PARP Inhibitor Upregulates PD-L1 Expression and Enhances Cancer-Associated Immunosuppression

PARP engages in DNA base excision repair by inducing poly (ADP-ribosy)lation of itself and other target proteins (1). PARP inhibition has been shown to be an effective therapeutic strategy against tumors associated with germline mutation in double-strand DNA repair genes by inducing synthetic lethality (1). One PARP inhibitor (PARPi), olaparib, was approved by the FDA in 2014 for the treatment of germline BRCA-mutated (gBRCAm) advanced ovarian cancer (2). More recently, another PARPi, niraparib, which was shown to significantly prolong the progression-free survival in ovarian cancer patients, received a fast track designation from the FDA for the treatment of patients with recurrent platinum-sensitive ovarian cancer (3).

In addition to ovarian cancer, PARPi has demonstrated tremendous potential in breast cancer, and there are currently several active clinical trials evaluating PARPi-containing combination therapies for advanced breast cancer. For instance, several combinations of PARPi and targeted anticancer agents, such as inhibitors against PI3K (4, 5), Wee1 kinase (6), DNA topoisomerase I (7), and DNA methyltransferase (8), have been proposed to enhance the cytotoxic effect of PARPi. In addition, c-Met–mediated phosphorylation of PARP was reported to contribute to PARPi resistance, suggesting that the combined inhibition of c-Met and PARP may benefit patients who do not respond to PARPi and whose tumors are associated with c-Met activation (9). Thus, developing a rational combination therapy with PARPi may lead to effective anticancer strategy.

Over the past few years, there have been major breakthroughs in our understanding of tumor-associated immunosuppression. A key mechanism underlying cancer immune evasion is the expression of multiple inhibitory ligands, notably PD-L1, on the surface of cancer cells. Engagement of the PD-1 receptor on T cells by PD-L1 leads to the suppression of T-cell proliferation, cytokine release, and cytolytic activity, whereas blockade of coinhibitory ligation with mAbs, such as PD-L1 or PD-1 antibodies, restores T-cell function and increases therapeutic efficacy (10, 11). The impressive and durable clinical response of checkpoint blockade immunotherapy resulted in the FDA approval of ipilimumab, nivolumab, pembrolizumab, and more recently atezolizumab for the treatment of multiple types of cancer, such as melanoma, Hodgkin lymphoma, and lung and bladder cancers (12–15). Notably, the PD-1 antibody pembrolizumab was approved as first-line treatment for patients with advanced non–small cell lung cancer and high PD-L1 expression (16).

There is accumulating evidence indicating that conventional and targeted anticancer therapies also affect tumor-targeting immune responses (17). Thus, delineating the cross-talk between cytotoxic anticancer agents and cancer-associated immunity may lead to more efficient combinatorial regimens. Although the effects of PARPi, a targeted anticancer agent, have shown promising results in multiple cancer types, how and whether PARPi plays a role in cancer-associated immunity is still unknown. In the current study, we investigate the cross-talk between PARP inhibition and immune checkpoint, in particular, the PD-L1/PD-1 axis, which is a dominant immune checkpoint pathway in the tumor microenvironment, and further explore a mechanism-driven combination strategy to potentiate PARPi.

All cell lines were obtained from ATCC and have been independently validated by STR DNA fingerprinting at MD Anderson Cancer Center (Houston, TX). PARP1 (#sc-400046), PD-L1 (#sc-401140), and GSK3β (#sc-425249) knockout cells were established using CRISPR/Cas9 KO plasmids from Santa Cruz Biotechnology. PARP1 knockdown MDA-MB-231 cells were established as described previously (9).

