NEW YORK – Oncologists hoping for more precise and individualized care for patients with brain cancers and metastases are continuing to research liquid biopsy technologies that could help them more easily and consistently analyze tumor genomics and monitor patients in ways that have so far been either difficult or impossible.
Presentations at the American Society of Clinical Oncology's virtual annual meeting this week featured new data related to this effort, including progress in defining the variables that affect why only some patients with brain metastases can be successfully genotyped using liquid biopsy, and advances in defining the potential utility of blood-based cell-free DNA analysis for patient monitoring and prognosticating.
Brain cancers have emerged as somewhat of a conundrum for the liquid biopsy field, providing both a clear opportunity — because of the difficulty and danger in obtaining tissue samples from the brain relative to other organs — but also a significant challenge, as research has shown that brain tumors seem to shed DNA into the blood at much lower rates and quantities than other cancers.
Speaking during a clinical science symposium for the ASCO virtual annual meeting on Friday, Dana Farber physician-scientist Keith Ligon said that one of the most obvious and most promising use cases for liquid biopsy in brain tumors is for assessing patient prognosis and monitoring response to treatment.
But a major hurdle has been figuring out which signals — whether DNA mutations, other cancer-specific signals, or even less-specific measures like overall circulating DNA levels — might prove most useful or valuable.
In one session presentation, researchers from the University of Pennsylvania, led by Stephen Bagley, shared updated data from their efforts to explore how cell-free DNA levels might provide oncologists with prognostic information and response monitoring that could help them tune treatments more precisely and effectively.
The group published preliminary findings from this effort last year, reporting that glioblastoma patients with a higher concentration of cell-free DNA in their blood had shorter progression-free survival than those with less cfDNA, and that spikes in circulating DNA seemed to correlate with or even predict disease progression.
According to Bagley, although GBM is known to be a heterogeneous cancer, the only established biomarker that can help oncologists predict how individual patients will respond to treatment is MGMT promoter methylation.
Approximately 40 percent of glioblastoma patients have a methylated MGMT promoter tumor, and this subgroup has been shown to have a better response to standard-of-care radiation and temozolomide chemotherapy.
MGMT testing suffers from several limitations though, Bagley said, including a population of assays that differ considerably across institutions and lack standardized cutoffs. The fact that these tests require tumor tissue also poses a significant problem.
"A noninvasive, and more reliable, prognostic biomarker for glioblastoma would allow for improved personalization of care, including improved stratification for clinical trials, as well as more tailored advanced care planning for individual patients," he argued.
According to Bagley, the choice to look at cell-free DNA concentration rather than circulating tumor mutations evolved from growing data that the blood-brain barrier prevents detectable levels of tumor DNA from making it into the circulation of the majority of GBM patients.
In their ASCO presentation, Bagley and his colleagues shared updated results from a validation of their cfDNA approach in an independent set of 61 patients. Based on the data from their initial 41-patient cohort, they derived a cutoff cfDNA concentration that they said provided the best discrimination of GBM patients who exceeded the disease's median progression-free survival time versus those who fell short of it.
When the researchers then analyzed plasma samples from the new validation cohort using this cutoff point, they found that it did predict prognosis. In a univariate analysis, the difference wasn't significant for either PFS or overall survival. Median PFS was 7.3 months in patients with low cfDNA, versus 4.4 months in patients with high cfDNA, and OS was 13.4 months in patients with low cfDNA, versus a median of 12 months in patients with high cfDNA.
But when the team did a multivariate analysis adjusting for other relevant prognostic factors, they could see that cfDNA was independently associated with the likelihood of surviving without progression for at least seven months, with an odds ratio of 4.3 and a p-value of 0.03.
In an exploratory analysis including all 102 patients from both cohorts, the researchers then looked at a combined predictor, incorporating both plasma cfDNA levels and MGMT promoter methylation.
"Patients with an MGMT methylated tumor and a low cfDNA did by far the best, with a median progression-free survival of 13.6 months. All other patients, regardless of their MGMT methylation status, did similarly poorly, with a median progression-free survival of only 4 to 5 months," Bagley said.
There were similar patterns for overall survival, he added. "Breaking patients down by MGMT methylation status and high versus low cfDNA, there is, again, a pretty clear difference in median overall survival between the four groups."
"Patients with a methylated MGMT promoter and low cfDNA did the best… Conversely, patients with an unmethylated MGMT promoter and high cfDNA did uniformly poorly," he said.
Finally, the team also explored whether measuring cfDNA levels at other time points might offer oncologists even more actionable information.
