In the summer of 1890, a seventeen-year-old named Bessie Dashiell caught her hand between two train-car seats. The pain wouldn't go away.
A biopsy found an aggressive bone cancer — a sarcoma. Surgeons amputated her arm below the elbow. It wasn't enough. The cancer had already spread, and in January 1891, Bessie died [1].
Her surgeon was a young New Yorker named William Coley, and her death broke something in him. He spent weeks combing the hospital's old records for anything that might have saved her.
What he found in those records got him branded a fraud for the next hundred years. And in August 2026 — 135 years later — it produced the first Phase 3 win for a cancer vaccine that was personalized to each individual patient.
It also caused the stock price of Moderna, who built it, to soar by 177% in a single day — the largest single-day percentage gain by an S&P 500 stock this century [2].
Table of Contents
- The Surgeon Who Infected His Patients
- A Century in the Wilderness
- The Graveyard of Failed Vaccines
- The Flaw: Central Tolerance
- How to Build a Personalized Vaccine
- The COVID Detour
- August 19, 2026 — and What It Does Not Prove
- What This Means — Your Questions Answered
- References
The Surgeon Who Infected His Patients
Deep in those files, Coley found a case from seven years before. Fred Stein, a German immigrant, had a sarcoma on his neck that kept growing back after every operation. His doctors called him hopeless. Then Stein caught a severe strep skin infection that tore through hospital wards before antibiotics existed.
And as the infection raged, his tumor melted away.
Coley had to know how that story ended. He spent weeks knocking on doors through the tenements of the Lower East Side — and found Stein. Alive, seven years on. No sign of cancer [3].
Was it the immune system activation to fight off the infection that also caused the cancer to go away?
So in 1891, Coley did something that would end a career today. He took a patient named Zola — an Italian immigrant with a tumor in his tonsil that was too advanced to cut out — and deliberately infected him.
Zola's tumor regressed completely. He stayed cancer-free for years — before, eventually, the tumor returned [1][4].
Coley turned that observation into a treatment: killed bacteria, injected as "Coley's Toxins," given to roughly a thousand patients. Some responded — a few remarkably. Most didn't. And nobody, Coley included, could predict who would respond, and who wouldn't [5].
The clinical world's patience ran out. In 1894, the Journal of the American Medical Association delivered its verdict on the toxin injections as a cure: an "entire failure" [3]. Radiation arrived soon after, with results you could measure and repeat. Coley used his toxins into the 1930s, but the field had moved on. Immunotherapy — as in, using the immune system to treat cancer — was pushed to the margins.
Which left a question hanging for eighty years: was Coley wrong — or just early?
A Century in the Wilderness
For most of the twentieth century, the answer looked like "wrong."
But in 1976, a urologist injected the tuberculosis vaccine — directly into the bladders of nine bladder-cancer patients. It worked. Fifty years later, it is still standard care for high-risk early bladder cancer [6].
In the years that followed, the idea of using the immune system to target cancer grew.
And at the National Institutes of Health, a surgeon named Steven Rosenberg had seen his own version of the Fred Stein immune-activation miracle. In the summer of 1968, as a young resident at the West Roxbury VA Hospital, he operated on a 63-year-old veteran — a routine gallbladder removal. But the man's chart said he should have been dead: twelve years earlier, surgeons had opened him up, found stomach cancer already spread to his liver, and closed him back up. No treatment. Nothing. Rosenberg's own words: "I operated to remove his gall bladder and there was no remaining trace of the cancer… one of the rarest events in medicine" [7].
Something inside him had hunted it down and destroyed it. Rosenberg's prime suspect was the immune system.
It took him twenty years, but by 1988 he had the answer. His team pulled immune cells — T cells — out of patients' own melanomas, multiplied them in the lab into the millions, and dripped them back into the patients' veins. In some patients, the tumors shrank. That was the proof: the immune system's soldiers can recognize cancer, and kill it [8].
And that proof is what launched the gold rush. Because if T cells can already recognize cancer, then a vaccine should be the cheap, simple way to win — a vaccine's whole job is to hold up a target and tell the immune system, "attack anything that looks like this." Through the 1990s, cancer vaccines poured into trials by the dozen, and the field expected them to work.
But across 1,306 vaccine treatments, the objective response rate — tumors actually, measurably shrinking — was only 3.3 percent [9].
