Stereoselective hydrogen atom transfer to acyclic radicals: a switch enabling diastereodivergent borylative radical cascades

Radical cascade reactions are powerful tools to construct structurally complex molecules. However, the stereochemical control of acyclic radical intermediates remains a persistent challenge, due to the low differentiation between the two faces of these species. This hurdle further makes stereodivergent synthesis rather more difficult to be accomplished, in particular for intermediates resulted from complex cascades. Here we report an efficient strategy for stereoselective hydrogen atom transfer (HAT) to acyclic carbon radicals, which are generated via N-heterocyclic carbene (NHC)-boryl radicals triggered addition-translocation-cyclization cascades. A synergistic control by the NHC subunit and a thiol catalyst has proved effective for one facial HAT, while a ZnI2-chelation protocol allows for the preferential reaction to the opposite face. Such a stereoselectivity switch enables diastereodivergent construction of heterocycles tethering a boron-substituted stereocenter. Mechanistic studies suggest two complementary ways to tune HAT diastereoselectivity. The stereospecific conversions of the resulting boron-handled products to diverse functionalized molecules are demonstrated. Stereochemical control of acyclic carbon radical intermediates remains a challenge, due to the difficulty in differentiation between the two faces of the planar species. Here the authors show diastereodivergent hydrogen atom transfer to acyclic carbon radicals via control of N-heterocyclic carbene–borane reagents with either a thiol catalyst or a Lewis acid.

C onstruction of complex organic molecules in step-and atom-economic manners with high levels of stereoselectivity control is a preeminent goal in synthetic chemistry 1,2 . Radical cascade reactions represent one of the most efficient strategies to assemble densely functionalized architectures from simple starting materials in one step [3][4][5] . However, the stereochemical control in these reactions remains an inherently difficult issue when acyclic carbon stereocenters are created through complex cascade processes, due to the persistent challenge associated with the low differentiation between the two faces of acyclic alkyl radicals 6,7 . This hurdle also makes the stereodivergent synthesis, which is highly demanded in medicinal studies 8 , rather difficult to be achieved.
Hydrogen atom transfer (HAT) reactions often serve as a key step to terminate a radical cascade sequence through the reduction of the resulting radical intermediate [9][10][11] . However, HAT to acyclic radical intermediates with high levels of diastereoselectivity is a subject of persistent challenge [12][13][14][15] . In this context, some stereoselective HAT approaches have been reported, which include, for example, introducing different steric or electronic interactions in acyclic substrates (Fig. 1a). In addition, chelation by the use of Lewis acids has been applied so that the corresponding acyclic systems become rigid and high diastereoselectivity is consequently possible to be realized 16,17 . Although these approaches have proven effective in certain cases, some limitations and drawbacks still remain: (i) most methods are restricted to specific substrates with the preinstallation of different functional groups onto specific positions for steric, electronic, or complexation purposes, but such tactics are hardly applied in the reduction of radicals derived from complex radical cascades; (ii) stereocontrol is highly temperature-dependent and relatively lower temperature (usually below room temperature) is often necessary to ensure high diastereoselectivity, but this becomes disfavored for radical cascades that need high temperature 18,19 ; (iii) diastereodivergent HAT reactions have only been reported in limited cases 20 , and a general strategy for diastereodivergency is still lacking. Therefore, developing new strategies that can render the reduction of acyclic radical intermediates to be stereoselectively distinguishable and switchable is highly desirable for advancing radical chemistry in synthesis.
N-heterocyclic carbene (NHC)-borane complexes have recently been developed as an important class of hydrogen atom donor, as pioneered by Fensterbank, Lacote, Malacria, and Curran 21 . These complexes feature easy accessibility, low toxicity, and broad applicability. Moreover, the group of Curran and Taniguchi 22-25 , our group [26][27][28][29] , and others [30][31][32][33] disclosed that these complexes are viable for the synthesis of organoboron molecules via radical cascade reactions, wherein both the boryl moiety and hydrogen atom are transferred to products. A polarity reversal catalyst 34,35 , such as thiol compound, is sometimes required to maintain efficient chain propagation. It is known that NHC molecules with various substitutions and stereochemistry are either commercially available or can be synthesized readily. Therefore, the NHC subunit would be considered as an important and tunable element for stereocontrol, where various potential factors, including steric or electronic effect, weak interactions between catalysts or substrates, and complexation with Lewis acid, would be used to modulate HAT stereoselectivity. Meanwhile, the NHC-boryl moiety is installed during the reaction process, thereby avoiding the prior installation of specific stereoselectivity-determining functionalities into substrates, which is desired in complex radical cascade reactions.
