Results

DPAGT1 Activity and Architecture
To determine the structure of DPAGT1, we expressed both full-length wild-type (WT) DPAGT1 and the missense mutant Val264Gly in the baculovirus/insect cell system. The Val264Gly mutation is a variant found in some CMS patients (Belaya et al., 2012), which improved crystallization behavior compared to WT. We tested the enzymatic activity of WT and Val264Gly DPAGT1 to confirm that the protein was functional. Product GlcNAc-PP-Dol was confirmed by mass spectrometry (Figure S1C). WT DPAGT1 has an apparent Km of 4.5 ± 0.8 μM and a kcat of 0.21 ± 0.007 min-1 towards UDP-GlcNAc (Figure 1B, STAR Methods); Dol-P displayed an apparent Km of 36.3 ± 7.2 μM and a kcat of 0.20 ± 0.012 min-1 (Figure 1B). The Val264Gly mutant showed 2.5-fold higher activity (Figure S1D) and similar thermostability to WT protein (Tm1/2 of 51.7 ± 0.2 °C for WT and 50.4 ± 0.3 °C for mutant, Figure S1E, STAR Methods, Table S1). Adding the product analogue GlcNAc-PP-Und (equimolar to Dol-P and UDP-GlcNAc) reduced activity 3-fold, the addition of the other product, UMP, had no effect (Figure S1F). Tunicamycin fully inhibited at 1:1 molar ratio with DPAGT1 (Figure S1G). While both substrates thermostabilized WT and mutant DPAGT1 by 3–7 °C, tunicamycin thermostabilized both by more than 30 °C (Figure S1E). Interestingly, phosphatidylglycerol co-purified with DPAGT1, even when not added to the purification, suggesting a role for phosphatidylglycerol in the stability of DPAGT1 (Figures S1H and S1I).
The crystal structures of the WT DPAGT1 and Val264Gly mutant were solved by X-ray crystallography using molecular replacement with the bacterial homologue MraY (Chung et al., 2013, PDB: 4J72, 19% identity) as an initial model (STAR Methods). The apo WT- and Val264Gly-DPAGT1 proteins gave structures to 3.6 Å and 3.2 Å resolution (Table S2). Complexes with UDP-GlcNAc and tunicamycin gave data to 3.1 Å and 3.4 Å resolution, respectively (Figure 1, Figures S2A–S2E and Table S2). In the crystals DPAGT1 is a dimer formed through a crystallographic 2-fold axis (Figure S2F), with an 1850 Å2 interaction surface. The interface observed is similar to that seen in the Yoo et al., 2018 structures, although in that case the crystals show a dimer in the asymmetric unit. Interestingly, comparison with the dimeric bacterial homologue MraY, showed that the surface that forms the dimer interface differs between DPAGT1 and MraY (Figure S2G) (Yoo et al., 2018). In solution, DPAGT1 exists predominantly as a dimer, as shown by native mass spectrometry, although the monomer was also detected (Figures S2H and S2I). A Leu103Phe mutation introduced at the dimer interface to disrupt this interaction gave unstable protein, suggesting that DPAGT1 dimerization plays a role in stability. In the dimer interface the sidechains of Cys106 are adjacent, but no intermolecular disulfide bond was observed in our electron density maps (Figure S2J). To assess the presence of this potential disulfide, DPAGT1 was purified without reducing agents and its mass matched only monomer (STAR Methods). No covalent, disulfide-bonded dimer was observed. We concluded that DPAGT1 exists predominantly as a non-covalent dimer in solution and that dimerization is important for its stability.
Figure S2 Electron Density Maps, Heavy Atom Phasing and Dimer Interface Analysis for the Structure of Dimeric WT and Gly264Val Mutant of DPAGT1 Solved at Resolutions up to 3.2 Å, Related to Figure 2
(A) Hg bound sites from a soaked crystal. 6 Å anomalous difference Fourier map calculated from a dataset collected from a crystal soaked with EMTS is contoured at 5σ (purple) and 3σ (magenta mesh) and overlaid on the final model. Cysteine positions are highlighted by cyan sticks. Labelling of five of the nine ordered cysteines (Cys42, Cys72, Cys106 (weak), Cys217 and Cys299) is observed under the soaking conditions used.
(B) S-SAD peaks. The PHASER-EP log-likelihood map after anomalous model completion with sulphur atoms is shown overlaid on a cartoon representation of the final UDP-GlcNAc complex. The positions of cysteine (cyan sticks) and methionine (blue sticks) residues are highlighted. The map is contoured at 4.5σ (magenta) and 2.5σ (pink mesh). Peaks are observed for 16 sulphur atoms (out of a total of 18 possible ordered sulphurs) and the pyrophosphate of the UDP-GlcNAc is also resolved.
(C and D) Two views of final 2Fo-Fc AUTOBUSTER electron density map around the active site in the UDP-GlcNAc complex. The map has been sharpened using a B-factor of -100 Å2 in COOT and is contoured at 1.5σ and overlaid on the final model. UDP-GlcNAc is shown in ball-and-stick form (carbon-magenta, oxygen-red, phosphorus-orange, nitrogen-blue).
(E) Omit Fo-Fc electron density map for tunicamycin. The omit difference density is contoured at 2.5σ (green mesh) and a sharpened omit Fo-Fc density map (B=-100 Å2) is contoured at 2σ and overlaid on the final tunicamycin coordinates. Density for tunicamycin’s alkyl tail is not as well resolved as the TUN core.
