A BCR-ABL1-Like Gene Expression Profile Confers a Poor Prognosis In Patients with High-Risk Acute Lymphoblastic Leukemia (HR-ALL): A Report From Children's Oncology Group (COG) AALL0232

Abstract 743

We have previously identified a subset of National Cancer Institute (NCI)-HR B-cell precursor (BCP) ALL patients with a gene expression profile similar to that of BCR-ABL1 ALL (BCR-ABL1-like ALL (Mullighan, N Engl J Med 2009; den Boer, Lancet Oncology 2009; Harvey, Blood, 2010, and unpublished data) and poor outcome on the COG P9906 trial, which was limited to a selected subset of HR BCP ALL patients. These cases are BCR-ABL1-negative but commonly have deletion or mutation of IKZF1. Up to half of these cases harbor rearrangements, deletions and/or mutations activating cytokine receptors and tyrosine kinase signaling (e.g. CRLF2 and activating JAK1/2 mutations), although the kinase-activating mutations in many cases remain unknown. In this analysis, we have assessed the prognostic significance of this BCR-ABL1-like signature in an unselected cohort of BCR-ABL1 negative BCP ALL patients consecutively enrolled on COG AALL0232. This phase 3 trial utilized a 2×2 factorial design comparing dexamethasone (DEX) versus prednisone (PRED) during induction, and high dose methotrexate (HD-MTX) versus Capizzi methotrexate (C-MTX) during interim maintenance 1 (IM-1). We recently reported improved event free survival (EFS) for patients receiving HD-MTX versus C-MTX (Larsen, J Clin Oncol 29: 6s, 2011) and for DEX versus PRED among patients <10 years old randomized to HD MTX (Winick, J Clin Oncol 29: 586s, 2011). We used two algorithms, Recognition of Outliers by Sampling Ends (ROSE) and Predictive Analysis of Microarrays (PAM), to define 66 of 565 (ROSE) and 81 of 572 (PAM) patients as BCR-ABL1-like.

Event-free survival (EFS) for BCR-ABL1-like cases was inferior to that of non-BCR-ABL1-like cases, irrespective of the clustering algorithm used to identify them, with 5-yr EFS rates of 63.1±7.2% vs. 84.9±2.0% (p<0.0001) for ROSE clustering and 62.6±6.9% vs. 85.8±2.0% (p<0.0001) for PAM. These differences were maintained regardless of randomized treatment arm.

We next examined variables that contributed to outcome in patients who displayed the BCR-ABL1-like signature, identified either by ROSE or PAM. Older (≥ 10 years) BCR-ABL1-like patients were significantly more likely to have an initial white blood count greater than 100,000/ul (ROSE: p< 0.001, PAM: p< 0.001). Interestingly, older females with the BCR-ABL1-like signature had superior EFS compared to males (4-yr EFS for ROSE: 73. ±9.8% vs. 43.0 ±10.3%, p=0.02; 4-yr EFS for PAM: 69. ±10.2% vs. 43. ±9.4%, p=0.04). In a multivariate COX regression analysis of the entire cohort that included identification of BCR/ABL1-like by PAM (HR 1.88, p=0.011), the other significant predictors of poor outcome were the presence of minimal residual disease (MRD) ≥ 0.01% in the bone marrow as measured by flow cytometric methods on day 29 (HR 3.09, p < 0.0001) and the presence of hypodiploidy (HR 3.14, p=0.027). In a COX model including identification of BCR/ABL1-like by ROSE (HR 1.65, p=0.053), other significant factors were day 29 MRD positivity (HR 3.26, p<0.0001), age ≥ 10 years (HR 1.61, p=0.047), presenting white blood cell count > 100,000/ul at diagnosis (HR 1.62, p=0.047), and hypodiploidy (HR 3.0, p=0.034).

In summary, the BCR-ABL1-like gene expression profile identified a subset of unselected BCP ALL patients using two different clustering algorithms that was strongly associated with a high rate of treatment failure, even with the best available therapy recently identified in COG AALL0232. The prognostic significance of these gene signatures was also independent of other known risk factors. Ongoing work to determine the genetic and biochemical landscape that contribute to this phenotype will hopefully yield new approaches to treatment for these BCR-ABL1-like patients in order to improve outcome.