Circadian clocks optimally adapt to sunlight for reliable synchronization.  
Jump to Full Text  
MedLine Citation:

PMID: 24352677 Owner: NLM Status: InDataReview 
Abstract/OtherAbstract:

Circadian oscillation provides selection advantages through synchronization to the daylight cycle. However, a reliable clock must be designed through two conflicting properties: entrainability to synchronize internal time with periodic stimuli such as sunlight, and regularity to oscillate with a precise period. These two aspects do not easily coexist, because better entrainability favours higher sensitivity which may sacrifice regularity. To investigate conditions for satisfying the two properties, we analytically calculated the optimal phaseresponse curve with a variational method. Our results indicate an existence of a dead zone, i.e. a time period during which input stimuli neither advance nor delay the clock. A dead zone appears only when input stimuli obey the time course of actual solar radiation, but a simple sine curve cannot yield a dead zone. Our calculation demonstrates that every circadian clock with a dead zone is optimally adapted to the daylight cycle. 
Authors:

Yoshihiko Hasegawa; Masanori Arita 
Related Documents
:

24272867  Changes of the diurnal and circadian (endogenous) mrna oscillations of the chlorophyll ... 24352677  Circadian clocks optimally adapt to sunlight for reliable synchronization. 21764627  Estimation of slipping organ motion by registration with directiondependent regulariza... 23485867  Entrainment of the circadian clock by daily ambient temperature cycles in the camel (ca... 19894837  Matching the waveform and the temporal window in the creation of experimental signals. 23019817  Application of ultraviolet lightemitting diode photocatalysis to remove volatile organ... 
Publication Detail:

Type: Journal Article Date: 20131218 
Journal Detail:

Title: Journal of the Royal Society, Interface / the Royal Society Volume: 11 ISSN: 17425662 ISO Abbreviation: J R Soc Interface Publication Date: 2014 
Date Detail:

Created Date: 20131219 Completed Date:  Revised Date:  
Medline Journal Info:

Nlm Unique ID: 101217269 Medline TA: J R Soc Interface Country: England 
Other Details:

Languages: eng Pagination: 20131018 Citation Subset: IM 
Export Citation:

APA/MLA Format Download EndNote Download BibTex 
MeSH Terms  
Descriptor/Qualifier:

Full Text  
Journal Information Journal ID (nlmta): J R Soc Interface Journal ID (isoabbrev): J R Soc Interface Journal ID (publisherid): RSIF Journal ID (hwp): royinterface ISSN: 17425689 ISSN: 17425662 Publisher: The Royal Society 
Article Information Download PDF openaccess: Received Day: 4 Month: 11 Year: 2013 Accepted Day: 27 Month: 11 Year: 2013 Print publication date: Day: 6 Month: 3 Year: 2014 pmcrelease publication date: Day: 6 Month: 3 Year: 2014 Volume: 11 Issue: 92 Elocation ID: 20131018 PubMed Id: 24352677 ID: 3899870 DOI: 10.1098/rsif.2013.1018 Publisher Id: rsif20131018 
Circadian clocks optimally adapt to sunlight for reliable synchronization Alternate Title:Circadian clocks optimally adapt to sunlight for reliable synchronization  
Yoshihiko Hasegawa1  
Masanori Arita12†  
1Department of Biophysics and Biochemistry, Graduate School of Science, The University of Tokyo, Tokyo 1130033, Japan 

2RIKEN Center for Sustainable Resource Science, Kanagawa 2300045, Japan 

Correspondence: email: hasegawa@cb.k.utokyo.ac.jp †Present address: Center for Information Biology, National Institute of Genetics, Shizuoka 4118540, Japan. 
Circadian oscillators are prevalent in organisms from bacteria to humans and serve to synchronize bodies with the environmental 24 h cycle [^{1},^{2}]. Although the molecular implementation of oscillation is speciesspecific [^{3}–^{6}], every circadian clocks satisfies two requirements to achieve reliable synchronization to the environment: entrainability to synchronize internal time with periodic stimuli and regularity to oscillate with a precise period. Circadian clocks are acquired through evolution independently in bacteria, fungi, plants and animals [^{7}]. Nonetheless, entrainability and regularity constitute major characteristics conserved in all circadian clocks [^{6}], which strongly suggest that these two properties are essential for survival. A main source of interference with regularity is discreteness of molecular species, i.e. molecular noise [^{8}–^{13}]. Many studies have analysed the resistance mechanisms of circadian oscillators against noise [^{14}–^{17}]. Regarding entrainability, circadian clocks synchronize their internal time with the environmental cycle via sunlight, and its effect depends on the wavelength or fluence, as well as on the phase of the stimulation. However, entrainability and regularity are conflicting factors, because circadian clocks with better entrainability are sensitive not only to the periodic light stimuli, but also to the molecular noise which interferes with regularity.
