Poly(imidazolium) Carbosilane Dendrimers: Synthesis, Catalytic Activity in Redox Esteriﬁcation of α , β -Unsaturated Aldehydes and Recycling via Organic Solvent Nanoﬁltration

: Three series of poly(ionic) carbosilane dendrimers peripherally functionalized with imidazolium groups substituted on N-3 with methyl, isopropyl and 2,6-diisopropylphenyl (Dipp) were prepared up to the 3rd generation together with model monovalent imidazolium iodides and used as N-heterocyclic carbene (NHC) precursors. Catalytic activity of model and dendritic NHCs generated in situ by deprotonation with DBU was tested in redox esteriﬁcation of α , β -unsaturated aldehydes and the inﬂuence of substitution, dendrimer generation, temperature and substrate structure on the reaction outcome was evaluated. Dipp substituted NHCs showed high activity and selectivity in the reaction with primary alcohols. Effectiveness of organic solvent nanoﬁltration for the recycling of dendritic NHCs was demonstrated on the 1st generation Dipp substituted catalyst in model redox esteriﬁcation of cinnamaldehyde with benzyl alcohol. A marked increase in both activity and selectivity in the ﬁrst four reaction runs was observed and this improved performance was preserved in the following catalytic cycles.


Introduction
Azolium salts can act as effective organocatalysts [1,2], providing catalytically active species with different mechanisms of substrate activation [3][4][5]. Besides simple activation of substrates' nucleophilic center by hydrogen bonding or ion-pair interaction, enabling e.g., cycloaddition of carbon dioxide to epoxides [6,7], they can serve as precursors of N-heterocyclic carbenes (NHCs). These subtle species generated by deprotonation of azolium salts offer an entirely unique active site through the carbene atom and its tunable surroundings, able to promote a broad range of chemical transformations [8][9][10][11]. Over the last two decades, the practical application of so called umpolung principle provided access to a wide library of compounds [12,13]. Nevertheless, while many aspects of NHCcatalyzed umpolung reactions, such as activity and selectivity of the catalysts, were studied and optimized, only few studies dealing with the problem of catalyst separation and reuse have been reported so far, moreover, they were limited to benzoin condensation [14][15][16] and Stetter reaction [17].
Increasing the sustainability of homogeneous catalytic processes and overcoming the major drawback of organocatalysis, i.e., high catalyst loading, by means of catalyst recycling is still a challenge. Recently we developed a branched fluorous tag designed for easy fluorophilization of various substrates [18,19]. Highly fluorophilic imidazolium salts bearing this tag were used as NHC precursors and applied in redox esterification carried out in fluorous biphase system, which allowed for their effective recycling [20]. Such solution allows the reaction to proceed in fully homogeneous mode, but it is unsuitable for continuous processes. Accordingly, immobilization or incorporation of a homogeneous catalyst in a solid (inorganic or polymeric) support still remains a predominant approach for the separation of catalysts from reaction mixtures [21], although the homogeneity of the active site is lost to some extent. Many types of immobilized ionic liquids (ILs) on different supports, such as silica [22], MOFs [23], zeolites, various nanoparticles and other materials, have been described [24]. Somewhat surprisingly, soluble macromolecules, such as dendrimers, are much less utilized, in spite of their potential advantages.
Dendrimers, as highly branched macromolecules with regular, tunable structure and low polydispersity due to a well-controlled synthesis [25], can be considered as chemical individuals, despite of their size. They seem to be ideally suited for the attachment of functional moieties at defined positions, yielding their multivalent analogs as optimal systems to study multivalency and the effect of immobilization. Several methods have been proposed for the recovery of dendrimers from reaction mixtures [26,27], nevertheless most of them makes use of different solubility of components and phase separation [26,[28][29][30]. There are also some reports on magnetically recoverable dendrimers and catalysts with dendritic ligands utilizing their interaction with magnetic nanoparticles [31,32]. Recovery methods based on molecular enlargement of the active group, employing membrane processes for the separation of the homogeneous catalyst and other reaction components, are reported much less than it would correspond to the high suitability of this approach for dendritic catalysts. Dialysis in commercially available membrane bags, a slow, diffusiondriven process, is mostly used [33,34]. Despite low efficiency of early attemps [35], a substantial effort was made to develop the method [36,37].
We have previously tested carbosilane dendritic supports for immobilization of both metal-based [38] and metal-free catalysts, which recycling by means of organic solvent nanofiltration (OSN) [39] was also accomplished [40]. OSN proved to be a very effective method, enabling separation of homogeneous reaction components under mild conditions with low energy and material input. This study expands our previous research to another type of catalysts and aims to the preparation of multivalent NHCs. We describe the influence of the attachment of imidazolium salts to dendritic supports on their catalytic performance in redox esterification, their activity towards a wide variety of substrates and their recycling by OSN. We also report the positive effect of OSN on the activity of multivalent NHC catalyst, leading to increase of reaction rate in subsequent reactions.

