The heme oxygenase-like diiron enzyme HrmI reveals altered regulatory mechanisms for dioxygen activation and substrate N-oxygenation

Abstract

Nonheme diiron enzymes activate dioxygen (O₂) to affect various biochemical outcomes. HrmI, a member of the recently discovered and functionally versatile heme oxygenase-like dimetal oxidase/oxygenase (HDO) superfamily, catalyzes the N-oxygenation of L-Lysine to yield 6-nitronorleucine for the biosynthesis of the antibiotic hormaomycin. Unlike other characterized HDO N-oxygenases that have an additional carboxylate ligand thought to be key for regulating dioxygen activation and ensuing N-oxygenation, the predicted primary coordination sphere of HrmI resembles those of HDOs that instead perform C-C fragmentation of substrates. We show that diferrous HrmI reacts with O₂ in a substrate-independent manner to form a presumptive μ−1,2 (Fe³⁺)₂ peroxo (or P) intermediate common to the catalytic scheme of many HDOs. P is rapidly converted to a second species with both optical and Mössbauer properties that resemble an activated peroxodiferric adduct (P’). The substrate-dependent acceleration of P’ decay suggests that it, rather than P, initiates L-Lysine metabolism. X-ray crystallographic studies of HrmI in several redox and ligand-bound states provide a stepwise view of structural changes during catalysis, and together with analytical approaches, capture a hydroxylamino metabolic intermediate en route to 6-nitronorleucine formation. The activation of peroxo species provides a key strategy that enables functional adaptation within the widely distributed HDO structural scaffold.

Graphical Abstract

INTRODUCTION

Enzymes that contain dimetal clusters activate dioxygen for a panoply of functions that are central to human health and biotechnology (e.g. the synthesis of deoxyribonucleotides,¹ monounsaturated fatty acids,² and the modification of tRNAs³ or transcription factors⁴). Apart from a few notable exceptions where a mixed-valent form (Fe₂^3+/2+) is used as a reactant,^5–6 most diiron enzymes employ a cluster in the fully-reduced (diferrous or (Fe²⁺)₂) state to bind and activate atmospheric dioxygen.⁷ As a result, O₂ is transformed into a diverse set of iron-oxygen (Fe-O) adducts that can have variable and often potent oxidative capabilities.⁷ The vast structural and electronic landscape of diiron-oxygen intermediates, and the expansive substrate scope for the proteins that house them, has positioned diiron enzymes as valuable catalysts for sustainable energy and drug development.

HrmI from S. griseoflavus, a member of the recently identified heme oxygenase-like dimetal oxygenase (HDO) superfamily, catalyzes the iron and O₂-dependent 6-electron oxidation of L-Lysine to form 6-nitronorleucine (Figure 1A).^8–10 The product is a precursor for hormaomycins that serve as signaling peptides for Streptomyces differentiation and display narrow-spectrum antibiotic properties.^11–12 Despite functional verification as an N-oxygenase, a primary sequence alignment of HrmI with other HDOs suggests that it contains a 3-histidine/3-carboxylate (3H/3C) primary ligand set that is prototypical of HDOs that perform C-C cleavage rather than N-oxygenation (Figure 1B). This contradiction positions HrmI as a useful system to evaluate how subtle factors beyond the first coordination sphere enable drastically different chemical outcomes within a shared structural scaffold.

Figure 1:

N-oxygenase function and predicted cluster ligation of HrmI. (A) HrmI catalyzes the N-oxygenation of L-lysine in the hormaomycin biosynthetic pathway in S. griseoflavus. (B) Primary sequence alignment of HDOs shows that HrmI is predicted to contain a 3-histidine, 3-carboxylate (3H/3C) ligand set that is more commonly attributed to C-C fragmenting HDOs. The “extra” carboxylate found in SznF and N-oxygenases such as RohS is denoted by an arrow.

The nuanced regulation of diiron cofactors has been most studied within the context of ferritin-like diiron oxidase/oxygenases (FDOs). In FDOs, the binuclear iron cofactor is housed within a four alpha-helical (α₄) core structure that provides the carboxylate and histidine ligands to the cluster. Despite sharing a common core, FDOs use vastly different overall architectures, auxiliary protein component(s), and cluster ligation strategies. These differences, along with the technical challenges associated with many FDOs (e.g. membrane association, structural regulation by protein effectors, and carrier protein-linked substrates), complicate efforts to parse general paradigms of diiron catalysis. Initially, it was posited that variable ratios of primary ligands (histidine versus carboxylate) underscored the differences in oxidative capabilities of FDOs, with the latter ligand promoting the formation of highly oxidizing ferryl intermediates. However, recent structures have revealed histidine-rich clusters in FDOs that perform very challenging oxidations on inert hydrocarbons (e.g. desaturation¹³ and hydroxylation^14–15), suggesting that more subtle structural factors are involved in tuning function.

Only a small fraction of HDOs from this emerging class have been functionally characterized, but they are beginning to reveal a chemical versatility that matches, and possibly outrivals, their FDO counterparts.¹⁶ HDOs have been shown to catalyze various types of C-C cleavage,^17–27 C-O bond installation,²⁴ N-oxygenation,^{8, 28–32} radical transfer,³³ and many other complex oxidative rearrangements.^{17, 23 27} Despite this functional diversity, most HDOs share a 3H/3C ligand-set. One notable exception is the N-oxygenase SznF which possesses a fourth carboxylate ligand (3H/4C) that has been proposed to enable substrate-independent O₂ activation and possibly promote N-oxygenation.³⁴ RohS, an N-oxygenase, is also predicted to contain an SznF-like 3H/4C primary ligation based on sequence homology (Figure 1B). The exact role of this fourth carboxylate remains unclear, but the fact that HrmI appears to perform N-oxygenation with only three carboxylate ligands presents an opportunity to determine how O₂ activation is perturbed by factors other than the primary coordination to support N-oxygenation.

