FOLD №53 — N-Terminal Hexenoyl Lipidation of Sermorelin for Half-Life Extension

Verdict: DISCARDED | Peptide: Sermorelin | Class: PERFORMANCE | Target: GHRHR (Q02643)

Mechanism of Action (Background)

Sermorelin (YADAIFTNSYRKVLGQLSARKLLQDIMSR) is the shortest synthetic fragment of human GHRH retaining full agonist activity at the growth hormone-releasing hormone receptor (GHRHR). Receptor engagement is mediated by a bipartite interaction: the N-terminal residues (Tyr-1, Asp-3, Phe-6, Thr-7) form the primary pharmacophore for extracellular domain contact, while the central and C-terminal helical segment drives receptor activation. GHRHR couples primarily to Gs/cAMP signaling in pituitary somatotrophs, triggering GH secretion and downstream IGF-1 production. GHRHR is also expressed in lung, immune, and cardiac tissues, with downstream JAK2/STAT3 and MAPK pathway coupling documented in antagonist literature (Gesmundo et al., 2025; Liang et al., 2020).

Sermorelin's principal pharmacokinetic liability is rapid proteolytic degradation via two independent routes: DPP-IV cleaves the Tyr1-Ala2 N-terminal bond within minutes of administration, while carboxypeptidases attack the C-terminal Arg-29. González-López et al. (2023) confirmed both degradation axes empirically, characterizing fragments (1-11), (13-20), and (22-29) as distinct products in human blood and serum — establishing that both termini are simultaneously active vulnerability points.

Modification Hypothesis (What We Tested)

Fold №53 proposed dual terminus protection via:

1. N-terminal trans-3-hexenoyl acylation of the Tyr-1 α-amine — the identical N-cap chemistry used in FDA-approved Tesamorelin — to sterically block DPP-IV recognition of the Tyr1-Ala2 bond
2. C-terminal amidation of Arg-29's α-carboxylate — a standard pharmacokinetic modification — to blunt carboxypeptidase attack

Yielding: hexenoyl-YADAIFTNSYRKVLGQLSARKLLQDIMSR-NH2

The hypothesis was grounded in a strong precedent: Tesamorelin is GHRH(1-44) with the same N-cap, is FDA-approved, and demonstrates superior pharmacokinetics to native GHRH and sermorelin. The modification was designed to preserve pharmacophore residue orientation (Tyr-1, Asp-3, Phe-6, Thr-7) toward the GHRHR ECD while closing both degradation routes without altering side-chain chemistry. This fold is strategically distinct from prior Sermorelin pharmacokinetic attempts — specifically Fold №2 (D-Ala2, stereochemical DPP-IV block, DISCARDED) — in using a steric cap rather than a non-canonical amino acid, and from Fold №42 (Lys-21/Asp-25 lactam staple, DISCARDED) in leaving the helical region entirely untouched.

Why the Prediction Was Uninformative (Technical Analysis)

The prediction produced no actionable structural data. An ipTM of 0.31 is well below the threshold at which receptor-peptide interface geometry can be trusted — the model has not converged on a pose. No Chai-1 ensemble run was available to cross-validate or partially rescue the signal. The Boltz-2 affinity module returned no output, meaning even the heuristic binding-change estimate is absent.

The failure mechanism is likely methodological, not biological. Neural network structure predictors (AlphaFold-family, Boltz, Chai) are trained overwhelmingly on canonical amino acid sequences. N-terminal acyl caps — especially short-chain non-natural modifications like trans-3-hexenoyl — fall substantially outside training data distributions. The predictor cannot reliably place the modified N-terminus relative to the receptor ECD, and this uncertainty propagates through the entire interface score. Sermorelin's inherently flexible, partially disordered backbone in isolation compounds the problem: the model has no rigid anchor to build the receptor interaction around.

