A role for ATP-dependent chromatin remodeling in the hierarchical cooperativity between noninteracting transcription factors
AA role for ATP-dependent chromatin remodeling in the hierarchicalcooperativity between noninteracting transcription factors
Denis Michel
Universite de Rennes1. Campus de Beaulieu Bat.13. 35042 Rennes France. E.mail: [email protected]
Abstract
Chromatin remodeling machineries are abundant anddiverse in eukaryotic cells. They have been involvedin a variety of situations such as histone exchange andDNA repair, but their importance in gene expressionremains unclear. Although the influence of nucleosomeposition on the regulation of gene expression is gener-ally envisioned under the quasi-equilibrium perspective,it is proposed that given the ATP-dependence of chro-matin remodeling enzymes, certain mechanisms neces-sitate non-equilibrium treatments. Examination of thecelebrated chromatin remodeling system of the mousemammary tumor virus, in which the binding of tran-scription factors opens the way to other ones, revealsthat breaking equilibrium offers a subtle mode of tran-scription factor cooperativity, avoids molecular trappingphenomena and allows to reconcile previously conflictingexperimental data. This mechanism provides a controllever of promoter responsiveness to transcription factorcombinations, challenging the classical view of the uni-lateral influence of pioneer on secondary transcriptionfactors.
Keywords
Chromatin remodeling; cooperativity;transcription factor; MMTV; glucocorticoid receptor.
The importance of ATP-dependent machineries remod-eling chromatin by actively moving nucleosomes rela-tively to DNA, remains puzzling. Beside their possi-ble structural roles in chromatin organization, nucleo-some repositioning, histone exchange and DNA repair,a role in transcriptional cooperativity is proposed here.The specificity and intensity of gene expression is gov-erned by interactions between regulatory DNA sequences(cis-regulators) and various trans-acting factors (tran-scription factor proteins (TFs) and non-coding RNAs).The occupation of a gene promoter by these trans-regulators involves both micro-reversible and micro-irreversible steps. Micro-reversible binding processescan lead to sigmoidal concentration-dependent response through classical multimeric cooperativity (Bolouri andDavidson, 2002; Michel, 2010). The role of nucleosomeshas also been examined from the micro-reversible per-spective (Dodd et al., 2007; Segal and Widom, 2009;Mirny, 2010). The rapid equilibration of these thermally-driven phenomena, relatively to the slow changes of cel-lular components, simplifies the definition of the inputfunctions used in gene network modeling (Bintu et al.,2005; Michel, 2010). But promoter occupancy also in-volves some micro-irreversible transitions such as chro-matin remodeling and active dissociation processes. Pre-cisely, it is proposed in the present study that insert-ing micro-irreversible steps in the process of promotersaturation, offers additional possibilities of potent coop-erativity. A single example has been selected becauseit remarkably illustrates how micro-irreversible transi-tions can generate a refined discernment in gene ex-pression. In this example, the micro-irreversible stepcorresponds to the energy-dependent phenomenon ofchromatin-remodeling, in which the position of DNAaround nucleosomes is modified. By this way, upon bind-ing to DNA, a first TF directs the accessibility to DNAof other ones. This mechanical activity allows: (i) coop-erativity between non-physically interacting TFs and (ii)constitutively expressed TFs to participate to conditionalinduction. One of the most celebrated chromatin remod-eling system is provided by the thoroughly documentedMouse Mammary Tumor Virus (MMTV) promoter.
The assembly of macromolecular complexes generallyproceeds in a hierarchical manner in the cell. For ex-ample, a component C which cannot bind to the isolatedcomponents A and B , can bind to a pre-associated com-plex AB . Hierarchical binding chains such as A + B (cid:10) AB, + C (cid:10) ABC, + D (cid:10) ABCD etc, are often involvedin the building of multi-molecular complexes, but are lesscompatible with the dynamic and reactive behaviours ofsolubles components. Indeed, in equilibrium conditions,1 a r X i v : . [ q - b i o . M N ] S e p he chain written above leads to the trapping of the earlycomponents in the complexes as long as the late compo-nents are present. This phenomenon can exist for TFbinding to gene promoters. It can hold for example, forthe successive binding steps observed in equilibrium con-ditions between the TodT TFs and the series of TodT-binding sites juxtaposed in the Tod gene promoter, thathas been proposed to be mediated by DNA conformationchanges (Lacal et al., 2008). Beside this puzzling situa-tion of equilibrium allostery, the