Proton-Electron Transfer to the Active Site Is Essential for the Reaction Mechanism of Soluble Δ⁹-Desaturase

Abstract

Full understanding of the catalytic action of non-heme iron (NHFe) and non-heme diiron (NHFe₂) enzymes is still beyond the grasp of contemporary computational and experimental techniques. Many of these enzymes exhibit fascinating chemo-, regio- and stereoselectivity, in spite of employing highly reactive intermediates which are necessary for activations of most stable chemical bonds. Herein, we study in detail one intriguing representative of the NHFe₂ family of enzymes: soluble Δ⁹ desaturase (Δ⁹ D) which desaturates rather than performing the thermodynamically favorable hydroxylation of substrate. Its catalytic mechanism has been explored in great detail by using QM(DFT)/MM and multireference wave function methods. Starting from the spectroscopically-observed 1,2-μ-peroxo diferric P intermediate, the proton-electron uptake by the P is the favored mechanism for catalytic activation since it allows a significant reduction of the barrier of the initial (and rate-determining) H-atom abstraction from the stearoyl substrate as compared to the ‘proton-only activated’ pathway. Also, we ruled out that a ‘Q-like intermediate’ (high-valent diamond-core bis-μ-oxo-[Fe^IV]₂ unit) is involved in the reaction mechanism. Our mechanistic picture is consistent with the experimental data available for Δ⁹D and satisfy fairly stringent conditions required by Nature: the chemo-, stereo- and regioselectivity of the desaturation of stearic acid. Finally, the mechanisms evaluated are placed into a broader context of NHFe₂ chemistry provided by an amino acid sequence analysis through the families of the NHFe₂ enzymes. Our study thus represents an important contribution toward understanding the catalytic action of the NHFe₂ enzymes and may inspire further work in NHFe₍₂₎ biomimetic chemistry.

Graphical Abstract

1. INTRODUCTION

Over the last decades, mononuclear and binuclear non-heme iron enzymes (NHFe and NHFe₂) have raised considerable interest owing to their versatility in catalyzing a multitude of essential biochemical transformations, such as functionalization of the C–H bonds of alkyl chains and aromatic rings, heterocyclic ring opening and formation, epoxidation, or the C–C bond desaturation.^1,2 Most of the NHFe₂ enzymes share a common structural motif comprising a four-helix bundle protein fold that accommodates two iron ions, which are coordinated by two histidine and four glutamate residues to form the active site. This implies that NHFe₂ may also share specific reaction patterns such as the activation of the resting state of an enzyme via the addition of O₂ (or resting state via H₂O₂), leading to more reactive oxygenated intermediates (vide infra).

Still, geometric and electronic structure properties of NHFe₂ intermediates and their contributions to reactivity remain largely underexplored. From a theoretical perspective, the NHFe₂ enzymes belong to one of the most difficult systems to study computationally^3,4 due to strong correlation effects given by the multireference character of many of their electronic states^5,6,7,8 which often precludes the straightforward usage of standard density functional theory (DFT). Instead, multireference ab initio wave-function methods need to be applied. However, their application is far from routine and deep understanding of the many subtleties in a particular computational protocol limits their widespread practice.^3,4^,9 Despite the above drawbacks, considerable advances in understanding NHFe₂ were made by correlating quantum chemical calculations with experimental – mostly spectroscopic – data.^2,10

Δ ⁹ Desaturase (Δ⁹D, EC 1.14.19.2., Pfam ID PF03405) is an O₂-dependent binuclear non-heme iron enzyme that evolved to convert stearic acid to oleic acid as a part of the fatty-acid metabolic pathway of plants.^11,12,13,14 The process, which is highly chemo-, regio-, and stereoselective, involves a cis double bond insertion into a C₉–C₁₀ bond of a stearic acid via two sequential H-atom abstraction (HAA) reactions. Prior to an initial C₁₀–H bond activation,¹⁵ preparatory stages in Δ⁹D catalytic cycle include: (i) substrate binding within the active site pocket, which perturbs ligation coordination spheres of both Fe centers from penta,penta (5C,5C) to penta,tetra (5C,4C) coordination and thus increases the affinity of the active site for O₂ binding;^16,17 (ii) O₂ activation yielding the experimentally observed peroxo-diferric P intermediate (cf. Figure 1);¹⁸ (iii) conversion of the non-reactive P into an intermediate capable of the initial HAA from the stearic acid.¹⁹

Figure 1.

Proposed oxygenated intermediates of selected NHFe₂ enzymes. Δ⁹D = Δ⁹ desaturase; sMMO = soluble methane monooxygenase; BoxB = benzoyl coenzyme A epoxidase; AurF = p-aminobenzoate N-oxygenase; T4MO = toluene 4-monooxygenase; Rbr = rubrerythrin; RNR = ribonucleotide reductase; CmlI = arylamine oxygenase.