With this in mind, we hypothesized an NHC-boryl radical triggered radical addition-translocation-cyclization (RATC) cascade 36-39 that could generate a boron-substituted structurally complex acyclic radical intermediate. As shown in Fig. 1b, the reaction starts with the α-addition of an NHC-boryl radical to arylpropiolates 1 28 , followed by 1,5-hydrogen atom shift and intramolecular 5-exo cyclization, yielding an indoline ring with trans-stereochemistry of cyclic substituents. The following HAT to the resulting acyclic radical III with high stereocontrol is perceived to be a major challenge for achieving the goal of stereoselective and stereodivergent synthesis of boron-tethered indolines.
Here we report our realization of diastereoselective HAT to this acyclic radical, which is accomplished by a synergistic control of NHC-BH 3 reagents together with either a thiol catalyst or a Lewis acid. Notably, these two modes operate in complementary ways that offer a stereochemistry switch to enable diastereodivergent synthesis of borylated products 3 and 4. Mechanistic investigations of the detailed mechanisms are carried out to understand how the present HAT reactions take place and the origins of stereochemistry. Significantly, the formed boron-tethered indoline derivatives would be of great interest in synthetic and medicinal chemistry, given the fact that indoline is a core motif in numerous biologically relevant molecules 40,41 . Moreover, the boron moiety can be used in stereospecific diversifications [42][43][44] , providing direct and economical accesses to various functionalities through late-stage functionalizations, thus circumventing the traditional lengthy de novo synthesis.

Results
NHC-BH 3 /thiol catalyst-controlled stereoselective HAT to acyclic radicals generated via borylative RATC. We commenced our study by investigating the reaction of 1a and NHC-BH 3 2a in the presence of di-tert-butyl hyponitrite (TBHN) as the radical initiator. The reaction took place as expected, affording diastereomers 3a-Me and 4a-Me in 52% combined yield and a 67:33 diastereomeric ratio (dr) ( Table 1, entry 1). Guindon has found that the exocyclic effect plays a vital role in the stereoselective reduction of carbon-centered radicals locating exo to a heterocycle 14 . However, the low dr value obtained in our reaction suggested that such an effect was not significant for the reduction of radical III.
Next, we sought to examine various NHC-BH 3 complexes and thiol catalysts to improve the diastereoselectivity. The use of 2b instead of 2a led to a lower conversion, but with a slightly increased dr (entry 2). Interestingly, when a thiol compound (thiol A, B, or C) was employed as a polarity reversal catalyst 34,35 , both the product yield (52-79%) and stereoselectivity (dr, 90:10-92:8) were enhanced (entries [3][4][5]. Furthermore, using orthosubstituted thiophenols increased the dr (93:7 to 94:6) while maintaining high product yields (entries 6-8). However, 2,4,6triisopropyl substituted thiol G resulted in a much lower dr (88:12) although good product yield could be retained (entry 9). Eventually, thiol H was found to be optimal, furnishing 3a-i-Pr in excellent yield and dr (entry 10). Using thiol H as the catalyst, the reaction with 2a resulted in a decrease of dr to 88:12 (entry 11), implying that the isopropyl group of NHC-BH 3 2b is also crucial to maintain high diastereoselectivity. The reaction of 2c bearing a 2,6-diisopropylphenyl motif led to nearly no reaction (entry 12), presumably due to the increased difficulty in the addition step. These findings support a fact that the stereoselectivity of HAT to acyclic radical III is cooperatively controlled by the substituents of both NHC-BH 3 and the thiol catalyst. When the radical initiator was changed to 2,2-azobisisoheptonitrile (ABVN), 3a-i-Pr was formed in 93% yield and 95:5 dr (entry 13). However, the employment of azobisisobutyronitrile (AIBN) and azobiscyclohexanecarbonitrile (ACCN), which require elevated temperature (80-95°C) for initiation, caused decreased diastereoselectivity (entries [14][15]. The control experiment showed that no reaction occurred in the absence of a radical initiator (entry 16), verifying a radical reaction mechanism of this transformation.
