Intercepting fleeting cyclic allenes with asymmetric nickel catalysis

Abstract

Strained cyclic organic molecules, such as arynes, cyclic alkynes and cyclic allenes, have intrigued chemists for more than a century with their unusual structures and high chemical reactivity¹. The considerable ring strain (30–50 kilocalories per mole)^2,3 that characterizes these transient intermediates imparts high reactivity in many reactions, including cycloadditions and nucleophilic trappings, often generating structurally complex products⁴. Although strategies to control absolute stereochemistry in these reactions have been reported using stoichiometric chiral reagents^5,6, catalytic asymmetric variants to generate enantioenriched products have remained difficult to achieve. Here we report the interception of racemic cyclic allene intermediates in a catalytic asymmetric reaction and provide evidence for two distinct mechanisms that control absolute stereochemistry in such transformations: kinetic differentiation of allene enantiomers and desymmetrization of intermediate π-allylnickel complexes. Computational studies implicate a catalytic mechanism involving initial kinetic differentiation of the cyclic allene enantiomers through stereoselective olefin insertion, loss of the resultant stereochemical information, and subsequent introduction of absolute stereochemistry through desymmetrization of an intermediate π-allylnickel complex. These results reveal reactivity that is available to cyclic allenes beyond the traditional cycloadditions and nucleophilic trappings previously reported, thus expanding the types of product accessible from this class of intermediates. Additionally, our computational studies suggest two potential strategies for stereocontrol in reactions of cyclic allenes. Combined, these results lay the foundation for the development of catalytic asymmetric reactions involving these classically avoided strained intermediates.

Arynes, cyclic alkynes and cyclic allenes have captured the attention of the scientific community for more than a century¹. The field originated with Stoermer and Kahlert’s controversial proposal of 2,3-benzofuranyne’s (1) existence in 1902⁷ (Fig. 1a). Fifty years later, Roberts⁸ and Wittig^9,10 independently validated the intermediacy of benzyne (2), prompting extensive theoretical and experimental studies¹. The related intermediates cyclohexyne (3) and 1,2-cyclohexadiene (4) were discovered soon thereafter and were shown to engage in similar strain-promoted reactivity^11,12. Intermediates 2–4 exhibit high strain energies, ranging from 30–50 kcal mol⁻¹ (refs. ^2,3) and, consequently, high reactivities and short lifetimes^13,14. Nowadays, these classes of strained cyclic intermediates, particularly arynes, are used in the synthesis of heterocycles¹⁵, drug scaffolds¹⁶, ligands for catalysis (such as XPhos)¹⁷, natural products^18–20, agrochemicals²¹ and organic materials²².

Fig. 1|. Historical context of strained cyclic intermediates and current reaction design.

a, Milestones in strained cyclic intermediate chemistry. b, Seminal report of strained cyclic allene racemization, wherein an enantioenriched cyclic allene precursor is converted to a racemic cycloadduct. c, Our reaction design wherein a racemic cyclic allene precursor would be converted to an enantioeriched annulated product via the dynamic kinetic resolution of strained cyclic allene intermediates. d, Envisioned reaction pathway delivering enantioenriched products through a stereoselective olefin insertion reaction. Ph, phenyl; OTf, trifluoromethanesulfonate; Et, ethyl; X, CH₂, (CH₂)₂, N-Boc, N-Cbz, or O; A, carbon substituent; B, nitrogen substituent; R, aryl or alkyl group; L*, chiral ligand.

Despite the many modern synthetic applications of transient strained intermediates, the development of catalytic asymmetric reactions of these species has remained challenging. So far, only two strategies exist, which are the palladium-catalysed [2+2+2] cycloaddition of arynes to form helically chiral products^23,24 and the α-arylation of in-situ-generated chiral enamines with electrochemically generated arynes²⁵. Overall, catalytic asymmetric strategies for controlling absolute stereochemistry in the trapping of strained cyclic intermediates are rare, particularly those suitable for the formation of stereodefined sp³ centres.

