Conformational heterogeneity suggests multiple substrate binding modes in CYP106A2

Abstract

CYP106A2 (cytochrome P450_meg) is a bacterial enzyme originally isolated from B. megaterium, and has been shown to hydroxylate a wide variety of substrates, including steroids. The regio- and stereochemistry of CYP106A2 hydroxylation has been shown to be dependent on a variety of factors, and hydroxylation often occurs at more than one site and/or with lack of stereospecificity for some substrates. Comprehensive backbone ¹⁵N, ¹H and ¹³C resonance assignments based on multidimensional nuclear magnetic resonance (NMR) experiments performed with uniform and selective isotopically labeled CYP106A2 samples are reported herein, and broadening and splitting of resonances assigned to regions of the enzyme shown to be affected by substrate binding in other P450 enzymes indicate that substrate binding does not reduce structural heterogeneity as has been observed previously in P450 enzymes CYP101A1 and MycG. Paramagnetic relaxation enhancement (PRE) due to proximity between substrate protons and the heme iron were measured for three different substrates, and the relatively uniform nature of the PREs support the proposal that multiple substrate binding modes are occupied at saturating substrate concentrations.

Graphical Abstract

Left) Two equal energy orientations of abietic acid in the CYP106A2 active site. Structure in gray sticks shows distance between hydroxylation site and heme. Structure in orange lines places the carboxylate close to the heme. Right) Structure of CYP106A2 with residues most affected by heterogeneity highlighted in red.

Introduction

The importance of cytochromes P450 in the biosynthesis of natural products and bioremediation processes cannot be overstated. These Feheme containing monooxygenases activate molecular oxygen (O₂), and catalyze the oxidative modification of unactivated C-H and C-C bonds, often in a highly regio- and stereoselective manner. As such, these enzymes are commonly found in biosynthetic or catabolic pathways that require oxidative modifications for elaboration or modification of intermediates.³ An early P450 to be isolated was CYP106A2 (cytochrome P450_meg), one of several P450s from the soil bacterium Bacillus megaterium.⁴ Although the natural substrate(s) of CYP106A2 are not known, it was found that the enzyme can act a steroid hydroxylase, with the 15β-position of the steroid ring being a major (but not the only) product.⁵ Given that CYP106A2 is soluble and readily expressed in bacterial culture, there is obvious interest in the potential of enzyme engineering to improve the regioselectivity of steroid hydroxylation by CYP106A2 for manufacturing purposes. Extensive research by the Bernhardt group has found that not only the nature of the substrate but that of associated redox partners affect the product distributions in hydroxylations catalyzed by CYP106A2, and that site-directed mutagenesis combined with directed evolution can be used to change hydroxylation regiochemical distributions.^6–8

We are applying multidimensional NMR methods in order to understand the role of enzyme dynamics and conformational selection in substrate binding and orientation in the P450 enzyme superfamily. In two cases that we have examined in detail, (CYP101A1, a camphor hydroxylase from Pseudomonas, and MycG, a macrolactone hydroxylase-epoxidase from Micromonospora) the substrate-depleted enzymes sample multiple conformational states so that many ¹H-¹⁵N correlations in TROSY spectra are broadened asymmetrically. Such broadening indicates that multiple conformations exchanging over a range of time scales are contributing to the observed correlation.⁹ However, upon substrate binding, we observe collapse of most asymmetric multiplets to single correlations. This collapse indicates that substrate binding induces population of a subset of preferred enzyme conformers with low interconversion barriers. We note that both of CYP101A1 and MycG give rise to stereo- and regiospecific oxidations of their respective substrates.¹⁰

However, in the case of CYP106A2, we find that substrate binding does not substantially reduce multiplicity of ¹H,¹⁵N correlations in HSQC spectral maps, suggesting that multiple conformational states separated by relatively high energy barriers are occupied in the substrate-bound enzyme. Practically, this renders difficult the sequential assignment of resonances via multidimensional NMR (critical for more complete characterization of structural and dynamic considerations in the regio- and stereoselectivity of this enzyme), as even in three-dimensional spectra, the overlap of broadened correlations in multiple dimensions increases the ambiguity of through-bond connections. It is likely that these issues have also affected the determination of crystallographic structures of CYP106A2: Unlike most bacterial (soluble) P450s, crystals structures for this enzyme have appeared only recently, and disorder in several regions, particularly near the active site, are apparent in those structures.^{11, 12}

Despite these issues, we now report assignments for ~70% of the backbone amide N-H correlations in CYP106A2, providing insight into those regions of the enzyme that are most affected by conformational heterogeneity. Further insights were obtained from paramagnetic relaxation of substrate ¹H resonances due to the ferric heme iron, with paramagnetically induced relaxation of nuclear spins providing comparative average distances from the heme iron. These allow the characterization of possible binding orientation(s) of several known substrates of CYP106A2 (see Scheme I), and relate this to known hydroxylation patterns of the enzyme.

