Cryo-EM characterization of the anydromuropeptide permease AmpG central to bacterial fitness and β-lactam antibiotic resistance

E. coli AmpG protein production

AmpG permeases are present in many Gram-negative bacteria (Supplementary Fig. 1). An E. coli variant was chosen for study to allow for production in its native lipid environment, often a consideration for yield and stability of polytopic membrane proteins for structure/function analysis. Hexahistidine-tagged full-length wild-type AmpG (Fig. 2a) was expressed in E. coli strain C41ΔompFΔacrB¹⁵ and screened for optimized purification conditions and membrane mimetics. AmpG was successfully solubilized in either N-dodecyl-β-D-maltopyranoside (DDM) or MSP1D1 nanodiscs¹⁶ (9 nM diameter) with incorporated E. coli polar lipids. The purity and monodispersity of AmpG after removal of the recombinant hexahistidine-tag was confirmed by SDS-PAGE and negative-stain TEM (Supplementary Fig. 2a, b). To aid structure determination of this small ~53 kDa membrane protein with cryo-EM, thermostabilized apocytochrome b562 (BRIL), a 106 residue four helical bundle¹⁷, was included at the C-terminus and used as an epitope for the synthetic antibody BAG2¹⁸. The high affinity interaction between AmpG-BRIL and BAG2 was verified with surface plasmon resonance (SPR) with a binding constant (K_d) of 9.6 nM (Supplementary Fig. 3e). AmpG-BRIL in DDM was incubated with BAG2 at a 1:1 molar ratio and complex formation was validated by glycerol gradient centrifugation, pull-down assay, native PAGE, and negative-stain TEM (Supplementary Fig. 2c–h).

Synthesis of GlcNAc-1,6-anhydroMurNAc-peptide analogs

The chemical syntheses, purification, and validation of several AmpG GlcNAc-1,6-anhydroMurNAc substrate variants was pursued for this study: compound 1, or GlcNAc-1,6-anhydroMurNAc, is the minimal structural unit necessary for recognition by AmpG¹⁴, compound 2 is GlcNAc-1,6-anhydroMurNAc bearing L-alanine, compound 3 is GlcNAc-1,6-anhydroMurNAc bearing L-alaninamide, and compound 4 is GlcNAc-1,6-anhydroMurNAc bearing a full L-Ala-D-iso-Glu-meso-oxa-Dap-D-Ala-D-Ala stem pentapeptide (Fig. 1b).

The chemical synthesis of 1 has been known since 1986 when first published by the Paulsen group¹⁹. Many other groups have since developed their own synthetic routes to the GlcNAc-1,6-anhydroMurNAc motif to study proteins and biochemical pathways that use GlcNAc-1,6-anhydroMurNAc glycopeptides as substrates or agonists: measuring the turnover of lytic transglycosylases²⁰, investigating the effect of tracheal cytotoxin (TCT) on the Nod1 receptor²¹, and developing a fluorescent assay for AmpG²². Our synthetic route is a slight modification of Paulsen’s¹⁹ and is described in a recent publication by the Tanner group²³. We produced intermediate peracetyl GlcNAc-1,6-anhydroMurNAc methyl ester sugar S1, which yields 1 quantitatively after saponification and an acidic resin workup (Supplementary Fig. 4a).

Disaccharide 1 could be converted into the novel L-alanine-bearing disaccharide 5 (Supplementary Methods) by a preliminary coupling reaction to L-alanine benzyl ester via PyBOP/HOBt and a subsequent tosic acid-catalyzed acetonide protection. This protected sugar intermediate 5 is produced in modest yields but is far easier to isolate and purify via flash chromatography than its amphiphilic benzyl ester counterpart lacking a GlcNAc acetonide. Compound 5 may be deprotected via acidic resin treatment followed by hydrogenolysis to give substrate 2 in quantitative yields, or amidated by ammonia in methanol and deprotected by acidic resin to give substrate 3 in excellent yields (Supplementary Fig. 4a).

