arXiv:1007.1928v1 [physics.bio-ph] 12 Jul 2010 itiuin o hs rcse r oegnrlthan characteristic general are more which distributions are exponential processes simple these for waiting-time the distributions Correspondingly, steps. ob- multi- biochemical cellular involve the ple of often from degradation derive (mRNAs/proteins) and mRNAmacromolecules terms creation These for both that periods [11]. servation decay ’senescence’ give can and and that birth ex- ’gestation’ processes gene of to for effects rise the Recent models include these 10]. to [5, generalized pression stage further active has the work during only synthe- occuring and mRNA sis with active scenario) the between (Telegraph by switching states driven inactive promoter In primarily of are kinetics [9]. fluctuations slow mRNAs case, of low other to synthesis the related infrequent is levels and protein scenario) abundance in (Poisson variability process syn- high Poisson the mRNA a and case, as one modeled In is To [8]. distinct thesis proposed two been fluctuations, [5–7]. have mRNA synthesis models of mRNA source the in elucidate fluctuations to tributed of source a as ’noise’ heterogeneity. expression phenotypic gene drive mecha- which molecular uncovering nisms con- on benefits been focusing potential has effort there siderable The Correspondingly for [4]. discussed [3]. systems been diverse have cells heterogeneity of phenotypic such population rise of give a individual- can in cells non-genetic turn, ity and of in heterogeneity levels, population phenotypic protein isogenic to in Variations an 2]. across variations [1, large levels to protein rise give in can expression gene in volved ihvralt npoenlvl a eeal enat- been generally has levels protein in variablity High in- reactions biochemical of stochasticity intrinsic The plctoso itesLwt tcatcmdl fgn ex gene of models stochastic to Law Little’s of Applications fgn xrsinadpolm fitrs nqeen theo relatio with queueing exact models in derive for we interest means Law, distribution of Little’s theory, problems consideri queueing by and from issue expression this address gene we of work, there distribu this Correspondingly, protein In steady-state expression. and expression. burst between protein relations of observ ’bursts’ mo experimental explain stochastic of models in scenario proposed Telegraph and been have different variability, processes) this explain ’Telegraph’ To cells. of population eemndfo hne nma taysaepoenlevel transcr protein of steady-state degree mean th the expression. in as that changes levels show from protein obtained determined deri results mean we The parameters, the of for varied. range m simulations) broad a stochastic a consider For by we RNAs. this, small illustrate by To bursts expression. gene in noise ereo rncitoa usigadhnet discrimina implicati to have hence relations and derived bursting The transcriptional of proteins. degree and mRNAs of h nrni tcatct fgn xrsincnla ol to lead can expression gene of stochasticity intrinsic The INTRODUCTION eateto hsc,Vrii oyehi nttt and Institute Polytechnic Virginia Physics, of Department ldElgart, Vlad arbitrary ∗ Dtd coe 5 2018) 25, October (Dated: a Jia, Tao atn-iedsrbtosfrarvladdegradation and arrival for distributions waiting-time † n au .Kulkarni V. Rahul and h oreo utain nmN synthesis. mRNA determine in to fluctuations used of be source the cannot distributions protein steady- of state measurements gestation single-cell as most well senescence) as and different bursting from transcriptional contributions (e.g. include sources can noise levels observed the protein Since in processes. Poisson single-step of cnro,i a enage htdnmcmeasure- needed. are dynamic cells single that in mRNAs argued of number been the of ments has Telegraph it and Poisson scenarios, the between discriminate cannot dis- burst mRNA [15]. underlying protein distribution the of burst identify uniquely observations scenarios, not that two do shown tribution the burst be for mRNA can distinct underlying it be the thus While ran- is can a promoter distributions state. the from active when the produced synthesized in is mRNAs of it burst scenario, dom Telegraph a whereas mRNA, the is single burst for a observed from the translation syn- of scenario, consequence mRNA Poisson the for For scenarios thesis. two the consider can to between distributions interest burst discriminate of protein is of it quantification lifetimes features, whether these mRNA Given typical to 