PD-L1 (#13684), PARP1 (#9532), phospho-GSK3β (Ser9, #9336), GSK3β (#9315), and Ki-67 (#9449) antibodies were purchased from Cell Signaling Technology, BRCA1 (sc-8326) and BRCA2 (sc-642) were from Santa Cruz Biotechnology, and α-tubulin (#B-5-1-2) and β-actin (#A2228) were obtained from Sigma-Aldrich. Olaparib, rucaparib, and talazoparib were purchased from Selleckchem, ChemieTek, and Selleckchem, respectively. Human CD274 (B7-H1, PD-L1) antibody for T-cell killing assay was from BioLegend (#329709). eSiRNA human BRCA1 (EHU096311), eSiRNA human BRCA2 (EHU031451), and siRNA universal negative control (SIC001) were from Sigma-Aldrich.

The Human Phospho-Kinase Array Kit (#ARY003B) was purchased from R&D Systems. Array screening was performed following the manufacturer's protocol. Briefly, cell lysates were incubated with the array membranes. After washing, the membranes were incubated with biotinylated antibody cocktail. The amounts of phospho-kinase were assessed with streptavidin conjugated to horseradish peroxidase (HRP), followed by chemiluminescence detection. A GS-800 Calibrated Densitometer (Bio-Rad Laboratories) was used to quantify the density of each dot against the average of the internal controls on the membrane as indicated in the protocol.

For detection of cell surface PD-L1, cells were suspended in 100 μL of cell staining buffer (#420201, BioLegend) and incubated with APC-conjugated anti-human PD-L1 antibody (#329708, BioLegend) at room temperature for 30 minutes. After washing in the staining buffer, stained cells were analyzed by FACS (BD Biosciences).

Cells (1 × 10⁶) were incubated with 5 μg/mL recombinant human PD-1 FC chimera protein (#1086-PD-050, R&D Systems) at room temperature for 30 minutes. After washing in staining buffer, cells were incubated with anti-human Alexa Fluor 488 dye conjugated antibody (Thermo Fisher Scientific) at room temperature for 30 minutes. Cells were analyzed by FACS after wash in the staining buffer. The FACS data were analyzed using FlowJo (Tree Star), and the cut-off line for relative positive percentage was set at the median of the maximum signal.

NucLight RFP MDA-MB-231 cells (#4457, Essen Bioscience) were seeded in a 96-well plate with or without olaparib. Human peripheral blood mononuclear cells (PBMC; #70025, STEMCELL) were activated with 100 ng/mL CD3 antibody, 100 ng/mL CD28 antibody, and 10 ng/mL IL2 (#317303; #302913; #589102, BioLegend) and then cocultured with MDA-MB-231 cells at 10:1 ratio in the presence of fluorescence caspase-3/7 substrate (#4440, Essen Bioscience).

All animal procedures were conducted under the guidelines approved by the Institutional Animal Care and Use Committee at UT MD Anderson Cancer Center. MDA-MB-231 (0.5 × 10⁶), BT549 (1 × 10⁶), or SUM149 (2 × 10⁶) cells in Matrigel (BD Biosciences) were injected into the mammary fat pads of female nude mice of 6 to 8 weeks of age. When tumor volume reached approximately 50 mm³, mice were administered olaparib (25 mg/kg) or rucaparib (5 mg/kg) orally 5 days per week for 3 weeks. Tumors were collected after final treatment and analyzed by immunoblotting and IHC. IHC staining was performed as described previously (18). Briefly, frozen tissue sections were fixed with 4% paraformaldehyde for 1 hour and then hydrated in PBS for 5 minutes at room temperature. Sections were permeabilized with 0.5% Triton X-100 for 15 minutes at room temperature. The endogenous peroxidase activity was blocked with 0.3% hydrogen peroxide in methanol for 15 minutes at room temperature. After serum blocking, the slides were incubated overnight at 4°C with human PD-L1 antibody (#13684, Cell Signaling Technology, 1:50 dilution). Slides were then incubated with biotinylated secondary antibodies for 1 hour at room temperature, followed by incubation with avidin–biotin–HRP complex. Visualization was performed using 0.125% amino-ethylcarbazole chromogen. After counterstaining with Mayer's hematoxylin, slides were mounted.

