Wilms tumour is the most common paediatric renal malignancy, and due to improved multimodality therapy, there is now a survival rate of nearly 90%.1 The search for molecular markers to identify those children whose tumours will relapse has highlighted genetic losses on chromosomes 1p and 16q,2^–5 as well as gains on 1q,6 7 although to date no responsible genes at these loci have been identified. A common mechanism for these alterations involving a balanced t(1;16) translocation followed by duplication and non-disjunction has been proposed, but lacks hard experimental evidence. Due to the large heterochromatic regions on chromosomes 1 and 16, there exists the possibility that chromosome 9 may act as an alternative partner chromosome for translocations involving 1q, although such karyotypes have only rarely been reported.8 9

Genomic alterations on chromosome 9 have not been described frequently in Wilms tumours, with cytogenetic data reporting trisomy 9 in approximately 13% of cases,10 and metaphase CGH studies have reported 9p loss in just a single case.11 A recent genome-wide survey of LOH has identified a novel region on 9q22.2–q31.1, which shows a greater degree of LOH in WT1 wild type compared with WT1 mutant tumours, and has been purported to harbour tumour suppressor genes whose expression is regulated by the pleiotropic effects of WT1.12

Deletions involving chromosome 9p have been implicated in a wide range of tumours including melanoma and non-small cell lung cancer, with the main region of loss covering 9p21.13^–15 This locus harbours three tumour suppressor genes of the cyclin dependent kinase (CDK) inhibitor family: CDKN2A/p16^INK4a, CDKN2A/p14^ARF and CDKN2B/p15^INK4b. Alterations targeting the CDKN2A/2B locus have been suggested to play an important role in favourable histology Wilms tumours, with a complete loss of CDKN2A/p16^INK4a expression reported in over 20% of tumours; this reduction of expression was found to be associated with higher tumour stage.16 However, a more recent case-cohort study found that decreased or absent expression was not correlated with adverse outcome.17 Loss of CDKN2A/p16^INK4a gene expression in Wilms tumours was found to correlate with methylation of the gene; however, no homozygous deletions were detected.16 Promoter methylation was further reported in approximately 10% of cases,18 19 although a separate group reported absence of methylation at the CDKN2A/p16^INK4a locus in Wilms tumours.20

We have sought to further clarify the role of chromosome 9 alterations in relation to Wilms tumour development and relapse. Utilising microarray CGH (aCGH), we describe a complex variety of genetic aberrations along chromosome 9 in Wilms tumours, and have investigated the CDKN2A/p16^INK4a locus by loss of heterozygosity (LOH), fluorescence in situ hybridisation (FISH) and immunohistochemistry on a paediatric renal tumour tissue microarray.

MATERIALS AND METHODS

Tumour samples

Frozen Wilms tumour samples and matching constitutional DNA (where available) were obtained after approval by local and multicentre ethical review committees from the North American Children’s Oncology Group and the United Kingdom Childhood Cancer Study Group (UKCCSG). The sample set (n = 76) was enriched for tumours that subsequently relapsed (n = 37), were taken at immediate nephrectomy, and have been described previously.21 The mean age at diagnosis was 37.3 months. There were 23 stage 1 tumours, 21 stage 2 tumours, 20 stage 3 tumours, 7 stage 4 tumours and 4 stage 5 tumours (one unknown). The mean relapse-free survival time was 44.2 months, and the mean overall survival time 71.0 months. Samples were taken at surgery and frozen in liquid nitrogen. Genomic DNA was extracted using a standard proteinase K digestion followed by phenol/chloroform extraction, and resuspended in water.

Loss of heterozygosity

PCR and LOH analysis on 34 tumours was performed as previously described.22 Primer sequences were obtained from the Genome Database, and are given in table 1. The forward primer was labelled fluorescently with either tetrachlorofluorescein or hexachlorofluorescein dyes (Invitrogen, Paisley, UK). Sequences from tumour DNA and corresponding normal DNA were amplified using the following cycle parameters: denaturing at 95°C for 10 min, followed by 29 cycles of 95°C for 30 s, 55°C for 30 s, 72°C for 30 s, and a subsequent final extension of 72°C for 10 min and incubation at 60°C for 30 min. Each PCR reaction was performed under standard conditions in a 12.5 μl reaction volume containing 10 ng of template DNA, 0.8 μM of each primer, 0.2 μM dNTP, 1.5 mM MgCl₂, 0.2 units of Taq polymerase (ABGene, Epsom, UK), and 1× PCR buffer. Denatured reaction products were analysed on an ABI 7700 capillary sequencer (Applied Biosystems, Foster City, California, USA), and visualised with Genemapper software (Applied Biosystems). Allele ratios for tumour compared with normal DNA were calculated as (A1/A2)T/(A1/A2)N and LOH was scored if this ratio was less than 0.67 and greater than 1.5.

