Chapter II: Polyhedral constraints enable holistic analysis of bioregulation
3.6 Polyhedral shape of log derivatives in one binding reaction
So we know the atomic chart with matrixπ΄=π΄π=[οΈIπ 0]οΈand the total chart withπ΄=πΏ both resides in the same connected component ofπ. Exactly which one depends on the choice of which direction is used for the stoichiometry vectors. Thus, without loss of generality, we assumeπ΄π,πΏβ π+(π).
To find other alternative chart, we note that given aπ΄β π+(π), then for anyπ βGL+π(R) such thatππ΄βRπΓπβ₯0 , thenππ΄β π+(π). HereGL+π(R)is the identity component of the general linear group ofπΓπmatrices, i.e. πΓπmatrices with positive determinant. This states that any invertible matrix with positive determinant can be left-multiplied toπ΄, and if the resulting matrix is non-negative, then it is inπ+(π). There are several elementary matrices of GL+π(R)worth noting. It includes positive scaling ΞπΌ, where πΌ β Rπ>0 is a positive vector inRπ. It also includes permutations with positive sign, i.e. consisting of an even number of transpositions. It also includes row additionsI+ππΈππ, whereπ ΜΈ=π, π, π β {1, . . . , π}, πΈππ has 1at (π, π)entry and zero everywhere else, and π β R is a real number. I+ππΈππ takes theπth row ofπ΄, multiply byπ, and adds to theπth row ofπ΄. We caution that combinations of these elementary operations may go out ofGL+π(R), and the resulting matrix may no longer be non-negative.
As for right multiplication, givenπ΄β π+(π), thenπ΄ΞπΌ β π+(π)for any positive vector πΌ β Rπ>0. This is because this multiplication is just a scaling of variables π₯, without changing its domainRπ>0.
These operations give a basic approach to explore other alternative charts ofβ³. It is an interesting question for further research to characterize the setπ+(π).
3.6 Polyhedral shape of log derivatives in one binding reac-
particular log derivative above, πlogπ₯π
π(logπ‘,logπ), if π₯π is the active catalytic species, hence the name reaction order. Reaction order can also be equivalently interpreted as infinitesimal fold change, or the exponents of a local expression ofπ₯πin terms ofπ‘andπ.
This does not answer the question of why reaction orders are used to study bioregulation though. For example, why the linear direction ππ₯π
π(π‘,π) is not used? We demonstrate several useful properties of reaction orders in this section by examining the reaction orders of a single binding reaction in close detail. The key properties we demonstrate are summarized here. (1) Constrained by binding network stoichiometry, reaction orders are bounded within some polyhedral set, with vertices correspond to biologically meaningful regimes.
This is in contrast to linear derivatives, which are unbounded in most directions. (2) The polyhedral sets has hierarchical structures corresponding to robustness to variations, so that points concentrate to vertices, edges, and faces under asymptotic limits. This means reaction orders are robust to variations in rates and concentrations, so it can be controlled to precise values using noisy and uncertain actuations, and robust behaviors can be built on top of it.
Explicit reaction order calculation for one binding reaction
We do explicit calculation for one binding reaction to reveal the geometric shape of log derivatives. For transparency, we calculate the log derivatives directly without using any of the formula developed in previous section, although they produce the same result.
A binding network consisting of just one binding reaction can be written as follows:
πΈ+π π
ββ+
β½β
πβ
πΆ, (3.25)
where component species enzymeπΈand substrateπbind to form complexπΆ, andπ+and πβ are forward and backward reaction rate constants. Although we used enzyme and substrate to denote the two component species out of tradition, we do not assume they have any special properties, and by symmetryπΈandπ are equivalent by re-labeling.
This binding reaction has the following deterministic dynamics from the law of mass-action:
π
ππ‘πΆ(π‘) = π+πΈ(π‘)π(π‘)βπβπΆ(π‘), (3.26) where by slight abuse of notation, the symbol for a species is also used to denote the concentration of that species. This system has steady-state equation
πΈπ =πΎπΆ, (3.27)
whereπΎ := ππβ+ is the equilibrium constant in the dissociation direction. As guaranteed, this corresponds to πlogπ₯ = logπ, where π = [οΈβ1 β1 1]οΈ, π₯ = (πΈ, π, πΆ), and π = πΎ. Note that in this particular case, the steady state necessarily satisfy detailed balance, so we do not need to begin with detailed balance condition to defineβ³, but instead deriveβ³ from the specified rates.
