Investigating the Structural Plasticity of a Cytochrome P450

Cytochrome P450s are heme-containing proteins that catalyze the oxidative metabolism of many physiological endogenous compounds. Because of their unique oxygen chemistry and their key role in drug and xenobiotic metabolism, particular attention has been devoted in elucidating their mechanism of substrate recognition. In this work, we analyzed the three-dimensional structures of a monomeric cytochrome P450 from Saccharopolyspora erythraea, commonly called EryK, and the binding kinetics to its physiological ligand, erythromycin D. Three different structures of EryK were obtained: two ligand-free forms and one in complex with its substrate. Analysis of the substrate-bound structure revealed the key structural determinants involved in substrate recognition and selectivity. Interestingly, the ligand-free structures of EryK suggested that the protein may explore an open and a closed conformation in the absence of substrate. In an effort to validate this hypothesis and to investigate the energetics between such alternative conformations, we performed stopped-flow absorbance experiments. Data demonstrated that EryK binds erythromycin D via a mechanism involving at least two steps. Contrary to previously characterized cytochrome P450s, analysis of double jump mixing experiments confirmed that this complex scenario arises from a pre-existing equilibrium between the open and closed subpopulations of EryK, rather than from an induced-fit type mechanism.

basis of these structures we propose a mechanism for substrate recognition and regioselectivity and we highlight the conformational changes involved in substrate binding to EryK. Furthermore, we show that the ligand-free structure of EryK may be shifted from an open to a closed state by increasing ionic strength conditions. In an effort to support these observations, we present an extensive characterization of the binding kinetics of EryK to ErD, using both single and double mixing time-resolved stopped-flow experiments. The significance of our results, in the context of previous characterization of the structure and function of P450s, will be also discussed.

Sample Preparation and X-ray Data Collection-Recombi-
nant His-tagged EryK was overexpressed in the BL21 STAR TM (DE3) Escherichia coli strain (Invitrogen) and purified as described previously (15). Purified EryK was concentrated and stored at 193 K. To assess its oligomerization state, EryK was subjected to ultracentrifugation in a Beckman Optima XL-1 instrument equipped with absorbance optics. Protein samples, measured at a concentration of 15 M in the presence of 50 mM Tris⅐HCl, pH 7.5, at 298 K, were found to be monomeric (supplemental Fig. S1).
Three crystal forms of EryK were obtained in different crystallization conditions: with His tag (His 6 -EryK) in low salt conditions, without tag (EryK) in high salt, and in complex with its substrate erythromycin D (ErD-EryK) ( Table 1). All crystallization protocols were carried out at 294 K by vapor diffusion and ErD was introduced by co-crystallization. Diffraction data were collected at 100 K at the ESRF synchrotron (Grenoble, France) beamlines ID14-3 using a marCCD detector, and ID14-2 using a ADSC Q4R detector. Data were integrated and scaled using the HKL suite (16), except for His 6 -EryK data, which were processed with MOSFLM (17) and scaled with SCALA in the CCP4 suite (18). A summary is shown in Table 1.
Structure Determination and Refinement-In all cases molecular replacement was carried out to determine the initial crystallographic phases using the program MolRep (19) from the CCP4 suite (18). The ligand-free form of His 6 -EryK was determined using EryF coordinates (20) (33.9% sequence identity, supplemental Fig. S2) as a search model (PDB code 1OXA). EryK was similarly determined by molecular replacement with His 6 -EryK coordinates as a search model. EryK coordinates were used to solve the structure of ErD-EryK. Initial atomic models were subjected to iterative rounds of building and refinement. Refinement was carried out with Refmac5 (21) in CCP4, followed by model building and manual readjustment performed in QUANTA (Accelrys Inc.). Solvent molecules were added into the F o Ϫ F c density map, contoured at 3, with the X-SOLVATE tool of QUANTA. Secondary structure assignment was performed using the Kabsch and Sander algorithm and geometrical quality of final models was assessed using PROCHECK (22). The first 14 to 18 N-terminal residues, depending on the crystal form, are missing in all models due to insufficient electron density in this region. Final statistics are shown in Table 1. All figures were produced with PyMOL.
