q-bio0703030/paper.tex
1: %&latex
2: %**********************************************************************
3: %   Sample paper (in LaTeX2e) for Banach Center Publ.   Mar. 7, 1999.
4: %**********************************************************************
5: \documentclass{bcp}
6: \def\LaTeX{L\kern -.36em\raise .3ex\hbox{\sc a}\kern -.15em T\kern -.1667em%
7: \lower .7ex\hbox{E}\kern -.125em X}
8: 
9: \font\tt=cmtt10
10: 
11: \usepackage{subeqnarray}
12: \usepackage{graphicx}
13: 
14: \begin{document}
15: 
16: %\keywords{These are optional}
17: %\mathclass{Primary 46C20; Secondary 32G81.}
18: \thanks{This work was supported by the Center for Complexity Studies}
19: \abbrevauthors{O. Biham, N.Q. Balaban, A. Loinger, A. Lipshtat and H.B. Perets}
20: \abbrevtitle{Simulations of simple genetic circuits}
21: 
22: \title{Deterministic and Stochastic Simulations \\
23: of Simple Genetic Circuits}
24: 
25: \author{Ofer Biham, Nathalie Q. Balaban and Adiel Loinger}
26: \address{Racah Institute of Physics, The Hebrew University, Jerusalem 91904, 
27: Israel}
28: 
29: \author{Azi Lipshtat}
30: \address{Department of Pharmacology and Biological Chemistry,
31: Mount Sinai \\ School of Medicine,
32: New York, NY 10029, USA
33: }
34: 
35: \author{Hagai B. Perets}
36: \address{Faculty of Physics,
37: Weizmann Institute of Science, Rehovot 76100, Israel
38: }
39: 
40: \maketitlebcp
41: 
42: \abstract{
43: 
44: We analyze three simple genetic circuits which involve transcriptional
45: regulation and feedback: the autorepressor, the switch and the
46: repressilator, that consist of one, two and three genes, respectively.
47: Such systems are commonly simulated using rate equations, that
48: account for the concentrations of the mRNAs and proteins produced
49: by these genes. 
50: Rate equations are suitable when the concentrations
51: of the relevant molecules in a cell are large and 
52: fluctuations are negligible.
53: However, when some of the proteins in the circuit
54: appear in low copy numbers, fluctuations become important
55: and the rate equations fail. 
56: In this case stochastic methods, such as 
57: direct numerical integration of
58: the master equation or
59: Monte Carlo simulations are required.
60: Here we present deterministic and stochastic simulations
61: of the autorepressor, the switch and the repressilator.
62: We show that fluctuations give rise to quantitative
63: and qualitative changes in the dynamics of these systems.
64: In particular, we demonstrate a fluctuations-induced bistability
65: in a variant of the genetic switch and and noisy oscillations
66: obtained in the repressilator circuit.
67: }
68: 
69: %-----------------------------------------------------
70: \section{Introduction.} 
71: 
72: The production of proteins in cells is regulated 
73: by networks of interacting genes.
74: Some of these genes code for transcription factors,
75: which are proteins that regulate the expression of
76: other genes by binding to specific promoter sites
77: on the DNA.
78: Some of these proteins, called repressors, perform
79: negative regulation, while others, called activators,
80: perform positive regulation.
81: Post-trancriptional regulation can occur by 
82: translational regulation, or by post-translational 
83: regulation such as protein-protein interaction, 
84: which may modify the function of these proteins. 
85: These networks of interacting genes 
86: are at the heart of all processes in cells: 
87: from the regulation of the cell cycle 
88: to the various stress responses.
89: 
90: It turns out that genetic networks are typically sparse networks,
91: namely most genes interact only with a small number of other genes. 
92: Also, the networks exhibit some degree of modularity, namely
93: one can identify recurring modules, which
94: are known as network motifs
95: \cite{Milo2002}.
96: 
97: Modules found in nature are often hard to study and fully control. 
98: To overcome these limitations, synthetic networks can be constructed 
99: from well-studied components. 
100: Examples of such components are the 
101: lactose, the lambda and the tetracycline systems in bacteria.
102: The synthetic networks are designed to 
103: perform desired functions, 
104: determined by their architecture.
105: They do not require the manipulation of the structure of
106: proteins and other regulatory elements at the molecular
107: level.
108: The genes and promoters are often inserted into plasmids
109: rather than on the chromosome.
110: Two important examples of synthetic circuits 
111: are the genetic toggle switch and the repressilator.
112: 
113: The toggle switch consists of two genes, which negatively
114: regulate each other expression.
115: The regulation is performed at the transcriptional level,
116: namely each gene codes for a repressor protein that binds
117: to the promoter of the other gene.
118: Such system may exhibit bistability, namely two stable
119: states, where in each state one of the proteins is dominant
120: and the other is suppressed.
121: A synthetic toggle switch
122: was constructed in 
123: {\it E. coli} 
124: and the conditions for bistability
125: were examined
126: \cite{Gardner2000}.
127: The switching between its two states was demonstrated using 
128: chemical and thermal induction.
129: More recently, such circuit was found to exist in a natural
130: system in which two mutual repressors regulate
131: the differentiation of myeloid progenitors
132: into either macrophages or neutrophils
133: \cite{Laslo2006}.
134: 
135: The synthetic repressilator circuit, constructed in
136: {\it E. coli}, 
137: was designed to exhibit
138: oscillations, reminiscent of natural genetic
139: oscillators such as the circadian rhythms
140: \cite{Elowitz2000}. 
