Abstract
Several different enzymes display an apparent diffusion coefficient that increases with the concentration of their substrate. Moreover, their motion becomes directed in substrate gradients. Currently, there are several competing models for these transport dynamics. Here, we use mathematical modeling and numerical simulations to analyze whether the enzymatic reactions can generate a significant feedback from enzyme transport onto the substrate profile. We find that this feedback can generate spontaneous spatial patterns in the enzyme distribution, with just a singlestep catalytic reaction. However, patterns are formed only for a subclass of transport models. For such models, nonspecific repulsive interactions between the enzyme and the substrate, or attractive interactions between the enzyme and the product, cause the enzyme to accumulate in regions of low substrate concentration. Reactions then amplify local substrate and product fluctuations, causing enzymes to further accumulate where substrate is low. Experimental analysis of this pattern formation process could discriminate between different transport models.
Similar content being viewed by others
Introduction
Experiments performed during the last decade found that at least eight different enzymes display a higher diffusion coefficient when the concentration of the corresponding substrate in solution is increased^{1,2,3,4,5,6,7}. These increases are in the range of 24–80% relative to the diffusion coefficients without substrate. Most of these experiments relied on fluorescence correlation spectroscopy (FCS) measurements. Although artifacts introduced by this technique have been pointed out^{8,9}, recent findings using other techniques^{10,11} validated the phenomenon, which is often referred to as “enhanced diffusion”. The underlying mechanism is still under debate^{12}. Some experiments suggest that catalysis plays a key role^{3,5,13,14}, while others indicate that enhanced diffusion persists when the substrate is replaced by an inhibitor^{1,4,15}. According to the latest experiments performed with the enzyme urease^{11,13}, it appears that both the binding and the catalysis step of the reaction scheme contribute to enhanced diffusion.
A related question is how enzymes behave in the presence of substrate gradients. The answer to this question appears to be complex. Some experiments suggest that enzymes drift downstream gradients of substrates, performing “antichemotaxis”^{5,13}. Others suggest that enzymes move upstream gradients of substrates, performing “chemotaxis”^{6,7,15}. Antichemotaxis can be explained based on just the enhanced diffusion^{5,16} (enzymes accumulate in regions with low substrate concentration where they have a lower diffusion coefficient), chemotaxis cannot be generated by enhanced diffusion alone. However, crossdiffusion is a possible cause for enzyme chemotaxis^{6}. Crossdiffusion describes the response of the enzyme to forces generated by gradients of substrate. Mathematically, it corresponds to an offdiagonal element in the diffusion matrix describing the combined motion of the enzyme and the substrate. It has been suggested that crossdiffusion can be due to specific interactions (ligand binding) between the enzyme and the substrate^{6,15,17} or due to nonspecific interactions (e.g., steric, electrostatic, and van der Waals)^{18}. Specific interactions only lead to chemotaxis, while nonspecific interactions can cause the enzymes to move both up or downstream the substrate gradient, depending on whether the interactions are attractive or repulsive, respectively. The model including nonspecific interactions^{18} can be considered as a mathematical generalization of other existing models^{5,6,17}.
While the existing models and experiments study how enzymes move in preimposed substrate gradients, they do not consider the feedback from the enzymatic reaction onto the substrate distribution. Here, we analyze the effects of this feedback starting from the most general transport model^{18}. We show that spatial patterns can emerge in initially homogeneous systems if nonspecific interactions contribute to the accumulation of the enzyme in regions where concentration of substrate is low. Enzymes accumulating in these regions further deplete the substrate, causing the substrate gradient to become steeper, hence further increasing the accumulation of the enzyme. We obtain a set of conditions for the parameter range in which patterns form. We see that patterns arising from initial homogeneous concentrations can emerge only for nonspecific interactions driving the enzyme away from the substrate, hence only for the model proposed by Agudo–Canalejo et al.^{18}, but not for the models proposed by Zhao et al.^{6}, Mohajerani et al.^{15}, and Jee et al.^{5}, suggesting that the analysis of spontaneous pattern formation experiments can be used to discriminate between the different proposed models. Our findings imply that patterns can arise for a singlestep enzymatic reaction even in the absence of autocatalytic activity or allosteric regulation. This is surprising given that the formation of conventional Turing patterns^{19} with such simple reaction schemes requires at least a reaction network of three states, where forward and backward reactions are catalyzed by two different enzymes^{20}.
Results
Model scenario
Our starting model system is depicted in Fig. 1. We consider a singlestep enzymatic reaction in a narrow reaction chamber of length L connected via a permeable membrane to a large substrate reservoir. We assume only substrate and product molecules can diffuse through the permeable membrane, while enzymes are confined to the reaction chamber (since enzymes are typically larger than their substrates and products). We indicate with γ_{s} and γ_{p} the permeation rates of substrate and product, respectively, i.e., the rate at which substrate and product molecules diffuse through the membrane. Such rates are obtained by dividing the substrate and product permeabilities by the reaction chamber thickness. The reservoir has a fixed concentration of substrate s_{R} and no products.
For the reaction occurring in the bulk of this effectively 1D system we assume a Michaelis–Menten scheme (Fig. 1, inset). The substrate \({\mathcal{S}}\) binds to the enzyme \({\mathcal{E}}\) with rate k_{on}, forming a complex, and it can unbind with rate k_{off}. The catalytic step of the reaction has rate constant k_{cat} and catalysis is irreversible. This leads to a turnover rate per enzyme \({k}_{{\rm{cat}}}F(s):= {k}_{{\rm{cat}}}s{({K}_{{\rm{M}}}+s)}^{1}\), where s is the substrate concentration and \({K}_{{\rm{M}}}=({k}_{{\rm{off}}}+{k}_{{\rm{cat}}}){k}_{{\rm{on}}}^{1}\). Following Agudo–Canalejo et al.^{18} we assume that the diffusion coefficient of the enzyme depends on whether the enzyme is free, D_{f}, or it is in its complexed form, D_{c}. This change in diffusion coefficient can be due to changes in either the hydrodynamic radius or the conformational fluctuations of the enzyme upon substrate binding^{4,21,22}. Typically, enzymes become more tightly folded upon substrate binding, i.e., D_{c} > D_{f}, consistent with the experimentally observed trend. We also assume that shortrange nonspecific interactions can either cause the enzyme to move towards or away from substrate, effectively generating a phoretic drift velocity that can also be interpreted as crossdiffusion^{18}. We denote these interactions via pairwise potentials ϕ_{fs}, ϕ_{cs} depending on whether the enzyme is free or in the complexed form, respectively, as depicted in Fig. 1(inset).
