Reciprocal epigenetic remodeling controls testicular cancer hypersensitivity to hypomethylating agents and chemotherapy

Testicular germ cell tumors (TGCTs) are aggressive but sensitive to cisplatin‐based chemotherapy. Alternative therapies are needed for tumors refractory to cisplatin with hypomethylating agents providing one possibility. The mechanisms of cisplatin hypersensitivity and resistance in TGCTs remain poorly understood. Recently, it has been shown that TGCTs, even those resistant to cisplatin, are hypersensitive to very low doses of hypomethylating agents including 5‐aza deoxy‐cytosine (5‐aza) and guadecitabine. We undertook a pharmacogenomic approach in order to better understand mechanisms of TGCT hypomethylating agent hypersensitivity by generating a panel of acquired 5‐aza‐resistant TGCT cells and contrasting these to previously generated acquired isogenic cisplatin‐resistant cells from the same parent. Interestingly, there was a reciprocal relationship between cisplatin and 5‐aza sensitivity, with cisplatin resistance associated with increased sensitivity to 5‐aza and 5‐aza resistance associated with increased sensitivity to cisplatin. Unbiased transcriptome analysis revealed 5‐aza‐resistant cells strongly downregulated polycomb target gene expression, the exact opposite of the finding for cisplatin‐resistant cells, which upregulated polycomb target genes. This was associated with a dramatic increase in H3K27me3 and decrease in DNMT3B levels in 5‐aza‐resistant cells, the exact opposite changes seen in cisplatin‐resistant cells. Evidence is presented that reciprocal regulation of polycomb and DNMT3B may be initiated by changes in DNMT3B levels as DNMT3B knockdown alone in parental cells resulted in increased expression of H3K27me3, EZH2, and BMI1, conferred 5‐aza resistance and cisplatin sensitization, and mediated genome‐wide repression of polycomb target gene expression. Finally, genome‐wide analysis revealed that 5‐aza‐resistant, cisplatin‐resistant, and DNMT3B‐knockdown cells alter the expression of a common set of polycomb target genes. This study highlights that reciprocal epigenetic changes mediated by DNMT3B and polycomb may be a key driver of the unique cisplatin and 5‐aza hypersensitivity of TGCTs and suggests that distinct epigenetic vulnerabilities may exist for pharmacological targeting of TGCTs.

Testicular germ cell tumors (TGCTs) are aggressive but sensitive to cisplatin-based chemotherapy. Alternative therapies are needed for tumors refractory to cisplatin with hypomethylating agents providing one possibility. The mechanisms of cisplatin hypersensitivity and resistance in TGCTs remain poorly understood. Recently, it has been shown that TGCTs, even those resistant to cisplatin, are hypersensitive to very low doses of hypomethylating agents including 5-aza deoxy-cytosine (5-aza) and guadecitabine. We undertook a pharmacogenomic approach in order to better understand mechanisms of TGCT hypomethylating agent hypersensitivity by generating a panel of acquired 5-aza-resistant TGCT cells and contrasting these to previously generated acquired isogenic cisplatin-resistant cells from the same parent. Interestingly, there was a reciprocal relationship between cisplatin and 5-aza sensitivity, with cisplatin resistance associated with increased sensitivity to 5-aza and 5-aza resistance associated with increased sensitivity to cisplatin. Unbiased transcriptome analysis revealed 5-azaresistant cells strongly downregulated polycomb target gene expression, the exact opposite of the finding for cisplatin-resistant cells, which upregulated polycomb target genes. This was associated with a dramatic increase in H3K27me3 and decrease in DNMT3B levels in 5-aza-resistant cells, the exact opposite changes seen in cisplatin-resistant cells. Evidence is presented that reciprocal regulation of polycomb and DNMT3B may be initiated by changes in DNMT3B levels as DNMT3B knockdown alone in parental cells resulted in increased expression of H3K27me3, EZH2, and BMI1, conferred 5-aza resistance and cisplatin sensitization, and mediated genome-wide repression of polycomb target gene expression. Finally, genome-wide analysis revealed that 5-aza-resistant, cisplatin-resistant, and DNMT3Bknockdown cells alter the expression of a common set of polycomb target genes. This study highlights that reciprocal epigenetic changes mediated by DNMT3B and polycomb may be a key driver of the unique cisplatin and 5-aza hypersensitivity of TGCTs and suggests that distinct epigenetic vulnerabilities may exist for pharmacological targeting of TGCTs.

