Epimutation screen of fibroblasts derived from childhood cancer patients using bisulfite pyrosequencing and deep bisulfite sequencing
In a previous study, we detected expression changes in tumor-related proteins as RAD9A in fibroblasts of former childhood tumor patients with the prevalent cancer types being leukemias and solid tumors [18]. Alteration of gene expression may be caused by methylation changes of promotor or other regulation sites. The methylation of tree CpG’s in intron 2 of the RAD9A gene was proven to influence the expression of the RAD9A protein [22]. Therefore, we analyzed the methylation of intron 2 of the RAD9A gene and chose 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 references S3) using bisulfite pyrosequencing and deep bisulfite sequencing (DBS)
Bisulfite pyrosequencing. We have 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 (Fig. 1a-e). There were significant variations in methylation among the groups. APC (Fig. 1a) proved to be hypermethylated in 1N in comparison to the 0N control (adj.p-value = 0.03). CDKN2A (Fig. 1b), which is often mutated in a variety of cancers, exhibited hypermethylation in the 1N group (adj.p-value = 0.006) while it was hypomethylated in 2N group in comparison to 0N group (adj.p-value = 0.008). The 2N group showed a hypomethylation in TP53 (Fig. 1c) in comparison to the control group 0N (adj.p-value = 0.01). No significant differences between the groups were detected for RAD9A genes (Fig. 1d) and EFNA5 (Fig. 1e).
Outlier analysis. In previous work, we showed that outliers detected using bisulfit pyrosequencing could be considered as likely candidates for an abnormal methylation pattern, indicative of a mosaic epimutation [16]. So, we focused in this study on those cases. We identified three patients, one patient (1N08) who exhibited conspicuous hypermethylation (9%) of the APC promoter, another one (1N15) showed hypermethylation of CDKN2A (7%) and patient 2N12 displayed hypomethylation in TP53 (95%). Five patients (1N04, 1N07, 1N14, 1N20, and 2N21) showed increased RAD9A intron 2 mean methylation, ranging from 10–31% (Fig. 1. a-d).
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" [23], 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).
Deep bisulfite sequencing 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 suspected epimutations (Table 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 [3], 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 2.0–24.5% 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 and Fig. 1f).
Table 1
Results of methylation analysis (by bisulfite pyrosequencing and deep bisulfite sequencing) of patients with suspected epimutations
| bisulfite pyrosequencing | deep bisulfite sequencing |
Sample ID | First cancer* | Second cancer* | Gene harboring the putative epimutation | Mean (%) of all 1N and 2N patients | Mean (%) of the given patient | Mean (%) in a given patient | Alleles with > 60% methylation (%) EMR | Fully methylated alleles (%) | Fully unmethylated alleles (%) |
1N08 | ALL | - | APC | 5 | 9 | 0.3 | 0 | 0 | |
1N15 | ALL | - | CDKN2A | 3 | 6.5 | 2.2 | 0.2 | 0 | |
2N12 | Rhabdo-myosarcoma | Liver cancer | TP53 | 97 | 90 | 93 | | | 0.30 |
1N06 (control) | ALL | - | RAD9A | 7 | 0 | 2.2 | 0.5 | 0.01 | |
1N04 | Hodgkin lymphoma | - | RAD9A | 7 | 13 | 9.3 | 2.0 | 0.3 | |
1N07 | ALL | - | RAD9A | 7 | 10 | 9,2 | 7.3 | 1.4 | |
1N14 | Hodgkin lymphoma | - | RAD9A | 7 | 14 | 12.6 | 5.9 | 0.7 | |
1N20 | Retinoblastoma | - | RAD9A | 7 | 31 | 32.3 | 24.5 | 11.5 | |
2N21 | ALL | RA | RAD9A | 7 | 12 | 10 | 3.2 | 0.9 | |
* ALL, acute lymphoblastic leukemia; RA, refractory anemia |
Supplementary Table S1: List of PCR- and sequencing primer for bisulfite pyrosequencing |
RAD9A hypermethylation in the bone marrow of leukemia patients.
Four childhood patients with RAD9A EMR > 2% suffered from leukemia. To corroborate the results of leukemic transformation, we analyzed firstly the methylation in the bone marrow of three leukemia patients. Patient P1 with pre-ALL and a Philadelphia 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 a 29% RAD9A mean methylation whereas patient P3 with PBL and 60% cells of his bone marrow cells showed complex aberrations, consisting of a hypodiploid (28%) clone and a hyperdiploid line (32%), had a RAD9A mean methylation of 41% (Fig. 2a). Further analysis of bone marrow samples from 27 NHL and 26 AML patients revealed a RAD9A mean methylation of 30% for NHL and 20% for the AML patients (p ≤ 0.025). We detected three NHL and four AML patients with elevated mean methylation. The values of the aberrant samples ranged from 38–66% for the NHL patients and from 14–63% for the AML patients (Fig. 2b). We were able to analyze several bone marrow samples taken in the course of therapy for three AML patients (Fig. 2c-e). The first bone marrow sample of AML patient 1 at diagnosis showed 36% mean hypermethylation of RAD9A intron 2 site. In course of the therapy, the methylation levels at first decreased (mean methylation 25%) but in the sample nr. 4 the mean methylation values raised again to 30% and shortly after this, the patient deceased (life period from diagnosis to death 19 months) (Fig. 2c). AML patient 2 had a rather high mean hypermethylation 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 raised again (mean methylation 55%) and shortly after the last sample with the mean methylation value of 50% the patient deceased (life period from diagnosis to death was 4 months) (Fig. 2d). The first bone marrow sample of AML patient 3 displayed the mean methylation values of RAD9A 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 shortly before death to mean hypermethylation of 63% (life period from diagnosis to death was 27 months). To elucidate the reason for the increase of the methylation we analyzed bone marrow samples of this patient using genome-wide SNP array molecular karyotyping. As shown in Fig. 2e 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 normal primary fibroblast cell lines (N = 20). As expected, fibroblast control cell lines exhibited RAD9A mean methylation values ranging from 3–11%, in contrast to FANCD1 (28%) and FANCP1 (47%).
