Reaction developmentTo functionalize ABB 1, we needed to produce two suitable transient radical species. Because of the different electronic natures of the two atoms to be functionalized, namely nitrogen and carbon (C3), it was necessary to carefully consider the selection of the radical source. To prevent a possible unfruitful one-electron oxidation of the ABB through photoredox scenarios35, we elected to perform triplet sensitization, and used sulfonyl imine 2 as the reaction partner (Fig. 2). Preliminary optimization experiments led us to identify the optimal stoichiometry, between 1 and 2, and reaction time (Supplementary Tables 1 and 2). During the selection of the ideal PS, we found out that thioxanthen-9-one (TXO) 6 afforded full conversion of imine 2 with only 10% yield of product 3, along with major amounts of imine dimer 4 (Supplementary Table 3). This suggests that the concentration of the iminyl radical in the reaction mixture is critical. We thus reasoned that slowing down the iminyl radical formation might channel the reactivity towards the formation of the target azetidine 336,37,38. To this end, we investigated different classes of PSs at a reduced catalyst loading of 0.25 mol%, characterized by a sufficiently high T1 E0,0 (that is, the energy of the first triplet excited state) >2.55 eV, and a small ΔST (Supplementary Fig. 3 and Supplementary Table 3)39. This last parameter corresponds to the difference between the energies of the first singlet (S1) and triplet (T1) excited state and promotes a rapid S1 → T1 and T1 → S1 interconversion through intersystem crossing (ISC) and reverse intersystem crossing (RISC), respectively. We hypothesized that a small ΔST together with an efficient RISC—or in other words, an inefficient triplet sensitization event—would downregulate the concentration of the PS’s T1 state in solution, controlling the imine sensitization and preventing the formation of 4. Indeed, we observed a consistent trend in the relationship between the ΔST value and the selectivity of the process towards the azetidine 3 versus imine dimer 4 (Supplementary Table 3). Notably, other parameters can also influence the selectivity of the process, including the PS loading, which was kept as low as 0.25 mol%, λmax (maximum absorption wavelength), the PS’s absorption tail and the PS’s solubility (Supplementary Fig. 4). Considering all these aspects, we found the best performance for the PS 9, characterized by the lowest ΔST of 0.05 eV and a T1 E0,0 = 2.74 eV (calculated, 2.83 eV; see below). The design of the PS was inspired by recent progress in the field of thermally activated delayed fluorescence emitters and their use in photocatalysis40,41,42. The PS was thoroughly characterized in terms of its prompt and delayed lifetimes, ISC/RISC rates (Supplementary Equations 3 and 4 and Supplementary Table 4), T1 excited state energy, HOMO–LUMO difference and spin distribution at T1 (Fig. 3e and Supplementary Fig. 37). The molecular structure of the PS was unambiguously confirmed by single-crystal X-ray analysis (Supplementary Fig. 48). Quite surprisingly, across all the performed reactions, we never observed the formation of the reversed product isomer 5, in which the sulfonyl group is attached to the C3 position (see below).Fig. 3: Mechanistic investigations.a, Proposed reaction scenario suggesting the sensitization of 2, resulting in the generation of sulfonyl and iminyl radicals 10 and 11, respectively. Addition of 10 to 1 produces 12, which undergoes recombination with 11 leading to 3. b, Stern–Volmer plot shows 2 as the quencher. kSV is k of Stern–Volmer and kq is k of quenching. c, Left: the resulting transient absorption LFP spectrum of PS 9 in the presence of 2. The region between 420 nm and 520 nm (within the black circle) suggests that the sensitization of 2 by the triplet PS 9 leads to the radicals 10 + 11 because 10 is expected to absorb at this region. ΔOD, Δ optical density. Middle: flash photolysis of a solution of 2 in the presence of 1, leading to a prompt transient spectrum (time delay, 50 ns) with a maximum at 410 nm (orange trace) which can be assigned to the sulfonyl radical 10 (see left panel for comparison). The dark red trace with a maximum at 435 nm and a tail at longer wavelength (time delay, 750 ns) can be assigned to the radical intermediate 12 obtained by reaction of 1 with radical 10. mΔOD, milli Δ optical density. Right: the dependence of the transient signal at 450 nm on the concentration of 1. d, EPR studies using PBN as spin trap. The blue trace shows the experimental results conducted in the absence of 1; the red trace is the model reaction. e, Insights from DFT calculations. Unless otherwise stated, all values are reported in kcal mol−1.Mechanistic investigationsWe next investigated the mechanism of this RSR process (Fig. 3a). We initially assessed the feasibility of an energy transfer (EnT) step between the excited PS 9 and the sulfonyl imine 2. Time-correlated single-photon counting experiments demonstrated that the singlet excited state of PS 9 is reactive towards neither the sulfonyl imine 2 