Toward Decoding the Principles of Cancer Metastasis Circuits

2014;74:4574-4587. Cancer Res Mingyang Lu, Mohit Kumar Jolly, Jose' Onuchic, et al. Toward Decoding the Principles of Cancer Metastasis Circuits

Updated version

http://cancerres.aacrjournals.org/content/74/17/4574

Access the most recent version of this article at:

Cited Articles

http://cancerres.aacrjournals.org/content/74/17/4574.full.html#ref-list-1

This article cites by 100 articles, 33 of which you can access for free at:

E-mail alerts related to this article or journal.Sign up to receive free email-alerts

Subscriptions

Reprints and

[email protected]

To order reprints of this article or to subscribe to the journal, contact the AACR Publications Department at

Permissions

[email protected]

To request permission to re-use all or part of this article, contact the AACR Publications Department at

on September 2, 2014. © 2014 American Association for Cancer Research. cancerres.aacrjournals.org Downloaded from on September 2, 2014. © 2014 American Association for Cancer Research. cancerres.aacrjournals.org Downloaded from

http://cancerres.aacrjournals.org/content/74/17/4574

http://cancerres.aacrjournals.org/content/74/17/4574.full.html#ref-list-1

http://cancerres.aacrjournals.org/cgi/alerts

mailto:[email protected]

mailto:[email protected]

http://cancerres.aacrjournals.org/


Physics in Cancer Research

Toward Decoding the Principles of Cancer MetastasisCircuits

Mingyang Lu1, Mohit Kumar Jolly1,2, Jose' Onuchic1,3,4,5, and Eshel Ben-Jacob1,5,6,7

AbstractUnderstanding epithelial–mesenchymal transitions (EMT) during cancer metastasis remains a major

challenge in modern biology. Recent observations of cell behavior together with progress in mapping theunderlying regulatory genetic networks led to new understandings of carcinoma metastasis. It is nowestablished that the genetic network that regulates the EMT also enables an epithelial–mesenchymal hybridphenotype. These hybrid cells possess mixed carcinoma epithelial and mesenchymal characteristics thatenable specialized capabilities such as collective cell migration. On the gene network perspective, a four-component decision unit composed of two highly interconnected chimeric modules—the miR34/SNAIL andthe miR200/ZEB mutual-inhibition feedback circuits—regulates the coexistence of and transitions betweenthe different phenotypes. Here, we present a new tractable theoretical framework to model and decodethe underlying principles governing the operation of the regulatory unit. Our approach connects theknowledge about intracellular pathways with observations of cellular behavior and advances towardunderstanding the logic of cancer decision-making. We found that the miR34/SNAIL module acts as anintegrator while the miR200/ZEB module acts as a three-way switch. Consequently, the combined unit cangive rise to three phenotypes (stable states): (i) a high miR200 and low ZEB, or (1, 0) state; (ii) a low miR200and high ZEB, or (0, 1) state; and (iii) a medium miR200 and medium ZEB, or (1/2,

1/2) state. We associate thesestates with the epithelial, mesenchymal, and hybrid phenotypes, respectively. We reflect on the consistencybetween our theoretical predictions and recent observations in several types of carcinomas and suggest newtestable predictions.See all articles in this Cancer Research section, "Physics in Cancer Research."Cancer Res; 74(17); 4574–87. �2014 AACR.

IntroductionUnderstanding cell fate decisions during embryonic devel-

opment and tumorigenesis remain a major research challengein modern developmental and cancer biology (1). In recentyears, we havewitnessed rapid progress inmapping the geneticregulatory networks that determine the fate of different cells.Examples include networks that govern the transition betweenepithelial and mesenchymal phenotypes, networks that con-

trol the differentiation of pluripotent stem cells into differentlineages and/or progenitor cells, networks that are involved inthe transition into and from cancer stem–like cells (CSC), aswell as networks that are involved in cellular dedifferentiationduring the formation of induced pluripotent stem cells (iPSC;refs. 2–5). In all these examples, a cell's fate is orchestrated bychanges in the expression of transcription factors (TF) andmicroRNAs (miRNA; miR) that in turn govern downstreamregulatory networks, ultimately leading to the genome-widegene expression patterns and corresponding protein levelsspecific to a particular cell lineage (fate).

An important representative example of cell fate decisionis seen during metazoan embryonic development, whensessile epithelial cells reversibly attain mesenchymal-likecharacteristics that allow them to migrate and invade adja-cent tissues. Successive rounds of these forward and back-ward transitions, namely epithelial–mesenchymal transition(EMT) and mesenchymal–epithelial transition (MET), play acrucial role in forming many internal organs. Cancer cellsin various carcinomas (including lung, breast, prostate,colon, pancreas, and ovaries) adopt this embryonic processof EMT and MET during invasion and metastasis. Thisaberrant activation of EMT by carcinomas is consideredto be an important hallmark of cancer metastasis, an

Authors' Affiliations: 1Center for Theoretical Biological Physics, Depart-ments of 2Bioengineering, 3Physics and Astronomy, 4Chemistry, and5Biochemistry and Cell Biology, Rice University, Houston, Texas; 6Schoolof Physics and Astronomy; and 7The Sagol School of Neuroscience, Tel-Aviv University, Tel-Aviv, Israel

Note: Supplementary data for this article are available at Cancer ResearchOnline (http://cancerres.aacrjournals.org/).

M. Lu and M.K. Jolly contributed equally to this work.

Corresponding Authors: Eshel Ben-Jacob, Tel-Aviv University, P.O.B.39040, Tel-Aviv 69978, Israel. Phone: 972-3-640-7845; Fax: 972-3-642-5787; E-mail: [email protected]; and Jose' Onuchic, Rice Univeristy, P.O.B.1982, Houston, TX 77005. Phone: 713-348-4197; E-mail:[email protected]

doi: 10.1158/0008-5472.CAN-13-3367

�2014 American Association for Cancer Research.

CancerResearch

Cancer Res; 74(17) September 1, 20144574

on September 2, 2014. © 2014 American Association for Cancer Research. cancerres.aacrjournals.org Downloaded from


outcome that is responsible for more than 90% of cancerdeaths (6, 7). The core decision network that regulatescarcinoma EMT and MET also allows for transition into ahybrid epithelial–mesenchymal (E/M) phenotype, which hascombined epithelial (cell–cell adhesion) and mesenchymal(motility) traits. These mixed characteristics of the hybridE/M phenotype enable cells to migrate collectively, as com-pared with the single-cell migration of purely mesenchymalcells (8). The hybrid E/M state, also referred to as partialEMT (pEMT), can revert to the epithelial phenotype whenrequired, as seen during the final stages of branchingmorphogenesis and wound healing (9–12). More recently,this hybrid state has been observed at different stages ofepiboly (13), organ morphogenesis (14), and metastasis(15–17).Carcinoma metastases typically begin when epithelial

cells from the primary tumor lose their apical-basal polarityand cell–cell adhesion and acquire migratory and invasivemesenchymal traits. The newly transformed motile cellsnavigate through the extracellular matrix (ECM) towardblood vessels. It has been shown that cancer cells canmigrate as dispersed individuals with mesenchymal charac-ter or in a coordinated collective motion of cells with ahybrid character (10, 11, 17). Collective migration obviatesthe need for all cells to be able to detect extrinsic signals formigration (12), enabling the cohort of cells to adeptly adaptto different microenvironments. Thus, hybrid E/M cells havean important functional role for successful migrationthrough the ECM. After reaching blood vessels, some of thecells succeed in penetrating into the bloodstream (intrava-sation) and stay as circulating tumor cells (CTC) until theyexit (extravasation) to reach an appropriate colonization sitein a distant organ. Subsequently, the cells undergo the METand regain their epithelial characteristics, growing later intomacrometastases (6, 16).Notably, while hybrid E/M cells have been observed during

forward EMT, they have not been observed during the reverse,MET (18, 19). This implies an innate asymmetry in the under-lying mechanisms regulating the epithelial and mesenchymalcell fate. Also, many CTCs coexpress both epithelial andmesenchymal markers (17), hence demonstrating the signifi-cance of cellular plasticity in metastasis and colonization. Thetumor cells that undergo EMT are usually resistant to bothchemotherapy and radiotherapy (20, 21), and have propertiessimilar to that of CSCs (22–24). Thus, understanding thebackward and forward transitions between these three cellphenotypes [epithelial (E), mesenchymal (M), and E/M] couldinform strategies against metastases.

The Core Decision UnitCell fate determination between these three phenotypes is

regulated by many internal and external signals such ashypoxia-inducible factor 1 (HIF1), p53, TGFb, HGF, FGF, EGF,Notch, and Wnt (25). These signals converge on a core regu-latory unit composed of four components, two families oftranscription factors that induce EMT, SNAIL and ZEB, andtwo families miRNAs that inhibit EMT, miR34, andmiR200 (5).The epithelial phenotype corresponds to high levels of miR34

and miR200, whereas the mesenchymal phenotype corre-sponds to high levels of SNAIL and ZEB. SNAIL and ZEBrepress epithelial-specific gene expression including E-cad-herin, the hallmark of the epithelial phenotype, and promotethe expression of mesenchymal markers such as N-cadherinand vimentin (26). The EMT-inducing signals (HIF1, TGFb,HGF, FGF, EGF, Notch, and Wnt) activate SNAIL and ZEB(24, 26, 27, 28), whereas EMT-repressing signals (p53) activatemiR34 and miR200 (29). The input signals have elaboratemutual interactions such as the degradation of p53 viaTGFb-mediated activation of MDM2 (Fig. 1; ref. 30).