RAD9A hypermethylation during EBV transformation and tumor development
According to Cheng and colleagues [22], RAD9A becomes an oncogene in breast cancer via hypermethylation in intron 2. This means if a cell becomes transformed to allow unlimited growth and show distinct tumor characteristics, changes in methylation values in intron 2 may be visible. Oncogenic 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. 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 [24]. We, therefore, tested EBV transformed lymphoblasts for changes in methylation of RAD9A in intron 2 using bisulfit sequencing. The mean methylation varied in six different EBV transformed cell lines from 6–41% (Fig. 3a). The DBS analysis of the cell line with the highest mean methylation value exhibited 9% fully methylated alleles (the corresponding fibroblast cell line showed 8% mean methylation in RAD9A). In contrast, the methylation patterns of APC, BRCA1, CDKN2A, and TP53 remained virtually unchanged in this cell line after EBV transformation (Fig. 3b).
RAD9A hypermethylation in tumor cell lines and FaDu subclones
Tumor cell lines. To understand the role of hypermethylation of RAD9A in cancer and oncogenic transformation, we analyzed BT-549 (breast cancer), MCF7 (breast cancer), EFO-21 (ovary), T47D (breast cancer), and FaDu (squamous cell carcinoma) cell lines. The hypermethylation varied between these tumor cell lines (mean values 20–81%, Fig. 4a). The highest methylation in all three CpG’s was detected in the cell line EFO-21 of 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.
FaDu subclones. If RAD9A methylation is relevant for tumor development and is not due to changed chromosomal numbers but gene alteration, it should be possible to detect divergent methylation in tumor subclones. Subclonal events with gene mutations were reported by Nisar et al. 2016 and recently by Ben-David et al. 2018 with the consequence of copy number gains and losses and consequently different drug responses [25, 26]. To test the hypothesis that subclonal events may be responsible for 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 a certain number of divergent cells (Fig. 4b). We were able to generate thirteen subclonal cell lines with divergent RAD9A methylation patterns (Fig. 4c). Two of the subclones, 4 and 6, exhibit high RAD9A methylation values (mean methylation 75% and 73% respectively) in comparison to the parental cell line (mean methylation 54%), while others showed a lower methylation level (e.g. subclone 2; mean 42% and subclone 9; mean 40%). The methylation levels of RAD9A remained stable during cultivation for all clones. Clones 2, 4, 6, 9 and 10 and the parental FaDu cell line were chosen for further characterization.
Subclone characterization. Monolayer culture growth kinetics, for the subclones 2, 4, 9 and the parental FaDu cell line using triplicates revealed significantly delayed growth for subclone 4 (p-value < 0.0001) (Fig. 4d). Clonogenic survival experiments upon irradiation performed with the subclones 2, 4, 9 and the FaDu cell line resulted in significantly reduced survival of the clone 4 (adj p-value. <0.0003) and clone 2 (adj p-value < 0.001) in comparison to the FaDu parental cell line (Fig. 4e, f). Methylation signatures for the clones in clonogenic survival experiments matched the untreated clone signatures.
SNP array analysis provides a more detailed view of chromosomal alterations. Using this technique, we analyzed subclones with hypermethylation (4, 6 and 10) in comparison to low methylation (subclones 2 and 9) and the parental FaDu cell line. The obtained karyograms matched the karyogram of the FaDu parental cell line. No structural chromosomal differences between the analyzed subclones and parental cell line could be detected; instead, 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 [27, 28] (Fig. 5a and b).
In contrast to the subclones 4, 6, and FaDu, the subclone 10 displayed a 302 kb duplication in 16q23.1(75,318,494 − 75,620,953). The duplication encompasses 6 genes of which 4 are frequently deregulated in cancer (TMEM170A, CHST6, CHST5, TMEM231). The TMEM231 gene is responsible for the Joubert Syndrome 20; (OMIM 614970) and the GABARAPL2 is involved in the autophagy interaction network (Fig. 5c).
Effects of radiation and chemotherapeutics on RAD9A methylation
As most of our patients (except patient 1N20) did receive chemo- and radiotherapy following diagnosis and the donation of the fibroblasts was done in adulthood, several years after the first malignancy, we designed experiments that may clarify whether treatment of the malignancies has an impact on the methylation of RAD9A or if RAD9A-methylation is a potential stable methylation marker.
Effects of radiation. We have previously shown that DNA methylation remains relatively stable in primary fibroblasts throughout the first cell cycle after irradiation [29]. 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 [30]. To study radiation effects on the mean 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% (Fig. 6a). 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 strains (1N08, 0N12, and 2N12) which were irradiated in fractions of 8 × 2 Gy, 4 × 4 Gy, 8 × 4 Gy, 10 × 2 Gy, and 10 × 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 (Fig. 6b). 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 54–57% and there was no difference between irradiated and non-irradiated cells (Fig. 6c). 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 (Fig. 6d).
Effects of chemotherapeutics. 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, as stated in the study of Przybylska et al. [31], yields a surviving cell fraction of 60%. Analysis of γH2AX as a marker for DNA double-strand breaks confirmed the incorporation and toxicity of daunorubicin in the cells (Fig. 6e). 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% (Fig. 6f).