Pathological Modulation of Genome Maintenance by Cancer/Testes Antigens (CTAs)

Abstract

The Cancer Testis Antigens (CTAs) are a group of germ cell proteins that are absent from normal somatic cells yet aberrantly expressed in many cancer cells. When mis-expressed in cancer cells, many CTAs promote tumorigenic characteristics including genome instability, DNA damage tolerance and therapy resistance. Here we highlight some of the CTAs for which their roles in genome maintenance in cancer cells are well established. We consider three broad CTA categories: (1) Melanoma Antigens (MAGEs) (2) Mitotic CTAs and (3) CTAs with roles in meiotic homologous recombination. Many cancer cells rely on CTAs to tolerate intrinsic and therapy-induced genotoxic stress. Therefore, CTAs represent molecular vulnerabilities of cancer cells and may provide opportunities for therapy. Owing to their high-level expression in tumors and absence from normal somatic cells, CTA-directed therapies could have a high level of specificity and would likely be devoid of side-effect toxicity.

1. Introduction

Cancer Testis Antigens (CTAs) are a group of at least 276 proteins whose expression is normally restricted to the testes and placenta but which are aberrantly over-expressed in many tumors [1, 2]. The first CTA to be identified was discovered based on its antigenicity and high-level expression in melanoma cells [3]. That antigen, later designated Melanoma-associated antigen 1 (MAGE-A1), belongs to a larger family of over 60 human MAGE genes, many of which are also aberrantly expressed in cancer cells. In addition to the MAGEs, hundreds of other germ cell-restricted genes have been shown to be aberrantly expressed in cancers, leading to the establishment of the Cancer-Testis Antigen database (CTdatabase, http://www.cta.Incc.br/). Owing to their high level expression in tumors and absence from normal somatic cells (Fig. 1) CTAs may represent appealing targets for immunotherapy, and CTA-based vaccines have been evaluated for cancer treatment [4, 5].

Figure 1. Expression of CTA genes across various cancers in The Cancer Genome Atlas (TCGA) pan-cancer with unsupervised clustering within tumor samples.

The heatmap shows the relative mRNA expression levels of CTA genes, based on genomic and clinical data sourced from the Xena browser and the GDC portal. The classifiers including histological subtype, TP53 mutation status, HRD score, CA20 score are labeled at the top of the heatmap. To generate the heatmap, normal and tumor samples were first separated and then hierarchically clustered based on Euclidean distance. Within each group, the samples were clustered in unsupervised manner regardless of any of the classifiers. Expression of the mitotic CTAs (PBK2, NUF2, TTK, KIF2C, CEP55, OIP5, KIF20B, ATAD2 and ODF2) correlates with CA20 scores (the Pearson’s correlation coefficient score (ρ) exceeds 0.45 for all these mitotic genes). For TTK and KIF2C the Pearson’s correlation coefficient scores (ρ) were > 0.9.

More than half of the CTAs are encoded on the X chromosome, are reactivated in tumors by promoter demethylation, histone modifications, and non-coding RNAs [6], and are widely expressed among tumors of different histological origins (Fig. 1). While hundreds of CTAs have been identified, their functions in the normal germline tissues and malignant cells are generally poorly understood. Some CTAs have important roles in germ cell-specific processes, for example meiotic homologous recombination (HR). Although CTA expression in malignant cells was once considered to be an epiphenomenon of cancer, it is increasingly recognized that CTAs may acquire ‘neomorphic’ functions unrelated to their germ cell roles when they are mis-expressed in cancer cells. Using a multidimensional siRNA-based screening strategy, Whitehurst and colleagues showed that CTAs can promote many tumorigenic signaling pathways (including HIF, WNT and TGFβ signaling) that sustain cancer cells [7]. The ‘moonlighting’ activities of CTAs in cancer signaling pathways provide an interesting paradigm for how neoplastic cells gain tumorigenic characteristics. CTA-induced tumorigenic signaling may also present exciting opportunities for cancer therapy; the most ideal targets for cancer therapy are those that are necessary to sustain the growth and viability of cancer cells but dispensable for normal cells. In principle, therapies targeting key CTA-dependent tumorigenic processes might have high specificity for neoplastic cells while lacking cytotoxicity towards normal cells.

This review is concerned specifically with the ways in which CTAs promote tumorigenic phenotypes via reprogramming genome maintenance processes. During multi-step tumorigenesis, neoplastic cells experience genomic damage and stress from a variety of sources including oncogenes, metabolites, and mitotic segregation defects [8–14]. Therefore, the CTA-dependent genome maintenance pathways that confer stress tolerance can promote cancer cell survival.

Cancer is fueled by mutations and chromosomal instability [15, 16]. The fidelity of DNA repair and chromosome segregation can critically define the genomic landscape of cancer cells. Thus, CTA-induced imbalance between error-prone and error-free genome maintenance pathways is also a potential driver of genetic change and tumorigenesis.

Finally, the success of radiotherapy and many chemotherapies also relies critically on the induction of irreparable and lethal chromosomal damage [17–20]. Thus, pathological reprogramming of genome maintenance pathways by CTAs provides an interesting and unusual way in which cancer cells may acquire tumorigenic phenotypes such as mutability, tolerance of DNA replication stress, or radio/chemoresistance.

Identifying the CTAs that modulate genome maintenance could provide biomarkers of sensitivity to therapeutic agents (based on DNA repair pathways that they reprogram). Moreover, since CTAs are absent from all normal somatic cells, CTA-dependent radio/chemoresistance pathways are molecular vulnerabilities of cancer cells and may present untapped opportunities for therapies that lack side-effect toxicity to normal cells.

This article is not intended as an exhaustive catalogue of all CTAs with proposed roles in genome maintenance. Instead we highlight the CTAs for which roles in genome maintenance have been clearly established. The examples described here best illustrate the paradigm that CTAs are modifiers of DNA repair that can impact DNA damage tolerance and genome stability. We consider three broad categories of CTAs and their roles in genome maintenance: (i) Melanoma Antigens (MAGEs), (ii) CTAs with mitotic functions, and (iii) CTAs with roles in meiotic recombination.

