q-bio0702003/paper.tex
1: \documentclass[preprint,showpacs,aps,floatfix]{revtex4} 
2: \usepackage{graphicx}
3: \include{epsf}
4:  
5: \begin{document}  
6: 
7: \title{Stochastic Simulations of Genetic Switch Systems} 
8: 
9: \author{Adiel Loinger$^1$, Azi Lipshtat$^2$, 
10: Nathalie Q. Balaban$^1$ and Ofer Biham$^1$
11: }  
12: \affiliation{  
13: $^1$ Racah Institute of Physics,   
14: The Hebrew University,   
15: Jerusalem 91904,   
16: Israel
17: \\
18: $^2$ Department of Pharmacology and Biological Chemistry,
19: Mount Sinai \\ School of Medicine,
20: New York, NY 10029, USA
21: }  
22:  
23: \begin{abstract}  
24: 
25: Genetic switch systems with mutual repression of two
26: transcription factors are studied using deterministic 
27: methods (rate equations)
28: and 
29: stochastic methods 
30: (the master equation
31: and Monte Carlo simulations).
32: These systems exhibit bistability, namely
33: two stable states such that 
34: spontaneous transitions between them are rare. 
35: Induced transitions may take place as a
36: result of an external stimulus.
37: We study several variants of the genetic
38: switch and examine the effects
39: of cooperative binding, exclusive binding, protein-protein
40: interactions and degradation of bound repressors.
41: We identify the range of parameters
42: in which bistability takes place, enabling the system
43: to function as a switch. 
44: Numerous studies have concluded 
45: that cooperative binding is 
46: a necessary condition for the emergence of bistability
47: in these systems.
48: We show that a suitable combination of network structure and
49: stochastic effects gives rise to bistability even
50: without cooperative binding. 
51: The average time between spontaneous transitions is evaluated
52: as a function of the biological parameters.
53: 
54: \end{abstract}
55: 
56: \pacs{87.10.+e,87.16.-b} 
57:  
58: \maketitle  
59: 
60: \section{Introduction}
61: \label{Introduction}
62: 
63: Recent advances in quantitative measurements 
64: of gene expression at the single-cell level 
65: \cite{Elowitz2002,Ozbudak2002}
66: have brought new insight on the importance
67: of stochastic fluctuations 
68: in genetic circuits
69: \cite{Mcadams1997}.
70: The role of fluctuations is enhanced due to 
71: the discrete nature of the transcription factors
72: and their binding sites, which may appear in low copy numbers
73: \cite{Becskei2000,Kaern2005}.
74: As a result,
75: populations of genetically identical cells may 
76: show significant variability.
77: Stochastic behavior
78: may invoke oscillations
79: \cite{Vilar2002}
80: and spatio-temporal patterns
81: \cite{Shnerb2000},
82: which are unaccounted for by
83: macroscopic chemical rate equations. 
84: Genetic circuits with
85: feedback mechanisms may exhibit bistability,
86: namely, two distinct stable states which
87: can be switched by an external 
88: signal
89: \cite{Atkinson2003}.
90: A low rate of spontaneous switching events
91: may also take place.
92: To qualify as a switch, this rate
93: must be much lower than the rates of the relevant
94: processes in the cell, namely transcription, translation,
95: and degradation of transcription factors.
96: Genetic switches, such as the 
97: phage $\lambda$ switch, 
98: give rise to different cell fates
99: \cite{Ptashne1992}.
100: In this switch, $\lambda$
101: phages infect {\it E. coli} bacteria and 
102: can exist in two exclusive states,
103: one called 
104: lysogeny
105: and the other called 
106: lysis. 
107: When the phage enters its host, it integrates itself into the
108: host's DNA and is duplicated by cell division.
109: It codes for proteins that can identify stress in the host cell.
110: In case of stress, the phage transforms into the lysis state.
111: In this state, it kills the host cell, using its DNA to produce
112: many copies of the phage, which are released and later infect other
113: cells.
114: Other switch circuits exist in the metabolic systems of cells.
115: These switches determine which type of sugar the cell will digest
116: \cite{Lewis2005}.
117: The genetic switch may also serve as a memory unit of the cell,
118: and help determine its fate during cell differentiation.
119: 
120: Recent advances enable the construction of genetic circuits
121: with desired properties, that are determined by the network
122: architecture.
123: These networks are constructed from available components, namely 
124: genes and promoters. They do not require the manipulation of the 
125: structure of proteins and other regulatory elements at the
126: molecular level.
127: These genes and promoters are often inserted into plasmids
128: rather than on the chromosome.
129: A synthetic toggle switch, that consists of two repressible promoters
130: with mutual negative regulation,
131: was constructed in {\it E. coli} and the conditions for bistability
132: were examined
133: \cite{Gardner2000}.
134: The switching between its two states was demonstrated using 
135: chemical and thermal induction.
136: More recently, such circuit was found to exist in a natural
137: system in which two mutual repressors regulate
138: the differentiation of myeloid progenitors
139: into either macrophages or neutrophils
140: \cite{Laslo2006}.
141: 
142: In this paper we analyze   
143: the genetic toggle switch using deterministic and stochastic methods.
144: In this simple genetic circuit, 
145: two proteins, $A$ and $B$,
146: negatively regulate each other's synthesis.
147: The regulation is performed at the transcription level, namely
148: the production of protein $A$ is 
149: negatively regulated by protein $B$,
150: through binding of $n$ copies of $B$
151: to the $A$ promoter (and vice versa). 
152: This process can be
153: modeled by a Hill function, which 
154: reduces the production rate of
155: $A$ 
156: by a factor of
157: $1+k[B]^n$,
158: where 
159: $[B]$
160: is the concentration of $B$ proteins
161: in the cell,
162: $k$ is a parameter and 
163: $n$ is the Hill coefficient
164: \cite{Hill}.
165: In case that $n=1$,
166: the binding of a single protein is 
167: sufficient in order to perform the negative regulation, while
168: for $n>1$ the {\it cooperative binding} of two or more
169: proteins is required.
170: 
171: One may expect this circuit to function as a switch,
172: with two stable states, one dominated by $A$ proteins
173: and the other dominated by $B$ proteins.
174: When the population of $A$ proteins is larger than the population of $B$ 
175: proteins, the $A$ proteins suppress the production of $B$ 
176: proteins. Under these conditions, 
177: the production of $A$ proteins will not be suppressed
178: much by the small $B$ population. 
179: Therefore, the system approaches a state
180: rich in $A$ proteins and poor in $B$ proteins. 
181: Similarly, the system may approach a state rich 
182: in $B$ proteins and poor
183: in $A$ proteins.
184: 
185: To qualify as a switch, the system should be 
186: bistable. 
187: In the deterministic description, bistability is defined
188: as the existence of two stable steady state solutions 
189: of the rate equations.
190: This description does not account for the possibility of
191: spontaneous transitions between the two states.
192: In the stochastic description, spontaneous transitions
193: do take place. Therefore, the condition for bistability
194: is that the rate of
195: spontaneous switching events 
196: (due to random fluctuations rather than 
197: an external signal) 
198: is much lower than the rates of
199: all other relevant processes
200: in the system. 
201: 
202: Rate equations provide the average concentrations of $A$ and $B$
203: proteins in a population of cells.
204: In these equations, bistability emerges at a bifurcation
205: point, where two stable states emerge.
206: Rate equations do not include fluctuations and do
207: not account for the possibility of spontaneous transitions
208: between the two states.
209: The master equation provides the probability distribution
210: of the populations of $A$ and $B$ proteins.
211: The two bistable states appear as two distinct peaks
212: in this distribution.
213: Monte Carlo simulations enable to follow the fluctuations
214: in a single cell and to evaluate the rate of spontaneous
215: switching events.
216: 
217: We examine the conditions for the system to become 
218: a switch, and
219: calculate the rate of spontaneous transitions between its
220: two states.
221: This is done for several variants of the toggle switch.
222: In particular, 
223: we focus on switch systems in which the repression in done
224: without cooperative binding (namely, $n=1$). 
225: Numerous studies have concluded,
226: using rate equations,
227: that cooperative binding is
228: a necessary condition for the emergence of bistability
229: \cite{Gardner2000,Cherry2000,Warren2004,Warren2005,Walczak2005}. 
230: Below we show,
231: using a combination of deterministic and stochastic simulation methods,
232: that this is not the 
233: case, namely a bistable switch can exist even in
234: the absence of cooperative binding.
235: In particular, we show that bound-repressor degradation (BRD)
236: and protein-protein interactions (PPI)
237: give rise to bistability,
238: without cooperative binding,
239: even at the level of rate equations.
240: These results are confirmed by stochastic simulations using
241: the master equation and Monte Carlo methods.
242: We also consider the exclusive switch,
243: in which the $A$ and $B$ repressors cannot be bound simultaneously
244: due to overlap between their promoter sites.
245: This system
246: exhibits bistability only when 
247: stochastic fluctuations are taken into account.
248: The rate of spontaneous transitions between the two states
249: is calculated as a function of the biological parameters.
250: 
251: The paper is organized as follows.
252: In Sec.
253: \ref{sec:general}
254: we consider the basic version called the general switch.
255: Several variants of this circuit are considered in the
256: Sections that follow.
257: The exclusive switch is studied in
258: Sec.
259: \ref{sec:exclusive},
260: the BRD switch is considered in 
261: Sec.
262: \ref{sec:BRD}
263: and the PPI switch is analyzed in 
264: Sec.
265: \ref{sec:PPI}.
266: The effects of cooperative binding are studied in
267: Sec.
268: \ref{sec:cooperative}.
269: The response of toggle switch systems to external 
270: signals is examined in 
271: Sec.
272: \ref{sec:response}.
273: The results are discussed in 
274: Sec.
275: \ref{sec:discussion}
276: and summarized in
277: Sec. 
278: \ref{sec:summary}.
279: 
280: \section{The General Switch (Without Cooperative Binding)}
281: \label{sec:general}
282: 
283: The general switch consists of two transcription factors,
284: $A$ and $B$, that negatively regulate each other's
285: synthesis
286: \cite{Warren2004,Warren2005}.
287: A schematic description of 
288: this circuit is given in
289: Fig. \ref{fig1}(a).
290: The regulation is done by the binding of a protein to
291: the promoter site of the other gene, blocking the access
292: of the RNA polymerase and suppressing the transcription 
293: process.
294: In this circuit there is no cooperative binding, namely
295: the regulation process is performed by a single bound
296: protein.
