Frequency of Pathogenic Germline Variants in Cancer-Susceptibility Genes in the Childhood Cancer Survivor Study

Abstract

Survival rates have statistically significantly improved in recent decades, yet cancer still remains the leading cause of death by disease in children (1). As a result of advances in cancer treatment and supportive care, a growing population of childhood cancer survivors are at risk for long-term therapy-related sequelae (2-4). Most childhood cancers are not associated with identified environmental risk factors, and accumulating evidence suggests germline genetic susceptibility is an important contributor to pediatric cancer etiology (5-8).

To date, 4 studies (n = 150-3006) have identified pathogenic or likely pathogenic (P/LP) germline variants in well-described cancer-susceptibility genes (CSG) in 5.8%-10.0% of pediatric cancer patients (6-9). Germline variants in TP53 were most commonly observed in 3 studies with up to 4.5% of cases carrying a P/LP variant (7-9). In the only pediatric cancer survivor cohort analysis, 6% of children harbored a P/LP germline variant in a CSG, most commonly in RB1 and NF1 (6,10). Acute lymphoblastic leukemia cases (11) and osteosarcoma cases (12) with germline TP53 P/LP variants have also been observed to die earlier or present with higher risk disease compared with the cases without germline TP53 P/LP variants, but this has not been evaluated for CSGs in multiple cancer types. Most pediatric cancer studies have evaluated only a subset of CSGs, limiting our understanding of the broader spectrum of defined and plausible CSGs, including those important in adult-onset cancers.

Despite mounting evidence of the importance of germline susceptibility in pediatric cancers, a comprehensive understanding across pediatric cancer types and outcomes is needed. To better understand the genetic etiology of primary pediatric cancers, we conducted a comprehensive analysis of 5451 long-term cancer survivors from the Childhood Cancer Survivor Study (CCSS) using exome sequencing (ES) and characterized the frequency of germline pathogenic variants in CSGs in individuals of European ancestry.

Methods

Study Population

CCSS is a multi-institutional research effort that established a large and longitudinally characterized cohort of 14 361 participants who survived childhood and adolescent cancer for 5 or more years in its original cohort diagnosed at 1 of 26 North American hospitals between 1970 and 1986 (13-15). Participants were younger than 21 years of age at diagnosis, and cancer types included bone, central nervous system, Hodgkin lymphoma, leukemia, neuroblastoma, non-Hodgkin lymphoma (NHL), soft tissue sarcoma (STS), and Wilms tumor. Specific cancer types are outlined in Table 1. Among survivors actively participating in CCSS, those who provided a germline sample (ie, blood, buccal cells, or saliva) survived a longer period (mean follow-up = 32.5 years) compared with patients who did not provide a sample (mean follow-up = 24.2 years). Additional details are provided in the Supplementary Methods (available online).

Controls consisted of 597 cancer-free adults [median age at the last follow-up = 69 years, see Table 2; selected for European ancestry from the National Cancer Institute Prostate Lung, Colon and Ovarian Cancer Prevention Trial (16) and the American Cancer Society Cancer Prevention Study II (17); hereafter, “controls”]. All participants provided written informed consent and were recruited through institutional review board–approved protocols.

Exome Sequencing and Bioinformatic Analysis

ES was performed as previously described (18). Briefly, after alignment to human genome assembly hg19 with NovoAlign (http://www.novocraft.com), variants were jointly called using GATK² HaplotypeCaller (v3.3-0-g37228af), GATK UnifiedGenotyper (v3.1-1-g07a4bf8), and FreeBayes (v0.9.14-24-gc292036). Variants were retained if called by HaplotypeCaller and UnifiedGenotyper and/or FreeBayes within 5 base pairs of the target capture region, and the frequency was less than 10% within the dataset. Individual genotypes were retained if the HaplotypeCaller genotype quality score was 20 or more and an alternative allele read depth of less than 1. The average depth was 50x (Supplementary Figure 1, available online); 808 samples were whole-genome amplified (19). SNPWEIGHTS version 2.1 (20) was used to infer genetic ancestry for all CCSS cases and cancer-free controls; individuals with more than 80% European ancestry were considered European. Additional details are provided in the Supplementary Methods (available online).

Candidate Cancer-Susceptibility Gene List

We investigated 237 CSGs (hereafter, CSG-237) from 4 published studies (9,21-23) (Supplementary Table 1, available online). We focused our analyses on 172 autosomal dominant genes, including both autosomal dominant and autosomal recessive genes and those of unknown inheritance (hereafter, CSG-172), and a subset of 60 autosomal dominant genes with high-to-moderate penetrance (hereafter, CSG-60) previously reported (6,7).