The chemical synthesis of disaccharide pentapeptide 4 is far more challenging than that of compounds 1–3, mostly due to the separate chemical synthesis of the meso-oxa-Dap pentapeptide. Meso-oxa-Dap is an isosteric analog of the natural meso-diaminopimelic acid (meso-Dap) found in the stem peptides of various Gram-negative bacteria such as E. coli and P. aeruginosa and used as a “handle” by their PBPs to cross-link layers of cell wall PG polymers²⁴. meso-Dap is a reasonably challenging synthetic target with many different lengthy synthetic routes already published²⁵. Moreover, synthetic routes to meso-Dap as part of a PG stem peptide must be able to selectively incorporate only the L-stereocenter of the symmetric meso-Dap into the oligopeptide backbone. Orthogonally-protected meso-oxa-Dap was first produced by the Vederas group via a Lewis acid-catalyzed aziridine ring opening with a serine nucleophile²⁶. The succinct synthesis of this meso-Dap analog makes meso-Dap peptides more attractive synthetic targets in general. Moreover, L,L-oxa-Dap proved to be an adequate substrate for Dap-epimerase²⁶ and meso-oxa-Dap oligopeptides were successfully processed by the Gram-negative carboxypeptidases Csd6 and Pgp2²⁷; thereby lending credence to meso-oxa-Dap’s viability as a structural analog to meso-Dap for enzymatic/biological experiments.

The synthesis of 4 along with the significance and various pitfalls were discussed prior²³. Vederas’ methodology was extended to the nucleophilic ring opening of an aziridine already embedded within an oligopeptide. Disaccharide S2 was coupled to meso-oxa-Dap-pentapeptide P1 via EDCI/HOBt – S2 was also derived from S1, and P1 was derived from its N-terminus Boc-protected analog. The resultant protected glycopeptide S3 underwent hydrogenolysis and subsequent deacetylation under mildly basic conditions to give the full disaccharide meso-oxa-Dap pentapeptide 4 in good yields (Supplementary Fig. 4b). The identity and validating metrics of all novel compounds verified by ¹H and ¹³C nuclear magnetic resonance spectroscopy (NMR) and by high-resolution electrospray ionization mass spectrometry (HR-ESI-MS) (Supplementary Methods).

Binding analysis of AmpG with synthesized substrate analogs

To minimize protein and ligand requirements and accommodate the requisite use of membrane mimetics for AmpG solubilization, microscale thermophoresis was used for binding studies between DDM-solubilized AmpG and the synthesized substrates. The calculated affinities for GlcNAc-1,6-anhydroMurNAc disaccharide (compound 1) and GlcNAc-1,6-anhydroMurNAc-pentapeptide (compound 4) were similar and in keeping with published values determined using E. coli spheroplast transport assays²². For the shorter peptide variants, specifically the GlcNAc-1,6-anhydro-MurNAc-L-ala with free negatively charged carboxylate (compound 2) or neutral amidated (compound 3) forms, we observed that the negative charge on the truncated peptide was deleterious to binding, with binding only recovered once the charge was neutralized (Supplementary Fig. 3a). Binding of compound 1 to AmpG-BRIL and the AmpG-BRIL BAG2 complex used for structure determination was also assayed showing comparable binding to the wild-type protein (Supplementary Fig. 3d). To rule out differences between detergent micelle-solubilized protein and a more native membrane bilayer-like environment, binding was also assayed using AmpG solubilized in MSP1D1 nanodiscs¹⁶ using isothermal titration calorimetry (ITC), which showed binding values for the GlcNAc-1,6-anhydroMurNAc disaccharide consistent with those from MST (Supplementary Fig. 3b).

Cryo-EM structure of AmpG-BRIL/BAG2 antibody complex

The structure of DDM-solubilized AmpG-BRIL in complex with the BAG2 antibody was determined using cryo-EM to a global resolution of 3.78 Å (Supplementary Fig. 5). Local resolution analysis showed the majority of the protein at 3.0 to 3.5 Å, with clearly resolved side chain densities throughout (Supplementary Fig. 6). The refined model contains AmpG residues 3-486 (the physiological C-terminus), a modeled DDM detergent (used for solubilization), a phosphoethanolamine (PE) lipid, and has excellent stereochemistry (Supplementary Table 1). The C-terminal BRIL fusion and BAG2 antibody were not well resolved in the final reconstruction but were clear in 2D classification (Supplementary Figs. 5, 6). Attempts to determine the structure of AmpG without BRIL and BAG2 were unsuccessful, suggesting their inclusion was important for accurate particle alignment, despite this flexibility. The DDM-solubilized AmpG-BRIL fusion showed similar binding to the GlcNAc-1,6-anhydroMurNAc disaccharide by MST, suggesting inclusion at the C-terminus does not have a deleterious effect on the structure (Supplementary Fig. 3d).

