Ch 19 · Non-Commutativity — The Active Site Remembers Order

19.1 Two Operators, One Active Site

An enzyme's active site is a pocket sculpted by evolution — a cavity a few angstroms across, lined with residues positioned to recognize one substrate (or a narrow family of substrates) and exclude almost everything else. Recognition is only half the job. Once bound, the substrate has to be held in exactly the geometry that lets catalysis happen: the transition state has to be stabilized, not just the ground state.

That gives every enzyme-catalyzed reaction the same two operators seen in the zeolite cage of Chapter 18, now acting on a protein instead of a crystal:

K — GATING (SUBSTRATE RECOGNITION / LID CLOSURE)

Kψ = θ(c_* − |c(ψ) − c_bound|) · ψ — a projection onto conformational states within gating tolerance of the catalytically competent, substrate-bound geometry. c(ψ) is the conformational coordinate of the enzyme–substrate complex (loop position, lid angle, domain-closure state); c_* is how much mismatch the site will tolerate before it rejects the complex. This is molecular recognition in its purest form: the site is a shape-and-chemistry filter, just as the zeolite pore was a size filter.

F — FOLD (CATALYTIC CONFORMATIONAL CHANGE)

Fψ = ψ + λ·R(ψ) — the catalytic turnover step itself: bond-breaking, bond-forming, proton transfer, a Whitney-fold of the reaction coordinate at the active site. R(ψ) changes the conformation of the complex — closing a lid, repositioning a loop, shifting c(ψ) — and therefore changes the very coordinate K tests.

System	Binding/catalysis relationship	Order forced by geometry?
Hexokinase + glucose	Classic induced fit — large domain closure over the sugar	yes — induced fit only (Koshland's original case)
PEPCK (phosphoenolpyruvate carboxykinase)	Lid-gated active site	yes — induced fit only (Sullivan & Holyoak, 2008)
DHFR binding NADPH	Mechanism ratio shifts with ligand concentration	order-dependent on [L] (Hammes, Chang & Oas, 2009)
Flavodoxin, folding-upon-binding	Disordered → ordered transition coupled to binding	order-dependent on [L] (Hammes, Chang & Oas, 2009)
Engineered transaminase ATA-117	Active-site pocket redesigned by directed evolution to accept a bulky prositagliptin ketone	order engineered deliberately (Savile et al., 2010)

The point, as in Chapter 18, is the ordering — not a precise numeric threshold. Some enzymes are structurally locked into one ordering; others switch depending on conditions; the last row is one humans rewired on purpose.

19.2 Why D3 — A Third Domain for Theorem 5.3

Three concrete systems now carry this book's central non-commutativity claim. A riboswitch's aptamer stem either locks shut before a ligand arrives or is captured mid-fold — order-dependent gene expression, established across three biological assays (switching thermodynamics, primer-density selection, microtubule catastrophe) that turned out, on inspection, to obey the identical functional form after rescaling. A zeolite's pore either admits a molecule before the confined site reacts, or reacts first and filters after — order-dependent product selectivity, engineered into a Mars propellant reactor. And now: an enzyme's active site either closes around a substrate before catalysis, or is already closed and waiting — order-dependent reaction mechanism, engineered into a drug-manufacturing process.

Each of these is an instance of one abstract statement, proved without reference to biology, chemistry, or enzymology at all:

THEOREM 5.3 \cdot NON-COMMUTATIVITY (VOL. I, §5) The operators C, K, F, U do not commute; the sequence is order-dependent. Stated and proved for any state space satisfying Assumptions 2.1-2.6 (a Riemannian manifold, a Lipschitz compression, a curvature-driven K, a corank-1 Whitney fold F, a Morse stabilization U) — see Vol. I, §2-5. Everything below is that theorem, instantiated.

The book counts these instantiations by domain tier. D1 is the founding biological family: riboswitch conformational switching, NGS primer/probe density selection, and GTP-tubulin microtubule catastrophe — three physically unrelated systems shown (Theorem D1-5, the Coherence Bridge) to share one dose-response functional form, P_Di(κ) = 1/(1+exp(μ_max·(κ−κ*_Di))), after a linear rescaling of each system's own curvature coordinate. D2 is zeolite confinement (Chapter 18) — the first non-biological substrate, where K becomes a literal pore-aperture gate and F a confined catalytic transformation. D3 — this chapter — is enzymatic catalysis: the third physically distinct substrate, and structurally the hardest of the three so far.