BALB/c mice (6- to 8-week-old female, The Jackson Laboratory) were inoculated in the mammary fat pads with EMT6 (1 × 10⁵) cells in Matrigel. On days 3 after the inoculation, mice were injected intraperitoneally with 50 mg/kg olaparib or vehicle daily. After day 4, mice were injected intraperitoneally every 4 days with 75 μg anti-mouse PD-L1 antibody (10F.9G2, Bio X Cell) or control rat IgG2b (LTF-2, Bio X Cell). Tumor volumes were measured every 3 days with a digital caliper and were calculated using the formula: π/6 × length × width². Body weight was measured every 5 days.

EMT6 tumors were excised and digested in collagenase/hyaluronidase and DNase I and dissociated by gentleMACS Dissociator as described by the protocol of Tumor Dissociation Kit (Miltenyi Biotec). Tumor cells and tumor-infiltrating lymphocytes (TIL) were enriched and harvested separately by Percoll gradient (Sigma). Cell surface PD-L1 of tumor cell were stained with Brilliant Violet 421–conjugated anti-mouse PD-L1 antibody (#124315, BioLegend) and analyzed by FACS, and TILs were stained and analyzed by mass cytometry (CyTOF). The antibodies used to stain TILs are listed as follows: CD45-147Sm, CD3e-152Sm, CD4-172Yb, CD8a-168Er, and IFNγ-165Ho (Fluidigm).

IHC staining was performed as described previously (19). Human breast tumor tissue specimens were obtained following the guidelines approved by the Institutional Review Board at MD Anderson Cancer Center, and written informed consent was obtained from patients in all cases. Briefly, tissue specimens were incubated with antibodies against PAR (Enzo Life Sciences, Clone 10H, 1:200 dilution) and PD-L1 (Abcam, Clone 288, #ab205921, 1:100 dilution) and a biotin-conjugated secondary antibody and then incubated with an avidin–biotin–peroxidase complex. Visualization was performed using amino-ethylcarbazole chromogen. According to the histologic scores, the intensity of staining was ranked into four groups: high (+++), medium (++), low (+), and negative (–).

Statistical analyses were performed using GraphPad Prism software (GraphPad). Student t test or one-way ANOVA was used to compare experimental data. The Pearson χ² test was used to analyze IHC data. A P value <0.05 (*) was considered statistically significant.

Elevated PD-L1 expression in cancer cells has been shown to enhance PD-L1/PD-1 axis–mediated anticancer immunosuppression (11, 16). To determine whether inhibition of PARP affects the PD-L1 protein level, we treated MDA-MB-231 and BT549 cells with two different PARPis, olaparib and talazoparib, and determined PD-L1 expression by immunoblotting. PARPi treatment increased the total level of PD-L1 protein in both the cell lines (Fig. 1A). To validate whether PARPi-induced PD-L1 upregulation is through PARP1 inhibition, we knocked down (KD) or knocked out (KO) PARP1 in MDA-MB-231 cells. Consistent with the results shown in Fig. 1A, PD-L1 expression in the PARP1 KD and KO cells was substantially higher compared with the parental cells (Fig. 1B).

PD-L1 expressed on cell surface of cancer cells exerts immunosuppressive effects by binding to PD-1 receptor on activated T cells (20). To determine whether the level of cell surface PD-L1 increases after PARPi treatment, MDA-MB-231 cells were treated with or without PARPi and subjected to FACS using fluorescence-labeled PD-L1 antibody. Cell surface PD-L1 levels significantly increased after olaparib and talazoparib treatment (Fig. 1C), and the increase occurred in a dose-dependent manner (Supplementary Fig. S1). Likewise, cell surface PD-L1 levels were higher in PARP1 KD and KO MDA-MB-231 cells than in parental cells (Fig. 1D). Because PARPi is commonly used to treat BRCA-deficient cancers (21), we also investigated the effects of olaparib on BRCA-mutant SUM149 cells. SUM149 cells were treated with the different concentrations of olaparib for 10 days to mimic chronic PARPi exposure in clinic and subjected to FACS to determine PD-L1 expression. Consistent with the results in MDA-MB-231 cells, cell surface PD-L1 was significantly increased following olaparib treatment in a dose-dependent manner (Fig. 1E). To further validate the role of BRCA deficiency in PARPi-induced PD-L1 upregulation, we knocked down BRCA1 or BRCA2 in MDA-MB-231 cells and exposed cells to olaparib. Downregulation of BRCA1 or BRCA2 had virtually no effect on PARPi-induced PD-L1 expression (Supplementary Fig. S2). These results together suggested that PARPi can upregulate cell surface PD-L1 level in both BRCA-proficient and BRCA-deficient cells.

