The cancer hallmark concept defines the biological processes that contribute to cancer development and progression. While the hallmarks represent an important theoretical framework, they reflect generic properties of all cancer types and hence most hallmarks are not measurable nor applicable to guiding diagnosis or therapy.

Here, we carried out deep multi-omic profiling (DNA panel sequencing, , transcriptome, DNA methylation and proteome) and functional testing of responses to 525 drugs in 118 primary acute myeloid leukemia (AML). In addition, 47 primary AML patients were used as an independent validation cohort, including measurements by single-cell assays. Using unsupervised multi-omic dimensionality reduction, we defined 11 independent axes of biological variability that we considered here to reflect the data-driven hallmarks (DDHM) of AML. We defined the molecular features that contributed most to each hallmark, hence forming an understanding of the biological nature and pathways of each DDHM We hypothesized that the 11 DDHMs define biologically and clinically important features of AML applicable to diagnostics, therapy and precision medicine. For example, each patient had a unique composition of the DDHMs that reflected the AML biology of that patient case. Each DDHM was associated with distinct AML biology, including NPM1 mutation, transcriptional deregulation, stem cell properties, cell cycle, stromal signals, HOX signaling, and stress response. The DDHMs captured both biology that is already well known in AML, but with multi-omic resolution, and highlighted also novel, previously poorly understood, properties of AML.

Each DDHM included a ranked list of drugs that explained the biology reflected in the hallmarks. For instance, the DDHM2 that defined sensitivity to Bcl-2 (Venetoclax) and MEK inhibitors highlighted myeloid cell differentiation patterns.

Moreover, DDHM8 was strongly associated with patient prognosis, a higher response to hypomethylating agents and complete remission,. Combining this prognostic DDHM8 with hallmark DDHM1 that explained the omics of the NPM1 mutation landscape, captured the ELN risk classification and other prognostic cell surface markers. This revealed that several previously independently defined biomarkers of prognosis converge in the multi-omic paradigms defined by our DDHMs.

In summary, we have defined through multi-omics and functional analysis a set of eleven data-driven hallmarks of AML. These capture some of the most important known prognostic and therapeutic biomarkers in AML but also highlight opportunities for novel biomarker discovery and drug repurposing. This approach provides a new quantitative and measurable framework to understand the biological heterogeneity of AML and exploit it for diagnosis, prognosis and precision therapy.

No relevant conflicts of interest to declare.

Author notes

*

Asterisk with author names denotes non-ASH members.

Sign in via your Institution