certain influencing factors: Towards a Rosetta stone

Since the size of the protein pool is relatively large, it isbuffered somewhat against variations in mRNA level. Thus,levels of protein in cells vary less than levels of mRNA.However, the lifetimes of different proteins vary, and short-lived proteins are therefore more susceptible to variation inthe level of their respective mRNAs (Tyagi, 2007).

Another cause of internal variation from cell to cell is theresult of statistical properties of regulatory chemicalreactions that involve a small number of reaction centersand slow reaction rates (McAdams and Arkin, 1999). Forexample, consideration of the statistical properties oftranscript initiation and translation suggests that proteinsare produced from activated promoters in short bursts ofhighly variable numbers. These bursts occur at random timeintervals in both bacterial (McAdams and Arkin, 1997) andeukaryotic cells (Ross et al., 1994). When the proteininvolved is a regulatory protein, fluctuations in proteinconcentration from cell to cell result, causing variability inthe time taken to complete regulated events (Arkin et al.,1998). The requirement for order suggests that robustnessitself may be an intrinsic property of intracellular networks(Barkai and Leibler, 1997; Ciliberti et al., 2007; von Dassowet al., 2000). Genetic regulatory networks, which must copewith such stochasticity, use redundancy, feedback loops, andother complex strategies to introduce the determinismnecessary for embryogenesis and development (McAdamsand Arkin, 1999).

Swain et al. (2002) classify sources of noise in cellularsystems into two classes according to their source: intrinsicor extrinsic. Intrinsic sources of noise are associated with theinherent stochasticity of biochemical processes, such astranscription and translation, while extrinsic noise sourcesarise independent of genes but nonetheless can act uponthem (Swain et al., 2002). Swain et al. (2002) includenumerous environmental variables, as well as position in cellcycle, as extrinsic noise sources. Their definition refers toextrinsic noise being caused by factors external to particulargenes. Hence, they also consider the quantity of proteinproduced by a gene and the mRNA degradation machinery assources of extrinsic noise. We use a different definition ofextrinsic noise source that does not include the cell cycle as acontributor. To make our difference in meaning explicit, weadopt the terms external and internal, and highlight that wemean external and internal to a particular cell. Swain et al.(2002) note that extrinsic noise sources have been relativelyneglected in theoretical studies of gene regulatory networks,which have typically focused on intrinsic noise sources(Swain et al., 2002). One exception is Hasty et al. (2000),who describe a model of gene regulation and demonstratehow external noise can be used to control switching behaviorof a network. The development of externally controllablenoise-based switches for gene expression could havesubstantial clinical implications (Hasty et al., 2000). Hastyet al. (2000) do not use the terms intrinsic and extrinsic, butinstead also use the terms internal and external.

Not all noisy variation in biological systems is overridden,and in many systems noise is apparently exploited (Rao et al.,2002). How cells make use of noise via regulatory networks isalso unclear, although studies that consider how cellularprocesses amplify or exploit noise often fall into two classes:(1) those that give rise to population heterogeneity anddiversity of cell type and (2) those that use noise to attenuate

noise (Rao et al., 2002). It has become clear that noise canplay an important role in biological systems. Stochasticresonance is a well-documented phenomenon in which noiseis used to increase the ability of nonlinear systems to detectweak signals (Collins, 1999; Collins et al., 1995, 1996; Russellet al., 1999; Wiesenfeld and Moss, 1995). Stochasticresonance appears to play an information processing role inboth the brain and the central nervous system (Ferster, 1996;Gilden et al., 1995; Traynelis and Jaramillo, 1998).

A particularly interesting use of internal noisy variationoccurs when there is coupling between molecular-levelfluctuations and the emergence of a specific observedphenotype (Arkin et al., 1998; McAdams and Arkin, 1999).This can occur when two regulatory proteins are involved inthe competitive control of a developmental switch that selectsbetween alternative cell lineage pathways, each associatedwith a specific phenotype. The production of regulatorymolecules can vary widely and stochastically among cells,and hence the pathway selected via the developmental switchcan appear random (Arkin et al., 1998; McAdams and Arkin,1999). The design of the switch circuit also impacts cell fateselection (Arkin et al., 1998), as can prevailing environmentalconditions (McAdams and Arkin, 1999).

Recognizing that large cell-to-cell variations in geneexpression are the norm, rather than the exception, manyinvestigators have begun to explore the origins and con-sequences of these variations (Elowitz et al., 2002; Pedrazaand van Oudenaarden, 2005; Raser and O'Shea, 2004, 2005).Raj et al. (2006) found that even isogenic populations of cellsdisplay large-scale variations in gene expression. This isbecause mRNAs are not synthesized at a steady rate, but areinstead synthesized in bursts, beginning and ending ran-domly. Cells that exhibited a large number of mRNAs werethought to be in the middle of a burst of RNA synthesis at themoment of observation, while cells that possess just a fewmRNAs were thought either to have not experienced a burstyet or to have experienced a burst of synthesis so long beforeobservation that most mRNA molecules had degraded (Rajet al., 2006).

In the hematopoietic system, some evidence suggests thatprogenitor cells are primed toward specific lineages atdifferent times of the cell cycle (Colvin et al., 2004). Thistemporal variation could follow a specific sequence, witheach cell progressing through stages corresponding tospecific lineages, or it could occur randomly. If primingoccurred in predictable series, the presence of multiplelineage-specific mRNAs could be due to overlap betweenprimed states or persistence of long-lived mRNAs. It is not yetclear whether temporal variation in gene expression in suchcells is stochastic, sequential, or some combination of these(Cross et al., 1997; Hu et al., 1997). Such analysis presumablyrequires the refinement of sufficiently sensitive assays forinterrogating stem cells in real time (Hu et al., 1997).

Low-level gene expression could have important implica-tions for early events in lineage determination, but needs tobe interpreted with caution (Hu et al., 1997). It has beensuggested that all genes are expressed at some low leveldetectable by sensitive assays, and this has been termed“illegitimate transcription” (Chelly et al., 1989). However,coexpression of lineage-affiliated gene expression programsis a feature of multipotential, not unilineage-committed,cells. This suggests that low-level gene expression in stem

160 J.D. Halley et al.

cells is not a case of illegitimate transcription and is moreconsistent with the multilineage priming hypothesis (Huet al., 1997).

Features common to stochastic regulatory switches thatappear to select randomly among alternative pathwaysinclude: (1) the transient low-level expression of keyregulatory proteins; (2) stochastic progress toward lineagecommitment, with a transient period of partial (reversible)commitment before a definitive choice emerges; and (3)multiple feedback loops that reinforce activation of selectedpaths and repress unselected alternatives (McAdams andArkin, 1999). There is conspicuous overlap between thesemechanisms of lineage commitment and patterns in othercomplex systems. In our view, such similarity betrays thepresence of fundamental mechanisms of self-organizationin nonequilibrium systems generally (Halley and Winkler,2008). In biological and nonbiological systems alike, self-organization plays a prominent pattern formation role insystems that are out of equilibrium (Anderson, 2002; Blazis,2002; Bonabeau, 1998; Bonabeau et al., 1997, 1999;Camazine et al., 2001; Haken, 1977, 1983, 1992; Nicolis andPrigogine, 1977).

Interestingly, the competitive interplay that occursbetween pheromone trails in mass-recruiting ant speciesappears analogous to interplay among lineage-specificfactors in stem cells. In both systems, apparent randomnessprovides information about future options available tosystems. In ant colonies, lost ants retrieve informationregarding food sources that would otherwise remain undis-covered. Similarly, in stem cells, stochastic gene expressionappears to prime lineage-specific modules, facilitatingcompetitive interplay among them during the process oflineage specification. Hence, in both systems, chaoticinteractions at the level of individual network nodes (antsor genes) play a key role in collective decision makingbecause of the ways in which they harness and propagatenew information.

