Recent findings indicate that a major mechanism by which poly(ADP-ribose) polymerase (PARP) inhibitors kill cancer cells is by trapping PARP1 and PARP2 to the sites of DNA damage. The PARP enzyme-inhibitor complex “locks” onto damaged DNA and prevents DNA repair, replication, and transcription, leading to cell death. Several clinical-stage PARP inhibitors, including veliparib, rucaparib, olaparib, niraparib, and talazoparib, have been evaluated for their PARP-trapping activity. Although they display similar capacity to inhibit PARP catalytic activity, their relative abilities to trap PARP differ by several orders of magnitude, with the ability to trap PARP closely correlating with each drug’s ability to kill cancer cells. In this article, we review the available data on molecular interactions between these clinical-stage PARP inhibitors and PARP proteins, and discuss how their biologic differences might be explained by the trapping mechanism. We also discuss how to use the PARP-trapping mechanism to guide the development of PARP inhibitors as a new class of cancer therapy, both for single-agent and combination treatments.
Poly(ADP-ribose) polymerase (PARP)-1 and its close relative PARP2 play essential roles in the repair of DNA single-strand breaks (SSBs) (Rouleau et al., 2010; Curtin and Szabo, 2013). PARP1 is an abundant nuclear protein that detects broken DNA strands and binds damaged DNA through its N-terminal zinc-finger domain (Zn), leading to activation of the C-terminal catalytic domain (CAT) that hydrolyzes NAD+ to produce linear and branched poly(ADP-ribose) polymers that extend hundreds of ADP-ribose units (Krishnakumar and Kraus, 2010; Langelier et al., 2011). This post-translational modification, referred to as PARylation, of multiple protein substrates, including PARP1 itself, promotes the recruitment of DNA repair proteins (such as X-ray repair cross-complementing protein 1 and MRE11A) to the site of SSBs, allowing subsequent repair (Rouleau et al., 2010; Curtin, 2012). Auto-PARylation of PARP1 occurs at its regulatory domain; due to the high negative charge of poly(ADP-ribose), extensive auto-PARylation of PARP1 facilitates its dissociation from DNA, which is necessary for the completion of DNA repair (Satoh and Lindahl, 1992; Langelier et al., 2011; Luo and Kraus, 2012). This process is critical for the resealing of DNA SSBs during base excision repair (BER) (Juarez-Salinas et al., 1979; Benjamin and Gill, 1980; Durkacz et al., 1980), as well as repairing topoisomerase (Top) 1 cleavage complexes (Zhang et al., 2011) and DNA double-strand breaks (DSBs) (Audebert et al., 2004; Haince et al., 2008). PARP2 functions in a similar manner to PARP1, but is comparatively less abundant, contributing only 5–10% of the total PARP activity (Ame et al., 1999; Schreiber et al., 2006).
Small-molecule inhibitors of PARP1 and PARP2 are in development as a new class of anticancer agents (Lord and Ashworth, 2012; Curtin and Szabo, 2013). Several selective and potent PARP inhibitors have been reported, all of which are analogs of nicotinamide (Curtin and Szabo, 2013; Ekblad et al., 2013; Lupo and Trusolino, 2014). These inhibitors compete with NAD+ at the catalytic pocket of PARP, inhibiting PARylation (Curtin and Szabo, 2013; Ekblad et al., 2013). PARP inhibitors were originally developed to potentiate the tumor-killing activity of ionizing irradiation and genotoxic agents, such as temozolomide (TMZ) and topotecan. Indeed, all PARP inhibitors tested thus far have demonstrated radio- or chemopotentiation (Curtin and Szabo, 2013), consistent with the key role of PARP1 and PARP2 in DNA damage repair (Yelamos et al., 2011). An additional use for PARP inhibitors was reported by both Bryant et al. (2005) and Farmer et al. (2005), demonstrating that inhibition of PARP1/2 is synthetically lethal with the loss of function of either the BRCA1 or BRCA2 tumor suppressor gene. Loss of BRCA1/BRCA2 gene function is believed to cause a deficiency in homologous recombination (HR)–mediated double-strand DNA break repair, making these cells highly susceptible to the DNA lesions caused by PARP inhibition (Ashworth, 2008; Helleday, 2011). This concept has gained widespread support from both preclinical research and clinical investigations, providing the basis for the first clinical attempt to exploit synthetic lethality in cancer treatment. Subsequent work has shown that, in addition to BRCA1/BRCA2 mutations, deficiency in other HR genes (e.g., Fanconi anemia genes) may also cause cancer cells to become sensitive to PARP inhibitors (McCabe et al., 2006; Peng and Lin, 2011; Turner and Ashworth, 2011).
Until recently, PARP inhibitors were assumed to work solely by inhibiting the catalytic activity of PARP1/PARP2. However, Pommier and colleagues recently demonstrated that catalytic inhibition is not the only mechanism by which PARP inhibitors exert cytotoxic effects (Murai et al., 2012, 2014a). Selected PARP inhibitors may also trap PARP1 and PARP2 on damaged DNA by way of a poisonous allosteric effect (Murai et al., 2012, 2014a). Trapped PARP-DNA complexes prevent DNA replication and transcription, killing cancer cells more effectively than catalytic inhibition. Not all PARP inhibitors have equivalent PARP-trapping activity. In fact, the capacity to trap PARP varies significantly among several clinical-stage PARP inhibitors. In this review, we discuss the evidence for the PARP trapping concept, the chemical structure-activity relationship for PARP trapping, and the clinical implications of PARP trapping.