Root mean square deviation and sequence identity among the structures of EryK, PikC, and EryF were determined using CE (23). Sequence numbering refers to EryK (SwissProt accession A4FNo_SACEN; NCBI accession YP_001102980); secondary structure elements are named following P450 convention.
Spectral Binding Titrations-Because catalysis would require the presence of external reducing agents and EryK was purified in its oxidized form, addition of ErD results in monitoring only the binding reaction (12). Thus, all the experiments reported in this work refer to the binding reaction of EryK to ErD, in the absence of catalysis. Binding affinities of ErD to EryK in different ionic strength conditions were determined (at 298 K) by titrating 2 M enzyme with the ligand, ErD, in a total volume of 800 l of 50 mM Tris⅐HCl buffer, pH 7.5, with and without 2 M NaCl. Ligand concentration covered a suitable range from 0 to 300 M. UV visible spectra (190 -820 nm) were recorded after each addition and the absorbance intensities at 420 and 390 nm were plotted against the logarithm of the added ligand concentrations (data not shown). From each of the substrate-bound spectra were subtracted the appropriate blank.
The dissociation constant (K D ) was estimated using the Kaleidagraph software package. A nonlinear regression analysis was applied using hyperbolic equation ⌬AU obs ϭ ⌬AU max [L]/(K D ϩ [L]), where ⌬AU obs is the absorbance difference, ⌬AU max is the maximum absorbance difference extrapolated to infinite ligand concentration, and [L] is the ligand analytical concentration. Data were also globally fitted with the program Prism (Graphpad).
Ligand Binding Kinetics-The time evolution of the EryK-ErD interaction was followed by monitoring the change in absorbance at 390 nm on a stopped-flow apparatus. Measurements were performed on a SX18-MV stopped-flow instrument (Applied Photophysics, Leatherhead, UK) using both symmetric and asymmetric (1:10) mixing. In the symmetric mixing mode, EryK, at a final constant concentration of 7.5 M, was mixed with ErD diluted to the desired final concentration, varying in a range from 9 to 150 M. The traces were recorded in 50 mM Tris⅐HCl, pH 7.5, at 288 K, both in the absence and presence of 2 M NaCl. In the asymmetric mixing mode, varying the NaCl concentration in a range from 0 to 3.6 M, the kinetic traces of the interaction between EryK (final constant concentration of 5.5 M) and ErD (final constant concentration of 100 M) were recorded at 288 K in 50 mM Tris⅐HCl, pH 7.5. Kinetic traces were extracted from the acquired spectra (recorded as average of at least three experiments) and analyzed using the Kaleidagraph software package.
Double Jump Measurements-To characterize the pre-equilibrium stage of the open and closed conformations of EryK, we performed double mixing measurements carried out in 50 mM Tris⅐HCl, pH 7.5, by using an Applied Photophysics SX18-MV stopped-flow instrument at 288 K. EryK (13.5 M) was incubated in 2 M NaCl for different delay times and then symmetrically mixed with ErD at a final concentration of 100 M, in 50 mM Tris⅐HCl, pH 7.5, with 2 M NaCl. The observed amplitudes, reflecting the fraction of molecules populating the open and closed conformations, were plotted as a function of delay times and fitted to a single exponential decay.
where i runs over multiple observations of the same intensity, and j runs over all crystallographically unique intensities. b R work ϭ ⌺ʈF obs ͉ Ϫ͉Fcalcʈ/⌺͉F obs ͉, where ͉F obs ͉ Ͼ 0. R free is based on 5% of the data randomly selected and is not used in the refinement.
Kinetic Models-Classically, complex kinetics for a monomeric protein, as EryK, undergoing a ligand-induced conformational change, is described by the Scheme 1. A binding event progressing through pathways 1 and 2 is representative of an induced fit model, whereby ligand binding is postulated to induce a conformational change (24), which may be transmitted to neighboring residues as shown in Scheme 2.
Under such conditions, observations of biphasic kinetics arise from the accumulation of the EryK-ErD reaction intermediate. When and if the fast phase is 1 Ͼ Ͼ 2 , the fast phase 1 is due to the initial encounter between the EryK and ErD molecules, with an observed rate constant shown in Equation 1.
Furthermore, the slow phase 2 is then expected to increase hyperbolically with increasing ErD concentration as shown in Equation 2.