The core decision unit is also known to play an importantrole as "a motor of cellular plasticity" in various humancarcinomas, as it is coupled to a variety of other key cellularfeatures, including stemness, cell-cycle arrest, apoptosis andsenescence, resistance to chemotherapy, and cell–cell com-munication, as discussed in the last section (31, 32).

From a clinical perspective, SNAIL and ZEB are over-expressed at the tumor–stroma interface and correlatesignificantly with many interrelated features including over-all survival, poor prognosis, as well as tumor subtypes andgrades associated with worse outcomes (6, 26). These obser-vations reinforce the importance of EMT in carcinomametastasis, as SNAIL and ZEB are the major drivers of EMT.

The four components of the core regulatory unit form twohighly interconnected modules—the miR34/SNAIL and themiR200/ZEB modules (Fig. 1). Each of these modules is adouble-negative or a mutually inhibiting miRNA–TF chimericfeedback loop (33–35), also referred to as a chimera switch (36).In the miR34/SNAIL module, miR34 (m34) binds to two con-served sites on the 30-untranslated region (UTR) of the SNAIL(S) mRNA, whereas SNAIL represses miR34 transcriptionallythrough binding at one site in its promoter region (35). In themiR200/ZEB module, the miR200 family (m200)—mir-141,miR200a/b/c, and miR429—has a total of eight binding sitesin theZEB1 30-UTR andnine in theZEB2 30-UTR, whereas ZEB1and ZEB2 (considered together as the ZEB family) bind to themiR200 family promoter region (33, 34). Because stable expres-sion of miR200c alone is sufficient to reverse EMT and restoreE-cadherin expression (37), in Fig. 1, we considered six bindingsites of themiR200 (m200) family to ZEB (Z) mRNA, the numberof miR200c binding sites on the ZEB family, and three bindingsites for ZEB on the miR200 promoter, the number of ZEBbinding sites on miR200c. SNAIL activates ZEB (38) and istranscriptionally self-inhibiting (39). ZEB has two potentialbinding sites in its promoter (40) and activates itself indirectlyby stabilizing the SMADcomplexes (41), thuswe represent ZEBas a transcriptional self-activating gene of rank 2 (two sites inits promoter region). It may be noted that these interactionsare not specific for a particular cancer, but rather hold true formost human carcinomas undergoing EMT (see SupplementaryData for a list of these interactions as identified in variouscarcinoma cell lines).

Modeling miRNA-Based Chimeric CircuitsAimed for readers fromdiverse disciplines, we include in this

section more detailed description of modeling miRNA-basedcircuits. Understanding the mathematical details is not

Toward Decoding Cancer Decision Circuits

www.aacrjournals.org Cancer Res; 74(17) September 1, 2014 4575



essential for understanding the conclusions reached and theirimplications about the crucial role of epithelial plasticity, aconserved phenomenon among carcinomas.

BecausemiRNAs are key components of the EMT regulatorycircuits, we developed a theoretical framework for modelingmiRNA-based chimera (MBC) circuits (36). This frameworkcaptures the essential features of miRNA-mediated regulationby considering the binding/unbinding chemical reactions forboth TF–promoter and miRNA–mRNA complexes (Fig. 2).In the model, when one or multiple miRNAs bind to mRNA,the miRNAs are able to silence mRNA translation by inhibitingthe translation process (translation inhibition) and/or activelydegrading the mRNA-containing complex (Fig. 2A). Considerthe case in whichmiRNA (m)molecules target themRNA (m) ofa protein (B), and there are n distinct miRNA binding sites onthe mRNA. In this case, the deterministic equations are givenby

_m ¼ gm �mYm � kmm

_m ¼ gm �mYm � kmm ð1Þ_B ¼ gBmL� kBB

where gm and gm are the synthesis rates of m and m, respec-tively, and these could depend on the concentration of someexternal signals. km, kB, and km are the innate degradation ratesof m, B, and m, respectively, and gB is the innate mRNA'stranslation rate for protein B. Equation 1 also contains threem-dependent functions to quantify the miRNA–mRNA cou-pling, that is, the translational inhibition term (L), the mRNAactive degradation term (Ym), and the miRNA active degrada-tion term (Ym). The formulae and their derivation can be foundin the Supplementary Data of ref. 36. As shown in Fig. 2, as mincreases, L is markedly repressed (Fig. 2B), while Ym increasesslightly and Ym increases markedly (Fig. 2C). Compared withthe previously derivedmodels for miRNA–TF chimeric circuits(42–50), the new approach is more consistent with the biologicmechanism. Because miRNAs are typically more stable than

mRNAs (49), we sometimes can reduce Equation 1 to twocoupled equations for B andm by assuming thatm is always atsteady states (36).

The Unit ModulesChimera circuits

Applying the theoretical framework for miRNA-based cir-cuits, we studied the dynamics of the two miRNA–TF chimeratoggle switches in the core regulatory unit. Typically, theoperating characteristics of circuit modules depend on thenonlinearity of the inhibitory regulations by TFs/miRNAs andthe nonlinearity of the auto-regulations of the TFs (Fig. 1). Tobetter understand the relationship between nonlinearity andmultistability, we examined the behavior of several similarcircuit modules.

We started with the miR200/ZEB module, which is a typicalmiRNA–TF chimera switch (Fig. 3A). To simplify the problem,we first omitted the ZEB self-activation in the initial stage. TheEMT transcription factor SNAIL transcriptionally regulatesboth miR200 (inhibition) and ZEB (activation). ConsideringSNAIL as an external control input signal S, we obtained thefollowing deterministic equations for miR200 (m200), ZEBmRNA (mZ), and ZEB protein (Z) as

_m200 ¼ gm200HSðZ; lZ;m200

ÞHSðS; lS;m200Þ �mZYmðm200Þ

� km200m200

_mZ ¼ gmZHSðS; lS;mZ Þ �mZYmðm200Þ � kmZmZ

_Z ¼ gZmZLðm200Þ � kZZ

ð2Þ

Here, the shifted Hill function, defined as HSðX; lÞ¼ H�ðXÞ þ lHþðXÞ, where H�ðXÞ ¼ 1=½1 þ ðX=X0ÞnX �,HþðXÞ ¼ 1�H�ðXÞ, and nX is the Hill rank for X (which isusually associated with the number of binding sites of TF X onthe promoter); l is a positive number, which quantifies the"fold change" of the synthesis rate caused by X. According to

© 2014 American Association for Cancer Research

Cancer Research: Physics in Cancer Research

miR34

miR200 m200

m34

TGFb

6

3

2

2

2

2

1

1

S

Z

1

ZEB

MDM2 SMAD 3/4

SNAIL

p53

miR34

miR200 mmmmmm20220020000000

mmmmm3343434444mmmm

6

3

2

2

2

2

1

1

SSSS

ZZZ

1

ZEB

SNAIL

Figure 1. The regulatory networkgoverning EMT and MET, and itsregulation by p53 and TGFb. Thecore unit (shown in dotted box)comprises two linked miRNA, TFchimeric circuits, the mutuallyinhibiting loops of miR34/SNAILand miR200/ZEB. A solid arrow(straight or curved) representstranscriptional activation, and asolid bar represents transcriptionalinhibition. Dashed lines indicatemiRNA translational regulation anddotted lines denote indirect links.The numbers listed alongregulatory lines represent thenumber of corresponding bindingsites as deduced fromexperiments(see Supplementary Data andref. 51 for details).

Lu et al.

Cancer Res; 74(17) September 1, 2014 Cancer Research4576



this definition (51), the activation corresponds to l > 1, whilethe inhibition corresponds to l < 1.From Fig. 3A, the miR200/ZEB switch is seen to be bistable

(two-way switch) for a certain set of biologically relevantparameters (see also Definitions box). We also tested themultistability of the system by adjusting a wide range ofparameters. The system can be monostable for some para-meters, but not tristable (three-way switch).Given the experimental knowledge about the miR200/ZEB

module, ZEB mRNA has six miRNA binding sites for themiR200 family (including miR141, miR200a/b/c, and miR429;see Fig. 1), so this translation inhibition is highly nonlinear.Similarly, ZEB transcriptionally inhibits miR200 with high

nonlinearity (Hill rank is 3). Such high nonlinearity rendersthe circuit to have multiple steady states.

We further examined the effect of the nonlinearity (thevalue of the Hill rank) on a generic chimera toggle switch.We found that (see Supplementary Data in ref. 36), ingeneral, when there is one miRNA binding site, the systemis monostable, even for a very high rank of transcriptionalinhibition. For the case of two miRNA binding sites, thesystem is monostable when the Hill rank is less than orequals to three for the TF transcription inhibition and isbistable when the Hill rank is 4 or more. The system is easilybistable (even for low Hill rank of transcriptional inhibition)when there are more miRNA binding sites.



