Expanding the paradigm of glycopeptide antibiotic recognition through molecular dynamics simulations

Abstract

Glycopeptide antibiotics (GPAs) canonically bind the terminus of peptidoglycan precursors. However, recent work suggests they can engage other targets. Keratinicyclin B (KCB), for example, is believed to bind the wall teichoic acid polysaccharide II (PSII). Although GPA–peptidoglycan interactions are well studied, the molecular determinants of KCB specificity remain unclear. Herein, we employ molecular dynamics and free-energy analysis to probe how scaffold variations, glycosylation patterns, and functional group modifications govern GPA recognition of canonical (peptidoglycan) and noncanonical (PSII) ligands. Our results demonstrate that features beyond the characteristic hydrogen bonding network, including steric preorganization, electrostatics, and binding site solvation, collectively shape the energetic landscape. Simulations of KCB–PSII variants reveal a layered recognition strategy in which PSII is stabilized by a secondary interface when the primary interface is disrupted. Together, these findings expand the paradigm of GPA recognition and provide a mechanistic framework for designing next-generation antibiotics.

Although the interactions between glycopeptide antibiotics (GPAs) and peptidoglycan have been extensively studied, the molecular determinants of specificity remain underexplored. Here, the authors use molecular dynamics and free-energy analysis to probe how scaffold variations, glycosylation patterns, and functional group modifications govern GPA recognition of canonical (peptidoglycan) and noncanonical (PSII) ligands via the study of keratinicyclin B (KCB), reporting a layered recognition strategy in which PSII is stabilized by a secondary interface when the primary interface is disrupted.

Introduction

Glycopeptide antibiotics (GPAs) are a class of clinically important drugs commonly used to treat gram-positive bacterial infections, including those caused by resistant pathogens with limited susceptibility to other treatments. Structurally, these natural products feature an extensively modified tricyclic or tetracyclic heptapeptide core decorated with glycosyl functional groups. The first GPA, vancomycin (VAN, Fig. 1A)¹, was identified in the 1950s and found to have potent activity against a broad range of gram-positive species. After approval by the Food and Drug Administration in 1958^2,3, VAN was used sparingly for the treatment of serious infections, including penicillin-resistant Staphylococcus aureus⁴, until the 1980s when it became more widely prescribed to combat the rise of methicillin-resistant S. aureus^2,5. Despite the drug’s continued clinical importance, the emergence of vancomycin-resistant strains (Enterococcus faecium in 1986^2,6,7 and S. aureus in 1997⁸) prompted efforts to identify and/or develop alternative GPAs, such as the semi-synthetic molecule oritavancin².

Fig. 1. Structural comparison of diverse GPAs.

A VAN, B KMA, C KCB and a truncated monomer of the wall teichoic acid (D) PSII are depicted with physiological protonation states. Residues are labeled upon first introduction, and the heptapeptide numbering scheme is retained across all GPAs for ease of comparison. Sugars are colored blue, and the N-terminal moiety of each GPA is colored red. Non-canonical amino acids and sugars are abbreviated as Hpg (hydroxyphenylglycine), Glc (glucose), Act (actinosamine), Rha (rhamnose), Man (mannose), Aco (acosamine), and GalNAc (N-acetylgalactosamine).

Canonically, GPAs inhibit peptidoglycan (PG) biosynthesis in gram-positive bacteria by binding to the PG precursor, lipid II, thereby disrupting cell wall assembly. The PG network consists of N-acetylglucosamine–N-acetylmuramic acid (GlcNAc-MurNAc) disaccharide backbone bearing peptide stems (L-Ala–D-iso-Glu–L-Lys–D-Ala–D-Ala), assembled through transglycosylation and cross-linked via transpeptidation^9,10. While β-lactam antibiotics target the latter step, GPAs bind to the terminal D-Ala-D-Ala sequence of the peptide, sequestering the PG precursor and sterically inhibiting its participation in both transglycosylation and transpeptidation. This mode of action has been extensively studied^11–24. Structural characterization by both high-resolution nuclear magnetic resonance (NMR)¹¹ and X-ray crystallography^12,19 depict a tripeptide mimic of the PG terminus (acyl-L-Lys-D-Ala-D-Ala) in the cleft formed by the core structure of VAN anchored by five crucial hydrogen (H) bonds. Bacterial resistance mechanisms often disrupt this H-bonding network to circumvent GPA binding. The widespread D-Ala to D-Lac mutation, for example, eliminates a key H-bond donor and introduces an unfavorable electrostatic repulsion between the fourth residue (D-4-hydroxyphenylglycine; D-Hpg) of VAN and the ester oxygen of D-Lac^15,16. To counteract this effect, chemical modification of the D-Hpg4 carbonyl to an amidine reduces O-O repulsion, thus generating a 600-fold increase in binding affinity^25,26.

