Cobaloxime-catalysed regiodivergent hydrogen atom transfer for alkenyl and allylic carbamoylation with branched alkenes

Abstract

Hydrogen atom transfer (HAT) represents a fundamental reaction in synthetic chemistry, with ongoing innovations aimed at improving its selectivity and efficiency. Leveraging the unique reactivity of cobaloxime, we developed a regioselective HAT process that enables divergent carbamoylation of branched alkenes. Through modulation of the electronic and steric properties of the ligands, this cobalt complex achieves precise discrimination between hydrogen atoms exhibiting minor differences in bond dissociation energies (BDEs), affording vinyl and allylic amides with predictable selectivity. This operationally simple method features a broad substrate scope, excellent functional group tolerance, and is applicable to the late-stage functionalization of natural products. The synthetic utility of this radical approach is highlighted by the asymmetric total synthesis of a lignan natural product (−)-arctigenin. Experimental and computational studies shed important light on the mechanism and the critical role of the cobaloxime catalyst in regiocontrol.

The authors report a regioselective hydrogen atom transfer process that enables divergent carbamoylation of branched alkenes. Through modulation of the electronic and steric properties of the ligands, this cobalt complex achieves precise discrimination between hydrogen atoms exhibiting minor differences in bond dissociation energies (BDEs), affording vinyl and allylic amides with predictable selectivity.

Introduction

Hydrogen atom transfer (HAT) is a fundamental reaction involving the concerted transfer of a proton and an electron between two species¹. It underpins diverse chemical and biological processes, including hydrocarbon combustion, atmospheric chemistry, and enzymatic catalysis. Recent advances in photoredox and electrochemical catalysis have significantly expanded the synthetic applications of HAT-mediated C–H functionalization (Fig. 1a)^2–6. Homolytic cleavage of C–H bonds provides direct access to carbon-centered radicals that can be intercepted for new bond-forming processes^7–12. However, achieving high regioselectivity in HAT remains a formidable challenge due to the similar electronic and steric properties of ubiquitous C − H bonds¹³.

Fig. 1. Regioselective functionalization of alkenes via cobaloxime catalysis.

a HAT-mediated C–H functionalization. b Radical alkenylation and allylation. c Selective alkylation of linear alkenes. d Cobaloxime catalysis. e Regiodivergent carbamoylation of branched alkenes. FG functional group.

Beyond sp³-hybridized centers, HAT-mediated processes also enable olefinic C–H functionalization^14–17. Recent studies demonstrate that alkenylation can proceed via radical addition to an alkene, followed by double-bond restoration through C–H cleavage at an adjacent carbon atom (Fig. 1b)^18–23. These one-electron approaches provide a versatile strategy for constructing substituted alkenes, leveraging the high reactivity of radicals with feedstock olefinic acceptors. In this context, Fu and co-workers reported radical alkylation of linear aliphatic alkenes^22,23, which is less studied in conventional Heck reactions^24–27. To achieve hydrogen abstraction at a specific position, a directing-group was employed, resulting in selective formation of vinyl products (Fig. 1c). Nevertheless, regioselective functionalization of branched alkenes remains largely underdeveloped^28,29. The difficulties arise from competing HAT processes, which generate mixtures of vinyl and allylic products. As depicted in Fig. 1b, abstraction of the more accessible H_b yields the kinetically favored terminal alkene, while cleavage of the C−H_a bond affords the thermodynamically stable internal isomer. Consequently, developing catalysts capable of distinguishing between Hₐ and H_b represents a critical objective. Such advancement would enable regiodivergent functionalization of alkenes³⁰, delivering either vinyl or allylic products as desired.