AmpG is stabilized in an outward-open conformation

AmpG was captured in an outwards (to the periplasm) open conformation (Fig. 2b). The structure shows 14 bitopic transmembrane helices, as opposed to the previous predictions of 12 bitopic and two re-entrant helices (Fig. 2c)¹¹. The core of the transporter is composed of two pseudo-symmetric bundles of six transmembrane helices (N-terminal bundle TMs 1–6 and C-terminal bundle TMs 7–12) in the canonical topology of the MFS family²⁸. Two additional antiparallel C-terminal helices, TM13 and TM14, pack against the lateral V-shaped opening to the central cavity formed by TM2 and TM11 at the intersection of the N- and C-terminal helical bundles, creating a unique hydrophobic vestibule (Fig. 2d). Both the N- and C- termini reside on the cytoplasmic face of the membrane and result in a typical distribution of positively charged residues consistent with the positive-inside rule of polytopic membrane protein topology²⁹ (Fig. 2e). The outward-facing conformation has been proposed to be more energetically stable in other MFS symporters e.g., bacterial fucose symporter FucP³⁰, and AmpG is stabilized in the outward-open state by several conserved interactions between TM2 and TM11 (see below) with the central channel sealed at the apex by interactions between TM4, 5, 10, and 11 (Fig. 2f). AmpG contains conserved structural motifs found in both multidrug antiporters and sugar and peptide symporters. However, despite more functional similarity to sugar and peptide symporters FucP and PepT^31,32, analysis with Dali³³ shows more structural similarity to the multidrug exporter family (MdfA, LmrP, YajR, NorA; range of RMSD superpositions on 370 common Cα atoms is 4.8-6.4 Å).

Conserved motif A acts as a molecular latch

Various sequence motifs amongst MFS transporters have been documented²⁸. The most highly conserved is motif A, present in all AmpG homologs and here involving residues at the C-terminus of TM2, the N-terminus of TM3 and the intervening loop (collectively residues 66–80) (Fig. 3a, d). Strictly conserved Asp70 is the key component of the motif, with its carboxylate side chain positioned for electrostatic interaction with the adjacent N-terminal helix dipole of TM11, stabilizing its position in the outward-open state (Fig. 4a). The guanidinium side chains of conserved Arg79 and Arg80, which protrude from the same face at the N-terminus of TM3, both directly coordinate the carboxylate of Asp70, likely enhancing its electronegative charge and the helix dipole interaction (Fig. 4a). Arg80 is further oriented by a direct interaction with the side chain carboxylate of Asp134 on the C-terminus of TM4. Together with Asp70, these residues have been proposed to constitute a charge relay triad critical for function³⁴. Arg81, the third arginine in succession and more unique to the AmpG family (Supplementary Figs. 1, 7), projects towards the exterior where it would be ideally situated to interact with lipid headgroups of the membrane inner leaflet (Fig. 4a). Supporting the importance of motif A in stabilizing the outward-open state, a substitution of Asp70 to alanine in the sequence allowed us to obtain a model of the inward-open conformation using AlphaFold3³⁵. The accuracy of the predicted structure, especially the newly formed interface that closes the periplasmic opening, is supported by tightly co-evolving clusters of residues identified by EVcouplings³⁶ (Fig. 5, Supplementary Fig. 8).

a Sequence conservation logo created using 2936 sequences identified from EVcouplings with a bitscore of 0.7. Plotted in WebLogo3 with key motifs and features annotated (see methods). b Binding cavity interactions with the sugar moiety of DDM. c Conserved titratable residues on the C-terminal helical repeat. Glu326 and Asp238 form a conserved carboxyl-carboxylate pair coordinated next to Lys235. d Conserved titratable residues on the N-terminal helical repeat. Motif A Asp70 coordinated by Arg79 and Arg80 interacting with TM11 helix dipole and substrate binding site Lys132, Asp129, and Asp125. e “Foot-in-the-door” motif, which stabilizes the C-terminal helical bundle and helps create the hydrophobic vestibule.