0 1 2 30.0

0.2

0.4

0.6

0.8

1.0

0 1 2 30.0

2.0

4.0

6.0

8.0

10.0

0 1 2 30.00

0.05

0.10

0.15

0.20

m: mRNA

mRNA active

degradation

miRNA active

degradation

mYm (m)

miRNA degradation

m/m0 m/m0

L (Tra

nsla

tio

n in

hib

itio

n f

un

cti

on

)

Sh

ifte

d H

ill fu

ncti

on

Ym

, Y

m(A

cti

ve d

eg

rad

ati

on

fu

nc

tio

ns)

mRNA degradationTotal silencing

TF/TF0

Ym (m)

Ym(m)

Ym(m)

L (m)

L (m)/[1 + Ym (m)/km]

Translation

inhibition

m: miRNAm–m complex

km km

ØA

B C D

ØØØ

Figure 2. Theoretical framework forMBCcircuit. A, illustration of theMBCmodel. AmiRNA–mRNA (m–m) complex is formedwhenone ormoremiRNAs (m) bindto anmRNA (m). km and km are innate degradation rates of miRNA andmRNA, respectively. OncemiRNA binds to mRNA, it silences the protein translation bytwomechanisms: it inhibits the translation rate [a process that is scaled by L(m)] and promotes active degradation of both mRNA [at rate Ym(m)] andmiRNA [atrate mYm(m) for m miRNAs]. P(m) represents the effects of both active degradation and translational inhibition. B, the L (translation inhibition) functionwith respect to miRNA (scaled by the threshold m0). In this example, there are six miRNA binding sites on the mRNA. A vertical dotted line is presented toillustrate the values atm0. C, theYmandYm functionswith respect tomiRNA,Ym(m), andYm(m) (scaledby the thresholdm0). D, the shiftedHill functionwith respectto the TFconcentrations (scaledby the threshold TF0—TF/TF0). In this example, theHill rank is 3 (asdepicted on the x-axis), the fold changel is 10 (as depictedon the y-axis).





Three-way switch: chimera switch with self-activationNext, we considered the miR200/ZEB module with the

ZEB self-activation (Fig. 3B). The deterministic equations formiR200 (m200), ZEB mRNA (mZ), and ZEB protein (Z) are

_m200 ¼ gm200HSðZ; lZ;m200


� km200m200

_mZ ¼ gmZHSðZ; lZ;mZ ÞHSðS; lS;mZ Þ �mZYmðm200Þ � kmZmZ

_Z ¼ gZmZLðm200Þ � kZZ ð3ÞAs mentioned earlier, the module can be bistable when the

ZEBself-activation isomitted(Fig.3A).Typically, self-activationonbothsidesof the toggle switchcan render thecircuit tristable(ternary or three-way) switch (52–55). However, self-activation

on one side can also render the circuit tristable for a wide rangeof parameters, as seen in both TF–TF toggle switches andmiRNA–TF chimera toggle switches (36). This is consistentwith our modeling on the miR200/ZEB module (Fig. 3B).A typical phase space diagram is shown in Fig. 3B. The circuit

is seen to have three coexisting stable states (three filled circlesat points of intersection of the curves), which correspond to (i)high miR200 and low ZEB [denoted as (1, 0) and furthest to theright], (ii) medium miR200 and medium ZEB [denoted as (1/2,1/2) and in the middle], and (iii) low miR200 and high ZEB[denoted as (0, 1) and furthest to the left]. The states (1, 0) and(0, 1) correspond to the E and M phenotypes, respectively(33, 34). We suggest that the intermediate (1/2,

1/2) state shouldbe associated with the hybrid E/M phenotype. Thus, the



0 5 10 15 200

200

400

600

800

1,000

1,200

ZE

B m

RN

A (m

ole

cu

les)

miR200 (K molecules)

A B

C D

miR200

m200

S

S

Z6

3

22

ZEB

0 10 20 30 400

100

200

300

400

Pro

tein

SN

AIL

(K

mo

lecu

les)

miR34 (K molecules)

miR34

m34m34

2

1

1I

S

I

SNAIL

0 5 10 15 25200

200

400

600

800

1,000

1,200

ZE

B m

RN

A (m

ole

cu

les)


m200 Z3

22

2

6

1

1

1

2

0 5 10 15 25200

200

400

600

800

1,000

ZE

B m

RN

A (m

ole

cu

les)


miR200

m200

S

Z6

3

22

2

ZEB

Figure 3. Multistability of various types ofmiRNA–TF chimera toggle switches. The plots show the nullclines and all possible steady states in the phase spacesformed by the concentrations of the two molecules. A, miR200/ZEB chimera toggle switch (without ZEB self-activation) driven by SNAIL. The circuit can bebistable for a wide range of parameters. Red nullcline is for the condition dm200/dt ¼ 0 and dZ/dt ¼ 0, and blue nullcline is for dmZ/dt ¼ 0 and dZ/dt ¼ 0. B,miR200/ZEB chimera toggle switch with ZEB self-activation driven by SNAIL. The circuit can be tristable for some parameters. Nullclines are plottedin the sameway as in A. C,miR34/SNAIL chimera toggle switch driven by a generic external signal I. The circuit is found to be onlymonostable. Red nullcline isfor dm34/dt¼ 0 and dS/dt¼ 0, and blue nullcline is for dmS/dt¼ 0 and dS/dt¼ 0. D, the combined miR34/SNAIL and miR200/ZEB circuit driven by a genericexternal signal I. The combined circuit can also be tristable. Nullclines are plotted in the same way as in A.

Lu et al.




miR200/ZEB module serves as the three-way decision switchthat enables the cells to adopt a hybrid E/M state or acompletely epithelial or mesenchymal state (51). The Hill rankfor ZEB self-activation was chosen to be 2 for the reasonsdescribed earlier. It was shown that both roles of miRNA—translational inhibition and active mRNA degradation (see Fig.1)—are required for tristability and for the ZEB/miR200 mod-ule to operate as a three-way switch (51).

An integrator: a chimera switch with self-inhibitionThe miR34/SNAIL module (Fig. 3C) represents a chimera

toggle switch with TF self-inhibition (in this case, SNAIL). Inthe model, the module is driven by an external input signal I tocapture the effect of various signals that target the gene (e.g.,TGFb and HIF1). The deterministic equations for miR34 (m34),SNAIL mRNA (mS), and SNAIL protein (S) are

_m34 ¼ gm34HSðS; lS;m34

Þ �mSYmðm34Þ � km34m34

_mS ¼ gmSHSðS; lS;mS ÞHSðI ; lI ;mS Þ �mSYmðm34Þ � kmSmS ð4Þ

_S ¼ gSmSLðm34Þ � kSS

A typical phase space diagram is shown in Fig. 3C. As seen,there exists only one stable steady state, which corresponds tofixed levels of SNAIL and miR34. When the signal I increases,the SNAIL level increases smoothly and the miR34 leveldecreases smoothly.

There are only twomiR34 binding sites on SNAILmRNA andZEB inhibits miR34 transcriptionally with weak nonlinearity(Hill rank is 1; ref. 35). Thus, themodeling results are consistentwith the findings mentioned above, where we considered thechimera toggle switch without self-inhibition. Further inves-tigations showed that self-inhibition usually makes the systemeven less likely to have multistability. Moreover, the self-inhibition reduces the effect of external noise in variousEMT-inducing signals and determines the sensitivity thresholdto those signals (56). The miR34/SNAIL module serves as anoise-buffer signal integrator (51). Such noise-buffering featureand relatively weak dependency of the stable sate enables themodule to be more reliable and prevents erroneous activationof EMT by some transient signals. The proposed role of miR34/SNAIL also explains why the epithelial phenotype is stable (57).

The Dynamics of the Combined Regulatory UnitToward characterizing the combined system of the two

chimera modules here, we analyzed the complete core regu-latory unit (Fig. 3D), which includes the decision module(miR200/ZEB) and the integrator module (miR34/SNAIL). Thecombined circuit is also driven by an external signal I thatactivates SNAIL transcriptionally. The deterministic equationsfor miR200 (m200), ZEB mRNA (mZ), ZEB protein (Z), miR34(m34), SNAIL mRNA (mS), and SNAIL protein (S) are

_m200 ¼ gm200HSðZ; lZ;m200


� km200m200

_mZ ¼ gmZHSðZ; lZ;mZ ÞHSðS; lS;mZ Þ �mZYmðm200Þ � kmZmZ

_Z ¼ gZmZLðm200Þ � kZZ

_m34 ¼ gm34HSðS; lS;m34

ÞHSðZ; lZ;m34Þ �mSYmðm34Þ � km34

m34

_mS ¼ gmSHSðS; lS;mS ÞHSðI ; lI ;mS Þ �mSYmðm34Þ � kmSmS

_S ¼ gSmSLðm34Þ � kSS

ð5ÞAs shown in Fig. 3D, the combined core decision unit acts as

a three-way switch (has three stable states). In Fig. 4, bifurca-tion curves were plotted for the cases with and without