Scheme I.

Experimental

Synthesis of 1 (levopimaric acid-benzoquinone Diels-Alder adduct).¹³

24 mg of levopimaric acid (BOSci, Shirley, NY) was dissolved along with 10 mg of p-benzoquinone (Sigma-Aldrich) in 0.3 mL benzene in a sealed reaction vial with stirring at room temperature. After 12 h, crystals formed, and 0.1 mL hexanes was added and the mixture briefly sonicated to encourage further crystallization. Crystals of 1 were obtained by filtration after washing with cold 50:50 hexane/benzene solution, m.p. 214 °C. 1 is light-sensitive and so should be stored dark at −20 °C.

Recrystallization of abietic acid 2.

3g of impure 2 (~70%, Fischer Scientific, Waltham, MA) was dissolved to supersaturation in 25mL of diethyl ether by heating the mixture to near boiling, or just under 34°C. (Boiling can lead to phase separation). Once fully dissolved, the amber yellow solution was allowed to cool to room temperature. Faint crystals (the index of refraction of solution and 2 appear to be very similar) could be observed at the bottom of the flask, and so the flask was placed in a sonicating bath for 2 minutes to fragment these crystals into seeds, and then the flask was placed on ice for 1 hour. Once cooled, the flask was swirled to dislodge crystals from the bottom, and the mixture was poured onto a filter paper in a Buchner funnel, vacuumed to dryness and washed with cold diethyl ether. After solvent removal, crystals were scraped from the filter paper into an Eppendorf microcentrifuge tube, yielding 23mg of 2.

Compound 3 (11-deoxycorticosterone) was used as purchased (Sigma-Aldrich) without further purification.

Expression and purification of u-²H,¹⁵N,¹³C CYP106A2.

E. coli strain NCM533^14-16 was transformed with a CYP106A2-encoding plasmid, pKKHC106A2, a gift from Professor Rita Bernhardt. This plasmid features a trc promoter and confers ampicillin resistance.¹⁷ In order to acclimate transformed cells to perdeuterated media, a stepwise procedure was employed. Fresh transformant colonies (NCM533/pKKHC106A2) were inoculated into 5mL LB starter cultures, incubated at 37°C at 200rpm until visible growth was observed. 5mL of the starter culture was transferred to a pre-warmed flask of 50mL M9 minimal media, and incubated at 37°C, 200rpm, until O.D.₆₀₀ ≥ 0.5 was reached. 10mL of this culture was transferred to a pre-warmed 250mL 30% D₂O-based M9 medium, and then via the same procedure to 70% D₂O-based M9 after O.D.₆₀₀ ≥ 0.5 reached. This culture is incubated at 37°C, 200rpm, until the O.D.₆₀₀ ≥ 0.25, which may take 4–6 hours after inoculation. Once the 70% culture has reached an appropriate O.D.₆₀₀, the culture was pelleted in sterile centrifuge tubes using a Beckman-Coulter JA-10 fixed angle rotor, centrifuged at 5000rpm for 30 minutes, 20°C. The pellet was resuspended in 250mL of pre-warmed 99+% D₂O-based M9 media and incubated until O.D.₆₀₀ ≥ 0.8 was reached, and then re-pelleted and resuspended in the final expression media. Note that in all media preparations with D₂O concentrations of 98% or higher, filter sterilization is required, as steam sterilization will dilute the D₂O with H₂O. Due to low cost, ¹⁵NH₄Cl was included in all of the minimal media preparations at all stages. The final culture, 1.375L of pre-warmed 99+% D₂O M9 with u-²H,¹³C-labeled glucose (CIL) was inoculated with the prior culture’s pellet, incubated at 37°C 200rpm until O.D.₆₀₀ = 0.6–0.8 is reached, at which point 100mg 5-aminolevulinic acid (5-ALA) and up to 10μM of an appropriate stabilizing ligand (2 for CYP106A2) were added to the culture. (If only ¹⁵N,²H labelling was desired, unlabeled glucose was substituted for the labeled material). The temperature was then decreased to 30°C. Reaching this point may take another 4–8 hours, so careful monitoring is essential. Once the culture reaches O.D.₆₀₀ = 0.8–1.0, isopropyl β-D-1-thiogalactopyranoside (IPTG) is added to 1mM, the culture was supplemented with 0.5g or more of u-¹³C, u-²H, u-¹⁵N Celtone Base Powder (Cambridge Isotope Laboratories, Inc.), and the rotation speed reduced to 150rpm. The culture was allowed to express for 32 hours. The culture, a tan pink to dark salmon color, was pelleted in a JA-10 rotor, 4°C 5000rpm 45min, after which supernatant was decanted and pellets are collected into 50mL conical tubes. The tubes were flash frozen in liquid nitrogen and stored at −80°C. Spent media may be recycled by distillation under nitrogen from CaO and filtration of the distillate through activated charcoal to recover D₂O for use in subsequent preparations.

