Nationwide germline whole genome sequencing of 198 consecutive pediatric cancer patients reveals a high incidence of cancer prone syndromes

PURPOSE: Historically, cancer predisposition syndromes (CPSs) were rarely established for children with cancer. This nationwide, population-based study investigated how frequently children with cancer had or were likely to have a CPS. METHODS: Children (0–17 years) in Denmark with newly diagnosed cancer were invited to participate in whole-genome sequencing of germline DNA. Suspicion of CPS was assessed according to Jongmans’/McGill Interactive Pediatric OncoGenetic Guidelines (MIPOGG) criteria and familial cancer diagnoses were verified using population-based registries. RESULTS: 198 of 235 (84.3%) eligible patients participated, of whom 94/198 (47.5%) carried pathogenic variants (PVs) in a CPS gene or had clinical features indicating CPS. Twenty-nine of 198 (14.6%) patients harbored a CPS, of whom 21/198 (10.6%) harbored a childhood-onset and 9/198 (4.5%) an adult-onset CPS. In addition, 23/198 (11.6%) patients carried a PV associated with biallelic CPS. Seven of the 54 (12.9%) patients carried two or more variants in different CPS genes. Seventy of 198 (35.4%) patients fulfilled the Jongmans’ and/or MIPOGG criteria indicating an underlying CPS, including two of the 9 (22.2%) patients with an adult-onset CPS versus 18 of the 21 (85.7%) patients with a childhood-onset CPS (p = 0.0022), eight of the additional 23 (34.8%) patients with a heterozygous PV associated with biallelic CPS, and 42 patients without PVs. Children with a central nervous system (CNS) tumor had family members with CNS tumors more frequently than patients with other cancers (11/44, p = 0.04), but 42 of 44 (95.5%) cases did not have a PV in a CPS gene. CONCLUSION: These results demonstrate the value of systematically screening pediatric cancer patients for CPSs and indicate that a higher proportion of childhood cancers may be linked to predisposing germline variants than previously supposed.


Introduction
In Europe, 15,000 children (1 in 300) are diagnosed with cancer each year. [1] Cancer can be attributed to genetic predisposition, exposure to carcinogens, and/or random mutations during cell division. Children are exposed to fewer carcinogens than adults. [2,3] Therefore, genetic predisposition and randomly acquired mutations are the major causes of most childhood cancers.
Cancer predisposition syndromes (CPSs) were previously considered rare among pediatric cancer patients, but increasing use of whole-exome sequencing (WES) and whole-genome sequencing (WGS) have identified up to 10% CPS among children, including several cases of CPS for adult-onset cancers not previously associated with childhood CPS. However, most studies investigated selected or single institution cohorts and included patients with specific diagnoses that were frequently associated with CPS. [4][5][6] Although some studies have included a broader range of pediatric cancer patients, [7][8][9][10][11] there have currently been no nationwide population-based studies. Moreover, most studies have focused on single nucleotide variants (SNVs) and few have included the effects of copy number variants (CNVs). [12] Many clinical criteria have been developed to identify patients with CPS, [13][14][15][16][17][18] but these have not been validated in a national cohort.
Here we present the genetic SNV and CNV findings from the first 198 consecutive pediatric cancer patients included in the Danish, prospective, nationwide study Sequencing Tumor And Germline DNA-Implications for National Guidelines (STAGING).

Ethics statement
Ethical approval was obtained through the regional scientific ethical committee (the Ethical Scientific Committees for the Capital Region, H-15016782) and the Danish Data Protection Agency (RH-2016-219, I-Suite no: 04804). All parents/guardians and patients 15 years or older gave formal written consent to participation in this study.

