Joining the PARty: PARP Regulation of KDM5A during DNA Repair (and Transcription?)

Abstract

The lysine demethylase KDM5A collaborates with PARP1 and the histone variant macroH2A1.2 to modulate chromatin to promote DNA repair. Indeed, KDM5A engages poly(ADP-ribose) (PAR) chains at damage sites through a previously uncharacterized coiled-coil domain, a novel binding mode for PAR interactions. While KDM5A is a well-known transcriptional regulator, its function in DNA repair is only now emerging. Here we review the molecular mechanisms that regulate this PARP1-macroH2A1.2-KDM5A axis in DNA damage and consider the potential involvement of this pathway in transcription regulation and cancer. Using KDM5A as an example, we discuss how multi-functional chromatin proteins transition between several DNA-based processes, which must be coordinated to protect the integrity of the genome and epigenome. The dysregulation of chromatin and loss of genome integrity that is prevalent in human diseases including cancer may be related and could provide opportunities to target multitasking proteins with these pathways as therapeutic strategies.

Graphical Abstract

PARP1-macroH2A1.2 regulates the lysine demethylase KDM5A in DNA repair, including through poly(ADP-ribose) (PAR) binding by its coiled coil domain. The collaboration of this pathway beyond genome integrity in transcription and cancer is unknown. We discuss this question and the potential therapeutic targeting of this pathway using catalytic or PARP inhibitors.

INTRODUCTION

DNA accessibility represents a major regulatory step in transcription, replication and DNA repair. This process is highly dynamic and is influenced by various chromatin states created by histone proteins that form the basic nucleosomal unit of chromatin, along with other DNA and chromatin interacting proteins.^{[1, 2]} Repair of chromatinized DNA breaks requires access to the lesion, which involves chromatin structural alterations. Post-repair restoration of the chromatin environment surrounding the break to the pre-damaged state is also vital for epigenetic integrity.^{[3, 4]} DNA damage from endogenous or exogenous sources can create a variety of aberrant DNA structures that must be resolved to avoid mutagenic outcomes. Among these lesions, DNA double-strand breaks (DSBs) represent a significant threat to the genome and cellular homeostasis if not accurately repaired, as these lesions can result in not only local mutations but also chromosomal aberrations including translocations, deletions and aneuploidy.^[5] DSBs originate from several sources including formation of atypical DNA and RNA structures, unresolved replication errors, aberrant nuclease activity, proteome dysregulation and cancer therapies (e.g. ionizing radiation (IR), topoisomerase inhibitors and DNA-alkylating agents).^[6–8]

In order to efficiently sense and repair DSBs, cells have evolved multiple DNA damage response (DDR) mechanisms to identify and address these genetic threats. DSB repair occurs primarily through two separate pathways, homologous recombination repair (HR) and nonhomologous end-joining (NHEJ); although several additional related pathways also exist.^[9] The DDR functions within chromatin and is influenced in several ways including by post-translational modifications (PTMs) of histones. The cell cycle phase in which the break occurs also plays an important role in determining repair pathway choice since NHEJ occurs throughout the cell cycle, while HR repair is primarily constrained to S- and G2-phases when a sister chromatid is available for use as a template for error free repair. An early initiating event in HR repair is the partial resection of the DSB end by CtIP (CtBP-interacting protein) and the MRN (Mre11/Rad50/Nbs1) complex, which generate a 3’ DNA overhang that in turn inhibits NHEJ. The 3’ overhang is coated with RPA (Replication Protein A) and later replaced with the RAD51 recombinase, which facilitates homology search and strand invasion into the template strand, ultimately allowing for synthesis dependent repair of the DSB.^[10] When repaired by NHEJ, DSBs are recognized and bound by the KU70-KU80 heterodimer, which protects DNA ends from resection. The DNA ends are processed to be compatible for ligation and the autophosphorylation activity of DNA-PKCs (DNA-dependent protein kinase catalytic subunit) recruits the XRCC4-LIG4 (X-ray repair cross-complementing protein 4- Ligase 4) complex to the DSBs for ligation. In some instances, limited resection or modification of the DNA ends by nucleases and polymerases is required before repair can proceed.^[11] Given that NHEJ repairs DSBs by non-templated re-ligation of free DNA ends, this pathway is considered to be more error prone than HR. DSB repair pathway choice is also heavily influenced by the environment and function of chromatin in which the break occurs. For example, DSBs within heterochromatin or euchromatin, as well as transcribed verses non-transcribed regions, may be repaired through different mechanisms.^[12–14] Additionally, the DDR must be coordinated with other processes that engage DNA, including for example DNA replication and transcription, to ensure these events do not interfere with each other. This coordination relies on a diverse set of histone PTMs, which serve to recruit specific effectors that modify chromatin structure and function, allowing different DNA-based processes to co-exist and not conflict with each other.^{[12, 15–18]} Therefore, DSB repair can be dictated by the type of lesion and its local chromatin environment, cell cycle phase and function of the break-proximal loci; which can ultimately influence the DSB repair pathway that engages and repairs the break. Integrative signaling between both the DDR and chromatin is necessary to ensure both repair efficiency and epigenome stability to maintain the fidelity of the underlying genetic sequence and its function ^[15].