Because both regularity and entrainability are important adaptive values, we expect actual circadian oscillators to optimally satisfy these two factors (figure 1). Here, we investigate the optimal phase–response curve (PRC), which is both entrainable and regular, in the phase oscillator model [^{18}] by using the Euler–Lagrange variational method. Our main finding is the inherent existence of a dead zone in the PRC: optimality is achieved only when the PRCs have a time period during which light stimuli neither advance nor delay the clock (figure 2a). In other words, a PRC with a dead zone (figure 2a) is better adapted than those without a dead zone (figure 2b). This result is intriguing, because a dead zone, with which oscillators tend to be unaffected by stimuli (i.e. lower entrainability), achieves better entrainability. We also tested this with two types of input stimuli: a solar radiationtype input that simulated the time course of solar radiation intensity (cf. equation (2.24) and figure 4a) and a simple sinusoidal input (sine curve). Surprisingly, the dead zone in the optimal PRC emerges only for the solar radiationtype input, not for the sinusoidal input. Many experimental studies reported the existence of a dead zone in various species (figure 2c,d show experimentally observed PRCs of fruitfly [^{19}] and mouse [^{20}], respectively). Our results indicate that circadian oscillators in various species have adapted to solar radiation for reliable synchronization.
Circadian oscillators basically comprise interaction between mRNAs and proteins, whose dynamics can be modelled by differential equations. A circadian oscillator of N molecular species can be represented by
where the Ndimensional vector x = (x_{1}, x_{2}, … ,x_{N}) denotes the concentration of molecular species (mRNAs or proteins). The effect of noise on genetic oscillators has been a subject of considerable interest, and noiseresistant mechanisms have been extensively studied [^{14}–^{17},^{21}–^{23}]. In general, the dynamics of the ith molecular concentration in a circadian oscillator subjected to molecular noise is described by the following Langevin equation (Stratonovich interpretation):where Q_{i}(x) is an arbitrary function representing the multiplicative terms of the noise, ξ_{i}(t) is white Gaussian noise with the correlation (a bracket denotes expectation), and ρ is a model parameter.Circadian oscillators synchronize to environmental cycles by responding to a periodic input signal (light stimuli). We let ρ in equation (2.2) be stimulated by the input signal: for example, ρ can be the degradation rate (for the sake of simplicity, we consider that the input signal affects only one parameter). We use equation (2.2) for calculating regularity and entrainability of circadian oscillators.
Because the circadian oscillator of equation (2.2) is subjected to noise, its period varies cycle to cycle. We use the term regularity for the period variance of the oscillation (higher regularity corresponds to smaller period variance). Let us first consider the case without input signals (i.e. ρ is constant). As equation (2.1) exhibits periodic oscillation, we can naturally define the phase ϕ ∈ [0,2π) on equation (2.1) by
where Ω = 2π/T is the angular frequency of the oscillation (T is a period of the oscillation). The phase ϕ in equation (2.3) is defined only on a closed orbit of the unperturbed limitcycle oscillation. However, we can expand the definition into the entire x = (x_{1}, x_{2}, … ,x_{N}) space, where the equiphase surface is referred to as the isochron ℐ(ϕ) (figure 3a). By using standard stochastic phase reduction [^{18}], equation (2.2) can be transformed into the following Langevin equation with respect to the phase variable ϕ (Stratonovich interpretation):where U(ϕ) = (U_{1}(ϕ), … ,U_{N}(ϕ)) is an infinitesimal PRC (iPRC) , and we abbreviated Q_{i}(x_{LC}(ϕ)) as Q_{i}(ϕ). iPRC U_{i}(ϕ) quantifies the extent of phase advance or delay when perturbed along an x_{i} coordinate direction at phase ϕ. The Ndimensional vector x_{LC}(ϕ) denotes a point on the limitcycle trajectory at phase ϕ, where LC stands for limit cycle. The value of iPRC U_{i}(ϕ) is calculated as a solution of an adjoint equation [^{24}] or as the set of eigenvectors of a monodromy matrix in the Floquet theory [^{18}] for arbitrary oscillators. Let P(ϕ; t) be the probability density function of ϕ at time t. From equation (2.4), the Fokker–Planck equation (FPE) [^{25}] of P(ϕ; t) is given bywhereandIntroducing a slow variable φ = ϕ – Ωt, the FPE of the probability density function Π(φ; t) = P(ϕ = φ + Ωt; t) is given byWith sufficiently weak noise, Π(φ; t) is a slowly fluctuating function of t. In such cases, ℱ(φ + Ωt) and 𝒢(φ + Ωt) fluctuate much faster than Π(φ; t), thus these two terms can be averaged for one period while keeping Π(φ; t) constant (phase averaging). In other words, we separate time scales between ℱ(φ + Ωt), 𝒢(φ + Ωt) and Π(φ; t). By phase averaging, ℱ(φ + Ωt) vanishes because of the periodicity (use integration by parts), yielding
withSee Kuramoto [^{18}] for further details of stochastic phase reduction and the phaseaveraging procedure. From equation (2.9), because Π(φ = ϕ–Ωt; t)[ = P(ϕ; t)] obeys a simple onedimensional diffusion equation, its solution is represented byEquation (2.11) shows that the variance of the phase after one period T is
In equation (2.4), the average period corresponds to the mean first passage time with ϕ starting from 0 to 2π, and the period variance is the variance of the first passage time. Because direct calculation of the period variance is difficult, we approximate the period variance with the phase variance , after Kori et al. [^{26}]. As the phase ϕ crosses a threshold ϕ = 2π with gradient 2π/T without noise, there is a scaling relation for sufficiently weak noise [^{26}] (figure 3b). Consequently, the variance of the period is approximated by
The entrainment property is an important characteristic of limitcycle oscillators and attracts attention in systems biology [^{27}–^{32}]. For instance, a period mismatch in coupled oscillators is known to enhance entrainability in genetic oscillators [^{31}]. Light stimuli affect the rate constants, i.e. the parameter ρ in equation (2.2) is perturbed as ρ + dρ by the input signal. Phase dynamics of equation (2.2) can be viewed as representing that of a tilted periodic potential (i.e. ratchet) subjected to noise. Because a synchronizable condition corresponds to the existence of stable points in the ratchetlike potential, the entrainability can be discussed without considering the noise. Consequently, in contrast to the calculation of regularity, in the evaluation of the entrainability, we consider a case without molecular noise (i.e. Q_{i}(x) = 0 in equation (2.2)).
