Childhood acute myeloid leukemia shows a high level of germline predisposition

TO THE EDITOR:

As germline variants can influence cancer patient treatment decisions, outcomes, and counseling, and as the level of genetic predisposition for sporadic childhood acute myeloid leukemia (AML) is not clearly established, we undertook a comprehensive analysis of rare germline variants in childhood AML. As childhood AML is rare,¹ to date, pan-cancer childhood cohorts have included few AML cases, and often the germline panels used have not included key genes relevant to myeloid malignancy. We therefore combined data from the Therapeutically Applicable Research to Generate Effective Treatments (TARGET) program together with an Australian childhood AML cohort (supplemental Tables 1 and 2, available on the Blood Web site) to identify the rare germline variants in a large panel of cancer predisposition genes (n = 216) compiled from literature review, and including genes involved in familial hematological malignancies (HMs) and bone marrow failure syndromes (supplemental Table 3). We analyzed whole-genome sequencing and whole-exome sequence data available through the TARGET program (n = 48) (phs000218.v22.p8.c1) and whole-exome sequence data for the Australian cohort (n = 24). Given that damaging and disease-causing variants are predicted to have a low population prevalence, we identified extremely rare, potentially deleterious germline variants (variant allele frequency [VAF] > 30%; minor allele frequency [gnomAD] <0.001; combined annotation dependent depletion score >10) and classified these as shown in Figure 1. All variants passing initial filtering are listed in supplemental Table 4. The distribution of germline and somatic variants is shown in supplemental Table 5 and supplemental Figure 1. Given the small cohort size, pairwise comparisons of germline variants and clinicopathological characteristics did not reveal significant associations after applying multiple correction (supplemental Table 6).

We compared the prevalence of extremely rare variants in the combined childhood AML cohort (n = 72) to that in the Medical Genome Reference Bank^2,3 (MGRB; n = 2570) comprising individuals aged at least 70 years with no history of cardiovascular disease, dementia, or cancer. To avoid bias, rare variants in both the AML and the MGRB cohorts were selected by comparable filtering procedures, based on variant rarity in the independent gnomAD cohort. We conducted statistical tests on multiple randomized subsamples (n = 6000) of both cohorts, enabling fair testing between cohorts of different sizes. This showed a significant increase of rare alleles in childhood AML for 100% of subset comparisons (P <.05; Mann-Whitney U test; supplemental Figure 2), and this trend was maintained when analysis was restricted to the European-ancestry subset of patients (79.2% P < .05).

As >200 patients are needed to conduct Burden testing,⁴ we performed odds ratio (OR) analysis for genes in which variants were observed in ≥4 of 72 patients with AML (5.5%; 10 genes). For the MGRB cohort, we assumed 1 extremely rare variant per gene per individual. This revealed increased odds of rare variants in RUNX1 occurring in childhood AML (OR, 15.06; P = .0004; supplemental Table 7). As the MGRB cohort is 97% non-Finnish European,³ we also performed a subanalysis with only European ancestry patients with AML after first confirming no high-degree patient relatedness, which may skew these results. This analysis provided strong evidence that ancestry is not driving the increased odds of rare RUNX1 variants (n = 47; OR, 23.81; P ≤ .0001). Of the 4 patients in our cohort with rare variants in RUNX1, 2 from the TARGET cohort harbored pathogenic mutations linked in the ClinGen resource to familial platelet disorder with associated myeloid malignancy; however, as we did not have access to family history, we cannot provide definitive evidence for familial cancer risk. The other 2 patients harbored extremely rare RUNX1 missense variants, including the previously reported pathogenic variant, p.Asp198Asn.⁵ 

We next applied additional stringent filtering to determine predicted damaging germline predisposition variants. First, we considered variants that are (1) annotated as pathogenic or likely pathogenic in the ClinVar⁶ database, or (2) frameshift indels, stop gain, or splice acceptor/donor site variants in tumor suppressor genes, which are likely to result in loss of protein function. As matched nontumor material was not available for the Australian cohort, we have used caution for variants from this cohort and excluded those that have been recurrently reported as somatic in HM (see supplemental Methods). Overall, 17 patients were identified (24%; 21% of the Australian cohort and 25% of the TARGET cohort) with a total of 18 variants across 15 genes (supplemental Table 8). This frequency is substantially higher than that reported with cancer predisposition gene sets and similar variant classification criteria in a pan-cancer childhood cohort (7% to 8%⁷; Figure 1) and adult AML (9%⁸). The frequency of predicted damaging germline predisposition variants in childhood AML may be higher than that estimated, as we identified a further 3 germline variants in the TARGET cohort (supplemental Table 9) by analyzing mutation hotspots in additional genes that are somatically mutated in myeloid malignancies (supplemental Table 10), including a heterozygous mutation affecting DNMT3A Arg882 (patient PARBTV; germline specimen bone marrow–derived fibroblasts); such mutations are associated with Tatton-Brown-Rahman syndrome^9,10 for which an adolescent case of AML has been reported.¹¹ 

Finally, we assessed the 18 rare variants identified by our analysis above (supplemental Table 8), and the 3 additional hotspot variants from the TARGET cohort (supplemental Table 9), for pathogenicity using the 2015 American College of Medical Genetics and Genomics and the Association for Molecular Pathology (ACMG-AMP) criteria.¹² Fifteen variants in 14 patients (19.4%; 21% of the Australian cohort and 19% of the TARGET cohort) were classified as pathogenic or likely pathogenic using these criteria (Table 1). RUNX1 mutations may be particularly important for sporadic childhood AML, as 2 patients harbored pathogenic mutations (p.Arg166Gln and p.Tyr287*) that would fit into the recently defined World Health Organization category “myeloid neoplasms with germline predisposition.”¹³ The overall frequency of predisposition variants when ACMG-AMP criteria are used is considerably higher than that reported in other childhood cancer studies (7% to 10% overall^14-16) and for most adult cancers (8% overall¹⁷; Figure 1), although in a similar range to that reported recently for high-risk pediatric cancers (16.2%)¹⁸ and specific childhood tumors.^16,19 We acknowledge that analyses of the frequency of predicted and classified pathogenic variants across different cohorts are not directly comparable, as differences or improvements in variant calling may contribute to the observed elevated frequency.¹⁸ Interestingly, apart from RUNX1, we did not identify pathogenic or likely pathogenic variants in several established familial myeloid malignancy genes (GATA2, DDX41, SAMD9, SAMD9L, ETV6, or CEBPA), or in genes associated with familial childhood syndromes that confer risk of HM (TP53, SBDS, RAS pathway genes). A pathogenic variant of interest was detected in XRCC2/FANCU, associated with Fanconi anemia, which confers a highly elevated risk of HM in early life.²⁰ Some predisposition genes for which we detected predicted pathogenic germline variants in TARGET patients have not previously been associated with HM (eg, GJB2 and FLCN).

The key finding from this study is that the frequency of rare germline cancer predisposition variants in newly diagnosed childhood AML is higher than suggested previously,²¹ indicating a clinically significant risk in an important fraction of patients. The detection of such a germline mutation at diagnosis is an important consideration for stem cell donor selection, and genetic counseling for the family and patient. Further studies in larger, independent cohorts with detailed family history are now warranted to clarify the role of germline predisposition variants in childhood AML.