Between 2019 and 2024, children under the age of 18 years who developed leukemia/lymphoma and exhibited increased risk for cancer predisposition according to Jongmans’ criteria (Jongmans et al., 2016) were included in the study. All participants and/or their legal guardians provided written informed consent. Patients were recruited from eight pediatric hemato-oncology departments across hospitals in Türkiye. Additional clinical information and laboratory investigation results were systematically recorded in an in-house form. The samples and data generated in this study were archived in the Acibadem University Biobank Unit. An overview of the study design is presented in Figure 1. This study was approved by the Acibadem Healthcare Institutions Medical Research Ethics Committee (ATADEK; no. 2017-16/5 and no. 2024-11/496).

Peripheral blood or bone marrow samples were obtained during remission, and when available, buccal swab samples were collected from the index patients and their family members who were diagnosed with a hematological malignancy in childhood. For segregation analysis, blood samples were collected from the parents and other available family members. Germline DNA from remission blood or bone marrow samples was extracted using the QIAamp/QIAquick DNA extraction and purification kit (QIAGEN, Hilden, Germany). DNA from buccal swabs was extracted using the buccal swab DNA isolation kit (Hibrigen Biotechnology, Kocaeli, Türkiye). Genomic DNA from formalin-fixed paraffin-embedded samples was extracted using the QIAamp DNA FFPE Tissue Kit (QIAGEN, Hilden, Germany).

Standard paired-end short-read sequencing was performed on germline DNA samples. Whole-exome sequencing (WES) was performed using the NovaSeq 6000 system (Illumina, Inc., San Diego, CA, United States) with the S1 reagent kit v1.5. Clinical exome sequencing (CES), which covers exons and flanking intronic regions of 4,490 clinically relevant genes, was performed using the SOPHiA Clinical Exome Solution v2 kit (SOPHiA Genetics, Lausanne, Switzerland) in Umraniye Training and Research Hospital, Istanbul, Türkiye. Additionally, short-read RNA sequencing (RNA-seq) was performed for a single case (case #13). WGS with 30x coverage was conducted at Science for Life Laboratory (SciLifeLab; Stockholm, Sweden), as previously described (Stranneheim et al., 2021).

All sequencing data were subjected to bioinformatic processing, annotated, and filtered for candidate variants using the GenNext platform (https://app.gennext.bio/auth) (https://github.com/GenivaInformatics/gennext-workflows, Geniva, Istanbul, Türkiye). WGS data analysis was performed using the mutation identification pipeline (Stranneheim et al., 2021). Annotated and ranked variants in WGS data were further filtered and interpreted using the visualization tool Scout (https://github.com/Clinical-Genomics/scout). WGS was performed using the hg19 (GRCh37) reference genome, while WES was conducted using the hg38 (GRCh38) reference genome. The WGS data were analyzed using the ChiCaP clinical gene panel, which includes 189 well-established childhood cancer predisposition genes (Tesi et al., 2024). A detailed description of sequencing data analysis is provided in Supplementary File 1 and Supplementary Figures 1–2.

The Turkish Variome dataset was used to evaluate the variant allele frequencies of the candidate variants (Kars et al., 2021). Missense variants of unknown significance (VUS) were evaluated using variant analysis with multiple pathogenicity predictors (VAMPP), as described previously (Ozdemir et al., 2024). VAMPP scores higher than 0.35 indicated moderate evidence for pathogenicity (PP3) of variants (https://vamppscore.com/).

Moreover, we utilized AlphaFold (Jumper et al., 2021) to determine the predicted local distance difference test (pLDDT) scores and spliceAI (Jaganathan et al., 2019) for the splice site variants. All the candidate variants were validated, and segregation analyses were performed by Sanger sequencing for parental DNA samples using standard protocols. Additionally, ΔΔG values were calculated for missense variants with the DynaMut2 tool (https://biosig.lab.uq.edu.au/dynamut2/). A negative ΔΔG value indicates that the mutation is destabilizing, indicating a decrease in stability. In this context, the ΔΔG value was calculated for 15 missense variants.

The variant classifications were finalized with the combined analysis of these findings according to the American College of Medical Genetics (ACMG) (Richards et al., 2015).