(F and G) Comparison of dimer organization in DPAGT1 (F) and unbound MraY (PDB: 5JNQ) (G) in crystals, indicating that DPAGT1 and MraY are both “head-to’head” dimers in their respective crystals, but the dimer interfaces are unrelated.
(H) Size exclusion chromatography from a DPAGT1 purification with activity data per fraction per unit of protein shown as box plot, indicating that there is no difference in the activity of DPAGT1 across the peak (n=6).
(I) Native mass spectrometry confirms that DPAGT1 is a mixture of monomers and dimers.
(J) Electron density at the dimer interface between adjacent Cys106 residues in TMH3, with the sharpened (-100 Å2) 2FoFc density shown in blue (contoured at 1.0sigma) and “omit” density from a refinement where the sidechain of Cys106 was omitted (green mesh, contoured at 6.0sigma).
The DPAGT1 structure (as reported here and by Yoo et al., 2018) consists of 10 transmembrane helices (TMH1 to 10), with both termini in the ER lumen (Figures 1C and 1D). Five loops connect the TMHs on the cytoplasmic side of the membrane (CL1, -3, -5, -7, and -9), which form the active site, 3 loops on the ER side of the membrane (EL2, -4, -6) and one (EL8) embedded in the membrane on the ER side. DPAGT1 has a similar overall fold to MraY (Chung et al., 2013, Yoo et al., 2018) (Figure 1E). A feature of the eukaryotic DPAGT1 PNPT family not found in prokaryotic PNPTs is a 52-residue insertion between Arg306-Cys358 in CL9, following TMH9. This motif adopts a mixed α/β fold with an extended structure with 2 β-hairpins, a 3-stranded β-sheet (Cβ5-Cβ7) and two amphipathic α-helices (CH2/CH3). This CL9 domain (Figure 1F) forms part of the substrate recognition site in human DPAGT1 but not in bacterial MraY (Figure 1G).
The active site is on the cytoplasmic face of the membrane, formed by the 4 cytoplasmic loops between the TMHs (Figure 2A). The long CL1 loop forms the “back wall”; CL5 and CL7 form the base and the “side walls” are formed by TMH3-CL3-TMH4, TMH9b and the extended loop at the start of the CL9 domain (Ile297-Pro305). The entrance to the catalytic site, between TMH4 and TMH9b, is open and accessible from the lipid bilayer via a 10 Å wide cleft. Within the membrane, adjacent to the active site, there is a hydrophobic concave region (Figure 2B), created by a 60° bend in TMH9 midway through the bilayer, which creates a groove in the DPAGT1 surface (Figure 1D).
Figure 2 DPAGT1 Active Site
(A) The loops that form the active site; UDP-GlcNAc in magenta.
(B) Sliced molecular surface showing occluded active site cleft and putative Dol-P recognition groove. Surface is coloured by electrostatic potential; UDP-GlcNAc in magenta.
(C) Conformational changes with UDP-GlcNAc binding. Protein depicted in tube form with the tube thickness and colouring reflecting the rmsd in mainchain atomic positions between the unbound and UDP-GlcNAc-bound structures.
(D) UDP-GlcNAc binding in active site. Omit Fo-Fc difference electron density shown for UDP-GlcNAc (green mesh, contoured at 3σ) and 4 Å anomalous difference Fourier electron density (magenta mesh, contoured at 15σ) from a dataset with MnCl2.
(E) Recognition of uridine moiety of UDP-GlcNAc.
(F) Schematic representation of interactions made by UDP-GlcNAc.
(G) Tunicamycin binding in active site.
(H) Schematic representation of interactions made by tunicamycin.

Binding Mode for UDP-GlcNAc and Metal Ion
The structure of the DPAGT1/UDP-GlcNAc complex reveals an overall stabilization of the active site, due to movements of CL1 and CL9, and the N terminus of TMH4 (Figure 2C), without any global changes in conformation. The C-terminal end of TMH9b and the following loop region (Phe286-Ile304) display the largest conformational change with an induced fit around the GlcNAc-PP (Figure 2C).
The uridyl moiety of UDP-GlcNAc lies in a narrow cleft at the back of the active site formed by CL5 and CL7. The uracil ring is sandwiched between Gly189 and Phe249 with additional recognition conferred by hydrogen bonds between the Leu46 backbone amide, the Asn191 sidechain to the uracil carbonyls (Figures 2D and 2E) and an extensive hydrogen bond network involving two waters links the uracil ring to five residues (Figures 2D and 2E). Hydrogen-bonding of the ribosyl hydroxyls to the Gln44 mainchain carbonyl and Glu56 sidechain carboxylate complete the recognition of the uridyl nucleotide.
The pyrophosphate bridge is stabilized by interactions with Arg301 and by the catalytic Mg2+ (Figure 2D). The Arg301 sidechain coordinates one pair of α and β phosphate oxygen atoms, whilst the Mg2+ ion is chelated by the second pair of α and β phosphate oxygens. Each oxygen atom is thus singly coordinated in an Arg-Mg2+-“pyrophosphate pincer.” The octahedral coordination of the Mg2+ ion is completed by the sidechains of conserved residues Asn185 and Asp252, and 2 water molecules. Data from DPAGT1 co-crystallized with UDP-GlcNAc and Mn2+ gave a single anomalous difference peak at the metal ion binding site, confirming the presence of a single Mg2+ ion in the active site (Figure 2D). The position of the Mg2+ ion differs by 4 Å from that observed in MraY unliganded structure and the co-ordination differs (Chung et al., 2013).