2. Melanoma-Associated Antigens (MAGEs)

The human Melanoma-associated antigen (MAGE) family consists of over 60 genes. MAGEs are subdivided by expression patterns and chromosome location. Type I MAGEs (including subfamilies -A, -B, and -C) are considered CTAs as they are exclusively expressed in germline tissues, placenta, and cancer cells and located on the X chromosome. In contrast, Type II MAGEs (consisting of subfamilies -D, -E, -F, -G, -H, and -L) are ubiquitously expressed and present in somatic cells and are not restricted to the X chromosome. The MAGE genes appear to have evolved in four phases from a single protozoan ancestral gene, with Type I MAGE genes appearing more recently than Type II MAGEs [21].

Type I and Type II MAGE family members share a conserved MAGE homology domain (MHD) containing approximately 170 amino acids. In most MAGE proteins the MHD is located near the C-terminus [21]. All MHDs contain a conserved di-leucine sequence and share 46% amino acid sequence identity. Within each MAGE family there is even higher conservation of MHD protein sequence. For example the MAGEA MHDs share 70% sequence identity. Each MHD contains two winged-helix (WH) domains designated WH-A and WH-B. Each WH domain possesses a helix-turn-helix juxtaposed with a three-stranded anti-parallel β-sheet wing. Biophysical studies show that the MHDs are flexible and subject to conformational changes, potentially mediating dynamic interactions with other proteins.

Developmentally, the MAGEA family members are important for germ cell fitness since deletion of the Mage-a gene cluster in mice disrupts spermatogonial stem cells and causes sensitivity of the germline to environmental and genotoxic stresses [22]. In various tumors, including bladder, breast, head and neck, ovarian, and non-small cell lung cancer, there are clear associations between MAGE-A antigen expression and poor prognosis [23–26]. In some instances, cancer cell viability and fitness are dependent on MAGE proteins [21]. Overexpression of MAGE proteins often facilitates cancer cell growth and survival whereas depleting MAGE genes has adverse effects on proliferation and sometimes leads to apoptosis [21, 27].

In a landmark study, Potts used proteomics to identify RING Finger E3 ubiquitin ligases as binding partners and major targets of MAGE proteins in tumorigenic signaling pathways [28]. RING finger E3 ligases act as scaffolds and specificity factors that target E2 ubiquitin-conjugating enzymes to their substrates to promote ubiquitylation [29]. The RING finger is a domain that binds to E2 enzymes [30]. Potts discovered that MAGEs bind RING finger E3 ligases with relative specificity and promote substrate ubiquitination by the E2-E3 complexes. Since Potts’ original identification of MAGE Ring Ligase (MRL) complexes, over 50 distinct MRLs have been identified [21, 27]. Different MAGEs tend to bind specific E3 ligase partners [28]. Within each MRL complex, the MHD typically mediates the interaction between the MAGE and its cognate E3 ligase partner. Curiously, there is no single conserved region of RING finger E3 ligases that mediates interactions with MAGEs. For example in the MAGE-C2/TRIM28 MRL complex, the coiled-coil region of TRIM28 mediates MAGE-C2 binding, whereas in the MAGE-G1/NSE1 complex, the WH motifs of NSE1 bind MAGEG1 [21, 27, 28]. Allosteric modification of the MHD domain gives MAGEs specificity when binding with proteins. Therefore, the flexibility and versatility of MHDs appears to confer diverse modes of binding to E3 ligases. For the different MRL complexes, MAGEs have been shown to reprogram ubiquitin signaling via several mechanisms including: (i) promoting E2/E3-mediated ubiquitylation of substrates, (ii) promoting E3 ligase stability, (iii) targeting different substrates for ubiquitylation, and (iv) redirecting the E3 ligase to different subcellular locations. E3 ligase signaling and protein ubiquitylation play prominent roles in the DNA damage response (DDR) [31–33]. Several key E3 ligase mediators of the DDR bind to and are pathologically regulated by MAGEs, in ways that may endow cancer cells with tumorigenic properties. Some of the ways in which MAGE proteins influence genome maintenance via reprogramming of E3 ligases (and other targets) are summarized below.

2.1. Regulation of TRIM28/KAP1 by MAGE proteins

TRIM28 (Tripartite Motif Containing 28), also known as KAP1 (Krüppel-Associated Box (KRAB)-Associated Protein 1) and transcription intermediary factor-β (TIF1β) is a multi-faceted RING finger E3 ligase which promotes both ubiquitylation and SUMOylation and has roles in transcription, p53 signaling, and DNA repair. It is well established that TRIM28/KAP1 binds several MAGEs. Here we summarize known roles of KAP1 in genome maintenance and we describe how those functions may be impacted by MAGEs. A detailed overview of TRIM28 signaling pathways that are unrelated to genome maintenance is beyond the scope of this review, but may be found elsewhere [34].

A major role of KAP-1 is to act as a co-repressor which helps maintain the rigid structure of heterochromatin (transcriptionally-inactive chromatin). This is achieved by recruiting histone-modifiers (e.g. HDAC1/2, SETDB1) and nucleosome-remodelers such as CHD3 [34]. The repair of heterochromatic DNA Double Stranded Breaks (DSB) requires inhibition of KAP1 activity by ATM and CHK2 signaling. In response to DSB, KAP1 is phosphorylated by ATM on S824 [35] and by CHK2 on S473 [36]. These phosphorylations relieve repressive KAP1 activity, and allow heterochromatin decondensation, thereby facilitating DSB repair [35, 37]. Mechanistically this process involves inhibition of KAP1 auto-SUMOylation activity which constitutively recruits the SUMO1-binding nucleosome modeler CHD3 [38]. KAP-1(Ser824) phosphorylation disrupts interactions between CHD3 and SUMO1, thereby dispersing CHD3 from heterochromatic DSBs and promoting DNA repair [38]. Similarly, KAP1 phosphorylation at S473 by CHK2 promotes mobilization of the chromodomain protein HP1-β to create a DNA repair permissive heterochromatin environment [36]. Longley and colleagues showed that MAGE-C2 promotes ATP/KAP1 association, ATM-mediated phosphorylation of KAP1 at S824, and repair of DSB in U2OS cells [39]. Therefore, pathological activation of DSB repair could provide a mechanism for how MAGE-C2 (and perhaps the other TRIM28-binding MAGEs A2, A3, A6, D1 - see Doyle et al. [28]) confer tolerance of intrinsic and therapy-induced DSB in cancer cells.

