PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
 
Clin Cancer Res. Author manuscript; available in PMC 2010 March 30.
Published in final edited form as:
PMCID: PMC2847352
NIHMSID: NIHMS172559

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.

Wilms tumor is the most common urogenital malignancy in children, with ~500 new cases per year in North America. Several national and international cooperative group clinical trials have optimized the therapy resulting in an increase in the overall survival rate to ~90%. The current therapeutic approach for Wilms tumor is based on histologic subtype (favorable versus unfavorable histology) and tumor stage (1). The majority of Wilms tumor has favorable histology, defined as the absence of anaplasia, and these represent the focus of the current study. Patients with anaplasia are treated differently than those with favorable histology Wilms tumor (FHWT) and are beyond the scope of this study.

In recent years, the improvement in relapse-free and overall survival for FHWT at each stage has reached a plateau. Some patients are not successfully treated initially, resulting in relapse and less frequently death. Of equal importance, many patients may receive more therapy than needed; this is particularly true for patients with stage III disease (2, 3). Further improvements in outcome will rely in part on the ability to identify markers associated with relapse, with the hope of better stratifying patients. This goal represented a major focus of the National Wilms Tumor Study-5 clinical protocol, which included a large-scale effort aimed at tumor banking and molecular analysis. These efforts showed that loss of heterozygosity (LOH) for both chromosomes 1p and 16q was associated with poor outcome (4). However, LOH is able to detect only a very small subset of FHWT patients who have an increased risk of relapse and death. Additional efforts are therefore required to further define markers of relapse. In this study, we analyzed gene expression patterns to identify such markers and to investigate the feasibility of developing classifiers able to predict patients at high risk for relapse.

Translational Relevance

This article evaluates gene expression signatures to predict relapse in patients registered on the National Wilms Tumor Study-5 cooperative group protocol using stage and treatment-specific analyses. This will enable independent validation using samples from patients registered in the ongoing Children's Oncology Group protocols. Successful signatures will be able to be used for therapeutic stratification during protocols estimated to open in 2012. Signatures with 50 genes were associated with relapse in stage III tumors (sensitivity of 47% and specificity of 70%). Existing markers for relapse currently used for stratification (1p and 16q loss of heterozygosity) have a sensitivity of 8% and specificity of 96%. Analysis of specific genes associated with relapse revealed apoptosis,Wnt signaling, and the insulin-like growth factor pathway to be important. These pathways will be separately validated at the protein level within the current protocol. Importantly, all the above-identified pathways have been previously targeted for developmental therapies in the current literature. Two additional potential therapeutic targets identified in this study were FRAP/MTOR and CD40. MTOR inhibitors and anti-CD40 antibodies are currently used in clinical trials. The Children's Oncology Group Renal Tumor Committee has an active Developmental Therapeutics Subcommittee that is able to bring such promising agents forward.

Materials and Methods

Patient samples

On June 25, 2003, all 1,451 patients with FHWT registered on National Wilms Tumor Study-5 from August 1995 to June 2002 who had available pretreatment tumor tissues were identified. Six hundred patients, consisting of all those known to have relapsed and a random sample including ~30% of the remainder, were randomly divided into two groups of 300 patients each (groups A and B), both enriched for relapse.

The National Wilms Tumor Study-5 protocol was approved by the review boards of institutions that registered patients. Parents or guardians signed informed consent for collection and banking of biological samples. Histologic diagnosis and local stage were confirmed by central review. Patients who did not receive the therapy consistent with the diagnosis and stage found on central review were rejected from the analysis. Chest X-ray, abdominal ultrasound, and computed tomography studies were required for staging. Tumor samples were obtained at initial diagnosis before the initiation of therapy and were snap frozen. A frozen section was evaluated to confirm at least 80% viable tumor cellularity.

Gene expression analysis

RNA was extracted and hybridized to Affymetrix U133A arrays, scanned, and subjected to quality-control variables according to the previously described protocol (5).

Quantitative reverse transcription-PCR

RNA levels of five genes showing a range of fold changes and overall expression levels were further analyzed by quantitative reverse transcription-PCR using TaqMan Gold and the ABI Prism 7700 Sequence Detection System (Applied Biosystems). β-Actin was used as the endogenous control. Primer and probe sequences are provided in Supplementary Table S1.8 Each threshold cycle (CT) was determined and the CT for β -actin was subtracted from this for normalization.

Data analysis

Positional-dependent nearest-neighbor model software9 was used to translate the scanned images into expression analysis files and to normalize the data across all arrays (6). Genes with maximum expression less than a log scale of 6 across all tumors and Affymetrix control genes were excluded, resulting in 20,931 probe sets for analysis. Support vector machine was chosen for relapse risk stratification (7). The following procedure was used to construct 150 different classifiers for each randomly drawn training set of tumors. For each gene, the t statistic comparing expression between cases and controls and the associated P value were calculated using the Welch method. The K genes (1-150 genes) with the smallest P values were selected to construct a n support vector machine model as developed by Chang and Lin10 and implemented in a R software package, e1071 (7). This model was then applied to categorize tumors within the complementary test set as having high or low risk for relapse.

