betaFeedback

WorksComputer Science

Countering Anion Inhibition of Laccase Electron Transfer: Electrostatic Design, Charge-Transfer Dynamics on a Superconducting Quantum Processor, and Its Limits

Abstract

Hydroxide at neutral-to-alkaline pH and halides inhibit fungal laccases and limit their use in wastewater treatment. From GFN2-xTB cluster models of Trametes versicolor laccase (PDB 1GYC) we build a four-site Hubbard model of the His454-T3A-T3B-His452 junction of the trinuclear copper cluster and execute its charge-transfer dynamics on the Euro-Q-Exa quantum computer (IQM, 8 qubits, Trotterized circuits with Pauli twirling; 726 jobs). The charge-transfer frequency is measured to 3-5 meV, reproducibly over nine days and across changes of the device qubit map: hydroxide raises the charge-transfer energy of the crystal-geometry model by 92.0 +/- 2.2 meV (exact 87) and exceeds fluoride by 10.1 +/- 1.8 meV. A screen of charged probes in three dielectric models identifies Y116E; in the probe model the hardware resolves a reduction of the hydroxide penalty by 74 +/- 7 meV and of the halide penalty by 64 +/- 5 meV, acting on the site-energy gap rather than on the coupling. A real Glu116 (five conformers) with PROPKA protonation states and the coupled Asp456 gives a realistic reduction of 10-30 meV, with the same sign in 28 of 28 model combinations; the direction and ordering are the prediction, rate factors only an illustration. Scanning the on-site repulsion reverses the sign of the hydroxide effect, and a one-electron Hartree model reproduces the reversal: the measured dynamics is effectively single-particle. On a Fermi-Hubbard benchmark the hardware does not resolve electron correlation; it would need a contrast loss below 0.13 percent per CZ gate instead of the measured 0.34-0.54 percent.

Introduction

Laccases (EC 1.10.3.2) are multicopper oxidases that couple the one-electron oxidation of phenols, anilines and related compounds at a type-1 (T1) copper to the four-electron reduction of O2O_2 to water at a trinuclear copper cluster (TNC) formed by one type-2 (T2) and two type-3 (T3) coppers. [1, 2, 3] Electrons reach the TNC from T1 through a conserved Cys–His pathway of about 13 Å: the T1 ligand Cys453 is flanked in sequence by His452 and His454, which coordinate the two T3 coppers. [3] [4] Fungal laccases with a high T1 potential, such as the enzyme from the white-rot fungus Trametes versicolor, are attractive green oxidants for pulp bleaching, lignin processing and the removal of dyes, phenolic pollutants, pharmaceuticals and other micropollutants from municipal and industrial wastewater. 5–7

Two inhibitions limit such applications. First, the activity of fungal laccases on phenolic sub-ends_mid=1 strates follows a bell-shaped pH profile whose alkaline side is attributed to hydroxide binding at the TNC, which slows internal electron transfer from T1.8 Second, halides inhibit the enzymes, fluoride most strongly, then chloride and bromide, an order that has been linked to access of the anion to the T2/T3 site. 9,10 Kinetic work on T. versicolor laccase found that chloride and hydroxide act through a common mechanism and that chloride inhibition is strongest at low pH (competitive Ki=0.35K_i = 0.35 mM at pH 3 and 23.7 mM at pH 6).11 Both effects are relevant for water treatment, where the enzymes have to operate near neutral pH and in the presence of salts.6,10 Mutations near the copper sites shift the pH optimum of fungal and bacterial laccases and their halide sensitivity,12–14 and QM/MM studies have explored how mutations near T1 change its redox potential,15 which makes the electrostatic environment of the TNC a target for rational design.

Electron transfer (ET) in proteins is described by Marcus theory, in which the rate depends on the electronic coupling between donor and acceptor, the driving force and the reorganization energy.16 The coupling is mediated by the protein through specific pathways,17,18 and the distance dependence of biological ET rates is well established.19,20 The same quantities, site energies and couplings, define a tight-binding (Hubbard) model of the junction between the His bridges and the T3 coppers.21 The real-time dynamics of such a model can be simulated on a digital quantum computer,22–24 and metalloenzymes are prominent targets of quantum computation.25 Recent hardware studies have treated iron–sulfur clusters with a quantum processor coupled to a supercomputer,26 the conformers of a miniprotein27 and small molecules on IQM hardware.28 At the same time, the evidence for a generic exponential quantum advantage in quantum chemistry is weak,29 and present processors operate before fault tolerance with error mitigation.30,31 It is therefore important to state precisely what a present-day quantum processor contributes to a biochemical question.

In this work we use the Euro-Q-Exa quantum computer to address anion inhibition of laccase ET in three steps. (i) We execute the charge-transfer (CT) dynamics of an enzyme-derived four-site model of the laccase TNC junction on the hardware and show that the CT energy, and its change by hydroxide and halides, are measured at meV precision, reproducibly over days and across reconfigurations of the device. (ii) We screen electrostatic modifications of the second coordination sphere and identify a mutation, Y116E, that removes most of the CT penalty imposed by hydroxide and halides; we test it on the hardware, decompose its mechanism, and follow it from a generic charge probe to a real glutamate with explicit conformers and protonation states. (iii) We measure where the hardware stops resolving electron correlation, using the junction model itself and a Fermi–Hubbard benchmark. We state explicitly that the four-site dynamics is classically exact and, as we show, effectively single-particle; the hardware results validate the execution of the model and quantify the limits of the device, while the design conclusions rest on the classical model and its stated approximations.

Methods

Cluster models

Cluster models were cut from the 1.90-Å crystal structure of T. versicolor laccase with a full complement of coppers (PDB 1GYC).4 The pathway cluster contains the T1 copper with its His395 and His458 imidazoles and Cys453, the His452–Cys453–His454 tripeptide with its backbone, the three TNC coppers with their His imidazoles, the bridging hydroxide of the T3 pair and the T2-bound ligand (H2_2O, OH−^-, F−^- or Cl−^-). The protein-embedded models add the charged second-sphere residues closest to the T2 water, Asp77 (5.2 Å), Asp424 (5.5 Å) and Arg243 (7.6 Å), as capped side chains; with the probes and Asp456 the clusters contain 124–141 atoms. The cluster is a quartet (Cu(I) T1 and three Cu(II)) with the T3 pair antiparallel (broken-symmetry). Hydrogen positions were relaxed with GFN2-xTB32 while heavy atoms were kept at the crystal positions.

Tight-binding parameters

Site energies and couplings were obtained from the self-consistent GFN2-xTB Fock matrix FF and the overlap SS in the minimal valence basis. After Löwdin orthogonalization, F′=S−1/2FS−1/2F' = S^{-1/2}FS^{-1/2}, the block of F′F' belonging to each pathway fragment was diagonalized and one frontier orbital cic_i selected per site: for Cu sites the fragment orbital with more than 30% Cu 3d weight closest to the Fermi level, for the bridges (Cys453, His452, His454) the highest occupied fragment orbital. Site energies and couplings are εi=ci†F′ci\varepsilon_i = c_i^{\dagger}F'c_i and tij=ci†F′cjt_{ij} = c_i^{\dagger}F'c_j. Because each site orbital comes from a separate diagonalization, the sign of each hopping is arbitrary; populations, currents and frequencies are gauge invariant on a chain, so the observables used here do not depend on it. Dielectric screening by the protein and solvent was modeled with the analytical linearized Poisson–Boltzmann model ALPB34 for a protein-like environment (ε≈4\varepsilon\approx4, parametrized as diethyl ether) and for water, and compared with the gas phase. Because the effective screening at this site is not known, we bracket it by the gas phase (no screening) and water (full screening) and use ε≈4\varepsilon\approx4 as a protein-like middle value. The size, and in some cases the sign, of second-sphere effects depends on screening, so we accepted a design conclusion only if it had the same sign in all three dielectric models.

