Regiodivergent Radical Termination for Intermolecular Biocatalytic C–C Bond Formation

Abstract

Radical hydrofunctionalizations of electronically unbiased dienes are challenging to render regioselective, because the products are nearly identical in energy. Here, we report two engineered FMN-dependent “ene”-reductases (EREDs) that catalyze regiodivergent hydroalkylations of cyclic and linear dienes. While previous studies focused exclusively on the stereoselectivity of alkene hydroalkylation, this work highlights that EREDs can control the regioselectivity of hydrogen atom transfer, providing a method for selectively preparing constitutional isomers that would be challenging to prepare using traditional synthetic methods. Engineering the ERED from Gluconabacter sp. (GluER) furnished a variant that favors the γ,δ-unsaturated ketone, while an engineered variant from a commercial ERED panel favors the δ,ε-unsaturated ketone. The effect of beneficial mutations has been investigated using substrate docking studies and the mechanism probed by isotope labeling experiments. A variety of α-bromo ketones can be coupled with cyclic and linear dienes. These interesting building blocks can also be further modified to generate difficult-to-access heterocyclic compounds.

Introduction

The reductive hydroalkylation of alkenes using radicals is an attractive strategy for building C–C bonds because these reactions are often highly regioselective and are tolerant to various functional groups.^1,2 While most examples of this type of reactivity utilize monoalkenes, 1,3-dienes are compatible and present a unique selectivity challenge because the resulting allylic radical can be terminated at two possible positions, forming a mixture of constitutional isomers. This limitation is usually overcome by using substrates that provide a thermodynamic preference for one constitutional isomer. Alternatively, in the context of hydroalkylations, transition metal-catalyzed strategies have been developed where termination of a π-allyl organometallic intermediate determines which olefin isomer is formed.³ For example, the copper-catalyzed functionalization of 1,3-dienes proceeds through a π-allyl species in which the 1,2-/1,4-addition intermediates are in equilibrium. This equilibrium is influenced by the choice of ligand and nature of the electrophile; however, the exact reaction pathway is often unclear.⁴

Nature has evolved enzymes to catalyze regio- and chemoselective radical transformations in the synthesis of structurally complex natural products.^5,6 Some of the most intensely studied metalloenzymes, including cytochrome P450 and ribonucleotide reductase, as well as radical S-adenosylmethionine (SAM) enzymes, employ radical intermediates.^7,8 However, many of these transformations control the regio- and stereoselectivity of hydrogen atom abstraction from an unactivated C–H bond. In contrast, there are few examples where hydrogen atom transfer (HAT) selectively terminates a substrate radical in an enzyme active site. Parallels can be drawn with terpenoid cyclases in which high-energy carbocation intermediates are terminated by a water molecule.⁹ Access of the solvent water molecule must be strictly controlled so as not to prematurely quench carbocation intermediates in the cyclization cascade.¹⁰

The Hyster group previously demonstrated that flavin-dependent oxidoreductases, including “ene”-reductases (EREDs) and Baeyer–Villiger monooxygenases (BVMOs), can catalyze non-natural radical reactions within their active sites to produce cyclic lactams, γ-stereogenic amides and ketones, and alkylated arenes.¹¹⁻¹⁵ In these systems, electrophilic α-acyl radicals are coupled to monoalkenes, typically styrenes, to form a new C–C bond. This step occurs with high regioselectivity because of the strong thermodynamic preference for the formation of a benzylic radical. HAT from the flavin semiquinone (FMN_sq) terminates the substrate-centered radical and sets the stereocenter in the product. While this step frequently occurs with high enantioselectivity, it has not been used to control the regioselectivity of HAT. This type of selectivity represents a significant challenge because it involves precisely positioning the diene over the N5 position of the flavin cofactor to favor one constitutional isomer. We sought to develop enzymes to couple alkyl halides with 1,3-dienes to provide products with a high level of selectivity for a single constitutional isomer of the product (Figure 1). If successful, this method would enable the rapid synthesis of saturated bicyclic heterocycles, a common structure in pharmaceutically important motifs.¹⁶

Figure 1.

(A) Photoredox-catalyzed radical addition to 1,3-dienes. (B)Protein engineering for regioselective radical termination.

