GlcNAc-1-P-transferase-tunicamycin complex structure reveals basis for inhibition of N-glycosylation

Summary

N-linked glycosylation is a predominant post-translational modification of protein in eukaryotes, and its dysregulation is the etiology of several human disorders. The enzyme UDP-N-acetylglucosamine:dolichyl-phosphate N-acetylglucosaminephosphotransferase (GlcNAc-1-P-transferase, GPT) catalyzes the first and committed step of N-linked glycosylation in the endoplasmic reticulum membrane, and it is the target of the natural product tunicamycin. Tunicamycin has potent antibacterial activity by inhibiting the bacterial cell wall synthesis enzyme MraY, but its usefulness as an antibiotic is limited by off-target inhibition of human GPT. Our understanding of how tunicamycin inhibits N-linked glycosylation and efforts to selectively target MraY are hampered by a lack of structural information. Here we present crystal structures of human GPT in complex with tunicamycin. Our structural and functional analyses reveal the difference between GPT and MraY in their mechanisms of inhibition by tunicamycin. We demonstrate that this difference could be exploited for the design of MraY-specific inhibitors as potential antibiotics.

Main text

N-linked glycosylation of proteins is one of the most common post-translational modifications in mammals and is involved in many important physiological processes including synaptic transmission, cell adhesion, organellar trafficking, protein folding, and neurodevelopment. The integral membrane enzyme UDP-N-acetylglucosamine:dolichyl-phosphate N-acetylglucosaminephosphotransferase (GlcNAc-1-P-transferase or GPT; DPAGT1 gene in human or ALG7 in yeast) catalyzes the first and committed step of N-linked glycosylation on the cytosolic face of endoplasmic reticulum (ER) membrane ¹. This reaction involves the transfer of N-acetylglucosamine-1-phosphate (GlcNAc-1-P) from UDP-N-acetylglucosamine (UDP-GlcNAc) to the carrier lipid dolichyl-phosphate (Dol-P). The dolichyl-diphosphate-N-acetylglucosamine (Dol-PP-GlcNAc) product is further processed by downstream enzymes and flipped into the ER lumen where processing is completed, before the sugars are transferred to the target protein by the oligosaccharyltransferase complex. Recent studies demonstrate that loss-of-function mutations in DPAGT1 are the etiologies of two types of human neurological diseases, congenital myasthenic syndrome (CMS) 13 and congenital disorder of glycosylation type lj (CDG-Ij) ^2–4, and that overexpression of DPAGT1 is a critical contributor to the early development and progression of oral cancer ^5,6. GPT is a member of the polyprenyl-phosphate-N-acetylhexosamine-1-phosphate-transferase (PNPT) superfamily, which also contains the bacterial paralogues MraY (peptidoglycan synthesis), WecA (lipopolysaccharide and enterobacterial common antigen synthesis), and TarO (teichoic acid synthesis) ⁷.

Tunicamycin is a nucleoside analog inhibitor of many PNPT superfamily enzymes ^8,9. Tunicamycin was thought of as a transition state analog that mimics the substrates UDP-GlcNAc and dolichyl phosphate ^10–14 (Supplementary Fig. 1), but it acts as a high-affinity competitive inhibitor for UDP-GlcNAc in mammalian GPT ^15–18. Tunicamycin is widely used as a research tool to block N-linked glycosylation or as an ER stress inducer to activate the unfolded protein response ^19–22. Tunicamycin also has potent antibacterial activity due to its competitive inhibition of the essential translocase MraY for its natural substrate UDP-MurNAc-pentapeptide ^23,24. However, tunicamycin has not been used as an antibiotic in the clinic due to its cytotoxicity in mammalian cells arising from its off-target inhibition of GPT ¹⁹. Efforts to develop tunicamycin into an antibiotic are hampered by a dearth of structural information ²⁵. Of the PNPT superfamily, only crystal structures of MraY have been published, either from Aquifex aeolicus (MraY_AA) in the Mg²⁺-bound state ²⁶ and bound to the competitive inhibitor muraymycin D2 ²⁷, or from Clostridium bolteae (MraY_CB) in complex with tunicamycin ²⁸. MraY contains a highly conserved active site, including ~34 invariant residues, that overlaps with the inhibitor binding sites; thus MraY orthologues are known to catalyze the same enzymatic reaction and share natural product inhibitors ^{23,26,27,29,30}. The uridine moiety of both muraymycin D2 and tunicamycin was bound to the same location in the conserved active site of MraY, which is consistent with the idea that both tunicamycin and muraymycin D2 act as competitive inhibitors to the substrate UDP-MurNAc-pentapeptide (Supplementary Fig. 1) ^23,27,28. Despite progress in our understanding of MraY, a crystal structure of GPT has remained elusive and many questions remain regarding the mechanism of GPT inhibition by tunicamycin. Does tunicamycin inhibit GPT and MraY by the same mechanism? Comparative studies of GPT and MraY are not only key to obtaining mechanistic insight into the inhibition of N-linked glycosylation by tunicamycin, but are also important as a platform for the design of antibiotics targeting MraY.

Toward these goals, we solved the crystal structure of human GPT in complex with tunicamycin. Structural analyses performed in parallel with biochemical and enzymatic studies demonstrate that GPT and MraY utilize different quaternary structure organization and mechanisms of inhibition by tunicamycin. Finally, we applied this knowledge to the design of an MraY-specific tunicamycin analog, which would serve as a blueprint for the development of MraY-specific antibiotics.