The generality of this NHC-BH 3 /thiol catalyst-controlled diastereoselective synthesis of product 3 was examined (Fig. 2). In most cases, high levels of diastereoselectivity (dr >92:8) could be obtained. An array of functional groups on the 2-aryl ring (for 3c-3e) and indoline framework (for 3h and 3i) were incorporated. Both naphthyl (for 3f) and thiophenyl (for 3g) motifs were compatible. The reaction of 1j delivered a pyrrolo[2,3-b]pyridine framework (for 3j), albeit with a slightly decreased diastereoselectivity. A series of alkyl groups could be used as the R 2 substituent (for 1k-1m) on the nitrogen atom, furnishing desired products in both excellent yields and diastereoselectivity. When the R 2 substituent was replaced with a phenyl ring, the reaction could produce the expected product 3n in 82% yield, but with a moderate stereoselectivity (86:14 dr). However, no reaction occurred for the hydrogen substituted one (for 1o). The present protocol allowed for the construction of a tetracyclic framework (for 3p) with no detrimental effect on diastereoselectivity. The borylative cascade reactions of 1q and 1r, however, became less efficient (<25% yield of cyclized products) and hydroboration of the alkyne moiety 23,45,46 was observed as a competing reaction pathway. This is presumably due to the disfavored 1,5-hydrogen atom transfer of the corresponding alkenyl radical I that makes the reduction by the thiol catalyst possible. To our delight, the desired cascade reactions could be promoted by increasing the reaction temperature to 95°C with the use of ACCN as the radical initiator and using 2a as the boryl radical precursor, furnishing products 3q and 3r in 83% yield (86:14 dr) and 57% yield (67:33 dr), respectively. In these transformations, 2a itself also acted as the hydrogen atom donor to maintain the radical chain process 47 . It was also found that the addition of a thiol catalyst was not effective at improving the dr value in this case (see Table S4 in the Supplementary Information). Likewise, the standard reaction protocol of 1s led to no reaction, while the modified procedure provided the borylated product in 65% yield, albeit with low diastereoselectivity.
NHC-BH 3 /Lewis acid-controlled stereoselective HAT to acyclic radicals generated via borylative RATC. Next, we questioned whether it is possible to reverse the stereoselectivity of the HAT process by changing the reaction conditions, leading to the preferential formation of diastereomers 4. Considering that acyclic radical III contains some coordinative functionalities, such as methoxycarbonyl group, NHC subunit, and amino group, a range of Lewis acids have been examined. Eventually, it was found that the addition of ZnI 2 as a chelating reagent and the employment of 2d as the boryl radical precursor could afford 4aa in good yield and excellent diastereoselectivity (3aa:4aa = 6:94, Table 2, entry 1). No thiol catalyst was required in this reaction, implying that acyclic radical III should abstract a hydrogen atom from 2d to propagate the radical chain process 47 . In this case, the formed radical intermediate III likely became more electrophilic by coordination of ZnI 2 to the carbonyl group, thus promoting the HAT process by more polarity matching. Reducing the amount of ZnI 2 resulted in a dramatic decrease of dr (entries 2 and 3), and the reaction without ZnI 2 even gave 3aa as the major product (entry 4). Other Lewis acids such as MgCl 2 , MgI 2 , CuI, ZnCl 2 , Me 3 Al, and MgBr 2 ·OEt 2 48 were found to be ineffective (entries 5-10). When 2a was used as the radical precursors instead of 2d, the dr value was retained, although the yield was decreased to 37% (entry 11). The reaction of 2b led to only a trace amount of product (entry 12). These results indicate that the ZnI 2 coordination with the triazole moiety of 2d may be beneficial to maintain the radical chain propagation, but the contribution in controlling the diastereoselectivity is not significant.
