PARP Inhibitors – Trapped in a Toxic Love Affair

The primary target of clinically used PARP inhibitors (PARPi) is PARP1, a chromatin-associated poly-ART (ADP-riboslytransferase) enzyme, which is activated by binding to damaged DNA. Once bound to DNA, PARP1 uses NAD⁺ to synthesize multiple ADP-ribose units on adjacent substrate proteins. The resultant ADP-ribose homopolymer chains, now known as poly(ADP-ribose) (PAR), facilitate DNA repair by driving local changes in chromatin structure and also the recruitment of DNA repair effector proteins. PARP1's role in these processes began to be uncovered in earnest from the late 1970s onwards, when alkylating agents and ionizing radiation were both shown to increase PARP1's catalytic activity, reducing cellular levels of NAD⁺ while increasing PAR (1). Relatively nonspecific, low potency “toolbox” PARP1 inhibitors, such as 3-aminobenzamide (3-AB), reversed these NAD⁺/PAR phenotypes and sensitized cells to alkylating agents, largely by impairing the repair of DNA damage. Early studies also demonstrated that PARP1 has a “shuttle mechanism” where its catalytic activity is linked to its binding to and dissociation from DNA (1). We now know that PARP1 binds DNA via N-terminal zinc fingers (ZnF). PARP1-DNA binding initiates an allosteric signaling cascade and conformational change in the protein that ultimately leads to the release of an autoinhibitory interaction between the helical domain (HD) and the catalytic ADP-ribosyl transferase domain (ART). The change in the HD/ART interaction allows NAD⁺ to access ART, an event that drives PARylation and the subsequent recruitment of substrate proteins and the repair of the double helix. PARP1 eventually PARylates itself (autoPARylation), an act which drives the dissociation of PARP1 from DNA. In keeping with this PAR-dependent shuttle mechanism, reducing the level of PAR on PARP1 via PAR glycohydrolase (PARG) activity results in PARP1 being retained on DNA (Fig. 1). The connection between autoPARylation and the residence of PARP1 on DNA also led to the suggestion that catalytic inhibitors of PARP1 might impair the fitness of cells by preventing the release of PARP1 from damaged DNA (1).

Small molecule inhibitors of PARP1's catalytic activity (which also inhibit PARP1 paralogs, such as PARP2) were originally developed as potentiators of DNA damaging chemo- or radiotherapy; the first-in-human trial using a PARPi (initiated in 2003) assessed a PARPi combined with the alkylating agent temozolomide. However, once drug-like potent and selective PARP1 inhibitors were discovered, their potential as single-agent drugs was uncovered. Preclinical work demonstrated the ability of PARPi to kill tumor cells with homologous recombination (HR) defects, including those with BRCA1 or BRCA2 mutations, leaving nonmutant cells relatively unharmed (2, 3). This led to clinical trials demonstrating the efficacy of PARPi in treating HR-defective cancers and the eventual regulatory approval of five PARPi (talazoparib, rucaparib, niraparib, olaparib, and pamiparib), with others, such as veliparib, still undergoing clinical development.

A number of processes could have conceivably explained the BRCA1/2 synthetic lethal effects. Phenotypically, PARPi exposure in both BRCA1/2 mutant and wild-type (WT) cells elicits a DNA damage response. In WT cells, this response cooccurs with activation of HR, but in BRCA1/2 mutant cells the inability to carry out HR and effectively repair the damage caused by PARPi results in gross chromosomal aberrations and cell death (2, 3). Importantly, only prolonged exposure to PARPi caused the maximal synthetic lethal effect in in vitro and in vivo models of BRCA1/2 mutant cancer (2), suggesting that an accumulation of DNA damage over a series of cell cycles is required. Based on these observations and the known function of PARP1, it was originally proposed that the inhibition of PARylation by PARPi caused synthetic lethality by impairing the recruitment and activity of DNA repair proteins, phenocopying the effects of deleting PARP1; the persistence of unrepaired DNA lesions would ultimately cause the structural collapse of replication forks. Alternatively, PARPi could potentially “trap” PARP1 on DNA either by preventing the autoPARylation required for PARP1 release and/or by altering the conformation of PARP1 such that its release from DNA was prevented and/or delayed; this trapped PARP1 could conceivably cause a steric barrier to the normal progression of the replication fork, again causing a greater reliance upon HR (1). Indeed, in experiments where cells were exposed to the combination of the DNA alkylating agent methyl methanesulfonate (MMS) and the toolbox PARPi (4-AN), the amount of chromatin-associated PARP1 was increased (4). Deletion of PARP1 caused resistance to the 4-AN/MMS combination, suggesting that the trapping of the PARP1 protein by a PARPi could indeed cause cytotoxicity (4).