Although in stem cells it is still unclear whether primingcorresponds to a stable state of low-level coexpression or anessentially zero-level expression, overlaid by random expres-sion noise (Roeder and Glauche, 2006), taken together, thesefindings suggest that gene expression noise is a biologicallyimportant variable, subject to selectionprocesses (Kærnet al.,2005). This has important implications for the architecture ofstem cell regulatory networks, because such networks mayhave evolved either to exploit noisy variation or to control it.Indeed, it may even be possible that some sources of noise aresuppressed while others are exploited. These ideas aredeveloped further in the next section.

Architecture of stem cell regulatory networks

Despite intensive work aimed at understanding stem cellsand their gene regulatory circuitry, clinical evidence appearsinadequate, as their clinical efficacy still seems far away(Joung et al., 2006). To help resolve the complexity of stemcell regulatory networks, several large-scale gene expressionanalyses have been employed, producing copious amounts ofexpression data. However, experiments have been per-formed with different questions in mind, and their resultshave precipitated different interpretations. The integration

of many different data sets will provide a more comprehen-sive view (Joung et al., 2006). In particular, the comparativestudy of diverse stem cell types could reveal core mechan-isms of stem cell regulation (Joung et al., 2006). Cai and co-workers (2004) reason that stem cells, irrespective of theirtissue of origin, face a similar set of challenges. All stem cellsneed to regulate progression tightly through the cell cycle,self-renewal, and periods of quiescence. In addition, main-tenance of genomic integrity to avoid accumulation ofmutations is paramount, as is proper response to externalenvironments via nutrients, growth factor receptors, extra-cellular matrix molecules, and gap-junction communication(Cai et al., 2004). It is expected that pathways controllingand facilitating such behaviors are more highly regulated instem cell populations and possibly by similar mechanisms(Cai et al., 2004).

Initial attempts to identify a molecular signature commonto several stem cell populations appear unsuccessful, withonly one gene common among three stem cell populationsanalyzed (almost 1000 genes) (Cai et al., 2004; Fortunelet al., 2003; Ivanova et al., 2002; Ramalho-Santos et al.,2002). However, the lack of a common set of stemness genescould reflect technical difficulties with experiments or in theway common genes are selected, rather than a genuineabsence of a common stem cell molecular signature (Caiet al., 2004; Karsten et al., 2003). Ideal comparisons amongstem cell populations require isolation of pure cells withminimal culturing, at a defined state of development, and inquantities such that comparative methods are performedwith sufficient rigor to overcome variability inherent in thecomparison technique itself. This is problematic becauseobtaining pure populations of stem cells is itself difficult (Caiet al., 2004; Vogel, 2003). Apart from technical problems, itis also likely that key genes vary their expression over time,such that their significance is hard to reveal. It is alsopossible that the sought-after stemness genes are absentfrom commercially available chips used by researchers(although this is becoming increasingly unlikely). Alterna-tively, functionalities common among stem cells (self-renewal, pluripotency) may be indefinable at the geneticlevel or could be produced by different sets of genes in eachstem cell population (Fortunel et al., 2003; Karsten et al.,2003). What does, at least, seem clear is that stemness genes(if they exist at all) appear not to be highly expressed ones(Fortunel et al., 2003; Vogel, 2003).

Key regulatory genes

Deciphering molecular mechanisms that underlie multiline-age priming and its relation to lineage commitment awaits acomplete profile of the molecular circuitry prevailing in eachof the blood cell lineages (Enver and Greaves, 1998).Nonetheless, some questions are immediately apparent.Does transcriptional noise vary according to conditions, e.g.,during different parts of the cell cycle? Are all thecomponents of lineage circuits primed to a similar degreeor is a skeleton crew recruited? (Enver and Greaves, 1998).Interestingly, studies on the hematopoietic system suggestthat alternative cell fate potentials are preserved in stemand progenitor cells by simultaneous low expression of keygenes for several lineages (Cory, 1999). Key regulatory genes

161The stem cell genome and certain influencing factors: Towards a Rosetta stone

for specific cell lineages are sometimes termed “mastergenes” (Cory, 1999; Gangenahalli et al., 2005; Rosenbauerand Tenen, 2007). In other cases the term “master gene” isavoided as it suggests that upon activation, a cell can beconverted to the corresponding cell type, an executive levelof control over lineage selection (Orkin, 2000; Weintraubet al., 1991). In the hematopoietic system, there do notappear to be lineage-specific master regulators. Instead, it isthe combinatorial activation of multiple factors that are notnecessarily lineage specific that determines lineage specifi-cation. This delicate interplay includes active suppression ofunselected lineage pathways and may also be tipped one wayor another by environmental or external influences (Enverand Greaves, 1998; Goldfarb, 2007; Nutt et al., 1999; Orkin,2000). This complexity presumably underlies examples of keyhematopoietic regulators that can convert one cell type toanother (Orkin, 2000).

A more relaxed conception of a master regulatory gene issimply a gene that is very well connected to other genes.These well-connected “hub” genes integrate multiple signalsand help to control the differentiation process by activatingentire networks of genes necessary for generating manydifferent cell types. Differentiation and cell behavioremerge through cross-antagonism and interplay betweentwo or more hub genes (Cinquin and Demongeot, 2005; Looseand Patient, 2006). In human embryonic stem cells (ESCs),for example, OCT4, SOX2, and NANOG together regulate asubstantial portion of genes. Conspicuously, there is sig-nificant overlap in the sets of genes that these three genesregulate. Greater than 90% of genes regulated by both OCT4and SOX2 are also regulated by NANOG (Boyer et al., 2005).Transcription factor binding sites often overlap with oneanother and when this occurs, competition between themoften results (Hermsen et al., 2006). In contrast, cooperativebehavior among proteins occurs when the binding of oneprotein to DNA increases the binding affinity of the other(Ptashne and Gann, 2002). Interplay between cooperativeand competitive activity among transcription factors canprovide regulatory networks with substantial information-processing capabilities (Hermsen et al., 2006).

Stem cell transcription factors appear sufficient toactivate many differentiation-associated genes long beforelineage choice and cell-type specification occur. Differentia-tion apparently involves mechanisms that select a subset ofgenes from within a range of accessible options, stabilizingtheir expression while down-regulating other irrelevantgenes (Rothenberg and Anderson, 2002). It has beensuggested that if there are immature cell types thatsimultaneously express master genes related to differentdifferentiation pathways, as long as they avoid expressinggenes antagonistic to these, such cells should be able to actas precursors for both mature cell types. This is a differentway of looking at hematopoietic stem cell complexity andsuggests that if combinations of transcription factorsexpressed in a precursor are sufficient to predict develop-mental potential, the question becomes which transcriptionfactor combinations are possible (Rothenberg and Anderson,2002).

Dose-dependent developmental pathways increase thecomplexity of hematopoietic differentiation such thattrajectories of differentiating cells depend on a network oftime-dependent interactions. The order and duration of

transcription factor action contribute to lineage choice aswell. Any model of a regulatory network for hematopoieticcell-type specification must account for mechanisms thatopen and close temporal windows for permitted activity(Rothenberg and Anderson, 2002).