Although some exceptions have recently been found,^{19, 35} most characterized HDOs use a diferrous (Fe²⁺)₂ cofactor to activate dioxygen to form a shared intermediate with optical and Mössbauer spectroscopic properties that are commonly associated with a μ−1,2 (Fe³⁺)₂ peroxo (or P) species. The P species characterized in HDOs orchestrate substrate C-C scission (e.g. BesC,^{20, 22} UndA,²⁶ SktA,²⁷ and AetD¹⁷) or N-oxygenation (e.g. SznF)³⁶. In the case of C-C fragmentation of 4-Cl-L-Lysine by BesC, the rate of P decomposition is dependent on substrate-type (e.g. substituent or isotopologue at the targeted C₄ position).^{20, 22} The N-oxygenase SznF demonstrates an acceleration of P decay upon the addition of substrate(s).³⁶ Collectively, these observations have been used as primary evidence that HDO-P species either directly react with their respective substrates, or transform to an as-of-yet undetected intermediate(s) that affords substrate metabolism. However, the corresponding FDO μ−1,2-P species are generally considered as relatively inert oxidants and are not typically implicated in reactions with substrates.⁷ Rather, transformation to a more activated type of peroxo species (denoted here as P’) or high-valent (e.g. (Fe⁴⁺)₂ or Q)³⁷ adduct is usually needed to promote catalysis.

Unlike FDOs that can sometimes use protein-tethered substrates or effectors to regulate O₂ activation,^38–40 HDOs broadly fall under two categories depending on the apparent regulatory role of the substrate. Some HDOs (e.g. UndA,^{26, 41} BesC,²⁰ AetD,¹⁷and SktA²⁷) require a substrate to trigger the formation of P while others (e.g. SznF) do not require this step.³⁶ Structural studies have suggested that one operative strategy may (at least in part) be impacted by ligation-changes of the cluster. For example, structures of the 3H/3C substrate-triggered HDOs AetD and UndA, the latter with a partially occupied cofactor, show that their corresponding amino acid or fatty acid substrates can directly ligate the metallocluster.^17–18 While the provision of an “additional” ligand (via the substrate or the protein) would appear to satisfactorily rationalize the gating of O₂ activation in HDOs, the regulatory factors that give rise to P and how a common intermediate can mediate such different types of transformations are poorly understood.

Here, we have employed X-ray crystallography, isotopic tracing mass spectrometry, and transient kinetics to evaluate the structure and catalytic mechanism of HrmI. Fortuitous stabilization of the diiron cofactor has enabled the crystallographic characterization of HrmI in several oxidation and ligand-bound states, providing the most comprehensive structural view of HDO catalysis to date. Structural analyses confirm that HrmI contains a 3H/3C ligand-set, and although the substrate weakly coordinates the (Fe²⁺)₂ cluster, transient kinetics studies demonstrate that HrmI effectively activates O₂ without the provision of a substrate. The reaction of (Fe²⁺)₂ HrmI with O₂ results in the appearance of two spectroscopically distinct and kinetically separated transient intermediates. The optical properties of the first species suggest probable assignment as a μ−1,2 P adduct thought to be common to the catalytic schemes of HDOs. However, the decay kinetics of HrmI-P are completely unperturbed by the substrate, suggesting that it does not directly initiate substrate N-oxygenation. Instead, P decay is accompanied by the formation of a second diferric species that has atypical spectroscopic (optical and Mössbauer) properties that instead resemble an activated P’ adduct. The substantial acceleration of P’ decay with L-Lysine implicates it as the reagent that likely initiates N-oxygenation. These findings illustrate that factors other than the primary coordination sphere contribute to the formation and fate of HDO-P and highlight the growing complexity of factors that govern O₂ activation strategies and downstream catalysis in HDOs.

RESULTS

Structural studies reveal that HrmI contains a 3H/ 3C Diiron cluster

Obtaining holo-structures of HDOs, particularly in the oxidized diferric (Fe³⁺)₂ state, has proven elusive due to incomplete metal loading and/or instability of the cluster in aerobic conditions.^{20, 22, 25–26, 31, 34} AetD, solved as (Fe²⁺)₂, represents the only available substrate-bound structure of an HDO with all of the primary ligand residues resolved and substantial cofactor occupancy.^17–18 Collectively, this has limited understanding the structural changes that need to occur during O₂ activation and the contribution of substrate positioning on steering different metabolic trajectories. Following affinity purification (described in SI Methods), HrmI fortuitously retains ~1.1 equivalents of iron which corresponds to ~55 % cofactor loading, assuming a diiron center (Figure S1, Table S1). Mössbauer spectroscopy shows that the Fe cofactor is in a magnetically coupled diferric form, exhibiting two distinct Mössbauer quadrupole doublets (Figure S2, Table S2). This observation suggests that the two iron centers in this diferric-resting state contain slightly different ligand environments. To confirm the ligation strategy and other underlying factors that may enable N-oxygenation, we first determined the X-ray crystal structure of the protein in the partially loaded form to a resolution of ~2.0 Å (PDB: 9N1E, Table S3). The overall structure of HrmI consists of two alpha-helical domains. The smaller N-terminal domain participates in homodimerization (Figures 2A, S3) and is similar to SznF.^{31, 34} The C-terminal HDO domain consists of a bundle of seven alpha helices, three of which house the six residues that are involved in cofactor binding. The occupancy of Fe in this structure is approximately 50 % which closely mirrors solution measurements. Single-wavelength anomalous datasets collected at the Fe-edge reveal the presence of both irons with roughly equal scattering intensities at each site (Figure S4). Despite partial occupancy of the cofactor, the secondary structure of core helix α3 is largely ordered. However, the mFo-DFc difference map indicates that for several sidechains, the electron density calculated from the model (DFc) exceeds the experimental density (mFo) (Figure S5, red mesh), suggesting some degree of structural disorder. To improve cofactor loading, the as-purified protein crystals were soaked in Fe²⁺ followed by an aerobic buffer to promote oxidation (SI Methods). The resulting structure was determined at a resolution of 2.13 Å and the Fe occupancy improved to ≥ 80 % (PDB: 9N1A, Table S3). The structure confirms that HrmI contains a 3H/3C set of primary ligands. H204, E194 (core α1) and H288 (core α3) coordinate Fe1, and E258 (core α2), D292, and H295 (core α3) interact with Fe2 (Figure 2B). The Fe-Fe distance (3.7 Å) and presence of a single-atom (aquo, hydroxo or oxo) bridging ligand that interacts with both irons (Fe1-O and Fe2-O distances of 2.0 Å and 2.1 Å respectively) are consistent with assignment as the (Fe³⁺)₂ oxidation state. Additional hydroxo or aquo ligands (one to Fe1, and two to Fe2 respectively) result in a pentacoordinate trigonal bipyramidal geometry for Fe1 and coordinative saturation of Fe2. Notably, (Fe³⁺)₂ HrmI lacks a carboxylate bridge between the two metals, a feature that clearly differs from FDOs in the same oxidation state.⁷ E194 is monodentate with the second carboxylate oxygen residing more than 3 Å from Fe1.

