Stereoselective Construction of β-, γ-, and δ-Lactam Rings via Enzymatic C–H Amidation

Lactam rings are found in many biologically active natural products and pharmaceuticals, including important classes of antibiotics. Given their widespread presence in bioactive molecules, methods for the asymmetric synthesis of these molecules, in particular through the selective functionalization of ubiquitous yet unreactive aliphatic C–H bonds, are highly desirable. In this study, we report the development of a novel strategy for the asymmetric synthesis of 4-, 5-, and 6-membered lactams via an unprecedented hemoprotein-catalyzed intramolecular C-H amidation reaction with readily available dioxazolone reagents. Engineered myoglobin variants serve as excellent biocatalysts for this transformation producing an array of β-, γ-, and δ-lactam molecules in high yields, with high enantioselectivity, and on preparative scale. Mechanistic and computational studies elucidate the nature of the C–H amination and enantiodetermining steps in these reactions and provide insights into protein-mediated control of regioselectivity and stereoselectivity. Using this system, it was possible to accomplish the chemoenzymatic total synthesis of an alkaloid natural product and a drug molecule in much fewer steps (7–8 vs. 11–12) than previously possible, which showcases the power of this biosynthetic strategy toward enabling the preparation of complex bioactive molecules.


Introduction
The selective functionalization of ubiquitous yet unreactive carbon-hydrogen (C-H) bonds via both chemical and enzymatic methods constitutes a powerful strategy for the diversification of organic molecules and enabling the devise of new disconnections and routes for the construction of complex molecules and natural products. [1][2][3][4][5][6][7] Due to the large prevalence of amine-based functionalities in bioactive molecules and pharmaceuticals, a highly desirable and sought-after transformation in organic and medicinal chemistry is the selective amination of aliphatic C-H bonds. [8][9][10] Notable advances in this field have led to the development of organometallic catalysts for catalyzing the insertion of nitrene species into C−H bonds, resulting in the formation of new carbon-nitrogen bonds (Figure 1b). [8][9][10] These transformations are mediated by reactive metalnitrenoid species generated upon reaction of the transition metal catalyst with nitrene precursor reagents such as iminoiodinane, azides, and hydrolamine derivatives. [8][9][10] Using this strategy, a variety of cyclic amines, including oxazolidinones, sulfamates, sultams, and pyrrolidines have been accessible. Despite this progress, extension of this C-H amination strategy to the synthesis of cyclic amides (lactams), which are key structural motifs in many pharmaceuticals, agrochemicals, and other fine chemicals (Figure 1a), 11,12 has represented a major challenge. 13 The difficulty of this transformation can be attributed to the instability of the acyl nitrene intermediate that undergoes facile decomposition to isocyanates through a Curtius-type rearrangement, thereby outcompeting the desired C-H nitrene insertion process. 13 Recently, Chang and coworkers has reported a breakthrough in this area through the development of an iridium-based system for enabling this transformation. 13 This progress notwithstanding, asymmetric versions of this methodology are restricted to 5-membered rings (γ-lactams) and require the use of rare and toxic metals (i.e., Ir, Ru). 14,15 Inspired by the chemistry of metalloporphyrins, 16 our group and the Arnold group have recently demonstrated the ability of engineered hemoproteins to serve as biocatalysts for intramolecular [17][18][19][20][21][22] and intermolecular [23][24][25] C -H am inat ions via nit rene tr ansfe r ( Figure 1c).
