Germline Pathogenic DROSHA Variants Are Linked to Pineoblastoma and Wilms Tumor Predisposition.

The proband in family 1 was a female diagnosed with pineoblastoma at age 6 that recurred as a teenager (individual II-1; Fig. 1 ). Her male first cousin (individual II-2) was diagnosed with pineoblastoma at age 5 that also recurred when he was a teenager. The mothers of both children (individuals I-1 and I-2, respectively) are healthy biological full siblings with no personal history of cancer. Importantly, tissue limitations precluded the possibility of determining whether the second tumors were recurrences or new primary tumors. We performed WES of germline DNA from the four individuals and identified 160 shared rare germline variants. These include 143 nonsynonymous, three nonframeshift InDels, four frameshift InDels, four nonsense, two splice site modifiers, and four unannotated variants. Of these 160 variants, 103 variants in 101 genes were predicted to be deleterious using SIFT, PolyPhen2, MutationTaster, Radial SVM, or LRT (Supplementary Table S3; refs. 37 , 39 , 40 ). Family 1 pedigree. Individuals with pineoblastoma are shaded in dark blue. Other cancer-affected family members are shaded in light blue. WES was performed on germline DNA isolated from both probands and their mothers who are full biological sisters but cancer-unaffected, as well as the tumor DNA from individual II-1. Among these 103 variants, we identified a heterozygous germline nonsense variant in DROSHA shared by all four individuals at amino acid 271 of DROSHA ( NM_013235.4 : c.811 C>T, p.Arg271Ter; Table 1 ; Figs. 2 and 3 ). WES analysis of tumor DNA from individual II-1 revealed a somatic nonsense mutation in DROSHA at amino acid 265 ( NM_013235.4 : c.795C>G, p.Tyr265Ter). Given the proximity of the variants (which could be captured on the same sequencing read) and the observation that reads from the tumor contained either the germline or somatic pathogenic variant but never both, the phase of the two variants could be determined to be in trans , indicating biallelic inactivation of DROSHA in the tumor. Germline and somatic DROSHA variants identified in families 1 to 8. Each row represents a single individual. Physical positions are listed using GRCh37/hg19. cDNA changes correspond to transcript ID NM_013235 . NA (Not available) suggests that an amino acid change was not present due to the nature of the variation. Map of the DROSHA protein showing functional domains and sites of germline and somatic pathogenic variants identified in families 1 to 8. dsRBD, double-stranded RNA-binding domain; P-rich, proline-rich; RIIIDb, RNase III domain B; RS-rich, arginine/serine-rich. Representative WES alignment sequence reads from individual II-1 (family 1). p.R271Ter is found on one allele of DROSHA in the patient’s germline and tumor samples. Y265Ter is found in trans in the tumor, indicating biallelic loss of DROSHA in the tumor. Nucleotide cDNA sequence for DROSHA is shown in the 3′ to 5′ direction from left to right. The amino acid sequence is in C- to N-terminus direction from left to right. The proband in family 2 was an 8-year-old male diagnosed with pineoblastoma. Family history revealed a paternal first cousin diagnosed with medulloblastoma as a child, but no further contributory family history was available. Clinical paired tumor/normal WES was performed by a commercial laboratory. In his germline DNA, a heterozygous DROSHA NM_013235.4 : c.2883-1G>A variant predicted to destroy the canonical splice acceptor site in exon 23 of DROSHA was identified (SpliceAI delta score = 0.95) and classified as likely pathogenic. His tumor was found to be tetraploid with the loss of one copy of chr5, in which DROSHA resides, and with a tumor VAF of 0.63 for the DROSHA c.2883-1G>A GPV. Also identified in the tumor was a tier 1 somatic DROSHA NM_013235.4 : c.2683-1G>T variant (VAF = 0.37) that destroyed the splice acceptor site for exon 21 of DROSHA (SpliceAI delta score = 0.98). Although phasing could not be confirmed in the tumor sample, these results suggest that one copy of DROSHA was lost, two of the three DROSHA alleles present in the tumor carried the GPV, and one carried the somatic tier 1 variant, which is consistent with complete inactivation of DROSHA in the tumor ( Fig. 2 ; Supplementary Fig. S1). The individual’s tumor was further classified as miRNA processing–altered 1 subtype of pineoblastoma (PB-miRNA1) based on alterations in the tumor to chromosome 12 (gain 12p13.33q24.33) and 16 (loss 16p13.3q24.3; ref. 49 ). Further details of chromosomal aberrations of this individual’s tumor can be found in Supplementary Table S4. The proband in family 3 was a 7-year-old girl diagnosed with pineoblastoma with embryonal morphology. The tumor was completely excised and confirmed as pineoblastoma by IHC. Family history was noncontributory. Clinical NGS panel sequencing of germline and tumor DNA was performed using a custom cancer gene panel. Germline analysis revealed a heterozygous frameshift InDel in DROSHA resulting in a premature termination codon (c.403_409delinsCCACTT, p.Ala135Leufs*9). Tumor analysis revealed a VAF in the tumor of 0.81, indicative of LOH of this variant. Methylation analysis classified the pineoblastoma in the PB-miRNA1 subtype ( 60 , 61 ). The proband in family 4 