Figure 2:

Crystal Structures of diiron HrmI. (A) The structure of HrmI is primarily alpha-helical, consisting of an N-terminal dimerization domain (gray) and an HDO domain (colored by helices). For simplicity, only one subunit of the homodimer is shown. (B) The structure of substrate-free, diferric (Fe³⁺)₂ HrmI demonstrates that the cluster contains three histidine and three carboxylate ligands. The Fe-Fe distance is 3.7 Å. (C) The structure of substrate-free, diferrous (Fe²⁺)₂ HrmI with a Fe-Fe distance of 4.6 Å. Fe1 is best fit as pentacoordinate, with a trigonal bipyramidal geometry (D) The structure of L-Lysine-bound, (Fe²⁺)₂ HrmI. The zeta amine (N_ζ) of L-Lysine (shown in pink) displaces a hydroxo or aquo ligand and weakly coordinates Fe1. Substrate-binding does not significantly alter the Fe-Fe distance (4.4 Å).

The structure of diferrous (Fe²⁺)₂ HrmI and the substrate ternary complex

For the vast majority of diiron enzymes, the diferrous (Fe²⁺)₂ state of the enzyme is used as a reactant to initiate O₂ activation.⁷ Upon chemical reduction of the cluster and anaerobic soaking of additional Fe²⁺ into crystals, we obtained diferrous structures at a resolution of 2.14 Å (Table S3). In the substrate-free (Fe²⁺)₂ structure (PDB: 9N1X, Figure 2C), the Fe-Fe distance elongates to ~4.6 Å. This distance is slightly smaller than the corresponding Fe-Fe distances of (Fe²⁺)₂ AetD (4.9 Å, PDB: 8JI2⁹) and SznF (5.0 Å, PDB: 6VZY³⁴). As noted previously, the formation of HDO-P would necessitate significant contraction along the Fe-Fe axis, which is ≤ 3.5 Å in most structurally characterized P species from diiron enzymes and biomimetic complexes.⁷ HrmI also lacks the carboxylate bridge in the diferrous state. Instead, a hydroxo or aquo ligand is positioned between the two irons but is too far from Fe1 to form a stable bridge (Fe1-O distance ~3.1 Å). This ligand sits somewhat closer to Fe2, which exhibits a hexacoordinate, octahedral coordination geometry. Fe1 forms five strong interactions with H204, H288, two terminal hydroxo or aquo ligands, and the proximal oxygen of E194. Fe1 is best fit as a pentacoordinate species in trigonal bipyramidal geometry (Table S4). Alternative fits for the cluster were also considered (Figure S6, Table S4). The distal carboxylate oxygen of E194 sits ~2.6 Å away from Fe1, suggesting that a weak bidentate interaction could occur. However, the resulting fit produces a distorted trigonal prismatic geometry for Fe1. Alternative modeling of a sixth ligand originating from the bridging aquo or hydroxo atom provides a more favorable octahedral geometry, but the apparent interaction between Fe1 and the bridging atom is relatively weak (Fe-O distance ~3.1 Å). Overall, the probable pentacoordinate structure of Fe1, or at the very least, the absence of a strongly coordinating sixth ligand, implies that Fe1 is poised to form additional interactions with substrate and/or O₂ during the ensuing steps of the catalytic cycle.

Given the lack of a seventh protein-derived ligand in HrmI to promote O₂ activation, we speculated that O₂-triggering may be enabled through ligation of the substrate. The (Fe²⁺)₂ crystal was soaked with L-Lysine and the structure was determined at a resolution of 2.14 Å (PDB: 9N2A, Table S3). The zeta-amine (N_ζ) of the substrate sidechain displaces a terminal Fe1 hydroxo or aquo ligand and weakly binds to the Fe1 (Fe-N distance of 2.5 Å) (Figure 2D). A similar interaction between the substrate and Fe1 is also observed in structures of AetD^17–18 and UndA²⁶ (Figure S7). The bridging aquo, hydroxo, or oxo now resides 2.4 Å from Fe1, but the Fe-Fe distance (4.4 Å) is nearly unchanged from the substrate-free structure. Fe1 is best modeled as pentacoordinate with a trigonal bipyramidal geometry (Table S4). The alpha amine (N_α) and carboxylate of the substrate are firmly anchored away from the iron by several polar contacts. The side chain of R325 requires fitting with two rotamers, suggesting that it can readily rotate to provide interactions with the substrate carboxylate. D165 and E161 interact with the backbone (N_α) amine, and a water network further stabilizes the binding site through several hydrogen-bond interactions. The C₄, C₅, and N_ζ atoms of the substrate side chain sit < 4 Å from the center of the aromatic ring of F168. Therefore, interactions with F168 (e.g. π-cation or van der Waals) help to anchor the sidechain of the substrate. The positioning of the N_α far away from the cluster provides selective targeting of N_ζ for oxidation. The extensive substrate backbone interactions and length of the substrate binding pocket also rationalize why substrate analogs containing shorter side chains (e.g. L-ornithine) are not effectively metabolized by HrmI nor a closely related ortholog, BelK.^{32, 42}

Substrate-independent O₂ activation results in the formation of two optically distinct intermediates.

Substrate-free diferric (Fe³⁺)₂ HrmI was anaerobically reduced with dithionite (see SI Methods) and rapidly mixed with excess O₂ at 4 °C. A broad absorption with λ_max ~ 600 nm is formed within the dead-time (1 ms) of the instrument and decays rapidly within 50 ms (Figure 3A, red spectrum). The highly transient nature of this intermediate precludes trapping at high concentrations for more detailed spectroscopic interrogation. Nonetheless, the absorption properties are very similar to those of μ−1,2-peroxo (P) species formed by several nonheme diiron enzymes (λ_max ~ 600 – 700 nm), including some HDOs (Tables S5, S6). An examination of the kinetics at 600 nm reveals a single exponential decay process with an apparent rate of k_obs ~ 90 s⁻¹. Neither the accumulation nor decay kinetics of HrmI-P are significantly perturbed by the inclusion of excess L-Lysine in the O₂-containing buffer (Figure 3B, Table S7). This suggests that HrmI-P either serves as a precursor to the operative N-oxygenating intermediate or originates from an off-pathway process.