Evidence for PARP Trapping
Many clinically investigated PARP inhibitors, including talazoparib (BMN 673; BioMarin Pharmaceutical Inc., Novato, CA), niraparib (MK-4827; Tesaro Inc., Waltham, MA), olaparib (AZD-2281; AstraZeneca, London, UK), rucaparib (AG-014699; Clovis Oncology, Boulder, CO), and veliparib (ABT-888; AbbVie Inc., North Chicago, IL), are highly efficacious at inhibiting PARP catalytic activity, with very similar IC50 values in the single-digit nanomolar range (Fig. 1A) (Rouleau et al., 2010; Murai et al., 2012, 2014a; Shen et al., 2013). PARP inhibitors were initially thought to exert their antitumor activity solely by inhibiting the catalytic activity of PARP1 and PARP2. Based on this theory, the inability of PARP to auto-PARylate prevents its dissociation from DNA, resulting in “trapped” PARP-DNA complexes that interfere with the repair of SSBs, and accumulation of unrepaired SSBs leads to DSBs and cell death. This has led to an early version of the PARP trapping theory (Lindahl, 1993; Helleday, 2011; Kedar et al., 2012), which predicted that the cytotoxicity in BRCA-deficient cells and the capacity for chemopotentiation are determined solely by catalytic inhibition, implying that these clinical PARP inhibitors would exhibit similar cytotoxic potency given their comparable PARP-inhibitory activities. However, we and others have shown that the ability of these PARP inhibitors to kill BRCA-deficient cancer cells and to potentiate the cytotoxicity of TMZ differs by several orders of magnitude (Fig. 1A) (Shen et al., 2013). Whereas rucaparib, olaparib, and niraparib have comparable cancer cytotoxicity, talazoparib is approximately 25- to 100-fold more potent in these assays. In contrast, veliparib is ∼1000- to 10,000-fold less potent than talazoparib in cytotoxicity assays. Similar observations were obtained in other laboratories (Murai et al., 2012, 2014a; Stewart et al., 2014). This magnitude of differences in the cytotoxicity cannot be explained simply by differences in catalytic inhibition. These differences in cytotoxicity also cannot be explained by differential off-target activity of the PARP inhibitors, as olaparib and veliparib have similar selectivity profiles against 13 of the 17 human PARP family members (Wahlberg et al., 2012). Talazoparib also shows a similar selectivity profile against the PARP family members (unpublished data). In addition, a recent report demonstrated that PARP1 knockout cells are highly resistant to talazoparib, indicating that the potent cytotoxicity of talazoparib is mediated primarily by PARP1 rather than off-target activities (Murai et al., 2014a).
A revised PARP trapping concept was proposed based on analyses of several lines of experimental data (Murai et al., 2012, 2014a). This new concept suggests that, in addition to catalytic inhibition, some PARP inhibitors may induce an allosteric conformational change in PARP1 and PARP2, thereby stabilizing their associations with DNA. These studies have provided direct evidence that PARP inhibitors exert their cytotoxicity primarily by this trapping mechanism rather than by classic enzyme inhibition. The capacity to trap PARP-DNA complexes varies widely across different PARP inhibitors and is not correlated with their PARP catalytic inhibition. As described below, several lines of evidence support this PARP-DNA–trapping model.
First, if PARP catalytic inhibition were the sole mechanism of PARP inhibitor cytotoxicity, one would expect the antitumor cytotoxicity of an inhibitor to be proportional to its catalytic inhibition. As discussed above, the cytotoxicities of veliparib, rucaparib, olaparib, niraparib, and talazoparib differ by several orders of magnitude, both in terms of single-agent activity in BRCA-deficient cells and in the sensitization of cancer cells to alkylating agents. This strongly suggests that some of the PARP inhibitors may possess additional mechanisms of cytotoxic action in addition to the inhibition of PARP enzymatic activity.
Second, if PARP catalytic inhibition were the sole mechanism of PARP inhibitor cytotoxicity, depletion of PARP would be expected to have equivalent or greater effect on the cytotoxicity, as compared with the PARP inhibition by small-molecule inhibitors. However, treating cancer cells that harbor wild-type PARP gene with PARP inhibitors, such as talazoparib and olaparib, induces far superior cytotoxicity than genetic depletion of PARP (Murai et al., 2012, 2014a). Similarly, knockdown of PARP gene expression by small interfering RNA (siRNA) is far less effective than PARP inhibitor treatment in inducing apoptosis in BRCA-deficient cells (Bryant et al., 2005; Farmer et al., 2005; Strom et al., 2011). Consistent with this cytotoxicity data, SSB repair is delayed to a greater extent by small-molecule inhibitors, as compared with PARP gene knockdown by siRNA (Strom et al., 2011), with the level of DSBs considerably higher in olaparib-treated wild-type cells than in PARP gene knockout cells (Murai et al., 2012).
Third, there is now mounting evidence demonstrating that PARP itself is required for the cytotoxic effects of the PARP inhibitors. Depletion of PARP renders cells resistant to both talazoparib and olaparib (Murai et al., 2012, 2014a). Likewise, cells with PARP1 loss-of-function mutations are 100-fold more resistant to olaparib and talazoparib than wild-type cells, and restoring PARP1 expression causes these cells to revert to wild-type sensitivity (Pettitt et al., 2013). These data indicate that PARP1 protein is the critical mediator of the cytotoxic effect of PARP inhibitors, arguing against the suggestion that these differences in cytotoxicity result from off-target activities. PARP is also required for sensitizing tumor cells to the alkylating agents TMZ and methyl methanesulfonate (Liu et al., 2009; Murai et al., 2012, 2014a). Other data have also confirmed that wild-type cells are more sensitive to PARP inhibitor cytotoxicity than PARP1−/− mouse cells (Horton et al., 2005; Kedar et al., 2012). The effects of PARP1 ablation are not cell type specific, as it was observed in chicken cells (Murai et al., 2012, 2014a), mouse cells (Horton et al., 2005; Kedar et al., 2012; Pettitt et al., 2013), and human cancer cells (Liu et al., 2009; Murai et al., 2012, 2014a; Pettitt et al., 2013; Das et al., 2014). Although the PARP1 loss-of-function mutant cell lines also showed resistance to veliparib, this effect was observed only at very high drug concentrations, consistent with the relatively weak trapping effects of veliparib (Murai et al., 2012).