Mutants, probes and protonation

Second-sphere mutants were modeled explicitly (D77N, D424N, R243A, D77N/D424N; Q70K, Q102K and Q70K/Q102K, with CH3NH3+\mathrm{CH_3NH_3^+} at the Gln CD/OE1 position). To screen positions rather than specific side chains, generic charge probes (CH3COO−\mathrm{CH_3COO^-} or CH3NH3+\mathrm{CH_3NH_3^+} with the charged group at the tip of the native side chain) were placed at Tyr116, Leu459 and Phe450, the residues closest to His454 and T3A that a charged side chain could replace. Real Glu116 was then built on the crystal backbone (N, Cα_\alpha, Cβ_\beta kept) on a 10∘/10∘/15∘10^\circ/10^\circ/15^\circ grid of χ1/χ2/χ3\chi_1/\chi_2/\chi_3; conformers with heavy-atom contacts below 3.0 Å (2.55 Å for a carboxylate O to an N/O partner) were rejected, crystal waters being assumed displaced. The 317 clash-free conformers all lie in the region of the Tyr116 side chain (χ1=30∘\chi_1 = 30^\circ–150∘150^\circ; Tyr116 has 93∘93^\circ); five conformer families, with the carboxylate 5.3 Å to 7.4 Å from T3A, were represented by their least strained member, and two of them were also modeled as protonated Glu. pKaK_{\mathrm{a}} values were estimated with PROPKA 335 on the full protein with Glu116 inserted, and Asp456, which couples to Glu116, was added to the cluster in both protonation states. Anion affinities relative to the wild type (WT) are ΔΔE=[E(mut,X)−E(mut,H2O)]−[E(WT,X)−E(WT,H2O)]\Delta\Delta E = [E(\mathrm{mut}, X) - E(\mathrm{mut}, \mathrm{H_2O})] - [E(\mathrm{WT}, X) - E(\mathrm{WT}, \mathrm{H_2O})] from GFN2-xTB single points in the three dielectric models; ΔΔE>0\Delta\Delta E > 0 means weaker binding of XX in the mutant. These are differences of electronic energies at fixed geometries, without relaxation, entropy or explicit water; we use only their sign, not their magnitude.

Junction model

The junction His454–T3A–T3B–His452 is described by a four-site Hubbard Hamiltonian21

H=∑i,σεiniσ+∑⟨ij⟩,σtij(ciσ†cjσ+h.c.)+∑iUini↑ni↓,(1)H = \sum_{i,\sigma} \varepsilon_i n_{i\sigma} + \sum_{\langle ij\rangle,\sigma} t_{ij}(c_{i\sigma}^{\dagger}c_{j\sigma} + \mathrm{h.c.}) + \sum_i U_i n_{i\uparrow}n_{i\downarrow}, \tag*{(1)}

with nearest-neighbour hopping from the fragment couplings and on-site repulsion U=4U = 4 eV on Cu and 2 eV on the His bridges. The Cu value lies in the range of effective one-band repulsions obtained for Cu 3d holes from first principles and spectroscopy (about 3.7–5 eV; the bare 3d repulsion of three-band models is larger, 7–10 eV);36 the bridge value is a chosen smaller value for a delocalized π\pi orbital. Both are model choices; the UU scan below shows that the sign of the anion effect holds for any common scaling of UU above 0.55 of these values. The initial state has both His bridges doubly occupied and one hole per spin on the T3 coppers (the antiferromagnetic T3 pair: β\beta hole on T3A, α\alpha hole on T3B; Figure 1). Starting the pair as singlet, triplet (MS=0M_S = 0) or broken-symmetry product gives identical charge dynamics to below 10−310^{-3} e within 3 fs, because the direct T3A–T3B coupling (4 meV to 8 meV) is too weak to matter on this time scale; extending the model by T2 and T1 adds no femtosecond dynamics either (couplings to the core ≤0.03\le0.03 eV). The Jordan–Wigner transformation37 maps the α\alpha and β\beta chains onto the two rows of a qubit ladder (8 qubits); the rungs carry the on-site interaction. For the classical ET analysis the same parameters were used in a larger pathway model (T1, Cys453, His454, His452, T3A, T3B) to obtain the T1→\toT3 driving force, an effective donor–acceptor coupling by Löwdin partitioning and Marcus rates for reorganization energies λ=0.5\lambda= 0.5–1.11.1 eV.

Figure 1. Model. (a) Active site of T.T. versicolor laccase (PDB 1GYC), crystal coordinates projected on their principal plane: Cu ions, Cys453 S, His452/His454 ring centroids, the T2 ligand and the Tyr116/Asp456 positions; arrows: the T1 →\to Cys–His →\to T3 pathway. (b) Four-site junction (WT, ε≈4\varepsilon\approx4) with GFN2-xTB site energies (relative to their mean), couplings tt and on-site repulsions UU; arrows: occupied spin orbitals, open circles: holes. (c) Mapping of the α\alpha and β\beta chains onto the qubits of ladder C of EQE1; rungs carry the on-site interaction; the measured β\beta His454–T3A bond is highlighted.

Circuits

Time evolution uses second-order Strang splitting38 of HH into even bonds, odd bonds and on-site terms, with a step of 1 fs. Each hopping term is a native XX+YY rotation and each on-site term a ZZ rotation with single-qubit phases, both compiled to two CZ gates on neighbouring qubits of a ladder; the deepest circuits (6 fs) contain 128 CZ gates. Circuits were built with Qiskit39 and verified against exact dynamics computed with ffsim. Ladders were selected from calibration data (readout and relaxation errors per qubit) on the square lattice of the device; most runs used two ladders, α\alpha 38–41/β\beta 44–47 (“B”) and α\alpha 9–12/β\beta 16–19 (“C”), in the numbering of the 52-qubit map. When a qubit was lost or restored and the numbering changed, circuits were re-indexed so that they ran on the same physical qubits.

Observables and error mitigation

Three observables were used. (1) Site populations in the computational basis. (2) Spin-resolved bond currents and bond coherences, obtained with a number-conserving XX+YY(π/2\pi/2, β\beta) rotation on a bond before measurement: β=0\beta= 0 gives ⟨ni−nj⟩=2Im⁡⟨ci†cj⟩\langle n_i-n_j\rangle= 2\operatorname{Im}\langle c_i^\dagger c_j\rangle (current) and β=π/2\beta= \pi/2 gives −2Re⁡⟨ci†cj⟩-2\operatorname{Re}\langle c_i^\dagger c_j\rangle (coherence). (3) The CT frequency: the complex bond signal ρ(t)=⟨cHis454†cT3A⟩β\rho(t) = \langle c_{\mathrm{His454}}^\dagger c_{\mathrm{T3A}}\rangle_\beta is reconstructed from the two bases at 24 times between 0.25 and 6 fs and fitted jointly to Im⁡ρ=asin⁡(ωt)e−γt\operatorname{Im}\rho= a\sin(\omega t)e^{-\gamma t}, Re⁡ρ=b(1−cos⁡ωt)e−γt+c\operatorname{Re}\rho= b(1-\cos\omega t)e^{-\gamma t}+c. ℏω\hbar\omega is the CT excitation energy of the junction model, a model quantity without a direct spectroscopic counterpart whose changes, not its absolute value, are interpreted; γ\gamma absorbs the hardware damping. Hardware noise damps the amplitude but hardly shifts the frequency, which is why the frequency became the main observable. The same fit applied to the exact signal and to the noiseless circuits on the same time grid gives the reference values; the difference between the two references is the Trotter error. The penalty of an anion XX is ℏω(X)−ℏω(H2O)\hbar\omega(X)-\hbar\omega(\mathrm{H_2O}); variants compared in a result were always measured in the same jobs.

Every CZ gate was Pauli-twirled40 (12 random instances per circuit), which turns coherent errors into stochastic ones common to all variants. Results were post-selected on the conserved particle numbers (Nα,Nβ)(N_\alpha,N_\beta); the fraction of shots that survives is a model-free indicator of circuit quality. In early runs linear zero-noise extrapolation (ZNE) by gate folding41 was also tested. Errors are bootstrap estimates over twirl instances,42 which include shot noise and the spread between instances; results from several runs were combined with inverse-variance weights and errors scaled by χ2/dof\sqrt{\chi^2/\mathrm{dof}} where larger than one.

