Version history

{
  "session_id": "sess_SDA-2026-04-10-gap-debate-20260410-095958-ccbb9a80",
  "round_number": 3,
  "agent_persona": "persona-domain_expert",
  "agent_backend": "mini-max",
  "action": "support",
  "content": "\n\n# Practical Feasibility Assessment: HSP90-Tau Conformational Specificity\n\n## Executive Summary\n\nOf the seven hypotheses, **four represent viable drug discovery targets** worthy of resource commitment. Two should be deprioritized, and one requires structural validation before any investment decision is made. The critical path forward hinges on a single cross-validation experiment: cryo-EM classification of HSP90-tau-ATPγS complexes without reference alignment, which will simultaneously test Hypotheses 1, 3, 4, and 6.\n\n---\n\n## Surviving Hypotheses Prioritization Matrix\n\n| Hypothesis | Revised Confidence | Druggability | Therapeutic Potential | Development Risk | Priority |\n|------------|-------------------|--------------|----------------------|-------------------|----------|\n| 4. p23 Asymmetry | 0.70 | **High** | High (precision medicine) | Low-Medium | **Tier 1** |\n| 7. Acetylation Control | 0.68 | **High** | High (repurposing possible) | Low | **Tier 1** |\n| 6. Cryptic Pockets (MD) | 0.55 | Medium-High | Moderate | Medium | **Tier 2** |\n| 3. Dimer Interface Crevice | 0.48–0.52 | Medium | Moderate | Medium-High | **Tier 2** |\n| 5. C-terminal Gate | 0.61 | Medium | Moderate | Medium | **Tier 3** |\n| 1. Crypto-State (MD) | 0.38–0.42 | Low-Medium | Moderate (first-in-class) | High | **Tier 4** |\n| 2. ATP-Independent | 0.28–0.33 | N/A | N/A | N/A | **Deprioritize** |\n\n---\n\n## Tier 1: Proceed with Drug Discovery Programs\n\n### Hypothesis 4: p23 Asymmetry → Stabilization of HSP90-Tau Complexes for Targeted Degradation\n\n**Druggability: HIGH — 8/10**\n\nThe p23-HSP90 interface is a proven drug target. Geldanamycin derivatives already exploit the N-terminal pocket; the p23 binding site on the middle domain is structurally distinct and accessible.\n\n#### Therapeutic Rationale\n\nThe concept is elegant: stabilize a specific conformational state (HSP90-tau-p23) that renders tau susceptible to proteasomal degradation while sparing wild-type clients. This is precision oncology logic applied to neurodegeneration. The therapeutic index could be substantial if the asymmetric complex is genuinely distinct from symmetric p23-HSP90-client complexes used for kinase triage.\n\n**However, a critical vulnerability exists:** p23 binding prevents the full closed state, but the existing cryo-EM data does not resolve whether this creates a *unique* binding surface or merely represents one point on a conformational continuum. If the p23-tau-HSP90 complex shares structural features with p23-HSP90-kinase client complexes, selectivity will be difficult to achieve and off-target toxicity will be limiting.\n\n#### Existing Compounds and Precedents\n\n| Compound/Agent | Mechanism | Relevance | Clinical Status |\n|---------------|-----------|-----------|-----------------|\n| Geldanamycin / 17-AAG (Tanespimycin) | N-terminal HSP90 inhibitor | Proof of principle for HSP90 druggability | Withdrawn (Phase II/III) |\n| PU-H71 | N-terminal HSP90 inhibitor, binds buried pocket | Scaffold exists for HSP90 targeting | Phase I complete (oncology) |\n| Onalespib (AT13387) | N-terminal HSP90 inhibitor | Clinical PoC for chaperone targeting | Phase II |\n| PU-AD27 | Analog of PU-H71, improved CNS penetration | Relevant scaffold for tau indications | Preclinical |\n| **p23 disruptors (SBA1-targeting)** | **No current agents** | **This is the novel target** | N/A — requires *de novo* design |\n\nNo existing compound directly targets the p23-HSP90 interface. The existing clinical candidates are all N-terminal ATP competitors and would *antagonize* hypothesis 4's mechanism — they stabilize the closed state and would disrupt the p23-asymmetric complex the hypothesis proposes.\n\n**Development cost estimate:** $80–120M over 6–8 years to IND. The p23 binding site is novel and will require fragment-based screening, which adds time but reduces risk relative to high-throughput screening against an uncharacterized pocket.