Next, we asked whether PARPi may affect PD-L1 expression in tumors; we inoculated BT549 and SUM149 cells into the mammary fad pads of nude mice, and after the tumor formed, administered olaparib to mice 5 days a week for 3 weeks. Tumor tissues from xenografts were isolated and subjected to immunoblotting with PD-L1 antibody. As shown, PD-L1 expression was substantially higher in the xenograft tumors from mice treated with olaparib compared with those from the untreated mice (Fig. 2A and B). We also assessed PD-L1 expression by IHC staining in MDA-MB-231 xenograft tumor tissues from mice that had been treated with another PARPi, rucaparib, and harvested from our previous study (9). Mice treated with rucaparib for 3 weeks had higher PD-L1 expression in their tumors compared with control mice (Fig. 2C). Together, these results indicated that PARPi upregulates PD-L1 expression in triple-negative breast cancer in vitro and in vivo.

Next, we explored the mechanism underlying PARPi-enhanced PD-L1 protein expression. Multiple signaling pathways, such as STAT, NF-κB, and mTOR, have been reported to regulate PD-L1 expression level (19, 22, 23). In an attempt to identify the potential signaling pathways that regulate PD-L1, we performed a phospho-kinase antibody array screen to identify kinases that are activated or inactivated following PARPi treatment. In the presence of PARPi treatment, the phosphorylation signal of GSK3α/β at Ser21 and Ser9, which represents its inactivated form (24), scored the second highest after the CHK2-p53 DNA repair pathway (Fig. 3A). CHK2 kinase is known to respond to DNA damage, and PARPi-induced CHK2 phosphorylation has previously been reported (25, 26), further lending support to the results of our screen. The finding that PARPi inactivates GSK3α/β is also in line with our recent report demonstrating inactivation of GSK3β (p-Ser9) stabilizes PD-L1 (27). Those findings together prompted us to investigate whether PARPi upregulates the expression of PD-L1 via inactivation of GSK3β. To this end, we examined the status of GSK3β phosphorylation at Ser9 in response to PARPi in SUM149 cells, BT549, and MDA-MB-231 cells by immunoblotting. The results indicated that PARPi treatment induced high GSK3β Ser9 phosphorylation that was associated with PD-L1 upregulation (Fig. 3B and C; Supplementary Fig. S3). Knocking out GSK3β significantly increased PD-L1 expression (Fig. 3C, lane 3 vs. 1). However, PD-L1 expression level in GSK3β KO cells was no longer enhanced by olaparib treatment (Fig. 3C, lane 4 vs. 3). These results suggested that inactivation of GSK3β is required for the PARPi-induced PD-L1 upregulation.

