Human Cytochrome P450 21A2, the Major Steroid 21-Hydroxylase

Background: P450 21A2 catalyzes 21-hydroxylation of both progesterone and 17α-hydroxyprogesterone, an important step in adrenal steroidogenesis. Results: A crystal structure of human P450 21A2 with progesterone can explain many functional variants. Conclusion: High kinetic deuterium isotope effects show the importance of a closely spaced site for C21 hydrogen abstraction. Significance: The structure provides insight into enzyme deficiencies in congenital adrenal hyperplasia. Cytochrome P450 (P450) 21A2 is the major steroid 21-hydroxylase, and deficiency of this enzyme is involved in ∼95% of cases of human congenital adrenal hyperplasia, a disorder of adrenal steroidogenesis. A structure of the bovine enzyme that we published previously (Zhao, B., Lei, L., Kagawa, N., Sundaramoorthy, M., Banerjee, S., Nagy, L. D., Guengerich, F. P., and Waterman, M. R. (2012) Three-dimensional structure of steroid 21-hydroxylase (cytochrome P450 21A2) with two substrates reveals locations of disease-associated variants. J. Biol. Chem. 287, 10613–10622), containing two molecules of the substrate 17α-hydroxyprogesterone, has been used as a template for understanding genetic deficiencies. We have now obtained a crystal structure of human P450 21A2 in complex with progesterone, a substrate in adrenal 21-hydroxylation. Substrate binding and release were fast for human P450 21A2 with both substrates, and pre-steady-state kinetics showed a partial burst but only with progesterone as substrate and not 17α-hydroxyprogesterone. High intermolecular non-competitive kinetic deuterium isotope effects on both kcat and kcat/Km, from 5 to 11, were observed with both substrates, indicative of rate-limiting C–H bond cleavage and suggesting that the juxtaposition of the C21 carbon in the active site is critical for efficient oxidation. The estimated rate of binding of the substrate progesterone (kon 2.4 × 107 m−1 s−1) is only ∼2-fold greater than the catalytic efficiency (kcat/Km = 1.3 × 107 m−1 s−1) with this substrate, suggesting that the rate of substrate binding may also be partially rate-limiting. The structure of the human P450 21A2-substrate complex provides direct insight into mechanistic effects of genetic variants.

bolic errors known in humans. In addition, P450 21A2 is the major autoantigen in autoimmune Addison disease (11).
Because of the relevance of P450 21A2 to human disease, there is considerable interest in identifying the molecular basis for loss of function in P450 21A2 variants (6,7). Major aspects of structure-activity relationships among the deficient variants include changes in the structure, the nature of the rate-limiting steps in catalysis, and the substrate selectivity ( Fig. 1). In particular, P450 21A2 is unusual in that its major 21-hydroxylase activities generate electron-deficient primary carbon radicals (-CH 3 3 -CH 2 ⅐ ) in the course of the reaction (12,13).
Despite the clinical interest in P450 21A2, only limited work has been done directly with the human enzyme. Bovine P450 21A2 was purified from adrenal cortex (14). Arase et al. (15) successfully expressed the bovine enzyme in Escherichia coli in 2006, and in 2012, we reported the x-ray crystal structure of modified bovine P450 21A2 (6). That structure has been used to model some of the possible effects of variants observed in clinics (6,7). Human P450 21A2 has been heterologously expressed in yeast microsomes and utilized in mutagenesis and basic kinetic isotope effect (KIE) studies (13) and also expressed in baculovirus (11) and bacterial systems (16).
In order to address several catalytic and structural issues related to the function of P450 21A2, we expressed the human enzyme in E. coli and purified the enzyme. This protein was used to obtain an x-ray crystal structure in the presence of the substrate progesterone (6) bound at the active site, with a second substrate molecule of low occupancy residing inside the access channel (in the bovine complex, the corresponding sites are fully occupied by 17␣-OH-progesterone (6)). Kinetic analyses indicate that human P450 21A2 is more catalytically efficient than the bovine enzyme and may be the most catalytically efficient mammalian P450 reported (i.e. has the highest k cat /K m value) (3). In contrast to previous reports on the bovine enzyme (17), we find that the enzyme is not limited by rates of product dissociation but, at least in large part, by the rate of C-H bond breaking (Fig. 2, step 7), as revealed by high non-competitive intermolecular KIEs.
Expression of Human P450 21A2-In the human CYP21A2 open reading frame, the region encoding the N-terminal transmembrane helix (residues 1-29) was replaced by nucleotides coding for the peptide MAKKTSSKGK (P450 2C3 N-terminal region), and the 3Ј-end was extended by 18 nucleotides encoding six histidines. The entire modified cDNA was synthesized by GenScript (Piscataway, NJ) and inserted into a pET17b expression vector (EMD Millipore, Billerica, MA). The pET-17b plasmid containing the gene encoding human P450 21A2 was co-transformed with a chaperone pGro12 plasmid into E. coli BL21-Gold DE(3)-competent cells. The cultures were used for inoculation in Terrific Broth medium with 100 g/ml ampicillin and 50 g/ml kanamycin at 37°C and 250 rpm. When an OD 600 of 1.0 was achieved (ϳ3.5 h), the shaker speed was lowered to 100 rpm, and ␦-aminolevulinic acid and arabinose were added to final concentrations of 1.0 and 27 mM, respectively. After 60 min, the temperature was lowered to 26°C, and 1.0 mM isopropyl 1-thio-␤-D-galactopyranoside was added to induce the expression of human P450 21A2. The cells were cultured for another 40 h at 26°C (shaking at 100 rpm) and harvested by centrifugation (2.5 ϫ 10 3 g, 20 min). The cells were subsequently suspended in 50 mM Tris-HCl buffer (pH 7.4) containing 250 mM sucrose and 0.5 mM EDTA (200 ml/liter of cell culture) containing 1 mg/ml lysozyme. After incubation at 4°C for 30 min, the cells were pelleted and stored at Ϫ80°C.