14]. between compared [13, to separation large seen mean is a indeed that with was bursts bursts expression random Protein in occur 14]. sin- in [13, proteins cells of gle synthesis to real-time led monitoring have Further- techniques studies 12]. single-molecule [6, produc- in bursts advances protein random more, that infrequent limit in Tele- the occurs re- and tion in scaling Poisson models observed the scenario The both graph with 7]. consistent [6, is abun- lationship genes protein several mean of for function dance a as expression protein in ie htpoensed-tt n us distributions burst and steady-state protein that Given noise of variation the determined have studies Recent ore fmN utain ’oso’and (’Poisson’ fluctuations mRNA of sources in o eea tcatcmdl fgene of models stochastic general for tions gampigbtensohsi models stochastic between mapping a ng swihcnetbrtadsteady-state and burst connect which ns ebtendffrn ore fintrinsic of sources different between te toso os npoenlvl nterms in levels protein in noise of ations dlfrrglto fpoenexpression protein of regulation for odel revrain npoenlvl cosa across levels protein in variations arge scnieal neeti establishing in interest considerable is o eea tcatcmdl fgene of models stochastic general for s eso eeepeso.Bt Poisson Both expression. gene of dels eeso euaoysalRA are RNAs small regulatory of levels e pinlbrtn a,i rnil,be principle, in can, bursting iptional y yapyn eea theorem general a applying By ry. n o prahst uniythe quantify to approaches for ons eaayia xrsin (validated expressions analytical ve tt University State ‡ pression 2

Such methods have indeed been developed in recent years from the OFF state to the ON state (with rate β). Note [2, 12, 16], and have been used to quantify the degree of that in the limit β << α, mRNAs will be produced in transcriptional bursting. In this context, it would be of infrequent bursts. Thus the Poisson process gives rise to interest to derive equations relating burst and steady- only protein bursts, whereas the Telegraph process gives state distribution means for both mRNAs and proteins. rise to both mRNA and protein bursts. It is interest- Such relations can provide useful checks for experimental ing to note that the mRNA burst distribution for both approaches for measuring mRNA/protein burst distribu- the Poisson and Telegraph scenarios can be represented tion means. Furthermore they can also suggest alterna- by the conditional geometric distribution; specifically by tive approaches which allow inference of the underlying considering bursts conditional on production of at least mRNA burst distribution. This work focuses on deriv- one mRNA [15]. This can be understood as follows: the ing such relations between the means of mRNA/protein mRNA burst distribution is the number of mRNAs pro- burst and steady-state distributions and exploring their duced in the active state before transition to the inactive consequences for approaches to quantify the degree of state. Let us take the initial condition to correspond to transcriptional bursting. the transition from the inactive (OFF) state to the active In this paper, we consider a mapping between general (ON) state i.e. at t = 0 the DNA has just transitioned stochastic models of gene expression [11] and problems of to the ON state. Now the next reaction that can occur interest in queueing theory. By applying a general theo- either results in the production of a mRNA (with rate rem from queueing theory, Little’s Law, we derive exact km) or a transtion to the OFF state (with rate α). The relations connecting mRNA/protein burst and steady- probablity of the next reaction being protein production state distribution means for stochastic models of gene is km whereas the probability that it is a transition to km+α expression with arbitrary waiting-time distributions for the OFF state is α . If we set p = α , the number km+α km+α arrival and degradation of mRNAs and proteins. Fur- of mRNAs produced (m) before the transtion to the OFF thermore the derived relations can be used to show how state, conditional on the production of atleast 1 mRNA, mRNA burst distributions can be inferred from measure- is given by ments of mean protein levels by introducing an additional π (m)=(1 − p)m−1p, m ≥ 1, (2) interaction in the reaction scheme. Specifically, we con- m sider a reaction scheme that includes interaction between πm(0) = 0, (3) mRNAs and regulatory genes called small RNAs. In bac- For