We medicinal chemists are used to thinking about small molecule drugs – it’s what we do. And we’re also comfortable with having a category in our worldview that we assign to “biologics” – proteins, mostly, many of them antibodies, which can also be extremely therapeutically effective under the right conditions. But we really need to expand our thinking a bit, because there’s a whole world in between these two.
Consider the odd sorts of antibodies found in the camelids (camels, llamas, alpacas, and so on). They have only the “heavy chain” part, and stripping it down to just that domain (patented under the name “Nanobody”) gives you a much smaller protein piece that’s still capable of selective antigen recognition. You can get something similar from sharks as well. These smaller proteins can have rather different properties from traditional antibodies, and being smaller, you have more opportunity to modulate those properties in a defined way. Similarly, there have been many proposals over the years for “antibody-ish” protein platforms, and you can imagine that there are a lot of possibilities.
Here’s another example from the recent literature. The authors are, as they say, taking advantage of
. . .advances in both DNA manufacturing and protein design that have led to a fortunate convergence between the upper limit of the size of oligonucleotides (230 bp) that can be synthesized as pools of 10,000 or larger, and the lower limit of the size of genetically encodable computationally designed proteins (roughly 40 amino acids).
As a test case, they’re looking for a protein binder to influenza A H1 haemagglutinin (HA) and botulinum toxin, which was discussed here just the other day. The work starts out with computational filtering of possible proteins (using Rosetta software), trying different combinations of alpha-helix and beta-strands, and for the actual binding surfaces they built on past work to find such species and on the geometry of natural binding partners. (This definitely gives you a leg up – doing this from scratch, without known structures, would be a steep climb, you’d have to think). They picked several thousand likely variants for each target, including some random changes as well as more defined ones, and expressed the lot (about 17,000 proteins) in yeast.
Fluorescent binding assays on these libraries, followed by cell sorting on said fluorescent readouts, gave dose-responsive readouts of hits, which is reassuring. With botulin, for example, at 100, 10, and 1 nanomolar, they pulled out 2685, 987, and 355 proteins (from a starting pool of 5306 that actually expressed, out of the 5311 designed). Exposing the last set to a protease treatment beforehand, to leave only the hardy ones, took that down to 57 sequences. For both this and the HA pool, the computationally designed sequences were enriched in the final winners, which is a good sign. Interestingly, in both sets, the only ones that remained after protease exposure were the ones that incorporated disulfides, which clearly had an impact on structural stability, as they should.
The authors claim that this may well be the largest scale attempt at confirming computational ability to design protein-protein interactions, and I certainly can’t refute them. Some crystal structures were solved with some of the best binders, and they fit the calculations quite well.They proved highly active in cell protection assays, and here comes the small-protein property advantage: incubating these at 80C for an hour before running the assay did not change their activity at all, in marked contrast to what happens with conventional antibodies. Similarly, multiple rounds of dosing elicited little or no immune response in rodents. Moreover, intranasal administration of an HA candidate to mice, followed by a lethal challenge with influenza virus, led to 100% survival at doses down to 0.03 mg/kg, which is 100-fold lower on a mass basis than the known broadly neutralizing antibody. A 3 mg/kg dose was 100% effective 72 hours after exposure as well. Notably, iv administration was useless.
This is all pretty impressive, I’d say, and what I like even more is that we’re just in the beginning phases of this sort of work. We’re going to get better at producing these proteins, designing them, and selecting them, and there could be some major opportunities out there. Stay tuned!