Gene Expression Analysis of Independent Data Sets Identifies HBG1 to Be Associated with Outcome in Cytogenetically Normal AML.

Abstract 2613

Poster Board II-589

Cytogenetically normal acute myeloid leukemia (CN-AML) represents a biologically and clinically heterogeneous group. During recent years novel molecular markers like FLT3, CEBPA and NPM1 gene mutations as well as deregulated expression of single genes such as EVI1, MN1, ERG and BAALC have been identified that provide important prognostic information in CN-AML. Furthermore, DNA microarray-based gene expression profiling (GEP) has been shown to capture the molecular heterogeneity of leukemia and has been applied to build clinical outcome predictors in CN-AML.

In this study, we wanted to assess whether GEP based outcome prediction using novel biostatistical approaches applied to large gene expression data sets could refine previous findings. First we profiled gene expression in a large set of clinically well annotated CN-AML (entered on a multicenter trial for patients <60 years, AMLSG 07-04, n=154 cases) using Affymetrix Human Genome U133plus2.0 microarrays. Then, we applied L1-penalized Cox proportional hazards regression to develop a sparse prognostic model for overall survival. Using this algorithm we were able to define a gene expression signature correlated with outcome in CN-AML. Interestingly, our model resulted in a signature that only comprised one probe set corresponding to the HBG1 gene (hemoglobin, gamma A). While quantitative RT-PCR validated correct measurement of HBG1 expression (correlation r=0.96, n=15), we next evaluated our finding in independent cohorts of CN-AML cases. We applied our “HBG1 gene signature” to previously published CN-AML data (Metzeler et al. Blood 2008) comprising 163 patient samples on Affymetrix U133A/B (data set I) and 79 patient samples on Affymetrix U133plus2.0 microarrays (data set II). Univariate Cox PH regression showed that our gene expression predictor was highly significant for overall survival in both data sets (P=0.002, HR 1.71, 95%CI 1.23–2.39; and P<0.001, HR 3.03, 95%CI 1.84–5.02, respectively). Adjusted for age, NPM1, and FLT3-ITD mutational status HBG1 retained its prognostic relevance in multivariate analysis (P=0.007 and P<0.001, respectively).

As HBG1 expression might be used as novel prognostic marker in AML, the role of HBG1 expression in AML remains to be determined. HBG1 expression might represent a surrogate marker indicating the activation of distinct oncoproteins such as MYB, which recently has been shown to transcriptionally repress the level of gamma-globulin. On the other hand HBG1 expression has been shown to be induced upon exposure to azacytidine. Thus, HBG1 expression could also reflect underlying epigenetic profiles of prognostic relevance. While studying HBG1 expression in larger CN-AML cohorts as well as in the context of other molecular markers might help to further determine the prognostic impact of HBG1, integrative analyses combining GEP with genomics and epigenomics data sets will provide novel insights into the mechanisms underlying HBG1 expression in CN-AML.