Where reduced biochemical rate laws come from

The experiments, mechanisms and mathematical arguments behind a changing quasi-steady state.

Lecture 6 · lecturer-written lineage · Companion to the core and exposition

A compact rate law combines several claims. It proposes which hidden states matter, asserts when those states can be eliminated and chooses the concentration or count that the remaining law describes. The history becomes clearer when those claims are traced separately.

Enzyme experiments supplied mechanisms and measurements. Singular perturbation supplied conditions for following a changing fast state. Gene regulation and single-molecule observations made occupancy, molecular storage and hidden switching part of the same modeling problem. These research programmes intersect. They did not develop as one uninterrupted chain.

1A mechanism had to explain a controlled measurement

Michaelis and Menten studied the kinetics of invertase, which cleaves sucrose into glucose and fructose. Their 1913 paper asked whether a proposed enzyme–sucrose complex could account for the measured reaction progress. Optical rotation supplied the observable. The complex itself was inferred through a kinetic model.

The experimental design separated effects that a single progress curve could mix. They controlled acidity and addressed the changing optical rotation of newly formed glucose. Initial-rate measurements limited the influence of accumulating products. The proposed complex concentration then connected substrate abundance to conversion rate. The translation's experimental discussion and pages 10–11 make this inferential sequence explicit.¹

The original analysis also stated the free-to-total approximation. On page 10, free sucrose is identified with total sucrose because only a negligible fraction is enzyme-bound. Thus the familiar explicit enzyme law already combined a binding hypothesis with an abundance assumption. Lecture 6 separates those steps so each can be checked.

Read an effective rate law as a connection between an observable, a proposed hidden state and stated approximations.

2Product inhibition was part of the original question

Michaelis and Menten tested product competition directly. They added glucose or fructose to otherwise comparable sucrose–enzyme mixtures and measured a reduced initial rate. In their model, a product-bound enzyme competed with the sucrose-bound complex. The translation develops the two binding balances on page 15 and gives measurements in Tables 5–7.¹

The measurements supported a competition hypothesis while leaving an estimation problem. The authors noted that curvature made initial slopes difficult to determine and that graphical extrapolation limited accuracy. Their rate model organized a quantitative experiment. It did not turn an indirect occupancy inference into a direct molecular observation.

The course's reversible enzyme is a separate illustrative mechanism. Free enzyme, substrate and product have concentrations $E,S,P$. The substrate-bound and product-bound complexes have concentrations $C_{ES}$ and $C_{EP}$. These complexes interconvert through catalysis. Historical invertase produced two different sugars. The shared idea is competition for a finite enzyme pool, not an assertion that the two reaction schemes are identical.

Connect product inhibition to its original experimental test while keeping a teaching mechanism distinct from the historical chemistry.

3Briggs and Haldane broadened the intermediate balance

Briggs and Haldane's 1925 note retained catalytic escape when balancing formation and removal of the enzyme complex. Let $k_+$, $k_-$ and $k_\mathrm{cat}$ denote association, dissociation and catalytic rate constants. Binding balance uses $K_d=k_-/k_+$. Their quasi-steady balance instead uses $K_M=(k_-+k_\mathrm{cat})/k_+$. Catalysis need not be negligible compared with unbinding for the second balance to be useful.²

The original complex was allowed to change after its initial formation. Briggs and Haldane wrote that their complex variable $p$ “diminishes from a value not exceeding e to zero.” Their $p$ meant enzyme complex, not product. The argument compared the scale of that change with the much larger substrate conversion.²

Lecture 6's opening puzzle therefore comes from an overly literal reading of the classroom shorthand $\dot C_{ES}\approx0$. The historical argument did not require a permanently fixed complex. Modern singular perturbation supplies the explicit clocks and trajectory conditions needed to explain its continuing change.

Distinguish rapid binding balance from quasi-steady intermediate balance, and avoid reading either as a claim that complex never changes.

4Tikhonov made convergence the object of the claim

Tikhonov studied differential systems with small parameters multiplying some derivatives. His 1952 paper defined a reduced system through an isolated root of the fast equations. The attached fast system holds the slow coordinates fixed. The first theorem connects the full and reduced solutions when the chosen root attracts the initial fast state and the solution remains in the required stability region.³

This changes what must be established. An algebraic root is a candidate fast state. The dynamics must approach that root before it can replace a differential variable. Multiple roots, an unstable branch or initial data outside its basin prevent the algebraic calculation from justifying the same reduction.

