Nonheme Fe 1,3-nitrogen migratases for asymmetric noncanonical amino acid synthesis

Abstract

Despite their excellent potential, nonheme Fe enzymes remain largely underexploited in the development of new-to-nature biocatalytic reactions with significant synthetic utility. Herein, we report the repurposing and directed evolution of plant-derived nonheme Fe enzyme 1-aminocyclopropane-1-carboxylic acid oxidase (ACCO) to achieve the highly efficient and enantioselective 1,3-nitrogen migration reaction. Evolved nonheme 1,3-nitrogen migratases enabled the enantioselective amination of both secondary C(sp³)–H bonds from a prochiral methylene and tertiary C(sp³)–H bonds from racemic substrates, giving rise to highly enantioenriched α-trisubstituted as well as α-tetrasubstituted non-canonical amino acids with broad utility. Furthermore, our tailored nitrogen migratase allowed the asymmetric construction of notoriously challenging methyl-ethyl stereocenters via enantioconvergent C(sp³)–H functionalization. As the key design element, the availability of multiple open coordination sites of the nonheme Fe center allowed simultaneous binding of the carboxylate substrate and the nitrogen-centered radical to Fe, permitting the efficient intramolecular 1,5-hydrogen atom transfer (HAT) that is not feasible with heme systems. Combined experimental and computational studies and quantitative free energy relationship analysis revealed an unusual stereoablative HAT followed by a stereoselective C–N bond forming radical rebound mechanism, which accounts for the excellent enantiocontrol observed in biocatalytic amino acid synthesis.

Introduction

Nonheme Fe enzymes catalyze an astonishing array of challenging biological transformations underlying essential metabolism, including C–H hydroxylation, C–H halogenation and desaturation reactions^1–3. Beyond their fascinating native nonheme enzymology, the structural and functional diversity of naturally occurring nonheme enzymes, along with their versatile Fe coordination chemistry, offer a powerful platform for discovering non-native enzymatic reactions. Furthermore, the additional open coordination sites of the nonheme Fe center relative to heme-based systems permit the binding of multiple substrate fragments, thereby allowing challenging new-to-nature enzymatic reactions to be advanced beyond the capabilities of heme biocatalysis^4,5.

Despite its tremendous potential, new-to-nature stereoselective nonheme biocatalysis—especially those with high synthetic value—remains largely underdeveloped. Elegant prior studies have exploited nonheme Fe enzymes to facilitate nitrene transfer^6,7 and azidation reactions^8–11 with varying degrees of enantiocontrol. Recently, the Huang group and our group independently engineered nonheme Fe enzymes as effective fluorine atom transfer enzymes to catalyze asymmetric C(sp³)–H fluorination via a radical rebound mechanism^12,13. Inspired by this fluorine atom transfer enzymology^12,13 and notable small-molecule Ru and Fe-catalyzed C(sp³)–H amination^14,15, we envisioned a new-to-nature nonheme Fe enzyme-catalyzed asymmetric nitrogen group migration as a unifying means to synthesize both α-trisubstituted and tetrasubstituted non-canonical amino acids^16,17 from abundant and easily available carboxylic acids (Fig. 1a–c). In this proposed catalytic cycle (Fig. 1d), the nonheme Fe enzyme (I) would first catalyze the conversion of azanyl ester 1 into an Fe-bound nitrogen centered radical II via α-elimination. This Fe-bound nitrogen radical would then undergo an intramolecular 1,5-hydrogen atom transfer (1,5-HAT) to furnish the corresponding carboxylate α-radical. At this stage, intramolecular radical rebound with the Fe-bound nitrogen group (−NHR) would allow for enantioselective C–N bond formation (III). Release of the non-canonical amino acid product 2 from III would regenerate the ferrous nonheme enzyme, thereby completing the catalytic cycle. This design leverages the multiple open coordination sites of the nonheme Fe center, enabling simultaneous binding of the reactive nitrogen-centered radical and the substrate carboxylate. This facilitates the intramolecular 1,5-HAT—a transformation that is difficult to achieve with heme enzymes, which possess only a single available coordination site. As azanyl esters 1 could be easily prepared from the corresponding carboxylic acids, the successful implementation of this proposal would furnish a synthetically valuable nonheme Fe enzyme-catalyzed stereoselective preparation of non-canonical amino acids with widespread applications as key building blocks in peptide therapeutics¹⁸, bioactive natural products¹⁹, and functional proteins²⁰.

Fig. 1.

Directed evolution of new-to-nature nonheme Fe nitrogen migratase for the enantioselective synthesis of non-canonical amino acids: an overview. a, Nonheme Fe enzyme-catalyzed enantioselective amination of secondary C(sp³)–H bonds. b, Enantioconvergent amination of tertiary C(sp³)–H bonds. c, Enantioconvergent construction of methyl-ethyl stereogenic center. d, Proposed catalytic cycle for the nonheme Fe enzyme-catalyzed asymmetric nitrogen migration reaction.