Validation procedures

Cross-validations (2- and 10-fold) were used to investigate the ability of classifiers established in a training set to predict relapse in an independent test set. For 2-fold cross-validation, the data set was randomly divided 500 times into training and corresponding test sets of equal size, each including half the patients who relapsed. A classifier for relapse was identified for each training set and used to assign tumors in the corresponding test set to low-risk and high-risk categories. The training and test sets were then swapped. The number of top K genes in each classifier evaluated ranged from 1 to 150. Therefore, 150,000 different classifiers were developed, one for each value of K from 1 to 150 for each of the 1,000 (500 × 2) training sets. For 10-fold cross-validation, the data set was randomly divided 500 times into 10 groups of approximately equal size. Each group included approximately the same number of relapses. For each such group, a classifier was built with the remaining 9 of 10 of the samples and then used to categorize tumors in the group as low or high risk; the process was repeated until all tumor samples were categorized as low or high risk. For all the cross-validation procedures, to avoid gene-selection bias, classifiers were completely rebuilt in each cross-validation iteration (8).

Results

Patient information and quality control

Group A

Ninety of the 300 tumors from group A were rejected. Following verification of the stage, relapse status, clinical follow-up, and diagnosis, 1 tumor (with diffuse hyperplastic perilobar nephroblastomatosis) was excluded for diagnosis, 14 tumors with clinical follow-up <3 years were excluded, 29 tumors were rejected due to <80% viable tumor cellularity, 34 tumors were rejected due to A260/A280 ratios < 1.8, and 12 tumors failed to meet Affymetrix-recommended hybridization quality-control standards. Illustrated in Table 1 is the stage and relapse status of group A. The tumors were categorized based on local (abdominal) and overall stage. Stage I tumors (49) were confined to the kidney. Stage II tumors (73) showed infiltration beyond the kidney with negative surgical margins and regional lymph nodes; 6 of these patients had distant metastases (local stage II, overall stage IV). Of 76 local stage III patients, 12 presented with distant metastasis (local stage III, overall stage IV). Six patients presented with bilateral disease (stage V) and 6 unilateral tumors were biopsied and treated before the subsequent removal of the kidney following therapy (“biopsy only”).

Table 1
Categorization of 300 samples in group A by stage

This study grouped together for analysis patients who received the same therapy. First, there were 10 stage I patients (3 relapses) registered on a treatment arm within National Wilms Tumor Study-5 that included patients ages ≤24 months with stage I disease whose kidney and tumor had a combined weight of ≤550 g; these patients were treated without initial chemotherapy as reported previously (9). These patients are not further considered in this particular study due to their small numbers. The remaining patients with stage I and those with overall stage II disease were treated with nephrectomy, vincristine, and dactinomycin (106 patients, 32 relapses); they formed a second group referred to below as stage I/II tumors. The third group, referred to collectively as stage III tumors, includes those patients with local and overall stage III disease (64 patients) and those with local stage III, overall stage IV disease (12 patients). Careful consideration was given to the inclusion of patients with distant metastasis (stage IV) to this group due to their potential biological differences. Because the goal of the study was to define clinically applicable signatures, it was considered to be optimal to group all patients receiving the same therapy. Patients with stages III/III and III/IV disease both receive nephrectomy, vincristine, actinomycin, doxorubicin, and abdominal radiation therapy. In addition, both stages III and IV tumors have known or presumed residual disease, and one of the most significant causes for relapse is resistance to chemotherapy, which does not change with stage. To ensure that the stage IV patients did not significantly alter the results, all analyses of the stage III tumors were also performed, leaving the patients with overall stage IV disease out. Six patients with local stage II, overall stage IV disease were excluded from the stage I/II group because they received additional adjuvant therapy; they were also excluded from the stage III group because they did not receive abdominal radiation. Six tumors biopsied and treated before removal of the kidney were not included nor were 6 patients with bilateral disease (stage V) due to the individualized nature of their therapy. In total, gene expression analysis from 182 tumors from group A was used in these analyses.

Group B

The only patients analyzed from group B were the 99 appropriately diagnosed and treated stages III/III and III/IV tumors as defined above. The same quality-control variables outlined above were applied. Nine tumors were rejected following histology review, 5 had <3 years of follow-up, 8 had poor RNA quality, and 9 failed to meet quality-control variables following hybridization, leaving 68 tumors for analysis. Clinical stage and outcome for all 144 stage III tumors within the entire group of 600 tumors are provided in Table 2.

