An evaluation of genetic predisposition to congenital anomalies and pediatric cancer supports KAT6B as a novel neuroblastoma susceptibility gene

Abstract

Purpose

Although congenital anomalies are among the strongest risk factors for developing pediatric cancer, the genetic underpinnings remain unclear. Therefore, we evaluated germline susceptibility in children with congenital anomalies and cancer.

Methods

Through the Genetic Overlap Between Anomalies and Cancer in Kids Study, we recruited 47 participants with anomalies and cancer, along with their biological families. Genome sequencing was performed, focusing on single-nucleotide variants, indels, and structural variants. Pathogenic or likely pathogenic variants were identified by the American College of Medical Genetics and Genomics classification.

Results

We identified pathogenic or likely pathogenic variants in 23.4% (11 of 47) of participants. These variants encompassed (1) 4 genes associated with both anomalies and cancer (WT1, USP9X, PTPN1, and LZTR1), (2) 2 established cancer predisposition genes (TP53 in 2 participants and PAX5), and (3) 4 genes that are associated with anomalies (MMUT, FBN1, COL3A1, and KAT6B). We further investigated the role of KAT6B on neuroblastoma in a gene-based analysis from 409 neuroblastoma cases and 952 controls. This analysis demonstrated a significant enrichment of rare, predicted deleterious variants (P = .017), with odds ratios ranging from 2 to 4 based on the conditions we applied.

Conclusion

This study demonstrates a molecular diagnostic yield of 23.4% in participants with both anomalies and cancer. Additionally, the findings further implicate the role of KAT6B as a novel neuroblastoma predisposition gene.

Introduction

In the United States, more than 15,000 children per year are diagnosed with cancer,¹ making it the leading cause of death by disease in those 1 to 19 years of age.² Notably, one of the strongest risk factors for cancer in children is being born with congenital anomalies. This is true both for chromosomal anomalies (eg, Down syndrome) and nonchromosomal birth defects (eg, nonsyndromic congenital heart defects), as recently validated in the population-based registry linkage study of over 10 million live births.³ In fact, it has been estimated that 10% of pediatric cancer cases could be attributable to the risk associated with having a congenital anomaly.⁴

Despite these observations, the reasons underlying these associations remain unclear. It has been hypothesized that genetic mechanisms could play an import role in the cooccurrence of congenital anomalies and pediatric cancer.⁵ However, few candidate genes and pathogenic/likely pathogenic (P/LP) variants associated with these concurrent conditions have been identified, mainly because of limited research endeavors.

Here, we utilized data from genome sequencing (GS) from families recruited as part of the birth registry study, Genetic Overlap Between Anomalies and Cancer in Kids (GOBACK) Study³ to (1) comprehensively evaluate rare single-nucleotide variants, insertion and deletion variants (indels), and copy-number variants (CNVs) in children with congenital anomalies and pediatric cancer and (2) discern the effects of identified germline variants on individual phenotypes through detailed variant prioritization, interpretation, and evaluation of existing genomic databases, which would allow us to ascertain specific genotype-phenotype associations. Overall, the goal was to identify specific genes that may explain the cooccurrence of congenital anomalies and pediatric cancer.

Materials and Methods

Study population

As previously described in detail, as part of the GOBACK Study, we linked population-based birth defects registries to cancer registries in the following states (birth years): Arkansas (1995-2011), Michigan (1992-2011), North Carolina (2003-2012), and Texas (1999-2013).³ Through this linkage, we could identify children having concurrent congenital anomalies and diagnosed with cancer between 1999 and 2017. Next, research coordinators contacted and took consent from families for participation in further genomic studies. A brief study questionnaire was administered in English or Spanish. After the interview was completed, saliva collection kits were sent to obtain DNA samples from the child, their parents, and siblings when available.

Additionally, we identified children with congenital anomalies and cancer through electronic health records from 2002 to 2017. As with the registry-based study, a research coordinator approached and enrolled eligible families, conducted interviews with the parents of each child, collected information on the child’s family medical history, and facilitated the collection of saliva samples from each of the family members.

Clinical information and phenotypic terms

The comprehensive phenotypic details on the participants and the compiled demographic data are provided in Table 1. For this study, congenital anomaly diagnoses were coded using the Centers for Disease Control and Prevention modification of the British Paediatric Association Classification of Diseases and the World Health Organization’s International Classification of Diseases, Ninth Revision, Clinical Modification. Information on pediatric cancer diagnoses was obtained from participating cancer registries; tumors were classified according to the International Classification of Childhood Cancer, Third Edition.⁶

Table 1.

