Source documents. The EMA BWP position on residual DNA in Comirnaty (EMA/CHMP/BWP/61303/2023, adopted 31 October 2023, released to the author under ASK-289849 batch 2 via release letter EMA/142071/2026, 19 June 2026). The public GenBank deposit OR134577.1 (McKernan et al., 20 June 2023). Every coordinate and motif call in this article is reproducible from these two sources.
On 19 June 2026, the European Medicines Agency released a document to this author under access-to-documents request ASK-289849. The document is the BWP position on residual DNA in Comirnaty, adopted by the Biologics Working Party at its meeting of 30 to 31 October 2023 (reference EMA/CHMP/BWP/61303/2023).
That document contains three regulatory concessions the EMA committed to in writing. This article uses the EMA's own words as the anchor for a sequence-level annotation of the BNT162b2 plasmid that any reader can verify from the public GenBank record.
The public does not need to take the author's word for what is in the vaccine plasmid. The sequence is public. The EMA's position is now public (released under ASK-289849). The reader can check both.
Three concessions in the BWP position paper frame everything that follows. The quotes are verbatim from the released document.
Concession 1. The SV40 elements are in the plasmid.
Page 4 (Additional information, 24 October 2023):
"SV40 polyA signal, SV40 Promoter/Enhancer, including SV40 Origin are present in the DNA plasmid, which is a starting material in the manufacturing process of BNT162b2 / Comirnaty Drug Substance."
This is the EMA on the record. The "no SV40 in Comirnaty" position is closed. All four SV40 elements are present: the promoter, the enhancer, the origin of replication, and the polyadenylation signal.
Concession 2. The manufacturer did not disclose the SV40 content at the time of the original authorisation.
Page 5 (BWP assessment):
"While the full DNA sequence of the plasmid starting material was provided in the initial marketing authorisation application for Comirnaty, the applicant did not specifically highlight the SV40 sequence, as it was considered to be a non-functional part of the plasmid. They have since clarified this information in response to questions raised by EMA."
The full DNA sequence was provided. The SV40 sequence was not highlighted. The EMA considers this a non-disclosure that required subsequent "clarification." The regulatory record now states that the SV40 content was not identified at MAA.
Concession 3. The testing regime does not measure what is in the syringe.
Page 5 (Rationale for testing at Drug Substance level):
"Analysis of finished drug product (DP) samples requires additional sample manipulation, including extraction of residual DNA from the LNP, which may impact the ability to sensitively detect and characterize residual DNA."
Translation: the regulatory compliance number (10 ng DNA per 30 microgram RNA dose) is measured on the Drug Substance, the upstream intermediate before lipid nanoparticle encapsulation. The Drug Product, what is actually in the syringe and injected into people, is not the material being tested. The justification is technical (the LNP makes DNA harder to extract), but the consequence is that the number on the compliance certificate does not measure what the patient receives.
The public GenBank deposit OR134577.1, recovered from bivalent vaccine vials by Kevin McKernan and colleagues in June 2023, is the complete 7,810 base pair sequence of the BNT162b2 production plasmid. Eleven genetic elements are identifiable by motif scan against public reference databases:
| Element | Position | Length | Function |
|---|---|---|---|
| SV40 large T-antigen C-terminal fragment | 165 to 501 | 336 bp | 3' end of T-ag, travels with SV40 poly(A) |
| SV40 CPD phospho-degron (KKPPTPPPEPET) | 319 to 354 | 35 bp | Inherited from T-ag, no upstream ATG |
| SV40 enhancer / promoter / origin | 1096 to 1453 | 358 bp | Drives mammalian expression of KanR |
| Kanamycin / neomycin resistance | 1487 to 2304 | ~817 bp | Bacterial and mammalian selection |
| BsaI Type IIS assembly scar | 2610 | 6 bp | Golden Gate assembly remnant |
| T7 promoter | 3603 to 3619 | 17 bp | Drives in vitro transcription of spike mRNA |
| Spike CDS (BA.4/5 bivalent insert) | 3673 to 7479 | 3,807 bp | The vaccine mRNA product |
| AarI Type IIS linearisation site | 7583 | 7 bp | Linearisation for in vitro transcription |
| SapI Type IIS assembly scar | 79 | 7 bp | Assembly remnant |
| SV40 polyadenylation signals (3 signals) | within 220 to 501 | ~282 bp | Native SV40 poly(A) cassettes |
| f1 / ColE1 origin | remainder | ~3,000 bp | High-copy bacterial replication origin |
The mRNA vaccine product (the T7 transcript) is spike-only. The SV40, KanR, and backbone elements are in the plasmid DNA template, not in the mRNA. Any exposure route for SV40 and KanR sequences in a vaccinated person is via residual plasmid DNA co-formulated in the lipid nanoparticle, not via the mRNA itself.
This is the sequence-level confirmation of EMA concession 1. The EMA said the four SV40 elements are present. The public sequence shows exactly where they are, how long they are, and what else is around them.
The plasmid contains 695 bp of SV40 sequence (8.9% of the total plasmid) at 100% identity to SV40 strain 776, the ATCC reference. The footprint has two functional blocks.