In this case, 2 is expected to equal k Ϫ2 at [ErD] ϭ 0 and (k Ϫ2 ϩ Alternatively, a binding event progressing through pathways 3 and 4 assumes that two alternative conformations are in preequilibrium in the absence of the ligand, formally similar to a concerted or conformational selection model (25). Accordingly, complex kinetics may arise from a heterogeneous conformational ensemble of EryK, which may populate, in the absence of ligand, experimentally detectable populations of both EryK and EryK*.
Under such conditions, and if k Ϫ2 Ͼ Ͼ (k 1 ϩ k Ϫ1 ), the fast phase would arise from an initial population of monomers in the EryK* conformation and would occur with a rate constant as described in Equation 4.
On the other hand, the slow phase 2 is expected to decrease hyperbolically with increasing ErD concentration as shown in Equation 5.

Overall Structure
EryK displays the typical topology of previously characterized P450s (Fig. 2). This enzyme shows the highest sequence identity with EryF (33.9%), which is involved in the same biosynthetic pathway (see supplemental Fig. S2). The conserved protein core is formed by a four-helix bundle, three parallel helices (D, L, and I) and an antiparallel one (E). The prosthetic heme group is located between the distal helix I and the proximal helix L and it is bound to Cys 353 , the highly conserved fifth ligand of the heme iron. Despite a conserved overall structure, P450s display a wide range of substrate specificities (1). At a structural level, P450 substrate recognition and binding is assured by six so called "substrate recognition sites" (SRS) that line the active site (26). Although the definition of SRS becomes obsolete once the structure has been determined and precise protein-substrate contacts have been revealed, we will use in this section this nomenclature to define the amino acid topology because in EryK the elements that contact the substrate correspond to these SRS. These regions are: SRS1, formed either by the BC loop (as in EryK) or the BЈ helix (as in EryF); SRS2 formed by the C-terminal part of helix F; SRS3 formed by the N-terminal part of helix G; SRS4 formed by the central part of the distal helix I; SRS5 formed by the loop at the C-terminal end of helix K; SRS6 formed by the ␤-hairpin ␤4 (supplemental Fig. S2). Some SRSs may rearrange upon substrate binding, formally following an induced fit scenario (4).
The superposition of the crystal structures of the two ligandfree forms of EryK ( Fig. 2C), obtained at low and high salt, reveals that there are two different conformations for this protein in the unbound state, involving, among others, a movement of the BC loop of about 11 Å. Furthermore, helix G kinks at the level of Pro 192 , causing its N-terminal part to shift about 10 Å toward the BC loop. Interestingly, Pro 192 is not conserved in homologous P450. The movement of helix G also pulls the FG loop and helix F, causing the latter to unwind almost completely (Figs. 2C and 3A, and supplemental Movie 1). This reorganization, observed in the EryK structure at high ionic strength, closes the access channel for the active site of the molecule.
Although the last two C-terminal turns of helix F keep parallel to the moving part of helix G, its N-terminal part (Asp 158 -Asp 164 ) adopts a coil conformation upon active site closing. As shown in detail in Fig. 3C, the coil conformation of helix F causes the loss of a H-bond between His 243 and Ser 166 and contact between Trp 165 and Pro 192 that connect helix I to helix G through helix F in an open ligand-free EryK structure obtained at low ionic strength. Moreover, Trp 165 rotates and docks in a new position pointing toward helix I and interacts with His 243 (supplemental Movie 2). In summary, these rearrangements result in a marked reduction of the exposure of the heme distal site that identifies two different conformations to which we will refer to as "open" and "closed" depending on the position of helices F and G and the BC loop. Importantly, an open/closed transition has been previously observed in other P450s, as in the case of the Bacillus megaterium flavocytochrome P450 BM3 (CYP102) (27,28) and PikC (CYP107L1), the Streptomyces venezuelae P450 involved in C-12 hydroxylation in the pikromycin biosynthetic pathway (29).

ErD-EryK Structure
Determinants of Substrate Selectivity-We co-crystallized EryK in a complex with ErD. The well defined electron density of the substrate was unambiguously positioned in the active site (Fig. 3B). The cavity hosting the substrate involves essentially two regions: a fixed one, formed by the loop at the helix K C-terminal, the ␤-hairpins ␤4 and ␤1 (Asp 46 -Gly 49 ), and a mobile one, formed by the BC loop and the F and G helices.
