Restoration of Temozolomide Sensitivity by PARP Inhibitors in Mismatch Repair Deficient Glioblastoma is Independent of Base Excision Repair

Mismatch repair (MMR) is one of the critical pathways responsible for repair of base-base mismatches arising during DNA replication or otherwise caused by DNA damage. MMR deficiency resulting from inactivating MMR gene mutations not only predisposes cells to tumorigenesis, but also affects a tumor cell's response to DNA damaging agents. MMR deficiency causes drug resistance to alkylating agents that mediate the formation of O6 methylguanine-containing mismatches (1). Mechanistically, the recognition of alkylator-induced O6meG:T mismatches by the multiprotein complex MutSα (comprised of the heterodimer of MSH2 and MSH6) is thought to result in repeated replacement of the mispaired thymidine, rendering a futile cycle of replication/repair, and ultimately programmed cell death (2, 3). In the absence of a functional MMR response, O⁶meG:T is not targeted for this “futile repair” cycle, and the cells avoid programmed cell death and survive, albeit with the accumulation of a massive number of DNA mis-pairs, resulting in a characteristic cytidine-to-thymidine “hypermutator” phenotype (4).

Temozolomide (TMZ) is the most commonly used alkylator for gliomas, with significant clinical activity in both lower-grade tumors carrying isocitrate dehydrogenase 1 (IDH1) mutations (5, 6), as well as primary IDH1-wild-type glioblastoma (GBM) exhibiting methylation of the O6-methyguanine DNA methyltransferase (MGMT) promoter (7, 8). Unfortunately, prolonged treatment with TMZ is often followed by the development of acquired resistance to TMZ, contributing to malignant progression, tumor recurrence, and mortality. Inactivation of mismatch repair (MMR) genes, that is, MSH2, MSH6, MLH1, and PMS2, has been identified in both IDH mutant (9, 10) and IDH wild-type (11, 12) recurrent malignant gliomas that have been previously treated, whereas these alterations are extremely rare in primary tumors (13, 14). Studies have identified MSH6 inactivation as a key molecular mechanism of acquired TMZ resistance in glioma cells (15), and strong association between TMZ treatment and the development of MMR deficiency in patient tumors (16). Furthermore, MMR alterations after TMZ treatment of low-grade gliomas are considered a driver for malignant progression to higher grade and a post-TMZ hypermutator phenotype (14).

To potentiate TMZ cytotoxicity, combination therapies that modulate DNA repair pathways have been evaluated as a potential strategy to overcome TMZ resistance. TMZ induces not only Ome6G but also large numbers of N3-methyladenine and N7-methylguanine adducts that are rapidly processed by the base excision repair (BER) DNA repair pathway. N3-methyladenine and N7-methylguanine lesions activate the nuclear enzyme PARPs, which synthesize poly(ADP-ribose) chains (PARylation) and facilitate the recruitment of XRCC1, pol-beta, and DNA ligase to the DNA-strand break sites, thus playing an critical role in the initiation of BER machinery. PARPi combinations have been shown to increase TMZ sensitivity in the treatment of various cancer cell types (17–20), and are under active investigation in IDH wild-type (21, 22) and IDH mutant (23, 24) gliomas. Because of the documented role of PARP in single-strand break repair and BER, PARP inhibitor (PARPi)-mediated potentiation of TMZ sensitivity has been speculated to be due to its inhibition of BER, nevertheless direct evidence is lacking.

Here, we show that PARPis restore TMZ sensitivity in MSH6-inactivated, MMR-deficient TMZ-resistant gliomas (both IDH wild-type and IDH mutant) in vitro and in vivo. Notably, this restoration effect is selectively observed in MSH6-inactivated cells, representing a genotype-specific synthetic phenotype. Furthermore, PARPi-mediated restoration of TMZ sensitivity is independent of BER, as we show that genetic and pharmacologic blockage of BER pathway enzymes and PARP1 knockout fail to recapitulate this phenotype.

Human GBM cell line LN229 was obtained from the ATCC and was authenticated in 2017 by comparison of STR profile to the ATCC public dataset. Gli36 was provided by Dr. Khalid Shah, Boston, MA. Normal human astrocyte (NHA) was purchased from ScienCell. LN229 and NHA were maintained in DMEM with 4.5 g/L glucose, l-glutamine, and sodium pyruvate supplemented with 10% FBS and 1% penicillin/streptomycin/amphotericin. Patient-derived glioma neurosphere lines (MGG4, MGG123, MGG152) were established from patient tumors and cultured in serum-free neural stem cell medium as described previously (25–27). TMZ and methoxamine were purchased from Sigma, Veliparib was purchased from APE-BIO, Olaparib was purchased from Selleckchem, and APE inhibitor purchased from EMD MILLIPORE.