What was going wrong? It feels like there should be a breakthrough here. The immune system can destroy cancer on its own — Stein, Zola, and Rosenberg's veteran prove it. T cells taken out of the body and multiplied can destroy cancer — Rosenberg proved that too. But vaccines — the very tool designed to rally those same T cells — failed in 97 patients out of every 100.
That didn't stop companies from trying, but there were four huge failures in the 2000s and 2010s.
The Graveyard of Failed Vaccines
MAGE-A3: 2,312 lung-cancer patients randomized, and the vaccine group's disease-free survival was statistically identical to placebo — a hazard ratio of 1.02 [10].
Canvaxin: the Phase 3 was halted, and the final publication showed patients on the vaccine did worse — median overall survival of 31.4 months versus 38.6 months on placebo [11].
GVAX: its Phase 3 trial in advanced prostate cancer was halted early [12].
And PROSTVAC: its Phase 2 actually missed its primary endpoint (six-month progression-free survival 23% versus 25%), but a secondary analysis showed a tantalizing 8.5-month survival signal [13]. The Phase 3 followed: 1,297 men randomized, median overall survival 34.4 and 33.2 months in the vaccine arms versus 34.3 months on placebo — stopped when the criteria for futility were met [14].
But from those failures came learnings that opened the door for the success in August 2026.
The Flaw: Central Tolerance
Before you were born, your immune system ran a school. Every T cell — the white blood cells that hunt infected and cancerous cells — faced one brutal final exam: attack any protein that belongs to your own body, and you're destroyed. No exceptions. That policy is why your immune system doesn't eat you alive.
Now look at the failed cancer vaccines. Most pointed the immune system at proteins that cancer shares with the body — that's the explanation for the failures. Those T cells were wiped out and were never going to attack the cancerous cells.
Immunologists call that screening process central tolerance.
In 2015, two immunologists — Schumacher and Schreiber — put the way forward into words. Cancer's own mutations create brand-new proteins that exist nowhere else in the body — neoantigens, "neo" as in new. The body's school never saw them. The T cells that can attack them were never deleted. They're still in the army, waiting for a target list [15].
Which leaves one problem. Your tumor's mutations are not my tumor's mutations. Every patient's target list is different. A shelf vaccine can't work — you'd need a personalized vaccine per patient.
How to Build a Personalized Vaccine
Here's what's required.
First: sequence the patient's tumor and their healthy blood, and subtract — whatever is mutated in the tumor but not the blood goes on the candidate list.
Second: predict which of those mutant fragments the patient's immune system can actually display on the cell surface — a target the immune system can't see is useless. Neural AI networks do that prediction. The best-known public tool, NetMHCpan, has been in development since 2007 — trained on more than 850,000 measured protein fragments by 2017 [16], and over 13 million data points by 2020 [17].
Moderna's own selection algorithm is proprietary and unpublished. We know what it does — predict what's displayable, pick up to 34 targets — not how it does it.
Third: encode up to 34 of those targets on one single strand of mRNA, wrap it in a microscopic fat droplet — a lipid nanoparticle — and inject it [18].
In July 2017, this process was tested in 13 humans with melanoma, 5 of whom had metastatic disease already. The patients received a personalized vaccine, and the eight who were disease-free at vaccination all stayed recurrence-free — for up to 23 months [19].
And a team in Boston vaccinated six melanoma patients with protein fragments rather than RNA. Four of six stayed recurrence-free at 25 months, and the two whose cancer progressed had complete tumor regression once an anti-PD-1 checkpoint drug was added [20].
But building a bespoke drug for one patient at a time, with the technology of that era, was incredibly expensive.
The COVID Detour
Then the world handed over the exact technology needed to scale up this concept.
You know this part. mRNA technology for COVID [21].
Researchers used the mRNA vaccine technology to develop personalized therapies for 157 patients with resected melanoma — people who'd had their tumors cut out and were waiting to learn whether it would come back. Randomized: personalized mRNA vaccine plus the checkpoint drug Keytruda, versus Keytruda alone.
In December 2022, the result: recurrence risk down roughly 44 percent. The trial met the endpoint it had pre-committed to [22].
So Merck and Moderna ran the large, follow-up, Phase 3 trial. Double-blind. Placebo-controlled.
On August 19, 2026, it read out.
August 19, 2026 — and What It Does Not Prove
1,137 patients with completely resected Stage 2B to Stage 4 melanoma — cancer removed by surgery, high risk of it coming back. Randomized two-to-one, double-blind: the personalized therapy or a matching placebo, on top of Keytruda, an immunotherapy drug, for everyone — so whatever benefit appeared came on top of today's standard treatment, not instead of it.
