Interaction Mode of the Novel Monobactam AIC499 Targeting Penicillin Binding Protein 3 of Gram-Negative Bacteria

Novel antimicrobial strategies are urgently required because of the rising threat of multi drug resistant bacterial strains and the infections caused by them. Among the available target structures, the so-called penicillin binding proteins are of particular interest, owing to their good accessibility in the periplasmic space, and the lack of homologous proteins in humans, reducing the risk of side effects of potential drugs. In this report, we focus on the interaction of the innovative β-lactam antibiotic AIC499 with penicillin binding protein 3 (PBP3) from Escherichia coli and Pseudomonas aeruginosa. This recently developed monobactam displays broad antimicrobial activity, against Gram-negative strains, and improved resistance to most classes of β-lactamases. By analyzing crystal structures of the respective complexes, we were able to explore the binding mode of AIC499 to its target proteins. In addition, the apo structures determined for PBP3, from P. aeruginosa and the catalytic transpeptidase domain of the E. coli orthologue, provide new insights into the dynamics of these proteins and the impact of drug binding.


Introduction
In 2014, the WHO published an alarming report about the increasing proportion of multi drug resistant (MDR) bacterial strains in human infections, which is insufficiently accounted for by the development of new antibacterials and was thus declared one of the most important public health threats of the 21st century [1]. Originally named after their ability to interact with penicillin, penicillin binding proteins (PBPs) are well established targets of a larger family of antibiotics containing a β-lactam moiety [2]. Their physiological function is associated with the synthesis, maintenance, and remodeling of peptidoglycan (PG), which is a major component of the bacterial cell wall [3]. PG, also known as murein, consists of linear β(1→4)-linked chains of alternating N-acetylglucosamine (GlcNAc) and N-acetylmuramic acid (MurNAc) units, which are cross-linked via short peptide chains attached to MurNAc [4]. Consequently, synthesis of PG requires glycosyltransferase (GTase) activity for joining GlcNAc-MurNAc disaccharides, which are available as a membraneanchored precursor (lipid II), to the growing chain, and transpeptidase (TPase) activity for establishing peptide cross-links.
PBPs have been categorized into high and low molecular mass (HMM, LMM) groups. HMM PBPs collectively confer the abovementioned activities and are thus essential for While the amino-thiazole (R 4 ) was found to be crucial for potent antimicrobial activity, variation of the R 2 and R 3 positions led to an optimum of having two methyl groups on the β-carbon and a sulfate group attached to the nitrogen of the β-lactam ring. The optimal linker (R 5 ) length was found to be ethylene, and introducing the carboxylate function modulated the overall antibacterial and ADME profile. Finally, substituting the head group (R 1 ) with a piperidine moiety increased antibacterial activity.
In this report, we aim to explore the X-ray structures of covalent intermediates formed with PBP3 from E. coli and P. aeruginosa to rationalize the binding mechanism of the five functional groups in AIC499. Additionally, we determined the apo structures of PaPBP3 (different space groups) and of a newly designed TPd construct featuring EcPBP3. Besides serving as a reference for investigating the impact of AIC499, these apo structures also provided new insights into various aspects of the structure and dynamics of PBP3 from these two model organisms.

Materials and Methods
Cloning, expression and purification. All PBP3 variants used in this study were expressed as soluble proteins, lacking their N-terminal transmembrane segments. The genetic sequence of soluble PBP3 from P. aeruginosa (PaPBP3ΔTM, res. 40-563) was cloned into pET41b(+) at restriction sites NdeI/HindIII. A thrombin cleavage site was introduced using restriction sites HindIII/NotI to facilitate removal of the C-terminal 8xHis-tag. The coding sequence of soluble PBP3 from E. coli (EcPBP3ΔTM, res. 49-588) was cloned by NdeI/BamHI into pET28a(+) vector, resulting in a thrombin-cleavable 6xHis-tag at the Nterminus. A cDNA coding for the transpeptidase domain of E. coli PBP3 (EcTPd*, S68-V88-G3-E164-Q203-G3-A228-T570) with N-terminally fused 6xHis-tag and TEV cleavage site sequences was obtained as a synthetic gene (Biomatik) and inserted into pET28a(+) at NcoI/BamHI restriction sites. Expression of all constructs was performed in E. coli In this report, we aim to explore the X-ray structures of covalent intermediates formed with PBP3 from E. coli and P. aeruginosa to rationalize the binding mechanism of the five functional groups in AIC499. Additionally, we determined the apo structures of PaPBP3 (different space groups) and of a newly designed TPd construct featuring EcPBP3. Besides serving as a reference for investigating the impact of AIC499, these apo structures also provided new insights into various aspects of the structure and dynamics of PBP3 from these two model organisms.
