Nonheme Fe Enzyme‐Catalyzed Enantiodivergent Nitrogen Migration: Directed Evolution and Computational Study of Isopenicillin N Synthases for Biocatalytic Synthesis of Arylglycines

Abstract

We describe the reprogramming and directed evolution of nonheme Fe enzyme isopenicillin N synthase (IPNS) as an efficient biocatalyst for 1,3‐nitrogen migration reactions via an unnatural mechanism. Directed evolution of isopenicillin N synthase from Emericella nidulans furnished a quadruple mutant (EniIPNS V185L I187V S102I R279H, IPNS_Nim), enabling the conversion of a range of azanyl esters into N‐protected l‐arylglycines. IPNS_Nim achieved a TTN of 16 000 and a TOF of 1200 min⁻¹. This TTN surpassed state‐of‐the‐art small‐molecule Fe catalysts by 330‐fold and represented the highest TTN value reported for a nonheme Fe enzyme in a new‐to‐nature reaction. IPNS_Nim and our previously evolved ACCO_Nim (ACCO: 1‐aminocyclopropane‐1‐carboxylic acid oxidase) exhibited complementary enantiopreference, allowing enantioselective synthesis of either l‐ or d‐arylglycines—essential building blocks in clinically important peptide therapeutics. Mechanistic studies revealed a biocatalyst‐controlled switch in the rate‐determining step (RDS): While the hydrogen atom transfer (HAT) step is the RDS for ACCO_Nim‐catalyzed nitrogen migration, it is likely not with IPNS_Nim. Moreover, while ACCO_Nim exhibits almost no enantioselectivity in this HAT step, IPNS_Nim confers excellent enantiocontrol over HAT. Computational studies using density functional theory calculations and molecular dynamics simulations suggested that IPNS and ACCO adopt two different substrate binding modes. Classical MD simulations shed light on important interactions between the substrate and active‐site residues that control the substrate binding mode and enantioselectivity.

Nonheme Fe enzyme isopenicillin N synthase was reprogrammed and evolved as an efficient nitrogen migratase IPNS_Nim, converting diverse azanyl esters to valuable l‐arylglycines with up to 16 000 TTN and 97:3 e.r. IPNS_Nim and ACCO_Nim allowed enantiodivergent preparation of both l‐ and d‐arylglycines. Mechanistic studies revealed a change in the rate‐determining step and H atom transfer enantioselectivity for these nonheme enzymes.

Introduction

Nonheme Fe enzymes catalyze a diverse array of transformations, including hydroxylation, halogenation, desaturation and epimerization.^{[ 1} , ² , ³ , ⁴ , ⁵ , ^{6 ]} Owing to their catalytic versatility, these enzymes play a critical role in a variety of essential biological processes such as DNA modification^{[ 7} , ^{8 ]} and post‐translational protein modification.^{[ 9} , ^{10 ]} Compared to heme enzymes,^{[ 11} , ^{12 ]} nonheme Fe centers feature additional open coordination sites that can accommodate multiple substrate fragments simultaneously,^{[ 13 ]} making them an attractive platform for the development of new‐to‐nature enzymatic reagents beyond heme biocatalysis.

Recent studies have shown that nonheme Fe enzymes can be repurposed to catalyze non‐native reactions including nitrene transfer,^{[ 14} , ^{15 ]} fluorine atom transfer,^{[ 16} , ^{17 ]} azide,^{[ 18} , ¹⁹ , ²⁰ , ²¹ , ^{22 ]} thiocyanate^{[ 19} , ²⁰ , ^{21 ]} and isocyanate^{[ 20 ]} transfer as well as H transfer^{[ 23 ]} processes. Despite these advances, new‐to‐nature nonheme Fe biocatalysis remains limited by their relatively narrow substrate scope and/or modest levels of catalytic efficiency. In this context, the development of highly efficient and synthetically useful nonheme Fe enzyme‐catalyzed biotransformations continues to present a significant challenge.

