Predicting relapse in favorable histology Wilms tumor using gene expression analysis: a report from the Renal Tumor Committee of the Children's Oncology Group

Abstract

Purpose: The past two decades has seen significant improvement in the overall survival of patients with favorable histology Wilms tumor (FHWT); however, this progress has reached a plateau. Further improvements may rely on the ability to better stratify patients by risk of relapse. This study determines the feasibility and potential clinical utility of classifiers of relapse based on global gene expression analysis.

Experimental design: Two hundred fifty FHWT of all stages enriched for relapses treated on National Wilms Tumor Study-5 passed quality variables and were suitable for analysis using oligonucleotide arrays. Relapse risk stratification used support vector machine; 2- and 10-fold cross-validations were applied.

Results: The number of genes associated with relapse was less than that predicted by chance alone for 106 patients (32 relapses) with stages I and II FHWT treated with chemotherapy, and no further analyses were done. This number was greater than expected by chance for 76 local stage III patients. Cross-validation including an additional 68 local stage III patients (total 144 patients, 53 relapses) showed that classifiers for relapse composed of 50 genes were associated with a median sensitivity of 47% and specificity of 70%.

Conclusions: This study shows the feasibility and modest accuracy of stratifying local stage III FHWT using a classifier of <50 genes. Validation using an independent patient population is needed. Analysis of genes differentially expressed in relapse patients revealed apoptosis, Wnt signaling, insulin-like growth factor pathway, and epigenetic modification to be mechanisms important in relapse. Potential therapeutic targets include FRAP/MTOR and CD40.

Fig. 1

Cross-validation with 144 stage III FHWT: all 144 stage III FHWT were randomly divided into 2 and 10 groups. For each training set composed of all but one group, a classifier was developed using from 1 to 150 genes; each classifier was then applied to the group withheld to define low-risk and high-risk tumors. This was repeated until all groups were categorized. This entire process was repeated 500 times. Left and right, results of the 2- and 10-fold cross-validations, respectively. For each, five curves are illustrated representing the 5th, 25th, 50th, 75th, and 95th percentiles. Red curves, 50% percentile. X axis, number of genes used in the classifier (K = 1-150). A, specificity: among tumors that did not relapse, the percentage categorized using gene expression as low risk for relapse. B, sensitivity: among tumors that relapsed, the percentage categorized as high risk for relapse. C, error rate/accuracy: percentage of tumors incorrectly classified.

Fig. 2

Success of the 10-fold cross-validation method using 50-gene classifiers for each individual tumor. Y axis, percentage of the 500 50-gene classifiers that categorized each tumor as high risk for relapse; X axis, arbitrary tumor number, with relapses (red dots to the left of the vertical line) having low numbers and nonrelapses having higher numbers (black dots).