Cobaloxime, originally synthesized to mimic vitamin B12, has been recognized as a highly effective catalyst for hydrogen evolution reactions under mild conditions^31–34. Recent advancements in photocatalysis have further expanded the applications of these four-coordinated cobalt complexes, leveraging their rich redox chemistry^35,36. In particular, the planar Co^II species can serve as a versatile catalyst for HAT^37–43. In 2021, Leonori and co-workers integrated halogen-atom transfer (XAT) with photoredox and cobaloxime catalysis to achieve regioselective E2 elimination of unsymmetrical alkyl halides³⁹. The initial XAT step generated an alkyl radical, which subsequently interacted with the cobalt catalyst to facilitate the elimination process (Fig. 1d). Modulation of the steric and electronic properties around the cobalt center provided contra-thermodynamic olefin products, which are typically difficult to access under conventional E2 elimination conditions. Furthermore, the combination of a photocatalyst and cobaloxime offered a solution for radical alkenylations^44–52 with branched alkenes. Lei and co-workers demonstrated that cobaloxime selectively cleaved the less sterically hindered allylic C − H bond adjacent to the carbon-centered radical, which was generated via the addition of aminium radical cations to olefins⁵³. This site-selective strategy has since been effectively applied to fluoroalkylation^54–56 and sulfonylation reactions⁵⁷. However, overriding this kinetically favored selectivity to generate internal alkenes remains a significant challenge. Given that the steric and electronic properties of cobaloximes significantly influence their reactivity, we hypothesized that regioselectivity could be modulated through the strategic choice of ligands (Fig. 1e). Specifically, we postulated that an electron-rich cobaloxime would preferentially abstract the more protic hydrogen (H_a), while a Co catalyst with sterically congested planar ligands would favor HAT at the less hindered site (H_b).

Herein, we report the successful application of this catalyst-controlled strategy to achieve regiodivergent carbamoylation of branched alkenes. The site-selective HAT process provides efficient access for the synthesis of both vinyl and allylic amides. The method exhibits broad substrate scope with excellent regioselectivity, as demonstrated by late-stage functionalization of natural products, the total synthesis of (−)-arctigenin, and other synthetic manipulations.

Results

Reaction development

Owing to their fundamental role in biological systems and drug discovery, amides are widely applied in both the pharmaceutical industry and academic research. In this context, photoredox-catalysed carbamoylation reactions have attracted considerable interest from chemists. For example, Melchiorre’s group employed an amide-derived dihydropyridine 1 as a carbamoyl radical precursor⁵⁸, enabling efficient amide synthesis and peptide modification^59,60. As illustrated in Fig. 2a, we propose a dual photoredox-cobalt catalytic system for the regioselective carbamoylation of alkenes to access both vinyl and allylic amides. The catalytic cycle begins with oxidative quenching of the excited photocatalyst (*PC) by cobaloxime, generating a Co^II intermediate and PC^•+. The latter oxidizes 4-carbamoyl-1,4-dihydropyridine 1a, producing the carbamoyl radical A and regenerating the ground state photocatalyst (PC). The carbamoyl radical then adds to the branched alkene 2a, forming the benzyl radical B, which undergoes cobalt-promoted HAT to yield the desired product 3. Finally, the reaction between Co^III–H and a proton release H₂ and complete the cobalt cycle. Based on the analysis of the polar and steric properties of the two hydrogen atoms, we propose that regioselective HAT can be achieved by fine-tuning of the cobaloxime catalyst (Fig. 2b). Specifically, we hypothesize that an electron-rich Co catalyst favors abstraction of the protic hydrogen (H_a), while steric hindrance shifts selectivity toward the more accessible hydrogen atom (H_b).

Fig. 2. Exploration of reaction conditions.

a Proposed mechanism. b Regiodivergent HAT catalysed by cobaloxime. c Optimization of reaction conditions. d Evaluation with various cobaloximes, namely Co(dmgH)₂PyCl (Co-1), [Co(dmgH)₂Py₂]PF₆ (Co-2), Co(dMesgH)₂PyCl (Co-3). ^[a]Substrate 1a (0.10 mmol), styrene 2a (4.0 equiv.), 4CzTPN (3 mol%), cobaloxime (5 mol%), pyrene (10 mol%), CsHCO₃ (1.0 equiv.), KOAc (0.5 equiv.) in solvent (3.0 mL) under blue light irradiation. ^[b]Isolated yield. ^[c]Without 4CzTPN. ter terminal olefin, int internal olefin, n.d. not detected.

