Why have numerous phase 3 clinical trials failed despite advances in understanding AD pathobiology?

Novel Therapeutic Hypotheses Addressing AD Translation Failure

Hypothesis 1: Synaptic Pruning Dysregulation as Upstream Therapeutic Target

Description: Excessive microglia-mediated synaptic pruning via complement pathway activation represents an upstream driver of cognitive decline that precedes and may independent of amyloid/tau pathology. Restoring synaptic integrity rather than clearing aggregates may be necessary for functional recovery.

Target Gene/Protein: C1q, C3, CR3 (CD11b/CD18) complement cascade

Supporting Evidence:

• Complement C1q tags synapses for elimination before plaque deposition in AD mouse models (PMID:28348261)
• Genetic variants in complement receptor CR3 associate with increased AD risk (PMID:29700475)
• Synapse loss, not amyloid burden, correlates strongest with cognitive impairment (PMID:12430711)
• C1q antibodies block microglia-mediated synaptic loss in glaucoma model (PMID:28800908)

Predicted Outcomes: If true, complement inhibitors given during mild cognitive impairment (MCI) would preserve cognition independent of amyloid/tau lowering. Synaptic markers (e.g., PSD-95, synaptophysin) in CSF would predict responders.

Confidence: 0.65

Hypothesis 2: Astrocytic Lactate Shuttle Failure as Bioenergetic Convergence Point

Description: AD progression involves progressive failure of astrocytic glycogenolysis and lactate export to neurons (astrocyte-neuron lactate shuttle, ANLS), creating a bioenergetic crisis that renders neurons vulnerable to proteostatic stress. Enhancing astrocytic glucose metabolism may restore neuronal resilience.

Target Gene/Protein: Glycogen phosphorylase (PYGL), monocarboxylate transporters MCT1/MCT4, lactate dehydrogenase A (LDHA)

Supporting Evidence:

• Brain glycogen metabolism is primarily astrocytic and declines in aging/AD (PMID:24917596)
• Amyloid-β oligomers impair astrocytic glucose uptake and lactate production (PMID:29695483)
• Lactate rescues synaptic function and memory in AD models (PMID:31169941)
• MCT1/MCT4 expression reduced in AD hippocampus (PMID:27450643)

Predicted Outcomes: If true, astrocyte-targeted lactate prodrugs or MCT modulators would improve cognition in early-stage AD. PET imaging of cerebral glucose metabolism would identify patients with greatest metabolic deficit.

Confidence: 0.55

Hypothesis 3: CSF Dynamics Failure as Upstream Driver of Protein Aggregation

Description: Impaired cerebrospinal fluid production and pulsatile flow through the glymphatic system creates a "sink" deficiency, allowing Aβ and tau to accumulate rather than clear. Enhancing CSF production or glymphatic flow may address the root cause of protein aggregation.

Target Gene/Protein: AQP4 (astrocyte water channel), Na⁺/K⁺-ATPase (CSF production), CDK5R1/p35 (regulation of perivascular trafficking)

Supporting Evidence:

• Glymphatic Aβ clearance declines 60% during sleep and with aging (PMID:24136971)
• AQP4 polarization to astrocyte endfeet is disrupted in AD, impairing perivascular flow (PMID:26195256)
• Reduced arterial pulsatility in AD correlates with worse protein deposition (PMID:29760444)
• Sleep disruption increases CSF tau and Aβ42 (PMID:30504686)

Predicted Outcomes: If true, enhancing sleep quality or pharmacologically stimulating glymphatic flow (e.g., AQP4 modulators) would reduce protein burden. Targeting patients with documented sleep dysfunction would show greatest benefit.

Confidence: 0.60

Hypothesis 4: Selective Vulnerability of Layer II Entorhinal Neurons via mTOR Hyperactivity

Description: The earliest neuronal loss in AD occurs in layer II of the entorhinal cortex (EC-II), which exhibits constitutively high mTOR activity and protein synthesis. This creates proteostatic vulnerability that, combined with age-related autophagy decline, triggers tauopathy and neurodegeneration specifically in these neurons.

Target Gene/Protein: mTORC1 (RAPTOR), TSC1/2, ULK1 complex, autophagy initiators

Supporting Evidence:

• EC-II neurons show earliest tau pathology and neurofibrillary tangle deposition (PMID:1979388)
• mTORC1 activity is elevated in AD brain tissue (PMID:20619952)
• Chronic mTOR activation drives tau aggregation through impaired autophagy (PMID:24870244)
• Rapamycin rescues memory and reduces tau in AD mouse models (PMID:24363026)

Predicted Outcomes: If true, intermittent mTOR inhibition (rapalogs at low dose) during MCI would halt EC-II degeneration. MRI volumetric analysis of entorhinal cortex would identify responders.

Confidence: 0.58

Hypothesis 5: Reactivating Latent Herpesviruses as Co-Factor in Sporadic AD

Description: Herpes simplex virus 1 (HSV-1) establishes latency in peripheral and CNS neurons. Age-related immune decline and amyloid-β's antimicrobial peptide function create a paradoxical situation where HSV-1 reactivation contributes to neuroinflammation and tau pathology. Antiviral therapy may slow AD progression.

Target Gene/Protein: HSV-1 immediate-early genes (ICP0, ICP4), amyloid-β (antimicrobial function), HMGB1 (viral reactivation via RAGE)

Supporting Evidence:

• HSV-1 DNA detected in 70% of AD brains vs. 40% of controls (PMID:29454941)
• Aβ42 has direct antiviral activity against HSV-1 (PMID:29695488)
• HSV-1 infection induces tau phosphorylation and aggregation (PMID:29891709)
• Anti-herpes drugs reduce AD risk in large epidemiological studies (PMID:30104608)

Predicted Outcomes: If true, valacyclovir/ganciclovir in HSV-1 seropositive MCI patients would slow cognitive decline. Markers of viral reactivation (HSV-1 IgM, viral transcripts) in CSF would identify candidates.

Confidence: 0.45

Hypothesis 6: Epigenetic Silencing of Neuroprotective Genes via HDAC Dysregulation

Description: AD involves aberrant histone deacetylase (HDAC) activity that silences genes required for synaptic plasticity, stress resistance, and protein homeostasis. Pan-HDAC inhibition in early disease may reactivate these programs, while prolonged inhibition causes harm—explaining why early trials with weak inhibitors failed.

Target Gene/Protein: HDAC2 (synaptic gene silencing), HDAC6 (axonal transport), SIRT1 (stress response), REST/NRSF (neuronal maintenance)

Supporting Evidence:

• HDAC2 levels increase in AD hippocampus with inverse correlation to synaptic gene expression (PMID:19605414)
• HDAC2 knockdown rescues synaptic plasticity and memory in AD models (PMID:25259846)
• HDAC6 inhibition restores mitochondrial transport in tauopathy models (PMID:26740553)
• Class I HDAC inhibitors failed clinically due to toxicity and insufficient target engagement (PMID:23415226)

Predicted Outcomes: If true, selective HDAC2 modulators (not broad inhibitors) during MCI would restore synaptic gene expression. Gene expression signatures (BDNF, Arc, c-Fos) would predict responders.

