Structural basis for antibiotic transport and inhibition in PepT2, the mammalian proton-coupled peptide transporter

The uptake and elimination of beta-lactam antibiotics in the human body are facilitated by the proton-coupled peptide transporters PepT1 (SLC15A1) and PepT2 (SLC15A2). The mechanism by which SLC15 family transporters recognize and discriminate between different drug classes and dietary peptides remains unclear, hampering efforts to improve antibiotic pharmacokinetics through targeted drug design and delivery. Here, we present cryo-EM structures of the mammalian proton-coupled peptide transporter, PepT2, in complex with the widely used beta-lactam antibiotics cefadroxil, amoxicillin and cloxacillin. Our structures, combined with pharmacophore mapping, molecular dynamics simulations and biochemical assays, establish the mechanism of antibiotic recognition and the important role of protonation in drug binding and transport.


Introduction
Antibiotics are a vital part of modern healthcare systems and one of the most signi cant advances in modern medicine 1 .Beta-lactams are among the most widely prescribed and effective antibiotics, with a broad range of activity against most pathogenic bacteria 2 .Beta-lactams function by inhibiting the synthesis of the peptidoglycan layer in the bacterial cell wall 3 .Structurally, the beta-lactam antibiotics are currently separated into four classes: penicillins, cephalosporins, carbapenems and monobactams, which differ in either the R groups attached to either the four-membered beta-lactam ring or in the case of the penicillin, carbapenem and cephalosporins, the attached ve or six-membered rings 4 .Due to the wide range of derivatives based on these core scaffolds, the uptake and retention of beta-lactam antibiotics within the human body vary partly due to their speci c interactions with different families of solute carrier (SLC) transporters [5][6][7] .A particular focus in current antibiotic drug development is improving the oral bioavailability of carbapenems, which are highly effective against Gram-negative, Gram-positive, and anaerobic bacteria.However, current formulations of carbapenems show increased breakdown in the gut and poor transport across the gut epithelia, restricting the options available to clinicians in delivering these drugs to patients 8 .Understanding the interactions between small molecule drugs and solute carriers is a promising route to improving drug pharmacokinetics and e cacy 9 .
In contrast, circulating peptides are retained in the body through reabsorption via PepT2, which selectively retains peptides in the kidney 27,28 , as well as regulates peptide transport across the bloodbrain barrier 7,29 (Fig. 1A).PepT1 and PepT2 are unusual solute carriers in being highly promiscuous 30 , able to recognize a large and diverse range of chemically distinct substrates 31 .The ability to recognize different chemical groups underpins the role played by these proteins in beta-lactam antibiotic uptake, as these drugs display both steric and chemical similarity to tripeptides 32 .
To address this important aspect of SLC15 function, we determined the structure of PepT2 from Rattus norvegicus in complex with three different beta-lactam antibiotics.Using in vitro assays, we demonstrate that one of the beta-lactam antibiotics, cloxacillin, functions as a competitive inhibitor.Combining our cryo-EM structures with molecular dynamics to probe the role of protonation in substrate recognition, we identify a crucial role for the primary amine group in orientating beta-lactam antibiotics in the biding site and proton binding in locking the drug via the carboxylate group.Our results establish a working pharmacophore model for beta-lactam recognition and explain the differences between substrate and inhibition mechanisms.

Characterization of beta-lactam transport
The proton-coupled peptide transporters are known to transport several families of beta-lactam antibiotics, including cephalosporins and penicillins.Although the substrate range recognized is extensive, encompassing many different drug classes, substrate speci city exists within the beta-lactam families, with some antibiotics exhibiting higher a nity, such as cefadroxil with a Ki 3 µM, medium a nity, such as amoxicillin with a Ki 0.2-0.43 mM and low a nity, such as Cloxacillin Ki 1 mM 10,19 (Fig. 1B).However, to date, the transport has been primarily characterized using studies reporting the inhibition of a radioactive reporter peptide in cell-based assays 10,56 .To verify the Ki values reported for these cell-based studies, we reconstituted rat PepT2 into liposomes and calculated IC 50 values using inhibition of the uptake of radioactive di-alanine peptide (Fig. 1C).While cefadroxil displayed an IC 50 of 19 µM ± 3, both amoxicillin and cloxacillin displayed signi cantly weaker values of 270 µM ± 39 and 203 µM ± 21, respectively.In comparison, the di-alanine peptide had an IC 50 of 32 µM ± 3.However, a drawback of IC 50 studies is the inability to discriminate between transported ligands and nontransported inhibitors.Therefore, we tested the ability of these antibiotics to drive transport via a counter ow experiment compared to the di-alanine peptide (Fig. 1D).As expected, cefadroxil was able to drive transport more effectively than di-alanine, consistent with the calculated IC 50 values.
Similarly, amoxicillin was less effective, showing ~ 35% uptake.However, cloxacillin was unable to drive transport, even at high concentrations (10 mM).To verify the inhibitory property of this antibiotic, we used the negative control of Lys[Z(NO2)]-Pro (LZNP), a known high-a nity inhibitor of PepT2 57 .Recently, the use of solid support membrane devices has enabled measurements of changes in membrane capacitance following charge movement via secondary active transporters 58 .We analyzed the activity of the reconstituted PepT2 using the SURFR2 platform, which also con rmed the radioactive uptake and counter ow assays (Fig. 1E).Whilst the addition of di-alanine, cefadroxil and amoxicillin all generated a signi cant capacitance change, the addition of cloxacillin did not.Thus, we conclude that cloxacillin is not transported by PepT2 but likely acts as a competitive inhibitor.We considered the question of why certain antibiotics function as ligands while others are inhibitors, as this behaviour has implications for inhibitor design more broadly within the SLC15 family, which includes members linked to in ammatory regulation 59 .We, therefore, sought to understand the nature of the binding interactions that determine good compared to weak substrates and verify the inhibitory mechanism of cloxacillin.