Microarray CGH

aCGH experiments were carried out as previously described,21 with all data submitted according to MIAME guidelines23 to the public data repository ArrayExpress (http://www.ebi.ac.uk/arrayexpress/) with accession number E-TABM-10. Seventy-six favourable histology Wilms tumours were hybridised in duplicate to Breakthrough Breast Cancer Research Centre human CGH 4.6K 1.1.2 arrays (Array Express accession number A-MEXP-192) consisting of 4115 BAC clones, spaced at approximately 1 Mb throughout the genome, spotted in triplicate onto Corning GAPSII-coated glass slides (Corning, New York, USA). All statistical analysis was carried out in R 2.0.1 (http://www.r-project.org/).

Interphase FISH

Wilms tumours were assessed with a dual-labelled FISH assay comprising a flourescein-labelled chromosome 9 centromere probe and a Spectrum Orange 9p21-specific probe spanning ∼190 kb and covering the loci D9S1749, D9S1747, CDKN2A/p16^INK4a, CDKN2A/p14^ARF, D9S1748, CDKN2B/p15^INK4b and D9S1752 (Vysis, Downers Grove, Illinois, USA); it was carried out according to manufacturer’s instructions on eight Wilms tumour touch preparations.

Immunohistochemistry on tissue microarrays

Paediatric renal tumour tissue microarrays were constructed24 containing replicate representative cores (n = 885) from all available cellular components from 274 Wilms tumours, 14 clear cell sarcomas of the kidney (CCSK), 9 mesoblastic nephromas, and 7 rhabdoid tumours of the kidney (RTK). Tumours were treated either according to National Wilms Tumor Study Group guidelines (immediate nephrectomy) or International society of Pediatric Oncology (SIOP) protocols (delayed nephrectomy following preoperative chemotherapy), and classified accordingly. There was a slight enrichment of tumours which relapsed. The presence of tumour tissue on the arrayed samples was verified on an H&E stained section.

Immunohistochemistry was performed on 5 μm tissue array sections using mouse monoclonal antibody to CDKN2A/p16^INK4a (6H12, Novocastra, Newcastle, UK). Antigen retrieval was enhanced by boiling the slides in a microwave for 25 min in DAKO solution pH 9.0. Positive (islet cells of normal pancreas samples) and negative (omission of the primary antibody) controls were included in each slide run. Tumour cell nuclear positivity was assessed independently by three pathologists (JSR-F, M-AB, GV) and reported as a percentage of the total number of tumour cells present in each core, assessed by manual cell counting. Tumour cores were scored as: “strong expression”, in which >30% tumour cells displayed immunoreactivity; “weak expression”, with 1–30% positive cells; “negative”; or “uninterpretable/missing”. For the purposes of statistical analyses, weak and strong expression were regarded as positive. A p value of <0.05 was regarded as significant. Overall agreement between the three observers was very good (Cohen’s κ = 0.768), and discrepant cores were scored according to the majority vote.

RESULTS

Chromosome 9 copy number changes in Wilms tumours

Wilms tumours showed a complex pattern of DNA copy number changes on chromosome 9, as detected by aCGH analysis using a 1 MB-spaced genome-wide BAC array platform (fig 1). Hierarchical clustering of the tumours on the basis of the chromosome 9 aCGH data revealed approximately half to have no changes, and separated those samples which did harbour copy number alterations into four main categories. There were no clinical correlates of these different groups of tumours, with no statistically significant associations observed between the chromosome 9 copy number changes and relapse, survival, stage or age (table 2).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 Heatmap of chromosome 9 aCGH data for 76 favourable histology Wilms tumours. Averaged weights smoothed data plotted as gains (green), losses (red) or no change (black) for each BAC clone in order along chromosome 9 (left, pter to right, qter). Coloured bars represent tumours group by hierarchical clustering according to their pattern of copy number alterations: blue, trisomy 9; orange, gain 9q34; pink, loss 9p; olive, complex pattern; violet, no change. The grey area represents the centromere and heterochromatic region (no data).