Characteristics of the study participants (N = 47)

Characteristics		No. of Participant (%)
Gender	Male	25 (53.2)
	Female	22 (46.8)
Race/ethnicity	Non-Hispanic Black	3 (6.4)
	Non-Hispanic White	16 (34.0)
	Hispanic	19 (40.4)
	Other	9 (19.1)
Congenital anomalies by group	Congenital musculoskeletal Anomalies/congenital anomalies of limbs	17 (22.4)
	Congenital anomalies of heart/circulatory system/cardiac septal closure	16 (21.1)
	Congenital anomalies of nervous system	13 (17.1)
	Congenital anomalies of urinary system	7 (9.2)
	Congenital anomalies of genital organs	7 (9.2)
	Congenital anomalies of eye, ear, face, and neck	7 (9.2)
	Congenital anomalies of digestive system	5 (6.6)
	Congenital anomalies of respiratory system	3 (3.9)
	Other	1 (1.3)
Number of anomalies	n = 1	26 (55.3)
	n = 2	8 (17)
	n = 3	5 (10.6)
	n > 3	8 (17)
Cancer diagnosed	Embryonal (n = 16, 34.0%)
	Neuroblastoma	6 (12.8)
	Wilms tumor	5 (10.6)
	Hepatoblastoma	4 (8.5)
	Other	1 (2.1)
	CNS (n = 10, 21.3%)
	Astrocytoma	4 (8.5)
	Ependymoma	3 (6.4)
	Other	3 (6.4)
	Sarcomas (n = 10, 21.3%)
	Rhabdomyosarcoma	4 (8.5)
	Osteosarcoma	3 (6.4)
	Other	3 (6.4)
	Hematologic malignancies (n = 6, 12.8%)
	B-ALL	5 (10.6)
	Other	1 (2.1)
	Other (n = 5, 10.6%)

B-ALL, B lymphoblastic leukemia/lymphoma; CNS, central nervous system.

Germline GS and variant identification and annotation

Saliva samples were collected using Oragene DNA saliva kits following manufacturer guidelines. Genomic DNA was extracted from saliva according to standard procedures. Germline GS was conducted. The data set included GS of 118 individuals, encompassing 47 participants along with their family members. This data set comprised 29 complete parent-offspring trios (and siblings when available), 13 duos (a participant and 1 parent), and 5 singletons (participant only). The methods for GS, variant identification, annotation, candidate variant prioritization, and interpretation are detailed in the Supplemental Methods.

Briefly, the post-sequencing analysis utilized the Mercury (HgV)⁷ analysis pipeline for a range of tasks, including base calling, mapping using Burrows-Wheeler Aligner-Maximal Exact Matches algorithm⁸ to the human genome reference sequence (hg38), and for the collection of quality control metrics. Samples had an average coverage of 40×, with 97% of bases reaching at least 20× coverage.

The single-nucleotide and small indel variant calling were performed using Platypus,⁹ and the results were annotated using ANNOVAR software.¹⁰ This annotation includes variant type, minor allele frequencies (MAFs) in population databases, multiple in silico predictions of variant deleteriousness, and information from genetic disease and clinical variant interpretation focused databases.

CNVs (deletions/duplications) were identified using 5 different structural variation detection algorithms (details in the Supplemental Methods), and the information from individual callers was integrated to achieve higher variant calling accuracy by using an ensemble approach. A CNV was considered valid if (1) it had more than an 80% reciprocal overlap across the different algorithms and (2) at least 3 of the 5 callers supported the CNV. The results obtained were annotated using ClassifyCNV.¹¹

Single-nucleotide variants and small indels candidate variant prioritization and interpretation

Because each participant had a different combination of phenotypic features, the variant prioritization was carried out for each participant separately. Variants were excluded if their overall MAF was >1% or if the variant was classified as intergenic, intronic, or synonymous without dbscSNV (Ada, RF)¹² or spliceAI scores¹³ or any reports in clinical variant databases. For the participants, we prioritized variants based on their inheritance mode: de novo, rare homozygous, compound heterozygous, and uniparentally inherited variants.