Considering the importance of the DDR in maintaining genome stability and suppressing mutations that can promote diseases, it is perhaps not surprising that alterations in chromatin modifying enzymes are frequently observed in human diseases including cancer.^[19] In addition to driving mutagenesis and cancer promotion, cancer-associated dysregulation of repair and chromatin factors can also produce therapeutic vulnerabilities in some circumstances. A well-established example of this principle is the specific sensitivity of HR repair deficient cancers to the inhibition of poly(ADP-ribose) polymerases (PARPs).^[20] Upon DNA damage, DSBs are rapidly recognized by PARP1, which catalyzes poly(ADP-ribose) (PAR) chain addition at the break site, which serves as a recruitment signal for early DDR factors (discussed below).^[21] Many clinically available PARP inhibitors (PARPi) work by trapping PARP1 on chromatin, which forms replication dependent DSBs and single strand gaps ^[22–24]. This treatment strategy has been found to be very effective in HR-deficient cells, and in particular BRCA-deficient cancers, which are unable to effectively repair DSBs that are generated by PARP-trapping, although additional mechanisms are also involved in PARPi responses in BRCA-deficient cancers ^[25].

Discern how DNA accessibility is regulated by chromatin and how dysregulation of these pathways promote human diseases as well as therapeutic vulnerabilities. One such histone PTM is methylation, which occurs primarily on lysine and arginine residues and is known to be involved in gene regulation, heterochromatin formation and chromatin responses to DNA damage.^[26–29] The complex regulation elicited by methylation is further enhanced by the existence of multiple methylation states on histones, as well as methylation of RNA and DNA.^{[27, 30]} For example, mono, di-, and tri-methylation can occur on the same lysine residue within proteins, with each modification state displaying unique effector protein interactions and functions, which can be further influenced by methylation status of localized DNA and RNA.^{[27, 29–31]} Histone methylation is dynamically regulated by methyltransferase “writer” and demethylase “eraser” enzymes. Some factors involved in histone methylation control multiple DNA templated processes through their substrate specific activity and recruitment of additional effectors.^{[27, 32]} One such emerging example is the lysine specific demethylase KDM5A (JARID1A, RBP2), which demethylates tri-methylated histone H3 at lysine 4 (H3K4me3) through its catalytic Jumonji C (JmjC) domain (Fig. 1a).^[33] Upon DNA damage, H3K4me2/3, marks associated with active transcription are demethylated in proximity to DSBs.^[34–37] Indeed, KDM5A mediated H3K4me3 demethylation promotes the recruitment of ZMYND8 and the nucleosome remodeling by the deacetylase (NuRD) complex, which acts to silence DSB proximal transcription and repair DSBs by HR when recruited to breaks within actively transcribed genomic loci (Fig. 1b).^{[34, 38–40]} KDM5A is also known to promote cancer through its ability to regulate gene expression. For example KDM5A inhibits p53 expression,^[41] regulates epithelial-to-mesenchymal transition (EMT) markers^[42] and impacts the expression of the cell cycle regulator p27.^[43] Recently, KDM5A has gained attention for its role in promoting the progression of specific cancer types, including breast,^[44] osteosarcoma,^[43] and lung cancers,^{[42, 45]} all of which present with significantly elevated levels of KDM5A.^[46–48] Additional cancers have been identified with KDM5A mutations and importantly, silencing of KDM5A expression can reduce progression in some cancer-contexts (Fig. 1b).^{[41, 49]} These observations have established KDM5A as an attractive therapeutic target in cancer, which has driven drug development efforts towards identifying specific KDM inhibitors.^[48]

FIGURE 1.

Summary of KDM5A Domains and their Functions in DNA repair, Transcription and Cancer. a) Map of KDM5A domains with interactions and specific functions indicated. b) Schematic of known KDM5A functions in DSB repair and transcription. Cancers with KDM5A functions implicated in their initiation and promotion are indicated. Recruitment of KDM5A to DSBs is dependent on PARP1, PAR chain binding (PAR = small blue circles) and the histone variant macroH2A1.2. Once at DSBs, KDM5A promotes demethylation of H3K4me3, DNA repair by homologous recombination, and break-induced transcriptional repression. In undamaged conditions, KDM5A represses target genes through demethylation of H3K4me3. Whether the PARP1-macroH2A1.2 driven DSB functions of KDM5A are also involved in gene regulation and cancer is unknown. PAR – Poly(ADP-ribose), Jmj – jumonji domain, ARID – A–T Rich Interaction Domain, PHD – plant homeodomain.

