With the help of the German Childhood Cancer Registry, 20 individuals who survived a childhood malignancy and then developed a second primary cancer (2N) and 20 carefully matched (first tumor, manifestation age, sex) individuals who survived a childhood cancer without developing a second malignancy (1N) were recruited for the KiKme study (Cancer in Childhood and Molecular Epidemiology) (Marron, 2017[55]). Twenty matched patients (sex and age) without cancer from the Department of Accident Surgery and Orthopedics served as controls (0N). Participants were informed in writing and verbally of the procedures and purpose of this study. Signed informed consent documents were obtained from both patients and healthy individuals after genetic counseling. This study was approved by the Ethics Committee of the Medical Association of Rhineland-Palatinate (no. 837.440.03 [4102]; no. 837.262.12 [8363-F]; and 837.103.04 [4261]).

All patients were followed up from primary cancer diagnosis to recruitment for this study several years after treatment. Patient characteristics can be found in Supplementary data Table 1. The corresponding tumor samples were not available.

Affymetrix array SNP analysis (Weis et al., 2011[73]) of 1N and 2N patients did not reveal detectable copy number variations in tumor suppressor genes (BRCA1, BRCA2, RAD9A, TP53, NF1) or oncogenes (PTPN11, ETV6-RUNX1, TCF3-PBX1). No pathogenic germline mutations in TP53, RAD9A, BRCA1, or BRCA2 were identified by Sanger sequencing using the ACMG criteria (TP53 Mutation Database (http://p53.iarc.fr/); https://www.ncbi.nlm.nih.gov/clinvar/ and https://www.lovd.nl/). RB1 gene analysis in patient 1N20 did not reveal a pathogenic mutation. Using bisulfite pyrosequencing, none of the matched (1N or 2N) childhood cancer patients of this study showed BRCA1 hypermethylation, consistent with an epimutation (Galetzka et al., 2012[24]). Cell lines 0N24 and 2N24 were derived from a pair of discordant monozygotic twins. One twin suffered from childhood leukemia and later on from thyroid carcinoma, whereas her sister is completely healthy to date.

Bone marrow samples (excess material from routine chromosome diagnostics) were obtained from 27 patients with NHL (19 males, 8 females, aged 58.0±12.0 years), 27 AML (16 males and 11 females, aged 60.0±11.0 years), one pre-B-ALL (46,XY,t(9;22)(q34;q11.2)[2]/46,XY[25], one AML (46,XY,der(7)(q-).ish del(16)(q22)[12]/ 46,XY[10], and one plasmablastic lymphoma (PBL) with complex aberrations in 60 % of cells, consisting of a hypodiploid clone (28 %) and a hyperdiploid line (32 %). The analysis of the samples was approved by the Ethics Committee of the Medical Association of Rhineland-Palatinate (no. 2019-14677).

The FaDu tumor cell line (from hypopharyngeal carcinoma), carrying mutations in CDKN2A (c.151-1G>T), TP53 (c.376-1 G>A; c.743G>T) and SMAD4 (c.1_1659del1659) was purchased from ATCC and characterized using SNP-array analysis in 24.05.18. The cell line BT-549 (from ductal carcinoma) was purchased from CLS (Eppelheim, Germany). The cell line, EFO21 (from ovarian serous cystadenocarcinoma), was purchased from DSMZ (Braunschweig, Germany), the cell lines MCF7 (from breast carcinoma) and T47D (from breast cancer) from ATCC (Manassas, VA, USA). All cell lines were verified using the STR (Short Tandem Repeat)-analysis (DMSZ German Collection of Microorganisms and Cell Cultures GmbH on 21.11.2018). All the subsequent research analyses were carried out in accordance with the approved guidelines and regulations.

The generation of primary fibroblasts from skin biopsies is described in Supplementary data Methods). The primary fibroblasts cell lines were cultured in Dulbecco`s Minimal Essential Medium with Earle's salts DMEM (Invitrogen, Karlsruhe, Germany), supplemented with 15 % fetal bovine serum (FBS) (Merck, Darmstadt, Germany), 1 % vitamins and 1 % antibiotics (penicillin/streptomycin) (Biochrom, Berlin, Germany) in a 90 % humidified incubator with 5 % CO₂ at 37 °C.

The FaDu tumor cell line was cultured in Dulbecco`s Minimal Essential Medium (Sigma-Aldrich, St. Louis, USA) containing 1 % non-essential amino acids (Biochrom, Berlin, Germany), 15 % FBS (Biochrom, Berlin, Germany) and 1 % antibiotics (Biochrom, Berlin, Germany). Passaging was done using 0.05 % trypsin with 0.1 % ethylene-diamine-tetra-acetate (EDTA) (Biochrom, Berlin, Germany). FaDu derived single-cell clones were generated by limiting dilution of the primary cell line and propagated in DMEM (Invitrogen, Karlsruhe, Germany), supplemented with 15 % FBS, 1 % vitamins and 1 % antibiotics (Biochrom, Berlin, Germany) using 96-well plates (Cellstar, Greiner GmbH, Kremsmünster, Germany). After 48 h wells with only one cell were selected via microscopic examination. Subsequent propagation was performed in conditioned medium (one day old medium of primary culture at 50 % confluency, sterile filtered using 0.2 µm filter). The cells were transferred when they reached 80 % confluency, initially into a 24-well plate, later in a 6-well and finally in the 10 cm petri dishes.

Cell lines (MCF7, BT-549, EFO21 und T47D) were cultivated in RPMI1640 (Gibco, Life Technologies, Darmstadt, Germany) supplemented with 2.5 % HEPES buffer (Sigma, Darmstadt, Germany) and 1 % antibiotics (Life Technologies, Darmstadt, Germany). BT-549, MCF7 and T47D were additionally supplemented with 10 % FBS, EFO21 with 20 % FBS.