The data annotated by ANNOVAR and our specialized in house tools, encompassing mode of inheritance, MAF (overall and popmax), gene-level details, ClinVar interpretations, and in silico predictions (initial thresholds for prioritization: Rare Exome Variant Ensemble Learner (REVEL) score > 0.5,¹⁴ Combined Annotation-Dependent Depletion (CADD) score > 25,¹⁵ dbscSNV Ada, RF scores, both > 0.9, and spliceAI score > 0.2 for either donor or acceptor sites), along with disease annotations matched from Swiss-Prot¹⁶ or OMIM, were thoroughly cross-referenced with the phenotypes of each participant. Subsequently, additional literature review was also conducted. This comprehensive approach was employed to pinpoint the most probable candidate variants.

We followed American College of Medical Genetics and Genomics/Association of Molecular Pathology (ACMG/AMP) criteria¹⁷ specifically to evaluate P/LP variants within known monogenic disease genes, in alignment with the participant’s phenotype. All identified variants were visually verified using integrative genomics viewer.¹⁸ To validate the candidate de novo variants, orthogonal DNA-sequencing techniques were used to confirm the candidate single-nucleotide variants and how they were inherited. Targeted amplicons were amplified from genomic DNA using conventional polymerase chain reaction (HotStarTaqDNA polymerase, QIAGEN), and polymerase chain reaction amplification products were analyzed by Sanger sequencing using established methods.

CNVs candidate variant prioritization and interpretation

De novo deletions and duplications were identified using BamTools.¹⁹ Based on the annotated information, we excluded CNVs that were either fully contained within a recognized benign CNV region (MAF > 1%) or had been frequently reported as benign variants in multiple peer-reviewed publications or curated databases (details in the Supplemental Methods). The primary focus was de novo CNVs overlapping protein-coding genes, and CNVs larger than 10,000 bases regardless of gene overlap.

The prioritized CNVs were interpreted using the CNV interpretation standards from ACMG and the Clinical Genome Resource (ClinGen).²⁰ P/LP variants were identified and evaluated in the context of the participant’s phenotype and visually evaluated using integrative genomics viewer and/or Samplot.²¹

Gene-based association analysis

The initial trio analysis implicated KAT6B (HGNC: 17582) as a neuroblastoma risk gene, a finding we followed up with an expanded analysis of germline variants in KAT6B in a GS data set from 409 neuroblastoma cases from the Gabriella Miller Kids First Pediatric Research Program neuroblastoma data set and 952 control samples unaffected parents from unrelated disease groups (Ewing sarcoma, orofacial cleft, and congenital diaphragmatic hernia).²² This approach also ensures that genomic data for both the neuroblastoma cases and controls were sequenced at the same institute, using the same instruments, and analyzed using the same bioinformatics pipelines minimizing batch effects. This uniform approach ensures data homogeneity and helps avoid biases and errors that could arise from different sequencing methods, quality control processes, or bioinformatic analysis strategies. Specifically, we conducted joint-genotype calling for all samples using GLnexus (version 1.1.3) (Lin MF, Rodeh O, Penn J, et al. GLnexus: joint variant calling for large cohort sequencing. bioRxiv. 2018:343970. https://doi.org/10.1101/343970). We then conducted an extensive data harmonization and quality control procedure using Cross-Platform Association Toolkit (XPAT)²³ to control for technological stratification. Next, we used Variant Annotation, Analysis, and Search Tool (VAAST) 2.1²³ to conduct gene-based association tests incorporating the first 10 principal components to minimize population stratification. VAAST 2.1 algorithm incorporates functional prediction weights (conservation-controlled amino acid substitution Matrix [CASM] scores²⁴) for each variant and calculates P values using covariate adjusted permutation. Additional criteria (VAAST 2.1 > 0 and MAF < 0.5%) were applied to calculate P values for gene-based association. We estimated odds ratios based on specific filtering criteria, by excluding the variants reported as benign/likely benign from ClinVar and applying cutoffs for in silico prediction scores (CADD > 20, REVEL > 0.5, and CASM score > 1). Because the in silico prediction scores for CADD and REVEL are not available for indels, we treated these indels the same as variants that met or exceeded the cutoff values from in silico predictions. Following the above conditions, each odds ratio and its 95% confidence intervals was calculated.

