Detection of Water Molecules on the Radical Transfer Pathway of Ribonucleotide Reductase by 17O Electron–Nuclear Double Resonance Spectroscopy

The role of water in biological proton-coupled electron transfer (PCET) is emerging as a key for understanding mechanistic details at atomic resolution. Here we demonstrate 17O high-frequency electron–nuclear double resonance (ENDOR) in conjunction with H217O-labeled protein buffer to establish the presence of ordered water molecules at three radical intermediates in an active enzyme complex, the α2β2E. coli ribonucleotide reductase. Our data give unambiguous evidence that all three, individually trapped, intermediates are hyperfine coupled to one water molecule with Tyr-O···17O distances in the range 2.8–3.1 Å. The availability of this structural information will allow for quantitative models of PCET in this prototype enzyme. The results also provide a spectroscopic signature for water H-bonded to a tyrosyl radical.

90 % 17 O labelled water was purchased from Sigma Aldrich. The incorporation of unnatural amino acids into E. coli ribonucleotide reductase followed the previously reported protocols. [1][2] Purified 2 (wild-type, Y730F, NH2Y731 and NH2Y730) was exchanged into 5 mM HEPES buffer (pH 7.6) containing 1.5 mM MgSO4, 0.1 mM EDTA and 1 mM -mercaptoethanol with Amicon spin filters (30 000 NMWL). 100 µL protein solution was supplemented with 300 µL buffer and spun at 12 000 g for 5 min. This process was repeated 6 times. ATP and CDP were added and the protein concentration was adjusted with assay buffer (50 mM HEPES pH 7.6, 15 mM MgSO4, 1 mM EDTA) to yield a final concentration of 30 µM 2, 500 µM ATP and 167 µM CDP. 100 µL quantities of this solution were frozen in liquid nitrogen and lyophilized overnight. The samples were rehydrated in 10 µL H2 17 O to yield solutions of 300 µM 2, 5 mM ATP and 1.67 mM CDP in assay buffer. Recovery of wild-type (wt) 2 activity after the lyophilization procedure was checked by spectrophotometric activity assay and found to be 90 -100 % (data not shown). Purified 2 (wt, F3Y122, F3Y122/E52Q) was exchanged into assay buffer with the abovementioned protocol and had the following concentrations: 890 µM wt-2, 980 µM F3Y122-2, 1600 µM F3Y122/E52Q-2. EPR samples were prepared by mixing the previously described 2 solutions containing substrate and effector with the corresponding 2 solution (Table S1) and addition of H2 17 O to a final concentration of 180 µM 22, 3 mM ATP and 1 mM CDP. The final amount of H2 17 O was approx. 80 %. The reaction mixtures were hand quenched in liq. N2 inside EPR tubes. The quench times followed the previously established protocols for maximum radical yield. [3][4][5] EPR samples containing only 2 with H2 17 O were prepared by diluting the abovementioned solutions of 2 (wt and F3Y122) with H2 17 O to a final protein concentration of 180 µM and approx. 90 % H2 17 O. The 2 solution was left to incubate for 10 min at 4°C to allow for sufficient exchange of water molecules within the protein, i.e. close to Y122, and subsequently frozen in liq. N2 inside the EPR tubes. W-Band samples contained 2 µL protein mixture in 0.9 mm OD/0.5 mm ID suprasil tubes. 263 GHz samples contained 30 -50 nL protein mixture in 0.33 mm OD/0.2mm ID suprasil capillaries. 3.4 T EPR experiments were performed on a Bruker E680 pulsed W-band spectrometer with 2 W microwave power output. The optimal pulse length was determined by a Rabi nutation to 8-10 ns for a π/2 pulse at maximum output power. Echo detected EPR spectra for radical yield determination were recorded with a Hahn echo pulse sequence (π/2 -τ -π -τ -echo) with τ = 290 ns. Shot repetition time (SRT) and shots/point varied for different temperatures and radicals and are given in the figure captions. 9.4 T experiments were performed on a Bruker E780 pulsed 263 GHz quasi-optical spectrometer with 100 mW microwave power output. The optimal pulse length was determined by a Rabi nutation to 30 -32 ns for a π/2 pulse at maximum output power.
