This comprehensive and large-scale study of DNA methylation profiles associated with ~14.000 genes in a human disease demonstrates that epigenetic patterning distributes into signatures of biological and clinical significance and that DNA methylation classifiers can be derived from population studies with clinical predictive power. From the biological standpoint, these data offer an opportunity to better understand the mechanisms through which hematopoietic cells undergo leukemogenesis. Much effort has been invested in identifying genetic lesions that cooperate with known recurrent translocations such as t(8;21), t(15;17) and inv(16) or in patients with normal karyotype leukemia. Although this effort has led to the identification of bona fide leukemogenic mutations such as those in CEBPA, FLT3, and NPM1, it now appears that recurrent genetic lesions insufficiently explain the biological diversity of clinical AML. In contrast, our data show that epigenetic lesions are abundant and common, raising the possibility that a number of the oncogenic lesions in AML could be epigenetic in nature. Thus, further research exploring the contribution of genes affected by aberrant DNA methylation seems warranted.
The clinical significance of DNA methylation profiles is underlined by the fact that it contributes to identifying groups of patients that share a common clinical outcome, in some cases even beyond what their cytogenetic class is, such as the case of cluster 3 leukemias. This cluster, which was enriched for t(8;21) patients, included others without this cytogenetic marker, however, there was no difference in survival between the two subgroups. Although some of the patients negative for the t(8;21) presented other cytogenetic and molecular indicators of favorable risk, this finding reflects the existence of a common DNA methylation profile for these patients. This epigenetic signature aggregated these leukemias together beyond the presence of other molecular and cytogenetic markers, and in addition identified additional cases that did not present with any favorable risk indicator. Similarly, a hypermethylated gene signature defines a subset of leukemias with CEBPA
silencing due to hypermethylation, T cell lineage infidelity, resistance to myeloid growth factors, and a poor prognosis (Figueroa et al., 2009b
). These cases formed cluster 10 in this cohort. AMLs with mutations on both CEBPA
alleles or with homozygous mutations were recently shown to have a highly favorable prognosis (Wouters et al., 2009
), and these cases also presented with a defining DNA methylation profile. Taken together, these data warrant considering both of these subtypes as distinct leukemia diseases that should be assigned to risk stratified therapy regimens and explored for the development of specific targeted therapy.
NPM1 mutations distributed to four related but slightly distinct signatures. These epigenetic variations cannot be explained by the presence or absence of a concurrent FLT3-ITD, suggesting that other as yet unrecognized mechanisms might be at play in determining these different epigenetic groups. We were unable to identify a DNA methylation signature associated with FLT3 lesions, indicating that mutations of this gene do not exert their effects in AML by imposing an aberrant epigenetic pattern.
One of the notable findings of this study was the identification of five methylation signatures with no other common morphologic or molecular features, but with distinct clinical outcomes, suggesting that these too are unique forms of AML with their own biological characteristics. It is particularly significant that these AML subtypes cannot be identified by any available diagnostic method, underlining that epigenetic signatures provide a critical layer of additional information. The fact that these cases included both normal karyotype leukemias as well as those with cytogenetic lesions and across multiple FAB subtypes supports a move away from definitions rooted in standard karyotyping, rather toward a more functional classification of AML. Future studies will be required to explore the biological basis of these epigenetically defined subtypes in the effort to develop risk-adapted and molecular targeted clinical trials that more accurately reflect interindividual differences among leukemia patients. However, the presence of a strong hypermethylated signature in some of these clusters (clusters 2, 7, and 15) () along with their unfavorable prognosis leads us to speculate that these patients, as well as those in the CEBPA-silenced cluster, might benefit from the inclusion of hypomethylating agents as part of their therapeutic regimen.
Furthermore, in this study we identified the presence of a common DNA methylation signature that is detected in the vast majority of cases. The nature of the genes found in this common epigenetic signature, which included tumor suppressors, putative and well-described transcription factors, nuclear import proteins, apoptosis-related proteins, and a regulator of myeloid cytokines, is highly suggestive of a role in leukemic transformation. In addition, we found that this aberrant methylation was accompanied by significant downregulation of these genes. The fact that these genes are affected in a broad fashion, across multiple different subtypes of AML, leads us to believe that deregulation of these genes is most likely a necessary, though probably not sufficient, event during the malignant transformation process of hematopoietic cells.
Finally, the study identified a robust 15 gene methylation classifier that was predictive of overall survival, which was generated in an unbiased manner using a large enough data set to perform training, testing, and independent validation. The methylation predictor was further validated as an independent risk factor in a multivariate analysis. Because DNA is stable and readily obtained from clinical specimens, we believe that this DNA methylation classifier could serve as a clinically useful biomarker used for decision-making in future clinical trials. In conclusion, although epigenetic deregulation has been recognized as a hallmark of cancer for some time, the use of epigenomics to further expand our understanding of the biology of these diseases has only more recently become feasible in the clinical context. Here we show that DNA methylation profiling is a powerful tool for the clinical stratification of AML and to further explore and define the biology of this disease.