Under the assumptions that (a) the enzyme is very dilute and (b) the system is locally in chemical equilibrium (i.e., the timescales of diffusion and crossdiffusion are slower than the chemical reactions), one can derive an effective transport equation for the enzymes^{18},
within the quasi1D reaction chamber (Fig. 1) oriented along the xaxis. Here, e(x, t) denotes the local enzyme concentration (regardless of free or complexed) and s(x, t) the substrate concentration. The effective diffusion coefficient of the enzyme,
is a function of substrate concentration, interpolating between the diffusion coefficient of the free enzyme and the complex, with F(s) as defined above. For s ≪ K_{M}, D_{e}(s) ~ D_{f}, whereas D_{e}(s) ~ D_{c} for s ≫ K_{M}. Note that the disassembly of enzyme oligomers into monomers can also contribute to enhanced diffusion^{9,10,23}. Equation (1) would then describe the motion of the enzyme irrespective of its oligomeric state. The crossdiffusion term \({D}_{{\rm{xd}}}^{{\rm{s}}}(s,e)\) of Eq. (1) describes how enzymes respond to gradients of substrate due to the shortrange nonspecific and hydrodynamic interactions,
Here we define what we will refer to as ‘crossdiffusion strength’ \({C}_{{\rm{c}}/{\rm{f}}}^{{\rm{s}}}:= {N}_{{\rm{A}}}{k}_{{\rm{B}}}T{\lambda }_{{\rm{c}}/{\rm{f}}}^{{\rm{s}}\ 2}{\eta }^{1}\), where η is the viscosity of the fluid, k_{B} the Boltzmann constant, N_{A} the Avogadro number, T the temperature and \({\lambda }_{{\rm{c}}/{\rm{f}}}^{{\rm{s}}}\) the Derjaguin length^{18,24}. The Derjaguin length is a parameter capturing the effective shortrange interaction between the complex/free enzyme and the substrate. It is typically a few angstroms^{25,26} and can at most be as large as the Debye length (screening length), which in typical buffer conditions is about 1 nm^{12}. It is expressed via the integral \(\lambda_{{\rm{c}}/{\rm{f}}}^{{\rm{s}}\ 2}= \int_{0}^{\infty}dhh({e}^{{\phi }_{{\rm{cs/fs}}}(h){({k}_{{\rm{B}}}T)}^{1}}1)\). \({\lambda }_{{\rm{c}}/{\rm{f}}}^{{\rm{s}}\ 2}\) is positive (negative) when the interaction is attractive (repulsive)^{18}. The derivation of \({\lambda }_{{\rm{c}}/{\rm{f}}}^{{\rm{s}}\ 2}\) is similar to that of the second Virial coefficient for a real gas^{27}, but it also includes hydrodynamic effects and is computed by assuming that the size of the enzyme is much larger than the interaction length. The sign of \({\lambda }_{{\rm{c}}/{\rm{f}}}^{{\rm{s}}\ 2}\) determines the sign of \({C}_{{\rm{c}}/{\rm{f}}}^{{\rm{s}}}\) and it has opposite sign as compared to \({D}_{{\rm{xd}}}^{{\rm{s}}}(s,e)\). For attractive interactions \({D}_{{\rm{xd}}}^{{\rm{s}}}(s,e)<0\), i.e., the enzyme drifts towards higher concentrations of substrate (chemotaxis). The enzyme performs antichemotaxis for repulsive interactions, for which \({D}_{{\rm{xd}}}^{{\rm{s}}}(s,e)> 0\). Note that the effect of nonspecific interactions can also be written as a phoretic drift^{18} by swapping the ∂_{x}s in Eq. (1) with e in Eq. (3), with a drift velocity directly proportional to the substrate gradient \({v}_{{\rm{ph}}}(s,{\partial }_{x}s)=\left[{C}_{{\rm{f}}}^{{\rm{s}}}+({C}_{{\rm{c}}}^{{\rm{s}}}{C}_{{\rm{f}}}^{{\rm{s}}})F(s)\right]{\partial }_{x}s\).
The enzyme dynamics as given by Eq. (1) is valid if shortrange nonspecific interactions between the product and the enzyme are neglected. We first neglect such interactions, but will discuss their effects further below. This allows us to derive simple analytical conditions for the determination of the relevant parameter ranges for the formation of spontaneous patterns in the enzyme, substrate and product profiles. This approximation is valid for γ_{p} ≫ γ_{s}. In such a regime products very quickly permeate out of the reaction chamber, impeding the formation of large product gradients and causing the crossdiffusion induced by the product to be negligible. Note that the values of γ_{p} and γ_{s} can vary by several order of magnitudes. For a reaction chamber that is 1–10 μm thick, the permeation rates are in the range of γ_{s,p} ≈ 10^{−9}–10^{4}s^{−1} ^{28}. In the regime where the enzyme is dilute and Eq. (1) is valid, the crossdiffusion that enzyme molecules would induce on substrate and product molecules can be neglected. Moreover the crossdiffusions related to the substrate–product interaction can also be neglected as product and substrate molecules are typically much smaller than the enzyme and they do not affect each other as much as they affect the enzyme motion. Hence, the reaction chamber of Fig. 1 is described by the coupled reactiontransport equations
where p(x, t) denotes the product concentration. The dynamics of the system of Eq. (4) is determined by the interplay of the different physical processes described above, each associated with a different timescale. The diffusion processes have timescales \({\tau }_{{\rm{diff.}}}={L}^{2}{D}_{{\rm{e,s,p}}}^{1}\). The crossdiffusion process has a timescale \({\tau }_{{\rm{xd}}}={L}^{2}{e}_{{\rm{h}}}{({D}_{{\rm{xd}}}^{{\rm{s}}}{s}_{{\rm{R}}})}^{1}\) and it represents the time it takes an enzyme to explore a length L, driven by a constant gradient s_{R}L^{−1}, where s_{R} is the maximal substrate concentration allowed in the system. Equivalently, we can say that the nonspecific interactions cause the enzyme to drift with velocity v_{ph}(s, ∂_{x}s), with associated timescale \({\tau }_{{\rm{ph.}}}=L{v}_{{\rm{ph}}}^{1}={\tau }_{{\rm{xd}}}\). The reaction has a timescale \({\tau }_{{\rm{react.}}}={k}_{{\rm{cat}}}^{1}\) and the permeations have timescales \({\tau }_{{\rm{perm.}}}={\gamma }_{{\rm{s,p}}}^{1}\).
