S. aureus strain SH1000 (wild type) was a generous gift of S. J. Foster (University of Sheffield). S. aureus cells were grown in 1.5 L of Tryptic Soy Broth (TSB) until an OD₆₀₀ of 0.8–0.9 (Bui et al., 2012; Figueiredo et al., 2012). Cells were subsequently cooled down at 4°C and harvested by centrifugation at 10,000 × g for 20 min. Bacterial cell walls were purified from the cell pellets according to a published protocol for Streptococcus pneumoniae (Bui et al., 2012). To remove wall teichoic acids, 10 mg of cell wall was treated with 48% hydrofluoric acid at 4°C for 48 h. PG (~5 mg) was recovered by centrifugation at 264,000 × g and 4°C for 45 min. The PG pellet was washed with water and resuspended in 1 mL of buffer containing 0.02% sodium azide for storage at 4°C. Uniformly ¹³C,¹⁵N-labeled peptidoglycan SH1000 samples were obtained by growing S. aureus cells in a M9 medium containing 4 g/L of ¹³C-glucose and 1 g/L of ¹⁵NH₄Cl.

To generate muropeptides, 150 μL of the S. aureus SH1000 stock suspension of PG was incubated overnight at 37°C with 50 μL of 320 mM sodium phosphate, pH 4.8 and 10 μL of 1 mg/mL cellosyl (Höchst AG, Frankfurt, Germany). The mixture was boiled at 100°C for 10 min to inactivate the enzyme, the sample was centrifuged at 10,000 × g for 20 min, and the supernatant containing the muropeptides was collected. The muropeptide solution was stored at 2–8°C. To generate the soluble PG glycan chains with monomeric peptides the PG was digested with recombinant lysostaphin from Staphylococcus simulans (Sigma-Aldrich). In a typical preparation, 20 mL of a 5 mg/mL suspension of unlabeled or ¹³C,¹⁵N-labeled PG in 5 mM sodium phosphate at pH 7.0 was incubated under stirring at 37°C for 24 h with 2 mg of lysostaphin (final concentration 100 μg/mL). Lysostaphin was inactivated and precipitated by placing the mixture at 100°C for 10 min. The mixture was centrifuged at 10,000 × g for 30 min and the supernatant containing the PG fragments was recovered and stored at −20°C.

Soluble fragments were dialyzed against water. Before aliquoting the samples or before lyophilization, the concentration of the stock solution was estimated by liquid-state NMR using the Eretic pulse sequence (Frank et al., 2014) and a reference 1 mM sucrose sample in 90%:10% H₂O:D₂O.

Fmoc-D-Glu-NHTrt and D-LacO^tBu were synthetized as previously described (see Supporting Information in Ngadjeua et al., 2018). Orthogonal Fmoc and 1-(4,4-dimethyl-2,6-dioxocyclohexylidene)-3-methylbutyl (ivDde) protecting groups were used for sequential assembly of the branched peptide stem (see Ngadjeua et al., 2018 for similar peptides containing a D-iAsn bridge). First, the main peptide stem (D-Lac-L-Ala-D-iGln-L-Lys-D-Ala_1or2) was elongated from the Wang-D-Ala resin by successive coupling reactions with Fmoc-protected amino-acids. Second, the ε-NH₂ group of L-Lys was deprotected by hydrazinolysis, and Fmoc-Gly and 2 Fmoc-Gly-Gly were successively introduced to build the pentaglycine-bridge in three steps (Figure 2A). Final deprotection of acid-labile Fmoc and Trt protecting groups and cleavage from the resin was achieved under gentle stirring at room temperature with 2 mL of a TFA solution containing DCM, TIPS, and water (80:20:5:5 v:v:v:v). The solution was filtered to remove the resin, which was washed with 1 mL of the TFA solution. The TFA solutions were pooled and evaporated under reduced pressure before solvent extraction and purification by rpHPLC (Supplementary Information). 11 mg (14.2 μmol) and 4 mg (4.7 μmol) of pure tetrapeptide and pentapeptide, respectively, were isolated. The purity of each sample was analyzed by HPLC and their chemical structure characterized by mass spectrometry (Supplementary Figure S2) and NMR (Supplementary Figure S1 and Supplementary Table 1).