Finally, PARP-DNA complexes have been demonstrated both in intact cells and in cell-free systems (Kedar et al., 2012; Murai et al., 2012, 2014a). The presence of PARP inhibitors significantly stabilizes the association of PARP with DNA in a reversible manner. Importantly, the potency to trap PARP-DNA complexes varies considerably among different PARP inhibitors. Of the five PARP inhibitors tested, veliparib is primarily a catalytic inhibitor with little trapping activity, while olaparib, niraparib, and rucaparib trap PARP ∼100-fold more efficiently than veliparib. Talazoparib is the most potent PARP trapper studied to date, promoting PARP-DNA complexes ∼100-fold more efficiently than olaparib, niraparib, and rucaparib (Murai et al., 2012, 2014a). These data suggest that the anticancer cytotoxicities of PARP inhibitors do not have a linear relationship with catalytic inhibition. In contrast, the cytotoxicities correlate very well with trapping potency.
In summary, current data support a model in which PARP inhibitors use a dual mechanism to induce cell killing: one by directly inhibiting PARP catalytic activity and the other by allosterically affecting the conformational flexibility and dynamics of PARP to enhance its affinity for the SSB repair intermediate. This second mechanism is often referred to as PARP trapping or PARP “poisoning.” Trapped PARP-DNA complexes promote cytotoxicity to a much greater degree than the unrepaired DNA SSBs that are caused by the loss of PARP activity, with the PARP inhibitors varying considerably in their trapping capacities. The clinical efficacy data for PARP inhibitors in cancer patients with deleterious BRCA gene mutations are consistent with the in vitro cytotoxicity levels of the PARP inhibitors, nicely correlating the clinical activity with the ability to trap PARP. For completeness of this discussion, we note that an alternative conclusion has been presented at meetings, but not yet published (Hopkins et al., 2014; Solomon et al., 2014). These investigators argue that the inhibition of catalytic activity alone is responsible for the observed PARP trapping. Further work will be required to resolve the differences in methodology and interpretations that are raised by this work. In particular, it remains to be established whether the alternative mechanism can explain both the requirement of PARP for the cytotoxic effects of the PARP inhibitors and the large differences in cytotoxicity among the PARP inhibitors with similar catalytic inhibition activities. At this point, cocrystal structures that demonstrate allosteric change of the PARP protein in the presence of the inhibitors are not available. As described in the following section, intriguing hypotheses based on the current knowledge of the binding mechanism between PARP and PARP inhibitors may be postulated.
Current View of PARP-Trapping Structure-Activity Relationship
Comparative analyses of the inhibitor-binding modes may provide insight into potential structural attributes for the differential levels of PARP-trapping activity reported for the selected set of inhibitors (Fig. 1). Although other structural aspects of the PARP superfamily proteins, including inhibitor selectivity, have been extensively studied and thoroughly reviewed elsewhere (Ferraris, 2010; Ekblad et al., 2013; Papeo et al., 2013; Steffen et al., 2013), the scope of this review will be limited to the structural data on PARP1, the most-studied DNA-dependent enzyme of the PARP superfamily, focusing on those inhibitors for which PARP-trapping activity has been evaluated or their closely related analogs for which cocrystal structures are publicly available for analyses (Fig. 1).
Inhibitor Binding in PARP1 CAT Domain
The CAT domain of PARP1 (Fig. 2A), containing the helical regulatory subdomain (HD) and the ADP-ribosyltransferase subdomain (ARTD), has been structurally well characterized, offering considerable information about ligand-binding interactions. In particular, the highly conserved NAD+-binding (“donor”) site in the ARTD (Fig. 2A) has been exploited extensively in the development of PARP inhibitors, including those in late-stage clinical development (Fig. 1) (Ferraris, 2010; Ekblad et al., 2013; Papeo et al., 2013; Steffen et al., 2013). Within the relatively large NAD+-binding site, these inhibitors exhibit a range of diverse binding modes targeting distinct binding regions (Fig. 2B).
All known PARP1 inhibitors (Fig. 1) are anchored in the nicotinamide pocket within the NAD+-binding site (Fig. 2). The core carboxamide moiety, shared among these PARP inhibitors, forms critical hydrogen bonds with conserved Gly863 and Ser904 residues (Fig. 2B) (Ferraris, 2010; Ekblad et al., 2013; Papeo et al., 2013; Steffen et al., 2013). These inhibitors also contain an aryl (and sometimes a bicyclic) ring system forming π-stacking interactions with conserved Tyr896 and Tyr907 residues (Fig. 2B). Other conserved residues bordering the nicotinamide pocket, such as pocket-forming Ala898 and Lys903, and a predicted catalytic residue (Glu988) (Ruf et al., 1998) have also been exploited in the design of inhibitors (Fig. 2B) (Ferraris, 2010; Ekblad et al., 2013; Papeo et al., 2013; Steffen et al., 2013). Veliparib, the smallest PARP inhibitor in clinical development, binds within the NAD+-binding site primarily through critical interactions with or near the nicotinamide-binding residues. The carboxamide benzimidazole core maintains canonical interactions with conserved Gly863, Ser904, and Tyr907 residues in the nicotinamide pocket (Fig. 2B) (PDB ID: 2RD6) (Penning et al., 2009), while the benzimidazole nitrogen forms a water-mediated hydrogen bond with Glu988 (Ruf et al., 1998). The comparable binding mode (PDB ID: 3GN7) and enzymatic potency reported for its enantiomer (Penning et al., 2009) suggest that interactions of the carboxamide-benzimidazole scaffold at the nicotinamide subsite (Fig. 2) play a significant role in overall binding potency.