All experiments using the primary fibroblasts were performed with growth-arrested cells in the G0/G1 stage. Confluency was achieved by contact inhibition and subsequent cultivation for two weeks and confirmed by FACS (flow cytometric cell cycle) analysis. For comparisons of 2N, 1N, and 0N patients, fibroblasts with a similar passage number 5 (±2) were used.

To obtain proliferating immortal lymphoblastoid cells Epstein Barr virus (EBV) transformation of resting B cells (in peripheral blood lymphocytes) is widely used (Neitzel, 1986[59]). Lymphoblastoid cells were generated in our routine labor and harvested in an early passage after stable infection.

Fibroblasts were cultured as described above. The cells were cultured in T25 flasks to 80-90 % confluency, then a non-lethal dose of 3 µM daunorubicin (Pfizer Pharma PFE GmbH, Berlin, Germany) (Przybylska et al., 2001[60]) was added. After two hours the medium was replaced 2, 4 h and 24 h post-treatment and the cells were harvested using trypsin. Quantification of γH2AX was performed as described previously (Weis et al., 2008[72]).

Tumor cell lines were cultured as described above. Cells were seeded at a density of 5*10⁴ in T75 cell culture flasks in triplicates. The cells were harvested at different time points using trypsin and total cell numbers were determined using the Moxi^z automatic cell counter (Orflo, Ketchum, USA) at every time point. The cellular proliferation rate was calculated as the cumulative population doublings (CPD). The statistical analysis was done using the linear mixed-effects model fit by REML setting the biological replicate as a random variable.

Cells were exposed to X-rays with a D3150 X-Ray Therapy System (Gulmay Ltd, Surrey, UK) at 140 kV and a dose rate of 3.62 Gy/min at room temperature. Sham irradiated control cells were kept at the same conditions in the radiation device control room. The FaDu tumor cell line was irradiated at 80 % confluency. Cells were exposed to single doses ranging from 2-8 Gy and harvested at 15 min, 2 h, and 24 h after irradiation. For fractionalized irradiation, fibroblasts were irradiated 4x, 8x, or 10x with doses of 2 Gy and 4 Gy, within 20 days. Cells were given one day of recovery time between exposures and the medium was changed twice a week. Cells were harvested one day after the final irradiation.

Clonogenic survival was determined in colony formation assays adapted after Menegakis et al. (2009[58]). At passage p8, cells were seeded (1x10⁵) in 10 cm diameter Petri dishes in triplicates. After five days and one medium change, the cells were irradiated with 2 Gy, 4 Gy, 6 Gy, and 8 Gy. Sham irradiated cells were kept at the same conditions in the radiation device control room (0 Gy). After 24 h cell suspensions were obtained for each dose and different seeding densities were plated in triplicates. The remaining cells were pelleted, washed with PBS and stored at -80 °C for further experiments. 14 days after irradiation colonies were fixed and stained with crystal violet. The image of a FaDu cell line (0 Gy) was made with the ZEISS Axiovert microscope using the Clone software at 10x magnification. The grayscale transformation was done with GIMP 2.10.14 software. Colonies defined as >50 cells were counted and surviving fractions were expressed in terms of plating efficiency. Survival data after radiation were fitted to linear quadratic regression models employing the maximum likelihood approach (R package CFAssay). Differences between curves were evaluated using the F-test. Adjustment of p-values was done using the method of Benjamini and Hochberg (1995[5]).

Cells were washed with PBS (37 °C; 1 ml/25 cm²) and trypsinized (Trypsin/EDTA (37 °C; 1 ml/ 25 cm²), for 5 min at 37 °C. The reaction was stopped by the addition of a two-fold volume of culture medium at 37 °C and cells were collected by centrifugation at 300 x g for 6 min. 70 % ethanol (-20 °C) was added dropwise to the cell pellet under permanent turbulence (~ 4 ml/10⁵ cells) and the suspension was stored for ≥ 30 min at 4 °C or overnight. Later, the cell suspension was centrifuged at 300 x g for 6 min, and the supernatant was discarded. Cells were re-suspended in PBS (without magnesium and calcium -/-) with RNase (20 µg/ 10⁴ cells) and incubated for 30 min at 37 °C. After an additional centrifuge step of 300 x g for 6 min, cells were re-suspended in HOECHST 33258-staining solution (0.2 µg/ml in 1x PBS-/-) (~ 4 ml/ 10⁵ cells) and again incubated for ≥ 30 min at 4 °C. The analysis was performed using FACS Canto II (BD Biosciences, Germany).

Genomic DNA was isolated with the NucleoSpin tissue kit (Macherey-Nagel, Germany). Bisulfite conversion of 0.2 µg DNA was performed with the EpiTect Bisulfite Kit (Qiagen, Germany) according to the manufacturer's instructions. PCR and sequencing primers for the genes mentioned above were designed with PyroMark Assay Design 2.0 software (Qiagen, Germany) (Supplementary information Table 1). The 50 µl PCR reactions consisted of 25 µl Pyro PCR Mix 2x, 5 µl Coral load 10x, 1 μl of each forward and reverse primer (10 µM), 17 μl PCR-grade water and 1 μl (~100 ng) bisulfite-converted template DNA (PyroMark PCR Master Mix, Qiagen). PCR amplifications were performed with an initial denaturation step at 95 °C for 5 min, 35 cycles (95 °C for 30 sec, TM (gene analyzed) for 45sec and 72 °C for 45 sec) and a final extension step at 72 °C for 10 min. Bisulfite pyrosequencing was performed on a PyroMark Q96 MD Pyrosequencing System using the PyroMark Gold Q96 CDT Reagent Kit (Qiagen, Germany) and 1 µl of sequencing primers (10 µM). Data analysis was performed with the Pyro Q-CpG software (Qiagen, Germany). Single CpG errors which arise presumably due to bisulfite conversion errors (technical variation) were in the order of 1 %. This value was calculated by analyzing samples of three patients in three replicates.