The feasibility of this proposal was evaluated by investigating the reaction of 4-carbamoyl-1,4-dihydropyridine 1a with styrene 2a, using 4CzTPN as the photocatalyst in THF under blue light irradiation at room temperature (Fig. 2c). As expected, the electron-rich Co-1 catalyst provided the thermodynamically stable alkene 3-int in 57% yield, favoring HAT at the protic site (entry 1). The alternation of the axial ligand increased the yield to 82%, while maintaining excellent regioselectivity (entry 2). Furthermore, the sterically hindered cobaloxime Co-3 completely reversed the regioselectivity, yielding the terminal olefin 3-ter with >19:1 r.r. (entry 3). Additionally, switching the solvent to CH₂Cl₂ improved the yield to 95%. Control experiments confirmed that the reaction could not proceed in the absence of cobaloxime or the photocatalyst (entries 5–7).

Substrate scope

Considering the importance of amides in both the pharmaceutical industry and academic research, we endeavored to expand the applicability of this regiodivergent carbamoylation reaction with various carbamoyl radical precursors. As shown in Fig. 3, two distinct cobaloxime catalysts facilitated the selective formation of both vinyl and allylic amides. With the sterically hindered Co catalyst (Co-3), terminal alkenes 3-ter to 15-ter were synthesized with yields ranging from 43% to 95%. The reaction proceeded with Co-2 provided α,β-unsaturated amides with excellent regioselectivity and stereoselectivity ( > 19:1 E/Z). A range of N-alkyl carbamoyl groups were compatible, producing secondary and tertiary amides in moderate to high yields. Notably, amides bearing the sterically hindered adamantyl moiety were synthesized in both catalytic systems (12-ter and 12-int). The reaction also proceeded efficiently with dihydropyridines derived from amino acids, including alanine (13), histidine (14), and tryptophan (15), which were successfully installed into the amide products with complete regiomeric control. Furthermore, the dual catalytic system was also effective for the selective acylation of branched alkenes with 4-acyl-1,4-dihydropyridines (16 and 17).

Fig. 3. Scope of carbamoyl radical precursors.

Reactions generally performed on 0.2 mmol scale at 40 °C for 16 h under irradiation by blue LEDs and using dihydropyridines 1 (1.0 equiv.), α-methylstyrene 2a (4.0 equiv.), CsHCO₃ (1.0 equiv.), KOAc (0.5 equiv.) in CH₂Cl₂ or THF (6.0 mL). Isolated product yields. Regiomeric ratios (r.r.) and stereoselectivity ( > 19:1 E/Z for internal alkenes) were determined by ¹H NMR analysis of the mixtures. ^[a] The reaction was carried out on a gram scale with 7.0 mmol of 2a. ^[b] E/Z = 10:1.

Next, the scope of styrenes was explored (Fig. 4). A broad range of branched alkenes with different substituents on the aromatic ring were effectively accommodated in both catalytic systems, yielding both regioisomers with excellent selectivity. The arenes could be substituted with both electron-donating and electron-withdrawing groups, affording the amides in high yields (18–24). A variety of functional groups, including trifluoromethyl, cyano, ester, thioether, and naphthalene, were successfully incorporated into the products (25–29). Additionally, the transformation proceeded smoothly with substituted thiophenes (31–32).

Fig. 4. Scope of aromatic alkenes.

See Fig. 3 for detailed procedures. Isolated product yields. Regiomeric ratios (r.r.) and stereoselectivity ( > 19:1 E/Z for internal alkenes) were determined by ¹H NMR analysis of the mixtures.

