Marina Sirota had spent years building computational tools to match diseases with drugs that already exist. In mid-2025, her team at the University of California, San Francisco published the most striking result yet: two cancer medications, given together to mice engineered to develop Alzheimer’s disease, reversed memory loss and reduced the plaques and tangles that define the condition. The study, published in the journal Cell, has since drawn attention from neurologists and drug developers because both compounds already carry years of human safety data from oncology, potentially cutting the timeline to clinical testing by a wide margin.
As of June 2026, no human trial of the combination for Alzheimer’s has been announced. But the research offers a concrete case study in how single-cell brain data and computational screening can surface unexpected therapeutic candidates, and it arrives at a moment when the Alzheimer’s field is hungry for new approaches.
How two cancer drugs landed on the same shortlist
The project began not in a mouse lab but in a database. Sirota, a computational biologist, worked alongside neuroscientist Yadong Huang and researcher Jielin Xu to analyze single-cell transcriptomic data drawn from the brains of people who had died with Alzheimer’s. That data captures gene-activity patterns in individual cell types, including neurons and microglia, the brain’s resident immune cells responsible for clearing toxic debris.
The team fed those disease-specific gene signatures into the Connectivity Map, a Broad Institute database that catalogs how thousands of drugs shift gene expression across more than a million cell profiles. The logic was straightforward: find compounds whose molecular effects are the mirror image of what Alzheimer’s does to brain cells. Out of thousands of candidates, two rose to the top.
Letrozole is an aromatase inhibitor prescribed to postmenopausal women with hormone-receptor-positive breast cancer. Irinotecan, sold under the brand name Camptosar, is a topoisomerase inhibitor used against colorectal cancer. Neither drug was designed with the brain in mind. But computationally, each one corrected a different piece of the Alzheimer’s puzzle.
According to the Cell paper, letrozole primarily normalized the gene-expression distortions found in neurons, while irinotecan addressed those in microglia. Neither drug alone fully reversed the disease signature. Together, the authors reported, they produced a network-level correction that neither achieved on its own. This cell-type-directed approach distinguishes the work from earlier repurposing screens, which typically matched drugs against bulk tissue samples rather than parsing the contributions of individual cell populations.
What happened in the mice
The combination was tested in a transgenic mouse model engineered to develop both amyloid-beta plaques and tau tangles, the two protein pathologies central to Alzheimer’s in humans. Critically, the researchers waited until the animals had already developed established brain pathology before starting treatment, making this a test of reversal rather than prevention.
Treated mice showed restored performance on standard memory and behavioral tests compared with untreated controls. Brain tissue analysis revealed fewer plaques and tangles, less neurodegeneration, and healthier synaptic structures. The behavioral improvements tracked with measurable biological changes, a pairing that strengthens the case that the drugs were acting on disease-relevant pathways rather than producing a superficial cognitive boost.
The UCSF cancer center described the results as a proof of concept for the computational pipeline. Sirota emphasized that the team used algorithmic filters to narrow candidates before committing to expensive animal experiments, a workflow designed to raise the odds that a given compound will show meaningful effects in living organisms.
Why repurposing matters for speed
Developing a new Alzheimer’s drug from scratch typically takes 10 to 15 years and costs well over a billion dollars, with a failure rate above 99 percent in clinical trials, according to pipeline analyses published by the Alzheimer’s Drug Discovery Foundation. Repurposing approved drugs offers a shortcut. Because letrozole and irinotecan already have extensive human safety and pharmacokinetic data, a future trial could potentially use the FDA’s 505(b)(2) regulatory pathway, which allows sponsors to reference existing safety findings rather than repeating early-phase work from zero.
That does not mean a treatment is around the corner. “Faster” in drug development still means years, not months. Any clinical program would need to determine safe doses for elderly patients with dementia, a population very different from the cancer patients for whom these drugs were originally tested. But the gap between a promising mouse study and a first-in-human trial could be meaningfully shorter than it would be for an entirely novel molecule.
The approach also arrives in a changed landscape. The FDA approved lecanemab (Leqembi) in 2023 and donanemab (Kisunla) in 2024, both anti-amyloid antibodies that modestly slow cognitive decline but carry risks of brain swelling and microbleeds. Those drugs validated the idea that targeting Alzheimer’s biology can produce measurable clinical effects, but their modest benefits have left physicians, patients, and payers wanting more. A repurposed oral combination that works through different mechanisms could, in theory, complement or even compete with the antibody therapies, though that possibility remains speculative without human data.
The caveats are substantial
No human data exist for this drug pair in Alzheimer’s patients. That is the single most important caveat, and it colors everything else.
Mouse models of Alzheimer’s have a painful track record. Dozens of drugs that cleared amyloid or restored memory in transgenic mice have gone on to fail in human trials, a pattern so consistent that some researchers have questioned whether current mouse models capture enough of the human disease to be predictive. The UCSF team is aware of this history; the Cell paper frames the work as validation of the computational method, not as a finished therapy.
Irinotecan, at oncology doses, causes severe diarrhea, neutropenia, and other toxicities that would be unacceptable in a frail elderly population. Letrozole can accelerate bone loss and cause joint pain. Whether either drug, let alone both, can be given at doses low enough to be tolerable yet high enough to affect brain pathology is an open question the study does not answer. Dose translation from mice to humans will require careful pharmacokinetic and toxicology work before any trial could begin.
The study’s computational pipeline also raises reproducibility questions. The exact Connectivity Map inversion scores and ranking thresholds that placed letrozole and irinotecan at the top of the candidate list have not been released in a form that outside groups can easily re-run. Independent researchers will need those details to verify whether small changes in modeling parameters would yield different drug pairs.
Mechanistically, the downstream steps between altered gene expression and improved cognition remain only partially mapped. The authors have not yet identified, for example, which specific microglial functions irinotecan restores or how letrozole’s hormonal effects in neurons translate into better synaptic health. A deeper mechanistic picture would help design biomarkers for early human trials and might eventually point to safer, more targeted molecules that hit the same pathways.
What this means for patients and families right now
The practical answer, as of June 2026, is that nothing changes at the pharmacy counter. No clinical trial for this combination in Alzheimer’s patients has been registered on ClinicalTrials.gov. No professional medical society has endorsed off-label use. Oncologic doses of these drugs carry risks that would be inappropriate outside of cancer care, and no physician should prescribe them for dementia based on a mouse study, however promising.
What the research does offer is a validated framework. By linking single-cell disease signatures to drug-perturbation databases, the UCSF team built a pipeline that could be applied to other neurodegenerative diseases or used to refine the Alzheimer’s candidates further. Even if letrozole and irinotecan never advance to the clinic, the same method might surface other compounds with better safety profiles or stronger mechanistic rationale.
For a field that has spent decades chasing single-target therapies with limited success, the idea of computationally designing multi-drug corrections tailored to specific cell types represents a genuine shift in strategy. Whether that shift produces a treatment people can take remains, for now, a question only human trials can answer.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
