Vancomycin resistance: Modeling backbone variants with d-Ala-d-Ala and d-Ala-d-Lac peptides

Abstract

To seek vancomycin analogs with broader antibacterial activity, effects of backbone modifications for the agylcon 2 on binding with d-Ala-d-Ala- and d-Ala-d-Lac-containing peptides were investigated by Monte Carlo/free energy perturbation (MC/FEP) calculations. The experimental trend in binding affinities for 2 with three tripeptides was well reproduced. Possible modifications of the peptide bond between residues 4 and 5 were then considered, specifically for conversion of the O=C—NH linkage to CH₂NH₂⁺ (6), FC=CH (7), HC=CH (8), and HN—C=O(9). The MC/FEP results did not yield binding improvements for 7, 8, and 9, though the fluorovinyl replacement is relatively benign. The previously reported analog 6 remains as the only variant that exhibits improved affinity for the d-Ala-d-Lac sequence and acceptable affinity for the d-Ala-d-Ala sequence.

Vancomycin (1) is a potent glycopeptide antibiotic that is a treatment of last resort for infections caused by methicillin resistant Staphylococcus aureus (MRSA).^1,2 The mechanism of action of 1 is to halt cell-wall biosynthesis of Gram-positive bacteria by binding to the terminal d-Ala-d-Ala sequence of the peptidoglycan cell-wall precursors.^3,4 In the common strains of vancomycin-resistant enterococci, VanA and VanB, the terminal residues are reprogrammed to the depsipeptide sequence, d-Ala-d-Lac.³ This replacement of the terminal peptide bond by an ester linkage decreases the antibiotic activity by a factor of 1000.⁴

To elucidate the Ala→Lac modification, Boger and co-workers studied the binding of 1 and vancomycin aglycon (VA, 2) to a series of tripeptide cell-wall precursor mimics, including Ac₂-l-Lys-d-Ala-d-Ala (3), Ac₂-l-Lys-d-Ala-d-Lac (4), and Ac₂-l-Lys-d-Ala-d-Ket (5) (Fig. 1).¹ For both 1 and 2, removal of the ligand's hydrogen-bond donor (3→5, amide to ketone) leads to a 10-fold decrease in binding affinity, and addition of the lone pair repulsion (5→4, ketone to ester) further decreases the binding 100-fold.¹ Motivated by the strength of the latter effect, Crowley and Boger synthesized the amine analog 6, which has the C=O of residue 4 replaced by a methylene group (Fig. 1).⁵ Relative to 2, 6 yields a 40-fold increase in binding affinity for 4 but a 35-fold decrease in binding affinity for 3; it has similar affinities for both peptides.

Figure 1.

Vancomycin (1), vancomycin aglycon (2), and the tripeptide ligands as cell wall precursor mimics (3–5).

The NH→O ligand modification converts a hydrogen-bond (H-bond) donor to a weak H-bond acceptor at a critical site for binding. Substitutions that aim to accommodate the lactate linkage are likely not to be optimal for the d-Ala-d-Ala sequence. Nevertheless, following the success of Crowley and Boger,⁵ the present computational studies were undertaken to seek additional productive backbone modifications and also to provide a quantitative understanding of the effects of the modifications at the molecular level. The computations focused on four VA analogs with the O=C—NH linkage between residues 4 and 5 replaced by CH₂NH₂⁺ (6), trans-FC=CH (7), trans-HC=CH (8), and HN—C=O (9), respectively, the amine, fluorovinyl, vinyl, and retropeptide analogs. The first three alternatives remove the H-bond accepting carbonyl group, while the retropeptide modification inverts the H-bond character to potentially allow the amide NH to H-bond with the ester oxygen of the d-Lac residue. The effects of these modifications are evaluated in terms of relative free energies of binding, ΔΔG_Binding, for peptides 3 and 4 as computed by Monte Carlo/free energy perturbation (MC/FEP) calculations.^6-10

Computational details

The binding of tripeptide ligands 3, 4, and 5 to 2 was first modeled as a check before examination of the backbone variants. The thermodynamic cycles for the MC/FEP calculations are shown in Figure 2. For the study of the tripeptide ligands, two alchemical perturbations were performed (Fig. 2a): first from d-Ala (3, X = NH) to d-Lac (4, X = O) and then from d-Lac to d-Ket (5, X = CH₂). Using equations 1 and 2, the relative free energy of binding, ΔΔG_Binding, for each pair of ligands was computed from the FEP simulations for the bound and unbound ligands. For study of the backbone variants (Fig. 2b), 2 was perturbed to the fluorovinyl (7), vinyl (8), and retropeptide (9) analogs. The relative binding affinities of ligands 3 and 4 were evaluated for each analog using equations 3 and 4. For the amine analog, the perturbation from 2 was not carried out as this involves a change in net charge (CONH→CH₂NH₂⁺), which can lead to electrostatic artifacts. Instead, the d-Ala→d-Lac perturbation of the ligand (3→4) was again performed.

