About 10–15% of acute myeloid leukemia (AML) arise secondary after chemotherapeutic treatment and/or radiation for a primary malignant disease and are therefore called therapy-related AML (t-AML). Overall, t-AMLs respond less well to treatment than their de novo counterparts. It has also been shown that the karyotype of the leukemic blasts is an independent prognostic parameter in t-AML. Here we assessed the gene expression profiles of 53 t-AML cases compared to a matched patient cohort of 53 de novo AML patients using Affymetrix DNA-oligonucleotide microarrays (HG-U133 chip design). The following cohort was analyzed (de novo and t-AML each): 2 t(15;17), 3 inv(16), 4 t(8;21), 4 normal karyotypes, 5 other intermediate karyotypes, 2 t(8;16), 5 inv(3), 20 t(11q23)/MLL, and 8 complex aberrant karyotypes. First, we aimed at mining for specific patterns separating de novo from t-AML. Both unsupervised, i.e. hierarchical clustering and principal component analysis, and supervised data analysis algorithms could not identify a signature that robustly separated de novo from therapy-related cases. Then the algorithm was changed to analyze both patient cohorts stratified according to favorable cytogenetics, i.e. t(15;17), t(8;21), inv(16), intermediate cytogenetics, i.e. inv(3), and unfavorable cytogenetics, i.e. t(11q23)/MLL and complex aberrant karyotypes. This resulted in gene expression signatures where de novo and t-AML patients were closely related to each other but separated within genetic subgroups. Particularly in t(11q23)/MLL, complex aberrant karyotypes and t(8;21) samples, de novo cases intercalated with therapy-related cases. When a supervised classification algorithm (Support Vector Machine) was trained to predict the subtype for both de novo and t-AML samples misclassifications were observed. However, these misclassifications mainly occurred within cytogenetic subgroups, e.g. therapy-related AML with complex aberrant karyotype was classified as de novo AML with complex aberrant karyotype, or de novo t(11q23)/MLL as therapy-related t(11q23)/MLL and vice versa. In contrast, an analysis only focusing on cytogenetic subtypes and grouping de novo and t-AML cases together clearly identified signatures related to the biologically distinct AML subtypes. Taken together, we were not able to define a specific expression profile globally associated with therapy-related AML. Instead, t-AML cases of distinct cytogenetic subgroups were demonstrated to be similar to their de novo counterpart. Despite known differences in the genetic events leading to the malignant transformation, i.e. leukemogenic exposures in t-AML through alkylating agents, topoisomerase II inhibitors, or radiation both de novo and t-AML can be considered as biologically identical diseases. Thus, given the underlying biology in AML one should separate patients in clinical trials according to cytogenetics and their history as t-AML or de novo as suggested by the WHO classification. However, differences in prognosis may not be explainable by the respective gene expression profiles in both groups.

Author notes

Corresponding author

Sign in via your Institution