Cells were seeded in 96-well plates at 1,000 to 2,000 cells per well. After overnight incubation, compounds were serially diluted and added to wells. Cell viability was evaluated on day 6 by Cell Titer Glo (Promega) according to the manufacturer's instruction.

Cells were lysed in RIPA (Boston Bioproducts) with a cocktail of protease and phosphatase inhibitors (Roche). Protein (10–15 μg) was separated by 4% to 20% SDS-PAGE and transferred to polyvinylidene difluoride membranes by electroblotting. After blocking with 5% nonfat dry milk in TBS-T [20 mmol/L Tris (pH 7.5) 150 mmol/L NaCl, 0.1% Tween20] for 1 to 2 hours at room temperature, membranes were incubated with primary antibody at 4°C overnight. Membranes were washed in TBS-T and incubated with appropriate peroxidase conjugated secondary antibodies for 1 hour at room temperature. Signals were visualized using the enhanced chemiluminescense (ECL) Kit (Amersham Bioscience). Primary antibodies used were: MSH6 (#5425), XRCC1 (#2735; Cell Signal Technology), PARP1 (sc-8007; Santa Cruz), and β-actin (A1978; Sigma).

Lentivirus vector plasmids containing shRNA sequences for MSH6 (TRCN0000286578 shMSH6no.1) and XRCC1 (TRCN0000007913 no.1, TRCN0000011211 no.2), and nontargeting shRNA sequence (SHC002 shNS) were obtained from Sigma. shMSH6no.2, a second shRNA for MSH6, was from Dharmacon (V3LHS 318784). For PARP1 knockout, lentivirus vector plasmids containing SpCas9 (pLentiCRISPR v2) with/without sgRNA sequences for PARP1 (No.1:TTCTAGTCGCCCATGTTTGA, No.2: CCCCTTGCACGTACTTCTGT) were obtained from GenScript. To generate lentiviral particles, 293T cells were transfected with a lentiviral plasmid, packaging plasmid (pCMV-dR8.2), and envelope plasmid (pCMV-VSV-G) with FuGene (Promega). Cells were infected with lentivirus in the presence of polybrene (8 μg/mL) for 8 hours. Three days later, cells were selected with puromycin (0.6 μg/mL for LN229, 0.2 μg/mL for Gli36, MGG4, and MGG152) for 3 to 4 days before use. Knockdown and knockout were confirmed by Western blot analysis.

Five million LN229-shNS cells and LN229-shMSH6no.1 cells were implanted subcutaneously into the flank of 7-week-old female nude mice (NCI). When the maximum diameter of established tumors reached 5 mm, mice were randomized and treated with PBS (n = 6), TMZ (n = 6, 50 mg/kg i.p.), Veliparib (n = 6, 50 mg/kg oral gavage), TMZ plus Veliparib (n = 6) 5 consecutive days per week for 2 cycles with 1 week interval before the second treatment cycle. TMZ was dissolved in DMSO, diluted with PBS, and injected intraperitoneally. Veliparib was dissolved in DMSO, diluted with 10% 2-hydroxyl-propul-β-dextrine/PBS and administrated orally. Tumor diameters were measured twice a week using a digital caliper. Tumor volumes were calculated using the formula: tumor volume (mm³) = tumor length × tumor width²/2. Orthotopic model of MSH6-deficient GBM was established by intracerebral implantation of 5 × 10⁵ LN229-shMSH6no.1 cells in 7-week-old female nude mice (NCI) as described previously (26). Seven days later, mice were randomized and treated with PBS (n = 5), TMZ (n = 5), TMZ plus Veliparib (n = 5) at the same doses and schedules as the flank model. Animals were euthanized when significant deterioration of neurologic or general conditions was noted. All animal procedures were approved by Institutional Animal Care and Use Committee at Massachusetts General Hospital.

The Prism (GraphPad) software package was used for statistical analysis. Comparisons between two groups were done with Student t test (unpaired). Differences of tumor volumes were analyzed by Student t test. Survival analysis was performed using the Kaplan–Meier method, and the log-rank test (two-sided) was used to compare survival differences between treatment arms. P values less than 0.05 were considered statistically significant.

