I’ve been meaning to blog about this recent advance in mRNA vaccines, a new multivalent candidate with the widest range ever tried against influenza. It’s worth a look to see how this was done, and what it might mean (or might not) as it goes forward.

Flu is a tough vaccination problem, as is well known. Everyone’s heard of the years when the seasonal vaccine turns out to be a poor match for the strains that actually end up causing trouble, and that’s an inevitable consequence of two factors: the lag time inherent in choosing a particular antigen combination along with production on scale and distribution of the actual vaccine, and the constantly changing nature of the influenza viruses themselves.

The major antigen on the surface of those viruses is hemagglutinin, and it is the point of the spear as far as the infection mechanism is concerned. It recognizes and binds to sialic acid (for the most part) on the surfaces of human cells through its HA1 domain to start things off. Cells deal with such attachments by budding inwards to form endosomes, and these compartments then start to acidify in order to break down their cargo. The influenza virus has evolved a way to deal with this, though: a fusion peptide is released, and its hemagglutinin HA2 domain changes shape as the pH drops. The fusion peptide fits into the endosomal membrane, and this also fits in with the HA2 protein to bring the viral envelope right into contact with the endosomal membrane. The two then fuse, dumping the viral payload (its RNA, crucially) directly into the cytoplasm, and the cell is now infected, damn it all.

There’s another flu-virus surface protein of great importance as well, neuraminidase. That’s involved in that sialic acid binding the the HA1 domain is doing. The viral particle actually wanders around the surface of the cell, binding one one sialic acid residue after another, until it can find the actual cell-surface receptor protein that it’s looking for to start the entry process and find itself in an endosomal compartment. Not all of those sialic acid binding events put it within range of such a receptor protein, and neuraminidase is the enzyme that cuts it loose to bind again and take another crack at cell entry. Interestingly, the identity of the actual receptor protein it’s looking for was a mystery for decades, but recently there’s evidence for it being EGFR, the epidermal growth factor receptor. Neuraminidase, by the way, is the target of the small-molecule drug Tamiflu (oseltamivir) – as an enzyme, it’s susceptible to being blocked by such an active-site inhibitor. Unfortunately, while the drug isn’t useless, it’s not as much a weapon against flu as you’d like either.

Those two proteins are the basis for the naming system that’s often seen for influenza A viruses. There are at least 18 types of hemagglutinin and at least 11 types of neuraminidase, and these can mix around pretty much any way that they’d like. That’s where you get nomenclature like H1N1, H3N2, and so on, and there are a nasty number of possibilities there. Each of those can of course branch off into clades, subclades of those, and so on, depending on amino acid mutations elsewhere in the viral genome, but the higher-level classifications still hold.

The highest level are the letter designations – influenza A is the only type that’s known to cause pandemic-type flu, and as mentioned, those are the ones with the HxNx designations. Influenza B is also known to infect humans, with only two subtypes (Yamagata and Victoria – recently it’s been mostly Victoria, but influenza B is generally second fiddle to A in most locations). Influenza C is really not much of a player in human health (God knows we don’t need it to be), and it generally causes only mild infections with no epidemic spread. And Influenza D as far as we know  only infects cattle, and I hope it stays there. So the focus is on A and B for vaccines.

With that in mind, hemagglutinin has naturally been the glycoprotein used in flu vaccines – sticking antibodies to it is just what you’d want to do to mess up the viral infection process. We’re all familiar with the coronavirus Spike protein in that context, and the RBD (receptor-binding domain) out at its far end. Hemagglutinin in influenza has its own complex structure, too, starting with a “head” region and a “stalk” or “stem” region. In general, antibodies to the head region block attachment of the viral particle, while antibodies to the stalk prevent the later membrane fusion event. It looks like the human immune system tends to focus more on “head” antibodies that only recognize a few closely related strains, but it’s been shown that “stalk” antibodies have broader specificity (since the stalk protein itself is more conserved across influenza subtypes) and a lot of work has gone into those in the quest for a more universal flu vaccine. That link has some evidence that higher doses of vaccine would be necessary to get a good response to those conserved domains.

This latest work includes 20 different hemagglutinin-encoding mRNAs, one for each of the 18 influenza A types and two for the two influenza B ones. So they’re going for a smaller number of conserved domains, but rather casting the net widely across all the known types. The team vaccinated mice with this cocktail (using the now-famous lipid nanoparticle formulations) and indeed saw an antibody response to all 20, whether or not the animal had previously been exposed to influenza. The response seemed to be in the form of distinct antibodies, rather than several cross-reactive ones stretched across different subtypes.

28 days later, the mice were challenged with either an H1N1 that was right on target with the vaccinated antigen (a California strain), or one that was antigenically distinct (a Puerto Rico one). Influenza is pretty bad news for mice; either strain was enough to kill control animals after about a week. The on-target California-infected mice showed little sign of disease, had slight weight loss, and recovered completely. The antigenically-distinct Puerto RIco-infected ones had a rougher time of it, but 80% of them survived as opposed to the complete death seen in the unvaccinated controls. Another control experiment used a 19-valent vaccination, this time with H1 taken out. Those animals showed more severe disease and several deaths when exposed to the California strain, but had a less severe time with the Puerto Rico one, this time with no deaths, and the reason for this difference isn’t entirely clear. It looks like the matched infections are cleared by neutralizing antibodies, as you’d expect, whereas the mismatched ones are eventually cleared by more brute-force nonneutralizing mechanisms like antibody-dependent cell toxicity.

A further experiment used a two-dose prime/boost treatment (28 days apart) in ferrets, who were then challenged with an avian influenza H1N1 strain whose H1 was distinct from the H1 included in the vaccine mix – this was to mimic a real “crossover from another species” flu pandemic event, where birds (and pigs) are indeed a constant reservoir for new varieties. Influenza is hard on these animals, too: the unvaccinated controls lost a significant fraction of their body weight, and two of the four animals died. But the vaccinated ones had only about half the body weight loss, with complete survival.

So based on these animal results, it looks like a 20-valent flu vaccine like this might provide very strong protection in the case of a reasonably close antigenic match, and it would have the substantial advantage of having a wide range of hemagglutinin types ready for this. In the case of a mismatch, you would get less severe symptoms and faster clearance, but still be vulnerable to flu infection. The hope is that this would all be solid “anti-flu-pandemic” insurance from a public health standpoint, and we’ll have to see if that’s how it develops. There’s going to need to be more work put into this before it’s ready for human trials, though – which antigens would be best for each of the 20 components? How different are thosee different choices in animal models, and are there any synergistic or interference phenomena that could show up? Could you go higher than 20 for greater coverage of (say) H1 subtypes? And can this shotgun style be extended to other infectious diseases, now that the mRNA platform looks like it can allow it?