mivelsiran

Mivelsiran (ALN-APP) is Alnylam's intrathecally delivered siRNA that silences amyloid precursor protein (APP), the parent molecule from which beta-amyloid is cut. Phase 1 in early-onset Alzheimer's disease showed durable, dose-dependent drops in soluble APP-beta after multi-dose administration [1][2]. A Phase 2 has opened in cerebral amyloid angiopathy (CAA), a related condition where amyloid coats brain blood vessels and causes microbleeds [3]. The bet: choke off amyloid production at the source rather than clearing plaques after they form - a mechanistically distinct play from Leqembi (lecanemab) and Kisunla (donanemab), and one that should not produce ARIA-E (amyloid-related imaging abnormalities with edema - the brain swelling side effect that haunts antibody dosing).

Novel compound, first-in-class for CNS-delivered RNAi. Two trials are running. NCT05231785 is the Phase 1 single- and multiple-ascending-dose study in 60 patients with early-onset Alzheimer's, still recruiting [2]. NCT06393712 is a Phase 2 in 200 patients with sporadic CAA, double-blind, recruiting [3]. No public FDA breakthrough, fast track, or orphan designations have been disclosed as of mid-2026. Alnylam continues to drip out Phase 1 cohort data at AAIC and via 8-K disclosures [4][5]. The next meaningful catalysts are extended-dosing Phase 1 results from the EOAD cohort and early biomarker data from the CAA Phase 2 - both 2026-2027 windows. A Phase 3 program has not yet been declared; Alnylam will want stronger CSF biomarker durability and a safety read across repeat doses before committing.

APP is the protein that, when chopped up by enzymes called secretases, yields the beta-amyloid fragments that aggregate into plaques. Mivelsiran is a small interfering RNA - a short genetic snippet that locks onto APP messenger RNA in neurons and tells the cell to shred it, so APP is never made in the first place. Less APP means less substrate for the bad cleavage, which means less amyloid. The genetic case is strong: people with Down syndrome have three copies of the APP gene (it sits on chromosome 21) and develop Alzheimer's pathology decades early; conversely, the Icelandic A673T variant in APP reduces beta-amyloid production by about 40% and protects carriers from late-life Alzheimer's [6]. That's nature's randomized trial showing lower APP processing equals less disease. The open question is whether knocking down APP after symptoms appear matters as much as having lower APP from birth. Leqembi and Kisunla proved amyloid removal slows decline modestly. Mivelsiran tests whether upstream production blockade - cheaper to maintain, no ARIA-E mechanism because no antibody binds vessel-wall amyloid - does at least as well without the brain swelling problem that haunts antibody dosing.

NCT05231785 (Phase 1, EOAD, n=60): single- and multiple-ascending-dose design. The roughly 70% sustained APP-beta reduction reported at AAIC came from the multi-dose cohorts, lasting six-plus months after the last dose [4][5]. Primary endpoint is adverse-event frequency; key pharmacodynamic readouts are CSF soluble APP-alpha and APP-beta [2]. NCT06393712 (Phase 2, sporadic CAA, n=200): double-blind, placebo-controlled, intrathecal dosing approximately every six months over a roughly two-year treatment period; primary endpoint is the annualized rate of new lobar cerebral microbleeds on MRI [3]. The CAA endpoint is mechanistically clean - CAA is amyloid in vessel walls, microbleeds are the direct consequence - but it is not a regulatorily validated surrogate. FDA has never approved a CAA drug, so endpoint negotiation will matter. Recruitment of 200 sporadic CAA patients is achievable but slow: Boston 2.0 diagnostic criteria require MRI evidence of multiple lobar microbleeds or cortical superficial siderosis, and most community neurologists do not actively work up suspected CAA. Expect enrollment to be the rate-limiter on Phase 2 timelines, not data. Intrathecal dosing - a needle in the spine to put drug directly into cerebrospinal fluid - also limits the addressable population and complicates community uptake even if efficacy lands.

Our model estimates a 5% chance this drug is eventually approved. That starts from the historical approval rate for Phase 2 drugs in this area - about 24% - and then adjusts based on ten specific facts about the trial and its sponsor. The biggest factors pulling the number down are heavier-than-usual blinding, a thin or weak sponsor approval record, weak earlier-phase results, and a randomized design. The remaining factors are close to average for this stage, so they don't move the estimate much in either direction.

Efficacy risk first. The amyloid hypothesis has produced two modestly effective drugs (Leqembi, Kisunla) and a graveyard of failures including aducanumab (Aduhelm, withdrawn from US market by Biogen in January 2024) and gantenerumab (Roche terminated Phase 3 in November 2022 after the GRADUATE trials missed). Even confirmed APP knockdown does not guarantee clinical benefit - the protein may need to be lowered earlier in disease, before plaques and tau pathology take hold. Safety risk is the elephant. APP and its cleavage products have roles in synaptic function and neuronal repair; chronic 70% suppression has no human precedent and could surface late, after multiple doses. Watch for any signal in CSF neurofilament light chain (a blood/CSF marker of neuronal axonal damage) or cognitive batteries. Off-target liver effects of GalNAc-conjugated siRNAs are well-characterized, but mivelsiran uses Alnylam's CNS-specific conjugate, so the toxicity profile is genuinely new territory. Execution risk: the CAA Phase 2 endpoint (microbleed rate) has not been used for approval before; FDA may require a follow-on confirmatory study even with positive results. Commercial risk: intrathecal dosing approximately every six months requires a neurologist and a procedure room. Current anti-amyloid antibodies sit at roughly $16,000-$32,000/year (Leqembi WAC was cut from $26,500 to $16,000/year in early 2024; Kisunla launched at ~$32,000/year), and both infuse in an outpatient chair - payers will demand head-to-head benefit over those cheaper, simpler-to-deliver options. The commercial bar is actually higher at the lower antibody price point because intrathecal administration adds procedural cost and access friction. Without a cognitive endpoint win, this stays a small CAA franchise.

Worth watching, mostly for the platform read. The multi-dose pharmacodynamic signal is genuinely strong - sustained 70% soluble APP-beta knockdown from intrathecal dosing is the kind of CNS RNAi result that has eluded the field for fifteen years [4]. The specific signal that would matter: CSF Aβ42/40 ratio normalization plus a cognitive trend in the multi-dose EOAD cohort, expected 2026-2027. Check back at AAIC 2026 (July) and at Alnylam's R&D days. Alnylam closed 2025 with strong cash from Amvuttra product revenue (which has displaced Onpattro and is now the company's lead asset) and the Leqvio royalty stream from Novartis, giving it the balance sheet to run mivelsiran for years [5]. If mivelsiran works, the bigger story is that CNS siRNA is real, which would reprice every Alnylam neurology preclinical/clinical program - notably ALN-HTT02 (huntingtin, Huntington's disease, Phase 1), ALN-SOD1 (SOD1-ALS, IND-stage), and CNS targets disclosed at R&D days including tau and ataxin programs. If it fails on safety, the platform takes the hit, not just the program. Sporadic CAA itself is a niche indication - symptomatic, Boston 2.0-diagnosable prevalence is estimated in the low hundreds of thousands in the US, of which only a fraction would be MRI-screened and treatment-eligible - so the CAA franchise alone does not justify the platform; the EOAD/late-onset AD read-through is the bigger prize. Mivelsiran specifically is a long-dated bet on a clean mechanism with an unproven regulatory path; the read-through to Alnylam's neurology pipeline is the more tradeable thesis.