The combination treatment successfully met its endpoint, meaning it delivered greater results than Keytruda alone. We don't have the numbers yet, but in the preceding Phase 2 study, there was a 49% reduction in the risk of recurrence or death compared to Keytruda alone, and a 59% reduction in the risk of distant metastasis or death compared to Keytruda alone [23]. There were also no new safety signals observed.
The companies describe it as the first positive Phase 3 trial ever for an individualized neoantigen therapy — and the first for any mRNA-based cancer treatment.
The trial's principal investigator is Georgina Long of Melanoma Institute Australia — joint 2024 Australian of the Year. Her words: the first Phase 3 study to show that a therapy designed on "the unique mutational 'fingerprint' of a patient's own tumor" can reduce the risk of recurrence or death [23].
What This Means — Your Questions Answered
As Patreon member Red asks: "For what kind of cancers do those vaccines work? Do they cure cancer? Or just increase treatment and survival rates by a bit?"
This is for melanoma treatment, not prevention, and not a cure.
But a preventive cancer vaccine is already here — the HPV vaccine. It blocks the virus behind most cervical cancer: US cervical-cancer deaths in women under 25 have fallen more than 60 percent in a decade [24][25].
As Patreon member Blassy points out: "this worked on melanoma — is it reasonable to be hopeful that this should also work for other types of tumors? Based on my understanding of the mechanism, it seems like it should."
Melanoma is one of the most heavily mutated cancers there is — the most raw material for neoantigens — so in theory, one of the easier cancers to build these personalized treatments for.
Merck and Moderna alone are running nine trials of this therapy: two in melanoma, four in lung cancer — three already at Phase 3 — one in kidney, two in bladder. And buried in that list is one of my favorite details in this entire story: in high-risk early bladder cancer, it's being trialled alongside BCG — the same tuberculosis vaccine instilled in 1976 [26].
This, in my view, is the main reason why Moderna's stock price jumped so much. The hope is that this personalized treatment can be translated onto other types of cancer.
And it's bigger than one partnership. In May, Nature Reviews Drug Discovery mapped the whole clinical pipeline of therapeutic cancer vaccines [27].
Patreon member David Sprouse asks the money question: "any idea how affordable is it expected to be, given the need for customization and the extra work involved?"
We don't know. This therapy isn't yet FDA approved or available to the general public. But we have reason to be hopeful.
With this breakthrough will come further innovation. And it's a really exciting path forward, particularly if we see positive results from the other cancer vaccines that are in the pipeline.
References
1. https://www.cancerresearch.org/blog/the-legacy-of-bessie-dashiell
3. https://pmc.ncbi.nlm.nih.gov/articles/PMC1888599/
4. https://www.thepharmacologist.org/william-coley
5. https://pmc.ncbi.nlm.nih.gov/articles/PMC7232517/
6. https://journals.asm.org/doi/10.1128/cmr.00194-23
7. https://pmc.ncbi.nlm.nih.gov/articles/PMC8210396/
8. https://pmc.ncbi.nlm.nih.gov/articles/PMC6237474/
9. https://pmc.ncbi.nlm.nih.gov/articles/PMC1435696/
10. https://pubmed.ncbi.nlm.nih.gov/27132212/
11. https://link.springer.com/article/10.1245/s10434-017-6072-3
12. https://www.fiercebiotech.com/biotech/cell-genesys-halts-vital-2-gvax-trial-advanced-prostate-cancer
13. https://pubmed.ncbi.nlm.nih.gov/20100959/
14. https://pmc.ncbi.nlm.nih.gov/articles/PMC6494360/
15. https://www.science.org/doi/10.1126/science.aaa4971
16. https://academic.oup.com/jimmunol/article/199/9/3360/7977122
17. https://academic.oup.com/nar/article/48/W1/W449/5837056
18. https://doi.org/10.1016/S0140-6736(23)02268-7
19. https://www.nature.com/articles/nature23003
20. https://www.nature.com/articles/nature22991
21. https://www.cnn.com/2021/11/12/health/covid-cancer-biontech-ugur-sahin/index.html
24. https://www.nejm.org/doi/full/10.1056/NEJMoa1917338
25. https://jamanetwork.com/journals/jama/fullarticle/2827212