We hypothesized that this nonheme Fe enzyme-catalyzed nitrogen migration process would be compatible with carboxylic acid derivatives featuring either an α-monosubstituted or an α,α-disubstituted α-carbon. Such broad compatibility would enable the enantioselective amination of secondary C(sp³)–H bonds in a prochiral methylene unit (Fig. 1a) and the enantioconvergent amination of tertiary C(sp³)–H bonds of racemic substrates (Fig. 1b) with excellent stereoselectivity. Furthermore, the directed evolution of nonheme Fe nitrogen migratases could facilitate the asymmetric construction of minimally differentiated methyl-ethyl stereogenic center with good enantiocontrol (Fig. 1c)—a challenging task eluding the state-of-the-art small-molecule transition-metal catalysts. Moreover, evolved nonheme Fe enzymes may exhibit catalytic activities orders of magnitude higher than small-molecule Fe complexes. Herein, we report the successful repurposing and directed evolution of nonheme Fe enzymes as highly efficient and enantioselective nitrogen migratases, introducing a novel and synthetically valuable function to the catalytic repertoire of nonheme Fe enzymes.

Results and discussion

Discovery of new-to-nature 1,3-nitrogen migratase activity through nonheme Fe enzyme mining

At the outset of this study, using a range of azanyl esters (1) as the substrate, we evaluated an in-house collection of different types of nonheme Fe proteins by high throughput experimentation using 96 or 24-well plates (Fig. 2a). Among all the substrates with various N-substituents we evaluated, N-alkoxycarbonyl groups, including tert-butyloxycarbonyl (Boc, 1a), 2,2,2-trichloroethoxycarbonyl (Troc, 1b), benzyloxycarbonyl (Cbz, 1a-2) and methoxycarbonyl (CO₂Me, 1a-3) exhibited higher activity than N-acyl substituents such as benzoyl (Bz, 1a-4) and pivaloyl (Piv, 1a-5). Among substrates possessing these N-substituents, Boc (1a) and Troc (1b) were found to display the highest activity across a range of nonheme Fe enzymes we evaluated. Among all the nonheme enzymes we evaluated, with N-Boc (1a) and Troc (1b) substrates, the α-ketoglutarate (αKG)-dependent taurine dioxygenase from Escherichia coli (TauD, Uniprot ID: P37610)²¹ provided the corresponding nitrogen migration product 2a and 2b in 16% yield (54:46 e.r.) and 10% yield (77:23 e.r.), respectively (Fig. 2a; see supplementary Table 1 for further details). Another (αKG)-dependent nonheme enzyme, the ethylene-forming enzyme from Pseudomonas savastanoi (EFE, Uniprot ID: P32021)²² afforded 2a and 2b in 10% yield (57:43 e.r.) and 7% yield (52:48 e.r.), respectively. Furthermore, other nonheme Fe enzymes, which are not αKG-dependent, also showed promising initial results in this nitrogen migratase activity. For example, the use of 4-hydroxyphenylpyruvate dioxygenase from Streptomyces avermitilis (HPPD, Uniprot ID: Q53586)²³ featuring a two-histidine-one-carboxylate facial triad resulted in the formation of 2a and 2b in 13% yield (32:68 e.r.) and 8% yield (48:52 e.r.), respectively. Among all the nonheme Fe enzymes that we evaluated, a plant-derived enzyme, namely 1-aminocyclopropane-1-carboxylic acid oxidase from Petunia hybrida (ACCO, Uniprot ID: Q08506)^24,25 displayed good initial activity and excellent enantioselectivity in this 1,3-nitrogen migration process. In particular, with ACCO, the N-Boc-substituted product 2a formed in 15% yield and an excellent e.r. of 95:5. Similarly, the N-Troc-substituted product 2b was produced in 17% yield with slightly inferior, but still good, enantioselectivity (81:19 e.r.). These results mirrored our previous findings in the evaluation of nonheme Fe enzymes in enantioselective fluorine atom transfer, highlighting the utility of ACCO in group migration activities allowing C–H functionalization. This two-dimensional high-throughput experimentation provided valuable insights into the substrate structure-enzyme activity relationship. Due to the superior initial activity and excellent enantioselectivity observed with ACCO and substrate 1a, this enzyme-substrate combination was selected for further directed evolution.

Fig. 2. Development of nonheme Fe enzyme ACCO_Nim for new-to-nature asymmetric 1,3-nitrogen migration: enzyme mining and directed evolution.

a, High-throughput evaluation of nonheme Fe enzymes across a range of azanyl esters with diverse N-protecting groups; see the supplementary Table 1 for details. b, Directed evolution of ACCO_Nim1 as highly effective nitrogen migratase for enantioselective non-canonical amino acid synthesis. Active-site illustration of ACCO was made using PyMol from PDB ID 1WA6. ACCO = 1-aminocyclopropane-1-carboxylic acid oxidase, His = histidine, Asp = aspartate, Glu = glutamate, e.r. = enantiomeric ratio, TTN = total turnover number. Reaction conditions: 10.0 mM azanyl ester 1, nonheme Fe enzyme cell-free lysate in 100 mM KPi buffer (pH = 7.4), 1.0 mM Fe(NH₄)₂(SO₄)₂, 2.0 mM sodium ascorbate, MeCN (5% v/v), in Coy anaerobic chamber, room temperature, 6 h. TTN was determined by the SDS-PAGE analysis using cell-free lysate. All the biocatalytic reactions were performed as technical triplicates. See the Supplementary Information for details.

