Evidence for distinct rate-limiting steps in the cleavage of alkenes by carotenoid cleavage dioxygenases

Carotenoid cleavage dioxygenases (CCDs) use a nonheme Fe(II) cofactor to split alkene bonds of carotenoid and stilbenoid substrates. The iron centers of CCDs are typically five-coordinate in their resting states, with solvent occupying an exchangeable site. The involvement of this iron-bound solvent in CCD catalysis has not been experimentally addressed, but computational studies suggest two possible roles. 1) Solvent dissociation provides a coordination site for O2, or 2) solvent remains bound to iron but changes its equilibrium position to allow O2 binding and potentially acts as a proton source. To test these predictions, we investigated isotope effects (H2O versus D2O) on two stilbenoid-cleaving CCDs, Novosphingobium aromaticivorans oxygenase 2 (NOV2) and Neurospora crassa carotenoid oxygenase 1 (CAO1), using piceatannol as a substrate. NOV2 exhibited an inverse isotope effect (kH/kD ∼ 0.6) in an air-saturated buffer, suggesting that solvent dissociates from iron during the catalytic cycle. By contrast, CAO1 displayed a normal isotope effect (kH/kD ∼ 1.7), suggesting proton transfer in the rate-limiting step. X-ray absorption spectroscopy on NOV2 and CAO1 indicated that the protonation states of the iron ligands are unchanged within pH 6.5–8.5 and that the Fe(II)–aquo bond is minimally altered by substrate binding. We pinpointed the origin of the differential kinetic behaviors of NOV2 and CAO1 to a single amino acid difference near the solvent-binding site of iron, and X-ray crystallography revealed that the substitution alters binding of diffusible ligands to the iron center. We conclude that solvent-iron dissociation and proton transfer are both associated with the CCD catalytic mechanism.

ene bonds to form carbonyl-containing products (1)(2)(3). Carotenoids are the canonical substrates for CCDs, but some members of this family instead cleave noncarotenoid compounds such as stilbenoids (4). CCDs play crucial roles in various synthetic biological processes (5,6). These include the formation of retinaldehyde chromophores for opsin proteins (7) and signaling molecules such as abscisic acid (8), retinoic acid (9), and strigolactones (10). Some CCDs, e.g. ␤-carotene oxygenase 2, also function as catabolic enzymes to degrade potentially toxic carotenoids (11). Despite progress made in elucidating CCD substrate specificity and biological activity, the mechanisms by which these enzymes bind their substrates and activate O 2 to achieve regioselective alkene bond cleavage remain incompletely understood.
X-ray crystal structures of Synechocystis sp. PCC 6803 apocarotenoid oxygenase (ACO) (12) and 9-cis-epoxycarotenoid dioxygenase from Zea mays viviparous-14 (VP14) (13) and more recent studies on stilbene-cleaving CCDs from fungi (14) and bacteria (15,16) have revealed a conserved active-site geometry for the CCD family. The resting-state active site contains an iron cofactor coordinated by four conserved His residues ( Fig. 1). Two cis-localized sites in the iron-coordination sphere are unoccupied by protein ligands and are potentially available for diffusible ligand binding. In all structures mentioned above except that for VP14, one of the vacant sites is partially occluded by the methyl group of a nearby Thr side chain, which prevents the metal from attaining a complete sixcoordinate octahedral ligand set. The accessible site is occupied by a solvent molecule in a majority of these crystal structures, although two of them contain modeled O 2 instead (13,15). The resting-state iron structure with four coordinating His residues and a bound solvent ligand gives the metal center a distorted square pyramidal geometry. X-ray absorption spectroscopy (XAS) and Mössbauer spectroscopy studies on the iron centers of NOV2, CAO1, and ACO further corroborated their fivecoordinate geometry and established that they are all in a highspin Fe(II) state (14).
Crystal structures of cobalt-substituted CAO1 in complex with stilbenes showed that substrates bind ϳ4 -5 Å away from the metal center, allowing the solvent to remain bound to the metal without inducing major conformational changes in the protein structure (14). Comparative Mössbauer studies on NOV2, CAO1, and ACO and their respective ES complexes showed only minor differences in the electronic environment of the active-site Fe(II) atom, indicating small changes in iron coordination upon substrate binding to the active site (14). These observations are different from many other mononuclear nonheme iron enzymes whose iron-bound solvent molecules are displaced upon substrate binding to their active sites (17,18). The role of the iron-coordinated solvent in O 2 binding and activation by CCDs is unclear, but a computational study on the ACO catalytic mechanism has suggested two possible scenarios (19). In the first, the solvent ligand dissociates from iron, creating a vacant site for O 2 binding. In the second, solvent ligand remains bound to iron but changes its equilibrium position to allow simultaneous binding of O 2 . Here, the bound solvent could also act as a proton donor to stabilize reaction intermediates.
We sought to address the role of solvent in the alkene cleavage reaction through solvent kinetic isotope effect (sKIE) studies. Metal-bound aquo/hydroxo ligands exhibit hydrogen/deuterium fractionation factors that are less than unity ( Ͻ 1), meaning that deuterium tends to accumulate in bulk solvent as opposed to the metal-bound solvent ligand (20,21). Thus, dissociation of the metal-bound solvent ligand is thermodynamically more favored in D 2 O compared with H 2 O (22). In a situation where slow aquo dissociation occurs as part of the rate-limiting step of catalysis, an inverse isotope effect is anticipated (23). Conversely, a normal KIE is expected if proton transfer occurs during a rate-limiting step of the reaction (22). These isotope effect studies were complemented with XAS and X-ray crystallography to further address O 2 gating and the proton transfer events associated with CCD catalysis. Our data on NOV2 and CAO1, respectively, indicate that the dissociation of solvent ligand from the Fe(II) ion provides the binding site for O 2 , and a single proton transfer occurs during a step that is rate-limiting under some conditions. We also demonstrate that a single active-site amino acid difference between these two enzymes alters the rate-limiting step of catalysis.

