The Role of DNA/Histone Modifying Enzymes and Chromatin Remodeling Complexes in Testicular Germ Cell Tumors

It is well established that cancer cells exhibit alterations in chromatin structure and accessibility. Indeed, the dysregulation of many protein-coding players with enzymatic activity (DNA and histone-modifying enzymes) and chromatin remodelers have been depicted in various tumor models in recent years. Still, little attention has been directed towards testicular germ cell tumors (TGCTs)—representing the most common neoplasm among young adult Caucasian men—with most studies focusing on exploring the role of DNA methyltransferases (DNMTs) and DNA demethylases (TETs). TGCTs represent a complex tumor model, associated with developmental and embryogenesis-related phenomena, and display seldom (cyto)genetic aberrations, leaving room for Epigenetics to explain such morphological and clinical diversity. Herein, we have summarized the major findings that were reported in literature regarding the dysregulation of DNA/histone-modifying enzymes and chromatin remodelers in TGCTs. Additionally, we performed in silico analysis of The Cancer Genome Atlas database to find the most relevant of those players in TGCTs. We concluded that several DNA/histone-modifying enzymes and chromatin remodelers may serve as biomarkers for subtyping, dictating prognosis and survival, and, possibly, for serving as targets of directed, less toxic therapies.


Testicular Germ Cell Tumors in Brief
Testicular germ cell tumors (TGCTs) comprise more than 95% of testicular neoplasms and they are grouped in two major families according to the most recent World Health Organization classification: the germ-cell neoplasia in situ (GCNIS)-related tumors (the most frequent, which include Seminomas (SEs) and Non-Seminomatous Tumors (NSTs), two subgroups with very distinct behavior and clinical impact), and the GCNIS-unrelated ones [1,2].
Despite representing only 1% of male cancer worldwide, they constitute the most common cancer afflicting Caucasian men between 15-44 years old, with the Western lifestyle contributing to a rising incidence. They also exhibit outstanding cure rates and a decreasing mortality trend, in response to multimodal treatments. However, many issues are left unresolved and they deserve our attention, namely the substantial proportion of patients with disseminated disease that relapse The field of Epigenetics in TGCTs has been expanding in the last years, with a growing number of publications on the topic. Most studies have focused on methylation [26][27][28][29] and on microRNAs (miRs) [30][31][32][33], where the major breakthroughs in TGCTs have taken place. Protein-coding epigenetic players, including DNA-modifying enzymes, histone-modifying enzymes, and ChRCs, have been explored in diverse tumor models in the recent years; however, little attention has been paid to TGCTs. Hence, we have conducted a PubMed search with the query "testicular germ cell tumors" AND "(protein-coding epigenetic players)", with no time period restraints. Only papers that were written in English were considered. All abstracts were read in order to select those papers truly related to the topic. Table 1 displays the result of our query, listing original articles addressing the role of these players in TGCTs pathogenesis and summarizing their major findings . Despite the overwhelming evidence that stem cells and germ cells display dynamic epigenetic modifications during differentiation and spermatogenesis, including changes in the expression of these enzymes (e.g., with DNA methyltransferases more expressed in spermatogonia and histone methyltransferases mainly in spermatocytes) [10,[56][57][58][59][60][61][62], there is still a lack of studies on the role of these players and related modifications in TGCTs (especially in certain families, with most studies published so far focusing on DNA-modifying enzymes).   [45] Upward (↑) and downward (↓) arrows stand for up-and downregulation, respectively. Abbreviations: AR-androgen receptor; BSP-bisulfite sequencing PCR; CH-choriocarcinoma; ChIP-chromatin immunoprecipitation; CoIP-co-immunoprecipitation; EC-embryonal carcinoma; ER-α-estrogen receptor alpha; FFPE-formalin-fixed paraffin embedded; GCNIS-germ cell neoplasia in situ; IB-immunoblot; IF-immunofluorescence; IHC-immunohistochemistry; ISH-in situ hybridization; LCT-Leydig cell tumor; MSP-methylation specific PCR; NST-non-seminomatous tumor; PCR-polymerase chain reaction; pS-ATM-S1981-phosphorylated ATM; RT-PCR-real-time PCR; SB-southern blot; SE-seminoma; TGCT-testicular germ cell tumor; TMA-tissue microarray; WB-western blot; 5AZA-5-aza-2 deoxycytidine.
Cancers 2019, 11, 6 6 of 22 In this line, we performed an in silico analysis of the publicly available The Cancer Genome Atlas (TCGA) database for TGCTs, regarding the diverse families of both DNA-modifying, histone modifying, and chromatin remodeling enzymes. We ultimately aimed to identify alterations in these players, exposing those potentially being the most relevant, and finally, providing the reader with a list of the most promising biomarkers to be further validated in independent patient cohorts. For this, we used the online resource cBioPortal for Cancer Genomics [63] and a user-defined entry gene set for all of these players. Statistical analysis with the available data was performed with Microsoft Excel 2016, GraphPad Prism 6 and IBM SPSS Statistics v.24. Distribution of continuous variables between groups was compared using the nonparametric Mann-Whitney U test. Co-occurrence/mutual exclusivity of alterations in pair of genes was assessed with the odds ratio (OR). Biomarker performance was assessed through the receiver operating characteristics (ROC) curve construction. ROC curves were constructed plotting sensitivity (true positive) against 1-specificity (false positive). A cut-off was established by the Youden's index method [64,65]. Area under the curve (AUC) and biomarker performance parameters, including sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy, were ascertained. Survival curves were plotted with the Kaplan-Meier method and log-rank test was used for survival analysis. A p-value that was equal or inferior to 0.05 was considered to be significant. Bonferroni's correction was applied to multiple pairwise comparisons.
A summary of major findings of this analysis is depicted in Table 2.
Regarding subtype discrimination, SEs disclosed significantly lower expression levels of all three enzymes (p < 0.0001) as compared to NSTs. The best performance was obtained for DNMT3A, rendering an AUC of 0.88. Interestingly, DNMT3A and DNMT3B expression was remarkably different among SEs and ECs (being strongly upregulated in the latter, p < 0.0001), with these enzymes discriminating among subtypes with AUCs of 0.98 and 0.99 ( Figure 1A). Stage I patients also exhibited lower DNMT3A (p = 0.0006) and DNMT3B (p = 0.0011) expression levels when compared to stage II/III patients. No significant associations with overall (OS) or disease/progression-free (D/PFS) survival were depicted.
Regarding subtype discrimination, SEs showed significantly higher expression levels of TET2 when compared to NSTs (p < 0.0001), achieving an AUC = 0.79. Again, the differences in expression between SEs and ECs were quite remarkable (with upregulation in SEs, p < 0.0001), rendering an AUC = 0.98 ( Figure 1B). Stage I disease also expressed significantly higher levels of TET2 when compared to stages II/III (p = 0.0096). No significant associations with OS or D/PFS were depicted.
MAIN CONCLUSIONS: SEs display lower expression levels of DNMTs and higher expression levels of TET2, compatible with the described hypomethylated genome pattern of these tumors when compared to NSTs [29]. The expression pattern of these enzymes is completely opposite in ECs (with higher expression of DNMTs and lower expression of TET2), a finding that might prove useful in discriminating these two forms of TGCT, which have very different aggressiveness and prognosis. Also, there is room for prognostic impact of these markers, as DNMTs/TET2 are upregulated and downregulated, respectively, in advanced stage disease. These findings are in accordance with most studies published so far, which also report DNMTs' overexpression in ECs [36,40,42,47,48,50] and of TETs in SEs [45] (Table 1).