Unlabeled recombinant native PBP4 (residues 21-383) was expressed as previously published (Hamilton et al., 2017). Unlabeled PBP4(S75C) was expressed, purified and treated with thrombin to remove the His-tag as previously reported for the native enzyme, except that 5 mM TCEP was included in all buffers used during purification (Alexander et al., 2018). The expression and purification protocol was adapted for the production of ¹³C,¹⁵N-labeled PBP4 and PBP4(S75C) samples for NMR studies. To maintain good protein yields, thrombin cleavage of the GSSHHHHHHSSGLVPRGSHM N-terminal His-tag was not performed. Plasmid carrying the PBP4 or PBP4(S75C) gene was transformed into E. coli BL21(DE3). Freshly transformed bacteria were sequentially adapted over 1 day from LB to minimal M9 medium (37 mM Na₂HPO₄, 22 mM KH₂PO₄, 8.5 mM NaCl, 1 g L^{−1 15}NH₄Cl, 2 g L⁻¹ D-glucose-¹³C₆ for ¹³C,¹⁵N-labeled proteins or D-glucose for ¹⁵N-only labeled proteins, 1 mM MgSO₄, 0.1 mM CaCl₂, 0.1 mM MnCl₂, 50 μM ZnSO₄, 50 μM FeCl₃, 1 mg pyridoxine, 1 mg biotin, 1 mg hemicalcium salt of panthothenic acid, 1 mg folic acid, 1 mg choline chloride, 1 mg niacinamide, 0.1 mg riboflavin, 5 mg thiamine). M9 preculture grown overnight at 37°C with OD₆₀₀ ~2–2.5 was used to inoculate 1 L of M9 culture medium in a 1:10 ratio. The latter culture was grown at 37°C until OD₆₀₀ = 1. Expression was then induced for 3 h at 37°C with 1 mM IPTG. After harvesting the cells by centrifugation at 6,000 × g and 4°C for 20 min, the pellet was resuspended in 20 mL of lysis buffer (25 mM Tris-HCl, 500 mM NaCl, 20 mM imidazole at pH 7.5, which contained in addition 2 mM β-mercaptoethanol in the case of PBP4(S75C)). One tablet of cOmplete EDTA-free (Roche), 10 mg of lysozyme and RNase/DNase were added, and cells were disrupted by sonication. The lysate was clarified by centrifugation at 46,000 x g during 45 min at 4°C and cell debris were discarded. The supernatant was loaded onto a HisTrap™ (GE Healthcare) chromatography column. A wash was performed with 5 column-volumes of Buffer A (25 mM Tris-HCl, 1 M NaCl, 20 mM imidazole at pH 7.5). The protein was eluted with a 10–100% linear gradient of Buffer B (25 mM Tris-HCl, 500 mM NaCl, 500 mM imidazole at pH 7.5). Fractions containing the protein were pooled and concentrated and then loaded to a Superdex200 (16/60) column pre-equilibrated with the NMR buffer (100 mM potassium phosphate, 150 mM KCl at pH 7.0). Fractions containing the pure protein were pooled and concentrated for the requirements of NMR studies. For the production of the ¹³C,¹⁵N,²H-labeled PBP4 sample, two additional pre-cultures were performed before the 1L-culture for the sequential adaptation of cells to a 50%H₂O:50%D₂O M9 medium and 100% D₂O M9 medium. The 1L-culture was furthermore achieved in 100% D₂O with D-Glucose-¹³C62H₇.

