P450 engineering via structure-guided rational design achieves high C21-selectivity and bioconversion in steroid biosynthesis

TL;DR

Scientists engineered a P450 enzyme to efficiently convert steroids with high selectivity for C21-hydroxylation, improving steroid biosynthesis.

Contribution

A structure-guided approach created a P450 variant with 98% C21 selectivity and high conversion efficiency for steroid hydroxylation.

Findings

The engineered M6a variant achieved 98% C21 hydroxylation selectivity and >99% conversion of progesterone.

The redox fusion variant M6a-RhFRED L3 improved catalytic activity by 1.43-fold compared to the original.

M6a-RhFRED L3 efficiently hydroxylated three steroid analogs with >60% conversion and 98% selectivity.

Abstract

C21-hydroxylation is a crucial step in the synthesis of corticosteroids. For instance, C21-hydroxylase catalyzes the conversion of progesterone (PRO) into 11-deoxycorticosterone (DOC, the precursor of cortisol and aldosterone). Through structure-guided rational design combined with focused rational iterative site-directed mutagenesis (FRISM), we engineered the wild-type CYP154C5 from Nocardia farcinica to obtain the highly efficient variant M6a (F92A/V291L/L294I/Q239K/F180W/Q398M). This variant demonstrated a complete regioselectivity shift from C16α to C21 hydroxylation (98% selectivity) of PRO with high catalytic efficiency (>99% conversion). To further enhance catalytic performance, we engineered a redox fusion variant (M6a-RhFRED L3) that demonstrated the highest catalytic activity, achieving a 1.43-fold improvement compared to the M6a-RhFRED. Additionally, M6a-RhFRED L3 catalyzed the C21-hydroxylation of three steroid analogs (dydrogesterone, 16-dehydroprogesterone, and pregna-4,6-diene-3,20-dione), with selectivity reaching 98% and conversion rates exceeding 60%. Molecular docking and molecular dynamics (MD) simulations revealed that the PRO substrate undergoes conformational rearrangement in the M6a active site. This structural reorganization underscores the critical role of key residues in modulating regioselectivity, particularly the shift from C16α to C21 hydroxylation. This study not only provides an efficient biocatalyst for steroids C21-hydroxylation but also offers valuable insight for the rational engineering of the P450 enzymes.

Figures

Six key amino acid residues targeted for rational design in CYP154C5. (A) Active site structure of WT CYP154C5 complexed with PRO (PDB ID: [4J6C](https://www.rcsb.org/structure/4J6C)). The α-helices and β-strands are shown as cartoons. Six key amino acid residues and PRO are displayed as yellow and magenta sticks, respectively. The distances from C16 to heme iron (green dashed line) and from C21 to heme iron (yellow dashed line) are 4.3 Å and 5.5 Å, respectively. (B) The mutant library was constructed by targeting the six key residues (F92, V291, Q239, L294, F180, and Q398) in this study.Fig.

FRISM mutagenesis of CYP154C5. (A) Regioselectivity of PRO hydroxylation across the mutant library. C16*α*-selectivity (orange) and C21-selectivity (blue) are shown in the histogram. (B) Total conversion rates for PRO hydroxylation among the engineered variants. (C) Representative HPLC chromatograms of six selected mutants (M1a–M6a) demonstrating conversion of PRO to 16*α*-hydroxy and/or 21-hydroxy products. Notably, M6a exhibited a complete regioselectivity switch from C16*α* to C21 position (98% selectivity). Reaction conditions: Recombinant *E. coli* cells were incubated in 50 mM phosphate

Chimeric enzymes M6a-RhFRED (L1–L10) for the bioconversion of PRO to 11-deoxycorticosterone (DOC). (A) Schematic illustration of the linker region engineering between M6a and RhFRED. (B) Comparison of conversion efficiency among ten chimeric enzymes (L1–L10), as well as the M6a-RhFRED enzyme. (C and D) TTN values for PRO conversion at different PRO concentrations (0.1–6.0 mM) were measured using whole-cell M6a-RhFRED and M6a-RhFRED L3. TTN values were calculated as the amount of substrate consumed per unit enzyme concentration ([μmol of substrate consumed/μmol of P450]). The conversion rates a

Bioconversion of PRO analogs by recombinant strain M6a-RhFRED L3. (A) Chemical structures of the PRO analogs used in this study. (B) HPLC analysis of the bioconversion of PRO analogs 1, 2, and 3 (substrate concentration: 200 μM). I: Biotransformed PRO analogs by M6a-RhFRED L3; II: Control bioconversion using *E*. *coli* harboring pET28a-RhFRED without P450; III: Purchased standard compounds (1, 2, and 3). Assays were performed in triplicate; representative chromatograms are shown. (C) Mass spectrometry analysis of bioconversion products (1a, 2a, and 3a). Corresponding NMR data are provided in

MD simulations of WT and M6a bound with PRO. (A, C) Active site structures of WT and M6a. Six key residues in the active site are shown as cyan and orange sticks in M6a and WT, respectively. The green dashed line shows the hydrogen bond between PRO and Q239 in the WT. Black dashed lines show the distances from the hydroxylation sites (C16 in WT and C21 in M6a) to the OFe (heme). (B, D) Plots showing distances between the C16- and C21-hydroxylation sites of PRO and the OFe (heme) group over simulation time for WT (B) and M6a (D), respectively.Fig. 5