Structure of hGPT bound to tunicamycin

We determined the crystal structures of human GPT (hGPT) in complex with tunicamycin for both the canonical Pro129 variant (UniProt ID Q9H3H5) and a His129 variant (European Nucleotide Archive ID: AAG43168) to 3.10 Å and 2.95 Å resolution respectively. Both hGPT variants are functionally competent and they have comparable enzymatic activities (Supplementary Fig. 2). The final models were refined to good geometry (Table 1). Both models are structurally similar (Cα r.m.s.d ~0.8 Å). The canonical Pro129 model is used for all the figures except where noted. The electron density map for the hGPT structure was of excellent quality (Supplementary Fig. 3), allowing unambiguous placement of the entire tunicamycin core as well as the aliphatic tail (Supplementary Fig. 4), which was previously unresolved in the MraY-tunicamycin complex structure ²⁸.

Table 1.

Data collection and refinement statistics.

	hGPT (canonical)*	hGPT (His129 variant)†
Data collection
Space group	C2	C2
Cell dimensions
a, b, c (Å)	212.27, 105.51, 149.39	210.62, 105.36, 148.08
α, β, γ (°)	90, 103.53, 90	90, 103.49, 90
Wavelength (Å)	0.9791	0.9791
Resolution (Å)	75.31 - 3.10 (3.21 - 3.10)a	93.69 - 2.95 (3.06 - 2.95)a
CC1/2	0.96 (0.59)	0.84 (0.76)
I/σ(I)	13.65 (0.66)	17.51 (0.56)
Rpim	0.068 (>1.0)	0.079 (>1.0)
Redundancy	79.9 (69.1)	112.4 (112.8)
Completeness (%)	97.5 (96.3)	92.6 (54.7)
Completeness‡ (%)	60.1 (18.4)	55.3 (15.7)
Ellipsoidal completeness‡ (%)	94.9 (99.8)	94.8 (99.5)
Refinement
Resolution (Å)	75.31 - 3.10 (3.21 - 3.10)a	93.69 - 2.95 (3.06 - 2.95)a
No. reflections	35100 (1071)	36769 (1030)
Rwork/Rfree (%)	25.6/28.9	26.0/29.1
No. atoms
Protein	11170	11280
Tunicamycin/POPG	360	364
B factors
Protein	44.71	41.80
Tunicamycin/POPG	51.08	46.50
R.m.s. deviations
Bond lengths (Å)	0.003	0.003
Bond angles (°)	0.68	0.58
Ramachandran (%)
Favored	96.48	97.23
Allowed	3.52	2.77
Outliers	0.0	0.0
Rotamer outliers	0.0	0.0
Clashscore	4.66	1.57

^*Merged from 23 crystals.

^†Merged from 31 crystals.

^‡After applying anisotropy correction in STARANISO.

^aValues in parentheses are for highest-resolution shell.

Ellipsoidal completeness was calculated by STARANISO. All other statistics were calculated in PHENIX.

hGPT crystallized as a homodimer with one tunicamycin molecule bound to the active site of each protomer near the cytosolic side of the ER membrane (Fig. 1a). hGPT adopts the canonical PNPT superfamily fold of 10 transmembrane helices, with the active site framed on one side by the C-terminal segment of TM 9 (TM 9b) that is bent outward into the membrane. Notably, hGPT contains a large insertion in loop E between TMs 9b and 10, which folds into an extended βαββ motif not seen in MraY (Supplementary Fig. 5). This βαββ motif protrudes into the cytosol and connects to TM10 via the loop E helix (Fig. 1a). Tunicamycin is bound in an enclosed pocket formed mainly by TMs 4, 5, 6, 8 and almost completely covered by cytoplasmic loops A (between TMs 1 and 2) and E (between TMs 9 and 10) (Fig. 1a – 1b). The aliphatic tail of tunicamycin protrudes into a groove between TMs 4, 5, and 9, which has been suggested to be the binding site for the polyprenyl moiety of undecaprenyl phosphate (C₅₅-P) in MraY ²⁶. Tunicamycin is a high-affinity binder to hGPT, which we characterized by isothermal titration calorimetry (ITC) with a dissociation constant (K_d) of 5.6 nM in detergent micelles (Fig. 1c).

Figure 1. The crystal structure of human GPT in complex with tunicamycin reveals the structural basis for high affinity binding.

a, hGPT crystallized as a homodimer with one tunicamycin molecule (magenta sticks) bound to the cytosolic active site cavity of each protomer, tightly enclosed by loops A and E. The two protomers are related by pseudo-twofold rotational symmetry and are covalently linked by a disulfide bond between TM3 of each protomer. b, Cartoon representation of the cytoplasmic view, rainbow colored from N-terminus (blue) to C-terminus (red). c, Representative ITC raw data (top) and binding isotherm (bottom) for tunicamycin interacting with wild-type hGPT; K_d = 4.3 nM, ΔH° = −10.8 kcal/mol. This ITC experiment was performed in triplicate (technical replicates (5.6+/−1.3 nM, mean +/− S.E.M.)).

