Formylation or methylation: what determines the chemoselectivity of the reaction of amine, CO2, and hydrosilane catalyzed by 1,3,2-diazaphospholene?

A DFT study demonstrates that methylation and formylation of amines with CO2 and hydrosilane, catalyzed by 1,3,2-diazaphospholene, are two competitive reaction channels.


Introduction
The rising concentration of carbon dioxide (CO 2 ) in the atmosphere is one of the key factors for global warming, leading to great efforts to develop effective catalytic routes that convert CO 2 to value-added chemicals. [1][2][3] Formylation and methylation of amines with CO 2 are promising synthetic strategies to use CO 2 as a C1 carbon source. 4 In 1998, Vaska and coworkers developed the rst Pt-catalyzed formylation of amine with CO 2 and H 2 . 5 This study has encouraged further developments using other transition metal catalysts 6 or metal-free catalysts. 7 In 2012, Cantat and coworkers achieved the rst organocatalytic formylation of amines with CO 2 and hydrosilane, catalyzed by triazabicyclodecene (TBD). 8 Since then, more similar transformations were reported. 9 In 2013, Beller and coworkers reported the rst methylation of amine with CO 2 and hydrosilane, catalyzed by a ruthenium complex. 10 More similar transformations were later developed. 11 It is worth mentioning that Cantat et al. also developed metal-free methylation of CO 2 with amines. 12 Furthermore, transition metal catalyzed methylation of amines with CO 2 and H 2 has also been accomplished by several groups. 13 Previously, we studied the catalytic mechanisms of CO 2 reduction to methanol 14 and methane. 15 In this context, we were intrigued by the catalytic reactions developed by Kinjo and coworkers. 16 They used 1,3,H) to catalyze the formylation of amines ([N]H) with CO 2 and hydrosilane (Ph 2 SiH 2 ¼ [Si]H 2 ) (e.g. eqn (1) in Scheme 1). Interestingly, two amines (2a and 3a) were found to be exceptional, affording N-methylated amines (2c and 3c). They attributed 2c and 3c to the further reductions of 2b and 3b, respectively, complying with the general consideration that methylation takes place sequentially through formylation, giving formamide, followed by the reduction of formamide. 10,17 Nevertheless, we conceived that this mechanism may not be true in the present system. First, due to the smaller steric effect of 1b compared to 2b, 1b should be reduced more easily than 2b, but eqn (1) affords 1b rather than 1c. Second, if the methylation mechanism is true, N-methylated amines could be at least detected in eqn (1). In addition, Cantat et al. 18 showed that in the TBD-catalyzed aminal synthesis from amine, CO 2 , and hydrosilane, which is somewhat similar to methylation, the formation of an aminal product takes place aer forming [Si] OCH 2 O[Si] via two sequential 2-electron reductions of CO 2 with hydrosilane and the HC(]O)O[Si] intermediate resulting from the rst 2-electron reduction of CO 2 with hydrosilane does not react with amine to give formamide. Thus, the formation of aminal does not pass formamide as an intermediate. Given these analyses, we carried out a DFT mechanistic study to deeply understand the catalytic system, in combination with experimental verications. To our knowledge, there has been no systematic study on the mechanisms of formylation and methylation of amines with CO 2 , although Cantat and coworkers reported some computational results in their experimental study. 19 Scheme 1B (3)).