Superimposition of the active site structures of WT and M6a, both extracted from MD simulations at 80 ns. Six key amino acid residues, PRO and heme, are depicted as sticks, with WT and m6a colored cyan and orange, respectively. The hydroxylation positions (C16 in WT and C21 in M6a) are highlighted as spheres. The keto oxygen of PRO C3 shifts ∼5 Å toward W180 in M6a compared to WT, indicating a steric reorganization of the active site.Fig. 6

Full text

Introduction

1

Corticosteroid drugs (such as hydrocortisone, prednisolone) are widely used to treat asthma, arthritis, and autoimmune diseases. C21-hydroxylation represents a critical enzymatic modification in the synthetic pathway of corticosteroid drugs, as it directly determines the pharmacological activity and metabolic stability of the final product [1,2]. Traditional chemical synthesis of C21-hydroxysteroids typically involves multiple steps using hazardous halogenating reagents and complex purification processes, often resulting in low efficiency and considerable environmental impact [3]. Biocatalysis has emerged as a promising and sustainable alternative due to its high regioselectivity and environmentally benign nature [[4], [5], [6], [7]]. Nevertheless, the industrial application of C21-hydroxylases is significantly constrained by their limited catalytic efficiency and narrow substrate specificity. To date, the most extensively studied steroid C21-hydroxylases are animal-derived P450 enzymes, including CsCYP21A, CYP21A2, P450c21, CYP3A4, and P450XXI [[7], [8], [9], [10]]. P450c21 catalyzes the hydroxylation of PRO and 17α-hydroxyprogesterone to 11-deoxycorticosterone (DOC) and 11-deoxycortisol [11,12]. DOC is the C21-hydroxylated product of PRO, also serves as a key intermediate in the synthesis of hydrocortisone and aldosterone [[13], [14], [15]]. However, the heterologous expression and regioselectivity of animal-derived P450 enzymes remain challenging, which limits their application. Therefore, it is urgent to explore novel steroids C21-hydroxylase or engineer existing ones to achieve higher conversion and selectivity.

Cytochrome P450s (P450s) are involved in diverse steroid biosynthetic pathways across all domains of life, including plants, insects, vertebrates, yeasts, fungi, and bacteria [16]. Compared to their eukaryotic counterparts, bacterial P450s exhibit higher solubility and greater amenability to protein engineering, making them attractive candidates for industrial biocatalysis. A notable example is the hydroxylation of vitamin D3 to generate 25-hydroxyvitamin D3 (calcidiol) and 1α,25-dihydroxyvitamin D3 by engineered bacterial P450 enzymes, such as CYP107Pdh from Pseudonocardia dioxanivorans CB1190 [17], CYP105A1 from Streptomyces griseolus, and CYP105A2 from Amycolata autotrophica [18]. Recent advances in protein engineering have enabled the selective hydroxylation of various steroidal positions (e.g., C7, C11, C15, and C16) by engineered P450s with enhanced catalytic efficiency [[19], [20], [21], [22]].

To date, bacterial C21 hydroxylases capable of PRO hydroxylation are predominantly engineered variants, including the F92A mutant of CYP154C5 from Nocardia farcinica and engineered P450BM3 variants from Bacillus megaterium. However, engineered C21-hydroxylase variant of P450BM3 (VD21-BM3) convert only PRO with high regioselectivity (95%), while CYP154C5 catalyzes the C21-hydroxylation of PRO with low regioselectivity (∼20%) [23,24]. In this study, we implemented structure-guided FRISM coupled with redox partner optimization to simultaneously enhance both C21-selectivity and catalytic efficiency of CYP154C5.

Materials and methods

2

Materials

2.1

All restriction enzymes were purchased from Takara (Shanghai, China). Primers were obtained from Tsingke Biotechnology (Beijing, China). Steroid substrates, including 11-deoxycorticosterone (Yuanye) and their derivatives (Aladdin), were commercially sourced from suppliers in Shanghai.

Construction of recombinant E. coli cells

2.2

The pET28a-CYP154C5 plasmid was synthesized by Tsingke Biotechnology (Beijing, China). The CYP154C5 gene (GenBank accession: NC_006361, locus tag: nfa53110) was amplified by PCR from pET28a-CYP154C5. The RhFRED gene (UniProt: Q8KU27) was sourced from plasmid pET28b-sav3882-RhFRED, which was previously constructed in our study [25]. For plasmid construction, pET28a was digested with XhoI and NdeI*,* followed by ligation with the PCR amplified fragments of CYP154C5 and RhFRED using a One-Step Cloning Kit (primers listed in Table S1). The resulting plasmid (pET28a-CYP154C5-RhFRED) was transformed into Escherichia coli BL21 (DE3) for whole-cell biocatalyst preparation.