\n\n**Safety concerns:**\n- **On-target toxicity to wild-type clients**: HSP90 inhibition broadly destabilizes oncogenic clients (HER2, EGFR, CDK4). CNS-selective agents are needed, which the PU-H71 scaffold partially addresses.\n- **Cardiovascular risk**: 17-AAG showed cardiotoxicity in trials. This is a class effect from N-terminal inhibitors; p23-interface inhibitors should have a different profile if they stabilize rather than inhibit.\n- **Biodistribution**: p23 targeting must achieve CNS exposure. **This is the single largest development risk.** Tau pathology requires micromolar concentrations in brain tissue. Current HSP90 inhibitors have poor brain penetration.\n\n**Recommendation:** Proceed with fragment screening against the p23-middle domain interface. Simultaneously pursue cryo-EM validation of the asymmetric complex in a blinded classification paradigm. Budget for medicinal chemistry investment in CNS-penetrant analogs.\n\n---\n\n### Hypothesis 7: HDAC6-Mediated Acetylation Control → HDAC6 Inhibition as Surrogate Strategy\n\n**Druggability: HIGH — 9/10**\n\nThis is the most immediately actionable hypothesis because the target (HDAC6) is already druggable, clinical compounds exist, and the mechanism (HSP90 acetylation at Lys294/420) is testable with existing reagents.\n\n#### Therapeutic Rationale\n\nThe logic is a chain: HDAC6 inhibition → HSP90 hyperacetylation → altered conformational flexibility → reduced tau client recognition → decreased tau aggregation. This is indirect but mechanistically coherent. Critically, the hypothesis makes a testable prediction: if acetylated HSP90 disfavors tau clients, HDAC6 inhibitors should *reduce* the chaperone burden on tau and potentially improve proteostasis.\n\n**The key experimental data needed is missing:** direct measurement of HSP90 acetylation at Lys294 and Lys420 in disease-state neurons vs. age-matched controls. Acetylome studies (Choudhary et al., Science 2009; Weinert et al., 2011) have mapped lysine acetylation sites on HSP90, but disease-state-specific acetylation at these residues in tauopathy models has not been demonstrated with quantitative mass spec.\n\n#### Existing Compounds\n\n| Compound | HDAC6 Selectivity | Clinical Status | Relevance |\n|----------|------------------|-----------------|-----------|\n| Tubastatin A | High (but off-targets HDAC10 at high concentrations) | Preclinical | Research tool only |\n| ACY-1215 (Rocilinostat) | High selectivity for HDAC6 | Phase I/II (myeloma) | **Most advanced HDAC6-selective probe** |\n| ACY-738 | CNS-penetrant HDAC6 inhibitor | Preclinical | **Highest priority for tau studies** |\n| citarinostat (ACY-203) | HDAC6-selective | Phase I | Oral bioavailability, better than ACY-1215 |\n| Nexturastat A | Highly selective HDAC6 | Preclinical | Research use |\n| **PCI-34051** | High HDAC6 selectivity, low off-target liability | Preclinical | Pro-apoptotic in T cells — hematologic toxicity concern |\n\n#### Development Cost and Timeline\n\n**This is the lowest-cost path to clinical evidence.** ACY-738 or ACY-1215 can be obtained under MTA for preclinical studies. The cost to validation is approximately **$2–5M over 18 months** to test the hypothesis in iPSC-derived neurons from tauopathy patients.\n\nIf the hypothesis holds: HDAC6 inhibitors have a clean path to Phase II in tauopathies because the safety profile of ACY-738 has been established in the oncology setting. Repurposing reduces development cost to approximately **$15–25M** for a new indication IND package.\n\n**Timeline to Phase II proof-of-concept: 3–4 years.** This is the fastest path to clinical data on the HSP90-tau conformational selectivity concept.\n\n**Safety Concerns:**\n\n| Risk | Assessment | Mitigation |\n|------|------------|------------|\n| CNS off-target effects | HDAC6 is ubiquitous; tubulin acetylation changes are global | ACY-738 has demonstrated CNS penetration in rodents |\n| Hematologic toxicity | HDAC6 inhibitors cause thrombocytopenia (observed in ACY-1215 trials) | Monitor in Phase I; manageable |\n| Immunomodulation | HDAC6 affects T-cell function | Relevant for oncology indication; less concern for neurodegeneration |\n| Acetylation of off-target clients | HDAC6 deacetylates many substrates beyond HSP90 | Selectivity window must be defined experimentally |\n| Therapeutic index for tau vs. oncology | Unknown if same HDAC6 activity level is needed | Requires biomarker strategy (HSP90 acetylation as pharmacodynamic marker) |\n\n**Recommendation:** Execute the acetylome validation immediately (budget: ~$200K). If Lys294/420 acetylation is confirmed in disease states, proceed to iPSC neuron efficacy studies with ACY-738. This is the highest-value experiment in the entire research program.