Population decomposition

The frequency gives only Ω=Δ2+4t2\Omega= \sqrt{\Delta^2+4t^2}, where Δ\Delta is the effective gap between the His and Cu sites and tt the effective coupling. The population of the T3A hole oscillates with amplitude A=4t2/Ω2A = 4t^2/\Omega^2, so amplitude and frequency together give tt and Δ\Delta separately, the two Marcus ingredients (rate ∝t2\propto t^2, driving force from Δ\Delta). These runs used the same circuits measured in the Z basis, with t=0t = 0 circuits to correct preparation and readout errors.

Hardware

Euro-Q-Exa is a EuroHPC quantum computer hosted by the Leibniz Supercomputing Centre and integrated with the SuperMUC-NG supercomputer.43 System 1 (backend EQE1) has 54 physical superconducting qubits from IQM on a square lattice with tunable couplers44 and was accessed through the Munich Quantum Software Stack.45 Depending on the calibration, 51–53 qubits were available during the project. Between 19 October and 12 November 2026 we ran 63 experiments with 726 jobs, 68 300 circuits and about 1.03×1081.03 \times10^{8} shots (Table S1).

daterunjobsjob idscircuitsshotsα\alpha qubitsresult
2026-10-19calibration_eqe12423521–429254612000–device calibration
2026-10-19depth_scan_eqe11423520618000–SI: depth scan
2026-10-19ghz142348300–device test
2026-10-19quality_eqe12423521–42925400–device calibration
2026-10-19t1_lucj_r1_eqe1142350500–SI: T1 LUCJ
2026-10-19t1_lucj_r1_eqe1_hole142351000–SI: T1 LUCJ
2026-10-21junction_eqe11425097104000032–40–46–50populations (3.2)
2026-10-21junction_long_eqe13425102–42510424060000032–40–46–50populations (3.2)
2026-10-21junction_twirl_eqe114250999667200–populations (3.2)
2026-10-21junction_zne_eqe114250981664000–populations (3.2)
2026-10-21layout_ab_eqe1142503728000–SI: layout
2026-10-21t1_quality_eqe11425096432000–device calibration
2026-10-22current_eqe16425441–42544657686400032–40–46–50currents (3.2)
2026-10-23anion_eqe118426341–4263721728259200031–39–45–49fingerprint (3.4)
2026-10-23freq2_eqe114425961–4259741344201600038–39–40–41CT energy (3.3)
2026-10-23freq3_eqe114425976–4259961344201600031–39–45–49CT energy (3.3)
2026-10-23freq6_eqe112426300–4263221152172800031–39–45–49CT energy (3.3)
2026-10-23freq_eqe114425947–4259601344201600031–39–45–49CT energy (3.3)
2026-10-23ligand4_eqe124426445–4264922304345600038–39–40–41fingerprint (3.4)
2026-10-23multiplex_eqe112425927–4259458641296000–three ladders (3.2)
2026-10-28bridge2_eqe13428602–42860824048000038–39–40–41–42bridge model (3.12)
2026-10-28bridge_eqe13428577–42858224048000025–32–40–46–50bridge model (3.12)
2026-10-28ligand4a_eqe124428533–4285652304345600032–40–46–50fingerprint (3.4)
2026-10-29anionC_eqe118429373–429438172825920009–10–11–12fingerprint (3.4)
2026-10-29freqC_eqe112429256–429364115217280009–10–11–12CT energy (3.3)
2026-10-29ohfB_eqe112429441–4294601152172800038–39–40–41fingerprint (3.4)
2026-10-30clfC_eqe112429471–429940115217280009–10–11–12fingerprint (3.4)
2026-10-30mainB_eqe112429963–4299961152172800038–39–40–41CT energy (3.3)
2026-10-30mfreq_eqe18430029–4300527681152000–parallel copies (3.12)
2026-10-30mfreq_eqe1L08430029–43005276811520009–10–11–12parallel copies (3.12)
2026-10-30mfreq_eqe1L18430029–430052768115200039–40–41–42parallel copies (3.12)
2026-10-30mfreq_eqe1L28430029–430052768115200026–27–28–29parallel copies (3.12)
2026-11-02envB_eqe112436538–4365611152172800038–39–40–41reference, stability (3.3)
2026-11-03env2B_eqe112437275–4372921152172800038–39–40–41reference, stability (3.3)
2026-11-03env2C_eqe112437365–437386115217280009–10–11–12reference, stability (3.3)
2026-11-03envC_eqe112437135–437154115217280009–10–11–12reference, stability (3.3)
2026-11-03kk2B_eqe112437294–4373231152172800038–39–40–41Q70K/Q102K (3.5)
2026-11-03kk2C_eqe112437387–437406115217280009–10–11–12Q70K/Q102K (3.5)
2026-11-03kkB_eqe112437156–4371761152172800038–39–40–41Q70K/Q102K (3.5)
2026-11-03kkC_eqe112437183–437270115217280009–10–11–12Q70K/Q102K (3.5)
2026-11-08halB_eqe124438428–4399852304345600038–39–40–41Q70K/Q102K halides (3.5)
2026-11-08halC_eqe124439986–440009230434560009–10–11–12Q70K/Q102K halides (3.5)
2026-11-09hal2B_eqe124440567–4406162304345600038–39–40–41Q70K/Q102K halides (3.5)
2026-11-09hal2C_eqe124440618–440664230434560009–10–11–12Q70K/Q102K halides (3.5)
2026-11-09popB_eqe112441144–4411711200180000038–39–40–41populations (3.8)
2026-11-09popC_eqe112441174–441197120018000009–10–11–12populations (3.8)
2026-11-09singB_eqe124440945–4409932304345600038–39–40–41single mutants (3.5)
daterunjobsjob idscircuitsshotsα\alpha qubitsresult
2026-11-09singC_eqe124440996–441046230434560009–10–11–12single mutants (3.5)
2026-11-09vibB_eqe112441198–44122012001800000–vibronic model (3.12)
2026-11-09vibC_eqe112441222–44124212001800000–vibronic model (3.12)
2026-11-10y116eB_eqe124442087–4421172304345600039–40–41–42Y116E (3.7)
2026-11-10y116eC_eqe124442119–442166230434560009–10–11–12Y116E (3.7)
2026-11-10y116ehalB_eqe124442238–4422612304345600039–40–41–42Y116E halides (3.7)
2026-11-10y116ehalC_eqe124442262–442308230434560009–10–11–12Y116E halides (3.7)
2026-11-10y116epopB_eqe112442214–4422251200180000039–40–41–42Y116E mechanism (3.8)
2026-11-10y116epopC_eqe112442226–442237120018000009–10–11–12Y116E mechanism (3.8)
2026-11-11u00C_eqe112442886–442898115217280009–10–11–12U scan (3.10)
2026-11-11u04C_eqe112442899–442911115217280009–10–11–12U scan (3.10)
2026-11-11u06C_eqe112442912–442928115217280009–10–11–12U scan (3.10)
2026-11-11u10C_eqe112442930–442943115217280009–10–11–12U scan (3.10)
2026-11-12hubN4_eqe14443595–44359833650400047–41–33–26correlation benchmark (3.11)
2026-11-12hubN4b_eqe14443599–44360233650400026–19–12–6correlation benchmark (3.11)
2026-11-12hubN6_eqe14443604–44360733650400047–41–33–26–19–12correlation benchmark (3.11)

Table S1. All Euro-Q-Exa (EQE1) runs of this work, generated from the result files.

Correlation benchmark

To test the device on genuinely correlated dynamics, we used a one-dimensional Fermi–Hubbard chain with t=0.3t = 0.3 eV (the scale of the enzyme couplings), U=0U = 0 or 4t4t, and N=4N = 4 or 6 sites (8 or 12 qubits) on nested windows of the longest straight ladder of the device, starting from a charge domain wall (left half doubly occupied). The observable is the charge imbalance (Nleft−Nright)/N(N_{\mathrm{left}} - N_{\mathrm{right}})/N after post-selection, compared with exact dynamics and with time-dependent mean field (Hartree).46 A second N=4N = 4 window on other qubits separates the effect of size from that of qubit quality.

