Conformational Analysis of the Mannosidase Inhibitor Kifunensine: A Quantum Mechanical and Structural Approach

The varied yet family‐specific conformational pathways used by individual glycoside hydrolases (GHs) offer a tantalising prospect for the design of tightly binding and specific enzyme inhibitors. A cardinal example of a GH‐family‐specific inhibitor, and one that finds widespread practical use, is the natural product kifunensine, which is a low‐nanomolar inhibitor that is selective for GH family 47 inverting α‐mannosidases. Here we show, through quantum‐mechanical approaches, that kifunensine is restrained to a “ring‐flipped” 1C4 conformation with another accessible, but higher‐energy, region around the 1,4B conformation. The conformations of kifunensine in complex with a range of GH47 enzymes—including an atomic‐level resolution (1 Å) structure of kifunensine with Caulobacter sp. CkGH47 reported herein and with GH family 38 and 92 α‐mannosidases—were mapped onto the kifunensine free‐energy landscape. These studies revealed that kifunensine has the ability to mimic the product state of GH47 enzymes but cannot mimic any conformational states relevant to the reaction coordinate of mannosidases from other families.

The varied yet family-specific conformational pathways used by individual glycoside hydrolases (GHs) offer at antalising prospectf or the design of tightly binding and specific enzyme inhibitors. Ac ardinal example of aG H-family-specific inhibitor, and one that finds widespread practical use, is the natural product kifunensine, which is al ow-nanomolar inhibitor that is selectivef or GH family 47 inverting a-mannosidases. Here we show,t hrough quantum-mechanical approaches, that kifunensine is restrained to a" ring-flipped" 1 C 4 conformation with another accessible, but higher-energy,r egion aroundt he 1,4 B conformation.T he conformationso fk ifunensine in complex with ar ange of GH47 enzymes-including an atomic-levelr esolution (1 )s tructure of kifunensinew ith Caulobacter sp. CkGH47 reportedh erein and with GH family 38 and 92 a-mannosidases-werem apped ontot he kifunensine free-energy landscape. These studies revealed that kifunensine has the ability to mimic the product state of GH47 enzymesb ut cannot mimic any conformational states relevant to the reactionc oordinate of mannosidases from other families.
There is compelling evidence that the enzymatic hydrolysis of glycosides, catalysed by glycoside hydrolases (GHs) or glycosidases,o ccurs via transition states with significant oxocarbenium ion character.F or pyranoside-active enzymes,S innott was the first to argue that the allowed canonical conformationso f the transition state sugar ring were two half-chairs ( 4 H 3 and 3 H 4 ;a nd their closely relatede nvelopec onformations)a nd two boats ( 2,5 B and B 2,5 ). [1] Sustained efforts to map the conformational reaction pathways of glycosidases leadingf rom substrate to product via the transition state(s) have revealed that individual glycosidases are optimised to act on substrates and follow ad efined conformational itinerary through as pecific transition-state conformation. [2] Glycosidases that are sequence related (for family classification see www.cazy.org and www.cazypedia.org) [3] and act on sugars with the same configuration are believed to act with identical conformational reaction itineraries. It is also apparent that GHs from different families that act on substrates with the same stereochemical configuration can follow different conformationali tineraries during catalysis. Given that mimicryofthe enzymatic transition state is apowerful approach to inhibitor design and enzymei nhibition, [4] the potentiale xists for the design or discoveryo fm olecules with intrinsically biased conformationst hat could act as GH-familyspecific enzymei nhibitors. However,t oa chieves pecificity, inhibitor designd oes not have to be limited to transition-state mimicryb ut could target any distinctive conformations tate along the reactionc oordinate. Although highly appealing,e fforts to design molecular scaffolds that predispose inhibitors to conformations matching that of the transition state of a specific GH have been disappointing. For example, molecular constraints that restrict conformational mobility typicallyr esult in steric clashes that prevent efficient binding. [5] Moreover,a recent study that identified au nique stereoelectronic bias of mannopyranosiduronici minosugars failed to deliver effective inhibition of ac onformationally matched a-mannosidase, presumably because of the inability to accommodate the requisite carboxylate group. [6] In the context of the failure thus far of rational designm ethods to achievec onformationally specific inhibition, one compound from nature, kifunensine, stands out as ap owerful andf amily-specific enzyme inhibitor. Kifunensine (1;S cheme 1) is an unusual oxalamide-fused iminosugar with high specificity for a-mannosidases of GH47.
