Introduction

Wilms tumour is the most common childhood genitourinary tract cancer and the third most common paediatric solid tumour. It comprises 95% of all renal cancers and 6% of all cancers diagnosed among children <15 years of age1,2. Depending on tumour histology and disease stage, treatment consists of nephrectomy and combination chemotherapy, with or without radiotherapy. Although the overall survival of Wilms tumour patients has increased to 85% at 5 years after diagnosis3, children with diffusely anaplastic or unfavourable histology Wilms tumour continue to have poor outcomes, with 4-year survival rates of 56% for Stage III and 17% for Stage IV4,5,6. Moreover, the success of modern therapy comes at a price in that up to 70% of Wilms tumour survivors develop chronic health problems as young adults, including but not limited to renal failure, congestive heart failure, interstitial pulmonary fibrosis, kyphoscoliosis, infertility, intestinal obstruction and second malignancies7,8,9.

Few targetable molecular lesions have been defined in Wilms tumours. Recurrent genetic changes in sporadic Wilms tumours have been described, including mutations in WTX, WT1, CTNNB1 and TP53. However, these mutations account for only about 1/3 of Wilms tumours10. Children with DICER1 syndrome, who carry one germline-null allele of DICER1, are susceptible to tumours, including pleuropulmonary blastoma and Wilms tumour, suggesting that defects in microRNA (miRNA) processing may increase susceptibility to this disease11,12,13. This idea has further been reinforced by the recent description of somatic heterozygous missense mutations in DICER1 (ref. 14) and DROSHA15 in Wilms tumours. However, the mechanisms through which heterozygous DICER1 and DROSHA mutations affect miRNA biogenesis in Wilms tumour are unknown and it is unclear whether mutations in these proteins have similar or distinct effects on miRNA expression in these tumours. Furthermore, it is unknown whether other genes are recurrently mutated in Wilms tumour beyond those already described.

Here we describe the results of whole-exome sequencing of a cohort of 44 Wilms tumours. In this study, we identify both known and novel recurrent, somatic mutations. We carry out functional studies in vitro and in genetically modified cell lines to elucidate the mechanisms by which missense mutations in DICER1 and DROSHA affect miRNA expression in tumours. These studies identify a new subclass of Wilms tumours and suggest that impaired expression of let-7 tumour-suppressing miRNAs may be a common underlying mechanism of this subclass.

Results

Exome sequencing of Wilms tumours

To discover the full range of pathogenic Wilms tumour mutations, we performed exome capture and massively parallel sequencing on a discovery set of 15 pairs of Wilms tumours and matched adjacent normal kidney cortices, and subsequently performed whole-exome sequencing on a validation set of 29 additional Wilms tumours. Table 1 lists the patient demographics and Supplementary Data 1 lists the total number of variants identified in the tumours. In the discovery set, we identified between 0 and 17 non-synonymous somatic single nucleotide variants per tumour (Fig. 1a and Supplementary Data 2). These numbers are consistent with the generally lower mutational load seen in paediatric tumours compared with tumours from adults16,17.

Table 1 Clinical characteristics of patients in this study.
Full size table
Figure 1: Mutations in the miRNA biogenesis pathway in Wilms tumours.
figure 1

(a) Summary of somatic mutations detected by exome capture and massively parallel sequencing of matched Wilms tumours and germline DNA in the discovery set. (b) Mutations in miRNA biogenesis components are mutually exclusive with WT1 and CTNNB1 mutations. (c) Domain structure of DICER1 and DROSHA, indicating positions of mutations. Metal-binding residues in the RNase III domains are indicated in red. PAZ, Piwi–Argonaute–Zwille domain; DRBM, double-stranded RNA-binding motif. (d) Electropherograms from resequencing of selected mutations. Pie charts depict read counts from Illumina sequencing for reference (blue) and variant (red) alleles. (e) Modelling of G1809V mutation (silver) in the DICER1 metal-binding site, based on the structure of G. lamblia RNase III (PDB 2QVW). Numbering of equivalent Homo sapiens DICER1 residues is indicated. Dots indicate solvent-accessible surface.

Full size image

Mutations were identified in genes previously associated with Wilms tumour as well as in genes associated with other paediatric tumour types (Fig. 1b,d, and Supplementary Data 2 and 3). For example, seven Wilms tumours harboured mutations in either WT1, or CTNNB1, or both (Fig. 1b). Four of 44 (9%) Wilms tumours had diffusely anaplastic histology, and three of these tumours harboured missense mutations of TP53, consistent with the known association of anaplastic Wilms tumours with TP53 mutation4. Somatic mutations in the chromatin-remodelling factors SMARCA4 and ARID1A, which are known to be mutated in other childhood cancers18,19,20,21, were each seen in two tumours (4.5%). MYCN was mutated at codon 44 in two tumours (4.5%), creating a P44L substitution at a highly conserved residue. The same mutation was recently reported in high-risk neuroblastoma22. MYCN is overexpressed in some Wilms tumours23 through genomic amplification24 and through mutations in FBXW7, an E3 ubiquitin ligase that regulates MYCN stability25. All mutations listed in Figure 1b and Supplementary Data 3 were confirmed by Sanger sequencing of tumour and germline DNA.