Three CCD enzymes each exhibit distinct isotope effects on steady-state activity
We initiated the study by measuring k cat values for NOV2, CAO1, and ACO in H 2 O-and D 2 O-containing buffers under air-saturated conditions using standard substrates, all-trans-8Ј-apocarotenol for ACO and piceatannol for NOV2 and CAO1 (Fig. 2). The activity was monitored by optical spectrophotometry as described under "Experimental procedures" (Fig. S1) (2,24). ACO activity was unaffected by the substitution of isotopic solvent within experimental uncertainty (k H /k D ϭ 1). By contrast, NOV2 and CAO1 exhibited distinct kinetic behaviors in the two buffers. We observed an inverse isotope effect for NOV2 (k H /k D ϭ 0.62), whereas CAO1 displayed a normal KIE (k H /k D ϭ 1.7). Notably, CAO1 exhibited a significantly greater turnover number compared with NOV2 despite the reactions being carried out under identical conditions, indicating functionally important active-site differences between the two enzymes. The divergent kinetic behavior between NOV2 and CAO1 prompted us to examine their sKIE dependences in more detail.

Factors determining the inverse isotope effect on NOV2 activity
To further examine the inverse sKIE on NOV2 activity, we measured NOV2 steady-state kinetics in H 2 O-and D 2 O-based buffer systems at a range of substrate concentrations to allow extraction of the Michaelis-Menten kinetic parameters. We found that the inverse isotope effect on k cat for NOV2 was essentially constant over the range of substrate concentrations and pL values tested under air-saturated conditions (Fig. 3). The catalytic efficiency parameter (k cat /K m ), which is influ-

Solvent isotope effects on alkene bond cleavage by CCDs
enced by all reaction steps up to the first irreversible step, also revealed an inverse isotope effect of approximately an equal magnitude observed for k cat (Table 1).
We next asked whether the inverse isotope effect depends on the O 2 concentration in the reaction mixture. As a prerequisite to the experiment, we first determined the NOV2 steady-state kinetic parameters for O 2 at a fixed concentration of piceatannol (35 M). We recorded k cat and K m values of 1.3 s Ϫ1 and 166 M O 2 , respectively, indicating that the enzyme operates at submaximal capacity in air-saturated buffer under these conditions (Fig. 4A). We then measured the isotope effect in O 2 -saturated buffer (ϳ780 M O 2 ) in which the enzyme is fully saturated with cosubstrate and compared it with that observed in air-saturated buffer. We found that the inverse isotope effect was completely abolished under saturating O 2 conditions, suggesting a change in the rate-limiting step of the reaction (Fig.  4B). The observed kinetic parameters measured for NOV2 in air-saturated conditions are thus apparent parameters.
Inverse isotope effects have a limited number of documented causes in enzyme systems. These include viscosity-dependent conformational changes (25), involvement of nucleophilic thiols in catalysis (26), and aquo release from metal sites (21)(22)(23). We examined the first potential cause by carrying out reactions in an H 2 O-based reaction buffer containing 9% glycerol, which matches the viscosity of a 100% D 2 O-based buffer system (25). This change in viscosity resulted in a slight reduction in the reaction rate, which allowed us to rule out the first possible origin of the inverse isotope effect (Fig. S2). We excluded the second possibility as well by observing that NOV2 does not contain Cys residues at sequence positions that are predicted, based on the CAO1 crystal structure, to form its active site.
Thus, the release of an aquo ligand from the metal ion best explains our inverse isotope effect observation.

