Chapter II: Polyhedral constraints enable holistic analysis of bioregulation
3.4 Detailed balance steady states of binding networks
Since binding reactions between molecules tend to happen at a faster time scale than the phenomena we are interested in, e.g. production and degradation of molecules, it is natural to assume that the binding reactions have reached a steady state. In fact, we can further assume that the steady states the binding reactions reach are at equilibrium, based on the intuition that nonequilibrium steady states require continuous energy input at the fast time scale of binding reactions, which is costly and therefore unlikely. In other words, although a mathematically specified CRN may not have equilibrium steady states, and may have steady states that are not at equilibrium, it is physically plausible to assume that a binding network on a fast time scale should have equilibrium steady states and practically stay in equilibrium steady states. Therefore, in this work we focus our study of equilibrium steady states in a way that does not concern whether they exist from mathematical specifications.
In this section, we describe the manifold of equilibrium steady states, which are also called detailed balance steady states in chemical reaction network theory.
Letπ₯βRπ>0 denote a vector of species concentrations in a binding network. Then detailed balance steady states are ones that satisfy the condition that the forward (dissociation) and backward (association) fluxes of each reaction balances out. Letππ = π
β π
π+π , the ratio between the forward (dissociation) and backward (association) reaction rates, be the equilibrium constant of theπth binding reaction (in the dissociation direction, so also called dissociation constants). Assuming mass-action kinetics, then the balancing of forward and backward fluxes corresponds toππβπ₯πΌππ =π+π π₯π½ππ, whereπΌππ,π½ππ β Zπ are the reactant and product vectors of theπth forward (dissociation) reaction. This can be written asπ₯πΎππ =ππ, where πΎππ =π½ππβπΌππ βZπis the stoichiometry vector of theπth forward (dissociation) reaction. Let πdenote the rank of the stoichiometry matrixΞ βRπΓπ, which is also the rank ofΞπ βRπΓπ2, the stoichiometry of only the forward reactions. Then we can always select π forward reactions with linearly independent stoichiometry vector to form the transpose-reduced stoichiometry matrixπ, whose rows are selected columns ofΞπ. Similarly, we take the equilibrium constants for theπreactions selected to form vectorπβRπ>0. Taking log and write in matrix form, we have that the detailed balanced condition becomes
πlogπ₯= logπ. (3.11)
Since there areπvariables andπequations, we see the solutionlogπ₯of this equation for a fixedπhasπ=πβπdegrees of freedom.
What variables could represent thisπdegrees of freedom for the detailed balance solutions logπ₯? One natural choice are the conserved quantities. LetπΏβRπΓπbe the conservation law matrix, then the conserved quantities are defined asπ‘=πΏπ₯. The conserved quantities are denoted asπ‘for totals since in binding networks they often correspond physically to the total of some species in various forms. So we see theπtotal concentrations π‘ β Rπ>0 form a natural representation of the remainingπdegrees of freedom. If we further require the binding network to be isomer-atomic, then alternatively we can represent theπdegrees of freedom as the concentrations of atomic species. We will see this explicitly as alternative coordinate charts for the manifold of detailed balance steady states.
Formally, we define the manifold for the detailed balanced solutions of a binding network.
We consider this manifold as the set of all possible values the variables of interest can take for a given binding network. The variables of interest from our above discussion are the species concentration π₯ β Rπ>0, total concentrationsπ‘ β Rπ>0, and equilibrium constants
πβRπ>0. The manifold is then defined by the constraint on these variables imposed by the detailed balance condition of a particular binding network.
Definition 3.4.1. Given a binding network with reduced stoichiometry matrixπ βRπΓπ. LetπβRπ>0be the equilibrium constants for theπreactions selected. LetπΏβRπΓπ,π=πβπ be the unique conservation law matrix of the network. Then the manifold of detailed balance steady statesof the binding network, also calledequilibrium manifoldof the binding network for short, is
β³={οΈ(π₯,π‘,π)βR2π>0 :π‘ =πΏπ₯,πlogπ₯= logπ}οΈ. (3.12) Note thatβ³isπ-dimensional and immersed inR2π>0. We denote a point in β³as vector πβR2π>0. When convenient, we could also consider thelogmanifold immersed inR2π.
logβ³:={οΈ(logπ₯,logπ‘,logπ)βR2π: (π₯,π‘,π)β β³}οΈ. (3.13) We caution that this approach of defining β³considers a given binding network as a detailed-balanceconstrainton the values(π₯,π‘,π)can take. The detailed balance condition takes higher priority than any specification of rates of a CRN. This is different from the typical approaches in CRN theory where the specification of network stoichiometry and rates take precedence, and the existence of detailed balance steady state is then determined later from the CRN specification. Because of our taking detailed balance as higher priority than CRN specifications, several CRNs can be equivalent to the same detailed balance behavior. In other words, if detailed balance is guaranteed, then specifying the full CRN may be redundant. For example, consider 3-state transitionπΆ1 βπΆ2 β πΆ3 βπΆ1. With detailed balance given, this network is equivalent to the same network with one reaction deleted. Because of this, any nonequilibrium steady states are also automatically excluded.
