Chromatin landscape and epigenetic heterogeneity of acute myeloid leukemia

By analyzing ATAC-seq data from 1,563 acute myeloid leukemia cases, this study establishes a novel epigenetic classification system comprising 16 distinct subgroups with unique chromatin landscapes, regulatory networks, and clinical implications that complement and extend beyond traditional genomic profiling.

The Gist

Imagine Acute Myeloid Leukemia (AML) not as a single, uniform enemy, but as a massive, chaotic city. For decades, doctors tried to map this city by looking only at its genetic blueprint (the DNA). They found that if a specific street sign was broken (a mutation), the city belonged to a certain district. This helped them treat patients, but it wasn't the whole story. Some patients with the "same" broken sign reacted very differently to treatment, and some with different signs acted exactly the same. It was like trying to understand a city's culture just by looking at its street names, ignoring the actual buildings, the people, and how they interact.

This paper, by a team of researchers from Japan and Sweden, decided to look at the city in a completely new way. Instead of just reading the street signs (DNA), they looked at how the city was lit up and open for business at night. They used a technology called ATAC-seq to map the "chromatin landscape."

Here is the simple breakdown of what they found, using some creative analogies:

1. The "Light Switch" Analogy (Chromatin Accessibility)

Think of your DNA as a giant library containing every book (gene) you could ever need.

• The Old Way (Genetics): Doctors looked at the library to see which books had torn pages or missing chapters (mutations).
• The New Way (Epigenetics/Chromatin): The researchers looked at which books were actually open on the tables and ready to be read.
• The Metaphor: In a cell, "chromatin" is like the shelving system. Sometimes the shelves are locked tight (closed chromatin), so the cell can't read the books. Sometimes the shelves are wide open (open chromatin), allowing the cell to read specific instructions.
• The Discovery: The researchers found that AML isn't just one disease; it's actually 16 distinct "neighborhoods" (subgroups). Each neighborhood has a unique pattern of which "books" are open and which are locked, regardless of the torn pages in the DNA.

2. The "City Zoning" Discovery

By mapping these "open books" in 1,563 patients, the team realized that the old genetic map was missing huge chunks of the city.

• They found that patients who looked identical under the old genetic rules actually lived in different neighborhoods with different "zoning laws" (gene expression).
• Conversely, patients with very different genetic mutations were found to be living in the same neighborhood, behaving in the same way.
• The Result: They created a new map with 16 subgroups. Some of these groups matched known genetic types (like the famous PML::RARA fusion), but many were brand new types that had never been recognized before.

3. The "Master Keys" (Transcription Factors & Super-Enhancers)

Why do these neighborhoods look so different? The researchers found the "Master Keys" that keep the doors open.

• Super-Enhancers: Imagine these as giant, glowing neon signs above specific buildings that say, "This is the most important building in the city!" These signs (called super-enhancers) force the cell to act a certain way.
• Transcription Factors: These are the "keyholders" who hold the keys to those neon signs.
• The Finding: Each of the 16 neighborhoods has its own unique set of Master Keys and Neon Signs. This explains why the cells in one neighborhood act like immature stem cells, while another acts like mature immune cells, even if their DNA is similar.

4. The "Crystal Ball" (Prognosis and Treatment)

This new map isn't just for show; it's a crystal ball for the future.

• Predicting the Future: The researchers found that knowing which "neighborhood" a patient lives in tells you much more about their survival chances than just looking at their DNA. Some neighborhoods are safe havens (good prognosis), while others are high-crime zones (poor prognosis).
• The "Wrong" Drug, Right Target: This is the most exciting part. They tested how these neighborhoods reacted to different drugs.
  • Example 1: They found that a specific neighborhood (Group E) was super-sensitive to a drug called Quizartinib, but only if the patient also had a specific mutation.
  • Example 2 (The Surprise): They found that Group K (which has a broken RUNX1 gene) was surprisingly sensitive to ABL inhibitors (drugs usually used for a different leukemia). Why? Because the "lighting" in this neighborhood made the cells act like early-stage B-cells, which naturally rely on ABL signaling. It's like finding that a car with a broken engine runs perfectly on a different type of fuel because of how the chassis is built.