297: 
298: The concentrations of free $A$ and $B$ proteins in the cell
299: are denoted by $[A]$ and $[B]$, respectively (by concentration
300: we mean the average copy number of proteins per cell).
301: The copy numbers of the bound proteins, are denoted by 
302: $[r_A]$ and $[r_B]$,
303: where $r_A$ is a bound $A$ protein that monitors the production of $B$, 
304: while $r_B$ is a bound $B$ protein that monitors the production of $A$.
305: Note that there is at most one bound repressor of each type at any
306: given time, and thus
307: $0 \le r_A,r_B \le 1$.
308: For simplicity, we ignore the mRNA level and combine the 
309: processes of transcription and translation as a single 
310: step of synthesis
311: \cite{mRNA}. 
312: 
313: The maximal production rate of protein $X$ is denoted by
314: $g_X$ (s$^{-1}$), $X=A,B$.
315: The degradation rate of protein $X$ is given by
316: $d_X$ (s$^{-1}$).
317: While the structure of the circuits studied here
318: is symmetric,
319: the rate constants can be different for $A$ and $B$. 
320: However, for simplicity we use symmetric parameters,
321: i.e. $g=g_A=g_B$ and $d=d_A=d_B$.
322: The binding rate of proteins to the promoter
323: is denoted by $\alpha_0$ (s$^{-1}$)
324: and the dissociation rate
325: by $\alpha_1$ (s$^{-1}$). 
326: 
327: \subsection{Rate Equations}
328: 
329: The dynamics of the general switch circuit 
330: is described by the rate equations
331: \cite{Lipshtat2005,Lipshtat2006}
332: 
333: \begin{eqnarray}
334: \dot{[A]} &=& g_A (1-[r_B]) - d_A [A]-
335: \alpha_0 [A] \left (1-[r_A] \right) + \alpha_1[r_A]  \nonumber \\
336: \dot{[B]} &=& g_B(1-[r_A])-d_B[B]-
337: \alpha_0[B]\left(1-[r_B]\right)+\alpha_1[r_B] \nonumber \\
338: \dot{[r_A]} &=& \alpha_0[A]\left(1-[r_A]\right)-\alpha_1[r_A] \nonumber \\
339: \dot{[r_B]} &=& \alpha_0[B]\left(1-[r_B]\right)-\alpha_1[r_B].
340: \label{eq:general_switch_rate}
341: \end{eqnarray}
342: 
343: \noindent
344: It is commonly assumed that the 
345: binding-unbinding processes 
346: are much faster than other 
347: processes in the circuit, namely 
348: $\alpha_0,\alpha_1 \gg d_X,g_X$. 
349: This means that the relaxation times of $[r_X]$ 
350: are much shorter than other relaxation 
351: times in the circuit. 
352: Under this assumption,
353: one can take the 
354: time derivatives of $[r_X]$
355: to zero, even if the system is away from
356: steady state.
357: This brings the rate equations to the 
358: standard Michaelis-Menten form
359: 
360: %
361: \begin{eqnarray}    
362: \dot{[A]} &=& \frac{g}{1+k[B]} - d [A] \nonumber \\
363: \dot{[B]} &=& \frac{g}{1+k[A]} - d [B],
364: \label{eq:general_switch_michaelis_menten}
365: \end{eqnarray}    
366: 
367: \noindent
368: where symmetric parameters are used, and
369: $k=\alpha_0/\alpha_1$ 
370: is the repression strength.
371: For a given population of free $X$ repressors,
372: the parameter $k$  controls the value of $[r_X]$.
373: The limit of weak repression,
374: $[r_X] \ll 1$,
375: is obtained when
376: $k[X] \ll 1$,
377: while the limit of strong repression,
378: $[r_X] \simeq 1$,
379: is obtained for
380: $k[X] \gg 1$. 
381: 
382: The meaning of bistability at the level of rate equations is
383: that at steady state the equations exhibit two distinct positive
384: solutions.
385: In this particular class of circuits, 
386: one solution is dominated by $A$ proteins and the other
387: is dominated by $B$ proteins.
388: Starting from any initial state, the system will converge
389: to one of these solutions.
390: The solutions are stable, so the possibility of spontaneous 
391: transitions, induced by stochastic fluctuations, 
392: is not included in the rate equation description.
393: 
394: The steady state solutions of Eqs.
395: (\ref{eq:general_switch_rate})
396: and 
397: (\ref{eq:general_switch_michaelis_menten})
398: are identical.
399: We will now show that
400: these equations have only one positive steady-state solution. 
401: To this end, we first
402: take $\dot{[A]}=\dot{[B]}=0$ 
403: in Eq.
404: (\ref{eq:general_switch_michaelis_menten}).
405: We multiply each equation by the denominator
406: of the Hill function that appears in it. 
407: We obtain:
408: 
409: \begin{eqnarray}    
410: g - d[A] - kd[A][B] &=& 0 \nonumber \\
411: g - d[B] - kd[A][B] &=& 0.
412: \label{eq:general_switch_hill1}
413: \end{eqnarray}    
414: 
415: \noindent
416: Subtracting one equation from the other we get 
417: $d([A]-[B])=0$ 
418: and therefore 
419: $[A]$ must be equal to $[B]$ at steady state.
420: The steady state values of 
421: $[A]$ and $[B]$ can be easily found. 
422: Inserting
423: $[A]=[B]$ into 
424: Eq.~(\ref{eq:general_switch_hill1}) 
425: we obtain a quadratic equation
426: whose only positive solution is 
427: 
428: \begin{equation}
429: [A] = [B] =  \frac {-1+\sqrt{1+4kg/d}}{2k}.
430: \end{equation}
431: 
432: \noindent
433: Standard linear stability analysis shows that 
434: this solution is always stable.
435:     
436: As a result, we conclude that at the level of rate equations
437: the general switch, without cooperative binding,
438: does not exhibit bistability.
439: In Sec. 
440: \ref{sec:cooperative}
441: we consider the case of cooperative binding, where
442: the rate equations do exhibit bistability.
443: 
444: \subsection{Master Equation}
445: 
446: In order to account for stochastic effects  
447: and to obtain insight on 
448: the reason that this system is not bistable, 
449: the master equation approach is applied
450: \cite{Mcadams1997,Arkin1998,Kepler2001,Paulsson2000,Paulsson2004}.
451: In this case, we consider
452: the probability distribution function
453: $P(N_A,N_B,r_A,r_B)$.
454: It is the probability for a cell to 
455: include $N_X$ copies of free protein $X$
456: and  $r_X$ copies of the bound $X$ repressor,
457: where $N_X=0,1,2,\dots$,  
458: and $r_X = 0,1$.
459: The master equation for the general switch takes the form
460: 
461: \begin{eqnarray}
462: && \dot{P}(N_A,N_B,r_A,r_B) = 
463: g_{\rm A} \delta_{r_B,0} 
464: [P(N_A-1,N_B,r_A,r_B) - P(N_A,N_B,r_A,r_B)]  \nonumber\\
465: && + g_{\rm B} \delta_{r_A,0} 
466: [P(N_A,N_B-1,r_A,r_B) - P(N_A,N_B,r_A,r_B)] \nonumber\\
467: && + d_{\rm A} [(N_A+1) P(N_A+1,N_B,r_A,r_B) - N_A P(N_A,N_B,r_A,r_B)]  
468: \nonumber\\
469: && + d_{\rm B} [(N_B+1) P(N_A,N_B+1,r_A,r_B) - N_B P(N_A,N_B,r_A,r_B)]  
470: \nonumber\\
471: && + \alpha_0 [(N_A+1) \delta_{r_A,1} P(N_A+1,N_B,r_A-1,r_B) - 
472:               N_A \delta_{r_A,0} P(N_A,N_B,r_A,r_B)]  \nonumber\\
473: && + \alpha_0 [(N_B+1) \delta_{r_B,1} P(N_A,N_B+1,r_A,r_B-1) - 
474:               N_B \delta_{r_B,0} P(N_A,N_B,r_A,r_B)]  \nonumber\\
475: && + \alpha_1 [\delta_{r_A,0} P(N_A-1,N_B,r_A+1,r_B) -
476:              \delta_{r_A,1} P(N_A,N_B,r_A,r_B)]  \nonumber\\
477: && + \alpha_1 [\delta_{r_B,0} P(N_A,N_B-1,r_A,r_B+1) -
478:              \delta_{r_B,1} P(N_A,N_B,r_A,r_B)], 
479: \label{eq:general_switch_master}
480: \end{eqnarray}
481: 
482: \noindent
483: where 
484: $\delta_{i,j}=1$ for $i=j$ and $0$ otherwise.
485: The $g_X$ terms account for the production of proteins.
486: The $d_X$ terms account for the degradation of free proteins,
487: while the $\alpha_0$ ($\alpha_1$) terms describe the 
488: binding (unbinding) of proteins to (from) the promoter site. 
489: In numerical integration, the master equation must be truncated
490: in order to keep the number of equations finite. 
491: This is done by setting suitable upper cutoffs,
492: $N_{\rm A}^{\rm max}$ and $N_{\rm B}^{\rm max}$,
493: on the populations sizes of free proteins.
494: In order to maintain the accuracy of the calculations,
495: the probability of population sizes beyond the cutoffs
496: must be sufficiently small.
497: 
498: The master equation has a single steady state solution, 
499: which is always stable 
500: \cite{VanKampen1992}.
501: The criterion for bistability  
502: is that the steady state solution 
503: $P(N_A,N_B,r_A,r_B)$ 
504: exhibits two distinct regions (peaks) of high  
505: probabilities, 
506: separated by a gap in which the probabilities are very small.
507: These two regions correspond to the two  
508: states in which the system is likely to be.
509: %Although this is a steady state solution of the master equation, the
510: %system can still move from state to state. 
511: If the transition rate between the peaks
512: is small enough, 
513: the system is indeed a bistable switch.
514: Note, that in this case, averages of the form
515:   
516: \begin{equation}
517: \langle N_{\rm X} \rangle =
518: \sum_{N_{\rm A}=0}^{N_{\rm A}^{\rm max}} 
519: \sum_{N_{\rm B}=0}^{N_{\rm B}^{\rm max}} 
520: \sum_{r_{\rm A}=0}^1 
521: \sum_{r_{\rm B}=0}^1 
522: N_{\rm X} P(N_A,N_B,r_A,r_B),
523: \label{eq:averagex}
524: \end{equation}
525: 
526: \noindent
527: where $X=A$, $B$,
528: do not reflect the complex structure of the 
529: probability distribution.