A lost differential equation also loses an independent initial condition. The initial fast layer carries the discrepancy between the original occupancy and the algebraic state selected by the initial totals. A slow model of product formation can therefore be valid after adjustment while omitting the complex-formation transient.

Ask which root attracts the intended initial state and on which time interval full trajectories approach the reduction.

5Fenichel separated approach to a constraint from motion along it

Fenichel's 1979 theory develops invariant manifolds and their stable and unstable families for singularly perturbed systems. On suitable compact normally hyperbolic pieces, the limiting set of stationary fast states has a nearby persistent slow description. The transverse directions describe rapid approach. The retained dynamics describes motion along the manifold.⁴

For the enzyme, binding at fixed totals restores a particular complex concentration. Catalysis changes the substrate total that selects that concentration. The limiting binding graph is called the critical manifold. The nearby invariant graph at finite separation is the slow manifold. It need not coincide with the full system's zero-complex-derivative curve.

The distinction has a calculable consequence. Differentiating the reconstructed complex along the changing total gives a nonzero slow derivative. More accurate invariant-graph calculations retain the small displacement that supplies that derivative. Recent enzyme analyses clarify the relationship among these graphs, small parameters and timescale estimates.⁷

Explain a changing quasi-steady complex through fast attraction and slow motion without identifying the critical graph with the finite-speed invariant graph.

6Scaling made the enzyme approximation testable

Segel and Slemrod's 1989 case study derives the fast complex-formation time and the slow substrate-consumption time before choosing dimensionless equations. Their estimates explicitly include concentrations. A bimolecular association constant becomes a relaxation rate only after the relevant molecular abundance is specified.⁵

The initial substrate-stationarity condition compares enzyme abundance with the available substrate scale. Write total enzyme concentration as $q_E$ and initial free substrate concentration as $S_\mathrm{ref}$. The condition is $q_E/(K_M+S_\mathrm{ref})\ll1$. It controls how much substrate can be captured during complex formation. The corresponding ratio of formation and consumption times is no larger than this quantity. The paper distinguishes the stationarity condition from a timescale comparison alone, especially on pp. 450–451.

The course's two enzyme families retain this distinction. Increasing binding speed at fixed affinity separates binding from catalysis. Reducing enzyme abundance at fixed microscopic constants slows substrate-pool change relative to complex adjustment. These are different parameter families, so their algebraic constants need not be the same.

Identify which concentrations and parameters make a proposed fast process fast relative to the observable being retained.

7Total substrate rescued information lost by a free-species coordinate

Borghans, de Boer and Segel asked how to extend quasi-steady reduction when enzyme captures an appreciable part of the substrate. Their 1996 paper replaces free substrate by total substrate, $q_S=S+C_{ES}$. Binding changes the free and bound pieces rapidly but cannot change their sum. The retained coordinate therefore follows the slower conversion process.⁶

The improvement is exact accounting of sequestered substrate. The equation $\dot q_S=-k_\mathrm{cat}C_{ES}$ holds before approximation in the closed vessel. A separate argument must still justify the fast-state closure. The total-QSSA root uses $K_M$, whereas rapid binding balance uses $K_d$.

Separate an unsuitable slow coordinate from a failure of time-scale separation, while preserving the approximation's remaining hypotheses.

8An open vessel requires a declared forcing experiment

Continuing supply changes the limiting problem. A closed enzyme consumes an initial stock. An open enzyme must also process what enters while the complex adjusts. Eilertsen and colleagues' open Michaelis–Menten analysis makes the scaling of that inflow part of the singular-perturbation question.⁸

The distinction is visible in the small-enzyme limit. Sending enzyme abundance to zero also sends processing capacity to zero. Holding supply fixed then differs from holding its fraction of capacity fixed. A closed-vessel approximation cannot settle which open experiment is being approached.

The choice of experiment also determines whether inhibition is visible in the measured output. At a finite processing steady state, total balance forces net conversion to equal input. Product competition can demand a much larger substrate pool while leaving that throughput unchanged. A batch test or a comparison at matched substrate totals reveals the activity loss more directly.