hierarchical binding oftranscription factors in equilibrium conditions is also pos-sible in the case of the large eukaryotic preinitiation com-plexes made of the so-called general transcription factorsGTFs (Michel, 2010). But hierarchical relationships havealso been reported for non-interacting isolated TFs in ab-sence of any trapping phenomenon. To allow reconcilinghierarchical binding and absence of trapping, one shouldpostulate the possibility to break equilibrium. This sit-uation is well illustrated by the case of the occupationof the MMTV promoter involving micro-irreversible pro-cesses, thoroughly documented but not yet clearly un-derstood. In section 4, this system will be analysed un-der the classical equilibrium assumption. Then, in sec-tion 5, a non-equilibrium scheme will be proposed basedon hypotheses built from MMTV experimental data, inwhich the equilibrium-breaking machines are the ATP-dependent SWI/SNF chromatin remodeling enzymes. A central piece of data about MMTV expression is therole of nucleosomes in the mutual influence betweenthe glucocorticoid receptor (GR) and a group of TFs(NF1/Oct). Although the activation of MMTV by GRand NF1/Oct-1 seemed clear in the initial reports, dis-crepancies appeared in following studies. The basisof glucocorticoid hormone-induced MMTV regulation isthat GR has an initiating role, triggered upon hormonebinding (stress hormone corticol or corticosterone) andsubsequently amplified by NF1 and Oct-1. This sequen-tial action is dependent on the position of nucleosomes onDNA, since it is not observed with naked DNA (Richard-Foy and Hager, 1987; Archer et al., 1992; Ch´avez andBeato, 1997). The repositioning of nucleosomes triggeredby GR, probed by nuclease or chemical mapping, leadsto the exposure of the NF1 and Oct-1 binding sites andis mediated by SWI-SNF ATPases (Fryer and Archer,1998). The different roles of GR and NF1 in initiatingand amplifying transcription respectively, are explainedby their differential mode of interaction with chromatin:GR can bind to DNA wrapped around nucleosomes, con- trary to NF1 which requires a fully accessible double he-lix (Eisfeld et al., 1997). This DNA-binding hierarchy,first of hormone-bound GR and then of NF1/Oct, of-fers a powerful opportunity of cooperativity. Indeed, GRis inducible but not very potent contrary to the coupleNF1/Oct. As NF1/Oct can access DNA only upon bind-ing of GR, the promoter activity can become stronglysigmoidal, particularly if NF1/Oct is transcriptionallymore potent than GR. Sigmoidal responses are gener-ally due to decreased responsiveness to low signals. Thisis the case for the MMTV promoter in which TFs areprevented to bind at low concentration. But this ele-gant mechanism has then been clouded in the followingreports, which introduced new actors and revised the hi-erarchy of binding of GR and NF1. In sharp contrastwith the earlier articles, NF1 and Oct-1 binding siteshave been shown to preset chromatin prior to GR binding(Belikov et al., 2004). The picture is thus more complexthan supposed previously and the mutual influence be-tween GR and NF1 for binding to the MMTV promoteris now described as dualistic (Hebbar and Archer, 2007),blurring the logic of this system. The same apparentparadox has been pointed for the relationships betweenpurported ”pioneer” and secondary transcription factors(Caizzi et al., 2014). In fact, the strict dependence onthe previous fixation of a factor to allow the fixation ofanother factor can be alleviated if introducing an ad-ditional step of chromatin remodeling (Fig.1). It willbe shown that the cooperative relationships between GRand NF1, which are unclear when examined only from atime-reversible perspective, can be usefully reconsideredfrom a non-equilibrium perspective (section 5), but theoutcomes of equilibrium modeling is first examined belowfor comparison.
In the simplest hierarchical modeling scheme assum-ing micro-reversibility (Fig.1a), MMTV transcription isstimulated by two groups of transcription factors GR(named A ) and NF1/Oct (named B ) (Fig.1). A and B bind to the MMTV promoter ( P ) through their DNA-binding domain (DBD) in a hierarchical manner, butonce bound to DNA, they are supposed to stimulatetranscription in an independent and additive manner,through their activation domain (TAD). In these con-ditions, the fractional activity ( F ) ranging from 0 to 1,of the MMTV promoter, is described in Eq.(1). F = p ( A ) k A + p ( B ) k B k A + k B (1)In this equation, k A and k B are the maximal frequen-cies at which A and B , when bound to DNA, recruit tran-2cription machineries, thereby initiating multiple roundsof transcription. These frequencies should be weightedby the probabilities of presence of A and B on the pro-moter, written p ( A ) and p ( B ) respectively (with smallletters p to not be confused with the promoter P ). Figure 1.