Our lab has been actively developing stereoselective radical reactions catalyzed by heme^{[ 24} , ²⁵ , ²⁶ , ²⁷ , ²⁸ , ^{29 ]} and nonheme^{[ 17} , ²⁰ , ^{30 ]} enzymes. Very recently, inspired by elegant prior studies in transition‐metal catalysis,^{[ 31} , ³² , ^{33 ]} we reported the repurposing and engineering of plant‐derived nonheme Fe enzyme 1‐aminocyclopropane‐1‐carboxylic acid oxidase (ACCO)^{[ 34 ]} as nitrogen migratases for the enantioselective synthesis of N‐protected non‐canonical d‐amino acids from carboxylic acid derivatives (Figure 1a,b).^{[ 30 ]} Mechanistically, this nonheme Fe enzyme‐catalyzed 1,3‐nitrogen migration reaction involves nitrene formation, 1,5‐hydrogen atom transfer (1,5‐HAT), and radical rebound to complete the catalytic cycle (Figure 1b). Through a C─H functionalization logic, these biocatalysts enabled the asymmetric synthesis of non‐canonical amino acids (ncAAs) which are ubiquitous building blocks in bioactive natural products,^{[ 35 ]} peptide therapeutics^{[ 36 ]} and unnatural proteins.^{[ 37 ]} As an important subset of ncAAs, arylglycines available from these biocatalytic methods constitutes structural elements of many clinically significant antibiotics^{[ 38 ]} including vancomycin^{[ 39 ]} and arylomycin (Figure 1c).^{[ 40 ]} Additionally, arylglycines have also been shown to modulate the activity of metabotropic glutamate receptors.^{[ 41 ]} Thus, the availability of efficient nonheme Fe nitrogen migratases offers a promising strategy for the enantioselective biocatalytic production of these valuable chemical entities.

Figure 1.

Nonheme Fe enzyme‐catalyzed 1,3‐nitrogen migration reaction: biocatalytic enantiodivergent synthesis of arylglycines. a) Enantiodivergent synthesis of arylglycines using engineered IPNS and ACCO. b) Proposed reaction mechanism and key mechanistic features of engineered IPNS and ACCO. c) Selected examples of arylglycine containing peptide therapeutics.

Despite these advances, several key challenges still remain. First, we were unable to invert ACCO's enantiopreference to produce l‐amino acids. Given that l‐amino acids are significantly more common than d‐amino acids in natural products and therapeutic agents, identifying an enantiocomplementary nonheme Fe enzyme capable of producing l‐ncAAs with high enantioselectivity remains a critical task. Second, ACCO_Nim exhibited modest chemoselectivity in the production of α‐trisubstituted ncAAs, generating undesired carboxylic acid byproducts via azanyl ester hydrolysis. Strategies to suppress these side reactions would significantly enhance the synthetic utility of nonheme Fe nitrogen migratases.

In this article, we present a study on the mining, engineering and computational analysis of nonheme Fe enzymes, culminating in the development of an enantiocomplementary set of biocatalysts for asymmetric nitrogen migration reactions. We first re‐evaluated our in‐house collection of nonheme Fe enzymes, leading to the discovery of isopenicillin N synthase (IPNS)^{[ 42} , ⁴³ , ^{44 ]} as a promising candidate for l‐amino acid production. Four rounds of directed evolution afforded IPNS_Nim, catalyzing 1,3‐nitrogen migration with a total turnover number (TTN) of 16 000 (Figure 1a). This TTN is 330 times higher than those with small‐molecule Fe‐catalyzed processes^{[ 31} , ³² , ^{33 ]} and represents the highest TTN reported in nonheme Fe‐catalyzed new‐to‐nature reactions. Furthermore, our mechanistic studies revealed a switch in the rate‐determining step and HAT enantiocontrol of ACCO and IPNS‐catalyzed nitrogen migration.

Results and Discussion

We began our investigation by evaluating our in‐house collection of nonheme Fe enzymes for the targeted 1,3‐nitrogen migration reaction using azanyl ester 1a as the model substrate (Figure 2). In previous work, we found that 1‐aminocyclopropane‐1‐carboxylic acid oxidase from Petunia hybrida (PhyACCO, Uniprot ID: Q08506) catalyzed the formation of d‐2a with substantial enantiomeric enrichment.^{[ 34 ]} To identify nonheme Fe enzymes capable of producing l‐amino acids, we performed biocatalytic nitrogen migration reactions in 24‐well plates using whole E. coli cells harboring nonheme Fe enzymes. Among the nonheme Fe enzymes studied, Escherichia coli α‐ketoglutarate (αKG)‐dependent taurine dioxygenase (EcoTauD, Uniprot ID: P37610)^{[ 45 ]} with a two‐histidine‐one‐carboxylate facial triad provided the amino acid product l‐2a in 17% yield and 52:48 l:d ratio with a small preference for l‐2a over d‐2a (Figure 2a, entry 1). Similarly, hercynine oxygenase from Mycolicibacterium thermoresistibile (MthEgtB, Uniprot ID: G7CFI3),^{[ 46 ]} which bears a three‐histidine facial triad, delivered l‐2a in 12% yield and 68:32 l:d ratio (Figure 2a, entry 2). Among all l‐amino acid producing enzymes discovered in our study, Emericella nidulans Isopenicillin N synthase (EniIPNS, Uniprot ID: P05326) exhibited the highest catalytic activity and enantioselectivity, affording l‐2a in 21% yield and 73:27 l:d ratio (Figure 2a, entry 5). Based on this result, we selected wild‐type EniIPNS as the template for further directed evolution. Additionally, our study also led to the discovery of a range of nonheme Fe enzymes to produce d‐2a. These results are summarized in entries 6–10 (see Table S6 for further details).