In addition to these known and novel mutations, we identified mutations in components of the miRNA biogenesis machinery in eight tumours in our cohort (18%). Primary (pri)-miRNAs are processed in the nucleus by the RNase III enzyme DROSHA (in a complex with the double-stranded RNA binding protein DGCR8) to yield smaller pre-miRNAs, which are transported to the cytoplasm and processed by DICER1 (in a complex with TARBP2) to produce mature miRNAs26. Six Wilms tumours in our cohort showed heterozygous missense mutations in DROSHA (14%); one of these tumours (CMCW39) also harboured a somatic D221G mutation in TARBP2. Another two tumours contained novel missense mutations in DICER1, and one of these tumours also harboured a germline frameshift deletion predicted to create a null allele (Fig. 1b–d). The frequency of mutations in miRNA pathway genes and specifically in DROSHA were overall similar to that reported by Torrezan et al.15 (33% and 12%, respectively). All DICER1 and DROSHA missense mutations were somatic with the exception of the DROSHA R967W mutation in CMCW41 and the DROSHA M120V mutation in CMCW53, which were also present in non-tumour tissue. One tumour (CMCW85) harbored a germline DGCR8 S720N variant. Pertinently, miRNA processing mutations were mutually exclusive from WT1 or CTNNB1 mutations, suggesting that impairment of miRNA biogenesis represents a novel mode of oncogenesis in a subset of Wilms tumours.

Functional analysis of miRNA processing mutations in vitro

DROSHA and DICER1 both possess two highly conserved tandem endonuclease domains, denoted RNase IIIA (RIIIA) and RNase IIIB (RIIIB)27. The RIIIA domains of DROSHA and DICER1 selectively process the 3′-arm of the pri-miRNA or the pre-miRNA hairpin, respectively, while the RIIIB domains selectively process the 5′-arm of the pri-miRNA or the pre-miRNA, respectively28,29 (Fig. 2a). Cleavage of double-stranded RNA is Mg2+-dependent and requires the presence of conserved metal-binding aspartate and glutamate residues in the RIIIA and RIIIB domains30,31. Recurrent somatic mutations in the metal-binding residues of the DICER1 RIIIB domain (E1705, D1709, D1810 and E1813) have been reported in sex cord-stromal cell tumours and germ cell tumours32. In vitro processing assays and cellular reconstitution experiments have demonstrated that mutations in metal-binding residues of the DICER1 RIIIB domain strongly impair maturation of 5p miRNAs (those derived from the 5′-arm of the pre-miRNA hairpin), yet still allow partial production of 3p miRNAs (derived from the 3′-arm of the pre-miRNA hairpin)28,32,33.

Figure 2: Wilms tumour DICER1 and DROSHA mutations impair pri- and pre-miRNA processing in vitro.
figure 2

(a) Schematic of pri-miRNA indicating RNase III cleavage sites. (b) DICER1 mutations affect pre-miRNA processing. Wild-type or mutant DICER1 was purified by immunoprecipitation and tested against pre-miRNA substrates. Arrowhead, mature miRNA; asterisk, processing intermediate formed by DICER1 RIIIA cleavage in the absence of RIIIB activity. (c) Mutations in the RNase III domains impair DROSHA activity. Wild-type or mutant DROSHA was purified by immunoprecipitation and tested against pri-miRNA substrates. Arrowhead, processed pre-miRNA.

Full size image

The DICER1 missense mutations that we identified in Wilms tumours occur at the highly conserved glycine 1809 residue adjacent to the metal-binding aspartate residue 1810 within the RIIIB domain (Fig. 1c). In the Giardia lamblia RNase III crystal structure34, the equivalent residue G719 is within 4 Å of a neighbouring α-helix, suggesting that the G1809V and G1809R mutations may distort the active-site geometry and impair enzyme function (Fig. 1e). To directly determine the effects of these mutations on DICER1 activity, in vitro pre-miRNA processing assays were performed (Fig. 2b and Supplementary Fig. 1a). The DICER1 mutation I1102fsdel is predicted to produce a truncation before the ribonuclease domains and, as expected, I1102fsdel has no detectable activity with any substrate (Fig. 2b). When processed by the DICER1 RIIIB mutants G1809R and G1809V, 5p miRNAs show accumulation of the expected aberrant processing product due to cleavage by the RIIIA domain but not the RIIIB domain. For some very efficiently processed 5p miRNAs, some mature miRNA is still produced by these mutants (let-7i and miR-122), whereas other 5p miRNAs are not measurably processed (miR-16-1 and miR-18a). For the tested 3p miRNAs, the G1809R and G1809V mutants produced normal (miR-1-1) or reduced (miR-124-2) amounts of the mature miRNA. These data demonstrate that the G1809R and G1809V substitutions impair activity of the DICER1 RIIIB domain and, consistent with prior findings28,33, suggest that these mutations inhibit the production of 5p miRNAs more severely than 3p miRNAs.