Results and Discussion

As a model system, we targeted the coupling of α-bromo-acetophenone with 1,3-cyclohexadiene. When this coupling is run using typical photoredox catalysis, a 54:46 mixture of the 1,4- and 1,2-addition products is formed, presumably because the energy minima of the constitutional isomers are similar (Figure 1A; see Supporting Information Figure 17 for photocatalyzed reaction). A screen of readily available ERED homologues was tested for the model coupling while irradiated with 497 nm cyan light (Scheme 1). The most active ERED was the Gluconabacter oxidans (GluER) with a single mutation to a surface residue (GluER-T36A), which provided the product in 62% yield as a 55:45 mixture of γ,δ- and δ,ε-unsaturated ketones (Scheme 1A). The nicotinamide-dependent cyclohexanone reductase (NCR) from Zymomonas mobiliz afforded a slightly improved regioselectivity (60:40) for the 1,4-addition product 3A, however with a significantly reduced yield (8% yield; Supporting Information Table 7). In addition to screening known EREDs, we tested a collection of 52 genetically diverse “ene”-reductases from Prozomix. One “ene”-reductase, from Escherichia coli named EschER, gave the opposite regioselectivity, 29:71 in favor of the 1,2-addition product 3B, with a yield of 52% (Scheme 1A). We proceeded to optimize the reaction by modifying the pH, diene equivalents, and temperature. We found that pH 8 was optimal for GluER-T36A, with lower pH values of 7 and 6 furnishing product with decreased regioselectivity (Supporting Information Figure 9). When testing different irradiation conditions, we found that an improved yield, with less formation of the hydrodehalogenated product, was achieved when reactions were run in the dark. The decrease in the level of acetophenone formation is likely due to the mechanism of radical initiation. In the absence of an alkene, the HUMO/LUMO gap is too large for ground-state electron transfer. In contrast, when the alkene is part of the complex, a hyperconjugative interaction between the alkene and alkyl halide decreases LUMO energy of the electron acceptor in the complex, decreasing the HOMO/LUMO gap and enabling ground-state electron transfer (Scheme 1B).¹⁷

Scheme 1. (A) Screening Light Source and pH of ERED Hits (*Remaining Mass is 2-Hydroxy Acetophenone); (B) σ–π Hyperconjugation with CT-Complex Formation (LUMO_A=Acceptor; HOMO_D=Donor).

With optimal conditions in hand, we conducted a protein engineering campaign focused on increasing the regioselectivity of the HAT event with GluER-T36A. We selected 14 sites within 6 Å of FMN for iterative site saturation mutagenesis (ISM) (Figure 2; Supporting Information Table 1). We found that mutation of asparagine at position 175 to glycine (GluER-T36A-N175G) resulted in an improvement in regioselectivity for product 3A to 84:16; however, the yield decreased to 36%, compared to 74% for the GluER-T36A control (Supporting Information Table 2). This residue is one of the canonical binding residues for NADPH, potentially accounting for the decreased yield. However, mutation of alanine at position 56 to glycine (GluER-T36A-A56G) provided comparable yields with increased regioselectivity (86:14). A second round of site saturation mutagenesis, using GluER-T36A-A56G as the template, revealed that mutation of the canonical binding residue at position 175 to methionine (GluER-T36A-A56G-N175M) resulted in the (GluER-B7) variant with a regioselectivity of >95:5 and 37% yield for 3A. Optimization of the reaction conditions with GluER-B7 revealed pH 7 to provide improved yield (76% yield, 96:4; Figure 2 and Supporting Information Figure 9). To increase the selectivity associated with EschER, the ERED that favors formation of the δ,ε-unsaturated ketone 3B, we conducted a second protein engineering campaign.

Figure 2.

Initial screening results and engineering. (A) Initial results for GluER-T36A and EschER catalyzed the hydroalkylation of 1,3-cyclohexadiene. (B) Results from the evolution campaign for GluER-T36A (PDB: 6MYW, gold) and EschER (AlphaFold Supporting Information Figure 8, blue).

Initially, 12 positions analogous to those explored for GluER-T36A were subjected to site saturation mutagenesis. This resulted in variant EschER-A277I with an improved regioselectivity of 20:80, and a yield of 72%. Further site saturation mutagenesis of EschER-A277I resulted in variant EschER-A277I-D275Q (EschER-D1) producing 3B in 72% yield and a regioselectivity of 17:83 (Figure 2B). A model of EschER overlapped well with the crystal structure of GluER-T36A (RMSD = 1.08 Å). A significant difference between GluER-T36A and EschER is in the loop region, aa264–aa271 and aa275–aa282, respectively (Supporting Information Figure 8). Interestingly, the two mutations appear on this loop, suggesting that this region is important for the switch in selectivity. During the two engineering campaigns, it was also noted that selectivity at the β-chiral carbonyl center was largely lost (Supporting Information Tables 4 and 6).

