A Rational Approach to Re-engineer Cytochrome P450 2B1 Regioselectivity Based on the Crystal Structure of Cytochrome P450 2C5*

The regioselectivity for progesterone hydroxylation by cytochrome P450 2B1 was re-engineered based on the x-ray crystal structure of cytochrome P450 2C5. 2B1 is a high K m progesterone 16 (cid:1) -hydroxylase, whereas 2C5 is a low K m progesterone 21-hydroxylase. Initially, nine in- dividual 2B1 active-site residues were changed to the corresponding 2C5 residues, and the mutants were purified from an Escherichia coli expression system and assayed for progesterone hydroxylation. At 150 (cid:2) M progesterone, I114A, F297G, and V363L showed 5–15% of the 21-hydroxylase activity of 2C5, whereas F206V showed high activity for an unknown product and a 13-fold decrease in K m . Therefore, a quadruple mutant, I114A/ F206V/F297G/V363L (Q), was constructed that showed 60% of 2C5 progesterone 21-hydroxylase activity and 57% regioselectivity. Based on their 2C5-like testosterone hydroxylation profiles, S294D and I477F alone and in combination were added to the quadruple mutant. All three mutants showed enhanced regioselectivity (70%) for progesterone 21-hydroxylation, whereas only Q/I477F had a higher k

Until the recent elucidation of the rabbit cytochrome P450 2C5 x-ray crystal structure (1), chimeragenesis, site-directed mutagenesis, and homology modeling based on bacterial structures have been the primary tools available to identify key residues responsible for the substrate specificities of mammalian P450 enzymes (2). Most of these residues belong to the substrate recognition sites (SRSs) 1 proposed by Gotoh (3) based on analogy with the crystal structure of bacterial P450 101 and have direct counterparts in the active site of P450 2C5. However, most P450 structures reveal that the heme group is buried deep within the protein matrix, indicating that residues outside of the active site may also be required to guide the substrate into the heme pocket by recognizing substrates at the protein surface and/or comprising part of a substrate access channel (4 -6).
The role of SRS and non-SRS residues in differential substrate specificity and stereo-and regioselectivity within P450 subfamilies has been studied thoroughly, especially in the case of P450 2A, 2B, and 2C enzymes. In most cases, the functions of the enzymes could be interconverted by making multiple reciprocal substitutions at SRS residues (7)(8)(9)(10). However, in a number of other studies, non-SRS residues were shown to play a crucial role in determining substrate specificities (11)(12)(13)(14)(15)(16)(17). These non-SRS residues were predicted to be part of a substrate access channel near a region between the F and G helices, as seen in bacterial P450 101, or between the B-C loop and N terminus of the I helix, as seen in bacterial P450 51 (18 -20). The mammalian P450 2C5 structure supports the existence of both substrate access channels (1).
Structure-function studies across P450 subfamilies have been largely neglected until very recently (21,22). Renewed interest in this area was sparked by the considerable effort in a number of laboratories to use the P450 2C5 structure to model other mammalian P450 enzymes and to predict their substrate specificities and stereo-and regioselectivity (2,(23)(24)(25). An implicit assumption in all such models based on a single template is that the backbones of the enzymes are essentially invariant and that active-site differences alone are responsible for specificity differences. In this study, we sought to confer the progesterone hydroxylation specificity of 2C5 on 2B1. 2B1 is a high K m progesterone 16␣-hydroxylase, whereas 2C5 is a low K m progesterone 21-hydroxylase (26). 2B1 was used as a model enzyme because of extensive previous site-directed mutagenesis studies that have verified experimentally all 13 active-site residues inferred from the 2C5 structure (2,21,(27)(28)(29)(30)(31)(32). Through a systematic approach of site-directed mutagenesis, a 2B1 enzyme (V103I/I114A/F206V/S294D/F297G/V363L/I477F) was constructed that showed a 3-fold higher k cat compared with 2C5 and 80% regioselectivity for progesterone 21-hydroxylation. The results suggest a dominant role of active-site side chains in determining regioselectivity differences across these two P450 subfamilies and extend previous evidence for the reliability of 2B models based on the 2C5 structure (24). All other chemicals were purchased from sources previously described (33) or from standard suppliers. Rat NADPH-cytochrome P450 reductase and cytochrome b 5 were prepared as described (34).