To understand the functional significance of PD-L1 upregulation by PARPi, we first asked whether PARPi-induced PD-L1 increases PD-1 binding on cells. Exposure of MDA-MB-231 cells to olaparib induced more PD-1 binding to PD-L1 on the cell surface (Fig. 4A). Similar results were observed in MDA-MB-231 PARP1 KD and KO cells (Fig. 4B), and in SUM149 cells treated with different concentrations of olaparib (Fig. 4C). Next, to determine whether PARPi-mediated PD-L1 upregulation, which resulted in increased PD-1 binding, affects T-cell function, we performed a T-cell–mediated killing assay by coculturing activated human PBMCs with MDA-MB-231 cells labeled with nuclear red fluorescence protein (RFP; NucLight Red MDA-MB-231) in the presence or absence of olaparib. As expected, olaparib efficiently inhibited cancer cell proliferation (black vs. blue in Fig. 4D, left; Supplementary Fig. S4A; Supplementary, Movie S1 vs. S2) but did not inhibit activated PBMC proliferation (Supplementary Fig. S4B). Interestingly, although MDA-MB-231 cells were sensitive to the T-cell killing (black vs. red in Fig. 4D, left; Supplementary Movie S3), those that were treated with olaparib were strongly resistant to activated T-cell killing (blue vs. yellow in Fig. 4D, left and right; Supplementary Movie S4), supporting the notion that upregulation of PD-L1 by PARPi may render the PARPi-treated cancer cells more resistant to T-cell killing. On the basis of the above results, we tested the T-cell killing effects of the combination of PD-L1 antibody and olaparib. The results showed that blockage of PD-L1 resensitized PARPi-treated MDA-MB-231 cells to activated T-cell killing and that the PARPi–PD-L1 antibody combination was more effective than each agent alone (Fig. 4E).

Next, we sought to determine whether PD-L1 blockade could further potentiate PARPi antitumor efficacy in vivo. We first treated a murine breast cancer cell line, EMT6, with olaparib with results showing significant induction of PD-L1 by olaparib (Fig. 5A). Consequently, we evaluated PARPi and anti-PD-L1 treatment alone or in combination in the EMT6 syngeneic mouse model. Consistent with our observations in vitro, both olaparib and anti-PD-L1 restricted tumor growth, but the combined treatment demonstrated better therapeutic benefit than each treatment alone (Fig. 5B and C). There were significantly fewer Ki-67–positive tumor cells in the combined treatment compared with each treatment alone (Fig. 5D). Mice that received the combination treatment did not show any significant changes in body weight or elevation in liver enzyme (ALT and AST), and kidney toxicity marker, blood urea nitrogen (Supplementary Fig. S5). At the end of the treatment, tumors resected from mice were subjected to dissociation, and tumor cells and TILs were separately harvested. Consistent with the observation in xenograft mouse model (Fig. 2), PARPi significantly upregulated PD-L1 expression on tumor cell surface as determined by FACS in EMT6 syngeneic mice (Fig. 5E). Analysis of TILs by CyTOF showed that tumor-infiltrating cytotoxic CD8⁺ T-cell population, as measured by the level of IFNγ, decreased after PARPi treatment. Meanwhile, the addition of anti-PD-L1 restored the cytotoxic CD8⁺ T-cell population (Fig. 5F). These results suggested that PARPi enhances cancer-associated immunosuppression through upregulation of PD-L1 and that PD-L1 blockade potentiates PARPi therapy.

PARP exerts its biological function through its PARylation enzyme activity, and PARP inhibitors are designed to inhibit its enzyme activity. Thus, the extent of protein PARylation was utilized to assess the efficacy of PARP inhibition (7). To further validate our findings in human cancer patient samples, we analyzed the correlations between PARylation level and PD-L1 expression in human breast tumor specimens using IHC. High level of protein PARylation was detected in 87 (75.0%) of the 116 specimens, of which 65 cases (74.7%) showed low PD-L1 expression (Fig. 6, top). The Pearson χ² test further showed the inverse correlation between PARylation level and PD-L1 expression exists in human cancer patient specimens (Fig. 6, bottom). These results supported the notion that high PARP enzyme activity suppresses PD-L1 expression.

Although the cytotoxic effects of PARPi have been well studied, the role of PARP inhibition in cancer-associated immunity is still largely unknown. In this study, we demonstrated that PARPi upregulates PD-L1 expression primarily through GSK3β inactivation. PARPi renders cancer cells more resistant to T-cell–mediated cell death, and PD-L1 blockade potentiates PARPi in vitro and in vivo. These data strongly suggested that PD-L1 upregulation by PARPi treatment attenuates PARPi therapeutic efficacy via tumor-associated immunosuppression, and simultaneous inhibition of PARP and PD-L1 may benefit breast cancer patients. There are currently three clinical trials testing the combination of PARPi (olaparib, niraparib, and BGB-290) and PD-L1 or PD-1 antibody in multiple cancer types (NCT02484404, NCT02657889, NCT02660034). The results of the current study provided scientific basis for these clinical trials.