530: These can be considered as averages over many
531: cells, some dominated by $A$ and others dominated by $B$
532: proteins, such that the total populations of the two species
533: are about the same.
534: 
535: To examine  
536: the existence of bistability we consider
537: the marginal probability distribution 
538: 
539: \begin{equation}
540: P(N_A,N_B) = 
541: \sum_{r_{\rm A}=0}^1
542: \sum_{r_{\rm B}=0}^1
543: P(N_A,N_B,r_A,r_B).
544: \end{equation}
545: \noindent
546: 
547: \noindent
548: This probability distribution was calculated 
549: for a broad range of parameters.
550: Two representative examples are shown in 
551: Fig. \ref{fig2}.
552: 
553: Under conditions of weak repression (small k),
554: $P(N_A,N_B)$ 
555: exhibits a single peak for which 
556: $N_A \approx N_B \approx g/d$
557: [Fig. \ref{fig2}(a)],
558: in agreement with the rate equations.
559: This is due to the fact that the
560: repression is weak, and the 
561: $A$ and $B$ populations are almost uncorrelated.
562: In this case, the cell will contain roughly the same
563: amount of $A$ and $B$ proteins. 
564: 
565: For strong repression, the distribution $P(N_A,N_B)$ 
566: exhibits a peak dominated by $A$ proteins
567: and a peak dominated by $B$ proteins,	
568: as expected for a bistable system. 
569: However, a third peak appears near the origin,
570: in which both populations of free proteins 
571: are suppressed
572: [Fig. \ref{fig2}(b)]. 
573: This peak represents a dead-lock situation, caused by the 
574: fact that both $A$ and $B$ repressors can be bound
575: simultaneously, each bringing to a halt the production of the other specie.
576: The third peak provides a corridor through which the probability can
577: flow between the other two peaks.
578: As a result, the system can quickly switch between the
579: $A$-dominated and the $B$-dominated states.
580: 
581: 
582: In addition to the solution of the master equation, 
583: Monte Carlo simulations
584: have been performed. 
585: In these simulations one can follow the time evolution
586: of the populations of free and bound proteins
587: in a single cell. 
588: In 
589: Fig. \ref{fig3}(a)
590: we present the population sizes of free proteins vs. time
591: for the general switch.
592: It is clear that the cell can indeed
593: be in one of three states: a state rich in $A$, 
594: a state rich in $B$ and a state
595: in which both proteins are in very low copy numbers. 
596: We conclude that a necessary condition for the system to become a switch
597: is to prevent this dead-lock situation in which both protein
598: populations are suppressed simultaneously. 
599: Below we present several 
600: variants of the circuit in which the third peak is suppressed, 
601: giving rise to a bistable switch. 
602: 
603: 
604: \section{The Exclusive Switch}
605: \label{sec:exclusive}
606: 
607: The first variant we consider is the exclusive switch,
608: depicted in 
609: Fig.~\ref{fig1}(b).
610: In this circuit 
611: there is an overlap between the promoters 
612: of $A$ and $B$.
613: As a result, there is
614: no room for both  
615: $A$ and $B$ proteins to be bound
616: simultaneously.
617: Exclusive binding is encountered in nature,  
618: for example, in the 
619: lysis-lysogeny switch of phage $\lambda$
620: \cite{Ptashne1992}.
621: 
622: It was shown that 
623: in presence of cooperative binding,
624: the exclusive switch is more stable than
625: the general switch
626: \cite{Warren2004,Warren2005}.
627: This is because in the exclusive switch the access of the
628: minority proteins to the promoter site is blocked by the
629: dominant proteins.
630: Here we show that in the exclusive switch, 
631: stochastic effects give rise to bistability even
632: without cooperativity between the transcription factors.
633: The dead-lock situation in prevented in this case, since 
634: $A$ and $B$
635: repressors cannot be bound simultaneously. 
636: 
637: \subsection{Rate Equations}
638: 
639: To model 
640: the exclusive switch,
641: recall that the variable
642: $[r_A]$  ($[r_B]$)  
643: is actually the fraction of 
644: time in which the promoter 
645: is occupied by a bound $A$ ($B$) protein
646: \cite{Lipshtat2005}.
647: The fraction of time in which the promoter is vacant is
648: thus
649: $1 - [r_A] - [r_B]$.
650: Incorporating this into 
651: Eq.~(\ref{eq:general_switch_rate})
652: gives rise to the following modification:
653: in the $\alpha_0$ terms,
654: each appearance of $[r_A]$ or $[r_B]$
655: should be replaced by $[r_A]+[r_B]$.
656: With this modification, the rate equations take
657: the form
658: 
659: \begin{eqnarray}
660: \dot{[A]} &=& g (1-[r_B]) - d [A]-
661: \alpha_0 [A] \left (1-[r_A]-[r_B] \right) 
662: + \alpha_1 [r_A]  \nonumber \\
663: \dot{[B]} &=& g (1-[r_A]) - d [B] -
664: \alpha_0 [B]\left(1-[r_A]-[r_B] \right)
665: +\alpha_1 [r_B] \nonumber \\
666: \dot{[r_A]} &=& \alpha_0 [A] 
667: \left(1- [r_A] - [r_B] \right)-\alpha_1 [r_A]  \nonumber \\
668: \dot{[r_B]} &=& \alpha_0 [B] 
669: \left(1- [r_A] - [r_B] \right)-\alpha_1 [r_B].
670: \label{eq:exclusive_rate}
671: \end{eqnarray}
672: 
673: \noindent
674: Under steady state conditions, the rate equations can be reduced
675: to the Michaelis-Menten form
676: 
677: \begin{eqnarray}
678: \dot{[A]} &=& \frac{g}{1+{k[B]/(1+k[A])}} - d[A] \nonumber \\
679: \dot{[B]} &=& \frac{g}{1+{k[A]/(1+k[B])}} - d[B],
680: \label{eq:exclusive_michaelis_menten}
681: \end{eqnarray}   
682: 
683: \noindent
684: where, as before, $k=\alpha_0/\alpha_1$.
685: We will now show that even for the case of the 
686: exclusive switch, the rate equations still exhibit 
687: a single solution, thus there is no bistability.
688: This is done by 
689: taking $\dot{[A]}=\dot{[B]}=0$ and getting rid of the
690: denominators, by repeated multiplications.
691: The resulting equations are
692: 
693: \begin{eqnarray}
694: g + (kg- d) [A]  -kd[A] ([A]+[B]) &=& 0 \nonumber \\
695: g + (kg- d) [B]  -kd[B] ([A]+[B]) &=& 0. 
696: \label{Brep2}
697: \end{eqnarray}   
698: 
699: \noindent
700: By subtraction of one equation from the other, 
701: we find that
702: 
703: \begin{equation}
704: \{kg-d - kd ([A]+[B])\}([A]-[B]) = 0. 
705: \label{Brep3}
706: \end{equation}   
707: 
708: \noindent
709: The positive, symmetric solution, $[A]=[B]$, is given by
710: 
711: \begin{equation}
712: [A] = \frac{(kg-d) + \sqrt{(kg+d)^2 +4kgd}}{4kd}.
713: \end{equation}   
714: 
715: \noindent
716: The other, non-symmetric solution, given by
717: 
718: \begin{equation}
719: kg-d -kd ([A]+[B]) = 0
720: \label{eq:nonsymmsol}
721: \end{equation}
722: 
723: \noindent
724: is inconsistent with
725: Eq.~(\ref{Brep2})
726: unless $g=0$, 
727: namely there is no production of $A$ and $B$ proteins,
728: which immediately leads to $[A]=[B]=0$.
729: Under these conditions the solution of Eq.
730: (\ref{eq:nonsymmsol})  
731: is 
732: $[A]+[B]=-1/k$, which requires a negative population size
733: and thus makes no physical sense.
734: Therefore, the only solution for $g>0$
735: is the symmetric solution, $[A]=[B]$.
736: Thus, the rate equations do not support a bistable
737: solution for the exclusive switch for any choice
738: of the parameters.
739: 
740: \subsection{Master Equation}
741: 
742: To account for the effects of fluctuations, 
743: we now describe the exclusive switch using 
744: the master equation.
745: It is similar to to master equation 
746: for the general switch given by 
747: Eq.~(\ref{eq:general_switch_master}),
748: except for the following modifications:
749: (a) In the $\alpha_0$ and $\alpha_1$ terms,
750: each time $\delta_{r_A,j}$ ($\delta_{r_B,j}$),
751: $j=0,1$,
752: appears it should be multiplied by
753: $\delta_{r_B,0}$, ($\delta_{r_A,0}$);
754: (b) The constraint 
755: $P(N_A,N_B,1,1)=0$ should be imposed.
756: Implementing these changes we obtain the following equation:
757: 
758: \begin{eqnarray}
759: \label{eq:exclusive_switch_master}
760: && \dot{P}(N_A,N_B,r_A,r_B) = 
761:    g_{\rm A} \delta_{r_B,0} 
762: [P(N_A-1,N_B,r_A,r_B) - P(N_A,N_B,r_A,r_B)]  \nonumber\\
763: && + g_{\rm B} \delta_{r_A,0} 
764: [P(N_A,N_B-1,r_A,r_B) - P(N_A,N_B,r_A,r_B)] \nonumber\\
765: && + d_{\rm A} [(N_A+1) P(N_A+1,N_B,r_A,r_B) 
766: - N_A P(N_A,N_B,r_A,r_B)]  \nonumber\\
767: && + d_{\rm B} [(N_B+1) P(N_A,N_B+1,r_A,r_B) 
768: - N_B P(N_A,N_B,r_A,r_B)]  \nonumber\\
769: && + \alpha_0 \delta_{r_B,0} 
770: [(N_A+1) 
771: \delta_{r_A,1}
772: P(N_A+1,N_B,r_A-1,r_B) - 
773: N_A \delta_{r_A,0}
774: P(N_A,N_B,r_A,r_B)] 
775:  \nonumber\\
776: && + \alpha_0 \delta_{r_A,0} 
777: [(N_B+1) \delta_{r_B,1} P(N_A,N_B+1,r_A,r_B-1) - 
778:               N_B \delta_{r_B,0} P(N_A,N_B,r_A,r_B)] 
779:  \nonumber\\
780: && + \alpha_1 [\delta_{r_A,0} P(N_A-1,N_B,r_A+1,r_B) -
781:              \delta_{r_A,1} P(N_A,N_B,r_A,r_B)]  \nonumber\\
782: && + \alpha_1 [\delta_{r_B,0} P(N_A,N_B-1,r_A,r_B+1) -
783:              \delta_{r_B,1} P(N_A,N_B,r_A,r_B)]. 