Figure 2.

Thermodynamic cycles for MC/FEP calculations. VA, d-Ala, d-Lac and d-Ket represent 2, 3, 4, and 5, respectively. (a) Thermodynamic cycle for the tripeptide ligands. (b) Thermodynamic cycle for a pair of VA analogs, VA_i and VA_j.

The initial coordinates of apo 2 and the 2/3 complex were based on the crystal structures of apo 1 (PDB:1AA5) and the 1/3 complex (PDB:1FVM). The sugar residues were replaced with a hydroxyl group to produce the aglycon structures. These structures were relaxed via MC simulations covering ca. 1 billion configurations in a periodic water box containing ca. 1170 TIP4P water molecules.¹¹ The subsequent MC/FEP calculations involved 11 windows of simple overlap sampling, single-topology perturbations, and 9-Å residue-based cutoffs.¹² All computations were performed with MCPRO and the OPLS–AA force field.^13,14 For the FEP calculations, the solutes were re-hydrated using a 25-Å radius cap containing ca. 2000 TIP4P water molecules. Each MC/FEP window was composed of 20 million configurations of solvent-only equilibration, 25–75 million configurations of full system equilibration, and 50 million configurations of averaging. Attempted moves are made for one amino acid residue or water molecule at a time. All degrees of freedom were sampled except for the TIP4P water molecules, which are internally rigid.¹¹

Results for 2 with 3–5

The computed ΔΔG_Binding values for the tripeptide ligands agree well with the trend in the experimental binding affinities, as summarized in Table 1. The binding of 3 is computed to be most favorable, whereas the binding of 4 is computed to be weakest. For 5, the ligand neither forms an H-bond nor receives electrostatic repulsion from the backbone carbonyl of 2, so the binding affinity is intermediate. Representative illustrations of the bound complexes are shown in Figure 3. An interesting observation is that, in addition to the characteristic backbone-backbone H-bonds, all three complexes display another side-chain-side-chain H-bond between a phenolic OH of residue 7 in 2 and the terminal amide C=O from the acetylated Lys of the ligands. In Figure 3, this additional H-bond between the solutes appears to be coupled to an intermolecular clustering of methyl groups, which involves two methyl groups from the Leu1 side chain of 2 and two methyl groups from the side chains of the Ac-Lys and d-Ala of the ligand. The hydrophobic clustering is expected to enhance binding, providing an explanation for why the acetylation of the Lys side chain of α-Ac-l-Lys-d-Ala-d-Ala improves its binding to 1 by a factor of 3.¹⁵

Table 1.

Calculated and experimental ΔΔG_Binding values (kcal/mol) for ligands 3, 4, and 5 with vancomycin aglycon, 2

Perturbation	ΔGB	ΔGU	ΔΔGBinding (Calc)a	ΔΔGBinding (Expt)b
3 → 4 (X = NH → O)	30.5	24.2	6.3 ± 0.1	4.4
4 → 5 (X = O → CH2)	−18.7	−16.2	−2.5 ± 0.2	−2.6

^aUncertainties (±1σ) from separate averages over batches of 2.5 million configurations. See Ref. 11 for a review.

^bRef. ¹.

Figure 3.

Representative structures of the bound complexes of 2 from MC/FEP calculations. 2 is shaded in grey and is represented in sticks; the tripeptide ligands are shown in ball-and-stick representations. The intermolecular H-bonds are marked by dashed lines, and the clustering methyl groups are circled in green. Structures are rendered using PyMOL.¹⁶ (a) Complex 2/3 with 3 in purple. (b) Complex 2/4 with 4 in yellow. (c) Complex 2/5 with 5 in cyan. Hydrogens on carbon are hidden except for in the keto methylene group of 5.

Results for vancomycin analogs

The subsequent modeling of the backbone variants of 2 provided the MC/FEP results in Table 2. The computed ΔG_Binding values for the d-Ala-d-Ala (3) and d-Ala-d-Lac (4) peptides are relative to the 2/3 complex. For binding of 3, a negative ΔΔG_Binding would indicate an improvement in affinity over that for VA 2. For the binding of 4, an improvement is indicated by ΔΔG_Binding <4.4 kcal/mol.

Table 2.

Calculated and experimental relative ΔG_Binding values (kcal/mol) for peptides 3 and 4 with the backbone-modified vancomycin analogs

Analog	Relative ΔGBinding
	3 (d-Ala-d-Ala)	4 (d-Ala-d-Lac)
2 (Y = CONH)a	(0.0)	(4.4)
6 (Y = CH2NH2+)b	(2.1)	2.3 ± 0.1 (2.0)
7 (Y = CFCH)	2.6 ± 0.2	6.0 ± 0.2
8 (Y = CHCH)	5.0 ± 0.2	4.8 ± 0.2
9 (Y = NHCO)	10.9 ± 0.5	10.3 ± 0.4

^aRef. ¹. Experimental data in parentheses.