Human P450 21A2 Structure and Kinetics
The protein for crystallization experiments was purified with a method similar to that used for purifying bovine P450 21A2, except that no detergent was used for chromatography with an SP Fast-Flow Sepharose column (GE Healthcare). The protein was eluted with 50 mM potassium phosphate buffer (pH 7.2) containing 20% glycerol (v/v), 0.1 mM DTT, 0.1 mM EDTA, and 400 mM NaCl.
Spectral Analyses of P450 21A2-Steady-state binding experiments were carried out using an Aminco DW2/OLIS spectrophotometer (On-Line Instruments, Bogart, GA) at 23°C. In general, P450 21A2 (1.0 M, in 0.2 M potassium phosphate (pH 7.4) buffer) was titrated with increasing concentrations of ligands (in C 2 H 5 OH) in the sample cuvette, and the same amount of C 2 H 5 OH was added to the reference cuvette. The UV-visible spectra from 350 to 500 nm were recorded. The spectral change ⌬(A 390 Ϫ A 418 ), as a function of ligand concentration, was fit to a quadratic equation of the form, Ϫ (4*E*X)), using Prism software (GraphPad, Inc.).
In order to obtain more accurate K d values for the ligands with initial K d values of Ͻ1 M, titration experiments were repeated in a Cary 14/OLIS spectrophotometer using a 10-cm path length cylindrical cuvette (Starna Cells (Atascadero, CA), catalogue no. 34-Q-100, volume 25 ml). P450 21A2 (0.10 M, in 0.2 M potassium phosphate (pH 7.4) buffer) was titrated with increasing concentrations of ligands (dissolved in C 2 H 5 OH). The spectral change ⌬(A 390 Ϫ A 418 ), as a function of ligand concentration, was fit to a quadratic equation (see above) using GraphPad Prism software.
An OLIS RSM-1000 instrument (1.24-mm slits, 20-mm path length) was used to study pre-steady-state kinetics of ligand binding to P450 21A2. Experiments were performed in the rapid scanning mode at 23°C in 0.2 M potassium phosphate buffer (pH 7.4) containing 60 M L-␣-1,2-dilauroyl-sn-glycero-3-phosphocholine. Briefly, 2 or 4 M P450 was mixed with 4 M ligand, and spectral data were collected for 1 s. The traces at 390 nm were fit to equations using the manufacturer's software (GlobalWorks). The basis of the experimental design is described by Daniels and Alberty (18). A second-order reaction 2A 3 A 2 is kinetically equivalent to A ϩ B 3 AB, if similar concentrations of A and B are mixed; the analysis is straightforward, especially if the back-reaction is slow (AB 3 A ϩ B), which was the case here. This approach was used instead of trying to achieve pseudo-first-order kinetics with excess substrate, because of the low K d values (i.e. K d Ͻ 1-2 M, which is a minimum for detection of changes in the kinetic binding experiments). Experiments with higher concentrations of substrate did not result in faster rates, in that the enzyme was already saturated with substrate.
Pre-steady-state Kinetics of Product Formation-Pre-steadystate 21-hydroxylation reactions with P450 21A2 were carried out in a KinTek instrument (model RQF-3, KinTek Corp., State College, PA). All reactions were performed in 100 mM potassium phosphate buffer (pH 7.4) at 37°C. Human P450 21A2 (2 M) was reconstituted with 4 M recombinant rat NADPH-P450 reductase (19), 15 M L-␣-1,2-dilauroyl-snglycero-3-phosphocholine, and either 20 M progesterone ([4-14 C]progesterone) or 17␣-OH-progesterone (17␣-OH-[1,2,6, 7-3 H]progesterone). The reactions were initiated by rapidly mixing with an NADPH-regenerating system (20 mM glucose 6-phosphate, 2 mM NADPH, and 2 IU/ml glucose-6-phosphate dehydrogenase (20)) and quenched with 1 M HCl as a function of time from 10 ms to 1 s. A 50-l aliquot of each reaction mixture was spotted onto the loading zone of a channeled silica gel G TLC plate (Analtech, Newark, DE), and the plates were developed with ethyl acetate/CH 2 Cl 2 (1:4 (v/v) for progesterone and 1:3 (v/v) for 17␣-OH-progesterone). With progesterone as the substrate, the plate was exposed to a phosphor imager screen (Imaging Screen K, Bio-Rad) for 7 days and analyzed with a molecular imager system (Pharos FX Plus Molecular Imager, Bio-Rad) using the manufacturer's Quantity One software. With 17␣-OH-progesterone as the substrate, the plate was scanned with an AR-2000 Radio-TLC imaging scanner (Bioscan, Poway, CA) and analyzed with the manufacturer's Winscan software. The data points were fit to a burst equation, y ϭ A(1 Ϫ e Ϫkpt ) ϩ k ss t, where A is the burst amplitude, k p is the first-order rate of the pre-steady-state burst, and k ss is the steady-state rate, using GraphPad Prism software.