Given the multi-functionality of KDM5A, it remains unclear how its activities in different biological pathways are related and whether these functions contribute uniquely or collectively to human diseases where KDM5A pathways are found to be dysregulated. Here, we examine recent advances in our understanding of how KDM5A is regulated during the DDR, with a focus on a newly defined PARP1 and macroH2A pathway. We consider the mechanistic implications of these findings, including in informing the function of these factors in gene regulation and cancer, as well as the relevance of targeting of these pathways as a therapeutic strategy in cancer.

KDM5A AS A PAR EFFECTOR

How KDM5A engages chromatin at sites of DNA damage and gene regulatory loci is complex and known to be controlled by several factors, signals and interaction events. KDM5A binding to histone H3 requires two plant homeodomain (PHD) zinc fingers (ZnFs) where PHD1 recognizes the unmodified N-terminal tail of H3^[50] and PHD3 specifically interacts with H3K4me3.^[51] Interestingly, these PHD domains are also involved in KDM5A DNA damage functions where the PHD1 domain is required to localize KDM5A to DSBs, while the PHD3 domain, which is dispensable for break recruitment, is required for KDM5A DSB repair functions downstream of break recruitment (Fig. 1a).^{[34, 52]} KDM5A can promote DNA repair and transcriptional silencing through the recruitment of additional effector complexes, including ZMYND8 and the NuRD complex.^[34] At DSBs, ZMYND8 requires the histone acetyltransferase (HAT) TIP60 and its bromodomain (BRD) to localize to breaks, suggesting an important, yet poorly understood, requirement for acetylation in this pathway^[40]. ZMYND8 interacts with the NuRD complex through direct interactions between the NuRD component GATAD2A and its MYND domain, which acts to recruit the NuRD chromatin remodeling complex, including the ATPase CHD4, to DNA DSBs.^{[40, 53–55]} The NuRD complex is subsequently required for break-proximal repression of transcription and the repair of DSBs by HR, which is thought to involve its chromatin remodeling activities.^{[40, 56]} KDM5A is also known to regulate the transcription of many cancer associated genes but whether or not its DDR functions relate to those involving gene regulation is unclear (Fig. 1b).^{[41, 42]}

Considering the diverse roles of KDM5A in the DDR and in transcription, additional modes of regulation for KDM5A have been investigated. Recently, our work uncovered new regulatory elements controlling KDM5A functions in the DDR and damage-proximal transcriptional responses, which includes the involvement of PARP1 in promoting the recruitment of KDM5A to DSBs.^[55] PAR mediated signaling has emerged as a diverse and essential mode of regulation for multiple cellular processes, including the DDR and transcription^{[57, 58]}. Cells express multiple PARPs that can deposit PAR chains of various lengths and structures under different cellular conditions. Of these, PARP1 is the primary PARP in mammalian cells and is responsible for the vast majority of PAR chains produced.^[59] The activity and substrate preferences of PARP1 can be influenced by its interacting partners. For example, recent studies have demonstrated that in response to DNA damage, PARP1 associates with HPF1 (Histone PARylation Factor 1) and produces PAR chains primarily attached to serine residues.^[60–63] To synthesize PAR chains, PARPs utilize free NAD+ as a substrate to covalently link an adenosine diphosphate (ADP) ribose (ADP-ribose) unit onto target proteins, which releases nicotinamide (Fig. 2a). Repeats of this process results in the generation of PAR chains of various lengths and branched structures (Fig. 2b). Thus far this modification has been found to occur on Glutamate, Arginine, Asparagine, Cysteine, Lysine, Serine/p-Serine, Histidine, and Tyrosine residues. Several proteomic studies have collectively identified over 2000 PARylated proteins ^[63–70], as well as a number of proteins exhibiting non-covalent binding interactions with PAR chains ^{[21, 71–74]}. These studies have revealed a large and complex network of factors implicated in PARP biology.

FIGURE 2.

PAR Synthesis, Structure and Modes of Effector Recognition. a) Synthesis of Poly(ADP-ribose) chains by PARPs utilizes free NAD+ to covalently link PAR to target proteins containing appropriate nucleophilic sites (indicated by red “X”), a reaction that generates free nicotinamide. For chain elongation, new PAR molecules are iteratively added to the hydroxyl group distal from the target to make both linear and branched PAR chains. b) PAR chains contain distinct structural motifs that are recognized by specific effector proteins containing various PAR binding domains. Several PAR interaction domains are shown, with their specific binding region of the PAR chain indicated by the same color as the domain. FHA – forkhead-associated domain, OB Fold – oligosaccharide-binding fold, BRCT – BRCA1 C-terminus domain, PBZ – PAR binding zinc finger.