The conserved quantities of this binding reaction are the total enzyme and total substrate:
π‘π :=π+πΆ, π‘πΈ :=πΈ+πΆ.
(3.28) This can be written as π‘ = πΏπ₯ with the conservation matrix below. Here the species ordering is(πΈ, π, πΆ), and the total ordering is(π‘πΈ, π‘π). We also attach the stoichiometry matrix here for clear comparison.
β‘
β£
πΏ π
β€
β¦=
β‘
β’
β’
β’
β£
1 0 1 0 1 1 1 1 β1
β€
β₯
β₯
β₯
β¦
(3.29)
This in turn defines the equilibrium manifold:
β³={οΈ(π₯,π‘,π)βR6>0 :π logπ₯= logπ,π‘ =πΏπ₯}οΈ
={οΈ(πΈ, π, πΆ, π‘πΈ, π‘π, πΎ)βR6>0 :πΈπ =πΎπΆ, π‘π =π+πΆ, π‘πΈ =πΈ+πΆ}οΈ.
(3.30) In this case, we can explicitly solve for π₯ = (πΈ, π, πΆ) expressed in (π‘,π) = (π‘πΈ, π‘π, πΎ). Namely,
2πΆ =π‘πΈ+π‘π+πΎββοΈ(π‘πΈ +π‘π+πΎ)2β4π‘πΈπ‘π, (3.31) which is then easily used to derive expressions forπΈ andπ. This formula can be used as the exact solution for us to compare with in this case. But for the derivative to follow the workflow in the general case as described in Section3.4, we do not rely on this formula.
Instead, we use the implicit function theorem as we did in the general case to calculate the reaction order, i.e. the log derivative πlog(πΈ,π,πΆ)
πlog(π‘πΈ,π‘π,πΎ). Define πΉ :R6>0 βR3 whose zero set is the equilibrium manifoldβ³:
πΉ(πΈ, π, πΆ, π‘πΈ, π‘π, πΎ) =
β‘
β£
πΉ1(πΈ, π, πΆ, π‘πΈ, π‘π) πΉ2(πΈ, π, πΆ, πΎ)
β€
β¦=
β‘
β’
β’
β’
β£
πΈ+πΆβπ‘πΈ π+πΆβπ‘π πΈπβπΆπΎ
β€
β₯
β₯
β₯
β¦
, (3.32)
By implicit function theorem,
π(πΈ, π, πΆ)
π(π‘πΈ, π‘π, πΎ) =β
β‘
β’
β’
β’
β£
1 0 1 0 1 1 π πΈ βπΎ
β€
β₯
β₯
β₯
β¦
β1β‘
β’
β’
β’
β£
β1 0 0 0 β1 0 0 0 βπΆ
β€
β₯
β₯
β₯
β¦
= 1
πΈ+π+πΎ
β‘
β’
β’
β’
β£
πΈ+πΎ βπΈ πΆ
βπ π+πΎ πΆ
π πΈ βπΆ
β€
β₯
β₯
β₯
β¦
.
SinceπΆ = πΈππΎ , we can express the above in dimensionless quantitiesπ:= πΎπΈ andπ = πΎπ.