Different models to explain the hierarchical occu-pation of the MMTV promoter ( P ) by GR (named A ) andNF1 (named B ). The schemes ( a ) and ( b ), comply with theprinciple of microscopic reversibility but not the schemes ( c )and ( d ). In ( c ), k o is the rate of chromatin opening drivenby SWI/SNF ATPases and k c is the rate of chromatin clos-ing, driven by stabilization of DNA bending. B cannot bindto P because of inappropriate nucleosome positioning, while A can bind to both P and P (cid:48) with different constants. Inthe scheme ( d ), chromatin closing can occur only when thepromoter is free of any TF. The probabilities p ( A ) and p ( B ), equivalent to frac-tional occupation times, can be formulated through anAdair approach, as the ratio of occupied over total bind-ing sites, which can be expressed as concentrations in er-godic conditions, by enumerating the possible promoterstates. p ( A ) = [ P A ] + [
P AB ][ P ] + [ P A ] + [
P AB ] (2a)and, given that B is supposed to access DNA only when A is present, p ( B ) = [ P AB ][ P ] + [ P A ] + [
P AB ] (2b)Using P as a reference, Eq.(2) can be converted into p ( A ) = K A [ A ](1 + K B [ B ]) /D (3a) p ( B ) = K A [ A ] K B [ B ] /D (3b)with D = 1 + K A [ A ] + K A [ A ] K B [ B ] (3c)In this scheme, GR is prevented to dissociate from asaturated promoter. This trapping effect which can ap-pear puzzling, is inherent to the equilibrium modeling ofsequential cooperativity, but such a trapping of GR onthe MMTV promoter is not consistent with the observa-tion that GR can escape DNA whatever the chromatinconfiguration (Fletcher et al., 2000; Hager et al., 2000).To avoid this problem, one can imagine an alternativescenario (Fig.1b), in which GR dissociation does not re-quire the absence of NF1. Two different equilibrium con-stants are defined for GR ( K A and K (cid:48) A = 1/ K (cid:48) dA ), to takeinto account the different chromatin states. The occupa-tion probabilities of the A and B sites are respectively: p ( A ) = K A [ A ](1 + K B [ B ]) /D (4a) p ( B ) = K A [ A ] K B [ B ](1 + K (cid:48) dA / [ A ]) /D (4b)with D = 1 + K A [ A ](1 + K B [ B ](1 + K (cid:48) dA / [ A ])) (4c)But the phenomenon of trapping now concerns NF1,possibly for long periods in case of removal of the glu-cocorticoid hormone. Though puzzling, this possibilitywould be consistent with the observation that NF1 ispresent on DNA prior to hormone addition (Hebbar andArcher, 2003; Belikov et al., 2004). But it remains toexplain why the MMTV promoter would be inactive inspite of the continuous presence of NF1. A possible ex-planation could be that NF1 doesn’t work as long as itis prevented to recycle on DNA, according to the modelof one-shot TFs like ATF6 (Michel, 2010). This system3ould still conform microscopic reversibility, but the no-tion of equilibrium would become shaky since it is nolonger dynamic after disappearance of A . To proposea more plausible formulation of the MMTV promoteroccupancy, relieved from any trapping effect, a micro-irreversible steps should be introduced in the system. To not recourse to trapping phenomena which are notexperimentally verified for MMTV, one should postulateanother mode of cooperativity, liberated from the micro-reversibility constraints. Energy inputs obviously existin the system and are provided by ATPases (SWI/SNF),recruited by DNA-bound GR (Fryer and Archer, 1998).Among the different micro-irreversible mechanisms thatcan be imagined, the model shown in Fig.1c is an at-tempt to reconcile the more experimental data as pos-sible, in a novel scheme as simple as possible. Chro-matin remodeling can be triggered and reversed dynami-cally, according to the well established reversibility (in itstraditional acceptation) of hormone-induced nucleosomepositioning (Belikov et al., 2001) and to the dynamic in-teraction of remodeling complexes with the MMTV pro-moter (Johnson et al., 2008). The spontaneous nucleo-some repositioning from P (cid:48) to P is dictated by the intrin-sic bendability of DNA sequence and can be consideredas nearly micro-irreversible. Since transient behavioursfollowing GR activation can be neglected for the result-ing gene expression, a steady state treatment is sufficientfor modeling this system. A reasonable additional hy-pothesis is that the time scales are different between therapid equilibration of the TFs with the promoter, andthe slower dynamics of micro-irreversible chromatin re-modeling, for opening site for B ( k o ) and for closing it( k c ). The time scale separation hypothesis is not alwaysapplicable, but is justified in the present case, given therapid equilibration of the TFs with the MMTV DNA,suggested by the short turnovers of GR (12 milliseconds)evidenced by fluorescence recovery after photobleaching(FRAP)(Sprague et al., 2004). This condition allows touse the approach of (Cha, 1968), mixing in the sametreatment rate constants (time-dependent) and equilib-rium constants (time-independent). In this method, sev-eral groups of rapidly equilibrated species are definedusing equilibrium constants. They correspond in thepresent case to the two chromatin states of the MMTVpromoter, which will be written Σ P and Σ P (cid:48) (Fig. 1c,d).Two DNA-binding constants K a and K (cid:48) a are postulatedfor A depending on the chromatin state, but K (cid:48) a is notaffected by the presence or not of B , given that these TFs do not directly interact with each other.