To identify somatic mutations, we performed whole-exome sequencing on tumor-normal pairs of 50 GCT patients aged 0–18 years (discovery cohort, Supplementary Tables 1 and 2). We identified 1180 somatic mutations in total, including 299 somatic single-nucleotide variants (SNVs) and 19 somatic small-scale insertion/deletions (INDELs) that were predicted to be protein-altering. To validate mutations in the discovery cohort and study the prevalence of these mutations in a larger cohort, we performed custom-capture deep sequencing in 129 GCTs (48 out of 50 cases of the discovery cohort with sufficient DNA and an additional 81 GCTs) (Supplementary Table 1). Therefore, 131 GCT cases were studied by either whole exome or targeted deep sequencing. The 51 genes chosen for the validation set were selected because they were frequently mutated in the discovery cohort or are candidate GCT drivers based on previous studies15,17,21,22. The mutational landscape of these candidate driver genes is shown in Fig. 1a and Supplementary Data 1.
Type I, Type II and ovarian GCTs in our datasets had a low mutation rate (0.23 non-silent mutations per Mb on average), consistent with previous reports from adult type II TGCT16,18,23,24. All histologic subtypes have low mutation rates that are not statistically different from one another (Fig. 1b). The most common recurrently mutated gene was KIT, with nine mutations identified in seven patients. Similar to previous reports13, all KIT mutations were identified in seminoma patients (Fig. 1a). Among these mutations, seven are in exon 17 (encoding the kinase activation loop), one is in exon 11 (encoding the regulatory domain of enzyme) and one is in exon 2 (encoding the Ig-like-C2-type 1 domain; Fig. 1c). In addition, KRAS mutations were found in two EC, four seminoma, two MMGCT, and two YST, and NRAS mutations in one seminoma and one YST. Consistent with previous reports25, KRAS, NRAS, and KIT mutations were mutually exclusive within a given tumor.
The most striking finding was the prevalence of mutations in six WNT pathway genes (CTNNB1, APC, LRP5, TCF7L2, CHD8, and FAT1) in ten YSTs and three MMGCTs containing YST elements. Previous studies in other tumor types have suggested that loss-of-function mutations in APC26, CHD827, and FAT128 could all activate the WNT pathway and promote tumorigenesis. We observed missense mutations in these six WNT genes, as well as stop-gain mutations in APC and FAT1 that likely result in truncated proteins (Table S3). Moreover, we also found mutations of FAT2, FAT3, and FAT4, which share high sequence similarity to the FAT1 gene and have been proposed as candidate tumor suppressors29. By integrating RNA-seq and whole-exome sequencing data, we found a significantly increased CTNNB1 expression in GCT cases with somatic protein-altering mutations in FAT2 or FAT3 genes compared to GCT cases without such mutations (Mann–Whitney U test, P = 0.001; Supplementary Fig. 1). FAT family genes were mutated in Type I GCTs, whereas KIT and KRAS were almost exclusively mutated in Type II GCTs (Fig. 1a).
We also observed frequent mutations in chromatin remodeling genes, including five missense, one nonsense, and one frameshift deletion mutation in the Histone H3K4 methyltransferase MLL2/KMT2D. All seven KMT2D mutations were in YSTs. Besides KMT2D, another 20 chromatin remodeling genes were mutated in our cohort (Fig. 1a). These results highlight the potential role of epigenetic dysregulation in GCTs.
DNA repair genes were also frequently mutated in pediatric GCTs. Six tumors had mutations in ATM, a DNA damage response regulator. Five of six ATM mutations were found in YSTs. We also observed seven other DNA repair genes with mutations (Fig. 1a), including PRKDC (also known as DNA-PKcs). A recent report suggests that mutation of PRKDC might elevate DNA damage and mutation rate in cancer30.
Lastly, we investigated known or suspected GCT driver genes in the validation study. We observed somatic protein-altering mutations in DMRT1, CBL, NOTCH1 (three patients), NOTCH2 (four cases), EGFR, NRAS (two cases), PTEN (two cases) and MTOR. We also observed recurrent mutations in genes of the Hedgehog signaling pathway, including PTCH1 (three cases), GLI1 (two cases), and GLI2 (Fig. 1a). Supplementary Data 1 provides details of genes with somatic SNVs in multiple tumors, including scores of the predicted deleterious effect of variants. However, further studies will be required to fully assess the functional impact of these variants.
