Studies Towards Hypoxia-Activated Prodrugs of PARP Inhibitors

Poly(ADP-ribose)polymerase (PARP) inhibitors (PARPi) have recently been approved for the treatment of breast and ovarian tumors with defects in homologous recombination repair (HRR). Although it has been demonstrated that PARPi also sensitize HRR competent tumors to cytotoxic chemotherapies or radiotherapy, normal cell toxicity has remained an obstacle to their use in this context. Hypoxia-activated prodrugs (HAPs) provide a means to limit exposure of normal cells to active drug, thus adding a layer of tumor selectivity. We have investigated potential HAPs of model PARPi in which we attach a bioreducible “trigger” to the amide nitrogen, thereby blocking key binding interactions. A representative example showed promise in abrogating PARPi enzymatic activity in a biochemical assay, with a ca. 160-fold higher potency of benzyl phthalazinone 4 than the corresponding model HAP 5, but these N-alkylated compounds did not release the PARPi upon one-electron reduction by radiolysis. Therefore, we extended our investigation to include NU1025, a PARPi that contains a phenol distal to the core binding motif. The resulting 2-nitroimidazolyl ether provided modest abrogation of PARPi activity with a ca. seven-fold decrease in potency, but released the PARPi efficiently upon reduction. This investigation of potential prodrug approaches for PARPi has identified a useful prodrug strategy for future exploration.


Introduction
Poly(ADP-ribose)polymerases (PARP) are a family of enzymes involved in the synthesis of poly(ADP-ribose) (PAR) chains from NAD + . Of the eighteen described members, three (PARP-1, -2, -3) have defined roles in DNA damage repair [1][2][3][4]. The role of PARP-1 is the most clearly defined and it contributes most significantly to PAR synthesis. In response to single strand breaks (SSBs) in DNA, PARP-1 binds to DNA and attaches PAR chains to nuclear proteins (PARylation) including itself (autoPARylation). The PAR chains recruit base excision repair (BER) enzymes to the SSB and ultimately lead to dissociation of PARP [1][2][3][4][5]. This role in DNA damage repair lead to an early proposal that PARP inhibitors (PARPi) could find utility in cancer therapy [2].
PARP inhibitor cytotoxicity derives from a variety of mechanisms which have been well described elsewhere [6][7][8][9][10][11]. Briefly, catalytic inhibition of PARP stalls the BER process resulting in downstream DNA lesions when the replications forks collide with unrepaired SSB [7,12,13]. DNA lesions are also generated through the inability of inhibited PARP to dissociate from DNA, so-called "PARP trapping" [9][10][11]. In normal tissues, these lesions are repaired with high fidelity by homologous recombination repair (HRR), however, in HRR incompetent cell (e.g., BRCA1 and BRCA2 mutants) repair occurs via low fidelity pathways, including non-homologous end joining (NHEJ), resulting in an accumulation of errors and ultimately cell death [12][13][14]. This combination of a genetic defect and a pharmacological treatment combining to cause cell death is a form of "synthetic lethality" and has provided the context for clinical PARPi approvals to date [14][15][16].
In tandem with development of potent small molecule PARPi, increased investigation of PARP biology has established involvement of the PARP family in the wider DNA damage response [3,4]. In addition to involvement in BER, PARPs participate in HRR, canonical NHEJ (cNHEJ) and alternate end joining (alt-EJ), and have numerous interactions with nuclear proteins of unknown consequence [3,4,17,18]. Due to this widespread involvement, PARPi can sensitize cells to a variety of DNA damaging agents, and therefore combination with cytotoxic chemotherapies or radiotherapy has been proposed as an approach for treatment of HRR competent tumors [19,20]. However, studies have shown that use of PARPi in combination therapies often lead to normal tissue toxicity requiring reduction in the dose of either the PARPi or chemotherapeutic agent [21][22][23][24][25][26][27][28].
Hypoxia is a well-established feature of many solid tumors which contributes to both tumor progression and resistance to therapy [29][30][31][32][33][34]. As tumors grow, an oxygen gradient develops as its metabolic consumption outstrips the oxygen supply. Tumor vasculature lacks the organization of normal tissue vasculature which leads to tumor hypoxia, with chronic hypoxia due to oxygen diffusion limitations, and acute hypoxia caused by transient blockages or flow reversals [29,34].
