A decade of genome-wide gene expression profiling in acute myeloid leukemia: flashback and prospects

Abstract

The past decade has shown a marked increase in the use of high-throughput assays in clinical research into human cancer, including acute myeloid leukemia (AML). In particular, genome-wide gene expression profiling (GEP) using DNA microarrays has been extensively used for improved understanding of the diagnosis, prognosis, and pathobiology of this heterogeneous disease. This review discusses the progress that has been made, places the technologic limitations in perspective, and highlights promising future avenues.

Figure 1

Summary of GEP findings in a cohort of 285 cases of AML. (A) A previous study of 285 cases of AML revealed 16 subgroups (clusters) of cases based on similarities in gene expression profiles. Pairwise correlations between these AML cases are shown on the left. The cells in the visualization are colored by Pearson correlation coefficient values, with deeper colors depicting higher positive (red) or negative (blue) correlations, as indicated by the scale bar. Five of the 16 clusters have been labeled as clusters 4, 5, 9, 12, and 13. One finding of the original study was the tight aggregation into distinct clusters of AML cases with cytogenetic abnormalities that predict good risk. For those cases, cytogenetic status is color-coded in the cytogenetics column: inv(16) is yellow, next to cluster 9; t(15;17) is orange, next to cluster 12; and t(8;21) is pink, next to cluster 13. A subsequent study in the same patient cohort identified NPM1 mutations in 95 of 285 cases. NPM1 mutational status is depicted next to each case (red indicates NPM1 mutant; green, NPM1 wild-type). The figure illustrates that NPM1 mutations were not randomly distributed over the 16 previously defined clusters, but enriched in several of them. Cluster 4 was found to associate with CEBPA mutations (red indicates CEBPA mutant). However, a subset of 6 patients in this cluster did not show any CEBPA mutation (green indicates CEBPA wild-type). It was found that these cases differed in their CEBPA mRNA expression as compared with the CEBPA mutant AMLs, as indicated by the histograms depicting signal intensity values for the CEBPA probe set on the microarray. In fact, whereas CEBPA mutant AMLs highly expressed CEBPA mRNA, expression was silenced in the cases lacking mutations. This silencing was associated with CEBPA DNA promoter hypermethylation (red indicates methylation; green, no methylation). In addition, NOTCH1 mutations were found as common characteristics of this subgroup (red indicates NOTCH1 mutation; green, NOTCH1 wild-type). (B) In the original analysis of 285 AML cases (panel A left), the 44 cases in cluster 5 aggregated very tightly, as indicated by the deep red colors representing positive Pearson correlation coefficients. Most of these 44 cases showed a monocytoid morphology (FAB-M4 or -M5). This raises the possibility that a significant part of the clustering effect was caused by specific up- or down-regulation of genes that are important in monocytic differentiation, resulting in a different signature than the remaining, mostly nonmonocytoid, cases of AML in the study. To answer whether gene expression profiling would enable identification of potential heterogeneity within this apparently homogeneous subgroup, in panel B the 44 cases were reclustered as an isolated cohort. For this analysis, only probe sets that showed a variable expression within these 44 AML cases were taken into account, as defined by a fold change of 3.5 to the mean in log₂ scale in at least 1 case. The resulting cluster image shows that several potentially interesting subgroups can indeed be identified within these 44 AML cases, which have been indicated by gray lines.