^bRef. ⁵.

None of the analogs improves the binding for the d-Ala-d-Ala peptide 3, while the amine analog 6 is the only one that improves binding for the d-Ala-d-Lac peptide 4. As observed,⁵ the computed results for 6 find its affinity for 3 and 4 to be almost the same. The binding for both 3 and 4 with 6 benefits from the charge–charge interaction between the protonated amine and the terminal carboxylate of the ligands.

The fluorovinyl (7) and vinyl (8) modifications explore alternative modulation of the non-bonded interactions, e.g., with F as a weak H-bond acceptor. From OPLS–AA optimizations, the geometries for N-methylacetamide (NMA) dimer and the NMA-2-fluorobut-2-ene complex are very similar, though the interaction energy weakens from −8.8 to −5.1 kcal/mol (Fig. 4). Thus, the C=O to F change does reduce the affinity for 3 with 7, but just to a similar level as for 3 with 6. However, no improvement is computed for the d-Ala-d-Lac sequence despite the expected reduction in electrostatic repulsion between 4 and the fluorine rather than oxygen. To investigate further, a FEP calculation was performed using the amide geometry, but perturbing to the fluorovinyl charges; this does improve the binding with 4 by 0.8 kcal/mol. Thus, the change in geometry and van der Waals (vdW) interactions are unfavorable. The longer C—F bond length (1.35 Å) than C=O (1.23 Å) is noted. The vinyl analog 8, suggested by Crowley and Boger,⁵ was hoped also to bind better with 4. However, the computed affinity for 4 with 8 shows little change from 2, while the binding of 3 by 8 is poorer than with 2 or 7. Diminished vdW and H-bond interactions for 3 with 8 are implicated.

Figure 4.

OPLS–AA interaction energies (kcal/mol) and N…O or N…F distances (Å) for gas-phase complexes with N-methylacetamide (NMA). DFT (B3LYP/6-31G(d)) optimizations for the corresponding complexes of acetamide with NMA, 2-fluorobut-2-ene, and methylacetate yield ΔE = −8.5, −5.1, and −4.3 kcal/mol, respectively. The molecular planes are roughly at right angles.

With 9, the retropeptide linkage might invert the usual binding preference by acting as a H-bond donor for the d-Ala-d-Lac peptide (4) and by promoting an unfavorable NH…HN interaction with the d-Ala-d-Ala sequence (3). However, the computed relative binding affinities are high (10 kcal/mol), predicting negligible binding of both 3 and 4 by 9. As shown in Figure 5, the orientation of the retropeptide linkage turns out to be poor for intermolecular H-bonding, probably owing to avoidance of unfavorable intramolecular interactions with the NH of residue 4 and C=O of residue 5. The average N—O distance of 4.4 Å for the amide–ester interaction from the 4/9 simulation is beyond the limit for H-bonds. Even when optimally oriented, such amide–ester H-bonds are weak. From OPLS–AA optimization of the NMA–methyl acetate complex with the alkoxy oxygen of the ester as the acceptor, the interaction energy is only −3.7 kcal/mol (Fig. 4).

Figure 5.

Representative structure from the MC simulations for the complex of 4 with the retropeptide analog 9; the poor NH…OC=O geometry caused by twisting of the retropeptide linkage is illustrated. Hydrogens on carbon are not shown for clarity.

Summary

MC/FEP calculations were shown to reproduce the experimental trend in binding affinities for vancomycin aglycon 2 with peptides 3–5. The binding of four backbone variants of 2 to the d-Ala-d-Ala (3) and d-Ala-d-Lac (4) peptides was then modeled to seek modifications that might improve binding of both sequences. The results indicate that the most promising design remains the previously reported amine analog, 6, which benefits from favorable electrostatic interactions between the ammonium group and the carboxylate terminus of the peptides. Isosteric replacement of the residue 4–5 peptide bond with a fluorovinyl group in 7 was moderately disruptive of binding with 3, but did not provide the desired improvement with 4. The vinyl alternative 8 was more damaging for the interaction with 3 in view of the complete loss of the NH…O or NH…F H-bond. Finally, 9 did not deliver a H-bond with the ester oxygen owing to geometrical mismatch (Fig. 5) and intrinsic weakness of such amide–ester H-bonds. Overall, the simulations shed light on the challenges of re-engineering the well-evolved binding site of vancomycin. Exploration of vancomycin analogs continues with emphasis on side-chain variants as a route to improved antibacterial agents.