A range of HAA reactive intermediates has been previously suggested for Δ⁹D or other NHFe₂ enzymes (examples are given in Figure 1).^{18,20,21,22,23,24,25,26,27} However, none of these has been experimentally captured and characterized for Δ⁹D. This contrasts for example soluble methane monooxygenase (sMMO), where the Fe^IIIFe^III P intermediate converts to the high-valent (Fe^IVFe^IV) intermediate Q,^28,29,30 or to the R2 subunit of class I ribonucleotide reductase (RNR), where the peroxo moiety P is proposed to be first protonated to yield a 1,2-μ-hydroperoxo Fe^IIIFe^III P’ that accepts an exogenous electron leading to a mixed-valence Fe^IVFe^III intermediate X^{31,32,33,34,35,36}.

Herein, we build on a detailed investigation of initial stages of O₂ activation in the Δ⁹D active site reported earlier by our group^18,19 and explore several reaction pathways in order to complete the catalytic cycle. To this aim, we employ the hybrid DFT/molecular mechanics (DFT/MM) calculations, which we complement for key reaction steps by using advanced multiconfigurational/multireference methods such as multireference configuration interaction, MRCI;^37,38 complete active space second-order perturbation theory, CASPT2;^39,40,41 and multiconfiguration pair-density functional theory, MC-PDFT^42,43. As revealed in this study, several possible activations of P, including proton-assisted, O–O cleavage, and proton-electron transfer (PET)-assisted pathways may all represent viable alternatives of the Δ⁹D reaction mechanism. Our calculations tend to favor the PET-assisted activation of the P. Additionally, computed data and postulated proton and electron transfer pathways for the PET-assisted mechanism of P activation are further corroborated by the extensive bioinformatics study of various NHFe₂ enzymes as well as different plant, archaeal, and bacterial desaturases.

2. COMPUTATIONAL DETAILS

2.1. Protein Setup.

All reported calculations are based on the fully equilibrated all-atom QM/MM model built from the crystal structure of Δ⁹ stearoyl-acyl carrier protein desaturase obtained at the resolution of 2.4 Å (Protein Data Bank accession code 1AFR).⁴⁴ Because of the absence of the X-ray structure of the [protein…substrate] complex, a simplified substrate model (i.e., CH₃-(CH₂)₁₆-COO-phosphopantetheine linker capped by a methyl group instead of an acyl carrier protein) was docked manually into the substrate entrance channel of the Δ⁹D in our previous study.¹⁹ In this work, a position of the substrate in the Δ⁹D active site was further refined - employing the newly published X-ray structure with the resolution of 3.35 Å (PDB code 2XZ1)⁴⁵ - by simulated annealing of eight different initial substrate conformations, and the QM region was expanded by including amino acids in higher coordination spheres of the two iron ions (see Figure S1). The system preparation for the QM/MM calculations were described in detail in Ref. 18 and was followed accordingly in the presented study (see the Supporting Information).

2.2. QM/MM Calculations.

The ComQum software^46,47 was used for all QM/MM calculations. Within the QM/MM approach, the entire system was divided into three regions as follows: (i) System 1 (flexible region; treated at the QM level); (ii) System 2 (flexible region; treated at the MM level); (iii) System 3 (fixed region; MM level). System 1 comprised the active site of an enzyme, corresponding to 196 or 197 atoms (based on the protonation of the peroxo bridge − vide infra). The flexible MM region (System 2) consisted of all residues and water molecules within 3 Å of any atom in System 1, truncated at the ‘per-residue’ level (resulting in 48 amino acids + 2 water molecules), and the fixed region (System 3) comprised of the remaining part of the protein and the solvent water molecules (643 amino acids + 6980 water molecules in total). Within the QM(DFT) calculations, the atoms specified as part of System 1 were treated quantum chemically. System 1 has been polarized by atoms in System 2 and System 3 – represented by point charges – i.e., the electrostatic embedding scheme in the QM/MM terminology. A junction between Systems 1 and 2 was modeled utilizing the hydrogen link-atom approach.⁴⁸ A detailed description of the contributions to the total QM/MM energy as well as many technical details on the ComQum software can be found elsewhere.^46,47,49 In brief, the total (QM/MM) energy is calculated as:

$$E_{\text{QM}/\text{MM}}=E_{\text{QM}-\text{pchg}}+E_{\text{MM}123}-E_{\text{MM}1},$$ (1)

where E_QM-pchg corresponds to the QM energy of System 1 embedded in a set of point charges of the System 2 and 3 (MM part), E_MM123 is the MM energy of the entire system, and E_MM1 is the MM energy of System 1 with MM charges of the System 1 zeroed.

All DFT calculations (QM region) were performed using the Turbomole 6.6 program.⁵⁰ Geometry optimizations were carried out using the TPSS functional⁵¹ with the def2-SV(P) basis set,⁵² including the empirical zero-damping dispersion correction (D3)⁵³, and expedited by the RI-J approximation.⁵⁴ The S = 0 (or S = 1/2 in the 1e⁻-reduced system) spin state with the antiferromagnetically-coupled iron ions in the high-spin states (S = 5/2 for Fe^III or S = 2 for Fe^IV) was considered throughout. The transition states were obtained as one- or two-dimensional scans of QM/MM potential energy surfaces (PESs). The MM calculations were performed employing the Amber ff14SB force field.⁵⁵

2.3. Single-point QM Calculations.

On top of QM(TPSS-D3/def2-SV(P))/MM equilibrium geometries, the E_QM-pchg term is evaluated with B3LYP*-D3 method (i.e., where the symbol “*” refers to B3LYP-D3 with 15 % of Hartree-Fock exchange)⁵⁶ employing the def2-TZVP basis set⁵²; the method is denoted in the text as: QM(B3LYP*-D3)/MM.

