How does the Ir(III) carbene complex 1a bind the Girdin C-terminus — and can the SAR be explained?

Stage-1 computational binding-site search · corrected, disorder-aware, blindly-searched, SAR-validated protocol

Target: human Girdin / CCDC88A, UniProt Q3V6T2 (1871 aa); binding construct His-Girdin-CT 1660–1870 (MST).   Ligand: substitutionally-inert octahedral iridium(III) carbene 1a = [Ir(ppy)2(4-pyrrolidinyl-iso-BIPY-NHC)]+ and its 4-R congeners 1b–1e + R=H control.
Paper under test: Ruan M-L et al., Iridium(III) carbene complexes as potent girdin inhibitors against metastatic cancers. PNAS 2024;121(25):e2316615121.
NEGATIVE No binding site on the Girdin C-terminus reproduces the experimental structure–affinity relationship (SAR) above the method's own noise floor. The honest deliverable is a correct, reproducible, falsifiable methodology and an honest hypothesis space — not a proven binding site. This is a hypothesis, not a proof.

Bottom line

Using a corrected, disorder-aware, blindly-searched, SAR-validated protocol, the blind whole-surface search returns a rigorous NEGATIVE (4 candidate sites, 0 clear both robust SAR signals, expected false positives ≈ 0), physics (GFN2-xTB) rescoring of the best-occupied site does not confirm it, and the only place the coarse SAR does reproduce — the authors' reproduced homology “SH2” pocket — reproduces it for a non-pocket-specific shape/size reason, in a fold that six independent methods have refuted.

The single best structural hypothesis carried forward (low confidence, explicitly not a validated lead) is a shallow, transient interaction near the GBA/Gαi3 helix (canonical ≈1692–1744, site B01) — consistent with the one experimentally-validated function of this region (1a disrupts the Girdin–Gαi3 interaction), but not SAR-reproducing. The fine intra-amine ordering is never claimed.

1Methodology — and how it differs from the authors'

A seven-stage, third-party-reproducible pipeline (full spec: METHODOLOGY.md). The core correction: never assume a fold, search the whole surface blindly, and make every site stand or fall against the measured SAR under a noise-gated, multiple-comparison-guarded test.

This work — the corrected protocol
  • Stage 0 · build & chemically validate the exact 1a–1e + R=H series (only 4-R differs; +1 singlet octahedra; formulas < 0.7 mDa vs SI HRMS).
  • Stage 1 · disorder-aware ensemble of Girdin-CT: 28 distinct members / 5 independent source families; median Cα-RMSD 20.3 Å, Rg 13–46 Å, pLDDT 23–42 (disorder represented honestly).
  • Stage 2 · per-conformer fpocket transient-pocket catalogue (288 pockets → 8 recurrent, all transient).
  • Stage 3 · blind whole-surface docking of the full series into every compact member (smina, exh 32, Ir→Fe shape surrogate, rigid, xTB-GFN2 charges).
  • Stage 4 · SAR rank-order validation vs the fixed Kd yardstick with a pre-registered noise floor (MDE), pre-registered Signals A/B, and a 720-permutation multiple-comparison guard.
  • Stage 5 · GFN2-xTB physics rescoring of the lead/best-occupied site.
  • Stage 6 · orthogonal-data consistency (Gαi3/GBA pulldown; mutagenesis residues).
The authors' approach (Ruan et al.)
  • A single rigid homology-modeled “SH2-like” pocket (SWISS-MODEL on mouse SOCS3-SH2, 2HMH) — on a fold six independent methods refute.
  • A single-complex 1a docking pose into that one predefined pocket.
  • The measured Kd series (1a–1e) is never tested against the pose — the strongest experimental constraint in their own paper sits unused.

Converting docking from an illustration into a falsifiable hypothesis test is the whole point — for which a rigorous NEGATIVE is itself a valid result.