was an 11-year-old girl diagnosed with pineoblastoma confirmed by biopsy. No family history was available. Clinical WES was performed on both germline and tumor DNA. Germline analysis revealed a DROSHA c.3261+1G>C splicing variant resulting in out-of-frame exon skipping with loss of the donor site at exon 27 (SpliceAI delta score = 0.95). Tumor analysis revealed LOH of this variant, with a variant allele fraction in the tumor of 1.00, confirming biallelic inactivation of DROSHA . RNA analysis of the tumor specimen confirmed exon skipping. Methylation analysis classified the pineoblastoma in the PB-miRNA1 subtype. The proband in family 5 was a 15-year-old female with metastatic pineoblastoma. Family history was notable for a maternal great-aunt with uterine or ovarian cancer and a maternal great-uncle with pancreatic cancer, both diagnosed in early adulthood, but for whom samples were not available. Two LOF variants in DROSHA (c.1869del, p.Phe623Leufs*3; VAF = 0.42) and (c.3548del, p.Asn1183Ilefs*8; VAF = 0.48) were identified in the tumor by clinical NGS panel testing, suggesting biallelic LOF, although the phase could not be confirmed. Subsequent clinical germline-targeted DROSHA sequencing revealed that the heterozygous c.3548del DROSHA frameshift variant was germline. Tumor methylation profiling was performed, and the tumor was classified as PB-miRNA1. The proband in family 6 presented at 3 years of age with a large left-sided renal mass and two discrete right-sided renal masses with hydronephrosis and pulmonary metastatic disease. She underwent left radical nephroureterectomy and right nephron-sparing surgery. Pathology revealed diffuse anaplasia in the left mass and favorable histology Wilms tumor in the right-sided lesions. Family history showed no classical DICER1- or DROSHA -associated tumors but was significant for a maternal ATM GPV ( NM_000051.3 : c.2502dup, p.Val835Serfs*7) and early-onset breast cancer in the maternal grandmother. Clinical germline Wilms tumor predisposition panel testing showed a heterozygous 1.29 Mb interstitial 5p13.3 deletion [arr(hg19) 5p13.3(30384246-31670457) × 1] including the entirety of CDH6 and DROSHA and exon 1 of PDZD2 , as well as the presence of the maternal ATM GPV. Beckwith–Wiedemann syndrome molecular testing showed no evidence of abnormal methylation, and clinical 115-gene cancer predisposition panel testing did not identify an additional GPV. Clinical paired tumor-germline sequencing revealed an acquired somatic variant in DROSHA ( NM_013235.4 : c.147_148delinsAT; p.Gln50Ter) at VAF = 0.80 in the left Wilms tumor sample indicating biallelic inactivation of DROSHA in the tumor, as well as other chromosome gains/losses. Parental testing confirmed the germline 5p13.3 deletion was inherited from her asymptomatic father. The proband in family 7 was identified from a group of 19 pediatric ( n = 16) and adult ( n = 3) patients from a single institution with tumors observed in the spectrum of DICER1 -related tumor predisposition (Supplementary Table S5) and for whom germline and tumor DNA were available. Four had pineoblastomas, 13 had unilateral Wilms tumor, and two had thyroid tumors. We performed Sanger sequencing of DICER1 and DROSHA in germline and tumor DNA from all individuals and identified five exonic variants in DROSHA ( n = 2) and DICER1 ( n = 3) that were present in at least one sample (Supplementary Table S6). Of these variants, only one was germline, whereas the remaining four were somatic only. No individuals were identified with a DICER1 GPV. We identified one patient, a 21-year-old male with pineoblastoma (family 7), with a germline heterozygous 4 bp deletion in DROSHA ( NM_013235.4 : c.2436_2439del) that caused a frameshift and premature termination at p.Asp814Asnfs*19 ( Table 1 ). Tumor sequencing from this individual revealed copy-neutral LOH in the wild-type allele resulting in biallelic DROSHA LOF ( Figs. 2 and 4 ). Of note, the mother of this patient had been diagnosed with Ewing sarcoma as a child and carried the same germline 4 bp DROSHA deletion. DROSHA Sanger sequencing chromatograms of individual 9, family 7, indicating a germline 4 bp deletion with LOH in the tumor. The plot shows the sequence results from germline ( A ) and pineoblastoma tumor ( B ) DNA. The x -axis represents the position of the sequencing read in DROSHA , and the y -axis represents the signal intensity of each potential nucleotide at that position. The color of the line represents each of the four nucleotides (adenosine: green, thymine: red, cytosine: blue, and guanosine: black). The aligned nucleotide sequence is provided above each peak in which degenerate positions are represented with International Union of Pure and Applied Chemistry codes. The position of the 4 bp deletion is represented by a light blue–highlighted region. To investigate the occurrence of DROSHA GPVs in a large cohort of patients with pediatric brain tumors, we queried the germlines of all patients with central nervous system tumors from the OpenPedCan (v15, n = 2,148; bioRxiv 2024.07.09.599086). In OpenPedCan, there are 21 patients with pineoblastoma, of whom seven were the PB-miRNA1 subtype. We identified one individual, the proband in family 8, an 8-year-old male with pineoblastoma with a stop-gain DROSHA GPV (c.2988C>A; p.Glu1033Ter; Fig. 