Main

Gaucher disease (GD), the most common lysosomal storage disorder, is caused by mutations in acid β-glucosidase (glucocerebrosidase, GCase, locus GBA), a peripheral membrane protein that catalyzes the hydrolysis of glucosylceramide (2) to β-glucose (3) and ceramide (4; Fig. 1a, top) in the presence of the modulator protein saposin C and lipid. Mutations in GCase lead to diminished biological activity because of impaired trafficking, altered enzyme stability and/or defective intrinsic activity1,3. Consequential accumulation of glucosylceramide in macrophages leads to proinflammatory responses and altered glycosphingolipid distribution4, which results in hepatosplenomegaly, abnormal bone turnover, pulmonary hypertension and, in some cases, central nervous system (CNS) disease5. In addition to glucosylceramide accumulation, protein accumulation may also activate cellular stress responses that predisposes vulnerable cells to pathological events such as the deposition of Lewy bodies6. In support of this hypothesis, recent epidemiological studies have reported that GD mutations are risk factors for Lewy body diseases7.

Figure 1: Effects of IFG on GCase activity and trafficking in primary N370S GCase fibroblasts.
figure 1

(a) Schematic of the reaction performed by GCase and the chemical structure of IFG. (b) Inhibition curve for IFG with imiglucerase. Error bars are s.d. (c) Dose response for N370S GCase fibroblasts treated with IFG for 5 d, reported as arbitrary fluorescence units (F460) per microgram of protein per hour. We compared values from lysates of IFG-treated cells to those of untreated control cells using a two-tailed Student's t-test assuming equal variances. *P < 0.05. Error bars are s.e.m. (d) Fluorescence intensities (grayscale) from untreated and IFG-treated wild-type (WT) and N370S GCase primary fibroblasts labeled with anti-GCase (top panels; scale bar 53 μm for all top panels) and dual labeling of fibroblasts with antibodies against GCase (green) and the lysosomal marker Lamp1 (red). Overlap (yellow) indicates colocalization of GCase and Lamp1 (bottom panels; scale bars from left to right are 30 μm, 26 μm, 53 μm and 53 μm). Proliferation of Lamp1-positive structures evident here in N370S fibroblasts has been previously observed in cells from GD-afflicted individuals30. (e) Comparison of IFG and CBE treatment on GCase localization in the lysosomes in primary fibroblasts. Immunofluorescence of IFG (100 μM; 5 d) or CBE (100 μM; 5 d) treated and untreated wild-type and GCase N370S fibroblasts with anti-GCase (green) and Lamp1 (red). Overlap (yellow) indicates colocalization of GCase and Lamp1. Scale bars from left to right are 27 μm, 26 μm, 30 μm and 53 μm.

Full size image

A number of missense mutations in the GBA gene locus are associated with GD; the two most prevalent are those resulting in the amino acid substitutions N370S and L444P in GCase. People with type 1 GD, who are often homozygous for the alleles resulting in N370S GCase, show multiple organ involvement but generally lack the neuropathologies found in type 2 and type 3 GD, such as those manifested in individuals with the L444P mutation. These mutations do not act by removing a key catalytic residue in the GCase active site, but likely by destabilizing the native conformation, thereby rendering the protein more susceptible to mistrafficking and degradation8.

The main treatment for type 1 GD is enzyme replacement therapy (ERT) using recombinant human GCase (imiglucerase, Cerezyme). Disadvantages of ERT include regular intravenous infusions, little direct effect on the CNS-affected variants of GD (types 2 and 3)1, and high cost. An alternative to ERT, substrate reduction therapy (SRT), works by inhibiting glucosylceramide synthesis with N-butyldeoxynojirimycin (NB-DNJ, miglustat, Zavesca; 5) but has a lower therapeutic index9. In addition, NB-DNJ results in adverse side effects10. Both therapeutic approaches address substrate accumulation but not the potential contributions of mutant GCase proteotoxicity and mistrafficking to the pathogenesis of GCase.