Introduction
Testicular germ cell tumors (TGCTs) are the most common solid tumor of males 15-40 years of age [1]. Prior to the introduction of cisplatin-based chemotherapy, the majority of metastatic TGCT patients died of progressive disease. This situation has been remarkably reversed with cisplatin-based therapy where nearly 80% of metastatic patients survive beyond 5 years [2]. A greater understanding of the curability of TGCTs may inform more effective strategies to treat other metastatic cancers. Also, cisplatin refractory patients have a poor prognosis and 50% will die of progressive disease [3]. There are essentially no effective traditional, targeted, or immunotherapies for these patients. We and others have shown in preclinical models that TGCT cells are hypersensitive to hypomethylating agents including 5-aza deoxycytidine (5-aza) and the 5aza prodrug, guadecitabine [4][5][6][7][8]. This hypersensitivity is on the order of 100-to 1000-fold compared to the sensitivity of other solid cancers, extends to cisplatinresistant TGCT cells, and depends on very high levels of the DNA methyltransferase, DNMT3B. These findings are in line with the hypothesis that due to their germ cell origins and low mutational burden, TGCTs may be especially driven and sustained by distinct epigenetic alterations resulting in distinct vulnerabilities to epigenetic drugs [9][10][11]. Importantly, hypomethylating agents have been shown to resensitize cisplatinresistant cells to cisplatin and we and a second group have provided early phase clinical findings suggesting that a subset of cisplatin refractory TGCT patients favorably respond to guadecitabine in combination with cisplatin [4][5][6][7][12][13][14]. However, besides the potential role of DNMT3B, the mechanisms accounting for the hypersensitivity of TGCT cells to hypomethylating agents are unknown.
In prior work, we developed a series of isogenic cisplatin refractory cell lines from parental cisplatin sensitive cells following a protocol designed to mimic cisplatin-based patient therapy [15]. Genome-wide transcriptome analysis revealed that the majority of cisplatin-resistant cells gained expression of genes normally repressed by the polycomb pathway, which coincided with a global decrease in H3K27me3, H2AUbK119, and decreased expression of the polycomb pathway components EZH2 and/or BMI1 [15]. Further, pharmacologic inhibition of EZH2 conferred cisplatin resistance to TGCT cells, while inhibition of H3K27me3 demethylases sensitized TGCT cells to cisplatin [15]. This implies that targeting the polycomb pathway may have therapeutic value in treating cisplatin-resistant TGCTs. We also reported that these cisplatin-resistant cells have global and genome-wide DNA hypermethylation compared to parental, cisplatin-sensitive cells [16]. The association between genome-wide DNA hypermethylation and cisplatin resistance has also recently been demonstrated in TGCT patients [17]. Our cisplatin-resistant cells also possessed hypomethylation at polycomb target gene promoters, suggesting a complex crosstalk exists between the polycomb and DNA methylation pathways that influences cisplatin sensitivity of TGCT cells [16].
The current study directly addresses mechanisms of TGCT hypersensitivity to DNA hypomethylating agents through the derivation of unique isogenic 5-aza-resistant cells models. Interestingly, transcriptome analysis strongly suggests that the polycomb pathway is also altered in 5-aza-resistant cells but in a manner opposite to that of isogenic cisplatin-resistant cells, whereby 5-aza resistance is associated with a dramatic downregulation of polycomb target gene expression. This reciprocal relationship was also evident for H3K27me3 and DNMT3B with 5-aza resistance associated with a dramatic downregulation of DNMT3B and upregulation of H3K27me3, the exact opposite situation noted in isogenic cisplatin-resistant cells. Further, 5-aza-resistant cells were more sensitive to cisplatin, while cisplatin-resistant cells were more sensitive to 5-aza. Finally, we provide evidence that DNMT3B may be an important upstream driver of epigenetic crosstalk in TGCT cells since DNMT3B depletion alone upregulated the polycomb pathway and conferred resistance to 5-aza and increased sensitivity to cisplatin. These data suggest that epigenetic remodeling mediated by polycomb and DNMT3B are major drivers of the unique pharmacology of TGCTs that includes hypersensitivity to cisplatin and hypomethylating agents.

Derivation of 5-aza resistant cells
All cells were cultured in DMEM (Gibco; Grand Island, NY, USA) with 10% FBS (Invitrogen; Carlsbad, CA, USA). The 2102EP cells are a human testicular cancer-derived embryonal carcinoma (EC) cell line purchased from ATCC (Manassas, VA, USA) and authenticated by ATCC with karyotyping and short tandem repeat profiling. Cells were frozen within 1 month of purchase and used within 2 months of resuscitation. Generation of 5-aza resistant cell lines was similar to our previously reported generation of cisplatin-resistant cells [15]. Parental cells were exposed to stepwise dosages of 5-aza deoxycytidine (Sigma; St. Louis, MO, USA) starting at 1.0 nM for three consecutive days and then allowed to recover for 1-2 weeks (Fig. 1A). This cycle was repeated 4-6 times with a final selection at 100 nM 5-aza. Clones were then derived from each pool with cloning cylinders. Cells with prefix AH1, AH2A, and AH2B were cloned from independently treated batches of cells. All clones were stably resistant since resistance to 5-aza was retained after passaging in 5-aza-free media for at least 4 months. Stable cisplatin-resistant cells 2102EP-B3, 2102EP-C1, and 2102EP-C4 were previously described [15].