Pathogenicity Classification

The analysis was restricted to variants with a minor allele frequency of less than 1% in cases or controls drawn from all ethnic subgroups reported in Exome Aggregation Consortium (excluding The Cancer Genome Atlas data) (24), 1000 Genomes Project (25), and the Exome Sequencing Project (26). Variants were categorized as pathogenic (P), likely pathogenic (LP), variant of uncertain significance, likely benign, or benign using a hierarchical classification system based on ClinVar, the Human Gene Mutation Database, and InterVar (Supplementary Figure 2, Supplementary Table 2, available online). Details are described in the Supplementary Methods (available online).

Statistical Methods

Rare-variant association tests were conducted for both case-case and case-control comparisons, based on the burden test statistics included in the optimal unified test R package. The case-case analyses were performed using all cases in 2 different ways: individual cancer type and groups of cancers. Details are described in the Supplementary Methods (available online). The case-control analyses were performed using European ancestry individuals only and used the individual cancer type and groups of cancers approaches but compared the P/LP frequencies to controls.

In a time-to-event analysis of overall survival, we compared cases harboring P/LP variants in the CSG-172 vs those without any P/LP variants. At-risk time for death started at the age at sample collection and ended either with age at death (all-cause mortality as the event of interest) or age at last follow-up (censoring). Cox proportional hazards regression models were used to estimate the hazard ratio (HR) of death and 95% confidence intervals (CIs) adjusting for individual cancer type, age at diagnosis, and treatments (anthracycline, cyclophosphamide, and any radiation within 5 years). The R package “survival” was used to fit the Cox models and generate Kaplan-Meier survival curves.

Statistical tests were performed with R version 3.3.2. To correct for multiple testing, false discovery rates were calculated using the P adjust function, and a Q value of less than 0.1 was considered statistically significant for the burden test and denoted as Q_burden.

Results

Summary Findings

We evaluated germline exonic variants in 5451 long-term pediatric cancer survivors within the CCSS cohort. The median age at diagnosis for primary cancers was 7.2 years (range = 0-21.0 years), and the median follow-up time since age at diagnosis was 32.6 years (range = 8.6-46.3 years) with a median age at follow-up of 41.0 years (range = 10.9-64.3 years). Cases were 48.2% male, 93.7% were of European ancestry, and 93.1% were alive at last follow-up (Table 1).

P/LP Variants in Autosomal Dominant CSGs

In 5105 pediatric cancer survivors with European ancestry, 211 (4.1%, 95% CI = 3.6% to 4.7%) harbored at least 1 P/LP variant in a high-to-moderate penetrance autosomal dominant CSG (CSG-60), which was statistically significantly more than the 1.3% (95% CI = 0.4% to 2.3%) observed in controls (P_Fisher = 3 x 10^-4) (Table 3). Out of 123 cases with P/LP variants, 7 cases carried a P/LP variant in more than 1 CSG (FH/NF1, FH/BRCA2, MSH2/BRCA2, VHL/PALB2, PMS2/BRCA1, BRAF/NF1, NF1/BRCA1). In analyses by cancer type, the highest frequency of P/LP variants was observed for patients with astrocytoma (7.4%, n = 27), followed by medulloblastoma (6.3%, n = 9) and osteosarcoma (6.3%, n = 18); the latter differed substantially from a recent survey of more than 1000 cases of osteosarcoma (720 unselected cases, 284 CCSS cases), where 11.1% of unselected cases had a P/LP variant in the CSG-60, suggesting that P/LP variants could be of poor prognosis in osteosarcoma (12). In the more expanded set of autosomal dominant CSGs (CSG-172), there were also statistically significantly more P/LP variants in cases (11.9%) compared with controls (7.7%) (P_Fisher = 2.2 × 10^-3), and the highest frequency of P/LP variants was found in leukemia other or not otherwise specified (NOS) (17.1%) followed by rhabdomyosarcoma (14.9%) and astrocytoma (14.5%) (Supplementary Figure 3, Supplementary Table 3, available online).

In pediatric cancer survivors carrying a germline P/LP variant in the CSG-172, the risk of all-cause mortality was elevated compared with those without a germline P/LP variant after adjusting for individual cancer type, age at diagnosis, and treatment (anthracycline, cyclophosphamide equivalent, and radiation within 5 years) (HR = 1.70, 95% CI = 1.27 to 2.28, P_{Cox regression} = 3 × 10^-4) (Supplementary Figure 4, available online).

