Skip to content

STAT – New study points to best way to assess which cancer patients immunotherapy will help

STAT – New study points to best way to assess which cancer patients immunotherapy will help

Experiments involving just a few patients have hinted at it, and research on one type of tumor at a time has supported it, but a large study has now delivered the strongest evidence yet about how to identify cancer patients who are likely to benefit from a particular form of immunotherapy: count.


Specifically, count how many mutations their tumor cells have. The higher this “tumor mutational burden,” concludes a study published on Monday in Nature Genetics, the likelier a patient is to go into remission, and possibly be cured, by checkpoint inhibitor drugs such as Bristol-Myers Squibb’s nivolumab (Opdivo) and Merck’s pembrolizumab (Keytruda).


In the study’s main surprise, what counts as a high mutational burden differs markedly from one type of cancer to another: As few as six mutated genes per million DNA bases means a breast cancer is likely to respond to immunotherapy, but a colorectal cancer needs 52 to be vulnerable.


If the results hold up, they could spare many patients from suffering the severe side effects of these drugs with nothing to show for it. Checkpoint inhibitors remove the brakes that tumors put on immune cells, sometimes causing rashes, fever, and inflammation of the lungs and other organs — in rare cases severe enough to be fatal. The majority of patients don’t benefit from the drugs, however; their immune system, even when the brakes come off, can’t defeat their cancer. Since checkpoint drugs cost well into the six figures, there is also a financial incentive to identify whom the drugs can really help.


“Predictive biomarkers are really important for identifying with some certainty patients who are likely to get long-term benefits [from checkpoint inhibitors] and those who are less likely to,” said Dr. Padmanee Sharma of MD Anderson Cancer Center and an expert on immune-oncology.


Companies are scrambling to identify the most accurate biomarkers, but first they have to cut through what is now a Tower of Biological Babel. More than two dozen predictive biomarkers, from levels of PD-L1 (the molecule that many checkpoint drugs target) to inflammation to the gut microbiome, are vying for medical and commercial supremacy in what is projected to be a multibillion-dollar market by 2022.


“It’s confusing and it’s frustrating,” said Dr. F. Stephen Hodi, director of immune-oncology at Dana-Farber Cancer Institute. “The problem is, no biomarker [for checkpoint drugs] is perfect, but finding one that works is the holy grail: We want something for the patient sitting in front of us.”


It remains to be seen whether tumor mutational burden, as in the new study, is that holy grail, but there are signs it may be emerging from the very crowded pack.


The early leaders for checkpoint biomarkers, the proteins PD-1 and PD-L1 that the checkpoint drugs target, have “lost some of their luster,” the trade journal ClinicalOmics recently warned. It seemed obvious: Of course cancers with high levels or PD-1 or PD-L1 would be most vulnerable to checkpoint drugs targeting them. In fact, however, some patients with low levels have responded well and some patients with high levels haven’t benefitted at all.


Attention has therefore turned to tumor mutational burden (TMB). It got a big vote of confidence last year when 28 drug companies, diagnostics makers, academic medical centers, the National Cancer Institute, and the Food and Drug Administration joined in an effort to propose standards for TMB testing and how best to use it in patient care, including such tricky questions as how to measure TMB and what the cutoff should be between likely and unlikely to benefit. The project, organized by the nonprofit advocacy group Friends of Cancer Research, expects to finalize its proposals by this summer.


As a predictive biomarker for checkpoint drugs, “tumor mutational burden is building up quite a head of steam,” said Dr. Tim Chan of Memorial Sloan Kettering Cancer Center, lead author of the new study. He and his 56 co-authors have, among them, financial ties to nearly 250 pharmaceutical, biotech, and other companies. Chan is among the inventors of a patent for TMB testing, and Sloan Kettering has licensed its TMB discoveries to the genetic testing startup PGDx.


“Two and a half years ago, TMB was mostly a theory” in terms of predicting patient response to checkpoint drugs, said David Fabrizio, who leads the cancer immunotherapy unit at genetic testing company Foundation Medicine, which was acquired by Roche last June for $2.4 billion. “But we’re getting closer to proving it.”


For the new TMB study, Chan and colleagues used Sloan Kettering’s tumor-DNA-sequencing test “IMPACT” to count the mutations in 468 cancer-associated genes in the tumors of 1,662 patients. They calculated whether that mutational load was correlated with length of survival after treatment with any of seven checkpoint inhibitors: atezolizumab, Genentech’s Tecentriq; avelumab, sold by Pfizer and EMD Serono as Bavencio; durvalumab, AstraZeneca’s Imfinzi; ipilimumab, Bristol-Myers Squibb’s Yervoy; nivolumab, BMS’s Opdivo; pembrolizumab, Merck’s Keytruda; or AstraZeneca’s in-development tremelimumab.