Subsequently, we investigated whether the optimized reaction conditions were effective for the regioselective carbamoylation of alternative alkenes (Fig. 5). Styrenes with ethyl or propyl substituents at the α-position underwent smooth reactions, yielding the allylic amides 33 and 34 in good yields, respectively. In addition to aromatic alkenes, this carbamoylative method also proved compatible with enol ether and methacrylate (37 and 38). Furthermore, natural products and their derivatives successfully participated as coupling partners, including derivatives of eugenol (39), ( − )-menthol (41), ( − )-isopulegol (42), and andrographolide (47). Additionally, this synergistic catalytic system demonstrated excellent compatibility with natural terpenes, producing a range of amide-containing molecules 43–45 with >19:1 r.r. Both (−)-β-pinene and caryophyllene oxide underwent carbamoylation to the ring-opened products 40 and 46 through the cleavage of the cyclobutane ring, which highlights the radical nature of the reaction. Unfortunately, the reaction with electron-rich cobaloxime Co-2 failed to transform these alkenes to the vinyl products. Thus, further efforts were made to explore a suitable condition for the isomerization of these allylic amides to its regioisomer. As further illustrated in Fig. 5, the vinyl amides were successfully afforded in high yields under basic conditions.

Fig. 5. Additional scope of alkenes.

Reactions generally performed on 0.2 mmol scale at 40 °C for 16 h under irradiation by blue LEDs and using dihydropyridine (1.0 equiv.), alkenes (8.0 equiv.), CsHCO₃ (1.0 equiv.), KOAc (0.5 equiv.) in CH₂Cl₂ (4.0 mL). Isolated product yields. Regiomeric ratios (r.r.) were determined by ¹H NMR analysis. ^[a] E/Z = 10:1.

Synthetic applications

Derivatizations of the products demonstrated the utility of this cobalt-catalysed transformation for accessing diverse amide-containing molecules (Fig. 6a). Cyclopropanation of the internal (3-int) and terminal (3-ter) alkenes afforded amides 48 and 49 in good yields, respectively. Photocatalytic cyclization of the α,β-unsaturated amide 8-int yielded quinolinone 50⁶¹, which was subsequently converted to quinoline-2-thione 51 using Lawesson’s reagent. On the other hand, hydroboration of the allylic regioisomer 8-ter, followed by oxidative work-up, provided γ-hydroxyamide 52, which underwent cyclization to furnish γ-lactam 53. The synthetic value of this methodology was further highlighted by its application in the total synthesis of the lignan natural product (−)-arctigenin. As outlined in Fig. 6b, regioselective carbamoylation of dialkyl-substituted alkene 54 provided allylic product 55 in high yield. Subsequent rhodium-catalysed asymmetric hydroboration of β,γ-unsaturated amide 55 proceed with high regio- and enantioselectivity employing the TADDOL-derived phosphite ligand L1⁶². The following oxidation gave γ-hydroxyamide 56, which was cyclized into γ-lactone 57. Eventually, ( − )-arctigenin 58 was obtained by diastereoselective alkylation of 57 followed by benzyl group removal. The spectroscopic data, including the specific rotation of this obtained sample, completely matched the literature data⁶³.

Fig. 6. Synthetic applications.

a Derivations of products. b Asymmetric synthesis of (−)-arctigenin. c Regiodivergent carbamoylation of peptides.

We next directed our efforts to the regioselective functionalization of peptide N-termini (Fig. 6c), which is a valuable tool for both basic research and the development of new drug candidates⁶⁴. In this context, a series of peptidyl-dihydropyridines were synthesized via late-stage modification with 4-carbamoyl-1,4-dihydropyridines. Encouragingly, the carbamoylation reaction worked well with peptides containing three to five amino acids, such as the bioactive leu-enkephalin, affording the desired vinyl and allyl products 59–61 with excellent regioselectivity.

