Moderna and Merck announced that an individualized mRNA neoantigen therapy achieved recurrence-free survival in combination with pembrolizumab in resected, high-risk melanoma. It is the first positive Phase 3 result for a cancer vaccine. Many are celebrating that this vindicates cancer vaccines and mRNA technology. I would argue that it makes the case for more combination studies in oncology.

The Paper That Could Have Ended Cancer Vaccines

It's remarkable that cancer vaccines survived long enough to produce a result like this, one that gives real hope to patients who need it. In September 2004, Steven Rosenberg published a paper in Nature Medicine reporting that the objective response rate of cancer vaccines in trials run at the National Cancer Institute was 2.6%. Across 40 published studies from other groups, the number barely improved, to 4.0%.

Dr. Rosenberg is not an outside critic. He is described by NIH's own materials, as "the father of cancer immunotherapy", the Chief of the NCI Surgery Branch since 1974; developer of the first FDA-approved cancer immunotherapy; first to demonstrate tumor infiltrating lymphocyte (TIL)-based adoptive cell transfer; first to treat a patient with CAR T cells; and the 2024 AACR Lifetime Achievement Award recipient. When the most credentialed person in the field publishes his own data showing cancer vaccines weren’t working everyone listens.

TAAs vs. Neoantigens

The vaccines Rosenberg's data indicted were built against tumor-associated antigens, or TAAs: proteins the body already makes, simply overexpressed on cancer cells. TAAs are self-antigens. They are the same proteins our immune systems are trained, from early development, to tolerate rather than attack, a process called central tolerance. Central tolerance prevents our immune system from attacking ourselves. It also strips away most of the T cells that would otherwise be capable of recognizing an overexpressed self-protein and killing the cell carrying it. In contrast, anti-viral antigen vaccines have had tremendous success because T cells that recognize viral antigens are not stripped away from our immune system during development. The vaccines Rosenberg tested were asking T cells to override training that exists specifically to prevent exactly what the vaccine wanted them to do.

The field's answer was to stop targeting TAAs and start targeting neoantigens. Neoantigens are mutated versions of our own proteins. They are still self, but they are mutated self, and different enough that central tolerance doesn't reliably delete the T cells that would recognize them. They sit somewhere between a viral antigen and a TAA, which is an unmutated self-protein. This was the alternative path around the exact constraint Rosenberg had identified in 2004.

But neoantigens carry a manufacturing problem TAAs never had. Every patient's tumor mutates differently, so a neoantigen vaccine has to be built new for every patient, and from a sequence nobody knows until the tumor is sequenced. That technology didn't exist in 2004. Moderna's mRNA platform was built to solve exactly this by quickly encoding and manufacturing vaccine against a specific sequence. Between March 2018 and November 2019, Rosenberg's group at the NCI sequenced up to twenty neoantigens, selected from each patient's own tumor, and had Moderna manufacture a personalized mRNA vaccine encoded specifically for them. Four patients were treated. There were no objective clinical responses.

Rosenberg published the result in the Journal of Clinical Investigation in October 2020. Reviewing the neoantigen vaccine trials run by his own team and others, he and his co-authors wrote that although the trials had demonstrated feasibility, safety, and immunogenicity, clear evidence of their clinical efficacy was lacking. In the discussion, they wrote plainly that "vaccines as single agents have not historically been effective in eradicating established tumors across a wide range of antigens," and concluded that "potential future combination of such vaccines with checkpoint inhibitors or adoptive T cell therapy should be evaluated for possible clinical benefit."

As no clinical response was observed, and because Moderna is now conducting a clinical trial with the combination of the anti-PD-1 agent, we decided not to proceed with the phase II part of the clinical trial.

That Moderna trial was KEYNOTE-942, already enrolling as he wrote. It reported a 44% reduction in the risk of recurrence or death compared to pembrolizumab alone in 2022, was published in the Lancet in 2024, and led to the Phase 3 that reported this week.

Two Drugs, Two Different Jobs

The mechanism at the heart of the Moderna and Merck result is not a vaccine succeeding where vaccines used to fail. It's two drugs combining two different jobs necessary for a breakthrough outcome.

