The Vaccine Technology That Was Almost Abandoned Is Changing Medicine
Karikó and Weissman spent decades ignored while developing modified mRNA. COVID proved them right. Now personalized cancer vaccines and new RSV shots are showing what this platform can really do.
Contents 6 sections
In 2013, Katalin Karikó was demoted at the University of Pennsylvania. Her lab funding was cut. Administrators suggested she find work elsewhere. The reason: she had spent the better part of two decades trying to make synthetic mRNA work as medicine, and most of the scientific establishment had concluded this was a waste of time.
Eleven years later, she shared the Nobel Prize in Physiology or Medicine.
The mRNA platform she developed with her colleague Drew Weissman didn’t just produce the COVID-19 vaccines that ended a pandemic. It turned out to be a general-purpose biological printer — one capable of instructing cells to make almost any protein the immune system might need to recognize. And now, with COVID in the rearview mirror, researchers are learning just how far that capability extends.
The results so far are stunning. A new mRNA vaccine against RSV — a virus that kills more than 100,000 older adults annually — proved 83.7% effective in a trial of 35,000 people and earned FDA approval in May 2024. Personalized mRNA cancer vaccines, customized for each individual patient’s tumor, are showing early signs of delaying recurrence in melanoma and pancreatic cancer. A vaccine against cytomegalovirus — a pathogen that causes lifelong disability in thousands of babies born each year — is in late-stage trials. And mRNA-based flu vaccines, potentially faster to manufacture and more adaptable than current egg-based shots, are moving through clinical development.
The COVID vaccines were the proof of concept. What’s happening now is the proof of principle.
The Discovery Nobody Wanted
To understand why this moment matters, you have to understand how improbable it is.
mRNA — messenger RNA — is the molecular courier that shuttles genetic instructions from DNA to the cell’s protein-making machinery. In principle, if you could deliver synthetic mRNA into cells, you could get those cells to produce any protein you wanted. For vaccines, the application is obvious: inject mRNA encoding a viral protein, and the immune system would learn to recognize that protein without ever encountering the real pathogen.
The problem, and it was a serious one, was that synthetic mRNA was toxic. When injected into mammals, it triggered a ferocious immune response — not against the encoded protein, but against the mRNA itself. The innate immune system, which evolved to detect and destroy pathogens, treats foreign RNA as a danger signal. It responds with inflammation, fever, and shutdown of the very cellular machinery needed to translate the mRNA into protein.
For decades, this seemed like a fundamental barrier. The field moved on.
Karikó didn’t. She and Weissman, an immunologist at Penn, noticed something curious: naturally occurring human RNA doesn’t trigger this response. The difference, they hypothesized, lay in chemical modifications to individual nucleosides — the building blocks of RNA. In humans and other mammals, a fraction of uridine residues are converted to pseudouridine, a slightly different molecular shape. This modification, they suspected, might be what prevented the innate immune system from flagging the RNA as foreign.
They were right. Their 2005 paper in Immunity showed that incorporating pseudouridine into synthetic mRNA completely suppressed activation of the Toll-like receptors responsible for the danger signal response. Modified mRNA didn’t just avoid triggering inflammation — it also produced protein more efficiently, because the cell’s ribosomes weren’t being shut down by competing immune activation. The paper has accumulated more than 2,500 citations and anchors a Nobel Prize.
At the time, almost nobody cared.
What changed was COVID. Two biotech companies — Moderna and BioNTech, both founded partly on this research — had been quietly developing mRNA platforms for cancer and infectious disease. When SARS-CoV-2 emerged in early 2020, they pivoted immediately. Within months, they had clinical candidates. Within a year, they had vaccines authorized for emergency use. The mRNA-1273 (Moderna) and BNT162b2 (Pfizer/BioNTech) shots worked better than almost anyone expected — 94-95% efficacy against symptomatic COVID-19 in the original trials — and were produced at a scale and speed that would have been unimaginable with traditional vaccine technology.
The platform had arrived. And the question everyone started asking was: what else can it do?
A New Weapon Against RSV
Respiratory syncytial virus doesn’t make headlines the way COVID did. It should.
RSV causes roughly 160,000 hospitalizations and 6,000 to 10,000 deaths among adults over 65 in the United States every year, and somewhere between 100,000 and 200,000 deaths globally — most in older adults and very young children. For most healthy adults, RSV produces a miserable cold. For older adults, especially those with chronic lung or heart disease, it can cause pneumonia and respiratory failure. Until recently, there was no approved vaccine.
