Genomic Profiling Maps Loss of Heterozygosity and Defines the Timing and Stage Dependence of Epigenetic and Genetic Events in Wilms' Tumors Free

Since 1972 when Knudson and Strong analyzed epidemiologic data from Wilms' tumor to develop the concept of tumor suppressor genes (1), research on this pediatric kidney tumor has continued to uncover important principles in cancer genetics. Early findings were the recognition of the Wilms' tumor-aniridia (WAGR) syndrome and its mapping to chromosome 11p13 (2-4) and the discovery of the WT1 tumor suppressor gene in that chromosome band (5-7). WT1 is mutated in ∼10% of Wilms' tumors overall, predominantly in the setting of WAGR and Denys-Drash syndromes but also in a small subset of sporadic Wilms' tumors (8-10).

WT1 mutations are rare in sporadic Wilms' tumors; instead, a second locus, called WT2 and mapping to chromosome 11p15.5, is the more common target of loss of heterozygosity (LOH). WT2 most likely corresponds to the closely linked and oppositely imprinted genes IGF2 and H19 in this chromosomal band. In the ∼45% of sporadic Wilms' tumors that have 11p15.5 LOH, there is a selective loss of maternal alleles and duplication of paternal alleles (reviewed in ref. 11). This results in biallelic expression of IGF2 mRNA and loss of expression of H19 RNA. In another large group of sporadic Wilms' tumors, both parental alleles are retained, but the maternal allele of H19 and its upstream insulator sequences become hypermethylated at CpG sites, leading to the same end point of biallelic IGF2 expression [loss of imprinting (LOI)] and H19 silencing. In the Beckwith-Wiedemann syndrome, a subset of affected individuals show gain of methylation at H19^mat and LOI of IGF2 in their somatic tissues, leading to somatic overgrowth and a strong predisposition to Wilms' tumor (12). Recently, rare Beckwith-Wiedemann syndrome cases were identified with germ-line microdeletions in the H19 upstream sequences (13), a genetic finding that further supports the equivalence of WT2 and IGF2/H19.

In addition to chromosome band 11p15.5, cytogenetic and LOH studies have implicated several other chromosomal regions in sporadic Wilms' tumors. One copy of the q-arm of chromosome 16 is sometimes lost, typically via unbalanced chromosomal translocations (see ref. 14 and references therein). The resulting 16q hemizygosity correlates positively with tumor anaplasia (14, 15). Other chromosomes previously implicated by cytogenetics and/or LOH include losses and gains of portions of chromosome 1 (16), abnormalities of chromosome 7, notably i(7p) in which the p-arm is lost and q-arm is reduplicated (17-21), and loss of chromosome 14q (22). To examine LOH at higher resolution and with complete genomic coverage and to gain insights into the relationship between genetic and epigenetic events in Wilms' tumors, we have carried out a genome-wide scan at high resolution in syndromic (WAGR or Denys-Drash) and sporadic Wilms' tumors using Affymetrix 10K single nucleotide polymorphism (SNP) chips. The results of this scan, supplemented by DNA methylation and cytogenetic analyses, provide insights into the mechanism and timing of genetic and epigenetic events in Wilms' tumors.

We applied probes made from Wilms' tumor and normal kidney DNAs to Affymetrix 10K SNP chips. Allele call rates in our series ranged from 74.7% to 99.6%, with the 74.7% call rate representing an outlier case (mean call rate, 94.9%; median, 96.3%). We first validated the SNP chip data by comparing the results for selected chromosomal regions with data from conventional microsatellite analysis, RFLP analysis, and DNA sequencing. We have genotyped previously many of the Wilms' tumors for a series of microsatellite and RFLP markers on chromosome 11p (23), and the LOH results for 25 of the 26 tumors included in both that previous study and the current one were concordant. The single discrepancy was accounted for by partial (mosaic) LOH, visible in microsatellite analysis but not detected by the SNP chips. Similar concordance was evident from comparing the current data with our prior microsatellite genotyping for markers on chromosome 16q (14), with a single discrepancy (1 of 25 cases), which was again accounted for by low-level mosaic LOH, detected by microsatellite analysis but not by the SNP chip. Additional independent validations of the SNP chip data for a small region of reduction to hemizygosity on chromosome 4q and for LOH of markers on chromosome 7p are shown in subsequent sections.