While H-bonding clearly plays a crucial role in GPA-tripeptide recognition, other molecular interactions also contribute to binding specificity. For example, binding is abolished by Edman degradation of N-methylleucine (N-methyl-D-Leu), indicating that the steric contributions from the N-terminal residue are essential for stabilizing tripeptide interactions with the GPA²⁷. Molecular dynamics (MD) simulations further underscore the importance of this residue, revealing that it also facilitates binding site desolvation with assistance from L-Asn3 (Fig. S1A), which transiently occupies the site to prevent resolvation. Intriguingly, a single mutation of L-Asn3 to L-Asp precludes binding by the native tripeptide but enhances affinity for the D-Lac variant—an effect attributed to the introduction of negative charge¹⁸ and conformational rearrangement of the binding pocket, driven by changes to the intramolecular H-bonding network^20,22,23. These findings illustrate how subtle structural modifications can significantly alter GPA target specificity, emphasizing the need for more detailed molecular understanding to guide the development of next-generation antibiotics.

Conveniently, the recent discovery of a GPA subclass isolated from the soil bacterium Amycolatopsis keratiniphila provides a unique opportunity to systematically interrogate the basis for target recognition^28,29. Among these molecules, keratinimicin A (KMA) and keratinicyclin B (KCB) share a highly similar core structure, yet key differences in their peptide scaffolds and glycosylation patterns give rise to divergent mechanisms of action (Fig. 1B, C). KMA retains the canonical heptapeptide structure and D-Ala-D-Ala target, whereas KCB is a truncated hexapeptide with an oxazolidinone-functionalized N-terminus incapable of binding the PG motif. Furthermore, KMA carries a glucose-rhamnose (Glc-Rha) disaccharide on D-Hpg4, while KCB features a rare glucose-acosamine (Glc-Aco) moiety at the same position. Together these structural differences give rise to recognition of distinct molecular targets.

Rather than the broad gram-positive coverage typical of GPAs, KCB exhibits highly selective, ultra-narrow-spectrum antibiosis against Clostridioides difficile²⁹, a virulent opportunistic pathogen of the gastrointestinal tract common in healthcare settings. This selectivity is especially notable given that treatment of C. difficile with VAN often disrupts gut microbiota, increasing the risk of reinfection and contributing to medical costs exceeding $1 billion annually in the United States alone^30–32. By contrast, nanomolar combination therapy of VAN and KCB effectively suppresses C. difficile growth, leaving commensal gut bacteria unaffected. Investigations into this synergistic activity suggest that KCB binds to a species-specific wall teichoic acid (WTA), polysaccharide II (PSII; Fig. 1D), and that its truncated hexapeptide scaffold is crucial for accommodating the complex polysaccharide^29,33.

Herein, we present molecular dynamics (MD) simulations of VAN, KMA, and KCB interacting with their respective cell wall targets to further elucidate the principles governing target binding by these distinct chemotypes. Resultant trajectories were compared with simulations of non-cognate complexes and in silico-derived variants to probe the role of specific functional groups. This approach not only confirms the binding modes of each GPA but also reveals unique insights into how steric interactions, electrostatic repulsion, and conformational dynamics influence ligand binding and dissociation. More broadly, our findings highlight key molecular determinants of GPA–ligand interactions and provide a framework for understanding how structural modifications influence binding specificity relevant for downstream drug design.

Results

Dynamics of GPA–tripeptide complexes

As GPA binding to the tripeptide terminus of the cell wall precursor has been studied extensively, we began by conducting 100 ns MD simulations in triplicate (see Supplementary Methods) of VAN, KMA, and KCB with the tripeptide analog diacetyl-L-Lys-D-Ala-D-Ala to validate our approach. Root-mean-square-deviation (RMSD) analysis was used to assess conformational dynamics of the bound tripeptide backbone relative to the initial equilibrated structures. Low RMSD fluctuations indicate stable binding, while larger deviations suggest instability, or even ligand dissociation. Consistent with its well-known bioactivity, VAN forms a stable complex with the tripeptide, having an average RMSD of 1.5 ± 0.6 Å (gray traces, Fig. 2A). Comparable results were obtained for KMA (RMSD of 1.0 ± 0.4 Å), in agreement with biochemical evidence supporting a similar mechanism of action by keratinimicins (orange traces, Fig. 2B). By contrast, KCB is unable to retain the tripeptide in the canonical configuration, evidenced by much higher RMSD fluctuations (5.5 ± 2.7 Å) despite maintaining a backbone configuration that exposes the canonical H-bonding partners throughout the simulation (blue traces, Fig. 2B). Such instability is consistent with previous isothermal titration calorimetry studies, which estimated a 100-fold reduction in the K_d value for tripeptide binding to KCB relative to VAN²⁹. Together, these results highlight the importance of the N-terminal seventh residue (i.e., 3-Cl-D-Hpg in 2) in sterically locking the tripeptide within the binding pocket.

Fig. 2. Behavior of different GPAs complexed to a tripeptide mimic of the PG precursor.

RMSD plots of the tripeptide backbone atoms after least-squares alignment of each trajectory frame to the initial equilibrated structure of the relevant GPA receptor: A VAN (gray) and dVAN (red); B KMA (orange) and KCB (blue). The average RMSDs are indicated above each plot. Initial density functional theory (DFT)-optimized configurations of C VAN and D dVAN demonstrate how the distance between the tripeptide carboxylate carbon and the GPA ω3 carbonyl carbon is altered by removing the N-terminal Leu. Both GPAs are bound to acyl-L-Lys-D-Ala-D-Ala, shown in light blue.