Cryo-EM structure of PepT2 in complex with Cefadroxil and Amoxicillin
To gain further insight into the mechanism of antibiotic recognition, we determined the structure of PepT2 in complex with both cefadroxil and amoxicillin at 3.1 Å and 3.2 Å, respectively, using cryo-EM (Fig. 2A & B, Table 1, Fig. S1 and S2).The structures were obtained using an inhibitory nanobody identi ed in our previous study reporting the apo structure 33 .In both structures, the transporter adopts an outward open, extracellular facing state, with a large solvent-accessible cavity extending from the exterior of the transporter down towards the intracellular gate, which is constructed from the packing together of TMs 4-5 and TMs10-11 (Fig. S3).The overall arrangement of TM helices in both structures is essentially identical to the previously obtained apo structure (PDB:7NQK), except the coulomb density map around the extracellular domain (ECD) is much lower resolution, resulting in our decision not to build the IgG domains in these structures.However, clear density for the nanobody was obtained, which binds to the extracellular regions of TMs 1 and 2 33 (Fig. S1 and S2).The structures have root mean square deviations (r.m.s.d) of 0.62 Å and 0.45 Å for the cefadroxil and amoxicillin structures, respectively, when aligned against the TM helices (562 C α atoms) of the apo state of the rat PepT2 protein (PDB:7NQK).
Cefadroxil can be clearly observed in the map and adopts an elongated 'L-shaped' conformation (Fig. 3A & B).The hydroxyphenyl group interacts with Asp317 on the extracellular gating helix, TM7, which also forms part of a negatively charged pocket within the binding site.The amino group sits adjacent to Glu622 on the intracellular gating helix, TM10, and interacts with Asn192 on TM4, which forms part of the intracellular gating helices in the N-terminal bundle.The carbonyl group of cefadroxil also interacts with Asn192, anchoring this region of the drug within the transporter.The remainder of the drug molecule makes fewer direct interactions with the binding site, with only the carbonyl group in the betalactam ring hydrogen bonding to Tyr61 on the extracellular gating helix, TM1.The methyl group at the C3 position on the dihydrothiazine ring sits within a hydrophobic pocket formed by Leu650, Val653 and Trp649 on gating helix TM11 and observed in bacterial homologues of the POT family 37 .The carboxylate group sits within the positively charged pocket formed by Arg57 on TM1 and Lys161 on TM4.Previous structures of peptide-bound prokaryotic POT family homologues and human PepT2 have identi ed the critical role electrostatics play in correctly orientating peptides within the binding site 34,37,44 .Speci cally, the conserved glutamate, Glu622 on TM10 and Arg57 on TM1, function to clamp the amino and carboxy termini of peptides, respectively.Amoxicillin adopts a similar binding position to that observed for cefadroxil, with the hydroxyphenyl, amino and carboxylate groups sitting in equivalent positions in the binding site (Fig. 3C & D).However, the beta-lactam ring is rotated ~ 90º relative to cefadroxil, with the carbonyl group pointing towards the extracellular entrance to the binding pocket.The two methyl groups on the 5-membered thiazolidine ring point towards Trp313, Val653 and Trp649, which constitute a hydrophobic pocket near TM11.Consistent with the lower IC 50 values, amoxicillin makes far fewer direct interactions with the transporter, which appear to result from the altered orientation of the beta-lactam ring, presumably to accommodate the geometry of the thiazolidine ring.As observed in the cefadroxil structure, the amino group sits in the negatively charged pocket and interacts with Asn192 via a hydrogen bond but now sits a little further from Glu622 (~ 3.5 Å).Tryosine 61 now interacts with the carbonyl group of the peptide bond, as opposed to the beta-lactam ring observed in cefadroxil.As noted above, repositioning the beta-lactam ring substantially changes the location of the carbonyl group of amoxicillin, which now interacts with Tyr94 on TM2.Together, these structures establish the fundamental recognition mechanism for betalactam antibiotics within PepT2 and enable the comparison with physiological peptide binding, as discussed below.