Kifunensine is produced by the actinobacterium Kitasatosporia kifunense strain no. 9482. [7] Originally isolated on the basis of immunomodulatory properties, it wass oon identified as ap otent inhibitor of selected a-mannosidases. [ contemporary applicationso fk ifunensine stem from its activity as as pecific inhibitor of the class IG H47 a-mannosidases involved in glycoproteinb iosynthesis within the secretory pathway that is specific for the hydrolysis of a-1,2-glycosidic bonds. [9] During glycoprotein biosynthesis, glucosylated high-mannoseN-glycans are appended to nascent unfoldedp eptide chains in the endoplasmic reticulum( ER), whereupon they undergo ar ange of trimming reactions in the ER that assist in folding the peptide,a nd subsequent trimming reactions in the Golgi apparatus that remove additional mannose residues prior to late-stage glycosylation reactions. [10] As part of this process,aquality control mechanism termed ER-associated degradation (ERAD) extracts terminally misfolded proteins from the secretory pathway for proteosomal degradation. [11] Kifunensine-sensitive a-mannosidases are found within both the normalt rimming pathway( ER mannosidase I, Golgi mannosidase I), andw ithin the ERAD pathway( ER degradation-enhancing mannosidase-like proteins 1, 2and 3). [12] The powerful and specific inhibition of GH47 a-mannosidases by kifunensine has led to its widespread use for manipulating the N-glycan structure. In the structural-biology context it is used to improve the homogeneity and crystallisation of proteins by arresting glycan remodelling so as to yield high-mannose glycanst hat are more easily cleaved by endoH. [13] Kifunensine is also used in the production of therapeutic proteins. The effectiveness of acid b-glucocerebrosidase as at reatment for the lysosomal storage disorderG aucher's disease depends upon the presence of high-mannose N-glycans of this protein, which enable ligation to mannoser eceptors and delivery to lysosomes. [14] The lysosomal-replacement-therapy protein Velaglucerase alfa (acid b-glucocerebrosidase) is produced by Shire Plc by culturing HT1080 fibrosarcoma cellse xpressing acid bglucocerebrosidase in the presence of kifunensine-this promotes the biosynthesis of enzymed ecorated with immature high-mannose-type N-linked glycan chains. [15] Structural studies on GH47 enzymes,f irstly the seminal work by the Howellg roup on the human ER ManB1 a-1,2-mannosidase, [16] and subsequent work on a Penicillium citrinum homologue [12c] revealed that kifunensine binds in a" ring-flipped" 1 C 4 conformation. Thisc onformation is unusualf or an iminosugar. Relative to the proposed conformational pathway 3 16,17] for this family of enzymes, it pro-vides conformational mimicryo ft he product. Of relevance to this observation, ac omplex of noeuromycin 2 bound to ab acterial a-mannosidase of family GH47, Caulobacter sp. K31 (CkGH47) also adopted a 1 C 4 conformation that, on the basis of computational work, was assigned as ap roduct-mimicking species. [17a] However, in the case of noeuromycin,computational analysis of the inhibitor reveals that it favours a 4 C 1 conformation, [18] and the observedc onformation on-enzyme would therefore appear to be ac onsequenceo ft he enzyme restricting its shape to ah igher-energy conformation. Because of the widespreada pplicationo fk ifunensine, and its unique specificity as aG H47-selective a-mannosidase inhibitor,w esought to understand its specificity by developingaquantitative view of the conformational preferences of this compound when isolated, relative to bound states on a-mannosidases from GH47 and other families.
Structures are available for kifunensine bound to Homo sapiens [16] and P. citrinum [12c] family 47 a-mannosidases at medium resolutions of 1.75 and 2.20 ,r espectively.I no rder to allow comparison to conformations that the a-mannoside substrate follows during catalysis, we sought to obtain ah igher-resolution structure of 1 with CkGH47 a-mannosidase, which we have previouslys hown to provide atomic resolution diffraction data (Table1). Kifunensine, shown here using isothermal titration calorimetry,i satight binding (K D = 39 nm)i nhibitor of CkGH47 ( Figure 1A). Crystals of ac omplex of CkGH47 bound to kifunensine diffracted to 1.05 resolution ( Figure 1B). Kifunensine binds to CkGH47 in a 1 C 4 conformation, in line with Scheme1.Assortedg lycosidase inhibitors.
