NEW ORLEANS – Children with acute lymphoblastic leukemia (ALL) whose cells express an epigenomic biomarker rooted in DNA methylation patterns may be less likely to respond to CD19-directed CAR T-cell therapy, according to research presented at the American Association of Cancer Research's annual meeting on Tuesday.
Many relapsed or refractory pediatric leukemia patients initially respond robustly to this type of autologous CAR T-cell therapy, yielding response rates as high as 90 percent. Novartis' Kymriah (tisagenlecleucel) has had regulatory approval for this patient population for five years now.
But for the small subset of patients who do not respond to the therapy, overall survival rates are only about 30 percent.
As such, there is a pressing need to identify, and potentially mitigate, the mechanisms driving patients' resistance to CAR T-cell therapy. Yet very little is known about the underlying biology of CAR T-cell resistance, explained Katherine Masih, a National Institutes of Health-Cambridge scholar in the National Cancer Institute's Genetics Branch. Research so far into CAR T-cell resistance mechanisms has shown that patients who respond at first but eventually relapse, often do so due to CD19 antigen loss on the surface of cancer cells.
The patients whose cancers do not respond at all, on the other hand, have intact CD19 expression, meaning that their treatment resistance doesn't have to do with the antigen loss. According to Masih, the few reasons experts have offered as an explanation for this initial therapy resistance — including dysfunctional T cells in the treatment product or decreased death receptor expression — aren't consistent across patients and show that "we still don't really understand primary non-response, and causes may be heterogenous."
To determine exactly why these patients don't respond, Masih and her research colleagues collaborated with the Seattle Children's Hospital and performed a retrospective, in-depth multiomic analysis of patient samples from the Phase I/II PLAT-02 clinical trial. In that trial, researchers at Seattle Children's evaluated autologous CAR T-cell therapy in kids with relapsed or refractory leukemias and lymphomas.
The hypothesis that Masih and her team had going into the multiomic analysis was that "primary non-responders to CD19 CAR T cells have distinct leukemia compared to patients who do respond, and these differences can be detected prior to therapy."
The researchers analyzed pre-treatment bone marrow samples from fourteen patients enrolled in the PLAT-02 trial, seven of whom went on to respond to therapy and seven who were treatment resistant from the onset.
"We were able to take the pre-treatment bone marrow aspirates from the [PLAT-02] eligibility workup and perform a gauntlet of multiomic analyses to understand the underlying biology of this disease," Masih said.
'Gauntlet' of analyses
The experiments that Masih and colleagues performed were extensive, including whole-exome sequencing, bulk RNA-sequencing, long-read sequencing of the CD19 locus, array-based methylation testing, ATAC sequencing, single-cell RNA sequencing, and mass cytometry (CyTOF).
The team found 238 regions of increased DNA methylation in patients who did not respond to treatment, and further determined that the pre-treatment methylation patterns were those known to be turned off by PRC2 repression in stem cells. They then did a gene set enrichment analysis of their ATAC sequencing data and saw increased accessibility of chromatin at regions known to be associated with proliferation and cell cycling in stem cells.
These findings led Masih and her team to wonder whether their findings meant that the phenotypes of the leukemia cells were altogether different in non-responders than in responders. To answer this, they looked at genome-specific regions associated with different cell types of hematopoiesis.
Here, they saw a significant increase in chromatin accessibility at regions associated with hematopoietic stem cells and myeloid progenitors, suggesting that the leukemias shared features of stem cell and myeloid precursor cells.
"The major finding of our study was that the way we're thinking of this disease as being very lineage-restricted is not correct," she said.
With all of this information in hand, the researchers were able to describe a biomarker for CAR T-cell resistance, which they dubbed "stem cell epigenome with multi-lineage potential," or SCE-MLP. The biomarker comprises 238 hypermethylated DNA regions and could, in theory, be used to identify patients unlikely to respond to CAR T-cell therapy before they undergo treatment.
"What's most important about this is that we can detect it prior to therapy," Masih said. "If we can reliably identify responders, perhaps through screening of SCE-MLP, we can prioritize less toxic, targeted therapies for our patients, and overall improve outcomes."
SCE-MLP biomarker validation
Acknowledging the fact that this study was conducted in a very small group of patients, Masih said that, going forward, "we would love to see this validated in a larger cohort with more cases of primary non-responders." Additionally, she said the team hopes this SCE-MLP biomarker can be evaluated for sensitivity and specificity in a prospective trial setting.
Further down the line, there is a chance that the biological information gleaned from this study could lead to new treatment methods for these resistant patients, potentially via combinations of epigenetic drugs that disrupt the stem cell epigenome, or even multi-targeting CAR T cells, Masih said. She explained that her team also found some subpopulations of patients whose cells expressed different surface markers such as CD19, CD33, and CD34, which could serve as targets for combination treatment strategies. There are also chemotherapy drugs that directly target DNA methylation, "so perhaps these could be used to pre-treat patients," she mused.
A future in which this SCE-MLP biomarker is used in routine clinical practice isn't out of the question, either. "DNA methylation is very stable and easy to get from patients [so] it would be easy to screen for these 238 loci," Masih said, adding that she and her team are not developing a scalable test for this biomarker. Still, her study certainly provides a starting point for another team or interested party to develop a commercial test.
"This is an expensive, complicated therapy that isn't without risk for kids," she said. "We need to do better at predicting response."