Located immediately adjacent to the nicotinamide pocket of PARP is a large hydrophobic pocket (known as the adenine-ribose–binding site), representing another potential druggable subsite (Fig. 2A). Many known PARP1 inhibitors leverage the presence of this adenine-ribose subsite by incorporating a large hydrophobic group onto the core nicotinamide pharmacophore (Ferraris, 2010; Ekblad et al., 2013; Papeo et al., 2013; Steffen et al., 2013). An early preclinical PARP1 inhibitor, FR257517 (IC50, 14 nM) (Fig. 1B), binds to the nicotinamide subsite via a quinazolinone moiety, while the remaining part of the inhibitor (including the dihydropyridinyl and fluorophenyl moieties) reaches into the adenine-ribose–binding pocket, creating a new hydrophobic subsite with Leu769, Arg878, Ile879, and Pro881 residues (Fig. 2B) (PDB ID: 1UK0) (Kinoshita et al., 2004). Several other preclinical inhibitors, for which structural data are available, also exhibit a similar binding mode that involves secondary contacts in the adenine-ribose–binding subsite (Hattori et al., 2004; Kinoshita et al., 2004; Miyashiro et al., 2009; Gangloff et al., 2013; Ye et al., 2013; Patel et al., 2014). From the chemical scaffold and binding mode reported in the tankyrase 2 cocrystal structure (Narwal et al., 2012), olaparib, while bound to the nicotinamide pocket via a phthalazinone core, also likely has both diacylpiperazine and cyclopropyl moieties reaching deeply into the adenine-ribose–binding pocket.
A phosphate-binding region formed by HD residues at the NAD+ site (Fig. 2A) (Ruf et al., 1998) has also been explored for novel inhibitor-scaffold designs. According to a publicly available cocrystal structure (PDB ID: 3L3M), the compound referred to as A927929 (IC50, 6 nM) (Fig. 1B) binds the CAT domain by interacting with the Asp766 residue, which is predicted to be involved in phosphate binding, in addition to key interactions with nicotinamide-binding Gly863 and Ser904 residues (Fig. 2B) (Penning et al., 2010). Recent studies also report crystal structure data (PDB ID: 2OQA; compound 59) that describe a similar inhibitor-binding pose directed toward the phosphate-binding region (Patel et al., 2014). Based on the chemical scaffolds in the absence of PARP1 cocrystallographic structure data, rucaparib and niraparib may also be directed toward the phosphate-binding residues within the NAD+ site. Compound A906894 (Fig. 1B) also projects toward the phosphate-binding residues (PDB ID: 3L3L). However, the relatively small size of this compound may limit a direct contact with the HD phosphate-binding residues (Gandhi et al., 2010). Instead, this inhibitor, via its piperidine nitrogen, forms a hydrogen bond with Gly888 located on a donor-site loop (D-loop: Gly876–Gly894) that, together with the N-terminal HD, borders the NAD+-binding site (Fig. 2A).
Outer Borders of the NAD+ Site.
The majority of known PARP1 inhibitors target the nicotinamide pocket and one of the two other (adenine-ribose or phosphate) subsites within the relatively large NAD+ cosubstrate-binding site (Fig. 2, A and B) (Wiholm and Myrhed, 1993; Ekblad et al., 2013; Papeo et al., 2013; Steffen et al., 2013). However, based on recent cocrystal structural data, talazoparib (PDB ID: 4PJT) (Fig. 2B) adds another dimension to the already diverse range of inhibitor-binding modes (Aoyagi-Scharber et al., 2014). Talazoparib represents a new tripartite chemical scaffold consisting of a core tricyclic group with stereospecific disubstituents (Fig. 1A) (Shen et al., 2013). While the cyclic amide moiety is tethered to the nicotinamide subsite and the unsaturated six-membered ring system forms a water-mediated interaction with the catalytic residue, Glu988 (Ruf et al., 1998), the 8S-(4-fluorophenyl) and 9R-methyltriazole moieties are directed toward the outer edges of the NAD+-binding pocket (Fig. 2B) (Aoyagi-Scharber et al., 2014). The first substituent, the 4-fluorophenyl group, reaches out to the unique ligand-binding space near the D-loop (Gly876–Gly894), which, as mentioned above, forms the lid of the NAD+-binding site (Fig. 2B). Two D-loop residues, Tyr889 and Met890, interact with this fluorophenyl group by forming π-stacking and water-mediated hydrogen-π/hydrogen-bonding interactions, respectively (Fig. 2B). The second substituent, the methyltriazole group, is involved in a water-mediated hydrogen-bonding interaction to the backbone amide of the highly conserved aromatic Tyr896 residue within the nicotinamide pocket (Fig. 2B). The methyl group of the triazole moiety, when bound to PARP1, points toward the HD, which not only defines the size of the NAD+-binding site within the CAT domain, but also has critical regulatory roles in the catalytic function of PARP1.
Structural Basis for PARP-DNA Trapping
DNA-Dependent PARP1 Interdomain Communication.
The distinct binding modes in the CAT domain observed among the PARP1 inhibitors (Fig. 2B) combined with newly available structural data on noncatalytic domains (Langelier et al., 2008, 2011, 2012; Loeffler et al., 2011; Ali et al., 2012; Hassler and Ladurner, 2012) may suggest key structural insights into the DNA-trapping activity of PARP inhibitors. The crystal structure of the “near” full-length PARP1 enzyme containing 1) zinc-finger domains Zn1 and Zn3, 2) WGR domain named after a conserved Trp-Gly-Arg sequence, and 3) CAT domain consisting of HD-ARTD subdomains bound to a DNA DSB (PDB ID: 4DQY) demonstrates that DNA binding by Zn1, Zn3, and WGR domains destabilizes the CAT domain, thereby activating the PARP enzymatic activity (Langelier et al., 2012). Specifically, structural changes upon DNA binding in the noncatalytic domains lead to displacements of Leu698 and Leu701 residues from the hydrophobic core of the HD (Fig. 3A), decreasing the stability of ARTD in the CAT domain; an increase in the protein dynamics of the ARTD is suggested to stimulate PARP catalysis (Langelier et al., 2012). The proposed mechanism is consistent with a previously identified gain-of-function mutation at Leu713 within the HD hydrophobic core that leads to DNA-independent PARP1 activity (Fig. 3A) (Miranda et al., 1995). The importance of interdomain communication in DNA-dependent activation is also demonstrated by Trp318 mutations at the HD-WGR-Zn3 domain interfaces, resulting in a catalytically inactive PARP1 without affecting DNA binding (Steffen et al., 2014). Notably, the HD subdomain, implicated as a focal point for interdomain communication, is an important CAT domain structural element defining the size of the inhibitor-binding pocket, and sometimes directly interacting with bound inhibitors (Iwashita et al., 2005; Penning et al., 2010).