Among the 62 sporadic Wilms' tumors, matched tumor and normal DNA samples were available for 46 cases, with unpaired tumor DNA available from the remaining 16 cases. The boundaries of regions of LOH could be ascertained at high resolution for the cases with paired samples (Fig. 1, blue bars), but the high density of SNPs on the microarrays also allowed detection of large regions of allelic loss in the unpaired cases (Fig. 1, green bars). Such regions, representing segments of DNA greater than ∼15 Mb, are detected as contiguous stretches of >100 SNPs giving uniformly homozygous genotypes (AA and BB). We assessed specificity and sensitivity of this method by reexamining the data for our paired tumor/normal cases, treating these tumors as unpaired. This procedure indicated a moderately high sensitivity for LOH (67.4%), with a high specificity (false-positive rate of 4.5%) using this cutoff of >100 contiguous homozygous SNPs.

Examination of the euchromatic regions of all chromosome arms, except the Y-chromosome, revealed chromosome 11p as the most common site of allelic loss (Fig. 1). This finding is consistent with prior LOH mapping by us and others (23-25). These prior LOH surveys, and cytogenetics (26-29), have implied that Wilms' tumors are often genetically stable, and we therefore expected a low baseline rate of LOH on other chromosomes. As shown in Fig. 1, this baseline rate is in fact very low, with only 12 of the 41 euchromatic chromosome arms showing LOH at frequencies above 5%. The most prominent “secondary peaks” of LOH were on chromosome arms 11q, 7p, 16q, 14q, 17p, and 1p, in that order of frequency, with an additional interesting region of recurrent LOH on 4q (Fig. 1). The LOH on chromosome arm 11q reflected loss of alleles along the entire chromosome in some cases, likely driven by selection for loss of the distal p-arm, but the mapping also indicated an independent region of LOH on the q-arm of this chromosome (Fig. 1). As discussed below, an unexpectedly large percentage of Wilms' tumors, 12 (19%) sporadic cases in this series, showed no detectable LOH. Of these 12 cases, 9 were tumor/normal pairs, and the conclusion that these tumors lacked LOH is therefore based on the full resolution of the 10K chips.

In contrast to the findings in the sporadic Wilms' tumors, nine Wilms' tumors from WAGR or Denys-Drash syndromes did not show LOH in any of these specific secondary regions (Fig. 1). As predicted from prior work, these syndromic cases, which invariably contain WT1 mutations, showed stretches of homozygous SNPs (reflecting hemizygous germ-line deletions encompassing WT1) centered on band 11p13, whereas the sporadic Wilms' tumors showed a more distal epicenter of LOH at band 11p15 (Fig. 1). Consistent with the relative rarity of WT1 mutations in sporadic Wilms' tumors, only two of the sporadic tumors showed LOH restricted to band 11p13, and in one of these cases, we have confirmed previously the presence of a WT1 mutation (9).

LOH can occur via at least two different pathways: Mitotic recombination (crossing over) between the paternal and the maternal homologues of a given chromosome can produce homozygosity for alleles distal to the cross-over in one daughter cell, whose clonal progeny go on to populate the tumor. This pathway does not produce a net reduction in chromosomal copy number; rather, it leads to loss of one set of parental alleles and duplication of the other. This pathway is relevant to the 11p15.5 WT2 region, which contains the imprinted IGF2 gene, which is active only on the paternal homologue and which therefore undergoes a 2-fold increase in functional gene dosage when there is loss of maternal and reduplication of paternal chromosome 11p15.5 DNA. Alternatively, LOH can reflect a true reduction to hemizygosity (i.e., a net loss of DNA, as occurs with unbalanced chromosomal translocations, chromosomal deletions, and mitotic nondisjunctions).