Experimental investigations on the reaction pathways

To gain further insight into the catalytic cycle, the mechanism of the carbamoylation reaction was investigated by a radical clock experiment (Fig. 7a). The cyclopropyl-containing alkene 62 underwent ring opening followed by intramolecular radical addition to afford dihydronaphthalene 63 in good yield. These results strongly support a radical-involved pathway. To monitor hydrogen evolution in the cobaloxime catalytic cycle, the gas phase in the reaction vessel headspace was analyzed by gas chromatography (GC). A 200 μmol-scale reaction produced 60 μmol and 86 μmol of hydrogen in CH₂Cl₂ and THF, respectively (Fig. 7b). Furthermore, the kinetic isotope effect (KIE) studies were performed using deuterated α-methylstyrene [D]-2a (Fig. 7c). Competitive KIE experiments exhibited k_H/k_D values of 5.7 and 3.3, suggesting that HAT is likely to be the rate-determining step in the catalytic cycle. In addition, the isolated vinyl and allylic products were individually subjected to the optimal reaction conditions (Fig. 7d). In both cases, no interconversion between the two products was observed, implying that they are formed via regiodivergent pathways.

Fig. 7. Mechanistic studies.

a Radical clock experiment. b H₂ detection experiment. c Kinetic isotope effect (KIE) experiment. d Control experiment. Reaction conditions A: dihydropyridine 1a (0.2 mmol), alkene (4.0 equiv.), 4CzTPN (3 mol%), Co(dMesgH)₂PyCl (Co-3, 5 mol%), pyrene (10 mol%), CsHCO₃ (1.0 equiv.), KOAc (0.5 equiv.) in CH₂Cl₂ (3.0 mL) under blue light irradiation. Reaction conditions B: dihydropyridine 1a (0.2 mmol), alkene (4.0 equiv.), 4CzTPN (3 mol%), [Co(dmgH)₂Py₂]PF₆ (Co-2, 5 mol%), pyrene (10 mol%), CsHCO₃ (1.0 equiv.), KOAc (0.5 equiv.) in THF (3.0 mL) under blue light irradiation.

Computational studies

To understand how the regioselective hydrogen abstraction by the cobaloxime can be controlled by ligand substituents, density functional theory (DFT) computations were performed on the product-determining hydrogen abstraction steps. Previous DFT studies on related cobaloxime systems have provided important insights into their reactivity and selectivity^51,65,66. In this work, 4-dimethylcarbamoyl-1,4-dihydropyridine and styrene 2a were used as the model substrates, with the diethylamino moiety of 1a simplified to a dimethylamino group. As shown in Fig. 8a, we first located the transition states TS1a-Int and TS1a-Ter corresponding to the formation of the internal olefin product 4-int and the terminal olefin product 4-ter, respectively, under the catalysis of Co-2. In both transition states, the C–H bond at the reactive site is significantly elongated (C–H bond length: 1.42 Å for TS1a-Int; 1.46 Å for TS1a-Ter). The C, H, and Co atoms are nearly colinear (θ_C-H-Co: 174.9° for TS1a-Int; 172.7° for TS1a-Ter), consistent with a typical hydrogen atom transfer (HAT) process. Single-point energy calculations revealed that both transition states exhibit spin-polarized singlet state as the ground state, with <S²> values of 0.35 and 0.19 for TS1a-Int and TS1a-Ter, respectively, suggesting partial diradical character. The calculated free energy barriers for TS1a-Int and TS1a-Ter are 15.4 kcal/mol and 17.0 kcal/mol, respectively, consistent with the experimentally observed regioselectivity (for a complete energy profile, see Figs. S13–S14 in the Supplementary Information).

Fig. 8. DFT studies for the origin of regioselectivity.

a Structures and free energy barriers of the competing transition states. b Differences in free energy barriers. c Bond dissociation energies (BDE) of the benzyl radical intermediates. d Distortion energy analysis for the substrate (ΔE_dis-sub) and catalyst (ΔE_dis-cat) fragments. Calculations were performed at the (u)B3LYP-D3(BJ)/6-311 G(d,p) (SDD for Co)/SMD//(u)B3LYP-D3(BJ)/6-31 G(d) (SDD for Co) level of theory at 313.15 K with Grimme’s entropic correction and Head-Gordon’s enthalpic correction. Energies are reported in kcal/mol.

