A programmable bifunctional flavoenzyme for direct amine-to-ester conversion Authors: Yaoyun Wu, Wei Song, Wanqing Wei, Liming Liu, Jing Wu Published: 2026 ABSTRACT The unification of mechanistically distinct oxidative transformations within a single enzyme active site represents a long-standing challenge in biocatalysis. In particular, flavin-dependent oxidative deamination and Baeyer-Villiger oxidation have remained evolutionarily and mechanistically segregated, raising fundamental questions as to whether their catalytic cycles can be coherently integrated without mutual interference. Here, we report a bifunctional ancestral flavoenzyme, AncFO-221, reconstructed using a function-oriented ancestral sequence reconstruction strategy, which enables direct amine-to-ester conversion within a single active site. Combined experimental and computational analyses reveal a unified catalytic framework in which histidine-assisted proton and hydride transfer during amine oxidation is intrinsically coupled to C4a-peroxyflavin–mediated oxygen insertion, establishing a continuous amine oxidation-Baeyer–Villiger oxidation (AO-BVO) reaction cycle rather than a fortuitous cascade. Guided by this mechanistic unity, modular protein engineering produced an optimized variant, M15, exhibiting an ~18-fold enhancement in catalytic efficiency, near-complete suppression of reductive side reactions, and lactone yields up to 93%. Notably, the engineered enzyme displays programmable and unconventional regioselectivity, preferentially migrating weakly migratory groups across structurally diverse amines, thereby overriding the classical Baeyer-Villiger migratory rule. This study demonstrates that ancestral reconstruction combined with mechanism-guided evolution can merge evolutionarily segregated chemistries into a single, tunable catalytic platform, providing a generalizable blueprint for the design of multistep oxidative biocatalysts. INTRODUCTION Designing enzymes capable of unifying mechanistically distinct transformations within a single active site represents a frontier in synthetic biocatalysis, promising efficiencies and selectivities approaching those observed in naturally evolved multistep enzymatic systems. A paradigmatic case is the direct conversion of amines to esters, which is inherently partitioned in nature: Flavin-dependent amine oxidases catalyze oxidative deamination to ketones (1, 2), whereas Baeyer-Villiger monooxygenases (BVMOs) subsequently oxidize ketones to lactones or esters (3, 4). Amine oxidases catalysis typically produces hydrogen peroxide (H2O2) as a stoichiometric coproduct, which can be deleterious in preparative and whole-cell settings, necessitating peroxide-scavenging enzymes or additives to prevent oxidative damage and side reactions. Although classical chemical strategies using peracids or transition-metal catalysts can accomplish the ketone-to-ester transformation (3, 5), they do not catalyze oxidative deamination, highlighting the persistent functional gap and limitations in selectivity, substrate scope, and sustainability (Fig. 1A) (5, 6). While BVMOs operate under mild conditions with exquisite stereo- and regioselectivity (4), they remain inactive toward amines, further emphasizing this challenge (Fig. 1B). Despite extensive evolutionary diversification, flavin-dependent enzymes conserve the cofactor and mediate a broad spectrum of redox transformations, including olefin reduction (7), (de)halogenation (8), hydroxylation (9), oxidative deamination (1, 2), and Baeyer-Villiger oxidation (BVO) (3). Recent engineering efforts highlight their mechanistic plasticity: Old Yellow Enzymes have been repurposed for the Morita-Baylis-Hillman C─C coupling (10), flavin-mediated oxidative desymmetrization installs quaternary stereocenters in chiral enones (11), and evolved BVMO variants (e.g., AcHYAM) enable photocatalytic hydroamination with yields up to 97% and enantiomeric ratios of 95:5 (12). Nevertheless, precisely coupling sequential redox steps in a single active site—while simultaneously managing reactive by-products such as H2O2—remains largely unexplored, underscoring both the feasibility and transformative potential of unifying amine oxidation (AO) and BVO in one catalyst. Chemical and biocatalytic strategies used in this study. (A) Chemical approaches for BVO. (B) Biocatalytic oxidative deamination and BVO. (C) Design of a bifunctional biocatalyst capable of catalyzing both reactions in this work. The dashed-line substrate representation denotes a generalized amine scaffold and may correspond to either cyclic or linear molecules tested in this study. Flavin-dependent amine oxidases use diverse catalytic mechanisms. In several structurally characterized amine oxidases, oxidative deamination proceeds via histidine-assisted proton abstraction coupled to hydride transfer from the substrate Cα─H to flavin N5 (1), forming an imine that spontaneously hydrolyzes to the corresponding ketone. Other amine oxidase families, including monoamine oxidases, lack a catalytic histidine residue and use alternative mechanistic pathways. This ketone then engages the BVO cycle through a C4a-(hydro/peroxy)flavin intermediate, enabling regio- and stereocontrolled oxygen insertion (3). The partial mechanistic compatibility between amine oxidases and BVMO suggests that integration within a single active site is feasible, provided that proton-relay networks, active-site electrostatics, and peroxyflavin lifetimes are precisely tuned. H2O2 is generated as a stoichiometric by-product during the AO stage. While peroxide-dependent oxygen insertion pathways are known in flavin systems under certain conditions, the present study focuses on the canonical NADPH [reduced form of nicotinamide adenine dinucleotide phosphate (NADP+)]/O2 (oxygen gas)–dependent peroxyflavin mechanism for the BVO step. Motivated by advances in coupled biocatalysis—including conjugate reduction-reductive amination (13), bifunctional methyltransferase-cyclases (14), oxidoreductase/Diels-Alderase (15), and vicinal oxygen chelate-like proteins (16)—and by ancestral sequence reconstruction (ASR)–enabled functional broadening in nucleases (17) and light-emitting dehalogenases (18), we applied a function-oriented