Drug treatments and cell viability and proliferation assays
Cells were treated with the indicated dosages of 5-aza for three consecutive days and cells assayed for survival 1 day after the last dose (Fig. 1B). Cells were treated with cisplatin for 6 h and cells assayed for survival 3 days later (Fig. 1B). To assess cell viability, CellTiter-Glo (Promega; Madison, WI, USA) assays were performed as previously described [15]. For each cell line, four biological replicates were tested at each concentration, and experiments were repeated at least twice on different days. IC50 values were estimated from the best-fit dose-response model selected by residual standard error using GRAPHPAD PRISM software (San Diego, CA, USA). To estimate cell doubling times, equal number of cells were plated in 24-well plates and viable cell numbers determined for four consecutive days using the CellTiter-Glo assay. Doubling times were calculated with an exponential growth curve equation using GRAPHPAD PRISM software.

Lentiviral shRNA knockdown
Lentiviruses were produced by co-transfecting HEK293 cells with 10 lg of the viral packaging vector pCMV-dR8.2 and envelope vector pCMV-VSV-G (10 : 1 ratio) and 10 lg of lenti-shRNA or the control vector using Lipofectamine 3000. Lentiviral silencing shRNA targeting DNMT3B (TRCN0000437183, TRCN0000414235 Trimmomatic was used to remove low-quality bases from both ends LEADING ≤ 28 and TRAILING ≤ 28, respectively, with minimum length of 30. The reads in FASTQ format were aligned to human genome assembly NCBI GRCh38.p13 using STAR aligner as previously described [15]. Reads were counted and assigned to genes using featureCount. The LIMMA R package was used to identify differentially expressed genes [18]. Genes whose expression was not greater than 0.5 counts per million in at least two samples were removed, and the resultant filtered expression matrix was TMMnormalized. Benjamini-Hochberg false discovery rate (FDR) was used to correct for multiple hypotheses. Genes with FDR ≤ 0.01 and absolute fold change ≥ 1.5 were considered differentially expressed. In all cases, each 5-aza-resistant cell line was compared to the parental line. The 'ENHANCED VOLCANO' R package was used for visualization of volcano plots. RNA-seq data for 2102EP-C1 cells were from our prior study [15]. The RNA-seq datasets for the current study have been submitted to the NCBI Database of GEO Datasets under the accession numbers GSE172266 and GSE172458.

Downstream enrichment analysis
Gene set enrichment analysis (GSEA) from the Broad Institute was performed to identify enriched gene sets in each resistant cell line compared to parental cells [19]. The BART (Binding Analysis for Regulation of Transcription) tool in gene-set mode was used to predict the enrichment of transcription factor binding sites in differentially expressed genes [20]. GENEOVERLAP package from R BIOCON-DUCTOR was used to identify significant gene set overlap between common gene expression changes between 2102EP-AH2A2, 2102EP-C1, and 2102EP-DNMT3BKD cells and C2 gene sets from the MSigDB database [19,21].

Western analysis and real-time PCR
For western analysis, cells were lysed in radioimmune precipitation buffer and separated by SDS/PAGE.  Table S1.

Statistics
Two-tailed Student's t-tests were performed where appropriate using GRAPHPAD PRISM v6.0, and P-values indicative of nonsignificant P > 0.05 and significant *P ≤ 0.05 were determined. Mean and standard error of the mean was used to describe sample variability.

