Clinical Significance of HOX and Meis gene Expression Profiling in Acute Myeloid Leukemia

Background: The clustered homeobox (HOX) gene family of transcription factors plays a crucial role in normal hematopoiesis, and their aberrant expression is implicated in acute myeloid leukemia. The selective binding of HOX proteins to DNA is influenced by interactions with cofactors, most notably the MEIS family. While the regulation of HOX genes in certain AML subsets is well-characterized, the clinical and molecular profiles of patients with varying levels of HOX and MEIS interaction and expression are not fully understood. We sought to understand the clinical outcomes and the mutational and transcriptomic landscape of a large cohort of AML patients with well-defined gene expression thresholds for HOX and MEIS families. Furthermore, we analyzed the gene expression data for an NPM1^MT-like signature, which may hold therapeutic implications using the novel menin inhibitors.

Methods: We analyzed a cohort of AML patients with HOXA1-10 and MEIS1-3 overexpression from cBioPortal (Cerami et al., 2012). We analyzed the molecular data in the context of overall survival (OS), clinical parameters, coexisting mutations, karyotype, AML subtypes, and RNA-seq expression data. Chi-square and t-tests for univariate statistics, and Kaplan-Meier and log-rank methods for overall survival (OS) were used.

Results: 1,205 adult AML patients were separated into 4 cohorts, based on HOX-MEIS expression z-scores: 430 (36%; z score < 0) negative expressors, 286 (24%; z-score 0 to 0.5) low expressors, 322 (27%; z-score 0.5 to 1.5) intermediate expressors and 167 (14%; z-score >2) high expressors. The median age of the low expressor cohort was 61.7 ys. [22-86], compared to the intermediate and high expressors at 62.4 and 57.2 ys. [23-85 ys. p=0.2; vs 20-76 ys. p=0.01 resp.]. Females were more represented in the high expressor cohort than low and intermediate expressors [46.5% vs 44% and 39.3%; p =0.01, <0.001 resp.]. Abnormal karyotype enriched the low expressor more than the intermediate and high cohorts [44% vs 39% vs 33%; p =<0.001, <0.001], whereas normal karyotype was more prominent in the high compared to low and intermediate expressor cohorts [36% vs 34% and 31%; p =0.4, 0.02 resp.]. FLT3-ITD^NEG status dominated all cohorts; low and high expressors had similar FLT3-ITD^NEG mutants [63% vs 58%, p =0.04], and intermediate expressors had the highest FLT3-ITD^POS hits compared to high expressors (44% vs 42%; p=0.3). NPM1^MT enriched the high expressors most significantly compared to low and intermediate [61% vs 43% and 33%; p =<0.001, <0.001 resp.]. All cohorts were profoundly hyperproliferative in bone marrow blast response (73% vs 70% vs 75%). The degree of pancytopenia was comparable in low and intermediate but the high expressors trended numerically towards higher Hb [7 g/dL, 7.2 g/dL vs 9.2 g/dL; p=0.8, 0.1 resp.]

The high expressors had the highest OS compared to low and intermediate expressors [22.7 mo. vs 17.1 vs 18.4 mo; p=<0.001 and 0.07 resp.]. Most of the high and intermediate expressors were enriched with TET2 (17%), DNMT3A (22%), and SRSF2 (15%) mutations, whereas low expressors enriched by poor-risk mutations including TP53 (10%), NRAS (9%) and BCOR (5%).

Given the known vulnerability of NPM1^MT to menin inhibitors, in order to find other therapeutic vulnerabilities in the same pathway, we defined a NPM1-baseline signature based on expression levels of various HOX and MEIS genes in NPM1^MT patients. Using a weighted Support Vector Machine (SVM) algorithm, we generated gene-expression similarity scores amongst the high expressors. From our search, SF3B1, DNMT3A, RAD21, RUNX1, PTPN11 and WT1 mutational signatures exhibited >95% similarity to the NPM1^MT-baseline signature. Gene set enrichment analysis (GSEA) of the co-occurring mutational partners in high-similarity targets revealed over-activation of the NOTCH pathway, PI3K pathway, and the AEP complex pathway (AF4 and ENL family genes), setting up a premise to conduct preclinical studies to find additional vulnerabilities that could be potentially synergistic with novel menin inhibitors.

Conclusion: Expression levels of HOX and MEIS shared pathways can be used to develop gene expression signatures that may reveal new potential pre-clinical targets for therapeutic targeting. Ongoing work is focused on using unsupervised machine learning to discover latent clusters of genes with a similar expression signature as HOX and MEIS family genes.

Balasubramanian:Kura Oncology: Research Funding; Alexion AstraZeneca: Speakers Bureau.