Stored sacculi were resuspended, washed twice with water and twice with the interaction buffer, 100 mM potassium phosphate (pH 7.0), and isolated by centrifugation for 10 min at 16,000 × g. The PBP4 protein sample was prepared as described previously and the buffer was exchanged on a NAP™-5 desalting column (GE-Healthcare) pre-equilibrated with the interaction buffer, and the protein concentration was adjusted to 20 μM. After resuspension in the interaction buffer, 100 μL of sacculi solution was incubated with 100 μL of 20-μM PBP4 at 4°C overnight. The supernatant was discarded and kept for analysis by SDS-PAGE. Sacculi were washed 3 times with 100 μL of interaction buffer and finally resuspended in Laemmli buffer for SDS-PAGE analysis.

Assays were carried out in a final volume of 50 μL containing 10 mM Tris/HCl at pH 7.5, 10 mM MgCl₂, 0.1% Triton X-100, and 10 μM PBP4 or PBP4(S75C). A ~5 mg mL⁻¹ suspension of PG or 25 μL of a ~5 mg mL⁻¹ solution of muropeptides were added and the reaction mixture was incubated at 37°C overnight. The enzymatic reaction was stopped by boiling the samples for 10 min. Muropeptides were reduced and analyzed by HPLC as described (Boneca et al., 1997).

Purified PBP4(S75C) at 1.2 mg mL⁻¹ was incubated at ~23°C for 30 min in buffer C (20 mM MES pH 6, 300 mM NaCl, 5 mM Tris(2-carboxyethyl)phosphine (TCEP)) with or without 1.25 mM imipenem prior to being frozen at −20°C. For mass spectrometry analysis, samples were thawed and diluted 500 times in 5% acetonitrile and 0.1% formic acid. 5 μL of each sample was injected onto a 5-mm C4 column connected to a Waters Xevo GS-2 QTof mass spectrometer via a NanoAquity UPLC system. Samples were eluted in a 2-min gradient from 5 to 100% acetonitrile at a flow rate of 20 μL min⁻¹. Mass spectra (Supplementary Figure S3) were summed and peak masses deconvoluted using Waters' MassLynx software (V4.1).

PBP4(S75C) was crystallized using the sitting drop vapor diffusion method with streak seeding. Protein at 25–35 mg mL⁻¹ was added to precipitant solution (8 mM zinc chloride, 80 mM sodium acetate pH 5, 100 mM sodium fluoride, and 16% polyethylene glycol 6000) in a 1:1 (v/v) ratio and incubated at 23°C. Immediately after mixing the protein and precipitant solution, the drops were streak-seeded with a housecat whisker that had been swished through a drop containing PBP4(S75C) crystals. Prior to harvesting, the PBP4(S75C) crystals were soaked in 1 mM ampicillin and 8 mM HO-D-Lac-L-Ala-D-iGln-[L-Lys(Gly)₅]-D-Ala-D-Ala-COOH pentapeptide for 40–60 min before glycerol was added to 15% (v/v). Crystals were then promptly harvested and stored in liquid nitrogen.

Data collection was performed under cryogenic temperatures (100 K) at the Advanced Light Source synchrotron (U.C. Berkeley) on beamline 5.0.2. Data from one crystal that diffracted to 1.9 Å resolution were processed using Xia2 (Winter et al., 2013) and XDS (Kabsch, 2010) with a space group of C121 and merged with Aimless in the CCP4 software package (Winn et al., 2011). Phaser (McCoy et al., 2007) was employed to solve the structure by molecular replacement using PDB ID 5TXI as the starting model. The Phenix suite of programs was used for model generation and refinement (Adams et al., 2010). Briefly, AutoBuild was used with several iterative rounds of manual manipulation of the model with Coot (Emsley et al., 2010) followed by refinement with Phenix.refine with TLS being used in the later stages of refinement. Supplementary Figure S4 was generated using PyMOL (The PyMOL Molecular Graphics System, Version 2.1.1 Schrödinger, LLC). The final structure was deposited into the PDB under the accession code 6DZ8.