Higuchi and colleagues recently investigated the combination of PARPi and CTLA4 antibody in the BR5-AKT ovarian cancer syngeneic mouse model and claimed to have observed a synergistic therapeutic effect (28); however, they indicated they did not observe such synergistic effect using the anti-PD-1 and PARPi combination in the same animal model. It is worthwhile to mention that PD-1 blockade in the BR5-AKT syngeneic mice did not affect T-cell activation or cytokine induction in the peritoneal tumor environment in their study (28), and therefore, synergistic effects may not be observed in combination with PARPi under their experimental condition. Moreover, because BR5-AKT tumors display high AKT activities that already inhibit GSK3β (29), it is possible that PARPi cannot further inhibit GSK3β and upregulate PD-L1 in the BR5-AKT tumors, and thus, the synergistic effects were not observed. In contrast, the results from our study indicated that PARPi upregulates PD-L1 in EMT6 tumors and PD-L1 blockade attenuated immunosuppression activity (Fig. 5F), which allowed us to observe an anti-PD-L1 therapy–potentiated antitumor activity of PARPi. Meanwhile, other studies have reported that chemotherapeutic agents, gemcitabine and paclitaxel, can induce PD-L1 in ovarian cancer cells (30, 31). The combination of paclitaxel and PD-L1/PD-1 blockade enhanced antitumor efficacy in an ID8 ovarian syngeneic mouse model (30). Therefore, whether the combination of PD-L1 blockade and PARPi induces synergistic effect in ovarian cancer warrants further investigation in a suitable animal model. Nonetheless, the mechanism of interaction between PARP and PD-L1/PD-1 as shown in the current study is timely and provides scientific basis to develop more effective combination therapies consisting of two powerful anticancer agents.

J.K. Litton reports receiving commercial research grants from Genentech, Medivation, Novartis, and Pfizer, reports receiving speakers bureau honoraria from Medscape, and is a consultant/advisory board member for Medivation and Novartis. B. Arun is an employee of Abbvie. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S. Jiao, M.-C. Hung

Development of methodology: S. Jiao, H.-H. Lee, C.-W. Li, S.-O. Lim

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Jiao, W. Xia, H. Yamaguchi, Y. Wei, M.-K. Chen, C.-K. Chou, B. Arun, G.N. Hortobagyi

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Jiao, W. Xia, M.-K. Chen, J.-M. Hsu, C.-W. Li, J. Litton, B. Arun, G.N. Hortobagyi, M.-C. Hung

Writing, review, and/or revision of the manuscript: S. Jiao, H. Yamaguchi, J.-L. Hsu, J. Litton, B. Arun, G.N. Hortobagyi, M.-C. Hung

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Jiao, J.-M. Hsu, W.-H. Yu, Y. Du, H.-H. Lee, S.-S. Chang, G.N. Hortobagyi

Study supervision: G.N. Hortobagyi, M.-C. Hung

This work was partially supported by the following: the NIH (CCSG CA016672), Cancer Prevention & Research Institutes of Texas (DP150052 and RP160710), Breast Cancer Research Foundation grant (to M.-C. Hung and G.N. Hortobagyi), Patel Memorial Breast Cancer Endowment Fund, National Breast Cancer Foundation, Inc., The University of Texas MD Anderson Cancer Center-China Medical University and Hospital Sister Institution Fund (to M.-C. Hung), Ministry of Science and Technology, International Research-intensive Centers of Excellence in Taiwan (I-RiCE; MOST 105-2911-I-002-302), and Ministry of Health and Welfare, China Medical University Hospital Cancer Research Center of Excellence (MOHW106-TDU-B-212-144003).