Derivation of acquired 5-aza-resistant testicular cancer cells reveals reciprocal drug sensitivities between 5-aza-and cisplatinresistant cells
We and others have previously documented that TGCT-derived EC cells undergo an acute cytotoxic response to low nanomolar levels of hypomethylating agents including 5-aza and guadecitabine [4][5][6][7][8]. Anticancer effects for most somatic cancer cells require doses of hypomethylating agents 100-to 1000-fold higher and the effects at these higher doses manifest as delayed antiproliferative responses rather than acute cytotoxic responses [22]. In order to better understand the mechanism of hypomethylating agent hypersensitivity of EC, we developed a panel of distinct clonal 5aza-resistant cell lines from parental 2102EP cells. 2102EP cells were chosen as it was difficult to generate 5-aza-resistant cells in other parental TGCT lines, including NT2/D1 and 833K without inducing differentiation. Cells were treated with increasing dosages of 5-aza starting at 1 nM for three consecutive days and allowed to recover for 1-2 weeks. Final selection was at 100 nM 5-aza, and cells were then cloned (Fig. 1A). Cells with prefix designation AH1, AH2A, and AH2B were cloned from independently treated batches of cells. Resistance to 5-aza was confirmed in dose response experiments with 3 day 5-aza treatments ( Fig. 1B,C). The 5-aza IC50s of resistant cells ranged from 35 to 100 nM compared to the 10 nM IC50 of parental cells (Fig. 1D). Resistance was maintained for at least 4 months of cell passage in 5-aza free media. In a prior report, we noted that one of the most 5aza-sensitive EC cell lines was the cisplatin resistant EC line 833K-CP [6]. This prompted us to compare and contrast 5-aza and cisplatin sensitivity in the newly generated 5-aza-resistant 2102EP cells with our previously derived isogenic cisplatin-resistant 2102EP cells [15]. We focused our studies on three 5-aza-resistant lines (2102EP-AH2A2, 21012EP-AH2B9, and 2102EP-AH1A5) and three cisplatin-resistant lines (2102EP-B3, 2102EP-C1, and 2102EP-C4) [15]. All three cisplatinresistant cell derivatives demonstrated increased sensitivity to 5-aza compared to parental 2102EP, while all three 5-aza-resistant cells lines demonstrated either Human testicular cancer derived EC cell line 2102EP was exposed to stepwise dosages of 5-aza starting at 1 nM and then allowed to recover for 1-2 weeks between doses. After selection in 100 nM 5-aza, cells were cloned. (B) Schematic of 5-aza and cisplatin treatment protocols for cell viability assays. CTG, CellTiter-Glo. (C) 5-aza-selected cell lines are stably resistant to 5-aza. Cell survival and viability were measured using CellTiter-Glo in parental or acquired 5-aza-resistant TGCT cells treated with indicated 5-aza doses for 3 days. (D) 5-aza IC50 values for parental and 5-aza-resistant cells were estimated from a best-fit dose-response model. (E) Cisplatin-resistant cell lines 2102EP-B3, 2102EP-C1, and 2102EP-C4 have increased sensitivity to 5-aza compared to parental cells. Cell survival and viability were measured using CellTiter-Glo in parental, 5-aza-resistant, and cisplatin-resistant TGCT cells treated with indicated 5-aza doses for 3 days.  sensitivity to cisplatin or a trend toward increased sensitivity to cisplatin ( Fig. 1E-H). This was despite the fact the all six cell lines had comparable levels of basal cell proliferation when drug was not present as compared to parental cells, indicating that differences in drug sensitivity was not due to gross alterations in cell proliferation (Fig. 1I).
3.2. Acquired 5-aza resistance in testicular cancer cells is associated with genome-wide alterations in levels of polycomb target genes, H3K27me3, H2AUbK119, and DNMT3B that is in an opposing manner compared to cisplatin resistant cells In order to understand mechanisms of 5-aza resistance, RNA-seq analysis was performed in biological triplicate in 2102EP parental and 2102EP-AH1A5, 2102EP-AH2B9, and 2102EP-AH2A2 cells. There was a substantial number of differentially expressed genes with a fold-change of 1.5 and FDR ≤ 0.01 in 5-aza-resistant cells as compared to parental 2102EP cells ( Fig. 2A). For AH1A2 and AH2B9 cells, there were 2091 and 2264 genes downregulated, respectively, compared to parental cells and 1754 and 2351 genes upregulated, respectively. In AH1A5 cells, there was substantially more genes downregulated than upregulated compared to parental cells at this cutoff, 2052 vs 544 genes, respectively. The identity of these differentially expressed genes is provided in Table S2.
We next performed unbiased GSEA comparing each aza-resistant cell line with parental 2102EP cells. The top 50 positively and negatively enriched gene sets (from over 5529 curated gene sets) for each cell line ranked by normalized enrichment score is provided in Table S3. Interestingly for genes downregulated in the 5-aza-resistant cells, there was a highly significant enrichment for multiple gene sets related to polycomb target genes (Fig. 2B). For example, of the top 50 ranked gene sets for downregulated genes, 13, 13, and 10 independent gene sets related to polycomb target genes were noted for AH2A2, AH2B9 and AH1A5 cells, respectively (Fig. 2B,C). This was by far the most frequent and significant gene set family that emerged from GSEA, suggesting that the 5-azaresistant cells have globally downregulated expression of polycomb target genes. Interestingly, we identified previously in 10 isogenic cisplatin-resistant EC cells including those derived from parental 2102EP, the exact opposite pattern of polycomb gene expression with polycomb target genes being among the most significantly upregulated gene sets [15].
BART (binding analysis for regulation of transcription) predicts functional transcription factor binding based on more than 6000 ChIP-seq datasets for over 400 factors in human and mouse cells [20]. The BART analysis of genes downregulated in 5-aza-resistant cells, as compared with parental controls, strongly predicted polycomb complex component binding. The number of distinct PRC1/PRC2-related components contained in the top 20 predicted transcription factor binders for each line ranged from three to five and included EZH2, JARID2, SUZ12, and EED (Table S4). These results further suggest the polycomb pathway is a major mediator in regulating downregulated genes in the 5-aza-resistant cells. In contrast, BART analysis of cisplatin-resistant cells demonstrated polycomb component binding enrichment in upregulated genes [15].
To further probe the reciprocal nature of 5-aza-and cisplatin-resistant cells, immunoblot and RT-PCR analyses were performed. We focused on the three 5-azaand three cisplatin-resistant 2102EP lines used for current and prior RNA-seq analysis ( Fig. 2A) [15]. Remarkably, all three 5-aza-resistant 2102EP cell lines had a dramatic increase in the polycomb marks H3K27me3 and H2AUbK119 compared to parental cells, while, as previously shown, the cisplatin-resistant cells had decreased expression of these marks (Fig. 3A) [15]. DNMT3B levels were also assessed, since we have previously shown that hypersensitivity of EC cells to 5aza is dependent on high-level expression of DNMT3B [5,6]. Interestingly, all three 5-aza-resistant cells had dramatically decreased expression of DNMT3B, while all three cisplatin-resistant cells had a dramatic increase in DNMT3B expression compared to parental cells (Fig. 3A). In contrast, there was minimal change in DNMT1 expression among the cell lines.
Real-time PCR analysis demonstrated substantial downregulated expression of DNMT3B in 5-azaresistant cells and a dramatic upregulated expression of DNMT3B in cisplatin-resistant cells, consistent with immunoblot analysis and suggesting regulation of DNMT3B in these cell models is at the level of mRNA regulation (Fig. 3B). In contrast, there were only small changes in mRNA for DNMT1 and DNMT3A between the cell lines. PRC1 component BMI1 was upregulated in 5-aza-resistant cells while downregulated in cisplatin-resistant cells consistent with the changes observed in H2AUbK119 (Fig. 3C). Expression of EZH2 also followed this same general trend but was less consistently changed between the resistant cells (Fig. 3C). The H3K27me3 demethylase KDM6B was upregulated in cisplatin-resistant cells consistent with the downregulation of H3K27me3 in cisplatin-resistant cells, but unchanged in the 5-aza-resistant cell lines (Fig. 3D). In contrast, expression of the H3K27me3 demethylase KDM6A was relatively unchanged between   (Fig. 3D). These data suggest that the reciprocal alterations seen in H3K27me3 and H2AUbK119 could be related to changes in expression of certain polycomb components.