2.2. Regulation of the p53 signaling axis by MAGEs

p53 has many established roles in tumor suppression, some of which involve genome maintenance [40, 41]. Several studies show that the p53 signaling axis may be modulated by MAGEs, both via direct and indirect mechanisms. Yang and colleagues reported that depleting KAP1-interacting MAGEs (including MAGEA, MAGEB, and MAGEC family members) from melanoma cells leads to increases in active acetylated p53, expression of p53-responsive genes and apoptosis [42]. In a subsequent study, Doyle et al showed that MAGE-C2-TRIM28 directly ubiquitylates and targets p53 for proteasomal degradation [28]. Taken together, these studies suggest that MAGE-TRIM28 signaling may confer tumorigenicity by repressing p53.

Another E3 ligase-mediated mechanism whereby MAGEs modulate p53 involves the MDM2 family members. The MDM2 and MDMX (MDM4) RING finger proteins are oncogenes that have major roles in negatively regulating p53 [43]. Both MDM2 and MDM4 genes are p53-inducible and their induction constitutes a p53 autoregulatory loop. MDM2 has E3 ligase activity and promotes ubiquitylation and proteasomal degradation of p53 [44]. MDMX lacks E3 ligase activity, but facilitates MDM2-mediated ubiquitylation of p53 [45]. Meek and colleagues showed that MAGE-A2 binds the RING finger and the p53-binding pocket of MDM2 to inhibit MDM2-mediated ubiquitin ligase activity and also inhibits MDM4 ubiquitylation promotes MDM4 stability [46]. Therefore, MAGE-A2 may have the potential to influence the DNA repair p53 functions via modulation of its critical upstream regulators MDM2 and MDM4 [46].

MAGEs can also influence p53 directly. Marcar et al. showed that MAGE-A1 interacts with 3 specific motifs in the DNA-binding domain of p53 to inhibit the interaction of p53 with chromatin [47]. Importantly, ablation of MAGE-A1 leads to de-repression of p53, evidenced by transcriptional induction of p53 target genes and cell cycle arrest [47]. Monte et al. showed that the association of MAGEA2 with p53 promotes recruitment of histone deacetylase 3 (HDAC3) to the complex, leading to decreased acetylation of chromatin in the vicinity of p53 response elements and reduced transcription of p53 target genes [48]. Given the prominent role of p53 in diverse genome maintenance processes [40, 41] it may be expected MAGE-induced dysregulation of p53 signaling might have significant consequences for DNA repair, genome stability, and viability following genotoxic insult.

2.3. Reprogramming of Trans-Lesion Synthesis (TLS) by MAGE-A4

TLS is an important DNA damage tolerance mechanism that relies on specialized ‘Y-family’ DNA polymerases to replicate damaged DNA templates [49]. When replicating undamaged DNA, the Y-family TLS polymerases can be highly error-prone when compared with replicative DNA polymerases [49]. However, some Y-family TLS polymerases can replicate DNA templates harboring ‘cognate lesions’ with relative accuracy and efficiency. For example solar ultraviolet (UV) radiation-induced cyclobutene pyrimidine dimers (CPDs) are cognate lesions for DNA Polymerase eta (Polη), which correctly inserts two adenines opposite UV-crosslinked thymine dimers [50, 51]. In the absence of Polη, alternative TLS polymerases can perform more error-prone bypass of CPD, leading to mutations [52, 53]. Congenital defects in the POLH gene result in the sunlight-sensitivity and skin cancer-propensity syndrome xeroderma pigmentosum variant (XPV) [54].

Collectively, the TLS polymerases enable replicative bypass of DNA templates containing diverse forms of bulky DNA damage from chemical carcinogens and chemotherapeutic agents. The fidelity of lesion bypass and risk of mutation can depend on whether TLS polymerases act on cognate or non-cognate lesions. Therefore the selection and availability of TLS polymerases can be determinants of genome stability. In addition to their roles in replicating DNA templates containing bulky adducts, TLS polymerases play important roles in filling of post-replicative single-stranded DNA (ssDNA), including ssDNA regions of the genome that are induced by oncogenes [55] or that arise due to HR-deficiency [56, 57]. Owing to their roles in promoting mutagenesis and tolerance of chemotherapy- and oncogene-induced DNA replication stress, TLS polymerases may endow neoplastic cells with tumorigenic characteristics [58].

Because TLS polymerases are potential drivers of mutagenesis, they must be used sparingly. DNA damage-inducible activation of the TLS polymerases is promoted by a RING finger E3 ubiquitin ligase RAD18 and its partner E2 ubiquitin-conjugating enzyme RAD6. Mechanistically, RAD18-RAD6 is redistributed to sites of DNA replication stalling via associations with RPA-coated ssDNA [59]. Once recruited to the vicinity of stalled DNA replication forks, RAD18-RAD6 promotes PCNA mono-ubiquitylation [60, 61]. TLS-polymerases possess PCNA-Interacting Peptide (PIP) motifs and ubiquitin-binding domains and therefore associate preferentially with PCNA in its mono-ubiquitylated state [62]. Therefore, RAD18-mediated PCNA mono-ubiquitylation may be a critical determinant of TLS.

The molecular details of the RAD18-RAD6 interactions have remained a mystery. RAD18 has been shown to form a dimer through its N-terminal RING domain [63]. The RAD18-RAD6 interaction may involve both the RING domain and a C-terminal peptide, dubbed the RAD6-binding domain (R6BD) [63, 64]. Biochemical evidence has supported the model of a heterotrimer, in which a RAD18 dimer recruits a single copy of RAD6 monomer [63]. This observation hints at a potential allosteric regulation of RAD6 binding such that binding of one unit of RAD6 masks the binding site for the second RAD6 molecule.

Gao et al. defined the RAD18 protein interaction network in cancer cells and identified MAGE-A4 as a major component of the RAD18 complex [65]. Interestingly MAGE-A4 also interacts with the RAD6-binding domain of RAD18 in vitro, and the molecular details of such an interaction have been captured by X-ray crystallography [66]. Given that RAD18 functions as a dimer that interacts with a single copy of RAD6, a possible explanation is that MAGE-A4 may interact with the unoccupied R6BD from the RAD18 dimer. As the R6BD domain is flanked by long loops on both ends in Rad18, MAGE-A4 binding to the R6BD of RAD18 may mask disordered loops and thus promote the stability of the trimeric RAD18:RAD18:RAD6 complex. Such a hypothesis is consistent with the observation that in contrast with many other MAGEs (which stimulate ubiquitin conjugation activity of their E3-E2 partners), MAGE-A4 does not promote RAD18-RAD6-dependent PCNA ubiquitylation directly [65]. Instead, MAGE-A4 protects RAD18 from proteolytic degradation, thereby leading to unusually high RAD18 protein levels in many cancer cells [65]. It is well documented that increases in RAD18 levels (e.g. following ectopic RAD18 expression) can stimulate both basal and genotoxin-induced PCNA mono-ubiquitylation [67, 68]. Therefore, it appears that MAGE-A4 promotes PCNA ubiquitylation and distal TLS events indirectly, by maintaining RAD18 expression at high levels.