To further explore the role of the N-terminal residue, we analyzed the dynamics of desleucyl vancomyin (dVAN), a truncated derivative of VAN that lacks the N-terminal D-Leu. It should be noted that NMR spectra indicate the ω dihedral associated with L-Asn3 (ω3) in dVAN is also inverted relative to VAN due to increased positive electrostatic density near the GPA backbone (Fig. 2C, D)^19,27,29. As dVAN is essentially incapable of binding the D-Ala-D-Ala motif^27,29, the starting model was generated by alignment of dVAN with the crystal structure of VAN in complex with the target to position the tripeptide. Upon DFT optimization of this complex, the tripeptide carboxylate carbon shifts 0.7 Å away from the ω3 peptide carbonyl carbon stabilized by L-Asn3 (Fig. 2C, D). Consistent with the absence of stable binding, RMSD fluctuations for the tripeptide complexed with dVAN are the highest among all tested GPAs (red traces, Fig. 2A).

We subsequently employed molecular mechanics Poisson–Boltzmann surface area (MM-PBSA) analysis to evaluate the binding energetics of the GPA-tripeptide complexes. Contributions from the Gibbs free energy, enthalpy, electrostatics, and van der Waals interactions are presented in Table 1. As has been reported previously^17,34, the electrostatic energy term (ΔE_ele) dominates the binding free energy (ΔG) for target binding by VAN due to charge complementarity between the GPA N-terminus and the C-terminus of the tripeptide. The aromatic m-chloro-D-Hpg6 sidechain in KMA, by contrast, allows for weak dispersive interactions with the D-Ala residues of the tripeptide, enhancing the van der Waals contribution (ΔE_vdw). Furthermore, experimental evidence indicates a higher entropic gain when KMA binds the tripeptide²⁹, attributed to solvent restructuring around m-chloro-D-Hpg6. Our simulations suggest that L-Phe3 (note that heptapeptide numbering is retained across all GPAs for ease of comparison) also contributes to this effect by moving closer to the tripeptide, accommodated by rearrangements of the GPA backbone, thereby shielding the binding pocket from solvent (Fig. S1B). Notably, contacts between L-Phe3 of KMA and the tripeptide C-terminus resemble DFT-predicted π-stacking interactions with the aromatic ring of 3,5-dihydroxyphenylglycine (Dpg), proposed to enhance binding affinity of teicoplainin compared to VAN¹⁶.

Table 1.

Average RMSD and MM-PBSA energetic values calculated from tripeptide-bound simulations

GPA system	RMSD, Å	ΔEvdw, kcal mol−1	ΔEele, kcal mol−1	ΔG, kcal mol−1	ΔH, kcal mol−1
VAN-tripeptide	1.5 ± 0.6	−24.5 ± 2.9	−75.3 ± 10.0	−11.8 ± 4.0	−24.1 ± 4.0
dVAN-tripeptide	40.7 ± 23.6	–6.1 ± 6.4	−25.1 ± 28.3	−65.5 ± 6.7	−4.5 ± 5.7
KMA-tripeptide	1.0 ± 0.4	−29.6 ± 3.0	−20.4 ± 5.9	−12.4 ± 2.8	−24.4 ± 2.8
KCB-tripeptide	5.5 ± 2.7	−22.0 ± 5.1	−16.5 ± 13.3	−27.5 ± 5.4	−16.9 ± 5.4

While the binding enthalpies (ΔH) calculated for VAN and KMA are indicative of favorable interactions that stabilize the tripeptide (−24.1 ± 4.0 and −24.4 ± 2.8 kcal mol⁻¹, respectively), complexation with dVAN is significantly less exothermic (−4.5 ± 5.7 kcal mol⁻¹), consistent with the absence of experimentally measurable affinity for the tripeptide and complete dissociation during simulations (Table 1). The calculated ΔG for dVAN, however, appears more favorable (−65.5 ± 6.7 kcal mol⁻¹) than those determined for VAN and KMA (−11.8 ± 4.0 and −12.4 ± 2.8 kcal mol⁻¹, respectively). A similar, though less pronounced trend is observed for tripeptide binding to KCB, in agreement with measured K_d values (Table 1). Such unexpectedly favorable free energy values can be rationalized in light of previous studies on vancomycin, which demonstrated that binding site desolvation incurs a significant entropic penalty^23,29,35. This entropic contribution likely offsets the weaker ΔH values of dVAN and KCB, resulting in deceptively favorable ΔG values that reflect the energetic favorability of dissociation rather than stable complex formation. To test this hypothesis, we performed geometry optimizations and frequency calculations on the first and last frames of a 100 ns simulation of KCB complexed with the tripeptide using the semiempirical PM3 method in Gaussian16 (Fig. S2). The resultant estimate of the binding entropy (ΔS) for the final complex, relative to its separate components, was 0.010 kcal mol⁻¹ K⁻¹ higher than the initial complex (Table S1). This entropic increase upon partial dissociation was accompanied by a 16.4 kcal mol^_1 increase in ΔG relative to VAN (Table 1)^27,29,35. While not reflective of full ensemble averaging, these calculations provide qualitative insight into the entropic origin of the unexpectedly favorable free energy values, in agreement with experimentally observed dissociation.