Crystallization-To obtain the human P450 21A2-progesterone complex, substrate (dissolved in C 2 H 5 OH) was added in a 1:5 (P450/progesterone) molar ratio, followed by mixing and concentrating to 25 mg/ml in 50 mM potassium phosphate buffer (pH 7.2) containing 5% glycerol (v/v), 0.1 mM DTT, 0.1 mM EDTA, and 100 mM NaCl. Aliquots of the complex solution were frozen in liquid N 2 and then stored at Ϫ80°C. Human P450 21A2-progesterone complex crystals were grown using the sitting drop vapor diffusion technique by mixing equal volumes (200 nl) of complex solution and mother liquor (0.10 M sodium HEPES (pH 7.5) containing 0.20 M ammonium sulfate and 25% (w/v) polyethylene glycol 3350) and equilibrating droplets against 60 l of reservoir solution at 20°C. Microcrystals were obtained within 2 weeks, and in order to increase their size, one round of microseeding was performed. Wells containing microcrystals were washed with 10 l of reservoir buffer and transferred to a microcentrifuge tube with 40 l of mother liquor. After vortex mixing the seeds for 2 min and diluting 1:1000 with mother liquor, 200 nl of the seed solution was mixed with 200 nl of protein for each drop. Larger crystals of the complex grew from microseeded droplets in 5 days. Crystals were mounted in nylon loops and swiped through a droplet of 25% glycerol (v/v) in mother liquor for flash freezing in liquid nitrogen.
X-ray Data Collection, Structure Determination, and Refinement-Diffraction data were collected on the insertion device beam line (21-ID-G) of the Life Sciences Collaborative Access Team, located at Sector 21 of the Advanced Photon Source, Argonne National Laboratory (Argonne, IL) at a wavelength of 0.9782 Å using a Mar300 CCD detector (at 100 K). Processing was done with the program HKL2000 (21), and selected crystal data and refinement statistics are summarized in Table 2. The structure of the P450 21A2-progesterone complex was deter-mined by molecular replacement with the program MOLREP (22) (23), using the crystallographic coordinates of bovine P450 21A2 as the search model (6) (Protein Data Bank code 3QZ1). Refinement was carried out with the program Refmac5 (23,24), and manual rebuilding was performed with Coot (25). The r.m.s.d. values from the superimposition were calculated with the Superpose routine in CCP4 (23). All structural figures were generated with the program UCSF Chimera (26).

Purification of Human P450 21A2 and Catalytic Properties-
To facilitate the expression of human P450 21A2 in E. coli, the N terminus was replaced with the sequence MAKKTSSKGK to increase the solubility of this enzyme; this sequence has been utilized to improve heterologous expression of several P450s (27). P450 21A2 was stabilized in high ionic strength buffers; thus, at least 0.3 M NaCl was added in the buffers during purification. The protein was efficiently purified to Ͼ90% electrophoretic purity after one step of Ni 2ϩ -nitrilotriacetic acid affinity chromatography. The Fe 2ϩ ⅐CO versus Fe 2ϩ -reduced difference spectra of purified human P450 21A2 displayed a typical P450 spectrum with a Soret peak at 450 nm (Fig. 3B). Only a small 420 nm peak was observed in the assays (when 1 M safranin T was added as a mediator to increase the rate of reduction (29)), indicating that the purified human P450 21A2 is very low in the presence of the inactive form (cytochrome P420) (Ͻ5%).
Steady-state Kinetics of Hydroxylation Reactions Catalyzed by Human and Bovine P450 21A2-Steady-state kinetic parameters for both progesterone and 17␣-OH-progesterone 21-hydroxylation by purified human and bovine P450 21A2 were determined ( Fig. 4 and Table 1). The bovine P450 21A2 C3B21RA mutant, designed for crystallization of bovine P450 21A2, has a similar functional activity as the wild-type protein and was studied for direct comparison. Bovine P450 21A2 had comparable catalytic efficiency (ϳ90 min Ϫ1 M Ϫ1 ) for the substrates progesterone and 17␣-OH-progesterone, with similar  Table 2). The catalytic activities determined for this mutant were significantly higher than those reported previously (6), probably due to the lack of residual detergent (Cymal 5) used in that work. In particular, for progesterone, the k cat value is about 3-fold higher, and the K m value is about 5-fold lower, yielding a 15-fold increase in the catalytic efficiency compared with the previous kinetic data reported for the purified bovine P450 21A2 C3B21RA mutant (6). Human P450 21A2 is a more efficient enzyme than bovine P450 21A2, with catalytic efficiencies of 800 and 160 M Ϫ1 min Ϫ1 for progesterone and 17␣-OH-progesterone, respectively (i.e. 1.3 ϫ 10 7 and 2.7 ϫ 10 6 M Ϫ1 s Ϫ1 , respectively). Although the k cat values were similar with both substrates, the 7-fold lower K m value with the substrate progesterone resulted in much higher catalytic efficiency than with the substrate 17␣-OH-progesterone. Compared with other previous studies (11,30), the significantly higher enzymatic activities of human P450 21A2 reported herein are probably due to the fact that the purified enzyme was utilized in the assays rather than whole cells or cell lysates containing expressed human P450 21A2.
