AML Review: Emerging Subtypes, Diagnostics, and Prognosis

Acute myeloid leukemia is no longer treated as a single disease. Over the past few years, advances in molecular profiling have fractured AML into dozens of genetically defined subtypes, each carrying different treatment implications and survival expectations. Two competing classification systems now guide diagnosis, next-generation sequencing is reshaping how doctors track residual disease, and targeted drugs are reaching patients whose specific mutations were untreatable just a decade ago. The result is an AML landscape that looks strikingly different from what textbooks described even five years ago.

Two Classification Systems, One Disease

In 2022, the field found itself in an unusual position: two major classification systems for myeloid cancers were published within months of each other. The World Health Organization released its 5th edition (WHO5), and a separate international group published the International Consensus Classification (ICC). Both systems move AML classification deeper into genetics, using recurrent chromosomal and molecular abnormalities as the primary organizing principle rather than relying mainly on how blast cells look under a microscope.¹

Where the two systems diverge matters for patients near the boundary between AML and myelodysplastic syndromes (MDS). The ICC requires at least 10% blasts in the bone marrow or blood to diagnose AML, while WHO5 drops a fixed blast cutoff for certain genetically defined entities.² They also handle TP53-mutated disease differently and draw the line between MDS and AML in slightly different places.³ For clinicians, this means a patient could technically receive different diagnoses depending on which system their institution follows, though in practice both systems converge on the same treatment path for most cases.

Molecular Subtypes That Shape Treatment

The mutations that define AML subtypes are not just diagnostic labels. They directly influence which drugs work, how likely the leukemia is to come back, and whether a stem cell transplant should be part of the plan. A few gene mutations stand out as especially consequential.

FLT3 and NPM1

FLT3 internal tandem duplication (FLT3-ITD) is one of the most common mutations in AML and historically one of the most worrisome. But its impact depends heavily on context. In patients who also carry an NPM1 mutation, a low FLT3-ITD burden behaves much more favorably than a high burden. A study of over 300 intermediate-risk AML patients found that those with NPM1 mutations and a low FLT3-ITD ratio had survival and relapse rates similar to patients without FLT3-ITD at all, while a high ratio was tied to significantly worse outcomes. In patients without NPM1 mutations, any level of FLT3-ITD worsened prognosis.⁴ This interplay between two mutations in the same patient is a central reason AML risk stratification has become so granular.

NPM1 itself is one of the most frequently mutated genes in newly diagnosed AML, present in roughly a third of cases. It generally carries a favorable prognosis at diagnosis but a poor one if the disease relapses.⁵ Targeted therapies now exist for both FLT3 and NPM1-mutated AML, a point addressed further below.

IDH1 and IDH2

Mutations in the IDH1 and IDH2 genes alter a metabolic enzyme, causing it to produce a compound that blocks normal cell maturation. These mutations are targetable with approved inhibitors. Their prognostic significance is less straightforward than FLT3-ITD. Some studies suggest that IDH2 R140 mutations carry a more favorable outlook than IDH1 R132 or IDH2 R172 mutations, while other data show no clear difference among the three. One likely explanation is that R140 IDH2 mutations frequently co-occur with NPM1 mutations, and the favorable prognosis associated with NPM1 may be doing much of the heavy lifting.⁵

TP53 Mutations and Multi-Hit Status

TP53 mutations represent the most adverse molecular subtype in AML. Both new classification systems single out TP53-mutated disease, and for good reason. What matters is not just whether TP53 is mutated but whether the patient carries a “multi-hit” alteration, meaning both copies of the gene are disrupted through some combination of mutation, deletion, or loss of the normal allele. Among patients who underwent stem cell transplantation, those with a single TP53 mutation had roughly 63% two-year overall survival, compared to about 46% for multi-hit patients. The relapse rate told a starker story: roughly 16% at two years for single-hit versus about 36% for multi-hit.⁶ Even among TP53-mutated patients, complex cytogenetics independently worsened survival, while co-occurring IDH1 mutations and transplantation both improved it.⁷

Therapy-Related AML Has Its Own Genetic Profile

AML that develops after prior cancer treatment, known as therapy-related AML, is genetically distinct from the disease that arises on its own. A study of 352 patients with therapy-related myeloid cancers found TP53 was the most commonly mutated gene at 37%, followed by PPM1D at 19% and TET2 at 16%. About 61% of those TP53 mutations were multi-hit. Compared to de novo AML, therapy-related disease was enriched for TP53, PPM1D, and SETBP1 mutations, while standard de novo mutations like NPM1, FLT3, and IDH1/2 were less common.⁸ This different mutational landscape helps explain why therapy-related AML tends to respond poorly to conventional chemotherapy and why these patients are often channeled into different treatment strategies from the start.