Data were analyzed using the Kruskal-Wallis rank sum-test. P-values were adjusted for multiple testing using the method of Benjamini and Hochberg (1995[5]). Analysis of the AML and NHL patient data was performed using the Mann-Whitney-U Test (RV3.6.2)

Next-generation sequencing (NGS) libraries for DBS were generated as described previously (Böck et al., 2018[9]). PCR amplification of APC, CDKN2A, TP53, and RAD9A was performed using primers containing a target-specific part and partial adapter overhangs (Supplementary informatin Table 2). Following the purification of the amplicons with magnetic beads (0.9:1), the DNA concentration was determined with the dsDNA BR Assay System (Life Technologies, USA). Samples were diluted to 0.2 ng/µl in a total volume of 15 µl each. After pooling, barcoding was performed by PCR with NEBNext Multiplex Oligos for Illumina, Dual Index Primer Set 1 (New England BioLabs, Germany), followed by another bead cleaning step (0.9:1). DNA concentration and fragment length were determined with the High Sensitivity DNA kit on a 2100 Bioanalyzer (Agilent Technologies, USA). All barcoded pools were diluted to 2 nM and pooled to the final library. Denaturation and preparation of the library and PhiX control were done according to the manufacturer's protocol (Illumina, USA). Paired-end sequencing was performed on an Illumina MiSeq using the MiSeq Reagent Kit v2 (2 x 150 cycles) cartridge.

The sequences in FASTQ format were processed using the Amplikyzer2 pipeline (Rahmann et al., 2013[62]), which provides a detailed nucleotide-level analysis including the calculation of CpG methylation rates. All sequences were aligned to the genomic sequence of each amplicon using default settings. For the subsequent extraction of reads and CpG-wise methylation status only reads with an overall bisulfite conversion rate of over 95 % were considered. Further downstream processing of Amplikyzer2 output files and subsequent analyses of methylation rates were performed using R-scripts. Statistical analyses were performed with the statistical software package R (Version 3.2.2) (R-Core Team).

High-resolution screening for microdeletions and duplications of the FaDu cell line and its subclones was performed with the Affymetrix GeneChip Genome-Wide Human SNP Cyto HD, following the protocol developed by the manufacturer (Affymetrix, Santa Clara, CA, USA). Data calculation was performed with Chromosome Analysis Suite 3.1.0.15 and NetAffx Build 33.1 (hg19). Karyotypes were described in accordance with the International System for Human Cytogenetic Nomenclature (ISCN2013).

Genomic DNA was isolated using the NucleoSpin tissue kit (Macherey-Nagel, Germany). PCR reactions were performed using the FastStart Taq DNA Polymerase, dNTPack, kit # 04738403001 (Roche Diagnostics, Germany). PCR was performed in three stages with one cycle of 5 min at 94 °C, forty cycles (15 sec, 94 °C, 30 sec primer TM, 1 min 68 °C) and one cycle of 2 min at 72 °C, using 30 ng DNA per reaction. Primers were designed using NCBI's primer designing tool (https://www.ncbi.nlm.nih.gov/tools/primer-blast/). The following primer pairs were used; forward primer (AGGCAGTCAGTCGGAAAGTG) and reverse (TTGGAACCTGCTGATTCGCT) for CHD2 (NM_001271) TM 62 °C; forward primer (ACAGCTGTTCTCACGGAAGG) and reverse (TAGGCTGCTGAGGATGGTCT) for SPATA8 (NR_158221) TM 62 °C and forward primer (GCAAGAAGAAGATGAAGAGC) and reverse (GGCAGGTAAGCTCAGGTTTT) for SMARCA1 (NM_003069) TM 64 °C. Sequence reactions were performed by Genterprise Germany, using PCR products digested with Exo/SaP (New England Biolabs, Germany). Uncropped image may be found in Supplementary data Figure 5.

Total RNAs were prepared from three independent cell culture samples (growth kinetics experiment) using the Nucleo Spin RNA Plus Kit from Macherey-Nagel (MN). 2 µg of the RNA samples were reversely transcribed into cDNA using the SuperScript IV First-Strand random hexamer Synthesis System (Invitrogen).

Forward and reverse primers (Exon-spanning for gene expression) were designed with the Primer-Blast program (https://www.ncbi.nlm.nih.gov/tools/primer-blast/). The TBP gene was used as an endogenous control. The following primer pairs (5´-3´ orientation) were used; forward primer (CGGCAACGTGAAGGTGCTC) and reverse (CAGGTTGTGAGTCTTCCGCA) for RAD9A (NM_004584.3) TM 65°C; forward primer (CCACTCACAGACTCTCACAAC) and reverse (CTGCGGTACAATCCCAGAACT) for TBP (NM_003194.5)

Each 10 µl reaction volume contained 40 ng cDNA in 5 µl Sybr-Green Master Mix (Biozym), 2 µl RNase-free PCR graded water (MN), 1 µl each of forward and reverse primer (10 µM). All reactions were performed in triplicate and two stages, with one cycle of 95 °C for 10 min (first stage) and 45 cycles of 94 °C for 10 s, (TM-primer)°C for 10 s, and 72 °C for 10 s (second stage) using the LightCycler 480II Roche. Amplification qualities were assayed using melting curves and agarose gel analysis. The qPCR amplification efficiency was calculated using the LinReg program (Ruijter et al., 2014[64]) and the CT values were corrected using the mean amplification efficiency. Relative quantification was carried out with the ΔΔCT method using the endogenous control gene for calibration. The student t-test was used for the comparison of data between two or more groups. All tests were two-tailed; p<0.05 was considered statistically significant.

In a previous study, we detected expression changes of RAD9A protein in fibroblasts of former childhood tumor patients with the prevalent cancer types being leukemias and solid tumors (Weis et al., 2011[73]). Alteration of gene expression may be caused by methylation changes of promotor or other regulation sites. The methylation of three CpG's in intron 2 of the RAD9A gene (Figure 1G(Fig. 1)) is known to regulate the expression of the RAD9A protein (Cheng et al., 2005[12]). Therefore, we analyzed the methylation of this site and as well as a panel of control genes (APC, CDKN2A, EFNA5, TP53) which were reported to be aberrantly methylated in related tumors and/or have a role in childhood malignancy (for further information see Supplementary material, Supplementary Table 1) using bisulfite pyrosequencing and deep bisulfite sequencing (DBS).

