Recurrent DGCR8, DROSHA, and SIX Homeodomain Mutations in Favorable Histology Wilms Tumors

SUMMARY

We report the most common single nucleotide substitution/deletion mutations in Favorable Histology Wilms Tumors (FHWT) to occur within SIX1/2 (7% of 534 tumors) and microRNA processing genes (miRNAPG) DGCR8 and DROSHA (15% of 534 tumors). Comprehensive analysis of 77 FHWTs indicates that tumors with SIX1/2 and/or miRNAPG mutations show a pre-induction metanephric mesenchyme gene expression pattern and are significantly associated with both perilobar nephrogenic rests and 11p15 imprinting aberrations. Significantly decreased expression of mature Let-7a and the miR-200 family (responsible for mesenchymal-to-epithelial transition) in miRNAPG-mutant tumors is associated with an undifferentiated blastemal histology. The combination of SIX and miRNAPG mutations in the same tumor is associated with evidence of RAS activation and a higher rate of relapse and death.

INTRODUCTION

Wilms tumor (WT) represents the most common pediatric renal malignancy, with an estimated annual incidence of 500 cases in the United States [Howlander et al., 2013]. WTs commonly display epithelial, stromal, and undifferentiated (blastemal) components in varying proportions and often closely resemble the different stages of renal development [Rivera and Haber, 2005]. They often arise within precursor lesions known as perilobar and intralobar nephrogenic rests (PLNR, ILNR) [Beckwith et al., 1990]. Evidence suggests that WT development depends not only on the nature of specific genetic events, but also on the timing of their occurrence within early renal development [Gadd et al., 2012]. The developmental window begins with the early intermediate mesoderm, which contains progenitor cells of both the urinary collecting system and pre-induction metanephric mesenchyme. The metanephric mesenchyme undergoes induction, including mesenchymal-to-epithelial transition (MET), resulting in nephron development [Kobayashi et al., 2008].

Mutations in WT1, WTX, and CTNNB1 contribute to WT development; in addition, loss of imprinting (LOI) or loss of heterozygosity (LOH) at 11p15 (resulting in biallelic expression of IGF2) is present in the majority of WTs [Gadd et al., 2012]. However, the identification of 11p15 LOH in normal tissue from some WT patients [Chao et al., 1993], and the absence of tumor development in mutant mice with LOI of 11p15 [Hu et al., 2011] suggest that biallelic expression of IGF2 alone is insufficient for tumor development. Five subsets of WT identified based on gene expression patterns differ in their histology, nephrogenic rest status, clinical outcome, and show evidence of arrest at different stages of renal development [Gadd et al., 2012]. Two subsets express high levels of WT1: Subset 1 (~5% of FHWT) is comprised exclusively of epithelial tumors lacking nephrogenic rests in infants that do not relapse and show a post-induction metanephric mesenchyme gene expression pattern; Subset 5 (S5; ~70% of FHWT) is exemplified by tumors with 11p15 LOI or LOH that arise within PLNRs and have the gene expression pattern of pre-induction metanephric mesenchyme. The remaining three subsets (S2-4) are defined by a low WT1 expression pattern, are often accompanied by WT1 mutations and/or deletions, and arise within ILNRs. Subset 2 (~15% of FHWT) arises in young infants, shows high expression of muscle-related genes, has an excellent prognosis, and has a gene expression pattern of the intermediate mesoderm. Subset 3 (S3; ~10% of FHWT) arises in older children, has a higher relapse rate, and a pre-induction metanephric mesenchyme gene expression pattern without high expression of muscle genes. Subset 4 (S4; ~5% of FHWT) has a gene expression profile similar to that of S2 tumors (that of the intermediate mesoderm), but occurs in older children, and has the highest relapse rate.

In the US, WTs are treated with primary resection (if possible), followed by stage-specific adjuvant chemotherapy, whereas in Europe, neoadjuvant chemotherapy followed by resection is the preferred treatment [Dome et. al, 2013]. Over 95% of WTs are classified as Favorable Histology because they lack evidence of anaplasia (presence of nuclear hyperchromasia and enlargement with atypical mitoses, often accompanied by TP53 mutations) [Beckwith and Palmer, 1978; Bardeesy et al., 1994]. Patients with FHWTs, the subject of this study, overall have an excellent survival (~90%); however, over 15% relapse and approximately 40% of these patients eventually die from their disease [Dome et al., 2002]. The National Cancer Institute’s (NCI) “Therapeutically Applicable Research to Generate Effective Treatments” (TARGET) initiative seeks to identify driver mutations and therapeutic targets for high-risk pediatric tumors through comprehensive integrative genomics (http://ocg.cancer.gov/program/target). We report the mutations commonly identified in FHWT and place these in their clinical, pathologic, and developmental context.