Regulatory networks underlie countless unfolding eventsduring development and have a recognizable, underlyingstructure despite their obvious complexity (Materna andDavidson, 2007). They can be roughly grouped into fourlevels of detail, the smallest of which includes thetranscription factors, their target genes, and binding siteson DNA (Madan Babu et al., 2004). Transcription factors bindto the cis-regulatory modules of downstream genes and areinformation processors that execute basic logic operations toyield new transcriptional outputs according to their tran-scription factor inputs (Istrail and Davidson, 2005). Abovethis level are recurring patterns of transcription factorinteraction called network motifs. Although motifs do notusually represent functionally separable network compo-nents, they display kinetic properties that influence thetemporal activity of target genes (Madan Babu et al., 2004;Mangan et al., 2003). For example, coherent feedforwardloops enable decision making despite noisy input by filteringout fluctuations in input stimuli (Mangan et al., 2003). Abovethe level of motifs, clusters of motifs organize into modulesof semi-independent transcription units (Madan Babu et al.,2004). The final level of detail is the whole network,comprising the full set of modules and interplay amongthem (Madan Babu et al., 2004) (Fig. 1).

The concept of modularity suggests that cellular func-tionality can be reasonably partitioned into a collection ofmodules, each of which performs a specific task, largelyseparable from (loosely coupled to) that of other modules(Hartwell et al., 1999; Lauffenburger, 2000; Rao and Arkin,2001; Ravasz et al., 2002). As discussed in the previoussection, stochastic fluctuations in some components ofregulatory networks are unavoidable. Such fluctuationsmay propagate and affect an entire system's performance(Swain et al., 2002). One factor that is expected to limitnoise propagation is modularity. Interestingly, a recent studyon the topological properties of transcription regulatorynetworks in yeast (Saccharomyces cerevisiae) suggests thathighly connected proteins primarily connect to those withlow connectivity. In other words, hubs appear well separatedfrom each other, a property that may prevent cross talkbetween different functional modules. Furthermore, itappears that in the vertebrate genome the density ofmethyl-CpG (a factor that represses transcription) is suffi-cient to repress weak promoters without affecting strongerones (Bird and Tweedie, 1995). Hence, low-density methyla-tion could function as a noise reduction system, suppressingbackground levels of transcription while leaving authentictranscription unaffected (Bird and Tweedie, 1995). Suchrepression probably plays a role in the maintenance andcontrol of network modularity.

Appreciation of the modular nature of gene networks iscrucial for understanding how cells respond to externalsignals (Segal et al., 2003). Although genome-wide expres-sion profiles provide valuable information about coexpressedgenes, the regulatory programs of these modules can besuggested only indirectly (Segal et al., 2003). One way toapproach this problem is to search for common cis-regulatory

Figure 1 Hierarchical descriptions of transcriptional regulatory networks. (a) The elementary level comprising transcription factorand target gene with DNA recognition site. (b) Regulatory network motifs of relatively fixed pattern that occur commonly in networks.(c) Connection of motifs into more complex modules, many of which have been experimentally identified. (d) The complete regulatorynetwork, which provides the program for regulation of gene expression in an organism. Used with permission from Madan Babu et al.(2004).

162 J.D. Halley et al.

binding sites in the upstream regions of groups of genes (Rothet al., 1998; Tavazoie et al., 1999). Cis-regulatory elementscomprise the chemical code that specifies interactions be-tween transcription factors and their constituent sequences(Davidson et al., 2003). Each cis-regulatory element pro-cesses multiple inputs and therefore interacts with multipletranscription factors. Similarly, each transcription factorinteracts with multiple cis-regulatory elements (Davidsonet al., 2003). The complex interactions specified by suchinstruction demonstrate why a gene regulatory network hasnetwork architecture rather than a simple linear or branch-ing structure (Davidson et al., 2003).

Genes encoding given transcription factors are utilizedrecursively, in multiple times and places during embryogen-esis and development, participating in entirely independentprocesses (Howard-Ashby et al., 2006). Repeated utilizationand redundancy appear common in regulatory gene net-works. Regulatory genes can be thought of as nodes in a generegulatory network that read, process, and transmit spatialand temporal information (Davidson, 2006). Any given geneis activated when the correct set of upstream inputs ispresented, resulting in regulatory proteins that convey newspatial and temporal cues when they interact with cis-regulatory targets in downstream genes. New information-processing nodes are activated continuously during embry-ogenesis, with concomitant increases in complexity of theembryo regulome (Howard-Ashby et al., 2006).

Binary vs complex lineage specification

Formation of specialized tissues is progressive, and muchevidence suggests that cell fate is progressively restricted.Understanding this relationship between progressive restric-tion of cell fate requires consideration of the nature of thedecision-making process (Brown et al., 1988) and hence thearchitecture of stem cell regulatory networks. Althoughcellular differentiation and development are often consid-ered in the context of stepwise binary diversification (Brownet al., 1988; Huang et al., 2007; Kaletta et al., 1997), there is

also evidence of a more complex decision-making process inwhich more than two outcomes are possible (Cinquin andDemongeot, 2002, 2005; Rothenberg et al., 1999). Such“multistable” switches are more resistant to mathematicalanalysis than bistable switches, but are expected to appearin biological systems in which a decision more complicatedthan a binary decision must be made (Cinquin and Demon-geot, 2002). In particular, studies of the hematopoieticsystem suggest that decision making is more complex than aseries of binary decisions (Cinquin and Demongeot, 2005).This more complex interplay is illustrated in Fig. 2b andcontrasted with a simpler binary decision-making architec-ture in Fig. 2a. Any conceptual framework of stem celldecision making should accommodate both types of archi-tecture, as well as a mix of these extremes, if necessary.

Interestingly, the predominately binary model of celldifferentiation appears to prevail in a number of inverte-brate development systems and in some mammalian pro-genitors, including the germ cell lineages (Enver andGreaves, 1998). This supports the intuitive notion that binarydecision making is evolutionarily more ancient, representinga simpler mechanism of lineage commitment. An interestingpossibility is that the two extremes represented in Fig. 2describe embryonic stem cells and a population of stem cellswith more complex decision-making capabilities, possiblythose of the hematopoietic system. Hematopoietic stemcells (HSCs) are responsible for producing around 10 differentcell types. They need to respond to the changing demands ofan organism and must persist throughout an organism'slifetime. In contrast, ESCs are an artifact cell population thatare representative of a very early stage of development. Itfollows that ESCs could have more ancient, relatively simple,regulatory architecture compared with HSCs. Consistentwith this suggestion are findings that HSCs have complextranscription factor requirements, while multipotent hema-topoietic precursors in embryos can have simpler require-ments (Rothenberg and Anderson, 2002, and referencestherein). However, Enver and Greaves (1998) note that itwould be surprising if the antique binary mechanism did notpenetrate the hematopoietic system to some extent and

Figure 2 Taken with permission from Cinquin and Demongeot(2002, 2005). Two extremes of stem cell regulatory networkarchitecture. (a) Simple binary (stepwise) model, in whichlineage specification occurs through a series of binary decisions.(b) Complex (concerted) interplay of multiple factors, in whichcells have more than two options available simultaneously tothem and phenotype selection depends on a delicate interplayamong multiple factors. Arrows represent activation and squaresinhibition.

163The stem cell genome and certain influencing factors: Towards a Rosetta stone

suggest that the hierarchical structure of hematopoiesis isevidence of this.

In particular, if ESCs exploit a relatively simple regulatoryarchitecture they may be less sensitive to external circum-stance, relying instead on an unfolding gene program. Thiscould be beneficial for organisms, which proceed withembryogenesis robustly, despite mutations and variableexternal conditions. The idea that HSC regulatory architec-ture is more complex than that of ESCs is supported by thestriking complexity of postembryonic development andformation of the adult body plan, which dwarfs that ofembryogenesis in terms of both multilayered morphology andnumber of cell types (Howard-Ashby et al., 2006).