Directed evolution of nonheme Fe nitrogen migratase ACCO_Nim1 for the enantioselective synthesis of α-trisubstituted non-canonical amino acids

To improve the activity and enantioselectivity of nonheme Fe nitrogen migratases, we further evaluated our ACCO lineage from our previous C(sp³)–H fluorination study (Fig. 2b, see supplementary Table 3 for further details). Among all the ACCO variants in our collection, ACCO I184A exhibited the highest activity and enantioselectivity ((32 ± 2)% yield, 95:5 e.r., (310 ± 20) total turnover number (TTN), Fig 2b, entry 2). We hypothesized that the I184A mutation close to the Fe center might further open up the active site, thus better accommodating the non-native substrate. ACCO I184A was chosen as the parent for the directed evolution of an effective and enantioselective nitrogen migratase. As distal residues from the substrate tunnel were found to have a significant impact on the activity of our previously developed ACCO fluorine transferase¹², we first targeted amino acid residues in the 89–93 and 156–160 β-sheets at the entrance of the substrate tunnel (Supplementary Table 7 and 8). Site-saturation mutagenesis (SSM) and screening were employed as the diversification strategy. In each SSM library, 88 clones were selected, grown in 96-well plates, and analyzed by chiral high performance liquid chromatography (HPLC) for enzyme activity and enantioselectivity. In these β-sheets, a single beneficial mutation K158T was identified, affording ca. 1.2-fold improvement in activity ((37 ± 1)% yield, 92:8 e.r., (340 ± 10) TTN, entry 3). In contrast to the engineering of fluorine transfer enzyme, no additional beneficial mutations were found, indicating a departure of enzymatic control mode in the present nonheme nitrogen migratase.

We next turned our attention to amino acid residues proximal to the nonheme Fe center, particularly those from the 169–173 α-helix and the adjacent loop as well as the 182–186 β-sheet. Through SSM and screening, two important beneficial mutations K172V and L186V were discovered, leading to notable enhancement in the yield of product 2a. The quadruple mutant ACCO I184A K158T K172V L186V furnished 2a in (64 ± 2)% yield, 92:8 e.r. and (520 ± 20) TTN (entry 5). We further focused our efforts on the improvement of enzyme enantioselectivity. Additional four rounds of SSM and screening by targeting active-site residues led to ACCO I184A K158T K172V L186V G156E R175P V236G G173R, which is eight mutations away from the wild-type ACCO, affording amino acid product 2a in (71 ± 1)% yield, 99:1 e.r., and (810 ± 10) TTN. All the four newly identified mutations, including G156E, R175P, V236G, and G173R, resulted in a steady improvement of enantioselectivity (92:8 e.r., entry 5 → 99:1 e.r., entry 9). Additionally, the chemoselectivity of the nonheme enzyme also improved throughout directed evolution, as evidenced by improved ratio of amino acid 2a : reduced carboxylic acid 3a (see the Supplementary Information for details). We named this final octuple mutant ACCO_Nim1 (ACCO nitrogen migratase 1). By further lowering the loading of ACCO_Nim1 via decreasing the density of E. coli cell suspension overexpressing ACCO_Nim1 (OD₆₀₀ = 3, ca. 0.026 mol% enzyme loading), a TTN of (2,260 ± 40) was achieved with a slightly decreased enantioselectivity (entry 11). This total turnover number was approximately 200-times higher than those of small-molecule Fe catalysts¹⁵, highlighting the catalytic efficiency of the nonheme Fe enzyme.

With newly evolved nonheme Fe nitrogen migratase ACCO_Nim1, we set out to examine the substrate scope of this biocatalytic enantioselective synthesis of non-canonical amino acids. Azanyl esters with other N-protecting groups, such as the 2,2,2-trichloroethyloxycarbonyl (Troc, 2b) group which can be conveniently removed under reductive conditions²⁶ were also transformed with similar activity and enantioselectivity. Aromatic rings bearing an ortho- (2c), a meta- (2d) and a-para- (2e) substituent were compatible, indicating the insensitivity of ACCO_Nim1 towards changes in substrate’s steric properties. In addition, halogen functional group handles including a fluorine (2f), a chlorine (2g), and a bromine (2h) were readily tolerated. Electron-donating groups including a methoxy (2i), a methylthio (2j) and a phenol (2o) group as well as electron-withdrawing substituents including a trifluoromethoxy (2k) and an ester (2n) were also compatible with this biocatalytic enantioselective nitrogen migration process. A sterically encumbered tert-butyl group (2l) at the para-position of the aromatic ring was also tolerated. Furthermore, arylglycines possessing sensitive functional groups, including a thioether (2j), an alkyne (2m), a methyl ester (2n) and a free phenol (2o), could be obtained with excellent enantiocontrol, highlighting the mild reaction conditions of this biocatalytic amino acid synthesis. Disubstituted (2p) and bicyclic aromatic groups such as benzodioxole (2q), naphthyl (2r) and indole (2s) could also be effectively transformed, providing the corresponding non-canonical amino acids with excellent enantioselectivity. Furthermore, heterocycles including an indole (2s), a pyridine (2t), a thiophene (2u, and 2v) and a furan (2w) were readily accommodated by this biocatalytic process. Finally, amino acids with an α-alkenyl (2x and 2y) could be produced with excellent enantioselectivity.

