A tau-targeting Alzheimer's drug failed to slow decline in a phase 2 trial

For a decade, the dominant Alzheimer's drug strategy has aimed at amyloid. A parallel line of work targets tau, the protein that forms neurofibrillary tangles and tracks more closely with cognitive decline. One way to slow tangle formation, the theory runs, is to inhibit an enzyme called O-linked N-acetylglucosaminidase (OGA), which strips a protective sugar from tau [s1].

Ceperognastat is a potent oral OGA inhibitor built to test exactly that idea. A phase 2 trial published in JAMA reports that it did not work [s1].

What the trial did

The study was double-blind, randomised, and placebo-controlled, run at 72 sites in five countries from September 2021 to August 2024, with an observational extension through May 2025 [s1]. It randomised participants with early symptomatic Alzheimer's disease and biomarker evidence of tau pathology 1:1:1 to once-daily oral ceperognastat at 0.75 mg (n=110) or 3 mg (n=108), or placebo (n=108) [s1].

The trial was funded by Eli Lilly, which developed the drug (registry code LY3372689), and the registration lists Lilly as the sole sponsor [s1][s2]. That funding matters when reading any trial, but here the sponsor's own study reports a failure — the kind of result a manufacturer has no incentive to overstate.

The primary analysis was restricted to the 259 participants with low-to-medium baseline tau levels, measured by flortaucipir PET, on the reasoning that a tau-lowering drug has the best chance early in the disease [s1]. The primary outcome was change in the Integrated Alzheimer's Disease Rating Scale (iADRS). The prespecified success criterion was unusual: a 60% or greater probability of achieving at least 25% slower progression than placebo, estimated from a Bayesian disease-progression model [s1].

What it found

The drug missed. At 100 weeks, the posterior mean change in iADRS was −8.39 with ceperognastat 0.75 mg, −13.27 with 3 mg, and −10.07 with placebo — meaning the higher dose declined more than placebo, not less [s1].

The model's disease-progression ratio told the same story. Relative to placebo it was 0.84 (95% credible interval, 0.66 to 1.04) for the 0.75 mg group and 1.32 (1.10 to 1.58) for the 3 mg group, corresponding to 16% less progression and 32% greater progression respectively [s1]. None of the six secondary clinical endpoints showed a benefit either [s1].

Two biomarker signals ran in the drug's favour and are worth stating plainly rather than burying. On tau PET, only the 3 mg group showed a statistically significant smaller increase than placebo, and only in the lateral temporal lobe, at 76 weeks (P=.04) [s1]. On MRI, both doses showed less whole-brain volume loss than placebo — 43.2% less at 0.75 mg and 49.5% less at 3 mg, both P<.001 [s1].

That combination — a possible effect on brain-imaging measures with no clinical benefit, and worse cognition at the higher dose — is the puzzle the trial leaves behind.

Safety

The 3 mg dose was also the least well tolerated. Serious treatment-emergent adverse events occurred in 13 participants (12.0%) on 0.75 mg, 29 (26.4%) on 3 mg, and 17 (15.7%) on placebo; severe events followed the same pattern, in 7 (6.5%), 15 (13.6%), and 7 (6.5%) respectively [s1].

Why the imaging-versus-cognition split matters

A drug can move a biomarker without helping a patient. Less brain-volume loss on an MRI is not the same as preserved memory, and the trial is a clean demonstration of the gap: the imaging measures shifted while the cognitive score, the thing that matters to a person, did not — and moved the wrong way at the higher dose [s1]. The accompanying editorial by Rabinovici and Grill frames the study as a negative trial with lessons, not a near miss [s3].

The limits, stated plainly

This was a phase 2 study of 327 randomised participants, with the primary endpoint confined to a 259-person subgroup selected by tau level [s1]. It is a single trial of a single molecule at two doses, and a negative phase 2 result does not by itself indict the entire tau-lowering hypothesis — it indicts this drug, at these doses, over this window. The Bayesian success threshold was prespecified, but a probabilistic criterion is harder for readers to interpret than a simple between-group difference [s1].

What to watch

Whether other tau-directed approaches — different targets, antibodies against tau itself, or antisense drugs — fare differently. On this compound, the registry lists the study as completed [s2]. The signal that the higher dose was associated with faster cognitive decline is the finding that should temper enthusiasm for pushing OGA inhibition harder.

This article describes trial results, including doses and adverse events, for informational purposes only. It is not medical advice and not a recommendation about any treatment.

Sources

• [s1] Fleisher AS, Munsie L, Mancini M, et al. Ceperognastat in Early Symptomatic Alzheimer Disease: A Randomized Clinical Trial. JAMA, published online 2026 (created 2026-07-13).
• [s2] ClinicalTrials.gov. A Study of LY3372689 in Participants With Early Symptomatic Alzheimer's Disease. NCT05063539, last update posted 2026-06-12.
• [s3] Rabinovici GD, Grill JD. Ceperognastat in Alzheimer Disease: Lessons From a Negative Clinical Trial. JAMA, published online 2026 (created 2026-07-13).