To optimize ENDOR experiments at 50 K, we measured the relaxation properties of each radical.
All relaxation experiments were recorded at the maximum of the EPR line, i.e. B0║gy. The electron spin-lattice relaxation time (T1e) was determined via an inversion recovery experiment (π -t -π/2τ -π -τ -echo, inset Fig S1). A bi-exponential fit (Eq. 1) to the echo intensity as a function of the time-interval t yielded T1e as the largest time constant, while the smaller time constant was assigned to spectral diffusion. T1e as the longest relaxation time determines the shot repetition time (SRT) of all experiments should be at least 2-3 times longer than T1e.
At 50 K, T1e is 1.6 ms for the tyrosyl radical Y356 • /Y730F- and 2.9 ms and 4.6 ms for the two aminotyrosyl radicals NH2Y731 • and NH2Y730 • , respectively. Therefore all 50 K ENDOR experiments of Y356 • were performed with 5 ms SRT, while 10 ms SRT was used for the two amino tyrosyl radicals. The phase memory time Tm strongly influences the Mims ENDOR sensitivity (see SI1.5). It was measured by recording the stimulated echo intensity as a function of the time delay  (π/2 - -π/2 -T -π/2 -τ -echo, inset Fig. S2). Tm is the time constant of a mono exponential decay fit to the experimental data (data not shown): This experiment was repeated for increasing times T, i.e. the separation of the second and third π/2 pulse. The experiments show that initially, Tm decreases almost exponentially with increasing pulse separation time T for all investigated radicals. At T = 40 µs, which was used for all ENDOR experiments, the phase memory time is approximately 700 -800 ns for the three trapped radical intermediates.
ENDOR experiments at 94 and 263 GHz were recorded with the Mims 6 pulse sequence (π/2 -τπ/2 -RF -π/2 -τ -echo) most sensitive to small hyperfine couplings. The microwave power at both instruments was reduced to produce π/2-pulses of 40 ns with an excitation bandwidth of 25 MHz/ 0.7 mT for increased orientation selectivity. The τ -value was set as explained in the following section (SI1.5). A 250 W RF-amplifier (250A250A, Amplifier research) was used to increase the RF pulse power. RF pulse length was optimized with a RF Rabi nutation experiment (see Fig. S3). At W-band frequency, 40 µs RF pulses with an excitation bandwidth of 25 kHz were used, while 75 µs pulses with an excitation bandwidth of 13 kHz were used at 263 GHz. The difference in experimental setup at 3.4 T vs 9.4 T is caused by the different ENDOR resonator design and efficiency as well as varying output powers of the amplifier at the different frequencies.