141: The repressilator circuit was encoded on a plasmid
142: that appears in a low copy number. 
143: The protein concentrations
144: were measured vs. time in single cells.
145: Oscillations with a period of about 150 minutes
146: were found, 
147: namely extending over several division cycles.
148: The oscillations were found to be 
149: noisy, typically maintaining phase coherence for 
150: times of the order of a single oscillation
151: period.
152: 
153: The dynamics of genetic networks is often simulated
154: using rate equation models.
155: These are sets of coupled ordinary differential equations,
156: which account for the concentrations of the mRNAs and proteins
157: in the network.
158: In general, rate equations account for average concentrations
159: and ignore fluctuations. 
160: They are suitable for systems in which the concentration of
161: interacting molecules are large and fluctuations are negligible.
162: However, proteins in cells often appear in low copy numbers and
163: may exhibit large fluctuations. 
164: Moreover, in case of transcriptional regulation, the expression
165: of the regulated gene may be controlled by a single protein
166: that binds to its promoter.
167: This extends the notion of low copy numbers and the
168: resulting fluctuations even to the case
169: when there is a large number of free repressors of a certain
170: type, since only one of them may bind to the promoter at any
171: given time.
172: Recent advances measurement techniques made
173: it possible to measure the fluctuations in 
174: copy number of proteins in single cells
175: \cite{Elowitz2002,Paulsson2004,Swain2002}. 
176: Measurements of protein levels in single cells revealed distributions
177: that depend on the topology of the regulatory 
178: network controlling the particular protein
179: \cite{Ozbudak2004}. 
180: 
181: To account for the fluctuations, simulations of genetic networks
182: should be done using stochastic methods such as the master
183: equation 
184: \cite{Kepler2001,McAdams1997,McAdams1999,Paulsson2000,Paulsson2000b,Paulsson2002,Paulsson2004}
185: or Monte Carlo simulations
186: \cite{Gibson2000,Gillespie1977,McAdams1997}.
187: The master equation provides the probability distribution of 
188: the concentrations of proteins in a population of cells.
189: From this distribution one can calculate the average concentrations
190: as well as the correlations and the rates of all processes.
191: Monte Carlo simulations enable to follow the temporal variations
192: in the level of gene
193: expression and in the concentrations of proteins in a single cell.
194: These results enable to extract the noise level and temporal correlation 
195: functions. 
196: They can also be used to characterize the dynamics of the system
197: and determine whether it exhibits a single steady state, multi-stability
198: or oscillatory behavior.
199: 
200: In this paper we present deterministic analysis (using rate
201: equations) and stochastic analysis (using direct numerical integration of 
202: the master equation
203: and Monte Carlo simulations) of 
204: three simple genetic circuits:
205: the autorepressor, the toggle switch and the repressilator
206: (Fig.~\ref{fig:1}).
207: %
208: \begin{figure}
209: \includegraphics[width=10cm]{fig1.eps}
210: \centering
211: \caption{
212: The three genetic circuits considered in this paper:
213: the autorepressor (a); the genetic switch (b); and the repressillator (c).
214: The flat-headed arrows denote negative transcriptional regulation.}
215: \label{fig:1}
216: \end{figure}
217: %
218: We show that fluctuations give rise to both quantitative and
219: qualitative effect in the dynamics of thse circuits.
220: 
221: The paper is organized as follows.
222: In Sec. 2 we consider the autorepressor and 
223: present the deterministic and stochastic methods used in
224: the paper.
225: In Sec. 3 we study the genetic switch and
226: in Sec. 4 we analyze the repressillator circuit.
227: The results are discussed and summarized in Sec. 5.
228: 
229: %-------------------------------------------------------
230: \section{The Autorepressor}
231: 
232: The autorepressor is
233: the simplest genetic circuit that involves feedback.
234: It consist of a single gene, denoted by $a$, 
235: that negatively regulates its own expression.
236: The gene transcribes into mRNAs that translate into $A$ proteins.
237: These proteins function as repressors. 
238: When an $A$ protein binds to the $a$ promoter site,
239: it prevents 
240: the RNA polymerase from binding
241: to the promoter and thus inhibits the transcription process.
242: It turns out that genetic networks include a large
243: number of autorepressor modules. 
244: This circuit is thus considered as a network motif
245: \cite{Rosenfeld2002}.
246: One may speculate that it performs some crucial function
247: in the cell.
248: It was proposed that the role of the autorepressor is to
249: speed up response times
250: \cite{Rosenfeld2002} 
251: or to reduce fluctuations
252: \cite{Becskei2000}.
253: Below we analyze the autorepressor using deterministic
254: and stochastic methods. We utilize its simplicity in order
255: to present the methodologies in details.  
256: 
257: \subsection{Michaelis-Menten equations}
258: 
259: The dynamics of genetic networks are commonly described
260: using the Michaelis-Menten equations.
261: These equations describe the temporal variations in the
262: concentrations of the relevant molecules in the cell. 
263: Here, we denote 
264: the concentration of protein $A$ in a cell
265: by $[A]$
266: (by concentration we refer to the average copy number of $A$ proteins
267: per cell).
268: The concentration of the corresponding mRNA 
269: is denoted by $[m]$.