For the reaction catalysed by the sterically more hindered Co-3, the corresponding transition states TS2a-Int and TS2a-Ter were also located. Similar to their Co-2 counterparts, the C–H bond lengths in the TS2a-Int and TS2a-Ter are significantly elongated. However, the angles between C, H and Co atoms deviate more from a linear alignment (θ_C-H-Co: 164.8° for TS2a-Int; 162.5° for TS2a-Ter), potentially indicating a more congested catalytic center in Co-3. Single-point energy calculations show that TS2a-Int and TS2a-Ter also possess spin-polarized singlet states ( < S² > = 0.34 and 0.19, respectively), with their free energy barriers similar to those of TS1a-Int and TS1a-Ter (19.8 kcal/mol for TS2a-Int; 17.6 kcal/mol for TS2a-Ter). Notably, these energy barriers can still be readily overcome at room to moderate temperatures. Interestingly, in contrast to the Co-2 system, TS2a-Int is energetically less favored than TS2a-Ter by 2.2 kcal/mol. This reversal in regioselectivity is in line with experimental observations when Co-3 is employed.

To further explore the role of the cobalt catalyst and substrate architectures in controlling this regioselectivity, we performed calculations on two additional substrates, in which the dimethylamino group (NMe₂) in the modified 1a was replaced by a methyl (Me) and phenyl (Ph) group, respectively (for a complete energy profile, see Figs. S15–S18 in the Supplementary Information). The calculated free energy barriers for the two competing transition states of each substrate under Co-2 and Co-3 catalysis are shown in Fig. 8b. Overall, the barrier heights are comparable to those of the dimethylamino-substituted substrate. The predicted regioselectivities for both substrates are also consistent with our experimental observations.

To probe the origin of regioselectivity, we first calculated the C–H bond dissociation energies (BDE) of the benzyl radical intermediates derived from the three substrates. As shown in Fig. 8c, the C–H_a bond is consistently weaker than the C–H_b bond in all cases, indicating an intrinsic preference for HAT process at C–H_a. This trend agrees with the selectivity observed with Co-2. We then performed a distortion energy analysis of the competing transition states (Fig. 8d). Each transition state was divided into a substrate fragment (sub) and a cobalt catalyst fragment (cat). The distortion energies, ΔE_dis-sub and ΔE_dis-cat, were defined as the energy differences between their geometries in the transition state and their respective optimized ground states. For the less hindered Co-2 (TS1-Int and TS1-Ter columns; Fig. 8d), ΔE_dis-cat is minimal (0.6–0.8 kcal/mol) for both competing transition states, indicating negligible catalyst distortion. The total distortion energy mainly arises from the substrate fragment. In addition, for all three substrates, the ΔE_dis-sub of the terminal pathway is slightly higher than that of the internal pathway, consistent with the BDE trend and the experimentally observed internal selectivity. In contrast, for the more sterically hindered Co-3 (TS2-Int and TS2-Ter columns; Fig. 8d), steric interactions between the approaching fragments become more pronounced, leading to increased distortion in both components. This is evidenced by the larger ΔE_dis-sub values as well as an increase of ΔE_dis-cat to 1.5–2.9 kcal/mol for all transition states. Notably, the increase in substrate distortion ΔE_dis-sub is more pronounced for the internal transition states, whereas the terminal transition states are comparatively less affected compared to the Co-2 scenario, thereby shifting the energetic preference to the terminal pathway.

Taken together, these results suggest a mechanistic picture: with the less hindered Co-2 catalyst, steric interactions between the substrate and the cobalt center are minimal, and HAT occurs preferentially at the intrinsically weaker C–H_a bond, leading to the more thermodynamically stable internal product. In contrast, the bulkier Co-3 catalyst introduces more steric repulsion as the substrate approaches the metal center, requiring greater structural distortion of both fragments in the transition state. Under this condition, the kinetically more accessible C-H_b bond becomes favored, resulting in a reversal of regioselectivity.