Inhibitor-Dependent PARP1 Interdomain Communication.
An improved understanding of the structural basis for DNA-dependent PARP1 activation (Hassler and Ladurner, 2012; Langelier and Pascal, 2013) has led to a new hypothesis that inhibitor-induced reverse-allosteric signaling may drive PARP-trapping activity (Murai et al., 2012). Using a fluorescence anisotropy-based DNA-binding assay, 10- to 40-fold differences in the biochemical trapping potency have been reported among the PARP inhibitors with comparable potencies in inhibiting catalysis, further suggesting possible inhibitor-specific allosteric effects (Murai et al., 2012, 2014a). The differential binding modes among the inhibitors may explain why those equally potent PARP1 catalytic inhibitors exhibit significant differences in their abilities to trap PARP on DNA (Murai et al., 2012, 2014a). As discussed earlier and further elaborated on in the following section, a substantial body of structural data indicates that PARP inhibitors differ considerably in overall size/scaffold (Fig. 1) and consequently in binding modes (Fig. 2B).
The size or bulk of a PARP inhibitor may be correlated with the level of reverse-allosteric signaling, which may be related to the degree of PARP trapping (Marchand et al., 2014). Among the five clinical inhibitors evaluated for PARP-trapping ability (Fig. 1A), the smallest inhibitor, veliparib (mol. wt., 244), was the least potent in trapping PARP relative to the bulkier PARP inhibitors, such as olaparib (mol. wt., 434), niraparib (mol. wt., 320), rucaparib (mol. wt., 323), and talazoparib (mol. wt., 380) (Murai et al., 2012, 2014a). The low trapping activity of veliparib correlated with the absence of inhibitor-induced PARP1 conformational changes during microsecond molecular dynamic simulations (Marchand et al., 2014). Under the same experimental conditions, the binding of comparatively bulkier olaparib to the CAT domain caused steric bumps on the HD subdomain, triggering conformational rearrangements in the DNA-binding domains to stabilize the PARP1-DNA complex (Marchand et al., 2014). A lack of inhibitor-induced steric bumps in the HD subdomain by small inhibitors can be explained by the limited extent of the binding interactions on the CAT domain. The methylpyrrolidine moiety of veliparib appears to be too small to make significant contacts with the rest of the NAD+ site (Fig. 2B) (Penning et al., 2009), including the HD subdomain implicated as a crucial structural element in the proposed reverse-allosteric mechanism (Marchand et al., 2014). HYDAMTIQ (mol. wt., 274) (Fig. 1B), another potent inhibitor similar to veliparib in size and binding mode, also shows low levels of PARP trapping and cytotoxicity, further supporting a potential relationship between the molecular size of the inhibitors and the degree of trapping activity (Pellicciari et al., 2011; Marchand et al., 2014).
The molecular size of the inhibitor alone, however, seems insufficient to describe the differential trapping of PARP1 reported for PARP inhibitors. Among the bulky PARP1 inhibitors, no correlation can be firmly established between inhibitor size and degree of PARP trapping. For instance, talazoparib has been shown to be the most potent PARP-trapping inhibitor tested to date, despite its small size relative to olaparib (Fig. 1A) (Murai et al., 2014a). Recent studies suggest that the shape and flexibility of inhibitors may contribute to the efficiency of trapping PARP-DNA complexes (Murai et al., 2014a). Compared with other bulky yet flexible inhibitors, talazoparib has unique tripartite chemical scaffolds that are rigid and stereospecific (Fig. 1A) (Shen et al., 2013; Aoyagi-Scharber et al., 2014). The shape and flexibility of the inhibitor dictate the extent and type of binding interactions, which vary significantly among these clinical inhibitors and may explain marked differences in PARP-trapping activities.
Of the five clinical-stage PARP inhibitors tested (Fig. 1), those inhibitors with noticeable ability to stabilize the PARP-DNA complexes appear to interact with the D-loop, though to differing extents, at the outer border of the NAD+ site (Fig. 2, A and B). Using a structural analogy with A927929 (Penning et al., 2010), the potent PARP trappers rucaparib and niraparib (Fig. 1A) are likely bound close to the D-loop (Fig. 2B). Interestingly, a recent molecular dynamics simulation study reported an alternative bound conformation of olaparib, another PARP trapper, forming a hydrogen bond with the D-loop residue Met890 (Marchand et al., 2014). Talazoparib, the most potent PARP trapper evaluated to date, has a unique stereospecific 4-fluorophenyl substituent favorably oriented to interact with the Met890 backbone via a water molecule and the Tyr889 side chain on the D-loop (Figs. 2B and 3B). The degree of PARP-trapping capacity may be associated with the extent to which the inhibitor interacts with the D-loop residues.
The D-loop is structurally linked to the HD subdomain (Fig. 3B), suggesting that the inhibitor interactions with the D-loop residues may influence conformational flexibility of the CAT domain. The PARP1 D-loop has been described as a relatively rigid structure, unlike the flexible counterpart in other PARP superfamily proteins (Wahlberg et al., 2012). However, recently determined PARP1 inhibitor cocrystal structures (PDB IDs: 4HHY, 4HHZ, and 4GV7) (Lindgren et al., 2013; Ye et al., 2013) reveal that the Tyr889 side chain on the D-loop can adopt multiple rotameric positions among noncrystallographic symmetry-related molecules (Fig. 3C). When bound to talazoparib, these alternative side-chain conformations would not be permitted due to steric constraints imposed by the 4-fluorophenyl substituent (Figs. 2B and 3C). Additional studies are required to elucidate the structural consequences of the decreased side-chain flexibility of the D-loop residues, such as Tyr889, on the conformational stability of the HD or entire CAT domain and subsequent interdomain communication that has been proposed to be critical for the allosteric signaling of PARP1 (Langelier et al., 2012).
