The acute myeloid leukemias (AMLs) are a heterogeneous group of hematological malignancies with diverse clinical outcomes. Pre-treatment karyotype analysis identifies biologically distinct subgroups and is currently used as a predictor of response to induction chemotherapy and risk of relapse. Cases may be stratified into one of three prognostic groups as follows: relatively favorable prognosis [t(8;21), t(15;17) and inv(16)]; adverse prognosis [−5/del(5q), −7, abnormalities of chromosome 3q and complex karyotype]; and intermediate prognosis [remainder including normal karyotype]. HOX genes encode master transcription factors which regulate key developmental processes including differentiation, proliferation and apoptosis. Humans have 39 HOX genes and multiple lines of evidence implicate their deregulated expression in the pathogenesis of AML.

Drabkin et al. (Leukemia 2002; 16: 186–95)
have reported that AMLs with a relatively favorable prognostic karyotype are associated with low levels of HOX gene expression whereas AMLs with an adverse prognostic karyotype have higher levels of expression. To further characterize HOX gene expression in cytogenetic prognostic groups we determined the expression profiles of 26 HOX genes by real-time quantitative PCR (Q-PCR) in diagnostic samples, representative of the three prognostic groups, from 26 patients with de novo AML. Profiles were then analyzed using Artificial Neural Network based computational approaches to identify a subset of HOX genes which could discriminate between prognostic groups in a predictive fashion. Predictive models were developed for each prognostic group. Predictive classification performance for prognostic groups based on blind data of 88%, 92%, and 97% (with equal sensitivity and specificity) were achieved for the three prognostic groups. The models were interrogated to determine the nature of the relationship between the key HOX genes identified and prognostic group. The relatively favorable prognosis group was primarily defined by downregulation of HOXA5 and upregulation of HOXC4. The intermediate prognosis group was characterized by upregulation of HOXB3 and downregulation of HOXD10 and the adverse prognosis group by downregulation of both HOXC5 and HOXD3. Although the sample size is small, the results show that Artificial Neural Network based computational approaches are capable of further characterizing HOX gene expression within AML prognostic groups as determined by presenting karyotype and that measuring the expression levels of a small number of HOX genes at diagnosis can provide useful clinical information in cases where karyotype analysis has been unsuccessful.

Disclosure: No relevant conflicts of interest to declare.

Author notes

*

Corresponding author

Sign in via your Institution