mRNA is still made the 1984 way. It was not built for the drugs coming next.

Pharma has committed $6 billion to the idea that RNA can turn cell therapy into an injection. The harder problem starts when the same molecule has to be given in grams.

Since June 2025, AbbVie has paid $2.1 billion for Capstan, Bristol Myers Squibb $1.5 billion for Orbital and Lilly has agreed to pay up to $2.4 billion for Orna. Each was buying a version of the same bet: that RNA can do inside the patient what CAR-T therapy currently does in a factory. That means removing someone’s T cells, engineering them in a facility and infusing them back, at a list price of half a million dollars or so.

RNA is an instruction. It can tell a cell to make a protein that no infused drug can reach, including proteins that only work inside the cell. Expression fades within days and the molecule never enters the nucleus, so there is nothing to integrate. Gene therapies built on viral vectors are one-shot treatments, because the patient makes antibodies against the capsid that block a second dose. A chronic disease needs a drug you can give again.

We like that promise at ZAKA enough to have backed it four times, across sequence design, delivery and manufacturing.

A full COVID vaccine course was 0.2 milligrams. Merck and Moderna’s personalised melanoma vaccine, which met its Phase 3 endpoints in August 2026, is given as up to nine doses of one milligram. A year of Moderna’s treatment for propionic acidemia, given every two weeks for life, runs from 550 milligrams to more than a gram and a half. The manufacturing was built for milligrams.

The reaction was never clean

Since Douglas Melton and colleagues published the method in 1984, almost all mRNA has been made by in vitro transcription: a DNA template, an enzyme, a tub of nucleotides, no cells involved. It is quick to set up and quick to run, which is how Moderna shipped a clinical batch to the NIH 42 days after choosing its sequence in January 2020, for which the world was duly grateful.

It also produces a mixture. Alongside the intended molecule come double-stranded RNA and other species that the innate immune system reads as infection and that translate poorly. Katalin Karikó and Drew Weissman showed in 2005 that modifying the building blocks quietens the immune response, and in 2011 that stripping the double-stranded material out quietens it further and raises protein output by between ten and a thousand times. Both fixes went into the COVID vaccines, and the Nobel followed in 2023.

Those two fixes, changed building blocks and a cleaner product, were enough for a vaccine. A vaccine sets out to rouse the immune system, so it can live with a product that alarms it a little. A drug that replaces a missing protein cannot. And a chronic drug asks more of both fixes, because how much residue a product can carry is set by the dose and how often it is given.

A chronic drug needs grams

An adult on Moderna’s propionic acidemia drug would need about a gram of mRNA a year. Each infusion carries as much mRNA as 200 to 600 of Moderna’s original COVID shots, and there is one every two weeks. Dwight Koeberl and colleagues published the dosing in Nature in 2024. Their patients were children, so the adult figures are my own sum.

A residue you can tolerate once, at 100 micrograms, becomes a tolerability ceiling when the same product goes in every two weeks for years. That ceiling caps the dose. A lower dose means less protein at the target and a smaller effect for the trial to detect. If the trial then fails, it is the drug that gets the blame.

Then there is the bill. TriLink, the Maravai company that supplies much of the industry with capping reagents, published a costing in 2023 that put the manufacturing cost of a single gram of clinical-grade mRNA at $305,000 by its own capping method and $442,000 by the enzymatic route. Discount it as you would any number from the company selling the cheaper option, and note that it covers drug substance alone, with no formulation, fill and finish or release testing.

A bigger factory does not bring the number down the way it usually would, because modelling by Zoltán Kis and colleagues at Imperial College finds that RNA manufacturing has high variable costs. The cost sits in the consumables rather than in the building. So a chronic mRNA medicine does not get affordable by increasing the scale. It needs a cheaper way to make the material.

The patches chase purity and add cost

The industry’s answer has been to keep the 1984 reaction and improve it from inside. Two decades after Karikó and Weissman, it is still working on both of their fixes, and neither makes the gram cheaper.

In March 2026, researchers at Verve Therapeutics, now a Lilly subsidiary, published redesigned versions of the transcription enzyme that pushed the double-stranded byproduct below the limit of their assay. That means their assay found none, which is a weaker claim than there being none. TriLink sells an alternative polymerase off the shelf that it says cuts the same impurity by up to 85 percent. What the enzyme leaves behind is removed by chromatography, which works well and scales poorly.

The changed building blocks are the older fix, and both Moderna and BioNTech built their vaccines on it. Its catch surfaced only after billions of doses had been given. In December 2023 Thomas Mulroney and colleagues at Cambridge reported in Nature that cells sometimes misread the modified mRNA and make proteins nobody designed. About a third of the 21 vaccinated people they tested had an immune response to those stray proteins. The authors found no harm, and their remedy is one more patch, this time to the sequence. But those people had had a vaccine course. A patient dosed every two weeks gets 26 infusions a year. Before backing a drug meant to be taken for life, I would ask whether anyone has checked for those proteins at that dose.