The challenge was a protein called the fusion glycoprotein — the spike-like structure RSV uses to merge with and enter cells. The protein is the obvious vaccine target, but it’s a shapeshifter: it has a “prefusion” conformation (before it fuses with a cell) and a “postfusion” conformation (after). Antibodies trained against the postfusion shape turn out to be largely useless at preventing infection, because the virus completes fusion before those antibodies can do anything. What you need are antibodies against the prefusion conformation. Getting the immune system to make those means presenting it with the prefusion form — which is unstable and hard to manufacture.
mRNA solves this elegantly. You don’t need to manufacture the protein at all. You encode it in mRNA, deliver it to cells, and let the cells make the prefusion-stabilized version themselves. Engineers at Moderna designed mRNA-1345 to encode a prefusion-stabilized RSV fusion protein — a specific molecular conformation locked in place by strategic amino acid substitutions.
The trial was one of the largest vaccine trials in recent memory. The ConquerRSV study enrolled 35,541 adults aged 60 and older, randomized 1:1 to vaccine or placebo, at sites across the US, Europe, and beyond. The primary result, published in the New England Journal of Medicine in December 2023: vaccine efficacy of 83.7% (95% CI 66.0–92.2%) against RSV-associated lower respiratory tract disease with at least two symptoms, and 82.4% against the more severe definition requiring three or more symptoms. Protection extended to both RSV-A and RSV-B subtypes, and was consistent across age groups and among participants with underlying conditions. The FDA approved the vaccine — branded as mRESVIA — in May 2024, the first mRNA vaccine approved for an indication other than COVID-19.
The achievement matters beyond RSV. For decades, a prefusion-stabilized RSV vaccine seemed like a worthy goal but an intractable engineering problem. mRNA technology cut through that problem because the difficult protein doesn’t need to be manufactured and purified — the cell makes it fresh. That same capability applies to a large class of viral proteins that are structurally unstable or hard to produce conventionally.
Teaching the Immune System to Fight Cancer
Cancer is a different challenge. Viruses present consistent targets — the same spike protein, the same surface antigen, more or less reliably across millions of infections. Cancer is chaos. Every tumor carries a different constellation of mutations, and many of those mutations produce altered proteins — neoantigens — that the immune system could, in principle, recognize as abnormal. But which neoantigens are immunogenic? Which will actually trigger a T cell response? And how do you make a vaccine against something unique to one person’s tumor?
This is the promise of personalized mRNA cancer vaccines: read the tumor’s genome, identify the mutations most likely to produce immune-stimulating neoantigens, synthesize mRNA encoding those targets, and vaccinate the patient against their own cancer.
The first major clinical evidence came from KEYNOTE-942, a randomized phase 2b trial involving 157 patients with completely resected high-risk melanoma — the kind of melanoma likely to recur after surgery. Patients were assigned 2:1 to receive either mRNA-4157 (V940), Moderna’s personalized neoantigen vaccine, plus pembrolizumab (an immune checkpoint inhibitor), or pembrolizumab alone. The vaccine was customized for each patient based on sequencing of their individual tumor.
The results, published in The Lancet in January 2024, showed a 44% reduction in recurrence or death with the combination (hazard ratio 0.561; 95% CI 0.309–1.017; p=0.053). At 18 months, 79% of patients in the combination group were recurrence-free, versus 62% in the pembrolizumab-only group. A three-year update presented at ASCO showed the gap widening: 74.8% versus 55.6% recurrence-free survival at two and a half years. The p-value in the original analysis was 0.053 — just above the conventional threshold, in a small trial — but the effect size is striking, and the biological story is coherent. A phase 3 trial is underway.
Meanwhile, an even more ambitious application is being tested in pancreatic cancer.
Pancreatic ductal adenocarcinoma is among the most lethal cancers known. The five-year survival rate is around 12%. It responds poorly to chemotherapy and almost not at all to immunotherapy — the tumor microenvironment is intensely immunosuppressive, and pancreatic cancer tumors have relatively few neoantigens compared to melanoma.
A 2023 phase 1 trial led by Memorial Sloan Kettering and BioNTech, published in Nature, tried a different approach. Sixteen patients with resected pancreatic cancer received a personalized mRNA vaccine (autogene cevumeran), encoding up to 20 neoantigens per patient, followed by chemotherapy. In 8 of the 16 patients — exactly half — the vaccine induced robust neoantigen-specific T cell responses. These vaccine-expanded T cells comprised up to 10% of all T cells in the blood, a remarkable expansion. In some patients, the T cells targeted more than one vaccine neoantigen.