Figure 3:

Stopped-flow absorption spectroscopy depicting the formation and decay of two transient intermediates, HrmI-P and P’ (λ_max ~ 600 nm and 475 nm, respectively) in the reaction of (Fe²⁺)₂ HrmI with O_{2 at} 4 °C. Concentrations after mixing are 150 μM (HrmI) and 1 mM (O₂). (A) Absorption spectra obtained upon mixing substrate-free (Fe²⁺)₂ HrmI with O₂ (B) Comparison of the decay kinetics of HrmI-P, monitored at 650-nm, in the presence (blue) and absence (red) of 5 mM L-Lysine. Fits to a single decay process are shown in black with residuals shown in corresponding colors. (C) Absorption spectra obtained when mixing (Fe²⁺)₂ HrmI with 1 mM O₂ and 5 mM L-Lysine. (D) Time courses of the formation and decay of P’ monitored at 450 nm in the absence (red) and presence of increasing concentrations of L-Lysine (inset). (E) Dependence of the rate of P’ decay on L-Lysine concentration. Fitting to a hyperbolic expression (dashed line) results in an apparent K_D = 2.6 ± 0.7 mM.

In the absence of L-Lysine, the depletion of HrmI-P is accompanied by the appearance of a broad absorption feature at ~ 475 nm that maximally accumulates at ~2 s (Figure 3A, blue spectrum). The λ_max of this second intermediate is much lower than what is expected for a μ−1,2-P species but is similar to values observed for atypical peroxodiferric intermediates in FDOs (Table S5). Therefore, we refer to this species as HrmI-P’ and provide further evidence for this assignment in the sections to follow. Due to the broad and overlapping absorptions of P’ and P, time courses were collected and analyzed at several different wavelengths that span this absorption region to determine optimal collection strategies for kinetic analysis. A more detailed discussion of the selection criteria used for data collection and non-linear regression analysis is found in SI Methods and Figure S8. Briefly, the low extinction coefficient and spectral overlap of P’ with P leads to complexity when fitting the entire time-regime. However, truncation of the dataset to times after 50 ms, corresponding to when P is expected to be significantly (≥ 95 %) depleted, provided reproducible fitting results. Using this approach, the time course of P’ monitored at 450 nm is accurately modeled by a three-summed exponential equation consisting of one formation (negative amplitude) and two decay (positive amplitude) phases (Figure 3D, Table S8). In the absence of L-Lysine, the apparent rate of P’ formation is ~ 3 s⁻¹, and the intermediate decays in a biphasic manner with apparent rates of ~ 0.006 s⁻¹ and ~ 0.03 s⁻¹ for the two phases. Although the origin of the kinetic complexity for P’ decay is unclear, it could be envisioned to result from multiple steps and/or parallel decomposition pathways.

Unlike P, the lifetime of P’ is significantly affected by L-Lysine (Figure 3D, Table S8). The P’ time course is still best fit with a three-summed exponential equation but now consists of two formations and one decay phase. The first formation phase has an apparent rate (k_form1 ~ 3 s⁻¹) that is very similar to that observed in the absence of substrate. However, the inclusion of a second slower formation phase (k_form2 ~ 0.5 s⁻¹) is now necessary to accurately model the data. An analysis of the fractional amplitude of the two formation phases versus L-Lysine reveals a hyperbolic dependence, with k_form2 becoming the dominant phase at [L-Lys] ≥ 1 mM (Table S9, Fig. S9). The behavior suggests that k_form2 represents a pathway for P’ formation that is favored through substrate binding. However, the most notable change induced by L-Lysine is a dramatic acceleration in the rate of P’ decay, which is ~ 50-fold faster in the presence of 7.5 mM substrate (Figure 3D and Table S8). A plot of the P’ decay rate versus L-Lys concentration reveals a hyperbolic dependence, with an apparent K_D ~ 2.6 ± 0.7 mM (Figure 3E). A similar low-affinity K_D was extrapolated for SznF using the acceleration of P decay.⁴³ In both cases, this suggests that the substrate weakly binds to the ferric-peroxide complex, or there is a lack of equilibration leading to inaccuracy in the measurement. Nonetheless, the hyperbolic dependence suggests that a substrate binding step is involved in regulating P’ decay and further implicates P’ (rather than P) for initiating substrate metabolism.

Sequence and Relationship of HrmI-P and P’

To test whether P and P’ arise from parallel O₂ activation pathways or are formed in series, we examined the formation of each as a function of O₂. For a reaction sequence involving a series of irreversible steps, only the initial product(s) of O₂ binding would be anticipated to result in an O₂-dependent formation rate. To approximate pseudo-first-order kinetics with respect to O₂, the enzyme concentration was lowered, and L-Lysine was omitted. At high (~1 mM) concentrations of O₂, P formation is completed within the 1 ms deadtime. However, a growth phase clearly emerges when O₂ is lowered (Figure 4A). The decay rate of P is unchanged, as expected (Table S9). The accessible range of O₂ to satisfy pseudo-first-order conditions does not allow for full determination of the O₂ concentration dependence (i.e. through a k_obs vs O₂ plot). However, assuming a bimolecular route to P, a second-order rate constant of ~2500 mM⁻¹s⁻¹ is estimated from the available data. If P’ is the initial intermediate of a parallel route, one would anticipate that its formation kinetics should similarly exhibit a dependence on O₂. A (Fe²⁺)₂ HrmI:L-Lysine complex was mixed with varying concentrations of O₂. Neither the formation nor decay of P’ exhibited significant changes (Figure 4B and Table S10). This suggests that P and P’ are unlikely to derive from parallel pathways.