We, and others, have demonstrated that hypoxia can be exploited to activate a prodrug selectively within a tumor [29,32,35]. These hypoxia-activated prodrugs (HAPs) rely on the different metabolic fates of a bioreducible functional group (i.e., a trigger) in oxygenated versus hypoxic environments. One such trigger, the nitroaromatic group, is reduced by one-electron reductases to a nitro radical anion [29,32]. Under normoxia, this radical anion is oxidized back to the parent nitro group, whereas under hypoxia, direct fragmentation of the radical anion, or further reduction to electron-donating hydroxylamino or amino groups leads to activated species [36]. This shift in electron density can activate the drug via fragmentation of a frangible linker (e.g., evofosfamide) [37] or through activation of a reactive centre (e.g., PR-104) [38].
We considered that tumor-selective delivery of a PARPi via a HAP would increase the therapeutic index of PARPi in combination with radiotherapy or chemotherapy. To explore this proposition we started with olaparib (Lynparza) 1 as an ideal "effector" for use in a HAP as it has nanomolar potency as a PARP-1 inhibitor and recently gained first-in-class registration in an BRCA mutant advanced ovarian cancer setting as a monotherapy [15,39].
The PARPi binding mode exemplified by olaparib 1 relies on a tridentate hydrogen-bond network which mimics the natural substrate nicotinamide, Figure 1. The phthalazinone carbonyl interacts with both Ser904-OH and Gly863-NH and the amide proton interacts with Gly863-CO. Additional interactions are formed by Tyr907 and Tyr986 forming π-stacking arrangements with bound inhibitor [40].
Molecules 2019, 24 x BRCA2 mutants) repair occurs via low fidelity pathways, including non-homologous end joining (NHEJ), resulting in an accumulation of errors and ultimately cell death [12][13][14]. This combination of a genetic defect and a pharmacological treatment combining to cause cell death is a form of "synthetic lethality" and has provided the context for clinical PARPi approvals to date [14][15][16].
In tandem with development of potent small molecule PARPi, increased investigation of PARP biology has established involvement of the PARP family in the wider DNA damage response [3,4]. In addition to involvement in BER, PARPs participate in HRR, canonical NHEJ (cNHEJ) and alternate end joining (alt-EJ), and have numerous interactions with nuclear proteins of unknown consequence [3,4,17,18]. Due to this widespread involvement, PARPi can sensitize cells to a variety of DNA damaging agents, and therefore combination with cytotoxic chemotherapies or radiotherapy has been proposed as an approach for treatment of HRR competent tumors [19,20]. However, studies have shown that use of PARPi in combination therapies often lead to normal tissue toxicity requiring reduction in the dose of either the PARPi or chemotherapeutic agent [21][22][23][24][25][26][27][28].
Hypoxia is a well-established feature of many solid tumors which contributes to both tumor progression and resistance to therapy [29][30][31][32][33][34]. As tumors grow, an oxygen gradient develops as its metabolic consumption outstrips the oxygen supply. Tumor vasculature lacks the organization of normal tissue vasculature which leads to tumor hypoxia, with chronic hypoxia due to oxygen diffusion limitations, and acute hypoxia caused by transient blockages or flow reversals [29,34].
We, and others, have demonstrated that hypoxia can be exploited to activate a prodrug selectively within a tumor [29,32,35]. These hypoxia-activated prodrugs (HAPs) rely on the different metabolic fates of a bioreducible functional group (i.e., a trigger) in oxygenated versus hypoxic environments. One such trigger, the nitroaromatic group, is reduced by one-electron reductases to a nitro radical anion [29,32]. Under normoxia, this radical anion is oxidized back to the parent nitro group, whereas under hypoxia, direct fragmentation of the radical anion, or further reduction to electron-donating hydroxylamino or amino groups leads to activated species [36]. This shift in electron density can activate the drug via fragmentation of a frangible linker (e.g., evofosfamide) [37] or through activation of a reactive centre (e.g., PR-104) [38].