Let p(ωt) be an input signal with angular frequency ω. Considering a weak periodic input signal dρ = χp(ωt), where χ is the signal strength (χ ≥ 0), and applying the phase reduction approach to equation (2.2), the time evolution of the phase variable ϕ is given by
with F_{i}(ϕ; ρ) = F_{i}(x_{LC}(ϕ); ρ) andwhere Z(ϕ) is the PRC with respect to the parameter ρ and corresponds to experimentally observed PRCs. In order to distinguish Z(ϕ) from iPRC U_{i}(ϕ), we will refer to Z(ϕ) as the parametric PRC (pPRC) [^{33}]. Note that the definition of measured PRCs is different from pPRCs Z(ϕ) in a rigorous definition; the experimentally measured PRCs quantify the phase shift Δϕ caused by light stimuli, whereas pPRCs Z(ϕ) are normalized by the strength of perturbation, i.e. . Therefore, the ranges of the measured pPRCs have limitation −π ≤ Δϕ < π, whereas pPRCs Z(ϕ) do not. The phase reduction can yield reliable results only when the perturbed trajectory is close to the unperturbed limitcycle trajectory (i.e. χ is sufficiently small).We next evaluate the extent of synchronization to the periodic input signal. By introducing another slow variable ψ = ϕ–ωt in equation (2.13), we can again apply the phaseaveraging procedure, which yields
with ΔΩ = Ω – ω andThe oscillator of interest can synchronize to input signals when there is a stable solution of ψ in (equation (2.15)). The stable solution is an intersection point of Θ(ψ) and –ΔΩ with dΘ/dψ < 0 (an empty circle in figure 3c). Then, a condition for the existence of a stable solution iswhere and .We define entrainability, the extent of synchronization to the periodic input signal, by the width of the Arnold tongue, which is a domain with respect to χ (signal strength) and ω (signal angular frequency). The shaded region in figure 3d represents the Arnold tongue; with parameters χ and ω inside the Arnold tongue, the oscillator can synchronize to a periodic input signal. Because equation (2.17) constitutes a linear approximation of the Arnold tongue for sufficiently small χ, the width of the Arnold tongue is given by χ(Θ(ψ_{M}) – Θ(ψ_{m})) under the linear approximation. Thus, we define the entrainability ℰ, or the extent of synchronization, as
Intuitively, a circadian oscillator with better entrainability (i.e. larger ℰ) can synchronize to an input signal that has a period further from that of the oscillator. The calculation above is standard in the phase reduction approach, and further details are available in Kuramoto [^{18}].
We use the variational method to calculate the optimal PRCs which maximize the entrainability ℰ subject to constant variance (the optimal solutions correspond to the edge in figure 1, which is described by the thickdashed line). The constrained optimization of U_{i}(ϕ) can be intuitively interpreted as maximization of weighted area (equation (2.18)), where the input being the weight, with constant area under the squared magnitude (equation (2.12)). In simple terms, the optimality is reached when the magnitude of the PRC is small during intervals when the input magnitude is small (and vice versa). In the context of neuronal oscillators, a study [^{34}] has used the variational method to calculate the optimal PRCs for stochastic synchrony (noiseinduced synchronization [^{35},^{36}]).
The variational equation to be optimized is
where λ is the Lagrange multiplier. Note that variational equation (2.19) is similar to Harada et al. [^{37}], which optimizes the input signal for the maximal entrainment under constant power of the input. The variational condition δℒ[U] = 0 yields the optimal iPRCand the pPRC is calculated with equation (2.14):Because ψ_{M} and ψ_{m} themselves depend on U_{i}(ϕ), they have to satisfy a selfconsistent condition, i.e. equation (2.18) is maximal with ψ_{M} and ψ_{m}. Consequently, we maximize the following function:withwhere Δ = ψ_{M} – ψ_{m} and δ = ψ_{m}. The optimal iPRC can be obtained by first finding the maximum solution of Ψ(Δ, δ) with respect to Δ and δ, and then substituting the obtained solution ψ_{m} = δ and ψ_{M} = δ + Δ into equations (2.20) and (2.21).Optimal PRCs depend on input signals, as seen in equations (2.20) and (2.21). The most common synchronizer in circadian oscillators is sunlight, for which the strength is determined by 24 hperiodic solar irradiance. The solar irradiance is calculated by I = I_{0} cos ϑ and I = 0 when the sun is above the horizon (0 ≤ ϑ < π) and below the horizon (π ≤ ϑ < 2π), respectively, where ϑ is the zenith angle and I_{0} is the maximum irradiance [^{38}]. It can be approximated by
where ramp(x) is the ramp function defined by ramp(x) = x for x ≥ 0 and ramp(x) = 0 for x < 0. We call equation (2.24) the solar radiation input, whose plot is shown in figure 4a (the shaded region represents night). In order to show the validity of the solar radiation modelling, we compare equation (2.24) with observed irradiance data from Vick & Moss [^{39}], which are shown in a dual axis plot of figure 4b. In figure 4b, equation (2.24) is plotted by the solid line (left axis) and the observed data by the dashed line (right axis). The solar radiation input of equation (2.24) is shifted horizontally, so that equation (2.24) becomes a good fit to the data. From figure 4b, the solar radiation input is in good agreement with the observed data, which verifies the validity of equation (2.24) as a solar radiation model.For comparison, we also use a sinusoidal input, which is common in nonlinear sciences:
Note that p(ωt) = B + sin(ωt), where B is an arbitrary constant, also yields the same optimal PRCs as equation (2.25), because a constant B in the signal is offset in equations (2.20)–(2.23). Although a constant B does not play any roles in formation of the optimal PRCs, different B result in different Arnold tongues in general. For calculating the optimal PRCs, we use equations (2.24) and (2.25).