784: \end{eqnarray}
785: 
786: \noindent
787: For the exclusive switch, as for the general switch,
788: under conditions of weak repression,
789: $P(N_A,N_B)$ exhibits a single peak
790: [Fig. \ref{fig4}(a)]
791: that satisfies $N_A \approx N_B \approx g/d$.
792: However, as the repression strength increases 
793: two distinct peaks begin to form.
794: For intermediate values of $k$ these peaks are still connected, 
795: by a corridor of non-vanishing probabilities
796: [Fig. \ref{fig4}(b)]. 
797: Monte Carlo simulations show that for intermediate values of $k$,
798: the system indeed exhibits two states, 
799: one rich in $A$ and the other rich in $B$, 
800: but rapid transitions occur between 
801: them.
802: 
803: For strong repression,
804: the distribution 
805: $P(N_A,N_B)$ 
806: exhibits two peaks which are separated by 
807: a region with vanishing probabilities
808: [Fig. \ref{fig4}(c)].
809: In one peak the $A$ population is suppressed, while in the 
810: other peak the $B$ population is suppressed, 
811: as expected for a bistable system. 
812: The average population of the dominant protein specie in each peak is 
813: $\langle N_{\rm X} \rangle \approx g/d$,
814: while the population of the suppressed specie is 
815: $\langle N_{\rm X} \rangle \approx 0$. 
816: Monte Carlo simulations show that in this case the average
817: time between spontaneous transitions is much longer. 
818: The typical switching time  for the case shown
819: in Fig. \ref{fig3}(b)
820: is around $10^{5}$ seconds. 
821: It is much longer than the time-scales of the transcription,
822: translation and degradation processes.
823: It is also longer than the time between cell divisions
824: which is of the order of $10^3-10^4$ seconds.
825: The Monte Carlo results clearly
826: show a large number of failed attempts in which a protein of the minority 
827: specie binds to the promoter and then unbinds again, without causing 
828: the system to flip.
829: 
830: \subsection{Analysis of Switching Times}
831: 
832: To evaluate the switching times
833: we performed the following procedure.
834: We initialized the master equation in a state which
835: is completely dominated by A proteins, 
836: namely,
837: $P(N_A=\lfloor g/d \rfloor,N_B=0,r_A=0,r_B=0)=1$
838: (where $\lfloor \  \rfloor$ represents the integer part),
839: and all other probabilities vanish.
840: The master equation was then integrated numerically 
841: and $P(N_A,N_B)$ was calculated 
842: as a function of time. 
843: The function 
844: $f(t)=P(N_A>N_B)-P(N_A<N_B)$
845: was found to decay exponentially 
846: from its initial value,
847: $f(0)=1$, to zero,
848: according to
849: $f(t)=\exp(-t/\tau)$. 
850: %[Fig.~\ref{fig7}],
851: The time constant
852: $\tau$ is defined as the switching time
853: \cite{TauMEMC}. 
854: Its inverse, $\tau^{-1}$, 
855: is referred to as the switching rate. 
856: 
857: Using this procedure, we examined the dependence of the
858: switching time, $\tau$ on the protein synthesis rate, $g$
859: [Fig. \ref{fig5}(a)], 
860: the degradation rate, $d$
861: [Fig. \ref{fig5}(b)], 
862: and the repression strength, $k$ 
863: [Fig. \ref{fig5}(c)].
864: In the parameter range in which bistability takes place,
865: we obtain that
866: (a) $\tau \sim g$,
867: (b) $\tau \sim 1/d^2$
868: and
869: (c) $\tau \sim k$.
870: Concerning 
871: Fig. \ref{fig5}(c),
872: note that system exhibits bistability only in the regime in which $k$
873: is large
874: \cite{OnlyK2006}. 
875: For $k<1$ , $\tau \sim 100-1000$ (s), which is the typical time-scale
876: of other processes in the cell.
877: Only for $k \gtrsim 10$, 
878: $\tau$ becomes significantly larger than 
879: the time scales of other processes, and 
880: the system can function as a stable switch. 
881: The scaling properties of the switching time can be summarized by
882: 
883: \begin{equation}
884: \tau \sim {\alpha_0 \over \alpha_1} {g \over d^2}.
885: \label{eq:exclusive_scaling}
886: \end{equation}
887: 
888: \noindent
889: This result can be reproduced by a simple argument.
890: Consider an initial state in which the system is dominated by
891: $A$ proteins, while the population of $B$ proteins is suppressed,
892: namely
893: $[A] \gg [B]$.
894: In this situation the promoter site is occupied by an $A$ protein
895: during most of the time. 
896: In order that the switch will flip, 
897: the bound A protein must
898: unbind (at rate $\alpha_1$). 
899: Then, a $B$ protein (rather than an $A$ protein)
900: should bind to the promoter.
901: The probability for this to happen is
902: $\sim [B]/[A]$. 
903: This $B$ protein
904: should remain bound long enough in order to build up a sufficiently
905: large population of $B$ proteins.
906: On average, the $B$ protein stays bound $1/\alpha_1$ (s), 
907: during which $g/\alpha_1$ proteins of type $B$ are produced.
908: After the $B$ repressor will unbind, the probability that 
909: the next protein that binds will be of type $B$ rather than $A$,
910: is thus
911: $\sim (g/\alpha_1)/[A]$ 
912: (neglecting the degradation of $A$ proteins,
913: because $\alpha_1 \gg d$).
914: Following this argument, 
915: the switching rate is given by
916: 
917: \begin{equation}
918: \tau^{-1} \sim
919: \alpha_1 
920: \times {[B] \over [A]} 
921: \times { {g} \over {\alpha_1 [A]} } 
922: = g { [B] \over [A]^2 }.
923: \label{eq:exclusive_scaling2}
924: \end{equation}
925: 
926: \noindent
927: From the Michaelis-Menten equations we obtain that 
928: 
929: \begin{equation}
930: \frac{[B]}{[A]} = \frac{1}{1 + k [A]} \approx \frac{1}{k[A]}, 
931: \end{equation}
932: 
933: \noindent
934: since 
935: for strong repression
936: $k[A] \gg 1$. 
937: Inserting this result into Eq. 
938: (\ref{eq:exclusive_scaling2})
939: and
940: using
941: and 
942: $[A] \approx g/d$
943: we find that
944: 
945: \begin{equation}
946: \tau = \frac{kg}{d^2}.
947: \end{equation}
948: 
949: \noindent
950: This result can be considered as the leading term in the expansion
951: of $\tau$ in powers of $g$, $d$ and $k$. 
952: This leading term turns out to provide a
953: very good approximation to simulation results. 
954: For example
955: for $g=0.2$,
956: $d=0.005$,
957: $\alpha_0=0.2$,
958: and
959: $\alpha_1=0.01$
960: (s$^{-1}$) 
961: we get 
962: $\tau=1.6 \cdot 10^5$ (s),
963: which agrees perfectly with 
964: the results of Monte Carlo simulations. 
965: 
966: From Eq.
967: (\ref{eq:exclusive_scaling2}), 
968: and from the fact that
969: the average copy number of the 
970: dominant specie is 
971: $[A] \approx g/d$, 
972: we find that
973: when the production rate, $g$, is varied while keeping
974: all other parameters fixed,
975: $\tau \sim [A]$.
976: Otherwise,
977: when the degradation rate, $d$, is varied while all other parameters
978: are fixed,
979: $\tau \sim [A]^2$. 
980: In general,
981: the switching time is 
982: $\tau = \tau(k,g,d)$,
983: while the population size, $[A]$, 
984: of the dominant specie
985: depends on both $g$ and $d$.
986: Thus,
987: by a suitable variation of the rate constants,
988: any desired dependence of $\tau$ on
989: $[A]$
990: can be obtained.
991: In particular, 
992: by increasing $k$,
993: $\tau$ can be increased with no effect on
994: $[A]$.
995: A similar result is obtained when $g$ and $d$ are decreased
996: by the same factor.
997: We thus conclude that the population size is only one
998: of several factors that affect the switching time.
999: A complete description of the switching time should
1000: include all the relevant rate constants.
1001:  
1002: In Monte Carlo simulations of
1003: a switch system with cooperative
1004: binding, the switching time was found to 
1005: depend exponentialy on the copy number
1006: \cite{Warren2004,Warren2005}.
1007: This is consistent with the discussion above, but requires
1008: a well defined protocol according to which
1009: the rate constants are varied.
1010: 
1011: \section{The Switch with Bound Repressor Degradation}
1012: \label{sec:BRD}
1013: 
1014: Consider a different variant of the general switch, 
1015: in which not only free repressors,
1016: but also bound repressors
1017: are affected by degradation.
1018: The bound-repressor degradation (BRD) 
1019: tends to prevent the dead-lock situation in which 
1020: both $A$ and $B$ repressors are bound simultaneously. 
1021: This is due to the fact that degradation removes the bound repressor 
1022: from the system, unlike 
1023: unbinding, where the resulting free repressor may quickly bind again.
1024: It turns out that degradation of bound repressors induces 
1025: bistability not only at the level of the master 
1026: equation but even at the level of rate equations.
1027: 
1028: \subsection{Rate Equations}
1029: 
1030: The rate equations that describe the BRD switch
1031: take the form
1032: 
1033: \begin{eqnarray}
1034: \dot{[A]} &=& g (1-[r_B]) - d [A]-
1035: \alpha_0 [A] \left (1-[r_A] \right) + \alpha_1[r_A]  \nonumber \\
1036: \dot{[B]} &=& g (1-[r_A])-d [B]-
1037: \alpha_0[B]\left(1-[r_B]\right)+\alpha_1[r_B] \nonumber \\
1038: \dot{[r_A]} &=& \alpha_0[A]\left(1-[r_A]\right)-\alpha_1[r_A] 
1039: - d_r [r_A] \nonumber \\
1040: \dot{[r_B]} &=& \alpha_0[B]\left(1-[r_B]\right)-\alpha_1[r_B] 
1041: - d_r [r_B],
1042: \label{eq:brd_switch_rate}
1043: \end{eqnarray}
1044: 
1045: \noindent
1046: where $d_r$ is the degradation rate of the bound repressors.
