Investigation of DNA Damage Response Genes Validates the Role of DNA Repair in Pediatric Cancer Risk and Identifies SMARCAL1 as a Novel Osteosarcoma Predisposition Gene

Abstract

PURPOSE

Recent studies reveal that 5%-18% of children with cancer harbor pathogenic variants in known cancer-predisposing genes. However, DNA damage repair (DDR) genes, which are frequently somatically altered in pediatric tumors, have not been systematically examined as a source of novel cancer-predisposing signals.

METHODS

To address this gap, we interrogated 189 DDR genes for presence of germline predisposing variants (PV) among 5,993 childhood cancer cases and 14,477 adult noncancer controls (discovery cohort). PV were determined using a tiered approach incorporating ClinVar annotations, InterVar classification, and in silico tools (REVEL, CADD, and MetaSVM). Using logistic and firth regression, we identified genes with PV statistically enriched in the germline of children with tumors and replicated findings among 1,497 additional childhood cancer cases across three independent cohorts.

RESULTS

Analysis across all cases with cancer revealed enrichment of TP53 PV. Cancer-specific analyses confirmed known associations including germline TP53 PV in adrenocortical carcinoma, high-grade glioma (HGG), and medulloblastoma (MB), PMS2 in HGG and non-Hodgkin lymphoma (NHL), MLH1 in HGG, BRCA2 in NHL, and BARD1 in neuroblastoma. In addition, four novel associations were uncovered, including BRCA1 in ependymoma, SPIDR in HGG, SMC5 in MB, and SMARCAL1 in osteosarcoma (OS). Importantly, the SMARCAL1:OS association was significant in the discovery (6/230, 2.6%, false discovery rate [FDR]_logistic = 0.0189) as well as all three replication cohorts (Childhood Cancer Survivor Study: 8/275, 2.9%; P_Fisher < .0001; Cancer Predisposition Syndrome-German Childhood Cancer Registry: 4/135, 3%, P_Fisher = .002; Individualized Therapy for Relapsed Malignancies in Childhood: 4/217, 1.8%, P_Fisher = .012). The remaining wild-type SMARCAL1 allele was deleted in three of four OS tumors with available data.

CONCLUSION

Our study confirms the relevance DDR genetic variation in pediatric cancer risk and establishes SMARCAL1 as a novel OS predisposing gene, providing insights into tumor biology and creating opportunities to optimize care for patients with this challenging tumor.

INTRODUCTION

Germline variants in cancer-predisposing genes (CPG) perturb cell growth and differentiation to set the stage for malignant transformation. Accordingly, the study of CPG and associated hereditary syndromes provides critical insights into normal and cancer biology. To this end, recent sequencing studies have revealed that up to 18% of children with cancer harbor an underlying genetic predisposition.^1-4 However, 40%-80% of children with cancer have family histories and/or clinical features concerning for cancer predisposition but lack a causal genetic diagnosis.^1,5 This observation suggests that additional CPG remain undiscovered and their association with cancer phenotypes further elucidated.

CONTEXT

• Key Objective

• To comprehensively assess the contribution of germline predisposing variants in DNA damage repair (DDR) genes to pediatric cancer risk, using a large case-control cohort and stringent variant interpretation framework.

• Knowledge Generated

• Through this investigation, we demonstrate the relevance of DDR pathway perturbations in childhood cancer risk by validating several known and discovering four novel DDR gene:cancer associations, including SMC5 in medulloblastoma, BRCA1 in ependymoma, SPIDR in high-grade glioma, and SMARCAL1 in osteosarcoma (OS). Among these, the SMARCAL1:OS association validated across three replication cohorts with evidence for biallelic inactivation in tumors, confirming SMARCAL1 as a novel OS predisposition gene.

• Relevance (S. Bhatia)

• These findings will inform future investigations aimed at developing targeted therapies as well as germline testing for prospective surveillance and early detection of OS.*

• *Relevance section written by JCO Associate Editor Smita Bhatia, MD, MPH, FASCO.

Previous investigations have primarily included children with specific tumor types (eg, high-risk solid or CNS tumors, relapsed cancers) and examined for pathogenic variants in known CPG. Nevertheless, expanding the scope of germline analyses to include children with a broader array of cancers and additional cancer-associated genes is crucial to identify the missing heritable factors underlying childhood tumor formation. The identification of novel CPG and predisposing variants (PV) is also central to improving the outcomes for affected children as it enables development of targeted cancer therapies, guides genetic counseling and testing of relatives, and informs cancer surveillance and risk reduction.^6,7

Somatic alterations affecting DNA damage repair (DDR) genes are drivers of high-grade pediatric tumors.² In addition, germline PV affecting selected DDR genes have been linked to several highly penetrant childhood cancer predisposition syndromes (CPS), including Li-Fraumeni syndrome, ataxia telangiectasia, Fanconi anemia, and replication repair deficiency.^2,8 Germline PV in DDR genes have also been implicated in the development of subsequent malignant neoplasms in long-term survivors of childhood cancer, especially those previously exposed to higher doses of ionizing radiation, anthracyclines, or alkylating agents.⁹ Despite these previous observations, to the best of our knowledge, an unbiased assessment of DDR genes and their role in development of primary cancers in children has not been conducted.

