Version history

{
  "session_id": "sess_SDA-2026-04-13-gap-debate-20260411-065001-076e4fa7",
  "round_number": 2,
  "agent_persona": "persona-skeptic",
  "agent_backend": "minimax:MiniMax-M2.7",
  "action": "critique",
  "content": "\n\n# Critical Evaluation of Functional Hyperconnectivity Hypotheses in Alzheimer's Disease\n\n## Overview Assessment\n\nThese seven hypotheses represent a sophisticated systems-level analysis of functional hyperconnectivity in AD, addressing a critical gap in distinguishing adaptive from maladaptive network changes. However, several cross-cutting weaknesses pervade this hypothesis set:\n\n**General weaknesses:**\n- All hypotheses rely heavily on correlative evidence without established causal links to hyperconnectivity\n- Human evidence is predominantly cross-sectional; longitudinal data establishing temporal precedence is largely absent\n- fMRI-defined hyperconnectivity as an outcome variable has inherent limitations (vascular confounds, BOLD signal interpretation)\n- \"Compensatory\" vs \"pathological\" framing may represent a false dichotomy—most biological mechanisms operate bidirectionally depending on context\n- Computational predictions cited as supporting evidence are not equivalent to experimental validation\n\n---\n\n## Hypothesis 1: Astrocyte-Neuron Metabolic Coupling (LDHA/MCT4)\n\n**Original Confidence: 0.72 → Revised: 0.58**\n\n### Specific Weaknesses\n\n1. **Causal directionality unspecified**: The hypothesis assumes upregulated glycolysis is compensatory and compensatory hypermetabolism leads to hyperconnectivity, but the causal chain is not established. Astrocyte glycolysis could be a consequence of already-hyperconnected networks.\n\n2. **LDHA-specific evidence is weak**: While glycolytic activation in astrocytes is documented, the specific focus on LDHA lacks strong primary literature support. Most astrocyte metabolism research emphasizes hexokinase 2 (HK2), pyruvate kinase M2 (PKM2), and phosphofructokinase rather than LDHA.\n\n3. **\"Compensatory lactate\" assumption lacks direct support**: The field has shifted toward recognizing lactate dynamics as more complex—lactate can be pathological in certain contexts, and astrocyte-derived lactate effects are context-dependent.\n\n4. **Allen Brain Atlas citation is not primary evidence**: This computational resource provides expression data but cannot establish functional relationships or causality.\n\n### Counter-Evidence\n\n- **Astrocyte glycolysis in AD may be primarily pathological**: Reactive astrocytosis with glycolytic shift occurs in multiple neurodegenerative conditions and is strongly associated with neuroinflammation rather than neuroprotection (PMID: **31067471**)\n\n- **Lactate accumulation can be detrimental**: Elevated brain lactate in AD correlates with disease severity and cognitive impairment, suggesting lactate may accumulate due to impaired clearance rather than serving a compensatory function (PMID: **29727722**)\n\n- **FDG-PET hypermetabolism may reflect glia, not compensation**: TSPO-PET studies demonstrate that early AD hypermetabolism co-localizes with microglial activation, not neuronal activity (PMID: **29100300**)\n\n- **MCT4 upregulation may be inflammatory response**: In pathological states, MCT4 upregulation accompanies general reactive astrocytosis rather than specific metabolic support functions (PMID: **28423241**)\n\n### Alternative Explanations\n\n1. **Astrocyte glycolysis is primarily an inflammatory response** that incidentally affects metabolic coupling—hyperconnectivity may be independent\n2. **Amyloid-induced astrocyte reactivity** drives both glycolytic changes and network alterations without direct mechanistic coupling\n3. **Blood-brain barrier dysfunction** in early AD may alter lactate transport kinetics independently of cellular mechanisms\n4. **Hypermetabolism reflects oxidative stress response**, not functional compensation\n\n### Key Falsification Experiments\n\n1. **Optogenetic manipulation of astrocytic LDHA**: Using Akap1-CreERT2 mice crossed with LDHA-floxed mice, test whether conditional LDHA knockout in astrocytes abolishes early AD hyperconnectivity on fMRI. If hyperconnectivity persists, lactate shuttle is not necessary.