FOLD №52 — Semaglutide αMe-His-1 N-cap

Verdict: DISCARDED | pLDDT 0.72 | pTM 0.82 | ipTM 0.84 | No binding signal

Mechanism of action (background)

Semaglutide is a GLP-1 receptor agonist that activates GLP-1R on pancreatic β-cells, hypothalamic nuclei, and cardiovascular tissue to suppress appetite, reduce fasting glucose, and improve cardiometabolic outcomes. Receptor activation proceeds through a two-domain mechanism: the peptide C-terminus docks to the extracellular domain (ECD) of GLP-1R, anchoring the peptide, while the N-terminal segment — most critically His-1 — inserts into the transmembrane (7TM) bundle and triggers Gαs coupling and cAMP elevation. The imidazole side chain of His-1 makes direct contacts within the TMD that are essential for agonist efficacy; truncation or ablation of His-1 converts full agonists to partial agonists or antagonists.

Semaglutide's medicinal chemistry diverges from native GLP-1(7-36) at three points: Aib at position 2 (helix nucleation + DPP-IV blockade), Arg at position 34 (reduced renal clearance), and a C18 fatty diacid via γGlu-γGlu spacer at Lys-20/26 (albumin binding, ~1-week half-life in humans). The Aib-2 substitution is directly relevant here: it establishes that a Cα,α-disubstituted residue at the N-terminal dipeptide is not only tolerated but deliberately engineered, providing the conceptual foundation for the Fold №52 hypothesis.

Modification hypothesis (what we tested)

Fold №52 replaces His-1 with α-methyl-L-histidine (αMe-His) — a Cα,α-disubstituted histidine bearing a methyl group at the α-carbon. The hypothesis: combining αMe-His-1 with the existing Aib-2 creates a consecutive Cα,α-disubstituted dipeptide that cooperatively pre-organizes the N-terminal cap into the helical conformation observed in the GLP-1R–bound state, reducing the entropic cost of binding and potentially improving intrinsic receptor activation efficiency. The imidazole side chain is chemically intact, so all direct His-1–TMD contacts are nominally preserved.

This modification targets conformation rather than stability — a deliberate departure from Folds №15 and №36, which tested central-helix salt-bridge geometry (Glu-16 → homoglutamate) and lipidation spacer chemistry (γGlu-γGlu → β-Ala-β-Ala), respectively. All three semaglutide distillations have now returned DISCARDED verdicts, but for distinct mechanistic reasons.

Why the prediction was uninformative (technical analysis of the metrics)

pLDDT = 0.72 places the predicted structure at the lower edge of moderate confidence. For a 33-residue peptide, this is not catastrophic, but it is noticeably below the pLDDT > 0.75 threshold the Researcher identified as the target for a confident verdict. The moderate score almost certainly reflects the predictor's handling of non-canonical residues: AlphaFold-class models are trained predominantly on canonical amino acids, and αMe-His is typically represented as its nearest canonical surrogate (histidine) with no information about the α-methyl constraint. The structural geometry of the N-terminal cap is therefore likely mis-modeled at precisely the residue of interest.

pTM = 0.82 / ipTM = 0.84 are superficially encouraging interface scores. However, in the absence of Chai-1 agreement data — which would cross-validate the predicted binding mode against an independent model — these values cannot be interpreted as evidence of productive receptor engagement. High ipTM in a single-model, single-run prediction without ensemble agreement is not a reliable binding signal.

Boltz-2 affinity module: no values. This is the critical gap. The entire biological hypothesis rests on whether αMe-His-1 improves receptor binding affinity. Without a predicted binding change (ΔΔG or binding probability), the fold cannot speak to its own core question. The two prior semaglutide folds faced the same outcome:

The pattern across all three semaglutide distillations is consistent: moderate pLDDT, no binding metrics, DISCARDED. This strongly suggests a systematic tool-coverage limitation for heavily modified, lipidated 33-mer peptides with non-canonical residues — not a biological verdict against the modification itself.