π(πΈ, π, πΆ)
π(π‘πΈ, π‘π, πΎ) = 1 1 +π+π
β‘
β’
β’
β’
β£
1 +π βπ ππ
βπ 1 +π ππ
π π βππ
β€
β₯
β₯
β₯
β¦
. (3.33)
To calculate log derivative, we note that πlogπ(π₯)
πlogπ₯ = Ξβ1π ππ(π₯)ππ₯ Ξπ₯, whereΞπ£ = diag(π£), since
πlogππ(π₯)
πlogπ₯π = ππππ₯π(π₯)
π
π₯π
ππ(π₯). Therefore,
πlog(πΈ, π, πΆ)
πlog(π‘πΈ, π‘π, πΎ) = 1 πΈ+π+πΎ
β‘
β’
β’
β’
β£
πΈ 0 0 0 π 0 0 0 πΆ
β€
β₯
β₯
β₯
β¦
β1β‘
β’
β’
β’
β£
πΈ+πΎ βπΈ πΆ
βπ π+πΎ πΆ
π πΈ βπΆ
β€
β₯
β₯
β₯
β¦
β‘
β’
β’
β’
β£
π‘πΈ 0 0 0 π‘π 0 0 0 πΎ
β€
β₯
β₯
β₯
β¦
= 1
πΈ+π+πΎ
β‘
β’
β’
β’
β£
(1 + πΎπ)(πΈ+πΎ) βπ‘π π
βπ‘πΈ (1 + πΈπΎ)(π+πΎ) πΈ
π+πΎ πΈ+πΎ βπΎ
β€
β₯
β₯
β₯
β¦
.
Expressing this in terms ofπandπ yields the following result.
Theorem 3.6.1(Reaction orders of a simple binding reaction). The reaction orders of a simple binding reactionπΈ+π π
ββ+
β½β
πβ
πΆ can be expressed as
πlog(πΈ, π, πΆ)
πlog(π‘πΈ, π‘π, πΎ) = 1 1 +π+π
β‘
β’
β’
β’
β£
(1 +π)(1 +π ) βπ (1 +π) π
βπ(1 +π ) (1 +π )(1 +π) π
1 +π 1 +π β1
β€
β₯
β₯
β₯
β¦
, (3.34)
where π = πΎπΈ, π = πΎπ, and πΎ := ππβ+ the equilibrium constant of this binding reaction in the dissociation direction.
Below we re-do the calculation for reaction order using the formula in Eq (3.19) to show that indeed they yield the same result.
Example 5(Simple binding reaction.). πΈ+π βπΆwith binding constantπΎ. Let variables be π₯= (πΈ, π, πΆ),π‘ = (π‘πΈ, π‘π)andπ=πΎ. The corresponding stoichiometry matrix (note we use the forward or dissociation direction), conservation laws, and part of the matrix to be inverted are
π =[οΈ1 1 β1]οΈ, πΏ=
β‘
β£
1 0 1 0 1 1
β€
β¦,Ξβ1π‘ πΏΞπ₯=
β‘
β£
πΈ
π‘πΈ 0 πΆπ‘πΈπ
πΈ
0 π‘π
π
πΆπΈπ
π‘π
β€
β¦
πlog(πΈ, π, πΆ)
πlog(π‘πΈ, π‘π, πΎ) =
β‘
β’
β’
β’
β£
πΈ π‘πΈ 0 π‘πΆ
πΈ
0 π‘π
π
πΆ π‘π
1 1 β1
β€
β₯
β₯
β₯
β¦
β1
=
(οΈπΆπΈ+πΆπ+πΈπ π‘πΈπ‘π
)οΈβ1
β‘
β’
β’
β’
β£
1 βπ‘πΆ
πΈ
πΆπ π‘πΈπ‘π
βπ‘πΆ
π 1 π‘πΆπΈ
πΈπ‘π
π π‘π
πΈ
π‘πΈ βπ‘πΈπ
πΈπ‘π
β€
β₯
β₯
β₯
β¦
.
Now we express the above in terms of(π, π ) := (πΎπΈ,πΎπ), we get the same result as Eq (3.34).
We can also utilize the stoichiometry-atomic property of this network, with atomic species π₯π= (πΈ, π). This splits the stoichiometry and the conservation law matrix into atomic and non-atomic parts:
π =[οΈπ1 π2]οΈ=[οΈ1 1 β1]οΈ, πΏ=[οΈI2 πΏ2]οΈ=
β‘
β£
1 0 1 0 1 1
β€
β¦, and the core symmetric structure is
(πΏΞπ₯πΏβΊ)β1 =
β‘
β£
πΈ+πΆ πΆ πΆ π+πΆ
β€
β¦
β1
= 1
πΈπ+πΆπ+ππΆ
β‘
β£
π+πΆ βπΆ
βπΆ πΈ+πΆ
β€
β¦= 1
1 +π+π
β‘
β£
1 + 1π β1
β1 1 + 1π
β€
β¦.