[Σ P ] = [ P ] + [ P A ] (5a)[Σ P (cid:48) ] = [ P (cid:48) ] + [ P (cid:48) A ] + [ P (cid:48) B ] + [ P (cid:48) AB ] (5b)with [ P ][Σ P ] = 1 D P (6a)[ P A ][Σ P ] = K A [ A ] D P (6b)and D P = 1 + K A [ A ] (6c)For the P (cid:48) states: [ P (cid:48) ][Σ P (cid:48) ] = 1 D P (cid:48) (7a)[ P (cid:48) A ][Σ P (cid:48) ] = K (cid:48) A [ A ] D P (cid:48) (7b)[ P (cid:48) B ][Σ P (cid:48) ] = K B [ B ] D P (cid:48) (7c)[ P (cid:48) AB ][Σ P (cid:48) ] = K (cid:48) A [ A ] K B [ B ] D P (cid:48) (7d)with D P (cid:48) = (1 + K (cid:48) A [ A ])(1 + K B [ B ]) (7e)When gathering the promoter states, the probabilities p ( A ) and p ( A ) that P is occupied by A and B respec-tively, can be defined as follows p ( A ) = p ( P ∩ A ) + p ( P (cid:48) ∩ A ) (8a)equivalent to p ( A ) = p ( A | P ) p ( P ) + p ( A | P (cid:48) ) p ( P (cid:48) ) (8b)and p ( B ) = p ( P (cid:48) ∩ B ) (given that p ( P ∩ B ) = 0) (8c) p ( B ) = p ( B ∩ P (cid:48) ) p ( P (cid:48) ) (8d)where p ( A | P ) = [ P A ][Σ P ] = K A [ A ]1 + K A [ A ] (8e) p ( A | P (cid:48) ) = [ P (cid:48) A ] + [ P (cid:48) AB ][Σ P (cid:48) ] = K (cid:48) A [ A ]1 + K (cid:48) A [ A ] (8f) p ( B | P (cid:48) ) = [ P (cid:48) B ] + [ P (cid:48) AB ][Σ P (cid:48) ] = K B [ B ]1 + K B [ B ] (8g)and p ( P ) = [ P ][ P ] + [ P (cid:48) ] and p ( P (cid:48) ) = 1 − p ( P ) (8h)4 P ] and [ P (cid:48) ] are the amounts of time in which the pro-moter is in the P and P (cid:48) states, which can be deducedfrom the steady state balance. If supposing, to agreewith experimental observations, that the restoration ofthe basal chromatin can occur only when the promoteris not occupied by B (Fig.1c), then, the steady state canbe written: k [ P ] p ( A | P ) = k c [ P (cid:48) ] (1 − p ( B | P (cid:48) )) (9)which yields, using the values defined previously,[ P ][ P (cid:48) ] = k c (1 + K A [ A ]) k o K A [ A ](1 + K B [ B ]) (10)leading to p ( A ) = K A [ A ] (cid:18) k c + k o K (cid:48) A [ A ] (cid:18) K B [ B ]1 + K (cid:48) A [ A ] (cid:19)(cid:19) /D (11a)and p ( B ) = k o K A [ A ] K B [ B ] /D (11b)where D = k c (1 + K A [ A ]) + k o K A [ A ](1 + K B [ B ]) (11c)These results are then incorporated in Eq.(1). Thecapacity of this system to generate sigmoidal curves isdue to the products of the concentrations of A with it-self (in Eq.(11a)) and with B (in Eqs(11a) and (11b))if assuming a double time scale separation: (i) betweenDNA/TF interactions and chromatin remodeling kinet-ics, as previously postulated and (ii) between chromatinremodeling and gene product concentration changes. The important parameters to evaluate are the sigmoidic-ity and sensitivity of the promoter activity to TF con-centration changes. Sigmoidicity is classically obtainedwhen the TFs should multimerize for binding to DNA.This is the case for GR which is active as a dimer. Butto focus on the specific source of cooperativity providedby the mechanism examined here, dimerisation will beignored and the TFs will be considered as preformeddimers. For easier analyses, the fractional promoter ac-tivity equations will be adimensioned by setting someconstants. The ratio of the transcriptional strength ofDNA-bound B and A is γ = k B /k A . In the mechanismsof Fig.1c and 1d, the ratio of equilibrium constants of GR binding to the two chromatin states is α = K A /K (cid:48) A ,and the ratio of chomatin opening and closing rates is β = k o /k c . γ is not an equilibrium constant and is mod-ifiable but the cellular contents in remodeling enzymesand ATP. An energy-independent conformational equi-librium between P A and P (cid:48) A would lead again to molec-ular trapping (of A and of the chromatin-remodeling en-zyme). The equivalence between k o and a Poissonian rateis a gross approximation since the transition P A → P (cid:48) A encloses many elementary events, including the recruit-ment of enzymes, of ATP, catalytic steps etc, which willnot be detailed here. Let be x and y the binding po-tentials of A and B respectively which are, for the non-equilibrium mechanisms x = K (cid:48) A [ A ], y = K B [ B ] and forthe equilibrium mechanisms x = K A [ A ] and y = K B [ B ].Fractional activity can be defined with these symbols andused for drawing 3D plots. They are listed below for thedifferent models. A and B tothe promoter Eq.(1) yields F = 11 + γ (cid:18) x x + γ y y (cid:19) (12)The corresponding curve is shown in Fig.2a, whensaturating A can trigger only one quarter of maximalactivation ( γ = 3). F = x (1 + y (1 + γ ))(1 + γ )(1 + x + xy ) (13)The corresponding plot is shown in Fig.2b for γ = 3. F = x (1 + y ) + γy ( α + x )(1 + γ )(1 + x + y ( α + x )) (14) In this system, chromatin relaxation can occur duringthe periods of absence of B , irrespective of whether A ispresent or not. F = αx (cid:104) βx (cid:16) y x (cid:17) + βγy (cid:105) (1 + γ )[1 + αx (1 + β (1 + y ))] (15)5 igure 2. Comparative shapes of promoter activity curves in linear coordinates (left panels) and in Hill coordinates (rightpanels). x and y are the binding potentials of the TFs A and B used in the main text, and X and Y are their logarithms. Thesmall inserts show 2D sections of the Hill plots at the indicated planes. In all cases, B is considered 3-time more potent than A for activating transcription. ( a ) The two TFs bind independently to the gene promoter (Eq.(12)). ( b ) Putative equilibriumhierarchical cooperativity model (Eq.(13)). ( c ) Hierarchical model with chromatin remodeling, in which the basal chromatinorganization state can be restored only from the TF-free promoter (Eq.