We analyzed 36 individuals, including 20 index cases (diagnosis age median: 8 years, female/male: 11/9) and nine affected and seven unaffected family members. Seven out of 20 families (35%) reported consanguinity. Of the index cases, 14 were diagnosed with B-ALL (acute lymphoblastic leukemia), two with T-ALL, one with T-lymphoblastic lymphoma, two were Hodgkin’s lymphoma cases, and one was a non-Hodgkin’s lymphoma case (Table 1; Supplementary File 2).

Twenty index patients presented 30 candidate variants (Figure 2; Table 2). The candidate genes observed in the B-ALL cases were ABCC6, BCNP1, BMP6, BRAT1, BRCA1, DNHD, ETAA1, ETV6, JAK2, JAK3, KSR1, MAP2K2, MLLT10, MUTYH, MYH11, MVP, PAPSS2, RAD52, RECQL, TGFBRS1, TP53, and WRN. The candidate genes observed in the T-ALL cases were ATR, BRCA2, BIRC6, and NOTCH1, and those observed in lymphoma cases were MLH1, MSH6, RAD52, and TNFRSF9 (Figure 3).

AlphaFold provided pLDDT scores for 28 variants (DNHD1 and BIRC6 were not available), and 13 of these (TP53, BMP6, JAK2/3, MYH11, MVP, RAD52, WRN, MUTYH, TGFBRS1, RECQL, ABCC6, and PAPSS2) had very high scores. To assess the impact on protein dynamics and stability, we analyzed the stabilization status using DynaMut2 (https://biosig.lab.uq.edu.au/dynamut2/). Eleven missense variants were found to be destabilizing (BMP6, JAK2, MYH11, ATR, NOTCH, RAD52, WRN, MUTYH, RECQL, BRCA2, and ABCC6) (Supplementary File 3).

Additionally, 15 missense variants were also evaluated by the AlphaMissense tool, and eight of them had a score of >0.5 (Supplementary File 1). Two splice-site variants (ETV6 and BCNP1) were at the essential splice site (Supplementary file 2). The spliceAI algorithm yielded higher than 0.5 scores of TNFRSF9 c.413 + 6T>C variant in case #13, and the RNA sequencing of the index revealed an alternative splice site between exons 5 and 6 (Supplementary File 2.15–2.16).

Seven candidate variants (MAP2K2, BMP6, DNHD1, RAD52, MUTYH, RECQL, and ABCC6) presented allele frequencies <0.001 in Turkish Variome (Table 2).

Altogether, based on all the provided evidence, 10 variants were classified as pathogenic (P), 11 were likely pathogenic (LP), and nine were VUS. Eleven out of the 16 missense variants were suitable for VAMPP score evaluation, and four of them (MYH11, MUTYH, TGFBRS1, and ABCC6) showed scores >0.35 but did not change the variant classification of VUS (Table 2; Figure 2).

In the 13 index cases, we hypothesized that the candidate variants caused cancer predisposition. The detailed information, pedigrees, and genotypes of the cases and variants are provided as case vignettes in Supplementary File 2.

Index case #1 (Supplementary File 2.1) was diagnosed with B-ALL, and we determined a germline pathogenic stop-gain variant in TP53 c.437G>A p.Trp146Ter. The patient successfully completed treatment for B-ALL in 2019; however, she was diagnosed with acute myeloid leukemia (AML) in 2021. In index case #2 (Supplementary File 2.2), who had B-ALL, we found a maternally inherited LP ETV6 c.464-2A>G variant. It was also detected in the patient’s affected brother, who died of AML after a T-ALL diagnosis.

The LP MAP2K2 c.907C>T p.R303C and LP BMP6 c.1007G>A p.Gly336Glu variants were detected in B-ALL index case #3 (Supplementary File 2.4) and his brother with Hodgkin’s lymphoma. In addition, our case presented elevated iron levels. Their parents were first-degree cousins and were heterozygous for both variants.

In index case #4 (Supplementary File 2.5), the diagnosis was of B-ALL. The patient’s father and one paternal uncle had been treated for Hodgkin’s lymphoma; they suffered relapse after 10 and 20 years, respectively, after the first diagnosis, and additional thyroid papillary carcinoma and synovial sarcoma were diagnosed in the father. The patient’s parents were consanguineous, and all the affected family members were positive for the likely pathogenic variant ETAA1 c.2565_2568del p.K855fs and the pathogenic variant DNHD1 c.7888C>T p.R2630X, whereas the mother and the siblings, who were unaffected, were negative.