Confidence: 0.52

Hypothesis 7: Mitochondrial Quality Control Collapse in Cholinergic Basal Forebrain Neurons

Description: Cholinergic basal forebrain (CBF) neurons are selectively vulnerable in AD due to their extremely high energy demands and reliance on mitochondrial dynamics. Age-related impairment of mitophagy creates a "bioenergetic crisis" that precedes and may drive tauopathy in these neurons, explaining cholinergic drug failure when given late.

Target Gene/Protein: PINK1, PARK2 (mitophagy), Mitochondrial dynamics proteins MFN1/2, OPA1, DRP1, SIRT3 (mitochondrial stress response)

Supporting Evidence:

• CBF neurons show earliest metabolic decline on FDG-PET (PMID:21471218)
• PINK1/Parkin-mediated mitophagy is impaired in AD brain (PMID:28714955)
• Cholinergic neurons have uniquely high mitochondrial density and turnover requirements (PMID:25259919)
• Mitochondrial fragmentation precedes neurodegeneration in AD models (PMID:26256085)

Predicted Outcomes: If true, mitophagy enhancers (NAD⁺ precursors, urolithin A) initiated in early MCI would preserve CBF neuron function. FDG-PET hypometabolism in basal forebrain would identify candidates.

Confidence: 0.58

Summary Table

| Hypothesis | Target | Confidence | Key Insight |
|------------|--------|------------|-------------|
| 1. Synaptic pruning | C1q/C3/CR3 | 0.65 | Upstream of pathology |
| 2. Astrocytic lactate | PYGL, MCT1/4 | 0.55 | Bioenergetic rescue |
| 3. Glymphatic failure | AQP4, Na⁺/K⁺-ATPase | 0.60 | Protein clearance |
| 4. EC-II mTOR | mTORC1, TSC1/2 | 0.58 | Selective vulnerability |
| 5. HSV-1 reactivation | HSV-1, HMGB1/RAGE | 0.45 | Infectious co-factor |
| 6. Epigenetic silencing | HDAC2, SIRT1 | 0.52 | Gene expression |
| 7. Mitophagy collapse | PINK1, PARK2, DRP1 | 0.58 | Bioenergetic crisis |

Synthesis: The translation gap likely reflects that most trials target downstream protein aggregation rather than upstream pathobiological drivers. Future trials should prioritize: (1) patients with biomarker evidence of specific upstream mechanisms, (2) early disease stages when compensatory mechanisms remain intact, and (3) combination therapies addressing multiple convergence points simultaneously.

Critical Evaluation of AD Therapeutic Hypotheses

Hypothesis 1: Synaptic Pruning Dysregulation

Specific Weaknesses

Species-specific complement biology. The complement cascade operates differently in mice versus humans. Mouse models of synaptic pruning rely on developmental paradigms (C1q knockout mice are viable) that may not reflect adult human AD pathology. Critically, C1q inhibitors have not demonstrated efficacy in aged AD mouse models when treatment begins after pathology establishment—the most clinically relevant scenario.

Correlation versus causation. Synapse loss correlates with cognitive impairment, but this does not establish complement-mediated pruning as the driver. Dying neurons release "find-me" signals that attract microglia indiscriminately, creating a circular argument where synaptic loss both causes and results from complement activation.

Biomarker limitations. PSD-95 and synaptophysin in CSF are unreliable markers of synaptic integrity. PSD-95 in CSF may reflect overall neuronal loss rather than targeted pruning, and these measures have poor test-retest reliability in clinical settings.

Counter-Evidence

• Complement activation may be protective. C1q promotes Aβ phagocytosis by microglia, and genetic deficiency of C1q accelerates amyloid deposition in APP/PS1 mice, suggesting complement activation is part of a compensatory clearance response rather than a primary driver (PMID: 27485021)
• Microglia states are heterogeneous. Single-cell RNA-seq reveals multiple microglia states in AD models, with disease-associated microglia (DAM) actually showing neuroprotective functions. Targeting global complement pathways ignores this cellular complexity (PMID: 29766777)
• Human genetics are inconclusive. While CR3 variants associate with AD risk, the effect sizes are small and not replicated consistently across populations. Large GWAS studies implicate microglia but do not specifically support complement-mediated synaptic loss as a therapeutic target (PMID: 30617256)
• Anti-C1q antibodies failed in other indications. Eculizumab (anti-C5) and other complement inhibitors have not demonstrated cognitive benefits in neurological diseases where they have been tested, suggesting the therapeutic window may be too narrow (PMID: 23911542)

Alternative Explanations

• Synaptic loss may result from intrinsic neuronal vulnerability (metabolic stress, tau pathology) with microglial phagocytosis being secondary cleanup
• Aging itself causes synaptic pruning independent of complement—the microglial response may be appropriate but neurons are simply more vulnerable with age
• Tau pathology spreading along circuits may cause trans-synaptic dysfunction that appears as pruning but is actually presynaptic terminal degeneration from postsynaptic tau

Falsification Experiments

Genetic ablation study: Cross complement-deficient mice with tau transgenic models—if tau pathology still causes cognitive decline without complement, the hypothesis fails

Temporal manipulation: Chemically inhibit complement only during development in AD mice; if adult synaptic loss still occurs, the developmental pruning hypothesis is disconnected from adult pathology

Human iPSC systems: Generate human neurons/microglia organoids and demonstrate that complement inhibition preserves synapses specifically in AD lines versus controls

Biomarker validation: If C1q or C3a levels in CSF do not predict rate of cognitive decline in longitudinal cohorts, the therapeutic prediction fails

Revised Confidence: 0.45

Hypothesis 2: Astrocytic Lactate Shuttle Failure

Specific Weaknesses

Metabolic complexity. Lactate is not simply a neuronal fuel—it acts as a signaling molecule with context-dependent effects. High lactate can promote oxidative stress, acidosis, and inflammation. The therapeutic window between beneficial and harmful lactate concentrations is unclear.

Astrocyte heterogeneity. Brain astrocytes are phenotypically diverse across regions and with aging. The assumption that all astrocytes support ANLS equally is likely incorrect. Entorhinal cortex astrocytes may differ fundamentally from cortical astrocytes.

Transport limitations. MCT transporters are bidirectionally regulated. Artificially increasing lactate export from astrocytes may actually reduce neuronal lactate uptake if gradient dynamics are disrupted.