Table 2
Clinical characteristics of 144 stage III tumors

Classifiers that identify relapse risk: initial feasibility assessment

Group A was used to assess the feasibility of developing classifiers that predict risk of relapse within the above groups. To accomplish this, the expression of each probe set in the tumors that relapsed was compared with the tumors that did not relapse using the t test, and a nominal P value was established for this association. The distribution of P values within the stage I/II tumors failed to show more probe sets with P < 0.05 than would be expected by chance alone, indicating that further investigation would be associated with high false discovery rates. Similar results were found with the separate analysis of stage II tumors alone. The distribution of P values for stage III tumors, however, showed a higher frequency of probe sets at low P values than would be expected by chance alone (1,445 probe sets with P < 0.05). This analysis was repeated excluding stage IV patients with similar results. This indicates that a gene signature associated with relapse may exist. The remainder of the study therefore focused on the stage III tumors.

Experimental verification

Five genes with P values < 0.05 were analyzed using quantitative reverse transcription-PCR. These included COL2A1, DNMT1, ENPP2, GCNT, and LY6G6D. In all cases, the gene expression data from array analysis and from quantitative reverse transcription-PCR were comparable (data not shown). Replicate analysis beginning with RNA extraction done on a subgroup of 10 tumors confirmed reproducibility of the expression patterns.

Developing and validating classifiers that predict relapse in stage III FHWT

The original design was to identify classifiers based on the association of gene expression with time to relapse using the 300 tumors in group A and to then apply these classifiers in a blinded fashion to tumors in group B. Following this design, a classifier was developed from stage III tumors in group A using the association between gene expression and time to relapse to select the top genes. This classifier was applied to stage III tumors within group B in a blinded fashion. Six of 10 tumors placed in the high-risk category relapsed and 23 of 63 tumors placed in the low-risk category relapsed, resulting in an odds ratio measure of association of 2.6 (P = 0.18, Fisher's exact test). Due to concerns regarding the small number of relapses and the use of time to relapse in addition to fact of relapse to select genes, we also analyzed the results using a case-control design including all stage III patients. Three years had elapsed during the laboratory analysis, allowing for extended follow-up for the 600 patients. Lastly, it is increasingly recognized that small sample sizes may hinder the reliability of gene classifiers, particularly those for which outcome prediction is the goal (10). Therefore, for the remaining analyses, a case-control design was applied. All stage III patients within both groups A and B who were known to have relapsed as of February 2007 were defined as cases; those who had not relapsed and who had been followed >3 years were defined as controls. The risk of relapse after 3 years is low, with only 3 of 213 (1.4%) patients relapsing beyond 3 years in National Wilms Tumor Study-5 (4).

The 144 stage III tumors from groups A and B include 53 relapses (cases) and 91 nonrelapses (controls; Table 2). The distribution of P values for all 144 tumors again showed a higher frequency of probe sets at low P values than would be expected by chance alone (2896 probe sets with P < 0.05). (When the stage IV tumors were left out, 2,259 probe sets had a P < 0.05.) Cross-validations (2- and 10-fold) were repeated 500 times, varying the number of genes in the classifiers from 1 to 150. The specificity, sensitivity, and overall error rate were determined for each classifier and are illustrated in Fig. 1. The error rate improved markedly when the number of genes in the classifier increased from 5 to 50 but only very slowly with increasing number of genes thereafter. With 10-fold cross-validation, when 50 genes were used in the classifier, the median sensitivity was 47% (range, 35-58%), the median specificity was 70% (range, 60-78%), and the median overall error rate was 38% (range, 31-45.8%) in this mix of cases and controls. The 10-fold cross-validation analysis was repeated leaving the stage IV patients out, and the median sensitivity was 35% and the median specificity was 78%.

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 ...

Performance of classifiers within individual tumors

To evaluate the robustness of the classifiers when applied to individual tumor samples, we observed the 10-fold cross-validation analyses, selecting only the 50-gene classifiers. Every tumor was assessed 500 times for relapse risk using a 50-gene classifier and the success of every classification was tracked for each individual tumor. The percentage of classifiers that categorized each tumor as high risk is illustrated in Fig. 2. Fifteen (28%) relapsed tumors were consistently and correctly classified as high risk using a stringent cut-point requiring >90% of the classifiers to predict relapse (top dashed line). Similarly, 47 (52%) nonrelapsed tumors were consistently and correctly classified as low risk using a cut-point requiring <10% of the classifiers to predict relapse (bottom dashed line). However, a large number of tumors fell in between these cut-points, and 18 (34%) of the tumors that relapsed were consistently but incorrectly placed in the low-risk group even when using the least stringent cut-point of 10%.

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 ( ...

Genes associated with relapse in stage III FHWT

Using the 144 stage III tumors, 109 genes were identified with P < 0.001 and are listed in Supplementary Table S21 (false discovery rate of 11.6%; ref. 11). The entire gene expression data and description of the experiment using the MIAME format are available.11 Of the 20,931 genes analyzed, a total of 1,626 genes were represented in at least 1 of the 5,000 50-gene classifiers used in the 10-fold cross-validation: 1,004 genes were selected <10 times, 579 genes were selected >50 times, 244 genes were selected >500 times, and 55 genes were selected >1,000 times. The number of times each of the top 100 genes was used and its ranking within the 5,000 50-gene classifiers is provided within Supplementary Table S2. The P value correlated highly with the number of times each gene was selected.

