Pathogenic Variants in Adult-Onset Cancer Predisposition Genes in Pediatric Cancer: Prevalence and Impact on Tumor Molecular Features and Clinical Management

The incorporation of genomic sequencing in pediatric oncology is increasingly uncovering pathogenic variants in what are more typically considered adult-onset cancer predisposition genes (aoCPG). These findings question the contribution of aoCPGs to pediatric tumor formation and raise clinical and ethical questions surrounding the germline testing of aoCPGs in children (1–20). Towards this end, recent studies have reported that pathogenic germline variants affecting BRCA2 and PALB2 are enriched in adult survivors of pediatric non-Hodgkin lymphoma and children with medulloblastoma, respectively (11, 12). Further, there are reports of rare pediatric tumors harboring aoCPG variants that exhibit loss or mutation of the remaining wild-type aoCPG allele (7, 12, 13, 20) and/or an abnormal mutational signature (7, 12, 13). Although these data suggest a causal role for aoCPGs in the formation of a subset of pediatric tumors, it remains unclear to what extent aoCPG variants occur in pediatric tumors and whether or how these variants influence tumor phenotypes, global molecular profiles, and approaches to clinical care.

Current recommendations surrounding tumor genomic testing include discussing the ability of this testing to identify variants of potential germline origin and referring patients for follow-up genetics evaluation to determine germline status when such variants are reported (21–27). Nevertheless, these recommendations do not address the clinical or ethical nuances surrounding the identification of tumor variants affecting aoCPGs in children. In adults, variants affecting certain aoCPGs (including those associated with hereditary breast and ovarian cancer) are more likely to be germline than somatic in origin when found in tumor tissue (28). It remains unknown whether the same phenomenon holds true for pediatric cancers.

The primary objective of this study was to evaluate the prevalence and spectrum of pathogenic variants in aoCPGs in a large cohort of pediatric patients whose tumors had been evaluated using clinical genomic sequencing at our institution. Further, we aimed to describe the tumor molecular features, clinical presentations, germline testing outcomes, and management impacts for children whose tumors harbored a pathogenic variant in one or more aoCPG.

This study was performed with approval from the St. Jude Children's Research Hospital Institutional Review Board. Tumor genomic reports from patients who had undergone clinical genomic testing through the St. Jude Clinical Genomics Laboratory between April 01, 2017, and December 31, 2019, were queried for the presence of pathogenic or likely pathogenic variants (hereafter denoted as “pathogenic variants”) in 24 aoCPGs. These genes were selected based on the association of monoallelic and/or biallelic germline pathogenic variants with development of cancers in adults plus the existence of cancer screening guidelines that are initiated in adulthood (29–31). The final gene list included: APC (exclusively the NM_000038.5: c.3920T>A [p.I1307K] variant), ATM, AXIN2, BRCA1, BRCA2, BRIP1, CDH1, CHEK2, EPCAM (exclusively monoallelic deletions of 3′ untranslated region), FLCN, GREM1, MET, MLH1, MSH2, MSH3, MSH6, MUTYH, NTHL1, PALB2, PMS2, POLD1, POLE, RAD51C, RAD51D. The specified APC and EPCAM variants were selected as they are associated with solely adult-onset cancer risk and screening guidelines. Five patients with a tumor aoCPG variant plus pathogenic or likely pathogenic germline variant in a pediatric-onset cancer predisposition gene were excluded from this study (Supplementary Materials and Methods).

For patients whose tumors harbored one or more aoCPG variants, information about clinical, demographic, and tumor genomic data, as well as germline genetic testing and management alterations based on aoCPG data, were gathered with an end date of data collection of June 07, 2021. Management alterations were defined to include those affecting the patient's cancer treatment, initiation of cancer surveillance, and genetic counseling and cascade testing for other family members, as per previous studies from our group and others (7, 20, 32) as well as professional guidelines related to sequence variants in cancer (22, 33).

All tumor specimens were evaluated by an anatomic pathologist, hematopathologist, or neuropathologist to confirm diagnosis and specimen adequacy. Tumors were sequenced using: (i) two-platform whole exome sequencing and RNA sequencing with analysis of nearly 1,000 cancer-related genes (for formalin-fixed paraffin-embedded tumor samples plus a paired germline sample); or (ii) three-platform whole exome and whole-genome sequencing (WGS) and RNA sequencing with analysis of ∼1,000 cancer-related genes (for fresh-frozen tumors plus a paired germline sample; ref. 7).