RESULTS

A discovery set of 77 pre-therapy FHWTs that subsequently relapsed was analyzed by whole genome (WGS, n=58) or whole exome (WXS, n=19) sequencing. Bioinformatic analysis identified 825 high-quality somatic, non-synonymous variants, with an average of 11 candidate mutations/case (range 2–42, Figure S1). Consistent with previous reports [Ruteshouser et al., 2008], somatic SNVs or small deletions were identified in WT1 (3 patients, 4%), in WTX (5 patients, 6.5%), and in CTNNB1 (5 patients, 6.5%). Unexpectedly, 12 somatic variants were identified in miRNAPG in 11 patients (14%). Lastly, eight tumors (10%) had variants in either the SIX1 or SIX2 homeodomain; strikingly, 5/8 SIX1/2-mutant tumors also had mutations in miRNAPG. None of the miRNAPG or SIX1/2 variants were annotated as polymorphisms in dbSNP versions 134 and 135 [Sherry et al., 2001] or in the 1000 Genomes Pilot Projects 1, 2, and 3 [Abecasis et al., 2012], and none had been previously identified in COSMIC Version 69 [Forbes et al., 2010]. All variants were verified and expressed by mRNA-sequencing (mRNAseq). All were predicted to be deleterious by PolyPhen Version 2 [Adzhubei et al., 2010].

Recurrent SIX1/2 homeodomain hotspot mutations

In the discovery set, 4 SIX1 and 4 SIX2 mutations involved the same location (p.Q177R) in the SIX homeodomain responsible for DNA binding and protein interaction (Table S1) [Christenson et al., 2008]. Q177 resides in a region conserved in 95% of 100 homologous proteins by a UniProt sequence similarity search (http://www.ncbi.nlm.nih/gov/pubmed/24253303); by Protein Homology/Analogy Recognition Engine Version 2 (Phyre 2) [Kelly et al., 2009], this glutamine residue was predicted to specifically interact with DNA (Figure 1A). All mutations except one (a tumor with copy neutral LOH for chromosome 2) were heterozygous, with both alleles highly expressed by mRNAseq (Figure 1B). In the validation set of 534 FHWT, the same SIX1 and SIX2 p.Q177R missense mutations were identified in 23 and 13 patients, respectively, for an overall frequency of 6.7%. An additional SIX2 variant involved p.Y129N (Table S2, Figure 1C).

Figure 1. Recurrent SIX1/2 Q177 mutations in FHWT.

(A) Phyre2 images showing the Q177 residue of SIX1 to be located within a conserved region (top panel) and is predicted to be a specific DNA-contact base (bottom panel; blue and red indicate regions of low and high likelihood of DNA contact, respectively).

(B) Coverage of the reference allele (blue bar) and variant allele (red bar) as determined by mRNA-sequencing for SIX1 and SIX2 mutant tumors.

(C) Location of validation set variants within the SIX1 and SIX2 proteins; number of variants detected are provided in parenthesis.

(D) Unsupervised NMF clustering of 75 FHWT with annotation of mutations identified (bottom, Red= somatic, Gray = germline)

(E) Supervised hierarchical clustering of 75 FHWT according to the top 100 genes differentially expressed in SIX1/2-mutant tumors with annotation of tumors with SIX1/2 mutations and copy number changes (blue = gain; dark red = loss) shown at the bottom.

(F) Boxplots of MEIS2 and CCND2 in SIX1/2-mutants versus wild-type FHWTs. The bottom and top of the box represent the first and third quartiles, respectively, the band inside the box represents the median, and the whiskers represent the maximum and minimum values.

(G) Boxplots of MEIS2 and CCND2 in NMF cluster 2 SIX1/2-mutants versus wild-type tumors.

See also Figure S1 and Tables S1–S3.

Gene expression characteristics of SIX1/2 Q177R mutant tumors

Global gene expression analysis was performed with 75 discovery samples that passed quality control. Unsupervised analysis with Non-negative Matrix Factorization (NMF) [Brunet et al., 2004], resulted in k=6 clusters having the highest cophenetic correlation (0.95) after k=2. NMF Cluster 2 contained all 7 evaluable SIX1/2 mutant tumors (Figure 1D). Gene Set Enrichment Analysis (GSEA) [Subramanian et al., 2005] comparing the 7 evaluable SIX1/2-mutant tumors with all SIX1/2 wild-type tumors identified no significantly enriched canonical pathways, GO biologic processes, or oncogenic signatures. Hierarchical analysis of the 100 top ranked differentially expressed genes in the SIX1/2-mutant tumors by GSEA (Table S3) reveals a similar distinctive expression pattern for both SIX1- and SIX2-mutant tumors (Figure 1E), suggesting that they share a common function. Tumors showing copy number change at the SIX1/2 loci did not cluster with the mutant tumors, indicating that the mutations may have a neomorph function. The NMF clusters were then compared with the Subsets previously outlined in the introduction. All tumors in NMF Cluster 2 were members of Subset 5, previously characterized by their similarity to the pre-induction metanephric mesenchyme, including high expression of SIX1, PAX2, EYA1, SALL1, MEOX1, MEIS2, WASF, and CCND2 [Gadd et al., 2012] (Figure 1F). To identify genes characterizing SIX mutations within NMF Cluster 2, the SIX1/2-mutant tumors were compared to the SIX1/2 wild-type tumors in this cluster, identifying 100 top ranked genes (Table S3), two of which (CCND2, p= 0.0001 and MEIS2, p=0.001) are illustrated in Figure 1G.