Structural basis for inhibition by Cloxacillin
Having established how PepT2 discriminates between a cephalosporin and an aminopenicillin, we next sought to uncover the mechanism of inhibition observed for cloxacillin, a semisynthetic penicillin carrying a 3-(2-chlorophenyl)-5-methylisoxazole-4-carboxamido group at position 6 of the beta-lactam ring.From the cryo-EM data, we could generate two maps derived from independent 3D classi cation schemes of a consensus particle set, which show two different positions for the drug within the binding site (Fig. 4A, Figs.S4-S5).Pose one was obtained from 106,684 particles and generated a map at 3.1 Å resolution (map 1), whereas pose two was obtained from 201,206 particles, which generated a map at 2.9 Å resolution (map 2) (Table 1).
Similar to the cefadroxil and amoxicillin complex structures, the protein backbone showed no obvious differences to those obtained for the apo PepT2 (r.m.s.d of 0.35 Å for 531 C α atoms).In pose 1, cloxacillin adopts a vertical orientation with the chlorophenyl group sitting within 3Å of the conserved E 53 xxER motif on TM1 and adjacent to the intracellular gate occlusion formed by TMs 4, 5 and 10, 11 packing against one another (Fig. 4B & Fig. S5).The methylisoxazole group sits close to Trp313, Trp649 and Tyr188, with the carbonyl group of the peptide bond sitting close to Lys161.The beta-lactam ring makes no direct contact; however, the sulphur atom in the thiazolidine ring sits close to Tyr61.At the end of the molecule, the carboxylate group makes the only observed hydrogen bond interaction with Tyr94, which presumably stabilizes the vertical orientation in the binding site.In pose 2, the chlorophenyl group adopts a similar location to pose 1, at the base of the binding site and sitting close to Glu53 in the E 53 xxER motif, but now sits closer to Ile191 and Arg57 (Fig. 4C & Fig. S5).Similarly, the methyl group in the methylisoxazole ring extends towards Trp313 and Val653.However, the two poses differ more substantially in the location of the beta-lactam and thiazolidine rings.Whereas in pose 1, the drug molecule sits in roughly the centre, in pose 2, cloxacillin makes several direct interactions with the binding site.Speci cally, the peptide carbonyl group interacts with Tyr188 (TM5) and Ser626 (TM10).Tyrosine 188 also interacts with the carbonyl group in the beta-lactam ring, while the sulphur atom interacts with Tyr61, similar to that observed in pose 1.Of note is the beta-lactam ring carbonyl, which interacts with Glu622 in this pose.However, we can model two rotamer positions for Glu622 in the maps, suggesting this interaction is not stable, supporting our prediction that a free amino group is necessary to lock transported ligands into the binding site.The carboxyl group makes further hydrogen bond interactions with Asn192, which also interacts with the nitrogen in the beta-lactam ring.The two poses suggest that cloxacillin binding is dominated by the positioning of the chlorophenyl group at the base of the transporter, with no strict speci city for the location of the beta-lactam backbone.Taken together, these structures reveal that while cloxacillin interacts with many of the same side chains observed in the cefadroxil and amoxicillin complexes, it cannot adopt a stable binding pose, which likely results in this antibiotic's observed inhibitory properties.As discussed below, the inability to adopt a stable binding is likely to result from the absence of a free amino group, which would orientate the drug to engage Glu622 and position the carboxyl to engage Arg57.