Insight into the normal KIE on CAO1 catalysis
Unlike NOV2, CAO1 cleaved piceatannol in an air-saturated buffer in a mechanism that showed a normal KIE (Fig. 2). To exclude the possibility of the KIE being a result of the viscosity difference between H 2 O and D 2 O, we carried out the CAO1 reaction in H 2 O buffer supplemented with 9% glycerol. In contrast to the results described above for NOV2, we did not observe any change in the reaction rate in the presence of glycerol (Fig. S3). We considered the possibility that the normal KIE might be a consequence of a difference in ionization behavior of side-chain residues in H 2 O and D 2 O. One strategy to account for such behavior is to perform the KIE studies at different pH values. We performed the CAO1 reaction at pH 6.5, 7.0, 7.4, 8.0,

Solvent isotope effects on alkene bond cleavage by CCDs
and 8.7 in an air-saturated buffer with saturating substrate and observed a normal KIE of similar magnitude for all pH conditions tested (Fig. 5).
Proton inventory studies probe the number of protons transferred during the transition state of the rate-limiting step (27,28). For a normal KIE, the rate of reaction decreases with increasing mole fraction of D 2 O in the reaction buffer. Such behavior was observed for CAO1 in air-saturated buffer. We plotted the ratio of the observed k cat at "n" mole fraction of D 2 O to k cat at 100% H 2 O (k n /k 0 ) versus the mole fraction of D 2 O (n) and observed a linear correlation between k n /k 0 and n. The data from the resulting plot were fitted to the linear form of the Gross-Butler equation for a single hydron transfer with an observed KIE of ϳ1.7 (Fig. 6), suggesting that a single proton transfer occurred during the rate-limiting step (29).
Analogous to our experiments on NOV2, we performed isotope effect studies at varying O 2 concentrations with CAO1 to determine its effect on the KIE. We first determined the rate dependence on O 2 concentration at fixed piceatannol concentration (35 M). The reaction rate increased in a hyperbolic manner with increasing O 2 concentration, and the data were fitted to the Michaelis-Menten equation with kinetic parameters of k cat ϭ 4.68 s Ϫ1 and K m ϭ 114 M O 2 , indicating that the reaction is unsaturated for O 2 in air-saturated buffer (Fig. 7A). We observed that the KIE is maintained at saturating O 2 concentrations indicating that proton transfer remains rate-limiting under these conditions (Fig. 7B). By contrast, we observed a significant decrease in magnitude of normal KIE when the O 2 concentration was below that of air-saturated buffer, at or below its K m value. This observation is consistent with a normal KIE competing with an inverse isotope effect.

Impact of pH changes and substrate complexation on X-ray absorption spectra of NOV2 and CAO1
Previous iron K-edge XAS analyses of NOV2 and CAO1 at pH 8.0 suggested slightly different resting-state structures for the iron centers of these enzymes, likely due to differences in exchangeable ligands such as bound solvent (14). We hypothesized that this structural difference could, at least in part, give rise to the difference in kinetic behavior between these two enzymes. To probe for pK a differences in the primary ironcoordinating ligands between these enzymes, we performed iron K-edge XAS experiments on NOV2 and CAO1 at pH values of 6.5 and 8.5 where variations in catalytic activity could be observed (Fig. 5). K-edge X-ray absorption near-edge spectroscopy (XANES) provides information about metal coordination number, oxidation state, and electronic orbital occupancy and has been previously used to detect differences in the iron centers of CCDs (14). XANES spectra for NOV2 at pH 6.5 and 8.5 and CAO1 at pH 6.5 and 8.5 are presented in Fig. 8, A and B, respectively. The spectra are consistent with our previously published XANES data on NOV2 and CAO1, exhibiting edge inflections at 7121.7-7122.0 eV, consistent with an iron(II) formulation, as well as bimodal pre-edge absorption bands (E pre-edge ϭ 7112.6 and 7114.5 eV), which are attributable to formally forbidden 1s-to-3d electronic transitions gaining intensity from admixture with 4p orbitals in noncentrosymmetric geometries. The shapes and intensities of the pre-edge transitions are consistent with five-coordinate iron(II) centers. Notably, we observed essentially no change in the XANES spectra between the two pH states, as evidenced by a lack of changes in the edge and pre-edge energies, and only subtle changes in pre-edge peak areas ( Fig. S4 and Table S1). This result indicates that there is no change in iron-site symmetry or the effective