The heuristic peptide profile (aggregation propensity 0.129, stability 0.439, half-life moderate-to-long) is sequence-derived and not structurally informative. It confirms semaglutide-class physicochemistry but cannot substitute for binding data.

What this tells us (negative results are data — what does it rule out?)

The DISCARDED verdict does not rule out that αMe-His-1 improves GLP-1R engagement. It rules out that current single-run in silico predictors can answer this question for this modification on this scaffold. That is a meaningful result for lab-wide methodology.

The biological uncertainty it surfaces is real: the GLP-1R TMD binding pocket is sterically constrained around His-1, and the added α-methyl group increases the effective steric volume at the α-carbon. Two outcomes are plausible and cannot be distinguished computationally without better tools:

1. Favorable: Conformational pre-organization dominates; the entropic gain from locking the N-cap outweighs any small steric penalty, net improvement in Kd and/or EC50.
2. Unfavorable: Steric clash in the TMD pocket reduces affinity despite favorable conformational bias; over-rigidification of the N-terminus disrupts the dynamic insertion mechanism.

Cryo-EM structures of GLP-1R in complex with peptide agonists (e.g., PDB 7KI0) show relatively tight packing around His-1 — the steric concern is legitimate and should be a primary consideration in any follow-up experimental design. This fold has not resolved the question; it has confirmed that resolving it requires either MD simulation with explicit non-canonical residue parameterization or direct synthesis.

For the broader semaglutide modification program: three DISCARDED folds across lipidation chemistry (№36), central-helix geometry (№15), and N-cap conformation (№52) collectively suggest that the AlphaFold-class pipeline is not the right primary tool for semaglutide SAR given its non-canonical residue content. This contrasts with tirzepatide Folds №23 and №31, where Cα-methylation and Aib substitutions yielded PROMISING verdicts — likely because tirzepatide's shorter, more canonical scaffold is better handled by the predictor.

Alternative hypotheses to test (avoid the failure mode)

Better computational approaches for this specific question:

• Molecular dynamics simulation (GROMACS/AMBER with GAFF or CHARMM-CGenFF parameters for αMe-His) against the GLP-1R TMD (starting from PDB 7KI0) to directly measure N-terminal conformational dynamics, binding free energy (FEP or MM-GBSA), and steric clash potential.
• Ensemble prediction (≥5 AlphaFold runs with varied random seeds) to assess conformational variance at the N-terminus and distinguish genuine structural flexibility from prediction noise.
• Rosetta FlexPepDock or similar peptide-receptor docking with explicit non-canonical residue parameterization for αMe-His, which handles backbone constraints more rigorously than standard AlphaFold input.

Alternative modification strategies to test the N-cap hypothesis:

• His-1 → D-His: epimerization is a simpler, better-parameterized modification for AlphaFold surrogates, and D-His has precedent in GLP-1 analogs as a DPP-IV-resistant, partial-agonist probe that could inform TMD tolerance at position 1.
• Desamino-His-1: removal of the α-amino group eliminates DPP-IV cleavage while keeping backbone flexibility — the result would isolate the pharmacophore contribution of His-1 without conformational constraint.
• Nτ-methyl-His-1: methylation of the imidazole nitrogen rather than the α-carbon probes a different aspect of the His-1–TMD interaction (H-bond donor capacity) without introducing steric bulk at the backbone.
• Single-point cAMP assay on synthetic αMe-His-1 semaglutide: given that this fold cannot be resolved computationally, direct synthesis and functional testing against HEK293-GLP1R is the definitive experiment. The modification is chemically feasible via solid-phase peptide synthesis with Fmoc-αMe-His(Trt)-OH building blocks.

For the broader semaglutide program, the lab should consider whether future distillations on this scaffold are better served by switching to the MD/FEP computational track rather than the AlphaFold-binding-predictor pipeline, given the consistent tool-coverage limitation observed across Folds №15, №36, and №52.