the conditional geometric distribution given above, teria, small RNAs have been studied extensively in recent the mean is given by years [17] in part due to the critical roles they play in 1 cellular post-transcriptional regulation in response to en- mb = (4) vironmental changes. The results derived in this work, p besides the potential applications for quantifying the For the Poisson scenario (p = 1), a single mRNA is degree of transcriptional bursting, also provide insight produced per burst, which corresponds to the conditional into small-RNA based regulation for specific parameter geometric distribution with mean mb = 1. The Telegraph ranges. scenario also gives rise to a conditional geometric distri- bution for mRNA bursts, but with mean mb > 1. Thus, determination of the degree of transcriptional bursting MODEL AND RESULTS (mb) can discriminate between the Poisson and Telegraph scenarios for intrinsic noise in gene expression. Connecting burst and steady-state means The general model for gene expression that we analyze is as follows. Bursts of protein expression result due to We begin by considering the minimal reaction scheme translation from the underlying mRNA burst, which has for translation from mRNAs a conditional geometric distribution with mean mb. The number of proteins produced from different mRNAs are kp µ µp M −→ M + P ; M −→m ∅; P −→ ∅; (1) taken to be independent random variables. The decay time for mRNAs and proteins is assumed to be drawn A single burst corresponds to proteins produced from the from arbitrary waiting-time distributions with means τm underlying mRNA burst distribution until decay of the and τp respectively. Likewise, the waiting-time distri- last mRNA. For the Poisson process, mRNA transcrip- bution between consecutive bursts is a random variable tion occurs with constant probability per unit time km. drawn from an arbitrary distribution with mean τb. Cor- On the other hand, for the Telegraph process, mRNA respondingly, the average arrival rate for bursts is given transcription occurs with constant rate k only when the by k = 1 . Since the waiting-time distributions are arbi- m b τb DNA is in the active(ON) state; once it transitions from trary, effects due to gestation and sensecence of mRNAs the ON state to the inactive OFF state (with rate α) no and proteins [11] are included. For this setup, we will de- mRNA transcription can occur until it transitions back rive analytical relations which can be used to determine 3 mb and thereby to quantify the degree of transcriptional which can be done using single-molecule approaches. bursting. Such a procedure was used in Ref. [14] to estimate the We begin with the observation that the processes con- degree of transcriptional bursting, with the assumption sidered in the above model have exact analogs in prob- of constant mRNA arrival rates and decay rates. Eq.7 in- lems of interest in queueing theory. For example, the dicates that, even if this is not the case and arbitrary ges- creation of proteins corresponds to the arrival of cus- tation and sensescence periods are considered, the above tomers in queueing models [18]. On the other hand, the procedure remains a valid approach to determine the de- service-time distribution corresponds to the waiting-time gree of transcriptional bursting mb. Alternatively, since for the customer to depart the system, making it the ana- the above relation is valid for arbitrary stochastic pro- log of the waiting-time distribution for degradation of cesses governing mRNA arrival and decay, it can serve as proteins. Given that degradation of each mRNA/protein a useful consistency check for different experimental ap- is independent of other mRNAs/proteins in the system, proaches for quantifying mRNA burst and steady-state the mapping corresponds to queueing systems with in- distributions. finite servers. This can be seen as follows. In infinite Using Little’s Law we can also relate the steady-state server queues, since the number of servers is unlimited, protein distribution mean to the burst mean following each customer is associated with a server immeduately similar logic. Since the average arrival rate of proteins is upon arrival. This effectively implies that each customer given by hmikp, we derive is served independently of the others, which for the gene expression model is equivalent to the assumption that Ps = hmikpτp, (8) mRNAs/proteins are degraded independently. where k and τ −1 are average