The first cancer vaccine to work: every dose is a new drug for one person
After more than a thousand failed clinical trials, an mRNA personalized cancer vaccine has cleared Phase 3 for the first time. About 90% of the antigens differ from person to person, so personalization isn't an edge case — it's the precondition for this drug to exist at all.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
The first success after more than a thousand failed trials
Stefan says this is the first therapy to beat Keytruda monotherapy in melanoma, and the first time a cancer vaccine has actually worked. The field has spent over 20 years and run more than a thousand clinical trials, all of them failures. This Phase 3 hit its primary endpoint, recurrence-free survival — meaning a significant drop in the risk that a patient relapses or dies. More surprising, it also hit a secondary endpoint, distant metastasis-free survival; distant metastasis takes longer to mature and wasn't expected. He repeatedly stresses that this is only the first interim analysis, not the end of the study.
— Stefan BancelKeytruda's ceiling is 40% of people treated for nothing
Checkpoint inhibitors like Keytruda work by "letting the dog off the leash" so the immune system attacks the cancer, but only 60% of patients are disease-free at five years. The other 40% go through the treatment, endure immune-related side effects, and get none of the benefit. The side effects aren't chemo-style; they're autoimmune disease — type 1 diabetes, lupus, Crohn's disease. Stefan's words: think about those 40% who don't respond — most of the time they get an autoimmune disease and they don't get the benefit of the drug. Moderna's mechanism is orthogonal: it doesn't unleash the dog, it tells T cells at the molecular level exactly what to look for.
— Stefan BancelThe point of mRNA is letting the immune system see it from the inside
Stefan breaks the difference into two parts. First, delivery: injected into muscle, the mRNA reaches lymph nodes and APCs (antigen presenting cells), gets translated inside the immune cell and presented "from the inside." Cancer vaccines built from proteins or peptides in the past were manufactured in a reactor and then injected; once in the blood they circulate, and the immune system sees them in a completely different way. Second, personalization: cancer is a disease of DNA, sequencing costs have fallen dramatically over 20 years, so you can biopsy the tumor, read out every DNA letter, compare it letter by letter, nucleotide by nucleotide, against healthy cells, and use an algorithm to pick the most relevant mutations.
— Stefan Bancel90% of antigens differ between people, so it can only be one drug per person
Moderna picks the 34 most relevant mutations out of hundreds to thousands and strings them into one large mRNA, made for that one patient in about 30 days. The data presented at ASCO: roughly 90% of the antigens are different between patients. Stefan says that when they started they had no idea whether it would be 2%, 5% or 90%, because the whole field had gone down the shared-antigen route. That number directly determines the business shape — if 90% don't overlap, personalization isn't an optional feature, it's the only way this drug can work.
— Stefan BancelThe algorithm was written ten years ago, and this is only version 1.0
Stefan says the algorithm used in Phase 3 is the same one used in Phase 2 and Phase 1 — an algorithm from ten years ago. It already gets 80% of patients to five years disease-free, but 20% still don't respond. Now, with all the blood samples and sequencing data from Phase 3 patients, they plan to go back and mine the data to figure out why some respond and some don't, then talk to the FDA about whether they can move from 1.0 to 2.0 — provided they don't lose efficacy. His analogy: everyone jokes that today's AI is the worst AI we'll see in our lifetime; Intismeran is the same, this version is the worst version in the history of medicine.
— Stefan Bancel42 days needle to needle, and no need to take out your cells
From taking the biopsy to the vaccine being ready at the hospital is now about 42 days, needle to needle. The biggest difference from CAR-T: you don't have to take out the patient's immune cells, engineer them in a factory reactor and ship them back to the hospital. Moderna only needs information — the sequences of the healthy cells and the cancer cells, i.e. a file. After that it's fully synthetic, fully enzymatic, done in water, then wrapped in lipids. Because you don't have to grow cells, the reactor can be made extremely small. Stefan says they've been working on "shrinking everything" since before entering the clinic; the first version of the machine was like a big refrigerator, large and hard to handle, because at that point the only requirement was quality, not efficiency.
