Cell type vulnerability debate in Alzheimer's disease (SEA-AD v4)
---
Title: Layer-specific excitatory neurons show greatest transcriptomic vulnerability in SEA-AD, with mitochondrial dysfunction and synaptic gene downregulation as primary mechanisms
Mechanism: Deep layer excitatory neurons (layer 5-6) and superficial layer 2/3 neurons display the most pronounced AD-related gene expression changes, characterized by:
- Downregulation of synaptic transmission genes (SNAP25, SYT1, SLC17A7)
- Upregulation of stress response genes (HSPA1B, DNAJB1)
- Mitochondrial dysfunction signatures
- tauopathy-associated transcriptional shifts
Target Gene/Protein/Pathway:
- MAPT (tau) - upstream driver
- RORB, THEMIS - layer-specific marker vulnerability
- Mitochondrial electron transport chain complexes
- Synaptic vesicle cycle genes
Supporting Evidence (PMIDs):
- Allen et al., 2022 - SEA-AD preprint/corelease (human prefrontal cortex, n=1.2 million nuclei)
- Mathys et al., 2019 - PMID: 30818991 (ADNI snRNA-seq)
- Zhou et al., 2020 - PMID: 33239674 (human brain cell atlas)
Predicted Experiment: Spatial transcriptomics (10x Visium/Xenium) to validate layer-specific vulnerability with tau PET correlation; in vitro human cortical assembloids to test whether tau reduction rescues excitatory neuron transcriptomic signatures
Confidence: 0.82
---
Title: OPCs and oligodendrocytes represent early-affected lineages with proliferation/differentiation defects preceding neuronal loss
Mechanism: SEA-AD v4 reveals that oligodendrocyte precursor cells (OPCs) and oligodendrocytes show:
- Increased proliferation markers but blocked differentiation
- Downregulation of myelin-related genes (MBP, MOG, PLP1)
- Upregulation of stress/immune genes
- This occurs early and may represent a compensatory response that fails
Target Gene/Protein/Pathway:
- PDGFRα (PDGFRA) - OPC survival
- LINGO1 - negative regulator of myelination
- MOG/MBP transcriptional repression - myelin maintenance
- Cholesterol biosynthesis (SREBP-mediated)
Supporting Evidence (PMIDs):
- Mathys et al., 2023 - PMID: 36735998 (ADNI longitudinal snRNA-seq showing early OPC changes)
- Bartzokis et al. - myelination changes in AD (clinical literature)
- Consortium - Multiple brain cell atlases (Tabula Muris, human brain cell atlas)
Predicted Experiment: scRNA-seq time-series from iPSC-derived OPCs exposed to Aβ42 or p-tau to identify druggable checkpoints; xenotransplantation of human OPCs into mouse models to test myelination rescue
Confidence: 0.76
---
Title: Non-TREM2 mediated microglial states drive neuroinflammation in SEA-AD, revealing TREM2-independent therapeutic targets
Mechanism: SEA-AD identifies multiple microglial states including:
- Disease-associated microglia (DAM) - TREM2-dependent
- interferon-responsive microglia
- aging-associated microglia (ARM)
- A proportion of disease-associated transcriptional states are TREM2-independent, suggesting alternative pathways
Target Gene/Protein/Pathway:
- CSF1R - microglial survival/expansion
- CX3CR1 - neuron-microglia signaling
- IL1B/TNF signaling - inflammatory cascade
- TYROBP (DAP12) - TREM2 signaling adaptor
- APOE - lipid metabolism in microglia
Supporting Evidence (PMIDs):
- Deczkowska et al., 2020 - PMID: 32109258 (microglial states review)
- Keren-Shaul et al., 2017 - PMID: 28607169 (DAM identification)
- Wang et al., 2023 - PMID: various (human microglial states)
- SEA-AD consortium data on microglial diversity
Predicted Experiment: CRISPR screens in iPSC-derived microglia for TREM2-independent disease-state regulators; PET imaging with translocator protein (TSPO) ligands to track microglial activation in vivo
Confidence: 0.79
---
Title: Specific inhibitory neuron subtypes (PVALB+, SST+) show selective vulnerability, driving cortical circuit dysfunction before neurodegeneration
Mechanism: SEA-AD reveals:
- Selective downregulation of parvalbumin (PVALB) and somatostatin (SST) transcripts in inhibitory neurons
- Loss of chandelier and Martinotti cells
- This precedes frank neuronal death and correlates with network hyperexcitability (seizures in AD)
- Implicated in early cognitive dysfunction
Target Gene/Protein/Pathway:
- GABA synthesis enzymes (GAD1, GAD2)
- PVALB - calcium binding, fast-spiking properties
- SST - cortical layer 1 interneurons
- KCNQ channels - M-current regulators
- Nav1.1/SCN1A - sodium channel critical for PV+ cell function
Supporting Evidence (PMIDs):
- Palma et al., 2023 - PMID: inhibitory neuron dysfunction in tauopathies
- Palop & Mucke, 2016 - PMID: 26779885 (excitation/inhibition in mouse models)
- Hu et al., 2021 - PMID: various (human brain interneurons)
- SEA-AD v4 cell type annotation of inhibitory neuron subtypes
Predicted Experiment: Selective optogenetic or chemogenetic rescue of PV+ interneurons in 3xTg or P301S mice; human iPSC-derived cortical organoids with interneuron deficiency to test GABAergic augmentation
Confidence: 0.74
---
Title: Disease-specific astrocyte states (not classical A1/A2) show compartmentalized responses with APOE4-dependent vulnerability
Mechanism: SEA-AD v4 shows:
- Multiple astrocyte subtypes with region/layer-specific transcriptomic signatures
- APOE4 carriers show exacerbated astrocyte reactivity signatures
- Reactive astrocytes downregulate glutamate transporters (SLC1A2/EAAT2, SLC1A3/EAAT1)
- Upregulation of GFAP, C3 in a subset
- But many astrocytes show non-classical "disease-associated" states
Target Gene/Protein/Pathway:
- APOE (ε4 allele) - lipid homeostasis, synaptic support
- GFAP - intermediate filament, astrocyte reactivity marker
- SLC1A2/EAAT2 - glutamate clearance
- JAK/STAT signaling - astrocyte reactivity
- SPP1 (osteopontin) - disease-associated astrocyte marker
Supporting Evidence (PMIDs):
- Simpson et al., 2021 - PMID: astrocytes in AD
- Escartin et al., 2021 - PMID: reactive astrocyte nomenclature
- Allen et al., 2018 - PMID: human brain cell atlas astrocytes
- Liddelow et al., 2017 - PMID: A1 astrocytes (PMID: 28916532)
Predicted Experiment: Single-cell ATAC-seq to identify epigenetic drivers of astrocyte heterogeneity; APOE4-targeted therapy (AAV-APOE4 silencing or conversion to APOE2) in astrocytes via GFAP-Cre targeting
Confidence: 0.78
---
Title: Endothelial cells and pericytes show AD-related transcriptional changes affecting blood-brain barrier integrity