Interplay of protonation and ligand recognition
PepT2 is a proton-coupled transporter, and therefore, understanding how drugs interact with the binding site requires consideration of the protonation states of key protonatable side chains 33,36,40,60 .To gain further insights into substrate discrimination, proton coupling and inhibition in PepT2, we undertook unbiased molecular dynamics (MD) simulations of the drug molecules using the cryo-EM structures as starting poses (Fig. 5A & Fig. S6).For both cefadroxil and amoxicillin, three sets of simulations were run with the protonation state of Glu53 and Glu56 modulated (see methods).When both Glu53 and Glu56 are deprotonated, i.e. in their standard protonation states at pH 7.5, the amino group of cefadroxil remains stably bound to Glu622, and with some exibility also to Asn192 and Asn348 (Fig. S7).In contrast, the carboxylate group is unstable and sits ~ 6-10 Å away from Arg57 (Fig. 5A).These results are consistent with the interaction network observed in the cryo-EM structure (Fig. 3) and previous MD simulations of peptide ligands, which highlighted the role of the amino terminus in the initial capture of ligands 33 .However, following the protonation of Glu56 and, to a lesser extent, Glu53, the carboxylate group of cefadroxil moves to interact with Arg57 (Fig. 5B).The same pattern of interactions occurs in the amoxicillin simulations, with the protonation of Glu56 releasing Arg57 to clamp the carboxylate group.
However, when Glu56 is protonated, and to a lesser extent with Glu53 protonated, the amino group of amoxicillin moves away from Glu622 and detaches entirely from Asn192 (Fig. S7).This change to the orientation can be visualized by projecting the pooled trajectories onto a 2D plane, representing a slice through the 3D volume of the binding site (Fig. 5C).These 2D plots illustrate the extent to which protonation of Glu56 moves the centre of mass for the carboxylate group of both cefadroxil and amoxicillin closer to Arg57, which establishes the necessary interactions between the ligand and N-and C-terminal bundles of the transporter.The observation that amoxicillin cannot interact stably with the Glu622, Asn192 and Asn348 triad at the amino group and Arg57 at the carboxyl group simultaneously provides a reasonable explanation for its reduced experimental a nity compared to cefadroxil (Fig. 1C).
We next validated the signi cance of our results from the unbiased MD by running absolute binding free energy (ABFE) simulations of cefadroxil and amoxicillin for deprotonated and protonated Glu56 (see methods).The results indicate that cefadroxil and amoxicillin have similar binding free energies to the transporter in their cryo-EM poses (Table 2).Once Glu56 is protonated, however, cefadroxil experiences a substantial gain in a nity of ~ 6 kcal mol − 1, while amoxicillin a nity is reduced by ~ 1.7 kcal mol − 1 , supporting the conclusions drawn from the unbiased MD runs above.The cryo-EM structures of cloxacillin revealed a more dynamic binding mode compared to either cefadroxil or amoxicillin (Fig. 4).Indeed, our MD simulations support these observations, as cloxacillin fails to adopt a stable binding pose within the 6 x1 µs trajectories for each of the three cryo-EM models (Fig. S8).The absence of an amino group results in no signi cant interaction with Glu622 by any functional group in the drug molecule.Similarly, the carboxylate group also fails to interact with Arg57 with increased frequency following Glu56 protonation.The failure of cloxacillin to engage in the speci c binding pocket interactions formed by cefadroxil and amoxicillin is also quantitatively re ected in ABFE a nities (Table 2).When Glu56 is deprotonated, the a nity is ~ 4 kcal mol − 1 lower than either cefadroxil or amoxicillin, and unlike the two transported antibiotics, there is no stabilization of binding upon Glu56 protonation.As discussed below, the likely mechanism for inhibition appears to be simple steric occlusion and failure to trigger the necessary interactions to Glu622 and Arg57 required for transport.
Structural discrimination between substrates and inhibitors within the beta-lactam family.
The inhibitory property of cloxacillin was unexpected but presented an opportunity to decode further the structural differences between substrates and inhibitors within the beta-lactam family 61 .For a comparative analysis, we tested ve additional antibiotics from the penicillin and cephalosporin families.
Using our counter ow assay, we rst determined that PepT2 cannot transport either moxalactam, ceftibuten or benzylpenicillin (Fig. 6A).
Cefaclor, however, acts as a good substrate, achieving 76% activity relative to the physiological di-alanine peptide in the counter ow assay, and an IC 50 of 60 µM (Fig. 6B).Ampicillin is also transported, albeit at a lower level than cefaclor, achieving 13% activity in the counter ow assay and an IC 50 of 750 µM.This compares to IC 50 value for moxalactam of 500 µM (Table S1).Together with our previous analysis of cefadroxil, amoxicillin and cloxacillin, we can now separate these beta-lactam drugs into substrates and inhibitors (Fig. 6C).Of note is the hydroxyl on the hydroxyphenyl group of cefadroxil and amoxicillin, which is absent in cefaclor and ampicillin respectively.In both cases, cefadroxil and amoxicillin are better substrates for PepT2 than either cefaclor or ampicillin, considering their IC 50 values (Table S1).A plausible explanation for the increased a nity comes from the cryo-EM structure of cefadroxil, which reveals the hydroxyl is positioned close to a key gating helix, TM7 and interacts directly with Asp317 (Fig. 3A & B).Aspartate 317 forms part of an interaction network that controls the extracellular gate dynamics in response to proton binding to a conserved histidine on TM2, His87 in PepT2 36,62 .Thus, the interaction between the ring hydroxyl in cefadroxil and amoxicillin with Asp317 positively affects both the recognition and transport of these drugs via PepT2.
Conversely, when assessing the inhibitory potency of the beta-lactam drugs, our data shows that cloxacillin, with an IC 50 of 203 µM, is more effective than moxalactam (500µM), benzylpenicillin (1mM) or ceftibuten (1mM).Our MD analysis of cloxacillin shows the drug is unable to adopt a stable binding orientation with respect to Glu622 or Arg57 (Fig. S8), which is consistent with the three binding poses we observe in the cryo-EM structures (Fig. 4).A notable difference between cloxacillin and amoxicillin is the replacement of the primary amine group in the former compound for a methylisoxazole group in the latter.The absence of the primary amine removes a positive charge from cloxacillin and, therefore, the ability of the antibiotic to engage the Glu622, Asn192 and Asn348 triad on TMs 10, 4 and 9, respectively.Without this anchoring interaction, the binding of the drug is dominated by the chlorophenyl moiety, which in the cryo-EM structures binds in a hydrophobic pocket below Arg57 (Fig. 4).Interestingly, the chlorophenyl moiety of cloxacillin sits in a similar position to that modelled previously for the prodrug valacyclovir 33 , indicating that the occupation of this pocket is not the reason for the inability of cloxacillin to trigger transport.More likely, the failure to adopt a stable binding pose and engage Glu622 explains the inhibitory properties of this drug.It is also consistent with the requirement of a free amino terminus for peptide ligands 27,63 .Neither moxalactam nor benzylpenicillin have primary amine groups, so the most likely explanation for their ability to inhibit PepT2 is their inability to engage the Glu622 pocket, as observed for cloxacillin.Our results from ceftibuten, which does contain a primary amine, also highlight the importance of distance between the free amino and carboxyl groups in beta-lactam substrates.The addition of the amino-thiazol group extends the primary amine at a distance equivalent to a tetrapeptide, which is similar to moxalactam and likely makes these drugs too large to transport.