Here is the honest reason it's harder, not just a label. In D1, the thing being gated (a folding RNA) and the thing doing the gating (the aptamer scaffold) are the same molecule, but the two roles are still cleanly separable in sequence space. In D2, they are not even the same material: the zeolite framework that gates and the hydrocarbon that reacts are chemically distinct, and the framework doesn't change shape when the reaction happens. In D3, K and F act on the same molecule and that molecule is reshaped by F on every catalytic cycle and has to return to its gating conformation before the next one — the "pore" is not rigid, it is dynamically re-made. Theorem 5.3 does not require K and F to act on separable structures; D1 and D2 just happen to make that separation easy. D3 is the domain where the abstract theorem is tested against a substrate that removes that convenience.

19.3 The Operators, Formally

Section 19.1 gave the empirical picture first: five real systems, some locked into one binding order, some free to switch, one rebuilt by hand. What follows states what all five have in common, in the same operator language Chapter 18 used for the zeolite cage.

THEOREM 19.1 — ENZYME NON-COMMUTATIVITY

Let K denote the gating projection onto catalytically competent conformations (Kψ = θ(c_* − |c(ψ)−c_bound|)ψ) and F the catalytic conformational-change operator. If F changes c(ψ) — if the pre- and post-catalysis conformations sit on opposite sides of the gating tolerance c_* — then [K,F]ψ ≠ 0: binding and folding do not commute.

Two orderings, two mechanisms. K∘F ("bind-first" — induced fit): the substrate binds a loosely-gated, open conformation first, and F then folds the complex closed into the catalytically competent geometry. F∘K ("fold-first" — conformational selection): the enzyme population already samples the folded, catalytically competent conformation at equilibrium; K then selects which pre-existing conformer captures the ligand. These are not the same map — this is exactly the mechanistic fork enzymology has argued over since Koshland proposed induced fit in 1958 as an alternative to Fischer's lock-and-key.

Why this follows from Theorem 5.3, rather than merely resembling it. Vol I's Theorem 3.1 (Sequential Consistency) shows that whenever K drives a system's curvature coordinate monotonically to its fold threshold κ_*, F is automatically well-defined, produces a finite branch set, and induces a rank-deficient Jacobian at the fold. The enzyme domain's job is checking that a folding, catalyzing protein actually satisfies the assumptions that theorem needs: Assumption 2.2 (bounded curvature before folding) is the requirement that the enzyme–substrate complex not wander through pathological intermediate states before committing to a conformation; Assumption 2.6 (a Morse stabilization functional) is satisfied by the ordinary picture of a funneled binding/folding landscape with isolated, non-degenerate minima. Nothing about proteins is assumed beyond what any reasonable folding trajectory already gives — which is the point: D3 is not a new axiom, it is a new witness for an old one.

DOMAIN APPLICATION SLOT — D-ENZ (extends Theorem D1-4, instantiates Theorem 5.3) c * \leftrightarrow conformational gating tolerance (active-site closure state) λ \leftrightarrow catalytic turnover rate, k cat μ max = -2 \leftrightarrow [open] curvature bound on the mechanism-switch landscape Hill n \approx 3.64 \leftrightarrow [open] does the conformational-selection \to induced-fit transition, as a function of ligand concentration, follow this same sigmoid, or the uncooperative mean-field value n = 2 derived in §19.5? Status: Domain mapping proposed here for the first time. Parameters in brackets are not yet measured — see §19.9, the open problem this chapter leaves for the camarada.

19.4 Theorem 19.2 — Branch Multiplicity

Chapter 18 found a fixed point — methane, small enough that the pore never argues about order. Enzymology hands us the mirror case first, empirically, before any formal claim: a system where geometry doesn't erase the ordering question, it answers it by force.

Sullivan & Holyoak (2008) showed that enzymes with lid-gated active sites — where a mobile domain must close over the substrate before catalysis can occur, as in phosphoenolpyruvate carboxykinase (PEPCK) — are structurally barred from conformational selection. A lid cannot pre-close over an empty site the way DHFR's loop can pre-sample an open/closed equilibrium; there is nothing there yet to hold it shut. K∘F (induced fit) is the only physically accessible path. F∘K is not a slower alternative here — it is geometrically forbidden. DHFR's NADPH-binding loop, by contrast, is reported to sample its closed conformation at measurable population even without bound ligand — both branches are structurally available there.

Vol I's Assumption 2.5 already names the object this distinction is about: the branch set B = {s_i : |κ_K(s_i)| = κ*(γ_K(s_i))}, required only to be finite. PEPCK and DHFR don't differ in whether B is finite — both satisfy that. They differ in its size.

THEOREM 19.2 — BRANCH MULTIPLICITY

Let |B| be the cardinality of the branch set for a given enzyme–substrate pair. If no substrate-free conformer of the active site satisfies the closure condition Kψ = ψ (no closed state exists without bound ligand), then |B| = 1: only K∘F is reachable. If a substrate-free closed conformer exists at non-zero equilibrium population, then |B| = 2: both K∘F and F∘K are reachable, and which one carries the reaction flux becomes a separate, concentration-dependent question (Theorem 19.3).