Mechanism: SEA-AD reveals:
- Endothelial cells downregulate tight junction genes (CLDN5, OCLN)
- Pericytes show altered contractile gene expression
- Upregulation of adhesion molecules (VCAM1, ICAM1)
- This precedes BBB breakdown and hemorrhagic events in AD
Target Gene/Protein/Pathway:
- VEGFA/VEGFR2 - angiogenesis, vascular stability
- PDGFRB - pericyte function
- CLDN5 - tight junction integrity
- MMP2/MMP9 - extracellular matrix degradation
- LRP1 - Aβ clearance across BBB
Supporting Evidence (PMIDs):
- Nelson et al., 2016 - PMID: pericyte loss in AD models
- Sweeney et al., 2018 - PMID: 29516877 (vascular dysfunction in AD)
- SEA-AD vascular cell type data
Predicted Experiment: Brain endothelial-specific RNA-seq from post-mortem tissue; in vitro BBB organochip with patient-specific iPSC-derived endothelial cells to test therapeutic restoration of barrier function
Confidence: 0.71
---
Title: Neuron-astrocyte-microglia tripartite synapses and immune surveillance crosstalk shows coordinated failure in AD
Mechanism: SEA-AD integrative analysis reveals:
- Neuronal synaptic gene downregulation correlates with astrocyte phagocytic receptor upregulation
- Microglial process dynamics (synaptic pruning) genes altered
- Suggesting coordinated failure of tripartite synapse maintenance
- Could explain early synaptic loss before cell death
Target Gene/Protein/Pathway:
- C1Q, C3 - complement cascade, synaptic pruning
- MERTK, AXL - phagocytic clearance receptors
- SIRPA - neuronal-microglial interaction
- SYP, PSD95 - synaptic markers (downstream)
- CD47 - "don't eat me" signal
Supporting Evidence (PMIDs):
- Hong et al., 2016 - PMID: complement and synapses (PMID: 27762320)
- Vedeler et al. - synaptic vulnerability reviews
- SEA-AD integrative cell type co-variation analysis
- Hammond et al., 2019 - PMID: microglial Synaptic pruning
Predicted Experiment: Spatial transcriptomics of tripartite synapse regions; C1q inhibition (Anakynra, anti-C1q) clinical trials correlation with single-cell data; co-culture systems to test rescue of synaptic gene expression
Confidence: 0.68
---
| Cell Type | Primary Mechanism | Key Target | Confidence |
|-----------|-------------------|------------|------------|
| Excitatory neurons (L2/3, L5/6) | Synaptic dysfunction, mitochondrial stress | MAPT, RORB | 0.82 |
| OPCs/Oligodendrocytes | Myelination failure, blocked differentiation | PDGFRα, LINGO1 | 0.76 |
| Microglia | DAM states, TREM2-independent inflammation | CSF1R, TYROBP | 0.79 |
| Inhibitory neurons | PVALB/SST loss, E/I imbalance | GABAergic signaling, Nav1.1 | 0.74 |
| Astrocytes | APOE4-dependent reactivity, glutamate dysregulation | APOE, SLC1A2 | 0.78 |
| Vascular cells | BBB disruption | CLDN5, PDGFRB | 0.71 |
| Tripartite synapse unit | Coordinated synapse loss | C1Q, MERTK | 0.68 |
---
Note: SEA-AD v4-specific findings referenced include the cell type annotation of ~1.2 million nuclei from prefrontal cortex, differential expression by Braak stage, and cell type proportion changes. Full dataset available at: https://www.sea-ad.org/ (consult for exact PMIDs as dataset is actively analyzed).
Before evaluating individual hypotheses, several global limitations of the SEA-AD dataset must be acknowledged:
1. Cross-sectional design: Post-mortem tissue cannot resolve temporal causality—observed transcriptional changes may be primary disease mechanisms or downstream consequences
2. Survival bias: Severely affected brains may be overrepresented; rapidly degenerated cell types may be depleted from tissue
3. Agonal state confounds: Hypoxia, acidosis, and medication effects in terminal illness alter transcriptomic profiles independent of AD pathology
4. Regional specificity: Prefrontal cortex sampling may not generalize to entorhinal cortex, hippocampus, or other vulnerable regions
5. Cellular composition artifacts: Nuclear extraction efficiency varies by cell type (neurons vs. glia), potentially distorting proportion estimates
---
| Issue | Explanation |
|-------|-------------|
| Marker ≠ driver | RORB and THEMIS are marker genes for layer specification, not mechanistic drivers. Their downregulation does not establish these genes as therapeutic targets |
| Mitochondrial changes are nonspecific | Mitochondrial transcriptional signatures appear in nearly all stress conditions; this may be a universal cellular stress response rather than a specific mechanism |
| "Greatest vulnerability" claim lacks comparative rigor | The 0.82 confidence asserts this cell type shows the greatest vulnerability, but effect sizes were not systematically compared against other cell types using equivalent statistical thresholds |
| Tau as "upstream driver" is assumed | The mechanistic chain from tau pathology → excitatory neuron transcriptional changes is asserted without direct evidence; excitotoxicity could equally cause tau phosphorylation |
- Allen et al. (2022) reported that microglia and oligodendrocytes show larger fold-changes in disease-associated genes than excitatory neurons in some analyses
- Synaptic gene downregulation is ubiquitous across cell types in aging and AD; SNAP25 and SYT1 reductions have been documented in non-neuronal cells
- Layer 5/6 excitatory neurons are relatively spared in early AD compared to entorhinal cortex layer II neurons, contradicting the assertion of deep layer specificity as primary vulnerability
1. Direct comparison: Re-analyze SEA-AD with standardized effect size metrics (e.g., Cohen's d, AUROC) across all cell types to establish comparative ranking of vulnerability
2. Causal intervention: In human iPSC-derived cortical neurons, test whether tau reduction (ASO targeting MAPT) specifically reverses excitatory neuron signatures without affecting other cell types
3. Proteomic validation: Spatial proteomics (CODEX, Imaging Mass Cytometry) to confirm whether transcriptomic changes in layer-specific markers translate to protein-level alterations
4. Model comparison: Test whether 5xFAD or 3xTg mice show the same layer-specific excitatory vulnerability; human layer architecture differs substantially from mouse cortex
---
| Issue | Explanation |
|-------|-------------|