\n\n---\n\n## Tier 2: Validate Structural Basis Before Commitment\n\n### Hypothesis 6: Cryptic Pockets at N-terminal/Middle Domain Interface\n\n**Druggability: MEDIUM-HIGH — 6/10**\n\nThe concept is mechanistically sound: conformational ensembles of HSP90-tau vs. HSP90-HER2 should differ, and cryptic pocket formation probability can be computed. Fragment-based drug discovery (FBLD) can exploit this if the pocket is sufficiently large and stable.\n\n**The critical uncertainty:** Are the cryptic pockets in the HSP90-tau ensemble *unique* to tau, or do they also appear in other client-bound states? If the pocket is present in multiple clients, selectivity is lost and the therapeutic index collapses.\n\n#### Existing Approaches\n\n- **Molecular dynamics (MD) + ensemble-based FBDD**: Already commercially viable. Companies like Schrödinger, Cadence, and Healx use this pipeline routinely. The cost to generate 5–10 fragment scaffolds hitting the cryptic pocket would be approximately **$500K–1M** with an experienced computational team.\n- **Fragment screening by NMR**: SARmligands or Astex-style fragment screens are applicable here. The pocket is likely too cryptic for high-throughput crystallography, making NMR-based detection of binding-induced chemical shift perturbations the appropriate method.\n- **Cryo-EM fragment screening**: Newer approach (Diamond XChem, Rubintec) using cryo-EM as the primary screening read-out. Fragments are soaked into cryo-EM grids of HSP90-tau complexes. This directly connects to the validation experiments and is the most integrated approach.\n\n**Timeline:** 12–18 months to first validated fragment hits; 3–4 years to lead optimization.\n\n**Safety concerns:** Identical to existing HSP90 inhibitor concerns (see Tier 1 above). The cryptic pocket may be proximal to the nucleotide-binding site, meaning selectivity over the canonical pocket may be difficult. This would create a pharmacology similar to existing inhibitors, with similar toxicity profiles.\n\n**Recommendation:** Run MD ensemble comparison (Schrödinger or equivalent) for $150–200K to quantify cryptic pocket probability difference between HSP90-tau and HSP90-HER2. If the differential probability is >2-fold, proceed to fragment screen. If not, deprioritize.\n\n---\n\n### Hypothesis 3: Cryptic Hydrophobic Crevice at Dimer Interface\n\n**Druggability: MEDIUM — 5/10**\n\nThe dimer interface is the most challenging target of the three viable options. Dimeric protein-protein interfaces are inherently difficult to drug because the contact surface is large and flat. However, \"cryptic\" crevices at interfaces can be more tractable than direct PPIs.\n\n**The structural challenge:** The dimer interface is only exposed in the open state, which comprises approximately 10–20% of the HSP90 conformational ensemble at any given time. This means:\n1. A drug targeting this site would need to bind with high affinity to compete with the low population state\n2. The binding site may not be pre-formed in sufficient quantity for crystallographic fragment screening\n3. The cryptic crevice may only open transiently\n\n#### Structural Biology Requirements Before Investment\n\n| Experiment | Cost | Time | Decision Gate |\n|------------|------|------|---------------|\n| Cryo-EM of HSP90-tau at 3.5Å or better | $80–120K | 6 months | Does a crevice appear in classification? |\n| HDX-MS at dimer interface (residues 200-250) | $30–50K | 3 months | Is the HDX pattern different from HER2 complexes? |\n| Cross-linking mass spec with BS3/DSS cross-linkers | $40–60K | 4 months | Does the interface topology change with tau? |\n\n**If all three are positive:** This becomes a high-value target because the dimer interface is genuinely distinct from sites targeted by existing HSP90 inhibitors. Selectivity could be high.