To this end, we generated a harmonized data set of germline variants from 5,993 childhood cancer cases and 14,477 adult non-cancer controls. We then conducted rare variant gene burden analysis using a curated set of 189 DDR genes with the aim to identify novel CPG that could account for the missing heritability of childhood cancer. Novel gene-cancer associations were replicated using three independent pediatric cancer cohorts and available tumor data.

METHODS

Patient Cohorts

The discovery cohort consisted of 5,993 children with cancer across five large scale sequencing studies, including the Pediatric Cancer Genome Project (PCGP),¹⁰ National Cancer Institute Therapeutically Applicable Research to Generate Effective Treatments initiative (NCI-TARGET, phs000218),¹¹ St Jude Lifetime Cohort Study (SJLIFE),¹² Genomes for Kids (G4K),³ and St Jude Real-Time Clinical Genomics study (RTCG; Fig 1A, Data Supplement, Table S1, online only). The control cohort for discovery comprised 14,477 adults without cancer from the 1000 Genomes Project¹³ and Alzheimer's Disease Sequencing Project (phs000572).¹⁴ For replication of novel CPG, we queried three independent pediatric cancer cohorts, including the Childhood Cancer Survivor Study (CCSS, phs001327),¹⁵ Individualized Therapy for Relapsed Malignancies in Childhood (INFORM),¹⁶ and the Cancer Predisposition Syndrome- German Childhood Cancer Registry (CPS-GCCR). The controlcohort for replication included adults without cancer from gnomAD v2.1.¹⁷ This study was approved by the Institutional Review Board at St Jude Children's Research Hospital (No. 20-0379) and informed consent was obtained from parents, guardians, or patients, as appropriate.

FIG 1.

Germline DDR gene variants across pediatric cancers. (A) Study design and workflow. The discovery cohort comprised 5,993 patients with pediatric cancer from the PCGP, G4K, RTCG, SJLIFE, and TARGET cohorts, with 14,477 controls from the 1000 Genomes Project and ADSP. Germline variants were called using GATK joint genotyping (N = 20,470) and underwent quality control and post hoc filtering along with genetic ancestry and sex determination. Variant filtering focused on 189 DNA repair genes, selecting rare PV (allele frequency <0.05%) predicted as P/LP or damaging. Replication analysis included three independent cohorts with matched cancer types from significant associations: CCSS, CPS-GCCR, and INFORM. (B) Number of jointly called germline whole-exome sequencing samples across 22 pediatric cancers in the discovery cohort. (C) Variant filtering pipeline. Among 5,993 pediatric cancer samples, 6,636,054 variants passed quality control. Filtering by predicted functional impact (frameshift, nonsense, missense, and splicing) yielded 2,536,772 variants, of which 1,403,469 were rare (gnomAD v2.1 AF <0.05%). Restriction to DDR genes identified 10,368 variants, with 1,881 meeting predisposing filtering criteria. (D) The prevalence of DDR PV across 22 pediatric cancers is shown as the fraction of samples in each cancer type harboring a PV in any of the 189 DDR genes. Median prevalence of DDR PV across the three cancer categories is shown with a dashed line. (E) Gene-level burden of pathogenic germline variants. All associations with FDR_logistic <0.25 are shown, whereas those that met significance threshold FDR_logistic <0.05 are highlighted with black border. Color intensity represents log10[FDR_logistic]. Number of variants for respective cancer:gene pairs are indicated. ACC, adrenocortical carcinoma; ACPG, adamantinomatous type craniopharyngioma; ADSP, Alzheimer's Disease Sequencing Project; AF, allele frequency; ALL-NOS, acute lymphoblastic leukemia- not otherwise specified; AML, acute myeloid leukemia; ATRT, atypical teratoid/rhabdoid tumor; BALL, B-cell ALL; CCSS, Childhood Cancer Survivorship Study; DDR, DNA damage repair; EPD, ependymoma; EWS, Ewing sarcoma; FDR, false discovery rate; G4K, Genomes4Kids; CPS-GCCR, Cancer Predisposition Syndrome-German Childhood Cancer Registry; GCT, germ cell tumor; HGG, high-grade glioma; HL, Hodgkin lymphoma; INFORM, Individualized Therapy for Relapsed Malignancies in Childhood; LGG, low-grade glioma; LIC, liver cancer; MB, medulloblastoma; MEL, melanoma; NBL, neuroblastoma; NHL, non-Hodgkin lymphoma; OS, osteosarcoma; P/LP, pathogenic/likely pathogenic; PCGP, Pediatric Cancer Genome Project; PV, predisposing variants; QC, quality control; RB, retinoblastoma; RMS, rhabdomyosarcoma; RTCG, St Jude Real-time Clinical Genomics; SJLIFE, St Jude Lifetime Cohort; TALL, T-cell acute lymphoblastic leukemia; WLM, Wilms tumor.