Binding of Substrates to Human P450 21A2-The binding of the substrates progesterone and 17␣-OH-progesterone to P450 21A2 produces a classic Type I binding spectrum (Fig. 5), indicating the effects of substrate binding on the heme iron spin state equilibrium. High affinities of P450 21A2 for the steroid substrates progesterone and 17␣-OH-progesterone were observed, with K d values of ϳ10 Ϯ 3 and 30 Ϯ 3 nM, respectively. In order to estimate the rate constants (k on ) for substrate binding, rates of spectral changes at 390 nm were monitored using stopped-flow spectroscopy, and the k on values were determined by fitting. The estimated k on values were 2.4 ϫ 10 7 and 2.2 ϫ 10 7 M Ϫ1 s Ϫ1 for progesterone and 17␣-OH-progesterone, respectively. The dissociation rate constants (k off ) were estimated using the relationship K d ϭ k off /k on and were 0.24 and 0.66 s Ϫ1 for progesterone and 17␣-OH-progesterone, respectively. These studies indicated that substrate binding and release are relatively fast for P450 21A2 with both substrates. The monophasic nature of the binding of both substrates differs from the apparent biphasic binding of 17␣-OH-progesterone to bovine P450 21A2 (6).
Binding of Products to Human P450 21A2-Titration of P450 21A2 with the 21-OH products deoxycorticosterone and 11-deoxycortisol also yielded Type I binding spectra (Fig. 6). The K d values were 1.2 and 2.9 M for deoxycorticosterone and 11-deoxycortisol, respectively, which are 2 orders of magnitude higher than those for the substrates. Pre-steady-state binding studies using stopped-flow spectroscopy yielded k on values of 2.2 ϫ 10 7 and 1.9 ϫ 10 7 M Ϫ1 s Ϫ1 for deoxycorticosterone and 11-deoxycortisol, respectively, which are comparable with the rate constants for binding of substrates. The calculated k off rates were 26 s Ϫ1 for deoxycorticosterone and 55 s Ϫ1 for 11-de-    oxycortisol. The lower affinities of products for P450 21A2 are probably due to fast product dissociation from P450 21A2, as is the case with most proteins (31). Analysis of Burst Kinetics-In order to determine whether steps following product formation are rate-limiting for 21-hydroxylation by P450 21A2, pre-steady-state kinetic analysis was performed (Fig. 7). Product formation was monitored over the range of 10 ms to 1 s using a rapid quench flow instrument. A small kinetic burst within the initial 80 ms (20%) was followed by a slower product formation with the substrate progesterone, whereas an almost linear curve was observed with 17␣-OHprogesterone as the substrate.
Kinetic Isotope Effects-In order to determine the kinetic contribution of step 7 (C-H bond breaking; Fig. 2) in the catalytic cycle of P450 21A2, KIE approaches were used to analyze the rate-limiting nature of this step. The steady-state kinetic parameters were determined with the deuterium-labeled substrates 21,21,21-d 3 -progesterone and 17␣-OH-21,21,21-d 3progesterone. Comparison of the k cat and K m values between non-deuterated (d 0 ) and deuterated (d 3 ) substrates yielded non-competitive intermolecular KIEs, designated D V and D (V/K) following the convention of Northrop (32). D V and D (V/K) values determined for progesterone were 6.5 and 11, respectively. D V and D (V/K) values determined for 17␣-OHprogesterone were 8.0 and 5.3, respectively ( Fig. 8 and Table 1). The relatively high non-competitive intermolecular isotope effects indicate that C-H bond cleavage is at least partially ratelimiting in both 21-hydroxylation reactions catalyzed by human P450 21A2.