Figure 4:

O₂ concentration-dependence of the formation and decay of HrmI-P and P’. Anaerobic, diferrous (Fe²⁺)₂ HrmI was rapidly mixed with a buffer containing various concentrations of O₂. (A) Time course of P formation at 650 nm. Traces that include a formation phase can be fit to a two-summed exponential equation overlaid in black with one formation phase (negative amplitude) and one decay phase (positive amplitude). (B) Timecourse of P’ formation and decay at 450 nm. Traces are fit to a three-summed exponential equation with two formation phases (negative amplitudes) and one decay phase (positive amplitude). (C) A new phase for P’ formation is apparent under limiting O₂. HrmI (150 μM) was reduced and rapidly mixed with buffer containing 1 mM L-Lysine and one equivalent O₂. The data and residuals are depicted in green, and the exponential fit is overlaid in black.

As P’ forms later in the reaction sequence, we sought to determine if it was formed from the decay of P. To minimize the spectral contributions of the latter, the O₂ concentration was lowered, thus minimizing P accumulation by slowing down its rate of formation. This allows the ability to more accurately monitor P’ in the previously obscured 1 – 50 ms time window. Upon mixing (Fe²⁺)₂ HrmI with 1 eq O₂ and 1 mM L-Lysine, a new third phase for P’ formation becomes apparent with an observed rate of k ~ 80 s⁻¹ (Figure 4C) that is very similar in magnitude to the rate of HrmI-P decay (k ~ 90 s⁻¹). Therefore, at least one route to P’ formation appears to involve the direct conversion from P. Irrespective of the precise scheme, the difference in O₂-dependence kinetics and reactivity towards L-Lysine indicate that P is the primary intermediate generated from O₂ binding and position P’ downstream in the catalytic cycle.

Interrogation of P’ by Mӧssbauer Spectroscopy

The ⁵⁷Fe enriched enzyme was chemically reduced and the resulting low-field 4.2K Mӧssbauer spectrum reveals one quadrupole doublet with the spectral parameters (δ =1.26, ΔE_Q = 2.77) that are indicative of high-spin (S = 2) diferrous (Fe²⁺)₂ species (Figure 5, spectrum A, Tables S11, S12). The addition of excess (10 mM) L-Lysine does not significantly alter the spectrum (Figure 5, spectrum B). This implies that either the K_D for substrate binding to (Fe²⁺)₂ HrmI is substantially higher than that estimated using the P’ decay rate, or the substrate does not significantly perturb the cluster. The latter interpretation is consistent with structural comparisons of (Fe²⁺)₂ HrmI in the ligand-free and bound forms, whereby L-Lysine binding does not induce any major structural rearrangements of the cluster other than the displacement of a terminal Fe1 hydroxo or aquo ligand.

Figure 5:

Characterization of (Fe²⁺)₂ HrmI and P’ by Mӧssbauer spectroscopy. 0.1 T, 4.2K Mӧssbauer spectra (vertical bars) of (A) (Fe²⁺)₂ HrmI, (B) (Fe²⁺)₂ HrmI with excess L-Lysine, (C) substrate-free (Fe²⁺)₂ HrmI mixed with oxygen saturated buffer and frozen at 20 s, and (D) frozen at 800 s. (E) the difference spectrum generated by subtracting spectrum C from spectrum D, and (F) the overlay of the spectra recorded on the 800 s substrate-free HrmI sample (black) and the 9 s HrmI:L-Lys sample (purple), which show the diferric resting state is the only major species in both spectra. The red solid lines are the overall spectral simulations with simulation parameters listed in Table S2. The simulations of sites 1 and 2 for P’ and sites 3 and 4 for diferric components are shown in dark green, dark blue, light green, and light blue lines, respectively. See Table S13 for the simulation parameters.

Freeze-quench (FQ) samples were analyzed by Mӧssbauer spectroscopy to further interrogate the oxidation state and electronic nature of P’ (see SI Methods for a description of simulation methods used for Mössbauer spectral analysis). Substrate-free (Fe²⁺)₂ HrmI was mixed with O₂ at 4 °C and frozen in liquid N₂ at various times. The low-field 4.2K spectrum of the samples quenched at early timepoints (e.g. 9s and 20s), corresponding to a time when P’ is anticipated to accumulate based on the 450-nm absorption kinetics, reveals a set of prominent spectral features that can be best simulated by two partially overlapping quadrupole doublets with δ₁ = 0.62 mm/s, ΔE_Q1 = 0.22 mm/s and δ₂ = 0.48 mm/s, ΔE_Q2 = 0.79 mm/s (Figure 5, spectrum C, Figures S10, S11, and Table S13), which represent a total of ~ 60% of the spectral absorption in the samples quenched at 9 s and 20 s. These parameters suggest the existence of two distinct high-spin ferric sites that belong to magnetically coupled di-iron centers, which is further confirmed by a 7 T Mössbauer measurement showing an S = 0 spin ground state (Figure S10). In addition, a small amount (~ 15%) of the diferric resting state is also present (see the SI Methods for detailed discussion). After incubating (Fe²⁺)₂ HrmI with O₂-saturated buffer for 800 s, the corresponding low-field 4.2K spectrum shows that the spectral features from sites 1 and 2 observed in the early reaction times have completely disappeared. Instead, the prominent spectral features in this 800 s sample, representing ~75% of the spectral absorption, can be best simulated as the diferric resting state species that is represented by two quadrupole doublets with δ₃= 0.52 mm/s, ΔE_Q3 = 0.45 mm/s; δ₄ = 0.48 mm/s, ΔE_Q4 =1.09 mm/s, which are nearly identical to the unreacted as-purified (Fe³⁺)₂ protein (Figure 5, spectrum D, Figure S12, and Table S13). The above-described spectral analysis can be further confirmed through the analysis of the 800s – 20s difference spectrum (Figure 5, spectrum E), which shows the complete conversion of the spectral features from sites 1 and 2 (−60% of the total absorption) to the final diferric component represented by sites 3 and 4 (+60% of the total absorption). Mössbauer measurements were also performed on a sample quenched at 60s. The spectral features exhibited by this sample can be attributed to a mixture of the diferric resting state component (sites 3 and 4) and the early time di-iron component (sites 1 and 2), suggesting a partial decay of the early time species at this point (Figure S10). The combined time-dependent contributions of the early time di-iron species (sites 1 and 2) closely mirror the predicted accumulation of P’ computed from the absorption data to a model involving the direct conversion of P to P’ (Figure S13, Table S14). To provide further evidence that sites 1 and 2 belong to P’, the (Fe²⁺⁾₂:L-Lysine ternary complex was reacted with O₂ and frozen at the t = 9 s timepoint. Sites 1 and 2 are no longer detected, and only the spectral component of the final diferric species is observed (Figure 5, spectrum F), consistent with the faster rate of P’ decay observed in transient absorption studies.