Synthesis and Characterization
For this study, three novel series of monovalent models and multivalent dendritic catalysts were prepared using the synthetic protocol that we have recently developed for the preparation of related cationic carbosilane compounds; a thorough description of synthesis and characterization data can be found in the Experimental Section and in the published papers [40,41]. The final synthetic step involving quaternization of imidazoles with either model 3-iodopropylbutyldimethylsilane or 3-iodopropyl-terminated dendrimers is depicted in Scheme 1.

Scheme 1. Synthesis of model and dendritic imidazolium precatalysts.
Whereas the quaternization step with methyl-and isopropylimidazole is nearly quantitative according to both NMR and ESI MS spectra (Supplementary Material), giving almost monodisperse products, the reaction with hindered diisopropylphenylimidazole (DippIm) is less effective, which leads to defects in the structure at higher generations and lower loading of imidazolium units. HRMS showed the presence of iodopropyl and/or hydroxypropyl-terminated branches, the same types of defects we have reported previously for poly(phosphonium) carbosilane dendrimers [42]. The sum of quaternization-related defects typically did not exceed 5% as estimated from 1 H NMR and HRMS (with the exception of G 2 (DippImI) 8 , in which nearly 20% of defective branches were present according to 1 H NMR). However, as the quantification is rather inaccurate due to peak overlap and the progress of catalysis is quite insensitive to such small differences in catalyst concentration, the calculation of catalyst loading was based on the nominal number of active groups in the molecule without defects.

Screening of Catalysts
All model imidazolium salts were tested as precatalysts of redox esterification of cinnamaldehyde and benzyl alcohol (Table 1). This reaction is often used as a benchmark allowing facile evaluation of the catalytic performance of newly prepared compounds [43,44]. Preliminary experiments were performed at 100 • C in toluene, with 10 mol% catalyst loading and 10 mol% of the base (DBU); dioctyl phthalate was added as an inert standard to enable accurate quantification of GC-FID data. As evident from entries 1 and 5, the catalysts generated from methyl-and iso-propyl-substituted model imidazolium salts G 0 MeImI and G 0 iPrImI by deprotonation showed similar performance to previously described fluorinated analogs, for which also a superior activity of the methyl-substituted compound was observed [20]. To our delight, the third model G 0 DippImI bearing a sterically demanding 2,6-diisopropylphenyl (Dipp) group manifested by far the highest activity, reaching the full conversion of the starting aldehyde in less than 20 min (entry 8) and confirming the previously observed positive influence of bulky aryl substituents on azolium on the catalyst performance [45]. All three models showed high selectivity for the formation of saturated ester and only small quantities of byproducts 4 and 5a reported previously to accompany this reaction [20] were detected by GC (Table 1). Exploiting the unprecedented high activity of G 0 DippImI, we examined the influence of the reaction temperature ( Figure 1a) and decrease in catalyst loading. It turned out that lowering the reaction temperature has a detrimental effect on selectivity and overall yield, benzyl cinnamate 5a being the major side product at low temperature (entry 13). A similar change in selectivity was previously observed by Scheidt et al. for triazolium and imidazolium-based NHC catalysts in enantioselective redox esterification carried out at low temperature [46,47]. On the other hand, lowering the amount of the precatalyst to 1 mol% (with twofold excess of the base) had a positive effect on selectivity while the reaction rate was still fair (entry 10). The reaction rate could also be boosted by the use of larger excess of the base (10-fold to the catalyst) which, however, slightly lowered the selectivity (entry 9). It is clear that the base plays multiple role in the reaction and it serves not only to generate the active NHC moiety, but also facilitates other transformations in the catalytic cycle, as already suggested by Sohn and Bode [48].
Next, we examined the effect of the dendritic scaffold in the benchmark reaction. As apparent from Figure 1, a negative effect of immobilization was observed in all cases. Methylimidazolium derived dendritic catalysts showed a severe drop in activity compared to the model compound even in the first generation, while the differences between generations 1, 2 and 3 were negligible (Figure 1b). This behavior can be ascribed to a close proximity of active sites leading to preferential formation of intramolecular adducts between NHC and imidazolium moieties stabilizing the carbene and lowering its activity as an organocatalyst [11]. The dendrimers bearing isopropylimidazolium groups not only had a very low activity, but at the same time they exhibited poor selectivity (entries 6, 7), so they were excluded from further studies. On the other hand, in case of DippIm series the deactivation is not so severe and it is clearly generation dependent ( Figure 1c). Again, the addition of excess base had a positive influence on the reaction rate (cf. Figure 1c,d), albeit less pronounced than in case of the corresponding model. Considering the amount of defects in G 2 (DippImI) 8 , the activity of one particular active site is very similar for G 1 and G 2 catalyst when large excess of base is used. Under the same conditions the activity of G 3 catalyst remained somewhat lower, although experiencing the largest increase (three times higher reaction rate). It seems clear that the bulky Dipp substituent effectively hinders the formation of adducts and it is responsible for the higher activity of this type of dendritic catalysts. The observed negative dendritic effect can be rationalized by a restricted mobility of active sites residing at the periphery of a dendrimer and their uneven distribution in solution (variable local concentration of catalyst), which implies lower probability of effective collisions with the base and substrates. Figure 1. Time course of redox esterification of cinnamaldehyde with benzyl alcohol (4 equiv) in toluene; (a) effect of temperature on the reaction catalyzed by G 0 DippImI (10 mol%; DBU 10 mol%); (b) 100 • C, influence of the support, precatalyst from methylimidazolium series G n (MeImI) m (n = 0-3, 10/m mol%; DBU 10 mol%); (c,d) 100 • C, influence of the support and amount of the base, precatalyst from Dipp imidazolium series G n (DippImI) m (n = 0-3, 1/m mol%; DBU 2 mol% (c), 10 mol% (d)).
In our recent study utilizing fluorous imidazolium as a precatalyst we have observed that the use of secondary alcohols led to a slight decrease in the reaction rate, yield and selectivity [20]. Compared to those fluorinated NHCs, Dipp substituted organocatalysts demonstrated much higher sensitivity to the structure of alcohol, even if the trends were similar. Apart from benzyl alcohol, the best results were obtained for primary alcohols without substitution at α-carbon (2b, 2c, 2d). The reaction with α-substituted haloalcohols 2i and 2j was much slower, but the selectivity remained high. Generally, the difference between the activity of the model and dendritic catalysts was larger than observed for the benchmark reaction with benzyl alcohol. Secondary alcohols were poor substrates except for unsubstituted alicyclic alcohols (2e, 2f, 2g), which gave moderate yields. Tert-butyl alcohol and phenol gave only traces of products. The lack of G 0 DippImI activity towards phenol contradicts our previous findings, as we have previously obtained fair yields of phenyl ester under the catalysis by fluorinated methylimidazolylidene [20]. A deeper insight into the complex reaction mechanism is needed to understand this unexpected behavior; a mechanistic study addressing the issue is currently underway. The typical side products (lactone 4 and unsaturated esters) were present in trace amounts (0-2%). However, especially for the combination of less reactive alcohols and dendritic catalysts, a marked difference was observed between the consumption of the aldehyde and the overall yield, pointing to the possible formation of higher molecular weight impurities.
As the bulky Dipp substituent obviously hinders the reaction with sterically more demanding alcohols, we also tested the methylimidazolium-derived catalysts in the reac-tion with such substrates. Although the activity of G n (MeImI) m in the benchmark reaction was an order of magnitude lower than that of G n (DippImI) m , the reaction rate in the case of more hindered alcohols was comparable or even higher and the selectivity was much better ( Table 2). Esters of alicyclic alcohols were formed in good yields, the performance of the 2nd generation catalyst was slightly better than that of the 1st one.
Subsequently we carried out a short screening of representative α,β-unsaturated aldehydes with model G 0 DippImI as a precatalyst (Table 3, products depicted in Figure 2). As expected, α-substituted aldehydes 1b and 1f did not react at all and only a trace of saturated ester was obtained with β-substituted substrate 1c; the major product in this case was the unsaturated ester 5m. Other tested aldehydes gave moderate to fair yields of the target products but the rate was low compared to the reaction with cinnamaldehyde. The reactions with 1d and 1g were selective and only traces of low M W byproducts were detected by GC. In the case of 2,4-decadienal, the participation of the double bond in a larger conjugated system accelerates the reaction rate, but it also promotes the formation of the unsaturated product 5o. The lowest yield and selectivity were observed for onitrocinnamaldehyde 1h, possibly due to the thermal instability of the enal and/or target product (a thermally induced degradation of products was observed during their isolation). Aside from aldehyde oligomers, a substantial quantity of the unsaturated ester 5q was also formed. When employing the dendritic catalysts G 1 DippImI 4 and G 2 DippImI 8 , again the decrease in both the rate and selectivity was observed, especially for the less activated substrate 1d ( Supplementary Information, Table S1).