Inclusion criteria and national setup
CPSs were defined as likely pathogenic or pathogenic variants (PVs) in a gene, predisposing the carrier to childhood-or adult-onset cancer. Between 1 July and 31 December 2016 we included 25 patients in the STAGING pilot study at Rigshospitalet (Copenhagen University Hospital, Denmark). On 1 January 2017, the study was expanded to all four pediatric oncology departments in Denmark. All patients were enrolled before June 2018.
Patients were eligible for inclusion if aged 0-17 years at diagnosis of a primary cancer including benign brain tumors, Langerhans cell histiocytosis (LCH), or myelodysplastic syndrome and parents spoke and read Danish.
Families were provided with written and oral information about the study by a research nurse or oncologist. A PhD student from STAGING (AB) or clinical geneticist provided genetic counselling to families interested in participating. Counselling sessions included pedigree construction (three generations), recording the child's clinical phenotypic features according to McGill Interactive Pediatric OncoGenetic Guidelines (MIPOGG) [18] and Jongmans' criteria [13] (Table 1), and explaining the potential consequences of genetic findings. These consequences included secondary findings, variants of unknown significance (VUS), implications of pathogenic findings associated with CPSs, and subsequent preventive and surveillance measures. Families choosing to enroll in the study were informed that PVs in 'actionable' genes listed by the American College of Medical Genetics and Genomics (ACMG) [19] would be disclosed to them. Families could select information regarding: 1: PVs in ACMG 'actionable' genes. 2: "1" and PVs in 314 known and putative cancer genes. Heterozygous variants in genes with solely recessive inheritance patterns were reported only if further familial genetic testing was warranted, in accordance with clinical guidelines.
3: In addition to "1" and "2", PVs in genes unrelated to CPSs (Table 2). Variants were only reported if clinical consequences were anticipated. Findings in these genes are not presented here.
Pedigrees covering 1 st -3 rd -generation family members were constructed for all patients. 1 stdegree family members were parents and siblings, 2 nd -degree family members were uncles/ aunts and grandparents, and 3 rd -degree family members were cousins, grandparents' siblings and great-grandparents. Cancer diagnoses were verified using unique civil registration numbers, which link family members to medical records, including pathological descriptions of cancer, in the Danish Pathology Data Bank. Living family members gave consent, whereas medical records of deceased family members could be retrieved without consent.

DNA sampling and sequencing
Genomic DNA was isolated from peripheral blood samples. For patients with hematologic malignancies, blood samples were drawn after remission, otherwise skin biopsies were obtained. Parental blood samples were collected to establish whether variants were paternally or maternally derived or occurred de novo.
WGS was performed by the Norwegian Sequencing Center (Oslo, Norway) for the pilot study and by the Beijing Genomics Institute (Hong Kong, China) for the national study using the HiSeqX platform (Illumina, San Diego, CA, USA) with paired-end sequencing of  [7] Rahman, [26] and novel genes recently linked to childhood or adult CPSs. All variants with a minor allele frequency of <1% in any large population (gnomAD) were tabulated. For CPS genes with higher variant frequencies in the general population (e.g., ATM, CHEK2), a separate filter was used. We did not apply a specific variant filter to identify mosaicism. Variants were assessed by a team of clinical geneticists and molecular biologists based on variant type (e.g., frameshift, nonsense, missense), computational predictions of effect on protein and RNA function (e.g., Combined Annotation Dependent Depletion [CADD], PHRED quality score, ADA splice prediction score), [27] and database searches for published literature on each variant. Moreover, we used Alamut Visual 2.10 to evaluate variants effect on splicing (https://www. interactive-biosoftware.com/alamut-visual/). The effects of variants were considered significant if the scores of at least three programs were reduced �10% or a strong cryptic acceptor or donor site was generated. Variants were classified as pathogenic (class 5), likely pathogenic (class 4), VUS (class 3), likely benign (class 2), and benign (class 1). [28] Class 4 and 5 variants were designated 'PVs'. Class 3 variants, especially those that potentially matched the child's diagnosis, were further investigated by segregation analysis and splice predictions, and tumor RNA sequencing was used to assess loss-of-heterozygosity (LOH) if tissue was available (Fig 1). In addition, we used the machine-learning tool ORVAL to predict whether combinations of genetic variants were likely to be pathogenic. [29] Variants were discussed at regular multidisciplinary meetings by pediatric oncologists, clinical geneticists, and bioinformaticians. PVs were verified by Sanger or next-generation sequencing before parents were informed. Level 1 Information regarding pathogenic or likely pathogenic variants in genes identified by the American College of Medical Genetics [19]. These genes are 'actionable' i.e., there are potential preventive, treatment, or surveillance modalities available. Half of these genes are related to CPS (primary findings); the others are related to cardiac disease, metabolic disorders, or familial hypercholesterolemia (secondary findings).
Level 2 In addition to the genes listed at level 1, information regarding pathogenic or likely pathogenic variants in other known or putative CPS genes (from the list of 314 CPS genes found in S 1A Data). These were considered primary findings. However, if there was no known correlation between the clinical phenotype and the gene in question, the variant was considered a secondary finding.
Level 3 In addition to the genes listed at levels 1 and 2, families would also receive information regarding pathogenic or likely pathogenic variants in other genes not related to CPS (not presented in this paper). These were considered secondary findings. https://doi.org/10.1371/journal.pgen.1009231.t002

