Two patients receive the same immunotherapy for the same cancer. In one, the tumor retreats and stays gone for years. In the other, the treatment does nothing. Oncologists still have no reliable way to tell these patients apart before therapy begins. Why does immunotherapy succeed for some and fail for others?

A new study led by Prof. Dvir Aran of the Technion Faculty of Biology and the Henry and Marilyn Taub Faculty of Computer Science, together with first author Dr. Zhongyang Lin and collaborators including Prof. Jürgen C. Becker of the German Cancer Consortium (DKTK), offers a new way to think about that question. Its central message: the answer may depend less on how a tumor looks before treatment begins than on how it changes during the first weeks of therapy.

At the center of the study is the tumor microenvironment, the complex ecosystem of immune cells, blood vessels, and structural cells that surround and interact with the tumor. This environment can either support the immune system’s attack or suppress it, and researchers have long suspected it plays a major role in whether immunotherapy succeeds or fails.

Until now, however, most studies examined the tumor microenvironment at a single point in time, before treatment began, trying to predict a patient’s fate from that initial snapshot alone. “But tumors are not static,” said Prof. Aran. “Immunotherapy is not just a test; it actively reshapes the tumor environment.”

The researchers turned their attention to the earliest stages of treatment, when the immune system and the tumor first engage. According to Prof. Aran, “A useful analogy is a chess game. You cannot predict the outcome just by looking at the opening position. You need to see how the players respond to each other’s first moves. In cancer, one of those players is the patient’s immune system.”

Building on this analogy, Prof. Becker emphasized that, “What is particularly exciting is that these early interactions may not only explain why patients respond or fail to respond to treatment. Much like in chess, where opening moves can already reveal a player’s strategy, our findings suggest that early treatment-induced changes may allow us to anticipate the course of the immune response at a very early stage. This could open new opportunities to tailor therapy before clinical success or failure becomes apparent.”

From individual studies to a broader picture

Watching those first moves is technically demanding. The most powerful tool for the job, single-cell RNA sequencing, which profiles a tumor one cell at a time, is expensive and labor-intensive. As a result, previous studies typically followed only 10 to 20 patients: enough to glimpse intriguing patterns, but too few to connect them reliably to clinical outcomes.

To break through that limit, the team did two things. They generated their own single-cell data from melanoma patients sampled in the first days of treatment, and they brought together data scattered across the field. “We realized the data already existed, but it was scattered,” said Dr. Lin. “By bringing it together, we could finally identify patterns that were neglected in individual studies.”

Assembling 16 independent cohorts, the researchers built one of the largest datasets of its kind, with nearly 200 patients across multiple cancer types, each sampled sequentially: before treatment and shortly after immunotherapy began.

It is not where you start, it is where you can go

Using advanced computational and AI-based methods, the researchers distilled this sprawling data into four recurring states of the tumor microenvironment that appear repeatedly across different cancers. At one end of the spectrum sat immune-rich, inflamed environments, including tumors crowded with B cells, where the immune system was actively engaged. At the other sat environments dominated by suppressive myeloid cells or largely emptied of immune activity altogether.

But the real discovery was not the states themselves. It was what happened to them during treatment. In about half of the patients, the tumor environment stayed put. In the other half, it shifted from one state to another, and the direction of that shift mattered more than where the tumor began. Movement toward inflamed, immune-rich states tracked with strong responses; drift toward suppressive states marked treatment failure. “In other words, it is not just where you start,” said Prof. Becker. “It is where you can go.”

Reading the opening position

That raised a tantalizing question. If you cannot judge a tumor by its opening position, can you at least read that position for clues about which way the game will break? The researchers found that you can. From pre-treatment samples alone, they built a “transition score” that estimates a tumor’s capacity to move toward a favorable immune state once therapy begins, capturing not what the tumor is, but what it can become. Tested on a much larger dataset of approximately 1,300 patients, the score predicted who would respond, matching established biomarkers while reading an entirely different signal: the tumor’s potential to change.

Toward more effective immunotherapy

The findings are not yet ready for the clinic. But they point to a shift in how researchers and clinicians might think about cancer immunotherapy: not as a matter of sorting tumors into fixed categories, but of watching, and perhaps steering, how they evolve. Rather than asking only whether a tumor is “hot” or “cold,” the work suggests asking which way it will move once treatment begins and eventually developing therapies that nudge that evolution in a favorable direction.

“Our results suggest that improving immunotherapy may require not only identifying the right patients,” said Prof. Aran, “but also understanding, and eventually influencing, how their tumors respond in real time.”

That is already the next step. Prof. Becker added that, “To this end, we have launched a collaborative research initiative called DYNAMO to validate and expand upon these findings. We are delighted that the project has very recently been awarded the corresponding grant.”

For now, the message is one of perspective. The fate of an immunotherapy patient may be decided not by a single snapshot taken before the first dose, but in the opening moves that follow, and learning to read those moves early could one day let doctors change the game while it is still being played.

About the study

The study, titled “Tumor Immune Microenvironment Remodeling Predicts Response to Checkpoint Inhibitor Therapy,” integrates longitudinal single-cell data across multiple cancer types to uncover how early treatment-associated changes in the tumor microenvironment link to clinical outcomes.

What factors influence students’ decisions about whether they will continue to use ChatGPT for learning in the future? This was the question addressed by a new study from ELTE’s Faculty of Informatics, which used an interpretable machine learning framework to identify the factors that most strongly influence the use of generative artificial intelligence in higher education.