The somatic DROSHA mutations E1147K and D1151Y (analogous to the E1705 and D1709 positions of DICER1) alter a conserved RNase IIIB domain metal-binding residue, while the L728V mutation is not in an annotated functional domain. E1147K was also identified by Torrezan et al.15 as a mutational hotspot15. In vitro pri-miRNA processing assays revealed that the DROSHA D1151Y and E1147K mutants are functional nulls for all pri-miRNA substrates tested, while the L728V mutations does not measurably impair enzyme activity (L728V is less active in this assay, but this variant reproducibly expresses at lower levels (Supplementary Fig. 1b), precluding definitive conclusions about its activity) (Fig. 2c). We also detected two DROSHA mutations that were present in the germline: R967W that is adjacent to the metal-binding residues E969 and D973 in the RNase IIIA domain and M120V that is in the amino-terminal proline-rich domain (Fig. 1c). As germline DICER1 mutations are associated with elevated cancer susceptibility, we also tested these variants in vitro. The M120V mutation has no detectable effect on DROSHA activity, while the R967W mutation measurably reduces DROSHA activity in a substrate-specific manner (Fig. 2c). Thus, although the significance of the R967W variant is presently unclear, it is possible that analogous to germline DICER1 mutations, this allele may confer an elevated inherited risk of Wilms tumour. None of the DROSHA mutations appeared to preferentially affect processing of 5p or 3p miRNAs.

We also tested the effects of the DGCR8 and TARBP2 mutations in vitro (Supplementary Fig. 2). DGCR8 S702N shows no impairment of DROSHA interaction or miRNA processing activity. Similarly, TARBP2 D221G is equivalent to wild type in its DICER1 interactions and processing activity. Therefore, although we cannot rule out an effect of these mutations in vivo, they do not appear to detectably affect activity in vitro. Of note, in addition to mutations in DROSHA, Torrezan et al.15 also identified non-synonymous missense mutations in DGCR8 and TARBP2, as well as XPO5 and DICER1; however, the functional impact of these mutations was not tested.

miRNA expression in DICER1- and DROSHA-mutant tumours

We next determined the effect of DICER1 and DROSHA mutations on miRNA expression in tumours. Small RNAs were prepared from Wilms tumours with and without DROSHA and DICER1 mutations, and analysed using high-throughput sequencing and quantitative PCR. Although both DICER1- and DROSHA-mutant tumours remain competent to produce some mature miRNAs, both tumour classes exhibit defective production of specific miRNAs (Fig. 3a). To gain insight into the possible oncogenic mechanisms of these mutations, we examined miRNAs that were expressed at lower levels in both DICER1- and DROSHA-mutant Wilms tumours (Fig. 3a). Included in this list are many members of the let-7 family, a group of highly evolutionarily conserved 5p miRNAs that play roles in cell proliferation and differentiation, and have been implicated as tumour suppressors in human cancers35. By quantitative reverse transcription–PCR, we confirmed the lower expression of let-7 family members and several other miRNAs in all mutant and a few wild-type tumours (Fig. 3b). Therefore, our results show that mutations in DICER1 or DROSHA affect the biogenesis of selected miRNAs in tumours, and suggest that partially defective miRNA processing may drive the development of a subset of Wilms tumours via decreased expression of tumour suppressor miRNAs, including members of the let-7 family.

Figure 3: DICER1 and DROSHA mutations impair miRNA biogenesis in Wilms tumours.
figure 3

(a) Summary of differentially expressed miRNAs in DICER1 and DROSHA-mutant tumours as determined by small RNA-seq. The 39 miRNAs whose expression is lower in both DICER1 and DROSHA-mutant tumours are listed. (b) Quantitative reverse transcription–PCR validation of expression of selected miRNAs in tumours with or without mutations in DROSHA and DICER1. (c) Fold-change in expression of 5′ (green) and 3′ (grey)-derived miRNAs comparing DROSHA or DICER1 mutant tumours with non-mutant tumours. Members of the let-7 family are indicated. (d) Fraction of total miRNA reads deriving from the 5′ (green) or 3′ (grey) arm of the pre-miRNA hairpin for each tumour. Arrowheads indicate lower representation of 5′ miRNAs in DICER1 mutants.