This NHC-BH 3 /ZnI 2 chelation-controlled diastereoselective synthesis showed broad substrate scope and functional group tolerance (Fig. 3). In most reactions, both moderate to good yields and high levels of dr (8:92~6:94) could be obtained (4aa, 4ha, 4ia, 4ta-4xa). However, when the R 2 substituent was changed to other alkyl groups (for 1b and 1l) or phenyl group (1n), the products were formed with low stereochemical outcomes. This implied that the benzyl group as R 2 is crucial. In addition, replacing the terminal methoxycarbonyl group with a non-coordinating CF 3 moiety resulted in a precipitous drop in diastereoselectivity (3sa:4sa = 45:55), hinting the participation of the carbonyl moiety in the diastereo-determining step. Moreover, the presence of a strong coordinative pyridine ring (for 1j) also interrupted the chelation, resulting in almost no diastereoselectivity.
Synthetic applications. The synthetic utility of this diastereodivergent RATC process was demonstrated (Fig. 4). Diastereomeric α-boryl carbonyls 3i and 4ia were converted to a series of useful building blocks based on the versatile chemistry of both the boryl and carbonyl moieties. For example, reduction of the ester group of 3i and 4ia followed by oxidation afforded diastereomeric 1,2-diols 5a and 5aa in good yields. Instead of oxidation, the treatment with N-chlorosuccinimide (NCS) in the presence of pinacol and further protection of the hydroxy group provided useful pinacol boronic esters 6a and 6aa, respectively. The subsequent coupling reaction with furan using Aggarwal's arylation protocol 49 delivered the corresponding 7a and 7aa in good efficiency. Moreover, homologation of 6a and 6aa also proceeded, furnishing alkyl boronic esters 8a and 8aa in good yields. It is worth mentioning that the starting materials 3i and 4ia were both used as a single diastereomer and the stereochemistry was retained in all these transformations, thus the resulting products were obtained with exclusive diastereoselectivity.

Discussion
Computational studies of diastereoselective HAT controlled by NHC-BH 3 /thiol catalyst. To elucidate the diastereoselectivity of this HAT process, density functional theory (DFT) calculations   Fig. 4). The results reveal that~7 kcal/mol of energy barrier is required for the free rotation of C 1 -C 2 bond in Int-III-A. This energy is lower than the activation Gibbs energies of HATs, which are 8.3 and 10.6 kcal/mol, respectively. Therefore, the stereochemistry is determined by the relative HAT transition states, based on the Curtin-Hammett principle 50 . Then hydrogen atom transfer to Int-III-A was computed by using thiol E for simplification. The transition states leading to products 3a-i-Pr and 4a-i-Pr were located as TS-A-major and TS-A-minor, respectively (Fig. 5). Before reaching these two transition states, two complexes between Int-III-A and thiol E are formed. TS-Amajor is 2.3 kcal/mol more favorable than TS-A-minor in terms of Gibbs free energy, suggesting that 3a-i-Pr is the major product. This is consistent with the experimental observation (Table 1,  entry 7). To find the origin of the stereochemistry preference for TS-A-major over TS-A-minor, distortion/interaction analysis 51-53 (DIA) was performed by separating the transition states into two fragments, Int-III-A and thiol E. The distortion energy for TS-A-major is slightly higher than that of TS-Aminor (18.8 vs. 18.2 kcal/mol), and a stronger interaction energy is observed in TS-A-major (ΔΔE int = −2.0 kcal/mol). These imply the presence of different interaction energies that determines the diastereoselectivity. The interactions between Int-III-A and thiol E in TS-A-major and TS-A-minor were further analyzed by noncovalent interaction 54,55 (NCI) and AIM analysis.
