Design, Synthesis, and Evaluation of Novel Inhibitors of Aminoglycoside-Resistance 16S Ribosomal RNA Methyltransferases

Abstract

One of the major mechanisms of resistance to ribosome-targeting antibiotics is the modification of ribosomal RNA (rRNA). Specific methyltransferase enzymes, for example, confer high-level resistance to aminoglycosides by selectively methylating the 16S rRNA in the ribosomal decoding center. These enzymes have been detected globally and pose a threat to the continued use of aminoglycosides. We previously identified 1, a dehydroamino amide inhibitor of the m¹A1408 methyltransferase NpmA, using high-throughput virtual screening. Here, we report the synthesis and biological evaluation of rationally designed analogs of 1. Guided by molecular docking, we predicted additional putative inhibitors of NpmA, as well as the functionally related m⁷G1405 methyltransferase RmtB, varying in each region of the original scaffold. A modular, fragment based synthesis enabled access to 17 analogs, which exhibited mixed activity against NpmA and RmtB, including several which were selective for RmtB. The structure-activity relationship determined for the dehydroamino amide series will guide continued research against this target class with the aim of developing a toolkit for selective- or pan-16S rRNA methyltransferase inhibition.

Graphical Abstract

The continued clinical use of aminoglycoside antibiotics is threatened by resistance conferred by 16S ribosomal RNA (rRNA) methyltranserase enzymes. This report describes the design, synthesis, and biochemical evaluation of novel dehydroamino amide compounds against two representative 16S rRNA methyltransferases, NpmA and RmtB. Some compounds are RmtB-selective, suggesting the possibility of a complete rRNA methyltransferase inhibitor toolkit.

Introduction

The rise of bacterial antimicrobial resistance (AMR) represents a grave threat to modern healthcare. Currently, an estimated 4.71 million deaths are associated with bacterial AMR, and this figure is forecast to increase to 8.22 million by 2050.^[1] Meanwhile, the discovery and approval of novel antibacterial therapies have failed to match the pace of AMR development, leading us toward a “post-antibiotic era.”^[2] Therefore, there is an urgent need to develop novel strategies to combat increasingly dangerous bacterial infections.

Aminoglycoside antibiotics are a class of broad-spectrum bactericidal agents which act via inhibition of protein synthesis by binding to the 16S ribosomal RNA (rRNA) of the small 30S subunit of the bacterial ribosome.^[3] Since the introduction of streptomycin in 1944, aminoglycosides have seen continued clinical use, with medicines including gentamicin, amikacin, and tobramycin commonly used to treat serious gram-negative and mycobacterial infections.^[4]

Unfortunately, the clinical efficacy of aminoglycosides is under threat due to the spread of AMR. A commonly identified mechanism of resistance is mediated by aminoglycoside modifying enzymes (AMEs),^[5] which acylate,^[6] phosphorylate,^[7] or adenylate^[8] aminoglycosides, rendering them inactive. A primary focus of current research is the discovery of aminoglycosides which are not readily modified by AMEs. The next-generation semisynthetic compound plazomicin was discovered to evade most AMEs and has found limited clinical use.^[9,10] However, nearly all aminoglycosides, including plazomicin,^[11] remain highly sensitive to aminoglycoside-resistance 16S rRNA methyltransferase enzymes, which represent a growing cause for concern.^[12–14] These enzymes confer resistance by modifying the bacterial ribosome small subunit at the site of aminoglycoside binding to interfere with their action. The encoding genes, originally observed in aminoglycoside producers—which express the enzymes as a self-protection mechanism—have more recently been discovered in human pathogens.^[15,16] The global detection of such genes in several gram-negative species such as Escherichia coli, Klebsiella pneumoniae, and Pseudomonas aeruginosa, often alongside β-lactamase genes,^[17] marks aminoglycoside-resistance rRNA methyltransferases as a serious and unaddressed threat to aminoglycoside use.