A central feature of ErD recognition involves anchoring of the two sugar substituents, a desosamine on C-5 and a mycarosyl group on C-3 of the macrolactone ring (supplemental Table S1 and Fig. 3B). Desosamine interacts mainly with the loop at the helix K C-terminal, by two H-bonds with Gln 290 and Gln 292 ; moreover, Ile 392 in the ␤-hairpin ␤4 ensures the sugar clamping by van der Waals contacts. It is worth emphasizing that this is the only amino acid close to the forbidden conformation in the Ramachandran plot for the refined ErD-EryK structure, suggesting a critical structural role (30,31). All these interactions would be absent for a ligand lacking the desosamine sugar as in the case of the biosynthetic intermediate MycEB (see Fig. 1). Mycarose interacts with the mobile part of the channel establishing four crossed H-bonds with His 88 and Glu 89 on BC loop (Fig. 3B). Its position is also restricted by van der Waals interactions with the FG loop, which closes down on top of it. Substrate binding can therefore be described as landing on the fixed region followed by closure of two "arms" consisting of the BC loop and the F and G helices.
The selectivity for ErD against ErB is clearly explained by this structure. Indeed, methylation of 3Ј-OH would interfere with the complete closure of the arms to embrace the substrate, caused by clashes with Met 86 , His 88 , and Glu 89 in the BC loop.
The macrolactone ring of ErD establishes hydrophobic interactions with residues on helix I, helix K on the C-terminal, and ␤-hairpin ␤4 (Table S1). Ile 244 and Thr 245 on helix I lie in the  Tables S1 and S2. active site, and the amino acids in these topological positions are believed to be involved in catalysis, playing a role in stabilization and proton delivery to the iron-bound dioxygen species (32). Altogether packing the macrolactone ring against the heme and SRS4 is very tight, explaining why limited hydroxylation at C-12 was observed when bulky substituents were introduced in C-13 (33,34) and also why there is no flexibility allowed for substrate hydroxylation sites, as observed for PikC (29), where empty pockets in the binding site are present.
In ErD-EryK, the target of hydroxylation C-12 of the macrolactone ring is positioned 5.3 Å from the heme iron. Similar distances between the hydroxylation site and the iron have been reported for other P450s, such as EryF (Fe-C6 4.8 Å) and P450 cam (Fe-C5 4.2 Å) (35). The water molecule bound to the oxidized heme iron in the ligand-free structure of EryK disappears and a new one (Wat 102 ) is found at 2.7 Å from the metal, which is H-bonded to the substrate 11-OH group. This water occupies a position that almost coincides with the O-2 atom of dioxygen bound to P450 cam . Two more water molecules are found in the vicinity of Thr 245 , Wat 4 and Wat 38 , which in P450 cam appear upon dioxygen binding (36). Moreover, Wat 27 is also positioned as in P450 cam , hydrogen bonded to conserved Asp 362 .
Therefore, taking into account the carbonyl atom of Gly 242 , in ErD-EryK we find a water molecule chain that leads from the active site to the bulk. This network corresponds closely to that observed in the oxygen-bound form of P450 cam , with Wat 102 mimicking the bound dioxygen (supplemental Fig. S3). From this work we cannot determine the mechanism of EryK catalysis, but given the conservation of water molecules in the active site and Thr 241 , we may infer that the general catalytic mechanism proposed for bacterial P450 is conserved.