Computational details
Experimentally, the reactions were carried out in a polar solvent (acetonitrile, 3 ¼ 35.7). Considering the possible signicant effects of the strong polar solvent, all geometries were optimized and characterized as minima (no imaginary frequency) or transition states (TSs, having one unique imaginary frequency) at the M06-2X 20 /6-31G(d,p) level with the solvation effect of acetonitrile simulated by the SMD 21 solvent model. At the M06-2X/6-31G(d,p) geometries, the energies were further rened by M06-2X/6-311++G(d,p) single-point energy calculations with the solvent effect accounted for by the SMD solvent model. All DFT calculations adopted ultrane integration grids (Int ¼ ultrane) to ensure stable numerical integrations. The M06-2X/6-31G(d,p) frequencies were used for thermal and entropic corrections at 298.15 K and 1 atm. It should be emphasized that such a correction approach is based on the ideal gas phase model, which inevitably overestimates entropy contributions to free energies for reactions in solvent, in particular for reactions involving a multicomponent change, because they ignore the suppressing effect of solvent on the rotational and transitional freedoms of substrates. The entropy overestimation of the approach was also demonstrated experimentally. 22,23 While no standard quantum mechanics-based method is available to accurately calculate entropy in solution, approximate methods were proposed. According to the proposal of Martin et al. 24 we previously applied a correction of (n À m) Â 4.3 kcal mol À1 for a process from m-to n-components and found that such corrected free energies were more reasonable than enthalpies and uncorrected free energies, 15,25 although the protocol is by no means accurate. Other correction factors (e.g. 1.9, 26 2.6, 3a,27 and 5.4 kcal mol À1 (ref. 28)) were adopted in the literature depending on the approximate approaches. As will be seen, our studied reactions involve multicomponent changes. As a conservative consideration, we applied a correction factor of 1.9 kcal mol À1 in this study. The corrected free energies are discussed and the uncorrected ones are given in the parentheses for references, unless otherwise specied. Note that using a correction factor of 4.3 kcal mol À1 does not alter our conclusions except for the numerical values. Natural bond orbital (NBO) analyses were performed at the M06-2X/6-311++G(d,p) level to assign partial atomic charges (Q). 29 All calculations were carried out using Gaussian 09. 30

Results and discussion
In this study, we use eqn (1) as a representative to compute the formylation mechanism of amine 1a (Section 3.1). In Section 3.2, using eqn (2), we investigate the methylation mechanism of amine (2a). Aer characterizing the mechanisms of formylation and methylation, we discuss the origins of chemoselectivity and experimentally verify our proposed mechanism in Section 3.3.
Our computed mechanisms involve ionic species, thus we explicitly label the charges of all species when applicable for simplicity of the descriptions.
Hydrophosphination of CO 2 (stage I). Fig. 1 illustrates the mechanism for CO 2 hydrophosphination, along with the key optimized structures. The catalyst [NHP]H is a hydride with P and H bearing 0.921 and À0.069e partial charges, respectively.
Conventionally, CO 2 prefers inserting into an E-H bond (e.g. E ¼ B or Ni) via a four-membered TS, forming C-H and E-O bonds concertedly. 14b,15 However, the optimized structure of TS1 targeting for an insertion TS describes a hydrogen abstraction process. Zhu et al. reported a similar TS. 31 The IRC (intrinsic reaction coordinate) calculation toward the product stopped aer 129 steps ( Fig. S1 †), giving a structure (namely, IRCF-129) which can be viewed as an ion pair resulting from CO 2 abstraction of the H dÀ atom of [NHP]H. However, geometric optimization starting from IRCF-129 reached an insertion product [NHP]OCHO (IM1). We attribute the abnormal insertion to the difference between the P d+ -H dÀ bond in [NHP]H and E d+ -H dÀ bond (e.g. B-H or Ni-H); 14b,15 the P center has a lone pair disfavoring P-O bond formation, while the E center features an empty orbital favoring E-O bond formation. IM1 is different from the X-ray structure of the CO 2 hydrophosphination product (IM3) but can convert to the more stable IM3 easily (see Fig. 1). Overall, the insertion crosses a barrier of 16.7 kcal mol À1 and is exergonic by 6.9 kcal mol À1 , indicating the feasibility of the process.
Kinjo et al. observed zwitterionic character of IM3. Consistently, the [NHP] and HCO 2 moieties in IM3 bear charges of 0.658 and À0.658e, respectively. Because of the zwitterionic nature, we conceived that IM3 can dissociate easily in the strong polar acetonitrile solvent, as demonstrated by the small dissociation energy (4.6 kcal mol À1 , see Scheme 2). Thus a microscopic equilibrium is expected in this catalytic system. As will be shown, the free [NHP] + and HCO 2 À ions play catalytic roles to mediate subsequent steps of the transformation. involved in the transformation. 16 Fig. 2   OCHO is energetically feasible with a RDS (rate determining step) barrier of 21.2 kcal mol À1 (TS5) relative to IM3. Yet, we speculated that the stage may proceed via an ionic mechanism because free HCO 2 À is available via the equilibrium (Scheme 2).