All ENDOR experiments were recorded using stochastic RF acquisition with 30 shots per point (SPP). [7][8][9] Comparison of experiments with 1 SPP vs 30 SPP showed negligible saturation effects (data not shown), while a significant shortening of measurement time was observed for the latter method. This is caused by the reprogramming time of the spectrometer upon change of the RF frequency (i.e. between each x-axis data point), which is around 30 ms and does not occur between shots at the same frequency. ENDOR experiments were recorded at 50 K at W-Band and at 20 K at 263 GHz. The temperature was chosen to achieve the best compromise between high signal intensity and short relaxation time for quick experimental shot repetition. The sensitivity S of the Mims ENDOR experiment is described by the product of the ENDOR efficiency (FENDOR) and the echo intensity Iecho: = ENDOR • echo (3) The Mims ENDOR efficiency for a nuclear spin ½ system can be described by a periodic function, depending on the effective hyperfine coupling Aeff : 10 This formula breaks down for nuclear spins I > ½, if the quadrupole coupling is on the order of the hyperfine coupling. 11 The approach was adapted for I = 1 nuclei by calculating Aeff from a combination of hyperfine and quadrupole coupling by Hoffman and coworkers. A similar treatment to I = 5/2 nuclei was however deemed unfeasible. 11 Therefore, the blind spots in a 17 O Mims ENDOR spectrum have to be treated within the density matrix formalism and an explicit calculation of the coherence transfer pathway of the Mims sequence for each individual spin system of interest. This can be achieved with the easyspin routine saffron 12 (vide infra). The second term on the right side of equation 3 nevertheless is true and has to be considered, since Iecho decays exponentially as a function of the phase memory time Tm (see Eq. 2). The choice of the optimal τ-value therefore depends on Tm (here ~ 0.7 s, see Fig. S2) and the expected coupling parameters (A and P). Fig. S4 shows the simulated ENDOR spectrum of Y356 • with blind spots (color, Mims ENDOR simulation) and with pure tensor simulation (black). The ENDOR spectra are scaled by the phase memory time. The simulations show that no periodic blind spots are clearly visible in the simulated spectra. The shape of the central nuclear transitions depends slightly on τ but the main difference is the change in overall signal intensity. The overall maximum ENDOR signal can be achieved with values between 400 and 600 ns. In this study, we chose a τ-value of 390 ns to give the best compromise between ENDOR sensitivity and echo intensity, since the latter also influences the overall signal-to-noise of the spectra. All ENDOR spectra were simulated using the EasySpin software package. 13,12 The simulated spin system was based on the literature g-values of the radicals Y356 •5 , NH2Y731 • and NH2Y730 •3 as well as the nuclei with the largest hyperfine coupling constants (Table S2). In case of Y356 • , this was only the -methylene proton, while the amino nitrogen was included for NH2Y731 • and NH2Y730 • . Additional couplings were neglected, since they significantly prolong calculation times while their contribution to the orientation selection and therefore the simulated 17 O ENDOR spectra was found to be negligible. 17 O ENDOR spectra were calculated with the saffron routine employing full tensor diagonalization (See S4). An excitation bandwidth of 25 MHz was used to select the orientations. A uniform ENDOR linewidth of 60 kHz was used for all simulations. All simulated ENDOR spectra were normalized to unity for comparison with equally treated experimental spectra. The quadrupole coupling size calculated by DFT was generally too large, evident from simulation of the ENDOR spectra. The deviation can in part be explained by the absence of other hydrogen bonding partners to the water molecule in the small DFT models, which are known to reduce the quadrupole coupling constant. 14 Therefore, the literature known coupling constants for pure water in ice: P={-0.02 -0.32 0.34} MHz 14 was chosen and found to be in good agreement with the data of this work. DFT models were calculated using the Orca 4.0.1.2 software package. 15 Geometry optimization was performed using the BP86 functional [16][17] in combination with the Ahlrichs' def2-TZVP basis set of triple-ζ quality [18][19] for all atoms and the RIJCOSX approximation(def 2/J auxiliary basis set) 20 . Grimmes dispersion correction (d3bj) [21][22] was applied on top of the SCF calculations. Single point energies and EPR parameters were calculated from the geometry optimized structures employing the B3LYP 17, 23-24 functional in conjunction with the EPR-II basis set for all nuclei. 25 The abovementioned RIJCOSX approximation and dispersion correction were also used. The protein environment was approximated by a conductor-like polarization model (CPCM) with polarity  = 24.
The geometry of the small models of a tyrosyl and amino-tyrosyl radical model was initially optimized without a water molecule and only restricted to the experimentally known -H dihedral angles (C2-C1-C-H1) of 70° and -120° for Y356 • and NH2Y731 • , respectively. 3, 5 A water molecule was added and its geometry was optimized while the C3-C4-OTyr ··· HH2O dihedral angle  and the coordinates of all the radical atoms were fixed. In case of the amino tyrosyl radical, the amino protons were not fixed, since they are potential hydrogen bond partners for the water molecule. 36 individual models were calculated with  values in increments of 10° from 0° to 350°. The OTyr ··· HH2O distance r was not fixed in the models.