PAR effectors are capable of recognizing various structural features of PAR chains through several well characterized domains. Effector proteins with OB-fold, FHA, or WWE domains specifically recognize iso-ADP-ribose (iso-ADPR), the smallest structural unit of a PAR chain which contains the ribose–ribose glycosidic bond (Fig. 2b).^{[75, 76]} Other effectors containing macro or BRCT domains can recognize entire ADPR units, while PAR binding ZnF (PBZ) domains require two adjacent ribose groups of the PAR chain for recognition (Fig. 2b).^{[75, 77–79]} Notably, the newly identified PAR binding capability of KDM5A does not utilize these previously characterized PAR binding domains. Rather, KDM5A binds directly to PARP1 catalyzed PAR chains through a previously uncharacterized, C-terminally localized, coiled-coil (CC) domain (Fig. 1).^[55] Whether or not additional proteins similarly interact with PAR chains by utilizing CC domains awaits further investigation.

In addition to PAR binding, KDM5A has been found to be PARylated on two individual serine residues using proteomics^[66], although the function of this modification on KDM5A is unknown. Interestingly, KDM5A interacts with the histone H2A variant macroH2A1.2 (mH2A1.2), an interaction that promotes KDM5A recruitment to DNA damage sites.^[55] Although macrodomains are known to bind PAR, macroH2A1.2 is one of two isoforms formed by alternative splicing of macroH2A1, an event that removes key residues within the macroH2A1.2 macrodomain that abolishes its ability to bind PAR. In addition, macroH2A1.2 recruitment to DNA breaks is PARP1-independent.^[80] Thus, involvement of macroH2A1.2 in aiding KDM5A interactions at DSBs is likely not due to direct PAR binding, including PARylated KDM5A, by macroH2A1.2. However, macroH2A1.2 has been reported to be PARylated,^[65] presenting the possibility that KDM5A binds PARylated macroH2A1.2. This is consistent with findings that the interaction between KDM5A and macroH2A1.2 is diminished upon PARPi treatment.^[55] The macrodomain of the other isoform, macroH2A1.1, does bind PAR chains and its role in KDM5A-dependent DDR functions have not been fully characterized. While interactions between macroH2A1.1 and KDM5A were not observed, and macroH2A1.1 deficiency results in increased NHEJ rather than a reduction in HR repair,^{[80, 81]} it remains untested if macroH2A1.1 participates in KDM5A recruitment to breaks and break-proximal transcriptional repression. However, macroH2A1.1, through an ability to bind PAR chains, has been demonstrated to suppress PARP activity while also protecting PAR chain stability in response to oxidative stress ^[82]. Given that PAR chain length and PARP activation regulate the association of KDM5A with DSBs, it is worth testing how loss of macroH2A1.1 specifically may alter KDM5A-dependent functions. Given that macroH2A1.1 cells display limited defects in HR, unlike KDM5A- and macroH2A1.2-deficient cells, we favor the idea that macroH2A1.2 specifically functions along KDM5A in the DDR while macroH2A1.1 is involved in other DDR pathways, including those involving Alt-NHEJ, single strand break (SSB) repair and NHEJ (reviewed in ^[83]).

Future work should aim to delineate how recognition of PAR chains and binding to macroH2A1.2 can fine-tune KDM5A, and other PAR binding proteins, functions in the DDR and potentially also in gene regulation. The rationale for this idea is supported by the finding that the expression of some KDM5A target genes are influenced by PARP1 activity.^{[55, 84, 85]} Additionally, PARylation is known to regulate gene expression in several contexts, in addition to KDM5A (reviewed in ^[57]). A comprehensive analysis of direct effects of PARP inhibition or depletion on KDM5A recruitment to these genes has not been reported. These multiple modes of KDM5A regulation may also provide targetable vulnerabilities in KDM5A driven cancers given that both PARP1 activity and interactions with mH2A1.2 are needed for KDM5A functions at DSBs (discussed below).

PAR BINDING BY COILED-COIL DOMAINS: NEW PARADIGM OR KDM5A SPECIFIC?