Block 1 (positions 165 to 501): the T-ag 3' fragment and poly(A) cassette. The 3' terminal 138 nucleotides of the SV40 large T-antigen gene, including a phosphodegron motif at T-ag residues 697 to 708 (KKPPTPPPEPET). Three native SV40 polyadenylation signals are in this block. This is the SV40 poly(A) cassette that the EMA confirmed is present.
The T-ag fragment does not translate. There is no start codon upstream of it, no promoter that would transcribe it, no splice sites, and GC content below the CpG island threshold for active promoters. Translation of T-ag protein from this plasmid is excluded by the sequence evidence.
Block 2 (positions 1096 to 1453): the enhancer, promoter, and origin. The full SV40 regulatory region in functional form. All three T-antigen binding sites. Both 72 bp enhancer repeats (the full enhancer). The early promoter with its 21 bp repeat triplet. The origin of replication intact. This is the SV40 promoter/enhancer and origin that the EMA confirmed are present.
This block drives mammalian expression of the downstream kanamycin/neomycin resistance gene. That is the design intent: G418 selection of transfected mammalian cells.
This is not whole SV40 virus. The plasmid contains less than 7% of the T-antigen gene. None of the Rb-binding, p53-binding, helicase, or ATPase domains. No SV40 miR-S1. No capsid DNA.
The BWP position paper describes the SV40 elements as functionally inert. The BWP assessment (page 5) characterises the SV40 sequence as having been "considered to be a non-functional part of the plasmid."
The public sequence shows this characterisation is architecturally wrong.
The SV40 enhancer/promoter block at OR134577.1 positions 1096 to 1453 ends 34 base pairs upstream of the kanamycin/neomycin resistance CDS at position 1487. In pCMV-Script, the commercial vector from which this backbone derives, this is the designed promoter for mammalian expression of the KanR/NeoR gene. The 358 bp SV40 block is not a vestigial fragment. It is the upstream regulatory element for the downstream selection marker.
This is the design intent of the parent vector. Stratagene's pCMV-Script was engineered so that the SV40 early promoter drives NeoR expression in mammalian cell lines for G418 selection during transient or stable transfection workflows. That functional architecture was inherited unchanged into the BNT162b2 production plasmid through the pST vector family (Holtkamp et al. 2006).
The regulatory frame that calls SV40 "non-functional" is looking at the bacterial propagation step only. In E. coli, the SV40 promoter is indeed silent. But the plasmid's purpose is not bacterial propagation. Its purpose is to serve as the DNA template for in vitro transcription of the spike mRNA, and before that, to be manufactured, purified, and quality-controlled in a process where the mammalian selection architecture is part of the vector design. The SV40 promoter is functional in the design context that matters: any mammalian cell that receives this DNA.
The backbone of the BNT162b2 plasmid traces to pCMV-Script, a commercial cloning vector sold by Stratagene in 1997 (GenBank AF028239.1). Three independent sequence tests confirm this.
Test 1. The 3,469 bp backbone block of OR134577.1 matches pCMV-Script as a single continuous block at 99.97% identity. The competing candidate, pcDNA3.1(+), matches only as six disjoint fragments.
Test 2. Every named backbone feature (the SV40 T-ag block, the SV40 enhancer/promoter/ori, the f1 origin, the KanR gene) matches pCMV-Script at 100% over the full feature length. pcDNA3.1(+) matches every feature only partially, with SNPs and truncations.
Test 3. Four diagnostic positions in the kanamycin resistance gene distinguish pCMV-Script from pcDNA3.1(+). OR134577.1 matches pCMV-Script at all four. The probability of pcDNA3.1(+) ancestry is 0.39%.
Primary literature confirmation. Holtkamp et al. 2006 (Blood, PMID 16926288), the founding paper of BioNTech's pST mRNA vector family from the Mainz group of Sahin and Tureci, states that pST1 was built "into the pCMV-Script-Vector (Stratagene, La Jolla, CA)."
The developmental chain:
pSV2-gpt (Mulligan and Berg, PNAS 1981) -> pCMV-Script (Stratagene, 1997) -> pST1 (Holtkamp et al., Blood 2006) -> pST family -> BNT162b2 production plasmid (OR134577.1)
The SV40 content has been in this vector family since 1981. It is not a pandemic-era insertion. It is not a 1997 commercialisation decision. It is the original design of the pSV2 vector family, transmitted through pCMV-Script into BioNTech's pST platform, which has carried it since 2006.
Moderna made a different choice. The Moderna mRNA-1273 plasmid (OR134578.1) contains zero SV40 sequence. Zero base pairs. The same vaccine modality, the same regulatory pathway, the same lipid nanoparticle delivery, and Moderna used an SV40-free backbone. SV40 inclusion was a manufacturer design choice, not a technological necessity.
The two SV40 elements in the BNT162b2 plasmid correspond to the two SV40 functional blocks defined in the founding paper of the pSV2 vector family. Mulligan and Berg, in their 1981 PNAS paper (PMID 6262762), described the SV40 origin/promoter block and the SV40 polyadenylation block at the exact coordinates inherited in OR134577.1.
The same paper documented something else, on pages 2075 to 2076:
"[Transformants] contain one to five copies of the transfecting DNA associated with, and most probably integrated into, cellular DNA sequences."