Variant Calling and Filtering for the Discovery Cohort

Germline variants in 189 DDR genes were identified through joint genotyping of whole exome sequencing (WES) data from cases and controls and filtered for rarity (minor allele frequency [AF] <0.05% in gnomAD v2.1 noncancer subset; Supplementary Methods and Data Supplement, Table S2).^8,9,17 To identify PV, a tiered filtering strategy was employed using a combination of American College of Medical Genetics and Genomics and the Association for Molecular Pathology guidelines as per ClinVar pathogenic/likely pathogenic (P/LP) database (2025-06-23), InterVar automated classification tool (P/LP), and in silico predictions (REVEL >0.7, CADD >20, MetaSVM damaging; Supplementary Methods and Data Supplement, Fig S1).^18,19 For replication analysis, an identical filtering strategy for germline PV was applied across replication cohorts (CCSS, INFORM, and CPS-GCCR) and adult noncancer controls (gnomAD v2.1).

Statistical Analysis

We performed a gene-based burden analysis for germline PV in pediatric cancer cohort versus noncancer controls using logistic regression (false discovery rate [FDR]_logistic) and Firth regression (FDR_Firth; Supplementary Methods). For replication analysis, Fisher exact test (P_Fisher) was used to calculate enrichment of germline PV in cases versus controls. Comparison of age at cancer diagnosis between DDR germline variant carriers and noncarriers was completed as described in the Supplementary Methods.

RESULTS

Germline Variants in DDR Genes Across Tumor Types

The discovery cohort included 5,993 children and adolescents with 22 cancer subtypes, classified as hematologic, solid, and CNS tumors (Figs 1A and 1B). The median age at cancer diagnosis was 6 years (range, 4 days to 32 years) with most cases (67%) of European ancestry (Data Supplement, Table S3). Across cancers, 6,636,054 germline variants were analyzed to retain 1,881 rare, PV among 189 DDR genes (Fig 1C, Data Supplement, Table S4). Frequency of PV was similar across hematologic (27.6% [range, 24.8%-31.4%]), solid (29.3% [18.8%-59.3%]), and CNS cancers (27.2% [23.3%-34.5%]; P = .9, Kruskal-Wallis test) but varied across cancer subtypes, such as rhabdomyosarcoma (19%) and adrenocortical carcinoma (ACC, 59%; Fig 1D).

Confirmation of Known Associations of Germline DDR PV With Childhood Cancers

Overall, 1,561 of 5,993 (26%) childhood cancer cases harbored PV in one or more DDR genes. We observed a significant enrichment of PV in TP53 compared with jointly called adult noncancer controls (n = 14,477; 37/5,993, 0.6%, FDR_logistic = 0.0013; odds ratio [OR], 3.2 [95% CI, 1.9 to 5.4]), supporting its critical role in maintaining genome stability (Data Supplement, Table S5). We next performed rare variant burden analysis to identify associations of DDR genes across children with pediatric cancers and observed several previously described associations, which serve as an internal validation of our analytic pipeline. For example, TP53 PV were enriched in ACC (12/27, 44%, FDR_logistic < 0.0001; OR, 426.7 [95% CI, 182.1 to 999.5]), high-grade glioma (HGG; 5/206, 2.4%, FDR_logistic = 0.0011; OR, 12.5 [95% CI, 4.4 to 29.9]), and medulloblastoma (MB; 4/257, 1.6%, FDR_logistic = 0.0011; OR, 12.5 [95% CI, 4.4 to 29.9]; Figs 1E and 2A; Table 1; Data Supplement, Tables S5 and S6).^2,20 Furthermore, we confirmed biallelic inactivation of TP53 in all corresponding tumors examined (Data Supplement, Fig S2).

FIG 2.