The optical properties of P’, derived from single value decomposition approaches or a comparison of the Mössbauer data to rapid kinetic studies conducted at similar protein concentrations, provide an estimated extinction coefficient of ~1000 – 1500 M⁻¹cm⁻¹ at 450 nm (Figures S14, S15). This value is within the broad range exhibited by synthetic and enzymatic diferric-peroxide species. However, the ΔE_Q values (≤ 1 mm/s) and λ_max (≤ 500 nm) of P’ are substantially lower than those expected for prototypical μ−1,2-P species (Table S5). Given the altered spectroscopic parameters, kinetic placement relative to P, and apparent reactivity of P’ with the substrate, we have assigned P’ as a structurally distinct, “activated” type of peroxodiferric species.

Detection of a Lys-NHOH Metabolic Intermediate in Solution and in Crystallo

The sequence for the 6-electron oxidation (-NH₂ to −NO₂) of L-Lys by HrmI is unknown. Precedent from the proposed mechanisms of FDO arylamine N-oxygenases have implicated the possible involvement of partially oxidized metabolic intermediates that could include hydroxylamino (-NHOH), nitroso (-N=O), or dihydroxy (-N(OH)₂) products.^44–48 In possible support of such a sequence, turnover studies have shown that the metabolism of Lys-NHOH by HrmI affords Lys-NO₂.¹⁰ Anticipating the need to detect and quantify Lys-NHOH, we used the flavin-dependent N-hydroxylating L-Lysine monooxygenase (FMO) NbtG that specifically halts at Lys-NHOH for its enzymatic preparation.^49–50 Samples from NbtG reactions were analyzed using normal phase liquid chromatography-mass spectrometry (LC-MS) (see SI methods for experimental details). In addition to the expected Lys-NHOH product (m/z = 163.1), the NbtG reactions also contained a minor amount of Lys-NO₂ (m/z = 177.1) (Figure S16). Further interrogation of the Lys-NO₂ in reactions conducted in ¹⁸O₂ gas revealed a substantial (and surprising given the well-established mechanisms of FMOs) level of isotopic scrambling (Figure S17). Thus, we suspected that Lys-NHOH may be prone to non-catalytic conversion to Lys-NO₂. The inclusion of ROS scavengers (catalase and superoxide dismutase) in reaction mixtures effectively eliminated Lys-NO₂ formation. Furthermore, analysis of the Lys-NHOH product in reactions with ¹⁸O₂ showed nearly complete (~97 %) incorporation of ¹⁸O.

The product(s) from single-turnover HrmI reactions, prepared using conditions identical to the rapid kinetic studies (i.e. (Fe²⁺)₂ HrmI/excess O₂/L-Lys) were analyzed, and ROS scavengers were included to minimize any possible non-enzymatic transformations. The only product detected under these conditions was Lys-NO₂ (m/z = 177.1), and no products were observed in control reactions where HrmI was omitted (Figure 6A). To favor the possible detection of a metabolic intermediate, multiple turnover conditions (using excess ascorbate as a sacrificial electron donor and the mediator phenazine methosulfate) were also analyzed by LC-MS. The Lys-NO₂ product was accompanied by minor quantities of a product with an m/z (163.1) and retention time consistent with the assignment as Lys-NHOH. When quantified using an NbtG-generated standard, Lys-NHOH accounted for ~4 % of the total product in HrmI reactions.

Figure 6:

HrmI product formation monitored by LC-MS. 0.5 mg/mL Catalase and 0.25 mg/mL superoxide dismutase were included in all reactions. (A) Extracted ion chromatograms (EIC) of Lys-NO₂ (m/z 177.1) under single-turnover conditions. 150 μM of anaerobic, diferrous (Fe²⁺)₂ HrmI was mixed with an equal volume of O₂ -saturated buffer containing 1 mM lysine (post-mix). A control reaction omitting HrmI was otherwise identical. (B) EICs showing the formation of Lys-NHOH (m/z 163.1) by HrmI under multiple turnover conditions. 150 μM HrmI was aerobically incubated with phenazine methosulfate (PMS), ascorbate, and 1 mM lysine for one hour. The product elutes at the same time as the enzymatically generated standard (see SI methods), and no product was observed in identical reactions omitting HrmI (C) EICs of Lys-NO₂ under multiple turnover conditions. In reactions conducted under ¹⁸O_2, Lys-NO₂ with an m/z +4 (181.1) ion is the major product, consistent with the incorporation of two oxygen atoms from ¹⁸O₂. The m/z +4 peak was not observed in control reactions conducted with ¹⁶O₂. (D) MS2 fragmentation of Lys-NO₂ formed in reactions with ¹⁶O₂ (blue) and ¹⁸O₂ (red). The m/z 177.1 and 181.1 parent ions are marked with an asterisk.

To identify the source of oxygen atoms integrated into Lys-NO₂, the products from multiple turnover studies (conducted in ¹⁸O₂ or ¹⁶O₂) were compared. A previous study³⁰ noted significant isotopic scrambling into the product. This implies that HrmI may use a mechanism that is very different from that of FDO N-oxygenases. The authors attributed the source of this isotopic scrambling to experimental error, but given the observations with NbtG, we anticipated the possible involvement of ROS-mediated transformations. Single and multiple turnover reactions were conducted in ¹⁸O₂ gas with ROS scavengers. The vast majority (≥ 93 %) of products exhibited an m/z of 181.1 (+4 relative to the ¹⁶O₂ product), consistent with the incorporation of both oxygen atoms from O₂ (Figures 6C, S17). Ions corresponding to the product were subjected to tandem mass spectrometry fragmentation to further validate this assignment (Figure 6D). When the ROS scavengers were omitted, the products from multiple turnover reactions exhibited m/z values of 177.1, 179.1, and 181.1 with relative abundances of 35%, 5%, and 60%, respectively (Figure S17). This result further emphasizes the susceptibility of the −NHOH product to non-catalytic degradation by ROS and explains the probable source of mixed-isotopic −NO₂ products in the previous study.³⁰ The inclusion of ROS scavengers resulted in merely an 8% increase in Lys-NO₂ production under either single or multiple turnover conditions. Similarly, Lys-NHOH production in multiple turnover reactions was only 5% lower in the presence of ROS scavengers.