Discussion
Solute carriers play essential roles in oral bioavailability and drug pharmacokinetics 64 , and understanding how SLCs interact with drugs is essential for realizing the full potential of carriermediated drug delivery 65,66 and computer-aided drug design 9,67 .A key question concerning drug transport via PepT1 and PepT2 is how beta-lactam antibiotics are accommodated compared to physiological peptides.The recently reported structure of human PepT2 in complex with the di-peptide L-Ala-L-Phe (PDB:7PMX) in a similar extracellular open conformation enables a direct comparison to the cefadroxil and amoxicillin structures obtained in this study (Fig. 7A & B).Cefadroxil was the best substrate we tested, with an IC 50 of 20 µM compared to 32 µM for di-alanine, while amoxicillin was roughly ten times worse, at 270 µM (Fig. 1C).The difference in a nity between the two antibiotics can be rationalized from the structural comparison, with cefadroxil adopting a very similar pose to the dipeptide.Whilst the amino groups of all three molecules sit close to Glu622, it is only in cefadroxil where we observe overlay for the carbonyl and amide groups with the di-peptide.The carbonyl and methyl groups of the cepham ring also occupy the same position as the peptide carboxylate and phenylalanine side chain of the peptide, respectively.Of note is the interaction made between the cefadroxil carbonyl and Asn192, which helps to orientate the rst side chain of peptides towards the binding site entrance, facilitating optimal interactions between the amino terminus, Asn348 and Glu622 to enable transport 33 .
Although this orientation results in a small extension of the carboxylate in cefadroxil towards Arg57 and K161, the overall length of the drug is very similar to the di-peptide ligand of ~ 9 Å.The position of cefadroxil contrasts with the position adopted by amoxicillin, which, due to the stereochemistry of the Sp 3 hybridized carbon in the penem ring, is forced to extend the carbonyl group towards the extracellular gate, breaking the interaction with Asn192 and resulting in a suboptimal positioning of the drug in the binding site.The position adopted by amoxicillin is very similar to that obtained for L-Phe-L-Ala from MD simulations, which also exhibits a lower a nity for PepT2 compared to L-Ala-L-Phe 33 .This enables amoxicillin to orientate the penem methyl groups towards the aromatic pocket formed by Trp313 and Trp649 whilst maintaining the necessary interactions between the amino group and Glu622 and the carboxylate with Arg57 and K161.The hydroxyphenyl group of both antibiotics points into an acidic pocket dominated by Asp317 and sits close to the hydrophobic pocket dominated by Trp313 and Trp649 that accommodates the side chain of the C-terminal amino acid in L-Ala-L-Phe.Similar pockets have been identi ed in both pro and eukaryotic POT family transporters and play important roles in dictating peptide speci city and a nity 68 .However, as noted above for the L-Phe-L-Ala peptide, utilizing the acidic pocket results in reduced a nity due to the distortions created in the peptide backbone, illustrated by amoxicillin.However, the stereochemistry of the cepham ring enables cefadroxil to utilize this pocket to accommodate the bulky R-group while maintaining optimal interactions observed in the L-Ala-L-Phe peptide structure.This provides a plausible explanation for why cefadroxil exhibits a lower IC 50 than the L-Ala-L-Phe peptide in our counter ow and Δµ H+ -driven uptake assays.Our analysis suggests that whilst PepT2 has evolved to recognize a diverse set of peptide substrates, this promiscuity comes at a cost to transport e ciency.Cefadroxil, however, can overcome these constraints and shows enhanced transport pro les compared to peptides, suggesting these pockets are a promising route for further prodrug development.Our results thus establish the core recognition mechanism for beta-lactam antibiotics within the mammalian SLC15 family and identify the similarities with peptide substrates.
Finally, the results of the MD simulations demonstrate the close coupling between the orientation of the substrate with respect to Arg57 and the protonation of the E 53 xxER glutamate residues, particularly Glu56.Here, we observe that the cryo-EM poses of cefadroxil and amoxicillin (which show an interaction between Glu56 and Arg57, but not between Arg57 and the carboxyl group of the substrate) re-orient after protonation of Glu56.This takes the form of a salt-bridge swap: Glu56-H releases Arg57, which is then free to engage the substrate carboxyl group and clamp the ligand in the binding site ready for transport.Conversely, as discussed below, deprotonation of Glu56 would weaken the interaction with ligand by favouring the return of Arg57 to engage the E 53 xxER motif.We hypothesize that the functional role of ligand binding (Fig. 7C -step 1) is to promote the movement of protons from the extracellular gate at His87 further down into the transporter, which drives the closure of the extracellular gate (Fig. 7C -step 2) 33 .The cryo-EM poses likely represent trapped intermediates of this substrate-proton coupling mechanism.This hypothesis is further supported by our recent MD study into proton coupling within PepT2 62 , where we found that an incoming peptide must engage Arg57 to disturb the pKa value of Glu56, thereby facilitating the movement of protons towards the E 53 xxER motif, as we observe for cefadroxil and amoxicillin (Fig. 5).This triggers the closure of the extracellular gate helices, TMs 1-2 and 7-8 and the movement of the transporter towards the inwards-facing orientation, before opening the intracellular gate, formed from TMs 4-5 and 10-11, where proton binding to Glu622 weakens the interaction with the amino terminus of the ligand (Fig. 7D -step 3).Deprotonation of Glu56 would similarly weaken the interaction with the carboxyl terminus (Fig. 7D -step 4), nally freeing the ligand to exit into the cytoplasm (Fig. 7D -step 5).These ndings complement the results presented in this study, establishing a rm basis for the proposed salt-bridge swap mechanism of Arg57 engagement and delineating the role of the conserved E 53 xxER motif within the POT family.By further considering our comparison between substrate and inhibitor classes of beta-lactam antibiotics, we suggest that the engagement of Arg57 is one of the points in the transport cycle at which good and bad substrates are distinguished.In effect, our study suggests it is not the binding free energy that is predictive of drug transport but rather the ability of a drug molecule to participate in the dynamic transport mechanism originally optimized for physiological peptides.

