Testing Perez's 2P Numerical Anomaly: What the Re-Scan Actually Showed

What this article covers#

Four things, kept separate:

1. The 2P edit itself: documented, intentional, published before deployment (Wrapp et al., Science 2020).
2. Our independent re-scan of the 2P region using the Supracode pattern-detection component (rupture density in sliding windows), and what finding an elevation there does and does not show.
3. The baseline problem: the article's two nulls disagree by two orders of magnitude, and the honest comparison, the cross-domain one, puts the vaccine sequence inside natural variation.
4. Mulroney et al. (Nature 2024) on m1Ψ and +1 frameshifting, stated at the strength of that paper, which this site has verified against the source.

All sequence analysis here is our own computation on public sequences. It has not been peer-reviewed or independently replicated.

What is actually shown#

The 2P edit is real, intentional, and documented#

The 2P substitutions are two proline changes at spike positions 986 and 987 (K986P / V987P), at the S2 junction between the central helix and heptad repeat 1. Proline rigidly kinks the backbone; the double substitution locks the spike in its prefusion conformation, which is the antigenic surface a vaccine wants to present. The construct was published by Wrapp, Wang, Corbett, McLellan and colleagues (Science 2020;367:1260-1263, DOI 10.1126/science.abb2507), with the cryo-EM structure deposited as PDB 6VSB at 3.46 Å, before either mRNA product was authorised. Both authorised products use it.

Figure 1: The 2P-stabilised SARS-CoV-2 spike prefusion structure (PDB 6VSB, 3.46 Å). The K986P and V987P substitutions lock the trimer in its prefusion state. Source: Wrapp et al., Science 2020.

Figure 2: Wild-type spike structure with the furin cleavage site and receptor binding motif, the baseline for comparison. Source: SARS-CoV-2 spike structural analyses.

At the nucleotide level, the canonical wild-type codons AAG (lysine) and GTG (valine) become CCG and CCG (proline, twice). Two codons out of roughly 3,800 change, and both become GC-rich. That is the entire sequence-level event. It was published in the patent and product literature at rollout. Nothing about it is hidden, and nothing about it is gain of function: gain of function means enhancing a replicating agent's fitness or pathogenicity, and a stabilisation substitution in a vaccine antigen transcribed from a non-replicating mRNA is the opposite category of intervention. Earlier versions of this article framed the 2P edit as a gain-of-function signature. That framing is withdrawn.

What our scan found#

The Supracode pattern-detection component, as applied here, computes UA/CG ratios in a 21-nucleotide sliding window, measures deviation from the golden ratio φ, and counts threshold crossings ("ruptures") per kilobase. Perez reported in 2020 that the 2P-stabilised vaccine spike was numerically different from wild-type under his metric. We re-scanned the region with our own implementation:

• Vaccine 2P region: 2735.85 ruptures/kb.
• Wild-type 2P region: 2169.81 ruptures/kb.
• Relative difference: +26.1 percent.
• Perez's own reported difference: +31.6 percent.

The two implementations agree in direction and roughly in magnitude on the same six-nucleotide edit. That is the surviving content of the original observation: two implementations of the same family of metric, run by us, both read the edited region higher.

What that does not establish: an independent validation of anything. The edit is published. Pointing a scanner at a region already known, from the published sequences, to contain deliberate codon changes, and recording that the scanner reads that region differently, is a calibration check on the scanner's sensitivity to a known change. It is expected-by-construction. Presenting it as "validation" of the method, as earlier versions of this article did, was circular. The same circularity note was applied to the codon-usage article on this site in September.

Figure 3: Scanner output over the spike region. The vertical bars are φ-deviation ruptures; the 2P site reads higher than wild-type. This is what a scanner reading a known edit looks like.

The baseline problem, stated plainly#

The original article built two null baselines, and they disagree by two orders of magnitude:

• A 30-sequence "article baseline" (20 coronavirus spikes plus 10 housekeeping genes) with mean 22.1 ± 3.4 ruptures/kb.
• A cross-domain baseline of 13 genomes (9 bacteria, 3 fungi, 4 viruses) with mean 3512 ± 237 ruptures/kb.

Against the first, every viral sequence in the study, wild-type spikes included, scores tens to hundreds of standard deviations out. The original results table printed Z-scores of 68 to 91 on this basis, including a Z of 70.8 for the wild-type Wuhan spike, a natural virus, against a null that supposedly included 20 natural coronavirus spikes. A null that flags the very sequences it was built from is not a usable null. The Z-scores are withdrawn and should not be quoted anywhere.

The two baselines also measure different things with, apparently, different units or window scales, and the original article never reconciled them. Until the scanner's units are pinned down and the baselines are rebuilt by someone other than the scanner's author, no significance statistic from this method is quotable. Directional comparisons within a single implementation remain readable; absolute claims do not.

