Optimization of Reaction Conditions

We started our investigations by testing the reaction between the ethyl allenoate 1 a and the parent acceptor 2 a using the well-established isochalcogenoureas ITU1-5 and the previously¹⁵ introduced ISeU and IOU (Table 1). First experiments using achiral ITU1 at room temperature in different solvents and with various bases (as exemplified in entry 1) were discouraging and resulted in no reaction between 1 a and 2 a at all (for comparison: PPh₃ and DABCO promote these reactions at r.t.^9-11). Interestingly however, when analysing the crude reaction mixtures by ESI-MS, a species corresponding to a potential adduct of 1 a and 2 a was observed. In addition, we also observed a species with m/z=303.1168 in the positive ion mode, which corresponds to a protonated catalyst-allenoate adduct like int-A. These ESI-MS observations suggested that higher temperatures (as present in the ESI device) may be required to facilitate the initial addition of ITU1 to 1 a and subsequent reactions of int-A with 2 a. Gratifyingly, when carrying out the reaction at 80 °C (entry 2), a distinct addition product was isolated, which, upon careful structural analysis, could be identified as (Z)-3 a (no reaction takes place at elevated temperature in the absence of catalyst; entry 3). Formation of this product can be explained by γ-addition of int-A to the enone 2 a followed by O-back-attack via the β-carbon of the allenoate and elimination of the catalyst, which also sets the double bond configuration (see below). Interestingly, in these first experiments we also observed smaller amounts of the α-addition product 6 a,²⁰ but only traces (less than 5 %) of the (E)-diastereoisomer, which is the favoured one under the DABCO-mediated conditions.⁹ With this encouraging proof-of-concept at hand, which underscores the complementary nature of this ITU-catalysed approach compared to established amine- or phosphine-catalysed ones, we next screened base and solvent (entries 2–9) as well as different stoichiometric ratios of reagents and base. Non-aromatic polar solvents gave the product in significantly reduced yields and resulted in lower conversions and larger amounts of by-products (entries 8 and 9). Base-free toluene conditions were tolerated too (entry 7), but with a slightly lower yield as compared to the base-mediated ones (entries 2 vs 7) and we also found the reaction being more robust and reliable in the presence of base (the reason may be that the base traps traces of eventually present acids which can have a detrimental effect on the reaction—see below). Overall we thus identified 1 eq. K₂CO₃ in toluene with 1 eq. of 2 a and a slight excess of 1 a (1.5 eq.) to be best-suited to access rac-(Z)-3 a in a good isolated yield of 74 % (entry 2).

With reliable high yielding conditions for the non-enantioselective approach at hand (K₂CO₃, toluene, entry 2), we next focused on the asymmetric variant by using chiral ITU2-5 (entries 10–13). Interestingly, the bicyclo[3.3.0]octene-based ITU2 and ITU3 did not allow for any product formation at all (entries 10, 11). In sharp contrast, the bicyclo[4.3.0]octene-based ITU4 and ITU5 gave (Z)-3 a with high enantioselectivities and in even higher yields and with significantly smaller amounts of α-addition product 6 a as compared to achiral ITU1 (entries 12,13 vs 2). The better performance observed with ITU4 and ITU5 as compared to the smaller ring derivatives ITU2 and ITU3 may be attributed to their higher nucleophilicity¹⁴ and points towards a mechanistic scenario where the addition of the IChU to the allenoate might be the rate limiting step (see below). Remarkably, the high reaction temperatures that are necessary for the cyclization to proceed allowed for high enantioselectivities up to 94 : 6 with ITU5 in these first experiments already. The presence of H₂O resulted in a reduced yield and conversion but without affecting e.r. (entry 14; addition of molecular sieves did not influence the reaction at all), thus underscoring the need for H₂O-free solvents and reagents. Attempts to increase the er by lowering the temperature did not improve the outcome (entries 15 and 16; higher T was not beneficial either). Based on the discouraging results when using more polar solvents for the reaction leading to racemic mixtures (entries 8 and 9), no systematic screening of other solvents for the asymmetric protocol was carried out anymore (a test reaction with ITU5 in THF however showed that only traces of product 3 a were formed in this case (result not given in the Table)). Interestingly, lowering the amount of catalyst to 5 mol % lead to a slightly improved er (entry 17) and allowed for a rather clean reaction, while carrying out the reaction under more concentrated conditions (entry 18) reduced selectivity and yield. Finally, we also tested the less-commonly used ISeU and IOU (entries 19–23).¹⁵ While ISeU was recently very successfully used for acyl-transfer reactions as well as C1 ammonium enolate transformations,^{15, 16l} IOU was reported to be catalytically less competent than the S- or Se-analogs.¹⁵ The same trend was also observed in our case: While IOU gave product 3 a in low yield and with significantly reduced enantioselectivity (entry 21), ISeU allowed for high yield (88 %) and enantioselectivity (98 : 2 er) even on 1 mmol scale (entry 20) and when reduced to 1 mol % catalyst loading (entry 22; lower catalyst loadings as outlined in entry 23 resulted in reduced conversions and yields but still with high er).