a AmpG motif A stabilizes the outward conformation with the Asp70 carboxylate forming an electrostatic interaction with the N-terminal helix dipole of TM11, likely enhanced by coordination with Arg79, Arg80, and Asp134. Colored as in Fig. 1d, with mentioned residues marked by *. b Conserved Trp65 and Trp83 in the motif A adaption stabilize the kink in TM2, creating an expanded binding cavity and positioning key residues e.g., Tyr59 and Lys62 for substrate binding. The observed PE lipid extends across the aromatic and nonpolar residues on TM2 and TM4. c Electrostatic surface of the periplasmic substrate binding cavity of AmpG with bound DDM. The maltose sits near an electronegative pocket arising from conserved Asp125 and Asp134, while the hydrophobic acyl chain points into the hydrophobic vestibule between the MFS helical bundle and additional C-terminal helices TM13 and 14. d AmpG substrate binding cavity colored by hydrophobicity. Side chains of disaccharide binding residues shown as sticks and labeled. DDM and lipid are shown as in a. e Density in the binding cavity of AmpG of the modeled bound DDM. f Density consistent with a water coordinated between Lys62, Asp125, and Asp129.

AmpG in the experimental outward open conformation (a) compared to the AlphaFold3³⁵ modeled inward open conformation (b), shown with Coulombic electrostatic potential surface (above) calculated with ChimeraX^35,65 and evolutionary coupled sets of resides (below). Residue pairs on the periplasmic side of the protein, Lys47 and Glu377 (green), are strongly coupled. They are separated ~30 Å in the outward structure but form close contacts in the inward model. At the cytoplasmic side, Asp134 and Ser350 (gold) are also evolutionarily coupled. c Interaction of evolutionary coupled residues E377 and K47 in the modeled inward state. d Structural alignment of N- and C-terminal helical bundles of the experimental outward and modeled inward (gray) states. TM1-TM6 (residues 1–195) RMSD was 1.7 Å, and TM7-TM12 (residues 223-413) RMSD was 1.1 Å.

Extending motif A in our structure, conserved residues specific to the AmpG family are observed immediately up- and downstream of the canonical sequence (Figs. 3a, 4b). Two highly conserved tryptophan residues, Trp65 (TM2) and Trp83 (TM3), appear to play a role in stabilizing the precise position of motif A and the direct substrate-binding residues (see below). The Trp65 indole notably forms a hydrogen bond with the evolutionarily coupled³⁶ Ser123 side chain hydroxyl on TM4 and, together with Pro67, appear to contribute to the significant helical deformation at Val61, with a ~ 125° angular redirection where TM2 and TM11 splay apart to form the lateral entrance to the substrate binding cavity. Characterized homologs have significantly less kinked helices (160-170°) in this region. In AmpG, this pronounced TM2 deformation also serves to orient the near invariant Lys62 along with well conserved Tyr59 to point into the central periplasmic exposed cavity, with a direct role in substrate binding (Fig. 4b).

The outside-open state presents a large binding cavity to the periplasm

In the captured outward-facing conformation, AmpG displays a large funnel-shaped binding cavity extending about three-quarters of the way through the membrane, which is conserved amongst AmpG homologs of clinical significance (Figs. 2f, 4c, and Supplementary Fig. 1). The GlcNAc-1,6-anhydroMurNAc-peptide substrates have an overall net negative charge (Fig. 1b) and we note a general electropositive surface of the cavity, with residues Lys133, Arg153, Lys235, Lys265, and Arg365 positioned to potentially help guide the substrate into the funnel-shaped binding pocket (Fig. 5a, b).