Having the reaction chamber coupled to a reservoir avoids product accumulation and substrate depletion, generating a nonzero homogeneous steadystate, with concentrations e_{h}, s_{h}, and p_{h} for enzyme, substrate, and product, respectively. For simplicity, we express e_{h} and p_{h} as functions of s_{h},
Since F(s_{h}) is a monotonic increasing function of s_{h}, it is possible to write s_{h} and p_{h} in terms of e_{h}, which can be directly tuned in experiments via the total enzyme concentration [see Supplementary Eqs. (4) and (5) of Supplementary Note 1.1 and the Supplementary Fig. 1].
Instability driven by substrate induced crossdiffusion
The homogeneous solution as given by Eqs. (5),(6) is stable for any positive values of the parameters for a wellmixed system, i.e., a system with no diffusion and no crossdiffusion (see Supplementary Note 1.2). Figure 2 shows the results of two simulations of the full system, Eq. (4), with periodic boundary conditions and parameters as given in the Supplementary Table 1. In Fig. 2a we see that the homogeneous solution is unstable and patterns form, for a value of \({\lambda }_{{\rm{f}}}^{{\rm{s}}\ 2}={\lambda }_{{\rm{c}}}^{{\rm{s}}\ 2}=1 {\mathrm{\AA}}^{2}\). A steadystate is reached after about 10^{3}s, corresponding to the enzyme diffusive timescale that for the chosen parameters is the slowest timescale \({\tau }_{{\rm{diff.}}}= {L}^{2}{D}_{{\rm{f}}}^{1}=1{0}^{3}\)s. In Fig. 2b we see that for \({\lambda }_{{\rm{f}}}^{{\rm{s}}\ 2}={\lambda }_{{\rm{c}}}^{{\rm{s}}\ 2}=1{\mathrm{\AA}}^{2}\) the homogeneous solution is instead stable. Hence, depending on the parameters, the system will spontaneously form patterns.
To characterize the instability of the homogeneous steadystate solution, v_{h} = (e_{h}, s_{h}, p_{h}), we linearize Eq. (4), v → v_{h} + δv, and make the exponential ansatz δv = v_{0} e^{σt}e^{iqx}, with q the spatial frequency of the linear perturbation and σ the perturbation growth rate. If σ > 0, the perturbation grows with a timescale as given by σ^{−1} and patterns form (see Supplementary Note 1.5 for the determination of σ^{−1} for the patterns of Fig. 2a). Note that since the system of Eq. (4) is isotropic, the results below will also be valid for systems of higher physical dimensions. A positive σ can be found (see Supplementary Note 1.3) provided that
with \(D_{\rm{e}}^{\prime} ({s}_{{\rm{h}}})={\partial {D}_{{\rm{e}}}(s)/{\partial }{s}}_{{s}_{{\rm{h}}}}\) and \(F^{\prime} ({s}_{{\rm{h}}})={\partial F(s)/{\partial }{s}}_{{s}_{{\rm{h}}}}\). Using the functional form of D_{e} [Eq. (2)], the term given in square brackets equals to \({D}_{{\rm{f}}}F^{\prime} ({D}_{{\rm{e}}}F)^{1}\), which is negative. Hence for the homogeneous solution to be unstable we must have \({D}_{{\rm{xd}}}^{{\rm{s}}}({s}_{{\rm{h}}},{e}_{{\rm{h}}}) \, > \, 0\), i.e., enzyme molecules drift downstream gradients of substrate, performing antichemotaxis. We can have a positive crossdiffusion if the nonspecific interactions between enzyme and substrate molecules are repulsive, i.e. for negative crossdiffusion strengths \({C}_{{\rm{c}}/{\rm{f}}}^{{\rm{s}}} \, < \, 0\). However, how repulsive do these interactions need to be? By using the definitions of D_{e}(s), F(s) and \({D}_{{\rm{xd}}}^{{\rm{s}}}(s,e)\) is possible to show [see Supplementary Eqs. (20–25) in the Supplementary Note 1.3] that Equation (7) is fulfilled if
with β = γ_{s}s_{R}. This result holds in the strong depletion regime (s_{h} ≪ s_{R}), where the effect of the reaction on the substrate dominates over the outflow to the reservoir. In this regime, the expressions are simpler and it is easier to pinpoint the driving mechanism behind the instability observed in Fig. 2. We refer the reader to the Supplementary Note 1.3 for the full analysis. The argument of the square root in the inequality Eq. (8) is positive if Eq. (9) is fulfilled. In the Supplementary Note 1.4 we show that the instability is a Type II instability^{29}, meaning that σ = 0 at q = 0, see Supplementary Fig. 2. This is natural as the total amount of enzymes in our system is conserved and homogeneously increasing or decreasing perturbations, i.e., perturbations at q = 0, would correspond to changes in the total enzyme amount. By looking at the inequalities Eq. (9), (8), we can see that negative crossdiffusion strengths \({C}_{{\rm{c}}/{\rm{f}}}^{{\rm{s}}} \, < \, 0\) are needed to have instabilities. Given that \({C}_{{\rm{c}}/{\rm{f}}}^{{\rm{s}}}:= {N}_{{\rm{A}}}{k}_{{\rm{B}}}T{\lambda }_{{\rm{c}}/{\rm{f}}}^{{\rm{s}}\ 2}{\eta }^{1}\), this means that repulsive interactions (\({\lambda }_{{\rm{c}}/{\rm{f}}}^{{\rm{s}}\ 2} \, < \, 0\)) are necessary. Diffusion tends to homogenize the concentration profiles and contributes with a positive term to the left hand side of inequality Eq. (9). In the case where the enzyme–substrate interaction is attractive, i.e., \({C}_{{\rm{f}}}^{{\rm{s}}} \, > \, 0\), also the second term on the left hand side of inequality Eq. (9) is positive. Hence the interaction between the enzyme and the substrate would need to change sign upon substrate binding to have an instability (\({C}_{{\rm{c}}}^{{\rm{s}}} \, < \, 0\)). This could happen if, upon substrate binding, the electrostatic surface charge distribution of the enzyme changes significantly, for instance due to a conformational transition. There can be cases in which the change in nonspecific interactions is less abrupt and both \({C}_{{\rm{f}}}^{{\rm{s}}} \, < \, 0\) and \({C}_{{\rm{c}}}^{{\rm{s}}} \, < \, 0\). Even in the case for which \({C}_{{\rm{c}}}^{{\rm{s}}}={C}_{{\rm{f}}}^{{\rm{s}}}={C}_{{\rm{s}}}\), i.e., there is no change in interaction upon substrate binding, it is possible to have an unstable homogeneous solution for
where we rewrote relation Eq. (9) with the use of the StokesEinstein relation, k_{B}T = 6πηD_{f}R_{f}, and the definition of C_{s}.