The GlcNAc moiety-binding site is formed by the CL9 domain and the CL5 loop, although all the direct hydrogen bond interactions are with the CL9 domain. The OH3 and OH4 hydroxyls of GlcNAc form hydrogen bonds with the sidechain of His302 and the mainchain amide of Arg303, respectively. The mainchain of residues 300–303 and the sidechain of Arg303 define the GlcNAc recognition pocket by specifically recognizing the N-acetyl substituent, forming a wall to the sugar-recognition pocket that appears intolerant of larger substituents, thereby “gating” substrate. This structure is absent in MraY, which has a much smaller CL9 loop.
Surprisingly, this structure does not support prior predictions that the highly conserved “aspartate rich” D115Dxx(D/N/E)119 motif is directly involved in Mg2+ binding and/or catalysis (Lloyd et al., 2004). This sequence is adjacent to the active site, but these residues do not directly coordinate Mg2+ or substrate (Figure 2D). Instead, Asp115 is hydrogen-bonded to Lys125 and Tyr256. Lys125 lies adjacent to the phosphates (Figure 2D) and has been implicated in catalysis (Al-Dabbagh et al., 2008). Asp116 forms hydrogen bonds to Ser57 and Thr253 and N-caps TMH8, thus stabilizing residues that interact with the UDP ribosyl moiety (Glu56) and the Mg2+ (Asp252). DPAGT1 with residues Asp115 and Asp116 mutated to Asn, Glu, or Ala retained at least 10% of WT activity (Figure 3A), suggesting they are not essential for catalysis. The third residue in this motif, Asn119, makes no significant interactions. Thus, 2 of the 3 conserved residues perform structural roles; none are directly involved in Mg2+-binding or catalysis.
Figure 3 Proposed DPAGT1 Catalytic Mechanism
(A) Relative activity of active site mutant residues; (n=9).
(B) UDP-GlcNAc complex active site structure with Dol-P modelled based on tunicamycin complex lipid chain position.
(C) Proposed DPAGT1 catalytic mechanism.
For all panels, data presented are means ± SD.

Comparison of the UDP-GlcNAc and Tunicamycin Complexes
The structure of the complex between tunicamycin and DPAGT1 (this work and Yoo et al., 2018) suggests that inhibition is achieved through partial mimicry of the Michaelis complex formed during catalysis with acceptor phospholipid Dol-P and UDP-GlcNAc. The uridyl and GlcNAc moieties in tunicamycin and UDP-GlcNAc occupy essentially identical sites (Figures 2G and 2H). In tunicamycin the pyrophosphate of UDP-GlcNAc is replaced by a galactosaminyl moiety, which displaces the Mg2+ and interacts with the sidechains of Arg301, Asp252, and Asn185.
Tunicamycin’s mimicry of Dol-P also gave critical insight into the potential binding of co-substrate Dol-P. The lipid chain of tunicamycin occupies the concave groove that runs along TMH5, between helices TMH4 and TMH9a (see above). The sidechain of Trp122 pivots around its Cβ–Cγ bond to lie over the lipid chain, trapping it in a tunnel. The surface beyond this hydrophobic tunnel, up to the EL4 loop on the ER lumen face of the membrane is highly conserved, so it is likely that the Dol-P lipid moiety could bind to this surface. At the other end of the tunicamycin lipid moiety, the amide forms polar interactions with Asn185 and lies close to Lys125, suggesting that the amide moiety partially mimics the phosphate head-group of Dol-P (Figures 2G and 2H).

The DPAGT1 Catalytic Mechanism
Alternative mechanisms have been proposed for the PNPT family including one-step, nucleophilic attack (Al-Dabbagh et al., 2008) or two-step, double displacement via a covalent intermediate (Lloyd et al., 2004). We did not observe any covalent modification of DPAGT1 in the presence of UDP-GlcNAc, nor did we see release of UMP in the absence of Dol-P as might be predicted for a two-step mechanism. Also, the active site lacks suitably placed residues that could act as a nucleophile. Therefore the most probable mechanism involves direct nucleophilic attack by Dol-P phosphate oxygen atom on the phosphorus atom of the β-phosphate of UDP-GlcNAc, causing phosphate inversion and loss of UMP (Figures 3B and 3C).
When bound to DPAGT1, UDP-GlcNAc adopts a bent-back conformation, with the donor sugar lying below the phosphates, rotated towards the uridine (Figure 2D). The pyranose ring is inclined so that the O6 hydroxyl of the GlcNAc is within 3.1 Å of the O5B atom of the α-phosphate. This orientation presents the β-phosphate of the UDP-GlcNAc to the position that would be occupied by the phosphate of the Dol-P, exposing the lowest unoccupied molecular orbital (LUMO) of its β-phosphate electrophile for reaction with the Dol-P phosphate O-nucleophile (Figures 3B and 3C).
Providing the correct geometry for this one-step phosphoryl transfer appears to be key to catalysis. Analyses of other enzymatic phosphoryl transfer reactions suggest that a bridging Arg (in a very similar position to Arg301) and bridging metals (in a very similar position to the Mg2+ ion) do not tighten the transition state but instead provide binding energy that optimizes geometry and alignment for attack (Lassila et al., 2011). They may also preferentially favor the formation of trigonal bipyramidal geometry during nucleophilic attack. Similarly, despite classical emphasis on reducing electrostatic repulsions between anionic nucleophiles with anionic electrophiles, such as those present in phosphoryl transfer (Westheimer, 1987), such effects are small in model systems (Lassila et al., 2011). This suggests that the role of Lys125, which would be close to the phosphate oxygens in Dol-P, would be mainly to guide the phosphate into position. The correct alignment of Dol-P for attack would be further facilitated by the “grip” provided by Trp122 holding the Dol-chain into the tunnel observed in the tunicamycin⋅DPAGT1 complex.