Cyclic allenes are attractive building blocks because trapping can occur under mild conditions and lead to the formation of two new bonds, with the creation of one or more sp³ centres^26–31. Moreover, the inherent axial chirality of cyclic allene intermediates and the relatively facile interconversion of enantiomers could provide an opportunity to achieve asymmetric reactions through a dynamic kinetic resolution (DKR)³². Pioneering studies by Balci and jones showed that cyclic allenes can racemize on a timescale that outcompetes their trapping (Fig. 1b), a requisite feature of the envisioned DKR process³³. Specifically, upon generating enantioenriched cyclic allene 6 from enantioenriched allene precursor (+)−5, only the racemic cycloadduct (±)−7 is obtained as a result of rapid racemization of allene 6. This observation, coupled with previous results from our own group²⁷, prompted our design of the enantioselective process shown in Fig. 1c. Racemic silyl triflate allene precursor 8 can be converted into a racemic mixture of allenes 9 and ent-9 upon exposure to a fluoride source. We expected these enantiomers to interconvert rapidly before selective interception of one allene enantiomer in a catalytic asymmetric reaction. Overall, such a process would convert racemic silyl triflate 8 into enantioenriched product 11 via a DKR.

To explore the proposed strategy, we sought to identify a suitable catalytic reaction capable of efficiently and selectively intercepting the cyclic allene. As metal-catalysed reactions of strained cyclic allenes are almost unknown^26,34, we identified several criteria for initial reaction discovery. The reaction selected for our study would ideally: (1) be compatible with conditions for strained intermediate generation, (2) allow for the facile in situ generation of an organometallic intermediate in catalytic quantities, and (3) be amenable to the development of an asymmetric variant. With these requirements in mind, the denitrogenative annulation of benzotriazinones³⁵ was deemed an attractive reaction manifold. Such annulations have been demonstrated to be compatible with fluoride-mediated generation of transient species, such as arynes³⁶. Furthermore, oxidative addition into benzotriazinones 12 is known to proceed rapidly and irreversibly to form metallacycle 14 (Fig. 1d), which we envisioned should provide a sufficient concentration of 14 to intercept short-lived allene 9 (ref. ³⁷). Lastly, elegant asymmetric variants employing linear allenes have also been reported by Murakami³⁸.

Having identified the ideal system in which to test our hypothesized DKR, we pursued the specific transformation shown in Fig. 1d, wherein benzotriazinones 12 would react with racemic strained cyclic allene precursors 8 to provide tricyclic products 13. However, several challenges posed by the transformation shown in Fig. 1d should be noted. First, the fleeting allene intermediate 9 and the catalytically generated metallacycle 14, two species present in low concentrations, must react with one another. Additionally, to render the transformation enantioselective via the envisioned DKR, racemization of allene 9 must outcompete olefin insertion. Lastly, in the olefin insertion step, organometallic species 14 must preferentially react with a single allene enantiomer to provide enantioenriched metallacycle 15, en route to product 13.

Towards the development of this enantioselective reaction, we first sought to establish the feasibility of a racemic variant (Fig. 2). After optimization, we found that a variety of benzotriazinones 16 and allene precursors 17 underwent the desired annulation. Reactions proceeded optimally when using Ni(cod)₂ as precatalyst, 1,1′-bis(diphenylphosphino)ferrocene (DPPF) as ligand, and cesium fluoride as the fluoride source in acetonitrile at 35 °C. Reactions occurred smoothly using 1:1 stoichiometry between benzotriazinones 16 and silyl triflates 17. Notably, we did not observe olefin isomerization of products 19 or 20 to provide the corresponding aromatic products at any point during reaction development. With regard to the scope of the benzotriazinone annulation partner, various nitrogen substituents were tolerated, including p-tolyl, p-fluorophenyl, methyl and a removable benzyl group, as demonstrated by the formation of annulated products 21–24. Additionally, both electron-rich and electron-deficient benzotriazinones could be used in the transformation, as evidenced by products 25 and 26. Moreover, heterocycles such as morpholine and thiophene were tolerated, giving rise to products 27 and 28, respectively. The use of oxacyclic and azacyclic allenes, as well as a seven-membered carbocyclic allene, cleanly provided tricycles 29–32. Furthermore, alkyl-substituted heterocyclic allenes underwent the annulation to afford products 33–35. Interestingly, product 33 was obtained as a 3.8 to 1 mixture of regioisomers 33a and 33b, favouring formation of a fully substituted carbon (that is, 33a), whereas products 34 and 35 were obtained as single regioisomers favouring the opposite sense of regiochemistry, an aspect of selectivity that is at present not well understood. The results shown in Fig. 2 establish the racemic methodology and demonstrate that, despite their high reactivity, strained cyclic allenes can be employed in synthetically useful transition metal-catalysed reactions, a finding that is expected to considerably expand the scope of products accessible from this class of intermediates. This transformation also provides a modular route to access phenanthridinone scaffolds, which complements existing strategies. We note that phenanthridinone frameworks are seen in biologically active compounds, such as the amaryllidaceae alkaloids, which have been explored as anti-cancer compounds³⁹.