Statistical analysis
Patient/parental characteristics were compared using Pearson chi-square test. Two-sided Pvalues below 0.05 were considered statistically significant. Statistical calculations were carried out using the R software version 3.6.0.

Patient characteristics
Of 248 consecutive pediatric cancer patients that fulfilled the inclusion criteria, 198 families consented to participation (Fig 1,
These 29 patients had 31 PVs in total, including seven frameshift, five nonsense, eight missense, and three splice-site variants. Three variants were larger deletions of at least one exon, one patient had UPD 11p, and four patients had trisomy 21.
Some CPSs occurred in several patients (Table 4). Two patients had PVs in two CPS genes: a patient with LCH had PVs in BRCA2 and AXIN2; a patient with a malignant peripheral nerve sheath tumor had PVs in NF1 and PALB2. All CPS PVs were monoallelic, except for one patient with biallelic PVs in DDX41. [39] Of the 21 childhood-associated CPSs 18 (85.7%) had previously established links between genotype (e.g., trisomy 21 and leukemia) and cancers (Table 4). This study identified a PV in eight of the 21 (38.1%) pediatric CPS patients, whereas 13/21 (61.9%) patients had a previously established genetic predisposition syndrome (e.g., NF1 or trisomy 21). Such a connection was established through clinical diagnosis and/or genetic testing. Among these 21 patients, only one had a family member diagnosed with cancer before 18 years of age (Table 5). Inclusion and sequencing strategy. Variants presented are in genes associated with CPS. All variants of unknown significance have a CADD-PHRED score >20 and an allele frequency <1%. 1 Patients whose parents were not able to give informed consent due to language barriers or social issues (mainly parental psychiatric/severe somatic disease). https://doi.org/10.1371/journal.pgen.1009231.g001

PVs in biallelic CPS
Twenty-seven of 198 (13.6%) patients carried one (n = 22) or two (n = 1, FANCM and ADA) PVs predisposing toward CPS through biallelic inheritance, of whom four also had a PV in a monoallelic CPS gene (Table 6). Variants were found in ADA, ATR, EFL1, ERCC3, FANCA/C/ E/I/L/M, NBN, POLH, RECQL, RECQL4, RINT1, WRAP53, and XPC (Fig 2). Seven of the 198 (3.5%) patients carried two or more PVs/trisomy 21 (Table 7). Of these seven patients, one carried variants bioinformatically predicted to be oligogenically pathogenic. This patient carried biallelic variants in DDX41 and a monoallelic variant in NBN. The digenic combination of a single DDX41 variant and the NBN variant were also predicted to be pathogenic. No variant combinations in the six remaining patients were predicted to be oligogenically pathogenic.