Results

Hydroxide slows the T1→T3 step in the classical model

In the pathway model, OH−^{-} on T2 lowers the T1→T3 driving force from −0.365-0.365 to −0.300-0.300 eV and slows the Marcus rate by a factor of 1.9–2.9 for reorganization energies between 0.5 and 1.1 eV (2.9×10102.9 \times10^{10} vs 1.1×10101.1 \times10^{10} s−1^{-1} at λ=0.8\lambda= 0.8 eV), consistent with the activity drop of fungal laccases at neutral and alkaline pH.8 F−^{-} and Cl−^{-} give the same rate as OH−^{-} (1.1×10101.1 \times10^{10} s−1^{-1}): at the level of the ET step, the anion’s charge, not its chemistry, sets the effect. On the femtosecond scale the junction shows the same trend: at 2 fs the T3 copper has gained 0.444 e from His454 with H2_2O and 0.325 e with OH−^{-}. The absolute rates are faster than experiment, as expected for semiempirical couplings; the relative effect of the ligand is the result.

Junction dynamics on the hardware

The first hardware runs measured site populations of the junction (Figure S6). With H2_2O the hardware followed the exact curve within 0.03 e up to 2 fs (68 CZ, 31–36% post-selection survival), but the raw data reproduced the pH effect (H2_2O − OH−^{-}) poorly (χ2=147\chi^2 = 147 for 4 points), because coherent errors differed between the two variants. Pauli twirling made the errors stochastic and common to both, and with twirling and ZNE the pH effect was reproduced at all times (χ2=1.7\chi^2 = 1.7, p=0.79p = 0.79), including its rise and decay up to 3.5 fs (Figure S7). The predicted sign change at 4 fs (128 CZ) was not reproduced: at that depth the spin density on His454 no longer

Figure S6. Site populations of the four-site junction on EQE1 and the pH effect on T3A.

Two-panel plots: (a) pH effect on T3 Cu charge (H₂O − OH⁻), with vertical axis “Δ charge on T3 Cu (e)” and horizontal axis “time (fs)”; legend “exact”, “EQE1 run 3 (68 CZ)”, “EQE1 run 4 (68–128 CZ)”. (b) spin density on His454 (nα − nβ), with vertical axis “spin density (e)” and horizontal axis “time (fs)”; legend “T2–OH₂, exact”, “T2–OH⁻, exact”, “T2–OH₂, EQE1”, “T2–OH⁻, EQE1”.

Figure S7. Pauli-twirled circuits, linear zero-noise extrapolation (run 3: noise scale 1, 3; run 4: 1, 2; jobs 425099, 425102–425104). Error bars: bootstrap over twirl instances (shot noise + instance spread). b: the β electron that flows from His454 to the α-occupied T3 Cu leaves α spin density on the bridge. Figure S7: pH effect on the T3 charge and the spin density on His454 with zero-noise extrapolation. 41

Bond currents proved more robust. The β\beta current His454→T3A agreed with the exact model over 0.5–3 fs (χ2=13.3\chi^2 = 13.3 for 12 points), with a peak at 1 fs of (0.34±0.03)(0.34 \pm0.03) e/fs for H2_2O (exact 0.38) and (0.29±0.02)(0.29 \pm0.02) e/fs for OH−^{-} (exact 0.31); the population change obtained by integrating the measured current satisfied the continuity equation (χ2=13.8/12\chi^2 = 13.8/12). The currents expected to vanish (α\alpha His454–T3A and β\beta T3B–His452) averaged 0.003 e/fs, against about 0.3 e/fs in the allowed channels (Figure S8): only the β\beta electron can enter T3A through His454 and only the α\alpha electron T3B through His452. This selection rule follows from the occupations of the initial state and, as discussed below, is also the reason why the measured dynamics is effectively single-particle. Running three copies of the junction simultaneously on disjoint ladders (24 qubits) confirmed that the selection rule is independent of the layout, but showed that post-selection survival and damping differ strongly between ladders (41/27/17%), and that linear ZNE no longer restores the signal below about 30% survival (Figure S9).

Three-panel plots: (a) β current His454 → T3 Cu, with vertical axis “current into Cu (e / fs)” and horizontal axis “time (fs)”; legend “T2–OH₂, exact”, “T2–OH⁻, exact”, “T2–OH₂, EQE1”, “T2–OH⁻, EQE1”, “α (spin-forbidden), EQE1”. (b) α current His452 → T3 Cu′, with vertical axis “current into Cu (e / fs)” and horizontal axis “time (fs)”; legend “T2–OH₂, exact”, “T2–OH⁻, exact”, “T2–OH₂, EQE1”, “T2–OH⁻, EQE1”, “β (spin-forbidden), EQE1”. (c) continuity: Δn on T3 Cu, with vertical axis “Δn (e)” and horizontal axis “time (fs)”; legend “T2–OH₂, exact”, “T2–OH⁻, exact”, “T2–OH₂, ∫ current”, “T2–OH₂, occupation”, “T2–OH⁻, ∫ current”, “T2–OH⁻, occupation”.

Figure S8. EQE1 jobs 425441–425446: number-conserving XX+YY readout rotation on each His–Cu bond, Pauli twirling, linear ZNE (noise scale 1, 2), post-selection (Nα, Nβ) = (3, 3). c: Δn from the time integral of the measured current (trapezoid) and, independently, from the occupation basis (jobs 425099, 425102–425104). Figure S8: Spin-resolved bond currents and the continuity check. The α His454–T3A current vanishes because the α electrons on His454 and T3A are both present and cannot move (the same reason makes the measured dynamics effectively single-particle, main text).

Three panels showing current into Cu versus time for three simultaneous copies of the junction, with exact and ladder results, and the pH effect

Figure S9. Jobs 425927–425945: three copies of the 8-qubit junction run simultaneously on disjoint ladders (24 qubits), twirling, linear ZNE (scale 1, 2). Post-selection survival at scale 1: ladder 1 41 %, ladder 2 27 %, ladder 3 17 %. c: inverse-variance mean over the three ladders. Figure S9: Three copies of the junction executed simultaneously on disjoint ladders (24 qubits).

Charge-transfer energy and its hydroxide shift

The CT frequency was the most precise observable (Figure 2). On the bare pathway model, the hardware frequencies agreed with the noiseless circuits to 0.2–0.3%. Over six runs on three ladders, hydroxide raised the CT energy of the His454→\toT3A channel by (74.8 ±\pm 2.2) meV and that of the His452→\toT3B channel by (138 ±\pm 6) meV (noiseless circuits 76 and 142, exact 80 and 144 meV). The two channels therefore respond differently to the same ligand, with an asymmetry of (58 ±\pm 7) meV (circuits 66 meV). The remaining 3% difference between the circuits and the exact dynamics is the Trotter error of the 1-fs step, not hardware error.

Figure 2. Charge-transfer signal of the His454–T3A bond on EQE1 (bare pathway model, no protein field; ladder A, 0.25–6 fs, raw twirled data). Points: hardware imaginary and real parts of ρ(t)=⟨cHis454†cT3A⟩β\rho(t) = \langle c_{\mathrm{His454}}^\dagger c_{\mathrm{T3A}}\rangle_{\beta} with bootstrap errors; solid lines: exact dynamics; dotted lines: damped-oscillation fits. The fitted CT energies ℏω\hbar\omega for H2_2O and OH−^- at T2 and their difference (the OH−^- penalty) are compared with the exact model and the noiseless circuits.

With the protein environment (Asp77, Asp424, Arg243, ε≈4\varepsilon\approx4) the model penalty of the His454 channel is 87 meV (circuit 83). This configuration served as the reference of all later comparisons and was measured in seven independent submissions on two ladders between 2 and 11 November 2026, across three changes of the qubit map (Figure 3). The CT energy with H2_2O was (1197.2 ±\pm 1.4) meV (scatter 2.7 meV; circuit 1200, exact 1237 meV) and the OH−^- penalty (92.0 ±\pm 2.2) meV (scatter 2.8 meV, χ2\chi^2/dof = 0.27). The scatter is below the statistical errors: no drift is detectable over nine days. The damping γ\gamma varied between 0.13 and 0.24 fs−1^{-1} without affecting the frequency. Across all runs the per-ladder systematic floor of a single frequency difference is about 3 meV to 5 meV, and 5 meV to 10 meV for differences that involve different jobs.