Recurrent miRNAPG hotspot mutations

We identified recurrent somatic mutations in DROSHA (recently reported [Rakheja et al., 2014; Torrezan et al., 2014]) and in DGCR8. Lastly, a somatic variant in XPO5 was identified (Table S1). The frequency of each mutation was established within the validation cohort (Table S2, Figure 2A).

Figure 2. Recurrent miRNAPG hotspot mutations in FHWT.

(A) Location of validation set variants within the DGCR8, DROSHA, XPO5, and DICER1 proteins; number of variants detected are provided in parenthesis.

(B) Coverage of the reference allele (blue bar) and variant allele (red bar) as determined by mRNA-sequencing for DROSHA (top panel) and DGCR8 (bottom panel).

(C) Supervised hierarchical clustering of 75 FHWT according to the top 100 genes differentially expressed in miRNAPG-mutant tumors with annotation of miRNAPG mutations (red= somatic, gray = germline) and copy number loss.

(D) Mature Let-7a average ddCt (left panel) and primary Let-7a average ddCt (right panel) in FHWT with miRNAPG mutations (red bar), copy number loss (blue bar), and lacking both miRNAPG mutations and copy number loss (black bar). Error bars = +SEM.

(E) Hierarchical analysis of the 43 miRNAs significantly differentially expressed in somatic miRNAPG mutant FHWTs compared with those lacking both miRNAPG mutations and copy number loss with annotation of miRNAPG mutations and copy number loss. Blue and yellow represent relatively high and low expression, respectively. Five clusters were observed, as indicated at the bottom.

See also Tables S4–S6.

DROSHA

In the discovery set, 8 somatic DROSHA mutations were identified in 7 patients. Six were missense mutations involving exon 29 in the RNase IIIB domain responsible for cleaving the 5′ end of primary miRNAs to form precursor miRNAs [Winter et al., 2009], including p.E1147K (4), p.D1151A (1), and D1151G (1). These residues reside within a 95% conserved region of the protein by UniProt. The two remaining variants were nonsense mutations in one patient (PAKZHF) resulting in loss of both RNase III domains. Therefore, all mutations affected the RNase IIIB domain of DROSHA. All DROSHA mutations were heterozygous, and mRNAseq confirmed equivalent levels of the mutant and wild-type genes with the exception of PAKZHF, which had discordant transcript ratios of 85% (p.Q46*) and 14% (p.R414*) (Figure 2B). These findings support the reported evidence of a dominant-negative mechanism for p.E1147 mutations [Rakheja et al., 2014; Wegert et al., 2015]. Within the validation set (534 tumors), 59 DROSHA variants were identified in 58 patients for a frequency of 11%; 42/59 variants were either p.E1147 (38) or p.D1151 (4). Two nonsense variants occurred in one patient. The remaining 15 were missense with recurrent mutations in the RNase IIIA domain (p.E969, 4 tumors) and the RNase IIIB domain (p.E1222, 3 tumors).

DGCR8

In the discovery set, 3 p.E518K mutations were identified in the double-stranded RNA binding domain. E518 is in a region conserved in 95% of the 100 top sequences by UniProt. The second DGCR8 allele was deleted in all mutant tumors, and mRNAseq confirmed an allelic fraction of >90% (Figure 2B). These observations are likewise reported by Wegert et al. [2015]. In the validation set of 534 tumors, 20 DGCR8 variants (17 p.E518K mutations, a missense variant in exon 12, and two nonsense variants in exon 2) resulted in an overall frequency of 4% in FHWT.

XPO5

One heterozygous somatic nonsense XPO5 mutation was identified in the discovery set, resulting in loss of the C-terminus required for binding pre-miRNAs [Okada et al., 2009]. Within the validation set, 10/534 non-recurring, damaging XPO5 variants spanning the length of the transcript were identified in 7 patients, for a frequency of 1%.

miRNAPG mutant tumors have reduced expression of critical miRNAs

In the NMF analysis previously described (Figure 1D), cluster 2 contained 8/11 tumors with somatic miRNAPG mutations as well as all SIX1/2-mutant tumors. Clusters 4 and 5 contained the remaining somatic miRNAPG-mutant tumors, one each in DROSHA p.D1151G, DGCR8 p.E518K, and XPO5. GSEA analysis comparing miRNAPG-mutant tumors (n=11) with the remainder (n=64) revealed significant negative enrichment of three gene lists, two of which contain genes up-regulated in breast cancer cell cultures over-expressing either MYC-C or E2F3 (Table S4), suggesting that MYC and E2F3 are relatively inactivated in miRNAPG-mutant tumors compared with other WTs. Hierarchical analysis using the 100 top ranked genes differentially expressed in tumors with miRNAPG somatic mutations by GSEA (Table S3) shows clustering of all somatic miRNAPG variants (Figure 2C), supporting a similar underlying mechanism of action. Further, tumors showing copy number loss at any of the miRNAPG loci did not cluster with the mutant tumors unless they also contained miRNAPG mutations. Intriguingly, DICER1 was expressed at significantly higher levels in tumors with somatic miRNAPG mutations (p<0.001, Table S3).