Expression and puri cation
Rattus norvegicus SLC15A2 was expressed with a C-terminal his tagged GFP as previously described 33 .
In brief, HEK293F cells were cultured in suspension in FreeStyle™ 293 Expression Medium at 37°C and 8% CO 2 .Prior to transfection, with PEI-MAX, cells were passed at a density of 7 x 10 5 cells/mL to give a density of 1.3-1.4x 10 6 cells/mL for transfection.Sodium butyrate was added at 8 mM nal concentration before transfection.Cells were returned to the incubator and harvested 36 hours posttransfection and frozen until required.Membranes were prepared by lysing the cells via sonication and unbroken cells and cell debris were pelleted at 10,000g for 10 mins at 4°C and membranes were harvested through centrifugation at 200,000g for one hour and washed once with 20 mM HEPES pH 7.5, 20 mM KCl.After washing the membranes were resuspended in PBS and snap frozen for storage at -80 until required.
For puri cation thawed membranes were solubilized in 1 x PBS, 150 mM NaCl, 10% glycerol containing 1% DDM: CHS (5:1 ratio) for 90 minutes at 4°C.Insoluble material was removed through centrifugation for one hour at 200,000g.PepT2 was puri ed to homogeneity using standard immobilized metal-a nity chromatography protocols in n-dodecyl-β-d-maltopyranoside (DDM) detergent (Anatrace) with cholesterol hemisuccinate (5:1 ratio DDM: CHS).Following TEV cleavage, the protease and cleaved his tagged GFP were removed through nickel a nity chromatography and the protein was subjected to size exclusion chromatography (Superdex 200) in a buffer consisting of 20 mM Tris pH 7.5, 150 mM NaCl with 0.02% DDM and 0.004% CHS.

Reconstitution into liposomes.
PepT2 was reconstituted into liposomes consisting of a 3:1 POPE:POPG (Avanti polar lipids, USA) using biobeads.The lipids were dried to obtain a thin lm using a rotary evaporator and washed twice in pentane before being resuspended at 10 mg ml − 1 in lipid buffer (50 mM potassium phosphate at pH 7.5).These lipid vesicles were frozen and thawed twice in liquid nitrogen and stored at − 80°C until required.For reconstitution, the lipids were thawed and then extruded rst through a 0.8-µm lter and then through a 0.4-µm lter.DDM: CHS.Puri ed PepT2 at 0.3 µg µl − 1 was added to the lipids at a nal lipid:protein ratio of 100:1 for transport assays or 10:1 for solid support membrane (SSM) experiments and incubated for 1 h at room temperature, then for a further 1 h on ice.After this time, biobeads were added in batches.After 24 hours biobeads were removed and the proteoliposomes harvested by centrifugation at 120,000 x g for 40 minutes before resuspension in lipid buffer at a nal protein concentration of 0.25 µg µl − 1 , or 0.5 ug ul -1 for liposomes for SSM.They were subjected to three rounds of freeze-thawing in liquid nitrogen before storage at − 80°C.

IC 50 Calculations
Proteoliposomes were harvested through centrifugation before resuspending in inside buffer (120 mM potassium acetate, 2 mM MgSO 4 and 20 mM HEPES pH 7.5) and were subjected to four rounds of freeze thawing in liquid nitrogen to fully distribute the buffer and then extruded through a 0.2-µm lter.The proteoliposomes (equivalent of 1 µg protein per concentration) were diluted into external buffer (120 mM NaCl, 2 mM MgSO 4 and 20 mM HEPES pH 7.5) containing increasing concentrations of peptide or antibiotic and a trace amounts of 3 H di-alanine.The reaction was initiated through the addition of valinomycin at 1 µM and stopped after 4 minutes by rapidly ltering onto 0.22 µm lters, which were then washed with 2 × 2 ml cold water.The amount of peptide transported inside the liposomes was calculated by scintillation counting in Ultima Gold (Perkin Elmer) with comparison to a standard curve for the substrate.Experiments were performed three times to generate an overall mean and s.d.
Counter ow experiments.
Proteoliposomes were harvested through centrifugation before resuspending in counter ow buffer (50 mM potassium phosphate pH 7.5) containing 0.5 mM peptide or antibiotic (unless stated otherwise) or water as the negative control and were subjected to four rounds of freeze thawing in liquid nitrogen to fully distribute the buffer and ligand.The proteoliposomes were then extruded through a 0.2-µm lter.
Transport was initiated by diluting into counter ow buffer containing 40 µM di-alanine with trace amounts of 3H di-alanine, transport was allowed to proceed for 5 minutes before termination of the experiments through rapidly ltering onto 0.22 µm lters, which were then washed with 2 × 2 ml cold water.The amount of peptide transported inside the liposomes was calculated as above, Experiments were performed a minimal of ve times and plotted as the level of transport for each ligand compared to the level observed with di-alanine as a percentage.

SSM-based electrophysiology assays
SSM-based assays were performed on a SURFE 2 R N1 (Nanion Technologies) with sensors prepared as described 58 .Proteoliposomes were diluted (5 ul liposomes with 36 ul buffer) in non-activating buffer (20 mM Hepes, 140 mM KCl, 2 mM MgCl 2 pH 7.2) and sonicated four times for 10 s in a water bath before application (2-6 ul) to the prepared 3 mm sensors.Sensors were incubated at room temperature and centrifugation at 2800 g for 30 minutes at 10°C and incubated at room temperature for at least an hour prior to assaying.The activating buffer was made with the desired concentration of ligand (alanine, dialanine, or the antibiotics of interest) in non-activating buffer.A single solution exchange work ow was used with activating buffer applied at 1 second and removed at 2 seconds.This technique was used to distinguish between antibiotics substrate (gives a current, similar to a known peptide control -dialanine) and inhibitor (no current observed similar to a known non substrate, alanine) and raw traces from 1-3 seconds are shown in Fig. 1.

