## Skeptic's Position: The Amyloid Cascade Hypothesis Has Become a Constraint, Not an Explanation
The amyloid cascade hypothesis (ACH) remains *influential*, but influence and explanatory power are not synonyms. I want to challenge whether "best explanation" is even the right frame—it presupposes that a single cascade adequately captures a heterogeneous syndrome.
### Where ACH Has Genuine Evidentiary Support
Genetic evidence is real. *APP* and *PSEN1* mutations causing early-onset autosomal dominant Alzheimer's disease establish that perturbed Aβ metabolism can trigger the full AD phenotype (PMID: **17101890**). Trisomy 21 and APP triplication provides a dose-response relationship that is difficult to dismiss (PMID: **24818095**). Aβ PET positivity precedes symptoms by decades in biomarker studies, satisfying a temporal criterion (PMID: **28762057**).
---
### Where the Hypothesis Has Structural Problems
**1. Clinical trial failures reveal mechanism-target conflation.**
Multiple Phase 3 trials targeting Aβ production, aggregation, or clearance have failed to meet primary endpoints in sporadic AD: semagacestat (secretase inhibition), bapineuzumab, solanezumab, and others. When *Aducanumab* and *Lecanemab* showed slowing of clinical decline, the effect sizes were modest (18–27% slowing on CDR-SB over 18 months), and the disconnect between amyloid clearance and clinical benefit was stark—patients with near-complete plaque removal sometimes declined faster than placebo (PMID: **34545258**). This suggests Aβ is a marker or contributor, not the sole driver.
**2. Aβ burden poorly predicts clinical status.**
Approximately 30–40% of cognitively normal elderly individuals have significant amyloid plaque burden (PMID: **26799073**). Conversely, many patients with clinical AD lack typical amyloid signatures (atypical AD or suspected non-AD pathophysiology). The ACH cannot explain why identical amyloid loads produce variable clinical phenotypes unless invoked as a silent modifier—which empties the hypothesis of falsifiable content.
**3. Tau spreads in a spatially predictable manner that Aβ pathology cannot explain.**
Braak staging of tau pathology follows a stereotypic pattern (transentorhinal → limbic → isocortical) that does not correlate with regional amyloid burden. Neuroinflammation and TDP-43 pathology (LATE-ND) frequently co-occur with AD neuropathology and independently predict cognitive decline (PMID: **31253662**). The cascade appears to be *multifactorial*, with Aβ as one upstream node among several.
---
### Alternative Frameworks Worth Taking Seriously
- **Tau propagation models**: Exogenous tau seeds spread trans-synaptically along connected circuits, explaining the stereotyped progression better than Aβ does (PMID: **28887438**)
- **Metabolic/vascular hypotheses**: Cerebral hypoperfusion and insulin resistance may initiate neuronal vulnerability; APOE4 effects on lipid metabolism and blood-brain barrier function are at least as compelling as its Aβ-binding properties (PMID: **34545258**, **32398658**)
- **Microglial dysfunction**: TREM2 variants and single-cell sequencing reveal disease-associated microglia states that may drive both Aβ deposition and tau spreading (PMID: **31230777**)
---
### Falsification Test the ACH Has Not Passed
A strong causal hypothesis should predict that removing the cause arrests progression. We have now tested this directly with monoclonal antibodies achieving near-complete plaque clearance, yet clinical progression continues. This is not definitive refutation—early intervention timing, downstream irreversible damage, and insufficient tau target engagement remain plausible explanations—but it is strong pressure against "trigger" as the right model.
---
### Revised Confidence
**Original confidence in ACH as best explanation**: ~0.45 (prior to my critique)
**Post-critique confidence**: **0.20–0.25** as *the* explanation; **0.55** as *an important upstream contributor* in a subset of patients
ACH remains the dominant framework for therapeutic development because of path dependency and the investment already made, not because the evidence compels it as the *best* explanation. The field would benefit from a formal shift to a *multifactorial hit* model where Aβ is one of several required cofactors—consistent with evidence that ApoE4, age, vascular risk, and neuroinflammation interact non-additively with Aβ pathology.