Karyotypes from standard cytogenetic analysis were available for 22 of the sporadic Wilms' tumors in our series. The results combining SNP chip and cytogenetic data indicated that LOH involving the p-arm of chromosome 11 is invariably due to mitotic recombination producing no net reduction in DNA copy number: in each of nine Wilms' tumors with 11p LOH and successful karyotyping, we observed two copies of chromosome 11. In contrast, LOH for chromosome arms 7p and 16q typically reflected a net loss of DNA, with reduction to hemizygosity. In four successfully karyotyped tumors with 7p LOH by SNP chip analysis, we found physical loss of one copy of the p-arm of chromosome 7 in each case. Similarly, for two tumors with 16q LOH by SNP chips analysis and successful karyotyping, both showed physical loss of the q-arm of chromosome 16 resulting in hemizygosity. This net loss of DNA was attributable in some cases to the formation of unbalanced chromosomal translocations: isochromosome 7q and t(1;16), respectively; however, in other cases, we observed simple deletions of the relevant chromosome arms as illustrated for a Wilms' tumor with 7p LOH in Fig. 2.

The boundaries of the smallest regions of recurrent LOH defined in the series of sporadic Wilms' tumors are shown in Fig. 3. Most of these regions are consistent with prior studies, and the 10K SNP chip data highlight some additional areas that may contain tumor suppressor genes. Notably, five sporadic Wilms' tumors showed LOH encompassing chromosome 4q, and the SNP chip data, supplemented by higher resolution mapping at individual SNPs, define a novel 2.4-Mb minimal region of LOH (Figs. 3 and 4) that includes the CXXC4/IDAX gene, encoding an inhibitor of the Wnt signaling pathway (30-32). The results of RFLP analysis and direct sequencing shown in Fig. 4 provide an independent validation of the SNP chip data by confirming LOH in this region. We sequenced the entire coding region of CXXC4 in two of the tumors with LOH, including the case with the minimal deletion, and did not find mutations. Southern blotting with quantitation by phosphorimaging showed CXXC4 hemizygosity in these cases (Fig. 4B). This reduction in CXXC4 copy number may contribute to Wilms' tumor progression, but an additional nine genes are present in the minimal region of 4q24-q25 LOH (Fig. 4A), and these remain to be examined for mutations.

Several other small minimal regions of LOH on other chromosomes are suggested by the data in Fig. 3. As is true for the 4q24-q25 region, due to the overall genomic stability of Wilms' tumors, many of these regions are defined by single tumor cases. In the future, pooling of data from multiple independent case series analyzed by SNP chips may allow these regions to be validated.

Advanced clinical stage at presentation, particularly metastatic disease (stage IV), is an independent risk factor for treatment failure in Wilms' tumors (33), and local tumor spread (stage III) has been associated with a greater probability of local recurrence after treatment (34). Information concerning the clinical stage was available for 43 of the sporadic Wilms' tumors in our series. We plotted the number of chromosomal regions affected by LOH in each tumor as a function of stage and found that this number increased steadily with increasing stage (Fig. 5). Somewhat surprisingly, fully 60% of stage I Wilms' tumors (confined to the kidney at presentation) showed no LOH (mean value for number of regions affected by LOH per tumor at stage I is 1.4). In contrast, most stage IV tumors (metastatic to distant sites at presentation) contained three or more regions of LOH (mean, 3.9) and all of these advanced-stage tumors showed at least two regions of LOH. Because the transition from stage I-II to stage III-IV has prognostic importance for both local recurrence and treatment failure, we analyzed these data statistically by comparing the number of chromosomal regions with LOH in each tumor for the stage I-II cases versus stage III-IV and found a significant difference (P = 0.006, t test). Combining the stage I-III cases and comparing with stage IV also revealed a significant difference (P = 0.004, t test).

The chromosomal region individually showing the strongest statistical association of LOH with advanced clinical stage was 16q (stage I-III compared with stage IV; P = 0.037, Fisher's exact test). Consistent with the significant overall association of chromosomal deletions with advanced clinical stage, trends for increasing LOH with increasing clinical stage were also seen for chromosomes 1p, 14q, and 17p (Fig. 6A), which were never lost in the stage I tumors. However, these did not reach statistical significance individually probably due to the small number of cases with LOH for each chromosome. LOH encompassing chromosome band 11p15 occurred at earlier stages, whereas loss of chromosome 7p did not show a discernible association with clinical stage (Fig. 6B; data not shown).