Figure 2.

Discovery and directed evolution of nitrogen migratases IPNS_Nim for catalytic asymmetric synthesis of l‐arylglycines. a) Evaluation of nonheme Fe enzymes for catalytic asymmetric 1,3‐nitrogen migration using whole E. coli cells. b) Directed evolution of EniIPNS. Reaction conditions: 1a (10.0 mM), E. coli cells harboring EniIPNS variants in M9‐N buffer (pH = 7.4), EtOH (5% v/v), room temperature, 24 h. All the biocatalytic reactions were performed as technical triplicates. c) Active site of wt EniIPNS. Active‐site illustration of wt EniIPNS was made using PyMol from PDB ID 1BK0.^{[ 43 ] a)} Using cell‐free lysate containing ACCO_Nim1 in 100 mM KPi buffer (pH = 7.4), MeCN (5% v/v), Mohr's salt (1 mM), sodium ascorbate (2 mM), room temperature, 6 h.

Starting from wt EniIPNS, we initiated an enzyme engineering campaign using iterative single site‐saturation mutagenesis (SSM) and screening. Guided by the crystal structure of EniIPNS (PDB ID: 1BK0),^{[ 43 ]} we created single SSM libraries at three to four active‐site and tunnel residues per round of engineering. For each single SSM library, 88 clones were screened in a 96‐well plate. Based on our results on ACCO engineering,^{[ 17} , ^{30 ]} we focused on residues within the β‐sheet spanning positions 184–189, particularly those at the entrance of the substrate tunnel. In the first round of engineering, V185L was identified as a beneficial mutation. EniIPNS V185L was found to catalyze l‐2a formation in 8.3% yield and 76:24 l:d ratio (Figure 2b, entry 2), representing a 1.4‐fold improvement in activity (40 turnover numbers (TTN)) compared to wt EniIPNS. We note that TTN values reported in Figure 2b were estimated in a conservative manner based on total nonheme enzyme concentrations in whole cells determined using the SDS‐PAGE assay. With purified IPNS enzyme samples, the apparent enzyme concentration determined by this SDS‐PAGE method was found to be slightly higher than that obtained from bicinchoninic acid (BCA) assays, resulting in a slightly underestimated TTN (see Table S16). In the second round, the inclusion of a critical beneficial mutation I187V led to EniIPNS V185L I187V, affording l‐2a in 61% yield, 86:14 l:d ratio and 280 TTN, corresponding to a 10‐fold enhancement in TTN relative to wt EniIPNS (Figure 2b, entry 3). Notably, both V185L and I187V involved subtle change in side chain steric bulk, underscoring the importance of tunnel residue steric property on both the activity and enantioselectivity in this non‐native transformation.

With EniIPNS V185L I187V, subsequent SSM and screening targeting residues proximal to the Fe center did not provide further improvements in activity or enantioselectivity (see Table S7 for details). In contrast, additional beneficial mutations were unveiled at residues distal to the nonheme Fe center. Mutation of S102 to isoleucine (S102I) resulted in a three‐fold increase in TTN and a substantial enhancement in enantioselectivity (95:5 l:d ratio, Figure 2b, entry 4). Similarly, the R279H mutation further improved catalytic performance (Figure 2b, entry 5). Under standard conditions (OD₆₀₀ = 5), the quadruple mutant EniIPNS V185L I187V S102I R279H furnished l‐2a in 82% yield, 96:4 l:d ratio and 1300 TTN (Figure 2b, entry 6). Notably, the Cα of residues 102 and 279 are located 14.9 Å and 16.4 Å from the Fe center, respectively. These findings underscored the critical role of distal residues in modulating both the activity and enantioselectivity of nonheme Fe enzymes in new‐to‐nature transformations. The final variant EniIPNS V185L I187V S102I R279H is herein designated IPNS_Nim (IPNS nitrogen migratase). Finally, we also measured the TTN of all the IPNS variants from this evolutionary lineage using purified enzymes in the presence of exogenously supplied Fe(II) (Table S10). Results from these measurements qualitatively agreed with those based on whole‐cell measurements, further confirming the enhancement of enzyme activity through directed evolution. In general, higher TTN values were obtained from purified enzyme measurements.