Due to the half-site reactivity of E.coli RNR [26][27] , the EPR spectra of hand quenched samples contain the contribution of two radicals. One signal is the trapped radical in the RT pathway of one / pair of the active  complex: Y356 • /Y730F- (Fig. S5A), NH2Y731 • (Fig. S5B) or NH2Y730 • (Fig. S5C). The second signal is Y122 • or F3Y122 • in the unreacted / pair. The two radical species have very different relaxation times due to their different environments. The signal associated with the radical at residue 122 relaxes very fast due their proximity to the di-iron center, making it fully visible only at very low temperatures. 28 Therefore, echo-detected EPR spectra of the samples were recorded at 10 K (Fig. S5, red lines) and the EPR spectrum of the respective tyrosyl radical at residue 122 (Fig. S5, blue lines) was subtracted. 29 The relative amount of radical trapped was then determined from the integral of the full EPR signal versus the integral of the subtracted spectrum (Fig.  S5, cyan lines). The resulting radical yields are displayed in Fig. S5 and Table S1. Figure S5: Echo-detected EPR spectra of reaction mixtures (red) with the radicals Y356 • /Y730F- (A), NH2Y731 • (B) and NH2Y730 • (C), reference spectra of the respective resting 2 (blue) and subtraction of the two spectra (cyan). Relative amounts of radical determined by integration and given in the figure. Canonical g-Tensor orientations (B0║gx, B0║gy and B0║gz) at which the orientation selective ENDOR spectra were recorded are marked by gray areas. Experimental parameters: Temperature = 10 K, Pulse Sequence: π/2 -τ -πτ -echo, π/2 = 10 ns, τ = 290 ns, 5 shot/point, 100 ms SRT.

• •
The 17 O ENDOR experiments in this study were performed on radical mixtures that contain more than 50% unreacted tyrosyl radical Y122 • or F3Y122 • at the diiron cofactor. (See Figure S5 and Table  S1). To exclude 17  These ENDOR experiments were performed at 10 K due to the fast relaxation properties of the two radicals caused by the adjacent di-iron cluster (see SI2). Experiments were performed at the maximum of the EPR signal, i.e. B0║gy and with full microwave power to increase the ENDOR signal.
The ENDOR spectra show a small amount of a mostly featureless signal in a range of ±0.2 MHz around the 17 O Larmor frequency. This contribution is smaller at temperatures higher than 10 K due to the fast relaxation properties which cause signal loss. The results confirm that the distinct coupling features detected in the 17 O ENDOR experiments of the pathway radical intermediates originate from water molecules at the intermediates themselves and not from the tyrosyl radicals associated with the di-iron cofactor. Nuclear spins I > 1/2 such as 17 O (I = 5/2) exhibit quadrupole coupling due to the interaction between the electric field gradient and the non-uniform charge distribution within the nucleus. The interaction is described by the traceless quadrupole coupling tensor P in the nuclear quadrupole Hamiltonian and causes an energy shift of the nuclear spin states in addition to the hyperfine interaction. 30 = e 2 4 (2 − 1) · ( The spin Hamiltonian for a coupled electron S = 1/2 to a nuclear spin I = 5/2 is described by the spin Hamiltonian: 30 If the hyperfine and quadrupole interactions are small as compared to the nuclear Zeeman interaction, they can usually be described within the high-field approximation, which significantly simplifies the spin Hamiltonian and makes simulation of the system fast. In the high-field approximation, only the zz-components of the coupling tensors are taken into account, assuming that ̂ and ̂ are dominating: If the high field approximation is justified, simulation of the ENDOR spectra with high field approximation and full matrix diagonalization should yield identical spectra. Figure S7 shows the simulated ENDOR spectra for a series of spin systems (Table S3) with coupling values similar to the ones in this work simulated with the two methods at a magnetic field of 3.35 T, which corresponds to the one used in the majority of experiments in this work. We simulated the individual nuclear transitions (color) and the full spectrum (black) to better illustrate the effect of nuclear quadrupole coupling on the spectral line shape.  : Simulated ENDOR spectra (black) and individual nuclear transitions (color) for a coupled spin system S=1/2 I=5/2 calculated in high field approximation with 1 st order perturbation theory (HF) and with full tensor diagonalization (Matrix). Coupling parameters are reported in Table S3.