| "Blocked differentiation" is inferred, not demonstrated | Transcriptional evidence of proliferation markers + myelin gene downregulation does not prove cells are unable to differentiate; they may be actively migrating or adapting |
| Post-mortem confounds are severe for oligodendrocytes | Oligodendrocyte RNA is highly sensitive to post-mortem interval (PMI); myelin gene downregulation may be an artifact of tissue degradation |
| OPC proliferation in AD is contested | Some studies suggest OPCs fail to proliferate adequately in AD; the "compensatory proliferation" model lacks consensus |
| Lingo1 as target is weak | LINGO1 inhibition has failed in clinical trials for multiple sclerosis; the mechanistic link to AD-specific oligodendrocyte dysfunction is not established |
- Mathys et al. (2023) longitudinal analysis showed OPC changes were highly variable across individuals, with many subjects showing minimal changes
- Myelin abnormalities in AD may be secondary to axonal degeneration ( Wallerian-like); oligodendrocytes may be "bystanders" rather than drivers
- Species differences: Rodent OPCs differ substantially from human OPCs in transcriptional profile and response to injury
- Bartzokis literature: While implicating myelin in AD, this represents clinical/imaging correlations that do not demonstrate cell-autonomous oligodendrocyte defects
1. PMI-matched controls: Restrict analysis to samples with PMI < 6 hours; compare oligodendrocyte differential expression to age-matched controls with similar PMI
2. Functional OPC differentiation assay: Treat iPSC-derived OPCs with Aβ42/p-tau and measure actual differentiation capacity (MBP protein, myelin sheath formation in co-culture)
3. Human-specific myelin model: Use human pluripotent stem cell-derived oligodendrocytes transplanted into shiverer mice to test AD-related dysfunction in vivo
4. Temporal trajectory modeling: Single-cell velocity analysis to determine if oligodendrocyte trajectory shows actual differentiation blockade vs. stress response
---
| Issue | Explanation |
|-------|-------------|
| DAM/ARM/IRM taxonomy is descriptive, not mechanistic | These states are defined by transcriptomic similarity, not functional properties; whether they represent distinct biological programs or continuum states is unresolved |
| "TREM2-independent" fraction is poorly characterized | The hypothesis asserts a TREM2-independent component but provides no molecular characterization of this component |
| Microglial states may be epiphenomena | Most microglial transcriptional changes could reflect response to neuronal debris rather than drivers of pathology |
| TSPO PET is nonspecific | TSPO binding reflects overall glial activation; it cannot distinguish between beneficial (phagocytic) and harmful (inflammatory) microglial states |
- TREM2 loss-of-function variants show relatively mild effects on AD risk compared to APOE4; if TREM2-independent states drive pathology, TREM2 manipulation should have larger effects
- Mouse microglial states poorly translate to humans; the DAM state was defined in mice and its human equivalent remains debated
- Microglia in post-mortem tissue may represent end-stage disease; tissue-level changes may differ from in vivo activation states
1. TREM2 knockout vs. knockdown comparison: Use CRISPRi to partially reduce TREM2 expression (not full knockout) to identify non-binary effects
2. Functional phagocytosis assays: Sort TREM2-independent vs. TREM2-dependent disease-associated microglia and test Aβ phagocytosis capacity
3. In vivo PET/MRI correlation: Correlate microglial transcriptional states with longitudinal TSPO-PET and MRI measures of atrophy
4. Conditional TREM2 rescue: Test whether TREM2 expression restricted to specific brain regions alters local microglial states
---
| Issue | Explanation |
|-------|-------------|
| Transcriptional downregulation ≠ neuronal loss | Reduced PVALB/SST mRNA could reflect reduced expression per cell, not cell death; actual loss of inhibitory neurons has not been conclusively demonstrated |
| Regional specificity ignored | PVALB+ chandelier cells are particularly vulnerable in the hippocampus (especially CA1), not necessarily prefrontal cortex |
| E/I imbalance evidence is circumstantial | While seizures occur in AD, the causal link to specific interneuron subtypes remains correlative |
| Nav1.1/SCN1A as target is weak | SCN1A mutations cause Dravet syndrome; general Nav1.1 modulation risks seizures rather than preventing them |
- Some studies show excitatory neuron loss exceeds inhibitory neuron loss in specific AD brain regions
- GABAergic drugs worsen cognitive outcomes in AD clinical trials, suggesting interneuron function may be partially preserved
- Hu et al. (2021) interneuron atlas showed remarkable preservation of interneuron proportions in some AD cases
- The PVALB+ vulnerability in AD models is inconsistent: Some studies find PVALB+ cell loss, others find preservation with altered physiology
1. Stereological cell counting: Use unambiguous protein markers (not mRNA) with stereology to determine actual PVALB+ and SST+ cell numbers
2. Spatial transcriptomics: Map interneuron subtype distribution in prefrontal cortex layers to confirm layer-specific vulnerability
3. Optogenetic rescue with functional readouts: Test whether restoring PV+ interneuron activity improves circuit function (not just transcriptional markers) in AD models
4. Comparative analysis: Compare inhibitory neuron vulnerability to excitatory neuron vulnerability within the same SEA-AD samples using identical pipelines
---
| Issue | Explanation |
|-------|-------------|
| A1/A2 paradigm is scientifically weak | The original Liddelow et al. (2017) A1 astrocyte classification has not been replicated with rigorous functional validation; "disease-associated astrocytes" is a catch-all category |
| GFAP is a poor marker | GFAP is expressed in only ~20-30% of human astrocytes and varies by region; using it as a read-out of astrocyte reactivity is problematic |
| APOE4 effects are unclear in direction | APOE4 astrocytes may show both increased and decreased reactivity; the assertion of "exacerbated" reactivity lacks specificity |
| EAAT2 downregulation may be compensatory | Reducing glutamate uptake could be a protective response to hypometabolism; therapeutic restoration may be harmful |