Mutations in miRNAPGs are expected to result in decreased mature miRNAs and increased primary miRNAs [Winter et al., 2009]; this has recently been documented in 3 mutant and 5 wild-type tumors [Rakheja, et al., 2014]. To confirm this, we analyzed mature and primary Let-7a miRNA expression within 77 discovery tumors. Given that haploinsufficient miRNAPGs may effect function [Lambertz et al., 2010], the samples were analyzed as three groups: those with somatic miRNAPG mutations (n=11), those without mutations but with copy number loss of miRNAPG loci (n=10), and those without either miRNAPG somatic mutations or copy number loss (n=56). The expression of mature Let-7a was significantly lower in both tumors with somatic miRNAPG mutations (p= 0.004) and those with miRNAPG copy number loss (p= 0.047), compared with those tumors lacking either (Table S5, Figure 2D). While the expression of the primary Let-7a transcript (PRI-Let-7a) was higher in the miRNAPG-mutant group compared with those lacking either mutations or copy number loss, this did not achieve statistical significance.

The effect of miRNAPG mutations on the global miRNA landscape was evaluated by miRNAseq. Two-class Significance Analysis of Microarray Sequencing [Tusher et al., 2001] comparing the 11 miRNAPG-mutant tumors with the 56 tumors lacking both mutations and copy number loss identified 43 differentially expressed miRNAs (FDR<1% and BH-corrected p<0.05) (Table S6). Hierarchical analysis of all 77 tumors using these 43 miRNAs shows localization of 10/11 miRNAPG-mutant tumors within a single cluster (cluster 4 in Figure 2E). Tumors with only SIX1/2 mutations (lacking miRNAPG mutations) and tumors with copy number loss of the miRNAPG loci without concomitant miRNAPG mutations did not cluster with the mutant tumors. Cluster 4 is characterized by decreased expression of the entire miR-200 family (miR-200a, -200b, -141, and -429) and miR-181b, all of which are involved in MET and stem cell maintenance [Hua et al., 2013; Ceppi et al., 2010; Park et al., 2008; Ceppi and Peter, 2014]. Decreased expression of the miR-200 family is predicted to reduce MYC and E2F expression [Hua et al., 2013], as was observed in our GSEA analysis of the miRNAPG-mutant tumors.

Correlating SIX1/2 and miRNAPG Mutations with Clinicopathologic Features

Analysis of the validation set revealed a significant female predominance in tumors with DGCR8 E518K and DROSHA exon 29 (miRNAPG-HS) mutations and a greater prevalence of tumors with blastemal predominant histology in patients with miRNAPG-HS and/or SIX1/2 Q177R mutations (Table 1). There was also a significantly higher association with PLNRs and a lower association with ILNRs in those tumors with SIX1/2 Q177R and miRNAPG-HS mutations (Table 1). Since PLNRs are associated with loss of the normal imprinting pattern at 11p15 [Ravenel et al., 2001], 11p15 methylation was analyzed within the 77 discovery set tumors. LOH, LOI, and retention of imprinting (ROI) were identified in 29/77 (38%), 30/77 (39%), and 18/77 (23%), respectively in the entire group. LOI was significantly more frequent in both those tumors containing miRNAPG-HS mutations (7/9 patients, 78%, p= 0.011), and in those with SIX mutations (7/8 patients, 87.5%, p= 0.003).

Table 1.

Significant Validation Set Patient Characteristics

Group	Total Number	Age at Diagnosis (months)	Gender (F:M)	Blastemal Histology	PLNR	ILNR
DGCR8 E518K Variants	17	57	15:2 (p=0.004)	32/59 (p=0.003)	27/55 (p=<0.001)	8/55 (p=0.079)
DROSHA exon 29 Variants	42	51	31:11 (p=0.009)
SIX Q177R Variants	36	52	22:14 (not significant)	20/36 (p=0.008)	16/35 (p=0.001)	3/35 (p=0.026)
Entire Validation Set	534	44	290:244	189	119	125

Integration of the above clinicopathologic features with mutation, copy number, 11p15 imprinting status, and membership in gene expression subsets and in miRNA expression categories is provided in Figure 3, arranged by NMF cluster. The NMF cluster (Figure 3, first row) correlates closely with the previously reported gene expression subsets (Figure 3, second row, assigned as shown in Figure S2) with the exception that S5 is represented most prominently within two NMF clusters, clusters 1 and 2. NMF cluster 2, which includes the majority of the miRNAPG mutations and all SIX1/2 mutations, demonstrates a predominance of tumors in miRNA cluster 4 (defined largely by low miR-200 family expression), a high prevalence of both blastemal histology and PLNRs, and a high frequency of 11p15 LOI. The second large S5 predominant group, NMF Cluster 1, lacks miRNAPG mutations, does not show reduction of the miR-200 family, and is associated with ILNRs rather than PLNRs. This cluster shows a relatively high frequency of DICER1 loss (through loss of chromosome 14). Loss of one DICER1 allele results in partial impairment of miRNA processing [Gurtan et al., 2012] and promotes tumorigenesis [Kumar et al., 2007; Lambetz et al., 2010]. These findings suggest that almost half of the large S5 group may be driven by miRNAPG and/or SIX mutations and the possibility remains that DICER1 loss may contribute to pathogenesis in some of the remaining S5 tumors.