Dynamics of the KCB-PSII complex

After validating our computational parameters via simulations with the tripeptide, we looked to explore the dynamic interactions between KCB and PSII (hereafter termed the KCB-1 complex). Our previous molecular docking simulations suggested that βGalNAc4 of PSII adopts a binding mode reminiscent of the tripeptide PG precursor in complex with VAN, forming analogous H-bonding interactions. Electrostatic interactions between the phosphodiester of PSII and sugar substituents of KCB likewise appear relevant for binding. Although we previously reported limited MD trajectories for KCB-1, additional refinement of our approach was required to optimize the starting model for longer simulations (see “Methods”). The resultant GPA configuration is largely consistent with the previous model, retaining three of the five canonical H-bonding partners associated with target binding, though the PSII sugars involved in these interactions differ (Fig. 3A). The most notable change is an inward shift of the GPA Man and oxazolidinone groups, which now partially occlude the original binding cleft (Fig. S3A). As a result, PSII is displaced upward, generating additional contacts with the central GPA sugars, Aco and Glc (Fig. S3B, C), while new hydrogen bonds between the oxazolidinone carbonyl and βGlc5 hydroxyl groups anchor the terminal PSII sugars near the binding site. Altogether, these changes expanded the overall interface between the two molecules but preserve key binding interactions. Using our updated model, three 100 ns simulations showed consistent trajectories of PSII, with an average RMSD of 3.4 ± 0.6 Å over all polysaccharide atoms (Fig. 3B, C). Reflecting this stability, MM-PBSA analysis of the complex determined a ΔH of –24.0 ± 6.3 kcal mol⁻¹ and ΔG of −12.1 ± 5.4 kcal mol⁻¹ (Table 2).

Fig. 3. Determinants of PSII recognition by KCB revealed through molecular dynamics.

A H-bonding interactions between KCB and PSII involving the same GPA backbone atoms that VAN employs to engage D-Ala-D-Ala. B Overlaid MD snapshots (every 20 ns) of KCB-1 with residues/sugars colored according to their average RMSD relative to the 0 ns model. C–I RMSD plots of KCB-1 and in silico variant simulations with modifications indicated by red text. The average RMSD and its standard deviation are annotated in the top right corner of each panel. RMSD values were calculated in triplicate for the PSII monomer after least-squares alignment of each trajectory frame to the initial equilibrated structure of KCB, colored in blue, green, and orange. Panels are grouped according to the location of their chemical modifications (drawings shown in the insets).

Table 2.

Average RMSD and MM-PBSA energetic values calculated from PSII-bound simulations

GPA system	RMSD, Å	ΔEvdw, kcal mol−1	ΔEele, kcal mol−1	ΔG, kcal mol−1	ΔH, kcal mol−1
KCB-1	3.4 ± 0.6	−43.0 ± 0.6	−88.9 ± 5.2	−12.1 ± 5.4	−24.0 ± 6.3
KCB-2	10.3 ± 3.1	−22.8 ± 3.0	−46.0 ± 5.2	−55.2 ± 6.2	−22.2 ± 6.3
KCB-3	3.0 ± 0.7	−43.7 ± 0.8	−91.6 ± 5.1	−11.5 ± 4.9	−23.4 ± 6.0
KCB-4	5.0 ± 0.8	−36.0 ± 4.6	−87.3 ± 15.8	−16.2 ± 4.6	−24.4 ± 4.6
KCB-5	6.9 ± 1.3	−38.1 ± 0.6	−110.5 ± 3.4	−21.5 ± 6.3	−25.4 ± 4.0
KCB-6	9.2 ± 3.8	−36.6 ± 1.6	−68.3 ± 2.3	−34.4 ± 6.9	−26.8 ± 5.0
KCB-7	10.6 ± 3.0	−34.8 ± 1.9	−37.9 ± 2.3	−30.1 ± 5.9	−26.3 ± 3.3

More in-depth structural analyses over the course of the simulations reveals marked rigidity in the core hexapeptide, while greater flexibility was observed in peripheral regions of the GPA (Fig. 3B). Notably, motion of the GPA Man is inversely correlated with that of βGlc5 in PSII such that displacement of one toward the binding site coincides with withdrawal of the other. Electrostatic stabilization arising from interaction between the PSII phosphodiester and the GPA disaccharide—mediated by H-bonding to Glc and the positive charge of Aco (noted previously²⁹)—appears critical for complex formation. Simulations further reveal that Glc-Aco and Act form an extensive H-bonding network with PSII dominated by contacts with αMan1, βGalNAc2, and βGlc3a. Collectively, these interactions maintain a tight association between the molecules, with electrostatic contributions dominating the overall free energy (Table 2).