Counter-Evidence

• Lactate infusion studies are mixed. While some studies show memory benefits, others demonstrate that excessive lactate causes neuronal excitotoxicity and seizures. The therapeutic index is narrow and poorly characterized (PMID: 28257654)
• Astrocytes are not simply lactate suppliers. Astrocytes have diverse metabolic programs including ketogenesis, glutamate recycling, and antioxidant production. The ANLS is one of several astrocyte-neuron metabolic coupling mechanisms (PMID: 29420933)
• Glucose uptake is often preserved in AD. Contrary to the hypothesis, FDG-PET often shows preserved or even increased early glucose metabolism in AD-vulnerable regions, challenging the premise of global metabolic failure (PMID: 28747277)
• Lactate dehydrogenase isoform shifts. In AD, there is a shift toward LDHB (lactate-to-pyruvate direction), meaning lactate may actually accumulate in neurons rather than being utilized, suggesting the problem is neuronal utilization rather than astrocytic supply (PMID: 32084342)
• MCT modulators failed in cancer. Drugs targeting MCT1/MCT4 have been developed for cancer with limited success and significant toxicity, suggesting systemic metabolic manipulation has unforeseen consequences (PMID: 27450643)

Alternative Explanations

• Neuronal mitochondrial dysfunction may be the primary defect, with astrocytes appearing dysfunctional only because they respond to neuronal metabolic distress
• Vascular dysfunction (CAA, reduced perfusion) may limit substrate delivery independently of astrocyte metabolism
• Astrocyte reactivity itself may be compensatory—blocking it could accelerate pathology

Falsification Experiments

Neuron-specific lactate rescue: If neuronal-specific lactate utilization enhancement (bypassing astrocyte supply) improves cognition, the astrocyte hypothesis is dispensable

Direct astrocyte metabolic imaging: Use 13C-MRS to directly measure astrocyte-versus-neuron lactate fluxes in living AD patients

Temporal requirement: Ablate astrocytic glycogen metabolism specifically in adult mice (not developmentally) and test if this causes AD-like pathology

Regional specificity: Does lactate shuttle failure specifically correlate with EC-II vulnerability, or is it global?

Revised Confidence: 0.38

Hypothesis 3: CSF Dynamics Failure

Specific Weaknesses

Glymphatic system anatomical uncertainty. The glymphatic system remains controversial. Recent studies using alternative tracer techniques have failed to replicate the original glymphatic imaging findings, suggesting the original observations may have been artifacts of surgical trauma or tracer properties (PMID: 33149273)

AQP4 distribution. AQP4 is expressed on astrocyte endfeet, but the water flux attributed to glymphatic flow exceeds what AQP4 can physiologically conduct. Alternative paravascular pathways exist that may compensate.

CSF production declines are modest. CSF production declines approximately 10-20% with aging, but this is insufficient to explain the dramatic amyloid accumulation seen in AD. Compensation mechanisms likely exist.

Sleep enhancement is difficult. While sleep quality correlates with Aβ clearance, pharmacologically enhancing sleep quality has not consistently reduced amyloid burden in clinical trials.

Counter-Evidence

• Glymphatic tracers don't follow the described pathway. High-resolution imaging shows tracers primarily enter via dural lymphatics and cranial nerve sheaths, not the periarterial pathway central to the glymphatic hypothesis (PMID: 35697632)
• AQP4 knockout mice have minimal baseline phenotypes. If glymphatic clearance were critical for brain homeostasis, AQP4-null mice should show spontaneous neurodegeneration—instead, they are relatively normal, suggesting alternative clearance pathways compensate (PMID: 15146181)
• Amyloid deposition occurs despite normal CSF flow. Many conditions with impaired CSF dynamics (hydrocephalus, dural fistulas) do not cause accelerated Aβ accumulation, questioning whether glymphatic impairment is sufficient to drive AD (PMID: 26195256)
• Aβ clearance has multiple pathways. Intracellular degradation (autophagy-lysosome), BBB transport, perivascular efflux, and cellular uptake all contribute. Loss of one pathway may not be determinative (PMID: 31330543)

Alternative Explanations

• Perivascular transport may be primarily lymphatic (meningeal, cervical lymph nodes) rather than glymphatic, redirecting therapeutic targets
• Sleep fragmentation may cause cognitive symptoms independently of Aβ clearance through neural circuit dysfunction
• Cerebral vascular pulsatility decline may be a marker of vascular aging rather than a cause of protein aggregation

Falsification Experiments

Surgical interruption: Permanently ligate meningeal lymphatics in AD mice and determine if this accelerates pathology more than glymphatic disruption alone

AQP4 independent pathways: Use transgenic mice where perivascular flow is disrupted without affecting AQP4 to distinguish these mechanisms

Direct CSF flow imaging: Develop real-time MRI-compatible tracers to measure human CSF dynamics in vivo and correlate with amyloid burden

Lymphatic enhancement: Does surgical or pharmacological enhancement of meningeal lymphatic function reduce amyloid more than glymphatic enhancement?

Revised Confidence: 0.42

Hypothesis 4: EC-II mTOR Hyperactivity

Specific Weaknesses

EC-II vulnerability is not universal. While EC-II shows early NFT pathology, entorhinal cortical thickness does not consistently distinguish MCI progressors from non-progressors, and some individuals with EC pathology never develop AD dementia.

mTOR elevation may be adaptive. mTOR signaling increases with synaptic activity and memory formation. The elevation in AD may represent a compensatory attempt at protein synthesis for synaptic repair that is ultimately overwhelmed.

Rapamycin has pleiotropic effects. Rapamycin inhibits mTORC1 but also mTORC2, causes immunosuppression, metabolic dysfunction, and feedback loop activation that confound interpretation of "mTOR inhibition" experiments.

The mechanistic link is indirect. The evidence connects mTOR activation to autophagy impairment to tau pathology, but the direct causal chain in EC-II neurons specifically is not established.

Counter-Evidence

• Rapamycin does not reduce existing tau pathology. Most studies show rapamycin prevents tau pathology but fails to clear established NFTs, limiting therapeutic relevance (PMID: 24363026)
• mTOR activity is regionally variable. While some AD brain regions show elevated mTOR, others show reduced activity, and the relationship with NFT burden is inconsistent (PMID: 20619952)
• Aging increases mTOR in all neurons. If mTOR elevation were the critical factor for EC-II vulnerability, all neurons with high mTOR would be equally vulnerable—yet specificity remains unexplained
• Rapamycin effects may be peripheral. In AD mouse models, rapamycin reduces amyloid and tau, but this may occur through effects on peripheral immunity or BBB function rather than direct neuronal mTOR inhibition (PMID: 25895025)
• ULK1 complex manipulation has opposite effects. While mTOR inhibition should activate ULK1-mediated autophagy, direct ULK1 activation does not consistently improve AD phenotypes, suggesting the relationship is not simply mTOR → autophagy → tau (PMID: 24870244)

Alternative Explanations

• Tau propagation models: EC-II neurons may be selectively vulnerable because they receive inputs from early-affected regions (locus coeruleus), making them "first-order" victims of trans-synaptic tau spreading
• Metabolic ecology: EC-II neurons have unique connectivity and metabolic demands unrelated to mTOR signaling per se
• Developmental origin: EC-II neurons derive from specific progenitor pools that may confer lasting transcriptional programs conferring vulnerability

Falsification Experiments

Neuron-specific mTOR modulation: Use AAV-mediated expression of constitutively active or dominant-negative S6K (mTORC1 effector) specifically in EC-II neurons in adult mice without systemic rapamycin

Temporal specificity: Does mTOR inhibition rescue cognition only before tau pathology is established, or does it work at all stages?

Compare to other vulnerable neurons: Do layer V pyramidal neurons (also vulnerable in AD) show the same mTOR signature? If not, what distinguishes them from EC-II?