The cavity is sealed to the cytoplasm by multiple interactions, notably between TM4/TM5 and TM10/TM11 from the N- and C-terminal bundles, respectively (Fig. 2d). As discussed above, the interaction between motif A Asp70 on TM2 and the N-terminal helical dipole of TM11 is key to stabilizing this conformation (Fig. 4a). We observe density extending into the cavity interacting with TM2, TM4 and TM5 of the N-terminal bundle that is consistent with a molecule of DDM positioned with its 12-carbon acyl tail pointing into the hydrophobic cleft created by TM13 and TM14 (see below), and its maltose disaccharide head group in the substrate binding cavity (Fig. 4c–e). Based on chemical similarity and binding to key conserved residues, we propose the latter is mimicking binding of the GlcNAc-1,6-anhydroMurNAc disaccharide of the natural substrate. The binding site is defined by the side chains of highly conserved residues Tyr59, Lys62, Asp125, Asp129 and Tyr152 (Fig. 3a). Asp125 (TM4) and Tyr152 (TM5) closely contact the terminating glucose sugar ring hydroxyls of DDM (Fig. 3b). Lys62 on TM2, mentioned above as an AmpG-specific adaption of MFS motif A, is also pointed into the cavity, where it is within hydrogen bonding distance of the first glucose sugar of DDM. Lys62 is partially buried in a remarkably conserved hydrophobic pocket composed of residues Tyr59, Phe63, Trp65, Ala122, Ile126, and Ala355 that restrains its side chain orientation. As a consequence of sugar binding, the lysine amino side chain becomes more buried, with a solvent accessible surface of 16 Ų compared to 51 Ų with the DDM removed^37,38.

Conserved Lys62 and Tyr152 are important for AmpG activity

Lys62, Asp125, Asp129, and Tyr152 are highly conserved (Fig. 3a), including in strains of clinical significance (Supplementary Fig. 1), and the proximity to the bound disaccharide of DDM suggests their importance for AmpG substrate binding and transport. To explore this further, Lys62 and Tyr152 were selected for mutation and tested in a β-lactam susceptibility assay. Although E. coli ampC is not inducible by β-lactams³⁹, E. coli (Ec) ampG can cross-complement an ampG knockout in inducible strain P. aeruginosa PAO1 and restore growth to wild-type levels in the presence of the second-generation cephamycin antibiotic cefoxitin (CEF), due to upregulated AmpC β-lactamase activity (Figs. 1a, 6a). EcAmpG K62A and Y152A mutants both had decreased CEF minimal inhibitory concentrations (MICs), similar to ΔampG complemented with empty vector, suggesting these point mutants are not functional for transport of the PG products required to upregulate AmpC. Increasing the expression of the EcAmpG Y152A but not the K62A mutant with 0.1% arabinose partially restored resistance (Fig. 6a).

a Effect of cefoxitin on growth of a P. aeruginosa PAO1 ampG knockout complemented with E. coli AmpG on low-copy plasmid pHERD30T. Leaky (uninduced) expression of WT ampG (green up-arrow) restored growth to WT levels, while K62A (orange diamond) or Y152A (blue square) mutants had the same expression as empty control (gray down-arrow). Induction of expression with 0.1% arabinose partially restored growth of Y152A but not K62A. b P. aeruginosa AmpG mutants tested as in a. K66A and Y159A affected growth similarly to the ampG knockout (coloring as in a). On induction with 0.1% arabinose, growth of K66A and Y159A was partially restored to varying degrees. c Growth of P. aeruginosa AmpG mutants as in b at varying levels of arabinose induction (gray down-arrow uninduced, blue circle 0.1%, green up-arrow 0.2%, orange diamond 0.5%). Both K66A and Y159A show a concentration-dependent increase in growth on induction, with the response much higher for Y159A. d Identification of a D74N mutant that rescues growth of the P. aeruginosa AmpG K66A mutant. (coloring as in a, blue square represents K66A/D74N). Experiments were performed in technical triplicate and biological duplicate with a representative shown. Data points correspond to technical replicates, with error bands representing SD.