The batch-conversion example uses this distinction. Its trajectories have identical initial totals and differ only in product affinity. The companion calculation checks the regime $q_E\approx C_{EP}$ and compares forward activity with a product-free reference at the same remaining substrate total. The dominance test selects informative conditions without requiring reaction-order geometry in this lecture.

Choose boundary conditions and observables that expose the mechanism being tested, and include their scaling in the reduction claim.

9Hill's curve began with an oxygen-binding hypothesis

Hill's 1913 paper addresses the sigmoid oxygen-binding curves of haemoglobin. It recalls his earlier aggregation hypothesis and considers how strongly populated fully saturated forms could suppress partially occupied intermediates. Write free ligand concentration as $R$, half-saturation concentration as $K_H$ and response exponent as $n$. The expression $(R/K_H)^n/[1+(R/K_H)^n]$ is a modern generic notation for that class of saturating response, not the original paper's gene-regulation model.⁹

The original reasoning was mechanistic rather than simply an empirical curve fit. Hill proposed aggregation and acknowledged that fitted exponents need not be integers. He also examined sequential binding and the contribution of partially occupied forms. Pages 471–473 show both the hypothesis and the difficulty that intermediate states create.

The exponent therefore needs a mechanism and a defined input variable before it acquires a molecular interpretation. In the course's preformed-dimer repressor, exponent two follows from the dimer concentration being proportional to free monomer squared. A sequential two-site promoter instead retains a singly bound term unless a further cooperative limit suppresses it. These mechanisms can give similar reduced curves under different assumptions.

Treat a Hill exponent as a property of a specified response law, then check what mechanism and concentration variable can support that interpretation.

10Genetic repression supplied a different kind of hidden state

Jacob and Monod's 1961 synthesis distinguished structural genes, regulatory determinants and the site where a repressor acts. Mutants and partial diploids helped separate a diffusible regulatory effect from a linked operator property. Their Section 4 formulates the operator as the element whose association with a specific repressor interrupts information transfer.¹⁰

This was an argument from genetic behavior to regulatory organization. It should not be retold as direct observation of every molecular binding event. The paper explicitly discusses which properties of the operator and repressor follow from the experiments and which physical details remain to be identified.

The course's self-repressor implements one concrete version of this organization: its own protein binds the promoter, and only free promoter produces. It is a teaching model, not a reconstruction of the entire lactose system. Omitting RNA is another declared simplification. Jacob and Monod's paper itself included a short-lived messenger in the path from gene to protein.

Distinguish evidence for a regulatory architecture from the additional molecular and kinetic assumptions used in a self-repressor model.

11Promoter occupancy became a quantitatively testable construction

Garcia and Phillips used simple repression to connect a state-based model with independently accessible parameters. Their 2011 study varied repressor abundance and operator binding strength, measured regulatory responses, and compared inferred quantities with separate measurements. The promoter states and their weights are explicit in Figure 1 and the accompanying equations.¹¹

Their model includes more structure than the course's binary promoter: RNA polymerase, a repressor-binding state and a nonspecific DNA reservoir. The point of the comparison is that a response law can be constructed from states whose parameters can be tested. A fitted hyperbola alone does not establish that the state list or its physical interpretation is correct.

Lecture 6 constructs a production fraction from two promoter states. Let $D$ and $R$ denote free promoter and repressor concentrations, and $C_{DR}$ their bound complex. Fast binding gives $DR=K_dC_{DR}$, with promoter total $q_D=D+C_{DR}$. The fraction that produces is $D/q_D=1/(1+R/K_d)$. The experimental study shows what additional measurements can test the interpretation of such a function.

Ask which independently measured abundances or affinities would test the state-based interpretation of a regulatory curve.

12The plotted concentration can change the apparent regulatory law

The binding totals determine which abundance an experiment or model is varying. Let $C_{RR}$ denote free dimer concentration and $C_{DRR}$ promoter-bound dimer concentration. Total protein in monomer units is $q_R=R+2C_{RR}+2C_{DRR}$. A Hill-2 expression in free monomer does not automatically become a Hill-2 expression in this total. Protein stored in dimers can dominate the difference.

This is the same accounting issue that motivated total-variable enzyme reduction. The course's dimer example makes its consequence especially visible: when dimers dominate total protein, the promoter can look degree one in total protein while remaining degree two in free monomer. The exposition derives the storage conditions needed to transfer the exponent.