synthesis and decay rates A general theorem from queueing theory, Little’s Law p p [18], states that the average number of customers in the of the proteins. The above equation can be recast in system (L), the mean arrival rate (λ) and the mean wait- terms of the mean number of proteins produced in a sin- ing time of a customer in the system (W ) are related by gle burst Pb (which is related to the mRNA burst distri- L = λW . The remarkable feature of Little’s Law is that bution mean by Pb = mbkpτm). Since the mean arrival it holds regardless of the specific forms of the arrival and rate of proteins is given by kbPb, we have departure processes. When applied to stochastic gene Ps = kbPbτp (9) expression models, this implies that the processes lead- ing to mRNA/protein can be arbitrary, e.g. including It is noteworthy that this simple relation is valid for ar- gestation and senescence effects. bitrary gestation and senescence waiting-time distribu- We now apply Little’s Law to derive an equation re- tions. It establishes that the mean steady-state protein lating mRNA burst and steady-state distribution means. level only depends on the average protein arrival and The arrival rate of mRNA bursts is driven by an arbitrary degradation rates and is independent of the higher mo- stochastic process with average arrival rate kb. The de- ments of the corresponding waiting-time distributions. cay process of mRNA is also assumed to be driven by Thus, it explains the observation in Ref [11] that gesta- an arbitrary stochastic process with average decay time tion and senescence do not affect the average susceptibil- τm. Employing Little’s Law [18], we obtain a relation ity to changes in parameters. between the mean mRNA burst size mb and the average Another important consequence of Eq.9 is that pro- number of mRNAs in the steady state: cesses that alter the burst distribution mean without af- fecting protein degradation times or burst arrival times hmi = λτm, (5) will produce a proportionate change in the steady-state distribution mean. Thus, regulatory interactions which where λ is average arrival rate of the mRNAs, which is are sensitive to the degree of transcriptional bursting and given by alter protein burst distributions will produce proportion- ate changes in protein steady-state distribution means. λ = mbkb (6) This, in turn, suggests the possibility of obtaining signa- Hence, we derive that the steady-state distribution mean tures of transcriptional bursting by observing changes in for mRNAs is related to the mean mRNA burst size by steady-state protein distribution means upon regulation. To explore this possibility, let us consider how regulation hmi = mbkbτm (7) by small RNAs modulates protein burst distributions.

Both the mean steady-state mRNA levels (hmi) and Regulation by small RNAs the mean mRNA lifetime (τm) can be determined exper- imentally using standard procedures. Eq. 7 implies that the degree of transcriptional bursting can then be deter- We consider regulation by small RNAs (sRNAs) based mined by estimating the mean burst arrival rate (kb), on a coarse-grained model (Fig. 1) studied previously 4

with sRNAs (˜πm(m)) is given by

∞

π˜m(m)= ρ(n)πm(m + n), m ≥ 1, (10) nX=0 where ρ(n) is the probability of finding n sRNA molecules at the time of burst. Any burst of m mRNA molecules instantly becomes an effective burst of m − n mRNA molecules (for m>n) due to coupled degradation with n sRNAs. If m ≤ n, the mRNA burst after the regulation will be effectively an ‘empty’ burst. The probability of an empty burst is given by FIG. 1. The kinetic scheme for regulation of protein produc- tion by small RNAs with coupled degradation rate γ. ∞ π˜m(0) = 1 − π˜m(m). (11) mX=1 Since the unregulated mRNA burst distribution is ge- [19–21] which applies to sRNAs that regulate mRNA tar- ometric (with parameter p, say), we derive gets stoichiometrically due to coupled degradation [22]. ∞ Synthesis of sRNAs is taken to be a Poisson process with n π˜m(m)= πm(m) (1 − p) ρ(n) constant rate ks and the sRNA degradation rate is also nX=0 taken as constant (µ ) in the following analysis. The pa- s = G(1 − p)π (m), m ≥ 1, (12) rameter γ controls mutual degradation of mRNAs inter- m acting with sRNAs. As in the previous section, mRNAs where G(1 − p) is the generating function of sRNA are created in bursts, with the average rate of arrival for probability distribution ρ(n), evaluated at the point 1−p. bursts given by kb. If kbµs << 1, i.e. if