Fold №63 — DISCARDED

Tirzepatide Lys24–Glu28 i,i+4 Lactam Staple: GIPR-Selective Bias Hypothesis

TLDR

Fold №63 was DISCARDED due to a tool-limit failure: the peptide–GIPR complex could not be modeled with sufficient inter-chain confidence (ipTM 0.45) to evaluate the GIPR-selectivity hypothesis. This is not a biological invalidation — the structural predictor could not resolve the complex geometry of a doubly modified peptide (covalent lactam staple plus Lys-20 C20 lipid chain) against a class B GPCR transmembrane bundle within a single-run, unconstrained prediction framework.

What We Tried

Tirzepatide's central amphipathic helix (residues ~20–30) is the structural core of its dual incretin pharmacology. Native Cys-24, a chemically inert and oxidation-prone residue, was mutated to Lys and bridged to the native Glu-28 via an i,i+4 side-chain-to-side-chain lactam bond — covalently locking one full helical turn. The Lys-20 C20 fatty diacid lipidation was retained in full.

The pharmacological hypothesis was that rigidifying the helical face that contacts GIPR's transmembrane bundle would pre-organize the GIPR-competent conformation more than the GLP-1R-competent one — differentially biasing the dual-agonist balance toward GIP-like signaling. The rationale was grounded in tirzepatide's already-imbalanced receptor occupancy (GIPR-favoring at clinical doses, per Willard et al. 2020) and the general principle that conformational pre-organization reduces entropic cost at the binding interface.

This fold builds directly on Fold №23 (tirzepatide αMe-Cys24, PROMISING, pLDDT 0.71), which validated Cys-24 as a productive modification anchor, and is conceptually related to Fold №54 (Retatrutide Lys17–Asp21 lactam, REFINED, pLDDT 0.71), which demonstrated that i,i+4 lactam stapling in an incretin scaffold can be modeled confidently when the chemical complexity is within tool resolution.

Why It Was Discarded

The structural predictor returned:

• pLDDT: 0.693 — moderate backbone confidence; the helical peptide fold itself was plausibly resolved
• ipTM: 0.454 — critically below the ~0.5 threshold for reliable inter-chain geometry; the docking pose of the stapled peptide against GIPR's transmembrane bundle could not be confidently established
• Chai-1 ensemble agreement: not produced
• Boltz-2 affinity values: not produced

The most likely technical cause is the combination of two non-canonical chemical features in a single prediction run: the covalent i,i+4 lactam crosslink (which falls outside standard AlphaFold-derived parametrization for non-adjacent side-chain bridges) and the bulky Lys-20 C20 fatty diacid lipid chain (similarly non-canonical). Either modification alone has approached tool resolution in prior folds; together, they appear to exceed the reliable modeling envelope for a single-run, unconstrained prediction.

The discard is therefore classified as a chemistry-complexity tool-limit failure — not a prediction of non-binding, not a prediction of poor folding, and not a biological signal of any kind.

This contrasts with Fold №52 (semaglutide αMe-His1, DISCARDED, pLDDT 0.72), where the discard was also at the tool-limit boundary but on a simpler modification — suggesting that ipTM failures on this peptide class tend to arise from GPCR transmembrane bundle modeling challenges rather than peptide backbone issues per se.

What This Doesn't Mean

DISCARDED does not mean disproved. This fold was discarded because current structure prediction tools could not adjudicate the peptide–receptor complex geometry with sufficient confidence — not because the compound was predicted to be inactive, misfolded, or pharmacologically incoherent. The backbone pLDDT of 0.693 is on par with the PROMISING tirzepatide and semaglutide folds in this lab; the peptide itself folds plausibly. The failure is entirely at the interface modeling level, driven by chemical features that lie at the edges of current parametrization. The selectivity-shift hypothesis — that helical rigidification at residues 24–28 biases dual-agonist activity toward GIPR — remains scientifically open, supported by sound pharmacological logic, and consistent with the broader literature on tirzepatide's imbalanced receptor occupancy. Wet-lab synthesis of this analog would not be unreasonable on the basis of this discard.

What Would Answer the Question

• cAMP dose-response assays at GIPR and GLP-1R (CHO or HEK293 cells transiently expressing each receptor): direct measurement of EC₅₀ and Emax at both receptors would quantify any selectivity shift relative to native tirzepatide. This is the minimum viable experiment and the most direct test of the hypothesis.
• β-arrestin recruitment assay (HTRF or BRET): since tirzepatide's biased GLP-1R agonism (cAMP-favoring, β-arrestin-avoiding) is a key pharmacological feature per Willard et al. (2020), measuring whether the lactam staple preserves or alters this bias profile is essential.
• Surface plasmon resonance (SPR) or isothermal titration calorimetry (ITC) against purified GIPR and GLP-1R extracellular domains: binding affinity measurements would decouple conformational pre-organization effects from downstream signaling, and ΔG/ΔH/ΔS decomposition (ITC) could directly detect the predicted reduction in entropic binding cost.
• Free energy perturbation (FEP) or enhanced sampling MD using cryo-EM structures of tirzepatide-GIPR and tirzepatide-GLP-1R complexes (available in the PDB): computational approaches that explicitly model the lactam constraint and lipid chain would bypass the parametrization limitations of AlphaFold-derived single-run predictions and provide a more reliable selectivity estimate.
• Circular dichroism (CD) spectroscopy of the synthetic stapled peptide in aqueous buffer and membrane-mimetic environments: would confirm whether the lactam bridge produces the expected increase in helical content and whether the Lys-20 lipid chain's membrane-anchoring function is preserved.

Raw Metrics

All structural metrics are in silico predictions from a single run. Heuristic peptide profile values are sequence-based estimates, not experimental measurements. This is research, not medical advice.