A likely target of the 17p LOH is the TP53 tumor suppressor gene, a locus that shows a well-known association with tumor progression in other systems and that has been implicated previously as mutated in some anaplastic Wilms' tumors (35). We sequenced several of the most frequently mutated exons of TP53 (exons 5-9) in six of the Wilms' tumors with 17p LOH and found a single case (a stage IV tumor) with a loss-of-function mutation: a 4-bp insertion in exon 5 that produced a frameshift and premature STOP codon (data not shown). The remaining cases may have TP53 mutations in other exons or mutations in another gene on this chromosome arm.

As shown in Fig. 6B, LOI as indicated by biallelic DNA methylation at the H19 5′ differentially methylated region (DMR) was more common in early-stage compared with late-stage Wilms' tumors (frequency of LOI in stage I-III versus stage IV; P = 0.037, Fisher's exact test). Strikingly, many of these early-stage Wilms' tumors with LOI entirely lacked LOH in the genome-wide scan. Figure 7 shows the genome-wide patterns of LOH for 58 sporadic Wilms' tumors in which we determined the LOI status: of the 22 tumors with LOI, 8 showed no evidence of LOH. Among the remaining 14 tumors with LOI, various chromosomes showed LOH, with 11q affected in 6 cases, 16q in 4 cases, and 7p in 4 cases. In comparing the rates of LOH for these three chromosomal regions in the tumors with and without LOI, none were significantly associated with LOI (Fisher's exact test, all nonsignificant).

These data indicate that LOI frequently precedes chromosomal losses in sporadic Wilms' tumors. We found additional evidence for this temporal sequence from analyzing mosaicism. In our prior study of chromosome 16q LOH in Wilms' tumors using conventional microsatellite markers, we have noticed that the loss of alleles on this chromosome is sometimes partial, consistent with the loss being a late-stage event that manifests as tissue mosaicism in the growing tumor (14). Consistent with this scenario, for one of the cases in that prior study, the 16q LOH only became clear when we used DNA obtained after early passages of the tumor explant, and for other cases, we also observed partial allele losses for 16q markers. Because this is relevant to the relative timing of the two events, we directly compared the extent of DNA methylation at the H19 5′ DMR with the extent of chromosome 16q LOH in primary Wilms' tumors. In the tumor illustrated in Fig. 8A, the gain of DNA methylation was complete, whereas the 16q LOH was partial, indicating that the LOI had preceded the LOH. In fact, LOI can occur as a preneoplastic lesion in Wilms' tumor patients, being detectable in the kidney parenchyma surrounding the tumor, even in the absence of histologic abnormalities, such as nephrogenic rests. We reported previously two cases of sporadic Wilms' tumors with this finding (36), and we have since found two additional examples in our expanded case series. In one of these cases, the SNP chip analysis revealed chromosome 7p LOH in the tumor. As shown in Fig. 8B, a side-by-side analysis of H19 5′ DMR methylation and chromosome 7p LOH in the tumor and adjacent kidney parenchyma from this case confirms that epigenetic lesion (LOI) occurred well before the genetic lesion (chromosomal loss) in the evolution of this tumor. As in our previously reported cases (36), nephrogenic rests were not found in standard histologic sections from this case and the patient did not meet clinical criteria for Beckwith-Wiedemann syndrome. However, the patient was in the 97th percentile for height and 95th percentile for weight, suggesting somatic overgrowth.

Work from several laboratories, including ours, has led to the realization that the genetic and epigenetic events in the formation of sporadic versus syndromic (Denys-Drash and WAGR) Wilms' tumors are largely distinct. First, WT1 loss-of-function mutations are invariably present in the syndromic tumors and are rare in the sporadic tumors (8-10). Second, β-catenin (CTNNB1) gain-of-function mutations are common in the syndromic tumors and, among the sporadic tumors, are only found when a WT1 mutation is also present (9, 37). Third, syndromic Wilms' tumors, together with the rare sporadic cases that have WT1 mutations, show a characteristic gene expression signature that differs from that of the common (WT1 wild-type/CTNNB1 wild-type) sporadic cases (9). Fourth, Wilms' tumors occurring in Denys-Drash and WAGR syndrome showed a lower frequency of IGF2 LOI compared with sporadic Wilms' tumors (38), a finding that is substantiated by our current data. Most recently, Ruteshouser et al. reported a genome scan similar in design to ours, which uncovered distinct patterns of LOH in syndromic versus sporadic Wilms' tumors (39). Our current data comparing syndromic and sporadic Wilms' tumors are consistent with these prior reports: we found no evidence of LOI in any of the nine syndromic Wilms' tumors in our series and found a different pattern of LOH in these cases compared with our series of sporadic tumors.

