DISTILLATION №26 — DISCARDED

Epitalon head-to-tail cyclization | cyclo(AEDG) | LONGEVITY

Mechanism of action (background)

Epitalon (Ala-Glu-Asp-Gly) is a synthetic tetrapeptide originally derived from the bovine pineal extract Epithalamin, with approximately 25 years of published research across multiple independent labs. Its most molecularly characterised mechanism is upregulation of telomerase reverse transcriptase (TERT): the 2025 Al-Dulaimi et al. study (PMID:40908429) demonstrated dose-dependent increases in hTERT mRNA expression and telomerase enzyme activity in normal epithelial and fibroblast cells, and ALT activation in cancer cell lines. A parallel epigenetic hypothesis (Khavinson et al., PMID:32019204) proposes that the linear peptide engages histone H1 at DNA-interacting basic surfaces via its acidic Glu-Asp dyad, modulating hTERT transcription indirectly. Additional documented effects include antioxidant activity, ROS reduction, melatonin synthesis modulation, interleukin-2 mRNA upregulation, and geroprotective effects in multiple species — a breadth more consistent with upstream regulatory or indirect mechanisms than a single high-affinity receptor interaction.

No experimental three-dimensional structure of Epitalon (linear or cyclic) in complex with TERT or any other proposed target has been published. The bioactive conformation is inferred from molecular modelling, not from experimental structural data — a critical gap that constrains interpretation of all structural predictions.

Modification hypothesis (what we tested)

Fold #26 applied head-to-tail backbone cyclization to Epitalon, forming cyclo(AEDG) via an amide bond between Ala-1's alpha-amine and Gly-4's alpha-carboxylate, yielding a 12-membered macrolactam. This was the most comprehensive Epitalon modification attempted in the lab to date, designed to:

1. Eliminate N-terminal exopeptidase vulnerability — removing the free alpha-amine that was left intact in Fold #21 (C-terminal amidation only)
2. Eliminate C-terminal carboxypeptidase vulnerability — removing the free alpha-carboxylate that was left intact in Fold #6 (D-Ala N-terminal substitution only)
3. Reduce conformational entropy — constraining the four-residue backbone into a defined beta-turn or gamma-turn topology expected to present the Glu-2/Asp-3 acidic dyad in a reproducible orientation
4. Improve structural predictability — the pre-organized turn geometry was hypothesized to give Boltz-2 a sharper structural signal, targeting pLDDT > 0.55 versus the 0.34 floor seen in Folds #6 and #21

The chemical rationale was sound: head-to-tail cyclization of small acidic tetrapeptides is well-precedented (cyclo-RGD analogs, cyclo-RGDS, cyclo-EDDG turn mimetics), and in analogous systems typically confers 10–100-fold metabolic stability improvement against exopeptidases.

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

The pLDDT of 0.341 is statistically indistinguishable from the 0.34 values returned for both prior Epitalon folds, and represents the effective floor of Boltz-2's confidence scoring on sub-5-residue peptides. This is a predictor behavior, not a property of cyclo(AEDG): Boltz-2 and most AlphaFold-lineage models are trained predominantly on full proteins and longer peptides from the PDB. The coverage of 12-membered cyclic macrolactams — particularly those lacking canonical secondary structure — is sparse in training data, and the model has no reliable basis for assigning per-residue confidence in this regime.

The ipTM of 0.209 (threshold for a confident binding prediction is typically ≥ 0.5–0.6) means the TERT complex was not modeled with any confidence. This could reflect: (a) the peptide genuinely failing to adopt a stable bound pose, (b) the model lacking sufficient cyclic peptide parameterization to generate one, or (c) the absence of an experimental TERT–Epitalon structure in the training set leaving the model without a template. All three explanations are plausible; they cannot be distinguished by in silico means alone.

The absence of Chai-1 agreement data removes the most important cross-predictor check available in this lab's workflow. When Boltz-2 and Chai-1 disagree, it flags ambiguity; when they agree at low confidence, it confirms the floor. Neither signal is available here, leaving the verdict dependent on a single low-confidence run — which is the minimum evidence threshold for a DISCARDED classification.