2 ) and a somatic whole-gene deletion of the wild-type allele in the tumor (Supplementary Fig. S2). Methylation analysis classified the tumor as the PB-miRNA1 subtype (Supplementary Table S5). The child did not harbor a GPV in DICER1 or DGCR8 . Next, we investigated the occurrence of DROSHA GPVs in other pediatric tumor registries. We queried the CCSS cohort ( n = 5,451) and St. Jude PeCan Portal ( n = 5,728), and although we cannot exclude an overlap of patients between the CCSS cohort and the St. Jude PeCan Portal, we did not identify any other pediatric patients with germline LOF DROSHA variants. We then investigated whether DROSHA GPVs are associated with the risk for adult-onset cancers. We analyzed the TCGA cohort of 10,389 patients with germline WES data across 33 cancer types. Only five individuals with predicted LOF (pLOF) DROSHA variants were detected (0.05%; Supplementary Table S7): A patient of African ancestry with uterine corpus endometrial carcinoma carried a DROSHA stop-gain variant (5:g.31521318G>A; ENSP00000425979 p.Arg287*). A patient of European ancestry with uterine corpus endometrial carcinoma carried a DROSHA splice acceptor variant (5:g.31431787T>C; c.3043-2A>G). A patient of European ancestry with thyroid carcinoma also carried a DROSHA splice acceptor variant (5:g.31527020C>T; c.21-1G>A). A patient of Asian ancestry with stomach adenocarcinoma carried a DROSHA stop-gain variant (5:g.31526994G>A, p.Arg16*). Lastly, a patient of unclassified ancestry with glioblastoma multiforme harbored a frameshift variant (5:g.31495408C>CGG​GGA, p.Thr582fs). Importantly and in contrast to all pediatric patients, no LOH or acquired somatic mutation of any kind in DROSHA was observed in the tumor DNA from these patients. To explore the occurrence of DROSHA LOF variants in the general population, we analyzed the UKBB ( n = 469,787) and Geisinger DiscovEHR cohorts ( n = 170,503). Figure 5 shows the distribution of DROSHA pLOF variants in UKBB, Geisinger, TCGA, and the identified eight families. In UKBB, there were 49 variants in 97 individuals and, in Geisinger, 24 variants in 44 individuals, giving the prevalence of pLOF DROSHA variants to be 1:4,843 (95% confidence interval, 1:3,970–1:5,970) and 1:3,875 (95% confidence interval, 1:2,886–1:5,201), respectively (Supplementary Table S8). No DROSHA pLOF homozygotes were observed. Comparing the demographics of DROSHA pLOF heterozygotes in UKBB and Geisinger with their controls, we did not see significant differences in sex, age, smoking status, body mass index, race, and death (Supplementary Table S9). DROSHA lollipop plot. Depiction of TCGA, population-level, and family-level DROSHA LOF variants superimposed on the domain structure of DROSHA. Variants identified from population-level queries of UKBB and Geisinger DiscovEHR are shown in blue. TCGA-identified variants are shown in green. Variants identified in the eight families are shown in pink. DRBM, Double-stranded RNA binding motif. Next, we explored cancer risk in these individuals. In Geisinger, there was a significantly reduced cancer rate in DROSHA heterozygotes compared with controls and no significant differences in UKBB (Supplementary Table S9). Using cohort-specific prevalence estimates, the power to detect common and/or rare cancers was estimated (Supplementary Fig. S3). In UKBB, there was >80% power to detect OR > 2 for all (22% rate) and common cancers (12%) and 80% power to detect OR > 2.4 for sex-specific common cancers (5% rate). In Geisinger, there was 60% power to detect OR = 2 for common cancers (22%). Both cohorts had limited power to detect rare cancers (<1%). Supplementary Figure S4 is an oncoprint that displays cancers reported in DROSHA LOF heterozygotes in both UKBB and Geisinger. The code C44 (“other malignant neoplasms of skin”) was the most common, followed by breast and prostate. As cancer predisposition syndromes typically feature a younger age at cancer diagnosis, time to cancer in UKBB was examined; there were an insufficient number of events in Geisinger. There was no significant difference in age at diagnosis compared with controls (Supplementary Fig. S5). As DICER1 -related tumor predisposition is associated with goiter and other thyroid diseases, noncancer phenotypes in DROSHA LOF heterozygotes were queried. In both cohorts, there was a nominal excess of six ICD codes (UKBB: N80 endometriosis, R60 edema NOS, and A09 diarrhea and gastroenteritis of presumed infectious origin; Geisinger: S40 superficial injury of shoulder and upper arm, I21 STEMI and NSTEMI myocardial infarction, and K12 stomatitis and related lesions). However, none overlapped across the cohorts, and none were significant after Bonferroni correction (Supplementary Table S10). Lastly, we examined the population frequency of the DROSHA variants found in the proband cases. The DROSHA p.Arg271Ter GPV identified in family 1 was found in one individual in gnomAD (v4.1.0 non-UKBB, n = 314,392) and four individuals in UKBB. The DROSHA c.2883-1G>A GPV identified in family 2 was not found in any individual in either database, but a c.2883-1G>C variant was found in one individual in gnomAD (non-UKBB) and three individuals in UKBB. The GPVs identified in the remaining six families were not present in any individual in either gnomAD or UKBB.