Enantioconvergent 1,3-nitrogen migratase ACCO_Nim2 enables access to α-tetrasubstituted non-canonical amino acids

We further challenged our evolved ACCO nitrogen migratase to synthesize α-tetrasubstituted amino acids via enantioconvergent C–H functionalization using racemic carboxylic acid derivatives. Initially, to identify an excellent starting point for this task, we evaluated our ACCO evolutionary lineage from the engineering of the current ACCO_Nim1 and the previous ACCO_CHF using (rac)-4a as the model substrate in a 24-well plate format (Fig. 4a). In this ACCO variant library evaluation, ACCO I184A K158T K172V L186V G156E R175P (entry 7, Fig. 2a; E5, Fig. 4a), an intermediate variant from the directed evolution of ACCO_Nim1, furnished the corresponding α-tetrasubstituted amino acid 5a with a promising activity and enantioselectivity (67% yield and 88:12 e.r.). In this screening, several variants possessing mutations at 236 and 173 displayed superior activity and/or enantioselectivity compared to the unmutated parent. Thus, these sites were targeted in enzyme engineering to evolve a more effective enantioconvergent C–H functionalizing nitrogen migratase for α-tetrasubstituted amino acid synthesis. With this ACCO I184A K158T K172V L186V G156E R175P variant as the parent, three rounds of iterative SSM and screening led to the discovery of V236K, G173V and F250L as beneficial mutations, giving rise to a substantially improved enantioconvergent nitrogen migratase ACCO_Nim2. ACCO_Nim2 catalyzed the formation of α-tetrasubstituted amino acid 5a with (84 ± 2)% yield, 99:1 e.r., and (770 ± 10) TTN in an enantioconvergent fashion. Furthermore, by further lowering the loading of ACCO_Nim2, a TTN of (2210 ± 40) was achieved albeit with a slightly decreased yield and enantioselectivity (Fig. 4a; see supplementary Table 16 for further details).

Fig. 4. Directed evolution of ACCO_Nim2 for the asymmetric synthesis of α–tetrasubstituted non-canonical amino acids via enantioconvergent C–H functionalization.

a, High-throughput valuation of ACCO_Nim1 and ACCO_CHF evolutionary lineage and directed evolution of ACCO_Nim2 for the enantioconvergent synthesis of α-tetrasubstituted amino acids. The heat map generated based on the screening yield with each grid cell represents a unique variant. The yield is indicated in white, while the percentage of major enantiomer is shown in dark red for representative variants. See the supplementary Table 12 for details. Active-site illustration was made using PyMol from PDB ID: 1WA6. E5 = ACCO I184A K158T K172V L186V G156E R175P. b, Substrate scope of ACCO_Nim2-catalyzed enantioconvergent synthesis of α-tetrasubstituted non-canonical amino acids. Reaction conditions: 10.0 mM azanyl ester (rac)-4, ACCO_Nim2 cell-free lysate in 100 mM KPi buffer (pH = 7.4, ca. 0.055–0.11 mol%), 1.0 mM Fe(NH₄)₂(SO₄)₂, 2.0 mM sodium ascorbate, MeCN (5% v/v), in Coy anaerobic chamber, room temperature, 6 h. TTN was determined by the SDS-PAGE analysis using cell-free lysate. All the biocatalytic reactions were performed as technical triplicates. See the Supplementary Information for details. ^aIsolated yield after esterification.

Using ACCO_Nim2 as the biocatalyst, we surveyed the scope of this enantioconvergent α-tetrasubstituted amino acid synthesis (Fig. 4b). Aryl groups with an electron-donating substituent such as a methyl (4b) and a methoxy (4c) as well as an electron-withdrawing substituent such as a trifluoromethyl (4d), were found to be compatible with this biocatalytic nitrogen migration. In addition, α-tetrasubstituted amino acid with an ethyl α-substituent (5e), a cyclic moiety at the α-position (5f) and a thienyl (5g) group could also be prepared with excellent enantiocontrol. Furthermore, azanyl esters lacking an α-aryl substituent could also be effectively transformed via the amination of unactivated C(sp³)–H bond, providing the corresponding fully aliphatic α-tetrasubstituted amino acid (5h) with 91:9 er. Together, these results illustrate the excellent evolvability of nonheme Fe enzyme ACCO for new-to-nature asymmetric nitrogen migration.

Biocatalytic construction of challenging methyl-ethyl stereocenters via enantioconvergent tertiary C(sp³)–H functionalization

Given the excellent evolutionary potential, we questioned whether we could further engineer ACCO nitrogen migratases to construct a minimally differentiated methyl-ethyl stereocenter via enantioconvergent C–H functionalization (Fig. 5a). Methyl and ethyl are two of the simplest substituents in organic chemistry with minimal steric and electronic differences. Thus, the effective differentiation of methyl and ethyl for the construction of “methyl-ethyl” stereocenters represents a notorious challenge in asymmetric catalysis.²⁷ In particular, the catalytic asymmetric construction of “methyl-ethyl” stereocenters via enantioconvergent functionalization of tertiary C(sp³)–H bonds is rare. To initiate this study, we first evaluated the ACCO_Nim1 and ACCO_Nim2 evolutionary lineage. Although WT ACCO exhibited a low level of activity with nearly no enantioselectivity ((10 ± 1)% yield, 52:48 e.r.), the newly evolved nitrogen migratase variant ACCO_Nim2 allowed the enantioconvergent biotransformation of (rac)-4i, providing α-tetrasubstituted amino acid 5i with a methyl-ethyl stereocenter in (58 ± 1)% yield and 72:28 e.r.. With ACCO_Nim2 as the parent, we initiated a brief directed evolution campaign by targeting active-site residues and substrate tunnel residues by SSM and screening. Four additional beneficial mutations, including A248T, S246F, A180F and G156T, were introduced to ACCO_Nim2, resulting in a new nitrogen migratase variant ACCO_Nim3 (Fig. 5b). ACCO_Nim3 catalyzed the enantioconvergent C–H functionalization with (55 ± 2)% yield and 88:12 e.r., providing a rare example of enantioconvergent synthesis of α-tetrasubstituted amino acid possessing a methyl-ethyl stereocenter.