It is interesting to note how relations Eq. (9), (10) depend on the substrate concentration. One could ask, given certain nonspecific interactions between the enzyme and the substrate, at which substrate concentration \({s}_{{\rm{h}}}^{* }\) should we begin to observe instabilities? From relations Eq. (9)–(10), we find that
with \({\lambda }_{{\rm{s}}}^{2}\, <\, 0\). The value of \({s}_{{\rm{h}}}^{* }\) given by Eq. (11) is similar to the value of s_{h} above which the phoresis generated by the nonspecific interactions v_{ph}(s, ∂_{x}s) dominates over the drift induced by the enhanced diffusion v_{bi} ≔ −∂_{x}D_{e}(s)^{18}. They differ only by a prefactor (D_{c} − D_{f})/D_{f} in the second term inside the square root of Eq. (11). Having v_{ph} > v_{bi} corresponds to \({D}_{{\rm{xd}}}^{{\rm{s}}}(s,e) \, > \, eD_{\rm{e}}^{\prime} (s)\). However, \({s}_{{\rm{h}}}^{* }\) is not only determined by this inequality, but also from the inequality Eq. (7), where reaction also plays a role. By considering biologically relevant ranges, such as R_{f} ~ 1–10 nm^{30,31}, K_{M} ~ 10^{−2}–10^{6} μM^{32}, ∣λ_{s}∣ being smaller than the Debye length ∣λ_{s}∣ ≈ 10^{−2}–10 Å, we find that \({s}_{{\rm{h}}}^{* } \, \lesssim \,{K}_{{\rm{M}}}\) if \( {C}_{{\rm{s}}} {K}_{{\rm{M}}}{(4{D}_{{\rm{f}}})}^{1}\gg 1\), or equivalently \(3/2\pi \,{R}_{{\rm{f}}} {\lambda }_{{\rm{s}}}^{2} {N}_{{\rm{A}}}{K}_{{\rm{M}}}\gg 1\). This can happen only for enzymes for which K_{M} = 0.1–1M, which is only a small fraction of enzymes^{32}. For the vast majority of enzymes \( {C}_{{\rm{s}}} {K}_{{\rm{M}}}{(4{D}_{{\rm{f}}})}^{1}\ll 1\) and \({s}_{{\rm{h}}}^{* }\gg \,{K}_{{\rm{M}}}\), meaning that we are in the saturated regime of the reaction, for which F(s) ≈ 1. In such a regime, we find that \({s}_{{\rm{h}}}^{* }\approx \sqrt{{D}_{{\rm{f}}}\,{K}_{{\rm{M}}} {C}_{{\rm{s}}}{ }^{1}}=\sqrt{{K}_{{\rm{M}}}{(6\pi {R}_{{\rm{f}}}{N}_{{\rm{A}}} {\lambda }_{{\rm{s}}}^{2* })}^{1}}\).
In Fig. 3 we plot the phase diagram of the system of Eq. (4) where on the abscissa we have s_{h} and on the ordinate we have \({\lambda }_{{\rm{s}}}^{2}\). The green dashed line corresponds to the instability curve, Eq. (10), in the strong depletion regime (s_{h} ≪ s_{R}) and the light blue line represents the instability curve derived for any s_{h} [see Supplementary Eq. (23) in the Supplementary Note 1.3]. As a proxy for the determination of patterns we plot the ratio of the maximum over the minimum of the enzyme profile at steadystate. Above the instability lines the homogeneous solution is stable. Below the line the system is unstable and patterns similar to the one shown in Fig. 2a arise.
Positive feedback causing pattern formation
What is the physical mechanism underlying the instability given by the Eqs. (9), (8)? We illustrate the feedback mechanism generating the pattern in Fig. 4. We have already seen that to have instabilities the crossdiffusion \({D}_{{\rm{xd}}}^{{\rm{s}}}(s,e)\) must be positive, see Eq. (7). The enzymatic current induced by the crossdiffusion is given by \({J}_{{\rm{xd}}}^{{\rm{e}}}={D}_{{\rm{xd}}}^{{\rm{s}}}(s,e){\partial }_{x}s\), which for a negative slope of substrate concentration generates a positive current for the enzyme, i.e., the enzyme moves away from a high substrate concentration. The current generated by the enhanced diffusion \({J}_{{\rm{D}}}^{{\rm{e}}}={\partial }_{x}[{D}_{{\rm{e}}}(s)e]\) also consists of a motion of the enzyme away from high substrate concentrations. In Fig. 2a we can see how in a regime where patterns form, more product is generated in locations where the substrate concentration is low and the enzyme concentration is high. Repulsive nonspecific interactions and enhanced diffusion cause the enzyme to accumulate in such regions (Fig. 4 from (a) to (b)). This accumulation then generates a higher reaction flux in these regions. Having a stronger reaction flux where substrate is already low, as compared with regions where substrate is abundant, causes substrate gradients to become steeper [Fig. 4 from (b) to (c)]. A steeper substrate gradient, in turn, causes both \({J}_{{\rm{xd}}}^{{\rm{e}}}\) and \({J}_{{\rm{D}}}^{{\rm{e}}}\) to increase, hence generating a further accumulation of the enzyme in substrate depleted regions [Fig. 4 from (c) to (d)]. This positive feedback between reaction and enzyme accumulation, leads to the formation of patterns. The feedback cycle halts when the substrate concentration is too low and the reaction is balanced by the influx of substrate from the reservoir. Then the substrate gradient stops getting steeper and the system approaches a steadystate.
