The rise of immunotherapy has been one of the most startling and promising developments in cancer research for some time. After decades of false starts and dead ends, scientists have finally found effective ways of marshaling the immune system to destroy cancers. Some use drugs called “checkpoint inhibitors” to lift the natural brakes that restrain immune cells, allowing them to go to town on tumors. Others are extracting, engineering, and re-injecting the immune cells themselves. The results have been staggering. Advanced cancers have gone into complete remission. People who were given months to live are still here years later.

But immunotherapy isn’t a panacea. For the moment, it only works for some types of cancer. Even then, only about 20 percent of patients respond. When these treatments work, they work really well—but they don’t always work. Why? To answer that,  Sergio Quezada at University College London and Charles Swanton at the Crick Institute realized that they needed to answer a simple question: How does the immune system see tumors, and what are they actually seeing?

As tumors expand, they accrue mutations in their genes that fuel their growth, and that distinguish them from normal, healthy cells. These mutations also change molecules called neoantigens that are displayed on the surface of the tumor cells. To our immune system, neoantigens are red flags that say “There’s something weird, foreign, other about these cells.” It responds by producing targeted weapons—T-cells that specifically recognize the neoantigens and attack whatever carries them.

Quezada and Swanton’s team, including Nicholas McGranahan, Andrew Furness, and Rachel Rosenthal, found that lung tumors carry anywhere from 80 to 700 neoantigens. They also found, after studying 150 lung cancer cases, that people with more of these red flags tend to live longer. That makes intuitive sense: More neoantigens means more potential targets for the immune system, which means a better chance of controlling a tumor.

But not all neoantigens are alike.

As cancers develop, they also evolve and diversify. Cells on one side of a tumor can end up with very different mutations (and neoantigens) from those just centimeters away. This concept, known as ‘heterogeneity,’ partly explains why the war against cancer has been so entrenched. Consider the much-vaunted “targeted therapies”—drugs that go after mutations specific to a patient’s cancer. If those mutations are found only in some parts of a tumor, the treatment will leave a reservoir of cells that can grow anew, or even evolve resistance to the drug. That’s why many people get great results with targeted therapies, but then rebound within a few months.

Heterogeneity matters to the immune system, too. Some mutations develop early on in a tumor’s life and are found in all of its cells. Let’s call them trunk mutations. Others are latecomers and found in just a fraction of the tumor cells. Those are the branches.

Quezada and Swanton’s teams found that patients had better survival rates if their tumors have lots of trunk neonatigens, but not branch ones. They also responded better to pembrolizumab, one of those checkpoint inhibitors that works by unleashing the immune system (and the drug Jimmy Carter recently took). The team studied 34 patients and found that almost everyone whose tumors had a wide trunk and sparse branches responded well to the drug. By contrast, the poorest responders almost all had thin trunks and luxuriant branches. They found the same pattern among 64 melanoma patients treated with two different drugs.

This makes total sense. The immune system only has a limited armada of T-cells at its command. If it deploys these against branch neoantigens, it is “wasting resources on the branches,” says Quezada. And you can’t chop down a tree by pruning back the branches. You need to go after the trunk. “Our data tell us with absolute certainty that if we want to target every single cell on the tumor, we need to find the trunk mutations.”

In Quezada’s vision, when his team meets a new cancer patient, they would identify trunk mutations that are found throughout their tumor, make T-cells that recognize those mutations, and inject those cells into the patient. Robert Gatenby from the Moffitt Cancer Center compares this approach to biological control, where farmers control pests using predators and parasites. “Cancer cells are analogous to pests, and predators are often a far more effective strategy for controlling pests than application of toxic pesticides.”

Indeed, predators already prowl the body. Quezada and Swanton’s team focused on two people with very different kinds of lung tumors—one with lots of trunk neoantigens and another with lots of branches. But in both cases, they found T-cells that recognize the trunk mutations. “This is the first evidence that there are seeds of a tumor’s own destruction nestling in the tumor itself,” says Swanton.

And yet, these trunk-targeting T-cells are clearly not living up to their full potential, because the two patients aren’t healthy. “That’s no surprise. If these cells are potent, the tumors will evolve ways of turning them off,” says Quezada. “We need to work out how to activate these T-cells.”

The team have already found potential ways of doing that. These T-cells harbor high levels of molecular restraints like PD-1 and LAG-3, and could potentially be unshackled by checkpoint blockade drugs that go after these targets. Maybe the future lies in finding these trunk-targeting T-cells, growing them outside a patient’s body, and injecting them back in along with drugs that let them go to town.

“We’re not naïve,” says Swanton. “We know there is a long way to go. This is the first step in an exciting journey.”