ASR strategy to probe multifunctional synergy in flavin enzymes. By integrating evolutionary signals from AO and BVMO lineages, we reconstructed an ancestral flavoprotein scaffold intrinsically predisposed to both chemistries, while structure prediction and computational modeling established a mechanistic framework for the coupled AO-BVO cycle. Here, a bifunctional ancestral flavin oxidase, AncFO-221, was designed to catalyze both oxidative deamination and BVO within a single active site (Fig. 1C). Evolutionary signals from amine oxidases and BVMOs were integrated via ASR, and computational modeling established a mechanistic framework for the coupled AO-BVO cycle. Mechanistic studies and targeted mutagenesis confirmed the catalytic cycle and the roles of key residues. These efforts produced the high-performance variant M15, which exhibits an ~18-fold increase in activity and programmable regioselectivity across diverse amine and thioether substrates. By channeling an AO-generated oxidant into the subsequent oxygen-insertion step, this work further suggests a route to streamline oxidative biotransformations with reduced peroxide management requirements, an aspect that is particularly relevant for whole-cell and process-oriented applications. This work establishes a mechanism-guided paradigm for integrating naturally segregated enzymatic modules, providing a blueprint for programmable, multistep catalytic platforms. RESULTS Design and characterization of a bifunctional flavin oxidase A function-oriented ASR strategy was used to design a bifunctional flavin oxidase with both AO and BVO activities. This strategy was implemented through a two-stage computational workflow integrating AO and BVO functionalities (Fig. 2). In the first stage, AO activity was reconstructed by selecting 28 representative flavin-dependent amine oxidases sequences (EC 1.4.3.-) with well-defined functions, prioritizing enzymes with available crystal structures or those extensively characterized in classical research (e.g., amino acid oxidases, cyclohexylamine oxidase, and related enzymes; see the Supplementary Materials). This ensured phylogenetic diversity and functional breadth. Automated PSI-BLAST homology searches, Clustal Ω multiple sequence alignments, IQTREE-RAxML phylogenetic reconstruction, and PALM-Lazarus ancestral inference were performed using the FireProtASR version 2.0 platform, yielding 28 well-supported ancestral amine oxidases nodes (bootstrap >70) (fig. S1A). In the second stage, BVO functionality was incorporated into the AO-derived framework. The 28 AO ancestors were merged with three previously reported high-activity ancestral flavin amine oxidases, such as EC 1.4.3.14-AncLLysO-7X7K (19–21), and 160 experimentally validated BVO-active enzymes. A second round of ASR was then conducted to reconstruct a comprehensive AO-BVO phylogenetic tree (fig. S1B). Candidate nodes were selected according to three criteria: (i) bootstrap support >70, indicating strong phylogenetic confidence; (ii) evolutionary placement at the convergence of AO and BVO clades, suggesting intermediate functional potential; and (iii) conservation of key catalytic residues, including canonical flavin adenine dinucleotide (FAD)/NADPH-binding motifs, basic residues (e.g., His) essential for AO, and BVMO-like sites critical for peroxy anion stabilization. On the basis of these criteria, four ancestral nodes—AncFO-192, AncFO-197, AncFO-219, and AncFO-221—were identified as promising bifunctional AO-BVO candidates (see the Supplementary Materials for detailed design and sequences). Design scheme for ancestral flavin oxidases with dual AO-BVO functionalities. To evaluate the functionality of the four candidate ancestral nodes (AncFO-192, AncFO-197, AncFO-219, and AncFO-221), their coding sequences were codon optimized for Escherichia coli, synthesized, and cloned into the pET-28a vector for heterologous expression in E. coli Rosetta (DE3). However, SDS–polyacrylamide gel electrophoresis analysis revealed low expression levels, with most protein localized in insoluble fractions (fig. S2A and table S5). To improve solubility and enable functional assessment, an N-terminal small ubiquitin-like modifier (SUMO) solubility tag (22) was introduced, which yielded two soluble variants, SUMO-AncFO-219 and SUMO-AncFO-221. Under identical expression conditions, the soluble protein levels increased by 41.98 and 68.65%, respectively (fig. S2B and table S5). The two soluble ancestral flavin oxidases were subsequently evaluated in the sequential oxidation of cyclohexylamine, yielding three products: cyclohexanone (AO product), ε-caprolactone (BVO product), and cyclohexanol (a reduction by-product of cyclohexanone) (fig. S4). These results confirmed bifunctional AO-BVO activity for both enzymes. The SUMO tag is positioned at the N terminus and separated from the catalytic domain by a linker sequence; structural modeling and mechanistic analyses were performed on the enzyme domain excluding the fusion tag. Among them, SUMO-AncFO-221 produced a higher proportion of ε-caprolactone (4.01%) and was selected for further study. In this system, cyclohexanol was also detected (16.13%) as a reduction product of cyclohexanone, resulting in an ε-caprolactone–to–cyclohexanol ratio of 20:80, thereby indicating pronounced redox promiscuity. Elucidation of the catalytic mechanism of the bifunctional flavin oxidase To investigate the catalytic mechanism of the ancestral bifunctional flavin oxidase AncFO-221, a ternary complex with cofactors FAD and NADPH was predicted using the AlphaFold3 server (23). Quaternary complexes containing cyclohexylamine, cyclohexanone, or ε-caprolactone were subsequently generated by molecular docking (Fig. 3A). Structural analysis revealed that AncFO-221 comprises three domains—the FAD-binding domain, the NADPH-binding domain, and a helical regulatory domain—connected by flexible loops to form a canonical BVMO-like fold. The FAD-binding domain anchors the cofactor through conserved residues [Thr51 (T51), Trp52 (W52), Asp65 (D65), Tyr71 (Y71), and Val116 (V116)] that stabilize FAD via hydrogen bonding and π-π interactions, exemplified by W52 stacking against the isoalloxazine ring. The NADPH-binding