Hereditary AML and DDX41

Most AML develops from mutations acquired during a person’s lifetime. But a growing number of cases trace back to inherited genetic variants. The most prominent hereditary AML gene identified so far is DDX41, found in about 2% of all AML and MDS patients.⁹ People who inherit a DDX41 mutation typically develop AML later in life, often in their 60s or 70s, making it easy to miss the hereditary connection. The disease tends to present with a normal karyotype, high-risk features, and a poor prognosis, and the inherited mutation acts as a “first hit” that sets the stage decades before the leukemia appears.¹⁰ Recognizing DDX41 mutations has practical implications for family screening, since siblings of an affected patient may carry the same variant and benefit from monitoring.

Detecting Residual Disease After Treatment

Getting a patient into remission is only half the battle. Knowing whether leukemia cells are truly gone, or merely hiding below the threshold of standard tests, is critical for deciding what comes next. Measurable residual disease (MRD) testing has become central to that decision.

Traditionally, MRD has been assessed by flow cytometry, which looks for abnormal cell-surface markers. Next-generation sequencing (NGS) takes a different approach, hunting for known mutations in the DNA of remaining cells. Head-to-head comparisons show NGS catches substantially more residual disease than flow cytometry alone. In one study, NGS identified MRD in about 54% of post-treatment patients, compared to 26% flagged by traditional methods. Among patients who appeared to be in complete remission, those who were NGS-positive had significantly shorter survival than those who were NGS-negative.¹¹ Another study found that NGS identified over 80% of flow-cytometry-positive cases, while flow cytometry caught only about half of NGS-positive cases, suggesting the two methods work best when combined.¹²

Optical Genome Mapping and Hidden Structural Changes

Standard chromosome analysis (karyotyping) has been a cornerstone of AML diagnosis for decades. But it misses structural changes that are too small or too cryptic for microscope-based detection. Optical genome mapping (OGM), a newer technology that images long strands of DNA directly, is filling that gap. In one study comparing OGM to karyotyping, concordance for structural variants above a certain size was about 98%. OGM also identified additional clinically significant variants in roughly 10% of cases that karyotyping missed.¹³

The payoff is especially striking in AML cases that appear to have a normal karyotype by conventional testing. A study of 48 such cases found that OGM detected abnormalities in 46% of them, with a third harboring large structural variants or alterations affecting leukemia-associated genes, many linked to adverse prognosis.¹⁴ For patients previously categorized as intermediate risk because nothing showed up on karyotyping, this reclassification can change treatment decisions entirely.

Single-Cell Sequencing and Clonal Complexity

When you sequence a bulk sample of bone marrow, you get an averaged picture of whatever mutations are present. You cannot tell which mutations coexist in the same cell or how different subclones relate to each other. Single-cell sequencing resolves that problem by reading the genome (or gene expression, or protein markers) of individual cells. In AML, this has revealed that the disease is often far more heterogeneous than bulk testing suggests, with multiple competing subclones evolving in parallel.¹⁵

A study of complex-karyotype AML using single-cell multiomics found that 75% of cases harbored multiple subclones, many of which showed ongoing chromosome remodeling driven by breakage-fusion-bridge cycles and chromothripsis. The researchers identified three patterns of clonal evolution: monoclonal, linear, and branched polyclonal. Using patient-derived mouse models, they demonstrated that different subclones had distinct drug sensitivities, and they identified therapies that specifically targeted leukemic stem cells, including BCL-xL inhibition.¹⁶ This kind of subclone-level drug matching is still largely a research tool, but it points toward a future where treatment is guided not just by which mutations are present but by how different clones within the same patient will respond.

Prognostic Risk Stratification

The European LeukemiaNet (ELN) risk classification is the most widely used framework for sorting AML patients into favorable, intermediate, and adverse categories. Its 2022 update (ELN22) incorporated many of the molecular insights described above, including multi-hit TP53 status and refined FLT3-ITD handling. A validation study in a large cohort of intensively treated patients found that ELN22 produced significantly better prognostic discrimination for overall survival compared to the prior 2017 version.¹⁷ A separate validation from another group concluded that ELN22 is helpful for guiding transplant decisions.¹⁸

Not every study agrees perfectly. One large validation found that ELN22 identifies a larger group of adverse-risk patients but at the cost of slightly reduced overall prognostic accuracy compared to ELN17.¹⁹ The trade-off reflects a deliberate choice: better to catch more high-risk patients, even if a few intermediate-risk patients get swept into the adverse bucket, than to miss people who need aggressive treatment.

