Gene Expression Profiling in the Classification of Acute Leukemia Brazilian Patients

In the past few decades, genetic data has become increasingly important for acute leukemia diagnosis and patients stratification. Indeed, the present World Health Organization (WHO) leukemia classification system is largely based upon genetically defined subgroups. Gene expression profile (GEP) may correctly predict most genetic leukemia subtypes, but so far no GEP report has evaluate patients from Latin America. In the present study, we used gene expression microarray data to build an acute leukemia classifier. Bone marrow samples were collected from 231 individuals at diagnosis, 110 presented de novo acute myeloid leukemia (AML), 97 had de novo acute lymphoid leukemia (ALL) and the remaining 24 were controls who had other conditions including chronic leukemias or non-hematological diseases. GEP was evaluated based on mRNA expression signatures obtained with the Sure Print G3 Human GE (60k) system (Agilent Technologies). k-nearest neighbors prediction algorithm was applied and the top 60 informative genes were selected for each of the most prevalent genetic subtypes (T-ALL, B-ALL BCR-ABL, B-ALL ETV6-RUNX1, B-ALL TCF3-PBX1, AML PML-RARa, AML RUNX1-RUNX1T1, AML FLT3-ITD, AML NPM1mut). The less prevalent groups such as MLL rearranged and CBFB-MYH11 were not included in the classifier because of the low number of patients carrying these aberrations in our cohort. Performance of each prediction model was assessed by leave-one-out crossvalidation through the GenePattern platform (Broad Institute). The average classifier accuracy was 94.75%. Higher accuracy and precision were achieved for T-ALL (99%/96%) and AML PML-RARa (97%/97%). However, for ALL BCR-ABL, AML FLT3-DIT and AML NPM1^mut the gene signature had low precision rates (74%, 66% and 80%, respectively). The data presented here confirm that a single platform of gene expression followed by bioinformatic analysis can correctly classify genetic subgroups, but a refinement of the classifier developed is needed in order to improve the detection of heterogeneous entities such as BCR-ABL or FLT3-ITD carriers.