Figure 3. Stability of the reference measurement (protein-embedded WT model, ε≈4\varepsilon\approx4). Seven independent submissions on ladders B and C between 2 and 11 November 2026, spanning three changes of the device qubit map. (a) CT energy with H2_2O; dashed: noiseless circuit, solid: exact model. (b) OH−^- penalty; grey band: weighted mean ±\pm error. Error bars: bootstrap over twirl instances.

Anion fingerprint

Every anion raised the CT energy (Figure 4). Combined over six runs on three ladders, OH−^- exceeded F−^- by (10.1 ±\pm 1.8) meV (5.5σ\sigma; noiseless circuit 11 meV) and Cl−^- by (8.4 ±\pm 2.3) meV, whereas Cl−^- and F−^- were indistinguishable ((1.9 ±\pm 1.6) meV; circuit 6 meV). The charge of the T2 ligand thus sets about 90% of the CT penalty and its chemical identity the remaining 10 meV. Because F−^- and Cl−^- act alike on the CT step, the strong difference between their inhibition constants 9,11 must arise from binding and access to the TNC rather than from the ET step itself.

Figure 4. Anion fingerprint (bare pathway model). (a) CT energy change relative to H2_2O for OH−^-, Cl−^- and F−^- on one ladder, with noiseless-circuit values (horizontal bars). (b) Differences between anions from all runs and ladders (points; run names as in Table S1) with their combined value; OH−^- is separated from both halides, Cl−^- and F−^- are not separated.

Second-sphere mutants: positive charges

Removing the negative charges near T2 (D77N, D424N, D77N/D424N) lowered the OH−^- penalty at ε≈4\varepsilon\approx4 but raised it in water, so no mutant was proposed from these. Two lysines replacing Gln70 and Gln102 (tips 7.8 and 8.1 Å from the T2 water) lowered the penalty in all three dielectric models (exact 87→\to71 meV at ε≈4\varepsilon\approx4). On the hardware, four runs per system on two ladders gave (92.0 ±\pm 2.9) meV for WT and (70.8 ±\pm 2.9) meV for Q70K/Q102K, a reduction of (21.2 ±\pm 4.1) meV (5.1σ\sigma; exact 16, circuit 13 meV); the halide penalty dropped by (13.9 ±\pm 4.0) meV (exact 14 meV). The single mutants each lowered the penalty (Q70K (24.0 ±\pm 5.6) meV, Q102K (17.3 ±\pm 4.9) meV; exact 9 and 7 meV), and the ordering of their CT energies with H2_2O (Q102K < Q70K < double) was reproduced, but their additivity was not resolved above the systematic floor. However, the change in anion affinity of the double mutant depends on the dielectric model (ΔΔE\Delta\Delta E for OH−^- = −-62, −-15 and +1.4 kcal/mol in gas, ε≈4\varepsilon\approx4 and water): in a protein-like environment the lysines would bind the inhibiting anions more strongly. Lowering the CT penalty is therefore not sufficient; a useful mutation must also not strengthen anion binding.

A negative charge near His454: the probe screen

We therefore required two conditions in all three dielectric models: a lower OH−^- penalty on the His454 channel and weaker OH−^- binding (Figure 5). Positive probes (Y116K, L459K, F450K) increased anion binding in every model. Negative probes at Leu459 and Phe450 weakened OH−^- binding (+11 to +12 kcal/mol at ε≈4\varepsilon\approx4) but hardly changed the penalty (78 and 82 vs 87 meV). Only the probe at Tyr116 satisfied both conditions: penalty 53/20/−7-7 meV (gas/ε≈4\varepsilon\approx4/water; WT 180/87/16 meV) and OH−^{-} binding weaker by +42/+11/+2.5 kcal/mol. Tyr116 lies next to His454 and T3A (its hydroxyl 5.9 Å from T3A), on the side of the junction opposite to the T2 ligand. The halide penalties of the Y116E probe model were also close to zero at ε≈4\varepsilon\approx4 (−6-6 and −5-5 meV for Cl−^{-} and F−^{-}, WT 56 and 53 meV).

Figure 5. Classical screen of second-sphere modifications (GFN2-xTB clusters; gas, ε≈4\varepsilon\approx4, water). (a) OH⁻ penalty on the His454 channel (exact four-site dynamics); lower is better. (b) OH⁻ affinity relative to WT, ΔΔE\Delta\Delta E (electronic energies, sign only); higher means weaker binding. Rows: WT; positive charges (Q70K/Q102K, K probes); negative probes; real Glu116 conformers a–e; protonated Glu116; WT and Glu116 with Asp456 in both protonation states. Missing points: SCF not converged or affinity not computed.

Y116E on the hardware

With WT and Y116E in the same jobs on two ladders, the OH−^{-} penalty fell from (93 ± 4) meV to (19 ± 9) meV, a reduction of (74 ± 7) meV (exact 66, circuit 60 meV); the two ladders agreed within 1.2σ\sigma (Figure 6a). The CT energy with H2_2O itself dropped by 94 meV to 101 meV (exact 116, circuit 106 meV), showing that the carboxylate stabilizes the whole His454→\toT3 manifold. For the halides, the difference ℏω(X)WT−ℏω(X)Y116E\hbar\omega(X)_{\mathrm{WT}}-\hbar\omega(X)_{\mathrm{Y116E}} was (163.1 ± 3.0) meV for Cl−^{-} and F−^{-} together, and (98.7 ± 4.2) meV for H2_2O, so Y116E lowered the halide penalty by (64 ± 5) meV (Cl−^{-} 65 ± 6, F−^{-} 64 ± 7; exact 62 and 58, circuit 57 and 53 meV; Figure 6b). Cl−^{-} and F−^{-} were again indistinguishable, both in WT ((−2-2 ± 4) meV) and in the mutant ((−2-2 ± 7) meV). Within the probe model at crystal geometry, Y116E removes essentially the whole CT penalty of all three anions; the relaxation tests below show that this overstates the effect.

Figure 6. Y116E (probe model, ε≈4\varepsilon\approx4) on EQE1, WT and mutant in the same jobs. (a) OH−^{-} penalty on ladders C (filled) and B (open); horizontal bars: exact (black) and noiseless circuit (orange). (b) Reduction of the Cl−^{-} and F−^{-} penalty by Y116E on both ladders, relative to the H2_2O difference of panel a; black bars: exact. (c) Change of the gap Δ\Delta and of the coupling tt caused by OH−^{-}, from Z-basis populations combined over both ladders: Y116E removes most of the gap increase and leaves the coupling change unchanged.

Mechanism: the gap, not the coupling

The population runs separate the gap Δ\Delta and the coupling tt (Figure 6c). In WT, OH−^{-} raised Δ\Delta by (77 ± 11) meV and lowered tt by (38 ± 6) meV; in Y116E, OH−^{-} raised Δ\Delta by only (27 ± 12) meV and lowered tt by (36 ± 5) meV. The mutant thus removed (51 ± 16) meV of the OH−^{-}-induced gap increase (3.2σ\sigma; circuit 57 meV) and left the coupling change untouched (−3-3 ± 8 meV). Both ladders gave the same decomposition. Hydroxide inhibits the CT step mainly by raising the site energy that the hole has to overcome, and the carboxylate counteracts exactly this quantity; the mechanism is electrostatic and does not involve the electronic coupling through the His bridge. The OH−^{-}-induced decrease of tt is about twice the circuit value on both ladders, which may reflect a systematic loss of amplitude on the hardware (amplitudes about 0.6–0.7 of the circuit values) and is not interpreted further.