The pSV2 vector family, from which the BNT162b2 plasmid descends, was characterised in its founding publication as integrating into the host cell chromosome. Integration was not a side effect. It was the basis of the transformation system. The SV40 promoter/enhancer drives marker gene expression from an integrated copy, and stable inheritance of the transformed phenotype requires chromosomal integration.
This property has been documented in the primary literature for 45 years.
The BWP position paper says: "There is no scientific evidence that any of these SV40 fragments can act as insertional mutagens." (page 5). This statement addresses insertion of a functional SV40 virus. It does not address the integration property of the pSV2 vector family documented by Mulligan and Berg. The manufacturer's risk frame is replication. The property documented in the founding paper of the vector family is integration.
The manufacturer's dossier section eCTD 3.2.S.2.3 ("Source, History and Generation of Plasmids") is where the SV40 content, the pCMV-Script ancestry, and the vector genealogy belonged. It identifies the production plasmid as pST4-1525 and contains:
Both figures are redacted in the public release.
The EMA describes pST4-1525 as 7,824 base pairs. The public GenBank deposit OR134577.1 is 7,810 base pairs. A 14 bp difference. The most likely explanations are assembly boundary trim, propagation drift, or a formulation difference. But the only way to resolve the gap definitively is to compare OR134577.1 against the unredacted pST4-1525 sequence.
That sequence exists. It is in Figure 3.2.S.2.3-2. It is redacted.
The BWP position paper (31 October 2023) committed to two actions by 1 December 2023:
"1. Further analytical evaluation of residual DNA on DS level - planned completion: 1 Dec 2023. Analysis and characterization of the size distribution of residual DNA fragments and residual intact circular plasmid. Assessment of the presence of SV40 sequences in residual DNA."
"2. Capability of the residual DNA plasmid replication in bacteria (i.e., transformation capability) - planned completion: 1 Dec 2023."
These were commitments to tests that, as of the document's date, had not been performed. The document is dated October 2023. This article is dated August 2026.
The BWP also said (page 7):
"Confirmatory testing at finished product level by the manufacturer and/or independent testing carried out by OMCL to confirm the levels of residual DNA in commercial batches of Comirnaty could also be considered."
"Could be considered" is the regulator's way of saying "is not currently required." As of the date of this article, finished-product (Drug Product) testing has not been mandated.
A pattern has emerged in the regulatory and publishing response to the residual DNA question. Safety arguments constructed for the mRNA product are applied to the residual plasmid DNA as if the two molecular species were interchangeable. They are not.
The mRNA is the intended product. It is transcribed in vitro from the plasmid template, capped, polyadenylated, purified, and encapsulated in lipid nanoparticles. It is designed to be translated in the cytoplasm. It does not enter the nucleus. It does not integrate. The safety profile of a non-integrating, cytoplasmic mRNA is well established.
Residual plasmid DNA is a process contaminant. It is the template itself, carryover from the in vitro transcription reaction. It is a different molecular species with different properties. It is double-stranded DNA, not single-stranded RNA. It contains the full plasmid backbone: the SV40 enhancer/promoter, the KanR/NeoR cassette, the f1 origin, and the bacterial replication origin. It is stable in ways the mRNA is not. It can be transported into the nucleus. It can persist.
The category error was documented in a recent commentary by Polykretis et al. (2026, Oncotarget) on the peer-review response to a residual DNA case report. Two rejection letters illustrate the error. The first stated that "mRNA vaccines only synthesize antigenic proteins in the cytoplasm and do not involve gene integration or replication" and concluded that "COVID-19 mRNA vaccines do not enter the cell nucleus, and thus cannot cause cancer." This is true of the mRNA. It is not an argument about the residual DNA. The second stated that "insertional mutagenesis is a well-known risk for certain viral vector-based gene therapies" but dismissed the concern because "mRNA vaccines are non-integrating by their fundamental design."
The residual DNA is not the mRNA. A safety argument that begins and ends with the properties of the mRNA product does not address the properties of the contaminant. The regulatory question is whether a double-stranded DNA molecule carrying an SV40 enhancer/promoter and a mammalian selection cassette, delivered inside a lipid nanoparticle, presents a different risk profile from the mRNA it accompanies. That question has not been answered by being assumed away.
You do not need to trust this article. You can reproduce every coordinate and every identity claim from the public record.
The EMA BWP position paper is available from the author on request (released under ASK-289849) or directly from the European Medicines Agency under Regulation (EC) No 1049/2001.
This article does not claim that SV40 large T-antigen protein is produced from the BNT162b2 plasmid. The sequence evidence excludes T-ag translation (no start codon, no promoter, no splice sites).
This article does not claim that whole SV40 virus is present in the vaccine. Less than 7% of the T-ag gene is in the plasmid. No capsid DNA. No SV40 miR-S1.
This article does not claim a deliberate pandemic-era SV40 insertion. The SV40 content is a 1981 legacy feature of the pSV2 vector family, inherited through pCMV-Script (1997) into the pST family (2006). It is a long-standing platform design choice that survived fourteen years of development.
What this article does claim is that:
The disclosure question is whether the integration property of this vector family, documented in the primary literature for 45 years, should have been addressed before the product was administered to billions of people.
Comments and corrections from readers with primary-sequence evidence are welcome. The companion preprint contains the full technical detail, methods, and tables.