Tunicamycin binding to GPT versus MraY

To gain further insight into the binding of tunicamycin to hGPT, we next compared the hGPT-tunicamycin complex to the MraY_CB-tunicamycin complex ²⁸ (Fig. 2a). Tunicamycin is composed of uracil, tunicamine, GlcNAc, and an aliphatic tail ¹⁴ (Supplementary Fig. 1). The binding of the uracil, tunicamine, and aliphatic tail moieties of tunicamycin to hGPT is analogous to their binding to MraY_CB with noted differences (Fig. 2b). The uracil moiety of tunicamycin is stabilized by a conserved π-π stacking interaction with either Phe249 (hGPT) or Phe228 (MraY_CB) ²⁸, which was also observed in the crystal structure of MraY_AA in complex with muraymycin D2 ²⁷. This presumably resembles the interaction formed between hGPT or MraY and the uracil moiety of their respective UDP-sugar substrates (Supplementary Fig. 1). The core tunicamine moiety of tunicamycin is stabilized by hydrogen bonds with Asp252 in hGPT (Asp231 in MraY_CB) in both hGPT and MraY_CB. However, the tunicamine forms additional hydrogen bond interactions with Asn119 in hGPT that are not seen in the MraY_CB-tunicamycin complex. The amide linkage to the aliphatic tail of tunicamycin is stabilized by an interaction with the invariant Asn185 (Asn172 in MraY_CB, but the amide linkage is flipped 180° in MraY_CB possibly due to the aliphatic tail being unresolved). The tail is further locked into place by Trp122 in hGPT but not in MraY_CB (Pro108).

Figure 2. Tunicamycin binding to GPT is analogous but distinct from that to MraY.

a, Tunicamycin (magenta sticks) binds to the active site of both hGPT (green) and bacterial MraY_CB (blue, PDB 5JNQ) near the cytoplasmic side of the membrane. The aliphatic tail is not resolved in the MraY-tunicamycin structure. b, Interactions between tunicamycin and either hGPT or MraY_CB. c, Loop E of hGPT is more extensive and closer to tunicamycin than it is the MraY_CB structure, bringing the invariant residue Arg303 into contact with the GlcNAc moiety of tunicamycin. The corresponding residue in MraY_CB (Arg282) does not interact with tunicamycin. d, The Loop E helix of hGPT is swiveled by 30 degrees towards the horizontal plane as compared to MraY_CB. e, Tunicamycin is almost completely enclosed in the binding pocket of hGPT, as compared to the solvent-exposed groove of MraY. In hGPT, the uracil and GlcNAc moieties of tunicamycin are completely covered by loop A and loop E respectively, while are both solvent-exposed in MraY.

Notably, the interactions formed with the GlcNAc moiety of tunicamycin are significantly different between hGPT and MraY_CB (Fig. 2c–d). Due to the elongated loop E in hGPT, the invariant Arg303 is flipped inward and interacts with the N-acetyl group of the GlcNAc moiety; however, the corresponding residue Arg282 in MraY_CB is moved completely away from tunicamycin. The loop E helix of hGPT does not interact with the GlcNAc moiety of tunicamycin, unlike the loop E helix of MraY_CB (His290 and His291) which is oriented at a 30° angle relative to that in hGPT. The GlcNAc moiety of tunicamycin has been proposed to mimic the GlcNAc moiety of the endogenous GPT substrate UDP-GlcNAc, or the MurNAc moiety in the MraY substrate UDP-MurNAc-pentapeptide ^11–13. These differences in tunicamycin binding could reflect the different substrate specificity of hGPT and MraY.

Another marked difference between the tunicamycin binding pockets of GPT and MraY is solvent accessibility. The GlcNAc and the uracil moieties of tunicamycin are completely enclosed by loop E and loop A in hGPT, respectively, which was not observed in MraY_CB (Fig 2e). Consequently, tunicamycin is sequestered in the active site pocket of hGPT, but is exposed to the cytosol when bound to the shallow groove of MraY_CB.

Distinct dimer interfaces of GPT and MraY

Both GPT and MraY were previously shown to form a dimer in the membrane ^26,31. Cooperativity within the GPT dimer was proposed for its function, as a dominant-negative effect was observed by co-expression of the inactive Arg303Lys mutant with the wild-type GPT ³¹. Strikingly, comparison of the hGPT and MraY structures reveal dramatically different homo-dimerization interfaces, occurring at opposite sides of the protomer (Fig. 3a). The GPT dimer is formed by TMs 2, 3, 4 and part of TM 1, while the MraY dimer is formed by TM 10 and parts of TMs 1 and 7 (Fig. 3b). The different dimerization interfaces are supported by sequence conservation data: the GPT dimerization interface is conserved among GPT sequences but not MraY sequences, and the reverse is true for the MraY dimerization interface (Fig. 3c). An unusual membrane-buried disulfide bond is found between Cys106 of each protomer in the canonical hGPT structure, which is absent in the MraY dimer. The conserved nature of this disulfide bond among GPT orthologues provides further evidence that this is the physiological dimerization interface of GPT (Supplementary Fig. 5). We recognized that while the core of the GPT dimerization interface is buried, two hydrophobic fenestrations penetrate the dimer interface from the sides, both containing strong snake-like electron density which we assigned as a phospholipid tail from the 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoglycerol (POPG) added during protein purification (Fig 3b and Supplementary Fig. 6). The MraY dimer interface contains a central hydrophobic tunnel separating the TM 6 of each protomer, which has been suggested to be occupied by lipid molecules (Fig. 3b) ²⁶. It is increasingly recognized that interfacial lipids are important for stabilizing the membrane protein oligomers ³². Consistent with this idea, the phosphatidylglycerol-dependence of GPT activity was reported in early studies ^33,34, suggesting that the observed lipid-dependent regulation of GPT activity might be due to its stabilizing effects on the dimer interface. The opposite dimerization interfaces between GPT and MraY, as well as the resultant difference in lipid accessibility to the interface, suggest that GPT and MraY potentially may be regulated differently by phospholipids.