Mutagenesis and bioconversion analysis

2.3

The CYP154C5 mutant library was constructed using the overlapping PCR method (primers listed in Table S1). Site-directed mutagenesis was performed to generate pET28a-CYP154C5(mutants)-RhFRED variants harboring the desired mutations. The constructed plasmids were sequentially transformed into E. coli BL21 (DE3) cells. Transformed colonies were inoculated into 5 mL LB medium supplemented with kanamycin (50 μg/mL) and cultured at 37 °C, 220 rpm for 16 h. The pre-cultures were transferred to 100 mL TB medium (pH 7.4) containing kanamycin (50 μg/mL) and incubated at 37 °C, 220 rpm until the OD_600_ reached 0.8. The culture medium was then supplemented with 0.5 mM 5-ALA, 0.5 M FeCl_3_·6H_2_O, and 0.1 mM IPTG, followed by incubation at 30 °C, 180 rpm for 16 h. Cells were harvested via centrifugation at 5000×g for 10 min at 4 °C and washed once with 25 mL of water. The reaction was performed using recombinant E. coli cells resuspended in 10 mL of 50 mM phosphate buffer (pH 7.4) containing 10 mM glucose. The cell density for all reactions was standardized to an OD_600_ of 20. The substrate stock solution (10 mM) was prepared using a 4.5% (w/v) hydroxypropyl-β-cyclodextrin (HP-β-CD) aqueous solution. This concentration of HP-β-CD has been shown to effectively solubilize steroids without inhibiting enzymatic activity or affecting cell viability. The reaction mixture (200 μL) with a final concentration of PRO at 100 μM was incubated at 30 °C with shaking at 250 rpm for 20 h. After incubation, the reaction mixture was extracted with an equal volume of ethyl acetate. Conversion and selectivity were analyzed by HPLC using a Waters W2695 system equipped with a C18 column (4.6 × 150 mm, 5 μm; Agilent Technologies, Santa Clara, CA, China). Separation was achieved using a 10-min linear gradient of solvent B from 30% to 100%, followed by an 8-min hold at 100% B and 10-min re-equilibration under the initial conditions (30% B) at a flow rate of 1 mL min^−1^. A 20 μL sample volume was injected for analysis, with absorbance detection at 240 nm for PRO and 16-dehydroprogesterone, and 280 nm for dydrogesterone and pregna-4,6-diene-3,20-dione. Products were further analyzed using an Agilent 1260–6230 time-of-flight LC-MS system.

Scale-up reaction and product identification

2.4

For product preparation, large-scale cultivation was performed by inoculating a 20-mL preculture of the optimal mutant strain M6a into 4 L of TB medium. The cells were harvested by centrifugation and resuspended in 400 mL of phosphate buffer (50 mM, pH 7.4). The substrate (25 mg) was added to initiate the biotransformation. This cell processing method is consistent with the small-scale procedure. After 24 h of incubation, the entire culture broth was extracted twice with an equal volume (400 mL) of ethyl acetate. The solvent was removed using a rotary evaporator to yield a crude product. This crude product was subsequently purified by solid-phase extraction (SPE), with a methanol gradient (20% to 100%) used as the eluent. The fractions containing the target product were pooled and concentrated. Further purification was achieved using semi-preparative HPLC with a COSMOSIL 5C18-MS-Ⅱ column (10 × 250 mm, 5 μm; Nacalai Tesque, Japan) under isocratic elution (65% methanol) at 3 mL/min. The purified products, with a total yield of 4∼6 mg (∼20% conversion), were subjected to NMR analysis. NMR spectra were recorded on a Bruker AV-600 MHz spectrometer using DMSO‑d6 or CDCl_3_ as solvents. All NMR data were processed and analyzed using MestReNova (version 14.0). Compounds DOC, 1a, 2a, and 3a were characterized by NMR for structural confirmation.

Redox partner engineering of M6a

2.5

The plasmid pACYCDuet-Pdx-PdR was constructed as previously described [26]. Both pACYCDuet-Pdx-PdR and pET28a-M6a were co-transformed into E. coli BL21 (DE3) cells to generate the recombinant strain. Ten chimeric enzymes (M6a-RhFRED L1-L10) were subsequently constructed using pET28a-M6a-RhFRED as the template. These enzymes incorporated polypeptide linkers of varying lengths, which were generated by site-directed deletion of specific amino acid residues (as listed in Table 2).

Total turnover number (TTN) analysis

2.6

Following fermentation, the harvested cells expressing M6a-RhFRED or M6a-RhFRED L3 were thoroughly resuspended in 20 mL of phosphate buffer (pH 7.4). An aliquot of cell suspension was subjected to sonication using an ultrasonic disruptor, and the P450 concentration in the crude extract was quantified by CO difference spectroscopy [27]. The remaining cell suspension was used for whole-cell reactions containing either 1 μM M6a-RhFRED or M6a-RhFRED L3, PRO (0.1–6 mM), and 10 mM glucose. The reactions were incubated at ‌30 °C with shaking at 250 rpm. Steroid conversion rates were quantified by HPLC. The total turnover numbers (TTN) were calculated as: TTN = Substrate consumed (μmol)/P450 concentration (μmol).