Identify the measured concentration before using the shape of a response curve to infer molecular cooperation.

13Downstream binding made molecular storage a systems problem

Del Vecchio, Ninfa and Sontag studied how attaching a downstream system can alter an upstream module. Their 2008 work develops retroactivity to describe this influence. A connection that binds an upstream regulator also removes molecules from its free pool, so its dynamical effect cannot be inferred from a one-way regulatory arrow alone.¹²

Fast binding can preserve substantial storage. In the simple model, changing free regulator also fills or empties bound sites. A storage factor therefore multiplies its derivative even in the limit of arbitrarily fast binding. Time-scale separation gives a reduced model of the load. It does not establish that the connection leaves the upstream response unchanged.

Check whether a downstream connection stores signal molecules and changes the transient after fast binding has been eliminated.

14Fast-state graphs united several effective rate laws

Gunawardena's 2012 linear framework treats biochemical state graphs with transition labels that depend on quantities outside the fast state list. Under its uncoupling condition, fixing those quantities makes the internal equations linear. Strong connectivity supplies a unique stationary occupancy up to the conserved amount, and the graph calculation produces rational expressions.¹³

This explains why enzyme and promoter models often yield rational rate laws. The algebraic calculation is exact for the specified stationary graph. A separate time-scale argument is needed to use its stationary state during the evolution of the larger system.

Total-coordinate closure can remain nonlinear even when this internal calculation is linear. For product competition, fixed free $S,P$ give a simple enzyme-state graph. Prescribed totals require free ligands to be reconstructed together with enzyme occupancy. The resulting scalar free-enzyme equation is generically cubic. An implicit DAE preserves that accounting without forcing a cumbersome explicit formula.

The framework also admits stationary graphs with internal currents. Stationary occupancy does not by itself establish thermodynamic equilibrium. This is why Lecture 7's thermodynamic conditions precede the broader binding-catalysis construction in Lecture 8.

Separate exact stationary-state algebra, dynamical justification and the additional total constraints that can make the closure implicit.

15A binary promoter model preceded direct RNA-episode measurements

Peccoud and Ycart's 1995 model combines random promoter activation, random production and molecular removal. Their paper begins from heterogeneous induction and earlier switching models. It specifies a continuous-time Markov process on gene state and protein count. Pages 223–224 give the active/inactive transitions, count-changing hazards and master equations.¹⁴

The course uses four rates to specify this telegraph model. An inactive promoter switches ON at rate $\alpha$, and an active promoter switches OFF at rate $\beta$. The ON state produces protein at rate $\rho$. Each protein is removed at rate $\gamma$. The stationary ON probability is $p_\mathrm{on}=\alpha/(\alpha+\beta)$.

The model treats a single gene as binary throughout. An active probability can be fractional, but an individual gene is either active or inactive at an instant. This is the distinction required when a fast promoter replaces the deterministic concentration constraint used in the enzyme examples.

The mathematical task also changes. Eliminating a fast gene state means averaging slow event hazards over its stationary distribution at the retained state. It does not mean replacing the physical gene by a fractional molecule. The independent-switching model in the core gives a case where this average can be derived explicitly.

Identify a promoter probability as a distribution over discrete states and derive a reduced hazard from that distribution.

16Single-molecule RNA records tested constant-rate production

Golding and colleagues followed RNA production in individual living bacteria in 2005. Their fluorescent reporter made intermittent production experimentally accessible. The observed distribution and first-event statistics challenged a constant-rate Poisson description. Figure 3 and the discussion on pp. 1031–1032 then examine active and inactive episodes and the numbers produced during an episode.¹⁵

The two-state description explains those episodes through random activation and deactivation, with constant-rate transcription while active. An exponentially distributed active duration combined with Poisson production yields a geometric episode size. The paper explicitly connects this interpretation to earlier switching models.

The evidence concerns RNA production in that reporter system. It does not directly image a unique two-state molecular mechanism, and it does not measure the illustrative protein parameters in core Figure 10. Those protein trajectories show what the specified switching mechanism can produce. The distinction between an observed history and a proposed hidden-state explanation remains essential.

State what a single-molecule experiment observed and distinguish that evidence from the hidden-state mechanism used to explain it.