In this study, we describe nine individuals from eight families with a DROSHA GPV. Eight had pineoblastoma, and one had bilateral Wilms tumor. One individual from each family had tumor sequencing performed; in every case, a second somatic DROSHA LOF variant in the tumor was observed, and in no patient was a GPV in DICER1 , DGCR8 , or any other gene associated with predisposition to pineoblastoma or Wilms tumor identified ( 3 , 7 , 8 , 19 ). For one patient (family 1), both germline and somatic pathogenic variants (PV) introduced premature stop codons. For two patients (families 2 and 4), the germline and somatic PV disrupted canonical splice sites. For three other patients (families 3, 5, and 7), the germline and somatic PVs were small deletions resulting in frameshifts and the introduction of premature stop codons. For two patients (families 6 and 8), a whole-gene deletion and a premature stop codon were observed. For three patients, the mechanism of somatic inactivation was LOH, whereas for the other five, it was the acquisition of an LOF variant distinct from the GPV. Importantly, for individuals from families 1, 3, 4, 6, 7, and 8, biallelic inactivation of DROSHA in the tumor could be confirmed based on the occurrence of the somatic variant in trans to the germline variant. In 2018, Snuderl and colleagues ( 20 ) identified homozygous deletions of DROSHA in five of 19 postmortem pineoblastoma tumor samples, supporting the contention that loss of DROSHA plays a role in pineoblastoma development. However, no germline DNA from any of these five patients was available for analysis, so it could not be determined whether these were germline or acquired deletions. Of the four pineoblastoma samples in their study with matched germline and tumor DNA, none harbored a DROSHA GPV. More recently, an analysis of methylation data from 221 patients with pineoblastoma found that more than half of the tumors ( n = 119) were classified as PB-miRNA1 or miRNA processing–altered 2 (PB-miRNA2), molecular subgroups of pineoblastoma associated with miRNA processing defects ( 49 ). Of the 117 PB-miRNA1 or PB-miRNA2 pineoblastomas with tumor genomic information, 28 had LOF DROSHA alterations in the tumor. With respect to Wilms tumor, in a series of 44 Wilms tumor samples, two were identified with heterozygous germline DROSHA missense variants (p.Met120Val and p.Arg967Trp; ref. 8 ). Additionally, in a separate analysis of 58 Wilms tumor samples, two unilateral patients with Wilms tumor carried germline DROSHA missense variants at p.Arg297Cys ( 17 ). In no case in either study, however, was a second somatic LOF variant observed in the corresponding tumor, and the significance of these variants in the context of Wilms tumor etiology remains unknown. Taken together, our results indicate that germline pathogenic variation in DROSHA underlies a novel autosomal dominant cancer predisposition associated with pineoblastoma and Wilms tumor. This new association phenocopies some clinical phenotypes associated with DICER1 -related tumor predisposition but is thus far distinct from those associated with DGCR8 germline pathogenic variation. The observation that this condition mimics some aspects of DICER1 -related tumor predisposition is supported by the observation that among patients with pineoblastoma, the chromosomal or methylation data from all are consistent with pineoblastoma subtype 1 (PB-miRNA1), which is also associated with both DICER1 -related pineoblastoma and pineoblastoma with biallelic somatic DROSHA , DICER1 , or DGCR8 LOF mutations. Additionally, DICER1 -related pineoblastoma has only been characterized in the context of a GPV with biallelic inactivation of DICER1 because of somatic LOH or LOF variants. This is in contrast with other DICER1 -related tumors, in which somatic second hits are usually missense variants in RNase III domain metal-ion binding site hotspots ( 62 , 63 ). Similarly, all cases of DROSHA- related pineoblastoma in our study were characterized by somatic second hit LOH or LOF DROSHA variants that arise along the length of the gene, consistent with biallelic DROSHA inactivation in the tumor. Given the observation that the mother in family 7 also carries the familial 4 bp germline deletion, our results may also suggest a previously unsuspected link between Ewing sarcoma and miRNA processing. Our analysis of DROSHA GPVs in large-scale germline databases highlights the rarity of DROSHA GPVs in the human population. In TCGA, five adult patients were identified with a DROSHA GPV. No patient, however, had a second somatic DROSHA variant identified in their tumor DNA, suggesting that a DROSHA GPV is not a common driver of adult-onset cancers although it is possible that the wild-type DROSHA allele may have been inactivated by genetic or epigenetic mechanisms not captured in the TCGA dataset. Separately, using a genome-first approach in large population databases, we found DROSHA GPV prevalence to be more common than DICER1 (∼1:4,000 for DROSHA and ∼1:8,000 for DICER1 ). Similar to the observations in TCGA and despite sufficient power to detect common cancers, we did not observe any enrichment of cancers in UKBB or Geisinger. In a previous study quantifying cancer risk in DICER1 heterozygotes using the same datasets ( 27 ), Kim and colleagues found no significant excess cancer risk among these individuals as well. This suggests that both DROSHA and DICER1 may have an age-limited window of increased tumor susceptibility and that penetrance in adults is low although we note that children with DROSHA or DICER1 GPV who die at a younger age from cancer or other conditions are not included in adult datasets, which may influence our ability to detect enrichments. Interestingly, although we observed a reduced cancer rate among DROSHA heterozygotes in Geisinger but not in UKBB, Kim and colleagues did not observe a similar reduction among DICER1 heterozygotes in either cohort. These differences likely reflect differences in the ascertainment and genetic background between the cohorts as well as the small number of cases in each cohort. One limitation of our study is that tumor DNA was not available for individual II-2 in family 1, limiting our ability to confirm the somatic acquisition of an LOF variant in all tumors. Second, methylation profiling of pineoblastoma is commonly used to infer the underlying etiologic defect from the molecular subtype of the tumor. Methylation information was only present for tumors from individuals from families 3, 4, 5, and 8. Although all were consistent with PB-miRNA1, which is associated with the inactivation of miRNA processing, this could not be investigated in the other three pineoblastoma tumors. Importantly, however, the pattern of chromosomal aberrations in family 2 was also consistent with the aberrations of miRNA processing defects. Finally, we note that due to the limited availability of tissue, we were not able to investigate the consequence of biallelic DROSHA inactivation on miRNA biogenesis and the regulation of transcription in the tumors we studied. In light of these findings, it will be of considerable interest to determine whether DROSHA GPVs are observed in other individuals with classical DICER1 phenotypes without identified DICER1 GPVs. Additionally, although there is little documented overlap with DGCR8 phenotypes to date, our understanding of both DGCR8 - and DROSHA -related tumor