Closing and Locking Mechanism-Structural rearrangements due to substrate binding occur in the most dynamic regions of EryK (F and G helices and BC loop). Movements of these elements are not hindered from crystal contacts in open, closed, and closed-bound EryK. The ErD-EryK structure resembles the closed ligand-free form (Fig. 2, B and D), but new interactions are found that work as a molecular device that switches upon substrate binding. The movement of the BC loop in ErD-EryK is less pronounced than in the closed ligand-free form, because interactions with the mycarose hinder its full transition. Moreover, helices F and G move less on top of the substrate channel (9 Å displacement instead of 11 Å); helix G kinks at the level of Pro 192 and, together with the FG loop, adopts a conformation similar to the closed ligand-free structure (C␣ differences "ligand-free closed"/ErD-EryK ϳ 1.5 Å). Helix F in the ErD-EryK structure follows the movement of helix G, being lifted about one turn (3.7 Å) and tilted by 13°. This helix is not unwound on its N-terminal side, whereas in the ligand-free closed structure it loses the ␣ helix geometry almost completely (Figs. 2B and 3A, and supplemental Movie 3). Therefore, in the open and substrate-bound structures, helix F is present and held by H-bonds with ␤-hairpin ␤4 and helix G, whereas such interactions are absent in the ligand-free closed structure where this helix is only two turns long ( Fig. 3A and supplemental Table S2).
Helix I is slightly distorted in its central part, between Ile 239 and Thr 245 , in both ligand-free EryK structures, whereas in ErD-EryK it shows pronounced bending due to extensive interactions of the macrolactone ring with side chains of residues in its middle segment (Ala 237 , Leu 240 , Ala 241 , Ile 244 , and Thr 245 ). This bending induces flipping and reorientation of His 243 on helix I (supplemental Movie 4) that can establish a H-bond with Ser 166 , located on helix F. This interaction, lost when EryK closes without ErD, allows helix F to not unwind and to shift following helix G kinking, thus moving as a rigid unit. As a consequence of helix F repositioning, Trp 165 exerts pressure on Pro 192 on helix G. Upon this pressure Pro 192 , the hinge that allows G helix kinking is locked, preventing reopening (Fig. 3D).
Importantly, whereas helix G kinking and repositioning of mobile elements can take place also in the absence of ErD, this network of interactions (called HSWP network after His 243 , Ser 166 , Trp 165 , and Pro 192 ) is present only if the substrate is clamped in the active site. The HSWP network is the connection that couples helix G kinking to helix I bending through helix F, and acts as a connecting rod, when substrate is bound, whereas it unwinds in the ligand-free closed structure. The HSWP network involving the same residues was also found in open EryK, but with a geometry that does not induce helix kinking. Therefore, the presence of the correct substrate, ErD, is sensed at two levels: bending of helix I to recognize the presence of the macrolide ring to be hydroxylated thus locking G helix kinking, and recognition of the mycarosyl sugar by the BC loop that specifically interacts with its 3Ј-OH.
Very recently the structure of a P450 (CYP199A2) from the highly metabolically versatile Rhodopseudomonas palustris (37,38) was reported, showing a structure with a kinked G helix at a proline corresponding to Pro 192 of EryK, and no BЈ helix. This peculiar conformation of G helix is therefore adopted in more than one P450, although it is difficult to conclude whether it is coming from convergent evolution or gene transfer. We hypothesize that the presence of proline in a position topologically equivalent to Pro 192 in EryK and Pro 204 in CYP199A2 is a signature indicating the presence of a kink in the G helix to compensate for the absence of the BЈ helix.

Are the Two Ligand-free Structures a Crystallographic Artifact? The Kinetics of Binding EryK to ErD
Crystallographic studies, reported above, reveal that it is possible to obtain two different and stable conformations of ligandfree EryK under conditions that differ mainly in ionic strength of the solution. We therefore hypothesized that free EryK in solution could exist in a heterogeneous pool of stable conformations that have different abilities to bind the ligand and are selectively stabilized by ionic strength conditions. To probe the existence of different EryK conformers, we compared static UV visible spectra and circular dichroism spectra of free EryK in the presence and absence of high ionic strength buffer conditions (data not shown). Contrary to previous work on other P450 (39,40), we observed that the absorption spectrum of EryK is insensitive to the ionic strength of the solution. These observations suggest that either EryK can accumulate only one discrete conformation under the conditions investigated in this study or no significant spectroscopic changes may be associated with the conformational change described above. In an effort to validate the existence of the two alternative conformations, we performed a complete characterization of the binding properties of EryK at high and low ionic strength conditions. Upon binding to ErD, EryK exhibits the spectral changes classically observed for previously characterized cytochrome P450s (4,6,8), i.e. a shift of the Soret peak from 420 to 390 nm. These spectral changes have been previously reported (12). Equilibrium binding experiments measured in the presence and absence of 2 M NaCl display a simple hyperbolic transition yielding an apparent K D of 3.5 Ϯ 0.1 and 1.0 Ϯ 0.1 M in the absence and presence of 2 M NaCl, respectively (data not shown).