Predisposing DDR gene variants and significant gene:cancer associations. (A) Distribution of PV across the affected proteins in statistically enriched tumor types. Respective protein domains are shown. Variants are categorized by type: frameshift (red), nonsense (orange), splice-site (purple), and missense (blue). ^aDUF, domain of unknown function. (B) Somatic DNA mutational signature analysis across 14 cases was performed using signature.tools.db. Bar plots display the proportion of COSMIC (v3) SBS mutational signatures. Proportion of mutations with unassigned signatures are also shown. (C) Tumor RNAseq of the SMC5:c.380+1G>C splicing variant. The top panel (blue) represents normal splicing in another MB case without SMC5 alteration, while the bottom panel (red) shows aberrant splicing in the medulloblastoma from the SMC5:c.380+1G>C germline variant carrier. The variant leads to exon skipping, as indicated by disrupted exon-exon junctions and altered splicing events. The read distribution demonstrates a substantial loss of normal exon inclusion, suggesting a pathogenic impact of the mutation on transcript integrity. ACC, adrenocortical carcinoma; DDR, DNA damage repair; EPD, ependymoma; HGG, high-grade glioma; MB, medulloblastoma; MMR, mismatch repair; NBL, neuroblastoma; NHL, non-Hodgkin lymphoma; PV, predisposing variants; SBS, single-base substitution.

TABLE 1.

Enrichment of Known Associations in Pediatric Cancers From Discovery Cohorts

Discovery Analysis: Known Associations
Gene	Cancer	Frequency in Cases	Frequency in Controls	Cancer Risk
				OR (95% CI)	P logistic	FDRlogistic
TP53	ACC	12 in 27 (44.44%)	27 in 14,477 (0.19%)	426.7 (182.1 to 999.5)	1.35E-43	4.86E-40
TP53	HGG	5 in 206 (2.43%)	27 in 14,477 (0.19%)	12.5 (4.4 to 29.9)	8.85E-07	0.001
PMS2	HGG	4 in 206 (1.94%)	39 in 14,477 (0.27%)	10.1 (3.2 to 24.9)	4.03E-05	0.017
MLH1	HGG	4 in 206 (1.94%)	42 in 14,477 (0.29%)	9.6 (3.1 to 23.6)	5.31E-05	0.019
BARD1	NBL	6 in 485 (1.24%)	31 in 14,477 (0.21%)	6.3 (2.4 to 13.9)	8.80E-05	0.023
PMS2	NHL	4 in 239 (1.67%)	40 in 14,477 (0.28%)	8.1 (2.6 to 20)	2.03E-04	0.037
TP53	MB	4 in 257 (1.56%)	28 in 14,477 (0.19%)	8.2 (2.6 to 20.7)	2.47E-04	0.040
BRCA2	NHL	5 in 239 (2.09%)	51 in 14,477 (0.35%)	6.2 (2.2 to 14)	2.62E-04	0.041

Abbreviations: ACC, adrenocortical carcinoma; FDR, false discovery rate; HGG, high-grade glioma; MB, medulloblastoma; NBL, neuroblastoma; NHL, non-Hodgkin lymphoma; OR, odds ratio.

Enrichment of PV was also observed in mismatch repair (MMR) genes, including PMS2 (4/206, 1.9%, FDR_logistic = 0.017; OR, 10.1 [3.2 to 24.9]) and MLH1 (4/206, 1.9%, FDR_logistic = 0.019; OR = 9.6 [3.0 to 23.5]) among HGG, as well as PMS2 in non-Hodgkin lymphoma (NHL; 4/239, 1.7%; FDR_logistic = 0.0366; OR = 8.1 [2.6 to 20.0]).^21,22 Analysis of tumor data revealed that six of the 10 MMR PV carriers (5 HGG and 1 NHL) exhibited an ultra/hypermutator phenotype and somatic mutational signatures consistent with defective MMR (SBS14, SBS15, and SBS26; Fig 2B, Supplementary Table S6), whereas four cases did not exhibit defective MMR and tumor data were unavailable for two. Of the six ultra/hypermutator cases, five had constitutional MMR deficiency and HGG, while the sixth case was a PMS2 carrier with NHL, whose tumor exhibited a second PMS2 hit. Germline PV in BARD1, a recently identified neuroblastoma (NBL) predisposition gene, were also identified (6/485, 1.2%; FDR_logistic = 0.0227; OR, 6.3 [95% CI, 2.4 to 13.9]; Fig 2A).²³ Two of the six BARD1 carriers with available tumor data exhibited somatic mutational signature previously observed in NBL (SBS18; Fig 2B).²⁴ Finally, we observed enrichment for BRCA2 PV in NHL (5/239, 2.1%, FDR_logistic = 0.0411; OR, 6.2 [95% CI, 2.2 to 14.0]; Fig 2A), an association previously described in long-term survivors of childhood cancer.²⁵