]]>Where this sits. This article is the "tolerance corrupts" node of the spikeopathy framework. It supplies a candidate mechanism for the clearance failure documented in Spike Persistence: Microclots, Reactivated Viruses, and Failed Clearance. The mechanism is real and peer-reviewed (Zhang et al., Nature 2025). The specific link to vaccine forensics is a hypothesis. Tag [HYPOTHESIS], confidence LOW-MODERATE until the bridging work in Part 5 lands.
Detection is not clearance. That is the entire thesis in four words, and the rest of this article unpacks why those four words may matter more than the vaccine debate has so far admitted.
A December 2025 paper from Edgar Engleman's group at Stanford (Zhang et al., Nature 2025, PMID 41372415) establishes a single signalling axis, EPO through EPOR on type 1 conventional dendritic cells (cDC1s), as the master switch between two immunological outcomes. When EPO-EPOR is active on cDC1s, the dendritic cells mature into a tolerogenic state and activate regulatory T cells (Tregs). When EPO-EPOR is blocked, the same dendritic cells mature into an immunogenic state and activate killer CD8+ T cells. The antigen is the same. The danger signals are the same. The outcome is decided by one hormonal switch.
Read alongside the cGAS-STING DNA-sensing literature, this places an uncomfortable question on the table. Residual plasmid DNA in mRNA vaccines (documented by Kämmerer and Steger 2024 at 32.71 to 43.38 ng per clinical dose, confirmed present by EMA under ASK-289849, with the full 11-feature plasmid map now independently annotated) is a classical trigger for cGAS-STING inflammatory sensing. cGAS-STING fires loudly. The Stanford finding implies the downstream decision, tolerance or clearance, is gated by EPO-EPOR. If EPO-EPOR is active in the wrong window, the contaminant is detected but the responding cell population is told to stand down.
This article does not claim that is what happens. It claims the mechanism is real, the components are documented, and the bridge between them is a testable hypothesis with explicit refutation criteria. The confidence level is LOW-MODERATE for the vaccine-specific link and HIGH for the underlying Stanford finding.
Evidence Level: [AN/PR] (animal and primary human ex vivo) Confidence: HIGH for the mechanistic claim; MODERATE for in-vivo clinical translation
Zhang and colleagues, working in Edgar Engleman's laboratory at Stanford, published in Nature in late 2025 that erythropoietin signalling through the erythropoietin receptor on type 1 conventional dendritic cells acts as a master regulator of the tolerogenic versus immunogenic maturation decision. The paper is:
Zhang J, et al. Erythropoietin receptor on type 1 conventional dendritic cells dictates immune tolerance. Nature. 2025. PMID 41372415. DOI: 10.1038/s41586-025-09824-z. PMC12929016.
The key findings, in plain language:
EPOR is expressed on cDC1s. This was not previously appreciated. EPO was thought of as a red-cell hormone. The Stanford group showed that cDC1s, the dendritic-cell subset responsible for cross-presentation of antigens to CD8+ T cells, express EPOR on their surface.
EPO through EPOR drives cDC1s into a tolerogenic maturation state. When EPO binds EPOR on a cDC1, the dendritic cell matures into a form that activates regulatory T cells (Tregs) rather than killer T cells. The antigen is presented, but the presentation is suppressive.
Blocking EPO-EPOR on cDC1s flips the switch. When EPOR is deleted or blocked on cDC1s, the same dendritic cells mature into an immunogenic form that activates effector CD8+ T cells. In mouse tumour models, EPOR deficiency in cDC1s reduced tumour growth by enhancing anti-tumour T-cell immunity and increased the pool of precursor exhausted tumour-specific T cells.
The mechanism is consistent with a broader tolerance literature. Total lymphoid irradiation (TLI) is known to induce systemic tolerance in transplant and autoimmune settings. TLI elevates EPO and upregulates EPOR on cDC1s. The Stanford finding identifies that axis as the mechanism: TLI works because it opens the EPO-EPOR tolerance gate on cDC1s.
"Cold" tumours use the same axis. Hypoxic, EPO-producing tumours keep the tolerance gate open locally and evade immune clearance even when danger signals are abundant.
The therapeutic implication Engleman's group draws is bidirectional. For autoimmunity and transplant, push EPO-EPOR open to induce tolerance. For cancer, block it to permit clearance.
The implication for vaccine forensics is the one this article exists to draw.
Evidence Level: [PR/PP] for the forensics; [HYPOTHESIS] for the link
The vaccine-forensic side of the bridge is well-documented on this site.
The default reading of these facts in the critical literature is: contaminant DNA, cGAS-STING fires, inflammation follows, bad outcome. That reading is correct as far as it goes. It is incomplete in a way the Stanford finding now exposes.
cGAS-STING is a detection pathway. It produces a danger signal. The Stanford finding establishes that the response to danger signals from dendritic cells is not hard-wired to clearance. The response is gated by EPO-EPOR. If EPO-EPOR is active on cDC1s at the time the danger signal arrives, the dendritic cell matures tolerogenically, presents antigen to CD4+ T cells in a suppressive context, and induces Tregs rather than effectors. The danger was detected. The response was tolerance.