We studied the binding kinetics of EryK to ErD by following the absorbance of the heme iron at 390 nm. Experiments were performed under pseudo first-order conditions by challenging EryK, at a constant concentration of 7.5 M, with increasing concentrations of ErD. Two typical kinetic traces recorded in the presence and absence of 2 M NaCl are reported in Fig. 4. At all monitored conditions, observed kinetics displayed a systematic deviation from a single exponential function and were fitted satisfactorily to a double exponential equation, suggesting that the binding of ErD to EryK occurs via at least two steps.
Inspection of Fig. 4 clearly reveals that addition of 2 M salt affects the relative amplitudes of the observed phases. The fast phase changes from about 90% of the total signal change in the absence of salt to about 60% in its presence. In an effort to further test the effect of the NaCl concentration on the relative amplitudes of the observed phases, we performed kinetic experiments at constant EryK and ErD concentrations (5.5 and 100 M, respectively) and at different salt concentrations (supplemental Fig. S4). Interestingly, the relative amplitude of the slow phase clearly increases at higher salt concentrations.
A plot of the observed rate constants for the two phases as a function of ligand concentration are reported in Fig. 5. Overall, whereas the rate constants for the fast phase appear to increase linearly with increasing ligand concentration, the rate constants for the slow phase are essentially independent of [ErD]. Notably, a similar behavior was observed in the presence of 2 M NaCl. The calculated rate constants from analysis of the fast phases were: k on ϭ (1.05 Ϯ 0.06) ϫ 10 6 M Ϫ1 s Ϫ1 and k off ϭ 70 Ϯ 5 s Ϫ1 in the absence of salt, and k on ϭ (5.5 Ϯ 0.3) ϫ 10 5 M Ϫ1 s Ϫ1 and k off ϭ 16 Ϯ 2 s Ϫ1 in the presence of 2 M NaCl. The two experiments revealed the presence of a slow phase, 2 , independent on ErD concentration, of about 4 Ϯ 0.4 s Ϫ1 and 1.3 Ϯ 0.1 s Ϫ1 in the absence and presence of salt, respectively.
As discussed under "Materials and Methods," analysis of 2 on substrate concentration should in theory be of diagnostic value to understand which of the two possible mechanisms best describes the experimental data (41,42), i.e. a mechanism involving a pre-existing conformational heterogeneity of EryK or a mechanism involving an induced fit scenario. However, as described below, in the case of the interaction between EryK and ErD we could not obtain any clear dependence of 2 on [ErD], the observed rate constant being essentially independent on [ErD] both in the presence and absence of NaCl. This effect may be due to either (i) a low value of K D Ј , which would imply 2 to reach a constant value at relatively low [ErD] or (ii) a low value of k 2 or k Ϫ1 in Schemes 2 or 3, respectively, which would result in a 2 nearly independent on [ErD] (see Equations 2 and 5 under "Materials and Methods").
