Diversity-oriented synthesis of glycomimetics

Glycomimetics are structural mimics of naturally occurring carbohydrates and represent important therapeutic leads in several disease treatments. However, the structural and stereochemical complexity inherent to glycomimetics often challenges medicinal chemistry efforts and is incompatible with diversity-oriented synthesis approaches. Here, we describe a one-pot proline-catalyzed aldehyde α-functionalization/aldol reaction that produces an array of stereochemically well-defined glycomimetic building blocks containing fluoro, chloro, bromo, trifluoromethylthio and azodicarboxylate functional groups. Using density functional theory calculations, we demonstrate both steric and electrostatic interactions play key diastereodiscriminating roles in the dynamic kinetic resolution. The utility of this simple process for generating large and diverse libraries of glycomimetics is demonstrated in the rapid production of iminosugars, nucleoside analogues, carbasugars and carbohydrates from common intermediates.

C arbohydrates are essential biomolecules that play critical roles in cell signaling, protein folding, and metabolism 1 . The recognition and regulation of carbohydrate structure are thus strictly controlled processes that rely on the fidelity of an array of carbohydrate-specific enzymes (e.g., glycosyl transferases and glycoside hydrolases) and binding proteins (e.g., lectins) 2,3 . While the use of carbohydrates as therapeutics is limited by generally poor pharmacokinetic properties, structural mimics of carbohydrates, also known as glycomimetics (1-6, Fig. 1), that interact in a similar manner with protein targets but have improved drug-like properties (e.g., affinity, stability and bioavailability) represent promising alternatives [2][3][4][5][6][7][8] . Common structural changes found in glycomimetics include replacement of the endocyclic oxygen with a carbon (carbasugars: 1) 9 or nitrogen (iminosugar: 3 and 4) 10 and substitution of a fluorine atom for hydroxyl groups (deoxyfluorosugars: 2, 5, and 6) 11 . For example, Oseltamivir (1) 12 is a carbasugar mimic of sialic acid and a potent neuraminidase inhibitor that became the frontline antiviral drug used during the H1N1 pandemic in 2009. Likewise, Miglitol (3) 13 is a glycomimetic that functions by inhibiting α-glucosidases and decreasing carbohydrate metabolism in patients with type II diabetes. Notably, iminosugars like 3 are typically protonated at physiological pH, which allows them to mimic the oxacarbenium ion transition state traversed during hydrolysis by glycoside hydrolases 10 .
The rational design of glycomimetic drugs often initiates with a structural analysis of the natural carbohydrate substrate bound to the protein target of interest, followed by iterative synthesis campaigns focused on stabilizing the bioactive conformation and removing unnecessary functional groups to improve drug-like properties 5 . Considering the vast array of biologically relevant carbohydrates, diversity-oriented synthesis (DOS) 14,15 strategies that broadly sample this biologically relevant chemical space would also be well-suited to support drug discovery efforts 16 . However, the functional group density and stereochemical complexity inherent to carbohydrates often requires that DOS strategies rely on the derivatization of naturally occurring carbohydrates, ultimately limiting diversity 16 . For example, Wong has reported a library of potent and selective α-fucosidase inhibitors derived from 1-aminomethyl-fuconojirimycin (7) (Fig. 1) 17 . Here, late-stage amide coupling was used to modify the core scaffold 7, itself generated through a 12-step synthesis. A DOS approach has also been reported by Marcaurelle, where 2,3unsaturated C-glycoside scaffolds were used to synthesize several new bicyclic carbohydrates 18 . Recently, Loh reported a strategy for the diversification of carbohydrates using hydrogen-and halogen-bond-catalyzed strain release glycosylation to produce complex O,N-glycoside analogs 11 ( Fig. 1) 19 . Ultimately, this platform enabled the discovery of potential leads for treating acquired cancer resistance. While these limited examples highlight a clear role for DOS in the generation of glycomimetics, they also underscore the need for new synthetic strategies that enable rapid sampling of more diverse regions of carbohydrateassociated chemical space 20 .
As an alternative de novo approach to carbohydrates and glycomimetics, we have reported a one-pot α-chlorination/aldol reaction (αCAR, Fig. 2) 21 . As detailed in Fig. 2 (top panel), here proline catalyzes (i) the racemic α-chlorination of aldehydes, (ii) the interconversion of the resulting racemic α-chloroaldehydes 12, and (iii) their subsequent aldol reaction with dioxanone 13. To rationalize the preference for syn-chlorohydrins 14, we proposed that electrostatic repulsion between the Cl and O atoms in the transition structure (S)-TS (X = Cl) disfavors formation of anti-chlorohydrin 15 21 . The high degree of enantioselectivity results from H-bonding that directs the facial approach of the correctly configured α-chloroaldehyde to the proline-derived enamine through a Houk-List type transition structure [22][23][24] . Importantly, chlorohydrin scaffolds 14 produced via this dynamic kinetic resolution (DKR) can be readily converted into ribose analogs 21 as well as other glycomimetics including carbasugars (e.g., 16 25 ) and iminosugars 26 (e.g. 17 27 ). Recently, we reported a complimentary proline-catalyzed α-fluorination/aldol reaction (αFAR) that supports the rapid synthesis of nucleoside analogs (e.g., 18) 28 . While these two processes have provided access to a range of useful glycomimetics, both the αCARs and αFARs rely exclusively on dioxanone 13 as the ketone coupling partner 24,29,30 , which prevents wider application of these strategies to library synthesis and DOS pursuits.
Results and discussion α-Functionalization/aldol reactions. Based on the limitations noted above, we envisioned that new α-functionalization/aldol reactions 31,32 , involving a broader selection of (i) electrophiles and (ii) enolizable ketones, would support the construction of diverse collections of glycomimetics. Specifically, we aimed to exploit organocatalytic aldehyde α-functionalization reactions, including αchlorination [33][34][35] , α-fluorination [36][37][38] , α-amination 39,40 , and αtrifluoromethylthiolation 41 in combination with proline-catalyzed aldol reactions of various cyclic and acyclic ketones 21 . Importantly, these processes would avoid isolation 27,28 of often unstable and configurationally labile α-functionalized aldehydes and expand their general utility. Toward this goal, we first investigated combinations of different proline-catalyzed α-functionalization reactions with prolinecatalyzed aldol reaction of dioxanone 13. As summarized in Table 1 these reactions were performed as two-step-one-pot sequences (see Supplementary Table 1 for details). Specifically, L-proline-catalyzed α-functionalization using N-chlorosuccinimide 34,42 , N-bromosuccinimide 43 , N-fluorobenzenesulfonimide [36][37][38] , N-trifluoromethylthiophthalimide (PhthN-SCF 3 ) 44 , or dibenzyl azodicarboxylate 39,40 was followed by the direct addition of dioxanone 13. As summarized in Table 1, we found that these reactions preferentially afford syn-chlorohydrin 27, syn-fluorohydrin 28, synbromohydrin 29, syn-trifluoromethylthiohydrin 30, and synaminohydrin 31 with variable diastereoselectivity and in generally excellent enantioselectivity. Considering that the proline-catalyzed αfluorination of aldehydes is not an enantioselective process (ee's < 30%) 45 , the selective formation of fluorohydrin 28 was attributed to a DKR of the intermediate α-fluoroaldehyde as observed previously with α-chloroaldehydes 21 . In a separate experiment, involving the αFAR of hydrocinnamaldehyde, the diastereomeric ratio of products (>15:1) and enantiomeric excess of the intermediate α-fluoroaldehyde did not change over the course of the reaction even as the (R)fluoroaldehyde was consumed, further confirming the role of a DKR in these processes (see Supplementary Table 2). Notably, the diastereoselectivity in the production of halohydrins (entries 2-4) correlates with increasing electronegativity of the halogen atom. Thus, despite a smaller van der Waals radius and shorter C-X bond length for X = F, the diastereoselective aldol reaction of the αfluoroaldehyde was more greatly differentiated. The αCAR was also carried out in CH 2 Cl 2 (rather than 9:1 CH 2 Cl 2 -DMF) and we observed a coincident increase in diastereoselectivity (2.2:1 to 6:1; entries 1 and 2) consistent with an increased influence of electrostatic interactions in the diastereodifferentiating step ( Fig. 2). As summarized in entry 5, the α-trifluoromethylthioaldehyde derived from pentanal also underwent a diastero-and enantioselective aldol reaction with dioxanone 13. In the case of the α-aza aldehyde generated from the reaction of pentanal and dibenzyl azodicarboxylate (entry 6) 40 , steric hindrance precludes formation of a proline enamine required for racemization 46 . Thus, the ultimate diastereoselectivity (dr = 3:1) reflects the ratio of enantiomeric α-aza aldehydes generated in situ. Notably, when this reaction was repeated in CH 2 Cl 2 or DMSO, the yield was significantly lower. The L-proline-catalyzed aldol reaction between (S)-2-Cbz-aminopentanal and dioxanone yielded the corresponding anti-aminohydrin as the sole product, and both L-and D-proline-catalyzed aldol reactions of (S)-N-Cbz prolinal (Supplementary Schemes 1 and 2) each gave single products without epimerization of the α-stereocenter, further confirming that αaminoaldehydes do not racemize under these reaction conditions (k rac << k aldol ) 47 . We also examined the L-proline-catalyzed aldol reaction of (±)-2-phenylpropanal and dioxanone 13 (Supplementary    Scheme 3), which afforded an equal mixture of syn-and anti-diastereomers again suggesting k rac << k aldol .