Crystal Structure of Human P450 21A2-The structure of the complex between human P450 21A2 and progesterone was phased by molecular replacement using the bovine P450 21A2 protein portion from the previously determined structure of the complex of the latter with 17␣-OH-progesterone as the search model (6). The monoclinic crystal of human P450 21A2 contained three molecules of the complex with progesterone per asymmetric unit, and the structure was refined to a resolution of 2.64 Å (Table 3 and Fig. 9A). The conformations adopted by the independent complexes exhibit close similarity (Fig. 9B), with 440, 440, and 442 amino acids traced for complexes 1, 2, and 3, respectively, in the electron density map. The sequence identity between the human and bovine P450 21A2 proteins is 80% (Fig. 10A). However, the initial model of the human P450 21A2⅐progesterone complex based on the molecular replacement solution obtained from the search with the coordinates of the 3 Å resolution structure of the bovine P450 21A2⅐17␣-OHprogesterone complex required extensive rebuilding to arrive The overall structure adopted by human P450 21A2 exhibits the typical P450 fold that consists of at least 12 ␣-helices and a ␤-sheet domain near the N terminus; it resembles a triangular prism (34). Human P450 21A2 features 13 ␣-helices (treating B and BЈ as well as F and FЈ as single helices; Fig. 9C) and a total of nine ␤-strands. The latter fold into an N-terminal ␤-sheet domain (␤1-␤3 (blue) and ␤4 -␤7 (yellow) ; Fig. 9C) and a C-terminal sheet (␤8 (orange) and ␤9 (red); Fig. 9C). Thus, human and bovine P450 21A2 exhibit very similar secondary structure arrangements (Fig. 11A), and differences in the distribution of ␣-helices and ␤-strands are most likely not at the origin of the considerable conformational differences indicated by the above r.m.s.d. values. Rather, the relative orientations of helices and sheets as well as deviating conformations in the extensive loop regions are probably more important in this regard.  Inspection of the progesterone substrates at the active sites of the three human P450 21A2 complexes shows a roughly 45°o rientation of the long axis of the substrate relative to the heme plane. The spacing between heme and substrate is quite large in all cases, with the average distance between the progesterone C21 atom and Fe 3ϩ in the heme prosthetic group amounting to 4.15 Å; Fig. 11B). However, if one assumes a perferryl state with a single oxygen wedged between iron and the progesterone carbon atom (Fe 4ϩ -O⅐⅐⅐C21), the Fe⅐⅐⅐C distance is reduced to ϳ3.0 Å, and that between the iron and the H21(C21) atom is reduced to ϳ2.3 Å. At the other end of the substrate molecule, the carbonyl group (O3) is engaged in a hydrogen bond with the guanidino moiety of Arg-234 in the G helix (average distance of 2.66 Å between O3 and NH 2 ), which protrudes from the ceiling of the active site (Fig. 11C). The inclination between the plane defined by the core rings of the progesterone substrate and the heme plane is ϳ45°. The substrate cavity wall facing the ␤-face interacts primarily with the I-helix, via the hydrophobic side chain atoms of Val-287 and Ile-291 as well as the Gly-292 residue (Fig. 11, B and C). The ␣-face does not make any noticeable contacts with nearby loops between helices BЈ and C, helix K and ␤4, and between the ␤8 and ␤9 strands. The C19 methyl group of progesterone lies close to Trp-202 from the F-helix and probably engages in an H⅐⅐⅐ type interaction. Other noteworthy residues located in the vicinity of the substrate are Asp-107, with its carboxylate moiety residing at ϳ3.90 Å from the C6 atom of progesterone, Asp-288 from helix I (average distances of 3.34 Å from C6 and 3.36 Å from C7), and Ser-109, with C␥ located 3.37 Å from C7 of progesterone (Fig. 11B).
Comparison between the Structures of the Human P450 21A2⅐Progesterone and Bovine P450 21A2⅐17␣-OH-Progesterone Complexes-The superimposition of the overall structures of human and bovine P450 21A2 reveals conformational similarities (Fig. 11A), consistent with the high sequence identity (Fig. 10A), but also clear differences. However, there are some notable differences as far as the substrate binding sites are concerned. Thus, in the case of bovine P450 21A2, two distinct 17␣-OH-progesterone molecules are visible in all four independent complex molecules, with the proximal one located close to the heme and the distal one residing in the substrate access channel, located between the FЈ-helix and the ␤4/␤7 sheet. In the human P450 21A2 complexes, only the proximal site is fully occupied by progesterone at first observation. However, inspection of the electron density map at a reduced threshold in the region of the human P450 21A2 structure that corresponds to the distal 17␣-OH-progesterone binding site in the crystal structure of the bovine P450 21A2 complex reveals density that is consistent with a progesterone molecule of partial occupancy (Fig. 11D). Further, in the structure of the bovine P450 21A2 complex, accommodation of 17␣-OH-progesterone leads to a movement of the I-helix relative to the structure of human P450 21A2⅐progesterone complexes. This helical shift is stabilized by an OH⅐⅐⅐ interaction between the Thr-177 O␥H and the Phe-288 ring (Fig. 11C). In the 17␣-OH-progesterone complexes with bovine P450 21A2, the 17␣-OH group makes a hydrogen bond with the carbonyl moiety of Gly-290 (average distance 3.37 Å). However, without a structure of 17␣-OH-progesterone in complex with human P450 21A2 available, it is difficult to assess exactly how the human enzyme accommodates 17␣-OH-progesterone. An overlay of the active sites in the human P450 21A2⅐progesterone and bovine P450 21A2⅐17␣-OH-progesterone complex is depicted in Fig. 11B. As can be seen, the majority of residues are conserved, and only a few are different: Gly-468 (bovine) versus Val-470 (human); Ser-97 (bovine) versus Thr-97 (human); and Met-197 (bovine) versus Leu-199 (human) (Fig. 10A). Interestingly, all residues interacting with the porphyrin moiety are identical except for one residue (Ile-108 (bovine) versus Leu-108 (human)). Another residue close to the C11 atom of the substrate C ring is Met-197 (bovine) (average distance 4.04 Å); Leu-199(human) features an average distance of 3.78 Å.
A further difference seen in the structures of the complexes of human and bovine P450 21A2 concerns the acetyl group of progesterone and 17␣-OH-progesterone. In the human P450 21A2⅐progesterone complex, the acetyl carbon is pointing toward the heme moiety, whereas in the bovine P450 21A2⅐17␣-OH-progesterone complex, the carbonyl group is pointing away from the heme (Fig. 11B). Acetyl groups in steroid molecules can rotate more or less freely, and both orientations are possible in principle. Another factor contributing to the different orientations of the acetyl groups could be constituted by the interaction between the progesterone O20 atom and a methyl group from Val-470 (average distance 3.14 Å).

