Wilms tumor is a childhood embryonal kidney cancer that affects approximately 1 in 10,000 children in Western populations.1
The median age of diagnosis is between 3 and 4 years and both kidneys are affected in ~5% of children. Histologically, Wilms tumor mirrors the development of normal kidney and classically contains the three cell types present in the embryonic kidney; blastema, epithelia and stroma.2
There is strong evidence of a genetic contribution to Wilms tumor. Firstly, there is significant variation in incidence that is attributable to ethnicity rather than geographical location.1
Secondly, approximately 2% of cases are familial.3
Thirdly, there exist several predisposition syndromes associated with an increased risk of Wilms tumor, the most common of which are conditions due to mutations in WT1
or epigenetic defects at 11p15.3
Fewer than 5% of Wilms tumor cases are attributable to known causes and the underlying basis of most Wilms tumor is unknown.4,5
We performed a genome-wide association study (GWAS) to identify genetic variants that confer susceptibility to Wilms tumor, using cases recruited through oncology clinics in North America. We compared the case genotype frequencies to data from North American controls obtained from the Database of Genotypes and Phenotypes (dbGAP).6
After quality control exclusions and removal of samples with non-European ancestry, we used data on 599,255 SNPs in 757 cases and 1,879 controls. We compared genotype frequencies in cases and controls primarily using the 1-degree-of-freedom (1df) Cochran Armitage trend test (Supplementary Figure 1
). There was evidence of only modest inflation of the test statistics (λ=1.08), indicating that confounding due to population stratification was limited (Supplementary Figure 2
We next considered 20 SNPs in nine regions of linkage disequilibrium (LD) that were significant at P<5×10−5. After eliminating strongly correlated SNPs, we further evaluated these putative associations by genotyping ten SNPs at nine loci in two independent replication series. These included 769 cases and 2,814 controls from the UK and 719 further cases from North America. We did not have samples from US control individuals to directly genotype so we analysed the US case genotypes against data from 1,037 additional controls obtained from dbGaP.
Two SNPs on chromosomes 2p24 and one SNP on chromosome 11q14 showed clear evidence of replication in each of the replication series separately (P
or better in the same direction as the GWAS) and reached genome-wide significance levels over both stages combined of P
at 2p24 and P
at 11q14 (, and Supplementary Tables 1 and 2
Summary results for three confirmed Wilms tumour susceptibility SNPs.
Regional plots of Wilms tumour susceptibility loci at 2p24 and 11q14
We also identified three low-frequency SNPs at 5q14 (rs1027643, P
), 22q12 (rs2283873, P
) and Xp22 (rs5955543, P
) associations for which achieved genome-wide significance, demonstrating clear replication in the US series but with only weaker evidence of replication in the UK series (P
=0.013 to P
=0.0672) (, Supplementary Tables 1 and 2
). Accordingly, further replication studies are required to confirm these associations, as low frequency variants in particular are susceptible to signal artefact when using different genotyping platforms for cases and controls.
Summary results for three low-frequency candidate Wilms tumour susceptibility SNPs.
We next used imputation to evaluate if a more strongly associated variant was present at the loci, by estimating the genotype probabilities at additional SNPs using HapMap3 and 1000 Genomes data as a framework. At four of the five loci, imputation identified more strongly associated variants than the original SNP (, Supplementary Table 3
). In all regions, the imputed SNPs were of similar risk-frequency profile to the genotyped SNPs. Direct genotyping of these imputed SNPs in cases and controls would be of interest to confirm these associations. We also estimated the ORs associated with haplotypes of SNPs in each of the five regions (Supplementary Table 4
). At each locus either the association was present on more than one haplotype carrying the risk allele, suggesting that the association was unlikely to be driven by a single rarer, higher penetrance variant, or the association was only present on a single haplotype of similar frequency to the sentinel SNP, thus affording no additional information. There was no evidence of departure from a log-additive model for any SNP, i.e. the OR for rare homozygotes did not differ significantly from the square of the OR for heterozygotes. There was also no evidence of statistical gene-gene interaction between loci; the combined effects of SNPs were consistent with multiplicative (log-additive) combination of effects.