We used lentivirus vectors expressing MSH6 shRNA or control (nontargeting, NS) shRNA to establish an isogenic model system of MSH6 inactivation using GBM cell lines (LN229 and Gli36) and patient-derived GBM sphere lines (MGG4, primary IDH1 wild-type GBM; and MGG152, recurrent, IDH1 mutant GBM; ref. 28; Fig. 1A–C; Supplementary Fig. S1D). As expected, these shMSH6 cell lines showed TMZ resistance across a wide range of TMZ doses (Supplementary Figs. S1A–S1D). MSH6 knockdown did not alter sensitivity to PARPi monotherapy in LN229 (Fig. 1D). Combination treatment with TMZ and PARPi, at PARPi doses only marginally (<20%) cytotoxic as monotherapy, greatly restored TMZ sensitivity across shMSH6 GBM cell lines (Fig. 1A–C; Supplementary Fig. S1E). Importantly, PARPis had no, or only marginal, additional effects on TMZ-induced reduction of cell viability in MSH6 proficient control (shNS) GBM cells (Fig. 1A–C; Supplementary Fig. S1E). Testing of a second independent shRNA sequence targeting MSH6 in LN229 and Gli36 showed induction of TMZ resistance and similar restoration of TMZ sensitivity by PARPis, suggesting on-target effects of the two shRNAs (Supplementary Figs. S1F–S1I). The second shRNA for MSH6 also confirmed that MSH6 silencing did not meaningfully alter response to PARPis (Supplementary Fig. S1J). Next, we tested PARPis in a patient-derived GBM neurosphere line (MGG123) that was derived from a GBM that recurred after TMZ treatment. We confirmed that MGG123 had loss of MSH6 and exhibited high resistance to TMZ (Fig. 1E). In accord with the results with MSH6 knockdown GBM cells, PARPi treatment restored response to TMZ in MSH6-deficient MGG123 even when PARPi monotherapy had a marginal effect (Fig. 1F). The combination of TMZ and veliparib had only modest toxicity for proliferating human normal astrocytes compared with TMZ monotherapy (Fig. 1G).

To assess the effects of veliparib combination with TMZ on MSH6-deficient GBM in vivo, we treated nude mice bearing LN229-shNS and LN229-shMSH6 flank tumors with vehicle, veliparib alone, TMZ alone, or TMZ combination with veliparib. Veliparib alone had no effect on tumor growth compared with vehicle in both LN229-shNS and LN229-shMSH6 tumor models, consistent with the results obtained in vitro. Both TMZ and combination showed potent inhibition of LN229-shNS tumor growth—adding veliparib to TMZ provided no significant therapeutic benefit to TMZ alone (Fig. 2A). In the LN229-shMSH6 tumor model, however, the combination treatment of veliparib with TMZ significantly inhibited tumor outgrowth better than TMZ monotherapy (*, P = 0.019, day32 TMZ vs. TMZ + veliparib, Student t test; Fig. 2A). We further tested the effect of veliparib combination with TMZ in mice harboring intracerebral LN229 MSH6-knockdown GBM xenografts. We omitted the group of veliparib monotherapy given the lack of efficacy in flank tumor models. One week after intracerebral implantation, animals started receiving TMZ alone or TMZ combined with veliparib treatment. Veliparib combination with TMZ resulted in an extension of animal survival (P = 0.02; long lank test), compared with TMZ alone that did not extend survival in this model (Fig. 2B). During the treatment, animals receiving the regimens involving TMZ lost 10% body weight, however they recovered promptly after completion of treatment. Combining veliparib with TMZ did not confer additional toxicity as the body weights of TMZ and combination groups were comparable (Fig. 2C).

PARPis including veliparib potentiate TMZ-induced cytotoxicity in diverse tumor types (17–20). A proposed mechanism of this PARPi effect is mediated via inhibition of DNA single-strand break (SSB) repair via interfering with BER pathway (Fig. 3A). The BER inhibitors methoxyamine (MeOX) and APE inhibitor (APEi) have been reported to potentiate the cytotoxicity of alkylating agents including TMZ by disrupting BER through an accumulation of mutagenic and toxic apurinic/apyrimidinic (AP) sites (APEi) or preventing AP endodeoxyribonuclease cleavage (MeOX; refs. 29–32; Fig. 3A). Therefore, we tested APEi and MeOX to determine whether these BER pathway inhibitors could recapitulate the observed PARPi-mediated resensitization to TMZ in LN229-shMSH6 and Gli36-shMSH6 cells. To our surprise, unlike PARPi, neither of these BER repair-modulating drugs resensitized MSH6 knockdown cells to TMZ, even at concentrations that displayed signs of toxicity as a single agent (Fig. 3B–E; Supplementary Figs. S2D–S2G).