Newly available structural data on noncatalytic domains (Hassler and Ladurner, 2012; Langelier and Pascal, 2013), combined with a wealth of cocrystal structure data available for the inhibitor-bound PARP1 CAT domain (Ferraris, 2010; Ekblad et al., 2013; Papeo et al., 2013; Steffen et al., 2013), signify an important first step toward elucidating the molecular basis for the differential PARP trapping among the inhibitors (Murai et al., 2012, 2014a). The D-loop, known for its involvement in recognizing the acceptor adenine moiety (Ruf et al., 1998) and well characterized as a sequence-diverse element in the PARP superfamily (Ekblad et al., 2013; Papeo et al., 2013; Steffen et al., 2013), may also contribute to the protein dynamics of the CAT domain that are implicated in the putative inhibitor-induced reverse-allosteric signaling. Further investigations of the interdomain interactions and allostery of the full-length or near full-length enzyme using diverse biophysical techniques, such as small-angle X-ray scattering, molecular dynamics simulations, and X-ray crystallography (Mansoorabadi et al., 2014; Marchand et al., 2014; Patel et al., 2014), may increase our understanding of the PARP-trapping mechanisms, providing new avenues to develop improved anticancer therapeutic agents and treatment approaches.
Understanding the mechanisms of action of PARP inhibitors may assist physicians in selecting the most appropriate drug for treating cancer. The effectiveness in PARP trapping may be an important factor to consider when formulating a therapeutic regimen involving a PARP inhibitor (single-agent or in combination). The PARP trapping concept may also shed new light on how to identify patients who will (or will not) be likely to respond to PARP inhibitor therapies.
Lessons from Topoisomerase Inhibitors.
Top inhibitors may provide useful insights into the development of PARP inhibitors as anticancer agents. Inhibitors of Top1 [e.g., camptothecin (CPT), topotecan, and irinotecan] and Top2 (e.g., doxorubicin and etoposide) are among the most effective and commonly used anticancer drugs. These drugs kill cancer cells by trapping their respective protein targets (Top1 or Top2) on DNA rather than by classic enzymatic inhibition (Pommier et al., 2010; Pommier, 2013). Depletion of Top1 causes cells to become resistant to CPT (Eng et al., 1988; Nitiss and Wang, 1988; Pommier et al., 1999; Miao et al., 2007), and reductions in Top1 and Top2 levels in tumors diminishes the effects of the Top inhibitors (Rasheed and Rubin, 2003; Beretta et al., 2006; Burgess et al., 2008), highlighting the Top requirements in the cytotoxic action of these drugs. X-ray crystallographic studies have revealed the ternary complexes of Top1-DNA-CPT (PDB ID: 1T8I) and Top2-DNA-etoposide (PDB ID: 3QX3), in which the drugs simultaneously interact with both the DNA and the enzymes (Redinbo et al., 1998; Ioanoviciu et al., 2005; Wendorff et al., 2012), providing direct evidence for the Top-trapping model (Pommier and Marchand, 2011). Biochemically, the formation of Top1 cleavage complex and Top2 cleavage complex has been demonstrated in cells treated with Top1 and Top2 inhibitors, respectively. Thus, Top inhibitors trap Tops to DNA much like PARP inhibitors trap PARP1/2 (Pommier et al., 1999; Pommier, 2013). Despite the analogous interactions between the Top inhibitors and Top1/2 and between the PARP inhibitors and PARP1/2, there are important differences. Top inhibitors directly target the DNA-protein interface and form covalent bonds with Top1 or Top2, whereas PARP inhibitors bind noncovalently with PARP1/2 at the catalytic active site away from the associated DNA (Patel et al., 2012).
Because the cytotoxicity of Top inhibitors appears to be positively correlated with the expression levels and activities of Top1 and Top2 (Burgess et al., 2008; O'Malley et al., 2009; Pfister et al., 2009), assays are being developed to measure the expression levels and activities of Top1 and Top2 (Pfister et al., 2009, 2012). However, as noted in a review by Pommier (2013), “there is no simple linear correlation between topoisomerase levels and drug response.” Other factors that influence the sensitivity to Top inhibitors, such as the protein kinase TAK1 and SLFN11, have been identified via two main approaches: 1) identification of alterations in the DNA repair and stress response of cells to Top inhibitors, and 2) genomic analyses of cancer cells (e.g., siRNA screens and gene expression correlations) (Pommier, 2013). Similar approaches may help identify biomarkers associated with the response to PARP inhibitors.
Can PARP Expression Level and PARP Activity Serve as Biomarkers for Sensitivity to PARP Inhibitors?
HR deficiency has emerged as the key determinant of the cancer cell response to PARP inhibitors; other factors affecting the sensitivity to PARP inhibitors (e.g., P-glycoprotein overexpression and loss of 53BP1) have also been described (Lord and Ashworth, 2012; Bouwman and Jonkers, 2014). In addition to these mechanisms, an intriguing question that derives from the PARP trapping concept is whether the tumor response to the PARP inhibitors may be associated with the PARP expression level and/or PARP activity in the cells. Though inconclusive at present, several lines of evidence appear to be consistent with this hypothesis, as discussed below.