Figure 3. Integrative Analysis of Non-negative Matrix Factorization Clusters.

Clinical, pathologic, and genetic features of FHWT arranged according to the NMF identified in Figure 1D. The key is illustrated at the bottom. See also Figure S2.

NMF clusters 3 and 4 contain the majority of the WT1, WTX, and CTNNB1 mutations, show a high frequency of membership in S3 and S4 (Figure 3, second row), and are characterized by mixed histology and an association with ILNRs. The presence of two tumors in NMF cluster 4 with somatic miRNAPG mutations associated with membership in S4 (characterized by a gene expression pattern of the intermediate mesoderm and a high relapse rate) suggests that miRNAPG mutations may also be pathogenic when they occur earlier in renal development, within the intermediate mesoderm. The overlap of miRNAPG-HS and SIX mutations with WT1, WTX, or CTNNB1 mutations was evaluated in the validation set. Of 36 patients (7%) with WT1 variants, 1 DGCR8 E518K and no DROSHA or SIX variants were present. WTX variants were identified in 31 patients (6%); in these 1 DGCR8 E518K, 1 DROSHA exon 29, and 3 SIX Q177R variants were present. Of 62 patients with CTNNB1 variants (12%), 3 DROSHA exon 29 variants, 1 SIX Q177R variant, and no DGCR8 variants were also present; in contrast, 18 also had WT1 variants (a recognized association [Maiti et al., 2000]). Of note, we were unable to evaluate exonic deletions of WT1and WTX in this data, which represent ~70% of the genetic aberrations that occur at these two loci [Ruteshouser et al., 2008; Gadd et al., 2012].

The combination of SIX1/2 and miRNAPG mutations results in poor outcome and RAS activation

Within the discovery set containing FHWT that subsequently relapsed, 5/8 (63%) tumors with SIX1/2 mutations also had somatic miRNAPG-HS mutations. In the validation set of 534 tumors, of 36 tumors with SIX1/2 Q177R variants, 10 (28%) also had a DGCR8 E518K (1) or DROSHA exon 29 (9) variants (p= 0.0015). There was no significant difference in either the rate of relapse (30%) or the number of deaths (14%) in the entire validation set compared with patients with DGCR8 E518K or DROSHA exon 29 variants without SIX Q177R variants (31% and 14%, respectively), or in those with SIX Q177R variants without associated miRNAPG-HS variants (31% and 15%, respectively). However, the 10 patients whose tumors contained both miRNAPG-HS and SIX Q177R variants had a significantly higher relapse rate (8/10, 80%, p= 0.0001, Figure 4) and a higher rate of death (40%). Hence, while the miRNAPG-HS and SIX Q177R variants alone do not portend a worse outcome, the combination of these mutations, while rare, appears to result in a worse outcome.

Figure 4. Disease Free Survival.

Kaplan-Meier curve of disease free survival in the following four validation set groups: (1) tumors with SIX1/2 and miRNAPG-HS variants (black line), (2) tumors with miRNAPG-HS variants without SIX1/2 variants (blue line), (3) SIX1/2 variants without miRNAPG-HS variants (green line), and (4) all other validation set tumors (red line).

To identify possible therapeutic targets for this group, we identified previously published Affymetrix U133A data on 291/534 validation tumors and deposited these in the TARGET Data Matrix [Gadd et al., 2012]. GSEA analysis of 22 tumors with miRNAPG-HS variants and without SIX1/2 variants and the analysis of 12 tumors with SIX1/2 Q177R variants and without miRNAPG-HS variants did not identify significant enrichment for canonical pathways, GO biologic processes, or oncogenic signatures, similar to our experience with the discovery set. In contrast, analysis of the 6 tumors with available gene expression data with miRNAPG-HS variants in combination with SIX1/2 Q177R variants demonstrated a large number of highly significantly (FDR<5%) enriched gene lists (Table S4). While the number of samples is small, of interest is the positive enrichment of 7 gene sets differentially expressed in a variety of tumor types following over-expression of an oncogenic KRAS mutation, 4 genes sets differentially expressed in a medulloblastoma cell line following knock-down of PCGF2 (a polycomb group protein that functions by transcription repression), and two gene sets up-regulated during embryoid body differentiation.