Results

Characteristics of the study population

This study included 47 participants, plus available family members from the GOBACK Study with heterogeneous anomaly and cancer diagnoses as described in Table 1. Every participant presented with at least 1 congenital anomaly and was diagnosed with pediatric cancer before the age of 18. No 2 participants had the exact same combination of congenital anomalies and pediatric cancer. There were 13 different congenital anomalies diagnosed among the children included in the assessment with an average of 2 congenital anomalies per individual. Musculoskeletal anomalies were the most prevalent (15 of 47, 32%), followed by congenital heart defects (14 of 47, 30%), and central nervous system (CNS) defects (10 of 47, 21%). For pediatric cancer, children included in the assessment were diagnosed with a total of 21 different cancers. The most frequent tumors were embryonal in histology: neuroblastoma (n = 6), Wilms tumor (n = 5), and hepatoblastoma (n = 4). Other tumors represented in the GOBACK cohort included diverse types of CNS tumors (n = 10) and sarcomas (n = 10) (details regarding diagnoses are presented in Table 1).

Frequency of P/LP variants

Among the 47 participants we studied, we identified 11 (23.4%) participants with P/LP variants in known monogenic disease genes, providing a molecular diagnosis for their congenital anomalies or pediatric cancer or both (Table 2). Ten of 11 had single-nucleotide P/LP variants, whereas 1 was a heterozygous 5-kb deletion. Although not statistically significant, participants with P/LP variants had a higher number of congenital anomalies compared with participants without P/LP variants (mean = 2.91 vs 1.81, P = .053, as determined by a one-tail test). Among the 11 participants with P/LP variants, the most prevalent group of congenital anomalies were CNS defects and musculoskeletal anomalies each occurring in 5 participants. Among those with P/LP variants, rhabdomyosarcoma was the only tumor represented more than once with 3 of 4 cases harboring a P/LP variant (Figure 1).

Table 2.

Pathogenic and likely pathogenic variants causing CA and/or PC

No.	Category	Pediatric Cancer	Congenital Anomalies	Gene (HGNC ID)	Phenotypes and Inheritance Mode From OMIM (MIM Number)	P/LP Variant (Genomic Coordinate)	Trio	Inheritance	MAF	In Silico Prediction (CADD/REVEL)	ACMG/AMP Criteria	Notes
1	CA/PC	Wilms tumor	Congenital chordee, hypospadias, perineal, scrotal	WT1 (12796)	Wilms tumor, type 1, AD, SMu (194070)	c.1399C>T p.(Arg467Trp) (g.32392020G>A)	Yes	de novo	0	32/0.184	PS2, PM1, PM2, PP5, PP2	ClinVar: P/LP Locating in mutational hotspot (exon 8-9)
2	CA/PC	B-ALL	Microcephaly, choanal atresia, cataract, congenital mitral/tricuspid regurgitation, other specified structural developmental anomalies of chest wall and anal canal	USP9X (12632)	Intellectual developmental disorder, X-linked 99, XLR/XLD (300919, 300968)	c.5015+1539_5189+1668del p.(?) (g.41207032_41212350del)	Yes	de novo	0	NA	PS2, PM2, PM4	Inframe deletion (exon 33) Sufficient evidence for dosage pathogenicity (3) curated by ClinGen panel Separately published25
3	CA/PC	Rosette-forming glioneuronal tumor	Congenital supravalvar aortic stenosis, perimembranous central ventricular septal defect, artery dilation, cerebral palsy	PTPN11 (9644)	Noonan syndrome 1, AD (163950)	c.923A>G p.(Asn308Ser) (g.112477720A>G)	No	NA	0	22.6/0.685	PS3, PM1, PM2, PP2, PP3	ClinVar: P Separately published26
4	CA/PC	ACC	Structural developmental anomaly of heart or great vessels, unspecified	LZTR1 (6742)	Noonan syndrome, AD/AR (616564, 605275) {Schwannomatosis-2, susceptibility to}, AD (615670)	c.1033G>T p.(Glu345∗) (g.20992253G>T)	Yes	from father	0	37/NA	PVS1, PM2, PP5	ClinVar: P/LP
5	PC	Ependymoma	Arnold-Chiari malformation type I, esophageal fistula without atresia	TP53 (11998)	Li-Fraumeni syndrome, AD (151623)	c.738G>T p.(Met246Ile) (g.7674225C>T)	Yes	from mother	0.000003778	26.7/0.838	PS1, PS3, PM1, PM5, PP1, PP3, PP5	ClinVar: P/LP Mother: Li-Fraumeni syndrome
6	PC	RMS	Plagiocephaly, congenital torticollis	TP53 (11998)	Li-Fraumeni syndrome, AD (151623)	c.794T>C p.(Leu265Pro) (g.7673826A>G)	No	from father	0	21/0.9	PS3, PM1, PM2, PM5, PP3, PP5	ClinVar: P/LP
7	PC	B-ALL	Stricture or atresia of vagina	PAX5 (8619)	{Leukemia, acute lymphoblastic, susceptibility to, 3} (615545)	c.419G>T p.(Arg140Leu) (g.37006529C>A)	No	NA	0	29.8/0.968	PM1, PM2, PM5, PP3, PP5	Observed in PAX5-B-ALL as a somatic hotpot p.(Arg140Gln) (LOVD, ClinVar): LP in neurodevelopmental disorders
8	CA	RCC	Intracranial arachnoid cyst, other specified disorders of the nervous system, atrophy of the putamen	MMUT (7526)	Methylmalonic aciduria, mut (0) type, AR (251000)	1c.322C>T p.(Arg108Cys) (g.49459145G>A) 2c.682C>T p.(Arg228∗) (g.49457762G>A)	No	Compound heterozygous	0/0	130/0.956 236/NA	1PVS1, PM2, PP5 2PS3, PM2, PM5, PP3, PP5	Both - ClinVar: P/LP Separately published27
9	CA	RMS	Hypermobility of joint functions	FBN1 (3603)	Marfan syndrome, AD (154700)	c.7540G>A p.(Gly2514Arg) (g.48421982C>T)	No	from mother	0	29.5/0.945	PS4, PM2, PP2, PP3, PP4, PP5	ClinVar: P/LP
10	CA	RMS	Patent oval foramen, structural developmental anomalies of duodenum, talipes equinovarus	COL3A1 (2201)	Ehlers-Danlos syndrome, vascular type, AD (130050)	c.3496C>T p.(Arg1166∗) (g.189008113C>T)	Yes	from father	0.000003778	39/NA	PVS1, PS3, PP5	ClinVar: P/LP Sibling: inguinal hernia
11	CA	Neuroblastoma	Patella deformity, scrotal hypoplasia, inguinal hernia, renal anomalies, intellectual disabilities, facial dysmorphism	KAT6B (2201)	Genitopatellar/SBBYSS syndrome, AD (606170, 603736)	c.3169dup p.(Glu1057Glyfs∗10) (g.75022028dup)	Yes	de novo	0	NA	PVS1, PS2, PM2, PP5	ClinVar: P Null variant (via NMD)

