Reading the RNA in the Pfizer & Moderna vaccines

Have you ever wondered if you could make sense of the RNA code in the Pfizer or Moderna vaccines, if you could just read it? With some help, you can now do exactly that.

Availability of the RNA sequences and what that doesn’t mean

As we reported earlier this month, the RNA sequences for both the Pfizer and Moderna vaccines have been posted on GitHub.

People have said all sorts of damn fool things about RNA vaccines in general, and about this new repository in particular. It is not a recipe for making your own version of the vaccine, since there’s very elaborate chemistry in the lipid nanocapsule, a complex set of buffers and preservatives to make it go into solution in a useful way, and about a squillion other things. It’s a bit like knowing the instruction set of a computer: if you think that makes you able to build a computer, you’re a couple conses shy of a cold load.

While an argument can be made that this is proprietary intellectual property, the counterargument runs that it is (a) an insignificant slice of the total IP necessary to do anything, and (b) all the tricks in it are well-known (by people who know such things).

The goal of this post is for us to become one of the people who know such things.

An explainer piece on the vaccine RNA

Bert Hubert wrote a masterful tutorial ^[1] last Christmas, explaining the sequence of the Pfizer RNA. Interestingly, he explained (some of) the hacks done to it to make it a better vaccine, not just a copy of the virus RNA. It makes a difference: the raw sequence of the SARS-CoV-2 virus would be terrible as a vaccine.

His explainer is so good, there are multiple video explainers-of-the-explainer available; here are two of them to which Bert himself referred.

Explaining the explainer

You really should read Bert’s piece (about a 20min read). But… since we both know you probably won’t do that, here’s a quick summary. (But really, his explanation is probably better than mine.)

The Pfizer sequence was apparently published by the WHO last September. It looks more or less like this, showing here the first 500 nucleotides of the RNA. There are a total of 4284 nucleotides in the whole thing.

All clear, right? No, of course not, unless you are a ribosome used to reading this sort of thing to make proteins. (If you are an intelligent ribosome who reads my blog, please get in contact. I’d very much like to interview you: you’re either an incredible miracle or someone with an interesting delusion. Possibly both.)

So let’s break down how that sequence works. It turns out it has some degree of modularity, i.e., we can point to bits of it and explain what they, and only they, do.

• Both graphics above came from the WHO report from the International Nonproprietary Names Programme in September 2020 (though the original now seems to have been moved somewhere else).
• RNA consists of a chain of 4 nucleoties/bases: adenosine (A), cytosine (C), guanine (G), and uracil (U). So the sequence above (almost) tells you how to put those 4 nucleotides together to make something. That “something” gets fed through a ribosome, which through outrageous Rube Goldberg machinery, assembles a corresponding protein out of amino acids. The proteins go off and do all sorts of things, like mostly everything.
• In order for a ribosome to recognize RNA a meaningful, there has to be a recognition handshake: this is the GA cap at the front. To a ribosome, this means: “you can process this”. So that’s the “cap” thing on the left end, where a ribosome starts looking. I dislike comparisons to computer code, since those are inevitably misleading, but this really does look like a shebang.
• Now, I just got done telling you an RNA sequence was a string of letters from the alphabet of ACGU. But… what’s that thing that looks like a capital Greek Ψ doing in there? Well!
  • Your body realio-trulio despises the idea of foreign DNA or RNA running around loose. Might be a virus, or something. So it will utterly destroy the natural RNA as it came from SARS-CoV-2, faster than it could be of any use in vaccination.
  • To prevent that, we need biohack #1: change every uracil (U) to 1-methyl-3’-pseudouridylyl (Ψ). It doesn’t matter exactly what that Ψ is, so long as you understand its kinda like U in terms of what a ribosome will do, but different enough not to trigger alarms in the immune system that get it destroyed.
  • I believe this trick is due to Weissman & Karikó at Penn. (Though apparently Penn treated Karikó somewhat badly.)
• Next comes a gizmo called the 5’ UTR (“five-prime UTR”), or “untranslated region”.
  
  • What’s with the “five-prime” business? Molecules with carbon rings get the carbons numbered, so we can talk to each other about which atom we mean. As shown in the picture, nucleotides have such a ring, so the carbons are numbered 1-5. The way things shake out, it’s carbons 3 and 5 that make the bonds to the next nucleic acid. (The “prime” business is another story, irrelevant here.) So the 5’ end just tells you you’re starting at the start end of the string of nucleic acids, and the 3’ end is at the back end of the string.
  • Ok, fine: “5-prime” means “in the beginning”. So what do the next 51 bases do, and if they’re untranslated (not made into protein), why bother? It’s the thing that initially clips onto the ribosome, so it can start reading. Basically, it’s just enough space for a ribosome to fit and then start reading what’s after the 5’ UTR.
  • Also, it instructs the ribosome a bit about when to make the protein and how much of it. Pfizer has supplied here a 5’ UTR that says make a boatload of this protein, and do it right this very minute, cribbed from the alpha globin gene ^[2], but tweaked even harder to force more protein. This is biohack #2.
• Next up is a short bit called the S glycoprotein signal peptide.
  • This says what to do with the protein once the ribosome has made it. The version in the vaccine says to export it via the endoplasmic reticulum, which ends up displaying the protein on the cell surface for the immune system to see, and freak out about. Here’s part of Bert’s comparison, calling out the changes with “!”:

          3   3   3   3   3   3   3   3   3   3   3   3   3   3   3   3
    Virus:   AUG UUU GUU UUU CUU GUU UUA UUG CCA CUA GUC UCU AGU CAG UGU GUU
    Vaccine: AUG UUC GUG UUC CUG GUG CUG CUG CCU CUG GUG UCC AGC CAG UGU GUG
               !   !   !   !   ! ! ! !     !   !   !   !   !           !
    

  • But wait, there’s more! There are a number of changes here between the vaccine version and the viral version. Each time it changes a U/Ψ to a C, in ways that do not affect the protein being made (e.g., UUU → UUC is a synonymous substitution). This is because RNA with higher GC content gets converted more efficiently to protein. Biohack #4.
  • There’s one other change (A → U) that prevents the formation of a hairpin RNA, that would bind to itself too tightly to be of use. Biohack #5.
• Next are 3777 bases that code for the spike protein amino acids.
  • They’ve also got a lot of synthesis-ehancing GC increases. Biohack #6.
  • But then there’s biohack #7: in exactly 2 places, the amino acid in the resulting protein is actually changed to proline instead. Why?
    • In the actual virus, the shape of the spike protein is stabilized by being attached to the rest of the virus particle.
    • But in the vaccine, we have just the spike, and in its original form it’s not stable. It would fold up the wrong way, and your body would produce antibodies that don’t correspond to what’s on the actual virus. This was a discovery of the Mclellan group at UT Austin, in their 2017 work on the MERS virus ^[3]: if you substitute proline, it gets more rigid and stays in the right shape. (Once again, we’re very lucky: RNA vaccines came along at just the right time, and our previous knowledge of MERS and SARS prepared us for being able to make the COVID-19 vaccine.)
  • The protein ends with 2 stop codons, i.e., instructions to the ribosome that this is the end of the protein. The actual virus has only 1 stop codon, so this is biohack #8.
• Then there’s the 3’ UTR. These are somehat mysterious beasts today, but they have something to do with protein stability, rate of translation, and a few other things. Pfizer scientists here picked a particularly efficient one of the the human genome (the 12S riobosomal protein). So that’s biohack #9.
• Then there’s the poly-A tail: polyadenylation means a lot of AAAAAAA’s at the end.
  • This is so the RNA can be re-used many times: it loses a few A’s at the end each time through the ribosome so to make it get used a lot they provided a lot of A’s. Biohack #10.
  • Finally, there’s a funny linker (GCAUAUGACU) that breaks up the poly-A tail into segments. This is well beyond anything I know much about, but apparently there’s a paper ^[4] saying that segmented tails make everything better in some complex ways. So that’s biohack #11.

What’s it all mean?

This bit of RNA has been considerably engineered by Pfizer to optimize for getting past the immune system, optimizing for protein translation speed & amount in several ways, improving the stability of the spike protein away from the rest of the virus, and so on.

Bert has guided us through the weeds to see at least 11 biohacks to do this. Really, go read his piece, since he probably explains things better than me.

What struck me was how much of all this engineering has come to light in just the last couple of years. Enormous preparation on RNA vaccines and knowledge of the previous coronavirus outbreaks (MERS and SARS) meant that we had just exactly the right tools, at the right time, in the right hands.

Be thankful.

---

Notes & References

1: B Hubert, “Reverse Engineering the source code of the BioNTech/Pfizer SARS-CoV-2 Vaccine”, BertHub.eu blog, 2020-Dec-25. ↩

2: K Asrani, et al., “Optimization of mRNA untranslated regions for improved expression of therapeutic mRNA”, RNA Biology 15:6 (2018), 756-762. ↩

3: J Pallesen, et al., “Immunogenicity and structures of a rationally designed prefusion MERS-CoV spike antigen”, Proc Natl Acad Sci USA 114:35 (2017-Aug-29), E7348-E7357. PMC5584442. ↩

4: Z Trepotec, et al., “Segmented poly(A) tails significantly reduce recombination of plasmid DNA without affecting mRNA translation efficiency or half-life”, RNA Journal, 2019-Jan-15. DOI: 10.1261/rna.069286.118. ↩