Data for ¹H- and ¹³C-NMR resonance assignment of synthetic tetra- and pentapeptide were recorded at 20°C on a 2 mg mL⁻¹ solution of the peptide in 50 mM potassium phosphate buffer at pH 6.5 with 10% D₂O on a Bruker 600-MHz spectrometer equipped with an Avance IIIHD console and a cryogenically cooled triple-resonance probe. Collected experiments on both peptides included 1D ¹H, 2D-TOCSY and 2D-NOESY with excitation sculpting for water suppression and sensitivity-enhanced ¹³C-HSQC.

Data for ¹³C,¹⁵N-labeled peptidoglycan fragments from S. aureus strain SH1000 obtained by lysostaphin digestion were recorded at 20°C on a 2 mg mL⁻¹ solution in 50 mM potassium phosphate buffer at pH 6.5 with 10% D₂O on a Bruker 850-MHz spectrometer equipped with an Avance IIIHD console and a cryogenically cooled triple-resonance probe. 2D ¹³C-HSQC and ¹⁵N-BEST-TROSY experiments served as starting points for assignments. 3D BEST-HNCACB, BEST-HNcoCACB, BEST-HNCO, HccoNH, hCcoNH, hNcacoNH experiments were collected for the assignment of resonances to nuclei of peptide stems ¹H, ¹³C and ¹⁵N (Favier and Brutscher, 2011). These experiments were complemented with ¹⁵N- and ¹³C-NOESY-HSQC. A 2D hCcH-TOCSY dataset was collected for the assignment of carbohydrate resonances.

Data for ¹H-, ¹³C- and ¹⁵N-NMR backbone resonance assignment of wild-type PBP4 were recorded at 25°C on an 800 μM ¹³C,¹⁵N,²H-labeled PBP4 sample in 100 mM potassium phosphate buffer containing 300 mM KCl and 5% D₂O at pH 7.5 on Bruker spectrometers with ¹H Larmor frequencies ranging from 700 to 950 MHz equipped with an Avance IIIHD console and a cryogenically cooled triple-resonance probe. The superimposition of the ¹⁵N-BEST-TROSY of this sample and an equivalent ¹³C,¹⁵N-only labeled sample proved the H/D back-exchange to be essentially complete in the conditions used to purify the protein. Collected experiments included a ¹⁵N-BEST-TROSY, a BEST-HNCACB/BEST-HNcoCACB pair, a BEST-HNCA/BEST-HNcoCA pair, a BEST-HNCO/BEST-HNcaCO pair, and a ¹⁵N-NOESY-HSQC. Deuteration of this 384-amino acid protein construct was instrumental for magnetization transfer through scalar couplings in 3D experiments. All NMR data were processed with Topspin 3.5 (Bruker) and analyzed with CcpNmr (Vranken et al., 2005) for resonance assignment. Assignments were 100% complete for synthetic peptides and peptidoglycan fragments. For PBP4 unambiguous resonance assignment was obtained for 65.5, 67.7, 69.0, and 67.0% of the backbone amide nuclei, carbonyl carbons, α-carbons, and β-carbons, respectively. Close to 13% of the amide resonances remained undetected at pH 7.5 and on the pH-stability range of the protein. The corresponding residues are probably in dynamic part of the protein in solution or their backbone amide is largely accessible to the solvent. Among the amide resonances, 25% were detected but remained unambiguously assigned due to coherence transfer issues in the 3D experiments and signal overlap. Assignments of resonances on wild-type PBP4 were transferred to PBP4(S75C) upon superimposition of ¹⁵N-BEST-TROSY experiments recorded in the same conditions.