Index case #5 (Supplementary File 2.6) was found to be homozygous for the pathogenic variant MSH6 c.3934_3937dup p.Ile1313SerfsTer7. T-ALL was diagnosed when the patient was 12 years old, and 3 years later, primary AML and astrocytoma were diagnosed. Their parents were first-degree cousins, and both were heterozygous for the candidate variant.

We hypothesized that the VUS ATR c.5273G>C p.Gly1758Ala was the causative factor in case #6 (Supplementary File 2.7), who had T-ALL. Both the patient and his brother, who also had T-ALL and died due to treatment toxicity, were homozygous for this variant. The buccal swab analysis confirmed the homozygous status of the variant in the index case. A year later, the index case developed a low-grade glioma. The parents were first-degree cousins, and both were heterozygous for this variant. The patient was also heterozygous for the LP NOTCH1 c.2929G>A p.Gly977Arg and homozygous VUS BIRC6 c.6124A>G p.Asn2042Asp variants. NOTCH1 was also heterozygous in the affected brother and the father, but the mother was wild-type (WT), whereas BIRC6 was heterozygous in the mother and homozygous in the index. The affected brother’s DNA material was insufficient to check the BIRC6 variant.

In index case #7 (B-ALL, Supplementary File 2.8) and in the mother (Hodgkin’s lymphoma), the pathogenic variant present was WRN c.1721G>T p. Gly574Val was detected. His parents were second-degree cousins. B-ALL was diagnosed in index case #8; AML had been diagnosed in her father. They were positive for the LP MYH11 c.1291C>T p.Arg431Cys variant. The family reported members on the paternal side who had died of lymphoma and osteosarcoma.

We found heterozygosity for the LP MVP c.2296C>T p.Arg766X and IL32 c.515_516insG p.Asp172Glufs* variants in both the index case #11 (Supplementary File 2.12) and her mother, in whom colon cancer had been diagnosed at an early age. The parents were consanguineous, and bladder cancer had been diagnosed in the father’s paternal uncle.

Hodgkin’s lymphoma was diagnosed not only in index case #13 (Supplementary File 2.14) but also in his brother, his father, and two of his paternal uncles. WGS revealed homozygosity for the likely pathogenic variant TNFRSF9 c.413 + 6T>C in all the affected cases, and the variant was comprehensively evaluated by RNA sequencing for possible alternative transcripts. The variant was confirmed in the buccal swab of the index patient. The parents were consanguineous, and the mother was heterozygous for the variant.

WES revealed heterozygosity for the pathogenic variant BRAT1 c.1007delGG336fs in index case #15 (Supplementary File 2.19), who had B-ALL. The mother was also heterozygous for the same variant, whereas the maternal aunt, who had had ovarian cancer, and the maternal grandmother, who had breast cancer, were WT. Subsequent WGS analysis revealed the LP KSR1 c.527A>G p.Glu176Gly in all affected family members, and this variant was, therefore, considered causative. Another clinically relevant variant was pathogenic MLH1 c.1459C>T p.Arg487Ter. We determined that index case #17(Supplementary File 2.21), who had non-Hodgkin’s lymphoma, was heterozygous for this variant, as was her unaffected father. The father’s mother, maternal aunt, and maternal grandmother had died of breast cancer.

We also determined the heterozygous pathogenic BRCA1 c.5266dup p.Gln1756ProfsTer74 variant in both the index case #18 (Supplementary File 2.22) and her mother by CES analysis. The maternal grandmother had died of breast cancer at the age of 45; the mother’s father was WT, which confirms that the variant was maternally inherited by the patient’s mother.