predisposition is still evolving. Importantly, the miRNA-altered subtypes of pineoblastoma (PB-miRNA1 and PB-miRNA2) are associated with significantly better prognosis compared with subtypes PB-RB1 and PB-MYC/FOXR2. PB-miRNA1 and PB-miRNA2 have 5-year progression-free survival rates of 86.1% and 56.7%, respectively, whereas PB-RB1 and PB-MYC/FOXR2 have progression-free survival rates of only 19.2% and 16.7% ( 49 ). This substantial difference in survival highlights the potential prognostic importance of determining the DROSHA GPV status of patients with pineoblastoma. Developing clinical recommendations or guidelines for DROSHA genetic testing and surveillance, however, requires considerable further research. We strongly advocate for clinical studies and registries in order to both characterize the phenotypic spectrum of DROSHA -related tumor predisposition and learn the natural history of individuals and families with DROSHA GPVs, both with and without tumor diagnoses. Until an association with specific pineoblastoma subtypes can be confirmed or refuted and there is sufficient evidence obtained to guide management, we contend that genetic testing for DROSHA GPV in the context of a research study may be appropriate for all patients diagnosed with pineoblastoma. In particular, given our results and the apparently high frequency (12.7%–26.3%) with which DROSHA is biallelically inactivated in pineoblastoma tumor samples ( 20 , 49 ), we suggest that genetic testing for DROSHA is considered for every patient diagnosed with this condition whose methylation profile is indicative of molecular subgroup PB-miRNA1 or PB-miRNA2. Additionally, as the miRNA-altered groups of pineoblastoma are more commonly associated with older children (>3 years of age) relative to other pineoblastoma molecular subtypes, the diagnosis of pineoblastoma in these individuals should raise suspicion for DROSHA GPV ( 49 ). Again, in the context of a research study, we also advocate genetic counseling and testing for first-degree relatives of patients with pineoblastoma with DROSHA GPV, as well as close neurologic follow-up and the consideration of annual brain MRI surveillance up to age 21 years—the age of the oldest patient in our case series—for individuals with both DROSHA GPV and a close family member with pineoblastoma. Given the current paucity of data, more work is required to define surveillance recommendations for carriers of DROSHA GPVs with or without pineoblastoma in first-degree relatives. Similarly, given that a DICER1 GPV is a relatively rare cause of Wilms tumor, a DROSHA GPV may be a comparably and proportionately uncommon cause. As both are components of the miRNA processing pathway, we suggest consideration of adding DROSHA to hereditary Wilms tumor panels, particularly for bilateral/multifocal Wilms tumor or in the presence of DROSHA tumor variants. As with pineoblastoma, we strongly advocate that patients with Wilms tumor and a DROSHA GPV and their families participate in research and registry studies in order to define the role and penetrance of DROSHA GPVs in Wilms tumor and to guide clinical testing and surveillance recommendations. In summary, we present the first evidence of a novel heritable autosomal dominant DROSHA -related tumor predisposition that phenocopies components of DICER1 -related tumor predisposition. Given our observation of biallelic DROSHA inactivation in these tumors, our results suggest that pineoblastoma and Wilms tumor are uniquely tolerant of the loss of miRNA processing through the canonical microprocessor-dependent pathway and may point toward important insights into both the normal development of the tissue of origin for these tumor types as well as the etiology of these tumors. Ultimately, further work will be required to define and characterize the phenotypes associated with a DROSHA GPV. Such work may yield significant insights into the overlapping and unique mechanisms by which DROSHA along with DICER1 and DGCR8 regulate miRNA processing and participate in other cellular processes while contributing to an understanding of tissue-specific functions that may explain the observed differences in their clinical presentations.

miRNAs are a class of small RNAs that play a critical role in the posttranscriptional regulation of eukaryotic gene expression by RNAi. miRNAs function as ∼22-nucleotide RNA molecules that base-pair with target mRNA 3′ untranslated region sequences to suppress translation or trigger transcript degradation ( 1 ). Canonical miRNA biogenesis is a two-step process involving the RNases DROSHA and DICER1. miRNAs are initially transcribed as primary miRNA transcripts thousands of nucleotides in length, which are cleaved by the “Microprocessor” complex, composed of DROSHA and DGCR8, to form precursor miRNAs. Precursor miRNAs are RNA hairpins ∼65 to 80 nucleotides in length that are trafficked to the cytoplasm by the nuclear transport factor exportin-5 and Ran–GTP in which they are cleaved by DICER1, an RNase III, and its cofactor TARBP2, to form mature miRNA duplexes. One of these miRNA strands is loaded into a member of the Argonaute protein family and other proteins to form the RNA-induced silencing complex, which carries out target gene silencing ( 2 ). Heterozygous germline pathogenic variants (GPV) in DICER1 were initially found to be associated with pleuropulmonary blastoma and have since been shown to be associated with an increased risk for other, mainly pediatric-onset, tumors, including pineoblastoma and Wilms tumor (refs. 3 – 11 ). Typically, DICER1 GPVs with either acquired somatic RNase III hotspot missense mutations or, rarely, loss-of-function (LOF) inactivating mutations of the wild-type allele in the tumor are observed. DGCR8 GPVs with somatic loss of heterozygosity (LOH) predisposing to multinodular goiter with schwannomatosis have been described ( 12 , 13 ). In contrast to DGCR8 and DICER1 , GPVs in DROSHA have not been linked to any clinical phenotype. Acquired monoallelic somatic variation in DROSHA , however, has been reported in approximately 10% of patients with Wilms tumor, a pediatric kidney tumor, with somatic biallelic inactivation of DROSHA observed in rare cases ( 14 , 15 ). Wilms tumor has been linked to tumor predisposition because of GPVs in WT1 , DICER1 , and other genes ( 15 , 16 ). Missense germline variants in DROSHA have been reported in two patients with Wilms tumor, but the pathogenicity of these variants is unclear as there was no evidence for inactivation of the remaining wild-type allele in the tumor of either patient ( 8 , 14 , 17 , 18 ). Similarly, somatic biallelic LOF variants in DROSHA have been reported in pineoblastoma ( 4 , 19 , 20 ), a rare brain tumor developing from pineal gland parenchyma that most commonly presents in childhood. It is a highly aggressive primitive neuroectodermal tumor with a 5-year survival rate of approximately 60% ( 21 – 23 ). Interestingly, pineoblastoma, like Wilms tumor, is a component of DICER1 -related tumor predisposition. In this study, we present nine patients from eight families with a DROSHA GPV and the somatic acquisition of an LOF DROSHA mutation. Of these, eight had pineoblastomas, and one had a bilateral Wilms tumor. We then investigated the contribution of DROSHA GPVs to pediatric and adult cancers in large datasets and estimated the population prevalence and penetrance of germline DROSHA LOF variants.