Figure 3. Divergence between GPT and MraY is underpinned by their different dimerization interfaces.

a, Superposition of hGPT (green) with MraY_CB (blue) reveals strikingly different dimerization interfaces, despite the conserved fold of each protomer. b, The different dimerization interfaces of GPT and MraY result in altered membrane accessibility. Hydrophobic side fenestrations penetrate the GPT dimer, while a central hydrophobic tunnel occupies the MraY dimer. c, The interface for GPT dimerization is conserved only among GPT orthologue sequences but not among MraY sequences, and vice versa is true for MraY. Surfaces are colored in increasing conservation from cyan (least) to magenta (most). Sequence conservation was mapped onto the hGPT and MraY_CB structures using 30 orthologue sequences for each alignment.

Distinct roles of magnesium

Magnesium is a required cofactor for GPT ³⁵ and MraY ³⁰ enzymatic activity. Extensive studies on MraY catalysis have revealed that three key aspartate residues (Asp116, Asp117, and Asp265 in MraY_AA), conserved throughout the PNPT superfamily, are critical for enzymatic activity ^26,29. Crystallographic studies unambiguously identified one of the aspartates, Asp265 in MraY_AA (Asp231 and Asp252 in MraY_CB and hGPT, respectively) as the coordinating residue of the magnesium cofactor (Fig. 4a).

Figure 4. MraY and GPT bind tunicamycin and magnesium differently and they are selective for distinct lipid substrates.

a, Structural overlay of hGPT-tunicamycin (green) and Mg²⁺-bound MraY_AA (magenta, PDB ID: 4J72). Labeled residues in magenta are known to be critical for MraY_AA activity and Asp265 coordinates the magnesium cofactor (magenta sphere). b, Fold increase in K_d of tunicamycin with either hGPT or MraY as measured by ITC. The K_d of tunicamycin for hGPT without added MgCl₂ is 5.6 ± 1.3 nM (triplicate, technical replicates). The K_d of tunicamycin for MraY_AA without added MgCl₂ is 37 ± 1 nM (triplicate, technical replicates). Data are shown as the mean of three technical replicates ± s.e.m. c, Chemical structures of dolichyl phosphate and undecaprenyl phosphate. d, TLC-based specific activity assay for hGPT and MraY_AA in the presence of either dolichyl phosphate (C₅₅-dol-P) or undecaprenyl phosphate (C₅₅-P). hGPT is selective for the saturated α-isoprenyl position in its substrate dolichyl-phosphate, while MraY is selective for the unsaturated α-isoprenyl position in its substrate undecaprenyl-phosphate. Measurements were performed in triplicate (technical replicates). Specific activity measurements were normalized relative to the C₅₅-Dol-P condition for hGPT and to the C₅₅-P condition for MraY. Data are shown as the mean of normalized specific activity ± s.e.m. Significance was determined with P < 0.001, two-tailed Student’s t-test. e, Cytoplasmic view of the tunicamycin lipid tail binding sites in hGPT and MraY_CB. In GPT, the aliphatic tail of tunicamycin packs against the conserved Trp122, which forms a lipid-binding tunnel. By contrast, the corresponding residue in MraY_CB (Pro108) is unable lock the lipid tail into place, leaving the hydrophobic groove exposed.

It has been suggested that Mg²⁺ is required for tunicamycin binding to MraY and GPT, because the tunicamine moiety of tunicamycin is thought to mimic the diphosphate moiety in the transition state of the enzymatic reaction between nucleotide sugar and lipid carrier ^10,12,13. However, comparison of the tunicamycin-bound hGPT and the Mg²⁺-bound apo MraY_AA structures shows that Asp265 (Asp252 in hGPT) coordinates Mg²⁺ in MraY_AA, but interacts directly with tunicamycin in hGPT, suggesting Mg²⁺ and tunicamycin likely compete for binding (Fig. 4a)²⁶. If our hypothesis is correct, we predict that Mg²⁺ is not required for tunicamycin binding, and we also expect reduced tunicamycin binding with increasing concentrations of Mg²⁺. We performed ITC by titrating tunicamycin into either MraY_AA or hGPT in the presence or absence of MgCl₂ (Fig. 4b). We used MraY_AA for our assays because the structure, function, and inhibition of this orthologue have been well studied and MraY orthologues share common mechanisms of catalysis and inhibition ^{23,26,27,29,30}. We found that tunicamycin binds to both MraY_AA and hGPT in the absence of Mg²⁺, and the addition of 10 mM MgCl₂ resulted in an eight-fold increase in K_d for tunicamycin with MraY_AA. Mutating the Mg²⁺-coordinating residue in MraY_AA (Asp265) to alanine increased the K_d for tunicamycin by over twenty-fold. On the contrary, the addition of 10 mM MgCl₂ has no appreciable effect on tunicamycin binding to GPT. Furthermore, mutation of Asp252 to alanine increased K_d of tunicamycin by only about three-fold. In summary, our data suggest that the role of Mg²⁺ in tunicamycin binding is substantially different in GPT and MraY; while Mg²⁺ competes with tunicamycin binding to MraY, it has no detectable effect on tunicamycin binding to GPT. These observations are contrary to the previous notion that Mg²⁺ is essential for tunicamycin binding to these two enzymes ^12,13.