Site-directed Mutagenesis-The truncated version of 2B1 that served as the background for all mutations described in this study, 2B1dH, was generated using overlap extension PCR as described (33). Mutants were constructed either by overlap extension PCR or by subcloning using pKK2B1dH as a template. The primers and templates used are shown in Fig. 1. Primary forward and reverse primers were as previously described (33). To confirm the desired mutation and to verify the absence of unintended mutations, all constructs were sequenced at the University of Texas Medical Branch Protein Chemistry Laboratory.
Expression and Purification of P450 2B1dH Mutants-2B1dH and its mutants were expressed in Escherichia coli TOPP3 (Stratagene) and purified as described (33). In brief, bacteria were grown in Terrific Broth salts for ϳ2.5 h at 37°C before induction using isopropyl-␤-Dthiogalactopyranoside and supplementation with the heme precursor ␦-aminolevulinic acid. After 72 h at 30°C, the cells were harvested by centrifugation, and protein was extracted from lysed membranes using potassium phosphate buffer (pH 7.4). Protein extract was loaded onto a Ni 2ϩ affinity column, which was washed with 20 volumes of buffer, and then eluted with 200 mM imidazole. Cytochrome P450 was quantitated using the reduced CO difference spectrum (35). Specific content was determined using the Pierce BCA kit with bovine serum albumin as a standard. The specific content for 2B1dH was 18 nmol of P450/mg of protein, whereas the specific content for most of the mutants was between 8 and 16. However, the I114A/F206V/S294D/F297G/V363L, I114A/F206V/S294D/F297G/V363L/I477F, and V103I/I114A/F206V/ S294D/F297G/V363L/I477F multiple mutants had specific contents of 4, 5, and 2 nmol, respectively. Low specific content in these mutants is accounted for by the presence of significant amounts of P420.
Enzymatic Assays-Progesterone and testosterone hydroxylation assays were carried out essentially as described (33, 36) using a 1:4:2 molar ratio of P450/cytochrome P450 reductase/cytochrome b 5 in the absence of lipid. 16␣and 15␣-hydroxyprogesterone were not resolved. However, the progesterone 16␣-hydroxylase activity of 2B1 has been measured previously using two-dimensional chromatography (21). K m and k cat values were determined by regression analysis using Sigma Plot (Jandel Scientific, San Rafael, CA).
Computer Modeling-A molecular model of P450 2B1 was constructed using the InsightII software package (Homology, Discover_3, Biopolymer, Builder, and Docking from Molecular Simulations Inc., San Diego, CA) and P450 2C5 as the template as described previously (24). For the 2B1 mutant (V103I/I114A/F206V/S294D/F297G/V363L/ I477F), the coordinates of the corresponding residues were changed in the 2B1 three-dimensional model by Biopolymer, and the resulting 2B1 mutant was minimized.
The structure of progesterone was constructed using the Builder module. The parameters for heme and ferryl oxygen were those described by Paulsen and Ornstein (37,38). During the docking calculations, the system energy minimization and molecular dynamics simulations were carried out with the Discover_3 program using the consistent valence force field with a non-bond cutoff of 10 Å to a maximum gradient of 5 kcal mol Ϫ1 Å Ϫ1 . Progesterone was automati- cally docked into the three-dimensional models of 2B1 and V103I/ I114A/F206V/S294D/F297G/V363L/I477F in a reactive binding orientation with the Docking module of InsightII, leading to 16␣-or 21hydroxylation. Because the initial oxidation step involves hydrogen abstraction, the C-16 or C-21 atom was placed 3.7 Å from ferryl oxygen, with the 16␣-hydrogen or one of the hydrogen atoms bonded to C-21 directed toward ferryl oxygen (C-H-ferryl oxygen angle of 180°) to promote hydrogen bond formation. During the subsequent energy minimization process, the substrate molecule, along with the side chains of protein residues within 5 Å of the substrate, was allowed to move. The non-bond interaction energies were evaluated with the Docking module of InsightII, and the lowest energy orientation obtained after molecular mechanics minimization of 2B1 and its mutant is shown in Fig. 6.