DefinitionsNullclines: lines or curves, in our case, depicted as a

function of the concentration of molecules (ZEB mRNA,SNAIL protein, miR-34 and miR-200). The concept of null-clines is useful in analyzing the stability of a system. In asystem of differential equations, the nullclines are thecurves obtained when the concentrations do not changewith time (solution curves for which all of the differentialequations are equal to zero). In a graph with two nullclines,any intersection of the nullclines is an equilibrium point ofthe system. From an examination of the graphs of thenullclines, it is possible to infer whether or not a systemwill be multi-stable. The nullclines must intersect in morethan one place for the system to be multi-stable. If theyintersect in only one place, there is one single equilibriumpoint and the system cannot be bistable. If there are morethan two intersections (three or five, for example), themiddle equilibrium point(s) are often unstable saddlepoint(s). For example if the nullclines cross over at threefixed points, the middle is a saddle-node fixed point, andthis is an "ideal" bistability plot.Hill Rank: in the binding of a ligand to amacromolecule,

ligand binding is often enhanced if there are already otherligands present on the same macromolecule, a phenome-non known as cooperative binding. The Hill rank (alsocalled Hill coefficient) provides a way to quantify this effect,by describing the fraction of macromolecules saturated byligand as a function of the ligand concentration. It is used indetermining the degree of cooperativeness of the ligandbinding to the macromolecule. The derived Hill rank mea-sures howmuch the binding of ligand to one site affects thebinding of ligand to other similar sites. A Hill rank of 1indicates the binding of any ligand is completely indepen-dent of any other ligand bound. Hill ranks greater than oneindicate "positive cooperativity", while numbers less thanone indicates "negative cooperativity".Bifurcation plot: in the current study shows all the

steady states of the circuit as a function of a parameter,named as the bifurcation parameter. Typically, the solidline segments represent the stable steady states, while thedashed line segments represent the unstable steady states.Depending on the bifurcation parameter, the system canhave one or multiple steady states. Bifurcation plot can beused to check how the value of the bifurcation parameterchanges the system's multistability. It can also evaluate thesensitivity of the system behavior to the parameter.





feedback from ZEB to miR34. When the ZEB mRNA levels ofthe steady states were plotted with respect to the SNAILprotein levels, as shown in Fig. 4A, the feedback did not changethe bifurcation curve. When they were plotted with respect tothe external signal levels I, as shown in Fig. 4B, the feedback(inhibition of miR34 by ZEB) shifted the bifurcation curve tothe left, especially for high levels of mRNA. Thus, the inhibitoryfeedback plays important roles in amplifying the symmetrybreaking between the forward (EMT) and backward (MET)transitions. During EMT, cells are more likely to go throughthe hybrid (E/M) phenotype, while during MET, cells are likelyto go directly from M to E, without attaining the hybrid stateen route. This result is consistent with the experimentalobservations that the hybrid phenotype has not yet beenobserved duringMET, for example, in reprogramming to iPSCs(18, 19), but frequently observed during EMT, for example,tumor invasion and metastasis (15–17).

Investigating a combined networkA gene regulatory network, although complicated, usually

consists of many circuit modules that interact with each other.In the current study, we found that the core regulatory circuitof EMT contains the miR34/SNAIL module, which serves as anoise-buffer signal integrator, and the miR200/ZEB module,which acts as a three-way decision circuit or switch. In theprevious sections, we have shown how we first analyzed thebehavior of each individual circuit module and then charac-terized the whole circuit by adding the coupling between thetwo modules. There are a total of three regulatory links

between miR34/SNAIL and miR200/ZEB, that is, inhibition ofmiR200 by SNAIL, activation of ZEB by SNAIL, and inhibition ofmiR34 by ZEB (Fig. 1). To simplify the problem,we included theregulation ofmiR200 andZEBby SNAIL into the circuitmoduleof miR200/ZEB, and regarded SNAIL as an external signal ofthis module. By doing this, we reduced the number of linksbetween two modules to one without sacrificing the accu-racy in describing the full circuit. From the analysis men-tioned in the previous section, we found that the feedbackfrom ZEB to miR34 does not dramatically affect the func-tions and stand-alone dynamics of the two modules. Instead,the feedback adds more asymmetry between forward (EMT)and backward (MET) transitions. We also found that theregulation of the miR200/ZEB module on SNAIL is unaf-fected by the feedback, which validates the decompositionapproach we adopted.

We propose the following multistep approach to investigatemore elaborated networks: (i) identify and study the "stand-alone" dynamics of the basic modules; (ii) formulate/devise"solvability" conditions of the mutual feedbacks (constraints)between the modules when functioning as a combined unit.More specifically, first, the combined network is decomposedinto different circuit modules in such a way that each moduleonly has a few elements (TFs or miRNAs) linked with the othermodules. The most straightforward way (although not neces-sarily themost efficient) is to simply have two elements in eachmodule. Second, eachmodule is analyzed by the so-called two-signal bifurcation technique, which provides informationabout the multistability with respect to various possible



160 180 200 220 2400

200

400

600

800

1,000

1,200

ZE

B m

RN

A (m

ole

cu

les)

Protein SNAIL (K molecules)

{E} {E, E/M, M} {E/M, M} {M}

M–(0, 1)

E–(1, 0)

E/M–(1/2, 1/2)

A

20 40 60 80 1000

200

400

600

800

1,000

1,200

ZE

B m

RN

A (m

ole

cu

les)

Signal I (K molecules)

B{E, M}

Figure 4. Bifurcation of the ZEB mRNA levels for the combined circuits driven by an external signal I. A, dependency of ZEB mRNA levels on SNAILproteins levels. The blue solid lines are the stable steady states, and the red dashed lines are the unstable steady states. The bifurcation shows thatdepending on SNAIL level, there are possibly three coexisting states: the high miR200 and low ZEB, or (1, 0) state that we associate with the epithelialphenotype (E), the (0, 1) state that we associate with the mesenchymal phenotype (M), and the (1/2,

1/2) state that we associate with the hybridphenotype (E/M). Black arrows and lines illustrate possible transitions when the signal level varies. The bifurcation curves do not change regardless ofthe feedback from ZEB to miR34. The colors in the background show the phases illustrated in Fig. 5A. B, dependency of ZEB mRNA levels on the signallevels I. The blue/red bifurcation curves are for the combined circuit with feedback from ZEB to miR34. The navy (stable states)/brown (unstable states)bifurcation curves are for the combined circuit without the feedback. Compared with the situation without feedback from ZEB to miR34, theinhibitory feedback makes the bifurcation curves shift a little to the left.

Lu et al.




combinations of the two signals. Third, the coupling amongdifferent modules is included by solving proper solvabilityconditions, which ensure consistency in the dynamics of thevarious modules. By consistency, we mean that the resultantlevels of expression of the elements (TFs and miRNAs) areconsistent with the levels of expression obtained from thestand-alone dynamics of the modules. Here, we elaboratebelow on the two-signal bifurcation for the specific exampleof the miR200/ZEB module.

The phenotypic phase diagramThe idea is illustrated in Fig. 5 in which we show the

phenotypic phase diagram for the miR200/ZEB module whendriven by two input signals representing the action of SNAIL inthe combined circuit. Therefore, miR200 is driven by aninhibitory signal S1 and ZEB is driven by an activator signalS2. The resultant phase diagram (shown in Fig. 5A) shows the

existence of seven different phases. From a dynamical systemperspective, each of the phases corresponds to a differentnullcline describing monostability or multistability (coexis-tence) of different states (nullclines of three different phasesare shown in Fig. 5B–D).

Consistency with Experimental ObservationsUsing our framework formiRNA-based chimeric circuits, we

decipher the operation principles of the EMT regulatory net-work (Fig. 3). First, the miR34/SNAIL acts as a noise-buffersignal integrator that prevents aberrant activation of EMT dueto transient signals and explains the stability of epithelialphenotype (57). Second, the miR200/ZEB acts as a three-way(ternary) switch, which explains the existence of three differentphenotypes—the canonical E and M ones, and the morerecently discovered hybrid phenotype (E/M).



{E}

{E, E/M}

{E, E/M, M}

{E/M, M}

E/M, M

{E/M}

{M}

0 100 200 4003000

100

200

300

400

Activation signal S2 (K)

Inh

ibit

ion

sig

nal S

1 (K

)

{E, M}

m200

S2S1

S1 = 300K, S2 = 152K

Z

B

0 5 10 15 25200

200

400

600

800

1,000

ZE

B m

RN

A (m

ole

cu

les)


E, E/M

S1 = 300K, S2 = 124K

C

A

0 5 10 15 25200

200

400

600

800

1,000

ZE

B m

RN

A (m

ole

cu

les)


E/M

S1 = 300K, S2 = 136K

D

0 5 10 15 25200

200

400

600

800

1,000

ZE

B m

RN

A (m

ole

cu

les)


Figure 5. The phase diagram, bifurcation, and some typical nullclines of the miR200/ZEB decision module driven by two independent signals S1 and S2

representing the action of SNAIL in the combined circuit with miR200 driven by an inhibitory signal (S1) and ZEB by an activator signal (S2; see Fig. 1). A, thetwo-signal bifurcation plot for the circuit (illustrated as an inset). Each phase is unique and may include a combination of coexisting steady states. Forinstance, in phase {E}, only the state (1, 0) exists and in phase {E, E/M}, the states (1, 0) and (1/2,

1/2) coexist. B, the nullclines for a system in the {E/M, M}phase, where S1 ¼ 300 K, S2 ¼ 152 K. C, the nullclines for a system in the {E, E/M} phase, where S1 ¼ 300 K, S2 ¼ 124 K. D, the nullclines for asystem in the {E/M} phase, where S1 ¼ 300 K, S2 ¼ 136 K. The nullclines from B–D were plotted similarly to those in Fig. 3.