All study subjects provided written informed consent to participate in a study of familial cancer genetics that was approved by the local Institutional Review Board. For family 2, this consent was for Beat Childhood Cancer Consortium research. All studies were conducted in accordance with recognized ethical guidelines of the U.S. Common Rule. The family investigated was ascertained by the Pediatric Familial Cancer Clinic at the University of Chicago. To protect the anonymity of the study subjects, the family pedigree was altered in ways that did not affect the genetic analysis. Germline DNA for whole-exome sequencing (WES) was obtained from whole blood. Tumor DNA from individual II-1 was isolated from formalin-fixed, paraffin-embedded (FFPE) scrolls after evaluation by a pathologist (>80% tumor). At least 1 μg of DNA was used for whole-exome capture using the NimbleGen SeqCap EZ Exome Library V2 kit (Roche Sequencing) for the germline samples and the Agilent SureSelect Human All Exon V4 kit (Agilent) for the tumor sample. Sequence reads were generated on an Illumina Genome Analyzer II (Illumina) for the germline samples and an Illumina HiSeq 2000 for the tumor sample. An average of 63.9 million 2 × 100 bp paired-end reads was generated for each sample. Full sequencing depth information is available in Supplementary Table S1. Sequencing reads were analyzed using the ExScalibur WES pipeline ( 24 ). The quality of raw reads was assessed by FastQC (v0.11.2; RRID: SCR_014583; ref. 25 ), followed by adapter clipping and 3′ overlap mate merging using SeqPrep (v1.1; RRID: SCR_013004; ref. 26 ). Processed reads were aligned to the human reference genome assembly (hg19) using the short-read aligner BWA-mem (RRID: SCR_022192; ref. 27 ). Exon coverage was calculated using bedtools (v2.23.0; RRID: SCR_010910; ref. 28 ). Unmapped reads or those with mapping quality <30 were excluded from further analysis. Read duplicates were removed using the Picard tools MarkDuplicates program (v1.121; RRID: SCR_006525; ref. 29 ). Alignments were postprocessed by GATK (v3.4.0) for insertions/deletions (InDel) realignment and base quality score calibration (RRID: SCR_001876; ref. 30 ). Four variant callers including GATK HaplotypeCaller (v3.4.0), FreeBayes (v0.9.20), Platypus (v0.8.1), and Samtools mpileup/bcftools (v1.2; RRID: SCR_005227) were used to detect single-nucleotide variants (SNV) and InDels ( 31 – 33 ). Variant calls passing the internal quality filters of each caller were then filtered to remove potential false positives based upon (i) variant quality score <50, (ii) read coverage <4, (iii) genotype quality score <20, or (iv) location within an SNV cluster in which at least three SNVs were called within a 10 bp window. After combining results from the four callers, variants called by at least two callers were carried forward for annotation using ANNOVAR (November 2014 release, RRID: SCR_012821; ref. 34 ). Population minor allele frequencies (MAF) were derived from the 1000 Genomes Project (May 2013 release), the National Heart, Lung, and Blood Institute GO Exome Sequencing Project (ESP, version ESP6500SI-V2; Exome Variant Server; September 2014 accessed), and the Exome Aggregation Consortium Project (ExAC v0.3; URL: http://exac.broadinstitute.org ; February 2015 accessed; refs. 35 , 36 ). Each variant was annotated for pathogenicity using SIFT (RRID: SCR_012813; ref. 37 ), PolyPhen-2 (RRID: SCR_013189; ref. 38 ), MutationTaster (RRID: SCR_010777; ref. 39 ), Likelihood Ratio Test (LRT) ( 40 ), and Radial SVM (RRID: SCR_005178; ref. 34 ). They were assessed for multispecies conservation using GERP++ (RRID: SCR_000563; ref. 41 ) and PhyloP (RRID: SCR_005178; ref. 42 ). To prioritize rare germline variation for further investigation as candidate pineoblastoma susceptibility mutations, we required that a variant passing our quality control pipeline meet the following criteria: (i) It has a general population MAF ≤0.01 in the 1000 Genomes Project, ESP, and ExAC databases; (ii) it is nonsynonymous, modifies a splice site, creates a stop codon, or is an InDel creating a frameshift; and (iii) it is deleterious as predicted by at least one of the variant functional consequence prediction algorithms. To identify somatic variants, we analyzed paired normal/tumor DNA for individual II-1 using MuTect (RRID: SCR_000559; ref. 43 ), Strelka (RRID: SCR_005109; ref. 44 ), Virmid (RRID: SCR_006780; ref. 45 ), and VarScan2 ( 46 ). All four programs detect somatic SNVs. Strelka and VarScan2 also detect somatic InDels. Variants passing the internal quality control of each caller were retrieved and filtered for high-confidence calls based upon (i) variant quality score ≥20, (ii) sequencing read depth ≥8, and (iii) allele fraction in the tumor sample of >0.20 and allele fraction in the germline sample of <0.05. We then combined somatic variants identified by at least two of the four calling algorithms for downstream analysis. Variants were annotated using ANNOVAR (November 2014 release), and those with MAF ≥0.01 in the 1000 Genomes, ESP, or ExAC databases were excluded from further analysis. Finally, somatic variants were manually inspected in Integrative Genome Viewer (RRID: SCR_011793; ref. 47 ) to confirm that the variant allele was not present in the matched normal sample or any other germline sample sequenced. The detection of the somatic and germline variants in the DROSHA gene was part of a paired tumor/normal WES/whole-transcriptome sequencing clinical testing offered by Sema4 (currently GeneDx Holdings). The test was clinically validated and approved by New York