Among the top 109 genes with P < 0.001, approximately one-quarter are known to function in metastasis, tumorigenesis, tumor progression, or tumor suppression; these are listed in Table 3. The chromosomal location of the 109 top genes (provided in Supplementary Table S2) was also analyzed to determine consistent areas of gain or loss of expression. No genes located on chromosomes 1p or 16q were in the top gene list and down-regulated. However, 16 of the top genes were located in the region 1p31-1pter and were up-regulated and 6 genes were located on distal 1q and were up-regulated in relapsed patients. These include CHC1, FRAP/MTOR, PTGER, and UBE4B from 1p and CDC42BPA and TRIM17 from 1q (Table 3).

Table 3
Categorization of selected genes of interest

Discussion

We report the gene expression analysis of a large group of FHWT consistently treated within a cooperative group setting. The goals were to determine the feasibility of using gene expression classifiers to identify patients who could benefit from increased or decreased adjuvant chemotherapy and to identify specific biological markers associated with relapse in patients with FHWT.

Ability to predict relapse in FHWT using gene expression varies with tumor stage

Efforts to establish an association between gene expression and relapse were not successful for stages I and II tumors yet were moderately successful for stage III tumors. The clinical endpoint of relapse in FHWT relies on two distinctly different features, the tumor's invasiveness (the ability to achieve lymphovascular access and distant growth, resulting in occult or overt residual disease) and the responsiveness of any residual disease to the therapy provided. The outcome of stage I FHWT <550 g in patients <24 months would suggest that most low-stage FHWT do not achieve lymphovascular access and are able to be completely treated by surgery alone. The majority of FHWT at all stages are exceedingly responsive to chemotherapy. As a result, our failure to detect gene expression signatures correlating with relapse in stage I to II patients treated with chemotherapy is not surprising. The majority of such tumors that are biologically “high risk” due to unresponsiveness to therapy are completely resected and the majority of those that are not completely resected (due to occult metastases, for example) are responsive to therapy. Conversely, stage III tumors by definition have evidence of residual disease. Therefore, the gene expression profile of the residual disease would be expected to correlate with responsiveness to therapy. It should be noted that the current study analyzes a random sample of the tumor, which is only an imperfect approximation of the residual disease itself.

Best classifiers in stage III FHWT will only be able to detect a subset of high-risk tumors

Our analysis indicates that not all stage III tumors that will go on to relapse can be correctly and consistently classified as high risk by global gene expression using Affymetrix U133A arrays. There are several possible explanations, all of which likely play a role:

  • Global gene expression analysis performs well when detecting large differences in genes expressed at moderate to high levels. It is more difficult to detect reliable expression differences with genes that are tightly regulated with small differences in expression or genes that are expressed and regulated at very low levels.
  • Drug resistance may develop during the administration of chemotherapy and the causes of this may not be detectable in the original sample.
  • Clonal evolution may result in a more invasive or less responsive tumor clone. If this clone is not sampled, the tumor will not be correctly classified.
  • Significant heterogeneity for the intrinsic causes of relapse likely exists, requiring a very large number of samples to define each cause.
  • Wilms tumors have a wide spectrum of histologic appearances and varied lines of differentiation. The resulting spectrum of gene expression may obscure the identification of genes specifically associated with relapse.
  • Some stage III tumors will have margins that are technically positive, but with no residual viable disease, or will have positive regional lymph nodes, with completely resected tumor. Therefore, it is expected that some stage III tumors would be identified based on gene expression as high risk that would not relapse.

Clinical utility of classifiers predicting relapse in stage III FHWT

In order for a biological marker to be clinically useful, it must be sufficiently robust and must have acceptable sensitivity and specificity to address the particular clinical needs of the patients being addressed. In this study, the classifiers developed using 50 genes resulted in a median sensitivity of 47% (25 of 53) and specificity of 70% (64 of 91). This compares with a sensitivity of 8% and specificity of 96% for LOH of both 1p and 16q in stages III and IV FHWT (4). To put this information into further perspective, the sensitivity and specificity of anaplasia for predicting death in Wilms tumor are 30% and 96%, respectively (12). Although the data suggest that the reliability of classifiers based on gene expression will be limited, strategies can be envisioned that propose either increased adjuvant therapy for those tumors at high risk or decreased adjuvant therapy for those tumors at low risk.

This study represents the first of a multistep process. The second step is the identification of a specific set of genes and the establishment of a mathematical rule that will be optimal for prospective prediction of individual tumors. This gene selection process may use the sensitivity and specificity of each individual gene, the overall absolute expression levels and differences in median expression levels (variables that take into account robustness of expression), the biological functions of the genes, and the interactions of expression within the top genes. The final and most critical step is the validation of the proposed classifier when applied in a blinded fashion to an independent patient population. This process is not currently possible and will require the analysis of patients in the current therapeutic protocols. In the absence of an ability to perform an independent validation currently, the proposal of a specific classifier may result in erroneous and premature conclusions. Therefore, we have chosen not to propose a specific classifier.