Formation of diformyloxysilane[Si](OCHO) 2 (stage II). Experimentally, it has been demonstrated that [Si](OCHO) 2 is
The red pathway in Fig. 2A (4) in Scheme 3). Intuitively, the bond can be formed via the nucleophilic attacks of amine, illustrated by mode A and B in Scheme 3, yet the high barriers, 41.1 (mode A) and 31.6 kcal mol À1 (mode B), rule out the two modes, considering that the reaction could occur under mild conditions (see eqn (1)). We explored other alternatives discussed below.
Thus, we considered whether a HCO 2 À ion can facilitate the C-N bond formation via H-bonding to the N-H bond of 1a (i.e. mode C in Scheme 3), because the bonding of the anionic species can enhance the nucleophilicity of amine 1a. Fig. 3 depicts the mechanism for eqn (4a) under the catalytic effect of HCO 2 À , along with key optimized structures. First, HCO 2 À and 1a form a H-bond complex IM8 À , then the complex attacks [Si](OCHO) 2 via TS9 À , giving IM9 À with a C-N bond formed.
Interestingly, the C-N bond formation shis the N- C-N and Si-O 2 bonds in IM9 À ) and cleavages (i.e. C-O 2 and Si-O 1 bonds in IM10 À ). It is interesting that CO 2 can be activated to an active species to facilitate its transformation. Following the same mechanism in Fig. 3, eqn (4b) takes place, producing another formamide (1b) and silanol [Si](OH) 2 . Without going into detail (see Fig. S5 † for the energy prole of eqn (4b)), we mention that the RDS barrier of eqn (4b) is 27.3 kcal mol À1 , 5.5 kcal mol À1 higher than that of eqn (4a). C-N bond formation facilitated by hydrogen transfer shuttle. The C-N bond formation through mode A and B involves a fourmembered TS featuring hydrogen transfer (see Scheme 3). Thus a protic molecule such as water may act as a hydrogen transfer shuttle (H-shuttle) 32,33 to facilitate the stage. In the present system, the possible H-shuttles could be water (trace water could not be excluded absolutely), N-methylaniline 1a, and silanol (HO[Si]OCHO and [Si](OH) 2 ), which are available when Scheme 3 C-N bond formation stage (eqn (4)) and possible modes to form the bond. the reaction is initiated. Using water as a representative, we characterize the H-shuttle-aided pathway (eqn (4)) through mode A, as illustrated in Fig. 4. Without going into detail, we mention that the water-aided C-N bond formation involves two hydrogen transfer steps, sequentially forming C-N and breaking C-O (CO 2 deoxygenation) bonds, as described by TS12 and TS13 for eqn (4a), respectively.  Fig. S6 †). HCO 2 À is more effective than water but less effective than silanol. For the formation of the C-N bond through mode B (Scheme 3), the water H-shuttle does not help much with only a slightly lower barrier (30.5 kcal mol À1 ), compared to 31.6 kcal mol À1 without the H-shuttle. The most effective H-shuttle, HO [Si] OCHO, in the case of mode A has a barrier of 27.3 kcal mol À1 in the case of mode B, which is much higher than 18.8 kcal mol À1 through mode A. We thus do not expect that other H-shuttles could aid the stage through the mode B mechanism more efficiently than that through mode A and did not pursue the mode further.
Aer characterizing the efficiency of these hydrogen transfer mediators in prompting C-N bond formation, we now discuss how the C-N bond could actually be formed. Both eqn (4a) and (4b) are thermodynamically favorable, being exergonic by 9.9 and 6.9 kcal mol À1 , respectively. We focus on the kinetics of the reactions using eqn (4a) as an example for simplicity.
It was reported that in the absence of [NHP]H and CO 2 , [Si](OCHO) 2 alone could react with 1a to give 1b. As the efficiency of the reaction was not reported, our energetic results show that the reaction is able to take place, because the barrier for eqn (4a), when using water as a H-shuttle, is 26.4 kcal mol À1 , which is somewhat high but in a reasonable range for a reaction to occur. Importantly, when the reaction is initiated to produce silanol, the silanol byproducts can promote the reaction more effectively, with lower barriers (see Table 1). In the presence of [NHP]H and CO 2 , HCO 2 À plays the role of initiating the reaction rather than water, because the RDS barrier of 21.8 kcal mol À1 using HCO 2 À as a catalyst is much lower than 26.4 kcal mol À1 using a water H-shuttle as a promoter. As the reaction proceeds, more and more silanols (HO[Si]OCHO or [Si](OH) 2 ) are produced, thus, silanols take the role of HCO 2 À to promote C-N bond formation.

