In an interview, Jamie L. Koprivnikar, MD, discussed the importance of a machine learning approach to genomic testing in AML.
The bone marrow immune microenvironment is thought to play a major role in the development of acute myeloid leukemia (AML) and myelodysplastic syndrome (MDS). Researchers at Genomic Testing Cooperative and Hackensack University Medical Center have identified key differences between the microenvironment of healthy patients, those with AML, and those with MDS, presenting their findings at the 2023 American Society of Clinical Oncology Annual Meeting.1
Investigators performed next-generation sequencing (NGS) on bone marrow samples from 626 patients with AML, 564 patients with MDS, and 1449 individuals whose bone marrow had no mutations or only low-level clonal hematopoiesis of indeterminate potential (CHIP) mutations. They studied the expression of 42 immune biomarkers, and using a machine learning algorithm, found 15 genes whose expression distinguished MDS from normal samples, and 10 whose expression distinguished AML from normal.
Eight of the 10 biomarkers in AML were shared in MDS; CD28 and IL7R were needed to classify patients as having AML. Most of these biomarkers had the highest level of expression in normal samples, less expression in MDS, and even lower levels of expression in AML.
Since the bone marrow microenvironment is involved in the growth of these malignancies and may influence the outcomes of treatment, identifying the presence of these biomarkers is crucial.
“This is just a further piece of the puzzle,” said Jamie L. Koprivnikar, MD, in an interview with Targeted OncologyTM. “This may suggest potential future targets for therapies that are aimed at affecting the bone marrow microenvironment and modifying it to make it less hospitable to pathological blast cells.”
In the interview, Koprivnikar, a leukemia specialist at John Theurer Cancer Center at Hackensack University Medical Center, discussed the importance of this machine learning approach to genomic testing in AML.
Targeted Oncology: What is the purpose of your presentation on the immune microenvironment of AML/MDS?
Koprivnikar: The presentation or the abstract is defining the immune microenvironment in both MDS and AML using machine learning. This was a presentation that looked at a set of patients who had AML, a set of patients who had MDS, and then compared some of the RNA levels of these patients with normal controls or controls who had CHIP to identify some of the important differences that influence the bone marrow microenvironment and contribute to the pathologic bone marrow microenvironment that we know exists in AML and MDS.
What was the rationale behind this research?
We know that the bone marrow microenvironment plays an important role in leukemogenesis, and possibly in chemotherapy resistance to AML. We think that this may represent a potential therapeutic target. In trying to better understand this microenvironment, we're hoping to be able to come up with better targeted therapies to intervene and improve outcomes for these patients.
What treatment options currently exist for these patients?
We are lucky that we've had a number of developments in therapy for our patients with AML and MDS. Currently for patients with AML who are older or unfit, the standard-of-care therapy is venetoclax [Venclexta] in combination with a hypomethylating agent. For patients who can tolerate intense chemotherapy, we still sometimes use the standard 7 + 3 regimen [cytarabine plus anthracycline], which has been in use for 40 years or more. But there have been some variations introduced such as the advent of targeted therapies, like midostaurin [Rydapt]. Then, there is a novel formulation of the agents that we use in 7 + 3, liposomal daunorubicin, and cytarabine [CPX-351; Vyxeos]. These are the current agents that we use for frontline therapy.
There are a number of targeted agents that can be used in second-line or later therapies as well. There are certainly a number of agents that we're currently developing.
Could you go into some of the methods and design of this research?
It was looking at the RNA that was extracted from fresh bone marrow aspirate samples. There were 626 pathologic samples of patients who had AML and 564 pathologic samples from patients who had MDS. Then, there were 1449 individuals who had a normal or fairly normal bone marrow. They were permitted to have low level mutations that were felt to be consistent with CHIP. They looked at RNA levels of a number of different biomarkers. These were quantified using NGS. Then using machine learning, they were able to select the genes that best distinguish between the 2 groups of patients. As a result of this machine learning algorithm, we were able to identify 15 key genes that helped to distinguish an individual with a pathologic bone marrow affected by AML or MDS from the healthy controls.
What were the main results that are important to discuss?
The 15 genes that were felt to differentiate the pathologic marrow from the normal marrow included a number of genes. CXCR4, CD58, CD36, CD19, PAX5, IL8, CD44, CD79A, and CD74 were just some of the pertinent genes that were found to be differentiating factors.
What would you say are the key takeaways from this research?
This is just a further piece of the puzzle. This may suggest potential future targets for therapies that are aimed at affecting the bone marrow microenvironment and modifying it to make it less hospitable to pathological blast cells.
What are the next steps for research in this space?
It's important to continue to understand more about the bone marrow microenvironment, and to understand the ways in which we may be able to modify it to make it less hospitable and less friendly to leukemia cells.
What should community oncologists take away from this research?
This research further underscores the importance of NGS testing in the management of patients with myeloid malignancies. While the mutations identified as part of our research are not yet ready to move into routine clinical practice, NGS is increasingly important in both risk stratification and treatment decisions for patients with myeloid malignancies. We're seeing that the lines between what we once classified as MDS and AML are beginning to blur and that a patient's mutational profile is increasingly important in determining the true nature of their underlying disease. For example, a patient with a TP53 mutation and 12% blasts is going to behave much more like a patient with AML although technically this individual is not meeting the somewhat arbitrary cut off having greater than 20% blasts that has previously been AML defining.
1. Albitar M, Zhang H, Koprivnikar JL, et al. Defining the immune microenvironment in myelodysplastic syndrome and acute myeloid leukemia using machine learning. J Clin Oncol. 2023;41(suppl_16):7060. doi:10.1200/JCO.2023.41.16_suppl.7060