Preparation of selectively ¹⁵N- and ¹³C-proline-u¹⁵N labeled CYP106A2.

Selective ¹⁵N-labeled amino acid labeling was accomplished as above, except the inorganic nitrogen source was not labeled, and amino acids were added prior to induction to suppress scrambling. The following amounts of each amino acid was added to 1 L of medium prior to induction: (a, autoclaved; f, filter sterilized) Ala 0.50g a, Arg 0.40g a, Asp 0.40g f, Asn 0.40g f, Gln 0.40g f, Glu 0.65g f, Gly 0.55g a, His 0.15g a, Ile 0.23g a, Leu 0.23g a, Lys 0.42g a, Met 0.25g a, Phe 0.13g a, Pro 0.10g a, Ser 2.00g a, Thr 0.23g a, Tyr 0.17g f, Val 0.23g a, Trp 0.05g f, Cys 0.05g f. The desired ¹⁵N-labeled amino acid was added in lieu of the unlabeled material in the same amount. Samples prepared with selective labels included ¹⁵N-leucine, ¹⁵N-valine, ¹⁵N-isoleucine, ¹⁵N-arginine, ¹⁵N-lysine, ¹⁵N-tyrosine, ¹⁵N-phenylalanine, ¹⁵N-glycine and ¹⁵N-alanine.

Given the frequency of prolines (20) in the CYP106A2 sequence and the interruptions in the amide NH-based sequential assignment process that proline introduces, a sample was prepared with uniform ²H and ¹⁵N labels, and selective ¹³C labelling of proline. Growth and expression followed a procedure similar to that for the uniformly labeled sample described above, except that unlabeled glucose was used to prepare the medium, and prior to induction, 0.10g/L of ¹³C-proline is added, allowing time for the amino acid to be fully dissolved before addition of IPTG. This sample was used for the two-dimensional HNCO experiment described below.

Purification of CYP106A2 and preparation of NMR samples.