Solvent isotope effects on alkene bond cleavage by CCDs
nuclear charge experienced by the iron(II) center. We therefore suggest that there is no change in iron coordination environment over this pH range and that ligands in the iron primary coordination sphere have pK a values outside the pH range studied here.
Although NOV2 and CAO1 crystallographic and Mössbauer spectroscopy studies have indicated that solvent remains bound to iron in the presence of organic substrate (14), we considered the possibility that substrate binding may weaken the metal-aquo bond in CAO1 more so than in NOV2, thus altering the rate-limiting step. To test this hypothesis, we undertook XANES studies of enzyme (E) and ES complexes of NOV2 and CAO1 at pH 7.4 (Fig. 8, C and D, respectively). Binding of piceatannol to NOV2 and CAO1 afforded a small ϳ0.2-eV increase in edge energies for the ES states relative to their E-only counterparts as well as a slight reduction in white line intensity. The former suggests that the iron environment became slightly more electropositive in the presence of piceatannol ( Fig. S5 and Table S2), whereas the latter suggests a possible subtle increase in the distribution of bond lengths seen by the iron(II) center in both NOV2 and CAO1. However, the 1s-to-3d pre-edge peak positions and areas changed only subtly between the two states, suggesting that substrate binding does not elicit a significant change in iron coordination environment for either NOV2 or CAO1.

A single amino acid substitution converts CAO1 solvent isotope effect from normal to inverse
The apparent lack of differential pH and substrate-binding effects on the iron environments of NOV2 and CAO1 prompted us to directly examine structural differences between these enzymes that could underlie their different behaviors in isotopic solvents. Crystallization conditions have previously been established for CAO1 (14), whereas NOV2 has proven recalcitrant to crystallization. Therefore, we focused on CAO1 for crystallographic analysis. Considering the apparent difference in stability of the iron-solvent complex between the two enzymes suggested by the kinetic isotope effects, we inspected the Fe-CAO1 crystal structure (14) for residues in proximity to this moiety (i.e. within 8 Å) (Fig. 9A) and compared these with homologous sites in NOV2 (Fig. 9B). From this comparison, we identified three sites differing between CAO1 and NOV2, namely amino acid positions 91, 383, and 509. To investigate the impact of these differences on the isotope effect, we generated F91L, E383D, and L509V point mutants of CAO1 and examined their activity in H 2 O-and D 2 O-based buffers. We observed that the normal KIE was maintained in both the F91L and E383D variants, whereas an inverse isotope effect, similar in magnitude to that of NOV2, was observed for L509V CAO1 (Fig. 9C). Inspection of the CAO1 structure revealed that, aside from the iron-coordinating His residues, the L509 C ␦2 atom makes the closest van der Waals interaction with the coordinated aquo ligand. Such a close contact would be altered by substitution of a Val side chain at position 509, which suggested that this difference could influence the strength of solvent interaction with iron. To gain further insights into the mechanism of this effect, we crystallized L509V and WT CAO1 under similar conditions and determined their structures by X-ray diffraction analysis ( Table 2).
The overall structure of L509V CAO1 was similar to that of the WT protein with an average root-mean-square difference of ϳ0.15 Å between chains in equivalent positions within the asymmetric unit. The presence of a Val residue at position 509 was fully supported by the difference electron density map calculated after the first cycle of refinement with Leu modeled at the site (Fig. S6). Refinement of the updated L509V model against the data resulted in an excellent fit of the Val side chain to the density. Interestingly, Val-509 adopted a conformation within the outlier region of the Val/Ile-specific Ramachandran plot, although itsangles reside in an allowed region of corresponding general plot (30). The only other notable active site difference between the two proteins was found at Phe-91 where the side chain adopted an alternative conformation with its phenyl ring rotated away from the iron center nestled within a hydrophobic cavity formed by Pro-89, Trp-340, Trp-339, and Tyr-133 side chains (Fig. S7). The alteration was possibly triggered by a loss of long-range van der Waals contacts normally formed between the Phe-91 and Leu-509 side chains. Nearly