Soπ2β1 =β1,βπΏ2Ξπ₯ππ2β1 =
β‘
β£
πΆ πΆ
β€
β¦. From this we can calculate
πlog(πΈ, π)
πlog(π‘πΈ, π‘π) = (πΏΞπ₯πΏβΊ)β1Ξπ‘= 1 1 +π+π
β‘
β£
(1 +π)(1 +π ) βπ (1 +π)
βπ(1 +π ) (1 +π)(1 +π )
β€
β¦,
πlog(πΈ, π)
πlogπΎ =β(πΏΞπ₯πΏβΊ)β1πΏ2Ξπ₯ππ2β1 = 1 1 +π+π
β‘
β£
π π
β€
β¦,
πlogπΆ
πlog(π‘πΈ, π‘π, πΎ) =πΏβΊ2 πlog(πΈ, π)
πlog(π‘πΈ, π‘π, πΎ)+[οΈ0 π2β1]οΈ= 1 1 +π+π
[οΈ1 +π 1 +π β1]οΈ.
β³
These calculations have yielded an explicit formula for the reaction orders of one binding reaction (Eq (3.34)). We derived this formula in both ways, one by explicitly step-by-step calculate through the definition of manifold and implicit function theorem, as an illustration of the abstract derivations in Section3.4, and another by directly applying the resulting log
derivative formula in Eq (3.19) from our general derivations. Below, we use these formula to investigate what properties the reaction orders have, and how are these properties related to biological behaviors.
Polyhedral shape of reaction order and its biological implications
We use the formula Eq (3.34) to study the properties of reaction orders and their biological implications.
We first look at the overall shape of all possible values that the reaction orders can take.
In order to visualize in 2D, we can consider the equilibrium constant πΎ as fixed and normalize all the variables so that πΎ is the unit of concentration. This encodes our biological assumption that we focus on changes to reaction orders caused by varying total concentrationsπ‘πΈ, π‘π, rather than varying the equilibrium constantπΎ. This makes sense when the biomolecules making up the binding network are already fixed, and we are studying how to regulate it. If instead we are studying what biomolecules to use to achieve a certain behavior in a binding network, then we should allowπΎ to vary and be an explicit variable in the reaction orders. For this particular example, the value thatπΎ is fixed at does not matter because Eq (3.34) shows that the value of the reaction orders can take only depends on(π, π ), the ratios ofπΈ andπ toπΎ, so the range of values that the reaction orders can take is independent of what valueπΎ is fixed at. Now withπΎ held fixed and normalized away, the reaction orders of interest are reaction order of the complexπΆto total enzyme and total substrate πlogπΆ
πlog(π‘πΈ,π‘π), and reaction order of free enzyme to total enzyme and total substrate πlogπΈ
πlog(π‘πΈ,π‘π). Note that π and πΈ are symmetric since we have just one binding reaction. Which speciesβ reaction order is of interest depends on which species is catalytically active. For example, ifπΆ is the active complex for downstream catalysis, such as a gene bound with an activating transcription factor or the activated form of an enzyme or receptor, then reaction orders ofπΆcorresponds to response of biological activity. On the other hand, ifπ is a repressive ligand or repressing transcription factor, then the free form πΈ is catalytically active, so the reaction order ofπΈcorresponds to biological activity.
We computationally sample the values of reaction orders in Eq (3.34) to obtain Figure 3.5. Several features arise from visual inspection. First, it appears that these sets take a polyhedral shape. Second, the polyhedral shape suggests it can be formed as combinations of simpler shapes, e.g. vertices and edges. Third, the points concentrate at the edges and vertices, suggesting robustness at those locations. Lastly, the reaction order ofπΈ is unbounded, going towards infinity in the(β1,1)direction, suggesting hypersensitivity.
We discuss these features in detail below.
(a) (b)
Figure 3.5Sampling of the log derivative ofπΆ(subfigure(a)) andπΈ(subfigure(b)) with respect toπ‘πΈandπ‘π for the binding network with just one binding reactionπΈ+πβπΆ. Sampling is via the variablesπ= πΎπΈ and π = πΎπ taking values between10β6to106, uniformly sampled on the log scale.