We conducted a series of deuterium incorporation experiments to better understand the product-determining step for both GluER-B7 and EschER-D1 (Figure 3A). For the 1,4-hydroalkylation catalyzed by GluER-B7, using d₁-glucose and GDH for in situ labeling of the flavin cofactor, the γ,δ-unsaturated ketone 3A was obtained in 52% yield, 96:4 (A:B) with 87% deuterium incorporation at the C4 position. In the case of EschER-D1, the δ,ε-unsaturated ketone product 3B was obtained in 30% yield and 15:85 (A:B) with 60% deuterium incorporation at the C2 position. These results indicate that radical termination primarily occurs via HAT from FMN_hq. The decreased deuterium incorporation with EschER-D1 suggests that there is a competing hydrogen atom source, presumably the OH bond of an active site tyrosine (Supporting Information Figure 15; D₂O experiment).¹⁸

Figure 3.

(A) Deuterium labeling experiments. ^aIsolated yield based on a 0.1 mmol scale reaction. (B) Overlayed induced fit models: GluER-T36A (blue) with 3A (green), C4–N5 distance of 4.7 Å, and GluER-B7 (gold) with 3A (orange), C4–N5 distance of 3.7 Å.

In combination with the knowledge gained from the mutagenesis studies and deuterium labeling experiments, induced fit modeling of GluER-T36A (blue) and GluER-B7 (gold) with 3A was carried out (Figure 3B; see Supporting Information). The asparagine N175 is involved in hydrogen bonding with the carbonyl of the α-bromo ketone substrate, and removal of this H-bonding interaction allows for a substrate conformation in which the 1,4-addition product is favored.^19,20 The proposed increased flexibility associated with the A56G mutation is supported by a slight rotation of residue W66. The site saturation mutagenesis (SSM) studies had already demonstrated that mutation of residue W66 could switch the regioselectivity in favor of product 3B (Supporting Information Table 2). The docked GluER-T36A with 3A (green) gave a C4–N5 distance of 4.7 Å, whereas GluER-B7 docked with 3A (orange) gave a C4–N5 distance of 3.7 Å. The modeling suggests that the combination of the A56G and N175 M mutations helps to place 3A over the FMN and force the C4 carbon closer to the N5 of FMN, favoring HAT at this position.

With a better mechanistic understanding and the engineered GluER and EschER variants in hand, we then explored the substrate scope (Figure 4). The improved GluER-B7 variant accommodated various ortho- and para-substituents on the aromatic ring, while maintaining a high yield. Both electron-donating and electron-withdrawing groups were well tolerated. Notably, ortho- and para-halogenated substrates were also well accepted, providing a handle for further functionalization to build molecular complexity. In cases where GluER-T36A favored the formation of the 1,2-addition products (23–26B, 28B, and 32B; Supporting Information Table 8), the GluER-B7 variant retained excellent selectivity, with a minimum selectivity range of 90:10 to 96:4 for the 1,4-addition products. It was also interesting to see that the N-containing heterocycle 33A could be formed in moderate yield but with excellent selectivity. In all cases, the GluER-B7 variant gave a much-improved regioselectivity for the 1,4-addition product. The EschER-D1 variant demonstrated a contrasting regioselectivity to afford products 30B–33B. Excellent selectivity was achieved for 1,2-addition products 30B (8:92) and 31B (3:97). The pyridyl derivative was also well accepted with a yield of 34%, favoring the 1,2-addition product 33B (19:81).

Figure 4.

Substrate scope with GluER-B7 and EschER-D1 variants. Standard reaction conditions: The reactions (500 μL) consisting of α-bromo ketone (1 eq, 2.5 μmol), diene (6 eq, 15 μmol), 2 mol % ERED, 2 mol % NADP⁺, 0.75 mg GDH, and glucose (6 eq, 15 μmol) in buffer (100 mM KP_i, pH 7) with 5% DMSO cosolvent were shaken in the dark for 24 h at room temperature. Yield and regioisomeric ratio were determined by UPLC using either 1,3,5-tribromobenzene (TBB) or 1,3,5-trimethoxybenzene (TMB) as an internal standard. Reactions were carried out in duplicate, and the average yield and regioisomeric ratio were reported. A:B ratio corresponds to the amount of γ,δ- vs δ,ε-unsaturated ketone. ^aIsolated yield was based on 0.1 mmol scale reaction.

