The amyloid-clearing drugs approved for Alzheimer's over the past few years, Leqembi and Kisunla, do something no previous treatment could: they slow the disease. They also carry a side effect serious enough to shadow the whole class, a condition called ARIA, in which the brain swells or bleeds. ProMIS Neurosciences reported interim data this week from an early trial suggesting its experimental drug might deliver the benefit without that harm, and the idea behind it is worth understanding, because if the bet is right it reframes what those side effects were ever about.
The results, released Tuesday, are genuinely encouraging and genuinely preliminary, and both halves of that sentence matter. Taking them for more than they are would be a mistake, and dismissing the underlying hypothesis would be a different one.
What ARIA is, and why it constrains the whole class
Amyloid-related imaging abnormalities, ARIA, is the name for what shows up on brain scans of patients taking amyloid-clearing antibodies: fluid buildup, which is swelling, called ARIA-E, and small bleeds, called ARIA-H. Most cases are mild and without symptoms, caught only because patients are scanned regularly. But some are serious, and rare cases have been fatal, which is why the approved drugs require repeated MRI monitoring and why the risk runs highest in people carrying two copies of the APOE4 gene, a group at elevated risk both for Alzheimer's itself and for these bleeds.
ARIA is not a minor inconvenience. It limits how high the approved drugs can be dosed, it requires an expensive monitoring apparatus, it frightens patients and families, and it makes physicians cautious about who to treat. A drug that cleared amyloid without causing it would not just be safer. It could potentially be dosed higher and given to patients currently considered too risky to treat, which is part of why the target is so sought after.
The hypothesis: the plaque was never the point
Here is the scientific idea that makes ProMIS's approach more than a marginal tweak. The approved antibodies bind to amyloid plaques, the visible clumps that accumulate in the Alzheimer's brain, and there is a growing view that the ARIA side effect comes precisely from that binding, from the antibody attacking plaque that sits on and around blood vessels, disrupting the vessel walls and causing the leaks and swelling.
But a substantial body of research now argues that the plaques may not be the thing driving the disease in the first place. On this view, the real culprits are soluble amyloid-beta oligomers, small, free-floating clusters of the protein that form before they clump into plaques and that appear to be the most toxic form for neurons. The plaques, in this framing, may be closer to a graveyard where the protein ends up than the active agent of harm.
If that is right, it splits the target in two. The approved drugs hit both the plaques and, to varying degrees, the oligomers, and the plaque-binding is where much of the danger lives while contributing less of the benefit. ProMIS designed its antibody, PMN310, to bind only the toxic oligomers and to leave the plaques alone. The bet is that you can get the therapeutic benefit, clearing the harmful soluble species, while skipping the plaque attack that causes ARIA. It is a genuinely different design premise, not a safer version of the same drug, and it rests on the claim that the field spent years treating the debris as the disease.
What the data actually showed, and the caveat that governs it
The interim numbers are the reason for interest. In a blinded interim analysis of 136 patients, there were no cases of ARIA-E, the brain-swelling form, across all genotypes including the highest-risk APOE4 homozygotes, and the total rate of any ARIA was 4.4%, with all cases mild and asymptomatic. Against a class where swelling is the feared complication and where high-risk patients are sometimes excluded from treatment entirely, no ARIA-E at all is a striking early signal.
Now the caveat that governs how much weight the number can bear, and it is the part most easily lost. The analysis is blinded, which means it pools together patients who received the drug and patients who received placebo, and no one analyzing it yet knows which is which. Placebo patients would not be expected to develop ARIA at all, since it is caused by the drug. So a low ARIA rate in a blinded pool that includes placebo recipients is exactly what you would see whether the drug is unusually safe or merely ordinarily effective diluted by placebo patients who were never at risk.
This does not make the result meaningless. Zero cases of ARIA-E even among APOE4 homozygotes is notable, because in that highest-risk group the approved drugs produce swelling often enough that some cases would be expected if the drug behaved like the rest of its class. But blinded, pooled data can only hint. It cannot establish that PMN310 is safer, because the comparison that would prove it, drug-treated patients against placebo patients, is exactly the comparison the blinding hides. The company was careful to frame these as safety and biomarker trends, and the honest reading is that the data are consistent with the hypothesis without confirming it.
Why the efficacy question is the one that matters more
There is a deeper reason to hold the enthusiasm in check, and it concerns what the trial has not yet shown at all. Safety without efficacy is not a drug. A treatment that causes no ARIA but also does not slow the disease helps no one, and this interim look was not designed to demonstrate that PMN310 works.
The whole hypothesis rests on an unproven premise: that clearing oligomers while sparing plaques delivers the cognitive benefit. That is biologically plausible and it is not established in humans. It is entirely possible that plaque-clearing contributes more to the benefit than the oligomer theory assumes, in which case a drug that spares plaque might be safer and less effective, trading one problem for another. The only way to know is efficacy data showing the drug actually slows cognitive decline, and that is not due until the unblinded 12-month topline readout expected in the first quarter of 2027. Until then, the central claim, benefit without the harm, is half-tested. The harm side looks promising in a way the data cannot yet confirm, and the benefit side has not been tested at all.
How to read it
The measured takeaway is that ProMIS has an interesting, scientifically grounded hypothesis and an early result that is consistent with it, which is a real and legitimate reason for attention and not yet a reason for confidence. The oligomer-selective idea is a serious attempt to solve the single biggest problem constraining amyloid therapy, it comes from a coherent reading of the disease biology, and the absence of ARIA-E in high-risk patients is a genuine early signal rather than noise. Fast Track designation from the FDA reflects that the agency sees the unmet need this addresses.
At the same time, the data are blinded and pooled with placebo, which means they can suggest but not demonstrate the safety advantage, and the efficacy question, whether the drug slows the disease at all, has not been answered and will not be until 2027. The bet that the plaque was never the point is elegant, and elegant bets in Alzheimer's have a long history of failing in late trials, because the disease has defeated a great many plausible ideas that looked strong at this stage.
For anyone following Alzheimer's research, the useful frame is that this is a hypothesis worth watching reach its real test, not a result to build hope on yet. The interim data earn PMN310 a place on the list of approaches that could matter. The first-quarter 2027 readout, with unblinded efficacy and a true drug-versus-placebo safety comparison, is when it will become clear whether the idea that the toxic oligomers, not the plaques, were the thing to target all along is a breakthrough or another elegant theory the disease outlasts. The science is promising. The proof is still a year away.
Primary sources
- ProMIS Neurosciences' July 28 press release via Manila Times/GlobeNewswire for the blinded six-month interim analysis of 136 patients, the absence of ARIA-E across all genotypes including APOE4 homozygotes, the favorable safety profile, and the Q1 2027 unblinded 12-month topline readout including efficacy.
- STAT for the reported total ARIA rate of 4.4% with all cases mild and asymptomatic and the framing of a potentially safer treatment.
- ProMIS's July 14 AAIC release for the first human evidence of dose-dependent amyloid-beta oligomer reduction and Dr. Johanne Kaplan's explanation of the oligomer-versus-plaque mechanism and ARIA.
- The PRECISE-AD trial background and BioSpace for PMN310's humanized IgG1 design, the July 2025 FDA Fast Track designation, and the Phase 1a basis for the Phase 1b trial, plus context on the broader amyloid and tau landscape at AAIC 2026.