Specifically, engineered cytochrome P450 enzymes have been shown catalyze the cyclization of sulfonyl azides, carbonazidates, and sulfonazidates 26 to produce sultams, cyclic carbamates, and cyclic sulfamides, respectively. [17][18][19][20][21][22] Furthermore, iridium-substituted P450s 27 and non-heme Fe-dependent enzymes 28,29 were also found to catalyze similar intramolecular C-H amination reactions. Despite this progress, the synthesis of lactam rings using biocatalytic nitrene transfer approaches has remained elusive, largely due to the aforementioned difficulty in controlling the reactivity of acyl nitrene intermediates while disfavoring other competing, unproductive reactions (e.g., nitrene reduction) known to affect these abiological enzyme-catalyzed reactions. 21 Here, we report the development of a general biocatalytic methodology for the asymmetric synthesis of enantioenriched lactams via intramolecular C-H amidation of dioxazolones ( Figure   1d), a safe and readily accessible class of nitrene donor reagents. 30 This strategy provides an efficient and scalable approach to the selective construction of γ-lactam molecules with high enantioselectivity. Furthermore, the scope of this biocatalytic system could be extended to the construction of optically active β-and δ-lactam motifs with high enantiocontrol. The power of this methodology is further showcased through the implementation of concise chemoenzymatic routes for the total synthesis of an alkaloid natural product and a drug molecule. Comprehensive mechanistic studies elucidate the nature of the C-H amination step as well as the role of the protein scaffold in controlling the regio-and enantioselectivity of the reaction. Leveraging a direct C-H amination strategy, this methodology is mechanistically distinct yet complementary to recently reported biocatalytic approaches for γ-lactam synthesis that rely on radical-mediated cyclizations of alkene-containing substrates [31][32][33] or carbene C-H insertion with artificial enzymes. 34

Results and Discussion
Biocatalyst discovery. Inspired by prior work of Chang and coworkers, 13,35 we envisioned the possibility to execute an enzyme-catalyzed γ-C-H amidation reaction via nitrene transfer with 6 dioxazolone reagent 1a (Figure 1a-b). Particularly attractive features of dioxazolones as nitrene precursors include their facile synthesis from commodity carboxylic acids and their stability and safety compared to azide-based reagents previously used in biocatalytic nitrene transfer reactions. [17][18][19][20][21][22] To identify an initial biocatalyst for this reaction, we tested various heme-containing enzymes and proteins, including wild-type myoglobin (Mb), cytochromes P450s, peroxidases, and cytochromes c, under anaerobic conditions (Table S1). While many of these reactions produced the acyclic amide 2b as byproduct, none of these biocatalysts displayed any activity toward formation of the desired γ-lactam product ( Table S1). Previous studies from our group demonstrated that mutations of the distal His64 residue in Mb (Figure 1c) can enhance its activity toward non-native carbene and nitrene transfer reactions. 36,37 Promisingly, Mb(H64V) was found to react with 1a to yield minute yet detectable amounts of the desired lactam 2a (2% GC yield) with good levels of enantioselectivity (96% ee; Figure 1d). Reaction conditions: 20 µM protein, 10 mM 1a, 10 mM Na2S2O4 in potassium phosphate buffer (50 mM, pH 7), 3 hours, room temperature, under anaerobic conditions. Yields and product distribution as determined by GC using calibration curves with isolated product. * With acetonitrile as co-solvent. ** Using sodium borate buffer (pH 9) with 5% (v/v) acetonitrile (= standard reaction conditions or s.r.c).
Encouraged by these results, we extended the screening to a broader panel of engineered Mb variants (Table S2) containing mutations at the level of the distal His64 residue (Ala, Val, Gly) along with additional mutations within the active site of the hemoprotein (Figure 2c). From this screening, multiple Mb variants were found to exhibit increase C-H amidation activity compared to Mb(H64V) (10-50% yield; Table S2). Among them, Mb(H64V,V68A) (called Mb*), which was previously developed for stereoselective cyclopropanation, 36 emerged as the best biocatalyst for this reaction, producing 2a in 50% yield with excellent enantioselectivity (>99% ee) (Figure 2d). The configuration of the γ-lactam product was determined to be S b y crystallography (Figures 2a and S11). In addition to the C-H amidation product, the Mb*catalyzed reaction also produced a significant amount (40%) of the amide byproduct 2b (Figure   2a), which likely arises from reduction and protonation of the nitrene intermediate (Scheme S1) as observed previously for C-H amination reactions with azide-based precursors. 18,22 Unexpectedly, a minor product (10%) corresponding to the γ-lactone 2c was also formed in the reaction (Figure 2a).