Simulation aiso Tx Ty Tz Px Py Pz
The first three spin systems (A, B and C) have very small quadrupole coupling constants and purely dipolar (A), purely isotropic (B) or combined (C) hyperfine coupling tensors. The nuclear transitions within the two electron spin manifolds have almost identical frequencies and only vary in intensity due to different transition probabilities. The simulated spectra for purely dipolar hf coupling show in both cases a small doublet structure close to the larmor frequency (A). In case of isotropic hf coupling (B) the signals are very sharp and only broadened by the ENDOR linewidth. In case of the rhombic hf tensor (C) the signals are additionally broadened by the dipolar hf coupling. For these systems, both methods yield the same result, justifying the high field approximation.
The second set of spin systems (D, E and F) have axial quadrupole tensors in the size observed experimentally and either dipolar (D), isotropic (E) or combined (F) hyperfine coupling tensors. Even though this quadrupole tensor shape is not expected in the systems of this study, the simulations are shown here to better illustrate the impact of quadrupole coupling on the ENDOR spectra. The individual nuclear transitions in the two electron spin manifolds are no longer energetically equivalent and the simulated ENDOR spectra contain sharp central signals (red) split by the hyperfine coupling, which correspond to the mS= -1/2 to mS=+1/2 nuclear transition as well as broad signals corresponding to the other nuclear transitions. The central transitions are not affected by quadrupole coupling in the high field approximation, so they have the same line shape as in A, B and C. The simulations performed with matrix diagonalization show that the central transitions are affected by the quadrupole coupling and become broadened and in case of a hf coupling tensor with dipolar coupling contribution (D and F) also asymmetric around the 17 O Larmor frequency. For a small, purely dipolar tensor (D) the small doublet structure is lost and the signal at the Larmor frequency becomes a single asymmetric peak.
The third set of spin systems has rhombic quadrupole coupling tensors (G, H and I), equivalent to the systems we investigated in this work. The quadrupole transitions become much broader and therefore less intense due to their rhombic tensor shape. In case of small dipolar coupling (G) the effect observed in the previous section is amplified and all hf coupling information is lost in the spectrum within a single central peak. For the case of isotropic hf coupling (H) the broadening effect on the central transition is limited while a significant broadening and asymmetry is observed for rhombic hf and quadrupole tensor (I).
We conclude, that the high-field approximation fails for the systems investigated in this work and full matrix diagonalization has to be used for all spectral simulations. We also conclude that the asymmetry of the ENDOR spectra is caused by the size and shape of the hyperfine and quadrupole tensors and therefore has to be considered during simulation.