We determined the mean methylation of the promoter regions in APC (NC_000084.6), CDKN2A (NC_000009.12), and EFNA5 (NC_000005.10), a cis-regulatory region in RAD9A intron 2 (NC_000011.10), and a mutation hotspot in TP53 exon 6 (NC_000017.11) in 20 primary fibroblast cell lines of cancer-free controls 0N and matched patients 1N and 2N (Figure 1A-E(Fig. 1)). There were significant variations in methylation among the groups. APC (Figure 1A(Fig. 1)) proved to be hypermethylated in 1N in comparison to the 0N control (adj.p = 0.03). CDKN2A (Figure 1B(Fig. 1)), which is often mutated in a variety of cancers, exhibited hypermethylation in the 1N group (adj.p = 0.006), while it was hypomethylated in 2N group in comparison to 0N group (adj.p = 0.008). The 2N group showed a hypomethylation in TP53 (Figure 1C(Fig. 1)) in comparison to the control group 0N (adj.p = 0.01). No significant differences between the groups were detected for RAD9A genes (Figure 1D(Fig. 1)) and EFNA5 (Figure 1E(Fig. 1)). However, we identified several patients with aberrant methylation values. Patients 1N08 and 1N15 exhibited promoter hypermethylation of APC (9 %) and CDKN2A (7 %), respectively, whereas patient 2N12 displayed hypomethylation (95 %) of the TP53 mutation hotspot. Five patients (1N04, 1N07, 1N14, 1N20, and 2N21) showed increased RAD9A intron 2 mean methylation, ranging from 10 % to 31 % (Figure 1A-D(Fig. 1)).

Previously, we have demonstrated that samples with aberrant methylation determined by bisulfite pyrosequencing can be considered as likely candidates for mosaic epimutations, which may have a role in carcinogenesis (Galetzka et al., 2012[24]). Therefore, we focused on patients with RAD9A hypermethylation.

The average methylation changes could be due to either single CpG methylation errors at different positions in a large number of alleles or due to allele methylation errors, where most CpGs in individual DNA molecules are aberrantly methylated. Because it is usually the density of CpG methylation in a cis-regulatory region rather than individual CpGs that turns a gene "on" or "off" (Weber et al., 2007[71]), allele methylation errors must be considered putatively as functionally relevant epimutations. To investigate the allele positions of the methylated CpG's, we performed an analysis utilizing deep bisulfite sequencing (DBS), which can determine the methylation profiles of many thousands of individual DNA alleles for multiple genes and samples in a single experiment and thus directly measure epimutation rates (EMRs). To determine the density of CpGs and allele methylation ratio in the present study, we performed DBS on the patients with presumed epimutations (Figure 1A-E(Fig. 1) and Table 1(Tab. 1)). Alleles with >50 % aberrantly (de)methylated CpGs in DBS are considered as functionally relevant epimutations.

Consistent with an epimutation screen in breast cancer susceptibility genes (Böck et al., 2018[9]), we considered EMRs >1 % as elevated and EMRs >2.5 % as likely pathogenic constitutive epimutations. Using the above-described classification, we did not detect any APC epimutations in fibroblasts of patient 1N08. Patient 1N15 displayed 0.2 % EMRs in CDKN2A and patient 2N12, 0.3 % EMRs in TP53. Five patients with suspected RAD9A epimutations displayed 2.0 % to 24.5 % EMRs. Four childhood cancer patients (1N20, 1N14, 1N07 and 2N21) showed RAD9A epimutations and childhood cancer patient 1N04 showed an elevated RAD9A EMR. Overall, 10 % (4 of 40) childhood cancer patients (1N and 2N) had RAD9A epimutations in their fibroblast cells (Table 1(Tab. 1) and Figure 1F(Fig. 1)).

Four childhood patients with RAD9A EMR >2 % suffered from leukemia (ALL and Hodgkin lymphoma). To corroborate this finding, we first analyzed RAD9A methylation in the bone marrow of three leukemia patients. Patient P1 with pre-ALL and a Philadel-phia chromosome in <10 % of analyzed (bone marrow) cells displayed a mean RAD9A methylation of 18 %. Patient P2 with AML and 50 % bone marrow cells with abnormal karyotype displayed 29 % RAD9A mean methylation whereas patient P3 with PBL and 60 % cells of his bone marrow cells showed complex aberrations had a RAD9A mean methylation of 41 % (Figure 2A(Fig. 2)). The TCGA data base collection makes it possible to compare the expression of RAD9A in different tumor entities. In most tumor types RAD9A expression is deregulated (http://firebrowse.org/viewGene.html?gene=RAD9A) in comparison to controls. As leukemia and NHL occurred in our patients (primary secondary tumor) we conducted an extended analysis in leukemia tumor samples.

Further analysis of bone marrow samples from 27 NHL and 27 AML patients revealed a mean RAD9A methylation of 30 % for NHL and 20 % for the AML patients (p ≤0.025). We detected four NHL and five AML patients with elevated mean methylation levels, ranging from 38 % to 66 % in NHL and from 14 % to 63 % in AML patients (Figure 2B(Fig. 2)). We were able to analyze several bone marrow samples taken in the course of therapy for three AML patients (Figure 2C-E(Fig. 2)). The first bone marrow sample of AML patient 1 at diagnosis showed 36 % mean methylation of the RAD9A intron 2 sites. In course of the therapy, (life period: 19 months from diagnosis to death), the methylation levels decreased to 25 % and shortly before death increased to 30 % (Figure 2C(Fig. 2)). AML patient 2 had a rather high mean methylation of 59 % in the first bone marrow sample. Again, we could observe a slight decrease of methylation values (mean 44 %) during the therapy, but one month later the values increased again (mean methylation 55 %) and shortly after the last sample with the mean methylation value of 50 % was taken the patient deceased (life period from diagnosis to death was 4 months) (Figure 2D(Fig. 2)). The first bone marrow sample of AML patient 3 displayed the mean RAD9A intron 2 methylation values of 27 %. These values did slightly vary during the therapy. However, in a short period of 2 months before the patient died, he displayed strong hypermethylation of 43 %. This methylation further increased to 63 % (the life period from diagnosis to death was 27 months) (Supplementary data Figure 2).