domain contains the conserved GXXGXG motif (24) and residues [Thr188 (T188), Thr191 (T191), Gln194 (Q194), Arg213 (R213), and Thr214 (T214)], which stabilize NADPH through a hydrogen bond and cation-π network, including dual contacts of R213 with the O2X/O3X groups of NADPH. The helical regulatory domain coordinates the relative orientation of the FAD- and NADPH-binding domains. Comparison with the classical BVMO [PDB (Protein Data Bank): 4RG3 (25)] (Fig. 3B) indicated that AncFO-221 retains the core cofactor-recognition features but exhibits substitutions around the FAD- and NADPH-binding pockets [e.g., Ser23 (S23), Glu43 (E43), Ser66 (S66), Ala427 (A427), Ala190 (A190), Ala192 (A192), and Val340 (V340)], potentially influencing domain stability. The substrate-binding pocket also displayed marked remodeling: In the classical enzyme, R329 and T435 stabilize the substrate carbonyl and product alkyl groups, whereas in AncFO-221, these positions are occupied by His322 (H322) and Cys425 (C425). H322 contributes to proton transfer during the AO step and stabilizes the carbonyl intermediate, while C425 likely assists in product stabilization and interaction with the FAD isoalloxazine ring (fig. S5A). Notably, ASR retained His322 at the active site despite integrating sequences from diverse AO lineages. This convergence likely reflects structural compatibility between a His-mediated proton relay and the electrostatic requirements of peroxyflavin-dependent oxygen insertion. The His-containing scaffold may thus facilitate coupling of Cα─H activation and Baeyer-Villiger chemistry within a single catalytic pocket. Structurally, the His-centered active-site architecture is preserved within a compact FAD-binding fold that accommodates NADPH interaction without perturbing substrate alignment. The active-site cavity further forms a confined hydrophobic pocket capable of enclosing cyclic amines, thereby facilitating precise orientation of ring-constrained ketone intermediates for subsequent oxygen insertion. Structure, site analysis, and catalytic cycle mechanism of the bifunctional ancestral flavin oxidase. (A) Overall structure of AncFO-221 showing the FAD-binding domain (N-terminal residues 1 to 144 and C-terminal residues 375 to 541; wheat), the NADPH-binding domain (residues 157 to 212 and 325 to 368; sky blue), the helical regulatory domain (residues 228 to 317; brown), flexible loop regions (residues 145 to 156, 213 to 227, 318 to 324, and 369 to 374; light green), and the BVMO signature motif (residues 184 to 189; yellow). (B) Selection of mutational sites. Structural alignment of the AncFO-221-FAD-NADP+-CYH complex (gray; colored labels) with BVMO from Rhodococcus sp. HI-31 PDB ID: 4RG3 (25) (light blue; black labels). Yellow and orange residues indicate FAD-binding sites, blue and purple residues indicate NADPH-binding sites, cyan residues indicate residues shared by FAD and NADPH, and green residues indicate substrate-binding sites. (C) Proposed catalytic cycle of the ancestral flavin oxidase. (D) Catalytic steps of oxidative deamination as elucidated by DFT calculations. (E) Reaction steps of the BVO as elucidated by DFT calculations. Additional DFT-calculated processes, including dehydration of C4a-hydroxyflavin to regenerate oxidized FAD and carbonyl reduction pathways yielding by-products, are shown in fig. S7. Molecular dynamics (MD) simulations were performed to compare substrate positioning and reactive geometries of the amine substrate and the ketone intermediate within the same catalytic pocket. In the AO stage, the amine substrate adopts a defined N5-oriented hydride-transfer pose above the isoalloxazine ring, stabilized by His322 and surrounding hydrophobic residues. The average Cα─N5 distance is 4.83 Å, with an attack angle of 111.4°, and 16.25% of sampled conformations fall within the reactive geometric window, consistent with productive hydride transfer. Following oxidation, the ketone intermediate (CYH) remains accommodated within the same cavity architecture but undergoes local reorientation of its carbonyl group toward the peroxyflavin species. The relevant BVO attack distance, defined as d1 (C1@CYH–O16@FADHOO−), averages 4.45 Å with a β angle of 94.76°, and 5.20% of conformations populate the corresponding reactive window. This transition does not require large-scale backbone rearrangement or formation of an alternative catalytic site. Instead, the shift from the amine-binding pose to the carbonyl-oriented pose is enabled by modest side-chain adjustments and inherent cavity plasticity within the same FAD-centered framework. Thus, rather than combining two chemistries in a rigid environment, the enzyme exploits localized conformational adaptability to support distinct reactive alignments at different catalytic stages. These results provide structural and dynamic evidence that integration of AO and BVO reactivity can be accommodated within a single catalytic cavity without global conformational remodeling. To test these functional predictions, site-directed mutagenesis was performed. The His322→Ala (H322A) substitution abolished productive catalysis, reducing ε-caprolactone formation to ≤1% and arresting the coupled reaction cascade. In contrast, the Cys425→Ala (C425A) variant selectively impaired the BVO step, leading to pronounced accumulation of cyclohexanone while leaving the initial AO largely unaffected (table S3). Together, these conserved and remodeled features define a catalytic framework compatible with the coupled AO-BVO reactivity of AncFO-221. On the basis of previously reported mechanisms and supported by multiple control experiments (fig. S6), an AO-BVO catalytic cycle was proposed for the ancestral flavoenzyme, in which amine substrates (e.g., cyclohexylamine and CYA) are directly converted into lactone products. The cycle consists of eight steps grouped into three stages (Fig. 3C): (i) FAD-dependent AO: H322 abstracts a proton from the α-NH3+ group of the substrate (S; CYA) via transition state [TS1], generating intermediate Int1 (step 1). The Cα─H of Int1 is then transferred as a hydride (H−) to the N5 position of the FAD isoalloxazine ring through [TS2], yielding a protonated imine intermediate (Int2) and reduced flavin (FADH−) (step 2). This step is rate-limiting (21.2 kcal mol−1; Fig. 3D). Int2 subsequently undergoes hydrolysis within the catalytic cavity, accompanied by local reorientation that permits water access, yielding cyclohexanone (Int3; CYH) and ammonia (step 3). Int3 can dissociate from the AO binding pose and be (re)captured in a carbonyl-oriented pose to enter the downstream BVO stage. Notably, this process proceeds solely with FAD, without NADPH or auxiliary dehydrogenases, underscoring the enzyme’s intrinsic oxidative capacity toward amines. The H2O2 generated in this stage is a stoichiometric by-product of flavin reoxidation. The FADH− formed during AO is reoxidized by O2 to regenerate oxidized FAD, completing the AO redox half-cycle. (ii) NADPH-driven BVO: Upon (re)binding of Int3, the ketone intermediate Int3 is converted into ε-caprolactone (Fig. 3E and fig. S7) through a mechanism analogous to classical BVMOs. NADPH transfers a hydride (H−) to FAD via [TS1′], generating NADP+ and FADH− (step 1′; energy barrier, 18.0 kcal mol−1). Reaction of FADH− with O2 produces the peroxyflavin intermediate (FADHOO−; step 4). Int3 reacts with FADHOO− via [TS5] to form a tetrahedral Criegee intermediate (Int4; step 5), which undergoes O─O bond cleavage and group migration through [TS6], yielding ε-caprolactone (P) and C4a-hydroxyflavin (step 6; energy barrier, 13.9 kcal mol−1), followed by product release from the active-site cavity. This stage requires NADPH, which was continuously regenerated by a glucose dehydrogenase/glucose system. (iii) Cofactor recycling and side reaction (fig. S7): C4a-hydroxyflavin spontaneously dehydrates to regenerate oxidized FAD (step 7; energy barrier, 14.4 to 17.4 kcal mol−1), completing the cycle. In parallel, Int3 can be reduced by the enzyme to cyclohexanol (P′; step 8) with an energy barrier as low as 6.2 kcal mol−1, lower than that of the main pathway. This favorable branch reflects the enzyme’s intrinsic reductive activity and accounts for the substantial accumulation of cyclohexanol in the system. Across all steps, the highest barrier (21.2 kcal mol−1) is consistent with feasibility under enzymatic conditions. Thus, AO and BVO are mechanistically linked through sequential reduction and reoxidation of the same FAD cofactor within the catalytic pocket. Overall, this bifunctional flavoenzyme establishes a direct route from amines to esters, integrates FAD- and NADPH-dependent redox chemistry, and exerts intrinsic control over toxic by-products, thereby highlighting mechanistic advantages that distinguish it from conventional BVMO systems. Modular protein engineering enhances a bifunctional flavin oxidase Although AncFO-221 exhibited AO-BVO bifunctionality, its BVO activity remained limited, with ε-caprolactone accounting for only 4.01% of the products and showing poor selectivity (ε-caprolactone:cyclohexanol = 20:80). The deficiency was attributed to three factors: (i) a cofactor-associated high-energy barrier impeding efficient oxidation, (ii) unstable binding of the Criegee intermediate restricting BVO progression, and (iii) a putative carbonyl-reducing catalytic triad promoting by-product accumulation. To systematically enhance AO-BVO activity, a modular protein-engineering strategy was devised (Fig. 4A). Module A targeted the FAD/NADPH–binding region to improve cofactor affinity, module B remodeled the substrate-recognition pocket to stabilize the Criegee intermediate and enhance BVO activity, and module C disrupted the carbonyl-reducing catalytic triad to improve selectivity. Guided by structure-based site-directed mutagenesis, this modular strategy enabled the stepwise construction of a more efficient and selective bifunctional flavin oxidase. Although AncFO-221 was reconstructed from AO-related lineages, module A mutagenesis was guided by structural features conserved in canonical BVMOs to enhance Baeyer-Villiger–type oxygen insertion activity. Key mutations in AncFO-221 and the directed-evolution trajectory, together with reaction composition analysis of the wild-type enzyme and its mutants. Catalytic performance was evaluated based on substrate consumption, intermediate accumulation, by-product formation, and ε-caprolactone production, along with selectivity toward ε-caprolactone (orange dots). (A) Key mutations in AncFO-221 and the directed-evolution trajectory. (B) Reaction composition analysis of the wild-type enzyme and its mutants. M15* indicates the mutant M15 in which the NADPH-recycling enzyme was replaced by formate dehydrogenase. Detailed results of the evolutionary trajectory are provided in fig. S8. Multiple mutant libraries were constructed by integrating MM/PBSA (molecular mechanics/Poisson-Boltzmann surface area) calculations with site-directed mutagenesis, and variants with improved performance were identified through iterative combination screening (Fig. 4B). In module A, four nonconserved residues in the FAD-binding region (S23, E43, S66, and A427) were substituted with conserved counterparts identified by structural alignment of AncFO-221 with the representative BVMO structure 4RG3 (Fig. 3B), generating a stepwise combination library of 15 mutants. From this library, a quadruple mutant M1-4 [Ser23→Gly/Glu43→Asp/Ser66→Thr/ Ala427→Leu (S23G/E43D/S66T/A427L)] increased the ε-caprolactone proportion to 34.71%. In parallel, three residues in the NADPH-binding region (A190, A192, and V340) were mutated to enhance cofactor interactions. A minimal combination strategy, in which individually beneficial substitutions were selectively combined to probe cooperative effects while minimizing combinatorial complexity, generated seven mutants, among which the triple mutant M2-3 [Ala190→Ser/Ala192→Gly/Val340→Ile (A190S/A192G/V340I)] raised the ε-caprolactone proportion to 35.53%. Combination of M1-4 and M2-3 produced mutant M7, which reached 39.01%, indicating a synergistic effect between the domains. Module B was optimized simultaneously. In classical BVMO, R329 (equivalent to H322 in AncFO-221) anchors the substrate carbonyl via a guanidinium group (fig. S5A). By contrast, residues Asp320/Asp321/Pro323/Thr324 (D320/D321/P323/T324) adjacent to H322 in AncFO-221 form a negatively charged cluster, creating electrostatic