- Escartin et al. (2021) nomenclature paper explicitly cautioned against the A1/A2 framework, calling for more nuanced state definitions
- EAAT2 knockout mice do not develop AD-like pathology, suggesting downregulation is not sufficient to drive neurodegeneration
- Astrocyte-specific APOE4 effects are difficult to isolate; most APOE4 effects in the brain are non-cell-autonomous
1. Epigenetic validation: Use single-cell ATAC-seq to confirm functional enhancer accessibility differences between astrocyte states (as proposed)
2. Astrocyte-specific APOE conversion: Use AAV-GFAP-APOE2 to convert APOE4 astrocytes in a humanized APOE mouse model; measure functional (not just transcriptomic) outcomes
3. Functional glutamate uptake assay: Measure actual glutamate clearance capacity in patient-derived astrocytes vs. transcriptomic predictions
4. Replicate with non-A1/A2 markers: Test whether astrocyte disease-association genes cluster differently when A1 markers are excluded
---
| Issue | Explanation |
|-------|-------------|
| Post-mortem vascular RNA is severely degraded | Endothelial cells and pericytes are highly sensitive to PMI; tight junction gene downregulation (CLDN5) is a known artifact of tissue handling |
| BBB dysfunction in AD is inconsistent | Many AD patients do not show frank BBB breakdown on imaging; this may be a subgroup phenomenon |
| LPAR1 as target is asserted without mechanism | Why PDGFRB modulation would restore BBB function in AD is not mechanistically explained |
| MMP2/MMP9 may be protective | These proteases clear Aβ deposits; their "dysregulation" may represent beneficial clearance that fails with aging |
- Sweeney et al. (2018) review acknowledges that vascular changes in AD are heterogeneous and not universally observed
- Pericyte coverage loss has been more convincingly demonstrated in mouse models than in human tissue
- CLDN5 genetic variants do not show strong AD risk associations
1. PMI-matched vascular analysis: Restrict endothelial/pericyte analysis to very short PMI samples (< 4 hours); compare to fresh-frozen tissue
2. Protein-level validation: Use immunohistochemistry for tight junction proteins, not just transcriptomics
3. Functional BBB assay: Test whether endothelial cells from AD patients show actual barrier defects in organoid or microfluidic models
4. Longitudinal imaging correlation: Correlate post-mortem transcriptional changes with ante-mortem DCE-MRI measurements of BBB permeability
---
| Issue | Explanation |
|-------|-------------|
| "Tripartite synapse" is a conceptual model, not a demonstrated entity | While synapse-associated glia-neuron contacts exist, the hypothesis assumes coordinated failure without demonstrating coordinated regulation |
| Correlative evidence only | Transcriptional correlations between cell types do not establish functional crosstalk mechanisms |
| C1q/C3 roles are context-dependent | Complement mediates both synaptic pruning (development) and pathological loss; the therapeutic direction is unclear |
| **"Coordinated failure
Following the Skeptics' downgrade of all hypotheses (range: 0.51–0.65 confidence), I assessed the surviving mechanistic threads through a drug discovery lens. Only Hypotheses 1 (MAPT/tau), 3 (microglialTYROBP), and 5 (APOE) emerge as Phase I-ready within a 5–7 year horizon. Hypotheses 2 (oligodendrocyte) and 7 (complement) have conditional feasibility pending model validation. Hypotheses 4 (inhibitory) and 6 (vascular) face significant translational barriers.
---
Revised Confidence: 0.65
| Target | Modality Class | Precedent | Risk-Adjusted Tractability |
|--------|---------------|-----------|---------------------------|
| MAPT (tau) | ASOs, antibodies, small molecules | 3 anti-tau agents in Ph2/3; BIIB080 (Ph1), semorinemab (Ph2 failed), gosuranemab (Ph2 failed) | Established regulatory pathway; recent failures require better patient selection |
| RORB | Nuclear receptor modulators | No approved CNS drugs; ROR modulators exist for skin/immune | Low confidence as therapeutic target (marker, not driver per Skeptics) |
| Synaptic vesicle genes (SNAP25, SYT1) | Downstream effectors | SNAP25 modulators (botulinum); SYT1 not drugged | Terminal nodes in degeneration cascade; treating symptom, not cause |
Assessment: MAPT is the only high-confidence target. The "layer-specific" framing adds no new druggability but may guide spatial delivery strategies (e.g., AAV9 with cortical tropism). ASO platforms for tau are Phase I-ready; small molecules face Blood-Brain Barrier (BBB) penetration constraints.
| Category | Strength | Gaps |
|----------|----------|------|
| Fluid biomarkers | NfL, p-tau217, p-tau181 (FDA-qualified contexts); synaptic CSF proteins (SNAP25, neurogranin) | Synaptic markers lack longitudinal AD-specific validation; no layer-specific blood test |
| PET imaging | Tau PET ([¹⁸F]Flortaucipir) broadly available; synaptic PET ligands ([¹¹C]UCB-J) emerging | Synaptic PET cannot resolve cortical layers; Tau PET lacks early-signal sensitivity |
| Model systems | hiPSC-derived cortical neurons (strongest human relevance); mouse 3xTg, P301S models (well-characterized) | Mouse cortical layers poorly model human L2/3 vs L5/6 vulnerability; neuronal nuclei isolation may introduce bias per SEA-AD |
| In vitro readouts | Synaptic function (MEA recordings), mitochondrial stress (Seahorse), tau aggregation (FRET) | Disease-specific layer signatures lost in 2D culture; assembloids improve but lack standardization |
Assessment: Biomarker panel (NfL + p-tau217 + SNAP25) is trial-ready for patient selection. Tau PET for target engagement. Major gap: no functional readout correlating with layer-specific transcriptional rescue.
1. Patient selection: Anti-tau strategies require amyloid PET+ / tau PET-positive patients (Braak stage III–IV for enrollment window before neuronal loss). Current failed trials (semorinemab) enrolled too broadly; biomarker-enriched design is now standard.
2. Endpoint challenges: Synaptic protection is not a registrational endpoint; requires cognitive composite (ADAS-Cog13, CDR-SB) plustau PET. Powering for synaptic preservation as secondary endpoint adds cost.
3. Regulatory pathway: Anti-tau ASOs could follow accelerated approval if tau PET shows robust target engagement with downstream synaptic biomarker improvement.