We considered that tumor-selective delivery of a PARPi via a HAP would increase the therapeutic index of PARPi in combination with radiotherapy or chemotherapy. To explore this proposition we started with olaparib (Lynparza) 1 as an ideal "effector" for use in a HAP as it has nanomolar potency as a PARP-1 inhibitor and recently gained first-in-class registration in an BRCA mutant advanced ovarian cancer setting as a monotherapy [15,39].
The PARPi binding mode exemplified by olaparib 1 relies on a tridentate hydrogen-bond network which mimics the natural substrate nicotinamide, Figure 1. The phthalazinone carbonyl interacts with both Ser904-OH and Gly863-NH and the amide proton interacts with Gly863-CO. Additional interactions are formed by Tyr907 and Tyr986 forming π-stacking arrangements with bound inhibitor [40].  We predicted that the addition of a 2-nitroimidazolyl trigger to the phthalazinone NH of olaparib 1 would disrupt the bonding interaction with Gly863-CO, resulting in a detrimental effect on PARP inhibition. This concept has precedence in the work of Threadgill and co-workers who installed nitroheterocyclic triggers on a series of isoquinolin-1-ones 2, Figure 2, and demonstrated modest abrogation of PARP inhibition [42,43]. Fragmentation of 2-nitrofuryl prodrugs 3a,b and 2-nitroimidazolyl prodrug 3c released effectors 2a-c, respectively, following chemical reduction (NaBH 4 , Pd/C; SnCl 2 ; Zn/NH 4 Cl) [42,43].

Molecules 2019, 24 x
We predicted that the addition of a 2-nitroimidazolyl trigger to the phthalazinone NH of olaparib 1 would disrupt the bonding interaction with Gly863-CO, resulting in a detrimental effect on PARP inhibition. This concept has precedence in the work of Threadgill and co-workers who installed nitroheterocyclic triggers on a series of isoquinolin-1-ones 2, Figure 2, and demonstrated modest abrogation of PARP inhibition [42,43]. Fragmentation of 2-nitrofuryl prodrugs 3a,b and 2-nitroimidazolyl prodrug 3c released effectors 2a-c, respectively, following chemical reduction (NaBH4, Pd/C; SnCl2; Zn/NH4Cl) [42,43].  To build on this initial observation and to explore the potential of this prodrug approach we prepared a series of model compounds and related 2-nitroimidazolyl derivatives based on a series of PARPi. We prepared phthalazinone 4 as a representative of the structural core of olaparib 1 and the corresponding 2-nitroimidazolyl derivative 5 as a model HAP in order to assess disruption of PARP inhibition. To assess trigger fragmentation, we prepared derivatives (3c, 7, 8, 10, 11 and 13, Figure 3) of model PARPi (2a, 2c, 6 and 9, Figure 3) that represent the core of literature PARPi. We also considered other possible sites of the PARPi core for placement of a trigger. The PARPi NU1025 12, Figure 3, contains a phenol distal to the core binding motif common to PARPi. This phenol may interact with Glu988 via a hydrogen-bond bridge with water molecules (depicted as red spheres in Figure 1) [40]. We proposed that forming a 2-nitroimidazolyl methyl ether at this phenol would hinder this interaction and could undergo fragmentation following reduction, providing an alternate prodrug strategy [44,45].
We evaluated the potency of model HAPs 5 and 13 and the associated PARPi 4 and 12 in a biochemical assay (Reaction Biology Corp, Malvern, PA) as representatives of these two approaches and investigated the stability of all model HAPs in a radiolytic reduction assay. To build on this initial observation and to explore the potential of this prodrug approach we prepared a series of model compounds and related 2-nitroimidazolyl derivatives based on a series of PARPi. We prepared phthalazinone 4 as a representative of the structural core of olaparib 1 and the corresponding 2-nitroimidazolyl derivative 5 as a model HAP in order to assess disruption of PARP inhibition. To assess trigger fragmentation, we prepared derivatives (3c, 7, 8, 10, 11 and 13, Figure 3) of model PARPi (2a, 2c, 6 and 9, Figure 3) that represent the core of literature PARPi. We also considered other possible sites of the PARPi core for placement of a trigger. The PARPi NU1025 12, Figure 3, contains a phenol distal to the core binding motif common to PARPi. This phenol may interact with Glu988 via a hydrogen-bond bridge with water molecules (depicted as red spheres in Figure 1) [40]. We proposed that forming a 2-nitroimidazolyl methyl ether at this phenol would hinder this interaction and could undergo fragmentation following reduction, providing an alternate prodrug strategy [44,45].