From the probe to a real glutamate

The probe places a full negative charge where the Tyr116 hydroxyl is; a real Glu116 is shorter. For the five conformer families of Glu116−^{-}, the OH−^{-} penalty at ε≈4\varepsilon\approx4 was 54–68 meV, a reduction of 19 meV to 32 meV instead of 66 meV, while OH−^{-} binding remained weaker by +11.2 to +12.0 kcal/mol in every family (Figure 5; Table S2). The conformers differ by only ±7 meV, so the result does not hinge on a particular side-chain geometry. PROPKA places the pKaK_{\mathrm{a}} of Glu116 between 5.0 and 8.9 depending on the conformer and shows that Glu116 and Asp456 share roughly one proton: in the three families closest to T3A, Glu116 is charged above pH ≈ 6.5 and raises the pKaK_{\mathrm{a}} of Asp456 from 7.2 to 8.3–9.2. A protonated Glu116 would leave only a 9 meV reduction and no change in binding. With Asp456 included in the cluster and the protonation states that PROPKA assigns at pH 8–9 to the three families closest to T3A (WT: Asp456−^{-}; Y116E: Glu116−^{-} with Asp456 protonated), Y116E lowers the OH−^{-} penalty by 17 meV to 20 meV at ε≈4\varepsilon\approx4, 29 meV to 33 meV in the gas phase and 26 meV to 28 meV in water; including the two remaining families in the same protonation state widens these ranges to 7 meV to 20 meV, 26 meV to 33 meV and 17 meV to 28 meV, and with both acids charged the reduction is 17 meV to 25 meV at ε≈4\varepsilon\approx4. The sign of the effect is the same for every conformer, both Asp456 states and all three dielectric models (28 of 28 combinations; the gas-phase SCF of conformer e with Asp456 did not converge). Interestingly, the native Asp456 already lowers the WT penalty by 12 meV when charged. The realistic prediction for the mutant is therefore a reduction of the CT penalty by 10 meV to 30 meV, about one third of the unrelaxed probe value. Partial relaxation tests this further (GFN2-xTB/ALPB at ε≈4\varepsilon\approx4; T2 ligand, the residue-116 group and all hydrogens mobile, the remaining heavy atoms fixed). Hydroxide stays bound to T2 (Cu–O 2.35 → 1.93 Å) but moves towards T3A (5.9 → 4.0 Å), which more than doubles the absolute penalty (WT with Asp456−: 74 → 201 meV). The reduction by the real mutant, however, is stable: 20 meV unrelaxed and 24 meV relaxed (Glu116 conformer c with protonated Asp456). The probe’s larger unrelaxed reduction (66 meV) shrinks to the same 24 meV after relaxation. Absolute penalties are therefore geometry dependent, whereas the mutant effect, the quantity on which the design rests, is not. The real-glutamate models were not run on the hardware: the dynamics is classically exact, and the hardware had reproduced the model differences of the probe to within 3 meV to 5 meV, so a repetition would add no information about the mutant; the design conclusion rests on the classical model.

abcde
Glu116−^{-} (no Asp456)5856546268
Glu116−^{-} + Asp456−^{-}5250495458
Glu116−^{-} + Asp456-H5856546267
pKapK_a Glu116 (PROPKA)5.66.45.08.58.9
pKapK_a Asp456 (PROPKA; WT 7.2)8.58.39.26.86.7

Table S2. OH−^{-} penalty on the His454 channel (meV) for the real Glu116 conformers with and without Asp456 (GFN2-xTB, ε≈4\varepsilon\approx4).

Role of the on-site repulsion

To ask whether the anion effect depends on the on-site repulsion, all UU were scaled by λU\lambda_U and the WT penalty was measured at four values on the same ladder, with both ligands in the same jobs at each value (Figure 7). The sign of the OH− effect reverses: −97 ± 5 meV at λU=0\lambda_U = 0 (21σ\sigma below zero), −44±4 at 0.4, +16±4 at 0.6 and +90 ± 5 meV at the physical value, with the crossover at λU=0.55±0.01\lambda_U = 0.55 \pm0.01 (exact 0.54). Without repulsion, hydroxide would accelerate the His454→T3A transfer; at the repulsion appropriate for Cu(II) it slows it, and the physical UU lies well inside the inhibiting regime (the sign would reverse only for a Cu repulsion below about 2.2 eV). A one-electron model, in which the β\beta electron moves in the static Hartree field UiniαU_{ini\alpha} of the frozen α\alpha electrons, reproduces the exact curve point for point. The reason is the occupation pattern: both α\alpha electrons of His454 and T3A are present and cannot move, and the T3 halves are coupled by only a few meV, so within the 6-fs window the measured dynamics is that of a single electron in a static field. The reversal therefore reflects the charging energy of the Cu site relative to the bridge, which a purely electrostatic picture without the site’s own repulsion would miss, but it is not an effect of electron correlation. The same argument applies to all junction results above: the hardware executed a single-particle problem, which a classical computer solves exactly.

Figure 7. OH−^{-} penalty versus a common scaling λU\lambda_U of all on-site repulsions (WT, protein-embedded model, ladder C). Points: EQE1 (both ligands in the same jobs at each λU\lambda_U); solid line: exact four-site Hubbard dynamics; dashed line: one β\beta electron in the static Hartree field of the frozen α\alpha electrons. The two curves coincide, so the sign change is a mean-field (charging-energy) effect, not a correlation effect.

Hardware limits for correlated dynamics

The correlation benchmark was designed so that hardware noise and correlation pull in opposite directions: noise drives the charge imbalance towards zero, as mean field does, whereas the exact correlated dynamics keeps the doubly occupied domain in place. At U=4tU = 4t and 3 fs, the exact imbalance exceeds mean field by 0.21 for N=4N = 4 (128 CZ) and 0.13 for N=6N = 6 (204 CZ); at U=0U = 0 the two coincide. The hardware reproduced free-fermion dynamics with an rms deviation of 0.04–0.08, but at U=4tU = 4t fell below even the mean-field curve (0.37 ± 0.01 and 0.36 ± 0.01 against 0.48 and 0.66 at 3 fs; Figure 8). The correlation gap only opens after about 1.5 fs, when the hardware deviation already exceeds it. The ratio of the hardware to the noiseless signal decays with the number of CZ gates as e−kNCZe^{-kN_{\mathrm{CZ}}} with k=0.34k = 0.34–0.54%0.54\% per CZ; state preparation and readout alone cost 2.1–3.2% per qubit; and two windows of the same size on different qubits differ about as much as N=4N = 4 and N=6N = 6. Resolving the correlation gap at 3 fs would require k≤0.13%k \leq0.13\% for N=4N = 4 and k≤0.04%k \leq0.04\% for N=6N = 6, three to eight times better than measured. The same limit appeared in earlier attempts: a five-site model including the bridging hydroxide of the T3 pair, whose two coppers compete for the bridge electrons (10 qubits, 80–228 CZ), gave only a 2.4σ\sigma hint of the predicted charge redistribution at 4–12% survival; a model coupling the transferred charge to an effective vibrational mode (λ=0.43\lambda= 0.43 eV) could not be resolved because gate noise heated the mode qubits more than the vibronic excitation itself; and linear ZNE did not restore signals below about 30% survival. Running three copies of a circuit in parallel increased the damping on two of three ladders, so for the frequency observable one copy per job on the best ladder was the most efficient use of the device.

Figure 8. Correlation benchmark: one-dimensional Fermi–Hubbard chain (t=0.3t = 0.3 eV) starting from a charge domain wall, on nested windows of the longest ladder of EQE1 (51-qubit map). First three panels: charge imbalance for U=0U = 0 (open, green) and U=4tU = 4t (filled, blue) with exact (solid) and time-dependent mean-field (dashed) curves, for N=4N = 4 on two different qubit windows and for N=6N = 6. Last panel: ratio of hardware to noiseless-circuit imbalance at U=4tU = 4t versus CZ count, with exponential decays for the measured (solid) and the required (dotted) error per CZ.

Discussion

How anions inhibit the charge-transfer step

Within the model, any anion at T2 raises the CT energy of the His→\toT3 junction and lowers the T1→\toT3 driving force, and does so mainly through the site energy (gap) of the T3 copper rather than through the electronic coupling. About 90% of the penalty is set by the charge of the ligand; hydroxide adds about 10 meV over the halides, and fluoride and chloride act alike. This assigns the different inhibition strengths of F−^{-} and Cl−^{-} observed experimentally 9–11 to binding and access rather than to the ET step, and it is consistent with a common mechanism for hydroxide and chloride inhibition. 11 The two His bridges respond differently (His452 channel about twice as strongly), so the T3 pair is not perturbed symmetrically by the T2 ligand.