Method optimization. Compared to Mb*, Mb variants containing Ala and/or Gly mutations at the 64 or 68 positions showed significant decreases in activity and/or selectivity (Entries 3-6 vs. 7; Table S2) and a similar effect was observed upon introduction of additional mutations to Mb*, suggesting that the reaction is sensitive to subtle changes in the shape of the enzyme's active site.
Based on prior mechanistic studies on P450-mediated C-H amination 21 and given the robustness of Mb biocatalysts to organic solvents, 38 we hypothesize that the addition of an organic co-solvent could favor the desired C-H amidation reaction by disfavoring formation of the amide byproduct 2b. Accordingly, screening of various organic co-solvents showed that acetonitrile (ACN) is beneficial toward increasing the yield of the C-H amidation product (50→65% yield), while reducing the undesired reduction reaction (30%) and without affecting enantioselectivity (Figures   2d and S1). Further optimization of the reaction conditions revealed that slightly alkaline conditions (pH 9) further increase the enzyme's C-H amidation activity (75% yield of 2a) at the expenses of byproduct 2b (10%), while retaining excellent enantioselectivity (Figure 2d). These trends are consistent with our hypothesis that formation of the amide byproduct 2b involves protonation of the nitrene intermediate, which should be disfavored under more alkaline conditions and in the presence of organic solvent. Under catalyst limiting conditions (0.07 mol%), Mb* was determined to catalyze the C-H amidation of 1a in 58% yield and 2,175 TON with excellent enantioselectivity. In addition to dioxazolone 1a, we also evaluated other nitrene precursors such as dioxazole 1aa, acyl-protected hydroxylamine 1ab, and dioxathiazole 1ac (Figure 2a). Whereas 1ac and 1ab were inactive and less effective than 1a, respectively, for the cyclization reaction Substrate scope. To assess the substrate scope of this methodology, Mb* was tested against an array of substituted dioxazolone substrates (1d-p). As summarized in Figure 3a, these experiments revealed a large tolerance of the enzyme toward substitutions at the para position of the aryl ring affording the desired γ-lactam products (2d-i) with high to excellent enantioselectivity (73-99% ee). While both electron-withdrawing and donating substitutions are accepted by the enzyme, increased activity was observed for substrates containing electron-donating substituents (1h-i).
Substitutions at the ortho and meta positions were also well tolerated by the enzyme, producing the corresponding γ-lactam products (2j-l) in good yields (68-88%) and high enantiopurity (99% ee, Figure 3a). In addition to substituted phenyl groups, the enzyme is able to catalyze the cyclization of substrates containing heteroaryl groups such as thiophenyl group (2m-n) with good to high levels of activity (75-90% yields) and enantioselectivity (66-99% ee). Allylic C-H amidation was also possible as demonstrated by the successful preparation of 2o in 99% ee. Lastly, enzymatic synthesis of 2p in 28% yield and 83:17 diastereomeric ratio showed the tolerance of the enzyme also to substitutions in α to the dioxazolone core.
Enantiodivergent biocatalyst. Enantiodivergent biocatalysts are highly desirable yet often hard to develop. 39 Notably, screening of the initial Mb active-site mutant library revealed a variant, Mb(L29T,H64V,V68L), that catalyzes the cyclization of 1a with inverted enantioselectivity compared to Mb*, producing the R-configured γ-lactam product ent-2a in 65% ee, albeit in modest yield (15% ; Table S2). To improve the performance of this biocatalyst, Mb(L29T,H64V,V68L) was subjected to active-site mutagenesis, ultimately leading to Mb(L29T,H64T,V68L), which produces ent-2a with both an improved enantioselectivity of 91% ee and two-fold higher activity compared to the parent enzyme. To explore the substrate promiscuity of this enantiocomplementary biocatalyst, Mb(L29T,H64T,V68L) was tested a g a i n s t t h e p a n e l o f dioxazolones 1d-p. Albeit in more moderate yields compared to the Mb* reactions, the majority of these substrates (10/14) could be converted to the R-configured γ-lactam products with good to high enantioselectivity (57-99% ee; Figure 3a). Of note, ent-2k, ent-2l, and the thiophenyl containing substrate ent-2m were all obtained in 99% enantiomeric excess. Altogether, these results highlighted the broad substrate scope and predictable enantiocomplementarity of the two Mb-based biocatalysts for γ-lactam ring formation. Synthesis of β-lactams and δ-lactams. Next, we targeted the synthesis of β-lactams which are highly desirable building blocks for medicinal chemistry as well as key pharmacophores in βlactam antibiotics. 