Representative residues proposed to bind Dol-P, sugar and pyrophosphate were probed by mutagenesis. Mutation of Mg2+-chelating residues to Ala (Asn185Ala and Asp252Ala) reduced DPAGT1 activity to 1.2% and 7%, respectively (Figure 3A). A conservative Asn185Asp mutation, expected to retain Mg2+-binding activity, also ablated activity (0.7% of WT) suggesting an additional role for Asn185 in catalysis. The amide group of Asn185 lies within 4 Å of the predicted Dol-P phosphate-binding site, forming hydrogen bonds with the nucleophilic oxygen of Dol-P to guide it towards the β-phosphate. Lys125 Mutations (Lys125Ala, Lys125Glu, and Lys125Gln) near the Dol-P phosphate binding site, all reduced the activity to below 2.2%, consistent with it playing a critical guiding role. Interestingly, an Asp252Asn mutation increased activity 5-fold (Figure 3A). This mutation removes a coordinating negative charge from the Mg2+, making it more electropositive and the β-phosphorus more electrophilic, potentially increasing its susceptibility to nucleophilic attack.
Mutation of His302, which hydrogen bonds to the O4 oxygen of GlcNAc in UDP-GlcNAc to hold it in its bent-back conformation, caused 98% loss of activity, consistent with it playing a role in aligning the nucleophile to attach the β-phosphate. Finally, mutations of Arg301, found in patients with CDG-Ij (Imtiaz et al., 2012), that is part of the “pyrophosphate pincer” caused ∼95% loss of activity.

Mutations in DPAGT1 in CMS and CDG
DPAGT1-CMS and CDG-Ij are recessive disorders, caused by loss of DPAGT1 function (Table S1), through loss of enzymatic activity, protein truncation or RNA splicing defects. Structures, catalytic activity (Figure 4A), and thermostability measurements (Figure 4B) for purified mutant proteins allowed us to examine how DPAGT1 variants cause disease (Figures 4 and S3, Table S1). Surprisingly, in most cases mutated DPAGT1 protein had close to WT thermostability (Figure 4B), suggesting correct folding. Only two (Ile29Phe and Arg218Trp, Basiri et al., 2013, Iqbal et al., 2013) gave protein that was too unstable to purify, due to insertion of large hydrophobic sidechains either at the ER surface or in the core of the protein (Figure 4C).
Figure 4 DPAGT1 CMS and CDG-Ij Missense Variants Analyses
(A) Relative activity and (B) thermostability for DPAGT1 missense variants found in CMS (dark grey) or CDG-Ij (mid grey). WT activity/stability is indicated by an orange line. Mutations are colored in panels A, D, E-G as follows: 25% or less activity: red; 50%–100% WT activity: pink; >WT activity: green; instability on purification: cyan; confirmed splicing variants indicated with a ∗; Spl: unknown splicing variant. For all measurements n=9.
(C) Location of variants mapped onto tunicamycin (purple) complex, colored as above.
(D) Changes in thermostabilization of selected variants by substrates UDP-GlcNAc and Dol-P (n=9).
(E) Location of variants near UDP-GlcNAc binding site.
(F) Variants near predicted Dol-P binding site. Dol-P (pale green) has been modelled based on the tunicamycin lipid tail.
(G) Environment of Val264. Structures of WT (grey) and V264G variant (orange) are shown illustrating the subtle conformational difference in TM4b and EL4.
For all panels data presented are means ± SD.
Figure S3 The c.478G>A; Gly160Ser and c.791T>G; Val264Gly Mutations Are Associated with Exon Splicing Errors, Resulting in the Loss of Exons 2 and 3 from the Transcript Harbouring c.478G>A; Gly160Ser and the Loss of Exons 6 and 7 from the Transcript Harbouring c.791T>G; Val264Gly, Related to Figure 4
(A) Schematic of pET01 exon trap vector with DPAGT1 exons 2-4 inserted showing the location within the genomic sequence of c.478G>A; Gly160Ser.
(B) Sequencing data and schematic diagrams showing aberrant splicing that results from genomic sequence harbouring the c.478G>A; Gly160Ser variant. RT-PCR on RNA produced in TE671 muscle cell line following transfection with the “exon trap” vector gave wild type RNA sequence, but also some transcripts that excluded exons 2 and 3. When the genomic sequence contained the c.478G>A; Gly160Ser variant was transfected only RNA missing exons 2 and 3 was detected.
(C) Schematic of pET01 exon trap vector with DPAGT1 exons 6-8 inserted showing the location within the genomic sequence of the c.791T>G; Val264Gly variant.
(D) Sequencing data and schematic diagrams showing sequences obtained following RT-PCR on TE671 cells transfected with the “exon trap” vector containing human genomic DNA that is either wild type or has the c.791T>G; Val264Gly variant. Wild type sequence generated RNA harbouring only exons 6, 7 and 8, as shown. The c.791T>G; Val264Gly variant generated some RNA transcripts containing exons 6, 7 and 8 (as for wild type) but for the majority of transcripts exons 6 and 7 were excluded and only exon 8 was present.