Crystallization. All crystallization experiments were set up at 293 K using the sitting drop approach where 0.5 µL of each precipitant solution was mixed with 0.5 µL of protein solution. Crystals of EcPBP3∆TM:AIC499 were observed in 3% (w/v) dextran sulfate (M-5000), 0.1 M sodium cacodylate, 5% (w/v) PEG 8000 and 30% (v/v) MPD at pH 6.5 with a protein concentration of 10 mg/mL in the presence of 500 µM AIC499 after a few days. Crystals of EcTPd* were observed in 0.1 M sodium cacodylate, 5% (w/v) PEG 8000, 20% (v/v) MPD, 0.2% (w/v) betaine, 0.2% (w/v) L-glutamic acid, 0.2% (w/v) L-proline, 0.2% (w/v) taurine, 0.2% (w/v) trimethlyamine N-oxide, and 0.02 M HEPES at pH 6.5 with a protein concentration of 14.4 mg/mL after a few days. The AIC499 complex was crystallized in 0.1 M MES and 25% (w/v) PEG 1000 at pH 6.5 with a protein concentration of 10 mg/mL in the presence of 500 µM AIC499 after one week. The structures derived from EcPBP3 were determined by molecular replacement using an appropriately modified version of the published EcPBP3∆TM structure (PDB entry 4BJP [13]). PaPBP3∆TM crystals were observed in 0.1 M Na 2 SO 4 and 24% (w/v) polyvinylpyrrolidone with a protein concentration of 8 mg/mL (apo crystal form 1). In the presence of 500 µM AIC499, and with a protein concentration of 14 mg/mL, PaPBP3∆TM:AIC499 crystals formed in 20% (v/v) Jeffamine ® M-2070 and 20% (v/v) DMSO and were cryoprotected by adding 1 µL glycerol to the crystallization drop. In co-crystallization experiments with a different PBP3-targeting compound, two additional PaPBP3∆TM crystal forms were observed after four to six weeks in (i) 0.1 M MES, 5% (w/v) PEG 3000, 30% (v/v) PEG 200, pH 6.0, and (ii) 0.2 M potassium/sodium tartrate, 20% (w/v) PEG 3350, using 9 mg/mL PaPBP3∆TM, and 500 µM of the compound dissolved in DMSO. Because the respective structures were found not to contain the compound, they are treated as de-facto apo structures, and these crystalline states are hence designated apo crystal forms 2 and 3 in this work. For determining the PaPBP3∆TM and PaPBP3∆TM:AIC499 structures, a model derived from PDB entry 3OC2 was used in molecular replacement [25].
Prior to measurements, all crystals were flash-cooled to 100 K and stored in liquid nitrogen. Diffraction data were collected at the European Synchrotron Radiation Facility (ESRF; Grenoble, France) on beam line ID23-1 and at the Deutsches Elektronen-Synchrotron (DESY; Hamburg, Germany) on beam lines P11 and P13 (operated by EMBL). Raw data were processed with XDS, followed by anisotropic truncation in STARANISO [26,27] using default settings. Following molecular replacement using MOLREP [28] (refer to Results section for choice of search structures), models were iteratively improved by reciprocal-space refinement in phenix.refine [29], including TLS (translation, libration, screw) parameterization, with grouping suggested on the basis of refined B-factors, and manual rebuilding using COOT [30]. Data collection and refinement statistics are provided in Table 1. Figures were prepared using PyMol and LigPlot+ [31,32]. LigPlot+ graphs in Figures 4 and 6 were created using the superposition option to ensure consistent orientation, and thresholds for plotting non-covalent interactions were modified as indicated in the respective legends. NMR experiments. Uniformly [ 2 H, 13 C, 15 N]-labeled EcPBP3∆TM and EcTPd* were expressed following the protocol from Cai et al. [33] and purified as described above. EcPBP3∆TM (111 µM) and EcTPd* (60 µM) NMR samples were prepared in Sörensen buffer (15 mM KH 2 PO 4 , 2 mM Na 2 HPO 4 , 100 mM NaCl, pH 6.0) and 1 mM 4,4-dimethyl-4silapentanesulfonic acid (DSS) with 5% (v/v) D 2 O. NMR spectra were recorded at 37 • C on a Bruker Avance III HD NMR spectrometer equipped with a TCI cryoprobe operating at a 1 H frequency of 700 MHz. The temperature was calibrated using predeuterated methanol [34]. Data were processed via NMRPipe and analyzed using CcpNMR analysis [35,36].

Structures of E. coli PBP3
To simplify handling, PBP3 from E. coli was produced as a soluble version in which the N-terminal 48 residues, including the transmembrane helix, were removed (EcPBP3∆TM).