Interaction studies with ¹⁵N-labeled PBP4(S75C) and different substrates were monitored by superimposition of ¹⁵N-BEST-TROSY spectra at 25°C for different substrate-to-protein ratio. Synthetic peptide stems and peptidoglycan fragments were extensively dialyzed against water using a Spectra/PorTM dialysis membrane with a 100–500 Da cutoff (Spectrum Laboratories, Inc.) and lyophilized before preparation of a few-mM stock solution in the protein buffer, with 50 mM potassium phosphate at pH 6.5. A 326-μM solution of PBP4(S75C) was titrated with 2, 8, 22, and 52 molar equivalents of tetrapeptide. A 150-μM solution of PBP4(S75C) was titrated with an estimated 2, 10, 25, 50, 100, and 180 molar equivalents of muropeptides from S. aureus strain SH1000. A 150-μM solution of PBP4(S75C) was titrated with estimated 2, 10, and 16 molar equivalents of peptidoglycan fragments obtained by lysostaphin digestion. A 150-μM solution of PBP4(S75C) was incubated with 1.2 molar equivalents of imipenem and showed significant chemical shift changes for some of the resonances. Addition of another 1.2 molar equivalents of imipenem did not yield further chemical shift changes. CcpNmr was used to monitor protein chemical shift perturbations (CSP) for every assigned amide resonance by superimposition of the ¹⁵N-BEST-TROSY spectra. CSPs (Δδ) were calculated on a per-residue basis for the highest substrate-to-protein ratio as follows:

where γ_N and γ_H are the ¹H and ¹⁵N gyromagnetic ratios, respectively, and δXPBP4(S75C):sub, and δXPBP4(S75C) are the protein resonance chemical shift value in the presence and absence, respectively, of the substrate at the highest ratio.

Models of PBP4 in complex with muropeptides were built with the version HADDOCK2.2 of “The HADDOCK web server for data-driven biomolecular docking” (de Vries et al., 2010). As starting structures, we used the X-ray crystallography structure of PBP4 (PDB ID 6C39) and a GlcNAc-MurNAc-L-Ala-D-iGln-[L-Lys(Gly)₅]-D-Ala muropeptide structure generated in-house using CNS (Schanda et al., 2014). The PBP4 and two muropeptide structures were then docked within HADDOCK. All atoms of the muropeptides were considered as passive Ambiguous Restraints (AIR). Residues of PBP4 that showed chemical shift perturbations above the threshold in Supplementary Figure S6 were considered as active AIRs. Calculations were performed with 2,000 structures during the HADDOCK rigid body energy minimization, 400 structures during the refinement, and 200 structures during the refinement in explicit water. Structure scores were calculated from the weighted energy average, HADDOCK score = E_vdW + 0.2 E_{Electrostatics} + 0.1 E_Air + E_desolvation. The output model structures were sorted with the HADDOCK built-in clustering tool using the Fraction of Common Contacts (FCC) method (Rodrigues et al., 2012) with a 0.61-Å cutoff and a minimum of 4 structures per cluster. To improve the convergence during the HADDOCK run, a 2.5-Å unambiguous restraint was introduced between the oxygen of the catalytic serine S75 of PBP4 and the carbonyl carbon of D-Ala⁴ of one muropeptide stem, as well as between the oxygen of the catalytic serine S75 of PBP4 and the amine nitrogen of the bridging Gly5 of another muropeptide stem. After on-line FCC clustering of the solutions, the two clusters with the best HADDOCK scores and lower energy were analyzed in more detail within PyMOL.

As PBP4 contributes to the high degree of cross-linking in the PG of S. aureus, we first considered the mature peptidoglycan of S. aureus as a substrate for PBP4 DD-TPase activity. We thus incubated the enzyme with PG from S. aureus strain SH1000, digested the resulting PG with cellosyl muramidase, and analyzed the obtained muropeptide profile by HPLC. PBP4 did not alter the muropeptide profile, indicating that high-molecular weight PG is not a substrate for the transpeptidase reaction (Figure 1A). However, it is possible that high-molecular weight PG is recognized and coordinated by the enzyme in an accessory role. To test this, we isolated PG from S. aureus cells and assayed for interaction with PBP4 by a pull-down experiment. PBP4 was not pulled down with PG, showing that the enzyme does not have a high affinity for PG (Figure 1B). This result also precluded being able to perform binding experiments by solid-state NMR spectroscopy and, therefore, we turned to liquid-state NMR spectroscopy approaches to probe for interactions between PBP4 and soluble peptidoglycan fragments.