As implemented in the MOLCAS 8.2 program,^57,58,59,60 multiconfigurational/multireference approximations to wave function theory (WFT) such as the CASSCF,^61,62 CASPT2,^39,40,41 and MC-PDFT^42,43 methods were applied in the combination with the ANO-RCC^63,64 basis set to size-reduced model systems (86–87 atoms, cf. Figure 2) as prepared from the QM/MM equilibrium geometries. The ANO-RCC basis set, contracted to [6s5p3d2f1g] for Fe, [4s3p2d] for ligating N and O atoms, [3s2p] for C and non-ligating N atoms, and [2s] for H atoms, was used. The second-order Douglas−Kroll−Hess (DKH2)^65,66,67 one-electron spin-less Hamiltonian was applied for all of the WFT-based calculations to allow for spin-free relativistic effects. In all CASSCF calculations, a level shift of 5 a.u. was used to improve convergence. The Cholesky decomposition technique⁶⁸ with a threshold of 10⁻⁶ a.u. was used to approximate two-electron integrals. Complete active spaces used in CASSCF calculations are specified in the text and comprise up to 22 electrons in 18 orbitals (see Figure S2). To account for the absence of dynamical correlation in the CASSCF method, the CASPT2 and MC-PDFT energies were also evaluated (for MC-PDFT, the ftBLYP, ftPBE, ftrevPBE, tBLYP, tPBE and trevPBE functionals were employed)⁶⁹. For these WFT-based calculations, the total ‘QM/MM energy’ was then evaluated as:

$$E_{\text{QM}/\text{MM}}=E_{\text{WFT},\text{ Model }}+E_{\text{DFT},\text{FullQM}}-E_{\text{DFT},\text{ Model }}+E_{\text{MM}123}-E_{\text{MM}1},$$ (2)

where E_WFT,Model is the energy calculated for the size-reduced model in vacuo using the CASSCF, CASPT2 or MC-PDFT method, E_DFT,FullQM is the energy calculated for the full QM system (with electrostatic embedding) at the DFT level (with the B3LYP* functional), and E_DFT,Model is the B3LYP* energy calculated on the size-reduced model in vacuo. The last two terms in Eq. (2) are the same as in Eq. (1).

Figure 2.

Truncated cluster model of the Δ⁹D active site utilized in the WFT calculations.

2.4. Calculations of Reduction Potentials.

To compare pathways differing by the number of electrons involved (ΔN_el = 1), the 1e⁻-reduction potentials were approximated according to:

$$E^{°}=E_{\text{oxidized }}-E_{\text{reduced }}-E_{\text{abs }}^{°}(\text{ reference }),$$ (3)

where E_oxidized and E_reduced are the potential energies of oxidized and reduced states of the solute/enzyme (i.e., without taking into account the thermal enthalpic and entropic contributions), and E°_abs(reference) is the absolute potential of a reference electrode. Herein, we adopted the values of E°_abs(reference) = 4.34 eV for SHE in water⁷⁰ and 5.04 eV for ferrocenium/ferrocene couple in acetonitrile⁷¹ (note that to be consistent with the level of calculations, these reference values do not include the ionization threshold energy of a free electron, i.e., neglecting the electron entropy).⁷²

As for enzymatic intermediates, the potential energies were obtained at the equilibrium geometries of the oxidized and reduced states at the QM(B3LYP*-D3)/MM level of theory as indicated above. The small binuclear iron complexes were correspondingly optimized using TPSS-D3/def2-SV(P) method with the COSMO solvation model⁷³ accounting for solvation in acetonitrile, and the energies were then obtained from single-point calculations using B3LYP*-D3/def2-TZVP with the COSMO-RS solvation model.^74,75

2.5. Sequence Analysis.

Sequences of different plant, archaeal, and bacterial desaturases were obtained by the NCBI BLAST⁷⁶ search, performed by employing the amino acid sequence of the Ricinus communis Δ⁹D, i.e., the one used in the QM/MM study. Randomly chosen sequences from different genera with annotated ACP-desaturase with decreasing sequence identity to Ricinus communis (92 – 26 %) were considered. Multiple sequence alignment was performed using the Clustal Omega algorithm.^77,78 Sequences and structures of different NHFE₂ enzymes were taken from the following crystal structures (PDB IDs): 1MHY (soluble methane monooxygenase; sMMO),⁷⁹ 3PM5 (benzoyl coenzyme A epoxidase; BoxB),⁸⁰ 1CHT (p-aminobenzoate N-oxygenase; AurF),⁸¹ 3DHH (toluene 4-monooxygenase; T4MO),⁸² 1RYT (rubrerythrin; Rbr),⁸³ and 1MXR (ribonucleotide reductase; RNR)⁸⁴. Structure-driven sequence alignment was performed using the PROMALS3D algorithm.⁸⁵