12 Notably, β-lactam formation via intramolecular nitrene transfer has not been reported to date. Upon challenging Mb* with substrate 3a, the desired β-lactam 4a was obtained in high yield (85%) and excellent enantioselectivity (99% ee) (Figure 3b). Mirroring the Senantioselectivity in γ-lactam formation, the enzyme maintains S-enantiopreference for the formation of 4a, as determined by X-ray crystallography (Figure 3c). These findings prompted us to further explore the substrate scope of this reaction (3b-j; Figure 3b). Remarkably, variously substituted substrates could be converted into the desired β-lactam products with excellent enantioselectivity (99% ee; Figure 3b) and up to 93% yield. Unlike the γ-lactams, substrates bearing electron-withdrawing groups on the aryl ring were cyclized more efficiently than those containing electron-donating groups (e.g., 32% yield for 4d vs. 75% for 4c) and para substitutions were better tolerated than meta substitutions (e.g., 75% yield for 4g vs 32% for 4d). These differences likely arise from the differential role of electronic and steric constraints in the 4-vs. 5membered ring formation. Furanyl-and thiophenyl-containing substrates 3i and 3j, respectively, were also cyclized very efficiently (75% and 93% yields, respectively) and with high enantiocontrol (99% ee).
To explore the reactivity of the Mb biocatalyst toward synthesis of δ-lactams, a reaction was carried out using substrate 5a, which resulted in a mixture of δ-lactam (6aa) and γ-lactam (6ab) in a 1:3.7 ratio, in addition to the amide 6b as the major product (7:26:67 ratio for 6aa:6ab:6b; Figure 3c). These results revealed the enzyme's preference for amidation of the homobenzylic γ-C-H bond (to give 6ab) over the benzylic δ-C-H bond (to give 6aa) despite the higher bond dissociation energy (BDE) of the latter (~95 vs. 90 kcal/mol) (Figure 3c). These findings inspired us to substitute the γ-C-H bond with an O atom to favor δ-lactam formation.
Gratifyingly, the Mb* reaction with 5c produced the desired δ-lactam 6c with significantly improved efficiency (78% yield) as well as excellent enantioselectivity (99% ee; Figure 3c). This biocatalytic reaction was also found to be tolerant toward substitution on the aryl ring, as demonstrated by the synthesis of δ-lactams 5d-f in 45-93% yields and high enantiomeric excess (90-99% ee). Taken together, these results reveal a remarkable generality of the Mb* catalyst toward enabling the stereoselective synthesis of lactams of varying sizes and with different substitutions. Noteworthy is also the consistent and predictable S stereoselectivity of the Mb*catalyzed C-H amidation reaction not only across the different substrates but also across the β-, γ-, and δ-lactam rings, which adds to the synthetic utility of this biocatalytic system. Mechanistic Studies. Studies were then performed to gain insights into the mechanism of this enzyme-catalyzed reaction. To probe the nature of the C-H amidation step, the Mb*-catalyzed reaction was carried out in the presence of the Z-configured dioxazolone 1q, which resulted in the formation of the lactam product 2q in the E configuration (Figure 4a). This result rules out a concerted C-H nitrene insertion process and is consistent with a stepwise hydrogen atom abstraction (HAA)/radical rebound mechanism proceeding via an allylic radical that undergoes Z→E isomerization to yield trans-2q prior to radical recombination. Of note, complete isomerization of the double bond in the cyclization product (no cis-2q was observed) shows that the radical intermediate is relatively long-lived.