In order to proceed we aimed to prepare different soluble peptidoglycan fragments to analyze enzyme/substrate complexes by X-ray crystallography or liquid-state NMR spectroscopy. As potential DD-TPase, DD-CPase and DD-EPase activities of PBP4 target the peptides in PG, we first considered chemical synthesis of defined tetrapeptide and pentapeptide stems. To this end, we synthesized two branched lactoyl-peptides, HO-D-Lac-L-Ala-D-iGln-L-Lys(Gly)₅-D-Ala-COOH tetrapeptide and HO-D-Lac-L-Ala-D-iGln-[L-Lys(Gly)₅]-D-Ala-D-Ala-COOH pentapeptide, which both resemble the native peptides in the PG of S. aureus, by divergent Fmoc solid-phase peptide synthesis using orthogonal protecting groups (see Methods) in 24 and 8% yield, respectively (Figure 2A). Milligram quantities were obtained for each of the tetrapeptide and pentapeptide products, following HPLC purification, solvent exchange against water and lyophilization. We also prepared alternative possible substrates for PBP4 by digesting PG from S. aureus by different hydrolases (Figure 2B). For this, PG was prepared from S. aureus grown either in unlabeled growth media or in growth media for ¹³C,¹⁵N-isotopic labeling. The PG was digested overnight with the cellosyl muramidase or the lysostaphin endopeptidase. Cellosyl generates a mixture of cross-linked and un-crosslinked disaccharide peptide subunits (muropeptides); lysostaphin generates glycan chains bearing un-crosslinked peptides. After heat inactivation of the hydrolases, the PG fragments were extensively dialyzed against water using a 500-kDa cut-off membrane, and lyophilized. In each case we recovered a few mg of a white powder of soluble peptidoglycan fragment mixture from approximately 10 mg of PG. Synthetic peptides and ¹³C,¹⁵N-labeled soluble fragments prepared from PG were analyzed by liquid-state NMR spectroscopy using different homonuclear or heteronuclear experiments. Figure 2C shows the resonance assignments for the ¹H,¹³C-correlation spectrum of the synthetic pentapeptide, confirming its chemical structure (see Supplementary Figure S1A for tetrapeptide spectra, Supplementary Figure S2 for HPLC and MS characterization and Supplementary Table 1 for resonance assignments on both peptides). Figure 2D shows the ¹H,¹⁵N-correlation spectrum of soluble PG fragments obtained by digestion with lysostaphin (see Supplementary Figure S1B for ¹H,¹³C-correlations). The intensities and line widths of the resonances suggested that the generated fragments behave mainly as monomers in solution. Resonance assignments through 3D heteronuclear experiments aided the characterization of fragments of different chemical structures within the mixture (Figure 2D). The amide resonances of terminal residues of a peptide chain could be clearly distinguished from the resonances derived from amino acids present within the peptide. This allowed us to identify the terminal D-Ala⁴ and D-Ala⁵ in the tetrapeptides and pentapeptides, respectively, and to quantify each species in the mixture. Similarly, terminal glycines with a free carboxylic acid and penultimate glycines with a free amine (resulting from the hydrolysis of a Gly-Gly peptide bond by lysostaphin) produce a ¹⁵N-chemical shift ranging from 114 to 117 ppm, while amide resonances from internal glycine residues show ¹⁵N chemical shifts ranging from 107 to 111 ppm (in pink in Figure 2D). Quantifying the signals from these two sets of Gly resonances revealed that lysostaphin cleaved the glycine bridge to leave mainly 2 to 3 Gly residues at the Lys, showing that lysostaphin predominantly cleaves the pentaglycine-bridge at these positions. This observation is in agreement with results obtained from lysostaphin digestion of the cell wall-anchored MalE-Cws protein, which yielded a major soluble protein fraction with the addition of three Gly from the PG pentaglycine bridge at its C-terminus as shown by mass spectrometry (Schneewind et al., 1995).