The K m for Progesterone Is Decreased by ϳ13-Fold upon F206V Substitution-As shown in Table I, the K m for progesterone of 2B1dH was determined to be 200 M. Mutants I114A and F297G showed a 3-fold decrease in K m , whereas V363L showed a 2-fold decrease. Interestingly, a F206V substitution decreased the K m by ϳ13-fold. The k cat values for progesterone 21-hydroxylation by I114A, F297G, and V363L were in the range of 1-2 min Ϫ1 versus 16 min Ϫ1 for 2C5dH (Table I).
I114A/F206V/F297G/V363L Shows Enhanced k cat and Regioselectivity for Progesterone 21-Hydroxylation-A number of multiple mutants were constructed by combining I114A, F297G, and V363L to test the additive effect on progesterone 21-hydroxylase activity. In addition, F206V was also added to assess whether this mutation decreases the K m . The progesterone hydroxylation profiles of the multiple mutants are presented in Fig. 3. As expected, I114A/F297G (Fig. 3, A bars) showed enhanced progesterone 21-hydroxylase activity compared with either single mutant. Addition of V363L to I114A/ F297G increased the regioselectivity for progesterone 21-hydroxylation, although activity was suppressed (Fig. 3, B bars versus A bars). Addition of F206V to I114A and I114A/V363L caused high activity for unknown products while suppressing progesterone 21-and 16␣/15␣-hydroxylase activities (Fig. 3, C and E bars). However, addition of F206V to I114/F297G did not suppress progesterone 21-and 16␣/15␣-hydroxylase activities (Fig. 3, D bars). Interestingly, addition of F206V to I114A/ F297G/V363L enhanced activity for progesterone 21-hydroxylation and suppressed progesterone 16␣-and 15␣-hydroxylase activity, but still allowed significant production of unknown products (Fig. 3, F bars). Unlike 2B1dH, which produced only 16␣-hydroxyprogesterone, and 2C5dH, which produced only 21-hydroxyprogesterone, the quadruple mutant I114A/F206V/ F297G/V363L (Q) produced two major and two minor unknown products (Fig. 4). 21-Hydroxyprogesterone was determined to compose 57% of all the products in the quadruple mutant. I114A/F206V, I114A/F206V/F297F, and I114A/F206V/V363L had a decreased K m for progesterone versus those mutants without Val 206 . The K m for progesterone was unaltered in I114A/F297G and I114A/F297G/V363L compared with the individual single mutants (Table II), further suggesting the crucial role of F206V in enhancing the affinity for the substrate. The k cat for the quadruple mutant was 60% of that for 2C5dH, and the K m was decreased by 3-4-fold compared with 2B1dH (Tables I and II). The 10-fold increase in k cat for progesterone 21-hydroxylation with no change in K m upon addition of F206V to I114A/F297G/V363L is striking and is different from other multiple mutants that include F206V.