the sRNA life- Using Eq. 11 we derive time is small compared to the mean arrival time between bursts, the distribution of sRNAs prior to a mRNA burst π˜m(0) = 1 − G(1 − p), (13) can be approximated by the steady-state distribution of sRNAs in the absence of mRNAs. Given this approxima- The regulated mRNA burst distribution is thus a con- tion, we wish to derive expressions for the protein burst ditional geometric distribution as in the unregulated case, ˜ distribution in the presence of sRNAs. This is, in general, but with modified average arrival rate km = G(1 − p)km. analytically intractable. However by employing further This is because 1 − π˜m(0) = G(1 − p) is the probabil- approximations which are valid for a range of parameters ity that the regulated burst results in atleast 1 mRNA. we can obtain analytical expressions for the burst dis- Therefore, the average number of mRNAs in the steady tribution. Specifically, we assume that synthesis of new state for the regulated case is given by (according to the sRNAs during a burst can be ignored, i.e. no new sRNAs equation Eq.7) are created in the time interval between mRNA creation hm˜ i = mbk˜mτm = G(1 − p)hmi. (14) and decay. Furthermore, we consider γτm ≫ 1 such that mRNA degradation in the presence of sRNAs is assumed We denote by Pb(ns) the mean burst size for proteins to occur due to mutual degradation with a sRNA rather 1 in the presence of sRNAs, where ns = ks/µs. Using the than natural decay with average rate µm = . Given τm equations Eq. (8,14) for mRNA’s steady state average, we that these approximations are valid, a simple analytic derive in the limit of fast coupled degradation (γ → ∞) expression for the mean regulated protein levels can be obtained as a function of mean sRNA levels as shown Ps(ns)= G(1 − p)Ps(0). (15) below. Taking the sRNA distribution prior to the burst (ρ(n)) to be a Poisson distribution with mean ns = ks/µs, and given that the mean mRNA burst size is given by m = As indicated in the reaction scheme in Fig.1, a pair b 1/p (Eq. 4), we derive of molecules of mRNA and sRNA can combine and be degraded rapidly with rate γ. We first consider the limit Pb(ns) −ns/mb γ → ∞. In this case, regulation by sRNA results in an = e (16) Pb(0) instantaneous modification of the distribution of mRNAs just after the burst. The mRNA burst distribution prior Thus, if the burst mean (Pb) is determined along with ns, to interaction with sRNAs is given by πm(m) (Eq. 2). the above relation determines mb and hence the degree The modified mRNA burst distribution after interaction of transcriptional bursting. Eq. 9 further implies that 5

distinguish between the Poisson and Telegraph scenarios. The parameter ranges for validity of the above analy- sis are accessible experimentally based on previous work, e.g. high values of γ relative to the natural degrada- tion rate µm are expected for the sRNA RhyB [24], and mRNA burst arrival rates which are small compared to the mRNA degradation rate have also been reported [14]. Finally, we note that it would be of interest to apply the preceding analysis to systems which show high degree of transcriptional bursting primarily arising from random activation and inactivation of the promoter state [25]. In particular, it was observed [25] that increasing concentra- FIG. 2. The relative error η of the estimated mb value (actual tions of a transcriptional activator resulted in increasing value mb = 10) as a function of mean sRNA levels ns. The the mean burst size rather than affecting the burst fre- estimate from the flat portions of the curves is within 15% error for the parameters shown with γ = 20. The results quency. Since the procedure proposed in this work is an µm shown are for the case of exponential waiting-time distribu- independent approach to determine the burst mean, it tions for mRNA arrival and decay, comparable relative errors would be of interest to further analyze the above system were obtained for more general waiting-time distributions. using the analysis proposed in the current work. the ratio of protein steady-state means for regulated to SUMMARY AND CONCLUSIONS unregulated cases ( Ps(ns) ) is equal to the corresponding Ps(0) ratio for the burst means in Eq. 16. This in turn im- In summary, we have considered a generalized model plies that the mean transcriptional burst size mb can be of gene expression with bursty production of mRNAs determined by considering changes in mean steady-state and