There are two families of aminoglycoside-resistance 16S rRNA methyltransferases, which each site-specifically modify the 16S rRNA and thus block the binding of aminoglycosides. ArmA and RmtA-H catalyze the transfer of a methyl group from S-adenosyl-L-methionine (SAM) to the N7 atom of G1405, which confers resistance to most clinically-relevant aminoglycosides, which bear a 4,6-disubstituted 2-deoxystreptamine core.^[18,19] NpmA-C, in contrast, mediate methyl transfer from SAM to the N1 atom of A1408, which confers resistance to a broader panel of aminoglycosides, including those with a 4,5-disubstituted 2-deoxystreptamine core.^[20–22] While NpmA has been identified in various gram-negative pathogens and in Clostridium species, the Rmt family of 16S rRNA methyltransferases has shown broader dissemination and is often co-associated with other resistance elements.^[17]

Based on their broad aminoglycoside resistance profiles, global dissemination, and well-defined mechanism, we pursued the Npm and Rmt families of 16S rRNA methyltransferases as targets for “antibiotic resistance breakers,” defined as compounds that act directly on resistance determinants to restore activity to affected antibiotics.^[23] As such, the new compounds would not be expected to possess inherent antimicrobial activity, but to restore the activity of existing aminoglycoside antibiotics. The most widely cited examples of antibiotic resistance breakers are β-lactamase inhibitors, which have proven highly effective in clinical practice.^[24,25] To our knowledge, this strategy has not previously been applied to aminoglycoside-resistance 16S rRNA methyltransferases outside of our work.

We recently reported the first-in-class inhibitors of NpmA, including compound 1, identified through high-throughput virtual screening, precision docking, and iterative rounds of inhibition assays. The Glide extra precision (XP) docking poses were subsequently used for Molecular Mechanics – Generalized Born Surface Area (MMGBSA) calculations to estimate the binding free energy and molecular dynamics (MD) simulations to elucidate key interactions and the contributions of specific residues.^[26] Although compound 1 and its analogs were initially screened against NpmA in its 30S subunit-bound conformation, which features a distinct Y-shaped binding pocket, we hypothesized that they might also exhibit cross-inhibition of the Rmt family, given the conserved features of the SAM-binding site and their shared function of rRNA purine modification via a “base flipping” mechanism.^[19–22] We therefore selected RmtB from the Rmt family of methyltransferases and report the first known inhibitors of this enzyme.

Based on our previous identification of 1 as an inhibitor of NpmA,^[26] we synthesized and evaluated a series of dehydroamino amide analogs against both NpmA and RmtB. While some analogs of 1 exhibit selective inhibition of RmtB over NpmA, suggesting that specific targeting of individual 16S rRNA methyltransferases is feasible, further optimization is needed to enhance both potency and selectivity. This structure–activity relationship (SAR) study provides valuable insights for the future design of inhibitors targeting both the NpmA and RmtB families, using the scaffold 1 as a foundation.

Results and Discussion

Synthesis of methyltransferase inhibitors

Compound 1 (Table 1) comprises three aromatic groups, with the “northern”, “western” and “eastern” arenes predicted to dock in the SAM methionine, SAM adenine and A1408 adenine binding pockets of NpmA, respectively (Fig. 1A). To access analogs of 1, we designed and optimized a robust and modular synthesis (Scheme 1) starting from commercially available phosphonate Int-1,^[27] enabled by a key late-stage olefination.^[26] In the first iteration, we accessed 2–3 via Int-6 and Int-7, respectively: Int-1 was hydrolyzed to the carboxylic acid and coupled with aniline Int-3, then subjected to hydrogenolysis and coupled with 3-bromobenzoic acid. The decorated phosphonate Int-5 underwent Horner-Wadsworth Emmons (HWE) olefination using NaH as the base to give a mixture of Int-6 and Int-7, which were separated by chromatography and assigned by x-ray crystallography. The protected intermediates were individually deprotected and oxidized to the desired carboxylic acids.

Table 1.

Methyltransferase inhibition data and molecular docking scores for 1 – 3.