A study by Pettitt et al. (2013) has shown that either complete loss or reduction of PARP1 expression is associated with acquired resistance to PARP inhibitors. Cells with complete loss of PARP1 expression are ∼100-fold more resistant to olaparib and talazoparib than the parental cell line, while reduction of PARP1 expression by siRNA treatment is associated with partial resistances to the PARP inhibitors (Pettitt et al., 2013). Similarly, Singh et al. (2013) reported that a common characteristic in several Ewing’s sarcoma cell lines that have developed resistance to talazoparib (500- to 1000-fold-higher IC50 as compared with the parental cell lines) is the loss or reduction of PARP1 expression. In addition, human cancer cell lines that have acquired resistance to a combined treatment of veliparib plus TMZ also show greatly reduced PARP1 levels (Liu et al., 2009). Michels and colleagues analyzed several cisplatin-resistant cancer cell lines and found that a subset of these lines overexpress PARP1 and display hyperactivity of PARylation. Furthermore, a positive correlation was observed between the cellular PARylation and the sensitivity to PARP inhibitors (Michels et al., 2013). It should be noted that clinical experience thus far has shown that platinum-sensitive tumors are more likely to respond to PARP inhibitors than are platinum-resistant or platinum-refractory diseases (Fong et al., 2010), so the clinical prospect of this observation remains to be evaluated. Nonetheless, this study lends additional support to the idea that cellular PARylation activity may be an important factor that influences the sensitivity to PARP inhibitors.
HR-defective cells may manifest hyper-PARylation activity. Disruption of BRCA2 and other HR genes leads to PARP hyperactivation, which correlates with an increased sensitivity to PARP inhibitors (Gottipati et al., 2010). Restoring BRCA2 activity (either by genetic reversion of the BRCA2 gene or by stable expression of a wild-type BRCA2 gene) led to reduced PARP activity and corresponding resistance to PARP inhibitors (Gottipati et al., 2010). It appears that cells with DSB repair deficiency may become highly dependent on PARP-mediated DNA repair, raising the possibility that PARylation activity may serve as a surrogate biomarker for DSB repair deficiency and a relevant predictor of response to PARP inhibitors.
It should be emphasized that PARP expression level/PARP activity is unlikely to be the sole determinant of the tumor response to PARP inhibitors. To date, sensitivity to single-agent PARP inhibitors in the clinical setting appears to be closely tied to HR defects, whereas PARP expression/activity may just be one of the additional factors that influence tumor response. As an example of the complexity of relating PARP levels to sensitivity to PARP inhibitors, a recent report shows a relationship between increased PARP expression in HCC1937-derived stem cells and resistance to olaparib (Gilabert et al., 2014). However, given that the HCC1937 cell line is highly resistant to olaparib despite its BRCA1 mutation status (Lehmann et al., 2011) the interpretation of this result is complicated. In any case, testing the possible link between PARP expression/activity and sensitivity to PARP inhibitors in the clinical setting is warranted.
Designing Drug Combination Strategies for PARP Inhibitors with Genotoxic Agents.
Combinations of PARP inhibitors with DNA-damaging agents are undergoing active clinical investigations (Curtin and Szabo, 2013; Murai et al., 2014b). As PARP inhibitors are not equal in their ability to trap PARP, careful selection of the appropriate PARP inhibitor(s) for each combination therapy is critical for maximizing the chance of clinical success. The choice of PARP inhibitor should be based not only on whether PARP is involved in the repair of the DNA lesions produced by the genotoxic agent, but also on whether the lesions trap PARP-DNA complexes. According to a recent publication (Murai et al., 2014b), three categories of combination strategies for use with PARP inhibitors and genotoxic agents are envisioned.
In the first category, highly synergistic combinations taking advantage of PARP trapping are considered (Murai et al., 2014b). This is exemplified by the combination of PARP inhibitors with the alkylating agent TMZ. TMZ induces two types of DNA lesions: 1) O6-methylguanine adducts, which cause DNA DSB and apoptosis—cells with defective mismatch repair pathway or cells with functional O6-methylguanine methyltransferase are resistant to this mechanism; and 2) N7-methylguanine (N7mG) and N3-methyladenine (N3mA) DNA adducts, which account for ∼90% of all methylation events caused by TMZ but are nonlethal in most normal cells and tumor cells, because they are readily repaired by BER (Sarkaria et al., 2008; Zhang et al., 2012). The repair of N7mG and N3mA adducts by BER requires PARP to recognize and bind the repair intermediate 5′-deoxyribose phosphate (Cistulli et al., 2004; Horton and Wilson, 2013a,b; Horton et al., 2014). In fact, 5′-deoxyribose phosphate is a preferred substrate for PARP trapping (Murai et al., 2014b; Smith et al., 2015). Thus, PARP inhibitors turn the nonlethal N7mG and N3mA adducts into cytotoxic lesions by trapping PARP at 5′-deoxyribose phosphate. This ability to potentiate TMZ cytotoxicity varies greatly among the various PARP inhibitors and aligns well with the PARP-trapping capacity of these molecules (Shen et al., 2013; Murai et al., 2014b). Optimal potentiation of TMZ by olaparib requires PARP. In PARP-depleted cells, olaparib and veliparib showed similar (weak) potentiation of TMZ (Murai et al., 2014b). These data indicate that combining TMZ with a potent PARP-trapping agent is a more rational approach than combining TMZ with a PARP inhibitor that primarily acts by catalytic inhibition.