Of the favourable histology Wilms tumours, enriched for relapsing cases, 39/76 (51.3%) exhibited no copy number changes on chromosome 9 (“no change” group). These patients had a mean age at diagnosis of 39.1 months, comprised 61% girls, and had a stage distribution of fourteen stage 1, eleven stage 2, eight stage 3, four stage 4 and one stage 5. Mean relapse-free and overall survival times were 46.6 and 75.3 months respectively.

Nearly 8% (6/76) of cases were found to have a gain of the entire chromosome (assumed “trisomy 9”), agreeing broadly with published cytogenetic data.10 These patients had a mean age at diagnosis of 30.5 months, were split equally between the sexes, and consisted of two stage 2 and four stage 3 tumours. The relapse-free (47.6 months) and overall survival (80.6 months) times were longer than the average (44.2 and 71.0 months), but this did not reach statistical significance (log rank test).

As well as whole chromosome gains, there were also 8/76 (10.5%) tumours which showed a focal gain at 9q34-qter. This represents a large region covering some 11 Mb, and houses more than 200 known genes. These cases comprised three stage 1, one stage 2, two stage 3, one stage 4 and one stage 5 tumours. There was a shorter than average time to relapse (39.2 months) and death (68.4 months); however this did not reach statistical significance (p = 0.278, log rank test).

Tumours with loss of all or most of 9p (“loss 9p”) accounted for 7/76 (9.2%) cases, and in each case targeted 9p21, one of the most common regions of deletion in human cancer.

There were two small regions of overlap among the tumours, mapping to 9p21.3 and 9p21.1. There were two each of stages 1, 2 and 3, and a single stage 4 tumour. These cases had the shortest relapse-free and overall survival times (39.1 and 57.2 months), but this was not significant (log rank test). Case RMH0855 has been included in the group with 9p loss on the basis of hierarchical clustering, as this tumour exhibited log₂ ratios along 9p that were less than zero, but did not cross the threshold for assignment as “loss” (−0.15). This may be due to contamination with normal, non-tumour cells in that sample, or perhaps intratumoural heterogeneity.

Finally there was a group of tumours with more complex rearrangements including multiple focal gains on 9p, as well as losses on 9q (16/76, 21.1%). These tumours form more than one cluster in the dendrogram, but are here grouped together for convenience, as one or more of the alterations observed appeared in any one tumour. This group of tumours also showed no significant clinicopathological correlates, despite having a shorter than average time to relapse (42.3 months) and death (65.0 months).

When we compared the aCGH profiles from across the genome with the different groups of tumours based on their chromosome 9 alterations, a number of significant correlations were found (table 2). Tumours with trisomy 9 were significantly associated with a concurrent gain of chromosome 18 (67% vs 8%, adjusted p<0.01, Fisher’s exact test); tumours with gain of 9q34 were associated with a loss of chromosome 21 (63% vs 13%, adjusted p<0.01, Fisher’s exact test); and losses of 9p significantly correlated with losses at 1q32–q41 (42% vs 2.5%, adjusted p<0.01, Fisher’s exact test). No correlations were found between copy number gains on chromosome 1q and losses on 9q (Fisher’s exact test).

LOH on chromosome 9p

In order to further characterise the copy number losses observed on chromosome 9p, we performed LOH on a subset of samples for which matching constitutional DNA was available. In order to minimise the confounding issue of allelic imbalance being detected due to copy number gains at these loci, tumours with trisomy 9 as detected by aCGH were excluded. Those tumours with no changes observed by aCGH were still analysed, due to incomplete coverage of 9p on the array, and the possibility of small interstitial deletions being missed.

LOH analysis was performed on 34 Wilms tumour samples using a panel of 14 highly informative microsatellite markers spanning 9p24.1–p21.1 All patients were informative for multiple markers. Allelic imbalance at one or more loci along 9p was observed in 53% (18/34) of cases (fig 2).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2 Pattern of loss of heterozygosity (LOH) in Wilms tumours with deletions of 9p. Microsatellite markers and their positions in megabases from 9p24.1 to 9p21.1 are given according to Build 34 of the human genome. Black boxes represent LOH; white boxes denote retention of heterozygosity; dark grey boxes indicate non-informative markers for that case; light grey boxes were not done. Shaded boxes represent a microsatellite alteration.