Novel Associations of Germline DDR PV With Childhood Cancers

In addition to known associations, we identified four novel associations, including BRCA1 in ependymoma (EPD), SPIDR in HGG, SMC5 in MB, and SMARCAL1 in osteosarcoma (OS; Table 2, Data Supplement, Table S5). To this end, three cases of EPD carried BRCA1 PV (3/146, 2.1%; FDR_logistic = 0.0366; OR, 11.6 [95% CI, 3.1 to 31.6]). All three missense variants were within the BRCT and BRCT-associated domains (Fig 2A). Tumor WES showed no second hits in any of these tumors, and RNA expression of BRCA1 in carriers was comparable with EPD lacking germline CPG PV (Data Supplement, Fig S3). Next, heterozygous truncating variants in SPIDR, a gene that encodes a scaffold protein involved in homologous recombination, were identified in two HGG (2/206, 1%; FDR_logistic = 0.0366; OR, 25.2 [95% CI, 4.5 to 101.1]; Fig 2A). Analysis of tumor WES and RNAseq from the SPIDR:p.Y383* mutated case revealed two somatic TP53 mutations and reduced SPIDR RNA expression (Data Supplement, Figs S2 and S3). No tumor whole genome sequencing (WGS) data were available to determine whether there was loss of heterozygosity (LOH) at the SPIDR locus. Next, germline PV in SMC5, the gene encoding Structural Maintenance of Chromosome 5, were enriched in MB (4/257, 1.6%; FDR_logistic = 0.0005; OR, 21.6 [95% CI, 6.4 to 60.1]; Fig 2A). SMC5 is important for chromosome maintenance with biallelic germline alterations associated with the rare neurodevelopmental disorder, Atelis syndrome-2.²⁶ All four germline SMC5 mutant cases were of the group 3 MB subtype and the SMC5:MB association was even stronger when only group 3/4 MB were included in the analysis (4/88 cases, 4.5%; FDR_logistic < 0.0001; OR, 54.6 [95% CI, 16.0 to 156.0]; Data Supplement, Table S5). Tumor RNAseq and WES data were available for all four SMC5 germline variant carriers, whereas WGS was available for three. The MB tumor associated with the germline c.380+1G>C alteration showed exon 3 skipping in approximately 50% of SMC5 transcripts, resulting in an out-of-frame transcript accompanied with reduced SMC5 expression (Fig 2C, Data Supplement, Fig S3). The p.N940Y germline carrier exhibited a second somatic hit in the tumor, SMC5:p.Q357K, but we could not establish allelic configuration of the two variants. This tumor also exhibited SBS8, a DNA mutational signature associated with late replication errors (Fig 2C), suggesting altered SMC5 function during mitotic progression.²⁷

TABLE 2.

Discovery of Putative Novel Associations in Pediatric Cancers From Discovery Cohorts

Discovery Analysis: Novel Associations
Gene	Cancer	Frequency in Cases	Frequency in Controls	Cancer Risk
				OR (95% CI)	P logistic	FDRlogistic
SMC5	MB	4 in 257 (1.56%)	12 in 14,477 (0.08%)	21.6 (6.4 to 60.1)	2.70E-07	4.87E-04
SMARCAL1	OS	6 in 230 (2.61%)	59 in 14,477 (0.41%)	6.3 (2.5 to 13.6)	6.54E-05	0.019
BRCA1	EPD	3 in 146 (2.05%)	35 in 14,477 (0.24%)	11.6 (3.1 to 31.6)	1.77E-04	0.037
SPIDR	HGG	2 in 206 (0.97%)	6 in 14,477 (0.04%)	25.2 (4.5 to 101.1)	1.91E-04	0.037

Abbreviations: EPD, ependymoma; FDR, false discovery rate; HGG, high-grade glioma; MB, medulloblastoma; OR, odds ratio; OS, osteosarcoma.

Finally, SMARCAL1 PV were enriched within OS cases (6/230, 2.6%; FDR_logistic = 0.0189; OR, 6.3 [95% CI, 2.5 to 13.6]; Fig 3A, Data Supplement, Tables S5 and S6). SMARCAL1 encodes the SNF2-related chromatin remodeling annealing helicase, which resolves stalled replication forks.²⁸ Germline biallelic SMARCAL1 inactivation causes Schimke immuno-osseous dysplasia (SIOD) typified by skeletal, renal, and hematologic clinical abnormalities. We identified four protein-truncating variants (p.R114Qfs*4, p.L139Efs*3, p.L397Rfs*40, and p.Q653*) and two missense variants (p.R820H and p.R490C) located in the helicase domain (Figs 3B and 3C, Data Supplement, Table S6). Considering that OS is commonly observed in Li-Fraumeni syndrome, we confirmed the absence of germline TP53 mutations in the six germline SMARCAL1-mutated cases. Matched tumor WES and RNAseq data were available for two cases, which did not reveal a somatic second hit in SMARCAL1. The germline p.R114Qfs*4-mutated OS harbored a somatic ATRX:p.P717Hfs*4 mutation. Tumor RNAseq from this case showed unaffected SMARCAL1 expression, while the germline p.Q653*-mutated case demonstrated reduced SMARCAL1 RNA expression (Data Supplement, Fig S3). However, additional genomic alterations associated with this reduced SMARCAL1 expression could not be ascertained in the absence of tumor WGS data.

