Virtual screen of 12 million compounds yields new bacterial kinase inhibitors

Antibiotic resistance is one of the defining medical challenges of the century, and much of the problem traces back to a set of molecular switches that pathogenic bacteria use to survive hostile environments. These switches, known as two-component signal transduction systems, pair a sensor histidine kinase with a response regulator to control genes governing drug resistance, virulence, and biofilm formation. Because the histidine kinases that drive these systems are absent from human cells, they have long been viewed as attractive targets for a new generation of antimicrobials. Yet despite decades of effort, no histidine kinase inhibitor has reached the clinic. A team of Japanese researchers now reports a fresh route into this stubborn target class, one that begins not with a protein structure but with a natural product and a computational descriptor built entirely from amino acids.

The starting point for the study, published in The Journal of Antibiotics, is waldiomycin, an antibiotic originally isolated from a Streptomyces actinomycete. Waldiomycin is a striking molecule: a set of angucyclic rings bearing a naphthoquinone scaffold, connected through a tetraene linker to a dioxolan ring. Earlier nuclear magnetic resonance work had suggested that the angucyclic portion of the compound wedges itself into the H-box, a short stretch of amino acids conserved across bacterial histidine kinases, and thereby suppresses the autophosphorylation reaction that the catalytic ATP-binding domain depends on. Crucially, the H-box sequence is specific to histidine kinases and does not appear in the serine/threonine or tyrosine kinases of human cells, which makes it a potentially selective point of attack against dangerous pathogens, including the ESKAPE organisms that dominate hospital-acquired infections.

Previous attempts to drug this region have produced a long list of chemotypes, including trityles, benzoxazines, salicylanilides, triazoles, benzimidazoles, benzoxazoles, thiazolidiones, thiophenes, and adenine derivatives. None has advanced to clinical testing. The problem, in part, is that most medicinal chemistry campaigns against histidine kinases have started from structures that resemble the natural ligand, limiting the chemical space explored. The new work, led by Teruhiko Ishikawa, Yoko Eguchi, Toshihide Okajima, Masayuki Igarashi, and Ryutaro Utsumi, set out to break that pattern by using a screening method that deliberately ignores the three-dimensional structure of the protein-ligand complex altogether.

The method, called amino acid mapping, was proposed by Kyosuke Tsumura and colleagues. Instead of modeling how a compound sits inside a binding pocket, amino acid mapping asks a simpler question: which amino acid residues tend to cluster around the compound in aqueous solution? The answer is encoded as a spatial probability distribution of individual amino acid residues surrounding the molecule, calculated from molecular dynamics simulations and independent of any explicit protein structure. Two compounds with similar descriptors are likely to be recognized by similar binding environments, even if their chemical scaffolds look nothing alike. This makes the approach a form of scaffold hopping: it can carry the biological activity of a known ligand across to chemically distant molecules without requiring a crystal structure of the target.

To apply the method, the team first computed a high-precision reference descriptor for the naphthoquinone moiety of waldiomycin, the fragment that directly contacts the H-box. One hundred low-precision conformations were generated with Open Babel and then refined using density functional theory calculations with the B3LYP functional, a 6-31G* basis set, and a polarizable continuum solvation model in Gaussian16, yielding the most stable conformation. The researchers then compared this reference against a commercially available library of 11,973,825 compounds, using cosine similarity between descriptors as the measure of resemblance. Because a full comparison of millions of compounds is computationally punishing, the team applied a stepwise filter: compounds were first screened against a phenylalanine-only descriptor, then against descriptors for charged residue types, and only survivors of both stages underwent the complete 27-residue calculation. A similarity threshold of 0.75, chosen on the basis of earlier screening campaigns in which roughly one in ten compounds above that value showed activity comparable to the reference ligand, narrowed the field to 714 candidate molecules.

From those 714 hits, the team extracted rigid substructures, defined as condensed ring systems together with adjacent atoms connected through double or higher-order bonds, using an RDKit-based Python workflow. The analysis yielded 227 distinct scaffolds. After excluding frameworks smaller than a two-ring system, the eleven most frequent scaffolds were shortlisted, and the team settled on 9-hydroxy-4H-pyrido[1,2-a]pyrimidin-4-one, a scaffold ranked ninth by frequency but chosen for its novelty as an H-box inhibitor and its synthetic accessibility. Fifteen derivatives were synthesized through condensation reactions in polyphosphoric acid, nitration and reduction sequences, and a newly developed magnesium(II) triflate-mediated coupling, producing compounds designated FF-053 through FF-081.