The Testosterone Hydroxylation Profiles of S294D and I477F Are Similar to That of 2C5dH-Testosterone hydroxylation profiles were determined for 2B1dH single mutants and compared with those of 2B1dH and 2C5dH. As reported above, 2B1dH produced equal amounts of 16␣-and 16␤-hydroxytestosterone, whereas 2C5dH demonstrated mainly testosterone 16␤-hydroxylase activity (Fig. 5). Of the nine single mutants tested, S294D, V363L, and I477F had decreased testosterone 16␣-hydroxylase activity and enhanced 16␤-hydroxylase activity (similar to 2C5dH). The K m of 2C5dH for testosterone (Table  III) was similar to the reported K m of 2B1dH (39). However, the S294D and I477F single mutants demonstrated a 2-fold lower K m for testosterone. Interestingly, F206V showed decreased 16␣and 16␤-testosterone hydroxylase activities and a new testosterone 6␣-hydroxylase activity. This profile of testosterone hydroxylation is similar to the progesterone hydroxylation profile in that F206V primarily showed activity for a new product (Figs. 2 and 5). On the other hand, V103I showed unaltered testosterone hydroxylation, and F297G, V367L, and G478V showed decreased testosterone 16␣and 16␤-hydroxylase activities. The similar testosterone hydroxylation profiles and affinities for substrate of 2C5dH, S294D, and I477F suggested that addition of S294D and/or I477F to the 2B1dH quadruple mutant might enhance progesterone 21-hydroxylase activity and/or regioselectivity. Q/I477F and Q/S294D/I477F Show Enhanced Progesterone 21-Hydroxylase Activity and Regioselectivity-Q/S294D, Q/I477F, and Q/S294D/I477F were examined for progesterone hydroxylation using 150 M substrate (Table IV). Progesterone 21-hydroxylase activity was largely unaffected, whereas regioselectivity for Q/S294D, Q/S294D/I477F, and Q/I477F was increased to 73, 69, and 67%, respectively, compared with 57% for the quadruple mutant. Steady-state kinetic parameters were also measured for these multiple mutants (Table IV). The k cat for Q/S294D and Q/S294D/I477F was similar to that for the quadruple mutant, but was increased for Q/I477F (similar to 2C5dH). However, the K m was either unaffected in the case of Q/S294D and Q/S294D/I477F or increased in the case of Q/I477F. These observations suggested that addition of S294D to the quadruple mutant increased regioselectivity for progesterone 21-hydroxylation, whereas I477F enhanced the activity as well as the regioselectivity. The k cat and regioselectivity for progesterone 21-hydroxylation were much improved over those for the quadruple mutant and close to those for 2C5dH. To test further improvement of progesterone 21-hydroxylase activity and regioselectivity, the remaining three mutants (V103I, V367L, and G478V) were added to Q/I477F and Q/S294D/ I477F.
Docking of Progesterone into the Active Site of 2B1dH and Q/V103I/S294D/I477F Models-To explain the changes in re-  gioselectivity observed for progesterone in 2B1 Q/V103I/S294D/ I477F, a molecular model was constructed. Fig. 6 (A and B) shows progesterone docked into the active site of the wild-type 2B1 and 2B1 Q/V103I/S294D/I477F models, respectively. The substrate fit well in the wild-type 2B1 active site, with no van der Waals overlaps when docked in an orientation that leads to formation of 16␣-hydroxyprogesterone (Fig. 6A). However, the substrate did not fit in a 21-OH orientation. The estimated angle (C-H-ferryl oxygen) and distance (between C-16 and ferryl oxygen) in 2B1 are 153.0°and 4.43 Å, respectively. Activesite residues Ile 114 , Phe 206 , Phe 297 , Val 363 , and Ile 477 are within 5 Å of the substrate. In contrast, progesterone fit well in the 2B1 Q/V103I/S294D/I477F active site when docked in a 21-OH orientation, but not in a 16␣-OH orientation (Fig. 6B). The estimated angle (C-H-ferryl oxygen) and distance (between C-21 and ferryl oxygen) are 160.6°and 3.69 Å, respectively.