With the final variant IPNS_Nim and our previously evolved^{[ 30 ]} ACCO_Nim1, we next investigated their total turnover number upper limit at reduced enzyme loadings using purified IPNS_Nim and ACCO_Nim1 (Scheme 1). For these TTN measurements, exogenous Fe(II) was supplied to the purified nonheme Fe enzymes (see Tables S15 and S16 for details on protein concentration determination). With a low IPNS_Nim loading of 0.0046 mol%, l‐2a still formed in 75% yield and 95:5 l:d ratio, corresponding to a TTN of 16 000 (Scheme 1, entry 1). Further lowering the enzyme loading provided similar TTNs (Table S15), establishing ca. 16 000 as the TTN upper limit for this IPNS_Nim‐catalyzed nitrogen migration reaction. Using the BCA assay to determine enzyme loading led to an estimated TTN of 28 000 (see Table S16 for details). This TTN is 330‐fold higher than that of the state‐of‐the‐art small‐molecule Fe catalyst^{[ 31} , ³² , ^{33 ]} and approximately 7‐fold higher than that of our engineered ACCO_Nim1 (Scheme 1, entry 2). To the best of our knowledge, this represents the highest TTN reported for nonheme Fe enzyme‐catalyzed unnatural transformation, highlighting the catalytic efficiency of this newly evolved IPNS nitrogen migratase. At a low enzyme loading of 0.0023 mol%, IPNS_Nim showed an initial rate of 0.28 mM·min⁻¹, corresponding to a turnover frequency (TOF) of 1200 min⁻¹. In contrast, under similar conditions, ACCO_Nim1 showed a TOF of 86 min⁻¹. Additionally, chemoselectivity (2a:3a) also improved consistently throughout the evolutionary trajectory of IPNS_Nim. At OD₆₀₀ = 5, IPNS_Nim exhibited higher chemoselectivity (2a:3a = 9.1:1, Figure 2b, entry 6) compared to ACCO_Nim1 (2a:3a = 2.2:1, Figure 2b, entry 7). This high level of chemoselectivity allowed the desired non‐canonical amino acid products to be obtained in excellent yields during preparative‐scale synthesis (vide infra).

Scheme 1.

Evaluating the total turnover numbers (TTNs) of IPNS_Nim and ACCO_Nim1 in biocatalytic enantiodivergent l‐ and d‐2a production. Reaction conditions: 1a (10.0 mM), Mohr's salt (1.0 mM), sodium ascorbate (2.0 mM), purified IPNS_Nim (0.0046 mol%) or ACCO_Nim1 (0.020 mmol%) in 100 mM KPi buffer (pH = 7.4), EtOH (5% v/v, IPNS_Nim) or MeCN (5% v/v, ACCO_Nim1), room temperature. All the biocatalytic reactions were performed in triplicates.

We next examined the substrate scope of our engineered nitrogen migratase IPNS_Nim. To evaluate the influence of steric hindrance, we first tested substrates bearing an ortho‐ (2b), a meta‐ (2c), and a para‐substituted (2d) aromatic ring. All substitution patterns were all well tolerated by IPNS_Nim, although slightly reduced activity was observed with the ortho‐substituted substrate 2b. Azanyl esters bearing an electron‐donating para‐substituent such as a methoxy (2e) and electronically neutral substituents such as a methylthio (2f) were also compatible, providing the corresponding non‐canonical amino acid products in excellent yields and enantioselectivities. Notably, aromatic rings bearing an electron‐withdrawing group such as a trifluoromethoxy (2g), and a strongly electron withdrawing cyano group (2h) were also compatible, resulting in the formation of the corresponding enantioenriched amino acids with excellent yields. The compatibility with electron‐deficient aromatic rings at the α‐position is a distinctive feature of the current evolved nonheme Fe enzyme IPNS_Nim and contrasts with previously engineered heme and nonheme Fe biocatalysts for C(sp³)─H amination, where such substrates usually afforded inferior results.^{[ 30} , ⁴⁷ , ^{48 ]} Substrates bearing a sensitive methyl ester (2i) could also be transformed efficiently with excellent enantioselectivity. Halogenated substrates including a fluorine (2j), a chlorine (2k) and a bromine (2l) group were transformed with excellent yields and enantioselectivities. Other sensitive function groups including an alkynyl (2n) and an acetal group (2o) were also accommodated. Furthermore, substrates possessing a sterically demanding substituent such as a naphthyl (2p) and a para‐tert‐butylphenyl (2q) were also excellent substrates for this IPNS_Nim‐catalyzed nitrogen migration reaction. Finally, heteroaromatic substrates such as a pyridine (2r) and a thiophene (2s) also underwent smooth transformation to the corresponding heterocyclic amino acid products with excellent enantiomeric purity (Figure 3).