Comparison of gene expression data with prior publications

Three publications have reported differences in gene expression associated with clinical outcome in Wilms tumor (1315). Zirn et al. analyzed 67 samples of all stages and of favorable and unfavorable histology collected following preoperative chemotherapy (13). Of their top genes associated with relapse, 3 were concordantly regulated and found in our Supplementary Table S2: IL11RA, POSTN, and SMPDL3A. IL11RA and SMPDL3A have been shown to be up-regulated in common adult cancers (16, 17). Both genes were down-regulated in FHWT that relapsed. POSTN, periostin, is up-regulated in many cancers and may activate the Akt/protein kinase B pathway and promote anchorage-independent growth (18, 19). Conversely, other studies have shown POSTN to be down-regulated in several human tumors and to inhibit metastasis and anchorage-independent growth (20). POSTN was down-regulated in relapses in FHWT. Williams et al. compared 27 pre-chemotherapy FHWT samples of all stages, of which 13 relapsed (14). None of their 15 top genes were identified in our Supplementary Table S2. Li et al. used a custom cDNA array and analyzed 26 pre-chemotherapy tumor samples (12 relapses) of all stages and of both favorable and unfavorable histology (15). Four top genes were reported that were capable of identifying tumors that relapsed: C/EBPB, p21, H4FG, and cDNA CF542255 (CLK1). The P values of the associations of these genes with relapse in our study were all >0.1. In summary, the available current literature associating gene expression with relapse in Wilms tumor includes studies with small sample sizes and significant clinical heterogeneity. Our study has the advantage of increased numbers as well as the analysis of only stage III tumors of favorable histology treated using a single chemotherapeutic regimen.

Natrajan et al. performed array-based comparative genomic hybridization analysis to define regions of genetic gain or loss that correlate with relapse in 76 FHWT of all stages, half of which relapsed (21). Genes within these chromosomal regions that were concordantly differentially expressed with P < 0.01 in our study are listed in Table 4. (No genes meeting these criteria were found within regions 13q31-13q33, 14q32, 18q21, 21q22, and 1q32.) Several genes were found to be both gained and overexpressed on chromosome 1q. ADAM15 may function in metastasis (22), TGFB2 has an established role in tumorigenesis and cancer progression (23), CDC42BPA (MRCK) is a protein kinase implicated in tumor cell invasion (24), and SETDB1 and SMYD2 are methyltransferases up-regulated in relapses, and the latter suppresses the proapoptotic action of p53 (25). Genes gained and overexpressed in the 16p13.2-13.2 region include MAPK8IP3 whose up-regulation is associated with cell invasion and brain tumor malignancy (26). Lastly, 4 genes in 12q24.13-24.31 were down-regulated in our analysis including DIABLO, involved in apoptosis as described further below.

Table 4
Genes concordantly differentially expressed within reported genetic regions of gain or loss in relapses

Specific genes involved in relapse of stage III FHWT

Those genes within Supplementary Table S2 with known or speculated biological function are provided in Table 3. There are several genes that merit special attention. PEG3 (down-regulated in FHWT relapse) is induced after DNA damage by a p53-dependent mechanism and contributes to the Bax translocation necessary for apoptosis (27). PEG3 is silenced by methylation in several gynecologic cancer cell lines (28). DIABLO, likewise down-regulated in FHWT relapse, is released from the mitochondrial membrane into the cytosol during apoptotic signaling, where it contributes to caspase activation (29). Loss of function mutations and down-regulation of DIABLO are associated with tumor aggressiveness and drug resistance (30, 31). Because DIABLO, PEG3, and SMYD2 (mentioned earlier) all operate in the p53/Bax apoptotic pathway, this pathway may be of significance in relapse of FHWT. TUSC3 is a potential tumor suppressor gene with decreased expression in advanced ovarian cancer (32). As with PEG3, down-regulation of TUSC3 occurs via methylation in glioblastoma (33). RECK acts as a suppressor of tumor invasion and metastasis by inhibiting matrix metalloproteinases (34). Down-regulation of RECK in colon cancer via promoter methylation results in increased cell invasion, and these actions are reversed by DMNT inhibition (35). Therefore, down-regulation by methylation of several top genes in our study and up-regulation of two xmethyltransferases (SETDB1 and SMYD2) suggest that epigenetic changes may play a role in outcome within stage III FHWT.