DNA extracted from the patient's tumor sample and germline sample were used to construct libraries for WES using the Illumina TruSeq Exome Enrichment Kit and for WGS using Illumina TruSeq DNA LT PCR-Free Sample Kit. RNA was only extracted from the patient's tumor sample and the Illumina TruSeq Stranded Total RNA LT Kit was used to generate libraries for RNA-seq. All libraries were sequenced using a paired end 2 × 125 bp cycle protocol and SBS technology on Illumina HiSeq and/or NextSeq Instruments. The WES/RNA assay detects fusion transcripts, SNPs, insertions, and deletions (indels). The WGS/WES/RNA assay detects copy-number variation, structural variation, fusion transcripts, SNPs, insertions, and deletions (indels). For WGS, samples were sequenced to obtain approximately 40% of bases at 45× depth of coverage, for WES, samples were sequenced to obtain approximately 65% of bases at 45× depth of coverage, and, for RNA-seq, samples were sequenced to obtain approximately 15% of bases at 45× depth of coverage. Variants must be present in a minimum of three reads to be reported.

Data from WGS, WES, and RNA-seq were analyzed independently to validate the presence of lesions in two or more platforms. A combined analysis was also performed to provide higher coverage particularly in cases with tumor heterogeneity, low tumor purity, and/or for regions of the genome that are difficult to sequence due to high GC content. Sequence information was aligned against a human reference sequence (hg19), utilizing a custom bioinformatics pipeline developed at St. Jude Children's Research Hospital.

Variants found in the tumor were compared with the matched germline to eliminate private benign polymorphisms. Data analyses were performed using publicly available software, publicly available databases, the proprietary database HGMD and databases/software developed at St. Jude Children's Research Hospital. Variants were selected for review based on these resources, scientific literature, and other predefined variant features. Variants were reviewed by a panel of experts prior to their inclusion in the tumor report.

The potential germline status of any pathogenic aoCPG variants identified in tumors was not specified in the tumor genomic reports due to a sequential two-step consenting process employed at our institution during the study period. Specifically, patients were first approached by oncologists for informed consent for tumor testing. Once tumor genomic reports were placed into the electronic medical record, patients were seen by the cancer predisposition team. The process for germline testing evolved over the course of this study's timeline. Decisions for germline testing were individualized based on hereditary cancer evaluation and family or provider preference. See Supplementary Table S1 for germline testing details.

For each paired tumor-normal sample with WGS data, somatic mutations were called by an in-house ensemble approach as described in Montefiori and colleagues (34). Briefly, somatic single nucleotide variants (SNV) and indels were called by five variant callers, including Mutect2 (v4.0.2.1; ref. 35), SomaticSniper (v1.0; ref. 36), Varscan2 (v2.4.3; ref. 37), MuSE (v1.0; ref. 38), and Strelka2 (v2.4.7; ref. 39), followed by consensus calling and variant review. Twenty-one tumors with >100 SNVs identified by at least three callers were selected for signature analysis.

To accurately assess mutational signature activities in each sample, we combined data from these 21 aoCPG-positive tumors with data from 199 tumors without pathogenic variants in aoCPGs sequenced through the Genomes from Kids (G4K) research study (7). We then separated tumors into different cancer types based on pathology information. When there were less than 10 samples per tumor type, data from tumors were combined into three general tumor groups: brain tumors, solid tumors, or hematologic malignancies. For each cancer type or tumor group, we used SigProfiler Bioinformatics Tools to construct mutational catalogs (single-base substitutions in 96-element form, Supplementary Table S2), extract de novo signatures, and decompose them into COSMIC (v3.1) single base substitution (SBS) signatures (40). To determine the mutational activities and their significance to the identified COSMIC SBS signatures in each sample (41), we used signature.tools.lib (v1.0) to run 1,000 permutations on the original 96-element mutational catalogue of each sample and estimate the signature activities in each permutation (Supplementary Table S3; ref. 42). Signatures with relative contribution above 5% with a P value <0.01 were considered significant and other signatures were annotated as “Unassigned.” The final signature activities of all 21 samples were summarized into a plot using R package ComplexHeatmap (Supplementary Table S4; ref. 43).

Homologous recombination deficiency (HRD) scores were calculated for the same 21 tumors from this cohort and 199 tumors from G4K (7). Specifically, for matched tumor and normal samples, we used Sequenza (v3.0.0) followed by scarHRD to calculate three genomic scar scores, including loss of heterozygosity (LOH), large-scale transitions, and telomeric allelic imbalances, the summation of which was used as the HRD score (Supplementary Table S5; refs. 44, 45).