Figure 3.

Substrate scope of IPNS_Nim‐catalyzed enantiodivergent 1,3‐nitrogen migration for the synthesis of l‐ and d‐arylglycines. Conditions: 1 (10.0 mM), suspension of E. coli cells harboring IPNS_Nim in M9‐N buffer (pH = 7.4, OD₆₀₀ = 5.0, ca. 0.061 mol% IPNS_Nim). Unless otherwise noted, yields and enantioselectivities were determined by HPLC analysis. See the Supporting Information for details. ^a) 10 mmol scale reaction: 1a (50 mM), E. coli cells harboring IPNS_Nim (from 500 mL TB culture) in M9‐N buffer, total biotransformation volume was 200 mL. Isolated yield was reported. ^b)E. coli cells harboring IPNS_Nim (ca. 0.16 mol%) in M9‐N buffer (OD₆₀₀ = 10.0). ^c)E. coli cells harboring IPNS_Nim (ca. 0.046 mol%) in M9‐N buffer (OD₆₀₀ = 5.0). All the biocatalytic reactions were performed as technical triplicates. See in Table S14 for details of biological replicates of technical triplicates for 2a, 2d, 2e, 2f, 2j, 2k, 2l, and 2o. Enzyme loading in whole‐cell reactions was estimated by SDS–PAGE densitometry, and the corresponding TTN values represented our conservative estimates (see Supporting Information for details).

To further demonstrate the synthetic utility of this IPNS_Nim‐catalyzed non‐canonical amino acid synthesis, we carried out a 10 mmol scale biotransformation. Using IPNS_Nim whole‐cell biocatalysts from only 0.50 L Terrific Broth (TB) culture, 2.24 g N‐Boc l‐phenylglycine (2a) could be prepared with 95:5 l:d ratio, thereby showcasing the scalability and synthetic value of this nonheme Fe enzyme‐catalyzed process.

To further probe the radical nature of these nonheme Fe enzyme‐catalyzed unnatural nitrogen migration reactions, we prepared a cyclopropyl‐containing substrate, (rac)‐4, and subjected it to nitrogen migration reaction conditions with both IPNS_Nim and our previously engineered ACCO_Nim2 (ACCO I184A K158T K172V L186V G156E R175P V236K G173V F250L)^{[ 30 ]} (Scheme 2). Using IPNS_Nim and ACCO_Nim2, dihydropyrone 6 was observed in 1.3% and 14% yield, respectively. This product likely formed via sequential 1,5‐hydrogen atom transfer (HAT) and strain‐releasing cyclopropane ring opening to homoallyl radical 9, followed by a final C─O bond forming radical rebound. Thus, the formation of this ring‐opened product 6 is consistent with the proposed radical mechanism. The formation of this C─O bond forming product indicates that the substrate carboxylate moiety likely remains bound to the nonheme Fe center during catalysis, which is also consistent with the mechanistic hypothesis. When IPNS_Nim was used, reduced carboxylic acid 5 was observed in 51% yield and 75:25 e.r. (entry 1). When ACCO_Nim2 was used, 5 formed in 46% yield and 44:56 e.r. (entry 2).^{[ 30 ]}

Scheme 2.

Radical clock experiment and TEMPO trapping experiment. a) Radical clock experiment. b) TEMPO trapping experiment. ^a) Reaction conditions: 4 (10.0 mM), Mohr's salt (1.0 mM), sodium ascorbate (2.0 mM), purified IPNS_Nim (0.58 mol%) or ACCO_Nim2 cell‐free lysate (OD₆₀₀ = 10.0) in 100 mM KPi buffer (pH = 7.4), EtOH (5% v/v, IPNS_Nim) or MeCN (5% v/v, ACCO_Nim2), room temperature. ^b) 1a (10.0 mM), Mohr's salt (1.0 mM), sodium ascorbate (2.0 mM), TEMPO (10.0 or 20.0 mM), purified IPNS_Nim (0.058 mol%), 100 mM KPi buffer (pH = 7.4), EtOH (5% v/v), room temperature.

To further probe whether this transformation proceeded via a radical pathway, radical trapping experiments were conducted with TEMPO (2,2,6,6‐tetramethylpiperidine‐1‐oxyl). Using 0.058 mol% IPNS_Nim, in the absence of TEMPO, the reaction afforded 2a in 90% yield. The addition of 1.0 equiv TEMPO dramatically lowered the yield of 2a to 3.4%. In the presence of 2.0 equiv TEMPO, the formation of 2a was not observed. In both cases, the corresponding TEMPO trapped product 10 was detected by LC–MS analysis. Collectively, these results were also consistent with a radical‐mediated reaction mechanism.