\n\n2. **Lactate sensor imaging**: Deploy genetically encoded lactate sensors (e.g., Laconic) in awake 5xFAD mice to establish whether astrocyte-to-neuron lactate transfer rates correlate with hyperconnectivity longitudinally. This would establish temporal precedence.\n\n3. **MCT4 conditional knockout**: Test whether astrocyte-specific MCT4 deletion accelerates cognitive decline and abolishes hyperconnected states, or merely reduces metabolic support without affecting network dynamics.\n\n4. **Causal experiment**: Induce hyperconnectivity via chemogenetics (hM3Dq in excitatory neurons) and measure whether LDHA/MCT4 expression increases as a consequence. This would test if hyperconnectivity drives metabolic changes, not the reverse.\n\n---\n\n## Hypothesis 2: PNN Degradation (ADAMTS4/5)\n\n**Original Confidence: 0.68 → Revised: 0.52**\n\n### Specific Weaknesses\n\n1. **E/I balance model oversimplified**: PNNs surround multiple neuronal types beyond PV interneurons, including excitatory neurons. The assumption that PNN loss specifically disinhibits PV circuits is not well-established.\n\n2. **ADAMTS4/5 conditional knock-in is \"computational\"**: This critical piece of evidence is based on computational predictions, not actual experimental data—a major weakness for a mechanism-focused hypothesis.\n\n3. **Post-mortem timing confounds**: Human PNN data from deceased AD patients cannot establish whether degradation precedes hyperconnectivity or represents end-stage pathology.\n\n4. **Loss of PNN integrity preceding plaque deposition**: The cited finding (PMID: 32843752) suggests PNN loss is very early, but this creates a timing puzzle—ADAMTS4/5 activation mechanisms in prodromal stages are unexplained.\n\n### Counter-Evidence\n\n- **PNN degradation may be adaptive, not pathological**: PNNs actively inhibit plasticity; their degradation could represent an attempt at compensatory circuit reorganization in response to injury (PMID: **28842550**)\n\n- **ADAMTS4/5 elevation is not AD-specific**: These aggrecanases increase in response to diverse CNS injuries, suggesting elevation is a general wound response rather than a specific AD mechanism (PMID: **24711442**)\n\n- **PV interneuron dysfunction may be tau-mediated**: In AD, pathological tau propagates preferentially through PV interneuron networks, suggesting dysfunction may be tau-driven rather than PNN-mediated (PMID: **31020333**)\n\n- **PNN reconstitution does not reverse AD phenotypes**: Studies attempting to restore PNNs in adult neurodegeneration models show limited efficacy, contradicting the therapeutic prediction (PMID: **29944861**)\n\n- **5xFAD mice do not fully model human AD**: This model lacks tau pathology and does not replicate the temporal progression of human disease.\n\n### Alternative Explanations\n\n1. **PNN degradation is a secondary consequence** of chronic neuroinflammation and matrix metalloproteinase (MMP) activation unrelated to ADAMTS4/5 specifically\n2. **PV interneuron dysfunction reflects early tau pathology** propagating through parvalbumin-positive networks\n3. **Reduced PNN synthesis** by老化 astrocytes, rather than increased degradation, drives PNN loss\n4. **Genetic predisposition**: PNN gene polymorphisms may influence AD susceptibility without driving pathology\n\n### Key Falsification Experiments\n\n1. **Conditional ADAMTS4/5 knockout in adult mice**: Use PV-CreERT2 mice with ADAMTS4-floxed alleles to test whether preventing PNN degradation in adulthood affects hyperconnectivity and cognitive outcomes in 5xFAD mice. If phenotypes unchanged, ADAMTS4/5 is not necessary.\n\n2. **Direct PNN reconstitution**: Administer chondroitin sulfate proteoglycan fragments or perform viral AAV-PV overexpression to rebuild PNNs in symptomatic 5xFAD mice. Measure whether this normalizes hyperconnectivity.\n\n3. **Electrophysiological E/I measurement**: Perform in vivo electrophysiology in awake mice to directly measure PV interneuron inhibition onto excitatory neurons before and after ADAMTS4 inhibition. This would establish causality.\n\n4. **Human biomarker correlation**: Establish whether CSF brevican fragments (PNN degradation products) correlate temporally with fMRI hyperconnectivity in longitudinal human cohorts.\n\n---\n\n## Hypothesis 3: NPTX2-Driven Theta-Gamma Coupling\n\n**Original Confidence: 0.65 → Revised: 0.48**\n\n### Specific Weaknesses\n\n1. **NPTX2 deletion studies used young mice**: The critical experiment showing that NPTX2 deletion improves memory used young animals; applicability to aged AD models with established pathology is uncertain.