Our theoretical predictions are consistent with experimen-tal observations. For example, the activation of SNAIL caninitiate EMT by repressing CDH1 (the gene for E-cadherin), butZEB1 is required for complete inhibition of E-cadherin (56),and hence the completion of EMT. Similarly, it has beenobserved that a complete reversal to an epithelial phenotyperequires a strong inhibition of ZEB1 (58), as the knockdown ofSNAIL is not sufficient (59). Furthermore, the cells that attainhigh ZEB levels, for example, by being continuously treatedwith the signal TGFb, do not immediately revert to an epithelialphenotype when the signal is removed. However, the cells withremarkably low levels of ZEBdo revert, indicating thatmiR200/ZEB is the module that acts as the commitment point for cellsundergoing EMT (27). Finally, ZEB1 transcriptionally inhibitsmost of the genes that are downregulated during EMT (60),suggesting that ZEB is the master regulator for cell fatedecision-making during EMT.

Recent studies of gastrulation in Drosophila embryos (61)show that collectively migrating cells, the hallmark of thehybrid E/M state, coexpress ZEB1 and E-cadherin, thusvalidating that the hybrid state has intermediate levels ofboth miR200 and ZEB. Also, the (1/2,

1/2) state for the hybridphenotype allows for cell–cell communication via the Jag1–Notch–Delta system (32), as observed during collectivemigration in wound healing (62). Because Jag1 mRNA hasfive binding sites for miR200 and is thus strongly repressedby miR200 (32), its expression in the hybrid state is possibleonly when the hybrid state has either low or intermediatelevels of miR200, instead of the high levels that are charac-teristic of epithelial cells. This implies that the level ofexpression of miR200 in the hybrid phenotype of collectivelymigrating cells that maintain the Notch–Jag1 signaling hasto be lower than its high level of expression in the epithelialphenotype. Thus, the results validate our hypothesis that theobserved hybrid phenotype corresponds to the theoreticalpredicted (1/2,

1/2) state.Of note, after the first version of this manuscript was

submitted, a phenotypic study of 43well-characterized ovariancarcinoma cell lines identified an EMT spectrum among them.Twenty-six of the 43 cell lines had a hybrid phenotype, werehighly aggressive, resistant to anoikis, had enhanced sphere-forming ability, and expressed both ZEB1 and E-cadherin, thusindicating collective cell migration (63). This suggests thatmost carcinoma cells undergo partial or incomplete EMT, anda complete EMT may not be necessary for cell survival duringmetastasis. Thus, characterizing the hybrid state is critical fordeveloping therapeutic targets to reverse EMT in a selectivegroup of patients.

We use the notation of "0," "1," and "1/2" to denote theexpression levels of miR200 and ZEB in the three differentphenotypes. It may be noted that "0," "1," or "1/2" expressionlevels of a given element in one cell line or context may bedifferent from those in another cell line or context, due tophenotypic heterogeneity and nonheritable variability pertain-ing to different cell lines. Thus, the absolute expression levels ofmany elements in the EMT regulatory network need to bemeasured quantitatively. Ourmodeling approach, which incor-porates a detailed analysis based on biologically realistic

parameters (see Supplementary Data in ref. 51 for details onrange of parameters), might be well suited to capture thisphenotypic variability, as opposed to a simplistic Booleanframework.

We also discuss the possibility of coexistence of differentphenotypes among cells, through phase plane analysis (Fig.5). From a biological perspective, each phase corresponds to(the coexistence of) different phenotypes. In three out ofseven phases, only one of the three phenotypes (epithelial,mesenchymal, or hybrid) can exist (denoted as {E}, {M}, and{E/M}). In three other phases, two of the phenotypes cancoexist (denoted as {E, M}, {E, E/M}, and {E/M, M}) and inone phase, we see the coexistence of all three phenotypes(denoted as {E, E/M, M}).

The abundant information from these identified phasescan also help to elucidate the experimental data on thediversity of behaviors associated with SNAIL in triggeringEMT in different contexts. SNAIL can induce complete EMT(64) or partial or transient EMT (28, 65) and is also requiredfor re-epithelialization of keratinocytes after wound healing(66). Depending on the rate of induction of SNAIL by differentsignals (e.g., by TFGb or hypoxia, both of which also regulateZEB), the cells would undergo a different trajectory in thephase diagram and therefore exhibit different phenotypictransitions. Also, the largest area in the phase diagram isthe monostable phase with only epithelial phenotype, whichagrees with the observation that epithelial type is the defaultphenotype (57).

Future Directions: Coupling the Decision Unitwith Other Cellular Processes

In Fig. 6, we show examples of how the EMT decisionunit is coupled to other key cellular processes includingprogrammability, cellular motility, genome plasticity, andmetabolism.

ProgrammabilitymiR200 inhibits the oncogenic driver LIN28 (67), which

forms a double-negative feedback loop with let-7 (68), whichis another family of miRNAs that inhibit EMT (69). TheLIN28/let7 loop has some indirect auto-regulations (70–72)that target the pluripotency factor OCT4 (73), which forms acomplex with SOX2 and activates itself indirectly, at least inhuman embryonic stem cells (hESC; ref. 3). Also, OCT4and SOX2 activate miR200 during MET while reprogram-ming of fibroblasts to iPSCs (74). In addition, let-7 alsomediates the double-negative feedback loop between metas-tasis repressor RKIP and metastasis-promoting self-inhibit-ing gene BACH1. RKIP activates let-7 indirectly, whichinhibits BACH1, and BACH1 inhibits RKIP directly. A similarmutual repression also exists between RKIP and SNAI, againvia let-7 (75, 76).

Cellular motilitymiR200 and miR34 are coupled to the regulatory network of

Rac1 and RhoA (77, 78), the two auto-regulatory GTPases thatdetermine the mode of cancer cell invasion (79). High levels ofactive Rac1 lead to the formation of cell protrusions and a

Lu et al.




mesenchymal phenotype, which degrade the ECM and are"path generators," while high levels of active RhoA lead toactomyosin contractility and amoeboid mode of invasion,which squeeze through the gaps in ECM and are referred toas "path finders" (79, 80). miR200 family members differentiallyregulate these two major modes of cell invasion by reducingboth the formation of cell protrusions and actomyosin con-tractility (78).

Genome plasticitymiR200 forms a chimera toggle switch with SIRT1 (81), a

key player in linking cell metabolism to stress response.SIRT1 is inhibited by itself and miR34 as well (82), andregulates hypoxia response through HIF1 and HIF2 (83, 84).Besides, SIRT1 maintains proper chromatin structure, thusbeing crucial for genome integrity and DNA damageresponse (85).



miR34

RhoA

A B

C D

Rac1

HIF2

HIF2

ROS

HIF1

HIF1

SIRT1

BACH1

Let-7

OCT4–SOX2OCT4

LIN28RKIP

miR200

m200

m34 S

Z6

3

2

2

2

2

1

1

1

ZEB

SNAIL

miR34

miR200 m200

m34 S

Z6

3

2

2

2

2

1

1

1

ZEB

SNAIL

miR34

miR200m200

m34 S

Z6

3

2

2

2

2

1

1

1

ZEB

SNAIL

miR34

miR200m200

m34 S

Z6

3

2

2

2

2

1

1

1

ZEB

SNAIL

Figure 6. Coupling of the EMT regulatory network to other key cellular properties. A, miR34 and miR200 both regulate expression levels of the GTPasesRhoA and Rac1 for cell motility. B, miR200/SIRT1 forms a mutually inhibitory loop, and SIRT1 regulates HIF1 and HIF2, and is inhibited by itself andmiR34 indirectly. C, miR200 inhibits LIN28, which forms a self-activating toggle switch with let-7. Also, OCT4 autoregulates itself through formingOCT4–SOX2 complex and also activates miR200 directly. Besides, let-7 mediates a double-negative feedback loop between RKIP and BACH1, andbetween RKIP and SNAIL. D, HIF1 activates SNAIL directly, whereas HIF2 induces ZEB indirectly. Also, HIF1 and HIF2 form a toggle switch throughROS, which activates both sides of the miR200/ZEB loop. A solid arrow represents transcriptional activation and a solid bar shows transcriptionalinhibition. Dashed lines indicate miRNA-mediated translational regulation and dotted lines denote indirect regulation. The numbers listed alongregulatory lines represent the number of corresponding binding sites as deduced from experiments. The boxes with dotted boundaries showseparations of different modules.