Transcripts (Matched Annotation from NCBI and EMBL-EBI [MANE] select, hg38) and NC accession number (hg38): WT1, NM_024426.6, NC_000011.10; USP9X, NM_001039591.3, NC_000023.11; PTPN11, NM_002834.5, NC_000012.12; LZTR1, NM_006767.4, NC_000022.11; TP53, NM_000546.6, NC_000017.11; PAX5, NM_016734.3, NC_000009.12; MMUT, NM_000255.4, NC_000006.12; FBN1, NM_000138.5, NC_000015.10; COL3A1, NM_000090.4, NC_000002.12; KAT6B, NM_ 012330.4, NC_000010.11.

Each variant is heterozygous except for participant no. 8.: The compound heterozygous status of participant no. 8, who was from a nontrio configuration, was confirmed as part of the published case study.

ACMG/AMP, the American College of Medical Genetics and Genomics and the Association for Molecular Pathology; AD, autosomal dominant; AR, autosomal recessive; B-ALL, B lymphoblastic leukemia/lymphoma; CA, congenital anomalies; LOVD, Leiden Open Variation Database; MAF, minor allele frequency; NA, not available; NMD, nonsense-mediated decay; P/LP, pathogenic/likely pathogenic; PC, pediatric cancer; RCC, renal cell carcinoma; RMS, rhabdomyosarcoma; SBBYSS, Say-Barber-Biesecker-Young-Simpson (a variant of Ohdo syndrome); SMu, somatic mutation; VUS, variant of uncertain significance; XLR/XLD, X-linked recessive/dominant.

Figure 1.

Types of congenital anomalies and pediatric cancers in participants with pathogenic/likely pathogenic variants. B-ALL, B lymphoblastic leukemia/lymphoma; P/LP, pathogenic/likely pathogenic; RMS, Rhabdomyosarcoma.