Clinical data were collected by manual chart review, including patient demographics, tumor genomic findings, family history information, and management outcomes. For this study, a family history was defined as "positive" if there was at least one cancer within the established phenotype of the aoCPG in first, second-, or third-degree relative(s), at any age. We also determined whether family histories met testing criteria for the relevant aoCPG [i.e., Bethesda (46) and Amsterdam (47)] criteria for mismatch repair (MMR) genes and National Comprehensive Cancer Network (NCCN) guidelines for hereditary breast, ovary, and pancreas cancer (48).

Prevalence was calculated as the number of patients with tumors containing aoCPG pathogenic variants over the total number of patients whose tumors underwent genomic sequencing. Descriptive statistics were used to analyze data. Significant differences in HRD scores were compared between cases with biallelic, monoallelic, and no alterations of homologous recombination genes using Wilcoxon test and ANOVA using R package “ggpubr.”

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/. Any additional data are available upon request from the corresponding author.

Tumor genomic reports from 1,018 individuals were queried for the presence of pathogenic variants among 24 aoCPGs. Thirty-four tumors from 33 patients (3%) were identified to harbor one or more pathogenic variants affecting an aoCPG (Fig. 1; Supplementary Table S6). The frequency of pathogenic variants in aoCPGs was 3% (24 of 683 cases) in tumors evaluated using three-platform sequencing and 3% (10 of 335 cases) in those evaluated using two-platform sequencing (P = 0.85, Fisher exact test). In total, 36 pathogenic variants were identified affecting 11 different aoCPGs with ATM being the most frequently involved gene (n = 12 variants; 33%).

Twenty-two patients (67%) were male (Table 1; Supplementary Table S6). The average age at first tumor diagnosis was 8 years (0–19 years). Fifteen (45%) patients had central nervous system (CNS) tumors; 9 (27%) had hematologic malignancies and 9 (27%) had non-CNS solid tumors. Two tumors harbored pathogenic variants affecting two different aoCPGs (SJHGG031119: MLH1, ATM; SJLGG031317: PMS2, AXIN2) and one tumor harbored two pathogenic variants in MSH6 (SJTALL030640).

We next examined tumor genomic data to determine the molecular impacts of aoCPG variants. For this analysis, we first evaluated tumor genomic data for presence of biallelic alterations (“two hits”) affecting the relevant aoCPG. Twenty-one tumors with WGS data were included in this analysis (Supplementary Table S6; Fig. 2). Among these 21 tumors, six (29%) harbored genomic lesions affecting the second copy of the relevant aoCPG. These included four cases with whole or partial chromosome losses [two involving ATM (SJHGG030665, SJBT030809), and one each involving MUTYH (SJNBL030820), and PMS2 (SJMDS031482)]. In addition, there was one tumor with copy neutral LOH (cnLOH) rendering the pathogenic variant homozygous in the tumor (SJBT030439, PMS2) and one with two SNVs of undetermined phase (SJTALL030640, MSH6). This latter tumor exhibited a hypermutator phenotype, suggesting that the MSH6 pathogenic variants were in trans. For the seven tumors with MMR gene variants, IHC staining demonstrated loss of expression of the relevant protein in two of three tumors analyzed (SJBT030439, SJLGG031317, both with PMS2 variants, Supplementary Table S6). As we did not have WGS data from SJLGG031317 and could not confirm a second hit or mutational signature at the DNA level, this tumor was not included in the final count of tumors harboring second hit mutations. Microsatellite instability analysis was not performed on any samples.

Mutations in DNA repair genes can exert distinct mutational signatures in tumor genomes, often linked to biallelic inactivation of the corresponding genes (49). Towards this end, four of six tumors (67%) with biallelic and one of 15 tumors (7%) with monoallelic aoCPG inactivation exhibited a mutational signature corresponding to the affected DNA repair process (Fig. 2; Supplementary Figs. S1 and S2; Supplementary Table S4). For example, a dominant mutational signature for defective base excision repair, SBS36, was present in a tumor with biallelic loss of MUTYH (SJNBL030820). Consistent with MUTYH's role in the repair of oxidative DNA damage, the presence of SBS36 suggests that MUTYH function was compromised in this tumor. Notably, a different sample with monoallelic inactivation of MUTYH (SJBT031704) exhibited a related signature, SBS18, which is also associated with oxidative DNA damage. Three tumors with biallelic inactivation of MMR genes (MSH6: SJTALL030640; and PMS2: SJBT030439, SJMDS031482) exhibited SBS15, SBS21, and SBS26, the signatures associated with defective MMR.