Recycling of Dendritic Catalysts
To verify the feasibility of separation and repeated reuse of the prepared dendritic catalysts, a suitable protocol for catalyst recycling by organic solvent nanofiltration (OSN) was developed. The benchmark reaction was carried out as usual, only with higher loading of the catalyst (2.5 mol% corresponding to 10 mol% of active groups) which enabled to shorten the reaction time and marginalized the influence of catalyst losses due to material transfers between the reaction vessel and nanofiltration cell. We also tried to lower the excess of benzyl alcohol to 2 equiv with respect to the aldehyde. However, this led to insufficient solubility of the catalyst in the reaction mixture and another 2 equiv had to be added to reach the homogeneity. In the next cycles, the additional 2 equiv of benzyl alcohol were substituted by the same amount of tert-butyl alcohol as a cheaper and easierto-remove solvent. A separate experiment with fresh catalyst showed that this change of reaction conditions had a negligible effect on the reaction progress ( Figure 3).
As summarized in Table 4, the catalyst recycling by means of OSN had a positive influence on the reaction outcome. From the first to the fourth cycle, both the activity and product purity increased, reaching a very satisfactory level (TOF = 10 h −1 , purity over 97%). The isolated catalyst was analyzed by HRMS after each cycle (Supplementary Information, Figure S1). The results suggest that a certain part of active groups stays in the form of NHC which is stable enough [49] to outlast the nanofiltration, and this fraction increases with increasing number of reaction/nanofiltration cycles due to repeated washing of iodide salt of DBU ([DBUH]I) off from the dendrimer. This can explain the observed increase in activity (better solubility and shifted equilibrium Im + + DBU NHC + DBUH + ) as well as the markedly lower dendrimer recovery in the first nanofiltration cycle (the most extensive loss of HI from dendrimer). The higher reactivity of recycled catalyst was confirmed by a direct comparison of reaction kinetics with a fresh catalyst; more importantly, we have demonstrated that the recycled catalyst exhibited significant catalytic activity even in the absence of base (Figure 3) while fresh G 1 (DippImI) 4 showed no activity under such conditions. Considering that these tests were performed one month after the last cycle and meantime the recycled catalyst was stored on air in ambient temperature, the carbene moiety exhibits remarkable stability.  4 (2.5 mol%) in the redox esterification of cinnamaldehyde (4 equiv of benzyl alcohol, toluene, 100 • C; 10% DBU unless otherwise noted). a 2 equiv of benzyl alcohol were replaced by tert-butyl alcohol. a Conditions: precatalyst loading 2.5 mol% (10 mol% with respect to active groups), 15 mol% DBU, 2 equiv of benzyl alcohol + 2 equiv of tert-butyl alcohol, 100 • C; b average value calculated from time to full conversion; c based on the weight of the isolated catalyst after nanofiltration; d 4 equiv of benzyl alcohol, without tertbutyl alcohol.