Apo-EcPBP3∆TM was readily crystallized, but crystals yielded weak and anisotropic diffraction data extending to a maximum resolution of 4 Å, which prevented determination of its three-dimensional structure. We reasoned that the N-terminal part of the protein comprising the n-PBd may prevent formation of a highly ordered crystal lattice because of its flexibility. This view is supported by a crystal structure published previously (PDB entry 4BJP) that features very high B-factors and partial absence of electron density, particularly in the head and anchor subdomains (nomenclature according to [37]; also refer to Figure 5). For this reason, a truncated version of EcPBP3∆TM was designed that included the catalytical TPd and had certain segments replaced with tri-glycine linkers. The resulting construct (S68-V88-G 3 -E164-Q203-G 3 -A228-T570; EcTPd*) is presented in Figure 2B. In contrast to a previously published EcTPd structure (PDB entry 6HZQ [38]), the truncation was defined strictly based on tertiary fold, ensuring that regions S68-V88 and E164-Q203 of the closely apposed linker subdomain (colored grey in Figure 2) were included.
Acquisition of two-dimensional (2D) 1 H -15 N transverse relaxation-optimized spectroscopy (TROSY) heteronuclear single quantum correlation (HSQC) NMR spectra of uniformly [ 2 H, 13 C, 15 N]-labeled EcPBP3∆TM and EcTPd* revealed that most peaks in the 2D 1 H -15 N TROSY-HSQC spectrum of EcTPd* matched peaks in the corresponding spectrum of EcPBP3∆TM (Figure 2), indicating that the transpeptidase and linker domains in the two proteins adopt a similar three-dimensional fold [39]. Many peaks in the spectrum of EcPBP3∆TM that are absent in the one of EcTPd* have random coil 1 H chemical shifts (between 8.00 and 8.43 ppm [40]), indicating that regions in the head and anchor subdomains are highly dynamic. This observation is consistent with our hypothesis, based on the initial crystallographic efforts and literature described above, that the n-PBd of EcPBP3∆TM displays enhanced flexibility. As expected, the average line width of the EcTPd* peaks narrowed because of a decrease in the rotational correlation time, and variation in intensities among peaks decreased by removing the more dynamic regions of EcPBP3∆TM. The good spectral dispersion of peaks in the 2D 1 H-15 N TROSY-HSQC of EcTPd* indicates that structure and dynamic studies of this protein by NMR are feasible. After showing that the solution structure of the EcPBP3 TPd is largely unaffected by the truncation, the EcTPd* protein was crystallized. Indeed, these crystals were found to exhibit considerably improved diffraction quality (including reduced anisotropy) when compared with those obtained using EcPBP3∆TM, with useful data extending to a resolution of approx. 2.3 Å. The final model, featuring hexagonal space group P 6 2 2 2 with one molecule per asymmetric unit, contains residues T69-T569 of EcTPd* ( Table 1).
Comparison of the EcTPd* structure ( Figure 3B, dark blue) with the EcPBP3∆TM structure published previously (PDB entry 4BJP [13]; Figure 3A, gold) confirms that the 3D fold is mostly identical; 324 common C α atoms superimpose with an overall rootmean-square (RMS) distance of 0.39 Å. As expected, notable differences are observed close to the truncation sites, i.e., the V88-G 3 -E164 and Q203-G 3 -A228 regions, which are in contact with the β5-α11 and α9-α10 segments, respectively, of symmetry equivalent molecules (designation of PBP3 secondary structure elements according to [37]). These lattice interactions, in turn, cause the α9-α10 backbone to shift outwards by up to 2 Å; the β5-α11 region has not been traced in the EcPBP3∆TM structure, but electron density indicates a relative displacement on the order of 5 Å. Additional differences in the vicinity of the catalytic center concern the β2b-β2c-β2d region as well as the β3-β4 hairpin. While the former is probably affected by a lattice contact, chiefly mediated by the adjacent α5-α6 linker, and the segment preceding helix α8, contacting their respective equivalents in a symmetry mate, the latter appears to be similarly influenced by the cognate region in a neighboring copy, with distances suggesting a repulsive electrostatic interaction. Since all of these moderate alterations are far from the truncation sites, and can be readily explained by differences in lattice contacts, we conclude that the structure of EcTPd* is largely representative of the EcPBP3∆TM protein, in agreement with our observations using NMR spectroscopy.  [13], gold) is used for superposition with the AIC499-complexed EcPBP3∆TM (blue, this study). (B) The structure of apo-EcTPd* is shown in dark blue, while the complex with AIC499 is colored salmon. The GGG linkers replacing the removed segments are shown in black.