We evaluated the potency of model HAPs 5 and 13 and the associated PARPi 4 and 12 in a biochemical assay (Reaction Biology Corp, Malvern, PA) as representatives of these two approaches and investigated the stability of all model HAPs in a radiolytic reduction assay.

Synthesis of PARPi
Benzyl phthalazinone 4 was prepared from isobenzofuranone 14 via addition of hydrazine hydrate, Scheme 1 [39]. Benzamide 6 was prepared by alkylation of hydroxybenzoic acid 15 in the presence of potassium iodide and potassium carbonate [46]. The remaining amide PARPi scaffolds (2a, 2c, and 9, Figure 3) were sourced commercially. The phenol containing PARPi NU1025 12 was prepared in four steps from benzoic acid 16 as previously described by Griffin et al. [47].

Synthesis of PARPi
Benzyl phthalazinone 4 was prepared from isobenzofuranone 14 via addition of hydrazine hydrate, Scheme 1 [39]. Benzamide 6 was prepared by alkylation of hydroxybenzoic acid 15 in the presence of potassium iodide and potassium carbonate [46]. The remaining amide PARPi scaffolds (2a, 2c, and 9, Figure 3) were sourced commercially. The phenol containing PARPi NU1025 12 was prepared in four steps from benzoic acid 16 as previously described by Griffin et al. [47].

Synthesis of Model HAPs
The key nitroimidazole alcohol intermediate 18 was prepared following the previously described route [35,48]. Conversion to chloride 19 was achieved by mesylation of 18 and in situ chloride displacement, Scheme 2 [44]. Alcohol 18 was converted to amine 21 via reductive amination of aldehyde 20, prepared by oxidation of 18. Aldehyde 20 was converted to secondary alcohol 22 by addition of a methyl titanium species as described by Winn et al. [49]. Oxidation to ketone 23 was followed by reductive amination to provide secondary amine 24. Both amines 21 and 24 were used without isolation in subsequent reactions, Scheme 4, due to their instability.

Synthesis of Model HAPs
The key nitroimidazole alcohol intermediate 18 was prepared following the previously described route [35,48]. Conversion to chloride 19 was achieved by mesylation of 18 and in situ chloride displacement, Scheme 2 [44]. Alcohol 18 was converted to amine 21 via reductive amination of aldehyde 20, prepared by oxidation of 18. Aldehyde 20 was converted to secondary alcohol 22 by addition of a methyl titanium species as described by Winn et al. [49]. Oxidation to ketone 23 was followed by reductive amination to provide secondary amine 24. Both amines 21 and 24 were used without isolation in subsequent reactions, due to their instability. Preparation of model prodrugs 3c, 5, 10, 11, and 13 was achieved by combination of PARPi 2c, 4, 9, 2a, 12, and 19 in the presence of a suitable base. LiHMDS furnished isoquinolin-1-one prodrugs 3c and 11, and sodium hydride was used for phthalazinone prodrugs 5 and 10. Formation of the N-alkyl products was confirmed by comparison of the NCH2 13 C NMR data to literature values. In O-alkyl products the OCH2 resonance appears between δ 50-65, whereas the NCH2 resonance in compounds 3c, 5, 10, and 11 occurs in the range δ 30-50 as characteristic of N-alkylated compounds [50]. Phenoxy prodrug 13 was prepared using K2CO3 to deliver selective phenol alkylation. Preparation of model prodrugs 3c, 5, 10, 11, and 13 was achieved by combination of PARPi 2c, 4, 9, 2a, 12, and 19 in the presence of a suitable base Scheme 3. LiHMDS furnished isoquinolin-1-one prodrugs 3c and 11, and sodium hydride was used for phthalazinone prodrugs 5 and 10. Formation of the N-alkyl products was confirmed by comparison of the NCH 2 13 C NMR data to literature values.