To further probe the mechanism of this enzymatic C(sp³)─H functionalization‐nitrogen migration, we conducted a series of kinetic isotope effect (KIE) studies (Figure 4). Initial rate measurements were conducted using 1a and its deuterated analog 1a‐d ₂, providing a k _H/k _D value of (1.11 ± 0.09). Importantly, this close‐to‐unity KIE value contrasts sharply with the large KIE value observed with our previously engineered ACCO_Nim1 (k _H/k _D = 5.5 ± 0.3), suggesting a change in rate‐determining step with these two nonheme Fe enzymes. The large KIE values of ACCO_Nim1 indicate that the H atom transfer (HAT) step is rate‐determining in ACCO_Nim1‐catalyzed nitrogen migration reaction. In contrast, the near‐unity KIE for IPNS_Nim suggests that the HAT event is not involved in the rate‐limiting step in IPNS_Nim‐catalyzed nitrogen migration.

Figure 4.

Kinetic isotope effect (KIE) studies and HAT enantioselectivity analysis. a) KIE determined by independently measured initial rates of 1a and 1a‐d ₂. b) KIE determined by intramolecular competition using enantiopure monodeuterated (R)‐1a‐d ₁ or (S)‐1a‐d ₁. c) HAT enantioselectivity analysis for IPNS_Nim and ACCO_Nim1. ^a) Reaction conditions: 1a or 1a‐d₂ (10.0 mM), Mohr's salt (1.0 mM), sodium ascorbate (2.0 mM), purified IPNS_Nim (0.0023 mol%) or purified ACCO_Nim1 (0.013 mol%) in 100 mM KPi buffer (pH = 7.4), EtOH (5% v/v, IPNS_Nim) or MeCN (5% v/v, ACCO_Nim1), room temperature. ^b) Reaction conditions: (R)‐1a‐d₁ or (S)‐1a‐d₁ (10.0 mM), Mohr's salt (1.0 mM), sodium ascorbate (2.0 mM), purified IPNS_Nim (0.0058 mol%) or purified ACCO_Nim1 (0.013 mol%) in 100 mM KPi buffer (pH = 7.4), EtOH (5% v/v, IPNS_Nim) or MeCN (5% v/v, ACCO_Nim1), room temperature. ^c) IPNS favored l‐2a. ^d) ACCO favored d‐2a.

To investigate the origin of enantioselectivity, we synthesized enantiomerically pure, monodeuterated substrates (R)‐ and (S)‐1a‐d ₁ from enantiopure mandelic acid using a previously established procedure.^{[ 30 ]} Using IPNS_Nim as the biocatalyst, (R)‐1a‐ d ₁ afforded l‐2a‐ d ₁ and l‐2a with a ratio of 96:4 as determined by ultra‐performance liquid chromatography‐mass spectrometry (UPLC‐MS), and the l:d ratio was determined to be 96:4. Similarly, when (S)‐1a‐ d ₁ was employed, the same enantiomeric ratio of 96:4 was obtained, but the l‐2a‐ d ₁ to l‐2a ratio shifted to 10:90. These results suggest that IPNS_Nim preferentially abstracts the hydrogen from the same pro‐(S) face, regardless of the absolute stereochemistry of the 1a‐ d ₁ substrate, indicating highly enantioselective HAT with this enzyme. In contrast, with our previously studied ACCO_Nim1, with either (R)‐ or (S)‐1a‐d ₁, d‐2a‐ d ₁ was produced as the major product, indicating a low degree of H atom transfer enantioselectivity and a more pronounced KIE in HAT.^{[ 30 ]}

Using our previously developed methods,^{[ 49 ]} we deconvoluted the enzymatic enantioselectivity (ΔG _{enantioselectivity}) from the intrinsic KIE (ΔG _KIE). For IPNS_Nim, we obtained a large negative value of ΔG _{enantioselectivity} (−1.59 kcal/mol), corresponding to an enantiomeric ratio of 94:6 for the HAT step. In contrast, our previously studied ACCO_Nim1 gave a near‐zero ΔG _{enantioselectivity}, indicating minimal enantioselectivity during the HAT step.^{[ 30 ]} Further analysis revealed that the prochiral hydrogen abstracted by IPNS_Nim is the pro‐(S) hydrogen, and the succeeding radical rebound is largely stereoretentive. Moreover, both (R)‐1a‐d ₁ and (S)‐1a‐d ₁ provided products with identical enantiomeric ratio, suggesting that the radical rebound step further contributes to and ultimately defines the enantioselectivity of the nitrogen migration process.