Application Scope and Further Manipulations

Having identified the best-suited asymmetric reaction conditions, we next investigated the application scope for a variety of differently functionalized acceptors 2 and allenoates 1 (Scheme 2). To allow for robust conditions and comparable results we systematically used 5 mol % of ITU5 and ISeU. In most cases ISeU was found to be slightly more selective and the reaction was found to be pretty tolerant to various modifications on both reagents. While allenoate ester modifications (3 a–3 d) were all equally well-accepted, the nature, i.e. the electronic properties, of different arylidene groups (3 e–3 r) had a stronger effect, and especially electron-poor aryl groups lead to lower selectivities as outlined for 3 n and 3 o. Gratifyingly the Np-containing 3 p gave crystals suited for X-ray diffraction analysis²¹ which allowed us to unambiguously confirm the double bond configuration being Z and the absolute configuration being S.

Noteworthy, γ-substituted allenoates were also very well-tolerated, giving the γ-addition products 3 s–3 u with good yields and diastereoselectivities (favouring the cis-substituted isomer) and excellent enantioselectivities (interestingly 3 t was the only case where no reaction took place when using ITU5). Finally, variations of the indandione part were also tested (3 v–3 z). Here we finally approached the limits of this methodology as electron-poorer acceptors like 3 y reacted significantly slower and also gave lower selectivities as compared to electron-richer ones like 3 v or 3 w.

To get a deeper understanding on the functional group tolerance we also carried out a short robustness test by carrying out the non-selective synthesis of product 3 e (with ITU1) in the presence of phenol, isobutanol, benzoic acid, aniline, N-benzylbenzamide and benzaldehyde.²² While butanol, aniline, and N-benzylbenzamide showed very little effect, the presence of stronger acids such as phenol and benzoic acid significantly lowered the product yield (details can be found in the SI). These results, which can be rationalized by a detrimental protonation (and thus quenching of reactivity) of the intermediate betaine species, underscore the need for aprotic conditions (compare with entry 14, Table 1) and the beneficial effect of the present base. A lower yield was also observed in the presence of benzaldehyde which can be rationalized by addition of the activated allenoate to the benzaldehyde.

To demonstrate the synthetic value of the highly functionalized tricyclic compounds 3, several follow-up transformations have been carried out. As outlined in Scheme 3, the t-butyl ester of 3 d can easily be hydrolysed under acidic conditions giving the free acid 7. In addition, various reductions either utilizing complex hydrides or by using transition metal-catalysed hydrogenations are possible. While LiBH₄ allows for the chemoselective ketone reduction (giving 8 as a mixture of diastereoisomers), LiAlH₄ reduces both, the ketone and the ester functionality (product 9). In both cases some erosion of er can be observed, but it should be emphasized that no further optimization addressing these subtle levels of epimerization has been carried out anymore. Alternatively, a transition metal-catalysed homogeneous hydrogenation using Wilkinson's catalyst allowed for the selective reduction of the exocyclic C−C double yielding product 10 with high diastereoselectivity.

Mechanistic Investigations

To get a fundamental understanding of the mechanism and the factors controlling the regio-, diastereo- and enantioselectivities in the IChU-catalysed (4+2) cycloaddition of allenoates, we have examined the free energy profile of the reaction between methyl allenoate 1 b and 2 a catalysed by ITU1 (Figure 1). The calculations were carried out at the M06-2X/6–311+G**//M06-2X/6-31G* level of theory, with a continuum description of toluene as the solvent (see Supporting Information for full details).²³ All the potential mechanisms for the ITU1-catalysed reaction between 1 b and 2 a were investigated (see Supporting Information for full details). The predicted most favoured pathways for the (4+2) and (3+2) cycloaddition processes are depicted in Figure 1.

