Survival Prediction Acute Myeloid Leukemia (AML) Using a Combination of DNA Methylation Analysis and Gene Expression Profiling.

Acute myeloid leukemia is classified by the presence or absence of recurrent cytogenetic aberrations. In order to improve diagnosis and therapy, more recently new studies have been performed to supplement the current classification with refined molecular information based on gene expression profiling. However, it has been established that expression levels of genes are often largely controlled by the state of cytosine methylation in the adjacent promoter region. Thus we were interested to evaluate the quantitative methylation levels for a previously identified predictive set of genes (Bullinger et al. 2004) using a novel technology based on a unique combination of base specific cleavage of single stranded nucleic acids with MALDI TOF detection. We have employed this new quantitative high throughput DNA methylation analysis technology to analyze 147 promoter regions in a total of 192 individuals. The resulting quantitative methylation data was analyzed using a semi-supervised approach to evaluate the quantitative methylation data as a predictor for patient survival. We used a first set of 96 individuals to train a statistical learning algorithm and a second set of 96 samples to validate the trained algorithm. The analysis revealed quantitative methylation patterns as a reliable predictor for survival (p < 0.001). Subsequently, we combined the methylation based predictive model with the results from the expression based predictor. The combination of both models yielded a superior predictive model for patient survival, which outperformed all clinical and cytogenetic risk stratification in the given sample set. The results of this work revealed a potential significance of DNA methylation in the pathophysiology of AML and suggest that DNA-methylation patterns might be useful molecular markers for patient survival prediction based on the fact that large-scale DNA methylation studies can now be performed with reasonable efforts in a limited amount of time. Therefore, these results lay the groundwork for future research which might ultimately enable individualized therapy based on improved molecular characterization of AML.