The CC domain mediated interaction between KDM5A and PAR chains raises several important mechanistic questions; foremost, do additional PAR regulated factors engage PAR chains through this mode of direct recognition? CC domains are highly abundant throughout the proteome and are found in approximately 5–10% of all proteins.^[86] In a large number of proteins, CC domains serve as dynamic regions that can scaffold and facilitate spatial organization of other functional regions.^[86] Outside of the leucine zipper class of CC domains, there is a paucity of examples for involvement of CC regions in direct nucleic acid recognition. Thus, the identification of a direct PAR binding CC domain within KDM5A establishes a potentially new paradigm, and the need to evaluate additional CC domain containing factors to determine if they can similarly bind PAR. Given the abundance of CC domains and PAR interactions within the human proteome, it seems unlikely that KDM5A represents the only member of this group that binds PAR chains through a CC domain. It is also notable that the KDM5A PAR recognition domain additionally contains an intrinsically disordered region (IDR), which may contribute to the binding interaction though the specific role of the IDR in PAR recognition.^[55] Structural investigations of the KDM5A PAR interaction domain, which includes the CC region, in complex with PAR would be able to illuminate the key elements required for direct CC domain recognition of PAR chains. Additionally, CC domains and PAR chain deposition are known prerequisites for the formation of phase separated liquid compartments; particularly among transcriptional regulators, many of which also encompass intrinsically disordered protein (IDP) elements.^[87–91] This principle has been demonstrated for the related H3K27 lysine demethylase KDM6A (UTX), which undergoes LLPS to facilitate transcriptional repressive functions ^[92]. The potential for KDM5A to exist and function within phase separated compartments is yet to be explored, but could help explain how it regulates both DNA repair and transcription, which may involve PAR recognition by the CC domain of KDM5A. Although speculative, phase separation may provide yet another property of gene regulation and DNA repair that could be targeting therapeutically, which is an emerging concept in the treatment of several human diseases including cancer and neurodegeneration ^[93].

REGULATION OF PAR CONJUGATE FEATURES DICTATE BINDERS AND BIOLOGY

The multitude of cellular functions regulated by PAR chains is a direct result of the structural diversity of these modifications. Conjugation of PAR to substrates can occur as a monomer (i.e., MARylation) or a complex polymer exceeding 60 units in length.^{[94, 95]} PAR chains also occur in both branched and linear forms, which can further direct effector binding and function (Fig. 2b).^[96] This diversity in PAR chain length and conformation suggests that functional codes may exist for PAR, akin to PTMs on histones that comprise “the histone code”. Our work has revealed that the KDM5A CC domain preferentially binds to mid-sized and long PAR chains (≥27-mer) with an affinity comparable to established PAR binding domains.^[55] This selective PAR chain length binding likely governs KDM5A interactions with specific genomic loci, including DSBs, which may further orchestrate KDM5A DDR and/or transcription functions. Aspects of this principle were recently demonstrated for other known PAR binding DDR factors^[75] including DEK, XPA, and p53, which all were found to prefer longer (40-mer) PAR chains using a photoaffinity-based approach.^{[71, 97]} The DDR factors APE1 and POLβ were identified to prefer mid (20-mer) to short PAR chains (8-mer). Interestingly, these factors also bind to linear PAR chains with higher affinity than branched chains of the same length.^[98] Early work on structure specific PAR chain binding also identified that histones can bind PAR chains and prefer complex long and branched PAR (examples of DDR factors with known PAR length binding preferences are summarized in Fig. 3).^[99] The structural diversity of PAR chains is not solely the product of PARPs. Enzymes which degrade PAR chains including Poly(ADP-ribose) Glycohydrolase (PARG) and the ADP-ribosylhydrolase ARH3, also shape the architecture of PAR chains through degrading specific linkages.^[100–102] Human PARG is known to exhibit both exoglycosidase and endoglycosidase activity, meaning that this enzyme is able to cleave protein-distal PAR to generate free ADP-ribose or within a PAR chain, generating free PAR.^[100] Conversely, ARH3 only possesses exoglycosidase activity.^[103] Multiple lines of evidence now support a model where the rapid removal or modification of PAR by PARG or ARH3 is essential for PAR regulated processes like DDR and transcriptional regulation.^{[65, 104–106]} Thus, the varying substrate specificity between PAR erasers further adds to the complexity and potential regulatory functions of PAR chains and their effectors.

FIGURE 3.

PAR Length Dependent Binding by DDR Factors. Examples of known DDR PAR readers are shown, along with their preferences for binding PAR chains of variable lengths. Factors also reported to regulate transcription are shown in red.

IS THE PARP1-MACROH2A1.2-KDM5A AXIS ACTIVE IN GENE REGULATION AND CANCER?