This is not a speculative chain. Each link is documented:
The bridge is the specific window. If the EPO-EPOR gate is open when the cGAS-STING signal arrives from contaminant DNA, the predicted outcome is detection-without-clearance. Spike protein from the vaccine mRNA would be produced, presented by the same cDC1s in a tolerogenic context, and the responding T-cell population would be tipped toward Treg induction rather than effector induction. Spike-loaded cells would be tolerated rather than cleared. Spike persistence would result.
The persistence article on this site documents that spike does persist, in subsets of vaccine recipients, for months to years. Patterson, Ogata, and Hulscher primary literature document the same. The tolerance-gate hypothesis supplies a candidate mechanism for why clearance fails in those subsets.
Evidence Level: [HYPOTHESIS] Confidence: LOW-MODERATE
The specific claim, stated as plainly as possible:
In a subset of vaccine recipients, the post-vaccination tissue environment produces a transient EPO elevation that opens the EPO-EPOR tolerance gate on cDC1s during the window in which contaminant plasmid DNA is being sensed by cGAS-STING. The result is detection-without-clearance: the contaminant is registered as danger, the spike antigen is registered as self-tolerated, and the responding T-cell population is tipped toward tolerance rather than effector clearance. Spike-loaded cells survive. Spike persists.
Subsets matter here. The hypothesis does not require every vaccine recipient to enter the tolerance window. It requires that the window opens in some recipients under some conditions, and that those recipients are overrepresented in the cohort that goes on to develop persistent spike, post-vaccination syndromes, or both.
Plausible conditions that would open the window:
The prediction is that the post-vaccination syndrome cohort is enriched for combinations of these conditions. High local EPO, high EPOR expression, high contaminant DNA, repeat dosing, and a fresh tolerance induction from a prior dose.
This is testable. Part 5 specifies how.
Three observations that the tolerance-gate hypothesis explains cleanly and the alternative explanations handle awkwardly.
4.1 Persistence in a subset, not the cohort.
If contaminant DNA produced direct toxicity or robust clearance, the outcome should be either universal harm (toxicity) or universal clearance (no persistence). Neither is observed. What is observed is persistence in a subset. The tolerance-gate hypothesis predicts exactly that distribution: the gate opens in some recipients and not others, depending on EPO-EPOR status at the time of exposure.
4.2 Detection of spike and antibody class switching without clearance.
Vaccine recipients produce anti-spike antibodies. This is the success criterion for the vaccine and it works. What it does not produce, in the subset that develops persistence, is CD8+ T-cell clearance of spike-loaded cells. Antibody without cellular clearance is the signature of a tolerance-corrupted response, not the signature of a failed response. The tolerance-gate hypothesis predicts exactly this pattern: antibody persists (B-cell arm intact), CD8+ clearance fails (tolerance gate open on the cDC1 cross-presentation arm).
4.3 Tolerance to self-antigens on spike-transfected cells.
Case reports of autoimmune phenomena post-vaccination, including molecular mimicry against self-antigens and breakouts of latent autoimmunity, are documented in the pharmacovigilance record. The tolerance-gate hypothesis does not require tolerance only to spike. If cDC1s are tipped tolerogenic while presenting spike derived from transfected cells, they are also presenting whatever self-antigens those transfected cells express. Tolerance induction is not antigen-specific at the dendritic-cell maturation level. A tolerogenic cDC1 tolerises to everything it presents. The hypothesis predicts that autoimmune phenomena, where they occur, should cluster with spike persistence rather than distribute randomly.
None of these three observations is uniquely explained by the tolerance-gate hypothesis. Each has alternative accounts. The point is that the tolerance-gate hypothesis fits the three observations cleanly, with a single mechanism, while the alternatives require separate explanations for each.
The hypothesis is testable. The bridging work is concrete.
5.1 Humanised mouse model.
Dose humanised mice with a Pfizer-bivalent-equivalent LNP-mRNA preparation at the high end of the Kämmerer DNA range (approximately 40 ng DNA per dose). Sacrifice subsets at days 1, 3, 7, 14, 30. Measure:
Prediction: EPOR upregulation on cDC1s coincides with Treg induction and absent or muted CD8+ activation. Spike and DNA persist past day 30 in a subset.
5.2 EPOR blockade arm.
Repeat the mouse study with a cDC1-targeted EPOR blockade (or use EPOR-floxed x CD11c-Cre mice). Prediction: tolerance gate closes, CD8+ activation rises, spike clearance improves.
5.3 Human cohort study.
Recruit a post-vaccination syndrome cohort and matched controls. Measure baseline EPO, EPOR expression on circulating dendritic cells, Treg frequencies, anti-spike IgG, and spike persistence markers (S1 in monocytes, circulating spike). Prediction: syndrome cohort shows higher baseline EPO, higher EPOR on cDC1s, higher Treg frequencies, and detectable circulating spike, as a correlated signature.
5.4 Batch-stratified pharmacovigilance.
Use the batch-level residual DNA data (where regulators hold it) to stratify adverse-event reports. Prediction: batches at the high end of the DNA distribution show disproportionately elevated reporting rates for syndromes consistent with tolerance-corrupted clearance (autoimmune flares, persistent spike symptom clusters), not just elevated reporting rates of acute reactions.
None of these studies exists in the published literature at present. All four are within reach of a reasonably funded immunology laboratory. None requires any technology that has not already been validated.