State and was made available to all 50 states. Tumor DNA whole exome (WES) and normal DNA WES were sequenced at 250× and 100× sequence coverage, respectively. The tumor RNA was sequenced at an average of 100 million reads per sample. Twist Core Exome kit was used for WES, and Illumina RNA Stranded (Gold) or RNA Exome kit was used for whole-transcriptome sequencing. Input requirements were 200 ng tumor DNA and 200 ng tumor RNA. The analysis pipeline utilized publicly available GATK4 best practices and custom tools. The assay was validated with a limit of detection down to a 0.03 variant allele frequency (VAF) for SNVs. Copy number alterations were detected at a single-copy resolution for copy-number alterations that range from focal (<5 Mb) to chromosomal arm or whole chromosome. The clinical next-generation sequencing (NGS) panel analysis on germline and tumor DNA was performed using SureSelect QXT enrichment with custom probes (Agilent) and sequencing on NextSeq 500 (Illumina) according to the manufacturer’s instructions. For methylation analysis, whole-genome sequencing (WGS) was performed on the MinION Mk1B device (Oxford Nanopore Technologies), as previously described ( 48 ). For the germline WGS, library preparation was performed with the NEBNext Ultra II End Repair/dA-Tailing Module and Ligation Module (New England Biolabs). For the tumor WES, enrichment was performed with capture probes (Twist Human Core Exome Kit + IntegraGen Custom v1, Twist Bioscience). The RNA sequencing library was performed with the NEBNext Ultra II RNA First Strand Synthesis Module and Directional RNA Second Strand Module (New England Biolabs). NovaSeq 6000 (Illumina) was the sequencing system for the WGS, WES, and RNA sequencing analyses. These analyses were conducted by the SeqOIA laboratory as a clinical test. Methylation analysis was the same as family 3. Targeted testing for germline variants in DROSHA was conducted on DNA extracted from a buccal swab/saliva sample using a commercially available clinical test (Prevention Genetics). Clinical tumor DNA sequencing was performed using the Solid Tumor Panel (version 1) in the Clinical Laboratory Improvement Amendments– and College of American Pathologists-certified Texas Children’s Hospital Cancer Genomics Laboratory. Tumor DNA methylation profiling was performed using the Illumina MethylationEPIC array (Illumina Inc.). Raw intensity was normalized by performing background and dye-bias correction, batch-corrected for tissue (FFPE/frozen) and array (450k/EPIC) type, and converted to methylated and unmethylated signals, and finally, β values used for further analysis were calculated. Unsupervised clustering (t-distributed stochastic neighbor embedding; iterations = 500 and perplexity = 30) was performed by selecting the 32,000 most variably methylated CpG sites according to the median absolute deviation from tumor methylation data and a published brain tumor DNA methylation reference cohort as previously published ( 49 ). Tumor methylation data were also analyzed using the Molecular Neuropathology brain tumor classifier (v12b6). Germline genetic testing was performed with the Wilms tumor NGS panel with exome-wide copy-number variation detection and the ATM -targeted family variant gene test from Prevention Genetics. Beckwith–Wiedemann syndrome molecular testing was performed with 11p15.5 methylation and copy-number analysis at the University of Pennsylvania Genetic Diagnostic Laboratory. Clinical paired tumor-germline sequencing and a 115-gene cancer predisposition panel from St. Jude Children’s Research Hospital Clinical Genomics Laboratory were also performed. Samples were ascertained from the Intermountain Biorepository and Huntsman Cancer Institute Pediatric Cancer Biobank by searching for patients with pathology-confirmed pineoblastoma, Wilms tumor, or other DICER1 -related conditions. FFPE scrolls were cut from the tumor blocks and matched normal tissue of each patient. Patient 11 did not have tumor tissue, and patients 5, 6, and 7 did not have matched normal tissue available. DNA was extracted from FFPE using the FFPE Direct kit (Thermo Fisher Scientific) per the manufacturer’s protocol and Hemo-De as necessary. For one patient (#3), peripheral blood mononuclear cells were used as the source of germline DNA. DNA was extracted from peripheral blood mononuclear cells using the QIAamp Blood Mini Kit (Qiagen) per the manufacturer’s protocol. Following extraction, sequencing libraries were created for each sample using the Ion AmpliSeq Library preparation kit (Life Technologies). Briefly, 20 ng of sample DNA was mixed well, and 7.5 μL was added to 5 μL of 5× Ion AmpliSeq HiFi Mix. This was again mixed well by pipetting, and 5 μL of this mix was added to 5 μL of 2× Primer Pool mix. The plate was sealed, followed by a quick vortex and spin, and then placed on the thermocycler using a cycling profile with a 4-minute extension time and a total of 20 cycles. Reactions using the same DNA but different primer pools were combined. The combined PCR products were end-repaired, and DNA ligase was used to ligate Ion Torrent adapters P1 and A. After purification with Axygen AxyPrep Fragment Select-I beads, the concentration of each library was determined using the Ion TaqMan Assay (Life Technologies) following the manufacturer’s protocol. Libraries were pooled at a concentration of 100 pmol/L and submitted to the university’s sequencing core. A custom sequencing panel was generated using the Ion AmpliSeq Designer (5.6.2). Included in the panel are the coding sequences, promoter regions, and untranslated regions