Several biologically significant genes were up-regulated with relapse in our study. FRAP1/MTOR is a serine/threonine kinase involved in signal transduction, translation initiation, and elongation. It exerts cell growth and survival effects via the Akt/PTEN pathway. FRAP1 has a well-documented role in cancer (reviewed in ref. 36) and is a promising therapeutic target (37). Another up-regulated gene that is a potential therapeutic target is CD40, a member of the tumor necrosis factor family. CD40 induces antiapoptotic genes (including Bcl-2 and its family members), early angiogenesis, and activation of cell proliferation signaling pathways (38). Both BCL2 and BLC2L11, a Bcl-2-like gene, were significantly up-regulated in our data (P = 0.008 and 0.0001, respectively). The up-regulation of CD40 and Bcl-2 provides further support to the importance of apoptosis in FHWT relapse. LY6D, a GPI-anchored protein involved in cell-cell adhesion, is associated with high-risk lung cancer and with colorectal cancer invasion (39, 40).

FGF18, up-regulated in FHWT relapse, is also up-regulated in colon cancer (41). FGF18 is regulated at the transcriptional level by TCF-CTNNB1 responsive elements in its promoter region (41). FGF18 induces nuclear accumulation of CTNNB1 by down-regulation of GSK3B activity at the protein level (42). CTNNB1 has been shown to undergo mutational activation in 10% to 15% of Wilms tumor, and there is a strong association between WT-1 mutation or deletion and CTNNB1 mutation (43). In addition, the putative tumor suppressor gene, WTX, which is mutated in ~30% of Wilms tumors, was recently shown to induce degradation of CTNNB1 (44, 45). Our data therefore suggest that aberrant Wnt signaling may be involved in FHWT relapse.

Insulin-like growth factor (IGF)-I and its receptor, IGFR1, have been inversely associated with WT-1 in studies showing repression of IGFR1 by WT-1 and repression of WT-1 by IGF-I (46). Copy number gain at chromosome 15q23.5 (the location of IGFR1) has been shown to correlate with FHWT relapse, and gain of expression of IGFR1 was confirmed by reverse transcription-PCR (47). Analysis of our expression data showed no significant increases within relapse samples of IGFR1, IGFR2, or IGF2, whereas IGF-I was significantly down-regulated in association with relapse. Although high levels of IGF-I have been implicated in oncogenesis in several different cancers, exogenous IGF-I was reported to induce apoptosis in a Wilms tumor cell line (48), suggesting that IGF-I may in certain situations induce cell death. Therefore down-regulation of IGF-I in relapses, or up-regulation in nonrelapses, may affect cell survival.

LOH of 1p and of 16q have been associated with relapse in FHWT (4). Only 3 of the 53 stage III relapses in our study were predicted as high risk by LOH of both 1p and 16q, consistent with the reported data that LOH detects only a small subset of relapses in FHWT. Rather than provide support for the association of loss of expression on 1p with relapse, our data show that 16 of the top 100 genes were located on 1p and were overexpressed, including several biologically important genes such as PTGER3, UQCRH, and FRAP1. It has been suggested that the critical event resulting in 1p LOH may involve gain of 1q rather than loss of 1p (49). This hypothesis is supported by Natrajan et al., as discussed above, which showed increased copy number of 1q to be associated with relapse (21). Our data provide additional support of this hypothesis by showing 6 of the top genes to be located on 1q, all of which were overexpressed.

In conclusion, this study provides some evidence supporting an association between gene expression and relapse in tumors of patients with stage III FHWT. The sensitivity of 47% and specificity of 70% moderate enthusiasm for direct clinical translation of a classifier based on gene expression, although this remains a potential interest if validation on an independent patient population is successful. Several different hypotheses are raised or further supported by these data and merit further investigation in the attempt to find potentially more robust methods of stratifying patients with FHWT. These include the relationship between relapse and apoptotic pathways, IGF-I signaling, the Wnt/β-catenin pathway, 1q gain, and epigenetic modification.

Supplementary Material

Supplementary Table 1

Supplementary Table 2

Acknowledgments

Grant support: NIH grants U10CA42326 (N. Breslow, D.M. Green, and E.J. Perlman), U10CA98543 (J.S. Dome, P.E. Grundy, and E.J. Perlman), and UO1CA88131 (E.J. Perlman).

Footnotes

Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).

Disclosure of Potential Conflicts of Interest: No potential conflicts of interest were disclosed.

8Supplementary tables can be found at http://www.childrensmrc.org/perlman/.

9http://odin.mdacc.tmc.edu/~zhangli/PerfectMatch/

10http://www.csie.ntu.edu.tw/~cjlin/libsvm

11http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?token=pnknvysekesyuly&acc=GSE10320