Comparison between the two models outlined above reveals that, whereas in the case of the induced fit scenario (Scheme 2) biphasic kinetics arise from accumulation of a genuine reaction intermediate (the EryK-ErD complex), in the case of the conformational selection model (Scheme 3), complex time courses may only occur when the EryK and EryK* states display a similar stability, i.e. they are both populated prior to encountering the ligand. Hence, to distinguish between the induced fit and conformational selection scenarios, we performed a double jump stopped-flow mixing experiment. Analysis of the NaCl dependence of observed kinetics reveals that the amplitude of the slow phase is increased with increasing ionic strength (supplemental Fig. S4). Following Scheme 3, it may be postulated that high NaCl concentrations shift the equilibrium toward the EryK conformer. It may be noted that the crystal structure of EryK under low and high ionic strength conditions reveals the presence of two different conformers, namely an open and a  closed state. Thus, the double jump experimental setup was designed with the specific aim of altering the relative populations of the two conformers before mixing them with the substrate. In particular, the protein was first mixed with a constant concentration of 2 M NaCl (first mix); after a controlled delay time, the reaction mixture was challenged with ErD at a constant concentration of 50 M (second mix). At all different delay times, the observed time courses were fitted to a double exponential function, which returned similar rate constants for both phases. Importantly, the observed amplitudes of the two phases display a single exponential dependence on the delay times between the first and the second mix, yielding an apparent rate constant of 1.8 Ϯ 0.1 s Ϫ1 (Fig. 6). Inspection of Fig. 6 clearly reveals that although the amplitude of the fast phase decreases with increasing delay times, the slow phase increases. This behavior is inconsistent with an induced fit model (Scheme 2). In fact, because this model postulates the existence of only one conformer prior to encountering of the ligand, it would imply that the relative amplitudes of the two phases are independent of the delay times. On the other hand, by assuming a conformational selection model (Scheme 3), data presented in Fig. 6 allow direct measurement of the observed rate constant for structural rearrangement (i.e. k 1 ϩ k Ϫ1 ϭ 1.8 Ϯ 0.1 s Ϫ1 at 2 M NaCl). Because we know the k 1 from the experiment shown in Fig. 5, the microscopic rate constant, k Ϫ1 , can then be calculated as 0.5 s Ϫ1 . The fractions of open and closed conformations from rate constants (0.7 and 0.3, respectively) agree well with those from kinetic amplitudes (supplemental Fig. S5).
We conclude that the biphasic kinetics of binding EryK to ErD arises from a conformational selection mechanism, whereby the protein pre-exists in two alternative states, where only one binds the substrate. Addition of NaCl shifts the equi-librium toward the closed state, as mirrored by the increased relative amplitude of the slow kinetic phase. Double jump mixing experiments allow (i) to exclude a kinetic mechanism involving an induced fit scenario and (ii) to monitor directly the observed rate constant of the conformational change reaction (k 1 ϩ k Ϫ1 ).

DISCUSSION
The structures of EryK highlight novel structural features, with respect to other related P450s, which are necessary to allow a complete and competent recognition and closure around the bulky substrate, a macrolactone ring carrying two sugar substituents. A relevant comparison can be carried out with respect to the structures of EryF bound to the macrolide 6-deoxyerythronolide B (20) and that of PikC, which binds a monoglycosylated macrolide, narbomycin (29). In EryF, elements BЈ helix, BC loop, and FG loop close like a lid on the substrate; this is also the case for PikC, where the extra bulk, represented by a desosamine, is accommodated by shortening of the BC loop. The presence of a second sugar moiety as in ErD exceeds this plasticity. As a result, in EryK the role of helix BЈ in closing the channel is taken by the N-terminal part of helix G, which contains a hinge (Pro 192 ) allowing a straight to kinked transition. With respect to EryF and PikC we also observed a longer FG loop, a rearrangement of helix F, and a reduction in size of the ␤ hairpin, ␤4.
On the basis of the two structures determined in the absence of substrate, we propose that both open and closed conformations are populated in solution: low salt and high salt crystallization conditions are able to shift this equilibrium, inducing the crystallization of either of the two forms. Kinetic data confirm this hypothesis because double jump experiments reveal that complex binding arises from a pre-existing conformational heterogeneity of substrate-free EryK. It may be noted that at physiological salt conditions, the closed conformer accounts for only 10% of the molecules, suggesting that biologically relevant processes are likely to involve mostly the open conformation. Yet, defining the relative stabilities of the two conformers may be significant in at least two ways. First, previous work has described the binding reaction of P450s to their ligands with an induced fit model, essentially on the basis of the observation of multiple exponentials. Although this could well be true for many P450s, such as 3A4, our experimental work on EryK shows that biphasic kinetics may also arise from a pre-existing equilibrium of protein conformers. Second, a population of about 10% implies that the free energies of the open and closed conformations are similar. Thus, in theory, it would be possible to shift the equilibrium between the two conformers, for example, via site-directed mutagenesis or ligand binding. The description of the structure and energetics of the two conformations paves the way for future work in this direction.