Polyhedral shape. The first thing we notice is that the overall shape for bothπΆ andπΈβs reaction orders arepolyhedral. This implies that the full range of reaction orders can be captured as the convex combination of several extreme cases. This can be seen in an explicit and analytical way as follows. TakeπΆβs reaction orders for example. We see from the sampled shape that it is the convex combination of vertices (1,0), (0,1), and(1,1). This means any point in the set can be expressed as π1(1,0) +π2(0,1) +π3(1,1)with ππ non-negative and sum to oneβοΈ3
π=1ππ = 1. We can then compare this expression with the formula in Eq (3.34) to have the following equation. Note that to be consistent with the figure, we have taken the order for the totals to be(π‘π, π‘πΈ)instead of(π‘πΈ, π‘π).
πlogπΆ
πlog(π‘π, π‘πΈ) =[οΈ1+π+π 1+π 1+π+π 1+π ]οΈ=[οΈπ1 +π3 π2+π3]οΈ. (3.35) This together with thatππβs sum to1can be used to formulate the following linear system of equations forππβs:
β‘
β’
β’
β’
β£
1 1 1 1 0 1 0 1 1
β€
β₯
β₯
β₯
β¦
β‘
β’
β’
β’
β£
π1 π2 π3
β€
β₯
β₯
β₯
β¦
=
β‘
β’
β’
β’
β£
1
πlogπΆ
πlogπ‘π
πlogπΆ
πlogπ‘πΈ
β€
β₯
β₯
β₯
β¦
=
β‘
β’
β’
β’
β£
1
1+π 1+π+π
1+π 1+π+π
β€
β₯
β₯
β₯
β¦
. (3.36)
Solving this linear system of equations yields
β‘
β’
β’
β’
β£
π1 π2 π3
β€
β₯
β₯
β₯
β¦
=
β‘
β’
β’
β’
β£
1 0 β1 1 β1 0
β1 1 1
β€
β₯
β₯
β₯
β¦
β‘
β’
β’
β’
β£
1
1+π 1+π+π
1+π 1+π+π
β€
β₯
β₯
β₯
β¦
= 1
1 +π+π
β‘
β’
β’
β’
β£
π π 1
β€
β₯
β₯
β₯
β¦
. (3.37)
This tells us that we can write the reaction orders of πΆ in an explicit way as convex combinations of a set of vertices:
πlogπΆ
πlog(π‘π, π‘πΈ) = π
1 +π+π (1,0) + π
1 +π+π (0,1) + 1
1 +π+π (1,1). (3.38) As a quick note, although log derivatives are often used to compute βweighted average of exponentsβ for polynomials in statistical mechanics, and the log derivatives form a convex polytope as well, the polytope forπΆ above is not obtainable from a polynomial.
For a polynomialπ(π₯)in variables taking positive valuesπ₯βRπ>0, the log derivatives are contained in its Newton polytope, which is the polytope formed as a convex combination of the exponents for each monomial term. For example, ifπ(π₯) = 1 +π₯1+π₯2, then the log derivative πlogπ
πlog(π₯1,π₯2) is contained in polytope with vertices(0,0),(1,0), and(0,1), and the convex coefficients are 1
1+π₯1+π₯2
, π₯1
1+π₯1+π₯2
, and π₯2
1+π₯1+π₯2
respectively. We see that although these coefficients are the same as the coefficients in Eq (3.38), one of the vertices are different.
If we want a polynomial with the(1,1)vertex and similar coefficients, then the polynomial can be π(π₯) = π₯1 +π₯2 +π₯1π₯2. But then the coefficient for the (1,1) vertex is π₯1π₯2
π₯1+π₯2+π₯1π₯2
, a different form compared to that in Eq (3.38). So we see polynomials cannot yield the convex combination in Eq (3.38). Either the vertices are different, or the convex coefficients (i.e. when the vertices are achieved) are different. Indeed, we know Eq (3.38) comes from taking the log derivative of the explicit expression in Eq (3.31) which involves square roots.
More generally, reaction order polyhedra come from log derivative of possibly non-analytic expressions that are roots of systems of polynomial equations.