repulsion with Int3 and hindering CRIInt3 formation. To address this, a library of 15 clones was generated using a progressive charge-superposition strategy, involving stepwise charge-reversal mutations to remodel the local electrostatic environment, yielding the triple mutant M3-3-4 [Asp321→Arg/Pro323→Arg/Thr324→Arg (D321R/P323R/T324R)], which increased the ε-caprolactone proportion to 36.15%. Considering the dual role of C425 in stabilizing both the C5 alkyl hydrogen of ε-caprolactone and the FAD isoalloxazine ring, Cys425→Thr (C425T) was introduced into M7 to generate mutant M8 (M7 + C425T), further raising the ε-caprolactone proportion to 44.46%. To maximize synergistic effects, dominant mutations from modules A and B were combined, producing composite mutants. The final screened mutant, M11-2 (D321R/P323R/T324R + M8), increased the ε-caprolactone proportion to 58.62%. Module C was then targeted. Structural analysis (fig. S5B) identified two potential carbonyl-reducing catalytic triads [Ser172-Tyr57-Arg174 (S172-Y57-R174) and Ser486-Tyr488-Arg489 S486-Y488-R489)] (26). Conformationally disruptive mutations were introduced to suppress the side-reaction pathway: Y → F substitutions removed the proton-donor capacity of the phenolic hydroxyl while preserving the hydrophobic aromatic ring for stability, and S → A substitutions replaced hydroxyl with methyl groups to dismantle the hydrogen-bonding network. The fully integrated mutant M15 [M11-2/Tyr57→Phe/Ser172→Ala/Ser486→Ala/Tyr488→Phe (M11-2/Y57F/S172A/S486A/Y488F)] increased the ε-caprolactone proportion to 73.15% with a yield of 88.70%, while the by-product cyclohexanol was nearly eliminated (ε-caprolactone:cyclohexanol = 96:4, see table S2 and fig. S8). Last, the impact of NADPH regeneration on selectivity was assessed. The conventional BmGDH-glucose system was replaced with a formate dehydrogenase (LbFDH)-sodium formate system, which avoids reductive side reactions. Compared with the glucose dehydrogenase cycle, this system showed greater compatibility with the ketone intermediate and tighter control of selectivity. As a result, the ε-caprolactone proportion increased to 86.70% with a yield of 92.99%, cyclohexanol was undetectable, and the overall selectivity exceeded 99:1. To elucidate the basis for improved activity and selectivity, representative mutants from the evolutionary trajectory—M1-4, M2-3, M3-3-4, M11-2, and the final mutant M15—were analyzed using MD simulations and MM/PBSA energy calculations. In module A, mutant M1-4 remodeled the hydrogen-bonding network of the FAD-binding region, markedly enhancing FAD stability (ΔGbind FAD = −73.7 kcal mol−1; ΔΔGbind FAD = −10.1 kcal mol−1; fig. S9). Mutant M2-3 optimized NADPH binding, lowering ΔGbind NADPH to −34.8 kcal mol−1 (ΔΔGbind NADPH = −16.0 kcal mol−1; fig. S10). Their combination in mutant M7 further strengthened cofactor affinity (ΔΔGbind FAD = −86.4 kcal mol−1; ΔGbind NADPH = −43.2 kcal mol−1; fig. S11). In module B, mutant M3-3-4 introduced a locally positively charged cluster that stabilized the Criegee intermediate CRIInt3, lowering its binding free energy (ΔGbind CRIInt3 = −127.0 kcal mol−1; ΔΔGbind CRIInt3 = −58.8 kcal mol−1) and thereby facilitating its efficient formation to enhance BVO activity. Combination of modules A and B in mutant M11-2 further reinforced cofactor-intermediate interactions (ΔGbind CRIInt3 = −136.9 kcal mol−1; fig. S12). To elucidate the dual enhancement of activity and selectivity in the final mutant M15, six 100-ns MD models (A to F) were constructed to capture key conformations of the AO and BVO stages. Models A (WT-CHA-FAD-NADPH) and B (M15-CHA-FAD-NADPH) represented the AO stage. Models C (WT-CYH-FADHOO−-NADP+) and D (M15-CYH-FADHOO−-NADP+) described nucleophilic attack of peroxyflavin on cyclohexanone, and models E (WT-CRIInt3-NADP+) and F (M15-CRIInt3-NADP+) evaluated the stability of the Criegee intermediate. These control pairs (A/B, C/D, and E/F) enabled systematic analysis of the mechanistic basis for the enhanced performance of M15. Four key aspects were resolved: (i) AO conformational optimization (models A versus B): In M15, the geometry for Cα─H hydride transfer to FAD was optimized, with the catalytic distance d (Cα─H to N5@FAD) shortened from 4.83 to 2.81 Å and the angle θ (Cα─H to the FAD isoalloxazine ring) shifted from 111.4° to 121.5°, close to the transition-state geometry (θ[TS2] = 122.10°). Consequently, AO-active conformations (d ≤ 3.5 Å and θ within [107°, 137°]) increased from 16.25% in wild type (WT) to 30.69% in M15 (Fig. 5A). (ii) Nucleophilic attack geometry (models C versus D): Productive conformations (d1 < 4 Å, β = 76° to 125°) increased from 8.7% in WT to 25.4% in M15. The attack distance d1 (C1@CYH-O16@FADHOO−) decreased from 2.75 to 2.35 Å, and the angle β (O1@CYH-C1@CYH-O16@FADHOO−) was optimized to 108° (Fig. 5, B and C), favoring Criegee intermediate formation. (iii) Criegee intermediate stability (models E versus F): The binding free energy of CRIInt3 decreased from −68.2 kcal mol−1 (WT) to −148.5 kcal mol−1 in M15, representing an 80.3 kcal mol−1 gain in stability (Fig. 5D). (iv) Blocking of the side reaction: Substitutions Y57F/S172A/S486A/Y488F disrupted the carbonyl-reducing catalytic triad by abolishing tyrosine hydroxyl proton donation and dismantling the serine-mediated hydrogen-bond network, thereby eliminating NADPH-dependent hydride transfer and suppressing cyclohexanol formation (fig. S13). In summary, mutant M15 increased the ε-caprolactone proportion from 4.01 to 73.15% and improved selectivity from 20:80 to 96:4, driven by four coordinated structural-functional factors: conformational optimization, facilitation of nucleophilic attack, stabilization of key intermediates, and suppression of side reactions. Mechanistic analysis of the enhanced catalytic performance of mutant M15. (A) Distribution of catalytic conformations during hydride transfer from cyclohexylamine (CHA) in 100-ns MD simulations of WT (M15)-CHA-FAD-NADPH complexes. d (CαH@CHA-N5@FAD) is defined as the distance between the Cα─H of CHA and the N5 of FAD, and θ (CαH@CHA-N5@FAD-N9@FAD) is defined as the angle formed by Cα─H@CHA, N5@FAD, and N9@FAD. (B) Active-site binding modes of cyclohexanone (CYH). Top: WT complex with CYH (magenta), FADHOO− (yellow), and H322 (green). Bottom: M15 complex with CYH (hot pink), FADHOO− (yellow), and residues H322, R321, R323, and R324 (green/cyan). The distance (d1) represents the nucleophilic attack distance between O16@FADHOO− and C1@CYH, and the angle β denotes the Bürgi-Dunitz angle (O1@CYH-C1@CYH-O16@FADHOO−). (C) Distribution of catalytic conformations for peroxyflavin nucleophilic attack on the carbonyl of CYH during 100-ns MD simulations of WT (M15)-CYH-FADHOO−-NADP+ complexes, defined by d1 (C1@CYH-O16@FADHOO−) and β (O1@CYH-C1@CYH-O16@FADHOO−). (D) Binding free-energy distribution of the Criegee intermediate (CRI2c) derived from CYH in WT (M15)-CRI2c-NADP+ complexes, calculated from 4000 snapshots taken over the final 40 ns of MD simulations. The two-dimensional distributions represent the probability density of catalytically relevant geometric parameters sampled during MD simulations. Enrichment within the boxed regions indicates increased sampling of transition state–compatible conformations. To experimentally support these structural-functional factors suggested by the computational analyses, we performed two complementary assays targeting the AO and BVO stages. Time-course profiling of cyclohexylamine oxidation (0 to 24 hours) showed that M15 consumed the amine more rapidly and accumulated downstream products more efficiently than WT, which mainly built up the ketone intermediate with only minimal lactone formation (fig. S14, A and B). This behavior is consistent with the AO-stage conformational features represented in models A and B. To evaluate the Baeyer-Villiger step independently of upstream oxidation, cyclohexanone was also tested as a direct substrate. Under identical conditions, M15 generated markedly higher amounts of ε-caprolactone, whereas WT afforded only trace levels (fig. S14, C and D), in agreement with the predicted stabilization of the Criegee-like intermediate and the improved BVO-stage configuration in M15 (models C to F). Furthermore, reverting one of the newly introduced arginine residues (M15-R321D) markedly decreased lactone formation and led to a corresponding increase in ketone accumulation, indicating a weakened BVO step consistent with the loss of Criegee-stabilizing interactions (fig. S15). These results corroborate the functional contribution of the arginine-rich pocket to the enhanced BVO reactivity of M15. Expansion of the substrate scope and regioselectivity of the bifunctional flavin oxidase mutant M15 As shown in Fig. 6, mutant M15 catalyzed the AO-BVO cascade for all tested amine substrates, affording the corresponding lactone or ester products and thereby demonstrating robust bifunctional activity. M15 efficiently converted 5- to 8- and 12-membered alicyclic amines (1a to 5a), where 1a corresponds to cyclopentanamine, into the corresponding lactones (1b to 5b) (see Fig. 6 for the chemical structures and reaction scheme), with 1- to 10-mmol-scale yields ranging from 34.17 to 92.99%, with lactone-to-alcohol selectivity consistently exceeding 99:1, indicative of broad alicyclic tolerance and excellent chemoselectivity. For medium- and long-chain α-aliphatic amines (6a to 8a), M15 produced regioisomeric esters pairs (6b to 8b/6b′ to 8b′) in total yields of 32.33 to 58.06%. Notably, unlike the canonical migratory aptitude observed in both classical BVO chemistry and natural BVMO catalysis, M15 preferentially migrated the less favorable methyl group, yielding abnormal regioisomers (6b′ to 8b′) as the predominant products with regioselectivities of 74 to 99%. This atypical migratory preference was also observed with aromatic amines (9a to 10a) and phenolic derivatives (11a to 12a), where abnormal regioisomers (9b′ to 12b′) dominated the ester products, with total yields of 29.39 to 88.50%. For the sterically demanding substrate 13a, containing an indane skeleton with the amino group directly attached to the alicyclic carbon, M15 generated esters (13b/13b′) in a 59.33% total yield, with the abnormal isomer (13b′) prevailing (13b:13b′ = 15:85), underscoring the enzyme’s regioselectivity and adaptability toward sterically hindered substrates. Beyond amines, M15 also catalyzed S-BVO of complex azole thioethers (14a and 15a), affording sulfoxides 14b and 15b in 71.05 and 42.12% yields, respectively, without detectable by-products. These heterocycles are key precursors of proton-pump inhibitors, conventionally synthesized using strong oxidants and metal-based catalysts (27, 28). The identities and systematic names of substrates 1a to 15a are provided in section S1.1. Together, the broad substrate scope and distinctive regioselectivity establish M15 as a versatile AO-BVO biocatalytic system. Notably, this system proved readily adaptable: For representative substrates 1a and 9a, modest fine-tuning of standard reaction parameters (pH, temperature, and enzyme loading) delivered substantial yield improvements (~20 to 25% absolute increase) while preserving the overall product distribution patterns (table S4). Substrate scope of mutant M15. Biotransformations were carried out with substrates 1a to 15a (1a to 5a, 10 mM; 6a to 13a, 5 mM; 14a to 15a, 1 mM), NADPH (1.2 mM), FAD (2.4 mM), mutant M15 (5 mg ml−1, ~66.7 μM), and LbFDH (100 μM). These standardized screening conditions were used to enable a comparison across structurally diverse substrates; representative substrates affording moderate yields under this baseline setup were further examined under modestly adjusted reaction parameters, as detailed in table S4. Detailed experimental procedures, including equations for product yield and regioselectivity ratio calculations, quantitative composition of the reaction systems (fig. S16), and comparative data for the WT enzyme (fig. S17), are provided in the Supplementary Materials. n.d., not determined. Despite substantial differences in substrate structures and electronic properties, mutant M15 consistently favored migration of groups with weaker migratory aptitude (e.g., methyl), thereby producing abnormal ester isomers (b′) across medium- and long-chain aliphatic amines, aromatic amines, and sterically hindered indane derivatives. This regioselectivity diverges from that