Outside of promoting DNA repair, KDM5A is known to regulate promoter expression of several target genes through its H3K4me3 demethylase activity.^{[45, 107, 108]} Interactions with repressive targets are regulated through multiple mechanisms and multivalent binding events. In some instances, the recognition of target genes by KDM5A requires DNA binding by the KDM5A ARID domain to its known recognition sequence (CCGCCC); which has been demonstrated to occur at several promoters containing this site.^[45] In order to regulate some cell cycle control genes, KDM5A associates with E2F4 transcription factors (TFs) through p130.^[107] This interaction occurs in a cell cycle regulated manner and is mediated by the KDM5A LxCxE motif. Transcriptional regulation by KDM5A also occurs through demethylase-independent mechanisms; for example, KDM5A can associate with circadian CLOCK and BMAL1, which promotes transcription of the Per2 gene by limiting the deacetylase activity of HDAC1.^[109] While the regulation of transcription by KDM5A is extensively characterized, the potential involvement of PARP1 activity in controlling these KDM5A functions has not been fully delineated. Regulation of transcription by PARylation is known to occur in multiple ways including direct PARylation of TFs,^{[110, 111]} regulation of chromatin structure and epigenetic states,^[112] and seeding membraneless compartments.^{[90, 113]} The H2A variant macroH2A1.2 is another KDM5A regulator heavily implicated in transcription, with the incorporation of this histone variant into nucleosomes influencing transcription directly or through the recruitment of specific effectors.^[114] During breast cancer-induced osteoclastogenesis the repression of lysyl oxidase (LOX) has been found to be under the control of macroH2A1.2. In this setting, macroH2A1.2 serves to recruit the methyltransferase E2H2 which enriches the repressive H3K27me3 mark at the LOX gene.^[115] The function of macroH2A in transcription is not limited to repression as studies utilizing ChIP-Seq have identified nucleosomes containing macroH2A at both expressed and repressed genes.^[116] The incorporation of macroH2A into nucleosomes can serve to limit the accessibility of transcriptional repressors or activators and similarly the activity of PARP1 in transcription can act to repress or enhance gene expression.^{[57, 58, 64, 117]} Considering that KDM5A relies on PARP1 activity and macroH2A1.2 for its chromatin localization during the DDR, exploring the effect of these factors on KDM5A during normal transcription, as well as in disease settings, is warranted and will be help to know how unique or universal PARP1-macroH2A1.2 are in regulating KDM5A functions.

The transcriptional targets of macroH2A, PARP1, and KDM5A share a common relevance to cancer progression as many of these genes encode for factors regulating cancer promoting processes (Fig. 4). Transcription networks governed by KDM5A renders this factor central to the progression of several cancers. Furthermore, identifying KDM5A regulation by macroH2A1.2 and PARP1 presents several exciting opportunities for therapeutic intervention (discussed below), and is an idea supported by the striking phenotypic overlap between these factors in cancer. The role of PARP1 in cancer is well studied and diverse, with the activity of PARP1 known to control the expression of cancer driving genes,^[84] regulate cell cycle checkpoints^[85] and cancer cell invasiveness,^[118] in addition to numerous other cancer promoting functions (Fig. 4).^[119] Cancer and gene regulatory functions of macroH2A have also been described to include DSB repair,^[80] expression of EMT markers,^[120] and chemoresistance^[114] (Fig. 4; reviewed in^{[114, 121]}). Intriguingly, these functions are shared with KDM5A, including expression of cancer driving genes (including EMT markers),^{[41, 42]} proliferation and metastasis,^[43] and chemoresistance (Fig. 4).^{[46, 47]} Several scenarios exist which could explain the functional overlap between these factors. Recruitment of KDM5A to DSBs relies on both macroH2A and PARP1 and it is conceivable that these signals also dictate KDM5A recruitment to genes involved in cancer associated pathways. This recruitment could be further fine-tuned by chromatin and/or PAR chain length/branching interactions to create specific codes for KDM5A binding and function. These may also involve other combinatorial signals, including macroH2A modifications, that are used to orchestrate site-specific recruitment and activity of this histone demethylase. Induction of DNA damage is well documented to alter PARP1 substrates and specifically induce modification at serine residues through PARP1-HPF1 interactions.^[122] This induced variability in PAR chains could provide a way to control KDM5A accumulation at DSBs vs. gene regulatory loci, including those involved in cancer. The expression of macroH2A variants and their incorporation into nucleosomes can also provide potential context specific signals governing KDM5A engagement with chromatin, where the ratio of macroH2A1.1 to macroH2A1.2 is known to vary between cancer types and can regulate PARP1 availability.^[123] More effort is needed to identify the precise regulatory mechanisms used by KDM5A to transition between different genomic locations and functions to ultimately control gene regulatory networks and DNA repair transactions. It cannot be ruled out that additional, as-of-yet identified regulatory factors, PTMs and interactions govern KDM5A dynamics within chromatin during these biological processes both in normal and disease states.

FIGURE 4.