Fig. 5. ACCO_Nim3-catalyzed enantioconvergent construction of methyl-ethyl stereocenter via tertiary C(sp³)–H functionalization and gram-scale biocatalytic amino acid synthesis.

a, Directed evolution of ACCO_Nim3 for the enantioconvergent construction of methyl-ethyl stereocenter. All the biocatalytic reactions were performed as technical triplicates. Active-site illustration was made using PyMol from PDB ID: 1WA6, b, Gram-scale biotransformations with ACCO_Nim1 and ACCO_Nim2. Reaction conditions: 36.4 mM azanyl ester, ACCO_Nim cell-free lysate in 100 mM KPi buffer (pH = 7.4), 3.6 mM Fe(NH₄)₂(SO₄)₂, 7.3 mM sodium ascorbate, MeCN (3.6% v/v), room temperature, 6 h. See the Supplementary Information for details.

To further demonstrate the synthetic utility of our newly evolved nonheme Fe nitrogen migratases, we performed biocatalytic amino acid synthesis on a 10 mmol scale with two azanyl ester substrates, including 1a bearing a secondary α-C(sp³)–H bond and (rac)-4a bearing a tertiary α-C(sp³)–H bond. For preparative applications, 1a and (rac)-4a could be conveniently synthesized from commercially available and inexpensive carboxylic acids or acyl chlorides via robust amidation reactions. Using the cell-free lysate of ACCO_Nim1 and ACCO_Nim2 produced from 0.5 L expression culture, 1.55 g 2a (62% yield, 98:2 e.r.) and 1.87 g 5a (71% yield, 99:1 e.r.) were isolated, respectively. The enantioselectivity of these 10 mmol scale reactions was found to be identical to analytical scale processes (Fig. 5c). The ability to conveniently carry out gram-scale synthesis of highly enantioenriched non-canonical amino acids at higher substrate titers showcased the synthetic potential of engineered nonheme Fe enzymes to produce value-added products.

Mechanistic studies with radical clock probes and deuterium-labeled substrates

To gain further insights into the mechanism of this nonheme Fe enzyme-catalyzed nitrogen migration process, azanyl esters (Z)-1x and (E)-1x bearing a cis- and trans-olefin, respectively, were synthesized and subjected to this biocatalytic nitrogen migration. Starting from (Z)-1x, (E)-2x bearing a trans-olefin formed exclusively in 27% yield and 97:3 e.r.. Intriguingly, the reduction side product carboxylic acid 3x formed with an (E)-configuration exclusively. These results suggest the intermediacy of an α-radical of carboxylic acid, which presumably arose from the nonheme Fe enzyme-catalyzed intramolecular 1,5-hydrogen atom transfer (1,5-HAT). Furthermore, the complete scrambling of olefin geometry indicated that this α-radical is relatively long lived and the C–N bond forming radical rebound is slower than the olefin E/Z isomerization at the α-radical stage. Next, we prepared the radical clock substrate (rac)-7 possessing a cyclopropyl group at the benzylic position and applied it to the nonheme Fe enzyme-catalyzed process. In addition to the formation of the reduction product 8 (46% yield), dihydropyrone 9 was also observed in 14% yield. We reasoned that 9 likely formed via a 1,5-HAT/cyclopropane ring opening/carboxylate rebound process. The discovery of carboxylate rebound product suggested that the newly formed carboxylate derived from the azanyl ester substrate is likely bound to Fe center of the nonheme enzyme.

To further understand the mechanism including the rate- and enantioselectivity-determining steps, we performed a series of kinetic isotope experiments. First, we prepared benzylic deuterated substrate 1a-d₂ and independently determined the initial rates with both 1a and 1a-d₂. These independent kinetic measurements with 1a and 1a-d₂ showed a k_H/k_D of (5.5 ± 0.3). These results indicated that the C(sp³)–H cleavage in hydrogen atom transfer is involved in the rate-determining step. Second, we prepared enantiomerically pure monodeutero, monoprotio substrates (R)-1a-d₁ and (S)-1a-d₁ by multistep synthesis starting from optically pure mandelic acid (see the Supplementary Information for details). Using (R)-1a-d₁ and (S)-1a-d₁, we evaluated ACCO_Nim1 and ACCO_Nim2, which were evolved for secondary and tertiary C(sp³)–H amination in this study, respectively. The secondary C(sp³)–H aminating enzyme ACCO_Nim1 was found to exhibit a similar k_H/k_D value of (5.6 ± 0.3) and (6.4 ± 0.5), respectively, when (R)-1a-d₁ and (S)-1a-d₁ were applied. In contrast, the tertiary C(sp³)–H aminating enzyme ACCO_Nim2 showed a marked difference in k_H/k_D values when (R)- and (S)-1a-d₁ were used as the substrates. Using (R)-1a-d₁, a large k_H/k_D value of (9.3 ± 0.8) was observed with ACCO_Nim2. By contrast, a much smaller k_H/k_D of (2.1 ± 0.1) was observed with (S)-1a-d₁.