Effect of AIC499 Binding on E. coli PBP3
The EcPBP3∆TM and EcTPd* proteins were crystallized after pre-incubation with the experimental β-lactam AIC499. In the case of the truncated version, the presence of the compound left the space group (P 6 2 2 2) and packing unchanged. It did, however, impart a notable improvement in useable resolution, along with slight changes in lattice constants. For the EcPBP3∆TM:AIC499 complex, we observed space group P 6 4 2 2, which differs from the symmetry reported previously for the apo protein (PDB entry 4BJP [13], space group P 6 1 2 2); again, diffraction quality was clearly improved by the presence of the ligand but was still inferior to EcTPd*:AIC499 (Table 1). To understand the effects of structural optimization leading to the final AIC499 compound, the environment of the five functional groups addressed during the process (Figure 1) was analyzed in the crystal structures of EcTPd* and EcPBP3∆TM complexes. Since there were only negligible differences between the two models (RMS distance: 0.37 Å for 336 equivalent C α positions), we will focus on the binding of AIC499 to EcTPd* ( Figures 3B and 4) because (i) electron density was clearly more informative in this structure, and (ii) despite differences in crystallization conditions, the complex has been crystallized in the same space group as the apo form, minimizing spurious differences caused by non-conserved lattice contacts.
As expected for a β-lactam antibiotic, AIC499 is found as a covalent acyl-enzyme intermediate, with the carbonyl group of the hydrolyzed lactam forming an ester with the side chain oxygen of the catalytic S307. The compound adopts a U-shaped overall conformation, and its presence correlates with several conformational changes in the protein environment, mostly regarding the β2b-β2c-β2d, β3-β4, and β5-α11 regions, which contribute to the upper lobe, the bottom, and the lower lobe, respectively, of the binding cleft.
The largest differences between apo and complex structures can be observed in the β5-α11 region, in particular residues K539-A544. Being poorly ordered and not completely resolved in the native protein, this segment is well-structured when AIC499 is bound in the active site. Indeed, ordering of the β5-α11 loop has been observed regularly as a result of PBP-β-lactam interaction [25]. Residues Y540 and Y541 appear to be particularly relevant here, because they re-orient towards the ligand and, together with Y511 from strand β4, shield the binding pocket for the R 1 substituent of the compound as an "aromatic wall". These hydrophobic interactions centered on the phenyl ring of the head group are accompanied by a notable reduction in B-factors in the side chains involved.
Strands β3 and β4 are slightly shifted towards the core of the domain, relative to the remainder of the central β-sheet, with the protruding twisted hairpin being displaced in the opposite direction. Interactions between residues in the β3 strand and AIC499 include coordination of the terminal sulfate moiety (R 2 ) by the T497 side chain, rotation of the T495-G496 peptide plane (belonging to the third catalytic motif, KS/TG, of the canonical penicillin binding domain) because of steric interference, and formation of a hydrogen bond between the K499 carbonyl and the primary amine of the amino-thiazole group (R 4 ). A less favorable bond is potentially formed between the K499 amide nitrogen and either of the two amines in group R 4 . Interestingly, the K499 side chain does not interact with the carboxyl group of the AIC499 linker region even though the apo structure suggests this side chain is in a favorable position for such an interaction. In fact, the side chain becomes less ordered in the presence of the compound, possibly alternating between various hydrogen bonding partners, which prohibits accurate modeling. The impact of AIC499 on strand β4 is mediated mostly by hydrophobic contacts. In addition to Y511 mentioned above, side chains of Y507 and Y514 are both involved in aromatic clusters. Y507 apparently moves in concert with Y419 in the proximal α8-α9 loop, which is displaced by and aligns parallel to the AIC499 amino-thiazole moiety, forming a π-π stack. Y514, on the other hand, propagates its displacement by the β3 strand on to F303 in the β2-α2 linker; the flipped side chain of F303 engages in a hydrophobic cluster with I512 and the aliphatic portion of K500, thus linking back to the β3-β4 region. Remarkably, while the side chain of S307 is involved in the acyl-enzyme intermediate, and E304 establishes an important hydrogen bond with the amino-thiazole moiety, the main chain of the 300s region is only moderately affected by compound binding. Together with helix α8, the whole upper lobe comprising the β2b-β2c-β2d region and the adjacent helices α4 and α5 moves slightly towards the active site, allowing some side chains to interact with AIC499. In particular, V344 between strands β2c and β2d appears to play an important role in stabilizing AIC499 by hydrophobic interactions. On the one hand, it is in Van der Waals (vdW) contact to the phenyl ring of the amidine group (R 1 ), thus forming the counterpart of the aromatic wall on the opposite face of the compound. On the other hand, it also favorably interacts with one of the methyl groups at the C-4 position of the former β-lactam ring. Furthermore, the side chains of S359 and N361 (part of the second catalytic motif, SXN) preceding helix α5 are hydrogen-bonded to the acyl-enzyme ester oxygen and carboxylamide oxygen atoms, respectively, of AIC499.
The piperidine moiety of the head group R 1 does not engage in strong interactions with the protein and is modeled in two alternate conformations, with one edge of the ring in vdW contact with the Y541 side chain. Thus, this part of AIC499 does not seem to be very important for stabilizing the bound state.  Table A1.