1047: Assuming quasi-steady state for the binding-unbinding
1048: processes we obtain
1049: the Michaelis-Menten equations
1050: 
1051: \begin{eqnarray}
1052: %\label{eq:Adeg}
1053: \dot{[A]}&=&{g \over {1+k[B]}} 
1054: - \left(d + {{d_rk}\over{1+k[A]}} \right) [A]  \nonumber \\
1055: %\label{eq:Bdeg}
1056: \dot{[B]}&=&{g \over {1+k[A]}}
1057: - \left( d + {{d_rk}\over{1+k[B]}} \right) [B],
1058: \label{eq:DBR_switch_michaelis_menten}
1059: \end{eqnarray}   
1060: 
1061: \noindent
1062: where now
1063: $k=\alpha_0/(\alpha_1+d_r)$. 
1064: Note that the coefficients of $[A]$
1065: and $[B]$
1066: in the second terms in 
1067: Eq.
1068: (\ref{eq:DBR_switch_michaelis_menten})
1069: can be considered as effective degradation rate 
1070: constants.
1071: 
1072: For steady state conditins,  
1073: Eq.
1074: (\ref{eq:DBR_switch_michaelis_menten})
1075: exhibits the
1076: symmetric solution
1077: 
1078: \begin{equation}
1079: [A] = [B] = \frac{[ (d+d_rk)^2 + 4dkg]^{1/2} - d - d_rk }{ 2dk}.
1080: \label{eq:symmBRD}
1081: \end{equation}
1082: 
1083: \noindent
1084: This solution exists for any choice of the parameters.
1085: In addition, in some parameter range, two non-symmetric solutions 
1086: exist. These solutions can be expressed as the solutions
1087: of the quadratic equation
1088: 
1089: \begin{equation}
1090: dd_r k^2 [A]^2 + (gdk+dd_rk+d_r^2k^2-gd_rk^2)[A]+gd=0.
1091: \label{eq:Asymmetrics}
1092: \end{equation}
1093: 
1094: \noindent
1095: The condition for the
1096: existence of two different solutions of this equation is
1097: 
1098: \begin{equation}
1099: (g-d_r)[g(kd_r-d)^2-d_r(kd_r+d)^2]>0.
1100: \label{eq:Asymmetrics2}
1101: \end{equation}
1102: 
1103: \noindent
1104: In order for them to be positive the 
1105: condition 
1106: $g>d_r$ 
1107: must be satisfied.
1108: Thus, the bifurcation takes place at
1109: 
1110: \begin{equation}
1111: k_c = \frac{d(\sqrt{g}+\sqrt{d_r})}{d_r(\sqrt{g}-\sqrt{d_r})},
1112: \label{eq:Asymmetrics3}
1113: \end{equation}
1114: 
1115: \noindent
1116: and the non-symmetric solutions exist for $k>k_c$. 
1117: Linear stability analysis shows that whenever the non-symmetric 
1118: solutions exist they are stable, while the symmetric solution
1119: is stable only for $k \le k_c$.
1120: 
1121: The steady state populations of free $A$ and $B$ repressors
1122: vs. $k$, for the BRD switch, are shown in 
1123: Fig. \ref{fig6}.
1124: The results of numerical integration of the rate equations
1125: ($\times$) are in perfect agreement with the analytical results
1126: derived above (solid line).
1127: We conclude that the degradation of bound repressors induces
1128: bistability, even at the level of rate equations.
1129: The emergence of bistability can be attributed to the 
1130: fact that the effective degradation rate for the minority
1131: specie in
1132: Eq. (\ref{eq:DBR_switch_michaelis_menten})
1133: is larger than the effective degradation rate 
1134: for the dominant specie. 
1135: This tends to
1136: enhance the difference between the population sizes
1137: and to destabilize the symmetric solution
1138: for $k>k_c$.
1139: 
1140: \subsection{Master Equation} 
1141: 
1142: The master equation for the BRD switch 
1143: can be obtained from
1144: Eq. (\ref{eq:general_switch_master})
1145: by adding the term
1146: 
1147: \begin{eqnarray}
1148: \label{eq:BRD_switch_master}
1149: && d_r [\delta_{r_A,0} P(N_A,N_B,r_A+1,r_B) 
1150: - \delta_{r_A,1} P(N_A,N_B,r_A,r_B)] + \nonumber\\
1151: && d_r [\delta_{r_B,0} P(N_A,N_B,r_A,r_B+1) 
1152: - \delta_{r_B,1} P(N_A,N_B,r_A,r_B)].  
1153: \end{eqnarray}
1154: 
1155: \noindent
1156: For steady state conditions we find that BRD tends to suppress 
1157: the peak near the origin of $P(N_A,N_B)$.
1158: For a suitable range of parameters, two separate peaks
1159: appear, which qualitatively resemble those obtained for the
1160: exclusive switch.
1161: However, unlike the exclusive switch, 
1162: there is a narrow corridor with small 
1163: but non-vanishing probabilities
1164: that connects the two peaks via the origin.
1165: As a result, the switching time for the BRD
1166: switch tend to be somewhat shorter than for the
1167: exclusive switch with the same parameters.
1168: The switching times, $\tau$, vs. the
1169: repression strength, $k$, 
1170: are shown in 
1171: Fig. \ref{fig7}.
1172: 
1173: We now examine in what range of parameters 
1174: this circuit is indeed a switch
1175: according to the master equation. 
1176: Unlike the rate equation where the condition
1177: for bistability is clear
1178: [Eq. (\ref{eq:Asymmetrics3})], 
1179: in the case of the master equation the
1180: notion of bistability is more subtle.
1181: Thus,
1182: in the analysis below we use the following operational criterion.
1183: First we define the two states of the switch.
1184: The $A$-dominated state is defined 
1185: as the set of all states in which
1186: $N_A>2$ and $N_B=0,1$.
1187: Similarly, the
1188: $B$-dominated state is defined by $N_B>2$ and $N_A=0,1$.
1189: The system is considered as a switch
1190: if, under steady state conditions, the total probability 
1191: to be in either of these states is larger than $0.99$.
1192: This leaves a probability of only 0.01 for all
1193: the intermediate states, 
1194: which the system must visit in order to switch between the
1195: $A$-dominated and the $B$-dominated states.
1196: As a result, the switching rate is low.
1197: 
1198: We used this criterion in order to find the region in the $(k,d_r)$ plane
1199: of the parameter space in which the BRD circuit exhibits bistability.
1200: %The results are given in 
1201: %Fig. \ref{fig}.
1202: It was found that the BRD switch exhibits bistability 
1203: for large enough values of $k$, 
1204: as long as the value of $d_r$ is not too different from $d$.
1205: If $d_r/d \ll 1$, the process of bound-repressor degradation is 
1206: negligible and cannot eliminate the dead-lock situation.
1207: If $d_r/d \gg 1$, proteins bind and quickly degrade.
1208: As a result, the population of the dominant specie is reduced
1209: and bistability is suppressed.
1210: 
1211: Within the parameter range in which the system exhibits bistability,
1212: we examined the dependence of the switching time 
1213: $\tau$ of the BRD switch 
1214: on the
1215: parameters 
1216: $g$, $d$, $\alpha_0$ and $d_r$. 
1217: %In 
1218: %Fig. \ref{fig:BRD_switch_tau}
1219: %we show  
1220: %the switching time 
1221: %$\tau$ vs. $g$, $d$ and $k$.
1222: It was found that $\tau$ exhibits linear dependence on
1223: the production rate $g$ and on the repression strength $k$
1224: (here, $k$ was varied by changing $\alpha_0$, keeping $\alpha_1$ 
1225: and $d_r$ fixed).
1226: The dependence of $\tau$ on the degradation rate $d$ was found to be 
1227: approximately $1/d^2$.
1228: Note that as 
1229: $d$ was veried, 
1230: we kept $d_r=d$ in order that the system remains bistable. 
1231: Since $k$ depends on $d_r$, 
1232: it slightly varied as well.
1233: 
1234: Unlike the exclusive switch,
1235: where we managed to obtain the scaling properties of 
1236: $\tau$ by a simple argument, 
1237: the BRD switch turns out to be more complicated. 
1238: This is due to the fact that
1239: there are several processes that may 
1240: lead to the flipping of the switch,
1241: such as the unbinding or the degradation of the bound repressor.
1242: A further complication is that the
1243: two repressors can be bound simultaneously.
1244: As a result, we have not managed to obtain
1245: an expression for $\tau$ in the BRD switch.
1246: 
1247: \section{The Switch with Protein-Protein Interaction}
1248: \label{sec:PPI}
1249: 
1250: Consider a switch circuit which 
1251: in addition to the mutual repression,
1252: exhibits 
1253: protein-protein interactions (PPI), namely
1254: an A protein and a B protein may form a complex, AB. 
1255: The AB complex is not active as a transcription factor. 
1256: 
1257: \subsection{Rate Equations}
1258: 
1259: The PPI switch can be described by the following rate equations
1260: 
1261: \begin{eqnarray}
1262: \dot{[A]} &=& g (1-[r_B]) - d [A]-
1263: \alpha_0 [A] \left (1-[r_A] \right) + \alpha_1[r_A] -\gamma AB \nonumber \\
1264: \dot{[B]} &=& g (1-[r_A])-d [B]-
1265: \alpha_0[B]\left(1-[r_B]\right)+\alpha_1[r_B] -\gamma AB \nonumber \\
1266: \dot{[r_A]} &=& \alpha_0[A]\left(1-[r_A]\right)-\alpha_1[r_A] \nonumber \\
1267: \dot{[r_B]} &=& \alpha_0[B]\left(1-[r_B]\right)-\alpha_1[r_B].
1268: \label{eq:PPI_switch_rate}
1269: \end{eqnarray}
1270: 
1271: \noindent
1272: The parameter $\gamma$ is the rate constant for the binding of
1273: a pair of $A$ and $B$ proteins.
1274: The Michaelis-Menten equations take the form
1275: 
1276: \begin{eqnarray}    
1277: \dot{[A]} &=& \frac{g}{1+k[B]} - d [A] -\gamma [A][B]\nonumber \\
1278: \dot{[B]} &=& \frac{g}{1+k[A]} - d [B] -\gamma [A][B].
1279: \label{eq:PPI_switch_michaelis_menten}
1280: \end{eqnarray}    
1281: 
1282: \noindent
1283: For steady state conditions,
1284: these equations exhibit a symmetric solution, $[A]=[B]$,
1285: for any choice of the parameters.
1286: It is the solution of 
1287: 
1288: \begin{equation}
1289: \gamma k [A]^3 + (\gamma+dk)[A]^2 + d[A] - g = 0.