Compound	NpmA Glide XP (kcal/mol)	NpmA %Inhibition (at 1 mM)	NpmA IC50 (μM)	RmtB Glide XP (kcal/mol)	RmtB %Inhibition (at 1 mM)	RmtB IC50 (μM)
1	−9.9	>99 ± 0.2	76 ± 5	−6.6	> 99 ± 4	75 ± 6
2	−6.8	50 ± 30	>750[a]	−6.0	NI	NI
3	−8.1	99 ± 1	402 ± 72	−5.8	95 ± 6	215 ± 37

^[a]IC₅₀ for 2 with NpmA is an estimate only due to limited compound solubility and variability between replicates. NI = No inhibition.

Figure 1.

A. Predicted binding poses of m¹A1408 methyltransferase NpmA with 1 (cyan) and 3 (green), both in the (Z) stereoisomeric configuration. B. RmtB with bound 1 (cyan) illustrates one representative pose within the distinct Y-shaped pocket of the m⁷G1405 methyltransferase. C. Overlay of RmtB bound to 2 (E) (orange) and 3 (Z) (cyan), highlighting differences in binding orientation due to the expanded volume and conformational plasticity of the pocket in this enzyme.

Scheme 1.

i. NaOH, H₂O, 1,4-dioxane, rt, 97%. ii. HATU, DIPEA, MeCN, rt, 90%. iii. H₂, Pd/C, MeOH, rt. iv. 3-bromobenzoic acid, HATU, DIPEA, MeCN, rt, 96% (2 steps). v. NaH, THF; then benzaldehyde, 0 °C, 35% (Int-6), 15% (Int-7). vi. TBAF, THF, rt, 95 – 97%. vii. DMP, DCM/THF, rt. viii. NaClO₂, NaH₂PO₄, 2-methyl-2-butene, MeCN, H₂O, 14–21% (2 steps). ix. KOtBu, THF; then aldehyde, rt, 19–61%. x. TBAF, THF, rt, 66 – 99%.

The synthesis was optimized to access 4–9 by replacing Int-3 with TMSE-protected aminocarboxylate Int-8,^[28] simplifying the latter part of the synthesis. A variety of carboxylate protecting groups were evaluated; however, simple alkyl esters could not be readily hydrolyzed at the end of the synthesis, and the corresponding trichloroethyl derivative could not be hydrolyzed or removed reductively. Additionally, we found that changing the base to KO^tBu in the olefination reaction led to preferential formation of the (Z) olefins in improved yield. The olefin configuration was assigned by x-ray crystallography or NMR using ¹H-¹³C HMBC, ¹H-¹³C HSQC, and ¹H 1D selective NOESY experiments.^[26]

Similar chemistry was employed to access 12, where the western aniline fragment was replaced with 4-aminobenzonitrile and the eastern carboxylic acid fragment was replaced with 2-chlorobenzoic acid. Bicyclo[1.1.1]pentane (BCP) analogs 10–11 were prepared in a similar manner, where the western amine fragment Int-19 was synthesized from commercially available monoester Int-17 (Scheme 2) via one-pot Curtius rearrangement/Boc protection and subsequent deprotection.^[29]

Scheme 2.

i. DPPA, Et₃N, tBuOH, rt; then reflux [Caution: explosion hazard], 52%. ii. HCl, dioxane, rt, 85%. iii. HATU, DIPEA, MeCN, rt. iv. H₂, Pd/C, MeOH, rt. v. 3-bromobenzoic acid, HATU, DIPEA, MeCN, rt, 91% (3 steps). vi. KOtBu, THF; then aldehyde, rt, 18 – 34%. vii. NaOH, H₂O/1,4-dioxane, rt, 84 – 88%. vii. 4-aminobenzonitrile, HATU, DIPEA, MeCN, rt, 99%. ix. H₂, Pd/C, MeOH/EtOAc, rt, 64% (2 steps). xi. KOtBu, THF, 0 °C, 25%.

To prepare compounds 13–18, we prepared phosphonate Int-24 using identical chemistry to the previous route. Although olefination with aldehyde Int-25 failed under the previously optimized conditions, treatment of Int-24 with LiCl and Et₃N at 0 °C, followed by the aldehyde, provided the desired olefins in moderate yield with good (Z) selectivity.^[30] Int-26 could be converted to additional derivatives by hydration of the nitrile to the primary amide (16)^[31] followed by Boc deprotection (15) and N-methylation (14) or removal of the Boc protecting group followed by N-methylation (13). Alternatively, the modified HWE protocol could be applied to phosphonates Int-32 and Int-33 to afford 17 and 18 following Boc deprotection (Scheme 3).

