Gene expression profiles predictive of outcome and age in infant acute lymphoblastic leukemia: a Children's Oncology Group study

Abstract

Gene expression profiling was performed on 97 cases of infant ALL from Children's Oncology Group Trial P9407. Statistical modeling of an outcome predictor revealed 3 genes highly predictive of event-free survival (EFS), beyond age and MLL status: FLT3, IRX2, and TACC2. Low FLT3 expression was found in a group of infants with excellent outcome (n = 11; 5-year EFS of 100%), whereas differential expression of IRX2 and TACC2 partitioned the remaining infants into 2 groups with significantly different survivals (5-year EFS of 16% vs 64%; P < .001). When infants with MLL-AFF1 were analyzed separately, a 7-gene classifier was developed that split them into 2 distinct groups with significantly different outcomes (5-year EFS of 20% vs 65%; P < .001). In this classifier, elevated expression of NEGR1 was associated with better EFS, whereas IRX2, EPS8, and TPD52 expression were correlated with worse outcome. This classifier also predicted EFS in an independent infant ALL cohort from the Interfant-99 trial. When evaluating expression profiles as a continuous variable relative to patient age, we further identified striking differences in profiles in infants less than or equal to 90 days of age and those more than 90 days of age. These age-related patterns suggest different mechanisms of leukemogenesis and may underlie the differential outcomes historically seen in these age groups.

Figures

Figure 1

Impact of age and MLL status on EFS. Kaplan-Meier survival curves show the impact of infant age and MLL status on EFS. (A) Patients with MLL-R have significantly shorter EFS than those without rearrangements (MLL-G; P = .008, log-rank test; HR = 3.23). (B) MLL-AFF1 cases have a nearly identical outcome pattern to overall MLL-R (P = .010; HR = 3.26). (C) Younger infants (≤ 90 days old) compared with older infants (> 90 days old) have significantly worse EFS (P = .009, log-rank test; HR = 2.13). (D) Infants with MLL-G have the best EFS, whereas infants less than or equal to 90 days of age with MLL-R have the worst EFS. Note that 2 patients died within 7 days of diagnosis, and they were excluded from these analyses.

Figure 2

Performance of the 3-gene regression tree model of EFS in ALL cases. A 3-gene model was developed for prediction of EFS in the entire study cohort. (A) FLT3 expression separates infants into low- versus intermediate- and high-risk disease. Infants with high FLT3 expression are further divided into intermediate- and high-risk disease categories based on IRX2 and TACC2 expression. The ovals and boxes contain the relative risk followed by number of events/number of cases. (B) Kaplan-Meier survival curves show significant differences in EFS among the infants in the low-, intermediate-, and high-risk categories. (C) The model significantly separates MLL-G infants into 2 groups with significantly different EFS. NA indicates HR is not applicable because of absence of failures in 1 group. (D) Validation cohort of 22 MLL-AFF1 cases also is separated into 2 groups with different EFS. No infants with MLL-AFF1 and low FLT3 expression (low risk) were present.

Figure 3

Hierarchical clustering of MLL-AFF1 cases using top 100 SD probe sets. The top 100 probe sets, ranked by SD, were used to cluster the 48 MLL-AFF1 cases. The yellow bar indicates the branch point defining the 2 major cluster patterns. These patterns are named after the family of homeobox genes most commonly and highly expressed by its members (IRX or HOXA). Samples are shown in columns and probe sets are in rows. Captions across the right indicate the positions of some of the more conserved genes across a cluster. Increasing (red) or decreasing (green) gene expression is shown relative to the median (black) for each gene.

Figure 4

Performance of the 7-gene (9-probe set) SPCA model of EFS in MLL-AFF1 cases. Kaplan-Meier survival curves showing prediction of EFS and relapse-free survival in our cohort and the independent validation cohort. (A) The 7-gene model separates the 47 MLL-AFF1 cases into low- and high-risk groups with significantly different EFS. (B) Validation cohort of 22 MLL-AFF1 cases are similarly separated into 2 groups with significantly different EFS based on our model. (C) The 7 gene model also significantly separates the 36 older (> 90 days) MLL-AFF1 infants into low- and high-risk groups. (D) Model overlaps significantly with the P9407 MLL-AFF1 unsupervised clusters; however, it also adds significant predictive risk information, particularly in the HOXA pattern cases.

Figure 5

Heat map of probe sets associated with patient age. The top 43 probe sets associated with age (as a continuous variable) at the significance level FDR = 15% were used to generate a heat map. Patients are ordered from left to right by ascending age. In addition to the 97 infant ALL patients, 21 pediatric MLL cases are included. Vertical white lines indicate the positions of age landmarks, and the horizontal line separates between the probe sets whose expressions are positively (top) and negatively (bottom) correlated with age. Age of patients is indicated across the top.