5. The "Single-Cell" Microscope

To make sure this wasn't just a blurry average picture, they used a "super-microscope" (single-cell sequencing) to look at individual cells.

• They confirmed that every single cancer cell in a patient's body shared the same "neighborhood" signature. It wasn't a mix of different types; the whole tumor was unified by this epigenetic state. This proves that the "lighting" (chromatin) is the true boss of the cell's identity.

The Big Takeaway

For a long time, we tried to fight AML by looking at the hardware (the DNA). This paper shows us that the software (the epigenome/chromatin) is just as important, if not more so.

By mapping the "open books" and "neon signs" of the cancer, doctors can now:

1. Classify patients into 16 precise groups.
2. Predict who will survive better.
3. Prescribe the right drug, even if the patient's DNA doesn't seem to match the drug's target.

It's like moving from a world where you only know a person's last name (Genetics) to a world where you know their personality, habits, and daily routine (Epigenetics). This allows for a much more personalized and effective way to treat the disease.

Technical Summary

1. Problem Statement

Acute Myeloid Leukemia (AML) is a highly heterogeneous hematologic malignancy. While current classification systems (WHO 2022, ICC 2022, ELN 2022) rely heavily on genetic abnormalities (mutations, fusions, chromosomal changes), these genomic factors alone fail to fully explain the diversity in gene expression, cellular phenotypes, therapeutic responses, and patient prognosis. The role of the epigenome, particularly chromatin state and accessibility, remains largely unexplored in large-scale patient cohorts. Previous studies have been limited by small sample sizes or a focus on DNA methylation (often restricted to promoters), leaving a gap in understanding how genome-wide chromatin accessibility drives AML heterogeneity and pathogenesis.

2. Methodology

The study, termed eCHROMA AML, employed a multi-omics approach on a massive cohort to map the chromatin landscape of AML.

• Cohort: 1,563 AML patients from independent Swedish and Japanese cohorts.
• Data Generation:
  • ATAC-seq: Performed on all 1,563 patients to profile genome-wide chromatin accessibility.
  • Genomics: Targeted-capture sequencing (331 driver genes) and SNP arrays for copy number alterations (CNAs) in all patients. Whole-genome sequencing (WGS) and RNA-seq were performed on subsets.
  • Epigenomics: ChIP-seq for histone marks (H3K27ac, H3K27me3) and structural proteins (CTCF, SMC1, RNA Pol II) in ~200 samples. DNA methylation arrays were performed on 424 samples.
  • Single-Cell Multi-omics: scRNA-seq and scATAC-seq (multiome) were performed on 36 AML samples and 4 complete remission (normal) controls, totaling ~281,000 cells.
  • Drug Screening: Ex vivo drug sensitivity testing was conducted on 112 samples against 250 drugs, integrated with data from the BeatAML cohort.
• Analytical Pipeline:
  • Clustering: Leiden clustering on 3,000 most variable ATAC peaks identified distinct epigenetic subgroups.
  • Integration: Multi-layered integration of ATAC, RNA, ChIP, and mutation data.
  • Regulatory Network Analysis:
    • ANANSE: Used to infer gene regulatory networks (GRNs) and identify key transcription factors (TFs) based on chromatin accessibility and expression.
    • coltron: Used to map TF networks regulated by super-enhancers (SEs).
    • SCENIC+: Applied to single-cell data to infer TF activity and eRegulons.
  • Differentiation Analysis: Deconvolution of bulk ATAC data and pseudotime trajectory analysis using scRNA-seq mapped to normal hematopoiesis references.
  • Clinical Modeling: Cox regression and LASSO models were used to assess prognostic value and drug sensitivity correlations.

3. Key Contributions & Results

A. Identification of 16 Epigenetic Subgroups

The study classified AML into 16 distinct subgroups (A–P) based solely on ATAC-seq profiles.

• Novelty: While subgroups A, B, C, and I aligned with known genetic fusions (PML::RARA, RUNX1::RUNX1T1, CBFB::MYH11) or CEBPA bZIP mutations, the remaining 12 subgroups represented novel AML subtypes not defined by conventional genomic classifications.
• Complex Mapping: A single epigenetic subgroup often contained multiple genetic subtypes (e.g., Subgroups D–G contained various NPM1-mutated and KMT2A-rearranged cases), and vice versa. This suggests that epigenetic states capture biological heterogeneity that genetics alone misses.
• Correlation: These subgroups showed stronger correlations with clinical features (WBC, blast count, differentiation state) and gene expression profiles than WHO/ICC classifications.