Fig. 2|. Scope of benzotriazinones and cyclic allenes in the racemic annulation reaction.

Yields shown reflect an average of two isolation experiments. ^aReaction was run for 40 h using a silyl tosylate as a cyclic allene precursor. R, aryl or alkyl; R′, H or alkyl; OTf, trifluoromethanesulfonate; X, CH₂, (CH₂)₂, N-Boc, N-Cbz, O; Et, ethyl; cod, 1,5-cyclooctadiene; DPPF, 1,1′-bis(diphenylphosphino)ferrocene; Me, methyl; Bn, benzyl; Ph, phenyl; p-tol, p-Me-C₆H₄; Cbz, carboxybenzyl; Boc, tert-butyloxycarbonyl.

Following the development of the racemic variant, we sought to identify a chiral ligand capable of rendering the transformation enantioselective. We examined the annulation of 36 with cyclic allene precursors 8 using a variety of chiral ligands and reaction conditions (see Supplementary Information, part I, section C for details), with select key results for the carbocyclic allene precursor 37 shown in Fig. 3a. The use of Quinap (L1), which has been used previously in a related transformation of linear allenes⁴⁰, provided poor enantioselectivity and low yield (entry 1). The DuPhos (L2) ligand gave the desired product (−)-21 in 61% enantiomeric excess (e.e.), albeit in low yield (entry 2). Increased yields and enantioselectivities were observed with either Walphos ligand L3 or BOX ligand L4 (entries 3 and 4, respectively). The Josiphos ligand L5 provided superior enantioselectivity, delivering (−)-21 in 90% e.e., albeit with only 37% yield (entry 5). Further optimization revealed that reducing the ligand loading from 10 mol.% to 5 mol.% led to an increased yield of 70% (entry 6), presumably by suppressing ligand-mediated decomposition of the product, based on control experiments. To increase the conversion of silyl triflate 37, we also examined the influence of additives. The addition of an excess of tetrabutylammonium iodide (TBAI), intended to increase the rate of allene formation by improving fluoride solubility⁴¹, gave (−)-21 in increased yield (entry 7) (the use of TBAF directly did not result in the desired product, see Supplementary Information part I, section C for details). Finally, by lowering the reaction temperature to 3 °C and extending the reaction time to 24 h, (−)-21 was obtained in 85% yield and 94% e.e. (entry 8).

Fig. 3|. Optimization and scope of the asymmetric annulation.

a, Optimization of the asymmetric annulation. b, Scope of the asymmetric annulation reaction. ^aYields were determined by ¹H NMR analysis of the crude reaction mixture using 1,3,5-trimethoxybenzene as an external standard. ^bEmpirically, it was found that a 1:1 ratio of nickel to Josiphos ligand gave higher yields of product. Control experiments showed that product 21 undergoes non-specific decomposition in the presence of unligated Josiphos. However, we cannot discount the possibility that with a 1:2 ratio of Ni:Josiphos ligand, a less reactive, saturated Ni(Josiphos)₂ complex could also form, potentially further decreasing reaction efficiency. ^cReaction was performed for 24 h. ^dYield was determined by an isolation experiment. ^eReaction was performed with 10 mol.% Ni(cod)₂ and 10 mol.% of L5 in the absence of TBAI. ^fReaction was performed with 10 mol.% Ni(cod)₂ and 10 mol.% of L5 in the absence of TBAI at 23 °C. TBAI, tetrabutylammonium iodide; X, N-Boc, N-Cbz, or CH₂; PMP, p-methoxyphenyl.