The notion that ESCs and HSCs have different regulatoryarchitecture for accomplishing similar tasks could be linkedwith their very different behavior in vitro. ESCs can bepropagated indefinitely in vitro despite the fact that theyare an artifact of a transient population in nature. Incontrast, HSCs are notoriously difficult to expand in vitro,

but are a permanent population in nature. These differencesare explicable if the complexity of HSC regulatory architec-ture is higher than that of ESCs. The greater complexity ofHSC architecture could facilitate integration of changingexternal conditions, possibly through increased sensitivity toexternal conditions relative to ESCs. However, such complex-ity could make maintenance of a stable stem cell state moredifficult. Theoretical work reveals an intimate relationshipbetween network complexity and network stability andsuggests that more complex networks are often less stablethan simpler networks (Albert and Barabási, 2002; Sinha,2005; Wilmers et al., 2002). Hence, if HSC regulatoryarchitecture is relatively complex it could be inherentlyunstable, requiring a very specialized niche environment forcontinued persistence (vide infra). The complexity of bothHSC regulatory architecture and that of the niche might bethe reason our ability to manipulate these stem cells in vitroremains so limited. Despite over 30 years of intenseexperimental effort, we remain unable to expand HSCssignificantly in vitro. In contrast, ESCs are readily expandedin vitro even though their study is relatively recent. Severalcomponents contribute to the stem cell niche, includingsoluble factors, extracellular matrix or cell substrate, thebiophysical environment, and nearby cells that can elicitcell-to-cell signaling. Such factors converge via intracellularsignaling pathways and, together with intrinsic geneticcircuitry, govern whether a cell divides, differentiates, ordies (Metallo et al., 2007).

The question of how complex interplay among multiplefactors can be modeled is a pertinent and difficult problem.Importantly, there is evidence to suggest that differentiationfactors can be antagonistic (Cinquin and Demongeot, 2005;Cory, 1999; Enver and Greaves, 1998; Nutt et al., 1999). Thatis, expression of one phenotype precipitates repression ofalternate cell fates. However, it appears more complex thanthis because there is evidence for competition among variouscell-fate-determining factors (Cinquin and Demongeot,2005; Enver and Greaves, 1998; Hermsen et al., 2006). Thenotion of competition is supported by dose-dependencyeffects, which imply that Boolean models, in which a specificmaster gene is turned on and represses all alternatives, areinadequate (Cinquin and Demongeot, 2005). Competitioncan occur through active repression of transcription bycompeting transcription factors, but also through physicalinteraction between factors (Cinquin and Demongeot, 2005;Enver and Greaves, 1998; Hermsen et al., 2006).

Because transcription factors regulate one another'sexpression and alter one another's DNA binding and transac-tivation activities, certain combinations are unstable(Rothenberg and Anderson, 2002). Cinquin and Demongeot(2005) test whether the complex model illustrated in Fig. 2bis able to display stable coexistence of several antagonisticfactors. Stable coexistence of antagonistic factors is poten-tially very important because their destabilization couldforce a decision, corresponding to lineage choice. One waythat this might occur is through modulation of competitionthrough transcription strength. At low levels of competition(corresponding to low level transcription), stable coexistenceof many antagonistic factors appears possible. However, ascompetition increases through increasing transcriptionstrength, this balance is upset and one cell fate is selectedat the expense of all others. Hence, transcription strength,

164 J.D. Halley et al.

because of its impact on competition, could provide a simplemeans to precipitate and regulate differentiation (Cinquinand Demongeot, 2005). Swiers et al. (2006) identified denseoverlapping region (DOR) gene motifs that may provide amechanistic explanation for this behavior. These DORs wereincluded in their motif-based models of cross-antagonism oftranscription factors in the hematopoietic system. DORs mayalso provide a rationale for the observed convergence to acommon cell fate via different gene expression trajectories,reported by Huang et al. (2005). They modeled a specificexample of cross-antagonism between PU.1 and GATA-1 andtheir respective target genes that results in one of twolineage choices, depending on which of the two transcriptionfactors dominates. Computational models are ideal toolsfor such investigations because they encourage the formalexpression of the current state of knowledge and facilitaterigorous exploration of system dynamics. Such models ena-ble us to test and investigate the key operating principles ofa network and other important questions such as howsources of noise can impact network behavior (Rao et al.,2002).

External factors

In the past, research into the regulatory mechanisms of stemcells often focused on internal mechanisms and tissue-specific patterns of gene activity (Heissig et al., 2005).However, numerous studies have since demonstrated thatstem cell niche microenvironments are also critical forregulating stem cell fate within different tissues duringpostnatal life (Haylock and Nilsson, 2005; Heissig et al.,2005). Cells respond to external signals through a series ofsignaling pathways that may be branched or linear or maylink to other such pathways to form complex networks(Knofler et al., 2005). External signals that influence stemcell fate include secreted factors and cell–cell interactions,mediated by integral membrane proteins (Heissig et al.,2005). Interactions with neighboring cells are key determi-nants of stem cell behavior (Adams and Scadden, 2006;Heissig et al., 2005; Lin, 2002; Metallo et al., 2007; Mooreand Lemischka, 2006) and cancer cell behavior (Lee andHerlyn, 2007; Li et al., 2007b; Rizo et al., 2006). Indeed, itmay even be possible that the location and niche environ-ment (rather than a unique molecular signature) are whatprovides different stem cell populations with unique iden-tities (Heissig et al., 2005).

Analysis of stem cell–niche interactions has been ham-pered by their unknown location, but over the past few yearsmuch progress has been made (Adams and Scadden, 2006;Haylock et al., 2007; Taichman, 2005; Wilson and Trumpp,2006; Yin and Li, 2006). In particular, transplantation studiesrevealed that localization within the bone marrow nichevaried according to cell phenotype, with HSCs of highhematopoietic potential being enriched in the endostealregion (Haylock et al., 2007; Nilsson et al., 2001; Nilsson andSimmons, 2004). Complex interplay between HSCs and theendosteal bone marrow niche influences not only HSChoming, retention, and mobilization (Wilson and Trumpp,2006), but also the relative numbers of blood cells produced.Every blood cell differentiation program includes lineage-specific proliferative checkpoints, at which the number of

cells produced depends on interaction with proliferation andsurvival signals from secreted cytokines. Signals within themicroenvironment can amplify the effects of transcriptionfactor combinations, tipping the balance between two ormore transcription factors, thereby favoring differentiationalong one pathway or another (Graf, 2002; Miyamoto et al.,2002). Since single progenitors can simultaneously expressmultiple lineage-related cytokine receptors, cross talkbetween different cytokine receptors can also occur (Miya-moto et al., 2002). The existence of an external controlfacilitates increased plasticity of cell fate during differentia-tion and has an important consequence for the regulatorylogic of hematopoietic differentiation. As long as someprecursors of each lineage are made, it may not matter ifthere are more or fewer than is ideal. By removing penaltiesfor specification errors, regulation by cytokines allows thehematopoietic system to exploit delicate mechanisms forcreating lineage divergence that need not be robustlypredictable in output volume (Rothenberg and Anderson,2002).

Most HSCs reside in a nondividing, quiescent, state despitethe fact that they are responsible for the daily production ofbillions of cells. Importantly, HSCs are activated in responseto injury and other external influences, and their offspringhave extensive proliferative potential (Fuchs et al., 2004;Heissig et al., 2005; Li et al., 2007a). The release of HSCs intocirculation could provide a readily accessible source of HSCsto repopulate areas of injured bone marrow and may be anartifact of bone remodeling, which causes constant destruc-tion and formation of HSC niches, requiring continual HSCrelocalization (Adams and Scadden, 2006; Fuchs et al., 2004;Taichman, 2005; Wilson and Trumpp, 2006).