The role of enhanced diffusion
What happens without enhanced diffusion, with only nonspecific interactions? The diffusion function D_{e}(s) in Eq. (4) becomes a constant, i.e., D_{e} = D_{c}, and the interval as given by Eq. (8) for the unstable wave vector q is slightly affected. The Eq. (9) characterizing the instability is unaffected. This result suggests that enzyme patterns for a singlestep reaction form only if driven by repulsive nonspecific interactions between the substrate and the enzyme. Why cannot patterns be generated simply by enhanced diffusion? By considering the strong depletion regime (s_{h} ≪ s_{R}) and \({D}_{{\rm{xd}}}^{{\rm{s}}}(s,e)=0\), Eq. (7) becomes
Both D_{e}(s) and F(s) have a Michaelis–Menten dependence on s for all the models proposed so far for the enzyme motion. They differ only in the prefactors and a nonzero offset for D_{e}(s). Equation (12) is never fulfilled for such D_{e}(s) and F(s) and consequently any initial perturbation of the concentrations is smoothed out by diffusion (See Supplementary Note 2.1).
However, Eq. (12) will apply to any model with a spatially dependent diffusion coefficient that is coupled to another diffusing and reacting species that is being depleted. In case the diffusing species is being generated, the sign of the above inequality is flipped. Systems of this type are used to study bacterial motion. Interestingly synthetic bacterial populations show stripe patterns as they grow on semi solid agar plates^{33}. These bacteria produce a certain signaling molecule through which they sense their own concentration and regulate their mobility. If we neglect bacterial growth, we have a system of equations similar to the equations describing the enzyme and the substrate motion in Eq. (4). The enzyme would correspond to the bacteria and the substrate to the signaling molecule. For this system \(F(s)={\rm{const.}} \, < \, 0\) and D_{e}(s) is a Hill function with \(D_{\rm{e}}^{\prime} (s) \, < \, 0\). Hence patterns can form for \(D_{\rm{e}}^{\prime} (s){D}_{{\rm{e}}}{(s)}^{1} \, < \, {q}^{2}{D}_{{\rm{s}}}{\beta }^{1}\), where now β < 0. This shows that such synthetic population of bacteria can form patterns even in the absence of growth. The Eq. (12) specifies the minimal ingredients for pattern formation for such systems. It implies that, if F(s) > 0, patterns form whenever D_{e}(s) is more sensitive than F(s) to perturbations in the substrate concentration. Having a more sensitive D_{e}(s) than F(s) causes a more sensitive response in the enzyme motion than the depletion due to the reaction. Consider a local increase in substrate concentration and that both D_{e}(s) and F(s) are monotonically increasing functions of s. Having a more sensitive D_{e}(s) than F(s) implies a higher increase in the current \({J}_{{\rm{D}}}^{{\rm{e}}}\) due to enhanced diffusivity away from the substrate, as compared to the depletion of substrate due to the reaction. Molecules \({\mathcal{E}}\) then migrate to regions with low substrate and if they do so with a high enough rate they can cause substrate gradients to get steeper, as in step (c) of Fig. 4. However, for the enzyme model as given by Eq. (4), D_{e}(s) and F(s) alone cannot generate instabilities (see Supplementary Note 2.1). The accumulation of the enzyme in low substrate region at a high enough rate can be guaranteed only via the crossdiffusive term \({D}_{{\rm{xd}}}^{{\rm{s}}}(s,e)\). This is because D_{e}(s) and F(s) are Michaelis–Menten functions with the same K_{M} and therefore are equally sensitive to substrate perturbations. In the Supplementary Note 2.3 we show that if the Michaelis–Menten constant associated to D_{e} is much larger than the one associated to F(s), Eq. (12) can be fulfilled, see also Supplementary Fig. 3. In this scenario, there can be a range of s_{h} concentrations for which D_{e}(s_{h}) is in the linear regime and sensitive to substrate perturbations, while F(s_{h}) is in the saturated regime and therefore insensitive to perturbations in s. In the Supplementary Note 2.2, we also show that if D_{e}(s) is a Hill function and F(s) a Michaelis–Menten reaction, we can have spontaneous pattern formation for a Hill coefficient > 1.
Inclusion of enzyme–product interaction
Up to now we focused on the effects that shortrange nonspecific interactions between the substrate and the enzyme have on the enzyme motion. We neglected the product–enzyme interaction. This approximation was helpful to derive the Eqs. (9)–(11) and check the validity of pattern formation for biologically relevant parameter ranges.
Now we ask: How is pattern formation affected by the shortrange interaction between product and enzyme? The enzyme dynamics becomes affected by gradients of products (see Supplementary Note 3) and instead of Eq. (1) we get
where \({D}_{{\rm{xd}}}^{{\rm{p}}}(s,e)\) is the crossdiffusion due to nonspecific shortrange and hydrodynamics interactions between the product and the enzyme. Similarly to the substrate induced crossdiffusion, \({D}_{{\rm{xd}}}^{{\rm{p}}}(s,e)=[{C}_{{\rm{f}}}^{{\rm{p}}}+({C}_{{\rm{c}}}^{{\rm{p}}}{C}_{{\rm{f}}}^{{\rm{p}}})F(s)]e\), where \({C}_{{\rm{c}}/{\rm{f}}}^{{\rm{p}}}={N}_{{\rm{A}}}{k}_{{\rm{B}}}T{\lambda }_{{\rm{c}}/{\rm{f}}}^{{\rm{p}}\ 2}{\eta }^{1}\) is the crossdiffusive strength corresponding to the product–enzyme interaction for the complexed and free form, respectively. The enzyme drifts upstream (downstream) the product gradients for an attractive (repulsive) interaction, for which \({D}_{{\rm{xd}}}^{{\rm{p}}}(s,e) \, < \, 0\) (\({D}_{{\rm{xd}}}^{{\rm{p}}}(s,e) \, > \, 0\)). Note that \({D}_{{\rm{xd}}}^{{\rm{p}}}(s,e)\) depends on the substrate concentration, because the fraction of enzymes in the complexed form is given by F(s).