An emerging strategy for GD therapy is the use of small molecules that stabilize mutant GCase and consequently restore trafficking and activity2. This approach, called pharmacological chaperoning, is also being evaluated as a therapeutic strategy for other lysosomal enzymes, including α-galactosidase A (mutated in Anderson-Fabry disease) and β-hexosaminidase (Tay-Sachs disease)2,11, and for noncatalytic proteins11. For GD, alkyl derivatives of various iminosugars1,12 and valienamines13 elevate the activity of mutant GCases in fibroblasts and transfected cell lines of affected individuals, although derivatives with alkyl tails longer than four carbons are cytotoxic14. Most compounds being evaluated as potential chaperones for GD are postulated to act as noncovalent active site inhibitors2,15. The mechanism by which chaperones promote GCase stability, trafficking and cellular activity is not understood on a molecular level. Such details will likely have an impact on the development of chaperone strategies for other diseases as well.

Isofagomine (IFG; Fig. 1a, bottom) is a potent inhibitor of wild-type GCase with a half-maximal inhibitory concentration (IC50) of 59 ± 1 nM (s.d.) (Fig. 1b). Despite the inhibitory action of IFG, treatment of N370S GCase fibroblasts (DMN89.15) with IFG for 5 d increases GCase activity ∼two-fold in cell lysate assays, in a dose-dependent fashion (Fig. 1c), which suggests that IFG is able to increase cellular GCase concentrations. Treatment of N370S GCase mutant fibroblasts with IFG elevates GCase immunofluorescence levels throughout the cell (Fig. 1d, top), and colocalization of GCase with Lamp1, a lysosomal marker (Fig. 1d, bottom), indicates that IFG also restores trafficking of N370S GCase to lysosomes. These results, corroborated by additional cell-based studies that have recently been reported16, demonstrate that IFG can act as an effective pharmacological chaperone for N370S GCase. By contrast, the covalently modified inhibitor conduritol B epoxide (CBE, 6) does not improve enzyme concentrations or trafficking (Fig. 1e).

To investigate the molecular interaction of IFG with GCase, we crystallized partially deglycosylated GCase and solved structures of GCase at pH 7.5 (neutral; 2.3-Å resolution) and at pH 4.5 with IFG (inhibited, inh; 2.1-Å resolution) and without it (acidic; 1.8-Å resolution) (see Methods, Supplementary Table 1 online and Fig. 2). The polypeptide backbones of the inh and acidic GCase structures are very similar to each other, to the two conformations of GCase observed at pH 7.5 (neutral), to the previously published pH 4.5 structures of GCase (Protein Data Bank (PDB) codes 1OGS (ref. 17) and 2F61 (ref. 8)) and to the structure of GCase covalently modified with CBE at pH 4.5 (PDB code 1Y7V)18. The r.m.s. deviations for Cα atoms are 0.6 Å2 or better, which indicates that inhibitor binding or change in pH does not change the overall structure of the enzyme. Both the triosephosphate isomerase (TIM) barrel and immunoglobulin domains superimpose well.

Figure 2: Superposition of GCase under different conditions.
figure 2

(a) Stereoview of inh (slate) superimposed with acidic (yellow). (b) Stereoview of inh (slate) superimposed with similar conformation observed in neutral (light blue). IFG is shown as ball-and-stick model in both a and b. (c) Stereoview of acidic (yellow) superimposed with similar conformation observed in neutral (brown). Glycerol is shown as ball-and-stick model. (d) Superposition of loop 1 from all independent copies of GCase in this study. Two IFG-bound inh copies are colored slate, two similar conformations observed in neutral are colored light blue, four copies of acidic are shown in yellow, two copies of glycerol-bound GCase from inh are colored beige, and the two copies of neutral that are similar to acidic are in brown.

Full size image

Our low-pH GCase crystals differ from those published previously8,17,18 in that they belong to a lower symmetry space group, and the acidic structure diffracts to higher resolution. Our acidic structure also contains glycerol in all four copies of the active site in the asymmetric unit (Fig. 3a), in place of a sulfate anion and two water molecules17 or ethylene glycol8. Glycerol or ethylene glycol molecules were likely introduced upon cryoprotection of the crystals. In the inh structure, IFG is bound in the active site of two of the four copies of GCase in the asymmetric unit, and glycerol is present in the remaining two monomers. The neutral structure is the first of GCase at pH 7.5 and is also the first truly apo structure (Fig. 3b,c). Of the four molecules in the asymmetric unit, two are nearly identical to those observed for IFG bound in inh (Fig. 2b), and the remaining two are similar to those in the acidic structure (Fig. 2c). Thus, the conformations observed at the low pH of the lysosome are likely similar to those sampled upon protein folding in the endoplasmic reticulum at neutral pH.