3. RESULTS AND DISCUSSION

3.1. Plausible Activation Pathways of the P Intermediate.

The P intermediate is best described as the 1,2-μ-peroxo-[Fe^III]₂ complex (Figure 3; left) with a (6C,6C) or (5C,6C) coordination.¹⁸ The (6C,6C) model is energetically favored, while the (5C,6C) model with a monodentate binding mode of Glu196 is more consistent with available spectroscopic data.¹⁸ The direct attack on the substrate’s C₁₀–H bond by P is, however, unlikely, as the lowest TS barrier is calculated to be ~25 kcal mol⁻¹, with respect to the P intermediate (see Table S1). This is in disagreement with the barrier (activation energy) of ~15 kcal mol⁻¹, derived from kinetic measurements.⁸⁶ Thus, we explore the possibility of three alternative pathways for P activation, denoted as 1a, 1b, and 1c in Figure 3. Pathway 1a involves protonation of the peroxo unit, leading to a μ-η²:η¹-hydroperoxo-[Fe^III]₂ species (denoted as P’) that was already proposed in ref. 19; 1b involves the O–O bond cleavage to yield the bis-μ-oxo-[Fe^IV]₂ complex, which is thought to exist as intermediate Q in the sMMO enzyme,^28,29,30 and 1c corresponds to the P activation by concomitant H⁺/e⁻ transfers into the active site.

Figure 3.

The spectroscopically observed 1,2-μ-peroxo-[Fe^III]₂ intermediate P is inactive toward the C–H bond cleavage. Thus, several activation pathways are proposed: (i) the proton-assisted rearrangement producing the μ-η²:η¹-hydroperoxo-[Fe^III]₂ species P’ (1a), (ii) the O–O bond cleavage to yield the bis-μ-oxo-[Fe^IV]₂ complex putatively corresponding to the intermediate Q in sMMO (1b), and (iii) concomitant proton and electron transfers to yield the complex, which would be reminiscent of the intermediate X of the Class Ia RNR enzyme (1c). The energies are calculated at the QM(B3LYP*-D3)/MM level of theory and referenced to the energy of the P intermediate.

3.2. Proton Transfer (PT)-Assisted Activation of the P Intermediate.

Proton transfer from the second-shell His203 residue to the Fe_A-terminal Glu105 ligand triggers a considerable change in the coordination sphere of the Fe_A center, opening an energetically accessible protonation of the bridging peroxo group (Figure 4). This results in the formation of the μ-η²:η¹-hydroperoxo-[Fe^III]₂ intermediate P’, which is ~13 kcal mol⁻¹ above P. To probe reactivity of P’ with respect to HAA from the C₁₀–H bond, we performed a relaxed scan along the O_A…H¹ coordinate yielding a ‘TS’ that is depicted in Figure 5, center (note that key geometrical parameters for reaction intermediates of all reaction pathways are illustrated in Figure S3). At this TS, the O–O distance is 0.22 Å longer than in P’, while spin density remains essentially unchanged, indicating no electron transfer from the substrate to the diiron center (cf. pathway 1a in Table S2).

Figure 4.

A possible pathway for the protonation of the active site of the P intermediate. The energies are calculated at the QM(B3LYP*-D3)/MM level of theory and referenced to the energy of the P intermediate. Note that 1a₀ structure is already protonated at the His203 residue and, therefore, does not correspond to the referential P intermediate from Figure 3.

Figure 5.

A different character of the TS for C₁₀–H bond cleavage is observed for the DFT calculations (heterolytic cleavage; Left) and multiconfigurational WFT-based methods (homolytic cleavage; Right).

The QM(B3LYP*-D3)/MM activation energy of ~33 kcal mol⁻¹ (referenced to the P intermediate), seems to be significantly overestimated as compared to the barriers obtained using multiconfigurational methods. According to the CASPT2- and MC-PDFT-based QM/MM calculations (with the active space of 14e,14o)^†, the barrier along the reaction coordinate from the P to the TS for C₁₀–H bond cleavage is predicted to lie between 21 and 25 kcal mol⁻¹. However, this is still too high in the comparison to the experimentally-derived energy of activation of ~15 kcal mol⁻¹.⁸⁶ We note in passing that C₁₀–H bond activation is inaccessible also from the 1,2-μ-hydroperoxo-[Fe^III]₂ intermediate 1a₂ (depicted in Figure 4) as the barrier exceeds 30 kcal mol⁻¹.

It is also notable that DFT and the multiconfigurational methods differ in the description of the mechanism of the C₁₀–H bond cleavage (Figure 5). DFT favors heterolytic cleavage producing a carbocationic substrate intermediate, whereas WFT methods prefer homolytic cleavage, as discussed already in ref. 19. We have investigated this aspect in more detail in this study and found that near the P’/TS structure (representing the ‘bifurcation’ point), MC-PDFT and MRCI⁸⁷ methods predict the carbocationic electronic structure as high-lying excited state (favoring carboradical by tens of kcal.mol⁻¹), in contrast to most DFT functionals.