Solvent isotope effects on alkene bond cleavage by CCDs
identical active-site structural differences were observed in a second L509V CAO1 crystal structure obtained under distinct crystallization conditions (Fig. S8 and Table S2). The extent to which this induced structural difference applies to NOV2 is unclear given that the protein contains a different residue at the corresponding sequence position (Leu instead of Phe; Fig. 9B). The overall result of the L509V substitution in CAO1 is thus an expansion of the active-site volume available for ligand binding within the iron coordination site trans to His-197.
Inspection of the difference electron density maps revealed an alteration in appearance between the WT and L509V CAO1 structures (Fig. 10). In our previously reported WT CAO1 structure (14), a relatively weak density was present near the iron center trans to His-197 that we modeled as a bound solvent  . Active-site variation between CAO1 and NOV2 and impact of amino acid substitutions on CAO1 sKIEs. A, residues within 8 Å of the CAO1 metal center are shown as sticks. Those labeled in bold red text are sites differing between CAO1 and NOV2. Iron and water are shown as brown and red spheres, respectively. B, selected regions of an amino acid sequence alignment between CAO1 and NOV2 that are centered on the residues labeled in A (marked by arrows under the sequence alignment). Three residues close to the iron center are nonidentical between CAO1 and NOV2 (marked with red arrows). C, substitution of Leu-509 with Val, as found in NOV2, changes the KIE from normal to inverse, whereas the other two active-site substitutions preserve the normal sKIE as observed for piceatannol cleavage by wild-type CAO1. The experimental conditions were similar to that of Fig. 5. Mean values are shown for C.

Solvent isotope effects on alkene bond cleavage by CCDs
molecule. We obtained a similar result with WT CAO1 in the present study but found that one protomer of the asymmetric unit (Fig. S8, chain C) had a particularly low difference electron density at this position, making the iron appear four-coordinate. This observation is notable given that the four-coordinate state of the Fe(II) center is likely the one capable of binding and reacting with O 2 . By contrast, both L509V structures featured stronger and larger difference electron density peaks at the corresponding coordination position. The density was heterogeneous when comparing different chains, which confounded our ability to assign it to a particular ligand. Attempts to model a single solvent molecule into the density resulted in reasonable iron-solvent bond lengths and B-factors but did not satisfactorily quench all of the difference density at the site. The addition of a second scatterer in the form of an O 2 molecule accounted for more of the difference electron density, but the refined O 2 ligand was found in a variety of orientations and distances contrary to chemical expectations. In light of the structural alterations described above for L509V that reduce hydrophobicity near the iron center and increase the potential ligand-binding volume, it is conceivable that the density represents multiple solvent ligands, each with partial occupancy.
It is notable that although the iron and surrounding N ⑀2 atoms of the coordinating His residues had atomic B-factors of   Table 2.

Solvent isotope effects on alkene bond cleavage by CCDs
similar magnitude, indicative of full iron occupancy, the WT CAO1 iron B-factors were disproportionately higher, suggesting either static or dynamic heterogeneity and/or incomplete occupancy ( Table 2). The apparent stronger ligand coordination observed for L509V CAO1 may help restrain the iron mobility and potentially prevent its dissociation, which would account for the lower iron B-factors.