To gain further insights into the mechanism and origin of enantioselectivity, we performed computational studies to investigate the preferred substrate binding mode and key substrate–active site residue interactions that contribute to stereocontrol (Figure 5). Based on the crystal structure of wt IPNS, H214, D216, and H270 coordinate to the nonheme Fe center, forming a two‐histidine‐one‐carboxylate facial triad (Figure 5b). A water molecule is expected to occupy the binding site trans to H214. This arrangement leaves two potential binding sites for the substrate: one trans to H270 and the other trans to D216, allowing two plausible substrate binding modes in which the NHBoc and carboxylate groups of the azanyl ester are positioned differently.

Figure 5.

Computational study on preferred azanyl ester binding modes. a) DFT‐computed reaction energy profiles for the 1,3‐nitrogen migration with two distinct substrate binding modes. Calculations were performed at the B3LYP‐D3(BJ)/def2‐TZVP/SMD(diethyl ether)//B3LYP‐D3(BJ)/SDD–6–31G(d) level of theory. b) and (c) Active site cavity analysis shows that IPNS_Nim accommodates substrate binding mode A (nitrene trans to His) more favorably, whereas ACCO_Nim1 better accommodates binding mode B (nitrene trans to carboxylate). d) and (e) Key substrate–active site residue interactions in the most populated conformations from classical MD simulations of α‐carboxylate radical intermediate with IPNS_Nim and ACCO_Nim1 in binding modes A and B.

To evaluate whether the facial triad imposes a strong preference for one binding orientation over the other, we first performed density functional theory (DFT) calculations using a simplified active site model to compare the energy landscapes of the two binding modes during 1,3‐nitrogen migration (Figure 5a). The DFT‐computed reaction energy profiles showed that both modes proceed with similar activation energies across all elementary steps during this 1,3‐nitrogen migration, suggesting that the substrate binding mode is not dictated by inherent differences in coordination energetics but rather by the nonheme enzyme active‐site environment. Structural analysis of DFT‐computed intermediates and transition states (Figure 5a) indicate that the NHBoc moiety is more sterically demanding and prefer the less hindered binding site. Enzyme cavity mapping revealed that in IPNS_Nim, the active site provides more space for substrate accommodation trans to H270 than trans to D216 (Figure 5b). Conversely, in ACCO_Nim1, the cavity is more open trans to D179 (Figure 5c). These differences suggest that IPNS_Nim and ACCO_Nim1 favor distinct substrate binding orientations: binding mode A (NHBoc trans to H270) is preferred for IPNS_Nim while binding mode B (NHBoc trans to D179) is preferred for ACCO_Nim1.

This hypothesis was further supported by classical molecular dynamics (MD) simulations of the α‐carboxylate radical intermediate (Figure 5d). In binding mode B, the NHBoc group encounters steric clashes with L223 and L231 in IPNS_Nim, leading to distortion of the tert‐butyl group and misalignment of the α‐carboxylate radical for productive C─N bond formation. In contrast, simulations of binding mode A showed no significant steric repulsion, indicating that this configuration better accommodates the bulky NHBoc group and facilitates efficient radical rebound at the nonheme Fe center. Classical MD simulations with ACCO_Nim1 revealed a different active site environment that favors binding mode B due to unfavorable steric clashes with the β‐sheet in binding mode A (Figure 5e).

Next, we further investigated the origin of enantioselectivity for IPNS_Nim using this preferred substrate binding mode A (Figure 6). Since KIE experiments indicated that the C─N bond forming radical rebound with the α‐carboxylate radical determines the final product enantioselectivity, we performed DFT and MD simulations to identify the factors governing enantioselectivity during this step. DFT calculations of C─N bond forming rebound transition state stereoisomers with a small active site model revealed only a minimal energy difference between the transition states leading to the d‐ and l‐arylglycine products (ΔΔG ^‡ = 0.2 kcal/mol, Figure 6a), suggesting that enantioselectivity is not primarily induced by the chirality of the nonheme Fe center coordinated by the two‐histidine‐one carboxylate triad. Instead, we attribute the observed enantioselectivity to the chiral environment of the active site, which influences the π‐facial selectivity of the radical rebound event with the α‐carboxylate radical.

Figure 6.

Computational study on the origin of enantioselectivity. a) DFT‐computed C─N rebound transition state. ΔΔG ^‡: relative Gibbs free energies of activation with respect to TS3a. b) Key active site residue–substrate interactions stabilizing the most populated near‐attack conformation (NAC) from classical MD simulations leading to the major enantiomeric product l‐2a. c) Distribution of the π/π distance between the carboxylate α‐radical and F211 residue in NAC MD simulations.