While the ‘absolute’ accuracy of computed values for the NHFe/NHFe₂ sites can be a matter of debate, it is clear that the mechanism of the first substrate’s C–H bond activation is critical for enzymatic selectivity. When the carbocation is formed, it becomes difficult, if not impossible, to avoid the energetically most favorable local minimum, which is a hydroxylated alkyl chain. The subsequent dehydration process that would yield the final desaturated product is not energetically feasible (the process is associated with a calculated activation barrier of > 65 kcal mol⁻¹). Thus, the C₁₀–H bond heterolysis seems to represent a ‘thermodynamic trap’, suggesting that the enzyme following the 1a pathway would have to undergo a carboradical formation instead to yield a product of desaturation. Following the ‘substrate-radical’ mechanism, desaturation is accomplished by the barrierless cleavage of the second C₉–H bond, providing the oleic acid as a final product.

3.3. Splitting the O–O Bond in the P intermediate.

The so-called Q intermediate is a key reactive species in the catalytic cycle of the binuclear non-heme iron soluble methane monooxygenase (sMMO).^29,30,88 It has been postulated to adopt the Fe₂O₂ diamond core geometry with two antiferromagnetically coupled high-spin (S = 2) Fe^IV centers bridged by two oxo groups, yielding a total spin state of S = 0 (note that also other geometries of Q were recently proposed in the literature)⁸⁹. Herein, an analogous Q-like diamond core was investigated employing our QM/MM model of Δ⁹D. At the QM(B3LYP*-D3)/MM level, the Δ⁹D Q is ~19 kcal mol⁻¹ higher in energy than the P intermediate. Nevertheless, a somewhat smaller energy barrier of ~10 and 12–19 kcal mol⁻¹ was obtained from the CASPT2- and MC-PDFT-based QM/MM calculations, respectively (for MC-PDFT, the range of energies reflects usage of six different functionals on top of the (20e,16o)^‡ CASSCF reference wave function). We note in passing that understabilization of the Q-like intermediate at the B3LYP level was already recognized in the computational studies of the sMMO⁹⁰ as well as the BoxB²⁷ enzymes.

With respect to C₁₀–H bond activation by Δ⁹D Q (Figure 6), the TS character is very different from that calculated for P’. Compared to P’, the TS is considerably later as evidenced by: (i) a shorter O…H¹ bond length (1.31 vs. 1.20 Å for P’/TS¹ vs. Q’/TS¹), and (ii) a spin density at the C₁₀ site indicating that carboradical character was already developed at the TS (cf. Table S2). This suggests a ‘carboradical’ channel is operative, effectively preventing the hydroxylation of the natural substrate.

Figure 6.

Activation of the C₁₀–H bond by the Q intermediate. The energies are calculated at the QM(B3LYP*-D3)/MM level of theory and referenced to the energy of the P intermediate.

However, the barrier for C₁₀–H bond activation is exceedingly high (29, 35–40, and 23 kcal mol⁻¹ relative to P as calculated using QM/MM scheme with the respective QM level of theory: B3LYP*-D3, MC-PDFT, and CASPT2^§). Comparing the lowest estimate of the barrier as obtained at the QM(CASPT2)/MM level of theory with the experimentally inferred value of ~15 kcal mol⁻¹, we do not consider the Q-involved 1b pathway to be mechanistically relevant in the Δ⁹D with the native substrate.

3.4. Proton-Electron Transfer (PET)-Assisted Activation.

3.4.1. Energetics.

The concurrent protonation and one-electron reduction of the active site is a frequently adopted mechanism for activation of enzymes for catalysis, thus avoiding very ‘high-energy’ intermediates.^91,92,93,94 Herein, we explore the possibility of 1e⁻ reduction of the active site upon its protonation by evaluating reduction potentials of all key reaction intermediates along the 1a coordinate (Table 1). As can be seen in Table 1, the reduction potential (E°) of a non-protonated P intermediate is estimated to be very low (ca. −3.5 V vs. SHE), while all PT-activated forms have much less negative E° (ca. +3 V with respect to the E° of P). This suggests that the activation of P by electron transfer (ET) into the active site or by stepwise electron / proton transfers (ET/PT) is unlikely. Instead, either concerted proton-coupled electron transfer (PCET) or stepwise proton transfer / electron transfer (PT/ET) mechanism can be considered as a possible activation process. For calibration, we calculated the E°’s of three synthetic binuclear oxo-bridged Fe^IVFe^IV/Fe^IVFe^III redox couples using the same methodology as for Δ⁹D, which are presented along with the intermediates from the 1a pathway in Table 1. From this calibration, the accuracy of calculated reduction potentials in Δ⁹D is within ~0.5 V.

Table 1.

Calculated reduction potentials of intermediates from the pathway 1a from Figure 3, X intermediate from Figure 7, and synthetic binuclear complexes from refs. 96 and 97.