References

1. Green DM. The treatment of stages I-IV favorable histology Wilms tumor. J Clin Oncol. 2004;22:1366–72. [PubMed]
2. Breslow NE, Ou SS, Beckwith JB, et al. Doxorubicin for favorable histology, stage II-III Wilms tumor: results from the National WilmsTumor Studies. Cancer. 2004;101:1072–80. [PubMed]
3. Grundy RG, Hutton C, Middleton H, et al. Outcome of patients with stage III or inoperable WT treated on the second United KingdomWT protocol (UKWT2); a United Kingdom Children's Cancer Study Group (UKCCSG) study. Pediatr Blood Cancer. 2004;42:311–9. [PubMed]
4. Grundy PE, Breslow NE, Li S, et al. Loss of heterozygosity for chromosomes 1p and 16q is an adverse prognostic factor in favorable-histology Wilms tumor: a report from the National WilmsTumor Study Group. J Clin Oncol. 2005;23:7312–21. [PubMed]
5. Huang CC, Cutcliffe C, Coffin C, Sorensen PH, Beckwith JB, Perlman EJ. Classification of malignant pediatric renal tumors by gene expression. Pediatr Blood Cancer. 2006;46:728–38. [PubMed]
6. Zhang L, Miles MF, Aldape KD. A model of molecular interactions on short oligonucleotide microarrays. Nat Biotechnol. 2003;21:818–21. [PubMed]
7. Vapnik VN. Statistical learning theory. New York (NY): John Wiley and Sons; 1998.
8. Simon R, Radmacher MD, Dobbin K, McShane LM. Pitfalls in the use of DNA microarray data for diagnostic and prognostic classification. J Natl Cancer Inst. 2003;95:14–8. [PubMed]
9. Green DM, Breslow NE, Beckwith JB, et al. Treatment with nephrectomy only for small, stage I/favorable histology Wilms tumor: a report from the National Wilms Tumor Study Group. J Clin Oncol. 2001;19:3719–24. [PubMed]
10. Michiels S, Koscielny S, Hill C. Prediction of cancer outcome with microarrays: a multiple random validation strategy. Lancet. 2005;365:488–92. [PubMed]
11. Storey JD. A direct approach of false discovery rates. JR Stat Soc Ser B. 2002;64:479–98.
12. Beckwith JB, Palmer NF. Histopathology and prognosis of Wilms tumors: results from the First National WilmsTumor Study. Cancer. 1978;41:1937–48. [PubMed]
13. Zirn B, Hartmann O, Samans B, et al. Expression profiling ofWilms tumors reveals new candidate genes for different clinical parameters. Int J Cancer. 2006;118:1954–62. [PubMed]
14. Williams RD, Hing SN, Greer BT, et al. Prognostic classification of relapsing favorable histology Wilms tumor using cDNA microarray expression profiling and support vector machines. Genes Chromosomes Cancer. 2004;41:65–79. [PubMed]
15. Li W, Kessler P, Yeger H, et al. A gene expression signature for relapse of primary Wilms tumors. Cancer Res. 2005;65:2592–601. [PubMed]
16. Hanavadi S, Martin TA, Watkins G, Mansel RE, Jiang WG. Expression of interleukin 11and its receptor and their prognostic value in human breast cancer. Ann Surg Oncol. 2006;13:802–8. [PubMed]
17. Wright KO, Messing EM, Reeder JE. Increased expression of the acid sphingomyelinase-like protein ASML3a in bladder tumors. J Urol. 2002;168:2645–9. [PubMed]
18. Kudo Y, Ogawa I, Kitajima S, et al. Periostin promotes invasion and anchorage-independent growth in the metastatic process of head and neck cancer. Cancer Res. 2006;66:6928–35. [PubMed]
19. Bao S, Ouyang G, Bai X, et al. Periostin potently promotes metastatic growth of colon cancer by augmenting cell survival via the Akt/PKB pathway. Cancer Cell. 2004;5:329–39. [PubMed]
20. Kim CJ, Yoshioka N, Tambe Y, Kushima R, Okada Y, Inoue H. Periostin is down-regulated in high grade human bladder cancers and suppresses in vitro cell invasiveness and in vivo metastasis of cancer cells. Int J Cancer. 2005;117:51–8. [PubMed]
21. Natrajan R, Williams RD, Hing SN, et al. Array CGH profiling of favourable histology Wilms tumours reveals novel gains and losses associated with relapse. J Pathol. 2006;210:49–58. [PubMed]
22. Kuefer R, Day KC, Kleer CG, et al. ADAM15 disintegrin is associated with aggressive prostate and breast cancer disease. Neoplasia. 2006;8:319–29. [PMC free article] [PubMed]
23. Bellone G, Carbone A, Tibaudi D, et al. Differential expression of transforming growth factors-β1, - β2 and -β3 in human colon carcinoma. Eur J Cancer. 2001;37:224–33. [PubMed]
24. Wilkinson S, Paterson HF, Marshall CJ. Cdc42-MRCK and Rho-ROCK signalling cooperate in myosin phosphorylation and cell invasion. Nat Cell Biol. 2005;7:255–61. [PubMed]
