Epigenetic mistakes in Plasmid manufacturing

cGAS-STING and cancer.

In the last few weeks, Medicinal Genomics has been sharpening it PathoSEEK pipeline for sequencing E.coli, Salmonella and Listeria genomes in under 48 hours.

To get this done we have been using the latest Oxford Nanopore chips to sequence these pathogens from colony to genomes assembled into a single circular contig in 48 hours. The genomes are assembled and antibiotic resistance genes and serotypes called in this time frame. This has been an absolutely remarkable change in turn around time for pathogen identification and supplement safety screening.

When these flow cells are done, you can wash them with DNaseI and recycle them for interesting side projects. One such side project was to sequence Pfizer lot GJ6665 to profile its methylation patterns to replicate a recent Preprint we published.

One topic this paper highlighted was that the Linearization step in the plasmid manufacturing failed. We have now replicated this observation in another Pfizer lot where 5/36 reads in this region are not cut.

But lets scan through the whole plasmid and look at the Dcm and Dam methylation sites flagged as Blue (GATC for Dam) and Red (CCWGG for Dcm).

A quick summary on how to read IGV plots is depicted below. Each horizontal grey line is a single molecule nanopore read. Verticle ticks in these reads are methylation signals. The bottom tracks are annotation tracks to highlight where various features exist in the DNA sequence (Spike, SV40 etc).

So lets tour the whole plasmid for methylation signals and then review which ones are real which may be single molecule noise or artifacts. Keep in mind, most sequencers until now are reading 1,000 to a 1M molecules at once and averaging the signal to get a consensus. This greatly reduces noise. When you are working with at the single molecule level there is a different signal to noise and you see more artifacts. But these artifacts are usually stochastic and are resolve with redundant sequencing reads.

Before we get started we need to take a moment to reflect on how remarkable it is that we can read a single molecule of anything and get it right.

Here is an example of a 17,128 base read (not a C19 vax but a monkeypox vax) that is 99% accurate. This was read at 400 bases per second and the USB sized sequencers are capable of doing this with 500 pores in parallel. Absolutely remarkable technical achievement.

So lets tour through the Pfizer plasmid lot GJ6665 and review the methylation signals.

You may notice some noisy regions that show low levels of potential methylation but do not occur at expected GATC and CCWGG sequences. I think we have an explanation for this.

Lets look at the Quadruplex Gs and how they impact nanopore signals that can be mistaken for enzymatic methylation signals. To resolve this you need to understand that most methylases target palindromic sequences and methylate both strands.

GATC is also GATC on the opposite strand so enzymatic methylation should exist on both strands but offset by a few bases (1bp for GATC and 2bp for CCWGG).

So when you see hemimethylation signals on ONT (one strand methylated and the other strand not) and the signal isn’t in a known methylation sequence motif (GATC and CCWGG), its very likely a nanopore sequencing artifact.

Here are a few examples of weak Nanopore off target methylation signals. You will notice they are mostly near Quadruplex G (G4) motifs and are only on one strand. These are annotated in gold on the bottom track from QGRS mapper. G4s form DNA knots which likely slow the motor protein translocating the DNA through the nanopores.

ChatGPT5.o agrees.

These noisy regions are unlikely to be off target enzymatic methylation but instead a slowing of DNA going through the pores which creates an odd current profile. You have to understand that this DNA , when not folded into a G4 knot is racing through the pores at 400 bases per second so secondary structure can slow this process and create signal artifacts.

Here is the translocation speed through the pores for the salmonella genomes we sequenced.

9.4 Gb of sequence in 20hours.

Once this array was finished, we decided to recycle it with a DNase treatment and ran the Vaccine sample for GJ6665 on the remainder of the life span of the spent flow cell.

Now when you get to the SV40 Promoter’s GC box (that part of sequence that binds to P53), we get weak hemimethylated signals that do not correlate with G4s. This is likely due to other secondary structures in G rich sequences of this region. Not all the Blue and Orange in the sequence track.

You can see these sequences are no CCWGG Dcm sites but also these motifs show no signs of Methylation on both strands.

This is very different than true palindromic methylation signals where you get strong methylation signals on both strands.

Why are we obsessing over the methylation signals in the contaminating DNA in the mRNA vaccines? The DNA in Process 1 was processed with PCR which erases all methylation signals. Process 2 used plasmid DNA which gets methylated by E.coli IF YOU USE THE WRONG E.coli STRAIN. And you can bet Pfizer used the wrong E.coli strain.

The FDA keeps claiming this DNA is non-functional.

Carnes et al demonstrate this is a false statement in 2010. It’s imperative you lock down the methylation patterns on your plasmid for any therapeutic.

The reason for this is that dcm and dam methylation is a signal to human cells that viral or bacterial DNA has invaded the cell. It triggers the cGAS-STING pathway which is a STimulation of INterferon Genes.

You’ll notice the above highlighted statement directly contradicts the FDAs statements on this DNA being non-functional. The 5 methyl cytosine does not exist in Process 1 but now is in fact in Process 2 products given to the public.

“It is critical to LOCK the plasmid methylation pattern prior to product clinical development”…. “dcm- plasmid for DNA therapeutics and cell transfection reagents such as AAV helper plasmids.”

The above paper explains why. cGAS-STING is more potently activated by methylated DNA compared to PCR DNAs.

There is also an induction of CD69 cells.

Kwon et al discusses this important cancer pathway.

Conclusions

These data confirm that multiple Pfizer lots (FL8095 and GJ6665) contain contaminating plasmid DNA that is methylated compared to the residual DNA that may have existed in the PCR products from Process 1. This DNA is NOT FUNCTIONALLY INERT as claimed by the FDA. Activation of cGAS-STING may paradoxically induce an immune response against cancer but chronic activation may in fact lead to oncogenic processes. This is important to understand as we see Pharma parade out papers showing a single dose may reduce cancer with checkpoint inhibitors. This is a statistical game and you will not see this signal if you look at patients given 2+ doses.

This data also demonstrates failure to linearize the plasmids in a second Pfizer lot suggesting this process failure isn’t a spurious event present in a single lot but may in fact exist in all products.

Manufacturing plasmid references.

Wang et al

Luecke et al

Balzarolo et al

Carnes et al

A few Photos of hero’s from the CHD conference.