2The read-only SAR oracle — computed vs experimental

The experimental SAR is the fixed yardstick, re-extracted verbatim from the primary source (Fig. 7E/8E/7G + SI Fig. S20). Only the 4-R substituent varies. The fine intra-amine ordering (~0.6 kcal/mol) is at/below docking noise — NOT resolved, never claimed.

complex4-R group exp. Kd (µM) docking, authors' pocket
(best, kcal/mol)		 GFN2-xTB ΔEint at B01
(mean, kcal/mol)		 note
1apyrrolidinyl (5-ring amine)1.3−8.37−22.7tightest (exp.)
1dpiperidinyl (6-ring amine)1.5−7.90−27.7focus pose ↓
1edimethylamino (NMe2)1.6−7.07−26.6docking misplaces 1e
1cmorpholinyl (amine-ether)2.6−7.54−29.2
1btert-butyl (bulky alkyl)3.6−7.43−30.0
R=Hnone — plain 2,2′-bipyridine> 20−6.71−33.8weakest (exp.)
Experimental rank (tightest → weakest): 1a < 1d < 1e < 1c < 1b ≪ R=H. Computed columns: authors'-pocket docking (stage1/dock_ctrl, R=H = measured Irppy-bipy); GFN2-xTB ΔEint at best-occupied site B01 (stage1/rescore, R=H = ctrl_scaffold_H).
Read the computed columns honestly. The authors'-pocket docking reproduces only the coarse trend (1a tightest, R=H weakest) — and only because R=H is a smaller ligand / amine rings are bulkier: a shape/size effect, not pocket-specific (1e is already misplaced). The GFN2-xTB ΔEint column sits entirely inside a 14.2 kcal/mol noise band (R=H is not even least-tight by mean energy) → NOT claimable in either direction, shown for completeness only. Two robust signals a model must rationalize:
  • Signal A — the 4-R substituent is essential (R=H is ≥ ~1.6 kcal/mol weaker than 1a; RT·ln(20/1.3) ≈ 1.62 kcal/mol).
  • Signal B — coarse chemotype: small basic amine ring (1a/1d/1e) > bulky/ether/tert-butyl (1c/1b).
  • NOT a signal — the fine intra-amine ordering (~0.6 kcal/mol) is at/below docking noise: not reproducible, never claimed.

3Candidate sites & the verdict

The blind whole-surface docking clustered all poses into four candidate sites. No site reproduces both robust SAR signals above its noise floor. Blind verdict: NEGATIVE — lead = none; expected false positives = 0.0 (720-permutation null). The one_of_n_sites list is empty — no “1 of N” candidate to promote.

sitecanonical spanoccupancy complete 6-complex SAR?site MDE
(kcal/mol)		 clears both signals?notes
B011692–17448/8 members, 286 posesyes1.13 nomost-occupied; overlaps GBA/Gαi3 + recurrent S01/S06 — focus pose below
B021752–17926 membersyes1.07 noover Q1778 / S07 region (A gap 0.58 < MDE)
B0318041 memberincomplete— n/asingle-occupancy, not scored
B041811–18336 membersyes0.72 noρ=1.0 / τ=1.0 but inside noise — NOT a lead
The anti-gaming save (must not be undone). Site B04 has a perfect rank order (ρ=1.0, τ=1.0) yet is correctly not a lead: its Signal-A gap (0.4627 kcal/mol) sits below the site MDE (0.7236) — exactly the ~0.6 kcal/mol fine ordering declared unclaimable. A perfect ρ inside the noise band is the textbook multiple-comparison false positive the guard exists to reject. This site is not resurrected as a hit.

Final determination after rescoring: the blind stage produced no SAR-validated lead and no demoted “1 of N” site — nothing rescoring could elevate. Physics rescoring of the best-occupied site B01 DOES NOT CONFIRM (MDE 14.22 kcal/mol; Signal A gap −6.54, Signal B gap −3.97, both far inside the band; the ρ=−0.94 anti-correlation is inside the 14 kcal/mol band → not claimable in either direction). Therefore the FINAL determination is NEGATIVE: there is no SAR-validated binding site.

4Interactive 3D pose — best-occupied site B01

The best-occupied focus pose: complex 1d in ensemble member E008, seed 101, affinity −9.27 kcal/mol. This is the best-occupied site, explicitly labeled as such — not a validated lead (the verdict is NEGATIVE). The 4-R substituent is in measured 3.32 Å contact (PHE1743) and is not solvent-exposed, so an “R drives affinity” rationale is geometrically admissible here — but because B01 fails both MDE-gated signals, no mechanistic story is asserted.