Determinants of PSII binding assessed through in silico variants

To dissect the polar interactions most crucial for binding PSII, we generated a series of mutations in silico and compared the resulting dynamics across three independent simulations to those of the native KCB-1 complex. Variants were designed based on analysis of the simulations above, which led us to target a series of functional groups seemingly important for antibiotic-ligand interaction. In addition to modeling a complex with dephosphorylated PSII (henceforth referred to as KCB-2), which is unable to form in vitro²⁹, we replaced the following polar atoms with carbon: the O1 atom of oxazolidinone (KCB-3); the oxygen atom of the N-acetyl group from βGalNAc4 (KCB-4); the backbone nitrogen (N2) of L-Phe3 (KCB-5); as well as the N1 (KCB-6) and O2 (KCB-7) atoms of oxazolidinone. The stability of each complex was then assessed by analyzing the RMSD of PSII atoms relative to their starting positions, as was done for KCB-1 (Fig. 3C–I). Modification of oxazolidinone O1 (KCB-3) had minimal impact on complex stability, with an RMSD of 3.0 ± 0.7 Å, in close agreement with KCB-1. Modification of the N-acetyl group of βGalNAc4 (KCB-4) or the L-Phe3 backbone (KCB-5), while slightly more destabilizing, also did not result in dramatic perturbations to the native complex (RMSDs of 5.0 ± 0.8 Å, 6.9 ± 1.3 Å, respectively). By contrast, substitution of the N1 (KCB-6) and O2 (KCB-7) atoms of oxazolidinone triggered substantial rearrangements (RMSDs of 9.2 ± 3.8 Å and 10.6 ± 3.0 Å, respectively) more consistent with the abolition of binding upon dephosphorylation of PSII (KCB-2, RMSD of 10.3 ± 3.1 Å).

Complex KCB-2 — dephosphorylated PSII

In the absence of phosphate-mediated interactions, PSII leaves the canonical binding site during the course of the 100 ns simulations, in agreement with experimental observations²⁹. However, it remains loosely associated with the GPA through a series of transient (lasting a few ns) interactions. These weak contacts are dominated by H-bonds between αMan1, αGlc3a, and βGalNAc4 of PSII and the Glc-Aco and Act substituents of the GPA (Fig. 4A). In contrast to simulations with the native WTA, βGlc5 is especially unstable, forming no significant interactions with KCB. Although MM-PBSA analysis suggests only a marginal increase in ΔH compared to KCB-1, both ΔE_vdw and ΔE_ele are notably higher (Table 2). The increased ΔE_ele contributions are consistent with loss of the negatively charged phosphate group and its associated ionic contacts. However, the overall ΔG is substantially more favorable than expected for an unstable complex (43.1 kcal mol⁻¹ lower than KCB-1). These results underscore the challenge of capturing accurate entropy contributions using MM-PBSA, especially when configurational sampling is broad, and solvent effects are treated implicitly^24,36,37. As with simulations of tripeptide interactions with dVAN (vide supra), the lower ΔG likely reflects the favorability of dissociation rather than complex stability.

Fig. 4. Global view of binding site changes in KCB-2, KCB-3, and KCB-4.

Snapshots from 100 ns simulations of A KCB-2, B KCB-3, or C KCB-4 that depict select binding interactions characteristic of each variant complex. The GPAs and PSII are shown in white and yellow, respectively, where dashed lines indicate H-bonding interactions of 3.5 Å or less.

Complex KCB-3 — oxazolidinone to 2-pyrrolidone substitution

With our approach to probing the importance of functional groups validated through simulations of dephosphorylated PSII, we set out to identify potentially tunable features of the GPA that govern its interaction with the WTA of C. difficle. Replacing O1 of the oxazolidinone moiety with carbon to generate 2-pyrrolidone (KCB-3) preserved the native stability of the complex (Fig. 3E). MM-PBSA analysis correspondingly showed similar binding energetics, with only a modest increase in ΔE_ele (Table 2). It should be noted, however, that this variant did induce a subtle rearrangement of the complex. Namely, loss of the H-bonding interaction between βGlc5 and the oxazolidinone nitrogen slightly displaced the sugar, but persistent contact with the oxazolidinone carbonyl likely tethers PSII in the binding pocket (Fig. 4B).

Complex KCB-4 — disruption of βGalNAc4 N-acetyl interactions

To assess how the N-acetyl group of βGalNAc4 contributes to GPA binding, we replaced the associated carbonyl oxygen with a methylene group (KCB-4), thereby eliminating a key polar contact observed in KCB-1 (Figs. 3A, E and S3B). This variant weakens interactions between the two molecules but does not result in dissolution of the complex. Instead, βGalNAc4 compensates by forming new H-bonds with the backbone amide of D-Hpg7, shifting the sugar away from the binding pocket (Fig. 4C). βGlc5 is displaced along with it, leaving space for the GPA Man to shift upward and occupy the binding site (Fig. S4). No other H-bonding interactions are substantially disrupted, and dynamic contacts between the GPA Man and the oxazolidinone carbonyl stabilize the conformation throughout the 100 ns simulations. MM-PBSA analysis correspondingly suggests that disruption of the βGalNAc4 N-acetyl minimally perturbs the ΔE_ele contributions to ΔG, while non-polar interactions are weakened (ΔE_vdW ~ 7.0 kcal mol⁻¹ higher than for KCB-1, Table 2). Altogether, these results suggest that interaction with the βGalNAc4 N-acetyl helps anchor the ligand in the primary site but is not essential for overall complex retention.