Scheme2.Conformation itineraries for inverting a-mannosidases proceeding through left: O S 2 ![B 2,5 ]°! 1 S 5 (e.g.,GH125) [24] and right: 3 S 1 ![ 3 H 4 ]°! 1 C 4 (GH47)pathways. Catalytic residues actinga sp roton donors/acceptors are omitted for clarity. Conformational free-energy landscapes (FELs) are quantitative maps of the energy of the full suite of conformationso f am olecule, and provide insight into locallya nd globally stable conformations as well as the barriers that must be crossedi n order to achievet hem. In order to understand the conformational landscape of kifunensine, and how this contributes to its GH47 specific inhibition, its conformational FEL was calculated by ab initio metadynamics [19] according to methods we have developed and applied to ar ange of glycosidase inhibitors. [18,20] The kifunensine FEL (Figure 2A)e xhibits as trong preference for the "southern hemisphere" 1 C 4 conformation; this is consistentw ith the conformation observed in as mallmolecule, single-crystal X-ray structure. [8,21] Al ow energy zone is also seen aroundt he 1 S 3 / 1,4 B/ 1 S 5 region of the FEL, which, at approximately 7-8 kcal mol À1 higheri ne nergy and with am inimal barrier, can be considered energetically accessible. Overall, the isolated kifunensine FEL is strongly constrained to ar ingflipped 1 C 4 conformation, with al imited ability to adoptasmall set of "equatorial" conformations. Plottingt he conformations of kifunensine observedi nr epresentative GH47c rystal structures (for enzymes from H. sapiens and Bacteroides thetaiotaomicron)o nto the isolated kifunensine FEL reveals that they lie close together in the low energy zone, with similar conformations to that of the small-molecule, single-crystal X-ray structure, thus suggesting that little distortion occurs upon binding to GH47 enzymes-theapparent differences near the poles are ac onsequence of the deformation of the Mercator projection (for aP olar projection and direct comparison, see Figure S1 in the Supporting Information). In concordancew ith the FEL for 1,a nF EL calculated for methyl a-mannosidew ithin the activecentre environment of the CkGH47 ( Figure 2B)r evealed that this molecule is heavily distorted away from its preferred 4 C 1 conformation when bound to the enzyme. [17a] The conformational bias of kifunensine towards the southern hemisphere is unusual for ag lycosidase inhibitor.F or example, FELs of the azasugari sofagomine revealt hat this molecule favours a" normal" 4 C 1 conformation, [20b] which is also expected for the iminosugar deoxymannojirimycin 4.Arange of iminosugars containing sp 2 -hybridised atoms,b est illustrated by mannoimidazole 5,favour half-chair ( 4 H 3 )and boat (B 2,5 )conformations that lie along the FEL tropic/equator. [20] One unusual example worth mentioning that is similar what is observed here is am annopyranosiduronici minosugar that, under acidic conditions, favours a 1 C 4 chair,a ss hown by 1 HNMR analysis of vicinal coupling constants;h owever,t he presence of the carboxylate group preventsb inding to a-mannosidases. [6] Although kifunensine typicallyb inds to GH47 a-mannosidases with nanomolar dissociation constants, it is usually af ar poorer binder/inhibitor of GH families other than family 47. The available evidences uggests that all other exo-mannosidases operate through O S 2 $[B 2,5 ]°$ 1 S 5 conformational pathways [GH families 2( retaining), [5b] 38 (retaining), [22] 92 (inverting), [23] 125 (inverting) [24] and 130 (inverting) [25] ]. Data are available for kifunensine binding to representatives of families 38 and 92. For the Drosophila melanogaster Golgi class II GH38 a-mannosi-  dase, dGMII, 1 has a K I value of 5mm, [26] and it is reported to be avery poor inhibitor of jack bean GH38 a-mannosidase. [8, 12a] Similarly,f or ar ange of B. thetaiotaomicron GH92 a-mannosidases, 1 is reported to have K I values in the range of 100-200 mm. [23] Structures of 1 bound to GH38 (PDB ID:1 PS3) and GH92 (PDB ID:2 WVZ) [23] a-mannosidases in both cases revealed a 1,4 B conformation. Consideration of these results in the context of the kifunensine FEL reveals that, although this conformation lies in the energetically accessible equatorial region (Figure 2A), it also lies some distance away from the O S 2 $[B 2,5 ]°$ 1 S 5 conformational pathway proposed for both GH38 and GH92 a-mannosidases;t his suggests that the poor inhibition results from an inability of the inhibitort oa dopt ac onformation that matches speciesf ormed on the enzyme duringc atalysis.
Our conformational analysis of kifunensine 1 highlights the unique ability of this compound to target ar egion of the FEL that is not accessed by other GH inhibitors. We suggest that the specificity and potency of 1 for GH47 family enzymesi s ad irect consequence of its strong preference for the southern hemisphere 1 C 4 conformation, which matches that of the product state of the 3 S 1 ![ 3 H 4 ]°! 1 C 4 conformational itinerary that this family of enzymes is believed to follow. [2b, 16,17] We highlight that this analysiss uggests that potency is achieved in the absence of transition state mimicry.I ti ss urprising that this conformational restraint is achieved by introducing an oxalamide bridge without interfering with binding in the active site, as ap reviousa ttemptw ith as imilar goal to synthetically introduce ab ridge into an iminosugar failed owing to steric clashes in the actives ite.