Scientific integrity requires naming the observations that would kill the hypothesis. The clearest refutations:
cGAS-STING activation by contaminant DNA produces robust CD8+ T-cell clearance regardless of EPO-EPOR status. If a mouse study shows that contaminant DNA sensing proceeds directly to effector T-cell activation and antigen clearance with no tolerance signal, regardless of EPOR manipulation, the bridge is broken. The Stanford finding stands. The vaccine link does not.
EPOR is not upregulated on cDC1s post-vaccination. If a human or mouse study fails to detect EPOR induction on cDC1s in the post-vaccination window, the gate never opens and the hypothesis fails. The Stanford finding is about baseline EPOR expression and induced upregulation. If vaccination does not induce it, there is no gate to open.
Treg frequencies do not differ between persistent-spike and clearance cohorts. If a cohort study shows no correlation between Treg frequency and spike persistence, the tolerance-gate mechanism is not the operative one in the subset. Some other mechanism produces persistence.
Batch-stratified pharmacovigilance shows no signal. If high-DNA batches do not cluster with tolerance-syndrome reports, the contaminant-DNA-specific arm of the hypothesis fails (although the broader spike-persistence mechanism is unaffected, since spike itself can induce tolerance through other pathways).
Each refutation is independently testable. The hypothesis can fall on any of them.
Calibrated claims survive better than uncalibrated ones. This article explicitly does not claim:
What this article does claim: the Stanford finding is real, published in Nature, and reshapes the framework within which the contaminant-DNA forensics have to be read. Detection is not clearance. The downstream decision is gated. The gate is nameable. The gate is testable.
The tolerance-gate hypothesis is one of three candidate mechanisms for the failure of spike clearance documented across this site. It is not the only one, and it is not claimed to be the dominant one.
The case is not that one of these mechanisms is right and the others are wrong. The case is that they are mutually reinforcing, that each is testable independently, and that the cGAS-STING detection arm is well-documented while the tolerance-decision arm has just been clarified by the Stanford finding.
The default framing in the critical literature is: contaminant DNA, cGAS-STING fires, inflammation, harm. That framing is correct as far as it goes, and it has been useful for getting the issue taken seriously. It is also incomplete in a way that a sophisticated defender of the regulatory position can exploit.
The incomplete framing predicts universal inflammation. The observed outcome is subset-specific persistence, mixed antibody-and-tolerance signatures, and clustered autoimmune phenomena. A sophisticated defender points to the gap between the predicted universal inflammation and the observed subset-specific outcome, and uses the gap to dismiss the mechanism.
The tolerance-gate framing predicts the subset-specific outcome from first principles. It predicts that antibody production proceeds while cellular clearance fails. It predicts that autoimmune phenomena cluster with persistence. It predicts that high-DNA batches produce tolerance-syndrome reports rather than acute-toxicity reports. Each prediction is independently testable.
That is the value of getting the mechanism right. Not rhetorical advantage. Predictive precision. The Stanford finding allows a sharper prediction than the simple cGAS-STING framing allowed, and sharper predictions are what move a debate forward.
Every claim in this article is traceable to a primary source.
The Stanford finding:
The contaminant DNA forensics:
The 11-feature plasmid annotation:
cGAS-STING and cytosolic DNA sensing:
EPO-EPOR physiology and tolerance induction:
The spike persistence record:
The SV40 enhancer and lymphoid mutagenesis:
End of tolerance-gate hypothesis article. This is a living document. If the bridging work in Part 5 lands, the hypothesis tag comes off. If the refutation criteria in Part 6 land, the article is retracted in line with the calibration discipline documented in the Calibrated Counter-Claims piece.
]]>Investigation Protocol. Every calibration below was produced by running the published sequence (OR134577.1 for Pfizer BNT162b2, OR134578.1 for Moderna mRNA-1273) through open-source bioinformatics tools (Biopython, BLAST, ViennaRNA). Every coordinate and motif call is reproducible from the public GenBank records. Where a previously circulated claim does not reproduce, the reason is documented.
Most "fact check" articles in this space target claims from the other side. That is the easy work. The harder and more useful work is to correct overreach from your own side, because that is where your credibility is most expensive to lose.
The seven claims calibrated below have all circulated in independent research circles in 2024 to 2026. Some appeared in early versions of this author's own working documents. Each contains a kernel of real observation wrapped in an error of mechanism, magnitude, or attribution. In every case the corrected version is weaker than the original claim, and in every case the corrected version is still scientifically interesting.
The rule throughout: if the sequence evidence does not support the claim, the claim gets cut, regardless of who made it or how useful it was rhetorically. That is the only standard that survives contact with a sophisticated opponent.
| # | Original claim | Calibrated finding | Status |
|---|---|---|---|
| 1 | "144 G-quadruplexes in ORF19" | Zero canonical G-quadruplexes anywhere in the plasmid | Retracted |
| 2 | "Pfizer-specific prion ORF, probability 4e-11%" | Both vaccines have antisense-spike ORFs with similar GxxxG density; probability does not reproduce | Retracted |
| 3 | "76.92 frameshift motifs per kb in Pfizer spike" | 0.79 motifs per kb (units bug in CDS extraction) | Corrected |
| 4 | "Mulroney 2023, PMID 37603869, in Nature" | Mulroney 2024, PMID 38057663, in Nature Biotechnology | Corrected |
| 5 | "35% seroprevalence of anti-frameshift antibodies" | No primary citation located | Unverified, do not repeat |
| 6 | "PRRAR vaccine modification at the FCS" | Normal Omicron BA.4/5 lineage sequence in the bivalent component | Corrected |
| 7 | "OR134577.1 is the BNT162b2 Wuhan-spike plasmid" | OR134577.1 is the BA.4/5 component of the bivalent vaccine | Corrected |
The details follow.