The most common region of overlap was defined between D9S932 (24.2 Mb) and D9S265 (25.4 Mb) at 9p21.2–p21.1, centromeric to the CDKN2A/2B locus. LOH was observed in this region in 12/34 (35%) tumours. This is a gene-poor region, flanked distally by ELAVL2, and proximally by TUSC1, proposed as a tumour suppressor candidate gene in lung cancer due to homozygous deletion.25 A further eight tumours (8/34, 24%) showed LOH in a region between D9S286 and D9S775, which spans nearly 2 Mb, and houses the gene PTPRD, which has been noted to be a common target of homozygous deletion in a number of cancer cell lines.26

Four tumours (4/34, 12%) showed LOH at D9S1748, which maps directly to the CDKN2A/2B locus at 9p21.3, with a further tumour showing microsatellite instability. This correlates with one of the regions of copy number deletion observed by aCGH. Deletions were confirmed by FISH analysis using commercial probes hybridising to this locus as well as the chromosome 9 centromere. All deletions were confirmed as hemizygous (fig 3).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3 Fluorescence in situ hybridisation (FISH) detection of copy number alteration at the CDKN2A/2B locus. Interphase FISH results (right) with probes mapping to the CDKN2A/2B locus (red) and chromosome 9 centromere (green) are given alongside the corresponding aCGH data for chromosome 9 in the respective tumours (left). Top panel shows a hemizygous deletion of CDKN2A/2B in a tumour showing loss of the whole short arm of chromosome 9. Bottom panel shows a tumour with trisomy 9 reflected in both the aCGH and FISH experiments.

No significant correlations were observed between LOH at any marker and tumour stage or clinical outcome (data not shown).

CDKN2A/p16^INK4a protein expression

To determine the frequency and potential implications of CDKN2A/p16^INK4a protein expression in Wilms tumours and other paediatric renal neoplasms, we analysed a cohort of paediatric renal tumours on a tissue microarray. None of the normal kidney tissues displayed detectable levels of CDKN2A/p16^INK4a, in keeping with previous reports27 (fig 4A). The vast majority of Wilms tumours were also negative in all cellular components assessable (fig 4B). In assessable cores, we observed 38/236 (16.1%) non-anaplastic Wilms tumours with nuclear CDKN2A/p16^INK4a expression. These tumours exhibited staining in the blastemal component in 13/136 (9.6%) cases, the epithelial cells in 9/98 (9.2%) cases, and the tumourigenic stromal cells in 22/95 (23.2%) cases. The frequency of stromal positivity was significantly greater than that of the blastemal or epithelial components (p<0.01, Fisher’s exact test). Figure 4 (C,D) shows representative tumours. There was no correlation with clinicopathological variables including age at diagnosis, tumour stage, clinical outcome, or treatment (immediate nephrectomy vs preoperative chemotherapy), even when separated by “weak” (24/236, 10.2%) and “strong” (14/236, 5.9%) expression (data not shown). There were also no associations with the SIOP histological risk classifications for patients treated with preoperative chemotherapy, although interestingly, an increased proportion of anaplastic tumours (4/9, 44.4%, p = 0.0493, Fisher’s exact test) showed a positive CDKN2A/p16^INK4a immunoreactivity, regardless of treatment status.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4 CDKN2A/p16^INK4a immunohistochemistry on a paediatric renal tumour tissue array. (A) normal kidney; and (B) triphasic Wilms tumour, both representing negative CDKN2A/p16^INK4a protein expression. (C) Wilms tumour blastemal cells; and (D) stromal cells showing strongly positive nuclear immunoreactivity diffusely throughout the tumour core. (E) Rhabdoid tumour of the kidney; and (F) clear cell sarcoma of the kidney, showing scattered cells strong to moderately positive for CDKN2A/p16^INK4a.

Also present on our tissue array were a number of non-Wilms paediatric renal tumours, including CCSK and RTK. Although the numbers are too small to make any rigorous conclusions, it was interesting to note that 4/10 (40%, p = 0.071) CCSKs and 5/7 (71.4%, p<0.001) RTKs showed a positive p16 immunoreactivity (fig 4E,F).

DISCUSSION

Chromosome 9 harbours a number of genes implicated in human cancer, including the tumour suppressor locus CDKN2A/2B at 9p21, and the proto-oncogene ABL1 on 9q34 involved in structural rearrangements including the t(9;22) reciprocal translocation. It is highly structurally polymorphic, and contains the largest autosomal block of heterochromatin in the human genome.28 Evidence that it may play an alternative partner chromosome for a balanced translocation involving 1q was further supported by the observation of a high degree of correlation between partial LOH at chromosome 1q and losses at 9q.12 Here we report a lack of association between genomic gain on chromosome 1q and loss on 9q by our aCGH experiments, which is in stark contrast to the significant correlation we observed between gain of 1q and loss of 16q.21