The ways in which external factors integrate into stem cellgene regulatory networks presumably reflect the architec-ture of such networks. As discussed above, it is intuitive thatmore complex regulatory networks are better able tointegrate more complex external conditions, potentiallyexploiting more complex external information. Such com-plexity could underlie HSC regulatory networks, ensuring thattheir in vitro manipulation is similarly complex. In contrast,ESCs could have relatively simple regulatory architecture,making integration of external factors cumbersome. Consis-tent with this idea are interesting findings within the field ofcancer research. The idea of cancer stem cells has beenaround for several decades (Marx, 2003), but it is onlyrelatively recently that it has become apparent that cancerdevelopment and normal embryonic development have somefeatures in common (Ferretti et al., 2007; Patel et al., 2006;Ruiz i Altaba et al., 2002; Wilczynski, 2006). For example,much of the signaling circuitry exploited by trophoblast cellsis also exploited by cancer cells. This overlap in molecularmachinery is consistent with the fact that both cell typeshave similar phenotypes, including proliferation, migration,and invasion (Ferretti et al., 2007; Hanahan and Weinberg,2000). The primary difference between normal trophoblastdevelopment and malignant transformation is that normaldevelopment is temporally and spatially predictable,whereas malignant transformation is typified by multiplestochastic events (Ferretti et al., 2007; Hanahan andWeinberg, 2000; Hiden et al., 2007).

Li et al. (2007a) propose that the ability of a tumor tometastasize, an inherent property of some cancer stem cells,

165The stem cell genome and certain influencing factors: Towards a Rosetta stone

is modulated through interactions with the niche. Consistentwith this idea is the fact that many factors known to governHSC migration, engraftment, and homing are also mediatorsof cancer metastasis and cell invasion (Li et al., 2007a, andreferences therein). Skeletal bone provides niche sites forHSCs and is the most common site for cancer metastasis (Liet al., 2007a; Raubenheimer and Noffke, 2006). It followsthat bone-specific factors, such as the level of calcium ions,could serve as chemoattractants for guiding migrating cancercells to bone (Li et al., 2007a). Although cancer cells andthose of the hematopoietic system appear to be drawn tosimilar niche environments, the proliferation of stem cells,unlike that of cancer stem cells, is rigorously controlled bythis environment (Li and Neaves, 2006; Rizo et al., 2006).Upsetting the delicate balance between intrinsic signals andthose from the niche can result in aberrant genetic changes,implicated in the development of leukemia and othercancers (Li et al., 2007a; Rizo et al., 2006).

Virtually all mammalian cells carry similar molecularmachinery for the regulation of proliferation, differentia-tion, and death (Hanahan and Weinberg, 2000). Hence, acancer phenotype might arise through inappropriate reacti-vation of some embryonic pathways in HSCs. Interestingly,our hypothesis that the regulatory architecture of embryonicstem cells is relatively simple might explain why cancer cellsfail to be adequately regulated by their niche environment.The relative simplicity of their regulatory architecture maybe ill-equipped to integrate the full complexity of regulationoffered by the HSC niche. Such discrepancy could upset thedelicate balance between intrinsic and extrinsic signaling, sovital to the control of cell behavior.

Classical work on cancer genomics centered on discoveryof oncogenes and tumor-suppressor genes. However, suchwork has trouble accounting for puzzling findings, such aswhy transition to cancer can sometimes occur in the absenceof major changes in chromosomal sequence (Qu et al., 2007).Hanahan and Weinberg's (2000) review on the hallmarks ofcancer suggests that although the next couple decades willundoubtedly be accompanied by technical advances, themore fundamental change in cancer research will beconceptual. They provocatively hint that the similaritiesbetween the many different types of human cancers could bedue to a small number of fundamental underlying principles,which elegantly summarize the complexity of the disease.This hope is supported by the fact that despite theirextraordinary diversity, all clinically significant cancersshare at least one common characteristic: excessive pro-liferation of affected cells (Li and Neaves, 2006).

Conclusion

Cracking the codes that will allow us to understand stem cellbehavior is clearly an extremely difficult task (Tsai, 2004).Although recent efforts have precipitated a better mechan-istic understanding of stem cell fate choice, little is knownabout combinatorial effects of multiple input signals,present in most culture systems (Metallo et al., 2007). Asstem cell technologies transition from research lab to clinicalapplication, the need for robust culture systems thatpredictably control stem cell growth and differentiationincreases. Stem-cell-based processes must reproducibly

generate large amounts of functional cells or tissues, andthe multistep mechanisms involved will undoubtedly requirecomplex, well-controlled systems. Such goals will beaccomplished only by exploiting the interactions of stemcells with their microenvironments (Metallo et al., 2007),together with an accurate conceptual framework thatintegrates known important variables.

Although the three classes of factors considered here arepresented separately, their shared evolutionary historynecessitates integration. Stochastic gene expression per-vades network components; network architecture controls,modulates, or exploits this noise while performing additionalcomputation; and such complexity also interplays withfactors external to cells, although the extent to whichvarious stem cell species respond to external factorsprobably varies. Hence, key features of stem cell regulationand behavior are probably emergent properties of severalinteracting pathways and networks (Phillips et al., 2000). If itbecomes possible to understand each of these factors, andhow they interrelate, we should be a few steps closer towarda Rosetta stone for the stem cell genome.

Acknowledgments

We gratefully acknowledge David Haylock, Ed Stanley, ErnstWolvetang, Andrew Laslett, Martin Burd, and AndrewElefanty for useful discussion and/or critical review of themanuscript. This work was supported by the Australian StemCell Centre.

References

Adams, G.B., Scadden, D.T., 2006. The hematopoietic stem cell in itsplace. Nat. Immunol. 7, 333.

Akashi, K., 2005. Lineage promiscuity and plasticity in hematopoie-tic development. Ann. N. Y. Acad. Sci. 1044, 125–131.

Akashi, K., He, X., Chen, J., Iwasaki, H., Niu, C., Steenhard, B.,Zhang, J., Haug, J., Li, L., 2003. Transcriptional accessibility forgenes of multiple tissues and hematopoietic lineages is hier-archically controlled during early hematopoiesis. Blood 101,383–390.

Albert, R., Barabási, A.L., 2002. Statistical mechanics of complexnetworks. Rev. Mod. Phys. 74, 47–97.

Anderson, C., 2002. Self-organization in relation to several similarconcepts: are the boundaries to self-organization indistinct? Biol.Bull. 202, 247–255.

Arkin, A., Ross, J., McAdams, H.H., 1998. Stochastic kinetic analysisof developmental pathway bifurcation in phage λ-infectedEscherichia coli cells. Genetics 149, 1633–1648.

Barkai, N., Leibler, S., 1997. Robustness in simple biochemicalnetworks. Nature 387, 913.

Bird, A., Tweedie, S., 1995. Transcriptional noise and the evolutionof gene number. Philos. Trans. Biol. Sci. 349, 249–253.

Blazis, D.E.J., 2002. Introduction: the limits to self-organization inbiological systems. Biol. Bull. 202, 245–246.

Bonabeau, E., 1998. Social insect colonies as complex adaptivesystems. Ecosystems 1, 437–443.

Bonabeau, E., Theraulaz, G., Deneubourg, J.-L., Aron, S., Cama-zine, S., 1997. Self-organization in social insects. TREE 12,188–193.

Bonabeau, E., Dorigo, M., Theraulaz, G., 1999. Swarm Intelligence:from Natural to Artificial Systems. Oxford Univ. Press, New York.

Bornholdt, S., 2005. Less is more in modeling large genetic networks.Science 310, 449–451.

166 J.D. Halley et al.

Boyer, L.A., Lee, T.I., Cole, M.F., Johnstone, S.E., Levine, S.S.,Zucker, J.P., Guenther, M.G., Kumar, R.M., Murray, H.L., Jenner,R.G., Gifford, D.K., Melton, D.A., Jaenisch, R., Young, R.A.,2005. Core transcriptional regulatory circuitry in human embryo-nic stem cells. Cell 122, 947–956.