[5] The preference of kifunensinef or a 1 C 4 conformation likely arises not merely from the fusion of the bridge to the ring, but also from the sp 2 hybridisation of the endocyclic nitrogen as part of an amide. Further,w en ote that kifunensine is an eutral species; this showst hat, unlike mosta za/ iminosugar glycosidase inhibitors, kifunensinea chieves its potencyt hrough shape, rather than charge. For ad iscussion of shape versus charge mimicry for the inhibition of aG H99 amannosidase see ref. [18];f or an example of ap otent class of neutralGHi nhibitors see ref. [27].
In conclusion, this work highlights the capacity of natural products to providei nspiration for selectively inhibiting glycosidases based on mechanistic principles. Aligned with this goal, we highlight the inspiration provided by the natural product nagstatin (6), [28] which informed the development of the concept of lateral protonation by GHs with ac atalytic acid or acid/base located anti to the C1ÀO5 bond, and the design of the glycoimidazole-class inhibitors that are selective for antiprotonating GHs. [29] The existence of kifunensine as aG H47-selective inhibitor should inspire continuing effortst od evelop selectivea nd potent GH inhibitors based on targeting unique conformational features of their catalysis reaction coordinates. We highlight that one other mannosidase family operates through ar eversed 1 C 4 ![ 3 H 4 ]°! 3 S 1 conformational itinerary, namely GH134 b-mannanases. [30] Based on the analysis presented herein, this familyislikely to be specifically targeted by substrate-mimicking kifunensine-derived oligosaccharides in contrast to the b-mannanases of GH26 and GH113, which instead operate through O S 2 $[B 2,5 ]°$ 1 S 5 conformational pathways. [20b] Experimental Section CkGH47 protein was cloned, expressed and purified as described previously. [17a] The crystallisation conditions were the same as discussed in ref. [6].M ature crystals were soaked in kifunensine (1 mm)f or 15 h. Data were collected on the Diamond I04-1 MX beamline. After data collection, the diffraction images were integrated by using XIA2 [31] and reintegrated by using AIMLESS from the CCP4 software suite. [32] The Free Rf lag data set was copied from PDB ID:4 AYO. The model was refined by using multiple rounds of REFMAC and manual model building in COOT. [33] Water molecules were added by using FIND WATERS in COOT and validated. The conformation of the ligand was validated by using PRIVA-TEER. Coordinates have been deposited in the PDB with ID 5NE5; details of refinement quality are shown in Ta ble 1a nd the PDB header.F igures of the structure were produced by using CCP4mg. [34] ITC was performed by using aM icroCal ITC 200 calorimeter at 25 8C with 20 injections. CkGH47 and kifunensine 1 were transferred into matching buffer by dialysis into HEPES (25 mm,p H7.0) containing NaCl (50 mm)a nd CaCl 2 (2 mm). The CkGH47 concentration in the cell was 50 mm,a nd the ligand concentration was 500 mm.T he binding affinity was calculated by using Origin (OriginLab, Northampton, MA).
The free-energy landscape of kifunensine was obtained by density functional theory-based metadynamics [19] by using the Car-Parrinello (CP) method. [35] The molecule was enclosed in an isolated cubic box of 14.0 14.0 14.0 .Af ictitious electron mass of 700 au and at ime step of 0.12 fs ensured ap roper conservation of the total energy during the simulation. The Kohn-Sham orbitals were expanded on ap lane wave (PW) basis set with ak inetic energy cutoffo f7 0Ry. Ab initio pseudopotentials, generated within the Troullier-Martins scheme, [36] were employed. The Perdew,B urke and Ernzerhoffg eneralised, gradient-corrected approximation (PBE) [37] was selected in view of its good performance in our previous work. [38] Twoc ollective variables of the puckering coordinates of Cremer and Pople (q and f)w ere used to explore the conformational space. [39] Initially,t he height/width of these Gaussian terms was set at 0.6 kcal mol À1 /0.10 rad, and an ew Gaussian-like potential was added every 250 MD steps. Once the whole free-energy space had been explored, the height of the Gaussian terms was reduced to half of its initial value (0.3 kcal mol À1 ), and an ew Gaussian-like potential was added every 500 MD steps. The simulation was stopped when energy differences among wells remained constant. The phase space was fully explored in less than 50 ps, and the simulation was further extended up to 95 ps. The error in the energy difference of the principal minima, taken as as tandard deviation (SD) from the last 30 ps, is below 0.6 kcal mol À1 .