To further investigate the kinetic role of the C-H cleavage step, non-competitive intermolecular H/D competition experiments were carried out using substrate 1a and 1a-d 1 in parallel reactions (Figure 4b). These experiments yielded a kinetic isotope effect (KIE) value of 2.6 ± 0.2 (Figure 4b and S3), which is lower than that determined for C−H amination reactions with azide-based substrates catalyzed by engineered P450s (kH/kD: 3.4-5.3) 20,40 , but higher than that determined for P450-catalyzed cyclization of sulfonylazides (kH/kD: 0.9), where the azide activation was established to be rate-determining. 21 Overall, these results indicated that the C-H cleavage step in the present system is only partially rate determining, with other steps contributing to control the overall rate of the reaction.  (Figure 4c and S4). In addition, both reactions proceed with high enantioselectivity (99% ee). From these results, it can be derived that (a) H abstraction is strongly favored over D abstraction regardless of the configuration of the C-H amination site, and (b) protein-mediated enantioinduction in the C-N bond forming process must occur at the level of the radical rebound step (Figure 4d). Indeed, formation of (S)-2a-d 1 as the major product from either (S)-1a-d 1 or (R)-1a-d 1 , along with the preserved high % ee in both cases, imply that, after HAA, the pro-S and pro-  (Figure 4d). However, the mere 2.5-fold difference between the kH/D values (in constrast to the 50-to 220-fold difference determined with other systems 40,41 ) suggests that the HAA step is barely stereoselective and thus that asymmetric induction is largely controlled by the enzyme during the radical recombination step. As such, this system shows a distinct enantioinduction mechanism compared to that described for asymmetric C-H aminations of sulfonyl azide substrates catalyzed by Co-porphyrins 41   that the OSS is not a pure singlet spin state, but 50:50 singlet:triplet, often observed in calculations on diradicals. Based on the energy profile, the C-N bond forming step is predicted to be rate determining, exhibiting a 4.5 kcal/mol higher energy barrier than the C-H bond cleavage event.
These findings are in excellent agreement with the results from the mechanistic experiments ( Figure 4) and explain the long-lived nature of the radical intermediate, as suggested by the complete double bond isomerization observed with trans-2q (Figure 4a).
This mechanistic model provides a framework also for defining a plausible mechanism for the formation of the unexpected γ-lactone product 3c from 1 (Figure 2a), a reaction that finds no precedents in catalytic nitrene transfer reactions. After the HAA step, single electron transfer from the C-centered radical to the heme (or protein matrix) can generate a benzylic carbocation which can then react with the amide group (via the carbonyl group) to form a dihydrofuranimine ring .
Hydrolysis of the latter produces 3c (Scheme S1). In addition to this radical-polar crossover mechanism, a radical pathway can be envisioned that proceeds via the same dihydrofuranimine intermediate produced via reaction of the benzylic radical with the amidyl group (Scheme S1). In either case, a slow C-N bond forming radical rebound step as revealed by our mechanistic and DFT studies is expected to enable this competing side reaction to occur under suboptimal reaction conditions.

Enzyme-controlled regio-and enantioselectivity.
To investigate the role of the enzyme in controlling the enantioselectivity of the reaction, we explored the heme-bound iron-nitrenoid intermediate IM1 docked in the active site of Mb* using the available crystal structure of this protein. 42 As shown in Figure 5a, the N atom of IM1 can abstract either the H 1 or H 2 , leading to the S and R lactam product, respectively, under fast rebound conditions. We measured the evolution of distances between the N atom and the H 1 and H 2 atoms during 1000 ns MD simulations (Figure 5b). These studies show that the average distance from N atom to H 1 and H 2 are 3.96-4.01 Å and 4.06-4.10 Å, respectively, suggesting little to no preference for abstraction of H 1 vs. H 2 (leading to the S vs. R product) by the nitrene intermediate. and Ala68 (Figure 5b-c). While this arrangement can accommodate substitution at different positions of the aryl ring, it also shows potential steric constraints as the size of the para substituent increases, providing a plausible rationale for the structure-activity trends observed experimentally (i.e., -F (2d) > -Cl/Br (2e-f) > -I (2g); Figure 4a).
