This is a review of the named 2025 developments and the evidence underlying them, not a current prescribing guide. Treatment availability and appropriate care depend on a person’s clinical assessment and jurisdiction. The purpose here is to show which conclusion each source can actually support.

Trontinemab: amyloid clearance is not the same as a memory outcome

In its 28 July 2025 research update, Roche described results from the Phase Ib/IIa Brainshuttle AD study of trontinemab. The company reported 91% becoming amyloid PET negative in the reported analysis and discussed planned Phase III TRONTIER studies. This was an investigational drug report, not evidence of an approved cure.

An amyloid PET result measures a biological target. It does not mean that 91% of participants recovered memory, avoided dementia or improved by the same percentage on a clinical scale. The original article did not keep that distinction clear enough. The appropriate next question is whether controlled clinical studies show an effect on cognition and daily function, alongside the treatment’s risks.

It is also unsound to conclude from an early study that a drug is safer or cheaper than another treatment without a suitable comparison. Populations, dosing, monitoring and follow-up may differ. The earlier claims of lower costs, fewer side effects than established therapies and possible prevention of dementia are not retained as established findings.

Donanemab: slowing decline in a defined patient population

The FDA’s approval account for Kisunla, or donanemab describes a randomised placebo-controlled study of 1,736 people with confirmed amyloid pathology and mild cognitive impairment or mild dementia. It reports significantly less clinical decline on the integrated Alzheimer’s Disease Rating Scale at 76 weeks. Treatment initiation is described for those early symptomatic stages.

“Less decline” is a comparison between groups over time. It does not imply recovery to an earlier level of memory or the same benefit for every individual. Nor does a result in the studied population establish effectiveness at every stage of the disease. The US approval dates to 2024; it should not be presented as a new 2025 approval.

FDA also identifies amyloid-related imaging abnormalities, including swelling and bleeding, as important risks, with greater risk among people carrying two ApoE ε4 copies. Those risks require attention to the full prescribing information and clinical assessment. This article does not reproduce a monitoring schedule that could become outdated or substitute for individual eligibility decisions.

A blood test assists diagnosis; it does not treat disease

On 16 May 2025, FDA cleared the Lumipulse G pTau217/β-Amyloid 1-42 Plasma Ratio to aid diagnosis in symptomatic adults aged 55 and older. The test measures a ratio associated with amyloid plaques. FDA describes its use alongside other clinical information, rather than as a stand-alone population screening test.

A diagnostic development can matter even when it does not change a disease’s course. Helping establish what is causing symptoms and demonstrating that a therapy slows decline are separate achievements. A positive test is not a treatment response, and an improved diagnostic pathway should not be described as memory restoration.

GL-II-73: the reported memory findings were in mice

The 2025 Neurobiology of Aging paper on GL-II-73 studied the 5xFAD mouse model. It reports working-memory and neuronal effects in that model, including effects despite amyloid accumulation. That is preclinical evidence. It does not demonstrate safety, dosing or memory benefit in people with Alzheimer’s disease.

An animal experiment can identify an avenue worth testing without predicting how much benefit a person would receive. The species, model, measured task and treatment conditions belong in any accurate summary. Removing those qualifications turns an experimental result into a claim the experiment did not test.

Four questions to ask of a “breakthrough” claim

  • Who was studied: cells, animals, healthy volunteers or people at a specified disease stage?
  • What changed: a biomarker, a test classification, a cognitive score or everyday function?
  • Compared with what, and over how long? A percentage needs its endpoint, denominator and comparator.
  • What risks and uncertainties remain? A promising signal is not a completed benefit–risk assessment.

These questions explain why a research pipeline cannot be counted as a list of treatments that already work. Several approaches can be scientifically worthwhile while remaining at very different distances from clinical use. Announcements should keep those distances visible.

Corrections to the previous account

The earlier descriptions of a graphene implant as an Alzheimer’s treatment and of TB006 reversing effects of the disease were not adequately supported by the evidence supplied, so they have been removed. The pipeline totals and speculative prevention claims were also removed rather than repeated without a defined source and date. GL-II-73 is now explicitly identified as a mouse study, and trontinemab’s biomarker result is separated from patient benefit.

The defensible conclusion is narrower than the old “turning point” framing: there were important developments in diagnosis and treatment research, but their endpoints and evidence stages differ. Readers can follow those developments without being promised a cure that these sources do not establish.