(18)), with the combination of parameters ( α , β , γ )= (0.001, 2, 3). A and B are assumed to participate to the recruitment of transcription machineries in an additive manner. In this alternative possibility, chromatin closing to B canoccur only for a TF-free promoter (specifically not fromthe P (cid:48) A state). This possibility could for example be ex-plained by the persistent molecular association between A and chromatin-remodeling enzymes. In this case, the same development as in section 5, gives: k o [ P ] p ( A | P ) = k c [ P (cid:48) ](1 − p ( A | P (cid:48) ))(1 − p ( B | P (cid:48) )) (16)leading to the following steady state P/P (cid:48) ratio[ P ][ P (cid:48) ] = k c (1 + K A [ A ]) k o K A [ A ](1 + K (cid:48) A [ A ])(1 + K B [ B ]) (17)6nd to the fractional activity F = αx [1 + βx (1 + y ) + βγy (1 + x )](1 + γ )[1 + αx (1 + β (1 + x )(1 + y ))] (18)A representative plot of this condition is shown inFig.2c for ( α , β , γ ) = (0.05, 2, 3). This set of param-eters is chosen to agree with experimental observations.Indeed, B is considered as more potent than A becauseit is in fact not a single TF, but a combination of sev-eral potent TFs (NF1 and Oct). The higher affinity of A (GR) for the promoter ( K (cid:48) a > K a ), is suggested bythe fact that the chromatin configuration permissive toNF1/Oct binding, strongly favours GR binding (Belikovet al., 2004). Sigmoidal genetic responses are the typical ingredients ofmultistable dynamic gene regularory networks. In a sim-ple example, if a gene subject to this mode of transcrip-tional regulation stimulates its own expression through apositive feedback, then, the sigmoidal and saturable ex-pression curve crosses twice the non-saturable first-orderdegradation line, thus generating bistability (Cherry andAdler, 2000). This general role of nonlinear responses inthe formation of Boolean-like networks is not specific tohierarchical cooperativity and will not be detailed furtherhere.
The mechanism of transcriptional cooperativity mostwidely reported and modeled in the literature is thatobtained with TFs capable of physically interacting witheach other and binding to a series of non-consensual DNAelements in a promoter. By this way, the direct inter-actions between the TFs help them to bind together toDNA, whereas their individual affinity for their DNA ele-ments would not have allowed their independent binding.But a recent study suggested that this mode of cooper-ativity is in fact doubtful (Chu et al., 2009), since di-rect interactions between adjacent TFs would lead to theclustering of TFs on DNA which inevitably contains non-specific TF binding sites. Indeed, all the TFs have a min-imal non-specific affinity for DNA (at least electrostatic).This point is interesting since it suggests that many in-teractions experimentally shown between TFs to explaincooperativity, could result from experimental drawbacksin detecting protein interactions (Mackay et al., 2007). This problem no longer holds for the model of hierarchi-cal cooperativity described here. Moreover, the numberof chromatin-remodeling machineries in the cell (Rippeet al., 2007), which is so far intriguing, further supportsthe general importance of the present proposal.
The NF1 and Oct-1 TFs are generally expressed at highlevel by the laboratory cell lines used in the MMTV ex-periments cited above. However, in spite of their con-stitutive presence, the expression of MMTV integratedin ordered chromatin, is very low in absence of gluco-corticoid hormone (personal data not shown). Hence,NF1 and Oct-1 contribute to the MMTV transcriptionalstrength but not to the decision to transcribe or not totranscribe.
Expectedly, when chromatin remodeling is inhibited( β = 0), Eqs.(15) and (18) reduce to a simple hyper-bola αx/ (1 + αx ), but with chromatin remodeling, thebehaviour of this system is unusual compared to classicalmodes of cooperativity. It can generate an ”interrupter-like” mode of promoter functioning with strong non-linearity. For non-zero α and βx , Eq.(18) can approachthe Hill-like equation αβ (1 + γ ) x y/ (1 + αβ (1 + γ ) x y ),where the square exponent of x in absence of any pos-tulated dimerisation, reflects its dual participation inregulating both the P and P (cid:48) promoter states. Interest-ingly, the self-cooperativity of x can be obtained evenwhen B is transcriptionally inactive ( γ = 0). In classicalequilibrium mechanisms of promoter occupation, the re-sponsiveness and sensitivity to increasing concentrationsof activated TFs, is fixed by the physico-chemical cellularconditions and by the values of equilibrium constants,which are themselves dictated by macromolecule struc-tures. For example in the case of TF dimerisation, thedegree of cooperativity is not modifiable and determinedby the affinity constants for given DNA elements. Bycontrast, in the present system, further adjustments arepossible by tuning the activity and quantity of SWI/SNFchromatin remodeling enzymes (Fig.3) and the energystatus of the cell (ATP), which influence altogether the β parameter. In particular, at given TF binding poten-tials, the chromatin-remodeling rate provides a precisecontrol lever of the degree of promoter activation whichcan be coupled to additional threshold effects such asthe buffering of low levels of mRNA expression by smallRNAs, as shown in bacteria (Levine and Hwa, 2008).7 igure 3. Alternative view of the 3D plot of Fig.2c, showingthe sigmoidal response of MMTV expression to the couple ofactivated TFs GR and Oct. For a given level of activity ofSWI/SNF, this shape of the transcriptional activity surfacemakes the response of MMTV to GR and Oct strongly over-additive. Its near-horizontal slope at low GR and Oct values,allows to avoid inappropriate activation. Moreover, MMTVexpression can be further adjusted by regulating the amountor the activity of the SWI/SNF enzymes and by other thresh-old effects which can cancel transcription activation by GRalone.
