Cluster analysis for histological risk grade clearly distinguishes between Wilms tumors of intermediate and high risk histology (Fig. 1a). The stratification was performed according to the revised Stockholm–Working-Classification.16 It should be stressed that therapeutic regimens in the European SIOP study mandate preoperative chemotherapy in most cases, which is different from the American NWTS study. It has been well documented that this leads to a quite different distribution of histological subtypes of Wilms tumor and concomitantly altered clinical course.17 Thus, it will remain open as to what extent the expression changes found in our patient cohort are different from those in the North American study.
The high risk tumor group comprises four predominantly blastemal and four anaplastic tumors. Despite histological differences between these high risk tumors they seem to have a common molecular basis that sets them apart from the intermediate risk group. The intermediate risk group comprises 52 Wilms tumors. Half of these were “classic” triphasic Wilms tumors presenting with a mixture of epithelial, stromal and blastemal cell types. The other half consists of predominantly stromal or epithelial and of regressive tumors. A partial clustering of histological subtypes can be seen, especially for regressive Wilms tumors in the left third of the cluster picture. Nevertheless, the separation of intermediate risk histologies into several subclusters lacks a clear clinicomorphologic correlate. This suggests that there may be additional underlying factors that are not covered by current diagnostic means. Part of this may also be due to the fact that the tumors generally display histological heterogeneity to different degrees, and thus, rather represent a continuum of histological subtypes with one often being predominant.
Relapse and survival
The cluster comparison between relapse-free tumors and those that later relapsed is based on 77 genes with lowest adjusted p-values. These genes clearly distinguished between the two groups (Fig. 1b). Only tumors with a relapse-free follow-up of at least 3 years were included in the first group. Expression data for survival overlap with those for relapse because most deaths in children with Wilms tumor occur as a consequence of relapse. More than half of the genes selected for survival were concurrently selected for relapsed tumors. In relapsed Wilms tumors, more genes with a lower FDR were selected due to the larger tumor cohort available in comparison to the stratification for survival.
An association between gene expression and prediction of outcome has recently been described in several other human cancers.18 Studies on breast cancer have delivered the most extensive and informative prognostic profiles.19 These studies suggest that the predictive power of microarray approaches may be even greater than that of other known predictive markers. However, the overall positive prediction of tumors rarely exceeded 60%. Here, we show that in Wilms tumors expression analysis may also help to predict tumor recurrence and outcome in future. Up to now, the risk for relapse in Wilms tumors is estimated from clinical and histological parameters and this stratifies the extent of postoperative therapy.16 Only 4 of the 10 relapsed tumors had high risk histology, whereas the majority of relapses occurred in tumors with intermediate risk histology. The stratification based on microarray data for relapsed vs. relapse-free Wilms tumors (Fig. 1b) points at a potential predictive value of expression data, in addition to histology. At this point, the number of analyzed Wilms tumors is definitely too small to calculate overall prediction rates. Prediction studies need both larger training and validation groups. However, expression profiling appears to deliver additional prognostic information, which may help to guide future research and to further optimize treatment of affected children.
Evaluation of other clinical features
Interestingly, cluster analysis for the stratification of tumors with and without metastasis was not unequivocal. The propensity of tumor cells to spread and colonize more distant sites is not easily identifiable in our tumor set. It may well be that the metastatic potential of individual subclones of tumor cells may get lost in a global analysis that is based on a comparatively large piece of tumor tissue. This is supported by the fact that in the literature metastasis-associated gene expression profiles are only available for comparisons between cell lines with different metastatic potential or genuine comparisons of primary tumor and metastases.
For the stratification of the clinical criterion response to chemotherapy, almost no expression difference could be detected. Response to preoperative chemotherapy was determined by measuring pre- and postchemotherapeutic tumor volumes. Tumor reduction of more than 50% was defined as good response (38 children), whereas tumor reduction of less than 50% or increase of tumor volume despite preoperative chemotherapy was defined as poor response (19 children). The lack of clear differentiating expression patterns was somewhat surprising in this case. Strongly hit tumors with massive apoptosis and large numbers of secondary cell types should be quite different from tumors that continue to grow almost unaffected by chemotherapy. However, it is possible that response to chemotherapy is controlled by relatively few genes that may not be represented on our microarrays.
Candidate genes for Wilms tumor development and progression
To date, 6 publications reported on microarray expression screening in Wilms tumors.20, 21, 22, 23, 24, 25 The aim of several of these studies, most of which are based on rather small numbers of Wilms tumor samples, was to identify genes differentially regulated in Wilms tumors compared to normal kidney tissue. Two recent studies sought to correlate expression profiles to clinical features. The study by Williams et al.22 compares 27 Wilms tumors with favorable histology, half of which later relapsed. The authors identified genes with potential biological interest, but state that their data are not useful for prediction. Probably due to the fact that they used prechemotherapy samples, there is only little overlap in genes associated with relapse compared to our study. The second recent study by Li et al.21 compares anaplastic tumors to tumors with favorable histology and additionally contains an overall comparison of Wilms tumors to normal kidney. Compared to our study, several genes were concordantly identified: CENPF, CCNA1, CDC2 (anaplastic vs. favorable histology) and EZH2 (Wilms tumor vs. fetal kidney). The first group of genes deregulated in advanced Wilms tumors is involved in cell cycle progression. The only difference between samples included in our study and that of Li et al. is the therapeutic approach: Wilms tumors analyzed by Li et al. were all primarily operated according to the American NWTS protocol, whereas tumor samples included in our study were derived from children who were treated with preoperative chemotherapy according to the European SIOP protocol. Since these genes were concordantly deregulated in both studies, they seem to be involved in tumor progression independent of the therapeutic approach.