Complex KCB-5 — substitution of the L-Phe3 backbone amide

We likewise replaced the backbone amide nitrogen of L-Phe3 with a methyl group to eliminate its H-bonding capacity (Fig. 3G). This variant targets the same peptide bond that flips upon Edman degradation of VAN²⁷, which we propose inhibits binding to the tripeptide PG precursor through associated repulsive interactions. Our simulations suggest that such a modification perturbs the H-bonding network more substantially than the KCB-4 variant, ultimately yielding two differentiable ligand conformations that differ in how βGalNAc4 repositions in response to the altered backbone (Fig. 5A, B). In the first conformation, PSII relinquishes contact with the canonical GPA binding site and instead appears to interact with a secondary site formed between the GPA Act and Aco glycosyl moieties. More specifically, βGalNAc4 shifts upward with its N-acetyl group pointing out of the binding pocket, allowing the GPA Man to penetrate more deeply into the binding cleft (Fig. 5A). This rearrangement is aided by a ω-dihedral flip in D-Hpg2, which repositions its carbonyl to H-bond with the Man, alongside the carbonyl of D-Hpg4 (Fig. 5A, bottom inset). The displacement of βGalNAc4 also disrupts interactions between PSII and the Aco unit of the GPA disaccharide: a new H-bond with βGalNAc4 replaces those with αGlc3a. Despite these rearrangements, αMan1 remains anchored to the GPA Glc throughout the simulation (Fig. 5A, top inset).

Fig. 5. Support for a secondary binding site in KCB.

Snapshots from 100 ns simulations of A, B KCB-5, C KCB-6, and D, E KCB-7. In KCB-5, βGalNAc4 either A leaves the canonical binding pocket to interact with the secondary site or B remains near the GPA backbone at the primary site. Shaded regions highlight the primary (red) and secondary (blue) binding sites, while inserts depict key interactions within these regions. C Comparison of 0 (gray) and 100 ns (white) simulations of KCB-6 highlight the concerted rearrangement of L-Phe3 and the oxazolidinone. In KCB-7, PSII adopts either an D extended or E contracted conformation, while maintaining similar interactions with the GPA. In all panels, the GPA and PSII are shown in white/gray and yellow, respectively, with dashed lines indicate H-boding interactions of 3.5 Å or less.

By contrast, in the second conformation, βGalNAc4 shifts downward and its pyranose ring rotates away from the binding cleft (Fig. 5B). However, the N-acetyl group remains H-bonded to the GPA backbone by forming a new contact with the amide of D-Hpg4 (Fig. 5B, bottom inset). We hypothesize that this interaction helps stabilize the complex and suppress the peptide flip seen in the first conformation. Supported by an additional H-bond with the oxazolidinone, the reoriented βGalNAc4 ring pulls on the reducing end of the glycan, disengaging the phosphate from the GPA Glc. Instead, the phosphate forms new interactions with the Aco unit of the disaccharide, aided by weak H-bonds between αMan1 and the GPA Act (Fig. 5B, top inset). As in the first conformation, contacts between αGlc3a and Aco are lost. Following these rearrangements, the glycan adopts a bowed, U-shaped conformation (Fig. S5). MM-PBSA analysis reveals ΔE_ele for KCB-5 is 21.6 kcal mol⁻¹ more favorable than that of KCB-1, suggesting the variant complex samples attractive electrostatic interactions absent in the parent structure (Table 2). ΔE_vdw, by contrast, remain comparable to KCB-1. In summary, the structural dynamics described above, underscore the importance of the βGalNAc4 N-acetyl in PSII-binding, as it can form preferential interactions with either the GPA backbone or Aco substituent to influence the overall configuration of the complex.

Complex KCB-6 — oxazolidinone to γ-butyrolactone substitution

In contrast to the subtler changes observed for KCB-3 through KCB-5 (Fig. 3E–G), substitution of the oxazolidinone nitrogen with carbon produces a dramatic structural rearrangement (Fig. 3I). The N-terminal ring is displaced outward from the binding pocket, likely due to the increased conformational flexibility of the γ-butyrolactone, while the L-Phe3 sidechain rotates up to occupy the vacated space (Fig. 5C). This rearrangement significantly disrupts the PSII binding mode, leading to high configurational sampling. While the WTA remains loosely associated, its disordered motion around the GPA, reflected in elevated RMSD fluctuations, is consistent with MM-PBSA analysis indicating weakened electrostatic and van der Waals interactions (Table 2). As observed for dVAN and KCB-2, ΔG is substantially reduced, supporting the hypothesis that dissociation of the complex is energetically favored (Table 2). In this case, the dynamics of L-Phe3 coupled with the increased solvent exposure of PSII likely increases the configurational entropy by expanding the ensemble of accessible microstates. Nonetheless, it is clear that the oxazolidinone nitrogen forms key interactions that promote a competent complex.