Isotopically labeled samples of CYP106A2 were purified as follows. A frozen pellet of cells (on average 9–13g per liter of growth) was thawed to room temperature, after which the pellet was resuspended in 50mL low salt buffer A (50mM Tris-HCl pH = 7.4, 1mM DTT, saturating amount of 1) per 5g of pellet. Phenylmethylsulfonyl fluoride (PMSF) was added to a final concentration of 200μM, and a spatula-tip of DNAse I added. The suspension was homogenized using a Dounce homogenizer, and then lysed using a French pressure cell (Thermo Scientific). After 3–4 passes of the resuspension at 10,000psi, the lysate was translucent, with a rusty orange color. The lysate was then centrifuged in a JA-20 rotor at 15,000rpm, 4°C, for 1 hour. The clarified supernatant was decanted into a syringe Luer-locked with a 0.45μm polyvinylidene fluoride (PVDF) filter. Then, the filtrate was applied to a diethylethanolamine (DEAE) anion-exchange gravity column (CV = 50mL) pre-equilibrated in buffer A. Once binding was complete, the column was washed with 2 CV of buffer A. A step-wise gradient of buffer A to buffer B (buffer A + 400mM NaCl) was applied, in which 50mL each of 90:10, 80:20, 75:25, 70:30, 60:40, and 50:50 buffer A to buffer B was used. 4 mL eluent fractions were collected, with deep red fractions obtained between fractions 20–28. Fractions with the highest A₄₁₇/A₂₈₀ ratio would be pooled, concentrated to <500μL, and injected onto a Superose^™ 6 Increase 10/300 gl size exclusion column (Cytiva, Marlborough, MA) pre-equilibrated with SEC buffer (50mM KPi pH = 7.4, 150mM NaCl, 1mM dithiothreitol (DTT), saturating amount of ligand 1) mounted on an Akta FPLC. A flow rate at 0.45mL/min was used, and fractions were collected in 1min increments. Fractions with the highest A₄₁₇/A₂₈₀ ratio were pooled, and concentrated to <500μL. The sample was then applied to a Cytiva HiTrap 5mL desalting column (Cytiva, Marlborough, MA) preequilibrated with NMR buffer (50mM KPi pH 7.4, 100mM KCl, saturating 1, 10% D₂O). After exchange, the sample was concentrated via centrifugal concentration in 30,000 MW cutoff Amicon Ultra-15 centrifugal filter units to a final concentration of ~400 μM in a volume of 250μL, and transferred to a gas-tight vial. The sample was kept on ice and the vial purged with a gentle stream of carbon monoxide gas for 15 min. Then it was transferred to an anaerobic chamber, where up to 3μL of freshly prepared 250 mM sodium dithionite (Na₂S₂O₄) in 1M KPi, pH = 7.4 was added to the sample, with mixing by inversion in between additions. The sample’s color changed from a deep, dark, blood red to a bright and much more translucent cherry red upon reduction. The sample was then transferred anaerobically to a Shigemi susceptibility-matched NMR tube. The tube was sealed with Parafilm prior to removal from the anaerobic chamber.

NMR data acquisition and analysis.

All two- and three-dimensional NMR spectra used for sequential backbone resonance assignments in CYP106A2 (¹H,¹⁵N-TROSY-HSQC, HNCA, HNCACB, 2D-HNCO, ¹⁵N-edited ¹H-NOESY) were acquired at 25 °C using standard pulse sequences on a Bruker Avance NEO 800 spectrometer at the Landsman Research Facility at Brandeis University equipped with a TCI cryoprobe (¹H, {¹³C,¹⁵N}). Typical data collection time for complete three-dimensional data sets range from 24 to 48 h. The spectrometer operates at 800.13 MHz (¹H), 201.193 (¹³C), 81.076 (¹⁵N) and 122.83 MHz (²H). All acquisitions were acquired using TROSY modifications for narrow line selection.¹⁸ NMR data was processed using TopSpin (Bruker), and sequential assignments performed using SPARKY¹⁹ and CCPNmr²⁰ software packages.

Acquisition and analysis of R₁ relaxation data.

¹H R₁ relaxation rates of substrates 1, 2 and 3 were measured in the presence of increasing concentrations of unlabeled ferric CYP106A2 in order to determine paramagnetic contributions R_1para to the observed substrate ¹H relaxation rates. In turn, R_1para can to be used to calculate relative distances between the heme iron and substrate protons for identifying potential binding orientations. A series of standard inversion-recovery T₁ relaxation experiments were performed for each sample with measurements first in the absence of CYP106A2, followed by addition of aliquots of the enzyme in molar ratios to substrate of (0.01, 0.02 and 0.025). Stock solutions of substrates 1 and 2 were first prepared in d₆-DMSO. These samples were used for NMR experiments required for ¹H resonance assignments, which were made by analysis of ¹H double-quantum filtered COSY, ¹H NOESY, ¹H,¹³C HSQC and ¹H,¹³C heteronuclear multiple bond correlation (HMBC) experiments. For 3, a 100% D₂O buffer 50 mM KPi pD 7.4, 1 mM DTT, was saturated with 3 (60 μM). Spectra were processed and relaxation times for substrate protons were calculated using Topspin 3.0 (Bruker BioSpin). Background ¹H signals due to added protein were suppressed using convolution difference subtraction prior to peak picking and curve fitting to obtain T₁ values. All experiments were performed in triplicate and average T₁ times were used for further analysis.