Germline variants in miRNAPG are identified in FHWT

Constitutional DICER1 mutations result in development of pleuropulmonary blastoma syndrome (PPBS), which includes cystic nephroma, and extremely rarely, WT [Doros et al., 2014; Hill et al., 2009; Slade et al., 2011]. Given the recent documentation of rare DICER1 and DROSHA germline variants in patients with WT [Rakheja et al., 2014], we examined the discovery set for germline exonic variants in the miRNAPGs and SIX1/2 and identified 4 variants (Table S1, Figure 2A). These were verified with Sanger sequencing, expressed by mRNAseq, predicted to be damaging by PolyPhen Version 2 [Adzhubei et al., 2010], not annotated as polymorphisms in dbSNP versions 134 and 135 [Sherry et al., 2001], or present in the 1000 Genomes Pilot Projects 1, 2, and 3 [Abecasis et al., 2012]. Further, these variants were not identified in a dataset of over 200,000 individuals in the NHLBI GO Exome Sequencing Project (http://evs.gs.washington.edu/EVS/). A DROSHA variant (p.P82T) involved the proline-rich domain implicated in protein-protein and protein-nucleic acid interactions [Nicholson, 2014]. The remaining allele was retained in the tumor sample. A germline XPO5 nonsense variant p.R159*, located in the Exportin-1 domain, was identified in the same patient who had the somatic nonsense XPO5 mutation. Two missense germline DICER1 variants were identified, one (p.R1368C) within the RNase IIIA domain, and one (p.I85M) in the helicase ATP-binding domain. Neither of the tumor samples showed loss of the remaining DICER1 allele. Analysis of the validation set revealed 9/534 damaging DICER1 variants in 8 patients, for an overall frequency of 1.5%. Eight variants affected the RNase IIIB domain and 6/8 were at previously identified hotspot locations [Doros et al., 2014]. Analysis of the mRNA and miRNA expression patterns of the 3 germline mutant tumors that lacked a somatic mutation revealed patterns distinct from that of the somatic miRNAPG mutations (Figures 1D, 2C, 2E, and 3).

DISCUSSION

Wilms tumor (WT) is an embryonal tumor of the kidney remarkable for its replication of early renal development. While mutations or deletions in WT1, WTX, and/or CTNNB1 are found in approximately 30% of WTs [Ruteshouser et al., 2008], the underlying pathogenesis of most WTs remains unknown. Hence, in the last few years several groups of investigators simultaneously embarked on in-depth molecular characterization studies to further elucidate the genetic landscape of WTs. Two groups recently reported their findings [Rakheja et al., 2014; Torrezan et al., 2014] and another group is reporting their findings in this journal [Wegert et al., 2015]. All point to the importance of the DROSHA E1147K missense mutation in the development of WTs. Torrezan et al. evaluated a family trio by WXS, revealing a DROSHA E1147K mutation and prompted sequencing of the DROSHA RNase IIIB domain in a validation set of 221 FHWTs (including a mixture of pre- and post-therapy samples from patients treated on different protocols). Rakheja et al. performed WXS in 15 patients followed by validation in 29 tumors, identifying three somatic DROSHA mutations (2 E1147K and 1 D1151Y). Our study offers the benefits of a much larger discovery set (77 tumors) and validation set (534 tumors) comprised exclusively of pre-therapy samples. This allowed for the identification of 1) recurrent somatic mutations in DGCR8 E518K, 2) recurrent DROSHA mutations other than E1147K, 3) recurrent SIX1/2 homeodomain mutations, 4) the association between 11p15 LOI and miRNAPG and SIX mutations, and 5) decreased expression of the miR-200 family in miRNAPG-HS mutant tumors, supporting the role of MET arrest in the function of these mutations. Lastly, the study of a large number of patients treated on a cooperative group protocol allowed documentation of the impact of mutations on clinical and pathologic features, including the association with blastemal histology, nephrogenic rest status, timing in renal development, the female predominance in miRNAPG-mutant tumors, and the poor clinical outcome of patients with both miRNAPG-HS and SIX1/2 mutations. The identification of RAS activation in such tumors suggests they may be treatable in the future with precision medicine.

Synthesis of mature miRNA requires normal function of DGCR8, DROSHA, XPO5, and DICER1. In brief, primary miRNAs (pri-miRNA) are cleaved in the nucleus by the DROSHA-DGCR8 microprocessor complex to form precursor miRNA (pre-miRNA), which are exported from the nucleus by XPO5. Within the cytoplasm, DICER1 cleaves the pre-miRNA to form mature miRNAs [Winters et al., 2009]. Given the multitude of cellular pathways miRNAs are known to affect, combined with their interactions and feedback loops, the range of effects associated with miRNAPG mutations is likely to be heterogeneous and complex [Hua et al., 2013]. Impaired miRNA synthesis has been shown to accelerate oncogenic transformation by deregulating target oncogenes and globally reducing mature miRNA levels [Kumar et al., 2007]. Further, miRNAs have an essential and unique role during mammalian kidney development [Bartram et al., 2013; Ho et al., 2013]. Recent studies have shown that DROSHA RNase IIIB mutations result in global impairment of miRNA processing, with specific impairment in tumor-suppressing miRNAs [Rakheja et al., 2014]. We demonstrate that miRNAPG mutations are associated with down-regulation of all members of the miR-200 family (miR-200a, -200b, -200c, -141, and -429), which are key regulators of MET [Hua et al., 2013]. Reduction of miR-200 results in a mesenchymal, highly motile, and aggressive phenotype of cancer cells [Ceppi et al., 2014; Park et al., 2008; Ceppi et al., 2010]. MET is a critical step in early renal development during which the capacity to form nephrons occurs [Kobayashi et al., 2008]. Therefore, decreased expression of these miRNAs in the pre-induction metanephric mesenchyme would prevent MET, resulting in failure of epithelial differentiation and a predominance of undifferentiated cells, as was seen in the miRNAPG-mutant tumors we report.