Light stimuli generally affect the oscillatory dynamics multiplicatively, i.e. they act on the rate constants or transcriptional efficiency of the gene regulatory circuits [^{3},^{40}]. We assume that the jth molecular species includes a parameter ρ as
where represents the terms that do not include ρ, and x_{k} is the concentration of the kth molecular species. Here, k∈{1,2, … ,N} can take any value, regardless of j (both j ≠ k and j = k are allowed). For example, let figure 4c be a gene regulatory circuit of a hypothetical circadian clock, where symbols → and ⊣ represent positive and negative regulations and x_{i} are molecular species (see Novák & Tyson [^{41}] for typical motifs of biochemical oscillators). Suppose x_{1} and x_{2} are mRNA and corresponding protein, respectively, and light stimuli increase the translational efficiency. In this case, the dynamics of light entrainment can be described by equation (3.1) with j = 2, k = 1 and ρ being the translation rate. In equation (3.1), although we can also consider an alternative case (a negative sign), the optimal pPRCs remain unchanged under the inversion which is seen from equations (2.21) and (2.23). Consequently, we consider only the positive case to calculate the optimal PRCs (i.e. equation (3.1)). However, note that relations between iPRCs and pPRCs are affected by the inversion of the sign, and the difference matters when considering biological feasibility.When using phase reduction, the dynamics of the limit cycle are considered on the unperturbed limitcycle trajectories x_{LC}, and hence the points on the limit cycle can be uniquely determined by the phase ϕ. Consequently, under the phase reduction, x_{k} is replaced by x_{LC,k}(ϕ) in equation (3.1), where x_{LC,k}(ϕ) is the kth coordinate of x_{LC} (i.e. in equation (2.20)). Here, corresponds to the time course of the concentration of the kth molecular species. Because x_{LC,k}(ϕ) constitutes a core clock component and is generally a smooth 2πperiodic function, we approximate it with a sinusoidal function:
where u is the initial phase and α denotes the amplitude of the oscillation (figure 4d). To ensure , we set 0 ≤ α ≤ 1, and α = 0 recovers the additive case. Because the initial phase u does not play any role (u is offset by δ in equation (2.23)) when the white Gaussian noise is additive (i.e. Q_{i}(x) ∝ 1), we also set u = 0. The parametric approximation of equation (3.2) enables an almost closed form for the overall calculations. Although we assumed in equation (3.1) that effects of ρ only depend on x_{k}, we can generalize equation (3.1) to , where K(x) is a nonlinear function and is assumed to be well approximated by 1 – αsin(ϕ + u). By this generalization, our theory can be applied to other possible light entrainment mechanisms such as the intercellular coupling [^{42}]. Our model needs only details about molecular species which have light input entry points but not about a whole molecular network. However, this advantage, in turn, means that we cannot specify iPRCs U_{i}(ϕ) of molecular species not having light input entry points. Consequently, for a noise term Q_{i}(x), we assume that the white Gaussian noise is additive and is present only in the jth coordinate equation ( , where q is the noise intensity and Q_{i}(ϕ) = 0 for i ≠ j).Figure 5a–c shows the landscape of Ψ(Δ,δ) as functions of Δ and δ, and figure 5d–f expresses the optimal iPRCs U_{j}(ϕ) and pPRCs Z(ϕ) for the solar radiation input (an explicit expression of Ψ(Δ, δ) is given in appendix A). The optimal PRC shape does not depend on the model parameters such as the period T, its variance , or noise intensity q. These three parameters only act on the magnitude of the PRCs (i.e. the vertical scaling of the PRCs). Consequently, we normalized T = 1, , and q = 1, as shown in figure 5. As the optimal PRCs depend on α, Ψ(Δ, δ) is plotted for three cases: α = 0, (figure 5a), α = 0.5 (figure 5b) and α = 1.0 (figure 5c), where the maximal points (Δ, δ) yield the optimal PRCs using equations (2.20) and (2.21). The maximal parameters Δ and δ are calculated numerically. Figure 5d–f describes the optimal iPRCs (solid line) and pPRCs (dashed line) for α = 0, 0.5 and 1.0, respectively. When α = 0, i.e. the input signal is additive, Ψ(Δ,δ) achieves a maximum for Δ = π and arbitrary δ, yielding sinusoidal PRCs as the optimal solution (figure 5d). Although the input signal p(ϕ) is not sinusoidal, the optimal PRCs obtained using the variational method become sinusoidal. In other words, considering optimality, resonatortype oscillators have an advantage over integratortype oscillators. For α > 0, the input signal p(ϕ) depends on the concentration of the kth molecular species. From figure 5b, the optimal parameters for α = 0.5 are (Δ, δ) = (2.31, 1.99) and (3.98, 4.30), which are different from Δ = π (these two sets yield symmetric PRCs with respect to the horizontal axis). Figure 5e shows the optimal iPRCs U_{j}(ϕ) and pPRCs Z(ϕ) for α = 0.5. Interestingly, the optimal iPRCs and pPRCs for α = 0.5 have a dead zone (region of in figure 5e) in which the input signal neither advances nor delays the clock. From equations (2.20)–(2.21) and the solar radiation input of equation (2.24), the optimal PRCs inevitably include a dead zone if the optimal Δ is not π. For α = 1.0, there are four sets of parameters (Δ, δ) that give optimal PRCs: (2.30, 2.72), (2.30, 1.26), (3.98, 3.56) and (3.98, 5.02) (PRCs with these four sets are symmetric with each other with respect to the horizontal axis or ϕ = 3π/2). Consequently, the optimal PRCs shown in figure 5f have a dead zone as in the case of α = 0.5.