DNMT3B knockdown results in induction of H3K27m3, EZH2, and BMI1 expression and confers 5-aza resistance and cisplatin sensitization to TGCT cells
Our prior work suggested a role for global remodeling of DNA methylation in the upregulation of polycomb target gene expression in cisplatin-resistant TGCT cells [16]. To investigate whether changes in DNMT3B play a causal role in altering polycomb signaling and 5-aza and cisplatin sensitivity in TGCT cells, DNMT3B was targeted by shRNA in parental 2102EP cells (Fig. 4A,  B). Note that in some experiments shRNA1 only mediated a partial knockdown of DNMT3B. However, DNMT3B knockdown was stable and persisted for extended cell culture intervals up to several months. DNMT3B shRNA resulted in upregulation of H3K27me3 levels and BMI1 and EZH2 protein expression in 2102EP cells (Fig. 4B). BMI1 induction was also noted at the mRNA level where little change was noted in mRNA levels of EZH2, KDM6A, and KDM6B upon DNMT3B knockdown (Fig. 4C). DNMT3B shRNA did not affect levels of DNMT1 and DNMT3A (Fig. 4C). Notably, DNMT3B shRNA also resulted in resistance to 5-aza, as we have shown previously [5,6], but importantly these cells also demonstrated a sensitization to cisplatin (Fig. 4D). These results suggest that repression of DNMT3B expression alone is sufficient to induce H3K27me3 levels in TGCT cells and that knockdown of DNMT3B mimics the situation seen in 5-azaresistant cells, namely, 5-aza resistance and heightened cisplatin sensitivity. The results suggest that alterations in DNMT3B may be driving the crosstalk between DNA methylation and polycomb signaling noted in cisplatin-resistant and 5-aza-resistant TGCT cells.