MetabolismThe miR200/ZEB loop links hypoxia to EMT as HIF1 and

HIF2 induce SNAIL and ZEB, respectively (28, 86). The mostintriguing coupling between hypoxia, metabolism, and EMT isthrough reactive oxygen species (ROS),whichnot onlymediatesthe mutual inhibition between HIF1 and HIF2 (87), but alsoactivates bothmiR200 andZEB (88, 89). BecauseHIF1 andHIF2are believed to govern the response to acute and chronichypoxia, respectively (90, 91), this coupling can unravel howepithelial–mesenchymal plasticity is linked to changes in tumormetabolism and tumor angiogenesis, as induced by hypoxia.

CSCsIn addition, the role of hypoxic niches in maintaining CSCs

in glioblastoma through the Notch signaling pathway (92, 93)underline the interplay between hypoxia, cell–cell communi-cation, and stemness. miR200 inhibits EMT and metastasis byaltering the tumormicroenvironment (94) and blocking tumorangiogenesis (95). Furthermore, ZEB1 forms feedback loopswith transcription factors GRHL2 (96), OVOL1, and OVOL2(97) to regulate EMT/MET during metastasis. We hypothesizethat the hybrid phenotype is associated with transition intoCSC. The transition can start during migration toward theblood vessels, during circulation, and also after seeding in thenew niche.

Noise bufferingmiR203 forms a chimera toggle switch with SNAIL and has

similar interconnections with the miR200/ZEB loop as that forthe miR34/SNAIL loop (98), thus indicating that both miR203and miR34 form noise-buffering integrators with SNAIL.

ConclusionsEpithelial–mesenchymal plasticity is crucial during embry-

onic development and cancer metastasis (6). The core regu-latory network for these transitions is the miR200/ZEB chi-mera toggle switch coupled with the miR34/SNAIL loop. Also,the miR200/ZEB decision circuit is coupled to many keycellular properties such as stemness, cell motility, cell–cellcommunication, metabolism, and resistance to apoptosis (31,32, 99). Despite its widespread influence, this decision circuithas had limited theoretical attention.

Here, we studied the involvement of miRNAs in EMT deci-sion-making circuit, using the recent theoretical frameworkdevised by Lu and colleagues (36). Our approach incorporatesin detail both modes of translational silencing by miRNAs andincludes the effects of number of binding sites of miRNA onmRNA, which have not been considered in previous studies(42–50). Using this framework, we unraveled the modulardesign principles of the core regulatory network for EMT/MET. We found that the miR34/SNAIL module is monostableand acts as a noise-buffering integrator, whereas miR200/ZEB

module is tristable and functions as a three-way switch. Thethree stable states of miR200/ZEB correspond to the epithelial[high miR200/low ZEB, denoted as (1, 0)], mesenchymal [lowmiR200/high ZEB, denoted as (0, 1)], and hybrid E/M [mediummiR200/medium ZEB, denoted as (1/2,

1/2)] phenotypes.We elaborated on the above to clarify that our hypothesis

for the (1/2,1/2) hybrid state (medium miR200/medium ZEB) is

different from that proposed by another recent study onEMT circuitry modeling, which neither included the self-reg-ulations of SNAIL and ZEB nor distinguished between thetranscription and translation inhibition processes (they wereboth modeled by inhibitory Hill functions of rank 2; ref. 45).Under these assumptions, the two modules have similar bis-table dynamics and act as binary switches, and the authorsproposed high miR200/low miR34 to be the hybrid state. Theexperimental data mentioned above is more consistent withour medium miR200, medium ZEB hypothesis. However, itmight be possible that different cell lines exhibit differenthybrid phenotypes, as multiple hybrid states have been pro-posed to be en route during EMT in certain contexts.

In Fig. 6, we show examples of how the EMT decision unit iscoupled to other key cellular processes. Future theoreticalinvestigations into these circuits hold the key to valuable newinsights about how these cellular characteristics are modifiedduring backward and forward transitions between epithelialand mesenchymal cells.

A better understanding of the transitions involving thehybrid phenotype is essential for a better comprehension ofcancer progression. Also, it can help answer the long-standingfundamental question—the difference between EMT that hap-pens during embryonic development and wound healing andEMT that happens during metastasis and organ fibrosis. Fromthe perspective of developing better therapeutics, we requirenormal wound healing to continue during antimetastatictherapies and also at the same time, restrict wound healingaugmentation from promoting metastasis in case a malignan-cy is present (100).

Disclosure of Potential Conflicts of InterestNo potential conflicts of interest were disclosed.

AcknowledgmentsThe authors have benefited from useful discussions with Herbert Levine, Ilan

Tsarfaty, and M. Cindy Farach-Carson.

Grant SupportThis work was supported by the NSF Center for Theoretical Biological Physics

(NSF grant no. PHY-1308264) and by the Cancer Prevention and ResearchInstitute of Texas (CPRIT). E. Ben-Jacob was also supported by a grant fromthe Tauber Family Funds and the Maguy-Glass Chair in Physics of ComplexSystems.

Received November 23, 2013; revised January 17, 2014; accepted March 31,2014; published online September 2, 2014.

References1. Balazsi G, van Oudenaarden A, Collins JJ. Cellular decision making

and biological noise: from microbes to mammals. Cell 2011;144:910–25.

2. Zhou JX, Huang S. Understanding gene circuits at cell-fate branchpoints for rational cell programming. Trends Genet 2011;27:55–62.

Lu et al.




3. Boyer LA, Lee TI, ColeMF, Johnstone SE, Levine SS, Zucker JP, et al.Core transcriptional regulatory circuitry in human embryonic stemcells. Cell 2005;122:947–56.

4. Holland JD, Klaus A, Garratt AN, Birchmeier W. Wnt signaling instem and cancer stem cells. Curr Opin Cell Biol 2013;25:254–64.

5. De Craene B, Berx G. Regulatory networks defining EMT duringcancer initiation and progression. Nat Rev Cancer 2013;13:97–110.

6. Thiery JP, Acloque H, Huang RYJ, Nieto MA. Epithelial–mesenchy-mal transitions in development and disease. Cell 2009;139:871–90.

7. Gupta GP, Massague J. Cancer Metastasis: building a framework.Cell 2006;127:679–95.

8. Friedl P, Gilmour D. Collective cell migration in morphogenesis,regeneration and cancer. Nat Rev Mol Cell Biol 2009;10:445–57.

9. Nakaya Y, Sheng G. EMT in developmental morphogenesis. CancerLett 2013;341:9–15.

10. Micalizzi DS, Farabaugh SM, Ford HL. Epithelial–mesenchymal tran-sition in cancer: parallels between normal development and tumorprogression. J Mammary Gland Biol Neoplasia 2010;15:117–34.

11. Savagner P. The epithelial–mesenchymal transition (EMT) phenom-enon. Ann Oncol 2010;21(Suppl 7):vii89–92.

12. Revenu C, Gilmour D. EMT 2.0: shaping epithelia through collectivemigration. Curr Opin Genet Dev 2009;19:338–42.

13. Faculty TA, FuttermanM,Fulfillment IP. Evidence for partial epithelial-to-mesenchyamal transition (pEMT) and recruitment of motile blas-toderm edge cells during avian epiboly. Dev Dyn 2011;240:1502–11.

14. Johnen N, Francart M-E, Thelen N, Cloes M, Thiry M. Evidence for apartial epithelial–mesenchymal transition in postnatal stages of ratauditory organ morphogenesis. Histochem Cell Biol 2012;138:477–88.

15. Christiansen JJ, Rajasekaran AK. Reassessing epithelial to mesen-chymal transition as a prerequisite for carcinoma invasion andmetastasis. Cancer Res 2006;66:8319–26.

16. Kalluri R, Weinberg RA. The basics of epithelial–mesenchymal tran-sition. J Clin Invest 2009;119:1420–8.

17. Yu M, Bardia A, Wittner BS, Stott SL, Smas ME, Ting DT, et al.Circulating breast tumor cells exhibit dynamic changes in epithelialand mesenchymal composition. Science 2013;339:580–4.

18. Samavarchi-Tehrani P, Folipour A, David L, Sung HK, Beyer TA, et al.Functional genomics reveals a BMP-drivenmesenchymal-to-epithe-lial transition in the initiation of somatic cell reprogramming. Cell StemCell 2010;7:64–77.

19. Li R, Liang J, Ni S, Zhou T, Qing X, Li H, et al. A mesenchymal-to-epithelial transition initiates and is required for the nuclear reprogram-ming of mouse fibroblats. Cell Stem Cell 2010;7:51–63.

20. Theys J, Jutten B, Habets R, Paesmans K, Groot AJ, Lambin P, et al.E-cadherin loss associated with EMT promotes radioresistance inhuman tumor cells. Radiother Oncol 2011;99:392–7.

21. Kurrey NK, Jalgaonkar SP, Joglekar AV, Ghanate AD, Chaskar PD,et al. Snail and Slug mediate radioresistance and chemoresistanceby antagonizing p53-mediated apoptosis and acquiring a stem-like phenotype in ovarian cancer cells. Stem Cells 2009;27:2059–68.

22. Mani SA, GuoW, Liao M-J, Eaton EN, Ayyanan A, Zhou AY, et al. Theepithelial–mesenchymal transition generates cells with properties ofstem cells. Cell 2008;133:704–15.

23. Morel A-P, Li�evre M, Thomas C, Hinkal G, Ansieau S, Puisieux A.Generation of breast cancer stem cells through epithelial–mesen-chymal transition. PLoS ONE 2008;3:e2888.