From the results discussed above, the optimal PRCs have a dead zone when α > 0. We next studied the length of the dead zone as a function of α (figure 6a) and improvements in the entrainability induced by the dead zone (figure 6b) for the solar radiation input. Because the dead zone, which is a null interval in PRCs, emerges when the optimal parameter is Δ ≠ π, we can naturally define its length as
where Δ is the maximum value of Ψ(Δ, δ). As seen in figure 6a, a dead zone clearly exists when α > 0, and the length increases with increasing α for α < 0.8. Even for α = 0.1, when the oscillation amplitude of x_{LC,k}(ϕ) (the concentration of a molecular species modulated by the lightsensitive parameter ρ; cf. figure 4d) is very small, we observe a dead zone with a length of L = 0.475, which corresponds to about 3 h within 24 h, indicating the universality of having a dead zone in order to attain optimality. The improvement in the entrainability that is induced by a dead zone is calculated by comparing the entrainability of the optimal PRCs with that of typical sinusoidal PRCs. We consider sinusoidal functions for both the iPRC U_{j}(ϕ) and pPRC Z(ϕ) by settingandwhere c is the parameter to be optimized so that entrainability is maximized for each α (see appendix B for the explicit expressions). Equations (3.4) and (3.5) are scaled, so that they satisfy the constraints on the period variance (equation (2.12)). We calculated the ratioswhere and represent the entrainabilities for the cases of the sinusoidal iPRC and pPRC, respectively, calculated for the solar radiation input. For the sinusoidal iPRC of equation (3.4), the entrainability is calculated with pPRC via equation (2.14). R_{U} and R_{Z} quantify the improvement rate of the optimal PRCs over the sinusoidal iPRC (R_{U}) and pPRC (R_{Z}). In figure 6b, the solid and dashed lines show R_{U} and R_{Z}, respectively, as a function of α. Both ratios monotonically increase as α increases, which shows that the optimal PRC with a dead zone exhibits better entrainability when the oscillation of x_{LC,k}(ϕ) has a larger amplitude. When the concentration of x_{LC,k}(ϕ) is low, the effects of the input signal on the circadian oscillators are smaller. This is because pPRC Z(ϕ), which quantifies the extent of the phase shift owing to the stimulation of the parameter, depends on the concentration x_{LC,k}(ϕ) (see equation (2.14)). However, even within the range ϕ where x_{LC,k}(ϕ) has smaller values, the iPRC U_{j}(ϕ) contributes to an increase in the variance of the period, regardless of the concentration. From this, we see that having an iPRC with a smaller magnitude when the concentration of x_{LC,k}(ϕ) is smaller results in a smaller variance, which results in a larger entrainability for a constant variance of the period. Although this qualitatively explains the benefit of a dead zone, for some input values, the optimal PRCs may not contain a dead zone for any value of α. This will be shown in the following.Because the optimal PRCs depend on input signals (equations (2.20) and (2.21)), we next consider a typical periodic input signal, a sinusoidal function (equation (2.25)). In this case, Ψ(Δ, δ) is calculated in a closed form (an explicit expression of Ψ(Δ, δ) is given in appendix A), which is plotted as functions of Δ and δ in figure 7a–c for three cases: α = 0 (figure 7a), α = 0.5 (figure 7b) and α = 1.0 (figure 7c). As can been seen from figure 7a–c, Ψ(Δ, δ) yields the maximal value for (Δ, δ) = (π, nπ) for 0 < α ≤ 1, where n is an integer and when α = 0, δ can take any value. Figure 7d–f expresses the optimal iPRCs U_{j}(ϕ) and pPRCs Z(ϕ) for the sinusoidal input. For α = 0, the optimal PRC is sinusoidal (figure 7d) and for α = 0.5, the optimal PRC is still close to a sinusoidal function (figure 7e). When increasing α to α = 1.0, the PRC diverges from the sinusoidal function and exhibits almost positive values (figure 7f). We see that the optimal PRCs owing to equations (2.20) and (2.21) do not exhibit a dead zone for any αvalues (figure 7d–f) when the input signal is a simple sinusoidal function.