Calcd. Eo [V] vs. SHE
P	−3.48
1a1	−0.75
P’-(5C,6C)	−0.31
P’-(6C,6C)	−0.44
X	−3.00
	Calcd. Eo [V] vs. Fc+/0	Expt. Eo [V]	Ref.
[FeIV2O(L)2]2+/[FeIVFeIIIO(L)2]2+	1.21	1.50	complex 1 in ref. 96
[FeIV2(μ-O)2(La)2]4+/[FeIVFeIII (μ-O)2(La)2]3+	0.55	0.90	complex 2a in ref. 97
[FeIV2(μ-O)2(Lb)2]4+/[FeIVFeIII (μ-O)2(Lb)2]3+	0.31	0.76	complex 2b in ref. 97

SHE = standard hydrogen electrode (referenced in water); Fc^+/0 = ferrocenium/ferrocene couple (referenced in MeCN)

Ferredoxin (Fd) is the most probable source of an electron for the possible reduction of the protonated active site, as it is already involved in the regeneration of the Δ⁹D after completing the catalytic cycle for desaturation of the oleic acid. Injection of one electron into P’ from Fd should be feasible without a significant thermodynamic penalty as the E° of P’ (−0.31 V in Table 1) falls close to the range of E°’s of various plant-type [2Fe-2S] Fds (between −0.39 and −0.43 V vs. SHE at 25 °C).⁹⁵ Importantly, ET to P’ triggers spontaneous O–O bond cleavage, yielding a mixed-valence Fe^IIIFe^IV intermediate, which is analogous to the intermediate X in RNR (1c pathway; see Figure 3 and Figure 7). Importantly, this Δ⁹D ‘X’ is estimated to be only ~4 kcal mol⁻¹ above P and its computed reduction potential of −3.0 V is also comparable to that of P (c.f., Table 1). Finally, the attack of C₁₀–H bond by X is energetically accessible, with the barrier of 13, 14 or 15–17 kcal mol⁻¹ as calculated by the QM(B3LYP*-D3)/MM, QM(CASPT2(15e,14o))/MM and QM(MC-PDFT(15e,14o))/MM approach, respectively. All these values are in agreement with the experimentally derived barrier of 14.8 kcal mol⁻¹.⁸⁶ As for the second substrate’s C₉–H bond cleavage, which leads to the final desaturated product, the calculated barrier is significantly lower (~7 kcal mol⁻¹ at the QM(B3LYP*-D3)/MM level). Thus, we consider PET-assisted activation of P (1c pathway in Figure 3), with two subsequent H-atom abstractions from the C₁₀ and C₉ sites of the substrate as the most viable catalytic mode of action in Δ⁹D.

Figure 7.

Plausible PET-assisted activation (stepwise PT/ET mechanism) of the P intermediate. Energies are calculated at the QM(B3LYP*-D3)/MM level of theory and referenced to the P intermediate. Note that coupled (concerted) proton-electron transfer might be a possible mechanism, bypassing the ‘high-energy’ P’ intermediate and converting P to X directly.

3.4.2. Desaturation vs. Hydroxylation Selectivity.

In contrast to P’, which is predicted by DFT to abstract hydride from the substrate C₁₀ site that leads to carbocation (C₁₀⁺) and hence hydroxylation trap, the X of the pathway 1c has already one extra electron and thus it has a lower oxidation power allowing only for HAA that produces carboradical-containing intermediate (X/int₁ in Table 2). It is noteworthy that this intermediate is quite flexible and at least four other structural variants lie within ~7 kcal mol⁻¹ (DFT) and therefore may be energetically feasible under turnover conditions. This flexibility is given by a variability in the coordination of the Glu105 and Glu196 residues and the orientation of the O–H bonds in the hydroxo ligands (Table 2).

Table 2.

Possible structures of the carboradical-containing intermediates in pathway 1c that are formed following the HAA step from C₁₀ of the substrate. Energies are calculated using QM(B3LYP*-D3)/MM and QM(MC-PDFT)/MM methods and referenced to the energy of the P intermediate. The range of energies for MC-PDFT method reflects usage of six different functionals. The structures are oriented so that the Glu105 and Glu196 residues are on the left and right side, respectively.

		QM(B3LYP*-D3)/MM	QM(MC-PDFT)/MM
X/int1		−4.3 kcal mol−1	2.6–5.5 kcal mol1
X/int2		−1.8 kcal mol−1	6.8–11.7 kcal mo−−1
X/int3		−0.3 kcal mol−1	0.1–2.8 kcal mol−1
X/int4		1.7 kcal mol−1	−1.1–3.0 kcal mol−1
X/int5		2.9 kcal mol−1	3.2–8.4 kcal mol−1

Despite the presence of C₁₀• (instead of C₁₀⁺) in the intermediate X/int₁ (Table 2), the OH rebound step is calculated to be nearly barrierless (at the QM(B3LYP*-D3)/MM level of theory), with the ‘C₁₀•-to-Fe_B’ ET followed by the recombination of the nascent cation C₁₀⁺ with the Fe_B-bound OH⁻ ligand; the latter is strongly reminiscent of the pathway 1a from Figures 3 and 5. From the literature, the DFT barrier for the enzymatic mononuclear non-heme iron OH rebound is usually calculated to be in the range of 5–15 kcal mol⁻¹.^{98,99,100,101} In this view, the QM/MM model (with the QM represented by DFT or a multireference wave-function calculation at the QM(DFT)/MM equilibrium geometries) seems to underestimate the barrier for hydroxylation from the X intermediate. As in pathway 1a, this is probably due to the considerable multireference character of the electronic structure of the binuclear non-heme iron species that cannot be accounted for by DFT and/or the inappropriate description of structural nuances preventing the “barrierless” hydroxylation. The significant discrepancy between the DFT and MC-PDFT energies of the five different carboradical-containing conformers (Table 2) precludes insights into how the hydroxylation trap is avoided in Δ⁹D.