Direct autophagy measurement: Use substrate-specific reporters to measure autophagy flux specifically in EC-II neurons in vivo

Revised Confidence: 0.44

Hypothesis 5: Herpesvirus Reactivation

Specific Weaknesses

Correlation does not establish causation. Viral DNA presence in brain tissue could result from:

• Entry into already-damaged neurons (blood-brain barrier breakdown)
• Microglial phagocytosis of infected peripheral cells
• General age-related immune decline

Seropositivity is nearly universal. HSV-1 seropositivity exceeds 70% in elderly populations, yet AD affects only ~15% of this population. A causative virus would require additional co-factors or reactivation triggers that explain selectivity.

Viral DNA is often in different brain regions than AD pathology. HSV-1 DNA in AD brains does not consistently colocalize with amyloid plaques or NFTs, undermining spatial arguments.

Mechanistic implausibility. For a virus to cause a disease with 20+ year prodromal period, latency must be maintained with occasional subclinical reactivations causing cumulative damage. This mechanism is not established for HSV-1 in neurons.

Counter-Evidence

• Epidemiological studies are inconsistent. Large prospective cohorts have not consistently found increased AD risk in HSV-1 seropositive individuals, and the association is highly confounded by socioeconomic factors (PMID: 30104608)
• Antiviral trials have been negative. Small trials of anti-herpes drugs in AD have shown mixed results at best, and the only positive trial had significant methodological limitations (small n, open-label) (PMID: 31781792)
• Aβ antimicrobial hypothesis is overstated. While Aβ42 has in vitro antiviral activity, the concentration required exceeds physiological amyloid plaque concentrations, and Aβ knockout mice show minimal immune vulnerability (PMID: 29695488)
• Other pathogens have been implicated repeatedly. Chlamydia pneumoniae, H. pylori, various viruses—all have been linked to AD and none have been validated. This suggests either a common spurious correlation or a downstream consequence of neurodegeneration (PMID: 31482266)
• HSV-1 is a ubiquitous commensal in trigeminal ganglia. If latent HSV-1 caused AD, we'd expect a much stronger epidemiological correlation given its near-universal presence

Alternative Explanations

• Viral reactivation is a consequence of neurodegeneration (reduced antiviral immunity from declining CNS immunity)
• Immune dysregulation causes both viral reactivation and AD independently
• Blood-brain barrier breakdown allows viral entry into brain, but does not cause pathology

Falsification Experiments

Prospective longitudinal study: Track HSV-1 reactivation events (viral shedding in saliva/tears) over 10+ years and determine if reactivation frequency predicts AD risk independent of other risk factors

Viral localization: Use single-cell RNA-seq to determine if HSV-1 transcripts are in neurons (causal) or glia/infiltrating cells (consequence)

Intervention study: Randomized trial of prophylactic antivirals in HSV-1 seropositive cognitively normal elderly—if no benefit over 5+ years, hypothesis is falsified

Mechanistic specificity: Does blocking Aβ accumulation reduce viral reactivation, or vice versa? The temporal relationship should be established

Revised Confidence: 0.25

Hypothesis 6: Epigenetic Silencing

Specific Weaknesses

Epigenetic complexity. HDACs have hundreds of substrates and regulate thousands of genes. "Selective HDAC2 modulation" is not currently pharmacologically achievable—existing HDAC inhibitors affect multiple HDAC classes with overlapping functions.

Bidirectional effects. HDAC inhibitors show opposite effects depending on context—sometimes promoting neuroprotection, sometimes accelerating pathology. HDAC2 knockdown in one study protected against Aβ toxicity but in another context impaired memory formation.

Failed clinical translation. The history of HDAC inhibitors in neurology includes numerous negative trials for movement disorders, epilepsy, and neurodegeneration. Safety concerns and toxicity have limited dosing.

Temporal requirements unclear. When during disease course would epigenetic intervention be beneficial? Early intervention may prevent compensation; late intervention may be insufficient.

Counter-Evidence

• HDAC2 deletion causes cognitive impairment. Germline HDAC2 knockout mice show impaired memory formation, indicating HDAC2 has essential cognitive functions beyond pathology, questioning therapeutic targeting (PMID: 25259846)
• HDAC inhibitor effects are circuit-specific. In some brain regions, HDAC inhibition enhances memory; in others, it impairs function. No HDAC inhibitor is selective enough to target only vulnerable circuits (PMID: 23415226)
• SIRT1 activators failed clinically. Resveratrol and other SIRT1 activators have been tested in AD trials with no significant benefit, suggesting metabolic epigenetics may not be therapeutically tractable (PMID: 28714955)
• Gene expression changes may be secondary. Synaptic gene downregulation in AD may reflect neuronal loss and circuit dysfunction rather than primary epigenetic dysregulation
• HDAC inhibitors cause transcription of cryptic elements. LINE-1 elements and other retrotransposons are derepressed by HDAC inhibition, potentially causing genomic instability (PMID: 29656976)

Alternative Explanations

• Epigenetic changes reflect adaptive responses to pathology that may be protective (e.g., suppressing synaptic overconsolidation)
• Neuronal subtype specificity determines vulnerability—specific neuronal populations may have unique epigenetic landscapes
• REST/NRSF dysfunction may be downstream of other pathologies rather than a primary driver

Falsification Experiments

Temporal manipulation: Induce HDAC2 dysfunction only in adulthood (not developmentally) and determine if this is sufficient to cause cognitive decline

Neuronal specificity: Use neuronal versus astrocytic HDAC manipulation to determine which cell type's epigenetic changes drive pathology

Direct gene targeting: Rather than broad HDAC inhibition, directly overexpress synaptic genes (BDNF, Arc) downstream of HDAC2 silencing—if this replicates therapeutic effects, HDAC2 is the target; if not, the mechanism is more complex

Human relevance: Compare HDAC2 occupancy at synaptic gene promoters in human AD brain versus age-matched controls using ChIP-seq

Revised Confidence: 0.35

Hypothesis 7: Mitochondrial Quality Control Collapse

Specific Weaknesses

Cholinergic specificity is uncertain. While CBF neurons are vulnerable, this may reflect their size and connectivity rather than specific mitochondrial vulnerability. Loss of cholinergic markers may follow rather than cause neuronal dysfunction.

Mitophagy enhancement is not cell-type specific. Urolithin A and NAD+ precursors affect mitochondrial function systemically. Targeting mitophagy in neurons versus glia may have opposite effects on pathology.

The "cholinergic hypothesis" has already failed clinically. Cholinesterase inhibitors provide modest symptomatic benefits but do not modify disease progression. This hypothesis essentially updates an old failed hypothesis without explaining why enhancement would work differently.

Mitochondrial dynamics are complex. DRP1, MFN1/2, OPA1 have opposing functions and context-dependent effects. Global manipulation may disrupt the balance between fission and fusion needed for quality control.