Full size image

As predicted by the previous analyses of DICER1 RNase IIIB-domain mutations in cell lines28 and our in vitro processing experiments (Fig. 2a), DICER1 mutant tumours showed preferential impairment of expression of mature miRNAs derived from the 5′-arm of pre-miRNA hairpins compared with those derived from the 3′-arm (Fig. 3c,d). Out of 122 miRNAs expressed at lower levels in DICER1-mutant tumours compared with tumours with wild-type DICER1 and DROSHA (log2 fold-change <−0.5), 101 derive from the 5′-arm, whereas 21 derive from the 3′-arm. One of the two tumours with a G1809 DICER1 mutation also harboured a germline frameshift deletion of the other allele. Notably, although this tumour (CMCW11) exhibited the most severe depletion of 5p miRNAs in our cohort, the tumour continued to express mature miRNAs (Fig. 3d), indicating that the G1809R mutant allele retained some capacity to produce miRNAs. A similar 3p skewing of miRNA expression was recently reported in pleuropulmonary blastomas occurring in patients with germline-null and somatic missense mutations in DICER1 (refs 36, 37).

DROSHA RNase IIIB mutations dominantly impair miRNA biogenesis

DROSHA acts at the pri-miRNA processing step, and, as expected, DROSHA mutant tumours show no 5p/3p skewing (Fig. 3c,d). In conjunction with our in vitro processing assays, which showed DROSHA RIIIB mutants to be functional nulls, this led us to question why heterozygous DROSHA mutations at precise metal-binding residues appear to be selected for in Wilms tumours, rather than other types of heterozygous null mutations such as frameshifts, as has been observed in DICER1 in cancer predisposition syndromes5,32. We confirmed that DROSHA mutations in Wilms tumours were heterozygous and that both alleles were expressed (Supplementary Fig. 3). We thus hypothesized that these mutations, rather than behaving as simple loss-of-function alleles, behave in a dominant-negative manner to impair the function of the residual wild-type allele.

To directly test whether DROSHA RNase IIIB mutations function through a haploinsufficient or a dominant-negative mechanism, we used genome editing with transcription activator-like effector nucleases (TALENs) to model DROSHA+/− and DROSHA+/E1147K states in cultured human cells. In the absence of available genomically stable Wilms tumour cell lines, we engineered the DROSHA mutations into HCT116 cells, a well-characterized diploid human cancer cell line. Crucially, we introduced mutations into the endogenous DROSHA locus, ensuring physiologic expression levels and allowing us to directly test the haploinsufficiency versus dominant-negative models. Three independent clones harbouring distinct heterozygous frameshift mutations in the second coding exon and three independent clones harbouring the E1147K mutation were generated (Fig. 4a and Supplementary Fig. 4). Immunoblottings of the engineered cells confirmed that heterozygous null cells (+/−) exhibit reduced DROSHA protein levels, while the E1147K mutation (+/EK) does not affect DROSHA expression (Fig. 4b and Supplementary Fig. 4c,d). miRNA profiling of parental cells and each of the six mutant clones using Taqman arrays was then used to determine the global effects of DROSHA mutations on miRNA processing. Although both DROSHA+/− and DROSHA+/EK cells produced less mature miRNA overall compared with parental HCT116 cells, the E1147K mutation resulted in a significantly greater magnitude of miRNA repression (average log2 fold change −0.86 for +/− versus −2.01 for +/EK; Fig. 4c). When individual miRNAs were evaluated, no individual miRNAs were expressed at higher levels in either mutant genotype, and only a few (9 out of 139) were expressed at statistically significantly lower levels in DROSHA+/− cells compared with parental HCT116 cells (Fig. 4d and Supplementary Fig. 5a). In contrast, DROSHA+/E1147K cells showed statistically significantly impaired expression of a large majority of detectable miRNAs (85 out of 139; Fig. 4d and Supplementary Fig. 5b). In keeping with our in vitro processing results (Fig. 2b) and analysis of miRNAs in Wilms tumours (Fig. 3c,d), the E1147K mutation did not result in 5p/3p skewing of miRNA expression (Supplementary Fig. 5b). The effect of the EK mutation was particularly striking when considering the expression of let-7 miRNAs. Although some let-7 family members were modestly reduced in DROSHA+/− cells, production of eight out of nine assayed let-7 family members was dramatically impaired in DROSHA+/EK cells (Fig. 4e). Taken together, these data provide strong evidence that DROSHA RIIIB hotspot mutations act in a dominant-negative manner to impair processing of a large set of miRNAs that includes the let-7 family members.