Rare-Variant Association Tests

We used a case-control rare-variant association test approach to identify genes with P/LP variants associated with childhood cancer susceptibility for both groups of cancers and specific cancer types. There was a statistically significant excess (P_burden < .05) of P/LP variants in cases compared with controls in NF1 (CNS, STS) (28), WT1 (Wilms tumor) (29), TP53 (bone and STS) (30), EGFR (Hodgkin lymphoma), and PMS2 (NHL) (31) (Table 4). However, after correcting for multiple testing, only NF1 (CNS) and WT1 (Wilms tumor) remained statistically significant (Q_burden < 0.1). All of these genes are known to be associated with susceptibility to these specific cancer types and/or groups, except for EGFR to Hodgkin lymphoma.

We performed case-case rare-variant association tests for individual cancer types and confirmed known associations (28-31) between specific CSG-237 P/LP variants and susceptibility to individual cancer types and observed a novel association. There was a statistically significant excess (Q_burden < 0.1) of DHCR7 P/LP variants in rhabdomyosarcoma cases that has not been previously described (Table 5). In our case-case analyses by groups of cancer, we observed a statistically significant excess (Q_burden < 0.1) of P/LP variants in the following genes: PMS2 in NHL and SDHB, ERCC4 and FGFR3 in STS (Table 5). All findings remained statistically significant after adjusting for population stratification (using the top 15 principal components) except for TP53 association with bone (Supplementary Tables 4 and 5, available online).

Multiple Cancer Types Associated With Well-Known CSGs

Figure 1 depicts the landscape of P/LP variants in genes from the CSG-60. We observed P/LP variants in 40 of the 60 genes (67.7%). The 10 genes with the most P/LP variants were genes typically associated with germline variants in adult-onset cancers rather than pediatric cancers, including BRCA2, FH, BRCA1, PALB2, PMS2, and CDKN2A, even after correction for gene size (data not shown). P/LP variants in both PMS2 and MSH6 were observed in NHL, leukemia, and Hodgkin lymphoma, suggesting, if replicated, a previously unsuspected link of leukemia and lymphoma with Lynch syndrome. In addition, we observed cases with Wilms tumor and bone malignancies to harbor P/LP variants in PMS2 and a neuroblastoma with a P/LP variant in MSH6. Numerous patients with leukemia carried P/LP variants in genes associated with pheochromocytoma or paraganglioma (SDHA, SDHB, SDHD, MAX, TMEM127) and a P/LP variant in SDHC with a neuroblastoma. Both leukemia and neuroblastoma are phenotypes not known to be associated with these genes.

Figure 1.

Summary of the pathogenic/likely pathogenic (P/LP) variants in the CSG-60 by cancer type and patient demographics. Each column represents 1 case. The right y-axis represents the total number of P/LP variants found in each gene and the method for variant classification. The top (-axis summarizes the patient demographics by cancer type. CNS = central nervous system; CSG = cancer susceptibility gene; HL = Hodgkin lymphoma; NBL = neuroblastoma; NHL = non-Hodgkin lymphoma; STS = soft tissue sarcoma; Wilms = Wilms tumor.

Open in new tabDownload slide

Homozygous P/LP Variants in Autosomal Recessive CSGs

We identified 2 individuals who were homozygous carriers of P/LP autosomal recessive gene variants: 1 each in MUTYH and SBDS. The MUTYH homozygote carrier developed a Wilms tumor and harbored no other P/LP variant in a CSG (including WT1 and REST). The SBDS homozygote carrier developed a STS other or NOS and did not have P/LP variants in known STS-associated genes (including NF1, TP53, or DICER1).

Discussion

Among 5451 long-term survivors of European ancestry with a prior diagnosis of pediatric cancer, we identified an excess of P/LP variants in established CSGs compared with adult cancer-free controls. P/LP variants in specific CSGs and cancer types not previously recognized have been identified. Of the cases, 4% had a P/LP variant in a well-described subset of autosomal dominant CSGs (CSG-60) with high-to-moderate penetrance, and 11.9% had a P/LP variant in a more expansive list of autosomal dominant CSGs (CSG-172). This large pediatric cancer survivor population and comprehensive CSG evaluation enabled us to confirm and, importantly, to extend understanding of the phenotypic consequences of rare germline variants in CSGs and, at the same time, highlight pleiotropy in well-characterized CSGs.