\n\n**If HDX and cross-linking are negative:** This hypothesis should be deprioritized. The dimer interface does not appear to remodel significantly with tau engagement.\n\n**Development cost estimate:** $60–100M over 5–7 years if structural validation is positive. The interface targeting would likely require a medium-sized molecule (MW 500–700) to achieve sufficient contact surface. This is achievable but requires significant medicinal chemistry investment.\n\n**Safety concerns:** The dimer interface is structurally critical. Disruption could:\n- Prevent HSP90 dimerization entirely → complete loss of chaperone function → catastrophic cellular stress\n- Stabilize a non-productive dimer state → dominant-negative effect\n- Trigger aggregation of partially dimerized HSP90 → off-target toxicity\n\n**This is the highest-risk structural hypothesis.** The potential selectivity reward is real, but the developmental risk of a structural disruption agent is substantial. Do not commit to medicinal chemistry investment without positive structural validation.\n\n---\n\n## Tier 3: Mechanistically Interesting, Low Priority\n\n### Hypothesis 5: C-terminal Gate\n\n**Druggability: MEDIUM — 5/10**\n\nThe C-terminal MEEVD motif and the allosteric site near the C-terminal dimer interface represent a legitimate structural target. The MEEVD binds TPR domain proteins (HOP, FKBP51/52, CHIP). If tau binding opens a gate near this site, it suggests that tau competes with or redirects TPR cochaperone binding.\n\n**Therapeutic value:** Modulating the TPR cochaperone network via the C-terminal gate is conceptually sound. FKBP51 is associated with stress resilience and tau pathology; CHIP is a ubiquitin ligase that could target tau for degradation. The therapeutic window could be favorable if selective.\n\n**Challenges:**\n- The C-terminal domain is highly dimerized andflexible — structural characterization is technically difficult\n- The MEEVD is a known binding motif with many competing interactions — achieving selectivity is nontrivial\n- No precedent for small molecules targeting the C-terminal HSP90 dimer interface (existing drugs all target the N-terminal ATP pocket)\n\n**Recommendation:** Low priority for drug discovery. Focus structural biology resources on Tier 1 and 2 hypotheses first. Revisit if Tier 1 programs fail to achieve selectivity.\n\n---\n\n## Tier 4: Require Fundamental Validation Before Any Investment\n\n### Hypothesis 1: Middle Domain Crypto-State\n\n**Druggability: LOW-MEDIUM — 4/10**\n\nThe crypto-state concept is scientifically interesting but structurally unsupported. The hypothesis makes specific predictions (helix 4/5 reorientation, druggable pocket formation) that have not been demonstrated. Without this demonstration, any drug discovery program would be targeting an unvalidated conformational state.\n\n**The structural biology barrier is fundamental:** If HSP90-tau cryo-EM data from 2019 and 2021 failed to reveal the crypto-state, either:\n(a) The state is transient and below cryo-EM detection threshold (requiring NMR or smFRET)\n(b) The state requires an additional cofactor or post-translational modification not present in the existing structures\n(c) The state does not exist\n\n**Recommendation:** Do not invest in drug discovery until the crypto-state is structurally validated. Perform the *ab initio* cryo-EM classification experiment (no reference alignment, >50K particles) as the decisive test. If a distinct class emerges at >10% abundance with the predicted features, this becomes a first-in-class opportunity. If not, deprioritize.\n\n**Why not fully deprioritize:** The original confidence of 0.72 is not trivial. Crypto-states in other chaperones (Hsp70, BiP) have proven to be real and druggable once properly characterized. The structural biology community should complete the validation experiment before abandoning this hypothesis.\n\n---\n\n## Deprioritized: Discontinue\n\n### Hypothesis 2: ATP-Independent Conformational States\n\n**Recommendation: Dis",
  "tokens_used": "4094"
}