For this study, we used publicly available whole exome sequencing datasets previously collected across studies at St. Jude Children’s Research Hospital or through dbGaP. Specifically, we used the Pediatric Cancer Genome Project (PCGP) (15), real-time clinical genomics (RTCG/G4K) (16), St. Jude Lifetime Cohort (SJLIFE) (17), and TARGET datasets (TARGET URL is https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000218). In sum, we interrogated 5,993 germline samples across 24 cancer types including hematologic, non-central nervous system (CNS) solid tumors, and CNS tumors. Cancers were stratified into hematologic (n = 3,452; 58%), solid (n = 1,974; 33%) or CNS (n = 1,068; 18%) cancers and further subclassified as follows: i) hematologic malignancies: B-cell (B) acute lymphoid leukemia (B-ALL), T-cell ALL (T-ALL), acute myeloid leukemia (AML), Hodgkin’s lymphoma, and non-Hodgkin’s lymphoma; ii) solid tumors: germ cell tumor (GCT), melanoma (MEL), neuroblastoma (NBL), nasopharyngeal carcinoma (NPC), papillary thyroid carcinoma (PTC), sarcomas (Ewing’s (EWS), osteosarcoma (OS), rhabdomyosarcoma (RMS), synovial), Wilms tumor (WT); and iii) CNS tumors: ependymoma (EP), low grade glioma (LGG), medulloblastoma (MB), and high grade glioma (HGG). An external cohort was queried, which consisted of 41 patients with AML from the German Study Alliance Leukemia that met following criteria: age below 35, blast-free remission after chemotherapy, karyotype aberrations (n = 12 with < 3, n = 29 with ≥ 3 aberrations detected in diagnostic karyotype or FISH analysis), and samples of peripheral blood or bone marrow at remission (18). The current study was approved by the Institutional Review Board at St. Jude Children’s Research Hospital.

Variant calling and genotyping were performed using Genome Analysis Toolkit’s (GATK) best practices workflow with modifications as described previously (19). We retained high quality variants that passed filtering using following criteria: allelic balance > 0.2, genotype quality > 20, variant allelic frequency (VAF) for heterozygous variants between 20-80%, minimum of 10 alternate reads supporting single nucleotide variants (SNVs) and 7 alternate reads supporting InDels, and missingness < 25% of samples. We performed variant annotation using ANNOtate VARiation (ANNOVAR) and variant effect predictor (VEP) tools (20). We also annotated all the variants using InterVar (21) automated clinical interpretation based on the American College of Medical Genetics and Genomics (ACMG) guidelines (22).

We retained coding variants in the RPA heterotrimer (RPA1, RPA2, and RPA3 genes) with genome aggregation database (gnomAD) non-cancer cohort allelic frequency (AF) of < 0.5% (23) of the following classes: missense, frameshift insertions and deletions, stop gain, and splice site. We further filtered to retain missense variants with a computed Combined Annotation-Dependent Depletion (CADD) (24) Phred score > 15.

We performed local coordinate minimization followed by global side-chain optimization with the Atomic Multipole Optimized Energetics for Biomolecular Applications (AMOEBA) polarizable force field (25) on 5 high resolution structures of RPA fragments collectively comprising 7 modular domain of RPA heterotrimer. These included X-ray structures of the DNA binding domains A and B, DBD-A and DBD-B (PDB: 1JMC) (26), and the RPA trimerization core composed of DBD-C, D and E (PDB: 1L1O) (27) and NMR structures of the DBD-F (5N8A) (28) and the wing helix domain (PDB: 1DPU) (29). Prior to minimization, the ssDNA was removed from the 1JMC structure and bound peptides were removed from the 2 NMR structures. We then used our optimized structures to predict protein stability differences ΔΔG_Fold (DDG untrained (DDGun)) (30). DDGun estimates the ΔΔG_Fold of missense variants from a linear regression of sequence and biochemical features determined from the protein structure. Destabilizing ΔΔG_Fold values indicate a decrease in the ratio of folded to unfolded protein due to the mutation (we define negative ΔΔG_Fold values as stabilizing and positive ΔΔG_Fold values as destabilizing). We established ΔΔG_Fold cut-offs for mutations highly likely to impact protein folding. Our cut-offs were determined based on a ΔΔG_Fold that affects the ratio of folded to unfolded protein 12-fold (~1.5 kcal/mol) for both stabilizing and destabilizing mutations.