A comparison of the frequency of P/LP variants reported in large published pediatric pan-cancer studies is daunting because of differences in the distribution of cancer types, pathogenicity scoring criteria, and ascertainment biases in collecting cases (eg, from high-risk cases referred to tertiary centers). The frequency of CSG-60 P/LP variants in our cohort was similar to the previously published long-term pediatric cancer survivor study, St. Jude Lifetime Cohort Study (approximately 4%), after restricting cancer types to include only cancers in CCSS (excluding retinoblastoma and adrenal cortical carcinoma because of their known associations with RB1 and TP53 P/LP variants, respectively) (6). The 2 collected series (7,9) of available pediatric cancers reported a prevalence of approximately 8%, which differs from our current observation in the CCSS; the lower prevalence in CCSS could be due to specific biologic features of cases with underlying P/LP that could result in poorer prognosis. The overall lower P/LP carrier rate in long-term survivor cohorts could be partly due to early mortality in highest-risk cases (some harboring a P/LP variant) prior to eligibility in a long-term survival study.

Previous studies showing adverse prognosis for P/LP variant carriers in certain genes (eg, TP53) have primarily focused on survival within 10 years from diagnosis (11). Our data suggest that P/LP germline variants could adversely impact long-term survival in excess of 10 years from diagnosis; this is particularly notable for TP53 P/LP overall as well as in osteosarcoma overall (12,32). Our study population has an introduced bias because of the inclusion of only long-term survivors who provided a DNA sample. Additional research is needed to assess potential confounding factors and evaluate disease-specific mortality. Nevertheless, if validated, these results could highlight the importance of assessing germline variant status even among long-term survivors and, perhaps, be useful for management of future survivorship (33).

In our case-control and case-case analyses, we observed well-established associations between CSG and specific cancer types, which provides confidence in our data and variant classification: NF1 and astrocytoma (34), STS (35); WT1 and Wilms tumor (29); PTPN11 and astrocytoma (36); REST and Wilms tumor (37); SUFU and medulloblastoma (38); and TP53 and rhabdomyosarcoma (39), STS (39), bone (40). These results are consistent with prior reports, such as NF1 and astrocytoma (41). In contrast, TP53 P/LP variants have been shown to be associated with poorer overall survival in unselected patients with osteosarcoma, and the rate of P/LP carriers observed in our osteosarcoma survivors is lower than previously reported rates in unselected cases for TP53 (1% vs 5%-10% of cases) (12,32).

We have extended the range of cancer subtypes that are associated with P/LP variants in CSGs and, if replicated, could be incorporated into future counseling for those with these P/LP variants. For example, we observed an increase in SDHB P/LP variants in children with STS, which is notable because pathogenic variation in SDHB has been reported for susceptibility to pheochromocytoma and paraganglioma as well as the Carney triad but not yet in pediatric sarcomas (42). We also observed a Wilms tumor in a child with homozygous pathogenic MUTYH variants, a gene associated with increased risk for polyposis and colorectal cancer (43), and the child had no other P/LP variants in our CSGs (including WT1 and REST). Last, a patient with a SBDS homozygote pathogenic variant developed STS other or NOS and did not have P/LP variants in NF1, TP53, or DICER1, which are known to be associated with STS. Although pathogenic SBDS variants that cause the autosomal recessive disorder Shwachman-Diamond syndrome are known to be associated with risk of myelodysplastic syndrome and/or AML, solid malignancies are not a common outcome (44,45). If replicated, these findings could have implications for the clinical care of people with SDHB, MUTYH, and SBDS pathogenic variants.

We further extended our analysis of candidate genes (CSG-237) to include those somatically mutated in pediatric cancer but not yet associated with germline susceptibility. In particular, we observed a statistically significant excess of loss-of-function P/LP variants in FGFR3 in children with STS, which is intriguing because somatic mutations in FGFR3 are frequently observed in bladder cancer (46). In addition, Muenke syndrome, a disorder of craniosynostosis, arises from germline gain-of-function variation in FGFR3, but to date, it is not known to be associated with germline cancer susceptibility (47). Our observation, if replicated, demonstrates: 1) an unexpected, putative genotype-phenotype correlation with FGFR3, 2) a novel germline phenotype in an otherwise well-known gene, and 3) a rare example of a somatically mutated gene also associated with germline susceptibility.

NF1 and BRCA2 had the highest frequency of pathogenic variants among the CSG-60, and TP53 was the seventh most common. Interestingly, we identified that 1.7% (93 of 5451) of cases had a P/LP variant in 1 of 6 genes (BRCA1/2, FH, PALB2, PMS2, and CDKN2A) typically associated with adult-onset tumors. Variation in these genes has not been well studied in pediatric cancer populations and highlights the unexpected pleiotropy of even well-characterized CSGs. Understanding penetrance in these genes also remains a challenge. For example, BRCA2 P/LP variants were observed in 0.5% of cases, but the carrier frequency was the same as in our controls (0.7%). Although BRCA2 P/LP variants have been reported in pediatric cancers (48,49), individual cases will require careful interpretation and perhaps orthogonal data (eg, tumor sequencing) to confirm causality.