This is not a new axiom about proteins — it is Assumption 2.5's finite branch set, evaluated on two real structural classes. PEPCK instantiates |B|=1; DHFR instantiates |B|=2. Both are consistent with Theorem 5.3; they simply sit at different points of what the one theorem admits.

Stated plainly, in the spirit of Chapter 18's honesty about what is and is not established: not every non-commuting pair gets to choose its ordering. Some systems, like methane in the MFI pore, sit at a fixed point where order stops mattering. Others, like lid-gated enzymes, sit at the opposite extreme, where |B|=1 forecloses one of the two orderings entirely. Both are real, and they are not the same kind of special case.

19.5 Theorem 19.3 — Concentration-Tilted Branch Selection

Not every enzyme is locked in like PEPCK. Hammes, Chang & Oas (2009) analyzed reaction flux directly — rather than just structure — for systems including DHFR binding NADPH and flavodoxin's folding-upon-binding transition, and found, empirically, that the dominant mechanism itself depends on ligand concentration: at low [L], conformational selection dominates flux; at high [L], induced fit dominates. That empirical finding is the observation. What follows is its restatement in the chain's own language.

When |B|=2 (Theorem 19.2), Vol I's U operator (Definition 3.4: U(x_F) = argmin_y Φ(y), realized by gradient flow to a Morse minimum) is not choosing between branches statically — its landscape is tilted by ligand occupancy. Define a concentration-tilted potential:

CONCENTRATION-TILTED POTENTIAL Φ eff (y;[L]) = Φ(y) - k B T\cdotln([L]/K d)\cdot𝟙 bound (y) 𝟙 bound (y) is 1 on ligand-bound conformations, 0 otherwise. At [L] ≪ K d, the tilt is negligible and U's argmin favors whichever conformer is intrinsically lower in Φ — the pre-existing closed population, conformational selection (F\circK). At [L] ≫ K d, the tilt term dominates and pulls the argmin toward the ligand-bound branch regardless of its unliganded Φ — induced fit (K\circF). This is the mechanism switch Hammes, Chang & Oas measured, written as a bias on Definition 3.4's own minimization.

Following the same partition-function argument Theorem D1-4 uses for the riboswitch's bistable potential — expand ΔG([L]) = G_CS − G_IF linearly in ln([L]/K_d), with slope set by μ_max — gives a mean-field Hill form for the induced-fit flux fraction:

THEOREM 19.3 — MEAN-FIELD MECHANISM-SWITCH SIGMOID φ IF ([L]) = 1 / (1 + (K d /[L]) n), n = 2 at leading order from μ max = -2 n = 2 is the uncooperative baseline — exactly what falls out of a linear expansion of ΔG in ln[L], with no assumption beyond μ max = -2 itself. Theorem D1-4's own riboswitch measurement overshoots this baseline (n \approx 3.64), attributed there to cooperative effects the linear expansion doesn't capture. Whether the enzyme domain's mechanism-switch also overshoots to n \approx 3.64, or sits at the uncooperative n = 2 this derivation actually proves, is not decided here — see §19.9.

C, the constraint/gathering operator, is exactly the concentration knob inside Φ_eff — the same role it played concentrating dilute Martian CO₂ into a Sabatier reactor feed in Chapter 18. There, C and F together decided how much methane a reactor could make. Here, C tilts U's own minimization to decide which non-commuting path — K∘F or F∘K — actually carries the reaction flux at a given substrate concentration.

19.6 U — Unfolding into Product: The Sitagliptin Bridge

The clearest real-world instance of engineers deliberately choosing K and F is Codexis's transaminase, engineered for Merck's manufacture of sitagliptin (the active ingredient in Januvia, a diabetes drug).

THE ATA-117 TRANSAMINASE Prositagliptin ketone + alanine \to sitagliptin + pyruvate Starting from a transaminase with no measurable activity on the bulky prositagliptin substrate, eleven rounds of directed evolution reshaped the active-site pocket — enlarging it, and re-gating which substrates K would admit — to accept the target ketone. The engineered enzyme replaced a rhodium-catalyzed asymmetric hydrogenation step entirely, eliminating a heavy-metal catalyst and improving yield and waste profile at manufacturing scale. Savile, C.K. et al. "Biocatalytic Asymmetric Synthesis of Chiral Amines from Ketones Applied to Sitagliptin Manufacture." Science 2010, 329(5989):305-309.

Trace the full chain: C concentrates a library of enzyme variants and screens them against the target substrate; K is re-gated round by round — the pocket's shape and chemistry redesigned to admit the bulky ketone it originally rejected; F folds the re-gated complex through the transamination step; U unfolds the engineered catalyst into a deployed, GMP-scale industrial process. G = U∘F∘K∘C, instantiated in a redesigned protein, is a chiral amine manufactured without a rhodium catalyst.