The contribution of paramagnetic effects to the observed relaxation times, T_1obs, were obtained using the nonlinear fit routine FindFit (Mathematica, Wolfram Inc.) by fitting to equation (2):

$$T_{1obs}^{-1}=\left(\frac{E_o}{K_d+S_o}\right)\left(T_{1para}^{-1}-T_{1free}^{-1}\right)+T_{1free}^{-1}$$ (1)

$$R_{1obs}=\left(\frac{E_o}{K_d+S_o}\right)\left(R_{1para}-R_{1free}\right)+R_{1free}$$ (2)

where E_o and S_o are enzyme and substrate concentrations, respectively. K_d values used in the fits were those determined by Bernhardt et al (53 μM for 1, 63 μM for 2, 60 μM for 3).^{1, 12} Once values for R_1para were obtained, relative distances to the heme iron could be calculated using the Solomon-Bloembergen equation (3):

$$R_{1para}=\frac{2}{15}{\left(\frac{{\mu }_o}{4\pi }\right)}^2\left(\frac{g_e^2{\mu }_B^2{\gamma }_I^{2}S(S-1)}{r^6}\right)\left(\frac{7{\tau }_c}{1+{\omega }_e^2{\tau }_2^2}+\frac{3{\tau }_c}{1+{\omega }_I^2{\tau }_2^2}\right)$$ (3)

where μ_o is the permeability of free space, g_e is the electronic g value, μ_B is the Bohr magneton, γ_I is the gyromagnetic ratio of ¹H, S is the electronic spin, τ_c is the molecular correlation time, ω_I is the ¹H frequency in radians, and ω_e is the electronic transition frequency. The electronic relaxation time τ_e is expected to dominate the value of the correlation time, and the measured value for the low spin state of cytochrome P450_cam (τ_e = 5.4 × 10⁻¹¹ s) was used.²¹

Docking simulations using AutoDockVina and PyRx.

The substrate free homology model of 106A2 was used as a starting point for docking. An energy minimized PDB structure of substrates were generated using ChemDraw3D (Perkin Elmer). Initial docking was performed using the AutoDock Vina plugin for PyRx.²² Docking was restrained to the active site of 106A2 homology model²³ using a 20 Å cubic search space.

Restrained MD simulations of 3-CYP106A2 complex.

Parameter and topology files for the complex were generated using the xleap module of Amber 11.²⁴ Potassium ions were used to neutralize the overall charge of the complex. The sander module of Amber 11 was used to perform energy minimization and subsequent distance-restrained molecular dynamics of the complex. The Fe-H distance restraints used were calculated as described above. 2000 steps of minimization were performed with a 20 Å cutoff, followed by three 2000× 2 fs step equilibration runs of vacuum dynamics performed using a continuum dielectric, from 0–40 K, 40–100 K and 100–300 K. After equilibration, 10 ps of production dynamics were performed at 300K. Coordinates were saved every 100 steps. The generated coordinates were examined for restraint violation penalties, and a representative zero-violation structure chosen for further examination.

Results

Resonance assignments in reduced carbonmonoxy- and 1-bound CYP106A2 were made using a combination of triple-resonance experiments (HNCA, HNCACB) performed with u-²H,¹⁵N,¹³C-labeled enzyme, a ¹⁵N-edited ¹H-NOESY experiment using uniformly ²H,¹⁵N-labeled material, and multiple samples prepared with selective ¹⁵N labeling of amino acid residues by type, including Ile, Leu, Val, Phe, Tyr, Gly, Ala, Arg and Lys. A two-dimensional HNCO experiment performed with a u-²H,¹⁵N, (¹³C-proline) labeled sample was used to identify amide N-H correlations i+1 to prolines. Given the multiplicity of many correlations in the triple-resonance experiments (see Figure 2), the use of type-selective labels to support and/or confirm assignments was particularly useful here.

Figure 2.

Splitting of Cα correlations from Asp 325 and Glu 326 to the amide NH of Glu 326 in the HNCA spectrum of reduced 1-bound carbonmonoxyCYP106A2. E326 lies at the C-terminal end of the K’ helix, which is sensitive to the presence of substrate. Data obtained at 18.8 T (800 MHz ¹H) at 25 °C.

The distribution of the type and extent of spectral multiplicity is shown in Figure 3, with a tabular listing provided as Supplementary Material. While there is necessarily some subjectivity with respect to how the appearance a given correlation is judged, we classified correlations as “single”, that is, no apparent asymmetric broadening, “slight”, with a minor shoulder(s) on the primary peak, “doubled”, at least two major components to a single correlation (see Fig. 2), and “broad”, with multiple different contributions to a given correlation.