Figure 4: The heterozygous DROSHA E1147K mutation acts as a dominant-negative allele.
figure 4

(a) TALEN-induced frameshift mutations in DROSHA+/− cells (left) and knock-in mutations in DROSHA+/E1147K cells (right) are shown. Silent mutations, incorporated to prevent re-cleavage by TALENs, are denoted by grey underlined letters, while the non-synonymous mutation is in red. Intronic sequences in lower case. (b) Immunoblotting of DROSHA levels in parental and mutant cell lines. The full blot is shown in Supplementary Fig. 4. (c) Distribution of miRNA expression levels, displayed as log2 (normalized expression), calculated by subtracting average ΔCt of each miRNA (CtmiRNA−CtU6, averaged within each genotype) from the median parental ΔCt value. Distributions compared by paired two-tailed t-test; *P<10−10. (d) Scatter plot of log2 (normalized expression) of individual miRNAs in DROSHA+/− and DROSHA+/E1147K cells versus parental cells. Red dots represent miRNAs whose expression differed significantly (P<0.05) between mutant cells and wild type; P-values computed by two-tailed t-test. (e) Expression of let-7 family members in DROSHA-mutant cells, relative to parental cells. Individual TaqMan assays for each miRNA were performed in triplicate for each of three clones within each genotype. Error bars represent mean±s.d. across three clones within each genotype. Dashed line represents expression in parental cells; P-values calculated by comparing expression in mutant versus parental cells by z-test. *P<0.05; **P<0.005.

Full size image

Discussion

In this report, we demonstrate that widespread dysregulation of miRNA biogenesis is a key driving mechanism in Wilms tumour formation. We used exome sequencing of germline and tumour DNA to identify novel oncogenic mechanisms in Wilms tumour. Somatic and inherited mutations in key components of the miRNA biogenesis pathway, most prominently in the RNase IIIB (RIIIB) domains of DICER1 and DROSHA, are recurrent events in Wilms tumour. So-called DICER1 ‘hotspot’ mutations prevent processing of 5p-derived miRNAs, which reverses the 5p/3p bias typically seen in Wilms tumour miRNA expression. RIIIB-mutant DROSHA, on the other hand, is completely unable to process pri-miRNAs for both 5p- and 3p-derived miRNAs, and even the germline RIIIA mutation, R967W, demonstrated a partial loss of function. To precisely model the effect of missense mutations in tumours, we used genome engineering to show that a heterozygous DROSHA RIIIB mutation is more deleterious than simply a heterozygous null allele, confirming the hypothesis that these mutations impair global miRNA biogenesis through a dominant-negative mechanism. Most importantly, mutations in either DICER1 or DROSHA result in the loss of key tumour suppressor miRNAs, including the let-7 family as a whole.

Our data thus provide a mechanistic understanding of the selective pressures that give rise to the observed mutation spectra of DICER1 and DROSHA in Wilms tumours. Prior work has established that in mouse models of lung adenocarcinoma, soft-tissue sarcoma and retinoblastoma, heterozygous Dicer1 deletion accelerates tumorigenesis, whereas complete Dicer1 loss-of-function is strongly selected against38,39. Moreover, loss of the residual wild-type allele in human tumours harbouring germline or somatic DICER1 mutations is rarely observed12,32,40,41. These data indicate that in specific cancer settings, partial impairment of miRNA biogenesis is advantageous, yet residual activity of the miRNA pathway is needed for cell viability and/or other aspects of tumour biology. Importantly, this is probably not the case in all tumour types, as complete DICER1 loss-of-function was observed in a case of pineoblastoma42. Nevertheless, our finding of partial, but never complete, loss of miRNA processing through DROSHA and DICER1 mutations in Wilms tumours suggest that residual miRNA activity is required in this cancer. Furthermore, it is reasonable to speculate that somatic heterozygous DICER1 RNase IIIB domain mutations, especially when they occur in a patient with a germline DICER1-null allele, provide the optimal balance of impaired production of tumour suppressor miRNAs (many of which derive from the 5′-arm of miRNA hairpins such as members of the let-7 family), while allowing sufficient biogenesis of miRNAs that are required for tumour viability. In the case of DROSHA, heterozygous null mutations have very little impact on miRNA production, as revealed by our studies in genetically engineered cell lines, and are thus not observed in tumours. Heterozygous DROSHA RNase IIIB mutations, in contrast, have a dominant-negative effect on miRNA production, impairing production of the let-7 family and numerous other miRNAs, yet apparently allowing sufficient residual miRNA processing activity to support tumour growth. Finally, it is important to note that functions beyond miRNA processing have been ascribed to both DROSHA and DICER1, such as participation in the DNA damage response43, warranting further investigation of these non-canonical functions in Wilms tumour pathogenesis.

As shown in Fig. 3b, some Wilms tumours without identifiable mutations in the miRNA processing machinery also show moderately reduced levels of let-7 family members, suggesting that other mechanisms of reducing let-7 levels may be active in these tumours. Upregulation of LIN28A or LIN28B is known to suppress biogenesis of the let-7 family in various cancers44,45,46, including Wilms tumour44, and Myc induces LIN28B expression to repress let-7 (ref. 47). Urbach et al.48 recently showed that conditional expression of LIN28 in developing mouse kidney mesodermal precursor cells is sufficient to cause Wilms tumours, an effect which could be suppressed by expression of a let-7 family member. In our tumour cohort, three of nine tumours examined expressed relatively high levels of LIN28A and/or LIN28B (Supplementary Fig. 6), consistent with the reported prevalence of LIN28 overexpression in Wilms tumours44. However, there was no direct correlation between LIN28 expression and downregulation of mature let-7 family members, suggesting that additional mechanisms, beyond DROSHA and DICER1 mutations and LIN28 overexpression, may lead to let-7 repression in some tumours. Taken together, these data indicate that loss of let-7 expression, whether by mutation of miRNA processing genes or by other mechanisms, may represent a central mechanism of oncogenesis in Wilms tumours.