Complex KCB-7 — oxazolidinone to 2-methenyloxazolidine

Replacing the carbonyl of oxazolidinone with a methenyl group, which preserves the double-bond character of the native functional group, yields RMSD values of similar magnitude to KCB-6 (Fig. 3I). Loss of the carbonyl’s H-bonding capacity, combined with mild steric repulsion, disrupts local interactions between the oxazolidinone and βGlc5. This destabilization triggers the displacement of βGalNAc2 and βGalNAc4 from the binding cleft. Comparison with other variant stimulations reveals similar perturbations of βGalNAc4 in KCB-2, KCB-4, conformation 1 of KCB-5, and KCB-6.

Although deviations from the initial binding mode are the most significant of any modification applied in silico, the configurational sampling is reduced compared to KCB-6. This restricted sampling likely arises from capture of the polysaccharide’s reducing end by the secondary binding pocket (Fig. 5D, E) first described in simulations of KCB-5 (Fig. 5A). More specifically, αMan1 maintains loose, but persistent contacts with the GPA Aco, while βGalNAc2 is stably positioned directly above Act for the duration of the simulations. To enable stable formation of this alternative binding mode, αGlc3a forms interactions with the carboxyl group of D-Hpg7. This network of interactions is consistent across two distinct conformations of the glycopolymeric chain: extended (Fig. 5D) and contracted (Fig. 5E). We further examined sampling of the secondary binding site and the relative stability of these two configurations by extending representative trajectories to 300 ns. Over this timescale, the contracted configuration is stably maintained, supported by persistent interactions between the PSII phosphate and cationic amine of Aco, with a smaller contribution from Act. Salt-bridge-mediated stabilization is further reinforced by an extensive H-bonding network involving nearby PSII sugars (Fig. S6). In contrast, the extended configuration exhibits significant conformational variability and samples a heterogeneous ensemble, rather than maintaining a well-defined bound state. Taken together, the KCB-7 simulations demonstrate the oxazolidinone carbonyl is required for primary-site capture of PSII, while disruption of this site does not compromise the secondary binding mode, thereby preserving the GPA–PSII complex.

Discussion and conclusions

Deciphering how GPA scaffolds recognize and bind their targets provides a foundation for developing next-generation therapeutics with improved potency and selectivity. While the H-bonding interactions that mediate GPA engagement with the peptidoglycan precursor are well-established, the influence of other molecular features has yet to be systematically explored. Our simulations of VAN, KMA, and KCB reveal both conserved features and divergent adaptations that govern recognition of canonical (PG precursor) and noncanonical (WTA) targets by GPAs. In particular, we demonstrate how variations in side-chain dynamics, glycosylation patterns, and backbone conformation can reshape the ligand binding landscape, modulating both target specificity and complex stability.

Reassuringly, our MD simulations of tripeptide-bound models recapitulate experimentally observed trends in binding affinities²⁹. Both VAN and KMA form stable complexes, while KCB dissociates despite accessibility of the characteristic H-bonding partners. Although this behavior resembles that of dVAN, a truncated derivative of VAN, the extent of complex instability differs markedly between the two hexapeptides. We therefore propose that the loss of ligand binding cannot be attributed solely to the absence of steric influence from the seventh residue, as suggested previously^17,27. Instead, our analysis of dVAN trajectories indicates that electrostatic effects also contribute via lone pair repulsion between the tripeptide’s acidic terminus and the N-terminal backbone carbonyl of dVAN projecting into the binding site. This revised model, which incorporates both steric shielding and electrostatic complementarity, may help account for the pronounced difference in measured dissociation constants between KCB (250 μM) and dVAN (20,000 μM)²⁹.

The results above reinforce a broader principle: conserved H-bonding interactions alone do not guarantee stable tripeptide-GPA complex formation. Binding additionally depends on how the GPA scaffold shapes the energetic landscape through steric, electrostatic, and solvation effects. The same complexity is evident in our simulations of target binding by KCB and its associated variants generated in silico, which reveal a distinct set of recognition principles beyond the canonical H-bonding contacts. The phosphate group of PSII, for example, plays a dominant role in anchoring the WTA, as its removal leads to high instability and complete loss of binding-site engagement consistent with in vitro data. Other polar features, such as the N-acetyl group of βGalNAc4 and backbone amides, modulate binding stability more subtly. For example, modification of the N-acetyl oxygen leads to displacement of the WTA but retention of binding through compensatory contacts, while perturbation of the L-Phe3 backbone amide induces alternative poses that preserve complex formation but alter ligand orientation. These findings highlight a layered recognition strategy in which multiple polar contacts fine-tune glycan positioning and complex lifetime, even if only a subset is individually essential for binding.