Recently, Bhogaraju and colleagues developed a proximity-based tool to identify substrates of E3 ubiquitin ligases, including RAD18 [69]. Consistent with previous results, PCNA was identified as a RAD18 substrate that is ubiquitinated in response to UV exposure. Intriguingly, MAGE-A4 was also identified as a RAD18 substrate, with its ubiquitination enriched in cells not exposed to UV [69]. The significance of RAD18-mediated MAGE-A4 ubiquitylation is not yet known. Nevertheless, this surprising finding suggests that RAD18 may undergo a conformational switch following DNA damage that leads to altered substrate specificity.

Many questions remain regarding the relationship between the different components of the RAD18-RAD6-MAGE-A4 complex, how it assembles, and how it affects other RAD18 target pathways. For example, since RAD6 and MAGE-A4 bind the same region of RAD18, do the relative levels of RAD6 and MAGE-A4 in cancer cells affect the composition of RAD18 complexes? How does the composition of the RAD18 complex affect its participation in different DNA repair effector pathways such as TLS, HR [70], and ICL repair [71]? Will a sufficiently high level of MAGE-A4 displace all RAD18-bound RAD6, perhaps attenuating TLS functions (which require RAD6-mediated PCNA ubiquitylation), yet favor other RAD6-independent RAD18 activities such as HR and ICL repair? Clearly further work is needed to determine how MAGE-A4 affects the RAD18 protein interaction network and the various RAD18 effector pathways.

2.4. Chemical tractability of MAGEs as therapeutic targets

Many chemotherapeutic agents preferentially target replicating cancer cells [72]. For successful cancer treatment, chemotherapy-induced DNA damage must induce persistent and irreversible DNA replication stalling that leads to cell death [17, 73]. However, TLS can allow cells to sustain chemotherapy-resistant DNA synthesis. TLS polymerases can perform replicative bypass of chemotherapy-induced DNA lesions including cisplatin adducts [74]. Additionally TLS polymerases can repair ssDNA tracts that are induced indirectly by chemotherapy [56, 57, 75]. Potentially, ssDNA gaps can arise between a stalled DNA polymerase and a downstream (3’) newly-reprimed leading strand [76, 77]. Excessive ssDNA can also be generated in response to chemotherapy drugs that do not cause bulky fork-stalling lesions. For example pharmacological inhibition of WEE1 leads to aberrant initiation of DNA synthesis and ssDNA-containing replication intermediates [78–80] (in addition to causing other defects such as replication fork deprotection [81]). Genomes of HR-deficient cells such as BRCA1-mutant breast and ovarian cancer cells also accumulate excessive ssDNA [56, 57].

Because TLS helps sustain viability of chemotherapy-treated or HR-deficient cancer cells, pathological MAGE-A4-RAD18-dependent TLS represents a very attractive therapeutic target. Importantly, MAGE-A4 is absent from all normal somatic cells. Therefore, MAGE-A4 inhibition provides a strategy for inhibiting TLS specifically in cancer cells.

To test the therapeutic tractability of the MAGE-A4-RAD18 complex, Fleming et al. screened an mRNA display library and identified cyclic peptides that bound to MAGE-A4 with high specificity and affinity [82]. Importantly, some of the cyclic peptides potently disrupted the MAGE-A4-RAD18 interaction [82]. A crystal structure of MAGE-A4 complexed with an inhibitory peptide revealed a hydrophobic pocket surrounded by α4, α5, and a8 helices. Interestingly, some of the discovered peptides (e.g. cMCF) shows sequence similarly with the R6BD of RAD18 [83] (Fig. 2), leading to structural elucidation of the RAD18 R6BD-MAGEA4 complex [66]. Therefore, peptide inhibitors discovered through mRNA display disrupt the RAD18-MAGE-A4 interaction by competing with RAD18 for binding to the same hydrophobic interface of MAGE-A4.

Figure 2. Schematic illustration of the MAGE-A4-mediated stabilization of RAD18 and the functional consequences of MAGE-A4 inhibition.

(A) MAGE-A4 binding to RAD18 stabilizes the otherwise flexible regions surrounding the R6BD of RAD18 and protects it from proteolytic degradation. (B) Disruption of the MAGE-A4-RAD18 interaction by MAGE-A4-binding peptidomimetics, such as cMCF01, reduces RAD18 stability and compromises RAD18-mediated DNA damage responses. Sequence alignment was generated using Clustal Omega using the default color scheme. The tyrosinase-oxidized Tyr in cMCF01 is indicated by an asterisk and colored in green, and its crosslinking to Cys is indicated by the bracket. The RAD6/RAD18 R6BD complex (PDB: 2YBF) and the MAGE-A4/RAD18 R6BD complex (9BD3) are shown in the cartoon model. The RAD18 R6BDs are colored in rainbow, with the N-termini in blue and the C-termini in red.

In summary, many cancer cells rely critically on MAGE-A4 to sustain RAD18 levels and damage-tolerant DNA replication. MAGEA4-RAD18 is a chemically tractable target whose inhibition represents an appealing strategy for ameliorating TLS-dependent chemoresistance. RAD18 also has other roles in genome maintenance, for example as a scaffold during ICL repair [71], and as a chaperone for RAD51 paralogs during homologous recombination (HR) [70]. Because MAGE-A4 critically maintains high RAD18 expression levels in cancer cells, it is likely that RAD18-mediated ICL repair and HR will also be MAGE-A4-dependent. The potential involvement of MAGE-A4 in ICL repair and HR provides additional rationale for therapeutic targeting of RAD18-MAGEA4 and sensitizing cancer cells to diverse classes of DNA-damaging drugs.

In addition to TRIM28, MDM2, and RAD18, there are numerous RING finger-containing E3 ubiquitin and SUMO ligases that play prominent roles in DDR signaling and genome maintenance in cancer cells [33]. Therefore it is likely that MAGEs reprogram additional E3 ligases to promote genome maintenance in cancer cells. Since MAGEs are pharmacologically-tractable [82], Identification of other MAGE-E3 ligases complexes with pathological DNA repair roles might reveal additional targets for chemo/radiosensitizer activity.