The biological results were encouraging. When the fifteen compounds were tested at 100 micrograms per milliliter against the cytoplasmic regions of three histidine kinases, Bacillus subtilis WalK, Streptococcus mutans VicK, and Escherichia coli EnvZ, several showed more than 30 percent inhibition of autophosphorylation and were advanced to full dose-response analysis. Three derivatives, FF-054, FF-055, and FF-060, emerged as the standouts, with IC50 values ranging from 4.39 to 259 micromolar. Against WalK, none of the new compounds surpassed waldiomycin itself, which inhibits at 13.1 micromolar. But against VicK, the story reversed dramatically: FF-055 inhibited at 6.92 micromolar and FF-060 at 4.39 micromolar, both considerably more potent than waldiomycin’s 25.8 micromolar. The authors suggest that structural differences between the dimeric forms of WalK and VicK, which remain incompletely characterized, may explain the enhanced affinity of the new compounds for VicK.

The mechanistic evidence is where the study becomes particularly compelling. When the conserved H-box serine at position 242 of EnvZ, adjacent to the autophosphorylation site H243, was mutated to alanine, the potency of all three lead compounds dropped three- to six-fold, while a T247A mutation had no such effect. This pattern contrasts with waldiomycin, which loses potency against both mutants, and suggests that the new compounds bind on the S242 side of the phosphorylated histidine rather than the T247 side favored by the natural product. Nuclear magnetic resonance experiments with the isolated DHp domain of EnvZ, residues 223 to 289, confirmed direct physical interaction: addition of FF-054 perturbed the chemical shifts of V241, D244, and R246 within the H-box, while FF-060 produced dose-dependent perturbations at S242, R246, and T247. Affinity selection mass spectrometry added a further twist. At 100 micromolar, FF-060 increased the amount of waldiomycin bound to VicK from roughly 50 to 120 nanomolar, mirroring the behavior of previously described Type B H-box inhibitors, which bind a distinct site and allosterically enhance waldiomycin binding to both H-boxes of the dimeric kinase.

Perhaps the most elegant demonstration came from converting the descriptors themselves into a map. The team computed a ten-by-ten similarity matrix among the naphthoquinone reference, the three new pyridopyrimidinones, and six previously synthesized naphthoquinone-derived inhibitors, then applied classical multidimensional scaling to visualize the relationships in two dimensions. The compounds separated cleanly into two clusters that corresponded exactly to their experimentally determined modes of action: Type A inhibitors, which compete with waldiomycin for the same site, grouped with the natural product, while Type B inhibitors, including all three new FF compounds, grouped on the opposite side of the map. In other words, the amino acid mapping descriptor, computed without any knowledge of the protein, predicted which compounds share a binding site.

The implications extend beyond this single target class. The study represents the first successful application of amino acid mapping-based screening to identify histidine kinase inhibitors targeting the H-box, and it demonstrates that a descriptor derived from a natural product can be used to mine a library of nearly twelve million compounds for chemically unrelated hits with genuine biochemical activity. Because the method requires no structure of the protein-ligand complex, it could open doors against targets that have resisted traditional structure-based design, a category that includes many membrane-associated and dimeric signaling proteins. The authors caution that future work will be needed to determine whether the descriptor framework can be used prospectively to rationally select inhibitors with specific functional properties. But for a field in which no histidine kinase inhibitor has yet reached clinical trials, the arrival of three potent, structurally novel lead compounds, together with a computational tool that maps their binding modes, is a meaningful step toward antibiotics that disarm bacterial signaling rather than simply poisoning the cell.

Subject of Research: Amino acid mapping-based virtual screening for novel histidine kinase inhibitors targeting the bacterial H-box region

Article Title: Amino acid mapping–guided scaffold selection and hit identification of histidine kinase inhibitors from a virtual library of over 12 million compounds

Article References: Ishikawa, T., Eguchi, Y., Okajima, T., Igarashi, M., Ohira, S., Tsumura, K., Sakurai, K., Ishizaki, Y., Hayashi, C., Akter, S., Nakayama, J., Kohayakawa, C., Tani, A., Sawa, N., Katsumoto, T., Matsumoto, F., Haruta, J.-I., & Utsumi, R. (2026). Amino acid mapping–guided scaffold selection and hit identification of histidine kinase inhibitors from a virtual library of over 12 million compounds. The Journal of Antibiotics. https://doi.org/10.1038/s41429-026-00953-9

Image Credits: AI Generated

DOI: 10.1038/s41429-026-00953-9

Keywords: histidine kinase, amino acid mapping, virtual screening, waldiomycin, two-component system, antibiotic resistance, H-box, scaffold hopping, molecular dynamics, NMR, pyridopyrimidinone, drug discovery