Loading 3D pose…
Girdin-CT (E008, disordered) site B01 (1692–1744) 4-R contact (ILE1698, PHE1743) complex 1d (ligand) Ir centre
B01 overlaps the GBA/Gαi3 interface (canonical 1698–1702) and recurrent transient pockets S01/S06. The receptor is a genuine intrinsically-disordered model (pLDDT 23–42) — its extended, thin trace is honest, not an artifact. Pose reconstruction & contacts: stage1/site/structural_rationalization.json.

5Authors'-pocket control — why coarse-signal reproduction is not validation

All six complexes were docked into the reproduced authors' homology “SH2” pocket under the identical protocol. The pocket reproduces BOTH Signal A (best-score gap 0.952 / mean 0.938 ≫ MDE 0.125) and Signal B (gap 0.298 / 0.340 > MDE), Spearman ρ = +0.83 (p = 0.042), Kendall τ = +0.73. But this does not validate the pocket, for three converging reasons:

  • The signals are not pocket-specific. Signal A is driven by R=H being a smaller ligand; Signal B by amine rings being bulkier — shape/size complementarity that would emerge in almost any concave surface. That is why the blind whole-surface search + multiple-comparison guard is the load-bearing test — and there the same signals fail.
  • The fold is refuted. Six independent methods (InterPro/Pfam, HMMER PF00017 0 hits, TM-align 0.11–0.13, Foldseek PDB+AFDB 0 hits, 19.6% identity) plus a functional test (SH2-segment + EGFR-pTyr co-fold fails 0/5, domain pLDDT 37–39, while the same method recovers a real SH2 at 2.4 Å on a positive control). fpocket druggability at the nominal pocket is 0.018; a ligand-sized cavity exists in only 1/11 models. The pocket is a homology artifact.
  • The fine ordering is still not recovered even here (docking misplaces 1e).

So the authors' pocket “explains” the coarse SAR only in the trivial shape sense, on a fold the target does not adopt — which is why a single-complex pose into it is not falsifiable and does not constitute validation.

6Confidence & limitations (honest boundaries)

  • A binding site here is a computational hypothesis, never a proof. The blind search returned NEGATIVE; physics rescoring does not confirm; the authors' pocket reproduces only non-specific shape/size signals. None of this proves where 1a binds — it constrains and falsifies.
  • The fine ordering is not resolved, by design. Only Signals A and B are testable; B04's ρ=1.0 is inside noise and is not a claim.
  • The MDE is conservative on this ensemble (whole-surface blind search over a heterogeneous IDR → docking MDE 0.7–1.1, xTB MDE ~14 kcal/mol, both larger than the entire ~0.8 kcal/mol SAR spread). The test cannot certify a SAR sitting inside its own noise — it yields honest negatives but is biased toward false negatives by construction.
  • Ensemble sampling is incomplete and partly substituted. The two GPU/host samplers named in the plan were environment-blocked → ANM normal-mode expansion was used as a labeled flexibility surrogate (not claimed to be MD). The extreme C-terminus 1815–1871 is unsampled.
  • Docking-score / rigid-body / gas-phase limits. smina/Vina has no Coulomb term; Ir is a Fe shape surrogate; the complex is rigid (justified for an inert octahedron; forbids induced fit). GFN2-xTB rescoring is gas-phase on a truncated, neutral, valence-capped pocket — applied identically across complexes (so it does not bias the between-complex comparison) but not a free-energy calculation.
  • Numbering discipline is load-bearing (paper = canonical − 1); a numbering error would silently move every contact and overlap.

Provenance. Every number on this page traces to a committed Stage-1 evidence file and the primary-source Kd oracle; nothing here exceeds the validated synthesis report (REPORT.md, VAL-REPORT-001). This mission does not claim to have found where 1a binds — its deliverable is a correct, reproducible, falsifiable protocol and an honest binding-mode hypothesis space: a rigorous NEGATIVE for the blind lead, a coarse-signal-only (non-specific) result for the authors' refuted pocket, and a best-occupied focus site (B01, GBA/Gαi3-overlapping) carried forward as a low-confidence hypothesis, not a proof.

Source paper: Ruan M-L, Ni W-X, Chu JCH, et al. PNAS 2024;121(25):e2316615121. doi:10.1073/pnas.2316615121 · Stage-1 evidence: stage1/{ligands_series, ensemble, dock_setup, dock_ctrl, dock_blind, site, rescore, report}.