The heuristic profile (aggregation propensity 0.073, stability 0.441, half-life 15–45 min) is entirely sequence-derived and does not account for cyclization. The half-life estimate is almost certainly an underestimate for the cyclic form: backbone cyclization removes both exopeptidase handles, and real-world cyclic tetrapeptide analogs routinely show dramatically extended plasma stability versus linear parents. This is chemistry the heuristic cannot model.

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

Three folds, one floor. Folds #6, #21, and #26 have collectively established that:

• The AEDG tetrapeptide sequence, regardless of terminal modification or backbone topology, sits below the effective resolution limit of Boltz-2 for confident structural prediction
• This is not a property of any single modification strategy — it is a property of the peptide's length and charge composition in the context of current tools
• Continued single-run Boltz-2 prediction on AEDG-based sequences is unlikely to return actionable confidence scores without a fundamental change in approach

What this does NOT rule out:

• The biological activity of cyclo(AEDG) — the literature strongly supports linear AEDG's telomerase effects, and cyclic constraint may enhance or preserve this activity
• The metabolic stability benefit of cyclization — this is well-established chemistry that the predictor cannot assess
• The relevance of the Glu-Asp dyad as a pharmacophore — the negative structural signal is a tool failure, not evidence against the binding hypothesis
• The value of Epitalon as a therapeutic target — the Al-Dulaimi 2025 data are compelling and independent of our structural predictions

What this does suggest:

• Epitalon's mechanism may not require a single defined bound conformation, consistent with the literature's epigenetic/indirect mechanistic hypothesis (flexible histone engagement at multiple orientations)
• If conformational flexibility is intrinsic to Epitalon's mechanism, rigid cyclization could reduce activity even if it improves stability — a testable wet-lab hypothesis

Alternative hypotheses to test (avoid the failure mode)

Given that three iterations have established a hard ceiling for in silico structure prediction on bare AEDG, the lab should pivot away from modification-of-AEDG approaches and consider:

1. Scaffold grafting: incorporate AEDG as a loop into a larger structured peptide Embed the AEDG sequence as an exposed loop within a helical or beta-sheet scaffold peptide of 12–20 residues. This would give structure-prediction tools enough backbone to generate confident models while preserving the Glu-Asp pharmacophore. Precedent: RGD grafted into knottin scaffolds, cyclic peptide libraries with constrained pharmacophoric loops.

2. Peptidomimetic approach: N-methylation or beta-amino acid incorporation Introduce N-methylation at Gly-4 or an alpha-methyl group (Aib) at Ala-1 within the cyclic scaffold to drive turn nucleation. These modifications can shift the preferred macrocycle geometry enough to produce a distinct prediction signal, and are chemically straightforward for cyclic tetrapeptides.

3. Wet-lab prioritization over in silico prediction Cyclo(AEDG) synthesis is chemically tractable (solution-phase or SPPS with on-resin cyclization). Given that structure prediction has reached its limit, empirical telomerase activity assay (hTERT mRNA qPCR, TRAP assay) and proteolytic stability comparison (plasma stability HPLC) of linear vs. cyclo(AEDG) would directly test the cyclization hypothesis without requiring structural modeling confidence.

4. Target TERT via a longer, AEDG-presenting peptide discovered by generative design Use a generative approach (e.g., RFdiffusion or ProteinMPNN-based binder design) to design a structured TERT-binding peptide that incorporates the AEDG acidic dyad geometry, then fold-test the designed sequence with Boltz-2. This bypasses the intrinsic length limitation of the native tetrapeptide.

5. Explore the histone H1 target with longer AEDG-flanked sequences If the epigenetic mechanism (histone H1 binding) is primary, design a 10–14 residue peptide with AEDG at the center flanked by residues predicted to stabilize helix or beta-strand secondary structure, and model against the H1 DNA-binding domain (PDB: 1GHC). This reframes the target and provides a structural context the predictor can work with.

In silico prediction only. All predicted properties require experimental validation. This is not medical advice.