270: The Michaelis-Menten equations for the concentrations 
271: take the form
272: 
273: \begin{eqnarray}
274: \dot{[m]} &=& \frac{g_m}{1+k[A]^n} - d_m [m] \nonumber \\
275: \dot{[A]} &=& g_p [m] - d_p [A].
276: \label{eq:michaelis_menten_with_mrna}
277: \end{eqnarray}
278: 
279: \noindent
280: The parameters 
281: $d_m$ 
282: and
283: $d_p$ 
284: (sec$^{-1}$)
285: are the degradation rates 
286: of the mRNAs and proteins,
287: respectively. 
288: The trascription rate is given by $g_m$
289: (sec$^{-1}$).
290: The rate in which each mRNA is translated into proteins
291: is given by $g_p$
292: (sec$^{-1}$).
293: Since $A$ proteins negatively regulate their own synthesis,
294: the transcription rate is reduced by a factor of 
295: $1/(1 +k[A]^n)$.
296: This factor is called the Hill-function.
297: In this expression, the parameter $k$ 
298: quantifies the regulation strength 
299: (determined by the affinity between the repressor and the promoter site).
300: The exponent $n$ is called the Hill-coefficient.
301: In general, Hill-function models can be derived from more complete
302: rate equation models.
303: In this case,
304: $n$ is expected to take only
305: integer values.
306: In fact, $n$ 
307: represents the number of copies of the transcription
308: factor, that are required to be bound simultaneously to the promoter
309: in order to perform the regulation. 
310: The case of $n>1$ is often referred to as cooperative binding.
311: In this paper we consider only integer values of $n$.
312: However, similar models may also be used to fit empirical data.
313: In this case, $n$ is simply a fitting parameter which may take 
314: non-integer values. 
315: 
316: To simplify the analysis of genetic circuits,
317: the mRNA level is often ignored and
318: the transcription and translation processes 
319: are regarded as a single step
320: of protein synthesis
321: \cite{Rosenfeld2002}.  
322: In this case, 
323: the effective production rate of proteins
324: is given by 
325: $g=g_p g_m/d_m$. 
326: The Michaelis-Menten equations are reduced to
327: 
328: \begin{equation}    
329: \dot{[A]} = \frac{g}{1+k[A]^n} - d [A],
330: \label{eq:michaelis_menten_without_mrna}
331: \end{equation}    
332: 
333: \noindent
334: where $d=d_p$.
335: Ignoring the mRNA level is typically 
336: justified under steady state conditions. 
337: However, in the analysis of systems 
338: that are away from steady state 
339: due to external signals, or those
340: that exhibit oscillations
341: or large fluctuations
342: the mRNA level should be included.
343: The Michaelis-Menten equations for the
344: autorepressor exhibit a single steady state solution
345: for the concentration of $A$ proteins.
346: In case that $n=1$, it is 
347: given by
348: 
349: \begin{equation}
350: \label{eq:repressillator_steady_state_1}
351: [A] = \frac {-1+\sqrt{1+4kg/d}}{2k}.
352: \end{equation}
353: 
354: \noindent
355: This solution is found to be stable 
356: for any choice of the parameters. 
357: 
358: 
359: \subsection{Extended set of rate equations}
360: 
361: In the Michaelis-Menten equations the 
362: negative regulation
363: process is described by the Hill-function.
364: This description is rather crude and incomplete.
365: In order to model 
366: the regulation process in greater detail 
367: we present below a more complete rate equation model
368: \cite{Lipshtat2005}.
369: In this model, 
370: we account separably for the populations 
371: of free and bound proteins.
372: The bound $A$ proteins are denoted by $r$ and their 
373: concentration is given by $[r]$.
374: 
375: Consider an autorepressor gene $a$, encoded on the chromosome, which
376: exhibits no cooperative binding.
377: In this case the number of bound repressors is in the range
378: $0 \le [r] \le 1$.
379: The concentration, $[r]$, can also be considered as 
380: the fraction of time 
381: in which the promoter is occupied by a bound repressor
382: and the transcription process is suppressed. 
383: Therefore the transcription rate
384: is reduced by a factor of $(1-[r])$. 
385: Ignoring the mRNA level,
386: the extended set of rate equations takes the form
387: \cite{Lipshtat2005}
388: 
389: \begin{eqnarray}
390: \dot{[A]} &=& g (1-[r])-d [A]-\alpha_0[A]\left(1-[r]\right)
391: +\alpha_1[r] \nonumber \\
392: \dot{[r]} &=& \alpha_0[A]\left(1-[r]\right)-\alpha_1[r],
393: \label{eq:extended_rate}
394: \end{eqnarray}
395: 
396: \noindent
397: where the parameter $\alpha_0$ (sec$^{-1}$) is the binding rate of the 
398: repressors to the promoter site and $\alpha_1$ (sec$^{-1}$)
399: is the their unbinding rate.
400: In the limit in which the binding and unbinding processes 
401: are much faster than other 
402: processes in the system 
403: (namely $\alpha_0,\alpha_1 \gg d,g$),
404: these equations can be reduced to the Michaelis-Menten form. 
405: In this limit, the relaxation time of the concentration $[r]$ 
406: is much shorter than other relaxation times in the system. 
407: Therefore, one can take the 
408: time derivative of $[r]$ to zero, 
409: even if the system is away from steady state.
410: This brings the rate equations to the 
411: Michaelis-Menten form 
412: [Eq. (\ref{eq:michaelis_menten_without_mrna})]
413: with $n=1$ and 
414: $k=\alpha_0/\alpha_1$.