In addition to RAD9A methylation, we analyzed bone marrow samples of this patient using a genome-wide SNP array for molecular karyotyping. As shown in Figure 2E(Fig. 2) the analyzed samples display a progressive duplication of the 16p13.3(1,345,222-3,178,084) fragment in the last 2 months before death. The final duplicated region contains 106 genes of which 9 genes are known to be involved in cancer (UBE2I, NUBP2, IGFALS, NTHL1, TSC2, PKD1, PDPK1, TCEB2, and TNFRSF12A).

To substantiate the connection between alteration of important genes and the hypermethylation of RAD9A, we examined two primary fibroblast cell lines known to have homozygous mutations either in BRCA2 (FANCD1) or in SLX4 (FANCP1) in contrast to primary fibroblast cell cultures (N=20, 0N). As expected, fibroblast control cells exhibited RAD9A mean methylation values ranging from 3-11 %, in contrast to FANCD1 (28 %) and FANCP1 (47 %) (Supplementary data Table 2).

According to Cheng and colleagues (Cheng et al., 2005[12]), in breast cancer, RAD9A becomes an oncogene via hypermethylation in intron 2, due to consequently increased RAD9A transcription. We hypothesized that RAD9A hypermethylation is linked to the oncogenic transformation of a cell. EBV transformation of B cells results in unlimited growth and has been associated with particular forms of cancer, such as Burkitt's lymphoma, Hodgkin's lymphoma, nasopharyngeal carcinoma, and gastric cancer (Vockerodt et al., 2015[69]). Thus, EBV infection is now widely used to generate immortal lymphoblastoid cell lines. Global changes in DNA methylation may contribute to the pathogenesis of EBV (Hernandez et al., 2017[31]). We, therefore, tested EBV transformed lymphoblasts for changes in methylation of RAD9A in intron 2 using bisulfite sequencing. The mean methylation in six different EBV transformed cell lines varied from 6 % to 41 %, the corresponding fibroblast cells showed 8 % mean methylation in RAD9A (Figure 3B(Fig. 3)). The DBS analysis of the cell line with the highest mean methylation value (41 %) exhibited 9 % fully methylated alleles. In contrast, the methylation patterns of APC, BRCA1, CDKN2A (hypomethylated), and TP53 (hypermethylated) remained virtually unchanged in this cell line after EBV transformation (Figure 3B(Fig. 3)).

To understand the role of hypermethylation of RAD9A in cancer and oncogenic transformation, we further analyzed BT-549 (breast cancer), MCF7 (breast cancer), EFO-21 (ovary), T47D (breast cancer), and FaDu (squamous cell carcinoma) cell lines. The RAD9A methylation varied between these tumor cell lines (mean values 20-81 %, Figure 4A(Fig. 4)). The highest methylation in all three CpG's was detected in the cell line EFO-21 (CpG1-84 %, CpG2-89 %, and CpG3-76 %). The methylation values seem to be independent of the RAD9A copy number. The MCF7 cell line (mean methylation 24 %) has two copies of Chr.11 and one derivative der(?)t(11;1;17;19;17), in contrast to EFO-21 (mean methylation 83 %) and FaDu cell line (mean methylation 54 %) with up to three copies of chromosome 11.

If RAD9A methylation is relevant for tumor development it should be possible to detect divergent methylation in tumor subclones. Subclonal events involving gene mutations were reported by Amin et al. (2016[1]) and recently by Ben-David et al. (2018[4]) showing copy number gains and losses and consequently different drug responses in cancer (Amin et al., 2016[1]; Ben-David et al., 2018[4]). To test the hypothesis that subclonal events may be associated with methylation changes in RAD9A we established several subclones of the FaDu cell line. The FaDu cell line exhibits homozygous loss of function mutations in TP53 and CDKN2A genes, therefore, chromosomal changes caused by the lack of proper DNA repair may occur frequently. During the cultivation of the parental FaDu cell line, subclonal events lead to divergent cells (Figure 4B(Fig. 4)).

We were able to generate thirteen stable subclonal cell lines with divergent RAD9A methylation (Figure 4C(Fig. 4)). Three subclones, 4, 6 and 10, exhibit high RAD9A methylation (74 %, 73 % and 69 % respectively) in comparison to the parental cell line (mean methylation 57 %), while others showed reduced methylation levels (e.g. subclone 2; mean 42 % and subclone 9; mean 40 %). The methylation levels of RAD9A remained stable during the further cultivation of all subclones. Subclones 2, 4, 6, 9 and 10 and the parental FaDu cell line were chosen for further characterization.

To elucidate if different methylation values have an impact on cell survival and damage response we performed clonogenic survival experiments. Using triplicates monolayer culture growth kinetics for the subclones 2, 4, 9 and the parental FaDu cell line cultured for 32 days (6 measuring points) revealed significantly delayed growth for subclone 4 (p <0.0001) (Figure 4D(Fig. 4)). Furthermore, clonogenic survival experiments upon irradiation performed with the subclones 2, 4 and the FaDu cell line resulted in significantly reduced survival of subclone 4 (adj p <0.0001) in comparison to the FaDu parental cell line and subclone 2 (adj p <0.026) (Figure 4 E, F(Fig. 4)). Methylation signatures for the subclones in clonogenic survival experiments matched the untreated subclone signatures (Supplementary data Figure 4).