Scheme 3.

H₂, Pd/C, MeOH/EtOAc, rt. ii. Isoquinoline-6-carboxylic acid, HATU, DIPEA, MeCN, rt, 43% (2 steps). iii. Et₃N, LiCl, DMF; then Int-25, 0 °C, 55%. iv. Cu(OAc)₂, Et₂NOH, MeOH, rt, 50%. v. HCl, MeOH/1,4-dioxane; then formalin, STAB, MeOH, rt, 43% (2 steps). vi. TFA, DCM, rt; then Amberlite IRA-400 (Cl^‒), 72%. vii. Formalin, STAB, MeOH, rt, 32%. viii. Amine, DIPEA, rt, 54 – 99%. ix. H₂, Pd/C, MeOH/EtOAc. x. 2-chlorobenzoic acid, HATU, DIPEA, 61 – 64% (2 steps). xi. Et₃N, LiCl, DMF; then Int-25, 0 °C, 28 – 96%. xii. HCl, 1,4-dioxane, rt, 26 – 30%.

Evaluation of methyltransferase inhibitor activity

Docking studies strongly favored the (Z) configuration (Fig. 1A; Glide XP score of −9.9 kcal/mol) and we confirmed that the active compounds bear the (Z) configuration through NMR-based assignments.^[26] To further confirm the preference of NpmA for the (Z) isomer, we synthesized two analogs of 1 bearing an unsubstituted phenyl group in the (E) and (Z) configurations. The synthesized compounds were first evaluated in a fixed-concentration enzyme activity assay; as predicted, (Z) olefin 3 demonstrated complete inhibition of NpmA at 1 mM, compared to partial inhibition by (E) olefin 2, although activity (IC₅₀) was reduced compared to 1 (Table 1). Glide XP scores and docking poses revealed that 3, bearing the central (Z)-olefin, adopts a binding orientation similar to that of 1 (Fig. 1A).

While our lead compound 1 was originally designed and tested against NpmA, an A1408 methylating enzyme, we extended its evaluation to RmtB, a G1405 methylating enzyme. This decision was motivated by two key factors: first, both enzymes recognize and modify purine bases and position them through distinct yet analogous sets of residues; second, using a homology model based on the RmtC–30S ribosome subunit structure (which became available during the course of this study), we identified a potential Y-shaped binding pocket within RmtB comprising the SAM cosubstrate and adjacent purine base G1405 binding pockets.^[19,21] Although the Y-shaped binding pockets in NpmA and RmtB differ in residue composition, we reasoned that the latter might potentially be capable of accommodating 1 or its analogs and any such cross-selectivity would provide a valuable lead for optimizing inhibitors targeting RmtB, a clinically more significant enzyme. Interestingly, 1 and 3 significantly inhibited RmtB (Table 1), while the (E) olefin again showed limited activity. Our docking studies show that 1 binds to RmtB within its Y-shaped pocket, while 2 (E) and 3 (Z) adopt distinct binding orientations within the same pocket due to its expanded nature in RmtB compared to NpmA (Fig. 1B, C).

Having confirmed the preference for (Z) olefins for both NpmA and RmtB, we examined the effect of polar substituents in the northern arene with a series of nitroarenes and pyridines, all bearing the (Z) configuration. These compounds demonstrated slightly improved activity compared to 3 but were all less potent than 1. Of this series, the most potent was 2-pyridyl analog 4. Of particular interest was compound 5, which demonstrated near-total inhibition of RmtB with relatively weak inhibition of NpmA at 1 mM (Table 2).

Table 2.

Methyltransferase inhibition data and molecular docking scores for 4 – 12 with varying eastern arene (R).