To test the activity of S. aureus PBP4, we incubated the enzyme with muropeptides from S. aureus SH1000 and analyzed the reaction products by HPLC and MS (Figure 3). PBP4 showed DD-CPase activity against the monomeric muropeptide, disaccharide pentapeptide(Gly₅) (peak5), which was nearly quantitatively converted to disaccharide tetrapeptide(Gly₅) (peak A). Interestingly, all the oligomeric peaks (for example, the dimer 11, trimer 15, tetramer 16, or pentamer 17) were completely converted by PBP4 to products with higher retention times (peaks B, C, D, and E). MS analysis of these products identified these as cyclic products, ostensibly as a result of intramolecular TPase activity of PBP4 (Figures 3B,C). As expected, the active site mutant PBP4(S75C) was inactive in this assay (Figure 3A). Remarkably, although PBP4(S75C) lacked CPase or TPase activity with muropeptide substrates, this version was nevertheless sensitive to acylation by β-lactams and, more specifically, by the carbapenem class compound imipenem as shown by MS (Supplementary Figure S3).

X-ray structures of wild-type PBP4 alone or in complex with β-lactams have been previously deposited in the PDB (PDB IDs: 6C39, 5TW8, 5TXI, 5TY7). Here we chose to study interactions with stem peptides using the catalytic serine mutant (S75C) to prevent any possible TPase activity that might interfere with the formation of a complex amenable for structural characterization by X-ray crystallography or NMR spectroscopy.

PBP4(S75C) was crystallized under similar conditions to those previously used for wild-type PBP4 (Alexander et al., 2018), yielding data to 1.86-Å resolution. Supplementary Table 2 shows the details of the data collection and refinement statistics for PBP4(S75C). PBP4(S75C) crystallized with two highly similar molecules present in the asymmetric unit (Cα alignment of chains A and B gives an r.m.s.d. of 0.35 Å over 359 aligned atoms). As expected, the structure of PBP4(S75C) (deposited under PDB ID 6DZ8) closely aligns with the structure of wild-type PBP4 (PDB ID 6C39), giving a Cα r.m.s.d. of 0.41 Å over 359 aligned atoms (Supplementary Figure S4). Despite soaking ampicillin and the synthetic pentapeptide stem with PBP4(S75C) or PBP4, we were unable to identify electron density for either of these compounds in the maps generated.

In the absence of a clear extra density that could correspond to the peptide stem in the PBP4 or PBP4(S75C) crystals, we investigated this interaction by liquid-state NMR. Assignment of the sequential backbone resonances of PBP4 was first achieved by using a series of conventional 3D heteronuclear experiments recorded on a ¹H,¹³C,¹⁵N-labeled sample of PBP4 complemented with data on a ²H,¹³C,¹⁵N-labeled sample due to the size (over 30-kDa) of the protein (Figure 4). Backbone resonance assignment was achieved at 70% and was mainly limited by sensitivity issues in the 3D experiments in relation to the relatively high molecular weight of the protein. Assignments were transferred to PBP4(S75C) by superimposing the 2D ¹H,¹⁵N-BEST-TROSY experiments (Favier and Brutscher, 2011) of the two proteins (Supplementary Figure S5). The spectra showed a very limited number of chemical shift variations, indicating that the structure is not significantly affected by the S75C substitution. This is in agreement with the highly similar structures of the wild-type and mutated protein determined by X-ray crystallography at similarly high resolution.

A 326 μM sample of ¹⁵N-labeled PBP4(S75C) was then incubated with increasing amounts (2, 8, 22, and 52 molar equivalents to the protein) of the synthetic unlabeled lactoyl-tetrapeptide HO-D-Lac-L-Ala-D-iGln-L-Lys(Gly)₅-D-Ala-COOH, and 2D ¹H,¹⁵N-correlation spectra (¹⁵N-BEST-TROSY) were collected for each titration point (Figure 5A). The addition of the peptide did not create any significant chemical shift perturbations for the amide resonances of PBP4(S75C), showing that any possible interaction of the stem peptide with the protein must have a dissociation constant that exceeds 110 mM. This result likely explains the failure to obtain co-crystals or loaded (soaked) crystals of PBP4(S75C) and PBP4 with synthetic peptides.