proteins. Since very different stochastic processes protein levels. Taken together, these results provide a can lead to steady-state distributions that are experi- novel procedure for determining mb. mentally indistinguishable, the degree of transcriptional The proposed procedure has been computationally val- bursting cannot be inferred from steady-state protein idated for a range of parameters using stochastic sim- distributions. In light of this, it has been argued that ulations (Fig. 2). Specifically, we set up simulations determination of transcriptional bursting requires dy- based on the standard Gillespie algorithm [23] wherein namic measurements of mRNA molecules in single cells the waiting-time for the next reaction is drawn from [11, 12, 26]. In this work, we have derived exact relations an exponential distribution. To consider effects such connecting mRNA/protein burst and steady-state distri- as mRNA senescence, we model mRNA degradation as bution means which are valid for arbitrary gestation and a multi-step process, wherein the waiting-time distribu- senescence waiting-time distributions. We further ana- tion for each step is drawn from an exponential distribu- lyzed how protein burst distributions are modified due tion such that the degradation time for mRNAs follows a to regulation by small RNAs for a range of parameters. Gamma distribution (see Appendix). Similarly, mRNA Our analysis computationally demonstrates an alterna- arrival was simulated as a multi-step process with gamma tive procedure for quantifying transcriptional bursting, waiting-time distribution between mRNA arrival bursts. which involves measurements of changes in mean pro- The output from the simulations is the mean steady-state tein steady-state levels induced by interactions with small protein levels as a function of the mean sRNA levels (ns), RNAs. The strategy presented can also be applied to a where the mean sRNA levels are varied by increasing the broader classes of biological networks whose analysis re- sRNA creation rate ks. Provided that the system pa- quires inference of internal variables from observations rameters are consistent with the following constraints: at higher levels. An alternative strategy to direct mea- γ ≫ µm ≫ µs, simulations indicate that the transcrip- surements of internal variables is to discriminate differ- tional burst size mb can be predicted with reasonable ent possibilities for the internal variables by coupling to accuracy from the ratio of measured protein steady-state a controlled external interaction. means for regulated to unregulated cases as discussed above. The errors in the estimate for mb using the above procedure are related to the validity of the approxima- APPENDIX tions made and are discussed further in the Appendix. Provided that regulatory small RNAs can be designed Finite γ corrections with parameters subject to the constraints noted, the rel- ative error in estimating mb is small and thus we can de- The analysis in the main text considered the limit termine the degree of transcriptional bursting and clearly (γ → ∞) and we now consider corrections due to finite 6

γ values. Lets take a more detailed look at the protein production process during the burst. We denote the du- ration from the beginning of the burst to the time when sRNA or mRNA number first reaches zero as the first stage of the burst. If the mRNAs outnumber the sRNAs, excess mRNAs will be left after the coupled degradation and evolve accordingly. We call the duration from this point to the time when all mRNAs are degraded as stage two of the burst. In the case that γ → ∞, the duration of stage one will be zero and all proteins are produced in stage two of the burst. However, for finite γ value, one has to take into account proteins that have been synthesized during stage one of the burst.