The clinical signal was equally striking. At 18 months, patients who showed strong T cell responses had significantly longer recurrence-free survival — median not yet reached — compared to non-responders, whose median recurrence-free survival was 13.4 months (p=0.003). It’s a small trial, and the survival difference could reflect differences in underlying tumor biology as much as vaccine response. But the fact that half of pancreatic cancer patients — a tumor type generally considered “cold” and immunologically unresponsive — mounted a meaningful T cell response is genuinely new information about what’s possible.
How Personalization Works
The technical pipeline behind personalized mRNA vaccines is itself worth pausing on.
When a tumor is removed surgically, its DNA is sequenced alongside the patient’s normal cells. The differences — somatic mutations specific to the tumor — are computationally filtered for those likely to produce altered peptides that will be presented on the surface of cancer cells and recognized by T cells. Algorithms trained on large datasets of known T cell responses predict which candidate neoantigens are most likely to be immunogenic. From this shortlist, mRNA sequences are designed, synthesized, encapsulated in lipid nanoparticles, and produced within weeks.
That timeline — from surgery to first vaccine dose — is still measured in weeks to months, not days. But it’s shrinking. And unlike traditional biologics, mRNA can be synthesized without biological manufacturing infrastructure: no cell cultures, no fermenters, no protein purification columns. Once you have a validated synthesis pipeline and a lipid nanoparticle formulation, the main constraint is computational: figuring out which neoantigens to target.
This is where artificial intelligence is accelerating the field. Next-generation neoantigen prediction models are trained on larger datasets and use structural information about antigen presentation — not just peptide sequence alone — to improve their hit rates. Better prediction means fewer wasted neoantigen slots, more effective vaccines.
What’s Still Coming
CMV — cytomegalovirus — is one of the next major targets. More than half of adults carry CMV latently, usually without consequence. But for women who acquire it during pregnancy, the consequences for the fetus can be severe: congenital CMV is the leading infectious cause of permanent disability in children, producing hearing loss, cognitive impairment, and neurological deficits in the most affected cases. The US alone sees roughly 30,000 children born with symptomatic congenital CMV each year.
Moderna’s mRNA-1647 is a six-component mRNA vaccine encoding different CMV proteins — a more complex design than the RSV or COVID shots, reflecting the CMV’s evasion strategies. A phase 1 trial established it was safe and immunogenic. A phase 3 trial (CMPASS) is ongoing.
Influenza is another frontier, and one with particular urgency. Current flu vaccines are manufactured in eggs — a process that takes months, limits supply, and requires the WHO to guess six months in advance which strains will circulate that winter. mRNA flu vaccines could be designed and manufactured in weeks, tracking circulating strains with much higher fidelity. Several mRNA flu candidates are in clinical trials; efficacy data are expected over the next few years.
And the cancer vaccine field continues to expand rapidly. Beyond melanoma and pancreatic cancer, personalized mRNA vaccines are being tested in lung cancer, colorectal cancer, and other solid tumors. The KEYNOTE-942 data prompted a phase 3 trial, KEYNOTE-V940-001, which will provide the definitive efficacy data.
The Real Lesson
There is a version of this story that focuses on the technology — the lipid nanoparticles, the pseudouridine substitution, the computational neoantigen pipelines. That story is accurate and remarkable.
But the deeper story is about what happened when a single platform unlocked simultaneous progress in multiple diseases. Before mRNA, each vaccine was essentially a separate engineering project. The antigen had to be identified, expressed, purified, formulated, and tested — a bespoke process for each pathogen. mRNA changes the algebra: the delivery system, the formulation, the manufacturing process, the immune stimulation machinery — all of these are the same. Only the sequence changes.
This is what platform thinking looks like in medicine. The same insight that let Moderna pivot from its cancer pipeline to a COVID vaccine in days — because the lipid nanoparticle and production process were already in place — is what now allows the rapid development of an RSV vaccine, a personalized cancer vaccine, a CMV vaccine. Each new application is easier than the last because the infrastructure already exists.
Katalin Karikó’s 2005 observation that nucleoside modification prevents mRNA from triggering innate immune alarm systems was not, on the surface, a clinical breakthrough. It was a basic science finding about how the immune system distinguishes self from non-self — a curiosity about the molecular logic of immune surveillance.
What she and Weissman built was not just a better vaccine. It was a way to write instructions for the immune system in a language it finally accepts.
That language turns out to be more versatile than almost anyone imagined. Scientists are just beginning to find out what it can say.