The 11 Faces of AML

Acute myeloid leukemia (AML) is a disease of many faces. At the moment, the two main classification schemes are the French-American-British (FAB), which separates disease according to cell type and maturity, and the World Health Organization (WHO), which incorporates past disease history and specific chromosomal translocations (1). But the more we learn about AML, the more we recognize that genetic changes inform not only the patient’s initial prognosis, but also disease evolution and response to treatment. A study published in the New England Journal of Medicine provides a detailed genetic analysis of patients in three prospective multicenter clinical trials, revealing that AML can be divided into a number of subgroups based on genomic changes and complex gene interactions (2).

Figure 1. Proposed genomic classifications of AML in decreasing order of frequency in the study’s sample population.

The researchers conducted cytogenetic analysis and sequencing of 111 genes to try to understand the mutations driving AML. What they found was an extensive landscape of genetic changes – 5,234 driver mutations in total, spread across 76 genes. Of the patient samples analyzed, 96 percent exhibited at least one driver mutation and 86 percent had two or more. The size of the sample population even allowed the study’s authors to examine correlations and clonal relationships between genes. For instance, they determined that mutations in epigenetic modifier genes (like DNMT3A, ASXL1, IDH1/2 and TET2) were often among the earliest acquired, followed later by others like receptor tyrosine kinase or NPM1 mutations. Tracing the acquisition of disease-causing mutations through time suggests that AML develops in specific, ordered trajectories.

But what does this all mean for classification? The authors point out that nearly half of the patients in their cohort would not fall under WHO molecular classification criteria, despite having disease driver mutations. They suggest an alternative classification method based on Bayesian statistical analysis of their own results, leading to 11 new subtypes in order of frequency (see Figure 1). The different groups had different clinical implications, too; not only was overall survival linked to the number of driver mutations, but some specific groups (for instance, those with chromatin or spliceosome changes) had a poorer clinical outlook than expected based on their WHO classifications. Not all interactions are deleterious, though – whereas some complex gene interactions indicated an especially dismal prognosis, others conferred a survival advantage.

The AML genome is complicated and will need careful study to unpack, but one thing is clear: it’s time to consider incorporating more genomic information into our disease classifications, especially where specific mutations are known to influence clinical outcomes. With the additional information provided by TP53, SRSF2, ASXL1, DNMT3A, IDH2 and splicing-factor genes, patients in high-risk groups could be identified earlier, treated more aggressively, and provided with more accurate prognoses.