Having optimized the asymmetric transformation, we evaluated the scope of the methodology with respect to both the benzotriazinone and silyl triflate (Fig. 3b). Beginning with the parent substrate combination, (−)-21 could be isolated in 85% yield and 94% e.e.. Variation of the benzotriazinone N-substituent was tolerated, as seen by the formation of annulation products (−)-38, (−)-22, (−)-23, (−)-39, and (−)-24 in good yields and with high selectivities. The synthesis of (−)-23 and (−)-24 demonstrate the methodology’s tolerance towards N-alkyl substituents, whereas (−)-39 and (−)-24 bear removable N-protecting groups. Heterocyclic allenes could also be employed in the asymmetric reaction. Interception of azacyclic allenes bearing either a Boc or Cbz group gave rise to enantioenriched products (−)-31 and (−)-30, respectively. Similarly, the use of an oxacyclic allene delivered adduct (−)-29 in 81% e.e.. One final example highlights the reaction between an N-benzyl benzotriazinone and an N-Boc cyclic allene to provide enantioenriched (−)-40, which bears differing N-substituents for orthogonal N-deprotection.

Using density functional theory, we computationally studied the mechanism of the reaction between metallacycle 41 and allenes 4 and ent-4 catalysed by Ni/(S)-(R)-Josiphos in MeCN (Fig. 4). Density functional theory calculations have been established as a viable computational tool to assess mechanistic aspects of strained cyclic allene reactions²⁷ and nickel-catalysed denitrogenative processes³⁷. Metallacycles analogous to 41 have previously been established as intermediates in related annulations³⁷. Additionally, formation of 4 from the corresponding silyl triflate is believed to occur readily, and alternative pathways involving direct oxidative addition into the silyl triflate rather than allene formation were ruled out on the basis of calculations showing this to be kinetically implausible (see Supplementary Information part II, section C for details regarding oxidative addition into the silyl triflate). Our first objective was to investigate the hypothesized DKR of the cyclic allene intermediate, which would require facile racemization of allene 4 and efficient discrimination of the two enantiomers by reaction with catalytically generated metallacycle 41 (Fig. 4a). Although the racemization of strained cyclic allenes can occur readily^27,33, kinetic differentiation of two strained cyclic allene enantiomers by a metal complex is unknown. Our calculations identified TS-1 (ΔG^‡ = 7.5 kcal mol⁻¹) and TS-1a (ΔG^‡ = 14.9 kcal mol⁻¹) as the lowest-energy transition states for the reaction of 41 with 4 and ent-4, respectively, via an olefin insertion step (for details concerning reversible precomplexation of 41 and 4/ent-4, see Supplementary Information part II, section A.ii.). In these competing transition states, TS-1 is 7.4 kcal mol⁻¹ lower in free energy than TS-1a, which is predicted to furnish intermediate metallacycle 42 in preference to stereoisomer 43. Distortion/interaction-activation strain analysis⁴² (Fig. 4b) of the two competing transition states suggests that the energy difference mainly originates from decreased distortion of the allene (ΔΔE_{dist, allene}^‡ = −5.3 kcal mol⁻¹) and increased interaction energy (ΔΔE_int^‡ = −4.7 kcal mol⁻¹) in TS-1 relative to TS-1a. The increased allene distortion in TS-1a can be explained by the steric repulsion between the terminal allenic C–H bond and the cyclohexyl group on the Josiphos ligand.

Fig. 4|. Computational study of the asymmetric annulation reaction mechanism.

a, Reaction pathway and calculated transition state barriers for the olefin insertion step. A precomplex is formed reversibly by the allene and Ni before the insertion, and the reported ΔG^‡ values are relative to the lowest energy allene-Ni complex (see Supplementary Information part II and Supplementary Fig. 5). Free energies relative to uncomplexed 41 and 4 (ΔG) are noted in parentheses. b, Optimized transition state structures TS-1 and TS-1a and distortion interaction analysis for the olefin insertion. c, Possible pathways from 42 to (−)-21/(+)−21. Calculated free energies relative to 42 are indicated in parentheses. d, Optimized transition state structures TS-3 and TS-3a for the outer-sphere nucleophilic attack. Calculations were performed at the M06L/6-311++G(d,p), SDD(Ni, Fe)/SMD(MeCN)//B3LYP-D3(BJ)/6-31G(d), LANL2DZ(Ni, Fe)/SMD(MeCN) level. All energies are reported in kilocalories per mole. L*, (S)-(R)-Josiphos; RE, reductive elimination.

