How do cells tell synthetic RNA from their own?

RNA medicines are read by the cell's own surveillance systems. Receptors such as RIG-I, MDA5 and PKR detect RNA that looks foreign, and their response limits how much protein a dose produces and how often a patient can be dosed. For a rare disease that has to be treated for years, that ceiling is the whole problem.

I want to understand what these receptors actually read on a synthetic transcript: its cap, its modified nucleosides, the double-stranded byproducts of manufacturing, and, for self-amplifying RNA, the intermediates it makes as it replicates. Knowing which features matter is how you design an RNA that cells tolerate.

Long term, I want to build RNA medicines for rare diseases, where there is often little or no research behind a condition.

The problem. Integrity, the share of transcripts that are full length, is a critical quality attribute of an RNA drug. A truncated RNA is either inactive or makes an incomplete protein, and the starting integrity largely sets shelf life: the lower it starts, the sooner the product falls below specification. This matters most for self-amplifying RNA, because long transcripts lose integrity faster than short ones. Our previous process delivered about 75% integrity.

Why it connects to the question above. Double-stranded RNA is both a product impurity and the thing the cell reads. The same byproduct that lowers integrity is a ligand for RIG-I, MDA5 and PKR, so a process change that removes it is also an immunology result. That is why a manufacturing problem led me to the question at the top of this page.

The hypothesis. Studying how RNA degrades during in vitro transcription (IVT) led me to a specific mechanism. Free magnesium released during the reaction accelerates RNA hydrolysis. If that was right, lowering free magnesium during the reaction should protect the transcript without slowing synthesis. We modeled how the magnesium species shift over the course of the reaction, and I tested the idea in two ways.

Using the reaction's own byproduct. IVT reactions normally include inorganic pyrophosphatase (iPPase), an enzyme that destroys the pyrophosphate released during synthesis. Pyrophosphate binds magnesium, so the standard view is that it should be removed to keep the reaction fast. My hypothesis predicted the opposite: leaving more pyrophosphate in place would turn it into a built-in chelator for excess free magnesium. Reducing iPPase raised integrity from 67% to 79%, and combining low iPPase with NTA reached 84%. The cost was yield at the highest NTA level, so we chose a condition that balanced both.

Purification. Experiments on oligo dT loading conditions showed an inverse relationship between the conductivity of the RNA loaded onto the column and the integrity of what came off it. We identified low-conductivity diluents that improve integrity by 4 to 8%.

The outcome. In a 35 mL scale-down model of manufacturing, three replicate runs with NTA came out at 89 to 92% integrity after purification, compared with about 75% for the previous process. At a degradation rate of 0.5% per day, that adds roughly a month of shelf life and brings self-amplifying RNA closer to the shelf life of conventional mRNA vaccines.

How I thought about it. I started from the chemistry of the reaction itself. Once free magnesium looked like the cause, the question became how to control it without adding anything a therapeutic manufacturing process cannot carry. Standard chelators such as EDTA work but are hard to justify in a drug process, so I looked at what the reaction already makes. Pyrophosphate is produced in every IVT reaction and binds magnesium, which made it the obvious candidate to keep instead of removing.

What it left open. This work told me which conditions change integrity and double-stranded RNA content, but not which features of the transcript the receptors are actually reading. Separating those, the cap, the modified nucleosides, the double-stranded byproducts and the replication intermediates, is what I want to do next.