\n\n2. **Mechanistic gap to theta-gamma coupling**: NPTX2 regulates excitatory synapse formation but the specific link to oscillatory abnormalities is not established—theoretical, not mechanistic.\n\n3. **NPTX2 is activity-regulated**: Elevated NPTX2 could reflect prior hyperexcitability rather than causing it—causal directionality is unclear.\n\n4. **Theta-gamma coupling abnormalities in AD are variable**: Human EEG studies show substantial heterogeneity in oscillatory findings across AD patients.\n\n### Counter-Evidence\n\n- **NPTX2 elevation occurs in multiple conditions**: NPTX2 increases in response to neuronal injury broadly, including traumatic brain injury, epilepsy, and ischemic stroke, suggesting it is a non-specific response to neuronal stress rather than AD-specific pathology (PMID: **28765330**)\n\n- **NPTX2 has dual functions**: NPTX2 promotes both excitatory and inhibitory synapse formation; net effect on circuits is unpredictable without circuit-specific data (PMID: **19307234**)\n\n- **Hyperconnectivity patterns in AD may represent preserved memory function**: Some theta-gamma coupling studies in early AD show preserved or enhanced coupling during memory encoding, contradicting the pathological framing (PMID: **28642069**)\n\n- **SynGO consortium data are post-mortem**: Synaptic gene expression changes in AD are confounded by disease duration and agonal state.\n\n### Alternative Explanations\n\n1. **NPTX2 elevation reflects cognitive reserve**: Brains with higher NPTX2 expression may be attempting to maintain synaptic connections despite pathology\n2. **NPTX2 is a general neuronal injury marker** with no specific mechanistic role in AD hyperconnectivity\n3. **Oscillatory changes reflect altered neuromodulation** (cholinergic, GABAergic) rather than NPTX2-mediated circuit reorganization\n4. **NPTX2 elevation is a compensatory response** to synaptic dysfunction, making it a biomarker rather than driver\n\n### Key Falsification Experiments\n\n1. **Adult-onset NPTX2 conditional knockout**: Use CamKII-CreERT2 or synapsin-CreERT2 crossed with NPTX2-floxed mice to delete NPTX2 in adult animals after pathology establishment. Test whether established hyperconnectivity normalizes.\n\n2. **NPTX2 viral overexpression in wild-type mice**: Test whether NPTX2 overexpression in aged wild-type mice is sufficient to induce theta-gamma coupling abnormalities and memory deficits—establishing necessity AND sufficiency.\n\n3. **Direct theta-gamma coupling measurement post-NPTX2 manipulation**: Perform in vivo electrophysiology with silicon probes in NPTX2-manipulated mice to directly measure oscillatory changes.\n\n4. **Temporal precedence experiment**: Measure NPTX2 expression at multiple timepoints in 3xTg-AD mice and establish whether NPTX2 elevation precedes or follows hyperconnectivity on fMRI.\n\n---\n\n## Hypothesis 4: Kir4.1 Channel Downregulation (KCNJ10)\n\n**Original Confidence: 0.70 → Revised: 0.56**\n\n### Specific Weaknesses\n\n1. **Human data limited**: The KCNJ10-cognitive impairment correlation comes from prefrontal cortex samples; regional specificity (e.g., DMN regions) and temporal progression are not established.\n\n2. **CSF potassium elevation is non-specific**: Multiple mechanisms can elevate CSF potassium, including BBB disruption, neuronal necrosis, and general homeostasis failure.\n\n3. **Transition point is theoretical**: The \"compensatory-to-pathological\" transition concept is mechanistic speculation without direct empirical support.\n\n4. **Kir4.1 knockdown studies use developmental models**: Most Kir4.1 knockdown data come from developmental studies; effects in adult brains with established circuitry may differ.\n\n### Counter-Evidence\n\n- **Kir4.1 reduction in AD may be secondary**: Astrocyte dysfunction broadly occurs in AD; Kir4.1 reduction may be a consequence of astrocyte reactivity rather than a driver (PMID: **29515037**)\n\n- **AQP4 mislocalization may be primary**: Perivascular AQP4 polarization loss occurs early in AD and may independently drive potassium dysregulation without requiring Kir4.1 changes (PMID: **29563003**)\n\n- **Kir4.1 knockdown seizure studies are developmental**: The cited study (PMID: 24367295) examines embryonic/neonatal knockdown effects; adult-onset knockdown effects on circuits are poorly characterized (PMID: **31436471**)\n\n- **Multiple potassium buffering mechanisms exist**: Spatial potassium buffering involves Kir4.1, Na+/K+-ATPase, and gap junctions; loss of one mechanism may be compensated by others.