Our findings highlight known and novel regions of LOH in sporadic Wilms' tumors. For the known regions of recurrent loss, including chromosome 1p, 7p, 11p, 11q, 14q, and 16q, the minimal regions of LOH remain large. But included in our data are several smaller regions of recurrent LOH, notably a 2.4-Mb region on chromosome 4q24-q25. This region of DNA contains several interesting genes, including the CXXC4 gene encoding a Wnt pathway inhibitor. We did not find mutations in this gene, but quantitation Southern blot analysis indicated CXXC4 hemizygosity in the tumors with LOH, and it is possible that a reduction in CXXC4 gene dosage might contribute to sporadic Wilms' tumor progression. Alternatively, one of the other genes in the region could be the relevant target of loss. The NPNT/LOC255743 gene, encoding an apparent orthologue of the mouse nephronectin gene, is another interesting candidate tumor suppressor, as nephronectin is an extracellular matrix component that is a ligand for α₈/β₁ integrin, which is in turn a molecule that is essential for metanephric kidney maturation (40-44).

Ruteshouser et al. reported previously a restricted region of chromosome 9q LOH in Wilms' tumors (39). In our series, four sporadic tumors showed 9q LOH (Fig. 1). The regions of LOH in two of our cases do not overlap with each other or with the minimal region defined by Ruteshouser et al., whereas the two other cases showed large segments of LOH that do include their minimal region. Analysis of additional cases will be needed to define the molecular anatomy of chromosome 9q in Wilms' tumors.

Our study goes beyond prior work in providing a comprehensive analysis of both genome-wide LOH and LOI in a large series of Wilms' tumors. The results, showing LOI more frequent in early-stage tumors and LOH, particularly for chromosome arms 1p, 14q, 16q, and 17p, more frequent in late-stage tumors, have implications for the timing and mechanism of LOI. The data show that LOI is an early event, often occurring in the complete absence of LOH. This finding is consistent with two prior observations. First, gain of methylation in H19 upstream DNA sequences (as well as in the transcribed region of the H19 gene), which is the DNA counterpart of LOI that is scored in this and many other studies, can be seen in the nonneoplastic kidney parenchyma of a substantial group of sporadic Wilms' tumor patients (36, 45). Second, IGF2 LOI associated with gain of methylation at the H19 5′ DMR in various somatic tissues is the defining feature of the class of Beckwith-Wiedemann syndrome that is associated with a predisposition to Wilms' tumor (12, 46, 47). Recently, it has been suggested that rare cases of sporadic Wilms' tumor might have LOI secondary to LOH affecting the chromosome 16q CTCF gene (48). Our data indicating that LOI usually precedes chromosome 16q LOH and is not statistically associated with loss of this chromosome arm do not provide positive support for this scenario. However, the CTCF protein plays an important role as an insulator binding protein at the H19 5′ DMR, and our results do not rule out the possibility of rare Wilms' tumors in which CTCF hemizygosity may contribute to LOI.

The fact that sporadic Wilms' tumors with LOI that present at early clinical stages often show no LOH raises interesting questions about the molecular pathogenesis of these cancers. Stage I Wilms' tumors, although confined to the kidney in which they arise, are nonetheless large neoplasms, which often come to clinical attention based on effects from the tumor mass. As the research community makes plans for large-scale cancer genome and cancer epigenome projects, such tumors may offer the opportunity to answer questions about the relative requirements for genetic versus epigenetic events in the early stages of tumor formation. Lastly, our finding that LOI is less common in Wilms' tumors that are metastatic at clinical presentation compared with cases in which the tumor presents at an earlier clinical stage suggests that the tumors with LOI may be biologically different from those with LOH.

Biopsies of 62 sporadic and 9 syndromic Wilms' tumor and, when available, adjacent normal kidney tissue were obtained with institutional review board approval. Of the sporadic Wilms' tumors, 10 were stage I, 12 were stage II, 14 were stage III, and 8 were stage IV. Stage information was not available for the remaining 18 sporadic tumors. DNA was extracted by proteinase K-SDS lysis followed by phenol-chloroform extraction and ethanol precipitation. DNA quality was assessed by gel electrophoresis and ethidium bromide staining, and only samples with intact high molecular weight DNA were used for preparing SNP chip probes. RNA was prepared from tissue pulverized under liquid nitrogen and lysed in Trizol reagent (Invitrogen, Carlsbad, CA).