1290: \end{equation}
1291: 
1292: \noindent
1293: Since all the coefficients of powers of [A] are positive,
1294: this equation has only one positive solution. 
1295: Also, within some range of parameters 
1296: there exist non-symmetric solutions, given by 
1297: the solutions of 
1298: 
1299: \begin{equation}
1300: d\gamma k[A]^2 + (d\gamma +d^2 k - g\gamma k)[A]+d^2 = 0.
1301: \label{eq:PPnonsym}
1302: \end{equation}
1303: 
1304: \noindent
1305: The non-symmetric solutions exist only for the range of parameters 
1306: in which 
1307: Eq. (\ref{eq:PPnonsym}) 
1308: has two positive solutions. 
1309: The condition for this can be easily expressed in terms of the coefficients
1310: in 
1311: Eq. (\ref{eq:PPnonsym}). 
1312: 
1313: As in the case of the BRD switch, 
1314: bistability is observed even
1315: at the level of rate equations. 
1316: Again, the emergence of bistability can be attributed to the 
1317: fact that the effective degradation rate constant for the minority
1318: specie is larger than for the dominant specie, thus
1319: enhancing the difference between the population sizes 
1320: [note that the effective degradation rate constant for 
1321: $A$ is $(d+\gamma [B])$,
1322: while for $B$ it is $(d+\gamma [A])$].
1323: 
1324: \subsection{Master Equation}
1325: 
1326: The master equation for the PPI switch 
1327: can be obtained from 
1328: Eq. (\ref{eq:general_switch_master})
1329: by adding the term
1330: 
1331: \begin{equation}
1332: \gamma [(N_A+1)(N_B+1) P(N_A+1,N_B+1,r_A,r_B) 
1333: - N_A N_B P(N_A,N_B,r_A,r_B)]. 
1334: \label{eq:PPI_switch_master}
1335: \end{equation}
1336: 
1337: \noindent
1338: For a suitable range of parameters 
1339: the steady state solution of the master 
1340: equation exhibits two separate peaks.
1341: To draw the range of parameters in 
1342: which bistability takes place we apply the
1343: operational criterion used above for the BRD switch. 
1344: We fix 
1345: $g$,
1346: $d$ 
1347: and 
1348: $\alpha_1$ 
1349: and examined the system in the
1350: $(k,\gamma)$ plane. 
1351: The results are plotted in 
1352: Fig. \ref{fig8} (solid line).
1353: 
1354: For small values of $\gamma$ 
1355: (weak PP interaction), 
1356: the circuit
1357: does not exhibit bistability. 
1358: As the interaction strength increases
1359: the circuit behaves as a switch for a certain range of repression 
1360: strength $k$. 
1361: This range broadens as $\gamma$ is increased.
1362: Unlike the switch systems discussed above,
1363: in which the bistability gets stronger as $k$ is increased,
1364: the PPI switch is bistable
1365: for intermediate values of $k$.
1366: This can be understood as follows.
1367: Recall that the key to the formation of a switch is the
1368: elimination of the dead-lock situation. 
1369: The exclusive and the
1370: BRD switches deal with this situation directly at the bound 
1371: repressor level. 
1372: However the PP interaction does not directly 
1373: affect the bound repressor.
1374: To prevent the possibility of two proteins bound 
1375: simultaneously, one of them should unbind 
1376: and form a complex with a protein of the other specie.
1377: In order for this to happen, the repressors must not
1378: be bound too strongly.
1379: Therefore, the PPI switch works 
1380: at intermediate repression strength. 
1381: As the PPI becomes more effective
1382: (larger $\gamma$) this mechanism applies
1383: at larger values of $k$.
1384: 
1385: Enhanced switching properties can be obtained by considering
1386: a hybrid system that combines
1387: PPI and exclusive binding.
1388: The resulting switch exhibits bistability
1389: in a broader
1390: range of parameters than the exclusive or PPI switches alone.
1391: The 
1392: master equation for the
1393: exclusive-PPI switch is 
1394: obtained from
1395: Eq. (\ref{eq:exclusive_switch_master})
1396: by adding the term 
1397: 
1398: \begin{equation}
1399: \gamma [(N_A+1)(N_B+1) P(N_A+1,N_B+1,r_A,r_B) 
1400: - N_A N_B P(N_A,N_B,r_A,r_B)],
1401: \end{equation}
1402: 
1403: \noindent
1404: which accounts for the PP interaction. 
1405: Numerical results, shown in 
1406: Fig.~\ref{fig8},
1407: indicate that indeed as expected the exclusive-PPI switch is a
1408: better switch than either the PPI or the 
1409: exclusive switch.
1410: The parameter range in which it 
1411: exhibits bistability is broader.
1412: Thus, it is more
1413: robust to variations in the parameters 
1414: than the exclusive or PPI switches.
1415: 
1416: \section{Cooperative Binding}
1417: \label{sec:cooperative}
1418: 
1419: Cooperative binding is found in genetic switch
1420: systems such as the phage $\lambda$ switch
1421: \cite{Ptashne1992}.
1422: In this case, transcription regulation is obtained
1423: only when several copies of the repressor are bound
1424: simultaneously.
1425: This situation can be achieved in two ways.
1426: One possibility is that repressors bind to each other and
1427: form a complex, which then binds to the promoter.
1428: The other possibility is that the repressors bind separately,
1429: but those already bound assist the other ones to bind more
1430: effectively.
1431: In the case of cooperative binding, bistability turns out
1432: to appear even at the level of rate equations
1433: \cite{Cherry2000}.
1434: 
1435: \subsection{Rate Equations}
1436: 
1437: Switch systems with cooperative binding are commonly 
1438: described by
1439: 
1440: \begin{eqnarray}    
1441: \dot{[A]} &=& \frac{g}{1+k[B]^n} - d [A] \nonumber \\
1442: \dot{[B]} &=& \frac{g}{1+k[A]^n} - d [B],
1443: \label{eq:general_switch_hilln}
1444: \end{eqnarray}    
1445:             
1446: \noindent
1447: where $n$ is the Hill coefficient.
1448: It corresponds to the number of copies of the transcription
1449: factor which are required in order to perform the repression
1450: process. Here we focus on the case $n=2$, 
1451: and show that these equations exhibit two stable steady state
1452: solutions 
1453: for some range of parameters.
1454: Imposing 
1455: $\dot{[A]}=\dot{[B]}=0$ 
1456: in Eq. (\ref{eq:general_switch_hilln}),
1457: we obtain 
1458: 
1459: \begin{eqnarray}    
1460: g - d[A] - kd[A][B]^2 &=& 0 \nonumber \\
1461: g - d[B] - kd[B][A]^2 &=& 0.
1462: \label{eq:general_switch_hilln1}
1463: \end{eqnarray}    
1464: 
1465: \noindent
1466: Subtracting one of these equations from the other we 
1467: find that
1468: 
1469: \begin{equation}
1470: -d([A]-[B])-kd[A][B]([B]-[A])=0.
1471: \label{eq:AmB}
1472: \end{equation}
1473: 
1474: \noindent
1475: Looking for a non-symmetric solution for 
1476: which $[A] \neq [B]$,
1477: we divide 
1478: Eq.
1479: (\ref{eq:AmB})
1480: by
1481: $[A]-[B]$.
1482: We 
1483: find that 
1484: $k[A][B]=1$, 
1485: or $[B]=1/k[A]$.
1486: Inserting this into 
1487: Eq. (\ref{eq:general_switch_hilln1}) 
1488: we get an equation for $[A]$:
1489: 
1490: \begin{equation}
1491: dk[A]^2-gk[A]+d=0.
1492: \end{equation}
1493: 
1494: \noindent
1495: This equation exhibits two distinct stable solutions 
1496: 
1497: \begin{equation}
1498: [A]=\frac{gk \pm \sqrt{g^2 k^2-4 d^2 k}}{2dk},
1499: \end{equation}
1500: 
1501: \noindent
1502: for
1503: $k>4d^2/g^2$.
1504: This means that the system becomes bistable
1505: at the bifurcation point, $k=4d^2/g^2$. 
1506: In addition to these solutions, 
1507: the symmetric solution 
1508: $[A]=[B]$ 
1509: exists for any choice of the parameters.
1510: This
1511: symmetric solution is stable for
1512: $k<4d^2/g^2$ and becomes unstable at the bifurcation point. 
1513: 
1514: \subsection{Monte Carlo Simulations}
1515: 
1516: Consider a switch system with cooperative binding with
1517: $n=2$,
1518: in which two proteins of the
1519: same specie bind together to form a complex or dimer.
1520: The repression of $A$ synthesis is done by dimers composed 
1521: of two $B$ proteins and vice versa. 
1522: For example, consider an exclusive switch, in which
1523: the dimers of $A$ and $B$ cannot be bound simultaneously.
1524: To account for stochastic effects, we have studied this
1525: system using Monte Carlo simulations.
1526: 
1527: The rate constant for the formation of dimers is
1528: denoted by $\gamma_D$. 
1529: It is assumed that dimers cannot dissociate 
1530: into single proteins, but they can degrade.
1531: The degradation rate of
1532: dimers is denoted by $d_D$.
1533: We examined the dependence of the
1534: switching time $\tau$ on all the parameters. 
1535: We found the following properties.
1536: The dependence of $\tau$ on $g$ was found to be 
1537: linear as for the exclusive and BRD switch.
1538: The dependence on $d$ is very weak, 
1539: except for the limit in which $d$ is very large.
1540: This is because the proteins 
1541: tend to form dimers before they have a chance to degrade.
1542: The dependence of $\tau$ on 
1543: $k=\alpha_0/\alpha_1$ 
1544: is found to be well fitted by a quadratic polynomial.
1545: This means that for sufficiently strong repression, 
1546: $\tau \sim k^2$.
1547: 
1548: The dependence of $\tau$ on the dimerization rate 
1549: $\gamma_D$ 
1550: [Fig. \ref{fig9}(a)]
1551: exhibits interesting behavior. 
1552: For small values of $\gamma_D$ the system 
1553: is not really a switch, because
1554: almost no dimers are formed. 
1555: Therefore the switching time is short. 
1556: For larger values of $\gamma_D$ 
1557: the dimer population increases and 
1558: the system starts to function as a
1559: switch.
1560: The switching time
1561: $\tau$ increases as the switch
1562: becomes more stable. 
1563: But from some point, 
1564: increasing $\gamma_D$ causes $\tau$ to
1565: decrease. 