We note certain structural differences between apo-EcTPd* and EcTPd*:AIC499 in regions distant from the active center of EcPBP3. In addition to the truncation sites, this also concerns neighboring segments with enhanced flexibility, specifically the N-terminus, the β5n-β6n loop, and the vicinity of helix α1b. These alterations can be explained by the extensive conformational change in the β5-α11 region, which is in direct contact with the V88-G 3 -E164 hairpin of a symmetry-equivalent copy, together with differences in crystallization conditions, resulting in the presence of a PEG molecule linking the V88-G 3 -E164 region to the N-terminal segment in the complex but not in the apo structure. The region preceding and including helix α1b is remarkable because it shifts more than 2 Å away from its conformation in the apo structure, which exceeds the differences in its immediate neighborhood, without itself being involved in a lattice contact. We speculate that these differences are an additive result of long-range conformational triggers propagated from the ligand binding site on the one hand and the "truncation pole" on the other.

Structures of P. aeruginosa PBP3
3.2.1. P. aeruginosa PBP3 Apo Protein As outlined above for E. coli PBP3, a soluble version of the P. aeruginosa orthologue was produced lacking the N-terminal transmembrane helix (PaPBP3∆TM). Remarkably, three distinct crystal forms were identified for PaPBP3∆TM, featuring different unit cells and lattice contacts but all belonging to space group C 1 2 1. Diffraction quality was generally superior when compared to EcPBP3∆TM crystals, with moderate anisotropy and usable resolutions of 2.2 Å (crystal form 1), 1.8 Å (crystal form 2), and 1.9 Å (crystal form 3; for details refer to Table 1).
As expected, superposition of the three models reveals a very similar overall structure ( Figure 5A). The RMS distance between corresponding C α positions was calculated between 0.50 Å and 0.65 Å, but values decrease to approx. 0.30 Å if only the TPase fold (R62-S76, R152-A187, and K217-A563) is considered. In fact, the head subdomain (T77-R152) appears to largely bend as a rigid body, leading to slightly different orientations. These can be explained by packing effects; in addition to a lattice contact shared by all three structures (α1n-α2n region and α3n helix with the α2-β2a loop and the proximal α10 helix of a symmetry mate), crystal forms 1 and 3 feature unique interactions involving, among others, the α3n helix and β3n-α4n regions, respectively, contacting the β2e-β2f and α2-β2a segments, or the β2b-β2c hairpin, of neighboring molecules. The anchor subdomain is non-contiguous, consisting of the N-terminal segment (A50-H61) and a long β-hairpin (G188-P215), and despite its spatial proximity moves independently of the head domain. This domain features particularly high flexibility, as evidenced by large B-factors and the difficulties of consistent tracing. In fact, the final models for crystal forms 2 and 3 lack several side chains in this region as well as small portions of the extreme N-terminus because of missing electron density, whereas in crystal form 1, larger parts of the backbone could not be traced. Again, the differences can be explained by packing effects.
In the TPd, the largest differences between the three apo-PaPBP3∆TM structures concern the β3-β4 and β5-α11 regions, both of which show indications of high flexibility. The β3-β4 hairpin is generally non-contiguous in electron density, preventing it from being modeled completely. In crystal forms 1 and 2, eight and six residues are not resolved, respectively, while in crystal form 3 nine residues are missing. In contrast to the structure in crystal forms 1 and 2, the protruding part of the β3-β4 region of crystal form 3 seems to bend towards the active site. This difference correlates with a lattice contact established with helix α1 that is exclusive to form 3. Based on previously solved apo-PaPBP3∆TM structures (PDB entries 3OC2, 3PBN and 6HZR), the β3-β4 hairpin was thought to be disordered and to be stabilized upon a ligand (or substrate) binding to the active site [11,25,38]. Our observations indicate that the protruding β3-β4 segment is at least partly structured in the apo protein, even if not engaged in lattice contacts.
While it was not possible to completely model the β3-β4 loop in any of the three crystal forms, we were able to trace the β5-α11 region at least in crystal form 1. In contrast, two and five residues, respectively, are missing in the other two forms. This observation might be related to a packing effect caused by the head subdomain of a neighbor molecule restricting mobility of the β5-α11 region in form 1 (indirectly via helix α1) to a bent conformation. In contrast, this segment appears to adopt a more extended structure in crystal form 2, as judged by the resolved portion. Comparison with published PaPBP3∆TM apo structures confirms the notion of enhanced conformational freedom of the β5-α11 loop. While it is bent towards the active site in crystal form 1, it is orientated in the opposite direction in the 6HZR structure and is essentially absent in 3OC2. The 3PBN structure adopts what may be considered an intermediate state, since the β5-α11 loop is not bent strongly in either direction; however, this is, again, the result of a lattice contact with a neighboring head subdomain.