Discussion
Congenital adrenal hyperplasia is one of the most common heritable metabolic diseases in humans, and Ͼ95% of cases are attributed to mutations in P450 21A2 (1,2). Much of the literature involving P450 21A2 biochemistry used enzymes from other species, and this appears to be the first reported purification of human P450 21A2. We previously published a crystal structure of bovine P450 21A2 with its substrate 17␣-OH-progesterone (6), but this is apparently the first structure of a P450 21A2 enzyme with the substrate progesterone. The human P450 21A2-progesterone structure has single ligand occupancy and differs from the bovine structure, which had two molecules of the substrate 17␣-OH-progesterone bound (6). We also report kinetic characterization of the human P450 21A2 enzyme and conclude that the rate-limiting step(s) differs from that reported for bovine P450 21A2 by another group (17).
The catalytic activities of our purified human P450 21A2 enzyme were higher than reported for the bovine enzyme, which had been used in some previous kinetic studies (14,15,17). Some of the differences may be due to the enzyme reconstitution conditions, in that previous studies included relatively high concentrations of detergents and glycerol, and our own studies with the bovine enzyme yielded higher rates (see above; Table 2), but not as high as for the human enzyme. The k cat values for progesterone and 17␣-OH-progesterone 21-hydroxylation were similar (Table 1), but due to a lower K m , the reaction was more efficient with progesterone. The catalytic efficiency (for progesterone 21-hydroxylation) appears to be the highest reported for any mammalian P450 enzyme (ϳ1.3 ϫ 10 7 M Ϫ1 s Ϫ1 at 37°C), which compares favorably with the apparent k on rate estimated for the reaction (at FIGURE 11. Comparison between the crystal structures of the human P450 21A2⅐progesterone and bovine P450 21A2⅐17␣-OH-progesterone complexes. A, overall view of the superimposed complexes (human P450 21A2 and bovine P450 21A2 are colored in green and pink, respectively). The overlay was generated using the Matchmaker option in UCSF Chimera (26). B, close-up view of the active sites in the superimposed complexes. Selected amino acid side chains in the two structures are depicted and labeled, with carbon, oxygen, nitrogen, and iron atoms colored red, blue, and beige, respectively (in green for human P450 21A2 and pink for bovine P450 21A2). Selected hydrogen bonds are indicated with dashed lines and distances in Å. C, significant deviations in the conformation of the central region of helix I in the human and bovine P450 21A2 structures. D, differences in the conformation of the loop region connecting helices FЈ and G in the human and bovine P450 21A2 structures and evidence for the binding of a second progesterone molecule at a distal site in the structure of the human protein complex. Fourier 2F o Ϫ F c sum electron density at the 0.7 level (green meshwork) in the human P450 21A2 structure is consistent with the presence of a partially occupied progesterone molecule at virtually the same location where 17␣-OH-progesterone (pink wire diagram) resided in the previously determined structure of the bovine P450 21A2 complex (6).
Because of the high catalytic efficiency of human P450 21A2, we were interested in what the rate-limiting step(s) in catalysis is. Substrate binding was relatively fast, calculated to be 2.4 ϫ 10 7 M Ϫ1 s Ϫ1 (at 23°C) (Fig. 5, C and D). If we consider the K d values of the substrates and products (Fig. 5, A and B) and assume that the k on rates are similar for the products, the k off values for the products 11-deoxycorticosterone and 11-deoxycortisol are 40 -60 s Ϫ1 (at 23°C), which is not very rate-limiting for steady-state reactions with rates of 3 s Ϫ1 (Fig. 4).
The KIE studies provide strong evidence that C-H bond breaking (step 7 in Fig. 2) is rate-limiting. A non-competitive intermolecular KIE provides the most direct evaluation of the extent of rate limitation of the C-H bond-breaking step in an enzyme reaction (32,36). The D V and D (V/K) values were high for the P450 21A2 reactions (5-11) ( Fig. 8 and Table 1), strongly indicating that C-H bond breaking is at least partially ratelimiting. Yoshimoto et al. (13) also reported significant KIE values for human P450 21A2 using other KIE experimental designs, and the results are consistent with ours.
The high KIEs and rapid product k off rates are not consonant with the conclusions of Kominami et al. (17) with bovine P450 21A2, who concluded that product dissociation was rate-limiting in the case of progesterone 21-hydroxylation but not 17␣-OH-progesterone. In contrast, we found high KIEs for both steroid hydroxylations with human P450 21A2. The conclusion that product (11-deoxycorticosterone) dissociation is rate-limiting in the progesterone 21-hydroxylation reaction was based largely on the appearance of a burst in the pre-steady-state reaction (17) (the reaction with 17␣-OH-progesterone did not show a burst). We also observed a partial burst in the progesterone reaction (Fig. 7A). The partial (20%) burst (cf. 40% for bovine P450 21A2 in the report of Kominami et al. (17)) has several possible explanations. One is that only part of the P450 21A2 is catalytically active, although this scenario would not explain the complete absence of a burst with the 17␣-OH-progesterone reaction (17) (Fig. 7B). The possibility can be considered that the product release is only partially rate-limiting, although modeling with the program DynaFit (37) did not yield partial bursts when k off was systematically lowered (results not presented). Another possibility is one that we established for DNA polymerases (38). The appearance of a quantitative burst is evidence that the rate-limiting step occurs after product formation, but a reversible equilibrium of an enzyme-substrate complex between an active and an alternate form can explain a partial burst (38,39), and this proved to be the case in DynaFit kinetic modeling (results not presented). Such an explanation is possible here, in that it is known that the enzyme also forms an alternate product from progesterone, 16␣-OH-progesterone (this reaction (i.e. 16␣-hydroxylation) does not occur with the substrate 17␣-OH-progesterone (13)). Thus, one can explain the partial burst with an alternate conformation of the enzyme-substrate complex (the "16␣-OH" form) in equilibrium. The equilibrium could be estimated by kinetic modeling (38) but the equilibrium may not be reflected in the comparison of the ratio of the 21-and 16␣-hydroxylation ratios.