To understand the regioselectivity of the enzyme, we studied the nitrene intermediate IM1 generated from substrate 5a (Figure 5d). In this case, the nitrene intermediate can abstract either the γ-or δ-H atom, leading to the 5-and 6-membered lactam product, respectively. DFT calculations show that the energy barrier for hydrogen abstraction leading to the δ-lactam is 2.4 kcal/mol higher than for the formation of γ-lactam (Figure 5d). We also studied the proximity of the γ and δ H atoms to the nitrene N atom via 1000 ns MD simulation. The γ-H…N average distance is 2.8-3.9 Å, whereas for δ-H…N distance is significantly larger, namely 4.3-5.2 Å (Figures 5e and S10). Thus, both the lower energy barrier for H γ abstraction and a closer nitrene N…H γ atom distance contribute to favor formation of the γ-lactam product, which can explain the regioselectivity of the Mb*-catalyzed C-H amidation of 5a observed experimentally (Figure 4c).
Chemoenzymatic total synthesis of bioactive molecules. The present strategy was then applied for the chemoenzymatic syntheses of bioactive alkaloid (S,S)-(-)-Homaline (7) and FDA-approved drug (S)-Dapoxetine (8) (Figure 6). Specifically, we envisioned that the key β-lactam intermediate 4a, previously accessible only in low yields and after lengthy routes (7-8 steps, 7% overall yield 43,44 ; Figure 6b), could be produced in a more efficient and step-economical manner by enzymatic means using the present method. Accordingly, enantiopure β-lactam 4a (>99% ee) was produced from 3a and isolated on a preparative scale (0.5 g) from a scaled up reaction with Mb* (Figure 6a). From 4a, (S,S)-(-)-Homaline (7) and (S)-Dapoxetine (8) could be synthesized in only four and five steps, respectively, using known routes (Figure 6b; see Scheme S2 for further details). Overall, asymmetric Mb-catalyzed C-H amidation enabled the chemoenzymatic synthesis of the alkaloid natural product and drug molecule in a total of 7 and 8 steps, respectively, compared to 11 and 12 steps required in previous total synthesis strategies. In addition, with the present approach, the key enantiopure β-lactam intermediate 4a was readily obtained from the achiral, commodity chemical hydrocinnamic acid as opposed to more expensive optically active precursors required in the previously reported routes (Figure 6b). These results further showcase the synthetic utility and scalability of the present methodology for the synthesis of biologically active molecules. Conclusions. In summary, we have developed a first biocatalytic strategy for the asymmetric construction of lactam molecules via nitrene transfer. Starting from readily accessible dioxazolones, this strategy could be leveraged to afford a broad range of β−, γ-, and δ-lactam scaffolds in good yields and high enantiomeric excess using a single Mb-based biocatalyst. In addition, we demonstrated the possibility to obtain enantiopodes of the γ-lactam products using an alternate engineered Mb variant with enantiodivergent selectivity. Our mechanistic investigations revealed that these reactions proceed via a HAA/radical rebound pathway, with the enzyme binding site controlling the stereo-and regioselectivity (β-, γ-, δ-C-H amidation) of the process.
Furthermore, while the HAA step is generally assumed to be enantiodetermining in enzymatic C-H aminations 45,46 , our studies show that protein-induced enantioselectivity in the present system is largely controlled at the level of the radical rebound step. The power of the present methodology was further showcased by the concise chemoenzymatic total synthesis of an alkaloid natural product and a drug molecule in about half of the steps required previously, while offering higher overall yields and starting from a commodity chemical instead of optically active precursors. This work expands the available biocatalytic toolbox for the asymmetric synthesis of amine-containing molecules and paves the way to the development of other asymmetric enzyme-catalyzed nitrene transfer reactions involving dioxazolones as nitrene precursors.

SUPPORTING INFORMATION
Experimental procedures including synthetic methods, compound characterization data, computational methods, atom coordinates of DFT models, additional figures and tables, Nature Research reporting summaries, details of author contributions and competing interests; and statements of data and code availability are available online.

DATA AVAILABILITY STATEMENT
Data that support the findings of this study are included in this published article (and its supplementary information files). Additional datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request. Crystallographic data for small molecules have been deposited in the Cambridge Crystallographic Data Centre (CCDC) as described in the supplementary information files.

CODE AVAILABILITY
No custom computer codes or mathematical algorithms have been used for this research.