24. Bao B, Wang Z, Ali S, Kong D, Li Y, Ahmad A, et al. Notch-1induces epithelial–mesenchymal transition consistent with cancerstem cell phenotype in pancreatic cancer cells. Cancer Lett 2011;307:26–36.

25. Thiery JP, Sleeman JP. Complex networks orchestrate epithelial–mesenchymal transitions. Nat Rev Mol Cell Biol 2006;7:131–42.

26. PeinadoH,OlmedaD,CanoA. Snail, Zeb andbHLH factors in tumourprogression: an alliance against the epithelial phenotype? Nat RevCancer 2007;7:415–28.

27. Gregory PA, BrackenCP, Smith E, Bert AG,Wright JA, Roslan S, et al.An autocrine TGF-beta/ZEB/miR-200 signaling network regulates

establishment and maintenance of epithelial–mesenchymal transi-tion. Mol Biol Cell 2011;22:1686–98.

28. Lundgren K, Nordenskj€old B, LandbergG. Hypoxia, Snail and incom-plete epithelial–mesenchymal transition in breast cancer. Br JCancer2009;101:1769–81.

29. Chang C-J, Chao C-H, Xia W, Yang J-Y, Xiong Y, Li C-W, et al. p53regulates epithelial–mesenchymal transition and stem cell propertiesthrough modulating miRNAs. Nat Cell Biol 2011;13:317–23.

30. Araki S, Eitel JA, Batuello CN, Bijangi-Vishehsaraei K, Xie X-J,Danielpour D, et al. TGF-beta1–induced expression of human Mdm2correlates with late-stage metastatic breast cancer. J Clin Invest2010;120:290–302.

31. Brabletz S, Brabletz T. The ZEB/miR-200 feedback loop—a motor ofcellular plasticity in development and cancer? EMBO Rep 2010;11:670–7.

32. Brabletz S, Bajdak K, Meidhof S, Burk U, Niedermann G, Firat E, et al.The ZEB1/miR-200 feedback loop controlsNotch signalling in cancercells. EMBO J 2011;30:770–82.

33. Bracken CP, Gregory PA, Kolesnikoff N, Bert AG, Wang J, ShannonMF, et al. A double-negative feedback loop between ZEB1-SIP1 andthe microRNA-200 family regulates epithelial–mesenchymal transi-tion. Cancer Res 2008;68:7846–54.

34. BurkU, Schubert J,Wellner U, Schmalhofer O, Vincan E, Spaderna S,et al. A reciprocal repression betweenZEB1andmembers of themiR-200 family promotes EMT and invasion in cancer cells. EMBO Rep2008;9:582–9.

35. Siemens H, Jackstadt R, H€unten S, Kaller M, Menssen A, G€otz U,et al. miR-34 and SNAIL form a double-negative feedback loop toregulate epithelial–mesenchymal transitions. Cell Cycle 2011;10:4256–71.

36. Lu M, Jolly MK, Gomoto R, Huang B, Onuchic J, Ben-JacobE. Tristability in cancer-associated microRNA–TF chimera toggleswitch. J Phys Chem B 2013;117:13164–74.

37. HurteauGJ, Carlson JA, Roos E, BrockGJ. Stable expression ofmiR-200c alone is sufficient to regulate TCF8 (ZEB1) and restoreE-cadherin expression. Cell Cycle 2009;8:2064–9.

38. Guaita S, Puig I, Franci C, Garrido M, Dominguez D, Batlle E, et al.Snail induction of epithelial to mesenchymal transition in tumor cellsis accompanied by MUC1 repression and ZEB1 expression. J BiolChem 2002;277:39209–16.

39. Peir�oS, Escriv�aM,Puig I, Barber�aMJ,DaveN,HerranzN, et al. Snail1transcriptional repressor binds to its own promoter and controls itsexpression. Nucleic Acids Res 2006;34:2077–84.

40. Wels C, Joshi S, Koefinger P, Bergler H, Schaider H. Transcriptionalactivation of ZEB1 by slug leads to cooperative regulation of theepithelial–mesenchymal transition-like phenotype in melanoma. JInvest Dermatol 2011;131:1877–85.

41. Braun J, Hoang-Vu C, Dralle H, H€uttelmaier S. Downregulation ofmicroRNAs directs the EMT and invasive potential of anaplasticthyroid carcinomas. Oncogene 2010;29:4237–44.

42. Aguda BD, Kim Y, Piper-Hunter MG, Friedman A, Marsh CB. Micro-RNA regulation of a cancer network: consequences of the feedbackloops involving miR-17-92, E2F, and Myc. Proc Natl Acad Sci U S A2008;105:19678–83.

43. Li Y, Chen Y, Zhang H, Li Y. MicroRNA-mediated positive feedbackloopandoptimizedbistable switch in acancer network involvingmiR-17-92. PLoS ONE 2011;6:e26302.

44. Zhou P, Liu Z, Cai S, Wang R. Mechanisms generating bistability andoscillations in microRNA-mediated motifs. Phys Rev E Stat NonlinSoft Matter Phys 2012;85:041916.

45. Tian X-J, ZhangH, Xing J.Coupled reversible and irreversible bistableswitches underlying TGFb-induced epithelial to mesenchymal tran-sition. Biophys J 2013;105:1079–89.

46. Verma M, Karimiani EG, Byers RJ, Rehman S, Westerhoff HV, DayPJR. Mathematical modelling of miRNA mediated BCR.ABL proteinregulation in chronic myeloid leukaemia vis-a-vis therapeutic strat-egies. Integr Biol 2013;5:543–54.

47. Kang H-W, Crawford M, Fabbri M, Nuovo G, Garofalo M, Nana-SinkamSP, et al. Amathematical model for microRNA in lung cancer.PLoS ONE 2013;8:e53663.





48. Lai X, Wolkenhauer O, Vera J. Modeling miRNA regulation in cancersignaling systems: miR-34a regulation of the p53/Sirt1 signalingmodule. Methods Mol Biol 2012;880:87–108.

49. KhaninR, Vinciotti V. Computationalmodeling of post-transcriptionalgene regulation by MicroRNAs. J Comput Biol 2008;15:305–16.

50. Wei Z, Jiang S, Zhang Y, Wang X, Peng X, Meng C. The effect ofmicroRNAs in the regulation of human CYP3A4: a systematic studyusing a mathematical model. Sci Rep 2014;4:4283.

51. Lu M, Jolly MK, Levine H, Onuchic JN, Ben-Jacob E. MicroRNA-based regulation of epithelial–hybrid–mesenchymal fate determina-tion. Proc Natl Acad Sci U S A 2013;110:18144–9.

52. Andrecut M, Halley JD, Winkler DA, Huang S. A general model forbinary cell fate decision gene circuitswith degeneracy: indeterminacyand switch behavior in the absence of cooperativity. PLoS ONE2011;6:e19358.

53. Guantes R, Poyatos JF. Multistable decision switches for flexiblecontrol of epigenetic differentiation. PLoS Comput Biol 2008;4:e1000235.

54. Macía J, Widder S, Sol�e R. Why are cellular switches Boolean?General conditions for multistable genetic circuits. J Theor Biol 2009;261:126–35.

55. Wang J, Zhang K, Xu L, Wang E. Quantifying the Waddingtonlandscape and biological paths for development and differentiation.Proc Natl Acad Sci U S A 2011;108:8257–62.

56. García de Herreros A, Baulida J. Cooperation, amplification, andfeed-back in epithelial–mesenchymal transition. Biochim BiophysActa 2012;1825:223–8.

57. Frisch SM. The epithelial cell default-phenotype hypothesis and itsimplications for cancer. Bioessays 1997;19:705–9.

58. Das S, Becker BN, Hoffmann FM, Mertz JE. Complete reversalof epithelial to mesenchymal transition requires inhibition of bothZEB expression and the Rho pathway. BMC Cell Biol 2009;10:94.

59. Aigner K, Dampier B, Descovich L, Mikula M, Sultan A, Schreiber M,et al. The transcription factor ZEB1 (deltaEF1) promotes tumour celldedifferentiation by repressingmaster regulators of epithelial polarity.Oncogene 2007;26:6979–88.

60. Taube JH, Herschkowitz JI, Komurov K, Zhou AY, Gupta S, Yang J,et al. Core epithelial-to-mesenchymal transition interactome gene-expression signature is associated with claudin-low and metaplasticbreast cancer subtypes. Proc Natl Acad Sci U S A 2010;107:15449–54.

61. Vannier C, Mock K, Brabletz T, Driever W. Zeb1 regulates E-cadherinand Epcam (epithelial cell adhesion molecule) expression to controlcell behavior in early zebrafish development. J Biol Chem2013;288:18643–59.

62. Chigurupati S, Arumugam TV, Son TG, Lathia JD, Jameel S, MughalMR, et al. Involvementof notch signaling inwoundhealing. PLoSONE2007;2:e1167.

63. Huang RY-J, Wong MK, Tan TZ, Kuay KT, Ng AHC, Chung VY, et al.An EMT spectrum defines an anoikis-resistant and spheroidogenicintermediate mesenchymal state that is sensitive to e-cadherin res-toration by a src-kinase inhibitor, saracatinib (AZD0530). Cell DeathDis 2013;4:e915.