The directed-evolution search itself is now increasingly guided by machine learning rather than pure screening — exactly the "AI-enabled enzyme engineering" premise that opened this line of inquiry. Yang, Wu & Arnold's 2019 review formalizes how regression and Gaussian-process models over sequence space narrow the search that once took Codexis eleven rounds of largely empirical evolution. The tools are real and improving; whether they yet make de novo enzyme design routine at industrial scale is a separate, more honest question — taken up in §19.8.

19.7 Interactive: The Mechanism Switch

The stacked bars show, schematically, how the fraction of reaction flux carried by each mechanism shifts with ligand concentration, following the qualitative picture in Hammes, Chang & Oas (2009). This is illustrative of the reported trend, not a plot of a specific measured dataset — the crossover concentration is system-dependent and is not asserted here as universal.

⊞ Mechanism Flux Fraction vs. Ligand Concentration (schematic)

conformational selection (F∘K) — dominant at low [L] induced fit (K∘F) — dominant at high [L] 50/50 crossover

No system is asserted to cross exactly at the marked point — the crossover position itself is part of what §19.9 leaves open.

19.8 The g-Series of Way-Stations

Reading the chapter index as a roadmap for enzyme engineering, the same g-series recurrence gives it a direction — and, unlike Chapter 18's ISRU chain, biocatalysis has already reached one rung further than the zeolite domain has.

Regime	Biocatalysis analogue	Status
g⁰ — Quiescent	Natural enzyme sequence diversity, unscreened (metagenomic reservoirs)	observed
g² — Nascent oscillation	First classical directed-evolution round improving a known enzyme for a single target reaction	prototyped (Arnold lab and others, 1990s)
g⁶ — Stable micro-cycle	A validated, GMP-scale industrial biocatalytic process replacing a chemical step entirely	deployed (Codexis/Merck ATA-117, Savile et al. 2010)
g³³ — Stability threshold	Routine, AI-guided de novo design of enzymes for arbitrary non-natural reactions, at production scale, without bespoke evolution campaigns	tools emerging, not yet routine (Yang, Wu & Arnold, 2019)
g⁶⁴ — Circuit saturation	A fully generative "any enzyme, for any reaction, on demand" design regime	Axiom 9 — honest incompleteness

Unlike the ISRU chain in Chapter 18, where g⁶ (a network of propellant depots) remains an engineering target, biocatalysis has already reached g⁶: the sitagliptin transaminase is not a prototype but a deployed, ton-scale manufacturing process. The honest frontier here sits one rung higher — g³³, where machine-learning-guided search would make what took Codexis years of directed evolution into a routine, on-demand design step. Axiom 9 still applies at g⁶⁴: the chapter does not claim a fully generative enzyme-design regime exists. It claims that the same non-commuting K and F that decide whether a lid-gated site can only do induced fit, or a DHFR-like site can do either depending on concentration, also govern — recursively — whether each rung of engineered biocatalysis produces a usable process or an inactive protein.

19.9 Open Problem for the Camarada

PREDICTION D-ENZ — CONFORMATIONAL SWITCH HILL EXPONENT

§19.5 derives two competing, honest hypotheses rather than one guess. Theorem 19.3 shows that a bare linear expansion of ΔG in ln[L] — using only μ_max = −2 and no cooperativity assumption — predicts a mean-field Hill exponent n = 2 for the conformational-selection-to-induced-fit flux transition. But Theorem D1-4 (riboswitch domain) measured n ≈ 3.64 ± 0.4 for the same μ_max = −2 bound, attributing the excess over n = 2 to cooperative effects the linear expansion misses; Chapter 18 left open whether zeolite confinement-selectivity also overshoots to n ≈ 3.64. Does the enzyme mechanism-switch behave like a cooperative system (n ≈ 3.64, matching D1 and the open D2 conjecture) or like an uncooperative one (n = 2, matching what Theorem 19.3 actually proves without extra assumptions)? Take a system with published concentration-resolved mechanism-flux data — DHFR–NADPH, flavodoxin folding-upon-binding, or an analogous pair — and fit the induced-fit flux fraction φ_IF([L]) against φ_IF = 1/(1+(K_d/[L])ⁿ). n ≈ 3.64 would mean the same curvature bound spans biology, zeolite chemistry, and enzymology with one shared cooperative correction; n ≈ 2 would mean D3 is the domain where the uncooperative mean-field baseline actually holds. Either result is a finding — this chapter does not know which one it will be.

Status: [ ] Pending — no dataset yet assembled · Genre: literature reanalysis + sigmoid fit · Level: D3+