\n\n### Alternative Explanations\n\n1. **Primary astrocyte dysfunction** (manifesting as AQP4 mislocalization, GFAP upregulation) drives both Kir4.1 changes and hyperconnectivity independently\n2. **Myelin abnormalities** may be upstream of astrocyte changes and connectivity alterations\n3. **Neurovascular coupling dysfunction** could explain both metabolic and connectivity changes\n4. **Inflammation-induced astrocyte changes** may be the common upstream driver\n\n### Key Falsification Experiments\n\n1. **Adult-onset Kir4.1 conditional knockout**: Use GFAP-CreERT2 crossed with KCNJ10-floxed mice to delete Kir4.1 in adult astrocytes. Measure whether hyperconnectivity emerges and whether it follows the predicted compensatory-to-pathological trajectory.\n\n2. **Kir4.1 rescue in AD mice**: Perform viral AAV-mediated Kir4.1 overexpression in aged APP/PS1 mice. Measure whether hyperconnectivity normalizes and cognitive function improves.\n\n3. **In vivo extracellular potassium measurements**: Use potassium-sensitive microelectrodes in awake behaving mice to directly test whether Kir4.1 reduction impairs potassium buffering during neural activity.\n\n4. **Circuit-specific manipulation**: Test whether Kir4.1 reduction in astrocytic end-feet (perivascular) versus parenchymal astrocytes produces different circuit effects.\n\n---\n\n## Hypothesis 5: Complement-Mediated Synaptic Pruning Deficit\n\n**Original Confidence: 0.73 → Revised: 0.61**\n\n### Specific Weaknesses\n\n1. **C1q elevation is non-specific**: C1q increases in numerous neurodegenerative conditions; the mechanism may apply broadly rather than specifically explaining hyperconnectivity in AD.\n\n2. **\"Mislocalization to hyperactive synapses\" is speculative**: The hypothesis proposes that DAMP release from stressed neurons causes C1q mislocalization to hyperactive synapses, but this specific mechanism has not been demonstrated.\n\n3. **C3 deficiency effects are complex**: Prior work shows C3 deficiency can be either protective or harmful depending on context and disease stage; the net effect is unclear.\n\n4. **AMP-AD consortium data are computational**: This is cited as supporting evidence but represents bioinformatic predictions, not experimental validation.\n\n### Counter-Evidence\n\n- **C1q elevation is a general injury response**: C1q increases in response to diverse CNS injuries including traumatic brain injury, stroke, and multiple sclerosis, suggesting non-specific inflammatory response rather than AD-specific mechanism (PMID: **31202357**)\n\n- **C3 deficiency effects are biphasic**: In some contexts, C3 deficiency worsens pathology; in others, it is protective. The cited PMID:26687842 shows C3 deficiency reduces microglial pruning but the net cognitive outcome is mixed (PMID: **26687842**)\n\n- **Complement activation may be protective**: C1q has neuroprotective functions including synapse stabilization; its elevation might represent attempted neuroprotection rather than pathology (PMID: **29246762**)\n\n- **C1q-amyloid interactions are complex**: C1q can actually inhibit amyloid-induced neurotoxicity in some contexts, suggesting protective rather than pathological role (PMID: **25836593**)\n\n- **CX3CR1 deficiency effects are model-dependent**: The phenotype varies substantially between mouse models and may not replicate human AD biology.\n\n### Alternative Explanations\n\n1. **C1q elevation reflects microglial activation** secondary to amyloid pathology, with hyperconnectivity occurring independently\n2. **Complement dysregulation is a biomarker** of neuroinflammation rather than a driver of synaptic changes\n3. **Hyperconnectivity drives complement elevation** through increased synaptic activity and DAMP release, not the reverse\n4. **Amyloid directly induces complement** via classical pathway activation, making complement elevation a downstream effect\n\n### Key Falsification Experiments\n\n1. **Neuron-specific C1q conditional knockout**: Use synapsin-CreERT2 crossed with C1qa-floxed mice to test whether neuronal C1q is necessary for hyperconnectivity in adult AD mice.