The original claim. An open reading frame on the antisense strand of the BNT162b2 plasmid, designated ORF19, contains 144 G-quadruplex (G4) motifs, suggesting extreme prion-like or amyloidogenic potential.
The calibrated finding. A canonical G-quadruplex motif is defined as G3+N1-7G3+N1-7G3+N1-7G3+, that is, four runs of at least three guanines separated by loops of one to seven nucleotides. Scanning all 7,810 bp of OR134577.1 in both strands for this pattern returns zero canonical G4 motifs anywhere in the plasmid.
What the original analysis counted was 124 individual G-tracts (runs of three or more guanines) on the minus strand. These are the raw ingredients from which a G4 could fold, not folded G4 motifs themselves. Counting G-tracts as G4 is a category error.
What survives. The antisense-spike ORF does have an elevated GxxxG prion-domain motif density (see claim 2 below), which is a separate and legitimate observation. The G4 claim does not survive.
Citation. Calibrated by Biopython regex scan of the public GenBank record OR134577.1. The canonical G4 pattern (G3+N1-7G3+N1-7G3+N1-7G3+) returns zero hits in all 7,810 bp on both strands.
The original claim. An ORF found only in the Pfizer plasmid has a prion-domain probability of 4 x 10^-11 percent, implying astronomical prion risk unique to BNT162b2.
The calibrated finding. Two problems.
First, the ORF is not Pfizer-specific. The Moderna plasmid (OR134578.1) has a comparable antisense-spike ORF with a similar GxxxG prion-domain density. Under a Poisson null with glycine fraction 0.072, the Pfizer antisense-spike ORF (1252 aa) has 19 GxxxG motifs at P = 4.76 x 10^-5. The Moderna antisense-spike ORF (403 aa) has 10 GxxxG at P = 6.12 x 10^-5. These are the same order of magnitude. The Pfizer-only framing was an artefact of comparing the full Pfizer ORF against the shorter Moderna ORF.
Second, the 4 x 10^-11 percent figure does not reproduce under any standard prion-domain probability model. The closest reproducible number is P = 4.76 x 10^-5, which is six orders of magnitude weaker.
What survives. Both vaccine plasmids carry antisense-spike ORFs with elevated GxxxG density relative to a Poisson null. This is a real artefact of the GC-rich codon optimisation and is worth flagging. It does not support a Pfizer-specific prion-risk claim.
Citation. Calibrated by GxxxG motif scan with PLAAC cross-check against the public GenBank records OR134577.1 and OR134578.1.
The original claim. The BNT162b2 spike CDS contains 76.92 slippery frameshift motifs per kilobase, suggesting extreme frameshift potential under m1Psi (N1-methylpseudouridine) substitution.
The calibrated finding. The frameshift motif scanner was truncating the CDS to 26 nucleotides before counting motifs, because of a bug in the coding-sequence extraction function (it was stopping at the first stop codon in any frame, not the first in-frame stop codon). After the fix, the Pfizer spike CDS is correctly read as 3,807 nucleotides, and the motif count is 3 motifs at 0.79 per kb.
The corrected density is still modestly elevated relative to the Moderna spike (which has zero such motifs, having been cleared by a more aggressive codon optimisation), but it is two orders of magnitude lower than the original claim.
What survives. The Mulroney 2024 frameshift mechanism (m1Psi-mediated plus-one ribosomal frameshifting at about 8 percent in vitro) is real and peer-reviewed. The three remaining slippery motifs in the Pfizer spike are at positions 564, 2547, and 3450, and are plausible (but unverified) candidate slip sites.
Citation. Mulroney LJR et al. N1-methylpseudouridine replacement in SARS-CoV-2 mRNA vaccines causes plus-one ribosomal frameshifting. Nature Biotechnology. 2024;42(7):799-805. PMID 38057663 (corrected PMID, see claim 4). DOI: 10.1038/s41587-023-01894-z.
The original citation. Mulroney et al., Nature, 2023, PMID 37603869.
The corrected citation. Mulroney LJR, Fleming A, Schepel I, et al. N1-methylpseudouridine replacement in SARS-CoV-2 mRNA vaccines causes plus-one ribosomal frameshifting. Nature Biotechnology. 2024;42(7):799-805. PMID 38057663. DOI: 10.1038/s41587-023-01894-z.
The wrong PMID (37603869) points to a different paper. The wrong journal (Nature) and year (2023) compounded the error. The correct journal is Nature Biotechnology, the correct year is 2024, the correct PMID is 38057663.
Why this matters. In any exchange with a sophisticated opponent, a wrong PMID is the cheapest possible win for them. Cite the correct PMID or do not cite the paper at all.