3. Mitotic CTAs

3.1. CTAs and paclitaxel tolerance

Using a synthetic lethal RNAi screening approach, Whitehurst and colleagues identified several CTAs that are necessary for lung cancer cells to tolerate the anti-mitotic agent paclitaxel [84]. Paclitaxel is a microtubule-stabilizing drug that is used to treat a variety of cancers including ovarian, breast and lung. The mechanisms by which paclitaxel kills cancer cells are dose dependent. In cell culture studies, high doses of paclitaxel cause mitotic arrest due to activation of the spindle assembly checkpoint (SAC). However, the clinically-relevant low doses of paclitaxel attainable in patients kill cancer cells by causing spindle multipolarity [85, 86]. In cells harboring more than two spindle poles chromosome segregation occurs unevenly between multiple spindle poles leading to lethal cytokinesis failure [85, 86].

One potential way in which CTAs might promote paclitaxel-resistance is by reprogramming components of the mitotic spindle (including microtubules, microtubule-associated factors, and motor proteins) to correct or avert spindle multipolarity. Whitehurst and colleagues showed that association between acrosin binding protein (ACRBP, a CTA) and the mitotic factor NuMA is necessary for paclitaxel resistance [87]. NuMA is a ~200-nm-long coiled-coil protein with essential and separable roles in mitosis and nuclear integrity [88, 89]. During mitosis NuMA facilitates attachment of centrosomes to spindle fibers and maintains focused kinetochore fibers at spindle poles [90]. ACRBP depletion results in mitotic errors and reduces proliferative fitness. The effects of ACRBP-depletion are rescued by NuMA co-depletion suggesting that ACRBP restricts a NuMA-dependent abrogation of a mitotic spindle assembly [87]. Other CTAs, FMR1NB, NXF2, MAGEA5, FSIP1, and STARD6 are also necessary for accurate chromosome segregation in tumor cells [91]. However, their mechanisms of action are most likely related to microtubule function since depleting these CTAs led to increased microtubule cytaster formation and reduced microtubule stability.

3.2. CTAs and supernumerary centrosomes

Similar to the effects of paclitaxel, excessive numbers of centrosomes in cancer cells may induce spindle multipolarity. In normal cells, the formation of bipolar spindles and proper chromosome segregation during mitosis critically depend on the presence of two centrosomes which serve as microtubule organizing centers (MTOCs) [92, 93]. The presence of supernumerary centrosomes is a hallmark of many cancers [12, 94] and represents a potential mechanism for tumor initiation [13, 95]. Studies with mammalian models and fruit flies show that centrosome amplification alone can drive aneuploidy and tumorigenesis [96, 97]. Ogden et al developed a ‘Centrosome Amplification CA20’ gene expression signature which comprises centrosome structural genes together with genes that promote centrosome amplification [98]. In TCGA patient tumors, the CA20 signature is associated with genomic instability and poor prognosis [99].

In cancer cells harboring supernumerary centrosomes, the ensuing spindle multipolarity can also lead to chromosome mis-segregation and cell death [13, 95, 100]. Therefore, cancer cells with excessive numbers of centrosomes must develop adaptive mechanisms to avert lethality from multipolar mitoses [101, 102]. Four major ways in which cancer cells adapt to the presence of supernumerary centrosomes are centrosome clustering, centrosome inactivation, centrosome degradation, and centrosome loss by extrusion [94]. The clustering of excess centrosomes to form pseudo-bipolar spindles is perhaps the best characterized mechanism by which cancer cells tolerate centrosome amplification. Notably, centrosome clustering mechanisms also allow cancer cells to tolerate the spindle multipolarity caused by paclitaxel [86, 103, 104].

A pan-cancer analysis of CTA expression reveals a cluster of CTAs whose expression correlates very closely with a centrosome amplification (CA20) gene signature (Fig. 1). Most of the CTAs associated with CA20 signature are known to have important mitotic functions (Table I). Notably, many of these mitotic CTAs allow cells to survive paclitaxel-induced spindle multipolarity (Table I). The strong correlation between the mitotic CTAs highlighted in Fig. 1 and CA20 signature suggests that mitotic CTAs allow cancer cells to tolerate supernumerary centrosomes. Indeed, KIF2C/MCAK (a mitotic centromere-associated kinesin, involved in error correction of kinetochore-MT attachments - see Table I) is important for centrosomal clustering in flies [100].

Table I.

CA20 signature-associated Mitotic CTAs

Mitotic CTA	Known roles in mitosis	Roles in chemoresistance
DCAF12 (DDB and CUL4-Associated Factor 12)	Centrosomal protein and component of the CRL4-DCAF12 E3 ubiquitin ligase A [107]; degrades MAGEA3/A6 during starvation-induced autophagy [108].	unknown
PBK (PDZ binding kinase)	Mitotic protein kinase phosphorylated by Cyclin B-CDK1 [109]; Promotes cytokinesis [110].	Confers paclitaxel resistance [111]
NUF2	Component of NDC80 kinetochore complex, functions at kinetochores for stable microtubule attachment and retention of MAD1/2 checkpoint proteins, interacts with CENP-E [112–114]	Unknown
TTK (threonine tyrosine kinase, aka MPS1)	Kinetochore-associated kinase required for the SAC [115]; regulates centrosome duplication [116].	Confers resistance to paclitaxel [117, 118] and docetaxel [119–121]
KIF2C (kinesin family member 2C, aka mitotic centromere-associated kinesin MCAK)	Mitotic centromere-associated kinesin [122]; required for anaphase chromosome separation [123]; cooperates with another microtubule-depolymerizing kinesin (KIF2B) to stimulate kinetochore-microtubule (kMT) dynamics and correct mal-orientations [124, 125]; promotes DSB repair [105] and DNA replication stress tolerance [106].	Confers resistance to Paclitaxel [126, 127] and the WEE1 inhibitor AZD-1775 [106]
CEP55 (centrosomal protein 55)	Present at centrosomes throughout mitosis [128] and required for completion of cytokinesis [129]; Cep55 overexpression in mice causes spontaneous tumorigenesis and accelerates tumorigenesis caused by Tp53-deficiency, compromises the Chk1-mediated S-phase checkpoint causing increased fork velocity and DNA damage, and stabilizes microtubules [130]; prevents death of aneuploid cells [131].	Unknown
OIP5 (Opa Interacting Protein 5, aka MIS18B)	Identified as Mis18 beta, one of 3 proteins essential for centromere/kinetochore structure and function [132]; essential for recruitment of the centromeric histone 3 variant CENP-A by Holliday junction recognition protein (HJURP) [133]; stable binding of Mis18 to centromeres in telophase licenses them for CENP-A deposition [134].	Unknown
KIF20B (Kinesin family member 20B)	Originally identified as a mitotic phosphoprotein MPP1 [135]; subsequently shown to be a plus-end-directed kinesin required for cytokinesis [136]; localizes to microtubules of the central spindle and midbody throughout cytokinesis [137].	Confers paclitaxel-resistance [138].
ODF2 (Outer Dense Fiber Of Sperm Tails 2)	Originally identified as a sperm outer dense fiber component [139]; the ODF2 gene was cloned independently based on its expression in bile canaliculi of chicken liver and the ODF2 protein was shown to be expressed in other cells including HeLa where it formed an insoluble scaffold in the centrosome matrix [140]; proposed to be necessary for mitotic functions of PLK1 at centrosomes in HeLa cells [141, 142]; important for preventing premature centrosome disjunction in untransformed NIH3T3 cells [143].	Unknown