In O-alkyl products the OCH 2 resonance appears between δ 50-65, whereas the NCH 2 resonance in compounds 3c, 5, 10, and 11 occurs in the range δ 30-50 as characteristic of N-alkylated compounds [50]. Phenoxy prodrug 13 was prepared using K 2 CO 3 to deliver selective phenol alkylation. Preparation of model prodrugs 3c, 5, 10, 11, and 13 was achieved by combination of PARPi 2c, 4, 9, 2a, 12, and 19 in the presence of a suitable base. LiHMDS furnished isoquinolin-1-one prodrugs 3c and 11, and sodium hydride was used for phthalazinone prodrugs 5 and 10. Formation of the N-alkyl products was confirmed by comparison of the NCH2 13 C NMR data to literature values. In O-alkyl products the OCH2 resonance appears between δ 50-65, whereas the NCH2 resonance in compounds 3c, 5, 10, and 11 occurs in the range δ 30-50 as characteristic of N-alkylated compounds [50]. Phenoxy prodrug 13 was prepared using K2CO3 to deliver selective phenol alkylation.  In contrast alkylation of primary amide 6 did not proceed despite screening a number of bases. In an alternative approach phenol 25 was first converted to acid 26 and then the corresponding acyl chloride which was used directly to acylate amine 21 giving prodrug 7, Scheme 4.
The fragmentation rates of 2-nitroheteroaromatic triggers are influenced by the ability of α-methylene substituents to stabilize developing positive charge in the fragmentation reaction [45]. Although an α-disubstituted trigger might be expected to provide maximal fragmentation rates these are extremely difficult to prepare [45,49]. We elected to prepare the α-substituted amide 8 to explore the influence of substitution on fragmentation rate. Formation of an acyl chloride from 26 and acylation of amine 24 yielded prodrug 8, Scheme 4.

Molecules 2019, 24 x
In contrast alkylation of primary amide 6 did not proceed despite screening a number of bases. In an alternative approach phenol 25 was first converted to acid 26 and then the corresponding acyl chloride which was used directly to acylate amine 21 giving prodrug 7, Scheme 4.
The fragmentation rates of 2-nitroheteroaromatic triggers are influenced by the ability of α-methylene substituents to stabilize developing positive charge in the fragmentation reaction [45]. Although an α-disubstituted trigger might be expected to provide maximal fragmentation rates these are extremely difficult to prepare [45,49]. We elected to prepare the α-substituted amide 8 to explore the influence of substitution on fragmentation rate. Formation of an acyl chloride from 26 and acylation of amine 24 yielded prodrug 8, Scheme 4.

Biochemical PARP-1 Inhibition
The PARP-1 inhibitory activity of the compounds was determined for the PARPi/model HAP pairs 4/5 and 12/13 in a radiometric PARP-1 inhibition assay, Table 1. Both compounds 4 and 12 were potent PARPi. Model HAP 5 provided significant (ca. 160-fold) deactivation of PARP inhibition, consistent with disruption of the key hydrogen-bonding network between the amide of PARPi 4 and PARP-1, Figure 1. In contrast, phenol prodrug 13 only showed ca. seven-fold disruption of PARP-1 inhibition, consistent with the added 2-nitroimidazolyl ether interfering with a secondary interaction, such as with Glu988, but not disrupting the key binding interactions.

Radiolytic Reduction
The ability of the model HAPs to release the effectors was assessed in a radiolytic reduction assay that provides obligate one-electron reduction with well-defined stoichiometry. Radiolysis of water generates both reducing (e -(aq)) and oxidizing (OH • , H • ) radicals but the latter can be scavenged by formate ions to generate the reducing CO2 •-radical to give a total of 0.62 μmol.J −1 reducing radicals [51,52]. Compounds (10 μM) in anoxic 100 mM sodium formate/5 mM sodium phosphate buffer, pH 7.0, were irradiated (40 Gy) using a cobalt-60 source and analyzed by HPLC with in-line photodiode

Biochemical PARP-1 Inhibition
The PARP-1 inhibitory activity of the compounds was determined for the PARPi/model HAP pairs 4/5 and 12/13 in a radiometric PARP-1 inhibition assay, Table 1. Both compounds 4 and 12 were potent PARPi. Model HAP 5 provided significant (ca. 160-fold) deactivation of PARP inhibition, consistent with disruption of the key hydrogen-bonding network between the amide of PARPi 4 and PARP-1, Figure 1. In contrast, phenol prodrug 13 only showed ca. seven-fold disruption of PARP-1 inhibition, consistent with the added 2-nitroimidazolyl ether interfering with a secondary interaction, such as with Glu988, but not disrupting the key binding interactions.