The cross-domain comparison, which is the honest headline#

Against the 13-genome cross-domain baseline, the full Pfizer vaccine spike reads 3823 ruptures/kb versus a natural mean of 3512 ± 237. That is Z = 1.35, p = 0.177. Inside natural variation, at the high end.

This is the result that kills the article's original thesis. If the vaccine sequence sat outside the natural range, that would be a finding. It does not. It sits where a GC-elevated, codon-optimised sequence would be expected to sit under this metric. Earlier versions carried this result in a table row marked "Revised" while the headline still claimed validation. The correct ordering is the reverse: the cross-domain result is the headline, and it does not support an "unnatural sequence" claim. Our own data refute our own framing, which is why the framing is withdrawn.

Figure 4: Cross-viral rupture-density comparison. The vaccine spike reads high among these sequences; against the full cross-domain natural range (3093 to 3975 ruptures/kb) it is interior, not exterior.

Mulroney et al. (2024), stated at the strength of the paper#

Mulroney, Pöyry, Yam-Puc and colleagues (MRC Toxicology Unit, Cambridge; Nature 625:189-194, PMID 38057663, DOI 10.1038/s41586-023-06800-3) showed that full substitution of uridine with N1-methylpseudouridine (m1Ψ) increases +1 ribosomal frameshifting. What the paper reports:

• In a reporter construct, frameshifted product accumulated to about 8 percent of the corresponding in-frame protein. This is an accumulation ratio in a reporter, not a translation rate, and not a figure for the vaccine sequence as a whole.
• Frameshifting mapped to specific slippery sequences; mutating them synonymously eliminated detectable frameshifting. It is a sequence-design problem, not an unavoidable property of the chemistry.
• Frameshifted products were detected in BNT162b2-vaccinated mice, and human BNT162b2 recipients showed immune reactivity to frameshifted peptides that ChAdOx1 recipients did not, in the sample tested.
• No adverse outcomes were reported in the donors tested.

Figure 5: Uridine, pseudouridine, and N1-methylpseudouridine. Source: RSC Chemical Biology 2024, DOI 10.1039/d4cb00022f.

Figure 6: Mulroney et al. Figure 1: m1Ψ substitution causes +1 frameshifting at slippery sequences, with immune reactivity to frameshifted peptides in vaccinated donors. Source: Nature 2024.

Figure 7: Mechanism schematic: ribosomal pausing at slippery motifs under m1Ψ substitution, leading to +1 frameshift products. Source: Nature News coverage of Mulroney et al. (2023).

Figure 8: Translation-effects schematic corresponding to Figures 6 and 7.

One earlier claim in this article, that +1 frameshifting is exclusive to synthetic mRNA because all natural viruses use -1 programmed frameshifting, is false as stated. Programmed +1 frameshifting exists in nature; the expression of the bacterial release-factor-2 gene prfB depends on it and is the standard textbook example. Whether any natural virus uses +1 frameshifting at comparable efficiency is a separate question this article does not need: the false absolute is withdrawn on the prfB evidence alone.

The HIV gp120 negative result survives#

The scan also covered HIV-1 gp120 (subtypes B and A, accessions AAA44191.1 and AAT67478.1). It produced diffuse, non-localised rupture patterns, unlike the concentrated hotspot at the 2P site. This does not support the Pradhan et al. (2020) insert claim (a preprint, withdrawn, DOI 10.1101/2020.01.30.927871). A negative result from our own scanner is worth exactly as much as the scanner, and no more; it is recorded because earlier versions used it, and it cuts against the insert narrative rather than for it.

Figure 9: HIV-1 gp120 core in complex with CD4 and the 17b antibody, showing the V1 to V5 variable loops. Source: Kwong et al., Nature 1998;393:648-659, DOI 10.1038/31405.

What is only suggested (labelled hypothesis)#

φ-harmony as a "universal principle"#

The cross-domain scan found rupture densities of roughly 3000 to 4000/kb across 13 bacterial, fungal, and viral genomes, a fairly narrow band. The original article called this a new universal principle of genomic organisation. That is an overstatement of an unreplicated observation from our own instrument. Rupture density under this metric plausibly tracks GC content, genome composition, and window-scale parameters; none of that was controlled. Until an independent group with a pre-registered protocol reproduces the cross-domain band, the honest statement is: one scanner, run by its author's analyst, reads a narrow band across 13 genomes, cause unknown.

2P-region RNA structure as a frameshifting contributor#

The CCG codons at the 2P site are GC-rich, and GC-rich stretches can form local secondary structure. It is reasonable to ask whether the 2P region contributes to ribosomal pausing. No measurement in Mulroney et al. or anywhere else in the verified literature localises frameshifting to the 2P site; the mapped slippery sequences were elsewhere in the constructs. Proposed sequence; untested.