Compound	R	NpmA Glide XP (kcal/mol)	%Inhibition (1 mM, NpmA)	RmtB Glide XP (kcal/mol)	%Inhibition (1 mM, RmtB)
4	2-pyridyl	−8.1	94.0 ± 0.9	−7.2	> 99 ± 1
5	3-pyridyl	−7.2	23 ± 13	−5.8	98 ± 2
6	4-pyridyl	−8.4	48 ± 5	−5.6	70 ± 10
7	2-nitrophenyl	−7.1	80 ± 4	−6.3	64 ± 3
8	3-nitrophenyl	−7.9	51 ± 5	−6.5	82 ± 3
9	4-nitrophenyl	−6.5	89 ± 2	−6.5	71 ± 10
10	phenyl	−7.7	NI	−6.6	NI
11	2-pyridyl	−7.4	NI	−6.9	NI
12	3-nitrophenyl	−8.1	> 99 ± 0.3[a]	−6.5	ND

^[a]Assay was performed at 0.5 mM. NI = no inhibition. ND = not determined.

Shifting focus to the western region of the scaffold, we prepared 10 and 11, bearing the BCP motif as a bioisosteric replacement. We hypothesized that this modification could lead to an improvement in aqueous solubility as reported in the literature,^[32,33] considering that all previous analogs prepared contained < 5% Csp³. We specifically chose to replace the western arene of 1, as neither the Glide XP scores nor the predicted binding poses suggested any significant weakening of interactions upon this substitution. However, biochemical assays revealed that neither 10 nor 11 exhibited measurable activity against NpmA or RmtB. Further assessments using flexible docking and alternative protocols also failed to discriminate between the 10 and 11 compared to 1 and other analogs based on docking scores. This discrepancy thus likely highlights a key limitation in current docking algorithms when aromatic systems are replaced with sp³-rich, saturated bioisosteres such as BCP, docking algorithms often fail to penalize the loss of key aromatic-specific interactions.^[34,35] This limitation stems from the fact that most scoring functions are trained on datasets dominated by planar, aromatic ligands, and thus do not accurately capture the geometric and electronic consequences of BCP substitution.

Comparison between 1 and 8 suggested that modifications to the western and eastern regions of the scaffold could be negatively impacting potency and led us to consider whether the western carboxylic acid was responsible for the weak activity of the synthesized analogs. To address this question, we prepared nitrile 12 and found comparative improvement in NpmA inhibition, corroborated by docking scores. At this stage, we sought to address apparent issues contributing to poor biochemical potency: potential repulsion between the western carboxylic acid and Glu88, poor engagement with the π-stacking residues in the A1408 adenine binding pocket, and overall lack of sp³ character leading to poor solubility.

Specifically, we replaced the carboxylic acid on the western arene with other polar functionalities to assess the contribution of charge-based interactions on its potential inhibition activity. Additionally, we introduced bulkier polycyclic systems in the eastern region to better mimic the spatial and electronic features of the A1408 adenine. To further enhance solubility, we incorporated saturated rings such as azetidines in the northern region. Based on these design principles, we synthesized a series of compounds incorporating one or more of these features to evaluate their collective impact on predicted binding and activity. While some analogs (15, 16, and 18) produced improved XP Glide docking scores relative to 4–12 against NpmA, these predicted affinities did not correlate with experimental results which showed weak inhibitory activity (Table 3). Only 15 and 16 exhibited >25% inhibition activity against NpmA at 1 mM. We also prepared 17 and 18 to obtain a more direct comparison to 1, but these were also inactive, suggesting that changes to the northern arene are not tolerated regardless of changes to the eastern and western regions. Notably, incorporation of the northern azetidine moiety likely reflects the same limitation observed with the BCP analogs, reflecting the tendency of docking algorithms to overestimate binding affinity due to inadequate modeling of the loss of π-stacking, conformational constraints, and steric mismatches introduced by sp³-rich, saturated scaffolds.

Table 3.

Methyltransferase inhibition data and molecular docking scores for 13 – 18 with varying western arene (R¹) and azetidine subsituent (R²).