In order to estimate the amount of the proteins FIG. 3. The relative error η of the estimated mb derived from produced on average from mRNAs that are degraded by changes in mean protein burst levels. For different ns and mb sRNAs (stage one), we observe first that the minimal values, the error is negligible when γ ≥ 10µm. degradation rate of a single mRNA in this process is γ. This is because at least one sRNA should be present to ensure coupled degradation. Second, the total amount of mRNAs in the originating burst is greater than or equal to number of mRNAs degraded by sRNAs (since Waiting-time distribution for multi-step processes some mRNAs may decay naturally). Previous work [11] on gestation and senescence effects Hence, we can employ formula Eq.7 in order to es- in mRNA/protein production and decay considered ex- timate contribution of the mRNAs decaying in coupled tensions of the single-step Poisson process to multi-step degradation process to overall steady state level. The processes. For the simplest case, the corresponding wait- upper bound of this contribution is given by ing time distribution is a Gamma distribution as derived k below. Consider a multi-step process, consisting of n hδmi∼ n m . (17) m γ steps such that each step is completed with rate k. Let T denote the random variable corresponding to the waiting- −1 Here we replaced the rate τm in the Eq. 7 by the time for the process to finish and let Ti be the random minimal rate γ in order to estimate the upper limit. variable corresponding to the waiting-time for the ith

step. Thus we have T = i Ti, i.e. T is the sum of Now we can use the expression Eq.8 to get the up- n identical independent randomP variables. Correspond- per bound of the proteins produced on average from the ingly the Laplace transform of the probability distribu- mRNAs during the coupled degradation process tion for T (denoted by F (s) say) is given by the product of n Laplace transforms of the exponential distribution. km th δPs(ns) ∼ δmkpτp ∼ nm kpτp. (18) The exponential waiting-time distribution for the i step  γ  is given by ke−kt with corresponding Laplace transform k k n Hence, the overall ratio of regulated to unregulated k+s . Correspondingly we have F (s) = ( k+s ) , and in- mean steady state levels of proteins is bounded as verting the Laplace transform we obtain that the waiting- time distribution for the multi-step process is given by − −1 e kt δPs(ns) 1 the Gamma distribution: k(kt)n . δR ∼ = , (19) (k−1)! Ps(0) τmγ which is independent of protein’s synthesis rate kp. Therefore, if coupled degradation process is much faster ∗ than natural mRNA decay, τmγ ≫ 1, we obtain δR → 0 [email protected] which is validated by simulations. As we can see in † [email protected] ‡ [email protected] Fig(3), when τmγ > 10 the proteins produced during stage one of the burst can be neglected and the result is [1] M. Kaern, T. C. Elston, W. J. Blake, and J. J. Collins, Nat Rev Genet 6, 451 (2005). almost the same as when γ → ∞. Finally we note that [2] A. Raj and A. van Oudenaarden, Cell 135, 216 (2008). recent studies [24] have shown that a well-studied bacte- [3] S. V. Avery, Nat. Rev. Microbiol. 4, 577 (2006). rial small RNA (RhyB) does induce rapid degradation of [4] D. Fraser and M. Kaern, Mol. Microb. 71, 1333 (2009). target mRNAs consistent with the condition τmγ ≫ 1. [5] J. Paulsson, Phys Of Life Rev 2, 157 (2005). 7

[6] A. Bar-Even, J. Paulsson, N. Maheshri, M. Carmi, [16] D. R. Larson, R. H. Singer, and D. Zenklusen, Trends in E. O’Shea, Y. Pilpel, and N. Barkai, Nat Genet 38, 636 Cell Biology 19, 630 (2009). (2006). [17] L. Waters and G. Storz, Cell 136, 615 (2009). [7] J. R. S. Newman, S. Ghaemmaghami, J. Ihmels, D. K. [18] J. D. C. Little, Operations Research 9, 383 (1961). Breslow, M. Noble, J. L. DeRisi, and J. S. Weissman, [19] E. Levine, Z. Zhang, T. Kuhlman, and T. Hwa, PLoS Nature 441, 840 (2006). Biol 5, e229 (2007). [8] B. B. Kaufmann and A. van Oudenaarden, Curr Opin [20] P. Mehta, S. Goyal, and N. S. Wingreen, Mol Sys Biol 4 Genet Dev 17, 107 (2007). (2008). [9] M. Thattai and A. van Oudenaarden, Proc Natl Acad Sci [21] N. Mitarai, A. M. Andersson, S. Krishna, S. Semsey, and USA 98, 8614 (2001). K. Sneppen, Phys Biol 4, 164 (2007). [10] J. Raser and E. O’Shea, Science 309, 2010 (2005). [22] E. Masse, F. Escorcia, and S. Gottesman, Genes & De- [11] J. M. Pedraza and J. Paulsson, Science 319, 339 (2008). velopment 17, 2374 (2003). [12] I. Golding, J. Paulsson, S. M. Zawilski, and E. C. Cox, [23] D. T. Gillespie, J. Phys. Chem. 81, 2340 (1977). Cell 123, 1025 (2005). [24] N. Mitarai, J. M. Benjamin, S. Krishna, S. Semsey, [13] J. Yu, J. Xiao, X. Ren, K. Lao, and X. S. Xie, Science Z. Csiszovszki, E. Masse, and K. Sneppen, Proc. Natl. 311, 1600 (2006). Acad. Sci. USA 106, 10655 (2009). [14] L. Cai, N. Friedman, and X. S. Xie, Nature 440, 358 [25] A. Raj, C. S. Peskin, D. Tranchina, D. Y. Vargas, and (2006). S. Tyagi, PLoS Biol 4, e309 (2006). [15] P. J. Ingram, M. P. H. Stumpf, and J. Stark, PLoS Comp [26] A. Raj and A. van Oudenaarden, Ann. Rev. Biophys. 38, Biol 4 (2008). 255 (2009).