415: Therefore, 
416: Eqs. (\ref{eq:extended_rate}) 
417: have the same steady state solution
418: for the protein $A$ as the Michaelis-Menten equation
419: (\ref{eq:michaelis_menten_without_mrna}).
420: However, the dynamics leading to steady state may differ between 
421: the two equations. 
422: Furthermore, the extended rate equation model exhibits more
423: flexibility in the sense that 
424: it is much easier to insert additional
425: features into Eqs. (\ref{eq:extended_rate}) than into
426: Eq. (\ref{eq:michaelis_menten_without_mrna}). 
427: For example, it is possible to introduce degradation of bound
428: repressors by adding
429: the term 
430: $-d_r [r]$ to the equation for 
431: $\dot{[r]}$ in
432: Eqs. (\ref{eq:extended_rate}).
433: 
434: \subsection{Stochastic Analysis}
435: 
436: Transcription factors and other proteins, 
437: as well as their mRNAs in a cell, 
438: often appear in low concentrations
439: \cite{Gillespie1977,McAdams1999,Gibson2001}.
440: In this case, fluctuations in the copy numbers
441: of these molecules may play an important role
442: in the dynamics of genetic networks.
443: To obtain a better description of thse systems,
444: one should take into account 
445: the discrete nature of the molecules
446: rather than using continuous concentrations.
447: Moreover, even in case that some transcription  
448: factor appears in a high concentration,
449: the regulation is performed by a small number of
450: copies that are bound to the promoter site.
451: The fluctuations in the number of bound transcription 
452: factors give rise to large temporal variations
453: in the transcription rate of the regulated gene.
454: 
455: In order to account for fluctuations in genetic networks,
456: stochastic methods are required, such as the master equation
457: or Monte Carlo simulations
458: \cite{Arkin1998,Kepler2001,McAdams1997,Paulsson2000,Paulsson2004}.
459: The master equation is expressed in terms of
460: the probability distribution 
461: $P(N_A,N_r)$. 
462: This is the probability
463: for a cell to include 
464: $N_A=0,1,2,\dots$
465: free copies of protein $A$ 
466: and $N_r=0,1$ copies of the same protein, 
467: which are bound to the promoter.
468: The master equation accounts for the temporal variations in the probability 
469: distribution. 
470: For the autorepressor, it takes the form
471: \cite{Lipshtat2005}
472: 
473: \begin{eqnarray}
474: \label{eq:autorepressor_master}
475: \dot P(N_A,N_r) &=& 
476:  g \delta_{N_r,0} [P(N_A-1,N_r) - P(N_A,N_r)]  \nonumber\\
477:  &+& d [(N_A+1) P(N_A+1,N_r)      -   N_A  P(N_A,N_r)]       \nonumber\\
478:  &+& \alpha_0 [\delta_{N_r,1}(N_A+1) P(N_A+1,N_r-1) -  
479:                \delta_{N_r,0} N_A P(N_A,N_r)]                \nonumber\\
480:  &+& \alpha_1 [\delta_{N_r,0} P(N_A-1,N_r+1) -
481:                \delta_{N_r,1} P(N_A,N_r) ],
482: \end{eqnarray}
483: 
484: \noindent
485: where the $g$ term accounts for the production of proteins
486: and the $d$ term accounts for their degradation.
487: The $\alpha_0$ ($\alpha_1$) terms describe the binding
488: (unbinding) of proteins to (from) the promoter site.
489: 
490: In numerical integration, the master equation must be truncated
491: in order to keep the number of equations finite. 
492: This is done by setting a suitable upper cutoff,
493: $N_{\rm A}^{\rm max}$, 
494: on the population size of the free proteins.
495: In order to maintain the accuracy of the calculations,
496: the cutoff should be chosen such that
497: the probability of population sizes beyond it
498: will be sufficiently small.
499: The master equation exhibits a single steady state solution,
500: characterized by 
501: $\dot{P}(N_A,N_r)=0$
502: for all $N_A$ and $N_r$.
503: This solution is always stable 
504: \cite{VanKampen1992}.
505: The average concentration of free $A$ 
506: proteins can be 
507: obtained from
508: 
509: \begin{equation}
510: \langle N_{\rm A} \rangle =
511: \sum_{N_{\rm A}=0}^{N_{\rm A}^{\rm max}} 
512: \sum_{N_{\rm r}=0}^{1} 
513: N_{\rm A} P(N_A,N_r).
514: \label{eq:averagex}
515: \end{equation}
516: 
517: \noindent
518: Other properties of the distribution, such as the
519: variance, can be obtained from the calculation of
520: higher moments.
521: 
522: Another useful approach to the study of stochastic dynamics 
523: is provided by Monte Carlo methods
524: \cite{Gibson2000,Gillespie1977,McAdams1997}.
525: The simulation dynamics is Markovian. 
526: At each instant, the
527: next process to take place is chosen randomly 
528: from all the possible processes,
529: where each process is assigned with a suitable weight,
530: proportional to its rate.
531: The elapsed time is then updated accordingly.
532: Unlike the master equation, which accounts for the entire 
533: distribution, Monte Carlo simulations follow the temporal variations
534: in protein concentrations in a single cell.
535: 
536: In 
537: Fig.~\ref{fig:2}(a)
538: we present the concentration of $A$ proteins vs. time,
539: obtained from the rate equations
540: [Eq. (\ref{eq:extended_rate})]
541: and from Monte Carlo simulations.