Interestingly, KIF2C also has roles in DNA repair that may not be readily explained by its mitotic functions. Zhu et al. showed that KIF2C associates with DSB sites in a manner that depends on PARP and ATM, and promotes DNA repair via the NHEJ and HR pathways [105]. Mechanistically it is proposed that KIF2C promotes the mobility of DSB to facilitate DNA repair [105]. In a separate study, KIF2C was shown to protect endometrial cancer cells from WEE1 inhibitor-induced DNA replication stress, averting fork collapse [106]. It is unknown whether the role of KIF2C in sustaining WEE1 inhibitor-tolerant DNA replication is related to its roles in regulating MT dynamics or DSB repair. However, the findings that KIF2C has roles in DNA DSB repair and DNA replication suggest the possibility that other mitotic CTAs may also have non-canonical roles in genome maintenance.

While the CTAs listed in Table I have well-established functions in mitosis, most of those mitotic roles were mechanistically defined in neoplastic cells. Therefore it is unclear whether ‘mitotic CTAs’ perform identical functions and fulfill similar roles in both normal and cancer cells. In some cases, these mitotic CTAs are induced by oncogenes. For example, Zhang et al. showed that ectopic expression of oncogenes (including CCNE1 and KRAS^G12V) in primary untransformed human endometrial epithelial cells leads to robust induction of CEP55 and NUF2, recapitulating expression levels of these CTAs that are observed in endometrial cancer cell lines [106]. Moreover, NUF2-depletion using siRNA was shown to reduce survival of KRAS^G12V-expressing but not parental epithelial cells [106]. These results suggest that induction of NUF2 (and perhaps other mitotic CTAs) confers tolerance of oncogene-induced stress.

4. CTAs with roles in meiotic recombination

Some CTAs have important roles in HR during meiotic prophase. During meiosis, the paternal and maternal homologous chromosomes become paired, generating points of connection (chiasmata) where genetic information is exchanged via HR (a process termed ‘crossover’). HR is initiated by programmed DSB which are created by a meiosis-specific endonuclease SPO11 [144–146].

The pairing of homologous chromosomes is a critical step in facilitating HR and crossovers. Homologous chromosome pairing is physically mediated by a proteinaceous structure termed the Synaptonemal Complex (SC). The SC bridges each homologous chromosome pair and consists of two independent Axial Elements (AEs) which are then connected by central element (CE) and transverse elements. The assembly and disassembly of the SC is temporally regulated and integrated with DSB formation and HR. Temporally, AEs form during leptotene while SCs assembly begins in zygotene and is complete during pachytene. Finally, SC disassembly occurs during diplotene [144–146].

Several meiotic CTAs are major components of the SC: SYCP2 and SYCP3 are present in the AEs, HORMAD1 and HORMAD2 are associated with unsynapsed chromosome axes, SYCE1/2/3 are components of the central element, SYCP1 is present in the transverse filaments, and SPO11 initiates HR. These CTAs are all critically needed for the induction and repair of DSB repair that mediates exchange of genetic information between maternal and paternal chromosomes. In mice, deletion of any of these meiotic CTA genes leads to infertility (both in males and females), but does not result in other developmental defects, demonstrating that these CTAs are normally required only for meiosis [147–155]. A detailed discussion of meiotic recombination and the roles played by CTAs in this process is beyond the scope of this review, but interested readers are referred to excellent articles that describe meiotic events in detail [144–146, 156]. Here we highlight proposed roles of two meiotic CTAs - HORMAD1 and SYCP2 - whose impact on genome maintenance in cancer cells has been described in considerable detail.

4.1. HORMAD1

During meiosis HORMAD1 and HORMAD2 are recruited to chromosomes to ensure generation of processed DSB for homology search. These CTAs also promote SC formation independently of their roles in the homology search. The HORMAD1 and HORMAD2 proteins have a conserved 200 AA HORMA domain that was first identified based upon its sequence similarity with yeast proteins Hop1, Rev7 and Mad2 [157]. The HORMA domain comprises a ~150 AA N-terminal ‘core’ which is followed by a C-terminal ‘safety belt’. The safety belt region of HORMA proteins can interact with the core in two distinct conformations termed ‘open’ or ‘closed’ states. In the closed state, a short interacting peptide from a binding partner binds the HORMA core and is locked in place by the safety belt [158]. HORMAD1 and HORMAD2 both contain a C-terminal interacting peptide that allows the safety belt motifs of the meiotic HORMAD molecules to interact in trans, thereby generating multimeric assemblies of HORMAD1 and 2 which promote DSB formation and crossovers.