Brown, G., Bunce, C.M., Lord, J.M., McConnell, F.M., 1988. Thedevelopment of cell lineages: a sequential model. Differentiation39, 83–89.

Bruno, L., Hoffmann, R., McBlane, F., Brown, J., Gupta, R., Joshi, C.,Pearson, S., Seidl, T., Heyworth, C., Enver, T., 2004. Molecularsignatures of self-renewal, differentiation, and lineage choice inmultipotential hemopoietic progenitor cells in vitro. Mol. Cell.Biol. 24, 741–756.

Cai, J., Weiss, M.L., Rao, M.S., 2004. In search of “stemness.” Exp.Hematol. 32, 585–598.

Camazine, S., Deneubourg, J.-L., Franks, N., Theraulaz, G.,Bonabeau, E., 2001. Self-Organization in Biological Systems.Princeton Univ. Press, Princeton, NJ.

Cantor, A.B., Orkin, S.H., 2001. Hematopoietic development: abalancing act. Curr. Opin. Genet. Dev. 11, 513–519.

Cantor, A.B., Orkin, S.H., 2002. Transcriptional regulation oferythropoiesis: an affair involving multiple partners. Oncogene21, 3368–3376.

Chelly, J., Concordet, J.-P., Kaplan, J.-C., Kahn, A., 1989.Illegitimate transcription: transcription of any gene in any celltype. Proc. Natl. Acad. Sci. U. S. A. 86, 2617–2621.

Ciliberti, S., Martin, O.C., Wagner, A., 2007. Robustness can evolvegradually in complex regulatory gene networks with varyingtopology. PLoS Comput. Biol. 3, 164–173.

Cinquin, O., Demongeot, J., 2002. Positive and negative feedback:striking a balance between necessary antagonists. J. Theor. Biol.216, 229–241.

Cinquin, O., Demongeot, J., 2005. High-dimensional switches and themodelling of cellular differentiation. J. Theor. Biol. 233, 391–411.

Collins, J.J., 1999. Fishing for function in noise. Nature 402,241–242.

Collins, J.J., Chow, C.C., Imhoff, T.T., 1995. Stochastic resonancewithout tuning. Nature 376, 236–238.

Collins, J.J., Imhoff, T.T., Grigg, P., 1996. Noise-enhanced tactilesensation. Nature 383, 770.

Colvin, G.A., Lambert, J.-F., Abedi, M., Dooner, M.S., Demers, D.,Moore, B.E., Greer, D., Aliotta, J.M., Pimentel, J., Cerny, J.,Lum, L.G., Quesenberry, P.J., 2004. Differentiation hotspots: thedeterioration of hierarchy and stochasm. Blood Cells Mol. Dis. 32,34–41.

Cory, S., 1999. Wavering on commitment. Nature 401, 538–539.Cross, M.A., Enver, T., 1997. The lineage commitment of haemo-

poietic progenitor cells. Curr. Opin. Genet. Dev. 7, 609–613.Cross, M.A., Heyworth, C.M., Murrell, A.M., Bockamp, E.O., Dexter,

T.M., Green, A.R., 1994. Expression of lineage restrictedtranscription factors precedes lineage specific differentiationin a multipotent haemopoietic progenitor cell line. Oncogene 9,3013–3016.

Cross, M.A., Heyworth, C.M., Dexter, T.M., 1997. How do stem cellsdecide what to do? The Molecular Basis of Cellular DefenceMechanisms. Wiley, Chichester, pp. 3–18.

Davey, R.E., Zandstra, P.W., 2004. Signal processing underlyingextrinsic control of stem cell fate. Curr. Opin. Hematol. 11,95–101.

Davidson, E.H., 2006. The Regulatory Genome. Academic Press, SanDiego.

Davidson, E.H., McClay, D.R., Hood, L., 2003. Regulatory genenetworks and the properties of the developmental process. Proc.Natl. Acad. Sci. U. S. A. 100, 1475–1480.

Elowitz, M.B., Levine, A.J., Siggia, E.D., Swain, P.S., 2002. Stochasticgene expression in a single cell. Science 297, 1183–1186.

Enver, T., Greaves, M., 1998. Loops, lineage, and leukemia. Cell 94,9–12.

Ferretti, C., Bruni, L., Dangles-Marie, V., Pecking, A.P., Bellet, D.,2007. Molecular circuits shared by placental and cancer cells,and their implications in the proliferative, invasive and migratorycapacities of trophoblasts. Hum. Reprod. Update 13, 121–141.

Ferster, D., 1996. Is neural noise just a nuisance? Science 273, 1812.Fortunel, N.O., Otu, H.H., Ng, H.-H., Chen, J., Mu, X., Chevassut, T.,

Li, X., Joseph, M., Bailey, C., Hatzfeld, J.A., Hatzfeld, A., Usta,F., Vega, V.B., Long, P.M., Libermann, T.A., Lim, B., 2003.Comment on “‘Stemness’: transcriptional profiling of embryonicand adult stem cells” and “A stem cell molecular signature”I. Science 302, 393 author reply p. 393.

Fuchs, E., Tumbar, T., Guasch, G., 2004. Socializing with theneighbors. Cell 116, 769–778.

Gangenahalli, G.U., Gupta, P., Saluja, D., Verma, Y.K., Kishore, V.,Chandra, R., Sharma, R.K., Ravindranath, T., 2005. Stem cell fatespecification: role of master regulatory switch transcriptionfactor PU.1 in differential hematopoiesis. Stem Cells Dev. 14,140–152.

Ghaemmaghami, S., Huh, W.-K., Bower, K., Howson, R.W., Belle, A.,Dephoure, N., O'Shea, E.K., Weissman, J.S., 2003. Global analysisof protein expression in yeast. Nature 425, 737.

Gilden, D.L., Thornton, T., Mallon, M.W., 1995. 1/f noise in humancognition. Science 267, 1837.

Goldfarb, A.N., 2007. Transcriptional control of megakaryocytedevelopment. Oncogene 26, 6795–6802.

Graf, T., 2002. Differentiation plasticity of hematopoietic cells.Blood 99, 3089–3101.

Guptasarma, P., 1995. Does replication-induced transcription reg-ulate synthesis of the myriad low copy number proteins ofEscherichia coli? BioEssays 17, 987–997.

Haken, H., 1977. Synergetics. Springer-Verlag, Berlin.Haken, H., 1983. Synergetics. Springer-Verlag, Berlin.Haken, H., 1992. Synergetics as a strategy to cope with complex

systems. In: Haken, H., Mikhailov, A. (Eds.), InterdisciplinaryApproaches to Nonlinear Complex Systems. Springer-Verlag,Berlin.

Halley, J.D., Winkler, D.A., 2008. Critical-like self-organization andnatural selection: two facets of a single evolutionary process?BioSystems 92, 148–158.

Hanahan, D., Weinberg, R.A., 2000. The hallmarks of cancer. Cell100, 57–70.

Hartwell, L.H., Hopfield, J.J., Leibler, S., Murray, A.W., 1999. Frommolecular to modular cell biology. Nature 402, C47–C52.

Hasty, J., Pradines, J., Dolnik, M., Collins, J.J., 2000. Noise-basedswitches and amplifiers for gene expression. Proc. Natl. Acad.Sci. U. S. A. 97, 2075–2080.

Haylock, D.N., Nilsson, S.K., 2005. Stem cell regulation by thehematopoietic stem cell niche. Cell Cycle 4, 1353–1355.

Haylock, D.N., Williams, B., Johnston, H., Liu, M., Rutherford, K.,Whitty, G., Simmons, P., Bertoncello, I., Nilsson, S., 2007. HSCwith higher hemopoietic potential reside at the bone marrowendosteum. Stem Cells 25, 1062–1069.