Effect of AIC499 Binding on P. aeruginosa PBP3
In addition to the apo structures, PaPBP3∆TM was produced and crystallized in the presence of AIC499; crystals belonged to space group P 2 1 2 1 2 1 with two copies per asymmetric unit and yielded diffraction data extending to a resolution of 1.7 Å ( Table 1). As expected, the AIC499 complex structure displays high overall similarity to the apo version, with overall RMS distances (chain B) of 0.55 Å, 0.58 Å, and 0.57 Å w.r.t. crystal forms 1, 2, and 3, respectively. The two molecules in the asymmetric unit differ from each other and from the apo structures in the orientation of the head and anchor subdomains relative to the TPd. Similar to the variation among the apo-PaPBP3∆TM structures, this effect can be explained by slight bending of β-strands dominating these extended folds, induced by different lattice environments. Specifically, in addition to the highly favorable contact described above for all three apo structures, chain A displays additional extensive interactions of its N-terminal part (strands β1n and β9n, α4n-β4n loop) with helix α5n in the linker subdomain, the α6-β2e loop and helix α1 of symmetry-related copies, while in chain B a contact of strand β9n with the β2h-β2i segment of a symmetry mate is noteworthy. Within the asymmetric unit, the extended C-terminus of chain A is in contact with the head and linker subdomains of chain B; the reciprocal interaction is not observed.
As described above for the complex with the E. coli protein, the AIC499 molecule is covalently associated with PaPBP3∆TM via the catalytic serine (S294) side chain. In the active site environment, the β3-β4 loop as well as the β5-α11 loop appear quite flexible, despite the presence of the ligand, as evidenced by high B-factors and often discontinuous electron density. Nevertheless, we note that strands β3 and β4 clearly bend towards the active site, supporting some interactions of their side chains with the AIC499 molecule, most importantly R489 (see below). In contrast, the β5-α11 loop is ordered in chain B only and bends in a similar direction to that observed in apo form 2. More importantly, helix α11 is N-terminally extended by more than one turn in both chains, which again allows additional contacts with the ligand.
The overall conformation and interactions of AIC499 in its complex with PaPBP3∆TM ( Figure 6) resemble those described above for the EcPBP3∆TM adduct. The terminal sulfate group (R 2 ) is hydrogen bonded to the side chains of K484, S485, and T487 (the region of the third signature motif). The nitrogen of the former β-lactam ring forms a hydrogen bond with the hydroxyl moiety of S349, whereas the ligand amide group interacts with the carbonyl of T487 on the one hand and with the side chain amide of N351 on the other.  Table A1. The amino-thiazole (R 4 ) is well stabilized, engaging in three hydrogen bonds with the side chain of E291 (specifically, one of its alternate conformations) and the backbone of R489. Unlike the situation in the EcPBP3∆TM:AIC499 complex, the amino-thiazole does not form a parallel π-π stack with Y409 (equivalent to Y419 in EcPBP3) but a displaced T-stack. Additionally, Y407 (equivalent to F417 in EcPBP3) is close to the aromatic rings of Y409 and the amino-thiazole, establishing an aromatic network for further stabilization.
The carboxyl group of the linker (R 5 ) forms hydrogen bonds to the guanidinium group of R489, and the phenyl ring of the head group (R 1 ) is close to the aliphatic side chain of V333. On the opposite face of this ring, another aromatic network is formed: the R 1 phenyl ring stacks with Y532, which is further stabilized by parallel π-π stacking with Y503. Additionally, C β of F533 appears close to the aromatic ring of AIC499, but the electron density was too weak to build the remainder of the side chain. Interestingly, with an occupancy of 40%, it was possible to model an alternative set of correlated side chain conformations for R489 and Y503, located in the β3 and β4 strand, respectively. Due to a clash of the alternate Y503 rotamer with Y532, the β5-α11 loop must be displaced as well; however, electron density was not conclusive as to the respective conformer.
Similar to the complex with EcPBP3∆TM, the AIC499 head group (R 1 ) appears with two conformations in both copies present in the asymmetric unit. In general, the piperidine moiety is devoid of strong interactions with the protein; while in chain B V333 at a vdW distance from either variant is the only notable contact, one of the conformers in chain A orients towards Y532.

Discussion
The novel monobactam AIC499 was developed as a drug candidate to address a major shortcoming in today's antimicrobial toolkit, i.e., decreasing efficiency because of acquired multi-drug resistance of pathogenic bacterial strains. In this work, we have investigated the structural foundations of AIC499 activity using PBP3 from Gram-negative bacteria E. coli and P. aeruginosa as model targets. The crystal structures described herein feature the long-lived acyl-enzyme intermediates formed with the compound and thus capture the end point of mechanism-based ("suicide") inhibition. It should be noted that the actual acylation reaction is preceded by formation of a non-covalent Michaelis complex, and the structural properties of the latter may differ to some extent from those of the final covalent adduct. In the crystal structures, including AIC499, however, the covalent linkage appears essentially "neutral" with respect to the non-covalent interactions of the compound, suggesting that distortions related to acylation are probably small.