We identified P/LP variants in 10 different genes that have reported associations with congenital anomalies, cancer, or both (Figure 2): (1) MMUT (HGNC: 7526, congenital anomalies); (2) COL3A1 (HGNC: 2201, congenital anomalies); (3) FBN1 (HGNC: 3603, congenital anomalies); (4) KAT6B (HGNC: 17582, congenital anomalies); (5) TP53 (HGNC: 11998, cancer); (6) PAX5 (HGNC: 8619, cancer); (7) WT1 (HGNC: 12796, both); (8) PTPN11 (HGNC: 9644, both); (9) LZTR1 (HGNC: 6742, both); and (10) USP9X (HGNC: 12632, both). Beyond these clear P/LP variants, we identified other P/LP variants that might not directly correlate with the participants’ congenital anomalies and/or pediatric cancer (details in the Supplemental Table 1). Some of the genes listed in this table (eg, CHEK2 [HGNC: 16627], MUTYH [HGNC: 7527]) are associated with an increased risk for certain cancers.^28,29 But because these genes are reported in adult-onset cancers and not clearly associated with pediatric cancers, we will not discuss these further.

Figure 2.

Identified genes with pathogenic/likely pathogenic variants in cases of congenital anomalies and pediatric cancer.

P/LP variants in congenital anomalies and pediatric cancer genes

As noted, 4 participants had P/LP variants in genes known to contribute to both congenital anomalies and pediatric cancer (a category labeled by “CA/PC” in Table 2). These included (1) a de novo P/LP variant in WT1 (p.(Asn308Ser), heterozygous) in a participant with genitourinary anomalies and Wilms tumor; (2) a de novo 5318 base-pair deletion spanning exon 33 in USP9X in a participant diagnosed with multiple congenital anomalies, intellectual disabilities, and B lymphoblastic leukemia (B-ALL) (separately reported)³⁰; (3) a pathogenic variant in PTPN11 (p.(Asn308Ser), heterozygous) in a participant with multiple congenital heart defects, cerebral palsy and glioneural tumor (as separately reported)²⁵; and (4) an inherited variant in LZTR1 (p.(Glu345∗), heterozygous) in a child with congenital heart defects and adrenal cortical carcinoma.

P/LP variants in congenital anomaly genes

We identified 4 participants with P/LP variants in genes known to cause congenital anomalies but not previously implicated in cancer development. These include (1) 2 compound heterozygous variants in MMUT (p.(Arg108Cys) and p.(Arg228∗)) in a participant diagnosed with multiple renal and CNS anomalies and renal cell carcinoma (as separately reported from exome analysis)²⁶; (2) a pathogenic variant in FBN1 (p.(Gly2514Arg)) in a participant with joint hypermobility and rhabdomyosarcoma; (3) a P/LP variant in COL3A1 (p.(Arg1166∗)) in a participant with skeleton, heart, gastrointestinal tract anomalies, and rhabdomyosarcoma; and (4) a loss-of-function (LoF) variant in KAT6B (p.(Glu1057Glyfs∗10)) in a participant diagnosed with genitopatellar syndrome and neuroblastoma. Interestingly, a different LoF variant (p.(Gln1246∗)) in KAT6B was previously reported in a case report of a participant with similar clinical features of genitopatellar syndrome and neuroblastoma,²⁷ which motivated a more extensive analysis of the role of KAT6B in neuroblastoma as described below.

P/LP variants in cancer predisposition genes

Three participants carried P/LP variants in genes associated with pediatric cancer but not congenital anomalies (Table 2). These included 2 participants with inherited P/LP variants in TP53 (p.(Met246Ile) and p.(Leu265Pro)) who were diagnosed with ependymoma and rhabdomyosarcoma. Additionally, we identified a pathogenetic variant in PAX5 (p.(Arg140Leu)) in a participant with B-ALL and congenital vaginal atresia. PAX5 is a well-described B-ALL predisposition gene in addition to a frequent target of somatic mutation.³¹ The p.(Arg140Leu) variant has not been reported as a germline variant but instead observed in PAX5 B-ALL as a somatic second hit, frequently co-occurring with another germline variant (p.(Arg38His)).³¹ Involvement of these genes in the observed anomaly remains to be elucidated.

Role of KAT6B in neuroblastoma

As noted, because of the phenotypic similarities in the participant #11 in Table 2 and the previous report,²⁷ we hypothesized that KAT6B might be a novel neuroblastoma predisposition gene. To further evaluate this hypothesis, we performed a gene-based analysis using GS data from 409 neuroblastoma participants and 952 control samples obtained from the Gabriella Miller Kids First Pediatric Research Program. No KAT6B LoF variants were observed in cases or controls. However, the results revealed a significant enrichment of rare, predicted deleterious missense and inframe indel variants in the KAT6B gene (detailed in the Supplemental Table 2) among neuroblastoma participants after excluding any ClinVar LB/B variants (P = .017). Using different variant inclusion scenarios to estimate effect sizes, the odds ratios for neuroblastoma ranged from 2.35 to 4.69 (Figure 3). Although the 95% confidence intervals for several filtering criteria included the null, when we only included variants with a CADD >20, the odds ratio was 2.64, and the 95% confidence interval was 1.06 to 6.57.