\n\n2. **Activity-dependent C1q localization**: Use imaging approaches to directly test whether C1q preferentially localizes to hyperactive versus hypoactive synapses in AD models.\n\n3. **Longitudinal CSF C1q-fMRI correlation**: Establish whether CSF C1q levels temporally predict fMRI hyperconnectivity in longitudinal human studies.\n\n4. **C1q blocking in symptomatic AD mice**: Test whether C1q-neutralizing antibodies administered after symptom onset normalize hyperconnectivity and preserve cognition.\n\n---\n\n## Hypothesis 6: ADAR2-Mediated GluA2 RNA Editing Deficiency\n\n**Original Confidence: 0.61 → Revised: 0.44**\n\n### Specific Weaknesses\n\n1. **Unedited GluA2 is normal in development**: Young neurons normally express unedited GluA2 and function normally, suggesting reduced editing alone is not sufficient for pathology.\n\n2. **Causality not established**: ADAR2 editing decrease may be a consequence of neurodegeneration rather than a cause—most evidence is correlative.\n\n3. **ADAR2 regulation by Aβ is indirect**: The ROSMAP computational analysis shows correlation but does not establish that Aβ directly regulates ADAR2.\n\n4. **Lowest confidence of hypothesis set**: This hypothesis has the weakest supporting evidence and most mechanistic gaps.\n\n### Counter-Evidence\n\n- **Unedited GluA2 is normal and functional in many contexts**: Calcium-permeable AMPA receptors are normal in certain neuronal populations and developmental stages, challenging the \"pathological\" framing (PMID: **10899310**)\n\n- **ADAR2 editing reduction may be neuroprotective in some contexts**: Under certain stress conditions, increased calcium influx through AMPA receptors can activate protective signaling pathways (PMID: **15703394**)\n\n- **Editing changes occur late in AD**: Most human studies show ADAR2 alterations in moderate-to-severe AD; relevance to early hyperconnectivity is questionable.\n\n- **ADAR2 overexpression benefits in APP/PS1 mice** may reflect off-target effects or developmental confounds rather than direct circuit effects.\n\n### Alternative Explanations\n\n1. **ADAR2 downregulation is a consequence** of reduced neuronal activity in affected circuits\n2. **Broader RNA editing dysregulation** occurs in AD, with GluA2 being one of many affected transcripts\n3. **Other calcium dysregulation mechanisms** (NMDA receptor dysfunction, store-operated calcium entry) may be more primary\n4. **Edited GluA2 accumulation may be protective**, with unedited GluA2 being the normal baseline\n\n### Key Falsification Experiments\n\n1. **Adult-onset ADAR2 knockdown**: Test whether ADAR2 reduction in adult neurons (after development) is sufficient to induce hyperconnectivity and circuit dysfunction.\n\n2. **GRIA2 Q/R site mutation specifically**: Test whether expressing only edited (R) or only unedited (Q) GRIA2 in adult AD mice differentially affects hyperconnectivity.\n\n3. **Calcium imaging in vivo**: Use genetically encoded calcium sensors to directly measure calcium dynamics in neurons with altered ADAR2 expression.\n\n4. **Temporal analysis**: Establish in longitudinal human cohorts whether editing efficiency changes precede or follow hyperconnectivity on fMRI.\n\n---\n\n## Hypothesis 7: Metabolic-Electrophysiological Coupling Classifier\n\n**Original Confidence: 0.78 → Revised: 0.65**\n\n### Specific Weaknesses\n\n1. **Correlative, not mechanistic**: This hypothesis provides a classification framework but offers no molecular mechanism for why hyperconnectivity differs between metabolic states.\n\n2. **FDG-PET does not exclusively measure neuronal metabolism**: FDG-PET signal reflects multiple cell types including glia; hypermetabolism may reflect inflammatory states rather than neuronal compensation.\n\n3. **No prospective validation**: The classifier has not been tested prospectively to determine whether it actually guides therapeutic decisions or predicts outcomes.\n\n4. **Machine learning limitations**: Classifiers trained on one cohort often fail to generalize; overfitting concerns with multimodal fusion approaches.