Heissig, B., Ohki, Y., Sato, Y., Rafii, S., Werb, Z., Hattori, K., 2005. Arole for niches in hematopoietic cell development. Hematology10, 247–253.

Hermsen, R., Tans, S., ten Wolde, P.R., 2006. Transcriptionalregulation by competing transcription factor modules. PLoSComput. Biol. 2, e164.

Hiden, U., Bilban, M., Knofler, M., Desoye, G., 2007. Kisspeptins andthe placenta: regulation of trophoblast invasion. Rev. Endocr.Metab. Disord. 8, 31–39.

Howard-Ashby, M., Materna, S.C., Brown, C.T., Tu, Q., Oliveri, P.,Cameron, R.A., Davidson, E.H., 2006. High regulatory gene use insea urchin embryogenesis: implications for bilaterian develop-ment and evolution. Dev. Biol. 300, 27–34.

Hu, M., Krause, D., Greaves, M., et al., 1997. Multilineage geneexpression precedes commitment in the hemopoietic system.Genes Dev. 11, 774–785.

167The stem cell genome and certain influencing factors: Towards a Rosetta stone

Hu, Z., Mellor, J., Wu, J., Kanehisa, M., Stuart, J.M., DeLisi, C.,2007. Towards zoomable multidimensional maps of the cell. Nat.Biotechnol. 25, 547–554.

Huang, S., Eichler, G., Bar-Yam, Y., Ingber, D.E., 2005. Cell fates ashigh dimensional attractor states of a complex gene regulatorynetwork. Phys. Rev. Lett. 94, 128701.

Huang, S., Guo, Y.P., May, G., Enver, T., 2007. Bifurcation dynamics inlineage-commitment in bipotent progenitor cells. Dev. Biol. 305,695–713.

Istrail, S., Davidson, E.H., 2005. Gene regulatory networks specialfeature: logic functions of the genomic cis-regulatory code. Proc.Natl. Acad. Sci. 102, 4954–4959.

Ivanova, N.B., et al., 2002. A stem cell molecular signature. Science298, 601–604.

Joung, J.-G., Shin, D., Seong, R.H., Zhang, B.-T., 2006. Identifica-tion of regulatory modules by co-clustering latent variablemodels: stem cell differentiation. Bioinformatics 22, 2005–2011.

Kærn,M., Elston, T.C., Blake,W.J., Collins, J.J., 2005. Stochasticity ingene expression: from theories to phenotypes. Nat. Rev. Genet. 6,451.

Kaletta, T., Schnabel, H., Schnabel, R., 1997. Binary specification ofthe embryonic lineage in Caenorhabditis elegans. Nature 390,294.

Karsten, S.L., Kudo, L.C., Jackson, R., Sabatti, C., Kornblum, H.I.,Geschwind, D.H., 2003. Global analysis of gene expression inneural progenitors reveals specific cell-cycle, signaling, andmetabolic networks. Dev. Biol. 261, 165–182.

Knofler, M., Sooranna, S.R., Daoud, G., Whitley, G.S., Markert, U.R.,Xia, Y., Cantiello, H., Hauguel-de Mouzon, S., 2005. Trophoblastsignalling: knowns and unknowns—a workshop report. Placenta26, S49.

Ko, M.S.H., 1992. Induction mechanism of a single gene molecule:stochastic or deterministic. BioEssays 14, 341–346.

Laforge, B., Guez, D., Martinez, M., Kupiec, J.J., 2005. Modelingembryogenesis and cancer: an approach based on an equilibriumbetween the autostabilization of stochastic gene expression andthe interdependence of cells for proliferation. Prog. Biophys.Mol. Biol. 89, 93–120.

Lauffenburger, D.A., 2000. Cell signaling pathways as controlmodules:complexity for simplicity? Proc. Natl. Acad. Sci. U. S. A. 97,5031–5033.

Lee, J.T., Herlyn, M., 2007. Microenvironmental influences inmelanoma progression. J. Cell. Biochem. 101, 862–872.

Li, L.H., Neaves, W.B., 2006. Normal stem cells and cancer stemcells: the niche matters. Cancer Res. 66, 4553–4557.

Li, F., Tiede, B., Massague, J., Kang, Y.B., 2007a. Beyond tumorigen-esis: cancer stem cells in metastasis. Cell Res. 17, 3–14.

Li, H.C., Fan, X.L., Houghton, J., 2007b.Tumormicroenvironment: therole of the tumor stroma in cancer. J. Cell. Biochem. 101, 805–815.

Lin, H., 2002. The stem-cell niche theory: lessons from flies. Nat.Rev. Genet. 3, 931–940.

Loose, M., Patient, R., 2006. Global genetic regulatory networkscontrolling hematopoietic cell fates. Curr. Opin. Hematol. 13,229–236.

Madan Babu, M., Luscombe, N.M., Aravind, L., Gerstein, M.,Teichmann, S.A., 2004. Structure and evolution of transcriptionalregulatory networks. Curr. Opin. Struct. Biol. 14, 283–291.

Mangan, S., Zaslaver, A., Alon, U., 2003. The coherent feedforwardloop serves as a sign-sensitive delay element in transcriptionnetworks. J. Mol. Biol. 334, 197–204.

Marx, J., 2003. Mutant stem cells may seed cancer. Science 301,1308–1310.

Materna, S.C., Davidson, E.H., 2007. Logic of gene regulatorynetworks. Curr. Opin. Biotechnol. 18, 351–354.

McAdams, H.H., Arkin, A., 1997. Stochastic mechanisms in geneexpression. Proc. Natl. Acad. Sci. U. S. A. 94, 814–819.

McAdams, H.H., Arkin, A., 1999. It's a noisy business! Geneticregulation at the nanomolar scale. Trends Genet. 15, 65–69.

Metallo, C.M., Mohr, J.C., Detzel, C.J., de Pablo, J.J., Van Wie, B.J.,Palecek, S.P., 2007. Engineering the stem cell microenvironment.Biotechnol. Prog. 23, 18–23.

Miyamoto, T., Iwasaki, H., Reizis, B., Ye, M., Graf, T., Weissman, I.L.,Akashi, K., 2002. Myeloid or lymphoid promiscuity as a criticalstep in hematopoietic lineage commitment. Dev. Cell 3, 137–147.

Moore, K.A., Lemischka, I.R., 2006. Stem cells and their niches.Science 311, 1880–1885.

Nicolis, G., Prigogine, I., 1977. Self-Organization in NonequilibriumSystems. Wiley, New York.

Nilsson, S.K., Simmons, P.J., 2004. Transplantable stem cells: hometo specific niches. Curr. Opin. Hematol. 11, 102–106.

Nilsson, S.K., Johnston, H.M., Coverdale, J.A., 2001. Spatiallocalization of transplanted hemopoietic stem cells: inferencesfor the localization of stem cell niches. Blood 97, 2293–2299.

Nutt, S.L., Heavey, B., Rolink, A.G., Busslinger, M., 1999. Commit-ment to the B-lymphoid lineage depends on the transcriptionfactor Pax5. Nature 401, 556.

Orkin, S.H., 2000. Diversification of haematopoietic stem cells tospecific lineages. Nat. Rev. Genet. 1, 57–64.

Patel, K.J., Pambuccian, S.E., Ondrey, F.G., Adams, G.L., Gaffney, P.M.,2006. Genes associated with early development, apoptosis and cellcycle regulation define a gene expression profile of adenoid cysticcarcinoma. Oral Oncol. 42, 994–1004.

Pedraza, J.M., van Oudenaarden, A., 2005. Noise propagation ingene networks. Science 307, 1965.