We asked whether the low affinity of PBP4 for small peptide substrates, too low for structural studies by either X-ray or NMR, could be due to the absence of the GlcNAc-MurNAc disaccharide motif or extended glycan strands. To address this question, we tested for interactions between PBP4(S75C) and PG fragments obtained by either cellosyl or lysostaphin digestion, using the same NMR approach as before (Figures 5B,C, left). In both cases, the ¹H,¹⁵N-correlation spectra revealed small but consistent chemical shift perturbations following the addition of increasing amounts of PG fragments, suggesting binding with a fast exchange regime and a dissociation constant in the range of hundreds of μM to a few mM.

Our next goal was to determine the interaction site(s) on PBP4(S75C) for the soluble peptidoglycan fragments. Chemical shift perturbations were calculated for the highest fragment-to-protein ratio of 180 and 16 molar equivalents for the soluble fragments obtained by cellosyl and lysostaphin digestion (Supplementary Figure S6), respectively, and reported on the PBP4 structure to localize interaction interfaces on the protein (Figures 5B,C, right). These results were compared to the one obtained for the interaction of PBP4(S75C) with the previously mentioned carbapenem, imipenem (Figure 5D). We found that the natural substrates and the antibiotic bound to the same regions of PBP4 that are shared with the published binding sites of cephems (ceftaroline) and penams (nafcillin) on wild-type PBP4. However, we observed additional residues with perturbed chemical shifts in the case of the soluble PG fragments, suggesting a more extended interface encompassing surfaces flanking both sides of the central serine nucleophile and catalytic pocket. The positioning of these two extended surfaces suggests that they indeed mimic the binding sites of the extended donor and acceptor substrates of the TPase reaction. Several residues remote from the β-lactam binding site and located in the C-terminal domain of PBP4 (I328, K331, V336, F347, V356, and V361) also experienced significant chemical shift perturbations with muropeptides (Figure 5, Supplementary Figure S7). These shifts could be caused by a secondary muropeptide binding site or a remote allostery.

We then calculated a model of the complex formed between PBP4 and the acceptor and donor muropeptides using HADDOCK 2.2/CNS by docking two muropeptide structures with tetrapeptide stems onto the PBP4 structure and using the measured NMR chemical shift perturbations as ambiguous distance restraints to drive the docking during the energy minimization process. Based on the evidence that S75 is required for the TPase catalysis (by analogy to the reactivity with carbapenem antibiotic), the distances from the S75 serine oxygen atom to both the carbonyl carbon of D-Ala⁴ of the donor peptide stem and the amine nitrogen atom at the N-terminus of the pentaglycine bridge of the acceptor peptide were constrained to 2.5 Å. These two restraints, justified by the necessity to bind both substrate peptides near the active site residue, improved the convergence of this multi-body docking process. The calculation was run with 4,000 structures during the rigid body energy minimization and 200 structures during the first and second energy refinements (the latter in explicit water). All residues of the protein that made intermolecular contacts within a 5-Å cutoff in the initial rigid body docking were allowed structural reorientations during the simulated annealing process. The muropeptides were assumed fully flexible and all of their atoms were considered as ambiguous interaction sites to facilitate the sampling of different structural conformations at all steps of the simulated annealing process. The final minimization of the complexes in water revealed the presence of two clusters of similar energies (HADDOCK score: −93.5 ± 2.7 and −93.3 ± 3.7; electrostatic energy: −375.5 ± 70.0 and −416.2 ± 5.5 kcal mol⁻¹; desolvation energy: −10.9 ± 10.1 and −0.1 ± 4.1 kcal mol⁻¹; cluster size 120 and 35 structures, respectively) (Figures 6A,B). They show an inverse position of the donor and acceptor muropeptides relative to the catalytic pocket (centered on the central nucleophile S75 as depicted in Figure 6C).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.