A. MYST Family
The MYST family is the largest family of histone acetyl transferases (HATs), being responsible for acetylating the epsilon-amino group of lysine, direct PTM phenomena. HATs are, in general, qualified as transcription activators. The MYST family, specifically, is characterized by a distinct conserved domain, containing a C 2 HC zinc finger and an acetyl-CoA binding site. It includes five members in humans: [66,67]. Globally, these genes are deregulated in 68/156 (44%) of TGCT samples. Most alterations consisted of mRNA upregulation (69%), followed by mRNA downregulation (22%). KAT6A was the member showing more frequent deregulation (in 20% of samples), followed by KAT5 and KAT7 (11% and 10% of cases, respectively). Alterations in KAT6A and KAT7 were significantly mutually exclusive (p = 0.03, logOR < −3), but not after Bonferroni's correction.
Regarding subtype discrimination, SEs showed significantly higher KAT6A and KAT6B expression levels (p < 0.0001 for both) when compared to NSTs. On the contrary, KAT5 and KAT8 were significantly downregulated in SEs as compared to NSTs (p < 0.0001 and p = 0.003, respectively). As a biomarker for discriminating SEs vs. NSTs, the best performance was rendered by KAT5, displaying an AUC = 0.75. Also, patients with stage I disease showed significantly higher expression levels of KAT6B and lower expression levels of KAT8 when compared to stages II/III (p = 0.004 and p = 0.02).