Variants of unknown significance
All 198 patients carried VUS in one or more CPS genes. VUS with a frequency <1% and a CADD/PHRED score >20 are listed in S1 Data. Thirty-nine of 198 (19.7%) patients had a

Fulfillment of clinical criteria indicating an underlying CPS
All patients were evaluated using a phenotype checklist developed for this study (S1 Text), and 116/198 (58.6%) patients had one or more CPS-associated findings.
Overall, 70/198 (35.4%) patients fulfilled Jongmans' (n = 56, 28.3%) and/or MIPOGG criteria (n = 64, 32.3%) including 17 (81.0%) of the 21 patients with a childhood-onset CPS (Tables  8 and 9). Of the four patients with a PV in a childhood-onset CPS gene who did not fulfill Jongmans' criteria, none had excessive chemotherapy-induced toxicity. Patients not identified by either tool included two with PVs in TP53, and one with a pathogenic SMARCA4 variant. The patient with a PV in CDC73 was identifies by MIPOGG and not by Jongmans. The SMARCA4-deletion patient was diagnosed with synovial sarcoma of the ovary, revised to  5 The initial diagnosis (synovial sarcoma) was revised later revised 6 Validation in process 7 Detected by clinical analysis 8 These patients are twin brothers 9 The specific variants in APC and ATM are only associated with an adult-onset cancer predisposition syndrome; had the variant been associated with childhood cancer predisposition syndrome, these genes would have been listed above. 10 These variants confer a moderate risk of cancer in adulthood https://doi.org/10.1371/journal.pgen.1009231.t004

PLOS GENETICS
Nationwide germline whole genome sequencing of 198 consecutive pediatric cancer patients small-cell carcinoma based on this study, and would have fulfilled both criteria if the initial diagnosis had been correct. Of the patients with adult-onset CPS, 2/9 (22.2%) fulfilled Jongmans' (n = 2) and MIPOGG (n = 1) criteria, which is significantly fewer than for childhoodonset CPS (p = 0.0022). Of the additional 23 patients with a heterozygous PV predisposing to biallelic CPS, eight (34.8%) fulfilled Jongmans' (n = 5) and MIPOGG (n = 7) criteria. The number of VUS identified were higher among patients without a CPS and with adultonset CPS compared to patients with a childhood-onset CPS, in the first group patients on average carried 2.5 VUS compared to 1.6 VUS in the latter group. The same was the case when comparing patients with a childhood-onset CPS to patients who solely fulfilled Jongmans/ MIPOGGs criteria, patients carrying a childhood-onset CPS carried an average of 1.6 VUS compared to 2.5 VUS among patients fulfilling Jongmans/MIPOGGs criteria alone.

Family histories of cancer
Parents reported cancer diagnoses for 704 family members, 106 of whom resided outside Denmark, precluding further verification. Cancer diagnoses were verified for 328 (54.8%) of the remaining 598 family members, whereas the others did not consent to retrieval of medical records (n = 45) or their diagnoses could not be verified (n = 225) due to difficulties identifying distant/deceased family members or cancer occurrence prior to registration in Danish registries (before 1943). For 1 st -, 2 nd -, and 3 rd -degree relatives 16 (100.0%), 133 (84.1%), and 179 (42.2%) cases were verified, respectively. The following is based on verified diagnoses and family recollection (for 1 st to 3 rd generation family members) when verification was impossible.
In total, 191/198 (96.5%) participants had a family history of cancer. Seven of 198 (3.5%) participants had a family member diagnosed with cancer before 18 years of age (two had a CPS). Fifty-six of 198 (28.3%) participants had at least one relative diagnosed with cancer between the ages of 18 and 45. Three of 198 (1.5%) participants had two or more relatives under the age of 45 diagnosed with cancer.
Forty-three of 198 (21.7%) participants had a relative with a cancer of the same organ system as the patient. Patients with hematologic malignancies and solid tumors did not have more family members with cancers of the same organ system than the other two patient groups. In contrast, patients with a CNS tumor had a family member with a malignancy in the CNS more frequently than patients with either solid tumors or hematologic malignancies (p = 0.04). This association also held (p = 0.04) when patients with a CPS were eliminated (Table 10). Family history for the patients with a confirmed CPS can be found in Table 5.

