Diagnosis of the small round blue cell tumors using multiplex polymerase chain reaction

Abstract

The small round blue cell tumors of childhood, which include neuroblastoma, rhabdomyosarcoma, non-Hodgkin's lymphoma, and the Ewing's family of tumors, are so called because of their similar appearance on routine histology. Using cDNA microarray gene expression profiles and artificial neural networks (ANNs), we previously identified 93 genes capable of diagnosing these cancers. Using a subset of these, together with some additional genes (total 39), we developed a multiplex polymerase chain reaction (PCR) assay to diagnose these cancer types. Blinded testing of 96 new samples (26 Ewing's family of tumors, 29 rhabdomyosarcomas, 24 neuroblastomas, and 17 lymphomas) using ANNs in a complete leave-one-out analysis demonstrated that all except one sample were accurately diagnosed as their respective category. Moreover, using an ANN-based gene minimization strategy in a separate analysis, we found that the top 31 genes could correctly diagnose all 96 tumors. Our results suggest that this molecular test based on a multiplex PCR reaction may assist the physician in the rapid confirmation of the diagnosis of these cancers.

Figure 1

The schematic illustration of multiplex RT-PCR assay. A: The multiplex RT-PCR involves two stages: the first stage includes reverse transcription and amplification using chimeric primers, and the second stage converts to the use of a single pair of universal primers during amplification (see Materials and Methods for the primer design). B: The amplicons obtained from multiplex amplification were then analyzed using fluorescence capillary electrophoresis. The peak location represents the gene identity, and the peak area represents gene expression level. C: The comparative chromatograms of four different categories of tumor samples from one multiplex assay. Blue, Lymph-13; yellow, EWS-T-4; red, RMS-A-18; and green, NB-20.

Figure 2

The artificial neural network. A: Workflow for a complete leave-one-out ANN analysis. Multiplex RT-PCR analysis using 40 genes was performed on tumors from 96 pediatric cancer patients (26 EWS, 29 RMSs, 17 lymphomas, and 24 NBs). One sample was left out as an independent test sample, and the ANNs were trained using the remaining 95 samples. ANN training scheme (gray box). 1, All samples were randomly partitioned into three groups. 2, One of the three groups (containing 32 samples) was selected as a validation set, whereas the remaining two groups (63 samples) were used to train the network. 3 and 4, The training weights were iteratively adjusted for 100 cycles (epochs). 5, The ANN output (0 to 1) for each of four classes (EWS, RMS, NB, and lymphoma) was calculated for each sample in the validation set. 6, A different validation set was selected from the same partitioning in 1, and the remaining two groups were used for training. Steps 2 through 6 were repeated until each of the three groups from 1 had been used as a validation set exactly one time. 7, The samples were randomly repartitioned into three new groups, and steps 2 through 6 were repeated. Sample partitioning was performed 100 times in total. Thus, steps 1 through 6 were repeated 100 times. Three hundred ANN models were thus trained and were used to predict the left-out test sample. This scheme was repeated for each left-out test sample. B: Classification of the samples from a leave-one-out ANN analysis. A sample is classified to a cancer category according to its highest committee vote (average of all ANN outputs; Table 1). Plotted is the distance for each sample from its committee vote to the ideal vote for that category (for example, for EWS, it is EWS = 1, RMS = NB = Lymph = 0). The perfectly classified sample would be plotted with a distance of 0. The histological diagnosis of four different cancer categories was displayed in shape as diamond for EWS, square for RMS, triangle for NB, and circle for lymphoma. All samples were correctly classified except one RMS sample, which was misclassified as EWS.

Figure 3

Hierarchical clustering and multidimensional scaling analysis. A: Gene minimization plot for ANN prediction. All of 39 genes were used for the analysis of 96 samples. ANNs were first trained using 96 samples, and 39 genes were ranked according to their importance to the ANN prediction. Red arrow marked the position of 31 genes. B: Multidimensional scaling analysis using 31 top-ranked genes. Three dimensions of the multidimensional scaling plot are shown. EWSs are depicted as yellow circles, RMS as red, NB as green, and lymphoma as blue. The samples clustered closely according to the four different cancer categories. C: Hierarchical clustering of all 96 samples and 31 top-ranked genes. Each row represents a gene, and each column, a separate sample. A pseudocolored representation of the ratio (log2-transformed and z-scored across the samples) is shown. On the right are the gene symbols of 31 genes as well as the ANN gene rank.