Commonly Shared Cancer Driving Functions of PARPs, macroH2A, and KDM5A. PARPs, macroH2A, and KDM5A have all been implicated in cancer through various mechanisms, several of which are shared. These include the regulation of EMT, DDR, transcription, cell cycle and drug resistance. Potential crosstalk and/or dependencies between these pathways both in specific biological processes beyond DSB repair and in cancer remain largely uncharacterized.

THERAPEUTIC TARGETING OF KDM5A WITH PARP INHIBITORS?

Considering the reported prevalence of KDM5A-dependent proliferation in several cancers^{[43, 46, 48]}, the identification of PARP1 and macroH2A1.2 as upstream regulators of KDM5A present potentially new and exciting opportunities for therapeutic intervention. Foremost, the dependency of KDM5A on PARP1 indicates that cancers driven by KDM5A overexpression ^[46] may be selectively treated with FDA approved PARPi. Thus far, PARPi use has been limited to HR-deficient tumors with mutations in BRCA1/2, where the mechanism of PARP inhibition results in general cellular toxicity as trapping PARP creates replication dependent DSBs reliant on BRCA-function.^{[23, 24]} Importantly, the role of KDM5A in DSB repair can be abolished by PARP-depletion in addition to PARP trapping by PARPi; as demonstrated by the lack of KDM5A function in PARP knockout and knockdown cells.^{[34, 55]} This observation indicates that KDM5A driven cancers may be sensitive to treatment with both trapping and non-trapping PARPi. Non-trapping PARPi may provide less toxic mechanisms in this context although the potential use of PARPi in KDM5A-dependent cancers awaits evaluation. While PARPi are likely to target several pathways in addition to KDM5A, a growing body of work has begun to highlight additional potential opportunities to use PARPi therapeutically, a concept recently reviewed by Huang and Kraus.^[57] While many challenges remain, future work is warranted to continue to test the potential use of PARPi in cancer settings beyond homologous recombination deficient (HRD) tumors. Ovarian cancers provide one example where treatment with PARPi is FDA approved and recommended for use regardless of HRD status under certain clinical settings. ^[124]

The mode of PAR binding by KDM5A through a novel CC domain is an additional feature of this interaction that could be capitalized on therapeutically and may be relevant to the evaluation of other demethylases involved in cancer progression. To date, there are 24 identified KDMs in human cells, with some known or predicted to contain CC domains and many operating under the control of PARP1 activity.^{[117, 125–128]} Intriguingly, the effect of PARP1 activity on KDMs can vary widely. For example, KDM5B, which also demethylates H3K4me3 and shares a high degree of sequence similarity with KDM5A, is negatively regulated by PARP1 activity during transcription.^[117] Inquiry into positively regulated PARP1 genes in MCF-7 cells found that PARP1 mediated PARylation blocked KDM5B recruitment to promoters and prevented the demethylation of H3K4, creating a transcriptionally active environment.^[117] How PARP1 activity is able to promote both of these opposing functions is not yet clear but could likely be attributed to structural differences between PAR chains in these cancer cell lines as well as the fact that KDM5A but not KDM5B contains a CC domain and binds PAR.^[55] Furthermore, KDM5A and KDM5B have both been implicated as drivers of lung cancer progression.^{[45, 129]} To facilitate lung cancer progression, KDM5A directly interacts with the promoters of cell cycle regulating genes to suppress their expression and drive cell proliferation.^[45] Alternatively, KDM5B has been found to enhance the expression of the transcription factors E2F1 and E2F2, which can drive lung cancer proliferation.^[129] Therefore, uncovering the conditions which select for the expression and function of one or more KDM5 demethylase in cancer could inform mechanistically on how these enzymes are acting in cancer, as well as reveal therapeutic strategies for this family of demethylases.