DNA copy-number analysis
We assessed 148 GCTs for copy-number alterations with high density SNP arrays, using GISTIC 2.031 to determine significance. Type I tumors exhibited copy-number gains at 12p, 20q and 21 and losses at 1p and 6q; Type II tumors exhibited gains at 12p and 20q, and losses at 10 and 19q (Fig. 2a). GISTIC identified recurrent focal copy-number changes in several genes associated with germ cell development, GCT predisposition, and WNT signaling, including gain of KRAS, ATF7IP, CCND2, DPPA3, GDF3, NANOG, LRP6, SOX18, and WNT5B, and loss of CXCL12, INSL3, NANOS3, SOX2, RET, and BTRC.
Recently, Taylor-Weiner and co-workers identified recurrent chromosome arm-level amplifications and reciprocal loss of heterozygosity as a major feature of adult testicular (type II) GCTs19. To separate type I from type II GCTs, we queried our SNP array dataset for LOH events and stratified the results by age (greater or less than 6 years of age32). We identified arm level LOH events in the majority of chromosomes of GCTs from patients older than 6 years. However, these large-scale LOH events were significantly less common in tumors from younger patients (Fig. 2b and Supplementary Fig. 2).
Figure 3 summarizes the pattern of somatic mutations and copy-number changes observed in the most frequently affected pathways in GCTs. As a complement to these analyses, we also performed whole-genome sequencing analysis in 10 tumor-normal pairs at 30× resolution. We used the DEFOR33 and SCHALE34 algorithms to assess copy-number changes, structural alterations and loss of heterozygosity (LOH) in the tumors. The results are shown in Supplementary Figs. 3–5. We observed recurrent somatic (tumor-specific) focal- and arm-level structural alteration events, recapitulating those described by lower-resolution array technologies in our study and by other groups, such as 1p gain, 6q loss and 12p gain. Of note, an ovarian pure yolk sac tumor from a 23-year-old female did not exhibit any evidence of chromosome 12p gain (Supplementary Fig. 3), supporting the idea that such tumors are more closely related to Type I YSTs of young children. We also observed previously unreported copy-number changes and loss of heterozygosity (LOH) events. In addition, we analyzed RNA-Seq data using DEFUSE35 to computationally identify possible gene fusions, which have not been reported previously (Supplementary Data 2). Further studies will be required to test the functional significance, if any, of these genetic alterations.
Frequent DNA copy-number, promoter methylation and gene expression alterations of WNT pathway genes in type I and II tumors
The occurrence of somatic mutations in WNT pathway genes in GCTs prompted us to examine the WNT pathway more closely by analyzing DNA copy-number, promoter methylation and gene expression data. A striking pattern emerged, with WNT pathway activators demonstrating low levels of promoter methylation and frequent focal copy-number gains, while repressors of WNT signaling display a reciprocal pattern, with high levels of promoter methylation and frequent focal copy-number losses (Fig. 4a, b; Supplementary Table 3). This pattern was present in type I and type II GCTs (both seminomas and non-seminomas), suggesting it may be a general feature of extracranial GCTs independent of age.
As an independent assessment of the effect of copy-number alterations on WNT pathway genes, we analyzed results from the TCGA study of testicular GCT16. Similar to pediatric GCTs, adult testicular GCTs showed a high frequency of tumors exhibiting focal copy-number gain of WNT activators, loss of WNT repressors, or both (Fig. 4c). Tumors exhibiting more than 5 such changes (designated WNT CN-rich, n = 85) had higher expression of beta-catenin compared to tumors with 5 or fewer changes in WNT pathway genes (WNT CN-poor, n = 71, Fig. 4d). To rule out the possibility that these patterns resulted from non-specific genomic instability in WNT CN-rich tumors, we compared the average numbers of genes exhibiting copy-number gains or losses in WNT CN-rich and WNT CN-poor tumors. The two groups did not exhibit significant differences (Supplementary Fig. 6).
WNT pathway activity has prognostic significance in GCTs
Based on these observations, we predicted that GCTs would show evidence of active WNT signaling. We compared the expression level of six frequently used markers for WNT activation (CTNNB1/beta-catenin, TCF1, TCF4, FZD7, MYC and CCND1) in normal and tumor tissue. Compared to human PGCs36 and normal testis, GCTs showed evidence of elevated WNT pathway activity, with highest levels in type I tumors (Fig. 5a,b).