With these results, we next performed quantitative activation free energy analysis for both ACCO_Nim1 and ACCO_Nim2. Our KIE results with 1a and 1a-d₂ and QM/MM studies (vide infra) both showed that the HAT step in this nitrogen migration reaction is irreversible. The 2a-d₁/2a ratio is thus determined by two energy terms, including ΔG_KIE, which reflects the kinetic isotope effects in HAT, and ΔG_{enantioselectivity}, which reflects the enzymatic enantiopreference in HAT for non-deuterated substrate (1a). When (S)-1a-d₁ was used as the substrate, both the kinetic isotope effect and the enzymatic enantioselectivity favor the abstraction of the pro-(R)-H, thus resulting in a higher 2a-d₁/2a ratio. In contrast, with (R)-1a-d₁, the inherent kinetic isotope effect is mitigated by the enzymatic enantiopreference for the abstraction of the pro-(R)-D, leading to a lower 2a-d₁/2a ratio. Using our previously developed free energy analysis, we dissected the two effects and calculated the energy terms ΔG_KIE and ΔG_{enantioselectivity} for both ACCO_Nim1 and ACCO_Nim (see the Supplementary Information). With nonheme Fe enzyme ACCO_Nim1 evolved for secondary C(sp³)–H amination, a close-to-zero ΔG_{enantioselectivity} of −0.04 kcal/mol was observed with the ΔG_KIE being −1.06 kcal/mol. This ΔG_{enantioselectivity} corresponds to an enantiomeric ratio (e.r.) of 52:48, indicating a lack of enzymatic enantioselectivity in the HAT step. On the other hand, with ACCO_Nim2 evolved for tertiary C(sp³)–H amination, a modest, but non-zero, ΔG_{enantioselectivity} of −0.43 kcal/mol was determined with a ΔG_KIE of −0.86 kcal/mol. ACCO_Nim2’s. ΔG_{enantioselectivity} corresponds to an e.r. of 67:33, suggesting a small degree of enantioselectivity in the HAT step. Together, these results demonstrated that directed evolution is a powerful means to manipulate enzymatic stereoselectivity in each individual step in a complex overall catalytic cycle. Furthermore, these data reveal that for the secondary C(sp³)–H aminating nonheme enzyme ACCO_Nim1, the overall enantioselectivity reflects the stereoselective radical rebound step succeeding the HAT step. Finally, the ΔG_KIE term for ACCO_Nim1 and ACCO_Nim2 corresponds to a k_H/k_D value of 52:48 and 67:33, respectively. These results are consistent with the k_H/k_D value obtained from independent initial rate measurements with 1a and 1a-d₂.

These results further prompted us to compare this nonheme Fe enzyme with previously engineered heme enzymes in different C(sp³)–H amination processes. Engineered serine-ligated P450 variant P411_Diane2 displayed a ΔG_{enantioselectivity} of −1.14 kcal/mol for HAT, which corresponds to an e.r. of 87:13. Additionally, histidine-ligated myoglobin variant Mb* showed a ΔG_{enantioselectivity} of −0.28 kcal/mol (62:38 e.r.) for HAT²⁸ (see the Supplementary Information for mathematical derivatization details). Collectively, these results revealed that among all the heme and nonheme Fe enzymes developed for diverse C(sp³)–H amination reactions, the current nonheme enzyme ACCO_Nim1 is the only system where the enzymatic HAT step is almost non-enantioselective while the radical rebound step exhibits excellent enantioselectivity. This finding is consistent with our results that ACCO_Nim-catalyzed amination of tertiary C(sp³)–H bonds is fully enantioconvergent with minimal kinetic resolution of (rac)-4.

Computational study by QM/MM calculations

We performed hybrid quantum mechanics/molecular mechanics (QM/MM) calculations using the ONIOM method²⁹ to compute the reaction energy profiles of the enantioselective 1,3-nitrogen migration of azanyl ester 1a catalyzed by the evolved nonheme enzyme ACCO_Nim1 (see the Supplementary Materials for computational details and additional results using a truncated model). After binding of substrate 1a to the Fe center and N–H deprotonation, intermediate 13 undergoes facile N–O bond cleavage via TS-1 with a low activation barrier of 4.7 kcal/mol relative to 13, generating a Fe(III)-bound nitrogen radical 14. 14 undergoes 1,5-HAT^14,15 via TS-2 to form α-carboxylate radical 15. Subsequent C–N bond forming radical rebound^28,30–33 via TS-3 affords the 1,3-nitrogen migration product complex 16. The computed activation barriers for the 1,5-HAT (TS-2) and radical rebound (TS-3) steps are 17.5 and 17.1 kcal/mol relative to 14 and 15, respectively. Therefore, the QM/MM-calculated reaction energy profile suggests that the C–H cleavage via 1,5-HAT is irreversible and involved in the rate-determining step, which agrees with the primary KIE observed experimentally (k_H/k_D = 5.5, vide supra).