B. GNAT Family
The GNAT (GCN5-related N-acetyltransferase) family is also involved in the reversible lysine acetylation of proteins such as histones and includes two main members, KAT2A/GCN5 and KAT2B/PCAF, and also others like ATAT1/MEC17, KAT1/HAT1, KAT9/ELP3, and AT1/SLC33A1. They are characterized by sharing a domain with four conserved motifs A-D [24,68,69]. Globally, they are deregulated in 82/156 (53%) of TGCT samples, almost always due to mRNA upregulation (94%). The most commonly deregulated enzyme was KAT9 (in 36% of TGCTs-43% of SEs and 28% of NSTs), the remainder only being deregulated in less than 10% of tumor samples. KAT2A and KAT9 expression was found to be mutually exclusive (p = 0.026, logOR < −3), but it did not remain significant after Bonferroni's correction.
KAT1, KAT2A ( Figure 1C), KAT2B, and KAT9 mRNA expression levels were significantly higher in SEs when compared to NSTs (p < 0.0001, p < 0.0001, p < 0.0001 and p = 0.0012), with the best discrimination performance disclosed by KAT2A (AUC = 0.78). On the other hand, SLC33A1 and ATAT1 were significantly downregulated in SEs as compared to NSTs (p < 0.0001, p = 0.0370). Also, KAT2B was significantly upregulated in patients with stage I disease compared to stages II/III (p = 0.0037). No impact on survival analysis was depicted.
SIRT4 and SIRT5 expression was significantly lower and higher in SEs as compared to NSTs, respectively (p < 0.0001 for both); still, they rendered only modest AUCs of 0.77 and 0.72. Patients with stage II/III disease showed SIRT4 overexpression (p = 0.01). No significant impact on survival was depicted.
MAIN CONCLUSIONS: SEs display higher expression levels of most acetylases and lower expression levels of most deacetylases, compatible with an acetylated, transcription-prone genome characteristic of these tumors. Again, important differences in the expression between SEs and NSTs (and especially between SEs and ECs) were noticed for HDACs (in accordance with the studies finding higher expression of HDACs in NST subtypes, such as choriocarcinoma and EC [53,54]), which might prove valuable in the clinical setting for discriminating these subtypes with different prognosis and treatment approaches. Regarding deacetylation, HDACs seem to have more impact in TGCTs biology than SIRTs. They were also found to associate with higher stage disease, as opposed to previous findings [53], meaning that studies in larger cohorts may be needed to ascertain their prognostic value.
SEs exhibited significantly higher mRNA expression levels than NSTs (p < 0.0001), rendering an AUC = 0.79 for discriminating among both subtypes. Interestingly, DOT1L expression differed between SEs and pure ECs, with the former displaying higher levels (p < 0.0001); for these, an AUC = 0.87 was depicted. No association with the disease stage or survival was found.

Arginine Methyltransferases (PRMTs)
Another group of enzymes introduces methyl groups preferentially into arginine residues. There are nine PRMTs encoded in human genome (PRMT1-9) [77], and they show deregulation in 81/156 (52%) TGCTs, mainly by mRNA upregulation (50.6%) and also amplification (24.7%). The most commonly deregulated are PRMT8 and PRMT2, in 21% and 13% of the samples, respectively; in particular, all alterations in PRMT8 consisted of amplifications, except for one case with a missense mutation. No significant co-occurring or mutual exclusive alterations were found.
SEs showed significantly lower expression levels of PRMT8 as compared to NSTs (p < 0.0001); PRMT8 allowed for an AUC = 0.83 for discriminating both subtypes. On the contrary, PRMT9 was overexpressed in SEs compared to NSTs, and an AUC = 0.75 was obtained. No associations with disease stage were depicted. However, patients with alterations in PRMT4 (also known as CARM1) showed significantly poorer D/PFS (log rank, p = 0.003) (Figure 2A). 76]. It is deregulated in 14/156 (9%) TGCTs, mainly by mRNA upregulation (79%). Two missense mutations were found.
SEs exhibited significantly higher mRNA expression levels than NSTs (p < 0.0001), rendering an AUC = 0.79 for discriminating among both subtypes. Interestingly, DOT1L expression differed between SEs and pure ECs, with the former displaying higher levels (p < 0.0001); for these, an AUC = 0.87 was depicted. No association with the disease stage or survival was found.