For case #9, case #10, case #12, and case #19 (Supplementary File 2.10, 2.11, 2.13, and 2.23, respectively), the analysis revealed pathogenic JAK2, likely pathogenic JAK3, VUS RAD52, and VUS BRCA2 variants, respectively, but only the index case’s samples were available, and further analysis could not be performed within the families. The evidence was insufficient to support the potential causative effect of these variants, and they remained unsolved (Table 2). Case #14 was diagnosed with B-ALL, and soon after the initial diagnosis, brain metastasis occurred, and the patient died following the development of renal cell carcinoma. The WGS analysis of the peripheral blood sample revealed candidate variants in BCNP1, RECQL, TGFBRS1, PAPSS2, and MLLT10, but none showed strong enough evidence to be considered causative. Likewise, in case #16, an incidental pathogenic heterozygous MUTYH variant was identified. However, pathogenicity in MUTYH is typically associated with bi-allelic variants, particularly in the context of MUTYH-associated polyposis. This variant was heterozygous and not present in the affected cousin. Furthermore, the family had no history of gastrointestinal symptoms suggestive of polyposis. Therefore, we did not consider this variant to be related to leukemia predisposition. Although we detected candidate variants, we found no strong evidence for variants detected in cases #14, #16 (Supplementary File 2.20), and #20 (Supplementary File 2.24), and they remained unsolved.

For the clinical utility of these results, we evaluated the genetic findings with the patient’s primary hematologist/oncologist. For instance, in case #1, the TP53 variant changed the diagnosis to Li–Fraumeni syndrome, and the family was referred to a clinical geneticist for counseling and surveillance plan. In case #5, the homozygous MSH6 variant changed the diagnosis from neurofibromatosis to constitutional mismatch repair deficiency (CMMRD), and the family was referred to a clinical geneticist for counseling and surveillance plan for the risk of Lynch syndrome. Our findings also led to a change in the treatment or donor candidate. In cases #2, #3, #5, #6, and #7, the unaffected donor siblings were also found to be carriers of the candidate genes, and the transplant planning was changed to be performed either from the WT siblings or unrelated donors.

Genetic counseling was provided by clinical geneticists at the referral centers, in accordance with the NCCN Guidelines for genetic/familial high-risk assessment (https://www.nccn.org/guidelines/category_1). For index cases with established cancer predisposition syndromes—such as Li–Fraumeni syndrome (TP53) or constitutional mismatch repair deficiency (CMMRD, MSH6)—counseling included a structured discussion of the associated cancer risks, current clinical guidelines, and established surveillance protocols (e.g., whole-body MRI for Li–Fraumeni and colonoscopy for CMMRD). Families were offered cascade testing and referred to relevant specialty clinics, including adult genetics services, to ensure long-term coordination of care.

For variants of uncertain significance or novel candidate variants without established disease associations or management protocols, clinical geneticists focused on transparent communication regarding the current level of evidence, limitations in interpretation, and potential—but not definitive—implications for cancer risk. These sessions emphasized the importance of regular phenotypic monitoring and re-evaluation as new data become available. Surveillance recommendations in these cases were tailored based on the family history, clinical phenotype, and expert consensus. To support clinical integration of the genetic results, all cases with potential diagnostic or therapeutic relevance were presented at multidisciplinary tumor boards, where available. These boards typically included pediatric oncologists, hematologists, pathologists, and clinical geneticists. Discussions focused on treatment implications, variant reinterpretation, and potential surveillance strategies for both the index cases and at-risk relatives. This collaborative setting enabled shared decision-making, particularly in complex or uncertain scenarios.

In centers lacking access to a clinical geneticist, the patient’s primary physician assumed the responsibility of conveying results and implementing follow-up care, supported by structured reports and consultation with the central study team. Families with asymptomatic carriers were counseled on possible future risks and the need for longitudinal follow-up, while psychosocial concerns were addressed through individualized discussions.

Overall, the genetic counseling process—supported by multidisciplinary input—was critical in translating germline findings into personalized clinical action, guiding diagnosis, treatment, and long-term risk management for families affected by childhood hematologic malignancies.

The STRING interaction network within NetworkAnalyst revealed that 26 genes were interconnected either directly or indirectly. Separate network analyses were conducted for all candidate and causative genes (Supplementary Figure 3a, b), as well as for genes associated specifically with leukemia and lymphoma (Supplementary Figure 3c,d). TP53 was both the central hub and the most connected gene in the network. In the leukemia-specific network, TP53, NOTCH1, JAK2, and BRCA1/2 were found to be key hub genes; in the lymphoma-specific network, MLH1 was identified as a key hub gene. Of the 18 causative genes, nine (TP53, ATR, ETAA1, BRCA1, BRCA2, BRAT1, WRN, MLH1, and MSH6) were identified as being responsible for DNA repair and damage response (Figures 4a,b).