Acute myeloid leukemia (AML) is a cancer of the blood and bone marrow in which immature white blood cells multiply out of control. While chemotherapy can make leukemic cells appear to vanish, a dangerous remnant often survives at levels too low to detect by standard microscopy.
This hidden residual disease - called minimal residual disease (MRD) - is a major reason why AML relapses even after a patient achieves apparent remission. Measuring MRD is therefore critical for predicting who will relapse and for guiding treatment decisions.
Existing MRD methods fall into two categories: PCR-based tests that detect specific genetic mutations, and flow cytometry that identifies abnormal cell surface proteins. PCR only works for the roughly 20-35% of patients who have known gene mutations; flow cytometry is broader but has limited sensitivity and does not cover all patients.
This paper set out to solve a fundamental problem: identifying a set of cell surface markers that could enable universal MRD monitoring - meaning it would work for essentially every AML patient, not just those with specific genetic subtypes.
The researchers began by performing genome-wide gene expression analysis - a technique that measures the activity level of every gene in a cell - comparing 157 AML patient samples against normal bone marrow progenitor cells from healthy donors.
This comparison revealed hundreds of genes that were switched on or off abnormally in leukemia cells. From these, the team focused on genes meeting strict criteria: aberrantly expressed in at least 33% of AML cases, changed by at least 5-fold relative to normal cells, and detectable using commercially available antibodies.
They then used flow cytometry - a technology that identifies cells by shining laser light on them after tagging them with fluorescent antibodies - to confirm that these gene-level differences also appeared at the protein level on the cell surface in a large set of 191 AML and 63 normal bone marrow samples.
The study enrolled a total of 240 AML patients and validated selected markers across both children and adults, ensuring results would be broadly applicable regardless of patient age or AML subtype.
After rigorous testing, the researchers validated 22 cell surface markers that are abnormally expressed in AML. These include proteins such as CD9, CD18, CD25, CD44, CD47, CD52, CD54, CD64, CD96, CD123, CD200, CD366, and CD371, among others.
These markers were differentially expressed (either too high or too low compared to normal) in 15% to 57% of AML cases individually. However, when used together as a panel, they covered essentially all patients - providing at least one usable marker in every case studied.
Critically, these markers remained stable even during chemotherapy treatment. In 97% of tests, marker expression levels stayed abnormal throughout therapy, meaning they would not generate false-negative results due to treatment effects masking the leukemia signal.
The markers were also found on phenotypically immature AML cells - the CD34+CD38- cell population thought to contain leukemia stem cells. This is especially important because these primitive cells are believed to be the source of relapse and are often missed by conventional markers.
The new markers were tested on 208 bone marrow and blood samples collected from 52 AML patients during chemotherapy. The results showed an excellent correlation with existing standard MRD methods, with a Spearman correlation coefficient of 0.98 - nearly perfect agreement.
More importantly, in 3 samples where standard methods showed no detectable leukemia, the new markers revealed residual disease at levels of 0.10%, 0.19%, and 0.28%. These are cases where the cancer would have been declared absent but was actually still present.
The research team also showed that different AML subtypes tend to overexpress different markers. For example, AML cases with the RUNX1-RUNX1T1 fusion commonly overexpress CD52, CD96, CD200, and CD366, while cases with FLT3 mutations more often overexpress CD25. This suggests the panel can be tuned to individual patient profiles.
The markers also remained detectable at relapse, with 86.9% of tests showing the same aberrant marker present at both diagnosis and relapse. This stability makes them reliable tools not just for initial monitoring but also for tracking disease after treatment failure.
A major innovation in this study was the application of t-SNE (t-Distributed Stochastic Neighbor Embedding), a machine learning algorithm that compresses complex multi-dimensional data into a simple two-dimensional map that humans can visually interpret.
When the new markers were combined with t-SNE analysis, leukemia cells were clearly visible as a distinct cluster even when mixed with normal bone marrow cells. The algorithm was able to detect as few as 1 leukemia cell among 100,000 normal cells - a sensitivity far beyond most standard clinical methods.
In direct comparisons using patient samples, the standard markers caused leukemia cells and normal cells to overlap on the t-SNE map, making interpretation difficult. The new markers produced clearly separated clusters, making it straightforward to identify even tiny amounts of residual disease.
This combination of novel markers with computational analysis addresses one of the long-standing weaknesses of flow cytometry - the need for expert human interpretation - by making the distinction between leukemia and normal cells visually obvious even at very low disease levels.
When the 22 new markers were applied to 129 consecutive AML patients at diagnosis, they provided a detectable aberrant immunophenotype - a distinctive cellular identity profile - in all 129 cases. Standard markers alone could not identify an aberrant profile in 14 of those cases (10.8%).
Perhaps even more striking was the improvement in sensitivity. With standard methods, MRD sensitivity was limited to 0.1% (1 in 1,000 cells) in 40% of patients and could only reach 0.001% (1 in 100,000) in 7.8% of patients. With the new markers, all 129 patients had sensitivity of 0.01% or better, and 40% could be monitored at 0.001%.
Notably, just 5 markers from the panel - CD54, CD18, CD96, CD97, and CD99 - were sufficient to monitor MRD in 94.6% of patients. This suggests that a relatively simple, standardized panel could be deployed across clinical centers without requiring all 22 markers in every patient.
In a cohort of 37 pediatric AML patients, MRD negativity after the first course of chemotherapy was strongly associated with better survival outcomes (P = 0.010 by log-rank test), confirming that more sensitive MRD detection has real clinical significance for predicting who will do well.
Knowing whether residual leukemia cells remain after chemotherapy is essential for deciding what to do next. Patients with persistent MRD may need more intensive therapy or a stem cell transplant, while those who achieve MRD negativity may be spared harsh additional treatments.
The markers in this study remained stable during treatment and at relapse, which means they can be used across multiple time points in a patient's care without the test losing reliability. This enables a continuous monitoring approach throughout the entire treatment course.
Because the markers are now applicable to virtually all AML patients - not just those with specific mutations - these tools could enable wider adoption of MRD-guided treatment protocols that adjust therapy intensity based on real-time disease response measurements.
Additionally, several of the identified markers, including CD123, CD47, and CD99, are already being explored as immunotherapy targets. The dual utility of these proteins - as both MRD tracking tools and potential drug targets - makes this panel especially valuable for future therapeutic development.
This study represents a significant advance by moving AML MRD monitoring from an empirical, case-by-case approach to a systematic, genome-guided strategy. Rather than using whatever markers happened to look different in earlier clinical experience, the researchers started from an unbiased, genome-wide search for differences between leukemia and normal cells.
The result is a defined panel of markers that is broadly applicable, sensitive, and stable - addressing the three major shortcomings of current clinical MRD assays. Combined with machine learning visualization, these markers also reduce the need for expert interpretation, potentially making high-quality MRD testing more widely accessible.
The authors acknowledge that implementing larger antibody panels increases cost, but argue that the gains in applicability, sensitivity, and reliability are likely to outweigh the expense - particularly when better MRD monitoring translates to more appropriate treatment selection and better patient outcomes.
Future directions include using these gene expression data as a reference for mass spectrometry studies of the AML cell surface, and further exploration of the therapeutic potential of the identified markers as drug targets in AML treatment.