542: In
543: Fig.~\ref{fig:2}(b)
544: we show the probability distribution
545: $P(N_{\rm A}$
546: obtained from the master equation
547: under steady state conditions.
548: 
549: \begin{figure}
550: \includegraphics[width=13cm]{fig2.eps}
551: \caption{
552: Results for the autorepressor circuit.
553: (a) The concentration of $A$ proteins vs. time, obtained
554: from the rate equations and from Monte Carlo simulations.
555: The rate equation results quickly reach a steady state.
556: The Monte Carlo results fluctuate around this steady state.
557: (b) The steady state probability distribution $P(N_A)$
558: for a cell to contain $N_A$ copies of protein $A$, 
559: obtained from the master equation.
560: The parameters used are 
561: $g=0.05$,
562: $d=0.001$,
563: $\alpha_0=0.01$,
564: $\alpha_1=0.01$
565: and
566: $d_r=0$
567: (sec$^{-1}$).
568: }
569: \label{fig:2}
570: \end{figure}
571: 
572: \section{The Genetic Switch}
573: 
574: The genetic toggle switch
575: consists of two proteins, $A$ and $B$, 
576: which negatively regulate each other at the transcriptional level
577: [Fig. \ref{fig:1}(b)].
578: This architecture may lead to two steady states, 
579: one dominated by $A$ proteins and the other dominated by $B$ proteins.
580: When the population of $A$ proteins is much larger than that of $B$ 
581: proteins, the $A$ proteins tend to suppress the production of $B$ 
582: proteins. 
583: Under these conditions, 
584: the production of $A$ proteins is enhanced, 
585: because the declining concentration of $B$
586: proteins is not sufficient to suppress it. 
587: Therefore, the system approaches a state
588: rich in $A$ proteins and poor in $B$ proteins. 
589: Similarly, the system may approach a state rich 
590: in $B$ proteins and poor in $A$ proteins. 
591: Transitions between the two states may take place
592: in response to a suitable external signal.
593: Spontaneous transitions, due to random fluctuations,
594: are also possible.
595: To qualify as a switch, the system should exhibit bistability. 
596: In the deterministic description, bistability is defined
597: as the existence of two stable steady state solutions 
598: of the rate equations.
599: This description does not account for the possibility of
600: spontaneous transitions between the two states.
601: In the stochastic description, spontaneous transitions
602: are taken into account.
603: Therefore, the condition for bistability
604: is that the rate of spontaneous transitions
605: (due to random fluctuations rather than 
606: an external signal) is much lower than the rates of
607: all other relevant processes in the system. 
608: 
609: Genetic switch systems exist in nature, and give
610: rise to different cell behaviors in different situations.
611: A notable example is the 
612: phage $\lambda$ switch
613: \cite{Ptashne1992}.
614: This switch appears in $\lambda$
615: phages, which infect {\it E. coli} bacteria and 
616: can exist in two exclusive states, one called 
617: lysogeny and the other called lysis. 
618: In addition to natural switches, a synthetic switch
619: was constructed and studied in 
620: {\it E. coli}
621: \cite{Gardner2000}.
622: Numerous studies, 
623: using rate equations,
624: have concluded
625: that cooperative binding is
626: a necessary condition for the emergence of bistability
627: \cite{Cherry2000,Gardner2000,Walczak2005,Warren2004,Warren2005}. 
628: Stochastic analysis reveals the reason to this fact.
629: For a switch without cooperative binding, three peaks
630: are obtained in the probability distribution function.
631: These peaks corresponds to three possible states for the system:
632: one in which $A$ is highly expressed, 
633: a second in which $B$ is highly expressed 
634: and a third in which both proteins are suppressed
635: (a 'deadlock' situation)
636: \cite{Lipshtat2006,Loinger2007}. 
637: Monte Carlo simulations show rapid transitions
638: between these three states.
639: The possibility of simultaneous suppression of both proteins, 
640: prevents the system from functioning as a switch.
641: It causes the system to exhibit three states 
642: of limited stability 
643: instead of the two stable states that are desired.
644: 
645: It is found that in switch systems in which the
646: $A$ and $B$ repressors exhibit cooperative binding,
647: the deadlock situation is removed and bistability
648: emerges.
649: This can be explained as follows.
650: The deadlock situation results from a simultaneous 
651: binding of $A$ and $B$ repressors to the corresponding promoter
652: sites.
653: Without cooperative binding, it is sufficient for
654: the minority protein to recruit a single copy that will bind 
655: and suppress the production of the dominant protein.
656: In the case of cooperative binding (for example, with $n=2$)
657: the minority protein needs to recruit two copies that will
658: bind simultaneously
659: in order to suppress the production
660: of the dominant protein.
661: This is much less likely. 
662: Therefore, cooperative binding induces bistability
663: and enables the system to function as a switch.
664: 
665: Apart from cooperative binding, 
666: there are several other mechanisms that may stabilize
667: bistability in genetic switch systems.
668: The most obvious is the exclusive switch,
669: where there is an overlap between the promoters 
670: of $A$ and $B$, leaving no room for both to be 
671: occupied simultaneously.
672: Such situations are encountered in nature,  
673: for example, in the 
674: lysis-lysogeny switch of phage $\lambda$
675: \cite{Ptashne1992}.