Another miRNA, Let-7a, has long been linked to tumor development [reviewed in Garzon et al., 2009] and decreased Let-7a expression has been implicated in the development of WT via up-regulation of LIN28 [Urbach et al., 2014]. LIN28 is a RNA-binding protein that specifically binds to PRI/PRE-Let-7 miRNAs, preventing maturation [Viswanathan et al., 2009]. In the murine embryonic kidney, over-expression of LIN28 within the pre-induction metanephric mesenchyme results in sustained proliferation, failure of MET, and tumor formation, a process that is rescued by Let-7 over-expression [Urbach et al., 2014]. Of particular interest, over-expression of LIN28 in the post-induction metanephric mesenchyme failed to result in tumor development suggesting that the effects of miRNAPG mutations depend on the cellular context in which they arise. Rakheja et al. functionally confirmed that DROSHA E1147K and D1151Y mutations result in decreased expression of the Let-7 family within an in vitro model; in this study we now confirm decreased Let-7a in a large population of miRNAPG-mutant WTs.

Non-recurrent germline variants were also identified in miRNAPG, although these tumors were outliers by gene expression and miRNA expression compared with those containing somatic mutations. Therefore, germline miRNAPG variants do not appear to function in the same manner as the somatic miRNAPG mutations, and there is no direct evidence that they are pathogenic. Germline DICER1 variants constitute the greatest clinical concern due to their association with familial PPBS. However, available data suggest that this syndrome follows a classic two-hit model of tumorigenesis, with germline truncating DICER1 mutations followed by deleterious somatic missense mutations involving the RNase IIIB domain [Doros et al., 2011]. Both germline and somatic DICER1 mutations were observed in 3 WTs by Wu et al. [2013] and in one WT by Rakheja et al. [2014], who also describe an additional patient with a germline DICER1 mutation only, similar to our two discovery cases. The contribution of germline DICER1 variants (as well as germline DROSHA and XPO5 variants) in such patients is not clear. However, their repeated identification in patients with WT suggests they may result in a predisposition to WT; our study documents this risk to be present in approximately 1.3% of FHWT.

Recurrent mutations involving a specific residue of the homeodomain of transcription factors SIX1 and SIX2 were identified in 7% of FHWT. The highly homologous SIX1 and SIX2 genes have a critical role in renal development [Christensen et al. 2008; Zu et al. 2003]. SIX1-deficient mice exhibit renal hypoplasia or agenesis [Li et al., 2003], and SIX1 mutations have been reported in the Branchio-Oto-Renal syndrome (BOR), although at positions distant from Q177R; BOR syndrome is not associated with WT [Patrick et al., 2009]. SIX2 maintains a population of undifferentiated renal blastemal cells, and loss of SIX2 results in premature differentiation of mesenchymal cells into epithelia [Kobayashi et al., 2008; Self et al. 2006]. Within a renal cell line, over-expression of SIX2 results in an increased percentage of cells in the S-phase and increased migration [Senanayake, et al., 2013]. Given the known function of SIX1 and SIX2, the localization of mutations within the SIX homeodomain, and the high expression of both mutant and wild-type alleles, it is probable that these SIX1/2 mutations in FHWT are activating, resulting in failure of MET and continued proliferation of the metanephric mesenchyme. Indeed, we demonstrate significant up-regulation of CCND2 in SIX-mutant tumors, which may be an important underlying cause of the continued proliferation. These findings are supported by those of Wegert et al. [2015], who report increased expression of both cell cycle genes and genes highly expressed in the pre-induction metanephric mesenchyme.

In summary, mutations in miRNAPG and/or SIX1/2 genes are identified in approximately 20% of FHWT. Clinical, pathologic, gene and miRNA expression data support disruption of MET at the time of induction as the underlying mechanism of tumorigenesis in this group of WTs, although no direct functional data is presented to confirm this hypothesis. Lastly, the very high prevalence of 11p15 LOI in WTs harboring both miRNAPG-HS and SIX Q177R mutations provides further evidence that multiple genetic events may be involved in the development and progression of WTs.

EXPERIMENTAL PROCEDURES

The TARGET initiative maintains public availability of the gene expression, chromosome copy number, DNA methylation, sequence analysis (i.e. MAF and summary files), and clinical information for the cases studied (available through the TARGET Data Matrix;http://target.nci.nih.gov/dataMatrix/TARGET_DataMatrix.html) in fully annotated MIAME compliant MAGE-TAB files describing the methods, specimen processing details, and quality control parameters. The aligned sequencing data (BAM files) are deposited in the Sequence Read Archive (SRA) at the National Center for Biotechnology Information and are accessible through the database of genotypes and phenotypes (dbGAP, http://www.ncbi.nlm.nih.gov/gap) under the accession number phs000471. See Supplemental Experimental Procedures for details.

Specimens

Pre-therapy tumor and normal DNA from peripheral blood or kidney from 77 FHWTs banked by COG with parental informed consent were included in the discovery set. A validation set of patients registered on the NWTS-5 protocol included all patients with available primary tumor DNA who subsequently relapsed and a random selection of all patients irrespective of relapse, resulting in 534 tumors enriched for relapse. Studies were performed with the approval of the Lurie Children’s Hospital Institutional Review Board.