17Eliminating a short episode can preserve a random packet

A fast-state limit is defined by the quantities held fixed. If both promoter switching rates grow while their ratio and the production rate remain fixed, repeated switching averages the production hazard. Slow protein arrivals approach a constant-rate Poisson process.

A different limit accelerates deactivation and active-state production together at fixed $\rho/\beta=b_\mathrm{pkt}$. Activation and protein removal rates remain fixed. Active periods become brief, but each leaves a finite random packet. The packet size is geometric because each production event competes with switching OFF. The reduced event is then a packet arrival.

Shahrezaei and Swain's 2008 expression analysis uses the related separation between RNA and protein lifetimes. An RNA produces proteins at rate $k_p$ and disappears at rate $\gamma_m$. It can leave several proteins before disappearing. Eliminating that lifetime can retain a random packet with mean $b_\mathrm{pkt}=k_p/\gamma_m$.¹⁶

Determine whether a proposed fast limit preserves an average hazard or the finite random output of a short-lived episode.

18Matching the mean did not settle the stochastic reduction

The telegraph model makes one limitation of mean-based validation explicit. Its stationary mean is $\rho p_\mathrm{on}/\gamma$ at every common switching-speed scale. Its Fano factor, the variance-to-mean ratio, depends on that speed. A model can therefore match the mean exactly while missing most of the fluctuations.

Kim and Sontag develop conditional fast-state calculations using exact moment structure in multiscale reaction networks. Their method emphasizes what distribution must be averaged to obtain the slow hazard.¹⁷

Holehouse and Grima examine fast-switching genetic feedback and show why heuristic replacement by a deterministic Hill propensity can fail. Binding also changes how protein is divided between free and bound forms. That accounting affects the retained count and its conditional distribution. Correctly posed stochastic averaging must retain these distinctions.¹⁸

The exposition works a one-promoter example in which exact conditional occupancy is one-half, while insertion of mean free protein into a deterministic occupancy curve gives one-third. This small counterexample separates conditional averaging from mean substitution without requiring a large simulation.

Test a stochastic reduction with a fluctuation statistic and verify which retained count its fast distribution conditions on.

19Event timing can distinguish mechanisms that share a histogram

Braichenko, Holehouse and Grima compare telegraph-like models with more detailed transcription mechanisms. Their 2021 analysis identifies regimes where transcript-number distributions can be similar while waiting-time distributions have different shapes. Event timing can therefore expose a mechanistic distinction hidden by a snapshot count distribution.¹⁹

This gives a practical meaning to model adequacy. An effective promoter can predict a chosen count distribution without uniquely identifying every activation or transcription step. A mechanism claim needs an observation that distinguishes plausible hidden descriptions.

The same principle applies to the enzyme. Equal final product yield does not certify the initial binding transient. Equal steady throughput under fixed supply does not certify equal enzyme availability. The observation determines which discarded feature matters.

Choose an observation that discriminates between hidden mechanisms when coarser predictions agree.

20The record supports three different kinds of claim

The completed method separates three operations. Exact molecular accounting identifies what binding preserves. A specified fast limit justifies reconstructing a moving occupancy or conditional distribution. An additional dominance approximation may replace free concentrations by totals or suppress a weakly populated state.

A whole-system steady state requires an additional slow flux balance. In the supplied enzyme, input, conversion and removal can all remain nonzero at constant concentrations. Fast attraction concerns redistribution at a fixed total. Slow stability concerns response to a changed total. Thermodynamic equilibrium imposes further physical conditions.

These distinctions define the next questions. Lecture 7 asks when reaction parameters and stationary binding states are thermodynamically compatible. Lecture 8 asks how compatible binding constraints determine a general DAE and its response. Later dominance and reaction-order tools organize the regimes glimpsed here. Each step has its own mathematical and empirical requirements.

State which part of a reduced model is exact, which part depends on a limiting regime and which further biological interpretation still needs evidence.

Primary sources and reading locations

The historical claims use the archived full texts linked below. Model figures are identified separately from the transcribed 1913 measurements. Page locations specify the passages used, rather than implying that each source establishes the whole lecture.