— Stefan BancelWhat he's really obsessed with is square inches, not science
Only after the Phase 2 data confirmed the science did Moderna move its smart engineers over to work on efficiency. Stefan names two vectors: one is cycle time — the faster the turnaround, the more times an asset turns over in a year; the other is square inches — cleanroom area is a fixed envelope, and if you can fit 2x or even 10x the machines into the same area, throughput goes up, fixed costs get spread thinner, and ultimately the product price comes down. He says every time he goes to the factory he's picking at things, looking at how much space can be saved, even moving some of the compute out of the cleanroom. The Marlboro plant can produce tens of thousands of doses, and the incidence of melanoma means tens of thousands of doses would cover the entire market.
— Stefan BancelThe FDA approves the process, not a single dose
Every dose is different, so approving dose by dose is impossible. Stefan says the precedent is CAR-T: approved as a process BLA rather than a product BLA. Moderna has five products approved as products; this one has been a process IND since it entered the clinic. The FDA's question is very concrete: if you start with the same sample — tumor and blood — does the same product come out of the black box at the end. That requires proving it to yourself first, then proving the robustness of the whole process to the FDA with data. He stresses this isn't a case of not talking to the FDA for ten years and then suddenly showing up; they've been in meetings, answering questions, and incidentally educating the regulators along the way.
— Stefan BancelThree directions to expand, including where Keytruda does nothing
Moderna has already shown it can generate de novo new T cells that didn't exist in the body before treatment. Three directions: first, everywhere Keytruda works, because the mechanism is orthogonal — lung cancer in Phase 3, kidney and bladder cancer in Phase 2; second, early disease, where checkpoints aren't used yet, such as a Phase 3 in early lung cancer starting in spring 2026, using Intismeran as monotherapy, because early-stage patients don't deserve a lifetime of autoimmune side effects for a checkpoint; third, places where checkpoints do nothing at all, pancreatic and gastric cancer, where past trials were all negative, but the mechanism is different so it's worth a try. He also mentions combining with Revolution Medicine's recently approved pancreatic cancer drug.
— Stefan BancelIn their own words · checked verbatim
It's the first time that there is an agent in melanoma that is better than Ketudra alone. So it's a big deal for patients, of course, with melanoma. It's the first time there is a cancer vaccine working.
Stefan Bancel4:05
But only 60% of people are disease-free after five years. If you look at the phase three published data of K-Truda.
Stefan Bancel7:06
But think about those 40% of people who don't respond to checkpoints. They get an autoimmune disease most of the time, and they don't get the benefit of a medicine.
Stefan Bancel8:08
And actually 90% of the antigen are different from a human to another one.
Stefan Bancel15:23
Well, it's exactly the same for Intismeran, which is the current version of Intismeran that Moderna is the worst version of Intismeran you're going to see for the rest of medical history.
Stefan Bancel17:25
The only thing we need is the information. As you and I talked about, the beautiful thing of mRNA is an information molecule.
Stefan Bancel21:28
Basically, Jorge, what the FDA wants to know, which is very legitimate, which I would want for my own family's sake, obviously, which is if you get the same sample at the beginning, the tumor and the blood, do you get the same product make at the end of a big black box?
Stefan Bancel28:56
Figures
| Keytruda five-year disease-free survival rate | 60% | 7:06 |
| Five-year disease-free survival in the Phase 2 study | About 80% | 6:05 |
| Share of antigens that differ between patients | About 90% | 15:23 |
| Mutations selected into the mRNA per patient | 34 | 14:22 |
| Personalized vaccine production cycle | About 42 days, needle to needle | 20:28 |
| Manufacturing time for a single mRNA | About 30 days | 14:22 |
| Marlboro plant capacity | Tens of thousands of doses | 24:36 |
Glossary
- recurrence-free survival
- The time a patient survives after treatment without the cancer coming back or dying; the primary endpoint of this Phase 3.
- distant metastasis-free survival
- The time before the tumor spreads beyond the original site; the secondary endpoint unexpectedly met here.
- APC (antigen presenting cell)
- An immune cell that the mRNA enters, gets translated inside, and presents antigen from.
- process BLA
- Approval based on the manufacturing process rather than a single product; CAR-T was approved this way.
- Intismeran
- The code name for Moderna and Merck's personalized mRNA cancer therapy.
- de novo T-cell
- A new T cell that didn't exist in the body before treatment, taught by the vaccine to recognize the tumor.
How to listen
Founders and investors in biotech, CDMOs, personalized medicine or automated manufacturing; engineers who want to understand how the mRNA platform extends from vaccines into oncology and autoimmune disease.
The show intro at 0:00-2:00 and the subscription and disclaimer segment after 39:00.