Arginine Methyltransferases (PRMTs)
Another group of enzymes introduces methyl groups preferentially into arginine residues. There are nine PRMTs encoded in human genome (PRMT1-9) [77], and they show deregulation in 81/156 (52%) TGCTs, mainly by mRNA upregulation (50.6%) and also amplification (24.7%). The most commonly deregulated are PRMT8 and PRMT2, in 21% and 13% of the samples, respectively; in particular, all alterations in PRMT8 consisted of amplifications, except for one case with a missense mutation. No significant co-occurring or mutual exclusive alterations were found.
SEs showed significantly lower expression levels of PRMT8 as compared to NSTs (p < 0.0001); PRMT8 allowed for an AUC = 0.83 for discriminating both subtypes. On the contrary, PRMT9 was overexpressed in SEs compared to NSTs, and an AUC = 0.75 was obtained. No associations with disease stage were depicted. However, patients with alterations in PRMT4 (also known as CARM1) showed significantly poorer D/PFS (log rank, p = 0.003) (Figure 2A). Regarding the removal of methyl groups from lysine residues, two classes of enzymes are considered, again based on their dependence of co-factors: the 2OG/Fe 2+ -dependent dioxygenases that contain a JmjC domain, and also the FAD-dependent amine oxidases. The former is the major family of KDMs, being composed of 29 different demethylase proteins [74,75,78], which are deregulated in 132/156 (85%) of TGCTs, mainly by mRNA upregulation (41%) and multiple alterations (40%). The most commonly altered enzyme was KDM5A (in 21% of TGCTs, by amplification in all but three tumors) and KDM7A (in 19% of TGCTs, always by mRNA upregulation). Two pairs, KDM4D + KDM4E and JARID2 + KDM2B, tended to show co-occurring alterations (logOR > 3, adjusted p-value 0.014; and logOR 2.446, adjusted p-value 0.015).
Patients with stage I disease depicted higher RIOX1 and KDM2A expression levels when compared to stages II/III (p = 0.0069 and p = 0.0441, respectively).
SEs overexpressed LSD2 as compared to NSTs (p < 0.0001), but the discrimination power was rather modest (AUC = 0.73), and no differences were depicted for LSD1. Patients with stage I disease showed higher LSD2 expression when compared to NSTs (p = 0.0062). No significant impact on survival was depicted.
MAIN CONCLUSIONS: SEs display higher expression levels of enzymes that establish activating modifications, such as KDM4D, KDM3A, KMT2B/C/D, SETD1A, and lower expression of those which establish repressive marks, like EHMT2 and EZH2 (despite the latter being reported in another work not to display significant differences in expression among SEs and NSTs [37]). More studies are needed to fully understand the interaction of all these enzymes, their respective modifications, and how they influence TGCTs biology.
MAIN CONCLUSIONS: ATM and AURKB, the two kinases already studied in TGCTs [43,44], seem to have impact on TGCTs biology, showing differential expression between SEs and NSTs. More studies are needed to fully uncover the role of these enzymes in TGCTs.

Ubiquitin Ligases
Histone ubiquitination (and deubiquitination) are less well explored PTMs, but they have been shown to crosstalk with the remaining modifications having impact on DNA repair and gene expression. The histone proteins most commonly conjugated with ubiquitin (especially monoubiquitination) are H2A and H2B. The enzymes catalyzing this reaction are ubiquitin ligases; they include RING1, RNF2, BMI1, UBE2D3, RNF20, RNF40, UBE2A, UBE2B, and UBE2E1 [81,82] and are deregulated in 73/156 (47%) of TGCTs, mainly by mRNA upregulation (68%). The most frequently altered enzymes were RING1 and RNF40 (10% for both). No co-occurring or mutually exclusive pairs of enzymes with alterations were depicted.
SEs showed significantly higher expression levels of USP16, reaching an AUC = 0.89; and, significantly lower levels of BAP1 ( Figure 1I), achieving an AUC = 0.84 (p < 0.0001). Also, stage I tumors displayed USP16 overexpression (p = 0.0001) when compared to stages II/III. MAIN CONCLUSIONS: Ubiquitination has not been explored thus far in TGCTs, but their differential expression among SEs and NSTs (reaching high AUC values) suggest that they might play an important role in tumorigenesis.