To test the role of these residues in other homologs, we repeated the susceptibility assay using P. aeruginosa (Pa) AmpG. Complementation of the P. aeruginosa ΔampG mutant with the equivalent PaAmpG point mutants (K66A or Y159A) resulted in similar decreases in CEF MICs (Fig. 6b) supporting the functional importance of these residues across the AmpG family. Further, titration of arabinose showed a dose-dependent response with 0.5% arabinose restoring CEF MICs to near wild type levels for Y159A, while growth of K66A was only partially restored, demonstrating mutation at this position is less tolerated for AmpG function (Fig. 6c). To further probe the specific roles of PaAmpG Lys66 and Tyr159, we performed a reverse genetic screen in P. aeruginosa, looking for mutations that could rescue growth of K66A or Y159A mutants in the presence of CEF but not in the presence of piperacillin (PIP). Both antibiotics are substrates of β-lactamase AmpC; however, PIP is only a weak inducer of AmpC⁴⁰. Therefore, resistance to CEF and PIP could indicate a mutation that results in constitutive upregulation of AmpC (e.g., through ampR or ampC promoter mutations) while CEF resistance but PIP sensitivity could indicate a mutation in PaAmpG that restores its function. Remarkably, for K66A we identified multiple suppressors with a D74N mutation that partially restored function (Fig. 6d). P. aeruginosa Asp74 is equivalent to E. coli Asp70 in motif A and the repeated occurrence of the same aspartate to asparagine mutation in combination with K66A suggests a possible mechanistic relationship between these two conserved residues in the substrate binding site and motif A, respectively.

In line with these results, we tested the ability of both EcAmpG K62A and Y159A mutants to bind GlcNAc-1,6-anhydroMurNAc using MST, showing a reduced binding affinity for GlcNAc-1,6-anhydroMurNAc of >2 mM and ~500 uM, respectively (Supplementary Fig. 3c).

An AmpG specific hydrophobic vestibule to the substrate binding cavity

Our structure also illustrates an unusual structural motif in AmpG, comprised of antiparallel C-terminal helices TM13 and TM14 that pack along one opening of the periplasmic binding cavity (Fig. 2c, d). Fourteen TMs is a departure from the canonical 12 TM MFS topology present in structural homologs of AmpG (Supplementary Fig. 9). In the AmpG family, these helices are typically located at the C-terminus as observed in E. coli here. However, some AmpG orthologues, including those from P. aeruginosa and A. baumannii, have an internal insertion between the N- and C-terminal helical bundles (Supplementary Fig. 1). Despite this, these insertions show a similar predicted structural placement and formation of related hydrophobic vestibule structures (Supplementary Fig. 10).

Here, both TM13 and TM14 are classical bitopic membrane-spanning helices of >25 residues. Rather than typical intimate hydrophobic packing with TM2 or TM11, a clear structural “foot-in-the-door” is observed at the cytosolic end of TM13 and TM14. This involves the amphipathic helix (TM6b; residues 207-219) and the loop to TM7, which provide a wedge-like protrusion with stabilizing interactions therein (Figs. 3e, 4d). Conserved residues Asp217, Phe218, Phe219 and Arg221 create a set of hydrophobic and electrostatic interactions, with the guanidinium group of Arg221 forming an electrostatic pair with Asp217 and also stacking against the aromatic rings of Phe218 and Phe415 from the loop connecting TM12 and TM13. Additional interactions, including conserved π-stacking of the side chain guanidinium of Arg418 and the aromatic side chain of Tyr425, further stabilize this complex intersection of helical segments. The ultimate consequence of this wedge-like protrusion is a dramatic hydrophobic cleft, approximately 30 × 30 × 12 Å, closed at the cytosol and running up to the periplasmic space. The hydrophobic faces of TM13 and TM14 form one wall with TM11, TM2, the C-terminus of TM6b and TM7 the other. We observe a significant amount of non-protein density in this region, which funnels down to the substrate binding pocket, which may represent loosely associated detergent or inherent lipids (Supplementary Fig. 11a). In addition, we observe a diacyl lipid bound at the lateral entrance to the vestibule with its headgroup sandwiched between the periplasmic ends of TM2 and TM13 while the two acyl tails extend down the outer surface of TM2 towards motif A and the cytoplasmic face of AmpG (Supplementary Fig. 11b, c). The density is well modeled by a PE lipid, which are highly abundant ( ~ 75 %) in E. coli membranes⁴¹.