Design rule and prediction for experiment

The screen yields a simple design rule: a negative charge on the far side of His454/T3A from the T2 ligand stabilizes the hole on the T3 copper, lowers the gap and offsets the anion’s field, and at the same time repels the anion from T2. Positive charges closer to T2 lower the CT penalty as well, but attract the inhibiting anion, and mutations that remove native acids near T2 are not robust against the dielectric model and may also impair proton delivery to the TNC, in which conserved acidic residues near T2 take part. [3] [13] Y116E is the only candidate that passed both criteria in every model, and the effect survives the step from a charge probe to a real glutamate with explicit conformers, protonation states and the coupled Asp456, at a reduced size of 10 meV to 30 meV, and survives partial relaxation (24 meV).

What would such a reduction mean for the rate? If a gap reduction δ\delta translated one-to-one into the T1→\toT3 driving force, the Marcus expression would change the rate by exp⁡[δ(ΔG+λ)/(2λkBT)]\exp[\delta(\Delta G+\lambda)/(2\lambda k_{\mathrm{B}}T)]; with ΔG=−0.30\Delta G=-0.30 eV and λ=0.5\lambda=0.5–1.1 eV, δ=10\delta=10 meV to 30 meV corresponds to a factor of about 1.1–1.5, a partial recovery of the 1.9–2.9-fold slow-down that hydroxide causes in the same model. The weaker anion binding comes on top of this and may matter more in practice, because it lowers the fraction of enzyme with an anion bound. The prediction is testable with standard assays: the alkaline side of the pH profile with ABTS or a phenolic substrate, and the chloride inhibition constant at pH 6, where it is 23.7 mM for the wild-type enzyme. 8,11 The main risk is the proton-delivery network of the TNC, which a CT model does not contain: a carboxylate next to the cluster could change O2_2 reduction independently of the ET step. A double check of activity at acidic pH, where the wild type is fully active, would separate the two effects.

What the quantum processor contributed

The hardware executed an enzyme-derived CT model with meV precision, reproducibly over nine days, on several ladders and through changes of the device configuration, and it resolved every model difference tested, from the 75-meV hydroxide shift to the 10-meV difference between hydroxide and fluoride and the 50-meV gap effect of Y116E. Three practical lessons emerged. Frequencies are far more robust than amplitudes or populations: noise damps an oscillation but hardly shifts it. Pauli twirling is essential whenever two variants are compared, because it makes coherent errors common to both. The post-selection survival is a free and reliable indicator of the quality of a ladder on a given day.

The hardware did not compute anything that a classical computer could not. The four-site junction is classically exact, and in this model its dynamics is effectively single-particle. On a correlated benchmark of comparable size, the present error rates are three to eight times too high, and the requirement grows with system size because the correlation signal shrinks while the circuit depth grows. Models that would need the quantum computer, such as the T3 pair with its bridging hydroxide, coupling to vibrations, or the full TNC with several mobile holes, are exactly those that could not be run. This agrees with the general assessment that quantum advantage in chemistry requires substantially better hardware. 29 The 150-qubit second system of Euro-Q-Exa, with longer coherence times,43 is the natural next platform for these models, and the frequency protocol with paired variants developed here can be transferred to them directly.

quantityEQE1exactcircuit
OH−^{-} penalty, bare model (6 runs)74.8 ±\pm 2.28076
OH−^{-} penalty, His452→\toT3B138 ±\pm 6144142
OH−^{-} penalty, protein field (7 runs)92.0 ±\pm 2.28783
OH−^{-} −- F−^{-}10.1 ±\pm 1.81111
Cl−^{-} −- F−^{-}1.9 ±\pm 1.666
reduction by Q70K/Q102K21.2 ±\pm 4.11613
reduction by Y116E (probe), OH−^{-}74 ±\pm 76660
reduction by Y116E (probe), halides64 ±\pm 56055
OH−^{-} effect on Δ\Delta: WT / Y116E77 ±\pm 11 / 27 ±\pm 12–116 / 59
penalty at λU=0\lambda_U = 0−97±5-97 \pm5−109-109−101-101
crossover λU\lambda_U0.55 ±\pm 0.010.54–

Table 1. Main hardware results (EQE1, raw twirled data) with exact and noiseless-circuit references; His454→\toT3A channel unless noted. Penalties in meV. Errors are statistical (bootstrap over twirl instances, scaled by χ2/dof\sqrt{\chi^2/\mathrm{dof}} where larger than one). The systematic floor estimated from run-to-run scatter is 3 meV to 5 meV for differences measured within the same jobs (all rows except Q70K/Q102K) and 5 meV to 10 meV for differences between jobs (Q70K/Q102K row).

Relevance for applications

For wastewater treatment, laccases are typically used immobilized on supports and have to operate near neutral pH and in the presence of salts.6,7 A mutation that lowers the CT penalty of hydroxide and halides and weakens their binding addresses both limitations at once. Its benefit, however, depends on the balance between the ET step and the other steps of the catalytic cycle, which only experiments can establish.

Limitations

(1) Cluster models of 124–141 atoms with implicit solvation; the size of second-sphere effects depends on the dielectric model, which we address by requiring the same sign in three models. (2) All tight-binding parameters come from GFN2-xTB; a DFT check on reduced clusters could not be converged, and a hybrid-DFT (B3LYP) test on a gas-phase fragment confirms the direction of the Glu116 effect on the His–Cu gap but not its size (Supporting Information). (3) The on-site repulsion is chosen, not computed; the UU scan shows that the sign of the anion effect holds for any common scaling of UU above 0.55 of the chosen values (Cu repulsion above about 2.2 eV). (4) One orbital per site: the T1→\toTNC step, whose Cys–His coupling is only 5 meV to 8 meV in this description, is not represented; Y116E lowers the T1–T3A site-energy difference of the model by 26 meV to 62 meV but leaves it downhill in all models. (5) Mostly static structures: the screen relaxes only hydrogens; a partial relaxation of the T2 ligand and the residue-116 group changes the absolute penalties strongly but leaves the mutant effect unchanged (20 vs 24 meV), and the surrounding protein never responds; empirical pKapK_a values (±1 unit), with the Cu ions treated only approximately by PROPKA. (6) The reorganization energy is taken from the literature range for the rate estimate. (7) No experimental validation yet. (8) The probe overestimates the effect of the real mutant by a factor of two to four; the probe-model hardware results validate the execution, not the size of the real effect.

Conclusions

An enzyme-derived model of the charge-transfer junction of laccase was executed on the Euro-Q-Exa quantum computer with meV precision, and the hardware reproduced the effects of hydroxide and halides, their fingerprint and the effect of mutations. Classical screening in three dielectric models, followed by hardware tests and a realistic glutamate model, identifies Y116E as a candidate that lowers the CT penalty of all three anions by 10 meV to 30 meV through the site energy of the T3 copper and weakens hydroxide binding. The measured dynamics is effectively single-particle, and on a correlated benchmark the hardware does not yet resolve electron correlation; the error rates required for that are three to eight times below the present ones. The work thus provides a testable mutant prediction and a quantitative, hardware-based statement of what present quantum processors can and cannot add to the modeling of electron transfer in metalloenzymes.

Data and code availability

Code, quantum circuits (QPY), raw hardware results, analyses and figure scripts are available in the project repository at https://github.com/lucia-malickova/quantum. Software versions are listed in the Supporting Information.

Acknowledgement The author gratefully acknowledges the EuroHPC Joint Undertaking for awarding access to the Euro-Q-Exa quantum computer under project ID EHPC-QCP2026Q01-017 and thanks the Leibniz Supercomputing Centre (LRZ) for the provisioning and support of the quantum computing infrastructure essential to this publication.

Use of AI tools

An AI assistant (Claude, Anthropic) was used for data processing: writing and running the analysis and modeling scripts, processing the hardware results, preparing the figures and drafting and editing the text. All quantum computations were executed on Euro-Q-Exa. The author designed the study, checked the results and is responsible for the content.

Supporting Information Available

The Supporting Information is appended after the references (Sections S1–S9, Figures and Tables S). Supporting Information: list of all hardware runs (jobs, ladders, dates), circuit depth scan, qubit maps, junction dynamics, bond currents, three-ladder multiplexing, earlier frequency runs, practical pH/chloride estimate, vibronic and bridge-model attempts, Glu116 conformer and pKapK_a tables, partial relaxation, DFT checks.