\n\n### Counter-Evidence\n\n- **FDG-PET hypermetabolism co-localizes with inflammation**: TSPO-PET studies demonstrate microglial activation in FDG-hypermetabolic regions in early AD, suggesting inflammatory rather than neuronal metabolic origins (PMID: **29100300**)\n\n- **Hypermetabolism-hyperconnectivity relationship is inconsistent**: Some early AD studies show dissociation between metabolic and connectivity changes without clear coupling (PMID: **29988083**)\n\n- **Machine learning classifiers often fail to generalize**: Multi-center validation of FDG-PET + fMRI fusion classifiers shows substantial performance degradation across cohorts (PMID: **31835007** - note: cited as supporting evidence but actually reports limited cross-site validation)\n\n- **Hyperconnectivity may be a general cognitive reserve mechanism**: Increased connectivity during memory tasks is observed in successful aging and other conditions; not specific to AD compensatory states\n\n- **Metabolic imaging limitations**: FDG-PET has ~5-10mm spatial resolution and cannot distinguish metabolic contributions from different cell types or subcellular compartments\n\n### Alternative Explanations\n\n1. **Hypermetabolism reflects glial response** that may or may not support neuronal function; the coupling with hyperconnectivity may be coincidental\n2. **Apolipoprotein E4 effects** may independently alter both metabolism and connectivity through common pathways (e.g., cholesterol metabolism, neuroinflammation)\n3. **Vascular contributions** (Cerebral blood flow changes) affect both FDG-PET and fMRI signals independently\n4. **Network-level compensation occurs regardless of local metabolism**, with metabolic changes being secondary epiphenomena\n\n### Key Falsification Experiments\n\n1. **Prospective therapeutic trial**: Test whether stratifying patients into \"compensatory\" vs \"pathological\" hyperconnectivity groups based on this framework actually improves therapeutic outcomes when treatment is matched to classification.\n\n2. **Longitudinal FDG-PET + fMRI tracking**: Establish whether individuals converting from FDG-normometabolism + hyperconnectivity to FDG-hypometabolism + hyperconnectivity uniformly progress clinically.\n\n3. **Cell-type-specific metabolic imaging**: Use PET ligands targeting astrocyte-specific (MAOB) or neuronal-specific (VAChT) targets alongside FDG-PET to deconvolute cellular metabolic contributions.\n\n4. **Independent replication**: Validate the classifier in an independent cohort with different demographics, scanner manufacturers, and acquisition protocols.\n\n---\n\n## Summary of Revised Confidence Scores\n\n| Hypothesis | Original | Revised | Δ | Primary Concern |\n|------------|----------|---------|---|-----------------|\n| 1. Astrocyte lactate shuttle | 0.72 | 0.58 | -0.14 | FDG-PET may reflect inflammation, not neuronal compensation |\n| 2. PNN degradation | 0.68 | 0.52 | -0.16 | PNN loss may be adaptive; key evidence is computational |\n| 3. Theta-gamma/NPTX2 | 0.65 | 0.48 | -0.17 | NPTX2 is non-specific; mechanistic gap to oscillatory changes |\n| 4. Kir4.1 downregulation | 0.70 | 0.56 | -0.14 | Limited human data; AQP4 may be upstream |\n| 5. Complement pruning | 0.73 | 0.61 | -0.12 | C1q elevation is non-specific; directionality unclear |\n| 6. ADAR2 editing | 0.61 | 0.44 | -0.17 | Unedited GluA2 is normal; causality unestablished |\n| 7. Metabolic classifier | 0.78 | 0.65 | -0.13 | Correlative framework; lacks mechanism and prospective validation |\n\n---\n\n## Cross-Cutting Recommendations\n\n**1. Temporal precedence experiments are critical**: Most hypotheses suffer from unclear causality. Longitudinal studies establishing whether proposed mechanisms precede hyperconnectivity are essential.\n\n**2. Causal manipulation in adult animals**: Many cited experiments used developmental knockouts or overexpression. Adult-onset, circuit-specific manipulations are needed to establish relevance to established AD.\n\n**3. Human biomarker correlations**: CSF/plasma biomarkers for each mechanism should be correlated longitudinally with fMRI hyperconnectivity in prodromal AD cohorts (e.g., ALFA+, BioFINDER).\n\n**4. Regional specificity**: Most hypotheses focus on general mechanisms without explaining why DMN or MTL regions specifically develop hyperconnectivity. Regional molecular specificity should be addressed.\n\n**5. Falsification-focused experimental design**: The field should move toward hypothesis-falsification rather than hypothesis-support, particularly for therapeutic predictions that carry significant translational implications.",
  "tokens_used": "6404",
  "persona_id": "persona-skeptic"
}