for DICER1 and DROSHA with 25 bp padding on each exon (BED file attached to supplement). The sample libraries were sequenced using the P1 chip on the Ion Proton. We queried the germline WGS or WES of all probands with a central nervous system tumor diagnosis from the Open Pediatric Cancer project (OpenPedCan; n = 2,148; bioRxiv 2024.07.09.599086). Germline VCFs annotated with VEP v105 and gnomAD 3.1.1 were processed by Kids First ( https://github.com/kids-first/kf-germline-workflow , v1.1.0) and obtained through data use agreements with the Children’s Brain Tumor Network ( cbtn.org ) and the Pacific Pediatric Neuro-Oncology Consortium (pnoc.us). We annotated variants with ClinVar (May 7, 2022), ANNOVAR (annovar_humandb_hg38_intervar), InterVar version 2.2.1, and AutoPVS1 version 2.0.0 using the Pathogenicity Preprocessing Workflow version 1.1.0 at https://github.com/d3b-center/D3b-Pathogenicity-Preprocessing ( 34 , 50 , 51 ). We performed filtering and pathogenicity annotation of rare variants using AutoGVP v1.0.1 ( 52 ). Briefly, variants were filtered for noncancer AF popmax < 0.001 in gnomAD 3.1.1, depth ≥ 10, sample VAF ≥ 0.2, and MAF < 0.05. ClinVar conflicting calls were resolved using a ClinGen Concept ID list to filter submissions using the “–conflict-res latest” argument. We identified one patient, PT_QD6KKKJH, diagnosed with pineoblastoma to have a rare pathogenic variant in DROSHA . The coordinates of this variant were lifted from hg38 to hg19 using the UCSC liftover tool. We queried the publicly available somatic alteration data available in OpenPedCan release v15 (SNVs, InDels, copy-number variations, and structural variants) and identified a focal chromosome 5 deletion containing the entire DROSHA gene in each of the two tumor samples from this individual sequenced at diagnosis, BS_9BN45DFK and BS_B91XGSA5 (bioRxiv 2024.07.09.599086). Molecular subtypes for pineoblastoma samples were determined using the v12 DKFZ methylation brain tumor classifier. Pineoblastoma subtypes with a subclass score ≥0.8 were determined to be high confidence and are listed in Supplementary Table S2. SpliceAI was used to predict the likely impact of splicing variants on the alternative splicing of DROSHA (RRID: SCR_024532; ref. 53 ). Variants were uploaded to the interactive web-based tool using their chromosome, hg19 position, reference allele, and alternative allele. The delta score and delta type were used to predict the impact of the variant on alternative splicing. We downloaded The Cancer Genome Atlas (TCGA) from the Genomic Data Commons using BAM slicing. Germline variant calling was performed as previously described ( 54 ). DROSHA variants that were identified as pathogenic or likely pathogenic were classified according to the American College of Medical Genetics and Genomics (ACMG) standards using AutoGVP v1.0.0 ( 52 ). We queried the Childhood Cancer Survivor Study (CCSS; phs002072.v1.p1) and St. Jude PeCan Portal ( 55 ). CCSS data were analyzed as previously described ( 56 ). The UK Biobank (UKBB) cohort consists of nearly half a million consented participants ages 37 to 69 years at the time of enrollment. The protection and review of human subjects were carried out through the North West Multicentre Research Ethics Committee. In this study, we used 469,787 individuals who were exome-sequenced ( 57 ). The DiscovEHR cohort consists of 170,503 participants ages 0 to ≥89 years enrolled in the Geisinger Health System ( 58 ). This study was approved by the Geisinger Institutional Review Board ( 59 ); participants consented to a broad research use of their exome and linked electronic health record (EHR) data. Gene annotations were done using SnpEFF, and general population frequency and in silico prediction are annotated using ANNOVAR. Any variants with GQ < 30 and ABHet < 0.2 were excluded from the analysis. Variant classification was performed using AutoGVP v1.0.0 ( 52 ). DROSHA wild-type, benign, or likely benign variations were selected as noncarriers ( 52 ). There were 269,334 and 123,731 individuals in UKBB and Geisinger, respectively, who harbored benign or likely benign or canonical DROSHA variation. For UKBB, EHR-linked phenotypes and date and age at diagnosis were retrieved on October 16, 2023, from fields 41270/41280 [Diagnosis: International Classification of Diseases (ICD10)], 40001/40007 (Underlying cause of death: ICD10), and 40006/40008 (Cancer registry: type of cancer ICD10). For Geisinger, the EHR and cancer registry were queried. Statistical analyses were performed using R version 4.1.0. All OR analyses were corrected for covariates, sex, ethnicity, body mass index, and smoking status. The forest plot was generated using the R package ggplot2, the oncoprint was generated using the R package ComplexHeatmap, and the lollipop plot was generated using the R package trackViewer. TCGA Pan-Cancer Atlas germline data are available on Genomic Data Commons at https://gdc.cancer.gov/about-data/publications/PanCanAtlas-Germline-AWG . St. Jude PeCan Portal can be accessed at https://pecan.stjude.cloud/#/home . The variant information for this study has been deposited at the NCBI ClinVar ( https://www.ncbi.nlm.nih.gov/clinvar ) under accession numbers SCV005442816-SCV005442821 and SCV005442823-SCV005442829. For information related to DNA sequencing and tumor characteristics, please contact the corresponding author. Patient information sharing is limited by the terms of informed consent.

Supplemental Tables including Representativeness of Study Participants Figure S1 Figure S2 Figure S3 Figure S4 Figure S5 Supplementary Table Legend