Figure 3: The GCase active site.
figure 3

(a) Ball-and-stick representation of glycerol-bound acidic active site. (b) Ball-and-stick representation of active site seen in inh-like conformation in neutral. (c) Ball-and-stick representation of active site seen in acidic-like conformation in neutral. (d) Ball-and-stick representation of IFG-bound inh active site. Difference (Fo − Fc) electron density (green) is contoured at 3σ and was calculated using only respective GCase coordinates. Asn396 and Trp381 are omitted for clarity. Hydrogen bonding interactions to ordered water molecules are indicated by gray dashed lines. (e) Schematic diagram of hydrogen bonding interactions involved in stabilizing IFG in the active site of GCase. Distances are in Å.

Full size image

In the inh structure, the chaperone IFG is bound in the GCase active site in a well-ordered, distorted chair conformation, and is held in place by extensive hydrogen bonding interactions (Fig. 3d,e). The hydroxyl groups of IFG interact with Asp127, Trp179, Trp381 and Asn396, and the imino group is stabilized by Glu235 and Glu340. In the GCase reaction, Glu340 is the catalytic nucleophile and Glu235 the general acid/base18,19. When glycerol is bound (acidic structure and two monomers of inh), it is stabilized by hydrogen bonding interactions with similar residues (Fig. 3a), excluding Asn396 but including Asn234. The main difference between the GCase active site occupied by IFG and that occupied by glycerol is the position of Tyr313, a residue that is not involved directly in binding to either compound (Fig. 3a,d). In the acidic structure, Tyr313 is hydrogen bonded to Glu235 and acts as a gate to keep the active site closed off (Fig. 4a), but in GCase with IFG bound, Tyr313 is hydrogen bonded to Glu340 in the active site and to Cys342 in the second sphere of residues surrounding the active site. This change widens the opening for solvent molecules or substrate to reach the active site and makes the surface gradient leading to the active site shallower (Fig. 4b,e; see below).

Figure 4: Surface representation of GCase near the active site.
figure 4

(a) Glycerol-bound acidic. (b) IFG-bound inh. (c) inh-like configuration in neutral. (d) acidic-like configuration in neutral. (e) Glucosylceramide-docked inh. Glucosylceramide has been modified by truncating its alkyl chains. Glycerol, IFG and modified glucosylceramide are presented in ball-and-stick models (in a,b and e, respectively).

Full size image

In the active site of apo-GCase at neutral pH (Figs. 3b,c and 4c,d), the four copies of the active site in the asymmetric unit are slightly different from one another with respect to the orientation of Tyr313 and to the positions of Asn396 and Phe397. In the two GCase monomers that are similar to the inh structure in the asymmetric unit, Tyr313 swings out from the active site and is found in two positions, one in each monomer. In the first monomer (Fig. 3b), the hydroxyl group of Tyr313 is 2.5 Å from the backbone carbonyl of Gly344, and clear 2Fo − Fc electron density is observed. In the second copy (not shown), Tyr313 appears well ordered and pointed toward Phe397, and the electron density for the nearby Gly344 and surrounding loop residues is poor. In the monomers similar to the acidic structure (Fig. 3c), Tyr313 is found in a single orientation, within hydrogen bonding distance of both the catalytic and the acid/base residues in the active site. Entry to the active site is partially blocked as a result of these changes (Fig. 4c,d), though less so than when glycerol is bound at pH 4.5 (Fig. 4a). These new structures of apo-GCase at pH 7.5, enabled by cryoprotection with Li2SO4 instead of glycerol, suggest that the active site is not preorganized for substrate binding when it is folded in the endoplasmic reticulum. In addition, the previously unobserved flexibility of Tyr313, and its ability to modulate access to and alter its hydrogen bonding interactions among residues in and near the active site, is likely important for catalysis and therefore warrants further investigation.

Although GCase remains relatively unaltered upon IFG binding and pH change, critical movements are observed for two loops at the mouth of the active site; these loops are GCase311–319 (loop 1) and GCase342–354 (loop 2) (Fig. 2a). Changes in an additional loop in the remote immunoglobulin domain, comprising GCase55–65, result directly from crystal packing and do not affect the active site. The configurations observed for the loops near the active site in the available crystal structures affect the shape and landscape of the active site.