The existence of a dead zone optimizes both entrainability and regularity. It is rather obvious that optimization of regularity alone leads to a dead zone [^{43}], because null response means no effect by any kind of fluctuations. Our result instead shows that optimality of both entrainability and regularity, which are in a tradeoff relationship, is uniquely achieved by a dead zone. Our finding is fairly general, because a dead zone always exists in an optimal PRC unless α = 0 (additive stimulation). Along with the fact that T, σ_{T} and q affect only the scaling of the optimal PRCs, when the input signal affects the dynamics multiplicatively (i.e. α > 0), the existence of a dead zone always provides a synchronization advantage. This is supported by many experimental studies of various species that report the existence of a dead zone in the PRC [^{1}] (cf. figure 2c,d). Our general result suggests that circadian oscillators have fully adapted to solar radiation to improve synchronization. Indeed, many experimental findings imply that circadian oscillators have adapted to actual solar radiation [^{44}]: for various animals, light–dark (LD) cycles that include a twilight period result in better entrainability than do abrupt LD cycles (on–off protocols) [^{44}]. In this regard, another interesting problem is optimal entrainment [^{37}] of circadian clocks by light stimuli. As two different input signals, the solar radiation and sinusoidal inputs, yield the same optimal PRCs for α = 0, optimal inputs and optimal PRCs do not have onetoone correspondence. Thus, the optimal inputs are not trivial and this problem should be pursued in our future studies.
The solar radiation input plays an essential role, because it yields a dead zone in the optimal PRC, whereas a sinusoidal signal does not (figure 7). In other words, oscillators that are entrained by stimuli other than solar radiation may not exhibit a dead zone in their PRCs. This is indeed found in mammals. Mammals possess a master clock in their suprachiasmatic nucleus (SCN), which receives light stimuli via retinal photoreceptors, and peripheral clocks in body cells [^{45}]. The peripheral oscillators are entrained by several stimuli such as feeding and signals from the SCN through chemical pathways (e.g. hormones) [^{45},^{46}]. By injection experiments of a hormone, Balsalobre et al. [^{47}] reported that the PRCs of the peripheral oscillators in the liver do not have a dead zone.
Our result also agrees with other experimental observations. Our theory implies that a dead zone should be located where the concentration x_{LC,k}(ϕ) is low (0 ≤ ϕ ≤ π in figure 4d), and that to achieve optimality, the concentration of x_{LC,k}(ϕ) should be maximal in the region where the PRCs exhibit a large phase shift. In Drosophila, the timeless (tim) gene is regarded as the molecular implementation of x_{LC,k}(ϕ). It is experimentally known that light enhances the degradation of the gene product (the TIM protein) [^{48},^{49}], and the TIM protein peaks during the late evening. Figure 2c shows observations of the PRC of Drosophila against light pulses as a function time from Hall & Rosbash [^{19}]; circles describe the experimental data, and the solid line expresses a trigonometric curve fitting (fourth order). Because the centre of the part of the PRC that can be phase shifted approximately corresponds to the peak of the concentration, as denoted above, when estimated from the PRC alone, the concentration peak of the TIM protein should occur at about 18 h. This time is also close to the experimental evidence (i.e. late evening). Therefore, our theory can be used to hypothesize further molecular behaviour affected by light stimuli.
In summary, we have constructed a model that regards circadian oscillators as a global optimization of entrainability and regularity. We have shown that our model is consistent with much experimental evidence as mentioned above. The extension and improvement of our method are possible and they are left as an area of future study.
This work was supported by the Global COE programme ‘Deciphering Biosphere from Genome Big Bang’ from MEXT, Japan (Y.H. and M.A.); grantinaid for Young Scientists B (no. 25870171) from MEXT, Japan (Y.H.); grantinaid for Scientific Research on Innovative Areas ‘Biosynthetic machinery’ from MEXT, Japan (M.A.).
A.1. Solar radiation input case
For the solar radiation input case (equation (2.24)), Ψ(Δ, δ) is given by
withwhere . We show equation (A 1) as functions of Δ and δ in figure 5d–f.A.2. Sinusoidal input case
For the sinusoidal input case (equation (2.25)), Ψ(Δ, δ) is given by
We plot equation (A 2) as functions of Δ and δ in figure 7d–f.