Compared to previous expression studies on Wilms tumors, we analyzed a much larger set of 63 preoperatively treated tumors and considered several clinical criteria for the evaluation of expression data. The aim of our study was to assess relative expression differences between these tumors and to search for genes affecting tumor progression. To investigate whether certain groups of genes may be over-represented among differentially regulated genes, we performed EASE analysis of different gene lists. Among genes that were deregulated in relapsed tumors, those involved in cell cycle progression and mitosis were significantly over-represented. BUB1, CENPF, CENPE, CDC6, CKS2, CUL5, ESR1, UBE2C, MAD2L1, MAD2, CHEK1 and STK6 are examples of significantly (EASE score < 0.05) over-represented mitosis-associated genes in relapsed Wilms tumors. In tumors with high-risk histology that were compared to standard-risk tumors, we found a slightly different picture with significant over-representation of defense and immune response genes, e.g.C3AR1, TREM2, CSF1R, BF, IL16, HLA-C, IL1R1, and genes coding for extracellular region proteins, e.g.COL15A1, ELN, MMP2, NID2, SFRP4 and OSF2. Although high-risk histology is the best predictor for later relapse, it is intriguing that these two criteria appear to be characterized by rather different sets of biological processes, i.e. mitosis vs. defense.
Previous studies have proposed further genes with prognostic relevance in Wilms tumors, e.g. members of the INK4 family,26TERT,27VEGF and its receptor FLT1,28, 29DKK1,30TrkBfull,31TGFA and EGFR.32 In our study, however, no significant regulation of these genes could be found in any stratification of clinical features in Wilms tumors. The expression differences reported before may be due to the small number of tumors included in most of these studies. Furthermore, it may generally be rather difficult to predict outcome in Wilms tumors by the assessment of single genes. The range of single gene expression values measured by microarray and qPCR approaches in relapsed tumors also supports this assumption (data not shown). We therefore speculate that only a combination of expression data from a set of different genes may represent a reliable new approach for prediction of outcome in Wilms tumors.
The retinoic acid and E2F pathways as potential new targets
Interestingly, our analysis identified several retinol-related genes belonging to the retinoic acid receptor responder family. RARRES2 was found to be 2.2-fold downregulated in high malignant tumors, while RARRES3 was 2-fold downregulated in relapsed tumors and in tumors that led to death of the affected children.
Another retinoic acid induced gene, CTGF (connective tissue growth factor; also known as IGFBP8), has been identified in a screen for WT1-induced genes.33 Compared to normal kidney tissue, CTGF was overexpressed in a subset of Wilms tumors. We found CTGF to be downregulated in the evaluation of relapse and survival (2.9- and 3.8-fold downregulation, respectively), which represent important criteria for the prediction of prognosis. It is conceivable that CTGF is activated in nephrogenesis and early tumorigenesis, while its expression decreases with tumor progression.
Additionally, NK4 (natural killer cell transcript 4), RAMP (RA-regulated nuclear matrix-associated protein) and ENPP2 (Autotaxin) represent genes regulated by retinoic acid,34, 35 and they were also found to be strongly deregulated in relapsed Wilms tumors. Therefore, it appears that the retinoic acid pathway can be deregulated at different levels. We have since tested whether retinoic acid could be employed as a novel therapeutic agent in Wilms tumors by exposing cultured tumor cells.36 Our data suggest that retinoic acid may represent a novel therapeutic approach to treat tumors with evidence for impaired retinoic acid signaling.
Another pathway potentially involved in Wilms tumor progression is the RB-E2F pathway. Members of the E2F transcription factor family are important targets of the retinoblastoma (RB) pathway and they regulate transcription of a number of genes that control cell cycle progression.37EZH2 (enhancer of zeste homolog 2) is a direct target of E2F transcription factors.38 Overexpression of EZH2 and involvement in tumor progression was shown in different human cancers, e.g. breast cancer.38 Takahashi et al.20 and Li et al.21 found EZH2 overexpressed in Wilms tumors with favorable histology compared to normal kidney tissue. Our results confirm the involvement of EZH2 in Wilms tumorigenesis and additionally imply that EZH2 acts as a tumor progression factor since its expression is upregulated in recurrent Wilms tumors. In addition to EZH2, several other genes are known to be controlled by E2F39 and 4 of these were highly overexpressed in recurrent vs. nonrecurrent Wilms tumors (CDC6, CDC2, MYCN, DHFR). These results suggest an important role of the RB-E2F pathway in Wilms tumor formation and progression.
In summary, we identified several candidate genes involved in Wilms tumor development and progression. Since expression profiling harbors the potential to predict prognosis, which has already been shown in several other cancer types, the genes identified in relapsed Wilms tumors are of special interest. These differentially expressed candidate genes can now be screened in extended tumor cohorts to corroborate our findings. Our goal is to select a minimal set of genes whose expression can be used to identify patients at risk for relapse who may benefit from intensified therapeutic regimens or enhanced surveillance. Such screening procedures could be implemented in future study protocols as prospective prognostic studies. From a basic science perspective it will be very interesting to identify gene expression in individual cell types by mRNA in situ hybridization or antibody staining to better characterize their individual contribution to Wilms tumor development.