Compound	R1	R2	NpmA Glide XP (kcal/mol)	%Inhibition (1 mM, NpmA)	RmtB Glide XP (kcal/mol)	%Inhibition (1 mM, RmtB)
13	CN	Me	−5.0	16 ± 6	−6.4	NI
14	C(O)NH2	Me	−4.9	16 ± 4	−5.5	NI
15	C(O)NH2	H	−8.7	46 ± 5	−5.7	52 ± 10
16	C(O)NH2	Boc	−8.2	32 ± 13	−7.8	NI
17	CN	H	−5.5	NI	−6.1	NI
18	C(O)Me	H	−9.7	7 ± 1	−6.0	NI

NI = no inhibition.

SAR of analogs of 1 with NpmA and RmtB

All molecular docking calculations were performed using Schrödinger Suite 2024–3, employing the Glide module in XP mode for scoring and pose prediction, as done in our previous studies.^[26,36–38] As mentioned earlier, analogs of 1 were primarily designed for targeting NpmA. Consistent with our initial hypothesis, NpmA docking results showed clear stereoselectivity: compound 3, the (Z) isomer, demonstrated a significantly better XP Glide score than its (E) isomer counterpart, 2. For analogs 4–9, docking poses closely resembled that of 1, with binding energies generally more favorable than −6 kcal/mol, suggesting consistent binding modes within the Y-shaped pocket of NpmA. In contrast, modifications to the western arene, specifically changes in the polarity of the carboxylic acid group, led to improved docking and inhibition outcomes for 12, indicating that tuning electrostatic interactions in this region may be a viable strategy for enhancing NpmA selectivity.

In the case of RmtB, the SAR pattern diverged from NpmA. While compound 1 displayed similar binding poses in both enzymes, stereoisomers 2 and 3 docked differently in RmtB but were both reasonably accommodated, suggesting a larger pocket. Notably, the overall Glide XP scores for RmtB were generally lower than those for NpmA, which may reflect limitations in homology modeling due to the lack of a 30S–RmtB complex structure. Still, several analogs (e.g., 4–8) maintained moderately favorable scores (below −5.5 kcal/mol).

The expanded binding pocket in RmtB appears to allow tolerance of bulkier substitutions such as bicyclic aromatic modifications to a greater extent. To understand how RmtB accommodates 1 and its analogs, we analyzed docking poses of compounds 1–18 within the predicted binding pocket. Docking revealed a distinct Y-shaped binding pocket that appears larger and more open than the corresponding pocket in NpmA, enabling greater ligand flexibility and potentially broader substrate tolerance. When the crystal structure of RmtB (PDB: 3FRH) is modeled with 1, the SAM pocket is in an open conformation but the phenyl ring of the ligand clashes with the closed G1405 pocket (Fig. 2A). However, comparison with RmtB modeled in the open (30S subunit-bound) conformation highlights the conformational rearrangement involving residues Y56, V57, and H81 (Fig. 2A,B). Specifically, these residues undergo a significant shift upon 30S association and thus transition from the closed to open configuration of the G1405 binding pocket. To further investigate the dynamic behavior of this pocket, we performed three replicate 100 ns molecular dynamics (MD) simulations of RmtB, after removing the SAM cosubstrate, to assess its intrinsic flexibility. The inter-residue distance between Y56-H81 and V57-H81 was tracked over time as a proxy for Y-shaped pocket openness (Fig. 2C). These simulations specifically aimed to quantifying the degree of pocket closure and the percentage of time the Y-shaped pocket remains in an accessible conformation suitable for inhibitor binding. Our analysis revealed that the Y-shaped pocket remains in an open conformation approximately 57.1% of the time, indicating favorable ligand binding accessibility and supporting its potential as a druggable site.

Fig. 2.

RmtB Y-shaped pocket dynamics. A. Crystal structure of RmtB (PDB: 3FRH) bound to 1 (Z), showing the SAM pocket in an open conformation while the G1405 channel remains closed (top) and homology model of RmtB based on the 30S-bound RmtC structure (PDB: 8GHU), illustrating an open conformation for both the SAM pocket and the G1405 channel in complex with 1 (Z) (bottom). B. Conformational change involving key residues Y56, V57 and H81, which move upon 30S binding (green), forming the G1405-binding pocket. C. Tracking the inter-residue distance between Y56-H81 and V57-H81 in MD simulations of RmtB as a metric of Y-shaped pocket formation over time, in the absence of SAM or the 30S ribosome subunit.