64. Batlle E, SanchoE, Francí C, DomínguezD,MonfarM, Baulida J, et al.The transcription factor snail is a repressor of E-cadherin geneexpression in epithelial tumour cells. Nat Cell Biol 2000;2:84–9.

65. Blanco MJ, Barrallo-Gimeno A, Acloque H, Reyes AE, Tada M,Allende ML, et al. Snail1a and Snail1b cooperate in the anteriormigration of the axial mesendoderm in the zebrafish embryo. Devel-opment 2007;134:4073–81.

66. Savagner P, Kusewitt DF, Carver EA, Magnino F, Choi C, Gridley T,et al. Developmental transcription factor slug is required for effectivere-epithelialization by adult keratinocytes. J Cell Physiol 2005;202:858–66.

67. Kong D, Banerjee S, Ahmad A, Li Y, Wang Z, Sethi S, et al.Epithelial to mesenchymal transition is mechanistically linked withstem cell signatures in prostate cancer cells. PLoS ONE 2010;5:e12445.

68. Yang X, Lin X, Zhong X, Kaur S, Li N, Liang S, et al. Double-negativefeedback loop between reprogramming factor LIN28 and microRNA

let-7 regulates aldehyde dehydrogenase 1-positive cancer stemcells. Cancer Res 2010;70:9463–72.

69. Li Y, VandenBoom TG, Kong D, Wang Z, Ali S, Philip PA, et al. Up-regulation ofmiR-200 and let-7 by natural agents leads to the reversalof epithelial-to-mesenchymal transition in gemcitabine-resistantpancreatic cancer cells. Cancer Res 2009;69:6704–12.

70. Tokumaru S, Suzuki M, Yamada H, Nagino M, Takahashi T. let-7regulates Dicer expression and constitutes a negative feedback loop.Carcinogenesis 2008;29:2073–7.

71. Zisoulis DG, Chang RK, Kai ZS, Pasquinelli AE. Autoregulation ofmicroRNA biogenesis by let-7 and Argonaute. Nature 2012;486:541–4.

72. Massirer KB, Yeo GW, Morris Q, Hutt KR, Yan BY, Kazan H, et al.LIN28 binds messenger RNAs at GGAGA motifs and regulatessplicing factor abundance. Mol Cell 2012;48:195–206.

73. Qiu C, Ma Y, Wang J, Peng S, Huang Y. Lin28-mediated post-transcriptional regulation of Oct4 expression in human embryonicstem cells. Nucleic Acids Res 2010;38:1240–8.

74. WangG,GuoX,HongW, LiuQ,Wei T, LuC, et al. Critical regulation ofmiR-200/ZEB2 pathway in Oct4/Sox2–induced mesenchymal-to-epithelial transition and induced pluripotent stem cell generation.Proc Natl Acad Sci U S A 2013;110:2858–63.

75. Dangi-Garimella S, Yun J, Eves EM, Newman M, Erekland SJ,Hammond SM, et al. Raf kinase inhibitory protein suppresses ametastasis signaling cascade involving LIN28 and let-7. EMBO J2009;28:347–58.

76. Lee J, Lee J, Farquhar KS, Yun J, Frankenberger CA, Bevilacqua E,et al. Network of mutually repressive metastasis regulators canpromote cell heterogeneity and metastatic transitions. Proc NatlAcad Sci U S A 2014;111:E364–73.

77. Kim D, Kim S, Jin E-J, Park HM, Sonn J, Song J, et al. MicroRNA-34amodulates cytoskeletal dynamics through regulating RhoA/Rac1cross-talk in chondroblasts. J Biol Chem 2012;287:12501–9.

78. Elson-Schwab I, Lorentzen A, Marshall CJ. MicroRNA-200 familymembers differentially regulate morphological plasticity andmode ofmelanoma cell invasion. PLoS ONE 2010;5:e13176.

79. Parri M, Chiarugi P. Rac and Rho GTPases in cancer cell motilitycontrol. Cell Commun Signal 2010;8:23.

80. Ben-Jacob E, Coffey DS, Levine H. Bacterial survival strategiessuggest rethinking cancer cooperativity. Trends Microbiol 2012;20:403–10.

81. Eades G, Yao Y, Yang M, Zhang Y, Chumsri S, Zhou Q. miR-200aregulates SIRT1 expression and epithelial to mesenchymal transition(EMT)-like transformation in mammary epithelial cells. J Biol Chem2011;286:25992–6002.

82. Brooks CL, Gu W. How does SIRT1 affect metabolism, senescenceand cancer? Nat Rev Cancer 2009;9:123–8.

83. Lim J-H, Lee Y-M, Chun Y-S, Chen J, Kim J-E, Park J-W. Sirtuin 1modulates cellular responses to hypoxia by deacetylating hypoxia-inducible factor 1alpha. Mol Cell 2010;38:864–78.

84. Dioum EM, Chen R, Alexander MS, Zhang Q, Hogg RT, Gerard RD,et al. Regulation of hypoxia-inducible factor 2alpha signaling bythe stress-responsive deacetylase sirtuin 1. Science 2009;324:1289–93.

85. Wang RH, Sengupta K, Li C, Kim HS, Cao L, Xiao C, et al. ImpairedDNA damage response, genome instability, and tumorigenesis inSIRT1 mutant mice. Cancer Cell 2008;14:312–23.

86. Kim WY, Perera S, Zhou B, Carretero J, Yeh JJ, Heathcote SA, et al.HIF2alpha cooperates with RAS to promote lung tumorigenesis inmice. J Clin Invest 2009;119:2160–70.

87. Yuan G, Peng YJ, Reddy VD, Makarenko VV, Nanduri J, Khan SA,et al. Mutual antagonism between hypoxia-inducible factors 1a and2a regulates oxygen sensing and cardio-respiratory homeostasis.Proc Natl Acad Sci U S A 2013;110:E1788–96.

88. Magenta A, Cencioni C, Fasanaro P, Zaccagnini G, Greco S, Sarra-Ferraris G, et al. miR-200c is upregulated by oxidative stress andinduces endothelial cell apoptosis and senescence via ZEB1 inhibi-tion. Cell Death Differ 2011;18:1628–39.

89. Fukawa T, Kajiya H, Ozeki S, Ikebe T, Okabe K. Reactive oxygenspecies stimulates epithelial mesenchymal transition in normal


Lu et al.



human epidermal keratinocytes via TGF-beta secretion. Exp Cell Res2012;318:1926–32.

90. Lin Q, Cong X, Yun Z. Differential hypoxic regulation of hypoxia-inducible factors 1alpha and 2alpha. Mol Cancer Res 2011;9:757–65.

91. Koh MY, Lemos R, Liu X, Powis G. The hypoxia-associated factorswitches cells from HIF-1a- to HIF-2a–dependent signaling promot-ing stem cell characteristics, aggressive tumor growth and invasion.Cancer Res 2011;71:4015–27.

92. Seidel S,GarvalovBK,WirtaV, vonStechowL,Sch€anzer A,Meletis K,et al. A hypoxic niche regulates glioblastoma stem cells throughhypoxia inducible factor 2 alpha. Brain 2010;133:983–95.

93. Bar EE, Lin A, Mahairaki V, Matsui W, Eberhart CG. Hypoxiaincreases the expression of stem-cell markers and promotes clo-nogenicity in glioblastoma neurospheres. Am J Pathol 2010;177:1491–502.

94. Schliekelman MJ, Gibbons DL, Faca VM, Creighton CJ, Rizvi ZH,ZhangQ, et al. Targets of the tumor suppressormiR-200 in regulationof the epithelial–mesenchymal transition in cancer. Cancer Res2011;71:7670–82.

95. Pecot CV, Rupaimoole R, Yang D, Akbani R, Ivan C, Lu C, et al.Tumour angiogenesis regulation by themiR-200 family. NatCommun2013;4:2427.

96. Cieply B, Farris J, Denvir J, Ford HL, Frisch SM. Epithelial–mesen-chymal transition and tumor suppression are controlled by a recip-rocal feedback loop between ZEB1 and Grainyhead-like-2. CancerRes 2013;73:6299–309.

97. RocaH, Hernandez J,Weidner S,McEachin RC, Fuller D, SudS, et al.Transcription factors OVOL1 andOVOL2 induce themesenchymal toepithelial transition in human cancer. PLoS ONE 2013;8:e76773.

98. MoesM, Le B�echec A, Crespo I, Laurini C, Halavatyi A, Vetter G, et al.A novel network integrating a miRNA-203/SNAI1 feedback loopwhich regulates epithelial to mesenchymal transition. PLoS ONE2012;7:e35440.

99. Vallejo DM, Caparros E, Dominguez M. Targeting Notch signalling bythe conserved miR-8/200 microRNA family in development andcancer cells. EMBO J 2011;30:756–69.

100. Leopold PL, Vincent J, Wang H. A comparison of epithelial-to-mesenchymal transition and re-epithelialization. Semin CancerBiol 2012;22:471–83.





Toward Decoding the Principles of Cancer Metastasis Circuits

Documents