FIG 3.

Characterization of SMARCAL1 PV. (A) Schematic representation of SMARCAL1 (NM_014140) protein with PV from the discovery (top) and replication cohorts (bottom). The protein domains are as shown: RBD: replication protein A (RPA) binding domain (green); HARP: HepA-related protein domain (blue); and helicase domain (red). Variants are categorized by mutation type with frameshift (red), nonsense (orange), splice-site (purple), and missense variants (blue). (B) Oncoplot depicting SMARCAL1 PV in discovery and replication cohorts. Data include pathogenicity predictions based on ClinVar classification, CADD and REVEL scores, and gnomADv4.1 allele frequencies. Additional pathogenic germline variants in SJCPG60 genes or somatic driver events are shown along with annotations for biallelic status (\), tumor relapse status, and tumor availability. (C) SMARCAL1 protein structure predicted by AlphaFold. Missense variants in discovery (green) and replication (purple) cohorts are shown. Protein domains HARP: HepA-related protein (blue) and helicase (red) are shown. (D) Proposed model of SMARCAL1-mediated OS predisposition. SMARCAL1 is important for accurate DNA replication and repair to reinforce genome integrity. Approximately 2.6% of OS cases carry a predisposing variant in SMARCAL1, negatively affecting SMARCAL1 function and exacerbating genome instability with tumor acquisition of somatic second hits in genes known to permit ALT (SMARCAL1 LOH, ATRX inactivation), resulting in OS development. ^aVariants predicted as LP by AlphaMissense. AF, allele frequency; ALT, alternative lengthening of telomeres; CCSS, Childhood Cancer Survivorship Study; CPS-GCCR, Cancer Predisposition Syndrome-German Childhood Cancer Registry; INFORM, Individualized Therapy for Relapsed Malignancies in Childhood; LOH, loss-of-heterozygosity; NA, not available; OS, osteosarcoma; P/LP, pathogenic/likely pathogenic; PV, predisposing variants; VUS, variants of uncertain significance.

Clinical Features Associated With Germline DDR Gene Variants

Patients with germline CPG PV are often younger at tumor onset than individuals with sporadic cancers. Therefore, we analyzed the ages of cancer onset in DDR PV carriers versus noncarriers and identified only one association with a significant difference, namely, a younger age of onset of TP53 carriers with ACC (Data Supplement, Table S7). Overall, family history was available for 37 of 59 PV patients and was positive for 14 of these cases with a mix of related and unrelated malignancies (Data Supplement, Table S6). For four OS cases with germline SMARCAL1 variants, we did not identify any renal, skeletal, or immunodeficiency abnormalities typically seen in SIOD; however, in two cases, we observed simple renal cysts (Bosniak I/II).

Replication of SMARCAL1 as a Novel OS Predisposition Gene

To replicate the four novel associations, we queried three additional pediatric cancer cohorts (CCSS, CPS-GCCR, and INFORM). Two associations reached significance in only one of the three cohorts: BRCA1:EPD in INFORM (2/143; 1.4%; P_Fisher = .018; OR, 10.02 [95% CI, 1.2 to 37.1]) and SMC5:MB in CPS-GCCR (1/31; 3.2%; P_Fisher = .028; OR, 36.6 [95% CI, 0.9 to 215.2]; Data Supplement, Table S8). Importantly, enrichment of SMARCAL1 in OS was significant across all three cohorts: CCSS (8/275 cases; 2.9%; P_Fisher < .0001; OR, 7.4 [95% CI, 3.1 to 14.7]); CPS-GCCR (4/135 cases; 3%; P_Fisher = .002; OR, 7.5 [95% CI, 2.0 to 19.6]); and INFORM (4/217 cases; 1.8%; P_Fisher = .012; OR, 4.6 [95% CI, 1.3 to 12.0]). In total, 12 unique germline SMARCAL1 PV were identified among 16 individuals, of which two were also observed in the discovery cohort (p.L139Efs*3 and p.L397Rfs*40). Among the 10 remaining variants, three were protein-truncating (p.F941Lfs*31, p.R563*, and p.E848*), three were canonical splice-site (c.863-2A>G, c.1335-2A>T, and c.2070+2dup) and four were missense (p.F801V, p.A838T, p.G857R, and p.F279S) variants located in functionally relevant domains (Fig 3).²⁸ Germline TP53 variants were not found among any SMARCAL1 variant carriers; however, two cases harbored PV in NF2 and MLH1 (Fig 3B).²⁹ In sum, the prevalence of germline PV in OS cases was 16/627 or 2.6% (P_Fisher < .0001, OR, 6.4 [95% CI, 3.6 to 10.6]), which is similar to the 2.6% prevalence in the discovery cohort (Figs 3A and 3B; Data Supplement, Tables S8 and S9).