Methods

Exome sequencing and analysis

The tumour specimens used in the study were residual tissue from the patients’ surgical resection specimens and stored either with informed consent in the Pediatric Biospecimen Repository or in the Department of Pathology archives at Children’s Medical Center. This study was performed under a protocol approved by the Institutional Review Board of the University of Texas Southwestern Medical Center. Indexed DNA libraries were constructed per the TruSeq Exome Enrichment Kit, as per the provided protocol. Target enrichment was then performed by pooling four indexed samples together and using the human TruSeq Exome Enrichment kit for each pool, also as per protocol. This kit targets 62 Mb of exomic sequences in the human genome, including 5′-untranslated region, 3′-untranslated region, miRNA and other non-coding RNA regions. The resulting libraries were sequenced using the HiSeq 2000, producing 100 bp paired-end reads.

Reads were demultiplexed using casava version 1.8.2, allowing up to one mismatch in the index sequence and clusters not passing filter were excluded from the resulting fastq files. The reads were then quality-checked using fastQC (0.8.0). Each sample’s reads were aligned to the hg19 build of the human genome reference sequence using BWA (v 0.6.2), and duplicates were removed using Picard’s MarkDuplicates (1.41). Sample level realignment around indels and base quality recalibration were performed using GATK 1.4.6 (Broad Institute, Boston, MA) (http://www.broadinstitute.org/gatk/guide/topic?name=best-practices). Coverage was also calculated using GATK, and an average coverage of 52 × was achieved across targeted regions, with 91% of bases covered at >10 × and 85% of bases covered at >20 × . Single-nucleotide polymorphisms (SNPs) and Indels in the targeted regions were called using GATK’s UnifiedGenotyper, followed by hard filtering (indels) and VariantRecalibration (SNPs). During recalibration, data from the OMNI and HapMap SNP chips were used as training sets. Those variants passing filters were annotated using successive ANNOVAR runs (2012-05-25), wherein dbSNP IDs, 1000Genomes frequencies, polyPhen and SIFT scores were added. ANNOVAR also provided genomic location, and when variants were found to be exonic, also annotated the gene name, transcript ID, functional change, nucleotide and amino acid change of each variant.

We generated a median of 88 million raw reads per tumour sample. In the discovery set, deviations from the reference genome were identified in both tumour and matched germline samples. Candidate somatic mutations were identified if they were not present on any reads in the germline samples; seen on at least three reads in tumour samples; non-synonymous; not present in the dbSNP or 1000 Genomes databases; and covered at a minimum total read depth of 6 × . Mutations in DROSHA and DICER1 were validated by Sanger sequencing with primers designed using Primer-BLAST. Similar filters were used to uncover single nucleotide variants in the validation set.

Mutagenesis

The V5-DROSHA expression plasmid was constructed by subcloning DROSHA from pcDNA4/TO/cmycDrosha49 into pcDNA3.1/V5-His-TOPO (Invitrogen). The QuikChange Lightning Site-Directed Mutagenesis kit (Stratagene) was used to introduce mutations into plasmids encoding V5-DROSHA, FLAG-DGCR8 (ref. 49), V5-DICER1 (ref. 50) and FLAG-TARBP2 (ref. 50). Primers used for mutagenesis are listed in Supplementary Data 4.

Cell Culture and immunoprecipitation

For DROSHA and DICER1 processing assays, HEK293T cells were seeded at a density of 5 × 106 cells per 100 mm plate 12 h before transfection. Cells were transfected with V5-DICER1 or a mixture of V5-DROSHA and FLAG-DGCR8 expression plasmids using X-tremeGENE 9 (Roche Applied Science) following the manufacturer’s protocol. Forty-eight hours after transfection, cells were washed with cold PBS and collected into 600 μl of IP buffer containing 20 mM Tris·HCl (pH 8.0), 150 mM NaCl, 1 mM EDTA, 0.5% Nonidet-40 and complete EDTA-free protease inhibitor (Roche Applied Science). Cells were lysed by sonication in a Bioruptor (Diagenode) and then centrifuged at 12,000 g for 10 min at 4 °C. The supernatant was collected and 15 μl was saved as total input. The remaining lysate was incubated with 3 μg of anti-V5 mouse monoclonal antibody (Invitrogen) for 30 min on ice. Thirty microlitres (900 μg) of protein G Dynabeads (Invitrogen) were added and lysates were rotated for an additional 2 h at 4 °C. The beads were washed five times in IP buffer and then twice in DROSHA/DICER1 assay buffer (20 mM Tris·HCl (pH 8.0), 100 mM KCl, 0.2 mM EDTA). Ninety per cent of the beads were resuspended in 50–70 μl DROSHA/DICER1 assay buffer for enzymatic assays and the remaining beads were resuspended in Laemmli buffer for western blot analysis.