Whether the trapping hypothesis bore any relevance to clinical PARPi only became clear with work from Murai and colleagues published in Cancer Research in 2012 (5). This work established that chicken DT40 cells with a deletion in Parp1 or human tumor cells with PARP1 silencing were profoundly resistant to the clinical PARPi olaparib. When exposed to olaparib, Parp1^−/− DT40 cells also did not display an enhanced DNA damage response, suggesting that the presence of Parp1 was an absolute requirement for the DNA damage normally caused by PARPi. Olaparib exposure also increased the amount of chromatin-associated PARP1 and PARP2 in cells coexposed to MMS, suggesting that olaparib stabilizes a PARP1/DNA complex. Interestingly, veliparib, as potent inhibitor of PARylation as olaparib and niraparib, but a poor PARP1 “trapper,” did not elicit same level of BRCA synthetic lethality in DT40 cells (5). Subsequent “head-to-head” studies in isogenic human BRCA1/2 WT and null cells showed that BRCA1/2 synthetic lethal effects are better elicited by PARPi that effectively trap PARP1, and less correlated with the ability of a PARPi to inhibit PARylation. Consistent with this, unbiased mutagenesis screens showed that PARP1 deletion or PARP1 mutations that prevent trapping are drivers of resistance to those PARPi that trap PARP1 on DNA (6).

In totality, this work made the critical link between PARP1 trapping and the cytotoxic effects of clinical PARPi and led Murai and colleagues to propose that many clinical PARPi induce “reverse allostery,” where PARPi binding in the catalytic site alters the PARP1 allosteric signaling required for PARP1 release from DNA and/or fixes the conformation of PARP1 in such a way that the inhibitor-bound state has a higher affinity for DNA (5). Previous work had also shown that silencing of PARP1 elicited BRCA1/2 synthetic lethality, as did PARPi was later shown to trap PARP1 (2, 3). Thus, the work of Murai and colleagues suggested that two parallel (and potentially nonexclusive) mechanisms might operate: genetic inactivation or catalytic inhibition of PARP1 in the absence of trapping might cause the persistence of DNA lesions normally repaired by PARP1, such as single-strand DNA breaks (SSB). Upon encountering replication forks, these SSBs might stall and collapse replication forks, thus activating HR in BRCA1/2 WT cells but causing synthetic lethality in BRCA1/2 mutant cells; alternatively, the PARP1-DNA complex formed by trapping PARPi could itself be a DNA lesion, which similarly requires BRCA1/2 function to mediate DNA repair and prevent cell death (5). The precise balance of each mechanism in cells is unknown, but several lines of evidence suggest trapping might be more important for tumor-cell cytotoxicity. For example, PARPi that are effective at trapping and catalytic inhibition (e.g., talazoparib, rucaparib, olaparib) elicit greater BRCA1/2 synthetic lethal effects in in vitro cell–based models of BRCA1/2 mutation than PARPi, which are less effective in PARP1 trapping (e.g., veliparib) but have similar effects on catalytic activity (5). However, clinical trials that compare different PARPi in a head-to-head fashion have not, as yet, been performed, so it is not completely clear whether PARPi like veliparib have different therapeutic effects than PARPi with greater trapping capabilities.