Discussion

In summary, we developed a regiodivergent carbamoylation reaction of branched alkenes via cobaloxime-catalysed HAT. Central to the success of this strategy is the rational modulation of the cobalt catalyst, which enables switching between thermodynamically and kinetically favored pathways. The method provides efficient access to both α,β- and β,γ-unsaturated amides and has been applied to the total synthesis of (−)-arctigenin. Experimental and computational investigations elucidate the mechanism and the unique role of the cobaloxime catalyst in promoting HAT. We anticipate that this strategy will facilitate further exploration and development of HAT-mediated C − H functionalization.

Methods

A general procedure for the regioselective carbamoylation of alkenes

To an oven-dried reaction tube equipped with a stir bar was added 1 (0.2 mmol, 1.0 equiv.), cobaloxime catalyst (5 mol%), 4CzTPN (3 mol%), pyrene (0.02 mmol, 10 mol%), CsHCO₃ (0.2 mmol, 1.0 equiv.), and KOAc (0.1 mmol, 0.5 equiv.). The tube was sealed and purged with nitrogen. Alkene 2 (0.8 mmol, 4.0 equiv.) and solvent (6 mL) were then added. The resulting mixture was stirred and irradiated with blue LEDs at 40 °C. Upon completion, the reaction mixture was purified by silica gel column chromatography to afford the desired product.

Density functional theory (DFT) calculations

All calculations were performed using the Gaussian 16 program⁶⁷. Geometry optimizations and vibrational frequency analyses for all intermediates and transition states were carried out in the gas phase at the (u)B3LYP-D3(BJ)/6-31 G(d) level of theory (u denotes unrestricted calculations), with the SDD effective core potential and corresponding basis set applied to Co atom. Single-point energy calculations were subsequently performed at the (u)B3LYP-D3(BJ)/6-311 G(d,p) level (SDD used for Co atom) with the SMD implicit solvent model. THF (Tetrahydrofuran) and DCM (Dichloromethane) were used as the solvent environments for the intermediates and transition states associated with Co-2 and Co-3, respectively. Gibbs free energies were calculated at 313.15 K with Grimme’s entropic correction and Head-Gordon’s enthalpic correction applied using Goodvibes program (version 3.2)⁶⁸. Additional computational details and the Cartesian coordinates of all calculated structures are provided in the Supplementary Information.

Author contributions

J.Z. and K.N.H. conceived the work. Z.J., J.S., and J.-N.M. designed the experiments and analysed the data. Z.J. and F.Z. performed the experiments. J.-R.S. contributed to the DFT calculations and mechanistic studies.

Peer review

Peer review information

Nature Communications thanks Shashikant U. Dighe, Hong Yi, and the other, anonymous, reviewers for their contribution to the peer review of this work. A peer review file is available.

Funding

J.Z. is grateful for the financial support provided by the National Natural Science Foundation of China (No. 22371028), the Natural Science Foundation of Liaoning Province (2024-MSLH-064), and the Fundamental Research Funds for the Central Universities (DUT25Z2517). K.N.H. thanks the funding support from the National Science Foundation (CHE-2153972 for K.N.H. and J.-R.S.). K.N.H. also thanks the supercomputing resources from the Expanse at SDSC, allocated under CHE040014, through the Advanced Cyberinfrastructure Coordination Ecosystem: Services & Support (ACCESS) program (for K.N.H. and J.-R.S.). This work also uses computational and storage services associated with the Hoffman2 Cluster, which is operated by the UCLA Office of Advanced Research Computing’s Research Technology Group.

Data availability

Crystallographic data for the structures reported in this Article have been deposited at the Cambridge Crystallographic Data Centre, under deposition numbers CCDC 2488763 (9-int) and CCDC 2488762 (33). Copies of the data can be obtained free of charge from the CCDC via https://www.ccdc.cam.ac.uk/structures/. All other data supporting the findings of this study, including experimental procedures and compound characterization, are available within the Article and its Supplementary Information. Coordinates of the optimized structures are provided in the source data file. Source data are provided with this paper. Data supporting the findings of this manuscript are also available from the corresponding author upon request. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-74595-8.