To test our hypothesis that alterations in tumor-relevant genes may be associated with changes in RAD9A methylation, we performed an SNP array analysis of hypermethylated (4, 6 and 10) and hypomethylated (2 and 9) FaDu subclones. We identified a homozygous deletion in 15q26.1q26.2 unique to subclone 4 and a heterozygous deletion in Xq25 in subclone 6. Downstream analysis using PCR and Sanger sequencing confirmed the homozygous deletion of the CHD2 and SPATA8 genes in subclone 4 and the stop codon mutation c.537-538insG, p.P182fs*18 in the remaining allele of SMARCA1 (SNF2L) gene in subclone 6. Both genes harbor a helicase domain and are involved in DNA-repair and transcription regulation (Eckey et al., 2012[19]; Luijsterburg et al., 2016[53]) (Figure 5A and B(Fig. 5)).

In contrast to subclones 4, 6, and FaDu, subclone 10 displayed a 302 kb duplication in 16q23.1(75,318,494-75,620,953). The duplication encompasses 6 genes four of which are frequently deregulated in cancer (TMEM170A, CHST6, CHST5, TMEM231). The TMEM231 gene is responsible for Joubert Syndrome 20; (OMIM 614970) and the GABARAPL2 is involved in the autophagy interaction network (Figure 5C(Fig. 5)).

The subclones 2 and 9 were hypomethylated in comparison to the parental cell line FaDu (42 % and 40 % versus 57 % mean methylation). The molecular karyotyping of these subclones revealed a restoration of former duplicated areas on chromosome 3 and 12 for subclone 2 and chromosome 14 for subclone 9 (Figure 5D-F(Fig. 5)). Chromosomal gain and gene amplification involving chromosome arm 3q are the most frequent in human tumors. The chromosomal part 3q25.33q26.33 (159,599,190-183,199,315) contains 83 genes of which 52 are involved in cancer progression. Genes within this region, like TERC, PHC3, TNIK, MIR569, NLGN1, TBL1XR1, KCNMB3, ZNF639 and USP13 are responsible for cancer development or progression when overexpressed or duplicated (Baena-Del Valle et al., 2018[3]; Chaluvally-Raghavan et al., 2014[11]; Crea et al., 2013[14]; Davidson et al., 2014[15]; Han et al., 2016[29]; Kang et al., 2009[38]; Lee et al., 2019[43]; Lin et al., 2005[49]; Xi et al., 2019[75]). The other restored part in subclone 2 is located on chromosome 12q24.22q24.32. It contains 155 Genes of which 29 are involved in cancer as a tumor suppressor, oncogene, or cancer-promoting factor. The chromosomal part restored in subclone 9 contains 114 genes, 67 of which are involved in cancer. Furthermore, the chromosome section 14q24.3 is highly amplified in 18 Head and Neck Squamous Cell Carcinoma (HNSCC) cell lines (Järvinen et al., 2008[37]) and the TTC8 gene is responsible for the Bardet-Biedl 8 (OMIM 209900) and Retinitis pigmentosa 51 (OMIM 613464) syndrome (Supplementary data Figure 5).

As most of our patients (except patient 1N20) received chemo- and radiotherapy, we designed experiments to test whether the treatment has an impact on the methylation of RAD9A and whether RAD9A-methylation can serve as a stable methylation marker. We have previously shown that DNA methylation remains stable in primary fibroblasts throughout the first cell cycle after irradiation (Maierhofer et al., 2017[54]). In contrast, significant methylation changes in >250 genes and the MAP kinase signaling pathway were associated with delayed radiation effects in single-cell clones of irradiated fibroblast (Flunkert et al., 2018[22]). To study radiation effects on the methylation of the RAD9A intron 2 site, the control cell line 0N18 was analyzed at 15 min, 2 and 24 h after irradiation with 0 Gy, 2 Gy, 5 Gy, and 8 Gy at each time point. The RAD9A mean methylation values in intron 2 remained virtually unchanged between 7 % and 9 % (Figure 6A(Fig. 6)). As RAD9A expression is deregulated in a variety of tumors and plays an important role in DNA repair, we additionally performed experiments with three fibroblast cultures (from patients 1N08, 0N12, and 2N12) which were irradiated in fractions of 8x 2 Gy, 4x 4 Gy, 8x 4 Gy, 10x 2 Gy, and 10x 4 Gy within 20 days. Again, RAD9A mean methylation remained rather constant at 5 % in 1N08 and 2N12, and at 8-9 % in 0N12 cell lines (Figure 6B(Fig. 6)). Furthermore, we performed experiments in exponentially growing FaDu tumor cells to estimate the mean methylation values of RAD9A. The cells were analyzed at 2, 4, and 24 h after irradiation with a single dose of 0 Gy, 2 Gy, 5 Gy, and 8 Gy. The RAD9A mean methylation varied within a narrow range of 54 % to 57 % and there was no difference between irradiated and non-irradiated cells (Figure 6C(Fig. 6)). The proportion of G2-phase cells (measured by flow cytometry for FaDu cells) increased with radiation dose and time after irradiation, ranging from 35 % G2-phase cells in non-irradiated cells to 65 % at 24 h after 8 Gy indicating a functional G2/M checkpoint (Figure 6D(Fig. 6)).

Although tumor therapy varied between patients, daunorubicin and doxorubicin were frequently used in the treatment regimens. As both drugs have similar properties and the cellular uptake of daunorubicin is superior to that of doxorubicin, we analyzed the influence of daunorubicin on RAD9A methylation. Treatment of normal fibroblasts with 3 µM daunorubicin yields a surviving cell fraction of 60 % (Przybylska et al., 2001[60]). Analysis of γH2AX as a marker for DNA double-strand breaks confirmed the incorporation and toxicity of daunorubicin in the cells (Figure 6E(Fig. 6)). Examination of the methylation signature at 0, 1, 4, 12 and 24 h post-treatment showed no significant changes in methylation in two independent fibroblast cell lines. The 2N24 cell line exhibited mean methylation values of 4-6 % and 0N24 control cell line of 9-12 % (Figure 6F(Fig. 6)) and Supplementary data Figure 6.