Given the ability of HrmI crystals to readily adopt different structural changes upon oxidation, we sought to determine if a metabolic intermediate could be trapped in crystallo. The (Fe²⁺)₂:L-Lysine crystals were exposed to O₂, and the structure was determined with a resolution of 1.96 Å (Table 3, PDB: 9NH9). As in the substrate-free structure, a significant contraction along the diiron axis was observed, consistent with oxidation to the (Fe³⁺)₂ state. The mFo-DFc difference map also reveals a substantial spherical electron density adjacent to the N_ζ of L-Lysine that is not present in the reduced precursor (Figure 7A). The inclusion of an additional oxygen atom (modeled as −NHOH given the results obtained via LC-MS) satisfactorily accounts for the extra density (Figure 7B). Additional mechanistic implications are presented in the Discussion.

Figure 7:

The 1.96 Å resolution structure of HrmI obtained after incubation of L-Lysine bound (Fe²⁺)₂ crystals in O₂. The mFo-DFc difference maps are depicted in green (contoured at 3.0 σ) and red (contoured at −3.0 σ). The 2mFo-DFc map (contoured at 1.0 σ) is depicted in yellow mesh. The substrate in the pocket is modeled as (A) L-Lysine or (B) Lysine-NHOH. At a contour of −3.0 σ, no mFo-DFc signal is observed.

DISCUSSION

It is shown here that the HDO HrmI uses a (Fe₂)²⁺ cofactor to activate dioxygen for the biosynthesis of nitronorleucine. The rapid generation of P in the absence of L-Lysine indicates that the enzyme can initiate substrate-independent O₂ activation despite the presence of a 3C/3H set of cluster ligands. This signifies a notable departure from other characterized diiron HDOs that regulate entry into the catalytic cycle through the provision of a fourth protein- or substrate-derived carboxylate ligand. A similar substrate-independent mechanism also appears to be operative in the 3H/3C HDO Chlamydia Associated with Death Domains (CADD) that houses a heterobimetallic Fe/Mn cluster for O₂ activation. CADD synthesizes p-aminobenzoic acid through a complex multi-step reaction that involves the long-range excision of a tyrosine residue located ~14 Å away from the cluster, and hence does not appear to immediately rely on binding an external substrate to initiate this radical hole-hopping process.^{19, 24, 33} The comparison of the substrate-free (Fe²⁺)₂ and L-Lysine bound structures of HrmI shows that Fe1 is pentacoordinate in the former and only weakly coordinated by the substrate N_ζ in the latter. This arrangement, together with weakly bound aqua ligands at both sites, could be envisioned to lower the barrier for O₂ binding to form P.

The ability to dually access substrate-bound and free structures of (Fe²⁺)₂ HrmI contrasts with HDOs that have been characterized to date and may have important implications on the order and/or timing of cluster assembly. As recently noted by Moore and Drennan,¹⁷ crystallographic access to the substrate-free (Fe²⁺)₂ state is currently limited to substrate-unregulated HDOs such as HrmI and SznF. This differs from AetD, where substrate appears to be required both for cluster assembly in crystallo and for O₂ activation in solution¹⁷. A model for cluster assembly that involves an obligatory substrate-binding step has been advanced by Bollinger, Krebs, and colleagues for BesC, largely based on the altered kinetics of Fe²⁺ removal in the presence/absence of substrates.²² Our studies with HrmI here, and to some extent with UndA previously which can be purified in a (Fe³⁺)₂ form,⁴¹ intimate that cluster assembly and/or stability may differ across the HDO landscape in ways that are not completely understood. Future studies of HDOs with similarly stable clusters may reveal if this provides a variable strategy to regulate O₂ activation or represents an artifact of in vitro preparations and crystal soaking procedures.

The formation of P as the initial product of O₂ activation is common to many HDOs and FDOs. However, the generation of P’ represents a catalytic strategy that was previously confined to the FDO architecture. While the N-oxygenase CmlI can form P through the reaction of the diferric enzyme and H₂O₂, and P’ in the reaction of diferrous CmlI and O₂, the former species is not observed with the native O₂ cosubstrate, nor does it convert to the reactive P’ species under these conditions.⁵¹ There is some indirect evidence for the conversion of P to P’ in the E. coli ribonucleotide reductase (RNR) R2. Mutational and kinetic analyses have implicated both intermediates as on-pathway, but only the latter initiates the single-electron oxidation of a tryptophan residue (W48) to form the mixed-valent oxo-bridged (Fe₂^3+/4+) intermediate termed X.⁵² X then initiates the formation of a tyrosyl radical at Y122. Intriguingly, the D84E R2 variant accumulates a P that precedes tyrosyl radical formation but is inert towards 3-methylindole (an exogenous chemical analog for W48) in the absence of W48, suggesting that W48 oxidation is mediated by an intermediate downstream of P. P decay can also result in the hydroxylation of a nearby Phe residue in D84E/W48F R2. It has been proposed that competing reaction pathways diverge after formation of R2-P, even if this process cannot be directly observed in the wild-type reaction coordinate. The activated P’ is competent for single-electron indole oxidation of the 3-methylindole surrogate in W48 variants, or W48 when present. The competing pathway that leads to Phe hydroxylation is purported to be mediated by P or a downstream high-valent species.⁵²

Further experiments will be needed to assess the electronic structure of HrmI-P’. However, the substrate-dependent acceleration of P’ decay implicates it as the intermediate responsible for initial N-hydroxylation in HrmI. P’ either directly oxidizes the substrate to generate Lys-NHOH or engages with the substrate to initiate conversion to a downstream catalytic species. Conversely, the kinetics of HrmI-P are nearly unchanged by the substrate, and the O₂ concentration data clearly implicate P as an on-pathway precursor to P’. To the best of our knowledge, HrmI thus represents the first instance where interconversion between P and P’ species can be observed in a nonheme diiron enzyme, offering unique opportunities to better understand the factors that promote this activation pathway.