The most notable difference observed upon IFG binding at low pH is a helical turn introduced in loop 1, GCase312–317 (Figs. 2a and 5), that affects the surface topology (Fig. 4a,b) and interactions among amino acid residues within core regions of GCase. An important residue involved in stabilizing the helical turn via a new hydrogen bonding network is Asn370 (see below, Fig. 5). The r.m.s. deviation for Cα atoms in loop 1 in glycerol-bound and IFG-bound GCase is ∼1.7 Å2. Tyr313, which changes conformation so as to open the gate to the active site, is located on loop 1. Inspection of isotropic B-factors among the copies of GCase in the inh asymmetric unit reveals that the side chains of loop 1 residues are more ordered in the presence of IFG than in the presence of glycerol. When glycerol is bound, there is poor side chain electron density for several residues in loop 1, but in inh with IFG bound, single conformations for all main and side chains are readily visible.

Figure 5: Stereoview of region near Asn370, highlighting changes in loop 1.
figure 5

(a,b) Conformation with IFG bound (a) and glycerol (GOL) bound (b). Residues in dark orange derive from PDB code 1OGS; those in yellow are acidic.

Full size image

The helical turn in loop 1 has not been observed in previous structures determined at pH 4.5, which raises the possibility that this is a case of nonphysiological induced fit of the inhibitor or is due to crystal packing. This change in loop 1 was not observed in CBE-bound GCase, likely because CBE becomes covalently bound to Glu340 (ref. 18). Binding of sulfate20, ethylene glycol21 and glycerol (this work) also fails to induce the conformational change at low pH. However, the neutral pH GCase structure supports the biological relevance of this loop 1 conformation: in two of the molecules in the asymmetric unit, loop 1 is nearly identical to that when IFG is bound (Figs. 2b,d and 4b,c). Upon folding in the endoplasmic reticulum, GCase likely samples these and intermediate configurations observed in the crystal structure at pH 7.5 (Fig. 2b,c), and IFG binding in the endoplasmic reticulum likely favors the conformation with additional secondary structure. Overall, these results suggest that IFG stabilizes a single, ordered conformation of GCase, an effect likely important for its chaperoning function.

The loop 1 conformation seen in IFG-bound GCase (Figs. 2b,d and 5a) and in the similar structure seen at pH 7.5 provides new insight into the biochemical role of Asn370, the site of the mutation affecting 70% of Ashkenazi Jews and 20% of overall GD-afflicted individuals. Asn370 is located over 13 Å from the active site in a stable helix of the GCase TIM barrel fold, yet mutation to a serine reduces catalytic activity to ∼30% of that of the wild type8. Adoption of the helical turn in loop 1—in particular, the motion of Trp312 and Asp315—affects the electrostatic environment of Asn370 (Fig. 5). In the acidic structure, Trp312 adopts two conformations: one that abuts Asn370 and one that tucks underneath loop 1, ∼8 Å from Asn370 (Fig. 5b). In the inh structure, the Trp312 side chain swings out completely to the other side of Ser366. The space vacated by Trp312 is then filled with water molecules, thereby creating an extended hydrogen bond network among the water molecules, Asn370, Ser366 and Asp315 from loop 1 (Fig. 5a). In the acidic structure, Asp315 is stabilized by hydrogen bonds to Glu349 on loop 2. Thus, Asn370 seems to be important for stabilizing the helical turn conformation of loop 1. On the basis of crystal structures in which this feature is observed, it is reasonable to hypothesize that for N370S mutant GCase, the distance between loop 1 and the helix containing Ser370 is increased. This wider gap would result in a destabilization of the loop 1 conformation. For example, the helical turn of loop 1 may unwind such that Asp315 participates mostly in the hydrogen bonding network involving Ser366, rather than Ser370. If the IFG-bound structure mimics that when substrate is bound (see below), then this new loop 1 conformation would result in an impaired (but not abolished) ability to bind substrate, and may also have an impact on the ability of N370S GCase to interact with phospholipids and saposin C20.

Loop 2, which is also at the mouth of the GCase active site, undergoes notable changes (Figs. 2 and 4). At low pH, the coordinates of loop 2 remained unchanged upon IFG binding (Fig. 2a), but we observed a shift in loop 2 (r.m.s. deviation of ∼1.5 Å2) for the acidic-like monomers at neutral pH when compared with the low-pH structures (Fig. 2b,c). The shift may be due to new crystal contacts between the side chain nitrogen in Trp348 from one monomer and the backbone oxygen in Gly243 of a nearby monomer. This new interaction also results in a different hydrogen bonding pattern between residues at the base of loop 2 and residues on loop 1 near the active site. Future studies will be necessary to address whether the changes observed in loop 2 are a result of crystal packing, the presence of only water in the active site, or higher pH.

