How Blood Tests Can Predict Treatment Success in Chronic Myeloid Leukemia
An exploratory biomarker analysis from ENESTnd reveals gene expression signatures that forecast deep molecular response
When Sarah was diagnosed with chronic myeloid leukemia (CML) in 2018, her oncologist explained she had good chances with modern treatments. But after months on targeted therapy, her results weren't improving as hoped. Meanwhile, another patient with seemingly identical CML characteristics achieved such deep remission that he could safely stop treatment altogether. This variability in treatment response—why some CML patients excel on standard therapy while others struggle—has long puzzled hematologists. Now, a groundbreaking approach that analyzes gene expression patterns in blood may finally provide answers, potentially predicting individual patient responses before treatment even begins.
CML originates from a genetic mishap known as the Philadelphia chromosome, where chromosomes 9 and 22 swap genetic material, creating the cancer-driving BCR::ABL fusion gene 1 . This malfunction produces an overactive tyrosine kinase protein that triggers uncontrolled white blood cell production 4 .
The genetic abnormality that defines CML, forming the BCR::ABL fusion gene 1
Revolutionary targeted drugs that disable the BCR::ABL protein, transforming CML into a manageable condition 9
Tracking BCR::ABL transcript levels to measure treatment effectiveness through specific milestones
BCR::ABL level ≤10% within 3-6 months of treatment
BCR::ABL level ≤0.1% by 12 months
Achieving DMR has particular significance—patients who maintain this ultralow leukemia cell level may qualify for Treatment-Free Remission (TFR), allowing them to stop medication while maintaining disease control 9 .
The research analyzed data from the ENESTnd clinical trial, a large phase III study that compared the effectiveness of imatinib versus nilotinib in newly diagnosed chronic-phase CML patients 9 . This robust dataset provided an ideal platform for investigating genetic predictors of treatment response.
Patients Selected
Good Responders Achieving DMR
Poor Responders Achieving DMR
Scientists selected 112 patients from the trial, categorizing them into two distinct groups based on their treatment performance. The dramatic difference between these groups became increasingly evident over time. By the five-year mark, a striking 95% of good responders had achieved DMR, compared to only 17% of poor responders 9 . This divergence confirmed that early response patterns have long-term implications for treatment success.
The research began with pretreatment blood samples collected from patients before they started TKI therapy. This timing was crucial—any genetic signatures identified would truly predict rather than reflect treatment effects. RNA was extracted from these samples and processed for RNA sequencing, a technique that captures a comprehensive snapshot of gene activity 9 .
The research team employed sophisticated bioinformatic pipelines to normalize and analyze the massive genetic datasets. They used the edgeR package to identify meaningful expression patterns among thousands of genes 9 . To determine which cell types were present in each sample, they applied the MCP-counter algorithm, which estimates immune and blood cell populations based on gene expression patterns 9 .
The core predictive modeling used penalized logistic regression with bootstrap resampling. This machine learning approach identified the most informative genes for classifying future patients as good or poor responders while guarding against overfitting to the specific patients in the study 9 .
To ensure their findings weren't unique to their specific patient sample, researchers validated their model using an independent dataset previously generated by Branford et al. 9 . This critical step tested whether the genetic signature maintained its predictive power in different patient populations, strengthening confidence in its potential clinical utility.
The gene expression-based prediction model demonstrated impressive accuracy, achieving an area under the curve (AUC) of 0.76 9 . In diagnostic terms, this represents good predictive ability—substantially better than random chance—suggesting genuine biological signals in the data rather than statistical noise.
| Metric | Result | Interpretation |
|---|---|---|
| AUC (Area Under Curve) | 0.76 (±0.07) | Good predictive accuracy |
| Validation Cohort Performance | Confirmed | Model generalized to independent data |
| Key Predictor | Pretreatment gene expression | Response predictable before treatment initiation |
Perhaps the most fascinating discovery emerged when researchers examined the biological pathways overexpressed in good responders. All top 20 pathways involved immune system regulation, suggesting that patients predisposed to treatment success had different immune activity even before starting therapy 9 . This finding was validated in the independent dataset, strengthening its credibility.
| Pathway Category | Specific Pathways Identified | Potential Significance in CML |
|---|---|---|
| Adaptive Immunity | T cell receptor signaling, Costimulation | Enhanced anti-leukemia immune surveillance |
| Innate Immunity | Toll-like receptor cascades, NLR signaling | Improved detection and response to abnormal cells |
| Immune Regulation | PD-1 signaling, Immunological checkpoints | Balanced immune activation without exhaustion |
The initial response classification had profound consequences for long-term outcomes. The divergence between good and poor responders persisted throughout the 10-year follow-up period, with good responders maintaining significantly lower BCR::ABL levels 9 . This finding confirms that early molecular response provides a durable indicator of long-term disease control.
This research required specialized laboratory and computational resources to connect genetic patterns with clinical outcomes.
This research transforms our understanding of CML treatment response, suggesting that a patient's pretreatment immune signature significantly influences TKI effectiveness. Rather than being determined solely by the cancer's genetics, treatment success appears closely linked to the patient's immune capacity to control the disease 9 .
Potentially identifying patients who would benefit from more potent initial therapy
Designing immunomodulatory approaches to enhance treatment response
Recognizing patients likely to achieve DMR who could be candidates for shorter treatment duration
The study aligns with emerging research highlighting the importance of the immune microenvironment in CML. Recent investigations have revealed that ferroptosis scores (measuring a specific cell death pathway) are significantly lower in CML patients and decrease further with disease progression 7 . Similarly, single-cell multiomics studies have enabled the distinction between leukemic stem cells (CD26+CD35-) and normal hematopoietic stem cells (CD26-CD35+), providing new targets for therapy 5 .
As research continues, the vision of personalized CML treatment comes closer to reality—where genetic profiling at diagnosis guides tailored therapeutic strategies, maximizing effectiveness while minimizing side effects. For patients like Sarah, such advances could mean receiving the right treatment intensity from the very beginning, potentially altering the long-term course of their disease.
This exploratory biomarker analysis from ENESTnd represents more than just a predictive model—it provides a window into the complex biological interplay between cancer and the immune system, bringing us closer to the goal of truly personalized cancer therapy.