GAD65 decarboxylates L-glutamate to GABA + CO2 using the cofactor pyridoxal-5′-phosphate (PLP). In the accepted mechanism the substrate forms an external aldimine with PLP, and when the Cα–COO− bond breaks the electrons left behind are delocalized into the protonated pyridine ring — the quinonoid / carbanion intermediate. The pyridinium nitrogen is the electron sink that makes this otherwise high-energy carbanion accessible. By Dunathan’s stereoelectronic rule the scissile Cα–COO bond sits perpendicular to the ring plane for maximal σ→π overlap.
This is more than enzymology for the T1D field: GAD65 is the major type-1-diabetes autoantigen (GADA is a front-line predictive/diagnostic marker), and GAD65’s autoantigenicity is tied to its unusual catalytic behaviour — its tendency to lose PLP and populate a flexible apo form, and the dynamics of the catalytic loop region. Any computational claim about GAD65 conformation, the apo↔holo cycle, or epitope exposure ultimately rests on whether the method gets the underlying PLP electronics right.
The question this run answers: does a parameter-free electronic-structure method, given only the nuclei, spontaneously reproduce the electron-sink role of the PLP ring during decarboxylation — i.e. stabilize the departing-CO2 carbanion specifically because the ring is conjugated?
RealQM — a real-space, grid-based electronic-structure method (no basis sets, no fitted force field). The electron density is relaxed to its minimum-energy configuration in the field of fixed nuclei; the total energy E is read out directly.
The PLP active-site core, not the 585-residue enzyme:
The breaking Cα–CO2 distance R is driven outward in fixed steps; at each R all nuclei are clamped and only the electrons relax to convergence → E(R). This is run for two electronic environments:
Reported quantity: ΔE(R) = E(R) − E(R0) (reaction energy of CO2 departure; negative = downhill), and the sink term = ΔE(free) − ΔE(PLP) = the extra stabilization attributable to the ring.
Run parameters. 24 atoms (PLP variant) on a 160³ real-space grid, ~38 a.u. box; 7 distances R = 1.55–6.00 Å × 2 environments = 14 independent relaxations; per-point convergence by energy plateau. Runs in-browser on WebGPU (Chrome).
| R (Å) | ΔE free | ΔE PLP | sink = free − PLP |
|---|---|---|---|
| 1.55 | 0.000 | 0.000 | 0.00 |
| 2.00 | −0.012 | −1.916 | +1.90 |
| 2.60 | −2.476 | −3.776 | +1.30 |
| 3.30 | −2.560 | −5.426 | +2.87 |
| 4.20 | −3.101 | −6.605 | +3.50 |
| 5.00 | −3.556 | −6.912 | +3.36 |
| 6.00 | −3.454 | −7.620 | +4.17 |
(model energy units; sign/trend meaningful, magnitudes not calibrated)
The diagnostic signature is the divergence of the two curves:
Interpretation. Given nothing but the atomic coordinates — no reaction coordinate built in, no bias toward the “right” answer — the method spontaneously reproduces the textbook role of PLP: the conjugated ring, and only the conjugated ring, stabilizes the carbanion left when CO2 departs. The free-vs-PLP contrast isolates that effect cleanly because the substrate is identical between the two; the only difference is the presence of the sink.
The PLP electron-sink / quinonoid mechanism this run reproduces is not new physics — it is what mainstream quantum chemistry has established for PLP-dependent decarboxylases over the past two decades. Stating the comparison plainly matters for calibrating the claim.
The standard tools are (i) QM-cluster models — the active site (PLP + substrate + key first-shell residues, ~100–300 atoms) optimized with a DFT functional (commonly B3LYP / ωB97X-D with a dispersion correction) in implicit solvent — and (ii) hybrid QM/MM, with that DFT region embedded in a classical protein + water environment and sampled by MD / metadynamics. Applied to PLP decarboxylases (e.g. histidine decarboxylase; the group-II decarboxylase family, where crystallographic snapshots of the Dunathan and quinonoid intermediates anchor the calculations) and to decarboxylase benchmarks such as LigW (5-carboxyvanillate) and OMP decarboxylase, these methods deliver:
In short, DFT/QM-MM already does — quantitatively and with kinetics — what this RealQM scan does only qualitatively and thermodynamically. On the established catalytic step, mature QM/MM is the gold standard and this run does not compete with it.
| DFT / QM-MM (state of the art) | RealQM (this run) | |
|---|---|---|
| Energies | calibrated kcal/mol, benchmarked to experiment | uncalibrated model units; sign/trend only |
| Kinetics | transition states, barriers, rates | none (thermodynamic, single coordinate) |
| Environment | protein electrostatics + explicit/implicit solvent | gas-phase reduced core |
| Basis / functional | basis set + exchange-correlation functional (main systematic-error source; functional choice can change answers) | none — real-space grid, no basis set, no functional |
| Parameters | empirical dispersion / corrections common | parameter-free |
| Control | sink role inferred from the full energy landscape | identical-substrate free-vs-conjugated negative control by construction |
| Cost | substantial; HPC, careful setup | commodity GPU, in-browser, scannable |
So RealQM’s value here is not a better number on a solved problem. It is (1) a methodological cross-check — a formalism with no basis set and no exchange-correlation functional (DFT’s principal source of systematic error, and the reason different functionals can disagree) independently recovers the same electron-sink physics; (2) a clean internal control, since the free amino acid is the same molecule minus the ring, making the sink term a direct difference rather than an inference from one energy landscape; and (3) throughput — cheap enough to scan, which is what makes the planned holo-vs-apo and environment-embedding sweeps tractable.
The gap both leave open. The DFT/QM-MM literature characterizes the generic PLP-decarboxylase mechanism well, but GAD65 specifically — and the apo↔holo cofactor-loss state tied to its autoantigenicity — is largely unaddressed computationally. That gap, not the textbook catalytic step, is the target; calibrating RealQM against high-level DFT/QM-MM on the catalytic step (Aim 1 of the grant material) is how we earn the right to take it into the less-charted apo / epitope regime.
This establishes that a parameter-free method captures the central electronic event of GAD65 catalysis correctly and from first principles — the foundation needed before the more T1D-relevant questions can be addressed credibly:
gad65_autoantibody_dock.html):
currently uses placeholder sequences and a linear-peptide epitope, whereas real GADA epitopes are
conformational — so that line needs real CDR/epitope residues and is a much larger lift.