In 2013, the healthcare company Illumina began offering a non-invasive prenatal test (NIPT) to pregnant women. The test aimed to find tiny DNA fragments in the women’s blood that might indicate chromosomal abnormalities in the fetuses they were carrying—abnormalities that could signal the presence of genetic disorders such as Down syndrome.
Dr. Meredith Halks-Miller, pathologist and laboratory director of Illumina’s NIPT clinical lab at the time, noticed odd findings in some of the blood samples of the pregnant women. They didn’t show evidence of the chromosomal disorders the test was designed to find, but they indicated chromosomal abnormalities that raised suspicions.
“I was pretty sure that these expectant mothers had cancer and didn’t know it,” Halks-Miller recalls. “I encouraged the clinical consulting staff to do more clinical follow-up for these patients even though they appeared to be healthy.”
Halks-Miller shared the information with Illumina’s chief medical officer at the time, Rick Klausner, a former director of the National Cancer Institute, who told her, “I don’t know of anything else that changes the genome the way you’re showing me here.”
Sure enough, 10 women with these DNA abnormalities were eventually diagnosed with cancer.
That was the “Eureka!” moment that led to Galleri, a new multi-cancer early detection test from the healthcare company GRAIL, which was spun off from Illumina, with Klausner as a cofounder, in 2016 (Klausner also serves on GRAIL’s board of directors). GRAIL hopes that Galleri, which is expected to become commercially available before summer, could revolutionize cancer screening, potentially leading to major reductions in mortality and expense.
“THE BEAUTY OF MACHINE LEARNING”
Although the science behind Galleri is sophisticated, its underlying premise is straightforward. Up until now, in the U.S., there have been early-screening tests for only five types of cancer: PSA tests for prostate cancer; colonoscopy for colorectal cancer; mammography for breast cancer; pap smears for cervical cancer; and low-dose CT scanning for people at high risk for lung cancer. But dozens of other cancers—the ones for which no screening tests are available—are often detected only after they’ve begun to spread, making treatment more difficult. Galleri can help address that disparity via a single blood test that can detect multiple types of cancer and indicate where they’re located in the body, with high accuracy, which has the potential to greatly expand the number of cancer cases that are caught in their early, more treatable stages. An earlier version of Galleri demonstrated the ability to detect more than 50 types of cancers, as defined by the American Joint Committee on Cancer Staging Manual.
Scientists have long known that cancer cells shed DNA fragments into the bloodstream, but until recently were unable to discern those signals from background noise. Galleri uses machine learning—essentially, algorithms—to filter out that noise. “We’ve amassed some of the largest clinical data sets that link clinical data with genomic data,” says GRAIL chief medical officer and head of external affairs, Dr. Josh Ofman. “We use those to train our machine-learning algorithm, which can then look at these signals circulating in the blood—little fragments of DNA—and classify them as cancer or not cancer. And if cancer is detected, it will give you the likely origin of the cancer—pancreas or liver and so on—so the doctor has a very clear direction for where to look.”
This use of algorithms means Galleri may improve over time and be able to detect additional types of cancer. “As more and more people use the test, the data we get will improve our ability to interpret the test for the next people,” Klausner says. “So Galleri is probably going to be much better in the future. That’s the beauty of machine learning.”
The results of all this could reconfigure the cancer landscape. GRAIL’s models estimate that Galleri, when added to diagnosis by usual care, has the potential for earlier-stage detection of nearly 70% of cancers that result in death within five years, which translates to the potential to avert 39% of deaths that would otherwise be expected. Financial savings, due to averting the need for costlier late-stage treatments, could also be significant.
Not bad for a test that was essentially discovered by accident. Galleri isn’t the first medical advance to come about in this manner (other accidental breakthroughs range from penicillin to Viagra), but it’s a good reminder that innovation doesn’t always involve a linear pursuit of a preconceived goal. Sometimes it’s more about being open to unexpected revelations or viewing something from a different perspective.
“A PATH TO EARLY DETECTION”
In the case of Galleri, the original context of testing for fetal abnormalities offered a key advantage. “What really blew me away wasn’t so much that we detected cancer,” Klausner says. “It’s that there were 125,000 women who had tests that didn’t have these signals in their blood, because the biggest challenge of looking for signals for cancer is signal-to-noise ratio and false positives. But we knew there couldn’t have been very many false positives, because there were 125,000 negatives.”
Ironically, Klausner himself was a longtime skeptic of using biomarkers for early cancer detection. “At one time I was very interested in it, but I had come to just view it as an area of rabbit holes, because of those signal-to-noise issues,” he says. “In fact, I had told Illumina’s CEO at the time, Jay Flatley, ‘If someone comes to you with an early-detection idea, run.’ And then, like, two months later, after those fetal blood tests, I came to him and said, ‘I think we have a path to the early detection of cancer.’ And that was the origin of GRAIL.”
As GRAIL prepares for Galleri’s launch, Klausner is trying to balance his enthusiasm with a bit of that early skepticism. “GRAIL has been set up from day one to be its own harshest critic, because we don’t want to emphasize hype over promise,” he says. “We’ve done the largest, most rigorous studies. We publish them. We’ve put skeptics on our Scientific Advisory Board. But I feel very comfortable saying we’ve begun to break the back on this holy grail of cancer, which is universal early detection.”
An earlier version of this story did not specifically mention Dr. Halks-Miller. GRAIL regrets the omission, and we are grateful for Dr. Halks-Miller and all of our employees who have helped—and continue to help—further our mission to detect cancer early, when it can be cured. We are fortunate to have amassed an incredibly talented group of scientists, engineers, and clinicians who have helped develop our technology during the past five-plus years. We will do our best to ensure they remain part of our story.