The failure to describe properly the Δ⁹D chemo-selectivity by the QM/MM model reflects the fine-tuned ‘balance’ between desaturation and hydroxylation pathways (from numbers presented herein, we estimate the difference in the activation energies for the hydroxylation/desaturation to be not greater than ~2 kcal.mol⁻¹). This is also evidenced by the observed oxygenation (hydroxylation) of some non-native substrates by wt and mutated forms of Δ⁹D.^15,102,103 Here, we speculate that the role of non-native substrates such as fluorinated fatty acids and ether/thioether analogs of fatty acids is to promote the facile formation of the carbocationic intermediate that would effectively open the hydroxylation/oxygenation channels.

3.5. Proton and Electron Pathways Available for PET-assisted Activation of P; Comparison with Other Enzymes

3.5.1. Proton Pathways.

The activation of P by its protonation was proposed in some other NHFe₂ enzymes, leading to a more reactive hydroperoxo species P’.² For instance, the involvement of 1,2-μ-hydroperoxo-[Fe^III]₂ P’ was demonstrated to lower the kinetic barrier for the oxidation of 4-aminobenzoic acid in AurF by 8 kcal mol⁻¹ due to the stabilization of a high-valent product of O–O bond cleavage.¹⁰⁴Alternatively, the formation of P’ in RNR was also suggested to increase the reduction potential for further electron-transfer activation, resulting in a mixed-valence intermediate X (vide infra).²¹

It is worth mentioning that also the structure of the suggested Δ⁹D P’, i.e., the μ-η²:η¹-hydroperoxo-[Fe^III]₂ (or energetically accessible 1,1-μ-hydroperoxo-[Fe^III]₂), is not unprecedented. The 1,1-μ-hydroperoxo P’ intermediate was captured by X-ray crystallography in T4MO Gln228Ala mutant.¹⁰⁵ In T4MO, it was, however, proposed to be part of a non-productive pathway due to the overlay of the 1,1-μ-hydroperoxide with the C3 position of bound toluene substrate. Still, an observation of 1,1-μ-hydroperoxo species in the T4MO active site gives significant credence to our suggested model of the protonated P’ intermediate in Δ⁹D. Finally, it is also noticeable that the non-protonated μ-η²:η¹-peroxo intermediate P in CmlI was suggested as the reactive species based on resonance Raman spectroscopy,¹⁰⁶ and the 1,1-μ-peroxo P based on X-ray absorption spectroscopy¹⁰⁷.

In our previous study of Δ⁹D, the PT activation of P was already suggested and a role of the second-shell His203 residue in the proton transfer chain via protonating the coordinated Glu105 was speculated.¹⁹ From the detailed bioinformatics analysis provided herein (Figure S4), we observe that His203 residue is conserved in plant as well as bacterial desaturases. In fact, two proton chains may thus connect the active site to the bulk solvent: (i) the previously proposed His203 – Asp101 – Thr206 – Cys222,¹⁸ or (ii) currently proposed His203 – Asp101 – Ser202 – Asp296 (see Figure S5). Although the Thr206 and Cys222 are often replaced by hydrophobic residues and Asp101 by Asn in many bacteria, the polar character of the latter proton chain is well conserved in plant and bacterial sequences and may facilitate proton transfer. If this is the case, similar proton chains should be present in other NHFe₂ enzymes that are known to require proton transfer to activate P. Indeed, such polar chains can be located in e.g., T4MO and AurF (c.f., Figures S6 and S7). However, the polar residues are present in a different position suggesting that Nature has evolved a number of different ways to connect Fe-coordinated glutamate to the H-bonded network ending in the bulk solvent. In contrast, the proton transfer chain is replaced with apolar residues in sMMO and BoxB, for which a solvent-based protonation of the dioxygen cofactor is not required.

In addition, the role of Thr201 in toluene/o-xylene monooxygenase hydroxylase (ToMOH) was suggested to be essential for facilitating the proton transfer to the peroxo moiety in P and/or to stabilize the resulting reactive oxygenated intermediate.¹⁰⁸ The equivalent residue, Thr199, is also present in Δ⁹D (c.f., Figure S8 for sequence alignment of various NHFe₂ enzymes) and is conserved among all plant and bacterial desaturases. By mutating Thr199 for glutamic or aspartic acid, Shanklin and coworkers demonstrated a central role of Thr199 in the catalytic cycle due to the resultant decrease in desaturase activity (mutants reported peroxidase activity instead – in analogy to Rbr enzyme, where the same position is occupied by glutamic acid).¹⁰⁹ Taken together, conserved Thr199 may facilitate protonation of P by the coordinated Glu105 via the conserved His203, which is connected to bulk solvent via a proton transfer chain.