Discussion
The steady-state kinetic isotope effects observed for NOV2, CAO1, and ACO when the protic solvent (H 2 O) is substituted by its isotope (D 2 O) provide an opportunity to gain insights into the different steps of CCD catalysis. Previous computational results suggested two different mechanisms for O 2 binding. In the first, the aquo ligand dissociates from the Fe center and provides the binding site for O 2 , whereas in the second, water remains bound to iron but changes its equilibrium position to accommodate O 2 binding as shown in Scheme 1.
Our results for NOV2 under air-saturated conditions support a mechanism in which the aquo ligand dissociates during O 2 binding/activation. In air-saturated buffer (ϳ280 M O 2 ), NOV2 demonstrated an inverse isotope effect on the steadystate k cat (k H /k D ϭ 0.61), which measures the rate-limiting step of the overall reaction. Based on previously measured inverse fractionation factors ( Ͻ1) for metal-bound aquo ligands (20) and experimental evidence of an inverse isotope effect reflecting metal-aquo dissociation in hypoxia-inducible factorprolyl hydroxylase (23), we assigned the origin of the inverse isotope effect on NOV2 k cat to slow dissociation of metalbound aquo ligand during catalysis. Due to the Ͻ 1, the dissociation of D 2 O ligand is thermodynamically more favorable than the dissociation of H 2 O ligand, thus increasing the probability of O 2 binding to the iron center, resulting in an inverse equilibrium isotope effect on the steady-state activity. This observation establishes that for NOV2, and possibly for all CCDs, the dissociation of aquo ligand occurs during or prior to O 2 binding/activation. The isotope effect on the steady-state k cat also suggests that aquo ligand rebinds during or after each catalytic cycle to reset the system for the next round of catalysis. Finally, the isotope effect on another kinetic parameter, k cat /K m (sKIE ϭ 0.69), with a magnitude similar to that observed for k cat , further suggests that the dissociation of aquo ligand occurs prior to the first irreversible step.
Although NOV2 and CAO1 both catalyze oxidation of piceatannol, unlike NOV2, we observed a normal kinetic isotope effect on CAO1 steady-state activity (k H /k D ϭ 1.7) in an airsaturated buffer. The corresponding proton inventory demonstrated that a single proton transfer occurred in the transition state of the rate-limiting step for this enzyme. The normal KIE of approximately equal magnitude at different pH values for CAO1 demonstrates that the isotope effect is a consequence of proton transfer occurring in a chemical step rather than being associated with a difference in side-chain ionization in the two buffers. It was previously suggested that proton transfer could occur from the 4-hydroxyl stilbenoid group to side-chain proton acceptors to generate an activated substrate intermediate (14,15). In fact, crystal structures of CAO1 in complex with piceatannol or ␤-fluororesveratrol feature a short hydrogen bond (ϳ2.45-2.55 Å) formed between the stilbenoid 4-hydroxyl group and the Tyr-133 hydroxyl group, suggesting the formation of a moderately strong hydrogen bond. Lys amine groups located in the protein interior, like that of Lys-166, can have dramatically downshifted pK a values, enabling them to act as proton acceptors at near neutral pH values (31). Based on these considerations, it is possible that deuterium substitution at the stilbenoid 4-hydroxyl group and consequent elevated pK a could underlie the normal sKIE observed for CAO1. However, a pure effect on substrate pK a is inconsistent with the enzyme activity shown in Fig. 5, where the pH-activity data are vertically downshifted in D 2 Oversus H 2 O-based solvent rather than being shifted to the right as would be expected for a pure effect on the hydroxyl group acidity. This mechanism also does not account for the O 2 dependence of the KIE.
Alternatively, proton transfer could also occur from an ironbound solvent molecule or an unidentified proton donor to an activated oxygen species, allowing its temporary stabilization. Studies on other mononuclear nonheme iron enzymes demonstrate that proton transfer to activated oxygen species and reaction intermediates is critical in governing their reaction trajectories (Ref. 32 and references therein). The possibility of such a proton transfer event occurring in CCDs was suggested by prior density functional theory studies on apocarotenoid cleavage by ACO (19) and is consistent with the kinetic data we present here.
These experimental results and structural considerations indicate that the iron-bound solvent plays a crucial role in the catalytic mechanism of CCDs. Previous extended X-ray absorption fine structure data indicated that the iron-bound solvent moiety is found in an aquo state in NOV2, whereas CAO1 exhibited scattering consistent with the presence of both hydroxo and aquo ligands (14). Conversely, the XAS edge position of CAO1 is slightly blue-shifted with respect to that of NOV2, indicating a more electropositive character (less surrounding negative charge) of the iron center of the former protein (14). Our XAS studies on NOV2 and CAO1 with varying pH (6.5-8.5) indicated no alteration in the primary ligand environment of the iron center, whereas deprotonation of an Fe(II)-aquo complex would be expected to detectably alter the XAS spectrum (23). Thus, it appears that the protonation states of the first-sphere iron ligands in these CCDs remain unchanged within a pH range of 6.5-8.5. For both NOV2 and CAO1, substrate complexation did result in a subtle perturbation in the XANES spectrum, but the alteration is likely too small to be attributable to induced dissociation of the solvent molecule. This conclusion is also supported by prior 57 Fe Mössbauer spectroscopy experiments, which showed only minor spectral changes upon organic substrate complexation by NOV2 and CAO1 (14). Indeed, crystal structures of Co-CAO1, which is structurally similar to Fe-CAO1 (33), in complex with

Solvent isotope effects on alkene bond cleavage by CCDs
piceatannol demonstrate that solvent can remain bound to metal in the presence of substrate.
Our mutagenesis studies showing that KIE behavior in CAO1 can be inverted by a single substitution from Leu to Val at position 509 also support the importance of the iron-solvent unit in the catalytic mechanism of CCDs. This position contains one of the few active-site differences between CAO1 and NOV2. The Leu-509 side chain is located in close proximity to the iron-bound solvent in WT CAO1 and is predicted to interact with the aquo ligand through van der Waals interactions. The structure of L509V CAO1 demonstrates that this substitution changes the orientation of aliphatic side-chain contact with the solvent-binding site but leaves the remaining activesite structure mostly unchanged with the exception of Phe-91, which was found in a new conformation pointing away from the iron-coordinated solvent. This combination of structural changes produced an opening in the exchangeable ligand-binding site, allowing greater ligand accessibility as evidenced by the much stronger difference electron density near the iron center compared with that for WT CAO1. The L509V CAO1 density was heterogeneous, having a variable appearance in different monomers of the asymmetric unit. Although we could not assign the density to a particular molecule, its strength suggests tighter binding of ligands, including solvent, to the L509V CAO1 iron center. Such an effect could explain the inverse sKIE we observed for L509V CAO1 and, by implication, NOV2 as deuterium substitution favors dissociation of the iron-solvent bond, providing an energetically more favorable path to the transition state. Although this explanation is compelling, we cannot exclude other influences as factors responsible for the change in sKIE behavior, including changes in substrate-binding orientation. However, the extra density observed in L509V CAO1 supports the idea that the smaller Val side chain allows more room for ligand binding, such as the interaction of additional solvent molecules to the metal-coordinated aquo ligand.
The overall activity of WT CAO1, having an apparently weaker iron-solvent interaction, is not limited by a slow solvent dissociation step in the catalytic cycle, and the effects of solvent deuterium substitution manifest instead on a chemical proton transfer step, giving rise to a normal KIE. In summary, we conclude that the dissociation of solvent bound to iron and proton transfer are two chemical steps associated with the catalytic mechanism of CCDs.