The results indicate that the lower interaction energy in TS-A-major is likely attributed to the CH/π attraction 56,57 between H a on thiol E and the 2-phenyl moiety, which is absent in the competing TS-A-minor. This is further supported by the experimental finding that thiol G lacking such ortho-Csp 2 -H gave inferior diastereoselectivity (Table 1, entry 9).
Experimental and Computational Studies of Diastereoselective HAT Controlled by NHC-BH 3 /Lewis Acid. Experimental and computational studies were then conducted to study how ZnI 2chelation-controlled HAT process takes place. Because Int-III-B is assumed to have a chelation with ZnI 2 similar to that of product 4aa, we used 1 H NMR spectroscopy measurements to probe the possible chelation sites employing 4aa as a model (Fig. 6). Upon addition of ZnI 2 (0-1.5 equiv), the protons locating on the triazole moiety (H a , H b and H c ) are shifted downfield. In addition, the proton of methoxycarbonyl group (H d ) and α-carbonyl proton (H e ) also exhibit a downfield shift. These imply that both the triazole and carbonyl groups are involved in binding with Zn II . It should be noted that protons at α-positions of the nitrogen atom (H g , H i , and H' i ) have negligible changes, which indicates that the coordination of this aryl-substituted nitrogen atom to the zinc center is weak. Importantly, only a single set of 1 H NMR spectrum is observed in all cases, suggesting a fast dynamics of complexation/decomplexation between the free 4aa and Zn II -4aa chelates. A similar behavior of dynamics between free 2d and Zn II -2d was also observed (Supplementary Fig. 3). As such, the following DFT calculations employed Int-III bound to one molecular of ZnI 2 and NHC-BH 3 2d ligated with another one ZnI 2 as the model substrate and hydrogen atom donor, respectively.
To investigate the chelation mode of ZnI 2 with Int-III, a variety of possible complexes with different chelation modes were extensively screened. The results revealed that the most favored chelation mode contains a ZnI 2 -chelated six-membered ring, where ZnI 2 coordinates with both the carbonyl oxygen atom and the B-H bond 58,59 (Int-III-B-1 in Fig. 7a). 1 H NMR spectroscopy of 2a titrated with ZnI 2 showed a downfield shift of B-H signals ( Supplementary Fig. 2), providing an evidence for the interaction of ZnI 2 with the B-H bond. Some other chelation models were also considered and computed, such as the coordination of ZnI 2 with only methoxycarbonyl oxygen atom (Int-III-B-2), or with the triazole nitrogen atom (Int-III-B-3), or with both the carbonyl group and the indoline nitrogen atom (Int -III-B-4). However, all of them are energetically disfavored in comparison with Int-III-B-1. The HAT transition states of the aforementioned four possible intermediates were located, and a preference of Int-III-B-1 was observed ( Supplementary Fig. 5). Moreover, our experimental findings showed no evident difference of dr value between 2a (7:93, Table 2, entry 9) and 2d (6:94, Table 2, entry 1), hinting that the chelation between ZnI 2 and the triazole moiety is most likely not the leading factor of determining HAT diastereoselectivity. However, it is worth mentioning that the use of 2d (Table 2, entry 1) gives a much better product yield than that of 2a (Table 2, entry 11), meaning that the dynamic complexation between 2d and ZnI 2 was probably helpful to accelerate the cascade process, while the detailed mechanism is still unclear and further mechanistic studies are currently under progress in our laboratory. In addition, the optimization studies (Table 2) showed that other Lewis acids, such as MgCl 2 , MgI 2 , CuI, ZnCl 2 , Me 3 Al, and MgBr 2 ·OEt 2 , are ineffective in controlling the stereoselectivity, suggesting that both the zinc center and the iodide anions are crucial to form the favored HAT transition state that can lead to high levels of diatereoselectivity. The switch of the anion to chloride might result in a more flexible complexation and the sixmembered ring chelation mode was likely not formed, and then inducing no effect of stereocontrol (Table 2, entry 8).