3.5.2. Electron Pathways.

Although Fd is assumed to be the electron donor required for the pathway 1c, there might also exist other sources of electrons available for the reduction of the enzyme active site. The negatively-charged Fd was demonstrated to bind near Lys56, Lys60, and Lys230, so that the electrons are shuttled to the active site through the conserved Trp62 residue (c.f., Figure S8 for sequence alignment, Figure S9 for Trp62 position relative to the active site).¹¹⁰ On the other hand, the lysine residues are conserved only among plant desaturases, suggesting also the viability of an alternative electron-transfer chain. This may include, Trp132 – Trp135 – Trp139 – Tyr236 on the other side of the protein (Figure S9). Since counterparts of Trp135 and Trp139 were detected as radical donors upon mutation of the physiologically-relevant tyrosine to phenylalanine in RNR,¹¹¹ these residues may indeed serve as a viable option for an alternative electron source for diiron site reduction in Δ⁹D or as a hole scavenger. However, there is no supporting experimental evidence that would clarify the role of this tryptophan chain. From the comparison to other NHFe₂ enzymes, the activation of P’ by ET is appealing owing to the fact that only RNR (which requires ET to activate its P’ intermediate) contains some of aromatic residues, equivalent to those of Δ⁹D, capable of electron transfer.

We note in passing that also other similarities to RNR were found in our calculations. In RNR, the transfer of proton to 1,2-μ-peroxo diferric intermediate P is assumed to be responsible for stabilization of the peroxo σ* orbital, which upon electron activation leads to a reductive O–O bond cleavage, generating an intermediate X.^21,112 Interestingly, such protonation was suggested to stabilize the peroxo σ* LUMO by as much as 2.38 eV.²¹ In Δ⁹D, we observed the same electron transfer-triggered O–O bond rupture that is associated with the increase in reduction potential upon protonation (in going from P to P’, the reduction potential increases by more than 3 V; cf. Table 1).

4. CONCLUSIONS

Employing the QM(DFT)/MM and multireference wave function methods, the catalytic mechanism of the non-heme binuclear Δ⁹ desaturase has been explored. Calculations suggest that the spectroscopically-observed P intermediate is not capable of the direct attack of the substrate (black in Figure 8). Therefore, we investigated three possible activation pathways of P and the following reaction steps leading to the desaturated product. The proton-electron uptake by the P intermediate (pathway 1c; in green in Figure 8) is the favored mechanism for catalytic activation since it allows a significant reduction of the barrier of the initial (and rate-determining) HAA from the stearoyl substrate as compared to the ‘proton-only activated’ pathway 1a and the O--O bond cleavage to generate a “Q” pathway 1b (in red and blue in Figure 8, respectively). This mechanistic picture is consistent with experimental data available for Δ⁹D and satisfy fairly stringent conditions required by Nature: the chemo-, stereo- and regioselectivity of the desaturation of stearic acid. Finally, all three studied mechanisms are placed into a broader context of NHFe₂ chemistry provided by the amino acid sequence analysis through the families of the NHFe₂ enzymes. In summary, this study represents an important advance in our understanding of the catalytic action of the NHFe₂ enzymes and should inspire further work in the NHFe₍₂₎ biomimetic chemistry.

Figure 8.

The energetics associated with all of the studied pathways, calculated at the QM(B3LYP*-D3)/MM level of theory. Raw data are shown in Table S3. Black: A direct C₁₀–H bond activation by P intermediate. First, the O–O bond of the P intermediate is cleaved with a calculated barrier of ~25 kcal mol⁻¹ (the rate-determining step), followed by a C₁₀–H bond activation (~22 kcal mol⁻¹). Red: The proton-assisted activation pathway (1a). The protonation of P generates a P’ intermediate (12.8 kcal mol⁻¹ above P), which may activate C₁₀–H bond of a substrate. The barrier of ~33 kcal mol⁻¹ is, however, too high to be operative. Blue: Cleavage of the peroxo O–O bond (pathway 1b). The ‘Q’ like [Fe^IV]₂O₂ intermediate is suggested to be inactive due to the rate-determining barrier of C₁₀–H bond activation of ~29 kcal mol⁻¹. Green: The activation of P by proton-electron uptake (1c). Here, either concerted or step-wise proton-electron transfer (dashed or solid line) is proposed, leading to a formation of an intermediate X analog (~4 kcal mol⁻¹ above P). The energy of the rate-determining step within the 1c pathway (i.e., the C₁₀–H bond cleavage) is consistent with the barrier derived from the experimental data (calcd. vs. expt.: ~13 kcal mol⁻¹ vs. 14.8 kcal mol⁻¹). The second C₉–H bond is then activated with a small activation energy of ~7 kcal mol⁻¹.