The scaffold protein XRCC1 plays a major role in coordinating BER by interacting with polymerase beta, PARP, and DNA ligase III at sites of DNA damage (Fig. 3A). XRCC1-deficient mouse embryonic fibroblasts have been shown to exhibit hypersensitivity to alkylating agents (33). We evaluated whether shRNA knockdown of XRCC1 altered TMZ sensitivity in our isogenic GBM cells with and without MSH6 knockdown (Fig. 3F). XRCC1 knockdown did not alter the TMZ-resistant phenotype of shMSH6 GBM cells and this was consistent with two independent shRNA constructs (Fig. 3G and H; Supplementary Figs. S2A and S2B). With intact MSH6, XRCC1 silencing rendered GBM cells slightly but significantly less responsive to TMZ (Fig. 3G and H; Supplementary Figs. S2A and S2B), unlike what was reported with Xrcc1^−/− mouse embryonic fibroblasts (33, 34). XRCC1 knockdown did not significantly alter response to PARPi in MSH6 wild-type and knockdown GBM cells (Fig. 3G and H; Supplementary Fig. S2C). Notably however, both veliparib and olaparib retained the ability to restore TMZ sensitivity in GBM cells that had dual knockdown of MSH6 and XRCC1 (Fig. 3G and H); we did not observe additional therapeutic benefit by silencing XRCC1. These results indicate that PARPi restoration of TMZ sensitivity in MSH6 inactivated GBM cells was not due to its blockage of BER signaling.

The biological effects mediated by PARPis are distinct from PARP knockdown, and PARPis often confer greater cytotoxicity than PARP1 knockdown when combined with TMZ (35, 36). PARPis physically impair the dissociation of PARP from DNA, which is required for the proper progress of repair process, and this “PARP trapping” at DNA damage sites has been shown, in certain contexts, to be more cytotoxic than persistent SSBs in the absence of PARP (37). PARPi have varying degrees of “PARP trapping” (olaparib greater than veliparib), and the possible involvement of this mechanism motivated us to test whether PARP1 knockdown restores TMZ sensitivity in MSH6-deficient cells as was observed with PARPis.

We used CRISPR-Cas9 lentivirus to inactivate PARP1 in our isogenic LN229 cells with and without MSH6 inactivation (Fig. 4A). Two independent sgRNAs for PARP1 consistently rendered shMSH6 GBM cells less responsive to PARPis, reflecting loss of their main target (Fig. 4B). Strikingly, PARP1 knockout did not alter responsiveness to TMZ in both MSH6 intact and MSH6 deficient GBM cells (Fig. 4C and D). Furthermore, analogous to what was observed with BER inhibitors, both veliparib and olaparib resensitized GBM cells with double silencing of MSH6/PARP1 to TMZ (Fig. 4E and F). These results show that PARP1 knockout does not recapitulate the PARPi-mediated effect of TMZ resensitization in MSH6-deficient GBM cells.

Multiple NCI consortia sponsored trials are investigating the potential efficacy of PARPi in combination with TMZ in both newly-diagnosed (NCT02152982, NCT00770471) and recurrent (NCT01026493, NCT01390571) GBM (38). Notably, preclinical studies have shown that gliomas can have upfront variability to PARPi when combined with TMZ, with TMZ-sensitive MGMT promoter methylated GBMs noted to be responders (21, 22, 39). It is important to understand mechanisms underlying the potentiation effect by PARPi and to identify molecularly defined subsets of gliomas that likely benefit from this combination treatment.

We here show that PARPi restores TMZ cytotoxicity selectively in GBM cells characterized by MMR-mediated TMZ resistance, both in vitro and in vivo. MSH6 knockdown did not alter sensitivity to PARPi, and PARPi monotherapy itself had little effect in either MMR deficient or wild-type cells. Therefore, enhanced cytotoxicity was not the result of a traditional synthetic lethal interaction, as described previously for PARP inhibition and homologous recombination repair deficiency resulting from BRCA gene alterations (40) or IDH mutations (23, 24). Our findings are consistent with reports showing that PARPi potentiation of TMZ cytotoxicity is limited to tumors that are intrinsically resistant to TMZ or have acquired resistance (41, 42). Our studies conclusively demonstrate, in experiments using isogenic cell models across a large panel of GBM lines, and in an orthotopic model, that MMR deficiency arising in gliomas from MSH6 inactivation is a unique molecular alteration predicting PARPi restoration of TMZ chemosensitivity. Importantly, relatively low concentrations of PARPi were sufficient to resensitize MMR-deficient glioma cells to TMZ, contrasting to negligible combination effect seen in MMR-proficient control cells at the same PARPi concentrations. Together with significant in vivo efficacy and safety in our orthotopic mouse model, our results suggest a wide therapeutic window that could be clinically meaningful for dosing regimens that otherwise circumvent toxicity issues reported in early veliparib combination trials (43).