Employing a similar approach we can parameterize a point in the reaction order polyhedron forπΈasπ1(0,1) +π2(β1,1) +π(β1,1), withπ1, π2, π β₯0andπ1 +π2 = 1. This expression uses our observation from Figure3.5bthat the polyhedron is defined by two vertices at (0,1)and(β1,1), and a ray in direction(β1,1). The equation relating these coefficients to
the reaction order expression in Eq (3.34) is
β‘
β’
β’
β’
β£
1 1 0 0 β1 β1 1 1 1
β€
β₯
β₯
β₯
β¦
β‘
β’
β’
β’
β£
π1 π2 π
β€
β₯
β₯
β₯
β¦
=
β‘
β’
β’
β’
β£
1
πlogπΈ
πlogπ‘π
πlogπΈ
πlogπ‘πΈ
β€
β₯
β₯
β₯
β¦
=
β‘
β’
β’
β’
β£
1
βπ (1+π) 1+π+π (1+π)(1+π )
1+π+π
β€
β₯
β₯
β₯
β¦
(3.39)
Solve this system of linear equations yield
β‘
β’
β’
β’
β£
π1 π2 π
β€
β₯
β₯
β₯
β¦
=
β‘
β’
β’
β’
β£
0 1 1 1 β1 β1
β1 0 1
β€
β₯
β₯
β₯
β¦
β‘
β’
β’
β’
β£
1
βπ (1+π) 1+π+π (1+π)(1+π )
1+π+π
β€
β₯
β₯
β₯
β¦
= 1
1 +π+π
β‘
β’
β’
β’
β£
1 +π π ππ
β€
β₯
β₯
β₯
β¦
. (3.40)
So we can explicitly write
πlogπΈ
πlog(π‘π, π‘πΈ) = 1 +π
1 +π+π (0,1) + π
1 +π+π (β1,1) + ππ
1 +π+π (β1,1). (3.41) We can also use the above results to write the reaction orders of all species in the form a polyhedron. Defineππ = 1+π+π π ,ππ = 1+π+π π ,π1 = 1+π+π 1 , andπ = 1+π+π ππ . Then
πlog(π, πΈ, πΆ)
πlog(π‘π, π‘πΈ, πΎ) =ππ
β‘
β’
β’
β’
β£
1 β1 1 0 1 0 1 0 0
β€
β₯
β₯
β₯
β¦
+ππ
β‘
β’
β’
β’
β£
1 0 0
β1 1 1 0 1 0
β€
β₯
β₯
β₯
β¦
+π1
β‘
β’
β’
β’
β£
1 0 0 0 1 0 1 1 β1
β€
β₯
β₯
β₯
β¦
+π
β‘
β’
β’
β’
β£
1 β1 0
β1 1 0 0 0 0
β€
β₯
β₯
β₯
β¦
. (3.42)
From the above, we see both through computational sampling and analytical derivations that indeed the range of values that reaction orders can take form a polyhedral set. The mathematical reason for this polyhedral shape is studied in Section3.8. We investigate the biological implications of the polyhedral shape below.
Vertices and edges as asymptotic approximations. Roughly speaking, a polyhedral set is the set formed by the convex combination of vertices (with vertices generalized to include rays as well). This suggests we can consider the general behavior of a catalysis reaction, i.e.
the reaction order of the active species in a binding network, as the βconvex combinationβ
of the behaviors at the vertices of reaction order polyhedron. If the behavior is simple at the vertices, then this gives us a way to describe the complicated general behavior through simple extreme-case behavior at the vertices.
We take theπΆ polyhedron to illustrate this. From Eq (3.38), we see each vertex is achieved at a certain extreme of the(π, π )variables. Namely, (1,0)is achieved whenπ β«1, π (πis much larger than1and π ),(0,1)is achieved whenπ β« 1, π and(1,1)is achieved when 1β«π, π . We can relate these verticesβ reaction order to the approximate expressions ofπΆ in(π‘π, π‘πΈ, πΎ)at these vertices by applying these asymptotic conditions to the equations defining the equilibrium manifold in Eq (3.30), or to the explicit solution in Eq (3.31).
