DISTILLATION №59 — FAILED

Humanin N-terminal myristoylation for mitochondrial membrane targeting / BAX antagonism

TLDR

Fold #59 was DISCARDED due to a tool-limit failure: Boltz-2 exited without producing a PDB file when presented with N-terminal myristoylated Humanin (Myr-MAPRGFSCLLLLTSEIDLPVKRRA) in complex with BAX. This is a non-standard chemical entity failure — the myristoyl amide bond to Met-1 falls outside the canonical amino acid alphabet that current AlphaFold-derived pipelines handle natively. No structural verdict on the hypothesis was reached. This is not a biological invalidation.

What we tried

Humanin (HN, MAPRGFSCLLLLTSEIDLPVKRRA) is a 24-amino acid mitochondria-derived peptide whose direct binding to BAX — preventing its conformational activation and translocation to the mitochondrial outer membrane (MOM) — is one of the most mechanistically supported anti-apoptotic interactions in the longevity peptide literature (Guo et al. 2003, Morris et al. 2019).

The hypothesis in Fold #59 was that covalent attachment of myristic acid (C14 saturated fatty acid) to the α-amine of Met-1 via an amide bond would enrich HN at the MOM — the precise site where BAX inserts and oligomerizes during apoptosis induction. The rationale draws on the validated membrane-targeting function of myristoylation in Src-family kinases, BID/tBID, and MARCKS, and on the Morris et al. 2019 finding that BAX's membrane-targeting C-terminal helix is required for HN-induced fiber formation, implying the HN–BAX interaction is intrinsically membrane-proximate. The expectation was that Boltz-2 would model the Humanin helix (residues 5–19) docking into the BAX hydrophobic groove with the myristoyl chain projecting toward a hydrophobic surface patch, yielding pLDDT >0.65 in the helical core.

This fold is the third in the lab's Humanin series, following Fold #22 (disulfide cyclization Cys-8/S14C → PROMISING, pLDDT 0.56) and Fold #37 (S7A substitution for receptor selectivity → DISCARDED, pLDDT 0.62). Those two folds addressed structural pre-organization and target selectivity, respectively. Fold #59 introduced a third strategic axis: pharmacokinetic membrane delivery.

Why it was discarded

Boltz-2 exited without generating a PDB file. Chai-1 cross-validation was not run. The primary cause is the non-standard chemical entity: N-terminal myristoylation is a covalent lipid modification that requires explicit small-molecule parameterization (e.g., a CCD ligand code, SMILES input, or modified residue definition) that current AlphaFold-derived folding pipelines do not handle within the standard amino acid alphabet. The myristoyl amide at Met-1 cannot be represented as a sequence character, and without a coordinate template or ligand definition, the predictor cannot place or score it.

This is the same class of failure seen in Fold #44 (palmitoyl-γGlu-Lys lipidation on Epitalon, pLDDT 0.34, DISCARDED) and Fold #58 (PEG2-TAT chimera on Epitalon, FAILED) — both involving non-peptidic conjugates that exceed current tool resolution. The pattern is consistent: lipid-modified and PEGylated peptides are a systematic blind spot for this generation of structure predictors.

No structural metrics were produced. Heuristic sequence analysis of the unmodified Humanin backbone yielded: aggregation propensity 0.31 (hotspot residues 7–20, consistent with the known amphipathic helix), stability score 0.37, BBB penetration 0.16, and a moderate-to-long estimated half-life (1–6 hours). These describe the native peptide core only and carry no information about myristoyl tail geometry, BAX groove compatibility, or lipid-mediated membrane partitioning.

What this doesn't mean

FAILED is not "disproved." The myristoylation-mediated membrane targeting hypothesis for Humanin has not been evaluated — it has been rendered invisible by current tool limitations. The biological logic remains coherent: BAX engagement is a membrane-proximate event, myristoylation is a validated MOM-targeting strategy, and co-localizing HN with BAX at the site of insertion is a mechanistically grounded approach. There is also genuine biological tension in the hypothesis — Guo et al. 2003 showed HN prevents BAX translocation to the MOM, implying cytosolic inhibition is the primary mechanism, and membrane anchoring could theoretically arrive too late in the activation sequence. Morris et al.'s in vitro fiber data (generated without membranes) further supports cytosolic HN-BAX engagement as real and substantial. These are scientific questions worth adjudicating experimentally, not computational artifacts. The absence of a structural prediction says nothing about whether myristoyl-HN would bind BAX, partition to the MOM, or enhance anti-apoptotic activity in cells.

What would answer the question

• Solid-phase synthesis + ITC or SPR: Myristoyl-HN is synthetically accessible via N-terminal myristoyl NHS ester coupling. Isothermal titration calorimetry or surface plasmon resonance against recombinant BAX (with and without lipid vesicles to mimic MOM context) would directly measure binding affinity and determine whether the lipid disrupts or preserves the HN–BAX interface. The liposome permeabilization assay format used by Guo et al. 2003 and Morris et al. 2019 is the gold-standard functional readout.
• FEP/MD simulation with explicit lipid bilayer: Free-energy perturbation or molecular dynamics in an explicit DOPC/CL MOM-mimetic bilayer, with the myristoyl chain pre-inserted, could model BAX groove accessibility and myristoyl anchor geometry in silico — bypassing the AlphaFold blind spot for non-standard chemistry.
• Cellular BAX translocation assay: GFP-BAX translocation from cytosol to mitochondria in response to apoptotic stimulus (e.g., staurosporine), with and without myristoyl-HN pre-treatment, would directly test the functional hypothesis in a relevant cellular context.
• Alternative computational tools: RosettaLigand, Glide, or AutoDock with explicit myristoyl parameterization could predict the lipid tail placement and BAX groove binding mode — providing the structural data that AlphaFold-based pipelines cannot currently generate for lipidated peptides.

Raw metrics

Heuristic metrics are sequence-based estimates of the unmodified Humanin backbone only — not real wet-lab measurements, not reflective of the myristoyl modification, and not predictive of BAX binding.