Figure 3.

Odds ratios for KAT6B gene-based association analysis in neuroblastoma cases vs unaffected controls with different variant filtering criteria. B/LB, benign/likely benign variant; CI, confidence interval; Ctrl, control; del/dup, deletion/duplication; NBL, neuroblastoma; OR, odds ratio.

Discussion

Overall, in the GOBACK cohort of participants diagnosed with both congenital anomalies and pediatric cancer, 23.4% (11 of 47) harbored a germline P/LP variant in a gene previously associated with congenital anomalies and/or pediatric cancer. When focusing on solely well-established cancer predisposition genes in pediatrics, 14.9% (7 of 47) harbored a P/LP, which is consistent with previous assessments.^32,33 It is important to highlight that some of the previous research included pathogenic variants in genes associated with adult-onset cancers, which were not considered in this study.

Notably, among the 47 participants evaluated, only 4 had P/LP variants in genes previously reported to be associated with both congenital anomalies and pediatric cancer. These genes included: WT1, PTPN11, USP9X, LZTR1, 3 of which are autosomal dominant (AD) inheritance. However, only females with LoF variants in USP9X develop multiple congenital anomalies and B-ALL and the LZTR1 gene exhibits complex inheritance patterns, leading to variation in expressivity.³⁴ Both heterozygous and biallelic P/LP variants in LZTR1 have been associated with Noonan syndrome.³⁴ Additionally, LZTR1 is associated with schwannomatosis type 2 in an AD manner.³⁴ The participant, who harbored a P LZTR1 variant inherited from his father, was diagnosed with an unspecified congenital heart defects and adrenal cortical carcinoma. LoF variants in LZTR1 gene, such as the one observed here (p.(Glu345∗), have been observed in patients with both schwannomatosis (AD) and Noonan syndrome (autosomal recessive [AR]). Notably, even when considering AR forms of the disorder, individuals heterozygous for LZTR1 may have mild features of Noonan syndrome.³⁵ We located a different case featuring the same variant in Genotype to Mendelian Phenotype (Geno2MP, https://geno2mp.gs.washington.edu/), a database developed by the University of Washington Center for Mendelian Genetics that includes variants from Mendelian gene discovery projects and provides phenotype information for individuals with specific genotypes. The variant was identified in a participant with malformation of the heart and great vessels (hypoplastic left-heart syndrome), a condition similar to the phenotype of the participant in this study (structural developmental anomaly of heart or great vessels). As of now, there is no established evidence suggesting an association between LZTR1 and adrenal cortical carcinoma. However, there is emerging evidence that LZTR1 plays a role as a cancer predisposition gene,^36,37 and the list of cancer types may be expanded over time.

As noted, we did not observe any instances of dual molecular diagnosis, ie, 2 different P/LP variants, each related to cancer and congenital anomalies in a single participant. For many participants the P/LP variants can only explain congenital anomalies or pediatric cancer, although there is a possibility that these genes could influence both phenotypes. Because we report only P/LP variants with strong evidence in known disease genes, other yet unknown but potentially important variants may have been missed. It is possible that there are genes yet to be discovered that could explain the cooccurrence of congenital anomalies and pediatric cancer. For instance, PAX5 variants, associated with B-ALL, have unclear connections to congenital anomalies. Interestingly, a variant at the same amino acid position (p.(Arg140Gln) compared with p.(Arg140Leu) identified in a participant in this study) in the PAX5 gene has been cited as a causative variant in neurodevelopmental disorders, as recently reported by Gofin et al.³⁸ However, associations between PAX5 gene and congenital anal atresia (as seen in this study) have not been demonstrated.