Although this clinical issue remains to be resolved, the work of Murai and colleagues was hugely influential and stimulated a series of other discoveries that may be informative in terms of understanding how PARPi are best used clinically. Subsequent high-density CRISPR-Cas9 mutagenesis of PARP1 established that in addition to mutations in the DNA-facing zinc finger domains, mutations in regions of PARP1 not predicted to directly bind DNA also impair trapping and cause PARPi resistance (6). These include a PARP1 p.R591C (WGR domain) mutation also seen in a case of clinical de novo PARPi resistance (6). Mutations at p.R591 could conceivably reduce PARP1 trapping by interfering with the PARP1-nucleosome interaction, by facilitating PARP1 reverse allostery in the presence of a PARPi (6), or by altering the conformation of PARP1 when it is bound to a PARPi (7). Interestingly, hydrogen-deuterium exchange mass spectrometry analysis of PARP1 suggests that PARPi with different trapping properties induce different allosteric changes in PARP1, adding weight to the suggestion that these conformational changes might be more important to the PARP1 trapping phenotype than the lack of autoPARylation (7). Maximal PARP1-trapping has also been shown to require an interaction between PARPi, which largely sit within ART, and the p.D766/D770 residues in the PARP1 HD that sit opposite to the catalytic site in ART; this information not only reinforces how the HD/ART interaction is critical to whether the ART is active or not, but also indicates how veliparib, a PARPi with limited trapping properties, could be modified to a structural derivative that interacts with p.D766/D770 and elicits greater PARPi trapping and cytotoxicity (7). These discoveries, which in part were stimulated by Murai and colleagues' 2012 work, thus illustrate how dissecting the PARP1 trapping phenotype has been informative both in terms of understanding how the interdomain interactions of PARP1 control its normal function and how PARPi could be improved to target these features.

While alterations in PARP1 that impair PARP1 trapping cause PARPi resistance, PARPi sensitivity can be enhanced by increasing the number of DNA lesions that PARP1 normally binds to. For example, inactivation of RNA ribonuclease H2 proteins increases the amount of genomic uracil, which is normally processed into a PARP1-binding substrate. Thus, RNA ribonuclease H2 inactivation, in concert with PARPi, leads to more trapped PARP1 and more PARPi sensitivity (8). Similarly, maximal activation of PARPi-induced cGAS-STING signaling (a potential driver of PARPi-mediated immune activation in cancer) is reversed by PARP1 deletion, thus implicating PARP1 trapping in this phenotype as well (9). It is also possible that one of the deleterious side effects of PARPi, myelosuppression, might be driven by PARP1 trapping. Finally, the synergistic effects of PARPi when used in combination with chemotherapy (at least in preclinical models) also appear to be determined by trapping. PARP1 trapping and catalytic inhibition are critical for PARPi/temozolomide synergy but not for PARPi/topoisomerase I inhibitor synergy, where only catalytic inhibition of PARP1 is required (10). This suggests, once again, that a consideration of trapping and the original 2012 Cancer Research work might be important in determining how these drugs are best used clinically.

C.J. Lord reports personal fees from AstraZeneca, Merck KGaA, Artios, Syncona, Sun Pharma, Gerson Lehrman Group, Vertex, Tango, 3rd Rock, Ono Pharma, Dark Blue Therapeutics, Horizon Discovery, Abingworth outside the submitted work; in addition, C.J. Lord is an inventor on patents describing the use of PARP inhibitors and stands to gain from their use as part of the ICR “rewards to inventors” scheme. No disclosures were reported by the other authors.

The authors apologize for being unable to cite all of the original work that influenced this field due to space constraints. They thank Breast Cancer Now as part of Programme Funding to the Breast Cancer Now Toby Robins Research Centre and Cancer Research UK for funding their work. This work represents independent research supported by the National Institute for Health Research (NIHR) Biomedical Research Centre at The Royal Marsden NHS Foundation Trust and the Institute of Cancer Research, London. The views expressed are those of the author(s) and not necessarily those of the NIHR or the Department of Health and Social Care.