Alternatively, we can include theπΎ variable in the reaction orders to see the vertices are
πlogπΆ
πlog(π‘π,π‘πΈ,πΎ) taking values(1,0,0),(0,1,0), and(1,1,β1), which we can integrate to getπΆ
expressed in(π‘π, π‘πΈ, πΎ)with a multiplicative constant which we can set to1. The result is summarized as follows,
πΆ β
β§
βͺβͺ
βͺβͺ
β¨
βͺβͺ
βͺβͺ
β©
π‘πΈ, reaction order(0,1), π‘π β«π‘πΈ, πΎ;
π‘ππ‘πΈ
πΎ , reaction order(1,1), πΎ β«π‘πΈ, π‘π; π‘π, reaction order(1,0), π‘πΈ β«π‘π, πΎ.
(3.43)
Each vertex corresponds to a biologically meaningfulregimethat the reaction can operate in. When the total substrateπ‘π is very large, the enzymes are saturated so that the speed of catalysis is only limited by the enzyme, thereforeπΆ βπ‘πΈ, corresponding to vertex(0,1). When it is the other way around and the total enzyme is very large and total substrate is limiting,πΆ β π‘π, corresponding to vertex(1,0). When enzyme and substrate are not abundant relative to the binding affinityπΎ, the speed of catalysis is limited by the formation of complexπΆ. In this regime, the complexπΆis low in number compared to total enzyme and total substrate, and increasing either enzyme and substrate creates more complexes.
Therefore πΆ β π‘ππΎπ‘πΈ, corresponding to vertex(1,1). Together, we see that the polyhedral shape highlights the vertices as theregimesthat a catalysis reaction can operate in, which corresponds to extreme cases of concentrations and equilibrium constants. The asymptotic conditions of these extreme cases in turn yield asymptotic approximations of the active species (πΆ in this case) that are simple monomials and biologically interpretable. The general complicated behavior can then be considered as varying in between these simple extreme-case regimes.
The above extreme-case analysis may remind a reader of the Michaelis-Menten formula (or the Langmuir form more generally), where one asymptotic condition that substrate is overabundant compared to enzymeπ‘π β«π‘πΈ is used to derive the formulaπΆβπ‘πΈπ‘ π‘π
π+πΎ
that spans two regimes. A common description of this formula is that it is responsive when substrate concentration is low π‘π βͺ πΎ, and becomes saturated when substrate concentration is highπ‘π β« πΎ. This can be viewed as theedgeconnecting vertices (0,1) and(1,1). This is also clear from the asymptotic conditions, as π‘π β« π‘πΈ has non-empty intersection with the conditions for these two vertices. We caution that although the Michaelis-Menten formula indeed forms an edge connecting the(0,1)and(1,1)vertices, it does not capture all points on the edge. In other words, the Michaelis-Menten assumption is a sufficient but not necessary condition for the edge. Although the conditionπ‘π β«π‘πΈ, πΎ for the(0,1)vertex is contained in the Michaelis-Menten condition π‘π β« π‘πΈ, the same does not hold for the(1,1)vertex. The condition(πΎ β«π‘πΈ, π‘πof the(1,1)vertex does not requireπ‘π β«π‘πΈ, e.g. bothπΎ β« π‘πΈ β«π‘π andπΎ β«π‘π β«π‘πΈ can achieve vertex(1,1). The necessary and sufficient condition for the(0,1)to(1,1)edge isπΈ βͺπΎ orπ‘πΈ βͺπ‘π. This is
more general thanπ‘πΈ βͺπ‘π. For example, whenπ‘π βͺπΎ, the(1,1)vertex is still achieved by π‘πΈ βͺπΎ, without assumingπ‘πΈ βͺπ‘π.
In summary, the polyhedral set for the full range of reaction orders highlights that the Michaelis-Menten formula is a sufficient edge-approximation of the overall behavior.
Through one asymptotic condition π‘π β« π‘πΈ, it captures a behavior spanning the edge connecting two regimes corresponding to vertices(0,1)and(1,1), and misses the regime corresponding to vertex(1,0)and expressionπΆ βπ‘π.