Secondary findings
Two patients had PVs in genes associated with familial hypercholesterolemia (APOB and LDLR), three had PVs in genes associated with arrhythmic right ventricular cardiomyopathy (DSC2, DSG2, and PKP2), and one patient had a PV in KCNQ1, which is associated with long QT syndrome. These six variants are associated with increased risk of disease and families were informed. Overall, 18 patients (9.1%) had an ACMG 'actionable' PV, including 12 PVs in CPS genes and six PVs in genes associated with other non-malignant diseases (Table 11).

Discussion
In this first nationwide unselected cohort of consecutive pediatric cancer patients, half had a likely or validated underlying CPS, based on WGS, clinical examination, and pedigree

PLOS GENETICS
mapping, and 14.6% had a genetically verified CPS. These findings strongly indicate that genetic predisposition to childhood cancer may be far more common than previously supposed. Furthermore, other modes of inheritance (di-, oligo-and polygenic risk) may play significant roles in pathogenesis. Our childhood-onset CPS results are consistent with previous studies that found a CPS in 7-10% of pediatric cancer patients. [8][9][10][11] The frequency of PVs in our study was significantly higher than that observed in control cohorts wherein 0.6-1.1% of adult patients in a Genomics England cohort and a cohort of pediatric and adult patients with autism had PVs in CPS genes. [7] This study, however, excluded patients with known CPS and included different genes (including genes with frequent somatic variants) and is thus not directly comparable. A study of osteosarcoma patients found a remarkably high frequency of CPSs in control cohorts (12.1% and 9.3%), probably because many of the genes included (e.g., PMS1, and COL7A1) were not definitely linked to cancer. [6] Other studies have included adult-onset CPS genes. For example, Zhang et al. found only 0.7% of their patients carried an adult-onset CPS variant (BRCA1/2 and PALB2). [7] We found that 1.5% of our patients carried PVs in BRCA2 and PALB2. Interestingly, another Scandinavian study reported a significantly higher prevalence of childhood cancer in families with PVs in BRCA2. [40] Similarly, Wilson et al. showed that BRCA2 was one of the most frequently mutated genes among childhood cancer survivors. [11] Therefore, BRCA2 variants may be important in childhood cancer etiology, [40] potentially influencing treatment options if deficiencies in homologous repair promote some tumors. Even though a higher frequency of PVs in BRCA2 than BRCA1 have been found in the general population [41], this does not explain why multiple studies have identified so few PVs in BRCA1, which, in the Danish population, are more frequently associated with breast cancer than PVs in BRCA2. [42,43] Overall, WGS data from ethnically comparable children are lacking making comparisons of genetic findings in patients difficult. As most children survive cancer; therefore, identifying adult-onset CPSs is important for future surveillance and counselling.  Many of our patients carried PVs in genes involved in DNA repair. Children have high celldivision rates, and deficiencies in DNA repair may result in accumulation of DNA damage and ultimately cancer. Fanconi anemia (FA) is associated with many large genes, and the frequency of PVs in FA genes was 4.3% in an adult population of 7,578 patients from the Exome Sequencing Project and the 1000 Genomes Project. [44] This is consistent with our results, which showed that 13 (6.6%) patients carried a PV in a FA gene. Pathogenic FA variants are associated with a small increase in lifetime adult-onset cancer risk, [45][46][47] and this may also be true for childhood-onset CPS. Interestingly, we observed one CDC73 VUS and one PV in two patients with hematologic malignancies. PVs in CDC73 cause 'hyperparathyroidism-jaw tumor syndrome' and parathyroid carcinoma, [48] and have been linked with hematologic cancer in mouse models. [49] RNA sequencing of leukemic cells from these patients showed no LOH, making a causal association less likely but not impossible. [50] Furthermore, we identified two patients with heterozygous deleterious variants in ERCC6L2, a gene linked to a bone marrow failure syndrome. [51] These patients were diagnosed with T-lineage acute lymphoblastic leukemia (ALL) and rhabdomyosarcoma, respectively. Tumor tissue was not available for further investigation.