This is a recurring pattern in the Sermorelin series. Fold №2 (D-Ala2 substitution) returned pLDDT 0.49. Fold №42 (lactam staple) returned pLDDT 0.50. All three modifications target structurally important but conformationally flexible regions of a peptide whose receptor interaction involves a disordered recognition arm — precisely the scenario modern deep learning predictors handle least well. Compare this to Fold №48 (Ipamorelin γGlu-Palm lipidation), which returned pLDDT 0.78 and a REFINED verdict: Ipamorelin's compact, constrained five-residue scaffold gave the predictor a rigid, resolvable anchor that Sermorelin simply does not provide.

What This Tells Us (Negative Results Are Data)

What this does NOT mean: The hexenoyl-sermorelin-NH2 construct is unlikely to be biologically active or therapeutically interesting. This interpretation is not supported. The mechanistic rationale is among the strongest in the Sermorelin series: it is anchored in an approved drug's chemistry (Tesamorelin), targets two experimentally confirmed degradation axes, and avoids the non-canonical amino acid issues that complicated earlier folds.

What this DOES mean:

• The Boltz-2/AlphaFold-family toolkit is not currently equipped to reliably evaluate N-terminally acylated peptides with flexible backbones bound to class B GPCRs. The tool has reached its modeling boundary.
• Sermorelin's extended, flexible conformation is consistently problematic for these predictors. Three consecutive uninformative pLDDT scores (~0.49-0.50) across structurally diverse modifications suggest this is a scaffold-level limitation, not a modification-level signal.
• The albumin-binding claim (transient association via C6 hexenoyl) is flagged as mechanistically weak by literature review. Trans-3-hexenoic acid is far shorter than the C14-C18 chains reliably required for albumin affinity (as seen in semaglutide, detemir). This specific sub-hypothesis should be treated as speculative absent experimental Kd measurement.
• The DPP-IV protection hypothesis remains strongly supported and is not refuted by the structural prediction failure. It is a chemistry-level argument (steric blockade) that does not require a high-confidence structural pose to be valid.

Alternative Hypotheses to Test (Avoiding This Failure Mode)

For the modification itself (wet lab path):

• Synthesize hexenoyl-YADAIFTNSYRKVLGQLSARKLLQDIMSR-NH2 directly and subject it to DPP-IV stability assay, plasma stability assay, and GHRHR cAMP activation assay. This is a well-precedented modification on a known scaffold — the chemistry is accessible and the assays are standard. The structural predictor has failed; the bench has not been tried.
• Measure albumin binding affinity (ITC or SPR) to confirm or refute the transient association hypothesis for C6 chain length.

For computational approaches:

• Use dedicated peptidomimetic modeling tools (e.g., Rosetta FlexPepDock with explicit small-molecule N-cap parametrization) rather than neural network structure predictors for acylated peptides.
• Attempt ensemble prediction across multiple Boltz-2 seeds and report variance in pLDDT to distinguish genuine disorder from convergence failure.
• Consider modeling the unmodified sermorelin backbone at high confidence first, then in silico dock the hexenoyl cap as a separate fragment — decomposing the problem rather than demanding a single end-to-end prediction.

For alternative PHARMACOKINETICS modifications on Sermorelin:

• A PEGylation strategy at a non-pharmacophore residue (e.g., Lys-12 or Lys-21 side chain) would test half-life extension through a mechanism more tractable to current predictors and with a larger albumin-retention contribution than C6 acylation.
• The Fold №46 DSIP dual terminus capping precedent (Ac-N/C-NH2, pLDDT 0.65, PROMISING) suggests that N-terminal acetylation rather than hexenoylation might yield a more confident structural prediction for Sermorelin's terminus — though at the cost of the DPP-IV steric blocking mechanism, since acetyl is smaller.
• Given that González-López et al. (2023) found sermorelin(22-29) to be relatively stable in blood, future pharmacokinetic modifications might prioritize N-terminal protection above C-terminal amidation if resources require prioritization.