To further probe this, we performed three replicas of 500 ns classical MD simulations to model the near‐attack conformations (NACs) preceding the C─N rebound step, leading to both enantiomers of the N‐Boc protected amino acid product. Simulations were focused on pathways that expose either the (Si)‐ or (Re)‐face of the α‐carboxylate radical to the NHBoc group, mimicking transition state structures en route to the major and minor enantiomeric products, l‐2a and d‐2a, respectively (Figure 6b, see Figure S20 for NACs leading to d‐2a). Throughout the 500 ns classical MD simulations, the phenyl group of the α‐carboxylate radical consistently occupies the hydrophobic pocket formed by residues S102I, V185L, and I187V altered during directed evolution. Persistent π/π interactions between this phenyl group and the aromatic side chain of F211 were observed in NACs for both enantiomeric outcome (Figure 6c and Figure S20), but were notably stronger in NACs leading to the major product l‐2a, as indicated by a shorter centroid‐to‐centroid distance (d) between the aromatic rings. In contrast, steric repulsions between the phenyl group of the radical intermediate and L223 were observed in the NACs leading to the disfavored d‐2a product (Figure S21), which is evidenced by a longer distance between the nonheme Fe center and the β‐sheet in the MD simulations (Figure S22). Additionally, under otherwise identical conditions, the use of IPNS_Nim F211G and L223G mutants provided reduced yield and enantioselectivity (see Table S19 for details), suggesting the importance of F211 and L223. Taken together, these results suggest that favorable π/π interactions with F211 and steric repulsion from L223 collectively dictate the π‐facial selectivity of the C─N bond forming radical rebound, thereby controlling the enantioselectivity of the reaction.

Conclusions

In summary, we have developed an orthogonal set of engineered biocatalysts IPNS_Nim and ACCO_Nim1 for the asymmetric 1,3‐nitrogen migration via a new‐to‐nature mechanism. In this research, mining the natural diversity of nonheme Fe enzymes afforded promising starting points for the development of enantiodivergent biocatalysts. With this, directed evolution starting from wt EniIPNS furnished a quadruple mutant IPNS_Nim (EniIPNS V185L I187V S102I R279H), allowing the transformation of a wide range of azanyl esters into the corresponding l‐arylglycines with excellent efficiency, chemoselectivity, and enantioselectivity. IPNS_Nim showed a TTN of 16 000, which is the highest reported TTN for new‐to‐nature transformations catalyzed by a nonheme Fe enzyme. The excellent enzyme activity and chemoselectivity allowed arylglycines to be produced biocatalytically in a scalable manner. Together with the d‐arylglycine‐producing biocatalyst ACCO_Nim1 (PhyACCO I184A K158T K172V L186V G156E R175P V236G G173R), our studies provided an enantiodivergent nonheme enzyme platform allowing either l‐ or d‐arylglycines to be conveniently prepared.

Further mechanistic studies showed that the two nonheme Fe biocatalysts ACCO_Nim1 and IPNS_Nim exhibited distinct features in their rate‐determining step and enantiocontrol during the HAT event. Specifically, kinetic isotope effects using deuterium labeled azanyl ester substrates demonstrated that the HAT step is the rate‐determining step for ACCO_Nim1‐catalyzed nitrogen migration but not the IPNS_Nim‐catalyzed process. Studies of reactions using enantiopure mono‐deutero, mono‐protio substrates revealed that while ACCO_Nim1 showed almost no enantioselectivity in the HAT step, IPNS_Nim exhibited excellent enantiocontrol over HAT. Together, these detailed studies revealed contrasting mechanistic nuances between structurally and evolutionarily related nonheme Fe enzymes in the same new‐to‐nature biocatalytic reaction. Computational studies further illuminated the origin of divergent enantioselectivity between ACCO_Nim1 and IPNS_Nim variants, revealing different substrate binding modes and key substrate‐active site residue interactions enhancing enzyme activity. We expect our studies reprogramming and evolving nonheme Fe enzymes for unnatural 1,3‐nitrogen migration reactions will inspire the further development of synthetically useful biocatalytic reactions using engineered nonheme enzymes.

Conflict of Interests

The authors declare no conflict of interest.

Supporting information

Supporting Information

Lin K., Zhao L.‐P., Wang S., Liu H., Zhang Y., Mai B. K., Liu P., Yang Y., Angew. Chem. Int. Ed. 2026, 65, e24718. 10.1002/anie.202524718.

Data Availability Statement

The data that support the findings of this study are available in the Supporting Information of this article.