Distinct lipid substrate specificity of GPT and MraY

GPT and MraY utilize different lipid substrates. GPT utilizes dolichyl phosphate, the lipid carrier for eukaryotic N-linked glycosylation, while MraY utilizes undecaprenyl phosphate, the lipid carrier for peptidoglycan biosynthesis. A major difference between these two polyprenyl phosphates is the α-isoprene unit closest to the phosphate group: dolichyl phosphate has a saturated bond at this position while undecaprenyl phosphate has an unsaturated bond (Fig. 4c). To test the selectivity of GPT and MraY for their respective lipid substrates, hGPT and MraY_AA enzymatic reactions were carried out in the presence of either undecaprenyl phosphate (C₅₅-P) or C₅₅ dolichyl phosphate (C₅₅-dol-P) (Fig. 4d). hGPT exhibits strong selectivity for C₅₅-dol-P with nearly no observed activity in the presence of C₅₅-P. Conversely, MraY_AA shows clear preference for C₅₅-P over C₅₅-dol-P. This result is consistent with the structural architecture of the hGPT and MraY_CB active sites (Fig. 4e). Tunicamycin contains a lipid tail that is thought to overlap with the lipid substrate binding sites in GPT and MraY. In hGPT, the lipid tail of tunicamycin fits snugly into a deep pocket, which is trapped by the conserved Trp122 (Fig. 4e left). This highly defined lipid tunnel is likely required to constrain dolichyl phosphate, which has higher rotational freedom due to the saturated α-isoprene unit. By contrast, the tunicamycin lipid tail binding site in MraY is a more shallow and wide groove than the analogous site in GPT (Fig. 4e right). The unsaturated α-isoprene unit of undecaprenyl phosphate has a limited degree of rotational freedom and its assumed geometry probably guides binding to the shallow MraY lipid tail binding groove. Structural differences between the lipid tail binding sites in GPT and MraY likely underpins the high selectivity each enzyme exhibits for its respective polyprenyl substrate.

Chemical modification of tunicamycin selectively targets MraY over GPT

Efforts to develop tunicamycin as an antibiotic requires rational design to eliminate the cytotoxicity arising from its off-target effect on hGPT activity ^25,36. Based on our structural analysis of the tunicamycin binding site, we designed a tunicamycin analog that replaces the GlcNAc sugar moiety in tunicamycin with a MurNAc sugar (Fig. 5a), which we synthesized as described in the recently published total synthesis of tunicamycin V (Supplementary Note 1)³⁷. The rationale for this chemical modification is that (1) the binding pocket of hGPT formed by loop E tightly packs around the GlcNAc sugar (Fig. 2e) and likely cannot accommodate the bulkier MurNAc moiety, and (2) the natural substrate for MraY contains a MurNAc moiety, while the natural substrate of hGPT contains a GlcNAc moiety (Supplementary Fig. 1). We observed that the IC₅₀ of tunicamycin for hGPT is about 9 nM and that of tunicamycin-MurNAc is about 15 μM, resulting in about a thousand-fold decrease in inhibition of hGPT by this simple modification of tunicamycin (Fig. 5b). By contrast, the activity of MraY_AA is similarly inhibited by both tunicamycin and tunicamycin-MurNAc (Fig. 5b). This observation is consistent with the crystal structure of MraY_CB bound to tunicamycin, which demonstrates that the binding site of the tunicamycin GlcNAc moiety is wider than that of hGPT and is therefore more likely to accommodate the larger MurNAc moiety (Fig. 2e). Therefore, we showed proof-of-principle to selectively target tunicamycin towards MraY over GPT at the protein level. Further investigation is required to evaluate the in vivo activity and toxicity of such analogs.

Figure 5. Chemically modifying tunicamycin can introduce ligand selectivity between hGPT and MraY_AA.

a, Chemical structures of tunicamycin and its MurNAc derivative. The substructure highlighted in red differs from tunicamycin, being a MurNAc-like moiety rather than a GlcNAc moiety as in tunicamycin. b, IC₅₀ measurements of tunicamycin and its MurNAc analog (tunicamycin-MurNac) with hGPT and MraY_AA, respectively. hGPT is much less inhibited by tunicamycin-MurNAc than tunicamycin, while MraY appears to be similarly inhibited by both tunicamycin and tunicamycin-MurNAc. The IC₅₀ for hGPT with tunicamycin is ~9 nM; for hGPT with tunicamycin-MurNAc is ~15 μM; for MraY_AA with tunicamycin is ~450 nM; for MraY_AA with tunicamycin-MurNAc is ~640 nM. Each IC₅₀ value was determined using the TLC-based phosphoglycosyl transferase assay. Data are shown as the mean of triplicate measurements (technical replicates) ± s.e.m.

Discussion

Our structural and functional studies of hGPT provide the framework to elucidate the molecular basis of mutations that cause CDG-Ij and CMS13 (Supplementary Fig. 7). Intriguingly, distinct groups of mutations within DPAGT1 are responsible for either CDG-Ij or CMS13, and they are spread over hGPT rather than localized around the active site of hGPT (Supplementary Fig. 7). Systematic enzymatic, structural, and cellular studies are required to understand the effects of disease mutations on hGPT and the N-linked glycosylation.

