In the current study we present WGBS data on 47 samples from relapsed pediatric cancers representing 11 different tumor types. We also examined corresponding cfDNA methylation from 17 patient matched plasma samples. Among the key findings, is the identification of a pan-pediatric cancer DNA methylation tumor signature found in both gDNA from tumor tissue and cfDNA. Previous studies have identified DNA methylation changes in many of the pediatric cancers evaluated in this study, including NBL 35, OS 36, 37, WT 38, AE 39, HB 40, ERMS 40, and FHC 41. Fewer studies have evaluated the DNA methylation changes that occur in pediatric cancers after relapse42, 43, 44. Novel to our study is the finding that in addition to tumor specific changes, DNA methylation patterns were also shared across the myriad of tumor types analyzed.
A previous study performed by Yang et al.45 demonstrated that there are DNA methylation changes common to multiple adult cancer types, which are frequently associated with tumor suppressor genes, and often associated with survival. However, to date, no study has identified DNA methylation changes common to multiple pediatric cancers. Given the rarity of many pediatric cancers, identifying DNA methylation patterns across multiple pediatric cancer types might have great value for developing biomarker approaches for disease recurrence and early detection– a feat otherwise very difficult to achieve for a given single tumor type.
Among the most interesting tumor specific DNA methylation changes was identification of DNA methylation deserts, specific to a subset of NBL, where large regions of DNA methylation are essentially eroded. The functional significance of these DNA methylation deserts are unknown but are distinct from previously reported PMDs in the degree of the observed loss of DNA methylation. Previous work by Brinkman et al. found that PMDs are associated with CpG island methylation in breast cancer46; PMDs have also been associated with lamin-associated domains (LADs) 47. Increased expression of FOXA1, one of the few genes found DNA methylation deserts, has been previously linked to late recurrence48. Further research is needed to explore possible mechanisms of these desert regions in NBL.
We identified a subset of tumors with a global hypermethylator phenotype – an atypical cancer phenotype in pediatric and adult tumors. Hypermethylation in cancer typically occurs in the promoter regions of tumor suppressor genes49, 50, 51, 52, 53 and in CpG islands, where it is associated with poor prognosis in many adult and pediatric cancers and can be associated to CIMP54, 55, 56, 57. However, global hypermethylation is rarely observed in cancer. While MRT, the most prominent hypermethylator tumor presented here, has been previously shown to have focal hypermethylation 58, the global hypermethylator phenotype we observed has not been previously described. Previous work by Kenny et al. found that SMARCB1 restoration in MRT cell lines resulted in widespread chromatin activation 59; conversely this hypermethylator phenotype could be linked to genome-wide chromatin inactivation. However, we only evaluated 2 MRT samples from 1 patient and further investigation is required to determine if this hypermethylator phenotype is common and is linked to SMARCB1 related chromatin inactivation.
We also profiled recurrent lesions from multiple metastatic sites in 5 patients (P01-019 n = 2, P01-020 n = 2, P01-022 n = 5, P01-28 n = 2, and P01-036 n = 2) and we found that methylation values were largely consistent between metastatic sites. P01-022 metastases were particularly concordant with each other despite multiple CNA differences between each metastasis (Fig. 4A) and discordant mutational profiles indicating that DNA methylation changes may precede metastasis whereas CNA and SNV changes can be acquired during the metastatic process. This is consistent with previous work by Ili et al. and Reyngold et al. which found genome wide methylation profiles of primary and metastatic tumors are largely conserved60, 61.