The importance of loop conformations for substrate binding is supported by in silico calculations (Fig. 4e). Much like in other glycosidases, the sugar moiety of glucosylceramide is expected to bind in the active site of GCase, with the catalytic and acid/base residues positioned to attack the anomeric carbon. However, the hydrophobic subsites21 for the extended substrate alkyl chains have not been previously identified unambiguously on GCase. In the original structure of GCase17, glucosylceramide was modeled such that the alkyl chains extended away from the surface of GCase, with glucose in the active site. We performed docking calculations with IFG and a modified glucosylceramide substrate (composed of truncated alkyl chains to limit the degrees of freedom, 7) using our IFG-bound and glycerol-bound structures as receptors. For calculations using coordinates of glycerol-bound GCase, both IFG and truncated glucosylceramide were docked onto the surface of the enzyme with poor computational scores. The active site opening is too small (Fig. 4a) to accommodate the substrate or analogs thereof. We next demonstrated that IFG docks computationally onto the coordinates of IFG-bound GCase in the same conformation as that observed crystallographically (not shown).

Notably, the coordinates of IFG-bound GCase accommodate the modified substrate in the active site, which overlays well with the position of IFG. The truncated alkyl branches of ceramide reside in two valleys emerging from the active site at the interface between loops 1 and 2, and between loop 1 and the loop comprising GCase233–252 (Fig. 4e). The sugar moiety can enter the active site once the side chain of Tyr313 has been repositioned and a shallower gradient and wider opening into the active site have been created. In the acidic and neutral structures, this latter surface can be blocked by hydrogen bonding of Asp315 on loop 1 to Glu349 on loop 2 (Fig. 4a), or by interactions with Cys342 (Fig. 4d), respectively.

Taken together, the effects of IFG on mutant GCase activity and trafficking have important implications for chaperone therapy. It seems likely that IFG improves activity and trafficking of N370S mutant GCase by locking GCase in a substrate-bound conformation—a hypothesis that is based on the conformation of loop 1, well-defined side chain density, decreased temperature factors, and the ability to dock the substrate onto coordinates of IFG-bound GCase but not onto the other observed enzyme conformations. Besides facilitating the replacement of IFG by substrate for catalysis, this distinct conformation of GCase may also favor proper binding of saposin C and lipids22,23. In addition, the structure of IFG-bound GCase is similar to one of the two distinct conformations of apo-GCase at neutral pH, which suggests that stabilization of this conformation in the endoplasmic reticulum may be required for correct subcellular trafficking. In sum, this study suggests that for enzymes undergoing substrate-dependent conformational changes, the ability to stabilize the right conformation, for activity and localization, may be the hallmark of a molecule with therapeutic potential.

Methods

GCase enzyme inhibition assay.

We purchased IFG and imiglucerase from Toronto Research Chemicals, Inc. and Genzyme Corporation, respectively. We incubated approximately 15 ng of protein, 3 mM 4-methylumbelliferyl-β-glucopyranoside (4MU-β-Glc) and various concentrations of IFG in McIlvaine buffer (0.1 M citrate, 0.2 M phosphate buffer, pH 5.2) containing 0.1% Triton X-100 and 0.25% sodium taurocholate (100 μl final volume) at 37 °C for 30 min, added an equal volume of 0.4 M glycine and 0.4 M NaOH, and measured 4-methylumbelliferone (4MU) release by fluorescence (excitation 355 nm, emission 460 nm) using a Victor 3 (Wallac) plate reader. We plotted normalized, background-subtracted triplicates against drug concentration and fitted them with a sigmoid function to estimate the IC50.

GCase enzyme activity in whole cell lysates.