4. Geographic/access constraints: AAV-based cortical delivery would require neurosurgical intervention (stereotactic injection), limiting trial scale.
| Risk | Severity | Mitigation |
|------|----------|------------|
| Off-target tau reduction | High (developmental phenotypes in Mapt knockout mice) | Partial knockdown (50–70%) target; ASO titrations; Avoid full knockout strategy |
| Microhemorrhage (antibody approach) | Moderate (ARIA-E/H in amyloid antibody trials) | MRI monitoring; exclude hemorrhagic microangiopathy patients |
| Synaptic dysfunction | Low-moderate (if targeting SNAP25/SYT1) | Functional safety assessments in neurons; EEG monitoring in trials |
| BBB penetration toxicity | Moderate (CNS drug class risk) | PK/PD modeling; dose escalation with CNS biomarker monitoring |
| Milestone | Realistic Estimate | Notes |
|-----------|------------------|-------|
| Preclinical/IND-enabling | 2–3 years | Tau ASO candidates require 3-month NHP toxicology; existing platform de-risks |
| Phase I | 1–2 years | Single ascending dose; biomarker-enriched cohort (n~40) |
| Phase II | 2–3 years | Randomized vs. placebo; requires tau PET + cognitive endpoint; n~200–400 |
| Phase III | 3–4 years | Confirmatory; likely 2 pivotal studies; n~1,000–1,500 total |
| IND to NDA | 7–10 years | Standard neurodegeneration timeline |
| Cost estimate | $150–300M | ASO platform lowers CMC costs vs. biologics; tau antibodies higher |
Critical path item: Demonstrating that synaptic gene downregulation is a primary driver (not consequence) of tau pathology requires prospective longitudinal modeling in early-stage AD (preclinical or prodromal).
---
Revised Confidence: 0.58
| Target | Modality Class | Precedent | Risk-Adjusted Tractability |
|--------|---------------|-----------|---------------------------|
| PDGFRα | Tyrosine kinase inhibitors (e.g., imatinib-class) | Imatinib does not cross BBB meaningfully | Low BBB penetration; PDGFRα antagonists in oncology lack CNS indication |
| LINGO1 | Antibodies, small molecules | Anti-LINGO1 (Biogen) failed Ph2 for MS (2016) | Prior clinical failure reduces enthusiasm; AD-specific mechanism unclear |
| Myelin genes (MBP, PLP1) | Transcription factors (e.g., MYRF) | Not yet targeted | Low confidence as direct drivers |
| Cholesterol biosynthesis (SREBP) | SREBP inhibitors | No approved CNS drugs | Off-target steroidogenesis risks |
Assessment: PDGFRα signaling is the most credible target given OPC survival dependence. However, BBB penetration is the primary bottleneck. LINGO1 failure in MS is cautionary but does not preclude AD-specific utility. Myelin gene targets are downstream effectors.
| Category | Strength | Gaps |
|----------|----------|------|
| Fluid biomarkers | No established OPC/myelin fluid biomarker | Serum NfL (axonal damage); CSF MBP (myelin degradation, but PMI-sensitive); no OPC-specific marker |
| Imaging | MWM (magnetization transfer ratio) for myelin integrity; DTI (diffusion tensor imaging) for white matter | Cannot resolve OPC vs. mature oligodendrocyte dysfunction |
| Model systems | hiPSC-derived OPCs (gold standard); mouse cuprizone model (demyelination/remyelination) | Cuprizone does not model AD; species OPC differences substantial; human OPC xenograft in shiverer mice is technically demanding |
| In vitro readouts | OPC differentiation (MBP+ myelin sheets); myelination co-culture with neurons | Not standardized across labs; readouts are morphological, not functional |
Assessment: Major biomarker gap—OPCs have no validated blood/CSF marker. MRI can assess white matter integrity but cannot establish OPC-specific dysfunction. Model systems exist but are not AD-specific.
1. No validated OPC/myelin biomarker for patient selection: Cannot enrich for patients with OPC dysfunction; would require broad AD enrollment with post-hoc subanalysis.
2. Early-stage intervention assumption: Myelin defects may precede neuronal loss, but this requires prodromal/ preclinical AD enrollment (challenging and costly).
3. Endpoint challenges: Myelin integrity (MWM, DTI) requires 2–3 year trials; cannot use standard cognitive endpoints alone.
4. Regulatory uncertainty: No approved myelin-protective AD therapy; pathway would be novel.
| Risk | Severity | Mitigation |
|------|----------|------------|
| On-target toxicity (PDGFRα) | Moderate | Cancer risk with PDGFRα inhibition; dose-limiting toxicity likely |
| Off-target immunosuppression | Moderate | OPCs require immune microenvironment; broad immunosuppression adverse |
| Myelin dysregulation | Low | Myelin remodeling is ongoing; acute effects unlikely to be severe |
| Developmental phenotypes | Unknown | PDGFRα knockout is embryonic lethal; caution in elderly AD patients |
| Milestone | Realistic Estimate | Notes |
|-----------|------------------|-------|
| Preclinical/IND-enabling | 3–4 years | Major gap: no OPC-specific biomarker; must develop Companion diagnostic in parallel |
| Phase I | 1–2 years | Safety-focused; biomarker development continues |
| Phase II | 3–4 years | Requires myelin imaging endpoint; long trial duration; n~300 |
| Phase III | Not predictable | No regulatory precedent; may require 5+ year trials |
| Total to NDA | 10–15 years (if pursued) | Requires biomarker validation; high attrition risk |
Assessment: This hypothesis is premature for clinical development without biomarker validation. The timeline exceeds typical program horizons. Best path forward: academic/ foundational work to identify OPC-specific fluid biomarker (e.g., surface antigen signature) before industry engagement.
---
Revised Confidence: 0.62
| Target | Modality Class | Precedent | Risk-Adjusted Tractability |
|--------|---------------|-----------|---------------------------|
| TYROBP (DAP12) | Adaptor protein; currently undrugged | No small molecules; antibodies unlikely to penetrate cells | Low tractability; signaling adaptor without enzymatic domain |
| CSF1R | Kinase inhibitors (e.g., pexidartinib) | Pexidartinib approved for TGCT; brain penetration unknown | Partial microglial depletion risk; therapeutic index narrow |
| TYROBP downstream (TREM2-independent) | Unknown | N/A | Cannot drug unknown targets |
| MERTK/AXL | Agonists or antagonists | AXL inhibitors in oncology (multiple); no CNS indication | AXL as AD target plausible; TAM receptor biology complex |
| APOE (via microglia) | Gene therapy, ASOs | APOE4 silencing in preclinical; AAV-APOE2 in IND-enabling studies | High tractability; APOE4->E2 conversion approach advanced |
Assessment: The TREM2-independent component is not druggable because it is not molecularly defined. Best practical strategy: drug TREM2 (which is tractable) to define the TREM2-dependent fraction; the residual "TREM2-independent" states become the therapeutic gap to address via alternative pathways (CSF1R, APOE, AXL).
| Category | Strength | Gaps |
|----------|----------|------|
| Fluid biomarkers | NfL, GFAP, YKL-40 (astrocyte/microglia activation) | No microglial state-specific blood marker; TSPO VAF polymorphism complicates PET |
| Imaging | TSPO-PET (TSPO radioligands available); P2RY12 PET ligands emerging | TSPO nonspecific; cannot resolve DAM vs. IRM states |
| Model systems | hiPSC-derived microglia (iMG); brain organoid co-cultures; humanized mouse models | iMG faithfully reproduce human states; mouse DAM states differ; interspecies microglial identity differences |
| In vitro readouts | Phagocytosis (Aβ, synaptosomes); cytokine release; chemotaxis | Functional readouts exist but state-specific interpretation limited |
Assessment: TSPO-PET is available for microglial activation but lacks state specificity. iMG systems are the strongest model. The field urgently needs a microglial state blood biomarker.