It is possible to derive a condition characterizing the instability of the homogeneous solution in the presence of product induced crossdiffusion, similar to Eq. (7) (see Supplementary Note 3):
As compared to Eq. (7), in Eq. (14) there is an extra term that involves the crossdiffusion \({D}_{{\rm{xd}}}^{{\rm{p}}}(s,e)\). The prefactor of such term goes to zero in the limit of large γ_{p} as expected. The product induced crossdiffusion contribute to the instability with opposite sign as compared to \({D}_{{\rm{xd}}}^{{\rm{s}}}(s,e)\). We have seen in Fig. 4 that patterns form if the interaction between substrate and enzyme is repulsive (transition from (a) to (b)) and \({D}_{{\rm{xd}}}^{{\rm{s}}}(s,e)\, > \, 0\). Then the reaction has the role of steepening preformed substrate gradients (transition from (b) to (c)), i.e., more products are formed where substrate is already low. This generates a gradient in the products that has a opposite sign as the substrate gradient, see Fig. 2a. Hence we can expect that patterns are favored by an attractive nonspecific interaction between the product and the enzyme, for which \({D}_{{\rm{xd}}}^{{\rm{p}}}(s,e) \, < \, 0\). This favors the accumulation of enzymes in region of low substrate, further steepening the substrate gradient. In Fig. 5a we present the results of simulations where \({\lambda }_{{\rm{f}}}^{{\rm{p}}}={\lambda }_{{\rm{c}}}^{{\rm{p}}}={\lambda }_{{\rm{p}}}\). We vary the value of λ_{p} and fix the other parameters to be the same as the ones considered for Fig. 2a, as given in Supplementary Table 1. For \({\lambda }_{{\rm{p}}}^{2}=1{\mathrm{\AA}}^{2}\) (Fig. 5a), i.e., for an attractive interaction between product and enzyme (\({D}_{{\rm{xd}}}^{{\rm{p}}}(s,e)\, <\, 0\)), we see that the steadystate pattern is more pronounced as compared to the case in which \({D}_{{\rm{xd}}}^{{\rm{p}}}=0\) (Fig. 2a). For \({\lambda }_{{\rm{p}}}^{2}={\lambda }_{{\rm{s}}}^{2}=1{\mathrm{\AA}}^{2}\) (Fig. 5b), we see how the pattern is still present, however, less pronounced. Note that we still get a pattern because we considered different permeation rates γ_{s} = 1s^{−1} and γ_{p} = 10s^{−1}. Finally for \({\lambda }_{{\rm{p}}}^{2}=10{\mathrm{\AA}}^{2}\) (Fig. 5c) we see that the repulsion between the product and enzyme is so strong that the pattern disappears. In this case \({D}_{{\rm{xd}}}^{{\rm{p}}}(s,e)\, > \, 0\) and the crossdiffusive term induced by the product prevails over the crossdiffusive term induced by the substrate. Therefore Eq. (14) is not fulfilled. Note that Eq. (14) can also be fulfilled if both crossdiffusions are negative, i.e., both substrate and product attract the enzyme. In this case, patterns form if the product attraction is stronger than the substrate one. As a result, enzyme would still antichemotax from the substrate and the intuitive mechanism leading to patterns illustrated in Fig. 4 still holds. Note that one could think of the intuitive mechanism leading to patterns in terms of the feedback between product induced crossdiffusion and product formation. Given an initial product gradient and a uniform enzyme profile, the enzyme would chemotax to the region of high product. Having more enzymes in this region would cause the formation of even more product. This would cause the steepening of the product gradient and the further accumulation of enzymes. This feedback mechanism is the ‘time reversal’ mechanism of the one illustrated in Fig. 4.
Discussion
We have seen that spontaneous patterns arising from homogeneous concentration profiles can form for a singlestep catalytic reaction if crossdiffusive effects are present. Patterns form given sufficiently strong repulsive nonspecific interactions between the enzyme and the substrate, and/or sufficiently strong nonspecific attractive interactions between the enzyme and the product as indicated by Eq. (14). These interactions cause the enzyme to move away from regions of high substrate concentrations and to accumulate in regions of low substrate, performing antichemotaxis. The accumulated enzymes then deplete the substrate, steepening substrate, and product gradients. Steeper gradients further drive the accumulation of enzymes as illustrated in Fig. 4. This positive feedback cycle between enzyme accumulation and reaction is what generates the patterns.
The enzyme accumulation is driven by the antichemotaxis with respect to substrate gradients due to nonspecific interactions. The enzyme chemotaxis considered in some of the models^{6,15,17} has a stabilizing effect. Enzymes accumulate in regions of high substrate concentrations. Then reactions flatten substrate gradients, breaking the feedback that leads to patterns. Hence for such systems spontaneous patterns cannot form. Even in the absence of shortrange nonspecific interactions, antichemotaxis exists due to enhanced diffusivity^{5,13}. Nevertheless, we have seen that enhanced diffusivity alone cannot generate spontaneous patterns for a simple enzymatic reaction. Hence, among all models proposed so far for the enzyme motion, only the model given by Eq. (1), first proposed by Agudo–Canalejo et al.^{18}, can lead to spontaneous pattern formation. The analysis of the patterns generated can shed light onto the microscopic interactions between enzyme and substrate. In fact, from Fig. 3, we can see how the ratio of the maximum over the minimum of the steadystate enzyme profile is affected by the Derjaguin length \( {\lambda }_{{\rm{s}}}^{2}\) and it does not depend on the substrate concentration s_{h}.
The patterns observed in Fig. 2 are not generated via the common shortrange activation and longrange inhibition mechanism^{34}, as neither the enzyme nor the substrate have autocatalytic activity. It is also not a motility induced phase separation (MIPS) mechanism^{35}, which relies on the slowing down of active particles in regions of high particle concentrations. Here a positive feedback mechanism between particle accumulation and reaction leads to the pattern formation. Note that for the system of Eqs. (4), patterns can form even if D_{e} ~ D_{s}, because inequality (9) does not depend on D_{s}, whereas for classical Turing patterns large differences in the diffusion coefficients of the different species are required^{19}. Moreover, it is surprising to see that patterns can form for a singlestep enzymatic reaction with no autocatalytic activity nor allosteric regulation. In fact, for a system where species have a constant diffusion coefficient and enzymatic reactions follow a simple Michaelis–Menten scheme, patterns form for a minimal network of three states, where forward and backward reactions are catalyzed by two different enzymes, respectively^{20}. In our system patterns form for a singlestep catalytic reaction because of crossdiffusion.