GPT and MraY have drastically different dimer organization. It is unusual to have distinct dimerization organization among paralogues for enzymes. What is the possible reason for distinct dimerization of GPT compared to MraY? While the cooperativity of the GPT dimer and the importance of phosphatidyglycerol to GPT function were reported, no studies have been reported for the cooperativity of the MraY dimer. We observed that the two active sites are closer together in the GPT dimer than in the MraY dimer. In the GPT dimer, part of TM1 and loop A is closer to the symmetric axis (Fig. 1), potentially enabling communication between the two active sites within the dimer, which is not present in the MraY dimer. Therefore, we speculate that the distinct dimerization of GPT is to enable cooperativity, which adds another layer of complexity to the regulation of this eukaryotic system.

We showed that the mechanisms of tunicamycin binding to MraY and GPT are different in many respects. First, we found the role of Mg²⁺ is different in tunicamycin inhibition of GPT and MraY. Some potential reasons for this difference include, (1) GPT forms several additional interactions with tunicamycin, thereby making it less readily displaced by Mg²⁺, and (2) the Mg²⁺ binding site in GPT is different to that in MraY. Second, we found that the active site shape and solvent accessibility is significantly different between GPT and MraY. Difference in the active site shape could reflect the substantial difference in size between the GPT substrate UDP-GlcNAc (~600 Da) and the MraY substrate UDP-MurNAc-pentapeptide (~1200 Da) (Supplementary Fig. 1); the former could be easily captured in an enclosed pocket whereas the latter would require an exposed groove for binding. Third, the lipid substrate specificity is distinct in GPT and MraY, which is further supported by structural differences in the lipid binding site of each enzyme. We exploited one of these distinctions to design a tunicamycin analog with biased specificity toward MraY at the enzyme level. Our proof-of-principle studies with tunicamycin-MurNAc demonstrate that developing MraY-specific tunicamycin analogues is possible. In principle, we believe many of the features that distinguish MraY and GPT could be utilized for the development of MraY-specific tunicamycin analogs. For example, the lipid tail portion of tunicamycin can be also modified to reduce affinity for GPT.

Methods

Human GPT-tunicamycin complex expression, purification, and crystallization

The full-length DPAGT1 gene encoding human GPT (all residues 1-408, His129 variant, European Nucleotide Archive ID: AAG43168) was synthesized and subcloned into a modified pFastBac vector (Invitrogen) in frame with a C-terminal PreScission protease site, FLAG affinity tag and 10xHis affinity tag. hGPT (Pro129) variant construct was made by a mutagenesis with Pfu-Turbo DNA polymerase (Agilent) on hGPT (His129) construct. Baculovirus was produced according to the manufacturer’s protocol (Bac-to-Bac Baculovirus expression system, Invitrogen). For hGPT expression, Spodoptera frugiperda 9 (Sf9) insect cells were infected with a baculovirus at a density of 1.2 M cells mL⁻¹ and grown at 27°C for 24 h in an orbital shaker. Tunicamycin (5 mgmL⁻¹ in DMSO) was added into cultures at 1 mgL⁻¹ and cells were grown for 48 h at 27°C. Cells were then harvested by centrifugation at 1710 rpm for 12 mins at 4°C. Cell pellets were resuspended in buffer A (50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1 μgmL⁻¹ leupeptin, 1 μgmL⁻¹ pepstatin, 1 μgmL⁻¹ aprotinin, 1 mM PMSF and DNaseI) and lysed by sonication (5 × 30 pulses). All hGPT-tunicamycin complex purification steps were performed at 4°C. To solubilize the hGPT-tunicamycin complex, 40 mM n-dodecyl-β-D-maltopyranoside (DDM, Anatrace), 2 mgmL⁻¹ iodoacetamide and 4 mM cholesteryl hemisuccinate tris salt (CHS, Anatrace) were added to the lysate and stirred at 4°C for 2 hrs. Insoluble material was removed by centrifugation (8000g, 30 min), and anti-FLAG resin was added to the supernatant and incubated for 1 h at 4°C. The resin was then washed with 10 column volumes of buffer B (20 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1 mM DDM, 0.1 mM CHS). The hGPT-tunicamycin complex was eluted with 5 column volumes of elution buffer (20 mM Tris-HCl, pH 8.0, 150 mM NaCl, 0.1 mM CHS, 1 mM DDM, 0.1 mgmL⁻¹ FLAG peptide, 0.1 mgmL⁻¹ 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-(1′-rac-glycerol) (POPG, Avanti Polar Lipids)). The FLAG and 10xHis affinity tags were removed by overnight incubation with PreScission protease at 4°C with 1 mM dithiothreitol (DTT) and 1 mM ethylenediaminetetraacetic acid (EDTA). The hGPT-tunicamycin complex was further purified by size-exclusion chromatography (Superdex 200 10/300 GL) with SEC buffer containing 20 mM Tris-HCl pH 8.0, 150 mM NaCl, 0.3 mM decyl maltose neopentyl glycol (DMNG, Anatrace), 0.1 mM CHS, 0.1 mgmL⁻¹ POPG, 2 mM DTT, 10 μM tunicamycin, 10 mM MgCl₂).