Mims ENDOR spectra of the three radical intermediates Y356 • (Y730F-)(A), NH2Y731 • (B) and NH2Y730 • (C) were recorded at the field positions corresponding to the three canonical g-tensor orientations marked in Fig. S5: B0║gx (blue), B0║gy (red) and B0║gz (cyan). Baseline corrected (1 st order polynomial) spectra are shown in light gray and 4 th order Savitzky-Golay filtered spectra (20 pt window) are shown in color. The difference in signal-to-noise in the spectra is a consequence of several factors: a) radical yield of the sample with the lowest overall S/N for NH2Y731 • , b) different EPR signal intensity at the field positions with B0║gy giving the best S/N since it is the maximum of the EPR spectrum, c) different ENDOR sensitivity at the specific g-tensor orientations with B0║gz having more ENDOR sensitivity than B0║gx for the Y356 • radical, whiles this is reversed for the amino-tyrosyl radicals. This is caused by the smaller difference of gx and gy in the amino-tyrosyl radicals, resulting in the excitation of more orientations and the larger similarity of the gx and gy spectra. Orientation selective ENDOR simulations with the spin system parameters specified under SI 1.6 and the 17 O coupling parameters from Table S4 are shown in black. DFT values are shown in blue. All coupling constants given in MHz. Euler angles given in degrees and defined from the A and P to the g tensor based on the y convention. A and P tensors are chosen so that |Ax/Px|< |Ay/Py| < |Az/Pz|.    where r is the inter spin distance and ρ the spin density. Since the water molecule is coordinated almost in plane, we assumed an interaction only with the spin density on the oxygen atom, which was estimated from the DFT Loewdin spin population 31 analysis calculation to be 0.3 for the tyrosyl radical. We note that the point-dipole model agrees well with the dipolar contribution of 17 O computed by DFT ( Figure S10A). For the 1 H coupling, the experimental values A correspond to T||, as the tensors are almost purely dipolar. For the two 1 H tensors, the DFT computed values slightly exceed the point-dipolar approximation. The experimental values are found right between these two (Figure S10B).
Overall, the distance analysis shows that a H bond distance of 1.85 ± 0.05 Å leads to the best agreement of the 17 O couplings detected in this work, while a distance of ~ 1.95 ± 0.05Å leads to the best agreement with the 1 H couplings detected in Nick et al 5 . Taken this information together, we conclude that the H bond distance is 1.9 ± 0.05 Å. Figure S10C,D,E illustrates that simulation of the 17 O ENDOR spectra with hf couplings predicted for a distance of 1.9 Å shows peaks and line shapes well compatible with the data.  Overall, the angle sweep for the amino-tyrosyl radical shows that coordination with =180°, i.e. in ring plane opposite to the NH2-group results in the energetic minimum and 17 O hyperfine couplings compatible with the experimental data. A distance sweep for this coordination angle shows, that the experimental 17 O couplings of NH2Y730 • and NH2Y731 • would be most compatible with r(OTyr ··· OH2O) = 2.8 Å. (Fig. 12A) This distance is however not compatible with the experimentally observed 1 H hf couplings for NH2Y730 • (Fig. 12B) (and also with those of NH2Y731 • since they are even smaller. See Table 1). We conclude that the small model cannot describe the amino tyrosyl radicals sufficiently due to the absence of surrounding second sphere residues.

•
The intermediate Y356 • is accessible from different biochemical constructs which share the mutation, 2-F3Y122. An additional mutation of Y730F in the pathway results in the highest radical yield. The F3Y122-2/Y730F-2 construct was thus investigated both at 94 and 263 GHz. The F3Y122/E52Q-2/2 was also studied as this double mutation leads to a tight 22 complex that is active 33 which gave rise to the first high-resolution structure of the holocomplex by cryo-EM. 34 Additionally, F3Y122-2/F2Y731-2 construct is planned for measurement of the distances between Y356 and the fluorines in F2Y731. Figure S13 shows 94 GHz Mims ENDOR experiments for four possible combinations of subunit pairs prepared under similar conditions. The 17 O ENDOR line shape and position of the peaks do not vary significantly. Minor differences in asymmetry can be explained by slight variations in field positions. The spectra generally show that a specific coordination of one water molecule exists and is conserved across the biochemical constructs.  35 . Experimentally observed H bonds and are shown within the optimized DFT model. The water and its mechanistic significance was proposed in that study, but direct experimental evidence was missing. The present study assigns the H bond unambiguously to a water molecule. Hf coupling parameters of the water molecule are reported in the main text in Table 1 as DFTlarge.