We can then investigate edge-approximation in general, with Michaelis-Menten approxi- mation (or single molecule states approximations and external-bath approximations, see Section3.1) as a special case. In the example ofπΆβs reaction order, while two asymptotic conditions yield vertices, one asymptotic condition yield edges. In terms of convex coeffi- cients in Eq (3.38), we can yield an edge by eliminating one vertex. For example, to obtain the Michaelis-Menten edge, we can eliminate the(1,0)vertex by letting coefficient π
1+π+π
goes to zero, which corresponds to asymptotic conditionπβͺ π orπβͺ1. For simplicity, we follow Michaelis-Menten to use just one of the two conditions to represent an edge approximation. This yields the following summary for edge approximations ofπΆ, and the graphical summarize of both edge and vertex approximations in Figure3.6.
πΆ β
β§
βͺβͺ
βͺβͺ
β¨
βͺβͺ
βͺβͺ
β©
π‘πΈπ‘ π‘π
π+πΎ, edge from(0,1)to(1,1), π‘π β«π‘πΈ;
min{π‘π, π‘πΈ}, edge from(0,1)to(1,0), πΎ βͺπ‘πΈ orπΎ βͺπ‘π; π‘ππ‘ π‘πΈ
πΈ+πΎ, edge from(1,0)to(1,1), π‘πΈ β«π‘π.
(3.44)
In addition to the edge that is symmetric reflection of Michaelis-Menten, we also obtain an edge approximation connecting (0,1) to (1,0). Importantly, this is another class of approximations just as valid as the Michaelis-Menten or single molecule states or external bath approximations. To derive the formula πΆ β min{π‘π, π‘πΈ}, we can apply the asymptotic conditionπΎ βͺ π‘πΈ or πΎ βͺ π‘π, which impliesπ‘πΈ +π‘π +πΎ β π‘πΈ +π‘π and
βοΈ(π‘πΈ +π‘π+πΎ)2β4π‘πΈπ‘π β |π‘πΈ βπ‘π|, to the exact solution ofπΆ in Eq (3.31):
πΆ(π‘πΈ, π‘π)β (π‘πΈ +π‘π)β |π‘πΈ βπ‘π|
2 = min{π‘πΈ, π‘π}.
The condition πΎ βͺ π‘πΈ or πΎ βͺ π‘π corresponds to tight binding between enzyme and substrate. One natural scenario where this is true is in strong sequestrations between molecules, e.g. between sigma and anti-sigma factors, and in nucleic acid circuits. This edge approximation also plays a central role for antithetic integral control motifs for robust perfect adaptation proposed in [22]. Importantly, this edge approximation is an alternative
Figure 3.6Visualization of how the reaction order polyhedron captures the holistic regulatory profile, and how the vertex and edge approximations forπΆin one binding reactionπΈ+π βπΆcompare to the exact solution. Upper left is the exact solution ofπΆin terms ofπ‘πΈandπ‘πin Eq (3.31). In the largeπ‘πlimit (close to π‘π axis in the 3D plot), the Michaelis Menten formula (upper middle) is a good approximation of the exact solution. In the smallπΎlimit (whenπ‘πΈand/orπ‘πare large), the minimum formula corresponding to the diagonal edge is a good approximation of the exact solution.
to the Michaelis-Menten (or single molecule states or external bath) edge approximations of enzymatic reactions. This edge approximation is valid whenever tight binding is present.
Robustness of vertices and edges. Another prominent feature of the sampling of reaction orders in Figure 3.5is that the randomly sampled points concentrate at the edges and vertices. Since the points are uniformly sampled in log scale on enzyme and substrate concentrations, if we consider perturbations to the system as multiplicative variations in these concentrations, then the vertices and edges should be robust to such variations.
We can see why the points concentrate at vertices and edges by looking at the convex coefficients in Eq (3.38) and Eq (3.41). Because the coefficients of the vertices all take rational-function form such as π
1+π+π , they approach extreme values of 0or 1when the variables(π, π ,1)are far apart in values. This means the condition for the reaction order to be in the interior of the polyhedra away from edges and vertices is quite fragile: the concentrations need to be βfinely adjustedβ so that they are close to each other. Once these concentrations drift apart from each other, we are at the vertices and edges.
Another way to describe this is that we can achieve precise values of reaction orders by very crude control of concentrations. For example, to push the reaction order ofπΆtoward