1566: This is because, very fast dimerization 
1567: helps the minority
1568: specie to form dimers, 
1569: making it more likely to flip the switch.
1570: 
1571: The dependence of $\tau$ on 
1572: $d_D$ 
1573: [Fig. \ref{fig9}(b)]
1574: was found to be well fitted by a 
1575: cubic polynomial in $1/d_D$.
1576: This means that in the limit of slowly degrading dimers, 
1577: $\tau \sim 1/d_D^3$. 
1578: In the limit of fast dimer-degradation
1579: the system is not bistable, 
1580: because the population of dimers is too small
1581: to make the repression effective.
1582: 
1583: The switching time for this system was also
1584: studied in Ref.
1585: \cite{Warren2004}, 
1586: where
1587: $\tau$ was presented as a function of 
1588: the average copy number of the dominant specie. 
1589: However, the copy number depends 
1590: in a non-trivial way on the parameters
1591: and cannot be directly controlled.
1592: Therefore, we believe that 
1593: in a systematic study of the switching times,
1594: it is more practical to examine
1595: the dependence of $\tau$ on the parameters themselves.
1596: 
1597: Note that there is another important realization of cooperative
1598: binding in which the promoter consists of two binding sites.
1599: When a protein binds to one of them it facilitates the binding
1600: of another protein to the second site. 
1601: The effect of this mechanism is qualitatively similar to the 
1602: one shown above for dimers.
1603: In general cooperative binding induces bistability becuase it
1604: forces the minority specie to recruit at least two proteins
1605: in order to flip the switch.
1606: As a result, cooperative binding helps to remove the dead-lock situation
1607: in which both species are suppressed simultaneously.
1608: 
1609: \section{Response to External Signals}
1610: \label{sec:response}
1611: 
1612: Until now our discussion considered only 
1613: spontaneous transitions between 
1614: the two states of the switch. 
1615: Here we demonstrate how an
1616: external signal may lead to the flipping of the switch.
1617: In case of the $\lambda$ switch,
1618: such an external signal may be, for example, the 
1619: exposure of {\it E. coli} 
1620: infected by phage $\lambda$ to UV light. 
1621: In the {\it lac} circuit, the external signal indicates 
1622: the presence of lactose.
1623: We assume that the effect of the external 
1624: signal is that one of the proteins
1625: undergoes a conformal change 
1626: that prevents its binding
1627: to the promoter.
1628: When the signal affects the dominant specie,
1629: this may lead to the 
1630: flipping of the switch.
1631: We assume that the conformal change is 
1632: fast and that it lasts for a period
1633: of time determined by the duration of the 
1634: external signal. 
1635: 
1636: We have performed Monte Carlo simulations, 
1637: where the binding rate $\alpha_0$ of the
1638: dominant specie was set to zero for 
1639: some period of time (the length of external signal).
1640: We calculated the probability for a 
1641: flipping of the switch during $1800$ (s), which is roughly the
1642: time between divisions of {\it E. coli},
1643: as a function of the signal length. 
1644: The results are
1645: shown in Fig.~\ref{fig10}.
1646: 
1647: For short duration of the signal, the switch 
1648: has a small chance to flip. As
1649: the duration increases the probability to flip 
1650: increases too, and so for a long 
1651: enough signal, the switch will eventually flip as expected 
1652: (the actual switching time depends on 
1653: the parameters of the switch, like
1654: the production rate $g$ or the unbinding rate $\alpha_1$. 
1655: Here we just demonstrated
1656: that in principle the switch will flip states 
1657: in response to an external signal).
1658: 
1659: \section{Discussion}
1660: \label{sec:discussion}
1661: 
1662: In the rate equations, the meaning of bistability is clear.
1663: It typically appears as a result of a bifurcation. 
1664: Below the bifurcation there is a single, stable solution, 
1665: which becomes unstable at the bifurcation point, where two
1666: stable solutions emerge.
1667: In case of the toggle switch, one of these solutions is
1668: dominated by $A$ proteins and the other is dominated by $B$ proteins.
1669: Since both solutions are stable, the possibility of spontaneous 
1670: transitions between them due to stochastic fluctuations is not
1671: included in the rate equation model.
1672: 
1673: The objects that participate in regulatory processes in cells,
1674: namely genes, mRNAs, proteins and promoter sites are discrete
1675: objects, and some of them often appear in low copy numbers. 
1676: This, together with the fact that many of the relevant processes 
1677: such as diffusion, degradation as well as binding and unbinding 
1678: of transcription factors are of stochastic nature, requires to
1679: consider the role of stochastic fluctuations in these regulatory
1680: processes. This can be done by using the master equation or Monte
1681: Carlo simulations. 
1682: 
1683: In the master equation, bistability is characterized by two 
1684: separate peaks in the probability distribution. These peaks
1685: should be sufficiently far from each other, with low probabilities
1686: in the domain between them. As a result, the flow of probability
1687: between the two peaks is low and the time between spontaneous
1688: switching events is long. In order to qualify as a switch,
1689: the average time between spontaneous switching events must 
1690: be much longer than the time constants of the transcription,
1691: translation and degradation processes in the cell.
1692: 
1693: For the systems studied here it was found that the general 
1694: switch without cooperative binding does not exhibit bistability
1695: bistability either with the rate equations or with the master
1696: equation.
1697: Two other variants, the BRD and the PPI switch systems,
1698: were found to exhibit bistability both with the rate
1699: equations and with the master equation.
1700: However, the exclusive switch,
1701: which is not bistable at the rate equation level,
1702: was found to exhibit bistability with the master equation.
1703: Thus, in case of the exclusive switch it is clear that
1704: stochastic fluctuations play a crucial role in making
1705: the system bistable. For this system we also found an exact
1706: phenomenological expression for the switching time in terms
1707: of the rate constants of the relevant processes.
1708: 
1709: Stochastic analysis of genetic networks can be done 
1710: either by direct integration of the master equation
1711: or by Monte Carlo simulations.
1712: The master equation provides the probability distribution
1713: of the population sizes of all the mRNA's and proteins
1714: in the simulated circuit.
1715: It can be considered as a distribution over a large 
1716: number of genetically identical cells. 
1717: The average population sizes and the rates of 
1718: processes are expressed in terms of moments of this
1719: distribution. 
1720: To obtain such distributions from Monte Carlo simulations,
1721: one needs to repeat the simulations a large number of times
1722: and average over them.
1723: This may be inefficient in terms of computer time, and the
1724: statistical errors may be significant.
1725: On the other hand, unlike the master equation,
1726: Monte Carlo simulations enable to follow
1727: the time evolution of a single cell and directly evaluate
1728: quantities such as switching times and oscillation periods.
1729: 
1730: The number of equations in the master equation set increases
1731: exponentially with the number of proteins and mRNAs 
1732: included in the simulated circuit.
1733: As a result, the master equation becomes infeasible for 
1734: complex networks.
1735: Recently, we have shown that for reaction networks described
1736: by sparse graphs, one can use suitable approximations and
1737: dramatically reduce the number of equations
1738: \cite{Lipshtat2004}.
1739: 
1740: A related circuit,
1741: the mixed feedback loop, 
1742: in which $A$ is a repressor to $B$
1743: and the $A$ and $B$ proteins bind to form a complex
1744: was recently studied using rate equations
1745: \cite{Francois2004,Francois2005}.
1746: It was found to exhibit bistability 
1747: within a range of parameters.
1748: 
1749: \section{Summary}
1750: \label{sec:summary}
1751: 
1752: Genetic switch systems with mutual repression of two
1753: transcription factors, 
1754: have been studied using a combination of
1755: deterministic and stochastic methods.
1756: These system exhibit bistability, namely
1757: two stable states such that 
1758: spontaneous transitions between them are rare. 
1759: Induced transitions take place as a
1760: result of an external stimulus.
1761: We have studied several variants of the genetic
1762: switch, which exhibit
1763: cooperative binding, 
1764: exclusive binding, 
1765: protein-protein interactions 
1766: and degradation of bound repressors.
1767: For each variant we examined the range of parameters
1768: in which bistability takes place.
1769: Numerous studies have concluded 
1770: that cooperative binding is 
1771: a necessary condition for the emergence of bistability
1772: in these systems.
1773: We have shown that a suitable combination of network structure and
1774: stochastic effects gives rise to bistability even
1775: without cooperative binding. 
1776: The average time $\tau$ between spontaneous transitions was evaluated
1777: as a function of the biological parameters.
1778: 
1779: \newpage
1780: \clearpage
1781: 
1782: %\bibliographystyle{prsty}
1783: %\bibliography{genetics}
1784: 
1785: \begin{thebibliography}{10}
1786: 
1787: \bibitem{Elowitz2002}
1788: {M.B. Elowitz, A.J. Levine, E.D. Siggia and P.S. Swain}, Science {\bf 297},
1789:   1183  (2002).
1790: 
1791: \bibitem{Ozbudak2002}
1792: {E.M. Ozbudak, M. Thattai, I. Kurtser, A.D. Grossman and A. van Oudenaarden},
1793:   Nature Genetics {\bf 31},  69  (2002).
1794: 
1795: \bibitem{Mcadams1997}
1796: {H.H. McAdams and A. Arkin}, Proc. Natl. Acad. Sci. US {\bf 94},  814  (1997).
1797: 
1798: \bibitem{Becskei2000}
1799: {A. Becskei and L. Serrano}, Nature {\bf 405},  590  (2000).
1800: 
1801: \bibitem{Kaern2005}
1802: {M. Kaern, T.C. Elston, W.J. Blake and J.J. Collins}, Nature Reviews Genetics
1803:   {\bf 6},  451  (2005).
1804: 
1805: \bibitem{Vilar2002}
1806: {J.M. Vilar, H.Y. Kueh, N. Barkai and S. Leibler}, Proc. Natl. Acad. Sci. US
1807:   {\bf 99},  5988  (2002).
1808: 
1809: \bibitem{Shnerb2000}
1810: {N.M. Shnerb, Y. Louzoun, E. Bettelheim and S. Solomon}, Proc. Natl. Acad. Sci.
1811:   US {\bf 97},  10322  (2000).
1812: 
1813: \bibitem{Atkinson2003}
1814: {M.R. Atkinson, M.A. Savageau, J.T. Myers and A.J. Ninfa}, {Cell} {\bf 113},
1815:   597  (2003).