Conclusion

We synthesized 17 analogs of 1 and found that modifications did not provide improved inhibitory activity against NpmA or RmtB. However, several important findings include that compounds bearing the (Z)-dehydroamino amide core are more active than the (E) isomers, in accordance with our docking results; replacement of the western or northern regions with the sp³-rich BCP or azetidine moieties, respectively, is not tolerated; and modifications to the polar substituent on the northern arene allows some modulation of biochemical potency.

Our synthesis and SAR strategy focused on systematically modifying substituents in the northern, western, and eastern regions of the scaffold guided by molecular docking. These modifications aimed to generate a more focused SAR dataset and improve docking models. Glide XP docking, complemented by additional computational tools, provided reasonable binding poses. While docking accurately captured the activity trend for NpmA analogs 1–3, including stereoselectivity, it failed to do so for RmtB. Despite the limitations of the docking model, we identified several compounds (e.g. 5, 8) capable of inhibiting RmtB while demonstrating little effect on NpmA. The enhanced selectivity of 5 and 8 toward RmtB can be reasonably attributed to functionally important differences in the dynamics of the RmtB pocket, as well as the conformational flexibility of these compounds. While the targeted ligand binding pockets of both NpmA and RmtB share a broadly similar Y-shaped topology, they differ in pocket depth and width, side-chain positioning of residues lining the nucleotide channel, and hydration patterns within the extended binding groove. These distinctions likely enable 5 and 8 to achieve a better fit within RmtB’s comparatively larger and more solvent-accessible cavity compared to NpmA. Docking successfully differentiated E/Z isomers, aligning with observed activity trends, but lacked quantitative correlation with inhibition, likely due to receptor flexibility and absence of explicit water. These findings highlight the limitations of traditional scoring functions for flexible targets like RmtB and underscore the need for advanced, data-driven modeling to better capture binding determinants. We are currently investigating this phenomenon to further understand the structural basis for selectivity in the hopes of designing a toolkit for selective- or pan-16S rRNA methyltransferase inhibition. Identification of inhibitors with increased potency and solubility will also enable future biological evaluation of their activity that was not possible in this work due to limited compound solubility in bacterial growth medium. However, we note that compound 1 had no apparent impact on bacterial growth at the highest concentration achievable. We hope that continued research into this target class can lead to useful adjuvants for the clinical treatment of aminoglycoside-resistant bacterial infections.

Experimental

Methyltransferase Inhibition Assay

Recombinant NpmA and RmtB proteins were expressed and purified, and E. coli 30S ribosomal subunits were isolated using established protocols.^[21,39] Test compounds were dissolved in DMSO to prepare 10 mM stock solutions. Methyltransferase reactions were carried out in a final volume of 30 μL, containing 0.3 μM NpmA or RmtB, 0.3 μM 30S subunit, 0.302 μM [³H]-SAM, and test compounds at final concentrations of 1.0 mM (unless otherwise stated). The reaction buffer consisted of 5 mM HEPES-KOH (pH 7.5), 50 mM KCl, 10 mM NH₄Cl, 10 mM magnesium acetate, 6 mM β-mercaptoethanol, and 10% (v/v) DMSO. Reactions were incubated at 37 °C for 10 minutes prior to quantification of methylation activity, assessed by measuring [³H]-methyl group incorporation into the 30S subunits via a filter binding assay. Following incubation, reaction mixtures were transferred to a glass fiber filter 96-well plate, thoroughly washed to remove unincorporated [³H]-SAM, dried, and the retained radioactivity was measured using liquid scintillation counting. Dose-response curves for selected compounds were generated using the same assay format, with a half-log dilution series ranging from 1000 to 7.81 μM. All assays were performed in at least duplicate. IC₅₀ values were determined by fitting the inhibition data to a four-parameter logistic regression model using GraphPad Prism version 10.