Our findings are consistent with recent studies analyzing the effects of crossdiffusion in pattern formation^{36}. Moreover it has been shown that phase separation, formation of static or selfpropelled aggregates can be observed in mixtures of crossdiffusive species interacting via a fast diffusing chemical that can be produced or consumed^{37}. The scenario considered here corresponds to the case of a single crossdiffusive species, i.e., the enzyme, that is able to consume the fast diffusing chemical, i.e., the substrate. Here we considered the full nonlinear forms of enhanced diffusion and crossdiffusion for the enzyme motion, whereas these other studies considered constant diffusion and constant crossdiffusion^{36,37}. The nonlinear model permitted us to address the question why enhanced diffusion alone is not able to generate patterns for a simple enzymatic reaction. We found that enhanced diffusion would need to be more sensitive to perturbations in substrate concentrations than the reaction, see Eq. (12). Although this is not the case for enzymes, we believe that Eq. (12) can characterize the pattern formation of species presenting different enhanced diffusion functions, such as synthetic bacteria^{33}. Moreover, the crossdiffusion terms characterizing the dynamics of \({\mathcal{E}}\) in Eq. (13) can correspond to chemotactic responses of organisms to gradients of \({\mathcal{S}}\) and \({\mathcal{P}}\), which can be chemoattractants or chemorepellants^{38,39}. Hence Eq. (14) can be relevant in the characterization of spontaneous pattern formation of species performing chemotaxis. As similarly shown in Fig. 4, patterns can form if a chemotacting organism is repelled from a chemorepellent and such chemorepellent is also depleted from the organism. Moreover, patterns can also form if the organism is attracted to a chemoattractant and the chemoattractant is produced by the organism. The latter scenario corresponds to the ‘time reversed’ mechanism of the one shown in Fig. 4 and is at the base of aggregation phenomena in chemotacting Amoebae^{40}.
Methods
Simulation parameters
The numerical simulations of the system of Eqs. (4) have been carried out by using COMSOL Multiphysics v5.3. We used the parameters listed in Supplementary Table 1 for the simulations shown in Fig. 2. We used \({\lambda }_{{\rm{f}}}^{{\rm{s}}\ 2}={\lambda }_{{\rm{c}}}^{{\rm{s}}\ 2}=1\ {\mathrm{\AA}}^{2}\) for the unstable homogeneous solution (Fig. 2a) and \({\lambda }_{{\rm{f}}}^{{\rm{s}}\ 2}={\lambda }_{{\rm{c}}}^{{\rm{s}}\ 2}=1\ {\mathrm{\AA}}^{2}\) for the stable homogeneous solution (Fig. 2b). The initial homogeneous concentrations e_{h}, s_{h}, and p_{h} were perturbed by adding white Gaussian noise with variance e_{h} ⋅ 10^{−2}, s_{h} ⋅ 10^{−2}, and p_{h} ⋅ 10^{−2}, respectively. We considered periodic boundary conditions and we used the “Time dependent” solver of COMSOL with a relative tolerance of 10^{−6} and an element size of 0.01 μm, i.e., 10^{4} lattice points.
For the results shown in Fig. 3 we again used the “Time dependent” solver of COMSOL. We considered a grid of 20 × 20 values for \({\lambda }_{{\rm{c}}}^{{\rm{s}}\ 2}={\lambda }_{{\rm{f}}}^{{\rm{s}}\ 2}={\lambda }^{2}\) in the interval [ −10, 3]Å^{2} (evenly spaced), and s_{h} in [10^{1}, 10^{5}] μM (evenly spaced on a logscale). For the COMSOL specific parameters, we used a relative tolerance of 10^{−6} and a maximum element size of 0.02 μm, except when two peaks were observed as a final result of the simulation. In these cases, we repeated the simulations with a finer grid (element size of 0.01 μm) and again we observed a single peak for the concentrations as the final result of the simulations. All the other parameters were the same as of Supplementary Table 1 and the boundary conditions were periodic.
For the simulations shown in Fig. 5 we modified the dynamics of the enzyme, following Eq. (13). We then considered the same simulation parameters as shown in the Supplementary Table 1, the same COMSOL specific parameters used for the simulations depicted in Fig. 2 and we considered the λ_{p} values shown in Fig. 5.
Data availability
The data that support the findings of this study are available from the corresponding author upon reasonable request.
Code availability
The code that supports the findings of this study is available from the corresponding author upon reasonable request.
References
Börsch, M. et al. Conformational changes of the H+ATPase from Escherichia coli upon nucleotide binding detected by single molecule fluorescence. FEBS Lett. 437, 251–254 (1998).
Muddana, H. S., Sengupta, S., Mallouk, T. E., Sen, A. & Butler, P. J. Substrate catalysis enhances singleenzyme diffusion. J. Am. Chem. Soc. 132, 2110–2111 (2010).
Riedel, C. et al. The heat released during catalytic turnover enhances the diffusion of an enzyme. Nature 517, 227–230 (2015).
Illien, P. et al. Exothermicity is not a necessary condition for enhanced diffusion of enzymes. Nano Lett. 17, 4415–4420 (2017).
Jee, A.Y., Dutta, S., Cho, Y.K., Tlusty, T. & Granick, S. Enzyme leaps fuel antichemotaxis. Proc. Natl Acad. Sci USA 115, 14–18 (2018).
Zhao, X. et al. Substratedriven chemotactic assembly in an enzyme cascade. Nat. Chem. 10, 311–317 (2018).
Yu, H., Jo, K., Kounovsky, K. L., Pablo, J. J. D. & Schwartz, D. C. Molecular propulsion: chemical sensing and chemotaxis of DNA driven by RNA polymerase. J. Am. Chem. Soc. 131, 5722–5723 (2009).
Günther, J.P., Börsch, M. & Fischer, P. Diffusion measurements of swimming enzymes with fluorescence correlation spectroscopy. Acc. Chem. Res. 51, 1911–1920 (2018).
Zhang, Y., Armstrong, M. J., Kazeruni, N. M. B. & Hess, H. Aldolase does not show enhanced diffusion in dynamic light scattering experiments. Nano Lett. 18, 8025–8029 (2018).
Jee, A.Y., Chen, K., Tlusty, T., Zhao, J. & Granick, S. Enhanced diffusion and oligomeric enzyme dissociation. J. Am. Chem. Soc. 141, 20062–20068 (2019).