The crystal structures of PBP3, in complex with AIC499, reveal that the extent of non-covalent interactions with AIC499 differs slightly between the P. aeruginosa and E. coli proteins (Figures 4 and 6). For example, a total of eleven hydrogen bonds are formed with PaPBP3, whereas only seven are formed in the EcPBP3 complex, which may reflect that PaPBP3 was used as the target during optimization. In both cases, the amino-thiazole (R 4 ) is the functional group exhibiting the highest density of interactions with PBP3. Besides the hydrogen bonds established with residues E291 and R489 (E304 and K499 in EcPBP3) by the nitrogen-containing half of the heterocycle, the less polar moiety engages in an aromatic network with the side chains of Y409 and Y407 (Y419 and F417). In a similar fashion, the phenyl ring of the head group (R 1 ) is in intimate contact with both proteins, being sandwiched between an aromatic cluster (Y511, Y540, Y541 in EcPBP3; Y503, Y532, F533 in PaPBP3) and the aliphatic V344/V333 side chain. The piperidine ring of the head group, on the other hand, seems to have relatively low impact on the binding of AIC499; in fact, it appears to be the most dynamic part with two discernible conformations in either complex. In contrast to Y541 in EcPBP3, the homologous F533 in PaPBP3 is disordered, and the piperidine moiety instead contacts Y532 or V333. While this modification was found to improve the pharmacodynamic and pharmacokinetic properties of AIC499 compared to its predecessors (data not shown), its benefit is thus not immediately obvious from the X-ray structures alone. We hypothesize that the piperidine substituent may instead impact the kinetics or thermodynamics of the primordial Michaelis complex; clarification of this issue will require further experimentation, applying catalytically inactive PBP3 variants. Notable differences between E. coli and P. aeruginosa complexes are observed for the carboxyl moiety of the linker (R 5 ); in PaPBP3 the side chain of R489 forms a salt bridge with this group, whereas the homologous K499 in EcPBP3 is partly disordered. This is consistent with the notion that arginine is more versatile in establishing electrostatic interactions because of its geometric properties [42].
Direct comparison between AIC499 and aztreonam acyl-enzyme intermediates (Figure 7) reveals significant differences at the extremities of the molecules, whereas the core structure including the amino-thiazole moiety and the carboxyl group in the linker are conserved. Specifically, note the dramatic increase in contact area because of the R 1 head group of AIC499 (panels A and C), which mostly participates in hydrophobic interactions. Conversely, replacement of the terminal sulfonate by sulfate (panels B and D) is likely to provide additional freedom for electrostatic interactions, but an equally relevant modification in this region may be the additional methyl substituent (R 3 ); in all AIC499 complexes investigated, one of these methyl groups contributes to the hydrophobic cluster orchestrated by the phenyl ring, effectively linking both branches of the molecule. Irrespective of the differences outlined above, binding of both compounds entails a significant increase in thermal stability of PBP3. In the case of E. coli PBP3 (Figure 7E), the midpoint of thermal unfolding (T m ) determined via differential scanning fluorometry (DSF) increased from 58.6 to 63.4 • C for aztreonam and to 65.4 • C for AIC499, whereas for the P. aeruginosa protein ( Figure 7F) the value rises from 46 to 54 • C for both complexes. The temperature-dependent shifts in the 350 nm/330 nm fluorescence ratio were closely mirrored by increases in turbidity ( Figure A1), supporting the view that they do not merely reflect local changes in the environment of aromatic residues, but they correspond to real unfolding transitions.
Our crystallographic investigation revealed a considerably higher flexibility of EcPBP3 when compared with its P. aeruginosa orthologue, particularly in the extended N-terminal portions of the head and anchor subdomains. This observation is consistent with data published previously on the E. coli apo form [13]. The reasons for this difference are unclear but are likely to reflect differences in the properties of one or more interaction partners in the multiprotein complex (the divisome) that PBP3 is chiefly involved in. To circumvent the drawbacks associated with enhanced dynamics in crystallographic studies, we introduced a new truncated construct for the E. coli protein, yielding crystals with significantly improved diffraction quality. Notably, our design differs from the one published previously by Bellini et al. (PDB entry 6HZQ [38]). While those authors used a straightforward N-terminal truncation, keeping residues 234-588, we decided to include the linker subdomain in the construct because it intimately interacts with the core TPase fold and is unlikely to be affected by enhanced dynamics. Since the linker subdomain is discontinuous, this involved removal of two internal sequence segments, in addition to the far N-terminus. Indeed, the structural data thus obtained for our truncated EcPBP3 variant (EcTPd*) allowed us to trace the entire linker region, which is the vast majority of N-terminal residues included in the previous structure of soluble EcPBP3∆TM (PDB entry 4BJP; refer to Figure A2 for a superposition), yet avoiding the negative impact of excessive dynamics on crystal packing and diffraction quality.