Many CMS patients have compound heterozygous mutations, and in general one allele had a much greater impact on DPAGT1 function, with either loss of at least 75% of the WT catalytic activity (Leu120Met, Gly192Ser, Met108Ile, Figure 4A, Table S1) or protein quantity (e.g., truncation: Thr234Hisfs∗116, (Belaya et al., 2012); low protein yield: Ile29Phe (Klein et al., 2015); exon skipping: Leu120Leu (Selcen et al., 2014). However, expressed protein from the second allele had 50% or more WT catalytic activity (Val117Ile, Pro30Ser, Val264Met) or even increased activity (Val264Gly and Gly160Ser, Belaya et al., 2012) (Figure 4B, Table S1). The only homozygous CMS mutation, Arg218Trp (Basiri et al., 2013), gave protein that was too unstable to purify. Patients with the more severe CDG-Ij disease often had significant loss of function for both alleles, with less than 25% activity (Leu168Pro, Tyr170Cys, Arg301His, Arg301Cys), low protein yield/stability (Ile29Phe), or disrupted splicing (with WT activity and thermostability for the purified protein, e.g., Ala114Gly, Würde et al., 2012). The Leu385Arg (Carrera et al., 2012) and Ile69Asn (Timal et al., 2012) mutations are unusual in that they did not show loss of stability or activity, suggesting that splicing changes should be investigated.
As expected, mutations with less than 25% activity generally lie in or near the substrate-binding regions. In two variants found in patients with CDG-Ij (Arg301His and Arg301Cys, Carrera et al., 2012, Imtiaz et al., 2012), the Arg301 that “pincers” the UDP-GlcNAc pyrophosphates is mutated (Figures 2D and 2F) and these mutations gave reduced thermostabilization by UDP-GlcNAc (Figure 4D), confirming a role in UDP-GlcNAc binding. Similarly, the CMS variant Leu120Met lies close to the Dol-P/UDP-GlcNAc interface on the critical CL3 loop that forms one side of the active site (Figure 4E) and the variant Gly192Ser (Belaya et al., 2012) lies within the highly conserved CL5 loop in the uridyl recognition pocket (Figure 4E) where it would disrupt substrate binding. Two CDG-Ij mutations (Leu168Pro and Tyr170Cys, Iqbal et al., 2013, Wu et al., 2003) are of particular interest as they lie on the ER side of the membrane, at the top of the predicted Dol-P binding site, an extensive, conserved intramembrane groove between TMH4, TMH5, and TMH9, below the EL4 loop (Figure 4F). Both mutations cause a reduction in thermostabilization by Dol-P, confirming their involvement in Dol-P binding. The Met108Leu mutation lies within a hydrophobic cluster in the core of DPAGT1 where a mutation would disrupt the positions of TMH4 and TMH5, which form the back of the Dol-P binding site (Figure 4F). Conversely, many of the mutations that maintain at least 50% of WT activity lie on the protein surface (e.g., Pro30Ser, Ala114Gly, Val117Ile) and involve small sidechain changes (Figure 4C).
Given that DPAGT1-CMS is a recessive disorder, we were surprised to find an increased enzymatic activity with the Val264Gly and Gly160Ser variants. Val264 is mutated to either Gly or Met (Selcen et al., 2014), giving either a 2.5-fold increase or a slight (18%) decrease in catalytic activity. Val264 is located on TMH8 in the core of the protein adjacent to TMH3/4 (Figure 4G). Comparing the WT and Val264Gly structures showed a small (1–1.5 Å) movement at the C-terminal end of TMH4b towards TMH8 (Figure 4G), which would affect both the dimer interface and the exact position of EL4, which forms the top of the Dol-P lipid-binding site. Conversely, the Val264Met variant would be poorly accommodated at this buried site. As the Gly160Ser mutation lies in the disordered EL4 luminal loop (Figure 4F), it is unclear why the activity increases. Since these missense variants cause an unexpected increase in enzyme activity, we explored other causes of pathogenicity. Tissue from muscle biopsies were unavailable, so we used the “exon trap” system to detect abnormal RNA splicing. Both mutations, c.478G>A, p.Gly160Ser, and c.791T>G, p.Val264Gly, gave rise to abnormal RNA splicing of their respective RNA transcripts (Figure S3), thus explaining the pathogenicity of these variants.

Development of Non-toxic “TUN-X,X” Analogues of Tunicamycin
The potent “off-target” inhibitory effects of tunicamycin on DPAGT1 prevent its use as an antibiotic. We used structural data coupled with genetics to design analogues (TUN-X,X) that retained anti-microbial activity yet no longer inhibited DPAGT1. We previously cloned and sequenced the tunicamycin biosynthetic gene cluster (tun) from Streptomyces chartreusis and expressed it heterologously in Streptomyces coelicolor (Wyszynski et al., 2010, Wyszynski et al., 2012). The cluster contains 14 genes, tunA-N. In-frame deletion mutations in all of the cloned tun genes in S. coelicolor (Widdick et al., 2018) revealed new insights into tunicamycin biosynthesis. Interestingly, deletion of tunI and tunJ, encoding components of an ABC transporter conferring immunity to tunicamycin, could only be achieved with mutations elsewhere in the tun gene cluster. Sequencing of one of the ΔtunI mutants revealed a G-to-A missense suppressor mutation in tunC, resulting in a Gly70Asp substitution in the N-acyltransferase TunC that attaches the lipid of tunicamycin. This presumed loss of function mutation abolished antibacterial activity, consistent with a key role for the lipid chain in the biological activity of tunicamycin.