Reagents and protein purification
All the chemicals used in enzyme purification and enzymatic assays were purchased from Sigma-Aldrich. LB medium was purchased from USB (Cleveland, OH). All chemicals were used without any further purification. Water from a Milli-Q purification system (resistivity of 18.2 megohm-cm) (ED Millipore, Billerica, MA) was used to prepare all reagents and buffer solutions. NOV2 protein was expressed in Escherichia coli and purified as described previously (2). The purity of the isolated protein (Ͼ95%) was established visually by SDS-PAGE followed by Coomassie Blue staining. The concentration of protein was quantified using a molar extinction coefficient of 69,786 M Ϫ1 cm Ϫ1 at 280 nm (14). ACO from Synechocystis sp. PCC6803 and CAO1 from Neurospora crassa were also purified as described previously (14,34). The variants of CAO1 (L509V, E383D, and F91L) were generated using a standard site-directed mutagenesis protocol. CAO1 L509V was purified following the protocol described for WT CAO1.

Activity assay
4-[(E)-2-(3,5-Dihydroxyphenyl)ethenyl]benzene-1,2-diol, also known as piceatannol (Sigma-Aldrich), was used as a substrate for NOV2 and CAO1. The rate of piceatannol oxidation was monitored with a UV-visible spectrophotometer (PerkinElmer Life Sciences, Lambda Bioϩ) by following the change in absorbance at 304 nm over time. The decrease in absorbance at 304 nm corresponds to the steady-state rate of piceatannol oxidation. All the assays were performed at room temperature (21-23°C) in a reaction buffer consisting of 20 mM HEPES-NaOH, pH 7.4. ACO activity assays were carried out using all-trans-8Ј-apocarotenol as a substrate as described previously (24).
The O 2 concentration dependence study was performed inside an AtmosBag (Sigma-Aldrich). The partial pressure of O 2 inside the bag was modulated by supplying an appropriate mixture of N 2 and O 2 gases using a customized gas mixer (Superflash, Cleveland, OH). The buffer consisting of 20 mM HEPES-NaOH, pH 7.4, was allowed to equilibrate with the atmosphere of the bag for 15 min with gentle stirring. Afterward, the concentration of dissolved O 2 was measured using an RDO optical dissolved O 2 sensor (Thermo Scientific, Waltham, MA). The O 2 sensor was calibrated according to manufacturer guidelines using air-saturated water. Once a stable O 2 concentration was recorded, 1 ml of the buffer was transferred to a 1.5-ml cuvette (BrandTech Scientific, Germany). An appropriate amount of piceatannol was added from a 2 mM stock prepared in 20 mM HEPES-NaOH, pH 7.2. The mixture was used to blank the UV-visible spectrometer. 20 nM NOV2 or CAO1 was finally added to initiate the reaction. The reaction was monitored by following the decrease in substrate absorbance at 304 nm. This procedure was repeated at various concentrations of O 2 .

Solvent isotope effect
20 mM HEPES buffer was prepared in H 2 O and D 2 O (99.9% D). The pH/pD of these corresponding buffers was adjusted to 7.4. pD 7.4 was obtained by adjusting the pH meter reading to 7.0 (pD ϭ pH read ϩ 0.4). 4 M NaOH was used to make the pH adjustments. Piceatannol stock was made in H 2 O and D 2 O separately in 20 mM HEPES-NaOH, and the pH/pD was adjusted to 7.4. The experiments were carried out exactly as described in the preceding section. For the D 2 O experiment, 2 M NOV2 or CAO1 stock was prepared in 20 mM HEPES, D 2 O buffer, pD 7.4, and the mixture was allowed to stand at room temperature for about 30 min before using it for activity assays. The final percentage of D in the D 2 O buffer was about 98 -99%. All solvent isotope effect reaction measurements were performed either under air or inside an AtmosBag as specified in the figure legends.