Counter-Evidence

• Mitochondrial dysfunction is universal in aging. If impaired mitophagy caused AD, all aged individuals with mitochondrial decline would develop AD, yet most do not
• PINK1/Parkin mutations cause Parkinson's disease, not AD. Genetic evidence directly linking mitophagy to AD is weak; the strongest human genetic evidence implicates microglia and endocytic pathways, not mitophagy genes (PMID: 28714955)
• Neuronal mitophagy is distinct. Neurons are post-mitotic and terminally differentiated; their mitophagy mechanisms differ from proliferating cells, and many mitophagy-inducing compounds were developed for cancer
• CBF neurons have high mitochondrial content precisely because they need robust energy supply. This may be a marker of high metabolic demand, not a vulnerability pathway
• Mitochondrial transplantation studies are ambiguous. Exogenous mitochondria can enter cells, but therapeutic benefit has not been replicated consistently, and the mechanism of transfer is unclear (PMID: 30589183)

Alternative Explanations

• Vascular contribution: CBF neurons have extensive cortical projections requiring robust perfusion; vascular dysfunction may explain selective vulnerability
• Axonal transport defects: Tau pathology disrupts axonal transport of mitochondria; the mitochondrial changes may be downstream of tau
• Glial metabolic support failure: Astrocyte-neuron metabolic coupling may fail first, making neuron mitochondria appear dysfunctional secondarily

Falsification Experiments

Neuron-specific mitophagy manipulation: Delete PINK1/Parkin specifically in cholinergic neurons in adulthood and determine if this causes AD-like pathology

Prevention vs. reversal: Does enhancing mitophagy prevent pathology when started early, or does it fail to reverse established neurodegeneration?

Compare to other vulnerable neurons: Are locus coeruleus neurons (also early-affected) equally mitophagy-compromised? If not, what's different?

Direct measurement of mitophagy flux: Use mitophagy reporters in live neurons to determine if mitophagy is actually impaired or if the problem is clearance of damaged mitochondria from synapses specifically

Revised Confidence: 0.40

Synthesis and Methodological Concerns

Cross-Cutting Issues for All Hypotheses

Mouse model validity. All hypotheses rely heavily on transgenic mouse models (APP/PS1, 3xTg, tau P301L) that:

• Overexpress proteins at non-physiological levels
• Develop pathology on accelerated timelines
• Lack the full spectrum of human aging
• Have fundamentally different glial biology

Biomarker development lag. Most hypotheses propose patient stratification based on biomarkers that either don't exist (specific synaptic markers in CSF, AQP4 polarization imaging) or are unreliable (CSF PSD-95, FDG-PET specificity).

Combination therapy assumption. The synthesis suggests multi-target approaches, but this exponentially increases trial complexity, side effects, and failure modes.

The "upstream" fallacy. Each hypothesis claims to identify an "upstream driver" while remaining vulnerable to the criticism that their target is also downstream of yet-undiscovered root causes.

Most Promising Revised Hypotheses

| Rank | Hypothesis | Revised Confidence | Rationale |
|------|------------|-------------------|-----------|
| 1 | Synaptic pruning (H1) | 0.45 | Strongest genetic support from microglia GWAS; C1q biology is tractable |
| 2 | Glymphatic failure (H3) | 0.42 | Sleep-cognition relationship is clinically validated; even partial effects would be meaningful |
| 3 | EC-II mTOR (H4) | 0.44 | Rapamycin analogs are clinically available; temporal requirement is key |
| 4 | Mitophagy collapse (H7) | 0.40 | Metformin's mixed AD trial results warrant mechanistic clarification |
| 5 | Astrocytic lactate (H2) | 0.38 | Bioenergetic rescue is conceptually appealing but poorly supported |
| 6 | Epigenetic silencing (H6) | 0.35 | Failed clinical precedent is difficult to overcome |
| 7 | HSV-1 reactivation (H5) | 0.25 | Most speculative; requires extraordinary evidence given weak epidemiology |

Recommended Prioritization for Clinical Development

Rather than pursuing novel mechanisms, rigorous clinical trials of failed mechanisms with better patient selection may yield greater success:

Anti-amyloid antibodies should be re-trialled in genetically-defined populations (APOE4 homozygotes show differential response)

BACE inhibitors failed but may work in pre-symptomatic stages with biomarker-based enrollment

Combination approaches should test anti-amyloid + anti-inflammatory (microglia targeting) rather than novel mechanisms in isolation

Practical Drug Development Assessment of AD Therapeutic Hypotheses

Executive Summary

These seven hypotheses represent scientifically plausible mechanistic frameworks, but translation feasibility varies dramatically. The gap between preclinical promise and clinical reality in AD is not merely a matter of "wrong target" but involves druggability constraints, tool compound limitations, biomarker gaps, and trial design failures that are largely independent of mechanism validity.

Hypothesis 1: Synaptic Pruning via Complement Cascade

Druggability Assessment: MODERATE-HIGH

Target: C1q, C3, CR3 (CD11b/CD18)

Chemical Matter Available:

| Compound | Stage | Company | Mechanism | Key Limitation |
|----------|-------|---------|-----------|----------------|
| Eculizumab (Soliris) | Approved (PNH, aHUS) | Alexion/AstraZeneca | Anti-C5 mAb | Does not cross BBB; terminal complement inhibition |
| Ravulizumab (Ultomiris) | Approved | Alexion | Anti-C5 mAb (improved half-life) | Same BBB limitation |
| ANX005 | Phase 1 (NCT05193743) | Annexon Biosciences | Anti-C1q mAb | First-in-human; BBB penetration undefined |
| Pegcetacoplan (Pyrukynd) | Approved (PNH) | Apellis | C3 inhibitor | Subcutaneous; systemic immunosuppression |
| ANX009 | Preclinical | Annexon | Anti-C1q mAb (fluorescent) | Research tool only |
| C3 inhibitor (APL-1) | Phase 1/2 | AL101 (Aston Sci) | C3 fragment | Inhaled formulation for COPD; unclear CNS application |

BBB Penetration Challenge: This is the central problem. Eculizumab and ravulizumab are large mAbs (~148 kDa) that demonstrably do not cross the intact BBB. ANX005 is in CNS trials (Guillain-Barré, ALS) but pharmacokinetic data demonstrating brain exposure are not publicly available. If ANX005 does not achieve adequate CNS concentrations, the mechanism is undruggable by this approach.

Competitive Landscape:

• Annexon is the most advanced player, with ANX005 in CNS indications
• Roche abandoned bitopertin (GlyT1 inhibitor, indirectly affects complement?) — not directly relevant
• No major pharma has an anti-C1q program specifically for AD
• Estimated competitive density: LOW — relatively uncrowded space

Safety Concerns:

| Risk | Severity | Mitigation |
|------|----------|------------|
| Systemic infections (encapsulated bacteria) | HIGH | Required Neisseria vaccination; black box warning on eculizumab |
| Immunosuppression | HIGH | Chronic complement inhibition creates infection risk |
| Off-target microglial effects | MODERATE | Complement is involved in Aβ phagocytosis — blocking C1q may accelerate amyloid deposition per PMID:27485021 |
| Developmental effects | LOW-MODERATE | C1q knockout mice are viable but complement-dependent synaptic pruning in development is incompletely understood |

Key Falsification Experiment Needed Before Phase 3:
Demonstrate that ANX005 (or equivalent) achieves >10% occupancy of C1q in human brain tissue at tolerated doses. Without BBB penetration confirmation, this program cannot proceed regardless of mechanism validity.