B. Molecular Characterization of Subgroups

• HOX-Related Subgroups (D–G): Characterized by HOX gene overexpression. They were further stratified by specific co-mutations (e.g., Subgroup D: NPM1 + TET2/IDH; Subgroup E: NPM1 + WT1/cohesin; Subgroup G: KMT2A-rearranged + complex karyotype).
• Differentiation States: Single-cell analysis revealed distinct differentiation blocks:
  • Subgroups D–F showed blocks at the HSC or GMP stages.
  • Subgroups F–H showed monocytic differentiation, with Subgroup H resembling Chronic Myelomonocytic Leukemia (CMML).
  • Subgroup K (RUNX1-mutated) showed a block at the Common Lymphoid Progenitor (CLP) stage, indicating a lymphoid commitment.
• Super-Enhancers (SEs): The study identified 1,718 non-redundant SEs.
  • Common SEs regulated general oncogenic TFs (e.g., ETV6, TP53).
  • Subgroup-specific SEs drove unique identities (e.g., FOXC1 in Subgroup D, IRF8 in Subgroup G, CD34 in Subgroup O).
• Gene Regulatory Networks (GRNs): Each subgroup was driven by a unique combination of TFs regulated by subgroup-specific SEs. For instance, HOXA family TFs drove Subgroups D–E, while SPI1/CEBPA drove monocytic subgroups.

C. Single-Cell Validation

• scATAC-seq confirmed that leukemic cells within the same ATAC subgroup shared a consistent chromatin signature, distinct from normal cells and other subgroups, regardless of the patient's differentiation state.
• TF activity analysis (SCENIC+) revealed that key TFs (e.g., HOXA9, SPI1) exhibited distinct temporal activation profiles along the differentiation trajectory depending on the subgroup, suggesting that the timing of TF activity is crucial for leukemogenesis.

D. Clinical and Therapeutic Implications

• Prognostic Value: ATAC subgroups had significant prognostic value independent of ELN risk stratification.
  • Favorable: Subgroups A, I, C, B.
  • Adverse: Subgroups L, G, N, O.
  • Improvement: Integrating ATAC subgroups into the ELN model significantly improved the concordance index (c-index) for survival prediction in both Swedish and Japanese cohorts.
• Drug Sensitivity:
  • FLT3 Inhibitors: Subgroup E (enriched for FLT3 mutations) showed high sensitivity.
  • MEK Inhibitors: Subgroups C, F, and H (monocytic differentiation) showed sensitivity to MEK inhibitors, even in the absence of RAS mutations, suggesting an epigenetic basis for this sensitivity.
  • ABL Inhibitors: Subgroup K (RUNX1-mutated, lymphoid-blocked) showed unexpected sensitivity to ABL inhibitors (e.g., dasatinib, ponatinib), likely due to the epigenetic similarity to early B-cells which rely on ABL signaling.

4. Significance

• Paradigm Shift: This study establishes that chromatin accessibility is a fundamental layer of AML heterogeneity that complements and often refines genetic classification. It reveals "epigenetic subtypes" that are biologically and clinically distinct.
• Mechanistic Insight: It elucidates the mechanism of leukemogenesis by linking specific chromatin landscapes to super-enhancers, transcription factor networks, and differentiation blocks.
• Clinical Utility: The ATAC-based classification provides superior prognostic stratification compared to current genetic models and identifies epigenetically driven drug sensitivities (e.g., MEK and ABL inhibitors in specific subgroups) that are not predicted by mutation status alone.
• Resource: The eCHROMA AML dataset (1,563 cases with multi-omics data) serves as a comprehensive reference resource for the AML research community to explore epigenetic mechanisms and develop targeted therapies.

In summary, Ochi et al. demonstrate that profiling the chromatin landscape via ATAC-seq uncovers a new layer of AML classification that offers deeper insights into disease biology, improves risk stratification, and opens new avenues for precision medicine.