25. Huang J, Perez-Burgos L, Placek BJ, et al. Repression of p53 activity by Smyd2-mediated methylation. Nature. 2006;444:629–32. [PubMed]
26. Takino T, Nakada M, Miyamori H, et al. JSAP1/JIP3 cooperates with focal adhesion kinase to regulate c-Jun N-terminal kinase and cell migration. J Biol Chem. 2005;280:37772–81. [PubMed]
27. Deng Y, Wu X. Peg3/Pw1promotes p53-mediated apoptosis by inducing Bax translocation from cytosol to mitochondria. Proc Natl Acad Sci U S A. 2000;97:12050–5. [PubMed]
28. Dowdy SC, Gostout BS, Shridhar V, et al. Biallelic methylation and silencing of paternally expressed gene 3 (PEG3) in gynecologic cancer cell lines. Gynecol Oncol. 2005;99:126–34. [PubMed]
29. Kim R. Recent advances in understanding the cell death pathways activated by anticancer therapy. Cancer. 2005;103:1551–60. [PubMed]
30. Bonavida B, Huerta-Yepez S, Goodglick L, Mizutani Y, Miki T. Can we develop biomarkers that predict response of cancer patients to immunotherapy? Biomarkers. 2005;10 1:S69–76. [PubMed]
31. Tirro E, Consoli ML, Massimino M, et al. Altered expression of c-IAP1, survivin, and Smac contributes to chemotherapy resistance in thyroid cancer cells. Cancer Res. 2006;66:4263–72. [PubMed]
32. Pils D, Horak P, Gleiss A, et al. Five genes from chromosomal band 8p22 are significantly down-regulated in ovarian carcinoma: N33 and EFA6R have a potential impact on overall survival. Cancer. 2005;104:2417–29. [PubMed]
33. Li Q, Jedlicka A, Ahuja N, et al. Concordant methylation of the ER and N33 genes in glioblastoma multiforme. Oncogene. 1998;16:3197–202. [PubMed]
34. Noda M, Oh J, Takahashi R, Kondo S, Kitayama H, Takahashi C. RECK: a novel suppressor of malignancy linking oncogenic signaling to extracellular matrix remodeling. Cancer Metastasis Rev. 2003;22:167–75. [PubMed]
35. Cho CY, Wang JH, Chang HC, Chang CK, Hung WC. Epigenetic inactivation of the metastasis suppressor RECK enhances invasion of human colon cancer cells. J Cell Physiol. 2007;213:65–9. [PubMed]
36. Hay N, Sonenberg N. Upstream and downstream of mTOR. Genes Dev. 2004;18:1926–45. [PubMed]
37. Faivre S, Kroemer G, Raymond E. Current development of mTOR inhibitors as anticancer agents. Nat Rev Drug Discov. 2006;5:671–88. [PubMed]
38. Dallman C, Johnson PW, Packham G. Differential regulation of cell survival by CD40. Apoptosis. 2003;8:45–53. [PubMed]
39. Lu Y, Lemon W, Liu PY, et al. A gene expression signature predicts survival of patients with stage I non-small cell lung cancer. PLoS Med. 2006;3:e467. [PMC free article] [PubMed]
40. Reichling T, Goss KH, Carson DJ, et al. Transcriptional profiles of intestinal tumors in Apc(Min) mice are unique from those of embryonic intestine and identify novel gene targets dysregulated in human colorectal tumors. Cancer Res. 2005;65:166–76. [PubMed]
41. Shimokawa T, Furukawa Y, Sakai M, et al. Involvement of the FGF18 gene in colorectal carcinogenesis, as a novel downstream target of the β-catenin/T-cell factor complex. Cancer Res. 2003;63:6116–20. [PubMed]
42. Katoh M, Katoh M. Cross-talk of WNT and FGF signaling pathways at GSK3β to regulate β-catenin and SNAIL signaling cascades. Cancer Biol Ther. 2006;5:1059–64. [PubMed]
43. Maiti S, Alam R, Amos CI, Huff V. Frequent association of β-catenin andWT1mutations inWilms tumors. Cancer Res. 2000;60:6288–92. [PubMed]
44. Rivera MN, Kim WJ, Wells J, et al. An X chromosome gene, WTX, is commonly inactivated in Wilms tumor. Science. 2007;315:642–5. [PubMed]
45. Major MB, Camp ND, Berndt JD, et al. Wilms tumor suppressor WTX negatively regulates WNT/β-catenin signaling. Science. 2007;316:1043–6. [PubMed]
46. Bentov I, LeRoith D, Werner H. The WT1 Wilms tumor suppressor gene: a novel target for insulin-like growth factor-I action. Endocrinology. 2003;144:4276–9. [PubMed]
47. Natrajan R, Reis-Filho JS, Little SE, et al. Blastemal expression of type I insulin-like growth factor receptor in Wilms tumors is driven by increased copy number and correlates with relapse. Cancer Res. 2006;66:11148–55. [PubMed]
48. Granerus M, Johannisson A, Ekblom P, Engstrom W. Insulin-like growth factors I and II induce cell death in Wilmss tumour cells. Mol Pathol. 2001;54:30–5. [PMC free article] [PubMed]
49. Lu YJ, Hing S, Williams R, Pinkerton R, Shipley J, Pritchard-Jones K. Chromosome 1q expression profiling and relapse in Wilms tumour. Lancet. 2002;360:385–6. [PubMed]