CHROMATIN REMODELING ENZYMES
ChRCs represent a wide range of proteins that have the common ability of inducing chromatin changes in a dynamic way, including nucleosome sliding, conformational modification of the nucleossome itself, and switching the composition of the histone octamers. Thus, they alter both histones and affect the histone-DNA interaction in the nucleosome. Through ATP hydrolysis, these players are grouped in four major families according to their core structure and presence of certain domains.
SEs showed higher BRG1 and BRM expression levels (p < 0.0001), but lower SMARCD1 levels (p = 0.0020), as compared to NSTs. The best discrimination was achieved by BRM, rendering an AUC = 0.84. Stage I tumors showed significantly higher expression levels of BRM compared to stages II/III (p = 0.0004).
Regarding subtype discrimination, SEs showed significantly higher expression levels of CHD1, CHD2, CHD6, and CHD7 as compared to NSTs (p < 0.0001), while exhibiting lower CHD4 expression (p < 0.0001). The best discrimination performance was achieved by CHD1 and CHD7 (AUC = 0.81 for both). Also, patients with stage I disease showed higher CHD7 and CHD8 transcript levels when compared to stages II/III (p = 0.0009 and p = 0.0026). Cases with CHD8 and CHD2 alterations showed poorer D/PFS (p = 0.0095 and p = 0.0493, respectively) ( Figure 2E).
Concerning differential expression among tumor subtype and disease stage, SEs and stage I patients displayed higher expression levels of INO80 as compared to NSTs and stages II/III (p < 0.0001 and p = 0.0054, respectively), achieving an AUC = 0.88 for the SE vs. NST discrimination. Also, patients with SRCAP and RUVBL2 alterations showed poorer D/PFS (p = 0.0488 and p = 0.0191) ( Figure 2F).
MAIN CONCLUSIONS: Again, ChRCs represent unexplored territory in TGCTs. Alterations in CHD proteins are particularly frequent. This analysis points out they could be important not only in TGCT subtyping, but also in prognostication (survival impact).

Conclusions
The integrated molecular characterization of TGCTs is only now being uncovered [86]. There is an urgent need for better biomarkers that can supplant the classical serum markers used in clinical practice (which display many drawbacks), both for diagnostic, prognostic, and predictive purposes. DNA/histone-modifying enzymes (along with related modifications) and chromatin remodelers show promise as biomarkers, as they are frequently differentially expressed among the major classes and subtypes of TGCTs, reflecting the so-called developmental model of tumorigenesis and the locked epigenetic status of the corresponding cell of origin. Nevertheless, they are still scarcely explored in TGCT patients. In this work, we have analyzed the expression of several protein coding epigenetic enzymes at the mRNA level e tumor samples. Detection of such transcript-based biomarkers in liquid biopsies might be technically challenging; however, novel techniques for detection of circulating tumor cells and their transcripts are increasingly being employed with success in several tumor models and they should be pursued in TGCTs as well [87][88][89][90]. If effectively detected in liquid biopsies, these epigenetic players may be explored as biomarkers for targeted therapies. By allowing lower toxicity than the routinely employed chemotherapy regimens, these therapies might improve patients' quality of life, which is fundamental for such young individuals with large life expectancy. Also, when used in combination, they may prove useful in overcoming cisplatin resistance, which eventually emerges in TGCT patients. The frequent upregulation of DNMTs in ECs (when compared to SEs) may, for instance, be used as a biomarker of susceptibility to DNMT inhibitors (DNMTi). These pharmacological agents comprehend both nucleoside and non-nucleoside analogs and the rationale for using them stands in the fact that by inhibiting the enzymatic activity of DNMTs they lead to the attenuation of malignant phenotype by inducing differentiation and tumor cell death (for review see [91]). Two of these agents (5-azacytidine and 5-aza-2 -deoxycytidine) are in fact already approved for treatment of patients with hematological malignancies, and they might prove useful in this particularly aggressive TGCT subtype. The frequent upregulation of KDACs in NSTs (the most aggressive and challenging subtypes of TGCTs), especially of those that are dependent on Zn 2+ (HDACs), also indicates a potential benefit from HDAC inhibitors (HDACi), such as hydroxamic acid inhibitors (one of which-suberanilohydroxamic acid-is already approved again for the treatment of hematological cancers) [92,93]. Inhibitors of HATs (HATi), although not being particularly selective, may also aid in treating patients with SE which show frequently upregulation of these enzymes [94]. The use of such agents might allow for dose reduction of chemotherapy that these young patients endure, as SE is a highly chemo-sensitive solid tumor.
All in all, more studies in larger series are needed to explore the practical role and the clinical value of these enzymes in TGCTs.