Figure 3.

Distribution of conformational multiplicity of assigned resonances in reduced carbonmonoxy- and 1-bound CYP106A2. The left-hand structure is rotated ~90° clockwise from Fig. 1 for clarity. Center and right-hand figures are approximately the same orientation as in Fig. 1. Left: Blue- “single”, one major peak observed. Center: Magenta- “slight”, some asymmetry, with one major peak. Right: Orange, “doubled” into at least two major peaks. Red- “broad”, multiple splittings. See Supplementary Material for tabulation of data used to generate color schemes.

Substrate 1, the Diels-Alder adduct of the slash pine resin component levopimaric acid and p-benzoquinone, was chosen to stabilize reduced (diamagnetic) carbonmonoxy-bound u-²H,¹⁵N,¹³C-labled samples for three-dimensional NMR experiments, as it is one of the few substrates known to induce an appreciable shift from low (S=1/2, λ_max = 418 nm) to high spin (S=5/2, λ_max = 390 nm) heme iron in CYP106A2. This spin shift is taken as an indication of a fairly compact fit of the substrate into the P450 active site (vide infra). It was hoped that this would minimize the conformational space accessible to the enzyme on the ¹H chemical shift time scale. Nevertheless, even with 1 bound, considerable broadening and resonance splitting is still observed, consistent with the presence of multiple substrate binding orientations with similar energetics.

It is worth noting the distribution of the broadening effects (Fig. 3). By far the most affected are residues in secondary structures that we have previously noted to be most responsive to substrate binding in CYP101A1 and MycG.²⁵ In both of those enzymes, these include the β-rich region bounded by the A, B and K’ helices, as well as the N-terminal half of the I helix. In the case of CYP101A1, the F and G helices are less affected, while the C-D region is more strongly perturbed. The sensitivity of the C-D loop in CYP101A1 has recently been ascribed to a secondary binding site for camphor in this region.²⁶ For MycG (with the larger substrate) the F and G helices are strongly affected by substrate binding. It is also worth noting that most residues in the F-G loop (which is disordered in all crystal structures of CYP106A2, and was modeled into the structures in Fig. 3) fall into the strongly broadened category.

Equally important is the observations of where there is less noticeable splitting/broadening of correlations in CYP106A2: These include the J and L helices, the “β-meander” that contains the proximal Cys thiolate ligand for the heme iron, and the regions of the β5 sheet that are remote from the active site (Fig. 3 left). In both CYP101A1 and MycG, these regions were relatively unperturbed upon substrate binding and exhibited the highest degree of sequence homology between those two enzymes.

Paramagnetic contributions to ¹H relaxation as a probe for relative proton-Fe distances in enzyme-substrate complexes.

Dipolar interactions between nuclear spins and unpaired electron spins are an important contributor to both transverse (T₂) and longitudinal (T₁) relaxation in paramagnetic molecules. The Solomon-Bloembergen equation (Eq. 3) provides an estimate of the distance between a given nuclear spin and an unpaired electron based on the extracted value of paramagnetic contribution R_1para to the observed nuclear spin longitudinal relaxation rate R₁ (the inverse of the relaxation time T₁). There are obvious caveats to the use of this approach in characterizing the orientation of substrate in the active site of a P450. First, it is assumed in Eq. 3 that the unpaired electron is localized on the iron; given that there is some delocalization of unpaired spin density into the porphyrin π system, this simplification leads to some ambiguity for the distances calculated. Furthermore, the spin state of the paramagnetic center is specified in Eq. 3, and mixing of spin states will complicate the calculation, even if the extent of mixing is known.

Nevertheless, we²⁷ and others²⁸ have found that such measurements can provide useful qualitative information regarding substrate orientation if the Fe-H distances extracted are treated as comparative values (as opposed to absolute distances). That is, the greater the paramagnetic contribution R_1para of to the observed relaxation rate R_1obs, the more time that ¹H spin spends near the heme iron. As such, we measured the ¹H relaxation times of the three CYP106A2 substrates under investigation to see if we could obtain any insights into substrate orientation(s) in the active site based on this data.