As depicted in Fig. 7b, calculations suggest HAT from 2d-ZnI 2 complex to Int-III-B-1 prefers to occur via TS-B-major-1 over TS-B-minor-1 (ΔΔG ‡ = 3.3 kcal/mol), which correlates well with the selectivity trend observed in the experiment. There are complexes formed between Int-III-B-1 and ZnI 2 -ligated 2d, we did not locate them because analyzing the transition states of TS-B-major-1 and TS-B-minor-1 can give us answers to stereochemistry. These two transition states TS-B-major-1 and TS-Bminor-1 were then subjected to DIA. Based on the DIA results, the preference of TS-B-major-1 could be mainly ascribed to the lower distortion energy of Int-III-B-1 (8.5 kcal/mol) than that in TS-B-minor-1 (11.7 kcal/mol). Careful inspection of the transition state structures reveals the specific Zn-chelated sixmembered ring plays a critical role in controlling the diastereoselectivity. In TS-B-minor-1, the six-membered ring has to undergo substantial torsion to orient towards N-Bn group, thus the steric congestion between these two components is likely the cause of high distortion energy in Int-III-B-1. This hypothesis is in line with the poor diastereoselectivities observed in the reactions of 1b, 1l, 1n, where N-alkyl and N-phenyl groups are employed instead of N-benzyl group (Fig. 3). Subsequent DFT calculation and DIA on HAT step of 1b provided computational support for this hypothesis, finding that replacing N-Bn with N-Me results in a dramatic decrease of the distortion energy ( Supplementary Fig. 6). Computational studies of the HAT from free NHC-BH 3 2d to Int-III-B were also conducted, and the results show that the use of free 2d or 2d-ZnI 2 complex as the hydrogen atom donor has a negligible effect on the diastereoselectivity ( Supplementary Fig. 7).
In summary, we have developed an NHC-boryl radical triggered radical cascade to assemble heterocyclic molecules anchoring a boron-substituted acyclic stereocenter. The stereochemistry of HAT to acyclic radical intermediates is determined by NHC-BH 3 /thiol catalyst and NHC-BH 3 /ZnI 2 -chelation systems. Experimental and computational studies suggest that these two modes operate in complementary ways. A cooperative control by the NHC unit and a thiol catalyst favor the formation of one diastereomer, while a B-H bond-involved six-membered ring chelation with ZnI 2 is beneficial for the formation of the other one. Such a stereoselectivity switch enables the divergent synthesis of diastereomeric borylated products. We expect that   the present stereoselectivity-switchable HAT strategy and the resulting diastereomeric borylated products would be important for synthetic and medicinal applications.

Methods
General procedure for NHC-BH 3 /thiol catalyst-controlled stereoselective synthesis of borylated molecules. Computational methods. Calculations were carried out with Gaussian 16 software packages. The (U)B3LYP functional was used with an empirical dispersion correction (Grimme-D3(0)) for all the calculations. The geometries of all the stationary points were optimized in the gas phase with the SDD basis set (Stuttgart/ Dresden ECP) for Zn and I, and the 6-31G(d,p) basis set for the other atoms. In the computational studies of HAT controlled by NHC-BH 3 /thiol catalyst, the internal six d-type orbitals were used for all elements in geometry optimization part. In the part of diastereoselective HAT controlled by NHC-BH 3 /Lewis acid, the keyword "5D" was used to specify that five d-type orbitals were used for all elements in the calculations. Vibrational frequency analysis was calculated at the same level of theory to validate each structure as either a minimum or a transition state. For each transition state, the intrinsic reaction coordinate (IRC) analysis was conducted to ensure that it connects the right reactant and product. To obtain more accurate energies, single-point energies were calculated with a mixed basis set of SDD for Zn and I, 6-311+G(d,p) for all the other atoms. The distortion/interaction analysis was performed with gas-phase single point energies, while all the other energies reported in the main manuscript were obtained from solution-phase single point energies with SMD solvation model (solvent = toluene). Noncovalent interaction (NCI) analysis was performed with Multiwfn by using the IGM method. The results were visualized by VMD software. 3D structures were generated by CYLview. See Supplementary refs. 22-37.

Data availability
The X-ray crystallographic coordinates for 3a-Me