Notably, the presence or absence of the linker subdomain has a significant influence on adjacent structural features of the TPase fold. For example, we were able to resolve the region between P279 and R297, which is not included in the previous EcTPd structure, presumably because it is stabilized by interactions with the linker region. In addition, residues R297-T300 adopt a non-native extended conformation in PBP entry 6HZQ, instead of the partly helical structure observed in our model. This local restructuring appears to propagate into other parts of the protein, especially the segment connecting helices α6 and α8; the region G392-W407 is disordered in PDB entry 6HZQ, whereas in our structure it was modeled completely, including most of the side chains. Notably, our structure of EcTPd* is consistent with previous data for soluble EcPBP3 (PBP entry 4BJP), confirming it closely reflects the native fold. In the vicinity of the active site the most notable differences are found in the β5-α11 region; it is entirely ordered and orientated towards the active site in PBP entry 6HZQ, resembling what is commonly found after ligand binding. The reasons for this discrepancy are unclear, given that this region is not restrained by lattice contacts. In contrast, the notoriously flexible parts of the β2b-β2c-β2d region and the β3-β4 hairpin appear more similar between the two EcPBP3 structures. Taken together, these observations support the view that the linker region contributes significantly to the stable fold of the TPd of EcPBP3 and, by extension, its orthologues in other species. Therefore, it appears advisable not to exclude this segment when designing truncated constructs for biophysical applications.  Figure A1. Furthermore, the utility of this new construct is not limited to crystallography. In this context, we showed that the EcTPd* protein (44 kDa) is suitable for NMR experiments, yielding very good signal dispersion in the 2D 1 H-15 N TROSY-HSQC spectrum on a spectrometer with modest field strength (700 MHz). In contrast, acquisition of good quality spectra of EcPBP3∆TM (60 kDa) is challenging because the dynamic n-PBd gives rise to a 2D 1 H-15 N TROSY-HSQC spectrum with severe resonance overlap ( Figure 2C) and an unfavorably longer correlation time. Sequence specific assignment of the 2D 1 H-15 N TROSY-HSQC spectrum for the EcTPd* protein is ongoing and should facilitate fragmentbased screening programs.
Regarding the PaPBP3 orthologue, the newly identified crystal forms described in this paper expand the repertoire of structural information on the apo protein, allowing for a more reliable assessment of protein dynamics and the impact of lattice interactions. For example, the β3-β4 region, while clearly showing enhanced dynamics, is traced to a larger extent than described previously, indicating a β-hairpin conformation that protrudes from the TPase fold. Previous structures of apo-PaPBP3∆TM, e.g., PDB entry 6HZR, were lacking coordinates for the β3-β4 loop, and it was assumed that this region is only stabilized upon acylation of the catalytic serine. For the β5-α11 region electron density suggests a high degree of conformational freedom with our crystal form 1 and PDB entry 6HZR representing two extremes of the conformational space sampled, i.e., the β5-α11 loop bends in opposite directions.
The structural data presented here may guide future efforts to further improve the properties of AIC499. One exciting option, inspired by the remarkable U-shape of proteinbound AIC499, is introduction of a covalent linkage or salt bridge between R 1 and R 3 , resulting in a cyclized compound with enhanced rigidity; such a modification may strongly affect PBP3 affinity and β-lactamase stability as well as impact pharmacokinetics. Given the unique role of PBPs in bacterial cell wall homeostasis, further development of compounds targeting these proteins is certainly warranted.  A   Table A1. Overview of the distances between atoms of AIC499 (center) and PaPBP3∆TM (chain B, left) or EcTPd* (right). The type of interaction as well as the functional group which the respective AIC499 atom belongs to are given in parentheses. See Figure 1 for the nomenclature of AIC499 atoms. Hydrogen bonds are included with donor-acceptor distances ranging from 2.3 to 3.2 Å, and hydrophobic and aromatic interactions are listed for distances between 3.0 and 4.5 Å. For distances differing between both conformers, the smaller value is given.

PaPBP3∆TM
AIC499 Inferred from the C β position, although the phenyl group is not resolved in the electron density. b An even shorter distance of 4.0 Å is found between AIC499 C12 (R 1 ) and EcTPd* Y540 C ζ .  in complex with aztreonam and AIC499, respectively. In the middle row, the analogous data are shown for PaPBP3∆TM (C,D). T m were calculated to be 45.9, 53.8 and 54.0 • C for the DMSO control, PaPBP3∆TM in complex with aztreonam and with AIC499, respectively. In the bottom row, the first derivatives of the turbidity are plotted for EcPBP3∆TM and PaPBP3∆TM, respectively (E,F).
The reference experiments with addition of DMSO only are displayed in gray, whereas the graphs for complexes with aztreonam and AIC499 are colored green and gold, respectively. Figure A2. Superposition of the X-ray structures of the EcTPd* protein introduced in this study (dark blue) and EcPBP3∆TM published previously (PDB entry 4BJP, gold), demonstrating the similar coverage of ordered structure. The GGG linkers replacing the deleted segments in EcTPd* are colored black.