In only one case, that of abietic acid 2, was there a discernable preference for an orientation that placed the site of hydroxylation (C8) significantly closer to the heme iron (5.6 Å) than other positions. However, the two methyl groups C16 and C17 (which overlap in ¹H NMR spectra) with a calculated distance of 6 Å, are nearly as close, as is the sp² hybridized C10 (6.7 Å). Rigid body modeling using the Autodock Vina plugin for PyRx detected no strong orientational preference for binding of 2, with orientations placing the carboxylate (C20) close to the heme with similar energies to that placing the hydroxylation site close (Figure 5).

Figure 5.

Two equal energy binding orientations of equal energy (−5.4 kcal/mol) from AutoDock Vina for substrate 2 into PDB structure 4YT3. Structure in gray sticks shows distance between hydroxylation site (C8) and heme. Structure in orange lines places the carboxylate (C20) close to the heme.

In the case of substrate 1, all the calculated distances range between 9.8 and 12.6 Å, with most falling between 10 and 11.8 Å (see Table S2). As the hydroxylation site for 1 has not been reported, it is not known whether any of the closer distances correspond to the site of hydroxylation. In any case, the range of distances is too small to allow for a confident prediction of regiochemistry of hydroxylation in 1.

Substrate 3, 11-deoxycorticosterone, is hydroxylated by CYP106A2 at the 7β and 15β positions. Again, the calculated distances for most ¹H spins to the heme iron in 3 range between 10–12.3 Å, with the 7β proton at 10.9 Å (7α at 11.1 Å), and the 15β at 12.9 Å (15β at 12.1 Å) (See Table S4 for all calculated distances and steroid atom numbering for 3). AutoDock Vina modeling of 3 into the rigidified CYP106A2 active site yielded nine different orientations (Figure 6) that varied in energy by less than 0.7 kcal/mol, again with no particular orientation of 3 obviously favored. The nominal lowest-energy orientation shown in red in Fig. 6 was chosen as input for an in vacuo molecular dynamics simulation using the calculated Fe-H distances in Table S4 as maximum distance restraints. A typical no-violation structure extracted from that simulation is shown in Figure 7. As can be seen, both hydroxylation sites are accessible to the heme iron, with only a small movement required to position one or the other of the sites appropriately for reaction.

Figure 6.

Nine most energetically favorable orientations of 3 in the rigidified CYP106A2 active site generated using AutoDock Vina. Despite widely different orientations, calculated energies ranged between −7.1 and −7.8 kcal/mol.

Figure 7.

Representative 3-CYP106A2 complex from restrained MD. See text above for details.

Discussion

The observed multiplicity of NMR correlations distributed as shown in Figs. 3 is expected to relate to the distribution of conformational states accessed by the enzyme at a given residue under the conditions of the experiment, with rapidly interconverting conformations contributing to a single observable peak. Narrow peaks suggest rapid interconversion on the ¹H chemical shift time scale, which will broaden as interconversion times increase, until slow exchange on the NMR time scale will result in separate correlations. The splitting in the correlations assigned to Glu 326 shown in Fig. 2 (~64 Hz) combined with the line widths of individual peaks (~40 Hz) when fit to a two-state simulation indicates an interconversion rate < 100 s⁻¹ between the ensembles contributing to each peak. We note that Glu 326 lies at the C-terminal end of the K’ helix, which is strongly perturbed by substrate binding both in CYP101A1 and MycG.^{25, 29} Estrada et al. noted similar slow conformational interconversions in spectra of CYP17A1 with type II inhibitor abiraterone bound, although unlike CYP106A2, the observed heterogeneity appears to be restricted to regions adjacent to the active site.³⁰ As abiraterone is tightly bound in the CYP17A1 active site via a dative covalent bond to the heme iron, it would seem that, at least in that case, the conformational exchange is occurring with inhibitor remaining bound. While this does not preclude the possibility that substrate 1 is exiting and re-entering the active site on a time scale sufficient to account for the observed splitting/broadening of lines, the fact that multiple states are present indicates that more than one substrate-bound conformational ensemble is populated. A reviewer of an earlier version of this paper suggested that it was possible that the heterogeneity observed here is a response to substrate binding, and that in the absence of substrate, interchange may be faster. We will investigate this possibility in future work.