676: It was shown that 
677: in the presence of cooperative binding,
678: the exclusive binding of $A$ and $B$ repressors
679: enhances the stability of the
680: genetic switch
681: \cite{Warren2004}.
682: This is because in exclusive binding the access of the
683: minority specie to the promoter site is blocked by the
684: dominant specie.
685: For the exclusive switch without cooperative binding,
686: rate equations exhibit only a single steady state solution,
687: namely, the system is not a switch in the deterministic framework.
688: However, in the stochastic framework the system exhibits 
689: bistability and functions as a switch.
690: The distribution $P(N_A,N_B)$ 
691: exhibits two peaks, one dominated by
692: $A$ proteins and the other dominated by
693: $B$ proteins.
694: The exclusive binding prevents the possibility of a
695: deadlock situation. 
696: 
697: 
698: \begin{figure}
699: \includegraphics[width=13cm]{fig3.eps}
700: \caption{
701: The probability distribution $P(N_{\rm A},N_{\rm B})$
702: of the concentrations of $A$ and $B$ proteins, for
703: the switch with degradation of bound repressors,
704: obtained from the master equation. 
705: Two sharp peaks are observed, one dominated by $A$ proteins
706: and the other dominated by $B$ proteins.
707: The peaks are sharp and are separated by a region
708: with vanishing probabilities.
709: As a result, the transition rate between the two
710: peaks is low and the switch is stable. 
711: }
712: \label{fig:3}
713: \end{figure}
714: 
715: 
716: 
717: \begin{figure}
718: \includegraphics[width=13cm]{fig4.eps}
719: \caption{
720: The concentration, $N_A$, of $A$ proteins vs. time,
721: obtained from Monte Carlo simulations for the switch with
722: degradation of bound repressors.
723: The two states are clearly observed: one in which $N_A$  
724: fluctuates around 50, and another in which it is nearly zero.
725: Transitions between these states occur 
726: at an average rate of one transition every 
727: $\sim 10^6$ sec (about 10 days).
728: }
729: \label{fig:4}
730: \end{figure}
731: 
732: In addition to the exclusive switch, there exist other
733: variants of the genetic switch circuit 
734: (we focus here on systems without cooperative binding).
735: Consider a switch in which not only free proteins,
736: but also bound proteins experience degradation.
737: Bound-repressor degradation 
738: tends to prevent the deadlock situation in which 
739: both $A$ and $B$ repressors are bound simultaneously. 
740: This is due to the fact that degradation removes the bound repressor 
741: from the system, unlike 
742: unbinding, where the resulting free repressor may quickly bind again.
743: It turns out that degradation of bound repressors induces 
744: bistability not only in the stochastic framework 
745: but also at the level of deterministic rate equations.
746: 
747: 
748: In Fig. \ref{fig:3}
749: we present the 
750: probability distribution 
751: $P(N_{\rm A},N_{\rm B})$
752: of the concentrations of $A$ and $B$ proteins, for
753: the switch with degradation of bound repressors.
754: Two sharp peaks, well separated from each other are
755: observed, illuminating the bistable nature of the
756: systems.
757: The parameters used 
758: in Fig. \ref{fig:3}
759: and in the rest of the paper
760: are:
761: $g=0.15$,
762: $d=0.003$,
763: $\alpha_0=0.5$,
764: $\alpha_1=0.01$
765: and 
766: $d_r=0.003$
767: (sec$^{-1}$).
768: 
769: In Fig. \ref{fig:4}
770: we present the temporal variations of 
771: the concentration, 
772: $N_A$, of $A$ proteins, obtained from Monte Carlo simulations
773: for the switch with degradation of bound repressors.
774: Two states are clearly observed: one in which $N_A$ is dominant
775: and another state in which it is suppressed.
776: Note that in spite of the very large fluctuations,
777: the switch is stable and the average time between 
778: spontaneous transitions is about 10 days.  
779: 
780: Another variant of the genetic switch, 
781: which exhibits bistability even at the level
782: of rate equations, 
783: involves protein-protein interactions,
784: where $A$ and $B$ proteins bind to each other and
785: form a complex that does not function as a transcription factor.
786: This additional process contributes to the stability of the
787: switch because in such 'pair annihilation' processes
788: the minority protein is affected more strongly.
789: It is thus less likely to bind and suppress the production
790: of the dominant protein.
791: 
792: \section{The Repressilator}
793: 
794: The repressilator circuit consists of three  
795: transcription factors, 
796: $A$, $B$ and $C$, which 
797: negatively regulate each other's synthesis 
798: in a cyclic manner
799: [Fig.~\ref{fig:1}(c)].  
800: This circuit was synthetically constructed 
801: on plasmids in {\it E. coli} 
802: and was found to exhibit oscillations
803: in the concentrations of the three transcription factors. 
804: To understand the origin of these oscillations, 
805: consider a situation in which the number of $A$ proteins is large. 
806: In this case it is likely that one of the $A$ proteins
807: will bind the to $b$ promoter and will 
808: repress the production of $B$ proteins.
809: The reduced level of $B$ proteins will enable the gene
810: $c$ to be fully expressed and the number of $C$ proteins
811: will increase and will start to repress gene $a$.
812: As a result, the number of $A$ proteins will decrease,
813: and gene $b$ will be activated, completing a full cycle.
814: The order of appearance of the dominant protein
815: type in this cycle is $A \to C \to B \to A$. 
816: 
817: From a theoretical point of view, oscillations in this
818: system can be obtained (under some conditions) both in 
819: rate equations and in Monte Carlo simulations
820: [Fig. \ref{fig:5}].