Size-exclusion chromatography fractions of GPT-tunicamycin complex were concentrated to 8–10 mgmL⁻¹. Initial crystallization trials were performed with in-house crystal screens in 96-well sitting drop plates (Art-Robbins). Crystals were grown by sitting-drop vapor diffusion at 17°C in 0.2 M CaCl₂, 0.05 M Tris-HCl, pH 8.5, 25–30% PEG 400. Crystals appeared within 3–4 days, at which point they were immediately harvested in liquid nitrogen without additional cryo-protectant.

Data collection and structure determination

We collected X-ray diffraction data at the NECAT 24-ID-C and 24-ID-E beamlines, as well as at the SERCAT 22-ID beamline (Advanced Photon Source, Argonne National Laboratory) with a wavelength of 0.9791 Å. Data were processed by XDS ³⁸, and XDS-processed data from multiple isomorphous crystals were merged by POINTLESS and AIMLESS in BLEND ^39–41. Due to severe diffraction anisotropy, merged data were subjected to ellipsoidal truncation and anisotropy correction in STARANISO ⁴². Anisotropy corrected data were used for molecular replacement using the publicly released structure of apo human GPT Val264Gly mutant (PDB ID: 5LEV) as search model in PHASER ⁴³. Our hGPT-tunicamycin crystals were in C2 space group with two dimers in the asymmetric unit. The molecular replacement solution was first subjected to initial jelly-body refinement in the LORESTR pipeline ⁴⁴, before manual model building in COOT ⁴⁵ and refinement in REFMAC ⁴⁶ and PHENIX.refine ⁴⁷. Due to the higher resolution data of the hGPT(His129)-tunicamycin complex (2.95 Å resolution) than the canonical hGPT(Pro129)-tunicamycin complex (3.10 Å resolution), the initial round of molecular replacement and refinement was performed on the His129 data as described above. The model built from the His129 data was then transferred to the Pro129 data and a second round of model building and refinement was performed to produce the Pro129 structure. Structures were refined to a final R_work/R_free of 25.6/28.9% (Pro129 structure) and 26.0/29.1% (His129 structure). The final models were refined to excellent geometry (96–97% Ramachandran favored, 0% Ramachandran outliers, 0% rotamer outliers for both structures) and minimal clashes (clashscore of 4.66 for Pro129 and 1.57 for His129). Structural alignments and molecular graphics were created in PyMOL ⁴⁸. MAFFT (“add new sequences to an existing alignment” mode) was used to expand the structure-guided sequence alignment to include more sequences ⁴⁹. Sequence conservation was mapped onto the hGPT and MraY_CB structures using the ConSurf server ⁵⁰ using 30 orthologue seuqences for each alignment.

hGPT expression and purification for functional studies

For functional studies, the same DNA construct for hGPT expression was used, but tunicamycin was omitted from all the culture and protein purification steps. After adding baculovirus of hGPT wild-type or Asp252Ala mutant into 1.2 M Sf9 cells mL⁻¹, cells were grown for 72 hrs at 27°C. Cells were then harvested by centrifugation at 1710 rpm for 12 mins at 4°C. The cell pellet was resuspended with buffer A (50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1 μgmL⁻¹ leupeptin, 1 μgmL⁻¹ pepstatin, 1 μgmL⁻¹ aprotinin, 1 mM PMSF and DNaseI) and lysed by sonication (5 × 30 pulses). To solubilize hGPT, 40 mM DDM was added to the lysate and stirred at 4°C for 2 h. Insoluble material was removed by centrifugation (8000g, 30 min), and anti-FLAG resin was added to the supernatant and stirred for 1 h at 4°C. The resin was then washed with 10 column volumes of buffer B (20 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1 mM DDM). hGPT was eluted with 5 column volumes of elution buffer (20 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1 mM DDM, 0.1 mgmL⁻¹ FLAG peptide, 0.1 mgmL⁻¹ POPG). FLAG and 10xHis affinity tag was removed by overnight incubation with PreScission protease at 4°C with 1 mM DTT and 1 mM EDTA. hGPT was further purified by size-exclusion chromatography (Superdex 200 10/300 GL) with SEC buffer (20 mM Tris-HCl pH 8.0, 150 mM NaCl, 0.3 mM DMNG, 0.1 mgmL⁻¹ POPG, 2 mM DTT).

Isothermal titration calorimetry

MraY_AA wild-type and Asp256Ala were purified as previously described ²⁶ in a buffer containing 150mM NaCl, 20mM Tris-HCl pH 8.0, 5 mM DM and 2mM DTT. This same buffer was used to dilute the ligand, tunicamycin (Sigma-Aldrich). For one set of titrations, 10 mM MgCl₂ was also included in the buffer. For MraY_AA wild-type, 210 μM tunicamycin was titrated into 30 μM enzyme. hGPT wild-type and Asp252Ala were prepared as described above and tunicamycin was dissolved in the SEC buffer used to purify these enzymes. For MraY_AA Asp265Ala, 300 μM tunicamycin was titrated into 30 μM enzyme. For hGPT WT and Asp252Ala, 240 μM tunicamycin was titrated into 40 μM enzyme. All titrations were performed in triplicate (technical replicates) at 30°C for hGPT and at 45°C for MraY_AA using a MicroCal iTC200 (GE Healthcare). The total heat exchanged during each injection was fit to a single-site binding isotherm with K_d and ΔH° as independent parameters. Data were analyzed and figures were generated using Origin software (OriginLab Corp).