Mechanism for 2a methylation (eqn (2))
Kinjo et al. 16 have applied an [NHP]H catalyst to perform formylations of a range of primary and secondary amines. Intriguingly, 2,2,4,4-tetramethylpiperidine (2a) and diisopropylamine (3a) were found to afford N-methylated amines, 2c (eqn (2)) and 3c (eqn (3)), respectively. In general, formamide (the formylation product) was considered to be the intermediate for the methylation of amine with CO 2 . 10,17 The mechanism was also adopted to elucidate the methylation products (2c and 3c). Nevertheless, we reasoned that this could not be true in the present catalytic system (supra infra). Using eqn (2) as an example, we investigate the methylation mechanism. The C-N bond in formylation is formed via the nucleophilic attack of amine (1a) to [Si](OCHO) 2 (see TS9 À in Fig. 3 (IM15). The in situ generated CH 2 O then attacks 2a electrophilically, forming a C-N bond and meanwhile transferring the (N-)H atom of amine to the carbonyl group of the formaldehyde moiety via TS18, resulting in IM16. The barrier for the process is 26.8 kcal mol À1 (TS18 relative to IM15), which is somewhat high but can be greatly lowered when a H-shuttle is used. For example, a water H-shuttle can lower the barrier to 14.1 kcal mol À1 (TS18 0 ).   Fig. S8. b Complete pathway is given in Fig. S9. c Complete pathway is given in Fig. S10 According to the methylation pathway (Fig. 5A), the reaction seems to consume the catalyst by forming [NHP]O[Si]OCHO (i.e. IM15) and [NHP]OH by-products. However, as detailed in ESI 2, † the two intermediates can be recovered to catalyst [NHP]H feasibly in terms of both kinetics and thermodynamics.
The methylation mechanism involves formaldehyde and a carbocation species IM17 + as the key intermediates. For the viability of formaldehyde, we call attention to the fact that Bontemps, Sabo-Etienne and coworkers experimentally detected formaldehyde in their Ru-catalyzed conversion of CO 2 to C2 species with pinacolborane as a reducing reagent. 34 Previously, we predicted that formaldehyde could be involved in the NHC- and Ni-catalyzed CO 2 conversion to CH 3 OH. 14 The involvement of a carbocation species in CO 2 conversion has not ever been reported. For the viability of the carbocation species (IM17 + ), the cationic species must not form stable species (namely, IM17OCHO) with the anionic HCO 2 À , because a deep trap would raise the hydrogen transfer barrier from IM17 + + IM3 to TS20 + (Fig. 5A). To estimate the stability of IM17OCHO, we computed the reaction energy of eqn (5). The small endergonicity (1.8 kcal mol À1 ) of the equation indicates that IM17OCHO is only slightly more stable than IM3.
It is interesting to compare the roles of the [NHP] + and HCO 2 À ions in formylation and methylation. In 1a formylation ( Fig. 3), only the HCO 2 À component plays the catalytic role and [NHP] + is a spectator. Differently, in 2a methylation (Fig. 5) the cationic component [NHP] + plays the catalytic role, and [NHP] + promotes the generation of CH 2 O (from IM14 À to IM15) and the carbocation species (IM17 + ) from IM16.