These are close to the angle and distance required for hydrogen bond formation (180°and 3.7 Å, respectively) and are similar to those for 2C5 ( Ref. 1 and data not shown). Active-site residues Ile 103 , Ala 114 , Val 206 , Asp 294 , Gly 297 , Leu 363 , and Phe 477 are within 5 Å of the substrate, with Ala 114 and Phe 477 lying closest at 3 Å. Our modeling results are consistent with the biochemical data that 2B1 favors 16␣-hydroxylation, whereas 2B1 Q/V103I/S294D/I477F favors the formation of 21-hydroxyprogesterone. Consistent with the experimental data, modeling of progesterone in 2B1 mutants that included V367L and/or G478V along with Q/V103I/S294D/I477F showed poor fits (data not shown). This suggests that V367L and G478V mutations, either individually or in combination with others, change the active-site structure in a way that is unfavorable for progesterone 21-hydroxylation. DISCUSSION The recent elucidation of the x-ray crystal structure of the first mammalian P450 (rabbit 2C5) has sparked intense interest in understanding the structural basis of P450 function to facilitate drug discovery/design and engineering of novel biocatalysts (1). This breakthrough has led to advanced homology models of drug-metabolizing cytochromes P450, especially the enzymes from the P450 2 subfamily, relative to those constructed based on bacterial enzymes (25). The P450 2C5 crystal structure has also provided an enhanced framework for identifying active-site residues and investigating their role in differential substrate specificities and stereo-and regioselectivity across subfamilies (2,22). By extensive site-directed mutagenesis studies and more recently by analogy with 2C5, 13 2B1 active-site residues have been identified, nine of which differ from those of 2C5 (2). In the present study, 2B1 residues have been replaced systematically by the corresponding active-site residues of 2C5 to confer a novel progesterone hydroxylase activity (progesterone 21-hydroxylation).
The major finding is that simultaneous substitution of seven 2B1 active-site residues (positions 103, 114, 206, 294, 297, 363, and 477) with the corresponding 2C5 residues resulted in a 3-fold higher k cat for progesterone 21-hydroxylation compared with 2C5 with 80% regioselectivity. However, the K m for the substrate remained an order of magnitude higher than that for 2C5. Consistent with the experimental data, substrate docking in the active site of a model of the multiple mutant showed that the formation of 21-hydroxyprogesterone is favored, unlike wild-type 2B1, which favors 16␣-hydroxyprogesterone. To produce a 2B1 multiple mutant with the desired phenotype, a three-tiered approach was used. First, 2B1 single mutants were made at all nine non-identical active-site positions, and three mutants that showed progesterone 21-hydroxylation (I114A, F297G, and V363L) and a fourth with a decreased K m (F206V) were combined to yield a quadruple mutant. Second, additional substitutions (S294D and I477F) were added based on a 2C5-like testosterone hydroxylation profile. Finally, V103I, which had no effect on progesterone or testosterone   Engineering Cytochrome P450 2B1 Regioselectivity hydroxylation profiles on its own, was added to Q/S294D/ I477F. New insights can be gained by comparing and contrasting the present study with our recent investigation of active-site determinants of specificity differences between P450 2B6 and 2E1 (22). In that case, a single point mutation at the alignment position corresponding to residue 477 in 2B enzymes conferred on 2E1 significant activity for the 2B6-selective substrate 7-ethoxy-4-trifluoromethylcoumarin and abolished activity for the 2E1-selective substrate p-nitrophenol. However, none of six 2B6 single mutants gained activity for p-nitrophenol. The two major advances in this investigation were the focus on regioselectivity differences for a common substrate, progesterone, and the generation of multiple 2B mutants, many of which included substitutions that, on their own, did not enhance the activity of interest. Based on the results, it appears that, in addition to direct interactions of active-site residues with substrate, residue-residue interactions and/or an influence of active-site backbone residues on orientation of the substrate may also be important (4,24). Residue-residue interactions have been implicated in determining stereo-and regioselectivity for androstenedione hydroxylation and differential inhibition by 4-phenylimidazole in 2B4 and 2B5 (7,24). For example, mutagenesis experiments and molecular modeling suggest that the side chains of residues 114 and 294 in 2B4 and 2B5 move in concert to influence the 4-phenylimidazole binding orientation (24). In the present case, there were increases in progesterone 21-hydroxylase activity of 10-, 2-, and 3-fold upon addition of F206V to I114A/F297G/V363L, I477F to I114A/F206V/F297G/ V363L, and V103I to Q/I477F, respectively, even though F206V, I477F, and V103I alone showed negligible progesterone 21-hydroxylase activity. These observations may reflect an additional interaction between Phe 477 and/or Phe 206 and progesterone that leads to tighter packing in the active site and an increased frequency of productive collisions (Fig. 6). Ile 103 is also closer to progesterone in Q/V103I/S294D/I477F compared with 2B1.