Thought they are functionally important, the sub-tle differences between sigmoidal responses cannot beeasily evaluated by eye in linear coordinates. Amongthe different mathematical tools developed for analysingnon-hyperbolic fractional curves, the Hill representationlong proved very useful because it allows to focus onthe specificities of the systems (Cornish-Bowden andKoshland, 1975; Dahlquist, 1978). Using the logarithmof TF binding potentials allows to give to the range ofbinding potential between 0 to 1, the same importancethan between 1 and infinite, for better visualizing the ef-fects of ligand concentrations below midsaturation (forln( K [ T F ]) = 0). The logarithm of the ratio of frac-tional activity vs inactivity ln( F/ (1 − F )) (”logit” coor-dinate), allows to finely appreciate the behaviour of thesystem, including the degree of cooperativity between theTF(s) through the slope of the curve. An hyperbolic phe-nomenon gives a slope of 1. Indeed, when F = x/ (1 + x ),then F/ (1 − F ) = x , thus eliminating saturation effects.In addition, in multidimensional Hill plots (3D in thepresent case), the relative participation of the differentactors in the course of saturation can be visualized. Al-though the Hill plots are generally used for equilibriumphenomena such as hemoglobin oxygenation, they canalso apply to steady states. The Hill equations corre-sponding to Eqs.(15) and (18) are Eqs.(19) and (20) re-spectively: H ( X,Y ) = ln α e X (cid:104) β e X (cid:16) Y X (cid:17) + βγ e Y (cid:105) γ + α e X (cid:104) β (cid:16) Y X (cid:17) + βγ (cid:105) (19) H ( X,Y ) = ln α e X [1 + β e X (1 + e Y ) + βγ e Y (1 + e X )]1 + γ + α e X [ γ + β (1 + e Y ) + βγ (1 + e X )](20) where X = ln( x ) and Y = ln( y ). The correspond-ing plots are shown in the right panels of Fig.2 usingthe parameter combination ( α , β , γ ) = (10 − , 2, 3).In these Hill surfaces, slopes of 1 correspond to freerandom (hyperbolic) binding, while non-unity slopes de-note the existence of collective influences in the system.Specifically, steep slopes reflect a phenomenon of coop-erativity increasing the sensitivity of the system to slightchanges in ligand concentration. Near horizontal slopesand plateaus indicate the regions of relative TF ineffi-cacy as long as the concentration of the other TF is lim-iting. This is a situation of negative cooperativity. TheseHill landscapes highlight the differences between the ba-sic (but doubtful) hierarchical mechanism (right panelof Fig.2b) and the nonequilibrium model (right panel ofFig.2c). While there is no limitation other than satura-tion in the response to large Y in the equilibrium model,this is no longer the case when Y > X in Fig.2c. In thisrespect, the latter model recovers some features of theindependent system in which parallel increases of X and Y are necessary to allow their action. This property is re-lated to the fact that A can always escape the promoterand is not trapped contrary to the equilibrium model.This difference could be used as a tool for experimentallyprobing hierarchical systems. The hierarchical nature ofthese systems is illustrated by the preponderant role of A at low fractional activity. Hence, the active chromatinremodeling mechanism described here allows pronouncednon-linearity, even for monomeric TFs, which can be fur-ther enhanced by other modes of cooperativity. Hierarchical cooperativity provides an exquisite modeof sigmoidicity, in equilibrium (Fig.1a) as well as non-equilibrium conditions (Fig.1c,d). In the equilibrium sys-tem, joint sigmoidicity is obtained only in the A + B bisector, by intersecting two series of orthogonal hyper-bolas (Fig.2b, illustrated by the 2D Hill curve at X = Y ).In addition, in the active remodeling model, the responseto A alone is also sigmoidal (visible along the A axis inthe 3D plot of Fig.2c). The self-cooperativity of A fur-ther enhances the steepness of the global response in thebisector ( A + B ) (small 2D plot in Fig.2c). The max-imal Hill coefficients ( n H ) for the different models are,for the independent TFs of Fig.2a (Eq.(11)): n H ( A ) = 1, n H ( B ) = 1, n H ( A + B ) = 1; for the equilibrium modelof Fig.1a: n H ( A ) = 1, n H ( B ) = 1, n H ( A + B ) = 2and for the non-equilibrium model of Fig.1d and Fig.2c: n H ( A ) = 2, n H ( B ) = 1, n H ( A + B ) = 3. The sigmoidic-ity of this latter situation is illustrated in Fig.3. Thissource of sigmoidicity can surimpose to other ones, in-cluding: i) TF multimerization (neglected here) and ii)the cooperative recruitment of transcription machineriesby DNA-bound TFs (Michel, 2010), rarely considered in8ranscription modeling studies. For simplicity, it has notbeen not taken into account in the present study andEq.