Coordinate charts of the equilibrium manifold
With the equilibrium manifoldβ³defined for a given binding networks, we would like to describe it and characterize its properties. The first thing to do with a smooth manifold is to specify coordinate charts on it, so that we have a parameterization of points on the manifold, and can talk about its tangent bundle for how to move around the manifold.
β³is anπ-dimensional manifold. It is naturally immersed in2π-dimensional space, where each point is specified asπ= (π₯,π‘,π)βR2π>0. But these variables are further related byπ equations, namelyπ‘ =πΏπ₯andπlogπ₯= logπ. So we would like a parameterization ofβ³ that maps its points toRπin a one-to-one or invertible way, so that theπdegrees of freedom
inβ³are now explicitly some variable inRπ. Such parameterizations are calledcoordinate charts, and a collection of charts that together cover the whole manifold is called anatlas.
Forβ³, we will see below that the natural choices of atlas we discuss have just one chart.
The most natural choice to begin with is of course the species concentrations logπ₯. By slight abuse of notation, this corresponds to the maplogπ₯ :β³ β Rπthat takes a point specified asπ= (π₯,π‘,π)β β³and output the log of the firstπvariableslogπ₯. This map is invertible because givenlogπ₯, we can findπ‘andπ uniquely. Also, this chart is an atlas because every point in β³can be represented in this way. So instead of using a vector π β R2π>0 to denote a point on β³, we could equivalently use π₯ β Rπ>0. We can use the diagram below to represent this mapping:
π= (π₯,π‘,π)β β³βββββββββ½ββββββββlogπ₯ β
(π₯,πΏπ₯,πlogπ₯)
logπ₯βRπ. (3.14)
We also want alternative charts for different purposes. For example, if we consider changing the steady states of a given binding network by adjusting the concentrations. In this case, althoughπ₯varies, the equilibrium constantsπare not modified by such changes, therefore should remain constant. So to study changes in concentrations in this case, we would likeπto appear explicitly as variables in our parameterization. Another reason we may want this is to study what changes to the system would be caused by modifying the equilibrium constantsπ, which is natural when asking questions about energy of molecules or temperature.
To obtain coordinate charts withπas variables, we first note thatπβRπ>0is ofπdimensions, so to parameterize theπdimensions ofβ³, we need anotherπvariables. A simple choice that is still close to thelogπ₯chart is to take πof theπ₯variables. Assuming our binding network is isomer-atomic, then one natural choice for this isπ₯πβRπ>0, concentrations for theπatomic species. This yields chart(logπ₯π,logπ).
We can investigate properties of this chart by how it is mapped from thelogπ₯chart. To write the map, we re-order the species so that the firstπspecies are atomic. Soπ₯= (π₯π,π₯π) is split into two parts, theπatomic speciesπ₯π βRπ, andπcomplex speciesπ₯πβRπ. Using detailed balanced condition Eq (3.11), we have
β‘
β£
logπ₯π logπ
β€
β¦=
β‘
β£
Iπ 0 π1 π2
β€
β¦logπ₯.
Invert this expression and use thatπΏβΊ2 =βπ2β1π1 yields an explicit alternative parameter-
ization of detailed balance steady states:
{logπ₯βRπ:πlogπ₯= logπ}
=
β§
β¨
β©
logπ₯=πΏβΊlogπ₯π+
β‘
β£
0 π2β1
β€
β¦logπ: logπ₯π βRπ,logπβRπ
β«
β¬
β
=
β§
β¨
β©
logπ₯=
β‘
β£
Iπ 0 πΏβΊ2 π2β1
β€
β¦
β‘
β£
logπ₯π logπ
β€
β¦: logπ₯πβRπ,logπβRπ
β«
β¬
β
.
(3.15)
So we see that the map between chartlogπ₯and chart(logπ₯π,logπ)is linear, so we represent this map explicitly as matrices in the following diagram:
logπ₯
β‘
β£
Iπ 0 π1 π2
β€
β¦
βββββββββ
β½ββββββββ
β‘
β£
Iπ 0 πΏβΊ2 π2β1
β€
β¦
(logπ₯π,logπ). (3.16)
Since this map is invertible, we know(logπ₯π,logπ)is also a one-chart atlas forβ³.
Although chart(logπ₯π,π)containsπas explicit variables, there are still scenarios where we may want an alternative. For one, we need the isomer-atomic assumption to have the atomic species. This may not hold in general. More importantly, in many scenarios when the concentrations in a binding network is adjusted, it is not by adjusting the atomic speciesβ
concentration, but by adjusting the conserved quantities or the totals (see Section3.1). This is especially often the case for time-scale separation for binding and catalysis reactions, where binding reaches equilibrium steady state while catalysis produces or degrades molecules to change concentrations. Here when a molecule is produced or degraded by catalysis, whether it is bound or not, or in which state among the state transitions, is not distinguishable from the slow time scale of catalysis. Once a molecule is produced or degraded in a particular form or state, the binding network quickly equilibrates and the net change of one molecule is then on the total, not any particular form or state.
From this reason, we would like to have the chart (logπ‘,logπ), where π‘ = πΏπ₯ is the totals or conserved quantities. To describe the map between this chart and chatlogπ₯is more involved, as can be seen from the mixing of linear mapπ‘ =πΏπ₯and log-linear map logπ=π logπ₯. We delve into this in the next section.