nor ABB 1. On the other hand, time-resolved luminescence experiments performed by laser flash photolysis (LFP) confirm that the triplet excited state of the PS is efficiently quenched by the sulfonyl imine 2 (bimolecular rate constant k = 2.6 × 108 M−1 s−1; Fig. 3b), whereas quenching by the ABB 1 is substantially less effective (k = 3.7 × 106 M−1 s−1; Supplementary Figs. 26–28). Transient absorption spectroscopic studies by LFP show that photoreaction of the PS 9 triplet with 2 leads to the formation of a subtle absorption signal in the 420–520 nm region (Fig. 3c, left panel), probably associated with the rapid generation of radicals 10 and 11 upon sensitization. However, its decay is faster than its formation, thus preventing full accumulation of the transient and its reliable attribution. We then focused on the homolytic cleavage of the nitrogen–sulfur bond and the reaction of the proposed radical intermediates with the ABB 1 to deliver the final N-tosyl azetidine 3. To gain insight into these steps, we performed LFP analysis upon direct excitation of 2 in the presence of the ABB 1. Under these conditions, we observed the prompt development of a transient absorption centred at 410 nm (Fig. 3c, middle panel; time delay, 50 ns), associated with the tosyl radical 10 (refs. 43,44). In the absence of other reagents, this species decays to the baseline within a few microseconds due to radical recombination (Fig. 3c, right panel). In the presence of 1, a new transient species subsequently forms, characterized by an absorption maximum at 435 nm and a pronounced tail at longer wavelengths. These transient changes can be reasonably assigned to the formation of the radical intermediate 12 (Fig. 3c, middle panel; time delay, 750 nm). Consistent with this attribution, the associated positive absorption, measured at 450 nm (Fig. 3c, right panel) increases with the concentration of the ABB 1, as expected based upon the bimolecular nature of the radical reaction between 10 and 1. The subsequent decay, occurring within a few microseconds, can be mainly assigned to the reaction of 12 with radical 11 and concurrent formation of the final product 3 (for further details, see Supplementary Figs. 33–35).To further corroborate the mechanistic proposal, while gathering more information about the key intermediate 12, we performed electron paramagnetic resonance (EPR) experiments in the presence of the spin trapping reagent α-phenyl-tert-butylnitrone (PBN) and the PS 9. We first explored the homolytic cleavage of 2 in the absence of the ABB 1. Under these conditions, we detected mixed radicals, clearly identifying two main species (Fig. 3d, blue trace; for details, see Supplementary Fig. 29). The parameters of the EPR spectrum are consistent with the trapping of nitrogen- and sulfur-centred radicals in equal proportion, suggesting that these species are the sulfonyl and iminyl radicals 10 and 11. When performing the spin-trapping experiment under the reaction conditions (in the presence of 1), we obtained a more intense and completely different spectrum (Fig. 3d, red trace). This spectrum suggests that multiple radical species are trapped, but the major one (about 70%) is compatible with the carbon-centred radical 12. Notably, the reaction with PBN in the absence of 1 and 2 shows no radical formation, confirming that these reagents are the only source of the radical species in the reaction mixture.Finally, we performed DFT calculations at the (U)M062x/def2TZVP level of theory, including a solvent model (IEFPCM, with toluene as the solvent) to evaluate the reaction’s energetic profile and rationalize the observed reaction outcome. To optimize computational resources, a sulfonyl imine featuring a mesyl group instead of the bulkier tosyl group was selected (2a; Fig. 3e). The S0 and T1 states of 2a and PS 9 have been computed. These calculations enabled us to elucidate the EnT process from the triplet state of PS 9 to the ground state of 2a, as outlined in the reaction scheme in Fig. 3e: 3(PS)∗ + 2a → PS + 3(2a)*. Remarkably, a Gibbs free-energy change (ΔG) of −9.2 kcal mol−1 (Fig. 3e, left) has been determined, underscoring the pronounced exergonic nature of the EnT process within the system. An examination of the spin-density distribution map of the relaxed triplet state of 9 (to optimize computational resources, the calculations of the ΔG of the energy transfer process were performed without taking into account the solvent model), combined with the observations acquired from the time-correlated single-photon counting experiments, which dictate that the triplet excited state is exclusively quenched, may diagnose that the EnT mechanism initiates via a collision complex involving the triazine moiety of PS 9 and the sulfonyl imine 2a (Supplementary Figs. 37 and 38). Next, the generation of the sulfonyl radical 10a and iminyl radical 11, resulting from the homolytic cleavage of the sulfonyl imine’s nitrogen–sulfur bond is determined to be exergonic, with a ΔG = −10.3 kcal mol−1. This step is accompanied by a relatively modest activation barrier (ΔG‡ = 8.1 kcal