COSMIC signature SBS3 is strongly associated with biallelic inactivation of BRCA1/2 and has been proposed to be a predictor of HR deficiency (49–52). None of our samples with monoallelic BRCA2 inactivation exhibited SBS3 activity, nor did the samples with monoallelic inactivation of other HR pathway genes such as CHEK2 or PALB2. Although previous studies have shown no association between ATM mutations and signature 3 in breast cancer (50, 53), SBS3 was found in two tumors harboring ATM variants: one with biallelic variants (SJBT030809) and one with a monoallelic variant (SJRHB031084); however, both samples also harbored somatic biallelic pathogenic TP53 alterations. An osteosarcoma (SJOS030422) with monoallelic MSH2 and TP53 variants also showed SBS3 activity without obvious mutations affecting the HR pathway; however, it has been reported that osteosarcomas frequently show mutational phenotypes resembling BRCA-deficient tumors (54). From the current data, it is not possible to determine whether the signature 3 activity observed in these tumors results from the ATM and/or TP53 variants.

To further evaluate HR deficiency, we also calculated the HRD score (45). Among 12 tumors with biallelic (ATM: n = 2) or monoallelic (ATM, BRCA2, CHEK2, PALB2: n = 10) inactivation of HR pathway genes, 11 passed the filtering threshold for HRD score analysis (cellularity >0.25; Supplementary Fig. S3). Two cases with biallelic inactivation of ATM [SJBT030809 (also with biallelic TP53 inactivation), and SJHGG030665 (no TP53 mutations)] displayed significantly elevated HRD scores when compared with nine cases with monoallelic inactivation of HR genes (Wilcoxon P-value = 0.022) or to 159 cases without pathogenic variants in these genes (Wilcoxon P-value = 0.012, eight cases from this cohort and 151 cases from the G4K cohort (Supplementary Fig. S3; Supplementary Table S5; ref. 7). Among the three samples exhibiting SBS3 activity, two [SJBT030809 (biallelic ATM and TP53 alterations), SJOS030422 (monoallelic MSH2 and TP53 alterations)] had an HRD score >42, suggesting full blown HR deficiency (55).

Genetic counseling and germline testing were completed for 26 of 33 (79%) patients whose tumors harbored aoCPG variants with 21 of these 26 (81%) testing positive for the pathogenic aoCPG variant identified in the tumor (Fig. 1; Supplementary Tables S1 and S6). One patient with a non-WNT/nonsonic hedgehog medulloblastoma harbored a pathogenic germline PALB2 variant (SJMB030498), consistent with the known enrichment of germline PALB2 variants in this subgroup (12). The remaining 20 patients had tumors not known to be associated with the identified aoCPGs. Nevertheless, two of these 21 (10%) patients’ tumors exhibited second hits and mutation signatures suggesting a contributory role for the aoCPG in tumor formation. These included one child each with: (i) myelodysplastic syndrome, germline PMS2 variant, somatic chromosome 7 loss and defective MMR signature (SJMDS031482); and (ii) neuroblastoma, germline MUTYH variant, somatic chromosome 1p loss with defective base excision repair/MUTYH signature (SJNBL030820).

We examined the age at first tumor diagnosis for the 21 patients with germline variants affecting an aoCPG (“cases”) and compared these the ages of first tumor diagnosis for 932 patients who lacked tumor variants affecting aoCPGs (“controls”). The controls are presumed to be negative for aoCPG or other variants in the germline based on the absence of such variants in their tumors; however, not all the controls underwent germline testing. Among cases, there was a trend for younger age at first cancer with 6% diagnosed under 12 months, compared to 2% and 2% diagnosed at 1 to 9 years or over 10 years, respectively (P value: 0.0583, chi-square test).

All patients testing positive in the germline received genetic counseling and were provided information on future cancer risks and guideline-based surveillance and cancer risk-reducing interventions. Although no patients had their cancer treatment changed based on the presence of tumor aoCPG variants, germline information was used from four patients for clinical decision-making. One young adult patient with germline PMS2 and AXIN2 variants (SJLGG031317) began colon cancer screening as per recommended guidelines. The other three patients were minors with anticipated hematopoietic stem cell transplantation for whom aoCPG variant information was used when considering family bone marrow donors (SJMDS031482, PMS2; SJBALL031495, ATM; SJAML031434, CHEK2).