B.1. Sinusoidal infinitesimal phase–response curve
An explicit expression for the sinusoidal iPRC (equation (3.4)) is
which yields the period variance of . Then, the corresponding pPRC is given bywhere we used equation (2.14).B.2. Sinusoidal parametric phase–response curve
For the pPRC Z(ϕ) to be a sinusoidal function, the iPRC U_{j}(ϕ) must be
where we used equation (2.14). An explicit expression of equation (B 3) iswhere 𝒩(c) is a normalizing termEquation (B 4) is normalized, so that the period variance becomes . Using equation (2.14), the corresponding pPRC is a sinusoidal function:which is an explicit expression of the sinusoidal pPRC (equation (3.5)).References
1.  Refinetti R. Year: 2005Circadian physiology, 2nd ednBoca Raton, FL: Taylor & Francis 
2.  Gonze D. Year: 2011Modeling circadian clocks: roles, advantages, and limitations. Cent. Eur. J. Biol.6, 712–729 (doi:10.2478/s1153501100624) 
3.  Johnsson A,Engelmann W. Year: 2007The biological clock and its resetting by light. In Photobiology: the science of life and light (ed. L Olof Björn), pp. 321–388. New York, NY: Springer 
4.  Hubbard KE,Robertson FC,Dalchau N,Webb AAR. Year: 2009Systems analyses of circadian networks. Mol. Biosyst.5, 1502–1511 (doi:10.1039/b907714f)19763342 
5.  Ukai H,Ueda HR. Year: 2010Systems biology of mammalian circadian clocks. Annu. Rev. Physiol.72, 579–603 (doi:10.1146/annurevphysiol073109130051)20148689 
6.  Johnson CH,Stewart PL,Egli M. Year: 2011The cyanobacterial circadian system: from biophysics to bioevolution. Annu. Rev. Biophys.40, 143–167 (doi:10.1146/annurevbiophys042910155317)21332358 
7.  Young MW,Kay SA. Year: 2001Time zones: a comparative genetics of circadian clocks. Nat. Rev. Genet.2, 702–715 (doi:10.1038/35088576)11533719 
8.  Kærn M,Elston TC,Blake WJ,Collins JJ. Year: 2005Stochasticity in gene expression: from theories to phenotypes. Nat. Rev. Genet.6, 451–464 (doi:10.1038/nrg1615)15883588 
9.  Perkins TJ,Swain PS. Year: 2009Strategies for cellular decisionmaking. Mol. Syst. Biol.5, 326 (doi:10.1038/msb.2009.83)19920811 
10.  Raser JM,O'Shea EK. Year: 2005Noise in gene expression: origins, consequences, and control. Science309, 2010–2013 (doi:10.1126/science.1105891)16179466 
11.  Eldar A,Elowitz MB. Year: 2010Functional roles for noise in genetic circuits. Nature467, 167–173 (doi:10.1038/nature09326)20829787 
12.  Hasegawa Y,Arita M. Year: 2012Fluctuating noise drives Brownian transport. J. R. Soc. Interface9, 3554–3563 (doi:10.1098/rsif.2012.0603)22977101 
13.  Viney M,Reece SE. Year: 2013Adaptive noise. Proc. R. Soc. B280, 20131104 (doi:10.1098/rspb.2013.1104) 
14.  Gonze D,Halloy J,Goldbeter A. Year: 2002Robustness of circadian rhythms with respect to molecular noise. Proc. Natl Acad. Sci. USA99, 673–678 (doi:10.1073/pnas.022628299)11792856 
15.  Forger DB,Peskin CS. Year: 2005Stochastic simulation of the mammalian circadian clock. Proc. Natl Acad. Sci. USA102, 321–324 (doi:10.1073/pnas.0408465102)15626756 
16.  Morelli LG,Jülicher F. Year: 2007Precision of genetic oscillators and clocks. Phys. Rev. Lett.98, 228101 (doi:10.1103/PhysRevLett.98.228101)17677881 
17.  Rougemont J,Naef F. Year: 2007Dynamical signatures of cellular fluctuations and oscillator stability in peripheral circadian clocks. Mol. Syst. Biol.3, 93 (doi:10.1038/msb4100130)17353935 
18.  Kuramoto Y. Year: 2003Chemical oscillations, waves, and turbulence. Mineola, NY: Dover Publications 
19.  Hall JC,Rosbash M. Year: 1987Genes and biological rhythms. Trends Genet.3, 185–191 (doi:10.1016/01689525(87)902319) 
20.  Daan S,Pittendrigh CS. Year: 1976A functional analysis of circadian pacemakers in noctornal rodents II. The variability of phase response curves. J. Comp. Physiol.106, 253–266 (doi:10.1007/BF01417857) 
21.  Vilar JMG,Kueh HY,Barkai N,Leibler S. Year: 2002Mechanisms of noiseresistance in genetic oscillators. Proc. Natl Acad. Sci. USA99, 5988–5992 (doi:10.1073/pnas.092133899)11972055 
22.  Scott M,Ingalls B,Kærn M. Year: 2006Estimations of intrinsic and extrinsic noise in models of nonlinear genetic networks. Chaos16, 026107 (doi:10.1063/1.2211787)16822039 
23.  Yoda M,Ushikubo T,Inoue W,Sasai M. Year: 2007Roles of noise in single and coupled multiple genetic oscillators. J. Chem. Phys.126, 115101 (doi:10.1063/1.2539037)17381233 
24.  Izhikevich EM. Year: 2007Dynamical systems in neuroscience: the geometry of excitability and bursting. Cambridge, MA: MIT Press 
25.  Risken H. Year: 1989The Fokker–Planck equation: methods of solution and applications, 2nd ednBerlin, Germany: Springer 
26.  Kori H,Kawamura Y,Masuda N. Year: 2012Structure of cell networks critically determines oscillation regularity. J. Theor. Biol.297, 61–72 (doi:10.1016/j.jtbi.2011.12.007)22185977 