Although pathogenic variants in KAT6B are known to cause genitopatellar syndrome, the association between variants in KAT6B and neuroblastoma represents a novel finding. We initially identified a LoF variant (p.(Glu1057Glyfs∗10)) in KAT6B in a participant with genitopatellar syndrome and neuroblastoma. As noted, there was a different LoF variant (p.(Gln1246∗)) identified in an individual with a similar set of phenotypes,²⁷ prompting us to further evaluate KAT6B as a neuroblastoma predisposition gene. KAT6B has a potential tumor-suppressive role, involving the acetylation of histone H3 at Lys23. Specifically, this hypothesis arose from the observation of KAT6B genomic loss in small cell lung cancer cell lines.³⁹ Additional research suggests its involvement in multiple cancers.⁴⁰

Although confirming a new cancer predisposition association requires multiple replication studies to ensure robust findings, our gene-based association analysis revealed that rare, predicted damaging missense and inframe indel variants in KAT6B confer an approximate 2- to 4-fold increase in neuroblastoma risk, suggesting KAT6B as a novel susceptibility gene for this cancer. Despite some confidence intervals not indicating a statistically significant odds ratio, our study, which used both the VAAST and traditional odds ratio methods, suggests a substantial association with these rare variants. VAAST, a novel computational approach, provides a more sophisticated analysis by differentially weighting variants based on bioinformatic predictions, which has proven superior in detecting statistical significance in rare variant studies.41, 42, 43 In contrast, traditional odds ratios rely solely on data from case-control groups and can be less informative in studies with small sample sizes. We included traditional odds ratio analysis to maintain alignment with standard practices in clinical research, thereby offering a comprehensive view of the significance of KAT6B variants. These estimates are exploratory but suggest that deleterious variants in KAT6B likely predict a higher risk of neuroblastoma.

Note that we observed no LoF variants in the association analysis, and it is possible that LoF variants in KAT6B are a rare cause of neuroblastoma. Further research is needed to explore these relationships, particularly considering the known association of KAT6B LoF variants with a subset of KAT6B disorders, such as Say-Barber-Biesecker-Young-Simpson syndrome, to avoid causing unwarranted concern among affected families.

Limitations of this study are, first, that although large in the context of overlap between congenital anomalies and pediatric cancer, the study population was relatively small and did not have recurring phenotypes, which limited the possibility of novel gene discovery. Second, although we conducted a thorough review of the GS results, in 36 of 47 participants we did not identify pathogenic variants that aligned with the observed phenotypes. This could be due to challenges in detecting mosaic variants, or variants located in complex genomic regions. Additionally, some variants are outside the scope of the analysis, such as those found in intergenic and intronic regions, where there is insufficient evidence to determine their pathogenicity. Third, because we relied on the data from questionnaires or electronic medical records for the congenital anomaly diagnoses, which were collected within a limited period, the access to updated or detailed phenotype information was challenging. Lastly, some variants lacked strong evidence under the variant classification guidelines set by ACMG/AMP and guidance provided by ClinGen and were interpreted as variants of uncertain significance (data not included). We anticipate that some of those variants might be reclassified as P/LP as more updated data become available.

Nevertheless, the GOBACK Study presents several notable strengths worth highlighting. First, the study adopted a population-based informed approach, leveraging congenital anomaly diagnoses. This approach offers valuable insights into the real-world prevalence and manifestations of congenital anomalies and pediatric cancer, enriching the context of the research. By incorporating these strengths, this article contributes significantly to the understanding of the genetic factors associated with these conditions, opening avenues for further exploration and potential therapeutic interventions. Second, it stands as one of the largest cohorts of children simultaneously affected by congenital anomalies and pediatric cancer to undergo GS analysis, providing a substantial and unique data set. This extensive cohort size not only enhances the robustness of the findings in this study but also underscores the importance of exploring the intricate interplay between these 2 conditions. Lastly, GS enables a comprehensive examination of the genome, allowing us to delve deep into the genetic underpinnings of these conditions and significantly increasing opportunities for discovering a wide range of novel variants, including splicing variants, noncoding variants, and CNVs.

In summary, the GOBACK Study establishes a molecular diagnostic yield of 23.4% (11 of 47) in participants with both congenital anomalies and pediatric cancer. Of the 11 diagnosed participants, 7 had molecular findings that could explain either congenital anomalies or cancer, and in 4 participants, the findings provided explanations for both conditions. Additionally, the results point to the role of KAT6B as a novel neuroblastoma gene. Furthermore, the findings in this study point to the importance of exploring the overlap between congenital anomalies and pediatric cancer in the discovery of novel predisposition genes, which could inform our understanding of development and carcinogenesis.

Data Availability

All data used in this analysis are accessible via the Genomics Research to Elucidate the Genetics of Rare diseases (GREGoR) dataset on NHGRI’s AnVIL platform (https://anvilproject.org/data/consortia/GREGoR).

Conflict of Interest

Sharon E. Plon is a member of the Scientific Advisory Panel of Baylor Genetics.