Each of these patches keeps the reaction and adds something to it, and every extra purification step costs you material. By TriLink’s own figures, the enzymatic route that needs a second reactor loses 30 to 50 percent of what it makes along the way, against 5 to 20 percent for the one-step version.

The price of a gram traps the incumbents too. Their economics were set at 100 micrograms a dose, where the cost of the material barely registers. At a gram a year that cost becomes part of whether the drug can be sold at all. But their plants and their regulatory filings are built around the reaction, so patching it is the natural move for them, even when what they buy is a cleaner product rather than a cheaper one.

Cells already do this

The alternative is to hand the job back to biology. Cells make RNA all day, without being asked. Engineer one to make the sequence you want, grow it in a tank, take the molecule out at the end. The shopping list changes. Instead of a plasmid prepared for every run, a transcription enzyme, a capping reagent and nucleotides under licence, you buy a cell line and something to feed it. And the double-stranded junk the reaction throws off is never made, so there is nothing to strip out.

It creates a different set of problems. The producer cell has to be engineered and stay genetically stable, the clone has to yield the same material batch after batch, and the mRNA has to be pulled out of a living system that is full of its own RNA. Clearing host-cell material out of a tank is familiar work, and the industry has done it since recombinant insulin went commercial in 1982. Recovering an intact RNA molecule from that tank is newer.

The cell also builds its RNA from natural building blocks, so the modification is gone. CureVac tried that with reaction-made mRNA. Its COVID vaccine was dosed at 12 micrograms, against 30 for BioNTech and 100 for Moderna, and protected 48 percent of recipients. CureVac blamed the variants then in circulation. It also dropped the product and built its next vaccines with GSK on modified mRNA. The bet behind cell-made mRNA is that what the reaction leaves in the product did more of that damage than the missing modification did.

A team at the University of Sheffield published a platform in 2024 that engineers E. coli to make mRNA inside the cell, lifting yields more than forty times above an unengineered system, with the product reaching about 30 percent of the cell’s total RNA. It is laboratory work, ten to fifteen milligrams a litre in shake flasks. The fifth author, Martyn Hulley, works in AstraZeneca’s bioprocess development group, and AstraZeneca part-funded it.

Sensible Biotechnologies, our portfolio company, is doing this commercially. It has just announced $47 million in financing led by Oxford Science Enterprises, and it has two bench results to show for the bet. Sartorius, which equips much of this industry and has worked with Sensible since July 2025 on reaching clinical-grade production, ran the standard antibody test for double-stranded RNA on Sensible’s material and detected none, the same bar Verve’s enzyme cleared. And in a lab test on immune cells, Sensible’s mRNA set off less of two standard inflammation signals than reaction-made mRNA did, while making as much protein or more.

Nobody has yet run cell-based mRNA at commercial scale, and two questions are open. One is yield: enough material per tank, batch after batch, to undercut a reaction with forty years of optimisation behind it. The other is whether a result from a dish of immune cells holds in a patient dosed every two weeks for years.

Manufacturing is where drugs get stuck

This is not an mRNA problem alone. In July 2025 the FDA published a batch of the rejection letters it sent between 2020 and 2024. Pharma Manufacturing counted 202 of them, and 150 cited quality or manufacturing problems, such as a failed facility inspection or a gap in the chemistry and controls file. Those letters cover drugs and biologics of every kind, and every application in that archive was eventually approved. So the factory did not kill these drugs. It cost them time. A Johns Hopkins team put the median gap between receiving a letter and getting approved at 1.28 years. That median counts only the drugs that made it, so 1.28 years is a floor.

I argued in an earlier piece that the disease a company picks sets what its proof will cost. Manufacturing is the same argument one step further along, and here the bill is for the material.

Two things would change my mind. The first is a cheaper gram. If someone cuts the cost of a clinical-grade gram tenfold while keeping the one-step reaction, the cost case for rebuilding the process around cells mostly goes away. At $30,000 a gram, a therapy that needs a gram a year has a manufacturing line item rather than a business-model problem.

The second is an approval. In January 2026 Moderna partnered with Recordati to take its propionic acidemia drug through the last stage of trials and sell it. If that drug is approved on reaction-made mRNA, dosed every two weeks, the quality problem was smaller than I have argued.

Three companies have now committed billions to the idea that RNA can turn cell therapy into an injection. None of the three announcements mentioned the price of a gram. If you are looking at one of these programmes, the check takes ten seconds. Multiply the milligrams per dose by the number of doses a year. If the answer is in grams, ask what the material costs before you ask anything else.

The companies working on that number have raised a fraction of what the three acquisitions committed, and they are the ones the bet now rests on.

If you are building one of them, we would like to talk.

Jan Buza, Partner at ZAKA VC

ZAKA is an early-stage fund investing in the teams building the next generation of healthcare and life sciences companies.