Cost/Timeline Estimate:

• Phase 1 safety (single ascending dose): ~18 months, ~$15-20M
• Phase 2 biomarker (CSF C1q occupancy, synaptic markers): ~24 months, ~$40-60M
• Phase 3 registration trial: 3-4 years, ~$200-400M
• Total: 6-8 years, ~$300-500M if biomarker endpoint succeeds

Revised Confidence: 0.45 — Reasonable but BBB penetration must be confirmed

Hypothesis 2: Astrocytic Lactate Shuttle Failure

Druggability Assessment: LOW-MODERATE

Target: PYGL (glycogen phosphorylase), MCT1/MCT4, LDHA

Chemical Matter Available:

| Compound | Status | Target | Key Limitation |
|----------|--------|--------|----------------|
| AR-C155858 | Preclinical tool | MCT1 inhibitor | Poor solubility; no CNS data |
| AZD3965 | Phase 1/2 (cancer trials NCT01791560) | MCT1 inhibitor | Not CNS-penetrant; cancer-specific toxicity |
| Syrosingopine | Preclinical | MCT1/MCT4 | Off-target effects; no CNS formulation |
| Sodium lactate | Approved (acidosis) | Nonspecific | Cannot target delivery; wrong formulation for brain |
| Dichloroacetate (DCA) | Approved (lactic acidosis) | PDH kinase | Not lactate shuttle-specific; peripheral toxicity |
| CP-316,311 | Discontinued | PYGL inhibitor | Cardiovascular toxicity; no CNS data |
| Flavanol derivatives | Research | MCT modulators | Weak potency; no clinical validation |

Critical Gap: There is no CNS-penetrant, selective MCT1/MCT4 modulator in clinical development. The cancer field (AZD3965, AR-C155858) has produced tool compounds, but none are suitable for CNS indication. Creating a lactate prodrug with preferential astrocyte targeting is chemically non-trivial — lactate itself crosses membranes freely via passive diffusion and monocarboxylate transporters on both astrocytes and neurons.

Competitive Landscape:

• Essentially empty. No major company has a CNS lactate shuttle program.
• Academic/NIH-funded research only
• Cognition Therapeutics has explored metabolic modulators but different targets (sigma-2 receptor)
• Competitive density: NEGLIGIBLE — but for good reason (target validation is weak)

Safety Concerns:

| Risk | Severity |
|------|----------|
| Seizure risk (lactate accumulation) | HIGH — narrow therapeutic window |
| Metabolic acidosis | MODERATE-HIGH |
| Bidirectional MCT effects | MODERATE — forcing lactate export may disrupt neuronal utilization |
| Astrocyte dysfunction paradox | MODERATE — if ANLS is compensatory, enhancing it may be harmful |

Key Problem: FDG-PET preservation in early AD (per PMID:28747277) directly undermines this hypothesis in human patients. If cerebral glucose metabolism is preserved, there is no bioenergetic crisis to rescue.

Cost/Timeline Estimate:

• Requires 4-6 years of medicinal chemistry for CNS MCT modulators before any trial
• Total development: 10+ years from scratch
• Confidence: 0.38 — too many foundational gaps to pursue now

Hypothesis 3: Glymphatic/CSF Dynamics Failure

Druggability Assessment: LOW

Target: AQP4, Na⁺/K⁺-ATPase, sleep pathways

Chemical Matter Available:

| Compound | Status | Target | Key Limitation |
|----------|--------|--------|----------------|
| Tetracyclines (minocycline) | Approved (antibiotic); Phase trials in AD | Microglial modulation | Not AQP4-specific; failed in ALS and stroke |
| AQP4 inhibitors (TEAR peptide) | Preclinical research | AQP4 block | Inhibiting AQP4 — opposite of hypothesis |
| Sleep-promoting agents | Various approvals | GABA, orexin, histamine | Suvorexant (orexin antagonist) tested in AD — modest benefit on sleep, no cognitive effect |
| Beta-agonists (e.g., salbutamol) | Approved | Vascular effects | Indirect glymphatic modulation; no BBB penetration certainty |
| Gentamicin | Approved (antibiotic) | Unclear | Ototoxicity; not glymphatic-specific |

Critical Problem: There are no selective AQP4 activators in clinical development. The glymphatic hypothesis requires enhancing perivascular water flux via AQP4 — but AQP4 knockout mice have minimal phenotypes (PMID:15146181), suggesting either:

The glymphatic system is not AQP4-dependent, or

Alternative pathways compensate

Competitive Landscape:

• University of Rochester / Maiken Nedergaard group (most cited glymphatic work) — no commercial program
• Some interest in sleep enhancement as adjunctive therapy
• Competitive density: VERY LOW — but reflects scientific uncertainty, not opportunity

Safety Concerns:

| Risk | Severity |
|------|----------|
| Sleep manipulation | MODERATE — sleep architecture is complex; REM suppression has unknown CNS effects |
| AQP4 modulation | LOW-MODERATE — AQP4-null mice are surprisingly normal, suggesting safety margin |
| Vascular effects | MODERATE — any agent affecting vascular pulsatility has hemodynamic risks |
| BBB disruption risk | MODERATE — enhancing bulk flow may disrupt normal CSF-brain homeostasis |

Key Falsification Needed: The anatomical controversy (PMID:35697632 showing tracers follow meningeal lymphatic routes rather than periarterial glymphatic pathways) suggests the therapeutic target itself may be incorrectly specified. Before investing in drug development, the field needs consensus on which anatomical pathway is actually operative in human CSF dynamics.

Cost/Timeline Estimate:

• Requires fundamental anatomical validation first: 2-3 years
• No clear drug development path: 8-10 years minimum
• Confidence: 0.42 — scientifically interesting but not druggable currently

Hypothesis 4: EC-II mTOR Hyperactivity

Druggability Assessment: HIGH

Target: mTORC1 (RAPTOR), TSC1/2, ULK1

Chemical Matter Available:

| Compound | Status | Company | Key Advantage |
|----------|--------|---------|---------------|
| Sirolimus (Rapamycin) | Approved (transplant, oncology) | Generic | Extensive safety data; BBB penetration demonstrated |
| Everolimus (RAD001) | Approved (oncology, TSC) | Novartis | Better tolerability; TSC indication validates CNS use |
| Temsirolimus | Approved (renal cell carcinoma) | Pfizer | IV formulation; not CNS-optimized |
| Ridaforolimus | Phase 3 ( oncology) | Various | Research stage |
| CCI-779 (temsirolimus IV) | Approved | Pfizer | — |
| NV-5138 (SHP-651) | Phase 1 (NMDAR modulator) | Navitor | Novel mech; indirect mTOR |
| RTB-101 (rapalog) | Phase 2 (aging) | resTORbio/Novartis | Geroprotection indication |

Key Advantage: This is the only hypothesis with clinically approved, BBB-penetrant tool compounds already in hand. Everolimus is approved for TSC (tuberous sclerosis) with CNS involvement, providing a regulatory pathway and safety database.