Copy number aberrations were seen in nearly half of the Wilms tumours we analysed (enriched for tumours which went on to relapse), and the types of genomic alterations we observed fell into four categories. Gain of the whole chromosome, an assumed trisomy 9, was seen in approximately 8% of tumours, which broadly agrees with cytogenetic data.10 Although there were no clinical correlates for this observation, there was a strong association with gain of chromosome 18, an alteration reported in 16% of Wilms tumours, and occasionally found in late onset (>5 years) tumours as a result of a constitutional abnormality.29

Approximately 10% of tumours showed a focal gain of 9q34, a region containing a number of purported cancer-related genes, including ABL1, LCN2, SET, CDK9, ENG, VAV2, TRAF2 and NOTCH1. Amplifications at this locus have been described in a variety of cancers, particularly lymphomas30^–32 and childhood adrenocortical tumours,33 34 and it also harbours a number of translocation breakpoints, most notably that for the Philadelphia translocation t(9;22)(q34;q11) resulting in the BCR-ABL fusion gene.35 Intriguingly, gain of 9q34 correlated with a loss of chromosome 21, itself described in approximately 8% of Wilms tumour karyotypes.10

In other tumours, there was a more complex pattern of copy number changes, including focal gains on 9p, and/or losses on 9q. Such a pattern of alterations may reflect underlying genomic instability, although this is not generally regarded a feature of Wilms and other embryonal tumours, and in these cases there was no correlation with an increased number of alterations elsewhere in the genome. In five cases the DNA copy number losses in this group of tumours correspond to the 9q22.2–q31.1 locus recently identified by genome-wide LOH analysis,12 although no smaller region of overlap could be identified. These data confirm the location of a possible tumour suppressor at this locus.

Approximately 9% of tumours showed a loss of chromosome 9p, in most cases extending across the entire arm, but in all cases covering 9p21. These deletions, as detected by aCGH, did not have any clinical implications, but did show a significant association with losses on chromosome 1q32–q41. This is an aberration observed in up to 10% of our cases, and has not previously been reported in Wilms tumours, which are associated with a high frequency of 1q gains.21 Deletions at this locus have, however, been described in renal collecting duct carcinoma.36

The CDKN2A/2B locus at 9p21.3 is one of the most commonly deleted regions in human cancer, and inactivation by either genetic or epigenetic means implicates CDKN2A/p16^INK4a, CDKN2A/p14^ARF, and CDKN2B/p15^INK4b as major tumour suppressor genes.37^–39 In Wilms tumour, the picture is less clear. An early publication suggested a decreased expression by promoter hypermethylation to be associated with advanced stage disease,16 although this was not borne out in a larger case-control study.17 Other groups report conflicting data on the presence of CDKN2A promoter hypermethylation.18^–20 Our data showed hemizygous deletion in 4/34 (12%) cases, the first reported LOH at the CDKN2A/2B locus in Wilms tumour, although we could find no correlations with clinicopathological variables.

Other regions that show LOH in this study include relatively gene-poor loci, both proximal and distal to CDKN2A/2B. Such regions of deletion have been reported in other tumour types, and may harbour tumour suppressor genes, such as the putative lung suppressor TUSC1 at 9p21.1.25 Equally, it may be that the high frequency of observed CDKN2A/p16^INK4a deletion in human cancer may be a result of a reduced negative selection pressure associated with deletion of adjacent genes.26

In order to further investigate changes in CDKN2A/p16^INK4a levels in Wilms tumours, we examined protein expression using immunohistochemistry on a paediatric renal tumour tissue microarray. None of the normal kidney tissues and only 38/236 (16.1%) non-anaplastic Wilms tumours displayed detectable levels of CDKN2A/p16^INK4a, and as reported in an adult renal cell carcinoma study, the negative phenotype of Wilms tumours may reflect the normal physiological state rather than a reduced expression.27 Frequent expression of CDKN2A/p16^INK4a can be seen in many adult tumours such as prostate,40 lung,41 ovarian,42 and breast carcinoma,43 with expression associated with tumour progression and poor prognosis. Such activation is frequently associated with the stress response to inappropriate growth signals, and may play a role in driving the cell towards senescence.44 Up-regulation and overexpression of CDKN2A/p16^INK4a has been described in the aging kidney,45 46 and may be associated with deteriorating transplants and kidney disease.47 The intriguing observation of increased CDKN2A/p16^INK4a expression in 4/9 anaplastic Wilms tumours, 5/7 RTKs and 4/10 CCSKs has not previously been described, and may warrant further investigation into the possible mechanistic associations with drug resistance in these tumour types.