of classical BVO, in which migratory aptitude typically follows the order aryl > primary alkyl > methyl, with strongly electron-donating aryl groups (e.g., p-hydroxyphenyl) migrating more readily than moderately donating (m-hydroxyphenyl) or unsubstituted phenyl groups (29). In conventional BVO, peroxyflavin attack on the carbonyl generates a Criegee intermediate (30), and the group with greater migratory aptitude preferentially migrates, yielding the “normal” ester isomer. In contrast, the marked preference of M15 for weakly migratory groups suggests that its selectivity arises from altered substrate recognition or conformational modulation of the intermediate. To probe the origin of this atypical regioselectivity, MD simulations were performed on enzyme-intermediate complexes of Criegee intermediates (CRI6c to CRI13c) derived from representative ketones (6c to 13c). A key geometric parameter—the dihedral angle defined by O(peroxy1)─O(peroxy2)─C(carbonyl)─C (migrating group)—was analyzed to reflect migratory conformation. When this dihedral approaches ±180°, the migrating group adopts an antiperiplanar configuration relative to the breaking peroxy bond, strongly favoring migration (31). This dihedral distribution was systematically examined across complexes to evaluate conformational bias. Notably, in all simulated systems, the dihedral distribution was skewed toward antiperiplanar conformations favoring migration of weakly migratory groups such as methyl. For example, in CRI6c, the methyl-migrating conformation (dihed1-abnormal) reached 98.57%, whereas the conventional n-pentyl–migrating conformation (dihed1-normal) was below 0.1%. Similar trends were observed across other models (Fig. 7 and figs. S18 to S24). These results indicate that the active-site microenvironment of M15 enforces geometric rearrangement of the Criegee intermediate, conferring an antiperiplanar advantage to otherwise disfavored groups. This conformational bias enables circumvention of the classical Baeyer-Villiger rule, in which selectivity is dictated by migratory aptitude, thereby stabilizing and directing the formation of unconventional products. Such conformational steering provides a compelling mechanistic explanation for the anomalous regioselectivity exhibited by M15. Distribution of dihedral angles of the key Criegee intermediate formed from 6c (heptan-2-one) in the active site of M15. (A) Representative active-site conformation of the intermediate, with the relevant atoms labeled. (B) Time evolution of the corresponding dihedral angles during the 100-ns MD simulation. Dihedral values near ±180° (i.e., 150° ≤ angle ≤180° and −180° ≤ angle ≤ −150°) indicate geometries permissive for migration. Blue dots represent dihed-normal ∠O(peroxy1)─O(peroxy2)─C(carbonyl)─C(pentyl), corresponding to formation of the conventional product, whereas orange dots represent dihed-abnormal ∠O(peroxy1)─O(peroxy2)─C(carbonyl)─C(methyl), corresponding to formation of the abnormal product. Distributions of dihedral angles for the Criegee intermediates formed from other ketone intermediates (7c to 13c) in M15 are shown in figs. S18 to S24. DISCUSSION Flavin-dependent amine oxidases and BVMOs catalyze valuable oxidations, but their functional segregation has hindered direct amine-to-ester conversion within a single enzyme, leaving the question of whether these activities could be unified unresolved. In this study, ASR was applied to merge amine oxidase and BVMO evolution, yielding an ancestral flavoenzyme capable of one-step amine-to-ester biotransformation. Guided by mechanistic insights, modular engineering of the AncFO-221–generated mutant M15, which exhibited an 18-fold increase in activity, improved chemoselectivity from 20:80 to 96:4 (ε-caprolactone:cyclohexanol). Mutant M15 also exhibited regioselectivity distinct from native BVOs and efficiently converted diverse amines and drug-like compounds under mild, oxidant-free conditions. In this study, a bifunctional ancestral flavoenzyme, AncFO-221, which catalyzes one-step amine-to-ester conversion, was constructed by integrating evolutionary information from flavin-dependent amine oxidases and BVMOs by ASR. To date, ASR has been applied primarily to enhance the stability or activity of single-function enzymes, including the engineering of thermostable ene-reductases for industrial applications (32), the use of machine learning for optimal mutant prediction (33), and the reconstruction of P450 monooxygenases with fatty acid C─H activation (34). In combination with sequence similarity networks, ASR has also enabled the reprogramming of FeII [iron(II)]/α-ketoglutaric acid–dependent hydroxylases into ring-expanding oxidases (35). Despite these successes, ASR has rarely been applied to bifunctional systems. Existing strategies for constructing bifunctional enzymes—such as natural screening, domain fusion, or directed evolution—often lack mechanistic rationale and depend on serendipity. For example, enzymes exhibiting both conjugate reduction and reductive amination activities were identified from natural imine reductases (13), chimeric constructs combining As(III) methyltransferase and radical S-adenosyl-L-methionine domains enabled dual methylation and adenosylation (36), and directed evolution produced ChaP (cysteine, histidine–dependent amidohydrolase/peptidase) mutants with combined ortho-quinone cleavage and decarboxylation activities (37). Conventional amine-to-ester conversions typically involve sequential metal- or peracid-mediated oxidations, which require complex operations and exhibit poor compatibility with multienzyme systems (5, 38–40). Although multienzyme cascades such as dehydrogenase-BVMO assemblies can mediate alcohol-to-ester transformations (41), no natural or engineered enzyme has been reported to catalyze amine-to-ester conversion through an integrated or cascade route. In this study, a modular mutagenesis strategy was designed to target cofactor binding, substrate recognition, and side-reaction suppression of a bifunctional flavoenzyme, AncFO-221, to enable synergistic engineering and concurrent improvements in both activity and selectivity. A variety of strategies have been developed to enhance enzymatic performance. Loop