From a mechanistic standpoint, it is well-established that the majority of TMZ-induced adducts are N3-methyladenine and N7-methylguanine, which are rapidly processed by BER machinery, whereas O6meG adducts, despite accounting for only 5% of total lesions, are responsible for TMZ cytotoxicity. In the combination treatment of PARPi with TMZ, PARPi has been postulated to impair BER by inhibiting PARP-mediated PARylation and cause cytotoxicity through increasing N3-methyladenine and N7-methylguanine–induced damages (44, 45). In accord with this working hypothesis, alterations to BER enzymes result in cell death by an accumulation of toxic and mutagenic BER intermediates after treatment with alkylating agents (46). Disruption of BER via APE inhibition and XRCC1 inactivation might be expected to therefore lead to the restoration or potentiation of TMZ sensitivity. However, our results show that BER perturbation by APE inhibition and XRCC1 inactivation does not recapitulate the effects of PARPis in MMR deficient, TMZ-resistant glioma cells. Furthermore, PARPi restoration of TMZ responsiveness was not affected by XRCC1 genetic inactivation, and there was no benefit in combining PARPi and XRCC1 depletion, consistent with the combination effects of PARPi 4-AN and TMZ seen in Xrcc1^−/− mouse embryonic fibroblasts (34). Together, these data strongly suggest that PARPi-mediated resensitization to TMZ in MMR-deficient glioma cells is independent of BER pathway inactivation.

PARPi exert anticancer effects via not only catalytic inhibition of PARPs but also trapping PARPs on sites of DNA damage, and these diverse PARPi functions explain why PARPi and PARP1 knockout often have different biological effects. Indeed, we observed a striking discrepancy between the phenotypes of PARPi when compared with genetic PARP1 knockout as only the former not the latter sensitized MSH6-deficient GBM cells to TMZ. We found that PARPi (particularly olaparib) combination therapy with TMZ for MMR-deficient glioma does not appear to depend on the expression levels of PARP1, as a mediator of PARP-trapping damage signaling, further suggesting that PARP1 trapping may not play a major role in this phenotype. Notably, veliparib, which has a lesser trapping ability than olaparib, was still capable of restoring sensitivity to TMZ in the absence of PARP1. The known targets of veliparib and olaparib include PARP1 and 2 (47, 48), but these inhibitors could cross-target other PARP enzymes, as has been reported for the ability of olaparib to inhibit PARP3 (49). Thus, although catalytic inhibition of PARP1 alone is not sufficient, catalytic inhibition and/or selective trapping of other PARP family members is necessary for producing chemosensitization of MMR-deficient glioma cells. Although the functions of many of seventeen known PARP family members have not been characterized, PARP2, PARP3, and PARP10 (50, 51) have been shown to participate in DNA damage repair. These PARPs are potential candidate mediators of the observed PARPi-mediated restoration of TMZ sensitivity in MMR-deficient cells.

In summary, we demonstrate that PARPi combination with TMZ selectively restores chemosensitivity in MSH6-inactivated, MMR-deficient GBM cells. These results identify a genetically-defined subgroup of recurrent gliomas, which may benefit from combination therapy of TMZ and PARPis. Our findings therefore support the assessment of TMZ combination with veliparib, or olaparib which is also brain penetrant (52), in glioma patients selected for MMR deficiency. From a mechanistic standpoint, we also show that this restoration effect does not require PARP1 and is independent of BER pathway, which has significant implications for our understanding of the pathways underlying the combination treatment effect. Further studies are needed to understand the precise mechanisms of PARPi chemorestoration to dissect the functions of PARPs that are targeted by this PARPi effect.

D.P. Cahill is a paid consultant for Merck and Lilly. No potential conflicts of interest were disclosed by the other authors.

Conception and design: F. Higuchi, D.P. Cahill

Development of methodology: F. Higuchi, J. Ning

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): F. Higuchi, H. Nagashima, M.V.A. Koerner

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): F. Higuchi, H. Wakimoto

Writing, review, and/or revision of the manuscript: F. Higuchi, H. Wakimoto, D.P. Cahill

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Nagashima

Study supervision: H. Wakimoto, D.P. Cahill

Burroughs Wellcome Fund Career Award (to D.P. Cahill), NIH R01CA227821 (to D.P. Cahill and H. Wakimoto), P50CA165962 (to D.P. Cahill), Tawingo fund (to D.P. Cahill), Loglio fund (to D.P. Cahill), OligoNation fund (to D.P. Cahill).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.