Molecular dynamics (MD) simulation

2.7

First, structural modeling of M6a was performed using AlphaFold3 (version 3.0, DeepMind) [28]. The structure of CYP154C5 (WT)_PRO was obtained from the PDB code 4J6C, and the alignment of Cpd I to the heme position was performed using PyMOL. Substrate docking was performed by AutoDock Vina [29]. The docking models with the lowest binding energies and a similar binding mode to 4J6C were selected. A grid box of 20 Å around the active site was applied by centering M6a. For MD simulations, missing hydrogen atoms were added by the module leap of Amber 24 [30]. The general AMBER force field (GAFF) was employed to parameterize the substrates using Antechamber, with missing ligand parameters generated by the parmchk utility from AMBER Tools. The force field for the Cpd I state was taken from the literature [31]. Systems were solved in a Cube box of TIP3P waters extending up to a minimum cutoff of 10 Å from the protein boundary. Finally, Na^+^ was added to the solvent to neutralize the total charge. System charge neutrality was achieved by adding Na+ counterions. The Amber ff14SB force field was utilized for protein simulations in all MD runs. The initial structures underwent full minimization using a combination of the steepest descent and conjugate gradient methods, with a total of 5000 steps. The system underwent 1 ns of density equilibration in the NPT ensemble (300 K, 1 atm) using a Langevin thermostat (collision frequency = 0.002 ns) and Berendsen barostat (pressure relaxation time = 0.001 ns). The systems underwent an additional 5 ns equilibration to achieve stable temperature and pressure distributions. Following energy minimization and system equilibration, 100 ns production MD simulations were performed for all complex systems using Gromacs 2024.2 [32].

Results and discussion

3

Construction of a whole-cell biotransformation system of CYP154C5

3.1

The CYP154C5 gene was amplified from the pET28a-CYP154C5 plasmid by PCR, and this plasmid was synthesized. RhFRED is the redox domain of P450RhF (CYP116B2), serving as an effective alternative redox partner for P450 enzymes. We constructed the plasmid pET28a-CYP154C5-RhFRED and transformed it into E. coli BL21 (DE3) as the WT strain. To assess the bioconversion capability of the WT CYP154C5, we added the substrate PRO (final concentration of 100 μM) to the 200 μL whole-cell reaction system and incubated at 30 °C for 20 h. The HPLC results indicated that the WT strain was capable of undergoing a complete C16α-hydroxylation reaction of PRO, with a conversion rate exceeding >99% (Fig. S1).

Rational selection of key amino acids for mutation

3.2

Structural analysis of the CYP154C5-PRO complex (PDB ID: 4J6C) revealed key active site residues involved in substrate recognition [33]. The C16 carbon of PRO was positioned 4.3 Å from the heme iron, while the C21 is 5.5 Å away (Fig. 1A). Sequence analysis revealed that residues F92, Q239, V291, and L294 are non-conserved within the CYP154C family (Fig. S2), suggesting their potential role in the functional diversity, including variations in activity and substrate specificity/selectivity. Among them, the mutant F92A of CYP154C5 has been proven to be capable of generating C21-hydroxylation products with limited selectivity (∼20%) ‌ [23]. Notably, V291 and L294 occupy the fifth and ninth positions of the EXXR motif, known “hotspot” related to the substrate specificity and selectivity [34]. Furthermore, the structural analysis of the substrate binding pocket indicates that residues F92 and F180 are spatially located on both sides of the substrate PRO, especially there are hydrophobic and π-π interactions between F92, F180 and the aromatic ring of PRO, which play a crucial role in determining the binding orientation of the substrate. Q398 and F180 are positioned on the same side of the substrate, with the closest distance between Q398 and PRO less than 5 Å. This spatial arrangement is predicted to influence substrate binding‌. Therefore, mutagenesis of these residues is predicted to alter substrate selectivity‌. Additionally, the six key amino acids mentioned above are all located at the substrate recognition site (SRS): F92 (SRS1), F180 (SRS2), Q239 (SRS4), V291, L294 (SRS5), and Q398 (SRS6) (Fig. S2). To shift regioselectivity from the C16 to the C21 position, we conducted strategic mutagenesis of the key amino acids (Fig. 1B).Fig. 1. Six key amino acid residues targeted for rational design in CYP154C5. (A) Active site structure of WT CYP154C5 complexed with PRO (PDB ID: 4J6C). The α-helices and β-strands are shown as cartoons. Six key amino acid residues and PRO are displayed as yellow and magenta sticks, respectively. The distances from C16 to heme iron (green dashed line) and from C21 to heme iron (yellow dashed line) are 4.3 Å and 5.5 Å, respectively. (B) The mutant library was constructed by targeting the six key residues (F92, V291, Q239, L294, F180, and Q398) in this study.Fig. 1