To determine the possible prognostic significance of WNT pathway activation, we tested the association between WNT gene copy-number alterations (defined as gain of WNT activators or loss of WNT repressors) and outcome in our dataset. Patients whose GCTs harbored no focal WNT gene copy-number alterations experienced no relapses (Fig. 5c, left bar) and had 100% survival (Fig. 5d, black curve), while patients whose tumors had focal copy-number alterations of one to five WNT genes (WNT CN-poor group) had a slight but not significant increase in relapse (7% of patients) (Fig. 5c, middle bar) and a small decrease in survival rate (Fig. 5d, purple curve). However, in the third group of GCT patients with focal copy-number alterations of more than five WNT genes (WNT CN-rich group), we found a striking increase in occurrence of relapsed tumors (~30%, p = 0.0007, Fig. 5c, right bar) as well as a significantly decreased survival rate of patients (log-rank test, p = 0.038, Fig. 5d, orange curve). We obtained similar results using CN values of 3, 7 or 9 as the threshold value (Supplementary Figs. 7, 8).
To further test the association between WNT pathway activation and poor outcome, we evaluated an independent, previously described cohort of 108 non-seminomatous TGCT patients37,38. For three known WNT activator genes (FZD1, FZD7, and CTNNB1) with frequent copy number gains as mentioned above, we observed significant associations between expression level and survival in this cohort (Fig. 5e). Taken together, these results suggest that aberrant activation of WNT pathway contributes to increased relapse and poor survival of GCT patients. The sample size of our dataset did not permit a separate evaluation of type I and type II tumors.
Small molecule WNT inhibitors suppress the growth of GCT cells in vitro
The discovery of aberrant WNT pathway activation in GCTs has important translational implications, as several small molecule WNT inhibitors are in clinical development for treatment of cancer39,40. We treated GCT cell lines GCT44 and 1411H (YST), NTERA-2 (EC) and TCam2 (seminoma) with two different WNT inhibitors: the tankyrase inhibitor IWR-1 (Fig. 6a) and the PORCN inhibitor LGK-974 (Fig. 6b). Both inhibitors reduced the growth of GCT cell lines, with the largest effects in the YST cells.
Finally, we assessed the effect of WNT inhibitors in vivo. WNT signaling plays important roles in stem cells, including cancer stem cells41,42,43. The WNT target gene PIWIL144 confers stem cell fate and supports cancer cell growth45,46. PIWIL1 is overexpressed in human GCTs47, which exhibit features of impaired differentiation48. We previously reported that male zebrafish bearing mutations in the bmpr1bb gene develop testicular GCTs with gene expression similar to human GCTs49. The tumors exhibit impaired germ cell differentiation and elevated piwil1 expression21. Using a piwil1:eGFP transgenic zebrafish reporter line50 that permits live visualization of GCTs in bmpr1bb mutants (Supplementary Fig. 9), we tested the effects of WNT inhibition. We treated males with GCTs for 7 days of with DMSO vehicle control or with IWR-1. Treatment with IWR-1 led to a striking decrease of eGFP expression (Fig. 6c–e), indicating that WNT inhibition downregulates activity of the piwil1 promoter.
Because PIWIL1 promotes stem cell fate at least in part by inhibiting differentiation45,46, we used two complementary assays to test whether the loss of eGFP signal was accompanied by evidence of increased differentiation in the tumors. First, histologic examination of H&E-stained tumor sections (Fig. 6f) showed that control DMSO-treated tumors consisted of sheets undifferentiated germ cells, with only scattered islands of mature spermatozoa, as we previously described51. In contrast, tumors from IWR-1 treated fish exhibited markedly more complete differentiation, with many lobules showing the full range of spermatocytic differentiation. Upon differentiating, germline stem cells enter meiosis. Therefore, we next used phosphohistone H2AX (pH2AX) as a marker of meiotic cells52, we found that the WNT inhibitor-treated tumors exhibited increased pH2AX signal (Fig. 6g–i), in the characteristic clustered pattern of spermatocytes synchronously entering meiosis (Fig. 6h, inset). Thus, WNT inhibitor treatment interferes with the stem cell program in germ cell tumors and promotes differentiation of the tumor cells.