Eight of 21 (38%) patients with germline aoCPG variants had family histories suggestive of the associated aoCPG and of these, 6 (75%) had at least one family member who met hereditary breast, ovary, and pancreas cancer genetic testing criteria (48) (Supplementary Table S6; Fig. 3). Thirty-three parents/custodial adult relatives from 20 families pursued germline testing (Supplementary Table S6). As of writing, 17 parents/custodial adult relatives tested positive and have been provided with cancer screening and cancer risk reducing recommendations.

The increased incorporation of tumor genomic sequencing in pediatric oncology is identifying an expanding array of pathogenic variants affecting aoCPGs, as highlighted by this study, one of the largest to date. Herein, we report that 33 of 1,018 (3%) pediatric, adolescent, and young adult patients had tumors harboring pathogenic variants affecting one or more aoCPG (Fig. 3). About 80% of tumor aoCPG variants were germline in origin. This latter observation brings with it new and often unexpected findings to address with families of children with cancer, challenging traditional genetic counseling and germline testing practices.

ATM, BRCA2, PALB2, and CHEK2 were among the aoCPGs most frequently altered in the tumors in this study. This finding agrees with recent reports, including one by MacFarland and colleagues, that demonstrated a high prevalence of ATM variants in pediatric oncology patients (56) and with others that describe PALB2, BRCA2, and CHEK2 variants in children with cancer (1, 2, 4–6, 9, 14, 57). Some of these studies demonstrate enrichment for germline BRCA2 variants in children with cancer compared with non-cancer controls (4, 11, 12). The prevalence of germline ATM variants in our cohort was 1% (11 of 1,018; assuming tumor ATM variants in patients who were not offered or did not pursue testing were germline in origin). This is similar to 0.7% of patients with personal and/or family history of cancer undergoing hereditary cancer panel testing in a U.S. study (58) and to the statistically significant prevalence in patients with breast cancer (0.8%) versus healthy unaffected controls (0.4%) in a recent population-based study (59). The prevalence of variants in BRCA2, PALB2, and CHEK2 was 0.3% for each in our cohort, as compared with the prevalence in healthy unaffected controls (0.24%, 0.12%, and 0.42%, respectively) in the same population-based study (59). Therefore, while the prevalence of pathogenic variants in ATM appears elevated in our cohort, the prevalence of variants in BRCA2, PALB2, and CHEK2 was similar to the general population. Because of the low numbers of patients with germline variants in individual aoCPGs and the heterogeneity of tumors in this study, we were not able to confirm enrichment of germline aoCPG variants within this cohort compared with healthy noncancer controls.

To determine whether aoCPGs contribute to pediatric tumor formation, we looked at the molecular features of tumors harboring pathogenic variants affecting aoCPGs. Through the analysis of 21 tumors with WGS data, we observed six tumors with bi-allelic inactivation or loss of the aoCPG. Although it is possible the biallelic losses were due to chance (e.g., passenger events), we found that four of six tumors with biallelic aoCPG inactivation exhibited DNA mutational signatures with relevant etiologies. For example, there was evidence of defective MMR in all three tumors with biallelic alterations in MSH6 or PMS2, but not in tumors with monoallelic alterations in these or MSH2. A fourth tumor with biallelic MUTYH alterations exhibited a mutational profile consistent with impaired base excision repair, consistent with our prior report (7). Although a recent report of high-risk pediatric cancers reported presence of the HR deficiency signature SBS3 in four of four tumors with monoallelic BRCA2 and one of three tumors with monoallelic CHEK variants (13), we could find no such evidence of this signature in any of the tumors harboring monoallelic variants in these genes. In contrast, we did observe SBS3 in three samples: one with biallelic inactivation of ATM and TP53 and a high HRD score; one with monoallelic loss of ATM but no TP53 alterations and a moderate HRD score; and one with no known pathogenic mutations in HR pathway genes but biallelic TP53 alterations and a high HRD score. Given the presence of biallelic TP53 inactivation in both tumors with high HRD score and SBS3, it is likely that the loss of TP53 function is driving the signature 3 activity in these cases.