Tumor WGS data were available for all four INFORM cases. We observed SMARCAL1 LOH in three relapsed OS tumors because of copy-number deletion (Data Supplement, Fig S4). Three of four tumors also had RNAseq data available, two of which (c.1335-2A>T and p.E848*) exhibited low SMARCAL1 expression (Data Supplement, Fig S5). Loss of SMARCAL1 has been associated with alternative lengthening of telomeres (ALT).²⁸ To this end, we observed all four OS tumors in the replication cohort to be ALT-positive as determined by TelomereHunter (Data Supplement).³⁰ In addition, we assessed tumor genomic data in these four cases for alterations in ALT-associated genes ATRX, DAXX, and H3F3A, and did not observe somatic or germline mutations (N = 4 WES) or reduction in RNA expression (N = 3 RNAseq) in these genes (Fig 3B, Data Supplement, Fig S5). These data suggest that loss of SMARCAL1 function may induce ALT, a potential mechanism for promoting OS.

DISCUSSION

Highly penetrant CPS result from germline variations in DDR genes. To date, comprehensive studies investigating the germline landscape of DDR alterations in pediatric cancers have been lacking, in part because of limited case-control designs that harmonize batch effects from library preparation, sequencing platforms, and variant calling pipelines necessary for identifying novel CPG. We remapped raw sequencing data and performed joint genotype calling across 5,993 pediatric cancer cases and 14,477 adult noncancer controls, followed by stringent filtering for rare, protein-damaging variants. Through this approach, we established a 26% prevalence of putative damaging variants in DDR genes across childhood cancers. Statistical testing confirmed known associations, including germline TP53 in ACC, HGG, and MB, PMS2 in HGG and NHL, MLH1 in HGG, BARD1 in NBL, and BRCA2 in NHL. We also discovered four novel associations, including BRCA1 in EPD, SPIDR in HGG, SMC5 in MB, and SMARCAL1 in OS. Notably, among these, we confirmed statistical enrichment of germline SMARCAL1 PV across all three of our replication cohorts, providing compelling evidence for its role as a novel OS predisposition gene.

Germline SMARCAL1 variants were found in 2.6% of all OS cases, with two thirds predicted to cause haploinsufficiency because of protein truncation and one third composed of putative damaging missense variants. Ballinger et al previously reported germline truncating SMARCAL1 variants in 19 sarcoma cases including two OS, while Akhavanfard et al identified SMARCAL1 in three OS cases from the SJLIFE cohort.^31,32 Ballinger et al also reported LOH in five of 19 sarcomas with germline SMARCAL1 variants.³¹ Importantly, a recent study identified 1.8% of osteosarcoma cases to harbor germline SMARCAL1 variants, strengthening our finding.³⁸ Two cases of OS have been reported in individuals with SIOD, suggesting that OS is associated with SMARCAL1 dysfunction.³³ Genotype-level comparison of germline SIOD and OS PV reveals overlap (Data Supplement, Fig S6); however, the rarity and short lifespan (median age of death: 11 years) of individuals with SIOD preclude us from accurately assessing the prevalence and penetrance of OS in this context. No other germline alterations in OS-relevant CPG were identified in our cases, which supports SMARCAL1 as an independent risk factor for OS.³⁴ Seven of 13 OS cases (54%) with available clinical information experienced disease relapse, suggesting that SMARCAL1-associated cases may be more aggressive. Nevertheless, longitudinal studies are needed to assess the penetrance and clinical features of OS associated with germline SMARCAL1 variation. Somatic SMARCAL1 deficiency in glioblastoma causes ALT-mediated telomere synthesis.^28,35 Similarly, ALT appeared active in four tumors, three of them with biallelic SMARCAL1 alterations. A fifth tumor harbored an inactivating ATRX mutation, which may contribute to ALT.³⁶ Altogether, these data suggest that SMARCAL1 is a tumor suppressor, whereby loss of SMARCAL1 protein function impairs DNA replication and repair, leading to acquisition of an ALT phenotype or somatic mutations in ALT-permissive genes, and ultimately resulting in OS formation (Fig 3D).

