Version history

{
  "session_id": "sess-gap-pubmed-20260410-150544-e3a2eab9-task-c747c608",
  "round_number": 3,
  "agent_persona": "persona-domain_expert",
  "agent_backend": "scidex.core.llm.complete",
  "action": "assess",
  "content": "The key feasibility filter is the source paper itself. In the February 5, 2026 `Cell` paper, Li et al. report that peripheral cancer/CSPs reduced amyloid in `5xFAD` and `APP/PS1`, but “did not affect tau protein misfolding in the `rTg4510` mice,” which sharply limits any claim of a broad anti-tau effect beyond amyloid-linked contexts. Separately, the March 5, 2026 phase 2 `AL002` TREM2 agonist trial showed CNS target engagement but missed its clinical primary endpoint in early AD, so the translational bar for any TREM2-based program is now much higher. Sources: `Cell` paper abstract/PDF and `Nature Medicine` phase 2 trial.  \nhttps://www.sciencedirect.com/science/article/pii/S0092867425014333  \nhttps://gwern.net/doc/psychiatry/alzheimers/2026-li.pdf  \nhttps://www.nature.com/articles/s41591-026-04273-1  \nhttps://pubmed.ncbi.nlm.nih.gov/31235932/\n\n**What survives**\nOnly three ideas look worth carrying forward, and all should be reframed as `amyloid-context, microglia-mediated adjunct hypotheses`, not broad anti-tau therapies.\n\n1. `Peri-plaque tau seeding restraint via TREM2 microglia`  \nThis is the best tau-facing survivor, but only in mixed amyloid-tau biology, not pure tauopathy. The fit is that TREM2-competent microglia can limit neuritic plaque tau spread around amyloid plaques, consistent with Leyns et al. That matches the paper’s amyloid-first mechanism and the negative `rTg4510` result.  \nDruggability: moderate for a brain-penetrant TREM2 agonist or engineered cystatin-C derivative; poor for wild-type systemic cystatin C as a drug because PK, renal clearance, and BBB delivery are unattractive.  \nBiomarkers/model systems: `Aβ PET`, `tau PET` focused on peri-plaque regions, CSF/plasma `p-tau217`, `p-tau231`, `sTREM2`, `osteopontin`, and microglial PET if available. Use `APP/PS1 x tau-seeding`, `5xFAD + tau inoculation`, or plaque-associated tau models, not `rTg4510` alone.  \nClinical constraints: likely only relevant in very early symptomatic or preclinical amyloid-positive disease. Monotherapy signal may be small.  \nSafety: same class risks as TREM2 agonism generally, including maladaptive microglial activation and uncertain ARIA interaction if combined with anti-amyloid antibodies.  \nRealistic timeline/cost: `3–4 years / $15M–$30M` to get convincing preclinical translational package; `7–10 years / >$150M` to a phase 2 proof-of-concept.\n\n2. `Synaptic protection via microglial/complement normalization`  \nThis is plausible and clinically important, but it is probably secondary to plaque remodeling plus microglial state change, not a distinct cystatin-C magic bullet.  \nDruggability: moderate if pursued through TREM2 pathway modulation; weak if pursued through native cystatin C itself.  \nBiomarkers/model systems: CSF `neurogranin`, `NfL`, synaptic vesicle markers, complement fragments, hippocampal spine density, synaptosome proteomics. Best models are amyloid-bearing mice with early synaptopathy; add `Trem2` KO and complement readouts.  \nClinical constraints: hard to prove mechanism in humans because synaptic biomarkers are noisy and slower-moving than amyloid PD markers.  \nSafety: complement suppression and microglial rewiring can impair host defense or debris clearance if overdone.  \nTimeline/cost: similar preclinical burden, `2–3 years / $8M–$20M` before a clear go/no-go.\n\n3. `Anti-inflammatory/pro-resolution microglial reprogramming`  \nThis is the most druggable pharmacology story, but also the least differentiated clinically because the field already has TREM2 agonists and the first major phase 2 has not shown clinical benefit despite biomarker engagement.  \nDruggability: highest of the survivors if you use a TREM2 agonist antibody or small molecule; lower if you rely on cystatin C replacement.  \nBiomarkers/model systems: CSF `sTREM2`, `osteopontin`, `YKL-40`, `GFAP`, cytokine panels, single-cell brain or CSF immune profiling, plaque-associated microglial signatures.  \nClinical constraints: needs a biomarker-enriched population and probably combination therapy with anti-amyloid, otherwise risk of another PD-positive/efficacy-negative outcome like `AL002`.  \nSafety: immune over- or under-activation, infusion reactions for antibodies, uncertain chronic microglial remodeling risk.  \nTimeline/cost: if piggybacking on an existing TREM2 platform, faster; a de novo program is still `6–9 years` and expensive.\n\n**What I would drop**\n`Hypothesis 2` direct cystatin-C:tau binding and `Hypothesis 5` neuronal `LRP2` protection are not good development bets. They have weak mechanistic support, poor translational leverage, and no help from the new paper’s tau data. `Hypothesis 6` should be demoted from “mechanism” to “expected downstream consequence of amyloid lowering,” and only tested as supportive biology. `Hypothesis 7` is interesting discovery biology but not trial-ready.\n\n**Bottom line**\nThe pathway is feasible as an `amyloid-linked microglial modulation strategy`, not as a broad disease-modifying solution for tau, inflammation, and synapse loss all at once. The best development thesis is: `use TREM2/cystatin-C biology to improve plaque-associated microglial function in early amyloid-positive AD, then test whether that secondarily reduces peri-plaque tau spread and synaptic injury.` Anything stronger than that is ahead of the evidence.\n\nIf useful, I can turn this into a scored table with `go / watch / kill` decisions for each of the 7 hypotheses.",
  "tokens_used": "1360",
  "persona_id": "persona-domain_expert"
}