A Gene Expression Classifier for Improved Risk Classification and Outcome Prediction in Pediatric Acute Lymphoblastic Leukemia (ALL).

Significant advances in the treatment of pediatric ALL have been achieved through the use of risk classification schemes that target children to increasing therapeutic intensities based on their relapse risk. However, current classification schemes do not fully reflect the molecular heterogeneity of the disease and do not precisely identify those children more prone to relapse or those who could be cured with less intensive regimens. To improve risk classification and outcome prediction in ALL, gene expression profiles were obtained using oligonucleotide arrays in a retrospective case control study of 220 children with B precursor ALL, balanced for outcome (continuous complete remission (CCR) vs. failure at 4 years) across several established prognostic variables (age, sex, WBC, karyotype). Using multiple statistical methods and computational tools, these comprehensive gene expression profiles were reduced to a 26 gene expression classifier that was highly predictive of overall outcome (two tailed p values ranging from 0.00001–0.001). Each of these 26 genes was shown to provide additional prognostic information relative to established prognostic variables (p<0.01). The 26 genes include signaling, adhesion, and growth regulatory proteins (RhoGEF4, FYB, HNK-1 sulfotransferase, SMAD1, HABP4, PHYN, IFI44L, JAG1, EFN-B2, type 3 inositol-1,4,5 triphosphate receptor, MONDOA, DOK1, CDK8, CD44, CCL5/RANTES, galectin, SPARC) and novel genes not previously known to play a role in hematopoiesis or leukemogenesis (DREBIN, MIDKINE, and the hypothetical protein FLJ20154 or OPAL1 which have cloned and characterized). High expression of 18 of the 26 genes was predictive of CCR while high expression of the remaining 8 genes (LGALS1/galectin, DOK1, GST𝛉1, CCL5/RANTES, PRG1, CD44, ATP2C1, SPARC) was predictive of treatment failure. Interestingly, 8 of the 26 genes are linked in a cell death regulatory network; 7 genes are components of a chemokine/CD44 signaling pathway; and 2 genes are critical regulators of intracellular calcium ion transport and apoptosis. Using stepwise logistic regression on the expression values of the 26 genes and 4 established prognostic variables (sex, age, WBC, t(12;21)), the best predictive outcome model was built using 9 genes alone (MIDKINE, CHST10, PHYH, IFI44L, OPAL1, CDK8, DOK1, ATP2C1, SPARC). This 9 gene predictive model was then tested for its ability to predict outcome in two independent B precursor ALL cohorts: 1) a series of 198 B precursor ALL cases previously published by Yeoh et al. (Cancer Cell 2002 1:133) where our 9 gene model was found to predict outcome with high statistical significance (p < 1.0⁻⁸); and, 2) a series of 59 B precursor ALL patients treated with a distinct modified BFM regimen CCG-1961 (p=.002; W.L. Carroll et al, in preparation). These results demonstrate that gene expression profiling can yield unique genes and classifiers that can improve outcome prediction and risk classification in ALL. Further studies may provide new insights into how these genes and pathways promote leukemogenesis and effect therapeutic responsiveness.