mol−1). Hence, we computed the insertion mechanism of the sulfonyl radical 10a at the C3 of the ABB 1, followed by radical–radical combination with the iminyl radical 11, ultimately yielding the final product 3a (Fig. 3e, right). Notably, the alternative structural isomer 5a, which remains unobserved experimentally, is theoretically computed to exhibit an energy of 21.9 kcal mol⁻1 higher than 3a. Thus, the absence of isomer 5a is more plausibly attributed to kinetic rather than thermodynamic factors. Indeed, our calculations support this scenario, revealing an energy barrier (ΔG‡) of 17.0 kcal mol−1 (TS2) between intermediate 12 (analogous to the benzyl radical) and the initial reactants 1 and 10a, while a substantially higher activation energy (ΔG‡ = 55.1 kcal mol−1) is determined between intermediates I2 and I3 (nitrogen-centred radical).Generality of the methodHaving shed light on the reaction mechanism, we next evaluated the generality of this RSR photocatalytic process (Fig. 4). Various types of sulfonyl imines were competent substrates, including ketimine and aldimine precursors with electron-donating and electron-withdrawing groups, affording the azetidine products (3, 14–19) in up to 79% yield. Variations at the sulfonyl group were also tolerated well, and the corresponding products 18–27 were isolated in yields ranging from 47% to 80%. We next turned our attention to the ABB precursor. Here, diverse aryl and carbonyl groups were evaluated, delivering the corresponding azetidines (28–30) with yields of up to 84%. Additionally, 2-substituted ABBs furnished the products 31–34 in up to 82% yield, and 2.3:1 d.r. We obtained crystalline products for both diastereoisomers 33 and 34. By performing single-crystal X-ray analysis, we unambiguously determined their relative configurations. Finally, we applied the developed method to more systems with biologically relevant scaffolds. Remarkably, both the celecoxib-containing sulfonyl imine 35 and the naproxen-containing ABB 37 reacted smoothly, affording the corresponding azetidine products 36 and 38 in 71% and 66% yield, respectively. Notably, the mild reaction conditions allowed the preservation of the optical purity of the starting chiral naproxen derivative 37.Fig. 4: Generality of the developed method.The reported yields refer to the average of isolated product obtained from three independent reactions. Unless otherwise noted, the reactions are performed at 0.05 mmol scale. CSD, Cambridge Structural Database (CSD 2305658, CSD 2305660, CSD 2305656, CSD 2305657, CSD 2305659); PMP, p-OMePh. aReaction performed in ethyl acetate.Having proved the generality of the process for accessing C3–N-type azetidines via the activity of the key radical intermediate 12 (Fig. 3), we wondered if it would be possible to intercept this transient species with diverse single occupied molecular orbital (SOMO)philes. Considering the lower efficiency of aldimines in comparison to ketimines as a source of iminyl radicals45, we leveraged this feature by designing a three-component reaction manifold in which the ABB 1 and aldimines 39–41 react in the presence of the supersilane 42 (Fig. 5a, left). Remarkably, the reaction proceeded smoothly, affording 44–46 in up to 91% yield. Encouraged by these results, and given the importance of the SCF3 moiety for pharmaceuticals46,47, we tested phthalimide-SCF3 43. The reaction led to the corresponding thiotrifluoromethylated azetidines 47–50 in up to 71% yield, including two different celecoxib derivatives. This demonstrates that the designed methodology can easily unlock alternative types of reactivity while providing access to families of densely functionalized azetidines without additional optimization processes (for limitations of the protocol, see Supplementary Fig. 39).Fig. 5: Utility of the developed synthetic method to access diversified azetidine scaffolds.a, Extension of the method to a three-component reaction with diverse SOMOphiles. b, Release and manipulations of the key amino functionalities of the products.Synthetic utilityFinally, we performed various synthetic manipulations to reveal the key amino functionalities embedded in the azetidine products. The benzylated intermediate 51 can be accessed by reduction of 16 or reductive amination of 3. In both cases, there are very high yields (up to 95%) without the need for column chromatography. Hydrolysis of 3, 3a and 27 allowed the isolation of the hydrochloride salts 52, 53 and 54, respectively. Neutralization of 52 led to amino sulfonyl azetidine 55 without the need for purification steps. This is a key synthetic intermediate that can be easily transformed into tertiary amines, bearing a biologically relevant pyrrolidine such as 56. The same azetidine succeeded in tosyl group cleavage, giving rise to the two-vector 3-aminoazetidine 57, while providing access to a structural variant of the kinase modulators 58 (ref. 48). Lastly, by targeting the synthesis of amino acid 59 through a RuO4-mediated oxidative fission of the phenyl ring49,50, we obtained the unexpected yet attractive spiroazetidine scaffold 60 in 52% yield.