Platform-wide transfer of the frameshifting property#

Every clinical-stage m1Ψ-substituted mRNA therapeutic carries the chemistry Mulroney characterised, so the frameshifting property transfers by construction. The clinical consequence does not: no study has connected frameshifted products to a clinical event, and Mulroney et al. reported no adverse outcomes in the donors tested. Rubio-Casillas et al. (Int J Biol Macromol 2024, PMID 38583833, DOI 10.1016/j.ijbiomac.2024.131427) is a narrative review of the m1Ψ-cancer question, framed by its own title as "friend or foe"; it is a discussion of hypothesised mechanisms, not new primary data. Earlier versions cited it for a specific melanoma tumour-growth finding; that specific claim was removed as unverified against the paper.

Figure 10: The 2023 Nobel Prize in Physiology or Medicine was awarded to Karikó and Weissman for nucleoside base modifications, the chemistry underlying m1Ψ substitution.

Figure 11: The clinical-stage mRNA cancer-vaccine landscape, all of it built on full m1Ψ substitution.

Counter-evidence and what would falsify#

Practical implications, stated at the strength of the evidence#

For sequence design, one actionable item exists and it comes from Mulroney et al., not from Supracode: screen constructs for slippery sequences and recode synonymously, which that paper showed eliminates detectable frameshifting. Any mRNA programme can do this now.

For Supracode as a quality-control screen, nothing actionable exists yet. It would need: open-sourced code, a pre-registered baseline with stated units, blinded test sets including engineered and natural sequences, and replication by a group that did not build it. Until then it is a hypothesis about a metric, with no clinical claim attached, and no clinical claim follows from this article for anyone who received these products.

Perez, credited#

Jean-Claude Perez developed the Supracode framework (Fibonacci and golden-ratio analysis of nucleotide composition) across decades of work, from L'ADN décrypté (1997) onward, and reported the 2P numerical observation in 2020, work criticised at the time for lacking statistical framing. This article applies only the pattern-detection component. His broader Master Code and Codex Biogenesis frameworks are not tested here and not endorsed here. His publications: Semantics Scholar author page. The 2020 SARS-CoV-2 application with Montagnier (Exogenous Informative Elements, Int J Res 2020) is the precursor to the observation re-scanned above; this site's assessment of the EIE insert claim itself is in the gain-of-function article's HIV-gp120 section, and it is negative.

Perez's directional observation on the 2P region survives re-scanning. That is the extent of what this re-analysis establishes about his work, and it is a smaller thing than the original article claimed.

References#

1. Wrapp D, Wang N, Corbett KS, et al. "Cryo-EM structure of the 2019-nCoV spike in the prefusion conformation." Science 2020;367(6483):1260-1263. DOI 10.1126/science.abb2507. PDB 6VSB.
2. Mulroney TE, Pöyry T, Yam-Puc JC, et al. "N(1)-methylpseudouridylation of mRNA causes +1 ribosomal frameshifting." Nature 2024;625(7993):189-194. DOI 10.1038/s41586-023-06800-3 · PMID 38057663.
3. Kwong PD, Wyatt R, Robinson J, Sweet RW, Sodroski J, Hendrickson WA. "Structure of an HIV gp120 envelope glycoprotein in complex with the CD4 receptor and a neutralizing human antibody." Nature 1998;393(6686):648-659. DOI 10.1038/31405.
4. Rubio-Casillas A, et al. "Review: N1-methyl-pseudouridine (m1Ψ): Friend or foe of cancer?" Int J Biol Macromol 2024;267(Pt 2):131427. DOI 10.1016/j.ijbiomac.2024.131427 · PMID 38583833.
5. Pradhan P, Pandey AK, Mishra A, et al. "Uncanny similarity of unique inserts in the 2019-nCoV spike protein to HIV-1 gp120 and Gag." bioRxiv 2020, withdrawn. DOI 10.1101/2020.01.30.927871.
6. Perez JC. L'ADN décrypté (1997) and subsequent Supracode publications. Semantics Scholar.
7. Perez JC, Montagnier L. Exogenous Informative Elements in SARS-CoV-2. Int J Res 2020.

NCBI sequences used: HIV-1 gp120 subtype B (AAA44191.1), subtype A (AAT67478.1), SARS-CoV-2 Wuhan spike (NC_045512.2), Pfizer 2P vaccine spike (OR134577.1), SARS-CoV-1 spike (NC_004718.3), MERS-CoV spike (NC_019843.3).

Related posts#

• Amyloid Fibrin, Mass Casualty, and the Crisis of Misdiagnosis
• mRNA Vaccine DNA Contamination: SV40 Promoter & Integration Risks
• The Molecular Wrecking Ball: HIV-Protein Functional Analogy in SARS-CoV-2
• Pinsolle's Framework: Spike, mRNA, and the Lab-Origin Question
• Genomic Under Siege: Mutagen Defense in the Age of Persistent Spike

Published 29 March 2026.