Radiolytic Reduction
The ability of the model HAPs to release the effectors was assessed in a radiolytic reduction assay that provides obligate one-electron reduction with well-defined stoichiometry. Radiolysis of water generates both reducing (e -(aq) ) and oxidizing (OH • , H • ) radicals but the latter can be scavenged by formate ions to generate the reducing CO 2 •radical to give a total of 0.62 µmol.J −1 reducing radicals [51,52]. Compounds (10 µM) in anoxic 100 mM sodium formate/5 mM sodium phosphate buffer, pH 7.0, were irradiated (40 Gy) using a cobalt-60 source and analyzed by HPLC with in-line photodiode array absorbance and single-stage quadrupole mass spectrometry immediately after irradiation and compared with unirradiated controls. Samples were also analyzed following incubation for 5 h at 37 • C after irradiation to allow for slow fragmentation of hydroxylamine or amine intermediates to be observed. However, incubation did not alter the quantitation or identity of the species produced. Therefore, only immediate analysis results are reported. Typical results are illustrated for compounds 5, Figure 4; 3c, Figure 5; and 13, with a summary of prodrug loss and product formation in Table 2. observed. However, incubation did not alter the quantitation or identity of the species produced. Therefore, only immediate analysis results are reported. Typical results are illustrated for compounds 5, Figure 4; 3c, Figure 5; and 13, Figure 7, with a summary of prodrug loss and product formation in Table 2. Radiolytic reduction of 5 resulted in an 85.3% decrease in diode-array signal, but we did not detect any released PARPi 4 in either the diode-array chromatogram (with comparison to authentic 4) or the extracted ion chromatogram at the 237.1 m/z base peak, Figure 4, Table 2. Inspection of the mass spectrum revealed the hydroxylamine derivative of 5 (362.2 m/z) in the irradiated solution eluting at an earlier retention time, Figure 4. Previous studies have shown facile fragmentation of nitroheteroaromatic ethers after reduction to a hydroxylamine intermediate [45,[53][54][55]. In this instance the stability of the hydroxylamine indicates that this system is too stable to release PARPi 4 [36,56]. as the corresponding hydroxylamines, Table 2. No evidence for the release of effectors was detected, although authentic standards were detected with high sensitivity. Solutions of the prodrugs incubated for 5 h at 37 °C post irradiation provided no evidence of effector release, suggesting slow fragmentation is unlikely. This suggests that directly-linked model HAPs of the amides, including bromoisoquinolinone 3c, Figure 5, do not readily fragment on reduction, in contrast to previous reports [42,43]. This discrepancy may be caused by the choice of reduction system. The formate radiolytic reduction system produces 4-electron reduction to the hydroxylamine species which we expected to fragment. However, we did not see evidence of 6-electron reduction to the more strongly electrondonating amino species. To address this we carried out reduction of 3c with Zn/NH4Cl and assessed the resulting solution by LC/MS. After one hour exposure to Zn/NH4Cl in acetonitrile the solution was filtered and analysed by LC/MS. Comparison to a control solution allowed estimated loss of prodrug at 45% based on the diode-array signal. At the extracted m/z 333.0 signal we detected the Radiolytic reduction of 5 resulted in an 85.3% decrease in diode-array signal, but we did not detect any released PARPi 4 in either the diode-array chromatogram (with comparison to authentic 4) or the extracted ion chromatogram at the 237.1 m/z base peak, Figure 4, Table 2. Inspection of the mass spectrum revealed the hydroxylamine derivative of 5 (362.2 m/z) in the irradiated solution eluting at an earlier retention time, Figure 4. Previous studies have shown facile fragmentation of nitroheteroaromatic ethers after reduction to a hydroxylamine intermediate [45,[53][54][55]. In this instance the stability of the hydroxylamine indicates that this system is too stable to release PARPi 4 [36,56].