Hardware runs

Table S1 lists every run on Euro-Q-Exa system 1 (EQE1) used in this work: 63 runs, 726 jobs, 68 300 circuits and about 1.03×1081.03 \times10^{8} shots between 19 October and 12 November 2026. Job identifiers are those of the Munich Quantum Software Stack; the ladder is given by the α\alpha-chain qubits in the device numbering valid on the date of the run (the map changed between 51, 52 and 53 qubits; circuits were re-indexed to the same physical qubits).

Device characterization

Figure S1. Depth scan of an 18-qubit LUCJ circuit 47 for the T1 site: particle-number survival and output-distribution fidelity versus CZ count.

Junction dynamics, currents and multiplexing

Practical estimate for pH and chloride

Attempts beyond the reach of the hardware

Correlated bridge model. A five-site model including the bridging hydroxide of the T3 pair (10 qubits, fermionic swaps, 80/152/228 CZ for 1–3 Trotter steps) predicts a pH-dependent redirection of charge to T3B. On two ladders the post-selection survival was 4–12% and the effect was seen only as a hint (+0.017 ± 0.007 e against +0.05 e for the noiseless circuits).

EQE1 qubit map and the ladders used

Figure S2. Ringed: qubits used. Dark lines: alpha/beta orbital chains. Orange: rungs (opposite-spin interactions). Calibration job 423521. Figure S2: EQE1 qubit map and the ladders used.

Vibronic model. An electron coupled to an effective vibrational mode (λ=0.43\lambda= 0.43 eV, ℏω=79\hbar\omega= 79 meV; 4 qubits, ≈70\approx70 CZ) was not resolvable: gate noise heated the mode qubits to populations of 0.3–0.5 within 3 fs, above the maximal vibronic excitation of 0.2.

T1 →\to TNC step. In the one-orbital-per-site description the Cys453–His454 coupling is 5–8 meV in all models (period ≈90\approx90 fs, far beyond the ≈6\approx6 fs window of the hardware signal), because the S 3p lone pair and the imidazole π\pi orbital are nearly orthogonal across the peptide link. The driving force ε\varepsilon(T1) − ε\varepsilon(T3A) stays positive for WT and Y116E in all dielectric models; Y116E lowers it by 26–61 meV.

Mitigation. Linear zero-noise extrapolation31,41 improved early junction runs only slightly and introduced a systematic bias of 0.03–0.05; below about 30% post-selection survival it no longer restored the signal.

Real Glu116 and protonation

WT: 87 (no Asp456), 74 (Asp456−^{-}), 83 (Asp456-H).

Bar charts comparing full ladder (incl. qubits 11, 18, 25) and quality-aware layout: Correct ($N_{\alpha}$, $N_{\beta}$), 3.1% and 7.3%; Fidelity to ideal output, 0.013 and 0.048

Figure S3. Same 18-qubit T1 circuit (68 CZ; 9 vs 7 opposite-spin rungs), one EQE1 job (425037), 4000 shots each. Effect of the qubit layout (same circuit, two layouts) on the result.

Partial relaxation

modelunrelaxedrelaxed
WT87200
WT + Asp456−^-74201
Y116E probe20176
Glu116 c + Asp456-H54177
reduction, probe6624
reduction, real mutant2024

Table S3. OH−OH^- penalty on the His454 channel (meV, GFN2-xTB, ε≈4\varepsilon\approx4) before and after relaxing the T2 ligand, the residue-116 group and all hydrogens (other heavy atoms fixed).

Relaxed hydroxide: Cu(T2)–O 1.93 Å, O–Cu(T3A) 4.0 Å (crystal geometry 2.35 and 5.9 Å). The relaxation of the Y116E-probe OH−OH^- model stopped at the 150-step limit.

DFT checks of the tight-binding parameters

Full clusters (not completed). A DFT check on reduced clusters (no T1 site and no Cys453; 96–104 atoms, 3 Cu; UPBE, density fitting, gas phase) could not be converged with the available resources: DIIS stalled at ∣g∣≈10−3|g| \approx10^{-3}, and a second-order restart converged towards a different, higher-energy open-shell solution.

Fragment test of the mechanism. A fragment cut from the crystal-geometry clusters contains T3A Cu with all its donor ligands (His454 with its backbone, the other ligand molecules and the bridging hydroxide) and the T2 water (33 atoms, charge +1, doublet); the perturbation is the residue-116 group of the probe or of Glu116 conformer c (charge −1). With the fragment-orbital procedure of the main text, the shift of the gap ε(His454)−ε(T3A)\varepsilon(\mathrm{His454})-\varepsilon(\mathrm{T3A}) caused by the charged group is −295 meV (GFN2-xTB) and −1554 meV (UB3LYP; def2-SVP on Cu, 6-31G elsewhere) for Glu116, and +48 and −418 meV for the probe. For the real glutamate both methods agree on the direction of the mechanism, a lower His–Cu gap, but not on its size; for the probe they disagree in sign. A gas-phase fragment without screening, in which an uncompensated anion shifts the Fermi level and thereby the choice of the Cu orbital, is too crude for a quantitative comparison, and we do not use it as one.

Bar charts comparing sampled and spin-completed subspaces for the lowest doublet and lowest quartet

Figure S4. Spin contamination of variational results for the T1 site (spin trap). Numbers above bars: ⟨S2⟩\langle S^2\rangle of the result (bold orange = wrong spin). 18-qubit T1 Cu site, CAS(15e,9o), 10 most frequent sampled determinants; errors below 0.001 mEh drawn at 0.001 (exact). In the hole-encoded run the spin-completed subspace reaches the full 324-determinant space.

Software

tblite 0.8.0 (GFN2-xTB), 32 ffsim 0.0.85, Qiskit 2.6.1, 39 gemmi 0.7.5, NumPy 2.5.3, SciPy 1.18.1, PROPKA 3.5.1, 35 PySCF 2.14.0.

Figure S5. Spin states of the TNC, CASCI(17e,12o), classical reference.

Figure S10. EQE1 jobs 425947–425960: complex bond signal ρ(t)=⟨cHis454†cT3A⟩\rho(t) = \langle c^{\dagger}_{\mathrm{His454}} c_{\mathrm{T3A}} \rangle from two readout bases, 0.25–3.5 fs, Pauli twirling; joint fit Im⁡ρ=asin⁡ωt e−γt\operatorname{Im}\rho= a \sin\omega t\,e^{-\gamma t}, Re⁡ρ=b(1−cos⁡ωt)e−γt+c\operatorname{Re}\rho= b(1-\cos\omega t)e^{-\gamma t} + c. ℏω\hbar\omega = energy of the His454 → T3 Cu charge-transfer excitation of the model (the sign of ρ\rho follows the sign of the hopping). Errors: bootstrap over twirl instances. Figure S10: First CT-frequency runs (0.25–3.5 fs).

Figure S11. Reproducibility of the pH shift of the CT energy across runs and ladders.

Figure S12. CT shift on three ladders, 0.25–6 fs.

Figure S13. Estimate of the internal electron-transfer slow-down with pH (Marcus rate ratio from the junction model and an assumed apparent pKapK_a) and of chloride inhibition with literature KiK_i values. 11 An estimate, not a measurement.

References

  1. [1]Solomon, E. I.; Sundaram, U. M.; Machonkin, T. E. Multicopper Oxidases and Oxygenases. Chem. Rev. 1996, 96, 2563–2606.pubmed.ncbi.nlm.nih.gov/11848837
  2. [2]Solomon, E. I.; Heppner, D. E.; Johnston, E. M.; Ginsbach, J. W.; Cirera, J.; Qayyum, M.; Kieber-Emmons, M. T.; Kjaergaard, C. H.; Hadt, R. G.; Tian, L. Copper Active Sites in Biology. Chem. Rev. 2014, 114, 3659–3853.DOI
  3. [3]Jones, S. M.; Solomon, E. I. Electron Transfer and Reaction Mechanism of Laccases. Cell. Mol. Life Sci. 2015, 72, 869–883.DOI

Paper details

vlnc:W_7cswqxgf
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Computer Science

Claude

View statement
Looking for related papers…
Contents