Our calculations indicate that the reaction begins by the addition of the catalyst onto the allene, in good agreement with the experimental observation of int-A (see Table 1). As suggested by experiment (see above), this first step is computed to be the rate determining step of the reaction, with a free energy barrier of 25.9 kcal/mol, explaining the high temperature necessary for the reaction to proceed. The catalyst-allenoate adduct can then add onto the Michael acceptor (2 a) according to two different modes: α- or γ-addition. For steric reasons, the γ-addition is predicted to be much more favourable, proceeding through a low free energy barrier (6.7 kcal/mol) to form int1. This latter can then cyclize toward a five- or a six-membered ring to lead to a (3+2) or a (4+2) cycloaddition product (5 b or 3 b, respectively). The observed exclusive formation of 3 b can be explained by the high endergonicity of the five-membered ring formation and the subsequent high free energy barrier to proton transfer.

Our IChU-catalysed (4+2) cycloaddition reaction provides dihydropyranes 3 with an excellent (>20/1) Z/E selectivity in favour of the Z isomer. According to the mechanism, double bond configuration is set in the last step, the elimination of the catalyst (int2→3). Computed free energy of the two transition states (TS-elim-E and TS-elim-Z) predicts a high selectivity in favour of the Z isomer, in good agreement with experiment (Scheme 4).

The stereogenic centre of dihydropyranes 3 is set during the addition step, leading to int1. Our calculations indicate that this step is essentially non-reversible. Accordingly, in order to get insights into factors controlling enantioselectivity, we investigated the two diastereomeric transition states for addition ((S)−TS-add-γ and (R)−TS-add-γ) in the case of HyperBTM (ITU5) as the chiral catalyst (Scheme 5).

Obtained relative free energies show a large discrimination between the two faces of the Michael acceptor (>10 kcal/mol difference), predicting >99 : 1 er. Experimentally observed enantioselectivities (95 : 5 er for (S)-3 a, Scheme 2) are in good but not complete agreement with this computed selectivity. This could be accounted for by a partial reversibility of the addition step which would enable the epimerization of int1. Indeed, the free energy difference between (S)−TS-add-γ and the subsequent (S)−TS-cycl-6 (compare with Figure 1) is 1.0 kcal/mol (for ITU5). The main factor determining enantioselectivity in our IChU-catalysed (4+2) cycloaddition methodology is thus not only the facial selectivity but also the degree of reversibility of the addition step.

This finding provides an explanation for the observed decrease in enantioselectivity when the Michael acceptor (2) is substituted by electron-withdrawing groups (see 3 m–o, 3 x and 3 y in Scheme 2). Indeed, in these cases, the increased electrophilicity of 2 leads to a lowering of the free energy barrier to addition (TS-add-γ), without impacting significantly the cyclization step (TS-cycl-6), what favours the reversibility of addition, and hence epimerization of the stereogenic centre.

Next, we sought to understand the origin of the difference in reaction product between PPh₃-, DABCO- and IChU-catalysed cycloadditions of allenoates (see Scheme 1B and 1D). The computed free energy profile for the DABCO-catalysed process indicates that formation of the catalyst-allene adduct is more endergonic than with IChU and the subsequent addition onto 2 is also less endergonic (Figure 2). It results that intermediate int1 lies much higher in energy than in the case of the IChU-catalysed reaction and the three key transition states (TS-DABCO, TS-add-γ and TS-cycl-6) have the same free energy, thus indicating a possible (partial) reversibility of the two first steps. In terms of E/Z selectivity, computed free energy of TS-elim-E and TS-elim-Z is 14.4 and 16.0 kcal/mol, respectively, predicting now a selectivity in favour of the E isomer, in good agreement with experiment (see Scheme 1B).⁹

In the case of the PPh₃-catalysed process, the addition of the catalyst onto the allene and the subsequent reaction with the Michael acceptor are similar to that with DABCO. However, for the cyclization step, there is a high preference for the formation a five-membered ring, in the case of PPh₃. This is due to the high stabilization of the negative charge by the phosphonium group in int2-5 (cyclization toward a six-membered ring does not lead to a phosphonium ylide but to an enolate, see int2-6 in Figure 2B). It results that the (3+2) cycloaddition reaction is favoured over the (4+2) cycloaddition when PPh₃ is used as a catalyst, in good agreement with reported experimental data.¹¹

Overall, the observed orthogonal reactivities between amine- (IChU or DABCO) and phosphine-catalysed cycloadditions can be accounted for by the high electrophilicity of the vinylphosphonium function, which outperform the one of the acrylate moiety and hence favour a 5-membered ring cyclization, whereas vinylammonium species are less electrophilic, leading to a 6-membered ring cyclization by O-1,4-addition of the enolate onto the acrylate group (Scheme 6).