Spin state and substrate binding

Upon substrate binding, P450 enzymes typically exhibit a change in the spin state of the heme iron, from a low spin Fe³⁺ (S=1/2, λ_max = 418 nm) to a high spin form (S=5/2, λ_max = 390 nm). The spin state change is accompanied by a redox potential shift that permits the first electron required for O₂ activation to be acquired.³¹ While this spin state change has traditionally been associated with the dissociation of the resting state water/hydroxide axial ligand to the heme iron upon substrate binding, we and others have found evidence that this spin state change is the result of conformational changes in the heme porphyrin itself,^{32, 33} with partial or complete dissociation of the axial ligand likely an effect rather than the cause of the spin state change.³ In the case of CYP106A2, substrate-induced spin state changes are typically small, and often undetectable by UV-visible spectroscopy: The binding of deoxycorticosterone 3 is detected only by perturbations of heme-CO absorbances in infrared spectra³⁴, while substrate 1 results in a small (< 5%) shift to 390 nm at saturation (K_d ~ 53 μM).¹ Given the significantly larger steric bulk of 1 relative to other known CYP106A2 substrates, this supports the notion that spin state changes are the result of direct or indirect interactions between the heme porphyrin and substrate.^{3, 33} Nevertheless, the fact that NMR spectra of CYP106A2 obtained with 1 bound still exhibits significant multiplicity, particularly for regions near to the active site or that are sensitive to substrate binding, indicates that 1 binds in multiple orientations. It was noted that 1 is oxidized in reconstituted CYP106A2 assays to yield a single product,¹ suggesting that the orientation giving rise to the high spin form may also be that which is also most conducive to reaction. Indeed, our recent combined computational/in vitro mutational characterization of CYP101A1 demonstrated that, while substrate (in that case, camphor) could occupy multiple orientations in the active site, both the spin state change and enzyme efficiency tracked remarkably closely with the fraction of time that site of hydroxylation spent close to the heme iron in simulations.¹⁰

We have previously suggested that, despite differences in sequence and specific function, the conserved structural features across the P450 superfamily perform similar functions in the different enzymes.²⁵ In particular, the proximal J, K and L helices, along with the portion of the β-meander containing the axial cysteinyl ligand to the heme iron, provide a stable platform required for the monooxygenase activity that members of the superfamily perform. As was the case between CYP101A1 and MycG, despite low overall sequence homology between CYP101A1 and CYP106A2 (30% identity, 44% similarity), these regions display a higher degree of identity (45%) and similarity (60%,) with no gaps in either sequence in these regions (See Supplementary Figure S1).³⁵

Conversely, those regions of CYP106A2 that exhibit the greatest sensitivity to substrate binding modes as described here and shown in Fig. 3 are indeed the same regions that were seen to be responsive to substrate binding in comparisons of solution structural ensembles of CYP101A1 and MycG with and without their respective substrates bound.²⁵ Our current efforts are focused on identifying ligands that might be used to stabilize a single conformational ensemble of CYP106A2. If successful, this will make possible the use of residual dipolar couplings to generate solution ensembles of substrate-free and bound CYP106A2 in a manner similar to our previous work with CYP101A1 and MycG.

Figure 1.

Superposition of crystallographic structures of CYP106A2 with abietic acid bound (5IKI, cyan) and substrate-free (4YT3, orange). Heme is shown in purple, abietic acid in green. Secondary structures are labeled as by Poulos.² Helix A, P22-N32; β1, V35-V46; B, Y49-S57; C, P94-A106; D, R110-Q129; E, D137-M155; F, L163-F173; G, E183-L208; H, I214-K220; I, D230-D262; J, E265-E272; K,L274-F287; β3 (s1), K293-V298; β4, K299-E311; β3 (s2), G312-W318; K’, M319-D325; β meander, E326-L356, L, G357-K374, β5, K377-S408. 4YT3 is disordered from K67-V82, between the B and C helices. Both structures are disordered in the F-G loop, F176-Q181.

Figure 4.

Carbon atom numbering for abietic acid 2 used in text. CYP106A2 catalyzed hydroxylation is regiospecific at C8, but both stereoisomers are detected. Readers should note that C8 corresponds to C12 in Bleif et al.¹

Comprehensive backbone NMR assignments for CYP106A2

Multiple substrate binding modes give rise to heterogeneity in NMR spectra.

Most affected regions are those involved in substrate binding.

Paramagnetic relaxation patterns support multiple substrate binding orientations.