Identification of Genomic Classifiers That Distinguish Induction Failure in T-Lineage Acute Lymphoblastic Leukemia in COG Study 9404.

The clinical features of age, white count, and presence of extramedullary disease cannot predict risk for induction failure (IF) in patients who present with T-cell acute lymphoblastic leukemia (T-ALL). On the basis of recent observations that gene expression profiles can distinguish clinicopathologic cohorts of patients with acute leukemia, we hypothesized that microarray analyses performed on diagnostic T-ALL bone marrow samples might identify a genomic classifier for IF patients. Using a case-control study design for children and young adults treated for T-ALL on Children’s Oncology Group Study 9404, we analyzed 50 cryopreserved T-ALL samples using Affymetrix U133A Plus 2 genechips, which have 54,000 genes, ESTs and genomic classifiers. Following RMA normalization, we used Prognostic Multi-array Analysis (PAM) to identify a 116-member genomic classifier that could accurately identify all 6 IF cases from the 44 patients who achieved remission. Within the IF cohort, 37 genes were up-regulated and 79 were down-regulated in comparison to other outcome groups. To further investigate the genetic mechanisms governing IF, we developed four cell lines with acquired drug resistance: Jurkat and Sup T1; each having resistance to daunorubicin (DNR) and asparaginase (ASP). Using a comparative analysis for fold-change in gene expression among 6 IF patients and the T-ALL DNR and ASP-resistant cell lines, we identified seven genes that were up-regulated, and another set of seven genes that were commonly down-regulated. To validate the potential use of our 116-member gene set in predicting IF in T-ALL, we tested our genomic classifier in 42 cases which were treated on COG study 8704 and hybridized to the Affymetrix U133Av.2 chip. Because only 85 probes were shared between U133A Plus 2 and U133Av. 2 chips, we employed shrunken class centroids to constrain our classifier to 25 rank-ordered probes. This smaller classifier correctly identified the single IF case in 8704, as well as another patient who was an early treatment failure, indicating that similar genomic classifiers may identify IF patients in different clinical trials. These results indicate that genetic profiling may be useful in prospectively identifying IF patients in T-ALL. In addition, we identified genes that were commonly upregulated in IF patients and T-ALL cell lines with intrinsic drug resistance.