We performed rare-variant burden tests for RPA1, RPA2, RPA3 variants using 5,993 cases from all pediatric cancers in our cohorts (pan-cancer) and within each sub-class of cancers, namely, hematologic (n = 3,452), solid (n = 1,974), and central nervous system CNS (n = 1,068) malignancies. For the control set, we retrieved all variants across RPA1, RPA2, and RPA3 from gnomAD v2 non-cancer subset containing 134,187 individuals with no reported malignancy (23). All variants from control dataset were processed through the same variant annotation and filtering workflow as our cancer cohort (AF < 0.5%). Enrichment tests for cases with and without germline ultra-rare (AF < 0.005%) plus novel (AF 0%) and rare (AF < 0.1%) variants in the 3 genes were performed using both two- and one-sided Fisher exact tests using the statistical package R (v4.3) described in previous studies (31, 32). We used Bonferroni correction to adjust for multiple testing with a significance cutoff of adjusted p-value of < 0.05.

Within the pan-cancer cohort, we identified 80 cases with 55 germline heterozygous RPA1, RPA2 or RPA3 variants meeting criteria of AF < 0.5% in gnomAD non-cancer cohort and CADD score > 15 for candidate variant selection ( Figure 1A ). Specifically, 40 RPA1, 7 RPA2 and 8 RPA3 unique heterozygous germline variants were identified in 63, 7 and 10 cases, respectively ( Figure 1B ). All variants were classified as variant of uncertain significance (VUS) according to the ACMG criteria ( Tables 1 – 3 ). Majority of the variants (92% of RPA1, 71% of RPA2 and all RPA3 variants) had CADD scores > 20, indicating a higher probability of a deleterious effect ( Tables 1 – 3 ). In addition, looking at variant burden in population, we found that 98% (54/55) of the identified RPA heterotrimer variants had AF < 0.1% (this includes rare, very-rare, ultra-rare, and novel variants, Figure 1B ). All RPA1, RPA2 and RPA3 variants are mutually exclusive and no cases with compound heterozygous or homozygous variants were identified.

RPA1 (616 amino acids, 70kDa) is the largest of the 3 subunits of the RPA heterotrimer. We discovered 1.05% (63/5993) of the cohort to harbor heterozygous germline RPA1 variants ( Figure 1C ; Tables 1 – 3 ), which was statistically not significant compared to gnomAD non-cancer controls for all cancers and cancer subtypes ( Table 4 ). RPA1 has 4 modular oligosaccharide binding-fold domains commonly referred to as functional DNA binding domains (DBD): F, A, B and C spanning the N- to C- terminal regions of the protein. RPA1 variants were found across all 4 DBDs as follows: 6 in DBD-F, 15 in DBD-A, 10 in DBD-B, and 26 in DBD-C ( Figure 1C ). Of note, 6 cases were found to have RPA1 variants in the linker regions between 2 DBDs. All RPA1 variants were missense ( Figure 1C ) except for p.L53lfs*53 within DBD-F, which was found in 1 case. Three recurrently mutated amino acids were discovered in RPA1 domains DBD-A (p.V286, 9 cases), DBD-B (p.R389, 5 cases), and DBD-C (p.G437, 5 cases). We next focused specifically on novel and ultra-rare RPA1 variants (33), present in 14 and 21 cases, respectively ( Figure 1B ; Tables 1 – 3 ). Notably, we found significant enrichment of RPA1 novel and ultra-rare variants in our cohort (adjusted p-value < 0.05, Table 4 ).

Prediction of variant structural effect was performed by calculating protein stability change scores (ΔΔG_Fold) with a ΔΔG_Fold that affects the ratio of folded to unfolded protein 12-fold (~1.5 kcal/mol) for both stabilizing and destabilizing mutations. Significant scores (>1.5 kcal/mol) were demonstrated for 4 variants (p.M46T in DBD-F, p. R234G in DBD-A, p.W361L in DBD-B, p.V594G in DBD-C) which were novel or ultra-rare ( Tables 1 – 3 ). RPA1 p.M46T is likely to destabilize folding of DBD-F resulting in the loss of multiple important protein-protein interactions (11). W361 is a key DNA binding residue in DBD-B and human cells with W361A support normal replication but are deficient in DNA repair (12, 34), suggesting that p.W361L may destabilize DBD-B folding resulting in hypomorphic RPA.

We next assessed which types of malignancies were present in patients with RPA1 variants ( Figure 1A ). We found comparable frequency of cases with RPA1 variants across solid tumors (n = 22, 1.1%), CNS cancers (n = 12, 1.1%) and hematological malignancies (n = 29, 0.8%). Among solid tumor cases with RPA1 variants, 31.8% (7/22) presented with sarcomas and 27.3% (6/22) were diagnosed with neuroblastoma. Notably, the 7 sarcoma cases carried 6 unique RPA1 variants (n = 2 novel and n = 1 ultra-rare) and only one was noted to have a concomitant germline mutation ( Table 2 ). All 6 neuroblastoma cases were found to have unique RPA1 variants (n = 3 novel, n = 2 ultra-rare) of which half were found to have germline variants reported in PALB2/NDRG4, MDC1, or TP53 genes ( Table 2 ). Three cases of retinoblastoma harbored unique RPA1 variants (2 ultra-rare) with 2 cases having concomitant germline RB1 mutation ( Table 2 ). Two cases of Wilms tumor were identified to have germline RPA1 variants. Among the single cases of solid tumors (germ cell tumor, melanoma, nasopharyngeal carcinoma, and papillary thyroid carcinoma), 2 ultra-rare and 1 novel RPA1 variants were found.