Notably, when interactions with the canonical GPA target-binding site are weakened, PSII appears to sample alternative binding poses stabilized by a secondary interface in which contacts between the GPA glycosyl groups and the reducing end of the polysaccharide become more extensively engaged. This behavior mirrors that proposed for desleucyl-oritavancin, a hexapeptide derivative of the semisynthetic GPA, which retains activity despite impaired binding at the canonical tripeptide-binding cleft³⁸. In that system, MD simulations suggest that alternative poses of the PG precursor engage the pentaglycine bridge—a crosslinking element that links peptide stems in the cell wall of Staphylococcus aureus and related organisms—through hydrophobic interactions with oritavancin’s 4-epi-vancosamine and fluorinated biphenyl bridge (Fig. S7)³⁸. Our results suggest that KCB may likewise accommodate noncanonical ligand configurations by engaging peripheral substituents when native binding is disrupted, suggesting a convergence in binding strategies across GPA scaffolds. However, it is important to note that secondary binding by KCB relies on electrostatic interactions with polar moieties (e.g., between the phosphate group and cationic GPA sugars), rather than on hydrophobic contacts as in oritavancin.

Beyond the target-specific adaptations detailed above, the residue corresponding to position 3 in the canonical heptapeptide scaffold emerges as a flexible and functionally versatile modulator of ligand binding across the GPAs studied. In the absence of substrate, L-Asn3 from VAN occupies the binding cleft to shield the site from solvent and maintain its preorganized architecture^22,23. A similar rearrangement occurs during simulations of KCB-6, where modification of the oxazolidinone nitrogen triggers inward migration of L-Phe3 into the vacated cleft, mirroring the behavior of L-Asn3 in vancomycin (Figs. 5C and S1). In KMA, the 3-Cl-D-Hpg side chain appears to assist L-Phe3 in shielding the backbone from hydration, while in KCB-4, KCB-5, and KCB-7 the GPA Man contributes to binding site desolvation (Fig. S1). Altogether, these observations suggest that the ‘third’ residue is a conserved intramolecular ligand surrogate whose role is to preserve primary binding site accessibility when the cleft is unoccupied. However, more in-depth studies are required to fully elucidate desolvation dynamics and their impact on target binding by these and other GPAs.

In conclusion, our results demonstrate how GPAs leverage both conserved and variable features to fine-tune ligand binding. The ability of these molecular scaffolds to engage the PG precursor through steric preorganization and electrostatic complementarity, or to accommodate noncanonical targets via interactions with peripheral functional groups, highlights the adaptability of this antibiotic class. More broadly, our findings offer a mechanistic framework for rationally designing next-generation GPAs that stabilize multiple binding modes to expand their therapeutic potential.

Methods

Starting models for VAN, dVAN, KMA, and KCB bound to the tripeptide were generated from DFT-optimized coordinates reported in our previous work²⁹. Although, a preliminary model of KCB-1 was available from this previous study, initial MD simulations failed to converge to a consistent binding mode, and additional refinement of the complex was required. A more stable configuration was identified by replacing the simple point charge (SPC) model for solvent with the transferable intermolecular potential with 3 points (TIP3P) model^39,40. Preliminary MD trajectories and 100 ns simulations performed in triplicate confirm persistence of the complex under these conditions. The geometry of the KCB-1 complex was subsequently optimized and frequencies calculated via DFT using the B3LYP-D3/6-311G* (Cl, P)/6-31G** (N, O, C, H)/CPCM(water) level using Gaussian 16, and ΔG of the updated configuration was found to be ~36 kcal mol⁻¹ lower than the previous complex^41–49. The initial model of the KCB-2 complex was generated from this updated KCB-1 template by replacing the phosphate group on αMan1 with a hydrogen. To generate KCB-3, KCB-4, KCB-5, KCB-6, and KCB-7, the O2 of the oxazolidinone, N-acetyl O of βGalNAc4, amide N of D-Hpg2, N1 of the oxazolidinone, or O1 of the oxazolidinone were replaced with a nonpolar carbon containing the same bond order as the native atom. Subsequent molecular dynamic simulations were carried out using GROMACS v2019.3, with 100 ns runs performed in triplicate for each complex. RMSD analysis was performed via the GROMACS rms module and Gibbs free energy of binding values were calculated via MM-PBSA (see Supplementary Methods)^50–54.

Author contributions

K.L.M. and B.D.D. conceptualized the study, built the models, and performed calculations and the subsequent analysis equally. B.M.C. assisted with preliminary calculations and associated analysis. K.M.D. supervised the project and contributed to the interpretation of results. The manuscript was written through contributions from all authors. All authors have given approval to the final version of the manuscript.

Peer review

Peer review information

Communications Chemistry thanks the anonymous reviewers for their contribution to the peer review of this work.

Data availability

Molecular dynamics simulations and starting structures that support the findings of this study are available in the KMDavisLab GitHub repository at https://github.com/KMDavisLab/Expanding-the-paradigm-of-glycopeptide-antibiotic-recognition-through-molecular-dynamics-simulations. All other relevant data are available from the corresponding author upon request.

Competing interests

Some of the authors (K.L.M. and K.M.D.) are on patent US12533389B2 related to this project. All other authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s42004-026-02040-4.