The vast majority of childhood cancers occur sporadically and cannot be explained by germline mutations in known tumor susceptibility genes (D'Orazio, 2010[18]; Sylvester et al., 2018[66]). In a previous study, we detected reduced expression levels of RAD9A in primary fibroblast cells of childhood cancer patients with a second primary tumor (2N), suggesting a function of RAD9A as a genomic caretaker (Weis et al., 2011[73]). We did not detect suspicious mutations in RAD9A or other cancer susceptibility genes, in our patient collective. Therefore, it seems plausible to assume that stochastic or adverse exposure events during early (intrauterine and postnatal) development may increase cancer susceptibility through epigenetic reprogramming (Marshall et al., 2014[56]; Walker and Ho et al., 2012[70]), including epimutations of RAD9A. We identified 5 patients (four 1N and one 2N) with ≥ 10 % mean RAD9A methylation in mosaic in intron 2 and using DBS analysis we could show that the mosaic epimutations represent ≥ 2 % hypermethylated alleles. Although this finding may not explain the down-regulation of RAD9A in our 2N collective, considering our experiments done in this study we propose that mosaic RAD9A hypermethylation is an early (either stochastic or environmentally induced) event, which may increase the probability of malignant transformation in regular body cells (predisposing factor) or/and a factor that signals the tumor progression itself (oncogenic factor) for this mainly 1N patients.

Constitutive epimutations (allele methylation errors) that arise in early development are likely to be present in a mosaic state in different tissues of an individual. Nevertheless, the risk for malignant transformations may increase with the percentage of cells with compromised genomic caretakers. Previously, we have shown that epimutations in BRCA1 and RAD51 can originate in single precursor cells (Böck et al., 2018[9]; Hansmann et al., 2012[30]). Furthermore, we identified a monozygotic twin pair, discordant for childhood cancer with a constitutive mosaic BRCA1 epimutation in the twin with cancer (Galetzka et al., 2012[24]). Because studies about the involvement of mosaic epimutations in cancer are sparse (Lønning et al., 2019[50]; Hitchins, 2015[32]) it is difficult to define a threshold for constitutive epimutations in regular tissues that can be associated with tumor formation. In our experience with screening for TS epimutations in more than 800 individuals (Böck et al., 2018[9]; Hansmann et al., 2012[30]), mean methylation values of ≥ 10 % and allele methylation errors of ≥ 2-3 % (depending on the gene and assay) are outside the normal methylation variation range (depending on the tissue origin analyzed). Since all patients were analyzed using a genome-wide SNP array and no variants in RAD9A or neighboring genes were detected, the observed methylation changes must be caused by other mechanisms affecting gene regulation. Studies suggest that methylation may change during the lifetime (Horvath and Raj, 2018[36]). Since childhood cancer patients and controls enrolled in this study were matched, differences in RAD9A methylation are not be explained by age.

Although we detected 1-2 % changes in methylation values in APC, CDKN2A and TP53 gene, it is questionable if these methylation differences are of biological significance. Epimutation rates were not significantly increased in 1N08 (APC), 1N15 (CDKN2A) and 2N12 (TP53) patients. When the analyzed childhood cancer patients were recruited for this study, they all had survived an initial cancer treatment (in most cases including radiation therapy) and the 2N patients had been tumor-free for several years. Although we did not find evidence for irradiation- or chemotherapy-induced RAD9A epimutations, we cannot completely exclude the possibility that the observed RAD9A hypermethylation in some patients is a consequence of cancer treatment, which may lead to mutation of important genes. Due to the dual role of RAD9A in cancer, we would assume the hypermethylation in RAD9A, seen in the five patients in mosaic, as potentially harmful, because it may activate RAD9A transcription.

Patient 1N20, who suffered from sporadic unilateral retinoblastoma, exhibited almost 12 % fully methylated RAD9A alleles in his fibroblasts, indicating that almost one-eighth of his normal cells are endowed with epimutated RAD9A alleles. Most retinoblastomas, which are derived from the cone photoreceptor lineage, show biallelic inactivation of the RB1 tumor suppressor gene, but additional (epi)genetic changes are most likely required for tumor development (Dimaras et al., 2008[17]). We did not detect an RB1 germline mutation in our retinoblastoma patient, however, the mutational status of the tumor is not known. Instead, we detected a unique mosaic duplication in 13q14.3 including the genes CKAP2, THSD1 and VPS36 in the patient's fibroblasts (Galetzka et al., 2020[25]). The CKAP2 gene is a cytoskeleton-associated protein that stabilizes microtubules and plays a role in the regulation of cell division (Yoo et al., 2016[77]). Whether this gene alteration contributes to the development of sporadic retinoblastoma and/or RAD9A hypermethylation remains to be elucidated. The retinoblastoma of patient 1N20 was cured solely by excision of the tumor; therefore, no therapy-related changes may be expected.

The other four patients with ≥ 2 % hypermethylated alleles in RAD9A suffered from leukemia. Because, there is no research about the involvement of RAD9A hypermethylation, leukemia development and progression to date, we designed experiments with EBV, tumor cell lines, and bone marrow samples from leukemia patients to elucidate a possible mechanistic involvement.

The involvement of RAD9A hypermethylation in oncogenic transformation is substantiated with experiments performed with bone marrow samples from leukemia patients. We could detect a significant change in methylation values of RAD9A between AML and NHL patients. Moreover, we could show an evolving duplication on chr.16p13.3, which may be linked to RAD9A hypermethylation. One of the duplicated genes on chr.16p13.3 is NTHL1. Its overexpression causes genomic instability and loss of contact inhibition of cells (Limpose et al., 2018[48]). Another duplicated gene in chr.16p13.3 is PRKD1. High PRKD1 mRNA levels have been associated with low overall survival in TNBC (triple-negative breast cancer) (Spasojevic et al., 2018[65]). TNFRS12A, also located in the 16p13.3 duplication, is responsible for poor prognosis in breast cancer, if overexpressed (Yang et al., 2018[76]). To our knowledge, this study is the first analysis of RAD9A methylation in leukemia and the first analysis of RAD9A methylation during the therapy of leukemia.