Although our computational studies suggest that the enantiomers of cyclic allene 4 can be kinetically differentiated during reaction with metal complex 41, two concerns were apparent. With the assumption that metallacycles 42 and 43 could undergo direct reductive elimination, the computationally predicted e.e. for intermediate 42 (>99% e.e. [at 3 °C], ΔΔG^‡ = 7.4 kcal mol⁻¹) is much higher than the experimentally observed e.e. for the final product (−)-21 (94% e.e., ΔΔG^‡ = 1.9 kcal mol⁻¹). Upon validating the absolute stereochemistry of our annulation products (see Supplementary Information part I, section D for crystallographic data), we also found that direct reductive elimination from favoured intermediate 42 would provide the opposite enantiomer of the product (that is, (+)−21) compared to what is observed experimentally (that is, (−)-21). To probe this discrepancy, we calculated the kinetic barrier for direct reductive elimination of metallacycle 42 to give product (+)−21 (Fig. 4c). Indeed, a prohibitively high barrier of 37.1 kcal mol⁻¹ was found for TS-2, confirming that direct reductive elimination is not operative.

A plausible mechanism for the conversion of favoured metallacycle 42 to the observed annulated product (−)-21 was subsequently investigated (Fig. 4c). More specifically, we evaluated whether the Ƞ¹-coordinated metallacycle 42 could give rise to a zwitterionic Ƞ³-allylnickel complex 44, which could then undergo addition at either C1 or C3 to construct the C–N bond in 21. Metallacycle 42 can undergo facile conversion to 44, which is a low-energy resting state. This Ƞ¹-to-Ƞ³ isomerization ablates the stereocentre set in the olefin insertion step, leading to a prochiral π-allyl fragment. The subsequent desymmetrizing outer-sphere nucleophilic attack by the amidate anion can occur at C1 or C3, via TS-3 (12.9 kcal mol⁻¹) or TS-3a (14.4 kcal mol⁻¹), leading to (−)-21 and (+)−21, respectively. The calculated ΔΔG^‡ of 1.5 kcal mol⁻¹ for these competing transition states predicts that (−)-21 would be formed in 87% e.e. (calculated for 3 °C), which is in good agreement with the experimentally observed formation of (−)-21 in 94% e.e. The difference in free energy between TS-3 and TS-3a is hypothesized to arise from solvation differences between these highly polar transition states (Fig. 4d). In the case of TS-3, the transition state is lower in energy due to a solvent-accessible cavity between the ligand’s phenyl groups, which allows solute-solvent interactions that stabilize the amidate anion. In TS-3a, the bulky cyclohexyl (Cy) groups on the Josiphos ligand sterically shield the amidate from the solvent, causing TS-3a to be comparatively less stable and therefore higher in energy. This hypothesis is further supported by calculations showing that changing the cyclohexyl substituent to less sterically demanding groups lowers the ΔΔG^‡ value, as well as gas-phase calculations demonstrating that TS-3 is less favourable than TS-3a in the absence of solvent effects (see Supplementary Information part II, section B for additional computational details on solvent effects).

Overall, the computational results indicate a mechanism (summarized in Extended Data Fig. 1) involving: (1) kinetic differentiation of the cyclic allene enantiomers 4 through stereoselective olefin insertion with 41, (2) subsequent loss of stereochemical information by formation of Ƞ³-allylnickel complex 44, and (3) stereoselective outer-sphere attack at C1 to give (−)-21. Thus, in future reaction design, one may consider either kinetic differentiation of cyclic allene enantiomers or the intentional formation of Ƞ³-allylmetal intermediates as viable means of introducing absolute stereochemistry in reactions of strained cyclic allenes.

Extended Data

Extended Data Fig. 1|. Proposed catalytic cycle.

Computational studies support a mechanism involving activation of the benzotriazinone by the Ni catalyst, migratory insertion across one olefin of the cyclic allene, isomerization to a Ni π-allyl complex, and enantioselective outer-sphere attack to provide the observed phenanthridinone.

Data availability

Crystallographic data are available free of charge from the Cambridge Crystallographic Data Centre under CCDC 1987661. The authors declare that all other data supporting the findings of this study are available within the paper and its Supplementary Information files.