We expanded our study to include analogous prodrugs based on the core scaffolds for other reported PARPi to see if the observed stability of the phthalazinone N-nitroimidazole framework was general [40]. We included bromoisoquinolinone 3c as it has previously been shown to fragment in a chemical reduction system [43]. We assessed all the model prodrugs in our radiolytic reduction assay, Table 2. Irradiation of amide model HAPs 3c, 7, 8, 10, and 11 resulted in loss of prodrug that was broadly consistent with the expected four-electron stoichiometry. The major products were identified as the corresponding hydroxylamines, Table 2. No evidence for the release of effectors was detected, although authentic standards were detected with high sensitivity. Solutions of the prodrugs incubated for 5 h at 37 • C post irradiation provided no evidence of effector release, suggesting slow fragmentation is unlikely. This suggests that directly-linked model HAPs of the amides, including bromoisoquinolinone 3c, Figure 5, do not readily fragment on reduction, in contrast to previous reports [42,43].
This discrepancy may be caused by the choice of reduction system. The formate radiolytic reduction system produces 4-electron reduction to the hydroxylamine species which we expected to fragment. However, we did not see evidence of 6-electron reduction to the more strongly electron-donating amino species. To address this we carried out reduction of 3c with Zn/NH 4 Cl and assessed the resulting solution by LC/MS. After one hour exposure to Zn/NH 4 Cl in acetonitrile the solution was filtered and analysed by LC/MS. Comparison to a control solution allowed estimated loss of prodrug at 45% based on the diode-array signal. At the extracted m/z 333.0 signal we detected the corresponding amino species, Figure 6A, confirmed by its mass spectrum, Figure 6B. Re-analysis of the same sample after standing for 3 h showed no change in the ion count for the amine (data not shown) and, importantly, the expected fragmentation product 2c was not detected despite ready detection of the authentic compound, Figure 6C,D. This result strengthens our observation that this type of prodrug does not release the desired PARP inhibitor, even when reduced to the corresponding amine. We are unable to account for observation of fragmentation in previous reports.
Molecules 2019, 24 x corresponding amino species, Figure 6A, confirmed by its mass spectrum, Figure 6B. Re-analysis of the same sample after standing for 3 h showed no change in the ion count for the amine (data not shown) and, importantly, the expected fragmentation product 2c was not detected despite ready detection of the authentic compound, Figure 6C,D. This result strengthens our observation that this type of prodrug does not release the desired PARP inhibitor, even when reduced to the corresponding amine. We are unable to account for observation of fragmentation in previous reports. This disappointing result led us to explore a new prodrug approach. We prepared a 2-nitroimidazolyl ether linked to the phenol of NU1025 (12). Related prodrugs based on nitroheterocyclic ethers have been demonstrated to release effectors efficiently [44,45,49]. Reduction of the PARPi prodrug 13 did result in release of effector 12, Table 2. Formation of 12 was detected in the diode-array chromatogram, Figure 7A  This disappointing result led us to explore a new prodrug approach. We prepared a 2-nitroimidazolyl ether linked to the phenol of NU1025 (12). Related prodrugs based on nitroheterocyclic ethers have been demonstrated to release effectors efficiently [44,45,49]. Reduction of the PARPi prodrug 13 did result in release of effector 12, Table 2. Formation of 12 was detected in the diode-array chromatogram, Figure 7A Figure 7C,D) with an efficient yield of 12 at ca. 55-60% reduction of prodrug, Table 2.
In conclusion, the N-alkylated nitroimidazolyl prodrug system did provide deactivation of the PARPi as intended. However, this model prodrug system does not fragment upon reduction to the corresponding hydroxylamine or amine. In contrast, the phenol prodrug 13 shows only ca. seven-fold reduction in PARP inhibition. Importantly, the 2-nitroimidazolyl ether does fragment upon reduction and this provides a lead towards identification of novel prodrugs which can combine the efficient fragmentation of the ether linker with a larger deactivation of the PARPi.  In conclusion, the N-alkylated nitroimidazolyl prodrug system did provide deactivation of the PARPi as intended. However, this model prodrug system does not fragment upon reduction to the corresponding hydroxylamine or amine. In contrast, the phenol prodrug 13 shows only ca. sevenfold reduction in PARP inhibition. Importantly, the 2-nitroimidazolyl ether does fragment upon reduction and this provides a lead towards identification of novel prodrugs which can combine the efficient fragmentation of the ether linker with a larger deactivation of the PARPi.

