Biophysical Derisking of Allosteric Ligand Engagement in Cancer Kinases Before Wet-Lab Assay Escalation

Technical Summary: Biophysical Derisking of Allosteric Ligand Engagement in Cancer Kinases

Problem Statement
The paper addresses a critical bottleneck in cancer kinase drug discovery: the difficulty of distinguishing true allosteric ligand engagement from false positives generated by computational pipelines (docking, structural modeling, molecular dynamics). A ligand may remain stable in a simulation without occupying the correct site, potentially representing an ATP-adjacent boundary state, a non-specific contact, or a chemically unreliable evidence chain. Because wet-lab assays (HDX-MS, binding, cellular engagement) are expensive and resource-intensive, there is a need for a rigorous pre-assay classification method that can derisk ligand-site cases before experimental escalation. The core problem is that simulation stability (ligand retention) is often over-interpreted as evidence of correct site assignment, allosteric engagement, or kinase inhibition, leading to wasted experimental resources.

Methodology
The study proposes a computational biophysical framework for pre-assay classification, moving beyond simple ligand retention to a multi-layered evidence chain. The methodology integrates six specific evidence layers:

1. Ligand Retention & Site Assignment: Verifying that the ligand remains associated with the intended region and classifying the site as intended allosteric, ATP-site, or boundary-variable.
2. Comparator Separation: Using known ATP-site ligands (e.g., Nilotinib) to ensure the method distinguishes ATP-site behavior from non-ATP allosteric engagement.
3. Apo-Holo Geometry: Analyzing local structural descriptors (Root-Mean-Square Deviation and Radius of Gyration) to determine if the ligand-bound state is structurally distinguishable from the apo reference.
4. Fixed-Volume Hydration Occupancy: Counting water molecules within a predefined allosteric volume to detect local dehydration (a proxy for pocket occupancy) without claiming thermodynamic free energy.
5. Dynamic Ensemble Coupling: Using residue-level dynamic cross-correlation and sector analysis to identify correlated motions, treated as ensemble support rather than proof of causal allostery.
6. Null-Supported Signal Robustness: Applying circular time-shift controls to dynamic signals to ensure observed attenuation is not an artifact of trajectory autocorrelation or noise.
7. Provenance Auditing: Checking ligand identity, stereochemistry, and source-file consistency to prevent "evidence-chain failures."

The study employs a Go / Caution / No-Go classification system:

• Go: Coherent retention, correct site assignment, comparator separation, and supportive local biophysical response.
• Caution: Meaningful positive evidence exists, but unresolved ambiguity (e.g., ATP/allosteric boundary variability) prevents high-confidence classification.
• No-Go: Evidence chain failure due to ligand drift, wrong-site behavior, setup inconsistency, or chemical provenance issues.

Key Case Studies and Results
The method was evaluated across three distinct kinase scenarios:

1. ABL1/Asciminib (Positive Benchmark):

   • Result: Classified as Go.
   • Evidence: Asciminib correctly occupied the myristoyl pocket, remained separated from the ATP-site, and showed expected local structural and hydration responses.
   • Significance: Confirmed the method's ability to recover a known non-ATP allosteric state and distinguish it from ATP-site behavior.

2. ABL1/Nilotinib (ATP-Site Comparator):

   • Result: Classified as Go (Comparator).
   • Evidence: Nilotinib remained in the ATP-site region and was clearly separated from the myristoyl-pocket allosteric state.
   • Significance: Demonstrated that the method does not misclassify stable ATP-site ligands as allosteric engagements.

3. EGFR L858R/T790M/C797S + JBJ (Resistance-State Stress Test):

   • Result: Classified as Caution (specifically R4 in an EGFR-specific sub-classification).
   • Evidence: The JBJ ligand showed retention, reproducible fixed-volume dehydration, and local apo-holo geometric separation. Dynamic coupling analysis showed robust, null-supported signal attenuation.
   • Limitation: LEU844-associated ATP/allosteric boundary variability prevented a "clean" high-confidence allosteric classification (R5).
   • Significance: The method successfully preserved positive evidence (retention/dehydration) while correctly downgrading the case due to unresolved boundary ambiguity, avoiding overclaiming of resistance rescue.

4. Legacy FGR/UNL Analog 7 (Negative Control):

   • Result: Classified as No-Go.
   • Evidence: While initial docking-to-MD criteria appeared attractive, re-evaluation revealed the ligand was stereochemically under-specified (12 chiral centers, only 3 assigned) and the molecular dynamics trajectory ligand identity did not match the submitted structure.
   • Significance: Highlighted that computationally attractive poses are insufficient if chemical provenance and stereochemical integrity are not verified.

Key Contributions and Claims
The paper does not claim to predict potency, residence time, kinase inhibition, clinical efficacy, or resistance rescue. Instead, its primary contribution is a pre-assay decision-support framework that:

• Derisks ligand-site cases by requiring multi-layered evidence (retention + site assignment + local response + provenance) rather than relying on simulation stability alone.
• Standardizes the interpretation of allosteric engagement by explicitly separating ATP-site behavior and identifying boundary-variable states (Caution) versus clean allosteric states (Go).
• Audits chemical provenance, preventing the advancement of cases with ambiguous ligand identity or stereochemistry.
• Prioritizes experimental follow-up: "Go" cases support focused testing; "Caution" cases suggest targeted assays to resolve specific ambiguities (e.g., boundary mapping); "No-Go" cases stop resource consumption on unreliable evidence.

Significance
The study argues that computational biophysical analysis should serve as a disciplined filter before wet-lab escalation. By explicitly avoiding claims of experimental validation, hydration free energy, or causal allostery, the method provides a conservative, reproducible logic for prioritizing cancer kinase ligand programs. It successfully demonstrated that a recovered positive benchmark (ABL1), a separated ATP comparator (ABL1), a boundary-variable resistance state (EGFR), and a chemically ambiguous legacy case (FGR) can be distinctly classified, thereby optimizing the allocation of experimental resources.