The origins for chemoselectivities of formylation and methylation
The detailed characterizations of the mechanisms of eqn (1) and (2) Table S1 †) support the discussions below. According to the discussion in Section 3.2, the formylation/ methylation preference stems from the competition between nucleophilic attacks of amine and hydride (i.e. TS9 À in Fig. 3 and TS16 in Fig. 5) to [Si](OCHO) 2 . Table 2 compares the barriers of the two attacks for different amines. Note that the barrier for methylation is independent of amines. For 1a formylation, the barrier is 21.8 kcal mol À1 , which is well below the barrier of 25.1 kcal mol À1 for methylation, thus eqn (1) prefers formylation. In contrast, the barrier (29.3 kcal mol À1 , TS9-2a in Fig. 6) for 2a formylation is much higher than the barrier of 25.1 kcal mol À1 for its methylation, rationalizing the production of N-methylated amine (i.e. 2c) in eqn (2). The higher formylation barrier of 2a compared to 1a can be attributed to the greater steric effect in TS9 À -2a than that in TS9 À , as indicated by the shorter H 1 /H 2 distance (2.112Å) than that (2.261Å) in TS9 À . In addition, TS9 À -2a suffers steric repulsion between H 1 and H 3 .
The competition mechanisms rationalize the chemoselectivities of eqn (1) and (2), but the energetic results disagree with the reported experimental result of eqn (3), affording Nmethylated amine 3c. The formylation barrier of 20.5 kcal mol À1 (TS9 À -3a in Fig. 6A) for 3a is lower than that (25.1 kcal mol À1 ) for its methylation. On the other hand, comparing the structures of TS9 À -3a and TS9 À (the TSs for 3a and 1a formylations respectively), the H 1 -H 2 distance (2.329Å) in the former is even longer than that (2.261Å) in the latter, indicating a smaller steric effect in TS9 À -3a than in TS9 À . In addition, the N atom in 3a bears more negative charge (À0.728e) than that (À0.658e) in 1a, indicating that 3a is more nucleophilic than 1a. Thus both the steric and electronic effect agree with the slightly lower formylation barrier (20.5 kcal mol À1 ) of 3a than that of 1a (21.8 kcal mol À1 ). We doubt that eqn (3) might actually produce formamide (3b).
To verify our computed mechanisms and the production of 3b in eqn (3), we performed experiments to study the reactions of 1a-3a (see ESI 3 for experimental details †). 35 Scheme 4 shows our experimental results. Under the same experimental conditions, we were successful in reproducing the reported results of eqn (1), giving 1a in 96% yield (see eqn (6)). However, our study shows that 3a prefers to undergo formylation, affording formamide (3b) in 56% yield (eqn (8)), rather than N-methylated amine 3c as reported previously (eqn (3)), supporting our computational prediction. For 2a, under the same experimental conditions, we could only obtain traces of 2c. Based on our computed mechanism, we reasoned that the poor performance of the reaction could be due to (a) the barrier for methylation (25.1 kcal mol À1 ) being higher than that for formylation (e.g. 21.8 kcal mol À1 for 1a formylation) and (b) [NHP]H being required to nally reduce IM17 + to 2c (see Fig. 5), but it could be consumed during the process reaching IM17 + . Thus, we modi-ed the experimental conditions as shown in eqn (7) of Scheme 4. Delightedly, under the modied conditions, the methylated amine 2c could be produced in 65% yield. Overall the experimental results corroborate our computational prediction satisfactorily.
We have shown that, in the present catalytic system, it is unlikely that methylation passes through formamide as an intermediate.   Table 2).
To further corroborate our conclusions, we calculated the RDS barriers for formylation of the other four amines (4a-7a in Table 2) reported in ref. 16. The barriers for formylation of the four amines, ranging from 18.8-22.5 kcal mol À1 , are all lower than the barrier for methylation (25.1 kcal mol À1 ), in excellent agreement with the experimental fact that these amines prefer formylation. Again, the barriers for hydride transfers to their corresponding formamides (4b-5b) are substantially high (>34.6 kcal mol À1 ). The high reduction barriers of formamides call attention to the sequential mechanism for understanding the methylation of amine with CO 2 .

Conclusions
In this study, we have performed a DFT study to investigate the catalytic mechanisms of the 1,3,2-diazaphospholene ([NHP]H)mediated formylation/methylation of amines (methylaniline (1a)/2,2,4,4-tetramethylpiperidine (2a) 2 to form a C-N bond, nally affording formamide (stage III). Methylation of 2a shares the rst two stages of formylation but is different in stage III. Aer stages I and II, the resultant [Si](OCHO) 2 is preferentially subjected to the attack of an [NPH]H hydride, resulting in formaldehyde which then couples with 2a to form a C-N bond in IM16. Subsequently, IM16 converts to a carbocation species. The methyl group is nally formed via hydride transfer of [NHP]H to the carbocation species. Thus, different from the general consideration that methylation passes through formamide as reduced intermediates of CO 2 , the formylation and methylation in the present catalytic system are two competitive reaction channels. The chemoselectivity originates from the competition between amines and [NHP]H to attack the formyloxy carbon of [Si](OCHO) 2 . If the attack of an amine (e.g. 1a) wins the competition, the transformation affords formamide (1b) and otherwise (e.g. 2a) results in N-methylated amine (2c). The reduction of formamides is highly kinetically unfavorable, which calls attention to the sequential mechanism for understanding amine methylation with CO 2 .
On the basis of the detailed pathways, we have the following key ndings in terms of reaction modes.

Conflicts of interest
There are no conicts to declare.