Mutagenesis-based approach to shift regioselectivity from C16α to C21

3.3

Based on structural rational analysis, we subsequently employed the Focused Rational Iterative Site-directed Mutagenesis (FRISM) method to carry out the mutagenesis of CYP154C5 [35]. Firstly, the WT CYP154C5 was used as a template to construct the variant F92A (M1). The M1 variant altered the C21-selectivity from 0% (WT) to 4%, without affecting conversion rate (>99%) (Table 1 and Fig. S3). The C21-selectivity result is less than the 20% that has been reported. Using M1 as the parental variant, we generated double mutants with V291 substitutions (L/I/K/D). Variants F92A/V291L (M2a), F92A/V291I (M2b), and F92A/V291K (M2c) achieved C21-selectivities of 36%, 32%, and 31%, respectively (4% for M1) (Table 1 and Fig. S4). However, both conversion and C21-selectivity were markedly reduced at variant F92A/V291D (M2d: 31% conversion, 2% selectivity) compared with the F92A. Therefore, in the next round of mutation, based on the optimal variant M2a, Q239 (K/M/R) was introduced to generate triple variants. Surprisingly, F92A/V291L/Q239K (M3a) increased C21-selectivity to 43% with 99% conversion. F92A/V291L/Q239M (M3b) reached 45% selectivity but with reduced conversion (85%). F92A/V291L/Q239R (M3c) did not improve selectivity and exhibited lower conversion (35% selectivity, 46% conversion) (Table 1 and Fig. S5).Table 1. Conversion rates and regioselectivity of CYP154C5 variants in PRO hydroxylation.Table 1. EnzymesMutationsConversion rate (%)Selectivity (%)16α21M1F92A>99964M2aF92A/V291L>996436M2bF92A/V291I996832M2cF92A/V291K<16931M2dF92A/V291D31982M3aF92A/V291L/Q239K995743M3bF92A/V291L/Q239 M895545M3cF92A/V291L/Q239R466535M4aF92A/V291L/Q239K/L294I972872M4bF92A/V291L/Q239K/L294F>998911M4cF92A/V291L/Q239K/L294Q628614M4dF92A/V291L/Q239K/L294R<15941M5aF92A/V291L/Q239K/L294I/F180W96694M5bF92A/V291L/Q239K/L294I/F180Y394060M5cF92A/V291L/Q239K/L294I/F180S<17129M6aF92A/V291L/Q239K/L294I/F180W/Q398M>99298M6bF92A/V291L/Q239K/L294I/F180W/Q398K25743M6cF92A/V291L/Q239K/L294I/F180W/Q398F84298M6dF92A/V291L/Q239K/L294I/F180W/Q398L74298Reaction conditions: recombinant E. coli cells were suspended in 50 mM phosphate buffer (pH 7.4), containing 100 μM PRO and 10 mM glucose. The conversion rate and selectivity of each enzyme were calculated as mean ± SD from three independent experiments.Table 2. Chimeric enzymes (M6a-RhFRED L1–L10) with variable linker lengths were used to increase the conversion of PRO.Table 2. EnzymesRemoved residuesConversion rate (%)FoldM6a-RhFRED0651.00L1L1 (V)891.37L2L2 (VL)861.32L3L3 (VLH)931.43L4L4 (VLHR)490.75L5L5 (VLHRH)420.64L6L6 (VLHRHQ)801.22L7L7 (VLHRHQP)821.25L8L8 (VLHRHQPV)841.28L9L9 (VLHRHQPVT)580.88L10L10 (VLHRHQPVTI)600.92Cells were prepared in 50 mM phosphate buffer (pH 7.4) containing 10 mM glucose and 200 μM PRO. The conversion rates were calculated as mean ± SD from three independent experiments.

To further enhance the selectivity of C21, we introduced L294 (I/F/Q/R) using M3a as the template. The quadruple mutant F92A/V291L/Q239K/L294I (M4a) exhibited enhanced selectivity (72% vs. 43% for M3a) without compromising activity, while F92A/V291L/Q239K/L294F (M4b) maintained conversion rate but showed diminished selectivity (11%). M4c and M4d decreased both conversion and C21-selectivity (Table 1 and Fig. S6). Therefore, variants F92A/V291L/L294I (M4a) were identified as the template for the next round of mutant experiments.

Next, we mutated F180 to W, Y, and S in M4a. Surprisingly, F92A/V291L/Q239K/L294I/F180W (M5a) increased selectivity to 94% (72% for M4a) while maintaining activity (Table 1 and Fig. S7). However, whereas F92A/V291L/Q239K/L294I/F180Y (M5b) and F92A/V291L/Q239K/L294I/F180S (M5c) decreased both selectivity and activity. Finally, using M5a, we mutated Q398 to M, F, K, and L. F92A/V291L/Q239K/L294I/F180W/Q398M (M6a), F92A/V291L/Q239K/L294I/F180W/Q398F (M6c) and F92A/V291L/Q239K/L294I/F180W/Q398L (M6d) increased the C21-selectivity to 98%. However, F92A/V291L/Q239K/L294I/F180W/Q398L (M6b) decreased selectivity to 43% (from 94% for M5a). Among these mutants, M6a retained a high conversion rate (99%), whereas M6b, M6c, and M6d exhibited reduced conversion rates (2%, 84% and 74%, respectively) (Table 1 and Fig. S8). Overall, targeted mutagenesis of V291 (L), Q239 (K/M), L294 (I), F180 (W), and Q398 (M) generated variants with measurable enhancements in conversion and/or C21-selectivity. Notably, M6a achieved near-complete regioselectivity switching, shifting from C16α to C21 (98% C21-selectivity) (Fig. 2). Importantly, only nineteen variants were synthesized in this process, highlighting the efficiency of the FRISM strategy. The product, 21-hydroxyprogesterone, was confirmed by MS and NMR (Figs. S9–S11, Table S2). Recently, Zhou et al. engineered a P450 BM3 mutant (F87A/P25A/P329A/E435D/S72Q/L437G/A330Y/A74G) with 95% C21-selectivity and 96% conversion for PRO [24]. To our knowledge, M6a of CYP154C5 exhibits the highest C21-selectivity (98%) for PRO C21-hydroxylation in bacterial P450s.Fig. 2FRISM mutagenesis of CYP154C5. (A) Regioselectivity of PRO hydroxylation across the mutant library. C16α-selectivity (orange) and C21-selectivity (blue) are shown in the histogram. (B) Total conversion rates for PRO hydroxylation among the engineered variants. (C) Representative HPLC chromatograms of six selected mutants (M1a–M6a) demonstrating conversion of PRO to 16α-hydroxy and/or 21-hydroxy products. Notably, M6a exhibited a complete regioselectivity switch from C16α to C21 position (98% selectivity). Reaction conditions: Recombinant E. coli cells were incubated in 50 mM phosphate buffer (pH 7.4) supplemented with 10 mM glucose and 100 μM PRO. The conversion rates and selectivity were calculated as mean ± standard deviation (SD) from three independent experiments.Fig. 2