The almost complete conversion of stereoselectivity for testosterone 16-hydroxylation by 2B1 S294D, V363L, and I477F to that of 2C5 aided in making multiple mutants with higher a Results are the means of two independent experiments. The variation between experiments is not more than 5%. b Results are the means Ϯ S.D. of two independent experiments. c Unknown activity is the sum of four different products, as shown in Fig. 4.  6. Docking of progesterone into the active site of P450 2B1 models. A, substrate was docked in 2B1dH in an orientation leading to formation of 16␣-hydroxyprogesterone; B, substrate was docked into Q/V103I/S294D/I477F in an orientation leading to 21-hydroxyprogesterone. The heme (red sticks), progesterone (brown space-filling representation), and active-site residues (purple sticks) are shown. The carbon atoms at positions 16 and 21 of progesterone are shown in blue and green, respectively. progesterone 21-hydroxylase activity and regioselectivity. However, it should be recognized that the 2C5-like testosterone hydroxylation profiles of S294D, V367L, and I477F resulted from abolished testosterone 16␣-hydroxylase activity and enhanced 16␤-hydroxylase activity, rather than acquisition of a novel activity as with progesterone 21-hydroxylation by I114A, F297G, and V363L. There are numerous prior examples of stereoselective loss of activity upon single amino acid substitutions in bacterial and mammalian P450 enzymes (28,40,41). On the other hand, F206V exhibited a novel activity with both progesterone and testosterone along with greatly suppressed original activities. Residue 206 in 2B1 has previously been shown to be critical in converting steroid 16-to 15␣-hydroxylation (21), and the analogous residue in 2a4 and 2a5 is a major determinant of differences in substrate specificity (40). The F205V substitution in 2C5 results in almost a complete loss of progesterone 21-hydroxylase activity and gain of new activity, suggesting that this residue is critical in determining P450 regioselectivity (26). However, the structural basis for these observations remains unclear.
In the absence of deleterious steric interactions with 2B1, the more hydrophobic progesterone should be characterized by a lower K m compared with testosterone, as observed with 2C5. However, wild-type 2B1 exhibited a 7-fold higher K m for progesterone than for testosterone. This suggests that the larger 17␤-acetyl group in progesterone, as opposed to the hydroxyl group in testosterone, clashes with one or more residues in the 2B1 active site. Furthermore, a F206V substitution in 2B1, either individually or in combination with I114A, I114A/ F297G, or I114A/V363L, decreased the K m for progesterone by 6 -13-fold, suggesting a major role of Val 206 in determining affinity. Other multiple mutants that included F206V exhibited an increased K m , which may be due to the occurrence of additional unfavorable interactions. The K m is generally affected by parameters such as size, shape, and hydrophobicity of the substrate and of active-site residues (42). A similar Phe-to-Val substitution at residue 226 in P450 1A2 (analogous to residue 206 in 2B1) has also been shown to have a strong bearing on substrate affinity (43). An excellent correlation has been observed between K m values and side chain size at residue 209 in P450 2a5 (analogous to residue 206 in 2B1), in which the K m values decrease as the side chains become larger regardless of the hydrophobicity (44). The larger side chain of Phe 206 in 2B1 may interact sterically with the additional acetyl group in progesterone compared with testosterone ( Fig. 6) (1). The F206V substitution may provide more room for entry in a particular orientation leading to increased affinity for substrate.
In summary, our results demonstrate that active-site residues are mainly responsible for determining differences in regioselectivity for progesterone hydroxylation between 2B1 and 2C5. A synergistic effect on progesterone 21-hydroxylation activity and regioselectivity by certain multiple substitutions suggests a role of residue-residue interactions in determining active-site topology and substrate orientation. This report suggests the feasibility of rational redesign of mammalian P450 specificity based on analogy with P450 2C5, as previously performed for bacterial P450 enzymes of known three-dimensional structure (45). This approach provides an excellent complement to directed evolution by random mutagenesis, which tends to mainly pinpoint non-activesite residues (46,47).