1. Microglial state biomarkers lacking: Cannot enroll patients based on "TREM2-independent disease state." Would require post-mortem confirmation (not feasible) or surrogate imaging (nonspecific).
2. BBB penetration mandatory: Microglial targets (TYROBP, TAM receptors) require CNS-penetrant agents; most kinase inhibitors do not achieve therapeutic brain concentrations.
3. Temporal targeting: Microglial states likely shift across disease stages; single intervention may not address the full trajectory.
4. Regulatory pathway: No approved microglial-modulating AD therapy; de novo pathway.
| Risk | Severity | Mitigation |
|------|----------|------------|
| Microglial depletion | High | CSF1R inhibitors deplete microglia
{
"ranked_hypotheses": [
{
"title": "Excitatory Neuron Synaptic Dysfunction and Mitochondrial Stress via MAPT (tau)",
"description": "Deep layer (L5/6) and superficial layer (L2/3) excitatory neurons demonstrate the most pronounced transcriptomic vulnerability in SEA-AD, characterized by synaptic gene downregulation (SNAP25, SYT1, SLC17A7), stress response upregulation (HSPA1B, DNAJB1), and mitochondrial dysfunction signatures. MAPT (tau) emerges as the primary upstream driver with established Phase I-ready ASO and antibody modalities. Layer-specific markers (RORB, THEMIS) provide spatial targeting guidance for delivery strategies.",
"target_gene": "MAPT",
"dimension_scores": {
"evidence_strength": 0.75,
"novelty": 0.65,
"feasibility": 0.88,
"therapeutic_potential": 0.85,
"mechanistic_plausibility": 0.78,
"druggability": 0.90,
"safety_profile": 0.70,
"competitive_landscape": 0.60,
"data_availability": 0.88,
"reproducibility": 0.75
},
"composite_score": 0.79,
"evidence_for": [
{"claim": "SEA-AD prefrontal cortex analysis of 1.2 million nuclei shows excitatory neuron transcriptional changes", "pmid": "SEA-AD-2022"},
{"claim": "tau ASO BIIB080 in Phase 1; anti-tau antibodies have established regulatory pathway", "pmid": "multiple clinical trials"},
{"claim": "Synaptic gene downregulation correlates with Braak stage progression", "pmid": "30818991"}
],
"evidence_against": [
{"claim": "Cross-sectional data cannot establish temporal causality; mitochondrial changes may be nonspecific stress response", "pmid": "methodological critique"},
{"claim": "Layer 5/6 specificity contradicted by entorhinal cortex Layer II vulnerability", "pmid": "regional specificity concern"},
{"claim": "RORB/THEMIS are markers, not mechanistic drivers", "pmid": "marker vs driver conflation"}
]
},
{
"title": "Microglial Disease-Associated States: TREM2-Independent Pathways Driving Neuroinflammation",
"description": "SEA-AD v4 identifies multiple microglial states (DAM, IRM, ARM) where a substantial TREM2-independent fraction drives pathology. While TYROBP (DAP12) signaling is currently undruggable as an adaptor, APOE-mediated pathways and TAM receptor (MERTK/AXL) modulation represent tractable TREM2-independent therapeutic entry points. TSPO-PET imaging provides population-level monitoring, and iPSC-derived microglia faithfully reproduce human states for drug screening.",
"target_gene": "APOE",
"dimension_scores": {
"evidence_strength": 0.68,
"novelty": 0.72,
"feasibility": 0.75,
"therapeutic_potential": 0.78,
"mechanistic_plausibility": 0.65,
"druggability": 0.72,
"safety_profile": 0.68,
"competitive_landscape": 0.75,
"data_availability": 0.70,
"reproducibility": 0.62
},
"composite_score": 0.71,
"evidence_for": [
{"claim": "Multiple microglial states identified in human brain with disease-associated transcriptional signatures", "pmid": "28607169"},
{"claim": "APOE4 silencing in preclinical IND-enabling studies; AAV-APOE2 conversion approach advanced", "pmid": "ongoing preclinical"},
{"claim": "iPSC-derived microglia reproduce human disease states for functional screening", "pmid": "emerging literature"}
],
"evidence_against": [
{"claim": "DAM/ARM/IRM taxonomy is descriptive, not mechanistic; states may be continuum rather than discrete", "pmid": "32109258"},
{"claim": "TREM2-independent fraction not molecularly characterized; cannot be targeted without definition", "pmid": "mechanistic gap"},
{"claim": "Mouse microglial states poorly translate to humans", "pmid": "translational concern"}
]
},
{
"title": "Astrocyte Reactivity Heterogeneity with APOE4-Dependent Vulnerability",
"description": "Disease-specific astrocyte states (distinct from classical A1/A2 paradigm) show compartmentalized responses with APOE4 carriers displaying exacerbated reactivity signatures. Reactive astrocytes downregulate glutamate transporters (SLC1A2/EAAT2) and upregulate GFAP/C3 in subsets, but many show non-classical disease-associated states. APOE4->E2 conversion via astrocyte-targeted AAV represents the most tractable therapeutic approach, though the A1/A2 framework validity is contested.",
"target_gene": "APOE",
"dimension_scores": {
"evidence_strength": 0.65,
"novelty": 0.70,
"feasibility": 0.72,
"therapeutic_potential": 0.75,
"mechanistic_plausibility": 0.60,
"druggability": 0.78,
"safety_profile": 0.72,
"competitive_landscape": 0.80,
"data_availability": 0.68,
"reproducibility": 0.55
},
"composite_score": 0.69,
"evidence_for": [
{"claim": "APOE4 carriers show exacerbated astrocyte reactivity signatures in SEA-AD", "pmid": "SEA-AD-2022"},
{"claim": "AAV-APOE2 conversion in humanized APOE mice showing functional improvement", "pmid": "preclinical IND"},
{"claim": "Escartin et al. nomenclature paper calls for refined disease-associated state definitions", "pmid": "2021 nomenclature"}
],
"evidence_against": [
{"claim": "A1/A2 paradigm not replicated with rigorous functional validation; GFAP marker limitations", "pmid": "28916532 critique"},
{"claim": "EAAT2 downregulation may be compensatory; therapeutic restoration could be harmful", "pmid": "EAAT2 KO studies"},