In addition to its role in meiosis, several groups have independently identified roles for HORMAD1 as a modulator of DNA DSB repair in cancer cells. Watkins et al. analyzed a cohort of TNBC patients and found an association between HORMAD1 expression and allelic-imbalanced copy-number aberrations (AiCNA). AiCNA is a genomic scar potentially caused by error-prone modes of DSB repair that often compensate for HR-deficiency [159], providing an indication that HORMAD1 expression might be related to reduced HR activity. Those workers also found that HORMAD1 expression is associated with carboplatinum-sensitivity (a feature potentially caused by defects in HR), and that ectopically-overexpressed HORMAD1 in TNBC cell lines leads to chromosomal abnormalities consistent with HR-defects [159]. Using a DR-GFP reporter assay for HR, Watkins et al. showed that HORMAD1 expression leads to reduced repair of ISce1-induced targeted breaks, while slightly increasing repair of DSB as assayed using an NHEJ reporter. Further consistent with HORMAD1 being a negative regulator of HR, HORMAD1-over-expressing cells were shown to recapitulate a major hallmark of HR-deficiency, namely PARP inhibitor-sensitivity [159]. Taken together the work of Watkins et al. strongly suggests that HORMAD1 inhibits HR. Using an siRNA screening approach, the same group subsequently demonstrated synthetic lethality between HORMAD1 expression and ablation of TLS polymerases [160]. A major phenotype of HR-deficient cells is accumulation of ssDNA gaps, which creates a dependency on gap-filling repair processes such as TLS [56, 57, 75, 161]. The dependency of HORMAD1-expressing cells on TLS is also fully consistent with inhibition of HR by HORMAD1.

In contrast, other workers have reported that HORMAD1 expression helps sustain HR in cancer cells [162–164]. Gao et al. used a candidate approach to identify meiotic CTAs with roles in genome maintenance [162]. Those workers used immunofluorescence microscopy to detect CTAs that redistributed to Ionizing radiation-induced foci (IRIF) containing the DSB marker γH2AX. Of a small panel of CTAs that were screened (HORMAD1, HORMAD2, SPO11, SYCE1 and SYCP1), only HORMAD1 co-localized with IRIF, suggesting a role for HORMAD1 as a DDR responder in cancer cells. Interestingly HORMAD1 recruitment to IRIF was dependent on ATM signaling. Mutational analyses revealed that the HORMA motif is required for nuclear localization and the C-terminal seat belt closure motif is necessary for localization to IRIF [162]. Gao et al. showed that HORMAD1-depleted cells are radiosensitive, have reduced HR activity as assayed using the DR-GFP reporter, and phenocopy a hallmark characteristic of HR-deficient cells, namely PARPi-sensitivity [162]. Using molecular markers that define the different stages of HR, these workers showed that HORMAD1 is required for CtIP-mediated DSB resection and production of the invading ssDNA strand. Gao et al. found that the role of HORMAD1 in DSB in cancer cells is not dependent on its meiotic binding partners HORMAD2 and CCDC36, indicating that HORMAD1-mediated DSB signaling and HR are ‘neomorphic’ or ‘moonlighting’ activities.

In an independent study, Nichols et al used a chemigenomics screening approach and identified a correlation between expression of HORMAD1 and resistance to the mitochondrial complex I inhibitor piericidin [163]. Similar to the results of Gao et al. [162], Nichols and colleagues demonstrated that HORMAD1 is recruited to DSB sites and promotes HR, thereby conferring tolerance of oxidative stress-induced genotoxicity [163]. Further consistent with a stimulatory role for HORMAD1 in HR, Wang et al. showed that HORMAD1 overexpression leads to PARPi-resistance and increased tumorigenicity [164].

Thus multiple studies have shown robust effects of HORMAD1 on HR pathway activity and HR-associated phenotypes. However, it is unclear how the reported effects of HORMAD1 on HR can be both inhibitory [159, 160] and stimulatory [162–164]. The most obvious explanation for the discrepant effects of HORMAD1 status on HR in different studies is that biological context is important. The various studies described above were performed using different cell lines. It is possible that differences in cell-intrinsic factors such as cell line-specific post-translational modifications or putative HORMAD1-binding partners determine the extent to which HORMAD1 promotes or inhibits HR. For example, the AAA+ ATPase TRIP13 catalyzes conformation switching of HORMA domain-containing proteins and is known to deplete HORMAD1/2 from meiotic chromosomes [165]. TRIP13 is overexpressed in many cancer cells [166, 167] and could dictate how HORMAD1 affects HR activity. Additionally, the availability of redundant DDR pathways might determine the extent to which different cells rely on HORMAD1 for HR.

To fully address the discrepancies in how HORMAD1 impacts HR, it will be necessary to elucidate mechanistically how HORMAD1 interfaces with the DNA repair pathway. For example although Gao et al. showed that the C-terminal seat belt closure motif is necessary for localization of HORMAD1 to IRIF, it is not yet clear what binding partners use the seatbelt to recruit HORMAD1 to the vicinity of DSB, or what DNA repair factors are activated/inhibited by HORMAD1. In meiotic cells, HORMADs bind short closure motifs within their own c-termini and self-assemble into oligomers [168]. One attractive possibility is that in cancer cells, the C-terminal interacting peptide in HORMAD1 associates with other HORMA-containing DNA repair factors to modulate DNA repair. The HORMA-containing protein MAD2L2/ REV7 has an important role as an anti-resection factor that determines the choice between HR and NHEJ [169, 170]. Therefore, putative HORMAD1-REV7 interactions could stimulate or inhibit DSB resection to influence HR.

Using IP MS, Herrera et al. identified several DNA replication factors as HORMAD1-associated proteins including PCNA, RFC3, RFC4, ATAD5, mini-chromosome maintenance family members, PARP1 and WRNIP1 [171]. Based on a known role of WRNIP in stabilizing RAD51-ssDNA interactions [172], those workers suggested a role for HORMAD1 and WRNIP in promoting DNA replication fork protection. The HU treatment conditions used by Herrera et al. to define a role for HORMAD1 in fork protection typically induce replication fork regression [173]. During DNA replication stress-induced fork regression, RAD51 can facilitate the strand exchange that initiates the fork reversal, or can act on the single-ended DSB at the regressed fork to facilitate replication restart [173, 174]. Therefore, the results of Herrera et al. may suggest that HORMAD1 cooperates with WRNIP to promote RAD51 loading that drives fork regression and restart. It is not clear that the putative WRNIP-mediated RAD51 loading mechanism that protects single-ended DSB generated by fork regression explains the HORMAD1-dependent HR and DSB repair described by other studies [162, 163]. However, another HORMAD1 partner identified by Herrera et al., namely PARP1, has diverse roles in DNA repair [175] and thus represents a potential effector of HORMAD1 in DSB repair. Ronson et al. demonstrated that PARP1 and PARP2 regulate the localization of RAD51 to damaged DNA replication forks [176], potentially providing a connection between HORMAD1 and HR.