We therefore designed a semi-synthetic strategy to access systematically “lipid-altered” variants of tunicamycin based on tunicamine scaffold TUN (Figure 5A). Large-scale fermentation of S. chartreusis NRRL 3882 (Doroghazi et al., 2011) (Methods S1) allowed access to crude tunicamycin on a multi-gram scale. Degradative conversion (Ito et al., 1979) of tunicamycin gave unfunctionalized core scaffold TUN. Critically, since the nucleobase of tunicamycin is hydrolytically sensitive, the creation of mixed Boc-imides at positions 10ʹ and 2ʹʹ allowed mild, selective deamidation on a gram-scale (see Supplemental Information SI 2). Chemoselective carbodiimide- or uronate-mediated acylation allowed direct lipid-tuning in a systematic, divergent manner through modification at 10ʹ-N and/or 2ʹʹ-N, yielding a library of novel analogues, TUN-X,X varying in chain length by one carbon, from C7 to C12 (TUN-7,7 to TUN-12,12, Figure 5A) with a typical purity of >99% (see Methods S1) as judged by NMR and/or HPLC.
Figure 5 Semi-synthetic Synthesis and Antibacterial Effects of TUN-X,X Analogues
(A) Semi-synthetic strategy for TUN mimics.
(B–D) MIC obtained from micro-broth dilution antimicrobial susceptibility tests of (B) B. subtilis EC1524, (C) B. cereus ATCC11778, (D) Mtb H37Rv (ATCC27294) cultured in 7H9/ADC/Tw (black), or GAST/Fe (grey) media (n=3). For all panels data presented are means ± SD

TUN Analogues Show Potent Antimicrobial Activity against a Range of Bacteria
We evaluated the analogues (TUN-7,7, -8,8, -9,9, -10,10, -11,11, -12,12) for potency against a range of Gram-negative and Gram-positive bacteria that cause infections in hospitals (Staphylococcus, Pseudomonas, and Escherichia), as well as reference bacteria (Bacillus, Staphylococcus, and Micrococcus) used in previous tunicamycin bioactivity studies (Ward, 1977). Kirby-Bauer disc diffusion susceptibility tests (Figures S4A–S4E), revealed potent activity against Bacillus subtilis (EC1524) and opportunistic pathogen Bacillus cereus (ATCC 11778). There was a weaker but significant effect on the pathogenic bacteria Staphylococcus aureus (ATCC 29219) and Pseudomonas aeruginosa (ATCC 27853); the latter is a strain resistant to natural tunicamycin. No activity was seen against Micrococcus luteus, a bacterium noted to have some resistance to tunicamycin (Ward, 1977). Consistent with the critical role of the lipid, none of the non-lipidated analogues (e.g., TUN or TUN-Ac,Ac) or synthetic intermediates showed any activity. Lipid-length (X = 7, 8…12) in the TUN-X,X analogues systematically modulated activity; the most potent analogues TUN-8,8 and TUN-9,9 were those with C8 and C9 chain lengths.
Figure S4 Kirby-Bauer Disc Diffusion Tests and Dose Response Curves Used for MIC, MBC, and IC50 Determination for Tunicamycin and the TUN-X,X Analogues against Several Bacterial Strains, Related to Figure 5
(A–E) Kirby-Bauer disc diffusion tests for the TUN analogues against bacterial strains (A) B. subtilis EC 1524, (B) M. luteus, (C) B. cereus ATCC 11778, (D) S. aureus ATCC 29219 and (E) P. aeruginosa ATCC 27853. Discs impregnated with 5 μg, unless otherwise indicated, of the compound were laid onto plates with lawns of bacteria. The compounds are labelled: TC: commercial tunicamycin, TS: tunicamycin from S. chartreusis, 2: (2) N-acetyl tunicamine, Ac: TUN-Ac,Ac, Boc: TUN-Boc,Boc, 7: TUN-7,7, 8: TUN-8,8, 9: TUN-9,9, 10: TUN-10,10, 11: TUN-11,11, 12: TUN-12,12, Cit: TUN-Cit,Cit, Amp: ampicillin, Gen: gentamycin.
(F–J) Dose response curves for tunicamycin and the analogues with (F) B. subtilis EC1524. (G) B. cereus ATCC11778. (H) S. aureus ATCC29219. (I) P. aeruginosa ATCC27853 and (J) E. coli ATCC25922. These dose-response curves were generated by Prism 6.0 software by plotting percent growth (normalized OD600 values) vs. logarithmic scale of the concentrations. The data shown are mean ± SEM errors of three independent experiments. See Table S3 for the MIC, MBC and IC50 values.
The minimal and half maximal inhibitory concentrations (MIC and IC50) and minimal bactericidal concentrations (MBC) were determined by both a micro-broth dilution test and drop plate test, respectively (Figures 5B and 5C, Figures S5F–S5J, Table S3). Only lipidated variants (tunicamycin and TUN-X,X) displayed antibacterial activity, with MICs down to 0.02 ± 0.01 μg/ml for TUN-9,9 against B. subtilis and 0.33 ± 0.11 μg/ml against B. cereus, with TUN-10,10.
Figure S5 On-Target Effects Demonstrated for the TUN-8,8 Analogue, Related to Figure 5
(A) Glucosamine incorporation within 48 h was reduced with 10x TUN-8,8. Positive controls were tunicamycin, meropenem/clavulanate (MCA) and D-cycloserine (DCS). All compounds tested at 1- and 10-fold MIC concentrations (n=2).
(B) GFP release assay. Mtb expressing GFP was treated with 1- and 10-fold MIC concentrations of TUN-8,8 or tunicamycin. Lysis was monitored by measurement of fluorescence in cell-free supernatants daily over a week of exposure as a measure of release of cytosolic protein (GFP) (n=2).