Machine learning models are also entering the picture. One large study used nine different algorithms trained on clinical, laboratory, and genetic data from nearly 1,400 AML patients. For predicting complete remission, these models achieved areas under the curve between 0.77 and 0.86, and between 0.63 and 0.75 for two-year survival, holding up well in an independent validation cohort.²⁰ A separate transcriptomics-based approach using RNA sequencing data from over 400 patients achieved above 90% accuracy in distinguishing ELN favorable from adverse groups and identified gene signatures that may serve as standalone prognostic biomarkers.²¹ These tools are not replacing ELN in the clinic yet, but they hint at a future where risk prediction integrates far more variables than any manual scoring system can handle.

Targeted Therapies Matching Mutations

The molecular subtyping described above is not just academic. FLT3 inhibitors and IDH1/2 inhibitors are already approved and integrated into standard care for patients whose leukemia carries those mutations.²² The newest class to reach patients is menin inhibitors, which target the protein interaction driving leukemia in KMT2A-rearranged and NPM1-mutated cases. Revumenib received approval in 2024–2025 for relapsed or refractory KMT2A-rearranged acute leukemia and NPM1-mutated AML, making it the first targeted therapy for these high-risk subtypes.²³

Resistance remains a persistent challenge. Leukemic stem cells, the rare cells that sustain the disease and drive relapse, can shift their survival dependencies in response to treatment. Longitudinal analyses of venetoclax-treated patients have shown that resistance frequently arises through plasticity in leukemic stem cells, which shift toward a megakaryocytic/erythroid progenitor state and switch their survival dependency from BCL-2 (the target of venetoclax) to BCL-xL.²⁴ Understanding these escape routes is essential for designing combination strategies that can head off resistance before it takes hold.

Speed of Molecular Results Matters

Getting molecular profiling results quickly is not just a convenience. A study of patients hospitalized for acute leukemia found that an ultra-rapid sequencing platform (Ion Torrent Genexus) returned results in a mean of about 3 days, compared to roughly 5.4 days for conventional methods. Every extra day of waiting was associated with nearly a day of delayed treatment initiation and about four additional days in the hospital.²⁵ For a disease where the right targeted drug can depend entirely on which mutation is present, those extra days of uncertainty translate into real clinical cost.

How AML Evades the Immune System

The bone marrow microenvironment in AML is not a passive bystander. Leukemia cells actively suppress the immune response through several mechanisms: they upregulate checkpoint molecules that shut down T cell activity, downregulate the surface proteins that T cells need to recognize them, drive T cells into a state of exhaustion, and recruit regulatory T cells that further dampen immune surveillance.²⁶ These evasion strategies help explain why checkpoint inhibitors, which have transformed treatment for many solid tumors, have so far produced disappointing results in AML. Understanding which evasion mechanisms dominate in a given patient could eventually guide the selection of immunotherapy combinations.

Pediatric AML Is a Different Disease

Children and adults with AML share some features, but the underlying genetics diverge sharply with age. KMT2A rearrangements are far more common in pediatric AML, while adult disease is enriched for DNMT3A mutations and splicing factor gene mutations. Core binding factor AMLs (those involving RUNX1-RUNX1T1 or CBFB-MYH11 fusions) span the age spectrum, but newer diagnostic technologies have identified fusions and duplications that are heavily concentrated in children, including NUP98 fusions, GLIS family fusions, and UBTF tandem duplications.²⁷ A dedicated pediatric genomic framework has been proposed to capture these age-specific subtypes, since adult-centric classification systems miss many of them entirely.²⁸

Epigenetic Subtypes and DNA Methylation

Beyond mutations in coding genes, the way DNA is chemically marked with methyl groups creates another layer of AML subtyping. An analysis of DNA methylation patterns across AML patients identified three distinct groups based on methylation levels at specific genomic sites: high, moderate, and low methylation phenotypes. These groups differed from one another in overall survival, cell appearance, and the composition of their immune microenvironment. The low-methylation group was enriched for DNMT3A mutations, while the moderate group was enriched for RUNX1 mutations.²⁹ Epigenetic profiling is not yet standard in clinical practice, but it adds prognostic information that genetic mutation testing alone does not capture, since two patients with the same mutation can have very different methylation landscapes and very different outcomes.