Germline testing was completed in 26 patients, with 21 (81%) having one or more of the aoCPG variants confirmed to be of germline origin despite a negative family history in most cases. Therefore, when a pathogenic variant affecting an aoCPG is included in a child's tumor report, strong consideration should be given to genetic counseling with possible germline testing. Notably, all young adult patients or parents who were offered germline testing for their child consented to this testing. This finding suggests that future cancer risks are an important factor in the germline testing decision-making process. However, this also raises the question of whether additional minor-aged siblings should also be tested for the aoCPG variants. Currently, there is insufficient evidence to recommend germline testing of pediatric-aged relatives or additional cancer screening for children with germline variants in aoCPGs; ideally, such families should be enrolled on research protocols. A flexible approach to testing minors, balancing the child's and family's shared interests and values is recommended (60–63).

Germline aoCPG information was one of the factors considered by our cellular therapy team when selecting related bone marrow donors for hematopoietic stem cell transplantation for 3 patients with germline ATM, CHEK2, and PMS2 variants. Germline genetic testing of biologically related donors is an evolving dimension in the transplant decision-making process. Currently, multidisciplinary teams must consider hereditary implications of pathogenic variants typically not identified in the pre-genomics era. There is often no clear guidance about which germline pathogenic variants are permissible (or not) when determining eligibility of biologically-related bone marrow donors (64). These decisions are becoming increasingly complex considering an expanding understanding of hematologic malignancy risk in individuals who harbor germline variants affecting HR pathway genes such as CHEK2 and ATM (65). Towards this end, we identified one patient who received bone marrow from a family donor where both were confirmed to harbor a heterozygous pathogenic germline ATM variant after transplant. It remains to be determined whether the presence of this variant will have an impact on this child's overall outcome.

Finally, the longer term psychological, ethical, and social aspects of germline testing for aoCPG variants in pediatric patients are important areas of study (62). Follow-up with a genetic counselor or other provider during adolescence is critical for patients known to harbor germline aoCPG variants so that they can be fully informed about future cancer risks, recommended approaches to clinical care, and family planning options. As well, for patients for whom germline testing for aoCPG pathogenic variants is not pursued in childhood, knowledge of the aoCPG variant in the tumor should not be lost or forgotten so that consideration of counseling and/or germline testing can be revisited when patients are older and ready for these conversations.

There are several limitations to this study. Pathogenic tumor variants in aoCPGs may have been missed or unreported due to technicalities related to the computational pipelines or pathology practices in place. Not all tumors had sufficient data to evaluate for somatic second hits or DNA mutation signatures and not all patients underwent germline testing for the identified aoCPG variants. Epigenetic silencing was not considered when evaluating for somatic second hits. Finally, we were not able to demonstrate statistical enrichment of aoCPG pathogenic variants in children with cancer compared with controls. Nevertheless, examination of individual cancer genomes uncovered distinctive mutational signatures for cases harboring aoCPG pathogenic variants, thus providing potential for genomics-directed therapies.

While only a small proportion of pediatric tumors harbors pathogenic variants affecting aoCPGs, the vast majority arise in the germline. On the basis of these findings, we recommend that pediatric oncologists develop mechanisms for clearly communicating with families about the possible presence and implications of aoCPG variants identified through tumor testing. Evidence from this study and others reveals that aoCPG variants contribute to the pathogenesis of a subset of childhood tumors, but further studies are needed to elucidate the mechanisms by which these variants promote tumor formation, how they influence tumor presentations and responses to current therapies, and whether they can be targeted to improve overall outcomes.

No author disclosures were reported.

R.B. McGee: Conceptualization, resources, data curation, formal analysis, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing. N. Oak: Data curation, software, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. L. Harrison: Data curation, formal analysis. K. Xu: Data curation, software, formal analysis, investigation, methodology, writing–review and editing. R. Nuccio: Resources, project administration. A.K. Blake: Resources, project administration. R. Mostafavi: Resources, data curation. S. Lewis: Resources, data curation. L.M. Taylor: Resources, data curation. M. Kubal: Resources, software. A. Ouma: Resources. S.J. Hines-Dowell: Resources. C. Cheng: Formal analysis. L.V. Furtado: Resources. K.E. Nichols: Conceptualization, resources, supervision, funding acquisition, investigation, project administration, writing–review and editing.

Funding for this study was provided by the American Lebanese Syrian associated charities. The authors thank the patients and families included in this study and members of the St. Jude Clinical Genomics Laboratory, without whom this work would not have been possible. We thank Emily Ashcraft from the Biostatistics Shared Resource at St. Jude for her assistance with statistical analysis. We thank Gang Wu from the Center for Applied Bioinformatics Shared Resource at St. Jude for his review and comments on the bioinformatics analysis. We also thank Jinghui Zhang and David Wheeler for their review and comments on the manuscript.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.