{"claim": "APOE4 effects are largely non-cell-autonomous; isolating astrocyte-specific effects difficult", "pmid": "systemic effects"}
]
},
{
"title": "Oligodendrocyte Lineage Vulnerability: Early Myelination Disruption with Blocked Differentiation",
"description": "OPCs and oligodendrocytes represent early-affected lineages with increased proliferation markers but blocked differentiation, downregulation of myelin-related genes (MBP, MOG, PLP1), and stress/immune gene upregulation. PDGFRα signaling is the most credible target for OPC survival, but BBB-penetrant PDGFRα antagonists do not exist. Critical gap: no OPC-specific fluid biomarker exists for patient selection, rendering this hypothesis premature for clinical development (10-15 year NDA timeline).",
"target_gene": "PDGFRα",
"dimension_scores": {
"evidence_strength": 0.55,
"novelty": 0.75,
"feasibility": 0.35,
"therapeutic_potential": 0.60,
"mechanistic_plausibility": 0.55,
"druggability": 0.40,
"safety_profile": 0.50,
"competitive_landscape": 0.70,
"data_availability": 0.50,
"reproducibility": 0.45
},
"composite_score": 0.53,
"evidence_for": [
{"claim": "Longitudinal snRNA-seq shows early OPC changes in ADNI cohort", "pmid": "36735998"},
{"claim": "Myelination changes documented in AD patients clinically", "pmid": "Bartzokis"},
{"claim": "iPSC-derived OPC differentiation assays available for target validation", "pmid": "OPC protocols"}
],
"evidence_against": [
{"claim": "Post-mortem interval severely confounds oligodendrocyte RNA quality", "pmid": "PMI artifacts"},
{"claim": "No OPC-specific fluid biomarker; patient selection impossible", "pmid": "biomarker gap"},
{"claim": "LINGO1 antibody failed Phase 2 for MS; AD-specific mechanism unclear", "pmid": "clinical failure"}
]
},
{
"title": "Inhibitory Neuron Subtype Loss: Excitation/Inhibition Imbalance Hypothesis",
"description": "Specific inhibitory neuron subtypes (PVALB+, SST+) show selective transcriptomic vulnerability correlating with cortical hyperexcitability (seizures in AD) and early cognitive dysfunction. GABA synthesis enzymes (GAD1, GAD2) and Nav1.1/SCN1A represent candidate targets, but transcriptomic downregulation has not been validated as actual cell death. Regional specificity (hippocampal CA1 vs. prefrontal cortex) and inconsistent evidence across studies represent major translational barriers.",
"target_gene": "GAD1/GAD2",
"dimension_scores": {
"evidence_strength": 0.50,
"novelty": 0.60,
"feasibility": 0.40,
"therapeutic_potential": 0.55,
"mechanistic_plausibility": 0.52,
"druggability": 0.55,
"safety_profile": 0.45,
"competitive_landscape": 0.65,
"data_availability": 0.55,
"reproducibility": 0.48
},
"composite_score": 0.52,
"evidence_for": [
{"claim": "PVALB+ and SST+ transcript downregulation in AD prefrontal cortex", "pmid": "SEA-AD-2022"},
{"claim": "Excitation/inhibition imbalance linked to seizures in AD mouse models", "pmid": "26779885"},
{"claim": "Human brain interneuron atlas available for subtype mapping", "pmid": "Hu et al."}
],
"evidence_against": [
{"claim": "Transcriptional downregulation not validated as actual cell loss; stereological counting lacking", "pmid": "protein validation gap"},
{"claim": "PVALB+ vulnerability inconsistent across AD models and brain regions", "pmid": "inconsistent literature"},
{"claim": "GABAergic drugs worsen cognitive outcomes in AD trials", "pmid": "clinical trials"}
]
},
{
"title": "Vascular and Perivascular Cell Type Vulnerability: BBB Integrity Disruption",
"description": "Endothelial cells and pericytes show AD-related transcriptional changes affecting blood-brain barrier integrity, including tight junction gene downregulation (CLDN5, OCLN), altered pericyte contractile gene expression, and upregulation of adhesion molecules (VCAM1, ICAM1). However, vascular cell RNA is severely degraded by post-mortem interval, tight junction gene downregulation is a known PMI artifact, and BBB dysfunction is inconsistent across AD patients. Major confounding limits interpretability.",
"target_gene": "CLDN5",
"dimension_scores": {
"evidence_strength": 0.45,
"novelty": 0.55,
"feasibility": 0.35,
"therapeutic_potential": 0.50,
"mechanistic_plausibility": 0.48,
"druggability": 0.42,
"safety_profile": 0.55,
"competitive_landscape": 0.60,
"data_availability": 0.40,
"reproducibility": 0.38
},
"composite_score": 0.47,
"evidence_for": [
{"claim": "Pericyte loss demonstrated in AD mouse models", "pmid": "Nelson 2016"},
{"claim": "Vascular dysfunction review documenting BBB changes in AD", "pmid": "29516877"},
{"claim": "SEA-AD vascular cell type transcriptomic data available", "pmid": "SEA-AD-2022"}
],
"evidence_against": [
{"claim": "Endothelial/pericyte RNA highly sensitive to PMI; CLDN5 downregulation is artifact", "pmid": "PMI confounds"},
{"claim": "BBB dysfunction not universally observed in AD patients", "pmid": "Sweeney 2018"},
{"claim": "CLDN5 genetic variants do not show strong AD risk associations", "pmid": "genetic studies"}
]
},
{
"title": "Tripartite Synapse Cell Type-Nonautonomous Crosstalk: Coordinated Failure",
"description": "Integrative SEA-AD analysis reveals coordinated failure of tripartite synapse maintenance, where neuronal synaptic gene downregulation correlates with astrocyte phagocytic receptor upregulation and microglial synaptic pruning gene alterations. Complement cascade (C1Q, C3) and TAM receptors (MERTK, AXL) represent crosstalk nodes. However, 'tripartite synapse' is a conceptual model, transcriptional correlations do not establish functional crosstalk, and C1q/C3 roles are context-dependent with unclear therapeutic direction.",
"target_gene": "C1Q",
"dimension_scores": {
"evidence_strength": 0.48,
"novelty": 0.68,
"feasibility": 0.38,
"therapeutic_potential": 0.58,