Cryo-EM sample preparation and data acquisition
PepT2 post size-exclusion was mixed with 1 mM of the antibiotic for one hour prior to the addition of 1.2 molar excess of the nanobody (D8) 33 and incubated on ice for at least 30 minutes.The nal concentration of PepT2 was between 5-6 mg/ml.The complex was adsorbed to glow-discharged holey carbon-coated grids (Quantifoil 300 mesh, Au R1.2/1.3) for 10 s.Grids were then blotted for 2 s at 100% humidity at 8°C and frozen in liquid ethane using a Vitrobot Mark IV (Thermo Fisher Scienti c).
Data were collected in counting mode in Electron Event Representation (EER) format on a CFEGequipped Titan Krios G4 (Thermo Fisher Scienti c) operating at 300 kV with a Selectris X imaging lter (Thermo Fisher Scienti c) with slit width of 10 e − V at 165,000x magni cation on either a Falcon 4 or Falcon 4i direct detection camera (Thermo Fisher Scienti c), with a physical pixel size of 0.693 Å (PepT2 + cefadroxil), 0.698 Å (PepT2 + amoxicillin) or 0.732 Å (PepT2 + cloxacillin).Movies were collected at a total dose of 54.8-57.6 e − /Å 2 fractionated to ~ 1 e − /Å 2 per frame.
Cryo-EM data processing (20 x 20) motion correction, CTF parameter estimation, particle picking, extraction, and initial 2D classi cation were performed in SIMPLE 3.0 70 .All downstream processing was carried out in cryoSPARC 3.3.1 71 or RELION 3.1 72 ,using the csparc2star.pyscript within UCSF pyem 73 to convert between formats.Global resolution was estimated from gold-standard Fourier shell correlations (FSCs) using the 0.143 criterion and local resolution estimation was calculated within cryoSPARC.
The cryo-EM processing work ow for PepT2 with cefadroxil is outlined in Fig. S1.Brie y, particles were subjected to two rounds of reference-free 2D classi cation (k = 300 each) using a 150 Å soft circular mask within cryoSPARC.Four volumes were generated from an 853,288 particle subset of the 2Dcleaned particles after multi-class ab initio reconstruction using a maximum resolution cutoff of 5 Å.These volumes were lowpass-ltered to 8 Å and used as references for a 4-class heterogeneous re nement against the full 2D-cleaned particle set (2,012,143 particles).Particles from the most populated and structured class were selected and non-uniform re ned against their corresponding volume lowpass-ltered to 15 Å, generating a 3.0 Å map.Bayesian polishing followed by per-particle defocus re nement and per-group CTF re nement ( tting beam tilt and trefoil) further improved map quality to 2.7 Å after non-uniform re nement.Alignment-free 3D classi cation using a soft spherical mask encompassing cefadroxil and surrounding partial TM helices was performed in RELION (k = 4, T = 4) resulting in one class with clear cefadroxil density.Particles belonging to this class were non-uniform re ned against a 15 Å lowpass-ltered reference, generating a 3.0 Å volume with improved cefadroxil occupancy.An additional round of alignment-free 3D classi cation (k = 3, T = 4) followed by non-uniform re nement of the class with strongest cefadroxil density generated a 3.1 Å volume that was used for model re nements.
The cryo-EM processing work ow for PepT2 with amoxicillin is outlined in Fig. S2.Brie y, particles were subjected to one round of reference-free 2D classi cation (k = 300) using a 150 Å soft circular mask within cryoSPARC.Four volumes were generated from a 618,157 particle subset of the 2D-cleaned particles after multi-class ab initio reconstruction using a maximum resolution cutoff of 7 Å.These volumes were lowpass-ltered to 8 Å and used as references a 4-class heterogeneous re nement against the full 2D-cleaned particle set (2,022,970 particles).Particles from the two most populated and structured classes were selected and subjected to an additional round of multi-class ab initio to further purify the dataset.Particles from the two most prominent classes were combined (845,401 particles) and non-uniform re ned against one of their corresponding volumes lowpass-ltered to 15 Å, generating a 3.2 Å map.Bayesian polishing followed by an additional round of 2D classi cation (k = 200) resulted in a selection of 582,083 pruned particles.These particles were non-uniform re ned followed by CTF re nement (per-particle defocus re nement and per-group CTF re nement tting beam tilt and trefoil) and another round of non-uniform re nement which generated a 2.9 Å volume.Alignment-free 3D classi cation using a soft spherical mask encompassing amoxicillin and surrounding TM helix side chains was performed in RELION (k = 4, T = 20) producing a class with strong ligand density.Particles belonging to this class were non-uniform re ned against a 15 Å lowpass-ltered reference, generating a 3.0 Å volume with improved amoxicillin occupancy.An additional round of alignment-free 3D classi cation (k = 4, T = 20) followed by non-uniform re nement of the class with most resolved amoxicillin density generated a 3.2 Å volume that was used for model re nements.
The cryo-EM processing work ow for PepT2 with cloxacillin is outlined in Fig. S4.Brie y, particles were subjected to one round of reference-free 2D classi cation (k = 300) using a 150 Å soft circular mask within cryoSPARC.Selected particles (1,795,057) were subjected to heterogeneous re nement against four 8 Å lowpass-ltered volumes generated ab initio from the amoxicillin dataset.Particles from the two most populated and structured classes were selected and subjected to a round of multi-class ab initio to further purify the dataset.Particles from the two most prominent classes were combined (756,906 particles) and non-uniform re ned against one of their corresponding volumes lowpass-ltered to 15 Å, generating a 3.0 Å map.Bayesian polishing followed by per-particle defocus re nement and another round of non-uniform re nement generated a consensus 2.8 Å volume.Two alignment-free 3D classi cation schemes were performed in RELION (k = 4) using a soft spherical mask encompassing cloxacillin and partial TM helix side chains using a regularization parameter (T) of either 20 or 10.For either T-value classi cation, particles were selected from the most populated classes, which also demonstrated the strongest ligand density.These particles were further subjected to non-uniform re nement against 15 Å lowpass-ltered references, followed by an additional round of alignment-free 3D classi cation.Particles were once again selected from the most populated and resolved classes and subjected to a nal round of non-uniform re nement, generating volumes at resolutions of 3.1 Å and 2.9 Å.