1816: 
1817: \bibitem{Ptashne1992}
1818: {M. Ptashne}, {\em {A Genetic Switch: Phage $\lambda$ and Higher Organisms, 2nd
1819:   edition.}} ({Cell Press and Blackwell Scientific Publications}, {Cambridge,
1820:   MA}, 1992).
1821: 
1822: \bibitem{Lewis2005}
1823: {M. Lewis}, {Comptes Rendus Biologies} {\bf 328},  521  (2005).
1824: 
1825: \bibitem{Gardner2000}
1826: {T.S. Gardner, C.R. Cantor and J.J. Collins}, Nature {\bf 403},  339  (2000).
1827: 
1828: \bibitem{Laslo2006}
1829: {P. Laslo et al.}, Cell {\bf 126}, 755 (2006).
1830: 
1831: \bibitem{Hill}
1832: Hill-function models are simplifications of rate-law equations.
1833: When derived directly from rate laws, 
1834: $n$ is expected to take only
1835: integer values,
1836: which represents the number of transcription
1837: factors required to perform the regulation. 
1838: However, when these models are used for fitting 
1839: empirical data, $n$ is a fitting parameter which may take 
1840: non-integer values.
1841: 
1842: \bibitem{Warren2004}
1843: {P.B. Warren and P.R. ten Wolde}, Phys. Rev. Lett. {\bf 92},  128101  (2004).
1844: 
1845: \bibitem{Warren2005}
1846: {P.B. Warren and P.R. ten Wolde}, J. Phys. Chem. B {\bf 109},  6812  (2005).
1847: 
1848: \bibitem{Walczak2005}
1849: {A. M. Walczak, M. Sasai, and P. Wolynes}, Biophysical Journal {\bf 88},  828
1850:   (2005).
1851: 
1852: \bibitem{Cherry2000}
1853: {J.L. Cherry and F.R. Adler}, J. Theor. Biol. {\bf 203},  117  (2000).
1854: 
1855: \bibitem{mRNA}
1856: Extended circuits that include the mRNA level have also been
1857: studied. It was found that 
1858: there is
1859: a perfect agreement between the 
1860: results of the REs for the simplified and
1861: the extended circuits,
1862: and excellent agreement with only a slight difference in peak
1863: shapes in the ME.
1864: 
1865: \bibitem{Lipshtat2005}
1866: {A. Lipshtat, H.B. Perets, N.Q. Balaban and O. Biham}, Gene {\bf 347},  265
1867:   (2005).
1868: 
1869: \bibitem{Lipshtat2006}
1870: {A. Lipshtat, A. Loinger, N.Q. Balaban and O. Biham}, Phys. Rev. Lett. {\bf
1871:   96},  188101  (2006).
1872: 
1873: \bibitem{Kepler2001}
1874: {T.B. Kepler and T.C. Elston}, Biophysical Journal {\bf 81},  3116  (2001).
1875: 
1876: \bibitem{Paulsson2000}
1877: {J. Paulsson and M. Ehrenberg}, Phys. Rev. Lett. {\bf 84},  5447  (2000).
1878: 
1879: \bibitem{Paulsson2004}
1880: {J. Paulsson}, Nature {\bf 427},  415  (2004).
1881: 
1882: \bibitem{Arkin1998}
1883: {A. Arkin et al.}, Genetics {\bf 149},  1633  (1998).
1884: 
1885: \bibitem{VanKampen1992}
1886: {N.G. Van Kampen}, {\em {Stochastic processes in physics and chemistry}}
1887:   ({North-Holland}, {}, 1992).
1888: 
1889: 
1890: \bibitem{TauMEMC}
1891: Results of $\tau$ calculated in this procedure, 
1892: agree with results 
1893: obtained from averaging a large number of Monte Carlo simulations, 
1894: up to a numerical factor of 2. This factor rises from the fact the in
1895: the procedure we described the system can also return from the $B$ dominated
1896: peak to the $A$ dominated peak.
1897: 
1898: \bibitem{OnlyK2006}
1899: {We assume that the switching time is a function only of $k$, the ratio
1900:   between $\alpha_0$ and $\alpha_1$, and not of $\alpha_0$ and $\alpha_1$
1901:   separately. This is indeed correct in the case of the exclusive switch,
1902:   because we can always choose the time unit as desired and in such a way
1903:   eliminate one of the variables. This assumption may fail in other cases,
1904:   where $k$ is not simply $\alpha_0 / \alpha_1$. For convenience, in this work,
1905:   most of the time $\alpha_1$ was held fixed at a value of $0.01$, $\alpha_0$
1906:   varied, and the results were plotted as a function of $k$.}
1907: 
1908: \bibitem{Lipshtat2004}
1909: A. Lipshtat and O. Biham,
1910: Phys. Rev. Lett. {\bf 93}, 170601 (2004).
1911: 
1912: \bibitem{Francois2004}
1913: P. Francois and V. Hakim, 
1914: Proc. Nat. Acad. Sci. {\bf 101}, 580 (2004)
1915: 
1916: \bibitem{Francois2005}
1917: P. Francois and V. Hakim, 
1918: Phys. Rev. E {\bf 72}, 031908 (2005) 
1919: 
1920: \end{thebibliography}
1921: 
1922: \newpage
1923: \clearpage
1924: 
1925: \begin{figure}
1926: \caption{Schematic illustrations of (a) the general switch circuit,
1927: that includes two transcription factors, $A$ and $B$, 
1928: which negatively regulate each other's synthesis; 
1929: (b) the exclusive switch,  
1930: in which there is an overlap between the 
1931: promoter sites of $A$ and $B$ proteins, 
1932: so they cannot be bound simultaneously.
1933: }
1934: \label{fig1}
1935: \end{figure}
1936: 
1937: \begin{figure}
1938: \caption{
1939: (Color online)
1940: The probability distribution
1941: $P(N_A,N_B)$ 
1942: for the general switch, 
1943: under conditions of 
1944: (a) weak repression ($k=0.005$) 
1945: where there is one symmetric peak;
1946: and 
1947: (b) strong repression ($k=50$)
1948: where three peaks appear, one dominated by $A$, the
1949: second dominated by $B$ and the third in which 
1950: both species are mutually suppressed. 
1951: The weights of the three peaks are about the same.
1952: }
1953: \label{fig2}
1954: \end{figure}
1955: 
1956: \begin{figure}
1957: \caption{
1958: (Color online)
1959: The population sizes of free $A$ and $B$ 
1960: proteins vs. time 
1961: obtained from a Monte Carlo simulation
1962: (a) for the general switch, where
1963: the system exhibits fast transitions between 
1964: its three states; 
1965: (b) for the exclusive switch
1966: The bistable behavior is clearly observed, where the population size 
1967: of the dominant specie is between $20-60$ and the other specie is nearly
1968: diminished. Failed switching attempts are clearly seen.
1969: The typical switching time is in the order of $10^5$ (s) or roughly 1 day.
1970: Bound proteins are also shown. Their fast binding and unbinding events
1971: cannot be resolved on the time scale that is presented.
1972: In both cases,
1973: $g=0.2$,
1974: $d=0.005$,
1975: $\alpha_0=0.2$ 
1976: and
1977: $\alpha_1 = 0.01$ (s$^{-1}$). 
1978: }
1979: \label{fig3}
1980: \end{figure}
1981: 
1982: \begin{figure}
1983: \caption{
1984: (Color online)
1985: The probability distribution $P(N_A,N_B)$ 
1986: for the exclusive switch, 
1987: under conditions of 
1988: (a) weak repression ($k=0.005$) where there is one symmetric peak 
1989: (b) intermediate repression ($k=1$) where two distinct peaks begin to emerge 
1990: but are still connected, and
1991: (c) strong repression ($k=50$), where bistability is observed.
1992: }
1993: \label{fig4}
1994: \end{figure}
1995: 
1996: \begin{figure}
1997: \caption{
1998: (Color online)
1999: Scaling properties of the switching time $\tau$
2000: for the exclusive switch
2001: vs. 
2002: the protein synthesis rate $g$,
2003: the degradation rate $d$
2004: and the repression strength $k$.
2005: }
2006: \label{fig5}
2007: \end{figure}
2008: 
2009: \begin{figure}
2010: \caption{
2011: Population sizes of the free $A$ and $B$ proteins vs. $k$
2012: for the BRD switch obtained from the rate equations.
2013: The parameters are 
2014: $g=0.05$,
2015: $d=d_r=0.005$,
2016: $\alpha_1=0.01$
2017: and
2018: $\alpha_0$ 
2019: is varied.
2020: Here $k_c \approx 1.92$.
2021: Stable solutions are shown by solid lines and unstable solutions
2022: by dashed lines.
2023: }
2024: \label{fig6}
2025: \end{figure}
2026: 
2027: 
2028: \begin{figure}
2029: \caption{
2030: (Color online)
2031: The switching time $\tau$ vs. $k$ for the 
2032: exclusive ($\times$),
2033: BRD ($\circ$)
2034: and PPI ($\triangle$)
2035: switch systems.
2036: The parameters used are
2037: $g=0.05$, $d=d_r=0.005$ and $\gamma=0.1$ (s$^{-1}$).
2038: }
2039: \label{fig7}
2040: \end{figure}
2041: 
2042: 
2043: \begin{figure}
2044: \caption{
2045: The range of parameters in the ($\gamma,k$) plane in which bistability
2046: takes place in the PPI switch (solid line) and in the
2047: exclusive-PPI switch (dashed line), using rate equations
2048: (a) and using the master equation (b).
2049: The other parameters are 
2050: $g=0.05$ and $d=0.005$ (s$^{-1}$).
2051: }
2052: \label{fig8}
2053: \end{figure}
2054: 
2055: \begin{figure}
2056: \caption{
2057: (Color online)
2058: The dependence of the switching time 
2059: $\tau$ for the dimers exclusive switch
2060: on the 
2061: dimers degradation rate $d_D$ (a)
2062: and the dimerization rate $\gamma_D$
2063: (b).
2064: }
2065: \label{fig9}
2066: \end{figure}
2067: 
2068: \begin{figure}
2069: \caption{
2070: Probability for the exclusive switch to flip during 1800 (s)
2071: after the initiation of the signal,
2072: as a function of the external signal duration. The parameters used were
2073: $g=0.2$,
2074: $d=0.005$,
2075: $\alpha_1=0.01$ 
2076: and 
2077: $\alpha_0=0.2$ 
2078: or zero during the signal.
2079: }
2080: \label{fig10}
2081: \end{figure}
2082: 
2083: \end{document}
2084: 
2085: