Trans-cyclosulfamidate mannose-configured cyclitol allows isoform-dependent inhibition of GH47 α-d-mannosidases through a bump–hole strategy

Class I inverting exo-acting α-1,2-mannosidases (CAZY family GH47) display an unusual catalytic itinerary featuring ring-flipped mannosides, 3S1 → 3H4‡ → 1C4. Conformationally locked 1C4 compounds, such as kifunensine, display nanomolar inhibition but large multigene GH47 mannosidase families render specific “isoform-dependent” inhibition impossible. Here we develop a bump-and-hole strategy in which a new mannose-configured 1,6-trans-cyclic sulfamidate inhibits α-d-mannosidases by virtue of its 1C4 conformation. This compound does not inhibit the wild-type GH47 model enzyme by virtue of a steric clash, a “bump”, in the active site. An L310S (a conserved residue amongst human GH47 enzymes) mutant of the model Caulobacter GH47 awoke 574 nM inhibition of the previously dormant inhibitor, confirmed by structural analysis of a 0.97 Å structure. Considering that L310 is a conserved residue amongst human GH47 enzymes, this work provides a unique framework for future biotechnological studies on N-glycan maturation and ER associated degradation by isoform-specific GH47 α-d-mannosidase inhibition through a bump-and-hole approach.


Introduction
Selective small molecule inhibition of glycoside hydrolases ("glycosidases") has proved to be an essential tool to unlock their cellular functions.Classical inhibitor design approaches involve transition-state mimicry through charge and conformation approaches, [1][2][3][4] targeting the (catalytic) nucleophile in specic mechanisms 2,5,6 or unusual reaction mechanisms (such as the neighbouring group participation of the O-GlcNAc hydrolase 7 ).All these approaches aim to enable specic inhibition of target carbohydrate-active enzymes, which consist of 1-2% of any genome.Glycosidases are a diverse class of enzymes, with sequence-based approaches currently classifying over 160 distinct families in the CAZy database. 8This classication allows for a family-wide denition of enzyme mechanisms and may predict general conformational itineraries. 8,9nitially described by Koshland in 1953, 10 and with few exceptions, glycosidases follow two distinct reaction pathways leading to inversion or net retention of the conguration of the anomeric carbon aer hydrolysis.Retaining enzymes (typically) harness a double-displacement mechanism via the formation and subsequent breakdown of a covalent intermediate anked by an oxocarbenium-ion-like transition state, a mechanism which can be specically hijacked using covalent inactivators. 2,11,12In contrast, the one-step inverting mechanism involves an "S N 2" direct attack of water at the anomeric carbon through a single oxocarbenium ion transition-state rendering small molecule selective inhibition more challenging. 13egardless of the reaction mechanism, one serious challenge, addressed here for the inverting a-D-mannosidases, is the difficulty in specically inhibiting single members of closely related enzyme families, whose active centres are identical, and for whom all members of the superfamily are inhibited similarly by small molecules.
Decades of work on diverse glycosidases (including a-Dmannosidases, [14][15][16] reviewed by Williams and collaborators 3,17 ) has shown that the substrates undergo specic conformational uctuations to accommodate their steric and electronic features.Indeed, conformational mimicry of the ligands along the reaction coordinates, Michaelis complex, transition state or product complexes, has been demonstrated to be of key relevance when designing inhibitors for specic carbohydrate hydrolases.GH47 mannosidases are Ca 2+ -dependent metalloenzymes that follow a one-step inverting mechanism with a 3 S 1 (Michaelis complex) / 3 H 4 ‡ (transition state) / 1 C 4 (product) conformational itinerary (Fig. 1A). 18,19Kifunensine 1 is one such conformationally restrained compound that achieves selective inhibition of the GH47 a-D-mannosidase family by virtue of mimicry of the mannose conguration and 1 C 4 conformation of the enzymeproduct state (Fig. 1). 15Similarly, mannoimidazole 2 and 1-deoxymannojirimycin (DMJ) 3 also inhibit GH47 a-D-mannosidases, but do so by mimicking the oxocarbenium transition state ( 3 H 4 ‡ ) or both the Michaelis complex ( 3 S 1 ) and product ( 1 C 4 ) conformations, respectively (Fig. 1).However, all are incapable of inhibiting individual GH47 a-D-mannosidases selectively which means they cannot be used for specic inhibition of single enzymes in cells.GH47 Class I inverting a-D-mannosidases hydrolyse 1,2mannosidic linkages and are responsible for the processing of N-glycans, ultimately regulating the maturation and quality control of glycoproteins in the secretory pathway. 20GH47 mannosidases can be divided into three subfamilies within the endoplasmic reticulum (ER) and Golgi apparatus.The cleavage of a-1,2-mannoside linkages in Man9GlcNAc2-Asn substrates is initiated by ERMI (rst subfamily) followed by Golgi-a-1,2mannosidases GMIA, GMIB, and GMIC (second subfamily), hydrolysing subsequent a-1,2-mannoside branches and affording Man5GlcNAc2-Asn (Fig. 2).The third subfamily comprises of ER degradation enhancing a-mannosidase-like (EDEM) enzymes (EDEM1, EDEM2 and EDEM3) that target misfolded proteins and mark them for degradation via the ER-associated degradation (ERAD) machinery.
Given the success of generic conformational mimicry for enzymes of this family 15,19,21 (notwithstanding the fact that humans have a large multigene family of seven GH47 enzymes) we sought to test if cyclic sulfates and sulfamidates might possess similar conformational attributes.Recently, we demonstrated that a cyclophellitol analogue bearing a cis-cyclic sulfate electrophile is a selective nanomolar covalent a-glucosidase inhibitor by virtue of its 4 C 1 Michaelis complex mimicry. 2 Substitution of this cyclic sulfate by an unreactive 1,6-cyclosulfamidate yielded diverse a-glucosidase 22 and a-galactosidase 23 competitive inhibitors with great potential as enzyme stabilizers for the treatment of lysosomal storage disorders such as Pompe or Fabry disease.In this work, we sought to build upon the capability of cis-cyclic sulfates and sulfamidates to lock their compounds in a 4 C 1 chair conformation 2,22-24 and hypothesize that trans-cyclic sulfate 5 and sulfamidate 6 may instead invert this chair, yielding a new class of 1 C 4 locked conformational glycosidase inhibitors (Fig. 1B).
Here, we present the design and synthesis of an a-D-mannose congured 1,6-trans-cyclic sulfamidate as a potential inhibitor of the GH47 a-D-mannosidase family.We show, using QM calculations of free energy landscapes, that cyclosulfamidate 6 favours a 1 C 4 chair conformation consistent with the GH47 conformational itinerary.We demonstrate that a steric clash in the enzyme active centre enables the implementation of a bump-and-hole methodology for GH47 inhibition.Mutation of a leucine, conserved across the family, to serine unlocks the  dormant inhibition of 6 realising a nM inhibitor specic for the mutant enzyme only.The work, combining conformational mimicry and bump-and-hole yields the opportunity for specic inhibition of individual, but closely related, a-D-mannosidases within the GH47 family.This proof of concept offers a singular system for future biotechnological investigations into N-glycan maturation and ER-associated degradation in diverse species.

Results and discussion
We rst analysed, in silico, the intrinsic conformational preference of mannose-congured cyclic sulfates and sulfamidate 4-6, for which we employed QM metadynamics simulations to reconstruct a free energy landscape (FEL).The FELs show that opposite to the 4 C 1 conformation adopted by 1,6-cis-cyclic sulfate 4 (Fig. 3A), both trans-5 and 6 have a strong conformational preference for 1 C 4 (Fig. 3B and C).The sugar ring of 4 is quite exible, but the relaxed chair conformation ( 4 C 1 ) is the most stable, followed by B 2,5 and 1 C 4 , which are z3 kcal mol −1 higher in energy.On the contrary, the FELs of 5 and 6 show that the 1,6-trans compounds are highly conned in the southern hemisphere.Both 1 S 3 and 1 C 4 conformations are thermally accessible, but 1 C 4 is the most stable.Given the preference for this conformation, we next sought to synthesize and establish whether, similar to kifunensine, the manno-congured 1,6trans-cyclic sulfate 5 and/or sulfamidate 6 would act as a GH47 a-D-mannosidase inhibitor.