We cultured fibroblasts (DMN89.15) from GD-afflicted individuals (homozygous for the alleles resulting in N370S GCase) in 12-well plates and incubated them with or without IFG for 5 d. We washed cells twice with medium (5 min each, 37 °C) and twice with phosphate-buffered saline (PBS) (5 min each, 37 °C) and then scraped, pelleted and lysed them in McIlvaine buffer (pH 5.2) containing 0.1% Triton X-100 and 0.25% sodium taurocholate. Then we centrifuged the cell lysates and preincubated supernatants in the presence or absence of 2.5 mM CBE for 30 min at 23 °C. Next, we added 50 μl of 6 mM 4MU-β-Glc substrate and incubated the mixture at 37 °C for 60 min. After adding 70 μl glycine buffer (0.2 M, pH 10.8) to stop the reaction, we measured released 4MU as described above. We measured protein concentrations in lysate by the bichinchonic acid method (Pierce) using bovine serum albumin as a standard, and we calculated GCase activity as the CBE-sensitive activity, expressed in terms of arbitrary fluorescence units released per milligram of protein per hour. We compared values from lysates of IFG-treated cells with those of untreated control cells using a two-tailed Student's t-test assuming equal variances. Differences with P values <0.05 were considered significant. We performed all analyses in triplicate.

Immunofluorescence.

We seeded primary fibroblasts from normal (fibroblast line CRL2097, American Type Culture Collection) and GD-afflicted (fibroblast line DMN89.15, from individuals that are homozygous for the alleles resulting in N370S GCase) individuals at 104 cells per well on glass coverslips in 12-well plates and treated them for 5 d with or without IFG, and with or without CBE, at the indicated concentrations. To fix the cells, we used 3.7% paraformaldehyde, permeabilized with 0.5% saponin in PBS, and incubated with rabbit polyclonal anti-GCase (raised against alglucerase; 1:200) and mouse monoclonal Lamp1 (human CD107a, BD Biosciences; 1:300) antisera in 0.1% saponin and 1% BSA in PBS. We detected primary antibodies using Alexa488-conjugated goat antibody to rabbit and Alexa594-conjugated goat antibody to mouse (Invitrogen; 1:500 for both), with samples mounted in Vectashield (Vector Labs), counterstained with 4′,6-diamidino-2-phenylindole (DAPI), and viewed using a Nikon C1 Plus laser scanning confocal microscope system. We rendered the acquired images using Adobe Photoshop.

Crystallization, data collection, structure determination and refinement.

We purchased imiglucerase (Genzyme Corporation) from Brigham & Women's Hospital pharmacy, partially deglycosylated it using N-glycosidase F (Glyko), and for the inh and acidic structures, crystallized it as described17. We obtained crystals with IFG by soaking GCase crystals for 10 min in a solution containing mother liquor plus 200 μM IFG. We froze crystals with a gradient of 5–20% glycerol in mother liquor. To obtain neutral pH crystals, we used hanging drop vapor diffusion with a crystallization solution composed of 0.8 M monosodium dihydrogen phosphate, 0.8 M monopotassium dihydrogen phosphate and 0.1 M HEPES, pH 7.5. Before data collection, we transferred these crystals to a solution containing 2 M Li2SO4 and cooled in liquid nitrogen. We collected all crystallographic data at GM/CA-CAT beamline at the Advanced Photon Source using a 4k × 4k MAR-Mosaic CCD detector, and processed them with HKL2000 (ref. 24).

We solved the structures by the method of molecular replacement with MOLREP25 using a monomeric search model derived from PDB code 1OGS (ref. 17) or by rigid-body refinement. First, we fit the atomic models into their respective electron density maps using Coot26 and refined them using REFMAC5 (ref. 25); we used medium noncrystallographic symmetry restraints for main chain and loose restraints for side chains for the neutral structure. Because of the inability to readily distinguish sulfate from phosphate in neutral electron density maps, we modeled all oxyanions as [PO4]3–. We generated topology and geometry restraints for IFG using PRODRG2 (ref. 27), and we identified bound water molecules using Coot26 and ARP/wARP28. The percentage of residues in the most favored and additional allowed regions of the Ramachandran plot were 99.7% for neutral and 99.8% for inh and acidic. To superposition coordinates, we used the secondary structure matching (SSM) algorithm29, and we generated figures using PyMOL (DeLano Scientific). For computational docking we used Shrödinger.

Accession codes.

Protein Data Bank: structures from this study have been deposited under accession codes 2NSX, 2NT0 and 2NT1. GCase structures cited from previous studies include 1OGS, 2F61 and 1Y7V.

Note: Supplementary information and chemical compound information is available on the Nature Chemical Biology website.