A similar protocol was followed for analysis of DGCR8 and TARBP2 mutations with the following modifications: V5-DROSHA was co-transfected with Flag-DGCR8 wild type or S702N. V5-DICER1 was co-transfected with Flag-TARBP2 wild type or D221G. Complexes were immunoprecipitated with 6 μg of anti-Flag M2 mouse monoclonal antibody (Sigma) and then resuspended in DROSHA/DICER1 assay buffer for enzymatic assays or Laemmli buffer for western blots.

DROSHA and DICER1 processing assays

Pri-miRNA substrates consisting of miRNA hairpins and ~80–150 bp of genomic flanking sequence were amplified from human genomic DNA and either cloned into the NheI site of pcDNA3.1 (+) (miR-1-1 and miR-124-1) or used as in vitro transcription templates directly (let-7b and miR-16-1, with T7 promoter in the 5′ primer). Primer sequences are provided in Supplementary Data 4. Non-radiolabelled pri-miRNAs were in vitro transcribed with XbaI-linearized plasmids (miR-1-1 and miR-124-1) or PCR fragments (let-7b and miR-16-1) using the MAXIscript kit (Ambion) and subsequentely gel purified. DROSHA processing assays were carried out by incubating 10 μl of immunoprecipitated beads in DROSHA/DICER1 assay buffer prepared as described above, 2 μl of 64 mM MgCl2, 0.2 μl of pri-miRNA (~0.3 pmol), 0.6 μl of RNaseOUT (Invitrogen) and 7.2 μl of diethylpyrocarbonate-treated water at 37 °C for 30 min and then stopped by chilling samples on ice. After adding 20 μl of Gel Loading Buffer II (Ambion), reaction mixtures were boiled at 95 °C for 3 min and then chilled on ice. Magnetic beads were removed and the supernatant was loaded onto a 15% denaturing polyacrylamide gel. Pre-miRNAs and processing intermediates were transferred onto nitrocellulose membranes in 0.5 × TBE buffer at 250 mA for 2 h. After ultraviolet cross-linking for 50 mJ, membranes were incubated overnight at 42 °C with Ultrahyb Oligo Buffer (Ambion) containing 32P end-labelled oligonucleotide probes perfectly complementary to the mature miRNA sequences. Membranes were then washed three times with Wash Buffer (2 × SSC, 0.5% SDS) at 42 °C and exposed to phosphor screens (GE Healthcare).

DICER1 processing assays were performed similarly using synthesized RNA oligos identical in sequence to human pre-miRNAs as substrates (sequences provided in Supplementary Data 4). Enzymatic mixtures consisted of 10 μl of immunoprecipitated beads, 2 μl of 20 mM MgCl2, 0.2 μl of 0.4 μM pre-miRNA, 0.1 μl of 100 mM dithiothreitol, 0.5 μl of RNaseOUT (Invitrogen) and 7.2 μl of DEPC water.

Small RNA sequencing

Small RNAwas prepared from 50 ng of each tumour sample using the mirVana PARIS kit according to standard manufacturer’s protocol (Life Technologies). The small RNA libraries were prepared using the protocol for small RNA library preparation in the SOLiD Total RNA-Seq kit from Life Technologies. The resulting libraries were sequenced on the SOLiD 5500xl. The reads produced were 35 bp single end. The machine demultiplexed the individual samples’ reads on-machine. Demultiplexed reads from the machine were analysed using LifeScope (v. 2.5.1) using the default small RNA parameters for analysis. The reads were first mapped to a ‘filter sequence’ fasta file containing SOLiD adapter sequences, ribosomal RNA, transfer RNA and other ‘junk’ RNA sequences. The reads mapping to this file were excluded from further analysis. The remaining reads were then mapped to the hg19 build of the human genome and also to miRBase sequences. The algorithm used a seed-and-extend method for mapping reads, to better identify the ends of the miRNAs. The mapped reads were annotated according to both known precursor and mature miRNA sequences from miRBase. Using the information from the mapped bam files, coverage counts of mature and precursor sequences, as well as coverage wiggle files were produced. Read counts were categorized as being derived from the 5p or 3p arm according to their entries in http://www.mirbase.org. Next, the miRNAs that were detectable in all 12 tumours were analysed for differential expression using the DESeq v1.10.1 package of R/Bioconductor.