Moderna’s mRNA, neoadjuvant vaccine produces nothing on its own. Rosenberg's 2020 trial showed that directly, there were zero objective responses albeit in only four patients. Pembrolizumab already works in this setting. In the trial that resulted in its approval as an adjuvant therapy in resected melanoma, it cut the risk of recurrence or death by 43% against placebo producing a real but incomplete effect. Adding Moderna’s mRNA vaccine pushes the effect further than pembrolizumab reaches on its own. Pembrolizumab removes a brake on the immune system, but it can only unleash the T cells a patient's own immune system has already produced against the tumor. The vaccine mobilizes the T cells pembrolizumab needs and primes them against sequences specific to the patient's tumor.

Only together does each drug complement what the other is missing. The result reported this week isn't a vaccine vindicated after twenty years, or a checkpoint inhibitor's ceiling being raised by something that had never, on its own, cleared the floor. Cancer doesn't arise from one mutation or one immune failure, it’s a complex and systems-level disease that requires combinatorial therapies

Combinations Can Lead to Breakthroughs

An early-stage company cannot easily raise capital on a combination story. Investors want to see a molecule work by itself before they believe it does anything at all, and a monotherapy signal is the cheapest way to prove a mechanism is real. That preference is reasonable for most drug classes. It's a poor match for a disease that, as the biology above shows, sometimes only reveals an effect in combination.

Combinations have driven some of medicine's largest gains. Triple-drug therapy turned HIV from a near-certain death sentence into a manageable chronic condition within a few years of replacing single-agent treatment. Tuberculosis has been treated with multi-drug regimens for decades, specifically because a single agent leaves room for resistant bacteria to survive. Combination chemotherapy protocols go back to the 1960s for the same reason. None of this is new, and none of it is rare.

What is rare is a way to test a combination without someone paying a specific price for it. Pairing two experimental drugs multiplies regulatory uncertainty: a company combining an unproven agent with an already-approved backbone, as Moderna did by pairing its vaccine with pembrolizumab, faces a far more direct path than two companies combining unapproved drugs, where regulators may require evidence of what each component contributes on its own. Combination trials also put reputations at risk in a specific direction. A company that tests its molecule alongside a partner drug risks proving the partner alone was doing the work, or that its own drug was less effective.

Rosenberg's 2020 trial was that risk realized. Four patients, no responses, published plainly rather than buried or spun. That's close to the worst outcome a company testing a novel agent can put on the record. He published it anyway, which is exactly why the sentence he wrote next is worth re-quoting now.

Vaccines as single agents have not historically been effective in eradicating established tumors across a wide range of antigens. Potential future combination of such vaccines with checkpoint inhibitors or adoptive T cell therapy should be evaluated for possible clinical benefit.

As Dr. Rosenberg advises in the paper, we should avoid drawing broad conclusions from a single case study here as well, but the constraint is worth noticing even from a single example. The therapies most worth testing in a complex, systems-level disease are often the therapies hardest to finance through clinical development. Individual failures should not be mistaken for class failures. Rosenberg's four patients did not disprove neoantigen vaccines. They were a measurement of what the vaccines were missing.

Moderna's path may not be repeatable. The mRNA-LNP platform was running multiple programs in parallel. The individualized neoantigen program with Merck had been running alongside infectious disease work since 2016, and it was that work which resulted in the COVID-19 vaccine. The global COVID pandemic provided validation of mRNA vaccines in exactly the kind of acute, monotherapy-solvable problem that the cancer program could never generate. Literally billions of people were administered mRNA vaccines. The platform's core technology risk was validated in a disease that could support proof of monotherapy efficacy.

That validation wasn't for the cancer vaccine, but it removed some of the largest barriers, manufacturing and safety risk, standing between a mechanism with real biological rationale and a company willing to test it in combination.

Before You Call It a Failure

Most combinations won't get that lucky. Most mechanisms that only work alongside something else don't have a pandemic to prove their manufacturing and safety.

The question worth carrying out of this week isn't whether cancer vaccines work. It's what happens to the next therapy that has a real mechanistic rationale and no single-agent efficacy data of its own: whether it gets tested in combination, or whether it gets treated as a failure before anyone asks what it was missing.