(C) Confocal microscopy of BODIPY-vancomycin used to track synthesis of PG, showing abnormal PG formation with tunicamycin and TUN-8,8 treated Mtb, both differing from the effects of pentapeptide labelling by the BODIPY-vancomycin caused by meropenem/clavulanate treatment.
Data presented in (A) and (B) are means ± SD.
TB, through its etiological agent Mycobacterium tuberculosis (Mtb) is a global concern. Testing of the lipid-altered analogues (TUN-7,7 to -12,12) against pathogenic Mtb strain H37Rv revealed striking MIC values (0.03 ± 0.001 μg/ml in minimal growth medium and 0.22 ± 0.02 μg/ml in rich 7H9-based growth medium) for TUN-9,9: some 5-fold more potent than even tunicamycin itself (Figure 5D, Table S4).
The on-target effect of tunicamycin analogues was explored by multiple approaches. Firstly, TUN-8,8-resistant Mtb mutants (Methods S1) carried mutations in Rv0751c (mmsB, 3-hydroxyisobutyrate dehydrogenase), an enzyme in fatty acid metabolism, suggesting possible inactivation of TUN-8,8 by destruction of fatty acid/lipid or an intergenic mutation between Rv2980 and Rv2918c (the central D-alanine-D-alanine ligase ddlA involved in peptidoglycan (PG) synthesis), suggesting possible regulation of this key PG biosynthetic enzyme (perhaps via small regulatory noncoding RNA). Secondly, macromolecular incorporation assay using 14C-glucosamine as radiolabeled precursor (Figure S5A) confirmed that TUN-8,8 and tunicamycin have similar effects on PG biosynthesis, suggesting that they act on the same target. Thirdly, extracellular release of green fluorescent protein (GFP) from Mtb expressing GFP showed lytic effects of TUN-8,8, even outstripping those of tunicamycin (Figure S5B), consistent with on-target PG inhibition activity. Finally, fluorescently-labeled vancomycin (Daniel and Errington, 2003, Tiyanont et al., 2006) revealed similar phenotypes (Figure S5C) following treatment with TUN-8,8 or tunicamycin with disrupted, non-uniform cell-wall, consistent with targeting of the PG pathway. Together these data suggested that TUN-8,8 is inhibiting the same target as tunicamycin in Mtb.

Lipid-altered TUN-X,X Analogues are Non-toxic to Eukaryotic Cells
We evaluated the effect of TUN-7,7 to -12,12 and corresponding synthetic intermediates on representative human cell lines from liver (HepG2), kidney (HEK293), and blood (Raji) cells where 24 hour incubation with tunicamycin showed both clear cytotoxicity (Figures 6A–6C, Figure S6) and morphological changes (Figure 6D, Figures S6B–S6D) consistent with prior observations (Takatsuki et al., 1972). Cell-cycle analysis (Figures 6E and S6A) suggested cell death coincides with decline in G0/1 phase populations with an LC50 ∼100 μg/ml.
Figure 6 TUN-X,X Analogues are Not Toxic to Cultured Human Cells
(A–C) Dose response curves from cell proliferation assays with (A) HEK293, (B) HepG2 and (C) Raji cells with tunicamycin and analogues (n=3 in each case).
(D) Effect of 400 μg/ml (saturating) tunicamycin, TUN-8,8 and TUN-9,9 on HEK293 cell morphology.
(E) Effects of tunicamycin, TUN-8,8 and TUN-9,9 on HEK293 cell cycle (n=3).
(F) Effects of tunicamycin and TUN-X,X analogues on DPAGT1 catalytic activity (n=9).
(G) DPAGT1 lipid-binding site, showing the restrictive tunnel with tunicamycin bound.
(H) More open MraY lipid-binding site (PDB: 5JNQ), with modelled lipid chains shown with lower contrast.
For panels (A–C) and (E), data presented are means ± SEM. For (F), data presented are means ± SD.
Figure S6 Effect of Tunicamycin and Analogues on HEK293 Cells, Related to Figure 6
(A) Cell cycle analysis at 24 hours.
(B–D) Cell morphology with (B) tunicamycin, (C) TUN-8,8 and (D) TUN-9,9.
(E–G) Effects of (E) tunicamycin, (F) TUN-8,8 and (G) TUN-9,9 on glycosylation of a model protein – IgG1Fc-His6. Black arrow indicates glycosylated protein, white arrow indicates unglycosylated protein.
Consistent with a mode of action requiring lipidation for toxicity, all non-lipidated variants (TUN core and synthetic intermediates) displayed no significant adverse effects; these variants do not act upon either bacteria or mammalian cells in any potent manner.
In contrast to tunicamycin’s toxicity (LD50 = 51.25 ± 31.27, 44.74 ± 4.73, 26.82 ±11.46 μg/ml for HEK293, HepG2 and Raji cells, respectively, Figures 6A–6C and Table S5) the designed TUN-X,X variants TUN-7,7 to -12,12, with altered lipids, showed mild or negligible toxicity (LD50 > 400 μg/ml, saturation) towards mammalian cells. A high level (>75%) of viable cells with no morphological changes were observed after 24 hours (Figure S6A) for HepG2 or HEK293 cells (400 μg/ml, saturation). Moreover, no variation in cell cycle was observed, with healthy G0/1 populations maintained at even highest concentrations (Figure 6E, Figure S6A).