General Remarks
All experiments were carried out in anhydrous conditions under an inert atmosphere of argon or nitrogen, using standard Schlenk techniques, unless otherwise stated. Solvents were dried by usual procedures, then distilled and kept under argon. CDCl 3 and DMSO-d 6 were dried over molecular sieves. Starting iodopropyl terminated dendrimers G m (PrI) n were prepared as described previously [41]. All other used chemicals were commercial products and were used without further purification. NMR spectra were measured on Bruker Avance 400 ( 1 H at 400.1 MHz; 13 C at 100.6 MHz; 29 Si { 1 H} (inept technique) at 79.5 MHz) at 25 • C. 1 H and 13 C NMR signals of prepared compounds were assigned to corresponding atoms utilizing HSQC, COSY and HMBS 2D NMR correlation spectra. 1 H and 13 C chemical shifts (δ/ppm) are given relative to solvent signals (δ H /δ C : DMSO-d 6 2.50/39.52); 29 Si spectra were referenced to external standard hexamethyldisilane (−19.87 ppm).
GC/FID and GC/MS analyses were performed on Agilent 6890 gas chromatograph equipped with DB-5MS column (30 m_0.25 mm_0.25 mm). Injector and detector temperatures were set at 260 • C and 300 • C, respectively. Helium was used as a carrier gas at a flow rate 1.61 mL/min. Oven temperature was set at 50 • C for 3 min and then programmed to 290 • C at a rate 10 • C/min. Quantitative data were calculated using Agilent MSD Chemstation D.02 from the FID peak areas assuming a linear response for the analytes and the internal standard dioctyl phthalate and applying relative response factors calculated from the molecular formula according to previously published method [50]. In the case of cinnamaldehyde, benzyl 3-phenylpropanoate and benzyl cinnamate the results were corrected with respect to experimentally determined calibration curves relative to dioctyl phthalate.

Synthesis of Model Imidazolium Salts and Poly(imidazolium) Carbosilane Dendrimers
General procedure. To a solution of the starting iodopropyl derivative G m (PrI) n (1.0 mmol) in acetonitrile (15 mL) a solution of a given imidazole (1.3 eq. for each iodide) in acetonitrile was added at 0 • C. The reaction mixture was then stirred at 70 • C for 48 h. The solvent was removed on the rotary evaporator and toluene (6 mL) was added to the residue to wash out the unreacted imidazole. The mixture was stirred at 80 • C for 6 h, let to cool down and decanted at room temperature. The residue was dried under vacuum to afford the product.  Dendrimer G 2 (MeImI) 8 . The compound was prepared from dendrimer G 2 (PrI) 8

Redox Esterification of Cinnamaldehyde
In a typical experiment, 2.8 mg (5.0 µmol) of G 0 DippImI was placed in a Schlenk flask which was then three times evacuated and refilled with argon. Cinnamaldehyde (66 mg, 0.50 mmol), alcohol (2.00 mmol), DBU (1.5 mg, 0.01 mmol) and internal standard for GC dioctyl phthalate (59 mg, 0.15 mmol) followed by 5 mL of toluene were added under the flow of argon. All components were thoroughly mixed, 30 µL sample was withdrawn for GC analysis and the reaction mixture was heated for a specified time in the oil bath of constant temperature (100 • C unless otherwise noted). Then the mixture was let to cool down and analyzed by GC. In case that samples were taken in the course of experiment for kinetic studies, for each point 30 µL of the reaction mixture was withdrawn via syringe under the flow of argon, dissolved in cold DCM and stored in refrigerator till analysis.
For the purpose of substrate screening of alcohols, a stock solution of cinnamaldehyde (0.1 mol.L −1 ), DBU (4.0 mmol.L −1 ) and dioctyl phthalate (0.03 mol.L −1 ) in toluene and stock solutions of precatalysts in MeOH (20 mmol.L −1 with respect to active centers) were prepared and stored under argon at 4 • C. The composition of aldehyde stock solution was confirmed by GC-FID. In a typical experiment, 100 µL of a stock solution of a given precatalyst was placed in a Schlenk flask and the solvent was evaporated in vacuum. The flask was then three times evacuated and refilled with argon and 1.0 mL of substrate stock solution was added under the flow of argon followed by 0.4 mmol of the corresponding alcohol. The reaction mixture was heated for a specified time to 100 • C in the oil bath and samples for GC were taken to follow the reaction progress. For the reactions with higher catalyst loading the amount of precatalyst stock solutions and the concentration of DBU in the stock solution were correspondingly increased.
The procedure for substrate screening of aldehydes was analogous, using stock solution of benzyl alcohol (0.4 mol.L −1 ), DBU (0.01 mol.L −1 ) and dioctyl phthalate (0.03 mol.L −1 ) in toluene to which the particular aldehyde (0.1 mmol) was added. Before the heating of the reaction mixture, a sample for GC was taken to confirm the ratio of the aldehyde to the internal standard.
Recycling experiments were performed on a larger scale without internal standard. The first cycle was set up with 1.00 mmol of cinnamaldehyde (132 mg), 4.00 mmol of benzyl alcohol (433 mg), 25.0 µmol of G 1 (DippImI) 4 (50.0 mg) and 0.15 mmol of DBU (23 mg) in 5 mL of toluene; the batch size in the following cycles was adapted to correspond with the amount of recycled precatalyst. The same concentration of all components at the start of all reaction runs was maintained, except benzyl alcohol, 50% of which was replaced by tert-butyl alcohol. After each cycle, G 1 (DippImI) 4 was separated from the reaction mixture by nanofiltration (vide supra), dried, weighted and checked by ESI MS. The product containing permeate was evaporated in vacuum, dissolved in dichloromethane, filtered through a short pad of silica gel and evaporated to dryness to afford the target ester.

Organic Solvent Nanofiltration
Nanofiltration was carried out using solvent resistant stirred cell (Millipore) equipped with 1 kDa MWCO regenerated cellulose ultrafiltration discs Ultracel ® (Millipore) and PTFE encapsulated O-rings (ERIKS), with nitrogen as a driving gas. Reaction mixture was quantitatively transferred into the cell, diluted with methanol to a total volume of 25 mL and passed through the membrane until the residual volume 1 mL of retentate was reached. The retentate was diluted with 20 mL of methanol and filtered; this procedure was repeated 3 times and subsequently the retentate was withdrawn via cannula. The inner space of the cell was washed with 3 × 2 mL of methanol and the retentate and washings were combined and evaporated under vacuum to recover the dendritic catalyst. The cell was not disassembled and it was used directly for the next separation cycle. The product was obtained after the permeate workup.

Conclusions
We have demonstrated the feasibility of immobilization of imidazolium salts on the dendrimer periphery and their use as multivalent carbene precursors in NHC catalysed redox esterification of α,β-enals. Although the immobilization led to expected decrease of catalytic activity, the deactivation can be limited by appropriate substitution on imidazolium. From the studied types, Dipp substituted catalysts were least influenced and exhibited high activity and selectivity in the reactions with primary alcohols. On the other hand, the bulky Dipp substituent hinders the reaction with secondary alcohols, for which a clear relationship between the steric hindrance of the alcohol nucleophile and the yield and purity of the product was observed. This drop in both activity and selectivity is much less pronounced for methyl substituted dendritic NHCs, which makes them more competent catalysts in the case of alicyclic alcohols. The concept of dendritic catalyst recycling by nanofiltration, although not new, has been shown as especially suited for this class of active species. The separation from the reaction mixture was effective and the catalyst recovery was high even for the first generation Dipp imidazolium dendrimer. Moreover, OSN induced an increase of catalytic activity, partially compensating for the negative effect of immobilization. Such "activation" can be ascribed to a shift in the equilibrium between imidazolium iodide and the corresponding carbene induced by separation of DBU salt during nanofiltration. Dipp substituted NHCs proved to be particularly stable. At least a part of these carbene moieties was present in the recycled material even after several weeks and it was capable to induce the redox esterification without added base which can, together with catalyst recycling, contribute to higher atom economy of the process.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/ 10.3390/catal11111317/s1, NMR spectra of all prepared precatalysts; synthesis and characterization of previously unpublished esters, EI-MS spectra of all identified products from redox esterification experiments; extended results of substrate screening of aldehydes; selected HRMS spectra of recycled G 1 (DippImI) 4 .