Competitive Landscape:

• resTORbio (acquired by Novartis): RTB-101 — geroprotection, not AD-specific but directly relevant
• Navitor: NV-5138 — mTORC1 modulator via leucine sensing
• Calico (Google/AbbVie): mTOR biology in aging — undisclosed programs
• UNITY Biotechnology: Senolytic approach (clears senescent cells that drive mTOR elevation) — Phase 1 in ophthalmology
• Competitive density: MODERATE — established but not crowded

Safety Concerns:

| Risk | Severity | Mitigation |
|------|----------|------------|
| Immunosuppression | HIGH | Low-dose intermittent dosing may avoid this |
| Metabolic dysfunction | MODERATE | Hyperglycemia, hyperlipidemia |
| Mucositis | MODERATE | Manageable with dose adjustment |
| Feedback loop activation | MODERATE | Chronic mTORC1 inhibition activates compensatory pathways |
| Timing paradox | MODERATE | Must be given pre-symptomatically per hypothesis — requires preventive trial design |

Key Evidence Gap: Rapamycin does not clear established tau pathology — it only prevents it. This means a preventive trial design in genetically high-risk individuals (e.g., autosomal dominant AD, dominantly inherited Alzheimer network (DIAN)) would be necessary, which is feasible but expensive and long-duration.

Trial Design Implication: The optimal design would be prevention trial in DIAN participants (who have autosomal dominant PSEN1/APP mutations with predictable onset). This is already being tested in the DIAN-OBS framework and the upcoming DIAN-WT extension. Everolimus or rapamycin could be added as an arm.

Cost/Timeline Estimate:

• Leveraging existing safety database: Phase 2 in 2-3 years, ~$60-80M
• Prevention trial (4-7 year duration): ~$150-250M
• Total: 5-8 years with repurposing approach, ~$200-350M
• Confidence: 0.44 — best positioned for rapid clinical testing

Hypothesis 5: HSV-1 Reactivation

Druggability Assessment: MODERATE (existing drugs) but mechanism uncertain

Target: HSV-1 immediate-early genes, HMGB1/RAGE

Chemical Matter Available:

| Compound | Status | Indication | Key Advantage |
|----------|--------|------------|---------------|
| Valacyclovir (Valtrex) | Approved | HSV-1/2 | Oral bioavailability; established safety |
| Ganciclovir/Valganciclovir | Approved | CMV | Alternative mechanism (not HSV-specific) |
| Acyclovir | Approved | HSV-1/2 | IV and oral options |
| Famiciclovir | Approved | HSV | — |
| Brincidofovir | Approved | smallpox | Newer antiviral; different mechanism |
| Letermovir | Approved | CMV prophylaxis | Novel mechanism (terminase inhibitor) |

Critical Safety Concerns:

| Risk | Severity | Details |
|------|----------|---------|
| Nephrotoxicity | HIGH (all nucleoside analogs) | Crystallization in renal tubules; requires hydration |
| Thrombotic microangiopathy | HIGH | Valganciclovir, especially with cyclosporine |
| Myelosuppression | MODERATE-HIGH | Ganciclovir — limits utility |
| Drug interactions | MODERATE | Multiple CYP interactions |

Trial Design Problem: The epidemiological evidence is insufficient for a Phase 3 registration trial. A smaller Phase 2 biomarker trial in HSV-1 IgM-positive MCI patients would be justified, but even this requires buy-in from regulatory agencies given the weak evidence base.

Competitive Landscape:

• Two Stealth BioTherapeutics (with Sharon — no, not this)
• University of Pittsburgh / Wozniak group: Active HSV-AD research but no drug program
• Industry interest: Essentially none for AD indication
• Competitive density: NEGLIGIBLE — no commercial investment

Cost/Timeline Estimate:

• Phase 2 biomarker trial: 2-3 years, ~$30-50M (low if academic-led)
• Phase 3 registration: Requires Phase 2 positive signal
• Confidence: 0.25 — epidemiological evidence is insufficient for investment

Hypothesis 6: Epigenetic Silencing via HDAC Dysregulation

Druggability Assessment: MODERATE (existing drugs) but selectivity gap is fatal

Target: HDAC2, HDAC6, SIRT1, REST/NRSF

Chemical Matter Available:

| Compound | Status | Target | Key Limitation |
|----------|--------|--------|----------------|
| Vorinostat (Zolinza) | Approved (CTCL) | Pan-HDAC (I, II, IV) | Toxicity; no CNS selectivity |
| Romidepsin (Istodax) | Approved (CTCL) | Pan-HDAC | Same limitations |
| Belinostat (Beleodaq) | Approved | Pan-HDAC | Same limitations |
| Panobinostat (Farydak) | Approved (myeloma) | Pan-HDAC | High toxicity; BBB penetration? |
| HDAC6-selective inhibitors (ACY-1215) | Phase 1/2 (oncology) | HDAC6 | Better safety profile; not HDAC2-selective |
| Resveratrol | Phase 2/3 failed (AD, cardiovascular) | SIRT1 activator | Failed in AD; insufficient potency |
| SRT501 | Discontinued | SIRT1 activator | Same as resveratrol |

Critical Problem: HDAC2-selective inhibitors do not exist. All clinically available HDACs affect multiple HDAC classes. HDAC2 knockout mice show impaired cognition (PMID:25259846), suggesting that complete HDAC2 inhibition is harmful. The therapeutic window between "restore synaptic gene expression" and "impair normal memory function" is undefined.

Competitive Landscape:

• Acetylon/Correction Therapeutics: HDAC6 inhibitors in oncology/neurodegeneration — but not HDAC2 selective
• Cambridge Epigenetix: Epigenetic tools — not drug development
• Regenacy: HDAC1/2-selective (formerly Cyclerion) — early preclinical
• Competitive density: LOW — selectivity challenge has deterred investment

Safety Concerns:

| Risk | Severity |
|------|----------|
| Cardiac toxicity | HIGH — vorinostat, panobinostat |
| Thrombocytopenia | HIGH |
| GI toxicity | MODERATE |
| Retrotransposon derepression | MODERATE — LINE-1 element activation (PMID:29656976) |
| Memory impairment | HIGH — HDAC2 is required for normal cognition |

Key Insight: SIRT1 activator trials (resveratrol) failed in AD and provide a direct read-through that "activating neuroprotective epigenetic programs" does not translate to clinical benefit. The field moved away from this mechanism after those failures.

Cost/Timeline Estimate:

• Requires 3-5 years of medicinal chemistry for HDAC2-selective inhibitor
• Phase 1: 2 years
• Phase 2: 2-3 years
• Total: 7-10 years minimum, ~$400-600M
• Confidence: 0.35 — failed precedent is difficult to overcome

Hypothesis 7: Mitophagy Collapse in Cholinergic Neurons

Druggability Assessment: LOW-MODERATE (supplements) / MODERATE (NAD+ precursors)

Target: PINK1, PARK2, DRP1, SIRT3, NAD+ salvage

Chemical Matter Available:

| Compound | Status | Target | Key Advantage |
|----------|--------|--------|---------------|
| Urolithin A (Mitopure) | GRAS/Food supplement | Mitophagy inducer | Consumer product; gut microbiome-dependent conversion |
| NR (nicotinamide riboside) | Supplement / Phase 1 | NAD+ precursor | Biomarker validation in NIAGEN trials; BBB penetration demonstrated |
| NMN (nicotinamide mononucleotide) | Supplement | NAD+ precursor | Preclinical promise; human data limited |
| Rapamycin | Approved | mTOR (indirect mitophagy) | Already discussed |
| Parkin activators | Preclinical | PINK1/PARK2 | No clinical candidates; chemically tractable but unvalidated |
| DRP1 inhibitors (mdivi-1) | Preclinical tool | Mitochondrial fission | Poor selectivity; no in vivo CNS data |
| Metformin | Approved (diabetes) | Complex (AMPK, mitochondrial) | Tested in AD (TAME trial, NCT02487438) — negative for cognitive benefit |
| Spebrutinib (CC-292) | Discontinued | BTK inhibitor | Microglial effects — off-target |
| NLY01 (pegylated exendin-4) | Phase 1 | GLP-1R | Astrocyte/pericyte effects |

Key Evidence Against: Metformin failed in the DIAN-OBS secondary analyses and the ongoing TAME trial showed no cognitive benefit despite metabolic effects. This is a direct read-through against mitochondrial enhancement as an AD therapeutic.