Thin-layer chromatography (TLC)-based enzyme assays

Enzymatic assays monitored either 1) the hGPT-mediated transfer of [¹⁴C]phospho-GlcNAc from [¹⁴C]UDP-GlcNAc to dolichyl phosphate (C₅₅-Dol-P), forming [¹⁴C]Dol-PP-GlcNAc, or 2) the MraY_AA-mediated transfer of [¹⁴C]phospho-MurNAc-pentapeptide from [¹⁴C]UDP-MurNAc-pentapeptide(DAP) ([¹⁴C]UM5A) to undecaprenyl phosphate (C₅₅-P), forming [¹⁴C]Lipid I. A thin-layer chromatography (TLC)-based radiochemical assay ⁵¹ was used, which was optimized for hGPT and MraY_AA-catalyzed reactions for the experiments described below. All reactions were quenched by spotting a 2-μL aliquot on a silica gel 60 thin layer chromatography (TLC) plate (EMD Millipore). The products and substrates were separated by TLC using isopropanol/ammonium hydroxide/water (6:3:1; v/v/v) as the mobile phase. The spots corresponding to product and substrate were visualized via Phosphorimager (Typhoon FLA 7000, GE Healthcare LifeSciences). The spot intensity was quantified using the ImageQuant TL (GE Healthcare LifeSciences). All experiments were performed in triplicate.

Specific activity measurements of hGPT wild-type His129 and Pro129 variants

The reaction buffer contained 70 mM Tris-HCl pH 8.0, 500 mM NaCl, 80 mM MgCl₂, 5 mM DM, 1 mg/mL POPG, 10% glycerol, 500 μM C₅₅-dol-P, 0.01 mM [¹⁴C]UDP-GlcNAc and 0.1 mM UDP-GlcNAc. The reactions were initiated with 200 nM hGPT His129 or Pro129, were incubated at 30°C for 20 min, and a 2-μL aliquot of each reaction was spotted on the TLC plate every 5 min. Reactions were performed in triplicate (technical replicates). The specific activity was calculated by dividing the rate of product formation (nmol/min) by the quantity of protein added in milligrams.

Lipid substrate dependence

The hGPT-catalyzed reactions contained the same buffer composition as described above. The MraY_AA-catalyzed reaction buffer contained 100 mM Tris-HCl pH 8.0, 500 mM NaCl, 10 mM MgCl₂, 20 mM CHAPS, and 150 μM [¹⁴C]UM5A. Each reaction included 500 μM of either C₅₅-Dol-P or C₅₅-P. Reactions were initiated by either 200 nM hGPT Pro129 or 50 nM MraY_AA wild-type and were carried out for 12 min at 30°C (hGPT) or 5 min at 60°C (MraY_AA). Reactions were performed in triplicate (technical replicates). Specific activity was calculated as described above and normalized to the maximally active reaction condition.

IC₅₀ measurements

The assay reaction mixtures contained the same buffer composition as described above for the hGPT- and MraY_AA-catalyzed reactions, respectively. Reactions were initiated by either 200 nM hGPT Pro129 or 50 nM MraY_AA and were carried out for 12 min at 30°C (hGPT) or 5 min at 60°C (MraY_AA), which is within the enzymatic linear range for each enzyme. For IC₅₀ measurements with tunicamycin and hGPT, the following concentrations of tunicamycin (μM) were used: 0, 0.0005, 0.01, 0.04, 0.07, 0.15, 0.5, 1, and 5. For IC₅₀ measurements with tunicamycin-MurNAc and hGPT, the following concentrations of tunicamycin-MurNAc (μM) were used: 0, 0.05, 0.1, 1, 5, 10, 50, and 100. For IC₅₀ measurements with tunicamycin and MraY_AA, the following concentrations of tunicamycin (μM) were used: 0. 0.05, 0.1, 0.4, 0.8, 1, 5, and 50. The same concentration range was used for IC₅₀ measurements with tunicamycin-MurNAc and MraY_AA. All measurements were made in triplicate (technical replicates). Data were plotted in GraphPad Prism 6 and IC₅₀ values were calculated using the log(inhibitor) vs. response – variable slope model.

Synthesis of tunicamycin-MurNAc

We synthesized tunicamycin-MurNAc as described in the recently published total synthesis of tunicamycin V (see Supplementary Note 1 for details) ³⁷. In brief, the analog was synthesized by assemblage of a trichloroacetimidate of MurNAc N-methylamide derivative S5 as a glycosyl donor and a suitably protected tunicaminyluracil S6 as a glycosyl acceptor ³⁷. Namely, treatment of S5 and S6 with TfOH in Et₂O at 0 °C provided the desired 11′-β-1″-α-glycoside S7 in 69% yield in a highly stereoselective manner. Transformation of the azide group, deprotection of the phthaloyl group at the GalNAc moiety, and installation of the fatty acyl group gave S10. Finally, global deprotection of six acid labile protecting groups by BCl₃ resulted in clean conversion to afford tunicamycin-MurNAc.

Data availability

Atomic coordinates and structure factors for the reported crystal structures are deposited in the Protein Data Bank under accession codes 6BW5 for the canonical hGPT (Pro129) and 6BW6 for the His129 variant. Any other data pertaining to this paper is available upon reasonable request.