To investigate the origin of enantioselectivity, we computed the enantioselectivity in both the HAT and radical rebound steps, as well as the rate of conformational change between the two α-carboxylate radical conformers (15 and 15′) generated from the competing 1,5-HAT pathways that cleave the enantiotopic C–H^R and C–H^S bonds, respectively. The computed activation free energy difference (ΔΔG^‡) between the competing 1,5-HAT transition states TS-2 and TS-2′ is only 0.5 kcal/mol, suggesting a low level of enantioselectivity during the C–H cleavage, which would form a mixture of 15 and 15′. Our metadynamics simulations³⁴ using the QM/MM method (see the Supplementary Materials for computational details) revealed that the conformational change from 15 to 15′ requires a relatively low barrier of 6.4 kcal/mol, which is substantially faster than the C–N radical rebound. Therefore, under these Curtin–Hammett conditions, the enantioselectivity of the overall transformation is determined by the C–N bond forming radical rebound step. In contrast to the low levels of enantioselectivity in HAT, the radical rebound is predicted to be substantially more enantioselective, with a computed ΔΔG^‡ of 5.0 kcal/mol between TS-3 and TS-3′, suggesting a much higher level of enantioselectivity in this C–N bond formation step. This finding is also consistent with the high levels of enantioselectivity observed experimentally. The disfavored rebound transition state TS-3′, which involves the C–N bond formation at the (Si)-face of the α radical center is destabilized by steric repulsions between the Ph group of the α-carboxylate radical and the F250 residue and the Boc protecting group (Fig. 7b), whereas in the favored rebound transition state TS-3, these steric repulsions with the Ph group are diminished as the Ph points away from these bulky groups. The lower enantioselectivity between the 1,5-HAT transition states (TS-2 and TS-2′) is attributed to their more flexible six-membered ring compared with the five-membered C–N rebound transition state, which requires minimal energy penalty to distort the Ph group away from F250 and Boc groups in TS-2′. No significant steric interactions with the α-carboxylate radical were found in either TS-2 or TS-2′, leading to nearly identical barriers for these competing transition states. Taken together, our experimental and computational studies reveal a unique enantioinduction mechanism. In this nonheme Fe enzyme-catalyzed reaction, the prochiral α-carboxylate radical intermediate undergoes a rapid stereoablative conformational change. The high level of enantioselectivity is determined by the C–N bond formation radical rebound and not by the intramolecular HAT.

Fig. 7. Computational studies.

a, Computed reaction energy profiles of the ACCO_Nim1-catalyzed 1,3-nitrogen migration. b, QM/MM-optimized 1,5-HAT (TS-2 and TS-2′) and C–N bond forming radical rebound (TS-3 and TS-3′) transition states. QM/MM calculations were performed at the ONIOM(B3LYP-D3(BJ)/def2-TZVP:Amber)//ONIOM(B3LYP-D3(BJ)/6-31G(d)–SDD(Fe):Amber) level of theory.

Conclusion

In summary, we developed a nonheme Fe enzyme-catalyzed stereoselective new-to-nature nitrogen migration reaction to convert abundant carboxylic acid derivatives into valuable non-canonical amino acids. Directed evolution culminated in a panel of highly efficient and enantioselective nonheme nitrogen migratases, allowing both α-trisubstituted and α-tetrasubstituted non-canonical amino acids to be prepared via enantioselective amination of secondary and tertiary C(sp³)–H bonds, respectively. Importantly, detailed mechanistic and computational studies revealed that the 1,5-HAT step lacks enantioselectivity. In contrast, the C–N bond forming radical rebound is highly enantioselective and thus constitutes the enantioselectivity determining step for both secondary and tertiary C(sp³)–H amination reactions. This radical-enabled stereoablation-enantioconvergent transformation mechanism provides an underexploited strategy to develop highly stereoselective processes. Beyond the synthetic utility of this biocatalytic method, we expect the use of multiple coordination sites of nonheme Fe center to drive otherwise difficult enzymatic processes will inspire the further development of new-to-nature reactions catalyzed by nonheme enzymes.

Methods

Expression of ACCO_Nim variants.

E. coli BL21(DE3) cells harboring recombinant plasmid encoding the appropriate ACCO variant were grown overnight at 37 ° and 230 rpm in 4 mL LB media supplemented with 0.05 mg/mL kanamycin (LB_kan). Preculture (1.0 mL, 5% v/v) was used to inoculate 20 mL TB media supplemented with 0.05 mg/mL kanamycin (TB_kan) in a 125 mL Erlenmeyer flask. The culture was incubated at 37°C and 230 rpm for 2 h to reach an OD₆₀₀ of ca. 1.5. The culture was then cooled on ice for 20 min and induced with 0.5 mM isopropyl β-d-1-thiogalactopyranoside (IPTG, final concentrations). Protein expression was performed at 20 ° and 200 rpm for 20 h. The cells were then transferred to a conical tube (50 mL) and harvested by centrifugation (3,434 g, 4 min, 4 °C) using an Eppendorf 5910R tabletop centrifuge.

Analytical scale enantioselective biocatalytic nitrogen migration.

The suspension of E. coli cells expressing ACCO in KPi buffer (100 mM, pH = 7.4, typically OD₆₀₀=5–20, 1.5 mL) was added to a 24-well plate and kept on ice. Cells were then disrupted by sonication using a Qsonica sonicator equipped with a 24-tip horn (45% amplitude, 2 secs on, 4 secs off, 18 min in total (sonication time: 6 min)). To pellet the cell debris, the resulting lysate was then centrifuged at 21130 g and 4 ° for 30 min using an Eppendorf 5424R centrifuge. The clarified cell-free lysate (360 μL per well) was then added to a new 2 mL vial and transferred into a Coy anaerobic chamber. In the anaerobic Coy chamber, the new mixed solution of ferrous ammonium sulfate and sodium ascorbate (20 μL, 20 mM ferrous ammonium sulfate and 40 mM sodium ascorbate in dd H₂O), and the azanyl ester substrate (20 μL, 200 mM in MeCN) were added. Final reaction volume was 400 μL; final concentration of the azanyl ester substrate was 10 mM. (Note: reaction performed with E. coli cells resuspended to OD₆₀₀ = 10 indicates that OD₆₀₀ = 10 cells were then disrupted by sonication and 360 μL clarified cell-free lysates was added after centrifuge, and likewise for other reaction OD₆₀₀ descriptions.) The vials were sealed and shaken in a Corning digital microplate shaker at room temperature and 680 rpm for 6 h. The reaction mixture was then analyzed by chiral HPLC.