The following limitations should be considered when interpreting the results of this study. We were unable to validate the three additional novel gene:cancer associations identified in this study across each of our replication cohorts. This finding may be due to the low sample sizes for specific cancer types in the replication cohorts. To achieve adequate statistical power, we combined high-risk primary cancers with presumably lower-risk cancers from adult survivors of childhood cancer, which may have diluted genetic signals and added survival bias (Data Supplement, Figs S7 and S8). The variant filtering criteria were rigorous, and it is possible that additional clinically relevant germline variants with less stringent in silico scores remain undescribed. To evaluate this possibility for SMARCAL1 in OS, we relaxed the missense filtering threshold to REVEL >0.5 across the discovery and replication cohorts and identified two additional germline SMARCAL1 variants (p.D424V and p.K27E) in three unrelated cases with relapsed OS from the INFORM cohort (data not shown). Matched tumor data from two of three cases with D424V revealed SMARCAL1 LOH; however, we have not included these cases in the current study to maintain consistency across our discovery and replication analyses. Given the sparse tumor data available, expanded tumor sequencing is needed to validate SMARCAL1 biallelic inactivation patterns observed herein. Finally, we could not establish the mode of inheritance or cosegregation with other cancers for many of the identified variants because of lack of familial testing. Future efforts examining the relatives of germline SMARCAL1 PV carriers are needed as this information may serve to strengthen existing evidence. Ultimately, the role of germline SMARCAL1 variation in OS tumorigenesis requires further investigation.

In summary, we demonstrate the importance of DDR pathway perturbations in predisposition to childhood cancer by validating known and discovering novel DDR gene:cancer associations. Our finding that germline damaging variants in SMARCAL1 predispose to OS serves as a foundation for future studies aimed at developing novel therapies for this aggressive cancer, one for which there have been little advances in treatment over the past four decades.³⁷ Similarly, genetic testing for germline SMARCAL1 PV will enable prospective surveillance of germline carriers to detect and treat incipient OS tumors at their earliest and most curable stages.

PREPRINT VERSION

This study was posted on medRxiv preprint server on May 13, 2025 (https://www.medrxiv.org/content/10.1101/2025.05.12.25325832v2).

DATA SHARING STATEMENT

A data sharing statement provided by the authors is available with this article at DOI https://doi.org/10.1200/JCO-25-01114.

The processed genomic data generated in this study are provided in the Supplementary Tables. Controlled-access raw genomic data can be requested via St Jude Cloud at https://platform.stjude.cloud/. The Childhood Cancer Survivor Study is a US National Cancer Institute–funded resource (U24 CA55727) to promote and facilitate research among long-term survivors of cancer diagnosed during childhood and adolescence. CCSS data are publicly available on the St Jude Survivorship Portal within the St Jude Cloud at https://survivorship.stjude.cloud/. In addition, use of the CCSS data that leverage the expertise of CCSS Statistical and Survivorship research and resources will be considered on a case-by case basis. For this use, a research Application of Intent followed by an Analysis Concept Proposal must be submitted for evaluation by the CCSS Publications Committee. Users interested in accessing this resource are encouraged to visit http://ccss.stjude.org. Full analytical data sets associated with CCSS publications since January 2023 are available on the St Jude Survivorship Portal at https://viz.stjude.cloud/community/cancer-survivorship-community∼4/publications. Any additional data are available upon request from the corresponding author.

AUTHOR CONTRIBUTIONS

Conception and design: Ninad Oak, Wenan Chen, Gang Wu, Kim E. Nichols, Richa Sharma

Financial support: Kim E. Nichols, Zhaoming Wang, Kirsten K. Ness, Greg T. Armstrong, Melissa M. Hudson, Stefan M. Pfister

Administrative support: Ninad Oak, Lynn Harrison, Gang Wu, Kim E. Nichols, Richa Sharma

Provision of study materials or patients: Ninad Oak, Kendra Maass, Judith Penkert, Kristian W. Pajtler, Olaf Witt, Uta Dirksen, Stefan M. Pfister, Greg T. Armstrong, Melissa M. Hudson, Kim E. Nichols

Collection and assembly of data: Ninad Oak, Wenan Chen, Alise Blake, Lynn Harrison, Martha O'Brien, Kendra Maass, Steffen Hirsch, Judith Penkert, Barbara C. Jones, Michaela Nathrath, Kristian W. Pajtler, David T.W. Jones, Olaf Witt, Uta Dirksen, Stefan M. Pfister, Christian Kratz, Zhaoming Wang, Greg T. Armstrong, Melissa M. Hudson, Gang Wu, Robert J. Autry, Kim E. Nichols, Richa Sharma

Data analysis and interpretation: Ninad Oak, Wenan Chen, Alise Blake, Martha O'Brien, Christopher Previti, Gnanaprakash Balasubramanian, Steffen Hirsch, Kathrin Schramm, Olaf Witt, Jiaming Li, Yadav Sapkota, Kirsten K. Ness, Lillian M. Guenther, Zhaoming Wang, Greg T. Armstrong, Melissa M. Hudson, Gang Wu, Robert J. Autry, Kim E. Nichols, Richa Sharma

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

Investigation of DNA Damage Response Genes Validates the Role of DNA Repair in Pediatric Cancer Risk and Identifies SMARCAL1 as a Novel Osteosarcoma Predisposition Gene

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/authors/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).