Redox partner engineering of M6a to improve DOC production

3.4

Redox partners play an essential role in determining the catalytic efficiency of P450 enzyme systems [36]. Especially, the reductase domains of self-sufficient P450s (e.g., RhFRED, reductase domain of P450RhF (CYP116B2) from Rhodococcus sp.) can be engineered through domain fusion with heterologous P450s, yielding chimeric biocatalysts with improved catalytic efficiency [20,37,38]. To identify an effective redox partner, we constructed E. coli BL21 (DE3) strains harboring either pET28a-M6a + pACYCDuet-Pdx/PdR or pET28a-M6a-RhFRED, and compared their conversion performance. Since the M6a-RhFRED system almost completely converted the substrate PRO at a final concentration of 100 μM (99%), in order to better evaluate the conversion capabilities of the two systems, we increased the substrate concentration from the original 100 μM to 200 μM. The results showed that RhFRED exhibited a higher conversion rate (65%) than the Pdx/PdR (45%) (Fig. S12). Consequently, RhFRED was selected as the redox partner for further optimization.

The linker length between P450 and RhFRED can influence electron transfer and catalytic efficiency [39] (Fig. 3A). In order to further enhance the conversion efficiency of M6a, we constructed ten types of chimeric enzymes (M6a-RhFRED L1–L10) with variable linker lengths (Table 2). All chimeric enzymes except M6a-RhFRED L4 and M6a-RhFRED L5 exhibited higher conversion than the control M6a-RhFRED (Fig. 3B). Compared to the control, M6a-RhFRED L3 showed a 1.43-fold improvement, indicating that L3 represents the most suitable linker length for RhFRED and effectively enhances the conversion rate. To assess the correlation between the length of the linker and the protein expression level, we conducted CO differential spectroscopy analysis on the two fusion proteins, M6a-RhFRED and M6a-RhFRED L3 (Fig. S13). We found that the characteristic peak at 450 nm for M6a-RhFRED L3 was much higher than that of M6a-RhFRED, indicating that the length of the RhFRED L3 linker increased the protein expression level. This might be the main reason for the improvement in the conversion rate.Fig. 3. Chimeric enzymes M6a-RhFRED (L1–L10) for the bioconversion of PRO to 11-deoxycorticosterone (DOC). (A) Schematic illustration of the linker region engineering between M6a and RhFRED. (B) Comparison of conversion efficiency among ten chimeric enzymes (L1–L10), as well as the M6a-RhFRED enzyme. (C and D) TTN values for PRO conversion at different PRO concentrations (0.1–6.0 mM) were measured using whole-cell M6a-RhFRED and M6a-RhFRED L3. TTN values were calculated as the amount of substrate consumed per unit enzyme concentration ([μmol of substrate consumed/μmol of P450]). The conversion rates and TTN values were calculated as mean ± SD from three independent experiments.Fig. 3

The catalytic efficiency was evaluated by determining the total turnover numbers (TTN) of M6a-RhFRED and M6a-RhFRED L3 in whole-cell systems with PRO concentration ranging from 0.1 to 6 mM. TTN values increased progressively with PRO concentration from 0.1 to 2 mM but declined at higher concentrations (4 mM and 6 mM) in both systems. The highest TTN values for C21-selective hydroxylation were achieved at 2 mM PRO, reaching 422 for M6a-RhFRED and 286 for M6a-RhFRED L3 (Fig. 3C and D). The results indicate that the electron transfer efficiency of M6a-RhFRED L3 is higher than that of M6a-RhFRED, achieving a higher substrate conversion efficiency. Similar linker optimization strategies have enhanced catalytic activity in other class I P450 [39]. For instance, P450tol-RhFRED L3 exhibited a 4.7-fold increase in catalytic activity compared to P450tol-RhFRED [20], while P450cam-RhFRED F4 achieved a 20-fold increase in yield relative to P450cam-RhFRED [39]. Our findings further confirm that optimizing linker length for RhFRED can effectively improve conversion efficiency.