The potent combination effects with PARP inhibitors and TMZ have been demonstrated in numerous studies [for examples, see Delaney et al. (2000) and Calabrese et al. (2004)], even before the PARP-trapping mechanism was elucidated. In fact, the first clinical trial of a PARP inhibitor was a combination study of rucaparib and TMZ (Plummer et al., 2013). An improved understanding of the PARP-trapping mechanism leads to completely novel combination strategies; rather than PARP inhibitors acting as the potentiator of the TMZ cytotoxicity, PARP inhibitors should be considered to be the cytotoxic agents by inducing PARP-DNA–trapping complexes, and TMZ as a potentiator that induces DNA lesions. As the creation of the N7mG and N3mA adducts is efficient (Sarkaria et al., 2008), TMZ may be effective at reduced dosages (to minimize the toxicity), allowing for maximal antitumor efficacy by dosing PARP inhibitors at levels approaching their maximum tolerated doses. We envision that distinctive mechanisms of sensitivity/resistance will exist for the combination of PARP inhibitors plus TMZ, as compared with TMZ immunotherapies. The traditional mechanisms of TMZ resistance related to O6-methylguanine adducts may not be relevant, as cancer cells expressing O6-methylguanine methyltransferase and those having defective mismatch repair maintain their sensitivity to PARP inhibitors plus TMZ. It should be noted that the concept at the present time is applicable only to combinations of PARP inhibitors with TMZ (and other DNA alkylating agents). Whether this combination strategy can apply to other DNA-damaging agents has yet to be evaluated. This novel combination approach with DNA alkylating agents is currently being tested both preclinically (Feng et al., 2014; von Euw et al., 2014; Smith et al., 2015) and in the clinic (for examples, see NCT02116777 and NCT02049593 at www.clinicaltrials.gov).
In the second category, synergistic combinations that do not use PARP trapping but require PARP catalytic inhibition are considered (Murai et al., 2014b). Combination with the Top1 inhibitor CPT is an example of this category. Olaparib and veliparib sensitize tumor cells to CPT to a similar degree, and PARP depletion itself enhances the cytotoxic effect of CPT, which is not further augmented by olaparib or veliparib (Murai et al., 2014b). These results suggest that the synergistic effect of CPT with PARP inhibitors is due to PARP catalytic inhibition rather than PARP trapping (Murai et al., 2014b). This synergy is consistent with recent findings demonstrating that PARP is coupled with tyrosyl-DNA phosphodiesterase-1 (Das et al., 2014), and with X-ray repair cross-complementing protein 1, polynucleotide kinase, and ligase III, which repairs the DNA lesions caused by Top1 inhibitors (Pommier et al., 2006; Das et al., 2014). Based on this mechanism, PARP inhibitors may preferentially potentiate Top1 inhibitors either in tyrosyl-DNA phosphodiesterase-1–deficient cells or in cells harboring ERCC1 mutations (Pommier et al., 2010; Zhang et al., 2011; Pommier, 2013).
In the third category, combinations of additive but mutually independent mechanisms are considered (Murai et al., 2014b). Examples for this category include Top2 inhibitors (e.g., etoposide) and cross-linking agents (e.g., cisplatin). The rationale for combining PARP inhibitors with these agents is not based on either PARP-DNA trapping or PARP catalytic inhibition. Therefore, additive as opposed to synergistic effects may be expected when PARP inhibitors are combined with these agents. Cisplatin and etoposide cause DNA lesions whose repair does not involve PARP-mediated pathways (Bowman et al., 2001; Zhang et al., 2011; Pommier, 2013), whereas PARP inhibitors kill cells by acting on DNA lesions that spontaneously accumulate during the cell cycle. In this scenario, two independent cell-killing mechanisms act in parallel.
There are likely more examples of genotoxic agents that use these three mechanisms. In addition, PARP inhibitors may also be effective in combination with other agents that interfere with DNA repair, such as CHK1/2 inhibitors and ATM inhibitors. From a mechanistic point of view, these combinations may likely have synergistic effects (Bryant and Helleday, 2006; Rehman et al., 2012; Tang et al., 2012; Booth et al., 2013; Węsierska-Gądek and Heinzl, 2014), but whether they use the trapping property of a PARP inhibitor has yet to be examined. Finally, other targeted cancer therapeutic agents, such as histone deacetylase inhibitors and phosphoinositide 3-kinase inhibitors, may be considered for combination with PARP inhibitors to achieve improved anticancer activity. The mechanisms of these combinations are beyond the scope of this review, but many interesting reports are available (Johnson et al., 2011; Nowsheen et al., 2012; Cardnell et al., 2013; Ha et al., 2014).
In summary, trapping of PARP on DNA has emerged as a major mechanism by which PARP inhibitors exert antitumor activity. The ability to induce PARP trapping varies considerably among the various PARP inhibitors. Their differential PARP-trapping activities may be attributed to the differences in their chemical scaffolds and hence binding modes in the CAT domain. The distinct binding modes among these inhibitors may serve as a structural basis for their varying abilities to modulate interdomain communication and allosteric regulation, which have been postulated to underpin PARP trapping. Recent advances in our understanding of DNA-induced PARP1 catalytic activation provide initial structural insight into potential molecular mechanisms for inhibitor-induced DNA trapping. Understanding the molecular mechanism of PARP trapping will assist in developing existing and new PARP inhibitors, as well as identifying optimal clinical applications of various PARP inhibitors, allowing the rational selection of appropriate combination therapies.
The authors thank Dr. L. E. Post (BioMarin Pharmaceutical Inc.) for guiding this review and providing valuable suggestions and to Drs. Y. Pommier (National Cancer Institute, Bethesda, MD) and J. Murai (Kyoto University, Kyoto, Japan) for reviewing the manuscript and sharing their expert insights. The authors thank E. Johnson for editing the manuscript and P. Fitzpatrick, G. Vehar, R. Srivastava, K. Krishnan, M. Henderson, E. Wang, and K. Yu (BioMarin Pharmaceutical Inc.) for their suggestions, all of which markedly improved the manuscript. The authors also appreciate the professional and competent assistance from ApotheCom in the preparation of the manuscript.
Performed data analysis: Shen, Aoyagi-Scharber, Wang.
Wrote or contributed to the writing of the manuscript: Shen, Aoyagi-Scharber, Wang.
- Received December 29, 2014.
- Accepted March 9, 2015.
- ADP-ribosyltransferase subdomain
- base excision repair
- C-terminal catalytic domain
- donor-site loop
- double-strand break
- helical regulatory subdomain
- homologous recombination
- poly(ADP-ribose) polymerase
- small interfering RNA
- single-strand break
- Zinc-finger domain
- Copyright © 2015 by The American Society for Pharmacology and Experimental Therapeutics