Quantitative PCR

Small RNAs were isolated from 50 ng of each tumour sample as above. All samples were analysed using the RNA 6000 Nano LabChip (Agilent Technologies) on a 2100 Bioanalyzer (Agilent Technologies), and we defined a minimum RNA Integrity Number >5 to be suitable for further evaluation by TaqMan reverse transcription–PCR analysis51. Taqman miRNA assays (Invitrogen) for following target miRNAs were used to measure their expression in tumour samples: let-7b (Assay ID: 002619), let-7d (Assay ID: 002283), let-7g (Assay ID: 002282), let-7i (Assay ID: 002221), miR-100 (Assay ID: 000437), miR-204 (Assay ID: 000508) and miR-424 (Assay ID: 000604). The reference gene snU6 (Assay ID: 001973) was used for normalization. The RT2 qPCR Primer Assay (Qiagen) was used to measure LIN28A (Assay ID: PPH10338B) and LIN28B (Assay ID: PPH57843B).

Genetic engineering of cell lines

Pairs of TALENs were designed to target specific DROSHA sequences using ZiFit Targeter (zifit.partners.org)52. A pair of TALENs were designed to target a SmaI site in the second coding exon to create a null mutant through the introduction of indel mutations, while a second pair of TALENs were designed against the region surrounding codon 1147 to create the knock-in mutation through homologous recombination. TALENs were constructed using the Joung lab REAL assembly TALEN kit (www.addgene.org/talengineering)53. Complete amino acid sequences of TALENs are provided in Supplementary Methods. A homology template for the E1147K mutation was created by amplifying arms from genomic DNA (primers provided in Supplementary Table 4) and mutating 1 bp to model the E1147K mutation; three more silent mutations were also introduced into the TALEN-binding site to prevent re-cleavage (mutations shown in Fig. 4a). Homology arms were ligated around a lox-stop-puro-lox cassette (from Addgene plasmid 11584 (ref. 54)), from which the stop elements had been excised.

The TALEN pair targeting the second coding exon was co-transfected into HCT116 cells with a green fluorescent protein-expressing plasmid using Fugene HD (Promega), and 48 h later the brightest 10% of cells were collected by FACS sorting. Colonies were expanded from single cells and genotyped for heterozygous destruction of the SmaI site and positive clones were confirmed by Sanger sequencing. Three heterozygous clones with independent frameshift mutations were expanded for further analysis.

The second TALEN pair was co-transfected with the homology template. Seventy-two hours after transfection, the cells were transferred to 1 μg ml−1 puromycin-containing media for 4 days and then seeded into 96-well plates for single-cell cloning. Puromycin-resistant colonies were genotyped by PCR, using a forward primer inside the puro cassette and a reverse primer outside the homology arm. The presence of the correctly targeted heterozygous mutations was confirmed by sequencing. Three correctly targeted clones were then transiently transfected with a Cre-expressing plasmid to remove the puro cassette and re-expanded from single cells to produce three independently derived clonal lines with the targeted allele.

Immunoblotting

Western blotting was performed using monoclonal antibodies against DROSHA (D28B1, Cell Signaling Technology, 1:1,000) and tubulin (clone DM1A, Sigma, 1:10,000).

Analysis of miRNA expression in engineered cell lines

miRNA expression was profiled in three independent heterozygous Drosha-null and E1147K cell lines and three biological replicates of parental HCT116 cells. Cells (4 × 106) were plated in 10-cm dishes and total RNA was extracted the following day using the mirVana miRNA isolation kit (Life Technologies). Complementary DNA was reverse-transcribed using the TaqMan MicroRNA Reverse Transcription Kit and Megaplex RT primers (Human Pool A v2.1), mixed with TaqMan Universal PCR Master Mix and loaded onto TaqMan Array Human MicroRNA A Cards 2.0 (Life Technologies). Following amplification on an Applied Biosystems 7900HT Real-Time PCR System, data were analysed using SDS software v2.4 and RQ manager v1.2.1 (Life Technologies). Raw Ct values were exported for statistical analysis.

miRNAs whose mean Ct values were >32 in any of the three groups were discarded. Global miRNA expression was assayed by TaqMan array, and for each miRNA, ΔCt was calculated by normalizing to U6 (CtmiRNA−CtU6). Average ΔCt value for each miRNA was then calculated within each genotype and log2 of normalized expression was calculated by subtracting each average ΔCt from the median parental ΔCt value. Distributions were compared by paired t-test. Levels of let-7 family members were validated using individual TaqMan assays from Life Technologies.

Additional information

How to cite this article: Rakheja, D. et al. Somatic mutations in DROSHA and DICER1 impair microRNA biogenesis through distinct mechanisms in Wilms tumours. Nat. Commun. 5:4802 doi: 10.1038/ncomms5802 (2014).

Accession Codes: Small RNA sequencing data have been deposited in the Gene Expression Omnibus (GEO) Database under the accession code GSE56955. Whole-exome sequence data of tumours for which patients (or their parent/legal guardian) gave consent for the deposit of their information have been deposited in dbGaP under the accession code phs000780.v1.p1 (http://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000780.v1.p1).