Model building and re nement
The model of rat PepT2 was docked into the globally-sharpened map for each drug complex and adjusted where necessary by manual building using Coot v. 0.9 74 and real-space re nement in PHENIX v.

Unbiased molecular dynamics (MD) simulations
The protein models from cryo-EM were patched at the missing ECD loop using MODELLER 79 , including residues 43-409 and 604-700 as continuous chain in the processed model (as done by 33 ).We scored 200 models with QMEANDisCo 80 and selected the highest-scoring protein model for embedding into a 3:1 POPE:POPG bilayer of target size 10 x 10 nm (213/71 lipid molecules) with the CHARMM-GUI membrane builder. 81rminal groups were patched with ACE/NME residues in pymol, and the E53 / E56 residue protonation states were assigned as required in the GROMACS 2021 82 pbd2gmx tool.The protein and lipids were parametrized in the AMBER ff.14sb 83 and slipids 84 force elds respectively.The ligand parameters were obtained using the OpenFF 2.0 (SAGE) force eld. 85Gromacs was then used to solvate the system with 0.15M NaCl and ~ 23,000 TIP3P water molecules.The boxsize after energy minimization was 9.9 * 9.9 * 11.3 nm.From these boxes, we ran six replicates of equilibration and production MD at temperature 310 K and pressure 1 bar.The thermostat was v-rescale 86 (separate temperture coupling groups for membrane and solvent), the barostat berendsen for equilibration and Parrinello-Rahman for production. 87The equilibration protocol was: 200 ps NVT, 1 ns NPT (C-alpha restraints and ligand heavy atoms), 20 ns NPT (C-alpha restraints only).The production simulations without restraints were then run for 1 µs.

Absolute binding free energy (ABFE) simulations
We ran ABFE simulations 88 for cefadroxil, amoxicillin and cloxacillin (pose one only) binding in the E56 unprotonated and protonated protein conditions.We derived Boresch restraints 89 from the last 200 ns of each of the 6-replicate unbiased simulations and picked the frame closest to the restraint centre as ABFE starting frames, using MDRestraintsGenerator. 90For cefadroxil and amoxicillin, we further made 4 replicate starting poses from the unbiased replicate with the highest a nity by running 4 times additional 200 ns-long equilibrations and deriving new sets of Boresch restraints from them (we report mean +/-standard deviation among these 5 replicates).For cloxacillin, we report the (single) highest binding a nity of the unbiased E56 unprotonated and protonated replicates, respectively.Our lambdaprotocol was to rst add Boresch restraints (for the complex thermodynamic leg only, the ligand side was calculated using the analytic formula; through values 0, 0.01, 0.025, 0.05, 0.075, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.8, 1.0), then annihilate coulomb interactions (even 0.1 spacings) followed by van der waals interactions (even 0.05 spacings).This gives 44 windows for the complex and 31 windows for the ligand sides of the cycle.At each lambda window, ran energy minimization, 200ps NVT equilibration (310 K, stochastic dynamics integrator), 1ns NPT equilibration (310K, stochastic dynamics integrator, 1 bar, berendsen), then 30ns (complex windows) / 100ns (ligand windows) production (Parrinello-Rahman) with replica exchange.We analyzed the simulations using alchemlyb (https://github.com/alchemistry/alchemlyb)with the mbar esitmator 91 .

Declarations
Funding Author conceived the project.GK maintained cell stocks and undertook largescale expression and tissue culture.JLP performed all protein preparation, transport, and biochemical assays.JCD and SML performed all cryo-EM sample processing, data collection and image analysis.JCD SML SN constructed the atomic models.SML and PCB performed all molecular dynamics simulations and analysis.JLP SN wrote the manuscript and prepared gures with contributions and discussions from SML PCB JCD and SML.

Figures
Figures

Figure 1 Functional
Figure 1

Figure 3 Analysis
Figure 3
: This research was supported by Wellcome awards to SN (215519;219531).SML is a Wellcome Trust PhD student (218514).Computing was supported via the Advanced Research Computing facility, Oxford, the EPSRC ARCHER2 UK National Supercomputing Service and JADE (EP/X035603/1) granted via the High-End Computing Consortium for Biomolecular Simulation, (HECBioSimhttp://www.hecbiosim.ac.uk), supported by EPSRC (EP/L000253/1).This research was funded (in part) by the Intramural Research Program of the NIH (to SML.)