Substantial genome-wide overlap between altered polycomb target gene expression in 5-aza-resistant, cisplatin-resistant, and DNMT3Bknockdown TGCT cells
Our findings suggest a reciprocal regulation of polycomb signaling in cisplatin-resistant and 5-aza-resistant  We chose 2102EP-AH2A2 and 2102EP-C1 cells as representative 5-aza and cisplatin cell lines, respectively, for this analysis. Interestingly, at a cutoff of 1.5-fold and FDR ≤ 0.01, there were many more genes downregulated than upregulated in 2102EP-DNMT3BKD  (Fig. 5A). This is interesting since it conflicts with the canonical role of DNMT3B in facilitating gene silencing by promoter DNA methylation [23]. This trend was present but to a lesser extent in 5-aza-resistant 2102EP-AH2A2 cells with 2091 genes downregulated and 1754 genes upregulated compared to parental cells and reversed to a certain extent in cisplatin-resistant 2102EP-C1 cells, with 2030 upregulated genes and 1587 downregulated genes compared to parental cells (Fig. 5A). This suggests that gene expression in 2102EP-DNMT3BKD cells may be more similar to 2102EP-AH2A2 cells than to 2102EP-C1 cells. Differentially expressed genes in 2102EP-DNMT3BKD cells are provided in Table S5. Gene set enrichment analysis demonstrated that genes downregulated in 2102EP-DNMT3BKD cells were highly significantly enriched for multiple polycomb target gene sets, further supporting that DNMT3B-knockdown mimics global expression changes in 5-aza-resistant cells. Of the top 50 ranked gene sets for downregulated genes in 2102EP-DNMT3BKD cells compared to PLK0.1 control cells, 15 gene sets were related to polycomb target genes (Fig. 5B,C). There were no major trends seen for upregulated genes in 2102EP-DNMT3BKD cells. The top 50 positively and negatively enriched gene sets for 2102EP-DNMT3BKD ranked by normalized enrichment score is provided in Table S6. This analysis strongly demonstrates that DNMT3B knockdown alone closely mimics the global downregulation of polycomb target genes seen in 5-aza cells (Fig. 2B,C). We previously reported the reciprocal pattern for cisplatin-resistant cells including 2102EP-C1 where these same polycomb target gene sets were enriched in upregulated genes [15]. BART analysis also predicted polycomb component binding at the promoters of the downregulated genes in 2102EP-DNMT3BKD cells, which included EZH2, JARID2, and SUZ12 in the top 20 predicted binders (Table S7).
Thus far, the data suggest there is a reciprocal relationship in polycomb target genes expression between 5-aza-and cisplatin-resistant cells with polycomb targets downregulated in 5-aza-resistant cells and upregulated in cisplatin-resistant cells. Furthermore, the data suggest that downregulation of polycomb target genes also occurred in DNMT3B-knockdown cells. To investigate this relationship further, we analyzed the degree of overlap in differentially expressed genes (1.5-fold change, FDR ≤ 0.01 compared to parental control cells) between 2102EP-AH2A2, 2102EP-C1, and 2102EP-DNMT3BKD cells. The identity of the overlap genes is provided in Table S8. Approximately 47% of the genes downregulated in 5-aza-resistant cells and 49% of the genes upregulated in cisplatin-resistant cells overlapped between the two cell lines (Fig. 6A). Remarkably, of the genes downregulated in DNMT3B knockdown cells, 85% were also downregulated in 5aza-resistant cells and 75% were also upregulated in cisplatin-resistant cells (Fig. 6A). These overlap genes were highly enriched in polycomb target genes sets from the GSEA database ( Fig. 6B and Table S9). These data suggest that DNMT3B-knockdown results in downregulation of the same polycomb target genes downregulated in 5-aza-resistant cells and upregulated in cisplatin-resistant cells.

