Gene Expression Profiling in AML with Normal Karyotype: A Multicenter Study Investigating Molecular Markers in 252 Cases

Acute myeloid leukemia (AML) is a heterogeneous disease and AML with normal karyotype (AML-NK) is a subgroup with intermediate prognosis. Over the past years molecular analyses have led to the identification of multiple molecular biomarkers that will further allow to dissecting clinically meaningful subgroups in this disease. Here, we present a multicenter study investigating whole-genome expression profiles of 252 cases of AML-NK. In three centers (Dresden, n=78; Munich, n=97; Ulm, n=77) Affymetrix HG-U133 Plus 2.0 microarray analyses were performed according to a standardized protocol. In a first series of analyses we focused on nucleophosmin gene (NPM1) mutations, the most common genetic lesion described in adult de novo AML to date. Gene expression signatures for 138 NPM1-mutated cases were compared to the profiles of 114 NPM1-unmutated cases. Supervised classification analyses resulted in >95% prediction accuracy of NPM1 mutation status (10-fold cross-validation). The sensitivity was very high for the positive detection of NPM1-mutated cases (>97%). Using a resampling approach (100 iterations) and splitting the complete data into a training set (n=168) and test set (n=84) the following genes were amongst the most frequently selected with higher expression in NPM1-mutated cases: HOXA1, HOXA2, HOXA3, HOXA4, HOXA5, HOXA6, HOXA7, HOXA9, HOXA10, HOXB2, HOXB3, HOXB4, HOXB5, HOXB6, HOXB7, HOXB9, MEIS1, and PBX3. Lower expression in NPM1-mutated cases was observed for ABCB1, BAALC, MN1, MLLT3, or SPARC. Furthermore, our signature also showed a strong overlap to predictive NPM1 mutation signatures as published by Verhaak et al., and Alcalay et al., with 16/20 common genes and 15/18 common genes, respectively. Interestingly, some few NPM1-unmutated cases were repeatedly classified as being NPM1-mutated suggesting mutations in gene sequences not routinely screened. A strong HOXA-cluster signature is clearly dominant as well in these cases, whereas HOXB-cluster genes demonstrated a similar expression as compared to NPM1-unmutated cases. With respect to the robustness of differential gene expression signatures in NPM1 mutated cases across the laboratories using the top 300 differentially expressed genes, an overlap of 67 (22.3%) probe sets was observed between all three laboratories (NPM1-mutated vs. NPM1-unmutated: Dresden: 36 vs. 42; Munich: 42 vs. 55; Ulm: 36 vs. 41). In a second series of analyses we investigated cases with CCAAT/enhancer binding protein alpha (CEBPA) gene mutations. Gene expression signatures for 26 CEBPA-mutated cases were compared to the profiles of 132 CEBPA-unmutated cases. Supervised classification analyses resulted in >90% prediction accuracy of CEBPA mutation status (10-fold cross-validation) and frequently selected genes included HOXA2, HOXA3, HOXA5, HOXA7, HOXA9, HOXA10, HOXB2, HOXB3, HOXB4, HOXB5, HOXB6, HOXB7, and HOXB9. Thus, several HOX genes were overlapping to the NPM1 signature, albeit with an inverse correlation, i.e., HOX genes with relative overexpression expression in NPM1-mutated AML-NK demonstrated low expression in CEBPA-mutated cases. Finally, all 252 cases have also been characterized according to their fms-related tyrosine kinase 3 gene (FLT3) internal tandem duplication mutation status and this data is currently being integrated with the obtained signatures from above to develop a multi-gene prognostic model for AML-NK. In conclusion, our gene expression profiling study demonstrates robust signatures for AML-NK with NPM1 or CEBPA mutations. As our knowledge of this heterogeneous disease is increasing it seems possible to refine the classification of AML by incorporating molecular risk-based gene expression signatures.