The mechanistic origin of reduced toxicity was tested in vitro with purified DPAGT1 enzyme. Whilst native tunicamycin completely inhibited DPAGT1, TUN-8,8 and TUN-9,9 had a negligible effect (Figure 6F), consistent observations for glycosylation of model protein (Figures S6E–S6G). Notably, synthetic reinstallation of a single C8 lipid into analogue TUN-8,Ac restored inhibitory activity towards DPAGT1 (Figure 6F), consistent with our design and the critical role of the second lipid preventing binding to DPAGT1. Given that in patients with CMS, a loss of the activity of one allele is not sufficient to cause disease, it is likely that a reduction in activity by <10%, caused by the TUN-X,X analogues, is unlikely to cause significant toxicity during a short-term treatment. Together these results confirmed our hypothesis that systematic “lipid alteration” could create tunicamycin analogues in which mammalian cytotoxicity is separated from antibacterial effects.

A Molecular Explanation for Differences in TUN-X,X Analogue Binding to DPAGT1 and MraY
Comparison of the structures of the complexes of tunicamycin with DPAGT1 and MraY (Hakulinen et al., 2017, Yoo et al., 2018; this work) gave an explanation for selectivity of analogues on MraY over DPAGT1. The MraY tunicamycin binding site has a more open, shallow surface than in DPAGT1; in the latter the lipid tail is completely enclosed by Trp122 adjacent to the active site (Figure 6G). The MraY binding site has a disordered loop CL1, a longer TMH9 and a relatively short CL9 region, with only one short α-helix (Figure 1G). In contrast, DPAGT1 has an ordered CL1 which folds over the UDP-binding site. It has a shorter TMH9, followed by a loop and extended strand (residues Gln292 to Arg306) (Figure 1F), which fold over tunicamycin, forming numerous interactions, (e.g., with Arg301, His302, Arg303, Figures 2G and 2H). This extended structure is stabilized by its interactions with the rest of the CL9 domain, a feature found only in eukaryotes.
The N2′′ in GlcNAc is the attachment site of the second lipid chain in TUN-X,X analogues—it occupies distinct environments in the two proteins. In DPAGT1 it is enclosed by the loop at the end of TMH9, and by a tight “gating” cluster of side chains from Trp122, Ile186, Leu293, Cys299, and Arg303 (Figure 6G). By contrast, in MraY, there is a 10 Å gap between Pro108 on TMH4a and Val272 on TMH9, providing ample space for more than one lipid chain to be attached to the amines in TUN-X,X analogues (Figures 6G and 6H).

Lipid-altered TUN-X,Xs Show Efficacy against Mtb in Mice
The enhanced therapeutic index of the TUN-X,X analogues in culture (Table S6) suggested strong potential in treating mammals. Toxicity and efficacy were probed in infection models of Mtb both in cellulo and in vivo. First, as a stringent in cellulo test of the ability to treat infection, TUN-8,8, TUN-9,9, TUN-10,10, and TUN-11,11 were used to treat Mtb-infected macrophages (Figure 7A), which showed that these analogues reduced intracellular bacterial burdens by 1- and 2-logs at 1 × and 10 × MIC, respectively.
Figure 7 Tunicamycin Analogues Eradicate Mtb during Host Pathogenesis
(A) Efficacy of tunicamycin analogues in macrophages. Infected J774A.1 macrophages were treated with compounds (1x and 10x MIC) for 3 or 7 days after which bacterial burdens were counted (n=3).
(B) Microsomal stability of the tunicamycin analogues in human (black) and mice (grey) liver microsomes assessed over a period of 30 mins (n=3).
(C) Mouse blood serum TUN-8,8 concentrations after a single intra-peritoneal injection of 30 mg/kg (n=5).
(D) Efficacy of TUN-8,8 in reducing lung bacterial burdens in Mtb-infected mice after 2 weeks of treatment.
(E) Efficacy of TUN-8,8 in reducing spleen bacterial burdens in Mtb-infected mice following 2 weeks of treatment. Unpaired t test for the TUN-8,8 and vehicle control group p value of 0.0017 in lungs and 0.01 in spleens.
For all panels (A–C), data presented are means ± SD. For panels (D) and (E), the individual median calculated for each individual mouse organ with standard error of the mean calculated for each group is shown.
Second, microsomal (human and mouse) stability (Figure 7B) suggested good metabolic survival of TUN-8,8 to -11,11, which was confirmed by in vivo pharmacokinetics in mice (Figure 7C). This revealed good bioavailability of TUN-8,8 following intraperitoneal (ip) delivery with acceptable blood plasma exposures. In human plasma protein binding assays, 98.2% was bound. Next, tolerance testing of TUN-8,8 in uninfected mice (n=5) over 10 days at daily doses of 30 mg.kg-1 (ip) showed no signs of toxicity, in striking contrast to tunicamycin. Finally, antitubercular activity was demonstrated in Mtb-infected mice. Treatment of Mtb-infected mice (n=10) over two weeks (10 mg.kg-1, ip) revealed an almost 10- and 5-fold reduction in bacterial burdens in lungs (Figure 7D) and spleens, respectively, (Figure 7E) compared to mice receiving vehicle control. Notably, despite the tolerability at 30 mg.kg-1 up to 10 days in uninfected animals, clinical signs of toxicity in infected mice precluded any longer-term testing beyond 2 weeks. The origins of this toxicity in diseased animals are unclear but suggest that further optimization of TUN-X,X analogues and/or their formulation, may be critical to disease-weakened animals. TUN-X,X analogues are therefore unoptimized but promising, proof-of-principle, leads rather than, as yet, optimized antibiotic drugs.