Discussion
Testicular germ cell tumors can be cured at a high-rate with conventional cisplatin-based therapies. However, patients resistant to cisplatin have a very poor prognosis [3]. The mechanisms to account for the sensitivity and resistance of TGCTs to cisplatin are unclear [24][25][26]. Although there are no effective alternative or targeted therapies available for cisplatin refractory TGCT patients, recent preclinical and early clinical trial data suggest a subset of cisplatin refractory TGCTs may be hypersensitive to DNA hypomethylating agents [4][5][6][7][8][12][13][14]. The mechanisms accounting for the hypersensitivity of TGCT cells to hypomethylating agents are entirely unknown apart from a reported dependence on high-level expression of DNMT3B [6][7][8][9]. Very few studies have addressed the issue of resistance to hypomethylating agents, and these are mainly centered on acute myeloid leukemia [27,28]. Biomarkers for both cisplatin and hypomethylating agents in TGCT patients are critically needed as early trials have indicated that only a subset of cisplatin refractory patients may benefit from hypomethylating agents [12][13][14].
The present study directly addresses mechanisms of TGCT hypersensitivity to DNA hypomethylating agents through the derivation of unique isogenic 5aza-resistant cell models. Transcriptome analysis revealed that 5-aza resistance is associated with H3K27me3-mediated downregulation of polycomb target genes and downregulation of the expression of DNMT3B. In addition, de novo pharmacogenomic studies in the isogenic 5-aza-resistant cells and prior isogenic cisplatin-resistant cells revealed that the polycomb pathway and DNA methylation are opposing interconnected epigenetic drivers of cisplatin and hypomethylating agent sensitivity in TGCTs. Increased 5-aza sensitivity in cisplatin-resistant cells validates prior anecdotal evidence that cisplatin resistance may sensitize TGCT cells to hypomethylating agents and suggests that polycomb/DNA methylation remodeling is one mechanism to account for hypomethylating therapy hypersensitivity in cisplatin refractory TGCTs. Reciprocally and unexpectedly, 5-aza-resistant cells were more sensitive to cisplatin. Together, these findings provide further rationale for the combined use of these two therapeutics for treating TGCTs. We have also begun to generate 5-aza-resistant cells from other parental TGCT cells, although this has been difficult while avoiding terminal differentiation. Some of these lines also appear to have similar polycomb alterations as in 5-aza-resistant 2102EP cells, but to a lesser extent. This suggests that there are likely other mechanisms of 5-aza resistance in TGCT cells. We previously showed that H3K27me3 was downregulated with cisplatin resistance and that cisplatinresistant cells have global DNA hypermethylation [15,16]. In contrast, less abundant hypomethylation in cisplatin-resistant cells was associated with H3K27me3 and polycomb target gene loci as part of a bidirectional shift between gene promoter and gene body DNA methylation that was associated with upregulation of polycomb targets and downregulation of tumor suppressor genes [16]. Our data thus suggest a complex relationship exists between DNA methylation and H3K27me3, the remodeling of which may have been selected for during acquired cisplatin and 5-aza resistance. The interconnection between polycomb and DNA methylation has been studied in other cell contexts including pluripotent embryonal stem cells [29][30][31][32]. It would be interesting to directly test this relationship in future work with H3K27me3 and polycomb cistrome and chromatin accessibility ATAC-seq analyses in TGCT cells. Further, important future work will be to confirm whether the epigenetic and gene expression changes highlighted here extend to clinical cisplatin-resistant TGCTs.
Recent studies suggest that global DNA hypermethylation may be a key factor in mediating cisplatin resistance in TGCTs [16,17]. The current work reveals that DNMT3B is also reciprocally altered with 5-aza and cisplatin resistance. Prior studies have demonstrated that knockdown of highly expressed DNMT3B in TGCT cells confers resistance to hypomethylating agents [5][6][7][8]. Interestingly, the 5-aza-resistant cells derived in the present study spontaneously downregulated DNMT3B, while cisplatin-resistant cells greatly induced DNMT3B expression. This again suggests an interconnected epigenetic mechanism linking 5-aza and cisplatin sensitivity/resistance in TGCTs involving polycomb and DNA methylation. Finally, DNMT3B knockdown alone phenocopied 5-aza resistance including repression of polycomb target genes, 5-aza  resistance, and cisplatin sensitization, suggesting that DNMT3B plays an important role in the crosstalk between DNA methylation and polycomb in TGCT cells. We found that DNMT3B knockdown induced the protein expression of EZH2 and BMI1; however, only BMI1 mRNA was increased, suggesting that DNMT3B knockdown may also have posttranscriptional effects on polycomb components. Interestingly, DNMT3B knockdown in parental TGCT cells resulted in many more downregulated genes than upregulated genes, suggesting that the high levels of DNMT3B in TGCT cells may have functions beyond the canonical role of promoting repression of gene expression [23]. Interestingly, the H3K27me3 demethylase KDM6B was upregulated in cisplatin-resistant cells. Targeting H3K27me3 demethylases like KDM6B could be a strategy for treating cisplatin-resistant tumors. We have previously demonstrated that inducing H3K27 methylation with GSK-J4, an inhibitor of both KDM6A and KDM6B H3K27me3 demethylases, increased cisplatin sensitivity of cisplatin-resistant TGCT cells [15]. Upregulation of KDM6B also provides a mechanistic explanation for the low basal levels of H3K27me3 in cisplatin-resistant 2102EP cells. It will be important to validate our findings in cisplatinsensitive versus cisplatin-resistant nonseminoma patients. Patient material could be interrogated for H3K27me3 levels and BMI1, EZH2, KDM6B, and DNMT3B levels by immunohistochemistry. We predict that patients resistant to cisplatin may have high levels of DNMT3B and KDM6B, and low levels of H3K27me3, EZH2, and BMI1. These patients would be predicted to be hypersensitive to 5-aza and thus candidates for hypomethylation therapy with or without cisplatin. These patents would also be candidates for combination therapy with a KDM6A/KDM6B inhibitor and cisplatin.
The precise nature of how H3K27me3 and DNA methylation alters cisplatin and 5-aza sensitivity will require further studies. Interestingly, DNMT3B knockdown alone was able to repress many of the same polycomb target genes that were repressed in 5-azaresistant cells and upregulated in cisplatin-resistant cells. These data suggest that alterations in chromatin context as revealed by polycomb target gene expression and likely downstream of DNMT3B/DNA methylation are a driving factor in TGCT cell response to cisplatin and 5-aza. Whether these results hold true for other cancer therapeutics in TGCT cells awaits further study. In other cell contexts, studies have shown that H3K27me3 and other histone marks have the ability to recruit or regulate distinct components of the DNA damage and repair pathways [33][34][35][36][37][38][39][40]. It will be interesting to assess how altering DNMT3B and polycomb components effect these endpoints in TGCT cells in future studies. Another potential mechanism to consider is the direct regulation of DNA damage response and repair gene expression by DNA methylation and polycomb remodeling in TGCT cells as recent reports have shown some genes involved in these pathways can be regulated by DNA methylation in TGCT cells and tumors [41][42][43]. However, we did not detect a clear signal for these family of genes in our transcriptome analyses.
Our ongoing work is centered on further elucidating the mechanism by which DNA and histone methylation are influencing 5-aza and cisplatin sensitivity. However, these mechanisms are likely complex and are beyond to scope of the current study. Our two leading hypothesis currently are that these epigenetic marks may be directly influencing the recruitment of DNA repair or DNA damage pathway components, or alternatively, a comparable amount of DNA damage is occurring but cells are responding differently based on the chromatin context, for example, more open chromatin allowing for more binding of p53 to target gene promoters. Thus far in appears that there is a comparable amount of DNA damage and repair in parental, 5-aza-, and cisplatin-resistant cells basally and in response to cisplatin and 5-aza.