Phillips, R.L., Ernst, R.E., Brunk, B., Ivanova, N., Mahan, M.A.,Deanehan, J.K., Moore, K.A., Overton, G.C., Lemischka, I.R.,2000. The genetic program of hematopoietic stem cells. Science288, 1635–1640.

Ptashne, M., Gann, A., 2002. Genes and Signals. Cold Spring HarborLaboratory Press, New York.

Qu, K., Haidar, A.A., Fan, J., Ensman, L., Tuncay, K., Jolly, M.,Ortoleva, P., 2007. Cancer onset and progression: a genome-wide, nonlinear dynamical systems perspective on onconet-works. J. Theor. Biol. 246, 234–244.

Raj, A., Peskin, C.S., Tranchina, D., Vargas, D.Y., Tyagi, S., 2006.Stochastic mRNA synthesis in mammalian cells. PLoS Biol. 4,1707–1719.

Ramalho-Santos, M., Yoon, S., Matsuzaki, Y., Mulligan, R.C., Melton,D.A., 2002. “Stemness”: transcriptional profiling of embryonicand adult stem cells. Science 298, 596–600.

Rao, C.V., Arkin, A.P., 2001. Control motifs for intracellularregulatory networks. Annu. Rev. Biomed. Eng. 3, 391–419.

Rao, C.V., Wolf, D.M., Arkin, A.P., 2002. Control, exploitation andtolerance of intracellular noise. Nature 420, 231–237.

Raser, J.M., O'Shea, E.K., 2004. Control of stochasticity in eukaryoticgene expression. Science 304, 1811.

Raser, J.M., O'Shea, E.K., 2005. Noise in gene expression: origins,consequences, and control. Science 309, 2010.

Raubenheimer, E.J., Noffke, C.E.E., 2006. Pathogenesis of bonemetastasis: a review. J. Oral Pathol. Med. 35, 129–135.

Ravasz, E., Somera, A.L., Mongru, D.A., Oltvai, Z.N., Barabási, A.L.,2002. Hierarchical organization of modularity in metabolicnetworks. Science 297, 1551–1555.

Rizo, A., Vellenga, E., de Haan, G., Schuringa, J.J., 2006. Signalingpathways in self-renewing hematopoietic and leukemic stem cells:do all stem cells need a niche? Hum. Mol. Genet. 15, R210–R219.

Roeder, I., Glauche, I., 2006. Towards an understanding of lineagespecification in hematopoietic stem cells: a mathematical modelfor the interaction of transcription factors GATA-1 and PU.1.J. Theor. Biol. 241, 852–865.

Rosenbauer, F., Tenen, D.G., 2007. Transcription factors in myeloiddevelopment: balancing differentiation with transformation.Nat. Rev. Immunol. 7, 105.

Ross, I.L., Browne, C.M., Hume, D.A., 1994. Transcription ofindividual genes in eukaryotic cells occurs randomly andinfrequently. Immunol. Cell Biol. 72, 177–185.

168 J.D. Halley et al.

Roth, F.P., Hughes, J.D., Estep, P.W., Church, G.M., 1998. FindingDNA regulatory motifs within unaligned noncoding sequencesclustered by whole-genome mRNA quantitation. Nat. Biotechnol.16, 939–945.

Rothenberg, E.V., Anderson, M.K., 2002. Elements of transcriptionfactor network design for T-lineage specification. Dev. Biol. 246,29–44.

Rothenberg, E.V., Telfer, J.C., Anderson, M.K., 1999. Transcriptionalregulation of lymphocyte lineage commitment. BioEssays 21,726–742.

Ruiz i Altaba, A., Sánchez, P., Dahmane, N., 2002. Gli and Hedgehogin cancer: tumours, embryos and stem cells. Nat. Rev. 2,361–372.

Russell, D.F., Wilkens, L.A., Moss, F., 1999. Use of behaviouralstochastic resonance by paddle fish for feeding. Nature 402,291–294.

Segal, E., Shapira, M., Pe'er, A.R.D., Botstein, D., Koller, D.,Friedman, N., 2003. Module networks: identifying regulatorymodules and their condition-specific regulators from geneexpression data. Nat. Genet. 34, 166–176.

Shivdasani, R.A., Orkin, S.H., 1996. The transcriptional control ofhematopoiesis. Blood 87, 4025–4039.

Sinha, S., 2005. Complexity vs. stability in small-world networks.Phys. A Stat. Mech. Appl. 346, 147–153.

Sternberg, P.W., Félix, M.A., 1997. Evolution of cell lineage. Curr.Opin. Genet. Dev. 7, 543–550.

Sulston, J.E., Horvitz, H.R., 1977. Post-embryonic cell lineages ofthe nematode, Caenorhabditis elegans. Dev. Biol. 56, 110.

Swain, P.S., Elowitz, M.B., Siggia, E.D., 2002. Intrinsic and extrinsiccontributions to stochasticity in gene expression. Proc. Natl.Acad. Sci. U. S. A. 99, 12795–12800.

Swiers, G., Patient, R., Loose, M., 2006. Genetic regulatorynetworks programming hematopoietic stem cells and erythroidlineage specification. Dev. Biol. 294, 525–540.

Taichman, R.S., 2005. Blood and bone: two tissues whose fates areintertwined to create the hematopoietic stem-cell niche. Blood105, 2631–2639.

Tavazoie, S., Hughes, J.D., Campbell, M.J., Cho, R.J., Church, G.M.,1999. Systematic determination of genetic network architecture.Nat. Genet. 22, 281–285.

Tomlin, C.J., Axelrod, J.D., 2007. Biology by numbers: mathematicalmodelling in developmental biology. Nat. Rev. Genet. 8, 331.

Traynelis, S.F., Jaramillo, F., 1998. Getting the most out of noise inthe central nervous system. Trends Neurosci. 21, 137.

Tsai, R.Y.L., 2004. A molecular view of stem cell and cancer cell self-renewal. Int. J. Biochem. Cell Biol. 36, 684–694.

Tyagi, S., 2007. RT-PCR enters the realm of stochastic geneexpression. Genetic Eng. Biotechnol. News 27.

Vogel, G., 2003. ‘Stemness’ genes still elusive. Science 302, 371.von Dassow, G., Meir, E., Munro, E.M., Odell, G.M., 2000. The

segment polarity network is a robust developmental module.Nature 406, 188–192.

Weatherall, D.J., 2001. Phenotype–genotype relationship in mono-genic disease: lessons from the thalassemias. Nat. Rev. Genet. 2,245–255.

Weintraub,H., Davis, R., Tapscott, S., Thayer,M., Krause,M., Benezra,R., Blackwell, T., Turner, D., Rupp, R., Hollenberg, S., et al., 1991.The myoD gene family: nodal point during specification of themuscle cell lineage. Science 761–766.

Wiesenfeld, K., Moss, F., 1995. Stochastic resonance and the benefitsof noise: from ice ages to crayfish and SQUIDs. Nature 373, 33–36.

Wilczynski, J.R., 2006. Cancer and pregnancy share similar mechan-isms of immunological escape. Chemotherapy 52, 107–110.

Wilmers, C.C., Sinha, S., Brede, M., 2002. Examining the effects ofspecies richness on community stability: an assembly modelapproach. Oikos 99, 363–367.

Wilson, A., Trumpp, A., 2006. Bone-marrow haematopoietic-stem-cell niches. Nat. Rev. Immunol. 6, 93–106.

Yin, T., Li, L.H., 2006. The stem cell niches in bone. J. Clin. Invest.116, 1195–1201.

Zlokarnik, G., Negulescu, P.A., Knapp, T.E., Mere, L., Burres, N.,Feng, L., Whitney, M., Roemer, K., Tsien, R.Y., 1998. Quantitationof transcription and clonal selection of single living cells withβ-lactamase as reporter. Science 279, 84–88.