remodeling has improved catalytic efficiency and substrate adaptability in enzymes such as triosephosphate isomerase, protein tyrosine phosphatase, and β-lactamase (42). Mechanism-guided active-site redesign, aided by MD and Rosetta modeling, has increased the C─N bond–forming activity of AspB from <5% to >99% (43) and enabled efficient polyethylene terephthalate depolymerization by hydrolases (44). Electrostatic field preorganization has enhanced the activities of alcohol dehydrogenase (45) and acid phosphatase (46) by 50- and 2.7-fold, respectively. Transition-state reprogramming has yielded superior catalysts, including amino acid oxidase M5 (total turnover number, 96,929 to 175,249) (2) and ω-transaminase M9 [kcat (turnover number)/Km (Michaelis constant) improved by 1660-fold] (47). Reaction selectivity has likewise been modulated by two main approaches. One involves conformational and electrostatic tuning, exemplified by R124 and D125 reshaping the nucleophilic trajectory in SNAr (nucleophilic aromatic substitution) to yield SNAr1.3 with >99% stereoselectivity (48). The other mimics hierarchical enzyme regulation through the introduction of second-sphere amines or biomimetic selenium centers, which increased H2O2 (49) selectivity to 97% or reduced by-product levels to 0.039% (50). In contrast to strategies that focus solely on either activity or selectivity, this study integrates mechanistic dissection, modular division, and multipoint engineering into a unified framework that simultaneously reinforces key catalytic steps while suppressing side reactions, thereby establishing a broadly applicable paradigm for optimizing both enzyme efficiency and specificity. In this study, mutant M15 exhibited regioselectivity distinct from native BVOs and efficiently transformed diverse amines and drug-like compounds under mild, oxidant-free conditions. Natural BVMOs, such as cyclohexanone monooxygenase (CHMO), display a strong tertiary-carbon migration bias in alicyclic substrates but are by narrow activity ranges and restricted substrate tolerance. Structural redesign of AcCHMO improved thermostability [+5°C in Tm (melting transition temperature)], extended half-life eightfold, and modestly increased activity (1.5- to 2.5-fold at 0.5 mM cyclohexanone), yet its practical biotransformation capacity remains untested (51). By contrast, M15 retained the natural tertiary-carbon migration preference while achieving markedly enhanced activity toward alicyclic amines. For example, it converted cyclohexylamine to ε-caprolactone in a 92.99% yield, underscoring its potential for scalable biocatalysis. Traditional BVOs rely on electronic effects to dictate regioselectivity, restricting their utility in regioisomer synthesis. To overcome this bias, catalytic strategies such as β-peptide catalysis and boronate ester templating have been developed. The former reverses conventional regioselectivity through conformational locking and hydrogen bonding (52), while the latter directs Criegee intermediate migration by reversible coordination with substrate hydroxyl groups (53). However, both approaches suffer from substrate dependency and synthetic complexity. In contrast, M15 intrinsically modulates regioselectivity by reshaping the active site to induce Criegee rearrangements without auxiliary molecules. This property enables regioselective outcomes complementary to those of classical BVOs across diverse substrates, substantially expanding the biocatalytic repertoire for regioisomer synthesis. Although AncFO-M15 enables direct amine-to-ester conversion in a single active site, further improvements in activity, robustness, and substrate generality will be needed for broad deployment. A particularly attractive feature of the integrated AO-BVO architecture is the potential channeling of AO-generated H2O2 into peroxyflavin formation for oxygen insertion, which could increase effective oxidant utilization while limiting free-peroxide accumulation; in whole-cell settings, such coupling may mitigate oxidative stress and reduce reliance on additional peroxide-detoxification measures. Looking ahead, combining high-throughput screening with structure- and mechanism-aware design (including AlphaFold3/cryo–electron microscopy–informed engineering) should enable more precise tuning of proton-relay networks, active-site electrostatics, and peroxyflavin lifetimes to refine Criegee control and selectivity. In parallel, expanding function-oriented ancestral reconstruction may uncover additional bifunctional scaffolds with distinct pocket architectures and reactivity windows. Last, integrating AncFO-based modules into microbial cell factories could streamline oxidative upgrading routes and simplify process configurations, provided that peroxide handling and redox balance are coengineered to minimize fitness penalties associated with oxidative stress. Together, these directions position AO-BVO unification as a transferable blueprint for programmable oxidative biocatalysis at the interface of catalyst design and process simplification. MATERIALS AND METHODS All experimental and computational procedures are detailed in the Supplementary Materials and include materials (section S1.1); construction of the WT ancestral flavin enzyme (section S1.2) and mutant libraries (section S1.3); cultivation, expression, and purification of ancestral flavin enzymes (section S1.4); activity assays (section S1.5); analytical methods, including gas chromatography and gas chromatography–mass spectrometry (section S1.6) as well as high-performance liquid chromatography and mass spectrometry (section S1.7); and determination of component distribution, product yield, reaction selectivity, and regioselectivity (section S1.8), and isolation and spectroscopic characterization of representative products (section S1.9); reconstruction of bifunctional ancestral flavin oxidases (section S2.1); preparation of initial structures for computational studies (section S2.2); molecular docking (section S2.3); MD simulations (section S2.4); density functional theory (DFT) calculations (section S2.5); and supplementary sequence information, relevant supplementary tables, figures, and computational data (sections S3 to S6) are also provided.