Conclusions
In summary, we have shown that 5-aza-and cisplatinresistant TGCT cells reciprocally regulate H3K27me3 and DNMT3B and that this results in reciprocal sensitivity to 5-aza and cisplatin. In that, 5-aza-resistant cells are more sensitive to cisplatin and cisplatinresistant cells are more sensitive to 5-aza. Together, our findings support the involvement of a novel crosstalk mechanism between two epigenetic pathways, global DNA methylation, and polycomb, in mediating the unique hypersensitivity of TGCTs to cisplatin and hypomethylating agents. This is precisely in line with the growing realization that the etiology and progression of TGCTs may be especially dominated by epigenetic dysregulation. The cell models reported here may have value in identifying biomarkers to guide the use of cisplatin and hypomethylating agents for the treatment of TGCT patients.

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article. Table S1. RT-PCR primers used in this study. Table S2. Differentially expressed genes in 5-aza resistant cells with FDR ≤ 0.01 and absolute fold change ≥ 1.5. Table S3. Top 50 positively and negatively enriched gene sets in 5-aza resistant cells compared to parental cells by GSEA. Table S4. Top 20 predicted transcription factors binding to promoters of downregulated genes in 5-aza resistant cells. Table S5. Differentially expressed genes in 2102EP-DNMT3BKD cells compared to shRNA control cells with FDR ≤ 0.01 and absolute fold change ≥ 1.5. Table S6. Top 50 positively and negatively enriched gene sets in 2102EP-DNMT3BKD cells compared to shRNA control cells by GSEA. Table S7. Top 20 predicted transcription factors binding to promoters of downregulated genes in 2102EP-DNMT3BKD cells. Table S8. Differentially expressed genes common to 5aza resistant down/cisplatin resistant up, 5-aza resistant down/DNMT3BKD down, or cisplatin resistant up/DNMT3BKD down with ≥ 1.5 fold-change and FDR ≤ 0.01. Table S9