General
All non-aqueous reactions were carried out under a dry nitrogen atmosphere unless otherwise noted. DMF, DCM, and THF were purchased pre-dried and stored over molecular sieves from Acros Organics. All commercial reagents were used without purification. Flash column chromatography was carried out on a silica gel solid phase (Merck 230-400 mesh). Thin layer chromatography was carried out using Merck 60 F254 aluminium plates pre-coated with silica. Compounds were identified using UV fluorescence and/or staining with either vanillin in ethanolic sulphuric acid (with heating), 3, 5-dinotrophenylhydrazine in ethanolic sulfuric acid (with heating), ninhydrin in ethanol/glacial acetic acid (95:5)(with heating), or iodine on silica gel. Melting points were determined on an Electrothermal 2300 melting point apparatus. High resolution mass spectra (HRMS) were measured on an Agilent Technologies 6530 Accurate-Mass Quadrupole Time-of-Flight (Q-TOF) LC/MS interfaced with an Agilent Jet Stream electrospray ionization (ESI) source allowing positive or negative ions detection. Low resolution mass spectra (LRMS) were measured on a Surveyor MSQ mass spectrometer using an atmospheric pressure chemical ionization (APCI) mode with a corona voltage of 50 V and a source temperature of 400 °C. NMR spectra data were recorded on a Bruker

General
All non-aqueous reactions were carried out under a dry nitrogen atmosphere unless otherwise noted. DMF, DCM, and THF were purchased pre-dried and stored over molecular sieves from Acros Organics. All commercial reagents were used without purification. Flash column chromatography was carried out on a silica gel solid phase (Merck 230-400 mesh). Thin layer chromatography was carried out using Merck 60 F 254 aluminium plates pre-coated with silica. Compounds were identified using UV fluorescence and/or staining with either vanillin in ethanolic sulphuric acid (with heating), 3, 5-dinotrophenylhydrazine in ethanolic sulfuric acid (with heating), ninhydrin in ethanol/glacial acetic acid (95:5) (with heating), or iodine on silica gel. Melting points were determined on an Electrothermal 2300 melting point apparatus. High resolution mass spectra (HRMS) were measured on an Agilent Technologies 6530 Accurate-Mass Quadrupole Time-of-Flight (Q-TOF) LC/MS interfaced with an Agilent Jet Stream electrospray ionization (ESI) source allowing positive or negative ions detection. Low resolution mass spectra (LRMS) were measured on a Surveyor MSQ mass spectrometer using an atmospheric pressure chemical ionization (APCI) mode with a corona voltage of 50 V and a source temperature of 400 • C. NMR spectra data were recorded on a Bruker Avance 400 spectrometer (400 MHz, 1 H nuclei, 100 MHz, 13 C nuclei). All chemical shift (δ) values were reported in parts per million (ppm) relative to tetramethylsilane (0.0 ppm) as an internal reference, coupling constants were reported in Hertz (Hz). Final products were analyzed by reverse-phase HPLC (Altima C18 5 µm column, 150 mm × 3.2 mm; Alltech Associated, Inc., Deerfield, IL) using an Agilent HP1100 equipped with a photodiode array detector. Mobile phases were gradients of 80% CH 3 CN/20% H 2 O (v/v) in 45 mM ammonium formate at pH 3.5 and 0.5 mL/min. Purity was determined by monitoring at 330 ± 50 nm.
To benzoic acid 26 (0.16 g, 0.704 mmol) in DCM (3 mL) was added oxalyl chloride (0.30 mL, 3.54 mmol) and the mixture was stirred overnight at room temperature. Solvent was removed in vacuo. The crude residue was taken up in DCM (2 mL) and added to a solution of amine 21 (0.10 g, 0.64 mmol) in pyridine (2 mL) at 0 • C. The mixture was allowed to come to room temperature and stirred for 4 h. Solvent was removed in vacuo and the crude residue was purified by column chromatography (2:1, EtOAc, X4) to yield 7 (0.16 g, 70%) as yellow solid, mp 120-122 • C.