Bioconversion of various steroid substrates

3.5

The substrate scope of M6a-RhFRED L3 was examined using eight PRO analogs (Fig. 4A). Among them, dydrogesterone (1), 16-dehydroprogesterone (pregna-4,16-diene-3,20-dione) (2), and pregna-4,6-diene-3,20-dione (6-dehydroprogesterone) (3) were converted to their C21-hydroxylated products with conversions of 64%, 60%, and 90%, respectively (Fig. 4B–D). Notably, M6a-RhFRED L3 exhibited exceptionally excellent selectivity (>98%) toward these three compounds. 1a, 2a, and 3a were characterized by MS and NMR for structural identification. The ^1^H and ^13^C NMR chemical shifts of 11-deoxycorticosterone, 21-hydroxy-16-dehydroprogesterone, and 21-hydroxypregna-4,6-diene-3,20-dione were referenced to previously reported values [40,41]. The corresponding products (1a, 2a, and 3a) were characterized by MS and NMR (Fig. 4C, Figure S14-19, Tables S3–5). Molecular docking analysis exhibited that the distances between the heme iron of M6a and the C21 atoms of the three substrates were 4.8 Å, 4.5 Å, and 4.5 Å, respectively (Fig. 4D). These values (4.5–4.8 Å) are within the characteristic range for P450s-mediated catalysis, supporting M6a′s catalytic competence for C21-hydroxylation.Fig. 4. Bioconversion of PRO analogs by recombinant strain M6a-RhFRED L3. (A) Chemical structures of the PRO analogs used in this study. (B) HPLC analysis of the bioconversion of PRO analogs 1, 2, and 3 (substrate concentration: 200 μM). I: Biotransformed PRO analogs by M6a-RhFRED L3; II: Control bioconversion using E. coli harboring pET28a-RhFRED without P450; III: Purchased standard compounds (1, 2, and 3). Assays were performed in triplicate; representative chromatograms are shown. (C) Mass spectrometry analysis of bioconversion products (1a, 2a, and 3a). Corresponding NMR data are provided in Figs. S14–19. (D) Analysis of conversion rate, selectivity, and molecular docking simulations. The measured distances between the heme iron of M6a and the C21 carbon atom of the substrates were 4.5 Å for 1, 4.5 Å for 2, and 4.8 Å for 3.Fig. 4

Dydrogesterone (1) is an active progestogen that may undergo C21-hydroxylation by human CYP21A2 in the adrenal cortex, generating the metabolite 21-hydroxydydrogesterone (1a) [42]. 16-Dehydroprogesterone (2) and pregna-4,6-diene-3,20-dione (3) are synthetic derivatives of PRO, primarily used as pharmaceutical intermediates or research reagents [43,44]. Recently, Zhao et al. reported that CsCYP21A efficiently catalyzes C21-hydroxylation of fourteen steroids, including PRO and 16-dehydroprogesterone (2), with C21-selectivities of 56.5% and 74.8%, respectively [7]. To our knowledge, this study represents the first in vitro bioconversion of dydrogesterone (1) and pregna-4,6-diene-3,20-dione (3) by P450 enzymes, yielding 21-hydroxydydrogesterone (1a) and 21-hydroxypregna-4,6-diene-3,20-dione (3a).

MD simulation analysis

3.6

To elucidate the molecular basis for the altered selectivity of M6a compared with WT CYP154C5, we attempted to obtain the crystal structure of the PRO-bound complex of M6a. Despite extensive efforts, crystallization of M6a was unsuccessful. Consequently, the M6a structure was predicted using AlphaFold3, which provided a reliable structural basis for our subsequent molecular dynamics (MD) simulations. To assess the reliability of the AlphaFold3-predicted model, the predicted structure of M6a was superimposed on the crystal structure of WT CYP154C5 (PDB: 4J6C), resulting in a backbone RMSD of 0.4 Å for 406 aligned Cα atoms as analyzed by the Dali server (Fig. S20). This indicates that the M6a model predicted by AlphaFold3 is extremely close to the experimentally determined WT structure in terms of the overall main-chain conformation, proving that AlphaFold3 can accurately predict the folding structure of P450 enzymes and can be used for subsequent MD simulations. This approach has also been successfully applied in the engineering of other P450 enzymes.

The differences between M6a and WT were analyzed through molecular docking and MD simulation. First, the catalytically active oxyferryl species of compound I (CpdI) was aligned to the heme positions of WT and M6a to obtain the catalytic intermediate model. Subsequently, the docking model was parameterized with the Amber ff14SB force field and subjected to 100-ns MD simulations using Gromacs. The root-mean-square deviation (RMSD) showed pronounced fluctuations within the first 10 ns for both WT and M6a, after which the trajectories stabilized, indicating a similar trend in both cases (Fig. S21A). Residue flexibility was evaluated by calculating the root mean square fluctuation (RMSF) of Cα atoms over the 100-ns MD trajectory (Fig. S21B).

Substrate-bound conformation was analyzed after simulation stabilization at 80 ns. In the WT, the four rings of PRO tend to lie above the heme, whereas in M6a, the PRO adopted an upright orientation. In the WT, the interaction between the keto group on PRO C3 and the Q239 amino group helped stabilize the substrate-bound conformation (Fig. 5A). In M6a, mutation of six residues (A92, L291, K239, I294, W180, and M398) reshaped the active pocket conformation and altered the substrate binding orientation (Fig. 5C). Mean distances between the iron–oxo species and C16/C21 were 3.7 ± 0.3 Å/4.5 ± 0.9 Å in the WT (Fig. 5B), whereas in M6a they were 4.5 ± 0.4 Å and 3.6 ± 0.3 Å. Thus, in M6a C21 lies closer to the reactive oxygen of Cpd I than in the WT, while C16 lies further away (Fig. 5D). In addition, based on established distance and angle cutoffs [21], all MD simulation conformations were classified as near-attack conformations (NAC) or nonproductive conformations. The NAC was defined by a C⋯O