Seven patients had more than one PV in CPS genes, suggesting that di-, oligo-, and polygenic inheritance can cause predisposition to childhood cancer. Four of these patients exhibited the phenotype associated with the childhood-onset PV. Bioinformatic predictions suggested that pathogenicity was highly likely in one of the seven patients. Other studies have also found more than one PV in the same patient, suggesting that digenic/polygenic inheritance may play a role in childhood cancer etiology, [8,52,53] as MUTYH and OGG1 do in colorectal cancer etiology. [54] Kuhlen et al. [55] proposed a model of concomitant digenic inheritance involving two PVs within the same pathway combining to increase the likelihood of disease development. Generally, the observations from this and other studies suggest that the risk of disease development may increase by having more than one PV, even if the corresponding genes function in different pathways.
PVs in genes not previously associated with cancer development were identified. Some of these genes were 'actionable' and their identities were disclosed to the children's families, in accordance with ACMG recommendations. It is important to identify genes associated with increased risk of cardiac disease in pediatric cancer patients due to the increased risk of both cardiomyopathy [56] and symptoms in patients with long QT syndrome [57] undergoing anticancer treatment. PVs in these genes may also have clinical implications for family members.
We applied Jongmans' and MIPOGG criteria to assess the risk of underlying childhood CPSs. The majority of patients (85.7%) with a childhood-onset CPS were identified using these criteria. However, among the three unidentified patients were the two with Li-Fraumeni syndrome (one with ALL and one with osteosarcoma). Li-Fraumeni syndrome is associated with a high lifetime-risk of cancer, and the risk of a secondary cancer is further increased when the first cancer occurs during childhood. [58,59] Data suggest that surveillance programs for Li-Fraumeni patients increase their survival rates. [60] However, in contrast to other studies, patients with Li-Fraumeni syndrome were not identified here. [8,16] A possible explanation is that one of our patients carried a de novo TP53 variant that could not be identified from a family history of cancer. Additionally, CPSs that are not associated with syndromic features may not fulfill relevant criteria and will be difficult to identify if the cancer is not pathognomonic of the CPS. Family history was only rarely the cause of fulfillment of Jongmans'/MIPOGG criteria in both patients with childhood-and adult-onset CPS. The primary causes of fulfillment of Jongmans'/MIPOGG criteria were the patient's diagnosis and clinical characteristics. This is interesting as family history is believed to be highly indicative of adult-onset CPS like hereditary breast-and ovarian cancer and Lynch syndrome. A possible explanation for this could be that more variants are de novo in pediatric cancer patients and that the age of pediatric cancer patient's parents is lower than parents of adult cancer patients.
We found that a family history of CNS tumors was associated with the case of childhood CNS tumors. However, only two of 44 patients with a CNS tumor carried a germline variant (TSC2, NF1), and none of these two patients had a 1 st -3 rd -degree family member with a CNS tumor. Therefore, there may be unidentified predisposition genes among CNS tumor patients. However, recall bias cannot be excluded, because CNS tumors among family members might be more memorable, especially if a child is diagnosed with a CNS tumor.
One limitation of this study is the lack of a comparison cohort, because population-based WGS data from ethnically comparable children are not publicly available. Another problem is whether PVs in cancer genes of children with cancers that are unassociated with that particular gene have occurred randomly. Other studies have found similar cases, in which the genotype and phenotype were not previously reported, [4,8] and it remains uncertain whether PVs in adult CPSs are driver or passenger mutations. [5,61] Thus, a large international collection of cases should be investigated to describe the phenotypic spectrum associated with each CPS variant.
Strengths of our study include the national setup with a consecutive cohort of unselected pediatric cancer patients, in-depth clinical examinations of children with cancer, and use of national databases to verify cancer diagnoses in family members. Additionally, we performed WGS instead of WES or gene panel analyses so that large structural rearrangements and CNVs could be identified, if present. Moreover, WGS will facilitate future analyses of deep intronic variants that affect splicing, variants within putative regulatory areas, and novel CPS genes.
These results demonstrate the value of systematically screening pediatric cancer patients for CPSs and strongly indicate that a higher proportion of childhood cancers may be linked to predisposing germline variants than previously supposed.