We investigated whether the loci are associated with different risks in subgroups of Wilms tumor cases characterised by specific phenotypic features or risk factors including sex, age at diagnosis, family history of Wilms tumor, if the tumor was bilateral or unilateral, histology, disease stage, or if there was relapse after treatment (Supplementary Tables 5–7
). rs790356 at 11q14 showed evidence of a stronger effect in females than males (P
=0.0017) and a trend in strength of association with increasing age at diagnosis (P
=0.0036). Genotyping of rs790356 in additional series would be of interest to further evaluate these associations. There was no significant difference in risk at any other SNP in any subgroup (Supplementary Tables 6–7
The strongest evidence of association (P
) was at rs790356, which lies in a 68 kb LD block on chromosome 11q14.1 containing DLG2
(Discs Large, Drosophila, homolog of 2). DLG2 is a member of the membrane-associated guanylate kinase protein family.7
Dlg, the Drosophila
homolog of DLG2, functions with Scribble and Lgl in the planar cell polarity pathway.8
This pathway is essential for correct tissue morphogenesis during development and its disruption has been implicated in oncogenesis.9
Interestingly, the human homolog of scribble (SCRIB) is a recognised target of the Wilms tumor 1 gene product, WT1.5,10
In mice, Scribble and wt1 show coincident expression in the developing kidney. Moreover, in both murine and human kidney cell lines, WT1 has been shown to bind the SCRIB promoter activating SCRIB
Thus, although interactions of DLG2 in humans have not been well characterised, it is plausible that susceptibility to Wilms tumor at the 11q14 locus is effected through DLG2 and WT1-related pathways.
We identified two SNPs associated with Wilms tumor susceptibility in a 109 kb LD block at 2p24.3. The effects of rs3755132 and rs807624 are maintained when corrected for each other (OR=1.25 (1.03–1.53), OR=1.21 (1.04–1.41) respectively; P
=0.01). Furthermore, the correlation between rs3755132 and rs807624 is weak (D’=1 and r2=0.328 in HapMap3 CEU samples, D’=0.979 and r2=0.328 in our GWAS control data). These data suggest that rs3755132 and rs807624 are independently associated with Wilms tumor. Because neither SNP alone can fully account for the association, it is possible that a unique causal variant may exist which is in LD with, and captures the effects of, both rs3755132 and rs807624. However, although imputation revealed more strongly associated SNPs at this locus (see Supplementary Table 3
), none fully captured the associations of both rs3755132 and rs807624. rs3755132 is located 1.5 kb upstream of the promoter of DDX1
(DEAD box polypeptide 1). rs807624 is 11 kb 3’ of DDX1
and 52 kb 3’ of rs3755132. DDX1 is involved in the initiation of translation and in RNA splicing and modification. It is also implicated in DNA double-strand break repair through its functions as an RNAse and RNA-DNA helicase.11
Notably, both SNPs are located within a region of 2p24 that shows somatic copy number gain or amplification in many childhood cancers, most notably neuroblastoma, but also occasionally in Wilms tumor.12,13
It is generally assumed that a neighboring gene in the amplicon, MYCN
, is the primary target in promoting oncogenesis. Whilst it is possible that the association we identified is mediated through a long-range effect on MYCN
seems the more likely target.
In summary, we have performed the first GWAS in Wilms tumor, identifying two definite and three probable predisposition loci and providing insights into biological pathways that may be important in the genesis of this embryonal kidney cancer. The power to detect these loci was 21% for 2p24, 13% for 11q14, 4% for 22q12, 3% for Xp22 and <1% for 5q14. This strongly suggests that multiple loci of equivalent or weaker effects are likely to exist, and may be tractable through follow-up of additional SNPs showing evidence of association in this study, and/or through further genome-wide association studies.