(1) describes additive contributions of the TFs A and B to the global promoter activity. Certain mouse strains contain without apparent trouble,genome-integrated MMTV which are vertically trans-mitted over generations. MMTV-infected mouse cellscan also remain healthy. Though viruses are generallydetrimental for infected cells, the issue of an infectionfor the host cells often depends on the conditions. Asit is counter-productive for a stowaway to destroy hisvehicle, viral infection is not necessarily lytic. Indeed,host genome-integrated viruses have the opportunity topropagate passively as furtive aliens, through the merespreading of the host cells. Accordingly, they developedstrategies during evolution to preserve host viability aslong as living conditions are satisfactory. In turn, whenthe viability of the host cells is menaced, the lytic phaseis triggered and leads to the production of metabolicallyinert viral particles which are more resistant to delete-rious conditions. This strategy has been observed inthe prokaryotic world, for example in the case of thelambda bacteriophage in lysogenic bacteria, but it canalso apply to certain eukaryotic integration viruses, suchas the MMTV provirus which generally remains dor-mant in adults, unless they are submitted to stresses.MMTV expression is weak in stressless conditions sincenucleosomes ensure its transcriptional silencing. In cellscontaining GR, glucocorticoid hormones trigger MMTVexpression. Glucocorticoid hormones (corticol, corticos-terone), are the hormones of nervous stress, which acti-vate the whole panoply of GR activities (nuclear import,DNA binding, transactivation, recruitement of BRG1).The secondary TFs which strongly enforce the GR ac-tion are NF1 but also Oct-1 or Oct-2. Interestingly,the preferential binding site for Oct-2 defined in (Rheeet al., 2001) precisely corresponds to the Oct modulepresent in the MMTV promoter. It is inducible by bacte-rial lipopolysaccharides (LPS) and inflammatory signals.Hence, several types of stresses: nervous (GR) and infec-tious (Oct-2), concur to activate MMTV. The sigmoidalshape of the response shown in Fig.2c, is such that thecombination of the two types of stress is required to trig-ger transcription. Moreover, the near horizontal slopeof the transcription surface and the very low responsive-ness to low GR and Oct concentrations (Fig.3), allow tobuffer stochastic fluctuations of these TFs. By this way,MMTV can remain latent in moderately stressed cells,and is revived upon conjunction of stresses (Fig.4).
Figure 4.
Example of hierarchical transcriptional cooper-ativity mediated by chromatin remodelling. ( a ) Schematicrepresentation of the proximal MMTV promoter criticallyregulated by a nucleosome (grey), which is positioned tooverlap the binding sites for GR and Oct-2. In this configu-ration, only GR can bind its target site, owing to its capacityto interact with nucleosomal DNA. Its fixation then triggersthe recruitment of the chromatin remodeller BRG1, which inturns allows the fixation of the Oct-2 factor requiring a fullyaccessible DNA helix.( b ) This systems predicts an overad-ditive combination effect and the response of the MMTV tomultiple stresses. This could be the case for example when the host-ing mouse is both frightened, with production of gluco-corticoid hormone (for example if a cat appears in theneighbourhood) and wounded (leading to a bacterial in-fection and to Oct-2 induction). When these conditionsare reunited, the mouse’s life is probabilistically compro-mised and it is beneficial for the MMTV to escape itbefore sinking with it. MMTV expression can be partic-ularly important in lymphocytes because these cells arecellular reservoirs for MMTV, undergo apoptosis uponglucocorticoid exposure and display strong Oct-2 induc-tion by inflammatory stress (Bendall et al., 1997). Thistranscriptional arbitration is equivalent to that of a crisisboard but is more economic. The conversion of randominteractions into discerning actions is a typical character-istic of dissipating systems, involving in the present caseenergy-dependent chromatin remodeling.9
Conclusion
The model proposed here is a simplification omittingmany actors in the MMTV promoter story, but is suffi-cient to reconcile conflicting data. While the first articlesconvincingly demonstrated that GR binding opens theway to NF1 and Oct-1, further studies showed that NF1and Oct-1 are present prior to glucocorticoid hormoneaddition (Belikov et al., 2004). The presetting actionof NF1 suggested in this latter article was interpretedas a locking action of NF1, that was suggested to clicknucleosome positioning in a unique configuration. Thisinterpretation is fully consistent with the mechanism pro-posed here, in which chromatin closing and NF1 bindingare mutually exclusive events. In the present model, afuzzy pattern is expected if the P and P (cid:48) states and theirtransition intermediates coexist in the cell population.This coexistence is possible in presence of GR alone (lig-and A in Fig.1c), but not of NF1 or Oct-1 (ligand B inFig.1c). The P state corresponds to the positioning ofnucleosomes thermodynamically favoured by nucleotidesequence-specific DNA bendability (Pina et al., 1990).The P (cid:48) state is a less stable configuration, whose forma-tion is forced by SWI/SNF ATPases and which is thenlocked by NF1/Oct-1 as long as present. This scheme issatisfactory in that it allows