821: %
822: \begin{figure}
823: \includegraphics[width=13cm]{fig5.eps}
824: \caption{
825: The concentration, $N_A$, of $A$ proteins vs. time, obtained from
826: Monte Carlo simulations of the repressilator circuit.  
827: The oscillations are noisy. Their
828: period and amplitude vary from cycle to cycle.
829: }
830: \label{fig:5}
831: \end{figure}
832: %
833: The oscillations obtained from the rate equations 
834: are regular,
835: forming a stable limit-cycle.
836: The oscillations
837: obtained from the Monte Carlo simulations are noisy and irregular.
838: Moreover, the period and amplitude differ significantly 
839: between the rate equations and the Monte Carlo simulations
840: \cite{Loinger2007b}.
841: 
842: The repressilator system was constructed synthetically on
843: plasmids. When the number of plasmids in a cell is small,
844: the numbers of promoter sites and bound transcription factors 
845: are also small.
846: As a result, one expects large fluctuations in the transcription
847: rates regulated by these transcription factors. 
848: In this case stochastic methods are required.
849: However, when the number of plasmids is large, fluctuations
850: are reduced and the rate equations become applicable.
851: Therefore, by gradually changing the number of plasmids in the cell
852: one can explore the cross-over from the stochastic regime,
853: where the oscillations are noisy, to the deterministic 
854: regime, where the oscillations are regular.
855: 
856: \section{Summary}
857: 
858: We have presented deterministic and stochastic analysis of 
859: three simple genetic circuits which involve transcriptional
860: regulation and feedback: 
861: the autorepressor, the switch and the repressilator.
862: Such systems can be simulated using rate equations, that
863: account for the concentrations of the mRNAs and proteins produced
864: by these genes. 
865: Rate equations 
866: are suitable when these concentrations are large and 
867: fluctuations are negligible.
868: However, when some of the transcription factors and their binding sites
869: in a cell appear in low copy numbers, 
870: fluctuations become important
871: and the rate equations fail. 
872: In this case stochastic methods such as the master equation or
873: Monte Carlo simulations are required.
874: We have shown that fluctuations give rise to quantitative
875: and qualitative changes in the dynamics of the systems.
876: In particular, we demonstrated the fluctuations-induced bistability
877: in the exclusive switch and the noisy oscillations
878: obtained in the repressilator circuit.
879: 
880: \bibliographystyle{prsty}
881: \begin{thebibliography}{xx}
882: 
883: \bibitem{Arkin1998}
884: {A. Arkin, J. Ross and H.H. McAdams}, 
885: {\it Stochastic kinetic analysis of developmental pathway bifurcation
886: in $\lambda$-infected E. coli cells}
887: Genetics {\bf 149} (1998), 1633.
888: 
889: \bibitem{Becskei2000}
890: {A. Becskei and L. Serrano},   
891: {\it Engineering stability in gene networks by autoregulation}, 
892: Nature {\bf 405} (2000), 590.
893: 
894: \bibitem{Cherry2000}
895: {J.L. Cherry and F.R. Adler}, 
896: {\it How to make a biological switch}
897: J. Theor. Biol. {\bf 203} (2000), 117.
898: 
899: \bibitem{Elowitz2000}
900: {M.B. Elowitz and S. Leibler},    
901: {\it A synthetic oscillatory network of transcriptional regulators}, 
902: Nature {\bf 403} (2000), 335.
903: 
904: \bibitem{Elowitz2002}
905: {M.B. Elowitz, A.J. Levine, E.D. Siggia and P.S. Swain},  
906: {\it Stochastic gene expression in a single cell}, 
907: Science {\bf 297} (2002), 1183.
908: 
909: \bibitem{Gardner2000}
910: {T.S. Gardner, C.R. Cantor and J.J. Collins}, 
911: {\it Construction of a genetic toggle switch in E. coli}
912: Nature {\bf 403} (2000), 339.
913: 
914: \bibitem{Gibson2000}
915: {M.A. Gibson and J. Bruck}, 
916: {\it Efficient
917: exact stochastic simulation of chemical systems with many species and many
918: channels}, 
919: {J. Phys. Chem.} {\bf 104} (2000), 1876.
920: 
921: \bibitem{Gibson2001}
922: {M.A. Gibson and E. Mjolsness},
923:   {\it Modeling the activity of single genes}, {\em in} {J. M. Bower and H.
924:   Bolouri}, ed., {Computational Modeling of Genetic and Biochemical
925:   Networks}, {MIT press}, {Cambridge, MA} (2000), pp.~1--48.
926: 
927: \bibitem{Gillespie1977}
928: {D.T. Gillespie}, 
929: {\it Exact stochastic simulation of coupled chemical reactions}, 
930: {\em J. Phys. Chem.} 
931: {\bf 81} (1977), 2340.
932: 
933: \bibitem{Kepler2001}
934: {T.B. Kepler and T.C. Elston},
935: {\it Stochasticity in transcriptional regulation: origins, consequences, and
936:   mathematical modeling representations}, 
937: {\em Biophysical Journal} {\bf 81} (2001), 3116.
938: 
939: \bibitem{Laslo2006}
940: {P. Laslo et al.}, 
941: {\it Multilineage transcriptional priming and determination of
942: alternate hematopoetic cell fates},
943: Cell {\bf 126} (2006), 755.