27.  Gonze D,Goldbeter A. Year: 2000Entrainment versus chaos in a model for a circadian oscillator driven by light–dark cycles. J. Stat. Phys.101, 649–663 (doi:10.1023/A:1026410121183) 
28.  Hasty J,Dolnik M,Rottschäfer V,Collins JJ. Year: 2002Synthetic gene network for entraining and amplifying cellular oscillations. Phys. Rev. Lett.88, 148101 (doi:10.1103/PhysRevLett.88.148101)11955179 
29.  Bagheri N,Taylor SR,Meeker K,Petzold LR,Doyle FJ III. Year: 2008Synchrony and entrainment properties of robust circadian oscillators. J. R. Soc. Interface5, S17–S28 (doi:10.1098/rsif.2008.0045.focus)18426774 
30.  Abraham U,Granada AE,Westermark PO,Heine M,Kramer A,Herzel H. Year: 2010Coupling governs entrainment range of circadian clocks. Mol. Syst. Biol.6, 438 (doi:10.1038/msb.2010.92)21119632 
31.  Hasegawa Y,Arita M. Year: 2013Enhanced entrainability of genetic oscillators by period mismatch. J. R. Soc. Interface10, 20121020 (doi:10.1098/rsif.2012.1020)23389900 
32.  Erzberger A,Hampp G,Granada AE,Albrecht U,Herzel H. Year: 2013Genetic redundancy strengthens the circadian clock leading to a narrow entrainment range. J. R. Soc. Interface10, 20130221 (doi:10.1098/rsif.2013.0221)23676895 
33.  Taylor SR,Gunawan R,Petzold LR,Doyle FJ III. Year: 2008Sensitivity measures for oscillating systems: application to mammalian circadian gene network. IEEE Trans. Automat. Control53, 177–188 (doi:10.1109/TAC.2007.911364) 
34.  Abouzeid A,Ermentrout B. Year: 2009TypeII phase resetting curve is optimal for stochastic synchrony. Phys. Rev. E80, 011911 (doi:10.1103/PhysRevE.80.011911) 
35.  Teramae J,Tanaka D. Year: 2004Robustness of the noiseinduced phase synchronization in a general class of limit cycle oscillators. Phys. Rev. Lett.93, 204103 (doi:10.1103/physrevlett.93.204103)15600929 
36.  Nakao H,Arai K,Kawamura Y. Year: 2007Noiseinduced synchronization and clustering in ensembles of uncoupled limitcycle oscillators. Phys. Rev. Lett.98, 184101 (doi:10.1103/PhysRevLett.98.184101)17501578 
37.  Harada T,Tanaka HA,Hankins MJ,Kiss IZ. Year: 2010Optimal waveform for the entrainment of a weakly forced oscillator. Phys. Rev. Lett.105, 088301 (doi:10.1103/physrevlett.105.088301)20868133 
38.  Hartmann DL. Year: 1994Global physical climatology. San Diego, CA: Academic Press. 
39.  Vick BD,Moss TA. Year: 2013Adding concentrated solar power plants to wind farms to achieve a good utility electrical load match. Solar Energy92, 298–312 (doi:10.1016/j.solener.2013.03.007) 
40.  Golombek DA,Rosenstein RE. Year: 2010Physiology of circadian entrainment. Physiol. Rev.90, 1063–1102 (doi:10.1152/physrev.00009.2009)20664079 
41.  Novák B,Tyson JJ. Year: 2008Design principles of biochemical oscillators. Nat. Rev. Mol. Cell Biol.9, 981–991 (doi:10.1038/nrm2530)18971947 
42.  Ohta H,Yamazaki S,McMahon DG. Year: 2005Constant light desynchronizes mammalian clock neurons. Nat. Neurosci.8, 267–269 (doi:10.1038/nn1395)15746913 
43.  Pfeuty B,Thommen Q,Lefranc M. Year: 2011Robust entrainment of circadian oscillators requires specific phase response curves. Biophys. J.100, 2557–2565 (doi:10.1016/j.bpj.2011.04.043)21641300 
44.  Fleissner G,Fleissner G. Year: 2002Perception of natural Zeitgeber signals. In Biological rhythms (ed. V Kumar), pp. 83–93. Berlin, Germany: Springer 
45.  Dibner C,Schibler U,Albrecht U. Year: 2010The mammalian circadian timing system: organization and coordination of central and peripheral clocks. Annu. Rev. Physiol.72, 517–549 (doi:10.1146/annurevphysiol021909135821)20148687 
46.  Dickmeis T. Year: 2009Glucocorticoids and the circadian clock. J. Endocrinol.200, 3–22 (doi:10.1677/JOE080415)18971218 
47.  Balsalobre A,Brown SA,Marcacci L,Tranche F,Kellendonk C,Reichardt HM,Schütz G,Schibler U. Year: 2000Resetting of circadian time in peripheral tissues by glucocorticoid signaling. Science29, 2344–2347 (doi:10.1126/science.289.5488.2344)11009419 
48.  Hardin PE. Year: 2005The circadian timekeeping system of Drosophila. Curr. Biol.15, R714–R722 (doi:10.1016/j.cub.2005.08.019)16139204 
49.  Xie Z,Kulasiri D. Year: 2007Modelling of circadian rhythms in Drosophila incorporating the interlocked PER/TIM and VRI/PDP1 feedback loops. J. Theor. Biol.245, 290–304 (doi:10.1016/j.jtbi.2006.10.028)17157878 
Figures
Article Categories:
Keywords: circadian clock, phase–response curve, variational method. 
Previous Document: Nanointerfacial strength between noncollagenous protein and collagen fibrils in antler bone.
Next Document: Alignment of crystal orientations of the multidomain photonic crystals in Parides sesostris wing sc...