to explain previous obser-vations seemingly contradictory: (i) the initiation role ofactivated GR on NF1/Oct-1 fixation, (ii) the presettingaction of NF1/Oct-1 on GR exchanges, (iii) the fact thatGR is not trapped in presence of NF1/Oct-1. Consid-ering the abundance of chromatin remodeling factors inthe cell (Rippe et al., 2007), such a mechanism couldbe very general and provide a widespread mode of co-operativity between TFs that do not directly interactwith each other. In addition, these enzymes render cell-context specific, the role of ubiquitous actors such asthe DNA-binding elements for TFs that are common toseveral cell types. Two nuclear receptors: GR and PR(progesterone receptor), are of equivalent strength andshare the same DNA modules in the MMTV promoter;but interestingly, in a cellular context permissive forGR, PR fails to activate MMTV integrated into orderedchromatin, but induces MMTV when transfected in anopen chromatin state (Smith et al., 1997). Accordingly,PR is unable to induce chromatin remodeling at stablyintegrated MMTV templates in these cells (Smith etal., 1997; Fryer and Archer, 1998) and the reciprocalsituation is obtained in other cellular contexts (T47D,personal data). The mechanism proposed in this studycould be a pivotal device for the management of theeukaryotic genomes based on their nucleosomal organ-isation. It allows: (i) to solve apparent discrepanciesbetween experimental observations, so far barely rec-oncilable in equilibrium conditions; (ii) to establish aprimary and highly tunable mode of cooperativity be-tween TFs, considering that the chromatin-remodelingenzymes are themselves subject to refined regulations; (iii) and to bypass the need for direct interactions be-tween them, which is questioned in (Chu et al., 2009). Revision of the concept of pioneer transcrip-tion factors.
Pioneer transcription factors are definedas developmental factors opening the way to secondarytranscription factors. In this sequential mode of ac-tion, the pioneer transcription factor is envisioned as au-tonomous whereas the secondary transcription factor istributary of the pioneer one. For example, the pioneerfactors FOXA1 (also involved in the MMTV system),AP2 γ , PBX1 and GATA3, are supposed to preset chro-matin and allow the fixation of the estrogen receptor- α (ER α ) in mammary lumenal epithelial cells. More than80% of the ER α -binding sites are associated to the fixa-tion of one of these pioneer factors (Magnani and Lupien,2014). The depletion of these factors prevents ER α frombinding, but the reverse has recently also been showntrue (Caizzi et al., 2014), which singularly challenges theunilateral dependence of secondary transcription factorson pioneer factors. By contrast, the present model iscompatible with a reciprocal dependence between thesefactors. As shown in Fig.1a,b, the secondary factor canstrengthen the fixation of the pioneer factor by extendingthe fraction of time of the remodeled chromatin state, towhich the pioneer factor can bind with a higher affinity.In addition, this mechanism is dynamic, contrary to astatic hierarchical view FOXA1 → ER α , which is poorlycompatible with the half residence time of FOXA1 of 4minutes, as measured by fluorescence microscopy (Sekiyaet al., 2009). The mechanism described here predicts mu-tual influences between transcription factors, creating akey combination effect for turning on or off gene expres-sion. The present manuscript is an extended version of thearticle: Hierarchical cooperativity mediated by chromatinremodeling; the model of the MMTV transcription regu-lation. Michel, D. 2011. J. Theor. Biol. 287, 74-81.
Acknowledgement: Supported by the University ofRennes1, Action d´efits scientifiques ´emergents.
References [1] Archer, T.K., Lefebvre, P., Wolford, R.G., Hager,G.L., 1992. Transcription factor loading on the MMTVpromoter: a bimodal mechanism for promoter activa-tion. Science 255, 1573-1576.[2] Belikov, S., Gelius, B., Wrange, O., 2001. Hormone-induced nucleosome positioning in the MMTV pro-moter is reversible. EMBO J. 20, 2802-2811.[3] Belikov, S., Holmqvist, P.H., Astrand, C., Wrange,O., 2004. Nuclear factor 1 and octamer transcriptionfactor 1 binding preset the chromatin structure of the10ouse mammary tumor virus promoter for hormoneinduction. J. Biol. Chem. 279, 49857-49867.[4] Bendall, H.H., Schrerer, D.C., Edson, C.R., Ballard,D.W., Oltz, E.M., 1997. Transcription factor NF-kBregulates inducible Oct-2 gene expression in precursorB lymphocytes. J. Biol. Chem. 272, 28826-28828.[5] Bintu, L., Buchler, N.E., Garcia, H.G., Gerland, U.,Hwa, T., Kondev, J., Phillips, R., 2005. Transcrip-tional regulation by the numbers: models. Curr. Opin.Genet. Dev. 15, 116-124.[6] Bolouri, H., Davidson, E.H., 2002. Modeling DNAsequence-based cis-regulatory gene networks. Dev.Biol. 246, 2-13.[7] Caizzi, L., Ferrero, G., Cutrupi, S., Cordero, F., Bal-lar´e, C, Miano, V., Reineri, S., Ricci, L., Friard, O.,Testori, A., Cor´a, D., Caselle, M, Di Croce, L., DeBortoli, M., 2014. Genome-wide activity of unligandedestrogen receptor- αα