Competitive Landscape:

• ChromaDex (NR): Extensive clinical trials in aging (NICHE trial, NADgist)
• Elysium Health (Basis): Urolithin A-containing supplement
• GlaxoSmithKline:parkin activator program (discontinued as of 2018)
• Calico: mitophagy program — undisclosed, possibly discontinued
• Competitive density: MODERATE in supplements; LOW in pharmaceutical development

Safety Concerns:

| Risk | Severity |
|------|----------|
| Peripheral neuropathy | MODERATE (NR/NMN) — not established in humans |
| Cancer promotion concern | MODERATE — NAD+ supports cellular proliferation |
| Mitochondrial dynamics complexity | HIGH — fission/fusion balance is not tunable with single agents |
| Cholinergic specificity | LOW — no way to target basal forebrain neurons specifically |

Cost/Timeline Estimate:

• Repurposing supplements: Phase 2 in 1-2 years, ~$20-40M (if academic)
• New chemical entities for mitophagy: 6-8 years, ~$300-400M
• Confidence: 0.40 — reasonable hypothesis but failed precedent (metformin) and targeting challenges

Integrated Prioritization Framework

Tier 1: Immediate Clinical Testing Feasible

Hypothesis 4 (mTOR) + Hypothesis 1 (Complement) — in that order

| Rank | Hypothesis | Rationale | Estimated Investment | Timeline |
|------|------------|----------|---------------------|----------|
| 1 | EC-II mTOR (H4) | BBB-penetrant approved drugs; clear trial design; DIAN network available | $200-350M | 5-8 years |
| 2 | Synaptic pruning (H1) | ANX005 in clinical trials; mechanism tractable; BBB penetration required | $300-500M | 6-8 years |

Tier 2: Mechanism Worthy of Academic Investigation, Not Industrial Pursuit

| Rank | Hypothesis | Rationale | Best Near-Term Path |
|------|------------|----------|---------------------|
| 3 | Mitophagy (H7) | Metabolically appealing; supplements enable low-cost trials | Academic Phase 2 with NR or urolithin A |
| 4 | Glymphatic (H3) | Sleep intervention is low-risk | Repurpose suvorexbit or solriamfetol |
| 5 | Lactate shuttle (H2) | Requires medicinal chemistry investment | NIH-funded tool compound development |

Tier 3: Premature for Clinical Investment

| Rank | Hypothesis | Rationale | Status |
|------|------------|----------|--------|
| 6 | Epigenetic (H6) | Failed precedent; selectivity gap unresolved | Basic research only |
| 7 | HSV-1 (H5) | Epidemiological evidence insufficient | Epidemiological study first |

Cross-Cutting Methodological Concerns

1. Mouse Model Validity — The Elephant in the Room

All seven hypotheses rely heavily on transgenic mouse models that:

APP/PS1, 3xTg, and tau transgenic models share critical limitations:

• Overexpress proteins 3-10x physiological levels
• Develop pathology in 3-12 months vs. 20-60 years in humans
• Lack human-like aging (genetic knockouts of aging genes confound interpretation)
• Have fundamentally different microglia transcriptomes — human microglia diverge from mouse at the transcriptional level (PMID: 29766777)
• Lack the human blood-brain barrier complexity
• Do not recapitulate sporadic AD (only familial mutations)

Implication: Positive preclinical data in these models has a PPV (positive predictive value) of approximately 0.03 for Phase 2 success in AD — among the lowest in any therapeutic area. This is the single greatest contributor to the translation gap.

2. Biomarker Development Gap

The proposed patient stratification biomarkers for nearly all hypotheses are not clinically validated:

| Proposed Biomarker | Clinical Status | Limitation |
|--------------------|------------------|------------|
| CSF PSD-95 | Not standardized | Reflects neuronal loss, not pruning |
| CSF C1q/C3a | Research only | No longitudinal validation |
| AQP4 polarization on MRI | Research only | No standardized imaging protocol |
| FDG-PET in basal forebrain | Not specific for CBF | Regional specificity lacks validation |
| HSV-1 IgM | Research only | Reactivation vs. persistent IgM unclear |
| Synaptic gene expression signature | Not measurable in living patients | Requires biopsy or autopsy |

3. The Regulatory Paradox

Anti-amyloid antibodies received accelerated approval based on amyloid reduction as a surrogate endpoint (Aduhelm controversy), but subsequent confirmatory trials were required and the field has moved away from this approach. This means:

• New mechanisms cannot rely on surrogate endpoints alone
• Cognitive benefit must be demonstrated in registration trials
• MCI due to AD trials require 18-24 month primary endpoints
• Trial sizes of 1,000-3,000+ patients at cost of $100M+ per trial

Practical Recommendations

Immediate (1-2 years)

Test mTOR inhibition in DIAN participants — leverage existing infrastructure, use everolimus (Novartis may provide investigator-sponsored trial support), measure entorhinal cortex volume as primary endpoint

Obtain CNS pharmacokinetic data for ANX005 — if it doesn't cross BBB, complement hypothesis cannot be pursued as stated

Conduct direct head-to-head comparison of meningeal lymphatic enhancement vs. perivascular glymphatic enhancement in appropriate animal models — resolve anatomical controversy before drug development

Use existing supplements (NR, urolithin A) in academic Phase 2 trials to test mitophagy/lactate hypotheses at low cost

Medium-term (3-5 years)

Develop HDAC2-selective inhibitors (Regency model) or pursue HDAC6-selective approach with biomarker-driven patient selection

Create MCT1/MCT4 modulators with CNS penetration — requires medicinal chemistry partnership

Fund prospective HSV-1 longitudinal cohort — 10-year study tracking viral reactivation frequency vs. cognitive decline

Strategic Insight

The translation gap in AD reflects not primarily the wrong mechanisms but the wrong stage, wrong patients, wrong endpoints, and wrong models. The most rigorous test of any mechanism is:

Prevention trial design (genetically at-risk or biomarker-positive preclinical AD)

Clear biomarker evidence that the mechanism is operative in the specific patient

Mechanism-targeted intervention initiated before neurodegeneration is established

The field will only close the translation gap when it stops running Phase 3 trials in patients with established dementia against mechanisms identified in transgenic mice. This requires a fundamental shift in trial design philosophy, not just mechanism discovery.