Preparative gram-scale enantioselective biocatalytic 1,3–nitrogen migration.

E. coli BL21(DE3) cells harboring the recombinant plasmid encoding the specific ACCO_Nim variants were grown overnight at 37 ° and 230 rpm in 25 mL LB media supplemented with 0.05 mg/mL kanamycin (LB_kan). Preculture (10 mL, 4% v/v) was used to inoculate 250 mL TB media supplemented with 0.05 mg/mL kanamycin (TB_kan) in a 1 L Erlenmeyer flask. The culture was incubated at 37 °C and 230 rpm for 2 h to reach an OD₆₀₀ of ca. 1.2. The culture was then cooled on ice for 30 min and induced with 0.5 mM isopropyl β-d-1-thiogalactopyranoside (IPTG, final concentrations). Protein expression was performed at 20 ° and 200 rpm for 20 h.

500 mL E. coli cells were harvested by centrifugation (4,347 g, 15 min, 4 °C) using an Eppendorf tabletop centrifuge 5910R and resuspended in 50 mL KPi buffer (100 mM, pH = 7.4, OD₆₀₀ = 100). Cell suspensions in KPi buffer were kept on ice and lysed by sonication using a Qsonica Q500 sonicator (45% amplitude, 2 secs on, 4 secs off, 18 min in total per cycle (sonication time: 6 min), 2 cycles were applied); samples were carefully submerged in wet ice to avoid overheating during this process. To pellet the cell debris, lysates were centrifuged using a Lynx 6000 superspeed centrifuge (28,928 g, 30 min, 4 °C). To a 1 L Erlenmeyer flask equipped with a screw cap were added the supernatant. The flask was transferred to a Coy anaerobic chamber, where 200 mL degassed KPi buffer was added to dilute the cell-free lysate suspension. At this point, an aqueous solution of ferrous ammonium sulfate (5 mL, 200 mM stock solution in degassed H₂O) and a solution of sodium ascorbate (10.0 mL, 200 mM stock solution in degassed KPi buffer) were added. The resulting mixture was carefully mixed by gentle shaking. 3 min later, the azanyl ester substrate (10.0 mmol, 10.0 mL stock solution in MeCN (1.0 M)) was added. The flask was capped, sealed with parafilm, taken out of the Coy anaerobic chamber, and allowed to shake in an Eppendorf Innova 44R shaker at room temperature and 230 rpm for 6 h.

Upon the completion of this biotransformation, 100 mL 1M HCl aqueous solution was added to the reaction mixture. The mixture was extracted with 300 mL EtOAc by vigorous mixing. The resulting mixture was transferred to centrifugation buckets and centrifuged (4,347 g, 20 min) using an Eppendorf tabletop centrifuge 5910R to separate the organic and the aqueous layers. The aqueous layer was extracted with EtOAc for an additional five times. Combined organic layers were dried over MgSO₄, and an aliquot of the organic layer (400 μL) was used for product enantiomeric ratio determination via chiral HPLC analysis. Combined organic layers were concentrated in vacuo with the aid of a rotary evaporator and purified by column chromatography with the aid of a Biotage Isolera.

Fig. 3. Substrate scope of ACCO_Nim1-catalyzed enantioselective synthesis of α-trisubstituted non-canonical amino acids.

Reaction conditions: 10.0 mM azanyl ester 1, ACCO_Nim1 cell-free lysate in 100 mM KPi buffer (pH = 7.4, ca. 0.049–0.098 mol%), 1.0 mM Fe(NH₄)₂(SO₄)₂, 2.0 mM sodium ascorbate, MeCN (5% v/v), in Coy anaerobic chamber, room temperature, 6 h. TTN was determined by the SDS-PAGE analysis using cell-free lysate. All the biocatalytic reactions were performed as technical triplicates. See the Supplementary Information for details.

Fig. 6. Mechanistic studies.

a, Olefin isomerization in nonheme Fe enzyme-catalyzed nitrogen migration reaction. b, radical clock experiments. c, Non-competitive intermolecular kinetic isotope effect (KIE) studies. d, Intramolecular KIE studies with enantioenriched (R)-1a-d₁ and (S)-1a-d₁ using ACCO_Nim1. e, Intramolecular KIE studies with enantioenriched (R)-1a-d₁ and (S)-1a-d₁ using ACCO_Nim2. f, Dissecting the kinetic isotope effects (KIE) and enzymatic enantioinduction effects by quantitative free energy analysis. ΔG_KIE: kinetic isotope effect (KIE) in the HAT step in kcal/mol; ΔG_{enantioselectivity}: enantioselectivity in the HAT step in kcal/mol.

Data availability

All data are available in the main text and the Supplementary Information. Plasmids encoding evolved ACCO nitrogen migratases reported in this study are available for research purposes from Y.Y. under a material transfer agreement with the University of California Santa Barbara.