Our assumption of the possible involvement of RAD9A hypermethylation in unknown signaling pathways triggered by hazardous mutations compromised DNA integrity and subsequent tumor formation is further substantiated by analysis of Fanconi anemia fibroblasts. They were derived from patients with rare recessive genetic disease resulting in impaired DNA damage response and genome instability. Among those affected, the majority develop cancer, most often acute myelogenous leukemia, and 90 % develop bone marrow failure (Fiesco-Roa et al., 2019[21]). Compared to control fibroblasts we observed RAD9A hypermethylation in FANCP (homozygote deletion of SLX4 gene) and D1 fibroblasts (homozygous deletion of BRCA2).

In our experiments the EBV transformation of resting B cells to proliferating lymphoblasts induced a dramatic increase in RAD9A methylation and epimutation rate, indicating that epigenetic dysregulation of RAD9A may occur during malignant transformation early in tumorigenesis. It was reported that EBV infection causes the overexpression of APOBEC3s (apolipoprotein B mRNA editing catalytic polypeptide) which may increase the mutation load in the host DNA, providing further clues to the mechanisms of EBV-induced gastric carcinogenesis. This may prove to be a secondary driving force in the mutational evolution of EBV+ gastric tumors (Bobrovnitchaia et al., 2020[8]).

The methylation variation among FaDu subclones demonstrated methylation dynamics during tumor progression, which is associated with specific chromosomal rearrangements. Copy number variation is one important mechanism of cancer cells to promote the expression of genes involved in cancer progression and outcome (Chin et al., 2006[13]). In general, subclones with RAD9A hypermethylation showed mutations or imbalance in genes crucial for cell stability and cancer, whereas in hypomethylated clones we observed a restoration to normal copy numbers.

It is tempting to speculate on the existence of a signaling pathway that is activated upon deregulation of tumor-relevant genes. RAD9A has been described to be involved in DNA repair and apoptosis. When translocated to mitochondria, RAD9A binds and neutralizes the anti-apoptotic activity of BCL-2 and BCL-xL proteins, thus promoting cell death (Komatsu et al., 2000[41]; Yoshida et al., 2002[78]). How the activation, choice and fine-tuning of the pathway occurs is not known. The methylation-sensitive region in intron 2 of RAD9A is endowed with three regulatory elements, annotated in ORegAnno (http://www.oreganno.org/). Element OREG1137234 represents a binding site for the transcription factor ZNF263, which functions as a transcriptional repressor (Weiss et al., 2020[74]). Furthermore, it was shown that SP3, which contains an inhibitory domain to repress gene transcription, binds to the intron 2 region of the gene in vivo (Cheng et al., 2005[12]). It is plausible to assume that intron 2 hypermethylation interferes with SP1, SP3 and/or ZNF263 binding, and in conclusion, RAD9A expression is activated (Figure 7A-C(Fig. 7)). SP1 and SP3 proteins are localized exclusively in the cell nucleus (Birnbaum et al., 1995[7]; Hagen et al., 1992[28]). Both proteins contain a highly conserved DNA binding domain and two glutamine-rich regions. The glutamine-rich domains interact with the TATA box binding protein (TBP)-associated factor dTAFII110 and have a strong transcription activation function and the SP1 and SP3 (Dennig et al., 1996[16]; Gill et al., 1994[26]) when bound to the promoter region. SP3 can however also act as a transcriptional repressor due to an inhibitory domain between the second glutamine-rich domain and the zinc finger region (Luca et al., 1996[52]). It has been shown, that Sp3 binds to the intron 2 region of RAD9A in vivo and that in MCF-7 breast cancer cells and breast cancer tumors with increased RAD9A mRNA this binding site was hypermethylated (Cheng et al., 2005[12]). This suggests, that the unmethylated intron 2 SP1/SP3 binding site is bound by SP3, reducing the transcription of RAD9A, whereas hypermethylation inhibits SP3 binding and leading to increased RAD9A expression.

First of all, we would like to thank the patients for participating in this study. Our clinical staff also deserves special thanks. We also thank Ursula Disque-Kaiser, Britta Weber, Danuta Wieczorek, Martina Mihm, Brigitte Schuh and Martina Hermanns for technical support. We kindly thank Prof. Detlev Schindler for the donation of Fanconi Fibroblasts. We are grateful to Benjamin Irmscher for providing the daunorubicin experiments. Special thanks to Heather Chorzempa for the proofreading of this manuscript.

The datasets generated and analyzed during the current study are not publicly available due to ethics and data protection reasons but are available from the corresponding author on reasonable request.

The authors declare no conflict of interest

This study was supported by the BMBF German ministry for education and science [02NUK016A, 02NUK042A, 2NUK042B and 2NUK042C] and has received funding from the Brigitte und Dr. Konstanze Wegener-Stiftung (Project #70).

DG, JB and LW contributed equally to this work, and should be considered co-first authors, DG is the guarantor of this work and, as such, had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. MD analyzed the bNGS data. DG, OS, ML, JM, prepared cell culture radiation and pyrosequencing experiments. AP performed the statistical analysis of the pyro sequence data, the colony formation assay and the growth kinetics. MR, LW and WB prepared the tumor cell lines and bone marrow samples. MLi and DG conducted the SNP array molecular karyotyping. HR helped with the analysis of pyro sequence data. JB and OS prepared the bisulfite sequence bNGS and provided the data interpretation. CS, MM, PSZ organized the patient's recruitment. DG, TH, DP and HS conceived and designed this project and wrote the paper. All authors revised the manuscript critically for important intellectual content and approved the final version and agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.