XAS sample preparation and data collection
The resting-state E-only samples at pH 7.4, 6.0, and 8.5 were prepared in atmospheric air by mixing NOV2 or CAO1 in 100 mM buffer (50 mM HEPES and 50 mM MES) with 20% glycerol (v/v). Approximately 30 min after mixing, 140 l of this mixture was transferred to the sample holder and flash frozen in liquid nitrogen. The ES complex samples were prepared under an argon atmosphere inside a glove bag to prevent any catalytic turnover. NOV2 or CAO1 (ϳ1 mM) was mixed with piceatannol (ϳ2 mM) in 20 mM HEPES, pH 7.4, containing 20% glycerol (v/v). The mixtures were transferred to the sample holder using a gas-tight syringe and flash frozen in liquid nitrogen. The E/ES complex and pH profile XAS data for NOV2 and CAO1 were collected on Stanford Synchrotron Radiation Lightsource (SSRL) 9-3 and 7-3, respectively, following previously described methods (14). The inflection point of the iron foil reference was set to 7112.0 eV. All XAS data processing and analysis were done according to the previously established protocol for CCDs (14).

Effect of pH on isotope effect
10 mM HEPES (pK a ϭ 7.5) and 10 mM MES (pK a ϭ 6.15) were used as a cobuffer system to create assay solutions at pH (pD) values of 6.5, 7.0, 7.5, 8.0, and 8.75. The molar extinction coefficient for piceatannol was measured for each pH (pD), and the calculated value was used to process the data.

Proton inventory
Solvent deuterium kinetic isotope effect experiments were performed for CAO1 by mixing 20 mM HEPES buffer in D 2 O or H 2 O (pD or pH, 7.4). A subtle difference in the molar extinction coefficient of piceatannol was observed for different mole fraction of D 2 O. The corrected molar absorption coefficient was used for the calculation. The activity (apparent k cat denoted as k n ) measured at a given mole fraction of D 2 O was divided by the apparent k cat in 100% H 2 O (k 0 ), and the ratio was plotted against the mole fraction of D 2 O (n). The plot was fit to the Gross-Butler equation for a single proton transfer (21), k n /k 0 ϭ 1 Ϫ n ϩ a ϫ n (Eq. 1) where n is the mole fraction of D 2 O and a ϭ 1/KIE.

Protein crystallization, diffraction data collection, and structure refinement
L509V CAO1 protein (ϳ35 mg/ml) and WT CAO1 (ϳ40 mg/ml) were crystallized by the hanging-drop vapor-diffusion method in a crystallization mixture containing 33% (w/v) poly-(acrylic acid 2100 sodium salt) and 28% (w/v) poly(acrylic acid 2100 sodium salt), respectively, with 0.1 M HEPES-NaOH, pH 6.0. The crystal quality was improved by introducing nucleating microseeds from a previously grown crystal of WT CAO1 in 43% (w/v) poly(acrylic acid sodium salt) and 0.1 M HEPES, pH 6.0. L509V CAO1 was also crystallized using a second crystallization mixture consisting of 100 mM imidazole/MES, pH 6.5, 30 mM MgCl 2 , 30 mM CaCl 2 , 10% (w/v) PEG 20,000, and 20% (v/v) PEG monomethyl ether 550. Rod-shaped crystals grew in about 1 week and were harvested after ϳ5-6 weeks. The isolated crystals were flash frozen directly in liquid nitrogen without the addition of any further cryoprotectant. The harvested crystals were stored in liquid nitrogen prior to X-ray exposure. Diffraction data were collected from N 2 -vapor cooled crystals at the Northeastern Collaborative Access Team (NE-CAT) beamline 24-IDE of the Advanced Photon Source or the FMX beamline (17-ID-2) at the National Synchrotron Light Source-II. Diffraction data were processed with XDS and XSCALE (35). Initial phases were obtained either by direct refinement or by molecular replacement using Phaser (36) and the atomic coordinates of a previously determined Fe-CAO1 structure (PDB accession code 5U8X) as an initial model. The structures were refined and manually edited using REFMAC (37) and Coot (38) and validated using the MolProbity (30) and wwPDB servers (39). In all cases, the N terminus of CAO1 was disordered and consequentially omitted from the model. Additionally, residues 335-351 in chain B of the second crystal form of L509V CAO1 were partially disordered and omitted from the final model.