Why don’t cancer medicines work the same for everyone

Cancer treatment has gotten more precise over time, as doctors first classified the disease by where it began in the body and, more recently, by the mutations found inside cancer cells to help find the right drugs to treat it.

But why can two people with seemingly similar cancers respond so differently to the same medication? Microsoft researcher Lorin Crawford thinks the answer lies in how tumors actually behave, not just how they're categorized.

A study by Crawford and his team, published today in Nature Methods, marks an important step in helping AI understand how individual cells act and interact with their environment as researchers harness the technology's power to spot patterns traditional approaches may miss.

The research is part of Project Ex Vivo, a collaboration between Microsoft and the Broad Institute with support from the Dana-Farber Cancer Institute. The group's work aims to make cell behavior part of how cancer is categorized and treated, helping combat one of the leading causes of death worldwide by more successfully matching therapies to patients.

"The complexity of the disease is scientifically a very interesting one, but also one where you can have almost immediate impact," Crawford says. "It feels like I'm doing something that's larger than me. Any single finding seems like a step forward in some way."

Looking beyond mutations

Part of the challenge with cancer research is that scientists can lose key signals when they test drugs outside the body. Ex vivo models - cancer cells grown in labs, including mini-tumors called organoids - don't always match what's happening inside a person. That means a medication that looks promising in a petri dish can fall short in a patient.

The Project Ex Vivo team focuses on "cell state" - how cancer cells behave and respond to their surroundings. Cell states can influence which treatments a tumor is sensitive to, how quickly resistance to drugs develops and how aggressive the disease becomes.

In pancreatic cancer, for example, researchers have observed two broad cell states associated with different outcomes and treatment responses. But when tumor cells are grown in the lab, Crawford says, the models often reflect only one of those states, which can make lab results line up poorly with what happens in patients.

"You and I can have the same mutation, but totally different cell states, and that's what really matters downstream," Crawford says.

Fewer mismatches, better bets

If cell state can be measured reliably, Crawford believes it could change cancer treatment in two key ways.

First, it could improve how patients are matched to existing therapies and enrolled in clinical trials. A more nuanced way of grouping tumors could raise the odds that a treatment is tested where it actually has a chance to work.

Second, it could open a new path for drug development itself. Instead of targeting a mutation, researchers could aim to target - or even shift - the underlying state of a tumor, pushing it into a form that's easier to treat.

"The challenge as a clinician is understanding which features within a patient's tumor are likely to shape its behavior and responses to therapies over time," says Srivatsan Raghavan, a medical oncologist and physician-scientist at Dana-Farber and a co-director of Project Ex Vivo. "This research aims to better represent these diverse features in cancer models and produce evidence that mirrors the complexity of tumor behaviors we see in patients."

A statistician in a wet lab

Crawford took an unusual path to this research.

Trained as a mathematician and statistician, he was urged by an advisor to look beyond spreadsheets and learn how the data was being made. He spent most of his graduate work at Duke University straddling two worlds, embedded in a cancer biology "wet lab" where researchers worked hands-on with living cells, not just data.

Even though it initially felt "like being in a foreign country," Crawford says it "was probably the best decision I've ever made, from a career perspective."

He realized his background could help biologists manage the immense amount of data that goes into understanding variations across patients and cancer subtypes.

"Stuff started to click for me," he says.

That experience shaped how he thinks about Project Ex Vivo today - as a way to close the gap between what works in the lab and what actually helps patients. Computational experts and experimentalists sit and work through problems together, using AI tools and tumor samples from anonymous patients.

From a lab problem to reshaping cancer treatment

What began as a focused research effort in 2022 has expanded quickly, fueled by advances in AI. The Project Ex Vivo team uses computational models to run virtual experiments and identify the most promising hypotheses before spending time and money in the lab. AI tools can help predict how a drug might shift one state into another, or how states translate across different cancer types.

As Crawford and his colleagues show in the Nature Methods study, AI models learn more from seeing a wide range of cell behavior than from simply being fed more data - a finding that challenges a common assumption in the field.

"There's a real temptation to think simply scaling up datasets will solve these problems," says Peter Winter, a co-director of Project Ex Vivo who's a principal investigator at the Broad Institute, an independent, nonprofit research institution with close ties to MIT and Harvard University. "But the diversity of cell states in those datasets fundamentally shapes what kinds of insights these models can produce."

The next step is to define cell states clearly and validate them across cancers, with the goal of giving doctors better information to help guide treatment decisions.

"There is a world where five years from now this space looks much different than it does now, which I think is highly, highly encouraging," Crawford says. "The reality of this is not that far away."

Lead image: Photo by Gregory Winter

Susanna Ray writes about AI and technology, with stories that show its real-world impact and examine how innovation is reshaping work, business and society. She previously reported for Bloomberg News and other major international news organizations in the U.S. and abroad, covering beats ranging from politics and government to business and aviation. Follow her work on Microsoft Source.