The original claim. A figure of 35 percent seroprevalence of anti-frameshift antibodies in vaccinated individuals has circulated as evidence that Mulroney's in vitro frameshift products are translated and immunogenic in humans.
The calibrated finding. No primary peer-reviewed citation for this figure has been located. It may exist as a preprint, conference abstract, or sub-stack summary, but until a primary citation is identified, the figure should not be repeated in any external-facing output.
What survives. Mulroney's 8 percent in vitro frameshift rate and their confirmation in vaccinated mice (97 percent reduction when 5-methoxyuridine replaces m1Psi) are both peer-reviewed and citable. The human-seroprevalence step is the missing link. Do not assert it.
The original claim. The BNT162b2 and mRNA-1273 vaccine spike proteins contain a modified furin cleavage site (HRRAR instead of Wuhan's PRRAR), suggesting the manufacturers deliberately modified the FCS in a way that matches later variants of concern.
The calibrated finding. The apparent "modification" is an artefact of comparing the wrong baseline. OR134577.1 and OR134578.1 are the bivalent vaccine plasmids (deposited June 2023), and their spike inserts are the Omicron BA.4/5 component, not the original Wuhan-spike monovalent.
The BA.4/5 lineage carries P681H as a natural lineage mutation (inherited from BA.2). When the vaccine spike is compared to a natural BA.4/5 reference rather than Wuhan-Hu-1, the FCS is wild-type for the lineage. There is no manufacturer modification of the FCS.
The full BA.4/5 signature is present in both vaccine spikes: D614G, N501Y, G339D, L452R, F486V, N679K, P681H, plus the K986P/V987P 2P stabilisation. Total mismatches versus Wuhan-Hu-1 are 30 (Pfizer) and 25 (Moderna) out of approximately 1274 amino acids.
What survives. Nothing about the FCS-modification claim survives. The broader observation that both manufacturers independently converged on rare CGG arginine codons at 17 of 41 spike arginine positions is unaffected (see claim 7).
The original claim (implicit in most prior analyses). The GenBank deposit OR134577.1 represents the plasmid used to manufacture the original Wuhan-spike BNT162b2 vaccine.
The calibrated finding. OR134577.1 is the bivalent Pfizer expression vector, deposited June 2023, and its spike insert is the Omicron BA.4/5 component. Moderna's OR134578.1 is similarly the bivalent mRNA-1273 vector with a BA.4/5 spike insert. (The GenBank annotation for OR134578.1 labels it "Wuhan-1 spike protein," which is incorrect, the sequence is BA.4/5.)
This means every analysis that compared CGG counts in OR134577.1 against natural sarbecoviruses (Wuhan-Hu-1, RaTG13, BANAL-52) was comparing a bivalent BA.4/5 vaccine spike to the wrong baseline.
What survives. The CGG enrichment signal survives the correction, because natural BA.1 and BA.2 Omicron spikes also carry only 2 CGG codons (1.6 per 1000 amino acids), same as Wuhan. Recomputed against natural Omicron:
| Sequence | Source | CGG count | per 1000 aa | vs natural BA.x |
|---|---|---|---|---|
| Wuhan-Hu-1 | natural | 2 | 1.6 | 1.0x |
| BA.1 Omicron | natural | 2 | 1.6 | 1.0x |
| BA.2 Omicron | natural | 2 | 1.6 | 1.0x |
| Pfizer bivalent BA.4/5 | vaccine | 19 | 15.0 | 9.4x |
| Moderna bivalent BA.4/5 | vaccine | 40 | 31.4 | 19.6x |
The codon-optimisation signal is real and reproduces against the correct baseline. The original attribution to Xia 2021 stands (Xia counted CGG in the original monovalent Wuhan-spike vaccines), and the Mulroney 2024 mechanism is unaffected. The only thing that changes is the baseline for the natural comparison.
A coincidence worth noting. Pfizer's bivalent BA.4/5 spike has exactly 19 CGG, matching Xia's count for the original Wuhan-spike BNT162b2. This suggests Pfizer uses a consistent codon-optimisation algorithm that produces the same CGG count regardless of input variant.
Nothing.
The seven claims calibrated above were not load-bearing for the overall case that SARS-CoV-2 vaccine plasmid forensics shows signatures inconsistent with the published regulatory characterisation. The load-bearing findings are:
The seven claims calibrated in this article were amplifiers, not foundations. Cutting them strengthens the case by making the remaining claims harder to dismiss.
There is a pattern in contested scientific debates. Each side accumulates a canon of claims, some strong and some weak. The weak claims travel faster than the strong ones because they are more dramatic. Eventually the weak claims get punctured, and the entire position loses credibility, including the strong claims that did not need the weak ones to survive.
The only defence is to puncture your own weak claims before the other side does it for you. This is expensive in the short term because it gives opponents easy material. It is necessary in the long term because it preserves the strong claims from collateral damage.
The seven calibrations above cost this author real rhetorical advantage. They were worth making because the alternative is to lose the whole case the first time a sophisticated opponent runs the same scans and gets the same corrected numbers.
The remaining case, with the calibrated numbers, is strong enough to stand on.
Every calibration in this article is reproducible from the public record:
If any reader produces a different result from these public inputs, the author wants to know.
End of calibrated counter-claims article. Comments and corrections from readers with primary-sequence evidence are welcome.
]]>