After following the patients for up to 80 months, they found that the more mutations a tumor had, the longer a patient survived with a checkpoint drug. In general, someone with a mutation load in the top 20 percent was 40 percent more likely to be alive than those with fewer mutations. More mutations also predicted a longer time without a tumor growing or spreading.


But the number of mutations needed to cross that 20 percent threshold varied markedly by cancer type. It was 52 in colorectal cancer and 31 in melanoma, but only six in breast cancer and kidney cancer.


That undermines the simplistic idea that lots of mutations spell a hopeful prognosis. That’s still true within a cancer type: A non-small cell lung cancer with a high mutational burden responds better than one with fewer mutations, with median survivals of 9.7 months vs. 5.8 months, according to one typical study. But “high” is different for each cancer, Chan said.


“Tumor mutation burden is an important predictor across many cancer types,” he said. “But the cutoff will be different depending on what kind of cancer you have.”


That variability might relate to things like the environment in and around a tumor, including whether T cells even reach it. The only cancer where mutational load didn’t predict response to checkpoint drugs was glioma, the almost-always incurable brain cancer. That is probably because immune cells can’t get into the brain.


Other experts in immuno-oncology caution against the very idea of a cutoff. In the new study, patients whose tumors had a mutational burden in the bottom 80 percent didn’t have zero chance of benefitting, just a much lower one. “We want to be really careful and not deny treatment to patients who have a chance,” said MD Anderson’s Sharma.


There is a very real risk of doctors, relying on a faulty biomarker, deciding not to prescribe a checkpoint drug for patients who actually have a shot at a cure. In a 2018 study called CheckMate-227, patients with advanced non-small cell lung cancer responded just as well to a combination of two checkpoint drugs, nivolumab and ipilimumab, regardless of how highly their tumor cells expressed PD-L1. All that mattered was their tumor mutational burden. It was the first time TMB had clearly bested PD-L1.


PD-L1 has several problems as a biomarker. Levels can vary over time and can be high in one area of a tumor but low in another. And measuring levels of a protein like PD-L1 is trickier than genetic tests.


That doesn’t mean PD-L1 tests are doomed. Agilent Technologies’ Dako PD-L1 IHC 22C3 pharmDx assay won FDA approval last year to identify patients with cervical cancer who are likely to benefit from Keytruda; it already had approval for non-small cell lung and gastric cancers. A PD-L1 test from Roche Diagnostics’ Ventana unit is approved for non-small cell lung cancer and bladder cancer as a way to predict benefit from Tecentriq. Last year, NeoGenomics began selling a PD-L1 test to go along with Keytruda in cervical cancer, adding to approvals for gastric, bladder, and non-small cell lung cancers.


Nevertheless, genetic tests, not tests for proteins like PD-L1, are poised to dominate the market for checkpoint inhibitor biomarkers.


Close to a dozen companies just in the U.S. are developing TMB tests, and, in some cases, already selling them.


Foundation Medicine, which is part of the Friends of Cancer Research project, got FDA approval in late 2017 for its $5,800 FoundationOne CDx (CDx stands for companion diagnostic), which assesses, among other things, tumor mutational burden at 324 genes. Some of those genes make proteins that travel to the surface of tumor cells and stick out like little flags, beckoning immune cells to come and kill the cell.


“TMB is a way to measure the number of potential flags your cancer has,” said Foundation’s Fabrizio. “That makes it a direct link to how likely a patient is to respond to a checkpoint inhibitor.” Foundation sold nearly 100,000 of the tests last year.


There are numerous challenges to perfecting tumor mutational burden as a checkpoint biomarker, including how many genes to test (Sloan Kettering’s 468 or Foundation’s 324) and what mutational burden signals a good prognosis in different cancer types. With scientists only starting to analyze the new TMB study, that’s largely a mystery, but even before its publication, companies were moving in that direction. In Caris Life Sciences’ TMB test (it has run 61,580 so far, president David Spetzler said), “low” and “high” is cancer-type specific: 10 per million DNA bases for non-small cell lung cancer and 17 for everything else.


Another challenge is doing the lab work and analysis right. Last October, Foundation discovered that it was misreporting some low TMB tests (unlikely to benefit from checkpoint drugs) as high (likely to benefit). “The root cause of the issue was quickly identified, addressed, and resolved,” the company said, adding that it had “informed physicians that their reports could include an incorrect test result.”…