A portion of the tumor was placed in MEM immediately after surgery for cytogenetic analysis. Tumor tissues were minced using scalpel blade and digested with 200 units/mL collagenase in complete DMEM for 2 to 12 hours. Dissociated cells were washed twice with DMEM and cultures were set up in complete DMEM supplemented with insulin-transferrin-sodium selenate (Invitrogen). Tumor cells grown for 2 to 10 days were harvested for metaphase preparation, g-banded, and analyzed by standard procedures (14).

Probes were prepared from the genomic DNA samples according to the manufacturer's instructions (Affymetrix GeneChip Mapping Assay). Where necessary, DNA was concentrated using sodium acetate extraction before digesting with XbaI. To generate the required 20 μg DNA, five (instead of four) PCR reactions were set up. PCR product was concentrated using either QIAquick PCR Purification kit or MiniElute UF PCR Purification kit (Qiagen, Hilden, Germany).

Alleles (AA, AB, or BB) at each SNP were called using the algorithm of Liu et al. (49) as implemented in the Affymetrix software package GDAS, and the data displayed in a spreadsheet sorted by chromosome and nucleotide position using Build 34 of the human genome. For normal/tumor pairs, LOH at a given SNP was identified as an AB-to-AA or AB-to-BB conversion. For unpaired tumor samples, regional LOH was scored whenever an uninterrupted contiguous stretch of >100 homozygous (AA or BB) genotypes was encountered (see Results).

We analyzed CpG methylation in the H19 5′ DMR by Southern blotting after digestion of genomic DNA with methylation-sensitive restriction enzymes as described previously (14). For measuring DNA copy number at the CXXC4 gene, genomic DNA from Wilms' tumors and matched nonneoplastic kidney parenchyma was digested with RsaI and the Southern blot was hybridized simultaneously with ³²P-labeled probes for CXXC4 intron 3 (intron 1 in published sequence) on chromosome arm 4q and intron 1 of the ADD1 gene on chromosome arm 4p. Probes were generated by genomic PCR with CXXC4 PCR primers: CCATATGGACTTGACTTAGTGC and CCTAGCAAAGGTGAACACCTG and ADD1 PCR primers: AATATCATGTAACTACAGAGACTC and GCAAGAAGAATCACGGAGCGT. Northern blots were done after electrophoresis of 6 μg denatured total RNA through 1% agarose gels containing formaldehyde.

For analyzing LOH in the minimal region of LOH on chromosome 4q, genomic PCR primers were rs233973-US (TGGTGGTGGTAAGGCAGAGC) and rs233973-DS (GGTGGGTGGTGTTTCAGTCG), PCR product digested with Tsp45I; rs950882-US (ATGATGCAGGCCAAGGAGCT) and rs950882-DS (GTAAGTTCCACAGCGTAGGCA), PCR product digested with BccI; and rs959482-US (ACTCTGAGGCCAAACTCCTTG) and rs959482-DS (CAGGGTTCTCCCCATTCAGC), PCR product digested with PsiI. CXXC4 intron 1 SNP (identified by us; rs number not assigned) CXXC4-intron1-US (GGTGTCCACGTCCCTGCGT) and CXXC4-intron1-DS (TTCACAATCCCAACGGACAGC), PCR product directly sequenced; rs17036613-US (ATATAATGTTCTTGGAGTACTGTG) and rs17036613-DS (CTTGTTCAGATATGCACTATGAC), PCR product directly sequenced. Primer sequences for amplifying and sequencing the coding region of CXXC4 are available on request. Exons 5 to 9 of the TP53 gene were sequenced after genomic PCR using flanking intronic primers; primer sequences are available on request. The RFLP for chromosome 7p was assessed by genomic PCR using primers rs720380-F1 (AGTTGGCTAGACCATGACACC) and rs720380-R1 (TTCCTCTAGAACCATCCTTGG) followed by EcoRV digestion.