1. Michaelis L, Menten ML (1913). Die Kinetik der Invertinwirkung. Biochemische Zeitschrift 49:333–369. English translation supplied with Johnson KA, Goody RS (2011), The original Michaelis constant. Source. Archived full text. Translation pp. 2–3, 10–11 and 15–18; Table 5 on p. 16.
2. Briggs GE, Haldane JBS (1925). A note on the kinetics of enzyme action. Biochemical Journal 19:338–339. Source. Archived full text. Both printed pages.
3. Tikhonov AN (1952). Systems of differential equations containing small parameters in the derivatives. Matematicheskii Sbornik 31(73):575–586. Source. Archived full text. Definitions and Theorem 1, pp. 575–576.
4. Fenichel N (1979). Geometric singular perturbation theory for ordinary differential equations. Journal of Differential Equations 31:53–98. Source. Archived full text. Introduction and invariant-family hypotheses, especially pp. 73–74.
5. Segel LA, Slemrod M (1989). The quasi-steady-state assumption: a case study in perturbation. SIAM Review 31:446–477. Source. Archived full text. Fast and slow estimates and scaled equations, pp. 450–451.
6. Borghans JAM, de Boer RJ, Segel LA (1996). Extending the quasi-steady state approximation by changing variables. Bulletin of Mathematical Biology 58:43–63. Source. Archived full text. Published journal version, especially pp. 43–50.
7. Eilertsen J, Schnell S (2020). The quasi-steady-state approximations revisited: timescales, small parameters, singularities, and normal forms in enzyme kinetics. Mathematical Biosciences 325:108339. Source. Archived full text. Archived author preprint; slow manifolds, matching and timescale definitions.
8. Eilertsen J, Roussel MR, Schnell S, Walcher S (2021). On the quasi-steady-state approximation in an open Michaelis–Menten reaction mechanism. AIMS Mathematics 6:6781–6814. Source. Archived full text. Open-system parameter families and singular perturbation.
9. Hill AV (1913). The combinations of haemoglobin with oxygen and with carbon monoxide. I. Biochemical Journal 7:471–480. Source. Archived full text. Aggregation hypothesis, noninteger exponent and partially saturated forms, pp. 471–473.
10. Jacob F, Monod J (1961). Genetic regulatory mechanisms in the synthesis of proteins. Journal of Molecular Biology 3:318–356. Source. Archived full text. Section 4, especially pp. 340–342, and opening messenger account.
11. Garcia HG, Phillips R (2011). Quantitative dissection of the simple repression input–output function. PNAS 108:12173–12178. Source. Archived full text. Figure 1 and equations 2–5; independent parameter comparisons in Figures 3–4.
12. Del Vecchio D, Ninfa AJ, Sontag ED (2008). Modular cell biology: retroactivity and insulation. Molecular Systems Biology 4:161. Source. Archived full text. Definition of retroactivity and transcriptional-module loading.
13. Gunawardena J (2012). A linear framework for time-scale separation in nonlinear biochemical systems. PLoS ONE 7:e36321. Source. Archived full text. Strong connectivity, stationary weights and the Uncoupling Condition.
14. Peccoud J, Ycart B (1995). Markovian modelling of gene product synthesis. Theoretical Population Biology 48:222–234. Source. Archived full text. Model, transition hazards and Kolmogorov equations, pp. 223–225.
15. Golding I, Paulsson J, Zawilski SM, Cox EC (2005). Real-time kinetics of gene activity in individual bacteria. Cell 123:1025–1036. Source. Archived full text. Figure 3 and pp. 1031–1032.
16. Shahrezaei V, Swain PS (2008). Analytical distributions for stochastic gene expression. PNAS 105:17256–17261. Source. Archived full text. RNA–protein separation and the geometric protein-packet derivation.
17. Kim JK, Sontag ED (2017). Reduction of multiscale stochastic biochemical reaction networks using exact moment derivation. PLoS Computational Biology 13:e1005571. Source. Archived full text. Conditional fast-state moments and reduced propensity construction.
18. Holehouse J, Grima R (2019). Revisiting the reduction of stochastic models of genetic feedback loops with fast promoter switching. Biophysical Journal 117:1311–1330. Source. Archived full text. Heuristic reduction versus fast-switching feedback with binding accounting.
19. Braichenko S, Holehouse J, Grima R (2021). Distinguishing between models of mammalian gene expression: telegraph-like models versus mechanistic models. Journal of the Royal Society Interface 18:20210510. Source. Archived full text. Waiting-time comparisons and the scope of transcript-number agreement.