As an alternative mechanism to modifying DNA repair factors directly, HORMAD1 could potentially promote DSB repair by interacting with the chromatin environment in the vicinity of damaged DNA. By analogy, another CTA (SYCP2) regulates HR by binding to DNA/RNA intermediates in the HR process [177] (as described below). Differences in the chromatin environment in different biological contexts might also help explain the contradictory effects of HORMAD1 on HR reported by different groups [159, 160, 162, 163]. It will be important to identify the factors that determine how HORMAD1 influences HR activity in cancer cells. Identifying mechanisms by which HORMAD1 sustains HR might reveal the HORMAD1 signaling axis as a tractable therapeutic target for improved chemo/radiosensitivity.

4.2. SYCP2

Similar to the proposed role of HORMAD1 as a pathological activator of DSB repair, SYCP2 promotes a sub-pathway of HR in cancer cells. Wang et al. found that expression of SYCP2 correlates strongly with resistance to chemotherapy drugs that are either genotoxic or that target the DDR including cisplatin (a DNA crosslinking agent), camptothecin 11 (CPT11, irinotecan, a Topoisomerase I inhibitor) and PARP inhibitors [177]. Those workers showed that SYCP2 depletion in various cancer cell lines leads to reduced cancer cell fitness under basal conditions, and increased sensitivity to DNA damaging agents. Using an elegant reporter assay that measures HR in transcriptionally ‘on’ and ‘off’ states, Wang et al showed that SYCP2 promotes HR in a transcription-dependent manner. RNA transcripts can form R-loops at transcribed regions of the genome, thereby stimulating HR [178, 179]. Mechanistically, SYCP2 binds to DNA-RNA hybrids promoting the formation of R-loops. A middle ‘M1’ protein fragment containing AAs 492–1035 of the large 1530 AA SYCP2 protein is both necessary and sufficient to localize SYCP2 to transcriptionally active DNA and promote HR efficiency. Recombinant SYCP2 M1 fragment binds synthetic RNA-DNA hybrids with higher affinity than dsDNA or ssRNA. A lysine (K) / arginine (R)-rich region of residues 822–826 are important for SYCP2-induced R-loop formation and transcription-coupled HR. The SYCP2-dependent mechanism of HR is intact in BRCA1/2-deficient cells, indicating that SYCP2-mediated TC-HR is independent of the canonical HR pathway.

5. Conclusions and Perspectives

As discussed above, there are now several well-established ways by which CTAs can pathologically facilitate DNA repair and bipolar chromosome segregation in cancer cells to confer viability (Figure 3). However a byproduct of pathologically-regulated DNA repair or chromosome segregation could be genome instability. For example, MAGE-A4 activates TLS which can be a highly error-prone mode of DNA replication [180] that leads to DNA damage tolerance at the cost of mutability. Similarly, HORMAD1 and SYCP2 activate HR which can also be error-prone and a contributor to both genome instability and cancer [181–184]. For example, 25% of the genome contains repetitive sequences and homologous genes that arose via duplication events.

Figure 3. Summary of mechanisms by which MAGE-A4, mitotic CTAs, and meiotic CTAs influence genome maintenance and chromosome stability in cancer cells.

In response to DNA replication stress MAGE-A4 promotes TLS and averts lethal replication fork collapse (left panel). Because TLS can be error-prone, MAGE-A4-dependent TLS may result in mutations. In cancer cells harboring supernumerary centrosomes, mitotic CTAs may promote centrosome clustering and the formation of pseudo-bipolar spindles, thereby averting lethal multipolar mitoses, albeit at the risk of aneuploidy (middle panel). In response to DSB, some meiotic CTAs promoting HR and avert lethal DSB, with the risk of genetic alterations (e.g. deletions, duplications and translocations) that can arise due to excessive recombination events between homologous sequences (right panel).

Recombination events between such homologous sequences could lead to genetic alterations (e.g. deletions, duplications and translocations) that are often found in cancer [182, 183]. HR genes (including RAD51) are frequently overexpressed in cancer and associated with increased genomic instability [185]. Moreover, RAD51 overexpression can stimulate both HR and genetic change [184, 186, 187], while ablating RAD51 expression reduces chromosomal instability [185]. A naturally-occurring BRCA1 mutant can cause excessive resection of DNA breaks and hyper-recombination [188], further demonstrating that HR is not always error-free. It is quite likely that the types of genome instability identified in HORMAD1-overexpressing TNBC [159] arose from pathologically-activated HR.

By analogy with pathological CTA-induced DNA repair mechanisms, mitotic CTAs may also sustain cancer cell viability while compromising genetic stability. Pathological centrosome clustering can avert lethal multipolar mitoses, yet generates pseudo-bipolar spindles that are not properly positioned on the normal bipolar axis and can cause chromosome segregation defects and aneuploidy [12]. Thus CTA-dependent genome maintenance mechanisms (such as TLS, HR, chromosome segregation) might promote multiple enabling features of cancer cells: (i) stress tolerance that sustains viability, (ii) genetic changes that drive neoplasia, and (iii) resistance to therapeutic agents (including genotoxins and mitotic inhibitors). More work is necessary to define the mechanisms by which MAGEs, and mitotic/meiotic CTAs impact genome maintenance, genome stability and responsiveness to therapeutic agents in cancer cells. Elucidating the mechanisms by which CTAs modulate genome maintenance will reveal how cancers develop and will provide new opportunities to predict and treat cancer.

Methods

TCGA pan-cancer data visualization

R (version 4.1.0) was used for visualization and data analyses. Overall, P < 0.05 was considered as statistical significance. All of the RNA-seq (FPKM-UQ), somatic variant calling, copy number variation, and clinical information data were downloaded through the GDC portal(https://portal.gdc.cancer.gov/) and Xena Browser(https://xenabrowser.net/). Heatmaps were generated using function Heatmap() from the R package ComplexHeatmap (Version: 2.8.0) [189]. Samples were separated into normal group and tumor group, using the option column_split. The standardized log2 transformed FPKM-UQ was used as the expression value for genes in the heatmap. The data in the heatmap were clustered using hierarchical clustering with Euclidean distance. Homologous recombination deficiency (HRD) was defined as the unweighted sum of three signatures: Number of telomeric Allelic Imbalances (NtAI), Large-scale State Transitions (LST), and loss of heterozygosity (LOH), as previously described [190]. The CA20 score was defined as the sum of the expression of 20 genes (log2 median centered). These genes include: AURKA, CCNA2, CCND1, CCNE2, CDK1, CEP63, CEP152, E2F1, E2F2, LMO4, MDM2, MYCN, NDRG1, NEK2, PIN1, PLK1, PLK4, SASS6, STIL and TUBG1 [99].