944: 
945: \bibitem{Lipshtat2005}
946: {A. Lipshtat, H.B. Perets, N.Q. Balaban and O. Biham}, 
947: {\it Modeling of negative autoregulated genetic networks in single cells},
948: Gene {\bf 347} (2005), 265. (See also arxiv:q-bio.MN/0504030)
949: 
950: \bibitem{Lipshtat2006}
951: {A. Lipshtat, A. Loinger, N.Q. Balaban and O. Biham}, 
952: {\it Genetic toggle switch without cooperative binding},
953: Phys. Rev. Lett. {\bf 96} (2006), 188101.
954: 
955: \bibitem{Loinger2007}
956: {A. Loinger, A. Lipshtat, N.Q. Balaban and O. Biham}, 
957: {\it Stochastic simulations of genetic switch systems},
958: Phys. Rev. E {\bf 75} (2007),  021904.
959: 
960: \bibitem{Loinger2007b}
961: {A. Loinger and O. Biham}, 
962: {\it Stochastic simulations of the repressilator circuit},
963: submitted to Phys. Rev. E (2007). 
964: 
965: \bibitem{McAdams1997}
966: {H.H. McAdams and A. Arkin},
967: {\it Stochastic mechanisms in gene expression}, 
968: {\em Proc. Natl. Acad. Sci. US}
969:   {\bf 94} (1997), 814.
970: 
971: \bibitem{McAdams1999}
972: {H.H. McAdams and A. Arkin}, {\it It's a
973:   noisy business! Genetic regulation at the nanomolar scale}, {\em Trends
974:   Genet.} {\bf 15} (1999), 65.
975: 
976: \bibitem{Milo2002}
977: {R. Milo, S. Shen-Orr, S. Itzkovitz, N. Kashtan, D. Chklovskii and U. Alon},
978: {\it Network Motifs: Simple Building
979:   Blocks of Complex Networks}, {\em Science} {\bf 298} (2002), 824.
980: 
981: \bibitem{Ozbudak2004}
982: E.M. Ozbudak, M. Thattai, H.N. Lim, B.I. Shraiman and A. van Oudenaarden  
983: {\it Multistability in the lactose utilization 
984: network of Escherichia coli} 
985: {\em Nature} {\bf 427} (2004) 737.
986: 
987: \bibitem{Paulsson2000}
988: {J. Paulsson and M. Ehrenberg},
989: {\it Random signal fluctuations can reduce random fluctuations in regulated
990: components of chemical regulatory networks}, 
991: {\em Phys. Rev. Lett.} {\bf
992:   84} (2000), 5447.
993: 
994: \bibitem{Paulsson2000b}
995: {J. Paulsson}, {\it Stochastic focusing:
996:   fluctuation-enhanced sensitivity of intracellular regulation}, {\em Proc.
997:   Natl. Acad. Sci. US} {\bf 97} (2000), 7148.
998: 
999: \bibitem{Paulsson2002}
1000: {J. Paulsson}, {\it Multileveled
1001: selection on plasmid replication}, 
1002: {\em Genetics} {\bf 161} (2002), 1373.
1003: 
1004: \bibitem{Paulsson2004}
1005: {J. Paulsson}, 
1006: {\it Summing up the noise
1007: in gene networks}, {\em Nature} {\bf 427} (2004), 415.
1008: 
1009: \bibitem{Ptashne1992}
1010: {M. Ptashne}, 
1011: {\em {A Genetic Switch: Phage $\lambda$ and Higher Organisms, 2nd
1012:   edition.}} ({Cell Press and Blackwell Scientific Publications}, {Cambridge,
1013:   MA}, 1992).
1014: 
1015: \bibitem{Rosenfeld2002}
1016: {N. Rosenfeld, M.B. Elowitz and U. Alon}, 
1017: {\it Negative autoregulation speeds the response times
1018: of transcription networks}, 
1019: {\em J. Mol. Biol.} {\bf 323} (2002), 785.
1020: 
1021: \bibitem{Swain2002}
1022: {P.S. Swain, M.B. Elowitz and E.D. Siggia}, 
1023: {\it Intrinsic and extrinsic contributions to
1024:   stochasticity in gene expression}, 
1025: {\em Proc. Natl. Acad. Sci. US} {\bf 99} (2002), 12795.
1026: 
1027: \bibitem{VanKampen1992}
1028: {N.G. van Kampen}, 
1029: {\em {Stochastic processes in physics and chemistry}}
1030: ({North-Holland}, {}, 1992).
1031: 
1032: \bibitem{Walczak2005}
1033: {A. M. Walczak, M. Sasai, and P. Wolynes}, 
1034: {\it Self consistent proteomic field theory of stochastic gene switches},
1035: Biophysical Journal {\bf 88},  
1036:   (2005) 828.
1037: 
1038: \bibitem{Warren2004}
1039: {P.B. Warren and P.R. ten Wolde}, 
1040: {\it Enhancement of the stability of genetic switches by overlapping upstream
1041: regulatory domains},
1042: Phys. Rev. Lett. {\bf 92} (2004),  128101.
1043: 
1044: \bibitem{Warren2005}
1045: {P.B. Warren and P.R. ten Wolde}, 
1046: {\it Chemical models of genetic toggle switches},
1047: J. Phys. Chem. B {\bf 109} (2005), 6812.
1048: 
1049: \end{thebibliography}
1050: 
1051: \end{document}
1052: