New Research In
Physical Sciences
Social Sciences
Featured Portals
Articles by Topic
Biological Sciences
Featured Portals
Articles by Topic
 Agricultural Sciences
 Anthropology
 Applied Biological Sciences
 Biochemistry
 Biophysics and Computational Biology
 Cell Biology
 Developmental Biology
 Ecology
 Environmental Sciences
 Evolution
 Genetics
 Immunology and Inflammation
 Medical Sciences
 Microbiology
 Neuroscience
 Pharmacology
 Physiology
 Plant Biology
 Population Biology
 Psychological and Cognitive Sciences
 Sustainability Science
 Systems Biology
The nonequilibrium mechanism for ultrasensitivity in a biological switch: Sensing by Maxwell's demons

Communicated by Charles H. Bennett, IBM Thomas J. Watson Research Center, Yorktown Heights, NY, May 20, 2008 (received for review December 19, 2007)
Abstract
The Escherichia coli flagellar motor senses the intracellular concentration of the response regulator CheYP and responds by varying the bias between its counterclockwise (CCW) and clockwise (CW) rotational states. The response is ultrasensitive with a large Hill coefficient (≈10). Recently, the detailed distribution functions of the CW and the CCW dwell times have been measured for different CW biases. Based on a general result on the properties of the dwelltime statistics for all equilibrium models, we show that the observed dwelltime statistics imply that the flagellar motor switch operates out of equilibrium, with energy dissipation. We propose a dissipative allosteric model that generates dwelltime statistics consistent with the experimental results. Our model reveals a general nonequilibrium mechanism for ultrasensitivity wherein the switch operates with a small energy expenditure to create high sensitivity. In contrast to the conventional equilibrium models, this mechanism does not require one to assume that CheYP binds to the CCW and CW states with different affinities. The estimated energy consumption by the flagellar motor switch suggests that the transmembrane proton motive force, which drives the motor's rotation, may also power its switching. The existence of net transitional fluxes between microscopic states of the switch is predicted, measurement of these fluxes can test the nonequilibrium model directly. Both the results on the general properties of the dwelltime statistics and the mechanism for ultrasensitivity should be useful for understanding a diverse class of physical and biological systems.
High sensitivity has been observed in many signaling systems in biology ranging from calcium signaling in skeletal muscle (1) to chemotaxis response in Escherichia coli (2, 3). One of the most studied systems is the flagellar motor switch, where the key component of the switching complex is a ring of around 34 identical FliM proteins (4). The motor switches stochastically between the CCW and the CW states, with a bias affected by the binding of CheYP to FliM. The conventional picture described the switch by an equilibrium twostate model where the free energies of the two states depend on the CheYP concentration, and the transition is driven by thermal fluctuation (5, 6). The large Hill coefficient in the motor CheYP response curve (7) is then explained in terms of cooperative interactions between the FliM molecules. Models falling within this equilibrium framework include the classical Monod–Wyman–Changeux (MWC) allosteric model (8–10) and the Isingtype model (11) with nearestneighbor interactions. Indeed, cooperative protein interaction has been proposed as a general mechanism for understanding ultrasensitivity in signaling (12). However, most biological complexes, such as the FliM ring, are embedded in and could strongly interact with other components in the system. A fundamental question therefore arises on how good an approximation these equilibrium models are in describing the underlying biological processes. Are there any relevant nonequilibrium effects? If so, how can they be detected and characterized? In this article, we try to address these general questions while also focusing on understanding the mechanism for E. coli flagellar motor's ultrasensitive response to CheYP. Our choice of this particular system is motivated by the recent experimental measurements of detailed single motor switching statistics (13).
Results
Nonexponential DwellTime Statistics and the Breakdown of Detailed Balance.
The observable state of the flagellar motor is represented by a binary variable s: s = 0, 1 corresponds to the CW and CCW rotational states of the motor. The internal state of the switch is described by an integer variable n: n = 0, 1, 2, …, N corresponds to the CheYP occupancy among the N(= 34) FliM monomers. The stochastic switching kinetics is determined by the transition probability rates between these 2 × (n + 1) states as illustrated in Fig. 1. The CheYP binding and unbinding rates for a given state (s, n) are k_{s}^{−}(n) and k_{s}^{+}(n), which depends on the CheYP concentration [Y]. The switching rate from the s state to the (1 − s) state for a given FliM occupancy n is ω_{s}(n). Given these transition rates, the steady state probability p_{s}^{(0)}(n) in state (s, n), the CW (CCW) bias B_{CW} (B_{CCW}), and the distribution function P_{s}(τ) for dwell time τ in the s state can be determined either by numerical simulation or by solving the master equations (see Materials and Methods for details).
For equilibrium systems, detailed balance is satisfied between pairs of states, e.g., ω_{0}(n)p_{0}^{(0)}(n) = ω_{1}(n)p_{1}^{(0)}(n), and, equivalently, the transition rates obey thermodynamic relations, e.g., k_{0}^{+}(n)ω_{0}(n + 1)k_{1}^{−}(n + 1)ω_{1}(n) = k_{0}^{−}(n + 1)ω_{1}(n + 1)k_{1}^{+}(n)ω_{0}(n). By using these relations in analyzing the master equations for P_{s}(τ), we discover that the dwelltime distribution function can be expressed as a sum of exponential decay functions with positive definite coefficients [see supporting information (SI) Text for details of the proof]: We emphasize that this result is valid for any equilibrium model with detailed balance, including the Isingtype model (see SI Text and Fig. S1 for details). From Eq. 1, P_{s}(τ) satisfies a set of constraints: in particular, it should be monotonically decreasing (m = 1) and convex (m = 2). For nonequilibrium systems, some of the coefficients c_{j} can become negative and consequently some of the constraints for P_{s}(τ) can be violated.
This general property of the dwelltime statistics provides a powerful tool in identifying possible nonequilibrium effects. Violation of any constraints in Eq. 2 is sufficient (but not necessary) to imply breakdown of detailed balance in the underlying kinetics. For the E. coli flagellar motor, detailed statistics of the single motor switching kinetics were measured recently by the Cluzel laboratory (13), for a series of given CW biases and at a higher resolution than previous experiments (14). The distribution functions of the CW and CCW dwell times are found to have a distinctive peak at a finite value. The accuracy of the measured P_{s}(τ) for small τ can be affected by experimental resolution as well as ambiguity in defining the CCW and CW states when the system is in the transitional states. However, the switching between CCW and CW states is extremely fast, occurring on the millisecond time scale (15); therefore, the observed peaks in P_{s}(τ) at times of up to 0.4 sec (13) are experimentally very well resolved. According to our general theory on dwelltime statistics, the presence of such peaks in P_{s}(τ) clearly indicates that the flagellar motor switch operates out of equilibrium. Thus, we suggest that the conventional models of flagellar motor switching need to be modified in a fundamental way to reflect the breakdown of thermodynamic relations and the violation of detailed balance.
Dissipative Allosteric Model and the Nonequilibrium Mechanism for High Sensitivity.
To capture the nonequilibrium nature of the switching kinetics, we modify the equilibrium transition rates ω_{s}^{(e)}(n) by a factor γ_{s}(n): ω_{s}(n) = γ_{s}(n)ω_{s}^{(e)}(n) with γ_{0}(n) ≠ γ_{1}(n). k_{B}T ln(γ_{0}(n)) and k_{B}T ln(γ_{1}(n)) can be interpreted as the decreases of the energy barrier from the CW and CCW sides of the doublewell potential respectively (k_{B} is the Boltzmann constant, T is the absolute temperature). The equilibrium model corresponds to the special case when γ_{0}(n) = γ_{1}(n) for all n.
For this article, ω_{s}^{(e)}(n) are taken to be the same as in the equilibrium MWC allosteric model (8) with two dissociation constants K_{d,0} and K_{d,1} for CheYP binding to FliM in the CW and CCW states, respectively. The general behavior of the dissipative allosteric model is studied with the simple choice of constant γ_{0,1}(n) in separate regions of n: γ_{1}(n ≥ n_{U}) = γ, γ_{1}(n < n_{U}) = 1; γ_{0}(n ≤ n_{L}) = γ, γ_{0}(n > n_{L}) = 1 with n_{U} ≥ n_{L}. Other choices of γ_{0,1}(n), such as exponential dependence γ_{0}(n) = γ_{1}(N − n) = b^{N/2−n} with b > 1, do not change the qualitative behaviors of the model (see Fig. S2 for results for other choices of γ_{0,1}(n) studied). The dwelltime statistics and the sensitivity of the switch are determined by solving the master equations for different model parameters. In Fig. 2A, we show the changes of the CCW dwelltime distribution functions P_{1}(τ) as γ increases. For γ = 1 (the equilibrium model), P_{1}(τ) follows an exponentiallike distribution. As γ increases, nonequilibrium characteristics, such as concavity and nonmonotonicity, develop progressively in P_{1}(τ) as predicted from our theory. As shown in Fig. 2B, the average dwell time〈τ〉 and the dwell time τ^{p} at the peak of the dwelltime distribution function for the CCW (CW) state decrease (increase) with the CheYP concentration, in qualitative agreement with the experimental observation (13).
The most important function of the flagellar motor switch is its ultrasensitive response to CheYP. For the equilibrium model (γ = 1), the response to CheYP is driven by the difference in CheYPbinding affinities to the CCW and the CW states, i.e., C ≡ K_{d,0}/K_{d,1} < 1. Quantitatively, a Hill coefficient of 10 in the equilibrium models requires a more than 4fold difference in CheYP binding (C ≤ 0.25). Despite much effort, this has not been observed (16). In contrast, for the nonequilibrium model even when CheYP binds equally to FliM in CCW and CW states (C = 1), highly sensitive response to CheYP develops for large values of γ as shown in Fig. 3A. The sensing of CheYP in the nonequilibrium model depends on the two amplification factors γ_{0}(n) and γ_{1}(n). Intuitively, these two nonequilibrium factors can be understood as resulting from the actions of two Maxwell demons (17, 18). In the context of our model, the two demons operate to irreversibly enhance the CW → CCW or CCW → CW transition rates beyond their equilibrium values in different regions of the switch's CheYP occupancy state space (n ≤ n_{L} or n ≥ n_{U}, respectively). Their opposing actions with different dependence on CheYP level give rise to the dependence of the switch's CW bias on CheYP concentration. Effectively, the two demons act as the switch's sensors for the CheYP concentration. A highly sensitive switch can be made, provided γ is large, and the operation regions for the two opposing demons are well separated (large Δn ≡ n_{U} − n_{L}).
The second law of thermodynamics prescribes that the Maxwell demons consume energy to carry out their work, i.e., to sense and respond to CheYP. The rate of energy consumption Ẇ can be calculated from our model (19, 20): where the sum is over all pairs of states 〈AB〉, J_{AB} is the probability flux from state A to state B. The net flux from A to B: ΔJ_{AB} ≡ J_{AB} − J_{BA} is nonzero only in nonequilibrium systems because of the breakdown of detailed balance. In fact, measurements of this net flux can be used to directly test the validity of the nonequilibrium models (see SI Text for a detailed calculation of these fluxes). The average energy dissipated per switch cycle (CCW to CW to CCW) is ΔW = (τ_{1} + τ_{0})Ẇ, where τ_{0} and τ_{1} are the average dwell time for CW and CCW states, respectively. In Fig. 3B, the energy dissipation ΔW/k_{B}T and the Hill coefficient of the corresponding response curve H_{eff} ≡ 2 × d(lnB_{CW})/d(ln[Y])_{BCW} = 0.5 for given values of γ are plotted together. Our result clearly shows that in our nonequilibrium model, sensitivity is powered by energy dissipation, and more energy is needed to generate higher sensitivity (H_{eff}). Eventually, H_{eff} is limited by the size of the operation range Δn, as shown in Fig. 4, where H_{eff} is plotted versus the energy dissipation ΔW/k_{B}T for different values of Δn.
Despite the different origins of the equilibrium (γ = 1, C < 1) and the nonequilibrium (γ > 1) mechanisms for high sensitivity, they can coexist as shown in Fig. 5A, where H_{eff} is shown for different values of C and ΔW (by varying γ). For a given C, H_{eff} increases from its equilibrium value at ΔW = 0 (γ = 1) as the energy dissipation ΔW increases. However, although both mechanisms can generate high sensitivity (black lines in Fig. 5B), their predicted dependence of average FliM occupancy 〈n〉 on CheYP concentration differs significantly. The CheYP occupancy curve for the equilibrium model (C = 0.25, γ = 1) has a steep region in the operating range of the CheYP concentration, whereas the FliM occupancy for the nonequilibrium model (C = 1, γ = 256) follows a gradual binding function
Discussion
The general theory on the properties of the dwelltime distribution obtained in this article holds true for all equilibrium systems with detailed balance. This theory provides a powerful tool in detecting relevant nonequilibrium effects in all systems and should be particularly useful in biology, where most interesting systems operate out of equilibrium. The nonequilibrium characteristics in the dwelltime distribution function, such as nonmonotonicity (peak), concavity, and other higherorder behaviors, generally appear at time scales much smaller than the average dwell time. Therefore, highresolution experiments are required to unravel these subtle but highly informative features. Depending on the strength of the nonequilibrium effects, extra care may also be needed to control the signal to prevent weakening of these features by averaging over different signal strengths. For example in the case of the E. coli flagellar motor switch, peaks in P_{s}(τ) became visible only when P_{s}(τ) were determined for fixed CW biases (13). There is a recent report of a similar peak in dwelltime distribution in the kinesin motor for its waitingtime (between steps) distribution at low ATP concentration (21); it would be interesting to study the implication of these observations in light of our work. The strong connection between the dwelltime statistic and the underlying kinetics established here will hopefully stimulate more such careful measurements of P_{s}(τ), particularly for small dwell times, for motors as well as other biological systems.
For the E. coli flagellar motor, the nonequilibrium effects are not only necessary in explaining the observed dwelltime statistics; more importantly, they provide a mechanism by which the system dissipates a small amount of energy to produce a highly sensitive, fast response to the signal (CheYP). To generate the observed high sensitivity for a single flagellar motor, the amount of energy per switch cycle (CCW–CW–CCW) dissipated is in the range of 7–12 k_{B}T for a variety of detailed models (see Fig. 4 and Fig. S3). Interestingly, this small switching energy is roughly equal to the work done by one to two protons moving through a membrane potential of −170 mV at room temperature, so the energy source for the switch could be the same proton gradient that drives the flagellar motor rotation. Therefore, in addition to the spontaneous switching (controlled solely by CheYP level), we propose a protonassisted switching mechanism. The energy released by each proton passage through the membrane, instead of always being converted to mechanical motion, may also (with a small probability) be used in carrying out switching as illustrated in Fig. 6. The probability of this protonassisted switching can depend on the internal state of the switch, i.e., its rotational sense (CW or CCW) and the CheYP occupancy of the FliM, this dependence being described by γ_{0,1}(n) in our model. The existence of this protonassisted switching is supported by the important observation by Fahrner, Ryu, and Berg (22) that the average switching frequency depends on the rotational speed of the motor or, equivalently, the proton flux. Molecularly, the FliM ring can tap into the proton energy source through its coupling to the FliG ring (23, 24), which interacts directly with the stator (MotA/MotB) complex, where protons pass through the membrane. Detailed information on these interactions (FliM/FliG/MotAB/proton, etc.) is needed to elucidate the molecular (energyconsuming) switching kinetics. Quantitative switching statistics, such as those measured in ref. 13 but with different loads and proton motive forces, will help determine the quantitative forms of γ_{0,1}(n) in our model, their dependence on the proton flux, and the molecular origin of these Maxwell demons. However, independently of the molecular details, our study here clearly demonstrate that the flagellar switch does not operate as an equilibrium apparatus isolated from the rest of the motor as previously conceived. The general nonequilibrium model framework proposed here provides a crucial step toward understanding the switch as a part of the integrated motor system.
One important general feature of the nonequilibrium mechanism for high sensitivity is that it does not require different binding affinities of the signaling molecule to the two different states of the switch, in contrast to the classical equilibrium allosteric mechanism. The measurement of the ligandbinding curve for a single motor in CCW and CW state can therefore be used to distinguish between these two mechanisms. Our nonequilibrium model also predicts the existence of nonzero net transitional fluxes between any pair of microscopic states of the switch; in vivo CheYP/FliMbinding measurements together with the motor switching measurements for the same single motor can be used to measure the net transitional fluxes and test microscopically whether detailed balance is violated in the switching kinetics. Besides the flagellar motor, we speculate that this more robust [details of γ_{0,1}(n) do not matter] nonequilibrium mechanism for creating highly sensitive switches with a small energy expenditure may be preferred by a large class of cellular processes to an equilibrium mechanism that requires fine tuning of the binding affinity ratio and the allosteric interaction strength (11). Many signaling systems, especially those with allosteric interactions, such as Ca signaling for cardiac filament activation (25) and signal transduction by MCP chemoreceptor clusters (3, 26–28), should be reexamined in light of the nonequilibrium model developed here.
Materials and Methods
Master Equations for Determining the DwellTime Distribution Functions.
The master equation for the probability p_{s}(n,t) of being in state (s, n) at time t is where s = 0, 1 and n = 0, 1, 2, …, N. The steadystate distribution p_{s}^{(0)}(n) is obtained by setting the right hand side of Eq. 4 to 0. The CW bias B_{CW} is derived from _{s}^{(0)}((n): To obtain the dwelltime distribution functions for CW and CCW, assume that the motor switches from the (1 − s) state to the s state at time t = 0, and define Q_{s}(n, τ) as the subsequent “survival” probability that the motor is still in the s state with n occupied FliM at time t = τ > 0. The master equation for Q_{s}(n,τ) is s = 0,1. The survival probability Q_{s}(n, τ) can be solved from the above equation with the initial condition determined by the steadystate distribution of the (1 − s) state: Q_{s}(n, τ = 0) = A_{s}ω_{1−s}(n)p_{1−s}^{(0)}(n), where A_{s} is normalization factor A_{s} = (Σ_{n=0}^{N}ω_{1−s}(n)p_{1−s}^{(0)}(n))^{−1}.
Eqs. 4 and 5 are solved by using standard linear algebra and eigenvalue/eigenvector analysis for any given choices of model parameters. Finally, from Q_{s}(n, τ), the distribution function for dwell time τ can be obtained:
Details of the Kinetic MWC Model.
The equilibrium part of the dissipative allosteric model studied here is the same as the Monod–Wyman–Changeux (MWC) model (8). Following the energetic representation of the MWC model (29), the free energy (in units of k_{B}T) for state (s, n) is given by where E is the energy difference between the CCW and CW states when n = 0. We set E = N/2ln C, known to be optimum in generating high sensitivity (11, 29). The free energy G_{m}(n) for the intermediate state is chosen in this article to have the same simple dependence on n and [Y]: where A_{m}, B_{m} are constants, chosen to be A_{m} = ½, B_{m} = ½ in this article. The equilibrium switching rate ω_{s}^{(e)}(n) = ω̄ exp(−ΔG_{s}) depends on the freeenergy difference: ΔG_{s}(n) ≡ −G_{s}(n) + G_{m}(n), where mac]ω is the (geometrical) average of all these equilibrium rates given the way we choose the intermediatestate free energy. For independent CheYP/FliM binding (30), we have: k_{s}^{+}(n) = (N − n)k_{s}^{on}, k_{s}^{−}(n) = nk_{s}off, where k_{0,1}^{off} are the on and off rates for individual CheYP to FliM binding in the CW and CCW states. These rates are related to the dissociation constants K_{d,s} and the CheYP concentration [Y]: k_{s}^{on}/k_{s}^{off} = [Y];K_{d,s} We let the off rates be k_{s}^{off} = 1 (for both CCW and CW) to set the (fast) time scale for our model. We also let the dissociation constant for CW be K_{d,0} = 1 to set the scale for CheYP concentration [Y]. The variable parameters for the equilibrium part of the model studied in this article are ω̄ ≪ 1 and C ≤ 1.
Net Transition Fluxes and Energy Dissipation in the Dissipative MWC Model.
One distinctive feature of a nonequilibrium system is the existence of a nonzero net transition flux ΔJ_{AB} ≠ 0 from state A to state B in the system because of breakdown of detailed balance. Quantitatively, the net flux from state (s, n) to state (s, n + 1) is: The net CCW to CW switching flux (Fig. S4) from state (1, n) to state (0, n) is: the net flux from CW to CCW is just the opposite: ΔJ_{0}^{(v)}(n) = −ΔJ_{1}^{(v)}(n). The total energy dissipated in the system can be decomposed into terms for each link in the transitional network. For the links between states (s, n) and (s, n + 1), the energy dissipation rate is whereas for links between states (1, n) and (0, n), the energy dissipation rate is Summing over all of the links in the system, we obtain the energy dissipation rate for the whole system: Ẇ = Σ_{s,n}δ̇ w_{s}^{(h)}(n) + Σ_{n}δ̇ w^{(v)}(n).
Acknowledgments
I thank Drs. G. Grinstein, J. Rice, and G. Stolovitzky for helpful discussions and Dr. Klaus Schwarz for careful reading of the manuscript. This work was partially supported by National Science Foundation Grant CCF0635134.
Footnotes
 *Email: yuhai{at}us.ibm.com

Author contributions: Y.T. designed research, performed research, analyzed data, and wrote the paper.

The author declares no conflict of interest.

This article contains supporting information online at www.pnas.org/cgi/content/full/0804641105/DCSupplemental.
 © 2008 by The National Academy of Sciences of the USA
References
 ↵
 ↵
 Segall JE,
 Block SM,
 Berg HC
 ↵
 ↵
 ↵
 ↵
 Scharf BE,
 Fahrner KA,
 Turner L,
 Berg HC
 ↵
 Cluzel P,
 Surette M,
 Leibler S
 ↵
 ↵
 Alon U,
 et al.
 ↵
 ↵
 ↵
 ↵
 Korobkova E,
 Emonet T,
 Park H,
 Cluzel P
 ↵
 Block SM,
 Segall JE,
 Berg HC
 ↵
 ↵
 Sourjik V,
 Berg HC
 ↵
 Maxwell JC
 ↵
 ↵
 Hill TL
 ↵
 ↵
 ↵
 ↵
 Park SY,
 Lowder B,
 Bilwes AM,
 Blair DF,
 Crane BR
 ↵
 Brown PN,
 Terrazas M,
 Paul K,
 Blair DF
 ↵
 ↵
 Duke TAJ,
 Bray D
 ↵
 Mello BA,
 Tu Y
 ↵
 ↵
 Mello BA,
 Tu Y
 ↵
 Sagi Y,
 Khan S,
 Eisenbach M
Citation Manager Formats
More Articles of This Classification
Physical Sciences
Statistics
Biological Sciences
Biophysics
Related Content
 No related articles found.
Cited by...
 Designing the optimal bit: balancing energetic cost, speed and reliability
 Structure of silent transcription intervals and noise characteristics of mammalian genes
 Switching dynamics of the bacterial flagellar motor near zero load
 Reconciling the concurrent fast and slow cycling of proteins on gene promoters
 Energetic costs of cellular computation
 Stochastic coordination of multiple actuators reduces latency and improves chemotactic response in bacteria
 Dynamics of the bacterial flagellar motor with multiple stators
 The switching dynamics of the bacterial flagellar motor