DNA Sequencing

WGS libraries were sequenced using Complete Genomics Inc. (CGI) technology [Drmanac et al., 2010]; alignment of reads to the NCBI Build 37 reference human genome assembly was performed by the CGI Cancer Sequencing service analytic pipeline version 2 [Carnevali et al., 2012]. WXS was performed on the Illumina HiSeq platform. Variant calling from the aligned BAM files was performed using both ATLAS and SAMtools and annotation and filtering was performed using the SACBE annotation pipeline [Bainbridge et al., 2013; Lupski et al., 2013] as well as Bambino Version 1.05 [Edmonson et al., 2011].

mRNA and miRNA Sequencing

Libraries were prepared following a paired-end protocol and sequenced on the Illumina HiSeq 2000/2500 platform using HiSeq Control Software version 2.0.10, aligned to GRCh37-lite genome-plus-junctions reference [Morin et al., 2008] using BWA Version 0.5.7 [Li et al., 2009]. For mRNA analysis, variants were detected on positive- and negative-split BAMs separately and annotated with SnpEff [Cingolani et al., 2012a] (Ensembl 66) and SnpSift [Cingolani et al., 2012b] (dbSNP137 and COSMIC64). For miRNA analysis, reads aligning to known miRNAs in miRBase v20 were summed and normalized to a million miRNA-aligned reads to generate the quantification files.

Target Capture Sequence Analysis

Probes were designed using Agilent’s SureDesign (https://earray.chem.agilent.com/suredesign/) and probe density was specified at 2x with 98.7144% coverage of the target region (Agilent SureSelect XT Custom 0.5–2.9Mb probes). Genomic DNA libraries were constructed as described above and hybridized to the RNA probes. Post-capture material was enriched with 10 cycles of PCR. Paired-end 100 base reads were sequenced on the Illumina HiSeq2500 instrument. SNVs were filtered out if not predicted to be damaging by at least 2/3 of the following: SIFT [McLaren et al., 2010], PolyPhen Version 2 [Adzhubei et al., 2010], or Provean Version 1.1.3 [Choi et al., 2012].

Copy Number Analysis was performed on tumor and normal pairs according to the manufacturer’s protocol for the AffyMetrix 6.0 SNP array and processed using AffyMetrix Genotyping Console 4.0 software. Reference normalization utilizing a diploid chromosome for each sample [Pounds et al., 2009] was performed in R using the DNAcopy BioConductor package. Segmented regions were identified by Circular binary segmentation (CBS) and those containing at least 8 markers in which the log2 value was ≥ +0.5 or ≤ −0.5 were considered gained or lost, respectively.

Gene Expression Analysis was analyzed with the Affymetrix U133+2 chip, according to the manufacturer’s protocol using the Gene-Chip Operating Software and normalized using robust Multichip Average normalization. Unsupervised analysis was performed using Non-negative Matrix Factorization Consensus Version 5 [Brunet et al., 2004]. Gene Set Enrichment Analyses, version 2.0.14, (http://www.broadinstitute.org/gsea) [Subramanian et al., 2005] were run using 1000 permutations and phenotype permutation. Significant enrichment was defined as those lists with >50 genes, a FDR <10%, and a p-value <5%.

Methylation Analysis was performed with Illumina Infinium Human Methylation 450K Bead Chips (Illumina, San Diego, CA), according to the manufacturer’s protocol. Methylation levels for all probes in imprint control regions ICR1 (IGF2/H19) and ICR2 (KCNO1/CDKN1C) were averaged. ROI was defined as 0.3–0.7 for ICR1 and ICR2, LOI as 0.8–1 for ICR1 and 0.3–0.7 for ICR2, and LOH as 0.8–1 for ICR1 and 0–0.2 for ICR2. Tumors outside of these ranges were not classified.

Let-7a Mature and Primary miRNA Expression

Reverse transcription, amplification, and real time PCR were performed per the manufacturer’s protocol (Life Technologies Corporation, Carlsbad, CA). Samples were run in triplicate and analyzed using the Applied Biosystems® 7500 Fast SDS Software (Life Technologies Corporation, Carlsbad, CA). Nuclear and cytoplasmic fractions from HEK293 cells (ATCC, Manassas, VA) were evaluated for mature and primary Let-7a expression to confirm probe specificity.

SIGNIFICANCE.

We establish recurrence of DGCR8 E518K mutations, confirm the high frequency of recurrent DROSHA exon 29 mutations, identify recurrent hotspot mutations in the SIX1/2 homeodomain, and identify high frequencies of 11p15 LOI in both miRNAPG and SIX1/2 mutant FHWT. Mutations in miRNAPG were associated with dysregulation of microRNAs involved in oncogenesis and mesenchymal-to-epithelial transition, and increased frequency of undifferentiated histology. The combination of multiple genetic events in some FHWT, including mutations in both miRNAPG and SIX1/2, and 11p15 imprinting abnormalities, provides evidence of a complex, multi-step process resulting in failure of normal differentiation, maintenance of progenitor cells, and support of proliferation. The critical combination of such genetic events is shown to result in an adverse outcome and may be targetable.