Unlike adult cancers, most cases of pediatric cancer can be traced to a single genetic driver, and these genetic alterations tend to be highly tumor specific. For example, loss of SMARCB1 specifically leads to development of MRT62. Specific gene fusions cause alveolar rhabdomyosarcoma (PAX3/7-FOXO1)63, Ewing’s sarcoma (EWS-FLI-1)64, and CML (BCR-ABL1) 65. Mutations to certain genes such as RB are specific to retinoblastoma and OS, while mutations in TP53 are extremely common in multiple pediatric cancers. WGBS is the gold standard assay for methylation evaluation as it evaluates every CpG in the genome. Additionally, we demonstrated its utility for copy number estimation and somatic variant calling – maximizing data generation from each sample. This is particularly useful when dealing with rare tumor types or sample types with very little available DNA (such as cfDNA). CNA analysis from WGBS was sensitive to both large-scale alterations and focal gains/losses such as MYCN gain in NBL, SMARCB1 loss in MRT, and PTEN loss in ERMS. SNV analysis from WGBS data revealed a set of 47 commonly mutated genes across multiple cancer types and 3 mutations that are known to be clinically significant. However, SNV analysis from WGBS remains constrained by currently available informatic tools.
Somatic mutation analysis found TREM8 was the most frequently mutated gene; it was mutated in 4 samples and 3 cancer types (NBL, OS and Teratoma). TREM8 is an E3 ubiquitin ligase protein that helps control TP53 function66 and has been associated with multiple congenital abnormalities67, 68, 69, however extensive alterations have not been previously reported in pediatric cancers. We also found clinically significant variants in MAP2K1, SMAD4 and RB that have been documented in OMIM. The MAP2K1 variant, observed here in FHC, has been specifically linked to cardiofaciocutaneous syndrome and melorheostosis; somatic mutations in MAP2K1 have been extensively reported in cancer 70, 71, 72, 73, 74 but not pediatric FHC. MRT tumors in this study harbored SMAD4 and RB clinical variants, which were previously reported in other cancers. The SMAD4 mutation has been associated with pancreatic and colon cancers75, 76. The RB mutation has been observed in multiple retinoblastoma cases 77, 78, 79. Neither of these mutations have been previously reported in MRT.
A key component of this study was evaluating the degree to which genomic alterations in solid tumors were reflected in cfDNA, as cfDNA is rapidly becoming a major mode for non-invasive screening, diagnosis, treatment, and monitoring of human tumors. There are three primary methods for detecting cancer markers in cfDNA: detection of SNVs, CNVs, and aberrant DNA methylation. Previous work in renal cell carcinoma by Lasseter et al.80 found that cfDNA methylation (using MeDIP-seq) was far more sensitive and specific than cfDNA SNV markers. In the current study, we found that detecting tumor tissue-derived DNA methylation using WGBS was superior to detecting CNVs and SNVs in cfDNA. In the this study, only 3/17 cfDNA samples reflected the copy number profile of the tumor, whereas the remaining samples lacked much of the signal found in tissue. By contrast, we found cfDNA methylation to resemble tissue DNA methylation more robustly, where on average approximately 25% of DMRs were common to tissue and plasma in each sample. We also found that hypermethylated regions were less likely to be retained in plasma and were often found to be hypomethylated. While the reasons for this are unclear, the implications are significant as it would impact feature discovery in tissue.
In addition, we identified a mDMR signature that was able to differentiate individuals with pediatric cancer, including extremely rare tumors such as DSRCT and MPNST from healthy individuals. We examined this mDMR signature in cfDNA and demonstrated excellent sensitivity and specificity to discriminate tumor from normal samples on a ROC curve. Since this signature is derived from relapsed patients, this signature may reflect a method to conduct recurrence monitoring via minimal residual disease detection in pediatric cancer. Further, this signature was also highly sensitive and specific to adult cancers as seen by samples we analyzed from TCGA, which suggests that in addition to recurrence monitoring, it may have utility as a pan-cancer detection signature in adult cancers too.
In summary, we have performed a comprehensive analysis of multiple pediatric cancers using WGBS. We identified a pan-cancer methylation signature, detectable in cfDNA, common to multiple pediatric cancer types, including extremely rare neoplasms such as DSRCT and MRT. We also show that WGBS can be exploited to retrieve clinically relevant copy number and somatic mutation data, in addition to genome wide methylation profiling. We found that DNA methylation was superior to SNV and CNA at detecting tumor specific signal in cfDNA. The pan-cancer cfDNA methylation signature in this study has potential utility in minimal residual disease monitoring and early detection and warrants further investigation in both pediatric and adult cancer.