Among cases with CNS tumors, 4 patients with medulloblastoma harbored novel (n = 3) or ultra-rare (n = 1) RPA1 variants. Each of these cases also carried other germline mutations (PBRM1, C7 and MYH9, BRCA1, ANKRD26) of which BRCA1 and ANKRD26 are cancer predisposition genes ( Table 3 ). Furthermore, 4 cases with high grade glioma harbored 3 RPA1 variants (n = 1 novel, n = 1 ultra-rare), all clustering within DBD-C domain of RPA1. These patients had no other potentially causative germline variants reported in other predisposition genes. Among the 3 low grade glioma, 3 RPA1 variants (one ultra-rare) were identified, with one harboring other germline mutations in SDHA and RUNX1. Lastly, one ultra-rare RPA1 variant was identified in a case of ependymoma without other germline mutations ( Table 3 ).

From patients with hematologic malignancies, RPA1 variants were most common in B-ALL (n = 13), followed by lymphoma (n = 7), AML (n = 5), and T-ALL (n = 4) ( Table 1 ). Out of 13 B-ALL cases, 3 and 5 were novel and ultra-rare, respectively. Only 2 cases out of the 13 had heterozygous germline variants in cancer predisposition genes (RAD51D, BRIP1). Among lymphomas, we observed 4 Hodgkin’s lymphoma (n = 1 novel, n = 1 ultra-rare) and 3 non-Hodgkin’s lymphoma (n = 1 novel, n = 1 ultra-rare) with RPA1 variants. In the AML sub-cohort, we found 3 unique RPA1 variants in 5 cases, of which 4 were mutated at nucleotide 856 in DBD-A (c.856G>T, c.856G>A coding different amino acids) and 1 ultra-rare variant in DBD-F domain ( Table 1 ). Of the 5 RPA1-mutated AML cases, 3 carried other germline variants (MLL, HIP1, NOTCH2 and FANCD2). Four unique RPA1 variants (n = 2 ultra-rare) were discovered in 4 patients with T-ALL, with one case having additional germline ATM variant ( Table 1 ).

Given the occurrence of MDS/AML in one prior patient with germline RPA1 p.V227A with TBD (10) and 5 AML cases in this study, we queried a cohort of 41 young adults with AML and karyotype aberrations (18) for RPA heterotrimer germline variants. We found 1 ultra-rare (c.460G>A, T154A, AF 0.001%) and 2 rare (c.1397C>G, A466G, AF 0.027%; c.1538G>A, R513H, AF 0.016%) RPA1 heterozygous variants ( Supplemental Table 1 ).

RPA2 is the second largest subunit (270 amino acids, 34kDa) of the RPA heterotrimer. We identified 6 heterozygous germline RPA2 variants in 7 cases of pediatric malignancies. Five variants are present in DBD-D ( Figure 1D ) and did not exhibit dysfunctional protein folding scores ( Tables 1 , 2 ). All variants were either ultra-rare (n = 5) or novel (n = 1). Four patients (4/3452, 0.1%) had hematological malignancies (n = 1 B-ALL, n = 1 T-ALL, n = 2 Hodgkin’s lymphoma) and 3 had solid cancers (n = 1 RBL, n = 1 neuroblastoma, n = 1 sarcoma). Other germline mutations were noted in 3 out of 7 cases ( Tables 1 , 2 ).

RPA3, although less than half the size of RPA2 (121 amino acids, 14kDa) had 8 unique germline heterozygous variants (n = 4 ultra-rare, n = 3 novel) in 10 cases of pediatric cancers, including 6 hematologic (B-ALL n = 3, T-ALL n = 1, AML n = 1, Hodgkin’s lymphoma n = 1), 3 solid tumors (Wilms tumor n = 2, sarcoma n = 1) and 1 CNS (medulloblastoma) cancers ( Figure 1E ; Tables 1 – 3 ). All were missense except for one frameshift (p.M40Cfs*16). Protein folding scores for 2 out of 7 available RPA3 variants were greater than 1.5 kcal/mol and were either novel or ultra-rare ( Tables 1 – 3 ). Half of RPA3 mutated cases had other germline variants noted ( Tables 1 – 3 ). The number of cases with RPA2 or RPA3 germline variants did not reach statistical significance compared to gnomAD non-cancer controls ( Table 4 ).