<svg xmlns="http://www.w3.org/2000/svg" version="1.0" width="20.666667pt" height="16.000000pt" viewBox="0 0 20.666667 16.000000" preserveAspectRatio="xMidYMid meet"><metadata> Created by potrace 1.16, written by Peter Selinger 2001-2019 </metadata><g transform="translate(1.000000,15.000000) scale(0.019444,-0.019444)" fill="currentColor" stroke="none"><path d="M0 440 l0 -40 480 0 480 0 0 40 0 40 -480 0 -480 0 0 -40z M0 280 l0 -40 480 0 480 0 0 40 0 40 -480 0 -480 0 0 -40z"/></g></svg>

Fe (heme) distance <3.8 Å, and a C⋯H⋯OFe (heme) > 150°. The C16⋯OFe (heme) NAC accounts for 24.6% of all conformations in the WT, and the C21⋯OFe (heme) NAC accounts for 20.5% in M6a (Fig. S22). The mutagenesis analysis revealed that subtle modifications in active-site architecture profoundly influence substrate positioning and catalytic regioselectivity.Fig. 5MD simulations of WT and M6a bound with PRO. (A, C) Active site structures of WT and M6a. Six key residues in the active site are shown as cyan and orange sticks in M6a and WT, respectively. The green dashed line shows the hydrogen bond between PRO and Q239 in the WT. Black dashed lines show the distances from the hydroxylation sites (C16 in WT and C21 in M6a) to the OFe (heme). (B, D) Plots showing distances between the C16- and C21-hydroxylation sites of PRO and the OFe (heme) group over simulation time for WT (B) and M6a (D), respectively.Fig. 5

Analysis of the effect of mutations on substrate selectivity

3.7

To further analyze the effects of each mutation on the substrate binding pocket and substrate binding orientation, we superimposed the structures of the WT and M6a (Fig. 6). First of all, we observed that the C ring of PRO acts as the intersection point, with opposite-directional displacements on both sides and the keto oxygen of PRO C3 shifted ∼5 Å toward W180. The reasons for this conformational rearrangement are as follows: The introduction of the bulky, hydrophobic tryptophan side chain at position 180 (F180W) not only reduces the overall volume but also establishes new π-stacking interactions with the pyrrolidine ring of the PRO substrate. This resulted in a significant increase in C21-selectivity, from 72% in M4a to 94% in M5a (Table 1). Concurrently, the Q398 M mutation replaces a polar glutamine with a nonpolar methionine, thereby generating a more hydrophobic environment that promotes substrate binding. Furthermore, introducing the Q239K disrupted original hydrogen bonds, releasing the binding of the keto oxygen of PRO C3 and indirectly promoting movement in the A ring region. The V291L and L294I mutations had the most significant impact, increasing C21-selectivity from 4% (M1) to 36% (M2a) and from 43% (M3a) to 72% (M4a), respectively (Table 1). These mutations demonstrate that minor changes in hydrophobic side chains can dramatically alter substrate selectivity. Notably, these two residues (V291 and L294) are located at the fifth and ninth positions after the EXXR motif, which are recognized as "hotspot" residues for P450 engineering [45]. A similar effect was observed in CYP109B4, where the V292S mutation (corresponding to the ninth residue after the EXXR motif) improved testosterone 15β-selectivity [21].Fig. 6. Superimposition of the active site structures of WT and M6a, both extracted from MD simulations at 80 ns. Six key amino acid residues, PRO and heme, are depicted as sticks, with WT and m6a colored cyan and orange, respectively. The hydroxylation positions (C16 in WT and C21 in M6a) are highlighted as spheres. The keto oxygen of PRO C3 shifts ∼5 Å toward W180 in M6a compared to WT, indicating a steric reorganization of the active site.Fig. 6

Overall, these mutations ultimately led to a rearrangement of the substrate scaffold conformation, with the hydroxylation preference shifting from C16α to C21. This highlights the extreme sensitivity of the P450 system to changes in the spatial, electronic, and hydrophobic properties within its active site.

Conclusions

4

This study successfully obtained the highly efficient variant M6a-RhFRED L3 through rational engineering of the active site of CYP154C5 and optimization of the linker length between P450 and RhFRED. The results demonstrate the feasibility of rational active-site engineering for remodeling enzymatic regioselectivity, as demonstrated by precedent study: the C16β to C15β hydroxylation shift in CYP109-mediated testosterone modification [21]. Most significantly, our engineered CYP154C5 variants demonstrate unprecedented selectivity and high conversion efficiency toward steroid C21-hydroxylation, thereby establishing a novel biocatalytic route for steroid biosynthesis with potential pharmaceutical applications.

CRediT authorship contribution statement

Jian Yang: Writing – original draft, Visualization, Methodology, Investigation. Rong Li: Writing – original draft, Investigation. Qilin Gao: Writing – original draft, Methodology, Investigation. Li Ma: Writing – review & editing, Validation. Qiang Wang: Software, Methodology. Guiying Ma: Investigation. Yikang Fan: Investigation. Shengying Li: Writing – review & editing, Validation, Supervision. Lian-Hua Xu: Writing – review & editing, Validation, Supervision, Resources, Funding acquisition.

Declaration of competing interest

This research was conducted without any financial or personal relationships that could be construed as a potential conflict of interest.