1,6-Cis-manno-cyclosulfate, 4, and 1,6-trans-mannocyclosulfate, 5, were synthesized from mannose-congured cisdiol and trans-diol, respectively (ESI, Scheme S1 †).Though the nal compounds 4 and 5 could be characterized, their storage for a week at room temperature led to compound degradation probably by intramolecular attack of the 2-OH pseudoanomeric position and opening of the cyclic sulfates.The synthesis of the potentially more stable cyclosulfamidate 6 started with the addition of sodium azide to tetra-O-benzyl-cyclophellitol 7 which resulted in a mixture of diastereomers 8 and 9. Azide 9 was subsequently reduced using PtO 2 to obtain amine 10.Cbzprotection of the amine gave intermediate 11 which was treated with SOCl 2 followed by RuCl 3 /NaIO 4 -mediated oxidation of the formed sultes, resulting in the formation of fully protected trans-sulfamidate 12. Trans-sulfamidate 12 was then exposed to hydrogenation conditions to obtain cyclosulfamidate 6 which proved to be chemically stable (Scheme 1).
To our surprise, no signicant inhibition of the GH47 model enzyme (whose active centre is identical to the human enzymes) from Caulobacter K31 strain was observed even at high inhibitor concentrations; 82% of a-1,2-mannobiose was hydrolysed by the WT GH47 mannosidase to mannose aer the addition of 1 mM 6 (Fig. S4 †) (in contrast to kifunensine (K D of 39 nM), 15 mannoimidazole (K D of 47 nM), 19 and DMJ (K D of 481 nM) 21 ).Accordingly, crystal soaks revealed no binding.Simple overlay of 6 over published complexes of the Caulobacter GH47 enzyme revealed a likely steric clash between one of the oxygen atoms of sulfamidate group of 6 with the Cd2 atom of the leucine side chain L310 in the active centre of the enzyme (Fig. S5 †).
Although this initial observation was fortuitous, it immediately presented a solution to the problem of specic inhibition of individual enzymes.In contrast to other known GH47 inhibitors, we could exploit the clash to formulate a "bump-andhole" strategy allowing selective individual a-mannosidase GH47 inhibition, otherwise not possible through chemical knockdown.Originally developed for kinases by Shokat, 25,26 bump-and-hole engineering is best applied to large multigene families where isoform-specic inhibition (sometimes described as "allele specic", reecting the inhibition of one of a panel of closely related proteins encoded by closely related genes) is challenging.This strategy relies on introducing a "bump" on the inhibitor/ligand/substrate that is accommodated specically by enzyme variants into which a complementary "hole" has been created through mutagenesis.Within the context of glycoscience, while Karanicolas's team expanded the approach to incorporate new allosteric pockets into the catalytic sites of several glycosidases, 27,28 Hou et al. capitalized on the bump and hole strategy to precisely deliver nitric oxide using alkylated b-galacatosyl NONOates 29 and Schumann et al. incorporated chemically tagged sugars into the cell surface glycome of living cells using "holed" glycosyltransferases. 30,31 Since, despite conformational matching, 6 failed because of encountering a "bump" with L310, we hypothesized that exchanging the leucine for a smaller amino-acid would form an appropriate "hole" to accommodate 6 in a variant-specic manner.Importantly, L310 is an entirely conserved residue across all seven human GH47 enzymes (Fig. S7 and S8 †).To demonstrate proof-of-concept, mutation to a serine was selected because of the potential to form a hydrogen bond to the inhibitor in addition to removing the steric clashing distance: a hole with benets.
The L310S variant was created and shown to be catalytically viable in the degradation of a-1,2-mannobiose (Fig. 4A, B and Table S1 †).The V max value of 0.15 mM s −1 and the K M value of 403 mM for the mutant were 10× and 6× lower than those of the WT. 32,33Similar loss in enzyme catalytic activity has also been observed in protein kinase bump and hole approaches. 25,26uilding on this kinetic assay, we obtained an inhibition constant by pre-incubation of the enzyme and 6 for 30 minutes and then following the same time-point procedure as for the Michaelis-Menten assay.Following the approach described by Suits et al., 34 a 574 nM K i was obtained for 6 with the L310S mutant specically (Fig. 4C and Table S1 †).Similarly, a dissociation constant of 970 nM was obtained using isothermal titration calorimetry (Fig. S6 †).To understand the interactions between 6 and CkGH47 L310S regarding the introduced hole, a crystal structure complex was obtained (statistics in Table S2 †).Trans-cyclic sulfamidate 6 binds similarly to other conformationally restricted GH47 inhibitors, with additional interactions between the sulfamidate and active site residues, for example, the nitrogen and oxygen, and D249 and R363 side chains.Additionally, the introduced serine residue is indeed now in a position to hydrogen bond to the oxygen of the sulfamidate with a distance of 3.1 Å.Also validating the initial design hypothesis, overlaying the L310S structure with 6 with  that of the WT CkGH47 native revealed 1.3-1.9Å steric clashes of 6 to leucine 310 (Fig. 5B and S5 †).
Finally, selectivity versus a panel of related exomannosidases was investigated.We rst tested inhibition against all ve human GH38 retaining a-mannosidases in overexpressed cell lysates using 4-methylumbelliferyl-a-mannoside (4-MU-a-man): Golgi mannosidase II and IIx (MAN2A1 and MAN2A2), cytosolic MAN2C1 and lysosomal MAN2B1 and MAN2B2.This a-mannosidase family follows a different conformational itinerary than GH47: 0 S 2 / B 2,5 / 1 S 5 . 35Cyclic sulfamidate 6 showed no inhibition of this panel of enzymes up to 1 mM with the exception of MAN2C1, for which an apparent IC 50 value of 3.5 mM was observed when using cobalt as the metal ion (Table 1).Intriguingly, no IC 50 value could be determined for MAN2C1 when the buffer was supplemented with 1 mM ZnCl 2 instead of CoCl 2 , and compound 6 showed no activity in the other tested GH38 a-mannosidases when the buffer was enriched with CoCl 2 .Of note, the presence of cobalt has been associated with conformational changes in the active site of some GH38 a-mannosidases, which might affect the binding interaction with inhibitors. 36Nevertheless, additional research is required to explore which specic metal ion is present in the MAN2C1 active site in its in vivo state.As expected, when looking at the representative GH2 a-mannosidases Bacteroides thetaiotaomicron BtMan2A, compound 6 was also inactive at 1 mM concentration.

Conclusions
"Isoform-specic" inhibition of single members of multigene families is a major challenge in chemical glycobiology.Here we use the dual approach of conformational mimicry with bump and hole to establish a nanomolar inhibitor-enzyme pair: 1,6trans-cyclic sulfamidate 6 and an L310S variant that allows selective inhibition of this specic a-1,2-mannosidase.][39]

Fig. 5
Fig. 5 Structure of CkGH47 L310S in complex with 6. (A) Active site residues (dark red) within hydrogen bonding distance to 6 (pink) in the −1 subsite.The maximum-likelihood/sA-weighted 2F obs − F calc map, shown in purple, is contoured at 1.3 e Å −3 .(B) Superposition of CkGH47 WT crystal structures highlighting the flexibility of residue 310; native leucine is shown in gold (PDB 4AYO) and CkGH47 in complex with kifunensine is shown in orange; there are 2 alternate conformations of Leu at 50% occupancy each (PDB 5NE5).