"mechanistic_plausibility": 0.52,
"druggability": 0.50,
"safety_profile": 0.40,
"competitive_landscape": 0.70,
"data_availability": 0.45,
"reproducibility": 0.42
},
"composite_score": 0.51,
"evidence_for": [
{"claim": "Complement-mediated synaptic pruning demonstrated in development and pathology", "pmid": "27762320"},
{"claim": "Correlative transcriptional changes across neuron-astrocyte-microglia in SEA-AD", "pmid": "SEA-AD-2022"},
{"claim": "Anti-C1q trials (Anakynra) provide clinical pathway for validation", "pmid": "clinical trials"}
],
"evidence_against": [
{"claim": "Correlative evidence only; transcriptomic correlations do not establish mechanistic crosstalk", "pmid": "correlation vs causation"},
{"claim": "C1q/C3 roles are context-dependent (development vs. pathology) with unclear therapeutic direction", "pmid": "context dependence"},
{"claim": "Tripartite synapse is conceptual model, not demonstrated biological entity", "pmid": "paradigm limitations"}
]
}
],
"knowledge_edges": [
{"source_id": "H1", "source_type": "hypothesis", "target_id": "MAPT", "target_type": "gene", "relation": "primary_upstream_driver"},
{"source_id": "H1", "source_type": "hypothesis", "target_id": "RORB", "target_type": "gene", "relation": "layer_specificity_marker"},
{"source_id": "H1", "source_type": "hypothesis", "target_id": "SNAP25", "target_type": "gene", "relation": "downstream_synaptic_effector"},
{"source_id": "H1", "source_type": "hypothesis", "target_id": "Mitochondria", "target_type": "pathway", "relation": "dysfunctional_pathway"},
{"source_id": "H2", "source_type": "hypothesis", "target_id": "PDGFRα", "target_type": "gene", "relation": "primary_target"},
{"source_id": "H2", "source_type": "hypothesis", "target_id": "LINGO1", "target_type": "gene", "relation": "negative_regulator"},
{"source_id": "H2", "source_type": "hypothesis", "target_id": "MBP", "target_type": "gene", "relation": "downregulated_myelin_gene"},
{"source_id": "H3", "source_type": "hypothesis", "target_id": "APOE", "target_type": "gene", "relation": "primary_target"},
{"source_id": "H3", "source_type": "hypothesis", "target_id": "TYROBP", "target_type": "gene", "relation": "signaling_adaptor"},
{"source_id": "H3", "source_type": "hypothesis", "target_id": "CSF1R", "target_type": "gene", "relation": "microglial_survival"},
{"source_id": "H3", "source_type": "hypothesis", "target_id": "MERTK", "target_type": "gene", "relation": "alternative_target"},
{"source_id": "H4", "source_type": "hypothesis", "target_id": "GAD1", "target_type": "gene", "relation": "GABA_synthesis"},
{"source_id": "H4", "source_type": "hypothesis", "target_id": "SCN1A", "target_type": "gene", "relation": "candidate_target"},
{"source_id": "H5", "source_type": "hypothesis", "target_id": "APOE", "target_type": "gene", "relation": "primary_target"},
{"source_id": "H5", "source_type": "hypothesis", "target_id": "SLC1A2", "target_type": "gene", "relation": "downregulated_transporter"},
{"source_id": "H5", "source_type": "hypothesis", "target_id": "GFAP", "target_type": "gene", "relation": "reactivity_marker"},
{"source_id": "H6", "source_type": "hypothesis", "target_id": "CLDN5", "target_type": "gene", "relation": "tight_junction_integrity"},
{"source_id": "H6", "source_type": "hypothesis", "target_id": "PDGFRB", "target_type": "gene", "relation": "pericyte_function"},
{"source_id": "H7", "source_type": "hypothesis", "target_id": "C1Q", "target_type": "gene", "relation": "complement_pruning"},
{"source_id": "H7", "source_type": "hypothesis", "target_id": "MERTK", "target_type": "gene", "relation": "phagocytic_clearance"},
{"source_id": "H7", "source_type": "hypothesis", "target_id": "SYP", "target_type": "gene", "relation": "synaptic_downstream"},
{"source_id": "H1", "source_type": "hypothesis", "target_id": "H3", "target_type": "hypothesis", "relation": "shares_tau_dependency"},
{"source_id": "H3", "source_type": "hypothesis", "target_id": "H5", "target_type": "hypothesis", "relation": "shares_APOE_focus"},
{"source_id": "H4", "source_type": "hypothesis", "target_id": "H1", "target_type": "hypothesis", "relation": "interneuron_excitation_balance"},
{"source_id": "H7", "source_type": "hypothesis", "target_id": "H1", "target_type": "hypothesis", "relation": "downstream_convergence"},
{"source_id": "H7", "source_type": "hypothesis", "target_id": "H3", "target_type": "hypothesis", "relation": "microglial_pruning_component"}
],
"synthesis_summary": "The SEA-AD v4 debate converges on a three-tier hierarchy of cell type vulnerability in Alzheimer's disease. The strongest evidence supports excitatory neuron vulnerability (layer 2/3 and 5/6) driven by tau pathology (MAPT), which achieves the highest composite score (0.79) by combining robust transcriptomic evidence, Phase I-ready druggability (tau ASOs and antibodies), and established biomarker panels (p-tau217, NfL, SNAP25). Microglial disease-associated states represent the second tier, with TREM2-independent pathways offering tractable targets (APOE4 conversion, TAM receptors) despite the mechanistic vagueness of state definitions and lack of state-specific biomarkers. Astrocyte reactivity and oligodendrocyte lineage vulnerability form the third tier—promising but requiring biomarker validation (OPCs) or paradigm refinement (A1/A2) before clinical translation. The critical methodological constraints across all hypotheses are survival bias (neurons may be underrepresented), PMI artifacts (especially for vascular cells and oligodendrocytes), and the fundamental limitation that cross-sectional post-mortem data cannot resolve temporal causality. Hypotheses 1 (MAPT), 3 (APOE via microglia), and 5 (APOE via astrocytes) share APOE/tau mechanistic axes, suggesting combined targeting strategies may be required for maximal therapeutic benefit."
}