Computational Modeling

We modeled the open conformation of E. coli RmtB bound to the 30S ribosomal subunit using homology modeling tools available in Schrödinger BioLuminate (version 2024–4). As a structural template, we used our previously determined cryo-EM structure of the E. coli 30S–RmtC complex (PDB ID: 8GHU)^[19], which captures the open conformation of the methyltransferase bound to the ribosome. The RmtB sequence from E. coli (UniProt ID: Q763K9) was aligned to the RmtC sequence using BioLuminate’s sequence alignment and template-based modeling pipeline. The modeling protocol accounted for backbone and side-chain adjustments to reflect sequence differences between RmtC and RmtB while preserving key interfacial interactions. The resulting model was subsequently energy-minimized using Schrödinger’s Protein Preparation Wizard to resolve any steric clashes.

A Y-shaped ligand-binding pocket in the open conformation of NpmA, derived from its crystal structure1 (PDB code 4OX9),^[27] was used as the receptor for docking score calculations, as described. For RmtB, two receptor models were employed for docking score calculations: (i) the apo RmtB crystal structure (PDB: 3FRH), which adopts a closed conformation of the G1405 channel (see main Fig. 2A, top), and (ii) a homology model of RmtB constructed using the 30S-bound structure of RmtC (PDB code 8GHU)2, which adopts an open conformation of both the SAM pocket and G1405 channel. Hydrogen atoms were added, and all receptor models were energy-minimized prior to docking. Docking grids were centered on the SAM-binding site and a cubic grid box of 15 Å per side was generated. Ligand docking was performed using the GLIDE XP algorithm within the Schrödinger Suite (version 2024–4). All structural visualizations were performed using PyMOL (Schrödinger, LLC), which was also used to analyze and illustrate key ligand–protein interactions, residue conformations, and binding pocket architectures as established in previous studies.^[40]

To investigate the conformational dynamics of RmtB, particularly the formation and stability of the Y-shaped ligand-binding pocket, MD simulations were performed for RmtB in the absence of SAM and the 30S ribosome, similar to our previous study with NpmA.^[26] The apo form of RmtB (PDB code 3FRH), was prepared using the Schrödinger Protein Preparation Wizard, and simulations were carried out using the Desmond with the OPLS4 force field (Schrödinger version 2024–4).^[39] Three replicate simulations were performed, each initiated with distinct random velocity seeds, using our established Desmond protocols for classical MD simulations, including multistep relaxation and OPLS4-based force field parameters. The system was solvated in an explicit TIP3P water box and neutralized using sodium ions via the System Builder module. To mimic physiological ionic strength, 150 mM NaCl was introduced by randomly replacing water molecules with Na⁺ and Cl⁻ ions. The system then underwent a five-step standard relaxation protocol in Desmond: 1) energy minimization was performed for 1000 steps with positional restraints (50 kcal/mol/Ų) applied to all solute heavy atoms; 2) a 12 ps NVT simulation with temperature ramping up to 310 K under the same restraints to allow solvent relaxation; 3) a subsequent 12 ps NPT simulation at 1.01325 bar was conducted to permit volume equilibration; followed by 4) another 24 ps NPT simulation with restraints only on solute heavy atoms; and 5) an unrestrained 24 ps NPT simulation was performed for full system equilibration. Production simulations were then run for 100 ns following a 10 ns unrestrained equilibration phase, all under NPT ensemble conditions. Temperature and pressure were maintained using the Nosé–Hoover chain thermostat and Martyna–Tobias–Klein barostat with relaxation times of 1 ps and 2 ps, respectively. The equations of motion were integrated using a 2 fs time step for short-range and 6 fs for long-range interactions, with a nonbonded interaction cutoff of 10 Å. To evaluate the Y-shaped pocket dynamics, distances between Y56 (Cα) and H81 (NE2), and V57 (Cα) and H81 (NE2), were measured throughout the simulations using Maestro. A conformation was defined as “open” if the Y56–H81 distance exceeded 9 Å and the V57–H81 distance exceeded 11 Å.

Supplementary Material

Detailed experimental procedures and analytical data relating to chemical synthesis are provided in the Supporting Information. Crystallographic data for S1 (2500720), S5 (2464629), and 9 (2464630) have been deposited at The Cambridge Crystallographic Data Centre (CCDC, https://www.ccdc.cam.ac.uk/

Supporting information for this article is given via a link at the end of the document.