The computed trends for reactions of (R)-and (S)-2chloropentanal were similar to those discussed above. Namely, three low-energy transition states (R)-TS1 O -Cl, (S)-TS2 O -Cl, and (S)-TS3 O -Cl were found with the latter two possessing structural attributes of prototypical Evans-Cornforth and Felkin-Anh carbonyl addition models (Fig. 4b). Among these structures, the transition state (R)-TS1 O -Cl was lower in energy than both of the diastereomeric transition states (S)-TS2 O -Cl and (S)-TS3 O -Cl, in agreement with the preferential formation of syn-chlorohydrin 27 ( Table 1, entry 1). In particular, the Evans-Cornforth-type transition state (S)-TS2 O -Cl was destabilized by a repulsive interaction between an oxygen on the dioxanone-derived enamine and the aldehyde chlorine with a distance measuring 3.05 Å. In contrast, Felkin-Anh-type transition state (S)-TS3 O -Cl suffered from steric interactions as seen from a close hydrogen-hydrogen contact with a distance of 2.21 Å. On the contrary, the energetically favored transition state (R)-TS1 O -Cl contained fewer steric contacts and favorable NCIs, including a  Fig. 3 α-Chlorination/aldol reaction products. α-Chlorination/aldol reaction products from a range of ketones. Isolated yields for isolated diastereomer shown, diastereomeric ratio determined by 1 H NMR spectroscopic analysis of crude reaction mixture.
chlorine-hydrogen contact measuring 2.72 Å, visible from a NCI surface (see Supplementary Figure 3). Collectively, these studies revealed several conserved features irrespective of fluoroor chloro-substitution and similar trends were found in the computed reactivity patterns of the cyclic ketones cyclohexanone, tetrahydropyranone, and thiopyranone 35 (Supplementary  Tables 1-9). Notably, across this panel of ketones, the preferred transition state structures consistently shared geometries analogous to (R)-TS1 O -F and (R)-TS1 O -Cl. Moreover, these trends were adhered to with bromo-substitution (see Supplementary  Table 6). In all cases, the minimization of repulsive steric interactions and favorable NCIs including a halogen-hydrogen (C-X···H) interaction were identified as key contributors to preferred formation of syn-halohydrin products.
Scope of α-functionalization/aldol reactions. With several new α-functionalization/aldol reactions in hand, we examined the reaction of a larger collection of aldehydes with one of the electrophiles demonstrated in Table 1 and ketones demonstrated in Table 2. As detailed in Fig We were particularly encouraged by productive reaction of αfluoro-α-heteroaryl-acetaldehydes, which produced adducts 68-70 that should support the synthesis of nucleoside analogs 28 . Considering the SCF 3 group is commonly used to increase lipophilicity of drug leads, we were pleased to find that syntrifluoromethylthiohydrins containing an alkyl (  Rapid synthesis of glycomimetics. To demonstrate the potential for this strategy to rapidly generate glycomimetics, several readily available chlorohydrins, fluorohydrins, trifluoromethylthiohydrins, and aminohydrins (Fig. 5) were converted into a diverse collection of iminosugars, furanose analogs, and bicyclic nucleoside analogs (Fig. 6). For example, hydrogenation of syn-aminohydrins allowed access to cyclic hydrazones 76 and 77 in excellent yield over two total steps (panel a). This route compares favorably with reported syntheses of the related azafagomines 74 and 75, inhibitors of αfucosidase, that require 17 steps 56 . As depicted in panel b, a series of unusual furanose analogs 78-81 was prepared through 1,3-syn reduction of aldol adducts 36, 38, 40, and 61 with sodium borohydride followed by thermal cyclization 21 . Alternatively, reductive amination of aldol adducts 59, 60, and 62 with benzyl amine followed by reflux in toluene under basic conditions gave access to selectively protected 5′-deoxy-iminosugars 82-84 (panel c) 26 . The recent discovery of 85 as a potent and selective PRMT5 inhibitor for cancer treatment highlights bicyclic nucleoside analogs as a relatively unexplored scaffold in drug discovery 57 . Here, we prepared bicyclic nucleoside analogs 86-88 in two steps via a 1,3-syn reduction and indium chloride mediated cyclization 28 from fluorohydrins 68a, 69, and 70 (panel d). Notably, reduction of 68a afforded preferentially a syn-diol intermediate that was readily cyclized to 86 without epimerization at the anomeric center 28 . Conversely, epimerization of 87 and 88 occurred following cyclization to give a mixture of α-Dand β-Danomers as shown. Given the synthetic challenges associated with nucleoside analog synthesis, this short sequence provides new opportunities to explore structure activity relationships in this potentially important family of bicyclic nucleoside analogs. As summarized in Fig. 7, we also explored the utility of fluorinated aldol adducts to serve as precursors to fluorinated glycomimetics. Toward this goal, Julia-Kocienski olefination 58 and subsequent ring closing metathesis 59   To access 2-deoxy-2-fluoro sugars, a 1,3-syn or 1,3-anti-selective reduction of the fluorohydrin 58 was followed by removal of the silyl protecting group and oxidation of the resultant primary alcohol (panel b). This short sequence delivered the epimeric 2-deoxy-2-fluoropyranoses 94 and 95 and avoids the iterative alcohol protection-deprotection steps commonly required in fluorosugar synthesis. Hydrogenation of the Cbzprotected amino fluorohydrin 54 (made using D-proline catalysis) gave 96 directly and in excellent yield: a previously undescribed fluorinated analog of the drug migalastat (Galafold) 60 , which is a pharmacological chaperone used to treat Fabry disease (panel c).
As an additional target of interest, we also prepared a fluorinated analog of D-ribo-phytosphingosine, a precursor to the potent natural killer T-cell stimulator α-galactosylceramide 61 , in a straightforward manner through the reductive amination of the fluorohydrin 49 with benzyl amine, followed by hydrogenolysis and acid deprotection (panel d).

Conclusion
In summary, we have developed a flexible and robust DOS approach to glycomimetics that relies on achiral and readily accessible starting materials to access an array of highly relevant carbohydrate-like scaffolds. These processes are enabled by a family of new α-functionalization/aldol reactions that directly transform commercially available aldehydes and ketones into stereochemically rich and densely functionalized aldol adducts in good yield and excellent diastereo-and enantioselectivity. Notably, DFT analysis of aldol reactions of both α-chloroand αfluoroaldehydes identified a key stabilizing halogen-hydrogen (C-X···H) interaction in the lowest energy transition structure and support steric and electrostatic interactions playing key diastereodiscriminating roles in these reactions. Ultimately, the demonstration that a broad range of glycomimetics can be readily accessed using these strategies suggests these processes should inspire new efforts in drug discovery.

Methods
All experimental procedures for α-functionalization/aldol reactions and synthesis of glycomimetics are included in the Supplementary Information (Supplementary Methods).

Data availability
All characterization data including 1 H and 13 C NMR spectral data and sample spectra, IR data, [α] D , HRMS, and HPLC chromatograms used to determine enantiomeric purity are included in the Supplementary Information (Supplementary Figs. 1-80