The three-dimensional overlay of the human and bovine P450 21A2 complex structures revealed that the two proteins exhibit significant deviations even in their main chain conformations despite the high sequence identity (Figs. 10 and 11). Some of these are accrued within secondary structure elements, as established for a section of helix I (Fig. 11C). It should not come as a surprise, however, that the most noticeable conformational differences occur in loop regions. A good case in point is offered by the loop connecting the FЈ and G helices that maps to the surface of the protein (Fig. 11D). The FЈ helix lies adjacent to the distal binding site occupied by 17␣-OH-progesterone in the structure of bovine P450 21A2, and together with the ␤4/␤7 sheet, the loop connecting ␤2 and ␤3 as well as a portion of the loop between helices BЈ and C, forms the outer section of the substrate access channel. The loop leading from helix FЈ to helix G features a short curl and makes a tighter turn between residues Val-212 and Arg-217 in the structure of human P450 21A2 compared with the bovine protein (Fig. 12A). The tighter turn is caused in part by a sulfate ion that stitches together the side chains of Arg-217, Arg-103, and Lys-99. The orientations of side chains from the corresponding residues Arg-215, Gln-103, and Lys-99, respectively, in the structure of the bovine protein are substantially different, and it is unlikely that these could shape the loop conformation in bovine P450 21A2 to resemble the one in human P450 21A2, even if a sulfate ion were present and could be trapped between Arg-215 and Lys-99. Although the r.m.s.d. values for main chain atoms discussed above offer a measure for the distinct geometries adopted by the human and bovine P450 21A2 proteins, the differences are even more obvious at the level of amino acid side chains, as demonstrated by the loop region between the FЈ and G helices. Differences between the conformations of the loop region connecting helices FЈ and G in the two structures (Fig. 12A) are not the result of deviating crystal packing interactions. Both space groups and packing interactions differ between the human and bovine P450 21A2 crystal structures. Bovine P450 21A2 crystals are of space group P2 1 with four molecules per asymmetric unit. Helices F, G, and J are involved in crystal packing interactions. In the case of human P450 21A2 crystals, the space group is C2 with three molecules per asymmetric unit. Crystal packing interactions are primarily mediated by helices B, BЈ, C, J, and M. The loop region between helices FЈ and G is not involved in packing in either the crystal of human P450 21A2 or that of the bovine protein. Instead, residues from loops participating in lattice interactions in the former include 35-36, 66 -67, 213-223, 247-252, 380 -382, 412-414, and terminal residues of helices F (position 189) and H (position 264).
With the human P450 21A2 crystal structure now in hand, it is interesting to evaluate the structural consequences of some selected missense mutations in human P450 21A2 that cause complete loss of function in some cases and congenital adrenal hyperplasia, resulting in salt-wasting disease, or steep reductions in enzyme activity and simple virilizing disease. An example of the former is constituted by the E352K mutation located in helix K (7). Glu-352 is part of the so-called ERR triad together with Arg-355 and Arg-409, the latter mapping to the meander region leading up to helix M (Fig. 12B). Using a model of human P450 21A2 based on the crystal structure of the bovine protein, it was predicted that the E352K mutation would disrupt the triad as a result of repulsions between the three basic side chains (7). The crystal structure of human P450 21A2 indeed confirms this view in that the ERR triad seen in the bovine protein is maintained and that replacement of the Glu-352 by a lysine would disrupt the salt bridge to Arg-409 (Fig. 12B). Another salt-wasting disease-causing mutation, R409H (7), can also be expected to disrupt the ERR triad, because the shorter histidine side chain might not be able to reach the carboxylate moiety of Glu-352 (Fig. 12B).
Although humanized models of P450 21A2 (6, 7) based on the crystal structure of the bovine protein (6) may be helpful in some cases for interpreting changes at the structural level as a result of disease-causing mutations in patients (e.g. those affecting the above ERR triad), clear differences between the human and bovine 21A2 crystal structures set limitations for the usefulness of a computational model. Thus, the V305D mutation in helix I, which is implicated in S-channel gating (simple virilizing phenotype), was interpreted to result in a salt bridge to Lys-473 based on the modeled structure of human 21A2 (7). However, Lys-473 in the crystal structure of human P450 21A2 and the corresponding Lys-472 in the structure of bovine P450 21A2 adopt very different orientations relative to Val-305 (Glu-305 in the mutant) and Val-303, respectively (Fig. 12C). Lys-473 in human P450 21A2 engages in a salt bridge to Glu-295 located further up in helix I. The same glutamate also forms a hydrogen bond to Asn-299 (Fig. 12C). The amino nitrogen of Lys-473 lies at a distance of 13.9 Å from the C ␤ atom of the side chain of Val-305, a gap that obviously cannot be bridged even if valine is mutated to glutamate. Conversely, in the structure of bovine P450 21A2, the corresponding distance between Lys-472 and Val-303 is 8.6 Å, a gap that could conceivably be bridged if valine were replaced by glutamate and the lysine side chain unraveled to assume a more extended conformation. In the crystal structure of bovine P450 21A2, methylene groups of the lysine side chain stack onto Trp-303 (Fig. 12C). In the structure of human P450 21A2, it is the side chain of Met-474 that snakes along the plane of Trp-303 (Fig. 12C).