It is of interest to comment on the magnitude of the kinetic constants for the binding of EryK to ErD, in comparison with crystallographic data. In fact, the rate constants obtained for the encounter of the open conformation to ErD, i.e. (1.05 Ϯ 0.06) ϫ 10 6 M Ϫ1 s Ϫ1 and (5.5 Ϯ 0.3) ϫ 10 5 M Ϫ1 s Ϫ1 in the absence and presence of salt, respectively, are somewhat lower than expected from a diffusion-controlled reaction (43). This It should be noted that the crystal structure of the final complex closely resembles the structure of the closed-free state. On the basis of these observations, we then conclude that a plausible mechanism describing the binding reaction of EryK to ErD would involve an additional step implying the locking of the enzyme, as formalized in Scheme 4, where EryK and EryK* denote the closed and open conformers, respectively. Although the experimental work presented here allows inference directly in the presence of a pre-existing conformational heterogeneity of EryK (step 3 in Scheme 1), the existence of a late closure step may be invoked only indirectly from the crystallographic work, as well as from the observed slow apparent second-order rate constants. A quantitative analysis of crystallographic and kinetic data of the binding of EryK to ErD is summarized in the energy diagram in Fig. 7.
Substrate binding to P450s has been extensively studied and is still an area of major interest (5). However, despite many efforts, there are still more exceptions than rules when it comes to binding a substrate to different P450s and a unifying comprehensive mechanism is yet to be unveiled. One of the most extensively characterized cytochrome P450s is the human cytochrome P450 3A4. In particular, Guengerich and co-workers (44,45) recently reported a comprehensive analysis of the binding kinetics of this enzyme to its specific ligands and inhibitors. Contrary to what we observed for EryK, it was concluded that multiphasic kinetics of P450 3A4 is a result of sequential events of binding as opposed to the existence of multiple enzyme populations in dynamic equilibrium. One of the key findings to support this hypothesis was based on the existence of a fast phase, which is silent in absorbance, but detectable with fluorescence, possibly as a result of binding multiple ligands to a peripheral site of the enzyme. Consistently with Scheme 3, EryK does not display this absorbance-silent phase, as monitored by fluorescence with ketoconazole. 6 Another possible model, which would account for the kinetic data presented in this work, is oligomerization of the enzyme (46). However, analytical ultracentrifugation profiles of EryK clearly show that the protein is monomeric at the concentrations investigated in this work (i.e.

5-15 M).
In conclusion, we present a clear example of a monomeric P450, whose complex binding kinetics are dictated by the presence of a heterogeneous conformational ensemble between an open and a closed state. Future studies based on protein engineering will further clarify the molecular details of the recognition of EryK with its physiological substrate ErD and possibly allow the design of an enzyme that hydroxylates substrates containing a cladinose sugar. This may result in greater levels of the desired ErA and would represent an alternative method of achieving better ErA/ErB ratios to that recently achieved , open-free (blue) EryK, and the ligand-bound ErD-EryK (green) are also shown. It should be noted that, given that observed kinetics is well fitted by a double exponential process under all investigated conditions, the existence of a late closure step may only be invoked indirectly from crystallographic work, as well as from the slow apparent second-order rate constant. The microscopic rate constants, obtained in the presence of 2 M NaCl, were calculated as reported below. As described in the text, we designed a double jump experiment with the specific aim of shifting the equilibrium between the open and closed states, prior the encountering the ligand. Hence, the amplitude dependence of the two phases as a function of the delay time (Fig. 6) directly reports on the kinetics of the first step in Schemes 3 and 4, the observed rate constant k obs being equal to (k 1 ϩ k Ϫ1 ) ϭ 1.8 Ϯ 0.1 s Ϫ1 . On the other hand, by considering that the binding reaction between the open conformer of EryK and ErD occurs faster than the conformational change between the open and closed states of EryK, the rate constant for the slow phase 2 should tend to be k 1 at [ErD] 3 ϱ (see Equation 5). Data in Fig. 5 clearly reveal that 2 is independent on [ErD], suggesting that the limiting value of k 1 is reached at relatively low [ErD] and equal to 1.3 Ϯ 0.1 s Ϫ1 . Based on these observations, we estimated a value of k Ϫ1 ϭ 0.5 Ϯ 0.1 s Ϫ1 . Importantly, a similar magnitude of k 1 and k Ϫ1 are consistent with the proposed heterogeneous conformation of EryK, the equilibrium constant of the first step being equal to k 1 /k Ϫ1 .