Xu, M., Ross, J. L., Valdez, L. & Sen, A. Direct single molecule imaging of enhanced enzyme diffusion. Phys. Rev. Lett. 123, 128101 (2019).
Zhang, Y. & Hess, H. Enhanced diffusion of catalytically active enzymes. ACS Cent. Sci. 5, 939–948 (2019).
Jee, A.Y., Cho, Y.K., Granick, S. & Tlusty, T. Catalytic enzymes are active matter. Proc. Natl Acad. Sci. USA 115, E10812–E10821 (2018).
Bonnin, E. et al. Mobility of pectin methylesterase in pectin/cellulose gels is enhanced by the presence of cellulose and by its catalytic capacity. Sci. Rep. 9, 1–10 (2019).
Mohajerani, F., Zhao, X., Somasundar, A., Velegol, D. & Sen, A. A theory of enzyme chemotaxis: from experiments to modeling. Biochemistry 57, 6256–6263 (2018).
Weistuch, C. & Pressé, S. Spatiotemporal organization of catalysts driven by enhanced diffusion. J. Phys. Chem. B 122, 5286–5290 (2018).
Schurr, J. M., Fujimoto, B. S., Huynh, L. & Chiu, D. T. A theory of macromolecular chemotaxis. J. Phys. Chem. B 117, 7626–7652 (2013).
AgudoCanalejo, J., Illien, P. & Golestanian, R. Phoresis and enhanced diffusion compete in enzyme chemotaxis. Nano Lett. 18, 2711–2717 (2018).
Turing, A. M. The chemical basis of morphogenesis. Bull. Math. Biol. 52, 153–197 (1990).
Sugai, S. S., Ode, K. L. & Ueda, H. R. A design principle for an autonomous posttranslational pattern formation. Cell Rep. 19, 863–874 (2017).
Illien, P., AdelekeLarodo, T. & Golestanian, R. Diffusion of an enzyme: the role of fluctuationinduced hydrodynamic coupling. EPL (Europhysics Letters) 119, 40002 (2017).
Kondrat, S. & Popescu, M. Brownian dynamics assessment of enhanced diffusion exhibited by ‘fluctuatingdumbbell enzymes’. Phys. Chem. Chem. Phys. 21, 18811–18815 (2019).
Shah, N. B., Hutcheon, M. L., Haarer, B. K. & Duncan, T. M. F1atpase of escherichia coli: the εinhibited state forms after atp hydrolysis, is distinct from the adpinhibited state, and responds dynamically to catalytic site ligands. J. Biol. Chem. 288, 9383–9395 (2013).
Derjaguin, B., Sidorenkov, G., Zubashchenkov, E. & Kiseleva, E. Kinetic phenomena in boundary films of liquids. Kolloidn. zh 9, 335–347 (1947).
Ebbens, S., Tu, M.H., Howse, J. R. & Golestanian, R. Size dependence of the propulsion velocity for catalytic Janussphere swimmers. Phys. Rev. E 85, 020401 (2012).
Anderson, J. L. Colloid transport by interfacial forces. Ann. Rev. Fluid Mech. 21, 61–99 (1989).
McQuarrie, D. Harper’s Chemistry Series (Harper & Row, 1975).
Milo, R. & Phillips, R. Cell Biology by The Numbers (Garland Science, 2015).
Cross, M. & Greenside, H. Pattern Formation and Dynamics in Nonequilibrium Systems by Michael Cross (Cambridge Core, 2009).
ArrioDupont, M., Foucault, G., Vacher, M., Devaux, P. F. & Cribier, S. Translational diffusion of globular proteins in the cytoplasm of cultured muscle cells. Biophys. J. 78, 901–907 (2000).
Erickson, H. P. Size and shape of protein molecules at the nanometer level determined by sedimentation, gel filtration, and electron microscopy. Biol. Proced. Online 11, 32–51 (2009).
BarEven, A. et al. The moderately efficient enzyme: evolutionary and physicochemical trends shaping enzyme parameters. Biochemistry 50, 4402–4410 (2011).
Liu, C. et al. Sequential establishment of stripe patterns in an expanding cell population. Science 334, 238–241 (2011).
Koch, A. J. & Meinhardt, H. Biological pattern formation: from basic mechanisms to complex structures. Rev. Mod. Phys. 66, 1481–1507 (1994).
Cates, M. E. & Tailleur, J. Motilityinduced phase separation. Ann. Rev. Condens. Matter Phys. 6, 219–244 (2015).
Vanag, V. K. & Epstein, I. R. Crossdiffusion and pattern formation in reactiondiffusion systems. Phys. Chem. Chem. Phys. 11, 897–912 (2009).
AgudoCanalejo, J. & Golestanian, R. Active phase separation in mixtures of chemically interacting particles. Phys. Rev. Lett. 123, 018101 (2019).
Woodward, D. et al. Spatiotemporal patterns generated by salmonella typhimurium. Biophys. J. 68, 2181–2189 (1995).
Tweedy, L., Knecht, D. A., Mackay, G. M. & Insall, R. H. Selfgenerated chemoattractant gradients: attractant depletion extends the range and robustness of chemotaxis. PLoS Biol. 14, e1002404 (2016).
Keller, E. F. & Segel, L. A. Initiation of slime mold aggregation viewed as an instability. J. Theor. Biol. 26, 399–415 (1970).
Acknowledgements
This work was funded by the Deutsche Forschungsgemeinschaft (DFG) within the framework of the Transregio 174 “Spatiotemporal dynamics of bacterial cells". G.G. was supported by a DFG Fellowship through the Graduate School of Quantitative Biosciences Munich (QBM). Open access funding provided by Projekt DEAL.
Author information
Authors and Affiliations
Contributions
All authors designed the research. G.G. and H.S.A. performed the analysis and the simulations. All authors wrote the manuscript.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Giunta, G., SeyedAllaei, H. & Gerland, U. Crossdiffusion induced patterns for a singlestep enzymatic reaction. Commun Phys 3, 167 (2020). https://doi.org/10.1038/s4200502000427w
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s4200502000427w
This article is cited by

Selforganization of primitive metabolic cycles due to nonreciprocal interactions
Nature Communications (2023)

Optimal spatial allocation of enzymes as an investment problem
Communications Physics (2022)
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.