Two additional salt-wasting mutations in the N-terminal region, G65E (40) and G91V (41), that reduce P450 21A2 activity to Ͻ1%, 6 can now be readily interpreted in light of the human P450 21A2 crystal structure. Gly-65 maps to a hydrophobic patch, and mutation to a glutamate would disrupt hydrophobic interactions and result in a direct clash with the side chain of Val-212 (Fig. 13A). The mutation can therefore be expected to destabilize the protein and alter the local conformation considerably. The opposite scenario, namely disturbing a largely negatively polarized region by introducing a bulky hydrophobic moiety, would play out in the case of the G91V mutation (Fig. 13B). Thus, in the mutant, the valine side chain would be surrounded by main chain keto groups as well as 6 C. Wang and F. P. Guengerich, unpublished data. 21A2. The corresponding residues in bovine P450 21A2 are Arg-215, Lys-99, and Gln-103, respectively, which exhibit completely different relative orientations compared with the situation in the structure of the human protein. B, view of a portion of an extended loop (meander) region connecting helices L and M, with Arg-409 in human P450 21A2 engaging in interactions with main chain oxygen atoms of amino acids Glu-400 and Arg-401 as well as the carboxylate moiety of the Glu-352 side chain from the K helix. A further amino acid in close vicinity of Arg-409 is Arg-355. The R409H and R352K mutations (7) as well as the R409C and R409L mutations (6, 44 -47) could disrupt the ERR triad. C, view of interactions between I helix residue Glu-295 and the side chains of the adjacent Asn-299 and Lys-473 located in the ␤8-␤9 loop. The V305D mutation was predicted to result in a salt bridge between Asp-305 and Lys-473 (7), but the structure of the human P450 21A2 complex reveals that the two residues are spaced too far apart. hydrophilic or negatively charged side chains from threonine or aspartate, respectively. Moreover, the carboxylate moiety of the latter residue, Asp-112, would clash with the Val-91 side chain.
Our study provides a complete kinetic framework for human P450 21A2 and the first crystal structure of the human enzyme with bound substrate. The kinetic data demonstrate that human P450 21A2 catalyzes 21-hydroxylation of both progesterone and 17␣-OH-progesterone with higher rates than the bovine enzyme and that human P450 21A2 is among the fastest mammalian P450 enzymes if not the fastest (3). The crystal structures of human P450 21A2 in complex with progesterone and the bovine enzyme in complex with 17␣-OH-progesterone exhibit significant differences. Importantly, these are not limited to loop regions and areas closer to the surface but are also found within secondary structure elements conserved in P450 enzymes. Could some of these conformational differences be the result of two 17␣-OH-progesterone molecules being bound to bovine P450 21A2 but only one progesterone to human P450 21A2? The electron density computed at a lower threshold for the latter structure revealed a second progesterone of lower occupancy that is located at the distal site in the substrate access tunnel (Fig. 11D), where the second 17␣-OH-progesterone molecule is lodged in the structure of the bovine P450 21A2 complex. Thus, the differences in the binding modes of progesterone and 17␣-OH-progesterone in the structures of the human and bovine P450 21A2 protein complexes are more likely reflective of deviating binding affinities for the substrates and different crystallization conditions. Although the human and bovine P450 21A2 enzymes share 80% sequence identity, the conformational differences in the main chains and orientations of side chains in the two structures are most likely a reflection of intrinsic differences between the two proteins rather than a consequence of the crystallizations with different substrates. This conclusion is also supported by very similar binding modes of progesterone and 17␣-OH-progesterone at the proximal sites in the human and bovine complexes (Fig. 11, B and C).
The crystal structure of human P450 21A2 now enables insight into the structural underpinnings of the phenotypes exhibited by the numerous mutations identified for the P450 21A2 enzyme and associated in many cases with congenital adrenal hyperplasia. By analyzing just a few of these mutants in the context of the structure of human P450 21A2, we have demonstrated that a computational model of the three-dimensional structure of human P450 21A2 derived by others (7) on the basis of our previously determined structure of the bovine enzyme (6) does not provide a meaningful understanding of the consequences for structure and stability by mutation in some cases. A more extensive correlative analysis of the structure and phenotypes of the many P450 21A2 mutants identified to date goes beyond the focus of this paper and will be presented elsewhere.
In summary, the human P450 21A2 structure, complexed with progesterone, provides a better template for a biochemical understanding of the disease variants. The juxtaposition of the substrate is considered to be very important in explaining the very efficient catalysis and the rate-limiting nature of the C-H bond cleavage in the reaction. Both the C-H bond-breaking and substrate-binding steps appear to contribute in the kinetic analysis of the overall hydroxylation reaction.