The diene scope is also broad for both the GluER and EschER variants. Increasing diene ring sizes are tolerated. The GluER-B7 variant can produce 34A with excellent selectivity 99:1 (A:B) in a moderate 33% yield, respectively. The EschER-D1 variant can produce 34B with a selectivity of 23:77 in a moderate 44% yield. However, GluER-B7 and EschER-D1 both gave selectivity for product 35A when challenged with the 8-membered 1,3-cyclooctadiene. This is due to the inherent selectivity of the reaction when using 1,3-cyclooctadiene, as, under typical photoredox conditions, a 66:34 mixture of 35A:35B was produced (Supporting Information Figure 16). Substituted 1,3-cyclohexadienes are also accepted. Addition to 1-methyl-1,3-cyclohexadiene further demonstrates the excellent regiodivergence, as GluER-B7 provides 36A in 73% yield and good selectivity (80:20). EschER-D1 can furnish alternative constitutional isomer 36B in 72% yield and excellent selectivity (2:98). This could also be run on a 0.1 mmol preparative scale to isolate 36B in a 59% yield. Radical addition to ethyl cyclohexa-1,3-diene-1-carboxylate generates 1,2-addition product 37B in all cases. This is likely due to radical destabilization at the C4 position due to the ester moiety. However, only EschER-D1 produces 37B with excellent selectivity (1:99) in moderate yield. A 0.1 mmol preparative scale synthesis produced 37B in a 55% yield. It is interesting to note that the engineered GluER-B7 variant could produce a significant proportion of the thermodynamically less favored isomer 37A (39:61). This is also in comparison to the parent GluER-T36A variant, which favored production of 37B (62% yield, 19:81 (A:B)); Supporting Information Table 8). The ester moiety of 37B provides a handle for the further valorization of these products. Similarly, it may be possible to screen other EREDs for asymmetric reduction of the activated alkene.²¹ Danishefsky diene is accepted to give the diketone product 38B in up to 38% yield using EschER-D1. This occurs through generation of the α-oxo radical and subsequent elimination of the silyl group to form the ketone.^22,23 Linear dienes with aromatic moieties were accepted, with 1-phenyl-1,3-butadiene giving only the 1,2-addition product 39B in up to 60% yield. In comparison, the addition of a methyl group yields a mixture of products 40A and 40B. It is impressive that GluER-B7 is able to produce a significant proportion (11% yield, 47:53 (A/B)) of the less favored product 40A, in which the alkene is not in conjugation with the phenyl ring. The GluER-T36A starting variant favors product 40B (19% yield, 38:62 (A:B)); Supporting Information Table 8). The cumulative diene 22 was also accepted with radical addition to the central carbon atom to yield the thermodynamically favored product 41A in 76% yield using GluER-T36A.

Preparative scale synthesis of γ,δ-unsaturated ketones can be achieved with GluER-B7, producing 3A in 60% yield (96:4). This product could be converted to the O-aryl oxime 42 and subsequent photocatalyzed hydroimination cyclization to generate pyrroline product 43 in 52% yield, 3:1 dr (Figure 5).²⁴

Figure 5.

Further functionalization of 3A to generate 5-membered pyrroline heterocycle 43.

The highly selective formation of γ,δ-unsaturated ketone 3A from readily available 1,3-cyclohexadiene provides a mild and facile synthesis of these multifunctional products. The preprogrammed selectivity for constitutional isomers influences the selectivity of further steps, allowing for the formation of the 5-membered pyrroline 43.

Conclusions

In conclusion, we have discovered that the flavin-dependent EREDs can catalyze the C–C bond-forming hydroalkylation of α-bromo ketones with unactivated dienes. While the wild-type enzymes are modestly selective, two rounds of protein engineering were sufficient to provide synthetically useful levels of regioselectivity, enabling the synthesis of both constitutional isomers. Deuterium labeling experiments provided evidence that the selectivity-determining step is HAT from flavin. Modeling with GluER-T36A and GluER-B7 suggests that the result of the engineering campaign is to place the C4 carbon of the product intermediate closer to the N5 of flavin, favoring HAT at this position. While the selectivity for a single constitutional isomer was highest for reactions with the 1,3-cyclohexadiene model substrate, other cyclic and acyclic dienes were reactive. Overall, this work highlights the opportunity for enzymes to control challenging mechanistic steps that enable more selective and efficient chemical synthesis.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/jacs.4c00482.

• General information; DNA and protein information; directed evolution information; experimental procedures; reaction optimization; general procedure for biocatalytic reactions on small and preparative scales; deuterium labeling experiments; general procedure for diene synthesis; general procedures for product standard synthesis and characterization; and NMR spectra and references (PDF)

Author Present Address

^# Department of Chemistry, Princeton University, Princeton, New Jersey 08544, United States

Author Contributions

All authors have given approval to the final version of the manuscript.

The authors declare no competing financial interest.