The optical and Mössbauer properties of HrmI-P’ (λ_max ~ 500 nm and ΔE_Q ~ 0.8 mm/s) are reminiscent of the reactive P’ intermediates that have been implicated in the reactions of the FDO N-oxygenases CmlI⁵³ and AurF.⁵⁴ Although the geometric and electronic structures of CmlI-P’ and AurF-P’ have been the source of some conjecture,^{53, 55–56} their unique optical and Mӧssbauer spectroscopic signatures are generally thought to arise from alterations in the peroxo binding mode and/or protonation state.^{55, 57} The recent application of nuclear resonance vibrational spectroscopy and density functional theory (DFT) calculations to both FDOs by the Solomon laboratory has suggested that these P’ intermediates are best described as protonated μ−1,2-hydroperoxy species, a formulation that is thought to promote SET reactivity on arylamine substrates.^57–58 Reaction coordinate calculations indicate that the protonation of the O atom of P’ stabilizes the ferric (Fe³⁺) oxidation state of the adjacent iron. Thus, the iron adjacent to the deprotonated O atom of P’ is better poised to convert to the ferryl (Fe⁴⁺) state to promote O-O bond cleavage and SET from the substrate, followed by radical recombination to form the hydroxylamino product and the diferric cluster.⁵⁸ In HrmI, the substrate amino and intermediary product hydroxylamino groups ligate Fe1 (Figures 2D and 7B). Therefore, if the inserted O atom observed in crystallo (Figure 7B) occupies the same position as the attacking O atom of the P’, the deprotonated site of P’ that attacks the substrate would be adjacent to Fe1, and the protonated O to Fe2. DFT calculations on CmlI have posited that protonation is mediated by an iron-bound water molecule in the diferrous state that, upon the reaction with O₂, moves to bridge the irons and donates a proton to form P’.⁶⁴ Our kinetic evidence for the direct conversion from HrmI-P to P’ suggests that a single step may be involved. The substrate-free diferrous structure of HrmI has two terminal solvent ligands on Fe1 and three on Fe2 that could serve as proton donors. In CmlI, the glutamate ligand proposed to facilitate water rearrangement occupies the same position (PDB: 5HYH)^{51, 64} as the fourth SznF-like glutamate ligand that is absent in HrmI and many other HDOs. This suggests that water rearrangement would require an alternative mechanism. Intriguingly, in diferrous HrmI, one of the Fe2 solvent ligands is positioned between the irons, suggesting that hydroxo bridge formation would require contraction of the Fe-Fe distance but not significant rearrangement of the solvent ligands. While we anticipate some aspects of the HrmI mechanism to resemble those of CmlI and AurF, it remains unclear why only HrmI accumulates both P and P’ in the native reaction with O₂.

The activation of P may represent a global strategy to enable HDOs to generate a competent electrophilic/single-electron transfer reagent for N-oxygenation.^{43, 57} The nitrile synthase, AetD, generates two intermediates that optically resemble P and P’, but the oxidation state and structure of these species has yet to be determined.¹⁷ The two intermediates in AetD are thought to be generated through separate O₂ activation pathways that are both substrate-gated.¹⁷ Although the final product of AetD does not contain an N-oxygenated moiety, the proposed mechanism is thought to involve a substrate-NHOH intermediate¹⁷ through analogy to other nitrile synthesis pathways.^59–61 In contrast to the FDO N-oxygenases and HrmI, the only Fe-O intermediate observed in the reaction of SznF clearly resembles a canonical μ−1,2-P species.³⁶ The precise mechanisms of N-oxygenases may differ depending on the nature of the substrate. The N-H homolytic bond dissociation enthalpy of a primary amine (~95 kcal/mol)⁶² is higher than those of primary anilines (~90 kcal/mol) but also lower than guanidines (~100 kcal/mol). Therefore, the nature of the substrate N-H bond strength may underly a requirement for accommodating different oxidants in these systems. Alternatively, it is possible that protonation and/or rearrangement of P also occurs in SznF but is substrate-induced, as proposed by Bollinger and coworkers³⁶. The diferrous site of SznF contains two terminal solvent ligands. However, SznF lacks the HrmI-like solvent ligand positioned between the irons, suggesting that rearrangement is required to form a hydroxo bridge. It is notable that in CmlI, one of the carboxylate ligands proposed to be involved in proton shuttling is a glutamate (E236)^{51, 64}, whereas an aspartate residue occupies this position in SznF³⁴ and HrmI (Figure 1B). Therefore, a combination of factors such as the number and chain length of carboxylate ligands as well as the position of solvent ligands in the diferrous state may necessitate different mechanisms for proton-donation and ultimately determine whether P, P’, or both accumulate in each system. In SznF, protonation by the substrate methylguanidinium group or a donor that is positioned by the substrate could result in the observed substrate-dependent enhancement of SznF-P decay³⁶. Unfortunately, a substrate-bound diferrous structure of SznF is not available to facilitate the identification of such a proton donor, a limitation that has been attributed to the possibility that substrate binding preferentially occurs after oxygen activation.^{34 36}

A mechanistic scheme that integrates the structural, rapid kinetics, and spectroscopic findings described in this study are shown in Figure 8. The use of an activated P’ to form an −NHOH product appears to unify the early steps of the mechanisms of HrmI, CmlI, and AurF. The 6-electron oxidation performed by these enzymes necessitates multiple rounds of O₂ activation. Recycling of the diferrous cofactor through a substrate intermediate has been proposed for both FDOs, even though the two enzymes may diverge after formation of the initial −NHOH product.^45,63,46 In CmlI, the substrate-NHOH derivative reduces the diferric cluster to circumvent the requirement for exogenous reducing equivalents.^{45, 63} The paltry level of Lys-NHOH accumulation in multiple-turnover studies, and absence in single turnover reactions when only two reducing equivalents are provided to HrmI, strongly suggest that such a mechanism is operant in HrmI.

Figure 8:

Proposed mechanism for HrmI N-oxygenation. Diferrous (Fe²⁺)₂ rapidly reacts with O₂, irrespective of a bound substrate, to form P, followed by the generation of the activated P’ adduct. Various formulations based on proposed structures from the literature. P’ reacts with the L-Lysine substrate to generate a metabolic Lys-NHOH intermediate. An electron recycling system is needed to generate the final nitronorleucine product.

Supplementary Material

The supporting information is available free of charge on the ACS Publications website:

Detailed experimental methods for crystallographic, kinetic, and spectroscopic characterization of HrmI, Figures S1 – S17 and Tables S1 – S13.