AI-Enabled Antibiotic Discovery Identifies Novel Compounds Against Drug-Resistant Gonorrhea
核心洞察
A Wyss Institute-led study published in Science Translational Medicine used deep learning to screen ~6 million compounds, identifying two novel anti-gonococcal candidates with new mechanisms of action.
The lead compound, A1 (搜索), targets alanine racemase (搜索)—a previously unexploited enzyme in N. gonorrhoeae cell wall biosynthesis—representing a novel antibiotic mechanism.
Both compounds demonstrated efficacy in human Vagina Chip models and a mouse vaginal infection model, significantly reducing pathogen burden.
A research team led by the Wyss Institute at Harvard University (搜索), MIT, and the Broad Institute has demonstrated that artificial intelligence can identify entirely new chemical structures capable of combating drug-resistant Neisseria gonorrhoeae (搜索), the pathogen responsible for the second most frequently reported sexually transmitted infection globally. The study, published in Science Translational Medicine, arrives at a critical juncture: two new oral antibiotics—zoliflodacin and gepotidacin—have recently been approved for uncomplicated urogenital gonorrhea, marking the first entirely new antibiotic classes for this infection in over three decades, yet the pathogen's history of rapid resistance development threatens their long-term utility.
"If these two antibiotics get used broadly, it's nearly guaranteed that the pathogen will develop significant resistance against them eventually. We've seen the cycle of resistance development occur within just five to 10 years after first-line roll-out—it has happened over and again," said Melis Anahtar, M.D., Ph.D., Assistant Director of the Clinical Microbiology Laboratory at Massachusetts General Hospital and the study's lead author. "To be able to prevail in this continuous arms race, we will need new antibiotics to fill the pipeline."
Building and Validating the Machine Learning Pipeline
The team began by testing 38,650 small molecules for their ability to inhibit N. gonorrhoeae growth in laboratory assays. This dataset was used to train a predictive deep learning model, which was subsequently validated for its ability to identify drug-like molecules with antibacterial potential and chemical structures distinct from those of conventional antibiotics.
With confidence in the model's discriminatory power, the researchers deployed it to virtually screen a much larger library of approximately 6 million compounds. This computational screen yielded 213 candidates, which were then subjected to a rigorous cascade of growth inhibitory assays, antimicrobial resistance assays, and cell biological evaluations to exclude compounds with undesirable toxicity profiles. From this funnel, two compounds—designated MP20 (搜索) and A1 (搜索)—emerged with promising selectivity for and potent activity against multi-drug resistant N. gonorrhoeae strains, while themselves inducing resistance at very low frequencies.
A Novel Mechanism: Targeting Alanine Racemase (搜索)
Using proteomic approaches, the team identified the molecular target of A1 (搜索), an aminothiazole compound with previously undescribed anti-gonococcal activity. A1 specifically binds to and inhibits alanine racemase (搜索), an enzyme critical for N. gonorrhoeae cell wall biosynthesis.
"We validated the alanine racemase (搜索)-specificity of A1 (搜索) using genetic tools and are now in the process of investigating how exactly A1 inhibits its enzyme activity," Anahtar explained. While several existing antibiotics disrupt cell wall biosynthesis in pathogenic bacteria, specifically targeting alanine racemase with a small molecule represents a novel mechanism of action uncovered by the study.
From Computational Screen to Physiological Models
To assess translational relevance, the team evaluated both compounds in physiologically relevant infection models. Collaborating with the group of Wyss Founding Director Donald Ingber, M.D., Ph.D., which had previously developed a microfluidic Organ Chip model of the human vagina, the researchers demonstrated that MP20 (搜索) significantly reduced pathogen titers after N. gonorrhoeae was introduced into the device and allowed to interact with vaginal epithelial cells.
In a mouse vaginal infection model, intravaginal inoculation of N. gonorrhoeae followed by five treatments with A1 (搜索) over a 24-hour period resulted in significantly lower pathogen concentrations compared to the no-antibiotic control.
The Broader Antimicrobial Resistance Challenge
The study addresses a pressing global health need. Gonorrhea accounts for an estimated 80 million new infections annually worldwide, with over 600,000 cases reported each year in the United States alone. Untreated infections can lead to severe reproductive health complications, including infertility and pelvic inflammatory disease, and increase the risk of HIV transmission. N. gonorrhoeae has sequentially developed resistance to penicillin, tetracyclines, macrolides, fluoroquinolones, and in some regions has shown reduced susceptibility to ceftriaxone, the current backbone of recommended therapy.
"While our observations on A1 (搜索) are promising, it requires further validation and hit-to-lead optimization through medicinal chemistry and other efforts in order to become a clinically relevant antimicrobial drug for treating gonorrhea," Anahtar noted. "However, our deep learning-enabled discovery pipeline has potential for screening much more extensive, ultra-large, make-on-demand chemical libraries to identify unexpected chemical compounds as new starting points in gonorrhea-focused antibiotic development programs."
Senior author James Collins, Ph.D., Termeer Professor of Medical Engineering & Science at MIT and a Core Faculty member at the Wyss Institute, emphasized the convergence of technological advances enabling this work: "We have arrived at an incredibly important point in time in which a vast chemical space has opened up in which billions of chemical compounds with clearly defined structures can be synthesized. This converges with the rapidly evolving capabilities of machine learning that allow us to explore that space with very specific biological activities, such as much-needed new antimicrobial activities, in mind."
Donald Ingber, M.D., Ph.D., co-corresponding author on the study, added: "This study by Jim Collins and his team showcases once again the enormous power of AI combined with high quality biological data sets in the discovery of potentially therapeutic compounds that otherwise would be entirely out of reach. It also shows how, at the Wyss Institute, we seamlessly integrate critical advancements in AI with human-relevant models, in this case a human Vagina Chip."
The research was supported by the Wyss Institute at Harvard University (搜索), the Broad Institute of MIT and Harvard (搜索), the Defense Threat Reduction Agency, the National Institutes of Health, the Siebel Scholars Foundation, the MIT–Novo Nordisk Artificial Intelligence Postdoctoral Fellows Program, the Swiss National Science Foundation, the Knut and Alice Wallenberg Foundation, the Swedish Research Council, and the Bill and Melinda Gates Foundation, as part of the Antibiotics-AI Project led by James Collins.