"If baseline plasma cfDNA concentration has prognostic value in newly diagnosed glioblastoma, perhaps plasma cfDNA concentration following completion of chemoradiotherapy is an even more accurate predictor of outcomes," Bagley suggested.
This exploratory analysis included only 24 patients who had enough follow-up timepoints at the time of the group's data analysis. Investigators compared whether having high or low cfDNA at the post-treatment timepoint correlated to overall survival at 12 months from diagnosis.
Using a cutoff point of approximately 17 ng/ml, the team found that they could predict 12-month survival with an area under the curve of 0.87.
Discussing the UPenn team's findings, Ligon said that one important next step will be to try to nail down what the biological basis is for non-tumor derived cfDNA concentration correlating with tumor biology.
In other words, researchers will need to try to explain the mechanisms whereby a poorer prognosis brain tumor leads to increases in all-source cell-free DNA in a patient's blood, as well as the factors that affect this process.
It would also be great to know how cfDNA concentration compares to other indirect serum biomarkers, Ligan said.
Looking at overall cfDNA rather than tumor-derived mutated DNA is just one way to get around the issue of brain tumors failing to passage their mutated DNA molecules through the blood brain barrier. Another is forgoing blood altogether and taking liquid biopsy samples from the cerebral spinal fluid.
Although this has practical and safety downsides relative to blood-based testing, Ligan said that the weight of evidence now shows that that ctDNA is much easier to detect in CSF than blood.
Studies of circulating tumor DNA in CSF have found similar patterns as Bagley and colleagues, he added, with higher levels corresponding to worse survival. And efforts are ongoing to explore serial monitoring, whereby changes in ctDNA could potentially help oncologists get ahead of treatment resistance or tumor progression.
If a case is going to be made for utility, Ligan argued that researchers will have to find ways to improve sensitivity so that methods work for most, if not all patients. Right now, studies have found that only about 50 percent of patients have detectable ctDNA in their CSF.
And the field will have to move on to prospective clinical trials, whether for CSF-based methods, or for things like the Bagley team's blood cfDNA technique.
"We need more studies and more trials, and probably more industry funding to really evaluate these complex situations," Ligan said. "This is something that has really been accelerating in other cancers and hopefully we can [see] some of that going forward."
As teams push forward with validation of both blood- and CSF-based methods, a related hurdle has been the issue Ligan highlighted of trying to identify (and hopefully find ways to manipulate) the mechanisms that lead to a liquid biopsy signal being detectable in some individuals but not others.
There have been some recent breakthroughs in this, for example recent data published by the same UPenn team, which showed that imaging methods can help predict which glioblastoma patients can be successfully genotyped via the blood, and which cannot.
In another presentation of the ASCO science symposium, researchers from a group of Chinese institutions reported on their own study that explored a similar question in the context of metastatic lesions that have spread to the brain from non-brain cancers. They specifically discussed the identification of phenotypic characteristics that play a role in whether ctDNA can be detected, in this case in patients' CSF.
Because obtaining CSF samples is not without risk to the patient, it would be beneficial clinically, to be able to predict which patients an oncologist or researcher should even try to sample, said Sun Yat-Sen University researcher Meichen Li, who described the results during the session.
"Precisely choosing suitable patients is needed to maximize detection benefits," she added.
In their study, Li and her colleagues sequenced 425 cancer-relevant genes in CSF samples and matched extracranial tissue or blood samples obtained from 67 lung cancer patients with brain metastases.
They then measured the impact of clinical factors — including age, gender, tumor size, number of lesions, and distance of lesions to the ventricle — on whether ctDNA could be detected in CSF.
The researchers detected a ctDNA somatic alteration in 39 of the 67 patients in the cohort. And their analysis revealed a significant association between CSF ctDNA detection and a variety of features, including intracranial lesion size, shortest distance between the largest lesion and the ventricle, and the shortest distance between all intracranial lesion and the ventricle. There was also a trend of higher detection rate in patients with central nervous system symptoms.
When the team conducted additional statistical analyses, they found that the best prediction power could be had using a combination of lesion size and largest lesion-ventricular distance, showing an area under the curve of 0.76. Using this model, they could improve the detection of CSF ctDNA from 58 percent to 83 percent, the group wrote.
Although not perfect, the authors suggested the method shows promise as a tool to predict the probability of CSF ctDNA detection, which could help facilitate clinical decisions and avoid useless attempts to monitor tumor evolution in the brain in cases where it isn't possible.