The growing interest in KDM5A as a cancer driver has spurred efforts to develop specific small molecule inhibitors for this factor (summarized in Fig. 5b).^[48] The majority of inhibitors developed thus far work through blocking the activity of the KDM5A JmjC domain, a domain that iteratively demethylates H3K4me3 through a two-step reaction mechanism. The first step requires 2-oxoglutarate (2OG), oxygen (O₂) and Fe(II), to form a hydroxymethyl intermediate, which also generates carbon dioxide (CO₂) and succinate as byproducts (Fig. 5a). In the second step, formaldehyde (HCHO) is eliminated regenerating unmethylated lysine. In line with this reaction mechanism, drugs which act as Fe(II) chelators or 2OG competitive binders have been found to be effective, but relatively non-specific, inhibitors of KDM5A and other KDMs.^{[130, 131]} Other classes of Jmj inhibitors have been developed, including isonicotinic acids (2,4-PDCA), pyrazoles, and heterocyclic compounds (ex. imidazopyridine). All of these compound classes are more selective for Jmj domain containing demethylases but are in some cases limited by the sequence and structural similarity between different KDM sub-families. Notably, KDM4 and KDM6 members share a high degree of sequence similarity with KDM5 members, making selective targeting challenging. The pyrimidinone derived molecule CPI-455 has emerged as one of the most selective inhibitors for KDM5 family members, as this compound displays a 500-fold preference for inhibiting KDM5 members over other demethylases.^[132] However, even this level of specificity can have limited applications if the objective is to individually and selectively target KDM5 members. Rhodium(III) based metal complexes have now been demonstrated to have selectivity toward KDM5A in cell-based assays.^[133] These drugs work by blocking the association of KDM5A with histone H3, though further studies are needed to evaluate the efficacy of this drug class in vivo. Importantly, the newly observed KDM5A interactions with PAR chains and the ability to block its function with PARPi present potential innovative and new routes through which KDM5A may be targeted. While the CC domain within KDM5A has only recently been reported,^[55] inhibitors of other CC domains have been developed but are currently limited to peptide and antibody-based drugs.^[134] The development of more specific and biologically available drugs to modulate CC domains and their activities, including the one within KDM5A, are warranted by their prevalence in the proteome and their association with proteins involved in cancer. Further, the potential to repurpose FDA approved PARPi to specifically target KDM5A driven cancers is an exciting possibility that warrants testing.

FIGURE 5.

KDM5A as a Therapeutic Target. a) Reaction mechanism of KDM5A mediated demethylation of H3K4me3. In the first reaction step, the JmjC domain dependent catalytic activity utilizes Fe(II), oxygen (O₂) and 2-oxoglutarate (2OG), to form a hydroxymethyl intermediate, which also generates succinate and carbon dioxide (CO₂) as byproducts. In the second step, formaldehyde (HCHO) is eliminated, regenerating unmethylated lysine. b) Summary of small molecule inhibitors capable of disrupting KDM5A functions through targeting specific KDM5A domains.

CONCLUDING REMARKS

The list of cellular functions governed by KDM5A has rapidly expanded in recent years to include roles in regulating the DDR, multiple cancer driving phenotypes and transcription.^{[33, 46, 108]} In order to accurately execute these various processes, KDM5A is regulated through multivalent chromatin interactions, including a newly discovered interaction with PAR chains and macroH2A1.2. These diverse interactions provide a framework for understanding how multifunctional proteins like KDM5A can be directed to perform and transition between different cellular tasks. The deposition of PAR chains occurs in response to a wide array of cellular events and results in structurally diverse PAR chains that can act to recruit specific effectors.^{[59, 75, 119, 125, 128]} The binding of KDM5A to mid-length PAR chains in the context of DNA damage invites further inquiry into how (or if) KDM5A interactions with PAR control its activity within different biological processes (i.e., transcription vs. DSB repair). Considering that the expression of many KDM5A regulated genes are also influenced by PARPs, it seems likely that KDM5A may interact with PAR either directly or through additional effectors to facilitate gene regulation; a possibility that necessitates further investigation. Binding to PAR chains is known to facilitate the formation of membraneless compartments^[90] and the creation of these condensates can govern transcriptional regulation, including gene repression by histone demethylases.^{[89, 92]} As a PAR and chromatin-binding factor, KDM5A should also be considered in this process as it could underlie its functions on chromatin and provide an additional point of intervention in KDM5A driven cancers.^[135] A promising outlook for treating KDM5A driven cancers is the re-tooling of clinically available PARPi to target KDM5A in these cancer types. Given the connections between transcription, DNA repair and replication with PARPs and epigenetic regulators including KDM5A, an evaluation of how PARP inhibition impacts transcription and DNA repair beyond BRCA mutations is justified and could inform on new therapeutic strategies to target both DNA repair and chromatin factor deficiencies that are known to be prevalent in cancer.

Abbreviations:

2OG

2-oxoglutarate

CC

coiled coil

DDR

DNA damage response

DSB

DNA double-strand break

EMT

epithelial-to-mesenchymal transition

FHA

forkhead-associated domain

HAT

histone acetyltransferase

HPF1

histone PARylation factor 1

HR

homologous recombination

IDR

intrinsically disordered region

Jmj

jumonji domain

IR

ionizing radiation

LLPS

liquid-liquid phase separation

LOX

lysyl oxidase

NHEJ

nonhomologous end-joining

NuRD

nucleosome remodeling and deacetylase complex

PAR

poly(ADP-ribose)

PARG

poly(ADP-ribose) glycohydrolase

PARP

poly(ADP-ribose) polymerase

PBZ

PAR binding zinc finger

PHD

plant homeodomain

PTM

post-translational modification

TFs

transcription factors

ZnF

zinc finger