The identification of a germline P/LP variant could result in a clinical intervention such as risk-reduction surgery, surveillance imaging, choice of therapeutic agent (50-52), and/or clinical genetic testing of asymptomatic family members (49,53,54). Such testing has historically centered around adult disease processes (eg, breast and colon cancer), and because it was performed in younger family members, it has been termed cascade testing. (53,54) Here, we identified P/LP variants in CSGs (eg, BRCA1/2, MEN1) linked to adult cancers. Observing variants in these genes in our pediatric cohort suggests a reexamination of the phenotypes associated with these genes and potentially the consideration of whether the parents (and other relatives) of young children with P/LP variants could benefit from “reverse-cascade testing.”

The strengths of our study included the large size of the case cohort and comparison of germline ES data from pediatric cancer survivors with controls that were jointly called using the same pipeline and quality control metrics. Although our controls were modest in number, we did not use publicly available controls for gene discovery because of the detection of statistically significant hyperinflation likely due to bioinformatic pipeline differences (Supplementary Figure 5, available online). However, we do see confirmation of our findings in these public databases. As a limitation, we may be underestimating the prevalence of damaging variants within the CSGs given our conservative, clinically based American College of Medical Genetics and Genomics and Association of Molecular Pathology classification criteria (55). Our study includes long-term pediatric cancer survivors, which limits generalizability.

In summary, we identified new associations that could be important to pediatric cancer susceptibility and, in turn, those who transition to survivorship with the risk of a second cancer (12,56,57). Our study includes long-term pediatric cancer survivors, which limits generalizability. Further studies are needed to validate and refine our observations, including functional investigation of the pathogenicity of novel variants in pediatric cancer. Characterization of pleiotropy is of central importance in cancer genetics to provide optimal genetic counseling and clinical surveillance. In pediatrics, germline genetic testing is frequently ordered in the context of children with phenotypes associated with a particular syndrome or a family history of cancer. Because of this ascertainment bias, the phenotypic cancer spectrum arising from pathogenic germline variants remains incomplete, even for iconic CSGs such as TP53. Our findings improve understanding of germline pediatric cancer genetic etiology and may lead to better pediatric cancer risk stratification at diagnosis, genetic counseling for patients and family members, and outcomes.

Funding

This work was supported by the Intramural Research Program of the Division of Cancer Epidemiology and Genetics of the National Cancer Institute, Bethesda, MD. We thank the CCSS participants and referring physicians for their valuable contributions. CCSS also is supported by the National Cancer Institute (CA55727, GT Armstrong, principal investigator) and St. Jude Children’s Research Hospital through the National Cancer Institute Cancer Center Support (CORE) grant (CA21765, C. Roberts, principal investigator) and the American Lebanese-Syrian Associated Charities (ALSAC).

Notes

Role of the funders: The funders had no role in the design of the study; the collection, analysis, and interpretation of the data; the writing of the manuscript; and the decision to submit the manuscript for publication.

Disclosures: DRS performs contract clinical telegenetics services for Genome Medical, Inc, in accordance with relevant National Cancer Institute ethics policies.

Prior presentation: This work was presented in part of the following meeting: Platform presentation at the 69th Annual Meeting of the American Society of Human Genetics, Houston, Texas, October 2019.

Disclaimer: The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products or organizations imply endorsement by the US government.

Acknowledgment: This work used the computational resources of the NIH High Performance Computing Biowulf cluster.

Author Contribution: Conceptualization: JK, MG, DMK, SAS, LMM, DRS, LM; Formal analysis: JK, MG, WL, BZ, LS, JNS; Resources: SWH, LM, LLR, YYGTA, SB, WML, RD, LT, NF, BDH, BZ, MW, KJ, AAH, CD; Data curation: JK, MG, DMK, MNF, MD, MY, SAB, KCA, FPF, AMG, PPK, MJM, MLM, MLN, LO, AP, MP, MR, KS, TW; Writing - original draft: JK, MG, DRS, LM; Writing - review & editing: all authors; Supervision: SAS, LMM, DRS, LM.

Data Availability

The data for this project is publicly available (dbGAP accession: phs002072.v1.p1), as per funder policies.

References

Author notes